]>
Commit | Line | Data |
---|---|---|
bb1d8b11 AM |
1 | diff -urN linux-orig/arch/alpha/Kconfig linux-orig2/arch/alpha/Kconfig |
2 | --- linux-orig/arch/alpha/Kconfig 2004-10-18 16:55:37.000000000 -0500 | |
3 | +++ linux-orig2/arch/alpha/Kconfig 2004-10-22 11:29:33.507218717 -0500 | |
4 | @@ -600,3 +600,4 @@ | |
5 | ||
6 | source "lib/Kconfig" | |
7 | ||
8 | +source "cluster/Kconfig" | |
9 | diff -urN linux-orig/arch/arm/Kconfig linux-orig2/arch/arm/Kconfig | |
10 | --- linux-orig/arch/arm/Kconfig 2004-10-18 16:54:31.000000000 -0500 | |
11 | +++ linux-orig2/arch/arm/Kconfig 2004-10-22 11:30:56.358918506 -0500 | |
12 | @@ -690,3 +690,5 @@ | |
13 | source "crypto/Kconfig" | |
14 | ||
15 | source "lib/Kconfig" | |
c1c6733f | 16 | + |
bb1d8b11 AM |
17 | +source "cluster/Kconfig" |
18 | diff -urN linux-orig/arch/arm26/Kconfig linux-orig2/arch/arm26/Kconfig | |
19 | --- linux-orig/arch/arm26/Kconfig 2004-10-18 16:54:32.000000000 -0500 | |
20 | +++ linux-orig2/arch/arm26/Kconfig 2004-10-22 11:29:33.531218341 -0500 | |
21 | @@ -222,3 +222,4 @@ | |
22 | ||
23 | source "lib/Kconfig" | |
24 | ||
25 | +source "cluster/Kconfig" | |
26 | diff -urN linux-orig/arch/cris/Kconfig linux-orig2/arch/cris/Kconfig | |
27 | --- linux-orig/arch/cris/Kconfig 2004-10-18 16:55:07.000000000 -0500 | |
28 | +++ linux-orig2/arch/cris/Kconfig 2004-10-22 11:31:11.965673644 -0500 | |
29 | @@ -174,3 +174,5 @@ | |
30 | source "crypto/Kconfig" | |
31 | ||
32 | source "lib/Kconfig" | |
c1c6733f | 33 | + |
bb1d8b11 AM |
34 | +source "cluster/Kconfig" |
35 | diff -urN linux-orig/arch/i386/Kconfig linux-orig2/arch/i386/Kconfig | |
36 | --- linux-orig/arch/i386/Kconfig 2004-10-18 16:53:22.000000000 -0500 | |
37 | +++ linux-orig2/arch/i386/Kconfig 2004-10-22 11:29:33.533218309 -0500 | |
38 | @@ -1194,6 +1194,8 @@ | |
39 | ||
40 | source "lib/Kconfig" | |
41 | ||
42 | +source "cluster/Kconfig" | |
c1c6733f | 43 | + |
bb1d8b11 AM |
44 | config X86_SMP |
45 | bool | |
46 | depends on SMP && !X86_VOYAGER | |
47 | diff -urN linux-orig/arch/ia64/Kconfig linux-orig2/arch/ia64/Kconfig | |
48 | --- linux-orig/arch/ia64/Kconfig 2004-10-18 16:55:27.000000000 -0500 | |
49 | +++ linux-orig2/arch/ia64/Kconfig 2004-10-22 11:29:33.534218294 -0500 | |
50 | @@ -390,3 +390,5 @@ | |
51 | source "security/Kconfig" | |
52 | ||
53 | source "crypto/Kconfig" | |
c1c6733f | 54 | + |
bb1d8b11 AM |
55 | +source "cluster/Kconfig" |
56 | diff -urN linux-orig/arch/m68k/Kconfig linux-orig2/arch/m68k/Kconfig | |
57 | --- linux-orig/arch/m68k/Kconfig 2004-10-18 16:54:32.000000000 -0500 | |
58 | +++ linux-orig2/arch/m68k/Kconfig 2004-10-22 11:31:38.187262279 -0500 | |
59 | @@ -655,3 +655,5 @@ | |
60 | source "crypto/Kconfig" | |
61 | ||
62 | source "lib/Kconfig" | |
c1c6733f | 63 | + |
bb1d8b11 AM |
64 | +source "cluster/Kconfig" |
65 | diff -urN linux-orig/arch/mips/Kconfig linux-orig2/arch/mips/Kconfig | |
66 | --- linux-orig/arch/mips/Kconfig 2004-10-18 16:54:08.000000000 -0500 | |
67 | +++ linux-orig2/arch/mips/Kconfig 2004-10-22 11:29:33.541218184 -0500 | |
d3b4771f | 68 | @@ -1587,6 +1587,8 @@ |
bb1d8b11 AM |
69 | |
70 | source "lib/Kconfig" | |
d3b4771f | 71 | |
bb1d8b11 | 72 | +source "cluster/Kconfig" |
d3b4771f AM |
73 | + |
74 | # | |
75 | # Use the generic interrupt handling code in kernel/irq/: | |
76 | # | |
bb1d8b11 AM |
77 | diff -urN linux-orig/arch/parisc/Kconfig linux-orig2/arch/parisc/Kconfig |
78 | --- linux-orig/arch/parisc/Kconfig 2004-10-18 16:54:37.000000000 -0500 | |
79 | +++ linux-orig2/arch/parisc/Kconfig 2004-10-22 11:31:57.146964867 -0500 | |
80 | @@ -195,3 +195,5 @@ | |
81 | source "crypto/Kconfig" | |
82 | ||
83 | source "lib/Kconfig" | |
c1c6733f | 84 | + |
bb1d8b11 AM |
85 | +source "cluster/Kconfig" |
86 | diff -urN linux-orig/arch/ppc/Kconfig linux-orig2/arch/ppc/Kconfig | |
87 | --- linux-orig/arch/ppc/Kconfig 2004-10-18 16:55:29.000000000 -0500 | |
88 | +++ linux-orig2/arch/ppc/Kconfig 2004-10-22 11:29:33.550218043 -0500 | |
89 | @@ -1231,3 +1231,5 @@ | |
90 | source "security/Kconfig" | |
91 | ||
92 | source "crypto/Kconfig" | |
c1c6733f | 93 | + |
bb1d8b11 AM |
94 | +source "cluster/Kconfig" |
95 | diff -urN linux-orig/arch/ppc64/Kconfig linux-orig2/arch/ppc64/Kconfig | |
96 | --- linux-orig/arch/ppc64/Kconfig 2004-10-18 16:54:31.000000000 -0500 | |
97 | +++ linux-orig2/arch/ppc64/Kconfig 2004-10-22 11:32:11.150745212 -0500 | |
98 | @@ -352,3 +352,5 @@ | |
99 | source "crypto/Kconfig" | |
100 | ||
101 | source "lib/Kconfig" | |
c1c6733f | 102 | + |
bb1d8b11 AM |
103 | +source "cluster/Kconfig" |
104 | diff -urN linux-orig/arch/s390/Kconfig linux-orig2/arch/s390/Kconfig | |
105 | --- linux-orig/arch/s390/Kconfig 2004-10-18 16:53:51.000000000 -0500 | |
106 | +++ linux-orig2/arch/s390/Kconfig 2004-10-22 11:32:31.175431141 -0500 | |
107 | @@ -466,3 +466,5 @@ | |
108 | source "crypto/Kconfig" | |
109 | ||
110 | source "lib/Kconfig" | |
c1c6733f | 111 | + |
bb1d8b11 AM |
112 | +source "cluster/Kconfig" |
113 | diff -urN linux-orig/arch/sh/Kconfig linux-orig2/arch/sh/Kconfig | |
114 | --- linux-orig/arch/sh/Kconfig 2004-10-18 16:55:29.000000000 -0500 | |
115 | +++ linux-orig2/arch/sh/Kconfig 2004-10-22 11:32:47.169180310 -0500 | |
116 | @@ -748,3 +748,5 @@ | |
117 | source "crypto/Kconfig" | |
118 | ||
119 | source "lib/Kconfig" | |
c1c6733f | 120 | + |
bb1d8b11 AM |
121 | +source "cluster/Kconfig" |
122 | diff -urN linux-orig/arch/sparc/Kconfig linux-orig2/arch/sparc/Kconfig | |
123 | --- linux-orig/arch/sparc/Kconfig 2004-10-18 16:53:05.000000000 -0500 | |
124 | +++ linux-orig2/arch/sparc/Kconfig 2004-10-22 11:33:06.891871022 -0500 | |
125 | @@ -386,3 +386,5 @@ | |
126 | source "crypto/Kconfig" | |
127 | ||
128 | source "lib/Kconfig" | |
c1c6733f | 129 | + |
bb1d8b11 AM |
130 | +source "cluster/Kconfig" |
131 | diff -urN linux-orig/arch/sparc64/Kconfig linux-orig2/arch/sparc64/Kconfig | |
132 | --- linux-orig/arch/sparc64/Kconfig 2004-10-18 16:55:06.000000000 -0500 | |
133 | +++ linux-orig2/arch/sparc64/Kconfig 2004-10-22 11:33:19.290676599 -0500 | |
134 | @@ -613,3 +613,5 @@ | |
135 | source "crypto/Kconfig" | |
136 | ||
137 | source "lib/Kconfig" | |
c1c6733f | 138 | + |
bb1d8b11 AM |
139 | +source "cluster/Kconfig" |
140 | diff -urN linux-orig/arch/um/Kconfig linux-orig2/arch/um/Kconfig | |
141 | --- linux-orig/arch/um/Kconfig 2004-10-18 16:54:08.000000000 -0500 | |
142 | +++ linux-orig2/arch/um/Kconfig 2004-10-22 11:29:33.564217823 -0500 | |
143 | @@ -225,6 +225,8 @@ | |
144 | ||
145 | source "lib/Kconfig" | |
146 | ||
147 | +source "cluster/Kconfig" | |
c1c6733f | 148 | + |
bb1d8b11 AM |
149 | menu "SCSI support" |
150 | depends on BROKEN | |
151 | ||
152 | diff -urN linux-orig/arch/x86_64/Kconfig linux-orig2/arch/x86_64/Kconfig | |
153 | --- linux-orig/arch/x86_64/Kconfig 2004-10-18 16:54:55.000000000 -0500 | |
154 | +++ linux-orig2/arch/x86_64/Kconfig 2004-10-22 11:33:37.130396876 -0500 | |
155 | @@ -424,3 +424,5 @@ | |
156 | source "crypto/Kconfig" | |
157 | ||
158 | source "lib/Kconfig" | |
c1c6733f | 159 | + |
bb1d8b11 AM |
160 | +source "cluster/Kconfig" |
161 | diff -urN linux-orig/cluster/cman/Makefile linux-orig2/cluster/cman/Makefile | |
162 | --- linux-orig/cluster/cman/Makefile 1969-12-31 18:00:00.000000000 -0600 | |
163 | +++ linux-orig2/cluster/cman/Makefile 2004-10-22 11:29:33.566217791 -0500 | |
164 | @@ -0,0 +1,6 @@ | |
165 | +cman-objs := cnxman.o config.o membership.o proc.o\ | |
166 | + sm_barrier.o sm_control.o sm_daemon.o sm_joinleave.o\ | |
167 | + sm_membership.o sm_message.o sm_misc.o sm_recover.o sm_services.o \ | |
168 | + sm_user.o | |
c1c6733f | 169 | + |
bb1d8b11 AM |
170 | +obj-$(CONFIG_CLUSTER) := cman.o |
171 | diff -urN linux-orig/cluster/Kconfig linux-orig2/cluster/Kconfig | |
172 | --- linux-orig/cluster/Kconfig 1969-12-31 18:00:00.000000000 -0600 | |
173 | +++ linux-orig2/cluster/Kconfig 2004-10-22 11:29:33.565217807 -0500 | |
174 | @@ -0,0 +1,13 @@ | |
175 | +menu "Cluster Support" | |
c1c6733f | 176 | + |
bb1d8b11 AM |
177 | +config CLUSTER |
178 | + tristate "Cluster support" | |
179 | + ---help--- | |
180 | + Enable clustering support. This is not the high-performance clustering | |
181 | + made famous by beowulf. It is a high-availability cluster often using | |
182 | + shared storage. | |
183 | + The cluster manager is the heart(beat) of the cluster system. It is | |
184 | + needed by all the other components. It provides membership services | |
185 | + for those other subsystems. | |
c1c6733f | 186 | + |
bb1d8b11 AM |
187 | +endmenu |
188 | diff -urN linux-orig/cluster/Makefile linux-orig2/cluster/Makefile | |
189 | --- linux-orig/cluster/Makefile 1969-12-31 18:00:00.000000000 -0600 | |
190 | +++ linux-orig2/cluster/Makefile 2004-10-22 11:29:33.566217791 -0500 | |
191 | @@ -0,0 +1,3 @@ | |
192 | +obj-y := nocluster.o | |
c1c6733f | 193 | + |
bb1d8b11 AM |
194 | +obj-$(CONFIG_CLUSTER) += cman/ |
195 | diff -urN linux-orig/cluster/nocluster.c linux-orig2/cluster/nocluster.c | |
196 | --- linux-orig/cluster/nocluster.c 1969-12-31 18:00:00.000000000 -0600 | |
197 | +++ linux-orig2/cluster/nocluster.c 2004-10-22 11:29:33.567217776 -0500 | |
198 | @@ -0,0 +1,20 @@ | |
199 | +/* | |
200 | + * cluster/nocluster.c | |
201 | + * | |
202 | + * Copy from net/nonet.c | |
203 | + * Dummy functions to allow us to configure cluster support entirely | |
204 | + * out of the kernel. | |
205 | + * | |
206 | + * Distributed under the terms of the GNU GPL version 2. | |
207 | + * Copyright (c) Matthew Wilcox 2003 | |
208 | + */ | |
c1c6733f | 209 | + |
bb1d8b11 AM |
210 | +#include <linux/module.h> |
211 | +#include <linux/errno.h> | |
212 | +#include <linux/fs.h> | |
213 | +#include <linux/init.h> | |
214 | +#include <linux/kernel.h> | |
c1c6733f | 215 | + |
bb1d8b11 AM |
216 | +void __init nocluster_init(void) |
217 | +{ | |
218 | +} | |
219 | diff -urN linux-orig/Makefile linux-orig2/Makefile | |
220 | --- linux-orig/Makefile 2004-10-18 16:54:38.000000000 -0500 | |
221 | +++ linux-orig2/Makefile 2004-10-22 11:29:33.507218717 -0500 | |
222 | @@ -445,7 +445,7 @@ | |
223 | ||
224 | # Objects we will link into vmlinux / subdirs we need to visit | |
225 | init-y := init/ | |
226 | -drivers-y := drivers/ sound/ | |
227 | +drivers-y := drivers/ sound/ cluster/ | |
228 | net-y := net/ | |
229 | libs-y := lib/ | |
230 | core-y := usr/ | |
231 | diff -urN linux-orig/cluster/cman/cnxman-private.h linux-patched/cluster/cman/cnxman-private.h | |
232 | --- linux-orig/cluster/cman/cnxman-private.h 1970-01-01 07:30:00.000000000 +0730 | |
233 | +++ linux-patched/cluster/cman/cnxman-private.h 2004-11-03 11:37:37.000000000 +0800 | |
234 | @@ -0,0 +1,432 @@ | |
235 | +/****************************************************************************** | |
236 | +******************************************************************************* | |
237 | +** | |
238 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
239 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
240 | +** | |
241 | +** This copyrighted material is made available to anyone wishing to use, | |
242 | +** modify, copy, or redistribute it subject to the terms and conditions | |
243 | +** of the GNU General Public License v.2. | |
244 | +** | |
245 | +******************************************************************************* | |
246 | +******************************************************************************/ | |
c1c6733f | 247 | + |
bb1d8b11 AM |
248 | +#ifndef __CNXMAN_PRIVATE_H |
249 | +#define __CNXMAN_PRIVATE_H | |
c1c6733f | 250 | + |
bb1d8b11 AM |
251 | +/* Version triplet */ |
252 | +#define CNXMAN_MAJOR_VERSION 3 | |
253 | +#define CNXMAN_MINOR_VERSION 0 | |
254 | +#define CNXMAN_PATCH_VERSION 1 | |
c1c6733f | 255 | + |
bb1d8b11 AM |
256 | +#define MAX_RETRIES 3 /* Maximum number of send retries */ |
257 | +#define CAP_CLUSTER CAP_SYS_ADMIN /* Capability needed to manage the | |
258 | + * cluster */ | |
259 | +#ifdef __KERNEL__ | |
c1c6733f | 260 | + |
bb1d8b11 AM |
261 | +/* How we announce ourself in console events */ |
262 | +#define CMAN_NAME "CMAN" | |
c1c6733f | 263 | + |
bb1d8b11 AM |
264 | +/* One of these per AF_CLUSTER socket */ |
265 | +struct cluster_sock { | |
266 | + /* WARNING: sk has to be the first member */ | |
267 | + struct sock sk; | |
c1c6733f | 268 | + |
bb1d8b11 AM |
269 | + unsigned char port; /* Bound port or zero */ |
270 | + int (*kernel_callback) (char *, int, char *, int, unsigned int); | |
271 | + void *service_data; | |
272 | +}; | |
c1c6733f | 273 | + |
bb1d8b11 | 274 | +#define cluster_sk(__sk) ((struct cluster_sock *)__sk) |
c1c6733f | 275 | + |
bb1d8b11 AM |
276 | +/* We have one of these for each socket we use for communications */ |
277 | +struct cl_comms_socket { | |
278 | + struct socket *sock; | |
279 | + int broadcast; /* This is a broadcast socket */ | |
280 | + int recv_only; /* This is the unicast receive end of a | |
281 | + * multicast socket */ | |
282 | + struct sockaddr_in6 saddr; /* Socket address, contains the sockaddr for | |
283 | + * the remote end(s) */ | |
284 | + int addr_len; /* Length of above */ | |
285 | + int number; /* Internal socket number, used to cycle around | |
286 | + * sockets in case of network errors */ | |
287 | + struct file *file; /* file pointer for user-passed in sockets */ | |
c1c6733f | 288 | + |
bb1d8b11 | 289 | + wait_queue_t wait; |
c1c6733f | 290 | + |
bb1d8b11 AM |
291 | + /* The socket list */ |
292 | + struct list_head list; | |
c783755a | 293 | + |
bb1d8b11 AM |
294 | + /* On here when it has something to say */ |
295 | + struct list_head active_list; | |
296 | + unsigned long active; | |
297 | +}; | |
c783755a | 298 | + |
bb1d8b11 AM |
299 | +/* A client socket. We keep a list of these so we can notify clients of cluster |
300 | + * events */ | |
301 | +struct cl_client_socket { | |
302 | + struct socket *sock; | |
303 | + struct list_head list; | |
304 | +}; | |
c1c6733f | 305 | + |
bb1d8b11 AM |
306 | +/* This structure is tacked onto the start of a cluster message packet for our |
307 | + * own nefarious purposes. */ | |
308 | +struct cl_protheader { | |
309 | + unsigned char tgtport; /* Target port number */ | |
310 | + unsigned char srcport; /* Source (originationg) port number */ | |
311 | + unsigned short seq; /* Packet sequence number, little-endian */ | |
312 | + unsigned short ack; /* Inline ACK */ | |
313 | + unsigned short cluster; /* Our cluster number, little-endian */ | |
314 | + unsigned int flags; | |
315 | + int srcid; /* Node ID of the sender */ | |
316 | + int tgtid; /* Node ID of the target or 0 for multicast | |
317 | + * messages */ | |
318 | +}; | |
c1c6733f | 319 | + |
bb1d8b11 AM |
320 | +/* A cluster internal protocol message - port number 0 */ |
321 | +struct cl_protmsg { | |
322 | + struct cl_protheader header; | |
323 | + unsigned char cmd; | |
324 | +}; | |
c1c6733f | 325 | + |
bb1d8b11 AM |
326 | +/* A Cluster ACK message */ |
327 | +struct cl_ackmsg { | |
328 | + struct cl_protheader header; | |
329 | + unsigned char cmd; /* Always CLUSTER_CMD_ACK */ | |
330 | + unsigned char remport; /* Remote port number the original message was | |
331 | + * for */ | |
332 | + unsigned char aflags; /* ACK flags 0=OK, 1=No listener */ | |
333 | + unsigned char pad; | |
334 | +}; | |
c1c6733f | 335 | + |
bb1d8b11 AM |
336 | +/* A Cluster LISTENREQ/LISTENRESP message */ |
337 | +struct cl_listenmsg { | |
338 | + unsigned char cmd; /* CLUSTER_CMD_LISTENRESP/REQ */ | |
339 | + unsigned char target_port; /* Port to probe */ | |
340 | + unsigned char listening; /* Always 0 for LISTENREQ */ | |
341 | + unsigned char pad; | |
342 | + unsigned short tag; /* PID of remote waiting process */ | |
343 | +}; | |
c1c6733f | 344 | + |
bb1d8b11 AM |
345 | +/* A Cluster PORTCLOSED message */ |
346 | +struct cl_closemsg { | |
347 | + unsigned char cmd; /* CLUSTER_CMD_PORTCLOSED */ | |
348 | + unsigned char port; | |
349 | +}; | |
c1c6733f | 350 | + |
bb1d8b11 AM |
351 | +/* Structure of a newly dead node, passed from cnxman to kmembershipd */ |
352 | +struct cl_new_dead_node { | |
353 | + struct list_head list; | |
c1c6733f | 354 | + struct cluster_node *node; |
bb1d8b11 | 355 | +}; |
c1c6733f | 356 | + |
bb1d8b11 AM |
357 | +/* Subcommands for BARRIER message */ |
358 | +#define BARRIER_REGISTER 1 | |
359 | +#define BARRIER_CHANGE 2 | |
360 | +#define BARRIER_WAIT 4 | |
361 | +#define BARRIER_COMPLETE 5 | |
c1c6733f | 362 | + |
bb1d8b11 AM |
363 | +/* A Cluster BARRIER message */ |
364 | +struct cl_barriermsg { | |
365 | + unsigned char cmd; /* CLUSTER_CMD_BARRIER */ | |
366 | + unsigned char subcmd; /* BARRIER sub command */ | |
367 | + unsigned short pad; | |
368 | + unsigned int flags; | |
369 | + unsigned int nodes; | |
370 | + char name[MAX_BARRIER_NAME_LEN]; | |
371 | +}; | |
c1c6733f | 372 | + |
bb1d8b11 AM |
373 | +/* Membership services messages, the cl_protheader is added transparently */ |
374 | +struct cl_mem_hello_msg { | |
375 | + unsigned char cmd; | |
376 | + unsigned char flags; | |
377 | + unsigned short members; /* Number of nodes in the cluster, | |
378 | + * little-endian */ | |
379 | + unsigned int generation; /* Current cluster generation number */ | |
380 | +}; | |
c1c6733f | 381 | + |
bb1d8b11 AM |
382 | +struct cl_mem_endtrans_msg { |
383 | + unsigned char cmd; | |
384 | + unsigned char pad1; | |
385 | + unsigned short pad2; | |
386 | + unsigned int quorum; | |
387 | + unsigned int total_votes; | |
388 | + unsigned int generation; /* Current cluster generation number */ | |
389 | + unsigned int new_node_id; /* If reason is a new node joining */ | |
390 | +}; | |
c1c6733f | 391 | + |
bb1d8b11 AM |
392 | +/* ACK types for JOINACK message */ |
393 | +#define JOINACK_TYPE_OK 1 /* You can join */ | |
394 | +#define JOINACK_TYPE_NAK 2 /* You can NOT join */ | |
395 | +#define JOINACK_TYPE_WAIT 3 /* Wait a bit longer - cluster is in transition | |
396 | + * already */ | |
c783755a | 397 | + |
bb1d8b11 AM |
398 | +struct cl_mem_joinack_msg { |
399 | + unsigned char cmd; | |
400 | + unsigned char acktype; | |
401 | +}; | |
c1c6733f | 402 | + |
bb1d8b11 AM |
403 | +/* This is used by JOINREQ message */ |
404 | +struct cl_mem_join_msg { | |
405 | + unsigned char cmd; | |
406 | + unsigned char votes; | |
407 | + unsigned short num_addr; /* Number of addresses for this node */ | |
408 | + unsigned int expected_votes; | |
409 | + unsigned int nodeid; /* node ID we want */ | |
410 | + unsigned int major_version; /* Not backwards compatible */ | |
411 | + unsigned int minor_version; /* Backwards compatible */ | |
412 | + unsigned int patch_version; /* Backwards/forwards compatible */ | |
413 | + unsigned int config_version; | |
414 | + unsigned int addr_len; /* length of node addresses */ | |
415 | + char clustername[16]; | |
416 | + /* Followed by <num_addr> addresses of `address_length` bytes and a | |
417 | + * NUL-terminated node name */ | |
418 | +}; | |
c1c6733f | 419 | + |
bb1d8b11 AM |
420 | +/* State transition start reasons: */ |
421 | +#define TRANS_NEWNODE 1 /* A new node is joining the cluster */ | |
422 | +#define TRANS_REMNODE 2 /* a node has left the cluster */ | |
423 | +#define TRANS_ANOTHERREMNODE 3 /* A node left the cluster while we were in | |
424 | + * transition */ | |
425 | +#define TRANS_NEWMASTER 4 /* We have had an election and I am the new | |
426 | + * master */ | |
427 | +#define TRANS_CHECK 5 /* A consistency check was called for */ | |
428 | +#define TRANS_RESTART 6 /* Transition restarted because of a previous | |
429 | + * timeout */ | |
430 | +#define TRANS_DEADMASTER 7 /* The master died during transition and I have | |
431 | + * taken over */ | |
c1c6733f | 432 | + |
bb1d8b11 AM |
433 | +/* This is used to start a state transition */ |
434 | +struct cl_mem_starttrans_msg { | |
435 | + unsigned char cmd; | |
436 | + unsigned char reason; /* Why a start transition was started - see | |
437 | + * above */ | |
438 | + unsigned char flags; | |
439 | + unsigned char votes; | |
440 | + unsigned int expected_votes; | |
441 | + unsigned int generation; /* Incremented for each STARTTRANS sent | |
442 | + */ | |
443 | + int nodeid; /* Node to be removed */ | |
444 | + unsigned short num_addrs; | |
445 | + /* If reason == TRANS_NEWNODE: Followed by <num_addr> addresses of | |
446 | + * `address_length` bytes and a NUL-terminated node name */ | |
447 | +}; | |
c1c6733f | 448 | + |
bb1d8b11 AM |
449 | +struct cl_mem_startack_msg { |
450 | + unsigned char cmd; | |
451 | + unsigned char reason; | |
452 | + unsigned short pad; | |
453 | + unsigned int generation; | |
454 | + unsigned int node_id; /* node_id we think new node should have */ | |
455 | + unsigned int highest_node_id; /* highest node_id on this system */ | |
456 | +}; | |
c1c6733f | 457 | + |
bb1d8b11 AM |
458 | +/* Reconfigure a cluster parameter */ |
459 | +struct cl_mem_reconfig_msg { | |
460 | + unsigned char cmd; | |
461 | + unsigned char param; | |
462 | + unsigned short pad; | |
463 | + unsigned int value; | |
464 | +}; | |
c1c6733f | 465 | + |
bb1d8b11 AM |
466 | +/* Structure containing information about an outstanding listen request */ |
467 | +struct cl_waiting_listen_request { | |
468 | + wait_queue_head_t waitq; | |
469 | + int result; | |
470 | + int waiting; | |
471 | + unsigned short tag; | |
472 | + int nodeid; | |
473 | + struct list_head list; | |
474 | +}; | |
c1c6733f | 475 | + |
bb1d8b11 AM |
476 | +/* Messages from membership services */ |
477 | +#define CLUSTER_MEM_JOINCONF 1 | |
478 | +#define CLUSTER_MEM_JOINREQ 2 | |
479 | +#define CLUSTER_MEM_LEAVE 3 | |
480 | +#define CLUSTER_MEM_HELLO 4 | |
481 | +#define CLUSTER_MEM_KILL 5 | |
482 | +#define CLUSTER_MEM_JOINACK 6 | |
483 | +#define CLUSTER_MEM_ENDTRANS 7 | |
484 | +#define CLUSTER_MEM_RECONFIG 8 | |
485 | +#define CLUSTER_MEM_MASTERVIEW 9 | |
486 | +#define CLUSTER_MEM_STARTTRANS 10 | |
487 | +#define CLUSTER_MEM_JOINREJ 11 | |
488 | +#define CLUSTER_MEM_VIEWACK 12 | |
489 | +#define CLUSTER_MEM_STARTACK 13 | |
490 | +#define CLUSTER_MEM_TRANSITION 14 | |
491 | +#define CLUSTER_MEM_NEWCLUSTER 15 | |
492 | +#define CLUSTER_MEM_CONFACK 16 | |
493 | +#define CLUSTER_MEM_NOMINATE 17 | |
c1c6733f | 494 | + |
bb1d8b11 AM |
495 | +/* Flags in the HELLO message */ |
496 | +#define HELLO_FLAG_MASTER 1 | |
497 | +#define HELLO_FLAG_QUORATE 2 | |
c1c6733f | 498 | + |
bb1d8b11 AM |
499 | +/* Parameters for RECONFIG command */ |
500 | +#define RECONFIG_PARAM_EXPECTED_VOTES 1 | |
501 | +#define RECONFIG_PARAM_NODE_VOTES 2 | |
502 | +#define RECONFIG_PARAM_CONFIG_VERSION 3 | |
c1c6733f | 503 | + |
bb1d8b11 AM |
504 | +/* Data associated with an outgoing socket */ |
505 | +struct cl_socket { | |
506 | + struct file *file; /* The real file */ | |
507 | + struct socket *socket; /* The real sock */ | |
508 | + int num_nodes; /* On this link */ | |
509 | + int retransmit_count; | |
510 | +}; | |
b7b72b66 | 511 | + |
bb1d8b11 AM |
512 | +/* There's one of these for each node in the cluster */ |
513 | +struct cluster_node { | |
514 | + struct list_head list; | |
515 | + char *name; /* Node/host name of node */ | |
516 | + struct list_head addr_list; | |
517 | + int us; /* This node is us */ | |
518 | + unsigned int node_id; /* Unique node ID */ | |
519 | + nodestate_t state; | |
520 | + unsigned short last_seq_recv; | |
521 | + unsigned short last_seq_acked; | |
522 | + unsigned short last_seq_sent; | |
523 | + unsigned int votes; | |
524 | + unsigned int expected_votes; | |
525 | + unsigned int leave_reason; | |
526 | + unsigned int incarnation; /* Incremented each time a node joins | |
527 | + * the cluster */ | |
528 | + unsigned long last_hello; /* Jiffies */ | |
529 | + struct timeval join_time; | |
530 | +}; | |
b7b72b66 | 531 | + |
bb1d8b11 AM |
532 | +/* This is how we keep a list of user processes that are listening for cluster |
533 | + * membership events */ | |
534 | +struct notify_struct { | |
535 | + struct list_head list; | |
536 | + pid_t pid; | |
537 | + int signal; | |
538 | +}; | |
b7b72b66 | 539 | + |
bb1d8b11 AM |
540 | +/* This is how we keep a list of kernel callbacks that are registered for |
541 | + * cluster membership events */ | |
542 | +struct kernel_notify_struct { | |
543 | + struct list_head list; | |
544 | + void (*callback) (kcl_callback_reason, long arg); | |
545 | +}; | |
c1c6733f | 546 | + |
bb1d8b11 AM |
547 | +/* A message waiting to be sent */ |
548 | +struct queued_message { | |
549 | + struct list_head list; | |
c1c6733f | 550 | + |
bb1d8b11 AM |
551 | + struct socket *socket; |
552 | + struct sockaddr_cl addr; | |
553 | + int addr_len; | |
554 | + int msg_len; | |
555 | + unsigned char port; | |
556 | + unsigned int flags; | |
557 | + char msg_buffer[MAX_CLUSTER_MESSAGE]; | |
558 | +}; | |
c1c6733f | 559 | + |
bb1d8b11 AM |
560 | +/* A barrier */ |
561 | +struct cl_barrier { | |
562 | + struct list_head list; | |
c1c6733f | 563 | + |
bb1d8b11 AM |
564 | + char name[MAX_BARRIER_NAME_LEN]; |
565 | + unsigned int flags; | |
566 | + enum { BARRIER_STATE_WAITING, BARRIER_STATE_INACTIVE, | |
567 | + BARRIER_STATE_COMPLETE } state; | |
568 | + unsigned int expected_nodes; | |
569 | + unsigned int registered_nodes; | |
570 | + atomic_t got_nodes; | |
571 | + atomic_t completed_nodes; | |
572 | + unsigned int inuse; | |
573 | + unsigned int waitsent; | |
574 | + unsigned int phase; /* Completion phase */ | |
575 | + unsigned int endreason; /* Reason we were woken, usually 0 */ | |
576 | + unsigned long timeout; /* In seconds */ | |
c1c6733f | 577 | + |
bb1d8b11 AM |
578 | + void (*callback) (char *name, int status); |
579 | + wait_queue_head_t waitq; | |
580 | + struct semaphore lock; /* To synch with cnxman messages */ | |
581 | + spinlock_t phase2_spinlock; /* Need to synchronise with timer | |
582 | + * interrupts */ | |
583 | + struct timer_list timer; | |
584 | +}; | |
c1c6733f | 585 | + |
bb1d8b11 AM |
586 | +/* Cluster protocol commands sent to port 0 */ |
587 | +#define CLUSTER_CMD_ACK 1 | |
588 | +#define CLUSTER_CMD_LISTENREQ 2 | |
589 | +#define CLUSTER_CMD_LISTENRESP 3 | |
590 | +#define CLUSTER_CMD_PORTCLOSED 4 | |
591 | +#define CLUSTER_CMD_BARRIER 5 | |
c1c6733f | 592 | + |
bb1d8b11 AM |
593 | +extern struct cluster_node *find_node_by_addr(unsigned char *addr, |
594 | + int addr_len); | |
595 | +extern struct cluster_node *find_node_by_nodeid(unsigned int id); | |
596 | +extern struct cluster_node *find_node_by_name(char *name); | |
597 | +extern void set_quorate(int); | |
598 | +extern void notify_kernel_listeners(kcl_callback_reason reason, long arg); | |
599 | +extern void notify_listeners(void); | |
600 | +extern void free_nodeid_array(void); | |
601 | +extern int send_reconfigure(int param, unsigned int value); | |
602 | +extern int calculate_quorum(int, int, int *); | |
603 | +extern void recalculate_quorum(int); | |
604 | +extern int send_leave(unsigned char); | |
605 | +extern int get_quorum(void); | |
606 | +extern void set_votes(int, int); | |
607 | +extern void kcl_wait_for_all_acks(void); | |
608 | +extern char *membership_state(char *, int); | |
609 | +extern char *leave_string(int reason); | |
610 | +extern void a_node_just_died(struct cluster_node *node); | |
611 | +extern void check_barrier_returns(void); | |
612 | +extern int in_transition(void); | |
613 | +extern void get_local_addresses(struct cluster_node *node); | |
614 | +extern int add_node_address(struct cluster_node *node, unsigned char *addr, int len); | |
615 | +extern void create_proc_entries(void); | |
616 | +extern void cleanup_proc_entries(void); | |
617 | +extern unsigned int get_highest_nodeid(void); | |
618 | +extern int allocate_nodeid_array(void); | |
619 | +extern void queue_oob_skb(struct socket *sock, int cmd); | |
620 | +extern int new_temp_nodeid(char *addr, int addrlen); | |
621 | +extern int get_addr_from_temp_nodeid(int nodeid, char *addr, int *addrlen); | |
622 | +extern void purge_temp_nodeids(void); | |
623 | +extern inline char *print_addr(unsigned char *addr, int len, char *buf) | |
624 | +{ | |
625 | + int i; | |
626 | + int ptr = 0; | |
c1c6733f | 627 | + |
bb1d8b11 AM |
628 | + for (i = 0; i < len; i++) |
629 | + ptr += sprintf(buf + ptr, "%02x ", addr[i]); | |
c1c6733f | 630 | + |
bb1d8b11 | 631 | + return buf; |
c1c6733f AM |
632 | +} |
633 | + | |
bb1d8b11 | 634 | +#define MAX_ADDR_PRINTED_LEN (address_length*3 + 1) |
b7b72b66 | 635 | + |
bb1d8b11 AM |
636 | +/* Debug enabling macros. Sorry about the C++ comments but they're easier to |
637 | + * get rid of than C ones... */ | |
b7b72b66 | 638 | + |
bb1d8b11 AM |
639 | +// #define DEBUG_MEMB |
640 | +// #define DEBUG_COMMS | |
641 | +// #define DEBUG_BARRIER | |
b7b72b66 | 642 | + |
bb1d8b11 AM |
643 | +/* Debug macros */ |
644 | +#ifdef DEBUG_COMMS | |
645 | +#define P_COMMS(fmt, args...) printk(KERN_DEBUG "cman comms: " fmt, ## args) | |
646 | +#else | |
647 | +#define P_COMMS(fmt, args...) | |
648 | +#endif | |
b7b72b66 | 649 | + |
bb1d8b11 AM |
650 | +#ifdef DEBUG_BARRIER |
651 | +#define P_BARRIER(fmt, args...) printk(KERN_DEBUG "cman barrier: " fmt, ## args) | |
652 | +#else | |
653 | +#define P_BARRIER(fmt, args...) | |
654 | +#endif | |
c783755a | 655 | + |
bb1d8b11 AM |
656 | +#ifdef DEBUG_MEMB |
657 | +#define P_MEMB(fmt, args...) printk(KERN_DEBUG "cman memb: " fmt, ## args) | |
658 | +#define C_MEMB(fmt, args...) printk(fmt, ## args) | |
659 | +#else | |
660 | +#define P_MEMB(fmt, args...) | |
661 | +#define C_MEMB(fmt, args...) | |
662 | +#endif | |
c783755a | 663 | + |
bb1d8b11 | 664 | +#endif /* __KERNEL */ |
c783755a | 665 | + |
bb1d8b11 AM |
666 | +#endif |
667 | diff -urN linux-orig/cluster/cman/cnxman.c linux-patched/cluster/cman/cnxman.c | |
668 | --- linux-orig/cluster/cman/cnxman.c 1970-01-01 07:30:00.000000000 +0730 | |
669 | +++ linux-patched/cluster/cman/cnxman.c 2004-11-03 11:37:37.000000000 +0800 | |
670 | @@ -0,0 +1,4214 @@ | |
671 | +/****************************************************************************** | |
672 | +******************************************************************************* | |
673 | +** | |
674 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
675 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
676 | +** | |
677 | +** This copyrighted material is made available to anyone wishing to use, | |
678 | +** modify, copy, or redistribute it subject to the terms and conditions | |
679 | +** of the GNU General Public License v.2. | |
680 | +** | |
681 | +******************************************************************************* | |
682 | +******************************************************************************/ | |
c783755a | 683 | + |
bb1d8b11 AM |
684 | +#define EXPORT_SYMTAB |
685 | +#include <linux/init.h> | |
686 | +#include <linux/socket.h> | |
687 | +#include <linux/kernel.h> | |
688 | +#include <linux/sched.h> | |
689 | +#include <linux/file.h> | |
690 | +#include <linux/utsname.h> | |
691 | +#include <net/sock.h> | |
692 | +#include <linux/proc_fs.h> | |
693 | +#include <linux/poll.h> | |
694 | +#include <linux/module.h> | |
695 | +#include <linux/list.h> | |
696 | +#include <linux/uio.h> | |
697 | +#include <cluster/cnxman.h> | |
698 | +#include <cluster/service.h> | |
c783755a | 699 | + |
bb1d8b11 AM |
700 | +#include "cnxman-private.h" |
701 | +#include "sm_control.h" | |
702 | +#include "sm_user.h" | |
703 | +#include "config.h" | |
c783755a | 704 | + |
bb1d8b11 | 705 | +#define CMAN_RELEASE_NAME "<CVS>" |
c783755a | 706 | + |
bb1d8b11 AM |
707 | +static void process_incoming_packet(struct cl_comms_socket *csock, |
708 | + struct msghdr *msg, struct kvec *vec, int veclen, int len); | |
709 | +static int cl_sendack(struct cl_comms_socket *sock, unsigned short seq, | |
710 | + int addr_len, char *addr, unsigned char remport, | |
711 | + unsigned char flag); | |
712 | +static void send_listen_request(int nodeid, unsigned char port); | |
713 | +static void send_listen_response(struct cl_comms_socket *csock, int nodeid, | |
714 | + unsigned char port, unsigned short tag); | |
715 | +static void resend_last_message(void); | |
716 | +static void start_ack_timer(void); | |
717 | +static int send_queued_message(struct queued_message *qmsg); | |
718 | +static void send_port_close_oob(unsigned char port); | |
719 | +static void post_close_oob(unsigned char port, int nodeid); | |
720 | +static void process_barrier_msg(struct cl_barriermsg *msg, | |
721 | + struct cluster_node *node); | |
722 | +static struct cl_barrier *find_barrier(char *name); | |
723 | +static void node_shutdown(void); | |
724 | +static void node_cleanup(void); | |
725 | +static int send_or_queue_message(struct socket *sock, void *buf, int len, struct sockaddr_cl *caddr, | |
726 | + unsigned int flags); | |
727 | +static struct cl_comms_socket *get_next_interface(struct cl_comms_socket *cur); | |
728 | +static void check_for_unacked_nodes(void); | |
729 | +static void free_cluster_sockets(void); | |
730 | +static uint16_t generate_cluster_id(char *name); | |
731 | +static int is_valid_temp_nodeid(int nodeid); | |
c783755a | 732 | + |
bb1d8b11 AM |
733 | +extern int start_membership_services(pid_t); |
734 | +extern int kcl_leave_cluster(int remove); | |
735 | +extern int send_kill(int nodeid); | |
c783755a | 736 | + |
bb1d8b11 AM |
737 | +static struct proto_ops cl_proto_ops; |
738 | +static struct sock *master_sock; | |
739 | +static kmem_cache_t *cluster_sk_cachep; | |
c783755a | 740 | + |
bb1d8b11 AM |
741 | +/* Pointer to the pseudo node that maintains quorum in a 2node system */ |
742 | +struct cluster_node *quorum_device = NULL; | |
c783755a | 743 | + |
bb1d8b11 AM |
744 | +/* Array of "ports" allocated. This is just a list of pointers to the sock that |
745 | + * has this port bound. Speed is a major issue here so 1-2K of allocated | |
746 | + * storage is worth sacrificing. Port 0 is reserved for protocol messages */ | |
747 | +static struct sock *port_array[256]; | |
748 | +static struct semaphore port_array_lock; | |
c783755a | 749 | + |
bb1d8b11 AM |
750 | +/* Our cluster name & number */ |
751 | +uint16_t cluster_id; | |
752 | +char cluster_name[MAX_CLUSTER_NAME_LEN+1]; | |
c783755a | 753 | + |
bb1d8b11 AM |
754 | +/* Two-node mode: causes cluster to remain quorate if one of two nodes fails. |
755 | + * No more than two nodes are permitted to join the cluster. */ | |
756 | +unsigned short two_node; | |
c783755a | 757 | + |
bb1d8b11 AM |
758 | +/* Cluster configuration version that must be the same among members. */ |
759 | +unsigned int config_version; | |
c783755a | 760 | + |
bb1d8b11 AM |
761 | +/* Reference counting for cluster applications */ |
762 | +atomic_t use_count; | |
c783755a | 763 | + |
bb1d8b11 AM |
764 | +/* Length of sockaddr address for our comms protocol */ |
765 | +unsigned int address_length; | |
c1c6733f | 766 | + |
bb1d8b11 AM |
767 | +/* Message sending */ |
768 | +static unsigned short cur_seq; /* Last message sent */ | |
769 | +static unsigned int ack_count; /* Number of acks received for message | |
770 | + * 'cur_seq' */ | |
771 | +static unsigned int acks_expected; /* Number of acks we expect to receive */ | |
772 | +static struct semaphore send_lock; | |
773 | +static struct timer_list ack_timer; | |
c1c6733f | 774 | + |
bb1d8b11 AM |
775 | +/* Saved packet information in case we need to resend it */ |
776 | +static char saved_msg_buffer[MAX_CLUSTER_MESSAGE]; | |
777 | +static int saved_msg_len; | |
778 | +static int retry_count; | |
c1c6733f | 779 | + |
bb1d8b11 AM |
780 | +/* Task variables */ |
781 | +static pid_t kcluster_pid; | |
782 | +static pid_t membership_pid; | |
783 | +extern struct task_struct *membership_task; | |
784 | +extern int quit_threads; | |
c1c6733f | 785 | + |
bb1d8b11 | 786 | +wait_queue_head_t cnxman_waitq; |
c1c6733f | 787 | + |
bb1d8b11 AM |
788 | +/* Variables owned by membership services */ |
789 | +extern int cluster_members; | |
790 | +extern struct list_head cluster_members_list; | |
791 | +extern struct semaphore cluster_members_lock; | |
792 | +extern int we_are_a_cluster_member; | |
793 | +extern int cluster_is_quorate; | |
794 | +extern struct cluster_node *us; | |
795 | +extern struct list_head new_dead_node_list; | |
796 | +extern struct semaphore new_dead_node_lock; | |
797 | +extern char nodename[]; | |
798 | +extern int wanted_nodeid; | |
c1c6733f | 799 | + |
bb1d8b11 AM |
800 | +/* A list of processes listening for membership events */ |
801 | +static struct list_head event_listener_list; | |
802 | +static struct semaphore event_listener_lock; | |
c1c6733f | 803 | + |
bb1d8b11 AM |
804 | +/* A list of kernel callbacks listening for membership events */ |
805 | +static struct list_head kernel_listener_list; | |
806 | +static struct semaphore kernel_listener_lock; | |
c1c6733f | 807 | + |
bb1d8b11 AM |
808 | +/* A list of sockets we are listening on (and can transmit on...later) */ |
809 | +static struct list_head socket_list; | |
c1c6733f | 810 | + |
bb1d8b11 AM |
811 | +/* A list of all open cluster client sockets */ |
812 | +static struct list_head client_socket_list; | |
813 | +static struct semaphore client_socket_lock; | |
c1c6733f | 814 | + |
bb1d8b11 AM |
815 | +/* A list of all current barriers */ |
816 | +static struct list_head barrier_list; | |
817 | +static struct semaphore barrier_list_lock; | |
c1c6733f | 818 | + |
bb1d8b11 AM |
819 | +/* When a socket is read for reading it goes on this queue */ |
820 | +static spinlock_t active_socket_lock; | |
821 | +static struct list_head active_socket_list; | |
c1c6733f | 822 | + |
bb1d8b11 AM |
823 | +/* If the cnxman process is running and available for work */ |
824 | +atomic_t cnxman_running; | |
c1c6733f | 825 | + |
bb1d8b11 AM |
826 | +/* Fkags set by timers etc for the mainloop to detect and act upon */ |
827 | +static unsigned long mainloop_flags; | |
b7b72b66 | 828 | + |
bb1d8b11 AM |
829 | +#define ACK_TIMEOUT 1 |
830 | +#define RESEND_NEEDED 2 | |
c1c6733f | 831 | + |
bb1d8b11 AM |
832 | +/* A queue of messages waiting to be sent. If kcl_sendmsg is called outside of |
833 | + * process context then the messages get put in here */ | |
834 | +static struct list_head messages_list; | |
835 | +static struct semaphore messages_list_lock; | |
b7b72b66 | 836 | + |
bb1d8b11 | 837 | +static struct semaphore start_thread_sem; |
c1c6733f | 838 | + |
bb1d8b11 AM |
839 | +/* List of outstanding ISLISTENING requests */ |
840 | +static struct list_head listenreq_list; | |
841 | +static struct semaphore listenreq_lock; | |
c1c6733f | 842 | + |
bb1d8b11 AM |
843 | +/* Any sending requests wait on this queue if necessary (eg inquorate, waiting |
844 | + * ACK) */ | |
845 | +static DECLARE_WAIT_QUEUE_HEAD(socket_waitq); | |
c1c6733f | 846 | + |
bb1d8b11 AM |
847 | +/* Wait for thread to exit properly */ |
848 | +struct completion cluster_thread_comp; | |
849 | +struct completion member_thread_comp; | |
c1c6733f | 850 | + |
bb1d8b11 AM |
851 | +/* The resend delay to use, We increase this geometrically(word?) each time a |
852 | + * send is delayed. in deci-seconds */ | |
853 | +static int resend_delay = 1; | |
c1c6733f | 854 | + |
bb1d8b11 AM |
855 | +/* Highest numbered interface and the current default */ |
856 | +static int num_interfaces; | |
857 | +static struct cl_comms_socket *current_interface = NULL; | |
c1c6733f | 858 | + |
bb1d8b11 AM |
859 | +struct temp_node |
860 | +{ | |
861 | + int nodeid; | |
862 | + char addr[sizeof(struct sockaddr_in6)]; | |
863 | + int addrlen; | |
864 | + struct list_head list; | |
865 | +}; | |
866 | +static struct list_head tempnode_list; | |
867 | +static struct semaphore tempnode_lock; | |
868 | + | |
869 | + | |
870 | +/* This is what's squirrelled away in skb->cb */ | |
871 | +struct cb_info | |
872 | +{ | |
873 | + int orig_nodeid; | |
874 | + char orig_port; | |
875 | + char oob; | |
876 | +}; | |
c1c6733f | 877 | + |
c1c6733f | 878 | + |
bb1d8b11 AM |
879 | +/* Wake up any processes that are waiting to send. This is usually called when |
880 | + * all the ACKs have been gathered up or when a node has left the cluster | |
881 | + * unexpectedly and we reckon there are no more acks to collect */ | |
882 | +static void unjam(void) | |
883 | +{ | |
884 | + wake_up_interruptible(&socket_waitq); | |
885 | + wake_up_interruptible(&cnxman_waitq); | |
c1c6733f AM |
886 | +} |
887 | + | |
bb1d8b11 AM |
888 | +/* Used by the data_ready routine to locate a connection given the socket */ |
889 | +static inline struct cl_comms_socket *find_comms_by_sock(struct sock *sk) | |
c1c6733f | 890 | +{ |
bb1d8b11 | 891 | + struct list_head *conlist; |
c1c6733f | 892 | + |
bb1d8b11 AM |
893 | + list_for_each(conlist, &socket_list) { |
894 | + struct cl_comms_socket *clsock = | |
895 | + list_entry(conlist, struct cl_comms_socket, list); | |
896 | + if (clsock->sock->sk == sk) { | |
897 | + return clsock; | |
898 | + } | |
c1c6733f | 899 | + } |
c1c6733f AM |
900 | + return NULL; |
901 | +} | |
902 | + | |
bb1d8b11 AM |
903 | +/* Data available on socket */ |
904 | +static void cnxman_data_ready(struct sock *sk, int count_unused) | |
c1c6733f | 905 | +{ |
bb1d8b11 | 906 | + struct cl_comms_socket *clsock = find_comms_by_sock(sk); |
c1c6733f | 907 | + |
bb1d8b11 AM |
908 | + if (clsock == NULL) /* ASSERT ?? */ |
909 | + return; | |
c1c6733f | 910 | + |
bb1d8b11 AM |
911 | + /* If we're already on the list then don't do it again */ |
912 | + if (test_and_set_bit(1, &clsock->active)) | |
913 | + return; | |
c1c6733f | 914 | + |
bb1d8b11 AM |
915 | + spin_lock_irq(&active_socket_lock); |
916 | + list_add(&clsock->active_list, &active_socket_list); | |
917 | + spin_unlock_irq(&active_socket_lock); | |
c1c6733f | 918 | + |
bb1d8b11 AM |
919 | + wake_up_interruptible(&cnxman_waitq); |
920 | +} | |
c1c6733f | 921 | + |
bb1d8b11 AM |
922 | +static int receive_message(struct cl_comms_socket *csock, char *iobuf) |
923 | +{ | |
924 | + struct msghdr msg; | |
925 | + struct kvec vec; | |
926 | + struct sockaddr_in6 sin; | |
927 | + int len; | |
c1c6733f | 928 | + |
bb1d8b11 | 929 | + memset(&sin, 0, sizeof (sin)); |
c1c6733f | 930 | + |
bb1d8b11 AM |
931 | + msg.msg_control = NULL; |
932 | + msg.msg_controllen = 0; | |
933 | + msg.msg_name = &sin; | |
934 | + msg.msg_namelen = sizeof (sin); | |
935 | + msg.msg_flags = 0; | |
c1c6733f | 936 | + |
bb1d8b11 AM |
937 | + vec.iov_len = MAX_CLUSTER_MESSAGE; |
938 | + vec.iov_base = iobuf; | |
c1c6733f | 939 | + |
bb1d8b11 AM |
940 | + len = kernel_recvmsg(csock->sock, &msg, |
941 | + &vec, 1, MAX_CLUSTER_MESSAGE, MSG_DONTWAIT); | |
942 | + | |
943 | + vec.iov_base = iobuf; | |
944 | + | |
945 | + if (len > 0) { | |
946 | + if (len > MAX_CLUSTER_MESSAGE) { | |
947 | + printk(KERN_CRIT CMAN_NAME | |
948 | + ": %d byte message far too big\n", len); | |
949 | + return 0; | |
c1c6733f | 950 | + } |
bb1d8b11 | 951 | + process_incoming_packet(csock, &msg, &vec, 1, len); |
c1c6733f | 952 | + } |
bb1d8b11 AM |
953 | + else { |
954 | + if (len != -EAGAIN) | |
955 | + printk(KERN_CRIT CMAN_NAME ": recvmsg failed: %d\n", | |
956 | + len); | |
957 | + } | |
958 | + return len; | |
c1c6733f AM |
959 | +} |
960 | + | |
bb1d8b11 | 961 | +static int cluster_kthread(void *unused) |
c1c6733f | 962 | +{ |
bb1d8b11 AM |
963 | + int len; |
964 | + char *iobuf; | |
965 | + struct list_head *socklist; | |
966 | + struct cl_comms_socket *csock; | |
967 | + wait_queue_t cnxman_waitq_head; | |
968 | + sigset_t tmpsig; | |
c1c6733f | 969 | + |
bb1d8b11 | 970 | + daemonize("cman_comms"); |
c1c6733f | 971 | + |
bb1d8b11 AM |
972 | + /* Block everything but SIGKILL/SIGSTOP/SIGTERM */ |
973 | + siginitset(&tmpsig, SIGKILL | SIGSTOP | SIGTERM); | |
974 | + sigprocmask(SIG_BLOCK, &tmpsig, NULL); | |
c1c6733f | 975 | + |
bb1d8b11 AM |
976 | + /* This is the waitq we can wake the process up with */ |
977 | + init_waitqueue_head(&cnxman_waitq); | |
978 | + init_waitqueue_entry(&cnxman_waitq_head, current); | |
979 | + add_wait_queue(&cnxman_waitq, &cnxman_waitq_head); | |
c1c6733f | 980 | + |
bb1d8b11 | 981 | + set_user_nice(current, -6); |
c1c6733f | 982 | + |
bb1d8b11 AM |
983 | + /* Allow the sockets to start receiving */ |
984 | + list_for_each(socklist, &socket_list) { | |
985 | + csock = list_entry(socklist, struct cl_comms_socket, list); | |
c1c6733f | 986 | + |
bb1d8b11 | 987 | + clear_bit(1, &csock->active); |
c1c6733f AM |
988 | + } |
989 | + | |
bb1d8b11 AM |
990 | + iobuf = kmalloc(MAX_CLUSTER_MESSAGE, GFP_KERNEL); |
991 | + if (!iobuf) { | |
992 | + printk(KERN_CRIT CMAN_NAME | |
993 | + ": Cannot allocate receive buffer for cluster comms\n"); | |
994 | + return -1; | |
995 | + } | |
c1c6733f | 996 | + |
bb1d8b11 | 997 | + complete(&cluster_thread_comp); |
c1c6733f | 998 | + |
bb1d8b11 AM |
999 | + for (;;) { |
1000 | + struct list_head *temp; | |
c1c6733f | 1001 | + |
bb1d8b11 AM |
1002 | + /* Wait for activity on any of the sockets */ |
1003 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
c1c6733f | 1004 | + |
bb1d8b11 AM |
1005 | + if (list_empty(&active_socket_list)) |
1006 | + schedule(); | |
1007 | + set_task_state(current, TASK_RUNNING); | |
c1c6733f | 1008 | + |
bb1d8b11 AM |
1009 | + if (quit_threads) |
1010 | + break; | |
c1c6733f | 1011 | + |
bb1d8b11 AM |
1012 | + if (test_and_clear_bit(ACK_TIMEOUT, &mainloop_flags)) { |
1013 | + check_for_unacked_nodes(); | |
1014 | + } | |
c1c6733f | 1015 | + |
bb1d8b11 AM |
1016 | + /* Now receive any messages waiting for us */ |
1017 | + spin_lock_irq(&active_socket_lock); | |
1018 | + list_for_each_safe(socklist, temp, &active_socket_list) { | |
1019 | + csock = | |
1020 | + list_entry(socklist, struct cl_comms_socket, | |
1021 | + active_list); | |
c1c6733f | 1022 | + |
bb1d8b11 AM |
1023 | + list_del(&csock->active_list); |
1024 | + clear_bit(1, &csock->active); | |
c1c6733f | 1025 | + |
bb1d8b11 | 1026 | + spin_unlock_irq(&active_socket_lock); |
c1c6733f | 1027 | + |
bb1d8b11 AM |
1028 | + do { |
1029 | + len = receive_message(csock, iobuf); | |
1030 | + } | |
1031 | + while (len > 0); | |
c1c6733f | 1032 | + |
bb1d8b11 | 1033 | + spin_lock_irq(&active_socket_lock); |
c1c6733f | 1034 | + |
bb1d8b11 AM |
1035 | + if (len == 0) |
1036 | + break; /* EOF on socket */ | |
1037 | + } | |
1038 | + spin_unlock_irq(&active_socket_lock); | |
c1c6733f | 1039 | + |
bb1d8b11 AM |
1040 | + /* Resend any unacked messages */ |
1041 | + if (test_and_clear_bit(RESEND_NEEDED, &mainloop_flags) | |
1042 | + && acks_expected) { | |
1043 | + resend_last_message(); | |
1044 | + } | |
c1c6733f | 1045 | + |
bb1d8b11 AM |
1046 | + /* Send any queued messages */ |
1047 | + if (acks_expected == 0) { | |
1048 | + struct list_head *temp; | |
1049 | + struct list_head *msglist; | |
1050 | + | |
1051 | + down(&messages_list_lock); | |
1052 | + list_for_each_safe(msglist, temp, &messages_list) { | |
1053 | + struct queued_message *qmsg = | |
1054 | + list_entry(msglist, struct queued_message, | |
1055 | + list); | |
1056 | + int status = send_queued_message(qmsg); | |
1057 | + | |
1058 | + if (status >= 0) { | |
1059 | + /* Suceeded, remove it from the queue */ | |
1060 | + list_del(&qmsg->list); | |
1061 | + kfree(qmsg); | |
1062 | + } | |
1063 | + /* Did it fail horribly ?? */ | |
1064 | + if (status < 0 && status != -EAGAIN) { | |
1065 | + printk(KERN_INFO CMAN_NAME | |
1066 | + ": send_queued_message failed, error %d\n", | |
1067 | + status); | |
1068 | + list_del(&qmsg->list); | |
1069 | + kfree(qmsg); | |
1070 | + } | |
1071 | + break; /* Only send one message at a time */ | |
c1c6733f | 1072 | + } |
bb1d8b11 | 1073 | + up(&messages_list_lock); |
c1c6733f AM |
1074 | + } |
1075 | + | |
bb1d8b11 AM |
1076 | + if (signal_pending(current)) |
1077 | + break; | |
1078 | + } | |
1079 | + P_COMMS("closing down\n"); | |
c1c6733f | 1080 | + |
bb1d8b11 | 1081 | + quit_threads = 1; /* force other thread to die too */ |
c1c6733f | 1082 | + |
bb1d8b11 AM |
1083 | + /* Wait for membership thread to finish, that way any |
1084 | + LEAVE message will get sent. */ | |
1085 | + wake_up_process(membership_task); | |
1086 | + wait_for_completion(&member_thread_comp); | |
c1c6733f | 1087 | + |
bb1d8b11 | 1088 | + node_shutdown(); |
c1c6733f | 1089 | + |
bb1d8b11 AM |
1090 | + if (timer_pending(&ack_timer)) |
1091 | + del_timer(&ack_timer); | |
c1c6733f | 1092 | + |
bb1d8b11 AM |
1093 | + node_cleanup(); |
1094 | + kfree(iobuf); | |
c1c6733f | 1095 | + |
bb1d8b11 | 1096 | + complete(&cluster_thread_comp); |
c1c6733f AM |
1097 | + return 0; |
1098 | +} | |
1099 | + | |
bb1d8b11 | 1100 | +void notify_kernel_listeners(kcl_callback_reason reason, long arg) |
c1c6733f | 1101 | +{ |
bb1d8b11 AM |
1102 | + struct kernel_notify_struct *knotify; |
1103 | + struct list_head *proclist; | |
1104 | + | |
1105 | + down(&kernel_listener_lock); | |
1106 | + list_for_each(proclist, &kernel_listener_list) { | |
1107 | + knotify = | |
1108 | + list_entry(proclist, struct kernel_notify_struct, list); | |
1109 | + knotify->callback(reason, arg); | |
1110 | + } | |
1111 | + up(&kernel_listener_lock); | |
c1c6733f AM |
1112 | +} |
1113 | + | |
bb1d8b11 | 1114 | +static void check_for_unacked_nodes() |
c1c6733f | 1115 | +{ |
bb1d8b11 AM |
1116 | + struct list_head *nodelist; |
1117 | + struct list_head *temp; | |
1118 | + struct cluster_node *node; | |
1119 | + | |
1120 | + clear_bit(RESEND_NEEDED, &mainloop_flags); | |
1121 | + retry_count = 0; | |
1122 | + | |
1123 | + P_COMMS("Retry count exceeded -- looking for dead node\n"); | |
1124 | + | |
1125 | + /* Node did not ACK a message after <n> tries, remove it from the | |
1126 | + * cluster */ | |
1127 | + down(&cluster_members_lock); | |
1128 | + list_for_each_safe(nodelist, temp, &cluster_members_list) { | |
1129 | + node = list_entry(nodelist, struct cluster_node, list); | |
1130 | + | |
1131 | + P_COMMS("checking node %s: last_acked = %d, last_seq_sent = %d\n", | |
1132 | + node->name, node->last_seq_acked, node->last_seq_sent); | |
1133 | + if (node->state != NODESTATE_DEAD && | |
1134 | + node->last_seq_acked != node->last_seq_sent && !node->us) { | |
1135 | + printk(KERN_WARNING CMAN_NAME | |
1136 | + ": node %s is not responding - removing from the cluster\n", | |
1137 | + node->name); | |
1138 | + | |
1139 | + /* Drop this lock or we can deadlock with membership */ | |
1140 | + up(&cluster_members_lock); | |
1141 | + | |
1142 | + /* Start a state transition */ | |
1143 | + a_node_just_died(node); | |
1144 | + down(&cluster_members_lock); | |
1145 | + } | |
1146 | + } | |
1147 | + up(&cluster_members_lock); | |
1148 | + acks_expected = ack_count = 0; | |
1149 | + unjam(); | |
1150 | + return; | |
c1c6733f AM |
1151 | +} |
1152 | + | |
bb1d8b11 | 1153 | +static void ack_timer_fn(unsigned long arg) |
b7b72b66 | 1154 | +{ |
bb1d8b11 | 1155 | + P_COMMS("%ld: ack_timer fired, retries=%d\n", jiffies, retry_count); |
b7b72b66 | 1156 | + |
bb1d8b11 AM |
1157 | + /* Too many retries ? */ |
1158 | + if (++retry_count > MAX_RETRIES) { | |
1159 | + set_bit(ACK_TIMEOUT, &mainloop_flags); | |
1160 | + wake_up_interruptible(&cnxman_waitq); | |
b7b72b66 AM |
1161 | + } |
1162 | + else { | |
bb1d8b11 AM |
1163 | + /* Resend last message */ |
1164 | + set_bit(RESEND_NEEDED, &mainloop_flags); | |
1165 | + wake_up_interruptible(&cnxman_waitq); | |
b7b72b66 | 1166 | + } |
bb1d8b11 | 1167 | +} |
b7b72b66 | 1168 | + |
bb1d8b11 AM |
1169 | +/* Called to resend a packet if sock_sendmsg was busy */ |
1170 | +static void short_timer_fn(unsigned long arg) | |
1171 | +{ | |
1172 | + P_COMMS("short_timer fired\n"); | |
b7b72b66 | 1173 | + |
bb1d8b11 AM |
1174 | + /* Resend last message */ |
1175 | + resend_delay <<= 1; | |
1176 | + set_bit(RESEND_NEEDED, &mainloop_flags); | |
1177 | + wake_up_interruptible(&cnxman_waitq); | |
1178 | +} | |
b7b72b66 | 1179 | + |
bb1d8b11 AM |
1180 | +static void start_ack_timer() |
1181 | +{ | |
1182 | + ack_timer.function = ack_timer_fn; | |
1183 | + ack_timer.data = 0L; | |
1184 | + mod_timer(&ack_timer, jiffies + HZ); | |
1185 | +} | |
b7b72b66 | 1186 | + |
bb1d8b11 AM |
1187 | +static void start_short_timer(void) |
1188 | +{ | |
1189 | + ack_timer.function = short_timer_fn; | |
1190 | + ack_timer.data = 0L; | |
1191 | + mod_timer(&ack_timer, jiffies + (resend_delay * HZ)); | |
1192 | +} | |
b7b72b66 | 1193 | + |
b7b72b66 | 1194 | + |
bb1d8b11 AM |
1195 | +static struct cl_waiting_listen_request *find_listen_request(unsigned short tag) |
1196 | +{ | |
1197 | + struct list_head *llist; | |
1198 | + struct cl_waiting_listen_request *listener; | |
b7b72b66 | 1199 | + |
bb1d8b11 AM |
1200 | + list_for_each(llist, &listenreq_list) { |
1201 | + listener = list_entry(llist, struct cl_waiting_listen_request, | |
1202 | + list); | |
1203 | + if (listener->tag == tag) { | |
1204 | + return listener; | |
1205 | + } | |
1206 | + } | |
1207 | + return NULL; | |
b7b72b66 AM |
1208 | +} |
1209 | + | |
bb1d8b11 | 1210 | +static void process_ack(struct cluster_node *rem_node, unsigned short seq) |
c1c6733f | 1211 | +{ |
bb1d8b11 AM |
1212 | + if (rem_node && rem_node->state != NODESTATE_DEAD) { |
1213 | + /* This copes with duplicate acks from a multipathed | |
1214 | + * host */ | |
1215 | + if (rem_node->last_seq_acked != | |
1216 | + le16_to_cpu(seq)) { | |
1217 | + rem_node->last_seq_acked = | |
1218 | + le16_to_cpu(seq); | |
c1c6733f | 1219 | + |
bb1d8b11 AM |
1220 | + /* Got em all */ |
1221 | + if (++ack_count >= acks_expected) { | |
c1c6733f | 1222 | + |
bb1d8b11 AM |
1223 | + /* Cancel the timer */ |
1224 | + del_timer(&ack_timer); | |
1225 | + acks_expected = 0; | |
1226 | + unjam(); | |
1227 | + } | |
1228 | + } | |
1229 | + } | |
1230 | +} | |
c1c6733f | 1231 | + |
bb1d8b11 AM |
1232 | +static void process_cnxman_message(struct cl_comms_socket *csock, char *data, |
1233 | + int len, char *addr, int addrlen, | |
1234 | + struct cluster_node *rem_node) | |
1235 | +{ | |
1236 | + struct cl_protmsg *msg = (struct cl_protmsg *) data; | |
1237 | + struct cl_protheader *header = (struct cl_protheader *) data; | |
1238 | + struct cl_ackmsg *ackmsg; | |
1239 | + struct cl_listenmsg *listenmsg; | |
1240 | + struct cl_closemsg *closemsg; | |
1241 | + struct cl_barriermsg *barriermsg; | |
1242 | + struct cl_waiting_listen_request *listen_request; | |
c1c6733f | 1243 | + |
bb1d8b11 AM |
1244 | + P_COMMS("Message on port 0 is %d\n", msg->cmd); |
1245 | + switch (msg->cmd) { | |
1246 | + case CLUSTER_CMD_ACK: | |
1247 | + ackmsg = (struct cl_ackmsg *) data; | |
c1c6733f | 1248 | + |
bb1d8b11 AM |
1249 | + if (rem_node && (ackmsg->aflags & 1)) { |
1250 | + if (net_ratelimit()) | |
1251 | + printk(KERN_INFO CMAN_NAME | |
1252 | + ": WARNING no listener for port %d on node %s\n", | |
1253 | + ackmsg->remport, rem_node->name); | |
1254 | + } | |
1255 | + P_COMMS("Got ACK from %s. seq=%d (cur=%d)\n", | |
1256 | + rem_node ? rem_node->name : "Unknown", | |
1257 | + le16_to_cpu(ackmsg->header.ack), cur_seq); | |
c1c6733f | 1258 | + |
bb1d8b11 AM |
1259 | + /* ACK processing has already happened */ |
1260 | + break; | |
c1c6733f | 1261 | + |
bb1d8b11 AM |
1262 | + /* Return 1 if we have a listener on this port, 0 if not */ |
1263 | + case CLUSTER_CMD_LISTENREQ: | |
1264 | + listenmsg = | |
1265 | + (struct cl_listenmsg *) (data + | |
1266 | + sizeof (struct cl_protheader)); | |
1267 | + cl_sendack(csock, header->seq, addrlen, addr, header->tgtport, 0); | |
1268 | + send_listen_response(csock, le32_to_cpu(header->srcid), | |
1269 | + listenmsg->target_port, listenmsg->tag); | |
1270 | + break; | |
c1c6733f | 1271 | + |
bb1d8b11 AM |
1272 | + case CLUSTER_CMD_LISTENRESP: |
1273 | + /* Wake up process waiting for listen response */ | |
1274 | + listenmsg = | |
1275 | + (struct cl_listenmsg *) (data + | |
1276 | + sizeof (struct cl_protheader)); | |
1277 | + cl_sendack(csock, header->seq, addrlen, addr, header->tgtport, 0); | |
1278 | + down(&listenreq_lock); | |
1279 | + listen_request = find_listen_request(listenmsg->tag); | |
1280 | + if (listen_request) { | |
1281 | + listen_request->result = listenmsg->listening; | |
1282 | + listen_request->waiting = 0; | |
1283 | + wake_up_interruptible(&listen_request->waitq); | |
1284 | + } | |
1285 | + up(&listenreq_lock); | |
1286 | + break; | |
c1c6733f | 1287 | + |
bb1d8b11 AM |
1288 | + case CLUSTER_CMD_PORTCLOSED: |
1289 | + closemsg = | |
1290 | + (struct cl_closemsg *) (data + | |
1291 | + sizeof (struct cl_protheader)); | |
1292 | + cl_sendack(csock, header->seq, addrlen, addr, header->tgtport, 0); | |
1293 | + post_close_oob(closemsg->port, le32_to_cpu(header->srcid)); | |
1294 | + break; | |
c1c6733f | 1295 | + |
bb1d8b11 AM |
1296 | + case CLUSTER_CMD_BARRIER: |
1297 | + barriermsg = | |
1298 | + (struct cl_barriermsg *) (data + | |
1299 | + sizeof (struct cl_protheader)); | |
1300 | + cl_sendack(csock, header->seq, addrlen, addr, header->tgtport, 0); | |
1301 | + if (rem_node) | |
1302 | + process_barrier_msg(barriermsg, rem_node); | |
1303 | + break; | |
c1c6733f | 1304 | + |
bb1d8b11 AM |
1305 | + default: |
1306 | + printk(KERN_ERR CMAN_NAME | |
1307 | + ": Unknown protocol message %d received\n", msg->cmd); | |
1308 | + break; | |
c1c6733f | 1309 | + |
c1c6733f | 1310 | + } |
bb1d8b11 | 1311 | + return; |
c1c6733f AM |
1312 | +} |
1313 | + | |
bb1d8b11 | 1314 | +static int valid_addr_for_node(struct cluster_node *node, char *addr) |
c1c6733f | 1315 | +{ |
bb1d8b11 AM |
1316 | + struct list_head *addrlist; |
1317 | + struct cluster_node_addr *nodeaddr; | |
c1c6733f | 1318 | + |
bb1d8b11 AM |
1319 | + /* We don't compare the first two bytes of the address because it's |
1320 | + * the Address Family and always in native byte order...so it will | |
1321 | + * not match if we have mixed big & little-endian machines in the cluster | |
1322 | + */ | |
c1c6733f | 1323 | + |
bb1d8b11 AM |
1324 | + list_for_each(addrlist, &node->addr_list) { |
1325 | + nodeaddr = list_entry(addrlist, struct cluster_node_addr, list); | |
c1c6733f | 1326 | + |
bb1d8b11 AM |
1327 | + if (memcmp(nodeaddr->addr+2, addr+2, address_length-2) == 0) |
1328 | + return 1; /* TRUE */ | |
c1c6733f | 1329 | + } |
bb1d8b11 | 1330 | + return 0; /* FALSE */ |
c1c6733f AM |
1331 | +} |
1332 | + | |
bb1d8b11 | 1333 | +static void memcpy_fromkvec(void *data, struct kvec *vec, int len) |
c1c6733f | 1334 | +{ |
bb1d8b11 AM |
1335 | + while (len > 0) { |
1336 | + if (vec->iov_len) { | |
1337 | + int copy = min_t(unsigned int, len, vec->iov_len); | |
1338 | + memcpy(data, vec->iov_base, copy); | |
1339 | + len -= copy; | |
1340 | + data += copy; | |
1341 | + vec->iov_base += copy; | |
1342 | + vec->iov_len -= copy; | |
1343 | + } | |
1344 | + vec++; | |
1345 | + } | |
1346 | +} | |
c1c6733f | 1347 | + |
bb1d8b11 AM |
1348 | +static int send_to_user_port(struct cl_comms_socket *csock, |
1349 | + struct cl_protheader *header, | |
1350 | + struct msghdr *msg, | |
1351 | + struct kvec *iov, int veclen, | |
1352 | + int len) | |
1353 | +{ | |
1354 | + struct sk_buff *skb; | |
1355 | + struct cb_info *cbinfo; | |
1356 | + int err; | |
c1c6733f | 1357 | + |
bb1d8b11 AM |
1358 | + /* Get the port number and look for a listener */ |
1359 | + down(&port_array_lock); | |
1360 | + if (port_array[header->tgtport]) { | |
1361 | + struct cluster_sock *c = cluster_sk(port_array[header->tgtport]); | |
c1c6733f | 1362 | + |
bb1d8b11 AM |
1363 | + /* ACK it */ |
1364 | + if (!(header->flags & MSG_NOACK) && | |
1365 | + !(header->flags & MSG_REPLYEXP)) { | |
c1c6733f | 1366 | + |
bb1d8b11 AM |
1367 | + cl_sendack(csock, header->seq, msg->msg_namelen, |
1368 | + msg->msg_name, header->tgtport, 0); | |
1369 | + } | |
c1c6733f | 1370 | + |
bb1d8b11 AM |
1371 | + /* Call a callback if there is one */ |
1372 | + if (c->kernel_callback) { | |
1373 | + up(&port_array_lock); | |
1374 | + if (veclen == 1) { | |
1375 | + c->kernel_callback(iov->iov_base, | |
1376 | + iov->iov_len, | |
1377 | + msg->msg_name, msg->msg_namelen, | |
1378 | + le32_to_cpu(header->srcid)); | |
c1c6733f | 1379 | + |
bb1d8b11 AM |
1380 | + } |
1381 | + else { /* Unroll iov, this Hardly ever Happens */ | |
1382 | + char *data; | |
1383 | + data = kmalloc(len, GFP_KERNEL); | |
1384 | + if (!data) | |
1385 | + return -ENOMEM; | |
c1c6733f | 1386 | + |
bb1d8b11 AM |
1387 | + memcpy_fromkvec(data, iov, len); |
1388 | + c->kernel_callback(data, len, | |
1389 | + msg->msg_name, msg->msg_namelen, | |
1390 | + le32_to_cpu(header->srcid)); | |
1391 | + kfree(data); | |
1392 | + } | |
1393 | + return len; | |
1394 | + } | |
c1c6733f | 1395 | + |
bb1d8b11 AM |
1396 | + /* Otherwise put it into an SKB and pass it onto the recvmsg |
1397 | + * mechanism */ | |
1398 | + skb = alloc_skb(len, GFP_KERNEL); | |
1399 | + if (!skb) { | |
1400 | + up(&port_array_lock); | |
1401 | + printk(KERN_INFO CMAN_NAME | |
1402 | + ": Failed to allocate skb\n"); | |
1403 | + return -ENOMEM; | |
1404 | + } | |
c1c6733f | 1405 | + |
bb1d8b11 AM |
1406 | + skb_put(skb, len); |
1407 | + memcpy_fromkvec(skb->data, iov, len); | |
c1c6733f | 1408 | + |
bb1d8b11 AM |
1409 | + /* Put metadata into cb[] */ |
1410 | + cbinfo = (struct cb_info *)skb->cb; | |
1411 | + cbinfo->orig_nodeid = le32_to_cpu(header->srcid); | |
1412 | + cbinfo->orig_port = header->srcport; | |
1413 | + cbinfo->oob = 0; | |
c1c6733f | 1414 | + |
bb1d8b11 AM |
1415 | + if ((err = |
1416 | + sock_queue_rcv_skb(port_array[header->tgtport], skb)) < 0) { | |
c1c6733f | 1417 | + |
bb1d8b11 AM |
1418 | + printk(KERN_INFO CMAN_NAME |
1419 | + ": Error queueing request to port %d: %d\n", | |
1420 | + header->tgtport, err); | |
1421 | + kfree_skb(skb); | |
c1c6733f | 1422 | + |
bb1d8b11 AM |
1423 | + /* If the port was MEMBERSHIP then we have to die */ |
1424 | + if (header->tgtport == CLUSTER_PORT_MEMBERSHIP) { | |
1425 | + up(&port_array_lock); | |
1426 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
1427 | + panic("membership stopped responding"); | |
1428 | + } | |
1429 | + } | |
1430 | + up(&port_array_lock); | |
c1c6733f | 1431 | + |
bb1d8b11 AM |
1432 | + } |
1433 | + else { | |
1434 | + /* ACK it, but set the flag bit so remote end knows no-one | |
1435 | + * caught it */ | |
1436 | + if (!(header->flags & MSG_NOACK)) | |
1437 | + cl_sendack(csock, header->seq, | |
1438 | + msg->msg_namelen, msg->msg_name, | |
1439 | + header->tgtport, 1); | |
c1c6733f | 1440 | + |
bb1d8b11 AM |
1441 | + /* Nobody listening, drop it */ |
1442 | + up(&port_array_lock); | |
1443 | + } | |
1444 | + return len; | |
c1c6733f AM |
1445 | +} |
1446 | + | |
bb1d8b11 AM |
1447 | +/* NOTE: This routine knows (assumes!) that there is only one |
1448 | + iov element passed into it. */ | |
1449 | +static void process_incoming_packet(struct cl_comms_socket *csock, | |
1450 | + struct msghdr *msg, | |
1451 | + struct kvec *vec, int veclen, int len) | |
c1c6733f | 1452 | +{ |
bb1d8b11 AM |
1453 | + char *data = vec->iov_base; |
1454 | + char *addr = msg->msg_name; | |
1455 | + int addrlen = msg->msg_namelen; | |
1456 | + struct cl_protheader *header = (struct cl_protheader *) data; | |
1457 | + struct cluster_node *rem_node = | |
1458 | + find_node_by_nodeid(le32_to_cpu(header->srcid)); | |
c1c6733f | 1459 | + |
bb1d8b11 AM |
1460 | + P_COMMS("seen message, from %d for %d, sequence num = %d, rem_node=%p, state=%d\n", |
1461 | + le32_to_cpu(header->srcid), le32_to_cpu(header->tgtid), | |
1462 | + le16_to_cpu(header->seq), rem_node, | |
1463 | + rem_node ? rem_node->state : -1); | |
c1c6733f | 1464 | + |
bb1d8b11 AM |
1465 | + /* If the remote end is being coy about its node ID then look it up by |
1466 | + * address */ | |
1467 | + if (!rem_node && header->srcid == 0) { | |
1468 | + rem_node = find_node_by_addr(addr, addrlen); | |
1469 | + } | |
c1c6733f | 1470 | + |
bb1d8b11 AM |
1471 | + /* If this node is an ex-member then treat it as unknown */ |
1472 | + if (rem_node && rem_node->state != NODESTATE_MEMBER | |
1473 | + && rem_node->state != NODESTATE_JOINING) | |
1474 | + rem_node = NULL; | |
c1c6733f | 1475 | + |
bb1d8b11 AM |
1476 | + /* Ignore messages not for our cluster */ |
1477 | + if (le16_to_cpu(header->cluster) != cluster_id) { | |
1478 | + P_COMMS("Dumping message - wrong cluster ID (us=%d, msg=%d)\n", | |
1479 | + cluster_id, header->cluster); | |
1480 | + goto incoming_finish; | |
1481 | + } | |
c1c6733f | 1482 | + |
bb1d8b11 AM |
1483 | + /* If the message is from us then just dump it */ |
1484 | + if (rem_node && rem_node->us) | |
1485 | + goto incoming_finish; | |
c1c6733f | 1486 | + |
bb1d8b11 AM |
1487 | + /* If we can't find the nodeid then check for our own messages the hard |
1488 | + * way - this only happens during joining */ | |
1489 | + if (!rem_node) { | |
1490 | + struct list_head *socklist; | |
1491 | + struct cl_comms_socket *clsock; | |
c1c6733f | 1492 | + |
bb1d8b11 AM |
1493 | + list_for_each(socklist, &socket_list) { |
1494 | + clsock = | |
1495 | + list_entry(socklist, struct cl_comms_socket, list); | |
c1c6733f | 1496 | + |
bb1d8b11 | 1497 | + if (clsock->recv_only) { |
c1c6733f | 1498 | + |
bb1d8b11 AM |
1499 | + if (memcmp(addr, &clsock->saddr, address_length) == 0) { |
1500 | + goto incoming_finish; | |
1501 | + } | |
1502 | + } | |
1503 | + } | |
c1c6733f | 1504 | + |
bb1d8b11 | 1505 | + } |
c1c6733f | 1506 | + |
bb1d8b11 AM |
1507 | + /* Ignore messages not for us */ |
1508 | + if (le32_to_cpu(header->tgtid) > 0 && us | |
1509 | + && le32_to_cpu(header->tgtid) != us->node_id) { | |
1510 | + goto incoming_finish; | |
1511 | + } | |
c1c6733f | 1512 | + |
bb1d8b11 AM |
1513 | + P_COMMS("got message, from %d for %d, sequence num = %d\n", |
1514 | + le32_to_cpu(header->srcid), le32_to_cpu(header->tgtid), | |
1515 | + le16_to_cpu(header->seq)); | |
1516 | + | |
1517 | + if (header->ack && rem_node) { | |
1518 | + process_ack(rem_node, header->ack); | |
c1c6733f | 1519 | + } |
c1c6733f | 1520 | + |
bb1d8b11 AM |
1521 | + /* Have we received this message before ? If so just ignore it, it's a |
1522 | + * resend for someone else's benefit */ | |
1523 | + if (!(header->flags & MSG_NOACK) && | |
1524 | + rem_node && le16_to_cpu(header->seq) == rem_node->last_seq_recv) { | |
1525 | + P_COMMS | |
1526 | + ("Discarding message - Already seen this sequence number %d\n", | |
1527 | + rem_node->last_seq_recv); | |
1528 | + /* Still need to ACK it though, in case it was the ACK that got | |
1529 | + * lost */ | |
1530 | + cl_sendack(csock, header->seq, addrlen, addr, header->tgtport, 0); | |
1531 | + goto incoming_finish; | |
1532 | + } | |
c1c6733f | 1533 | + |
bb1d8b11 AM |
1534 | + /* Check that the message is from the node we think it is from */ |
1535 | + if (rem_node && !valid_addr_for_node(rem_node, addr)) { | |
1536 | + return; | |
1537 | + } | |
c1c6733f | 1538 | + |
bb1d8b11 AM |
1539 | + /* If it's a new node then assign it a temporary node ID */ |
1540 | + if (!rem_node) | |
1541 | + header->srcid = cpu_to_le32(new_temp_nodeid(addr, addrlen)); | |
c1c6733f | 1542 | + |
bb1d8b11 AM |
1543 | + P_COMMS("Got message: flags = %x, port = %d, we_are_a_member = %d\n", |
1544 | + header->flags, header->tgtport, we_are_a_cluster_member); | |
c1c6733f | 1545 | + |
c1c6733f | 1546 | + |
bb1d8b11 AM |
1547 | + /* If we are not part of the cluster then ignore multicast messages |
1548 | + * that need an ACK as we will confuse the sender who is only expecting | |
1549 | + * ACKS from bona fide members */ | |
1550 | + if ((header->flags & MSG_MULTICAST) && | |
1551 | + !(header->flags & MSG_NOACK) && !we_are_a_cluster_member) { | |
1552 | + P_COMMS | |
1553 | + ("Discarding message - multicast and we are not a cluster member. port=%d flags=%x\n", | |
1554 | + header->tgtport, header->flags); | |
1555 | + goto incoming_finish; | |
1556 | + } | |
c1c6733f | 1557 | + |
bb1d8b11 AM |
1558 | + /* Save the sequence number of this message so we can ignore duplicates |
1559 | + * (above) */ | |
1560 | + if (!(header->flags & MSG_NOACK) && rem_node) { | |
1561 | + P_COMMS("Saving seq %d for node %s\n", le16_to_cpu(header->seq), | |
1562 | + rem_node->name); | |
1563 | + rem_node->last_seq_recv = le16_to_cpu(header->seq); | |
1564 | + } | |
c1c6733f | 1565 | + |
bb1d8b11 AM |
1566 | + /* Is it a protocol message? */ |
1567 | + if (header->tgtport == 0) { | |
1568 | + process_cnxman_message(csock, data, len, addr, addrlen, | |
1569 | + rem_node); | |
1570 | + goto incoming_finish; | |
1571 | + } | |
c1c6733f | 1572 | + |
bb1d8b11 AM |
1573 | + /* Skip past the header to the data */ |
1574 | + vec[0].iov_base = data + sizeof (struct cl_protheader); | |
1575 | + vec[0].iov_len -= sizeof (struct cl_protheader); | |
1576 | + len -= sizeof (struct cl_protheader); | |
c1c6733f | 1577 | + |
bb1d8b11 | 1578 | + send_to_user_port(csock, header, msg, vec, veclen, len); |
c1c6733f | 1579 | + |
bb1d8b11 AM |
1580 | + incoming_finish: |
1581 | + return; | |
1582 | +} | |
c1c6733f | 1583 | + |
bb1d8b11 AM |
1584 | +static struct sock *cl_alloc_sock(struct socket *sock, int gfp) |
1585 | +{ | |
1586 | + struct sock *sk; | |
1587 | + struct cluster_sock *c; | |
c1c6733f | 1588 | + |
bb1d8b11 AM |
1589 | + if ((sk = |
1590 | + sk_alloc(AF_CLUSTER, gfp, sizeof (struct cluster_sock), | |
1591 | + cluster_sk_cachep)) == NULL) | |
1592 | + goto no_sock; | |
1593 | + | |
1594 | + if (sock) { | |
1595 | + sock->ops = &cl_proto_ops; | |
c1c6733f | 1596 | + } |
bb1d8b11 | 1597 | + sock_init_data(sock, sk); |
c1c6733f | 1598 | + |
bb1d8b11 AM |
1599 | + sk->sk_destruct = NULL; |
1600 | + sk->sk_no_check = 1; | |
1601 | + sk->sk_family = PF_CLUSTER; | |
1602 | + sk->sk_allocation = gfp; | |
c1c6733f | 1603 | + |
bb1d8b11 AM |
1604 | + c = cluster_sk(sk); |
1605 | + c->port = 0; | |
1606 | + c->service_data = NULL; | |
c1c6733f | 1607 | + |
bb1d8b11 AM |
1608 | + return sk; |
1609 | + no_sock: | |
1610 | + return NULL; | |
1611 | +} | |
1612 | + | |
1613 | +static int cl_release(struct socket *sock) | |
1614 | +{ | |
1615 | + struct sock *sk = sock->sk; | |
1616 | + struct cl_client_socket *csock; | |
1617 | + struct list_head *socklist; | |
1618 | + struct list_head *tmp; | |
1619 | + | |
1620 | + down(&client_socket_lock); | |
1621 | + if (sk) { | |
1622 | + /* Remove port allocations if it's a bound socket */ | |
1623 | + struct cluster_sock *c = cluster_sk(sk); | |
1624 | + | |
1625 | + down(&port_array_lock); | |
1626 | + if (c->port) { | |
1627 | + port_array[c->port] = NULL; | |
1628 | + } | |
1629 | + up(&port_array_lock); | |
c1c6733f | 1630 | + |
bb1d8b11 AM |
1631 | + /* Tell other nodes in the cluster that this listener is going |
1632 | + * away */ | |
1633 | + if (atomic_read(&cnxman_running) && c->port) | |
1634 | + send_port_close_oob(c->port); | |
c1c6733f | 1635 | + |
bb1d8b11 AM |
1636 | + if (c->service_data) |
1637 | + sm_sock_release(sock); | |
c1c6733f | 1638 | + |
bb1d8b11 AM |
1639 | + /* Master socket released ? */ |
1640 | + if (sk->sk_protocol == CLPROTO_MASTER) { | |
1641 | + master_sock = NULL; | |
c1c6733f | 1642 | + |
bb1d8b11 AM |
1643 | + /* If this socket is being freed and cnxman is not |
1644 | + * started then free all the comms sockets as either | |
1645 | + * the userland "join" process has crashed or the | |
1646 | + * join failed. | |
1647 | + */ | |
1648 | + if (!atomic_read(&cnxman_running)) { | |
1649 | + quit_threads = 1; | |
1650 | + free_cluster_sockets(); | |
1651 | + } | |
1652 | + } | |
c1c6733f | 1653 | + |
bb1d8b11 AM |
1654 | + sock_orphan(sk); |
1655 | + sock_hold(sk); | |
1656 | + lock_sock(sk); | |
1657 | + release_sock(sk); | |
1658 | + sock_put(sk); | |
1659 | + sock_put(sk); | |
1660 | + sock->sk = NULL; | |
1661 | + } | |
c1c6733f | 1662 | + |
bb1d8b11 AM |
1663 | + /* Remove it from the list of clients */ |
1664 | + list_for_each_safe(socklist, tmp, &client_socket_list) { | |
1665 | + csock = list_entry(socklist, struct cl_client_socket, list); | |
1666 | + | |
1667 | + if (csock->sock == sock) { | |
1668 | + list_del(&csock->list); | |
1669 | + kfree(csock); | |
1670 | + break; | |
c1c6733f AM |
1671 | + } |
1672 | + } | |
bb1d8b11 | 1673 | + up(&client_socket_lock); |
b7b72b66 | 1674 | + |
bb1d8b11 | 1675 | + return 0; |
c1c6733f AM |
1676 | +} |
1677 | + | |
bb1d8b11 | 1678 | +static int cl_create(struct socket *sock, int protocol) |
c1c6733f | 1679 | +{ |
bb1d8b11 | 1680 | + struct sock *sk; |
c1c6733f | 1681 | + |
bb1d8b11 AM |
1682 | + /* All are datagrams */ |
1683 | + if (sock->type != SOCK_DGRAM) | |
1684 | + return -ESOCKTNOSUPPORT; | |
1685 | + | |
1686 | + if (protocol == CLPROTO_MASTER && !capable(CAP_CLUSTER)) | |
c1c6733f AM |
1687 | + return -EPERM; |
1688 | + | |
bb1d8b11 AM |
1689 | + /* Can only have one master socket */ |
1690 | + if (master_sock && protocol == CLPROTO_MASTER) | |
1691 | + return -EBUSY; | |
c1c6733f | 1692 | + |
bb1d8b11 AM |
1693 | + /* cnxman not running and a client was requested */ |
1694 | + if (!atomic_read(&cnxman_running) && protocol != CLPROTO_MASTER) | |
1695 | + return -ENETDOWN; | |
c1c6733f | 1696 | + |
bb1d8b11 AM |
1697 | + if ((sk = cl_alloc_sock(sock, GFP_KERNEL)) == NULL) |
1698 | + return -ENOBUFS; | |
c1c6733f | 1699 | + |
bb1d8b11 | 1700 | + sk->sk_protocol = protocol; |
c1c6733f | 1701 | + |
bb1d8b11 AM |
1702 | + if (protocol == CLPROTO_MASTER) |
1703 | + master_sock = sk; | |
1704 | + | |
1705 | + /* Add client sockets to the list */ | |
1706 | + if (protocol == CLPROTO_CLIENT) { | |
1707 | + struct cl_client_socket *clsock = | |
1708 | + kmalloc(sizeof (struct cl_client_socket), GFP_KERNEL); | |
