]>
Commit | Line | Data |
---|---|---|
50905e9e AM |
1 | diff -urN linux-2.6.7clean/arch/alpha/Kconfig linux-2.6.7/arch/alpha/Kconfig |
2 | --- linux-2.6.7clean/arch/alpha/Kconfig 2004-07-05 10:44:04.000000000 +0100 | |
3 | +++ linux-2.6.7/arch/alpha/Kconfig 2004-07-05 10:42:50.000000000 +0100 | |
4 | @@ -698,3 +698,4 @@ | |
4bf12011 | 5 | |
50905e9e AM |
6 | source "lib/Kconfig" |
7 | ||
8 | +source "cluster/Kconfig" | |
9 | diff -urN linux-2.6.7clean/arch/arm/Kconfig linux-2.6.7/arch/arm/Kconfig | |
10 | --- linux-2.6.7clean/arch/arm/Kconfig 2004-07-05 10:44:04.000000000 +0100 | |
11 | +++ linux-2.6.7/arch/arm/Kconfig 2004-07-05 10:50:40.000000000 +0100 | |
12 | @@ -814,3 +814,4 @@ | |
4bf12011 | 13 | |
14 | source "lib/Kconfig" | |
15 | ||
16 | +source "cluster/Kconfig" | |
50905e9e AM |
17 | diff -urN linux-2.6.7clean/arch/arm26/Kconfig linux-2.6.7/arch/arm26/Kconfig |
18 | --- linux-2.6.7clean/arch/arm26/Kconfig 2004-07-05 10:43:54.000000000 +0100 | |
19 | +++ linux-2.6.7/arch/arm26/Kconfig 2004-07-05 10:51:42.000000000 +0100 | |
20 | @@ -333,3 +333,4 @@ | |
21 | ||
22 | source "lib/Kconfig" | |
23 | ||
24 | +source "cluster/Kconfig" | |
25 | diff -urN linux-2.6.7clean/arch/cris/Kconfig linux-2.6.7/arch/cris/Kconfig | |
26 | --- linux-2.6.7clean/arch/cris/Kconfig 2004-07-05 10:44:04.000000000 +0100 | |
27 | +++ linux-2.6.7/arch/cris/Kconfig 2004-07-05 10:50:30.000000000 +0100 | |
28 | @@ -216,3 +216,4 @@ | |
29 | ||
30 | source "lib/Kconfig" | |
31 | ||
32 | +source "cluster/Kconfig" | |
33 | diff -urN linux-2.6.7clean/arch/i386/Kconfig linux-2.6.7/arch/i386/Kconfig | |
34 | --- linux-2.6.7clean/arch/i386/Kconfig 2004-07-05 10:44:04.000000000 +0100 | |
35 | +++ linux-2.6.7/arch/i386/Kconfig 2004-07-05 10:42:50.000000000 +0100 | |
36 | @@ -1315,6 +1315,8 @@ | |
4bf12011 | 37 | |
38 | source "lib/Kconfig" | |
39 | ||
40 | +source "cluster/Kconfig" | |
41 | + | |
42 | config X86_SMP | |
43 | bool | |
44 | depends on SMP && !X86_VOYAGER | |
50905e9e AM |
45 | diff -urN linux-2.6.7clean/arch/ia64/Kconfig linux-2.6.7/arch/ia64/Kconfig |
46 | --- linux-2.6.7clean/arch/ia64/Kconfig 2004-07-05 10:44:04.000000000 +0100 | |
47 | +++ linux-2.6.7/arch/ia64/Kconfig 2004-07-05 10:49:31.000000000 +0100 | |
48 | @@ -503,3 +503,5 @@ | |
49 | source "security/Kconfig" | |
50 | ||
51 | source "crypto/Kconfig" | |
52 | + | |
53 | +source "cluster/Kconfig" | |
54 | diff -urN linux-2.6.7clean/arch/m68k/Kconfig linux-2.6.7/arch/m68k/Kconfig | |
55 | --- linux-2.6.7clean/arch/m68k/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
56 | +++ linux-2.6.7/arch/m68k/Kconfig 2004-07-05 10:49:11.000000000 +0100 | |
57 | @@ -696,3 +696,4 @@ | |
58 | ||
59 | source "lib/Kconfig" | |
60 | ||
61 | +source "cluster/Kconfig" | |
62 | diff -urN linux-2.6.7clean/arch/mips/Kconfig linux-2.6.7/arch/mips/Kconfig | |
63 | --- linux-2.6.7clean/arch/mips/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
64 | +++ linux-2.6.7/arch/mips/Kconfig 2004-07-05 10:48:56.000000000 +0100 | |
65 | @@ -1651,3 +1651,5 @@ | |
66 | source "crypto/Kconfig" | |
67 | ||
68 | source "lib/Kconfig" | |
69 | + | |
70 | +source "cluster/Kconfig" | |
71 | diff -urN linux-2.6.7clean/arch/parisc/Kconfig linux-2.6.7/arch/parisc/Kconfig | |
72 | --- linux-2.6.7clean/arch/parisc/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
73 | +++ linux-2.6.7/arch/parisc/Kconfig 2004-07-05 10:42:50.000000000 +0100 | |
74 | @@ -229,3 +229,4 @@ | |
75 | ||
76 | source "lib/Kconfig" | |
77 | ||
78 | +source "cluster/Kconfig" | |
79 | diff -urN linux-2.6.7clean/arch/ppc/Kconfig linux-2.6.7/arch/ppc/Kconfig | |
80 | --- linux-2.6.7clean/arch/ppc/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
81 | +++ linux-2.6.7/arch/ppc/Kconfig 2004-07-05 10:48:34.000000000 +0100 | |
82 | @@ -1281,3 +1281,5 @@ | |
83 | source "security/Kconfig" | |
84 | ||
85 | source "crypto/Kconfig" | |
86 | + | |
87 | +source "cluster/Kconfig" | |
88 | diff -urN linux-2.6.7clean/arch/ppc64/Kconfig linux-2.6.7/arch/ppc64/Kconfig | |
89 | --- linux-2.6.7clean/arch/ppc64/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
90 | +++ linux-2.6.7/arch/ppc64/Kconfig 2004-07-05 10:48:43.000000000 +0100 | |
91 | @@ -443,3 +443,4 @@ | |
92 | ||
93 | source "lib/Kconfig" | |
94 | ||
95 | +source "cluster/Kconfig" | |
96 | diff -urN linux-2.6.7clean/arch/s390/Kconfig linux-2.6.7/arch/s390/Kconfig | |
97 | --- linux-2.6.7clean/arch/s390/Kconfig 2004-07-05 10:44:02.000000000 +0100 | |
98 | +++ linux-2.6.7/arch/s390/Kconfig 2004-07-05 10:48:22.000000000 +0100 | |
99 | @@ -431,3 +431,4 @@ | |
4bf12011 | 100 | |
101 | source "lib/Kconfig" | |
102 | ||
103 | +source "cluster/Kconfig" | |
50905e9e AM |
104 | diff -urN linux-2.6.7clean/arch/sh/Kconfig linux-2.6.7/arch/sh/Kconfig |
105 | --- linux-2.6.7clean/arch/sh/Kconfig 2004-07-05 10:43:55.000000000 +0100 | |
106 | +++ linux-2.6.7/arch/sh/Kconfig 2004-07-05 10:48:12.000000000 +0100 | |
107 | @@ -798,3 +798,4 @@ | |
4bf12011 | 108 | |
109 | source "lib/Kconfig" | |
110 | ||
111 | +source "cluster/Kconfig" | |
50905e9e AM |
112 | diff -urN linux-2.6.7clean/arch/sparc/Kconfig linux-2.6.7/arch/sparc/Kconfig |
113 | --- linux-2.6.7clean/arch/sparc/Kconfig 2004-07-05 10:43:55.000000000 +0100 | |
114 | +++ linux-2.6.7/arch/sparc/Kconfig 2004-07-05 10:47:47.000000000 +0100 | |
115 | @@ -456,3 +456,4 @@ | |
116 | ||
117 | source "lib/Kconfig" | |
118 | ||
119 | +source "cluster/Kconfig" | |
120 | diff -urN linux-2.6.7clean/arch/sparc64/Kconfig linux-2.6.7/arch/sparc64/Kconfig | |
121 | --- linux-2.6.7clean/arch/sparc64/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
122 | +++ linux-2.6.7/arch/sparc64/Kconfig 2004-07-05 10:42:50.000000000 +0100 | |
123 | @@ -713,3 +713,4 @@ | |
124 | ||
125 | source "lib/Kconfig" | |
126 | ||
127 | +source "cluster/Kconfig" | |
128 | diff -urN linux-2.6.7clean/arch/um/Kconfig linux-2.6.7/arch/um/Kconfig | |
129 | --- linux-2.6.7clean/arch/um/Kconfig 2004-07-05 10:44:02.000000000 +0100 | |
130 | +++ linux-2.6.7/arch/um/Kconfig 2004-07-05 10:47:29.000000000 +0100 | |
131 | @@ -194,6 +194,8 @@ | |
132 | ||
133 | source "lib/Kconfig" | |
134 | ||
135 | +source "cluster/Kconfig" | |
136 | + | |
137 | menu "SCSI support" | |
138 | ||
139 | config SCSI | |
140 | diff -urN linux-2.6.7clean/arch/x86_64/Kconfig linux-2.6.7/arch/x86_64/Kconfig | |
141 | --- linux-2.6.7clean/arch/x86_64/Kconfig 2004-07-05 10:44:10.000000000 +0100 | |
142 | +++ linux-2.6.7/arch/x86_64/Kconfig 2004-07-05 10:46:26.000000000 +0100 | |
143 | @@ -505,3 +505,4 @@ | |
144 | ||
145 | source "lib/Kconfig" | |
146 | ||
147 | +source "cluster/Kconfig" | |
148 | diff -urN linux-2.6.7clean/cluster/cman/Makefile linux-2.6.7/cluster/cman/Makefile | |
149 | --- linux-2.6.7clean/cluster/cman/Makefile 1970-01-01 01:00:00.000000000 +0100 | |
150 | +++ linux-2.6.7/cluster/cman/Makefile 2004-07-05 10:42:50.000000000 +0100 | |
151 | @@ -0,0 +1,6 @@ | |
152 | +cman-objs := cnxman.o config.o membership.o proc.o\ | |
153 | + sm_barrier.o sm_control.o sm_daemon.o sm_joinleave.o\ | |
154 | + sm_membership.o sm_message.o sm_misc.o sm_recover.o sm_services.o \ | |
155 | + sm_user.o | |
156 | + | |
157 | +obj-$(CONFIG_CLUSTER) := cman.o | |
158 | diff -urN linux-2.6.7clean/cluster/Kconfig linux-2.6.7/cluster/Kconfig | |
159 | --- linux-2.6.7clean/cluster/Kconfig 1970-01-01 01:00:00.000000000 +0100 | |
160 | +++ linux-2.6.7/cluster/Kconfig 2004-07-05 10:42:50.000000000 +0100 | |
4bf12011 | 161 | @@ -0,0 +1,13 @@ |
162 | +menu "Cluster Support" | |
163 | + | |
164 | +config CLUSTER | |
165 | + tristate "Cluster support" | |
166 | + ---help--- | |
167 | + Enable clustering support. This is not the high-performance clustering | |
168 | + made famous by beowulf. It is a high-availability cluster often using | |
169 | + shared storage. | |
170 | + The cluster manager is the heart(beat) of the cluster system. It is | |
171 | + needed by all the other components. It provides membership services | |
172 | + for those other subsystems. | |
173 | + | |
174 | +endmenu | |
50905e9e AM |
175 | diff -urN linux-2.6.7clean/cluster/Makefile linux-2.6.7/cluster/Makefile |
176 | --- linux-2.6.7clean/cluster/Makefile 1970-01-01 01:00:00.000000000 +0100 | |
177 | +++ linux-2.6.7/cluster/Makefile 2004-07-05 10:42:50.000000000 +0100 | |
4bf12011 | 178 | @@ -0,0 +1,3 @@ |
179 | +obj-y := nocluster.o | |
180 | + | |
181 | +obj-$(CONFIG_CLUSTER) += cman/ | |
50905e9e AM |
182 | diff -urN linux-2.6.7clean/cluster/nocluster.c linux-2.6.7/cluster/nocluster.c |
183 | --- linux-2.6.7clean/cluster/nocluster.c 1970-01-01 01:00:00.000000000 +0100 | |
184 | +++ linux-2.6.7/cluster/nocluster.c 2004-07-05 10:42:50.000000000 +0100 | |
4bf12011 | 185 | @@ -0,0 +1,20 @@ |
186 | +/* | |
187 | + * cluster/nocluster.c | |
188 | + * | |
189 | + * Copy from net/nonet.c | |
190 | + * Dummy functions to allow us to configure cluster support entirely | |
191 | + * out of the kernel. | |
192 | + * | |
193 | + * Distributed under the terms of the GNU GPL version 2. | |
194 | + * Copyright (c) Matthew Wilcox 2003 | |
195 | + */ | |
196 | + | |
197 | +#include <linux/module.h> | |
198 | +#include <linux/errno.h> | |
199 | +#include <linux/fs.h> | |
200 | +#include <linux/init.h> | |
201 | +#include <linux/kernel.h> | |
202 | + | |
203 | +void __init nocluster_init(void) | |
204 | +{ | |
205 | +} | |
50905e9e AM |
206 | diff -urN linux-2.6.7clean/Makefile linux-2.6.7/Makefile |
207 | --- linux-2.6.7clean/Makefile 2004-07-05 10:44:04.000000000 +0100 | |
208 | +++ linux-2.6.7/Makefile 2004-07-05 10:42:50.000000000 +0100 | |
209 | @@ -418,7 +418,7 @@ | |
210 | ||
211 | # Objects we will link into vmlinux / subdirs we need to visit | |
212 | init-y := init/ | |
213 | -drivers-y := drivers/ sound/ | |
214 | +drivers-y := drivers/ sound/ cluster/ | |
215 | net-y := net/ | |
216 | libs-y := lib/ | |
217 | core-y := usr/ | |
4bf12011 | 218 | diff -urN linux-orig/cluster/cman/cnxman-private.h linux-patched/cluster/cman/cnxman-private.h |
219 | --- linux-orig/cluster/cman/cnxman-private.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 220 | +++ linux-patched/cluster/cman/cnxman-private.h 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 221 | @@ -0,0 +1,427 @@ |
222 | +/****************************************************************************** | |
223 | +******************************************************************************* | |
224 | +** | |
225 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
226 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
227 | +** | |
228 | +** This copyrighted material is made available to anyone wishing to use, | |
229 | +** modify, copy, or redistribute it subject to the terms and conditions | |
230 | +** of the GNU General Public License v.2. | |
231 | +** | |
232 | +******************************************************************************* | |
233 | +******************************************************************************/ | |
234 | + | |
235 | +#ifndef __CNXMAN_PRIVATE_H | |
236 | +#define __CNXMAN_PRIVATE_H | |
237 | + | |
238 | +/* Version triplet */ | |
239 | +#define CNXMAN_MAJOR_VERSION 2 | |
240 | +#define CNXMAN_MINOR_VERSION 0 | |
241 | +#define CNXMAN_PATCH_VERSION 1 | |
242 | + | |
243 | +#define MAX_RETRIES 3 /* Maximum number of send retries */ | |
244 | +#define CAP_CLUSTER CAP_SYS_ADMIN /* Capability needed to manage the | |
245 | + * cluster */ | |
246 | +#ifdef __KERNEL__ | |
247 | + | |
248 | +/* How we announce ourself in console events */ | |
249 | +#define CMAN_NAME "CMAN" | |
250 | + | |
251 | +/* One of these per AF_CLUSTER socket */ | |
252 | +struct cluster_sock { | |
253 | + /* WARNING: sk has to be the first member */ | |
254 | + struct sock sk; | |
255 | + | |
256 | + unsigned char port; /* Bound port or zero */ | |
257 | + int (*kernel_callback) (char *, int, char *, int, unsigned int); | |
258 | + void *service_data; | |
259 | +}; | |
260 | + | |
261 | +#define cluster_sk(__sk) ((struct cluster_sock *)__sk) | |
262 | + | |
263 | +/* We have one of these for each socket we use for communications */ | |
264 | +struct cl_comms_socket { | |
265 | + struct socket *sock; | |
266 | + int broadcast; /* This is a broadcast socket */ | |
267 | + int recv_only; /* This is the unicast receive end of a | |
268 | + * multicast socket */ | |
269 | + struct sockaddr_in6 saddr; /* Socket address, contains the sockaddr for | |
270 | + * the remote end(s) */ | |
271 | + int addr_len; /* Length of above */ | |
272 | + int number; /* Internal socket number, used to cycle around | |
273 | + * sockets in case of network errors */ | |
274 | + struct file *file; /* file pointer for user-passed in sockets */ | |
275 | + | |
276 | + wait_queue_t wait; | |
277 | + | |
278 | + /* The socket list */ | |
279 | + struct list_head list; | |
280 | + | |
281 | + /* On here when it has something to say */ | |
282 | + struct list_head active_list; | |
283 | + unsigned long active; | |
284 | +}; | |
285 | + | |
286 | +/* A client socket. We keep a list of these so we can notify clients of cluster | |
287 | + * events */ | |
288 | +struct cl_client_socket { | |
289 | + struct socket *sock; | |
290 | + struct list_head list; | |
291 | +}; | |
292 | + | |
293 | +/* This structure is tacked onto the start of a cluster message packet for our | |
294 | + * own nefarious purposes. */ | |
295 | +struct cl_protheader { | |
296 | + unsigned char port; | |
297 | + unsigned char flags; | |
298 | + unsigned short cluster; /* Our cluster number, little-endian */ | |
299 | + unsigned short seq; /* Packet sequence number, little-endian */ | |
300 | + int srcid; /* Node ID of the sender */ | |
301 | + int tgtid; /* Node ID of the target or 0 for multicast | |
302 | + * messages */ | |
303 | +}; | |
304 | + | |
305 | +/* A cluster internal protocol message - port number 0 */ | |
306 | +struct cl_protmsg { | |
307 | + struct cl_protheader header; | |
308 | + unsigned char cmd; | |
309 | +}; | |
310 | + | |
311 | +/* A Cluster ACK message */ | |
312 | +struct cl_ackmsg { | |
313 | + struct cl_protheader header; | |
314 | + unsigned char cmd; /* Always CLUSTER_CMD_ACK */ | |
315 | + unsigned char remport; /* Remoye port number the original message was | |
316 | + * for */ | |
317 | + unsigned char aflags; /* ACK flags 0=OK, 1=No listener */ | |
318 | + unsigned char pad; | |
319 | + unsigned short seq; /* Sequence number we are acking */ | |
320 | +}; | |
321 | + | |
322 | +/* A Cluster LISTENREQ/LISTENRESP message */ | |
323 | +struct cl_listenmsg { | |
324 | + unsigned char cmd; /* CLUSTER_CMD_LISTENRESP/REQ */ | |
325 | + unsigned char target_port; /* Port to probe */ | |
326 | + unsigned char listening; /* Always 0 for LISTENREQ */ | |
327 | + unsigned char pad; | |
328 | + unsigned short tag; /* PID of remote waiting process */ | |
329 | +}; | |
330 | + | |
331 | +/* A Cluster PORTCLOSED message */ | |
332 | +struct cl_closemsg { | |
333 | + unsigned char cmd; /* CLUSTER_CMD_PORTCLOSED */ | |
334 | + unsigned char port; | |
335 | +}; | |
336 | + | |
337 | +/* Structure of a newly dead node, passed from cnxman to kmembershipd */ | |
338 | +struct cl_new_dead_node { | |
339 | + struct list_head list; | |
340 | + struct cluster_node *node; | |
341 | +}; | |
342 | + | |
343 | +/* Subcommands for BARRIER message */ | |
344 | +#define BARRIER_REGISTER 1 | |
345 | +#define BARRIER_CHANGE 2 | |
346 | +#define BARRIER_WAIT 4 | |
347 | +#define BARRIER_COMPLETE 5 | |
348 | + | |
349 | +/* A Cluster BARRIER message */ | |
350 | +struct cl_barriermsg { | |
351 | + unsigned char cmd; /* CLUSTER_CMD_BARRIER */ | |
352 | + unsigned char subcmd; /* BARRIER sub command */ | |
353 | + unsigned short pad; | |
354 | + unsigned int flags; | |
355 | + unsigned int nodes; | |
356 | + char name[MAX_BARRIER_NAME_LEN]; | |
357 | +}; | |
358 | + | |
359 | +/* Membership services messages, the cl_protheader is added transparently */ | |
360 | +struct cl_mem_hello_msg { | |
361 | + unsigned char cmd; | |
362 | + unsigned char flags; | |
363 | + unsigned short members; /* Number of nodes in the cluster, | |
364 | + * little-endian */ | |
365 | + unsigned int generation; /* Current cluster generation number */ | |
366 | +}; | |
367 | + | |
368 | +struct cl_mem_endtrans_msg { | |
369 | + unsigned char cmd; | |
370 | + unsigned char pad1; | |
371 | + unsigned short pad2; | |
372 | + unsigned int quorum; | |
373 | + unsigned int total_votes; | |
374 | + unsigned int generation; /* Current cluster generation number */ | |
375 | + unsigned int new_node_id; /* If reason is a new node joining */ | |
376 | +}; | |
377 | + | |
378 | +/* ACK types for JOINACK message */ | |
379 | +#define JOINACK_TYPE_OK 1 /* You can join */ | |
380 | +#define JOINACK_TYPE_NAK 2 /* You can NOT join */ | |
381 | +#define JOINACK_TYPE_WAIT 3 /* Wait a bit longer - cluster is in transition | |
382 | + * already */ | |
383 | + | |
384 | +struct cl_mem_joinack_msg { | |
385 | + unsigned char cmd; | |
386 | + unsigned char acktype; | |
387 | +}; | |
388 | + | |
389 | +/* This is used by JOINREQ message */ | |
390 | +struct cl_mem_join_msg { | |
391 | + unsigned char cmd; | |
392 | + unsigned char votes; | |
393 | + unsigned short num_addr; /* Number of addresses for this node */ | |
394 | + unsigned int expected_votes; | |
395 | + unsigned int members; /* Number of nodes in the cluster, | |
396 | + * little-endian */ | |
397 | + unsigned int major_version; /* Not backwards compatible */ | |
398 | + unsigned int minor_version; /* Backwards compatible */ | |
399 | + unsigned int patch_version; /* Backwards/forwards compatible */ | |
400 | + unsigned int config_version; | |
401 | + unsigned int addr_len; /* length of node addresses */ | |
402 | + char clustername[16]; | |
403 | + /* Followed by <num_addr> addresses of `address_length` bytes and a | |
404 | + * NUL-terminated node name */ | |
405 | +}; | |
406 | + | |
407 | +/* State transition start reasons: */ | |
408 | +#define TRANS_NEWNODE 1 /* A new node is joining the cluster */ | |
409 | +#define TRANS_REMNODE 2 /* a node has left the cluster */ | |
410 | +#define TRANS_ANOTHERREMNODE 3 /* A node left the cluster while we were in | |
411 | + * transition */ | |
412 | +#define TRANS_NEWMASTER 4 /* We have had an election and I am the new | |
413 | + * master */ | |
414 | +#define TRANS_CHECK 5 /* A consistency check was called for */ | |
415 | +#define TRANS_RESTART 6 /* Transition restarted because of a previous | |
416 | + * timeout */ | |
417 | +#define TRANS_DEADMASTER 7 /* The master died during transition and I have | |
418 | + * taken over */ | |
419 | + | |
420 | +/* This is used to start a state transition */ | |
421 | +struct cl_mem_starttrans_msg { | |
422 | + unsigned char cmd; | |
423 | + unsigned char reason; /* Why a start transition was started - see | |
424 | + * above */ | |
425 | + unsigned char flags; | |
426 | + unsigned char votes; | |
427 | + unsigned int expected_votes; | |
428 | + unsigned int generation; /* Incremented for each STARTTRANS sent | |
429 | + */ | |
430 | + int nodeid; /* Node to be removed */ | |
431 | + unsigned short num_addrs; | |
432 | + /* If reason == TRANS_NEWNODE: Followed by <num_addr> addresses of | |
433 | + * `address_length` bytes and a NUL-terminated node name */ | |
434 | +}; | |
435 | + | |
436 | +struct cl_mem_startack_msg { | |
437 | + unsigned char cmd; | |
438 | + unsigned char reason; | |
439 | + unsigned short pad; | |
440 | + unsigned int generation; | |
441 | + unsigned int node_id; /* node_id we think new node should have */ | |
442 | + unsigned int highest_node_id; /* highest node_id on this system */ | |
443 | +}; | |
444 | + | |
445 | +/* Reconfigure a cluster parameter */ | |
446 | +struct cl_mem_reconfig_msg { | |
447 | + unsigned char cmd; | |
448 | + unsigned char param; | |
449 | + unsigned short pad; | |
450 | + unsigned int value; | |
451 | +}; | |
452 | + | |
453 | +/* Structure containing information about an outstanding listen request */ | |
454 | +struct cl_waiting_listen_request { | |
455 | + wait_queue_head_t waitq; | |
456 | + int result; | |
457 | + int waiting; | |
458 | + unsigned short tag; | |
459 | + int nodeid; | |
460 | + struct list_head list; | |
461 | +}; | |
462 | + | |
463 | +/* Messages from membership services */ | |
464 | +#define CLUSTER_MEM_JOINCONF 1 | |
465 | +#define CLUSTER_MEM_JOINREQ 2 | |
466 | +#define CLUSTER_MEM_LEAVE 3 | |
467 | +#define CLUSTER_MEM_HELLO 4 | |
468 | +#define CLUSTER_MEM_KILL 5 | |
469 | +#define CLUSTER_MEM_JOINACK 6 | |
470 | +#define CLUSTER_MEM_ENDTRANS 7 | |
471 | +#define CLUSTER_MEM_RECONFIG 8 | |
472 | +#define CLUSTER_MEM_MASTERVIEW 9 | |
473 | +#define CLUSTER_MEM_STARTTRANS 10 | |
474 | +#define CLUSTER_MEM_JOINREJ 11 | |
475 | +#define CLUSTER_MEM_VIEWACK 12 | |
476 | +#define CLUSTER_MEM_STARTACK 13 | |
477 | +#define CLUSTER_MEM_TRANSITION 14 | |
478 | +#define CLUSTER_MEM_NEWCLUSTER 15 | |
479 | +#define CLUSTER_MEM_CONFACK 16 | |
480 | +#define CLUSTER_MEM_NOMINATE 17 | |
481 | + | |
482 | +/* Parameters for RECONFIG command */ | |
483 | +#define RECONFIG_PARAM_EXPECTED_VOTES 1 | |
484 | +#define RECONFIG_PARAM_NODE_VOTES 2 | |
485 | +#define RECONFIG_PARAM_CONFIG_VERSION 3 | |
486 | + | |
487 | +/* Data associated with an outgoing socket */ | |
488 | +struct cl_socket { | |
489 | + struct file *file; /* The real file */ | |
490 | + struct socket *socket; /* The real sock */ | |
491 | + struct cl_multicast_sock multicast_info; | |
492 | + int num_nodes; /* On this link */ | |
493 | + int retransmit_count; | |
494 | +}; | |
495 | + | |
496 | +/* There's one of these for each node in the cluster */ | |
497 | +struct cluster_node { | |
498 | + struct list_head list; | |
499 | + char *name; /* Node/host name of node */ | |
500 | + struct list_head addr_list; | |
501 | + int us; /* This node is us */ | |
502 | + unsigned int node_id; /* Unique node ID */ | |
503 | + nodestate_t state; | |
504 | + unsigned short last_seq_recv; | |
505 | + unsigned short last_seq_acked; | |
506 | + unsigned short last_seq_sent; | |
507 | + unsigned int votes; | |
508 | + unsigned int expected_votes; | |
509 | + unsigned int leave_reason; | |
510 | + unsigned int incarnation; /* Incremented each time a node joins | |
511 | + * the cluster */ | |
512 | + unsigned long last_hello; /* Jiffies */ | |
513 | +}; | |
514 | + | |
515 | +/* This is how we keep a list of user processes that are listening for cluster | |
516 | + * membership events */ | |
517 | +struct notify_struct { | |
518 | + struct list_head list; | |
519 | + pid_t pid; | |
520 | + int signal; | |
521 | +}; | |
522 | + | |
523 | +/* This is how we keep a list of kernel callbacks that are registered for | |
524 | + * cluster membership events */ | |
525 | +struct kernel_notify_struct { | |
526 | + struct list_head list; | |
527 | + void (*callback) (kcl_callback_reason, long arg); | |
528 | +}; | |
529 | + | |
530 | +/* A message waiting to be sent */ | |
531 | +struct queued_message { | |
532 | + struct list_head list; | |
533 | + | |
534 | + struct socket *socket; | |
535 | + struct sockaddr_cl addr; | |
536 | + int addr_len; | |
537 | + int msg_len; | |
538 | + unsigned char port; | |
539 | + unsigned int flags; | |
540 | + char msg_buffer[MAX_CLUSTER_MESSAGE]; | |
541 | +}; | |
542 | + | |
543 | +/* A barrier */ | |
544 | +struct cl_barrier { | |
545 | + struct list_head list; | |
546 | + | |
547 | + char name[MAX_BARRIER_NAME_LEN]; | |
548 | + unsigned int flags; | |
549 | + enum { BARRIER_STATE_WAITING, BARRIER_STATE_INACTIVE, | |
550 | + BARRIER_STATE_COMPLETE } state; | |
551 | + unsigned int expected_nodes; | |
552 | + unsigned int registered_nodes; | |
553 | + atomic_t got_nodes; | |
554 | + atomic_t completed_nodes; | |
555 | + unsigned int inuse; | |
556 | + unsigned int waitsent; | |
557 | + unsigned int phase; /* Completion phase */ | |
558 | + unsigned int endreason; /* Reason we were woken, usually 0 */ | |
559 | + unsigned long timeout; /* In seconds */ | |
560 | + | |
561 | + void (*callback) (char *name, int status); | |
562 | + wait_queue_head_t waitq; | |
563 | + struct semaphore lock; /* To synch with cnxman messages */ | |
564 | + spinlock_t phase2_spinlock; /* Need to synchronise with timer | |
565 | + * interrupts */ | |
566 | + struct timer_list timer; | |
567 | +}; | |
568 | + | |
569 | +/* Cluster protocol commands sent to port 0 */ | |
570 | +#define CLUSTER_CMD_ACK 1 | |
571 | +#define CLUSTER_CMD_LISTENREQ 2 | |
572 | +#define CLUSTER_CMD_LISTENRESP 3 | |
573 | +#define CLUSTER_CMD_PORTCLOSED 4 | |
574 | +#define CLUSTER_CMD_BARRIER 5 | |
575 | + | |
576 | +extern struct cluster_node *find_node_by_addr(unsigned char *addr, | |
577 | + int addr_len); | |
578 | +extern struct cluster_node *find_node_by_nodeid(unsigned int id); | |
579 | +extern struct cluster_node *find_node_by_name(char *name); | |
580 | +extern void set_quorate(int); | |
581 | +extern void notify_kernel_listeners(kcl_callback_reason reason, long arg); | |
582 | +extern void notify_listeners(void); | |
583 | +extern void free_nodeid_array(void); | |
584 | +extern int send_reconfigure(int param, unsigned int value); | |
585 | +extern int calculate_quorum(int, int, int *); | |
586 | +extern void recalculate_quorum(int); | |
587 | +extern int send_leave(unsigned char); | |
588 | +extern int get_quorum(void); | |
589 | +extern void set_votes(int, int); | |
590 | +extern void kcl_wait_for_all_acks(void); | |
591 | +extern char *membership_state(char *, int); | |
592 | +extern void a_node_just_died(struct cluster_node *node); | |
593 | +extern void check_barrier_returns(void); | |
594 | +extern int in_transition(void); | |
595 | +extern void get_local_addresses(struct cluster_node *node); | |
596 | +extern int add_node_address(struct cluster_node *node, unsigned char *addr, int len); | |
597 | +extern void create_proc_entries(void); | |
598 | +extern void cleanup_proc_entries(void); | |
599 | +extern unsigned int get_highest_nodeid(void); | |
600 | +extern int allocate_nodeid_array(void); | |
601 | +extern void queue_oob_skb(struct socket *sock, int cmd); | |
602 | +extern int new_temp_nodeid(char *addr, int addrlen); | |
603 | +extern int get_addr_from_temp_nodeid(int nodeid, char *addr, int *addrlen); | |
604 | +extern void remove_temp_nodeid(int nodeid); | |
605 | +extern inline char *print_addr(unsigned char *addr, int len, char *buf) | |
606 | +{ | |
607 | + int i; | |
608 | + int ptr = 0; | |
609 | + | |
610 | + for (i = 0; i < len; i++) | |
611 | + ptr += sprintf(buf + ptr, "%02x ", addr[i]); | |
612 | + | |
613 | + return buf; | |
614 | +} | |
615 | + | |
616 | +#define MAX_ADDR_PRINTED_LEN (address_length*3 + 1) | |
617 | + | |
618 | +/* Debug enabling macros. Sorry about the C++ comments but they're easier to | |
619 | + * get rid of than C ones... */ | |
620 | + | |
621 | +// #define DEBUG_MEMB | |
622 | +// #define DEBUG_COMMS | |
623 | +// #define DEBUG_BARRIER | |
624 | + | |
625 | +/* Debug macros */ | |
626 | +#ifdef DEBUG_COMMS | |
627 | +#define P_COMMS(fmt, args...) printk(KERN_DEBUG "cman comms: " fmt, ## args) | |
628 | +#else | |
629 | +#define P_COMMS(fmt, args...) | |
630 | +#endif | |
631 | + | |
632 | +#ifdef DEBUG_BARRIER | |
633 | +#define P_BARRIER(fmt, args...) printk(KERN_DEBUG "cman barrier: " fmt, ## args) | |
634 | +#else | |
635 | +#define P_BARRIER(fmt, args...) | |
636 | +#endif | |
637 | + | |
638 | +#ifdef DEBUG_MEMB | |
639 | +#define P_MEMB(fmt, args...) printk(KERN_DEBUG "cman memb: " fmt, ## args) | |
640 | +#define C_MEMB(fmt, args...) printk(fmt, ## args) | |
641 | +#else | |
642 | +#define P_MEMB(fmt, args...) | |
643 | +#define C_MEMB(fmt, args...) | |
644 | +#endif | |
645 | + | |
646 | +#endif /* __KERNEL */ | |
647 | + | |
648 | +#endif | |
649 | diff -urN linux-orig/cluster/cman/cnxman.c linux-patched/cluster/cman/cnxman.c | |
650 | --- linux-orig/cluster/cman/cnxman.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 651 | +++ linux-patched/cluster/cman/cnxman.c 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 652 | @@ -0,0 +1,4080 @@ |
653 | +/****************************************************************************** | |
654 | +******************************************************************************* | |
655 | +** | |
656 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
657 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
658 | +** | |
659 | +** This copyrighted material is made available to anyone wishing to use, | |
660 | +** modify, copy, or redistribute it subject to the terms and conditions | |
661 | +** of the GNU General Public License v.2. | |
662 | +** | |
663 | +******************************************************************************* | |
664 | +******************************************************************************/ | |
665 | + | |
666 | +#define EXPORT_SYMTAB | |
667 | +#include <linux/init.h> | |
668 | +#include <linux/socket.h> | |
669 | +#include <linux/kernel.h> | |
670 | +#include <linux/sched.h> | |
671 | +#include <linux/file.h> | |
672 | +#include <linux/utsname.h> | |
673 | +#include <net/sock.h> | |
674 | +#include <linux/proc_fs.h> | |
675 | +#include <linux/poll.h> | |
676 | +#include <linux/module.h> | |
677 | +#include <linux/list.h> | |
678 | +#include <cluster/cnxman.h> | |
679 | +#include <cluster/service.h> | |
680 | + | |
681 | +#include "cnxman-private.h" | |
682 | +#include "sm_control.h" | |
683 | +#include "sm_user.h" | |
684 | +#include "config.h" | |
685 | + | |
686 | +#define CMAN_RELEASE_NAME "<CVS>" | |
687 | + | |
688 | +static int __cl_setsockopt(struct socket *sock, int level, int optname, | |
689 | + char *optval, int optlen, int flags); | |
690 | +static int __cl_getsockopt(struct socket *sock, int level, int optname, | |
691 | + char *optval, int *optlen, int flags); | |
692 | +static void send_to_userport(struct cl_comms_socket *csock, char *data, int len, | |
693 | + char *addr, int addrlen); | |
694 | +static int cl_sendack(struct cl_comms_socket *sock, unsigned short seq, | |
695 | + int addr_len, char *addr, unsigned char remport, | |
696 | + unsigned char flag); | |
697 | +static void send_listen_request(int nodeid, unsigned char port); | |
698 | +static void send_listen_response(struct cl_comms_socket *csock, int nodeid, | |
699 | + unsigned char port, unsigned short tag); | |
700 | +static void resend_last_message(void); | |
701 | +static void start_ack_timer(void); | |
702 | +static int send_queued_message(struct queued_message *qmsg); | |
703 | +static void send_port_close_oob(unsigned char port); | |
704 | +static void post_close_oob(unsigned char port, int nodeid); | |
705 | +static void process_barrier_msg(struct cl_barriermsg *msg, | |
706 | + struct cluster_node *node); | |
707 | +static struct cl_barrier *find_barrier(char *name); | |
708 | +static void node_shutdown(void); | |
709 | +static void node_cleanup(void); | |
710 | +static int send_or_queue_message(void *buf, int len, struct sockaddr_cl *caddr, | |
711 | + unsigned char port); | |
712 | +static struct cl_comms_socket *get_next_interface(struct cl_comms_socket *cur); | |
713 | +static void check_for_unacked_nodes(void); | |
714 | +static void free_cluster_sockets(void); | |
715 | +static uint16_t generate_cluster_id(char *name); | |
716 | + | |
717 | +static int is_valid_temp_nodeid(int nodeid); | |
718 | + | |
719 | +extern int start_membership_services(pid_t); | |
720 | +extern int kcl_leave_cluster(int remove); | |
721 | +extern int send_kill(int nodeid); | |
722 | + | |
723 | +static struct proto_ops cl_proto_ops; | |
724 | +static struct sock *master_sock; | |
725 | +static kmem_cache_t *cluster_sk_cachep; | |
726 | + | |
727 | +/* Pointer to the pseudo node that maintains quorum in a 2node system */ | |
728 | +struct cluster_node *quorum_device = NULL; | |
729 | + | |
730 | +/* Array of "ports" allocated. This is just a list of pointers to the sock that | |
731 | + * has this port bound. Speed is a major issue here so 1-2K of allocated | |
732 | + * storage is worth sacrificing. Port 0 is reserved for protocol messages */ | |
733 | +static struct sock *port_array[256]; | |
734 | +static struct semaphore port_array_lock; | |
735 | + | |
736 | +/* Our cluster name & number */ | |
737 | +unsigned short cluster_id; | |
738 | +char cluster_name[MAX_CLUSTER_NAME_LEN+1]; | |
739 | + | |
740 | +/* Two-node mode: causes cluster to remain quorate if one of two nodes fails. | |
741 | + * No more than two nodes are permitted to join the cluster. */ | |
742 | +unsigned short two_node; | |
743 | + | |
744 | +/* Cluster configuration version that must be the same among members. */ | |
745 | +unsigned int config_version; | |
746 | + | |
747 | +/* Reference counting for cluster applications */ | |
748 | +atomic_t use_count; | |
749 | + | |
750 | +/* Length of sockaddr address for our comms protocol */ | |
751 | +unsigned int address_length; | |
752 | + | |
753 | +/* Message sending */ | |
754 | +static unsigned short cur_seq; /* Last message sent */ | |
755 | +static unsigned int ack_count; /* Number of acks received for message | |
756 | + * 'cur_seq' */ | |
757 | +static unsigned int acks_expected; /* Number of acks we expect to receive */ | |
758 | +static struct semaphore send_lock; | |
759 | +static struct timer_list ack_timer; | |
760 | + | |
761 | +/* Saved packet information in case we need to resend it */ | |
762 | +static char saved_msg_buffer[MAX_CLUSTER_MESSAGE]; | |
763 | +static int saved_msg_len; | |
764 | +static int retry_count; | |
765 | + | |
766 | +/* Task variables */ | |
767 | +static pid_t kcluster_pid; | |
768 | +static pid_t membership_pid; | |
769 | +extern int quit_threads; | |
770 | + | |
771 | +wait_queue_head_t cnxman_waitq; | |
772 | + | |
773 | +/* Variables owned by membership services */ | |
774 | +extern int cluster_members; | |
775 | +extern struct list_head cluster_members_list; | |
776 | +extern struct semaphore cluster_members_lock; | |
777 | +extern int we_are_a_cluster_member; | |
778 | +extern int cluster_is_quorate; | |
779 | +extern struct cluster_node *us; | |
780 | +extern struct list_head new_dead_node_list; | |
781 | +extern struct semaphore new_dead_node_lock; | |
782 | +extern char nodename[]; | |
783 | + | |
784 | +/* A list of processes listening for membership events */ | |
785 | +static struct list_head event_listener_list; | |
786 | +static struct semaphore event_listener_lock; | |
787 | + | |
788 | +/* A list of kernel callbacks listening for membership events */ | |
789 | +static struct list_head kernel_listener_list; | |
790 | +static struct semaphore kernel_listener_lock; | |
791 | + | |
792 | +/* A list of sockets we are listening on (and can transmit on...later) */ | |
793 | +static struct list_head socket_list; | |
794 | + | |
795 | +/* A list of all open cluster client sockets */ | |
796 | +static struct list_head client_socket_list; | |
797 | +static struct semaphore client_socket_lock; | |
798 | + | |
799 | +/* A list of all current barriers */ | |
800 | +static struct list_head barrier_list; | |
801 | +static struct semaphore barrier_list_lock; | |
802 | + | |
803 | +/* When a socket is read for reading it goes on this queue */ | |
804 | +static spinlock_t active_socket_lock; | |
805 | +static struct list_head active_socket_list; | |
806 | + | |
807 | +/* If the cnxman process is running and available for work */ | |
808 | +atomic_t cnxman_running; | |
809 | + | |
810 | +/* Fkags set by timers etc for the mainloop to detect and act upon */ | |
811 | +static unsigned long mainloop_flags; | |
812 | + | |
813 | +#define ACK_TIMEOUT 1 | |
814 | +#define RESEND_NEEDED 2 | |
815 | + | |
816 | +/* A queue of messages waiting to be sent. If kcl_sendmsg is called outside of | |
817 | + * process context then the messages get put in here */ | |
818 | +static struct list_head messages_list; | |
819 | +static struct semaphore messages_list_lock; | |
820 | + | |
821 | +static struct semaphore start_thread_sem; | |
822 | + | |
823 | +/* List of outstanding ISLISTENING requests */ | |
824 | +static struct list_head listenreq_list; | |
825 | +static struct semaphore listenreq_lock; | |
826 | + | |
827 | +/* Any sending requests wait on this queue if necessary (eg inquorate, waiting | |
828 | + * ACK) */ | |
829 | +static DECLARE_WAIT_QUEUE_HEAD(socket_waitq); | |
830 | + | |
831 | +/* Wait for thread to exit properly */ | |
832 | +struct completion cluster_thread_comp; | |
833 | +struct completion member_thread_comp; | |
834 | + | |
835 | +/* The resend delay to use, We increase this geometrically(word?) each time a | |
836 | + * send is delayed. in deci-seconds */ | |
837 | +static int resend_delay = 1; | |
838 | + | |
839 | +/* Highest numbered interface and the current default */ | |
840 | +static int num_interfaces = 0; | |
841 | +static struct cl_comms_socket *current_interface = NULL; | |
842 | + | |
843 | +struct temp_node | |
844 | +{ | |
845 | + int nodeid; | |
846 | + char addr[sizeof(struct sockaddr_in6)]; | |
847 | + int addrlen; | |
848 | + struct list_head list; | |
849 | +}; | |
850 | +static struct list_head tempnode_list; | |
851 | +static struct semaphore tempnode_lock; | |
852 | + | |
853 | +/* Wake up any processes that are waiting to send. This is usually called when | |
854 | + * all the ACKs have been gathered up or when a node has left the cluster | |
855 | + * unexpectedly and we reckon there are no more acks to collect */ | |
856 | +static void unjam(void) | |
857 | +{ | |
858 | + wake_up_interruptible(&socket_waitq); | |
859 | + wake_up_interruptible(&cnxman_waitq); | |
860 | +} | |
861 | + | |
862 | +/* Used by the data_ready routine to locate a connection given the socket */ | |
863 | +static inline struct cl_comms_socket *find_comms_by_sock(struct sock *sk) | |
864 | +{ | |
865 | + struct list_head *conlist; | |
866 | + | |
867 | + list_for_each(conlist, &socket_list) { | |
868 | + struct cl_comms_socket *clsock = | |
869 | + list_entry(conlist, struct cl_comms_socket, list); | |
870 | + if (clsock->sock->sk == sk) { | |
871 | + return clsock; | |
872 | + } | |
873 | + } | |
874 | + return NULL; | |
875 | +} | |
876 | + | |
877 | +/* Data available on socket */ | |
878 | +static void cnxman_data_ready(struct sock *sk, int count_unused) | |
879 | +{ | |
880 | + struct cl_comms_socket *clsock = find_comms_by_sock(sk); | |
881 | + | |
882 | + if (clsock == NULL) /* ASSERT ?? */ | |
883 | + return; | |
884 | + | |
885 | + /* If we're already on the list then don't do it again */ | |
886 | + if (test_and_set_bit(1, &clsock->active)) | |
887 | + return; | |
888 | + | |
889 | + spin_lock_irq(&active_socket_lock); | |
890 | + list_add(&clsock->active_list, &active_socket_list); | |
891 | + spin_unlock_irq(&active_socket_lock); | |
892 | + | |
893 | + wake_up_interruptible(&cnxman_waitq); | |
894 | +} | |
895 | + | |
896 | +static int receive_message(struct cl_comms_socket *csock, char *iobuf) | |
897 | +{ | |
898 | + struct msghdr msg; | |
899 | + struct iovec iov; | |
900 | + struct sockaddr_in6 sin; | |
901 | + int len; | |
902 | + mm_segment_t fs; | |
903 | + | |
904 | + memset(&sin, 0, sizeof (sin)); | |
905 | + | |
906 | + msg.msg_control = NULL; | |
907 | + msg.msg_controllen = 0; | |
908 | + msg.msg_iovlen = 1; | |
909 | + msg.msg_iov = &iov; | |
910 | + msg.msg_name = &sin; | |
911 | + msg.msg_namelen = sizeof (sin); | |
912 | + msg.msg_flags = 0; | |
913 | + | |
914 | + iov.iov_len = MAX_CLUSTER_MESSAGE; | |
915 | + iov.iov_base = iobuf; | |
916 | + | |
917 | + fs = get_fs(); | |
918 | + set_fs(get_ds()); | |
919 | + | |
920 | + len = sock_recvmsg(csock->sock, &msg, MAX_CLUSTER_MESSAGE, MSG_DONTWAIT); | |
921 | + set_fs(fs); | |
922 | + | |
923 | + if (len > 0) { | |
924 | + if (len > MAX_CLUSTER_MESSAGE) { | |
925 | + printk(KERN_CRIT CMAN_NAME | |
926 | + ": %d byte message far too big\n", len); | |
927 | + return 0; | |
928 | + } | |
929 | + send_to_userport(csock, iobuf, len, msg.msg_name, msg.msg_namelen); | |
930 | + } | |
931 | + else { | |
932 | + if (len != -EAGAIN) | |
933 | + printk(KERN_CRIT CMAN_NAME ": recvmsg failed: %d\n", | |
934 | + len); | |
935 | + } | |
936 | + return len; | |
937 | +} | |
938 | + | |
939 | +static int cluster_kthread(void *unused) | |
940 | +{ | |
941 | + int len; | |
942 | + char *iobuf; | |
943 | + struct list_head *socklist; | |
944 | + struct cl_comms_socket *csock; | |
945 | + wait_queue_t cnxman_waitq_head; | |
946 | + sigset_t tmpsig; | |
947 | + | |
948 | + daemonize("cman_comms"); | |
949 | + | |
950 | + /* Block everything but SIGKILL/SIGSTOP/SIGTERM */ | |
951 | + siginitset(&tmpsig, SIGKILL | SIGSTOP | SIGTERM); | |
952 | + sigprocmask(SIG_BLOCK, &tmpsig, NULL); | |
953 | + | |
954 | + /* This is the waitq we can wake the process up with */ | |
955 | + init_waitqueue_head(&cnxman_waitq); | |
956 | + init_waitqueue_entry(&cnxman_waitq_head, current); | |
957 | + add_wait_queue(&cnxman_waitq, &cnxman_waitq_head); | |
958 | + | |
959 | + set_user_nice(current, -6); | |
960 | + | |
961 | + /* Allow the sockets to start receiving */ | |
962 | + list_for_each(socklist, &socket_list) { | |
963 | + csock = list_entry(socklist, struct cl_comms_socket, list); | |
964 | + | |
965 | + clear_bit(1, &csock->active); | |
966 | + } | |
967 | + | |
968 | + iobuf = kmalloc(MAX_CLUSTER_MESSAGE, GFP_KERNEL); | |
969 | + if (!iobuf) { | |
970 | + printk(KERN_CRIT CMAN_NAME | |
971 | + ": Cannot allocate receive buffer for cluster comms\n"); | |
972 | + return -1; | |
973 | + } | |
974 | + | |
975 | + complete(&cluster_thread_comp); | |
976 | + | |
977 | + for (;;) { | |
978 | + struct list_head *temp; | |
979 | + | |
980 | + /* Wait for activity on any of the sockets */ | |
981 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
982 | + | |
983 | + if (list_empty(&active_socket_list)) | |
984 | + schedule(); | |
985 | + set_task_state(current, TASK_RUNNING); | |
986 | + | |
987 | + if (quit_threads) | |
988 | + break; | |
989 | + | |
990 | + if (test_and_clear_bit(ACK_TIMEOUT, &mainloop_flags)) { | |
991 | + check_for_unacked_nodes(); | |
992 | + } | |
993 | + | |
994 | + /* Now receive any messages waiting for us */ | |
995 | + spin_lock_irq(&active_socket_lock); | |
996 | + list_for_each_safe(socklist, temp, &active_socket_list) { | |
997 | + csock = | |
998 | + list_entry(socklist, struct cl_comms_socket, | |
999 | + active_list); | |
1000 | + | |
1001 | + list_del(&csock->active_list); | |
1002 | + clear_bit(1, &csock->active); | |
1003 | + | |
1004 | + spin_unlock_irq(&active_socket_lock); | |
1005 | + | |
1006 | + do { | |
1007 | + len = receive_message(csock, iobuf); | |
1008 | + } | |