1709 | + if (!clsock) { | |
1710 | + cl_release(sock); | |
1711 | + return -ENOMEM; | |
1712 | + } | |
1713 | + clsock->sock = sock; | |
1714 | + down(&client_socket_lock); | |
1715 | + list_add(&clsock->list, &client_socket_list); | |
1716 | + up(&client_socket_lock); | |
1717 | + } | |
c1c6733f AM |
1718 | + |
1719 | + return 0; | |
1720 | +} | |
1721 | + | |
bb1d8b11 | 1722 | +static int cl_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) |
b7b72b66 | 1723 | +{ |
bb1d8b11 AM |
1724 | + struct sock *sk = sock->sk; |
1725 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *) uaddr; | |
1726 | + struct cluster_sock *c = cluster_sk(sk); | |
b7b72b66 | 1727 | + |
bb1d8b11 | 1728 | + if (!capable(CAP_NET_BIND_SERVICE)) |
b7b72b66 AM |
1729 | + return -EPERM; |
1730 | + | |
bb1d8b11 | 1731 | + if (sk->sk_zapped == 0) |
b7b72b66 AM |
1732 | + return -EINVAL; |
1733 | + | |
bb1d8b11 AM |
1734 | + if (addr_len != sizeof (struct sockaddr_cl)) |
1735 | + return -EINVAL; | |
b7b72b66 | 1736 | + |
bb1d8b11 AM |
1737 | + if (saddr->scl_family != AF_CLUSTER) |
1738 | + return -EINVAL; | |
b7b72b66 | 1739 | + |
bb1d8b11 AM |
1740 | + if (saddr->scl_port == 0) |
1741 | + return -EINVAL; /* Port 0 is reserved for protocol messages */ | |
b7b72b66 | 1742 | + |
bb1d8b11 AM |
1743 | + down(&port_array_lock); |
1744 | + | |
1745 | + if (port_array[saddr->scl_port]) { | |
1746 | + up(&port_array_lock); | |
1747 | + return -EADDRINUSE; | |
b7b72b66 | 1748 | + } |
b7b72b66 | 1749 | + |
bb1d8b11 | 1750 | + port_array[saddr->scl_port] = sk; |
b7b72b66 | 1751 | + |
bb1d8b11 | 1752 | + up(&port_array_lock); |
b7b72b66 | 1753 | + |
bb1d8b11 AM |
1754 | + c->port = saddr->scl_port; |
1755 | + sk->sk_zapped = 0; | |
c783755a | 1756 | + |
bb1d8b11 AM |
1757 | + /* If we are not a cluster member yet then make the client wait until |
1758 | + * we are, this allows nodes to start cluster clients at the same time | |
1759 | + * as cluster services but they will wait until membership is achieved. | |
1760 | + * This looks odd in bind() (open would seem more obvious) but we need | |
1761 | + * to know which port number is being used so that things like | |
1762 | + * membership services don't get blocked | |
1763 | + */ | |
1764 | + | |
1765 | + if (saddr->scl_port > HIGH_PROTECTED_PORT) | |
1766 | + while (!we_are_a_cluster_member || !cluster_is_quorate | |
1767 | + || in_transition()) { | |
1768 | + DECLARE_WAITQUEUE(wq, current); | |
1769 | + struct task_struct *tsk = current; | |
1770 | + | |
1771 | + set_task_state(tsk, TASK_INTERRUPTIBLE); | |
1772 | + add_wait_queue(&socket_waitq, &wq); | |
1773 | + | |
1774 | + if (!we_are_a_cluster_member || !cluster_is_quorate | |
1775 | + || in_transition()) | |
1776 | + schedule(); | |
1777 | + | |
1778 | + set_task_state(tsk, TASK_RUNNING); | |
1779 | + remove_wait_queue(&socket_waitq, &wq); | |
1780 | + | |
1781 | + /* We were woken up because the cluster is going down, | |
1782 | + * ...and we never got a chance to do any work! (sob) */ | |
1783 | + if (atomic_read(&cnxman_running) == 0 || quit_threads) { | |
1784 | + return -ENOTCONN; | |
1785 | + } | |
1786 | + } | |
c783755a | 1787 | + |
c783755a | 1788 | + return 0; |
b7b72b66 AM |
1789 | +} |
1790 | + | |
bb1d8b11 AM |
1791 | +static int cl_getname(struct socket *sock, struct sockaddr *uaddr, |
1792 | + int *uaddr_len, int peer) | |
b7b72b66 | 1793 | +{ |
bb1d8b11 AM |
1794 | + struct sockaddr_cl *sa = (struct sockaddr_cl *) uaddr; |
1795 | + struct sock *sk = sock->sk; | |
1796 | + struct cluster_sock *c = cluster_sk(sk); | |
b7b72b66 | 1797 | + |
bb1d8b11 | 1798 | + *uaddr_len = sizeof (struct sockaddr_cl); |
b7b72b66 | 1799 | + |
bb1d8b11 | 1800 | + lock_sock(sk); |
b7b72b66 | 1801 | + |
bb1d8b11 AM |
1802 | + sa->scl_port = c->port; |
1803 | + sa->scl_flags = 0; | |
1804 | + sa->scl_family = AF_CLUSTER; | |
b7b72b66 | 1805 | + |
bb1d8b11 | 1806 | + release_sock(sk); |
b7b72b66 | 1807 | + |
bb1d8b11 AM |
1808 | + return 0; |
1809 | +} | |
b7b72b66 | 1810 | + |
bb1d8b11 AM |
1811 | +static unsigned int cl_poll(struct file *file, struct socket *sock, |
1812 | + poll_table * wait) | |
1813 | +{ | |
1814 | + return datagram_poll(file, sock, wait); | |
1815 | +} | |
b7b72b66 | 1816 | + |
bb1d8b11 AM |
1817 | +/* Copy internal node format to userland format */ |
1818 | +void copy_to_usernode(struct cluster_node *node, | |
1819 | + struct cl_cluster_node *unode) | |
1820 | +{ | |
1821 | + strcpy(unode->name, node->name); | |
1822 | + unode->size = sizeof (struct cl_cluster_node); | |
1823 | + unode->votes = node->votes; | |
1824 | + unode->state = node->state; | |
1825 | + unode->us = node->us; | |
1826 | + unode->node_id = node->node_id; | |
1827 | + unode->leave_reason = node->leave_reason; | |
1828 | + unode->incarnation = node->incarnation; | |
1829 | +} | |
1830 | + | |
1831 | +static int add_clsock(int broadcast, int number, struct socket *sock, | |
1832 | + struct file *file) | |
1833 | +{ | |
1834 | + struct cl_comms_socket *newsock = | |
1835 | + kmalloc(sizeof (struct cl_comms_socket), GFP_KERNEL); | |
1836 | + if (!newsock) | |
b7b72b66 AM |
1837 | + return -ENOMEM; |
1838 | + | |
bb1d8b11 AM |
1839 | + memset(newsock, 0, sizeof (*newsock)); |
1840 | + newsock->number = number; | |
1841 | + newsock->sock = sock; | |
1842 | + if (broadcast) { | |
1843 | + newsock->broadcast = 1; | |
1844 | + newsock->recv_only = 0; | |
1845 | + } | |
1846 | + else { | |
1847 | + newsock->broadcast = 0; | |
1848 | + newsock->recv_only = 1; | |
1849 | + } | |
b7b72b66 | 1850 | + |
bb1d8b11 AM |
1851 | + newsock->file = file; |
1852 | + newsock->addr_len = sizeof(struct sockaddr_in6); | |
b7b72b66 | 1853 | + |
bb1d8b11 AM |
1854 | + /* Mark it active until cnxman thread is running and ready to process |
1855 | + * messages */ | |
1856 | + set_bit(1, &newsock->active); | |
b7b72b66 | 1857 | + |
bb1d8b11 AM |
1858 | + /* Find out what it's bound to */ |
1859 | + newsock->sock->ops->getname(newsock->sock, | |
1860 | + (struct sockaddr *)&newsock->saddr, | |
1861 | + &newsock->addr_len, 0); | |
b7b72b66 | 1862 | + |
bb1d8b11 AM |
1863 | + num_interfaces = max(num_interfaces, newsock->number); |
1864 | + if (!current_interface && newsock->broadcast) | |
1865 | + current_interface = newsock; | |
b7b72b66 | 1866 | + |
bb1d8b11 AM |
1867 | + /* Hook data_ready */ |
1868 | + newsock->sock->sk->sk_data_ready = cnxman_data_ready; | |
1869 | + | |
1870 | + /* Make an attempt to keep them in order */ | |
1871 | + list_add_tail(&newsock->list, &socket_list); | |
1872 | + | |
1873 | + address_length = newsock->addr_len; | |
b7b72b66 AM |
1874 | + return 0; |
1875 | +} | |
1876 | + | |
bb1d8b11 AM |
1877 | +/* ioctl processing functions */ |
1878 | + | |
1879 | +static int do_ioctl_set_version(unsigned long arg) | |
b7b72b66 | 1880 | +{ |
bb1d8b11 AM |
1881 | + struct cl_version version, *u_version; |
1882 | + | |
b7b72b66 AM |
1883 | + if (!capable(CAP_CLUSTER)) |
1884 | + return -EPERM; | |
bb1d8b11 AM |
1885 | + if (arg == 0) |
1886 | + return -EINVAL; | |
b7b72b66 | 1887 | + |
bb1d8b11 | 1888 | + u_version = (struct cl_version *) arg; |
b7b72b66 | 1889 | + |
bb1d8b11 AM |
1890 | + if (copy_from_user(&version, u_version, sizeof(struct cl_version))) |
1891 | + return -EFAULT; | |
b7b72b66 | 1892 | + |
bb1d8b11 AM |
1893 | + if (version.major != CNXMAN_MAJOR_VERSION || |
1894 | + version.minor != CNXMAN_MINOR_VERSION || | |
1895 | + version.patch != CNXMAN_PATCH_VERSION) | |
1896 | + return -EINVAL; | |
b7b72b66 | 1897 | + |
bb1d8b11 AM |
1898 | + if (config_version == version.config) |
1899 | + return 0; | |
b7b72b66 | 1900 | + |
bb1d8b11 AM |
1901 | + config_version = version.config; |
1902 | + send_reconfigure(RECONFIG_PARAM_CONFIG_VERSION, config_version); | |
b7b72b66 AM |
1903 | + return 0; |
1904 | +} | |
1905 | + | |
bb1d8b11 | 1906 | +static int do_ioctl_get_members(unsigned long arg) |
c1c6733f | 1907 | +{ |
bb1d8b11 AM |
1908 | + struct cluster_node *node; |
1909 | + /* Kernel copies */ | |
1910 | + struct cl_cluster_node user_format_node; | |
1911 | + struct cl_cluster_nodelist user_format_nodelist; | |
1912 | + /* User space array ptr */ | |
1913 | + struct cl_cluster_node *user_node; | |
1914 | + struct list_head *nodelist; | |
1915 | + int num_nodes = 0; | |
c1c6733f | 1916 | + |
bb1d8b11 AM |
1917 | + if (arg == 0) |
1918 | + return cluster_members; | |
c1c6733f | 1919 | + |
bb1d8b11 AM |
1920 | + if (copy_from_user(&user_format_nodelist, (void __user *)arg, sizeof(struct cl_cluster_nodelist))) |
1921 | + return -EFAULT; | |
c1c6733f | 1922 | + |
bb1d8b11 | 1923 | + down(&cluster_members_lock); |
c1c6733f | 1924 | + |
bb1d8b11 AM |
1925 | + if (user_format_nodelist.max_members < cluster_members) { |
1926 | + up(&cluster_members_lock); | |
1927 | + return -E2BIG; | |
1928 | + } | |
c1c6733f | 1929 | + |
bb1d8b11 | 1930 | + user_node = user_format_nodelist.nodes; |
c1c6733f | 1931 | + |
bb1d8b11 AM |
1932 | + list_for_each(nodelist, &cluster_members_list) { |
1933 | + node = list_entry(nodelist, struct cluster_node, list); | |
1934 | + if (node->state == NODESTATE_MEMBER) { | |
1935 | + copy_to_usernode(node, &user_format_node); | |
1936 | + if (copy_to_user(user_node, &user_format_node, | |
1937 | + sizeof (struct cl_cluster_node))) { | |
1938 | + up(&cluster_members_lock); | |
1939 | + return -EFAULT; | |
1940 | + } | |
1941 | + user_node++; | |
1942 | + num_nodes++; | |
1943 | + } | |
1944 | + } | |
1945 | + up(&cluster_members_lock); | |
c1c6733f | 1946 | + |
bb1d8b11 AM |
1947 | + return num_nodes; |
1948 | +} | |
c1c6733f | 1949 | + |
bb1d8b11 AM |
1950 | +static int do_ioctl_get_all_members(unsigned long arg) |
1951 | +{ | |
1952 | + struct cluster_node *node; | |
1953 | + /* Kernel copies */ | |
1954 | + struct cl_cluster_node user_format_node; | |
1955 | + struct cl_cluster_nodelist user_format_nodelist; | |
1956 | + /* User space array ptr*/ | |
1957 | + struct cl_cluster_node *user_node; | |
1958 | + struct list_head *nodelist; | |
1959 | + int num_nodes = 0; | |
c1c6733f | 1960 | + |
bb1d8b11 AM |
1961 | + if (copy_from_user(&user_format_nodelist, (void __user *)arg, sizeof(struct cl_cluster_nodelist))) |
1962 | + return -EFAULT; | |
c1c6733f | 1963 | + |
bb1d8b11 | 1964 | + down(&cluster_members_lock); |
c1c6733f | 1965 | + |
bb1d8b11 | 1966 | + user_node = user_format_nodelist.nodes; |
c1c6733f | 1967 | + |
bb1d8b11 AM |
1968 | + list_for_each(nodelist, &cluster_members_list) { |
1969 | + node = list_entry(nodelist, struct cluster_node, list); | |
1970 | + if (arg) { | |
1971 | + copy_to_usernode(node, | |
1972 | + &user_format_node); | |
1973 | + | |
1974 | + if (copy_to_user(user_node, &user_format_node, | |
1975 | + sizeof (struct cl_cluster_node))) { | |
1976 | + up(&cluster_members_lock); | |
1977 | + return -EFAULT; | |
1978 | + } | |
1979 | + user_node++; | |
1980 | + if (--user_format_nodelist.max_members < 0) { | |
1981 | + num_nodes = -EFAULT; | |
1982 | + goto err_exit; | |
1983 | + } | |
1984 | + | |
1985 | + } | |
1986 | + num_nodes++; | |
b7b72b66 | 1987 | + } |
bb1d8b11 AM |
1988 | + err_exit: |
1989 | + up(&cluster_members_lock); | |
1990 | + | |
1991 | + return num_nodes; | |
c1c6733f AM |
1992 | +} |
1993 | + | |
bb1d8b11 AM |
1994 | + |
1995 | +static int do_ioctl_get_cluster(unsigned long arg) | |
c1c6733f | 1996 | +{ |
bb1d8b11 | 1997 | + struct cl_cluster_info __user *info; |
c1c6733f | 1998 | + |
bb1d8b11 | 1999 | + info = (struct cl_cluster_info *)arg; |
b7b72b66 | 2000 | + |
bb1d8b11 AM |
2001 | + if (copy_to_user(&info->number, &cluster_id, sizeof(cluster_id))) |
2002 | + return -EFAULT; | |
b7b72b66 | 2003 | + |
bb1d8b11 AM |
2004 | + if (copy_to_user(&info->name, cluster_name, strlen(cluster_name)+1)) |
2005 | + return -EFAULT; | |
b7b72b66 | 2006 | + |
bb1d8b11 AM |
2007 | + return 0; |
2008 | +} | |
b7b72b66 | 2009 | + |
bb1d8b11 AM |
2010 | +static int do_ioctl_get_node(unsigned long arg) |
2011 | +{ | |
2012 | + struct cluster_node *node; | |
2013 | + struct cl_cluster_node k_node, *u_node; | |
b7b72b66 | 2014 | + |
bb1d8b11 | 2015 | + u_node = (struct cl_cluster_node *) arg; |
b7b72b66 | 2016 | + |
bb1d8b11 AM |
2017 | + if (copy_from_user(&k_node, u_node, sizeof(struct cl_cluster_node))) |
2018 | + return -EFAULT; | |
c1c6733f | 2019 | + |
bb1d8b11 AM |
2020 | + if (!k_node.name[0]) { |
2021 | + if (k_node.node_id == 0) | |
2022 | + k_node.node_id = us->node_id; | |
2023 | + node = find_node_by_nodeid(k_node.node_id); | |
2024 | + } | |
2025 | + else | |
2026 | + node = find_node_by_name(k_node.name); | |
c1c6733f | 2027 | + |
bb1d8b11 AM |
2028 | + if (!node) |
2029 | + return -ENOENT; | |
c1c6733f | 2030 | + |
bb1d8b11 | 2031 | + copy_to_usernode(node, &k_node); |
b7b72b66 | 2032 | + |
bb1d8b11 AM |
2033 | + if (copy_to_user(u_node, &k_node, sizeof(struct cl_cluster_node))) |
2034 | + return -EFAULT; | |
c1c6733f | 2035 | + |
bb1d8b11 AM |
2036 | + return 0; |
2037 | +} | |
c1c6733f | 2038 | + |
bb1d8b11 AM |
2039 | +static int do_ioctl_set_expected(unsigned long arg) |
2040 | +{ | |
2041 | + struct list_head *nodelist; | |
2042 | + struct cluster_node *node; | |
2043 | + unsigned int total_votes; | |
2044 | + unsigned int newquorum; | |
c1c6733f | 2045 | + |
bb1d8b11 AM |
2046 | + if (!capable(CAP_CLUSTER)) |
2047 | + return -EPERM; | |
2048 | + if (arg == 0) | |
2049 | + return -EINVAL; | |
c1c6733f | 2050 | + |
bb1d8b11 | 2051 | + newquorum = calculate_quorum(1, arg, &total_votes); |
c1c6733f | 2052 | + |
bb1d8b11 AM |
2053 | + if (newquorum < total_votes / 2 |
2054 | + || newquorum > total_votes) { | |
2055 | + return -EINVAL; | |
2056 | + } | |
c1c6733f | 2057 | + |
bb1d8b11 AM |
2058 | + /* Now do it */ |
2059 | + down(&cluster_members_lock); | |
2060 | + list_for_each(nodelist, &cluster_members_list) { | |
2061 | + node = list_entry(nodelist, struct cluster_node, list); | |
2062 | + if (node->state == NODESTATE_MEMBER | |
2063 | + && node->expected_votes > arg) { | |
2064 | + node->expected_votes = arg; | |
c1c6733f AM |
2065 | + } |
2066 | + } | |
bb1d8b11 | 2067 | + up(&cluster_members_lock); |
c1c6733f | 2068 | + |
bb1d8b11 | 2069 | + recalculate_quorum(1); |
c1c6733f | 2070 | + |
bb1d8b11 AM |
2071 | + send_reconfigure(RECONFIG_PARAM_EXPECTED_VOTES, arg); |
2072 | + sm_member_update(cluster_is_quorate); | |
2073 | + | |
2074 | + return 0; | |
c1c6733f AM |
2075 | +} |
2076 | + | |
bb1d8b11 | 2077 | +static int do_ioctl_kill_node(unsigned long arg) |
c1c6733f | 2078 | +{ |
bb1d8b11 | 2079 | + struct cluster_node *node; |
c1c6733f | 2080 | + |
bb1d8b11 AM |
2081 | + if (!capable(CAP_CLUSTER)) |
2082 | + return -EPERM; | |
c1c6733f | 2083 | + |
c1c6733f | 2084 | + |
bb1d8b11 AM |
2085 | + if ((node = find_node_by_nodeid(arg)) == NULL) |
2086 | + return -EINVAL; | |
c1c6733f | 2087 | + |
bb1d8b11 AM |
2088 | + /* Can't kill us */ |
2089 | + if (node->us) | |
2090 | + return -EINVAL; | |
c1c6733f | 2091 | + |
bb1d8b11 AM |
2092 | + if (node->state != NODESTATE_MEMBER) |
2093 | + return -EINVAL; | |
c1c6733f | 2094 | + |
bb1d8b11 AM |
2095 | + /* Just in case it is alive, send a KILL message */ |
2096 | + send_kill(arg); | |
c1c6733f | 2097 | + |
bb1d8b11 AM |
2098 | + node->leave_reason = CLUSTER_LEAVEFLAG_KILLED; |
2099 | + a_node_just_died(node); | |
c1c6733f | 2100 | + |
bb1d8b11 AM |
2101 | + return 0; |
2102 | +} | |
c1c6733f | 2103 | + |
bb1d8b11 AM |
2104 | +static int do_ioctl_barrier(unsigned long arg) |
2105 | +{ | |
2106 | + struct cl_barrier_info info; | |
c1c6733f | 2107 | + |
bb1d8b11 AM |
2108 | + if (!capable(CAP_CLUSTER)) |
2109 | + return -EPERM; | |
c1c6733f | 2110 | + |
bb1d8b11 AM |
2111 | + if (copy_from_user(&info, (void *)arg, sizeof(info)) != 0) |
2112 | + return -EFAULT; | |
2113 | + | |
2114 | + switch (info.cmd) { | |
2115 | + case BARRIER_IOCTL_REGISTER: | |
2116 | + return kcl_barrier_register(info.name, | |
2117 | + info.flags, | |
2118 | + info.arg); | |
2119 | + case BARRIER_IOCTL_CHANGE: | |
2120 | + return kcl_barrier_setattr(info.name, | |
2121 | + info.flags, | |
2122 | + info.arg); | |
2123 | + case BARRIER_IOCTL_WAIT: | |
2124 | + return kcl_barrier_wait(info.name); | |
2125 | + case BARRIER_IOCTL_DELETE: | |
2126 | + return kcl_barrier_delete(info.name); | |
2127 | + default: | |
2128 | + return -EINVAL; | |
2129 | + } | |
c1c6733f AM |
2130 | +} |
2131 | + | |
bb1d8b11 | 2132 | +static int do_ioctl_islistening(unsigned long arg) |
c1c6733f | 2133 | +{ |
bb1d8b11 AM |
2134 | + DECLARE_WAITQUEUE(wq, current); |
2135 | + struct cl_listen_request rq; | |
2136 | + struct cluster_node *rem_node; | |
2137 | + int nodeid; | |
2138 | + int result; | |
2139 | + struct cl_waiting_listen_request *listen_request; | |
c1c6733f | 2140 | + |
bb1d8b11 AM |
2141 | + if (!arg) |
2142 | + return -EINVAL; | |
c1c6733f | 2143 | + |
bb1d8b11 AM |
2144 | + if (copy_from_user(&rq, (void *) arg, sizeof (rq)) != 0) |
2145 | + return -EFAULT; | |
c1c6733f | 2146 | + |
bb1d8b11 AM |
2147 | + nodeid = rq.nodeid; |
2148 | + if (!nodeid) | |
2149 | + nodeid = us->node_id; | |
c1c6733f | 2150 | + |
bb1d8b11 AM |
2151 | + rem_node = find_node_by_nodeid(nodeid); |
2152 | + | |
2153 | + /* Node not in the cluster */ | |
2154 | + if (!rem_node) | |
2155 | + return -ENOENT; | |
2156 | + | |
2157 | + if (rem_node->state != NODESTATE_MEMBER) | |
2158 | + return -ENOTCONN; | |
2159 | + | |
2160 | + /* If the request is for us then just look in the ports | |
2161 | + * array */ | |
2162 | + if (rem_node->us) | |
2163 | + return (port_array[rq.port] != 0) ? 1 : 0; | |
2164 | + | |
2165 | + /* For a remote node we need to send a request out */ | |
2166 | + | |
2167 | + /* If we are in transition then wait until we are not */ | |
2168 | + while (in_transition()) { | |
2169 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
2170 | + add_wait_queue(&socket_waitq, &wq); | |
2171 | + | |
2172 | + if (in_transition()) | |
2173 | + schedule(); | |
2174 | + | |
2175 | + set_task_state(current, TASK_RUNNING); | |
2176 | + remove_wait_queue(&socket_waitq, &wq); | |
c1c6733f | 2177 | + |
bb1d8b11 AM |
2178 | + if (signal_pending(current)) |
2179 | + return -EINTR; | |
c1c6733f | 2180 | + } |
c1c6733f | 2181 | + |
bb1d8b11 AM |
2182 | + /* Were we shut down before it completed ? */ |
2183 | + if (!atomic_read(&cnxman_running)) | |
2184 | + return -ENOTCONN; | |
c1c6733f | 2185 | + |
bb1d8b11 AM |
2186 | + listen_request = |
2187 | + kmalloc(sizeof (struct cl_waiting_listen_request), | |
2188 | + GFP_KERNEL); | |
2189 | + if (!listen_request) | |
2190 | + return -ENOMEM; | |
c1c6733f | 2191 | + |
bb1d8b11 AM |
2192 | + /* Build the request */ |
2193 | + listen_request->waiting = 1; | |
2194 | + listen_request->result = 0; | |
2195 | + listen_request->tag = current->pid; | |
2196 | + listen_request->nodeid = nodeid; | |
2197 | + init_waitqueue_head(&listen_request->waitq); | |
c1c6733f | 2198 | + |
bb1d8b11 AM |
2199 | + down(&listenreq_lock); |
2200 | + list_add(&listen_request->list, &listenreq_list); | |
2201 | + up(&listenreq_lock); | |
c1c6733f | 2202 | + |
bb1d8b11 AM |
2203 | + /* Now wait for the response to come back */ |
2204 | + send_listen_request(rq.nodeid, rq.port); | |
c1c6733f | 2205 | + |
bb1d8b11 AM |
2206 | + while (listen_request->waiting) { |
2207 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
2208 | + add_wait_queue(&listen_request->waitq, &wq); | |
c1c6733f | 2209 | + |
bb1d8b11 AM |
2210 | + if (listen_request->waiting) |
2211 | + schedule(); | |
c1c6733f | 2212 | + |
bb1d8b11 AM |
2213 | + set_task_state(current, TASK_RUNNING); |
2214 | + remove_wait_queue(&listen_request->waitq, &wq); | |
c1c6733f | 2215 | + |
bb1d8b11 AM |
2216 | + if (signal_pending(current)) { |
2217 | + result = -ERESTARTSYS; | |
2218 | + goto end_listen; | |
2219 | + } | |
2220 | + } | |
2221 | + result = listen_request->result; | |
c1c6733f | 2222 | + |
bb1d8b11 AM |
2223 | + end_listen: |
2224 | + down(&listenreq_lock); | |
2225 | + list_del(&listen_request->list); | |
2226 | + kfree(listen_request); | |
2227 | + up(&listenreq_lock); | |
2228 | + return result; | |
c1c6733f AM |
2229 | +} |
2230 | + | |
bb1d8b11 | 2231 | +static int do_ioctl_set_votes(unsigned long arg) |
c1c6733f | 2232 | +{ |
bb1d8b11 AM |
2233 | + unsigned int total_votes; |
2234 | + unsigned int newquorum; | |
2235 | + int saved_votes; | |
c1c6733f | 2236 | + |
bb1d8b11 AM |
2237 | + if (!capable(CAP_CLUSTER)) |
2238 | + return -EPERM; | |
c1c6733f | 2239 | + |
bb1d8b11 AM |
2240 | + /* Check votes is valid */ |
2241 | + saved_votes = us->votes; | |
2242 | + us->votes = arg; | |
c1c6733f | 2243 | + |
bb1d8b11 | 2244 | + newquorum = calculate_quorum(1, 0, &total_votes); |
c1c6733f | 2245 | + |
bb1d8b11 AM |
2246 | + if (newquorum < total_votes / 2 || newquorum > total_votes) { |
2247 | + us->votes = saved_votes; | |
2248 | + return -EINVAL; | |
c1c6733f | 2249 | + } |
c1c6733f | 2250 | + |
bb1d8b11 | 2251 | + recalculate_quorum(1); |
c1c6733f | 2252 | + |
bb1d8b11 AM |
2253 | + send_reconfigure(RECONFIG_PARAM_NODE_VOTES, arg); |
2254 | + | |
2255 | + return 0; | |
2256 | +} | |
2257 | + | |
2258 | +static int do_ioctl_pass_socket(unsigned long arg) | |
c1c6733f | 2259 | +{ |
bb1d8b11 AM |
2260 | + struct cl_passed_sock sock_info; |
2261 | + struct file *file; | |
2262 | + int error; | |
c1c6733f | 2263 | + |
bb1d8b11 AM |
2264 | + if (!capable(CAP_CLUSTER)) |
2265 | + return -EPERM; | |
c1c6733f | 2266 | + |
bb1d8b11 AM |
2267 | + if (atomic_read(&cnxman_running)) |
2268 | + return -EINVAL; | |
c1c6733f | 2269 | + |
bb1d8b11 | 2270 | + error = -EBADF; |
c1c6733f | 2271 | + |
bb1d8b11 AM |
2272 | + if (copy_from_user(&sock_info, (void *)arg, sizeof(sock_info))) |
2273 | + return -EFAULT; | |
c783755a | 2274 | + |
bb1d8b11 AM |
2275 | + file = fget(sock_info.fd); |
2276 | + if (file) { | |
2277 | + struct inode *inode = file->f_dentry->d_inode; | |
c1c6733f | 2278 | + |
bb1d8b11 AM |
2279 | + error = add_clsock(sock_info.multicast, |
2280 | + sock_info.number, SOCKET_I(inode), | |
2281 | + file); | |
2282 | + if (error) | |
2283 | + fput(file); | |
2284 | + } | |
2285 | + return error; | |
c1c6733f | 2286 | + |
bb1d8b11 | 2287 | +} |
c1c6733f | 2288 | + |
bb1d8b11 AM |
2289 | +static int do_ioctl_set_nodename(unsigned long arg) |
2290 | +{ | |
2291 | + if (!capable(CAP_CLUSTER)) | |
2292 | + return -EPERM; | |
2293 | + if (atomic_read(&cnxman_running)) | |
2294 | + return -EINVAL; | |
2295 | + if (strncpy_from_user(nodename, (void *)arg, MAX_CLUSTER_MEMBER_NAME_LEN) < 0) | |
2296 | + return -EFAULT; | |
2297 | + return 0; | |
2298 | +} | |
c1c6733f | 2299 | + |
bb1d8b11 AM |
2300 | +static int do_ioctl_set_nodeid(unsigned long arg) |
2301 | +{ | |
2302 | + int nodeid = (int)arg; | |
c1c6733f | 2303 | + |
bb1d8b11 AM |
2304 | + if (!capable(CAP_CLUSTER)) |
2305 | + return -EPERM; | |
2306 | + if (atomic_read(&cnxman_running)) | |
2307 | + return -EINVAL; | |
2308 | + if (nodeid < 0 || nodeid > 4096) | |
2309 | + return -EINVAL; | |
c1c6733f | 2310 | + |
bb1d8b11 AM |
2311 | + wanted_nodeid = (int)arg; |
2312 | + return 0; | |
2313 | +} | |
c1c6733f | 2314 | + |
bb1d8b11 AM |
2315 | +static int do_ioctl_join_cluster(unsigned long arg) |
2316 | +{ | |
2317 | + struct cl_join_cluster_info join_info; | |
c1c6733f | 2318 | + |
bb1d8b11 AM |
2319 | + if (!capable(CAP_CLUSTER)) |
2320 | + return -EPERM; | |
c1c6733f | 2321 | + |
bb1d8b11 AM |
2322 | + if (atomic_read(&cnxman_running)) |
2323 | + return -EALREADY; | |
c783755a | 2324 | + |
bb1d8b11 AM |
2325 | + if (copy_from_user(&join_info, (void *)arg, sizeof (struct cl_join_cluster_info) )) |
2326 | + return -EFAULT; | |
c1c6733f | 2327 | + |
bb1d8b11 AM |
2328 | + if (strlen(join_info.cluster_name) > MAX_CLUSTER_NAME_LEN) |
2329 | + return -EINVAL; | |
c1c6733f | 2330 | + |
bb1d8b11 AM |
2331 | + if (list_empty(&socket_list)) |
2332 | + return -ENOTCONN; | |
c1c6733f | 2333 | + |
bb1d8b11 AM |
2334 | + set_votes(join_info.votes, join_info.expected_votes); |
2335 | + cluster_id = generate_cluster_id(join_info.cluster_name); | |
2336 | + strncpy(cluster_name, join_info.cluster_name, MAX_CLUSTER_NAME_LEN); | |
2337 | + two_node = join_info.two_node; | |
2338 | + config_version = join_info.config_version; | |
c1c6733f | 2339 | + |
bb1d8b11 AM |
2340 | + quit_threads = 0; |
2341 | + acks_expected = 0; | |
2342 | + init_completion(&cluster_thread_comp); | |
2343 | + init_completion(&member_thread_comp); | |
2344 | + if (allocate_nodeid_array()) | |
2345 | + return -ENOMEM; | |
c1c6733f | 2346 | + |
bb1d8b11 AM |
2347 | + kcluster_pid = kernel_thread(cluster_kthread, NULL, 0); |
2348 | + if (kcluster_pid < 0) | |
2349 | + return kcluster_pid; | |
c1c6733f | 2350 | + |
bb1d8b11 AM |
2351 | + wait_for_completion(&cluster_thread_comp); |
2352 | + init_completion(&cluster_thread_comp); | |
b7b72b66 | 2353 | + |
bb1d8b11 | 2354 | + atomic_set(&cnxman_running, 1); |
b7b72b66 | 2355 | + |
bb1d8b11 AM |
2356 | + /* Make sure we have a node name */ |
2357 | + if (nodename[0] == '\0') | |
2358 | + strcpy(nodename, system_utsname.nodename); | |
c1c6733f | 2359 | + |
bb1d8b11 AM |
2360 | + membership_pid = start_membership_services(kcluster_pid); |
2361 | + if (membership_pid < 0) { | |
2362 | + quit_threads = 1; | |
2363 | + wait_for_completion(&cluster_thread_comp); | |
2364 | + init_completion(&member_thread_comp); | |
2365 | + return membership_pid; | |
c1c6733f AM |
2366 | + } |
2367 | + | |
bb1d8b11 AM |
2368 | + sm_start(); |
2369 | + return 0; | |
2370 | +} | |
c1c6733f | 2371 | + |
bb1d8b11 AM |
2372 | +static int do_ioctl_leave_cluster(unsigned long leave_flags) |
2373 | +{ | |
2374 | + if (!capable(CAP_CLUSTER)) | |
2375 | + return -EPERM; | |
c1c6733f | 2376 | + |
bb1d8b11 AM |
2377 | + if (!atomic_read(&cnxman_running)) |
2378 | + return -ENOTCONN; | |
c783755a | 2379 | + |
bb1d8b11 AM |
2380 | + if (in_transition()) |
2381 | + return -EBUSY; | |
c783755a | 2382 | + |
bb1d8b11 AM |
2383 | + /* Ignore the use count if FORCE is set */ |
2384 | + if (!(leave_flags & CLUSTER_LEAVEFLAG_FORCE)) { | |
2385 | + if (atomic_read(&use_count)) | |
2386 | + return -ENOTCONN; | |
c783755a AM |
2387 | + } |
2388 | + | |
bb1d8b11 AM |
2389 | + us->leave_reason = leave_flags; |
2390 | + quit_threads = 1; | |
2391 | + wake_up_interruptible(&cnxman_waitq); | |
c1c6733f | 2392 | + |
bb1d8b11 AM |
2393 | + wait_for_completion(&cluster_thread_comp); |
2394 | + atomic_set(&use_count, 0); | |
2395 | + return 0; | |
2396 | +} | |
c1c6733f | 2397 | + |
bb1d8b11 AM |
2398 | +static int cl_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) |
2399 | +{ | |
2400 | + int err = -EOPNOTSUPP; | |
2401 | + struct list_head *proclist; | |
2402 | + struct list_head *tmp; | |
2403 | + struct notify_struct *notify; | |
2404 | + struct cl_version cnxman_version; | |
c1c6733f | 2405 | + |
bb1d8b11 AM |
2406 | + switch (cmd) { |
2407 | + /* Process requests notification of cluster events */ | |
2408 | + case SIOCCLUSTER_NOTIFY: | |
2409 | + notify = kmalloc(sizeof (struct notify_struct), GFP_KERNEL); | |
2410 | + if (!notify) | |
2411 | + return -ENOMEM; | |
2412 | + notify->pid = current->pid; | |
2413 | + notify->signal = arg; | |
2414 | + down(&event_listener_lock); | |
2415 | + list_add(¬ify->list, &event_listener_list); | |
2416 | + up(&event_listener_lock); | |
2417 | + err = 0; | |
2418 | + break; | |
c1c6733f | 2419 | + |
bb1d8b11 AM |
2420 | + /* Process is no longer interested cluster events */ |
2421 | + case SIOCCLUSTER_REMOVENOTIFY: | |
2422 | + err = EINVAL; | |
c1c6733f | 2423 | + |
bb1d8b11 AM |
2424 | + down(&event_listener_lock); |
2425 | + list_for_each_safe(proclist, tmp, &event_listener_list) { | |
2426 | + notify = | |
2427 | + list_entry(proclist, struct notify_struct, list); | |
2428 | + if (notify->pid == current->pid) { | |
2429 | + list_del(¬ify->list); | |
2430 | + kfree(notify); | |
2431 | + err = 0; | |
2432 | + } | |
2433 | + } | |
2434 | + up(&event_listener_lock); | |
2435 | + break; | |
c1c6733f | 2436 | + |
bb1d8b11 AM |
2437 | + /* Return the cnxman version number */ |
2438 | + case SIOCCLUSTER_GET_VERSION: | |
2439 | + if (!arg) | |
2440 | + return -EINVAL; | |
2441 | + err = 0; | |
2442 | + cnxman_version.major = CNXMAN_MAJOR_VERSION; | |
2443 | + cnxman_version.minor = CNXMAN_MINOR_VERSION; | |
2444 | + cnxman_version.patch = CNXMAN_PATCH_VERSION; | |
2445 | + cnxman_version.config = config_version; | |
2446 | + if (copy_to_user((void *) arg, &cnxman_version, | |
2447 | + sizeof (struct cl_version))) { | |
2448 | + return -EFAULT; | |
2449 | + } | |
2450 | + break; | |
c1c6733f | 2451 | + |
bb1d8b11 AM |
2452 | + /* Set the cnxman config version number */ |
2453 | + case SIOCCLUSTER_SET_VERSION: | |
2454 | + err = do_ioctl_set_version(arg); | |
2455 | + break; | |
c1c6733f | 2456 | + |
bb1d8b11 AM |
2457 | + /* Return the active membership list */ |
2458 | + case SIOCCLUSTER_GETMEMBERS: | |
2459 | + err = do_ioctl_get_members(arg); | |
2460 | + break; | |
c1c6733f | 2461 | + |
bb1d8b11 AM |
2462 | + /* Return the full membership list include dead nodes */ |
2463 | + case SIOCCLUSTER_GETALLMEMBERS: | |
2464 | + err = do_ioctl_get_all_members(arg); | |
2465 | + break; | |
c1c6733f | 2466 | + |
bb1d8b11 AM |
2467 | + case SIOCCLUSTER_GETNODE: |
2468 | + err = do_ioctl_get_node(arg); | |
2469 | + break; | |
c783755a | 2470 | + |
bb1d8b11 AM |
2471 | + case SIOCCLUSTER_GETCLUSTER: |
2472 | + err = do_ioctl_get_cluster(arg); | |
2473 | + break; | |
c783755a | 2474 | + |
bb1d8b11 AM |
2475 | + case SIOCCLUSTER_ISQUORATE: |
2476 | + return cluster_is_quorate; | |
c1c6733f | 2477 | + |
bb1d8b11 AM |
2478 | + case SIOCCLUSTER_ISACTIVE: |
2479 | + return atomic_read(&cnxman_running); | |
c1c6733f | 2480 | + |
bb1d8b11 AM |
2481 | + case SIOCCLUSTER_SETEXPECTED_VOTES: |
2482 | + err = do_ioctl_set_expected(arg); | |
2483 | + break; | |
c1c6733f | 2484 | + |
bb1d8b11 AM |
2485 | + /* Change the number of votes for this node */ |
2486 | + case SIOCCLUSTER_SET_VOTES: | |
2487 | + err = do_ioctl_set_votes(arg); | |
2488 | + break; | |
c1c6733f | 2489 | + |
bb1d8b11 AM |
2490 | + /* Return 1 if the specified node is listening on a given port */ |
2491 | + case SIOCCLUSTER_ISLISTENING: | |
2492 | + err = do_ioctl_islistening(arg); | |
2493 | + break; | |
c1c6733f | 2494 | + |
bb1d8b11 AM |
2495 | + /* Forcibly kill a node */ |
2496 | + case SIOCCLUSTER_KILLNODE: | |
2497 | + err = do_ioctl_kill_node(arg); | |
2498 | + break; | |
c1c6733f | 2499 | + |
bb1d8b11 AM |
2500 | + case SIOCCLUSTER_GET_JOINCOUNT: |
2501 | + if (!capable(CAP_CLUSTER)) | |
2502 | + return -EPERM; | |
2503 | + else | |
2504 | + return atomic_read(&use_count); | |
2505 | + | |
2506 | + /* ioctl interface to the barrier system */ | |
2507 | + case SIOCCLUSTER_BARRIER: | |
2508 | + err = do_ioctl_barrier(arg); | |
2509 | + break; | |
2510 | + | |
2511 | + case SIOCCLUSTER_PASS_SOCKET: | |
2512 | + if (sock->sk->sk_protocol != CLPROTO_MASTER) | |
2513 | + err = -EOPNOTSUPP; | |
2514 | + else | |
2515 | + err = do_ioctl_pass_socket(arg); | |
2516 | + break; | |
c1c6733f | 2517 | + |
bb1d8b11 AM |
2518 | + case SIOCCLUSTER_SET_NODENAME: |
2519 | + if (sock->sk->sk_protocol != CLPROTO_MASTER) | |
2520 | + err = -EOPNOTSUPP; | |
2521 | + else | |
2522 | + err = do_ioctl_set_nodename(arg); | |
2523 | + break; | |
c1c6733f | 2524 | + |
bb1d8b11 AM |
2525 | + case SIOCCLUSTER_SET_NODEID: |
2526 | + if (sock->sk->sk_protocol != CLPROTO_MASTER) | |
2527 | + err = -EOPNOTSUPP; | |
2528 | + else | |
2529 | + err = do_ioctl_set_nodeid(arg); | |
2530 | + break; | |
c1c6733f | 2531 | + |
bb1d8b11 AM |
2532 | + case SIOCCLUSTER_JOIN_CLUSTER: |
2533 | + if (sock->sk->sk_protocol != CLPROTO_MASTER) | |
2534 | + err = -EOPNOTSUPP; | |
2535 | + else | |
2536 | + err = do_ioctl_join_cluster(arg); | |
2537 | + break; | |
c1c6733f | 2538 | + |
bb1d8b11 AM |
2539 | + case SIOCCLUSTER_LEAVE_CLUSTER: |
2540 | + err = do_ioctl_leave_cluster(arg); | |
2541 | + break; | |
2542 | + | |
2543 | + default: | |
2544 | + err = sm_ioctl(sock, cmd, arg); | |
2545 | + } | |
2546 | + return err; | |
c1c6733f AM |
2547 | +} |
2548 | + | |
bb1d8b11 | 2549 | +static int cl_shutdown(struct socket *sock, int how) |
c1c6733f | 2550 | +{ |
bb1d8b11 AM |
2551 | + struct sock *sk = sock->sk; |
2552 | + int err = -ENOTCONN; | |
c1c6733f | 2553 | + |
bb1d8b11 | 2554 | + lock_sock(sk); |
c1c6733f | 2555 | + |
bb1d8b11 AM |
2556 | + if (sock->state == SS_UNCONNECTED) |
2557 | + goto out; | |
c1c6733f | 2558 | + |
bb1d8b11 AM |
2559 | + err = 0; |
2560 | + if (sock->state == SS_DISCONNECTING) | |
2561 | + goto out; | |
c1c6733f | 2562 | + |
bb1d8b11 | 2563 | + err = -EINVAL; |
c1c6733f | 2564 | + |
bb1d8b11 AM |
2565 | + if (how != SHUTDOWN_MASK) |
2566 | + goto out; | |
c1c6733f | 2567 | + |
bb1d8b11 AM |
2568 | + sk->sk_shutdown = how; |
2569 | + err = 0; | |
c1c6733f | 2570 | + |
bb1d8b11 AM |
2571 | + out: |
2572 | + release_sock(sk); | |
c1c6733f | 2573 | + |
bb1d8b11 | 2574 | + return err; |
c1c6733f AM |
2575 | +} |
2576 | + | |
bb1d8b11 AM |
2577 | + |
2578 | +/* We'll be giving out reward points next... */ | |
2579 | +/* Send the packet and save a copy in case someone loses theirs. Should be | |
2580 | + * protected by the send mutexphore */ | |
2581 | +static int __send_and_save(struct cl_comms_socket *csock, struct msghdr *msg, | |
2582 | + struct kvec *vec, int veclen, | |
2583 | + int size, int needack) | |
c1c6733f | 2584 | +{ |
bb1d8b11 AM |
2585 | + int result; |
2586 | + struct kvec save_vectors[veclen]; | |
c1c6733f | 2587 | + |
bb1d8b11 AM |
2588 | + /* Save a copy of the IO vectors as sendmsg mucks around with them and |
2589 | + * we might want to send the same stuff out more than once (for different | |
2590 | + * interfaces) | |
2591 | + */ | |
2592 | + memcpy(save_vectors, vec, | |
2593 | + sizeof (struct kvec) * veclen); | |
c1c6733f | 2594 | + |
bb1d8b11 | 2595 | + result = kernel_sendmsg(csock->sock, msg, vec, veclen, size); |
c1c6733f | 2596 | + |
bb1d8b11 | 2597 | + if (result >= 0 && acks_expected && needack) { |
c1c6733f | 2598 | + |
bb1d8b11 AM |
2599 | + /* Start retransmit timer if it didn't go */ |
2600 | + if (result == 0) { | |
2601 | + start_short_timer(); | |
2602 | + } | |
2603 | + else { | |
2604 | + resend_delay = 1; | |
2605 | + } | |
c1c6733f AM |
2606 | + } |
2607 | + | |
bb1d8b11 AM |
2608 | + /* Restore IOVs */ |
2609 | + memcpy(vec, save_vectors, | |
2610 | + sizeof (struct kvec) * veclen); | |
c1c6733f | 2611 | + |
bb1d8b11 | 2612 | + return result; |
c1c6733f AM |
2613 | +} |
2614 | + | |
bb1d8b11 | 2615 | +static void resend_last_message() |
c1c6733f | 2616 | +{ |
c1c6733f | 2617 | + struct msghdr msg; |
bb1d8b11 AM |
2618 | + struct kvec vec[1]; |
2619 | + int result; | |
c1c6733f | 2620 | + |
bb1d8b11 AM |
2621 | + P_COMMS("%ld resending last message: %d bytes: port=%d, cmd=%d\n", |
2622 | + jiffies, saved_msg_len, saved_msg_buffer[0], | |
2623 | + saved_msg_buffer[6]); | |
2624 | + | |
2625 | + /* Assume there is something wrong with the last interface */ | |
2626 | + current_interface = get_next_interface(current_interface); | |
2627 | + if (num_interfaces > 1) | |
2628 | + printk(KERN_WARNING CMAN_NAME ": Now using interface %d\n", | |
2629 | + current_interface->number); | |
c1c6733f | 2630 | + |
bb1d8b11 AM |
2631 | + vec[0].iov_base = saved_msg_buffer; |
2632 | + vec[0].iov_len = saved_msg_len; | |
c1c6733f AM |
2633 | + |
2634 | + memset(&msg, 0, sizeof (msg)); | |
bb1d8b11 AM |
2635 | + msg.msg_name = ¤t_interface->saddr; |
2636 | + msg.msg_namelen = current_interface->addr_len; | |
c1c6733f | 2637 | + |
bb1d8b11 | 2638 | + result = kernel_sendmsg(current_interface->sock, &msg, vec, 1, saved_msg_len); |
c1c6733f | 2639 | + |
bb1d8b11 AM |
2640 | + if (result < 0) |
2641 | + printk(KERN_ERR CMAN_NAME ": resend failed: %d\n", result); | |
c1c6733f | 2642 | + |
bb1d8b11 AM |
2643 | + /* Try indefinitely to send this, the backlog must die down eventually |
2644 | + * !? */ | |
2645 | + if (result == 0) | |
2646 | + start_short_timer(); | |
2647 | + | |
2648 | + /* Send succeeded, continue waiting for ACKS */ | |
2649 | + if (result > 0) | |
2650 | + start_ack_timer(); | |
c1c6733f | 2651 | + |
c1c6733f AM |
2652 | +} |
2653 | + | |
bb1d8b11 AM |
2654 | +static int cl_recvmsg(struct kiocb *iocb, struct socket *sock, |
2655 | + struct msghdr *msg, size_t size, int flags) | |
c1c6733f | 2656 | +{ |
bb1d8b11 AM |
2657 | + struct sock *sk = sock->sk; |
2658 | + struct sockaddr_cl *sin = (struct sockaddr_cl *) msg->msg_name; | |
2659 | + struct sk_buff *skb; | |
2660 | + struct cb_info *cbinfo; | |
2661 | + int copied, err = 0; | |
c1c6733f | 2662 | + |
bb1d8b11 AM |
2663 | + /* Socket was notified of shutdown, remove any pending skbs and return |
2664 | + * EOF */ | |
2665 | + if (!atomic_read(&cnxman_running)) { | |
2666 | + while ((skb = skb_recv_datagram(sk, flags, MSG_DONTWAIT, &err))) | |
2667 | + skb_free_datagram(sk, skb); | |
2668 | + return 0; /* cnxman has left the building */ | |
2669 | + } | |
c1c6733f | 2670 | + |
bb1d8b11 AM |
2671 | + /* Generic datagram code does most of the work. If the user is not |
2672 | + * interested in OOB messages then ignore them */ | |
2673 | + do { | |
2674 | + skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); | |
2675 | + if (!skb) | |
2676 | + goto out; | |
c1c6733f | 2677 | + |
bb1d8b11 | 2678 | + cbinfo = (struct cb_info *)skb->cb; |
c1c6733f | 2679 | + |
bb1d8b11 AM |
2680 | + /* If it is OOB and the user doesn't want it, then throw it away. */ |
2681 | + if (cbinfo->oob && !(flags & MSG_OOB)) { | |
2682 | + skb_free_datagram(sk, skb); | |
2683 | + | |
2684 | + /* If we peeked (?) an OOB but the user doesn't want it | |
2685 | + then we need to discard it or we'll loop forever */ | |
2686 | + if (flags & MSG_PEEK) { | |
2687 | + skb = skb_recv_datagram(sk, flags & ~MSG_PEEK, | |
2688 | + MSG_DONTWAIT, &err); | |
2689 | + if (skb) | |
2690 | + skb_free_datagram(sk, skb); | |
2691 | + } | |
2692 | + } | |
2693 | + else | |
2694 | + break; | |
c1c6733f | 2695 | + } |
bb1d8b11 | 2696 | + while (cbinfo->oob && !(flags & MSG_OOB)); |
c1c6733f | 2697 | + |
bb1d8b11 AM |
2698 | + copied = skb->len; |
2699 | + if (copied > size) { | |
2700 | + copied = size; | |
2701 | + msg->msg_flags |= MSG_TRUNC; | |
c1c6733f | 2702 | + } |
bb1d8b11 | 2703 | + err = memcpy_toiovec(msg->msg_iov, skb->data, copied); |
c1c6733f | 2704 | + |
bb1d8b11 AM |
2705 | + if (err) |
2706 | + goto out_free; | |
c1c6733f | 2707 | + |
bb1d8b11 AM |
2708 | + if (msg->msg_name && msg->msg_namelen) { |
2709 | + memset(msg->msg_name, 0, msg->msg_namelen); | |
c1c6733f | 2710 | + |
bb1d8b11 AM |
2711 | + if (msg->msg_namelen >= sizeof (struct sockaddr_cl)) { |
2712 | + | |
2713 | + /* Nodeid is in native byte order - anything else is just | |
2714 | + * perverse */ | |
2715 | + sin->scl_nodeid = cbinfo->orig_nodeid; | |
2716 | + } | |
2717 | + msg->msg_namelen = sizeof (struct sockaddr_cl); | |
2718 | + sin->scl_port = cbinfo->orig_port; | |
2719 | + } | |
2720 | + | |
2721 | + if (cbinfo->oob) { | |
2722 | + msg->msg_flags |= MSG_OOB; | |
2723 | + } | |
2724 | + | |
2725 | + sock_recv_timestamp(msg, sk, skb); | |
c1c6733f | 2726 | + |
bb1d8b11 | 2727 | + err = copied; |
c1c6733f | 2728 | + |
bb1d8b11 AM |
2729 | + out_free: |
2730 | + skb_free_datagram(sk, skb); | |
c1c6733f | 2731 | + |
bb1d8b11 AM |
2732 | + out: |
2733 | + return err; | |
c1c6733f AM |
2734 | +} |
2735 | + | |
bb1d8b11 AM |
2736 | +/* Send a message out on all interfaces */ |
2737 | +static int send_to_all_ints(int nodeid, struct msghdr *our_msg, | |
2738 | + struct kvec *vec, int veclen, int size, int flags) | |
c1c6733f | 2739 | +{ |
bb1d8b11 AM |
2740 | + struct sockaddr_in6 daddr; |
2741 | + struct cl_comms_socket *clsock; | |
2742 | + int result = 0; | |
c1c6733f | 2743 | + |
bb1d8b11 | 2744 | + our_msg->msg_name = &daddr; |
c1c6733f | 2745 | + |
bb1d8b11 | 2746 | + list_for_each_entry(clsock, &socket_list, list) { |
c1c6733f | 2747 | + |
bb1d8b11 AM |
2748 | + /* Don't send out a recv-only socket */ |
2749 | + if (!clsock->recv_only) { | |
c1c6733f | 2750 | + |
bb1d8b11 AM |
2751 | + /* For temporary node IDs send to the node's real IP address */ |
2752 | + if (nodeid < 0) { | |
2753 | + get_addr_from_temp_nodeid(nodeid, (char *)&daddr, &our_msg->msg_namelen); | |
2754 | + } | |
2755 | + else { | |
2756 | + memcpy(&daddr, &clsock->saddr, clsock->addr_len); | |
2757 | + our_msg->msg_namelen = clsock->addr_len; | |
2758 | + } | |
c1c6733f | 2759 | + |
bb1d8b11 AM |
2760 | + result = __send_and_save(clsock, our_msg, vec, veclen, |
2761 | + size + sizeof (struct cl_protheader), | |
2762 | + !(flags & MSG_NOACK)); | |
2763 | + } | |
2764 | + } | |
2765 | + return result; | |
c1c6733f AM |
2766 | +} |
2767 | + | |
bb1d8b11 AM |
2768 | + |
2769 | +/* Internal common send message routine */ | |
2770 | +static int __sendmsg(struct socket *sock, struct msghdr *msg, | |
2771 | + struct kvec *vec, int veclen, int size, | |
2772 | + unsigned char port) | |
c1c6733f | 2773 | +{ |
bb1d8b11 AM |
2774 | + int result = 0, i; |
2775 | + int flags = msg->msg_flags; | |
2776 | + struct msghdr our_msg; | |
2777 | + struct sockaddr_cl *caddr = msg->msg_name; | |
2778 | + struct cl_protheader header; | |
2779 | + struct kvec vectors[veclen + 1]; | |
2780 | + unsigned char srcport; | |
2781 | + int nodeid = 0; | |
c1c6733f | 2782 | + |
bb1d8b11 AM |
2783 | + if (size > MAX_CLUSTER_MESSAGE) |
2784 | + return -EINVAL; | |
2785 | + if (!atomic_read(&cnxman_running)) | |
2786 | + return -ENOTCONN; | |
c1c6733f | 2787 | + |
bb1d8b11 AM |
2788 | + if (caddr) |
2789 | + nodeid = caddr->scl_nodeid; | |
c1c6733f | 2790 | + |
bb1d8b11 AM |
2791 | + /* Check that the node id (if present) is valid */ |
2792 | + if (msg->msg_namelen && (!find_node_by_nodeid(nodeid) && | |
2793 | + !is_valid_temp_nodeid(nodeid))) { | |
2794 | + return -ENOTCONN; | |
2795 | + } | |
2796 | + | |
2797 | + /* If there's no sending client socket then the source | |
2798 | + port is 0: "us" */ | |
2799 | + if (sock) { | |
2800 | + struct cluster_sock *csock = cluster_sk(sock->sk); | |
2801 | + srcport = csock->port; | |
c1c6733f AM |
2802 | + } |
2803 | + else { | |
bb1d8b11 | 2804 | + srcport = 0; |
c1c6733f AM |
2805 | + } |
2806 | + | |
bb1d8b11 AM |
2807 | + /* We can only have one send outstanding at a time so we might as well |
2808 | + * lock the whole send mechanism */ | |
2809 | + down(&send_lock); | |
c1c6733f | 2810 | + |
bb1d8b11 AM |
2811 | + while ((port > HIGH_PROTECTED_PORT |
2812 | + && (!cluster_is_quorate || in_transition())) | |
2813 | + || (acks_expected > 0 && !(msg->msg_flags & MSG_NOACK))) { | |
c1c6733f | 2814 | + |
bb1d8b11 AM |
2815 | + DECLARE_WAITQUEUE(wq, current); |
2816 | + struct task_struct *tsk = current; | |
c1c6733f | 2817 | + |
bb1d8b11 AM |
2818 | + if (flags & MSG_DONTWAIT) { |
2819 | + up(&send_lock); | |
2820 | + return -EAGAIN; | |
2821 | + } | |
c1c6733f | 2822 | + |
bb1d8b11 AM |
2823 | + if (current->pid == kcluster_pid) { |
2824 | + P_COMMS | |
2825 | + ("Tried to make kclusterd wait, port=%d, acks_count=%d, expected=%d\n", | |
2826 | + port, ack_count, acks_expected); | |
2827 | + up(&send_lock); | |
2828 | + return -EAGAIN; | |
2829 | + } | |
c1c6733f | 2830 | + |
bb1d8b11 AM |
2831 | + P_COMMS("%s process waiting. acks=%d, expected=%d\n", tsk->comm, |
2832 | + ack_count, acks_expected); | |
c1c6733f AM |
2833 | + |
2834 | + set_task_state(tsk, TASK_INTERRUPTIBLE); | |
2835 | + add_wait_queue(&socket_waitq, &wq); | |
2836 | + | |
bb1d8b11 AM |
2837 | + if ((port > HIGH_PROTECTED_PORT |
2838 | + && (!cluster_is_quorate || in_transition())) | |
2839 | + || (acks_expected > 0)) { | |
2840 | + | |
2841 | + up(&send_lock); | |
c1c6733f | 2842 | + schedule(); |
bb1d8b11 | 2843 | + down(&send_lock); |
c1c6733f AM |
2844 | + } |
2845 | + | |
2846 | + set_task_state(tsk, TASK_RUNNING); | |
2847 | + remove_wait_queue(&socket_waitq, &wq); | |
bb1d8b11 AM |
2848 | + |
2849 | + /* Going down */ | |
2850 | + if (quit_threads) { | |
2851 | + up(&send_lock); | |
2852 | + return -ENOTCONN; | |
2853 | + } | |
2854 | + | |
2855 | + if (signal_pending(current)) { | |
2856 | + up(&send_lock); | |
2857 | + return -ERESTARTSYS; | |
2858 | + } | |
2859 | + | |
2860 | + /* Were we shut down in the meantime ? */ | |
2861 | + if (!atomic_read(&cnxman_running)) { | |
2862 | + up(&send_lock); | |
2863 | + return -ENOTCONN; | |
2864 | + } | |
2865 | + | |
c1c6733f | 2866 | + } |
c1c6733f | 2867 | + |
bb1d8b11 | 2868 | + memset(&our_msg, 0, sizeof (our_msg)); |
c1c6733f AM |
2869 | + |
2870 | + /* Build the header */ | |
bb1d8b11 AM |
2871 | + header.tgtport = port; |
2872 | + header.srcport = srcport; | |
2873 | + header.flags = msg->msg_flags; | |
2874 | + header.cluster = cpu_to_le16(cluster_id); | |
2875 | + header.srcid = us ? cpu_to_le32(us->node_id) : 0; | |
2876 | + header.tgtid = caddr ? cpu_to_le32(nodeid) : 0; | |
c1c6733f | 2877 | + |
bb1d8b11 AM |
2878 | + ++cur_seq; |
2879 | + header.seq = cpu_to_le16(cur_seq); | |
2880 | + header.ack = 0; | |
c1c6733f | 2881 | + |
bb1d8b11 AM |
2882 | + if (header.tgtid) { |
2883 | + struct cluster_node *remnode; | |
c1c6733f | 2884 | + |
bb1d8b11 AM |
2885 | + remnode = find_node_by_nodeid(nodeid); |
2886 | + if (remnode) { | |
2887 | + header.ack = cpu_to_le16(remnode->last_seq_recv); | |
2888 | + } | |
c1c6733f AM |
2889 | + } |
2890 | + | |
bb1d8b11 AM |
2891 | + /* Set the MULTICAST flag on messages with no particular destination */ |
2892 | + if (!msg->msg_namelen) { | |
2893 | + header.flags |= MSG_MULTICAST; | |
2894 | + header.tgtid = 0; | |
2895 | + } | |
c783755a | 2896 | + |
bb1d8b11 AM |
2897 | + /* Loopback shortcut */ |
2898 | + if (nodeid == us->node_id && nodeid != 0) { | |
c1c6733f | 2899 | + |
bb1d8b11 AM |
2900 | + up(&send_lock); |
2901 | + header.flags |= MSG_NOACK; /* Don't ack it! */ | |
c1c6733f | 2902 | + |
bb1d8b11 AM |
2903 | + return send_to_user_port(NULL, &header, msg, vec, veclen, size); |
2904 | + } | |
c1c6733f | 2905 | + |
bb1d8b11 AM |
2906 | + /* Copy the existing kvecs into our array and add the header on at the |
2907 | + * beginning */ | |
2908 | + vectors[0].iov_base = &header; | |
2909 | + vectors[0].iov_len = sizeof (header); | |
2910 | + for (i = 0; i < veclen; i++) { | |
2911 | + vectors[i + 1] = vec[i]; | |
2912 | + } | |
c1c6733f | 2913 | + |
c1c6733f | 2914 | + |
bb1d8b11 AM |
2915 | + /* Work out how many ACKS are wanted - *don't* reset acks_expected to |
2916 | + * zero if no acks are required as an ACK-needed message may still be | |
2917 | + * outstanding */ | |
2918 | + if (!(msg->msg_flags & MSG_NOACK)) { | |
2919 | + if (msg->msg_namelen) | |
2920 | + acks_expected = 1; /* Unicast */ | |
2921 | + else | |
2922 | + acks_expected = max(cluster_members - 1, 0); | |
c1c6733f | 2923 | + |
bb1d8b11 | 2924 | + } |
c1c6733f | 2925 | + |
bb1d8b11 AM |
2926 | + P_COMMS |
2927 | + ("Sending message - tgt=%d port %d required %d acks, seq=%d, flags=%x\n", | |
2928 | + nodeid, header.port, | |
2929 | + (msg->msg_flags & MSG_NOACK) ? 0 : acks_expected, | |
2930 | + le16_to_cpu(header.seq), header.flags); | |
c1c6733f | 2931 | + |
bb1d8b11 AM |
2932 | + /* Don't include temp nodeids in the message itself */ |
2933 | + if (header.tgtid < 0) | |
2934 | + header.tgtid = 0; | |
c1c6733f | 2935 | + |
bb1d8b11 AM |
2936 | + /* For non-member sends we use all the interfaces */ |
2937 | + if ((nodeid < 0) || (flags & MSG_ALLINT)) { | |
c1c6733f | 2938 | + |
bb1d8b11 AM |
2939 | + result = send_to_all_ints(nodeid, &our_msg, vectors, veclen+1, |
2940 | + size, msg->msg_flags); | |
2941 | + } | |
2942 | + else { | |
2943 | + /* Send to only the current socket - resends will use the | |
2944 | + * others if necessary */ | |
2945 | + our_msg.msg_name = ¤t_interface->saddr; | |
2946 | + our_msg.msg_namelen = current_interface->addr_len; | |
c1c6733f | 2947 | + |
bb1d8b11 AM |
2948 | + result = |
2949 | + __send_and_save(current_interface, &our_msg, | |
2950 | + vectors, veclen+1, | |
2951 | + size + sizeof (header), | |
2952 | + !(msg->msg_flags & MSG_NOACK)); | |
2953 | + } | |
c1c6733f | 2954 | + |
bb1d8b11 AM |
2955 | + /* Make a note in each nodes' structure that it has been sent a message |
2956 | + * so we can see which ones went astray */ | |
2957 | + if (!(flags & MSG_NOACK) && nodeid >= 0) { | |
2958 | + if (msg->msg_namelen) { | |
2959 | + struct cluster_node *node; | |
2960 | + | |
2961 | + node = find_node_by_nodeid(le32_to_cpu(header.tgtid)); | |
2962 | + if (node) | |
2963 | + node->last_seq_sent = cur_seq; | |
c1c6733f AM |
2964 | + } |
2965 | + else { | |
bb1d8b11 AM |
2966 | + struct cluster_node *node; |
2967 | + struct list_head *nodelist; | |
2968 | + | |
2969 | + list_for_each(nodelist, &cluster_members_list) { | |
2970 | + node = | |
2971 | + list_entry(nodelist, struct cluster_node, | |
2972 | + list); | |
2973 | + if (node->state == NODESTATE_MEMBER) { | |
2974 | + node->last_seq_sent = cur_seq; | |
2975 | + } | |
c1c6733f AM |
2976 | + } |
2977 | + } | |
2978 | + } | |
c1c6733f | 2979 | + |
bb1d8b11 AM |
2980 | + /* if the client wants a broadcast message sending back to itself |
2981 | + then loop it back */ | |
2982 | + if (nodeid == 0 && (flags & MSG_BCASTSELF)) { | |
2983 | + header.flags |= MSG_NOACK; /* Don't ack it! */ | |
c1c6733f | 2984 | + |
bb1d8b11 | 2985 | + result = send_to_user_port(NULL, &header, msg, vec, veclen, size); |
c1c6733f | 2986 | + } |
bb1d8b11 AM |
2987 | + |
2988 | + /* Save a copy of the message if we're expecting an ACK */ | |
2989 | + if (!(flags & MSG_NOACK) && acks_expected) { | |
2990 | + struct cl_protheader *savhdr = (struct cl_protheader *) saved_msg_buffer; | |
2991 | + | |
2992 | + memcpy_fromkvec(saved_msg_buffer, vectors, | |
2993 | + size + sizeof (header)); | |
2994 | + | |
2995 | + saved_msg_len = size + sizeof (header); | |
2996 | + retry_count = ack_count = 0; | |
2997 | + clear_bit(RESEND_NEEDED, &mainloop_flags); | |
2998 | + | |
2999 | + /* Clear the REPLYEXPected flag so we force a real ACK | |
3000 | + if it's necessary to resend this packet */ | |
3001 | + savhdr->flags &= ~MSG_REPLYEXP; | |
3002 | + start_ack_timer(); | |
3003 | + } | |
3004 | + | |
3005 | + up(&send_lock); | |
3006 | + return result; | |
c1c6733f AM |
3007 | +} |
3008 | + | |
bb1d8b11 AM |
3009 | +static int queue_message(struct socket *sock, void *buf, int len, |
3010 | + struct sockaddr_cl *caddr, | |
3011 | + unsigned char port, int flags) | |
c1c6733f | 3012 | +{ |
bb1d8b11 | 3013 | + struct queued_message *qmsg; |
c1c6733f | 3014 | + |
bb1d8b11 AM |
3015 | + qmsg = kmalloc(sizeof (struct queued_message), |
3016 | + (in_atomic() | |
3017 | + || irqs_disabled())? GFP_ATOMIC : GFP_KERNEL); | |
3018 | + if (qmsg == NULL) | |
3019 | + return -1; | |
c1c6733f | 3020 | + |
bb1d8b11 AM |
3021 | + memcpy(qmsg->msg_buffer, buf, len); |
3022 | + qmsg->msg_len = len; | |
3023 | + if (caddr) { | |
3024 | + memcpy(&qmsg->addr, caddr, sizeof (struct sockaddr_cl)); | |
3025 | + qmsg->addr_len = sizeof (struct sockaddr_cl); | |
c1c6733f | 3026 | + } |
bb1d8b11 AM |
3027 | + else { |
3028 | + qmsg->addr_len = 0; | |
3029 | + } | |
3030 | + qmsg->flags = flags; | |
3031 | + qmsg->port = port; | |
3032 | + qmsg->socket = sock; | |
3033 | + | |
3034 | + down(&messages_list_lock); | |
3035 | + list_add_tail(&qmsg->list, &messages_list); | |
3036 | + up(&messages_list_lock); | |
3037 | + | |
3038 | + wake_up_interruptible(&cnxman_waitq); | |
3039 | + | |
3040 | + return 0; | |
c1c6733f AM |
3041 | +} |
3042 | + | |
bb1d8b11 AM |
3043 | +static int cl_sendmsg(struct kiocb *iocb, struct socket *sock, |
3044 | + struct msghdr *msg, size_t size) | |
c1c6733f | 3045 | +{ |
bb1d8b11 AM |
3046 | + struct cluster_sock *c = cluster_sk(sock->sk); |
3047 | + char *buffer; | |
3048 | + int status; | |
3049 | + uint8_t port; | |
3050 | + struct kvec vec; | |
3051 | + struct sockaddr_cl *caddr = msg->msg_name; | |
c1c6733f | 3052 | + |
bb1d8b11 AM |
3053 | + if (sock->sk->sk_protocol == CLPROTO_MASTER) |
3054 | + return -EOPNOTSUPP; | |
c1c6733f | 3055 | + |
bb1d8b11 | 3056 | + port = c->port; |
c1c6733f | 3057 | + |
bb1d8b11 AM |
3058 | + /* Only capable users can override the port number */ |
3059 | + if (caddr && capable(CAP_CLUSTER) && caddr->scl_port) | |
3060 | + port = caddr->scl_port; | |
c1c6733f | 3061 | + |
bb1d8b11 AM |
3062 | + if (port == 0) |
3063 | + return -EDESTADDRREQ; | |
3064 | + | |
3065 | + /* Allocate a kernel buffer for the data so we can put it into a kvec */ | |
3066 | + buffer = kmalloc(size, GFP_KERNEL); | |
3067 | + if (!buffer) | |
3068 | + return -ENOMEM; | |
3069 | + | |
3070 | + if (memcpy_fromiovec(buffer, msg->msg_iov, size)) { | |
3071 | + status = -EFAULT; | |
3072 | + goto end_send; | |
c1c6733f AM |
3073 | + } |
3074 | + | |
bb1d8b11 AM |
3075 | + vec.iov_len = size; |
3076 | + vec.iov_base = buffer; | |
c1c6733f | 3077 | + |
bb1d8b11 | 3078 | + status = __sendmsg(sock, msg, &vec, 1, size, port); |
c1c6733f | 3079 | + |
bb1d8b11 AM |
3080 | + end_send: |
3081 | + kfree(buffer); | |
c1c6733f | 3082 | + |
bb1d8b11 AM |
3083 | + return status; |
3084 | +} | |
c1c6733f | 3085 | + |
bb1d8b11 AM |
3086 | +/* Kernel call to sendmsg */ |
3087 | +int kcl_sendmsg(struct socket *sock, void *buf, int size, | |
3088 | + struct sockaddr_cl *caddr, int addr_len, unsigned int flags) | |
3089 | +{ | |
3090 | + struct kvec vecs[1]; | |
3091 | + struct msghdr msg; | |
3092 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
3093 | + unsigned char port; | |
c1c6733f | 3094 | + |
bb1d8b11 AM |
3095 | + if (size > MAX_CLUSTER_MESSAGE) |
3096 | + return -EINVAL; | |
3097 | + if (!atomic_read(&cnxman_running)) | |
3098 | + return -ENOTCONN; | |
3099 | + | |
3100 | + port = c->port; | |
3101 | + if (caddr && caddr->scl_port) | |
3102 | + port = caddr->scl_port; | |
3103 | + | |
3104 | + if (port == 0) | |
3105 | + return -EDESTADDRREQ; | |
3106 | + | |
3107 | + /* If we have no process context then queue it up for kclusterd to | |
3108 | + * send. */ | |
3109 | + if (in_interrupt() || flags & MSG_QUEUE) { | |
3110 | + return queue_message(sock, buf, size, caddr, port, | |
3111 | + flags & ~MSG_QUEUE); | |
b7b72b66 | 3112 | + } |
b7b72b66 | 3113 | + |
bb1d8b11 AM |
3114 | + vecs[0].iov_base = buf; |
3115 | + vecs[0].iov_len = size; | |
c1c6733f | 3116 | + |
bb1d8b11 AM |
3117 | + memset(&msg, 0, sizeof (msg)); |
3118 | + msg.msg_name = caddr; | |
3119 | + msg.msg_namelen = addr_len; | |
3120 | + msg.msg_flags = flags; | |
c1c6733f | 3121 | + |
bb1d8b11 | 3122 | + return __sendmsg(sock, &msg, vecs, 1, size, port); |
c1c6733f AM |
3123 | +} |
3124 | + | |
bb1d8b11 | 3125 | +static int send_queued_message(struct queued_message *qmsg) |
c1c6733f | 3126 | +{ |
bb1d8b11 AM |
3127 | + struct kvec vecs[1]; |
3128 | + struct msghdr msg; | |
c1c6733f | 3129 | + |
bb1d8b11 AM |
3130 | + /* Don't send blocked messages */ |
3131 | + if (qmsg->port > HIGH_PROTECTED_PORT | |
3132 | + && (!cluster_is_quorate || in_transition())) | |
3133 | + return -EAGAIN; | |
c1c6733f | 3134 | + |
bb1d8b11 AM |
3135 | + vecs[0].iov_base = qmsg->msg_buffer; |
3136 | + vecs[0].iov_len = qmsg->msg_len; | |
c1c6733f | 3137 | + |
bb1d8b11 AM |
3138 | + memset(&msg, 0, sizeof (msg)); |
3139 | + msg.msg_name = qmsg->addr_len ? &qmsg->addr : NULL; | |
3140 | + msg.msg_namelen = qmsg->addr_len; | |
3141 | + msg.msg_flags = qmsg->flags; | |
c1c6733f | 3142 | + |
bb1d8b11 AM |
3143 | + return __sendmsg(qmsg->socket, &msg, vecs, 1, |
3144 | + qmsg->msg_len, qmsg->port); | |
c1c6733f AM |
3145 | +} |
3146 | + | |
bb1d8b11 AM |
3147 | +int kcl_register_read_callback(struct socket *sock, |
3148 | + int (*routine) (char *, int, char *, int, | |
3149 | + unsigned int)) | |
c1c6733f | 3150 | +{ |
bb1d8b11 | 3151 | + struct cluster_sock *c = cluster_sk(sock->sk); |
c1c6733f | 3152 | + |
bb1d8b11 | 3153 | + c->kernel_callback = routine; |
c1c6733f | 3154 | + |
bb1d8b11 | 3155 | + return 0; |
c1c6733f AM |
3156 | +} |
3157 | + | |
bb1d8b11 AM |
3158 | +/* Used where we are in kclusterd context and we can't allow the task to wait |
3159 | + * as we are also responsible to processing the ACKs that do the wake up. Try | |
3160 | + * to send the message immediately and queue it if that's not possible */ | |
3161 | +static int send_or_queue_message(struct socket *sock, void *buf, int len, | |
3162 | + struct sockaddr_cl *caddr, | |
3163 | + unsigned int flags) | |
c1c6733f | 3164 | +{ |
bb1d8b11 AM |
3165 | + struct kvec vecs[1]; |
3166 | + struct msghdr msg; | |
3167 | + int status; | |
c1c6733f | 3168 | + |
bb1d8b11 AM |
3169 | + vecs[0].iov_base = buf; |
3170 | + vecs[0].iov_len = len; | |
c1c6733f | 3171 | + |
bb1d8b11 AM |
3172 | + memset(&msg, 0, sizeof (msg)); |
3173 | + msg.msg_name = caddr; | |
3174 | + msg.msg_namelen = caddr ? sizeof (struct sockaddr_cl) : 0; | |
3175 | + msg.msg_flags = MSG_DONTWAIT | flags; | |
c1c6733f | 3176 | + |
bb1d8b11 AM |
3177 | + status = __sendmsg(NULL, &msg, vecs, 1, len, 0); |
3178 | + | |
3179 | + /* Did it work ? */ | |
3180 | + if (status > 0) { | |
3181 | + return 0; | |
c1c6733f | 3182 | + } |
c1c6733f | 3183 | + |
bb1d8b11 AM |
3184 | + /* Failure other than EAGAIN is fatal */ |
3185 | + if (status != -EAGAIN) { | |
3186 | + return status; | |
c1c6733f | 3187 | + } |
bb1d8b11 AM |
3188 | + |
3189 | + return queue_message(sock, buf, len, caddr, 0, flags); | |
c1c6733f AM |
3190 | +} |
3191 | + | |
bb1d8b11 AM |
3192 | +/* Send a listen request to a node */ |
3193 | +static void send_listen_request(int nodeid, unsigned char port) | |
c1c6733f | 3194 | +{ |
bb1d8b11 AM |
3195 | + struct cl_listenmsg listenmsg; |
3196 | + struct sockaddr_cl caddr; | |
c1c6733f | 3197 | + |
bb1d8b11 | 3198 | + memset(&caddr, 0, sizeof (caddr)); |
c1c6733f | 3199 | + |
bb1d8b11 AM |
3200 | + /* Build the header */ |
3201 | + listenmsg.cmd = CLUSTER_CMD_LISTENREQ; | |
3202 | + listenmsg.target_port = port; | |
3203 | + listenmsg.listening = 0; | |
3204 | + listenmsg.tag = current->pid; | |
c1c6733f | 3205 | + |
bb1d8b11 AM |
3206 | + caddr.scl_family = AF_CLUSTER; |
3207 | + caddr.scl_port = 0; | |
3208 | + caddr.scl_nodeid = nodeid; | |
c1c6733f | 3209 | + |
bb1d8b11 AM |
3210 | + send_or_queue_message(NULL, &listenmsg, sizeof(listenmsg), &caddr, MSG_REPLYEXP); |
3211 | + return; | |
c1c6733f AM |
3212 | +} |
3213 | + | |
bb1d8b11 AM |
3214 | +/* Return 1 or 0 to indicate if we have a listener on the requested port */ |
3215 | +static void send_listen_response(struct cl_comms_socket *csock, int nodeid, | |
3216 | + unsigned char port, unsigned short tag) | |
c1c6733f | 3217 | +{ |
bb1d8b11 AM |
3218 | + struct cl_listenmsg listenmsg; |
3219 | + struct sockaddr_cl caddr; | |
3220 | + int status; | |
c1c6733f | 3221 | + |
bb1d8b11 | 3222 | + memset(&caddr, 0, sizeof (caddr)); |
c1c6733f | 3223 | + |
bb1d8b11 AM |
3224 | + /* Build the message */ |
3225 | + listenmsg.cmd = CLUSTER_CMD_LISTENRESP; | |
3226 | + listenmsg.target_port = port; | |
3227 | + listenmsg.tag = tag; | |
3228 | + listenmsg.listening = (port_array[port] != 0) ? 1 : 0; | |
c1c6733f | 3229 | + |
bb1d8b11 AM |
3230 | + caddr.scl_family = AF_CLUSTER; |
3231 | + caddr.scl_port = 0; | |
3232 | + caddr.scl_nodeid = nodeid; | |
c1c6733f | 3233 | + |
bb1d8b11 AM |
3234 | + status = send_or_queue_message(NULL, &listenmsg, |
3235 | + sizeof (listenmsg), | |
3236 | + &caddr, 0); | |
c1c6733f | 3237 | + |
bb1d8b11 | 3238 | + return; |
c1c6733f AM |
3239 | +} |
3240 | + | |
bb1d8b11 AM |
3241 | +/* Send an ACK */ |
3242 | +static int cl_sendack(struct cl_comms_socket *csock, unsigned short seq, | |
3243 | + int addr_len, char *addr, unsigned char remport, | |
3244 | + unsigned char flag) | |
c1c6733f | 3245 | +{ |
bb1d8b11 AM |
3246 | + struct kvec vec; |
3247 | + struct cl_ackmsg ackmsg; | |
3248 | + struct msghdr msg; | |
3249 | + struct sockaddr_in6 daddr; | |
3250 | + int result; | |
c1c6733f | 3251 | + |
bb1d8b11 AM |
3252 | +#ifdef DEBUG_COMMS |
3253 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
c1c6733f | 3254 | + |
bb1d8b11 AM |
3255 | + P_COMMS("Sending ACK to %s, seq=%d\n", |
3256 | + print_addr(addr, address_length, buf), le16_to_cpu(seq)); | |
3257 | +#endif | |
3258 | + | |
3259 | + if (addr) { | |
3260 | + memcpy(&daddr, addr, addr_len); | |
3261 | + } | |
3262 | + else { | |
3263 | + memcpy(&daddr, &csock->saddr, csock->addr_len); | |
3264 | + addr_len = csock->addr_len; | |
c1c6733f | 3265 | + } |
c1c6733f | 3266 | + |
bb1d8b11 AM |
3267 | + /* Build the header */ |
3268 | + ackmsg.header.tgtport = 0; /* Protocol port */ | |
3269 | + ackmsg.header.srcport = 0; | |
3270 | + ackmsg.header.seq = 0; | |
3271 | + ackmsg.header.flags = MSG_NOACK; | |
3272 | + ackmsg.header.cluster = cpu_to_le16(cluster_id); | |
3273 | + ackmsg.header.srcid = us ? cpu_to_le32(us->node_id) : 0; | |
3274 | + ackmsg.header.ack = seq; /* already in LE order */ | |
3275 | + ackmsg.header.tgtid = 0; /* ACKS are unicast so we don't bother | |
3276 | + * to look this up */ | |
3277 | + ackmsg.cmd = CLUSTER_CMD_ACK; | |
3278 | + ackmsg.remport = remport; | |
3279 | + ackmsg.aflags = flag; | |
3280 | + vec.iov_base = &ackmsg; | |
3281 | + vec.iov_len = sizeof (ackmsg); | |
c1c6733f | 3282 | + |
bb1d8b11 AM |
3283 | + memset(&msg, 0, sizeof (msg)); |
3284 | + msg.msg_name = &daddr; | |
3285 | + msg.msg_namelen = addr_len; | |
c1c6733f | 3286 | + |
bb1d8b11 | 3287 | + result = kernel_sendmsg(csock->sock, &msg, &vec, 1, sizeof (ackmsg)); |
c1c6733f | 3288 | + |
bb1d8b11 AM |
3289 | + if (result < 0) |
3290 | + printk(KERN_CRIT CMAN_NAME ": error sending ACK: %d\n", result); | |
3291 | + | |
3292 | + return result; | |
c1c6733f | 3293 | + |
c1c6733f AM |
3294 | +} |
3295 | + | |
bb1d8b11 AM |
3296 | +/* Wait for all ACKS to be gathered */ |
3297 | +void kcl_wait_for_all_acks() | |
c1c6733f | 3298 | +{ |
bb1d8b11 | 3299 | + while (ack_count < acks_expected) { |
c1c6733f | 3300 | + |
bb1d8b11 AM |
3301 | + DECLARE_WAITQUEUE(wq, current); |
3302 | + struct task_struct *tsk = current; | |
c1c6733f | 3303 | + |
bb1d8b11 AM |
3304 | + set_task_state(tsk, TASK_INTERRUPTIBLE); |
3305 | + add_wait_queue(&socket_waitq, &wq); | |
c1c6733f | 3306 | + |
bb1d8b11 AM |
3307 | + if (ack_count < acks_expected) { |
3308 | + schedule(); | |
c1c6733f | 3309 | + } |
bb1d8b11 AM |
3310 | + |
3311 | + set_task_state(tsk, TASK_RUNNING); | |
3312 | + remove_wait_queue(&socket_waitq, &wq); | |
c1c6733f | 3313 | + } |
c1c6733f AM |
3314 | +} |
3315 | + | |
bb1d8b11 AM |
3316 | +/* Send a closedown OOB message to all cluster nodes - this tells them that a |
3317 | + * port listener has gone away */ | |
3318 | +static void send_port_close_oob(unsigned char port) | |
c1c6733f | 3319 | +{ |
bb1d8b11 | 3320 | + struct cl_closemsg closemsg; |
c1c6733f | 3321 | + |
bb1d8b11 AM |
3322 | + /* Build the header */ |
3323 | + closemsg.cmd = CLUSTER_CMD_PORTCLOSED; | |
3324 | + closemsg.port = port; | |
c1c6733f | 3325 | + |
bb1d8b11 AM |
3326 | + send_or_queue_message(NULL, &closemsg, sizeof (closemsg), NULL, 0); |
3327 | + return; | |
c1c6733f AM |
3328 | +} |
3329 | + | |
bb1d8b11 AM |
3330 | +/* A remote port has been closed - post an OOB message to the local listen on |
3331 | + * that port (if there is one) */ | |
3332 | +static void post_close_oob(unsigned char port, int nodeid) | |
c1c6733f | 3333 | +{ |
bb1d8b11 AM |
3334 | + struct cl_portclosed_oob *oobmsg; |
3335 | + struct sk_buff *skb; | |
3336 | + struct sock *sock = port_array[port]; | |
3337 | + struct cb_info *cbinfo; | |
c1c6733f | 3338 | + |
bb1d8b11 AM |
3339 | + if (!sock) { |
3340 | + return; /* No-one listening */ | |
3341 | + } | |
c1c6733f | 3342 | + |
bb1d8b11 AM |
3343 | + skb = alloc_skb(sizeof (*oobmsg), GFP_KERNEL); |
3344 | + if (!skb) | |
c1c6733f AM |
3345 | + return; |
3346 | + | |
bb1d8b11 AM |
3347 | + skb_put(skb, sizeof (*oobmsg)); |
3348 | + oobmsg = (struct cl_portclosed_oob *) skb->data; | |
3349 | + oobmsg->port = port; | |
3350 | + oobmsg->cmd = CLUSTER_OOB_MSG_PORTCLOSED; | |
c1c6733f | 3351 | + |
bb1d8b11 AM |
3352 | + cbinfo = (struct cb_info *)skb->cb; |
3353 | + cbinfo->oob = 1; | |
3354 | + cbinfo->orig_nodeid = nodeid; | |
3355 | + cbinfo->orig_port = port; | |
c1c6733f | 3356 | + |
bb1d8b11 | 3357 | + sock_queue_rcv_skb(sock, skb); |
c1c6733f | 3358 | + |
bb1d8b11 | 3359 | +} |
c1c6733f | 3360 | + |
bb1d8b11 AM |
3361 | +/* Leave the cluster */ |
3362 | +static void node_shutdown() | |
3363 | +{ | |
3364 | + struct cl_barrier *barrier; | |
3365 | + struct list_head *blist; | |
3366 | + struct list_head *temp; | |
3367 | + struct list_head *socklist; | |
3368 | + struct cl_client_socket *csock; | |
3369 | + struct sk_buff *null_skb; | |
c1c6733f | 3370 | + |
bb1d8b11 AM |
3371 | + if (we_are_a_cluster_member) |
3372 | + printk(KERN_INFO CMAN_NAME ": we are leaving the cluster. %s\n", | |
3373 | + us->leave_reason?leave_string(us->leave_reason):""); | |
c1c6733f | 3374 | + |
bb1d8b11 AM |
3375 | + atomic_set(&cnxman_running, 0); |
3376 | + unjam(); | |
c1c6733f | 3377 | + |
bb1d8b11 AM |
3378 | + /* Notify kernel listeners first */ |
3379 | + notify_kernel_listeners(LEAVING, 0); | |
c1c6733f | 3380 | + |
bb1d8b11 AM |
3381 | + /* Notify client sockets */ |
3382 | + down(&client_socket_lock); | |
3383 | + list_for_each_safe(socklist, temp, &client_socket_list) { | |
3384 | + csock = list_entry(socklist, struct cl_client_socket, list); | |
c1c6733f | 3385 | + |
bb1d8b11 AM |
3386 | + null_skb = alloc_skb(0, GFP_KERNEL); |
3387 | + if (null_skb) | |
3388 | + sock_queue_rcv_skb(csock->sock->sk, null_skb); | |
3389 | + list_del(&csock->list); | |
3390 | + kfree(csock); | |
3391 | + } | |
3392 | + up(&client_socket_lock); | |
3393 | + we_are_a_cluster_member = 0; | |
3394 | + cluster_is_quorate = 0; | |
c1c6733f | 3395 | + |
bb1d8b11 | 3396 | + sm_stop(1); |
c1c6733f | 3397 | + |
bb1d8b11 AM |
3398 | + /* Wake up any processes waiting for barriers */ |
3399 | + down(&barrier_list_lock); | |
3400 | + list_for_each(blist, &barrier_list) { | |
3401 | + barrier = list_entry(blist, struct cl_barrier, list); | |
c1c6733f | 3402 | + |
bb1d8b11 AM |
3403 | + /* Cancel any timers */ |
3404 | + if (timer_pending(&barrier->timer)) | |
3405 | + del_timer(&barrier->timer); | |
3406 | + | |
3407 | + /* Force it to be auto-delete so it discards itself */ | |
3408 | + if (barrier->state == BARRIER_STATE_WAITING) { | |
3409 | + barrier->flags |= BARRIER_ATTR_AUTODELETE; | |
3410 | + wake_up_interruptible(&barrier->waitq); | |
3411 | + } | |
3412 | + else { | |
3413 | + if (barrier->callback) { | |
3414 | + barrier->callback(barrier->name, -ENOTCONN); | |
3415 | + barrier->callback = NULL; | |
3416 | + } | |
c1c6733f AM |
3417 | + } |
3418 | + } | |
bb1d8b11 | 3419 | + up(&barrier_list_lock); |
c1c6733f | 3420 | + |
bb1d8b11 AM |
3421 | + /* Wake up any processes waiting for ISLISTENING requests */ |
3422 | + down(&listenreq_lock); | |
3423 | + list_for_each(blist, &listenreq_list) { | |
3424 | + struct cl_waiting_listen_request *lrequest = | |
3425 | + list_entry(blist, struct cl_waiting_listen_request, list); | |
c1c6733f | 3426 | + |
bb1d8b11 AM |
3427 | + if (lrequest->waiting) |
3428 | + wake_up_interruptible(&lrequest->waitq); | |
3429 | + } | |
3430 | + up(&listenreq_lock); | |
c1c6733f AM |
3431 | +} |
3432 | + | |
bb1d8b11 | 3433 | +static void free_cluster_sockets() |
c1c6733f | 3434 | +{ |
bb1d8b11 AM |
3435 | + struct list_head *socklist; |
3436 | + struct cl_comms_socket *sock; | |
3437 | + struct list_head *temp; | |
c1c6733f | 3438 | + |
bb1d8b11 AM |
3439 | + list_for_each_safe(socklist, temp, &socket_list) { |
3440 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
c1c6733f | 3441 | + |
bb1d8b11 AM |
3442 | + list_del(&sock->list); |
3443 | + fput(sock->file); | |
3444 | + kfree(sock); | |
c1c6733f | 3445 | + } |
bb1d8b11 AM |
3446 | + num_interfaces = 0; |
3447 | + current_interface = NULL; | |
c1c6733f AM |
3448 | +} |
3449 | + | |
bb1d8b11 AM |
3450 | +/* Tidy up after all the rest of the cluster bits have shut down */ |
3451 | +static void node_cleanup() | |
c1c6733f | 3452 | +{ |
bb1d8b11 AM |
3453 | + struct list_head *nodelist; |
3454 | + struct list_head *proclist; | |
3455 | + struct list_head *temp; | |
3456 | + struct list_head *socklist; | |
3457 | + struct list_head *blist; | |
3458 | + struct temp_node *tn; | |
3459 | + struct temp_node *tmp; | |
3460 | + struct cl_comms_socket *sock; | |
3461 | + struct kernel_notify_struct *knotify; | |
c1c6733f | 3462 | + |
bb1d8b11 AM |
3463 | + /* Free list of kernel listeners */ |
3464 | + list_for_each_safe(proclist, temp, &kernel_listener_list) { | |
3465 | + knotify = | |
3466 | + list_entry(proclist, struct kernel_notify_struct, list); | |
3467 | + list_del(&knotify->list); | |
3468 | + kfree(knotify); | |
c1c6733f AM |
3469 | + } |
3470 | + | |
bb1d8b11 AM |
3471 | + /* Mark the sockets as busy so they don't get added to the active |
3472 | + * sockets list in the next few lines of code before we free them */ | |
3473 | + list_for_each_safe(socklist, temp, &socket_list) { | |
3474 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
c1c6733f | 3475 | + |
bb1d8b11 | 3476 | + set_bit(1, &sock->active); |
c1c6733f | 3477 | + } |
bb1d8b11 AM |
3478 | + |
3479 | + /* Tidy the active sockets list */ | |
3480 | + list_for_each_safe(socklist, temp, &active_socket_list) { | |
3481 | + sock = | |
3482 | + list_entry(socklist, struct cl_comms_socket, active_list); | |
3483 | + list_del(&sock->active_list); | |
c1c6733f AM |
3484 | + } |
3485 | + | |
bb1d8b11 AM |
3486 | + /* Free the memory allocated to cluster nodes */ |
3487 | + free_nodeid_array(); | |
3488 | + down(&cluster_members_lock); | |
3489 | + us = NULL; | |
3490 | + list_for_each_safe(nodelist, temp, &cluster_members_list) { | |
c1c6733f | 3491 | + |
bb1d8b11 AM |
3492 | + struct list_head *addrlist; |
3493 | + struct list_head *addrtemp; | |
3494 | + struct cluster_node *node; | |
3495 | + struct cluster_node_addr *nodeaddr; | |
c1c6733f | 3496 | + |
bb1d8b11 AM |
3497 | + node = list_entry(nodelist, struct cluster_node, list); |
3498 | + | |
3499 | + list_for_each_safe(addrlist, addrtemp, &node->addr_list) { | |
3500 | + nodeaddr = | |
3501 | + list_entry(addrlist, struct cluster_node_addr, | |
3502 | + list); | |
3503 | + | |
3504 | + list_del(&nodeaddr->list); | |
3505 | + kfree(nodeaddr); | |
c1c6733f | 3506 | + } |
bb1d8b11 AM |
3507 | + list_del(&node->list); |
3508 | + kfree(node->name); | |
3509 | + kfree(node); | |
c1c6733f | 3510 | + } |
bb1d8b11 | 3511 | + cluster_members = 0; |
c1c6733f AM |
3512 | + up(&cluster_members_lock); |
3513 | + | |
bb1d8b11 AM |
3514 | + /* Clean the temp node IDs list. */ |
3515 | + down(&tempnode_lock); | |
3516 | + list_for_each_entry_safe(tn, tmp, &tempnode_list, list) { | |
3517 | + list_del(&tn->list); | |
3518 | + kfree(tn); | |
3519 | + } | |
3520 | + up(&tempnode_lock); | |
3521 | + | |
3522 | + /* Free the memory allocated to the outgoing sockets */ | |
3523 | + free_cluster_sockets(); | |
3524 | + | |
3525 | + /* Make sure that all the barriers are deleted */ | |
3526 | + down(&barrier_list_lock); | |
3527 | + list_for_each_safe(blist, temp, &barrier_list) { | |
3528 | + struct cl_barrier *barrier = | |
3529 | + list_entry(blist, struct cl_barrier, list); | |
3530 | + | |
3531 | + list_del(&barrier->list); | |
3532 | + kfree(barrier); | |
3533 | + } | |
3534 | + up(&barrier_list_lock); | |
3535 | + | |
3536 | + kcluster_pid = 0; | |
3537 | + clear_bit(RESEND_NEEDED, &mainloop_flags); | |
3538 | + acks_expected = 0; | |
3539 | + wanted_nodeid = 0; | |
c1c6733f AM |
3540 | +} |
3541 | + | |
bb1d8b11 AM |
3542 | +/* If "cluster_is_quorate" is 0 then all activity apart from protected ports is |
3543 | + * blocked. */ | |
3544 | +void set_quorate(int total_votes) | |
c1c6733f | 3545 | +{ |
bb1d8b11 | 3546 | + int quorate; |
c1c6733f | 3547 | + |
bb1d8b11 AM |
3548 | + if (get_quorum() > total_votes) { |
3549 | + quorate = 0; | |
3550 | + } | |
3551 | + else { | |
3552 | + quorate = 1; | |
3553 | + } | |
c1c6733f | 3554 | + |
bb1d8b11 AM |
3555 | + /* Hide messages during startup state transition */ |
3556 | + if (we_are_a_cluster_member) { | |
3557 | + if (cluster_is_quorate && !quorate) | |
3558 | + printk(KERN_CRIT CMAN_NAME | |
3559 | + ": quorum lost, blocking activity\n"); | |
3560 | + if (!cluster_is_quorate && quorate) | |
3561 | + printk(KERN_CRIT CMAN_NAME | |
3562 | + ": quorum regained, resuming activity\n"); | |
3563 | + } | |
3564 | + cluster_is_quorate = quorate; | |
3565 | + | |
3566 | + /* Wake up any sleeping processes */ | |
3567 | + if (cluster_is_quorate) { | |
3568 | + unjam(); | |
c1c6733f | 3569 | + } |
c1c6733f | 3570 | + |
c1c6733f AM |
3571 | +} |
3572 | + | |
bb1d8b11 | 3573 | +void queue_oob_skb(struct socket *sock, int cmd) |
c1c6733f | 3574 | +{ |
bb1d8b11 AM |
3575 | + struct sk_buff *skb; |
3576 | + struct cb_info *cbinfo; | |
3577 | + struct cl_portclosed_oob *oobmsg; | |
c1c6733f | 3578 | + |
bb1d8b11 AM |
3579 | + skb = alloc_skb(sizeof (*oobmsg), GFP_KERNEL); |
3580 | + if (!skb) | |
3581 | + return; | |
c1c6733f | 3582 | + |
bb1d8b11 AM |
3583 | + skb_put(skb, sizeof (*oobmsg)); |
3584 | + oobmsg = (struct cl_portclosed_oob *) skb->data; | |
3585 | + oobmsg->port = 0; | |
3586 | + oobmsg->cmd = cmd; | |
3587 | + | |
3588 | + /* There is no remote node associated with this so | |
3589 | + clear out the field to avoid any accidents */ | |
3590 | + cbinfo = (struct cb_info *)skb->cb; | |
3591 | + cbinfo->oob = 1; | |
3592 | + cbinfo->orig_nodeid = 0; | |
3593 | + cbinfo->orig_port = 0; | |
c1c6733f | 3594 | + |
bb1d8b11 | 3595 | + sock_queue_rcv_skb(sock->sk, skb); |
c1c6733f AM |
3596 | +} |
3597 | + | |
bb1d8b11 AM |
3598 | +/* Notify interested parties that the cluster configuration has changed */ |
3599 | +void notify_listeners() | |
c1c6733f | 3600 | +{ |
bb1d8b11 AM |
3601 | + struct notify_struct *notify; |
3602 | + struct list_head *proclist; | |
3603 | + struct list_head *socklist; | |
3604 | + struct list_head *temp; | |
c1c6733f | 3605 | + |
bb1d8b11 AM |
3606 | + /* Do kernel listeners first */ |
3607 | + notify_kernel_listeners(CLUSTER_RECONFIG, 0); | |
c1c6733f | 3608 | + |
bb1d8b11 AM |
3609 | + /* Now we deign to tell userspace */ |
3610 | + down(&event_listener_lock); | |
3611 | + list_for_each_safe(proclist, temp, &event_listener_list) { | |
3612 | + notify = list_entry(proclist, struct notify_struct, list); | |
c1c6733f | 3613 | + |
bb1d8b11 AM |
3614 | + /* If the kill fails then remove the process from the list */ |
3615 | + if (kill_proc(notify->pid, notify->signal, 0) == -ESRCH) { | |
3616 | + list_del(¬ify->list); | |
3617 | + kfree(notify); | |
c1c6733f | 3618 | + } |
c1c6733f | 3619 | + } |
bb1d8b11 | 3620 | + up(&event_listener_lock); |
c1c6733f | 3621 | + |
bb1d8b11 AM |
3622 | + /* Tell userspace processes which want OOB messages */ |
3623 | + down(&client_socket_lock); | |
3624 | + list_for_each(socklist, &client_socket_list) { | |
3625 | + struct cl_client_socket *csock; | |
3626 | + csock = list_entry(socklist, struct cl_client_socket, list); | |
3627 | + queue_oob_skb(csock->sock, CLUSTER_OOB_MSG_STATECHANGE); | |
c1c6733f | 3628 | + } |
bb1d8b11 AM |
3629 | + up(&client_socket_lock); |
3630 | +} | |
c1c6733f | 3631 | + |
bb1d8b11 AM |
3632 | +/* This fills in the list of all addresses for the local node */ |
3633 | +void get_local_addresses(struct cluster_node *node) | |
3634 | +{ | |
3635 | + struct list_head *socklist; | |
3636 | + struct cl_comms_socket *sock; | |
c1c6733f | 3637 | + |
bb1d8b11 AM |
3638 | + list_for_each(socklist, &socket_list) { |
3639 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
c1c6733f | 3640 | + |
bb1d8b11 AM |
3641 | + if (sock->recv_only) { |
3642 | + add_node_address(node, (char *) &sock->saddr, address_length); | |
3643 | + } | |
3644 | + } | |
c1c6733f AM |
3645 | +} |
3646 | + | |
bb1d8b11 AM |
3647 | + |
3648 | +static uint16_t generate_cluster_id(char *name) | |
c1c6733f | 3649 | +{ |
bb1d8b11 AM |
3650 | + int i; |
3651 | + int value = 0; | |
c1c6733f | 3652 | + |
bb1d8b11 AM |
3653 | + for (i=0; i<strlen(name); i++) { |
3654 | + value <<= 1; | |
3655 | + value += name[i]; | |
c1c6733f | 3656 | + } |
bb1d8b11 AM |
3657 | + return value & 0xFFFF; |
3658 | +} | |
c1c6733f | 3659 | + |
bb1d8b11 AM |
3660 | +/* Return the next comms socket we can use. */ |
3661 | +static struct cl_comms_socket *get_next_interface(struct cl_comms_socket *cur) | |
3662 | +{ | |
3663 | + int next; | |
3664 | + struct list_head *socklist; | |
c1c6733f | 3665 | + |
bb1d8b11 AM |
3666 | + /* Fast path for single interface systems */ |
3667 | + if (num_interfaces <= 1) | |
3668 | + return cur; | |
c1c6733f | 3669 | + |
bb1d8b11 AM |
3670 | + /* Next number */ |
3671 | + next = cur->number + 1; | |
3672 | + if (next > num_interfaces) | |
3673 | + next = 1; | |
c1c6733f | 3674 | + |
bb1d8b11 AM |
3675 | + /* Find the socket with this number, I could optimise this by starting |
3676 | + * at the current i/f but most systems are going to have a small number | |
3677 | + * of them anyway */ | |
3678 | + list_for_each(socklist, &socket_list) { | |
3679 | + struct cl_comms_socket *sock; | |
3680 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
c1c6733f | 3681 | + |
bb1d8b11 AM |
3682 | + if (!sock->recv_only && sock->number == next) |
3683 | + return sock; | |
c1c6733f | 3684 | + } |
bb1d8b11 AM |
3685 | + |
3686 | + BUG(); | |
3687 | + return NULL; | |
c1c6733f AM |
3688 | +} |
3689 | + | |
bb1d8b11 AM |
3690 | +/* MUST be called with the barrier list lock held */ |
3691 | +static struct cl_barrier *find_barrier(char *name) | |
c1c6733f | 3692 | +{ |
bb1d8b11 AM |
3693 | + struct list_head *blist; |
3694 | + struct cl_barrier *bar; | |
c1c6733f | 3695 | + |
bb1d8b11 AM |
3696 | + list_for_each(blist, &barrier_list) { |
3697 | + bar = list_entry(blist, struct cl_barrier, list); | |
c1c6733f | 3698 | + |
bb1d8b11 AM |
3699 | + if (strcmp(name, bar->name) == 0) |
3700 | + return bar; | |
c1c6733f | 3701 | + } |
bb1d8b11 AM |
3702 | + return NULL; |
3703 | +} | |
c1c6733f | 3704 | + |
bb1d8b11 AM |
3705 | +/* Do the stuff we need to do when the barrier has completed phase 1 */ |
3706 | +static void check_barrier_complete_phase1(struct cl_barrier *barrier) | |
3707 | +{ | |
3708 | + if (atomic_read(&barrier->got_nodes) == ((barrier->expected_nodes != 0) | |
3709 | + ? barrier->expected_nodes : | |
3710 | + cluster_members)) { | |
c1c6733f | 3711 | + |
bb1d8b11 | 3712 | + struct cl_barriermsg bmsg; |
c1c6733f | 3713 | + |
bb1d8b11 AM |
3714 | + atomic_inc(&barrier->completed_nodes); /* We have completed */ |
3715 | + barrier->phase = 2; /* Wait for complete phase II */ | |
c1c6733f | 3716 | + |
bb1d8b11 AM |
3717 | + /* Send completion message, remember: we are in cnxman context |
3718 | + * and must not block */ | |
3719 | + bmsg.cmd = CLUSTER_CMD_BARRIER; | |
3720 | + bmsg.subcmd = BARRIER_COMPLETE; | |
3721 | + bmsg.flags = 0; | |
3722 | + strcpy(bmsg.name, barrier->name); | |
c1c6733f | 3723 | + |
bb1d8b11 AM |
3724 | + P_BARRIER("Sending COMPLETE for %s\n", barrier->name); |
3725 | + queue_message(NULL, (char *) &bmsg, sizeof (bmsg), NULL, 0, 0); | |
c1c6733f | 3726 | + } |
c1c6733f AM |
3727 | +} |
3728 | + | |
bb1d8b11 AM |
3729 | +/* Do the stuff we need to do when the barrier has been reached */ |
3730 | +/* Return 1 if we deleted the barrier */ | |
3731 | +static int check_barrier_complete_phase2(struct cl_barrier *barrier, int status) | |
c1c6733f | 3732 | +{ |
bb1d8b11 | 3733 | + spin_lock_irq(&barrier->phase2_spinlock); |
c1c6733f | 3734 | + |
bb1d8b11 AM |
3735 | + if (barrier->state != BARRIER_STATE_COMPLETE && |
3736 | + (status == -ETIMEDOUT || | |
3737 | + atomic_read(&barrier->completed_nodes) == | |
3738 | + ((barrier->expected_nodes != 0) | |
3739 | + ? barrier->expected_nodes : cluster_members))) { | |
3740 | + | |
3741 | + if (status == 0 && barrier->timeout) | |
3742 | + del_timer(&barrier->timer); | |
3743 | + barrier->endreason = status; | |
3744 | + | |
3745 | + /* Wake up listener */ | |
3746 | + if (barrier->state == BARRIER_STATE_WAITING) { | |
3747 | + wake_up_interruptible(&barrier->waitq); | |
3748 | + } | |
3749 | + else { | |
3750 | + /* Additional tasks we have to do if the user was not | |
3751 | + * waiting... */ | |
3752 | + /* Call the callback */ | |
3753 | + if (barrier->callback) { | |
3754 | + barrier->callback(barrier->name, 0); | |
3755 | + barrier->callback = NULL; | |
3756 | + } | |
3757 | + /* Remove it if it's AUTO-DELETE */ | |
3758 | + if (barrier->flags & BARRIER_ATTR_AUTODELETE) { | |
3759 | + list_del(&barrier->list); | |
3760 | + spin_unlock_irq(&barrier->phase2_spinlock); | |
3761 | + kfree(barrier); | |
3762 | + return 1; | |
3763 | + } | |
3764 | + } | |
3765 | + barrier->state = BARRIER_STATE_COMPLETE; | |
c1c6733f | 3766 | + } |
bb1d8b11 AM |
3767 | + spin_unlock_irq(&barrier->phase2_spinlock); |
3768 | + return 0; | |
3769 | +} | |
c1c6733f | 3770 | + |
bb1d8b11 AM |
3771 | +/* Called if a barrier timeout happens */ |
3772 | +static void barrier_timer_fn(unsigned long arg) | |
3773 | +{ | |
3774 | + struct cl_barrier *barrier = (struct cl_barrier *) arg; | |
c1c6733f | 3775 | + |
bb1d8b11 AM |
3776 | + /* Ignore any futher messages, they are too late. */ |
3777 | + barrier->phase = 0; | |
c1c6733f | 3778 | + |
bb1d8b11 AM |
3779 | + /* and cause it to timeout */ |
3780 | + check_barrier_complete_phase2(barrier, -ETIMEDOUT); | |
c1c6733f AM |
3781 | +} |
3782 | + | |
bb1d8b11 AM |
3783 | +/* Process BARRIER messages from other nodes */ |
3784 | +static void process_barrier_msg(struct cl_barriermsg *msg, | |
3785 | + struct cluster_node *node) | |
c1c6733f AM |
3786 | +{ |
3787 | + struct cl_barrier *barrier; | |
3788 | + | |
c1c6733f | 3789 | + down(&barrier_list_lock); |
bb1d8b11 AM |
3790 | + barrier = find_barrier(msg->name); |
3791 | + up(&barrier_list_lock); | |
c1c6733f | 3792 | + |
bb1d8b11 AM |
3793 | + /* Ignore other peoples messages, in_transition() is needed here so |
3794 | + * that joining nodes will see their barrier messages before the | |
3795 | + * we_are_a_cluster_member is set */ | |
3796 | + if (!we_are_a_cluster_member && !in_transition()) | |
3797 | + return; | |
3798 | + if (!barrier) | |
3799 | + return; | |
c1c6733f | 3800 | + |
bb1d8b11 AM |
3801 | + P_BARRIER("Got %d for %s, from node %s\n", msg->subcmd, msg->name, |
3802 | + node ? node->name : "unknown"); | |
c1c6733f | 3803 | + |
bb1d8b11 AM |
3804 | + switch (msg->subcmd) { |
3805 | + case BARRIER_WAIT: | |
3806 | + down(&barrier->lock); | |
3807 | + if (barrier->phase == 0) | |
3808 | + barrier->phase = 1; | |
c1c6733f | 3809 | + |
bb1d8b11 AM |
3810 | + if (barrier->phase == 1) { |
3811 | + atomic_inc(&barrier->got_nodes); | |
3812 | + check_barrier_complete_phase1(barrier); | |
3813 | + } | |
3814 | + else { | |
3815 | + printk(KERN_WARNING CMAN_NAME | |
3816 | + ": got WAIT barrier not in phase 1 %s (%d)\n", | |
3817 | + msg->name, barrier->phase); | |
3818 | + | |
3819 | + } | |
c1c6733f | 3820 | + up(&barrier->lock); |
bb1d8b11 | 3821 | + break; |
c1c6733f | 3822 | + |
bb1d8b11 AM |
3823 | + case BARRIER_COMPLETE: |
3824 | + down(&barrier->lock); | |
3825 | + atomic_inc(&barrier->completed_nodes); | |
c1c6733f | 3826 | + |
bb1d8b11 AM |
3827 | + /* First node to get all the WAIT messages sends COMPLETE, so |
3828 | + * we all complete */ | |
3829 | + if (barrier->phase == 1) { | |
3830 | + atomic_set(&barrier->got_nodes, | |
3831 | + barrier->expected_nodes); | |
3832 | + check_barrier_complete_phase1(barrier); | |
3833 | + } | |
3834 | + | |
3835 | + if (barrier->phase == 2) { | |
3836 | + /* If it was deleted (ret==1) then no need to unlock | |
3837 | + * the mutex */ | |
3838 | + if (check_barrier_complete_phase2(barrier, 0) == 1) | |
3839 | + return; | |
3840 | + } | |
3841 | + up(&barrier->lock); | |
3842 | + break; | |
3843 | + } | |
c1c6733f AM |
3844 | +} |
3845 | + | |
bb1d8b11 AM |
3846 | +/* In-kernel membership API */ |
3847 | +int kcl_add_callback(void (*callback) (kcl_callback_reason, long arg)) | |
c1c6733f | 3848 | +{ |
bb1d8b11 | 3849 | + struct kernel_notify_struct *notify; |
c1c6733f | 3850 | + |
bb1d8b11 AM |
3851 | + notify = kmalloc(sizeof (struct kernel_notify_struct), GFP_KERNEL); |
3852 | + if (!notify) | |
3853 | + return -ENOMEM; | |
3854 | + notify->callback = callback; | |
c1c6733f | 3855 | + |
bb1d8b11 AM |
3856 | + down(&kernel_listener_lock); |
3857 | + list_add(¬ify->list, &kernel_listener_list); | |
3858 | + up(&kernel_listener_lock); | |
c1c6733f | 3859 | + |
bb1d8b11 AM |
3860 | + return 0; |
3861 | +} | |
c1c6733f | 3862 | + |
bb1d8b11 AM |
3863 | +int kcl_remove_callback(void (*callback) (kcl_callback_reason, long arg)) |
3864 | +{ | |
3865 | + struct list_head *calllist; | |
3866 | + struct list_head *temp; | |
3867 | + struct kernel_notify_struct *notify; | |
c1c6733f | 3868 | + |
bb1d8b11 AM |
3869 | + down(&kernel_listener_lock); |
3870 | + list_for_each_safe(calllist, temp, &kernel_listener_list) { | |
3871 | + notify = list_entry(calllist, struct kernel_notify_struct, list); | |
3872 | + if (notify->callback == callback){ | |
3873 | + list_del(¬ify->list); | |
3874 | + kfree(notify); | |
3875 | + up(&kernel_listener_lock); | |
3876 | + return 0; | |
3877 | + } | |
c1c6733f | 3878 | + } |
bb1d8b11 AM |
3879 | + up(&kernel_listener_lock); |
3880 | + return -EINVAL; | |
3881 | +} | |
c1c6733f | 3882 | + |
bb1d8b11 AM |
3883 | +/* Return quorate status */ |
3884 | +int kcl_is_quorate() | |
3885 | +{ | |
3886 | + return cluster_is_quorate; | |
3887 | +} | |
c1c6733f | 3888 | + |
bb1d8b11 AM |
3889 | +/* Return the address list for a node */ |
3890 | +struct list_head *kcl_get_node_addresses(int nodeid) | |
3891 | +{ | |
3892 | + struct cluster_node *node = find_node_by_nodeid(nodeid); | |
c1c6733f | 3893 | + |
bb1d8b11 AM |
3894 | + if (node) |
3895 | + return &node->addr_list; | |
3896 | + else | |
3897 | + return NULL; | |
3898 | +} | |
c1c6733f | 3899 | + |
bb1d8b11 AM |
3900 | +static void copy_to_kclnode(struct cluster_node *node, |
3901 | + struct kcl_cluster_node *knode) | |
3902 | +{ | |
3903 | + strcpy(knode->name, node->name); | |
3904 | + knode->size = sizeof (struct kcl_cluster_node); | |
3905 | + knode->votes = node->votes; | |
3906 | + knode->state = node->state; | |
3907 | + knode->node_id = node->node_id; | |
3908 | + knode->us = node->us; | |
3909 | + knode->leave_reason = node->leave_reason; | |
3910 | + knode->incarnation = node->incarnation; | |
3911 | +} | |
c1c6733f | 3912 | + |
bb1d8b11 AM |
3913 | +/* Return the info for a node given it's address. if addr is NULL then return |
3914 | + * OUR info */ | |
3915 | +int kcl_get_node_by_addr(unsigned char *addr, int addr_len, | |
3916 | + struct kcl_cluster_node *n) | |
3917 | +{ | |
3918 | + struct cluster_node *node; | |
c1c6733f | 3919 | + |
bb1d8b11 AM |
3920 | + /* They want us */ |
3921 | + if (addr == NULL) { | |
3922 | + node = us; | |
c1c6733f | 3923 | + } |
bb1d8b11 AM |
3924 | + else { |
3925 | + node = find_node_by_addr(addr, addr_len); | |
3926 | + if (!node) | |
3927 | + return -1; | |
3928 | + } | |
3929 | + | |
3930 | + /* Copy to user's buffer */ | |
3931 | + copy_to_kclnode(node, n); | |
3932 | + return 0; | |
3933 | +} | |
c1c6733f | 3934 | + |
bb1d8b11 AM |
3935 | +int kcl_get_node_by_name(unsigned char *name, struct kcl_cluster_node *n) |
3936 | +{ | |
3937 | + struct cluster_node *node; | |
c1c6733f | 3938 | + |
bb1d8b11 AM |
3939 | + /* They want us */ |
3940 | + if (name == NULL) { | |
3941 | + node = us; | |
3942 | + if (node == NULL) | |
3943 | + return -1; | |
3944 | + } | |
3945 | + else { | |
3946 | + node = find_node_by_name(name); | |
3947 | + if (!node) | |
3948 | + return -1; | |
c1c6733f AM |
3949 | + } |
3950 | + | |
bb1d8b11 AM |
3951 | + /* Copy to user's buffer */ |
3952 | + copy_to_kclnode(node, n); | |
3953 | + return 0; | |
3954 | +} | |
c1c6733f | 3955 | + |
bb1d8b11 AM |
3956 | +/* As above but by node id. MUCH faster */ |
3957 | +int kcl_get_node_by_nodeid(int nodeid, struct kcl_cluster_node *n) | |
3958 | +{ | |
3959 | + struct cluster_node *node; | |
c1c6733f | 3960 | + |
bb1d8b11 AM |
3961 | + /* They want us */ |
3962 | + if (nodeid == 0) { | |
3963 | + node = us; | |
3964 | + if (node == NULL) | |
3965 | + return -1; | |
c1c6733f AM |
3966 | + } |
3967 | + else { | |
bb1d8b11 AM |
3968 | + node = find_node_by_nodeid(nodeid); |
3969 | + if (!node) | |
3970 | + return -1; | |
c1c6733f AM |
3971 | + } |
3972 | + | |
bb1d8b11 AM |
3973 | + /* Copy to user's buffer */ |
3974 | + copy_to_kclnode(node, n); | |
3975 | + return 0; | |
c1c6733f AM |
3976 | +} |
3977 | + | |
bb1d8b11 AM |
3978 | +/* Return a list of all cluster members ever */ |
3979 | +int kcl_get_all_members(struct list_head *list) | |
c1c6733f | 3980 | +{ |
bb1d8b11 AM |
3981 | + struct list_head *nodelist; |
3982 | + struct cluster_node *node; | |
3983 | + struct kcl_cluster_node *newnode; | |
3984 | + int num_nodes = 0; | |
c1c6733f | 3985 | + |
bb1d8b11 AM |
3986 | + down(&cluster_members_lock); |
3987 | + list_for_each(nodelist, &cluster_members_list) { | |
3988 | + if (list) { | |
3989 | + node = list_entry(nodelist, struct cluster_node, list); | |
3990 | + newnode = | |
3991 | + kmalloc(sizeof (struct kcl_cluster_node), | |
3992 | + GFP_KERNEL); | |
3993 | + if (newnode) { | |
3994 | + copy_to_kclnode(node, newnode); | |
3995 | + list_add(&newnode->list, list); | |
3996 | + num_nodes++; | |
3997 | + } | |
3998 | + } | |
3999 | + else { | |
4000 | + num_nodes++; | |
4001 | + } | |
4002 | + } | |
4003 | + up(&cluster_members_lock); | |
c1c6733f | 4004 | + |
bb1d8b11 AM |
4005 | + return num_nodes; |
4006 | +} | |
c1c6733f | 4007 | + |
bb1d8b11 AM |
4008 | +/* Return a list of cluster members */ |
4009 | +int kcl_get_members(struct list_head *list) | |
4010 | +{ | |
4011 | + struct list_head *nodelist; | |
4012 | + struct cluster_node *node; | |
4013 | + struct kcl_cluster_node *newnode; | |
4014 | + int num_nodes = 0; | |
4015 | + | |
4016 | + down(&cluster_members_lock); | |
4017 | + list_for_each(nodelist, &cluster_members_list) { | |
4018 | + node = list_entry(nodelist, struct cluster_node, list); | |
4019 | + | |
4020 | + if (node->state == NODESTATE_MEMBER) { | |
4021 | + if (list) { | |
4022 | + newnode = | |
4023 | + kmalloc(sizeof (struct kcl_cluster_node), | |
4024 | + GFP_KERNEL); | |
4025 | + if (newnode) { | |
4026 | + copy_to_kclnode(node, newnode); | |
4027 | + list_add(&newnode->list, list); | |
4028 | + num_nodes++; | |
c1c6733f AM |
4029 | + } |
4030 | + } | |
4031 | + else { | |
bb1d8b11 | 4032 | + num_nodes++; |
c1c6733f AM |
4033 | + } |
4034 | + } | |
4035 | + } | |
bb1d8b11 | 4036 | + up(&cluster_members_lock); |
c1c6733f | 4037 | + |
bb1d8b11 AM |
4038 | + return num_nodes; |
4039 | +} | |
c1c6733f | 4040 | + |
bb1d8b11 AM |
4041 | +/* Copy current member's nodeids into buffer */ |
4042 | +int kcl_get_member_ids(uint32_t *idbuf, int size) | |
4043 | +{ | |
4044 | + struct list_head *nodelist; | |
4045 | + struct cluster_node *node; | |
4046 | + int num_nodes = 0; | |
4047 | + | |
4048 | + down(&cluster_members_lock); | |
4049 | + list_for_each(nodelist, &cluster_members_list) { | |
4050 | + node = list_entry(nodelist, struct cluster_node, list); | |
4051 | + | |
4052 | + if (node->state == NODESTATE_MEMBER) { | |
4053 | + if (idbuf && size) { | |
4054 | + idbuf[num_nodes] = node->node_id; | |
4055 | + num_nodes++; | |
4056 | + size--; | |
4057 | + } | |
4058 | + else { | |
4059 | + num_nodes++; | |
4060 | + } | |
c1c6733f AM |
4061 | + } |
4062 | + } | |
bb1d8b11 AM |
4063 | + up(&cluster_members_lock); |
4064 | + | |
4065 | + return num_nodes; | |
c1c6733f AM |
4066 | +} |
4067 | + | |
bb1d8b11 AM |
4068 | +/* Barrier API */ |
4069 | +int kcl_barrier_register(char *name, unsigned int flags, unsigned int nodes) | |
c1c6733f | 4070 | +{ |
bb1d8b11 | 4071 | + struct cl_barrier *barrier; |
c1c6733f | 4072 | + |
bb1d8b11 AM |
4073 | + /* We are not joined to a cluster */ |
4074 | + if (!we_are_a_cluster_member) | |
4075 | + return -ENOTCONN; | |
c1c6733f | 4076 | + |
bb1d8b11 AM |
4077 | + /* Must have a valid name */ |
4078 | + if (name == NULL || strlen(name) > MAX_BARRIER_NAME_LEN - 1) | |
4079 | + return -EINVAL; | |
c1c6733f | 4080 | + |
bb1d8b11 AM |
4081 | + /* We don't do this yet */ |
4082 | + if (flags & BARRIER_ATTR_MULTISTEP) | |
4083 | + return -ENOTSUPP; | |
4084 | + | |
4085 | + down(&barrier_list_lock); | |
4086 | + | |
4087 | + /* See if it already exists */ | |
4088 | + if ((barrier = find_barrier(name))) { | |
4089 | + up(&barrier_list_lock); | |
4090 | + if (nodes != barrier->expected_nodes) { | |
4091 | + printk(KERN_WARNING CMAN_NAME | |
4092 | + ": Barrier registration failed for '%s', expected nodes=%d, requested=%d\n", | |
4093 | + name, barrier->expected_nodes, nodes); | |
4094 | + up(&barrier_list_lock); | |
4095 | + return -EINVAL; | |
c1c6733f | 4096 | + } |
bb1d8b11 AM |
4097 | + else |
4098 | + return 0; | |
4099 | + } | |
4100 | + | |
4101 | + /* Build a new struct and add it to the list */ | |
4102 | + barrier = kmalloc(sizeof (struct cl_barrier), GFP_KERNEL); | |
4103 | + if (barrier == NULL) { | |
4104 | + up(&barrier_list_lock); | |
4105 | + return -ENOMEM; | |
4106 | + } | |
4107 | + memset(barrier, 0, sizeof (*barrier)); | |
4108 | + | |
4109 | + strcpy(barrier->name, name); | |
4110 | + barrier->flags = flags; | |
4111 | + barrier->expected_nodes = nodes; | |
4112 | + atomic_set(&barrier->got_nodes, 0); | |
4113 | + atomic_set(&barrier->completed_nodes, 0); | |
4114 | + barrier->endreason = 0; | |
4115 | + barrier->registered_nodes = 1; | |
4116 | + spin_lock_init(&barrier->phase2_spinlock); | |
4117 | + barrier->state = BARRIER_STATE_INACTIVE; | |
4118 | + init_MUTEX(&barrier->lock); | |
4119 | + | |
4120 | + list_add(&barrier->list, &barrier_list); | |
4121 | + up(&barrier_list_lock); | |
4122 | + | |
4123 | + return 0; | |
4124 | +} | |
4125 | + | |
4126 | +static int barrier_setattr_enabled(struct cl_barrier *barrier, | |
4127 | + unsigned int attr, unsigned long arg) | |
4128 | +{ | |
4129 | + int status; | |
4130 | + | |
4131 | + /* Can't disable a barrier */ | |
4132 | + if (!arg) { | |
4133 | + up(&barrier->lock); | |
4134 | + return -EINVAL; | |
c1c6733f | 4135 | + } |
c1c6733f | 4136 | + |
bb1d8b11 AM |
4137 | + /* We need to send WAIT now because the user may not |
4138 | + * actually call kcl_barrier_wait() */ | |
4139 | + if (!barrier->waitsent) { | |
4140 | + struct cl_barriermsg bmsg; | |
c1c6733f | 4141 | + |
bb1d8b11 AM |
4142 | + /* Send it to the rest of the cluster */ |
4143 | + bmsg.cmd = CLUSTER_CMD_BARRIER; | |
4144 | + bmsg.subcmd = BARRIER_WAIT; | |
4145 | + strcpy(bmsg.name, barrier->name); | |
c1c6733f | 4146 | + |
bb1d8b11 AM |
4147 | + barrier->waitsent = 1; |
4148 | + barrier->phase = 1; | |
c1c6733f | 4149 | + |
bb1d8b11 | 4150 | + atomic_inc(&barrier->got_nodes); |
c1c6733f | 4151 | + |
bb1d8b11 AM |
4152 | + /* Start the timer if one was wanted */ |
4153 | + if (barrier->timeout) { | |
4154 | + init_timer(&barrier->timer); | |
4155 | + barrier->timer.function = barrier_timer_fn; | |
4156 | + barrier->timer.data = (long) barrier; | |
4157 | + mod_timer(&barrier->timer, jiffies + (barrier->timeout * HZ)); | |
4158 | + } | |
c1c6733f | 4159 | + |
bb1d8b11 AM |
4160 | + /* Barrier WAIT and COMPLETE messages are |
4161 | + * always queued - that way they always get | |
4162 | + * sent out in the right order. If we don't do | |
4163 | + * this then one can get sent out in the | |
4164 | + * context of the user process and the other in | |
4165 | + * cnxman and COMPLETE may /just/ slide in | |
4166 | + * before WAIT if its in the queue | |
4167 | + */ | |
4168 | + P_BARRIER("Sending WAIT for %s\n", barrier->name); | |
4169 | + status = queue_message(NULL, &bmsg, sizeof (bmsg), NULL, 0, 0); | |
4170 | + if (status < 0) { | |
4171 | + up(&barrier->lock); | |
4172 | + return status; | |
c1c6733f | 4173 | + } |
bb1d8b11 AM |
4174 | + |
4175 | + /* It might have been reached now */ | |
4176 | + if (barrier | |
4177 | + && barrier->state != BARRIER_STATE_COMPLETE | |
4178 | + && barrier->phase == 1) | |
4179 | + check_barrier_complete_phase1(barrier); | |
4180 | + } | |
4181 | + if (barrier && barrier->state == BARRIER_STATE_COMPLETE) { | |
4182 | + up(&barrier->lock); | |
4183 | + return barrier->endreason; | |
c1c6733f | 4184 | + } |
bb1d8b11 AM |
4185 | + up(&barrier->lock); |
4186 | + return 0; /* Nothing to propogate */ | |
4187 | +} | |
c1c6733f | 4188 | + |
bb1d8b11 AM |
4189 | +int kcl_barrier_setattr(char *name, unsigned int attr, unsigned long arg) |
4190 | +{ | |
4191 | + struct cl_barrier *barrier; | |
c1c6733f | 4192 | + |
bb1d8b11 AM |
4193 | + /* See if it already exists */ |
4194 | + down(&barrier_list_lock); | |
4195 | + if (!(barrier = find_barrier(name))) { | |
4196 | + up(&barrier_list_lock); | |
4197 | + return -ENOENT; | |
c1c6733f | 4198 | + } |
bb1d8b11 | 4199 | + up(&barrier_list_lock); |
c1c6733f | 4200 | + |
bb1d8b11 AM |
4201 | + down(&barrier->lock); |
4202 | + if (barrier->state == BARRIER_STATE_COMPLETE) { | |
4203 | + up(&barrier->lock); | |
4204 | + return 0; | |
4205 | + } | |
c1c6733f | 4206 | + |
bb1d8b11 AM |
4207 | + switch (attr) { |
4208 | + case BARRIER_SETATTR_AUTODELETE: | |
4209 | + if (arg) | |
4210 | + barrier->flags |= BARRIER_ATTR_AUTODELETE; | |
4211 | + else | |
4212 | + barrier->flags &= ~BARRIER_ATTR_AUTODELETE; | |
4213 | + up(&barrier->lock); | |
4214 | + return 0; | |
4215 | + break; | |
c1c6733f | 4216 | + |
bb1d8b11 AM |
4217 | + case BARRIER_SETATTR_TIMEOUT: |
4218 | + /* Can only change the timout of an inactive barrier */ | |
4219 | + if (barrier->state == BARRIER_STATE_WAITING | |
4220 | + || barrier->waitsent) { | |
4221 | + up(&barrier->lock); | |
4222 | + return -EINVAL; | |
4223 | + } | |
4224 | + barrier->timeout = arg; | |
4225 | + up(&barrier->lock); | |
4226 | + return 0; | |
c1c6733f | 4227 | + |
bb1d8b11 AM |
4228 | + case BARRIER_SETATTR_MULTISTEP: |
4229 | + up(&barrier->lock); | |
4230 | + return -ENOTSUPP; | |
c1c6733f | 4231 | + |
bb1d8b11 AM |
4232 | + case BARRIER_SETATTR_ENABLED: |
4233 | + return barrier_setattr_enabled(barrier, attr, arg); | |
4234 | + | |
4235 | + case BARRIER_SETATTR_NODES: | |
4236 | + /* Can only change the expected node count of an inactive | |
4237 | + * barrier */ | |
4238 | + if (barrier->state == BARRIER_STATE_WAITING | |
4239 | + || barrier->waitsent) | |
4240 | + return -EINVAL; | |
4241 | + barrier->expected_nodes = arg; | |
4242 | + break; | |
4243 | + | |
4244 | + case BARRIER_SETATTR_CALLBACK: | |
4245 | + if (barrier->state == BARRIER_STATE_WAITING | |
4246 | + || barrier->waitsent) | |
4247 | + return -EINVAL; | |
4248 | + barrier->callback = (void (*)(char *, int)) arg; | |
4249 | + up(&barrier->lock); | |
4250 | + return 0; /* Don't propgate this to other nodes */ | |
c1c6733f | 4251 | + } |
c1c6733f | 4252 | + |
bb1d8b11 AM |
4253 | + up(&barrier->lock); |
4254 | + return 0; | |
c1c6733f AM |
4255 | +} |
4256 | + | |
bb1d8b11 | 4257 | +int kcl_barrier_delete(char *name) |
c1c6733f | 4258 | +{ |
bb1d8b11 | 4259 | + struct cl_barrier *barrier; |
c1c6733f | 4260 | + |
bb1d8b11 AM |
4261 | + down(&barrier_list_lock); |
4262 | + /* See if it exists */ | |
4263 | + if (!(barrier = find_barrier(name))) { | |
4264 | + up(&barrier_list_lock); | |
4265 | + return -ENOENT; | |
4266 | + } | |
b7b72b66 | 4267 | + |
bb1d8b11 AM |
4268 | + /* Delete it */ |
4269 | + list_del(&barrier->list); | |
4270 | + kfree(barrier); | |
b7b72b66 | 4271 | + |
bb1d8b11 | 4272 | + up(&barrier_list_lock); |
b7b72b66 | 4273 | + |
bb1d8b11 | 4274 | + return 0; |
c1c6733f AM |
4275 | +} |
4276 | + | |
bb1d8b11 | 4277 | +int kcl_barrier_cancel(char *name) |
c1c6733f | 4278 | +{ |
bb1d8b11 | 4279 | + struct cl_barrier *barrier; |
c1c6733f | 4280 | + |
bb1d8b11 AM |
4281 | + /* See if it exists */ |
4282 | + down(&barrier_list_lock); | |
4283 | + if (!(barrier = find_barrier(name))) { | |
4284 | + up(&barrier_list_lock); | |
4285 | + return -ENOENT; | |
4286 | + } | |
4287 | + down(&barrier->lock); | |
c1c6733f | 4288 | + |
bb1d8b11 | 4289 | + barrier->endreason = -ENOTCONN; |
c1c6733f | 4290 | + |
bb1d8b11 AM |
4291 | + if (barrier->callback) { |
4292 | + barrier->callback(barrier->name, -ECONNRESET); | |
4293 | + barrier->callback = NULL; | |
c1c6733f AM |
4294 | + } |
4295 | + | |
bb1d8b11 AM |
4296 | + if (barrier->timeout) |
4297 | + del_timer(&barrier->timer); | |
c1c6733f | 4298 | + |
bb1d8b11 AM |
4299 | + /* Remove it if it's AUTO-DELETE */ |
4300 | + if (barrier->flags & BARRIER_ATTR_AUTODELETE) { | |
4301 | + list_del(&barrier->list); | |
4302 | + up(&barrier->lock); | |
4303 | + kfree(barrier); | |
4304 | + up(&barrier_list_lock); | |
4305 | + return 0; | |
4306 | + } | |
4307 | + | |
4308 | + if (barrier->state == BARRIER_STATE_WAITING) | |
4309 | + wake_up_interruptible(&barrier->waitq); | |
c1c6733f | 4310 | + |
bb1d8b11 AM |
4311 | + up(&barrier->lock); |
4312 | + up(&barrier_list_lock); | |
c1c6733f AM |
4313 | + return 0; |
4314 | +} | |
4315 | + | |
bb1d8b11 | 4316 | +int kcl_barrier_wait(char *name) |
c1c6733f | 4317 | +{ |
bb1d8b11 AM |
4318 | + struct cl_barrier *barrier; |
4319 | + int ret; | |
c1c6733f | 4320 | + |
bb1d8b11 AM |
4321 | + if (!atomic_read(&cnxman_running)) |
4322 | + return -ENOTCONN; | |
c1c6733f | 4323 | + |
bb1d8b11 AM |
4324 | + /* Enable it */ |
4325 | + kcl_barrier_setattr(name, BARRIER_SETATTR_ENABLED, 1L); | |
c1c6733f | 4326 | + |
bb1d8b11 | 4327 | + down(&barrier_list_lock); |
c1c6733f | 4328 | + |
bb1d8b11 AM |
4329 | + /* See if it still exists - enable may have deleted it! */ |
4330 | + if (!(barrier = find_barrier(name))) { | |
4331 | + up(&barrier_list_lock); | |
4332 | + return -ENOENT; | |
c1c6733f | 4333 | + } |
bb1d8b11 AM |
4334 | + |
4335 | + down(&barrier->lock); | |
4336 | + | |
4337 | + up(&barrier_list_lock); | |
4338 | + | |
4339 | + /* If it has already completed then return the status */ | |
4340 | + if (barrier->state == BARRIER_STATE_COMPLETE) { | |
4341 | + up(&barrier->lock); | |
4342 | + return barrier->endreason; | |
c1c6733f AM |
4343 | + } |
4344 | + | |
bb1d8b11 | 4345 | + barrier->state = BARRIER_STATE_WAITING; |
c1c6733f | 4346 | + |
bb1d8b11 AM |
4347 | + /* Have we all reached the barrier? */ |
4348 | + while (atomic_read(&barrier->completed_nodes) != | |
4349 | + ((barrier->expected_nodes == 0) | |
4350 | + ? cluster_members : barrier->expected_nodes) | |
4351 | + && barrier->endreason == 0) { | |
c1c6733f | 4352 | + |
bb1d8b11 | 4353 | + wait_queue_t wq; |
c1c6733f | 4354 | + |
bb1d8b11 AM |
4355 | + init_waitqueue_entry(&wq, current); |
4356 | + init_waitqueue_head(&barrier->waitq); | |
c1c6733f | 4357 | + |
bb1d8b11 AM |
4358 | + /* Wait for em all */ |
4359 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
4360 | + add_wait_queue(&barrier->waitq, &wq); | |
c1c6733f | 4361 | + |
bb1d8b11 AM |
4362 | + if (atomic_read(&barrier->completed_nodes) != |
4363 | + ((barrier->expected_nodes == | |
4364 | + 0) ? cluster_members : barrier->expected_nodes) | |
4365 | + && barrier->endreason == 0) { | |
4366 | + up(&barrier->lock); | |
4367 | + schedule(); | |
4368 | + down(&barrier->lock); | |
4369 | + } | |
c1c6733f | 4370 | + |
bb1d8b11 AM |
4371 | + remove_wait_queue(&barrier->waitq, &wq); |
4372 | + set_task_state(current, TASK_RUNNING); | |
c1c6733f | 4373 | + |
bb1d8b11 AM |
4374 | + if (signal_pending(current)) { |
4375 | + barrier->endreason = -EINTR; | |
4376 | + break; | |
4377 | + } | |
4378 | + } | |
4379 | + barrier->state = BARRIER_STATE_INACTIVE; | |
c1c6733f | 4380 | + |
bb1d8b11 AM |
4381 | + if (barrier->timeout) |
4382 | + del_timer(&barrier->timer); | |
c1c6733f | 4383 | + |
bb1d8b11 AM |
4384 | + /* Barrier has been reached on all nodes, call the callback */ |
4385 | + if (barrier->callback) { | |
4386 | + barrier->callback(barrier->name, barrier->endreason); | |
4387 | + barrier->callback = NULL; | |
4388 | + } | |
c1c6733f | 4389 | + |
bb1d8b11 | 4390 | + atomic_set(&barrier->got_nodes, 0); |
c1c6733f | 4391 | + |
bb1d8b11 AM |
4392 | + /* Return the reason we were woken */ |
4393 | + ret = barrier->endreason; | |
c1c6733f | 4394 | + |
bb1d8b11 AM |
4395 | + /* Remove it if it's AUTO-DELETE */ |
4396 | + if (barrier->flags & BARRIER_ATTR_AUTODELETE) { | |
4397 | + down(&barrier_list_lock); | |
4398 | + list_del(&barrier->list); | |
4399 | + up(&barrier_list_lock); | |
4400 | + up(&barrier->lock); | |
4401 | + kfree(barrier); | |
4402 | + } | |
4403 | + else { | |
4404 | + up(&barrier->lock); | |
4405 | + } | |
c1c6733f | 4406 | + |
bb1d8b11 AM |
4407 | + /* We were woken up because the node left the cluster ? */ |
4408 | + if (!atomic_read(&cnxman_running)) | |
4409 | + ret = -ENOTCONN; | |
c1c6733f | 4410 | + |
bb1d8b11 AM |
4411 | + return ret; |
4412 | +} | |
4413 | + | |
4414 | +/* This is called from membership services when a node has left the cluster - | |
4415 | + * we signal all waiting barriers with -ESRCH so they know to do something | |
4416 | + * else, if the number of nodes is left at 0 then we compare the new number of | |
4417 | + * nodes in the cluster with that at the barrier and return 0 (success) in that | |
4418 | + * case */ | |
4419 | +void check_barrier_returns() | |
c1c6733f | 4420 | +{ |
bb1d8b11 AM |
4421 | + struct list_head *blist; |
4422 | + struct list_head *llist; | |
4423 | + struct cl_barrier *barrier; | |
4424 | + int status = 0; | |
c1c6733f | 4425 | + |
bb1d8b11 AM |
4426 | + down(&barrier_list_lock); |
4427 | + list_for_each(blist, &barrier_list) { | |
4428 | + barrier = list_entry(blist, struct cl_barrier, list); | |
b7b72b66 | 4429 | + |
bb1d8b11 AM |
4430 | + if (barrier->waitsent) { |
4431 | + int wakeit = 0; | |
4432 | + | |
4433 | + /* Check for a dynamic member barrier */ | |
4434 | + if (barrier->expected_nodes == 0) { | |
4435 | + if (barrier->registered_nodes == | |
4436 | + cluster_members) { | |
4437 | + status = 0; | |
4438 | + wakeit = 1; | |
4439 | + } | |
4440 | + } | |
4441 | + else { | |
4442 | + status = -ESRCH; | |
4443 | + wakeit = 1; | |
4444 | + } | |
4445 | + | |
4446 | + /* Do we need to tell the barrier? */ | |
4447 | + if (wakeit) { | |
4448 | + if (barrier->state == BARRIER_STATE_WAITING) { | |
4449 | + barrier->endreason = status; | |
4450 | + wake_up_interruptible(&barrier->waitq); | |
4451 | + } | |
4452 | + else { | |
4453 | + if (barrier->callback) { | |
4454 | + barrier->callback(barrier->name, | |
4455 | + status); | |
4456 | + } | |
4457 | + } | |
4458 | + } | |
4459 | + } | |
c1c6733f | 4460 | + } |
bb1d8b11 | 4461 | + up(&barrier_list_lock); |
c1c6733f | 4462 | + |
bb1d8b11 AM |
4463 | + /* Part 2 check for outstanding listen requests for dead nodes and |
4464 | + * cancel them */ | |
4465 | + down(&listenreq_lock); | |
4466 | + list_for_each(llist, &listenreq_list) { | |
4467 | + struct cl_waiting_listen_request *lrequest = | |
4468 | + list_entry(llist, struct cl_waiting_listen_request, list); | |
4469 | + struct cluster_node *node = | |
4470 | + find_node_by_nodeid(lrequest->nodeid); | |
c783755a | 4471 | + |
bb1d8b11 AM |
4472 | + if (node && node->state != NODESTATE_MEMBER) { |
4473 | + lrequest->result = -ENOTCONN; | |
4474 | + lrequest->waiting = 0; | |
4475 | + wake_up_interruptible(&lrequest->waitq); | |
4476 | + } | |
4477 | + } | |
4478 | + up(&listenreq_lock); | |
4479 | +} | |
c783755a | 4480 | + |
bb1d8b11 AM |
4481 | +int get_addr_from_temp_nodeid(int nodeid, char *addr, int *addrlen) |
4482 | +{ | |
4483 | + struct temp_node *tn; | |
4484 | + int err = 1; /* true */ | |
4485 | +#ifdef DEBUG_COMMS | |
4486 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
4487 | +#endif | |
c783755a | 4488 | + |
bb1d8b11 | 4489 | + down(&tempnode_lock); |
c783755a | 4490 | + |
bb1d8b11 AM |
4491 | + list_for_each_entry(tn, &tempnode_list, list) { |
4492 | + if (tn->nodeid == nodeid) { | |
4493 | + memcpy(addr, tn->addr, tn->addrlen); | |
4494 | + *addrlen = tn->addrlen; | |
4495 | + P_COMMS("get_temp_nodeid. id %d:\n: %s\n", | |
4496 | + tn->nodeid, print_addr(tn->addr, tn->addrlen, buf)); | |
c783755a | 4497 | + |
bb1d8b11 AM |
4498 | + goto out; |
4499 | + } | |
4500 | + } | |
4501 | + err = 0; | |
4502 | + | |
4503 | + out: | |
4504 | + up(&tempnode_lock); | |
4505 | + return err; | |
c783755a AM |
4506 | +} |
4507 | + | |
bb1d8b11 AM |
4508 | +/* Create a new temporary node ID. This list will only ever be very small |
4509 | + (usaully only 1 item) but I can't take the risk that someone won't try to | |
4510 | + boot 128 nodes all at exactly the same time. */ | |
4511 | +int new_temp_nodeid(char *addr, int addrlen) | |
c783755a | 4512 | +{ |
bb1d8b11 AM |
4513 | + struct temp_node *tn; |
4514 | + int err = -1; | |
4515 | + int try_nodeid = 0; | |
4516 | +#ifdef DEBUG_COMMS | |
4517 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
c783755a AM |
4518 | +#endif |
4519 | + | |
bb1d8b11 AM |
4520 | + P_COMMS("new_temp_nodeid needed for\n: %s\n", |
4521 | + print_addr(addr, addrlen, buf)); | |
c783755a | 4522 | + |
bb1d8b11 | 4523 | + down(&tempnode_lock); |
c783755a | 4524 | + |
bb1d8b11 AM |
4525 | + /* First see if we already know about this node */ |
4526 | + list_for_each_entry(tn, &tempnode_list, list) { | |
c783755a | 4527 | + |
bb1d8b11 AM |
4528 | + P_COMMS("new_temp_nodeid list. id %d:\n: %s\n", |
4529 | + tn->nodeid, print_addr(tn->addr, tn->addrlen, buf)); | |
c783755a | 4530 | + |
bb1d8b11 AM |
4531 | + /* We're already in here... */ |
4532 | + if (tn->addrlen == addrlen && | |
4533 | + memcmp(tn->addr, addr, addrlen) == 0) { | |
4534 | + P_COMMS("reused temp node ID %d\n", tn->nodeid); | |
4535 | + err = tn->nodeid; | |
4536 | + goto out; | |
4537 | + } | |
4538 | + } | |
c783755a | 4539 | + |
bb1d8b11 AM |
4540 | + /* Nope, OK, invent a suitable number */ |
4541 | + retry: | |
4542 | + try_nodeid -= 1; | |
4543 | + list_for_each_entry(tn, &tempnode_list, list) { | |
c783755a | 4544 | + |
bb1d8b11 AM |
4545 | + if (tn->nodeid == try_nodeid) |
4546 | + goto retry; | |
4547 | + } | |
c783755a | 4548 | + |
bb1d8b11 AM |
4549 | + tn = kmalloc(sizeof(struct temp_node), GFP_KERNEL); |
4550 | + if (!tn) | |
4551 | + goto out; | |
c783755a | 4552 | + |
bb1d8b11 AM |
4553 | + memcpy(tn->addr, addr, addrlen); |
4554 | + tn->addrlen = addrlen; | |
4555 | + tn->nodeid = try_nodeid; | |
4556 | + list_add_tail(&tn->list, &tempnode_list); | |
4557 | + err = try_nodeid; | |
4558 | + P_COMMS("new temp nodeid = %d\n", try_nodeid); | |
4559 | + out: | |
4560 | + up(&tempnode_lock); | |
4561 | + return err; | |
4562 | +} | |
c783755a | 4563 | + |
bb1d8b11 AM |
4564 | +static int is_valid_temp_nodeid(int nodeid) |
4565 | +{ | |
4566 | + struct temp_node *tn; | |
4567 | + int err = 1; /* true */ | |
c783755a | 4568 | + |
bb1d8b11 | 4569 | + down(&tempnode_lock); |
c783755a | 4570 | + |
bb1d8b11 AM |
4571 | + list_for_each_entry(tn, &tempnode_list, list) { |
4572 | + if (tn->nodeid == nodeid) | |
4573 | + goto out; | |
4574 | + } | |
4575 | + err = 0; | |
c783755a | 4576 | + |
bb1d8b11 AM |
4577 | + out: |
4578 | + P_COMMS("is_valid_temp_nodeid. %d = %d\n", nodeid, err); | |
4579 | + up(&tempnode_lock); | |
4580 | + return err; | |
4581 | +} | |
c783755a | 4582 | + |
bb1d8b11 AM |
4583 | +/* |
4584 | + * Remove any temp nodeIDs that refer to now-valid cluster members. | |
4585 | + */ | |
4586 | +void purge_temp_nodeids() | |
4587 | +{ | |
4588 | + struct temp_node *tn; | |
4589 | + struct temp_node *tmp; | |
4590 | + struct cluster_node *node; | |
4591 | + struct cluster_node_addr *nodeaddr; | |
c783755a | 4592 | + |
c783755a | 4593 | + |
bb1d8b11 AM |
4594 | + down(&tempnode_lock); |
4595 | + down(&cluster_members_lock); | |
c783755a | 4596 | + |
bb1d8b11 AM |
4597 | + /* |
4598 | + * The ordering of these nested lists is deliberately | |
4599 | + * arranged for the fewest list traversals overall | |
4600 | + */ | |
c783755a | 4601 | + |
bb1d8b11 AM |
4602 | + /* For each node... */ |
4603 | + list_for_each_entry(node, &cluster_members_list, list) { | |
4604 | + if (node->state == NODESTATE_MEMBER) { | |
4605 | + /* ...We check the temp node ID list... */ | |
4606 | + list_for_each_entry_safe(tn, tmp, &tempnode_list, list) { | |
c783755a | 4607 | + |
bb1d8b11 AM |
4608 | + /* ...against that node's address */ |
4609 | + list_for_each_entry(nodeaddr, &node->addr_list, list) { | |
c783755a | 4610 | + |
bb1d8b11 AM |
4611 | + if (memcmp(nodeaddr->addr, tn->addr, tn->addrlen) == 0) { |
4612 | + list_del(&tn->list); | |
4613 | + kfree(tn); | |
4614 | + } | |
4615 | + } | |
4616 | + } | |
4617 | + } | |
4618 | + } | |
4619 | + up(&cluster_members_lock); | |
4620 | + up(&tempnode_lock); | |
4621 | +} | |
c783755a | 4622 | + |
c783755a | 4623 | + |
bb1d8b11 AM |
4624 | +/* Quorum device functions */ |
4625 | +int kcl_register_quorum_device(char *name, int votes) | |
4626 | +{ | |
4627 | + if (quorum_device) | |
4628 | + return -EBUSY; | |
c783755a | 4629 | + |
bb1d8b11 AM |
4630 | + if (find_node_by_name(name)) |
4631 | + return -EINVAL; | |
c783755a | 4632 | + |
bb1d8b11 AM |
4633 | + quorum_device = kmalloc(sizeof (struct cluster_node), GFP_KERNEL); |
4634 | + if (!quorum_device) | |
4635 | + return -ENOMEM; | |
4636 | + memset(quorum_device, 0, sizeof (struct cluster_node)); | |
c783755a | 4637 | + |
bb1d8b11 AM |
4638 | + quorum_device->name = kmalloc(strlen(name) + 1, GFP_KERNEL); |
4639 | + if (!quorum_device->name) { | |
4640 | + kfree(quorum_device); | |
4641 | + quorum_device = NULL; | |
4642 | + return -ENOMEM; | |
4643 | + } | |
c783755a | 4644 | + |
bb1d8b11 AM |
4645 | + strcpy(quorum_device->name, name); |
4646 | + quorum_device->votes = votes; | |
4647 | + quorum_device->state = NODESTATE_DEAD; | |
c783755a | 4648 | + |
bb1d8b11 AM |
4649 | + /* Keep this list valid so it doesn't confuse other code */ |
4650 | + INIT_LIST_HEAD(&quorum_device->addr_list); | |
c783755a | 4651 | + |
bb1d8b11 AM |
4652 | + return 0; |
4653 | +} | |
c783755a | 4654 | + |
bb1d8b11 AM |
4655 | +int kcl_unregister_quorum_device(void) |
4656 | +{ | |
4657 | + if (!quorum_device) | |
4658 | + return -EINVAL; | |
4659 | + if (quorum_device->state == NODESTATE_MEMBER) | |
4660 | + return -EINVAL; | |
c783755a | 4661 | + |
bb1d8b11 | 4662 | + quorum_device = NULL; |
c783755a | 4663 | + |
bb1d8b11 AM |
4664 | + return 0; |
4665 | +} | |
c783755a | 4666 | + |
bb1d8b11 AM |
4667 | +int kcl_quorum_device_available(int yesno) |
4668 | +{ | |
4669 | + if (!quorum_device) | |
4670 | + return -EINVAL; | |
c783755a | 4671 | + |
bb1d8b11 AM |
4672 | + if (yesno) { |
4673 | + quorum_device->last_hello = jiffies; | |
4674 | + if (quorum_device->state == NODESTATE_DEAD) { | |
4675 | + quorum_device->state = NODESTATE_MEMBER; | |
4676 | + recalculate_quorum(0); | |
4677 | + } | |
4678 | + } | |
4679 | + else { | |
4680 | + if (quorum_device->state == NODESTATE_MEMBER) { | |
4681 | + quorum_device->state = NODESTATE_DEAD; | |
4682 | + recalculate_quorum(0); | |
4683 | + } | |
4684 | + } | |
c783755a | 4685 | + |
bb1d8b11 AM |
4686 | + return 0; |
4687 | +} | |
c783755a | 4688 | + |
bb1d8b11 AM |
4689 | +/* APIs for cluster ref counting. */ |
4690 | +int kcl_addref_cluster() | |
4691 | +{ | |
4692 | + int ret = -ENOTCONN; | |
c783755a | 4693 | + |
bb1d8b11 AM |
4694 | + if (!atomic_read(&cnxman_running)) |
4695 | + goto addref_ret; | |
c783755a | 4696 | + |
bb1d8b11 AM |
4697 | + if (try_module_get(THIS_MODULE)) { |
4698 | + atomic_inc(&use_count); | |
4699 | + ret = 0; | |
4700 | + } | |
c783755a | 4701 | + |
bb1d8b11 AM |
4702 | + addref_ret: |
4703 | + return ret; | |
4704 | +} | |
c783755a | 4705 | + |
bb1d8b11 AM |
4706 | +int kcl_releaseref_cluster() |
4707 | +{ | |
4708 | + if (!atomic_read(&cnxman_running)) | |
4709 | + return -ENOTCONN; | |
4710 | + atomic_dec(&use_count); | |
4711 | + module_put(THIS_MODULE); | |
4712 | + return 0; | |
4713 | +} | |
c783755a | 4714 | + |
bb1d8b11 AM |
4715 | +int kcl_cluster_name(char **cname) |
4716 | +{ | |
4717 | + char *name; | |
c783755a | 4718 | + |
bb1d8b11 AM |
4719 | + name = kmalloc(strlen(cluster_name) + 1, GFP_KERNEL); |
4720 | + if (!name) | |
4721 | + return -ENOMEM; | |
4722 | + | |
4723 | + strncpy(name, cluster_name, strlen(cluster_name)+1); | |
4724 | + *cname = name; | |
4725 | + return 0; | |
4726 | +} | |
4727 | + | |
4728 | +int kcl_get_current_interface(void) | |
4729 | +{ | |
4730 | + return current_interface->number; | |
4731 | +} | |
c783755a | 4732 | + |
bb1d8b11 AM |
4733 | +/* Socket registration stuff */ |
4734 | +static struct net_proto_family cl_family_ops = { | |
4735 | + .family = AF_CLUSTER, | |
4736 | + .create = cl_create, | |
4737 | + .owner = THIS_MODULE, | |
c783755a AM |
4738 | +}; |
4739 | + | |
bb1d8b11 AM |
4740 | +static struct proto_ops cl_proto_ops = { |
4741 | + .family = AF_CLUSTER, | |
c783755a | 4742 | + |
bb1d8b11 AM |
4743 | + .release = cl_release, |
4744 | + .bind = cl_bind, | |
4745 | + .connect = sock_no_connect, | |
4746 | + .socketpair = sock_no_socketpair, | |
4747 | + .accept = sock_no_accept, | |
4748 | + .getname = cl_getname, | |
4749 | + .poll = cl_poll, | |
4750 | + .ioctl = cl_ioctl, | |
4751 | + .listen = sock_no_listen, | |
4752 | + .shutdown = cl_shutdown, | |
4753 | + .setsockopt = sock_no_setsockopt, | |
4754 | + .getsockopt = sock_no_getsockopt, | |
4755 | + .sendmsg = cl_sendmsg, | |
4756 | + .recvmsg = cl_recvmsg, | |
4757 | + .mmap = sock_no_mmap, | |
4758 | + .sendpage = sock_no_sendpage, | |
4759 | + .owner = THIS_MODULE, | |
c783755a AM |
4760 | +}; |
4761 | + | |
bb1d8b11 AM |
4762 | +#ifdef MODULE |
4763 | +MODULE_DESCRIPTION("Cluster Connection and Service Manager"); | |
4764 | +MODULE_AUTHOR("Red Hat, Inc"); | |
4765 | +MODULE_LICENSE("GPL"); | |
4766 | +#endif | |
c783755a | 4767 | + |
bb1d8b11 AM |
4768 | +static int __init cluster_init(void) |
4769 | +{ | |
4770 | + printk("CMAN %s (built %s %s) installed\n", | |
4771 | + CMAN_RELEASE_NAME, __DATE__, __TIME__); | |
c783755a | 4772 | + |
bb1d8b11 AM |
4773 | + if (sock_register(&cl_family_ops)) { |
4774 | + printk(KERN_INFO "Unable to register cluster socket type\n"); | |
4775 | + return -1; | |
4776 | + } | |
c783755a | 4777 | + |
bb1d8b11 AM |
4778 | + /* allocate our sock slab cache */ |
4779 | + cluster_sk_cachep = kmem_cache_create("cluster_sock", | |
4780 | + sizeof (struct cluster_sock), 0, | |
4781 | + SLAB_HWCACHE_ALIGN, 0, 0); | |
4782 | + if (!cluster_sk_cachep) { | |
4783 | + printk(KERN_CRIT | |
4784 | + "cluster_init: Cannot create cluster_sock SLAB cache\n"); | |
4785 | + sock_unregister(AF_CLUSTER); | |
4786 | + return -1; | |
4787 | + } | |
c1c6733f | 4788 | + |
bb1d8b11 AM |
4789 | +#ifdef CONFIG_PROC_FS |
4790 | + create_proc_entries(); | |
4791 | +#endif | |
c1c6733f | 4792 | + |
bb1d8b11 AM |
4793 | + init_MUTEX(&start_thread_sem); |
4794 | + init_MUTEX(&send_lock); | |
4795 | + init_MUTEX(&barrier_list_lock); | |
4796 | + init_MUTEX(&cluster_members_lock); | |
4797 | + init_MUTEX(&port_array_lock); | |
4798 | + init_MUTEX(&messages_list_lock); | |
4799 | + init_MUTEX(&listenreq_lock); | |
4800 | + init_MUTEX(&client_socket_lock); | |
4801 | + init_MUTEX(&new_dead_node_lock); | |
4802 | + init_MUTEX(&event_listener_lock); | |
4803 | + init_MUTEX(&kernel_listener_lock); | |
4804 | + init_MUTEX(&tempnode_lock); | |
4805 | + spin_lock_init(&active_socket_lock); | |
4806 | + init_timer(&ack_timer); | |
c1c6733f | 4807 | + |
bb1d8b11 AM |
4808 | + INIT_LIST_HEAD(&event_listener_list); |
4809 | + INIT_LIST_HEAD(&kernel_listener_list); | |
4810 | + INIT_LIST_HEAD(&socket_list); | |
4811 | + INIT_LIST_HEAD(&client_socket_list); | |
4812 | + INIT_LIST_HEAD(&active_socket_list); | |
4813 | + INIT_LIST_HEAD(&barrier_list); | |
4814 | + INIT_LIST_HEAD(&messages_list); | |
4815 | + INIT_LIST_HEAD(&listenreq_list); | |
4816 | + INIT_LIST_HEAD(&cluster_members_list); | |
4817 | + INIT_LIST_HEAD(&new_dead_node_list); | |
4818 | + INIT_LIST_HEAD(&tempnode_list); | |
c1c6733f | 4819 | + |
bb1d8b11 | 4820 | + atomic_set(&cnxman_running, 0); |
c1c6733f | 4821 | + |
bb1d8b11 AM |
4822 | + sm_init(); |
4823 | + | |
4824 | + return 0; | |
c1c6733f AM |
4825 | +} |
4826 | + | |
bb1d8b11 AM |
4827 | +static void __exit cluster_exit(void) |
4828 | +{ | |
4829 | +#ifdef CONFIG_PROC_FS | |
4830 | + cleanup_proc_entries(); | |
4831 | +#endif | |
c1c6733f | 4832 | + |
bb1d8b11 AM |
4833 | + sock_unregister(AF_CLUSTER); |
4834 | + kmem_cache_destroy(cluster_sk_cachep); | |
4835 | +} | |
c1c6733f | 4836 | + |
bb1d8b11 AM |
4837 | +module_init(cluster_init); |
4838 | +module_exit(cluster_exit); | |
c1c6733f | 4839 | + |
bb1d8b11 AM |
4840 | +EXPORT_SYMBOL(kcl_sendmsg); |
4841 | +EXPORT_SYMBOL(kcl_register_read_callback); | |
4842 | +EXPORT_SYMBOL(kcl_add_callback); | |
4843 | +EXPORT_SYMBOL(kcl_remove_callback); | |
4844 | +EXPORT_SYMBOL(kcl_get_members); | |
4845 | +EXPORT_SYMBOL(kcl_get_member_ids); | |
4846 | +EXPORT_SYMBOL(kcl_get_all_members); | |
4847 | +EXPORT_SYMBOL(kcl_is_quorate); | |
4848 | +EXPORT_SYMBOL(kcl_get_node_by_addr); | |
4849 | +EXPORT_SYMBOL(kcl_get_node_by_name); | |
4850 | +EXPORT_SYMBOL(kcl_get_node_by_nodeid); | |
4851 | +EXPORT_SYMBOL(kcl_get_node_addresses); | |
4852 | +EXPORT_SYMBOL(kcl_addref_cluster); | |
4853 | +EXPORT_SYMBOL(kcl_releaseref_cluster); | |
4854 | +EXPORT_SYMBOL(kcl_cluster_name); | |
c1c6733f | 4855 | + |
bb1d8b11 AM |
4856 | +EXPORT_SYMBOL(kcl_barrier_register); |
4857 | +EXPORT_SYMBOL(kcl_barrier_setattr); | |
4858 | +EXPORT_SYMBOL(kcl_barrier_delete); | |
4859 | +EXPORT_SYMBOL(kcl_barrier_wait); | |
4860 | +EXPORT_SYMBOL(kcl_barrier_cancel); | |
c1c6733f | 4861 | + |
bb1d8b11 AM |
4862 | +EXPORT_SYMBOL(kcl_register_quorum_device); |
4863 | +EXPORT_SYMBOL(kcl_unregister_quorum_device); | |
4864 | +EXPORT_SYMBOL(kcl_quorum_device_available); | |
c1c6733f | 4865 | + |
bb1d8b11 AM |
4866 | +EXPORT_SYMBOL(kcl_register_service); |
4867 | +EXPORT_SYMBOL(kcl_unregister_service); | |
4868 | +EXPORT_SYMBOL(kcl_join_service); | |
4869 | +EXPORT_SYMBOL(kcl_leave_service); | |
4870 | +EXPORT_SYMBOL(kcl_global_service_id); | |
4871 | +EXPORT_SYMBOL(kcl_start_done); | |
4872 | +EXPORT_SYMBOL(kcl_get_services); | |
4873 | +EXPORT_SYMBOL(kcl_get_current_interface); | |
c1c6733f | 4874 | + |
bb1d8b11 AM |
4875 | +/* |
4876 | + * Overrides for Emacs so that we follow Linus's tabbing style. | |
4877 | + * Emacs will notice this stuff at the end of the file and automatically | |
4878 | + * adjust the settings for this buffer only. This must remain at the end | |
4879 | + * of the file. | |
4880 | + * --------------------------------------------------------------------------- | |
4881 | + * Local variables: | |
4882 | + * c-file-style: "linux" | |
4883 | + * End: | |
4884 | + */ | |
c1c6733f | 4885 | diff -urN linux-orig/cluster/cman/config.c linux-patched/cluster/cman/config.c |
bb1d8b11 AM |
4886 | --- linux-orig/cluster/cman/config.c 1970-01-01 07:30:00.000000000 +0730 |
4887 | +++ linux-patched/cluster/cman/config.c 2004-11-03 11:37:37.000000000 +0800 | |
c783755a | 4888 | @@ -0,0 +1,49 @@ |
c1c6733f AM |
4889 | +/****************************************************************************** |
4890 | +******************************************************************************* | |
4891 | +** | |
4892 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
4893 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4894 | +** | |
4895 | +** This copyrighted material is made available to anyone wishing to use, | |
4896 | +** modify, copy, or redistribute it subject to the terms and conditions | |
4897 | +** of the GNU General Public License v.2. | |
4898 | +** | |
4899 | +******************************************************************************* | |
4900 | +******************************************************************************/ | |
4901 | + | |
4902 | +#include "config.h" | |
4903 | + | |
4904 | +/* Config file defaults */ | |
4905 | + | |
b7b72b66 | 4906 | +#define DEFAULT_JOIN_WAIT_TIME 16 /* Time to wait while sending JOINREQ |
c1c6733f | 4907 | + * messages. Should be at least twice |
b7b72b66 | 4908 | + * the HELLO timer, probably 3x */ |
c1c6733f AM |
4909 | +#define DEFAULT_JOIN_TIMEOUT 30 /* How long we wait after getting a |
4910 | + * JOINACK to regarding that node as | |
4911 | + * dead */ | |
4912 | +#define DEFAULT_HELLO_TIMER 5 /* Period between HELLO messages */ | |
4913 | +#define DEFAULT_DEADNODE_TIMER 21 /* If we don't get a message from a | |
4914 | + * node in this period kill it */ | |
4915 | +#define DEFAULT_TRANSITION_TIMER 15 /* Maximum time a state transition | |
4916 | + * should take */ | |
4917 | +#define DEFAULT_JOINCONF_TIMER 5 /* Time allowed to a node to respond to | |
4918 | + * a JOINCONF message */ | |
4919 | +#define DEFAULT_MAX_NODES 128 /* Max allowed nodes */ | |
4920 | +#define DEFAULT_TRANSITION_RESTARTS 10 /* Maximum number of transition | |
4921 | + * restarts before we die */ | |
4922 | +#define DEFAULT_SM_DEBUG_SIZE 256 /* Size in bytes of SM debug buffer */ | |
4923 | + | |
c783755a AM |
4924 | +#define DEFAULT_NEWCLUSTER_TIMEOUT 16 /* Time to send NEWCLUSTER messages */ |
4925 | + | |
c1c6733f AM |
4926 | +struct config_info cman_config = { |
4927 | + .joinwait_timeout = DEFAULT_JOIN_WAIT_TIME, | |
4928 | + .joinconf_timeout = DEFAULT_JOINCONF_TIMER, | |
4929 | + .join_timeout = DEFAULT_JOIN_TIMEOUT, | |
4930 | + .hello_timer = DEFAULT_HELLO_TIMER, | |
4931 | + .deadnode_timeout = DEFAULT_DEADNODE_TIMER, | |
4932 | + .transition_timeout = DEFAULT_TRANSITION_TIMER, | |
4933 | + .transition_restarts = DEFAULT_TRANSITION_RESTARTS, | |
4934 | + .max_nodes = DEFAULT_MAX_NODES, | |
4935 | + .sm_debug_size = DEFAULT_SM_DEBUG_SIZE, | |
c783755a | 4936 | + .newcluster_timeout = DEFAULT_NEWCLUSTER_TIMEOUT, |
c1c6733f AM |
4937 | +}; |
4938 | diff -urN linux-orig/cluster/cman/config.h linux-patched/cluster/cman/config.h | |
bb1d8b11 AM |
4939 | --- linux-orig/cluster/cman/config.h 1970-01-01 07:30:00.000000000 +0730 |
4940 | +++ linux-patched/cluster/cman/config.h 2004-11-03 11:37:37.000000000 +0800 | |
c783755a | 4941 | @@ -0,0 +1,32 @@ |
c1c6733f AM |
4942 | +/****************************************************************************** |
4943 | +******************************************************************************* | |
4944 | +** | |
4945 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
4946 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4947 | +** | |
4948 | +** This copyrighted material is made available to anyone wishing to use, | |
4949 | +** modify, copy, or redistribute it subject to the terms and conditions | |
4950 | +** of the GNU General Public License v.2. | |
4951 | +** | |
4952 | +******************************************************************************* | |
4953 | +******************************************************************************/ | |
4954 | + | |
4955 | +#ifndef __CONFIG_DOT_H__ | |
4956 | +#define __CONFIG_DOT_H__ | |
4957 | + | |
4958 | +struct config_info { | |
4959 | + int joinwait_timeout; | |
4960 | + int joinconf_timeout; | |
4961 | + int join_timeout; | |
4962 | + int hello_timer; | |
4963 | + int deadnode_timeout; | |
4964 | + int transition_timeout; | |
4965 | + int transition_restarts; | |
4966 | + int max_nodes; | |
4967 | + int sm_debug_size; | |
c783755a | 4968 | + int newcluster_timeout; |
c1c6733f AM |
4969 | +}; |
4970 | + | |
4971 | +extern struct config_info cman_config; | |
4972 | + | |
4973 | +#endif /* __CONFIG_DOT_H__ */ | |
4974 | diff -urN linux-orig/cluster/cman/kjoin.c linux-patched/cluster/cman/kjoin.c | |
bb1d8b11 AM |
4975 | --- linux-orig/cluster/cman/kjoin.c 1970-01-01 07:30:00.000000000 +0730 |
4976 | +++ linux-patched/cluster/cman/kjoin.c 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
4977 | @@ -0,0 +1,238 @@ |
4978 | +/****************************************************************************** | |
4979 | +******************************************************************************* | |
4980 | +** | |
4981 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
4982 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4983 | +** | |
4984 | +** This copyrighted material is made available to anyone wishing to use, | |
4985 | +** modify, copy, or redistribute it subject to the terms and conditions | |
4986 | +** of the GNU General Public License v.2. | |
4987 | +** | |
4988 | +******************************************************************************* | |
4989 | +******************************************************************************/ | |
4990 | + | |
4991 | +#include <linux/socket.h> | |
4992 | +#include <net/sock.h> | |
4993 | +#include <linux/list.h> | |
4994 | +#include <cluster/cnxman.h> | |
4995 | +#include <linux/in.h> | |
4996 | + | |
4997 | +#include "cnxman-private.h" | |
4998 | + | |
4999 | +static struct socket *mcast_sock; | |
5000 | +static struct socket *recv_sock; | |
5001 | +static struct socket *cluster_sock; | |
5002 | + | |
5003 | +extern short cluster_id; | |
5004 | +extern int join_count; | |
5005 | +extern struct semaphore join_count_lock; | |
5006 | +extern atomic_t cnxman_running; | |
5007 | + | |
5008 | +int kcl_join_cluster(struct cl_join_cluster_info *join_info) | |
5009 | +{ | |
5010 | + int result; | |
5011 | + int one = 1, error; | |
5012 | + unsigned int ipaddr = join_info->ipaddr, brdaddr = join_info->brdaddr; | |
5013 | + unsigned short port = join_info->port; | |
5014 | + mm_segment_t fs; | |
5015 | + struct sockaddr_in saddr; | |
5016 | + struct kcl_multicast_sock mcast_info; | |
5017 | + | |
5018 | + down(&join_count_lock); | |
5019 | + if (atomic_read(&cnxman_running)) | |
5020 | + { | |
5021 | + error = 0; | |
5022 | + if (join_info->cluster_id == cluster_id) | |
5023 | + join_count++; | |
5024 | + else | |
5025 | + error = -EINVAL; | |
5026 | + up(&join_count_lock); | |
5027 | + return error; | |
5028 | + } | |
5029 | + up(&join_count_lock); | |
5030 | + | |
5031 | + result = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &mcast_sock); | |
5032 | + if (result < 0) | |
5033 | + { | |
5034 | + printk(KERN_ERR CMAN_NAME ": Can't create Multicast socket\n"); | |
5035 | + return result; | |
5036 | + } | |
5037 | + | |
5038 | + result = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &recv_sock); | |
5039 | + if (result < 0) | |
5040 | + { | |
5041 | + printk(KERN_ERR CMAN_NAME ": Can't create Receive socket\n"); | |
5042 | + return result; | |
5043 | + } | |
5044 | + | |
5045 | + fs = get_fs(); | |
5046 | + set_fs(get_ds()); | |
5047 | + | |
5048 | + if ((error = sock_setsockopt(mcast_sock, SOL_SOCKET, SO_BROADCAST, | |
5049 | + (void *) &one, sizeof (int)))) | |
5050 | + { | |
5051 | + set_fs(fs); | |
5052 | + printk("Error %d Setting master socket to SO_BROADCAST\n", | |
5053 | + error); | |
5054 | + sock_release(mcast_sock); | |
5055 | + return -1; | |
5056 | + } | |
5057 | + set_fs(fs); | |
5058 | + | |
5059 | + /* Bind the multicast socket */ | |
5060 | + saddr.sin_family = AF_INET; | |
5061 | + saddr.sin_port = htons(port); | |
5062 | + saddr.sin_addr.s_addr = cpu_to_be32(brdaddr); | |
5063 | + result = | |
5064 | + mcast_sock->ops->bind(mcast_sock, (struct sockaddr *) &saddr, | |
5065 | + sizeof (saddr)); | |
5066 | + if (result < 0) | |
5067 | + { | |
5068 | + printk(KERN_ERR CMAN_NAME ": Can't bind multicast socket\n"); | |
5069 | + sock_release(mcast_sock); | |
5070 | + sock_release(recv_sock); | |
5071 | + return result; | |
5072 | + } | |
5073 | + | |
5074 | + /* Bind the receive socket to our IP address */ | |
5075 | + saddr.sin_family = AF_INET; | |
5076 | + saddr.sin_port = htons(port); | |
5077 | + saddr.sin_addr.s_addr = cpu_to_be32(ipaddr); | |
5078 | + result = | |
5079 | + recv_sock->ops->bind(recv_sock, (struct sockaddr *) &saddr, | |
5080 | + sizeof (saddr)); | |
5081 | + if (result < 0) | |
5082 | + { | |
5083 | + printk(KERN_ERR CMAN_NAME ": Can't bind receive socket\n"); | |
5084 | + sock_release(mcast_sock); | |
5085 | + sock_release(recv_sock); | |
5086 | + return result; | |
5087 | + } | |
5088 | + | |
5089 | + /* Create the cluster master socket */ | |
5090 | + result = | |
5091 | + sock_create(AF_CLUSTER, SOCK_DGRAM, CLPROTO_MASTER, &cluster_sock); | |
5092 | + if (result < 0) | |
5093 | + { | |
5094 | + printk(KERN_ERR CMAN_NAME | |
5095 | + ": Can't create cluster master socket\n"); | |
5096 | + sock_release(mcast_sock); | |
5097 | + sock_release(recv_sock); | |
5098 | + return result; | |
5099 | + } | |
5100 | + | |
5101 | + /* This is the broadcast transmit address */ | |
5102 | + saddr.sin_addr.s_addr = cpu_to_be32(brdaddr); | |
5103 | + | |
5104 | + /* Pass the multicast socket to kernel space */ | |
5105 | + mcast_info.sock = mcast_sock; | |
5106 | + mcast_info.number = 1; | |
5107 | + | |
5108 | + fs = get_fs(); | |
5109 | + set_fs(get_ds()); | |
5110 | + | |
5111 | + if ((error = cluster_sock->ops->setsockopt(cluster_sock, CLPROTO_MASTER, | |
5112 | + KCL_SET_MULTICAST, | |
5113 | + (void *) &mcast_info, | |
5114 | + sizeof (mcast_info)))) | |
5115 | + { | |
5116 | + set_fs(fs); | |
5117 | + printk(CMAN_NAME | |
5118 | + ": Unable to pass multicast socket to cnxman, %d\n", | |
5119 | + error); | |
5120 | + sock_release(mcast_sock); | |
5121 | + sock_release(recv_sock); | |
5122 | + sock_release(cluster_sock); | |
5123 | + return -1; | |
5124 | + } | |
5125 | + | |
5126 | + mcast_info.sock = recv_sock; | |
5127 | + if ((error = | |
5128 | + cluster_sock->ops->setsockopt(cluster_sock, CLPROTO_MASTER, | |
5129 | + KCL_SET_RCVONLY, | |
5130 | + (void *) &mcast_info, | |
5131 | + sizeof (mcast_info)))) | |
5132 | + { | |
5133 | + set_fs(fs); | |
5134 | + printk(CMAN_NAME | |
5135 | + ": Unable to pass receive socket to cnxman, %d\n", | |
5136 | + error); | |
5137 | + sock_release(mcast_sock); | |
5138 | + sock_release(recv_sock); | |
5139 | + sock_release(cluster_sock); | |
5140 | + return -1; | |
5141 | + } | |
5142 | + | |
5143 | + /* This setsockopt expects usermode variables */ | |
5144 | + | |
5145 | + if (cluster_sock->ops-> | |
5146 | + setsockopt(cluster_sock, CLPROTO_MASTER, CLU_JOIN_CLUSTER, | |
5147 | + (void *) join_info, | |
5148 | + sizeof (struct cl_join_cluster_info))) | |
5149 | + | |
5150 | + { | |
5151 | + set_fs(fs); | |
5152 | + printk(CMAN_NAME ": Unable to join cluster\n"); | |
5153 | + sock_release(mcast_sock); | |
5154 | + sock_release(recv_sock); | |
5155 | + sock_release(cluster_sock); | |
5156 | + return -1; | |
5157 | + } | |
5158 | + set_fs(fs); | |
5159 | + | |
5160 | + return 0; | |
5161 | +} | |
5162 | + | |
5163 | +int kcl_leave_cluster(int remove) | |
5164 | +{ | |
5165 | + mm_segment_t fs; | |
5166 | + int rem = remove; | |
5167 | + int ret = 0; | |
5168 | + struct socket *shutdown_sock = cluster_sock; | |
5169 | + | |
5170 | + cluster_sock = NULL; | |
5171 | + | |
5172 | + if (!shutdown_sock) | |
5173 | + { | |
5174 | + /* Create the cluster master socket */ | |
5175 | + int result = | |
5176 | + sock_create(AF_CLUSTER, SOCK_DGRAM, CLPROTO_MASTER, | |
5177 | + &shutdown_sock); | |
5178 | + if (result < 0) | |
5179 | + { | |
5180 | + printk(KERN_ERR CMAN_NAME | |
5181 | + ": Can't create cluster master socket\n"); | |
5182 | + sock_release(mcast_sock); | |
5183 | + sock_release(recv_sock); | |
5184 | + return result; | |
5185 | + } | |
5186 | + } | |
5187 | + | |
5188 | + fs = get_fs(); | |
5189 | + set_fs(get_ds()); | |
5190 | + | |
5191 | + if ((ret = | |
5192 | + shutdown_sock->ops->setsockopt(shutdown_sock, CLPROTO_MASTER, | |
5193 | + CLU_LEAVE_CLUSTER, (void *) &rem, | |
5194 | + sizeof (int)))) | |
5195 | + { | |
5196 | + printk(KERN_ERR CMAN_NAME ": Unable to leave cluster, %d\n", | |
5197 | + ret); | |
5198 | + } | |
5199 | + set_fs(fs); | |
5200 | + | |
5201 | + sock_release(shutdown_sock); | |
5202 | + | |
5203 | + return ret; | |
5204 | +} | |
5205 | + | |
5206 | +/* | |
5207 | + * Overrides for Emacs so that we follow Linus's tabbing style. | |
5208 | + * Emacs will notice this stuff at the end of the file and automatically | |
5209 | + * adjust the settings for this buffer only. This must remain at the end | |
5210 | + * of the file. | |
5211 | + * --------------------------------------------------------------------------- | |
5212 | + * Local variables: | |
5213 | + * c-file-style: "linux" | |
5214 | + * End: | |
5215 | + */ | |
5216 | diff -urN linux-orig/cluster/cman/membership.c linux-patched/cluster/cman/membership.c | |
bb1d8b11 AM |
5217 | --- linux-orig/cluster/cman/membership.c 1970-01-01 07:30:00.000000000 +0730 |
5218 | +++ linux-patched/cluster/cman/membership.c 2004-11-03 11:37:37.000000000 +0800 | |
5219 | @@ -0,0 +1,3160 @@ | |
c1c6733f AM |
5220 | +/****************************************************************************** |
5221 | +******************************************************************************* | |
5222 | +** | |
5223 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5224 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
5225 | +** | |
5226 | +** This copyrighted material is made available to anyone wishing to use, | |
5227 | +** modify, copy, or redistribute it subject to the terms and conditions | |
5228 | +** of the GNU General Public License v.2. | |
5229 | +** | |
5230 | +******************************************************************************* | |
5231 | +******************************************************************************/ | |
5232 | + | |
5233 | +#include <linux/socket.h> | |
5234 | +#include <net/sock.h> | |
5235 | +#include <linux/slab.h> | |
5236 | +#include <linux/spinlock.h> | |
5237 | +#include <linux/vmalloc.h> | |
5238 | +#include <asm/uaccess.h> | |
5239 | +#include <linux/list.h> | |
5240 | +#include <cluster/cnxman.h> | |
5241 | + | |
5242 | +#include "cnxman-private.h" | |
5243 | +#include "config.h" | |
5244 | +#include "sm_control.h" | |
5245 | + | |
5246 | +#ifndef TRUE | |
5247 | +#define TRUE 1 | |
5248 | +#endif | |
5249 | + | |
5250 | +/* Barrier name for membership transitions. %d is the cluster generation number | |
5251 | + */ | |
5252 | +#define MEMBERSHIP_BARRIER_NAME "TRANSITION.%d" | |
5253 | + | |
5254 | +/* Variables also used by connection manager */ | |
5255 | +struct list_head cluster_members_list; | |
5256 | +struct semaphore cluster_members_lock; | |
5257 | +int cluster_members; /* Number of ACTIVE members, not a count of | |
5258 | + * nodes in the list */ | |
b7b72b66 | 5259 | +int we_are_a_cluster_member; |
c1c6733f | 5260 | +int cluster_is_quorate; |
b7b72b66 | 5261 | +int quit_threads; |
c1c6733f AM |
5262 | +struct task_struct *membership_task; |
5263 | +struct cluster_node *us; | |
5264 | + | |
5265 | +static struct task_struct *hello_task; | |
5266 | +static struct semaphore hello_task_lock; | |
5267 | + | |
5268 | +/* Variables that belong to the connection manager */ | |
5269 | +extern wait_queue_head_t cnxman_waitq; | |
5270 | +extern struct completion member_thread_comp; | |
5271 | +extern struct cluster_node *quorum_device; | |
5272 | +extern unsigned short two_node; | |
5273 | +extern char cluster_name[]; | |
5274 | +extern unsigned int config_version; | |
5275 | +extern unsigned int address_length; | |
5276 | + | |
5277 | +static struct socket *mem_socket; | |
5278 | +static pid_t kcluster_pid; | |
5279 | + | |
5280 | +static char iobuf[MAX_CLUSTER_MESSAGE]; | |
5281 | +static char scratchbuf[MAX_CLUSTER_MESSAGE + 100]; | |
5282 | + | |
5283 | +/* Our node name, usually system_utsname.nodename, but can be overridden */ | |
5284 | +char nodename[MAX_CLUSTER_MEMBER_NAME_LEN + 1]; | |
5285 | + | |
c783755a AM |
5286 | +/* Node ID that we want. defaults of zero means |
5287 | + * it will be allocated by the cluster join mechanism | |
5288 | + */ | |
5289 | +int wanted_nodeid; | |
5290 | + | |
c1c6733f | 5291 | +static spinlock_t members_by_nodeid_lock; |
b7b72b66 | 5292 | +static int sizeof_members_array; /* Can dynamically increase (vmalloc |
c1c6733f AM |
5293 | + * permitting) */ |
5294 | +static struct cluster_node **members_by_nodeid; | |
5295 | + | |
5296 | +#define MEMBER_INCREMENT_SIZE 10 | |
5297 | + | |
5298 | +static int votes = 1; /* Votes this node has */ | |
5299 | +static int expected_votes = 1; /* Total expected votes in the cluster */ | |
5300 | +static unsigned int quorum; /* Quorum, fewer votes than this and we stop | |
5301 | + * work */ | |
5302 | +static int leavereason; /* Saved for the duration of a state transition */ | |
5303 | +static int transitionreason; /* Reason this transition was initiated */ | |
5304 | +static unsigned int highest_nodeid; /* Highest node ID known to the cluster */ | |
5305 | +static struct timer_list transition_timer; /* Kicks in if the transition | |
5306 | + * doesn't complete in a | |
5307 | + * reasonable time */ | |
5308 | +static struct timer_list hello_timer; /* Timer to send HELLOs on */ | |
5309 | +static unsigned long join_time; /* The time that we got our JOIN-ACK */ | |
5310 | +static unsigned long start_time; /* The time that we were started */ | |
5311 | +static int joinconf_count; /* Number of JOINCONF messages we have sent to | |
5312 | + * a new node */ | |
5313 | +static unsigned long wake_flags;/* Reason we were woken */ | |
5314 | + | |
5315 | +/* Flags in above */ | |
5316 | +#define WAKE_FLAG_DEADNODE 1 | |
5317 | +#define WAKE_FLAG_TRANSTIMER 2 | |
5318 | + | |
5319 | +/* The time the transition finished */ | |
5320 | +static unsigned long transition_end_time; | |
5321 | + | |
5322 | +/* A list of nodes that cnxman tells us are dead. I hope this never has more | |
5323 | + * than one element in it but I can't take that chance. only non-static so it | |
5324 | + * can be initialised in module_load. */ | |
5325 | +struct list_head new_dead_node_list; | |
5326 | +struct semaphore new_dead_node_lock; | |
5327 | + | |
bb1d8b11 AM |
5328 | +static int do_membership_packet(struct msghdr *msg, char *buf, int len); |
5329 | +static int do_process_joinreq(struct msghdr *msg, char *buf, int len); | |
5330 | +static int do_process_joinack(struct msghdr *msg, char *buf, int len); | |
5331 | +static int do_process_joinconf(struct msghdr *msg, char *buf, int len); | |
5332 | +static int do_process_leave(struct msghdr *msg, char *buf, int len); | |
5333 | +static int do_process_hello(struct msghdr *msg, char *buf, int len); | |
5334 | +static int do_process_kill(struct msghdr *msg, char *buf, int len); | |
5335 | +static int do_process_reconfig(struct msghdr *msg, char *buf, int len); | |
5336 | +static int do_process_starttrans(struct msghdr *msg, char *buf, int len); | |
5337 | +static int do_process_masterview(struct msghdr *msg, char *buf, int len); | |
5338 | +static int do_process_endtrans(struct msghdr *msg, char *buf, int len); | |
5339 | +static int do_process_viewack(struct msghdr *msg, char *buf, int len); | |
5340 | +static int do_process_startack(struct msghdr *msg, char *buf, int len); | |
5341 | +static int do_process_newcluster(struct msghdr *msg, char *buf, int len); | |
5342 | +static int do_process_nominate(struct msghdr *msg, char *buf, int len); | |
c1c6733f | 5343 | +static int send_cluster_view(unsigned char cmd, struct sockaddr_cl *saddr, |
b7b72b66 | 5344 | + unsigned int flags, unsigned int flags2); |
c1c6733f AM |
5345 | +static int send_joinreq(struct sockaddr_cl *addr, int addr_len); |
5346 | +static int send_startack(struct sockaddr_cl *addr, int addr_len, int node_id); | |
5347 | +static int send_hello(void); | |
5348 | +static int send_master_hello(void); | |
5349 | +static int send_newcluster(void); | |
5350 | +static int end_transition(void); | |
5351 | +static int dispatch_messages(struct socket *mem_socket); | |
5352 | +static void check_for_dead_nodes(void); | |
5353 | +static void confirm_joiner(void); | |
5354 | +static void reset_hello_time(void); | |
5355 | +static int add_us(void); | |
5356 | +static int send_joinconf(void); | |
5357 | +static int init_membership_services(void); | |
5358 | +static int elect_master(struct cluster_node **); | |
5359 | +static void trans_timer_expired(unsigned long arg); | |
5360 | +static void hello_timer_expired(unsigned long arg); | |
5361 | +static void join_or_form_cluster(void); | |
5362 | +static int do_timer_wakeup(void); | |
5363 | +static int start_transition(unsigned char reason, struct cluster_node *node); | |
c783755a | 5364 | +static uint32_t low32_of_ip(void); |
c1c6733f AM |
5365 | +int send_leave(unsigned char); |
5366 | +int send_reconfigure(int, unsigned int); | |
5367 | + | |
5368 | +#ifdef DEBUG_MEMB | |
5369 | +static char *msgname(int msg); | |
5370 | +static int debug_sendmsg(struct socket *sock, void *buf, int size, | |
5371 | + struct sockaddr_cl *caddr, int addr_len, | |
5372 | + unsigned int flags) | |
5373 | +{ | |
5374 | + P_MEMB("%ld: sending %s, len=%d\n", jiffies, msgname(((char *) buf)[0]), | |
5375 | + size); | |
5376 | + return kcl_sendmsg(sock, buf, size, caddr, addr_len, flags); | |
5377 | +} | |
5378 | + | |
5379 | +#define kcl_sendmsg debug_sendmsg | |
5380 | +#endif | |
5381 | + | |
5382 | +/* State of the node */ | |
c783755a | 5383 | +static enum { STARTING, NEWCLUSTER, JOINING, JOINWAIT, JOINACK, TRANSITION, |
c1c6733f | 5384 | + TRANSITION_COMPLETE, MEMBER, REJECTED, LEFT_CLUSTER, MASTER |
c783755a | 5385 | +} node_state = LEFT_CLUSTER; |
c1c6733f AM |
5386 | + |
5387 | +/* Sub-state when we are MASTER */ | |
5388 | +static enum { MASTER_START, MASTER_COLLECT, MASTER_CONFIRM, | |
5389 | + MASTER_COMPLETE } master_state; | |
5390 | + | |
5391 | +/* Number of responses collected while a master controlling a state transition */ | |
5392 | +static int responses_collected; | |
5393 | +static int responses_expected; | |
5394 | + | |
5395 | +/* Current cluster generation number */ | |
c783755a | 5396 | +int cluster_generation = 1; |
c1c6733f AM |
5397 | + |
5398 | +/* When another node initiates a transtion then store it's pointer in here so | |
5399 | + * we can check for other nodes trying to spoof us */ | |
5400 | +static struct cluster_node *master_node = NULL; | |
5401 | + | |
5402 | +/* Struct the node wanting to join us */ | |
5403 | +static struct cluster_node *joining_node = NULL; | |
b7b72b66 | 5404 | +static int joining_temp_nodeid; |
c1c6733f AM |
5405 | + |
5406 | +/* Last time a HELLO message was sent */ | |
b7b72b66 | 5407 | +unsigned long last_hello; |
c1c6733f AM |
5408 | + |
5409 | +/* When we got our JOINWAIT or NEWCLUSTER */ | |
b7b72b66 | 5410 | +unsigned long joinwait_time; |
c1c6733f AM |
5411 | + |