1009 | + while (len > 0); | |
1010 | + | |
1011 | + spin_lock_irq(&active_socket_lock); | |
1012 | + | |
1013 | + if (len == 0) | |
1014 | + break; /* EOF on socket */ | |
1015 | + } | |
1016 | + spin_unlock_irq(&active_socket_lock); | |
1017 | + | |
1018 | + /* Resend any unacked messages */ | |
1019 | + if (test_and_clear_bit(RESEND_NEEDED, &mainloop_flags) | |
1020 | + && acks_expected) { | |
1021 | + resend_last_message(); | |
1022 | + } | |
1023 | + | |
1024 | + /* Send any queued messages */ | |
1025 | + if (acks_expected == 0) { | |
1026 | + struct list_head *temp; | |
1027 | + struct list_head *msglist; | |
1028 | + | |
1029 | + down(&messages_list_lock); | |
1030 | + list_for_each_safe(msglist, temp, &messages_list) { | |
1031 | + struct queued_message *qmsg = | |
1032 | + list_entry(msglist, struct queued_message, | |
1033 | + list); | |
1034 | + int status = send_queued_message(qmsg); | |
1035 | + | |
1036 | + if (status >= 0) { | |
1037 | + /* Suceeded, remove it from the queue */ | |
1038 | + list_del(&qmsg->list); | |
1039 | + kfree(qmsg); | |
1040 | + } | |
1041 | + /* Did it fail horribly ?? */ | |
1042 | + if (status < 0 && status != -EAGAIN) { | |
1043 | + printk(KERN_INFO CMAN_NAME | |
1044 | + ": send_queued_message failed, error %d\n", | |
1045 | + status); | |
1046 | + list_del(&qmsg->list); | |
1047 | + kfree(qmsg); | |
1048 | + } | |
1049 | + break; /* Only send one message at a time */ | |
1050 | + } | |
1051 | + up(&messages_list_lock); | |
1052 | + } | |
1053 | + | |
1054 | + if (signal_pending(current)) | |
1055 | + break; | |
1056 | + } | |
1057 | + P_COMMS("closing down\n"); | |
1058 | + | |
1059 | + if (we_are_a_cluster_member) | |
1060 | + send_leave(us->leave_reason); | |
1061 | + | |
1062 | + kfree(iobuf); | |
1063 | + quit_threads = 1; /* force other thread to die too */ | |
1064 | + node_shutdown(); | |
1065 | + | |
1066 | + if (timer_pending(&ack_timer)) | |
1067 | + del_timer(&ack_timer); | |
1068 | + | |
1069 | + /* Wait for membership thread to die */ | |
1070 | + wait_for_completion(&member_thread_comp); | |
1071 | + | |
1072 | + node_cleanup(); | |
1073 | + | |
1074 | + complete(&cluster_thread_comp); | |
1075 | + return 0; | |
1076 | +} | |
1077 | + | |
1078 | +void notify_kernel_listeners(kcl_callback_reason reason, long arg) | |
1079 | +{ | |
1080 | + struct kernel_notify_struct *knotify; | |
1081 | + struct list_head *proclist; | |
1082 | + | |
1083 | + down(&kernel_listener_lock); | |
1084 | + list_for_each(proclist, &kernel_listener_list) { | |
1085 | + knotify = | |
1086 | + list_entry(proclist, struct kernel_notify_struct, list); | |
1087 | + knotify->callback(reason, arg); | |
1088 | + } | |
1089 | + up(&kernel_listener_lock); | |
1090 | +} | |
1091 | + | |
1092 | +static void check_for_unacked_nodes() | |
1093 | +{ | |
1094 | + struct list_head *nodelist; | |
1095 | + struct cluster_node *node; | |
1096 | + | |
1097 | + clear_bit(RESEND_NEEDED, &mainloop_flags); | |
1098 | + retry_count = 0; | |
1099 | + | |
1100 | + P_COMMS("Retry count exceeded -- looking for dead node\n"); | |
1101 | + | |
1102 | + /* Node did not ACK a message after <n> tries, remove it from the | |
1103 | + * cluster */ | |
1104 | + down(&cluster_members_lock); | |
1105 | + list_for_each(nodelist, &cluster_members_list) { | |
1106 | + node = list_entry(nodelist, struct cluster_node, list); | |
1107 | + | |
1108 | + P_COMMS | |
1109 | + ("checking node %s: last_acked = %d, last_seq_sent = %d\n", | |
1110 | + node->name, node->last_seq_acked, node->last_seq_sent); | |
1111 | + if (node->state != NODESTATE_DEAD | |
1112 | + && node->last_seq_acked != node->last_seq_sent && !node->us) { | |
1113 | + printk(KERN_WARNING CMAN_NAME | |
1114 | + ": node %s is not responding - removing from the cluster\n", | |
1115 | + node->name); | |
1116 | + | |
1117 | + /* Start a state transition */ | |
1118 | + a_node_just_died(node); | |
1119 | + } | |
1120 | + } | |
1121 | + up(&cluster_members_lock); | |
1122 | + acks_expected = ack_count = 0; | |
1123 | + unjam(); | |
1124 | + return; | |
1125 | +} | |
1126 | + | |
1127 | +static void ack_timer_fn(unsigned long arg) | |
1128 | +{ | |
1129 | + P_COMMS("%ld: ack_timer fired, retries=%d\n", jiffies, retry_count); | |
1130 | + | |
1131 | + /* Too many retries ? */ | |
1132 | + if (++retry_count > MAX_RETRIES) { | |
1133 | + set_bit(ACK_TIMEOUT, &mainloop_flags); | |
1134 | + wake_up_interruptible(&cnxman_waitq); | |
1135 | + } | |
1136 | + else { | |
1137 | + /* Resend last message */ | |
1138 | + set_bit(RESEND_NEEDED, &mainloop_flags); | |
1139 | + wake_up_interruptible(&cnxman_waitq); | |
1140 | + } | |
1141 | +} | |
1142 | + | |
1143 | +/* Called to resend a packet if sock_sendmsg was busy */ | |
1144 | +static void short_timer_fn(unsigned long arg) | |
1145 | +{ | |
1146 | + P_COMMS("short_timer fired\n"); | |
1147 | + | |
1148 | + /* Resend last message */ | |
1149 | + resend_delay <<= 1; | |
1150 | + set_bit(RESEND_NEEDED, &mainloop_flags); | |
1151 | + wake_up_interruptible(&cnxman_waitq); | |
1152 | +} | |
1153 | + | |
1154 | +static void start_ack_timer() | |
1155 | +{ | |
1156 | + ack_timer.function = ack_timer_fn; | |
1157 | + ack_timer.data = 0L; | |
1158 | + mod_timer(&ack_timer, jiffies + HZ); | |
1159 | +} | |
1160 | + | |
1161 | +static void start_short_timer(void) | |
1162 | +{ | |
1163 | + ack_timer.function = short_timer_fn; | |
1164 | + ack_timer.data = 0L; | |
1165 | + mod_timer(&ack_timer, jiffies + (resend_delay * HZ)); | |
1166 | +} | |
1167 | + | |
1168 | + | |
1169 | +static struct cl_waiting_listen_request *find_listen_request(unsigned short tag) | |
1170 | +{ | |
1171 | + struct list_head *llist; | |
1172 | + struct cl_waiting_listen_request *listener; | |
1173 | + | |
1174 | + down(&listenreq_lock); | |
1175 | + list_for_each(llist, &listenreq_list) { | |
1176 | + listener = | |
1177 | + list_entry(llist, struct cl_waiting_listen_request, list); | |
1178 | + if (listener->tag == tag) { | |
1179 | + up(&listenreq_lock); | |
1180 | + return listener; | |
1181 | + } | |
1182 | + } | |
1183 | + up(&listenreq_lock); | |
1184 | + return NULL; | |
1185 | +} | |
1186 | + | |
1187 | +static void process_cnxman_message(struct cl_comms_socket *csock, char *data, | |
1188 | + int len, char *addr, int addrlen, | |
1189 | + struct cluster_node *rem_node) | |
1190 | +{ | |
1191 | + struct cl_protmsg *msg = (struct cl_protmsg *) data; | |
1192 | + struct cl_protheader *header = (struct cl_protheader *) data; | |
1193 | + struct cl_ackmsg *ackmsg; | |
1194 | + struct cl_listenmsg *listenmsg; | |
1195 | + struct cl_closemsg *closemsg; | |
1196 | + struct cl_barriermsg *barriermsg; | |
1197 | + struct cl_waiting_listen_request *listen_request; | |
1198 | + | |
1199 | + P_COMMS("Message on port 0 is %d\n", msg->cmd); | |
1200 | + switch (msg->cmd) { | |
1201 | + case CLUSTER_CMD_ACK: | |
1202 | + ackmsg = (struct cl_ackmsg *) data; | |
1203 | + | |
1204 | + if (ackmsg->aflags & 1) { | |
1205 | + if (net_ratelimit()) | |
1206 | + printk(KERN_INFO CMAN_NAME | |
1207 | + ": WARNING no listener for port %d on node %s\n", | |
1208 | + ackmsg->remport, rem_node->name); | |
1209 | + } | |
1210 | + P_COMMS("Got ACK from %s. seq=%d (cur=%d)\n", | |
1211 | + rem_node ? rem_node->name : "Unknown", | |
1212 | + le16_to_cpu(ackmsg->seq), cur_seq); | |
1213 | + | |
1214 | + if (rem_node && rem_node->state != NODESTATE_DEAD) { | |
1215 | + /* This copes with duplicate acks from a multipathed | |
1216 | + * host */ | |
1217 | + if (rem_node->last_seq_acked != | |
1218 | + le16_to_cpu(ackmsg->seq)) { | |
1219 | + rem_node->last_seq_acked = | |
1220 | + le16_to_cpu(ackmsg->seq); | |
1221 | + | |
1222 | + /* Got em all */ | |
1223 | + if (++ack_count >= acks_expected) { | |
1224 | + | |
1225 | + /* Cancel the timer */ | |
1226 | + del_timer(&ack_timer); | |
1227 | + acks_expected = 0; | |
1228 | + unjam(); | |
1229 | + } | |
1230 | + } | |
1231 | + } | |
1232 | + else { | |
1233 | + if (cluster_members) { | |
1234 | +#ifdef DEBUG_COMMS | |
1235 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
1236 | + | |
1237 | + printk(KERN_INFO CMAN_NAME | |
1238 | + ": got ack from unknown or dead node: %s\n", | |
1239 | + print_addr(addr, addrlen, buf)); | |
1240 | +#endif | |
1241 | + } | |
1242 | + } | |
1243 | + break; | |
1244 | + | |
1245 | + /* Return 1 if we have a listener on this port, 0 if not */ | |
1246 | + case CLUSTER_CMD_LISTENREQ: | |
1247 | + listenmsg = | |
1248 | + (struct cl_listenmsg *) (data + | |
1249 | + sizeof (struct cl_protheader)); | |
1250 | + cl_sendack(csock, header->seq, addrlen, addr, header->port, 0); | |
1251 | + send_listen_response(csock, le32_to_cpu(header->srcid), | |
1252 | + listenmsg->target_port, listenmsg->tag); | |
1253 | + break; | |
1254 | + | |
1255 | + case CLUSTER_CMD_LISTENRESP: | |
1256 | + /* Wake up process waiting for listen response */ | |
1257 | + listenmsg = | |
1258 | + (struct cl_listenmsg *) (data + | |
1259 | + sizeof (struct cl_protheader)); | |
1260 | + cl_sendack(csock, header->seq, addrlen, addr, header->port, 0); | |
1261 | + listen_request = find_listen_request(listenmsg->tag); | |
1262 | + if (listen_request) { | |
1263 | + listen_request->result = listenmsg->listening; | |
1264 | + listen_request->waiting = 0; | |
1265 | + wake_up_interruptible(&listen_request->waitq); | |
1266 | + } | |
1267 | + break; | |
1268 | + | |
1269 | + case CLUSTER_CMD_PORTCLOSED: | |
1270 | + closemsg = | |
1271 | + (struct cl_closemsg *) (data + | |
1272 | + sizeof (struct cl_protheader)); | |
1273 | + cl_sendack(csock, header->seq, addrlen, addr, header->port, 0); | |
1274 | + post_close_oob(closemsg->port, le32_to_cpu(header->srcid)); | |
1275 | + break; | |
1276 | + | |
1277 | + case CLUSTER_CMD_BARRIER: | |
1278 | + barriermsg = | |
1279 | + (struct cl_barriermsg *) (data + | |
1280 | + sizeof (struct cl_protheader)); | |
1281 | + cl_sendack(csock, header->seq, addrlen, addr, header->port, 0); | |
1282 | + process_barrier_msg(barriermsg, rem_node); | |
1283 | + break; | |
1284 | + | |
1285 | + default: | |
1286 | + printk(KERN_ERR CMAN_NAME | |
1287 | + ": Unknown protocol message %d received\n", msg->cmd); | |
1288 | + break; | |
1289 | + | |
1290 | + } | |
1291 | + return; | |
1292 | +} | |
1293 | + | |
1294 | +static void send_to_userport(struct cl_comms_socket *csock, char *data, int len, | |
1295 | + char *addr, int addrlen) | |
1296 | +{ | |
1297 | + int err; | |
1298 | + struct cl_protheader *header = (struct cl_protheader *) data; | |
1299 | + struct cluster_node *rem_node = | |
1300 | + find_node_by_nodeid(le32_to_cpu(header->srcid)); | |
1301 | + struct sk_buff *skb = NULL; | |
1302 | + | |
1303 | + P_COMMS | |
1304 | + ("seen message, from %d for %d, sequence num = %d, rem_node=%p, state=%d\n", | |
1305 | + le32_to_cpu(header->srcid), le32_to_cpu(header->tgtid), | |
1306 | + le16_to_cpu(header->seq), rem_node, | |
1307 | + rem_node ? rem_node->state : -1); | |
1308 | + | |
1309 | + /* If the remote end is being coy about its node ID then look it up by | |
1310 | + * address */ | |
1311 | + if (!rem_node && header->srcid == 0) { | |
1312 | + rem_node = find_node_by_addr(addr, addrlen); | |
1313 | + } | |
1314 | + | |
1315 | + /* If this node is an ex-member then treat it as unknown */ | |
1316 | + if (rem_node && rem_node->state != NODESTATE_MEMBER | |
1317 | + && rem_node->state != NODESTATE_JOINING) | |
1318 | + rem_node = NULL; | |
1319 | + | |
1320 | + /* Ignore messages not for our cluster */ | |
1321 | + if (le16_to_cpu(header->cluster) != cluster_id) { | |
1322 | + P_COMMS("Dumping message - wrong cluster ID (us=%d, msg=%d)\n", | |
1323 | + cluster_id, header->cluster); | |
1324 | + goto userport_finish; | |
1325 | + } | |
1326 | + | |
1327 | + /* If the message is from us then just dump it */ | |
1328 | + if (rem_node && rem_node->us) | |
1329 | + goto userport_finish; | |
1330 | + | |
1331 | + /* If we can't find the nodeid then check for our own messages the hard | |
1332 | + * way - this only happens during joining */ | |
1333 | + if (!rem_node) { | |
1334 | + struct list_head *socklist; | |
1335 | + struct cl_comms_socket *clsock; | |
1336 | + | |
1337 | + list_for_each(socklist, &socket_list) { | |
1338 | + clsock = | |
1339 | + list_entry(socklist, struct cl_comms_socket, list); | |
1340 | + | |
1341 | + if (clsock->recv_only) { | |
1342 | + | |
1343 | + if (memcmp(addr, &clsock->saddr, address_length) == 0) { | |
1344 | + goto userport_finish; | |
1345 | + } | |
1346 | + } | |
1347 | + } | |
1348 | + | |
1349 | + } | |
1350 | + | |
1351 | + /* Ignore messages not for us */ | |
1352 | + if (le32_to_cpu(header->tgtid) > 0 && us | |
1353 | + && le32_to_cpu(header->tgtid) != us->node_id) { | |
1354 | + goto userport_finish; | |
1355 | + } | |
1356 | + | |
1357 | + P_COMMS("got message, from %d for %d, sequence num = %d\n", | |
1358 | + le32_to_cpu(header->srcid), le32_to_cpu(header->tgtid), | |
1359 | + le16_to_cpu(header->seq)); | |
1360 | + | |
1361 | + /* Have we received this message before ? If so just ignore it, it's a | |
1362 | + * resend for someone else's benefit */ | |
1363 | + if (!(header->flags & (MSG_NOACK >> 16)) && | |
1364 | + rem_node && le16_to_cpu(header->seq) == rem_node->last_seq_recv) { | |
1365 | + P_COMMS | |
1366 | + ("Discarding message - Already seen this sequence number %d\n", | |
1367 | + rem_node->last_seq_recv); | |
1368 | + /* Still need to ACK it though, in case it was the ACK that got | |
1369 | + * lost */ | |
1370 | + cl_sendack(csock, header->seq, addrlen, addr, header->port, 0); | |
1371 | + goto userport_finish; | |
1372 | + } | |
1373 | + | |
1374 | + /* If it's a new node then assign it a temporary node ID */ | |
1375 | + if (!rem_node) | |
1376 | + header->srcid = cpu_to_le32(new_temp_nodeid(addr, addrlen)); | |
1377 | + | |
1378 | + P_COMMS("Got message: flags = %x, port = %d, we_are_a_member = %d\n", | |
1379 | + header->flags, header->port, we_are_a_cluster_member); | |
1380 | + | |
1381 | + | |
1382 | + /* If we are not part of the cluster then ignore multicast messages | |
1383 | + * that need an ACK as we will confuse the sender who is only expecting | |
1384 | + * ACKS from bona fide members */ | |
1385 | + if (header->flags & (MSG_MULTICAST >> 16) && | |
1386 | + !(header->flags & (MSG_NOACK >> 16)) && !we_are_a_cluster_member) { | |
1387 | + P_COMMS | |
1388 | + ("Discarding message - multicast and we are not a cluster member. port=%d flags=%x\n", | |
1389 | + header->port, header->flags); | |
1390 | + goto userport_finish; | |
1391 | + } | |
1392 | + | |
1393 | + /* Save the sequence number of this message so we can ignore duplicates | |
1394 | + * (above) */ | |
1395 | + if (!(header->flags & (MSG_NOACK >> 16)) && rem_node) { | |
1396 | + P_COMMS("Saving seq %d for node %s\n", le16_to_cpu(header->seq), | |
1397 | + rem_node->name); | |
1398 | + rem_node->last_seq_recv = le16_to_cpu(header->seq); | |
1399 | + } | |
1400 | + | |
1401 | + /* Is it a protocol message? */ | |
1402 | + if (header->port == 0) { | |
1403 | + process_cnxman_message(csock, data, len, addr, addrlen, | |
1404 | + rem_node); | |
1405 | + goto userport_finish; | |
1406 | + } | |
1407 | + | |
1408 | + /* Skip past the header to the data */ | |
1409 | + data += sizeof (struct cl_protheader); | |
1410 | + len -= sizeof (struct cl_protheader); | |
1411 | + | |
1412 | + /* Get the port number and look for a listener */ | |
1413 | + down(&port_array_lock); | |
1414 | + if (port_array[header->port]) { | |
1415 | + int native_srcid; | |
1416 | + struct cluster_sock *c = cluster_sk(port_array[header->port]); | |
1417 | + | |
1418 | + /* ACK it */ | |
1419 | + if (!(header->flags & (MSG_NOACK >> 16))) | |
1420 | + cl_sendack(csock, header->seq, addrlen, addr, | |
1421 | + header->port, 0); | |
1422 | + | |
1423 | + /* Call a callback if there is one */ | |
1424 | + if (c->kernel_callback) { | |
1425 | + up(&port_array_lock); | |
1426 | + c->kernel_callback(data, len, addr, addrlen, | |
1427 | + le32_to_cpu(header->srcid)); | |
1428 | + goto userport_finish; | |
1429 | + } | |
1430 | + | |
1431 | + /* Otherwise put it into an SKB and pass it onto the recvmsg | |
1432 | + * mechanism */ | |
1433 | + skb = alloc_skb(len, GFP_KERNEL); | |
1434 | + if (!skb) { | |
1435 | + up(&port_array_lock); | |
1436 | + printk(KERN_INFO CMAN_NAME | |
1437 | + ": Failed to allocate skb\n"); | |
1438 | + return; | |
1439 | + } | |
1440 | + | |
1441 | + skb_put(skb, len); | |
1442 | + memcpy(skb->data, data, len); | |
1443 | + | |
1444 | + /* Put the nodeid into cb so we can pass it to the clients */ | |
1445 | + skb->cb[0] = 0; /* Clear flags */ | |
1446 | + native_srcid = le32_to_cpu(header->srcid); | |
1447 | + memcpy(skb->cb + 1, &native_srcid, sizeof(int)); | |
1448 | + | |
1449 | + if ((err = | |
1450 | + sock_queue_rcv_skb(port_array[header->port], skb)) < 0) { | |
1451 | + | |
1452 | + printk(KERN_INFO CMAN_NAME | |
1453 | + ": Error queueing request to port %d: %d\n", | |
1454 | + header->port, err); | |
1455 | + kfree_skb(skb); | |
1456 | + | |
1457 | + /* If the port was MEMBERSHIP then we have to die */ | |
1458 | + if (header->port == CLUSTER_PORT_MEMBERSHIP) { | |
1459 | + up(&port_array_lock); | |
1460 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
1461 | + panic("membership stopped responding"); | |
1462 | + } | |
1463 | + } | |
1464 | + up(&port_array_lock); | |
1465 | + | |
1466 | + } | |
1467 | + else { | |
1468 | + /* ACK it, but set the flag bit so remote end knows no-one | |
1469 | + * caught it */ | |
1470 | + if (!(header->flags & (MSG_NOACK >> 16))) | |
1471 | + cl_sendack(csock, header->seq, addrlen, addr, | |
1472 | + header->port, 1); | |
1473 | + | |
1474 | + /* Nobody listening, drop it */ | |
1475 | + up(&port_array_lock); | |
1476 | + } | |
1477 | + | |
1478 | + userport_finish: | |
1479 | + return; | |
1480 | +} | |
1481 | + | |
1482 | +static struct sock *cl_alloc_sock(struct socket *sock, int gfp) | |
1483 | +{ | |
1484 | + struct sock *sk; | |
1485 | + struct cluster_sock *c; | |
1486 | + | |
1487 | + if ((sk = | |
1488 | + sk_alloc(AF_CLUSTER, gfp, sizeof (struct cluster_sock), | |
1489 | + cluster_sk_cachep)) == NULL) | |
1490 | + goto no_sock; | |
1491 | + | |
1492 | + if (sock) { | |
1493 | + sock->ops = &cl_proto_ops; | |
1494 | + } | |
1495 | + sock_init_data(sock, sk); | |
1496 | + | |
1497 | + sk->sk_destruct = NULL; | |
1498 | + sk->sk_no_check = 1; | |
1499 | + sk->sk_family = PF_CLUSTER; | |
1500 | + sk->sk_allocation = gfp; | |
1501 | + | |
1502 | + c = cluster_sk(sk); | |
1503 | + c->port = 0; | |
1504 | + c->service_data = NULL; | |
1505 | + | |
1506 | + return sk; | |
1507 | + no_sock: | |
1508 | + return NULL; | |
1509 | +} | |
1510 | + | |
1511 | +static int cl_release(struct socket *sock) | |
1512 | +{ | |
1513 | + struct sock *sk = sock->sk; | |
1514 | + struct cl_client_socket *csock; | |
1515 | + struct list_head *socklist; | |
1516 | + struct list_head *tmp; | |
1517 | + | |
1518 | + down(&client_socket_lock); | |
1519 | + if (sk) { | |
1520 | + /* Remove port allocations if it's a bound socket */ | |
1521 | + struct cluster_sock *c = cluster_sk(sk); | |
1522 | + | |
1523 | + down(&port_array_lock); | |
1524 | + if (c->port) { | |
1525 | + port_array[c->port] = NULL; | |
1526 | + } | |
1527 | + up(&port_array_lock); | |
1528 | + | |
1529 | + /* Tell other nodes in the cluster that this listener is going | |
1530 | + * away */ | |
1531 | + if (atomic_read(&cnxman_running) && c->port) | |
1532 | + send_port_close_oob(c->port); | |
1533 | + | |
1534 | + if (c->service_data) | |
1535 | + sm_sock_release(sock); | |
1536 | + | |
1537 | + /* Master socket released ? */ | |
1538 | + if (sk->sk_protocol == CLPROTO_MASTER) { | |
1539 | + master_sock = NULL; | |
1540 | + | |
1541 | + /* If this socket is being freed and cnxman is not | |
1542 | + * started then free all the comms sockets as either | |
1543 | + * the userland "join" process has crashed or the | |
1544 | + * join failed. | |
1545 | + */ | |
1546 | + if (!atomic_read(&cnxman_running)) { | |
1547 | + quit_threads = 1; | |
1548 | + free_cluster_sockets(); | |
1549 | + } | |
1550 | + } | |
1551 | + | |
1552 | + sock_orphan(sk); | |
1553 | + sock_hold(sk); | |
1554 | + lock_sock(sk); | |
1555 | + release_sock(sk); | |
1556 | + sock_put(sk); | |
1557 | + sock_put(sk); | |
1558 | + sock->sk = NULL; | |
1559 | + } | |
1560 | + | |
1561 | + /* Remove it from the list of clients */ | |
1562 | + list_for_each_safe(socklist, tmp, &client_socket_list) { | |
1563 | + csock = list_entry(socklist, struct cl_client_socket, list); | |
1564 | + | |
1565 | + if (csock->sock == sock) { | |
1566 | + list_del(&csock->list); | |
1567 | + kfree(csock); | |
1568 | + break; | |
1569 | + } | |
1570 | + } | |
1571 | + up(&client_socket_lock); | |
1572 | + | |
1573 | + return 0; | |
1574 | +} | |
1575 | + | |
1576 | +static int cl_create(struct socket *sock, int protocol) | |
1577 | +{ | |
1578 | + struct sock *sk; | |
1579 | + | |
1580 | + /* All are datagrams */ | |
1581 | + if (sock->type != SOCK_DGRAM) | |
1582 | + return -ESOCKTNOSUPPORT; | |
1583 | + | |
1584 | + if (protocol == CLPROTO_MASTER && !capable(CAP_CLUSTER)) | |
1585 | + return -EPERM; | |
1586 | + | |
1587 | + /* Can only have one master socket */ | |
1588 | + if (master_sock && protocol == CLPROTO_MASTER) | |
1589 | + return -EBUSY; | |
1590 | + | |
1591 | + /* cnxman not running and a client was requested */ | |
1592 | + if (!atomic_read(&cnxman_running) && protocol != CLPROTO_MASTER) | |
1593 | + return -ENETDOWN; | |
1594 | + | |
1595 | + if ((sk = cl_alloc_sock(sock, GFP_KERNEL)) == NULL) | |
1596 | + return -ENOBUFS; | |
1597 | + | |
1598 | + sk->sk_protocol = protocol; | |
1599 | + | |
1600 | + if (protocol == CLPROTO_MASTER) | |
1601 | + master_sock = sk; | |
1602 | + | |
1603 | + /* Add client sockets to the list */ | |
1604 | + if (protocol == CLPROTO_CLIENT) { | |
1605 | + struct cl_client_socket *clsock = | |
1606 | + kmalloc(sizeof (struct cl_client_socket), GFP_KERNEL); | |
1607 | + if (!clsock) { | |
1608 | + cl_release(sock); | |
1609 | + return -ENOMEM; | |
1610 | + } | |
1611 | + clsock->sock = sock; | |
1612 | + down(&client_socket_lock); | |
1613 | + list_add(&clsock->list, &client_socket_list); | |
1614 | + up(&client_socket_lock); | |
1615 | + } | |
1616 | + | |
1617 | + return 0; | |
1618 | +} | |
1619 | + | |
1620 | +static int cl_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) | |
1621 | +{ | |
1622 | + struct sock *sk = sock->sk; | |
1623 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *) uaddr; | |
1624 | + struct cluster_sock *c = cluster_sk(sk); | |
1625 | + | |
1626 | + if (!capable(CAP_NET_BIND_SERVICE)) | |
1627 | + return -EPERM; | |
1628 | + | |
1629 | + if (sk->sk_zapped == 0) | |
1630 | + return -EINVAL; | |
1631 | + | |
1632 | + if (addr_len != sizeof (struct sockaddr_cl)) | |
1633 | + return -EINVAL; | |
1634 | + | |
1635 | + if (saddr->scl_family != AF_CLUSTER) | |
1636 | + return -EINVAL; | |
1637 | + | |
1638 | + if (saddr->scl_port == 0) | |
1639 | + return -EINVAL; /* Port 0 is reserved for protocol messages */ | |
1640 | + | |
1641 | + down(&port_array_lock); | |
1642 | + | |
1643 | + if (port_array[saddr->scl_port]) { | |
1644 | + up(&port_array_lock); | |
1645 | + return -EADDRINUSE; | |
1646 | + } | |
1647 | + | |
1648 | + port_array[saddr->scl_port] = sk; | |
1649 | + | |
1650 | + up(&port_array_lock); | |
1651 | + | |
1652 | + c->port = saddr->scl_port; | |
1653 | + sk->sk_zapped = 0; | |
1654 | + | |
1655 | + /* If we are not a cluster member yet then make the client wait until | |
1656 | + * we are, this allows nodes to start cluster clients at the same time | |
1657 | + * as cluster services but they will wait until membership is achieved. | |
1658 | + * This looks odd in bind() (open would seem more obvious) but we need | |
1659 | + * to know which port number is being used so that things like | |
1660 | + * membership services don't get blocked | |
1661 | + */ | |
1662 | + | |
1663 | + if (saddr->scl_port > HIGH_PROTECTED_PORT) | |
1664 | + while (!we_are_a_cluster_member || !cluster_is_quorate | |
1665 | + || in_transition()) { | |
1666 | + DECLARE_WAITQUEUE(wq, current); | |
1667 | + struct task_struct *tsk = current; | |
1668 | + | |
1669 | + set_task_state(tsk, TASK_INTERRUPTIBLE); | |
1670 | + add_wait_queue(&socket_waitq, &wq); | |
1671 | + | |
1672 | + if (!we_are_a_cluster_member || !cluster_is_quorate | |
1673 | + || in_transition()) | |
1674 | + schedule(); | |
1675 | + | |
1676 | + set_task_state(tsk, TASK_RUNNING); | |
1677 | + remove_wait_queue(&socket_waitq, &wq); | |
1678 | + | |
1679 | + /* We were woken up because the cluster is going down, | |
1680 | + * ...and we never got a chance to do any work! (sob) */ | |
1681 | + if (atomic_read(&cnxman_running) == 0 || quit_threads) { | |
1682 | + return -ENOTCONN; | |
1683 | + } | |
1684 | + } | |
1685 | + | |
1686 | + return 0; | |
1687 | +} | |
1688 | + | |
1689 | +static int cl_getname(struct socket *sock, struct sockaddr *uaddr, | |
1690 | + int *uaddr_len, int peer) | |
1691 | +{ | |
1692 | + struct sockaddr_cl *sa = (struct sockaddr_cl *) uaddr; | |
1693 | + struct sock *sk = sock->sk; | |
1694 | + struct cluster_sock *c = cluster_sk(sk); | |
1695 | + | |
1696 | + *uaddr_len = sizeof (struct sockaddr_cl); | |
1697 | + | |
1698 | + lock_sock(sk); | |
1699 | + | |
1700 | + sa->scl_port = c->port; | |
1701 | + sa->scl_flags = 0; | |
1702 | + sa->scl_family = AF_CLUSTER; | |
1703 | + | |
1704 | + release_sock(sk); | |
1705 | + | |
1706 | + return 0; | |
1707 | +} | |
1708 | + | |
1709 | +static unsigned int cl_poll(struct file *file, struct socket *sock, | |
1710 | + poll_table * wait) | |
1711 | +{ | |
1712 | + return datagram_poll(file, sock, wait); | |
1713 | +} | |
1714 | + | |
1715 | +/* Copy internal node format to userland format */ | |
1716 | +void copy_to_usernode(struct cluster_node *node, | |
1717 | + struct cl_cluster_node *unode) | |
1718 | +{ | |
1719 | + strcpy(unode->name, node->name); | |
1720 | + unode->size = sizeof (struct cl_cluster_node); | |
1721 | + unode->votes = node->votes; | |
1722 | + unode->state = node->state; | |
1723 | + unode->us = node->us; | |
1724 | + unode->node_id = node->node_id; | |
1725 | + unode->leave_reason = node->leave_reason; | |
1726 | + unode->incarnation = node->incarnation; | |
1727 | +} | |
1728 | + | |
1729 | +/* ioctl processing functions */ | |
1730 | + | |
1731 | +static int do_ioctl_set_version(unsigned long arg) | |
1732 | +{ | |
1733 | + struct cl_version version, *u_version; | |
1734 | + | |
1735 | + if (!capable(CAP_CLUSTER)) | |
1736 | + return -EPERM; | |
1737 | + if (arg == 0) | |
1738 | + return -EINVAL; | |
1739 | + | |
1740 | + u_version = (struct cl_version *) arg; | |
1741 | + | |
1742 | + if (copy_from_user(&version, u_version, sizeof(struct cl_version))) | |
1743 | + return -EFAULT; | |
1744 | + | |
1745 | + if (version.major != CNXMAN_MAJOR_VERSION || | |
1746 | + version.minor != CNXMAN_MINOR_VERSION || | |
1747 | + version.patch != CNXMAN_PATCH_VERSION) | |
1748 | + return -EINVAL; | |
1749 | + | |
1750 | + if (config_version == version.config) | |
1751 | + return 0; | |
1752 | + | |
1753 | + config_version = version.config; | |
1754 | + send_reconfigure(RECONFIG_PARAM_CONFIG_VERSION, config_version); | |
1755 | + return 0; | |
1756 | +} | |
1757 | + | |
1758 | +static int do_ioctl_get_members(unsigned long arg) | |
1759 | +{ | |
1760 | + struct cluster_node *node; | |
1761 | + /* Kernel copies */ | |
1762 | + struct cl_cluster_node user_format_node; | |
1763 | + struct cl_cluster_nodelist user_format_nodelist; | |
1764 | + /* User space array ptr */ | |
1765 | + struct cl_cluster_node *user_node; | |
1766 | + struct list_head *nodelist; | |
1767 | + int num_nodes = 0; | |
1768 | + | |
1769 | + if (arg == 0) | |
1770 | + return cluster_members; | |
1771 | + | |
1772 | + if (copy_from_user(&user_format_nodelist, (void __user *)arg, sizeof(struct cl_cluster_nodelist))) | |
1773 | + return -EFAULT; | |
1774 | + | |
1775 | + down(&cluster_members_lock); | |
1776 | + | |
1777 | + if (user_format_nodelist.max_members < cluster_members) { | |
1778 | + up(&cluster_members_lock); | |
1779 | + return -E2BIG; | |
1780 | + } | |
1781 | + | |
1782 | + user_node = user_format_nodelist.nodes; | |
1783 | + | |
1784 | + list_for_each(nodelist, &cluster_members_list) { | |
1785 | + node = list_entry(nodelist, struct cluster_node, list); | |
1786 | + if (node->state == NODESTATE_MEMBER) { | |
1787 | + copy_to_usernode(node, &user_format_node); | |
1788 | + if (copy_to_user(user_node, &user_format_node, | |
1789 | + sizeof (struct cl_cluster_node))) { | |
1790 | + up(&cluster_members_lock); | |
1791 | + return -EFAULT; | |
1792 | + } | |
1793 | + user_node++; | |
1794 | + num_nodes++; | |
1795 | + } | |
1796 | + } | |
1797 | + up(&cluster_members_lock); | |
1798 | + | |
1799 | + return num_nodes; | |
1800 | +} | |
1801 | + | |
1802 | +static int do_ioctl_get_all_members(unsigned long arg) | |
1803 | +{ | |
1804 | + struct cluster_node *node; | |
1805 | + /* Kernel copies */ | |
1806 | + struct cl_cluster_node user_format_node; | |
1807 | + struct cl_cluster_nodelist user_format_nodelist; | |
1808 | + /* User space array ptr*/ | |
1809 | + struct cl_cluster_node *user_node; | |
1810 | + struct list_head *nodelist; | |
1811 | + int num_nodes = 0; | |
1812 | + | |
1813 | + if (copy_from_user(&user_format_nodelist, (void __user *)arg, sizeof(struct cl_cluster_nodelist))) | |
1814 | + return -EFAULT; | |
1815 | + | |
1816 | + down(&cluster_members_lock); | |
1817 | + | |
1818 | + user_node = user_format_nodelist.nodes; | |
1819 | + | |
1820 | + list_for_each(nodelist, &cluster_members_list) { | |
1821 | + node = list_entry(nodelist, struct cluster_node, list); | |
1822 | + if (arg) { | |
1823 | + copy_to_usernode(node, | |
1824 | + &user_format_node); | |
1825 | + | |
1826 | + if (copy_to_user(user_node, &user_format_node, | |
1827 | + sizeof (struct cl_cluster_node))) { | |
1828 | + up(&cluster_members_lock); | |
1829 | + return -EFAULT; | |
1830 | + } | |
1831 | + user_node++; | |
1832 | + if (--user_format_nodelist.max_members < 0) { | |
1833 | + num_nodes = -EFAULT; | |
1834 | + goto err_exit; | |
1835 | + } | |
1836 | + | |
1837 | + } | |
1838 | + num_nodes++; | |
1839 | + } | |
1840 | + err_exit: | |
1841 | + up(&cluster_members_lock); | |
1842 | + | |
1843 | + return num_nodes; | |
1844 | +} | |
1845 | + | |
1846 | +static int do_ioctl_get_node(unsigned long arg) | |
1847 | +{ | |
1848 | + struct cluster_node *node; | |
1849 | + struct cl_cluster_node k_node, *u_node; | |
1850 | + | |
1851 | + u_node = (struct cl_cluster_node *) arg; | |
1852 | + | |
1853 | + if (copy_from_user(&k_node, u_node, sizeof(struct cl_cluster_node))) | |
1854 | + return -EFAULT; | |
1855 | + | |
1856 | + if (k_node.node_id) | |
1857 | + node = find_node_by_nodeid(k_node.node_id); | |
1858 | + else | |
1859 | + node = find_node_by_name(k_node.name); | |
1860 | + | |
1861 | + if (!node) | |
1862 | + return -ENOENT; | |
1863 | + | |
1864 | + copy_to_usernode(node, &k_node); | |
1865 | + | |
1866 | + if (copy_to_user(u_node, &k_node, sizeof(struct cl_cluster_node))) | |
1867 | + return -EFAULT; | |
1868 | + | |
1869 | + return 0; | |
1870 | +} | |
1871 | + | |
1872 | +static int do_ioctl_set_expected(unsigned long arg) | |
1873 | +{ | |
1874 | + struct list_head *nodelist; | |
1875 | + struct cluster_node *node; | |
1876 | + unsigned int total_votes; | |
1877 | + unsigned int newquorum; | |
1878 | + | |
1879 | + if (!capable(CAP_CLUSTER)) | |
1880 | + return -EPERM; | |
1881 | + if (arg == 0) | |
1882 | + return -EINVAL; | |
1883 | + | |
1884 | + newquorum = calculate_quorum(1, arg, &total_votes); | |
1885 | + | |
1886 | + if (newquorum < total_votes / 2 | |
1887 | + || newquorum > total_votes) { | |
1888 | + return -EINVAL; | |
1889 | + } | |
1890 | + | |
1891 | + /* Now do it */ | |
1892 | + down(&cluster_members_lock); | |
1893 | + list_for_each(nodelist, &cluster_members_list) { | |
1894 | + node = list_entry(nodelist, struct cluster_node, list); | |
1895 | + if (node->state == NODESTATE_MEMBER | |
1896 | + && node->expected_votes > arg) { | |
1897 | + node->expected_votes = arg; | |
1898 | + } | |
1899 | + } | |
1900 | + up(&cluster_members_lock); | |
1901 | + | |
1902 | + recalculate_quorum(1); | |
1903 | + | |
1904 | + send_reconfigure(RECONFIG_PARAM_EXPECTED_VOTES, arg); | |
1905 | + sm_member_update(cluster_is_quorate); | |
1906 | + | |
1907 | + return 0; | |
1908 | +} | |
1909 | + | |
1910 | +static int do_ioctl_kill_node(unsigned long arg) | |
1911 | +{ | |
1912 | + struct cluster_node *node; | |
1913 | + | |
1914 | + if (!capable(CAP_CLUSTER)) | |
1915 | + return -EPERM; | |
1916 | + | |
1917 | + | |
1918 | + if ((node = find_node_by_nodeid(arg)) == NULL) | |
1919 | + return -EINVAL; | |
1920 | + | |
1921 | + /* Can't kill us */ | |
1922 | + if (node->us) | |
1923 | + return -EINVAL; | |
1924 | + | |
1925 | + if (node->state != NODESTATE_MEMBER) | |
1926 | + return -EINVAL; | |
1927 | + | |
1928 | + /* Just in case it is alive, send a KILL message */ | |
1929 | + send_kill(arg); | |
1930 | + | |
1931 | + node->leave_reason = CLUSTER_LEAVEFLAG_KILLED; | |
1932 | + a_node_just_died(node); | |
1933 | + | |
1934 | + return 0; | |
1935 | +} | |
1936 | + | |
1937 | +static int do_ioctl_barrier(unsigned long arg) | |
1938 | +{ | |
1939 | + struct cl_barrier_info info; | |
1940 | + | |
1941 | + if (!capable(CAP_CLUSTER)) | |
1942 | + return -EPERM; | |
1943 | + | |
1944 | + if (copy_from_user(&info, (void *)arg, sizeof(info)) != 0) | |
1945 | + return -EFAULT; | |
1946 | + | |
1947 | + switch (info.cmd) { | |
1948 | + case BARRIER_IOCTL_REGISTER: | |
1949 | + return kcl_barrier_register(info.name, | |
1950 | + info.flags, | |
1951 | + info.arg); | |
1952 | + case BARRIER_IOCTL_CHANGE: | |
1953 | + return kcl_barrier_setattr(info.name, | |
1954 | + info.flags, | |
1955 | + info.arg); | |
1956 | + case BARRIER_IOCTL_WAIT: | |
1957 | + return kcl_barrier_wait(info.name); | |
1958 | + case BARRIER_IOCTL_DELETE: | |
1959 | + return kcl_barrier_delete(info.name); | |
1960 | + default: | |
1961 | + return -EINVAL; | |
1962 | + } | |
1963 | +} | |
1964 | + | |
1965 | +static int do_ioctl_islistening(unsigned long arg) | |
1966 | +{ | |
1967 | + DECLARE_WAITQUEUE(wq, current); | |
1968 | + struct cl_listen_request rq; | |
1969 | + struct cluster_node *rem_node; | |
1970 | + int nodeid; | |
1971 | + int result; | |
1972 | + struct cl_waiting_listen_request *listen_request; | |
1973 | + | |
1974 | + if (!arg) | |
1975 | + return -EINVAL; | |
1976 | + | |
1977 | + if (copy_from_user(&rq, (void *) arg, sizeof (rq)) != 0) | |
1978 | + return -EFAULT; | |
1979 | + | |
1980 | + nodeid = rq.nodeid; | |
1981 | + | |
1982 | + rem_node = find_node_by_nodeid(nodeid); | |
1983 | + | |
1984 | + /* Node not in the cluster */ | |
1985 | + if (!rem_node) | |
1986 | + return -ENOENT; | |
1987 | + | |
1988 | + if (rem_node->state != NODESTATE_MEMBER) | |
1989 | + return -ENOTCONN; | |
1990 | + | |
1991 | + /* If the request is for us then just look in the ports | |
1992 | + * array */ | |
1993 | + if (nodeid == us->node_id) | |
1994 | + return (port_array[rq.port] != 0) ? 1 : 0; | |
1995 | + | |
1996 | + /* For a remote node we need to send a request out */ | |
1997 | + | |
1998 | + /* If we are in transition then wait until we are not */ | |
1999 | + while (in_transition()) { | |
2000 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
2001 | + add_wait_queue(&socket_waitq, &wq); | |
2002 | + | |
2003 | + if (in_transition()) | |
2004 | + schedule(); | |
2005 | + | |
2006 | + set_task_state(current, TASK_RUNNING); | |
2007 | + remove_wait_queue(&socket_waitq, &wq); | |
2008 | + | |
2009 | + if (signal_pending(current)) | |
2010 | + return -EINTR; | |
2011 | + } | |
2012 | + | |
2013 | + /* Were we shut down before it completed ? */ | |
2014 | + if (!atomic_read(&cnxman_running)) | |
2015 | + return -ENOTCONN; | |
2016 | + | |
2017 | + listen_request = | |
2018 | + kmalloc(sizeof (struct cl_waiting_listen_request), | |
2019 | + GFP_KERNEL); | |
2020 | + if (!listen_request) | |
2021 | + return -ENOMEM; | |
2022 | + | |
2023 | + /* Build the request */ | |
2024 | + listen_request->waiting = 1; | |
2025 | + listen_request->result = 0; | |
2026 | + listen_request->tag = current->pid; | |
2027 | + listen_request->nodeid = nodeid; | |
2028 | + init_waitqueue_head(&listen_request->waitq); | |
2029 | + | |
2030 | + down(&listenreq_lock); | |
2031 | + list_add(&listen_request->list, &listenreq_list); | |
2032 | + up(&listenreq_lock); | |
2033 | + | |
2034 | + /* Now wait for the response to come back */ | |
2035 | + send_listen_request(rq.nodeid, rq.port); | |
2036 | + | |
2037 | + while (listen_request->waiting) { | |
2038 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
2039 | + add_wait_queue(&listen_request->waitq, &wq); | |
2040 | + | |
2041 | + if (listen_request->waiting) | |
2042 | + schedule(); | |
2043 | + | |
2044 | + set_task_state(current, TASK_RUNNING); | |
2045 | + remove_wait_queue(&listen_request->waitq, &wq); | |
2046 | + | |
2047 | + if (signal_pending(current)) { | |
2048 | + list_del(&listen_request->list); | |
2049 | + kfree(listen_request); | |
2050 | + return -ERESTARTSYS; | |
2051 | + } | |
2052 | + } | |
2053 | + result = listen_request->result; | |
2054 | + list_del(&listen_request->list); | |
2055 | + kfree(listen_request); | |
2056 | + return result; | |
2057 | +} | |
2058 | + | |
2059 | +static int do_ioctl_set_votes(unsigned long arg) | |
2060 | +{ | |
2061 | + unsigned int total_votes; | |
2062 | + unsigned int newquorum; | |
2063 | + int saved_votes; | |
2064 | + | |
2065 | + if (!capable(CAP_CLUSTER)) | |
2066 | + return -EPERM; | |
2067 | + | |
2068 | + /* Check votes is valid */ | |
2069 | + saved_votes = us->votes; | |
2070 | + us->votes = arg; | |
2071 | + | |
2072 | + newquorum = calculate_quorum(1, 0, &total_votes); | |
2073 | + | |
2074 | + if (newquorum < total_votes / 2 || newquorum > total_votes) { | |
2075 | + us->votes = saved_votes; | |
2076 | + return -EINVAL; | |
2077 | + } | |
2078 | + | |
2079 | + recalculate_quorum(1); | |
2080 | + | |
2081 | + send_reconfigure(RECONFIG_PARAM_NODE_VOTES, arg); | |
2082 | + | |
2083 | + return 0; | |
2084 | +} | |
2085 | + | |
2086 | +static int cl_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) | |
2087 | +{ | |
2088 | + int err = -EOPNOTSUPP; | |
2089 | + struct list_head *proclist; | |
2090 | + struct list_head *tmp; | |
2091 | + struct notify_struct *notify; | |
2092 | + struct cl_version cnxman_version; | |
2093 | + | |
2094 | + switch (cmd) { | |
2095 | + /* Process requests notification of cluster events */ | |
2096 | + case SIOCCLUSTER_NOTIFY: | |
2097 | + notify = kmalloc(sizeof (struct notify_struct), GFP_KERNEL); | |
2098 | + if (!notify) | |
2099 | + return -ENOMEM; | |
2100 | + notify->pid = current->pid; | |
2101 | + notify->signal = arg; | |
2102 | + down(&event_listener_lock); | |
2103 | + list_add(¬ify->list, &event_listener_list); | |
2104 | + up(&event_listener_lock); | |
2105 | + err = 0; | |
2106 | + break; | |
2107 | + | |
2108 | + /* Process is no longer interested cluster events */ | |
2109 | + case SIOCCLUSTER_REMOVENOTIFY: | |
2110 | + err = EINVAL; | |
2111 | + | |
2112 | + down(&event_listener_lock); | |
2113 | + list_for_each_safe(proclist, tmp, &event_listener_list) { | |
2114 | + notify = | |
2115 | + list_entry(proclist, struct notify_struct, list); | |
2116 | + if (notify->pid == current->pid) { | |
2117 | + list_del(¬ify->list); | |
2118 | + kfree(notify); | |
2119 | + err = 0; | |
2120 | + } | |
2121 | + } | |
2122 | + up(&event_listener_lock); | |
2123 | + break; | |
2124 | + | |
2125 | + /* Return the cnxman version number */ | |
2126 | + case SIOCCLUSTER_GET_VERSION: | |
2127 | + if (!arg) | |
2128 | + return -EINVAL; | |
2129 | + err = 0; | |
2130 | + cnxman_version.major = CNXMAN_MAJOR_VERSION; | |
2131 | + cnxman_version.minor = CNXMAN_MINOR_VERSION; | |
2132 | + cnxman_version.patch = CNXMAN_PATCH_VERSION; | |
2133 | + if (copy_to_user((void *) arg, &cnxman_version, | |
2134 | + sizeof (struct cl_version))) { | |
2135 | + return -EFAULT; | |
2136 | + } | |
2137 | + break; | |
2138 | + | |
2139 | + /* Set the cnxman config version number */ | |
2140 | + case SIOCCLUSTER_SET_VERSION: | |
2141 | + err = do_ioctl_set_version(arg); | |
2142 | + break; | |
2143 | + | |
2144 | + /* Return the active membership list */ | |
2145 | + case SIOCCLUSTER_GETMEMBERS: | |
2146 | + err = do_ioctl_get_members(arg); | |
2147 | + break; | |
2148 | + | |
2149 | + /* Return the full membership list include dead nodes */ | |
2150 | + case SIOCCLUSTER_GETALLMEMBERS: | |
2151 | + err = do_ioctl_get_all_members(arg); | |
2152 | + break; | |
2153 | + | |
2154 | + case SIOCCLUSTER_GETNODE: | |
2155 | + err = do_ioctl_get_node(arg); | |
2156 | + break; | |
2157 | + | |
2158 | + case SIOCCLUSTER_ISQUORATE: | |
2159 | + return cluster_is_quorate; | |
2160 | + | |
2161 | + case SIOCCLUSTER_ISACTIVE: | |
2162 | + return atomic_read(&cnxman_running); | |
2163 | + | |
2164 | + case SIOCCLUSTER_SETEXPECTED_VOTES: | |
2165 | + err = do_ioctl_set_expected(arg); | |
2166 | + break; | |
2167 | + | |
2168 | + /* Change the number of votes for this node */ | |
2169 | + case SIOCCLUSTER_SET_VOTES: | |
2170 | + err = do_ioctl_set_votes(arg); | |
2171 | + break; | |
2172 | + | |
2173 | + /* Return 1 if the specified node is listening on a given port */ | |
2174 | + case SIOCCLUSTER_ISLISTENING: | |
2175 | + err = do_ioctl_islistening(arg); | |
2176 | + break; | |
2177 | + | |
2178 | + /* Forcibly kill a node */ | |
2179 | + case SIOCCLUSTER_KILLNODE: | |
2180 | + err = do_ioctl_kill_node(arg); | |
2181 | + break; | |
2182 | + | |
2183 | + case SIOCCLUSTER_GET_JOINCOUNT: | |
2184 | + if (!capable(CAP_CLUSTER)) | |
2185 | + return -EPERM; | |
2186 | + else | |
2187 | + return atomic_read(&use_count); | |
2188 | + | |
2189 | + /* ioctl interface to the barrier system */ | |
2190 | + case SIOCCLUSTER_BARRIER: | |
2191 | + err = do_ioctl_barrier(arg); | |
2192 | + break; | |
2193 | + | |
2194 | + default: | |
2195 | + err = sm_ioctl(sock, cmd, arg); | |
2196 | + } | |
2197 | + return err; | |
2198 | +} | |
2199 | + | |
2200 | +static int cl_shutdown(struct socket *sock, int how) | |
2201 | +{ | |
2202 | + struct sock *sk = sock->sk; | |
2203 | + int err = -ENOTCONN; | |
2204 | + | |
2205 | + lock_sock(sk); | |
2206 | + | |
2207 | + if (sock->state == SS_UNCONNECTED) | |
2208 | + goto out; | |
2209 | + | |
2210 | + err = 0; | |
2211 | + if (sock->state == SS_DISCONNECTING) | |
2212 | + goto out; | |
2213 | + | |
2214 | + err = -EINVAL; | |
2215 | + | |
2216 | + if (how != SHUTDOWN_MASK) | |
2217 | + goto out; | |
2218 | + | |
2219 | + sk->sk_shutdown = how; | |
2220 | + err = 0; | |
2221 | + | |
2222 | + out: | |
2223 | + release_sock(sk); | |
2224 | + | |
2225 | + return err; | |
2226 | +} | |
2227 | + | |
2228 | +static int cl_setsockopt(struct socket *sock, int level, int optname, | |
2229 | + char *optval, int optlen) | |
2230 | +{ | |
2231 | + struct sock *sk = sock->sk; | |
2232 | + int err; | |
2233 | + | |
2234 | + if (sk != master_sock) | |
2235 | + return -EPERM; | |
2236 | + | |
2237 | + lock_sock(sk); | |
2238 | + err = __cl_setsockopt(sock, level, optname, optval, optlen, 0); | |
2239 | + release_sock(sk); | |
2240 | + | |
2241 | + return err; | |
2242 | +} | |
2243 | + | |
2244 | +static int add_clsock(int broadcast, int number, struct socket *sock, | |
2245 | + struct file *file) | |
2246 | +{ | |
2247 | + struct cl_comms_socket *newsock = | |
2248 | + kmalloc(sizeof (struct cl_comms_socket), GFP_KERNEL); | |
2249 | + if (!newsock) | |
2250 | + return -ENOMEM; | |
2251 | + | |
2252 | + memset(newsock, 0, sizeof (*newsock)); | |
2253 | + newsock->number = number; | |
2254 | + newsock->sock = sock; | |
2255 | + if (broadcast) { | |
2256 | + newsock->broadcast = 1; | |
2257 | + newsock->recv_only = 0; | |
2258 | + } | |
2259 | + else { | |
2260 | + newsock->broadcast = 0; | |
2261 | + newsock->recv_only = 1; | |
2262 | + } | |
2263 | + | |
2264 | + newsock->file = file; | |
2265 | + newsock->addr_len = sizeof(struct sockaddr_in6); | |
2266 | + | |
2267 | + /* Mark it active until cnxman thread is running and ready to process | |
2268 | + * messages */ | |
2269 | + set_bit(1, &newsock->active); | |
2270 | + | |
2271 | + /* Find out what it's bound to */ | |
2272 | + newsock->sock->ops->getname(newsock->sock, | |
2273 | + (struct sockaddr *)&newsock->saddr, | |
2274 | + &newsock->addr_len, 0); | |
2275 | + | |
2276 | + num_interfaces = max(num_interfaces, newsock->number); | |
2277 | + if (!current_interface && newsock->broadcast) | |
2278 | + current_interface = newsock; | |
2279 | + | |
2280 | + /* Hook data_ready */ | |
2281 | + newsock->sock->sk->sk_data_ready = cnxman_data_ready; | |
2282 | + | |
2283 | + /* Make an attempt to keep them in order */ | |
2284 | + list_add_tail(&newsock->list, &socket_list); | |
2285 | + | |
2286 | + address_length = newsock->addr_len; | |
2287 | + return 0; | |
2288 | +} | |
2289 | + | |
2290 | +static int __cl_setsockopt(struct socket *sock, int level, int optname, | |
2291 | + char *optval, int optlen, int flags) | |
2292 | +{ | |
2293 | + struct file *file; | |
2294 | + struct cl_join_cluster_info join_info; | |
2295 | + int error; | |
2296 | + int leave_flags; | |
2297 | + struct cl_multicast_sock multicast_info; | |
2298 | + | |
2299 | + if (optlen && !optval) | |
2300 | + return -EINVAL; | |
2301 | + | |
2302 | + switch (optname) { | |
2303 | + case CLU_SET_MULTICAST: | |
2304 | + case CLU_SET_RCVONLY: | |
2305 | + if (!capable(CAP_CLUSTER)) | |
2306 | + return -EPERM; | |
2307 | + | |
2308 | + if (optlen != sizeof (struct cl_multicast_sock)) | |
2309 | + return -EINVAL; | |
2310 | + | |
2311 | + if (atomic_read(&cnxman_running)) | |
2312 | + return -EINVAL; | |
2313 | + | |
2314 | + error = -EBADF; | |
2315 | + | |
2316 | + if (copy_from_user(&multicast_info, optval, optlen)) | |
2317 | + return -EFAULT; | |
2318 | + | |
2319 | + file = fget(multicast_info.fd); | |
2320 | + if (file) { | |
2321 | + struct inode *inode = file->f_dentry->d_inode; | |
2322 | + | |
2323 | + error = | |
2324 | + add_clsock(optname == CLU_SET_MULTICAST, | |
2325 | + multicast_info.number, SOCKET_I(inode), | |
2326 | + file); | |
2327 | + if (error) | |
2328 | + fput(file); | |
2329 | + } | |
2330 | + return error; | |
2331 | + | |
2332 | + case CLU_SET_NODENAME: | |
2333 | + if (!capable(CAP_CLUSTER)) | |
2334 | + return -EPERM; | |
2335 | + | |
2336 | + if (atomic_read(&cnxman_running)) | |
2337 | + return -EINVAL; | |
2338 | + | |
2339 | + if (optlen > MAX_CLUSTER_MEMBER_NAME_LEN) | |
2340 | + return -EINVAL; | |
2341 | + | |
2342 | + if (copy_from_user(nodename, optval, optlen)) | |
2343 | + return -EFAULT; | |
2344 | + break; | |
2345 | + | |
2346 | + case CLU_JOIN_CLUSTER: | |
2347 | + if (!capable(CAP_CLUSTER)) | |
2348 | + return -EPERM; | |
2349 | + | |
2350 | + if (atomic_read(&cnxman_running)) | |
2351 | + return -EALREADY; | |
2352 | + | |
2353 | + if (optlen != sizeof (struct cl_join_cluster_info)) | |
2354 | + return -EINVAL; | |
2355 | + | |
2356 | + if (copy_from_user(&join_info, optval, optlen)) | |
2357 | + return -EFAULT; | |
2358 | + | |
2359 | + if (strlen(join_info.cluster_name) > MAX_CLUSTER_NAME_LEN) | |
2360 | + return -EINVAL; | |
2361 | + | |
2362 | + if (list_empty(&socket_list)) | |
2363 | + return -ENOTCONN; | |
2364 | + | |
2365 | + set_votes(join_info.votes, join_info.expected_votes); | |
2366 | + cluster_id = generate_cluster_id(join_info.cluster_name); | |
2367 | + strncpy(cluster_name, join_info.cluster_name, MAX_CLUSTER_NAME_LEN); | |
2368 | + two_node = join_info.two_node; | |
2369 | + config_version = join_info.config_version; | |
2370 | + | |
2371 | + quit_threads = 0; | |
2372 | + acks_expected = 0; | |
2373 | + init_completion(&cluster_thread_comp); | |
2374 | + init_completion(&member_thread_comp); | |
2375 | + if (allocate_nodeid_array()) | |
2376 | + return -ENOMEM; | |
2377 | + | |
2378 | + kcluster_pid = kernel_thread(cluster_kthread, NULL, 0); | |
2379 | + if (kcluster_pid < 0) | |
2380 | + return kcluster_pid; | |
2381 | + | |
2382 | + wait_for_completion(&cluster_thread_comp); | |
2383 | + init_completion(&cluster_thread_comp); | |
2384 | + | |
2385 | + atomic_set(&cnxman_running, 1); | |
2386 | + | |
2387 | + /* Make sure we have a node name */ | |
2388 | + if (nodename[0] == '\0') | |
2389 | + strcpy(nodename, system_utsname.nodename); | |
2390 | + | |
2391 | + membership_pid = start_membership_services(kcluster_pid); | |
2392 | + if (membership_pid < 0) { | |
2393 | + quit_threads = 1; | |
2394 | + wait_for_completion(&cluster_thread_comp); | |
2395 | + init_completion(&member_thread_comp); | |
2396 | + return membership_pid; | |
2397 | + } | |
2398 | + | |
2399 | + sm_start(); | |
2400 | + break; | |
2401 | + | |
2402 | + case CLU_LEAVE_CLUSTER: | |
2403 | + if (!capable(CAP_CLUSTER)) | |
2404 | + return -EPERM; | |
2405 | + | |
2406 | + if (optlen != sizeof (int)) | |
2407 | + return -EINVAL; | |
2408 | + | |
2409 | + if (copy_from_user(&leave_flags, optval, optlen)) | |
2410 | + return -EFAULT; | |
2411 | + | |
2412 | + if (!atomic_read(&cnxman_running)) | |
2413 | + return -ENOTCONN; | |
2414 | + | |
2415 | + if (in_transition()) | |
2416 | + return -EBUSY; | |
2417 | + | |
2418 | + /* Ignore the use count if FORCE is set */ | |
2419 | + if (!(leave_flags & CLUSTER_LEAVEFLAG_FORCE)) { | |
2420 | + if (atomic_read(&use_count)) | |
2421 | + return -ENOTCONN; | |
2422 | + } | |
2423 | + | |
2424 | + us->leave_reason = leave_flags; | |
2425 | + quit_threads = 1; | |
2426 | + wake_up_interruptible(&cnxman_waitq); | |
2427 | + | |
2428 | + wait_for_completion(&cluster_thread_comp); | |
2429 | + break; | |
2430 | + | |
2431 | + default: | |
2432 | + return -ENOPROTOOPT; | |
2433 | + } | |
2434 | + | |
2435 | + return 0; | |
2436 | +} | |
2437 | + | |
2438 | +static int cl_getsockopt(struct socket *sock, int level, int optname, | |
2439 | + char *optval, int *optlen) | |
2440 | +{ | |
2441 | + struct sock *sk = sock->sk; | |
2442 | + int err; | |
2443 | + | |
2444 | + lock_sock(sk); | |
2445 | + err = __cl_getsockopt(sock, level, optname, optval, optlen, 0); | |
2446 | + release_sock(sk); | |
2447 | + | |
2448 | + return err; | |
2449 | +} | |
2450 | + | |
2451 | +static int __cl_getsockopt(struct socket *sock, int level, int optname, | |
2452 | + char *optval, int *optlen, int flags) | |
2453 | +{ | |
2454 | + | |
2455 | + switch (optname) { | |
2456 | + default: | |
2457 | + return -ENOPROTOOPT; | |
2458 | + } | |
2459 | + | |
2460 | + return 0; | |
2461 | +} | |
2462 | + | |
2463 | +/* We'll be giving out reward points next... */ | |
2464 | +/* Send the packet and save a copy in case someone loses theirs. Should be | |
2465 | + * protected by the send mutexphore */ | |
2466 | +static int __send_and_save(struct cl_comms_socket *csock, struct msghdr *msg, | |
2467 | + int size, int needack) | |
2468 | +{ | |
2469 | + mm_segment_t fs; | |
2470 | + int result; | |
2471 | + struct iovec save_vectors[msg->msg_iovlen]; | |
2472 | + | |
2473 | + /* Save a copy of the IO vectors as send_msg mucks around with them and | |
2474 | + * we may want to send the same stuff out more than once (for different | |
2475 | + * interfaces) | |
2476 | + */ | |
2477 | + memcpy(save_vectors, msg->msg_iov, | |
2478 | + sizeof (struct iovec) * msg->msg_iovlen); | |
2479 | + | |
2480 | + fs = get_fs(); | |
2481 | + set_fs(get_ds()); | |
2482 | + | |
2483 | + result = sock_sendmsg(csock->sock, msg, size); | |
2484 | + | |
2485 | + set_fs(fs); | |
2486 | + | |
2487 | + if (result >= 0 && acks_expected && needack) { | |
2488 | + | |
2489 | + /* Start retransmit timer if it didn't go */ | |
2490 | + if (result == 0) { | |
2491 | + start_short_timer(); | |
2492 | + } | |
2493 | + else { | |
2494 | + resend_delay = 1; | |
2495 | + } | |
2496 | + } | |
2497 | + | |
2498 | + /* Restore IOVs */ | |
2499 | + memcpy(msg->msg_iov, save_vectors, | |
2500 | + sizeof (struct iovec) * msg->msg_iovlen); | |
2501 | + | |
2502 | + return result; | |
2503 | +} | |
2504 | + | |
2505 | +static void resend_last_message() | |
2506 | +{ | |
2507 | + struct msghdr msg; | |
2508 | + struct iovec vec[1]; | |
2509 | + mm_segment_t fs; | |
2510 | + int result; | |
2511 | + | |
2512 | + P_COMMS("%ld resending last message: %d bytes: port=%d, cmd=%d\n", | |
2513 | + jiffies, saved_msg_len, saved_msg_buffer[0], | |
2514 | + saved_msg_buffer[6]); | |
2515 | + | |
2516 | + /* Assume there is something wrong with the last interface */ | |
2517 | + current_interface = get_next_interface(current_interface); | |
2518 | + if (num_interfaces > 1) | |
2519 | + printk(KERN_WARNING CMAN_NAME ": Now using interface %d\n", | |
2520 | + current_interface->number); | |
2521 | + | |
2522 | + vec[0].iov_base = saved_msg_buffer; | |
2523 | + vec[0].iov_len = saved_msg_len; | |
2524 | + | |
2525 | + memset(&msg, 0, sizeof (msg)); | |
2526 | + msg.msg_name = ¤t_interface->saddr; | |
2527 | + msg.msg_namelen = current_interface->addr_len; | |
2528 | + msg.msg_iovlen = 1; | |
2529 | + msg.msg_iov = vec; | |
2530 | + | |
2531 | + fs = get_fs(); | |
2532 | + set_fs(get_ds()); | |
2533 | + | |
2534 | + result = sock_sendmsg(current_interface->sock, &msg, saved_msg_len); | |
2535 | + | |
2536 | + set_fs(fs); | |
2537 | + | |
2538 | + if (result < 0) | |
2539 | + printk(KERN_ERR CMAN_NAME ": resend failed: %d\n", result); | |
2540 | + | |
2541 | + /* Try indefinitely to send this, the backlog must die down eventually | |
2542 | + * !? */ | |
2543 | + if (result == 0) | |
2544 | + start_short_timer(); | |
2545 | + | |
2546 | + /* Send succeeded, continue waiting for ACKS */ | |
2547 | + if (result > 0) | |
2548 | + start_ack_timer(); | |
2549 | + | |
2550 | +} | |
2551 | + | |
2552 | +static int cl_recvmsg(struct kiocb *iocb, struct socket *sock, | |
2553 | + struct msghdr *msg, size_t size, int flags) | |
2554 | +{ | |
2555 | + struct sock *sk = sock->sk; | |
2556 | + struct sockaddr_cl *sin = (struct sockaddr_cl *) msg->msg_name; | |
2557 | + struct cluster_sock *c = cluster_sk(sk); | |
2558 | + struct sk_buff *skb; | |
2559 | + int copied, err = 0; | |
2560 | + int isoob = 0; | |
2561 | + | |
2562 | + /* Socket was notified of shutdown, remove any pending skbs and return | |
2563 | + * EOF */ | |
2564 | + if (!atomic_read(&cnxman_running)) { | |
2565 | + while ((skb = skb_recv_datagram(sk, flags, MSG_DONTWAIT, &err))) | |
2566 | + skb_free_datagram(sk, skb); | |
2567 | + return 0; /* cnxman has left the building */ | |
2568 | + } | |
2569 | + | |
2570 | + /* Generic datagram code does most of the work. If the user is not | |
2571 | + * interested in OOB messages then ignore them */ | |
2572 | + do { | |
2573 | + skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); | |
2574 | + if (!skb) | |
2575 | + goto out; | |
2576 | + | |
2577 | + /* Is it OOB */ | |
2578 | + if (skb->cb[0] & 0x80) | |
2579 | + isoob = 1; | |
2580 | + else | |
2581 | + isoob = 0; | |
2582 | + | |
2583 | + /* If it is and the user doesn't want it, then throw it away. */ | |
2584 | + if (isoob && !(flags & MSG_OOB)) { | |
2585 | + skb_free_datagram(sk, skb); | |
2586 | + | |
2587 | + /* If we peeked (?) an OOB but the user doesn't want it | |
2588 | + then we need to discard it or we'll loop forever */ | |
2589 | + if (flags & MSG_PEEK) { | |
2590 | + skb = skb_recv_datagram(sk, flags & ~MSG_PEEK, | |
2591 | + MSG_DONTWAIT, &err); | |
2592 | + if (skb) | |
2593 | + skb_free_datagram(sk, skb); | |
2594 | + } | |
2595 | + } | |
2596 | + } | |
2597 | + while (isoob && !(flags & MSG_OOB)); | |
2598 | + | |
2599 | + copied = skb->len; | |
2600 | + if (copied > size) { | |
2601 | + copied = size; | |
2602 | + msg->msg_flags |= MSG_TRUNC; | |
2603 | + } | |
2604 | + err = memcpy_toiovec(msg->msg_iov, skb->data, copied); | |
2605 | + | |
2606 | + if (err) | |
2607 | + goto out_free; | |
2608 | + | |
2609 | + if (msg->msg_name && msg->msg_namelen) { | |
2610 | + memset(msg->msg_name, 0, msg->msg_namelen); | |
2611 | + | |
2612 | + if (msg->msg_namelen >= sizeof (struct sockaddr_cl)) { | |
2613 | + | |
2614 | + /* Nodeid is in native byte order - anything else is just | |
2615 | + * perverse */ | |
2616 | + memcpy(&sin->scl_nodeid, skb->cb + 1, sizeof(int)); | |
2617 | + } | |
2618 | + msg->msg_namelen = sizeof (struct sockaddr_cl); | |
2619 | + sin->scl_port = c->port; | |
2620 | + } | |
2621 | + | |
2622 | + /* Top bit set in cb[0] means this is an OOB message */ | |
2623 | + if (skb->cb[0] & 0x80) { | |
2624 | + msg->msg_flags |= MSG_OOB; | |
2625 | + } | |
2626 | + | |
2627 | + sock_recv_timestamp(msg, sk, skb); | |
2628 | + | |
2629 | + err = copied; | |
2630 | + | |
2631 | + out_free: | |
2632 | + skb_free_datagram(sk, skb); | |
2633 | + | |
2634 | + out: | |
2635 | + return err; | |
2636 | +} | |
2637 | + | |
2638 | +/* Send a message out on all interfaces */ | |
2639 | +static int send_to_all_ints(int nodeid, struct msghdr *our_msg, int size, int flags) | |
2640 | +{ | |
2641 | + struct sockaddr_in6 daddr; | |
2642 | + struct cl_comms_socket *clsock; | |
2643 | + int result = 0; | |
2644 | + | |
2645 | + our_msg->msg_name = &daddr; | |
2646 | + | |
2647 | + list_for_each_entry(clsock, &socket_list, list) { | |
2648 | + | |
2649 | + /* Don't send out a recv-only socket */ | |
2650 | + if (!clsock->recv_only) { | |
2651 | + | |
2652 | + /* For temporary node IDs send to the node's real IP address */ | |
2653 | + if (nodeid < 0) { | |
2654 | + get_addr_from_temp_nodeid(nodeid, (char *)&daddr, &our_msg->msg_namelen); | |
2655 | + } | |
2656 | + else { | |
2657 | + memcpy(&daddr, &clsock->saddr, clsock->addr_len); | |
2658 | + our_msg->msg_namelen = clsock->addr_len; | |
2659 | + } | |
2660 | + | |
2661 | + result = __send_and_save(clsock, our_msg, | |
2662 | + size + sizeof (struct cl_protheader), | |
2663 | + !(flags & MSG_NOACK)); | |
2664 | + } | |
2665 | + } | |
2666 | + return result; | |
2667 | +} | |
2668 | + | |
2669 | + | |
2670 | +/* Internal common send message routine */ | |
2671 | +static int __sendmsg(struct socket *sock, struct msghdr *msg, int size, | |
2672 | + unsigned char port) | |
2673 | +{ | |
2674 | + int result = 0, i; | |
2675 | + int flags = msg->msg_flags; | |
2676 | + struct msghdr our_msg; | |
2677 | + struct sockaddr_cl *caddr = msg->msg_name; | |
2678 | + struct cl_protheader header; | |
2679 | + struct iovec vectors[msg->msg_iovlen + 1]; | |
2680 | + int nodeid = 0; | |
2681 | + | |
2682 | + if (size > MAX_CLUSTER_MESSAGE) | |
2683 | + return -EINVAL; | |
2684 | + if (!atomic_read(&cnxman_running)) | |
2685 | + return -ENOTCONN; | |
2686 | + | |
2687 | + if (caddr) | |
2688 | + nodeid = caddr->scl_nodeid; | |
2689 | + | |
2690 | + /* Check that the node id (if present) is valid */ | |
2691 | + if (msg->msg_namelen && (!find_node_by_nodeid(nodeid) && | |
2692 | + !is_valid_temp_nodeid(nodeid))) { | |
2693 | + return -ENOTCONN; | |
2694 | + } | |
2695 | + | |
2696 | + /* We can only have one send outstanding at a time so we might as well | |
2697 | + * lock the whole send mechanism */ | |
2698 | + down(&send_lock); | |
2699 | + | |
2700 | + while ((port > HIGH_PROTECTED_PORT | |
2701 | + && (!cluster_is_quorate || in_transition())) | |
2702 | + || (acks_expected > 0 && !(msg->msg_flags & MSG_NOACK))) { | |
2703 | + | |
2704 | + DECLARE_WAITQUEUE(wq, current); | |
2705 | + struct task_struct *tsk = current; | |
2706 | + | |
2707 | + if (flags & MSG_DONTWAIT) { | |
2708 | + up(&send_lock); | |
2709 | + return -EAGAIN; | |
2710 | + } | |
2711 | + | |
2712 | + if (current->pid == kcluster_pid) { | |
2713 | + P_COMMS | |
2714 | + ("Tried to make kclusterd wait, port=%d, acks_count=%d, expected=%d\n", | |
2715 | + port, ack_count, acks_expected); | |
2716 | + up(&send_lock); | |
2717 | + return -EAGAIN; | |
2718 | + } | |
2719 | + | |
2720 | + P_COMMS("%s process waiting. acks=%d, expected=%d\n", tsk->comm, | |
2721 | + ack_count, acks_expected); | |
2722 | + | |
2723 | + set_task_state(tsk, TASK_INTERRUPTIBLE); | |
2724 | + add_wait_queue(&socket_waitq, &wq); | |
2725 | + | |
2726 | + if ((port > HIGH_PROTECTED_PORT | |
2727 | + && (!cluster_is_quorate || in_transition())) | |
2728 | + || (acks_expected > 0)) { | |
2729 | + | |
2730 | + up(&send_lock); | |
2731 | + schedule(); | |
2732 | + down(&send_lock); | |
2733 | + } | |
2734 | + | |
2735 | + /* Going down */ | |
2736 | + if (quit_threads) { | |
2737 | + up(&send_lock); | |
2738 | + return -ENOTCONN; | |
2739 | + } | |
2740 | + | |
2741 | + set_task_state(tsk, TASK_RUNNING); | |
2742 | + remove_wait_queue(&socket_waitq, &wq); | |
2743 | + | |
2744 | + if (signal_pending(current)) { | |
2745 | + up(&send_lock); | |
2746 | + return -ERESTARTSYS; | |
2747 | + } | |
2748 | + | |
2749 | + /* Were we shut down in the meantime ? */ | |
2750 | + if (!atomic_read(&cnxman_running)) { | |
2751 | + up(&send_lock); | |
2752 | + return -ENOTCONN; | |
2753 | + } | |
2754 | + | |
2755 | + } | |
2756 | + | |
2757 | + memset(&our_msg, 0, sizeof (our_msg)); | |
2758 | + | |
2759 | + /* Build the header */ | |
2760 | + header.port = port; | |
2761 | + header.flags = msg->msg_flags >> 16; | |
2762 | + header.cluster = cpu_to_le16(cluster_id); | |
2763 | + header.srcid = us ? cpu_to_le32(us->node_id) : 0; | |
2764 | + header.tgtid = caddr ? cpu_to_le32(nodeid) : 0; | |
2765 | + | |
2766 | + ++cur_seq; | |
2767 | + header.seq = cpu_to_le16(cur_seq); | |
2768 | + | |
2769 | + /* Set the MULTICAST flag on messages with no particular destination */ | |
2770 | + if (!msg->msg_namelen) { | |
2771 | + header.flags |= MSG_MULTICAST >> 16; | |
2772 | + header.tgtid = 0; | |
2773 | + } | |
2774 | + | |
2775 | + /* Copy the existing iovecs into our array and add the header on at the | |
2776 | + * beginning */ | |
2777 | + vectors[0].iov_base = &header; | |
2778 | + vectors[0].iov_len = sizeof (header); | |
2779 | + for (i = 0; i < msg->msg_iovlen; i++) { | |
2780 | + vectors[i + 1] = msg->msg_iov[i]; | |
2781 | + } | |
2782 | + | |
2783 | + our_msg.msg_iovlen = msg->msg_iovlen + 1; | |
2784 | + our_msg.msg_iov = vectors; | |
2785 | + | |
2786 | + /* Work out how many ACKS are wanted - *don't* reset acks_expected to | |
2787 | + * zero if no acks are required as an ACK-needed message may still be | |
2788 | + * outstanding */ | |
2789 | + if (!(msg->msg_flags & MSG_NOACK)) { | |
2790 | + if (msg->msg_namelen) | |
2791 | + acks_expected = 1; /* Unicast */ | |
2792 | + else | |
2793 | + acks_expected = max(cluster_members - 1, 0); | |
2794 | + | |
2795 | + } | |
2796 | + | |
2797 | + P_COMMS | |
2798 | + ("Sending message - tgt=%d port %d required %d acks, seq=%d, flags=%x\n", | |
2799 | + nodeid, header.port, | |
2800 | + (msg->msg_flags & MSG_NOACK) ? 0 : acks_expected, | |
2801 | + le16_to_cpu(header.seq), header.flags); | |
2802 | + | |
2803 | + /* Don't include temp nodeids in the message itself */ | |
2804 | + if (header.tgtid < 0) | |
2805 | + header.tgtid = 0; | |
2806 | + | |
2807 | + /* For non-member sends we use all the interfaces */ | |
2808 | + if ((nodeid < 0) || (flags & MSG_ALLINT)) { | |
2809 | + | |
2810 | + result = send_to_all_ints(nodeid, &our_msg, size, msg->msg_flags); | |
2811 | + } | |
2812 | + else { | |
2813 | + /* Send to only the current socket - resends will use the | |
2814 | + * others if necessary */ | |
2815 | + our_msg.msg_name = ¤t_interface->saddr; | |
2816 | + our_msg.msg_namelen = current_interface->addr_len; | |
2817 | + | |
2818 | + result = | |
2819 | + __send_and_save(current_interface, &our_msg, | |
2820 | + size + sizeof (header), | |
2821 | + !(msg->msg_flags & MSG_NOACK)); | |
2822 | + } | |
2823 | + | |
2824 | + /* Make a note in each nodes' structure that it has been sent a message | |
2825 | + * so we can see which ones went astray */ | |
2826 | + if (!(flags & MSG_NOACK) && nodeid >= 0) { | |
2827 | + if (msg->msg_namelen) { | |
2828 | + struct cluster_node *node; | |
2829 | + | |
2830 | + node = find_node_by_nodeid(le32_to_cpu(header.tgtid)); | |
2831 | + if (node) | |
2832 | + node->last_seq_sent = cur_seq; | |
2833 | + } | |
2834 | + else { | |
2835 | + struct cluster_node *node; | |
2836 | + struct list_head *nodelist; | |
2837 | + | |
2838 | + list_for_each(nodelist, &cluster_members_list) { | |
2839 | + node = | |
2840 | + list_entry(nodelist, struct cluster_node, | |
2841 | + list); | |
2842 | + if (node->state == NODESTATE_MEMBER) { | |
2843 | + node->last_seq_sent = cur_seq; | |
2844 | + } | |
2845 | + } | |
2846 | + } | |
2847 | + } | |
2848 | + | |
2849 | + /* Save a copy of the message if we're expecting an ACK */ | |
2850 | + if (!(flags & MSG_NOACK) && acks_expected) { | |
2851 | + mm_segment_t fs; | |
2852 | + | |
2853 | + fs = get_fs(); | |
2854 | + set_fs(get_ds()); | |
2855 | + | |
2856 | + memcpy_fromiovec(saved_msg_buffer, our_msg.msg_iov, | |
2857 | + size + sizeof (header)); | |
2858 | + set_fs(fs); | |
2859 | + | |
2860 | + saved_msg_len = size + sizeof (header); | |
2861 | + retry_count = ack_count = 0; | |
2862 | + clear_bit(RESEND_NEEDED, &mainloop_flags); | |
2863 | + | |
2864 | + start_ack_timer(); | |
2865 | + } | |
2866 | + | |
2867 | + up(&send_lock); | |
2868 | + return result; | |
2869 | +} | |
2870 | + | |
2871 | +static int queue_message(void *buf, int len, struct sockaddr_cl *caddr, | |
2872 | + unsigned char port, int flags) | |
2873 | +{ | |
2874 | + struct queued_message *qmsg; | |
2875 | + | |
2876 | + qmsg = kmalloc(sizeof (struct queued_message), | |
2877 | + (in_atomic() | |
2878 | + || irqs_disabled())? GFP_ATOMIC : GFP_KERNEL); | |
2879 | + if (qmsg == NULL) | |
2880 | + return -1; | |
2881 | + | |
2882 | + memcpy(qmsg->msg_buffer, buf, len); | |
2883 | + qmsg->msg_len = len; | |
2884 | + if (caddr) { | |
2885 | + memcpy(&qmsg->addr, caddr, sizeof (struct sockaddr_cl)); | |
2886 | + qmsg->addr_len = sizeof (struct sockaddr_cl); | |
2887 | + } | |
2888 | + else { | |
2889 | + qmsg->addr_len = 0; | |
2890 | + } | |
2891 | + qmsg->flags = flags; | |
2892 | + qmsg->port = port; | |
2893 | + qmsg->socket = NULL; | |
2894 | + | |
2895 | + down(&messages_list_lock); | |
2896 | + list_add_tail(&qmsg->list, &messages_list); | |
2897 | + up(&messages_list_lock); | |
2898 | + | |
2899 | + wake_up_interruptible(&cnxman_waitq); | |
2900 | + | |
2901 | + return 0; | |
2902 | +} | |
2903 | + | |
2904 | +static int cl_sendmsg(struct kiocb *iocb, struct socket *sock, | |
2905 | + struct msghdr *msg, size_t size) | |
2906 | +{ | |
2907 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
2908 | + char *buffer; | |
2909 | + int status; | |
2910 | + int saved_iovlen; | |
2911 | + uint8_t port; | |
2912 | + struct iovec iov; | |
2913 | + struct iovec *saved_iov; | |
2914 | + struct sockaddr_cl *caddr = msg->msg_name; | |
2915 | + | |
2916 | + if (sock->sk->sk_protocol == CLPROTO_MASTER) | |
2917 | + return -EOPNOTSUPP; | |
2918 | + | |
2919 | + port = c->port; | |
2920 | + | |
2921 | + /* Only capable users can override the port number */ | |
2922 | + if (caddr && capable(CAP_CLUSTER) && caddr->scl_port) | |
2923 | + port = caddr->scl_port; | |
2924 | + | |
2925 | + if (port == 0) | |
2926 | + return -EDESTADDRREQ; | |
2927 | + | |
2928 | + /* Hmmm. On machines with segmented user/kernel space (sparc64, hppa & | |
2929 | + * m68k AFAICT) we can't mix user and kernel space addresses in the | |
2930 | + * IOV. This stymies __sendmsg a little as it tries to add a header to | |
2931 | + * what could possibly be a userspace iov. So, here (where all the | |
2932 | + * userspace sends come) we copy it to a kernel space buffer first. If | |
2933 | + * performance is a big problem here then I might #ifdef it for the | |
2934 | + * affected architectures but for now I think it will probably be OK */ | |
2935 | + buffer = kmalloc(size, GFP_KERNEL); | |
2936 | + if (!buffer) | |
2937 | + return -ENOMEM; | |
2938 | + | |
2939 | + memcpy_fromiovec(buffer, msg->msg_iov, size); | |
2940 | + iov.iov_len = size; | |
2941 | + iov.iov_base = buffer; | |
2942 | + | |
2943 | + saved_iov = msg->msg_iov; | |
2944 | + saved_iovlen = msg->msg_iovlen; | |
2945 | + msg->msg_iov = &iov; | |
2946 | + msg->msg_iovlen = 1; | |
2947 | + | |
2948 | + status = __sendmsg(sock, msg, size, port); | |
2949 | + msg->msg_iov = saved_iov; | |
2950 | + msg->msg_iovlen = saved_iovlen; | |
2951 | + | |
2952 | + kfree(buffer); | |
2953 | + | |
2954 | + return status; | |
2955 | +} | |
2956 | + | |
2957 | +/* Kernel call to sendmsg */ | |
2958 | +int kcl_sendmsg(struct socket *sock, void *buf, int size, | |
2959 | + struct sockaddr_cl *caddr, int addr_len, unsigned int flags) | |
2960 | +{ | |
2961 | + struct iovec iovecs[1]; | |
2962 | + struct msghdr msg; | |
2963 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
2964 | + unsigned char port; | |
2965 | + | |
2966 | + if (size > MAX_CLUSTER_MESSAGE) | |
2967 | + return -EINVAL; | |
2968 | + if (!atomic_read(&cnxman_running)) | |
2969 | + return -ENOTCONN; | |
2970 | + | |
2971 | + port = c->port; | |
2972 | + if (caddr && caddr->scl_port) | |
2973 | + port = caddr->scl_port; | |
2974 | + | |
2975 | + if (port == 0) | |
2976 | + return -EDESTADDRREQ; | |
2977 | + | |
2978 | + /* If we have no process context then queue it up for kclusterd to | |
2979 | + * send. */ | |
2980 | + if (in_interrupt() || flags & MSG_QUEUE) { | |
2981 | + return queue_message(buf, size, caddr, port, | |
2982 | + flags & ~MSG_QUEUE); | |
2983 | + } | |
2984 | + | |
2985 | + iovecs[0].iov_base = buf; | |
2986 | + iovecs[0].iov_len = size; | |
2987 | + | |
2988 | + memset(&msg, 0, sizeof (msg)); | |
2989 | + msg.msg_name = caddr; | |
2990 | + msg.msg_namelen = addr_len; | |
2991 | + msg.msg_iovlen = 1; | |
2992 | + msg.msg_iov = iovecs; | |
2993 | + msg.msg_flags = flags; | |
2994 | + | |
2995 | + return __sendmsg(sock, &msg, size, port); | |
2996 | +} | |
2997 | + | |
2998 | +static int send_queued_message(struct queued_message *qmsg) | |
2999 | +{ | |
3000 | + struct iovec iovecs[1]; | |
3001 | + struct msghdr msg; | |
3002 | + | |
3003 | + /* Don't send blocked messages */ | |
3004 | + if (qmsg->port > HIGH_PROTECTED_PORT | |
3005 | + && (!cluster_is_quorate || in_transition())) | |
3006 | + return -EAGAIN; | |
3007 | + | |
3008 | + iovecs[0].iov_base = qmsg->msg_buffer; | |
3009 | + iovecs[0].iov_len = qmsg->msg_len; | |
3010 | + | |
3011 | + memset(&msg, 0, sizeof (msg)); | |
3012 | + msg.msg_name = qmsg->addr_len ? &qmsg->addr : NULL; | |
3013 | + msg.msg_namelen = qmsg->addr_len; | |
3014 | + msg.msg_iovlen = 1; | |
3015 | + msg.msg_iov = iovecs; | |
3016 | + msg.msg_flags = qmsg->flags; | |
3017 | + | |
3018 | + return __sendmsg(qmsg->socket, &msg, qmsg->msg_len, qmsg->port); | |
3019 | +} | |
3020 | + | |
3021 | +int kcl_register_read_callback(struct socket *sock, | |
3022 | + int (*routine) (char *, int, char *, int, | |
3023 | + unsigned int)) | |
3024 | +{ | |
3025 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
3026 | + | |
3027 | + c->kernel_callback = routine; | |
3028 | + | |
3029 | + return 0; | |
3030 | +} | |
3031 | + | |
3032 | +/* Used where we are in kclusterd context and we can't allow the task to wait | |
3033 | + * as we are also responsible to processing the ACKs that do the wake up. Try | |
3034 | + * to send the message immediately and queue it if that's not possible */ | |
3035 | +static int send_or_queue_message(void *buf, int len, struct sockaddr_cl *caddr, | |
3036 | + unsigned char port) | |
3037 | +{ | |
3038 | + struct iovec iovecs[1]; | |
3039 | + struct msghdr msg; | |
3040 | + | |
3041 | + int status; | |
3042 | + | |
3043 | + /* Don't send blocked messages */ | |
3044 | + if (port > HIGH_PROTECTED_PORT | |
3045 | + && (!cluster_is_quorate || in_transition())) { | |
3046 | + return queue_message(buf, len, caddr, port, 0); | |
3047 | + } | |
3048 | + | |
3049 | + iovecs[0].iov_base = buf; | |
3050 | + iovecs[0].iov_len = len; | |
3051 | + | |
3052 | + memset(&msg, 0, sizeof (msg)); | |
3053 | + msg.msg_name = caddr; | |
3054 | + msg.msg_namelen = caddr ? sizeof (struct sockaddr_cl) : 0; | |
3055 | + msg.msg_iovlen = 1; | |
3056 | + msg.msg_iov = iovecs; | |
3057 | + msg.msg_flags = MSG_DONTWAIT; | |
3058 | + | |
3059 | + status = __sendmsg(NULL, &msg, len, port); | |
3060 | + | |
3061 | + /* Did it work ? */ | |
3062 | + if (status > 0) { | |
3063 | + return 0; | |
3064 | + } | |
3065 | + | |
3066 | + /* Failure other than EAGAIN is fatal */ | |
3067 | + if (status != -EAGAIN) { | |
3068 | + return status; | |
3069 | + } | |
3070 | + | |
3071 | + return queue_message(buf, len, caddr, port, 0); | |
3072 | +} | |
3073 | + | |
3074 | +/* Send a listen request to a node */ | |
3075 | +static void send_listen_request(int nodeid, unsigned char port) | |
3076 | +{ | |
3077 | + struct cl_listenmsg listenmsg; | |
3078 | + struct sockaddr_cl caddr; | |
3079 | + | |
3080 | + memset(&caddr, 0, sizeof (caddr)); | |
3081 | + | |
3082 | + /* Build the header */ | |
3083 | + listenmsg.cmd = CLUSTER_CMD_LISTENREQ; | |
3084 | + listenmsg.target_port = port; | |
3085 | + listenmsg.listening = 0; | |
3086 | + listenmsg.tag = current->pid; | |
3087 | + | |
3088 | + caddr.scl_family = AF_CLUSTER; | |
3089 | + caddr.scl_port = 0; | |
3090 | + caddr.scl_nodeid = nodeid; | |
3091 | + | |
3092 | + send_or_queue_message(&listenmsg, sizeof(listenmsg), &caddr, 0); | |
3093 | + return; | |
3094 | +} | |
3095 | + | |
3096 | +/* Return 1 or 0 to indicate if we have a listener on the requested port */ | |
3097 | +static void send_listen_response(struct cl_comms_socket *csock, int nodeid, | |
3098 | + unsigned char port, unsigned short tag) | |
3099 | +{ | |
3100 | + struct cl_listenmsg listenmsg; | |
3101 | + struct sockaddr_cl caddr; | |
3102 | + int status; | |
3103 | + | |
3104 | + memset(&caddr, 0, sizeof (caddr)); | |
3105 | + | |
3106 | + /* Build the message */ | |
3107 | + listenmsg.cmd = CLUSTER_CMD_LISTENRESP; | |
3108 | + listenmsg.target_port = port; | |
3109 | + listenmsg.tag = tag; | |
3110 | + listenmsg.listening = (port_array[port] != 0) ? 1 : 0; | |
3111 | + | |
3112 | + caddr.scl_family = AF_CLUSTER; | |
3113 | + caddr.scl_port = 0; | |
3114 | + caddr.scl_nodeid = nodeid; | |
3115 | + | |
3116 | + status = send_or_queue_message(&listenmsg, | |
3117 | + sizeof (listenmsg), | |
3118 | + &caddr, 0); | |
3119 | + | |
3120 | + return; | |
3121 | +} | |
3122 | + | |
3123 | +/* Send an ACK */ | |
3124 | +static int cl_sendack(struct cl_comms_socket *csock, unsigned short seq, | |
3125 | + int addr_len, char *addr, unsigned char remport, | |
3126 | + unsigned char flag) | |
3127 | +{ | |
3128 | + mm_segment_t fs; | |
3129 | + struct iovec vec; | |
3130 | + struct cl_ackmsg ackmsg; | |
3131 | + struct msghdr msg; | |
3132 | + struct sockaddr_in6 daddr; | |
3133 | + int result; | |
3134 | + | |
3135 | +#ifdef DEBUG_COMMS | |
3136 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
3137 | + | |
3138 | + P_COMMS("Sending ACK to %s, seq=%d\n", | |
3139 | + print_addr(addr, address_length, buf), le16_to_cpu(seq)); | |
3140 | +#endif | |
3141 | + | |
3142 | + if (addr) { | |
3143 | + memcpy(&daddr, addr, addr_len); | |
3144 | + } | |
3145 | + else { | |
3146 | + memcpy(&daddr, &csock->saddr, csock->addr_len); | |
3147 | + addr_len = csock->addr_len; | |
3148 | + } | |
3149 | + | |
3150 | + /* Build the header */ | |
3151 | + ackmsg.header.port = 0; /* Protocol port */ | |
3152 | + ackmsg.header.seq = 0; | |
3153 | + ackmsg.header.flags = MSG_NOACK >> 16; | |
3154 | + ackmsg.header.cluster = cpu_to_le16(cluster_id); | |
3155 | + ackmsg.header.srcid = us ? cpu_to_le32(us->node_id) : 0; | |
3156 | + ackmsg.header.tgtid = 0; /* ACKS are unicast so we don't bother | |
3157 | + * to look this up */ | |
3158 | + ackmsg.cmd = CLUSTER_CMD_ACK; | |
3159 | + ackmsg.remport = remport; | |
3160 | + ackmsg.aflags = flag; | |
3161 | + ackmsg.seq = seq; /* Already in LE order */ | |
3162 | + vec.iov_base = &ackmsg; | |
3163 | + vec.iov_len = sizeof (ackmsg); | |
3164 | + | |
3165 | + memset(&msg, 0, sizeof (msg)); | |
3166 | + msg.msg_name = &daddr; | |
3167 | + msg.msg_namelen = addr_len; | |
3168 | + msg.msg_iovlen = 1; | |
3169 | + msg.msg_iov = &vec; | |
3170 | + | |
3171 | + fs = get_fs(); | |
3172 | + set_fs(get_ds()); | |
3173 | + | |
3174 | + result = sock_sendmsg(csock->sock, &msg, sizeof (ackmsg)); | |
3175 | + | |
3176 | + set_fs(fs); | |
3177 | + | |
3178 | + if (result < 0) | |
3179 | + printk(KERN_CRIT CMAN_NAME ": error sending ACK: %d\n", result); | |
3180 | + | |
3181 | + return result; | |
3182 | + | |
3183 | +} | |
3184 | + | |
3185 | +/* Wait for all ACKS to be gathered */ | |
3186 | +void kcl_wait_for_all_acks() | |
3187 | +{ | |
3188 | + while (ack_count < acks_expected) { | |
3189 | + | |
3190 | + DECLARE_WAITQUEUE(wq, current); | |
3191 | + struct task_struct *tsk = current; | |
3192 | + | |
3193 | + set_task_state(tsk, TASK_INTERRUPTIBLE); | |
3194 | + add_wait_queue(&socket_waitq, &wq); | |
3195 | + | |
3196 | + if (ack_count < acks_expected) { | |
3197 | + schedule(); | |
3198 | + } | |
3199 | + | |
3200 | + set_task_state(tsk, TASK_RUNNING); | |
3201 | + remove_wait_queue(&socket_waitq, &wq); | |
3202 | + } | |
3203 | +} | |
3204 | + | |
3205 | +/* Send a closedown OOB message to all cluster nodes - this tells them that a | |
3206 | + * port listener has gone away */ | |
3207 | +static void send_port_close_oob(unsigned char port) | |
3208 | +{ | |
3209 | + struct cl_closemsg closemsg; | |
3210 | + | |
3211 | + /* Build the header */ | |
3212 | + closemsg.cmd = CLUSTER_CMD_PORTCLOSED; | |
3213 | + closemsg.port = port; | |
3214 | + | |
3215 | + send_or_queue_message(&closemsg, sizeof (closemsg), NULL, 0); | |
3216 | + return; | |
3217 | +} | |
3218 | + | |
3219 | +/* A remote port has been closed - post an OOB message to the local listen on | |
3220 | + * that port (if there is one) */ | |
3221 | +static void post_close_oob(unsigned char port, int nodeid) | |
3222 | +{ | |
3223 | + struct cl_portclosed_oob *oobmsg; | |
3224 | + struct sk_buff *skb; | |
3225 | + struct sock *sock = port_array[port]; | |
3226 | + | |
3227 | + if (!sock) { | |
3228 | + return; /* No-one listening */ | |
3229 | + } | |
3230 | + | |
3231 | + skb = alloc_skb(sizeof (*oobmsg), GFP_KERNEL); | |
3232 | + if (!skb) | |
3233 | + return; | |
3234 | + | |
3235 | + skb_put(skb, sizeof (*oobmsg)); | |
3236 | + oobmsg = (struct cl_portclosed_oob *) skb->data; | |
3237 | + oobmsg->port = port; | |
3238 | + oobmsg->cmd = CLUSTER_OOB_MSG_PORTCLOSED; | |
3239 | + skb->cb[0] = 0x80; | |
3240 | + memcpy(skb->cb + 1, &nodeid, sizeof(int)); | |
3241 | + | |
3242 | + sock_queue_rcv_skb(sock, skb); | |
3243 | + | |
3244 | +} | |
3245 | + | |
3246 | +/* Leave the cluster */ | |
3247 | +static void node_shutdown() | |
3248 | +{ | |
3249 | + struct cl_barrier *barrier; | |
3250 | + struct list_head *blist; | |
3251 | + struct list_head *temp; | |
3252 | + struct list_head *socklist; | |
3253 | + struct cl_client_socket *csock; | |
3254 | + struct sk_buff *null_skb; | |
3255 | + | |
3256 | + printk(KERN_INFO CMAN_NAME ": we are leaving the cluster\n"); | |
3257 | + | |
3258 | + atomic_set(&cnxman_running, 0); | |
3259 | + unjam(); | |
3260 | + | |
3261 | + /* Notify kernel listeners first */ | |
3262 | + notify_kernel_listeners(LEAVING, 0); | |
3263 | + | |
3264 | + /* Notify client sockets */ | |
3265 | + down(&client_socket_lock); | |
3266 | + list_for_each_safe(socklist, temp, &client_socket_list) { | |
3267 | + csock = list_entry(socklist, struct cl_client_socket, list); | |
3268 | + | |
3269 | + null_skb = alloc_skb(0, GFP_KERNEL); | |
3270 | + if (null_skb) | |
3271 | + sock_queue_rcv_skb(csock->sock->sk, null_skb); | |
3272 | + list_del(&csock->list); | |
3273 | + kfree(csock); | |
3274 | + } | |
3275 | + up(&client_socket_lock); | |
3276 | + we_are_a_cluster_member = 0; | |
3277 | + | |
3278 | + sm_stop(1); | |
3279 | + | |
3280 | + /* Wake up any processes waiting for barriers */ | |
3281 | + down(&barrier_list_lock); | |
3282 | + list_for_each(blist, &barrier_list) { | |
3283 | + barrier = list_entry(blist, struct cl_barrier, list); | |
3284 | + | |
3285 | + /* Cancel any timers */ | |
3286 | + if (timer_pending(&barrier->timer)) | |
3287 | + del_timer(&barrier->timer); | |
3288 | + | |
3289 | + /* Force it to be auto-delete so it discards itself */ | |
3290 | + if (barrier->state == BARRIER_STATE_WAITING) { | |
3291 | + barrier->flags |= BARRIER_ATTR_AUTODELETE; | |
3292 | + wake_up_interruptible(&barrier->waitq); | |
3293 | + } | |
3294 | + else { | |
3295 | + if (barrier->callback) { | |
3296 | + barrier->callback(barrier->name, -ENOTCONN); | |
3297 | + barrier->callback = NULL; | |
3298 | + } | |
3299 | + } | |
3300 | + } | |
3301 | + up(&barrier_list_lock); | |
3302 | + | |
3303 | + /* Wake up any processes waiting for ISLISTENING requests */ | |
3304 | + down(&listenreq_lock); | |
3305 | + list_for_each(blist, &listenreq_list) { | |
3306 | + struct cl_waiting_listen_request *lrequest = | |
3307 | + list_entry(blist, struct cl_waiting_listen_request, list); | |
3308 | + | |
3309 | + if (lrequest->waiting) | |
3310 | + wake_up_interruptible(&lrequest->waitq); | |
3311 | + } | |
3312 | + up(&listenreq_lock); | |
3313 | +} | |
3314 | + | |
3315 | +static void free_cluster_sockets() | |
3316 | +{ | |
3317 | + struct list_head *socklist; | |
3318 | + struct cl_comms_socket *sock; | |
3319 | + struct list_head *temp; | |
3320 | + | |
3321 | + list_for_each_safe(socklist, temp, &socket_list) { | |
3322 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
3323 | + | |
3324 | + list_del(&sock->list); | |
3325 | + fput(sock->file); | |
3326 | + kfree(sock); | |
3327 | + } | |
3328 | + num_interfaces = 0; | |
3329 | + current_interface = NULL; | |
3330 | +} | |
3331 | + | |
3332 | +/* Tidy up after all the rest of the cluster bits have shut down */ | |
3333 | +static void node_cleanup() | |
3334 | +{ | |
3335 | + struct list_head *nodelist; | |
3336 | + struct list_head *proclist; | |
3337 | + struct list_head *temp; | |
3338 | + struct list_head *socklist; | |
3339 | + struct list_head *blist; | |
3340 | + struct cl_comms_socket *sock; | |
3341 | + struct kernel_notify_struct *knotify; | |
3342 | + | |
3343 | + /* Free list of kernel listeners */ | |
3344 | + list_for_each_safe(proclist, temp, &kernel_listener_list) { | |
3345 | + knotify = | |
3346 | + list_entry(proclist, struct kernel_notify_struct, list); | |
3347 | + list_del(&knotify->list); | |
3348 | + kfree(knotify); | |
3349 | + } | |
3350 | + | |
3351 | + /* Mark the sockets as busy so they don't get added to the active | |
3352 | + * sockets list in the next few lines of code before we free them */ | |
3353 | + list_for_each_safe(socklist, temp, &socket_list) { | |
3354 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
3355 | + | |
3356 | + set_bit(1, &sock->active); | |
3357 | + } | |
3358 | + | |
3359 | + /* Tidy the active sockets list */ | |
3360 | + list_for_each_safe(socklist, temp, &active_socket_list) { | |
3361 | + sock = | |
3362 | + list_entry(socklist, struct cl_comms_socket, active_list); | |
3363 | + list_del(&sock->active_list); | |
3364 | + } | |
3365 | + | |
3366 | + /* Free the memory allocated to cluster nodes */ | |
3367 | + free_nodeid_array(); | |
3368 | + down(&cluster_members_lock); | |
3369 | + us = NULL; | |
3370 | + list_for_each_safe(nodelist, temp, &cluster_members_list) { | |
3371 | + | |
3372 | + struct list_head *addrlist; | |
3373 | + struct list_head *addrtemp; | |
3374 | + struct cluster_node *node; | |
3375 | + struct cluster_node_addr *nodeaddr; | |
3376 | + | |
3377 | + node = list_entry(nodelist, struct cluster_node, list); | |
3378 | + | |
3379 | + list_for_each_safe(addrlist, addrtemp, &node->addr_list) { | |
3380 | + nodeaddr = | |
3381 | + list_entry(addrlist, struct cluster_node_addr, | |
3382 | + list); | |
3383 | + | |
3384 | + list_del(&nodeaddr->list); | |
3385 | + kfree(nodeaddr); | |
3386 | + } | |
3387 | + list_del(&node->list); | |
3388 | + kfree(node->name); | |
3389 | + kfree(node); | |
3390 | + } | |
3391 | + cluster_members = 0; | |
3392 | + up(&cluster_members_lock); | |
3393 | + | |
3394 | + /* Free the memory allocated to the outgoing sockets */ | |
3395 | + free_cluster_sockets(); | |
3396 | + | |
3397 | + /* Make sure that all the barriers are deleted */ | |
3398 | + down(&barrier_list_lock); | |
3399 | + list_for_each_safe(blist, temp, &barrier_list) { | |
3400 | + struct cl_barrier *barrier = | |
3401 | + list_entry(blist, struct cl_barrier, list); | |
3402 | + | |
3403 | + list_del(&barrier->list); | |
3404 | + kfree(barrier); | |
3405 | + } | |
3406 | + up(&barrier_list_lock); | |
3407 | + | |
3408 | + kcluster_pid = 0; | |
3409 | + clear_bit(RESEND_NEEDED, &mainloop_flags); | |
3410 | + acks_expected = 0; | |
3411 | +} | |
3412 | + | |
3413 | +/* If "cluster_is_quorate" is 0 then all activity apart from protected ports is | |
3414 | + * blocked. */ | |
3415 | +void set_quorate(int total_votes) | |
3416 | +{ | |
3417 | + int quorate; | |
3418 | + | |