5412 | +/* Number of times a transition has restarted when we were master */ | |
b7b72b66 | 5413 | +int transition_restarts; |
c1c6733f AM |
5414 | + |
5415 | +/* Variables used by the master to collect cluster status during a transition */ | |
b7b72b66 AM |
5416 | +static int agreeing_nodes; |
5417 | +static int dissenting_nodes; | |
c1c6733f AM |
5418 | +static uint8_t *node_opinion = NULL; |
5419 | +#define OPINION_AGREE 1 | |
5420 | +#define OPINION_DISAGREE 2 | |
5421 | + | |
5422 | +/* Set node id of a node, also add it to the members array and expand the array | |
5423 | + * if necessary */ | |
5424 | +static inline void set_nodeid(struct cluster_node *node, int nodeid) | |
5425 | +{ | |
5426 | + if (!nodeid) | |
5427 | + return; | |
5428 | + | |
5429 | + node->node_id = nodeid; | |
c783755a | 5430 | + if (nodeid >= sizeof_members_array) { |
c1c6733f | 5431 | + int new_size = sizeof_members_array + MEMBER_INCREMENT_SIZE; |
c783755a AM |
5432 | + struct cluster_node **new_array; |
5433 | + | |
5434 | + if (new_size < nodeid) | |
5435 | + new_size = nodeid + MEMBER_INCREMENT_SIZE; | |
5436 | + | |
5437 | + new_array = vmalloc((new_size) * sizeof (struct cluster_node *)); | |
c1c6733f AM |
5438 | + if (new_array) { |
5439 | + spin_lock(&members_by_nodeid_lock); | |
5440 | + memcpy(new_array, members_by_nodeid, | |
5441 | + sizeof_members_array * | |
5442 | + sizeof (struct cluster_node *)); | |
5443 | + memset(&new_array[sizeof_members_array], 0, | |
c783755a | 5444 | + (new_size - sizeof_members_array) * |
c1c6733f AM |
5445 | + sizeof (struct cluster_node *)); |
5446 | + vfree(members_by_nodeid); | |
c783755a | 5447 | + |
c1c6733f AM |
5448 | + members_by_nodeid = new_array; |
5449 | + sizeof_members_array = new_size; | |
5450 | + spin_unlock(&members_by_nodeid_lock); | |
5451 | + } | |
5452 | + else { | |
5453 | + panic("No memory for more nodes"); | |
5454 | + } | |
5455 | + } | |
5456 | + notify_kernel_listeners(NEWNODE, (long) nodeid); | |
5457 | + | |
5458 | + spin_lock(&members_by_nodeid_lock); | |
5459 | + members_by_nodeid[nodeid] = node; | |
5460 | + spin_unlock(&members_by_nodeid_lock); | |
5461 | +} | |
5462 | + | |
5463 | +static int hello_kthread(void *unused) | |
5464 | +{ | |
5465 | + struct task_struct *tsk = current; | |
5466 | + sigset_t tmpsig; | |
5467 | + | |
5468 | + daemonize("cman_hbeat"); | |
5469 | + | |
5470 | + /* Block everything but SIGKILL/SIGSTOP/SIGTERM */ | |
5471 | + siginitset(&tmpsig, SIGKILL | SIGSTOP | SIGTERM); | |
5472 | + sigprocmask(SIG_BLOCK, &tmpsig, NULL); | |
5473 | + | |
5474 | + down(&hello_task_lock); | |
5475 | + hello_task = tsk; | |
5476 | + up(&hello_task_lock); | |
5477 | + | |
5478 | + set_user_nice(current, -6); | |
5479 | + | |
5480 | + while (node_state != REJECTED && node_state != LEFT_CLUSTER) { | |
c1c6733f AM |
5481 | + |
5482 | + /* Scan the nodes list for dead nodes */ | |
5483 | + if (node_state == MEMBER) | |
5484 | + check_for_dead_nodes(); | |
5485 | + | |
5486 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
5487 | + schedule(); | |
5488 | + set_task_state(current, TASK_RUNNING); | |
b7b72b66 AM |
5489 | + |
5490 | + if (node_state != REJECTED && node_state != LEFT_CLUSTER) | |
5491 | + send_hello(); | |
c1c6733f AM |
5492 | + } |
5493 | + down(&hello_task_lock); | |
5494 | + hello_task = NULL; | |
5495 | + up(&hello_task_lock); | |
5496 | + P_MEMB("heartbeat closing down\n"); | |
5497 | + return 0; | |
5498 | +} | |
5499 | + | |
5500 | +/* This is the membership "daemon". A client of cnxman (but symbiotic with it) | |
5501 | + * that keeps track of and controls cluster membership. */ | |
5502 | +static int membership_kthread(void *unused) | |
5503 | +{ | |
5504 | + struct task_struct *tsk = current; | |
c1c6733f AM |
5505 | + sigset_t tmpsig; |
5506 | + | |
5507 | + daemonize("cman_memb"); | |
5508 | + | |
5509 | + /* Block everything but SIGKILL/SIGSTOP/SIGTERM */ | |
5510 | + siginitset(&tmpsig, SIGKILL | SIGSTOP | SIGTERM); | |
5511 | + sigprocmask(SIG_BLOCK, &tmpsig, NULL); | |
5512 | + | |
5513 | + membership_task = tsk; | |
5514 | + set_user_nice(current, -5); | |
5515 | + | |
5516 | + /* Open the socket */ | |
5517 | + if (init_membership_services()) | |
5518 | + return -1; | |
5519 | + | |
5520 | + add_us(); | |
5521 | + joining_node = us; | |
5522 | + | |
5523 | + init_timer(&hello_timer); | |
5524 | + hello_timer.function = hello_timer_expired; | |
5525 | + hello_timer.data = 0L; | |
5526 | + | |
5527 | + /* Do joining stuff */ | |
5528 | + join_or_form_cluster(); | |
5529 | + | |
5530 | + transition_end_time = jiffies; | |
5531 | + | |
5532 | + /* Main loop */ | |
5533 | + while (node_state != REJECTED && node_state != LEFT_CLUSTER) { | |
5534 | + | |
5535 | + struct task_struct *tsk = current; | |
5536 | + | |
5537 | + DECLARE_WAITQUEUE(wait, tsk); | |
5538 | + | |
5539 | + tsk->state = TASK_INTERRUPTIBLE; | |
5540 | + add_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5541 | + | |
5542 | + if (!skb_peek(&mem_socket->sk->sk_receive_queue) && | |
5543 | + wake_flags == 0) { | |
5544 | + if (node_state == JOINACK || | |
5545 | + node_state == JOINWAIT) | |
5546 | + schedule_timeout(HZ); | |
5547 | + else | |
5548 | + schedule(); | |
5549 | + } | |
5550 | + | |
5551 | + tsk->state = TASK_RUNNING; | |
5552 | + remove_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5553 | + | |
5554 | + /* Are we being shut down? */ | |
5555 | + if (node_state == LEFT_CLUSTER || quit_threads || | |
5556 | + signal_pending(current)) | |
5557 | + break; | |
5558 | + | |
5559 | + /* Were we woken by a dead node passed down from cnxman ? */ | |
5560 | + if (test_and_clear_bit(WAKE_FLAG_DEADNODE, &wake_flags)) { | |
5561 | + struct list_head *nodelist, *tmp; | |
5562 | + struct cl_new_dead_node *deadnode; | |
5563 | + | |
5564 | + down(&new_dead_node_lock); | |
5565 | + list_for_each_safe(nodelist, tmp, &new_dead_node_list) { | |
5566 | + deadnode = | |
5567 | + list_entry(nodelist, | |
5568 | + struct cl_new_dead_node, list); | |
5569 | + | |
5570 | + if (deadnode->node->state == NODESTATE_MEMBER) | |
5571 | + a_node_just_died(deadnode->node); | |
5572 | + list_del(&deadnode->list); | |
5573 | + kfree(deadnode); | |
5574 | + } | |
5575 | + up(&new_dead_node_lock); | |
5576 | + } | |
5577 | + | |
5578 | + /* Process received messages. If dispatch_message() returns an | |
5579 | + * error then we shut down */ | |
5580 | + if (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
5581 | + if (dispatch_messages(mem_socket) < 0) | |
5582 | + goto leave_cluster; | |
5583 | + | |
5584 | + } | |
5585 | + | |
5586 | + /* Were we woken by the transition timer firing ? */ | |
5587 | + if (test_and_clear_bit(WAKE_FLAG_TRANSTIMER, &wake_flags)) { | |
5588 | + switch (do_timer_wakeup()) { | |
5589 | + case -1: | |
5590 | + continue; | |
5591 | + case 0: | |
5592 | + break; | |
5593 | + case +1: | |
5594 | + goto leave_cluster; | |
5595 | + } | |
5596 | + } | |
5597 | + | |
5598 | + /* Got a JOINACK but no JOIN-CONF, start waiting for HELLO | |
5599 | + * messages again */ | |
c783755a AM |
5600 | + if (node_state == JOINACK && |
5601 | + time_after(jiffies, | |
5602 | + join_time + cman_config.join_timeout * HZ)) { | |
c1c6733f AM |
5603 | + P_MEMB |
5604 | + ("Waited a long time for a join-conf, going back to JOINWAIT state\n"); | |
5605 | + node_state = JOINWAIT; | |
5606 | + joinwait_time = jiffies; | |
5607 | + } | |
5608 | + | |
c783755a AM |
5609 | + /* Have we had an ACK for our JOINREQ message ? */ |
5610 | + if (node_state == JOINING && | |
5611 | + time_after(jiffies, | |
5612 | + join_time + cman_config.join_timeout * HZ)) { | |
5613 | + P_MEMB("didn't get JOINACK, going back to JOINWAIT\n"); | |
5614 | + node_state = JOINWAIT; | |
5615 | + joinwait_time = jiffies; | |
5616 | + } | |
5617 | + | |
c1c6733f | 5618 | + /* Have we been in joinwait for too long... */ |
c783755a AM |
5619 | + if (node_state == JOINWAIT && |
5620 | + time_after(jiffies, | |
5621 | + joinwait_time + cman_config.joinwait_timeout * HZ)) { | |
c1c6733f AM |
5622 | + printk(CMAN_NAME |
5623 | + ": Been in JOINWAIT for too long - giving up\n"); | |
5624 | + goto leave_cluster; | |
5625 | + } | |
5626 | + } | |
5627 | + | |
5628 | + leave_cluster: | |
5629 | + | |
5630 | + /* Wake up the heartbeat thread so it can exit */ | |
5631 | + down(&hello_task_lock); | |
5632 | + if (hello_task) | |
5633 | + wake_up_process(hello_task); | |
5634 | + up(&hello_task_lock); | |
5635 | + | |
5636 | + if (timer_pending(&hello_timer)) | |
5637 | + del_timer(&hello_timer); | |
5638 | + | |
5639 | + if (timer_pending(&transition_timer)) | |
5640 | + del_timer(&transition_timer); | |
5641 | + | |
5642 | + node_state = LEFT_CLUSTER; | |
5643 | + P_MEMB("closing down\n"); | |
5644 | + quit_threads = 1; /* force other thread to exit too */ | |
5645 | + | |
c783755a AM |
5646 | + send_leave(us->leave_reason); |
5647 | + sock_release(mem_socket); | |
c1c6733f AM |
5648 | + highest_nodeid = 0; |
5649 | + complete(&member_thread_comp); | |
5650 | + return 0; | |
5651 | +} | |
5652 | + | |
5653 | +/* Things to do in the main thread when the transition timer has woken us. | |
5654 | + * Usually this happens when a transition is taking too long and we need to | |
5655 | + * take remedial action. | |
5656 | + * | |
5657 | + * returns: -1 continue; 0 carry on processing +1 leave cluster; */ | |
5658 | +static int do_timer_wakeup() | |
5659 | +{ | |
5660 | + P_MEMB("Timer wakeup - checking for dead master node %ld\n", jiffies); | |
5661 | + | |
5662 | + /* Resend JOINCONF if it got lost on the wire */ | |
5663 | + if (node_state == MASTER && master_state == MASTER_CONFIRM) { | |
5664 | + mod_timer(&transition_timer, | |
5665 | + jiffies + cman_config.joinconf_timeout * HZ); | |
5666 | + if (++joinconf_count < MAX_RETRIES) { | |
5667 | + P_MEMB("Resending JOINCONF\n"); | |
5668 | + send_joinconf(); | |
5669 | + } | |
5670 | + else { | |
5671 | + P_MEMB("JOINCONF not acked, cancelling transition\n"); | |
5672 | + end_transition(); | |
5673 | + } | |
5674 | + return -1; | |
5675 | + } | |
5676 | + | |
5677 | + /* A joining node probably died */ | |
5678 | + if (cluster_members == 1) { | |
5679 | + end_transition(); | |
5680 | + return -1; | |
5681 | + } | |
5682 | + | |
5683 | + /* See if the master is still there */ | |
5684 | + if (node_state == TRANSITION || node_state == TRANSITION_COMPLETE) { | |
5685 | + | |
5686 | + /* If we are in transition and master_node is NULL then we are | |
5687 | + * waiting for ENDTRANS after JOIN-CONF */ | |
5688 | + if (!master_node) { | |
5689 | + /* Hmmm. master died after sending JOINCONF, we'll have | |
5690 | + * to die as we are in mid-transition */ | |
5691 | + printk(KERN_INFO CMAN_NAME | |
5692 | + ": Master died after JOINCONF, we must leave the cluster\n"); | |
5693 | + quit_threads = 1; | |
5694 | + return +1; | |
5695 | + } | |
5696 | + | |
5697 | + /* No messages from the master - see if it's stil there */ | |
5698 | + if (master_node->state == NODESTATE_MEMBER) { | |
5699 | + send_master_hello(); | |
5700 | + mod_timer(&transition_timer, | |
5701 | + jiffies + | |
5702 | + cman_config.transition_timeout * HZ); | |
5703 | + } | |
5704 | + | |
5705 | + /* If the master is dead then elect a new one */ | |
5706 | + if (master_node->state == NODESTATE_DEAD) { | |
5707 | + | |
5708 | + struct cluster_node *node; | |
5709 | + | |
5710 | + P_MEMB("Master node is dead...Election!\n"); | |
5711 | + if (elect_master(&node)) { | |
5712 | + | |
5713 | + /* We are master now, all kneel */ | |
5714 | + start_transition(TRANS_DEADMASTER, master_node); | |
5715 | + } | |
5716 | + else { | |
5717 | + /* Leave the job to someone on more pay */ | |
5718 | + master_node = node; | |
5719 | + mod_timer(&transition_timer, | |
5720 | + jiffies + | |
5721 | + cman_config.transition_timeout * HZ); | |
5722 | + } | |
5723 | + } | |
5724 | + } | |
5725 | + | |
5726 | + /* If we are the master node then restart the transition */ | |
5727 | + if (node_state == MASTER) { | |
5728 | + start_transition(TRANS_RESTART, us); | |
5729 | + } | |
5730 | + | |
5731 | + return 0; | |
5732 | +} | |
5733 | + | |
5734 | +static void form_cluster(void) | |
5735 | +{ | |
5736 | + printk(KERN_INFO CMAN_NAME ": forming a new cluster\n"); | |
5737 | + node_state = MEMBER; | |
5738 | + we_are_a_cluster_member = TRUE; | |
c1c6733f | 5739 | + us->state = NODESTATE_MEMBER; |
c783755a AM |
5740 | + if (wanted_nodeid) |
5741 | + set_nodeid(us, wanted_nodeid); | |
5742 | + else | |
5743 | + set_nodeid(us, 1); | |
c1c6733f AM |
5744 | + recalculate_quorum(0); |
5745 | + sm_member_update(cluster_is_quorate); | |
5746 | + send_hello(); | |
5747 | + kernel_thread(hello_kthread, NULL, 0); | |
5748 | + mod_timer(&hello_timer, jiffies + cman_config.hello_timer * HZ); | |
5749 | +} | |
5750 | + | |
5751 | +/* This does the initial JOIN part of the membership process. Actually most of | |
5752 | + * is done in the message processing routines but this is the main loop that | |
5753 | + * controls it. The side-effect of this routine is "node_state" which tells the | |
5754 | + * real main loop (in the kernel thread routine) what to do next */ | |
5755 | +static void join_or_form_cluster() | |
5756 | +{ | |
5757 | + start_time = jiffies; | |
5758 | + | |
5759 | + printk(KERN_INFO CMAN_NAME | |
5760 | + ": Waiting to join or form a Linux-cluster\n"); | |
c783755a AM |
5761 | + |
5762 | + restart_joinwait: | |
c1c6733f AM |
5763 | + join_time = 0; |
5764 | + start_time = jiffies; | |
5765 | + joinwait_time = jiffies; | |
5766 | + last_hello = 0; | |
c1c6733f | 5767 | + |
c783755a | 5768 | + /* Listen for HELLO or NEWCLUSTER messages */ |
c1c6733f AM |
5769 | + do { |
5770 | + DECLARE_WAITQUEUE(wait, current); | |
5771 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
5772 | + add_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5773 | + | |
5774 | + if (!skb_peek(&mem_socket->sk->sk_receive_queue)) | |
5775 | + schedule_timeout((cman_config.joinwait_timeout * HZ) / | |
5776 | + 5); | |
5777 | + | |
5778 | + set_task_state(current, TASK_RUNNING); | |
5779 | + remove_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5780 | + | |
5781 | + while (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
5782 | + dispatch_messages(mem_socket); | |
5783 | + } | |
5784 | + if (quit_threads) | |
5785 | + node_state = LEFT_CLUSTER; | |
5786 | + | |
5787 | + } | |
5788 | + while (time_before(jiffies, start_time + cman_config.joinwait_timeout * HZ) && | |
5789 | + node_state == STARTING); | |
5790 | + | |
c1c6733f | 5791 | + if (node_state == STARTING) { |
c783755a AM |
5792 | + start_time = jiffies; |
5793 | + joinwait_time = jiffies; | |
5794 | + node_state = NEWCLUSTER; | |
5795 | + } | |
5796 | + | |
5797 | + /* If we didn't hear any HELLO messages then start sending NEWCLUSTER messages */ | |
5798 | + while (time_before(jiffies, start_time + cman_config.newcluster_timeout * HZ) && | |
5799 | + node_state == NEWCLUSTER) { | |
5800 | + | |
5801 | + DECLARE_WAITQUEUE(wait, current); | |
5802 | + | |
5803 | + send_newcluster(); | |
5804 | + | |
5805 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
5806 | + add_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5807 | + | |
5808 | + if (!skb_peek(&mem_socket->sk->sk_receive_queue)) | |
5809 | + schedule_timeout((cman_config.joinwait_timeout * HZ) / | |
5810 | + 5); | |
5811 | + | |
5812 | + set_task_state(current, TASK_RUNNING); | |
5813 | + remove_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5814 | + | |
5815 | + while (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
5816 | + dispatch_messages(mem_socket); | |
5817 | + } | |
5818 | + /* Did we get a lower "NEWCLUSTER" message ? */ | |
5819 | + if (node_state == STARTING) { | |
5820 | + P_MEMB("NEWCLUSTER: restarting joinwait\n"); | |
5821 | + goto restart_joinwait; | |
5822 | + } | |
5823 | + | |
5824 | + if (quit_threads) | |
5825 | + node_state = LEFT_CLUSTER; | |
5826 | + | |
5827 | + } | |
5828 | + | |
5829 | + | |
5830 | + /* If we didn't hear any HELLO messages then form a new cluster */ | |
5831 | + if (node_state == NEWCLUSTER) { | |
c1c6733f AM |
5832 | + form_cluster(); |
5833 | + } | |
5834 | + else | |
5835 | + last_hello = jiffies; | |
5836 | + | |
5837 | +} | |
5838 | + | |
5839 | +int start_membership_services(pid_t cluster_pid) | |
5840 | +{ | |
5841 | + kcluster_pid = cluster_pid; | |
5842 | + | |
5843 | + init_timer(&transition_timer); | |
5844 | + transition_timer.function = trans_timer_expired; | |
5845 | + transition_timer.data = 0L; | |
5846 | + | |
5847 | + /* Start the thread */ | |
5848 | + return kernel_thread(membership_kthread, NULL, 0); | |
5849 | +} | |
5850 | + | |
5851 | +static int init_membership_services() | |
5852 | +{ | |
5853 | + int result; | |
5854 | + struct sockaddr_cl saddr; | |
5855 | + struct socket *sock; | |
5856 | + | |
5857 | + init_MUTEX(&hello_task_lock); | |
5858 | + /* Create a socket to communicate with */ | |
5859 | + result = sock_create_kern(AF_CLUSTER, SOCK_DGRAM, CLPROTO_CLIENT, &sock); | |
5860 | + if (result < 0) { | |
5861 | + printk(KERN_ERR CMAN_NAME | |
5862 | + ": Can't create cluster socket for membership services\n"); | |
5863 | + return result; | |
5864 | + } | |
5865 | + mem_socket = sock; | |
5866 | + | |
5867 | + /* Bind to our port */ | |
5868 | + saddr.scl_family = AF_CLUSTER; | |
5869 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5870 | + result = | |
5871 | + sock->ops->bind(sock, (struct sockaddr *) &saddr, sizeof (saddr)); | |
5872 | + if (result < 0) { | |
5873 | + printk(KERN_ERR CMAN_NAME | |
5874 | + ": Can't bind to cluster membership services port\n"); | |
5875 | + sock_release(sock); | |
5876 | + return result; | |
5877 | + } | |
5878 | + | |
5879 | + node_state = STARTING; | |
5880 | + return 0; | |
5881 | +} | |
5882 | + | |
5883 | +static int send_joinconf() | |
5884 | +{ | |
5885 | + struct sockaddr_cl saddr; | |
5886 | + int status; | |
5887 | + | |
5888 | + if (joining_temp_nodeid == 0) { | |
5889 | + BUG(); | |
5890 | + } | |
5891 | + | |
5892 | + master_state = MASTER_CONFIRM; | |
5893 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5894 | + saddr.scl_family = AF_CLUSTER; | |
5895 | + saddr.scl_nodeid = joining_temp_nodeid; | |
5896 | + status = send_cluster_view(CLUSTER_MEM_JOINCONF, &saddr, | |
b7b72b66 | 5897 | + MSG_NOACK, 0); |
c1c6733f AM |
5898 | + |
5899 | + if (status < 0) { | |
5900 | + printk("Error %d sending JOINCONF, aborting transition\n", status); | |
5901 | + end_transition(); | |
5902 | + } | |
5903 | + return status; | |
5904 | +} | |
5905 | + | |
5906 | +static int send_joinreq(struct sockaddr_cl *addr, int addr_len) | |
5907 | +{ | |
5908 | + char *msgbuf = scratchbuf; | |
5909 | + struct list_head *addrlist; | |
5910 | + int ptr = sizeof (struct cl_mem_join_msg); | |
5911 | + unsigned short num_addr = 0; | |
5912 | + struct cluster_node_addr *nodeaddr; | |
5913 | + struct cl_mem_join_msg *msg = (struct cl_mem_join_msg *) msgbuf; | |
5914 | + | |
5915 | + msg->cmd = CLUSTER_MEM_JOINREQ; | |
5916 | + msg->votes = votes; | |
5917 | + msg->expected_votes = cpu_to_le32(expected_votes); | |
c783755a AM |
5918 | + msg->nodeid = cpu_to_le32(wanted_nodeid); |
5919 | + msg->major_version = cpu_to_le32(CNXMAN_MAJOR_VERSION); | |
5920 | + msg->minor_version = cpu_to_le32(CNXMAN_MINOR_VERSION); | |
5921 | + msg->patch_version = cpu_to_le32(CNXMAN_PATCH_VERSION); | |
c1c6733f AM |
5922 | + msg->config_version = cpu_to_le32(config_version); |
5923 | + msg->addr_len = cpu_to_le32(address_length); | |
5924 | + strcpy(msg->clustername, cluster_name); | |
5925 | + | |
5926 | + /* Add our addresses */ | |
5927 | + list_for_each(addrlist, &us->addr_list) { | |
5928 | + nodeaddr = list_entry(addrlist, struct cluster_node_addr, list); | |
5929 | + | |
5930 | + memcpy(msgbuf + ptr, nodeaddr->addr, address_length); | |
5931 | + ptr += address_length; | |
5932 | + num_addr++; | |
5933 | + } | |
5934 | + msg->num_addr = cpu_to_le16(num_addr); | |
5935 | + | |
5936 | + /* And our name */ | |
5937 | + strcpy(msgbuf + ptr, nodename); | |
5938 | + ptr += strlen(nodename) + 1; | |
5939 | + | |
5940 | + return kcl_sendmsg(mem_socket, msgbuf, ptr, | |
5941 | + addr, addr_len, MSG_NOACK); | |
5942 | +} | |
5943 | + | |
5944 | +static int send_startack(struct sockaddr_cl *addr, int addr_len, int node_id) | |
5945 | +{ | |
5946 | + struct cl_mem_startack_msg msg; | |
5947 | + | |
5948 | + msg.cmd = CLUSTER_MEM_STARTACK; | |
5949 | + msg.generation = cpu_to_le32(cluster_generation); | |
5950 | + msg.node_id = cpu_to_le32(node_id); | |
5951 | + msg.highest_node_id = cpu_to_le32(get_highest_nodeid()); | |
5952 | + | |
b7b72b66 | 5953 | + return kcl_sendmsg(mem_socket, &msg, sizeof (msg), addr, addr_len, MSG_REPLYEXP); |
c1c6733f AM |
5954 | +} |
5955 | + | |
5956 | +static int send_newcluster() | |
5957 | +{ | |
c783755a AM |
5958 | + char buf[5]; |
5959 | + uint32_t lowip; | |
c1c6733f AM |
5960 | + |
5961 | + buf[0] = CLUSTER_MEM_NEWCLUSTER; | |
c783755a AM |
5962 | + lowip = cpu_to_le32(low32_of_ip()); |
5963 | + memcpy(&buf[1], &lowip, sizeof(lowip)); | |
c1c6733f | 5964 | + |
c783755a AM |
5965 | + return kcl_sendmsg(mem_socket, buf, sizeof(uint32_t)+1, |
5966 | + NULL, 0, | |
c1c6733f AM |
5967 | + MSG_NOACK); |
5968 | +} | |
5969 | + | |
5970 | +static int send_hello() | |
5971 | +{ | |
5972 | + struct cl_mem_hello_msg hello_msg; | |
5973 | + int status; | |
5974 | + | |
5975 | + hello_msg.cmd = CLUSTER_MEM_HELLO; | |
5976 | + hello_msg.members = cpu_to_le16(cluster_members); | |
b7b72b66 | 5977 | + hello_msg.flags = cluster_is_quorate ? HELLO_FLAG_QUORATE : 0; |
c1c6733f AM |
5978 | + hello_msg.generation = cpu_to_le32(cluster_generation); |
5979 | + | |
b7b72b66 AM |
5980 | + status = kcl_sendmsg(mem_socket, &hello_msg, |
5981 | + sizeof(struct cl_mem_hello_msg), | |
5982 | + NULL, 0, MSG_NOACK | MSG_ALLINT); | |
c1c6733f AM |
5983 | + |
5984 | + last_hello = jiffies; | |
5985 | + | |
5986 | + return status; | |
5987 | +} | |
5988 | + | |
5989 | +/* This is a special HELLO message that requires an ACK. clients in transition | |
b7b72b66 | 5990 | + * send these to the master to check it is still alive. If it does not ACK then |
c1c6733f AM |
5991 | + * cnxman will signal it dead and we can restart the transition */ |
5992 | +static int send_master_hello() | |
5993 | +{ | |
5994 | + struct cl_mem_hello_msg hello_msg; | |
5995 | + int status; | |
5996 | + struct sockaddr_cl saddr; | |
5997 | + | |
5998 | + hello_msg.cmd = CLUSTER_MEM_HELLO; | |
5999 | + hello_msg.members = cpu_to_le16(cluster_members); | |
b7b72b66 AM |
6000 | + hello_msg.flags = HELLO_FLAG_MASTER | |
6001 | + (cluster_is_quorate ? HELLO_FLAG_QUORATE : 0); | |
c1c6733f AM |
6002 | + hello_msg.generation = cpu_to_le32(cluster_generation); |
6003 | + | |
6004 | + saddr.scl_family = AF_CLUSTER; | |
6005 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
6006 | + saddr.scl_nodeid = master_node->node_id; | |
b7b72b66 AM |
6007 | + |
6008 | + status = kcl_sendmsg(mem_socket, &hello_msg, | |
6009 | + sizeof(struct cl_mem_hello_msg), | |
6010 | + &saddr, sizeof (saddr), 0); | |
c1c6733f AM |
6011 | + |
6012 | + last_hello = jiffies; | |
6013 | + | |
6014 | + return status; | |
6015 | +} | |
6016 | + | |
6017 | +/* Called when the transition timer has expired, meaning we sent a transition | |
6018 | + * message that was not ACKed */ | |
6019 | +static void trans_timer_expired(unsigned long arg) | |
6020 | +{ | |
6021 | + P_MEMB("Transition timer fired %ld\n", jiffies); | |
6022 | + | |
6023 | + set_bit(WAKE_FLAG_TRANSTIMER, &wake_flags); | |
6024 | + wake_up_process(membership_task); | |
6025 | +} | |
6026 | + | |
6027 | +static void hello_timer_expired(unsigned long arg) | |
6028 | +{ | |
6029 | + P_MEMB("Hello timer fired %ld\n", jiffies); | |
6030 | + | |
6031 | + mod_timer(&hello_timer, jiffies + cman_config.hello_timer * HZ); | |
6032 | + | |
6033 | + if (node_state >= TRANSITION) { | |
6034 | + wake_up_process(hello_task); | |
6035 | + } | |
6036 | +} | |
6037 | + | |
6038 | +static int wait_for_completion_barrier(void) | |
6039 | +{ | |
6040 | + int status; | |
6041 | + char barriername[MAX_BARRIER_NAME_LEN]; | |
6042 | + | |
6043 | + sprintf(barriername, MEMBERSHIP_BARRIER_NAME, cluster_generation); | |
6044 | + | |
6045 | + /* Make sure we all complete together */ | |
6046 | + P_MEMB("Waiting for completion barrier: %d members\n", cluster_members); | |
6047 | + if ((status = | |
6048 | + kcl_barrier_register(barriername, 0, cluster_members)) < 0) { | |
6049 | + printk(CMAN_NAME ": Error registering barrier: %d\n", status); | |
6050 | + return -1; | |
6051 | + } | |
6052 | + kcl_barrier_setattr(barriername, BARRIER_SETATTR_TIMEOUT, | |
6053 | + cman_config.transition_timeout); | |
6054 | + status = kcl_barrier_wait(barriername); | |
6055 | + kcl_barrier_delete(barriername); | |
6056 | + | |
6057 | + P_MEMB("Completion barrier reached : status = %d\n", status); | |
6058 | + return status; | |
6059 | +} | |
6060 | + | |
6061 | +/* Called at the end of a state transition when we are the master */ | |
6062 | +static int end_transition() | |
6063 | +{ | |
6064 | + struct cl_mem_endtrans_msg msg; | |
6065 | + int total_votes; | |
6066 | + int status; | |
6067 | + | |
6068 | + /* Cancel the timer */ | |
6069 | + del_timer(&transition_timer); | |
6070 | + | |
6071 | + confirm_joiner(); | |
6072 | + | |
6073 | + quorum = calculate_quorum(leavereason, 0, &total_votes); | |
6074 | + | |
6075 | + msg.cmd = CLUSTER_MEM_ENDTRANS; | |
6076 | + msg.quorum = cpu_to_le32(quorum); | |
6077 | + msg.generation = cpu_to_le32(++cluster_generation); | |
6078 | + msg.total_votes = cpu_to_le32(total_votes); | |
6079 | + if (joining_node && transitionreason == TRANS_NEWNODE) { | |
6080 | + msg.new_node_id = cpu_to_le32(joining_node->node_id); | |
6081 | + } | |
6082 | + else { | |
6083 | + msg.new_node_id = 0; | |
6084 | + } | |
6085 | + status = kcl_sendmsg(mem_socket, &msg, sizeof (msg), NULL, 0, 0); | |
6086 | + | |
6087 | + /* When that's all settled down, do the transition completion barrier */ | |
6088 | + kcl_wait_for_all_acks(); | |
6089 | + | |
6090 | + if (wait_for_completion_barrier() != 0) { | |
6091 | + P_MEMB("Barrier timed out - restart\n"); | |
6092 | + start_transition(TRANS_RESTART, us); | |
6093 | + return 0; | |
6094 | + } | |
6095 | + | |
b7b72b66 AM |
6096 | + joining_temp_nodeid = 0; |
6097 | + purge_temp_nodeids(); | |
6098 | + | |
c1c6733f AM |
6099 | + set_quorate(total_votes); |
6100 | + | |
6101 | + notify_listeners(); | |
6102 | + reset_hello_time(); | |
6103 | + | |
6104 | + /* Tell any waiting barriers that we had a transition */ | |
6105 | + check_barrier_returns(); | |
6106 | + | |
6107 | + leavereason = 0; | |
6108 | + node_state = MEMBER; | |
6109 | + transition_end_time = jiffies; | |
6110 | + | |
6111 | + sm_member_update(cluster_is_quorate); | |
6112 | + | |
6113 | + return 0; | |
6114 | +} | |
6115 | + | |
6116 | +int send_reconfigure(int param, unsigned int value) | |
6117 | +{ | |
6118 | + char msgbuf[66]; | |
6119 | + struct cl_mem_reconfig_msg *msg = | |
6120 | + (struct cl_mem_reconfig_msg *) &msgbuf; | |
6121 | + | |
6122 | + if (param == RECONFIG_PARAM_EXPECTED_VOTES && expected_votes > value) | |
6123 | + expected_votes = value; | |
6124 | + | |
6125 | + msg->cmd = CLUSTER_MEM_RECONFIG; | |
6126 | + msg->param = param; | |
6127 | + msg->value = cpu_to_le32(value); | |
6128 | + | |
6129 | + return kcl_sendmsg(mem_socket, &msgbuf, sizeof (*msg), NULL, 0, 0); | |
6130 | +} | |
6131 | + | |
6132 | +static int send_joinack(char *addr, int addr_len, unsigned char acktype) | |
6133 | +{ | |
6134 | + struct cl_mem_joinack_msg msg; | |
6135 | + | |
6136 | + msg.cmd = CLUSTER_MEM_JOINACK; | |
6137 | + msg.acktype = acktype; | |
6138 | + | |
6139 | + return kcl_sendmsg(mem_socket, &msg, sizeof (msg), | |
6140 | + (struct sockaddr_cl *)addr, addr_len, MSG_NOACK); | |
6141 | +} | |
6142 | + | |
6143 | +/* Only send a leave message to one node in the cluster so that it can master | |
6144 | + * the state transition, otherwise we get a "thundering herd" of potential | |
6145 | + * masters fighting it out */ | |
6146 | +int send_leave(unsigned char flags) | |
6147 | +{ | |
6148 | + unsigned char msg[2]; | |
6149 | + struct sockaddr_cl saddr; | |
6150 | + struct cluster_node *node = NULL; | |
6151 | + int status; | |
6152 | + | |
6153 | + if (!mem_socket) | |
c783755a | 6154 | + return 0; |
c1c6733f AM |
6155 | + |
6156 | + saddr.scl_family = AF_CLUSTER; | |
6157 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
6158 | + | |
6159 | + /* If we are in transition then use the current master */ | |
6160 | + if (node_state == TRANSITION) { | |
6161 | + node = master_node; | |
6162 | + } | |
6163 | + if (!node) { | |
6164 | + /* If we are the master or not in transition then pick a node | |
6165 | + * almost at random */ | |
6166 | + struct list_head *nodelist; | |
6167 | + | |
6168 | + down(&cluster_members_lock); | |
6169 | + list_for_each(nodelist, &cluster_members_list) { | |
6170 | + node = list_entry(nodelist, struct cluster_node, list); | |
6171 | + | |
6172 | + if (node->state == NODESTATE_MEMBER && !node->us) | |
6173 | + break; | |
6174 | + } | |
6175 | + up(&cluster_members_lock); | |
6176 | + } | |
6177 | + | |
6178 | + /* we are the only member of the cluster - there is no-one to tell */ | |
6179 | + if (node && !node->us) { | |
6180 | + saddr.scl_nodeid = node->node_id; | |
6181 | + | |
6182 | + P_MEMB("Sending LEAVE to %s\n", node->name); | |
6183 | + msg[0] = CLUSTER_MEM_LEAVE; | |
6184 | + msg[1] = flags; | |
b7b72b66 AM |
6185 | + status = kcl_sendmsg(mem_socket, msg, 2, |
6186 | + &saddr, sizeof (saddr), | |
6187 | + MSG_NOACK); | |
c1c6733f AM |
6188 | + if (status < 0) |
6189 | + return status; | |
6190 | + } | |
6191 | + | |
6192 | + /* And exit */ | |
6193 | + node_state = LEFT_CLUSTER; | |
6194 | + wake_up_process(membership_task); | |
6195 | + return 0; | |
6196 | +} | |
6197 | + | |
6198 | +int send_kill(int nodeid) | |
6199 | +{ | |
6200 | + char killmsg; | |
6201 | + struct sockaddr_cl saddr; | |
6202 | + | |
6203 | + killmsg = CLUSTER_MEM_KILL; | |
6204 | + | |
6205 | + saddr.scl_family = AF_CLUSTER; | |
6206 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
6207 | + saddr.scl_nodeid = nodeid; | |
6208 | + return kcl_sendmsg(mem_socket, &killmsg, 1, &saddr, | |
6209 | + sizeof (struct sockaddr_cl), MSG_NOACK); | |
6210 | +} | |
6211 | + | |
6212 | +/* Process a message */ | |
bb1d8b11 | 6213 | +static int do_membership_packet(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
6214 | +{ |
6215 | + int result = -1; | |
c1c6733f AM |
6216 | + struct sockaddr_cl *saddr = msg->msg_name; |
6217 | + struct cluster_node *node; | |
6218 | + | |
6219 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
6220 | + | |
6221 | + P_MEMB("got membership message : %s, from (%d) %s, len = %d\n", | |
6222 | + msgname(*buf), saddr->scl_nodeid, node ? node->name : "unknown", len); | |
6223 | + | |
6224 | + switch (*buf) { | |
6225 | + case CLUSTER_MEM_JOINREQ: | |
bb1d8b11 | 6226 | + result = do_process_joinreq(msg, buf, len); |
c1c6733f AM |
6227 | + break; |
6228 | + | |
6229 | + case CLUSTER_MEM_LEAVE: | |
6230 | + if (we_are_a_cluster_member) | |
bb1d8b11 | 6231 | + result = do_process_leave(msg, buf, len); |
c1c6733f AM |
6232 | + break; |
6233 | + | |
6234 | + case CLUSTER_MEM_HELLO: | |
bb1d8b11 | 6235 | + result = do_process_hello(msg, buf, len); |
c1c6733f AM |
6236 | + break; |
6237 | + | |
6238 | + case CLUSTER_MEM_KILL: | |
6239 | + if (we_are_a_cluster_member) | |
bb1d8b11 | 6240 | + result = do_process_kill(msg, buf, len); |
c1c6733f AM |
6241 | + break; |
6242 | + | |
6243 | + case CLUSTER_MEM_JOINCONF: | |
6244 | + if (node_state == JOINACK) { | |
bb1d8b11 | 6245 | + do_process_joinconf(msg, buf, len); |
c1c6733f AM |
6246 | + } |
6247 | + break; | |
6248 | + | |
6249 | + case CLUSTER_MEM_CONFACK: | |
6250 | + if (node_state == MASTER && master_state == MASTER_CONFIRM) { | |
6251 | + end_transition(); | |
6252 | + } | |
6253 | + break; | |
6254 | + | |
6255 | + case CLUSTER_MEM_MASTERVIEW: | |
6256 | + if (node_state == TRANSITION) | |
bb1d8b11 | 6257 | + do_process_masterview(msg, buf, len); |
c1c6733f AM |
6258 | + break; |
6259 | + | |
6260 | + case CLUSTER_MEM_JOINACK: | |
b7b72b66 AM |
6261 | + if (node_state == JOINING || node_state == JOINWAIT || |
6262 | + node_state == JOINACK) { | |
bb1d8b11 | 6263 | + do_process_joinack(msg, buf, len); |
c1c6733f AM |
6264 | + } |
6265 | + break; | |
6266 | + case CLUSTER_MEM_RECONFIG: | |
6267 | + if (we_are_a_cluster_member) { | |
bb1d8b11 | 6268 | + do_process_reconfig(msg, buf, len); |
c1c6733f AM |
6269 | + } |
6270 | + break; | |
6271 | + | |
6272 | + case CLUSTER_MEM_STARTTRANS: | |
bb1d8b11 | 6273 | + result = do_process_starttrans(msg, buf, len); |
c1c6733f AM |
6274 | + break; |
6275 | + | |
6276 | + case CLUSTER_MEM_ENDTRANS: | |
bb1d8b11 | 6277 | + result = do_process_endtrans(msg, buf, len); |
c1c6733f AM |
6278 | + break; |
6279 | + | |
6280 | + case CLUSTER_MEM_VIEWACK: | |
b7b72b66 | 6281 | + if (node_state == MASTER && master_state == MASTER_COLLECT) |
bb1d8b11 | 6282 | + result = do_process_viewack(msg, buf, len); |
c1c6733f AM |
6283 | + break; |
6284 | + | |
6285 | + case CLUSTER_MEM_STARTACK: | |
6286 | + if (node_state == MASTER) | |
bb1d8b11 | 6287 | + result = do_process_startack(msg, buf, len); |
c1c6733f AM |
6288 | + break; |
6289 | + | |
6290 | + case CLUSTER_MEM_NEWCLUSTER: | |
bb1d8b11 | 6291 | + result = do_process_newcluster(msg, buf, len); |
c1c6733f AM |
6292 | + break; |
6293 | + | |
6294 | + case CLUSTER_MEM_NOMINATE: | |
6295 | + if (node_state != MASTER) | |
bb1d8b11 | 6296 | + result = do_process_nominate(msg, buf, len); |
c1c6733f AM |
6297 | + break; |
6298 | + | |
6299 | + default: | |
6300 | + printk(KERN_ERR CMAN_NAME | |
bb1d8b11 AM |
6301 | + ": Unknown membership services message %d received from node %d port %d\n", |
6302 | + *buf, saddr->scl_nodeid, saddr->scl_port); | |
c1c6733f AM |
6303 | + break; |
6304 | + | |
6305 | + } | |
6306 | + return result; | |
6307 | +} | |
6308 | + | |
6309 | +/* Returns -ve to reject membership of the cluster 0 to accept membership +ve | |
6310 | + * to ignore request (node already joining) */ | |
6311 | +static int check_duplicate_node(char *name, struct msghdr *msg, int len) | |
6312 | +{ | |
6313 | + struct cluster_node *node; | |
6314 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *)msg->msg_name; | |
6315 | + char addr[address_length]; | |
6316 | + int addrlen; | |
6317 | + | |
6318 | + if (strlen(name) >= MAX_CLUSTER_MEMBER_NAME_LEN) | |
6319 | + return -3; | |
6320 | + | |
6321 | + /* See if we already have a cluster member with that name... */ | |
6322 | + node = find_node_by_name(name); | |
6323 | + if (node && node->state != NODESTATE_DEAD) { | |
6324 | + | |
c783755a | 6325 | + if (node->state == NODESTATE_JOINING) |
c1c6733f AM |
6326 | + return +1; |
6327 | + | |
6328 | + printk(KERN_WARNING CMAN_NAME | |
6329 | + ": Rejecting cluster membership application from %s - already have a node with that name\n", | |
6330 | + name); | |
6331 | + return -1; | |
6332 | + | |
6333 | + } | |
6334 | + | |
6335 | + /* Need to check the node's address too */ | |
6336 | + if (get_addr_from_temp_nodeid(saddr->scl_nodeid, addr, &addrlen) && | |
6337 | + (node = find_node_by_addr(addr, addrlen)) && | |
6338 | + node->state != NODESTATE_DEAD) { | |
6339 | + | |
c783755a | 6340 | + if (node->state == NODESTATE_JOINING) |
c1c6733f AM |
6341 | + return +1; |
6342 | + | |
6343 | + printk(KERN_WARNING CMAN_NAME | |
6344 | + ": Rejecting cluster membership application from %s - already have a node with that address\n", | |
6345 | + name); | |
6346 | + return -1; | |
6347 | + } | |
6348 | + return 0; | |
6349 | +} | |
6350 | + | |
6351 | +/* Start the state transition */ | |
6352 | +static int start_transition(unsigned char reason, struct cluster_node *node) | |
6353 | +{ | |
6354 | + char *startbuf = scratchbuf; | |
6355 | + struct cl_mem_starttrans_msg *msg = | |
6356 | + (struct cl_mem_starttrans_msg *) startbuf; | |
6357 | + | |
6358 | + P_MEMB("Start transition - reason = %d\n", reason); | |
6359 | + | |
6360 | + /* If this is a restart then zero the counters */ | |
6361 | + if (reason == TRANS_RESTART) { | |
6362 | + agreeing_nodes = 0; | |
6363 | + dissenting_nodes = 0; | |
6364 | + if (node_opinion) { | |
6365 | + kfree(node_opinion); | |
6366 | + node_opinion = NULL; | |
6367 | + } | |
6368 | + responses_collected = 0; | |
6369 | + } | |
6370 | + | |
6371 | + /* If we have timed out too many times then just die */ | |
6372 | + if (reason == TRANS_RESTART | |
6373 | + && ++transition_restarts > cman_config.transition_restarts) { | |
6374 | + printk(KERN_WARNING CMAN_NAME | |
6375 | + ": too many transition restarts - will die\n"); | |
c783755a | 6376 | + us->leave_reason = CLUSTER_LEAVEFLAG_INCONSISTENT; |
c1c6733f AM |
6377 | + node_state = LEFT_CLUSTER; |
6378 | + quit_threads = 1; | |
6379 | + wake_up_process(membership_task); | |
6380 | + wake_up_interruptible(&cnxman_waitq); | |
6381 | + return 0; | |
6382 | + } | |
6383 | + if (reason != TRANS_RESTART) | |
6384 | + transition_restarts = 0; | |
6385 | + | |
6386 | + /* Only keep the original state transition reason in the global | |
6387 | + * variable. */ | |
6388 | + if (reason != TRANS_ANOTHERREMNODE && reason != TRANS_NEWMASTER && | |
6389 | + reason != TRANS_RESTART && reason != TRANS_DEADMASTER) | |
6390 | + transitionreason = reason; | |
6391 | + | |
6392 | + /* Save the info of the requesting node */ | |
6393 | + if (reason == TRANS_NEWNODE) | |
6394 | + joining_node = node; | |
6395 | + | |
6396 | + node_state = MASTER; | |
6397 | + master_state = MASTER_START; | |
6398 | + responses_collected = 0; | |
6399 | + responses_expected = cluster_members - 1; | |
6400 | + | |
6401 | + /* If we are on our own then just do it */ | |
6402 | + if (responses_expected == 0) { | |
6403 | + P_MEMB("We are on our own...lonely here\n"); | |
6404 | + responses_collected--; | |
bb1d8b11 | 6405 | + do_process_startack(NULL, NULL, 0); |
c1c6733f AM |
6406 | + } |
6407 | + else { | |
6408 | + int ptr = sizeof (struct cl_mem_starttrans_msg); | |
6409 | + struct list_head *addrlist; | |
6410 | + unsigned short num_addrs = 0; | |
b7b72b66 | 6411 | + int flags = MSG_REPLYEXP; |
c1c6733f AM |
6412 | + |
6413 | + /* Send the STARTTRANS message */ | |
6414 | + msg->cmd = CLUSTER_MEM_STARTTRANS; | |
6415 | + msg->reason = reason; | |
6416 | + msg->votes = node->votes; | |
6417 | + msg->expected_votes = cpu_to_le32(node->expected_votes); | |
6418 | + msg->generation = cpu_to_le32(++cluster_generation); | |
6419 | + msg->nodeid = cpu_to_le32(node->node_id); | |
6420 | + | |
6421 | + if (reason == TRANS_NEWNODE) { | |
6422 | + /* Add the addresses */ | |
6423 | + list_for_each(addrlist, &node->addr_list) { | |
6424 | + struct cluster_node_addr *nodeaddr = | |
6425 | + list_entry(addrlist, | |
6426 | + struct cluster_node_addr, list); | |
6427 | + | |
6428 | + memcpy(startbuf + ptr, nodeaddr->addr, | |
6429 | + address_length); | |
6430 | + ptr += address_length; | |
6431 | + num_addrs++; | |
6432 | + } | |
6433 | + | |
6434 | + /* And the name */ | |
6435 | + strcpy(startbuf + ptr, node->name); | |
6436 | + ptr += strlen(node->name) + 1; | |
6437 | + } | |
6438 | + | |
6439 | + /* If another node died then we must queue the STARTTRANS | |
6440 | + * messages so that membershipd can carry on processing the | |
6441 | + * other replies */ | |
6442 | + if (reason == TRANS_ANOTHERREMNODE) | |
6443 | + flags |= MSG_QUEUE; | |
6444 | + | |
6445 | + msg->num_addrs = cpu_to_le16(num_addrs); | |
6446 | + kcl_sendmsg(mem_socket, msg, ptr, NULL, 0, flags); | |
6447 | + } | |
6448 | + /* Set a timer in case we don't get 'em all back */ | |
6449 | + mod_timer(&transition_timer, | |
6450 | + jiffies + cman_config.transition_timeout * HZ); | |
6451 | + return 0; | |
6452 | +} | |
6453 | + | |
6454 | +/* A node has died - decide what to do */ | |
6455 | +void a_node_just_died(struct cluster_node *node) | |
6456 | +{ | |
6457 | + /* If we are not in the context of kmembershipd then stick it on the | |
6458 | + * list and wake it */ | |
6459 | + if (current != membership_task) { | |
6460 | + struct cl_new_dead_node *newnode = | |
6461 | + kmalloc(sizeof (struct cl_new_dead_node), GFP_KERNEL); | |
6462 | + if (!newnode) | |
6463 | + return; | |
6464 | + newnode->node = node; | |
6465 | + down(&new_dead_node_lock); | |
6466 | + list_add_tail(&newnode->list, &new_dead_node_list); | |
6467 | + set_bit(WAKE_FLAG_DEADNODE, &wake_flags); | |
6468 | + up(&new_dead_node_lock); | |
6469 | + wake_up_process(membership_task); | |
6470 | + P_MEMB("Passing dead node %s onto kmembershipd\n", node->name); | |
6471 | + return; | |
6472 | + } | |
6473 | + | |
6474 | + /* Remove it */ | |
6475 | + down(&cluster_members_lock); | |
6476 | + if (node->state == NODESTATE_MEMBER) | |
6477 | + cluster_members--; | |
6478 | + node->state = NODESTATE_DEAD; | |
6479 | + up(&cluster_members_lock); | |
6480 | + | |
6481 | + /* Notify listeners */ | |
6482 | + notify_kernel_listeners(DIED, (long) node->node_id); | |
6483 | + | |
6484 | + /* If we are in normal operation then become master and initiate a | |
6485 | + * state-transition */ | |
6486 | + if (node_state == MEMBER) { | |
6487 | + start_transition(TRANS_REMNODE, node); | |
6488 | + return; | |
6489 | + } | |
6490 | + | |
6491 | + /* If we are a slave in transition then see if it's the master that has | |
6492 | + * failed. If not then ignore it. If it /is/ the master then elect a | |
6493 | + * new one */ | |
6494 | + if (node_state == TRANSITION) { | |
6495 | + if (master_node == node) { | |
6496 | + if (elect_master(&node)) { | |
6497 | + del_timer(&transition_timer); | |
6498 | + node_state = MASTER; | |
6499 | + | |
6500 | + start_transition(TRANS_DEADMASTER, master_node); | |
6501 | + } | |
6502 | + else { | |
6503 | + /* Someone else can be in charge - phew! */ | |
6504 | + } | |
6505 | + } | |
6506 | + return; | |
6507 | + } | |
6508 | + | |
6509 | + /* If we are the master then we need to start the transition all over | |
6510 | + * again */ | |
6511 | + if (node_state == MASTER) { | |
6512 | + /* Cancel timer */ | |
6513 | + del_timer(&transition_timer); | |
6514 | + | |
6515 | + /* Restart the transition */ | |
6516 | + start_transition(TRANS_ANOTHERREMNODE, node); | |
6517 | + transition_restarts = 0; | |
6518 | + return; | |
6519 | + } | |
6520 | +} | |
6521 | + | |
6522 | +/* | |
6523 | + * Build up and send a set of messages consisting of the whole cluster view. | |
6524 | + * The first byte is the command (cmd as passed in), the second is a flag byte: | |
6525 | + * bit 0 is set in the first message, bit 1 in the last (NOTE both may be set if | |
6526 | + * this is the only message sent The rest is a set of packed node entries, which | |
6527 | + * are NOT split over packets. */ | |
6528 | +static int send_cluster_view(unsigned char cmd, struct sockaddr_cl *saddr, | |
b7b72b66 | 6529 | + unsigned int flags, unsigned int flags2) |
c1c6733f AM |
6530 | +{ |
6531 | + int ptr = 2; | |
6532 | + int len; | |
6533 | + int status = 0; | |
6534 | + int last_node_start = 2; | |
6535 | + unsigned char first_packet_flag = 1; | |
6536 | + struct list_head *nodelist; | |
6537 | + struct list_head *temp; | |
6538 | + struct cluster_node *node; | |
6539 | + char *message = scratchbuf; | |
6540 | + | |
6541 | + message[0] = cmd; | |
6542 | + | |
6543 | + down(&cluster_members_lock); | |
6544 | + list_for_each_safe(nodelist, temp, &cluster_members_list) { | |
6545 | + node = list_entry(nodelist, struct cluster_node, list); | |
6546 | + | |
c783755a | 6547 | + if (node->state == NODESTATE_MEMBER || node->state == NODESTATE_DEAD) { |
c1c6733f AM |
6548 | + unsigned int evotes; |
6549 | + unsigned int node_id; | |
6550 | + unsigned short num_addrs = 0; | |
6551 | + unsigned short num_addrs_le; | |
6552 | + struct list_head *addrlist; | |
6553 | + | |
6554 | + last_node_start = ptr; | |
6555 | + | |
6556 | + message[ptr++] = len = strlen(node->name); | |
6557 | + strcpy(&message[ptr], node->name); | |
6558 | + ptr += len; | |
6559 | + | |
c783755a AM |
6560 | + message[ptr++] = node->state; |
6561 | + | |
c1c6733f AM |
6562 | + /* Count the number of addresses this node has */ |
6563 | + list_for_each(addrlist, &node->addr_list) { | |
6564 | + num_addrs++; | |
6565 | + } | |
6566 | + | |
6567 | + num_addrs_le = cpu_to_le16(num_addrs); | |
6568 | + memcpy(&message[ptr], &num_addrs_le, sizeof (short)); | |
6569 | + ptr += sizeof (short); | |
6570 | + | |
6571 | + /* Pack em in */ | |
6572 | + list_for_each(addrlist, &node->addr_list) { | |
6573 | + | |
6574 | + struct cluster_node_addr *nodeaddr = | |
c783755a AM |
6575 | + list_entry(addrlist, |
6576 | + struct cluster_node_addr, list); | |
c1c6733f AM |
6577 | + |
6578 | + memcpy(&message[ptr], nodeaddr->addr, | |
6579 | + address_length); | |
6580 | + ptr += address_length; | |
6581 | + } | |
6582 | + | |
6583 | + message[ptr++] = node->votes; | |
6584 | + | |
6585 | + evotes = cpu_to_le32(node->expected_votes); | |
6586 | + memcpy(&message[ptr], &evotes, sizeof (int)); | |
6587 | + ptr += sizeof (int); | |
6588 | + | |
6589 | + node_id = cpu_to_le32(node->node_id); | |
6590 | + memcpy(&message[ptr], &node_id, sizeof (int)); | |
6591 | + ptr += sizeof (int); | |
6592 | + | |
6593 | + /* If the block is full then send it */ | |
6594 | + if (ptr > MAX_CLUSTER_MESSAGE) { | |
6595 | + message[1] = first_packet_flag; | |
6596 | + | |
6597 | + up(&cluster_members_lock); | |
c783755a AM |
6598 | + status = kcl_sendmsg(mem_socket, message, |
6599 | + last_node_start, saddr, | |
6600 | + saddr ? sizeof (struct sockaddr_cl) : 0, | |
6601 | + flags); | |
c1c6733f AM |
6602 | + |
6603 | + if (status < 0) | |
6604 | + goto send_fail; | |
6605 | + | |
6606 | + down(&cluster_members_lock); | |
6607 | + | |
6608 | + first_packet_flag = 0; | |
6609 | + /* Copy the overflow back to the start of the | |
6610 | + * buffer for the next send */ | |
6611 | + memcpy(&message[2], &message[last_node_start], | |
6612 | + ptr - last_node_start); | |
6613 | + ptr = ptr - last_node_start + 2; | |
6614 | + } | |
6615 | + } | |
6616 | + } | |
6617 | + | |
6618 | + up(&cluster_members_lock); | |
6619 | + | |
6620 | + message[1] = first_packet_flag | 2; /* The last may also be first */ | |
6621 | + status = kcl_sendmsg(mem_socket, message, ptr, | |
6622 | + saddr, saddr ? sizeof (struct sockaddr_cl) : 0, | |
b7b72b66 | 6623 | + flags | flags2); |
c1c6733f AM |
6624 | + send_fail: |
6625 | + | |
6626 | + return status; | |
6627 | +} | |
6628 | + | |
6629 | +/* Make the JOINING node into a MEMBER */ | |
6630 | +static void confirm_joiner() | |
6631 | +{ | |
6632 | + if (joining_node && joining_node->state == NODESTATE_JOINING) { | |
6633 | + down(&cluster_members_lock); | |
6634 | + joining_node->state = NODESTATE_MEMBER; | |
6635 | + cluster_members++; | |
6636 | + up(&cluster_members_lock); | |
6637 | + } | |
c1c6733f AM |
6638 | +} |
6639 | + | |
6640 | +/* Reset HELLO timers for all nodes We do this after a state-transition as we | |
6641 | + * have had HELLOS disabled during the transition and if we don't do this the | |
6642 | + * nodes will go on an uncontrolled culling-spree afterwards */ | |
6643 | +static void reset_hello_time() | |
6644 | +{ | |
6645 | + struct list_head *nodelist; | |
6646 | + struct cluster_node *node; | |
6647 | + | |
6648 | + down(&cluster_members_lock); | |
6649 | + list_for_each(nodelist, &cluster_members_list) { | |
6650 | + node = list_entry(nodelist, struct cluster_node, list); | |
6651 | + | |
6652 | + if (node->state == NODESTATE_MEMBER) { | |
6653 | + node->last_hello = jiffies; | |
6654 | + } | |
6655 | + | |
6656 | + } | |
6657 | + up(&cluster_members_lock); | |
6658 | +} | |
6659 | + | |
6660 | +/* Calculate the new quorum and return the value. do *not* set it in here as | |
6661 | + * cnxman calls this to check if a new expected_votes value is valid. It | |
6662 | + * (optionally) returns the total number of votes in the cluster */ | |
6663 | +int calculate_quorum(int allow_decrease, int max_expected, int *ret_total_votes) | |
6664 | +{ | |
6665 | + struct list_head *nodelist; | |
6666 | + struct cluster_node *node; | |
6667 | + unsigned int total_votes = 0; | |
6668 | + unsigned int highest_expected = 0; | |
6669 | + unsigned int newquorum, q1, q2; | |
6670 | + | |
6671 | + down(&cluster_members_lock); | |
6672 | + list_for_each(nodelist, &cluster_members_list) { | |
6673 | + node = list_entry(nodelist, struct cluster_node, list); | |
6674 | + | |
6675 | + if (node->state == NODESTATE_MEMBER) { | |
6676 | + highest_expected = | |
6677 | + max(highest_expected, node->expected_votes); | |
6678 | + total_votes += node->votes; | |
6679 | + } | |
6680 | + } | |
6681 | + up(&cluster_members_lock); | |
6682 | + if (quorum_device && quorum_device->state == NODESTATE_MEMBER) | |
6683 | + total_votes += quorum_device->votes; | |
6684 | + | |
6685 | + if (max_expected > 0) | |
6686 | + highest_expected = max_expected; | |
6687 | + | |
6688 | + /* This quorum calculation is taken from the OpenVMS Cluster Systems | |
6689 | + * manual, but, then, you guessed that didn't you */ | |
6690 | + q1 = (highest_expected + 2) / 2; | |
6691 | + q2 = (total_votes + 2) / 2; | |
6692 | + newquorum = max(q1, q2); | |
6693 | + | |
6694 | + /* Normally quorum never decreases but the system administrator can | |
6695 | + * force it down by setting expected votes to a maximum value */ | |
6696 | + if (!allow_decrease) | |
6697 | + newquorum = max(quorum, newquorum); | |
6698 | + | |
6699 | + /* The special two_node mode allows each of the two nodes to retain | |
6700 | + * quorum if the other fails. Only one of the two should live past | |
6701 | + * fencing (as both nodes try to fence each other in split-brain.) */ | |
6702 | + if (two_node) | |
6703 | + newquorum = 1; | |
6704 | + | |
6705 | + if (ret_total_votes) | |
6706 | + *ret_total_votes = total_votes; | |
6707 | + return newquorum; | |
6708 | +} | |
6709 | + | |
6710 | +/* Recalculate cluster quorum, set quorate and notify changes */ | |
6711 | +void recalculate_quorum(int allow_decrease) | |
6712 | +{ | |
6713 | + int total_votes; | |
6714 | + | |
6715 | + quorum = calculate_quorum(allow_decrease, 0, &total_votes); | |
6716 | + set_quorate(total_votes); | |
6717 | + notify_listeners(); | |
6718 | +} | |
6719 | + | |
6720 | +/* Add new node address to an existing node */ | |
6721 | +int add_node_address(struct cluster_node *node, unsigned char *addr, int len) | |
6722 | +{ | |
6723 | + struct cluster_node_addr *newaddr; | |
6724 | + | |
6725 | + newaddr = kmalloc(sizeof (struct cluster_node_addr), GFP_KERNEL); | |
6726 | + if (!newaddr) | |
6727 | + return -1; | |
6728 | + | |
6729 | + memcpy(newaddr->addr, addr, len); | |
6730 | + newaddr->addr_len = len; | |
6731 | + list_add_tail(&newaddr->list, &node->addr_list); | |
6732 | + | |
6733 | + return 0; | |
6734 | +} | |
6735 | + | |
6736 | +static struct cluster_node *add_new_node(char *name, unsigned char votes, | |
6737 | + unsigned int expected_votes, | |
6738 | + int node_id, int state) | |
6739 | +{ | |
6740 | + struct cluster_node *newnode; | |
6741 | + | |
6742 | + /* Look for a dead node with this name */ | |
6743 | + newnode = find_node_by_name(name); | |
6744 | + | |
6745 | + /* Is it already joining */ | |
6746 | + if (newnode && newnode->state == NODESTATE_JOINING) | |
6747 | + return NULL; | |
6748 | + | |
6749 | + /* Update existing information */ | |
6750 | + if (newnode && newnode->state == NODESTATE_DEAD) { | |
6751 | + newnode->last_hello = jiffies; | |
6752 | + newnode->votes = votes; | |
6753 | + newnode->expected_votes = expected_votes; | |
6754 | + newnode->state = state; | |
6755 | + newnode->us = 0; | |
6756 | + newnode->leave_reason = 0; | |
6757 | + newnode->last_seq_recv = 0; | |
6758 | + newnode->last_seq_acked = 0; | |
6759 | + newnode->last_seq_sent = 0; | |
6760 | + newnode->incarnation++; | |
c783755a | 6761 | + do_gettimeofday(&newnode->join_time); |
c1c6733f AM |
6762 | + /* Don't overwrite the node ID */ |
6763 | + | |
6764 | + if (state == NODESTATE_MEMBER) { | |
6765 | + down(&cluster_members_lock); | |
6766 | + cluster_members++; | |
6767 | + up(&cluster_members_lock); | |
6768 | + } | |
6769 | + | |
6770 | + printk(KERN_INFO CMAN_NAME ": node %s rejoining\n", name); | |
6771 | + return newnode; | |
6772 | + } | |
6773 | + | |
6774 | + newnode = kmalloc(sizeof (struct cluster_node), GFP_KERNEL); | |
6775 | + if (!newnode) | |
6776 | + goto alloc_err; | |
6777 | + | |
6778 | + memset(newnode, 0, sizeof (struct cluster_node)); | |
6779 | + newnode->name = kmalloc(strlen(name) + 1, GFP_KERNEL); | |
6780 | + if (!newnode->name) | |
6781 | + goto alloc_err1; | |
6782 | + | |
6783 | + strcpy(newnode->name, name); | |
6784 | + newnode->last_hello = jiffies; | |
6785 | + newnode->votes = votes; | |
6786 | + newnode->expected_votes = expected_votes; | |
6787 | + newnode->state = state; | |
6788 | + newnode->node_id = node_id; | |
6789 | + newnode->us = 0; | |
6790 | + newnode->leave_reason = 0; | |
6791 | + newnode->last_seq_recv = 0; | |
6792 | + newnode->last_seq_acked = 0; | |
6793 | + newnode->last_seq_sent = 0; | |
6794 | + newnode->incarnation = 0; | |
c783755a | 6795 | + do_gettimeofday(&newnode->join_time); |
c1c6733f AM |
6796 | + INIT_LIST_HEAD(&newnode->addr_list); |
6797 | + set_nodeid(newnode, node_id); | |
6798 | + | |
6799 | + /* Add the new node to the list */ | |
6800 | + down(&cluster_members_lock); | |
6801 | + list_add(&newnode->list, &cluster_members_list); | |
6802 | + if (state == NODESTATE_MEMBER) | |
6803 | + cluster_members++; | |
6804 | + up(&cluster_members_lock); | |
6805 | + | |
6806 | + printk(KERN_INFO CMAN_NAME ": got node %s\n", name); | |
6807 | + return newnode; | |
6808 | + | |
6809 | + alloc_err1: | |
6810 | + kfree(newnode); | |
6811 | + alloc_err: | |
6812 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
6813 | + | |
6814 | + printk(KERN_CRIT CMAN_NAME | |
6815 | + ": Cannot allocate memory for new cluster node %s\n", name); | |
6816 | + | |
6817 | + panic("cluster memory allocation failed"); | |
6818 | + | |
6819 | + return NULL; | |
6820 | +} | |
6821 | + | |
6822 | +/* Remove node from a STARTTRANS message */ | |
6823 | +static struct cluster_node *remove_node(int nodeid) | |
6824 | +{ | |
6825 | + struct cluster_node *node = find_node_by_nodeid(nodeid); | |
6826 | + | |
6827 | + if (node && node->state == NODESTATE_MEMBER) { | |
6828 | + P_MEMB("starttrans removes node %s\n", node->name); | |
6829 | + down(&cluster_members_lock); | |
6830 | + node->state = NODESTATE_DEAD; | |
6831 | + cluster_members--; | |
6832 | + up(&cluster_members_lock); | |
6833 | + | |
6834 | + notify_kernel_listeners(DIED, (long) nodeid); | |
6835 | + | |
6836 | + /* If this node is us then go quietly */ | |
6837 | + if (node->us) { | |
6838 | + printk(KERN_INFO CMAN_NAME | |
6839 | + ": killed by STARTTRANS or NOMINATE\n"); | |
c783755a | 6840 | + node_state = LEFT_CLUSTER; |
c1c6733f AM |
6841 | + quit_threads = 1; |
6842 | + wake_up_process(membership_task); | |
6843 | + wake_up_interruptible(&cnxman_waitq); | |
6844 | + } | |
6845 | + } | |
6846 | + return node; | |
6847 | +} | |
6848 | + | |
6849 | +/* Add a node from a STARTTRANS or NOMINATE message */ | |
bb1d8b11 | 6850 | +static void add_node_from_starttrans(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
6851 | +{ |
6852 | + /* Add the new node but don't fill in the ID until the master has | |
6853 | + * confirmed it */ | |
6854 | + struct cl_mem_starttrans_msg *startmsg = | |
bb1d8b11 | 6855 | + (struct cl_mem_starttrans_msg *)buf; |
c1c6733f | 6856 | + int ptr = sizeof (struct cl_mem_starttrans_msg); |
c1c6733f | 6857 | + int i; |
bb1d8b11 AM |
6858 | + char *name = buf + ptr + le16_to_cpu(startmsg->num_addrs) * address_length; |
6859 | + char *nodeaddr = buf + sizeof(struct cl_mem_starttrans_msg); | |
c1c6733f AM |
6860 | + |
6861 | + joining_node = add_new_node(name, startmsg->votes, | |
6862 | + le32_to_cpu(startmsg->expected_votes), | |
6863 | + 0, NODESTATE_JOINING); | |
6864 | + | |
6865 | + /* add_new_node returns NULL if the node already exists */ | |
6866 | + if (!joining_node) | |
6867 | + joining_node = find_node_by_name(name); | |
6868 | + | |
6869 | + /* Add the node's addresses */ | |
6870 | + if (list_empty(&joining_node->addr_list)) { | |
6871 | + for (i = 0; i < le16_to_cpu(startmsg->num_addrs); i++) { | |
bb1d8b11 | 6872 | + add_node_address(joining_node, buf + ptr, address_length); |
c1c6733f AM |
6873 | + ptr += address_length; |
6874 | + } | |
6875 | + } | |
b7b72b66 AM |
6876 | + |
6877 | + /* Make sure we have a temp nodeid for the new node in case we | |
6878 | + become master */ | |
6879 | + joining_temp_nodeid = new_temp_nodeid(nodeaddr, | |
6880 | + address_length); | |
c1c6733f AM |
6881 | +} |
6882 | + | |
6883 | +/* We have been nominated as master for a transition */ | |
bb1d8b11 | 6884 | +static int do_process_nominate(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
6885 | +{ |
6886 | + struct cl_mem_starttrans_msg *startmsg = | |
bb1d8b11 | 6887 | + (struct cl_mem_starttrans_msg *)buf; |
c1c6733f | 6888 | + struct cluster_node *node = NULL; |
c1c6733f AM |
6889 | + |
6890 | + P_MEMB("nominate reason is %d\n", startmsg->reason); | |
6891 | + | |
6892 | + if (startmsg->reason == TRANS_REMNODE) { | |
6893 | + node = remove_node(le32_to_cpu(startmsg->nodeid)); | |
6894 | + } | |
6895 | + | |
6896 | + if (startmsg->reason == TRANS_NEWNODE) { | |
bb1d8b11 | 6897 | + add_node_from_starttrans(msg, buf, len); |
c1c6733f | 6898 | + node = joining_node; |
c1c6733f AM |
6899 | + } |
6900 | + | |
6901 | + /* This should be a TRANS_CHECK but start_transition needs some node | |
6902 | + * info */ | |
6903 | + if (node == NULL) | |
6904 | + node = us; | |
6905 | + start_transition(startmsg->reason, node); | |
6906 | + return 0; | |
6907 | +} | |
6908 | + | |
6909 | +/* Got a STARTACK response from a node */ | |
bb1d8b11 | 6910 | +static int do_process_startack(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
6911 | +{ |
6912 | + if (node_state != MASTER && master_state != MASTER_START) { | |
6913 | + P_MEMB("Got StartACK when not in MASTER_STARTING substate\n"); | |
6914 | + return 0; | |
6915 | + } | |
6916 | + | |
bb1d8b11 AM |
6917 | + /* buf is NULL if we are called directly from start_transition */ |
6918 | + if (buf) { | |
6919 | + struct cl_mem_startack_msg *ackmsg = | |
6920 | + (struct cl_mem_startack_msg *)buf; | |
c1c6733f AM |
6921 | + |
6922 | + /* Ignore any messages wil old generation numbers in them */ | |
6923 | + if (le32_to_cpu(ackmsg->generation) != cluster_generation) { | |
6924 | + P_MEMB("Got old generation START-ACK msg - ignoring\n"); | |
6925 | + return 0; | |
6926 | + } | |
6927 | + } | |
6928 | + | |
6929 | + /* If the node_id is non-zero then use it. */ | |
6930 | + if (transitionreason == TRANS_NEWNODE && joining_node && msg) { | |
bb1d8b11 AM |
6931 | + struct cl_mem_startack_msg *ackmsg = |
6932 | + (struct cl_mem_startack_msg *)buf; | |
c1c6733f AM |
6933 | + |
6934 | + if (ackmsg->node_id) { | |
6935 | + set_nodeid(joining_node, le32_to_cpu(ackmsg->node_id)); | |
6936 | + } | |
6937 | + highest_nodeid = | |
6938 | + max(highest_nodeid, le32_to_cpu(ackmsg->highest_node_id)); | |
6939 | + P_MEMB("Node id = %d, highest node id = %d\n", | |
6940 | + le32_to_cpu(ackmsg->node_id), | |
6941 | + le32_to_cpu(ackmsg->highest_node_id)); | |
6942 | + } | |
6943 | + | |
6944 | + /* If we have all the responses in then move to the next stage */ | |
6945 | + if (++responses_collected == responses_expected) { | |
6946 | + | |
6947 | + /* If the new node has no node_id (ie nobody in the cluster has | |
6948 | + * heard of it before) then assign it a new one */ | |
6949 | + if (transitionreason == TRANS_NEWNODE && joining_node) { | |
6950 | + highest_nodeid = | |
6951 | + max(highest_nodeid, get_highest_nodeid()); | |
6952 | + if (joining_node->node_id == 0) { | |
6953 | + set_nodeid(joining_node, ++highest_nodeid); | |
6954 | + } | |
6955 | + P_MEMB("nodeIDs: new node: %d, highest: %d\n", | |
6956 | + joining_node->node_id, highest_nodeid); | |
6957 | + } | |
6958 | + | |
6959 | + /* Behave a little differently if we are on our own */ | |
6960 | + if (cluster_members == 1) { | |
6961 | + if (transitionreason == TRANS_NEWNODE) { | |
6962 | + /* If the cluster is just us then confirm at | |
6963 | + * once */ | |
6964 | + joinconf_count = 0; | |
6965 | + mod_timer(&transition_timer, | |
6966 | + jiffies + | |
6967 | + cman_config.joinconf_timeout * HZ); | |
6968 | + send_joinconf(); | |
6969 | + return 0; | |
6970 | + } | |
6971 | + else { /* Node leaving the cluster */ | |
6972 | + recalculate_quorum(leavereason); | |
6973 | + leavereason = 0; | |
6974 | + node_state = MEMBER; | |
6975 | + } | |
6976 | + } | |
6977 | + else { | |
6978 | + master_state = MASTER_COLLECT; | |
6979 | + responses_collected = 0; | |
6980 | + responses_expected = cluster_members - 1; | |
6981 | + P_MEMB("Sending MASTERVIEW: expecting %d responses\n", | |
6982 | + responses_expected); | |
6983 | + | |
b7b72b66 | 6984 | + send_cluster_view(CLUSTER_MEM_MASTERVIEW, NULL, 0, MSG_REPLYEXP); |
c1c6733f AM |
6985 | + |
6986 | + /* Set a timer in case we don't get 'em all back */ | |
6987 | + mod_timer(&transition_timer, | |
6988 | + jiffies + | |
6989 | + cman_config.transition_timeout * HZ); | |
6990 | + } | |
6991 | + } | |
6992 | + return 0; | |
6993 | +} | |
6994 | + | |
6995 | +/* Got a VIEWACK response from a node */ | |
bb1d8b11 | 6996 | +static int do_process_viewack(struct msghdr *msg, char *reply, int len) |
c1c6733f | 6997 | +{ |
c1c6733f AM |
6998 | + struct sockaddr_cl *saddr = msg->msg_name; |
6999 | + | |
c1c6733f AM |
7000 | + if (node_opinion == NULL) { |
7001 | + node_opinion = | |
7002 | + kmalloc((1 + highest_nodeid) * sizeof (uint8_t), GFP_KERNEL); | |
7003 | + if (!node_opinion) { | |
7004 | + panic(": malloc agree/dissent failed\n"); | |
7005 | + } | |
7006 | + memset(node_opinion, 0, (1 + highest_nodeid) * sizeof (uint8_t)); | |
7007 | + } | |
7008 | + | |
7009 | + /* Keep a list of agreeing and dissenting nodes */ | |
7010 | + if (reply[1] == 1) { | |
7011 | + /* ACK - remote node agrees with me */ | |
7012 | + P_MEMB("Node agrees\n"); | |
7013 | + node_opinion[saddr->scl_nodeid] = OPINION_AGREE; | |
7014 | + agreeing_nodes++; | |
7015 | + } | |
7016 | + else { | |
7017 | + /* Remote node disagrees */ | |
7018 | + P_MEMB("Node disagrees\n"); | |
7019 | + node_opinion[saddr->scl_nodeid] = OPINION_DISAGREE; | |
7020 | + dissenting_nodes++; | |
7021 | + } | |
7022 | + | |
7023 | + P_MEMB("got %d responses, expected %d\n", responses_collected + 1, | |
7024 | + responses_expected); | |
7025 | + | |
7026 | + /* Are all the results in yet ? */ | |
7027 | + if (++responses_collected == responses_expected) { | |
7028 | + del_timer(&transition_timer); | |
7029 | + | |
7030 | + P_MEMB("The results are in: %d agree, %d dissent\n", | |
7031 | + agreeing_nodes, dissenting_nodes); | |
7032 | + | |
7033 | + if (agreeing_nodes > dissenting_nodes) { | |
7034 | + /* Kill dissenting nodes */ | |
7035 | + int i; | |
7036 | + | |
7037 | + for (i = 1; i <= responses_collected; i++) { | |
7038 | + if (node_opinion[i] == OPINION_DISAGREE) | |
7039 | + send_kill(i); | |
7040 | + } | |
7041 | + } | |
7042 | + else { | |
7043 | + /* We must leave the cluster as we are in a minority, | |
7044 | + * the rest of them can fight it out amongst | |
7045 | + * themselves. */ | |
c783755a | 7046 | + us->leave_reason = CLUSTER_LEAVEFLAG_INCONSISTENT; |
c1c6733f AM |
7047 | + agreeing_nodes = 0; |
7048 | + dissenting_nodes = 0; | |
7049 | + kfree(node_opinion); | |
7050 | + node_opinion = NULL; | |
7051 | + node_state = LEFT_CLUSTER; | |
7052 | + quit_threads = 1; | |
7053 | + wake_up_process(membership_task); | |
7054 | + wake_up_interruptible(&cnxman_waitq); | |
7055 | + return -1; | |
7056 | + } | |
7057 | + | |
7058 | + /* Reset counters */ | |
7059 | + agreeing_nodes = 0; | |
7060 | + dissenting_nodes = 0; | |
7061 | + kfree(node_opinion); | |
7062 | + node_opinion = NULL; | |
7063 | + | |
7064 | + /* Confirm new node */ | |
7065 | + if (transitionreason == TRANS_NEWNODE) { | |
7066 | + mod_timer(&transition_timer, | |
7067 | + jiffies + cman_config.joinconf_timeout * HZ); | |
7068 | + joinconf_count = 0; | |
7069 | + send_joinconf(); | |
7070 | + return 0; | |
7071 | + } | |
7072 | + | |
7073 | + master_state = MASTER_COMPLETE; | |
7074 | + | |
7075 | + end_transition(); | |
7076 | + } | |
7077 | + | |
7078 | + return 0; | |
7079 | +} | |
7080 | + | |
7081 | +/* Got an ENDTRANS message */ | |
bb1d8b11 | 7082 | +static int do_process_endtrans(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7083 | +{ |
7084 | + struct cl_mem_endtrans_msg *endmsg = | |
bb1d8b11 | 7085 | + (struct cl_mem_endtrans_msg *)buf; |
c1c6733f AM |
7086 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *) msg->msg_name; |
7087 | + | |
7088 | + /* Someone else's state transition */ | |
7089 | + if (node_state != TRANSITION && node_state != JOINACK) | |
7090 | + return 0; | |
7091 | + | |
7092 | + /* Check we got it from the MASTER node */ | |
7093 | + if (master_node && master_node->node_id != saddr->scl_nodeid) { | |
7094 | + printk(KERN_INFO | |
7095 | + "Got ENDTRANS from a node not the master: master: %d, sender: %d\n", | |
7096 | + master_node->node_id, saddr->scl_nodeid); | |
7097 | + return 0; | |
7098 | + } | |
7099 | + | |
7100 | + del_timer(&transition_timer); | |
7101 | + | |
7102 | + /* Set node ID on new node */ | |
7103 | + if (endmsg->new_node_id) { | |
7104 | + set_nodeid(joining_node, le32_to_cpu(endmsg->new_node_id)); | |
7105 | + P_MEMB("new node %s has ID %d\n", joining_node->name, | |
7106 | + joining_node->node_id); | |
7107 | + } | |
7108 | + | |
7109 | + node_state = TRANSITION_COMPLETE; | |
7110 | + | |
7111 | + /* Need to set this here or the barrier code will reject us if we've | |
7112 | + * just joined */ | |
7113 | + we_are_a_cluster_member = TRUE; | |
7114 | + | |
7115 | + confirm_joiner(); | |
7116 | + cluster_generation = le32_to_cpu(endmsg->generation); | |
7117 | + | |
7118 | + if (wait_for_completion_barrier() != 0) { | |
7119 | + P_MEMB("Barrier timed out - restart\n"); | |
7120 | + node_state = TRANSITION; | |
7121 | + mod_timer(&transition_timer, | |
7122 | + jiffies + cman_config.transition_timeout * HZ); | |
7123 | + return 0; | |
7124 | + } | |
7125 | + | |
7126 | + quorum = le32_to_cpu(endmsg->quorum); | |
7127 | + set_quorate(le32_to_cpu(endmsg->total_votes)); | |
c783755a | 7128 | + highest_nodeid = get_highest_nodeid(); |
c1c6733f AM |
7129 | + |
7130 | + /* Tell any waiting barriers that we had a transition */ | |
7131 | + check_barrier_returns(); | |
7132 | + | |
b7b72b66 AM |
7133 | + purge_temp_nodeids(); |
7134 | + | |
c1c6733f AM |
7135 | + /* Clear the master node */ |
7136 | + master_node = NULL; | |
7137 | + | |
7138 | + node_state = MEMBER; | |
7139 | + | |
7140 | + /* Notify other listeners that transition has completed */ | |
7141 | + notify_listeners(); | |
7142 | + reset_hello_time(); | |
7143 | + transition_end_time = jiffies; | |
7144 | + | |
7145 | + sm_member_update(cluster_is_quorate); | |
7146 | + return 0; | |
7147 | +} | |
7148 | + | |
7149 | +/* Turn a STARTTRANS message into NOMINATE and send it to the new master */ | |
7150 | +static int send_nominate(struct cl_mem_starttrans_msg *startmsg, int msglen, | |
7151 | + int nodeid) | |
7152 | +{ | |
7153 | + struct sockaddr_cl maddr; | |
7154 | + | |
7155 | + maddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
7156 | + maddr.scl_family = AF_CLUSTER; | |
7157 | + maddr.scl_nodeid = nodeid; | |
7158 | + | |
7159 | + startmsg->cmd = CLUSTER_MEM_NOMINATE; | |
7160 | + return kcl_sendmsg(mem_socket, startmsg, msglen, | |
7161 | + &maddr, sizeof (maddr), 0); | |
7162 | +} | |
7163 | + | |
7164 | +/* Got a STARTTRANS message */ | |
bb1d8b11 | 7165 | +static int do_process_starttrans(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7166 | +{ |
7167 | + struct cl_mem_starttrans_msg *startmsg = | |
bb1d8b11 | 7168 | + (struct cl_mem_starttrans_msg *)buf; |
c1c6733f AM |
7169 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *) msg->msg_name; |
7170 | + struct cluster_node *node; | |
7171 | + unsigned int newgen = le32_to_cpu(startmsg->generation); | |
7172 | + | |
7173 | + /* Got a WHAT from WHOM? */ | |
7174 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7175 | + if (!node || node->state != NODESTATE_MEMBER) | |
7176 | + return 0; | |
7177 | + | |
7178 | + /* Someone else's state transition */ | |
7179 | + if (node_state != MEMBER && | |
7180 | + node_state != TRANSITION && node_state != MASTER) | |
7181 | + return 0; | |
7182 | + | |
7183 | + /* Ignore old generation STARTTRANS messages */ | |
7184 | + if ((newgen < cluster_generation) || | |
7185 | + (newgen == 0xFFFFFFFF && cluster_generation == 0)) { | |
7186 | + P_MEMB("Ignoring STARTTRANS with old generation number\n"); | |
7187 | + return 0; | |
7188 | + } | |
7189 | + | |
7190 | + P_MEMB("Got starttrans: newgen = %d, oldgen = %d, reason = %d\n", | |
7191 | + newgen, cluster_generation, startmsg->reason); | |
7192 | + | |
7193 | + /* Up the generation number */ | |
7194 | + cluster_generation = newgen; | |
7195 | + | |
7196 | + /* If we are also a master then decide between us */ | |
7197 | + if (node_state == MASTER) { | |
7198 | + | |
7199 | + /* See if we really want the responsibility of being master */ | |
7200 | + if (elect_master(&node)) { | |
7201 | + | |
7202 | + /* I reluctantly accept this position of responsibility | |
7203 | + */ | |
7204 | + P_MEMB("I elected myself master\n"); | |
7205 | + | |
7206 | + /* start_transition will re-establish this */ | |
7207 | + del_timer(&transition_timer); | |
7208 | + | |
7209 | + start_transition(TRANS_NEWMASTER, node); | |
7210 | + return 0; | |
7211 | + } | |
7212 | + else { | |
7213 | + /* Back down */ | |
7214 | + P_MEMB("Backing down from MASTER status\n"); | |
7215 | + master_node = node; | |
7216 | + node_state = MEMBER; | |
7217 | + | |
7218 | + /* If we were bringing a new node into the cluster then | |
7219 | + * we will have to abandon that now and tell the new | |
7220 | + * node to try again later */ | |
7221 | + if (transitionreason == TRANS_NEWNODE && joining_node) { | |
7222 | + struct cluster_node_addr *first_addr = | |
7223 | + (struct cluster_node_addr *) joining_node-> | |
7224 | + addr_list.next; | |
7225 | + | |
7226 | + P_MEMB("Postponing membership of node %s\n", | |
7227 | + joining_node->name); | |
7228 | + send_joinack(first_addr->addr, address_length, | |
7229 | + JOINACK_TYPE_WAIT); | |
7230 | + | |
7231 | + /* Not dead, just sleeping */ | |
7232 | + joining_node->state = NODESTATE_DEAD; | |
7233 | + joining_node = NULL; | |
7234 | + } | |
7235 | + | |
7236 | + /* If the new master is not us OR the node we just got | |
7237 | + * the STARTTRANS from then make sure it knows it has | |
7238 | + * to be master */ | |
7239 | + if (saddr->scl_nodeid != node->node_id) { | |
7240 | + send_nominate(startmsg, len, node->node_id); | |
7241 | + return 0; | |
7242 | + } | |
7243 | + | |
7244 | + /* Fall through into MEMBER code below if we are | |
7245 | + * obeying the STARTTRANS we just received */ | |
7246 | + } | |
7247 | + } | |
7248 | + | |
7249 | + /* Do non-MASTER STARTTRANS bits */ | |
7250 | + if (node_state == MEMBER) { | |
7251 | + int ptr = sizeof (struct cl_mem_starttrans_msg); | |
7252 | + int node_id = 0; | |
7253 | + | |
7254 | + P_MEMB("Normal transition start\n"); | |
7255 | + | |
7256 | + /* If the master is adding a new node and we know it's node ID | |
7257 | + * then ACK with it. */ | |
7258 | + if (startmsg->reason == TRANS_NEWNODE) { | |
7259 | + struct cluster_node *node = | |
7260 | + find_node_by_addr((char *) startmsg + ptr, | |
7261 | + address_length); | |
7262 | + if (node) | |
7263 | + node_id = node->node_id; | |
7264 | + } | |
7265 | + | |
7266 | + /* Save the master info */ | |
7267 | + master_node = find_node_by_nodeid(saddr->scl_nodeid); | |
7268 | + node_state = TRANSITION; | |
7269 | + | |
7270 | + if (startmsg->reason == TRANS_NEWNODE) { | |
bb1d8b11 | 7271 | + add_node_from_starttrans(msg, buf, len); |
c1c6733f AM |
7272 | + } |
7273 | + | |
7274 | + if (startmsg->reason == TRANS_REMNODE || | |
7275 | + startmsg->reason == TRANS_ANOTHERREMNODE) { | |
7276 | + remove_node(le32_to_cpu(startmsg->nodeid)); | |
7277 | + } | |
7278 | + | |
7279 | + send_startack(saddr, msg->msg_namelen, | |
7280 | + node_id); | |
7281 | + | |
7282 | + /* Establish timer in case the master dies */ | |
7283 | + mod_timer(&transition_timer, | |
7284 | + jiffies + cman_config.transition_timeout * HZ); | |
7285 | + | |
7286 | + return 0; | |
7287 | + } | |
7288 | + | |
7289 | + /* We are in transition but this may be a restart */ | |
7290 | + if (node_state == TRANSITION) { | |
7291 | + | |
7292 | + master_node = find_node_by_nodeid(saddr->scl_nodeid); | |
7293 | + send_startack(saddr, msg->msg_namelen, 0); | |
7294 | + | |
7295 | + /* Is it a new joining node ? This happens if a master is | |
7296 | + * usurped */ | |
7297 | + if (startmsg->reason == TRANS_NEWNODE) { | |
7298 | + struct cluster_node *oldjoin = joining_node; | |
7299 | + | |
bb1d8b11 | 7300 | + add_node_from_starttrans(msg, buf, len); |
c1c6733f AM |
7301 | + |
7302 | + /* If this is a different node joining than the one we | |
7303 | + * were previously joining (probably cos the master is | |
7304 | + * a nominated one) then mark our "old" joiner as DEAD. | |
7305 | + * The original master will already have told the node | |
7306 | + * to go back into JOINWAIT state */ | |
7307 | + if (oldjoin && oldjoin != joining_node | |
7308 | + && oldjoin->state == NODESTATE_JOINING) | |
7309 | + oldjoin->state = NODESTATE_DEAD; | |
7310 | + } | |
7311 | + | |
7312 | + /* Is it a new master node? */ | |
7313 | + if (startmsg->reason == TRANS_NEWMASTER || | |
7314 | + startmsg->reason == TRANS_DEADMASTER) { | |
7315 | + P_MEMB("starttrans %s, node=%d\n", | |
7316 | + startmsg->reason == | |
7317 | + TRANS_NEWMASTER ? "NEWMASTER" : "DEADMASTER", | |
7318 | + le32_to_cpu(startmsg->nodeid)); | |
7319 | + | |
7320 | + /* If the old master has died then remove it */ | |
c783755a AM |
7321 | + if (startmsg->reason == TRANS_DEADMASTER) { |
7322 | + remove_node(le32_to_cpu(startmsg->nodeid)); | |
c1c6733f AM |
7323 | + } |
7324 | + | |
7325 | + /* Store new master */ | |
7326 | + master_node = find_node_by_nodeid(saddr->scl_nodeid); | |
7327 | + } | |
7328 | + | |
7329 | + /* Another node has died (or been killed) */ | |
7330 | + if (startmsg->reason == TRANS_ANOTHERREMNODE) { | |
7331 | + /* Remove new dead node */ | |
c783755a | 7332 | + remove_node(le32_to_cpu(startmsg->nodeid)); |
c1c6733f AM |
7333 | + } |
7334 | + /* Restart the timer */ | |
7335 | + del_timer(&transition_timer); | |
7336 | + mod_timer(&transition_timer, | |
7337 | + jiffies + cman_config.transition_timeout * HZ); | |
7338 | + } | |
7339 | + | |
7340 | + return 0; | |
7341 | +} | |
7342 | + | |
7343 | +/* Change a cluster parameter */ | |
bb1d8b11 | 7344 | +static int do_process_reconfig(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7345 | +{ |
7346 | + struct cl_mem_reconfig_msg *confmsg; | |
7347 | + struct sockaddr_cl *saddr = msg->msg_name; | |
7348 | + struct cluster_node *node; | |
7349 | + unsigned int val; | |
7350 | + | |
7351 | + if (len < sizeof(struct cl_mem_reconfig_msg)) | |
7352 | + return -1; | |
7353 | + | |
bb1d8b11 | 7354 | + confmsg = (struct cl_mem_reconfig_msg *)buf; |
c1c6733f AM |
7355 | + val = le32_to_cpu(confmsg->value); |
7356 | + | |
7357 | + switch (confmsg->param) { | |
7358 | + | |
7359 | + case RECONFIG_PARAM_EXPECTED_VOTES: | |
7360 | + /* Set any nodes with expected_votes higher than the new value | |
7361 | + * down */ | |
7362 | + if (val > 0) { | |
7363 | + struct cluster_node *node; | |
7364 | + | |
7365 | + down(&cluster_members_lock); | |
7366 | + list_for_each_entry(node, &cluster_members_list, list) { | |
7367 | + if (node->state == NODESTATE_MEMBER && | |
7368 | + node->expected_votes > val) { | |
7369 | + node->expected_votes = val; | |
7370 | + } | |
7371 | + } | |
7372 | + up(&cluster_members_lock); | |
7373 | + if (expected_votes > val) | |
7374 | + expected_votes = val; | |
7375 | + } | |
7376 | + recalculate_quorum(1); /* Allow decrease */ | |
7377 | + sm_member_update(cluster_is_quorate); | |
7378 | + break; | |
7379 | + | |
7380 | + case RECONFIG_PARAM_NODE_VOTES: | |
7381 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7382 | + node->votes = val; | |
7383 | + recalculate_quorum(1); /* Allow decrease */ | |
7384 | + sm_member_update(cluster_is_quorate); | |
7385 | + break; | |
7386 | + | |
7387 | + case RECONFIG_PARAM_CONFIG_VERSION: | |
7388 | + config_version = val; | |
7389 | + break; | |
7390 | + | |
7391 | + default: | |
7392 | + printk(KERN_INFO CMAN_NAME | |
7393 | + ": got unknown parameter in reconfigure message. %d\n", | |
7394 | + confmsg->param); | |
7395 | + break; | |
7396 | + } | |
7397 | + return 0; | |
7398 | +} | |
7399 | + | |
7400 | +/* Response from master node */ | |
bb1d8b11 | 7401 | +static int do_process_joinack(struct msghdr *msg, char *buf, int len) |
c1c6733f | 7402 | +{ |
bb1d8b11 AM |
7403 | + struct cl_mem_joinack_msg *ackmsg = |
7404 | + (struct cl_mem_joinack_msg *)buf; | |
c1c6733f AM |
7405 | + |
7406 | + join_time = jiffies; | |
7407 | + if (ackmsg->acktype == JOINACK_TYPE_OK) { | |
7408 | + node_state = JOINACK; | |
7409 | + } | |
7410 | + | |
7411 | + if (ackmsg->acktype == JOINACK_TYPE_NAK) { | |
7412 | + printk(KERN_WARNING CMAN_NAME | |
7413 | + ": Cluster membership rejected\n"); | |
7414 | + P_MEMB("Got JOINACK NACK\n"); | |
7415 | + node_state = REJECTED; | |
7416 | + } | |
7417 | + | |
7418 | + if (ackmsg->acktype == JOINACK_TYPE_WAIT) { | |
7419 | + P_MEMB("Got JOINACK WAIT\n"); | |
7420 | + node_state = JOINWAIT; | |
7421 | + joinwait_time = jiffies; | |
7422 | + } | |
7423 | + | |
c783755a AM |
7424 | + return 0; |
7425 | +} | |
7426 | + | |
7427 | +/* Check a JOINREQ message for validity, | |
7428 | + return -1 if we can't let the node join our cluster */ | |
7429 | +static int validate_joinmsg(struct cl_mem_join_msg *joinmsg, int len) | |
7430 | +{ | |
7431 | + struct cluster_node *node; | |
7432 | + | |
7433 | + /* Check version number */ | |
c1c6733f AM |
7434 | + if (le32_to_cpu(joinmsg->major_version) == CNXMAN_MAJOR_VERSION) { |
7435 | + char *ptr = (char *) joinmsg; | |
7436 | + char *name; | |
7437 | + | |
b7b72b66 AM |
7438 | + ptr += sizeof (*joinmsg); |
7439 | + name = ptr + le16_to_cpu(joinmsg->num_addr) * address_length; | |
7440 | + | |
c1c6733f AM |
7441 | + /* Sanity-check the num_addrs field otherwise we could oops */ |
7442 | + if (le16_to_cpu(joinmsg->num_addr) * address_length > len) { | |
7443 | + printk(KERN_WARNING CMAN_NAME | |
7444 | + ": num_addr in JOIN-REQ message is rubbish: %d\n", | |
7445 | + le16_to_cpu(joinmsg->num_addr)); | |
c783755a | 7446 | + return -1; |
c1c6733f AM |
7447 | + } |
7448 | + | |
7449 | + /* Check the cluster name matches */ | |
7450 | + if (strcmp(cluster_name, joinmsg->clustername)) { | |
7451 | + printk(KERN_WARNING CMAN_NAME | |
7452 | + ": attempt to join with cluster name '%s' refused\n", | |
7453 | + joinmsg->clustername); | |
c783755a | 7454 | + return -1; |
c1c6733f AM |
7455 | + } |
7456 | + | |
c1c6733f | 7457 | + /* Check we are not exceeding the maximum number of nodes */ |
b7b72b66 | 7458 | + if (cluster_members >= cman_config.max_nodes) { |
c1c6733f AM |
7459 | + printk(KERN_WARNING CMAN_NAME |
7460 | + ": Join request from %s rejected, exceeds maximum number of nodes\n", | |
7461 | + name); | |
c783755a | 7462 | + return -1; |
c1c6733f AM |
7463 | + } |
7464 | + | |
b7b72b66 | 7465 | + /* Check that we don't exceed the two_node limit, if applicable */ |
c1c6733f AM |
7466 | + if (two_node && cluster_members == 2) { |
7467 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7468 | + "rejected, exceeds two node limit\n", name); | |
c783755a | 7469 | + return -1; |
c1c6733f AM |
7470 | + } |
7471 | + | |
b7b72b66 | 7472 | + if (le32_to_cpu(joinmsg->config_version) != config_version) { |
c1c6733f AM |
7473 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " |
7474 | + "rejected, config version local %u remote %u\n", | |
7475 | + name, config_version, | |
b7b72b66 | 7476 | + le32_to_cpu(joinmsg->config_version)); |
c783755a AM |
7477 | + return -1; |
7478 | + } | |
7479 | + | |
7480 | + /* Validate requested static node ID */ | |
7481 | + if (joinmsg->nodeid && | |
7482 | + (node = find_node_by_nodeid(le32_to_cpu(joinmsg->nodeid))) && | |
7483 | + (node->state != NODESTATE_DEAD || | |
7484 | + (strcmp(node->name, name)))) { | |
7485 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7486 | + "rejected, node ID %d already in use by %s\n", | |
7487 | + name, node->node_id, node->name); | |
7488 | + return -1; | |
7489 | + } | |
7490 | + if (joinmsg->nodeid && | |
7491 | + (node = find_node_by_name(name)) && | |
7492 | + (node->state != NODESTATE_DEAD || | |
7493 | + node->node_id != le32_to_cpu(joinmsg->nodeid))) { | |
7494 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7495 | + "rejected, wanted node %d but previously had %d\n", | |
7496 | + name, le32_to_cpu(joinmsg->nodeid), node->node_id); | |
7497 | + return -1; | |
c1c6733f AM |
7498 | + } |
7499 | + | |
c783755a | 7500 | + /* If these don't match then I don't know how the message |
c1c6733f AM |
7501 | + arrived! However, I can't take the chance */ |
7502 | + if (le32_to_cpu(joinmsg->addr_len) != address_length) { | |
7503 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7504 | + "rejected, address length local: %u remote %u\n", | |
7505 | + name, address_length, | |
7506 | + le32_to_cpu(joinmsg->addr_len)); | |
c783755a | 7507 | + return -1; |
c1c6733f AM |
7508 | + } |
7509 | + } | |
7510 | + else { | |
7511 | + /* Version number mismatch, don't use any part of the message | |
7512 | + * other than the version numbers as things may have moved */ | |
c1c6733f | 7513 | + printk(KERN_INFO CMAN_NAME |
c783755a | 7514 | + ": Got join message from node running incompatible software. (us: %d.%d.%d, them: %d.%d.%d)\n", |
c1c6733f AM |
7515 | + CNXMAN_MAJOR_VERSION, CNXMAN_MINOR_VERSION, |
7516 | + CNXMAN_PATCH_VERSION, | |
7517 | + le32_to_cpu(joinmsg->major_version), | |
7518 | + le32_to_cpu(joinmsg->minor_version), | |
c783755a AM |
7519 | + le32_to_cpu(joinmsg->patch_version)); |
7520 | + return -1; | |
7521 | + } | |
7522 | + return 0; | |
7523 | +} | |
7524 | + | |
7525 | + | |
7526 | +/* Request to join the cluster. This makes us the master for this state | |
7527 | + * transition */ | |
bb1d8b11 | 7528 | +static int do_process_joinreq(struct msghdr *msg, char *buf, int len) |
c783755a AM |
7529 | +{ |
7530 | + static unsigned long last_joinreq = 0; | |
7531 | + static char last_name[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
bb1d8b11 | 7532 | + struct cl_mem_join_msg *joinmsg = (struct cl_mem_join_msg *)buf; |
c783755a AM |
7533 | + struct cluster_node *node; |
7534 | + char *ptr = (char *) joinmsg; | |
7535 | + char *name; | |
7536 | + int i; | |
7537 | + struct sockaddr_cl *addr = msg->msg_name; | |
7538 | + | |
7539 | + ptr += sizeof (*joinmsg); | |
7540 | + name = ptr + le16_to_cpu(joinmsg->num_addr) * address_length; | |
c1c6733f | 7541 | + |
c783755a AM |
7542 | + /* If we are in a state transition then tell the new node to wait a bit |
7543 | + * longer */ | |
7544 | + if (node_state != MEMBER) { | |
7545 | + if (node_state == MASTER || node_state == TRANSITION) { | |
7546 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7547 | + JOINACK_TYPE_WAIT); | |
7548 | + } | |
7549 | + return 0; | |
7550 | + } | |
7551 | + | |
7552 | + /* Reject application if message is invalid for any reason */ | |
7553 | + if (validate_joinmsg(joinmsg, len)) { | |
7554 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7555 | + JOINACK_TYPE_NAK); | |
7556 | + return 0; | |
7557 | + } | |
7558 | + | |
7559 | + /* Do we already know about this node? */ | |
7560 | + if (check_duplicate_node(name, msg, len) < 0) { | |
c1c6733f | 7561 | + send_joinack(msg->msg_name, msg->msg_namelen, |
c783755a AM |
7562 | + JOINACK_TYPE_NAK); |
7563 | + return 0; | |
7564 | + } | |
7565 | + | |
7566 | + /* Duplicate checking: Because joining messages do not have | |
7567 | + * sequence numbers we may get as many JOINREQ messages as we | |
7568 | + * have interfaces. This bit of code here just checks for | |
7569 | + * JOINREQ messages that come in from the same node in a small | |
7570 | + * period of time and removes the duplicates */ | |
7571 | + if (time_before(jiffies, last_joinreq + 10 * HZ) | |
7572 | + && strcmp(name, last_name) == 0) { | |
c1c6733f AM |
7573 | + return 0; |
7574 | + } | |
7575 | + | |
c783755a AM |
7576 | + /* OK, you can be in my gang */ |
7577 | + last_joinreq = jiffies; | |
7578 | + strcpy(last_name, name); | |
7579 | + | |
7580 | + node = add_new_node(name, joinmsg->votes, | |
7581 | + le32_to_cpu(joinmsg->expected_votes), | |
7582 | + le32_to_cpu(joinmsg->nodeid), | |
7583 | + NODESTATE_JOINING); | |
7584 | + | |
7585 | + /* Add the node's addresses */ | |
7586 | + if (list_empty(&node->addr_list)) { | |
7587 | + for (i = 0; i < le16_to_cpu(joinmsg->num_addr); | |
7588 | + i++) { | |
7589 | + add_node_address(node, ptr, address_length); | |
7590 | + ptr += address_length; | |
7591 | + } | |
7592 | + } | |
7593 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7594 | + JOINACK_TYPE_OK); | |
7595 | + joining_node = node; | |
7596 | + joining_temp_nodeid = addr->scl_nodeid; | |
7597 | + | |
7598 | + /* Start the state transition */ | |
7599 | + start_transition(TRANS_NEWNODE, node); | |
7600 | + | |
c1c6733f AM |
7601 | + return 0; |
7602 | +} | |
7603 | + | |
7604 | +/* A simple function to invent a small number based | |
7605 | + on the node name */ | |
7606 | +static int node_hash(void) | |
7607 | +{ | |
7608 | + int i; | |
7609 | + int value = 0; | |
7610 | + | |
7611 | + for (i=0; i<strlen(nodename); i++) { | |
7612 | + value += nodename[i]; | |
7613 | + } | |
c783755a AM |
7614 | + return (value & 0xF) + 1; |
7615 | +} | |
7616 | + | |
7617 | + | |
7618 | +/* Return the low 32 bits of our IP address */ | |
7619 | +static uint32_t low32_of_ip() | |
7620 | +{ | |
7621 | + struct cluster_node_addr *addr; | |
7622 | + uint32_t lowip; | |
7623 | + | |
7624 | + addr = list_entry(us->addr_list.next, struct cluster_node_addr, list); | |
7625 | + memcpy(&lowip, addr->addr+address_length-sizeof(uint32_t), sizeof(uint32_t)); | |
7626 | + if (!lowip) | |
7627 | + memcpy(&lowip, addr->addr - sizeof(uint32_t)*2, sizeof(uint32_t)); | |
7628 | + | |
7629 | + return lowip; | |
c1c6733f AM |
7630 | +} |
7631 | + | |
7632 | +/* A new node has stated its intent to form a new cluster. we may have | |
7633 | + * something to say about that... */ | |
bb1d8b11 | 7634 | +static int do_process_newcluster(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7635 | +{ |
7636 | + /* If we are also in STARTING state then back down for a random period | |
7637 | + * of time */ | |
7638 | + if (node_state == STARTING) { | |
7639 | + P_MEMB("got NEWCLUSTER, backing down for %d seconds\n", node_hash()); | |
7640 | + start_time = jiffies + node_hash() * HZ; | |
7641 | + } | |
c783755a AM |
7642 | + |
7643 | + if (node_state == NEWCLUSTER) { | |
7644 | + uint32_t otherip; | |
c783755a | 7645 | + |
bb1d8b11 | 7646 | + memcpy(&otherip, buf+1, sizeof(otherip)); |
c783755a AM |
7647 | + otherip = le32_to_cpu(otherip); |
7648 | + P_MEMB("got NEWCLUSTER, remote ip = %x, us = %x\n", otherip, low32_of_ip()); | |
7649 | + if (otherip < low32_of_ip()) | |
7650 | + node_state = STARTING; | |
7651 | + } | |
7652 | + | |
b7b72b66 AM |
7653 | + if (node_state == MEMBER) |
7654 | + send_hello(); | |
c1c6733f AM |
7655 | + |
7656 | + return 0; | |
7657 | +} | |
7658 | + | |
7659 | +/* Called for each node by the node-message unpacker. Returns -1 if there is a | |
7660 | + * mismatch and the caller will stop processing */ | |
7661 | +static int check_node(struct cluster_node *newnode, char *addrs, | |
7662 | + unsigned short num_addr) | |
7663 | +{ | |
7664 | + struct cluster_node *node = find_node_by_name(newnode->name); | |
7665 | + | |
7666 | + P_MEMB("check_node: %s", newnode->name); | |
7667 | + | |
7668 | + if (!node) { | |
7669 | + C_MEMB(" - not found\n"); | |
7670 | + return -1; | |
7671 | + } | |
7672 | + | |
7673 | + if (node->votes != newnode->votes || | |
7674 | + node->node_id != newnode->node_id || | |
c783755a AM |
7675 | + node->state != newnode->state) { |
7676 | + C_MEMB(" - wrong info: votes=%d(exp: %d) id=%d(exp: %d) state = %d\n", | |
7677 | + node->votes, newnode->votes, node->node_id, | |
7678 | + newnode->node_id, node->state); | |
c1c6733f AM |
7679 | + return -1; |
7680 | + } | |
7681 | + C_MEMB(" - OK\n"); | |
7682 | + return 0; | |
7683 | +} | |
7684 | + | |
7685 | +/* Called for each new node found in a JOINCONF message. Create a new node | |
7686 | + * entry */ | |
7687 | +static int add_node(struct cluster_node *node, char *addrs, | |
7688 | + unsigned short num_addr) | |
7689 | +{ | |
7690 | + P_MEMB("add_node: %s, v:%d, e:%d, i:%d\n", node->name, node->votes, | |
7691 | + node->expected_votes, node->node_id); | |
7692 | + | |
7693 | + if (!find_node_by_name(node->name)) { | |
7694 | + struct cluster_node *newnode; | |
7695 | + int i; | |
7696 | + | |
7697 | + if ((newnode = | |
7698 | + add_new_node(node->name, node->votes, node->expected_votes, | |
c783755a | 7699 | + node->node_id, node->state)) == NULL) { |
c1c6733f AM |
7700 | + P_MEMB("Error adding node\n"); |
7701 | + return -1; | |
7702 | + } | |
7703 | + if (list_empty(&newnode->addr_list)) { | |
7704 | + for (i = 0; i < num_addr; i++) { | |
7705 | + add_node_address(newnode, | |
7706 | + addrs + i * address_length, address_length); | |
7707 | + } | |
7708 | + } | |
7709 | + return 0; | |
7710 | + } | |
7711 | + else { | |
7712 | + P_MEMB("Already got node with name %s\n", node->name); | |
7713 | + return -1; | |
7714 | + } | |
7715 | +} | |
7716 | + | |
7717 | +/* Call a specified routine for each node unpacked from the message. Return | |
7718 | + * either the number of nodes found or -1 for an error */ | |
7719 | +static int unpack_nodes(unsigned char *buf, int len, | |
7720 | + int (*routine) (struct cluster_node *, char *, | |
7721 | + unsigned short)) | |
7722 | +{ | |
7723 | + int ptr = 0; | |
7724 | + int num_nodes = 0; | |
7725 | + char nodename[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
7726 | + struct cluster_node node; | |
7727 | + | |
7728 | + node.name = nodename; | |
7729 | + | |
7730 | + while (ptr < len) { | |
7731 | + int namelen = buf[ptr++]; | |
7732 | + unsigned int evotes; | |
7733 | + unsigned int node_id; | |
7734 | + unsigned short num_addr; | |
7735 | + unsigned char *addrs; | |
7736 | + | |
7737 | + memcpy(nodename, &buf[ptr], namelen); | |
7738 | + nodename[namelen] = '\0'; | |
7739 | + ptr += namelen; | |
7740 | + | |
c783755a AM |
7741 | + node.state = buf[ptr++]; |
7742 | + | |
c1c6733f AM |
7743 | + memcpy(&num_addr, &buf[ptr], sizeof (short)); |
7744 | + num_addr = le16_to_cpu(num_addr); | |
7745 | + ptr += sizeof (short); | |
7746 | + | |
7747 | + /* Just make a note of the addrs "array" */ | |
7748 | + addrs = &buf[ptr]; | |
7749 | + ptr += num_addr * address_length; | |
7750 | + | |
7751 | + node.votes = buf[ptr++]; | |
7752 | + | |
7753 | + memcpy(&evotes, &buf[ptr], sizeof (int)); | |
7754 | + node.expected_votes = le32_to_cpu(evotes); | |
7755 | + ptr += sizeof (int); | |
7756 | + | |
7757 | + memcpy(&node_id, &buf[ptr], sizeof (int)); | |
7758 | + node.node_id = le32_to_cpu(node_id); | |
7759 | + ptr += sizeof (int); | |
7760 | + | |
7761 | + /* Call the callback routine */ | |
7762 | + if (routine(&node, addrs, num_addr) < 0) | |
7763 | + return -1; | |
c783755a AM |
7764 | + |
7765 | + /* Return the number of MEMBER nodes */ | |
7766 | + if (node.state == NODESTATE_MEMBER) | |
7767 | + num_nodes++; | |
c1c6733f AM |
7768 | + } |
7769 | + return num_nodes; | |
7770 | +} | |
7771 | + | |
7772 | +/* Got join confirmation from a master node. This message contains a list of | |
7773 | + * cluster nodes which we unpack and build into our cluster nodes list. When we | |
7774 | + * have the last message we can go into TRANSITION state */ | |
bb1d8b11 | 7775 | +static int do_process_joinconf(struct msghdr *msg, char *buf, int len) |
c1c6733f | 7776 | +{ |
bb1d8b11 | 7777 | + if (unpack_nodes(buf + 2, len - 2, add_node) < 0) { |
c1c6733f AM |
7778 | + printk(CMAN_NAME |
7779 | + ": Error procssing joinconf message - giving up on cluster join\n"); | |
c783755a AM |
7780 | + us->leave_reason = CLUSTER_LEAVEFLAG_PANIC; |
7781 | + node_state = LEFT_CLUSTER; | |
c1c6733f AM |
7782 | + return -1; |
7783 | + } | |
7784 | + | |
7785 | + /* Last message in the list? */ | |
bb1d8b11 | 7786 | + if (buf[1] & 2) { |
c1c6733f AM |
7787 | + char ackmsg; |
7788 | + struct sockaddr_cl *addr = msg->msg_name; | |
7789 | + | |
7790 | + us->state = NODESTATE_MEMBER; | |
7791 | + node_state = TRANSITION; | |
7792 | + we_are_a_cluster_member = TRUE; | |
7793 | + | |
7794 | + ackmsg = CLUSTER_MEM_CONFACK; | |
7795 | + kcl_sendmsg(mem_socket, &ackmsg, 1, addr, | |
7796 | + sizeof (struct sockaddr_cl), | |
7797 | + MSG_NOACK); | |
7798 | + kernel_thread(hello_kthread, NULL, 0); | |
7799 | + mod_timer(&hello_timer, jiffies + cman_config.hello_timer * HZ); | |
7800 | + } | |
7801 | + return 0; | |
7802 | +} | |
7803 | + | |
7804 | +/* Got the master's view of the cluster - compare it with ours and tell it the | |
7805 | + * result */ | |
bb1d8b11 | 7806 | +static int do_process_masterview(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7807 | +{ |
7808 | + char reply[2] = { CLUSTER_MEM_VIEWACK, 0 }; | |
c1c6733f AM |
7809 | + static int num_nodes; |
7810 | + | |
7811 | + /* Someone else's state transition */ | |
7812 | + if (node_state != MEMBER && | |
7813 | + node_state != TRANSITION && node_state != MASTER) | |
7814 | + return 0; | |
7815 | + | |
7816 | + /* First message, zero the counter */ | |
bb1d8b11 | 7817 | + if (buf[1] & 1) |
c1c6733f AM |
7818 | + num_nodes = 0; |
7819 | + | |
bb1d8b11 | 7820 | + num_nodes += unpack_nodes(buf + 2, len - 2, check_node); |
c1c6733f AM |
7821 | + |
7822 | + /* Last message, check the count and reply */ | |
bb1d8b11 | 7823 | + if (buf[1] & 2) { |
c1c6733f AM |
7824 | + if (num_nodes == cluster_members) { |
7825 | + /* Send ACK */ | |
7826 | + reply[1] = 1; | |
7827 | + } | |
7828 | + else { | |
7829 | + P_MEMB | |
7830 | + ("Got %d nodes in MASTERVIEW message, we think there s/b %d\n", | |
7831 | + num_nodes, cluster_members); | |
7832 | + /* Send NAK */ | |
7833 | + reply[1] = 0; | |
7834 | + } | |
7835 | + kcl_sendmsg(mem_socket, reply, 2, msg->msg_name, | |
7836 | + msg->msg_namelen, 0); | |
7837 | + } | |
7838 | + return 0; | |
7839 | +} | |
7840 | + | |
bb1d8b11 | 7841 | +static int do_process_leave(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7842 | +{ |
7843 | + struct cluster_node *node; | |
7844 | + struct sockaddr_cl *saddr = msg->msg_name; | |
bb1d8b11 | 7845 | + unsigned char *leavemsg = (unsigned char *)buf; |
c1c6733f AM |
7846 | + |
7847 | + if ((node = find_node_by_nodeid(saddr->scl_nodeid))) { | |
7848 | + unsigned char reason = leavemsg[1]; | |
7849 | + | |
7850 | + if (node->state != NODESTATE_DEAD) { | |
7851 | + printk(KERN_INFO CMAN_NAME | |
c783755a AM |
7852 | + ": Node %s is leaving the cluster, %s\n", |
7853 | + node->name, leave_string(reason)); | |
c1c6733f AM |
7854 | + |
7855 | + node->leave_reason = reason; | |
7856 | + } | |
7857 | + leavereason = (reason == CLUSTER_LEAVEFLAG_REMOVED ? 1 : 0); | |
7858 | + | |
7859 | + a_node_just_died(node); | |
c1c6733f AM |
7860 | + } |
7861 | + return 0; | |
7862 | +} | |
7863 | + | |
bb1d8b11 | 7864 | +static int do_process_hello(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7865 | +{ |
7866 | + struct cluster_node *node; | |
7867 | + struct cl_mem_hello_msg *hellomsg = | |
bb1d8b11 | 7868 | + (struct cl_mem_hello_msg *)buf; |
c1c6733f AM |
7869 | + struct sockaddr_cl *saddr = msg->msg_name; |
7870 | + | |
7871 | + /* We are starting up. Send a join message to the node whose HELLO we | |
7872 | + * just received */ | |
c783755a AM |
7873 | + if (node_state == STARTING || node_state == JOINWAIT || |
7874 | + node_state == JOINING || node_state == NEWCLUSTER) { | |
c1c6733f AM |
7875 | + struct sockaddr_cl *addr = msg->msg_name; |
7876 | + | |
7877 | + printk(KERN_INFO CMAN_NAME ": sending membership request\n"); | |
7878 | + | |
7879 | + send_joinreq(addr, msg->msg_namelen); | |
7880 | + join_time = jiffies; | |
7881 | + node_state = JOINING; | |
7882 | + return 0; | |
7883 | + } | |
7884 | + | |
7885 | + /* Only process HELLOs if we are not in transition */ | |
7886 | + if (node_state == MEMBER) { | |
c1c6733f AM |
7887 | + |
7888 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7889 | + if (node && node->state != NODESTATE_DEAD) { | |
7890 | + | |
7891 | + /* Check the cluster generation in the HELLO message. | |
7892 | + * NOTE: this may be different if the message crossed | |
7893 | + * on the wire with an END-TRANS so we allow a period | |
7894 | + * of grace in which this is allowable */ | |
7895 | + if (cluster_generation != | |
7896 | + le32_to_cpu(hellomsg->generation) | |
7897 | + && node_state == MEMBER | |
7898 | + && time_after(jiffies, | |
7899 | + cman_config.hello_timer * HZ + | |
7900 | + transition_end_time)) { | |
c1c6733f AM |
7901 | + |
7902 | + printk(KERN_INFO CMAN_NAME | |
7903 | + ": bad generation number %d in HELLO message, expected %d\n", | |
7904 | + le32_to_cpu(hellomsg->generation), | |
7905 | + cluster_generation); | |
7906 | + | |
7907 | + notify_kernel_listeners(DIED, | |
7908 | + (long) node->node_id); | |
7909 | + | |
b7b72b66 | 7910 | + send_kill(node->node_id); |
c1c6733f AM |
7911 | + return 0; |
7912 | + } | |
7913 | + | |
7914 | + if (cluster_members != le16_to_cpu(hellomsg->members) | |
7915 | + && node_state == MEMBER) { | |
7916 | + printk(KERN_INFO CMAN_NAME | |
b7b72b66 AM |
7917 | + ": nmembers in HELLO message does not match our view (got %d, exp %d)\n", |
7918 | + le16_to_cpu(hellomsg->members), cluster_members); | |
c1c6733f AM |
7919 | + start_transition(TRANS_CHECK, node); |
7920 | + return 0; | |
7921 | + } | |
7922 | + /* The message is OK - save the time */ | |
7923 | + node->last_hello = jiffies; | |
c1c6733f AM |
7924 | + } |
7925 | + else { | |
c1c6733f AM |
7926 | + /* This node is a danger to our valid cluster */ |
7927 | + if (cluster_is_quorate) { | |
b7b72b66 | 7928 | + send_kill(saddr->scl_nodeid); |
c1c6733f | 7929 | + } |
c1c6733f AM |
7930 | + } |
7931 | + } | |
7932 | + | |
7933 | + return 0; | |
7934 | + | |
7935 | +} | |
7936 | + | |
bb1d8b11 | 7937 | +static int do_process_kill(struct msghdr *msg, char *buf, int len) |
c1c6733f AM |
7938 | +{ |
7939 | + struct sockaddr_cl *saddr = msg->msg_name; | |
7940 | + struct cluster_node *node; | |
7941 | + | |
7942 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7943 | + if (node && node->state == NODESTATE_MEMBER) { | |
7944 | + | |
7945 | + printk(KERN_INFO CMAN_NAME | |
7946 | + ": Being told to leave the cluster by node %d\n", | |
7947 | + saddr->scl_nodeid); | |
7948 | + | |
7949 | + node_state = LEFT_CLUSTER; | |
7950 | + quit_threads = 1; | |
7951 | + wake_up_process(membership_task); | |
7952 | + wake_up_interruptible(&cnxman_waitq); | |
7953 | + } | |
7954 | + else { | |
7955 | + P_MEMB("Asked to leave the cluster by a non-member. What a nerve!\n"); | |
7956 | + } | |
7957 | + return 0; | |
7958 | +} | |
7959 | + | |
7960 | +/* Some cluster membership utility functions */ | |
7961 | +struct cluster_node *find_node_by_name(char *name) | |
7962 | +{ | |
7963 | + struct list_head *nodelist; | |
7964 | + struct cluster_node *node; | |
7965 | + | |
7966 | + down(&cluster_members_lock); | |
7967 | + list_for_each(nodelist, &cluster_members_list) { | |
7968 | + node = list_entry(nodelist, struct cluster_node, list); | |
7969 | + | |
7970 | + if (strcmp(node->name, name) == 0) { | |
7971 | + up(&cluster_members_lock); | |
7972 | + return node; | |
7973 | + } | |
7974 | + } | |
7975 | + up(&cluster_members_lock); | |
7976 | + return NULL; | |
7977 | +} | |
7978 | + | |
7979 | +/* Try to avoid using this as it's slow and holds the members lock */ | |
7980 | +struct cluster_node *find_node_by_addr(unsigned char *addr, int addr_len) | |
7981 | +{ | |
7982 | + struct list_head *nodelist; | |
7983 | + struct list_head *addrlist; | |
7984 | + struct cluster_node *node; | |
7985 | + struct cluster_node_addr *nodeaddr; | |
7986 | + | |
7987 | + down(&cluster_members_lock); | |
7988 | + | |
7989 | + list_for_each(nodelist, &cluster_members_list) { | |
7990 | + node = list_entry(nodelist, struct cluster_node, list); | |
7991 | + | |
7992 | + list_for_each(addrlist, &node->addr_list) { | |
7993 | + nodeaddr = | |
7994 | + list_entry(addrlist, struct cluster_node_addr, | |
7995 | + list); | |
7996 | + | |
b7b72b66 | 7997 | + if (memcmp(nodeaddr->addr+2, addr+2, address_length-2) == 0) { |
c1c6733f AM |
7998 | + up(&cluster_members_lock); |
7999 | + return node; | |
8000 | + } | |
8001 | + } | |
8002 | + } | |
8003 | + | |
8004 | + up(&cluster_members_lock); | |
8005 | + return NULL; | |
8006 | +} | |
8007 | + | |
8008 | +/* This is the quick way to find a node */ | |
8009 | +struct cluster_node *find_node_by_nodeid(unsigned int id) | |
8010 | +{ | |
8011 | + struct cluster_node *node; | |
8012 | + | |
8013 | + if (id > sizeof_members_array) | |
8014 | + return NULL; | |
8015 | + | |
8016 | + spin_lock(&members_by_nodeid_lock); | |
8017 | + node = members_by_nodeid[id]; | |
8018 | + spin_unlock(&members_by_nodeid_lock); | |
8019 | + return node; | |
8020 | +} | |
8021 | + | |
8022 | +static int dispatch_messages(struct socket *mem_socket) | |
8023 | +{ | |
8024 | + int err = 0; | |
8025 | + | |
8026 | + while (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
8027 | + struct msghdr msg; | |
bb1d8b11 | 8028 | + struct kvec vec; |
c1c6733f AM |
8029 | + struct sockaddr_cl sin; |
8030 | + int len; | |
c1c6733f AM |
8031 | + |
8032 | + memset(&sin, 0, sizeof (sin)); | |
8033 | + | |
8034 | + msg.msg_control = NULL; | |
8035 | + msg.msg_controllen = 0; | |
c1c6733f AM |
8036 | + msg.msg_name = &sin; |
8037 | + msg.msg_namelen = sizeof (sin); | |
8038 | + msg.msg_flags = 0; | |
8039 | + | |
bb1d8b11 AM |
8040 | + vec.iov_len = MAX_CLUSTER_MESSAGE; |
8041 | + vec.iov_base = iobuf; | |
c1c6733f | 8042 | + |
bb1d8b11 AM |
8043 | + len = kernel_recvmsg(mem_socket, &msg, &vec, 1, |
8044 | + MAX_CLUSTER_MESSAGE, | |
8045 | + MSG_DONTWAIT); | |
c1c6733f | 8046 | + if (len > 0) { |
c1c6733f | 8047 | + msg.msg_name = &sin; |
bb1d8b11 | 8048 | + do_membership_packet(&msg, iobuf, len); |
c1c6733f AM |
8049 | + } |
8050 | + else { | |
8051 | + if (len == -EAGAIN) | |
8052 | + err = 0; | |
8053 | + else | |
8054 | + err = -1; | |
8055 | + break; | |
8056 | + } | |
8057 | + } | |
8058 | + return err; | |
8059 | +} | |
8060 | + | |
8061 | +/* Scan the nodes list for dead nodes */ | |
8062 | +static void check_for_dead_nodes() | |
8063 | +{ | |
8064 | + struct list_head *nodelist; | |
8065 | + struct cluster_node *node; | |
8066 | + | |
8067 | + down(&cluster_members_lock); | |
8068 | + list_for_each(nodelist, &cluster_members_list) { | |
8069 | + node = list_entry(nodelist, struct cluster_node, list); | |
8070 | + | |
8071 | + if (node->state != NODESTATE_DEAD && | |
8072 | + time_after(jiffies, | |
8073 | + node->last_hello + | |
8074 | + cman_config.deadnode_timeout * HZ) && !node->us) { | |
8075 | + | |
8076 | + up(&cluster_members_lock); | |
8077 | + | |
8078 | + printk(KERN_WARNING CMAN_NAME | |
8079 | + ": no HELLO from %s, removing from the cluster\n", | |
8080 | + node->name); | |
8081 | + | |
8082 | + P_MEMB("last hello was %ld, current time is %ld\n", | |
8083 | + node->last_hello, jiffies); | |
8084 | + | |
8085 | + node->leave_reason = CLUSTER_LEAVEFLAG_DEAD; | |
8086 | + leavereason = 0; | |
8087 | + | |
8088 | + /* This is unlikely to work but it's worth a try! */ | |
8089 | + send_kill(node->node_id); | |
8090 | + | |
8091 | + /* Start state transition */ | |
8092 | + a_node_just_died(node); | |
8093 | + return; | |
8094 | + } | |
8095 | + } | |
8096 | + up(&cluster_members_lock); | |
8097 | + | |
8098 | + /* Also check for a dead quorum device */ | |
8099 | + if (quorum_device) { | |
8100 | + if (quorum_device->state == NODESTATE_MEMBER && | |
8101 | + time_after(jiffies, | |
8102 | + quorum_device->last_hello + | |
8103 | + cman_config.deadnode_timeout * HZ)) { | |
8104 | + quorum_device->state = NODESTATE_DEAD; | |
8105 | + printk(KERN_WARNING CMAN_NAME | |
8106 | + ": Quorum device %s timed out\n", | |
8107 | + quorum_device->name); | |
8108 | + recalculate_quorum(0); | |
8109 | + } | |
8110 | + } | |
8111 | + | |
8112 | + return; | |
8113 | +} | |
8114 | + | |
8115 | +/* add "us" as a node in the cluster */ | |
8116 | +static int add_us() | |
8117 | +{ | |
8118 | + struct cluster_node *newnode = | |
8119 | + kmalloc(sizeof (struct cluster_node), GFP_KERNEL); | |
8120 | + | |
8121 | + if (!newnode) { | |
8122 | + /* Oh shit, we have to commit hara kiri here for the greater | |
8123 | + * good of the cluster */ | |
8124 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
8125 | + | |
8126 | + printk(KERN_CRIT CMAN_NAME | |
8127 | + ": Cannot allocate memory for our node structure\n"); | |
8128 | + panic("Must die"); | |
8129 | + | |
8130 | + return -1; | |
8131 | + } | |
8132 | + | |
8133 | + memset(newnode, 0, sizeof (struct cluster_node)); | |
8134 | + newnode->name = kmalloc(strlen(nodename) + 1, GFP_KERNEL); | |
8135 | + if (!newnode->name) { | |
8136 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
8137 | + | |
8138 | + printk(KERN_CRIT CMAN_NAME | |
8139 | + ": Cannot allocate memory for node name\n"); | |
8140 | + kfree(newnode); | |
8141 | + | |
8142 | + panic("Must die"); | |
8143 | + | |
8144 | + return -1; | |
8145 | + } | |
8146 | + | |
8147 | + strcpy(newnode->name, nodename); | |
8148 | + newnode->last_hello = jiffies; | |
8149 | + newnode->votes = votes; | |
8150 | + newnode->expected_votes = expected_votes; | |
8151 | + newnode->state = NODESTATE_JOINING; | |
8152 | + newnode->node_id = 0; /* Will get filled in by ENDTRANS message */ | |
8153 | + newnode->us = 1; | |
8154 | + newnode->leave_reason = 0; | |
8155 | + INIT_LIST_HEAD(&newnode->addr_list); | |
8156 | + get_local_addresses(newnode); /* Get from cnxman socket info */ | |
8157 | + | |
8158 | + /* Add the new node to the list */ | |
8159 | + down(&cluster_members_lock); | |
8160 | + list_add(&newnode->list, &cluster_members_list); | |
8161 | + cluster_members++; | |
8162 | + up(&cluster_members_lock); | |
8163 | + us = newnode; | |
8164 | + | |
8165 | + return 0; | |
8166 | +} | |
8167 | + | |
8168 | +/* Return the highest known node_id */ | |
8169 | +unsigned int get_highest_nodeid() | |
8170 | +{ | |
8171 | + struct list_head *nodelist; | |
8172 | + struct cluster_node *node = NULL; | |
8173 | + unsigned int highest = 0; | |
8174 | + | |
8175 | + down(&cluster_members_lock); | |
8176 | + list_for_each(nodelist, &cluster_members_list) { | |
8177 | + node = list_entry(nodelist, struct cluster_node, list); | |
8178 | + | |
8179 | + if (node->node_id > highest) | |
8180 | + highest = node->node_id; | |
8181 | + } | |
8182 | + up(&cluster_members_lock); | |
8183 | + | |
8184 | + return highest; | |
8185 | +} | |
8186 | + | |
8187 | +/* Elect a new master if there is a clash. Returns 1 if we are the new master, | |
8188 | + * the master's struct will also be returned. This, rather primitively, uses | |
8189 | + * the lowest node ID */ | |
8190 | +static int elect_master(struct cluster_node **master_node) | |
8191 | +{ | |
8192 | + int i; | |
8193 | + | |
8194 | + for (i = 1; i < sizeof_members_array; i++) { | |
8195 | + if (members_by_nodeid[i] | |
8196 | + && members_by_nodeid[i]->state == NODESTATE_MEMBER) { | |
8197 | + *master_node = members_by_nodeid[i]; | |
8198 | + P_MEMB("Elected master is %s\n", (*master_node)->name); | |
8199 | + return (*master_node)->us; | |
8200 | + } | |
8201 | + } | |
8202 | + BUG(); | |
8203 | + return 0; | |
8204 | +} | |
8205 | + | |
8206 | +/* Called by node_cleanup in cnxman when we have left the cluster */ | |
8207 | +void free_nodeid_array() | |
8208 | +{ | |
8209 | + vfree(members_by_nodeid); | |
8210 | + members_by_nodeid = NULL; | |
8211 | + sizeof_members_array = 0; | |
8212 | +} | |
8213 | + | |
8214 | +int allocate_nodeid_array() | |
8215 | +{ | |
8216 | + /* Allocate space for the nodeid lookup array */ | |
8217 | + if (!members_by_nodeid) { | |
8218 | + spin_lock_init(&members_by_nodeid_lock); | |
8219 | + members_by_nodeid = | |
8220 | + vmalloc(cman_config.max_nodes * | |
8221 | + sizeof (struct cluster_member *)); | |
8222 | + } | |
8223 | + | |
8224 | + if (!members_by_nodeid) { | |
8225 | + printk(KERN_WARNING | |
8226 | + "Unable to allocate members array for %d members\n", | |
8227 | + cman_config.max_nodes); | |
8228 | + return -ENOMEM; | |
8229 | + } | |
8230 | + memset(members_by_nodeid, 0, | |
8231 | + cman_config.max_nodes * sizeof (struct cluster_member *)); | |
8232 | + sizeof_members_array = cman_config.max_nodes; | |
8233 | + | |
8234 | + return 0; | |
8235 | +} | |
8236 | + | |
8237 | +/* Set the votes & expected_votes variables */ | |
8238 | +void set_votes(int v, int e) | |
8239 | +{ | |
8240 | + votes = v; | |
8241 | + expected_votes = e; | |
8242 | +} | |
8243 | + | |
8244 | +int get_quorum() | |
8245 | +{ | |
8246 | + return quorum; | |
8247 | +} | |
8248 | + | |
8249 | +/* Called by cnxman to see if activity should be blocked because we are in a | |
8250 | + * state transition */ | |
8251 | +int in_transition() | |
8252 | +{ | |
8253 | + return node_state == TRANSITION || | |
8254 | + node_state == TRANSITION_COMPLETE || node_state == MASTER; | |
8255 | +} | |
8256 | + | |
8257 | +/* Return the current membership state as a string for the main line to put | |
8258 | + * into /proc . I really should be using snprintf rather than sprintf but it's | |
8259 | + * not exported... */ | |
8260 | +char *membership_state(char *buf, int buflen) | |
8261 | +{ | |
8262 | + switch (node_state) { | |
8263 | + case STARTING: | |
8264 | + strncpy(buf, "Starting", buflen); | |
8265 | + break; | |
c783755a AM |
8266 | + case NEWCLUSTER: |
8267 | + strncpy(buf, "New-Cluster?", buflen); | |
8268 | + break; | |
c1c6733f AM |
8269 | + case JOINING: |
8270 | + strncpy(buf, "Joining", buflen); | |
8271 | + break; | |
8272 | + case JOINWAIT: | |
8273 | + strncpy(buf, "Join-Wait", buflen); | |
8274 | + break; | |
8275 | + case JOINACK: | |
8276 | + strncpy(buf, "Join-Ack", buflen); | |
8277 | + break; | |
8278 | + case TRANSITION: | |
8279 | + sprintf(buf, "State-Transition: Master is %s", | |
8280 | + master_node ? master_node->name : "Unknown"); | |
8281 | + break; | |
8282 | + case MEMBER: | |
8283 | + strncpy(buf, "Cluster-Member", buflen); | |
8284 | + break; | |
8285 | + case REJECTED: | |
8286 | + strncpy(buf, "Rejected", buflen); | |
8287 | + break; | |
8288 | + case LEFT_CLUSTER: | |
c783755a | 8289 | + strncpy(buf, "Not-in-Cluster", buflen); |
c1c6733f AM |
8290 | + break; |
8291 | + case TRANSITION_COMPLETE: | |
8292 | + strncpy(buf, "Transition-Complete", buflen); | |
8293 | + break; | |
8294 | + case MASTER: | |
8295 | + strncpy(buf, "Transition-Master", buflen); | |
8296 | + break; | |
8297 | + default: | |
8298 | + sprintf(buf, "Unknown: code=%d", node_state); | |
8299 | + break; | |
8300 | + } | |
8301 | + | |
8302 | + return buf; | |
8303 | +} | |
8304 | + | |
c783755a AM |
8305 | +char *leave_string(int reason) |
8306 | +{ | |
bb1d8b11 | 8307 | + static char msg[32]; |
c783755a AM |
8308 | + switch (reason) |
8309 | + { | |
8310 | + case CLUSTER_LEAVEFLAG_DOWN: | |
8311 | + return "Shutdown"; | |
8312 | + case CLUSTER_LEAVEFLAG_KILLED: | |
8313 | + return "Killed by another node"; | |
8314 | + case CLUSTER_LEAVEFLAG_PANIC: | |
8315 | + return "Panic"; | |
8316 | + case CLUSTER_LEAVEFLAG_REMOVED: | |
8317 | + return "Removed"; | |
8318 | + case CLUSTER_LEAVEFLAG_REJECTED: | |
8319 | + return "Membership rejected"; | |
8320 | + default: | |
bb1d8b11 AM |
8321 | + sprintf(msg, "Reason is %d\n", reason); |
8322 | + return msg; | |
c783755a AM |
8323 | + } |
8324 | +} | |
8325 | + | |
c1c6733f AM |
8326 | +#ifdef DEBUG_MEMB |
8327 | +static char *msgname(int msg) | |
8328 | +{ | |
8329 | + switch (msg) { | |
8330 | + case CLUSTER_MEM_JOINCONF: | |
8331 | + return "JOINCONF"; | |
8332 | + case CLUSTER_MEM_JOINREQ: | |
8333 | + return "JOINREQ"; | |
8334 | + case CLUSTER_MEM_LEAVE: | |
8335 | + return "LEAVE"; | |
8336 | + case CLUSTER_MEM_HELLO: | |
8337 | + return "HELLO"; | |
8338 | + case CLUSTER_MEM_KILL: | |
8339 | + return "KILL"; | |
8340 | + case CLUSTER_MEM_JOINACK: | |
8341 | + return "JOINACK"; | |
8342 | + case CLUSTER_MEM_ENDTRANS: | |
8343 | + return "ENDTRANS"; | |
8344 | + case CLUSTER_MEM_RECONFIG: | |
8345 | + return "RECONFIG"; | |
8346 | + case CLUSTER_MEM_MASTERVIEW: | |
8347 | + return "MASTERVIEW"; | |
8348 | + case CLUSTER_MEM_STARTTRANS: | |
8349 | + return "STARTTRANS"; | |
8350 | + case CLUSTER_MEM_JOINREJ: | |
8351 | + return "JOINREJ"; | |
8352 | + case CLUSTER_MEM_VIEWACK: | |
8353 | + return "VIEWACK"; | |
8354 | + case CLUSTER_MEM_STARTACK: | |
8355 | + return "STARTACK"; | |
8356 | + case CLUSTER_MEM_NEWCLUSTER: | |
8357 | + return "NEWCLUSTER"; | |
8358 | + case CLUSTER_MEM_CONFACK: | |
8359 | + return "CONFACK"; | |
8360 | + case CLUSTER_MEM_NOMINATE: | |
8361 | + return "NOMINATE"; | |
8362 | + | |
8363 | + default: | |
8364 | + return "??UNKNOWN??"; | |
8365 | + } | |
8366 | +} | |
8367 | + | |
8368 | +#endif | |
8369 | + | |
8370 | +/* | |
8371 | + * Overrides for Emacs so that we follow Linus's tabbing style. | |
8372 | + * Emacs will notice this stuff at the end of the file and automatically | |
8373 | + * adjust the settings for this buffer only. This must remain at the end | |
8374 | + * of the file. | |
8375 | + * --------------------------------------------------------------------------- | |
8376 | + * Local variables: | |
8377 | + * c-file-style: "linux" | |
8378 | + * End: | |
8379 | + */ | |
8380 | diff -urN linux-orig/cluster/cman/proc.c linux-patched/cluster/cman/proc.c | |
bb1d8b11 AM |
8381 | --- linux-orig/cluster/cman/proc.c 1970-01-01 07:30:00.000000000 +0730 |
8382 | +++ linux-patched/cluster/cman/proc.c 2004-11-03 11:37:37.000000000 +0800 | |
c783755a | 8383 | @@ -0,0 +1,372 @@ |
c1c6733f AM |
8384 | +/****************************************************************************** |
8385 | +******************************************************************************* | |
8386 | +** | |
8387 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8388 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8389 | +** | |
8390 | +** This copyrighted material is made available to anyone wishing to use, | |
8391 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8392 | +** of the GNU General Public License v.2. | |
8393 | +** | |
8394 | +******************************************************************************* | |
8395 | +******************************************************************************/ | |
8396 | + | |
8397 | +#include <linux/init.h> | |
8398 | +#include <linux/socket.h> | |
8399 | +#include <linux/kernel.h> | |
8400 | +#include <linux/sched.h> | |
8401 | +#include <linux/file.h> | |
8402 | +#include <linux/proc_fs.h> | |
8403 | +#include <linux/seq_file.h> | |
8404 | +#include <linux/list.h> | |
8405 | +#include <linux/in.h> | |
8406 | +#include <net/sock.h> | |
8407 | +#include <cluster/cnxman.h> | |
8408 | +#include <cluster/service.h> | |
8409 | + | |
8410 | +#include "cnxman-private.h" | |
8411 | +#include "config.h" | |
8412 | + | |
8413 | +extern int cluster_members; | |
8414 | +extern struct list_head cluster_members_list; | |
8415 | +extern struct semaphore cluster_members_lock; | |
8416 | +extern struct cluster_node *quorum_device; | |
8417 | +extern int we_are_a_cluster_member; | |
8418 | +extern int cluster_is_quorate; | |
bb1d8b11 | 8419 | +extern uint16_t cluster_id; |
c1c6733f AM |
8420 | +extern atomic_t use_count; |
8421 | +extern unsigned int address_length; | |
8422 | +extern unsigned int config_version; | |
8423 | +extern char cluster_name[]; | |
8424 | +extern struct cluster_node *us; | |
8425 | +static struct seq_operations cluster_info_op; | |
8426 | + | |
b7b72b66 | 8427 | +int sm_proc_open(struct inode *inode, struct file *file); |
c1c6733f AM |
8428 | +int sm_debug_info(char *b, char **start, off_t offset, int length); |
8429 | + | |
8430 | +/* /proc interface to the configuration struct */ | |
8431 | +static struct config_proc_info { | |
8432 | + char *name; | |
8433 | + int *value; | |
8434 | +} config_proc[] = { | |
8435 | + { | |
8436 | + .name = "joinwait_timeout", | |
8437 | + .value = &cman_config.joinwait_timeout, | |
8438 | + }, | |
8439 | + { | |
8440 | + .name = "joinconf_timeout", | |
8441 | + .value = &cman_config.joinconf_timeout, | |
8442 | + }, | |
8443 | + { | |
8444 | + .name = "join_timeout", | |
8445 | + .value = &cman_config.join_timeout, | |
8446 | + }, | |
8447 | + { | |
8448 | + .name = "hello_timer", | |
8449 | + .value = &cman_config.hello_timer, | |
8450 | + }, | |
8451 | + { | |
8452 | + .name = "deadnode_timeout", | |
8453 | + .value = &cman_config.deadnode_timeout, | |
8454 | + }, | |
8455 | + { | |
8456 | + .name = "transition_timeout", | |
8457 | + .value = &cman_config.transition_timeout, | |
8458 | + }, | |
8459 | + { | |
8460 | + .name = "transition_restarts", | |
8461 | + .value = &cman_config.transition_restarts, | |
8462 | + }, | |
8463 | + { | |
8464 | + .name = "max_nodes", | |
8465 | + .value = &cman_config.max_nodes, | |
8466 | + }, | |
8467 | + { | |
8468 | + .name = "sm_debug_size", | |
8469 | + .value = &cman_config.sm_debug_size, | |
8470 | + }, | |
c783755a AM |
8471 | + { |
8472 | + .name = "newcluster_timeout", | |
8473 | + .value = &cman_config.newcluster_timeout, | |
8474 | + }, | |
c1c6733f AM |
8475 | +}; |
8476 | + | |
8477 | + | |
8478 | +static int proc_cluster_status(char *b, char **start, off_t offset, int length) | |
8479 | +{ | |
8480 | + struct list_head *nodelist; | |
8481 | + struct cluster_node *node; | |
8482 | + struct cluster_node_addr *node_addr; | |
8483 | + unsigned int total_votes = 0; | |
8484 | + unsigned int max_expected = 0; | |
8485 | + int c = 0; | |
8486 | + char node_buf[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
8487 | + | |
c783755a AM |
8488 | + c += sprintf(b+c, |
8489 | + "Version: %d.%d.%d\n", | |
8490 | + CNXMAN_MAJOR_VERSION, CNXMAN_MINOR_VERSION, | |
8491 | + CNXMAN_PATCH_VERSION); | |
8492 | + | |
8493 | + c += sprintf(b+c, | |
8494 | + "Config version: %d\nCluster name: %s\nCluster ID: %d\nMembership state: %s\n", | |
8495 | + config_version, | |
8496 | + cluster_name, cluster_id, | |
8497 | + membership_state(node_buf, sizeof (node_buf))); | |
8498 | + | |
8499 | + if (!we_are_a_cluster_member) | |
c1c6733f | 8500 | + return c; |
c1c6733f AM |
8501 | + |
8502 | + /* Total the votes */ | |
8503 | + down(&cluster_members_lock); | |
8504 | + list_for_each(nodelist, &cluster_members_list) { | |
8505 | + node = list_entry(nodelist, struct cluster_node, list); | |
8506 | + if (node->state == NODESTATE_MEMBER) { | |
8507 | + total_votes += node->votes; | |
8508 | + max_expected = | |
8509 | + max(max_expected, node->expected_votes); | |
8510 | + } | |
8511 | + } | |
8512 | + up(&cluster_members_lock); | |
8513 | + | |
8514 | + if (quorum_device && quorum_device->state == NODESTATE_MEMBER) | |
8515 | + total_votes += quorum_device->votes; | |
8516 | + | |
8517 | + c += sprintf(b+c, | |
c1c6733f AM |
8518 | + "Nodes: %d\nExpected_votes: %d\nTotal_votes: %d\nQuorum: %d %s\n", |
8519 | + cluster_members, max_expected, total_votes, | |
8520 | + get_quorum(), | |
8521 | + cluster_is_quorate ? " " : "Activity blocked"); | |
8522 | + c += sprintf(b+c, "Active subsystems: %d\n", | |
8523 | + atomic_read(&use_count)); | |
8524 | + | |
8525 | + | |
8526 | + c += sprintf(b+c, "Node addresses: "); | |
8527 | + list_for_each_entry(node_addr, &us->addr_list, list) { | |
8528 | + struct sockaddr_in6 *saddr = (struct sockaddr_in6 *)node_addr->addr; | |
8529 | + if (saddr->sin6_family == AF_INET6) { | |
8530 | + c += sprintf(b+c, "%x:%x:%x:%x:%x:%x:%x:%x ", | |
8531 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[0]), | |
8532 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[1]), | |
8533 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[2]), | |
8534 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[3]), | |
8535 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[4]), | |
8536 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[5]), | |
8537 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[6]), | |
8538 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[7])); | |
8539 | + } | |
8540 | + else { | |
8541 | + struct sockaddr_in *saddr4 = (struct sockaddr_in *)saddr; | |
8542 | + uint8_t *addr = (uint8_t *)&saddr4->sin_addr; | |
8543 | + c+= sprintf(b+c, "%u.%u.%u.%u ", | |
8544 | + addr[0], addr[1], addr[2], addr[3]); | |
8545 | + } | |
8546 | + } | |
8547 | + c += sprintf(b+c, "\n\n"); | |
8548 | + return c; | |
8549 | +} | |
8550 | + | |
8551 | + | |
8552 | +/* Allocate one of these for /proc/cluster/nodes so we can keep a track of where | |
8553 | + * we are */ | |
8554 | +struct cluster_seq_info { | |
8555 | + int nodeid; | |
8556 | + int highest_nodeid; | |
8557 | +}; | |
8558 | + | |
8559 | +static int cluster_open(struct inode *inode, struct file *file) | |
8560 | +{ | |
8561 | + return seq_open(file, &cluster_info_op); | |
8562 | +} | |
8563 | + | |
8564 | +static void *cluster_seq_start(struct seq_file *m, loff_t * pos) | |
8565 | +{ | |
8566 | + struct cluster_seq_info *csi = | |
8567 | + kmalloc(sizeof (struct cluster_seq_info), GFP_KERNEL); | |
8568 | + | |
8569 | + if (!csi) | |
8570 | + return NULL; | |
8571 | + | |
8572 | + /* Keep highest_nodeid here so we don't need to keep traversing the | |
8573 | + * list to find it */ | |
8574 | + csi->nodeid = *pos; | |
8575 | + csi->highest_nodeid = get_highest_nodeid(); | |
8576 | + | |
8577 | + /* Print the header */ | |
8578 | + if (*pos == 0) { | |
c783755a | 8579 | + seq_printf(m, "Node Votes Exp Sts Name\n"); |
c1c6733f AM |
8580 | + } |
8581 | + return csi; | |
8582 | +} | |
8583 | + | |
8584 | +static void *cluster_seq_next(struct seq_file *m, void *p, loff_t * pos) | |
8585 | +{ | |
8586 | + struct cluster_seq_info *csi = p; | |
8587 | + | |
8588 | + *pos = ++csi->nodeid; | |
8589 | + if (csi->nodeid > csi->highest_nodeid) | |
8590 | + return NULL; | |
8591 | + | |
8592 | + return csi; | |
8593 | +} | |
8594 | + | |
8595 | +static int cluster_seq_show(struct seq_file *m, void *p) | |
8596 | +{ | |
8597 | + char state = '?'; | |
8598 | + struct cluster_node *node; | |
8599 | + struct cluster_seq_info *csi = p; | |
8600 | + | |
8601 | + /* | |
8602 | + * If we have "0" here then display the quorum device if | |
8603 | + * there is one. | |
8604 | + */ | |
8605 | + if (csi->nodeid == 0) | |
8606 | + node = quorum_device; | |
8607 | + else | |
8608 | + node = find_node_by_nodeid(csi->nodeid); | |
8609 | + | |
8610 | + if (!node) | |
8611 | + return 0; | |
8612 | + | |
8613 | + /* Make state printable */ | |
8614 | + switch (node->state) { | |
8615 | + case NODESTATE_MEMBER: | |
8616 | + state = 'M'; | |
8617 | + break; | |
8618 | + case NODESTATE_JOINING: | |
8619 | + state = 'J'; | |
8620 | + break; | |
c1c6733f AM |
8621 | + case NODESTATE_DEAD: |
8622 | + state = 'X'; | |
8623 | + break; | |
8624 | + } | |
c783755a | 8625 | + seq_printf(m, "%4d %3d %3d %c %s\n", |
c1c6733f AM |
8626 | + node->node_id, |
8627 | + node->votes, | |
8628 | + node->expected_votes, | |
8629 | + state, | |
8630 | + node->name); | |
8631 | + | |
8632 | + return 0; | |
8633 | +} | |
8634 | + | |
8635 | +static void cluster_seq_stop(struct seq_file *m, void *p) | |
8636 | +{ | |
8637 | + kfree(p); | |
8638 | +} | |
8639 | + | |
8640 | +static struct seq_operations cluster_info_op = { | |
8641 | + .start = cluster_seq_start, | |
8642 | + .next = cluster_seq_next, | |
8643 | + .stop = cluster_seq_stop, | |
8644 | + .show = cluster_seq_show | |
8645 | +}; | |
8646 | + | |
8647 | +static struct file_operations cluster_fops = { | |
8648 | + .open = cluster_open, | |
8649 | + .read = seq_read, | |
8650 | + .llseek = seq_lseek, | |
8651 | + .release = seq_release, | |
c783755a | 8652 | + .owner = THIS_MODULE, |
c1c6733f AM |
8653 | +}; |
8654 | + | |
b7b72b66 AM |
8655 | +static struct file_operations service_fops = { |
8656 | + .open = sm_proc_open, | |
8657 | + .read = seq_read, | |
8658 | + .llseek = seq_lseek, | |
8659 | + .release = seq_release, | |
c783755a | 8660 | + .owner = THIS_MODULE, |
b7b72b66 AM |
8661 | +}; |
8662 | + | |
c1c6733f AM |
8663 | +static int cman_config_read_proc(char *page, char **start, off_t off, int count, |
8664 | + int *eof, void *data) | |
8665 | +{ | |
8666 | + struct config_proc_info *cinfo = data; | |
8667 | + | |
8668 | + return snprintf(page, count, "%d\n", *cinfo->value); | |
8669 | +} | |
8670 | + | |
8671 | +static int cman_config_write_proc(struct file *file, const char *buffer, | |
8672 | + unsigned long count, void *data) | |
8673 | +{ | |
8674 | + struct config_proc_info *cinfo = data; | |
8675 | + int value; | |
8676 | + char *end; | |
8677 | + | |
8678 | + value = simple_strtoul(buffer, &end, 10); | |
8679 | + if (*end) { | |
8680 | + *cinfo->value = value; | |
8681 | + } | |
8682 | + return count; | |
8683 | +} | |
8684 | + | |
8685 | +/* Base of the config directory for cman */ | |
8686 | +static struct proc_dir_entry *proc_cman_config; | |
8687 | +void create_proc_entries(void) | |
8688 | +{ | |
8689 | + struct proc_dir_entry *procentry; | |
8690 | + struct proc_dir_entry *proc_cluster; | |
8691 | + int i; | |
8692 | + | |
8693 | + proc_cluster = proc_mkdir("cluster", 0); | |
8694 | + if (!proc_cluster) | |
8695 | + return; | |
8696 | + proc_cluster->owner = THIS_MODULE; | |
8697 | + | |
8698 | + /* Config dir filled in by us and others */ | |
8699 | + if (!proc_mkdir("cluster/config", 0)) | |
8700 | + return; | |
8701 | + | |
8702 | + /* Don't much care if this fails, it's hardly vital */ | |
8703 | + procentry = create_proc_entry("cluster/nodes", S_IRUGO, NULL); | |
8704 | + if (procentry) | |
8705 | + procentry->proc_fops = &cluster_fops; | |
8706 | + | |
8707 | + procentry = create_proc_entry("cluster/status", S_IRUGO, NULL); | |
8708 | + if (procentry) | |
c783755a | 8709 | + procentry->get_info = proc_cluster_status; |
c1c6733f | 8710 | + |
c783755a AM |
8711 | + procentry = create_proc_entry("cluster/services", S_IRUGO, NULL); |
8712 | + if (procentry) | |
8713 | + procentry->proc_fops = &service_fops; | |
c1c6733f | 8714 | + |
c783755a AM |
8715 | + /* Config entries */ |
8716 | + proc_cman_config = proc_mkdir("cluster/config/cman", 0); | |
8717 | + if (!proc_cman_config) | |
8718 | + return; | |
c1c6733f | 8719 | + |
c783755a AM |
8720 | + for (i=0; i<sizeof(config_proc)/sizeof(struct config_proc_info); i++) { |
8721 | + procentry = create_proc_entry(config_proc[i].name, 0660, | |
8722 | + proc_cman_config); | |
8723 | + if (procentry) { | |
8724 | + procentry->data = &config_proc[i]; | |
8725 | + procentry->write_proc = cman_config_write_proc; | |
8726 | + procentry->read_proc = cman_config_read_proc; | |
8727 | + } | |
8728 | + } | |
c1c6733f | 8729 | + |
c783755a AM |
8730 | + procentry = create_proc_entry("cluster/sm_debug", S_IRUGO, NULL); |
8731 | + if (procentry) | |
8732 | + procentry->get_info = sm_debug_info; | |
8733 | +} | |
c1c6733f | 8734 | + |
c783755a AM |
8735 | +void cleanup_proc_entries(void) |
8736 | +{ | |
8737 | + int i, config_count; | |
c1c6733f | 8738 | + |
c783755a | 8739 | + remove_proc_entry("cluster/sm_debug", NULL); |
c1c6733f | 8740 | + |
c783755a | 8741 | + config_count = sizeof(config_proc) / sizeof(struct config_proc_info); |
c1c6733f | 8742 | + |
c783755a AM |
8743 | + if (proc_cman_config) { |
8744 | + for (i=0; i<config_count; i++) | |
8745 | + remove_proc_entry(config_proc[i].name, proc_cman_config); | |
8746 | + } | |
8747 | + remove_proc_entry("cluster/config/cman", NULL); | |
8748 | + remove_proc_entry("cluster/config", NULL); | |
c1c6733f | 8749 | + |
c783755a AM |
8750 | + remove_proc_entry("cluster/nodes", NULL); |
8751 | + remove_proc_entry("cluster/status", NULL); | |
8752 | + remove_proc_entry("cluster/services", NULL); | |
8753 | + remove_proc_entry("cluster/config", NULL); | |
8754 | + remove_proc_entry("cluster", NULL); | |
8755 | +} | |
bb1d8b11 AM |
8756 | diff -urN linux-orig/cluster/cman/sm.h linux-patched/cluster/cman/sm.h |
8757 | --- linux-orig/cluster/cman/sm.h 1970-01-01 07:30:00.000000000 +0730 | |
8758 | +++ linux-patched/cluster/cman/sm.h 2004-11-03 11:37:37.000000000 +0800 | |
8759 | @@ -0,0 +1,109 @@ | |
8760 | +/****************************************************************************** | |
8761 | +******************************************************************************* | |
8762 | +** | |
8763 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8764 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8765 | +** | |
8766 | +** This copyrighted material is made available to anyone wishing to use, | |
8767 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8768 | +** of the GNU General Public License v.2. | |
8769 | +** | |
8770 | +******************************************************************************* | |
8771 | +******************************************************************************/ | |
8772 | + | |
8773 | +#ifndef __SM_DOT_H__ | |
8774 | +#define __SM_DOT_H__ | |
8775 | + | |
8776 | +/* | |
8777 | + * This is the main header file to be included in each Service Manager source | |
8778 | + * file. | |
8779 | + */ | |
8780 | + | |
8781 | +#include <linux/list.h> | |
8782 | +#include <linux/socket.h> | |
8783 | +#include <linux/kernel.h> | |
8784 | +#include <linux/sched.h> | |
8785 | +#include <linux/file.h> | |
8786 | +#include <linux/kthread.h> | |
8787 | +#include <net/sock.h> | |
8788 | + | |
8789 | +#include <cluster/cnxman.h> | |
8790 | +#include <cluster/service.h> | |
8791 | + | |
8792 | +#define SG_LEVELS (4) | |
8793 | + | |
8794 | +#include "sm_internal.h" | |
8795 | +#include "sm_barrier.h" | |
8796 | +#include "sm_control.h" | |
8797 | +#include "sm_daemon.h" | |
8798 | +#include "sm_joinleave.h" | |
8799 | +#include "sm_membership.h" | |
8800 | +#include "sm_message.h" | |
8801 | +#include "sm_misc.h" | |
8802 | +#include "sm_recover.h" | |
8803 | +#include "sm_services.h" | |
8804 | + | |
8805 | +extern struct list_head sm_sg[SG_LEVELS]; | |
8806 | +extern struct semaphore sm_sglock; | |
8807 | + | |
8808 | +#ifndef TRUE | |
8809 | +#define TRUE (1) | |
8810 | +#endif | |
8811 | + | |
8812 | +#ifndef FALSE | |
8813 | +#define FALSE (0) | |
8814 | +#endif | |
8815 | + | |
8816 | +#define SM_ASSERT(x, do) \ | |
8817 | +{ \ | |
8818 | + if (!(x)) \ | |
8819 | + { \ | |
8820 | + printk("\nSM: Assertion failed on line %d of file %s\n" \ | |
8821 | + "SM: assertion: \"%s\"\n" \ | |
8822 | + "SM: time = %lu\n", \ | |
8823 | + __LINE__, __FILE__, #x, jiffies); \ | |
8824 | + {do} \ | |
8825 | + printk("\n"); \ | |
8826 | + panic("SM: Record message above and reboot.\n"); \ | |
8827 | + } \ | |
8828 | +} | |
8829 | + | |
8830 | +#define SM_RETRY(do_this, until_this) \ | |
8831 | +for (;;) \ | |
8832 | +{ \ | |
8833 | + do { do_this; } while (0); \ | |
8834 | + if (until_this) \ | |
8835 | + break; \ | |
8836 | + printk("SM: out of memory: %s, %u\n", __FILE__, __LINE__); \ | |
8837 | + schedule();\ | |
8838 | +} | |
8839 | + | |
8840 | + | |
8841 | +#define log_print(fmt, args...) printk("SM: "fmt"\n", ##args) | |
8842 | + | |
8843 | +#define log_error(sg, fmt, args...) \ | |
8844 | + printk("SM: %08x " fmt "\n", (sg)->global_id , ##args) | |
8845 | + | |
8846 | + | |
8847 | +#define SM_DEBUG_LOG | |
8848 | + | |
8849 | +#ifdef SM_DEBUG_CONSOLE | |
8850 | +#define log_debug(sg, fmt, args...) \ | |
8851 | + printk("SM: %08x " fmt "\n", (sg)->global_id , ##args) | |
8852 | +#endif | |
8853 | + | |
8854 | +#ifdef SM_DEBUG_LOG | |
8855 | +#define log_debug(sg, fmt, args...) sm_debug_log(sg, fmt, ##args); | |
8856 | +#endif | |
8857 | + | |
8858 | +#ifdef SM_DEBUG_ALL | |
8859 | +#define log_debug(sg, fmt, args...) \ | |
8860 | +do \ | |
8861 | +{ \ | |
8862 | + printk("SM: %08x "fmt"\n", (sg)->global_id, ##args); \ | |
8863 | + sm_debug_log(sg, fmt, ##args); \ | |
8864 | +} \ | |
8865 | +while (0) | |
8866 | +#endif | |
8867 | + | |
8868 | +#endif /* __SM_DOT_H__ */ | |
c1c6733f | 8869 | diff -urN linux-orig/cluster/cman/sm_barrier.c linux-patched/cluster/cman/sm_barrier.c |
bb1d8b11 AM |
8870 | --- linux-orig/cluster/cman/sm_barrier.c 1970-01-01 07:30:00.000000000 +0730 |
8871 | +++ linux-patched/cluster/cman/sm_barrier.c 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
8872 | @@ -0,0 +1,232 @@ |
8873 | +/****************************************************************************** | |
8874 | +******************************************************************************* | |
8875 | +** | |
8876 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8877 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8878 | +** | |
8879 | +** This copyrighted material is made available to anyone wishing to use, | |
8880 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8881 | +** of the GNU General Public License v.2. | |
8882 | +** | |
8883 | +******************************************************************************* | |
8884 | +******************************************************************************/ | |
8885 | + | |
8886 | +#include "sm.h" | |
8887 | + | |
8888 | +static struct list_head barriers; | |
8889 | +static spinlock_t barriers_lock; | |
8890 | + | |
8891 | +struct bc_entry { | |
8892 | + struct list_head list; | |
8893 | + uint32_t gid; | |
8894 | + int status; | |
8895 | + char type; | |
8896 | +}; | |
8897 | +typedef struct bc_entry bc_entry_t; | |
8898 | + | |
8899 | +void init_barriers(void) | |
8900 | +{ | |
8901 | + INIT_LIST_HEAD(&barriers); | |
8902 | + spin_lock_init(&barriers_lock); | |
8903 | +} | |
8904 | + | |
8905 | +static int atoi(char *c) | |
8906 | +{ | |
8907 | + int x = 0; | |
8908 | + | |
8909 | + while ('0' <= *c && *c <= '9') { | |
8910 | + x = x * 10 + (*c - '0'); | |
8911 | + c++; | |
8912 | + } | |
8913 | + return x; | |
8914 | +} | |
8915 | + | |
8916 | +static void add_barrier_callback(char *name, int status, int type) | |
8917 | +{ | |
8918 | + char *p; | |
8919 | + uint32_t gid; | |
8920 | + bc_entry_t *be; | |
8921 | + | |
8922 | + /* an ESRCH callback just means there was a cnxman transition */ | |
8923 | + if (status == -ESRCH) | |
8924 | + return; | |
8925 | + | |
8926 | + /* extract global id of SG from barrier name */ | |
8927 | + p = strstr(name, "sm."); | |
8928 | + | |
8929 | + SM_ASSERT(p, printk("name=\"%s\" status=%d\n", name, status);); | |
8930 | + | |
8931 | + p += strlen("sm."); | |
8932 | + gid = atoi(p); | |
8933 | + | |
8934 | + SM_RETRY(be = kmalloc(sizeof(bc_entry_t), GFP_ATOMIC), be); | |
8935 | + | |
8936 | + be->gid = gid; | |
8937 | + be->status = status; | |
8938 | + be->type = type; | |
8939 | + | |
8940 | + spin_lock(&barriers_lock); | |
8941 | + list_add_tail(&be->list, &barriers); | |
8942 | + spin_unlock(&barriers_lock); | |
8943 | + | |
8944 | + wake_serviced(DO_BARRIERS); | |
8945 | +} | |
8946 | + | |
8947 | +static void callback_recovery_barrier(char *name, int status) | |
8948 | +{ | |
8949 | + add_barrier_callback(name, status, SM_BARRIER_RECOVERY); | |
8950 | +} | |
8951 | + | |
8952 | +static void callback_startdone_barrier_new(char *name, int status) | |
8953 | +{ | |
8954 | + add_barrier_callback(name, status, SM_BARRIER_STARTDONE_NEW); | |
8955 | +} | |
8956 | + | |
8957 | +static void callback_startdone_barrier(char *name, int status) | |
8958 | +{ | |
8959 | + add_barrier_callback(name, status, SM_BARRIER_STARTDONE); | |
8960 | +} | |
8961 | + | |
8962 | +int sm_barrier(char *name, int count, int type) | |
8963 | +{ | |
8964 | + int error; | |
8965 | + unsigned long fn = 0; | |
8966 | + | |
8967 | + switch (type) { | |
8968 | + case SM_BARRIER_STARTDONE: | |
8969 | + fn = (unsigned long) callback_startdone_barrier; | |
8970 | + break; | |
8971 | + case SM_BARRIER_STARTDONE_NEW: | |
8972 | + fn = (unsigned long) callback_startdone_barrier_new; | |
8973 | + break; | |
8974 | + case SM_BARRIER_RECOVERY: | |
8975 | + fn = (unsigned long) callback_recovery_barrier; | |
8976 | + break; | |
8977 | + } | |
8978 | + | |
8979 | + error = kcl_barrier_register(name, 0, count); | |
8980 | + if (error) { | |
8981 | + log_print("barrier register error %d", error); | |
8982 | + goto fail; | |
8983 | + } | |
8984 | + | |
8985 | + error = kcl_barrier_setattr(name, BARRIER_SETATTR_AUTODELETE, TRUE); | |
8986 | + if (error) { | |
8987 | + log_print("barrier setattr autodel error %d", error); | |
8988 | + goto fail_bar; | |
8989 | + } | |
8990 | + | |
8991 | + error = kcl_barrier_setattr(name, BARRIER_SETATTR_CALLBACK, fn); | |
8992 | + if (error) { | |
8993 | + log_print("barrier setattr cb error %d", error); | |
8994 | + goto fail_bar; | |
8995 | + } | |
8996 | + | |
8997 | + error = kcl_barrier_setattr(name, BARRIER_SETATTR_ENABLED, TRUE); | |
8998 | + if (error) { | |
8999 | + log_print("barrier setattr enabled error %d", error); | |
9000 | + goto fail_bar; | |
9001 | + } | |
9002 | + | |
9003 | + return 0; | |
9004 | + | |
9005 | + fail_bar: | |
9006 | + kcl_barrier_delete(name); | |
9007 | + fail: | |
9008 | + return error; | |
9009 | +} | |
9010 | + | |
9011 | +void process_startdone_barrier_new(sm_group_t *sg, int status) | |
9012 | +{ | |
9013 | + sm_sevent_t *sev = sg->sevent; | |
9014 | + | |
9015 | + if (!test_and_clear_bit(SEFL_ALLOW_BARRIER, &sev->se_flags)) { | |
9016 | + log_debug(sev->se_sg, "ignore barrier cb status %d", status); | |
9017 | + return; | |
9018 | + } | |
9019 | + | |
9020 | + sev->se_barrier_status = status; | |
9021 | + sev->se_state = SEST_BARRIER_DONE; | |
9022 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
9023 | + wake_serviced(DO_JOINLEAVE); | |
9024 | +} | |
9025 | + | |
9026 | +void process_startdone_barrier(sm_group_t *sg, int status) | |
9027 | +{ | |
9028 | + sm_uevent_t *uev = &sg->uevent; | |
9029 | + | |
9030 | + if (!test_and_clear_bit(UEFL_ALLOW_BARRIER, &uev->ue_flags)) { | |
9031 | + log_debug(sg, "ignore barrier cb status %d", status); | |
9032 | + return; | |
9033 | + } | |
9034 | + | |
9035 | + uev->ue_barrier_status = status; | |
9036 | + uev->ue_state = UEST_BARRIER_DONE; | |
9037 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
9038 | + wake_serviced(DO_MEMBERSHIP); | |
9039 | +} | |
9040 | + | |
9041 | +void process_recovery_barrier(sm_group_t *sg, int status) | |
9042 | +{ | |
9043 | + if (status) { | |
9044 | + log_error(sg, "process_recovery_barrier status=%d", status); | |
9045 | + return; | |
9046 | + } | |
9047 | + | |
9048 | + if (sg->state != SGST_RECOVER || | |
9049 | + sg->recover_state != RECOVER_BARRIERWAIT) { | |
9050 | + log_error(sg, "process_recovery_barrier state %d recover %d", | |
9051 | + sg->state, sg->recover_state); | |
9052 | + return; | |
9053 | + } | |
9054 | + | |
9055 | + if (!sg->recover_stop) | |
9056 | + sg->recover_state = RECOVER_STOP; | |
9057 | + else | |
9058 | + sg->recover_state = RECOVER_BARRIERDONE; | |
9059 | + | |
9060 | + wake_serviced(DO_RECOVERIES); | |
9061 | +} | |
9062 | + | |
9063 | +void process_barriers(void) | |
9064 | +{ | |
9065 | + sm_group_t *sg; | |
9066 | + bc_entry_t *be; | |
9067 | + | |
9068 | + while (1) { | |
9069 | + be = NULL; | |
9070 | + | |
9071 | + spin_lock(&barriers_lock); | |
9072 | + if (!list_empty(&barriers)) { | |
9073 | + be = list_entry(barriers.next, bc_entry_t, list); | |
9074 | + list_del(&be->list); | |
9075 | + } | |
9076 | + spin_unlock(&barriers_lock); | |
9077 | + | |
9078 | + if (!be) | |
9079 | + break; | |
9080 | + | |
9081 | + sg = sm_global_id_to_sg(be->gid); | |
9082 | + if (!sg) { | |
9083 | + log_print("process_barriers: no sg %08x", be->gid); | |
9084 | + break; | |
9085 | + } | |
9086 | + | |
9087 | + switch (be->type) { | |
9088 | + case SM_BARRIER_STARTDONE_NEW: | |
9089 | + process_startdone_barrier_new(sg, be->status); | |
9090 | + break; | |
9091 | + | |
9092 | + case SM_BARRIER_STARTDONE: | |
9093 | + process_startdone_barrier(sg, be->status); | |
9094 | + break; | |
9095 | + | |
9096 | + case SM_BARRIER_RECOVERY: | |
9097 | + process_recovery_barrier(sg, be->status); | |
9098 | + break; | |
9099 | + } | |
9100 | + | |
9101 | + kfree(be); | |
9102 | + schedule(); | |
9103 | + } | |
9104 | +} | |
9105 | diff -urN linux-orig/cluster/cman/sm_barrier.h linux-patched/cluster/cman/sm_barrier.h | |
bb1d8b11 AM |
9106 | --- linux-orig/cluster/cman/sm_barrier.h 1970-01-01 07:30:00.000000000 +0730 |
9107 | +++ linux-patched/cluster/cman/sm_barrier.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
9108 | @@ -0,0 +1,29 @@ |
9109 | +/****************************************************************************** | |
9110 | +******************************************************************************* | |
9111 | +** | |
9112 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9113 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9114 | +** | |
9115 | +** This copyrighted material is made available to anyone wishing to use, | |
9116 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9117 | +** of the GNU General Public License v.2. | |
9118 | +** | |
9119 | +******************************************************************************* | |
9120 | +******************************************************************************/ | |
9121 | + | |
9122 | +#ifndef __SM_BARRIER_DOT_H__ | |
9123 | +#define __SM_BARRIER_DOT_H__ | |
9124 | + | |
9125 | +#define SM_BARRIER_STARTDONE (0) | |
9126 | +#define SM_BARRIER_STARTDONE_NEW (1) | |
9127 | +#define SM_BARRIER_RECOVERY (2) | |
9128 | +#define SM_BARRIER_RESET (3) | |
9129 | + | |
9130 | +void init_barriers(void); | |
9131 | +void process_barriers(void); | |
9132 | +int sm_barrier(char *name, int count, int type); | |
9133 | +void process_startdone_barrier(sm_group_t *sg, int status); | |
9134 | +void process_startdone_barrier_new(sm_group_t *sg, int status); | |
9135 | +void process_recovery_barrier(sm_group_t *sg, int status); | |
9136 | + | |
9137 | +#endif | |
9138 | diff -urN linux-orig/cluster/cman/sm_control.c linux-patched/cluster/cman/sm_control.c | |
bb1d8b11 AM |
9139 | --- linux-orig/cluster/cman/sm_control.c 1970-01-01 07:30:00.000000000 +0730 |
9140 | +++ linux-patched/cluster/cman/sm_control.c 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
9141 | @@ -0,0 +1,156 @@ |
9142 | +/****************************************************************************** | |
9143 | +******************************************************************************* | |
9144 | +** | |
9145 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9146 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9147 | +** | |
9148 | +** This copyrighted material is made available to anyone wishing to use, | |
9149 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9150 | +** of the GNU General Public License v.2. | |
9151 | +** | |
9152 | +******************************************************************************* | |
9153 | +******************************************************************************/ | |
9154 | + | |
9155 | +#include "sm.h" | |
9156 | +#include "config.h" | |
9157 | + | |
9158 | +struct socket * sm_socket; | |
9159 | +uint32_t * sm_new_nodeids; | |
9160 | +uint32_t sm_our_nodeid; | |
9161 | +int sm_quorum, sm_quorum_next; | |
9162 | +struct list_head sm_members; | |
9163 | +int sm_member_count; | |
9164 | + | |
9165 | + | |
9166 | +/* | |
9167 | + * Context: cnxman | |
9168 | + * Called by cnxman when it has a new member list. | |
9169 | + */ | |
9170 | + | |
9171 | +void sm_member_update(int quorate) | |
9172 | +{ | |
9173 | + sm_quorum_next = quorate; | |
9174 | + wake_serviced(DO_START_RECOVERY); | |
9175 | +} | |
9176 | + | |
9177 | +/* | |
9178 | + * Context: cnxman | |
9179 | + * Called when module is loaded. | |
9180 | + */ | |
9181 | + | |
9182 | +void sm_init(void) | |
9183 | +{ | |
9184 | + sm_socket = NULL; | |
9185 | + sm_new_nodeids = NULL; | |
9186 | + sm_quorum = 0; | |
9187 | + sm_quorum_next = 0; | |
9188 | + sm_our_nodeid = 0; | |
9189 | + INIT_LIST_HEAD(&sm_members); | |
9190 | + sm_member_count = 0; | |
9191 | + | |
9192 | + init_services(); | |
9193 | + init_messages(); | |
9194 | + init_barriers(); | |
9195 | + init_serviced(); | |
9196 | + init_recovery(); | |
9197 | + init_joinleave(); | |
9198 | + init_sm_misc(); | |
9199 | +} | |
9200 | + | |
9201 | +/* | |
9202 | + * Context: cnxman | |
9203 | + * Called at beginning of cluster join procedure. | |
9204 | + */ | |
9205 | + | |
9206 | +void sm_start(void) | |
9207 | +{ | |
9208 | + struct sockaddr_cl saddr; | |
9209 | + struct socket *sock; | |
9210 | + int result; | |
9211 | + | |
9212 | + /* Create a communication channel among service managers */ | |
9213 | + | |
9214 | + result = sock_create_kern(AF_CLUSTER, SOCK_DGRAM, CLPROTO_CLIENT, &sock); | |
9215 | + if (result < 0) { | |
9216 | + log_print("can't create socket %d", result); | |
9217 | + goto fail; | |
9218 | + } | |
9219 | + | |
9220 | + sm_socket = sock; | |
9221 | + | |
9222 | + saddr.scl_family = AF_CLUSTER; | |
9223 | + saddr.scl_port = CLUSTER_PORT_SERVICES; | |
9224 | + | |
9225 | + result = sock->ops->bind(sock, (struct sockaddr *) &saddr, | |
9226 | + sizeof(saddr)); | |