3419 | + if (get_quorum() > total_votes) { | |
3420 | + quorate = 0; | |
3421 | + } | |
3422 | + else { | |
3423 | + quorate = 1; | |
3424 | + } | |
3425 | + | |
3426 | + /* Hide messages during startup state transition */ | |
3427 | + if (we_are_a_cluster_member) { | |
3428 | + if (cluster_is_quorate && !quorate) | |
3429 | + printk(KERN_CRIT CMAN_NAME | |
3430 | + ": quorum lost, blocking activity\n"); | |
3431 | + if (!cluster_is_quorate && quorate) | |
3432 | + printk(KERN_CRIT CMAN_NAME | |
3433 | + ": quorum regained, resuming activity\n"); | |
3434 | + } | |
3435 | + cluster_is_quorate = quorate; | |
3436 | + | |
3437 | + /* Wake up any sleeping processes */ | |
3438 | + if (cluster_is_quorate) { | |
3439 | + unjam(); | |
3440 | + } | |
3441 | + | |
3442 | +} | |
3443 | + | |
3444 | +void queue_oob_skb(struct socket *sock, int cmd) | |
3445 | +{ | |
3446 | + struct sk_buff *skb; | |
3447 | + struct cl_portclosed_oob *oobmsg; | |
3448 | + | |
3449 | + skb = alloc_skb(sizeof (*oobmsg), GFP_KERNEL); | |
3450 | + if (!skb) | |
3451 | + return; | |
3452 | + | |
3453 | + skb_put(skb, sizeof (*oobmsg)); | |
3454 | + oobmsg = (struct cl_portclosed_oob *) skb->data; | |
3455 | + oobmsg->port = 0; | |
3456 | + oobmsg->cmd = cmd; | |
3457 | + | |
3458 | + /* There is no remote node associated with this so | |
3459 | + clear out the field to avoid any accidents */ | |
3460 | + memset(skb->cb, 0, sizeof(int)); | |
3461 | + skb->cb[0] = 0x80; | |
3462 | + | |
3463 | + sock_queue_rcv_skb(sock->sk, skb); | |
3464 | +} | |
3465 | + | |
3466 | +/* Notify interested parties that the cluster configuration has changed */ | |
3467 | +void notify_listeners() | |
3468 | +{ | |
3469 | + struct notify_struct *notify; | |
3470 | + struct list_head *proclist; | |
3471 | + struct list_head *socklist; | |
3472 | + struct list_head *temp; | |
3473 | + | |
3474 | + /* Do kernel listeners first */ | |
3475 | + notify_kernel_listeners(CLUSTER_RECONFIG, 0); | |
3476 | + | |
3477 | + /* Now we deign to tell userspace */ | |
3478 | + down(&event_listener_lock); | |
3479 | + list_for_each_safe(proclist, temp, &event_listener_list) { | |
3480 | + notify = list_entry(proclist, struct notify_struct, list); | |
3481 | + | |
3482 | + /* If the kill fails then remove the process from the list */ | |
3483 | + if (kill_proc(notify->pid, notify->signal, 0) == -ESRCH) { | |
3484 | + list_del(¬ify->list); | |
3485 | + kfree(notify); | |
3486 | + } | |
3487 | + } | |
3488 | + up(&event_listener_lock); | |
3489 | + | |
3490 | + /* Tell userspace processes which want OOB messages */ | |
3491 | + down(&client_socket_lock); | |
3492 | + list_for_each(socklist, &client_socket_list) { | |
3493 | + struct cl_client_socket *csock; | |
3494 | + csock = list_entry(socklist, struct cl_client_socket, list); | |
3495 | + queue_oob_skb(csock->sock, CLUSTER_OOB_MSG_STATECHANGE); | |
3496 | + } | |
3497 | + up(&client_socket_lock); | |
3498 | +} | |
3499 | + | |
3500 | +/* This fills in the list of all addresses for the local node */ | |
3501 | +void get_local_addresses(struct cluster_node *node) | |
3502 | +{ | |
3503 | + struct list_head *socklist; | |
3504 | + struct cl_comms_socket *sock; | |
3505 | + | |
3506 | + list_for_each(socklist, &socket_list) { | |
3507 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
3508 | + | |
3509 | + if (sock->recv_only) { | |
3510 | + add_node_address(node, (char *) &sock->saddr, address_length); | |
3511 | + } | |
3512 | + } | |
3513 | +} | |
3514 | + | |
3515 | + | |
3516 | +static uint16_t generate_cluster_id(char *name) | |
3517 | +{ | |
3518 | + int i; | |
3519 | + int value = 0; | |
3520 | + | |
3521 | + for (i=0; i<strlen(name); i++) { | |
3522 | + value <<= 1; | |
3523 | + value += name[i]; | |
3524 | + } | |
3525 | + return value & 0xFFFF; | |
3526 | +} | |
3527 | + | |
3528 | +/* Return the next comms socket we can use. */ | |
3529 | +static struct cl_comms_socket *get_next_interface(struct cl_comms_socket *cur) | |
3530 | +{ | |
3531 | + int next; | |
3532 | + struct list_head *socklist; | |
3533 | + | |
3534 | + /* Fast path for single interface systems */ | |
3535 | + if (num_interfaces <= 1) | |
3536 | + return cur; | |
3537 | + | |
3538 | + /* Next number */ | |
3539 | + next = cur->number + 1; | |
3540 | + if (next > num_interfaces) | |
3541 | + next = 1; | |
3542 | + | |
3543 | + /* Find the socket with this number, I could optimise this by starting | |
3544 | + * at the current i/f but most systems are going to have a small number | |
3545 | + * of them anyway */ | |
3546 | + list_for_each(socklist, &socket_list) { | |
3547 | + struct cl_comms_socket *sock; | |
3548 | + sock = list_entry(socklist, struct cl_comms_socket, list); | |
3549 | + | |
3550 | + if (!sock->recv_only && sock->number == next) | |
3551 | + return sock; | |
3552 | + } | |
3553 | + | |
3554 | + BUG(); | |
3555 | + return NULL; | |
3556 | +} | |
3557 | + | |
3558 | +/* MUST be called with the barrier list lock held */ | |
3559 | +static struct cl_barrier *find_barrier(char *name) | |
3560 | +{ | |
3561 | + struct list_head *blist; | |
3562 | + struct cl_barrier *bar; | |
3563 | + | |
3564 | + list_for_each(blist, &barrier_list) { | |
3565 | + bar = list_entry(blist, struct cl_barrier, list); | |
3566 | + | |
3567 | + if (strcmp(name, bar->name) == 0) | |
3568 | + return bar; | |
3569 | + } | |
3570 | + return NULL; | |
3571 | +} | |
3572 | + | |
3573 | +/* Do the stuff we need to do when the barrier has completed phase 1 */ | |
3574 | +static void check_barrier_complete_phase1(struct cl_barrier *barrier) | |
3575 | +{ | |
3576 | + if (atomic_read(&barrier->got_nodes) == ((barrier->expected_nodes != 0) | |
3577 | + ? barrier->expected_nodes : | |
3578 | + cluster_members)) { | |
3579 | + | |
3580 | + struct cl_barriermsg bmsg; | |
3581 | + | |
3582 | + atomic_inc(&barrier->completed_nodes); /* We have completed */ | |
3583 | + barrier->phase = 2; /* Wait for complete phase II */ | |
3584 | + | |
3585 | + /* Send completion message, remember: we are in cnxman context | |
3586 | + * and must not block */ | |
3587 | + bmsg.cmd = CLUSTER_CMD_BARRIER; | |
3588 | + bmsg.subcmd = BARRIER_COMPLETE; | |
3589 | + bmsg.flags = 0; | |
3590 | + strcpy(bmsg.name, barrier->name); | |
3591 | + | |
3592 | + P_BARRIER("Sending COMPLETE for %s\n", barrier->name); | |
3593 | + queue_message((char *) &bmsg, sizeof (bmsg), NULL, 0, 0); | |
3594 | + } | |
3595 | +} | |
3596 | + | |
3597 | +/* Do the stuff we need to do when the barrier has been reached */ | |
3598 | +/* Return 1 if we deleted the barrier */ | |
3599 | +static int check_barrier_complete_phase2(struct cl_barrier *barrier, int status) | |
3600 | +{ | |
3601 | + spin_lock_irq(&barrier->phase2_spinlock); | |
3602 | + | |
3603 | + if (barrier->state != BARRIER_STATE_COMPLETE && | |
3604 | + (status == -ETIMEDOUT || | |
3605 | + atomic_read(&barrier->completed_nodes) == | |
3606 | + ((barrier->expected_nodes != 0) | |
3607 | + ? barrier->expected_nodes : cluster_members))) { | |
3608 | + | |
3609 | + if (status == 0 && barrier->timeout) | |
3610 | + del_timer(&barrier->timer); | |
3611 | + barrier->endreason = status; | |
3612 | + | |
3613 | + /* Wake up listener */ | |
3614 | + if (barrier->state == BARRIER_STATE_WAITING) { | |
3615 | + wake_up_interruptible(&barrier->waitq); | |
3616 | + } | |
3617 | + else { | |
3618 | + /* Additional tasks we have to do if the user was not | |
3619 | + * waiting... */ | |
3620 | + /* Call the callback */ | |
3621 | + if (barrier->callback) { | |
3622 | + barrier->callback(barrier->name, 0); | |
3623 | + barrier->callback = NULL; | |
3624 | + } | |
3625 | + /* Remove it if it's AUTO-DELETE */ | |
3626 | + if (barrier->flags & BARRIER_ATTR_AUTODELETE) { | |
3627 | + list_del(&barrier->list); | |
3628 | + spin_unlock_irq(&barrier->phase2_spinlock); | |
3629 | + kfree(barrier); | |
3630 | + return 1; | |
3631 | + } | |
3632 | + } | |
3633 | + barrier->state = BARRIER_STATE_COMPLETE; | |
3634 | + } | |
3635 | + spin_unlock_irq(&barrier->phase2_spinlock); | |
3636 | + return 0; | |
3637 | +} | |
3638 | + | |
3639 | +/* Called if a barrier timeout happens */ | |
3640 | +static void barrier_timer_fn(unsigned long arg) | |
3641 | +{ | |
3642 | + struct cl_barrier *barrier = (struct cl_barrier *) arg; | |
3643 | + | |
3644 | + /* Ignore any futher messages, they are too late. */ | |
3645 | + barrier->phase = 0; | |
3646 | + | |
3647 | + /* and cause it to timeout */ | |
3648 | + check_barrier_complete_phase2(barrier, -ETIMEDOUT); | |
3649 | +} | |
3650 | + | |
3651 | +/* Process BARRIER messages from other nodes */ | |
3652 | +static void process_barrier_msg(struct cl_barriermsg *msg, | |
3653 | + struct cluster_node *node) | |
3654 | +{ | |
3655 | + struct cl_barrier *barrier; | |
3656 | + | |
3657 | + down(&barrier_list_lock); | |
3658 | + barrier = find_barrier(msg->name); | |
3659 | + up(&barrier_list_lock); | |
3660 | + | |
3661 | + /* Ignore other peoples messages, in_transition() is needed here so | |
3662 | + * that joining nodes will see their barrier messages before the | |
3663 | + * we_are_a_cluster_member is set */ | |
3664 | + if (!we_are_a_cluster_member && !in_transition()) | |
3665 | + return; | |
3666 | + if (!barrier) | |
3667 | + return; | |
3668 | + | |
3669 | + P_BARRIER("Got %d for %s, from node %s\n", msg->subcmd, msg->name, | |
3670 | + node ? node->name : "unknown"); | |
3671 | + | |
3672 | + switch (msg->subcmd) { | |
3673 | + case BARRIER_WAIT: | |
3674 | + down(&barrier->lock); | |
3675 | + if (barrier->phase == 0) | |
3676 | + barrier->phase = 1; | |
3677 | + | |
3678 | + if (barrier->phase == 1) { | |
3679 | + atomic_inc(&barrier->got_nodes); | |
3680 | + check_barrier_complete_phase1(barrier); | |
3681 | + } | |
3682 | + else { | |
3683 | + printk(KERN_WARNING CMAN_NAME | |
3684 | + ": got WAIT barrier not in phase 1 %s (%d)\n", | |
3685 | + msg->name, barrier->phase); | |
3686 | + | |
3687 | + } | |
3688 | + up(&barrier->lock); | |
3689 | + break; | |
3690 | + | |
3691 | + case BARRIER_COMPLETE: | |
3692 | + down(&barrier->lock); | |
3693 | + atomic_inc(&barrier->completed_nodes); | |
3694 | + | |
3695 | + /* First node to get all the WAIT messages sends COMPLETE, so | |
3696 | + * we all complete */ | |
3697 | + if (barrier->phase == 1) { | |
3698 | + atomic_set(&barrier->got_nodes, | |
3699 | + barrier->expected_nodes); | |
3700 | + check_barrier_complete_phase1(barrier); | |
3701 | + } | |
3702 | + | |
3703 | + if (barrier->phase == 2) { | |
3704 | + /* If it was deleted (ret==1) then no need to unlock | |
3705 | + * the mutex */ | |
3706 | + if (check_barrier_complete_phase2(barrier, 0) == 1) | |
3707 | + return; | |
3708 | + } | |
3709 | + up(&barrier->lock); | |
3710 | + break; | |
3711 | + } | |
3712 | +} | |
3713 | + | |
3714 | +/* In-kernel membership API */ | |
3715 | +int kcl_add_callback(void (*callback) (kcl_callback_reason, long arg)) | |
3716 | +{ | |
3717 | + struct kernel_notify_struct *notify; | |
3718 | + | |
3719 | + notify = kmalloc(sizeof (struct kernel_notify_struct), GFP_KERNEL); | |
3720 | + if (!notify) | |
3721 | + return -ENOMEM; | |
3722 | + notify->callback = callback; | |
3723 | + | |
3724 | + down(&kernel_listener_lock); | |
3725 | + list_add(¬ify->list, &kernel_listener_list); | |
3726 | + up(&kernel_listener_lock); | |
3727 | + | |
3728 | + return 0; | |
3729 | +} | |
3730 | + | |
3731 | +int kcl_remove_callback(void (*callback) (kcl_callback_reason, long arg)) | |
3732 | +{ | |
3733 | + struct list_head *calllist; | |
3734 | + struct list_head *temp; | |
3735 | + struct kernel_notify_struct *notify; | |
3736 | + | |
3737 | + down(&kernel_listener_lock); | |
3738 | + list_for_each_safe(calllist, temp, &kernel_listener_list) { | |
3739 | + notify = list_entry(calllist, struct kernel_notify_struct, list); | |
3740 | + if (notify->callback == callback){ | |
3741 | + list_del(¬ify->list); | |
3742 | + kfree(notify); | |
3743 | + up(&kernel_listener_lock); | |
3744 | + return 0; | |
3745 | + } | |
3746 | + } | |
3747 | + up(&kernel_listener_lock); | |
3748 | + return -EINVAL; | |
3749 | +} | |
3750 | + | |
3751 | +/* Return quorate status */ | |
3752 | +int kcl_is_quorate() | |
3753 | +{ | |
3754 | + return cluster_is_quorate; | |
3755 | +} | |
3756 | + | |
3757 | +/* Return the address list for a node */ | |
3758 | +struct list_head *kcl_get_node_addresses(int nodeid) | |
3759 | +{ | |
3760 | + struct cluster_node *node = find_node_by_nodeid(nodeid); | |
3761 | + | |
3762 | + if (node) | |
3763 | + return &node->addr_list; | |
3764 | + else | |
3765 | + return NULL; | |
3766 | +} | |
3767 | + | |
3768 | +static void copy_to_kclnode(struct cluster_node *node, | |
3769 | + struct kcl_cluster_node *knode) | |
3770 | +{ | |
3771 | + strcpy(knode->name, node->name); | |
3772 | + knode->size = sizeof (struct kcl_cluster_node); | |
3773 | + knode->votes = node->votes; | |
3774 | + knode->state = node->state; | |
3775 | + knode->node_id = node->node_id; | |
3776 | + knode->us = node->us; | |
3777 | + knode->leave_reason = node->leave_reason; | |
3778 | + knode->incarnation = node->incarnation; | |
3779 | +} | |
3780 | + | |
3781 | +/* Return the info for a node given it's address. if addr is NULL then return | |
3782 | + * OUR info */ | |
3783 | +int kcl_get_node_by_addr(unsigned char *addr, int addr_len, | |
3784 | + struct kcl_cluster_node *n) | |
3785 | +{ | |
3786 | + struct cluster_node *node; | |
3787 | + | |
3788 | + /* They want us */ | |
3789 | + if (addr == NULL) { | |
3790 | + node = us; | |
3791 | + } | |
3792 | + else { | |
3793 | + node = find_node_by_addr(addr, addr_len); | |
3794 | + if (!node) | |
3795 | + return -1; | |
3796 | + } | |
3797 | + | |
3798 | + /* Copy to user's buffer */ | |
3799 | + copy_to_kclnode(node, n); | |
3800 | + return 0; | |
3801 | +} | |
3802 | + | |
3803 | +int kcl_get_node_by_name(unsigned char *name, struct kcl_cluster_node *n) | |
3804 | +{ | |
3805 | + struct cluster_node *node; | |
3806 | + | |
3807 | + /* They want us */ | |
3808 | + if (name == NULL) { | |
3809 | + node = us; | |
3810 | + if (node == NULL) | |
3811 | + return -1; | |
3812 | + } | |
3813 | + else { | |
3814 | + node = find_node_by_name(name); | |
3815 | + if (!node) | |
3816 | + return -1; | |
3817 | + } | |
3818 | + | |
3819 | + /* Copy to user's buffer */ | |
3820 | + copy_to_kclnode(node, n); | |
3821 | + return 0; | |
3822 | +} | |
3823 | + | |
3824 | +/* As above but by node id. MUCH faster */ | |
3825 | +int kcl_get_node_by_nodeid(int nodeid, struct kcl_cluster_node *n) | |
3826 | +{ | |
3827 | + struct cluster_node *node; | |
3828 | + | |
3829 | + /* They want us */ | |
3830 | + if (nodeid == 0) { | |
3831 | + node = us; | |
3832 | + if (node == NULL) | |
3833 | + return -1; | |
3834 | + } | |
3835 | + else { | |
3836 | + node = find_node_by_nodeid(nodeid); | |
3837 | + if (!node) | |
3838 | + return -1; | |
3839 | + } | |
3840 | + | |
3841 | + /* Copy to user's buffer */ | |
3842 | + copy_to_kclnode(node, n); | |
3843 | + return 0; | |
3844 | +} | |
3845 | + | |
3846 | +/* Return a list of all cluster members ever */ | |
3847 | +int kcl_get_all_members(struct list_head *list) | |
3848 | +{ | |
3849 | + struct list_head *nodelist; | |
3850 | + struct cluster_node *node; | |
3851 | + struct kcl_cluster_node *newnode; | |
3852 | + int num_nodes = 0; | |
3853 | + | |
3854 | + down(&cluster_members_lock); | |
3855 | + list_for_each(nodelist, &cluster_members_list) { | |
3856 | + if (list) { | |
3857 | + node = list_entry(nodelist, struct cluster_node, list); | |
3858 | + newnode = | |
3859 | + kmalloc(sizeof (struct kcl_cluster_node), | |
3860 | + GFP_KERNEL); | |
3861 | + if (newnode) { | |
3862 | + copy_to_kclnode(node, newnode); | |
3863 | + list_add(&newnode->list, list); | |
3864 | + num_nodes++; | |
3865 | + } | |
3866 | + } | |
3867 | + else { | |
3868 | + num_nodes++; | |
3869 | + } | |
3870 | + } | |
3871 | + up(&cluster_members_lock); | |
3872 | + | |
3873 | + return num_nodes; | |
3874 | +} | |
3875 | + | |
3876 | +/* Return a list of cluster members */ | |
3877 | +int kcl_get_members(struct list_head *list) | |
3878 | +{ | |
3879 | + struct list_head *nodelist; | |
3880 | + struct cluster_node *node; | |
3881 | + struct kcl_cluster_node *newnode; | |
3882 | + int num_nodes = 0; | |
3883 | + | |
3884 | + down(&cluster_members_lock); | |
3885 | + list_for_each(nodelist, &cluster_members_list) { | |
3886 | + node = list_entry(nodelist, struct cluster_node, list); | |
3887 | + | |
3888 | + if (node->state == NODESTATE_MEMBER) { | |
3889 | + if (list) { | |
3890 | + newnode = | |
3891 | + kmalloc(sizeof (struct kcl_cluster_node), | |
3892 | + GFP_KERNEL); | |
3893 | + if (newnode) { | |
3894 | + copy_to_kclnode(node, newnode); | |
3895 | + list_add(&newnode->list, list); | |
3896 | + num_nodes++; | |
3897 | + } | |
3898 | + } | |
3899 | + else { | |
3900 | + num_nodes++; | |
3901 | + } | |
3902 | + } | |
3903 | + } | |
3904 | + up(&cluster_members_lock); | |
3905 | + | |
3906 | + return num_nodes; | |
3907 | +} | |
3908 | + | |
3909 | +/* Copy current member's nodeids into buffer */ | |
3910 | +int kcl_get_member_ids(uint32_t *idbuf, int size) | |
3911 | +{ | |
3912 | + struct list_head *nodelist; | |
3913 | + struct cluster_node *node; | |
3914 | + int num_nodes = 0; | |
3915 | + | |
3916 | + down(&cluster_members_lock); | |
3917 | + list_for_each(nodelist, &cluster_members_list) { | |
3918 | + node = list_entry(nodelist, struct cluster_node, list); | |
3919 | + | |
3920 | + if (node->state == NODESTATE_MEMBER) { | |
3921 | + if (idbuf && size) { | |
3922 | + idbuf[num_nodes] = node->node_id; | |
3923 | + num_nodes++; | |
3924 | + size--; | |
3925 | + } | |
3926 | + else { | |
3927 | + num_nodes++; | |
3928 | + } | |
3929 | + } | |
3930 | + } | |
3931 | + up(&cluster_members_lock); | |
3932 | + | |
3933 | + return num_nodes; | |
3934 | +} | |
3935 | + | |
3936 | +/* Barrier API */ | |
3937 | +int kcl_barrier_register(char *name, unsigned int flags, unsigned int nodes) | |
3938 | +{ | |
3939 | + struct cl_barrier *barrier; | |
3940 | + | |
3941 | + /* We are not joined to a cluster */ | |
3942 | + if (!we_are_a_cluster_member) | |
3943 | + return -ENOTCONN; | |
3944 | + | |
3945 | + /* Must have a valid name */ | |
3946 | + if (name == NULL || strlen(name) > MAX_BARRIER_NAME_LEN - 1) | |
3947 | + return -EINVAL; | |
3948 | + | |
3949 | + /* We don't do this yet */ | |
3950 | + if (flags & BARRIER_ATTR_MULTISTEP) | |
3951 | + return -ENOTSUPP; | |
3952 | + | |
3953 | + down(&barrier_list_lock); | |
3954 | + | |
3955 | + /* See if it already exists */ | |
3956 | + if ((barrier = find_barrier(name))) { | |
3957 | + up(&barrier_list_lock); | |
3958 | + if (nodes != barrier->expected_nodes) { | |
3959 | + printk(KERN_WARNING CMAN_NAME | |
3960 | + ": Barrier registration failed for '%s', expected nodes=%d, requested=%d\n", | |
3961 | + name, barrier->expected_nodes, nodes); | |
3962 | + up(&barrier_list_lock); | |
3963 | + return -EINVAL; | |
3964 | + } | |
3965 | + else | |
3966 | + return 0; | |
3967 | + } | |
3968 | + | |
3969 | + /* Build a new struct and add it to the list */ | |
3970 | + barrier = kmalloc(sizeof (struct cl_barrier), GFP_KERNEL); | |
3971 | + if (barrier == NULL) { | |
3972 | + up(&barrier_list_lock); | |
3973 | + return -ENOMEM; | |
3974 | + } | |
3975 | + memset(barrier, 0, sizeof (*barrier)); | |
3976 | + | |
3977 | + strcpy(barrier->name, name); | |
3978 | + barrier->flags = flags; | |
3979 | + barrier->expected_nodes = nodes; | |
3980 | + atomic_set(&barrier->got_nodes, 0); | |
3981 | + atomic_set(&barrier->completed_nodes, 0); | |
3982 | + barrier->endreason = 0; | |
3983 | + barrier->registered_nodes = 1; | |
3984 | + spin_lock_init(&barrier->phase2_spinlock); | |
3985 | + barrier->state = BARRIER_STATE_INACTIVE; | |
3986 | + init_MUTEX(&barrier->lock); | |
3987 | + | |
3988 | + list_add(&barrier->list, &barrier_list); | |
3989 | + up(&barrier_list_lock); | |
3990 | + | |
3991 | + return 0; | |
3992 | +} | |
3993 | + | |
3994 | +static int barrier_setattr_enabled(struct cl_barrier *barrier, | |
3995 | + unsigned int attr, unsigned long arg) | |
3996 | +{ | |
3997 | + int status; | |
3998 | + | |
3999 | + /* Can't disable a barrier */ | |
4000 | + if (!arg) { | |
4001 | + up(&barrier->lock); | |
4002 | + return -EINVAL; | |
4003 | + } | |
4004 | + | |
4005 | + /* We need to send WAIT now because the user may not | |
4006 | + * actually call kcl_barrier_wait() */ | |
4007 | + if (!barrier->waitsent) { | |
4008 | + struct cl_barriermsg bmsg; | |
4009 | + | |
4010 | + /* Send it to the rest of the cluster */ | |
4011 | + bmsg.cmd = CLUSTER_CMD_BARRIER; | |
4012 | + bmsg.subcmd = BARRIER_WAIT; | |
4013 | + strcpy(bmsg.name, barrier->name); | |
4014 | + | |
4015 | + barrier->waitsent = 1; | |
4016 | + barrier->phase = 1; | |
4017 | + | |
4018 | + atomic_inc(&barrier->got_nodes); | |
4019 | + | |
4020 | + /* Start the timer if one was wanted */ | |
4021 | + if (barrier->timeout) { | |
4022 | + init_timer(&barrier->timer); | |
4023 | + barrier->timer.function = barrier_timer_fn; | |
4024 | + barrier->timer.data = (long) barrier; | |
4025 | + mod_timer(&barrier->timer, jiffies + (barrier->timeout * HZ)); | |
4026 | + } | |
4027 | + | |
4028 | + /* Barrier WAIT and COMPLETE messages are | |
4029 | + * always queued - that way they always get | |
4030 | + * sent out in the right order. If we don't do | |
4031 | + * this then one can get sent out in the | |
4032 | + * context of the user process and the other in | |
4033 | + * cnxman and COMPLETE may /just/ slide in | |
4034 | + * before WAIT if its in the queue | |
4035 | + */ | |
4036 | + P_BARRIER("Sending WAIT for %s\n", name); | |
4037 | + status = queue_message(&bmsg, sizeof (bmsg), NULL, 0, 0); | |
4038 | + if (status < 0) { | |
4039 | + up(&barrier->lock); | |
4040 | + return status; | |
4041 | + } | |
4042 | + | |
4043 | + /* It might have been reached now */ | |
4044 | + if (barrier | |
4045 | + && barrier->state != BARRIER_STATE_COMPLETE | |
4046 | + && barrier->phase == 1) | |
4047 | + check_barrier_complete_phase1(barrier); | |
4048 | + } | |
4049 | + if (barrier && barrier->state == BARRIER_STATE_COMPLETE) { | |
4050 | + up(&barrier->lock); | |
4051 | + return barrier->endreason; | |
4052 | + } | |
4053 | + up(&barrier->lock); | |
4054 | + return 0; /* Nothing to propogate */ | |
4055 | +} | |
4056 | + | |
4057 | +int kcl_barrier_setattr(char *name, unsigned int attr, unsigned long arg) | |
4058 | +{ | |
4059 | + struct cl_barrier *barrier; | |
4060 | + | |
4061 | + /* See if it already exists */ | |
4062 | + down(&barrier_list_lock); | |
4063 | + if (!(barrier = find_barrier(name))) { | |
4064 | + up(&barrier_list_lock); | |
4065 | + return -ENOENT; | |
4066 | + } | |
4067 | + up(&barrier_list_lock); | |
4068 | + | |
4069 | + down(&barrier->lock); | |
4070 | + if (barrier->state == BARRIER_STATE_COMPLETE) { | |
4071 | + up(&barrier->lock); | |
4072 | + return 0; | |
4073 | + } | |
4074 | + | |
4075 | + switch (attr) { | |
4076 | + case BARRIER_SETATTR_AUTODELETE: | |
4077 | + if (arg) | |
4078 | + barrier->flags |= BARRIER_ATTR_AUTODELETE; | |
4079 | + else | |
4080 | + barrier->flags &= ~BARRIER_ATTR_AUTODELETE; | |
4081 | + up(&barrier->lock); | |
4082 | + return 0; | |
4083 | + break; | |
4084 | + | |
4085 | + case BARRIER_SETATTR_TIMEOUT: | |
4086 | + /* Can only change the timout of an inactive barrier */ | |
4087 | + if (barrier->state == BARRIER_STATE_WAITING | |
4088 | + || barrier->waitsent) { | |
4089 | + up(&barrier->lock); | |
4090 | + return -EINVAL; | |
4091 | + } | |
4092 | + barrier->timeout = arg; | |
4093 | + up(&barrier->lock); | |
4094 | + return 0; | |
4095 | + | |
4096 | + case BARRIER_SETATTR_MULTISTEP: | |
4097 | + up(&barrier->lock); | |
4098 | + return -ENOTSUPP; | |
4099 | + | |
4100 | + case BARRIER_SETATTR_ENABLED: | |
4101 | + return barrier_setattr_enabled(barrier, attr, arg); | |
4102 | + | |
4103 | + case BARRIER_SETATTR_NODES: | |
4104 | + /* Can only change the expected node count of an inactive | |
4105 | + * barrier */ | |
4106 | + if (barrier->state == BARRIER_STATE_WAITING | |
4107 | + || barrier->waitsent) | |
4108 | + return -EINVAL; | |
4109 | + barrier->expected_nodes = arg; | |
4110 | + break; | |
4111 | + | |
4112 | + case BARRIER_SETATTR_CALLBACK: | |
4113 | + if (barrier->state == BARRIER_STATE_WAITING | |
4114 | + || barrier->waitsent) | |
4115 | + return -EINVAL; | |
4116 | + barrier->callback = (void (*)(char *, int)) arg; | |
4117 | + up(&barrier->lock); | |
4118 | + return 0; /* Don't propgate this to other nodes */ | |
4119 | + } | |
4120 | + | |
4121 | + up(&barrier->lock); | |
4122 | + return 0; | |
4123 | +} | |
4124 | + | |
4125 | +int kcl_barrier_delete(char *name) | |
4126 | +{ | |
4127 | + struct cl_barrier *barrier; | |
4128 | + | |
4129 | + down(&barrier_list_lock); | |
4130 | + /* See if it exists */ | |
4131 | + if (!(barrier = find_barrier(name))) { | |
4132 | + up(&barrier_list_lock); | |
4133 | + return -ENOENT; | |
4134 | + } | |
4135 | + | |
4136 | + /* Delete it */ | |
4137 | + list_del(&barrier->list); | |
4138 | + kfree(barrier); | |
4139 | + | |
4140 | + up(&barrier_list_lock); | |
4141 | + | |
4142 | + return 0; | |
4143 | +} | |
4144 | + | |
4145 | +int kcl_barrier_cancel(char *name) | |
4146 | +{ | |
4147 | + struct cl_barrier *barrier; | |
4148 | + | |
4149 | + /* See if it exists */ | |
4150 | + down(&barrier_list_lock); | |
4151 | + if (!(barrier = find_barrier(name))) { | |
4152 | + up(&barrier_list_lock); | |
4153 | + return -ENOENT; | |
4154 | + } | |
4155 | + down(&barrier->lock); | |
4156 | + | |
4157 | + barrier->endreason = -ENOTCONN; | |
4158 | + | |
4159 | + if (barrier->callback) { | |
4160 | + barrier->callback(barrier->name, -ECONNRESET); | |
4161 | + barrier->callback = NULL; | |
4162 | + } | |
4163 | + | |
4164 | + if (barrier->timeout) | |
4165 | + del_timer(&barrier->timer); | |
4166 | + | |
4167 | + /* Remove it if it's AUTO-DELETE */ | |
4168 | + if (barrier->flags & BARRIER_ATTR_AUTODELETE) { | |
4169 | + list_del(&barrier->list); | |
4170 | + up(&barrier->lock); | |
4171 | + kfree(barrier); | |
4172 | + up(&barrier_list_lock); | |
4173 | + return 0; | |
4174 | + } | |
4175 | + | |
4176 | + if (barrier->state == BARRIER_STATE_WAITING) | |
4177 | + wake_up_interruptible(&barrier->waitq); | |
4178 | + | |
4179 | + up(&barrier->lock); | |
4180 | + up(&barrier_list_lock); | |
4181 | + return 0; | |
4182 | +} | |
4183 | + | |
4184 | +int kcl_barrier_wait(char *name) | |
4185 | +{ | |
4186 | + struct cl_barrier *barrier; | |
4187 | + int ret; | |
4188 | + | |
4189 | + if (!atomic_read(&cnxman_running)) | |
4190 | + return -ENOTCONN; | |
4191 | + | |
4192 | + /* Enable it */ | |
4193 | + kcl_barrier_setattr(name, BARRIER_SETATTR_ENABLED, 1L); | |
4194 | + | |
4195 | + down(&barrier_list_lock); | |
4196 | + | |
4197 | + /* See if it still exists - enable may have deleted it! */ | |
4198 | + if (!(barrier = find_barrier(name))) { | |
4199 | + up(&barrier_list_lock); | |
4200 | + return -ENOENT; | |
4201 | + } | |
4202 | + | |
4203 | + down(&barrier->lock); | |
4204 | + | |
4205 | + up(&barrier_list_lock); | |
4206 | + | |
4207 | + /* If it has already completed then return the status */ | |
4208 | + if (barrier->state == BARRIER_STATE_COMPLETE) { | |
4209 | + up(&barrier->lock); | |
4210 | + return barrier->endreason; | |
4211 | + } | |
4212 | + | |
4213 | + barrier->state = BARRIER_STATE_WAITING; | |
4214 | + | |
4215 | + /* Have we all reached the barrier? */ | |
4216 | + while (atomic_read(&barrier->completed_nodes) != | |
4217 | + ((barrier->expected_nodes == 0) | |
4218 | + ? cluster_members : barrier->expected_nodes) | |
4219 | + && barrier->endreason == 0) { | |
4220 | + | |
4221 | + wait_queue_t wq; | |
4222 | + | |
4223 | + init_waitqueue_entry(&wq, current); | |
4224 | + init_waitqueue_head(&barrier->waitq); | |
4225 | + | |
4226 | + /* Wait for em all */ | |
4227 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
4228 | + add_wait_queue(&barrier->waitq, &wq); | |
4229 | + | |
4230 | + if (atomic_read(&barrier->completed_nodes) != | |
4231 | + ((barrier->expected_nodes == | |
4232 | + 0) ? cluster_members : barrier->expected_nodes) | |
4233 | + && barrier->endreason == 0) { | |
4234 | + up(&barrier->lock); | |
4235 | + schedule(); | |
4236 | + down(&barrier->lock); | |
4237 | + } | |
4238 | + | |
4239 | + remove_wait_queue(&barrier->waitq, &wq); | |
4240 | + set_task_state(current, TASK_RUNNING); | |
4241 | + | |
4242 | + if (signal_pending(current)) { | |
4243 | + barrier->endreason = -EINTR; | |
4244 | + break; | |
4245 | + } | |
4246 | + } | |
4247 | + barrier->state = BARRIER_STATE_INACTIVE; | |
4248 | + | |
4249 | + if (barrier->timeout) | |
4250 | + del_timer(&barrier->timer); | |
4251 | + | |
4252 | + /* Barrier has been reached on all nodes, call the callback */ | |
4253 | + if (barrier->callback) { | |
4254 | + barrier->callback(barrier->name, barrier->endreason); | |
4255 | + barrier->callback = NULL; | |
4256 | + } | |
4257 | + | |
4258 | + atomic_set(&barrier->got_nodes, 0); | |
4259 | + | |
4260 | + /* Return the reason we were woken */ | |
4261 | + ret = barrier->endreason; | |
4262 | + | |
4263 | + /* Remove it if it's AUTO-DELETE */ | |
4264 | + if (barrier->flags & BARRIER_ATTR_AUTODELETE) { | |
4265 | + down(&barrier_list_lock); | |
4266 | + list_del(&barrier->list); | |
4267 | + up(&barrier_list_lock); | |
4268 | + up(&barrier->lock); | |
4269 | + kfree(barrier); | |
4270 | + } | |
4271 | + else { | |
4272 | + up(&barrier->lock); | |
4273 | + } | |
4274 | + | |
4275 | + /* We were woken up because the node left the cluster ? */ | |
4276 | + if (!atomic_read(&cnxman_running)) | |
4277 | + ret = -ENOTCONN; | |
4278 | + | |
4279 | + return ret; | |
4280 | +} | |
4281 | + | |
4282 | +/* This is called from membership services when a node has left the cluster - | |
4283 | + * we signal all waiting barriers with -ESRCH so they know to do something | |
4284 | + * else, if the number of nodes is left at 0 then we compare the new number of | |
4285 | + * nodes in the cluster with that at the barrier and return 0 (success) in that | |
4286 | + * case */ | |
4287 | +void check_barrier_returns() | |
4288 | +{ | |
4289 | + struct list_head *blist; | |
4290 | + struct list_head *llist; | |
4291 | + struct cl_barrier *barrier; | |
4292 | + int status = 0; | |
4293 | + | |
4294 | + down(&barrier_list_lock); | |
4295 | + list_for_each(blist, &barrier_list) { | |
4296 | + barrier = list_entry(blist, struct cl_barrier, list); | |
4297 | + | |
4298 | + if (barrier->waitsent) { | |
4299 | + int wakeit = 0; | |
4300 | + | |
4301 | + /* Check for a dynamic member barrier */ | |
4302 | + if (barrier->expected_nodes == 0) { | |
4303 | + if (barrier->registered_nodes == | |
4304 | + cluster_members) { | |
4305 | + status = 0; | |
4306 | + wakeit = 1; | |
4307 | + } | |
4308 | + } | |
4309 | + else { | |
4310 | + status = -ESRCH; | |
4311 | + wakeit = 1; | |
4312 | + } | |
4313 | + | |
4314 | + /* Do we need to tell the barrier? */ | |
4315 | + if (wakeit) { | |
4316 | + if (barrier->state == BARRIER_STATE_WAITING) { | |
4317 | + barrier->endreason = status; | |
4318 | + wake_up_interruptible(&barrier->waitq); | |
4319 | + } | |
4320 | + else { | |
4321 | + if (barrier->callback) { | |
4322 | + barrier->callback(barrier->name, | |
4323 | + status); | |
4324 | + } | |
4325 | + } | |
4326 | + } | |
4327 | + } | |
4328 | + } | |
4329 | + up(&barrier_list_lock); | |
4330 | + | |
4331 | + /* Part 2 check for outstanding listen requests for dead nodes and | |
4332 | + * cancel them */ | |
4333 | + down(&listenreq_lock); | |
4334 | + list_for_each(llist, &listenreq_list) { | |
4335 | + struct cl_waiting_listen_request *lrequest = | |
4336 | + list_entry(llist, struct cl_waiting_listen_request, list); | |
4337 | + struct cluster_node *node = | |
4338 | + find_node_by_nodeid(lrequest->nodeid); | |
4339 | + | |
4340 | + if (node && node->state != NODESTATE_MEMBER) { | |
4341 | + lrequest->result = -ENOTCONN; | |
4342 | + lrequest->waiting = 0; | |
4343 | + wake_up_interruptible(&lrequest->waitq); | |
4344 | + } | |
4345 | + } | |
4346 | + up(&listenreq_lock); | |
4347 | +} | |
4348 | + | |
4349 | +int get_addr_from_temp_nodeid(int nodeid, char *addr, int *addrlen) | |
4350 | +{ | |
4351 | + struct temp_node *tn; | |
4352 | + int err = 1; /* true */ | |
4353 | +#ifdef DEBUG_COMMS | |
4354 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
4355 | +#endif | |
4356 | + | |
4357 | + down(&tempnode_lock); | |
4358 | + | |
4359 | + list_for_each_entry(tn, &tempnode_list, list) { | |
4360 | + if (tn->nodeid == nodeid) { | |
4361 | + memcpy(addr, tn->addr, tn->addrlen); | |
4362 | + *addrlen = tn->addrlen; | |
4363 | + P_COMMS("get_temp_nodeid. id %d:\n: %s\n", | |
4364 | + tn->nodeid, print_addr(tn->addr, tn->addrlen, buf)); | |
4365 | + | |
4366 | + goto out; | |
4367 | + } | |
4368 | + } | |
4369 | + err = 0; | |
4370 | + | |
4371 | + out: | |
4372 | + up(&tempnode_lock); | |
4373 | + return err; | |
4374 | +} | |
4375 | + | |
4376 | +/* Create a new temporary node ID. This list will only ever be very small | |
4377 | + (usaully only 1 item) but I can't take the risk that someone won't try to | |
4378 | + boot 128 nodes all at exactly the same time. */ | |
4379 | +int new_temp_nodeid(char *addr, int addrlen) | |
4380 | +{ | |
4381 | + struct temp_node *tn; | |
4382 | + int err = -1; | |
4383 | + int try_nodeid = 0; | |
4384 | +#ifdef DEBUG_COMMS | |
4385 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
4386 | +#endif | |
4387 | + | |
4388 | + P_COMMS("new_temp_nodeid needed for\n: %s\n", | |
4389 | + print_addr(addr, addrlen, buf)); | |
4390 | + | |
4391 | + down(&tempnode_lock); | |
4392 | + | |
4393 | + /* First see if we already know about this node */ | |
4394 | + list_for_each_entry(tn, &tempnode_list, list) { | |
4395 | + | |
4396 | + P_COMMS("new_temp_nodeid list. id %d:\n: %s\n", | |
4397 | + tn->nodeid, print_addr(tn->addr, tn->addrlen, buf)); | |
4398 | + | |
4399 | + /* We're already in here... */ | |
4400 | + if (tn->addrlen == addrlen && | |
4401 | + memcmp(tn->addr, addr, addrlen) == 0) { | |
4402 | + P_COMMS("reused temp node ID %d\n", tn->nodeid); | |
4403 | + err = tn->nodeid; | |
4404 | + goto out; | |
4405 | + } | |
4406 | + } | |
4407 | + | |
4408 | + /* Nope, OK, invent a suitable number */ | |
4409 | + retry: | |
4410 | + try_nodeid -= 1; | |
4411 | + list_for_each_entry(tn, &tempnode_list, list) { | |
4412 | + | |
4413 | + if (tn->nodeid == try_nodeid) | |
4414 | + goto retry; | |
4415 | + } | |
4416 | + | |
4417 | + tn = kmalloc(sizeof(struct temp_node), GFP_KERNEL); | |
4418 | + if (!tn) | |
4419 | + goto out; | |
4420 | + | |
4421 | + memcpy(tn->addr, addr, addrlen); | |
4422 | + tn->addrlen = addrlen; | |
4423 | + tn->nodeid = try_nodeid; | |
4424 | + list_add_tail(&tn->list, &tempnode_list); | |
4425 | + err = try_nodeid; | |
4426 | + P_COMMS("new temp nodeid = %d\n", try_nodeid); | |
4427 | + out: | |
4428 | + up(&tempnode_lock); | |
4429 | + return err; | |
4430 | +} | |
4431 | + | |
4432 | +static int is_valid_temp_nodeid(int nodeid) | |
4433 | +{ | |
4434 | + struct temp_node *tn; | |
4435 | + int err = 1; /* true */ | |
4436 | + | |
4437 | + down(&tempnode_lock); | |
4438 | + | |
4439 | + list_for_each_entry(tn, &tempnode_list, list) { | |
4440 | + if (tn->nodeid == nodeid) | |
4441 | + goto out; | |
4442 | + } | |
4443 | + err = 0; | |
4444 | + | |
4445 | + out: | |
4446 | + P_COMMS("is_valid_temp_nodeid. %d = %d\n", nodeid, err); | |
4447 | + up(&tempnode_lock); | |
4448 | + return err; | |
4449 | +} | |
4450 | + | |
4451 | +/* TODO: This needs to clean the list more fully of | |
4452 | + nodes that are now full members but we did not master the transition */ | |
4453 | +void remove_temp_nodeid(int nodeid) | |
4454 | +{ | |
4455 | + struct temp_node *tn; | |
4456 | + struct temp_node *tmp; | |
4457 | + | |
4458 | + down(&tempnode_lock); | |
4459 | + | |
4460 | + list_for_each_entry_safe(tn, tmp, &tempnode_list, list) { | |
4461 | + if (nodeid == tn->nodeid) { | |
4462 | + list_del(&tn->list); | |
4463 | + kfree(tn); | |
4464 | + up(&tempnode_lock); | |
4465 | + return; | |
4466 | + } | |
4467 | + } | |
4468 | + | |
4469 | + up(&tempnode_lock); | |
4470 | +} | |
4471 | + | |
4472 | +/* Quorum device functions */ | |
4473 | +int kcl_register_quorum_device(char *name, int votes) | |
4474 | +{ | |
4475 | + if (quorum_device) | |
4476 | + return -EBUSY; | |
4477 | + | |
4478 | + if (find_node_by_name(name)) | |
4479 | + return -EINVAL; | |
4480 | + | |
4481 | + quorum_device = kmalloc(sizeof (struct cluster_node), GFP_KERNEL); | |
4482 | + if (!quorum_device) | |
4483 | + return -ENOMEM; | |
4484 | + memset(quorum_device, 0, sizeof (struct cluster_node)); | |
4485 | + | |
4486 | + quorum_device->name = kmalloc(strlen(name) + 1, GFP_KERNEL); | |
4487 | + if (!quorum_device->name) { | |
4488 | + kfree(quorum_device); | |
4489 | + quorum_device = NULL; | |
4490 | + return -ENOMEM; | |
4491 | + } | |
4492 | + | |
4493 | + strcpy(quorum_device->name, name); | |
4494 | + quorum_device->votes = votes; | |
4495 | + quorum_device->state = NODESTATE_DEAD; | |
4496 | + | |
4497 | + /* Keep this list valid so it doesn't confuse other code */ | |
4498 | + INIT_LIST_HEAD(&quorum_device->addr_list); | |
4499 | + | |
4500 | + return 0; | |
4501 | +} | |
4502 | + | |
4503 | +int kcl_unregister_quorum_device(void) | |
4504 | +{ | |
4505 | + if (!quorum_device) | |
4506 | + return -EINVAL; | |
4507 | + if (quorum_device->state == NODESTATE_MEMBER) | |
4508 | + return -EINVAL; | |
4509 | + | |
4510 | + quorum_device = NULL; | |
4511 | + | |
4512 | + return 0; | |
4513 | +} | |
4514 | + | |
4515 | +int kcl_quorum_device_available(int yesno) | |
4516 | +{ | |
4517 | + if (!quorum_device) | |
4518 | + return -EINVAL; | |
4519 | + | |
4520 | + if (yesno) { | |
4521 | + quorum_device->last_hello = jiffies; | |
4522 | + if (quorum_device->state == NODESTATE_DEAD) { | |
4523 | + quorum_device->state = NODESTATE_MEMBER; | |
4524 | + recalculate_quorum(0); | |
4525 | + } | |
4526 | + } | |
4527 | + else { | |
4528 | + if (quorum_device->state == NODESTATE_MEMBER) { | |
4529 | + quorum_device->state = NODESTATE_DEAD; | |
4530 | + recalculate_quorum(0); | |
4531 | + } | |
4532 | + } | |
4533 | + | |
4534 | + return 0; | |
4535 | +} | |
4536 | + | |
4537 | +/* APIs for cluster ref counting. */ | |
4538 | +int kcl_addref_cluster() | |
4539 | +{ | |
4540 | + int ret = -ENOTCONN; | |
4541 | + | |
4542 | + if (!atomic_read(&cnxman_running)) | |
4543 | + goto addref_ret; | |
4544 | + | |
4545 | + if (try_module_get(THIS_MODULE)) { | |
4546 | + atomic_inc(&use_count); | |
4547 | + ret = 0; | |
4548 | + } | |
4549 | + | |
4550 | + addref_ret: | |
4551 | + return ret; | |
4552 | +} | |
4553 | + | |
4554 | +int kcl_releaseref_cluster() | |
4555 | +{ | |
4556 | + if (!atomic_read(&cnxman_running)) | |
4557 | + return -ENOTCONN; | |
4558 | + atomic_dec(&use_count); | |
4559 | + module_put(THIS_MODULE); | |
4560 | + return 0; | |
4561 | +} | |
4562 | + | |
4563 | +int kcl_cluster_name(char **cname) | |
4564 | +{ | |
4565 | + char *name; | |
4566 | + | |
4567 | + name = kmalloc(strlen(cluster_name) + 1, GFP_KERNEL); | |
4568 | + if (!name) | |
4569 | + return -ENOMEM; | |
4570 | + | |
4571 | + strncpy(name, cluster_name, strlen(cluster_name)+1); | |
4572 | + *cname = name; | |
4573 | + return 0; | |
4574 | +} | |
4575 | + | |
4576 | +int kcl_get_current_interface(void) | |
4577 | +{ | |
4578 | + return current_interface->number; | |
4579 | +} | |
4580 | + | |
4581 | +/* Socket registration stuff */ | |
4582 | +static struct net_proto_family cl_family_ops = { | |
4583 | + .family = AF_CLUSTER, | |
4584 | + .create = cl_create | |
4585 | +}; | |
4586 | + | |
4587 | +static struct proto_ops cl_proto_ops = { | |
4588 | + .family = AF_CLUSTER, | |
4589 | + | |
4590 | + .release = cl_release, | |
4591 | + .bind = cl_bind, | |
4592 | + .connect = sock_no_connect, | |
4593 | + .socketpair = sock_no_socketpair, | |
4594 | + .accept = sock_no_accept, | |
4595 | + .getname = cl_getname, | |
4596 | + .poll = cl_poll, | |
4597 | + .ioctl = cl_ioctl, | |
4598 | + .listen = sock_no_listen, | |
4599 | + .shutdown = cl_shutdown, | |
4600 | + .setsockopt = cl_setsockopt, | |
4601 | + .getsockopt = cl_getsockopt, | |
4602 | + .sendmsg = cl_sendmsg, | |
4603 | + .recvmsg = cl_recvmsg, | |
4604 | + .mmap = sock_no_mmap, | |
4605 | + .sendpage = sock_no_sendpage, | |
4606 | +}; | |
4607 | + | |
4608 | +#ifdef MODULE | |
4609 | +MODULE_DESCRIPTION("Cluster Connection and Service Manager"); | |
4610 | +MODULE_AUTHOR("Red Hat, Inc"); | |
4611 | +MODULE_LICENSE("GPL"); | |
4612 | +#endif | |
4613 | + | |
4614 | +static int __init cluster_init(void) | |
4615 | +{ | |
4616 | + printk("CMAN %s (built %s %s) installed\n", | |
4617 | + CMAN_RELEASE_NAME, __DATE__, __TIME__); | |
4618 | + | |
4619 | + /* allocate our sock slab cache */ | |
4620 | + cluster_sk_cachep = kmem_cache_create("cluster_sock", | |
4621 | + sizeof (struct cluster_sock), 0, | |
4622 | + SLAB_HWCACHE_ALIGN, 0, 0); | |
4623 | + if (!cluster_sk_cachep) { | |
4624 | + printk(KERN_CRIT | |
4625 | + "cluster_init: Cannot create cluster_sock SLAB cache\n"); | |
4626 | + return -1; | |
4627 | + | |
4628 | + } | |
4629 | + | |
4630 | + if (sock_register(&cl_family_ops)) { | |
4631 | + printk(KERN_INFO "Unable to register cluster socket type\n"); | |
4632 | + kmem_cache_destroy(cluster_sk_cachep); | |
4633 | + return -1; | |
4634 | + } | |
4635 | + | |
4636 | + | |
4637 | +#ifdef CONFIG_PROC_FS | |
4638 | + create_proc_entries(); | |
4639 | +#endif | |
4640 | + | |
4641 | + init_MUTEX(&start_thread_sem); | |
4642 | + init_MUTEX(&send_lock); | |
4643 | + init_MUTEX(&barrier_list_lock); | |
4644 | + init_MUTEX(&cluster_members_lock); | |
4645 | + init_MUTEX(&port_array_lock); | |
4646 | + init_MUTEX(&messages_list_lock); | |
4647 | + init_MUTEX(&listenreq_lock); | |
4648 | + init_MUTEX(&client_socket_lock); | |
4649 | + init_MUTEX(&new_dead_node_lock); | |
4650 | + init_MUTEX(&event_listener_lock); | |
4651 | + init_MUTEX(&kernel_listener_lock); | |
4652 | + init_MUTEX(&tempnode_lock); | |
4653 | + spin_lock_init(&active_socket_lock); | |
4654 | + init_timer(&ack_timer); | |
4655 | + | |
4656 | + INIT_LIST_HEAD(&event_listener_list); | |
4657 | + INIT_LIST_HEAD(&kernel_listener_list); | |
4658 | + INIT_LIST_HEAD(&socket_list); | |
4659 | + INIT_LIST_HEAD(&client_socket_list); | |
4660 | + INIT_LIST_HEAD(&active_socket_list); | |
4661 | + INIT_LIST_HEAD(&barrier_list); | |
4662 | + INIT_LIST_HEAD(&messages_list); | |