9227 | + if (result < 0) { | |
9228 | + log_print("can't bind socket %d", result); | |
9229 | + goto fail_release; | |
9230 | + } | |
9231 | + | |
9232 | + result = kcl_register_read_callback(sm_socket, sm_cluster_message); | |
9233 | + if (result < 0) { | |
9234 | + log_print("can't register read callback %d", result); | |
9235 | + goto fail_release; | |
9236 | + } | |
9237 | + | |
9238 | + sm_new_nodeids = (uint32_t *) kmalloc(cman_config.max_nodes * | |
9239 | + sizeof(uint32_t), | |
9240 | + GFP_KERNEL); | |
9241 | + start_serviced(); | |
9242 | + | |
9243 | + /* cnxman should call sm_member_update() once we've joined - then we | |
9244 | + * can get our first list of members and our own nodeid */ | |
9245 | + | |
9246 | + return; | |
9247 | + | |
9248 | + fail_release: | |
9249 | + sock_release(sm_socket); | |
9250 | + sm_socket = NULL; | |
9251 | + | |
9252 | + fail: | |
9253 | + return; | |
9254 | +} | |
9255 | + | |
9256 | +/* | |
9257 | + * Context: cnxman | |
9258 | + * Called before cnxman leaves the cluster. If this returns an error to cman, | |
9259 | + * cman should not leave the cluster but return EBUSY. | |
9260 | + * If force is set we go away anyway. cman knows best in this case | |
9261 | + */ | |
9262 | + | |
9263 | +int sm_stop(int force) | |
9264 | +{ | |
9265 | + struct list_head *head; | |
9266 | + sm_group_t *sg; | |
9267 | + sm_node_t *node; | |
9268 | + int i, busy = FALSE, error = -EBUSY; | |
9269 | + | |
9270 | + for (i = 0; i < SG_LEVELS; i++) { | |
9271 | + if (!list_empty(&sm_sg[i])) { | |
9272 | + sg = list_entry(sm_sg[i].next, sm_group_t, list); | |
9273 | + log_error(sg, "sm_stop: SG still joined"); | |
9274 | + busy = TRUE; | |
9275 | + } | |
9276 | + } | |
9277 | + | |
9278 | + if (!busy || force) { | |
9279 | + stop_serviced(); | |
9280 | + | |
9281 | + if (sm_socket) | |
9282 | + sock_release(sm_socket); | |
9283 | + | |
9284 | + head = &sm_members; | |
9285 | + while (!list_empty(head)) { | |
9286 | + node = list_entry(head->next, sm_node_t, list); | |
9287 | + list_del(&node->list); | |
9288 | + sm_member_count--; | |
9289 | + kfree(node); | |
9290 | + } | |
9291 | + | |
9292 | + kfree(sm_new_nodeids); | |
9293 | + sm_init(); | |
9294 | + error = 0; | |
9295 | + } | |
9296 | + return error; | |
9297 | +} | |
9298 | diff -urN linux-orig/cluster/cman/sm_control.h linux-patched/cluster/cman/sm_control.h | |
bb1d8b11 AM |
9299 | --- linux-orig/cluster/cman/sm_control.h 1970-01-01 07:30:00.000000000 +0730 |
9300 | +++ linux-patched/cluster/cman/sm_control.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
9301 | @@ -0,0 +1,22 @@ |
9302 | +/****************************************************************************** | |
9303 | +******************************************************************************* | |
9304 | +** | |
9305 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9306 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9307 | +** | |
9308 | +** This copyrighted material is made available to anyone wishing to use, | |
9309 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9310 | +** of the GNU General Public License v.2. | |
9311 | +** | |
9312 | +******************************************************************************* | |
9313 | +******************************************************************************/ | |
9314 | + | |
9315 | +#ifndef __SM_CONTROL_DOT_H__ | |
9316 | +#define __SM_CONTROL_DOT_H__ | |
9317 | + | |
9318 | +void sm_init(void); | |
9319 | +void sm_start(void); | |
9320 | +int sm_stop(int force); | |
9321 | +void sm_member_update(int quorate); | |
9322 | + | |
9323 | +#endif | |
9324 | diff -urN linux-orig/cluster/cman/sm_daemon.c linux-patched/cluster/cman/sm_daemon.c | |
bb1d8b11 AM |
9325 | --- linux-orig/cluster/cman/sm_daemon.c 1970-01-01 07:30:00.000000000 +0730 |
9326 | +++ linux-patched/cluster/cman/sm_daemon.c 2004-11-03 11:37:37.000000000 +0800 | |
c783755a | 9327 | @@ -0,0 +1,100 @@ |
c1c6733f AM |
9328 | +/****************************************************************************** |
9329 | +******************************************************************************* | |
9330 | +** | |
9331 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9332 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9333 | +** | |
9334 | +** This copyrighted material is made available to anyone wishing to use, | |
9335 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9336 | +** of the GNU General Public License v.2. | |
9337 | +** | |
9338 | +******************************************************************************* | |
9339 | +******************************************************************************/ | |
9340 | + | |
9341 | +#include "sm.h" | |
9342 | + | |
9343 | +static unsigned long daemon_flags; | |
9344 | +static struct task_struct * daemon_task; | |
c1c6733f AM |
9345 | +extern int sm_quorum; |
9346 | + | |
9347 | +void init_serviced(void) | |
9348 | +{ | |
9349 | + daemon_flags = 0; | |
9350 | + daemon_task = NULL; | |
c1c6733f AM |
9351 | +} |
9352 | + | |
9353 | +void wake_serviced(int do_flag) | |
9354 | +{ | |
9355 | + set_bit(do_flag, &daemon_flags); | |
c783755a | 9356 | + wake_up_process(daemon_task); |
c1c6733f AM |
9357 | +} |
9358 | + | |
9359 | +static inline int got_work(void) | |
9360 | +{ | |
9361 | + int rv = 0; | |
9362 | + | |
9363 | + rv = (test_bit(DO_START_RECOVERY, &daemon_flags) || | |
9364 | + test_bit(DO_MESSAGES, &daemon_flags) || | |
9365 | + test_bit(DO_BARRIERS, &daemon_flags) || | |
9366 | + test_bit(DO_CALLBACKS, &daemon_flags)); | |
9367 | + | |
9368 | + if (sm_quorum && !rv) | |
9369 | + rv = (test_bit(DO_JOINLEAVE, &daemon_flags) || | |
9370 | + test_bit(DO_RECOVERIES, &daemon_flags) || | |
9371 | + test_bit(DO_MEMBERSHIP, &daemon_flags)); | |
9372 | + return rv; | |
9373 | +} | |
9374 | + | |
9375 | +static int serviced(void *arg) | |
9376 | +{ | |
c783755a | 9377 | + while (!kthread_should_stop()) { |
c1c6733f AM |
9378 | + if (test_and_clear_bit(DO_START_RECOVERY, &daemon_flags)) |
9379 | + process_nodechange(); | |
9380 | + | |
9381 | + if (test_and_clear_bit(DO_MESSAGES, &daemon_flags)) | |
9382 | + process_messages(); | |
9383 | + | |
9384 | + if (test_and_clear_bit(DO_BARRIERS, &daemon_flags)) | |
9385 | + process_barriers(); | |
9386 | + | |
9387 | + if (test_and_clear_bit(DO_CALLBACKS, &daemon_flags)) | |
9388 | + process_callbacks(); | |
9389 | + | |
9390 | + if (sm_quorum) { | |
9391 | + if (test_and_clear_bit(DO_RECOVERIES, &daemon_flags)) | |
9392 | + process_recoveries(); | |
9393 | + | |
9394 | + if (test_and_clear_bit(DO_JOINLEAVE, &daemon_flags)) | |
9395 | + process_joinleave(); | |
9396 | + | |
9397 | + if (test_and_clear_bit(DO_MEMBERSHIP, &daemon_flags)) | |
9398 | + process_membership(); | |
9399 | + } | |
9400 | + | |
bb1d8b11 AM |
9401 | + set_current_state(TASK_INTERRUPTIBLE); |
9402 | + if (!got_work()) | |
9403 | + schedule(); | |
9404 | + set_current_state(TASK_RUNNING); | |
9405 | + } | |
c783755a | 9406 | + |
bb1d8b11 | 9407 | + return 0; |
c783755a AM |
9408 | +} |
9409 | + | |
bb1d8b11 AM |
9410 | +int start_serviced(void) |
9411 | +{ | |
9412 | + struct task_struct *p; | |
c783755a | 9413 | + |
d3b4771f | 9414 | + p = kthread_run(serviced, NULL, 0, "cman_serviced"); |
bb1d8b11 AM |
9415 | + if (IS_ERR(p)) { |
9416 | + printk("can't start cman_serviced daemon"); | |
9417 | + return (IS_ERR(p)); | |
9418 | + } | |
c783755a | 9419 | + |
bb1d8b11 AM |
9420 | + daemon_task = p; |
9421 | + return 0; | |
9422 | +} | |
c783755a | 9423 | + |
bb1d8b11 AM |
9424 | +void stop_serviced(void) |
9425 | +{ | |
9426 | + kthread_stop(daemon_task); | |
9427 | +} | |
9428 | diff -urN linux-orig/cluster/cman/sm_daemon.h linux-patched/cluster/cman/sm_daemon.h | |
9429 | --- linux-orig/cluster/cman/sm_daemon.h 1970-01-01 07:30:00.000000000 +0730 | |
9430 | +++ linux-patched/cluster/cman/sm_daemon.h 2004-11-03 11:37:37.000000000 +0800 | |
9431 | @@ -0,0 +1,32 @@ | |
9432 | +/****************************************************************************** | |
9433 | +******************************************************************************* | |
9434 | +** | |
9435 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9436 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9437 | +** | |
9438 | +** This copyrighted material is made available to anyone wishing to use, | |
9439 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9440 | +** of the GNU General Public License v.2. | |
9441 | +** | |
9442 | +******************************************************************************* | |
9443 | +******************************************************************************/ | |
c783755a | 9444 | + |
bb1d8b11 AM |
9445 | +#ifndef __SM_DAEMON_DOT_H__ |
9446 | +#define __SM_DAEMON_DOT_H__ | |
c783755a | 9447 | + |
bb1d8b11 AM |
9448 | +#define DO_RUN (0) |
9449 | +#define DO_START_RECOVERY (1) | |
9450 | +#define DO_MESSAGES (2) | |
9451 | +#define DO_BARRIERS (3) | |
9452 | +#define DO_CALLBACKS (4) | |
9453 | +#define DO_JOINLEAVE (5) | |
9454 | +#define DO_RECOVERIES (6) | |
9455 | +#define DO_MEMBERSHIP (7) | |
9456 | +#define DO_RESET (8) | |
c783755a | 9457 | + |
bb1d8b11 AM |
9458 | +void init_serviced(void); |
9459 | +void wake_serviced(int do_flag); | |
9460 | +void stop_serviced(void); | |
9461 | +int start_serviced(void); | |
c783755a | 9462 | + |
c783755a | 9463 | +#endif |
c1c6733f | 9464 | diff -urN linux-orig/cluster/cman/sm_internal.h linux-patched/cluster/cman/sm_internal.h |
bb1d8b11 AM |
9465 | --- linux-orig/cluster/cman/sm_internal.h 1970-01-01 07:30:00.000000000 +0730 |
9466 | +++ linux-patched/cluster/cman/sm_internal.h 2004-11-03 11:37:37.000000000 +0800 | |
b7b72b66 | 9467 | @@ -0,0 +1,231 @@ |
c1c6733f AM |
9468 | +/****************************************************************************** |
9469 | +******************************************************************************* | |
9470 | +** | |
9471 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9472 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9473 | +** | |
9474 | +** This copyrighted material is made available to anyone wishing to use, | |
9475 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9476 | +** of the GNU General Public License v.2. | |
9477 | +** | |
9478 | +******************************************************************************* | |
9479 | +******************************************************************************/ | |
9480 | + | |
9481 | +#ifndef __SM_INTERNAL_DOT_H__ | |
9482 | +#define __SM_INTERNAL_DOT_H__ | |
9483 | + | |
9484 | +/* | |
9485 | + * Any header files needed by this file should be included before it in sm.h. | |
9486 | + * This file should only be included by sm.h. | |
9487 | + */ | |
9488 | + | |
9489 | +struct sm_group; | |
9490 | +struct sm_sevent; | |
9491 | +struct sm_uevent; | |
9492 | +struct sm_node; | |
9493 | +struct sm_msg; | |
9494 | + | |
9495 | +typedef struct sm_group sm_group_t; | |
9496 | +typedef struct sm_sevent sm_sevent_t; | |
9497 | +typedef struct sm_uevent sm_uevent_t; | |
9498 | +typedef struct sm_node sm_node_t; | |
9499 | +typedef struct sm_msg sm_msg_t; | |
9500 | + | |
9501 | + | |
9502 | +/* | |
9503 | + * Number of seconds to wait before trying again to join or leave an SG | |
9504 | + */ | |
9505 | +#define RETRY_DELAY (2) | |
9506 | + | |
9507 | + | |
9508 | +/* | |
9509 | + * Service Event - what a node uses to join or leave an sg | |
9510 | + */ | |
9511 | + | |
9512 | +/* SE Flags */ | |
9513 | +#define SEFL_CHECK (0) | |
9514 | +#define SEFL_ALLOW_JOIN (1) | |
9515 | +#define SEFL_ALLOW_JSTOP (2) | |
9516 | +#define SEFL_ALLOW_LEAVE (3) | |
9517 | +#define SEFL_ALLOW_LSTOP (4) | |
9518 | +#define SEFL_ALLOW_STARTDONE (5) | |
9519 | +#define SEFL_ALLOW_BARRIER (6) | |
9520 | +#define SEFL_DELAY (7) | |
b7b72b66 AM |
9521 | +#define SEFL_DELAY_RECOVERY (8) |
9522 | +#define SEFL_LEAVE (9) | |
9523 | +#define SEFL_CANCEL (10) | |
c1c6733f AM |
9524 | + |
9525 | +/* SE States */ | |
9526 | +#define SEST_JOIN_BEGIN (1) | |
9527 | +#define SEST_JOIN_ACKWAIT (2) | |
9528 | +#define SEST_JOIN_ACKED (3) | |
9529 | +#define SEST_JSTOP_ACKWAIT (4) | |
9530 | +#define SEST_JSTOP_ACKED (5) | |
9531 | +#define SEST_JSTART_SERVICEWAIT (6) | |
9532 | +#define SEST_JSTART_SERVICEDONE (7) | |
9533 | +#define SEST_BARRIER_WAIT (8) | |
9534 | +#define SEST_BARRIER_DONE (9) | |
9535 | +#define SEST_LEAVE_BEGIN (10) | |
9536 | +#define SEST_LEAVE_ACKWAIT (11) | |
9537 | +#define SEST_LEAVE_ACKED (12) | |
9538 | +#define SEST_LSTOP_ACKWAIT (13) | |
9539 | +#define SEST_LSTOP_ACKED (14) | |
9540 | +#define SEST_LSTART_WAITREMOTE (15) | |
9541 | +#define SEST_LSTART_REMOTEDONE (16) | |
9542 | + | |
9543 | +struct sm_sevent { | |
9544 | + struct list_head se_list; | |
9545 | + unsigned int se_id; | |
9546 | + sm_group_t * se_sg; | |
9547 | + unsigned long se_flags; | |
9548 | + unsigned int se_state; | |
9549 | + | |
9550 | + int se_node_count; | |
9551 | + int se_memb_count; | |
9552 | + int se_reply_count; | |
9553 | + | |
9554 | + uint32_t * se_node_ids; | |
9555 | + char * se_node_status; | |
9556 | + int se_len_ids; /* length of node_ids */ | |
9557 | + int se_len_status; /* length of node_status */ | |
9558 | + | |
9559 | + int se_barrier_status; | |
9560 | + struct timer_list se_restart_timer; | |
9561 | +}; | |
9562 | + | |
9563 | +/* | |
9564 | + * Update Event - what an sg member uses to respond to an sevent | |
9565 | + */ | |
9566 | + | |
9567 | +/* UE Flags */ | |
9568 | +#define UEFL_ALLOW_STARTDONE (0) | |
9569 | +#define UEFL_ALLOW_BARRIER (1) | |
9570 | +#define UEFL_CANCEL (2) | |
9571 | +#define UEFL_LEAVE (3) | |
9572 | +#define UEFL_CHECK (4) | |
9573 | + | |
9574 | +/* UE States */ | |
9575 | +#define UEST_JSTOP (1) | |
9576 | +#define UEST_JSTART_WAITCMD (2) | |
9577 | +#define UEST_JSTART (3) | |
9578 | +#define UEST_JSTART_SERVICEWAIT (4) | |
9579 | +#define UEST_JSTART_SERVICEDONE (5) | |
9580 | +#define UEST_BARRIER_WAIT (6) | |
9581 | +#define UEST_BARRIER_DONE (7) | |
9582 | +#define UEST_LSTOP (8) | |
9583 | +#define UEST_LSTART_WAITCMD (9) | |
9584 | +#define UEST_LSTART (10) | |
9585 | +#define UEST_LSTART_SERVICEWAIT (11) | |
9586 | +#define UEST_LSTART_SERVICEDONE (12) | |
9587 | + | |
9588 | +struct sm_uevent { | |
9589 | + unsigned int ue_state; | |
9590 | + unsigned long ue_flags; | |
9591 | + uint32_t ue_id; | |
9592 | + uint32_t ue_nodeid; | |
9593 | + int ue_num_nodes; | |
9594 | + int ue_barrier_status; | |
9595 | + uint16_t ue_remote_seid; | |
9596 | +}; | |
9597 | + | |
9598 | +/* | |
9599 | + * Service Group | |
9600 | + */ | |
9601 | + | |
9602 | +#define RECOVER_NONE (0) | |
9603 | +#define RECOVER_STOP (1) | |
9604 | +#define RECOVER_START (2) | |
9605 | +#define RECOVER_STARTDONE (3) | |
9606 | +#define RECOVER_BARRIERWAIT (4) | |
9607 | +#define RECOVER_BARRIERDONE (5) | |
9608 | + | |
9609 | +/* SG Flags */ | |
9610 | +#define SGFL_SEVENT (1) | |
9611 | +#define SGFL_UEVENT (2) | |
9612 | +#define SGFL_NEED_RECOVERY (3) | |
9613 | + | |
9614 | +/* SG States */ | |
9615 | +#define SGST_NONE (0) | |
9616 | +#define SGST_JOIN (1) | |
9617 | +#define SGST_RUN (2) | |
9618 | +#define SGST_RECOVER (3) | |
9619 | +#define SGST_UEVENT (4) | |
9620 | + | |
9621 | +struct sm_group { | |
9622 | + struct list_head list; /* list of sg's */ | |
9623 | + uint16_t level; | |
9624 | + uint32_t local_id; | |
9625 | + uint32_t global_id; | |
9626 | + unsigned long flags; | |
9627 | + int state; | |
9628 | + int refcount; /* references from reg/unreg */ | |
9629 | + void * service_data; /* data from the service */ | |
9630 | + struct kcl_service_ops *ops; /* ops from the service */ | |
9631 | + struct completion event_comp; | |
9632 | + | |
9633 | + struct list_head memb; /* Membership List for RC */ | |
9634 | + int memb_count; /* number of nodes in memb */ | |
9635 | + struct list_head joining; /* nodes joining the sg */ | |
9636 | + sm_sevent_t * sevent; | |
9637 | + sm_uevent_t uevent; | |
9638 | + | |
9639 | + int recover_state; | |
9640 | + int recover_stop; | |
9641 | + struct list_head recover_list; /* recovery event list */ | |
9642 | + void * recover_data; | |
9643 | + char recover_barrier[MAX_BARRIER_NAME_LEN]; | |
9644 | + | |
9645 | + int namelen; | |
9646 | + char name[1]; /* must be last field */ | |
9647 | +}; | |
9648 | + | |
9649 | +/* | |
9650 | + * Service Message | |
9651 | + */ | |
9652 | + | |
9653 | +/* SMSG Type */ | |
9654 | +#define SMSG_JOIN_REQ (1) | |
9655 | +#define SMSG_JOIN_REP (2) | |
9656 | +#define SMSG_JSTOP_REQ (3) | |
9657 | +#define SMSG_JSTOP_REP (4) | |
9658 | +#define SMSG_JSTART_CMD (5) | |
9659 | +#define SMSG_LEAVE_REQ (6) | |
9660 | +#define SMSG_LEAVE_REP (7) | |
9661 | +#define SMSG_LSTOP_REQ (8) | |
9662 | +#define SMSG_LSTOP_REP (9) | |
9663 | +#define SMSG_LSTART_CMD (10) | |
9664 | +#define SMSG_LSTART_DONE (11) | |
9665 | +#define SMSG_RECOVER (12) | |
9666 | + | |
9667 | +/* SMSG Status */ | |
9668 | +#define STATUS_POS (1) | |
9669 | +#define STATUS_NEG (2) | |
9670 | +#define STATUS_WAIT (3) | |
9671 | + | |
9672 | +struct sm_msg { | |
9673 | + uint8_t ms_type; | |
9674 | + uint8_t ms_status; | |
9675 | + uint16_t ms_sevent_id; | |
9676 | + uint32_t ms_global_sgid; | |
9677 | + uint32_t ms_global_lastid; | |
9678 | + uint16_t ms_sglevel; | |
9679 | + uint16_t ms_length; | |
9680 | + /* buf of ms_length bytes follows */ | |
9681 | +}; | |
9682 | + | |
9683 | +/* | |
9684 | + * Node structure | |
9685 | + */ | |
9686 | + | |
9687 | +#define SNFL_NEED_RECOVERY (0) | |
9688 | +#define SNFL_CLUSTER_MEMBER (1) | |
9689 | +#define SNFL_LEAVING (2) | |
9690 | + | |
9691 | +struct sm_node { | |
9692 | + struct list_head list; | |
9693 | + uint32_t id; /* node id from cnxman */ | |
9694 | + unsigned long flags; | |
9695 | + int incarnation; /* node incarnation number */ | |
9696 | +}; | |
9697 | + | |
9698 | +#endif /* __SM_INTERNAL_DOT_H__ */ | |
9699 | diff -urN linux-orig/cluster/cman/sm_joinleave.c linux-patched/cluster/cman/sm_joinleave.c | |
bb1d8b11 AM |
9700 | --- linux-orig/cluster/cman/sm_joinleave.c 1970-01-01 07:30:00.000000000 +0730 |
9701 | +++ linux-patched/cluster/cman/sm_joinleave.c 2004-11-03 11:37:37.000000000 +0800 | |
b7b72b66 | 9702 | @@ -0,0 +1,1291 @@ |
c1c6733f AM |
9703 | +/****************************************************************************** |
9704 | +******************************************************************************* | |
9705 | +** | |
9706 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9707 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9708 | +** | |
9709 | +** This copyrighted material is made available to anyone wishing to use, | |
9710 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9711 | +** of the GNU General Public License v.2. | |
9712 | +** | |
9713 | +******************************************************************************* | |
9714 | +******************************************************************************/ | |
9715 | + | |
9716 | +#include "sm.h" | |
9717 | + | |
9718 | +/* | |
9719 | + * Routines used by nodes that are joining or leaving a SG. These "sevent" | |
9720 | + * routines initiate membership changes to a SG. Existing SG members respond | |
9721 | + * using the "uevent" membership update routines. | |
9722 | + */ | |
9723 | + | |
9724 | +extern uint32_t sm_our_nodeid; | |
9725 | +extern struct list_head sm_members; | |
9726 | +static struct list_head new_event; | |
9727 | +static spinlock_t new_event_lock; | |
9728 | +static struct list_head joinleave_events; | |
9729 | + | |
9730 | +void init_joinleave(void) | |
9731 | +{ | |
9732 | + INIT_LIST_HEAD(&new_event); | |
9733 | + spin_lock_init(&new_event_lock); | |
9734 | + INIT_LIST_HEAD(&joinleave_events); | |
9735 | +} | |
9736 | + | |
9737 | +void new_joinleave(sm_sevent_t *sev) | |
9738 | +{ | |
9739 | + spin_lock(&new_event_lock); | |
9740 | + list_add_tail(&sev->se_list, &new_event); | |
9741 | + spin_unlock(&new_event_lock); | |
9742 | + wake_serviced(DO_JOINLEAVE); | |
9743 | +} | |
9744 | + | |
9745 | +sm_sevent_t *find_sevent(unsigned int id) | |
9746 | +{ | |
9747 | + sm_sevent_t *sev; | |
9748 | + | |
9749 | + list_for_each_entry(sev, &joinleave_events, se_list) { | |
9750 | + if (sev->se_id == id) | |
9751 | + return sev; | |
9752 | + } | |
9753 | + return NULL; | |
9754 | +} | |
9755 | + | |
9756 | +static void release_sevent(sm_sevent_t *sev) | |
9757 | +{ | |
9758 | + if (sev->se_len_ids) { | |
9759 | + kfree(sev->se_node_ids); | |
9760 | + sev->se_node_ids = NULL; | |
9761 | + } | |
9762 | + | |
9763 | + if (sev->se_len_status) { | |
9764 | + kfree(sev->se_node_status); | |
9765 | + sev->se_node_status = NULL; | |
9766 | + } | |
9767 | + | |
9768 | + sev->se_node_count = 0; | |
9769 | + sev->se_memb_count = 0; | |
9770 | + sev->se_reply_count = 0; | |
9771 | +} | |
9772 | + | |
9773 | +static int init_sevent(sm_sevent_t *sev) | |
9774 | +{ | |
9775 | + sm_node_t *node; | |
9776 | + int len1, len2, count, cluster_members = 0; | |
9777 | + | |
9778 | + /* clear state from any previous attempt */ | |
9779 | + release_sevent(sev); | |
9780 | + | |
9781 | + list_for_each_entry(node, &sm_members, list) { | |
9782 | + if (test_bit(SNFL_CLUSTER_MEMBER, &node->flags)) | |
9783 | + cluster_members++; | |
9784 | + } | |
9785 | + | |
9786 | + sev->se_node_count = cluster_members; | |
9787 | + sev->se_memb_count = sev->se_sg->memb_count; | |
9788 | + | |
9789 | + /* | |
9790 | + * When joining, we need a node array the size of the entire cluster | |
9791 | + * member list because we get responses from all nodes. When leaving, | |
9792 | + * we only get responses from SG members, so the node array need only | |
9793 | + * be that large. | |
9794 | + */ | |
9795 | + | |
9796 | + if (sev->se_state < SEST_LEAVE_BEGIN) | |
9797 | + count = sev->se_node_count; | |
9798 | + else | |
9799 | + count = sev->se_memb_count; | |
9800 | + | |
9801 | + len1 = count * sizeof(uint32_t); | |
9802 | + sev->se_len_ids = len1; | |
9803 | + | |
9804 | + sev->se_node_ids = (uint32_t *) kmalloc(len1, GFP_KERNEL); | |
9805 | + if (!sev->se_node_ids) | |
9806 | + goto fail; | |
9807 | + | |
9808 | + len2 = count * sizeof (char); | |
9809 | + sev->se_len_status = len2; | |
9810 | + | |
9811 | + sev->se_node_status = (char *) kmalloc(len2, GFP_KERNEL); | |
9812 | + if (!sev->se_node_status) | |
9813 | + goto fail_free; | |
9814 | + | |
9815 | + memset(sev->se_node_status, 0, len2); | |
9816 | + memset(sev->se_node_ids, 0, len1); | |
9817 | + | |
9818 | + return 0; | |
9819 | + | |
9820 | + fail_free: | |
9821 | + kfree(sev->se_node_ids); | |
9822 | + sev->se_node_ids = NULL; | |
9823 | + sev->se_len_ids = 0; | |
9824 | + | |
9825 | + fail: | |
9826 | + return -ENOMEM; | |
9827 | +} | |
9828 | + | |
9829 | +/* Context: timer */ | |
9830 | + | |
9831 | +static void sev_restart(unsigned long data) | |
9832 | +{ | |
9833 | + sm_sevent_t *sev = (sm_sevent_t *) data; | |
9834 | + | |
9835 | + clear_bit(SEFL_DELAY, &sev->se_flags); | |
9836 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
9837 | + wake_serviced(DO_JOINLEAVE); | |
9838 | +} | |
9839 | + | |
9840 | +static void schedule_sev_restart(sm_sevent_t *sev) | |
9841 | +{ | |
9842 | + init_timer(&sev->se_restart_timer); | |
9843 | + sev->se_restart_timer.function = sev_restart; | |
9844 | + sev->se_restart_timer.data = (long) sev; | |
9845 | + mod_timer(&sev->se_restart_timer, jiffies + (RETRY_DELAY * HZ)); | |
9846 | +} | |
9847 | + | |
9848 | +void free_sg_memb(sm_group_t *sg) | |
9849 | +{ | |
9850 | + sm_node_t *node; | |
9851 | + | |
9852 | + while (!list_empty(&sg->memb)) { | |
9853 | + node = list_entry(sg->memb.next, sm_node_t, list); | |
9854 | + list_del(&node->list); | |
9855 | + kfree(node); | |
9856 | + } | |
9857 | + sg->memb_count = 0; | |
9858 | +} | |
9859 | + | |
9860 | +/* | |
9861 | + * 1. First step in joining a SG - send a message to all nodes in the cluster | |
9862 | + * asking to join the named SG. If any nodes are members they will reply with | |
9863 | + * a POS, or a WAIT (wait means try again, only one node can join at a time). | |
9864 | + * If no one knows about this SG, they all send NEG replies which means we form | |
9865 | + * the SG with just ourself as a member. | |
9866 | + */ | |
9867 | + | |
9868 | +static int send_join_notice(sm_sevent_t *sev) | |
9869 | +{ | |
9870 | + sm_group_t *sg = sev->se_sg; | |
9871 | + sm_node_t *node; | |
9872 | + char *msg; | |
9873 | + int i = 0, error, namelen, len = 0; | |
9874 | + | |
9875 | + /* | |
9876 | + * Create node array from member list in which to collect responses. | |
9877 | + */ | |
9878 | + | |
9879 | + error = init_sevent(sev); | |
9880 | + if (error) | |
9881 | + goto out; | |
9882 | + | |
9883 | + list_for_each_entry(node, &sm_members, list) { | |
9884 | + if (test_bit(SNFL_CLUSTER_MEMBER, &node->flags)) | |
9885 | + sev->se_node_ids[i++] = node->id; | |
9886 | + } | |
9887 | + | |
9888 | + /* | |
9889 | + * Create and send a join request message. | |
9890 | + * | |
9891 | + * Other nodes then run process_join_request and reply to us; we | |
9892 | + * collect the responses in process_reply and check them in | |
9893 | + * check_join_notice. | |
9894 | + */ | |
9895 | + | |
9896 | + namelen = sg->namelen; | |
9897 | + msg = create_smsg(sg, SMSG_JOIN_REQ, namelen, &len, sev); | |
9898 | + memcpy(msg + sizeof(sm_msg_t), sg->name, namelen); | |
9899 | + | |
9900 | + error = send_broadcast_message_sev(msg, len, sev); | |
9901 | + | |
9902 | + out: | |
9903 | + return error; | |
9904 | +} | |
9905 | + | |
9906 | +/* | |
9907 | + * 2. Second step in joining a SG - after we collect all replies to our join | |
9908 | + * request, we look at them. If anyone told us to wait, we'll wait a while, go | |
9909 | + * back and start at step 1 again. | |
9910 | + */ | |
9911 | + | |
9912 | +static int check_join_notice(sm_sevent_t *sev) | |
9913 | +{ | |
9914 | + int pos = 0, wait = 0, neg = 0, restart = 0, i, error = 0; | |
9915 | + | |
9916 | + for (i = 0; i < sev->se_node_count; i++) { | |
9917 | + switch (sev->se_node_status[i]) { | |
9918 | + case STATUS_POS: | |
9919 | + /* this node is in the SG and will be in new proposed | |
9920 | + * memb list */ | |
9921 | + pos++; | |
9922 | + break; | |
9923 | + | |
9924 | + case STATUS_WAIT: | |
9925 | + /* this node is in the SG but something else is | |
9926 | + * happening with it at the moment. */ | |
9927 | + wait++; | |
9928 | + break; | |
9929 | + | |
9930 | + case STATUS_NEG: | |
9931 | + /* this node has no record of the SG we're interested | |
9932 | + * in */ | |
9933 | + neg++; | |
9934 | + | |
9935 | + if (sev->se_node_ids[i] == sm_our_nodeid) | |
9936 | + sev->se_node_status[i] = STATUS_POS; | |
9937 | + break; | |
9938 | + | |
9939 | + default: | |
9940 | + /* we didn't get a valid response from this node, | |
9941 | + * restart the entire sev. */ | |
9942 | + restart++; | |
9943 | + break; | |
9944 | + } | |
9945 | + } | |
9946 | + | |
9947 | + if (pos && !wait && !restart) { | |
9948 | + /* all current members of this sg pos'ed our entry */ | |
9949 | + } else if (!pos && !wait && !restart && neg) { | |
9950 | + /* we're the first in the cluster to join this sg */ | |
9951 | + sev->se_sg->global_id = sm_new_global_id(sev->se_sg->level); | |
9952 | + } else | |
9953 | + error = -1; | |
9954 | + | |
9955 | + return error; | |
9956 | +} | |
9957 | + | |
9958 | +/* | |
9959 | + * 3. Third step in joining the SG - tell the nodes that are already members | |
9960 | + * to "stop" the service. We stop them so that everyone can restart with the | |
9961 | + * new member (us!) added. | |
9962 | + */ | |
9963 | + | |
9964 | +static int send_join_stop(sm_sevent_t *sev) | |
9965 | +{ | |
9966 | + sm_group_t *sg = sev->se_sg; | |
9967 | + sm_node_t *node; | |
9968 | + char *msg; | |
9969 | + uint32_t be_count; | |
9970 | + int i, len = 0, error = 0; | |
9971 | + | |
9972 | + /* | |
9973 | + * Form the SG memb list with us in it. | |
9974 | + */ | |
9975 | + | |
9976 | + for (i = 0; i < sev->se_node_count; i++) { | |
9977 | + if (sev->se_node_status[i] != STATUS_POS) | |
9978 | + continue; | |
9979 | + | |
9980 | + node = sm_new_node(sev->se_node_ids[i]); | |
9981 | + if (!node) | |
9982 | + goto fail; | |
9983 | + | |
9984 | + list_add_tail(&node->list, &sg->memb); | |
9985 | + sg->memb_count++; | |
9986 | + } | |
9987 | + | |
9988 | + /* | |
9989 | + * Re-init the node vector in which to collect responses again. | |
9990 | + */ | |
9991 | + | |
9992 | + sev->se_memb_count = sg->memb_count; | |
9993 | + | |
9994 | + memset(sev->se_node_status, 0, sev->se_len_status); | |
9995 | + memset(sev->se_node_ids, 0, sev->se_len_ids); | |
9996 | + i = 0; | |
9997 | + | |
9998 | + list_for_each_entry(node, &sg->memb, list) | |
9999 | + sev->se_node_ids[i++] = node->id; | |
10000 | + | |
10001 | + /* | |
10002 | + * Create and send a stop message. | |
10003 | + * | |
10004 | + * Other nodes then run process_stop_request and process_join_stop and | |
10005 | + * reply to us. They stop the sg we're trying to join if they agree. | |
10006 | + * We collect responses in process_reply and check them in | |
10007 | + * check_join_stop. | |
10008 | + */ | |
10009 | + | |
10010 | + msg = create_smsg(sg, SMSG_JSTOP_REQ, sizeof(uint32_t), &len, sev); | |
10011 | + be_count = cpu_to_be32(sg->memb_count); | |
10012 | + memcpy(msg + sizeof(sm_msg_t), &be_count, sizeof(uint32_t)); | |
10013 | + | |
10014 | + error = send_members_message_sev(sg, msg, len, sev); | |
10015 | + if (error < 0) | |
10016 | + goto fail; | |
10017 | + | |
10018 | + return 0; | |
10019 | + | |
10020 | + fail: | |
10021 | + free_sg_memb(sg); | |
10022 | + return error; | |
10023 | +} | |
10024 | + | |
10025 | +/* | |
10026 | + * 4. Fourth step in joining the SG - after we collect replies to our stop | |
10027 | + * request, we look at them. Everyone sending POS agrees with us joining and | |
10028 | + * has stopped their SG. If some nodes sent NEG, something is wrong and we | |
10029 | + * don't have a good way to address that yet since some nodes may have sent | |
10030 | + * POS. | |
10031 | + * | |
10032 | + * FIXME: even nodes replying with NEG should stop their SG so we can send an | |
10033 | + * abort and have everyone at the same place to start from again. | |
10034 | + */ | |
10035 | + | |
10036 | +static int check_join_stop(sm_sevent_t *sev) | |
10037 | +{ | |
10038 | + sm_group_t *sg = sev->se_sg; | |
10039 | + int i, pos = 0, neg = 0; | |
10040 | + | |
10041 | + for (i = 0; i < sev->se_memb_count; i++) { | |
10042 | + switch (sev->se_node_status[i]) { | |
10043 | + case STATUS_POS: | |
10044 | + pos++; | |
10045 | + break; | |
10046 | + | |
10047 | + case STATUS_NEG: | |
10048 | + log_error(sg, "check_join_stop: neg from nodeid %u " | |
10049 | + "(%d, %d, %u)", sev->se_node_ids[i], | |
10050 | + pos, neg, sev->se_memb_count); | |
10051 | + neg++; | |
10052 | + break; | |
10053 | + | |
10054 | + default: | |
10055 | + log_error(sg, "check_join_stop: unknown status=%u " | |
10056 | + "nodeid=%u", sev->se_node_status[i], | |
10057 | + sev->se_node_ids[i]); | |
10058 | + neg++; | |
10059 | + break; | |
10060 | + } | |
10061 | + } | |
10062 | + | |
10063 | + if (pos == sg->memb_count) | |
10064 | + return 0; | |
10065 | + | |
10066 | + free_sg_memb(sg); | |
10067 | + return -1; | |
10068 | +} | |
10069 | + | |
10070 | +/* | |
10071 | + * 5. Fifth step in joining the SG - everyone has stopped their service and we | |
10072 | + * all now start the service with us, the new member, added to the SG member | |
10073 | + * list. We send start to our own service here and send a message to the other | |
10074 | + * members that they should also start their service. | |
10075 | + */ | |
10076 | + | |
10077 | +static int send_join_start(sm_sevent_t *sev) | |
10078 | +{ | |
10079 | + sm_group_t *sg = sev->se_sg; | |
10080 | + sm_node_t *node; | |
10081 | + uint32_t *memb; | |
10082 | + char *msg; | |
10083 | + int error, count = 0, len = 0; | |
10084 | + | |
10085 | + /* | |
10086 | + * Create a start message and send it. | |
10087 | + */ | |
10088 | + | |
10089 | + msg = create_smsg(sg, SMSG_JSTART_CMD, 0, &len, sev); | |
10090 | + | |
10091 | + error = send_members_message(sg, msg, len); | |
10092 | + if (error < 0) | |
10093 | + goto fail; | |
10094 | + | |
10095 | + /* | |
10096 | + * Start the service ourself. The chunk of memory with the member ids | |
10097 | + * must be freed by the service when it is done with it. | |
10098 | + */ | |
10099 | + | |
10100 | + SM_RETRY(memb = kmalloc(sg->memb_count * sizeof(uint32_t), GFP_KERNEL), | |
10101 | + memb); | |
10102 | + | |
10103 | + list_for_each_entry(node, &sg->memb, list) | |
10104 | + memb[count++] = node->id; | |
10105 | + | |
10106 | + set_bit(SEFL_ALLOW_STARTDONE, &sev->se_flags); | |
10107 | + | |
10108 | + sg->ops->start(sg->service_data, memb, count, sev->se_id, | |
10109 | + SERVICE_NODE_JOIN); | |
10110 | + return 0; | |
10111 | + | |
10112 | + fail: | |
10113 | + free_sg_memb(sg); | |
10114 | + return error; | |
10115 | +} | |
10116 | + | |
10117 | +/* | |
10118 | + * 6. Sixth step in joining the SG - once the service has completed its start, | |
10119 | + * it does a kcl_start_done() to signal us that it's done. That gets us here | |
10120 | + * and we do a barrier with all other members which join the barrier when their | |
10121 | + * service is done starting. | |
10122 | + */ | |
10123 | + | |
10124 | +static int startdone_barrier_new(sm_sevent_t *sev) | |
10125 | +{ | |
10126 | + sm_group_t *sg = sev->se_sg; | |
10127 | + char bname[MAX_BARRIER_NAME_LEN]; | |
10128 | + int error; | |
10129 | + | |
10130 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
10131 | + sev->se_barrier_status = -1; | |
10132 | + | |
10133 | + set_bit(SEFL_ALLOW_BARRIER, &sev->se_flags); | |
10134 | + | |
10135 | + /* If we're the only member, skip the barrier */ | |
10136 | + if (sg->memb_count == 1) { | |
10137 | + process_startdone_barrier_new(sg, 0); | |
10138 | + return 0; | |
10139 | + } | |
10140 | + | |
10141 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
10142 | + sg->global_id, sm_our_nodeid, sev->se_id, sg->memb_count); | |
10143 | + | |
10144 | + error = sm_barrier(bname, sg->memb_count, SM_BARRIER_STARTDONE_NEW); | |
10145 | + if (error) | |
10146 | + goto fail; | |
10147 | + | |
10148 | + return 0; | |
10149 | + | |
10150 | + fail: | |
10151 | + clear_bit(SEFL_ALLOW_BARRIER, &sev->se_flags); | |
10152 | + sg->ops->stop(sg->service_data); | |
10153 | + free_sg_memb(sg); | |
10154 | + return error; | |
10155 | +} | |
10156 | + | |
10157 | +/* | |
10158 | + * 7. Seventh step in joining the SG - check that the barrier we joined with | |
10159 | + * all other members returned with a successful status. | |
10160 | + */ | |
10161 | + | |
10162 | +static int check_startdone_barrier_new(sm_sevent_t *sev) | |
10163 | +{ | |
10164 | + sm_group_t *sg = sev->se_sg; | |
10165 | + int error = sev->se_barrier_status; | |
10166 | + | |
10167 | + if (error) { | |
10168 | + sg->ops->stop(sg->service_data); | |
10169 | + free_sg_memb(sg); | |
10170 | + } | |
10171 | + return error; | |
10172 | +} | |
10173 | + | |
10174 | +/* | |
10175 | + * 8. Eigth step in joining the SG - send the service a "finish" indicating | |
10176 | + * that all members have successfully started the service. | |
10177 | + */ | |
10178 | + | |
10179 | +static void do_finish_new(sm_sevent_t *sev) | |
10180 | +{ | |
10181 | + sm_group_t *sg = sev->se_sg; | |
10182 | + | |
10183 | + sg->state = SGST_RUN; | |
10184 | + sg->sevent = NULL; | |
10185 | + clear_bit(SGFL_SEVENT, &sg->flags); | |
10186 | + | |
10187 | + sg->ops->finish(sg->service_data, sev->se_id); | |
10188 | +} | |
10189 | + | |
10190 | +/* | |
10191 | + * 9. Ninth step in joining the SG - it's done so get rid of the sevent stuff | |
10192 | + * and tell the process which initiated the join that it's done. | |
10193 | + */ | |
10194 | + | |
10195 | +static void sevent_done(sm_sevent_t *sev) | |
10196 | +{ | |
10197 | + sm_group_t *sg = sev->se_sg; | |
10198 | + | |
10199 | + list_del(&sev->se_list); | |
10200 | + release_sevent(sev); | |
10201 | + kfree(sev); | |
10202 | + complete(&sg->event_comp); | |
10203 | +} | |
10204 | + | |
10205 | +/* | |
10206 | + * Move through the steps of a join. Summary: | |
10207 | + * | |
10208 | + * 1. Send a join notice to all cluster members. | |
10209 | + * 2. Collect and check replies to the join notice. | |
10210 | + * 3. Send a stop message to all SG members. | |
10211 | + * 4. Collect and check replies to the stop message. | |
10212 | + * 5. Send a start message to all SG members and start service ourself. | |
10213 | + * 6. Use barrier to wait for all nodes to complete the start. | |
10214 | + * 7. Check that all SG members joined the barrier. | |
10215 | + * 8. Send finish to the service indicating that all nodes started it. | |
10216 | + * 9. Clean up sevent and signal completion to the process that started the join | |
10217 | + */ | |
10218 | + | |
10219 | +static void process_join_sevent(sm_sevent_t *sev) | |
10220 | +{ | |
10221 | + int error = 0; | |
10222 | + | |
10223 | + /* | |
10224 | + * We may cancel the current join attempt if another node is also | |
10225 | + * attempting to join or leave. (Only a single node can join or leave | |
10226 | + * at once.) If cancelled, 0ur join attempt will be restarted later. | |
10227 | + */ | |
10228 | + | |
10229 | + if (test_and_clear_bit(SEFL_CANCEL, &sev->se_flags)) { | |
b7b72b66 | 10230 | + error = 1; |
c1c6733f AM |
10231 | + goto cancel; |
10232 | + } | |
10233 | + | |
10234 | + log_debug(sev->se_sg, "sevent state %u", sev->se_state); | |
10235 | + | |
10236 | + switch (sev->se_state) { | |
10237 | + | |
10238 | + /* | |
10239 | + * An sevent is created in kcl_join_service with a state of | |
10240 | + * JOIN_BEGIN. | |
10241 | + */ | |
10242 | + | |
10243 | + case SEST_JOIN_BEGIN: | |
10244 | + sev->se_state = SEST_JOIN_ACKWAIT; | |
10245 | + error = send_join_notice(sev); | |
10246 | + break; | |
10247 | + | |
10248 | + /* | |
10249 | + * se_state is changed from JOIN_ACKWAIT to JOIN_ACKED in | |
10250 | + * process_reply (when all the replies have been received) | |
10251 | + */ | |
10252 | + | |
10253 | + case SEST_JOIN_ACKED: | |
10254 | + error = check_join_notice(sev); | |
10255 | + if (error) | |
10256 | + break; | |
10257 | + | |
10258 | + sev->se_state = SEST_JSTOP_ACKWAIT; | |
10259 | + error = send_join_stop(sev); | |
10260 | + break; | |
10261 | + | |
10262 | + /* | |
10263 | + * se_state is changed from JSTOP_ACKWAIT to JSTOP_ACKED in | |
10264 | + * proces_reply (when all the replies have been received) | |
10265 | + */ | |
10266 | + | |
10267 | + case SEST_JSTOP_ACKED: | |
10268 | + error = check_join_stop(sev); | |
10269 | + if (error) | |
10270 | + break; | |
10271 | + | |
10272 | + sev->se_state = SEST_JSTART_SERVICEWAIT; | |
10273 | + error = send_join_start(sev); | |
10274 | + break; | |
10275 | + | |
10276 | + /* | |
10277 | + * se_state is changed from JSTART_SERVICEWAIT to | |
10278 | + * JSTART_SERVICEDONE in kcl_start_done | |
10279 | + */ | |
10280 | + | |
10281 | + case SEST_JSTART_SERVICEDONE: | |
10282 | + sev->se_state = SEST_BARRIER_WAIT; | |
10283 | + error = startdone_barrier_new(sev); | |
10284 | + break; | |
10285 | + | |
10286 | + /* | |
10287 | + * se_state is changed from BARRIER_WAIT to BARRIER_DONE in | |
10288 | + * process_startdone_barrier_new | |
10289 | + */ | |
10290 | + | |
10291 | + case SEST_BARRIER_DONE: | |
10292 | + error = check_startdone_barrier_new(sev); | |
10293 | + if (error) | |
10294 | + break; | |
10295 | + | |
10296 | + do_finish_new(sev); | |
10297 | + sevent_done(sev); | |
10298 | + break; | |
10299 | + | |
10300 | + default: | |
10301 | + log_error(sev->se_sg, "no join processing for state %u", | |
10302 | + sev->se_state); | |
10303 | + } | |
10304 | + | |
10305 | + cancel: | |
10306 | + if (error) { | |
10307 | + /* restart the sevent from the beginning */ | |
b7b72b66 AM |
10308 | + log_debug(sev->se_sg, "process_join error %d %lx", error, |
10309 | + sev->se_flags); | |
c1c6733f AM |
10310 | + sev->se_state = SEST_JOIN_BEGIN; |
10311 | + sev->se_sg->global_id = 0; | |
10312 | + set_bit(SEFL_DELAY, &sev->se_flags); | |
10313 | + schedule_sev_restart(sev); | |
10314 | + } | |
10315 | +} | |
10316 | + | |
10317 | +/* | |
10318 | + * 1. First step in leaving an SG - send a message to other SG members asking | |
10319 | + * to leave the SG. Nodes that don't have another active sevent or uevent for | |
10320 | + * this SG will return POS. | |
10321 | + */ | |
10322 | + | |
10323 | +static int send_leave_notice(sm_sevent_t *sev) | |
10324 | +{ | |
10325 | + sm_group_t *sg = sev->se_sg; | |
10326 | + sm_node_t *node; | |
10327 | + char *msg; | |
10328 | + int i = 0, error = -1, len = 0; | |
10329 | + | |
10330 | + /* | |
10331 | + * Create a node array from member list in which to collect responses. | |
10332 | + */ | |
10333 | + | |
10334 | + error = init_sevent(sev); | |
10335 | + if (error) | |
10336 | + goto out; | |
10337 | + | |
10338 | + list_for_each_entry(node, &sg->memb, list) | |
10339 | + sev->se_node_ids[i++] = node->id; | |
10340 | + | |
10341 | + /* | |
10342 | + * Create and send a leave request message. | |
10343 | + */ | |
10344 | + | |
10345 | + msg = create_smsg(sg, SMSG_LEAVE_REQ, 0, &len, sev); | |
10346 | + | |
10347 | + error = send_members_message_sev(sg, msg, len, sev); | |
10348 | + | |
10349 | + out: | |
10350 | + return error; | |
10351 | +} | |
10352 | + | |
10353 | +/* | |
10354 | + * 2. Second step in leaving an SG - after we collect all replies to our leave | |
10355 | + * request, we look at them. If anyone replied with WAIT, we abort our attempt | |
10356 | + * at leaving and try again in a bit. | |
10357 | + */ | |
10358 | + | |
10359 | +static int check_leave_notice(sm_sevent_t *sev) | |
10360 | +{ | |
10361 | + int pos = 0, wait = 0, neg = 0, restart = 0, i; | |
10362 | + | |
10363 | + for (i = 0; i < sev->se_memb_count; i++) { | |
10364 | + switch (sev->se_node_status[i]) { | |
10365 | + case STATUS_POS: | |
10366 | + pos++; | |
10367 | + break; | |
10368 | + | |
10369 | + case STATUS_WAIT: | |
10370 | + wait++; | |
10371 | + break; | |
10372 | + | |
10373 | + case STATUS_NEG: | |
10374 | + neg++; | |
10375 | + break; | |
10376 | + | |
10377 | + default: | |
10378 | + /* we didn't get a valid response from this node, | |
10379 | + * restart the entire sev. */ | |
10380 | + restart++; | |
10381 | + break; | |
10382 | + } | |
10383 | + } | |
10384 | + | |
10385 | + /* all members approve */ | |
10386 | + if (pos && !wait && !restart) | |
10387 | + return 0; | |
10388 | + | |
10389 | + return -1; | |
10390 | +} | |
10391 | + | |
10392 | +/* | |
10393 | + * 3. Third step in leaving the SG - tell the member nodes to "stop" the SG. | |
10394 | + * They must be stopped in order to restart without us as a member. | |
10395 | + */ | |
10396 | + | |
10397 | +static int send_leave_stop(sm_sevent_t *sev) | |
10398 | +{ | |
10399 | + sm_group_t *sg = sev->se_sg; | |
10400 | + char *msg; | |
10401 | + int error, len = 0; | |
10402 | + | |
10403 | + /* | |
10404 | + * Re-init the status vector in which to collect responses. | |
10405 | + */ | |
10406 | + | |
10407 | + memset(sev->se_node_status, 0, sev->se_len_status); | |
10408 | + | |
10409 | + /* | |
10410 | + * Create and send a stop message. | |
10411 | + */ | |
10412 | + | |
10413 | + msg = create_smsg(sg, SMSG_LSTOP_REQ, 0, &len, sev); | |
10414 | + | |
10415 | + error = send_members_message_sev(sg, msg, len, sev); | |
10416 | + if (error < 0) | |
10417 | + goto out; | |
10418 | + | |
10419 | + /* | |
10420 | + * we and all others stop the SG now | |
10421 | + */ | |
10422 | + | |
10423 | + sg->ops->stop(sg->service_data); | |
10424 | + | |
10425 | + out: | |
10426 | + return error; | |
10427 | +} | |
10428 | + | |
10429 | +/* | |
10430 | + * 4. Fourth step in leaving the SG - check the replies to our stop request. | |
10431 | + * Same problem with getting different replies as check_join_stop. | |
10432 | + */ | |
10433 | + | |
10434 | +static int check_leave_stop(sm_sevent_t *sev) | |
10435 | +{ | |
10436 | + sm_group_t *sg = sev->se_sg; | |
10437 | + int i, pos = 0, neg = 0; | |
10438 | + | |
10439 | + for (i = 0; i < sev->se_memb_count; i++) { | |
10440 | + switch (sev->se_node_status[i]) { | |
10441 | + case STATUS_POS: | |
10442 | + pos++; | |
10443 | + break; | |
10444 | + | |
10445 | + case STATUS_NEG: | |
10446 | + log_error(sg, "check_leave_stop: fail from nodeid %u " | |
10447 | + "(%d, %d, %u)", sev->se_node_ids[i], | |
10448 | + pos, neg, sev->se_memb_count); | |
10449 | + neg++; | |
10450 | + break; | |
10451 | + | |
10452 | + default: | |
10453 | + log_error(sg, "check_leave_stop: status %u nodeid %u", | |
10454 | + sev->se_node_status[i], sev->se_node_ids[i]); | |
10455 | + neg++; | |
10456 | + break; | |
10457 | + } | |
10458 | + } | |
10459 | + | |
10460 | + if (pos == sg->memb_count) | |
10461 | + return 0; | |
10462 | + | |
10463 | + return -1; | |
10464 | +} | |
10465 | + | |
10466 | +/* | |
10467 | + * 5. Fifth step in leaving the SG - tell the other SG members to restart the | |
10468 | + * service without us. We, of course, don't start our own stopped service. If | |
10469 | + * we're the last SG member and leaving, we jump right to the next step. | |
10470 | + */ | |
10471 | + | |
10472 | +static int send_leave_start(sm_sevent_t *sev) | |
10473 | +{ | |
10474 | + sm_group_t *sg = sev->se_sg; | |
10475 | + char *msg; | |
10476 | + int error = 0, len = 0; | |
10477 | + | |
10478 | + if (sg->memb_count == 1) { | |
10479 | + sev->se_state = SEST_LSTART_REMOTEDONE; | |
10480 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
10481 | + wake_serviced(DO_JOINLEAVE); | |
10482 | + } else { | |
10483 | + msg = create_smsg(sg, SMSG_LSTART_CMD, 0, &len, sev); | |
10484 | + error = send_members_message(sg, msg, len); | |
10485 | + } | |
10486 | + return error; | |
10487 | +} | |
10488 | + | |
10489 | +/* | |
10490 | + * Move through the steps of a leave. Summary: | |
10491 | + * | |
10492 | + * 1. Send a leave notice to all SG members. | |
10493 | + * 2. Collect and check replies to the leave notice. | |
10494 | + * 3. Send a stop message to all SG members and stop our own SG. | |
10495 | + * 4. Collect and check replies to the stop message. | |
10496 | + * 5. Send a start message to SG members. | |
10497 | + * 6. Clean up sevent and signal completion to the process that | |
10498 | + * started the leave. | |
10499 | + */ | |
10500 | + | |
10501 | +static void process_leave_sevent(sm_sevent_t *sev) | |
10502 | +{ | |
10503 | + int error = 0; | |
10504 | + | |
10505 | + /* | |
10506 | + * We may cancel the current leave attempt if another node is also | |
10507 | + * attempting to join or leave. (Only a single node can join or leave | |
10508 | + * at once.) Our leave attempt will be restarted after being | |
10509 | + * cancelled. | |
10510 | + */ | |
10511 | + | |
10512 | + if (test_and_clear_bit(SEFL_CANCEL, &sev->se_flags)) { | |
10513 | + error = 1; | |
10514 | + goto cancel; | |
10515 | + } | |
10516 | + | |
10517 | + if (test_bit(SGFL_UEVENT, &sev->se_sg->flags)) { | |
10518 | + error = 2; | |
10519 | + goto cancel; | |
10520 | + } | |
10521 | + | |
10522 | + if (!list_empty(&sev->se_sg->joining)) { | |
10523 | + error = 3; | |
10524 | + goto cancel; | |
10525 | + } | |
10526 | + | |
10527 | + log_debug(sev->se_sg, "sevent state %u", sev->se_state); | |
10528 | + | |
10529 | + switch (sev->se_state) { | |
10530 | + | |
10531 | + /* | |
10532 | + * An sevent is created in kcl_leave_service with a state of | |
10533 | + * LEAVE_BEGIN. | |
10534 | + */ | |
10535 | + | |
10536 | + case SEST_LEAVE_BEGIN: | |
10537 | + sev->se_state = SEST_LEAVE_ACKWAIT; | |
10538 | + error = send_leave_notice(sev); | |
10539 | + break; | |
10540 | + | |
10541 | + /* | |
10542 | + * se_state is changed from LEAVE_ACKWAIT to LEAVE_ACKED in | |
10543 | + * process_reply (when all the replies have been received) | |
10544 | + */ | |
10545 | + | |
10546 | + case SEST_LEAVE_ACKED: | |
10547 | + error = check_leave_notice(sev); | |
10548 | + if (error) | |
10549 | + break; | |
10550 | + | |
10551 | + sev->se_state = SEST_LSTOP_ACKWAIT; | |
10552 | + error = send_leave_stop(sev); | |
10553 | + break; | |
10554 | + | |
10555 | + /* | |
10556 | + * se_state is changed from LSTOP_ACKWAIT to LSTOP_ACKED in | |
10557 | + * process_reply | |
10558 | + */ | |
10559 | + | |
10560 | + case SEST_LSTOP_ACKED: | |
10561 | + error = check_leave_stop(sev); | |
10562 | + if (error) | |
10563 | + break; | |
10564 | + | |
10565 | + sev->se_state = SEST_LSTART_WAITREMOTE; | |
10566 | + error = send_leave_start(sev); | |
10567 | + break; | |
10568 | + | |
10569 | + /* | |
10570 | + * se_state is changed from LSTART_WAITREMOTE to | |
10571 | + * LSTART_REMOTEDONE in process_leave_done | |
10572 | + */ | |
10573 | + | |
10574 | + case SEST_LSTART_REMOTEDONE: | |
10575 | + sevent_done(sev); | |
10576 | + break; | |
10577 | + | |
10578 | + default: | |
b7b72b66 | 10579 | + log_error(sev->se_sg, "process_leave_sevent state=%u", |
c1c6733f AM |
10580 | + sev->se_state); |
10581 | + } | |
10582 | + | |
b7b72b66 | 10583 | + cancel: |
c1c6733f | 10584 | + if (error) { |
b7b72b66 AM |
10585 | + log_debug(sev->se_sg, "process_leave error %d %lx", error, |
10586 | + sev->se_flags); | |
c1c6733f AM |
10587 | + /* restart the sevent from the beginning */ |
10588 | + sev->se_state = SEST_LEAVE_BEGIN; | |
10589 | + set_bit(SEFL_DELAY, &sev->se_flags); | |
10590 | + schedule_sev_restart(sev); | |
10591 | + } | |
10592 | +} | |
10593 | + | |
10594 | +/* | |
10595 | + * Sevent backout code. Take appropriate steps when a recovery occurs while | |
10596 | + * we're in the midst of an sevent. The recovery may or may not affect the | |
10597 | + * sevent. If it does, it usually means cancelling the sevent and restarting | |
10598 | + * it from the beginning once the recovery processing is done. | |
10599 | + */ | |
10600 | + | |
10601 | +/* | |
10602 | + * If any of the nodes that replied with OK is dead, we give up on the current | |
10603 | + * join attempt and restart. Otherwise, this sevent can continue. | |
10604 | + */ | |
10605 | + | |
10606 | +static int backout_join_acked(sm_sevent_t *sev) | |
10607 | +{ | |
10608 | + sm_node_t *node; | |
10609 | + int i; | |
10610 | + | |
10611 | + for (i = 0; i < sev->se_node_count; i++) { | |
10612 | + if (sev->se_node_status[i] != STATUS_POS) | |
10613 | + continue; | |
10614 | + | |
10615 | + list_for_each_entry(node, &sm_members, list) { | |
10616 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags) && | |
10617 | + (node->id == sev->se_node_ids[i])) | |
10618 | + return TRUE; | |
10619 | + } | |
10620 | + } | |
10621 | + return FALSE; | |
10622 | +} | |
10623 | + | |
10624 | +/* | |
10625 | + * In this state our sg member list exists and mark_affected_sgs() will have | |
10626 | + * set NEED_RECOVERY if any of the nodes in the sg we're joining is dead. We | |
10627 | + * restart the join process if this is the case, otherwise this sevent can | |
10628 | + * continue. | |
10629 | + */ | |
10630 | + | |
10631 | +static int backout_jstop_ackwait(sm_sevent_t *sev) | |
10632 | +{ | |
10633 | + sm_group_t *sg = sev->se_sg; | |
10634 | + | |
10635 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10636 | + return FALSE; | |
10637 | + | |
10638 | + clear_bit(SEFL_ALLOW_JSTOP, &sev->se_flags); | |
10639 | + free_sg_memb(sg); | |
10640 | + return TRUE; | |
10641 | +} | |
10642 | + | |
10643 | +/* | |
10644 | + * Same as previous. | |
10645 | + */ | |
10646 | + | |
10647 | +static int backout_jstop_acked(sm_sevent_t *sev) | |
10648 | +{ | |
10649 | + return backout_jstop_ackwait(sev); | |
10650 | +} | |
10651 | + | |
10652 | +/* | |
10653 | + * If NEED_RECOVERY is set a member of the sg we're joining died while we were | |
10654 | + * starting our service. The recovery process will restart the service on all | |
10655 | + * the prior sg members (not including those that died or us). We will | |
10656 | + * reattempt our join which should be accepted once the nodes are done with | |
10657 | + * recovery. | |
10658 | + */ | |
10659 | + | |
10660 | +static int backout_jstart_servicewait(sm_sevent_t *sev) | |
10661 | +{ | |
10662 | + sm_group_t *sg = sev->se_sg; | |
10663 | + | |
10664 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10665 | + return FALSE; | |
10666 | + | |
10667 | + clear_bit(SEFL_ALLOW_STARTDONE, &sev->se_flags); | |
10668 | + sg->ops->stop(sg->service_data); | |
10669 | + free_sg_memb(sg); | |
10670 | + return TRUE; | |
10671 | +} | |
10672 | + | |
10673 | +/* | |
10674 | + * Same as previous. | |
10675 | + */ | |
10676 | + | |
10677 | +static int backout_jstart_servicedone(sm_sevent_t *sev) | |
10678 | +{ | |
10679 | + return backout_jstart_servicewait(sev); | |
10680 | +} | |
10681 | + | |
10682 | +/* | |
10683 | + * If NEED_RECOVERY is set a member of the sg we're joining died while we were | |
10684 | + * waiting on the "all done" barrier. Stop our service that we just started | |
10685 | + * and cancel the barrier. The recovery process will restart the service on | |
10686 | + * all the prior sg members (not including those that died or us). We will | |
10687 | + * reattempt our join which should be accepted once the nodes are done with | |
10688 | + * recovery. | |
10689 | + */ | |
10690 | + | |
10691 | +static int backout_barrier_wait(sm_sevent_t *sev) | |
10692 | +{ | |
10693 | + sm_group_t *sg = sev->se_sg; | |
10694 | + char bname[MAX_BARRIER_NAME_LEN]; | |
10695 | + | |
10696 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10697 | + return FALSE; | |
10698 | + | |
10699 | + clear_bit(SEFL_ALLOW_BARRIER, &sev->se_flags); | |
10700 | + | |
10701 | + sg->ops->stop(sg->service_data); | |
10702 | + | |
10703 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
10704 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
10705 | + sg->global_id, sm_our_nodeid, sev->se_id, | |
10706 | + sg->memb_count); | |
10707 | + kcl_barrier_cancel(bname); | |
10708 | + | |
10709 | + free_sg_memb(sg); | |
10710 | + return TRUE; | |
10711 | +} | |
10712 | + | |
10713 | +/* | |
10714 | + * If NEED_RECOVERY is set, a member of the sg we just joined has failed. The | |
10715 | + * recovery began after the barrier callback. If the result in the callback is | |
10716 | + * "success" then we are joined, this sevent is finished and we'll process the | |
10717 | + * sg within the forthcoming recovery with the other members. | |
10718 | + * | |
10719 | + * We rely upon cnxman to guarantee that once all nodes have joined a barrier, | |
10720 | + * all nodes will receive the corresponding barrier callback *before any* | |
10721 | + * receive an sm_member_update() due to one of those nodes failing just after | |
10722 | + * joining the barrier. If some nodes receive the sm_member_update() before | |
10723 | + * the barrier callback and others receive the barrier callback before the | |
10724 | + * sm_member_update() then they will disagree as to whether the node joining/ | |
10725 | + * leaving is in/out of the sg. | |
10726 | + */ | |
10727 | + | |
10728 | +static int backout_barrier_done(sm_sevent_t *sev) | |
10729 | +{ | |
10730 | + sm_group_t *sg = sev->se_sg; | |
10731 | + | |
10732 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10733 | + return FALSE; | |
10734 | + | |
10735 | + if (!sev->se_barrier_status) { | |
10736 | + do_finish_new(sev); | |
10737 | + sevent_done(sev); | |
10738 | + return FALSE; | |
10739 | + } else { | |
10740 | + sg->ops->stop(sg->service_data); | |
10741 | + free_sg_memb(sg); | |
10742 | + return TRUE; | |
10743 | + } | |
10744 | +} | |
10745 | + | |
10746 | +/* | |
10747 | + * We've done nothing yet, just restart when recovery is done (if sg is flagged | |
10748 | + * with recovery.) | |
10749 | + */ | |
10750 | + | |
10751 | +static int backout_leave_begin(sm_sevent_t *sev) | |
10752 | +{ | |
10753 | + sm_group_t *sg = sev->se_sg; | |
10754 | + | |
10755 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10756 | + return FALSE; | |
10757 | + | |
10758 | + return TRUE; | |
10759 | +} | |
10760 | + | |
10761 | +/* | |
10762 | + * Ignore any replies to our leave notice and restart when recovery is done (if | |
10763 | + * sg is flagged with recovery.) | |
10764 | + */ | |
10765 | + | |
10766 | +static int backout_leave_ackwait(sm_sevent_t *sev) | |
10767 | +{ | |
10768 | + sm_group_t *sg = sev->se_sg; | |
10769 | + | |
10770 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10771 | + return FALSE; | |
10772 | + | |
10773 | + clear_bit(SEFL_ALLOW_LEAVE, &sev->se_flags); | |
10774 | + | |
10775 | + return TRUE; | |
10776 | +} | |
10777 | + | |
10778 | +/* | |
10779 | + * Same as previous. | |
10780 | + */ | |
10781 | + | |
10782 | +static int backout_leave_acked(sm_sevent_t *sev) | |
10783 | +{ | |
10784 | + return backout_leave_ackwait(sev); | |
10785 | +} | |
10786 | + | |
10787 | +/* | |
10788 | + * Ignore any stop replies. All the members will be stopped anyway to do the | |
10789 | + * recovery. Let that happen and restart our leave when done. | |
10790 | + */ | |
10791 | + | |
10792 | +static int backout_lstop_ackwait(sm_sevent_t *sev) | |
10793 | +{ | |
10794 | + sm_group_t *sg = sev->se_sg; | |
10795 | + | |
10796 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10797 | + return FALSE; | |
10798 | + | |
10799 | + clear_bit(SEFL_ALLOW_LSTOP, &sev->se_flags); | |
10800 | + | |
10801 | + return TRUE; | |
10802 | +} | |
10803 | + | |
10804 | +/* | |
10805 | + * Same as previous. | |
10806 | + */ | |
10807 | + | |
10808 | +static int backout_lstop_acked(sm_sevent_t *sev) | |
10809 | +{ | |
10810 | + return backout_lstop_ackwait(sev); | |
10811 | +} | |
10812 | + | |
10813 | +/* | |
10814 | + * All members will be stopped due to recovery and restarted by recovery | |
10815 | + * processing. That includes us, we have to retry the leave once the recovery | |
10816 | + * is done. | |
10817 | + */ | |
10818 | + | |
10819 | +static int backout_lstart_waitremote(sm_sevent_t *sev) | |
10820 | +{ | |
10821 | + sm_group_t *sg = sev->se_sg; | |
10822 | + | |
10823 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10824 | + return FALSE; | |
10825 | + | |
10826 | + return TRUE; | |
10827 | +} | |
10828 | + | |
10829 | +/* | |
10830 | + * Reset an sevent to its beginning so it can be restarted. This is necessary | |
10831 | + * when recovery affects an SG while we're trying to join or leave (ie. a node | |
10832 | + * in the SG fails). | |
10833 | + */ | |
10834 | + | |
10835 | +void backout_sevents(void) | |
10836 | +{ | |
10837 | + sm_sevent_t *sev, *safe; | |
10838 | + int delay; | |
10839 | + | |
10840 | + list_for_each_entry_safe(sev, safe, &joinleave_events, se_list) { | |
10841 | + | |
10842 | + delay = FALSE; | |
10843 | + | |
10844 | + log_debug(sev->se_sg, "backout sevent state %u", sev->se_state); | |
10845 | + | |
10846 | + switch (sev->se_state) { | |
10847 | + | |
10848 | + /* backout after kcl_join_service and before | |
10849 | + * send_join_notice */ | |
10850 | + case SEST_JOIN_BEGIN: | |
10851 | + break; | |
10852 | + | |
10853 | + /* backout after send_join_notice and before final | |
10854 | + * process_reply */ | |
10855 | + case SEST_JOIN_ACKWAIT: | |
10856 | + clear_bit(SEFL_ALLOW_JOIN, &sev->se_flags); | |
10857 | + sev->se_state = SEST_JOIN_BEGIN; | |
b7b72b66 AM |
10858 | + set_bit(SEFL_CHECK, &sev->se_flags); |
10859 | + wake_serviced(DO_JOINLEAVE); | |
c1c6733f AM |
10860 | + break; |
10861 | + | |
10862 | + /* backout after final process_reply and before | |
10863 | + * check_join_notice */ | |
10864 | + case SEST_JOIN_ACKED: | |
10865 | + delay = backout_join_acked(sev); | |
10866 | + break; | |
10867 | + | |
10868 | + /* backout after send_join_stop and before final | |
10869 | + * process_reply */ | |
10870 | + case SEST_JSTOP_ACKWAIT: | |
10871 | + delay = backout_jstop_ackwait(sev); | |
10872 | + break; | |
10873 | + | |
10874 | + /* backout after final process_reply and before | |
10875 | + * check_join_stop */ | |
10876 | + case SEST_JSTOP_ACKED: | |
10877 | + delay = backout_jstop_acked(sev); | |
10878 | + break; | |
10879 | + | |
10880 | + /* backout after send_join_start and before | |
10881 | + * kcl_start_done */ | |
10882 | + case SEST_JSTART_SERVICEWAIT: | |
10883 | + delay = backout_jstart_servicewait(sev); | |
10884 | + break; | |
10885 | + | |
10886 | + /* backout after kcl_start_done and before | |
10887 | + * startdone_barrier_new */ | |
10888 | + case SEST_JSTART_SERVICEDONE: | |
10889 | + delay = backout_jstart_servicedone(sev); | |
10890 | + break; | |
10891 | + | |
10892 | + /* backout after startdone_barrier_new and before | |
10893 | + * callback_startdone_barrier_new */ | |
10894 | + case SEST_BARRIER_WAIT: | |
10895 | + delay = backout_barrier_wait(sev); | |
10896 | + break; | |
10897 | + | |
10898 | + /* backout after callback_startdone_barrier_new and | |
10899 | + * before check_startdone_barrier_new */ | |
10900 | + case SEST_BARRIER_DONE: | |
10901 | + delay = backout_barrier_done(sev); | |
10902 | + break; | |
10903 | + | |
10904 | + /* backout after kcl_leave_service and before | |
10905 | + * send_leave_notice */ | |
10906 | + case SEST_LEAVE_BEGIN: | |
10907 | + delay = backout_leave_begin(sev); | |
10908 | + break; | |
10909 | + | |
10910 | + /* backout after send_leave_notice and before final | |
10911 | + * process_reply */ | |
10912 | + case SEST_LEAVE_ACKWAIT: | |
10913 | + delay = backout_leave_ackwait(sev); | |
10914 | + break; | |
10915 | + | |
10916 | + /* backout after final process_reply and before | |
10917 | + * check_leave_notice */ | |
10918 | + case SEST_LEAVE_ACKED: | |
10919 | + delay = backout_leave_acked(sev); | |
10920 | + break; | |
10921 | + | |
10922 | + /* backout after send_leave_stop and before final | |
10923 | + * process_reply */ | |
10924 | + case SEST_LSTOP_ACKWAIT: | |
10925 | + delay = backout_lstop_ackwait(sev); | |
10926 | + break; | |
10927 | + | |
10928 | + /* backout after final process_reply and before | |
10929 | + * check_leave_stop */ | |
10930 | + case SEST_LSTOP_ACKED: | |
10931 | + delay = backout_lstop_acked(sev); | |
10932 | + break; | |
10933 | + | |
10934 | + /* backout after send_leave_start and before | |
10935 | + * process_lstart_done */ | |
10936 | + case SEST_LSTART_WAITREMOTE: | |
10937 | + delay = backout_lstart_waitremote(sev); | |
10938 | + break; | |
10939 | + | |
10940 | + /* backout after process_lstart_done and before | |
10941 | + * process_leave_sevent */ | |
10942 | + case SEST_LSTART_REMOTEDONE: | |
10943 | + sevent_done(sev); | |
10944 | + delay = FALSE; | |
10945 | + break; | |
10946 | + | |
10947 | + default: | |
10948 | + log_error(sev->se_sg, "backout_sevents: bad state %d", | |
10949 | + sev->se_state); | |
10950 | + } | |
10951 | + | |
10952 | + if (delay) { | |
c1c6733f AM |
10953 | + if (test_bit(SEFL_LEAVE, &sev->se_flags)) { |
10954 | + sev->se_state = SEST_LEAVE_BEGIN; | |
b7b72b66 AM |
10955 | + set_bit(SEFL_DELAY_RECOVERY, &sev->se_flags); |
10956 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
10957 | + wake_serviced(DO_JOINLEAVE); | |
c1c6733f AM |
10958 | + } else { |
10959 | + sev->se_state = SEST_JOIN_BEGIN; | |
b7b72b66 AM |
10960 | + set_bit(SEFL_CHECK, &sev->se_flags); |
10961 | + wake_serviced(DO_JOINLEAVE); | |
c1c6733f AM |
10962 | + } |
10963 | + } | |
10964 | + } | |
10965 | +} | |
10966 | + | |
10967 | +void process_joinleave(void) | |
10968 | +{ | |
10969 | + sm_sevent_t *sev = NULL, *safe; | |
10970 | + | |
10971 | + spin_lock(&new_event_lock); | |
10972 | + if (!list_empty(&new_event)) { | |
10973 | + sev = list_entry(new_event.next, sm_sevent_t, se_list); | |
10974 | + list_del(&sev->se_list); | |
10975 | + list_add_tail(&sev->se_list, &joinleave_events); | |
10976 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
10977 | + } | |
10978 | + spin_unlock(&new_event_lock); | |
10979 | + | |
10980 | + list_for_each_entry_safe(sev, safe, &joinleave_events, se_list) { | |
10981 | + if (!test_and_clear_bit(SEFL_CHECK, &sev->se_flags)) | |
10982 | + continue; | |
10983 | + | |
b7b72b66 AM |
10984 | + if (test_bit(SEFL_DELAY, &sev->se_flags) || |
10985 | + test_bit(SEFL_DELAY_RECOVERY, &sev->se_flags)) | |
c1c6733f AM |
10986 | + continue; |
10987 | + | |
10988 | + if (sev->se_state < SEST_LEAVE_BEGIN) | |
10989 | + process_join_sevent(sev); | |
10990 | + else | |
10991 | + process_leave_sevent(sev); | |
10992 | + } | |
10993 | +} | |
10994 | diff -urN linux-orig/cluster/cman/sm_joinleave.h linux-patched/cluster/cman/sm_joinleave.h | |
bb1d8b11 AM |
10995 | --- linux-orig/cluster/cman/sm_joinleave.h 1970-01-01 07:30:00.000000000 +0730 |
10996 | +++ linux-patched/cluster/cman/sm_joinleave.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
10997 | @@ -0,0 +1,23 @@ |
10998 | +/****************************************************************************** | |
10999 | +******************************************************************************* | |
11000 | +** | |
11001 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
11002 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
11003 | +** | |
11004 | +** This copyrighted material is made available to anyone wishing to use, | |
11005 | +** modify, copy, or redistribute it subject to the terms and conditions | |
11006 | +** of the GNU General Public License v.2. | |
11007 | +** | |
11008 | +******************************************************************************* | |
11009 | +******************************************************************************/ | |
11010 | + | |
11011 | +#ifndef __SM_JOINLEAVE_DOT_H__ | |
11012 | +#define __SM_JOINLEAVE_DOT_H__ | |
11013 | + | |
11014 | +void init_joinleave(void); | |
11015 | +void new_joinleave(sm_sevent_t *sev); | |
11016 | +void process_joinleave(void); | |
11017 | +void backout_sevents(void); | |
11018 | +sm_sevent_t *find_sevent(unsigned int id); | |
11019 | + | |
11020 | +#endif | |
11021 | diff -urN linux-orig/cluster/cman/sm_membership.c linux-patched/cluster/cman/sm_membership.c | |
bb1d8b11 AM |
11022 | --- linux-orig/cluster/cman/sm_membership.c 1970-01-01 07:30:00.000000000 +0730 |
11023 | +++ linux-patched/cluster/cman/sm_membership.c 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
11024 | @@ -0,0 +1,696 @@ |
11025 | +/****************************************************************************** | |
11026 | +******************************************************************************* | |
11027 | +** | |
11028 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
11029 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
11030 | +** | |
11031 | +** This copyrighted material is made available to anyone wishing to use, | |
11032 | +** modify, copy, or redistribute it subject to the terms and conditions | |
11033 | +** of the GNU General Public License v.2. | |
11034 | +** | |
11035 | +******************************************************************************* | |
11036 | +******************************************************************************/ | |
11037 | + | |
11038 | +#include "sm.h" | |
11039 | + | |
11040 | +extern struct list_head sm_members; | |
11041 | + | |
11042 | +/* | |
11043 | + * Routines for SG members to handle other nodes joining or leaving the SG. | |
11044 | + * These "uevent" membership update routines are the response to an "sevent" on | |
11045 | + * a joining/leaving node. | |
11046 | + */ | |
11047 | + | |
11048 | +static void del_memb_node(sm_group_t *sg, uint32_t nodeid) | |
11049 | +{ | |
11050 | + sm_node_t *node; | |
11051 | + | |
11052 | + list_for_each_entry(node, &sg->memb, list) { | |
11053 | + if (node->id != nodeid) | |
11054 | + continue; | |
11055 | + list_del(&node->list); | |
11056 | + kfree(node); | |
11057 | + sg->memb_count--; | |
11058 | + log_debug(sg, "del node %u count %d", nodeid, sg->memb_count); | |
11059 | + break; | |
11060 | + } | |
11061 | +} | |
11062 | + | |
11063 | +static void add_memb_node(sm_group_t *sg, sm_node_t *node) | |
11064 | +{ | |
11065 | + list_add_tail(&node->list, &sg->memb); | |
11066 | + sg->memb_count++; | |
11067 | + log_debug(sg, "add node %u count %d", node->id, sg->memb_count); | |
11068 | +} | |
11069 | + | |
11070 | +/* | |
11071 | + * Join 1. The receive end of send_join_stop() from a node requesting to join | |
11072 | + * the SG. We stop the service so it can be restarted with the new node. | |
11073 | + */ | |
11074 | + | |
11075 | +static int process_join_stop(sm_group_t *sg) | |
11076 | +{ | |
11077 | + sm_uevent_t *uev = &sg->uevent; | |
11078 | + sm_node_t *node; | |
11079 | + sm_msg_t reply; | |
11080 | + int error; | |
11081 | + | |
11082 | + if (uev->ue_num_nodes != sg->memb_count + 1) { | |
11083 | + log_error(sg, "process_join_stop: bad num nodes %u %u", | |
11084 | + uev->ue_num_nodes, sg->memb_count); | |
11085 | + return -1; | |
11086 | + } | |
11087 | + | |
11088 | + sm_set_event_id(&uev->ue_id); | |
11089 | + | |
11090 | + node = sm_find_joiner(sg, uev->ue_nodeid); | |
11091 | + SM_ASSERT(node,); | |
11092 | + | |
11093 | + sg->state = SGST_UEVENT; | |
11094 | + sg->ops->stop(sg->service_data); | |
11095 | + | |
11096 | + reply.ms_type = SMSG_JSTOP_REP; | |
11097 | + reply.ms_status = STATUS_POS; | |
11098 | + reply.ms_sevent_id = uev->ue_remote_seid; | |
11099 | + smsg_bswap_out(&reply); | |
11100 | + | |
11101 | + error = send_nodeid_message((char *) &reply, sizeof(reply), | |
11102 | + uev->ue_nodeid); | |
11103 | + if (error < 0) | |
11104 | + return error; | |
11105 | + return 0; | |
11106 | +} | |
11107 | + | |
11108 | +/* | |
11109 | + * Join 2. The receive end of send_join_start() from a node joining the SG. | |
11110 | + * We are re-starting the service with the new member added. | |
11111 | + */ | |
11112 | + | |
11113 | +static int process_join_start(sm_group_t *sg) | |
11114 | +{ | |
11115 | + sm_uevent_t *uev = &sg->uevent; | |
11116 | + sm_node_t *node; | |
11117 | + uint32_t *memb; | |
11118 | + int count = 0; | |
11119 | + | |
11120 | + /* this memory is passed to the service which must free it */ | |
11121 | + SM_RETRY(memb = | |
11122 | + kmalloc((sg->memb_count + 1) * sizeof(uint32_t), GFP_KERNEL), | |
11123 | + memb); | |
11124 | + | |
11125 | + /* transfer joining node from joining list to member list */ | |
11126 | + node = sm_find_joiner(sg, uev->ue_nodeid); | |
11127 | + SM_ASSERT(node, printk("nodeid=%u\n", uev->ue_nodeid);); | |
11128 | + list_del(&node->list); | |
11129 | + add_memb_node(sg, node); | |
11130 | + | |
11131 | + /* the new member list for the service */ | |
11132 | + list_for_each_entry(node, &sg->memb, list) | |
11133 | + memb[count++] = node->id; | |
11134 | + | |
11135 | + set_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11136 | + | |
11137 | + sg->ops->start(sg->service_data, memb, count, uev->ue_id, | |
11138 | + SERVICE_NODE_JOIN); | |
11139 | + return 0; | |
11140 | +} | |
11141 | + | |
11142 | +/* | |
11143 | + * Join 3. When done starting their local service, every previous SG member | |
11144 | + * calls startdone_barrier() and the new/joining member calls | |
11145 | + * startdone_barrier_new(). The barrier returns when everyone has started | |
11146 | + * their service and joined the barrier. | |
11147 | + */ | |
11148 | + | |
11149 | +static int startdone_barrier(sm_group_t *sg) | |
11150 | +{ | |
11151 | + sm_uevent_t *uev = &sg->uevent; | |
11152 | + char bname[MAX_BARRIER_NAME_LEN]; | |
11153 | + int error; | |
11154 | + | |
11155 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
11156 | + uev->ue_barrier_status = -1; | |
11157 | + | |
11158 | + set_bit(UEFL_ALLOW_BARRIER, &uev->ue_flags); | |
11159 | + | |
11160 | + /* If we're the only member, skip the barrier */ | |
11161 | + if (sg->memb_count == 1) { | |
11162 | + process_startdone_barrier(sg, 0); | |
11163 | + return 0; | |
11164 | + } | |
11165 | + | |
11166 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
11167 | + sg->global_id, uev->ue_nodeid, uev->ue_remote_seid, | |
11168 | + sg->memb_count); | |
11169 | + | |
11170 | + error = sm_barrier(bname, sg->memb_count, SM_BARRIER_STARTDONE); | |
11171 | + | |
11172 | + return error; | |
11173 | +} | |
11174 | + | |
11175 | +/* | |
11176 | + * Join 4. Check that the "all started" barrier returned a successful status. | |
11177 | + * The newly joined member calls check_startdone_barrier_new(). | |
11178 | + */ | |
11179 | + | |
11180 | +static int check_startdone_barrier(sm_group_t *sg) | |
11181 | +{ | |
11182 | + int error = sg->uevent.ue_barrier_status; | |
11183 | + return error; | |
11184 | +} | |
11185 | + | |
11186 | +/* | |
11187 | + * Join 5. Send the service a "finish" indicating that all members have | |
11188 | + * successfully started. The newly joined member calls do_finish_new(). | |
11189 | + */ | |
11190 | + | |
11191 | +static void do_finish(sm_group_t *sg) | |
11192 | +{ | |
11193 | + sg->state = SGST_RUN; | |
11194 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11195 | + sg->ops->finish(sg->service_data, sg->uevent.ue_id); | |