4663 | + INIT_LIST_HEAD(&listenreq_list); | |
4664 | + INIT_LIST_HEAD(&cluster_members_list); | |
4665 | + INIT_LIST_HEAD(&new_dead_node_list); | |
4666 | + INIT_LIST_HEAD(&tempnode_list); | |
4667 | + | |
4668 | + atomic_set(&cnxman_running, 0); | |
4669 | + | |
4670 | + sm_init(); | |
4671 | + | |
4672 | + return 0; | |
4673 | +} | |
4674 | + | |
4675 | +static void __exit cluster_exit(void) | |
4676 | +{ | |
4677 | +#ifdef CONFIG_PROC_FS | |
4678 | + cleanup_proc_entries(); | |
4679 | +#endif | |
4680 | + | |
4681 | + sock_unregister(AF_CLUSTER); | |
4682 | + kmem_cache_destroy(cluster_sk_cachep); | |
4683 | +} | |
4684 | + | |
4685 | +module_init(cluster_init); | |
4686 | +module_exit(cluster_exit); | |
4687 | + | |
4688 | +EXPORT_SYMBOL(kcl_sendmsg); | |
4689 | +EXPORT_SYMBOL(kcl_register_read_callback); | |
4690 | +EXPORT_SYMBOL(kcl_add_callback); | |
4691 | +EXPORT_SYMBOL(kcl_remove_callback); | |
4692 | +EXPORT_SYMBOL(kcl_get_members); | |
4693 | +EXPORT_SYMBOL(kcl_get_member_ids); | |
4694 | +EXPORT_SYMBOL(kcl_get_all_members); | |
4695 | +EXPORT_SYMBOL(kcl_is_quorate); | |
4696 | +EXPORT_SYMBOL(kcl_get_node_by_addr); | |
4697 | +EXPORT_SYMBOL(kcl_get_node_by_name); | |
4698 | +EXPORT_SYMBOL(kcl_get_node_by_nodeid); | |
4699 | +EXPORT_SYMBOL(kcl_get_node_addresses); | |
4700 | +EXPORT_SYMBOL(kcl_addref_cluster); | |
4701 | +EXPORT_SYMBOL(kcl_releaseref_cluster); | |
4702 | +EXPORT_SYMBOL(kcl_cluster_name); | |
4703 | + | |
4704 | +EXPORT_SYMBOL(kcl_barrier_register); | |
4705 | +EXPORT_SYMBOL(kcl_barrier_setattr); | |
4706 | +EXPORT_SYMBOL(kcl_barrier_delete); | |
4707 | +EXPORT_SYMBOL(kcl_barrier_wait); | |
4708 | +EXPORT_SYMBOL(kcl_barrier_cancel); | |
4709 | + | |
4710 | +EXPORT_SYMBOL(kcl_register_quorum_device); | |
4711 | +EXPORT_SYMBOL(kcl_unregister_quorum_device); | |
4712 | +EXPORT_SYMBOL(kcl_quorum_device_available); | |
4713 | + | |
4714 | +EXPORT_SYMBOL(kcl_register_service); | |
4715 | +EXPORT_SYMBOL(kcl_unregister_service); | |
4716 | +EXPORT_SYMBOL(kcl_join_service); | |
4717 | +EXPORT_SYMBOL(kcl_leave_service); | |
4718 | +EXPORT_SYMBOL(kcl_global_service_id); | |
4719 | +EXPORT_SYMBOL(kcl_start_done); | |
4720 | +EXPORT_SYMBOL(kcl_get_services); | |
4721 | +EXPORT_SYMBOL(kcl_get_current_interface); | |
4722 | + | |
4723 | +/* | |
4724 | + * Overrides for Emacs so that we follow Linus's tabbing style. | |
4725 | + * Emacs will notice this stuff at the end of the file and automatically | |
4726 | + * adjust the settings for this buffer only. This must remain at the end | |
4727 | + * of the file. | |
4728 | + * --------------------------------------------------------------------------- | |
4729 | + * Local variables: | |
4730 | + * c-file-style: "linux" | |
4731 | + * End: | |
4732 | + */ | |
4733 | diff -urN linux-orig/cluster/cman/config.c linux-patched/cluster/cman/config.c | |
4734 | --- linux-orig/cluster/cman/config.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 4735 | +++ linux-patched/cluster/cman/config.c 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 4736 | @@ -0,0 +1,46 @@ |
4737 | +/****************************************************************************** | |
4738 | +******************************************************************************* | |
4739 | +** | |
4740 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
4741 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4742 | +** | |
4743 | +** This copyrighted material is made available to anyone wishing to use, | |
4744 | +** modify, copy, or redistribute it subject to the terms and conditions | |
4745 | +** of the GNU General Public License v.2. | |
4746 | +** | |
4747 | +******************************************************************************* | |
4748 | +******************************************************************************/ | |
4749 | + | |
4750 | +#include "config.h" | |
4751 | + | |
4752 | +/* Config file defaults */ | |
4753 | + | |
4754 | +#define DEFAULT_JOIN_WAIT_TIME 11 /* Time to wait while sending JOINREQ | |
4755 | + * messages. Should be at least twice | |
4756 | + * the HELLO timer */ | |
4757 | +#define DEFAULT_JOIN_TIMEOUT 30 /* How long we wait after getting a | |
4758 | + * JOINACK to regarding that node as | |
4759 | + * dead */ | |
4760 | +#define DEFAULT_HELLO_TIMER 5 /* Period between HELLO messages */ | |
4761 | +#define DEFAULT_DEADNODE_TIMER 21 /* If we don't get a message from a | |
4762 | + * node in this period kill it */ | |
4763 | +#define DEFAULT_TRANSITION_TIMER 15 /* Maximum time a state transition | |
4764 | + * should take */ | |
4765 | +#define DEFAULT_JOINCONF_TIMER 5 /* Time allowed to a node to respond to | |
4766 | + * a JOINCONF message */ | |
4767 | +#define DEFAULT_MAX_NODES 128 /* Max allowed nodes */ | |
4768 | +#define DEFAULT_TRANSITION_RESTARTS 10 /* Maximum number of transition | |
4769 | + * restarts before we die */ | |
4770 | +#define DEFAULT_SM_DEBUG_SIZE 256 /* Size in bytes of SM debug buffer */ | |
4771 | + | |
4772 | +struct config_info cman_config = { | |
4773 | + .joinwait_timeout = DEFAULT_JOIN_WAIT_TIME, | |
4774 | + .joinconf_timeout = DEFAULT_JOINCONF_TIMER, | |
4775 | + .join_timeout = DEFAULT_JOIN_TIMEOUT, | |
4776 | + .hello_timer = DEFAULT_HELLO_TIMER, | |
4777 | + .deadnode_timeout = DEFAULT_DEADNODE_TIMER, | |
4778 | + .transition_timeout = DEFAULT_TRANSITION_TIMER, | |
4779 | + .transition_restarts = DEFAULT_TRANSITION_RESTARTS, | |
4780 | + .max_nodes = DEFAULT_MAX_NODES, | |
4781 | + .sm_debug_size = DEFAULT_SM_DEBUG_SIZE, | |
4782 | +}; | |
4783 | diff -urN linux-orig/cluster/cman/config.h linux-patched/cluster/cman/config.h | |
4784 | --- linux-orig/cluster/cman/config.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 4785 | +++ linux-patched/cluster/cman/config.h 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 4786 | @@ -0,0 +1,31 @@ |
4787 | +/****************************************************************************** | |
4788 | +******************************************************************************* | |
4789 | +** | |
4790 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
4791 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4792 | +** | |
4793 | +** This copyrighted material is made available to anyone wishing to use, | |
4794 | +** modify, copy, or redistribute it subject to the terms and conditions | |
4795 | +** of the GNU General Public License v.2. | |
4796 | +** | |
4797 | +******************************************************************************* | |
4798 | +******************************************************************************/ | |
4799 | + | |
4800 | +#ifndef __CONFIG_DOT_H__ | |
4801 | +#define __CONFIG_DOT_H__ | |
4802 | + | |
4803 | +struct config_info { | |
4804 | + int joinwait_timeout; | |
4805 | + int joinconf_timeout; | |
4806 | + int join_timeout; | |
4807 | + int hello_timer; | |
4808 | + int deadnode_timeout; | |
4809 | + int transition_timeout; | |
4810 | + int transition_restarts; | |
4811 | + int max_nodes; | |
4812 | + int sm_debug_size; | |
4813 | +}; | |
4814 | + | |
4815 | +extern struct config_info cman_config; | |
4816 | + | |
4817 | +#endif /* __CONFIG_DOT_H__ */ | |
4818 | diff -urN linux-orig/cluster/cman/kjoin.c linux-patched/cluster/cman/kjoin.c | |
4819 | --- linux-orig/cluster/cman/kjoin.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 4820 | +++ linux-patched/cluster/cman/kjoin.c 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 4821 | @@ -0,0 +1,238 @@ |
4822 | +/****************************************************************************** | |
4823 | +******************************************************************************* | |
4824 | +** | |
4825 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
4826 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4827 | +** | |
4828 | +** This copyrighted material is made available to anyone wishing to use, | |
4829 | +** modify, copy, or redistribute it subject to the terms and conditions | |
4830 | +** of the GNU General Public License v.2. | |
4831 | +** | |
4832 | +******************************************************************************* | |
4833 | +******************************************************************************/ | |
4834 | + | |
4835 | +#include <linux/socket.h> | |
4836 | +#include <net/sock.h> | |
4837 | +#include <linux/list.h> | |
4838 | +#include <cluster/cnxman.h> | |
4839 | +#include <linux/in.h> | |
4840 | + | |
4841 | +#include "cnxman-private.h" | |
4842 | + | |
4843 | +static struct socket *mcast_sock; | |
4844 | +static struct socket *recv_sock; | |
4845 | +static struct socket *cluster_sock; | |
4846 | + | |
4847 | +extern short cluster_id; | |
4848 | +extern int join_count; | |
4849 | +extern struct semaphore join_count_lock; | |
4850 | +extern atomic_t cnxman_running; | |
4851 | + | |
4852 | +int kcl_join_cluster(struct cl_join_cluster_info *join_info) | |
4853 | +{ | |
4854 | + int result; | |
4855 | + int one = 1, error; | |
4856 | + unsigned int ipaddr = join_info->ipaddr, brdaddr = join_info->brdaddr; | |
4857 | + unsigned short port = join_info->port; | |
4858 | + mm_segment_t fs; | |
4859 | + struct sockaddr_in saddr; | |
4860 | + struct kcl_multicast_sock mcast_info; | |
4861 | + | |
4862 | + down(&join_count_lock); | |
4863 | + if (atomic_read(&cnxman_running)) | |
4864 | + { | |
4865 | + error = 0; | |
4866 | + if (join_info->cluster_id == cluster_id) | |
4867 | + join_count++; | |
4868 | + else | |
4869 | + error = -EINVAL; | |
4870 | + up(&join_count_lock); | |
4871 | + return error; | |
4872 | + } | |
4873 | + up(&join_count_lock); | |
4874 | + | |
4875 | + result = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &mcast_sock); | |
4876 | + if (result < 0) | |
4877 | + { | |
4878 | + printk(KERN_ERR CMAN_NAME ": Can't create Multicast socket\n"); | |
4879 | + return result; | |
4880 | + } | |
4881 | + | |
4882 | + result = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &recv_sock); | |
4883 | + if (result < 0) | |
4884 | + { | |
4885 | + printk(KERN_ERR CMAN_NAME ": Can't create Receive socket\n"); | |
4886 | + return result; | |
4887 | + } | |
4888 | + | |
4889 | + fs = get_fs(); | |
4890 | + set_fs(get_ds()); | |
4891 | + | |
4892 | + if ((error = sock_setsockopt(mcast_sock, SOL_SOCKET, SO_BROADCAST, | |
4893 | + (void *) &one, sizeof (int)))) | |
4894 | + { | |
4895 | + set_fs(fs); | |
4896 | + printk("Error %d Setting master socket to SO_BROADCAST\n", | |
4897 | + error); | |
4898 | + sock_release(mcast_sock); | |
4899 | + return -1; | |
4900 | + } | |
4901 | + set_fs(fs); | |
4902 | + | |
4903 | + /* Bind the multicast socket */ | |
4904 | + saddr.sin_family = AF_INET; | |
4905 | + saddr.sin_port = htons(port); | |
4906 | + saddr.sin_addr.s_addr = cpu_to_be32(brdaddr); | |
4907 | + result = | |
4908 | + mcast_sock->ops->bind(mcast_sock, (struct sockaddr *) &saddr, | |
4909 | + sizeof (saddr)); | |
4910 | + if (result < 0) | |
4911 | + { | |
4912 | + printk(KERN_ERR CMAN_NAME ": Can't bind multicast socket\n"); | |
4913 | + sock_release(mcast_sock); | |
4914 | + sock_release(recv_sock); | |
4915 | + return result; | |
4916 | + } | |
4917 | + | |
4918 | + /* Bind the receive socket to our IP address */ | |
4919 | + saddr.sin_family = AF_INET; | |
4920 | + saddr.sin_port = htons(port); | |
4921 | + saddr.sin_addr.s_addr = cpu_to_be32(ipaddr); | |
4922 | + result = | |
4923 | + recv_sock->ops->bind(recv_sock, (struct sockaddr *) &saddr, | |
4924 | + sizeof (saddr)); | |
4925 | + if (result < 0) | |
4926 | + { | |
4927 | + printk(KERN_ERR CMAN_NAME ": Can't bind receive socket\n"); | |
4928 | + sock_release(mcast_sock); | |
4929 | + sock_release(recv_sock); | |
4930 | + return result; | |
4931 | + } | |
4932 | + | |
4933 | + /* Create the cluster master socket */ | |
4934 | + result = | |
4935 | + sock_create(AF_CLUSTER, SOCK_DGRAM, CLPROTO_MASTER, &cluster_sock); | |
4936 | + if (result < 0) | |
4937 | + { | |
4938 | + printk(KERN_ERR CMAN_NAME | |
4939 | + ": Can't create cluster master socket\n"); | |
4940 | + sock_release(mcast_sock); | |
4941 | + sock_release(recv_sock); | |
4942 | + return result; | |
4943 | + } | |
4944 | + | |
4945 | + /* This is the broadcast transmit address */ | |
4946 | + saddr.sin_addr.s_addr = cpu_to_be32(brdaddr); | |
4947 | + | |
4948 | + /* Pass the multicast socket to kernel space */ | |
4949 | + mcast_info.sock = mcast_sock; | |
4950 | + mcast_info.number = 1; | |
4951 | + | |
4952 | + fs = get_fs(); | |
4953 | + set_fs(get_ds()); | |
4954 | + | |
4955 | + if ((error = cluster_sock->ops->setsockopt(cluster_sock, CLPROTO_MASTER, | |
4956 | + KCL_SET_MULTICAST, | |
4957 | + (void *) &mcast_info, | |
4958 | + sizeof (mcast_info)))) | |
4959 | + { | |
4960 | + set_fs(fs); | |
4961 | + printk(CMAN_NAME | |
4962 | + ": Unable to pass multicast socket to cnxman, %d\n", | |
4963 | + error); | |
4964 | + sock_release(mcast_sock); | |
4965 | + sock_release(recv_sock); | |
4966 | + sock_release(cluster_sock); | |
4967 | + return -1; | |
4968 | + } | |
4969 | + | |
4970 | + mcast_info.sock = recv_sock; | |
4971 | + if ((error = | |
4972 | + cluster_sock->ops->setsockopt(cluster_sock, CLPROTO_MASTER, | |
4973 | + KCL_SET_RCVONLY, | |
4974 | + (void *) &mcast_info, | |
4975 | + sizeof (mcast_info)))) | |
4976 | + { | |
4977 | + set_fs(fs); | |
4978 | + printk(CMAN_NAME | |
4979 | + ": Unable to pass receive socket to cnxman, %d\n", | |
4980 | + error); | |
4981 | + sock_release(mcast_sock); | |
4982 | + sock_release(recv_sock); | |
4983 | + sock_release(cluster_sock); | |
4984 | + return -1; | |
4985 | + } | |
4986 | + | |
4987 | + /* This setsockopt expects usermode variables */ | |
4988 | + | |
4989 | + if (cluster_sock->ops-> | |
4990 | + setsockopt(cluster_sock, CLPROTO_MASTER, CLU_JOIN_CLUSTER, | |
4991 | + (void *) join_info, | |
4992 | + sizeof (struct cl_join_cluster_info))) | |
4993 | + | |
4994 | + { | |
4995 | + set_fs(fs); | |
4996 | + printk(CMAN_NAME ": Unable to join cluster\n"); | |
4997 | + sock_release(mcast_sock); | |
4998 | + sock_release(recv_sock); | |
4999 | + sock_release(cluster_sock); | |
5000 | + return -1; | |
5001 | + } | |
5002 | + set_fs(fs); | |
5003 | + | |
5004 | + return 0; | |
5005 | +} | |
5006 | + | |
5007 | +int kcl_leave_cluster(int remove) | |
5008 | +{ | |
5009 | + mm_segment_t fs; | |
5010 | + int rem = remove; | |
5011 | + int ret = 0; | |
5012 | + struct socket *shutdown_sock = cluster_sock; | |
5013 | + | |
5014 | + cluster_sock = NULL; | |
5015 | + | |
5016 | + if (!shutdown_sock) | |
5017 | + { | |
5018 | + /* Create the cluster master socket */ | |
5019 | + int result = | |
5020 | + sock_create(AF_CLUSTER, SOCK_DGRAM, CLPROTO_MASTER, | |
5021 | + &shutdown_sock); | |
5022 | + if (result < 0) | |
5023 | + { | |
5024 | + printk(KERN_ERR CMAN_NAME | |
5025 | + ": Can't create cluster master socket\n"); | |
5026 | + sock_release(mcast_sock); | |
5027 | + sock_release(recv_sock); | |
5028 | + return result; | |
5029 | + } | |
5030 | + } | |
5031 | + | |
5032 | + fs = get_fs(); | |
5033 | + set_fs(get_ds()); | |
5034 | + | |
5035 | + if ((ret = | |
5036 | + shutdown_sock->ops->setsockopt(shutdown_sock, CLPROTO_MASTER, | |
5037 | + CLU_LEAVE_CLUSTER, (void *) &rem, | |
5038 | + sizeof (int)))) | |
5039 | + { | |
5040 | + printk(KERN_ERR CMAN_NAME ": Unable to leave cluster, %d\n", | |
5041 | + ret); | |
5042 | + } | |
5043 | + set_fs(fs); | |
5044 | + | |
5045 | + sock_release(shutdown_sock); | |
5046 | + | |
5047 | + return ret; | |
5048 | +} | |
5049 | + | |
5050 | +/* | |
5051 | + * Overrides for Emacs so that we follow Linus's tabbing style. | |
5052 | + * Emacs will notice this stuff at the end of the file and automatically | |
5053 | + * adjust the settings for this buffer only. This must remain at the end | |
5054 | + * of the file. | |
5055 | + * --------------------------------------------------------------------------- | |
5056 | + * Local variables: | |
5057 | + * c-file-style: "linux" | |
5058 | + * End: | |
5059 | + */ | |
5060 | diff -urN linux-orig/cluster/cman/membership.c linux-patched/cluster/cman/membership.c | |
5061 | --- linux-orig/cluster/cman/membership.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 5062 | +++ linux-patched/cluster/cman/membership.c 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 5063 | @@ -0,0 +1,3069 @@ |
5064 | +/****************************************************************************** | |
5065 | +******************************************************************************* | |
5066 | +** | |
5067 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5068 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
5069 | +** | |
5070 | +** This copyrighted material is made available to anyone wishing to use, | |
5071 | +** modify, copy, or redistribute it subject to the terms and conditions | |
5072 | +** of the GNU General Public License v.2. | |
5073 | +** | |
5074 | +******************************************************************************* | |
5075 | +******************************************************************************/ | |
5076 | + | |
5077 | +#include <linux/socket.h> | |
5078 | +#include <net/sock.h> | |
5079 | +#include <linux/slab.h> | |
5080 | +#include <linux/spinlock.h> | |
5081 | +#include <linux/vmalloc.h> | |
5082 | +#include <asm/uaccess.h> | |
5083 | +#include <linux/list.h> | |
5084 | +#include <cluster/cnxman.h> | |
5085 | + | |
5086 | +#include "cnxman-private.h" | |
5087 | +#include "config.h" | |
5088 | +#include "sm_control.h" | |
5089 | + | |
5090 | +#ifndef TRUE | |
5091 | +#define TRUE 1 | |
5092 | +#endif | |
5093 | + | |
5094 | +/* Barrier name for membership transitions. %d is the cluster generation number | |
5095 | + */ | |
5096 | +#define MEMBERSHIP_BARRIER_NAME "TRANSITION.%d" | |
5097 | + | |
5098 | +/* Variables also used by connection manager */ | |
5099 | +struct list_head cluster_members_list; | |
5100 | +struct semaphore cluster_members_lock; | |
5101 | +int cluster_members; /* Number of ACTIVE members, not a count of | |
5102 | + * nodes in the list */ | |
5103 | +int we_are_a_cluster_member = 0; | |
5104 | +int cluster_is_quorate; | |
5105 | +int quit_threads = 0; | |
5106 | +struct task_struct *membership_task; | |
5107 | +struct cluster_node *us; | |
5108 | + | |
5109 | +static struct task_struct *hello_task; | |
5110 | +static struct semaphore hello_task_lock; | |
5111 | + | |
5112 | +/* Variables that belong to the connection manager */ | |
5113 | +extern wait_queue_head_t cnxman_waitq; | |
5114 | +extern struct completion member_thread_comp; | |
5115 | +extern struct cluster_node *quorum_device; | |
5116 | +extern unsigned short two_node; | |
5117 | +extern char cluster_name[]; | |
5118 | +extern unsigned int config_version; | |
5119 | +extern unsigned int address_length; | |
5120 | + | |
5121 | +static struct socket *mem_socket; | |
5122 | +static pid_t kcluster_pid; | |
5123 | + | |
5124 | +static char iobuf[MAX_CLUSTER_MESSAGE]; | |
5125 | +static char scratchbuf[MAX_CLUSTER_MESSAGE + 100]; | |
5126 | + | |
5127 | +/* Our node name, usually system_utsname.nodename, but can be overridden */ | |
5128 | +char nodename[MAX_CLUSTER_MEMBER_NAME_LEN + 1]; | |
5129 | + | |
5130 | +static spinlock_t members_by_nodeid_lock; | |
5131 | +static int sizeof_members_array = 0; /* Can dynamically increase (vmalloc | |
5132 | + * permitting) */ | |
5133 | +static struct cluster_node **members_by_nodeid; | |
5134 | + | |
5135 | +#define MEMBER_INCREMENT_SIZE 10 | |
5136 | + | |
5137 | +static int votes = 1; /* Votes this node has */ | |
5138 | +static int expected_votes = 1; /* Total expected votes in the cluster */ | |
5139 | +static unsigned int quorum; /* Quorum, fewer votes than this and we stop | |
5140 | + * work */ | |
5141 | +static int leavereason; /* Saved for the duration of a state transition */ | |
5142 | +static int transitionreason; /* Reason this transition was initiated */ | |
5143 | +static unsigned int highest_nodeid; /* Highest node ID known to the cluster */ | |
5144 | +static struct timer_list transition_timer; /* Kicks in if the transition | |
5145 | + * doesn't complete in a | |
5146 | + * reasonable time */ | |
5147 | +static struct timer_list hello_timer; /* Timer to send HELLOs on */ | |
5148 | +static unsigned long join_time; /* The time that we got our JOIN-ACK */ | |
5149 | +static unsigned long start_time; /* The time that we were started */ | |
5150 | +static int joinconf_count; /* Number of JOINCONF messages we have sent to | |
5151 | + * a new node */ | |
5152 | +static unsigned long wake_flags;/* Reason we were woken */ | |
5153 | + | |
5154 | +/* Flags in above */ | |
5155 | +#define WAKE_FLAG_DEADNODE 1 | |
5156 | +#define WAKE_FLAG_TRANSTIMER 2 | |
5157 | + | |
5158 | +/* The time the transition finished */ | |
5159 | +static unsigned long transition_end_time; | |
5160 | + | |
5161 | +/* A list of nodes that cnxman tells us are dead. I hope this never has more | |
5162 | + * than one element in it but I can't take that chance. only non-static so it | |
5163 | + * can be initialised in module_load. */ | |
5164 | +struct list_head new_dead_node_list; | |
5165 | +struct semaphore new_dead_node_lock; | |
5166 | + | |
5167 | +static int do_membership_packet(struct msghdr *msg, int len); | |
5168 | +static int do_process_joinreq(struct msghdr *msg, int len); | |
5169 | +static int do_process_joinack(struct msghdr *msg, int len); | |
5170 | +static int do_process_joinconf(struct msghdr *msg, int len); | |
5171 | +static int do_process_leave(struct msghdr *msg, int len); | |
5172 | +static int do_process_hello(struct msghdr *msg, int len); | |
5173 | +static int do_process_kill(struct msghdr *msg, int len); | |
5174 | +static int do_process_reconfig(struct msghdr *msg, int len); | |
5175 | +static int do_process_starttrans(struct msghdr *msg, int len); | |
5176 | +static int do_process_masterview(struct msghdr *msg, int len); | |
5177 | +static int do_process_endtrans(struct msghdr *msg, int len); | |
5178 | +static int do_process_viewack(struct msghdr *msg, int len); | |
5179 | +static int do_process_startack(struct msghdr *msg, int len); | |
5180 | +static int do_process_newcluster(struct msghdr *msg, int len); | |
5181 | +static int do_process_nominate(struct msghdr *msg, int len); | |
5182 | +static int send_cluster_view(unsigned char cmd, struct sockaddr_cl *saddr, | |
5183 | + unsigned int flags); | |
5184 | +static int send_joinreq(struct sockaddr_cl *addr, int addr_len); | |
5185 | +static int send_startack(struct sockaddr_cl *addr, int addr_len, int node_id); | |
5186 | +static int send_hello(void); | |
5187 | +static int send_master_hello(void); | |
5188 | +static int send_newcluster(void); | |
5189 | +static int end_transition(void); | |
5190 | +static int dispatch_messages(struct socket *mem_socket); | |
5191 | +static void check_for_dead_nodes(void); | |
5192 | +static void confirm_joiner(void); | |
5193 | +static void reset_hello_time(void); | |
5194 | +static int add_us(void); | |
5195 | +static int send_joinconf(void); | |
5196 | +static int init_membership_services(void); | |
5197 | +static int elect_master(struct cluster_node **); | |
5198 | +static void trans_timer_expired(unsigned long arg); | |
5199 | +static void hello_timer_expired(unsigned long arg); | |
5200 | +static void join_or_form_cluster(void); | |
5201 | +static int do_timer_wakeup(void); | |
5202 | +static int start_transition(unsigned char reason, struct cluster_node *node); | |
5203 | +int send_leave(unsigned char); | |
5204 | +int send_reconfigure(int, unsigned int); | |
5205 | + | |
5206 | +#ifdef DEBUG_MEMB | |
5207 | +static char *msgname(int msg); | |
5208 | +static int debug_sendmsg(struct socket *sock, void *buf, int size, | |
5209 | + struct sockaddr_cl *caddr, int addr_len, | |
5210 | + unsigned int flags) | |
5211 | +{ | |
5212 | + P_MEMB("%ld: sending %s, len=%d\n", jiffies, msgname(((char *) buf)[0]), | |
5213 | + size); | |
5214 | + return kcl_sendmsg(sock, buf, size, caddr, addr_len, flags); | |
5215 | +} | |
5216 | + | |
5217 | +#define kcl_sendmsg debug_sendmsg | |
5218 | +#endif | |
5219 | + | |
5220 | +/* State of the node */ | |
5221 | +static enum { STARTING, JOINING, JOINWAIT, JOINACK, TRANSITION, | |
5222 | + TRANSITION_COMPLETE, MEMBER, REJECTED, LEFT_CLUSTER, MASTER | |
5223 | +} node_state = STARTING; | |
5224 | + | |
5225 | +/* Sub-state when we are MASTER */ | |
5226 | +static enum { MASTER_START, MASTER_COLLECT, MASTER_CONFIRM, | |
5227 | + MASTER_COMPLETE } master_state; | |
5228 | + | |
5229 | +/* Number of responses collected while a master controlling a state transition */ | |
5230 | +static int responses_collected; | |
5231 | +static int responses_expected; | |
5232 | + | |
5233 | +/* Current cluster generation number */ | |
5234 | +static int cluster_generation = 1; | |
5235 | + | |
5236 | +/* When another node initiates a transtion then store it's pointer in here so | |
5237 | + * we can check for other nodes trying to spoof us */ | |
5238 | +static struct cluster_node *master_node = NULL; | |
5239 | + | |
5240 | +/* Struct the node wanting to join us */ | |
5241 | +static struct cluster_node *joining_node = NULL; | |
5242 | +static int joining_temp_nodeid = 0; | |
5243 | + | |
5244 | +/* Last time a HELLO message was sent */ | |
5245 | +unsigned long last_hello = 0; | |
5246 | + | |
5247 | +/* When we got our JOINWAIT or NEWCLUSTER */ | |
5248 | +unsigned long joinwait_time = 0; | |
5249 | + | |
5250 | +/* Number of times a transition has restarted when we were master */ | |
5251 | +int transition_restarts = 0; | |
5252 | + | |
5253 | +/* Variables used by the master to collect cluster status during a transition */ | |
5254 | +static int agreeing_nodes = 0; | |
5255 | +static int dissenting_nodes = 0; | |
5256 | +static uint8_t *node_opinion = NULL; | |
5257 | +#define OPINION_AGREE 1 | |
5258 | +#define OPINION_DISAGREE 2 | |
5259 | + | |
5260 | +/* Set node id of a node, also add it to the members array and expand the array | |
5261 | + * if necessary */ | |
5262 | +static inline void set_nodeid(struct cluster_node *node, int nodeid) | |
5263 | +{ | |
5264 | + if (!nodeid) | |
5265 | + return; | |
5266 | + | |
5267 | + node->node_id = nodeid; | |
5268 | + if (nodeid > sizeof_members_array) { | |
5269 | + int new_size = sizeof_members_array + MEMBER_INCREMENT_SIZE; | |
5270 | + struct cluster_node **new_array = | |
5271 | + vmalloc((new_size) * sizeof (struct cluster_node *)); | |
5272 | + if (new_array) { | |
5273 | + spin_lock(&members_by_nodeid_lock); | |
5274 | + memcpy(new_array, members_by_nodeid, | |
5275 | + sizeof_members_array * | |
5276 | + sizeof (struct cluster_node *)); | |
5277 | + memset(&new_array[sizeof_members_array], 0, | |
5278 | + MEMBER_INCREMENT_SIZE * | |
5279 | + sizeof (struct cluster_node *)); | |
5280 | + vfree(members_by_nodeid); | |
5281 | + members_by_nodeid = new_array; | |
5282 | + sizeof_members_array = new_size; | |
5283 | + spin_unlock(&members_by_nodeid_lock); | |
5284 | + } | |
5285 | + else { | |
5286 | + panic("No memory for more nodes"); | |
5287 | + } | |
5288 | + } | |
5289 | + notify_kernel_listeners(NEWNODE, (long) nodeid); | |
5290 | + | |
5291 | + spin_lock(&members_by_nodeid_lock); | |
5292 | + members_by_nodeid[nodeid] = node; | |
5293 | + spin_unlock(&members_by_nodeid_lock); | |
5294 | +} | |
5295 | + | |
5296 | +static int hello_kthread(void *unused) | |
5297 | +{ | |
5298 | + struct task_struct *tsk = current; | |
5299 | + sigset_t tmpsig; | |
5300 | + | |
5301 | + daemonize("cman_hbeat"); | |
5302 | + | |
5303 | + /* Block everything but SIGKILL/SIGSTOP/SIGTERM */ | |
5304 | + siginitset(&tmpsig, SIGKILL | SIGSTOP | SIGTERM); | |
5305 | + sigprocmask(SIG_BLOCK, &tmpsig, NULL); | |
5306 | + | |
5307 | + down(&hello_task_lock); | |
5308 | + hello_task = tsk; | |
5309 | + up(&hello_task_lock); | |
5310 | + | |
5311 | + set_user_nice(current, -6); | |
5312 | + | |
5313 | + while (node_state != REJECTED && node_state != LEFT_CLUSTER) { | |
5314 | + send_hello(); | |
5315 | + | |
5316 | + /* Scan the nodes list for dead nodes */ | |
5317 | + if (node_state == MEMBER) | |
5318 | + check_for_dead_nodes(); | |
5319 | + | |
5320 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
5321 | + schedule(); | |
5322 | + set_task_state(current, TASK_RUNNING); | |
5323 | + } | |
5324 | + down(&hello_task_lock); | |
5325 | + hello_task = NULL; | |
5326 | + up(&hello_task_lock); | |
5327 | + P_MEMB("heartbeat closing down\n"); | |
5328 | + return 0; | |
5329 | +} | |
5330 | + | |
5331 | +/* This is the membership "daemon". A client of cnxman (but symbiotic with it) | |
5332 | + * that keeps track of and controls cluster membership. */ | |
5333 | +static int membership_kthread(void *unused) | |
5334 | +{ | |
5335 | + struct task_struct *tsk = current; | |
5336 | + struct socket *tmp_socket; | |
5337 | + sigset_t tmpsig; | |
5338 | + | |
5339 | + daemonize("cman_memb"); | |
5340 | + | |
5341 | + /* Block everything but SIGKILL/SIGSTOP/SIGTERM */ | |
5342 | + siginitset(&tmpsig, SIGKILL | SIGSTOP | SIGTERM); | |
5343 | + sigprocmask(SIG_BLOCK, &tmpsig, NULL); | |
5344 | + | |
5345 | + membership_task = tsk; | |
5346 | + set_user_nice(current, -5); | |
5347 | + | |
5348 | + /* Open the socket */ | |
5349 | + if (init_membership_services()) | |
5350 | + return -1; | |
5351 | + | |
5352 | + add_us(); | |
5353 | + joining_node = us; | |
5354 | + | |
5355 | + init_timer(&hello_timer); | |
5356 | + hello_timer.function = hello_timer_expired; | |
5357 | + hello_timer.data = 0L; | |
5358 | + | |
5359 | + /* Do joining stuff */ | |
5360 | + join_or_form_cluster(); | |
5361 | + | |
5362 | + transition_end_time = jiffies; | |
5363 | + | |
5364 | + /* Main loop */ | |
5365 | + while (node_state != REJECTED && node_state != LEFT_CLUSTER) { | |
5366 | + | |
5367 | + struct task_struct *tsk = current; | |
5368 | + | |
5369 | + DECLARE_WAITQUEUE(wait, tsk); | |
5370 | + | |
5371 | + tsk->state = TASK_INTERRUPTIBLE; | |
5372 | + add_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5373 | + | |
5374 | + if (!skb_peek(&mem_socket->sk->sk_receive_queue) && | |
5375 | + wake_flags == 0) { | |
5376 | + if (node_state == JOINACK || | |
5377 | + node_state == JOINWAIT) | |
5378 | + schedule_timeout(HZ); | |
5379 | + else | |
5380 | + schedule(); | |
5381 | + } | |
5382 | + | |
5383 | + tsk->state = TASK_RUNNING; | |
5384 | + remove_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5385 | + | |
5386 | + /* Are we being shut down? */ | |
5387 | + if (node_state == LEFT_CLUSTER || quit_threads || | |
5388 | + signal_pending(current)) | |
5389 | + break; | |
5390 | + | |
5391 | + /* Were we woken by a dead node passed down from cnxman ? */ | |
5392 | + if (test_and_clear_bit(WAKE_FLAG_DEADNODE, &wake_flags)) { | |
5393 | + struct list_head *nodelist, *tmp; | |
5394 | + struct cl_new_dead_node *deadnode; | |
5395 | + | |
5396 | + down(&new_dead_node_lock); | |
5397 | + list_for_each_safe(nodelist, tmp, &new_dead_node_list) { | |
5398 | + deadnode = | |
5399 | + list_entry(nodelist, | |
5400 | + struct cl_new_dead_node, list); | |
5401 | + | |
5402 | + if (deadnode->node->state == NODESTATE_MEMBER) | |
5403 | + a_node_just_died(deadnode->node); | |
5404 | + list_del(&deadnode->list); | |
5405 | + kfree(deadnode); | |
5406 | + } | |
5407 | + up(&new_dead_node_lock); | |
5408 | + } | |
5409 | + | |
5410 | + /* Process received messages. If dispatch_message() returns an | |
5411 | + * error then we shut down */ | |
5412 | + if (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
5413 | + if (dispatch_messages(mem_socket) < 0) | |
5414 | + goto leave_cluster; | |
5415 | + | |
5416 | + } | |
5417 | + | |
5418 | + /* Were we woken by the transition timer firing ? */ | |
5419 | + if (test_and_clear_bit(WAKE_FLAG_TRANSTIMER, &wake_flags)) { | |
5420 | + switch (do_timer_wakeup()) { | |
5421 | + case -1: | |
5422 | + continue; | |
5423 | + case 0: | |
5424 | + break; | |
5425 | + case +1: | |
5426 | + goto leave_cluster; | |
5427 | + } | |
5428 | + } | |
5429 | + | |
5430 | + /* Got a JOINACK but no JOIN-CONF, start waiting for HELLO | |
5431 | + * messages again */ | |
5432 | + if (node_state == JOINACK | |
5433 | + && time_after(jiffies, | |
5434 | + join_time + cman_config.join_timeout * HZ)) { | |
5435 | + P_MEMB | |
5436 | + ("Waited a long time for a join-conf, going back to JOINWAIT state\n"); | |
5437 | + node_state = JOINWAIT; | |
5438 | + joinwait_time = jiffies; | |
5439 | + } | |
5440 | + | |
5441 | + /* Have we been in joinwait for too long... */ | |
5442 | + if (node_state == JOINWAIT | |
5443 | + && time_after(jiffies, joinwait_time + | |
5444 | + cman_config.join_timeout * HZ)) { | |
5445 | + printk(CMAN_NAME | |
5446 | + ": Been in JOINWAIT for too long - giving up\n"); | |
5447 | + goto leave_cluster; | |
5448 | + } | |
5449 | + } | |
5450 | + | |
5451 | + leave_cluster: | |
5452 | + | |
5453 | + /* Wake up the heartbeat thread so it can exit */ | |
5454 | + down(&hello_task_lock); | |
5455 | + if (hello_task) | |
5456 | + wake_up_process(hello_task); | |
5457 | + up(&hello_task_lock); | |
5458 | + | |
5459 | + if (timer_pending(&hello_timer)) | |
5460 | + del_timer(&hello_timer); | |
5461 | + | |
5462 | + if (timer_pending(&transition_timer)) | |
5463 | + del_timer(&transition_timer); | |
5464 | + | |
5465 | + node_state = LEFT_CLUSTER; | |
5466 | + P_MEMB("closing down\n"); | |
5467 | + quit_threads = 1; /* force other thread to exit too */ | |
5468 | + | |
5469 | + /* Close the socket, NULL the pointer first so it doesn't get used | |
5470 | + * by send_leave() | |
5471 | + */ | |
5472 | + tmp_socket = mem_socket; | |
5473 | + mem_socket = NULL; | |
5474 | + sock_release(tmp_socket); | |
5475 | + highest_nodeid = 0; | |
5476 | + complete(&member_thread_comp); | |
5477 | + return 0; | |
5478 | +} | |
5479 | + | |
5480 | +/* Things to do in the main thread when the transition timer has woken us. | |
5481 | + * Usually this happens when a transition is taking too long and we need to | |
5482 | + * take remedial action. | |
5483 | + * | |
5484 | + * returns: -1 continue; 0 carry on processing +1 leave cluster; */ | |
5485 | +static int do_timer_wakeup() | |
5486 | +{ | |
5487 | + P_MEMB("Timer wakeup - checking for dead master node %ld\n", jiffies); | |
5488 | + | |
5489 | + /* Resend JOINCONF if it got lost on the wire */ | |
5490 | + if (node_state == MASTER && master_state == MASTER_CONFIRM) { | |
5491 | + mod_timer(&transition_timer, | |
5492 | + jiffies + cman_config.joinconf_timeout * HZ); | |
5493 | + if (++joinconf_count < MAX_RETRIES) { | |
5494 | + P_MEMB("Resending JOINCONF\n"); | |
5495 | + send_joinconf(); | |
5496 | + } | |
5497 | + else { | |
5498 | + P_MEMB("JOINCONF not acked, cancelling transition\n"); | |
5499 | + end_transition(); | |
5500 | + } | |
5501 | + return -1; | |
5502 | + } | |
5503 | + | |
5504 | + /* A joining node probably died */ | |
5505 | + if (cluster_members == 1) { | |
5506 | + end_transition(); | |
5507 | + return -1; | |
5508 | + } | |
5509 | + | |
5510 | + /* See if the master is still there */ | |
5511 | + if (node_state == TRANSITION || node_state == TRANSITION_COMPLETE) { | |
5512 | + | |
5513 | + /* If we are in transition and master_node is NULL then we are | |
5514 | + * waiting for ENDTRANS after JOIN-CONF */ | |
5515 | + if (!master_node) { | |
5516 | + /* Hmmm. master died after sending JOINCONF, we'll have | |
5517 | + * to die as we are in mid-transition */ | |
5518 | + printk(KERN_INFO CMAN_NAME | |
5519 | + ": Master died after JOINCONF, we must leave the cluster\n"); | |
5520 | + quit_threads = 1; | |
5521 | + return +1; | |
5522 | + } | |
5523 | + | |
5524 | + /* No messages from the master - see if it's stil there */ | |
5525 | + if (master_node->state == NODESTATE_MEMBER) { | |
5526 | + send_master_hello(); | |
5527 | + mod_timer(&transition_timer, | |
5528 | + jiffies + | |
5529 | + cman_config.transition_timeout * HZ); | |
5530 | + } | |
5531 | + | |
5532 | + /* If the master is dead then elect a new one */ | |
5533 | + if (master_node->state == NODESTATE_DEAD) { | |
5534 | + | |
5535 | + struct cluster_node *node; | |
5536 | + | |
5537 | + P_MEMB("Master node is dead...Election!\n"); | |
5538 | + if (elect_master(&node)) { | |
5539 | + | |
5540 | + /* We are master now, all kneel */ | |
5541 | + start_transition(TRANS_DEADMASTER, master_node); | |
5542 | + } | |
5543 | + else { | |
5544 | + /* Leave the job to someone on more pay */ | |
5545 | + master_node = node; | |
5546 | + mod_timer(&transition_timer, | |
5547 | + jiffies + | |
5548 | + cman_config.transition_timeout * HZ); | |
5549 | + } | |
5550 | + } | |
5551 | + } | |
5552 | + | |
5553 | + /* If we are the master node then restart the transition */ | |
5554 | + if (node_state == MASTER) { | |
5555 | + start_transition(TRANS_RESTART, us); | |
5556 | + } | |
5557 | + | |
5558 | + return 0; | |
5559 | +} | |
5560 | + | |
5561 | +static void form_cluster(void) | |
5562 | +{ | |
5563 | + printk(KERN_INFO CMAN_NAME ": forming a new cluster\n"); | |
5564 | + node_state = MEMBER; | |
5565 | + we_are_a_cluster_member = TRUE; | |
5566 | + us->node_id = 1; | |
5567 | + us->state = NODESTATE_MEMBER; | |
5568 | + set_nodeid(us, 1); | |
5569 | + recalculate_quorum(0); | |
5570 | + sm_member_update(cluster_is_quorate); | |
5571 | + send_hello(); | |
5572 | + kernel_thread(hello_kthread, NULL, 0); | |
5573 | + mod_timer(&hello_timer, jiffies + cman_config.hello_timer * HZ); | |
5574 | +} | |
5575 | + | |
5576 | +/* This does the initial JOIN part of the membership process. Actually most of | |
5577 | + * is done in the message processing routines but this is the main loop that | |
5578 | + * controls it. The side-effect of this routine is "node_state" which tells the | |
5579 | + * real main loop (in the kernel thread routine) what to do next */ | |
5580 | +static void join_or_form_cluster() | |
5581 | +{ | |
5582 | + start_time = jiffies; | |
5583 | + | |
5584 | + printk(KERN_INFO CMAN_NAME | |
5585 | + ": Waiting to join or form a Linux-cluster\n"); | |
5586 | + join_time = 0; | |
5587 | + start_time = jiffies; | |
5588 | + joinwait_time = jiffies; | |
5589 | + last_hello = 0; | |
5590 | + send_newcluster(); | |
5591 | + | |
5592 | + /* Listen for a reply */ | |
5593 | + do { | |
5594 | + DECLARE_WAITQUEUE(wait, current); | |
5595 | + set_task_state(current, TASK_INTERRUPTIBLE); | |
5596 | + add_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5597 | + | |
5598 | + if (!skb_peek(&mem_socket->sk->sk_receive_queue)) | |
5599 | + schedule_timeout((cman_config.joinwait_timeout * HZ) / | |
5600 | + 5); | |
5601 | + | |
5602 | + set_task_state(current, TASK_RUNNING); | |
5603 | + remove_wait_queue(mem_socket->sk->sk_sleep, &wait); | |
5604 | + | |
5605 | + while (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
5606 | + dispatch_messages(mem_socket); | |
5607 | + } | |
5608 | + if (quit_threads) | |
5609 | + node_state = LEFT_CLUSTER; | |
5610 | + | |
5611 | + } | |
5612 | + while (time_before(jiffies, start_time + cman_config.joinwait_timeout * HZ) && | |
5613 | + node_state == STARTING); | |
5614 | + | |
5615 | + /* If we didn't hear any HELLO messages then form a new cluster */ | |
5616 | + if (node_state == STARTING) { | |
5617 | + form_cluster(); | |
5618 | + } | |
5619 | + else | |
5620 | + last_hello = jiffies; | |
5621 | + | |
5622 | +} | |
5623 | + | |
5624 | +int start_membership_services(pid_t cluster_pid) | |
5625 | +{ | |
5626 | + kcluster_pid = cluster_pid; | |
5627 | + | |
5628 | + init_timer(&transition_timer); | |
5629 | + transition_timer.function = trans_timer_expired; | |
5630 | + transition_timer.data = 0L; | |
5631 | + | |
5632 | + /* Start the thread */ | |
5633 | + return kernel_thread(membership_kthread, NULL, 0); | |
5634 | +} | |
5635 | + | |
5636 | +static int init_membership_services() | |
5637 | +{ | |
5638 | + int result; | |
5639 | + struct sockaddr_cl saddr; | |
5640 | + struct socket *sock; | |
5641 | + | |
5642 | + init_MUTEX(&hello_task_lock); | |
5643 | + /* Create a socket to communicate with */ | |
5644 | + result = sock_create_kern(AF_CLUSTER, SOCK_DGRAM, CLPROTO_CLIENT, &sock); | |
5645 | + if (result < 0) { | |
5646 | + printk(KERN_ERR CMAN_NAME | |
5647 | + ": Can't create cluster socket for membership services\n"); | |
5648 | + return result; | |
5649 | + } | |
5650 | + mem_socket = sock; | |
5651 | + | |
5652 | + /* Bind to our port */ | |
5653 | + saddr.scl_family = AF_CLUSTER; | |
5654 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5655 | + result = | |
5656 | + sock->ops->bind(sock, (struct sockaddr *) &saddr, sizeof (saddr)); | |
5657 | + if (result < 0) { | |
5658 | + printk(KERN_ERR CMAN_NAME | |
5659 | + ": Can't bind to cluster membership services port\n"); | |
5660 | + sock_release(sock); | |
5661 | + return result; | |
5662 | + } | |
5663 | + | |
5664 | + node_state = STARTING; | |
5665 | + return 0; | |
5666 | +} | |
5667 | + | |
5668 | +static int send_joinconf() | |
5669 | +{ | |
5670 | + struct sockaddr_cl saddr; | |
5671 | + int status; | |
5672 | + | |
5673 | + if (joining_temp_nodeid == 0) { | |
5674 | + BUG(); | |
5675 | + } | |
5676 | + | |
5677 | + master_state = MASTER_CONFIRM; | |
5678 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5679 | + saddr.scl_family = AF_CLUSTER; | |
5680 | + saddr.scl_nodeid = joining_temp_nodeid; | |
5681 | + status = send_cluster_view(CLUSTER_MEM_JOINCONF, &saddr, | |
5682 | + MSG_NOACK); | |
5683 | + | |
5684 | + if (status < 0) { | |
5685 | + printk("Error %d sending JOINCONF, aborting transition\n", status); | |
5686 | + end_transition(); | |
5687 | + } | |
5688 | + return status; | |
5689 | +} | |
5690 | + | |
5691 | +static int send_joinreq(struct sockaddr_cl *addr, int addr_len) | |
5692 | +{ | |
5693 | + char *msgbuf = scratchbuf; | |
5694 | + struct list_head *addrlist; | |
5695 | + int ptr = sizeof (struct cl_mem_join_msg); | |
5696 | + unsigned short num_addr = 0; | |
5697 | + struct cluster_node_addr *nodeaddr; | |
5698 | + struct cl_mem_join_msg *msg = (struct cl_mem_join_msg *) msgbuf; | |
5699 | + | |
5700 | + msg->cmd = CLUSTER_MEM_JOINREQ; | |
5701 | + msg->votes = votes; | |
5702 | + msg->expected_votes = cpu_to_le32(expected_votes); | |
5703 | + msg->major_version = cpu_to_le32(CNXMAN_MAJOR_VERSION); | |
5704 | + msg->minor_version = cpu_to_le32(CNXMAN_MINOR_VERSION); | |
5705 | + msg->patch_version = cpu_to_le32(CNXMAN_PATCH_VERSION); | |
5706 | + msg->config_version = cpu_to_le32(config_version); | |
5707 | + msg->addr_len = cpu_to_le32(address_length); | |
5708 | + strcpy(msg->clustername, cluster_name); | |
5709 | + | |
5710 | + /* Add our addresses */ | |
5711 | + list_for_each(addrlist, &us->addr_list) { | |
5712 | + nodeaddr = list_entry(addrlist, struct cluster_node_addr, list); | |
5713 | + | |
5714 | + memcpy(msgbuf + ptr, nodeaddr->addr, address_length); | |
5715 | + ptr += address_length; | |
5716 | + num_addr++; | |
5717 | + } | |
5718 | + msg->num_addr = cpu_to_le16(num_addr); | |
5719 | + | |
5720 | + /* And our name */ | |
5721 | + strcpy(msgbuf + ptr, nodename); | |
5722 | + ptr += strlen(nodename) + 1; | |
5723 | + | |
5724 | + return kcl_sendmsg(mem_socket, msgbuf, ptr, | |
5725 | + addr, addr_len, MSG_NOACK); | |
5726 | +} | |
5727 | + | |
5728 | +static int send_startack(struct sockaddr_cl *addr, int addr_len, int node_id) | |
5729 | +{ | |
5730 | + struct cl_mem_startack_msg msg; | |
5731 | + | |
5732 | + msg.cmd = CLUSTER_MEM_STARTACK; | |
5733 | + msg.generation = cpu_to_le32(cluster_generation); | |
5734 | + msg.node_id = cpu_to_le32(node_id); | |
5735 | + msg.highest_node_id = cpu_to_le32(get_highest_nodeid()); | |
5736 | + | |
5737 | + return kcl_sendmsg(mem_socket, &msg, sizeof (msg), addr, addr_len, 0); | |
5738 | +} | |
5739 | + | |
5740 | +static int send_newcluster() | |
5741 | +{ | |
5742 | + char buf[1]; | |
5743 | + | |
5744 | + buf[0] = CLUSTER_MEM_NEWCLUSTER; | |
5745 | + | |
5746 | + return kcl_sendmsg(mem_socket, buf, 1, NULL, 0, | |
5747 | + MSG_NOACK); | |
5748 | +} | |
5749 | + | |
5750 | +static int send_hello() | |
5751 | +{ | |
5752 | + struct cl_mem_hello_msg hello_msg; | |
5753 | + int status; | |
5754 | + | |
5755 | + hello_msg.cmd = CLUSTER_MEM_HELLO; | |
5756 | + hello_msg.members = cpu_to_le16(cluster_members); | |
5757 | + hello_msg.flags = 0; | |
5758 | + hello_msg.generation = cpu_to_le32(cluster_generation); | |
5759 | + | |
5760 | + status = | |
5761 | + kcl_sendmsg(mem_socket, &hello_msg, sizeof (hello_msg), NULL, 0, | |
5762 | + MSG_NOACK | MSG_ALLINT); | |
5763 | + | |
5764 | + last_hello = jiffies; | |
5765 | + | |
5766 | + return status; | |
5767 | +} | |
5768 | + | |
5769 | +/* This is a special HELLO message that requires an ACK. clients in transition | |
5770 | + * send these to the master to check it is till alive. if it does not ACK then | |
5771 | + * cnxman will signal it dead and we can restart the transition */ | |
5772 | +static int send_master_hello() | |
5773 | +{ | |
5774 | + struct cl_mem_hello_msg hello_msg; | |
5775 | + int status; | |
5776 | + struct sockaddr_cl saddr; | |
5777 | + | |
5778 | + hello_msg.cmd = CLUSTER_MEM_HELLO; | |
5779 | + hello_msg.members = cpu_to_le16(cluster_members); | |
5780 | + hello_msg.flags = 1; | |
5781 | + hello_msg.generation = cpu_to_le32(cluster_generation); | |
5782 | + | |
5783 | + saddr.scl_family = AF_CLUSTER; | |
5784 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5785 | + saddr.scl_nodeid = master_node->node_id; | |
5786 | + status = | |
5787 | + kcl_sendmsg(mem_socket, &hello_msg, sizeof (hello_msg), | |
5788 | + &saddr, sizeof (saddr), 0); | |
5789 | + | |
5790 | + last_hello = jiffies; | |
5791 | + | |
5792 | + return status; | |
5793 | +} | |
5794 | + | |
5795 | +/* Called when the transition timer has expired, meaning we sent a transition | |
5796 | + * message that was not ACKed */ | |
5797 | +static void trans_timer_expired(unsigned long arg) | |
5798 | +{ | |
5799 | + P_MEMB("Transition timer fired %ld\n", jiffies); | |
5800 | + | |
5801 | + set_bit(WAKE_FLAG_TRANSTIMER, &wake_flags); | |
5802 | + wake_up_process(membership_task); | |
5803 | +} | |
5804 | + | |
5805 | +static void hello_timer_expired(unsigned long arg) | |
5806 | +{ | |
5807 | + P_MEMB("Hello timer fired %ld\n", jiffies); | |
5808 | + | |
5809 | + mod_timer(&hello_timer, jiffies + cman_config.hello_timer * HZ); | |
5810 | + | |
5811 | + if (node_state >= TRANSITION) { | |
5812 | + wake_up_process(hello_task); | |
5813 | + } | |
5814 | +} | |
5815 | + | |
5816 | +static int wait_for_completion_barrier(void) | |
5817 | +{ | |
5818 | + int status; | |
5819 | + char barriername[MAX_BARRIER_NAME_LEN]; | |
5820 | + | |
5821 | + sprintf(barriername, MEMBERSHIP_BARRIER_NAME, cluster_generation); | |
5822 | + | |
5823 | + /* Make sure we all complete together */ | |
5824 | + P_MEMB("Waiting for completion barrier: %d members\n", cluster_members); | |
5825 | + if ((status = | |
5826 | + kcl_barrier_register(barriername, 0, cluster_members)) < 0) { | |
5827 | + printk(CMAN_NAME ": Error registering barrier: %d\n", status); | |
5828 | + return -1; | |
5829 | + } | |
5830 | + kcl_barrier_setattr(barriername, BARRIER_SETATTR_TIMEOUT, | |
5831 | + cman_config.transition_timeout); | |
5832 | + status = kcl_barrier_wait(barriername); | |
5833 | + kcl_barrier_delete(barriername); | |
5834 | + | |
5835 | + P_MEMB("Completion barrier reached : status = %d\n", status); | |
5836 | + return status; | |
5837 | +} | |
5838 | + | |
5839 | +/* Called at the end of a state transition when we are the master */ | |
5840 | +static int end_transition() | |
5841 | +{ | |
5842 | + struct cl_mem_endtrans_msg msg; | |
5843 | + int total_votes; | |
5844 | + int status; | |
5845 | + | |
5846 | + /* Cancel the timer */ | |
5847 | + del_timer(&transition_timer); | |
5848 | + | |
5849 | + confirm_joiner(); | |
5850 | + | |
5851 | + quorum = calculate_quorum(leavereason, 0, &total_votes); | |
5852 | + | |
5853 | + msg.cmd = CLUSTER_MEM_ENDTRANS; | |
5854 | + msg.quorum = cpu_to_le32(quorum); | |
5855 | + msg.generation = cpu_to_le32(++cluster_generation); | |
5856 | + msg.total_votes = cpu_to_le32(total_votes); | |
5857 | + if (joining_node && transitionreason == TRANS_NEWNODE) { | |
5858 | + msg.new_node_id = cpu_to_le32(joining_node->node_id); | |
5859 | + } | |
5860 | + else { | |
5861 | + msg.new_node_id = 0; | |
5862 | + } | |
5863 | + status = kcl_sendmsg(mem_socket, &msg, sizeof (msg), NULL, 0, 0); | |
5864 | + | |
5865 | + /* When that's all settled down, do the transition completion barrier */ | |
5866 | + kcl_wait_for_all_acks(); | |
5867 | + | |
5868 | + if (wait_for_completion_barrier() != 0) { | |
5869 | + P_MEMB("Barrier timed out - restart\n"); | |
5870 | + start_transition(TRANS_RESTART, us); | |
5871 | + return 0; | |
5872 | + } | |
5873 | + | |
5874 | + set_quorate(total_votes); | |
5875 | + | |
5876 | + notify_listeners(); | |
5877 | + reset_hello_time(); | |
5878 | + | |
5879 | + /* Tell any waiting barriers that we had a transition */ | |
5880 | + check_barrier_returns(); | |
5881 | + | |
5882 | + leavereason = 0; | |
5883 | + node_state = MEMBER; | |
5884 | + transition_end_time = jiffies; | |
5885 | + | |
5886 | + sm_member_update(cluster_is_quorate); | |
5887 | + | |
5888 | + return 0; | |
5889 | +} | |
5890 | + | |
5891 | +int send_reconfigure(int param, unsigned int value) | |
5892 | +{ | |
5893 | + char msgbuf[66]; | |
5894 | + struct cl_mem_reconfig_msg *msg = | |
5895 | + (struct cl_mem_reconfig_msg *) &msgbuf; | |
5896 | + | |
5897 | + if (param == RECONFIG_PARAM_EXPECTED_VOTES && expected_votes > value) | |
5898 | + expected_votes = value; | |
5899 | + | |
5900 | + msg->cmd = CLUSTER_MEM_RECONFIG; | |
5901 | + msg->param = param; | |
5902 | + msg->value = cpu_to_le32(value); | |
5903 | + | |
5904 | + return kcl_sendmsg(mem_socket, &msgbuf, sizeof (*msg), NULL, 0, 0); | |
5905 | +} | |
5906 | + | |
5907 | +static int send_joinack(char *addr, int addr_len, unsigned char acktype) | |
5908 | +{ | |
5909 | + struct cl_mem_joinack_msg msg; | |
5910 | + | |
5911 | + msg.cmd = CLUSTER_MEM_JOINACK; | |
5912 | + msg.acktype = acktype; | |
5913 | + | |
5914 | + return kcl_sendmsg(mem_socket, &msg, sizeof (msg), | |
5915 | + (struct sockaddr_cl *)addr, addr_len, MSG_NOACK); | |
5916 | +} | |
5917 | + | |
5918 | +/* Only send a leave message to one node in the cluster so that it can master | |
5919 | + * the state transition, otherwise we get a "thundering herd" of potential | |
5920 | + * masters fighting it out */ | |
5921 | +int send_leave(unsigned char flags) | |
5922 | +{ | |
5923 | + unsigned char msg[2]; | |
5924 | + struct sockaddr_cl saddr; | |
5925 | + struct cluster_node *node = NULL; | |
5926 | + int status; | |
5927 | + | |
5928 | + if (!mem_socket) | |
5929 | + return 0; | |
5930 | + | |
5931 | + saddr.scl_family = AF_CLUSTER; | |
5932 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5933 | + | |
5934 | + /* If we are in transition then use the current master */ | |
5935 | + if (node_state == TRANSITION) { | |
5936 | + node = master_node; | |
5937 | + } | |
5938 | + if (!node) { | |
5939 | + /* If we are the master or not in transition then pick a node | |
5940 | + * almost at random */ | |
5941 | + struct list_head *nodelist; | |
5942 | + | |
5943 | + down(&cluster_members_lock); | |
5944 | + list_for_each(nodelist, &cluster_members_list) { | |
5945 | + node = list_entry(nodelist, struct cluster_node, list); | |
5946 | + | |
5947 | + if (node->state == NODESTATE_MEMBER && !node->us) | |
5948 | + break; | |
5949 | + } | |
5950 | + up(&cluster_members_lock); | |
5951 | + } | |
5952 | + | |
5953 | + /* we are the only member of the cluster - there is no-one to tell */ | |
5954 | + if (node && !node->us) { | |
5955 | + saddr.scl_nodeid = node->node_id; | |
5956 | + | |
5957 | + P_MEMB("Sending LEAVE to %s\n", node->name); | |
5958 | + msg[0] = CLUSTER_MEM_LEAVE; | |
5959 | + msg[1] = flags; | |
5960 | + status = | |
5961 | + kcl_sendmsg(mem_socket, msg, 2, | |
5962 | + &saddr, sizeof (saddr), | |
5963 | + MSG_NOACK); | |
5964 | + | |
5965 | + if (status < 0) | |
5966 | + return status; | |
5967 | + } | |
5968 | + | |
5969 | + /* And exit */ | |
5970 | + node_state = LEFT_CLUSTER; | |
5971 | + wake_up_process(membership_task); | |
5972 | + return 0; | |
5973 | +} | |
5974 | + | |
5975 | +int send_kill(int nodeid) | |
5976 | +{ | |
5977 | + char killmsg; | |
5978 | + struct sockaddr_cl saddr; | |
5979 | + | |
5980 | + killmsg = CLUSTER_MEM_KILL; | |
5981 | + | |
5982 | + saddr.scl_family = AF_CLUSTER; | |
5983 | + saddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
5984 | + saddr.scl_nodeid = nodeid; | |
5985 | + return kcl_sendmsg(mem_socket, &killmsg, 1, &saddr, | |
5986 | + sizeof (struct sockaddr_cl), MSG_NOACK); | |
5987 | +} | |
5988 | + | |
5989 | +/* Process a message */ | |
5990 | +static int do_membership_packet(struct msghdr *msg, int len) | |
5991 | +{ | |
5992 | + int result = -1; | |
5993 | + unsigned char *buf = msg->msg_iov->iov_base; | |
5994 | + struct sockaddr_cl *saddr = msg->msg_name; | |
5995 | + struct cluster_node *node; | |
5996 | + | |
5997 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
5998 | + | |
5999 | + P_MEMB("got membership message : %s, from (%d) %s, len = %d\n", | |
6000 | + msgname(*buf), saddr->scl_nodeid, node ? node->name : "unknown", len); | |
6001 | + | |
6002 | + switch (*buf) { | |
6003 | + case CLUSTER_MEM_JOINREQ: | |
6004 | + result = do_process_joinreq(msg, len); | |
6005 | + break; | |
6006 | + | |
6007 | + case CLUSTER_MEM_LEAVE: | |
6008 | + if (we_are_a_cluster_member) | |
6009 | + result = do_process_leave(msg, len); | |
6010 | + break; | |
6011 | + | |
6012 | + case CLUSTER_MEM_HELLO: | |
6013 | + result = do_process_hello(msg, len); | |
6014 | + break; | |
6015 | + | |
6016 | + case CLUSTER_MEM_KILL: | |
6017 | + if (we_are_a_cluster_member) | |
6018 | + result = do_process_kill(msg, len); | |
6019 | + break; | |
6020 | + | |
6021 | + case CLUSTER_MEM_JOINCONF: | |
6022 | + if (node_state == JOINACK) { | |
6023 | + do_process_joinconf(msg, len); | |
6024 | + } | |
6025 | + break; | |
6026 | + | |
6027 | + case CLUSTER_MEM_CONFACK: | |
6028 | + if (node_state == MASTER && master_state == MASTER_CONFIRM) { | |
6029 | + end_transition(); | |
6030 | + } | |
6031 | + break; | |
6032 | + | |
6033 | + case CLUSTER_MEM_MASTERVIEW: | |
6034 | + if (node_state == TRANSITION) | |
6035 | + do_process_masterview(msg, len); | |
6036 | + break; | |
6037 | + | |
6038 | + case CLUSTER_MEM_JOINACK: | |
6039 | + if (node_state == JOINING || node_state == JOINWAIT) { | |
6040 | + do_process_joinack(msg, len); | |
6041 | + } | |
6042 | + break; | |
6043 | + case CLUSTER_MEM_RECONFIG: | |
6044 | + if (we_are_a_cluster_member) { | |
6045 | + do_process_reconfig(msg, len); | |
6046 | + } | |
6047 | + break; | |
6048 | + | |
6049 | + case CLUSTER_MEM_STARTTRANS: | |
6050 | + result = do_process_starttrans(msg, len); | |
6051 | + break; | |
6052 | + | |
6053 | + case CLUSTER_MEM_ENDTRANS: | |
6054 | + result = do_process_endtrans(msg, len); | |
6055 | + break; | |
6056 | + | |
6057 | + case CLUSTER_MEM_VIEWACK: | |
6058 | + result = do_process_viewack(msg, len); | |
6059 | + break; | |
6060 | + | |
6061 | + case CLUSTER_MEM_STARTACK: | |
6062 | + if (node_state == MASTER) | |
6063 | + result = do_process_startack(msg, len); | |
6064 | + break; | |
6065 | + | |
6066 | + case CLUSTER_MEM_NEWCLUSTER: | |
6067 | + result = do_process_newcluster(msg, len); | |
6068 | + break; | |
6069 | + | |
6070 | + case CLUSTER_MEM_NOMINATE: | |
6071 | + if (node_state != MASTER) | |
6072 | + result = do_process_nominate(msg, len); | |
6073 | + break; | |
6074 | + | |
6075 | + default: | |
6076 | + printk(KERN_ERR CMAN_NAME | |
6077 | + ": Unknown membership services message %d received\n", | |
6078 | + *buf); | |
6079 | + break; | |
6080 | + | |
6081 | + } | |
6082 | + return result; | |
6083 | +} | |
6084 | + | |
6085 | +/* Returns -ve to reject membership of the cluster 0 to accept membership +ve | |
6086 | + * to ignore request (node already joining) */ | |
6087 | +static int check_duplicate_node(char *name, struct msghdr *msg, int len) | |
6088 | +{ | |
6089 | + struct cluster_node *node; | |
6090 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *)msg->msg_name; | |
6091 | + char addr[address_length]; | |
6092 | + int addrlen; | |
6093 | + | |
6094 | + if (strlen(name) >= MAX_CLUSTER_MEMBER_NAME_LEN) | |
6095 | + return -3; | |
6096 | + | |
6097 | + /* See if we already have a cluster member with that name... */ | |
6098 | + node = find_node_by_name(name); | |
6099 | + if (node && node->state != NODESTATE_DEAD) { | |
6100 | + | |
6101 | + if ((node->state == NODESTATE_JOINING || | |
6102 | + node->state == NODESTATE_REMOTEMEMBER)) | |
6103 | + return +1; | |
6104 | + | |
6105 | + printk(KERN_WARNING CMAN_NAME | |
6106 | + ": Rejecting cluster membership application from %s - already have a node with that name\n", | |
6107 | + name); | |
6108 | + return -1; | |
6109 | + | |
6110 | + } | |
6111 | + | |
6112 | + /* Need to check the node's address too */ | |
6113 | + if (get_addr_from_temp_nodeid(saddr->scl_nodeid, addr, &addrlen) && | |
6114 | + (node = find_node_by_addr(addr, addrlen)) && | |
6115 | + node->state != NODESTATE_DEAD) { | |
6116 | + | |
6117 | + if ((node->state == NODESTATE_JOINING || | |
6118 | + node->state == NODESTATE_REMOTEMEMBER)) | |
6119 | + return +1; | |
6120 | + | |
6121 | + printk(KERN_WARNING CMAN_NAME | |
6122 | + ": Rejecting cluster membership application from %s - already have a node with that address\n", | |
6123 | + name); | |
6124 | + return -1; | |
6125 | + } | |
6126 | + return 0; | |
6127 | +} | |
6128 | + | |
6129 | +/* Start the state transition */ | |
6130 | +static int start_transition(unsigned char reason, struct cluster_node *node) | |
6131 | +{ | |
6132 | + char *startbuf = scratchbuf; | |
6133 | + struct cl_mem_starttrans_msg *msg = | |
6134 | + (struct cl_mem_starttrans_msg *) startbuf; | |
6135 | + | |
6136 | + P_MEMB("Start transition - reason = %d\n", reason); | |
6137 | + | |
6138 | + /* If this is a restart then zero the counters */ | |
6139 | + if (reason == TRANS_RESTART) { | |
6140 | + agreeing_nodes = 0; | |
6141 | + dissenting_nodes = 0; | |
6142 | + if (node_opinion) { | |
6143 | + kfree(node_opinion); | |
6144 | + node_opinion = NULL; | |
6145 | + } | |
6146 | + responses_collected = 0; | |
6147 | + } | |
6148 | + | |
6149 | + /* If we have timed out too many times then just die */ | |
6150 | + if (reason == TRANS_RESTART | |
6151 | + && ++transition_restarts > cman_config.transition_restarts) { | |
6152 | + printk(KERN_WARNING CMAN_NAME | |
6153 | + ": too many transition restarts - will die\n"); | |
6154 | + send_leave(CLUSTER_LEAVEFLAG_INCONSISTENT); | |
6155 | + node_state = LEFT_CLUSTER; | |
6156 | + quit_threads = 1; | |
6157 | + wake_up_process(membership_task); | |
6158 | + wake_up_interruptible(&cnxman_waitq); | |
6159 | + return 0; | |
6160 | + } | |
6161 | + if (reason != TRANS_RESTART) | |
6162 | + transition_restarts = 0; | |
6163 | + | |
6164 | + /* Only keep the original state transition reason in the global | |
6165 | + * variable. */ | |
6166 | + if (reason != TRANS_ANOTHERREMNODE && reason != TRANS_NEWMASTER && | |
6167 | + reason != TRANS_RESTART && reason != TRANS_DEADMASTER) | |
6168 | + transitionreason = reason; | |
6169 | + | |
6170 | + /* Save the info of the requesting node */ | |
6171 | + if (reason == TRANS_NEWNODE) | |
6172 | + joining_node = node; | |
6173 | + | |
6174 | + node_state = MASTER; | |
6175 | + master_state = MASTER_START; | |
6176 | + responses_collected = 0; | |
6177 | + responses_expected = cluster_members - 1; | |
6178 | + | |
6179 | + /* If we are on our own then just do it */ | |
6180 | + if (responses_expected == 0) { | |
6181 | + P_MEMB("We are on our own...lonely here\n"); | |
6182 | + responses_collected--; | |
6183 | + do_process_startack(NULL, 0); | |
6184 | + } | |
6185 | + else { | |
6186 | + int ptr = sizeof (struct cl_mem_starttrans_msg); | |
6187 | + struct list_head *addrlist; | |
6188 | + unsigned short num_addrs = 0; | |
6189 | + int flags = 0; | |
6190 | + | |
6191 | + /* Send the STARTTRANS message */ | |
6192 | + msg->cmd = CLUSTER_MEM_STARTTRANS; | |
6193 | + msg->reason = reason; | |
6194 | + msg->votes = node->votes; | |
6195 | + msg->expected_votes = cpu_to_le32(node->expected_votes); | |
6196 | + msg->generation = cpu_to_le32(++cluster_generation); | |
6197 | + msg->nodeid = cpu_to_le32(node->node_id); | |
6198 | + | |
6199 | + if (reason == TRANS_NEWNODE) { | |
6200 | + /* Add the addresses */ | |
6201 | + list_for_each(addrlist, &node->addr_list) { | |
6202 | + struct cluster_node_addr *nodeaddr = | |
6203 | + list_entry(addrlist, | |
6204 | + struct cluster_node_addr, list); | |
6205 | + | |
6206 | + memcpy(startbuf + ptr, nodeaddr->addr, | |
6207 | + address_length); | |
6208 | + ptr += address_length; | |
6209 | + num_addrs++; | |
6210 | + } | |
6211 | + | |
6212 | + /* And the name */ | |
6213 | + strcpy(startbuf + ptr, node->name); | |
6214 | + ptr += strlen(node->name) + 1; | |
6215 | + } | |
6216 | + | |
6217 | + /* If another node died then we must queue the STARTTRANS | |
6218 | + * messages so that membershipd can carry on processing the | |
6219 | + * other replies */ | |
6220 | + if (reason == TRANS_ANOTHERREMNODE) | |
6221 | + flags |= MSG_QUEUE; | |
6222 | + | |
6223 | + msg->num_addrs = cpu_to_le16(num_addrs); | |
6224 | + kcl_sendmsg(mem_socket, msg, ptr, NULL, 0, flags); | |
6225 | + } | |
6226 | + /* Set a timer in case we don't get 'em all back */ | |
6227 | + mod_timer(&transition_timer, | |
6228 | + jiffies + cman_config.transition_timeout * HZ); | |
6229 | + return 0; | |
6230 | +} | |
6231 | + | |
6232 | +/* A node has died - decide what to do */ | |
6233 | +void a_node_just_died(struct cluster_node *node) | |
6234 | +{ | |
6235 | + /* If we are not in the context of kmembershipd then stick it on the | |
6236 | + * list and wake it */ | |
6237 | + if (current != membership_task) { | |
6238 | + struct cl_new_dead_node *newnode = | |
6239 | + kmalloc(sizeof (struct cl_new_dead_node), GFP_KERNEL); | |
6240 | + if (!newnode) | |
6241 | + return; | |
6242 | + newnode->node = node; | |
6243 | + down(&new_dead_node_lock); | |
6244 | + list_add_tail(&newnode->list, &new_dead_node_list); | |
6245 | + set_bit(WAKE_FLAG_DEADNODE, &wake_flags); | |
6246 | + up(&new_dead_node_lock); | |
6247 | + wake_up_process(membership_task); | |
6248 | + P_MEMB("Passing dead node %s onto kmembershipd\n", node->name); | |
6249 | + return; | |
6250 | + } | |
6251 | + | |
6252 | + /* Remove it */ | |
6253 | + down(&cluster_members_lock); | |
6254 | + if (node->state == NODESTATE_MEMBER) | |
6255 | + cluster_members--; | |
6256 | + node->state = NODESTATE_DEAD; | |
6257 | + up(&cluster_members_lock); | |
6258 | + | |
6259 | + /* Notify listeners */ | |
6260 | + notify_kernel_listeners(DIED, (long) node->node_id); | |
6261 | + | |
6262 | + /* If we are in normal operation then become master and initiate a | |
6263 | + * state-transition */ | |
6264 | + if (node_state == MEMBER) { | |
6265 | + start_transition(TRANS_REMNODE, node); | |
6266 | + return; | |
6267 | + } | |
6268 | + | |
6269 | + /* If we are a slave in transition then see if it's the master that has | |
6270 | + * failed. If not then ignore it. If it /is/ the master then elect a | |
6271 | + * new one */ | |
6272 | + if (node_state == TRANSITION) { | |
6273 | + if (master_node == node) { | |
6274 | + if (elect_master(&node)) { | |
6275 | + del_timer(&transition_timer); | |
6276 | + node_state = MASTER; | |
6277 | + | |
6278 | + start_transition(TRANS_DEADMASTER, master_node); | |
6279 | + } | |
6280 | + else { | |
6281 | + /* Someone else can be in charge - phew! */ | |
6282 | + } | |
6283 | + } | |
6284 | + return; | |
6285 | + } | |
6286 | + | |
6287 | + /* If we are the master then we need to start the transition all over | |
6288 | + * again */ | |
6289 | + if (node_state == MASTER) { | |
6290 | + /* Cancel timer */ | |
6291 | + del_timer(&transition_timer); | |
6292 | + | |
6293 | + /* Restart the transition */ | |
6294 | + start_transition(TRANS_ANOTHERREMNODE, node); | |
6295 | + transition_restarts = 0; | |
6296 | + return; | |
6297 | + } | |
6298 | +} | |
6299 | + | |
6300 | +/* | |
6301 | + * Build up and send a set of messages consisting of the whole cluster view. | |
6302 | + * The first byte is the command (cmd as passed in), the second is a flag byte: | |
6303 | + * bit 0 is set in the first message, bit 1 in the last (NOTE both may be set if | |
6304 | + * this is the only message sent The rest is a set of packed node entries, which | |
6305 | + * are NOT split over packets. */ | |
6306 | +static int send_cluster_view(unsigned char cmd, struct sockaddr_cl *saddr, | |
6307 | + unsigned int flags) | |
6308 | +{ | |
6309 | + int ptr = 2; | |
6310 | + int len; | |
6311 | + int status = 0; | |
6312 | + int last_node_start = 2; | |
6313 | + unsigned char first_packet_flag = 1; | |
6314 | + struct list_head *nodelist; | |
6315 | + struct list_head *temp; | |
6316 | + struct cluster_node *node; | |
6317 | + char *message = scratchbuf; | |
6318 | + | |
6319 | + message[0] = cmd; | |
6320 | + | |
6321 | + down(&cluster_members_lock); | |
6322 | + list_for_each_safe(nodelist, temp, &cluster_members_list) { | |
6323 | + node = list_entry(nodelist, struct cluster_node, list); | |
6324 | + | |
6325 | + if (node->state == NODESTATE_MEMBER) { | |
6326 | + unsigned int evotes; | |
6327 | + unsigned int node_id; | |
6328 | + unsigned short num_addrs = 0; | |
6329 | + unsigned short num_addrs_le; | |
6330 | + struct list_head *addrlist; | |
6331 | + | |
6332 | + last_node_start = ptr; | |
6333 | + | |
6334 | + message[ptr++] = len = strlen(node->name); | |
6335 | + strcpy(&message[ptr], node->name); | |
6336 | + ptr += len; | |
6337 | + | |
6338 | + /* Count the number of addresses this node has */ | |
6339 | + list_for_each(addrlist, &node->addr_list) { | |
6340 | + num_addrs++; | |
6341 | + } | |
6342 | + | |
6343 | + num_addrs_le = cpu_to_le16(num_addrs); | |
6344 | + memcpy(&message[ptr], &num_addrs_le, sizeof (short)); | |
6345 | + ptr += sizeof (short); | |
6346 | + | |
6347 | + /* Pack em in */ | |
6348 | + list_for_each(addrlist, &node->addr_list) { | |
6349 | + | |
6350 | + struct cluster_node_addr *nodeaddr = | |
6351 | + list_entry(addrlist, | |
6352 | + struct cluster_node_addr, list); | |
6353 | + | |
6354 | + memcpy(&message[ptr], nodeaddr->addr, | |
6355 | + address_length); | |
6356 | + ptr += address_length; | |
6357 | + } | |
6358 | + | |
6359 | + message[ptr++] = node->votes; | |
6360 | + | |
6361 | + evotes = cpu_to_le32(node->expected_votes); | |
6362 | + memcpy(&message[ptr], &evotes, sizeof (int)); | |
6363 | + ptr += sizeof (int); | |
6364 | + | |
6365 | + node_id = cpu_to_le32(node->node_id); | |
6366 | + memcpy(&message[ptr], &node_id, sizeof (int)); | |
6367 | + ptr += sizeof (int); | |
6368 | + | |
6369 | + /* If the block is full then send it */ | |
6370 | + if (ptr > MAX_CLUSTER_MESSAGE) { | |
6371 | + message[1] = first_packet_flag; | |
6372 | + | |
6373 | + up(&cluster_members_lock); | |
6374 | + status = | |
6375 | + kcl_sendmsg(mem_socket, message, | |
6376 | + last_node_start, saddr, | |
6377 | + saddr ? sizeof (struct sockaddr_cl) : 0, | |
6378 | + flags); | |
6379 | + | |
6380 | + if (status < 0) | |
6381 | + goto send_fail; | |
6382 | + | |
6383 | + down(&cluster_members_lock); | |
6384 | + | |
6385 | + first_packet_flag = 0; | |
6386 | + /* Copy the overflow back to the start of the | |
6387 | + * buffer for the next send */ | |
6388 | + memcpy(&message[2], &message[last_node_start], | |
6389 | + ptr - last_node_start); | |
6390 | + ptr = ptr - last_node_start + 2; | |
6391 | + } | |
6392 | + } | |
6393 | + } | |
6394 | + | |
6395 | + up(&cluster_members_lock); | |
6396 | + | |
6397 | + message[1] = first_packet_flag | 2; /* The last may also be first */ | |
6398 | + status = kcl_sendmsg(mem_socket, message, ptr, | |
6399 | + saddr, saddr ? sizeof (struct sockaddr_cl) : 0, | |
6400 | + flags); | |
6401 | + send_fail: | |
6402 | + | |
6403 | + return status; | |
6404 | +} | |
6405 | + | |
6406 | +/* Make the JOINING node into a MEMBER */ | |
6407 | +static void confirm_joiner() | |
6408 | +{ | |
6409 | + if (joining_node && joining_node->state == NODESTATE_JOINING) { | |
6410 | + down(&cluster_members_lock); | |
6411 | + joining_node->state = NODESTATE_MEMBER; | |
6412 | + cluster_members++; | |
6413 | + up(&cluster_members_lock); | |
6414 | + } | |
6415 | + remove_temp_nodeid(joining_temp_nodeid); | |
6416 | + joining_temp_nodeid = 0; | |
6417 | +} | |
6418 | + | |
6419 | +/* Reset HELLO timers for all nodes We do this after a state-transition as we | |
6420 | + * have had HELLOS disabled during the transition and if we don't do this the | |
6421 | + * nodes will go on an uncontrolled culling-spree afterwards */ | |
6422 | +static void reset_hello_time() | |
6423 | +{ | |
6424 | + struct list_head *nodelist; | |
6425 | + struct cluster_node *node; | |
6426 | + | |
6427 | + down(&cluster_members_lock); | |
6428 | + list_for_each(nodelist, &cluster_members_list) { | |
6429 | + node = list_entry(nodelist, struct cluster_node, list); | |
6430 | + | |
6431 | + if (node->state == NODESTATE_MEMBER) { | |
6432 | + node->last_hello = jiffies; | |
6433 | + } | |
6434 | + | |
6435 | + } | |
6436 | + up(&cluster_members_lock); | |
6437 | +} | |
6438 | + | |
6439 | +/* Calculate the new quorum and return the value. do *not* set it in here as | |
6440 | + * cnxman calls this to check if a new expected_votes value is valid. It | |
6441 | + * (optionally) returns the total number of votes in the cluster */ | |
6442 | +int calculate_quorum(int allow_decrease, int max_expected, int *ret_total_votes) | |
6443 | +{ | |
6444 | + struct list_head *nodelist; | |
6445 | + struct cluster_node *node; | |
6446 | + unsigned int total_votes = 0; | |
6447 | + unsigned int highest_expected = 0; | |
6448 | + unsigned int newquorum, q1, q2; | |
6449 | + | |
6450 | + down(&cluster_members_lock); | |
6451 | + list_for_each(nodelist, &cluster_members_list) { | |
6452 | + node = list_entry(nodelist, struct cluster_node, list); | |
6453 | + | |
6454 | + if (node->state == NODESTATE_MEMBER) { | |
6455 | + highest_expected = | |
6456 | + max(highest_expected, node->expected_votes); | |
6457 | + total_votes += node->votes; | |
6458 | + } | |
6459 | + } | |
6460 | + up(&cluster_members_lock); | |
6461 | + if (quorum_device && quorum_device->state == NODESTATE_MEMBER) | |
6462 | + total_votes += quorum_device->votes; | |
6463 | + | |
6464 | + if (max_expected > 0) | |
6465 | + highest_expected = max_expected; | |
6466 | + | |
6467 | + /* This quorum calculation is taken from the OpenVMS Cluster Systems | |
6468 | + * manual, but, then, you guessed that didn't you */ | |
6469 | + q1 = (highest_expected + 2) / 2; | |
6470 | + q2 = (total_votes + 2) / 2; | |
6471 | + newquorum = max(q1, q2); | |
6472 | + | |
6473 | + /* Normally quorum never decreases but the system administrator can | |
6474 | + * force it down by setting expected votes to a maximum value */ | |
6475 | + if (!allow_decrease) | |
6476 | + newquorum = max(quorum, newquorum); | |
6477 | + | |
6478 | + /* The special two_node mode allows each of the two nodes to retain | |
6479 | + * quorum if the other fails. Only one of the two should live past | |
6480 | + * fencing (as both nodes try to fence each other in split-brain.) */ | |
6481 | + if (two_node) | |
6482 | + newquorum = 1; | |
6483 | + | |
6484 | + if (ret_total_votes) | |
6485 | + *ret_total_votes = total_votes; | |
6486 | + return newquorum; | |
6487 | +} | |
6488 | + | |
6489 | +/* Recalculate cluster quorum, set quorate and notify changes */ | |
6490 | +void recalculate_quorum(int allow_decrease) | |
6491 | +{ | |
6492 | + int total_votes; | |
6493 | + | |
6494 | + quorum = calculate_quorum(allow_decrease, 0, &total_votes); | |
6495 | + set_quorate(total_votes); | |
6496 | + notify_listeners(); | |
6497 | +} | |
6498 | + | |
6499 | +/* Add new node address to an existing node */ | |
6500 | +int add_node_address(struct cluster_node *node, unsigned char *addr, int len) | |
6501 | +{ | |
6502 | + struct cluster_node_addr *newaddr; | |
6503 | + | |
6504 | + newaddr = kmalloc(sizeof (struct cluster_node_addr), GFP_KERNEL); | |
6505 | + if (!newaddr) | |
6506 | + return -1; | |
6507 | + | |
6508 | + memcpy(newaddr->addr, addr, len); | |
6509 | + newaddr->addr_len = len; | |
6510 | + list_add_tail(&newaddr->list, &node->addr_list); | |
6511 | + | |
6512 | + return 0; | |
6513 | +} | |
6514 | + | |
6515 | +static struct cluster_node *add_new_node(char *name, unsigned char votes, | |
6516 | + unsigned int expected_votes, | |
6517 | + int node_id, int state) | |
6518 | +{ | |
6519 | + struct cluster_node *newnode; | |
6520 | + | |
6521 | + /* Look for a dead node with this name */ | |
6522 | + newnode = find_node_by_name(name); | |
6523 | + | |
6524 | + /* Is it already joining */ | |
6525 | + if (newnode && newnode->state == NODESTATE_JOINING) | |
6526 | + return NULL; | |
6527 | + | |
6528 | + /* Update existing information */ | |
6529 | + if (newnode && newnode->state == NODESTATE_DEAD) { | |
6530 | + newnode->last_hello = jiffies; | |
6531 | + newnode->votes = votes; | |
6532 | + newnode->expected_votes = expected_votes; | |
6533 | + newnode->state = state; | |
6534 | + newnode->us = 0; | |
6535 | + newnode->leave_reason = 0; | |
6536 | + newnode->last_seq_recv = 0; | |
6537 | + newnode->last_seq_acked = 0; | |
6538 | + newnode->last_seq_sent = 0; | |
6539 | + newnode->incarnation++; | |
6540 | + /* Don't overwrite the node ID */ | |
6541 | + | |
6542 | + if (state == NODESTATE_MEMBER) { | |
6543 | + down(&cluster_members_lock); | |
6544 | + cluster_members++; | |
6545 | + up(&cluster_members_lock); | |
6546 | + } | |
6547 | + | |
6548 | + printk(KERN_INFO CMAN_NAME ": node %s rejoining\n", name); | |
6549 | + return newnode; | |
6550 | + } | |
6551 | + | |
6552 | + newnode = kmalloc(sizeof (struct cluster_node), GFP_KERNEL); | |
6553 | + if (!newnode) | |
6554 | + goto alloc_err; | |
6555 | + | |
6556 | + memset(newnode, 0, sizeof (struct cluster_node)); | |
6557 | + newnode->name = kmalloc(strlen(name) + 1, GFP_KERNEL); | |
6558 | + if (!newnode->name) | |
6559 | + goto alloc_err1; | |
6560 | + | |
6561 | + strcpy(newnode->name, name); | |
6562 | + newnode->last_hello = jiffies; | |
6563 | + newnode->votes = votes; | |
6564 | + newnode->expected_votes = expected_votes; | |
6565 | + newnode->state = state; | |
6566 | + newnode->node_id = node_id; | |
6567 | + newnode->us = 0; | |
6568 | + newnode->leave_reason = 0; | |
6569 | + newnode->last_seq_recv = 0; | |
6570 | + newnode->last_seq_acked = 0; | |
6571 | + newnode->last_seq_sent = 0; | |
6572 | + newnode->incarnation = 0; | |
6573 | + INIT_LIST_HEAD(&newnode->addr_list); | |
6574 | + set_nodeid(newnode, node_id); | |
6575 | + | |
6576 | + /* Add the new node to the list */ | |
6577 | + down(&cluster_members_lock); | |
6578 | + list_add(&newnode->list, &cluster_members_list); | |
6579 | + if (state == NODESTATE_MEMBER) | |
6580 | + cluster_members++; | |
6581 | + up(&cluster_members_lock); | |
6582 | + | |
6583 | + printk(KERN_INFO CMAN_NAME ": got node %s\n", name); | |
6584 | + return newnode; | |
6585 | + | |
6586 | + alloc_err1: | |
6587 | + kfree(newnode); | |
6588 | + alloc_err: | |
6589 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
6590 | + | |
6591 | + printk(KERN_CRIT CMAN_NAME | |
6592 | + ": Cannot allocate memory for new cluster node %s\n", name); | |
6593 | + | |
6594 | + panic("cluster memory allocation failed"); | |
6595 | + | |
6596 | + return NULL; | |
6597 | +} | |
6598 | + | |
6599 | +/* Remove node from a STARTTRANS message */ | |
6600 | +static struct cluster_node *remove_node(int nodeid) | |
6601 | +{ | |
6602 | + struct cluster_node *node = find_node_by_nodeid(nodeid); | |
6603 | + | |
6604 | + if (node && node->state == NODESTATE_MEMBER) { | |
6605 | + P_MEMB("starttrans removes node %s\n", node->name); | |
6606 | + down(&cluster_members_lock); | |
6607 | + node->state = NODESTATE_DEAD; | |
6608 | + cluster_members--; | |
6609 | + up(&cluster_members_lock); | |
6610 | + | |
6611 | + notify_kernel_listeners(DIED, (long) nodeid); | |
6612 | + | |
6613 | + /* If this node is us then go quietly */ | |
6614 | + if (node->us) { | |
6615 | + printk(KERN_INFO CMAN_NAME | |
6616 | + ": killed by STARTTRANS or NOMINATE\n"); | |
6617 | + quit_threads = 1; | |
6618 | + wake_up_process(membership_task); | |
6619 | + wake_up_interruptible(&cnxman_waitq); | |
6620 | + } | |
6621 | + } | |
6622 | + return node; | |
6623 | +} | |
6624 | + | |
6625 | +/* Add a node from a STARTTRANS or NOMINATE message */ | |
6626 | +static void add_node_from_starttrans(struct msghdr *msg, int len) | |
6627 | +{ | |
6628 | + /* Add the new node but don't fill in the ID until the master has | |
6629 | + * confirmed it */ | |
6630 | + struct cl_mem_starttrans_msg *startmsg = | |
6631 | + (struct cl_mem_starttrans_msg *) msg->msg_iov->iov_base; | |
6632 | + char *msgbuf = (char *) msg->msg_iov->iov_base; | |
6633 | + int ptr = sizeof (struct cl_mem_starttrans_msg); | |
6634 | + char *name = | |
6635 | + msgbuf + ptr + le16_to_cpu(startmsg->num_addrs) * address_length; | |
6636 | + int i; | |
6637 | + | |
6638 | + joining_node = add_new_node(name, startmsg->votes, | |
6639 | + le32_to_cpu(startmsg->expected_votes), | |
6640 | + 0, NODESTATE_JOINING); | |
6641 | + | |
6642 | + /* add_new_node returns NULL if the node already exists */ | |
6643 | + if (!joining_node) | |
6644 | + joining_node = find_node_by_name(name); | |
6645 | + | |
6646 | + /* Add the node's addresses */ | |
6647 | + if (list_empty(&joining_node->addr_list)) { | |
6648 | + for (i = 0; i < le16_to_cpu(startmsg->num_addrs); i++) { | |
6649 | + add_node_address(joining_node, msgbuf + ptr, address_length); | |
6650 | + ptr += address_length; | |
6651 | + } | |
6652 | + } | |
6653 | +} | |
6654 | + | |
6655 | +/* We have been nominated as master for a transition */ | |
6656 | +static int do_process_nominate(struct msghdr *msg, int len) | |
6657 | +{ | |
6658 | + struct cl_mem_starttrans_msg *startmsg = | |
6659 | + (struct cl_mem_starttrans_msg *)msg->msg_iov->iov_base; | |
6660 | + struct cluster_node *node = NULL; | |
6661 | + char *nodeaddr = msg->msg_iov->iov_base + sizeof(struct cl_mem_starttrans_msg); | |
6662 | + | |
6663 | + P_MEMB("nominate reason is %d\n", startmsg->reason); | |
6664 | + | |
6665 | + if (startmsg->reason == TRANS_REMNODE) { | |
6666 | + node = remove_node(le32_to_cpu(startmsg->nodeid)); | |
6667 | + } | |
6668 | + | |
6669 | + if (startmsg->reason == TRANS_NEWNODE) { | |
6670 | + add_node_from_starttrans(msg, len); | |
6671 | + node = joining_node; | |
6672 | + /* Make sure we have a temp nodeid for the new node */ | |
6673 | + joining_temp_nodeid = new_temp_nodeid(nodeaddr, | |
6674 | + address_length); | |
6675 | + } | |
6676 | + | |
6677 | + /* This should be a TRANS_CHECK but start_transition needs some node | |
6678 | + * info */ | |
6679 | + if (node == NULL) | |
6680 | + node = us; | |
6681 | + start_transition(startmsg->reason, node); | |
6682 | + return 0; | |
6683 | +} | |
6684 | + | |
6685 | +/* Got a STARTACK response from a node */ | |
6686 | +static int do_process_startack(struct msghdr *msg, int len) | |
6687 | +{ | |
6688 | + if (node_state != MASTER && master_state != MASTER_START) { | |
6689 | + P_MEMB("Got StartACK when not in MASTER_STARTING substate\n"); | |
6690 | + return 0; | |
6691 | + } | |
6692 | + | |
6693 | + /* msg is NULL if we are called directly from start_transition */ | |
6694 | + if (msg) { | |
6695 | + struct cl_mem_startack_msg *ackmsg = msg->msg_iov->iov_base; | |
6696 | + | |
6697 | + /* Ignore any messages wil old generation numbers in them */ | |
6698 | + if (le32_to_cpu(ackmsg->generation) != cluster_generation) { | |
6699 | + P_MEMB("Got old generation START-ACK msg - ignoring\n"); | |
6700 | + return 0; | |
6701 | + } | |
6702 | + } | |
6703 | + | |
6704 | + /* If the node_id is non-zero then use it. */ | |
6705 | + if (transitionreason == TRANS_NEWNODE && joining_node && msg) { | |
6706 | + struct cl_mem_startack_msg *ackmsg = msg->msg_iov->iov_base; | |
6707 | + | |
6708 | + if (ackmsg->node_id) { | |
6709 | + set_nodeid(joining_node, le32_to_cpu(ackmsg->node_id)); | |
6710 | + } | |
6711 | + highest_nodeid = | |
6712 | + max(highest_nodeid, le32_to_cpu(ackmsg->highest_node_id)); | |
6713 | + P_MEMB("Node id = %d, highest node id = %d\n", | |
6714 | + le32_to_cpu(ackmsg->node_id), | |
6715 | + le32_to_cpu(ackmsg->highest_node_id)); | |
6716 | + } | |
6717 | + | |
6718 | + /* If we have all the responses in then move to the next stage */ | |
6719 | + if (++responses_collected == responses_expected) { | |
6720 | + | |
6721 | + /* If the new node has no node_id (ie nobody in the cluster has | |
6722 | + * heard of it before) then assign it a new one */ | |
6723 | + if (transitionreason == TRANS_NEWNODE && joining_node) { | |
6724 | + highest_nodeid = | |
6725 | + max(highest_nodeid, get_highest_nodeid()); | |
6726 | + if (joining_node->node_id == 0) { | |
6727 | + set_nodeid(joining_node, ++highest_nodeid); | |
6728 | + } | |
6729 | + P_MEMB("nodeIDs: new node: %d, highest: %d\n", | |
6730 | + joining_node->node_id, highest_nodeid); | |
6731 | + } | |
6732 | + | |
6733 | + /* Behave a little differently if we are on our own */ | |
6734 | + if (cluster_members == 1) { | |
6735 | + if (transitionreason == TRANS_NEWNODE) { | |
6736 | + /* If the cluster is just us then confirm at | |
6737 | + * once */ | |
6738 | + joinconf_count = 0; | |
6739 | + mod_timer(&transition_timer, | |
6740 | + jiffies + | |
6741 | + cman_config.joinconf_timeout * HZ); | |
6742 | + send_joinconf(); | |
6743 | + return 0; | |
6744 | + } | |
6745 | + else { /* Node leaving the cluster */ | |
6746 | + recalculate_quorum(leavereason); | |
6747 | + leavereason = 0; | |
6748 | + node_state = MEMBER; | |
6749 | + } | |
6750 | + } | |
6751 | + else { | |
6752 | + master_state = MASTER_COLLECT; | |
6753 | + responses_collected = 0; | |
6754 | + responses_expected = cluster_members - 1; | |
6755 | + P_MEMB("Sending MASTERVIEW: expecting %d responses\n", | |
6756 | + responses_expected); | |
6757 | + | |
6758 | + send_cluster_view(CLUSTER_MEM_MASTERVIEW, NULL, 0); | |
6759 | + | |
6760 | + /* Set a timer in case we don't get 'em all back */ | |
6761 | + mod_timer(&transition_timer, | |
6762 | + jiffies + | |
6763 | + cman_config.transition_timeout * HZ); | |
6764 | + } | |
6765 | + } | |
6766 | + return 0; | |
6767 | +} | |
6768 | + | |
6769 | +/* Got a VIEWACK response from a node */ | |
6770 | +static int do_process_viewack(struct msghdr *msg, int len) | |
6771 | +{ | |
6772 | + char *reply = msg->msg_iov->iov_base; | |
6773 | + struct sockaddr_cl *saddr = msg->msg_name; | |
6774 | + | |
6775 | + if (master_state != MASTER_COLLECT) { | |
6776 | + printk(KERN_INFO CMAN_NAME | |
6777 | + ": got VIEWACK while not in state transition\n"); | |
6778 | + return 0; | |
6779 | + } | |
6780 | + | |
6781 | + if (node_opinion == NULL) { | |
6782 | + node_opinion = | |
6783 | + kmalloc((1 + highest_nodeid) * sizeof (uint8_t), GFP_KERNEL); | |
6784 | + if (!node_opinion) { | |
6785 | + panic(": malloc agree/dissent failed\n"); | |
6786 | + } | |
6787 | + memset(node_opinion, 0, (1 + highest_nodeid) * sizeof (uint8_t)); | |
6788 | + } | |
6789 | + | |
6790 | + /* Keep a list of agreeing and dissenting nodes */ | |
6791 | + if (reply[1] == 1) { | |
6792 | + /* ACK - remote node agrees with me */ | |
6793 | + P_MEMB("Node agrees\n"); | |
6794 | + node_opinion[saddr->scl_nodeid] = OPINION_AGREE; | |
6795 | + agreeing_nodes++; | |
6796 | + } | |
6797 | + else { | |
6798 | + /* Remote node disagrees */ | |
6799 | + P_MEMB("Node disagrees\n"); | |
6800 | + node_opinion[saddr->scl_nodeid] = OPINION_DISAGREE; | |
6801 | + dissenting_nodes++; | |
6802 | + } | |
6803 | + | |
6804 | + P_MEMB("got %d responses, expected %d\n", responses_collected + 1, | |
6805 | + responses_expected); | |
6806 | + | |
6807 | + /* Are all the results in yet ? */ | |
6808 | + if (++responses_collected == responses_expected) { | |
6809 | + del_timer(&transition_timer); | |
6810 | + | |
6811 | + P_MEMB("The results are in: %d agree, %d dissent\n", | |
6812 | + agreeing_nodes, dissenting_nodes); | |
6813 | + | |
6814 | + if (agreeing_nodes > dissenting_nodes) { | |
6815 | + /* Kill dissenting nodes */ | |
6816 | + int i; | |
6817 | + | |
6818 | + for (i = 1; i <= responses_collected; i++) { | |
6819 | + if (node_opinion[i] == OPINION_DISAGREE) | |
6820 | + send_kill(i); | |
6821 | + } | |
6822 | + } | |
6823 | + else { | |
6824 | + /* We must leave the cluster as we are in a minority, | |
6825 | + * the rest of them can fight it out amongst | |
6826 | + * themselves. */ | |
6827 | + send_leave(CLUSTER_LEAVEFLAG_INCONSISTENT); | |
6828 | + | |
6829 | + agreeing_nodes = 0; | |
6830 | + dissenting_nodes = 0; | |