11196 | +} | |
11197 | + | |
11198 | +/* | |
11199 | + * Join 6. The uevent is done. If this was a uevent for a node leaving the | |
11200 | + * SG, then send a final message to the departed node signalling that the | |
11201 | + * remaining nodes have restarted since it left. | |
11202 | + */ | |
11203 | + | |
11204 | +static void uevent_done(sm_group_t *sg) | |
11205 | +{ | |
11206 | + sm_uevent_t *uev = &sg->uevent; | |
11207 | + sm_msg_t reply; | |
11208 | + | |
11209 | + if (test_bit(UEFL_LEAVE, &uev->ue_flags)) { | |
11210 | + reply.ms_type = SMSG_LSTART_DONE; | |
11211 | + reply.ms_status = STATUS_POS; | |
11212 | + reply.ms_sevent_id = uev->ue_remote_seid; | |
11213 | + smsg_bswap_out(&reply); | |
11214 | + send_nodeid_message((char *) &reply, sizeof(reply), | |
11215 | + uev->ue_nodeid); | |
11216 | + } | |
11217 | + memset(&sg->uevent, 0, sizeof(sm_uevent_t)); | |
11218 | +} | |
11219 | + | |
11220 | +/* | |
11221 | + * Leave 1. The receive end of send_leave_stop() from a node leaving the SG. | |
11222 | + */ | |
11223 | + | |
11224 | +static int process_leave_stop(sm_group_t *sg) | |
11225 | +{ | |
11226 | + sm_uevent_t *uev = &sg->uevent; | |
11227 | + sm_msg_t reply; | |
11228 | + int error; | |
11229 | + | |
11230 | + sm_set_event_id(&uev->ue_id); | |
11231 | + | |
11232 | + sg->state = SGST_UEVENT; | |
11233 | + sg->ops->stop(sg->service_data); | |
11234 | + | |
11235 | + reply.ms_type = SMSG_LSTOP_REP; | |
11236 | + reply.ms_status = STATUS_POS; | |
11237 | + reply.ms_sevent_id = uev->ue_remote_seid; | |
11238 | + smsg_bswap_out(&reply); | |
11239 | + | |
11240 | + error = send_nodeid_message((char *) &reply, sizeof(reply), | |
11241 | + uev->ue_nodeid); | |
11242 | + if (error < 0) | |
11243 | + return error; | |
11244 | + return 0; | |
11245 | +} | |
11246 | + | |
11247 | +/* | |
11248 | + * Leave 2. The receive end of send_leave_start() from a node leaving the SG. | |
11249 | + * We are re-starting the service (without the node that's left naturally.) | |
11250 | + */ | |
11251 | + | |
11252 | +static int process_leave_start(sm_group_t *sg) | |
11253 | +{ | |
11254 | + sm_uevent_t *uev = &sg->uevent; | |
11255 | + sm_node_t *node; | |
11256 | + uint32_t *memb; | |
11257 | + int count = 0; | |
11258 | + | |
11259 | + SM_ASSERT(sg->memb_count > 1, | |
11260 | + printk("memb_count=%u\n", sg->memb_count);); | |
11261 | + | |
11262 | + /* this memory is passed to the service which must free it */ | |
11263 | + SM_RETRY(memb = | |
11264 | + kmalloc((sg->memb_count - 1) * sizeof(uint32_t), GFP_KERNEL), | |
11265 | + memb); | |
11266 | + | |
11267 | + /* remove departed member from sg member list */ | |
11268 | + del_memb_node(sg, uev->ue_nodeid); | |
11269 | + | |
11270 | + /* build member list to pass to service */ | |
11271 | + list_for_each_entry(node, &sg->memb, list) | |
11272 | + memb[count++] = node->id; | |
11273 | + | |
11274 | + /* allow us to accept the start_done callback for this start */ | |
11275 | + set_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11276 | + | |
11277 | + sg->ops->start(sg->service_data, memb, count, uev->ue_id, | |
11278 | + SERVICE_NODE_LEAVE); | |
11279 | + return 0; | |
11280 | +} | |
11281 | + | |
11282 | +/* | |
11283 | + * Move through the steps of another node joining or leaving the SG. | |
11284 | + */ | |
11285 | + | |
11286 | +static void process_one_uevent(sm_group_t *sg) | |
11287 | +{ | |
11288 | + sm_uevent_t *uev = &sg->uevent; | |
11289 | + int error = 0; | |
11290 | + | |
11291 | + log_debug(sg, "uevent state %u node %u", uev->ue_state, uev->ue_nodeid); | |
11292 | + | |
11293 | + switch (uev->ue_state) { | |
11294 | + | |
11295 | + /* | |
11296 | + * a uevent is initialized with state JSTOP in | |
11297 | + * process_stop_request | |
11298 | + */ | |
11299 | + | |
11300 | + case UEST_JSTOP: | |
11301 | + uev->ue_state = UEST_JSTART_WAITCMD; | |
11302 | + error = process_join_stop(sg); | |
11303 | + break; | |
11304 | + | |
11305 | + /* | |
11306 | + * ue_state is changed from JSTART_WAITCMD to JSTART in | |
11307 | + * process_start_request | |
11308 | + */ | |
11309 | + | |
11310 | + case UEST_JSTART: | |
11311 | + uev->ue_state = UEST_JSTART_SERVICEWAIT; | |
11312 | + error = process_join_start(sg); | |
11313 | + break; | |
11314 | + | |
11315 | + /* | |
11316 | + * ue_state is changed from JSTART_SERVICEWAIT to | |
11317 | + * JSTART_SERVICEDONE in kcl_start_done | |
11318 | + */ | |
11319 | + | |
11320 | + case UEST_JSTART_SERVICEDONE: | |
11321 | + uev->ue_state = UEST_BARRIER_WAIT; | |
11322 | + error = startdone_barrier(sg); | |
11323 | + break; | |
11324 | + | |
11325 | + /* | |
11326 | + * ue_state is changed from BARRIER_WAIT to BARRIER_DONE in | |
11327 | + * process_startdone_barrier | |
11328 | + */ | |
11329 | + | |
11330 | + case UEST_BARRIER_DONE: | |
11331 | + error = check_startdone_barrier(sg); | |
11332 | + if (error) | |
11333 | + break; | |
11334 | + | |
11335 | + do_finish(sg); | |
11336 | + uevent_done(sg); | |
11337 | + break; | |
11338 | + | |
11339 | + /* | |
11340 | + * a uevent is initialized with state LSTOP in | |
11341 | + * process_stop_request | |
11342 | + */ | |
11343 | + | |
11344 | + case UEST_LSTOP: | |
11345 | + uev->ue_state = UEST_LSTART_WAITCMD; | |
11346 | + error = process_leave_stop(sg); | |
11347 | + break; | |
11348 | + | |
11349 | + /* | |
11350 | + * a uevent is changed from LSTART_WAITCMD to LSTART in | |
11351 | + * process_start_request | |
11352 | + */ | |
11353 | + | |
11354 | + case UEST_LSTART: | |
11355 | + uev->ue_state = UEST_LSTART_SERVICEWAIT; | |
11356 | + error = process_leave_start(sg); | |
11357 | + break; | |
11358 | + | |
11359 | + /* | |
11360 | + * a uevent is changed from LSTART_SERVICEWAIT to to | |
11361 | + * LSTART_SERVICEDONE in kcl_start_done | |
11362 | + */ | |
11363 | + | |
11364 | + case UEST_LSTART_SERVICEDONE: | |
11365 | + uev->ue_state = UEST_BARRIER_WAIT; | |
11366 | + error = startdone_barrier(sg); | |
11367 | + break; | |
11368 | + | |
11369 | + default: | |
11370 | + error = -1; | |
11371 | + } | |
11372 | + | |
11373 | + /* If we encounter an error during these routines, we do nothing, | |
11374 | + expecting that a node failure related to this sg will cause a | |
11375 | + recovery event to arrive and call cancel_one_uevent(). */ | |
11376 | + | |
11377 | + if (error) | |
11378 | + log_error(sg, "process_one_uevent error %d state %u", | |
11379 | + error, uev->ue_state); | |
11380 | +} | |
11381 | + | |
11382 | +static sm_node_t *failed_memb(sm_group_t *sg, int *count) | |
11383 | +{ | |
11384 | + sm_node_t *node, *sm_node, *failed_uev_node = NULL; | |
11385 | + | |
11386 | + list_for_each_entry(node, &sg->memb, list) { | |
11387 | + | |
11388 | + sm_node = sm_find_member(node->id); | |
11389 | + SM_ASSERT(sm_node, ); | |
11390 | + | |
11391 | + if (test_bit(SNFL_NEED_RECOVERY, &sm_node->flags)) { | |
11392 | + (*count)++; | |
11393 | + if (node->id == sg->uevent.ue_nodeid) | |
11394 | + failed_uev_node = sm_node; | |
11395 | + } | |
11396 | + } | |
11397 | + return failed_uev_node; | |
11398 | +} | |
11399 | + | |
11400 | +static void send_recover_msg(sm_group_t *sg) | |
11401 | +{ | |
11402 | + char *msg; | |
11403 | + int len = 0; | |
11404 | + msg = create_smsg(sg, SMSG_RECOVER, 0, &len, NULL); | |
11405 | + send_members_message(sg, msg, len); | |
11406 | +} | |
11407 | + | |
11408 | +static void cancel_barrier(sm_group_t *sg) | |
11409 | +{ | |
11410 | + sm_uevent_t *uev = &sg->uevent; | |
11411 | + char bname[MAX_BARRIER_NAME_LEN]; | |
11412 | + | |
11413 | + clear_bit(UEFL_ALLOW_BARRIER, &uev->ue_flags); | |
11414 | + | |
11415 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
11416 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
11417 | + sg->global_id, uev->ue_nodeid, uev->ue_remote_seid, | |
11418 | + sg->memb_count); | |
11419 | + kcl_barrier_cancel(bname); | |
11420 | +} | |
11421 | + | |
11422 | +static void cancel_one_uevent(sm_group_t *sg, int *effected) | |
11423 | +{ | |
11424 | + sm_uevent_t *uev = &sg->uevent; | |
11425 | + int failed_count; | |
11426 | + sm_node_t *node, *failed_joiner, *failed_leaver; | |
11427 | + | |
11428 | + log_debug(sg, "cancel uevent state %u node %u", uev->ue_state, | |
11429 | + uev->ue_nodeid); | |
11430 | + | |
11431 | + switch (uev->ue_state) { | |
11432 | + | |
11433 | + case UEST_JSTOP: | |
11434 | + case UEST_JSTART_WAITCMD: | |
11435 | + case UEST_JSTART: | |
11436 | + | |
11437 | + sg->ops->stop(sg->service_data); | |
11438 | + | |
11439 | + failed_count = 0; | |
11440 | + failed_joiner = failed_memb(sg, &failed_count); | |
11441 | + SM_ASSERT(!failed_joiner, ); | |
11442 | + | |
11443 | + node = sm_find_member(uev->ue_nodeid); | |
11444 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11445 | + failed_joiner = node; | |
11446 | + | |
11447 | + if (!failed_count) { | |
11448 | + /* only joining node failed */ | |
11449 | + SM_ASSERT(failed_joiner, ); | |
11450 | + SM_ASSERT(!test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11451 | + set_bit(SGFL_NEED_RECOVERY, &sg->flags); | |
11452 | + (*effected)++; | |
11453 | + /* some nodes may not have gotten a JSTOP message | |
11454 | + in which case this will tell them to begin | |
11455 | + recovery for this sg. */ | |
11456 | + send_recover_msg(sg); | |
11457 | + | |
11458 | + } else { | |
11459 | + /* a member node failed (and possibly joining node, it | |
11460 | + doesn't matter) */ | |
11461 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11462 | + } | |
11463 | + | |
11464 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11465 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11466 | + break; | |
11467 | + | |
11468 | + | |
11469 | + case UEST_JSTART_SERVICEWAIT: | |
11470 | + case UEST_JSTART_SERVICEDONE: | |
11471 | + | |
11472 | + clear_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11473 | + sg->ops->stop(sg->service_data); | |
11474 | + | |
11475 | + failed_count = 0; | |
11476 | + failed_joiner = failed_memb(sg, &failed_count); | |
11477 | + SM_ASSERT(failed_count, ); | |
11478 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11479 | + | |
11480 | + if (failed_count == 1 && failed_joiner) { | |
11481 | + /* only joining node failed */ | |
11482 | + | |
11483 | + } else if (failed_count && failed_joiner) { | |
11484 | + /* joining node and another member failed */ | |
11485 | + | |
11486 | + } else { | |
11487 | + /* other member failed, joining node still alive */ | |
11488 | + SM_ASSERT(!failed_joiner, ); | |
11489 | + del_memb_node(sg, uev->ue_nodeid); | |
11490 | + } | |
11491 | + | |
11492 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11493 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11494 | + break; | |
11495 | + | |
11496 | + | |
11497 | + case UEST_LSTOP: | |
11498 | + case UEST_LSTART_WAITCMD: | |
11499 | + case UEST_LSTART: | |
11500 | + | |
11501 | + sg->ops->stop(sg->service_data); | |
11502 | + | |
11503 | + failed_count = 0; | |
11504 | + failed_leaver = failed_memb(sg, &failed_count); | |
11505 | + SM_ASSERT(failed_count, ); | |
11506 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11507 | + | |
11508 | + if (failed_count == 1 && failed_leaver) { | |
11509 | + /* only leaving node failed */ | |
11510 | + | |
11511 | + } else if (failed_count && failed_leaver) { | |
11512 | + /* leaving node and another member failed */ | |
11513 | + | |
11514 | + } else { | |
11515 | + /* other member failed, leaving node still alive */ | |
11516 | + SM_ASSERT(!failed_leaver, ); | |
11517 | + } | |
11518 | + | |
11519 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11520 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11521 | + break; | |
11522 | + | |
11523 | + | |
11524 | + case UEST_LSTART_SERVICEWAIT: | |
11525 | + case UEST_LSTART_SERVICEDONE: | |
11526 | + | |
11527 | + clear_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11528 | + sg->ops->stop(sg->service_data); | |
11529 | + | |
11530 | + failed_count = 0; | |
11531 | + failed_leaver = failed_memb(sg, &failed_count); | |
11532 | + SM_ASSERT(!failed_leaver, ); | |
11533 | + | |
11534 | + node = sm_find_member(uev->ue_nodeid); | |
11535 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11536 | + failed_leaver = node; | |
11537 | + | |
11538 | + if (!failed_count) { | |
11539 | + /* only leaving node failed */ | |
11540 | + SM_ASSERT(failed_leaver, ); | |
11541 | + SM_ASSERT(!test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11542 | + set_bit(SGFL_NEED_RECOVERY, &sg->flags); | |
11543 | + (*effected)++; | |
11544 | + | |
11545 | + } else if (failed_count && failed_leaver) { | |
11546 | + /* leaving node and another member failed */ | |
11547 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11548 | + | |
11549 | + } else { | |
11550 | + /* other member failed, leaving node still alive */ | |
11551 | + SM_ASSERT(failed_count, ); | |
11552 | + SM_ASSERT(!failed_leaver, ); | |
11553 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11554 | + node = sm_new_node(sg->uevent.ue_nodeid); | |
11555 | + add_memb_node(sg, node); | |
11556 | + } | |
11557 | + | |
11558 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11559 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11560 | + break; | |
11561 | + | |
11562 | + | |
11563 | + case UEST_BARRIER_WAIT: | |
11564 | + | |
11565 | + if (test_bit(UEFL_LEAVE, &uev->ue_flags)) | |
11566 | + goto barrier_wait_leave; | |
11567 | + | |
11568 | + sg->ops->stop(sg->service_data); | |
11569 | + cancel_barrier(sg); | |
11570 | + | |
11571 | + barrier_wait_join: | |
11572 | + | |
11573 | + failed_count = 0; | |
11574 | + failed_joiner = failed_memb(sg, &failed_count); | |
11575 | + SM_ASSERT(failed_count, ); | |
11576 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11577 | + | |
11578 | + if (failed_count == 1 && failed_joiner) { | |
11579 | + /* only joining node failed */ | |
11580 | + | |
11581 | + } else if (failed_count && failed_joiner) { | |
11582 | + /* joining node and another member failed */ | |
11583 | + | |
11584 | + } else { | |
11585 | + /* other member failed, joining node still alive */ | |
11586 | + SM_ASSERT(!failed_joiner, ); | |
11587 | + del_memb_node(sg, uev->ue_nodeid); | |
11588 | + } | |
11589 | + | |
11590 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11591 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11592 | + break; | |
11593 | + | |
11594 | + barrier_wait_leave: | |
11595 | + | |
11596 | + failed_count = 0; | |
11597 | + failed_leaver = failed_memb(sg, &failed_count); | |
11598 | + SM_ASSERT(!failed_leaver, ); | |
11599 | + | |
11600 | + node = sm_find_member(uev->ue_nodeid); | |
11601 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11602 | + failed_leaver = node; | |
11603 | + | |
11604 | + if (!failed_count) { | |
11605 | + /* only leaving node failed */ | |
11606 | + SM_ASSERT(failed_leaver, ); | |
11607 | + SM_ASSERT(!test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11608 | + set_bit(SGFL_NEED_RECOVERY, &sg->flags); | |
11609 | + (*effected)++; | |
11610 | + | |
11611 | + } else if (failed_count && failed_leaver) { | |
11612 | + /* leaving node and another member failed */ | |
11613 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11614 | + | |
11615 | + } else { | |
11616 | + /* other member failed, leaving node still alive */ | |
11617 | + SM_ASSERT(failed_count, ); | |
11618 | + SM_ASSERT(!failed_leaver, ); | |
11619 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11620 | + node = sm_new_node(sg->uevent.ue_nodeid); | |
11621 | + add_memb_node(sg, node); | |
11622 | + } | |
11623 | + | |
11624 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11625 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11626 | + break; | |
11627 | + | |
11628 | + | |
11629 | + case UEST_BARRIER_DONE: | |
11630 | + | |
11631 | + if (!uev->ue_barrier_status) { | |
11632 | + do_finish(sg); | |
11633 | + uevent_done(sg); | |
11634 | + break; | |
11635 | + } | |
11636 | + | |
11637 | + if (test_bit(UEFL_LEAVE, &uev->ue_flags)) | |
11638 | + goto barrier_wait_leave; | |
11639 | + else | |
11640 | + goto barrier_wait_join; | |
11641 | + | |
11642 | + | |
11643 | + default: | |
11644 | + log_error(sg, "cancel_one_uevent: state %d", uev->ue_state); | |
11645 | + } | |
11646 | +} | |
11647 | + | |
11648 | +void cancel_uevents(int *effected) | |
11649 | +{ | |
11650 | + sm_group_t *sg; | |
11651 | + sm_node_t *node, *sgnode; | |
11652 | + int i; | |
11653 | + | |
11654 | + list_for_each_entry(node, &sm_members, list) { | |
11655 | + if (!test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11656 | + continue; | |
11657 | + | |
11658 | + /* | |
11659 | + * Clear this dead node from the "interested in joining" list | |
11660 | + * of any SG. The node is added to this list before the uevent | |
11661 | + * begins. | |
11662 | + */ | |
11663 | + | |
11664 | + for (i = 0; i < SG_LEVELS; i++) { | |
11665 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
11666 | + sgnode = sm_find_joiner(sg, node->id); | |
11667 | + if (sgnode) { | |
11668 | + log_debug(sg, "clear joining node %u", | |
11669 | + sgnode->id); | |
11670 | + list_del(&sgnode->list); | |
11671 | + kfree(sgnode); | |
11672 | + } | |
11673 | + } | |
11674 | + } | |
11675 | + } | |
11676 | + | |
11677 | + /* Adjust any uevents in sg's effected by the failed node(s) */ | |
11678 | + | |
11679 | + for (i = 0; i < SG_LEVELS; i++) { | |
11680 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
11681 | + if (!test_bit(SGFL_UEVENT, &sg->flags)) | |
11682 | + continue; | |
11683 | + | |
11684 | + /* We may have some cancelling to do if this sg is | |
11685 | + flagged as having a failed member, or if a joining | |
11686 | + or leaving node has died. */ | |
11687 | + | |
11688 | + if (test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
11689 | + cancel_one_uevent(sg, effected); | |
11690 | + else if (sg->uevent.ue_nodeid) { | |
11691 | + node = sm_find_member(sg->uevent.ue_nodeid); | |
11692 | + SM_ASSERT(node, ); | |
11693 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11694 | + cancel_one_uevent(sg, effected); | |
11695 | + } | |
11696 | + } | |
11697 | + } | |
11698 | +} | |
11699 | + | |
11700 | +void process_membership(void) | |
11701 | +{ | |
11702 | + sm_group_t *sg; | |
11703 | + int i; | |
11704 | + | |
11705 | + down(&sm_sglock); | |
11706 | + | |
11707 | + for (i = 0; i < SG_LEVELS; i++) { | |
11708 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
11709 | + if (!test_bit(SGFL_UEVENT, &sg->flags)) | |
11710 | + continue; | |
11711 | + | |
11712 | + if (!test_and_clear_bit(UEFL_CHECK, | |
11713 | + &sg->uevent.ue_flags)) | |
11714 | + continue; | |
11715 | + | |
11716 | + process_one_uevent(sg); | |
11717 | + } | |
11718 | + } | |
11719 | + up(&sm_sglock); | |
11720 | +} | |
11721 | diff -urN linux-orig/cluster/cman/sm_membership.h linux-patched/cluster/cman/sm_membership.h | |
bb1d8b11 AM |
11722 | --- linux-orig/cluster/cman/sm_membership.h 1970-01-01 07:30:00.000000000 +0730 |
11723 | +++ linux-patched/cluster/cman/sm_membership.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
11724 | @@ -0,0 +1,20 @@ |
11725 | +/****************************************************************************** | |
11726 | +******************************************************************************* | |
11727 | +** | |
11728 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
11729 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
11730 | +** | |
11731 | +** This copyrighted material is made available to anyone wishing to use, | |
11732 | +** modify, copy, or redistribute it subject to the terms and conditions | |
11733 | +** of the GNU General Public License v.2. | |
11734 | +** | |
11735 | +******************************************************************************* | |
11736 | +******************************************************************************/ | |
11737 | + | |
11738 | +#ifndef __SM_MEMBERSHIP_DOT_H__ | |
11739 | +#define __SM_MEMBERSHIP_DOT_H__ | |
11740 | + | |
11741 | +void process_membership(void); | |
11742 | +void cancel_uevents(int *effected); | |
11743 | + | |
11744 | +#endif | |
11745 | diff -urN linux-orig/cluster/cman/sm_message.c linux-patched/cluster/cman/sm_message.c | |
bb1d8b11 AM |
11746 | --- linux-orig/cluster/cman/sm_message.c 1970-01-01 07:30:00.000000000 +0730 |
11747 | +++ linux-patched/cluster/cman/sm_message.c 2004-11-03 11:37:37.000000000 +0800 | |
b7b72b66 | 11748 | @@ -0,0 +1,856 @@ |
c1c6733f AM |
11749 | +/****************************************************************************** |
11750 | +******************************************************************************* | |
11751 | +** | |
11752 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
11753 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
b7b72b66 | 11754 | +** |
c1c6733f AM |
11755 | +** This copyrighted material is made available to anyone wishing to use, |
11756 | +** modify, copy, or redistribute it subject to the terms and conditions | |
11757 | +** of the GNU General Public License v.2. | |
11758 | +** | |
11759 | +******************************************************************************* | |
11760 | +******************************************************************************/ | |
11761 | + | |
11762 | +#include "sm.h" | |
11763 | + | |
11764 | +#define SMSG_BUF_SIZE (sizeof(sm_msg_t) + MAX_SERVICE_NAME_LEN + 1) | |
11765 | + | |
11766 | +extern struct socket * sm_socket; | |
11767 | +extern uint32_t sm_our_nodeid; | |
11768 | +static uint32_t global_last_id; | |
11769 | +static struct list_head messages; | |
11770 | +static spinlock_t message_lock; | |
11771 | +static char smsg_buf[SMSG_BUF_SIZE]; | |
11772 | + | |
11773 | +int send_nodeid_message(char *msg, int len, uint32_t nodeid); | |
11774 | + | |
11775 | +struct rq_entry { | |
11776 | + struct list_head list; | |
11777 | + char *msg; | |
11778 | + int len; | |
11779 | + uint32_t nodeid; | |
11780 | +}; | |
11781 | +typedef struct rq_entry rq_entry_t; | |
11782 | + | |
11783 | +void init_messages(void) | |
11784 | +{ | |
11785 | + global_last_id = 1; | |
11786 | + INIT_LIST_HEAD(&messages); | |
11787 | + spin_lock_init(&message_lock); | |
11788 | +} | |
11789 | + | |
11790 | +uint32_t sm_new_global_id(int level) | |
11791 | +{ | |
11792 | + uint32_t id = global_last_id++; | |
11793 | + uint8_t l = (uint8_t) level; | |
11794 | + | |
11795 | + if (level > 255) | |
11796 | + return 0; | |
11797 | + | |
11798 | + if (id > 0x00FFFFFF) | |
11799 | + return 0; | |
11800 | + | |
11801 | + id |= (l << 24); | |
11802 | + return id; | |
11803 | +} | |
11804 | + | |
11805 | +static void smsg_copy_in(char *msg, sm_msg_t *smsg) | |
11806 | +{ | |
11807 | + sm_msg_t *in = (sm_msg_t *) msg; | |
11808 | + | |
11809 | + smsg->ms_type = in->ms_type; | |
11810 | + smsg->ms_status = in->ms_status; | |
11811 | + smsg->ms_sevent_id = le16_to_cpu(in->ms_sevent_id); | |
11812 | + smsg->ms_global_sgid = le32_to_cpu(in->ms_global_sgid); | |
11813 | + smsg->ms_global_lastid = le32_to_cpu(in->ms_global_lastid); | |
11814 | + smsg->ms_sglevel = le16_to_cpu(in->ms_sglevel); | |
11815 | + smsg->ms_length = le16_to_cpu(in->ms_length); | |
11816 | +} | |
11817 | + | |
11818 | +/* swapping bytes in place is an easy source of errors - be careful not to | |
11819 | + * access the fields after calling this */ | |
11820 | + | |
11821 | +void smsg_bswap_out(sm_msg_t *smsg) | |
11822 | +{ | |
11823 | + smsg->ms_sevent_id = cpu_to_le16(smsg->ms_sevent_id); | |
11824 | + smsg->ms_global_sgid = cpu_to_le32(smsg->ms_global_sgid); | |
11825 | + smsg->ms_global_lastid = cpu_to_le32(smsg->ms_global_lastid); | |
11826 | + smsg->ms_sglevel = cpu_to_le16(smsg->ms_sglevel); | |
11827 | + smsg->ms_length = cpu_to_le16(smsg->ms_length); | |
11828 | +} | |
11829 | + | |
11830 | +char *create_smsg(sm_group_t *sg, int type, int datalen, int *msglen, | |
11831 | + sm_sevent_t *sev) | |
11832 | +{ | |
11833 | + char *msg; | |
11834 | + sm_msg_t *smsg; | |
11835 | + int fulllen = sizeof(sm_msg_t) + datalen; | |
11836 | + | |
11837 | + msg = smsg_buf; | |
11838 | + memset(smsg_buf, 0, SMSG_BUF_SIZE); | |
11839 | + SM_ASSERT(fulllen <= SMSG_BUF_SIZE,); | |
11840 | + | |
11841 | + smsg = (sm_msg_t *) msg; | |
11842 | + smsg->ms_type = type; | |
11843 | + smsg->ms_global_sgid = sg->global_id; | |
11844 | + smsg->ms_sglevel = sg->level; | |
11845 | + smsg->ms_length = datalen; | |
11846 | + smsg->ms_sevent_id = sev ? sev->se_id : 0; | |
11847 | + | |
11848 | + smsg_bswap_out(smsg); | |
11849 | + *msglen = fulllen; | |
11850 | + return msg; | |
11851 | +} | |
11852 | + | |
11853 | +static unsigned int msgtype_to_flag(int type) | |
11854 | +{ | |
11855 | + unsigned int flag; | |
11856 | + | |
11857 | + switch (type) { | |
11858 | + case SMSG_JOIN_REP: | |
11859 | + case SMSG_JOIN_REQ: | |
11860 | + flag = SEFL_ALLOW_JOIN; | |
11861 | + break; | |
11862 | + | |
11863 | + case SMSG_JSTOP_REP: | |
11864 | + case SMSG_JSTOP_REQ: | |
11865 | + flag = SEFL_ALLOW_JSTOP; | |
11866 | + break; | |
11867 | + | |
11868 | + case SMSG_LEAVE_REP: | |
11869 | + case SMSG_LEAVE_REQ: | |
11870 | + flag = SEFL_ALLOW_LEAVE; | |
11871 | + break; | |
11872 | + | |
11873 | + case SMSG_LSTOP_REP: | |
11874 | + case SMSG_LSTOP_REQ: | |
11875 | + flag = SEFL_ALLOW_LSTOP; | |
11876 | + break; | |
11877 | + | |
11878 | + default: | |
11879 | + SM_ASSERT(0, printk("msgtype_to_flag bad type %d\n", type);); | |
11880 | + } | |
11881 | + return flag; | |
11882 | +} | |
11883 | + | |
b7b72b66 | 11884 | +static int test_allowed_msgtype(sm_sevent_t *sev, int type) |
c1c6733f AM |
11885 | +{ |
11886 | + unsigned int flag = msgtype_to_flag(type); | |
11887 | + | |
11888 | + return test_bit(flag, &sev->se_flags); | |
11889 | +} | |
11890 | + | |
b7b72b66 | 11891 | +static void clear_allowed_msgtype(sm_sevent_t *sev, int type) |
c1c6733f AM |
11892 | +{ |
11893 | + unsigned int flag = msgtype_to_flag(type); | |
11894 | + | |
11895 | + clear_bit(flag, &sev->se_flags); | |
11896 | +} | |
11897 | + | |
b7b72b66 | 11898 | +static void set_allowed_msgtype(sm_sevent_t *sev, int type) |
c1c6733f AM |
11899 | +{ |
11900 | + unsigned int flag = msgtype_to_flag(type); | |
11901 | + | |
11902 | + set_bit(flag, &sev->se_flags); | |
11903 | +} | |
11904 | + | |
b7b72b66 | 11905 | +static int save_global_id(sm_sevent_t *sev, sm_msg_t *smsg) |
c1c6733f AM |
11906 | +{ |
11907 | + sm_group_t *sg = sev->se_sg; | |
11908 | + | |
11909 | + if (!smsg->ms_global_sgid) { | |
11910 | + log_error(sg, "save_global_id: zero sg id"); | |
11911 | + return -1; | |
11912 | + } | |
11913 | + | |
11914 | + if (!sg->global_id) | |
11915 | + sg->global_id = smsg->ms_global_sgid; | |
11916 | + | |
11917 | + if (sg->global_id != smsg->ms_global_sgid) { | |
11918 | + log_error(sg, "save_global_id: id %x", smsg->ms_global_sgid); | |
11919 | + return -1; | |
11920 | + } | |
11921 | + return 0; | |
11922 | +} | |
11923 | + | |
b7b72b66 | 11924 | +static void save_lastid(sm_msg_t *smsg) |
c1c6733f AM |
11925 | +{ |
11926 | + uint32_t gid = smsg->ms_global_lastid & 0x00FFFFFF; | |
11927 | + | |
11928 | + /* | |
11929 | + * Keep track of the highst SG id which has been used | |
11930 | + * in the cluster in case we need to choose a new SG id. | |
11931 | + */ | |
11932 | + | |
11933 | + if (gid > global_last_id) | |
11934 | + global_last_id = gid; | |
11935 | +} | |
11936 | + | |
11937 | +static int next_sev_state(int msg_type, int cur_state) | |
11938 | +{ | |
11939 | + int next = 0; | |
11940 | + | |
11941 | + switch (msg_type) { | |
11942 | + case SMSG_JOIN_REP: | |
11943 | + SM_ASSERT(cur_state == SEST_JOIN_ACKWAIT,); | |
11944 | + next = SEST_JOIN_ACKED; | |
11945 | + break; | |
11946 | + | |
11947 | + case SMSG_JSTOP_REP: | |
11948 | + SM_ASSERT(cur_state == SEST_JSTOP_ACKWAIT,); | |
11949 | + next = SEST_JSTOP_ACKED; | |
11950 | + break; | |
11951 | + | |
11952 | + case SMSG_LEAVE_REP: | |
11953 | + SM_ASSERT(cur_state == SEST_LEAVE_ACKWAIT,); | |
11954 | + next = SEST_LEAVE_ACKED; | |
11955 | + break; | |
11956 | + | |
11957 | + case SMSG_LSTOP_REP: | |
11958 | + SM_ASSERT(cur_state == SEST_LSTOP_ACKWAIT,); | |
11959 | + next = SEST_LSTOP_ACKED; | |
11960 | + break; | |
11961 | + } | |
11962 | + return next; | |
11963 | +} | |
11964 | + | |
11965 | +/* | |
11966 | + * Functions in sevent.c send messages to other nodes and then expect replies. | |
11967 | + * This function collects the replies for the sevent messages and moves the | |
11968 | + * sevent to the next stage when all the expected replies have been received. | |
11969 | + */ | |
11970 | + | |
b7b72b66 | 11971 | +static void process_reply(sm_msg_t *smsg, uint32_t nodeid) |
c1c6733f AM |
11972 | +{ |
11973 | + sm_sevent_t *sev; | |
11974 | + int i, expected, type = smsg->ms_type; | |
11975 | + | |
11976 | + /* | |
11977 | + * Find the relevant sevent. | |
11978 | + */ | |
11979 | + | |
11980 | + sev = find_sevent(smsg->ms_sevent_id); | |
11981 | + if (!sev) { | |
11982 | + log_print("process_reply invalid id=%u nodeid=%u", | |
11983 | + smsg->ms_sevent_id, nodeid); | |
11984 | + goto out; | |
11985 | + } | |
11986 | + | |
11987 | + /* | |
11988 | + * Check if this message type is what this sevent is waiting for. | |
11989 | + */ | |
11990 | + | |
11991 | + if (!test_allowed_msgtype(sev, type)) { | |
11992 | + log_debug(sev->se_sg, "process_reply ignored type=%u nodeid=%u " "id=%u", type, nodeid, sev->se_id); | |
11993 | + goto out; | |
11994 | + } | |
11995 | + | |
11996 | + expected = | |
11997 | + (type == SMSG_JOIN_REP) ? sev->se_node_count : sev->se_memb_count; | |
11998 | + | |
11999 | + SM_ASSERT(expected * sizeof(uint32_t) <= sev->se_len_ids, | |
12000 | + printk("type=%d expected=%d len_ids=%d node_count=%d " | |
12001 | + "memb_count=%d\n", type, expected, sev->se_len_ids, | |
12002 | + sev->se_node_count, sev->se_memb_count);); | |
12003 | + | |
12004 | + SM_ASSERT(expected * sizeof(char) <= sev->se_len_status, | |
12005 | + printk("type=%d expected=%d len_status=%d node_count=%d " | |
12006 | + "memb_count=%d\n", type, expected, sev->se_len_status, | |
12007 | + sev->se_node_count, sev->se_memb_count);); | |
12008 | + | |
12009 | + for (i = 0; i < expected; i++) { | |
12010 | + if (sev->se_node_ids[i] == nodeid) { | |
12011 | + /* | |
12012 | + * Save the status from the replying node | |
12013 | + */ | |
12014 | + | |
12015 | + if (!sev->se_node_status[i]) | |
12016 | + sev->se_node_status[i] = smsg->ms_status; | |
12017 | + else { | |
12018 | + log_error(sev->se_sg, "process_reply duplicate" | |
12019 | + "id=%u nodeid=%u %u/%u", | |
12020 | + sev->se_id, nodeid, | |
12021 | + sev->se_node_status[i], | |
12022 | + smsg->ms_status); | |
12023 | + goto out; | |
12024 | + } | |
12025 | + | |
12026 | + if (type == SMSG_JOIN_REP) { | |
12027 | + save_lastid(smsg); | |
12028 | + | |
12029 | + if (smsg->ms_status == STATUS_POS) | |
12030 | + save_global_id(sev, smsg); | |
12031 | + } | |
12032 | + | |
12033 | + /* | |
12034 | + * Signal sm if we have all replies | |
12035 | + */ | |
12036 | + | |
12037 | + if (++sev->se_reply_count == expected) { | |
12038 | + clear_allowed_msgtype(sev, type); | |
b7b72b66 | 12039 | + sev->se_state = next_sev_state(type, |
c1c6733f AM |
12040 | + sev->se_state); |
12041 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
12042 | + wake_serviced(DO_JOINLEAVE); | |
12043 | + } | |
12044 | + | |
12045 | + break; | |
12046 | + } | |
12047 | + } | |
12048 | + | |
12049 | + out: | |
12050 | + return; | |
12051 | +} | |
12052 | + | |
12053 | +/* | |
12054 | + * A node wants to join an SG and has run send_join_notice. If we know nothing | |
12055 | + * about the SG , then we have no objection - send back STATUS_POS. If we're a | |
12056 | + * member of the SG, then send back STATUS_POS (go ahead and join) if there's | |
12057 | + * no sevent or uevent of higher priority in progress (only a single join or | |
12058 | + * leave is permitted for the SG at once). If there happens to be a higher | |
12059 | + * priority sevent/uevent in progress, send back STATUS_WAIT to defer the | |
12060 | + * requested join for a bit. | |
12061 | + */ | |
12062 | + | |
12063 | +static void process_join_request(sm_msg_t *smsg, uint32_t nodeid, char *name) | |
12064 | +{ | |
12065 | + sm_group_t *sg = NULL; | |
12066 | + sm_sevent_t *sev = NULL; | |
12067 | + sm_node_t *node; | |
12068 | + int found = FALSE; | |
12069 | + int level = smsg->ms_sglevel; | |
12070 | + sm_msg_t reply; | |
12071 | + | |
12072 | + memset(&reply, 0, sizeof(reply)); | |
12073 | + | |
12074 | + down(&sm_sglock); | |
12075 | + | |
12076 | + if (nodeid == sm_our_nodeid) | |
12077 | + goto next; | |
12078 | + | |
12079 | + /* | |
12080 | + * search SG list for an SG with given name/len | |
12081 | + */ | |
12082 | + | |
12083 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
12084 | + if ((sg->namelen != smsg->ms_length) || | |
12085 | + memcmp(sg->name, name, sg->namelen)) | |
12086 | + continue; | |
12087 | + found = TRUE; | |
12088 | + break; | |
12089 | + } | |
12090 | + | |
12091 | + /* | |
12092 | + * build reply message | |
12093 | + */ | |
12094 | + | |
12095 | + next: | |
12096 | + | |
12097 | + if (!found) { | |
12098 | + reply.ms_type = SMSG_JOIN_REP; | |
12099 | + reply.ms_status = STATUS_NEG; | |
12100 | + reply.ms_global_lastid = global_last_id; | |
12101 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12102 | + } else { | |
12103 | + reply.ms_type = SMSG_JOIN_REP; | |
12104 | + reply.ms_status = STATUS_POS; | |
12105 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12106 | + reply.ms_global_sgid = sg->global_id; | |
12107 | + reply.ms_global_lastid = global_last_id; | |
12108 | + | |
12109 | + /* | |
12110 | + * The node trying to join should wait and try again until | |
12111 | + * we're done with recovery. | |
12112 | + */ | |
12113 | + | |
12114 | + if (sg->state == SGST_RECOVER) { | |
12115 | + reply.ms_status = STATUS_WAIT; | |
12116 | + goto send; | |
12117 | + } | |
12118 | + | |
12119 | + /* | |
12120 | + * An sevent node trying to join may have gotten as far as | |
12121 | + * creating a uevent with us and then backed out. That node | |
12122 | + * will retry joining from the beginning so we should not turn | |
12123 | + * them away. If we're handling a uevent for another node, | |
12124 | + * tell the joining node to wait. | |
12125 | + */ | |
12126 | + | |
12127 | + if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
12128 | + if (sg->uevent.ue_nodeid != nodeid) | |
12129 | + reply.ms_status = STATUS_WAIT; | |
12130 | + goto send; | |
12131 | + } | |
12132 | + | |
12133 | + /* | |
12134 | + * We're trying to join or leave the SG at the moment. | |
12135 | + */ | |
12136 | + | |
12137 | + if (test_bit(SGFL_SEVENT, &sg->flags)) { | |
12138 | + sev = sg->sevent; | |
12139 | + | |
12140 | + /* | |
12141 | + * We're trying to leave. Make the join wait until | |
12142 | + * we've left if we're beyond LEAVE_ACKWAIT. | |
12143 | + */ | |
12144 | + | |
12145 | + if (test_bit(SEFL_LEAVE, &sev->se_flags)) { | |
12146 | + if (sev->se_state > SEST_LEAVE_ACKED) | |
12147 | + reply.ms_status = STATUS_WAIT; | |
12148 | + else { | |
12149 | + reply.ms_status = STATUS_POS; | |
12150 | + clear_bit(SEFL_ALLOW_LEAVE, | |
12151 | + &sev->se_flags); | |
12152 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
12153 | + } | |
12154 | + } | |
12155 | + | |
12156 | + /* | |
12157 | + * We're trying to join. Making the other join wait | |
12158 | + * until we're joined if we're beyond JOIN_ACKWAIT or | |
12159 | + * if we have a lower id. (Send NEG to allow the other | |
12160 | + * node to go ahead because we're not in the SG.) | |
12161 | + */ | |
12162 | + | |
12163 | + else { | |
12164 | + if (sev->se_state > SEST_JOIN_ACKED) | |
12165 | + reply.ms_status = STATUS_WAIT; | |
12166 | + else if (sm_our_nodeid < nodeid) | |
12167 | + reply.ms_status = STATUS_WAIT; | |
12168 | + else { | |
12169 | + reply.ms_status = STATUS_NEG; | |
12170 | + clear_bit(SEFL_ALLOW_JOIN, | |
12171 | + &sev->se_flags); | |
12172 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
12173 | + } | |
12174 | + } | |
12175 | + | |
12176 | + if (test_bit(SEFL_CANCEL, &sev->se_flags)) { | |
12177 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
12178 | + wake_serviced(DO_JOINLEAVE); | |
12179 | + } | |
12180 | + goto send; | |
12181 | + } | |
12182 | + | |
12183 | + /* no r,u,s event, stick with STATUS_POS */ | |
12184 | + } | |
12185 | + | |
12186 | + send: | |
12187 | + | |
12188 | + if (reply.ms_status == STATUS_POS) { | |
12189 | + node = sm_find_joiner(sg, nodeid); | |
12190 | + if (!node) { | |
12191 | + node = sm_new_node(nodeid); | |
12192 | + list_add_tail(&node->list, &sg->joining); | |
12193 | + } | |
12194 | + } | |
12195 | + | |
12196 | + up(&sm_sglock); | |
12197 | + smsg_bswap_out(&reply); | |
12198 | + send_nodeid_message((char *) &reply, sizeof(reply), nodeid); | |
12199 | +} | |
12200 | + | |
12201 | +/* | |
12202 | + * Another node wants us to stop a service so it can join or leave the SG. We | |
12203 | + * do this by saving the request info in a uevent and having the sm thread do | |
12204 | + * the processing and then replying. | |
12205 | + */ | |
12206 | + | |
b7b72b66 AM |
12207 | +static void process_stop_request(sm_msg_t *smsg, uint32_t nodeid, |
12208 | + uint32_t *msgbuf) | |
c1c6733f AM |
12209 | +{ |
12210 | + sm_group_t *sg; | |
12211 | + sm_uevent_t *uev; | |
12212 | + sm_msg_t reply; | |
12213 | + int type = smsg->ms_type; | |
12214 | + | |
12215 | + if (nodeid == sm_our_nodeid) | |
12216 | + goto agree; | |
12217 | + | |
12218 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
12219 | + if (!sg) { | |
12220 | + log_print("process_stop_request: unknown sg id %x", | |
12221 | + smsg->ms_global_sgid); | |
12222 | + return; | |
12223 | + } | |
12224 | + | |
12225 | + /* | |
12226 | + * We shouldn't get here with uevent already set. | |
12227 | + */ | |
12228 | + | |
12229 | + if (test_and_set_bit(SGFL_UEVENT, &sg->flags)) { | |
12230 | + log_error(sg, "process_stop_request: uevent already set"); | |
12231 | + return; | |
12232 | + } | |
12233 | + | |
12234 | + uev = &sg->uevent; | |
12235 | + uev->ue_nodeid = nodeid; | |
12236 | + uev->ue_remote_seid = smsg->ms_sevent_id; | |
12237 | + uev->ue_state = (type == SMSG_JSTOP_REQ) ? UEST_JSTOP : UEST_LSTOP; | |
12238 | + | |
12239 | + if (type == SMSG_JSTOP_REQ) | |
12240 | + uev->ue_num_nodes = be32_to_cpu(*msgbuf); | |
12241 | + else | |
12242 | + set_bit(UEFL_LEAVE, &uev->ue_flags); | |
12243 | + | |
12244 | + /* | |
12245 | + * Do process_join_stop() or process_leave_stop(). | |
12246 | + */ | |
12247 | + | |
12248 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
12249 | + wake_serviced(DO_MEMBERSHIP); | |
12250 | + return; | |
12251 | + | |
12252 | + agree: | |
12253 | + reply.ms_status = STATUS_POS; | |
12254 | + reply.ms_type = | |
12255 | + (type == SMSG_JSTOP_REQ) ? SMSG_JSTOP_REP : SMSG_LSTOP_REP; | |
12256 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12257 | + smsg_bswap_out(&reply); | |
12258 | + send_nodeid_message((char *) &reply, sizeof(reply), nodeid); | |
12259 | +} | |
12260 | + | |
b7b72b66 | 12261 | +static void process_start_request(sm_msg_t *smsg, uint32_t nodeid) |
c1c6733f AM |
12262 | +{ |
12263 | + sm_group_t *sg; | |
12264 | + sm_uevent_t *uev; | |
12265 | + int type = smsg->ms_type; | |
12266 | + | |
12267 | + if (nodeid == sm_our_nodeid) | |
12268 | + return; | |
12269 | + | |
12270 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
12271 | + if (!sg) { | |
12272 | + log_print("process_start_request: unknown sg id %x", | |
12273 | + smsg->ms_global_sgid); | |
12274 | + return; | |
12275 | + } | |
12276 | + | |
12277 | + if (!test_bit(SGFL_UEVENT, &sg->flags)) { | |
12278 | + log_error(sg, "process_start_request: no uevent"); | |
12279 | + return; | |
12280 | + } | |
12281 | + | |
12282 | + uev = &sg->uevent; | |
12283 | + | |
12284 | + if (type == SMSG_JSTART_CMD) | |
12285 | + uev->ue_state = UEST_JSTART; | |
12286 | + else | |
12287 | + uev->ue_state = UEST_LSTART; | |
12288 | + | |
12289 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
12290 | + wake_serviced(DO_MEMBERSHIP); | |
12291 | +} | |
12292 | + | |
b7b72b66 | 12293 | +static void process_leave_request(sm_msg_t *smsg, uint32_t nodeid) |
c1c6733f AM |
12294 | +{ |
12295 | + sm_group_t *sg; | |
12296 | + sm_node_t *node; | |
12297 | + sm_msg_t reply; | |
12298 | + sm_sevent_t *sev; | |
12299 | + int found = FALSE; | |
12300 | + | |
12301 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
12302 | + if (sg) { | |
12303 | + if (nodeid == sm_our_nodeid) | |
12304 | + found = TRUE; | |
12305 | + else { | |
12306 | + list_for_each_entry(node, &sg->memb, list) { | |
12307 | + if (node->id != nodeid) | |
12308 | + continue; | |
12309 | + set_bit(SNFL_LEAVING, &node->flags); | |
12310 | + found = TRUE; | |
12311 | + break; | |
12312 | + } | |
12313 | + } | |
12314 | + } | |
12315 | + | |
12316 | + if (!found) { | |
12317 | + reply.ms_type = SMSG_LEAVE_REP; | |
12318 | + reply.ms_status = STATUS_NEG; | |
12319 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12320 | + } else { | |
12321 | + reply.ms_type = SMSG_LEAVE_REP; | |
12322 | + reply.ms_status = STATUS_POS; | |
12323 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12324 | + | |
12325 | + if (sg->state == SGST_RECOVER) | |
12326 | + reply.ms_status = STATUS_WAIT; | |
12327 | + | |
12328 | + else if (test_bit(SGFL_SEVENT, &sg->flags) && | |
12329 | + nodeid != sm_our_nodeid) { | |
12330 | + sev = sg->sevent; | |
12331 | + | |
12332 | + /* | |
12333 | + * We're trying to join or leave at the moment. If | |
12334 | + * we're past JOIN/LEAVE_ACKWAIT, we make the requestor | |
12335 | + * wait. Otherwise, if joining we'll cancel to let the | |
12336 | + * leave happen first, or if we're leaving allow the | |
12337 | + * lower nodeid to leave first. | |
12338 | + */ | |
12339 | + | |
12340 | + if (test_bit(SEFL_LEAVE, &sev->se_flags)) { | |
12341 | + if (sev->se_state > SEST_LEAVE_ACKWAIT) | |
12342 | + reply.ms_status = STATUS_WAIT; | |
12343 | + else if (sm_our_nodeid < nodeid) | |
12344 | + reply.ms_status = STATUS_WAIT; | |
12345 | + else { | |
12346 | + reply.ms_status = STATUS_POS; | |
12347 | + clear_bit(SEFL_ALLOW_LEAVE, | |
12348 | + &sev->se_flags); | |
12349 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
12350 | + } | |
12351 | + } else { | |
12352 | + if (sev->se_state > SEST_JOIN_ACKWAIT) | |
12353 | + reply.ms_status = STATUS_WAIT; | |
12354 | + else { | |
12355 | + reply.ms_status = STATUS_NEG; | |
12356 | + clear_bit(SEFL_ALLOW_JOIN, | |
12357 | + &sev->se_flags); | |
12358 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
12359 | + } | |
12360 | + } | |
12361 | + | |
12362 | + if (test_bit(SEFL_CANCEL, &sev->se_flags)) { | |
12363 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
12364 | + wake_serviced(DO_JOINLEAVE); | |
12365 | + } | |
12366 | + } | |
12367 | + | |
12368 | + else if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
12369 | + if (sg->uevent.ue_nodeid != nodeid) | |
12370 | + reply.ms_status = STATUS_WAIT; | |
12371 | + } | |
12372 | + | |
12373 | + } | |
12374 | + | |
12375 | + smsg_bswap_out(&reply); | |
12376 | + send_nodeid_message((char *) &reply, sizeof(reply), nodeid); | |
12377 | +} | |
12378 | + | |
12379 | +/* | |
12380 | + * Each remaining node will send us a done message. We quit when we get the | |
12381 | + * first. The subsequent done messages for the finished sevent get here and | |
12382 | + * are ignored. | |
12383 | + */ | |
12384 | + | |
12385 | +static void process_lstart_done(sm_msg_t *smsg, uint32_t nodeid) | |
12386 | +{ | |
12387 | + sm_sevent_t *sev; | |
12388 | + | |
12389 | + sev = find_sevent(smsg->ms_sevent_id); | |
12390 | + if (!sev) | |
12391 | + return; | |
12392 | + | |
12393 | + if (sev->se_state != SEST_LSTART_WAITREMOTE) | |
12394 | + return; | |
12395 | + | |
12396 | + sev->se_state = SEST_LSTART_REMOTEDONE; | |
12397 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
12398 | + wake_serviced(DO_JOINLEAVE); | |
12399 | +} | |
12400 | + | |
12401 | +/* | |
12402 | + * This function and everything it calls always runs in sm context. | |
12403 | + */ | |
12404 | + | |
12405 | +static void process_message(char *msg, uint32_t nodeid) | |
12406 | +{ | |
12407 | + sm_msg_t smsg; | |
12408 | + | |
12409 | + smsg_copy_in(msg, &smsg); | |
12410 | + | |
12411 | + switch (smsg.ms_type) { | |
12412 | + case SMSG_JOIN_REQ: | |
12413 | + process_join_request(&smsg, nodeid, msg + sizeof(sm_msg_t)); | |
12414 | + break; | |
12415 | + | |
12416 | + case SMSG_JSTOP_REQ: | |
12417 | + process_stop_request(&smsg, nodeid, | |
12418 | + (uint32_t *) (msg + sizeof(sm_msg_t))); | |
12419 | + break; | |
12420 | + | |
12421 | + case SMSG_LEAVE_REQ: | |
12422 | + process_leave_request(&smsg, nodeid); | |
12423 | + break; | |
12424 | + | |
12425 | + case SMSG_LSTOP_REQ: | |
12426 | + process_stop_request(&smsg, nodeid, NULL); | |
12427 | + break; | |
12428 | + | |
12429 | + case SMSG_JSTART_CMD: | |
12430 | + case SMSG_LSTART_CMD: | |
12431 | + process_start_request(&smsg, nodeid); | |
12432 | + break; | |
12433 | + | |
12434 | + case SMSG_LSTART_DONE: | |
12435 | + process_lstart_done(&smsg, nodeid); | |
12436 | + break; | |
12437 | + | |
12438 | + case SMSG_JOIN_REP: | |
12439 | + case SMSG_JSTOP_REP: | |
12440 | + case SMSG_LEAVE_REP: | |
12441 | + case SMSG_LSTOP_REP: | |
12442 | + process_reply(&smsg, nodeid); | |
12443 | + break; | |
12444 | + | |
12445 | + case SMSG_RECOVER: | |
12446 | + process_recover_msg(&smsg, nodeid); | |
12447 | + break; | |
12448 | + | |
12449 | + default: | |
12450 | + log_print("process_message: unknown type %u nodeid %u", | |
12451 | + smsg.ms_type, nodeid); | |
12452 | + } | |
12453 | +} | |
12454 | + | |
12455 | +/* | |
12456 | + * Always called from sm context. | |
12457 | + */ | |
12458 | + | |
12459 | +void process_messages(void) | |
12460 | +{ | |
12461 | + rq_entry_t *re; | |
12462 | + | |
12463 | + while (1) { | |
12464 | + re = NULL; | |
12465 | + | |
12466 | + spin_lock(&message_lock); | |
12467 | + if (!list_empty(&messages)) { | |
12468 | + re = list_entry(messages.next, rq_entry_t, list); | |
12469 | + list_del(&re->list); | |
12470 | + } | |
12471 | + spin_unlock(&message_lock); | |
12472 | + | |
12473 | + if (!re) | |
12474 | + break; | |
12475 | + process_message(re->msg, re->nodeid); | |
12476 | + kfree(re->msg); | |
12477 | + kfree(re); | |
12478 | + schedule(); | |
12479 | + } | |
12480 | +} | |
12481 | + | |
12482 | +/* | |
12483 | + * Context: cnxman and sm | |
12484 | + */ | |
12485 | + | |
12486 | +static int add_to_recvqueue(char *msg, int len, uint32_t nodeid) | |
12487 | +{ | |
12488 | + rq_entry_t *re; | |
12489 | + | |
12490 | + SM_RETRY(re = (rq_entry_t *) kmalloc(sizeof(rq_entry_t), GFP_KERNEL), | |
12491 | + re); | |
12492 | + SM_RETRY(re->msg = (char *) kmalloc(len, GFP_KERNEL), re->msg); | |
12493 | + | |
12494 | + memcpy(re->msg, msg, len); | |
12495 | + re->len = len; | |
12496 | + re->nodeid = nodeid; | |
12497 | + | |
12498 | + spin_lock(&message_lock); | |
12499 | + list_add_tail(&re->list, &messages); | |
12500 | + spin_unlock(&message_lock); | |
12501 | + | |
12502 | + wake_serviced(DO_MESSAGES); | |
12503 | + return 0; | |
12504 | +} | |
12505 | + | |
12506 | +/* | |
12507 | + * Context: cnxman | |
12508 | + * Called by cnxman when a service manager message arrives. | |
12509 | + */ | |
12510 | + | |
12511 | +int sm_cluster_message(char *msg, int len, char *addr, int addr_len, | |
12512 | + unsigned int node_id) | |
12513 | +{ | |
b7b72b66 AM |
12514 | + if (!node_id) |
12515 | + return -EINVAL; | |
12516 | + return add_to_recvqueue(msg, len, node_id); | |
c1c6733f AM |
12517 | +} |
12518 | + | |
12519 | +/* | |
12520 | + * These send routines are used by sm and are always called from sm context. | |
12521 | + */ | |
12522 | + | |
12523 | +int send_nodeid_message(char *msg, int len, uint32_t nodeid) | |
12524 | +{ | |
12525 | + int error = 0; | |
12526 | + struct sockaddr_cl saddr; | |
12527 | + | |
12528 | + if (nodeid == sm_our_nodeid) { | |
12529 | + add_to_recvqueue(msg, len, nodeid); | |
12530 | + goto out; | |
12531 | + } | |
12532 | + | |
12533 | + saddr.scl_family = AF_CLUSTER; | |
12534 | + saddr.scl_port = CLUSTER_PORT_SERVICES; | |
12535 | + saddr.scl_nodeid = nodeid; | |
b7b72b66 | 12536 | + error = kcl_sendmsg(sm_socket, msg, len, &saddr, sizeof(saddr), 0); |
c1c6733f AM |
12537 | + if (error > 0) |
12538 | + error = 0; | |
12539 | + | |
12540 | + if (error) | |
12541 | + log_print("send_nodeid_message error %d to %u", error, nodeid); | |
12542 | + out: | |
12543 | + return error; | |
12544 | +} | |
12545 | + | |
12546 | +int send_broadcast_message(char *msg, int len) | |
12547 | +{ | |
12548 | + int error; | |
12549 | + | |
12550 | + error = kcl_sendmsg(sm_socket, msg, len, NULL, 0, 0); | |
12551 | + if (error > 0) | |
12552 | + error = 0; | |
12553 | + | |
12554 | + add_to_recvqueue(msg, len, sm_our_nodeid); | |
12555 | + | |
12556 | + if (error) | |
12557 | + log_print("send_broadcast_message error %d", error); | |
12558 | + | |
12559 | + return error; | |
12560 | +} | |
12561 | + | |
12562 | +int send_members_message(sm_group_t *sg, char *msg, int len) | |
12563 | +{ | |
12564 | + sm_node_t *node; | |
12565 | + int error = 0; | |
12566 | + | |
12567 | + list_for_each_entry(node, &sg->memb, list) { | |
12568 | + error = send_nodeid_message(msg, len, node->id); | |
12569 | + if (error < 0) | |
12570 | + break; | |
12571 | + } | |
12572 | + return error; | |
12573 | +} | |
12574 | + | |
12575 | +int send_members_message_sev(sm_group_t *sg, char *msg, int len, | |
12576 | + sm_sevent_t * sev) | |
12577 | +{ | |
12578 | + int error; | |
12579 | + sm_msg_t *smsg = (sm_msg_t *) msg; | |
12580 | + | |
12581 | + set_allowed_msgtype(sev, smsg->ms_type); | |
12582 | + sev->se_reply_count = 0; | |
12583 | + | |
12584 | + error = send_members_message(sg, msg, len); | |
12585 | + if (error < 0) | |
12586 | + clear_allowed_msgtype(sev, smsg->ms_type); | |
12587 | + | |
12588 | + return error; | |
12589 | +} | |
12590 | + | |
12591 | +int send_broadcast_message_sev(char *msg, int len, sm_sevent_t * sev) | |
12592 | +{ | |
12593 | + int error; | |
12594 | + sm_msg_t *smsg = (sm_msg_t *) msg; | |
12595 | + | |
12596 | + set_allowed_msgtype(sev, smsg->ms_type); | |
12597 | + sev->se_reply_count = 0; | |
12598 | + | |
12599 | + error = send_broadcast_message(msg, len); | |
12600 | + if (error < 0) | |
12601 | + clear_allowed_msgtype(sev, smsg->ms_type); | |
12602 | + | |
12603 | + return error; | |
12604 | +} | |
12605 | diff -urN linux-orig/cluster/cman/sm_message.h linux-patched/cluster/cman/sm_message.h | |
bb1d8b11 AM |
12606 | --- linux-orig/cluster/cman/sm_message.h 1970-01-01 07:30:00.000000000 +0730 |
12607 | +++ linux-patched/cluster/cman/sm_message.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
12608 | @@ -0,0 +1,34 @@ |
12609 | +/****************************************************************************** | |
12610 | +******************************************************************************* | |
12611 | +** | |
12612 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
12613 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
12614 | +** | |
12615 | +** This copyrighted material is made available to anyone wishing to use, | |
12616 | +** modify, copy, or redistribute it subject to the terms and conditions | |
12617 | +** of the GNU General Public License v.2. | |
12618 | +** | |
12619 | +******************************************************************************* | |
12620 | +******************************************************************************/ | |
12621 | + | |
12622 | +#ifndef __SM_MESSAGE_DOT_H__ | |
12623 | +#define __SM_MESSAGE_DOT_H__ | |
12624 | + | |
12625 | +void init_messages(void); | |
12626 | +uint32_t sm_new_global_id(int level); | |
12627 | +void smsg_bswap_out(sm_msg_t * smsg); | |
12628 | +char *create_smsg(sm_group_t *sg, int type, int datalen, int *msglen, | |
12629 | + sm_sevent_t *sev); | |
12630 | +void process_messages(void); | |
12631 | +int sm_cluster_message(char *msg, int len, char *addr, int addr_len, | |
12632 | + unsigned int node_id); | |
12633 | +int send_nodeid_message(char *msg, int len, uint32_t nodeid); | |
12634 | +int send_broadcast_message(char *msg, int len); | |
12635 | +int send_broadcast_message_sev(char *msg, int len, sm_sevent_t * sev); | |
12636 | +int send_members_message(sm_group_t *sg, char *msg, int len); | |
12637 | +int send_members_message_sev(sm_group_t *sg, char *msg, int len, | |
12638 | + sm_sevent_t * sev); | |
12639 | +int sm_cluster_message(char *msg, int len, char *addr, int addr_len, | |
12640 | + unsigned int node_id); | |
12641 | + | |
12642 | +#endif | |
12643 | diff -urN linux-orig/cluster/cman/sm_misc.c linux-patched/cluster/cman/sm_misc.c | |
bb1d8b11 AM |
12644 | --- linux-orig/cluster/cman/sm_misc.c 1970-01-01 07:30:00.000000000 +0730 |
12645 | +++ linux-patched/cluster/cman/sm_misc.c 2004-11-03 11:37:37.000000000 +0800 | |
b7b72b66 | 12646 | @@ -0,0 +1,442 @@ |
c1c6733f AM |
12647 | +/****************************************************************************** |
12648 | +******************************************************************************* | |
12649 | +** | |
12650 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
12651 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
b7b72b66 | 12652 | +** |
c1c6733f AM |
12653 | +** This copyrighted material is made available to anyone wishing to use, |
12654 | +** modify, copy, or redistribute it subject to the terms and conditions | |
12655 | +** of the GNU General Public License v.2. | |
12656 | +** | |
12657 | +******************************************************************************* | |
12658 | +******************************************************************************/ | |
12659 | + | |
12660 | +#include "sm.h" | |
12661 | +#include "config.h" | |
b7b72b66 | 12662 | +#include <linux/seq_file.h> |
c1c6733f AM |
12663 | + |
12664 | +#define MAX_DEBUG_MSG_LEN (40) | |
12665 | + | |
12666 | +extern struct list_head sm_members; | |
12667 | +static uint32_t local_ids; | |
12668 | +static uint32_t event_id; | |
12669 | +static spinlock_t event_id_lock; | |
12670 | +static char * debug_buf; | |
12671 | +static unsigned int debug_size; | |
12672 | +static unsigned int debug_point; | |
12673 | +static int debug_wrap; | |
12674 | +static spinlock_t debug_lock; | |
12675 | + | |
12676 | + | |
12677 | +void init_sm_misc(void) | |
12678 | +{ | |
12679 | + local_ids = 1; | |
12680 | + event_id = 1; | |
12681 | + spin_lock_init(&event_id_lock); | |
12682 | + debug_buf = NULL; | |
12683 | + debug_size = 0; | |
12684 | + debug_point = 0; | |
12685 | + debug_wrap = 0; | |
12686 | + spin_lock_init(&debug_lock); | |
12687 | + | |
12688 | + sm_debug_setup(cman_config.sm_debug_size); | |
12689 | +} | |
12690 | + | |
12691 | +sm_node_t *sm_new_node(uint32_t nodeid) | |
12692 | +{ | |
12693 | + struct kcl_cluster_node kclnode; | |
12694 | + sm_node_t *node; | |
12695 | + int error; | |
12696 | + | |
12697 | + error = kcl_get_node_by_nodeid(nodeid, &kclnode); | |
12698 | + SM_ASSERT(!error,); | |
12699 | + | |
12700 | + SM_RETRY(node = (sm_node_t *) kmalloc(sizeof(sm_node_t), GFP_KERNEL), | |
12701 | + node); | |
12702 | + | |
12703 | + memset(node, 0, sizeof(sm_node_t)); | |
12704 | + node->id = nodeid; | |
12705 | + node->incarnation = kclnode.incarnation; | |
12706 | + return node; | |
12707 | +} | |
12708 | + | |
12709 | +sm_node_t *sm_find_joiner(sm_group_t *sg, uint32_t nodeid) | |
12710 | +{ | |
12711 | + sm_node_t *node; | |
12712 | + | |
12713 | + list_for_each_entry(node, &sg->joining, list) { | |
12714 | + if (node->id == nodeid) | |
12715 | + return node; | |
12716 | + } | |
12717 | + return NULL; | |
12718 | +} | |
12719 | + | |
12720 | +sm_node_t *sm_find_member(uint32_t nodeid) | |
12721 | +{ | |
12722 | + sm_node_t *node; | |
12723 | + | |
12724 | + list_for_each_entry(node, &sm_members, list) { | |
12725 | + if (node->id == nodeid) | |
12726 | + return node; | |
12727 | + } | |
12728 | + return NULL; | |
12729 | +} | |
12730 | + | |
12731 | +uint32_t sm_new_local_id(int level) | |
12732 | +{ | |
12733 | + uint32_t id = local_ids++; | |
12734 | + uint8_t l = (uint8_t) level; | |
12735 | + | |
12736 | + if (level > 0xFF) | |
12737 | + return 0; | |
12738 | + | |
12739 | + if (id > 0x00FFFFFF) | |
12740 | + return 0; | |
12741 | + | |
12742 | + id |= (l << 24); | |
12743 | + return id; | |
12744 | +} | |
12745 | + | |
12746 | +int sm_id_to_level(uint32_t id) | |
12747 | +{ | |
12748 | + uint8_t l = (id & 0xFF000000) >> 24; | |
12749 | + | |
12750 | + return (int) l; | |
12751 | +} | |
12752 | + | |
12753 | +void sm_set_event_id(int *id) | |
12754 | +{ | |
12755 | + spin_lock(&event_id_lock); | |
12756 | + *id = event_id++; | |
12757 | + spin_unlock(&event_id_lock); | |
12758 | +} | |
12759 | + | |
12760 | +sm_group_t *sm_local_id_to_sg(int id) | |
12761 | +{ | |
12762 | + sm_group_t *sg; | |
12763 | + int level = sm_id_to_level(id); | |
12764 | + int found = FALSE; | |
12765 | + | |
12766 | + down(&sm_sglock); | |
12767 | + | |
12768 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
12769 | + if (sg->local_id == id) { | |
12770 | + found = TRUE; | |
12771 | + break; | |
12772 | + } | |
12773 | + } | |
12774 | + up(&sm_sglock); | |
12775 | + if (!found) | |
12776 | + sg = NULL; | |
12777 | + return sg; | |
12778 | +} | |
12779 | + | |
12780 | +sm_group_t *sm_global_id_to_sg(int id) | |
12781 | +{ | |
12782 | + sm_group_t *sg; | |
12783 | + int level = sm_id_to_level(id); | |
12784 | + int found = FALSE; | |
12785 | + | |
12786 | + down(&sm_sglock); | |
12787 | + | |
12788 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
12789 | + if (sg->global_id == id) { | |
12790 | + found = TRUE; | |
12791 | + break; | |
12792 | + } | |
12793 | + } | |
12794 | + up(&sm_sglock); | |
12795 | + if (!found) | |
12796 | + sg = NULL; | |
12797 | + return sg; | |
12798 | +} | |
12799 | + | |
12800 | +void sm_debug_log(sm_group_t *sg, const char *fmt, ...) | |
12801 | +{ | |
12802 | + va_list va; | |
12803 | + int i, n, size, len; | |
12804 | + char buf[MAX_DEBUG_MSG_LEN+1]; | |
12805 | + | |
12806 | + spin_lock(&debug_lock); | |
12807 | + | |
12808 | + if (!debug_buf) | |
12809 | + goto out; | |
12810 | + | |
12811 | + size = MAX_DEBUG_MSG_LEN; | |
12812 | + memset(buf, 0, size+1); | |
12813 | + | |
12814 | + n = snprintf(buf, size, "%08x ", sg->global_id); | |
12815 | + size -= n; | |
12816 | + | |
12817 | + va_start(va, fmt); | |
12818 | + vsnprintf(buf+n, size, fmt, va); | |
12819 | + va_end(va); | |
12820 | + | |
12821 | + len = strlen(buf); | |
12822 | + if (len > MAX_DEBUG_MSG_LEN-1) | |
12823 | + len = MAX_DEBUG_MSG_LEN-1; | |
12824 | + buf[len] = '\n'; | |
12825 | + buf[len+1] = '\0'; | |
12826 | + | |
12827 | + for (i = 0; i < strlen(buf); i++) { | |
12828 | + debug_buf[debug_point++] = buf[i]; | |
12829 | + | |
12830 | + if (debug_point == debug_size) { | |
12831 | + debug_point = 0; | |
12832 | + debug_wrap = 1; | |
12833 | + } | |
12834 | + } | |
12835 | + out: | |
12836 | + spin_unlock(&debug_lock); | |
12837 | +} | |
12838 | + | |
12839 | +void sm_debug_setup(int size) | |
12840 | +{ | |
12841 | + char *b = kmalloc(size, GFP_KERNEL); | |
12842 | + | |
12843 | + spin_lock(&debug_lock); | |
12844 | + if (debug_buf) | |
12845 | + kfree(debug_buf); | |
12846 | + | |
12847 | + if (size > PAGE_SIZE) | |
12848 | + size = PAGE_SIZE; | |
12849 | + debug_size = size; | |
12850 | + debug_point = 0; | |
12851 | + debug_wrap = 0; | |
12852 | + debug_buf = b; | |
12853 | + memset(debug_buf, 0, debug_size); | |
12854 | + spin_unlock(&debug_lock); | |
12855 | +} | |
12856 | + | |
12857 | +#ifdef CONFIG_PROC_FS | |
b7b72b66 AM |
12858 | +static struct seq_operations sm_info_op; |
12859 | + | |
12860 | +struct sm_seq_info | |
12861 | +{ | |
12862 | + int pos; | |
12863 | + int level; | |
12864 | + sm_group_t *sg; | |
12865 | +}; | |
c1c6733f AM |
12866 | + |
12867 | +int sm_debug_info(char *b, char **start, off_t offset, int length) | |
12868 | +{ | |
12869 | + int i, n = 0; | |
12870 | + | |
12871 | + spin_lock(&debug_lock); | |
12872 | + | |
12873 | + if (debug_wrap) { | |
12874 | + for (i = debug_point; i < debug_size; i++) | |
12875 | + n += sprintf(b + n, "%c", debug_buf[i]); | |
12876 | + } | |
12877 | + for (i = 0; i < debug_point; i++) | |
12878 | + n += sprintf(b + n, "%c", debug_buf[i]); | |
12879 | + | |
12880 | + spin_unlock(&debug_lock); | |
12881 | + | |
12882 | + return n; | |
12883 | +} | |
12884 | + | |
c1c6733f | 12885 | + |
c1c6733f | 12886 | + |
b7b72b66 AM |
12887 | +static sm_group_t *sm_walk(loff_t offset, int *rlevel) |
12888 | +{ | |
12889 | + sm_group_t *sg; | |
12890 | + int level; | |
12891 | + loff_t n = 0; | |
c1c6733f AM |
12892 | + |
12893 | + down(&sm_sglock); | |
12894 | + | |
12895 | + for (level = 0; level < SG_LEVELS; level++) { | |
12896 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
b7b72b66 AM |
12897 | + if (++n == offset) |
12898 | + goto walk_finish; | |
12899 | + } | |
12900 | + } | |
12901 | + sg = NULL; | |
c1c6733f | 12902 | + |
b7b72b66 AM |
12903 | + walk_finish: |
12904 | + up(&sm_sglock); | |
12905 | + *rlevel = level; | |
c1c6733f | 12906 | + |
b7b72b66 AM |
12907 | + return sg; |
12908 | +} | |
c1c6733f | 12909 | + |
c1c6733f | 12910 | + |
b7b72b66 AM |
12911 | +static void *sm_seq_start(struct seq_file *m, loff_t * pos) |
12912 | +{ | |
12913 | + struct sm_seq_info *ssi = | |
12914 | + kmalloc(sizeof (struct sm_seq_info), GFP_KERNEL); | |
c1c6733f | 12915 | + |
b7b72b66 AM |
12916 | + if (!ssi) |
12917 | + return NULL; | |
c1c6733f | 12918 | + |
b7b72b66 AM |
12919 | + ssi->pos = *pos; |
12920 | + ssi->level = 0; | |
12921 | + ssi->sg = NULL; | |
c1c6733f | 12922 | + |
b7b72b66 AM |
12923 | + /* Print the header */ |
12924 | + if (*pos == 0) { | |
12925 | + seq_printf(m, | |
12926 | + "Service Name GID LID State Code\n"); | |
12927 | + } | |
12928 | + return ssi; | |
12929 | +} | |
c1c6733f | 12930 | + |
b7b72b66 AM |
12931 | +static void *sm_seq_next(struct seq_file *m, void *p, loff_t * pos) |
12932 | +{ | |
12933 | + struct sm_seq_info *ssi = p; | |
c1c6733f | 12934 | + |
b7b72b66 | 12935 | + *pos = ++ssi->pos; |
c1c6733f | 12936 | + |
b7b72b66 AM |
12937 | + if ( !(ssi->sg = sm_walk(ssi->pos, &ssi->level)) ) |
12938 | + return NULL; | |
c1c6733f | 12939 | + |
b7b72b66 AM |
12940 | + return ssi; |
12941 | +} | |
12942 | + | |
12943 | +/* Called from /proc when /proc/cluster/services is opened */ | |
12944 | +int sm_proc_open(struct inode *inode, struct file *file) | |
12945 | +{ | |
12946 | + return seq_open(file, &sm_info_op); | |
12947 | +} | |
c1c6733f | 12948 | + |
b7b72b66 AM |
12949 | +static int sm_seq_show(struct seq_file *s, void *p) |
12950 | +{ | |
12951 | + struct sm_seq_info *ssi = p; | |
12952 | + sm_node_t *node; | |
12953 | + int i; | |
c1c6733f | 12954 | + |
b7b72b66 AM |
12955 | + if (!ssi || !ssi->sg) |
12956 | + return 0; | |
c1c6733f | 12957 | + |
b7b72b66 AM |
12958 | + /* |
12959 | + * Cluster Service | |
12960 | + */ | |
c1c6733f | 12961 | + |
b7b72b66 AM |
12962 | + switch (ssi->level) { |
12963 | + case SERVICE_LEVEL_FENCE: | |
12964 | + seq_printf(s, "Fence Domain: "); | |
12965 | + break; | |
12966 | + case SERVICE_LEVEL_GDLM: | |
12967 | + seq_printf(s, "DLM Lock Space: "); | |
12968 | + break; | |
12969 | + case SERVICE_LEVEL_GFS: | |
12970 | + seq_printf(s, "GFS Mount Group: "); | |
12971 | + break; | |
12972 | + case SERVICE_LEVEL_USER: | |
12973 | + seq_printf(s, "User: "); | |
12974 | + break; | |
12975 | + } | |
c1c6733f | 12976 | + |
b7b72b66 AM |
12977 | + /* |
12978 | + * Name | |
12979 | + */ | |
12980 | + | |
12981 | + seq_printf(s, "\""); | |
12982 | + for (i = 0; i < ssi->sg->namelen; i++) | |
12983 | + seq_printf(s, "%c", ssi->sg->name[i]); | |
12984 | + seq_printf(s, "\""); | |
12985 | + | |
12986 | + for (; i < MAX_SERVICE_NAME_LEN-1; i++) | |
12987 | + seq_printf(s, " "); | |
12988 | + | |
12989 | + /* | |
12990 | + * GID LID (sans level from top byte) | |
12991 | + */ | |
12992 | + | |
12993 | + seq_printf(s, "%3u %3u ", | |
12994 | + (ssi->sg->global_id & 0x00FFFFFF), | |
12995 | + (ssi->sg->local_id & 0x00FFFFFF)); | |
12996 | + | |
12997 | + /* | |
12998 | + * State | |
12999 | + */ | |
13000 | + | |
13001 | + switch (ssi->sg->state) { | |
13002 | + case SGST_NONE: | |
13003 | + seq_printf(s, "none "); | |
13004 | + break; | |
13005 | + case SGST_JOIN: | |
13006 | + seq_printf(s, "join "); | |
13007 | + break; | |
13008 | + case SGST_RUN: | |
13009 | + seq_printf(s, "run "); | |
13010 | + break; | |
13011 | + case SGST_RECOVER: | |
13012 | + seq_printf(s, "recover %u ", | |
13013 | + ssi->sg->recover_state); | |
13014 | + break; | |
13015 | + case SGST_UEVENT: | |
13016 | + seq_printf(s, "update "); | |
13017 | + break; | |
13018 | + } | |
c1c6733f | 13019 | + |
b7b72b66 AM |
13020 | + /* |
13021 | + * Code | |
13022 | + */ | |
13023 | + | |
13024 | + if (test_bit(SGFL_SEVENT, &ssi->sg->flags)) | |
13025 | + seq_printf(s, "S"); | |
13026 | + if (test_bit(SGFL_UEVENT, &ssi->sg->flags)) | |
13027 | + seq_printf(s, "U"); | |
13028 | + if (test_bit(SGFL_NEED_RECOVERY, &ssi->sg->flags)) | |
13029 | + seq_printf(s, "N"); | |
13030 | + | |
13031 | + seq_printf(s, "-"); | |
13032 | + | |
13033 | + if (test_bit(SGFL_SEVENT, &ssi->sg->flags) | |
13034 | + && ssi->sg->sevent) { | |
13035 | + seq_printf(s, "%u,%lx,%u", | |
13036 | + ssi->sg->sevent->se_state, | |
13037 | + ssi->sg->sevent->se_flags, | |
13038 | + ssi->sg->sevent->se_reply_count); | |
13039 | + } | |
c1c6733f | 13040 | + |
b7b72b66 AM |
13041 | + if (test_bit(SGFL_UEVENT, &ssi->sg->flags)) { |
13042 | + seq_printf(s, "%u,%lx,%u", | |
13043 | + ssi->sg->uevent.ue_state, | |
13044 | + ssi->sg->uevent.ue_flags, | |
13045 | + ssi->sg->uevent.ue_nodeid); | |
13046 | + } | |
c1c6733f | 13047 | + |
b7b72b66 | 13048 | + seq_printf(s, "\n"); |
c1c6733f | 13049 | + |
b7b72b66 AM |
13050 | + /* |
13051 | + * node list | |
13052 | + */ | |
c1c6733f | 13053 | + |
b7b72b66 | 13054 | + i = 0; |
c1c6733f | 13055 | + |
b7b72b66 | 13056 | + seq_printf(s, "["); |
c1c6733f | 13057 | + |
b7b72b66 AM |
13058 | + list_for_each_entry(node, &ssi->sg->memb, list) { |
13059 | + if (i && !(i % 24)) | |
13060 | + seq_printf(s, "\n"); | |
c1c6733f | 13061 | + |
b7b72b66 AM |
13062 | + if (i) |
13063 | + seq_printf(s, " "); | |
13064 | + | |
13065 | + seq_printf(s, "%u", node->id); | |
13066 | + i++; | |
13067 | + } | |
13068 | + | |
13069 | + seq_printf(s, "]\n\n"); | |
13070 | + | |
13071 | + return 0; | |
13072 | +} | |
13073 | + | |
13074 | +static void sm_seq_stop(struct seq_file *m, void *p) | |
13075 | +{ | |
13076 | + kfree(p); | |
c1c6733f | 13077 | +} |
b7b72b66 AM |
13078 | + |
13079 | + | |
13080 | +static struct seq_operations sm_info_op = { | |
13081 | + .start = sm_seq_start, | |
13082 | + .next = sm_seq_next, | |
13083 | + .stop = sm_seq_stop, | |
13084 | + .show = sm_seq_show | |
13085 | +}; | |
13086 | + | |
13087 | + | |
c1c6733f AM |
13088 | +#endif |
13089 | diff -urN linux-orig/cluster/cman/sm_misc.h linux-patched/cluster/cman/sm_misc.h | |
bb1d8b11 AM |
13090 | --- linux-orig/cluster/cman/sm_misc.h 1970-01-01 07:30:00.000000000 +0730 |
13091 | +++ linux-patched/cluster/cman/sm_misc.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
13092 | @@ -0,0 +1,29 @@ |
13093 | +/****************************************************************************** | |
13094 | +******************************************************************************* | |
13095 | +** | |
13096 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13097 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13098 | +** | |
13099 | +** This copyrighted material is made available to anyone wishing to use, | |
13100 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13101 | +** of the GNU General Public License v.2. | |
13102 | +** | |
13103 | +******************************************************************************* | |
13104 | +******************************************************************************/ | |
13105 | + | |
13106 | +#ifndef __SM_MISC_DOT_H__ | |
13107 | +#define __SM_MISC_DOT_H__ | |
13108 | + | |
13109 | +void init_sm_misc(void); | |
13110 | +sm_node_t *sm_new_node(uint32_t nodeid); | |
13111 | +sm_node_t *sm_find_joiner(sm_group_t *sg, uint32_t nodeid); | |
13112 | +sm_node_t *sm_find_member(uint32_t nodeid); | |
13113 | +uint32_t sm_new_local_id(int level); | |
13114 | +int sm_id_to_level(uint32_t id); | |
13115 | +void sm_set_event_id(int *id); | |
13116 | +sm_group_t *sm_local_id_to_sg(int id); | |
13117 | +sm_group_t *sm_global_id_to_sg(int id); | |
13118 | +void sm_debug_log(sm_group_t *sg, const char *fmt, ...); | |
13119 | +void sm_debug_setup(int size); | |
13120 | + | |
13121 | +#endif | |
13122 | diff -urN linux-orig/cluster/cman/sm_recover.c linux-patched/cluster/cman/sm_recover.c | |
bb1d8b11 AM |
13123 | --- linux-orig/cluster/cman/sm_recover.c 1970-01-01 07:30:00.000000000 +0730 |
13124 | +++ linux-patched/cluster/cman/sm_recover.c 2004-11-03 11:37:37.000000000 +0800 | |
b7b72b66 | 13125 | @@ -0,0 +1,524 @@ |
c1c6733f AM |
13126 | +/****************************************************************************** |
13127 | +******************************************************************************* | |
13128 | +** | |
13129 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13130 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13131 | +** | |
13132 | +** This copyrighted material is made available to anyone wishing to use, | |
13133 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13134 | +** of the GNU General Public License v.2. | |
13135 | +** | |
13136 | +******************************************************************************* | |
13137 | +******************************************************************************/ | |
13138 | + | |
13139 | +#include "sm.h" | |
13140 | +#include "config.h" | |
13141 | + | |
13142 | +/* | |
13143 | + * A collection of sg's which need to be recovered due to a failed member. | |
13144 | + * These sg's are recovered in order of level. An sg subject to cascading | |
13145 | + * failures is moved from one of these structs to a newer one. | |
13146 | + */ | |
13147 | + | |
13148 | +struct recover { | |
13149 | + struct list_head list; /* list of current re's */ | |
13150 | + struct list_head sgs[SG_LEVELS]; /* lists of sg's by level */ | |
13151 | + int event_id; /* event id */ | |
13152 | + int cur_level; | |
13153 | +}; | |
13154 | +typedef struct recover recover_t; | |
13155 | + | |
13156 | + | |
13157 | +extern uint32_t * sm_new_nodeids; | |
13158 | +extern int sm_quorum, sm_quorum_next; | |
13159 | +extern uint32_t sm_our_nodeid; | |
13160 | +extern struct list_head sm_members; | |
13161 | +extern int sm_member_count; | |
13162 | +static struct list_head recoveries; | |
13163 | + | |
13164 | + | |
13165 | +void init_recovery(void) | |
13166 | +{ | |
13167 | + INIT_LIST_HEAD(&recoveries); | |
13168 | +} | |
13169 | + | |
13170 | +/* | |
13171 | + * This is the first thing called when a change is announced in cluster | |
13172 | + * membership. Nodes are marked as being a CLUSTER_MEMBER or not. SM adds new | |
13173 | + * nodes to its sm_members list which it's not seen before. Nodes which were | |
13174 | + * alive but are now gone are marked as "need recovery". | |
13175 | + * | |
13176 | + * The "need recovery" status of nodes is propagated to the node's SG's in | |
13177 | + * mark_effected_sgs. The effected SG's are themselves marked as needing | |
13178 | + * recovery and in new_recovery the dead nodes are removed from the SG's | |
13179 | + * individual member lists. The "need recovery" status of nodes is cleared in | |
13180 | + * adjust_members_done(). | |
13181 | + */ | |
13182 | + | |
13183 | +static int adjust_members(void) | |
13184 | +{ | |
13185 | + sm_node_t *node; | |
13186 | + struct kcl_cluster_node knode; | |
13187 | + int i, error, num_nodes, sub = 0, add = 0, found; | |
13188 | + | |
13189 | + /* | |
13190 | + * Get list of current members from cnxman | |
13191 | + */ | |
13192 | + | |
13193 | + memset(sm_new_nodeids, 0, cman_config.max_nodes * sizeof(uint32_t)); | |
13194 | + num_nodes = kcl_get_member_ids(sm_new_nodeids, cman_config.max_nodes); | |
13195 | + | |
13196 | + /* | |
13197 | + * Determine who's gone | |
13198 | + */ | |
13199 | + | |
13200 | + list_for_each_entry(node, &sm_members, list) { | |
13201 | + found = FALSE; | |
13202 | + for (i = 0; i < num_nodes; i++) { | |
13203 | + if (node->id == sm_new_nodeids[i]) { | |
13204 | + found = TRUE; | |
13205 | + sm_new_nodeids[i] = 0; | |
13206 | + break; | |
13207 | + } | |
13208 | + } | |
13209 | + | |
13210 | + if (found) { | |
13211 | + error = kcl_get_node_by_nodeid(node->id, &knode); | |
13212 | + SM_ASSERT(!error, printk("error=%d\n", error);); | |
13213 | + | |
13214 | + if (!test_bit(SNFL_CLUSTER_MEMBER, &node->flags)) { | |
13215 | + /* former member is back */ | |