6831 | + kfree(node_opinion); | |
6832 | + node_opinion = NULL; | |
6833 | + node_state = LEFT_CLUSTER; | |
6834 | + quit_threads = 1; | |
6835 | + wake_up_process(membership_task); | |
6836 | + wake_up_interruptible(&cnxman_waitq); | |
6837 | + return -1; | |
6838 | + } | |
6839 | + | |
6840 | + /* Reset counters */ | |
6841 | + agreeing_nodes = 0; | |
6842 | + dissenting_nodes = 0; | |
6843 | + kfree(node_opinion); | |
6844 | + node_opinion = NULL; | |
6845 | + | |
6846 | + /* Confirm new node */ | |
6847 | + if (transitionreason == TRANS_NEWNODE) { | |
6848 | + mod_timer(&transition_timer, | |
6849 | + jiffies + cman_config.joinconf_timeout * HZ); | |
6850 | + joinconf_count = 0; | |
6851 | + send_joinconf(); | |
6852 | + return 0; | |
6853 | + } | |
6854 | + | |
6855 | + master_state = MASTER_COMPLETE; | |
6856 | + | |
6857 | + end_transition(); | |
6858 | + } | |
6859 | + | |
6860 | + return 0; | |
6861 | +} | |
6862 | + | |
6863 | +/* Got an ENDTRANS message */ | |
6864 | +static int do_process_endtrans(struct msghdr *msg, int len) | |
6865 | +{ | |
6866 | + struct cl_mem_endtrans_msg *endmsg = | |
6867 | + (struct cl_mem_endtrans_msg *) msg->msg_iov->iov_base; | |
6868 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *) msg->msg_name; | |
6869 | + | |
6870 | + /* Someone else's state transition */ | |
6871 | + if (node_state != TRANSITION && node_state != JOINACK) | |
6872 | + return 0; | |
6873 | + | |
6874 | + /* Check we got it from the MASTER node */ | |
6875 | + if (master_node && master_node->node_id != saddr->scl_nodeid) { | |
6876 | + printk(KERN_INFO | |
6877 | + "Got ENDTRANS from a node not the master: master: %d, sender: %d\n", | |
6878 | + master_node->node_id, saddr->scl_nodeid); | |
6879 | + return 0; | |
6880 | + } | |
6881 | + | |
6882 | + del_timer(&transition_timer); | |
6883 | + | |
6884 | + /* Set node ID on new node */ | |
6885 | + if (endmsg->new_node_id) { | |
6886 | + set_nodeid(joining_node, le32_to_cpu(endmsg->new_node_id)); | |
6887 | + P_MEMB("new node %s has ID %d\n", joining_node->name, | |
6888 | + joining_node->node_id); | |
6889 | + } | |
6890 | + | |
6891 | + node_state = TRANSITION_COMPLETE; | |
6892 | + | |
6893 | + /* Need to set this here or the barrier code will reject us if we've | |
6894 | + * just joined */ | |
6895 | + we_are_a_cluster_member = TRUE; | |
6896 | + | |
6897 | + confirm_joiner(); | |
6898 | + cluster_generation = le32_to_cpu(endmsg->generation); | |
6899 | + | |
6900 | + if (wait_for_completion_barrier() != 0) { | |
6901 | + P_MEMB("Barrier timed out - restart\n"); | |
6902 | + node_state = TRANSITION; | |
6903 | + mod_timer(&transition_timer, | |
6904 | + jiffies + cman_config.transition_timeout * HZ); | |
6905 | + return 0; | |
6906 | + } | |
6907 | + | |
6908 | + quorum = le32_to_cpu(endmsg->quorum); | |
6909 | + set_quorate(le32_to_cpu(endmsg->total_votes)); | |
6910 | + | |
6911 | + /* Tell any waiting barriers that we had a transition */ | |
6912 | + check_barrier_returns(); | |
6913 | + | |
6914 | + /* Clear the master node */ | |
6915 | + master_node = NULL; | |
6916 | + | |
6917 | + node_state = MEMBER; | |
6918 | + | |
6919 | + /* Notify other listeners that transition has completed */ | |
6920 | + notify_listeners(); | |
6921 | + reset_hello_time(); | |
6922 | + transition_end_time = jiffies; | |
6923 | + | |
6924 | + sm_member_update(cluster_is_quorate); | |
6925 | + return 0; | |
6926 | +} | |
6927 | + | |
6928 | +/* Turn a STARTTRANS message into NOMINATE and send it to the new master */ | |
6929 | +static int send_nominate(struct cl_mem_starttrans_msg *startmsg, int msglen, | |
6930 | + int nodeid) | |
6931 | +{ | |
6932 | + struct sockaddr_cl maddr; | |
6933 | + | |
6934 | + maddr.scl_port = CLUSTER_PORT_MEMBERSHIP; | |
6935 | + maddr.scl_family = AF_CLUSTER; | |
6936 | + maddr.scl_nodeid = nodeid; | |
6937 | + | |
6938 | + startmsg->cmd = CLUSTER_MEM_NOMINATE; | |
6939 | + return kcl_sendmsg(mem_socket, startmsg, msglen, | |
6940 | + &maddr, sizeof (maddr), 0); | |
6941 | +} | |
6942 | + | |
6943 | +/* Got a STARTTRANS message */ | |
6944 | +static int do_process_starttrans(struct msghdr *msg, int len) | |
6945 | +{ | |
6946 | + struct cl_mem_starttrans_msg *startmsg = | |
6947 | + (struct cl_mem_starttrans_msg *) msg->msg_iov->iov_base; | |
6948 | + struct sockaddr_cl *saddr = (struct sockaddr_cl *) msg->msg_name; | |
6949 | + struct cluster_node *node; | |
6950 | + unsigned int newgen = le32_to_cpu(startmsg->generation); | |
6951 | + | |
6952 | + /* Got a WHAT from WHOM? */ | |
6953 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
6954 | + if (!node || node->state != NODESTATE_MEMBER) | |
6955 | + return 0; | |
6956 | + | |
6957 | + /* Someone else's state transition */ | |
6958 | + if (node_state != MEMBER && | |
6959 | + node_state != TRANSITION && node_state != MASTER) | |
6960 | + return 0; | |
6961 | + | |
6962 | + /* Ignore old generation STARTTRANS messages */ | |
6963 | + if ((newgen < cluster_generation) || | |
6964 | + (newgen == 0xFFFFFFFF && cluster_generation == 0)) { | |
6965 | + P_MEMB("Ignoring STARTTRANS with old generation number\n"); | |
6966 | + return 0; | |
6967 | + } | |
6968 | + | |
6969 | + P_MEMB("Got starttrans: newgen = %d, oldgen = %d, reason = %d\n", | |
6970 | + newgen, cluster_generation, startmsg->reason); | |
6971 | + | |
6972 | + /* Up the generation number */ | |
6973 | + cluster_generation = newgen; | |
6974 | + | |
6975 | + /* If we are also a master then decide between us */ | |
6976 | + if (node_state == MASTER) { | |
6977 | + | |
6978 | + /* See if we really want the responsibility of being master */ | |
6979 | + if (elect_master(&node)) { | |
6980 | + | |
6981 | + /* I reluctantly accept this position of responsibility | |
6982 | + */ | |
6983 | + P_MEMB("I elected myself master\n"); | |
6984 | + | |
6985 | + /* start_transition will re-establish this */ | |
6986 | + del_timer(&transition_timer); | |
6987 | + | |
6988 | + start_transition(TRANS_NEWMASTER, node); | |
6989 | + return 0; | |
6990 | + } | |
6991 | + else { | |
6992 | + /* Back down */ | |
6993 | + P_MEMB("Backing down from MASTER status\n"); | |
6994 | + master_node = node; | |
6995 | + node_state = MEMBER; | |
6996 | + | |
6997 | + /* If we were bringing a new node into the cluster then | |
6998 | + * we will have to abandon that now and tell the new | |
6999 | + * node to try again later */ | |
7000 | + if (transitionreason == TRANS_NEWNODE && joining_node) { | |
7001 | + struct cluster_node_addr *first_addr = | |
7002 | + (struct cluster_node_addr *) joining_node-> | |
7003 | + addr_list.next; | |
7004 | + | |
7005 | + P_MEMB("Postponing membership of node %s\n", | |
7006 | + joining_node->name); | |
7007 | + send_joinack(first_addr->addr, address_length, | |
7008 | + JOINACK_TYPE_WAIT); | |
7009 | + | |
7010 | + /* Not dead, just sleeping */ | |
7011 | + joining_node->state = NODESTATE_DEAD; | |
7012 | + joining_node = NULL; | |
7013 | + } | |
7014 | + | |
7015 | + /* If the new master is not us OR the node we just got | |
7016 | + * the STARTTRANS from then make sure it knows it has | |
7017 | + * to be master */ | |
7018 | + if (saddr->scl_nodeid != node->node_id) { | |
7019 | + send_nominate(startmsg, len, node->node_id); | |
7020 | + return 0; | |
7021 | + } | |
7022 | + | |
7023 | + /* Fall through into MEMBER code below if we are | |
7024 | + * obeying the STARTTRANS we just received */ | |
7025 | + } | |
7026 | + } | |
7027 | + | |
7028 | + /* Do non-MASTER STARTTRANS bits */ | |
7029 | + if (node_state == MEMBER) { | |
7030 | + int ptr = sizeof (struct cl_mem_starttrans_msg); | |
7031 | + int node_id = 0; | |
7032 | + | |
7033 | + P_MEMB("Normal transition start\n"); | |
7034 | + | |
7035 | + /* If the master is adding a new node and we know it's node ID | |
7036 | + * then ACK with it. */ | |
7037 | + if (startmsg->reason == TRANS_NEWNODE) { | |
7038 | + struct cluster_node *node = | |
7039 | + find_node_by_addr((char *) startmsg + ptr, | |
7040 | + address_length); | |
7041 | + if (node) | |
7042 | + node_id = node->node_id; | |
7043 | + } | |
7044 | + | |
7045 | + /* Save the master info */ | |
7046 | + master_node = find_node_by_nodeid(saddr->scl_nodeid); | |
7047 | + node_state = TRANSITION; | |
7048 | + | |
7049 | + if (startmsg->reason == TRANS_NEWNODE) { | |
7050 | + add_node_from_starttrans(msg, len); | |
7051 | + } | |
7052 | + | |
7053 | + if (startmsg->reason == TRANS_REMNODE || | |
7054 | + startmsg->reason == TRANS_ANOTHERREMNODE) { | |
7055 | + remove_node(le32_to_cpu(startmsg->nodeid)); | |
7056 | + } | |
7057 | + | |
7058 | + send_startack(saddr, msg->msg_namelen, | |
7059 | + node_id); | |
7060 | + | |
7061 | + /* Establish timer in case the master dies */ | |
7062 | + mod_timer(&transition_timer, | |
7063 | + jiffies + cman_config.transition_timeout * HZ); | |
7064 | + | |
7065 | + return 0; | |
7066 | + } | |
7067 | + | |
7068 | + /* We are in transition but this may be a restart */ | |
7069 | + if (node_state == TRANSITION) { | |
7070 | + | |
7071 | + master_node = find_node_by_nodeid(saddr->scl_nodeid); | |
7072 | + send_startack(saddr, msg->msg_namelen, 0); | |
7073 | + | |
7074 | + /* Is it a new joining node ? This happens if a master is | |
7075 | + * usurped */ | |
7076 | + if (startmsg->reason == TRANS_NEWNODE) { | |
7077 | + struct cluster_node *oldjoin = joining_node; | |
7078 | + | |
7079 | + add_node_from_starttrans(msg, len); | |
7080 | + | |
7081 | + /* If this is a different node joining than the one we | |
7082 | + * were previously joining (probably cos the master is | |
7083 | + * a nominated one) then mark our "old" joiner as DEAD. | |
7084 | + * The original master will already have told the node | |
7085 | + * to go back into JOINWAIT state */ | |
7086 | + if (oldjoin && oldjoin != joining_node | |
7087 | + && oldjoin->state == NODESTATE_JOINING) | |
7088 | + oldjoin->state = NODESTATE_DEAD; | |
7089 | + } | |
7090 | + | |
7091 | + /* Is it a new master node? */ | |
7092 | + if (startmsg->reason == TRANS_NEWMASTER || | |
7093 | + startmsg->reason == TRANS_DEADMASTER) { | |
7094 | + P_MEMB("starttrans %s, node=%d\n", | |
7095 | + startmsg->reason == | |
7096 | + TRANS_NEWMASTER ? "NEWMASTER" : "DEADMASTER", | |
7097 | + le32_to_cpu(startmsg->nodeid)); | |
7098 | + | |
7099 | + /* If the old master has died then remove it */ | |
7100 | + node = | |
7101 | + find_node_by_nodeid(le32_to_cpu(startmsg->nodeid)); | |
7102 | + | |
7103 | + if (startmsg->reason == TRANS_DEADMASTER && | |
7104 | + node && node->state == NODESTATE_MEMBER) { | |
7105 | + down(&cluster_members_lock); | |
7106 | + node->state = NODESTATE_DEAD; | |
7107 | + cluster_members--; | |
7108 | + up(&cluster_members_lock); | |
7109 | + } | |
7110 | + | |
7111 | + /* Store new master */ | |
7112 | + master_node = find_node_by_nodeid(saddr->scl_nodeid); | |
7113 | + } | |
7114 | + | |
7115 | + /* Another node has died (or been killed) */ | |
7116 | + if (startmsg->reason == TRANS_ANOTHERREMNODE) { | |
7117 | + /* Remove new dead node */ | |
7118 | + node = | |
7119 | + find_node_by_nodeid(le32_to_cpu(startmsg->nodeid)); | |
7120 | + if (node && node->state == NODESTATE_MEMBER) { | |
7121 | + down(&cluster_members_lock); | |
7122 | + node->state = NODESTATE_DEAD; | |
7123 | + cluster_members--; | |
7124 | + up(&cluster_members_lock); | |
7125 | + } | |
7126 | + } | |
7127 | + /* Restart the timer */ | |
7128 | + del_timer(&transition_timer); | |
7129 | + mod_timer(&transition_timer, | |
7130 | + jiffies + cman_config.transition_timeout * HZ); | |
7131 | + } | |
7132 | + | |
7133 | + return 0; | |
7134 | +} | |
7135 | + | |
7136 | +/* Change a cluster parameter */ | |
7137 | +static int do_process_reconfig(struct msghdr *msg, int len) | |
7138 | +{ | |
7139 | + struct cl_mem_reconfig_msg *confmsg; | |
7140 | + struct sockaddr_cl *saddr = msg->msg_name; | |
7141 | + struct cluster_node *node; | |
7142 | + unsigned int val; | |
7143 | + | |
7144 | + if (len < sizeof(struct cl_mem_reconfig_msg)) | |
7145 | + return -1; | |
7146 | + | |
7147 | + confmsg = (struct cl_mem_reconfig_msg *) msg->msg_iov->iov_base; | |
7148 | + val = le32_to_cpu(confmsg->value); | |
7149 | + | |
7150 | + switch (confmsg->param) { | |
7151 | + | |
7152 | + case RECONFIG_PARAM_EXPECTED_VOTES: | |
7153 | + /* Set any nodes with expected_votes higher than the new value | |
7154 | + * down */ | |
7155 | + if (val > 0) { | |
7156 | + struct cluster_node *node; | |
7157 | + | |
7158 | + down(&cluster_members_lock); | |
7159 | + list_for_each_entry(node, &cluster_members_list, list) { | |
7160 | + if (node->state == NODESTATE_MEMBER && | |
7161 | + node->expected_votes > val) { | |
7162 | + node->expected_votes = val; | |
7163 | + } | |
7164 | + } | |
7165 | + up(&cluster_members_lock); | |
7166 | + if (expected_votes > val) | |
7167 | + expected_votes = val; | |
7168 | + } | |
7169 | + recalculate_quorum(1); /* Allow decrease */ | |
7170 | + sm_member_update(cluster_is_quorate); | |
7171 | + break; | |
7172 | + | |
7173 | + case RECONFIG_PARAM_NODE_VOTES: | |
7174 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7175 | + node->votes = val; | |
7176 | + recalculate_quorum(1); /* Allow decrease */ | |
7177 | + sm_member_update(cluster_is_quorate); | |
7178 | + break; | |
7179 | + | |
7180 | + case RECONFIG_PARAM_CONFIG_VERSION: | |
7181 | + config_version = val; | |
7182 | + break; | |
7183 | + | |
7184 | + default: | |
7185 | + printk(KERN_INFO CMAN_NAME | |
7186 | + ": got unknown parameter in reconfigure message. %d\n", | |
7187 | + confmsg->param); | |
7188 | + break; | |
7189 | + } | |
7190 | + return 0; | |
7191 | +} | |
7192 | + | |
7193 | +/* Response from master node */ | |
7194 | +static int do_process_joinack(struct msghdr *msg, int len) | |
7195 | +{ | |
7196 | + struct cl_mem_joinack_msg *ackmsg = msg->msg_iov->iov_base; | |
7197 | + | |
7198 | + join_time = jiffies; | |
7199 | + if (ackmsg->acktype == JOINACK_TYPE_OK) { | |
7200 | + node_state = JOINACK; | |
7201 | + } | |
7202 | + | |
7203 | + if (ackmsg->acktype == JOINACK_TYPE_NAK) { | |
7204 | + printk(KERN_WARNING CMAN_NAME | |
7205 | + ": Cluster membership rejected\n"); | |
7206 | + P_MEMB("Got JOINACK NACK\n"); | |
7207 | + node_state = REJECTED; | |
7208 | + } | |
7209 | + | |
7210 | + if (ackmsg->acktype == JOINACK_TYPE_WAIT) { | |
7211 | + P_MEMB("Got JOINACK WAIT\n"); | |
7212 | + node_state = JOINWAIT; | |
7213 | + joinwait_time = jiffies; | |
7214 | + } | |
7215 | + | |
7216 | + return 0; | |
7217 | +} | |
7218 | + | |
7219 | +/* Request to join the cluster. This makes us the master for this state | |
7220 | + * transition */ | |
7221 | +static int do_process_joinreq(struct msghdr *msg, int len) | |
7222 | +{ | |
7223 | + int status; | |
7224 | + static unsigned long last_joinreq = 0; | |
7225 | + static char last_name[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
7226 | + struct cl_mem_join_msg *joinmsg = msg->msg_iov->iov_base; | |
7227 | + struct cluster_node *node; | |
7228 | + | |
7229 | + /* If we are in a state transition then tell the new node to wait a bit | |
7230 | + * longer */ | |
7231 | + if (node_state != MEMBER) { | |
7232 | + if (node_state == MASTER || node_state == TRANSITION) { | |
7233 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7234 | + JOINACK_TYPE_WAIT); | |
7235 | + } | |
7236 | + return 0; | |
7237 | + } | |
7238 | + | |
7239 | + /* Check version number */ | |
7240 | + if (le32_to_cpu(joinmsg->major_version) == CNXMAN_MAJOR_VERSION) { | |
7241 | + char *ptr = (char *) joinmsg; | |
7242 | + char *name; | |
7243 | + | |
7244 | + /* Sanity-check the num_addrs field otherwise we could oops */ | |
7245 | + if (le16_to_cpu(joinmsg->num_addr) * address_length > len) { | |
7246 | + printk(KERN_WARNING CMAN_NAME | |
7247 | + ": num_addr in JOIN-REQ message is rubbish: %d\n", | |
7248 | + le16_to_cpu(joinmsg->num_addr)); | |
7249 | + return 0; | |
7250 | + } | |
7251 | + | |
7252 | + /* Check the cluster name matches */ | |
7253 | + if (strcmp(cluster_name, joinmsg->clustername)) { | |
7254 | + printk(KERN_WARNING CMAN_NAME | |
7255 | + ": attempt to join with cluster name '%s' refused\n", | |
7256 | + joinmsg->clustername); | |
7257 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7258 | + JOINACK_TYPE_NAK); | |
7259 | + return 0; | |
7260 | + } | |
7261 | + | |
7262 | + ptr += sizeof (*joinmsg); | |
7263 | + name = ptr + le16_to_cpu(joinmsg->num_addr) * address_length; | |
7264 | + | |
7265 | + /* Check we are not exceeding the maximum number of nodes */ | |
7266 | + if (cluster_members > cman_config.max_nodes) { | |
7267 | + printk(KERN_WARNING CMAN_NAME | |
7268 | + ": Join request from %s rejected, exceeds maximum number of nodes\n", | |
7269 | + name); | |
7270 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7271 | + JOINACK_TYPE_NAK); | |
7272 | + return 0; | |
7273 | + } | |
7274 | + | |
7275 | + /* Check that we don't exceed the two_node limit */ | |
7276 | + if (two_node && cluster_members == 2) { | |
7277 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7278 | + "rejected, exceeds two node limit\n", name); | |
7279 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7280 | + JOINACK_TYPE_NAK); | |
7281 | + return 0; | |
7282 | + } | |
7283 | + | |
7284 | + if (le16_to_cpu(joinmsg->config_version) != config_version) { | |
7285 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7286 | + "rejected, config version local %u remote %u\n", | |
7287 | + name, config_version, | |
7288 | + le16_to_cpu(joinmsg->config_version)); | |
7289 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7290 | + JOINACK_TYPE_NAK); | |
7291 | + return 0; | |
7292 | + } | |
7293 | + | |
7294 | + /* If these don't match then I don't know how the message | |
7295 | + arrived! However, I can't take the chance */ | |
7296 | + if (le32_to_cpu(joinmsg->addr_len) != address_length) { | |
7297 | + printk(KERN_WARNING CMAN_NAME ": Join request from %s " | |
7298 | + "rejected, address length local: %u remote %u\n", | |
7299 | + name, address_length, | |
7300 | + le32_to_cpu(joinmsg->addr_len)); | |
7301 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7302 | + JOINACK_TYPE_NAK); | |
7303 | + return 0; | |
7304 | + } | |
7305 | + | |
7306 | + /* Duplicate checking: Because joining messages do not have | |
7307 | + * sequence numbers we may get as many JOINREQ messages as we | |
7308 | + * have interfaces. This bit of code here just checks for | |
7309 | + * JOINREQ messages that come in from the same node in a small | |
7310 | + * period of time and removes the duplicates */ | |
7311 | + if (time_before(jiffies, last_joinreq + 10 * HZ) | |
7312 | + && strcmp(name, last_name) == 0) { | |
7313 | + return 0; | |
7314 | + } | |
7315 | + | |
7316 | + /* Do we already know about this node? */ | |
7317 | + status = check_duplicate_node(name, msg, len); | |
7318 | + | |
7319 | + if (status < 0) { | |
7320 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7321 | + JOINACK_TYPE_NAK); | |
7322 | + return 0; | |
7323 | + } | |
7324 | + | |
7325 | + /* OK, you can be in my gang */ | |
7326 | + if (status == 0) { | |
7327 | + int i; | |
7328 | + struct sockaddr_cl *addr = msg->msg_name; | |
7329 | + | |
7330 | + last_joinreq = jiffies; | |
7331 | + strcpy(last_name, name); | |
7332 | + | |
7333 | + node = | |
7334 | + add_new_node(name, joinmsg->votes, | |
7335 | + le32_to_cpu(joinmsg->expected_votes), | |
7336 | + 0, NODESTATE_JOINING); | |
7337 | + | |
7338 | + /* Add the node's addresses */ | |
7339 | + if (list_empty(&node->addr_list)) { | |
7340 | + for (i = 0; i < le16_to_cpu(joinmsg->num_addr); | |
7341 | + i++) { | |
7342 | + add_node_address(node, ptr, address_length); | |
7343 | + ptr += address_length; | |
7344 | + } | |
7345 | + } | |
7346 | + | |
7347 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7348 | + JOINACK_TYPE_OK); | |
7349 | + joining_node = node; | |
7350 | + joining_temp_nodeid = addr->scl_nodeid; | |
7351 | + | |
7352 | + /* Start the state transition */ | |
7353 | + start_transition(TRANS_NEWNODE, node); | |
7354 | + } | |
7355 | + } | |
7356 | + else { | |
7357 | + /* Version number mismatch, don't use any part of the message | |
7358 | + * other than the version numbers as things may have moved */ | |
7359 | + char buf[MAX_ADDR_PRINTED_LEN]; | |
7360 | + | |
7361 | + printk(KERN_INFO CMAN_NAME | |
7362 | + ": Got join message from node running incompatible software. (us: %d.%d.%d, them: %d.%d.%d) addr: %s\n", | |
7363 | + CNXMAN_MAJOR_VERSION, CNXMAN_MINOR_VERSION, | |
7364 | + CNXMAN_PATCH_VERSION, | |
7365 | + le32_to_cpu(joinmsg->major_version), | |
7366 | + le32_to_cpu(joinmsg->minor_version), | |
7367 | + le32_to_cpu(joinmsg->patch_version), | |
7368 | + print_addr(msg->msg_name, msg->msg_namelen, buf)); | |
7369 | + | |
7370 | + send_joinack(msg->msg_name, msg->msg_namelen, | |
7371 | + JOINACK_TYPE_NAK); | |
7372 | + return 0; | |
7373 | + } | |
7374 | + | |
7375 | + return 0; | |
7376 | +} | |
7377 | + | |
7378 | +/* A simple function to invent a small number based | |
7379 | + on the node name */ | |
7380 | +static int node_hash(void) | |
7381 | +{ | |
7382 | + int i; | |
7383 | + int value = 0; | |
7384 | + | |
7385 | + for (i=0; i<strlen(nodename); i++) { | |
7386 | + value += nodename[i]; | |
7387 | + } | |
7388 | + return value & 0xF; | |
7389 | +} | |
7390 | + | |
7391 | +/* A new node has stated its intent to form a new cluster. we may have | |
7392 | + * something to say about that... */ | |
7393 | +static int do_process_newcluster(struct msghdr *msg, int len) | |
7394 | +{ | |
7395 | + /* If we are also in STARTING state then back down for a random period | |
7396 | + * of time */ | |
7397 | + if (node_state == STARTING) { | |
7398 | + P_MEMB("got NEWCLUSTER, backing down for %d seconds\n", node_hash()); | |
7399 | + start_time = jiffies + node_hash() * HZ; | |
7400 | + } | |
7401 | + | |
7402 | + return 0; | |
7403 | +} | |
7404 | + | |
7405 | +/* Called for each node by the node-message unpacker. Returns -1 if there is a | |
7406 | + * mismatch and the caller will stop processing */ | |
7407 | +static int check_node(struct cluster_node *newnode, char *addrs, | |
7408 | + unsigned short num_addr) | |
7409 | +{ | |
7410 | + struct cluster_node *node = find_node_by_name(newnode->name); | |
7411 | + | |
7412 | + P_MEMB("check_node: %s", newnode->name); | |
7413 | + | |
7414 | + if (!node) { | |
7415 | + C_MEMB(" - not found\n"); | |
7416 | + return -1; | |
7417 | + } | |
7418 | + | |
7419 | + if (node->votes != newnode->votes || | |
7420 | + node->node_id != newnode->node_id || | |
7421 | + node->state != NODESTATE_MEMBER) { | |
7422 | + C_MEMB | |
7423 | + (" - wrong info: votes=%d(exp: %d) id=%d(exp: %d) state = %d\n", | |
7424 | + node->votes, newnode->votes, node->node_id, | |
7425 | + newnode->node_id, node->state); | |
7426 | + return -1; | |
7427 | + } | |
7428 | + C_MEMB(" - OK\n"); | |
7429 | + return 0; | |
7430 | +} | |
7431 | + | |
7432 | +/* Called for each new node found in a JOINCONF message. Create a new node | |
7433 | + * entry */ | |
7434 | +static int add_node(struct cluster_node *node, char *addrs, | |
7435 | + unsigned short num_addr) | |
7436 | +{ | |
7437 | + P_MEMB("add_node: %s, v:%d, e:%d, i:%d\n", node->name, node->votes, | |
7438 | + node->expected_votes, node->node_id); | |
7439 | + | |
7440 | + if (!find_node_by_name(node->name)) { | |
7441 | + struct cluster_node *newnode; | |
7442 | + int i; | |
7443 | + | |
7444 | + if ((newnode = | |
7445 | + add_new_node(node->name, node->votes, node->expected_votes, | |
7446 | + node->node_id, NODESTATE_MEMBER)) == NULL) { | |
7447 | + P_MEMB("Error adding node\n"); | |
7448 | + return -1; | |
7449 | + } | |
7450 | + if (list_empty(&newnode->addr_list)) { | |
7451 | + for (i = 0; i < num_addr; i++) { | |
7452 | + add_node_address(newnode, | |
7453 | + addrs + i * address_length, address_length); | |
7454 | + } | |
7455 | + } | |
7456 | + return 0; | |
7457 | + } | |
7458 | + else { | |
7459 | + P_MEMB("Already got node with name %s\n", node->name); | |
7460 | + return -1; | |
7461 | + } | |
7462 | +} | |
7463 | + | |
7464 | +/* Call a specified routine for each node unpacked from the message. Return | |
7465 | + * either the number of nodes found or -1 for an error */ | |
7466 | +static int unpack_nodes(unsigned char *buf, int len, | |
7467 | + int (*routine) (struct cluster_node *, char *, | |
7468 | + unsigned short)) | |
7469 | +{ | |
7470 | + int ptr = 0; | |
7471 | + int num_nodes = 0; | |
7472 | + char nodename[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
7473 | + struct cluster_node node; | |
7474 | + | |
7475 | + node.name = nodename; | |
7476 | + | |
7477 | + while (ptr < len) { | |
7478 | + int namelen = buf[ptr++]; | |
7479 | + unsigned int evotes; | |
7480 | + unsigned int node_id; | |
7481 | + unsigned short num_addr; | |
7482 | + unsigned char *addrs; | |
7483 | + | |
7484 | + memcpy(nodename, &buf[ptr], namelen); | |
7485 | + nodename[namelen] = '\0'; | |
7486 | + ptr += namelen; | |
7487 | + | |
7488 | + memcpy(&num_addr, &buf[ptr], sizeof (short)); | |
7489 | + num_addr = le16_to_cpu(num_addr); | |
7490 | + ptr += sizeof (short); | |
7491 | + | |
7492 | + /* Just make a note of the addrs "array" */ | |
7493 | + addrs = &buf[ptr]; | |
7494 | + ptr += num_addr * address_length; | |
7495 | + | |
7496 | + node.votes = buf[ptr++]; | |
7497 | + | |
7498 | + memcpy(&evotes, &buf[ptr], sizeof (int)); | |
7499 | + node.expected_votes = le32_to_cpu(evotes); | |
7500 | + ptr += sizeof (int); | |
7501 | + | |
7502 | + memcpy(&node_id, &buf[ptr], sizeof (int)); | |
7503 | + node.node_id = le32_to_cpu(node_id); | |
7504 | + ptr += sizeof (int); | |
7505 | + | |
7506 | + /* Call the callback routine */ | |
7507 | + if (routine(&node, addrs, num_addr) < 0) | |
7508 | + return -1; | |
7509 | + num_nodes++; | |
7510 | + } | |
7511 | + return num_nodes; | |
7512 | +} | |
7513 | + | |
7514 | +/* Got join confirmation from a master node. This message contains a list of | |
7515 | + * cluster nodes which we unpack and build into our cluster nodes list. When we | |
7516 | + * have the last message we can go into TRANSITION state */ | |
7517 | +static int do_process_joinconf(struct msghdr *msg, int len) | |
7518 | +{ | |
7519 | + char *message = msg->msg_iov->iov_base; | |
7520 | + | |
7521 | + if (unpack_nodes(message + 2, len - 2, add_node) < 0) { | |
7522 | + printk(CMAN_NAME | |
7523 | + ": Error procssing joinconf message - giving up on cluster join\n"); | |
7524 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
7525 | + return -1; | |
7526 | + } | |
7527 | + | |
7528 | + /* Last message in the list? */ | |
7529 | + if (message[1] & 2) { | |
7530 | + char ackmsg; | |
7531 | + struct sockaddr_cl *addr = msg->msg_name; | |
7532 | + | |
7533 | + us->state = NODESTATE_MEMBER; | |
7534 | + node_state = TRANSITION; | |
7535 | + we_are_a_cluster_member = TRUE; | |
7536 | + | |
7537 | + ackmsg = CLUSTER_MEM_CONFACK; | |
7538 | + kcl_sendmsg(mem_socket, &ackmsg, 1, addr, | |
7539 | + sizeof (struct sockaddr_cl), | |
7540 | + MSG_NOACK); | |
7541 | + kernel_thread(hello_kthread, NULL, 0); | |
7542 | + mod_timer(&hello_timer, jiffies + cman_config.hello_timer * HZ); | |
7543 | + } | |
7544 | + return 0; | |
7545 | +} | |
7546 | + | |
7547 | +/* Got the master's view of the cluster - compare it with ours and tell it the | |
7548 | + * result */ | |
7549 | +static int do_process_masterview(struct msghdr *msg, int len) | |
7550 | +{ | |
7551 | + char reply[2] = { CLUSTER_MEM_VIEWACK, 0 }; | |
7552 | + char *message = msg->msg_iov->iov_base; | |
7553 | + static int num_nodes; | |
7554 | + | |
7555 | + /* Someone else's state transition */ | |
7556 | + if (node_state != MEMBER && | |
7557 | + node_state != TRANSITION && node_state != MASTER) | |
7558 | + return 0; | |
7559 | + | |
7560 | + /* First message, zero the counter */ | |
7561 | + if (message[1] & 1) | |
7562 | + num_nodes = 0; | |
7563 | + | |
7564 | + num_nodes += | |
7565 | + unpack_nodes(msg->msg_iov->iov_base + 2, len - 2, check_node); | |
7566 | + | |
7567 | + /* Last message, check the count and reply */ | |
7568 | + if (message[1] & 2) { | |
7569 | + if (num_nodes == cluster_members) { | |
7570 | + /* Send ACK */ | |
7571 | + reply[1] = 1; | |
7572 | + } | |
7573 | + else { | |
7574 | + P_MEMB | |
7575 | + ("Got %d nodes in MASTERVIEW message, we think there s/b %d\n", | |
7576 | + num_nodes, cluster_members); | |
7577 | + /* Send NAK */ | |
7578 | + reply[1] = 0; | |
7579 | + } | |
7580 | + kcl_sendmsg(mem_socket, reply, 2, msg->msg_name, | |
7581 | + msg->msg_namelen, 0); | |
7582 | + } | |
7583 | + return 0; | |
7584 | +} | |
7585 | + | |
7586 | +static int do_process_leave(struct msghdr *msg, int len) | |
7587 | +{ | |
7588 | + struct cluster_node *node; | |
7589 | + struct sockaddr_cl *saddr = msg->msg_name; | |
7590 | + unsigned char *leavemsg = (unsigned char *) msg->msg_iov->iov_base; | |
7591 | + | |
7592 | + if ((node = find_node_by_nodeid(saddr->scl_nodeid))) { | |
7593 | + unsigned char reason = leavemsg[1]; | |
7594 | + | |
7595 | + if (node->state != NODESTATE_DEAD) { | |
7596 | + printk(KERN_INFO CMAN_NAME | |
7597 | + ": Node %s is leaving the cluster, reason %d\n", | |
7598 | + node->name, reason); | |
7599 | + | |
7600 | + node->leave_reason = reason; | |
7601 | + } | |
7602 | + leavereason = (reason == CLUSTER_LEAVEFLAG_REMOVED ? 1 : 0); | |
7603 | + | |
7604 | + a_node_just_died(node); | |
7605 | + | |
7606 | + /* If it was the master node, then we have been nominated as | |
7607 | + * the sucessor */ | |
7608 | + if (node == master_node) { | |
7609 | + start_transition(TRANS_DEADMASTER, master_node); | |
7610 | + } | |
7611 | + | |
7612 | + } | |
7613 | + return 0; | |
7614 | +} | |
7615 | + | |
7616 | +static int do_process_hello(struct msghdr *msg, int len) | |
7617 | +{ | |
7618 | + struct cluster_node *node; | |
7619 | + struct cl_mem_hello_msg *hellomsg = | |
7620 | + (struct cl_mem_hello_msg *) msg->msg_iov->iov_base; | |
7621 | + struct sockaddr_cl *saddr = msg->msg_name; | |
7622 | + | |
7623 | + /* We are starting up. Send a join message to the node whose HELLO we | |
7624 | + * just received */ | |
7625 | + if (node_state == STARTING || node_state == JOINWAIT) { | |
7626 | + struct sockaddr_cl *addr = msg->msg_name; | |
7627 | + | |
7628 | + printk(KERN_INFO CMAN_NAME ": sending membership request\n"); | |
7629 | + | |
7630 | + send_joinreq(addr, msg->msg_namelen); | |
7631 | + join_time = jiffies; | |
7632 | + node_state = JOINING; | |
7633 | + return 0; | |
7634 | + } | |
7635 | + | |
7636 | + /* Only process HELLOs if we are not in transition */ | |
7637 | + if (node_state == MEMBER) { | |
7638 | + if (len < sizeof (struct cl_mem_hello_msg)) { | |
7639 | + printk(KERN_ERR CMAN_NAME | |
7640 | + ": short hello message from node %d\n", | |
7641 | + saddr->scl_nodeid); | |
7642 | + return -1; | |
7643 | + } | |
7644 | + | |
7645 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7646 | + if (node && node->state != NODESTATE_DEAD) { | |
7647 | + | |
7648 | + /* Check the cluster generation in the HELLO message. | |
7649 | + * NOTE: this may be different if the message crossed | |
7650 | + * on the wire with an END-TRANS so we allow a period | |
7651 | + * of grace in which this is allowable */ | |
7652 | + if (cluster_generation != | |
7653 | + le32_to_cpu(hellomsg->generation) | |
7654 | + && node_state == MEMBER | |
7655 | + && time_after(jiffies, | |
7656 | + cman_config.hello_timer * HZ + | |
7657 | + transition_end_time)) { | |
7658 | + char killmsg; | |
7659 | + | |
7660 | + printk(KERN_INFO CMAN_NAME | |
7661 | + ": bad generation number %d in HELLO message, expected %d\n", | |
7662 | + le32_to_cpu(hellomsg->generation), | |
7663 | + cluster_generation); | |
7664 | + | |
7665 | + notify_kernel_listeners(DIED, | |
7666 | + (long) node->node_id); | |
7667 | + | |
7668 | + killmsg = CLUSTER_MEM_KILL; | |
7669 | + kcl_sendmsg(mem_socket, &killmsg, 1, | |
7670 | + saddr, sizeof (struct sockaddr_cl), | |
7671 | + MSG_NOACK); | |
7672 | + return 0; | |
7673 | + } | |
7674 | + | |
7675 | + if (cluster_members != le16_to_cpu(hellomsg->members) | |
7676 | + && node_state == MEMBER) { | |
7677 | + printk(KERN_INFO CMAN_NAME | |
7678 | + ": nmembers in HELLO message does not match our view\n"); | |
7679 | + start_transition(TRANS_CHECK, node); | |
7680 | + return 0; | |
7681 | + } | |
7682 | + /* The message is OK - save the time */ | |
7683 | + node->last_hello = jiffies; | |
7684 | + | |
7685 | + } | |
7686 | + else { | |
7687 | + struct sockaddr_cl *addr = msg->msg_name; | |
7688 | + | |
7689 | + /* This node is a danger to our valid cluster */ | |
7690 | + if (cluster_is_quorate) { | |
7691 | + char killmsg; | |
7692 | + | |
7693 | + killmsg = CLUSTER_MEM_KILL; | |
7694 | + kcl_sendmsg(mem_socket, &killmsg, 1, addr, | |
7695 | + sizeof (struct sockaddr_cl), | |
7696 | + MSG_NOACK); | |
7697 | + } | |
7698 | + | |
7699 | + } | |
7700 | + } | |
7701 | + | |
7702 | + return 0; | |
7703 | + | |
7704 | +} | |
7705 | + | |
7706 | +static int do_process_kill(struct msghdr *msg, int len) | |
7707 | +{ | |
7708 | + struct sockaddr_cl *saddr = msg->msg_name; | |
7709 | + struct cluster_node *node; | |
7710 | + | |
7711 | + node = find_node_by_nodeid(saddr->scl_nodeid); | |
7712 | + if (node && node->state == NODESTATE_MEMBER) { | |
7713 | + | |
7714 | + printk(KERN_INFO CMAN_NAME | |
7715 | + ": Being told to leave the cluster by node %d\n", | |
7716 | + saddr->scl_nodeid); | |
7717 | + | |
7718 | + node_state = LEFT_CLUSTER; | |
7719 | + quit_threads = 1; | |
7720 | + wake_up_process(membership_task); | |
7721 | + wake_up_interruptible(&cnxman_waitq); | |
7722 | + } | |
7723 | + else { | |
7724 | + P_MEMB("Asked to leave the cluster by a non-member. What a nerve!\n"); | |
7725 | + } | |
7726 | + return 0; | |
7727 | +} | |
7728 | + | |
7729 | +/* Some cluster membership utility functions */ | |
7730 | +struct cluster_node *find_node_by_name(char *name) | |
7731 | +{ | |
7732 | + struct list_head *nodelist; | |
7733 | + struct cluster_node *node; | |
7734 | + | |
7735 | + down(&cluster_members_lock); | |
7736 | + list_for_each(nodelist, &cluster_members_list) { | |
7737 | + node = list_entry(nodelist, struct cluster_node, list); | |
7738 | + | |
7739 | + if (strcmp(node->name, name) == 0) { | |
7740 | + up(&cluster_members_lock); | |
7741 | + return node; | |
7742 | + } | |
7743 | + } | |
7744 | + up(&cluster_members_lock); | |
7745 | + return NULL; | |
7746 | +} | |
7747 | + | |
7748 | +/* Try to avoid using this as it's slow and holds the members lock */ | |
7749 | +struct cluster_node *find_node_by_addr(unsigned char *addr, int addr_len) | |
7750 | +{ | |
7751 | + struct list_head *nodelist; | |
7752 | + struct list_head *addrlist; | |
7753 | + struct cluster_node *node; | |
7754 | + struct cluster_node_addr *nodeaddr; | |
7755 | + | |
7756 | + down(&cluster_members_lock); | |
7757 | + | |
7758 | + list_for_each(nodelist, &cluster_members_list) { | |
7759 | + node = list_entry(nodelist, struct cluster_node, list); | |
7760 | + | |
7761 | + list_for_each(addrlist, &node->addr_list) { | |
7762 | + nodeaddr = | |
7763 | + list_entry(addrlist, struct cluster_node_addr, | |
7764 | + list); | |
7765 | + | |
7766 | + if (memcmp(nodeaddr->addr, addr, address_length) == 0) { | |
7767 | + up(&cluster_members_lock); | |
7768 | + return node; | |
7769 | + } | |
7770 | + } | |
7771 | + } | |
7772 | + | |
7773 | + up(&cluster_members_lock); | |
7774 | + return NULL; | |
7775 | +} | |
7776 | + | |
7777 | +/* This is the quick way to find a node */ | |
7778 | +struct cluster_node *find_node_by_nodeid(unsigned int id) | |
7779 | +{ | |
7780 | + struct cluster_node *node; | |
7781 | + | |
7782 | + if (id > sizeof_members_array) | |
7783 | + return NULL; | |
7784 | + | |
7785 | + spin_lock(&members_by_nodeid_lock); | |
7786 | + node = members_by_nodeid[id]; | |
7787 | + spin_unlock(&members_by_nodeid_lock); | |
7788 | + return node; | |
7789 | +} | |
7790 | + | |
7791 | +static int dispatch_messages(struct socket *mem_socket) | |
7792 | +{ | |
7793 | + int err = 0; | |
7794 | + | |
7795 | + while (skb_peek(&mem_socket->sk->sk_receive_queue)) { | |
7796 | + struct msghdr msg; | |
7797 | + struct iovec iov; | |
7798 | + struct sockaddr_cl sin; | |
7799 | + int len; | |
7800 | + mm_segment_t fs; | |
7801 | + | |
7802 | + memset(&sin, 0, sizeof (sin)); | |
7803 | + | |
7804 | + msg.msg_control = NULL; | |
7805 | + msg.msg_controllen = 0; | |
7806 | + msg.msg_iovlen = 1; | |
7807 | + msg.msg_iov = &iov; | |
7808 | + msg.msg_name = &sin; | |
7809 | + msg.msg_namelen = sizeof (sin); | |
7810 | + msg.msg_flags = 0; | |
7811 | + | |
7812 | + iov.iov_len = MAX_CLUSTER_MESSAGE; | |
7813 | + iov.iov_base = iobuf; | |
7814 | + | |
7815 | + fs = get_fs(); | |
7816 | + set_fs(get_ds()); | |
7817 | + | |
7818 | + len = | |
7819 | + sock_recvmsg(mem_socket, &msg, MAX_CLUSTER_MESSAGE, | |
7820 | + MSG_DONTWAIT); | |
7821 | + set_fs(fs); | |
7822 | + if (len > 0) { | |
7823 | + iov.iov_base = iobuf; /* Reinstate pointer */ | |
7824 | + msg.msg_name = &sin; | |
7825 | + do_membership_packet(&msg, len); | |
7826 | + } | |
7827 | + else { | |
7828 | + if (len == -EAGAIN) | |
7829 | + err = 0; | |
7830 | + else | |
7831 | + err = -1; | |
7832 | + break; | |
7833 | + } | |
7834 | + } | |
7835 | + return err; | |
7836 | +} | |
7837 | + | |
7838 | +/* Scan the nodes list for dead nodes */ | |
7839 | +static void check_for_dead_nodes() | |
7840 | +{ | |
7841 | + struct list_head *nodelist; | |
7842 | + struct cluster_node *node; | |
7843 | + | |
7844 | + down(&cluster_members_lock); | |
7845 | + list_for_each(nodelist, &cluster_members_list) { | |
7846 | + node = list_entry(nodelist, struct cluster_node, list); | |
7847 | + | |
7848 | + if (node->state != NODESTATE_DEAD && | |
7849 | + time_after(jiffies, | |
7850 | + node->last_hello + | |
7851 | + cman_config.deadnode_timeout * HZ) && !node->us) { | |
7852 | + | |
7853 | + up(&cluster_members_lock); | |
7854 | + | |
7855 | + printk(KERN_WARNING CMAN_NAME | |
7856 | + ": no HELLO from %s, removing from the cluster\n", | |
7857 | + node->name); | |
7858 | + | |
7859 | + P_MEMB("last hello was %ld, current time is %ld\n", | |
7860 | + node->last_hello, jiffies); | |
7861 | + | |
7862 | + node->leave_reason = CLUSTER_LEAVEFLAG_DEAD; | |
7863 | + leavereason = 0; | |
7864 | + | |
7865 | + /* This is unlikely to work but it's worth a try! */ | |
7866 | + send_kill(node->node_id); | |
7867 | + | |
7868 | + /* Start state transition */ | |
7869 | + a_node_just_died(node); | |
7870 | + return; | |
7871 | + } | |
7872 | + } | |
7873 | + up(&cluster_members_lock); | |
7874 | + | |
7875 | + /* Also check for a dead quorum device */ | |
7876 | + if (quorum_device) { | |
7877 | + if (quorum_device->state == NODESTATE_MEMBER && | |
7878 | + time_after(jiffies, | |
7879 | + quorum_device->last_hello + | |
7880 | + cman_config.deadnode_timeout * HZ)) { | |
7881 | + quorum_device->state = NODESTATE_DEAD; | |
7882 | + printk(KERN_WARNING CMAN_NAME | |
7883 | + ": Quorum device %s timed out\n", | |
7884 | + quorum_device->name); | |
7885 | + recalculate_quorum(0); | |
7886 | + } | |
7887 | + } | |
7888 | + | |
7889 | + return; | |
7890 | +} | |
7891 | + | |
7892 | +/* add "us" as a node in the cluster */ | |
7893 | +static int add_us() | |
7894 | +{ | |
7895 | + struct cluster_node *newnode = | |
7896 | + kmalloc(sizeof (struct cluster_node), GFP_KERNEL); | |
7897 | + | |
7898 | + if (!newnode) { | |
7899 | + /* Oh shit, we have to commit hara kiri here for the greater | |
7900 | + * good of the cluster */ | |
7901 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
7902 | + | |
7903 | + printk(KERN_CRIT CMAN_NAME | |
7904 | + ": Cannot allocate memory for our node structure\n"); | |
7905 | + panic("Must die"); | |
7906 | + | |
7907 | + return -1; | |
7908 | + } | |
7909 | + | |
7910 | + memset(newnode, 0, sizeof (struct cluster_node)); | |
7911 | + newnode->name = kmalloc(strlen(nodename) + 1, GFP_KERNEL); | |
7912 | + if (!newnode->name) { | |
7913 | + send_leave(CLUSTER_LEAVEFLAG_PANIC); | |
7914 | + | |
7915 | + printk(KERN_CRIT CMAN_NAME | |
7916 | + ": Cannot allocate memory for node name\n"); | |
7917 | + kfree(newnode); | |
7918 | + | |
7919 | + panic("Must die"); | |
7920 | + | |
7921 | + return -1; | |
7922 | + } | |
7923 | + | |
7924 | + strcpy(newnode->name, nodename); | |
7925 | + newnode->last_hello = jiffies; | |
7926 | + newnode->votes = votes; | |
7927 | + newnode->expected_votes = expected_votes; | |
7928 | + newnode->state = NODESTATE_JOINING; | |
7929 | + newnode->node_id = 0; /* Will get filled in by ENDTRANS message */ | |
7930 | + newnode->us = 1; | |
7931 | + newnode->leave_reason = 0; | |
7932 | + INIT_LIST_HEAD(&newnode->addr_list); | |
7933 | + get_local_addresses(newnode); /* Get from cnxman socket info */ | |
7934 | + | |
7935 | + /* Add the new node to the list */ | |
7936 | + down(&cluster_members_lock); | |
7937 | + list_add(&newnode->list, &cluster_members_list); | |
7938 | + cluster_members++; | |
7939 | + up(&cluster_members_lock); | |
7940 | + us = newnode; | |
7941 | + | |
7942 | + return 0; | |
7943 | +} | |
7944 | + | |
7945 | +/* Return the highest known node_id */ | |
7946 | +unsigned int get_highest_nodeid() | |
7947 | +{ | |
7948 | + struct list_head *nodelist; | |
7949 | + struct cluster_node *node = NULL; | |
7950 | + unsigned int highest = 0; | |
7951 | + | |
7952 | + down(&cluster_members_lock); | |
7953 | + list_for_each(nodelist, &cluster_members_list) { | |
7954 | + node = list_entry(nodelist, struct cluster_node, list); | |
7955 | + | |
7956 | + if (node->node_id > highest) | |
7957 | + highest = node->node_id; | |
7958 | + } | |
7959 | + up(&cluster_members_lock); | |
7960 | + | |
7961 | + return highest; | |
7962 | +} | |
7963 | + | |
7964 | +/* Elect a new master if there is a clash. Returns 1 if we are the new master, | |
7965 | + * the master's struct will also be returned. This, rather primitively, uses | |
7966 | + * the lowest node ID */ | |
7967 | +static int elect_master(struct cluster_node **master_node) | |
7968 | +{ | |
7969 | + int i; | |
7970 | + | |
7971 | + for (i = 1; i < sizeof_members_array; i++) { | |
7972 | + if (members_by_nodeid[i] | |
7973 | + && members_by_nodeid[i]->state == NODESTATE_MEMBER) { | |
7974 | + *master_node = members_by_nodeid[i]; | |
7975 | + P_MEMB("Elected master is %s\n", (*master_node)->name); | |
7976 | + return (*master_node)->us; | |
7977 | + } | |
7978 | + } | |
7979 | + BUG(); | |
7980 | + return 0; | |
7981 | +} | |