13216 | + set_bit(SNFL_CLUSTER_MEMBER, &node->flags); | |
13217 | + node->incarnation = knode.incarnation; | |
13218 | + add++; | |
13219 | + } else { | |
13220 | + /* current member is still alive - if the | |
13221 | + * incarnation number is different it died and | |
13222 | + * returned between checks */ | |
13223 | + if (node->incarnation != knode.incarnation) { | |
13224 | + set_bit(SNFL_NEED_RECOVERY, | |
13225 | + &node->flags); | |
13226 | + node->incarnation = knode.incarnation; | |
13227 | + sub++; | |
13228 | + } | |
13229 | + } | |
13230 | + } else { | |
13231 | + /* current member has died */ | |
13232 | + if (test_and_clear_bit(SNFL_CLUSTER_MEMBER, | |
13233 | + &node->flags)) { | |
13234 | + set_bit(SNFL_NEED_RECOVERY, &node->flags); | |
13235 | + sub++; | |
13236 | + } | |
13237 | + } | |
13238 | + } | |
13239 | + | |
13240 | + /* | |
13241 | + * Look for new nodes | |
13242 | + */ | |
13243 | + | |
13244 | + for (i = 0; i < num_nodes; i++) { | |
13245 | + if (sm_new_nodeids[i]) { | |
13246 | + node = sm_new_node(sm_new_nodeids[i]); | |
13247 | + set_bit(SNFL_CLUSTER_MEMBER, &node->flags); | |
13248 | + add++; | |
13249 | + list_add_tail(&node->list, &sm_members); | |
13250 | + sm_member_count++; | |
13251 | + } | |
13252 | + } | |
13253 | + | |
13254 | + /* | |
13255 | + * Get our own nodeid | |
13256 | + */ | |
13257 | + | |
13258 | + if (!sm_our_nodeid) { | |
13259 | + list_for_each_entry(node, &sm_members, list) { | |
13260 | + error = kcl_get_node_by_nodeid(node->id, &knode); | |
13261 | + SM_ASSERT(!error, printk("error=%d\n", error);); | |
13262 | + | |
13263 | + if (knode.us) { | |
13264 | + sm_our_nodeid = knode.node_id; | |
13265 | + break; | |
13266 | + } | |
13267 | + } | |
13268 | + } | |
13269 | + | |
13270 | + return sub; | |
13271 | +} | |
13272 | + | |
13273 | +/* | |
13274 | + * Given some number of dead nodes, flag SG's the dead nodes were part of. | |
13275 | + * This requires a number of loops because each node structure does not keep a | |
13276 | + * list of SG's it's in. | |
13277 | + */ | |
13278 | + | |
13279 | +static int mark_effected_sgs(void) | |
13280 | +{ | |
13281 | + sm_group_t *sg; | |
13282 | + sm_node_t *node, *sgnode; | |
13283 | + uint32_t dead_id; | |
13284 | + int i, effected = 0; | |
13285 | + | |
13286 | + down(&sm_sglock); | |
13287 | + | |
13288 | + list_for_each_entry(node, &sm_members, list) { | |
13289 | + if (!test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
13290 | + continue; | |
13291 | + | |
13292 | + dead_id = node->id; | |
13293 | + | |
13294 | + for (i = 0; i < SG_LEVELS; i++) { | |
13295 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
13296 | + /* check if dead node is among sg's members */ | |
13297 | + list_for_each_entry(sgnode, &sg->memb, list) { | |
13298 | + if (sgnode->id == dead_id) { | |
13299 | + set_bit(SGFL_NEED_RECOVERY, | |
13300 | + &sg->flags); | |
13301 | + effected++; | |
13302 | + break; | |
13303 | + } | |
13304 | + } | |
13305 | + } | |
13306 | + } | |
13307 | + } | |
13308 | + up(&sm_sglock); | |
13309 | + | |
13310 | + return effected; | |
13311 | +} | |
13312 | + | |
13313 | +static recover_t *alloc_recover(void) | |
13314 | +{ | |
13315 | + recover_t *rev; | |
13316 | + int i; | |
13317 | + | |
13318 | + SM_RETRY(rev = kmalloc(sizeof(recover_t), GFP_KERNEL), rev); | |
13319 | + | |
13320 | + memset(rev, 0, sizeof(recover_t)); | |
13321 | + | |
13322 | + sm_set_event_id(&rev->event_id); | |
13323 | + | |
13324 | + for (i = 0; i < SG_LEVELS; i++) { | |
13325 | + INIT_LIST_HEAD(&rev->sgs[i]); | |
13326 | + } | |
13327 | + | |
13328 | + return rev; | |
13329 | +} | |
13330 | + | |
13331 | +/* | |
13332 | + * An in-progress revent re-start for an SG is interrupted by another node | |
13333 | + * failure in the SG. Cancel an outstanding barrier if there is one. The SG | |
13334 | + * will be moved to the new revent and re-started as part of that. | |
13335 | + */ | |
13336 | + | |
13337 | +static void cancel_prev_recovery(sm_group_t *sg) | |
13338 | +{ | |
13339 | + int error; | |
13340 | + | |
13341 | + if (sg->recover_state == RECOVER_BARRIERWAIT) { | |
13342 | + error = kcl_barrier_cancel(sg->recover_barrier); | |
13343 | + if (error) | |
13344 | + log_error(sg, "cancel_prev_recovery: error %d", error); | |
13345 | + } | |
13346 | +} | |
13347 | + | |
13348 | +static void pre_recover_sg(sm_group_t *sg, recover_t *rev) | |
13349 | +{ | |
13350 | + if (sg->state == SGST_RECOVER) { | |
13351 | + cancel_prev_recovery(sg); | |
13352 | + list_del(&sg->recover_list); | |
13353 | + } | |
13354 | + | |
13355 | + sg->ops->stop(sg->service_data); | |
13356 | + sg->state = SGST_RECOVER; | |
13357 | + sg->recover_state = RECOVER_NONE; | |
13358 | + sg->recover_data = rev; | |
13359 | + list_add(&sg->recover_list, &rev->sgs[sg->level]); | |
13360 | +} | |
13361 | + | |
13362 | +/* | |
13363 | + * When adjust_members finds that some nodes are dead and mark_effected_sgs | |
13364 | + * finds that some SG's are effected by departed nodes, this is called to | |
13365 | + * collect together the SG's which need to be recovered. An revent (recovery | |
13366 | + * event) is the group of effected SG's. | |
13367 | + */ | |
13368 | + | |
13369 | +static int new_recovery(void) | |
13370 | +{ | |
13371 | + sm_group_t *sg; | |
13372 | + recover_t *rev; | |
13373 | + sm_node_t *node, *sgnode, *safe; | |
13374 | + int i; | |
13375 | + | |
13376 | + rev = alloc_recover(); | |
13377 | + list_add_tail(&rev->list, &recoveries); | |
13378 | + | |
13379 | + down(&sm_sglock); | |
13380 | + | |
13381 | + /* | |
13382 | + * Stop effected SG's and add them to the rev | |
13383 | + */ | |
13384 | + | |
13385 | + for (i = 0; i < SG_LEVELS; i++) { | |
13386 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
13387 | + if (test_and_clear_bit(SGFL_NEED_RECOVERY, &sg->flags)){ | |
13388 | + if (sg->state == SGST_JOIN) | |
13389 | + continue; | |
13390 | + pre_recover_sg(sg, rev); | |
13391 | + } | |
13392 | + } | |
13393 | + } | |
13394 | + | |
13395 | + /* | |
13396 | + * For an SG needing recovery, remove dead nodes from sg->memb list | |
13397 | + */ | |
13398 | + | |
13399 | + for (i = 0; i < SG_LEVELS; i++) { | |
13400 | + list_for_each_entry(sg, &rev->sgs[i], recover_list) { | |
13401 | + | |
13402 | + /* Remove dead members from SG's member list */ | |
13403 | + list_for_each_entry_safe(sgnode, safe, &sg->memb, list){ | |
13404 | + | |
13405 | + node = sm_find_member(sgnode->id); | |
13406 | + SM_ASSERT(node, printk("id %u\n", sgnode->id);); | |
13407 | + | |
13408 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)){ | |
13409 | + list_del(&sgnode->list); | |
13410 | + kfree(sgnode); | |
13411 | + sg->memb_count--; | |
13412 | + log_debug(sg, "remove node %u count %d", | |
13413 | + sgnode->id, sg->memb_count); | |
13414 | + } | |
13415 | + } | |
13416 | + } | |
13417 | + } | |
13418 | + | |
13419 | + up(&sm_sglock); | |
13420 | + rev->cur_level = 0; | |
13421 | + return 0; | |
13422 | +} | |
13423 | + | |
13424 | +/* | |
13425 | + * The NEED_RECOVERY bit on MML nodes is set in adjust_members() and is used in | |
13426 | + * mark_effected_sgs() and add_revent(). After that, we're done using the bit | |
13427 | + * and we clear it here. | |
13428 | + */ | |
13429 | + | |
13430 | +static void adjust_members_done(void) | |
13431 | +{ | |
13432 | + sm_node_t *node; | |
13433 | + | |
13434 | + list_for_each_entry(node, &sm_members, list) | |
13435 | + clear_bit(SNFL_NEED_RECOVERY, &node->flags); | |
13436 | +} | |
13437 | + | |
13438 | +/* | |
13439 | + * Start the service of the given SG. The service must be given an array of | |
13440 | + * nodeids specifying the new sg membership. The service is responsible to | |
13441 | + * free this chunk of memory when done with it. | |
13442 | + */ | |
13443 | + | |
13444 | +static void start_sg(sm_group_t *sg, uint32_t event_id) | |
13445 | +{ | |
13446 | + sm_node_t *node; | |
13447 | + uint32_t *memb; | |
13448 | + int count = 0; | |
13449 | + | |
13450 | + SM_RETRY(memb = kmalloc(sg->memb_count * sizeof(uint32_t), GFP_KERNEL), | |
13451 | + memb); | |
13452 | + | |
13453 | + list_for_each_entry(node, &sg->memb, list) | |
13454 | + memb[count++] = node->id; | |
13455 | + | |
13456 | + sg->ops->start(sg->service_data, memb, count, event_id, | |
13457 | + SERVICE_NODE_FAILED); | |
13458 | +} | |
13459 | + | |
13460 | +static void recovery_barrier(sm_group_t *sg) | |
13461 | +{ | |
13462 | + char bname[MAX_BARRIER_NAME_LEN]; | |
13463 | + int error, len; | |
13464 | + | |
13465 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
13466 | + | |
13467 | + /* bypass the barrier if we're the only member */ | |
13468 | + if (sg->memb_count == 1) { | |
13469 | + process_recovery_barrier(sg, 0); | |
13470 | + return; | |
13471 | + } | |
13472 | + | |
13473 | + len = snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.RECOV.%u", | |
13474 | + sg->global_id, sg->recover_stop, sg->memb_count); | |
13475 | + | |
13476 | + /* We save this barrier name so we can cancel it if needed. */ | |
13477 | + memset(sg->recover_barrier, 0, MAX_BARRIER_NAME_LEN); | |
13478 | + memcpy(sg->recover_barrier, bname, len); | |
13479 | + | |
13480 | + error = sm_barrier(bname, sg->memb_count, SM_BARRIER_RECOVERY); | |
13481 | + if (error) | |
13482 | + log_error(sg, "recovery_barrier error %d: %s", error, bname); | |
13483 | +} | |
13484 | + | |
13485 | +static void recover_sg(sm_group_t *sg, int event_id) | |
13486 | +{ | |
13487 | + log_debug(sg, "recover state %d", sg->recover_state); | |
13488 | + | |
13489 | + switch (sg->recover_state) { | |
13490 | + | |
13491 | + case RECOVER_NONE: | |
13492 | + /* must wait for recovery to stop sg on all nodes */ | |
13493 | + sg->recover_state = RECOVER_BARRIERWAIT; | |
13494 | + sg->recover_stop = 0; | |
13495 | + recovery_barrier(sg); | |
13496 | + break; | |
13497 | + | |
13498 | + case RECOVER_BARRIERWAIT: | |
13499 | + break; | |
13500 | + | |
13501 | + case RECOVER_STOP: | |
13502 | + /* barrier callback sets state STOP */ | |
13503 | + sg->recover_stop = 1; | |
13504 | + sg->recover_state = RECOVER_START; | |
13505 | + start_sg(sg, event_id); | |
13506 | + break; | |
13507 | + | |
13508 | + case RECOVER_START: | |
13509 | + break; | |
13510 | + | |
13511 | + case RECOVER_STARTDONE: | |
13512 | + /* service callback sets state STARTDONE */ | |
13513 | + sg->recover_state = RECOVER_BARRIERWAIT; | |
13514 | + recovery_barrier(sg); | |
13515 | + break; | |
13516 | + | |
13517 | + case RECOVER_BARRIERDONE: | |
13518 | + /* barrier callback sets state BARRIERDONE */ | |
13519 | + sg->ops->finish(sg->service_data, event_id); | |
13520 | + list_del(&sg->recover_list); | |
13521 | + sg->recover_state = RECOVER_NONE; | |
13522 | + sg->state = SGST_RUN; | |
13523 | + | |
13524 | + /* Continue a previous, interrupted attempt to leave the sg */ | |
13525 | + if (sg->sevent) { | |
b7b72b66 AM |
13526 | + sm_sevent_t *sev = sg->sevent; |
13527 | + log_debug(sg, "restart leave %lx", sev->se_flags); | |
13528 | + clear_bit(SEFL_DELAY_RECOVERY, &sev->se_flags); | |
13529 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
c1c6733f AM |
13530 | + wake_serviced(DO_JOINLEAVE); |
13531 | + } | |
13532 | + break; | |
13533 | + | |
13534 | + default: | |
13535 | + log_error(sg, "invalid recover_state %u", sg->recover_state); | |
13536 | + } | |
13537 | +} | |
13538 | + | |
13539 | +static void recover_level(recover_t *rev, int level) | |
13540 | +{ | |
13541 | + sm_group_t *sg, *safe; | |
13542 | + | |
13543 | + list_for_each_entry_safe(sg, safe, &rev->sgs[level], recover_list) | |
13544 | + recover_sg(sg, rev->event_id); | |
13545 | +} | |
13546 | + | |
13547 | +static void recover_levels(recover_t *rev) | |
13548 | +{ | |
13549 | + for (;;) { | |
13550 | + recover_level(rev, rev->cur_level); | |
13551 | + | |
13552 | + if (list_empty(&rev->sgs[rev->cur_level])) { | |
13553 | + if (rev->cur_level == SG_LEVELS - 1) { | |
13554 | + list_del(&rev->list); | |
13555 | + kfree(rev); | |
13556 | + return; | |
13557 | + } | |
13558 | + rev->cur_level++; | |
13559 | + continue; | |
13560 | + } | |
13561 | + break; | |
13562 | + } | |
13563 | +} | |
13564 | + | |
13565 | +/* | |
13566 | + * Called by SM thread when the cluster is quorate. It restarts | |
13567 | + * SG's that were stopped in new_recovery() due to a member death. | |
13568 | + * It waits for all SG's at level N to complete restart before | |
13569 | + * restarting SG's at level N+1. | |
13570 | + */ | |
13571 | + | |
13572 | +void process_recoveries(void) | |
13573 | +{ | |
13574 | + recover_t *rev, *safe; | |
13575 | + | |
13576 | + down(&sm_sglock); | |
13577 | + list_for_each_entry_safe(rev, safe, &recoveries, list) | |
13578 | + recover_levels(rev); | |
13579 | + up(&sm_sglock); | |
13580 | +} | |
13581 | + | |
13582 | +/* | |
13583 | + * The cnxman membership has changed. Check if there's still quorum and | |
13584 | + * whether any nodes have died. If nodes have died, initiate recovery on any | |
13585 | + * SG's they were in. This begins immediately if the cluster remains quorate; | |
13586 | + * if not this waits until the cluster regains quorum. | |
13587 | + */ | |
13588 | + | |
13589 | +void process_nodechange(void) | |
13590 | +{ | |
13591 | + int gone, effected; | |
13592 | + | |
13593 | + if ((sm_quorum = sm_quorum_next)) | |
13594 | + wake_serviced(DO_RUN); | |
13595 | + | |
13596 | + gone = adjust_members(); | |
13597 | + if (gone > 0) { | |
13598 | + effected = mark_effected_sgs(); | |
13599 | + | |
13600 | + backout_sevents(); | |
13601 | + cancel_uevents(&effected); | |
13602 | + | |
13603 | + if (effected > 0) { | |
13604 | + new_recovery(); | |
13605 | + wake_serviced(DO_RECOVERIES); | |
13606 | + } | |
13607 | + } | |
13608 | + adjust_members_done(); | |
13609 | +} | |
13610 | + | |
13611 | +int check_recovery(sm_group_t *sg, int event_id) | |
13612 | +{ | |
13613 | + if (sg->state == SGST_RECOVER) { | |
13614 | + recover_t *rev = (recover_t *) sg->recover_data; | |
13615 | + if (rev && rev->event_id == event_id) | |
13616 | + return 1; | |
13617 | + } | |
13618 | + return 0; | |
13619 | +} | |
13620 | + | |
13621 | +void process_recover_msg(sm_msg_t *smsg, uint32_t nodeid) | |
13622 | +{ | |
13623 | + sm_group_t *sg; | |
13624 | + recover_t *rev; | |
13625 | + | |
13626 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
13627 | + if (!sg) { | |
13628 | + log_print("process_recover_msg: unknown sg id %x", | |
13629 | + smsg->ms_global_sgid); | |
13630 | + return; | |
13631 | + } | |
13632 | + | |
13633 | + /* we already know about the recovery and can ignore the msg */ | |
13634 | + if (sg->state == SGST_RECOVER) | |
13635 | + return; | |
13636 | + | |
13637 | + if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
13638 | + /* we will initiate recovery on our own if we know about the | |
13639 | + uevent so we can ignore this */ | |
13640 | + log_debug(sg, "process_recover_msg: ignore from %u", nodeid); | |
13641 | + return; | |
13642 | + } | |
13643 | + | |
13644 | + log_debug(sg, "recovery initiated by msg from %u", nodeid); | |
13645 | + rev = alloc_recover(); | |
13646 | + list_add_tail(&rev->list, &recoveries); | |
13647 | + pre_recover_sg(sg, rev); | |
13648 | + wake_serviced(DO_RECOVERIES); | |
13649 | +} | |
13650 | diff -urN linux-orig/cluster/cman/sm_recover.h linux-patched/cluster/cman/sm_recover.h | |
bb1d8b11 AM |
13651 | --- linux-orig/cluster/cman/sm_recover.h 1970-01-01 07:30:00.000000000 +0730 |
13652 | +++ linux-patched/cluster/cman/sm_recover.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
13653 | @@ -0,0 +1,23 @@ |
13654 | +/****************************************************************************** | |
13655 | +******************************************************************************* | |
13656 | +** | |
13657 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13658 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13659 | +** | |
13660 | +** This copyrighted material is made available to anyone wishing to use, | |
13661 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13662 | +** of the GNU General Public License v.2. | |
13663 | +** | |
13664 | +******************************************************************************* | |
13665 | +******************************************************************************/ | |
13666 | + | |
13667 | +#ifndef __SM_RECOVER_DOT_H__ | |
13668 | +#define __SM_RECOVER_DOT_H__ | |
13669 | + | |
13670 | +void init_recovery(void); | |
13671 | +void process_recoveries(void); | |
13672 | +void process_nodechange(void); | |
13673 | +int check_recovery(sm_group_t *sg, int event_id); | |
13674 | +void process_recover_msg(sm_msg_t *smsg, uint32_t nodeid); | |
13675 | + | |
13676 | +#endif | |
13677 | diff -urN linux-orig/cluster/cman/sm_services.c linux-patched/cluster/cman/sm_services.c | |
bb1d8b11 AM |
13678 | --- linux-orig/cluster/cman/sm_services.c 1970-01-01 07:30:00.000000000 +0730 |
13679 | +++ linux-patched/cluster/cman/sm_services.c 2004-11-03 11:37:37.000000000 +0800 | |
b7b72b66 | 13680 | @@ -0,0 +1,426 @@ |
c1c6733f AM |
13681 | +/****************************************************************************** |
13682 | +******************************************************************************* | |
13683 | +** | |
13684 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13685 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13686 | +** | |
13687 | +** This copyrighted material is made available to anyone wishing to use, | |
13688 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13689 | +** of the GNU General Public License v.2. | |
13690 | +** | |
13691 | +******************************************************************************* | |
13692 | +******************************************************************************/ | |
13693 | + | |
13694 | +#include "sm.h" | |
13695 | + | |
13696 | +static struct list_head callbacks; | |
13697 | +static spinlock_t callback_lock; | |
13698 | +static struct list_head sg_registered[SG_LEVELS]; | |
13699 | + | |
13700 | +/* | |
13701 | + * These are the functions to register, join, leave, unregister, callback | |
13702 | + * with/to the sm. | |
13703 | + */ | |
13704 | + | |
13705 | +struct sc_entry { | |
13706 | + struct list_head list; | |
13707 | + uint32_t local_id; | |
13708 | + int event_id; | |
13709 | +}; | |
13710 | +typedef struct sc_entry sc_entry_t; | |
13711 | + | |
13712 | +void init_services(void) | |
13713 | +{ | |
13714 | + int i; | |
13715 | + | |
13716 | + INIT_LIST_HEAD(&callbacks); | |
13717 | + spin_lock_init(&callback_lock); | |
13718 | + | |
13719 | + for (i = 0; i < SG_LEVELS; i++) { | |
13720 | + INIT_LIST_HEAD(&sm_sg[i]); | |
13721 | + INIT_LIST_HEAD(&sg_registered[i]); | |
13722 | + } | |
13723 | + init_MUTEX(&sm_sglock); | |
13724 | +} | |
13725 | + | |
13726 | +/* Context: service */ | |
13727 | + | |
13728 | +int kcl_register_service(char *name, int namelen, int level, | |
13729 | + struct kcl_service_ops *ops, int unique, | |
13730 | + void *servicedata, uint32_t *service_id) | |
13731 | +{ | |
13732 | + sm_group_t *sg; | |
13733 | + int found = FALSE; | |
13734 | + int error = -EINVAL; | |
13735 | + | |
13736 | + if (level > SG_LEVELS - 1) | |
13737 | + goto fail; | |
13738 | + | |
13739 | + if (namelen > MAX_SERVICE_NAME_LEN) | |
13740 | + goto fail; | |
13741 | + | |
13742 | + error = kcl_addref_cluster(); | |
13743 | + if (error) | |
13744 | + goto fail; | |
13745 | + | |
13746 | + down(&sm_sglock); | |
13747 | + | |
13748 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
13749 | + if ((sg->namelen == namelen) && | |
13750 | + (!strncmp(sg->name, name, namelen))) { | |
13751 | + found = TRUE; | |
13752 | + goto next; | |
13753 | + } | |
13754 | + } | |
13755 | + | |
13756 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13757 | + if ((sg->namelen == namelen) && | |
13758 | + (!strncmp(sg->name, name, namelen))) { | |
13759 | + found = TRUE; | |
13760 | + goto next; | |
13761 | + } | |
13762 | + } | |
13763 | + | |
13764 | + next: | |
13765 | + | |
13766 | + if (found && unique) { | |
13767 | + error = -EEXIST; | |
13768 | + goto fail_unlock; | |
13769 | + } | |
13770 | + | |
13771 | + if (found) { | |
13772 | + sg->refcount++; | |
13773 | + goto out; | |
13774 | + } | |
13775 | + | |
13776 | + sg = (sm_group_t *) kmalloc(sizeof(sm_group_t) + namelen, GFP_KERNEL); | |
13777 | + if (!sg) { | |
13778 | + error = -ENOMEM; | |
13779 | + goto fail_unlock; | |
13780 | + } | |
13781 | + memset(sg, 0, sizeof(sm_group_t) + namelen); | |
13782 | + | |
13783 | + sg->refcount = 1; | |
13784 | + sg->service_data = servicedata; | |
13785 | + sg->ops = ops; | |
13786 | + sg->level = level; | |
13787 | + sg->namelen = namelen; | |
13788 | + memcpy(sg->name, name, namelen); | |
13789 | + sg->local_id = sm_new_local_id(level); | |
13790 | + sg->state = SGST_NONE; | |
13791 | + INIT_LIST_HEAD(&sg->memb); | |
13792 | + INIT_LIST_HEAD(&sg->joining); | |
13793 | + init_completion(&sg->event_comp); | |
13794 | + | |
13795 | + list_add_tail(&sg->list, &sg_registered[level]); | |
13796 | + | |
13797 | + out: | |
13798 | + *service_id = sg->local_id; | |
13799 | + up(&sm_sglock); | |
13800 | + return 0; | |
13801 | + | |
13802 | + fail_unlock: | |
13803 | + up(&sm_sglock); | |
13804 | + kcl_releaseref_cluster(); | |
13805 | + fail: | |
13806 | + return error; | |
13807 | +} | |
13808 | + | |
13809 | +/* Context: service */ | |
13810 | + | |
13811 | +void kcl_unregister_service(uint32_t local_id) | |
13812 | +{ | |
13813 | + sm_group_t *sg; | |
13814 | + int level = sm_id_to_level(local_id); | |
13815 | + | |
13816 | + down(&sm_sglock); | |
13817 | + | |
13818 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13819 | + if (sg->local_id == local_id) { | |
13820 | + SM_ASSERT(sg->refcount,); | |
13821 | + sg->refcount--; | |
13822 | + | |
13823 | + if (!sg->refcount) { | |
13824 | + list_del(&sg->list); | |
13825 | + kfree(sg); | |
13826 | + } | |
13827 | + kcl_releaseref_cluster(); | |
13828 | + break; | |
13829 | + } | |
13830 | + } | |
13831 | + up(&sm_sglock); | |
13832 | +} | |
13833 | + | |
13834 | +/* Context: service */ | |
13835 | + | |
13836 | +int kcl_join_service(uint32_t local_id) | |
13837 | +{ | |
13838 | + sm_group_t *sg; | |
13839 | + sm_sevent_t *sev; | |
13840 | + int level = sm_id_to_level(local_id); | |
13841 | + int error, found = FALSE; | |
13842 | + | |
13843 | + down(&sm_sglock); | |
13844 | + | |
13845 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13846 | + if (sg->local_id == local_id) { | |
13847 | + found = TRUE; | |
13848 | + break; | |
13849 | + } | |
13850 | + } | |
13851 | + | |
13852 | + if (!found) { | |
13853 | + up(&sm_sglock); | |
13854 | + error = -ENOENT; | |
13855 | + goto out; | |
13856 | + } | |
13857 | + | |
13858 | + if (sg->state != SGST_NONE) { | |
13859 | + up(&sm_sglock); | |
13860 | + error = -EINVAL; | |
13861 | + goto out; | |
13862 | + } | |
13863 | + | |
c1c6733f AM |
13864 | + sev = kmalloc(sizeof(sm_sevent_t), GFP_KERNEL); |
13865 | + if (!sev) { | |
b7b72b66 | 13866 | + up(&sm_sglock); |
c1c6733f AM |
13867 | + error = -ENOMEM; |
13868 | + goto out; | |
13869 | + } | |
13870 | + | |
13871 | + memset(sev, 0, sizeof (sm_sevent_t)); | |
13872 | + sev->se_state = SEST_JOIN_BEGIN; | |
b7b72b66 | 13873 | + sm_set_event_id(&sev->se_id); |
c1c6733f AM |
13874 | + sev->se_sg = sg; |
13875 | + sg->sevent = sev; | |
b7b72b66 AM |
13876 | + sg->state = SGST_JOIN; |
13877 | + set_bit(SGFL_SEVENT, &sg->flags); | |
13878 | + list_del(&sg->list); | |
13879 | + list_add_tail(&sg->list, &sm_sg[sg->level]); | |
13880 | + | |
13881 | + up(&sm_sglock); | |
13882 | + | |
13883 | + /* | |
13884 | + * The join is a service event which will be processed asynchronously. | |
13885 | + */ | |
c1c6733f AM |
13886 | + |
13887 | + new_joinleave(sev); | |
13888 | + wait_for_completion(&sg->event_comp); | |
13889 | + error = 0; | |
13890 | + | |
13891 | + out: | |
13892 | + return error; | |
13893 | +} | |
13894 | + | |
13895 | +/* Context: service */ | |
13896 | + | |
13897 | +int kcl_leave_service(uint32_t local_id) | |
13898 | +{ | |
13899 | + sm_group_t *sg = NULL; | |
13900 | + sm_sevent_t *sev; | |
13901 | + int error; | |
13902 | + | |
13903 | + error = -ENOENT; | |
13904 | + sg = sm_local_id_to_sg(local_id); | |
13905 | + if (!sg) | |
13906 | + goto out; | |
13907 | + | |
13908 | + /* sg was never joined */ | |
13909 | + error = -EINVAL; | |
13910 | + if (sg->state == SGST_NONE) | |
13911 | + goto out; | |
13912 | + | |
b7b72b66 AM |
13913 | + down(&sm_sglock); |
13914 | + | |
c1c6733f | 13915 | + /* may still be joining */ |
b7b72b66 AM |
13916 | + if (test_and_set_bit(SGFL_SEVENT, &sg->flags)) { |
13917 | + up(&sm_sglock); | |
13918 | + error = -EBUSY; | |
c1c6733f | 13919 | + goto out; |
b7b72b66 | 13920 | + } |
c1c6733f | 13921 | + |
c1c6733f | 13922 | + sev = kmalloc(sizeof(sm_sevent_t), GFP_KERNEL); |
b7b72b66 AM |
13923 | + if (!sev) { |
13924 | + up(&sm_sglock); | |
13925 | + error = -ENOMEM; | |
c1c6733f | 13926 | + goto out; |
b7b72b66 | 13927 | + } |
c1c6733f AM |
13928 | + |
13929 | + memset(sev, 0, sizeof (sm_sevent_t)); | |
13930 | + sev->se_state = SEST_LEAVE_BEGIN; | |
b7b72b66 | 13931 | + sm_set_event_id(&sev->se_id); |
c1c6733f AM |
13932 | + set_bit(SEFL_LEAVE, &sev->se_flags); |
13933 | + sev->se_sg = sg; | |
13934 | + sg->sevent = sev; | |
b7b72b66 AM |
13935 | + |
13936 | + up(&sm_sglock); | |
c1c6733f AM |
13937 | + |
13938 | + new_joinleave(sev); | |
13939 | + wait_for_completion(&sg->event_comp); | |
13940 | + error = 0; | |
13941 | + | |
13942 | + down(&sm_sglock); | |
13943 | + list_del(&sg->list); | |
13944 | + list_add_tail(&sg->list, &sg_registered[sg->level]); | |
13945 | + up(&sm_sglock); | |
13946 | + | |
13947 | + out: | |
13948 | + return error; | |
13949 | +} | |
13950 | + | |
13951 | +static void process_callback(uint32_t local_id, int event_id) | |
13952 | +{ | |
13953 | + sm_group_t *sg; | |
13954 | + sm_sevent_t *sev; | |
13955 | + sm_uevent_t *uev; | |
13956 | + | |
13957 | + sg = sm_local_id_to_sg(local_id); | |
13958 | + if (!sg) | |
13959 | + return; | |
13960 | + | |
13961 | + if (sg->state == SGST_RECOVER) { | |
13962 | + if (!check_recovery(sg, event_id)) { | |
13963 | + log_error(sg, "process_callback invalid recover " | |
13964 | + "event id %d", event_id); | |
13965 | + return; | |
13966 | + } | |
13967 | + | |
13968 | + if (sg->recover_state == RECOVER_START) | |
13969 | + sg->recover_state = RECOVER_STARTDONE; | |
13970 | + else | |
13971 | + log_error(sg, "process_callback recover state %u", | |
13972 | + sg->recover_state); | |
13973 | + wake_serviced(DO_RECOVERIES); | |
13974 | + } | |
13975 | + | |
13976 | + else if (test_bit(SGFL_SEVENT, &sg->flags) && sg->sevent && | |
13977 | + (sg->sevent->se_id == event_id)) { | |
13978 | + sev = sg->sevent; | |
13979 | + | |
13980 | + if (test_and_clear_bit(SEFL_ALLOW_STARTDONE, &sev->se_flags) && | |
13981 | + (sev->se_state == SEST_JSTART_SERVICEWAIT)) | |
13982 | + sev->se_state = SEST_JSTART_SERVICEDONE; | |
13983 | + | |
13984 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
13985 | + wake_serviced(DO_JOINLEAVE); | |
13986 | + } | |
13987 | + | |
13988 | + else if (test_bit(SGFL_UEVENT, &sg->flags) && | |
13989 | + (sg->uevent.ue_id == event_id)) { | |
13990 | + uev = &sg->uevent; | |
13991 | + | |
13992 | + if (test_and_clear_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags)) { | |
13993 | + if (uev->ue_state == UEST_JSTART_SERVICEWAIT) | |
13994 | + uev->ue_state = UEST_JSTART_SERVICEDONE; | |
13995 | + else if (uev->ue_state == UEST_LSTART_SERVICEWAIT) | |
13996 | + uev->ue_state = UEST_LSTART_SERVICEDONE; | |
13997 | + } | |
13998 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
13999 | + wake_serviced(DO_MEMBERSHIP); | |
14000 | + } | |
14001 | + | |
14002 | + else | |
14003 | + log_error(sg, "ignoring service callback id=%x event=%u", | |
14004 | + local_id, event_id); | |
14005 | +} | |
14006 | + | |
14007 | +void process_callbacks(void) | |
14008 | +{ | |
14009 | + sc_entry_t *se; | |
14010 | + | |
14011 | + while (1) { | |
14012 | + se = NULL; | |
14013 | + | |
14014 | + spin_lock(&callback_lock); | |
14015 | + if (!list_empty(&callbacks)) { | |
14016 | + se = list_entry(callbacks.next, sc_entry_t, list); | |
14017 | + list_del(&se->list); | |
14018 | + } | |
14019 | + spin_unlock(&callback_lock); | |
14020 | + | |
14021 | + if (!se) | |
14022 | + break; | |
14023 | + process_callback(se->local_id, se->event_id); | |
14024 | + kfree(se); | |
14025 | + schedule(); | |
14026 | + } | |
14027 | +} | |
14028 | + | |
14029 | +/* Context: service */ | |
14030 | + | |
14031 | +void kcl_start_done(uint32_t local_id, int event_id) | |
14032 | +{ | |
14033 | + sc_entry_t *se; | |
14034 | + | |
14035 | + SM_RETRY(se = kmalloc(sizeof(sc_entry_t), GFP_KERNEL), se); | |
14036 | + | |
14037 | + se->local_id = local_id; | |
14038 | + se->event_id = event_id; | |
14039 | + | |
14040 | + spin_lock(&callback_lock); | |
14041 | + list_add_tail(&se->list, &callbacks); | |
14042 | + spin_unlock(&callback_lock); | |
14043 | + | |
14044 | + wake_serviced(DO_CALLBACKS); | |
14045 | +} | |
14046 | + | |
14047 | +/* Context: service */ | |
14048 | + | |
14049 | +void kcl_global_service_id(uint32_t local_id, uint32_t *global_id) | |
14050 | +{ | |
14051 | + sm_group_t *sg = sm_local_id_to_sg(local_id); | |
14052 | + | |
14053 | + if (!sg) | |
14054 | + log_print("kcl_global_service_id: can't find %x", local_id); | |
14055 | + else | |
14056 | + *global_id = sg->global_id; | |
14057 | +} | |
14058 | + | |
14059 | +static void copy_to_service(sm_group_t *sg, struct kcl_service *s) | |
14060 | +{ | |
14061 | + s->level = sg->level; | |
14062 | + s->local_id = sg->local_id; | |
14063 | + s->global_id = sg->global_id; | |
14064 | + s->node_count = sg->memb_count; | |
14065 | + strcpy(s->name, sg->name); | |
14066 | +} | |
14067 | + | |
14068 | +int kcl_get_services(struct list_head *head, int level) | |
14069 | +{ | |
14070 | + sm_group_t *sg; | |
14071 | + struct kcl_service *s; | |
14072 | + int error = -ENOMEM, count = 0; | |
14073 | + | |
14074 | + down(&sm_sglock); | |
14075 | + | |
14076 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
14077 | + if (head) { | |
14078 | + s = kmalloc(sizeof(struct kcl_service), GFP_KERNEL); | |
14079 | + if (!s) | |
14080 | + goto out; | |
14081 | + copy_to_service(sg, s); | |
14082 | + list_add(&s->list, head); | |
14083 | + } | |
14084 | + count++; | |
14085 | + } | |
14086 | + | |
14087 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
14088 | + if (head) { | |
14089 | + s = kmalloc(sizeof(struct kcl_service), GFP_KERNEL); | |
14090 | + if (!s) | |
14091 | + goto out; | |
14092 | + copy_to_service(sg, s); | |
14093 | + list_add(&s->list, head); | |
14094 | + } | |
14095 | + count++; | |
14096 | + } | |
14097 | + | |
14098 | + error = count; | |
14099 | + out: | |
14100 | + up(&sm_sglock); | |
14101 | + return error; | |
14102 | +} | |
14103 | + | |
14104 | +/* These three global variables listed in extern form in sm.h. */ | |
14105 | +struct list_head sm_sg[SG_LEVELS]; | |
14106 | +struct semaphore sm_sglock; | |
14107 | diff -urN linux-orig/cluster/cman/sm_services.h linux-patched/cluster/cman/sm_services.h | |
bb1d8b11 AM |
14108 | --- linux-orig/cluster/cman/sm_services.h 1970-01-01 07:30:00.000000000 +0730 |
14109 | +++ linux-patched/cluster/cman/sm_services.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
14110 | @@ -0,0 +1,20 @@ |
14111 | +/****************************************************************************** | |
14112 | +******************************************************************************* | |
14113 | +** | |
14114 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14115 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14116 | +** | |
14117 | +** This copyrighted material is made available to anyone wishing to use, | |
14118 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14119 | +** of the GNU General Public License v.2. | |
14120 | +** | |
14121 | +******************************************************************************* | |
14122 | +******************************************************************************/ | |
14123 | + | |
14124 | +#ifndef __SM_SERVICES_DOT_H__ | |
14125 | +#define __SM_SERVICES_DOT_H__ | |
14126 | + | |
14127 | +void init_services(void); | |
14128 | +void process_callbacks(void); | |
14129 | + | |
14130 | +#endif | |
14131 | diff -urN linux-orig/cluster/cman/sm_user.c linux-patched/cluster/cman/sm_user.c | |
bb1d8b11 AM |
14132 | --- linux-orig/cluster/cman/sm_user.c 1970-01-01 07:30:00.000000000 +0730 |
14133 | +++ linux-patched/cluster/cman/sm_user.c 2004-11-03 11:37:37.000000000 +0800 | |
5a2052f6 | 14134 | @@ -0,0 +1,569 @@ |
c1c6733f AM |
14135 | +/****************************************************************************** |
14136 | +******************************************************************************* | |
14137 | +** | |
14138 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14139 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14140 | +** | |
14141 | +** This copyrighted material is made available to anyone wishing to use, | |
14142 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14143 | +** of the GNU General Public License v.2. | |
14144 | +** | |
14145 | +******************************************************************************* | |
14146 | +******************************************************************************/ | |
14147 | + | |
14148 | +#include "sm.h" | |
14149 | +#include "cnxman-private.h" | |
14150 | + | |
14151 | +void copy_to_usernode(struct cluster_node *node, struct cl_cluster_node *unode); | |
14152 | + | |
14153 | +#define UST_REGISTER 1 | |
14154 | +#define UST_UNREGISTER 2 | |
14155 | +#define UST_JOIN 3 | |
14156 | +#define UST_LEAVE 4 | |
14157 | +#define UST_JOINED 5 | |
14158 | + | |
14159 | +struct event { | |
14160 | + struct list_head list; | |
14161 | + service_event_t type; | |
14162 | + service_start_t start_type; | |
14163 | + unsigned int event_id; | |
14164 | + unsigned int last_stop; | |
14165 | + unsigned int last_start; | |
14166 | + unsigned int last_finish; | |
14167 | + unsigned int node_count; | |
14168 | + uint32_t * nodeids; | |
14169 | +}; | |
14170 | +typedef struct event event_t; | |
14171 | + | |
14172 | +struct user_service { | |
14173 | + uint32_t local_id; | |
14174 | + pid_t pid; | |
14175 | + int signal; | |
14176 | + struct socket * sock; | |
14177 | + uint8_t state; | |
14178 | + uint8_t async; | |
14179 | + struct semaphore lock; | |
14180 | + struct list_head events; | |
14181 | + spinlock_t event_lock; | |
14182 | + unsigned int last_stop; | |
14183 | + unsigned int last_start; | |
14184 | + unsigned int last_finish; | |
14185 | + unsigned int need_startdone; | |
14186 | + unsigned int node_count; | |
14187 | + uint32_t * nodeids; | |
14188 | + int name_len; | |
14189 | + char name[MAX_SERVICE_NAME_LEN]; | |
14190 | +}; | |
14191 | +typedef struct user_service user_service_t; | |
14192 | + | |
14193 | + | |
14194 | +static void add_event(user_service_t *us, event_t *ev) | |
14195 | +{ | |
14196 | + spin_lock(&us->event_lock); | |
14197 | + list_add_tail(&ev->list, &us->events); | |
14198 | + | |
14199 | + switch(ev->type) { | |
14200 | + case SERVICE_EVENT_STOP: | |
14201 | + us->last_stop = us->last_start; | |
14202 | + break; | |
14203 | + case SERVICE_EVENT_START: | |
14204 | + us->last_start = ev->event_id; | |
14205 | + break; | |
14206 | + case SERVICE_EVENT_FINISH: | |
14207 | + us->last_finish = ev->event_id; | |
14208 | + break; | |
14209 | + case SERVICE_EVENT_LEAVEDONE: | |
14210 | + break; | |
14211 | + } | |
14212 | + spin_unlock(&us->event_lock); | |
14213 | +} | |
14214 | + | |
14215 | +static event_t *get_event(user_service_t *us) | |
14216 | +{ | |
14217 | + event_t *ev = NULL; | |
14218 | + | |
14219 | + spin_lock(&us->event_lock); | |
14220 | + if (!list_empty(&us->events)) { | |
14221 | + ev = list_entry(us->events.next, event_t, list); | |
14222 | + ev->last_stop = us->last_stop; | |
14223 | + ev->last_start = us->last_start; | |
14224 | + ev->last_finish = us->last_finish; | |
14225 | + } | |
14226 | + spin_unlock(&us->event_lock); | |
14227 | + return ev; | |
14228 | +} | |
14229 | + | |
14230 | +static void del_event(user_service_t *us, event_t *ev) | |
14231 | +{ | |
14232 | + spin_lock(&us->event_lock); | |
14233 | + list_del(&ev->list); | |
14234 | + spin_unlock(&us->event_lock); | |
14235 | +} | |
14236 | + | |
14237 | +static event_t *alloc_event(void) | |
14238 | +{ | |
14239 | + event_t *ev; | |
14240 | + SM_RETRY(ev = (event_t *) kmalloc(sizeof(event_t), GFP_KERNEL), ev); | |
14241 | + memset(ev, 0, sizeof(event_t)); | |
14242 | + return ev; | |
14243 | +} | |
14244 | + | |
14245 | +/* us->lock must be held before calling */ | |
14246 | +static void user_notify(user_service_t *us) | |
14247 | +{ | |
14248 | + if (us->sock) | |
14249 | + queue_oob_skb(us->sock, CLUSTER_OOB_MSG_SERVICEEVENT); | |
14250 | + if (us->pid && us->signal) | |
14251 | + kill_proc(us->pid, us->signal, 0); | |
14252 | +} | |
14253 | + | |
14254 | +static service_start_t start_type(int type) | |
14255 | +{ | |
14256 | + switch (type) { | |
14257 | + case SERVICE_NODE_FAILED: | |
14258 | + return SERVICE_START_FAILED; | |
14259 | + case SERVICE_NODE_JOIN: | |
14260 | + return SERVICE_START_JOIN; | |
14261 | + case SERVICE_NODE_LEAVE: | |
14262 | + return SERVICE_START_LEAVE; | |
14263 | + } | |
14264 | + return 0; | |
14265 | +} | |
14266 | + | |
14267 | +static int user_stop(void *servicedata) | |
14268 | +{ | |
14269 | + user_service_t *us = (user_service_t *) servicedata; | |
14270 | + event_t *ev; | |
14271 | + | |
14272 | + down(&us->lock); | |
14273 | + if (!us->sock) | |
14274 | + goto out; | |
14275 | + | |
14276 | + ev = alloc_event(); | |
14277 | + ev->type = SERVICE_EVENT_STOP; | |
14278 | + | |
14279 | + add_event(us, ev); | |
14280 | + user_notify(us); | |
14281 | + out: | |
14282 | + up(&us->lock); | |
14283 | + return 0; | |
14284 | +} | |
14285 | + | |
14286 | +static int user_start(void *servicedata, uint32_t *nodeids, int count, | |
14287 | + int event_id, int type) | |
14288 | +{ | |
14289 | + user_service_t *us = (user_service_t *) servicedata; | |
14290 | + event_t *ev; | |
14291 | + | |
14292 | + down(&us->lock); | |
14293 | + if (!us->sock) { | |
14294 | + kcl_start_done(us->local_id, event_id); | |
14295 | + goto out; | |
14296 | + } | |
14297 | + | |
14298 | + us->need_startdone = event_id; | |
14299 | + | |
14300 | + ev = alloc_event(); | |
14301 | + ev->type = SERVICE_EVENT_START; | |
14302 | + ev->node_count = count; | |
14303 | + ev->start_type = start_type(type); | |
14304 | + ev->event_id = event_id; | |
14305 | + ev->nodeids = nodeids; | |
14306 | + | |
14307 | + add_event(us, ev); | |
14308 | + user_notify(us); | |
14309 | + out: | |
14310 | + up(&us->lock); | |
14311 | + return 0; | |
14312 | +} | |
14313 | + | |
14314 | +static void user_finish(void *servicedata, int event_id) | |
14315 | +{ | |
14316 | + user_service_t *us = (user_service_t *) servicedata; | |
14317 | + event_t *ev; | |
14318 | + | |
14319 | + down(&us->lock); | |
14320 | + if (!us->sock) | |
14321 | + goto out; | |
14322 | + | |
14323 | + ev = alloc_event(); | |
14324 | + ev->type = SERVICE_EVENT_FINISH; | |
14325 | + ev->event_id = event_id; | |
14326 | + | |
14327 | + add_event(us, ev); | |
14328 | + user_notify(us); | |
14329 | + out: | |
14330 | + up(&us->lock); | |
14331 | +} | |
14332 | + | |
14333 | +struct kcl_service_ops user_service_ops = { | |
14334 | + .stop = user_stop, | |
14335 | + .start = user_start, | |
14336 | + .finish = user_finish | |
14337 | +}; | |
14338 | + | |
5a2052f6 | 14339 | +static int user_register(char *u_name, user_service_t **us_data) |
c1c6733f AM |
14340 | +{ |
14341 | + user_service_t *us; | |
5a2052f6 AM |
14342 | + char name[MAX_SERVICE_NAME_LEN+1]; |
14343 | + int len, error; | |
14344 | + | |
14345 | + memset(name, 0, MAX_SERVICE_NAME_LEN+1); | |
14346 | + | |
14347 | + if (copy_from_user(&name, u_name, MAX_SERVICE_NAME_LEN)) | |
14348 | + return -EFAULT; | |
c1c6733f | 14349 | + |
5a2052f6 AM |
14350 | + len = strlen(name); |
14351 | + if (len > MAX_SERVICE_NAME_LEN) | |
c1c6733f AM |
14352 | + return -ENAMETOOLONG; |
14353 | + if (!len) | |
14354 | + return -EINVAL; | |
14355 | + | |
14356 | + us = kmalloc(sizeof(user_service_t), GFP_KERNEL); | |
14357 | + if (!us) | |
14358 | + return -ENOMEM; | |
14359 | + memset(us, 0, sizeof(user_service_t)); | |
14360 | + us->nodeids = NULL; | |
14361 | + INIT_LIST_HEAD(&us->events); | |
14362 | + spin_lock_init(&us->event_lock); | |
14363 | + init_MUTEX(&us->lock); | |
14364 | + us->name_len = len; | |
14365 | + memcpy(us->name, name, len); | |
14366 | + | |
14367 | + error = kcl_register_service(name, len, SERVICE_LEVEL_USER, | |
14368 | + &user_service_ops, TRUE, (void *) us, | |
14369 | + &us->local_id); | |
14370 | + if (error) { | |
14371 | + kfree(us); | |
14372 | + us = NULL; | |
14373 | + } | |
14374 | + *us_data = us; | |
14375 | + return error; | |
14376 | +} | |
14377 | + | |
14378 | +static void user_unregister(user_service_t *us) | |
14379 | +{ | |
14380 | + event_t *ev; | |
14381 | + | |
14382 | + kcl_unregister_service(us->local_id); | |
14383 | + | |
14384 | + if (us->nodeids) | |
14385 | + kfree(us->nodeids); | |
14386 | + | |
14387 | + while ((ev = get_event(us))) { | |
14388 | + del_event(us, ev); | |
14389 | + if (ev->nodeids) | |
14390 | + kfree(ev->nodeids); | |
14391 | + kfree(ev); | |
14392 | + } | |
14393 | +} | |
14394 | + | |
14395 | +static int user_join_async(void *arg) | |
14396 | +{ | |
14397 | + user_service_t *us = arg; | |
14398 | + int user_gone = 0; | |
14399 | + | |
14400 | + daemonize("cman_userjoin"); | |
14401 | + | |
14402 | + kcl_join_service(us->local_id); | |
14403 | + | |
14404 | + down(&us->lock); | |
14405 | + us->state = UST_JOINED; | |
14406 | + us->async = 0; | |
14407 | + if (!us->sock) { | |
14408 | + if (us->need_startdone) | |
14409 | + kcl_start_done(us->local_id, us->need_startdone); | |
14410 | + user_gone = 1; | |
14411 | + } | |
14412 | + up(&us->lock); | |
14413 | + | |
14414 | + if (user_gone) { | |
14415 | + kcl_leave_service(us->local_id); | |
14416 | + user_unregister(us); | |
14417 | + kfree(us); | |
14418 | + } | |
14419 | + return 0; | |
14420 | +} | |
14421 | + | |
14422 | +static int user_leave_async(void *arg) | |
14423 | +{ | |
14424 | + user_service_t *us = arg; | |
14425 | + | |
14426 | + daemonize("cman_userleave"); | |
14427 | + | |
14428 | + kcl_leave_service(us->local_id); | |
14429 | + | |
14430 | + down(&us->lock); | |
14431 | + us->async = 0; | |
14432 | + if (!us->sock) { | |
14433 | + user_unregister(us); | |
14434 | + kfree(us); | |
14435 | + } else { | |
14436 | + event_t *ev = alloc_event(); | |
14437 | + ev->type = SERVICE_EVENT_LEAVEDONE; | |
14438 | + add_event(us, ev); | |
14439 | + user_notify(us); | |
14440 | + up(&us->lock); | |
14441 | + } | |
14442 | + | |
14443 | + return 0; | |
14444 | +} | |
14445 | + | |
14446 | +static int user_join(user_service_t *us, int wait) | |
14447 | +{ | |
14448 | + int error = 0; | |
14449 | + | |
14450 | + if (wait) { | |
14451 | + error = kcl_join_service(us->local_id); | |
14452 | + us->state = UST_JOINED; | |
14453 | + } | |
14454 | + else { | |
14455 | + us->async = 1; | |
14456 | + kernel_thread(user_join_async, us, 0); | |
14457 | + } | |
14458 | + | |
14459 | + return error; | |
14460 | +} | |
14461 | + | |
14462 | +static void user_leave(user_service_t *us, int wait) | |
14463 | +{ | |
14464 | + if (wait) | |
14465 | + kcl_leave_service(us->local_id); | |
14466 | + else { | |
14467 | + us->async = 1; | |
14468 | + kernel_thread(user_leave_async, us, 0); | |
14469 | + } | |
14470 | +} | |
14471 | + | |
14472 | +static int user_start_done(user_service_t *us, unsigned int event_id) | |
14473 | +{ | |
14474 | + if (!us->need_startdone) | |
14475 | + return -EINVAL; | |
14476 | + if (us->need_startdone == event_id) | |
14477 | + us->need_startdone = 0; | |
14478 | + kcl_start_done(us->local_id, event_id); | |
14479 | + return 0; | |
14480 | +} | |
14481 | + | |
14482 | +static void user_set_signal(user_service_t *us, int signal) | |
14483 | +{ | |
14484 | + us->pid = current->pid; | |
14485 | + us->signal = signal; | |
14486 | +} | |
14487 | + | |
14488 | +static int user_get_event(user_service_t *us, | |
14489 | + struct cl_service_event *user_event) | |
14490 | +{ | |
14491 | + event_t *ev; | |
14492 | + struct cl_service_event event; | |
14493 | + | |
14494 | + ev = get_event(us); | |
14495 | + if (!ev) | |
14496 | + return 0; | |
14497 | + | |
14498 | + event.type = ev->type; | |
14499 | + event.start_type = ev->start_type; | |
14500 | + event.event_id = ev->event_id; | |
14501 | + event.last_stop = ev->last_stop; | |
14502 | + event.last_start = ev->last_start; | |
14503 | + event.last_finish = ev->last_finish; | |
14504 | + event.node_count = ev->node_count; | |
14505 | + | |
14506 | + if (copy_to_user(user_event, &event, sizeof(struct cl_service_event))) | |
14507 | + return -EFAULT; | |
14508 | + | |
14509 | + del_event(us, ev); | |
14510 | + | |
14511 | + if (ev->type == SERVICE_EVENT_START) { | |
14512 | + if (us->nodeids) | |
14513 | + kfree(us->nodeids); | |
14514 | + us->nodeids = ev->nodeids; | |
14515 | + us->node_count = ev->node_count; | |
14516 | + } | |
14517 | + | |
14518 | + kfree(ev); | |
14519 | + return 1; | |
14520 | +} | |
14521 | + | |
14522 | +static int user_get_members(user_service_t *us, | |
14523 | + struct cl_cluster_nodelist *u_nodelist) | |
14524 | +{ | |
14525 | + struct cl_cluster_nodelist user_nodelist; | |
14526 | + struct cl_cluster_node user_node, *u_node; | |
14527 | + struct cluster_node *node; | |
14528 | + unsigned int i; | |
14529 | + int num_nodes = 0; | |
14530 | + | |
14531 | + if (!u_nodelist) | |
14532 | + return us->node_count; | |
14533 | + | |
14534 | + if (copy_from_user(&user_nodelist, (void __user *) u_nodelist, | |
14535 | + sizeof(struct cl_cluster_nodelist))) | |
14536 | + return -EFAULT; | |
14537 | + | |
14538 | + if (user_nodelist.max_members < us->node_count) | |
14539 | + return -E2BIG; | |
14540 | + | |
14541 | + u_node = user_nodelist.nodes; | |
14542 | + | |
14543 | + for (i = 0; i < us->node_count; i++) { | |
14544 | + node = find_node_by_nodeid(us->nodeids[i]); | |
14545 | + if (!node) | |
14546 | + continue; | |
14547 | + | |
14548 | + copy_to_usernode(node, &user_node); | |
14549 | + if (copy_to_user(u_node, &user_node, | |
14550 | + sizeof(struct cl_cluster_node))) | |
14551 | + return -EFAULT; | |
14552 | + | |
14553 | + u_node++; | |
14554 | + num_nodes++; | |
14555 | + } | |
14556 | + return num_nodes; | |
14557 | +} | |
14558 | + | |
14559 | +static int user_global_id(user_service_t *us, uint32_t *id) | |
14560 | +{ | |
14561 | + uint32_t gid = 0; | |
14562 | + | |
14563 | + if (us->state != UST_JOINED) | |
14564 | + return -EINVAL; | |
14565 | + | |
14566 | + kcl_global_service_id(us->local_id, &gid); | |
14567 | + | |
14568 | + if (copy_to_user(id, &gid, sizeof(uint32_t))) | |
14569 | + return -EFAULT; | |
14570 | + return 0; | |
14571 | +} | |
14572 | + | |
14573 | +static int user_set_level(user_service_t *us, int level) | |
14574 | +{ | |
14575 | + int prev_id = us->local_id; | |
14576 | + int error; | |
14577 | + | |
14578 | + if (us->state != UST_REGISTER) | |
14579 | + return -EINVAL; | |
14580 | + | |
14581 | + error = kcl_register_service(us->name, us->name_len, level, | |
14582 | + &user_service_ops, TRUE, (void *) us, | |
14583 | + &us->local_id); | |
14584 | + if (error) | |
14585 | + return error; | |
14586 | + | |
14587 | + kcl_unregister_service(prev_id); | |
14588 | + return 0; | |
14589 | +} | |
14590 | + | |
14591 | +int sm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) | |
14592 | +{ | |
14593 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
14594 | + user_service_t *us = c->service_data; | |
14595 | + int error = 0; | |
14596 | + | |
14597 | + if (!us && cmd != SIOCCLUSTER_SERVICE_REGISTER) | |
14598 | + return -EINVAL; | |
14599 | + | |
14600 | + switch (cmd) { | |
14601 | + case SIOCCLUSTER_SERVICE_REGISTER: | |
14602 | + error = user_register((char *) arg, &us); | |
14603 | + if (!error) { | |
14604 | + us->state = UST_REGISTER; | |
14605 | + us->sock = sock; | |
14606 | + c->service_data = us; | |
14607 | + } | |
14608 | + break; | |
14609 | + | |
14610 | + case SIOCCLUSTER_SERVICE_UNREGISTER: | |
14611 | + down(&us->lock); | |
14612 | + us->state = UST_UNREGISTER; | |
14613 | + user_unregister(us); | |
14614 | + up(&us->lock); | |
14615 | + break; | |
14616 | + | |
14617 | + case SIOCCLUSTER_SERVICE_JOIN: | |
14618 | + us->state = UST_JOIN; | |
14619 | + user_join(us, 0); | |
14620 | + break; | |
14621 | + | |
14622 | + case SIOCCLUSTER_SERVICE_LEAVE: | |
14623 | + down(&us->lock); | |
14624 | + if (us->state != UST_JOINED) { | |
14625 | + error = -EBUSY; | |
14626 | + up(&us->lock); | |
14627 | + } else { | |
14628 | + us->state = UST_LEAVE; | |
14629 | + up(&us->lock); | |
14630 | + user_leave(us, 0); | |
14631 | + } | |
14632 | + break; | |
14633 | + | |
14634 | + case SIOCCLUSTER_SERVICE_SETSIGNAL: | |
14635 | + user_set_signal(us, (int) arg); | |
14636 | + break; | |
14637 | + | |
14638 | + case SIOCCLUSTER_SERVICE_STARTDONE: | |
14639 | + error = user_start_done(us, (unsigned int) arg); | |
14640 | + break; | |
14641 | + | |
14642 | + case SIOCCLUSTER_SERVICE_GETEVENT: | |
14643 | + error = user_get_event(us, (struct cl_service_event *) arg); | |
14644 | + break; | |
14645 | + | |
14646 | + case SIOCCLUSTER_SERVICE_GETMEMBERS: | |
14647 | + error = user_get_members(us, (struct cl_cluster_nodelist *)arg); | |
14648 | + break; | |
14649 | + | |
14650 | + case SIOCCLUSTER_SERVICE_GLOBALID: | |
14651 | + error = user_global_id(us, (uint32_t *) arg); | |
14652 | + break; | |
14653 | + | |
14654 | + case SIOCCLUSTER_SERVICE_SETLEVEL: | |
14655 | + error = user_set_level(us, (int) arg); | |
14656 | + break; | |
14657 | + | |
14658 | + default: | |
14659 | + error = -EINVAL; | |
14660 | + } | |
14661 | + | |
14662 | + return error; | |
14663 | +} | |
14664 | + | |
14665 | +void sm_sock_release(struct socket *sock) | |
14666 | +{ | |
14667 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
14668 | + user_service_t *us = c->service_data; | |
14669 | + int state; | |
14670 | + | |
14671 | + if (!us) | |
14672 | + return; | |
14673 | + | |
14674 | + down(&us->lock); | |
14675 | + us->sock = NULL; | |
14676 | + c->service_data = NULL; | |
14677 | + | |
14678 | + if (us->need_startdone) | |
14679 | + kcl_start_done(us->local_id, us->need_startdone); | |
14680 | + | |
14681 | + if (us->async) { | |
14682 | + /* async thread will clean up before exiting */ | |
14683 | + up(&us->lock); | |
14684 | + return; | |
14685 | + } | |
14686 | + state = us->state; | |
14687 | + up(&us->lock); | |
14688 | + | |
14689 | + switch (state) { | |
14690 | + case UST_JOIN: | |
14691 | + break; | |
14692 | + case UST_JOINED: | |
14693 | + user_leave(us, 1); | |
14694 | + /* fall through */ | |
14695 | + case UST_LEAVE: | |
14696 | + case UST_REGISTER: | |
14697 | + user_unregister(us); | |
14698 | + /* fall through */ | |
14699 | + case UST_UNREGISTER: | |
14700 | + kfree(us); | |
14701 | + break; | |
14702 | + } | |
14703 | +} | |
14704 | diff -urN linux-orig/cluster/cman/sm_user.h linux-patched/cluster/cman/sm_user.h | |
bb1d8b11 AM |
14705 | --- linux-orig/cluster/cman/sm_user.h 1970-01-01 07:30:00.000000000 +0730 |
14706 | +++ linux-patched/cluster/cman/sm_user.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
14707 | @@ -0,0 +1,21 @@ |
14708 | +/****************************************************************************** | |
14709 | +******************************************************************************* | |
14710 | +** | |
14711 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14712 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14713 | +** | |
14714 | +** This copyrighted material is made available to anyone wishing to use, | |
14715 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14716 | +** of the GNU General Public License v.2. | |
14717 | +** | |
14718 | +******************************************************************************* | |
14719 | +******************************************************************************/ | |
14720 | + | |
c783755a AM |
14721 | +#ifndef __SM_USER_DOT_H__ |
14722 | +#define __SM_USER_DOT_H__ | |
14723 | + | |
14724 | +int sm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg); | |
14725 | +void sm_sock_release(struct socket *sock); | |
14726 | +void sm_sock_bind(struct socket *sock); | |
14727 | + | |
14728 | +#endif | |
c1c6733f | 14729 | diff -urN linux-orig/include/cluster/cnxman-socket.h linux-patched/include/cluster/cnxman-socket.h |
bb1d8b11 AM |
14730 | --- linux-orig/include/cluster/cnxman-socket.h 1970-01-01 07:30:00.000000000 +0730 |
14731 | +++ linux-patched/include/cluster/cnxman-socket.h 2004-11-03 11:37:37.000000000 +0800 | |
14732 | @@ -0,0 +1,233 @@ | |
c1c6733f AM |
14733 | +/****************************************************************************** |
14734 | +******************************************************************************* | |
14735 | +** | |
14736 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14737 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14738 | +** | |
14739 | +** This copyrighted material is made available to anyone wishing to use, | |
14740 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14741 | +** of the GNU General Public License v.2. | |
14742 | +** | |
14743 | +******************************************************************************* | |
14744 | +******************************************************************************/ | |
14745 | + | |
14746 | +/* CMAN socket interface header, | |
14747 | + may be include by user or kernel code */ | |
14748 | + | |
14749 | +#ifndef __CNXMAN_SOCKET_H | |
14750 | +#define __CNXMAN_SOCKET_H | |
14751 | + | |
b7b72b66 AM |
14752 | +/* A currently unused number. TIPC also uses this number and you're unlikely |
14753 | + to be using both. | |
14754 | + */ | |
14755 | +#define AF_CLUSTER 30 | |
c1c6733f AM |
14756 | +#define PF_CLUSTER AF_CLUSTER |
14757 | + | |
14758 | +/* Protocol(socket) types */ | |
14759 | +#define CLPROTO_MASTER 2 | |
14760 | +#define CLPROTO_CLIENT 3 | |
14761 | + | |
c1c6733f AM |
14762 | +/* ioctls -- should register these properly */ |
14763 | +#define SIOCCLUSTER_NOTIFY _IOW('x', 0x01, int) | |
14764 | +#define SIOCCLUSTER_REMOVENOTIFY _IO( 'x', 0x02) | |
14765 | +#define SIOCCLUSTER_GETMEMBERS _IOR('x', 0x03, struct cl_cluster_nodelist) | |
14766 | +#define SIOCCLUSTER_SETEXPECTED_VOTES _IOW('x', 0x04, int) | |
14767 | +#define SIOCCLUSTER_ISQUORATE _IO( 'x', 0x05) | |
14768 | +#define SIOCCLUSTER_ISLISTENING _IOW('x', 0x06, struct cl_listen_request) | |
14769 | +#define SIOCCLUSTER_GETALLMEMBERS _IOR('x', 0x07, struct cl_cluster_nodelist) | |
14770 | +#define SIOCCLUSTER_SET_VOTES _IOW('x', 0x08, int) | |
14771 | +#define SIOCCLUSTER_GET_VERSION _IOR('x', 0x09, struct cl_version) | |
14772 | +#define SIOCCLUSTER_SET_VERSION _IOW('x', 0x0a, struct cl_version) | |
14773 | +#define SIOCCLUSTER_ISACTIVE _IO( 'x', 0x0b) | |
14774 | +#define SIOCCLUSTER_KILLNODE _IOW('x', 0x0c, int) | |
14775 | +#define SIOCCLUSTER_GET_JOINCOUNT _IO( 'x', 0x0d) | |
14776 | +#define SIOCCLUSTER_SERVICE_REGISTER _IOW('x', 0x0e, char) | |
14777 | +#define SIOCCLUSTER_SERVICE_UNREGISTER _IO('x', 0x0f) | |
14778 | +#define SIOCCLUSTER_SERVICE_JOIN _IO( 'x', 0x10) | |
14779 | +#define SIOCCLUSTER_SERVICE_LEAVE _IO( 'x', 0x20) | |
14780 | +#define SIOCCLUSTER_SERVICE_SETSIGNAL _IOW('x', 0x30, int) | |
14781 | +#define SIOCCLUSTER_SERVICE_STARTDONE _IOW('x', 0x40, unsigned int) | |
14782 | +#define SIOCCLUSTER_SERVICE_GETEVENT _IOR('x', 0x50, struct cl_service_event) | |
14783 | +#define SIOCCLUSTER_SERVICE_GETMEMBERS _IOR('x', 0x60, struct cl_cluster_nodelist) | |
14784 | +#define SIOCCLUSTER_SERVICE_GLOBALID _IOR('x', 0x70, uint32_t) | |
14785 | +#define SIOCCLUSTER_SERVICE_SETLEVEL _IOR('x', 0x80, int) | |
14786 | +#define SIOCCLUSTER_GETNODE _IOWR('x', 0x90, struct cl_cluster_node) | |
bb1d8b11 | 14787 | +#define SIOCCLUSTER_GETCLUSTER _IOWR('x', 0x91, struct cl_cluster_info) |
c1c6733f AM |
14788 | +#define SIOCCLUSTER_BARRIER _IOW('x', 0x0a0, struct cl_barrier_info) |
14789 | + | |
b7b72b66 AM |
14790 | +/* These were setsockopts */ |
14791 | +#define SIOCCLUSTER_PASS_SOCKET _IOW('x', 0x0b0, struct cl_passed_sock) | |
14792 | +#define SIOCCLUSTER_SET_NODENAME _IOW('x', 0x0b1, char *) | |
14793 | +#define SIOCCLUSTER_SET_NODEID _IOW('x', 0x0b2, int) | |
14794 | +#define SIOCCLUSTER_JOIN_CLUSTER _IOW('x', 0x0b3, struct cl_join_cluster_info) | |
14795 | +#define SIOCCLUSTER_LEAVE_CLUSTER _IOW('x', 0x0b4, int) | |
14796 | + | |
14797 | + | |
c1c6733f AM |
14798 | +/* Maximum size of a cluster message */ |
14799 | +#define MAX_CLUSTER_MESSAGE 1500 | |
14800 | +#define MAX_CLUSTER_MEMBER_NAME_LEN 255 | |
14801 | +#define MAX_BARRIER_NAME_LEN 33 | |
14802 | +#define MAX_SA_ADDR_LEN 12 | |
14803 | +#define MAX_CLUSTER_NAME_LEN 16 | |
14804 | + | |
14805 | +/* Well-known cluster port numbers */ | |
14806 | +#define CLUSTER_PORT_MEMBERSHIP 1 /* Mustn't block during cluster | |
14807 | + * transitions! */ | |
14808 | +#define CLUSTER_PORT_SERVICES 2 | |
14809 | +#define CLUSTER_PORT_SYSMAN 10 /* Remote execution daemon */ | |
14810 | +#define CLUSTER_PORT_CLVMD 11 /* Cluster LVM daemon */ | |
14811 | +#define CLUSTER_PORT_SLM 12 /* LVM SLM (simple lock manager) */ | |
14812 | + | |
14813 | +/* Port numbers above this will be blocked when the cluster is inquorate or in | |
14814 | + * transition */ | |
14815 | +#define HIGH_PROTECTED_PORT 9 | |
14816 | + | |
14817 | +/* Reasons for leaving the cluster */ | |
14818 | +#define CLUSTER_LEAVEFLAG_DOWN 0 /* Normal shutdown */ | |
14819 | +#define CLUSTER_LEAVEFLAG_KILLED 1 | |
14820 | +#define CLUSTER_LEAVEFLAG_PANIC 2 | |
14821 | +#define CLUSTER_LEAVEFLAG_REMOVED 3 /* This one can reduce quorum */ | |
14822 | +#define CLUSTER_LEAVEFLAG_REJECTED 4 /* Not allowed into the cluster in the | |
14823 | + * first place */ | |
14824 | +#define CLUSTER_LEAVEFLAG_INCONSISTENT 5 /* Our view of the cluster is | |
14825 | + * in a minority */ | |
14826 | +#define CLUSTER_LEAVEFLAG_DEAD 6 /* Discovered to be dead */ | |
14827 | +#define CLUSTER_LEAVEFLAG_FORCE 0x10 /* Forced by command-line */ | |
14828 | + | |
14829 | +/* OOB messages sent to a local socket */ | |
14830 | +#define CLUSTER_OOB_MSG_PORTCLOSED 1 | |
14831 | +#define CLUSTER_OOB_MSG_STATECHANGE 2 | |
14832 | +#define CLUSTER_OOB_MSG_SERVICEEVENT 3 | |
14833 | + | |
14834 | +/* Sendmsg flags, these are above the normal sendmsg flags so they don't | |
14835 | + * interfere */ | |
14836 | +#define MSG_NOACK 0x010000 /* Don't need an ACK for this message */ | |
14837 | +#define MSG_QUEUE 0x020000 /* Queue the message for sending later */ | |
14838 | +#define MSG_MULTICAST 0x080000 /* Message was sent to all nodes in the cluster | |
14839 | + */ | |
14840 | +#define MSG_ALLINT 0x100000 /* Send out of all interfaces */ | |
b7b72b66 | 14841 | +#define MSG_REPLYEXP 0x200000 /* Reply is expected */ |
c783755a | 14842 | +#define MSG_BCASTSELF 0x400000 /* Broadcast message also gets send to us */ |
c1c6733f | 14843 | + |
c783755a | 14844 | +typedef enum { NODESTATE_JOINING=1, NODESTATE_MEMBER, |
b7b72b66 | 14845 | + NODESTATE_DEAD } nodestate_t; |
c1c6733f AM |
14846 | + |
14847 | + | |
14848 | +struct sockaddr_cl { | |
14849 | + unsigned short scl_family; | |
14850 | + unsigned char scl_flags; | |
14851 | + unsigned char scl_port; | |
14852 | + int scl_nodeid; | |
14853 | +}; | |
14854 | + | |
b7b72b66 AM |
14855 | +/* |
14856 | + * This is how we pass the multicast & receive sockets into kernel space. | |
14857 | + */ | |
14858 | +struct cl_passed_sock { | |
c1c6733f AM |
14859 | + int fd; /* FD of master socket to do multicast on */ |
14860 | + int number; /* Socket number, to match up recvonly & bcast | |
14861 | + * sockets */ | |
b7b72b66 | 14862 | + int multicast; /* Is it multicast or receive ? */ |
c1c6733f AM |
14863 | +}; |
14864 | + | |
14865 | +/* Cluster configuration info passed when we join the cluster */ | |
14866 | +struct cl_join_cluster_info { | |
14867 | + unsigned char votes; | |
14868 | + unsigned int expected_votes; | |
14869 | + unsigned int two_node; | |
14870 | + unsigned int config_version; | |
14871 | + | |
14872 | + char cluster_name[17]; | |
14873 | +}; | |
14874 | + | |
14875 | + | |
14876 | +/* This is the structure, per node, returned from the membership ioctl */ | |
14877 | +struct cl_cluster_node { | |
14878 | + unsigned int size; | |
14879 | + unsigned int node_id; | |
14880 | + unsigned int us; | |
14881 | + unsigned int leave_reason; | |
14882 | + unsigned int incarnation; | |
14883 | + nodestate_t state; | |
14884 | + char name[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
14885 | + unsigned char votes; | |
14886 | +}; | |
14887 | + | |
14888 | +/* The struct passed to the membership ioctls */ | |
14889 | +struct cl_cluster_nodelist { | |
14890 | + uint32_t max_members; | |
14891 | + struct cl_cluster_node *nodes; | |
14892 | +}; | |
14893 | + | |
14894 | +/* Structure passed to SIOCCLUSTER_ISLISTENING */ | |
14895 | +struct cl_listen_request { | |
14896 | + unsigned char port; | |
14897 | + int nodeid; | |
14898 | +}; | |
14899 | + | |
14900 | +/* A Cluster PORTCLOSED message - received by a local user as an OOB message */ | |
14901 | +struct cl_portclosed_oob { | |
14902 | + unsigned char cmd; /* CLUSTER_OOB_MSG_PORTCLOSED */ | |
14903 | + unsigned char port; | |
14904 | +}; | |
14905 | + | |
14906 | +/* Get all version numbers or set the config version */ | |
14907 | +struct cl_version { | |
14908 | + unsigned int major; | |
14909 | + unsigned int minor; | |
14910 | + unsigned int patch; | |
14911 | + unsigned int config; | |
14912 | +}; | |
14913 | + | |
14914 | +/* structure passed to barrier ioctls */ | |
14915 | +struct cl_barrier_info { | |
14916 | + char cmd; | |
14917 | + char name[MAX_BARRIER_NAME_LEN]; | |
14918 | + unsigned int flags; | |
14919 | + unsigned long arg; | |
14920 | +}; | |
14921 | + | |
bb1d8b11 AM |
14922 | +struct cl_cluster_info { |
14923 | + char name[MAX_CLUSTER_NAME_LEN+1]; | |
14924 | + uint16_t number; | |
14925 | +}; | |
14926 | + | |
c1c6733f AM |
14927 | +typedef enum { SERVICE_EVENT_STOP, SERVICE_EVENT_START, SERVICE_EVENT_FINISH, |
14928 | + SERVICE_EVENT_LEAVEDONE } service_event_t; | |
14929 | + | |
14930 | +typedef enum { SERVICE_START_FAILED, SERVICE_START_JOIN, SERVICE_START_LEAVE } | |
14931 | + service_start_t; | |
14932 | + | |
14933 | +struct cl_service_event { | |
14934 | + service_event_t type; | |
14935 | + service_start_t start_type; | |
14936 | + unsigned int event_id; | |
14937 | + unsigned int last_stop; | |
14938 | + unsigned int last_start; | |
14939 | + unsigned int last_finish; | |
14940 | + unsigned int node_count; | |
14941 | +}; | |
14942 | + | |
14943 | + | |
14944 | +/* Commands to the barrier ioctl */ | |
14945 | +#define BARRIER_IOCTL_REGISTER 1 | |
14946 | +#define BARRIER_IOCTL_CHANGE 2 | |
14947 | +#define BARRIER_IOCTL_DELETE 3 | |
14948 | +#define BARRIER_IOCTL_WAIT 4 | |
14949 | + | |
bb1d8b11 AM |
14950 | +/* Attributes of a barrier - bitmask */ |
14951 | +#define BARRIER_ATTR_AUTODELETE 1 | |
14952 | +#define BARRIER_ATTR_MULTISTEP 2 | |
14953 | +#define BARRIER_ATTR_MANUAL 4 | |
14954 | +#define BARRIER_ATTR_ENABLED 8 | |
14955 | +#define BARRIER_ATTR_CALLBACK 16 | |
14956 | + | |
14957 | +/* Attribute setting commands */ | |
14958 | +#define BARRIER_SETATTR_AUTODELETE 1 | |
14959 | +#define BARRIER_SETATTR_MULTISTEP 2 | |
14960 | +#define BARRIER_SETATTR_ENABLED 3 | |
14961 | +#define BARRIER_SETATTR_NODES 4 | |
14962 | +#define BARRIER_SETATTR_CALLBACK 5 | |
14963 | +#define BARRIER_SETATTR_TIMEOUT 6 | |
14964 | + | |
14965 | +#endif | |
14966 | diff -urN linux-orig/include/cluster/cnxman.h linux-patched/include/cluster/cnxman.h | |
14967 | --- linux-orig/include/cluster/cnxman.h 1970-01-01 07:30:00.000000000 +0730 | |
14968 | +++ linux-patched/include/cluster/cnxman.h 2004-11-03 11:37:37.000000000 +0800 | |
14969 | @@ -0,0 +1,87 @@ | |
14970 | +/****************************************************************************** | |
14971 | +******************************************************************************* | |
14972 | +** | |
14973 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14974 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14975 | +** | |
14976 | +** This copyrighted material is made available to anyone wishing to use, | |
14977 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14978 | +** of the GNU General Public License v.2. | |
14979 | +** | |
14980 | +******************************************************************************* | |
14981 | +******************************************************************************/ | |
14982 | + | |
14983 | +#ifndef __CNXMAN_H | |
14984 | +#define __CNXMAN_H | |
14985 | + | |
14986 | +#include "linux/in6.h" | |
14987 | +#include "cluster/cnxman-socket.h" | |
14988 | + | |
14989 | +/* In-kernel API */ | |
14990 | + | |
14991 | +/* This is the structure, per node, returned from the membership request */ | |
14992 | +struct kcl_cluster_node { | |
14993 | + unsigned int size; | |
14994 | + unsigned int node_id; | |
14995 | + unsigned int us; | |
14996 | + unsigned int leave_reason; | |
14997 | + unsigned int incarnation; | |
14998 | + nodestate_t state; | |
14999 | + struct list_head list; | |
15000 | + char name[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
15001 | + unsigned char votes; | |
15002 | +}; | |
15003 | + | |
15004 | +struct cluster_node_addr { | |
15005 | + struct list_head list; | |
15006 | + unsigned char addr[sizeof(struct sockaddr_in6)];/* A large sockaddr */ | |
15007 | + int addr_len; | |
15008 | +}; | |
15009 | + | |
15010 | + | |
15011 | +/* Reasons for a kernel membership callback */ | |
15012 | +typedef enum { CLUSTER_RECONFIG, DIED, LEAVING, NEWNODE } kcl_callback_reason; | |
15013 | + | |
15014 | +/* Kernel version of above, the void *sock is a struct socket */ | |
15015 | +struct kcl_multicast_sock { | |
15016 | + void *sock; | |
15017 | + int number; /* Socket number, to match up recvonly & bcast | |
15018 | + * sockets */ | |
15019 | +}; | |
15020 | + | |
15021 | +extern int kcl_sendmsg(struct socket *sock, void *buf, int size, | |
15022 | + struct sockaddr_cl *caddr, int addr_len, | |
15023 | + unsigned int flags); | |
15024 | +extern int kcl_register_read_callback(struct socket *sock, | |
15025 | + int (*routine) (char *, int, char *, int, | |
15026 | + unsigned int)); | |
15027 | +extern int kcl_add_callback(void (*callback) (kcl_callback_reason, long)); | |
15028 | +extern int kcl_remove_callback(void (*callback) (kcl_callback_reason, long)); | |
15029 | +extern int kcl_get_members(struct list_head *list); | |
15030 | +extern int kcl_get_member_ids(uint32_t * idbuf, int size); | |
15031 | +extern int kcl_get_all_members(struct list_head *list); | |
15032 | +extern int kcl_get_node_by_addr(unsigned char *addr, int addr_len, | |
15033 | + struct kcl_cluster_node *n); | |
15034 | +extern int kcl_get_node_by_name(unsigned char *name, | |
15035 | + struct kcl_cluster_node *n); | |
15036 | +extern int kcl_get_node_by_nodeid(int nodeid, struct kcl_cluster_node *n); | |
15037 | +extern int kcl_is_quorate(void); | |
15038 | +extern int kcl_addref_cluster(void); | |
15039 | +extern int kcl_releaseref_cluster(void); | |
15040 | +extern int kcl_cluster_name(char **cname); | |
15041 | +extern int kcl_get_current_interface(void); | |
15042 | +extern struct list_head *kcl_get_node_addresses(int nodeid); | |
15043 | + | |
15044 | +extern int kcl_barrier_register(char *name, unsigned int flags, | |
15045 | + unsigned int nodes); | |
15046 | +extern int kcl_barrier_setattr(char *name, unsigned int attr, | |
15047 | + unsigned long arg); | |
15048 | +extern int kcl_barrier_delete(char *name); | |
15049 | +extern int kcl_barrier_wait(char *name); | |
15050 | +extern int kcl_barrier_cancel(char *name); | |
c1c6733f | 15051 | + |
bb1d8b11 AM |
15052 | +extern int kcl_register_quorum_device(char *name, int votes); |
15053 | +extern int kcl_unregister_quorum_device(void); | |
15054 | +extern int kcl_quorum_device_available(int yesno); | |
c1c6733f AM |
15055 | + |
15056 | +#endif | |
c1c6733f | 15057 | diff -urN linux-orig/include/cluster/service.h linux-patched/include/cluster/service.h |
bb1d8b11 AM |
15058 | --- linux-orig/include/cluster/service.h 1970-01-01 07:30:00.000000000 +0730 |
15059 | +++ linux-patched/include/cluster/service.h 2004-11-03 11:37:37.000000000 +0800 | |
c1c6733f AM |
15060 | @@ -0,0 +1,102 @@ |
15061 | +/****************************************************************************** | |
15062 | +******************************************************************************* | |
15063 | +** | |
15064 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
15065 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
15066 | +** | |
15067 | +** This copyrighted material is made available to anyone wishing to use, | |
15068 | +** modify, copy, or redistribute it subject to the terms and conditions | |
15069 | +** of the GNU General Public License v.2. | |
15070 | +** | |
15071 | +******************************************************************************* | |
15072 | +******************************************************************************/ | |
15073 | + | |
15074 | +#ifndef __SERVICE_DOT_H__ | |
15075 | +#define __SERVICE_DOT_H__ | |
15076 | + | |
15077 | +/* | |
15078 | + * Interface between service manager and services | |
15079 | + */ | |
15080 | + | |
15081 | +/* | |
15082 | + * Service levels are started in order from lowest, so level 0 is started on | |
15083 | + * all nodes before level 1 is started. | |
15084 | + */ | |
15085 | + | |
15086 | +#define SERVICE_LEVEL_FENCE (0) | |
15087 | +#define SERVICE_LEVEL_GDLM (1) | |
15088 | +#define SERVICE_LEVEL_GFS (2) | |
15089 | +#define SERVICE_LEVEL_USER (3) | |
15090 | + | |
15091 | +#define MAX_SERVICE_NAME_LEN (33) | |
15092 | + | |
15093 | +/* | |
15094 | + * The type of start a service receives. The start (and preceding stop) may be | |
15095 | + * due to a node joining or leaving the SG or due to a node having failed. | |
15096 | + */ | |
15097 | + | |
15098 | +#define SERVICE_NODE_FAILED (1) | |
15099 | +#define SERVICE_NODE_JOIN (2) | |
15100 | +#define SERVICE_NODE_LEAVE (3) | |
15101 | + | |
15102 | + | |
15103 | +struct kcl_service { | |
15104 | + struct list_head list; | |
15105 | + uint16_t level; | |
15106 | + uint32_t local_id; | |
15107 | + uint32_t global_id; | |
15108 | + int node_count; | |
15109 | + char name[MAX_SERVICE_NAME_LEN]; | |
15110 | +}; | |
15111 | + | |
15112 | +int kcl_get_services(struct list_head *list, int level); | |
15113 | + | |
15114 | + | |
15115 | +/* | |
15116 | + * These routines which run in CMAN context must return quickly and cannot | |
15117 | + * block. | |
15118 | + */ | |
15119 | + | |
15120 | +struct kcl_service_ops { | |
15121 | + int (*stop) (void *servicedata); | |
15122 | + int (*start) (void *servicedata, uint32_t *nodeids, int count, | |
15123 | + int event_id, int type); | |
15124 | + void (*finish) (void *servicedata, int event_id); | |
15125 | +}; | |
15126 | + | |
15127 | +/* | |
15128 | + * Register will cause CMAN to create a Service Group (SG) for the named | |
15129 | + * instance of the service. A local ID is returned which is used to join, | |
15130 | + * leave and unregister the service. | |
15131 | + */ | |
15132 | + | |
15133 | +int kcl_register_service(char *name, int namelen, int level, | |
15134 | + struct kcl_service_ops *ops, int unique, | |
15135 | + void *servicedata, uint32_t *local_id); | |
15136 | + | |
15137 | +void kcl_unregister_service(uint32_t local_id); | |
15138 | + | |
15139 | +/* | |
15140 | + * Once a service is joined it will be managed by CMAN and receive start, stop, | |
15141 | + * and finish calls. After leave is called the service is no longer managed by | |
15142 | + * CMAN. The first start for a service may arrive before kcl_join_service() | |
15143 | + * returns. | |
15144 | + */ | |
15145 | + | |
15146 | +int kcl_join_service(uint32_t local_id); | |
15147 | +int kcl_leave_service(uint32_t local_id); | |
15148 | + | |
15149 | +/* | |
15150 | + * After a service is started, it can ask for its cluster-wide unique ID. | |
15151 | + */ | |
15152 | + | |
15153 | +void kcl_global_service_id(uint32_t local_id, uint32_t * global_id); | |
15154 | + | |
15155 | +/* | |
15156 | + * Called by a service when it's done with a start(). Cannot be called from | |
15157 | + * the start function. | |
15158 | + */ | |
15159 | + | |
15160 | +void kcl_start_done(uint32_t local_id, int event_id); | |
15161 | + | |
15162 | +#endif |