7982 | + | |
7983 | +/* Called by node_cleanup in cnxman when we have left the cluster */ | |
7984 | +void free_nodeid_array() | |
7985 | +{ | |
7986 | + vfree(members_by_nodeid); | |
7987 | + members_by_nodeid = NULL; | |
7988 | + sizeof_members_array = 0; | |
7989 | +} | |
7990 | + | |
7991 | +int allocate_nodeid_array() | |
7992 | +{ | |
7993 | + /* Allocate space for the nodeid lookup array */ | |
7994 | + if (!members_by_nodeid) { | |
7995 | + spin_lock_init(&members_by_nodeid_lock); | |
7996 | + members_by_nodeid = | |
7997 | + vmalloc(cman_config.max_nodes * | |
7998 | + sizeof (struct cluster_member *)); | |
7999 | + } | |
8000 | + | |
8001 | + if (!members_by_nodeid) { | |
8002 | + printk(KERN_WARNING | |
8003 | + "Unable to allocate members array for %d members\n", | |
8004 | + cman_config.max_nodes); | |
8005 | + return -ENOMEM; | |
8006 | + } | |
8007 | + memset(members_by_nodeid, 0, | |
8008 | + cman_config.max_nodes * sizeof (struct cluster_member *)); | |
8009 | + sizeof_members_array = cman_config.max_nodes; | |
8010 | + | |
8011 | + return 0; | |
8012 | +} | |
8013 | + | |
8014 | +/* Set the votes & expected_votes variables */ | |
8015 | +void set_votes(int v, int e) | |
8016 | +{ | |
8017 | + votes = v; | |
8018 | + expected_votes = e; | |
8019 | +} | |
8020 | + | |
8021 | +int get_quorum() | |
8022 | +{ | |
8023 | + return quorum; | |
8024 | +} | |
8025 | + | |
8026 | +/* Called by cnxman to see if activity should be blocked because we are in a | |
8027 | + * state transition */ | |
8028 | +int in_transition() | |
8029 | +{ | |
8030 | + return node_state == TRANSITION || | |
8031 | + node_state == TRANSITION_COMPLETE || node_state == MASTER; | |
8032 | +} | |
8033 | + | |
8034 | +/* Return the current membership state as a string for the main line to put | |
8035 | + * into /proc . I really should be using snprintf rather than sprintf but it's | |
8036 | + * not exported... */ | |
8037 | +char *membership_state(char *buf, int buflen) | |
8038 | +{ | |
8039 | + switch (node_state) { | |
8040 | + case STARTING: | |
8041 | + strncpy(buf, "Starting", buflen); | |
8042 | + break; | |
8043 | + case JOINING: | |
8044 | + strncpy(buf, "Joining", buflen); | |
8045 | + break; | |
8046 | + case JOINWAIT: | |
8047 | + strncpy(buf, "Join-Wait", buflen); | |
8048 | + break; | |
8049 | + case JOINACK: | |
8050 | + strncpy(buf, "Join-Ack", buflen); | |
8051 | + break; | |
8052 | + case TRANSITION: | |
8053 | + sprintf(buf, "State-Transition: Master is %s", | |
8054 | + master_node ? master_node->name : "Unknown"); | |
8055 | + break; | |
8056 | + case MEMBER: | |
8057 | + strncpy(buf, "Cluster-Member", buflen); | |
8058 | + break; | |
8059 | + case REJECTED: | |
8060 | + strncpy(buf, "Rejected", buflen); | |
8061 | + break; | |
8062 | + case LEFT_CLUSTER: | |
8063 | + strncpy(buf, "Left-Cluster", buflen); | |
8064 | + break; | |
8065 | + case TRANSITION_COMPLETE: | |
8066 | + strncpy(buf, "Transition-Complete", buflen); | |
8067 | + break; | |
8068 | + case MASTER: | |
8069 | + strncpy(buf, "Transition-Master", buflen); | |
8070 | + break; | |
8071 | + default: | |
8072 | + sprintf(buf, "Unknown: code=%d", node_state); | |
8073 | + break; | |
8074 | + } | |
8075 | + | |
8076 | + return buf; | |
8077 | +} | |
8078 | + | |
8079 | +#ifdef DEBUG_MEMB | |
8080 | +static char *msgname(int msg) | |
8081 | +{ | |
8082 | + switch (msg) { | |
8083 | + case CLUSTER_MEM_JOINCONF: | |
8084 | + return "JOINCONF"; | |
8085 | + case CLUSTER_MEM_JOINREQ: | |
8086 | + return "JOINREQ"; | |
8087 | + case CLUSTER_MEM_LEAVE: | |
8088 | + return "LEAVE"; | |
8089 | + case CLUSTER_MEM_HELLO: | |
8090 | + return "HELLO"; | |
8091 | + case CLUSTER_MEM_KILL: | |
8092 | + return "KILL"; | |
8093 | + case CLUSTER_MEM_JOINACK: | |
8094 | + return "JOINACK"; | |
8095 | + case CLUSTER_MEM_ENDTRANS: | |
8096 | + return "ENDTRANS"; | |
8097 | + case CLUSTER_MEM_RECONFIG: | |
8098 | + return "RECONFIG"; | |
8099 | + case CLUSTER_MEM_MASTERVIEW: | |
8100 | + return "MASTERVIEW"; | |
8101 | + case CLUSTER_MEM_STARTTRANS: | |
8102 | + return "STARTTRANS"; | |
8103 | + case CLUSTER_MEM_JOINREJ: | |
8104 | + return "JOINREJ"; | |
8105 | + case CLUSTER_MEM_VIEWACK: | |
8106 | + return "VIEWACK"; | |
8107 | + case CLUSTER_MEM_STARTACK: | |
8108 | + return "STARTACK"; | |
8109 | + case CLUSTER_MEM_NEWCLUSTER: | |
8110 | + return "NEWCLUSTER"; | |
8111 | + case CLUSTER_MEM_CONFACK: | |
8112 | + return "CONFACK"; | |
8113 | + case CLUSTER_MEM_NOMINATE: | |
8114 | + return "NOMINATE"; | |
8115 | + | |
8116 | + default: | |
8117 | + return "??UNKNOWN??"; | |
8118 | + } | |
8119 | +} | |
8120 | + | |
8121 | +#endif | |
8122 | + | |
8123 | +/* | |
8124 | + * Overrides for Emacs so that we follow Linus's tabbing style. | |
8125 | + * Emacs will notice this stuff at the end of the file and automatically | |
8126 | + * adjust the settings for this buffer only. This must remain at the end | |
8127 | + * of the file. | |
8128 | + * --------------------------------------------------------------------------- | |
8129 | + * Local variables: | |
8130 | + * c-file-style: "linux" | |
8131 | + * End: | |
8132 | + */ | |
8133 | diff -urN linux-orig/cluster/cman/proc.c linux-patched/cluster/cman/proc.c | |
8134 | --- linux-orig/cluster/cman/proc.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 8135 | +++ linux-patched/cluster/cman/proc.c 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 8136 | @@ -0,0 +1,364 @@ |
8137 | +/****************************************************************************** | |
8138 | +******************************************************************************* | |
8139 | +** | |
8140 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8141 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8142 | +** | |
8143 | +** This copyrighted material is made available to anyone wishing to use, | |
8144 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8145 | +** of the GNU General Public License v.2. | |
8146 | +** | |
8147 | +******************************************************************************* | |
8148 | +******************************************************************************/ | |
8149 | + | |
8150 | +#include <linux/init.h> | |
8151 | +#include <linux/socket.h> | |
8152 | +#include <linux/kernel.h> | |
8153 | +#include <linux/sched.h> | |
8154 | +#include <linux/file.h> | |
8155 | +#include <linux/proc_fs.h> | |
8156 | +#include <linux/seq_file.h> | |
8157 | +#include <linux/list.h> | |
8158 | +#include <linux/in.h> | |
8159 | +#include <net/sock.h> | |
8160 | +#include <cluster/cnxman.h> | |
8161 | +#include <cluster/service.h> | |
8162 | + | |
8163 | +#include "cnxman-private.h" | |
8164 | +#include "config.h" | |
8165 | + | |
8166 | +extern int cluster_members; | |
8167 | +extern struct list_head cluster_members_list; | |
8168 | +extern struct semaphore cluster_members_lock; | |
8169 | +extern struct cluster_node *quorum_device; | |
8170 | +extern int we_are_a_cluster_member; | |
8171 | +extern int cluster_is_quorate; | |
8172 | +extern unsigned short cluster_id; | |
8173 | +extern atomic_t use_count; | |
8174 | +extern unsigned int address_length; | |
8175 | +extern unsigned int config_version; | |
8176 | +extern char cluster_name[]; | |
8177 | +extern struct cluster_node *us; | |
8178 | +static struct seq_operations cluster_info_op; | |
8179 | + | |
8180 | +int sm_procdata(char *b, char **start, off_t offset, int length); | |
8181 | +int sm_debug_info(char *b, char **start, off_t offset, int length); | |
8182 | + | |
8183 | +/* /proc interface to the configuration struct */ | |
8184 | +static struct config_proc_info { | |
8185 | + char *name; | |
8186 | + int *value; | |
8187 | +} config_proc[] = { | |
8188 | + { | |
8189 | + .name = "joinwait_timeout", | |
8190 | + .value = &cman_config.joinwait_timeout, | |
8191 | + }, | |
8192 | + { | |
8193 | + .name = "joinconf_timeout", | |
8194 | + .value = &cman_config.joinconf_timeout, | |
8195 | + }, | |
8196 | + { | |
8197 | + .name = "join_timeout", | |
8198 | + .value = &cman_config.join_timeout, | |
8199 | + }, | |
8200 | + { | |
8201 | + .name = "hello_timer", | |
8202 | + .value = &cman_config.hello_timer, | |
8203 | + }, | |
8204 | + { | |
8205 | + .name = "deadnode_timeout", | |
8206 | + .value = &cman_config.deadnode_timeout, | |
8207 | + }, | |
8208 | + { | |
8209 | + .name = "transition_timeout", | |
8210 | + .value = &cman_config.transition_timeout, | |
8211 | + }, | |
8212 | + { | |
8213 | + .name = "transition_restarts", | |
8214 | + .value = &cman_config.transition_restarts, | |
8215 | + }, | |
8216 | + { | |
8217 | + .name = "max_nodes", | |
8218 | + .value = &cman_config.max_nodes, | |
8219 | + }, | |
8220 | + { | |
8221 | + .name = "sm_debug_size", | |
8222 | + .value = &cman_config.sm_debug_size, | |
8223 | + }, | |
8224 | +}; | |
8225 | + | |
8226 | + | |
8227 | +static int proc_cluster_status(char *b, char **start, off_t offset, int length) | |
8228 | +{ | |
8229 | + struct list_head *nodelist; | |
8230 | + struct cluster_node *node; | |
8231 | + struct cluster_node_addr *node_addr; | |
8232 | + unsigned int total_votes = 0; | |
8233 | + unsigned int max_expected = 0; | |
8234 | + int c = 0; | |
8235 | + char node_buf[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
8236 | + | |
8237 | + if (!we_are_a_cluster_member) { | |
8238 | + c += sprintf(b+c, "Not a cluster member. State: %s\n", | |
8239 | + membership_state(node_buf, | |
8240 | + sizeof (node_buf))); | |
8241 | + return c; | |
8242 | + } | |
8243 | + | |
8244 | + /* Total the votes */ | |
8245 | + down(&cluster_members_lock); | |
8246 | + list_for_each(nodelist, &cluster_members_list) { | |
8247 | + node = list_entry(nodelist, struct cluster_node, list); | |
8248 | + if (node->state == NODESTATE_MEMBER) { | |
8249 | + total_votes += node->votes; | |
8250 | + max_expected = | |
8251 | + max(max_expected, node->expected_votes); | |
8252 | + } | |
8253 | + } | |
8254 | + up(&cluster_members_lock); | |
8255 | + | |
8256 | + if (quorum_device && quorum_device->state == NODESTATE_MEMBER) | |
8257 | + total_votes += quorum_device->votes; | |
8258 | + | |
8259 | + c += sprintf(b+c, | |
8260 | + "Version: %d.%d.%d\nConfig version: %d\nCluster name: %s\nCluster ID: %d\nMembership state: %s\n", | |
8261 | + CNXMAN_MAJOR_VERSION, CNXMAN_MINOR_VERSION, | |
8262 | + CNXMAN_PATCH_VERSION, | |
8263 | + config_version, | |
8264 | + cluster_name, cluster_id, | |
8265 | + membership_state(node_buf, sizeof (node_buf))); | |
8266 | + c += sprintf(b+c, | |
8267 | + "Nodes: %d\nExpected_votes: %d\nTotal_votes: %d\nQuorum: %d %s\n", | |
8268 | + cluster_members, max_expected, total_votes, | |
8269 | + get_quorum(), | |
8270 | + cluster_is_quorate ? " " : "Activity blocked"); | |
8271 | + c += sprintf(b+c, "Active subsystems: %d\n", | |
8272 | + atomic_read(&use_count)); | |
8273 | + | |
8274 | + | |
8275 | + c += sprintf(b+c, "Node addresses: "); | |
8276 | + list_for_each_entry(node_addr, &us->addr_list, list) { | |
8277 | + struct sockaddr_in6 *saddr = (struct sockaddr_in6 *)node_addr->addr; | |
8278 | + if (saddr->sin6_family == AF_INET6) { | |
8279 | + c += sprintf(b+c, "%x:%x:%x:%x:%x:%x:%x:%x ", | |
8280 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[0]), | |
8281 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[1]), | |
8282 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[2]), | |
8283 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[3]), | |
8284 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[4]), | |
8285 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[5]), | |
8286 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[6]), | |
8287 | + be16_to_cpu(saddr->sin6_addr.s6_addr16[7])); | |
8288 | + } | |
8289 | + else { | |
8290 | + struct sockaddr_in *saddr4 = (struct sockaddr_in *)saddr; | |
8291 | + uint8_t *addr = (uint8_t *)&saddr4->sin_addr; | |
8292 | + c+= sprintf(b+c, "%u.%u.%u.%u ", | |
8293 | + addr[0], addr[1], addr[2], addr[3]); | |
8294 | + } | |
8295 | + } | |
8296 | + c += sprintf(b+c, "\n\n"); | |
8297 | + return c; | |
8298 | +} | |
8299 | + | |
8300 | + | |
8301 | +/* Allocate one of these for /proc/cluster/nodes so we can keep a track of where | |
8302 | + * we are */ | |
8303 | +struct cluster_seq_info { | |
8304 | + int nodeid; | |
8305 | + int highest_nodeid; | |
8306 | +}; | |
8307 | + | |
8308 | +static int cluster_open(struct inode *inode, struct file *file) | |
8309 | +{ | |
8310 | + return seq_open(file, &cluster_info_op); | |
8311 | +} | |
8312 | + | |
8313 | +static void *cluster_seq_start(struct seq_file *m, loff_t * pos) | |
8314 | +{ | |
8315 | + struct cluster_seq_info *csi = | |
8316 | + kmalloc(sizeof (struct cluster_seq_info), GFP_KERNEL); | |
8317 | + | |
8318 | + if (!csi) | |
8319 | + return NULL; | |
8320 | + | |
8321 | + /* Keep highest_nodeid here so we don't need to keep traversing the | |
8322 | + * list to find it */ | |
8323 | + csi->nodeid = *pos; | |
8324 | + csi->highest_nodeid = get_highest_nodeid(); | |
8325 | + | |
8326 | + /* Print the header */ | |
8327 | + if (*pos == 0) { | |
8328 | + seq_printf(m, | |
8329 | + "Node Votes Exp Sts Name\n"); | |
8330 | + return csi; | |
8331 | + } | |
8332 | + return csi; | |
8333 | +} | |
8334 | + | |
8335 | +static void *cluster_seq_next(struct seq_file *m, void *p, loff_t * pos) | |
8336 | +{ | |
8337 | + struct cluster_seq_info *csi = p; | |
8338 | + | |
8339 | + *pos = ++csi->nodeid; | |
8340 | + if (csi->nodeid > csi->highest_nodeid) | |
8341 | + return NULL; | |
8342 | + | |
8343 | + return csi; | |
8344 | +} | |
8345 | + | |
8346 | +static int cluster_seq_show(struct seq_file *m, void *p) | |
8347 | +{ | |
8348 | + char state = '?'; | |
8349 | + struct cluster_node *node; | |
8350 | + struct cluster_seq_info *csi = p; | |
8351 | + | |
8352 | + /* | |
8353 | + * If we have "0" here then display the quorum device if | |
8354 | + * there is one. | |
8355 | + */ | |
8356 | + if (csi->nodeid == 0) | |
8357 | + node = quorum_device; | |
8358 | + else | |
8359 | + node = find_node_by_nodeid(csi->nodeid); | |
8360 | + | |
8361 | + if (!node) | |
8362 | + return 0; | |
8363 | + | |
8364 | + /* Make state printable */ | |
8365 | + switch (node->state) { | |
8366 | + case NODESTATE_MEMBER: | |
8367 | + state = 'M'; | |
8368 | + break; | |
8369 | + case NODESTATE_JOINING: | |
8370 | + state = 'J'; | |
8371 | + break; | |
8372 | + case NODESTATE_REMOTEMEMBER: | |
8373 | + state = 'R'; | |
8374 | + break; | |
8375 | + case NODESTATE_DEAD: | |
8376 | + state = 'X'; | |
8377 | + break; | |
8378 | + } | |
8379 | + seq_printf(m, " %3d %3d %3d %c %s\n", | |
8380 | + node->node_id, | |
8381 | + node->votes, | |
8382 | + node->expected_votes, | |
8383 | + state, | |
8384 | + node->name); | |
8385 | + | |
8386 | + return 0; | |
8387 | +} | |
8388 | + | |
8389 | +static void cluster_seq_stop(struct seq_file *m, void *p) | |
8390 | +{ | |
8391 | + kfree(p); | |
8392 | +} | |
8393 | + | |
8394 | +static struct seq_operations cluster_info_op = { | |
8395 | + .start = cluster_seq_start, | |
8396 | + .next = cluster_seq_next, | |
8397 | + .stop = cluster_seq_stop, | |
8398 | + .show = cluster_seq_show | |
8399 | +}; | |
8400 | + | |
8401 | +static struct file_operations cluster_fops = { | |
8402 | + .open = cluster_open, | |
8403 | + .read = seq_read, | |
8404 | + .llseek = seq_lseek, | |
8405 | + .release = seq_release, | |
8406 | +}; | |
8407 | + | |
8408 | +static int cman_config_read_proc(char *page, char **start, off_t off, int count, | |
8409 | + int *eof, void *data) | |
8410 | +{ | |
8411 | + struct config_proc_info *cinfo = data; | |
8412 | + | |
8413 | + return snprintf(page, count, "%d\n", *cinfo->value); | |
8414 | +} | |
8415 | + | |
8416 | +static int cman_config_write_proc(struct file *file, const char *buffer, | |
8417 | + unsigned long count, void *data) | |
8418 | +{ | |
8419 | + struct config_proc_info *cinfo = data; | |
8420 | + int value; | |
8421 | + char *end; | |
8422 | + | |
8423 | + value = simple_strtoul(buffer, &end, 10); | |
8424 | + if (*end) { | |
8425 | + *cinfo->value = value; | |
8426 | + } | |
8427 | + return count; | |
8428 | +} | |
8429 | + | |
8430 | +/* Base of the config directory for cman */ | |
8431 | +static struct proc_dir_entry *proc_cman_config; | |
8432 | +void create_proc_entries(void) | |
8433 | +{ | |
8434 | + struct proc_dir_entry *procentry; | |
8435 | + struct proc_dir_entry *proc_cluster; | |
8436 | + int i; | |
8437 | + | |
8438 | + proc_cluster = proc_mkdir("cluster", 0); | |
8439 | + if (!proc_cluster) | |
8440 | + return; | |
8441 | + proc_cluster->owner = THIS_MODULE; | |
8442 | + | |
8443 | + /* Config dir filled in by us and others */ | |
8444 | + if (!proc_mkdir("cluster/config", 0)) | |
8445 | + return; | |
8446 | + | |
8447 | + /* Don't much care if this fails, it's hardly vital */ | |
8448 | + procentry = create_proc_entry("cluster/nodes", S_IRUGO, NULL); | |
8449 | + if (procentry) | |
8450 | + procentry->proc_fops = &cluster_fops; | |
8451 | + | |
8452 | + procentry = create_proc_entry("cluster/status", S_IRUGO, NULL); | |
8453 | + if (procentry) | |
8454 | + procentry->get_info = proc_cluster_status; | |
8455 | + | |
8456 | + procentry = create_proc_entry("cluster/services", S_IRUGO, NULL); | |
8457 | + if (procentry) | |
8458 | + procentry->get_info = sm_procdata; | |
8459 | + | |
8460 | + /* Config entries */ | |
8461 | + proc_cman_config = proc_mkdir("cluster/config/cman", 0); | |
8462 | + if (!proc_cman_config) | |
8463 | + return; | |
8464 | + | |
8465 | + for (i=0; i<sizeof(config_proc)/sizeof(struct config_proc_info); i++) { | |
8466 | + procentry = create_proc_entry(config_proc[i].name, 0660, | |
8467 | + proc_cman_config); | |
8468 | + if (procentry) { | |
8469 | + procentry->data = &config_proc[i]; | |
8470 | + procentry->write_proc = cman_config_write_proc; | |
8471 | + procentry->read_proc = cman_config_read_proc; | |
8472 | + } | |
8473 | + } | |
8474 | + | |
8475 | + procentry = create_proc_entry("cluster/sm_debug", S_IRUGO, NULL); | |
8476 | + if (procentry) | |
8477 | + procentry->get_info = sm_debug_info; | |
8478 | +} | |
8479 | + | |
8480 | +void cleanup_proc_entries(void) | |
8481 | +{ | |
8482 | + int i, config_count; | |
8483 | + | |
8484 | + remove_proc_entry("cluster/sm_debug", NULL); | |
8485 | + | |
8486 | + config_count = sizeof(config_proc) / sizeof(struct config_proc_info); | |
8487 | + | |
8488 | + if (proc_cman_config) { | |
8489 | + for (i=0; i<config_count; i++) | |
8490 | + remove_proc_entry(config_proc[i].name, proc_cman_config); | |
8491 | + } | |
8492 | + remove_proc_entry("cluster/config/cman", NULL); | |
8493 | + remove_proc_entry("cluster/config", NULL); | |
8494 | + | |
8495 | + remove_proc_entry("cluster/nodes", NULL); | |
8496 | + remove_proc_entry("cluster/status", NULL); | |
8497 | + remove_proc_entry("cluster/services", NULL); | |
8498 | + remove_proc_entry("cluster/config", NULL); | |
8499 | + remove_proc_entry("cluster", NULL); | |
8500 | +} | |
8501 | diff -urN linux-orig/cluster/cman/sm.h linux-patched/cluster/cman/sm.h | |
8502 | --- linux-orig/cluster/cman/sm.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 8503 | +++ linux-patched/cluster/cman/sm.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 8504 | @@ -0,0 +1,108 @@ |
8505 | +/****************************************************************************** | |
8506 | +******************************************************************************* | |
8507 | +** | |
8508 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8509 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8510 | +** | |
8511 | +** This copyrighted material is made available to anyone wishing to use, | |
8512 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8513 | +** of the GNU General Public License v.2. | |
8514 | +** | |
8515 | +******************************************************************************* | |
8516 | +******************************************************************************/ | |
8517 | + | |
8518 | +#ifndef __SM_DOT_H__ | |
8519 | +#define __SM_DOT_H__ | |
8520 | + | |
8521 | +/* | |
8522 | + * This is the main header file to be included in each Service Manager source | |
8523 | + * file. | |
8524 | + */ | |
8525 | + | |
8526 | +#include <linux/list.h> | |
8527 | +#include <linux/socket.h> | |
8528 | +#include <linux/kernel.h> | |
8529 | +#include <linux/sched.h> | |
8530 | +#include <linux/file.h> | |
8531 | +#include <net/sock.h> | |
8532 | + | |
8533 | +#include <cluster/cnxman.h> | |
8534 | +#include <cluster/service.h> | |
8535 | + | |
8536 | +#define SG_LEVELS (4) | |
8537 | + | |
8538 | +#include "sm_internal.h" | |
8539 | +#include "sm_barrier.h" | |
8540 | +#include "sm_control.h" | |
8541 | +#include "sm_daemon.h" | |
8542 | +#include "sm_joinleave.h" | |
8543 | +#include "sm_membership.h" | |
8544 | +#include "sm_message.h" | |
8545 | +#include "sm_misc.h" | |
8546 | +#include "sm_recover.h" | |
8547 | +#include "sm_services.h" | |
8548 | + | |
8549 | +extern struct list_head sm_sg[SG_LEVELS]; | |
8550 | +extern struct semaphore sm_sglock; | |
8551 | + | |
8552 | +#ifndef TRUE | |
8553 | +#define TRUE (1) | |
8554 | +#endif | |
8555 | + | |
8556 | +#ifndef FALSE | |
8557 | +#define FALSE (0) | |
8558 | +#endif | |
8559 | + | |
8560 | +#define SM_ASSERT(x, do) \ | |
8561 | +{ \ | |
8562 | + if (!(x)) \ | |
8563 | + { \ | |
8564 | + printk("\nSM: Assertion failed on line %d of file %s\n" \ | |
8565 | + "SM: assertion: \"%s\"\n" \ | |
8566 | + "SM: time = %lu\n", \ | |
8567 | + __LINE__, __FILE__, #x, jiffies); \ | |
8568 | + {do} \ | |
8569 | + printk("\n"); \ | |
8570 | + panic("SM: Record message above and reboot.\n"); \ | |
8571 | + } \ | |
8572 | +} | |
8573 | + | |
8574 | +#define SM_RETRY(do_this, until_this) \ | |
8575 | +for (;;) \ | |
8576 | +{ \ | |
8577 | + do { do_this; } while (0); \ | |
8578 | + if (until_this) \ | |
8579 | + break; \ | |
8580 | + printk("SM: out of memory: %s, %u\n", __FILE__, __LINE__); \ | |
8581 | + schedule();\ | |
8582 | +} | |
8583 | + | |
8584 | + | |
8585 | +#define log_print(fmt, args...) printk("SM: "fmt"\n", ##args) | |
8586 | + | |
8587 | +#define log_error(sg, fmt, args...) \ | |
8588 | + printk("SM: %08x " fmt "\n", (sg)->global_id , ##args) | |
8589 | + | |
8590 | + | |
8591 | +#define SM_DEBUG_LOG | |
8592 | + | |
8593 | +#ifdef SM_DEBUG_CONSOLE | |
8594 | +#define log_debug(sg, fmt, args...) \ | |
8595 | + printk("SM: %08x " fmt "\n", (sg)->global_id , ##args) | |
8596 | +#endif | |
8597 | + | |
8598 | +#ifdef SM_DEBUG_LOG | |
8599 | +#define log_debug(sg, fmt, args...) sm_debug_log(sg, fmt, ##args); | |
8600 | +#endif | |
8601 | + | |
8602 | +#ifdef SM_DEBUG_ALL | |
8603 | +#define log_debug(sg, fmt, args...) \ | |
8604 | +do \ | |
8605 | +{ \ | |
8606 | + printk("SM: %08x "fmt"\n", (sg)->global_id, ##args); \ | |
8607 | + sm_debug_log(sg, fmt, ##args); \ | |
8608 | +} \ | |
8609 | +while (0) | |
8610 | +#endif | |
8611 | + | |
8612 | +#endif /* __SM_DOT_H__ */ | |
8613 | diff -urN linux-orig/cluster/cman/sm_barrier.c linux-patched/cluster/cman/sm_barrier.c | |
8614 | --- linux-orig/cluster/cman/sm_barrier.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 8615 | +++ linux-patched/cluster/cman/sm_barrier.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 8616 | @@ -0,0 +1,232 @@ |
8617 | +/****************************************************************************** | |
8618 | +******************************************************************************* | |
8619 | +** | |
8620 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8621 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8622 | +** | |
8623 | +** This copyrighted material is made available to anyone wishing to use, | |
8624 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8625 | +** of the GNU General Public License v.2. | |
8626 | +** | |
8627 | +******************************************************************************* | |
8628 | +******************************************************************************/ | |
8629 | + | |
8630 | +#include "sm.h" | |
8631 | + | |
8632 | +static struct list_head barriers; | |
8633 | +static spinlock_t barriers_lock; | |
8634 | + | |
8635 | +struct bc_entry { | |
8636 | + struct list_head list; | |
8637 | + uint32_t gid; | |
8638 | + int status; | |
8639 | + char type; | |
8640 | +}; | |
8641 | +typedef struct bc_entry bc_entry_t; | |
8642 | + | |
8643 | +void init_barriers(void) | |
8644 | +{ | |
8645 | + INIT_LIST_HEAD(&barriers); | |
8646 | + spin_lock_init(&barriers_lock); | |
8647 | +} | |
8648 | + | |
8649 | +static int atoi(char *c) | |
8650 | +{ | |
8651 | + int x = 0; | |
8652 | + | |
8653 | + while ('0' <= *c && *c <= '9') { | |
8654 | + x = x * 10 + (*c - '0'); | |
8655 | + c++; | |
8656 | + } | |
8657 | + return x; | |
8658 | +} | |
8659 | + | |
8660 | +static void add_barrier_callback(char *name, int status, int type) | |
8661 | +{ | |
8662 | + char *p; | |
8663 | + uint32_t gid; | |
8664 | + bc_entry_t *be; | |
8665 | + | |
8666 | + /* an ESRCH callback just means there was a cnxman transition */ | |
8667 | + if (status == -ESRCH) | |
8668 | + return; | |
8669 | + | |
8670 | + /* extract global id of SG from barrier name */ | |
8671 | + p = strstr(name, "sm."); | |
8672 | + | |
8673 | + SM_ASSERT(p, printk("name=\"%s\" status=%d\n", name, status);); | |
8674 | + | |
8675 | + p += strlen("sm."); | |
8676 | + gid = atoi(p); | |
8677 | + | |
8678 | + SM_RETRY(be = kmalloc(sizeof(bc_entry_t), GFP_ATOMIC), be); | |
8679 | + | |
8680 | + be->gid = gid; | |
8681 | + be->status = status; | |
8682 | + be->type = type; | |
8683 | + | |
8684 | + spin_lock(&barriers_lock); | |
8685 | + list_add_tail(&be->list, &barriers); | |
8686 | + spin_unlock(&barriers_lock); | |
8687 | + | |
8688 | + wake_serviced(DO_BARRIERS); | |
8689 | +} | |
8690 | + | |
8691 | +static void callback_recovery_barrier(char *name, int status) | |
8692 | +{ | |
8693 | + add_barrier_callback(name, status, SM_BARRIER_RECOVERY); | |
8694 | +} | |
8695 | + | |
8696 | +static void callback_startdone_barrier_new(char *name, int status) | |
8697 | +{ | |
8698 | + add_barrier_callback(name, status, SM_BARRIER_STARTDONE_NEW); | |
8699 | +} | |
8700 | + | |
8701 | +static void callback_startdone_barrier(char *name, int status) | |
8702 | +{ | |
8703 | + add_barrier_callback(name, status, SM_BARRIER_STARTDONE); | |
8704 | +} | |
8705 | + | |
8706 | +int sm_barrier(char *name, int count, int type) | |
8707 | +{ | |
8708 | + int error; | |
8709 | + unsigned long fn = 0; | |
8710 | + | |
8711 | + switch (type) { | |
8712 | + case SM_BARRIER_STARTDONE: | |
8713 | + fn = (unsigned long) callback_startdone_barrier; | |
8714 | + break; | |
8715 | + case SM_BARRIER_STARTDONE_NEW: | |
8716 | + fn = (unsigned long) callback_startdone_barrier_new; | |
8717 | + break; | |
8718 | + case SM_BARRIER_RECOVERY: | |
8719 | + fn = (unsigned long) callback_recovery_barrier; | |
8720 | + break; | |
8721 | + } | |
8722 | + | |
8723 | + error = kcl_barrier_register(name, 0, count); | |
8724 | + if (error) { | |
8725 | + log_print("barrier register error %d", error); | |
8726 | + goto fail; | |
8727 | + } | |
8728 | + | |
8729 | + error = kcl_barrier_setattr(name, BARRIER_SETATTR_AUTODELETE, TRUE); | |
8730 | + if (error) { | |
8731 | + log_print("barrier setattr autodel error %d", error); | |
8732 | + goto fail_bar; | |
8733 | + } | |
8734 | + | |
8735 | + error = kcl_barrier_setattr(name, BARRIER_SETATTR_CALLBACK, fn); | |
8736 | + if (error) { | |
8737 | + log_print("barrier setattr cb error %d", error); | |
8738 | + goto fail_bar; | |
8739 | + } | |
8740 | + | |
8741 | + error = kcl_barrier_setattr(name, BARRIER_SETATTR_ENABLED, TRUE); | |
8742 | + if (error) { | |
8743 | + log_print("barrier setattr enabled error %d", error); | |
8744 | + goto fail_bar; | |
8745 | + } | |
8746 | + | |
8747 | + return 0; | |
8748 | + | |
8749 | + fail_bar: | |
8750 | + kcl_barrier_delete(name); | |
8751 | + fail: | |
8752 | + return error; | |
8753 | +} | |
8754 | + | |
8755 | +void process_startdone_barrier_new(sm_group_t *sg, int status) | |
8756 | +{ | |
8757 | + sm_sevent_t *sev = sg->sevent; | |
8758 | + | |
8759 | + if (!test_and_clear_bit(SEFL_ALLOW_BARRIER, &sev->se_flags)) { | |
8760 | + log_debug(sev->se_sg, "ignore barrier cb status %d", status); | |
8761 | + return; | |
8762 | + } | |
8763 | + | |
8764 | + sev->se_barrier_status = status; | |
8765 | + sev->se_state = SEST_BARRIER_DONE; | |
8766 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
8767 | + wake_serviced(DO_JOINLEAVE); | |
8768 | +} | |
8769 | + | |
8770 | +void process_startdone_barrier(sm_group_t *sg, int status) | |
8771 | +{ | |
8772 | + sm_uevent_t *uev = &sg->uevent; | |
8773 | + | |
8774 | + if (!test_and_clear_bit(UEFL_ALLOW_BARRIER, &uev->ue_flags)) { | |
8775 | + log_debug(sg, "ignore barrier cb status %d", status); | |
8776 | + return; | |
8777 | + } | |
8778 | + | |
8779 | + uev->ue_barrier_status = status; | |
8780 | + uev->ue_state = UEST_BARRIER_DONE; | |
8781 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
8782 | + wake_serviced(DO_MEMBERSHIP); | |
8783 | +} | |
8784 | + | |
8785 | +void process_recovery_barrier(sm_group_t *sg, int status) | |
8786 | +{ | |
8787 | + if (status) { | |
8788 | + log_error(sg, "process_recovery_barrier status=%d", status); | |
8789 | + return; | |
8790 | + } | |
8791 | + | |
8792 | + if (sg->state != SGST_RECOVER || | |
8793 | + sg->recover_state != RECOVER_BARRIERWAIT) { | |
8794 | + log_error(sg, "process_recovery_barrier state %d recover %d", | |
8795 | + sg->state, sg->recover_state); | |
8796 | + return; | |
8797 | + } | |
8798 | + | |
8799 | + if (!sg->recover_stop) | |
8800 | + sg->recover_state = RECOVER_STOP; | |
8801 | + else | |
8802 | + sg->recover_state = RECOVER_BARRIERDONE; | |
8803 | + | |
8804 | + wake_serviced(DO_RECOVERIES); | |
8805 | +} | |
8806 | + | |
8807 | +void process_barriers(void) | |
8808 | +{ | |
8809 | + sm_group_t *sg; | |
8810 | + bc_entry_t *be; | |
8811 | + | |
8812 | + while (1) { | |
8813 | + be = NULL; | |
8814 | + | |
8815 | + spin_lock(&barriers_lock); | |
8816 | + if (!list_empty(&barriers)) { | |
8817 | + be = list_entry(barriers.next, bc_entry_t, list); | |
8818 | + list_del(&be->list); | |
8819 | + } | |
8820 | + spin_unlock(&barriers_lock); | |
8821 | + | |
8822 | + if (!be) | |
8823 | + break; | |
8824 | + | |
8825 | + sg = sm_global_id_to_sg(be->gid); | |
8826 | + if (!sg) { | |
8827 | + log_print("process_barriers: no sg %08x", be->gid); | |
8828 | + break; | |
8829 | + } | |
8830 | + | |
8831 | + switch (be->type) { | |
8832 | + case SM_BARRIER_STARTDONE_NEW: | |
8833 | + process_startdone_barrier_new(sg, be->status); | |
8834 | + break; | |
8835 | + | |
8836 | + case SM_BARRIER_STARTDONE: | |
8837 | + process_startdone_barrier(sg, be->status); | |
8838 | + break; | |
8839 | + | |
8840 | + case SM_BARRIER_RECOVERY: | |
8841 | + process_recovery_barrier(sg, be->status); | |
8842 | + break; | |
8843 | + } | |
8844 | + | |
8845 | + kfree(be); | |
8846 | + schedule(); | |
8847 | + } | |
8848 | +} | |
8849 | diff -urN linux-orig/cluster/cman/sm_barrier.h linux-patched/cluster/cman/sm_barrier.h | |
8850 | --- linux-orig/cluster/cman/sm_barrier.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 8851 | +++ linux-patched/cluster/cman/sm_barrier.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 8852 | @@ -0,0 +1,29 @@ |
8853 | +/****************************************************************************** | |
8854 | +******************************************************************************* | |
8855 | +** | |
8856 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8857 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8858 | +** | |
8859 | +** This copyrighted material is made available to anyone wishing to use, | |
8860 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8861 | +** of the GNU General Public License v.2. | |
8862 | +** | |
8863 | +******************************************************************************* | |
8864 | +******************************************************************************/ | |
8865 | + | |
8866 | +#ifndef __SM_BARRIER_DOT_H__ | |
8867 | +#define __SM_BARRIER_DOT_H__ | |
8868 | + | |
8869 | +#define SM_BARRIER_STARTDONE (0) | |
8870 | +#define SM_BARRIER_STARTDONE_NEW (1) | |
8871 | +#define SM_BARRIER_RECOVERY (2) | |
8872 | +#define SM_BARRIER_RESET (3) | |
8873 | + | |
8874 | +void init_barriers(void); | |
8875 | +void process_barriers(void); | |
8876 | +int sm_barrier(char *name, int count, int type); | |
8877 | +void process_startdone_barrier(sm_group_t *sg, int status); | |
8878 | +void process_startdone_barrier_new(sm_group_t *sg, int status); | |
8879 | +void process_recovery_barrier(sm_group_t *sg, int status); | |
8880 | + | |
8881 | +#endif | |
8882 | diff -urN linux-orig/cluster/cman/sm_control.c linux-patched/cluster/cman/sm_control.c | |
8883 | --- linux-orig/cluster/cman/sm_control.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 8884 | +++ linux-patched/cluster/cman/sm_control.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 8885 | @@ -0,0 +1,156 @@ |
8886 | +/****************************************************************************** | |
8887 | +******************************************************************************* | |
8888 | +** | |
8889 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
8890 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
8891 | +** | |
8892 | +** This copyrighted material is made available to anyone wishing to use, | |
8893 | +** modify, copy, or redistribute it subject to the terms and conditions | |
8894 | +** of the GNU General Public License v.2. | |
8895 | +** | |
8896 | +******************************************************************************* | |
8897 | +******************************************************************************/ | |
8898 | + | |
8899 | +#include "sm.h" | |
8900 | +#include "config.h" | |
8901 | + | |
8902 | +struct socket * sm_socket; | |
8903 | +uint32_t * sm_new_nodeids; | |
8904 | +uint32_t sm_our_nodeid; | |
8905 | +int sm_quorum, sm_quorum_next; | |
8906 | +struct list_head sm_members; | |
8907 | +int sm_member_count; | |
8908 | + | |
8909 | + | |
8910 | +/* | |
8911 | + * Context: cnxman | |
8912 | + * Called by cnxman when it has a new member list. | |
8913 | + */ | |
8914 | + | |
8915 | +void sm_member_update(int quorate) | |
8916 | +{ | |
8917 | + sm_quorum_next = quorate; | |
8918 | + wake_serviced(DO_START_RECOVERY); | |
8919 | +} | |
8920 | + | |
8921 | +/* | |
8922 | + * Context: cnxman | |
8923 | + * Called when module is loaded. | |
8924 | + */ | |
8925 | + | |
8926 | +void sm_init(void) | |
8927 | +{ | |
8928 | + sm_socket = NULL; | |
8929 | + sm_new_nodeids = NULL; | |
8930 | + sm_quorum = 0; | |
8931 | + sm_quorum_next = 0; | |
8932 | + sm_our_nodeid = 0; | |
8933 | + INIT_LIST_HEAD(&sm_members); | |
8934 | + sm_member_count = 0; | |
8935 | + | |
8936 | + init_services(); | |
8937 | + init_messages(); | |
8938 | + init_barriers(); | |
8939 | + init_serviced(); | |
8940 | + init_recovery(); | |
8941 | + init_joinleave(); | |
8942 | + init_sm_misc(); | |
8943 | +} | |
8944 | + | |
8945 | +/* | |
8946 | + * Context: cnxman | |
8947 | + * Called at beginning of cluster join procedure. | |
8948 | + */ | |
8949 | + | |
8950 | +void sm_start(void) | |
8951 | +{ | |
8952 | + struct sockaddr_cl saddr; | |
8953 | + struct socket *sock; | |
8954 | + int result; | |
8955 | + | |
8956 | + /* Create a communication channel among service managers */ | |
8957 | + | |
8958 | + result = sock_create_kern(AF_CLUSTER, SOCK_DGRAM, CLPROTO_CLIENT, &sock); | |
8959 | + if (result < 0) { | |
8960 | + log_print("can't create socket %d", result); | |
8961 | + goto fail; | |
8962 | + } | |
8963 | + | |
8964 | + sm_socket = sock; | |
8965 | + | |
8966 | + saddr.scl_family = AF_CLUSTER; | |
8967 | + saddr.scl_port = CLUSTER_PORT_SERVICES; | |
8968 | + | |
8969 | + result = sock->ops->bind(sock, (struct sockaddr *) &saddr, | |
8970 | + sizeof(saddr)); | |
8971 | + if (result < 0) { | |
8972 | + log_print("can't bind socket %d", result); | |
8973 | + goto fail_release; | |
8974 | + } | |
8975 | + | |
8976 | + result = kcl_register_read_callback(sm_socket, sm_cluster_message); | |
8977 | + if (result < 0) { | |
8978 | + log_print("can't register read callback %d", result); | |
8979 | + goto fail_release; | |
8980 | + } | |
8981 | + | |
8982 | + sm_new_nodeids = (uint32_t *) kmalloc(cman_config.max_nodes * | |
8983 | + sizeof(uint32_t), | |
8984 | + GFP_KERNEL); | |
8985 | + start_serviced(); | |
8986 | + | |
8987 | + /* cnxman should call sm_member_update() once we've joined - then we | |
8988 | + * can get our first list of members and our own nodeid */ | |
8989 | + | |
8990 | + return; | |
8991 | + | |
8992 | + fail_release: | |
8993 | + sock_release(sm_socket); | |
8994 | + sm_socket = NULL; | |
8995 | + | |
8996 | + fail: | |
8997 | + return; | |
8998 | +} | |
8999 | + | |
9000 | +/* | |
9001 | + * Context: cnxman | |
9002 | + * Called before cnxman leaves the cluster. If this returns an error to cman, | |
9003 | + * cman should not leave the cluster but return EBUSY. | |
9004 | + * If force is set we go away anyway. cman knows best in this case | |
9005 | + */ | |
9006 | + | |
9007 | +int sm_stop(int force) | |
9008 | +{ | |
9009 | + struct list_head *head; | |
9010 | + sm_group_t *sg; | |
9011 | + sm_node_t *node; | |
9012 | + int i, busy = FALSE, error = -EBUSY; | |
9013 | + | |
9014 | + for (i = 0; i < SG_LEVELS; i++) { | |
9015 | + if (!list_empty(&sm_sg[i])) { | |
9016 | + sg = list_entry(sm_sg[i].next, sm_group_t, list); | |
9017 | + log_error(sg, "sm_stop: SG still joined"); | |
9018 | + busy = TRUE; | |
9019 | + } | |
9020 | + } | |
9021 | + | |
9022 | + if (!busy || force) { | |
9023 | + stop_serviced(); | |
9024 | + | |
9025 | + if (sm_socket) | |
9026 | + sock_release(sm_socket); | |
9027 | + | |
9028 | + head = &sm_members; | |
9029 | + while (!list_empty(head)) { | |
9030 | + node = list_entry(head->next, sm_node_t, list); | |
9031 | + list_del(&node->list); | |
9032 | + sm_member_count--; | |
9033 | + kfree(node); | |
9034 | + } | |
9035 | + | |
9036 | + kfree(sm_new_nodeids); | |
9037 | + sm_init(); | |
9038 | + error = 0; | |
9039 | + } | |
9040 | + return error; | |
9041 | +} | |
9042 | diff -urN linux-orig/cluster/cman/sm_control.h linux-patched/cluster/cman/sm_control.h | |
9043 | --- linux-orig/cluster/cman/sm_control.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 9044 | +++ linux-patched/cluster/cman/sm_control.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 9045 | @@ -0,0 +1,22 @@ |
9046 | +/****************************************************************************** | |
9047 | +******************************************************************************* | |
9048 | +** | |
9049 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9050 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9051 | +** | |
9052 | +** This copyrighted material is made available to anyone wishing to use, | |
9053 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9054 | +** of the GNU General Public License v.2. | |
9055 | +** | |
9056 | +******************************************************************************* | |
9057 | +******************************************************************************/ | |
9058 | + | |
9059 | +#ifndef __SM_CONTROL_DOT_H__ | |
9060 | +#define __SM_CONTROL_DOT_H__ | |
9061 | + | |
9062 | +void sm_init(void); | |
9063 | +void sm_start(void); | |
9064 | +int sm_stop(int force); | |
9065 | +void sm_member_update(int quorate); | |
9066 | + | |
9067 | +#endif | |
9068 | diff -urN linux-orig/cluster/cman/sm_daemon.c linux-patched/cluster/cman/sm_daemon.c | |
9069 | --- linux-orig/cluster/cman/sm_daemon.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 9070 | +++ linux-patched/cluster/cman/sm_daemon.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 9071 | @@ -0,0 +1,120 @@ |
9072 | +/****************************************************************************** | |
9073 | +******************************************************************************* | |
9074 | +** | |
9075 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9076 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9077 | +** | |
9078 | +** This copyrighted material is made available to anyone wishing to use, | |
9079 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9080 | +** of the GNU General Public License v.2. | |
9081 | +** | |
9082 | +******************************************************************************* | |
9083 | +******************************************************************************/ | |
9084 | + | |
9085 | +#include "sm.h" | |
9086 | + | |
9087 | +static unsigned long daemon_flags; | |
9088 | +static struct task_struct * daemon_task; | |
9089 | +static struct completion daemon_done; | |
9090 | +static wait_queue_head_t daemon_wait; | |
9091 | +extern int sm_quorum; | |
9092 | + | |
9093 | +void init_serviced(void) | |
9094 | +{ | |
9095 | + daemon_flags = 0; | |
9096 | + daemon_task = NULL; | |
9097 | + init_completion(&daemon_done); | |
9098 | + init_waitqueue_head(&daemon_wait); | |
9099 | +} | |
9100 | + | |
9101 | +void wake_serviced(int do_flag) | |
9102 | +{ | |
9103 | + set_bit(do_flag, &daemon_flags); | |
9104 | + wake_up(&daemon_wait); | |
9105 | +} | |
9106 | + | |
9107 | +static inline int got_work(void) | |
9108 | +{ | |
9109 | + int rv = 0; | |
9110 | + | |
9111 | + rv = (test_bit(DO_START_RECOVERY, &daemon_flags) || | |
9112 | + test_bit(DO_MESSAGES, &daemon_flags) || | |
9113 | + test_bit(DO_BARRIERS, &daemon_flags) || | |
9114 | + test_bit(DO_CALLBACKS, &daemon_flags)); | |
9115 | + | |
9116 | + if (sm_quorum && !rv) | |
9117 | + rv = (test_bit(DO_JOINLEAVE, &daemon_flags) || | |
9118 | + test_bit(DO_RECOVERIES, &daemon_flags) || | |
9119 | + test_bit(DO_MEMBERSHIP, &daemon_flags)); | |
9120 | + return rv; | |
9121 | +} | |
9122 | + | |
9123 | +static int serviced(void *arg) | |
9124 | +{ | |
9125 | + DECLARE_WAITQUEUE(wait, current); | |
9126 | + | |
9127 | + daemonize("cman_serviced"); | |
9128 | + daemon_task = current; | |
9129 | + set_bit(DO_RUN, &daemon_flags); | |
9130 | + complete(&daemon_done); | |
9131 | + | |
9132 | + for (;;) { | |
9133 | + if (test_and_clear_bit(DO_START_RECOVERY, &daemon_flags)) | |
9134 | + process_nodechange(); | |
9135 | + | |
9136 | + if (test_and_clear_bit(DO_MESSAGES, &daemon_flags)) | |
9137 | + process_messages(); | |
9138 | + | |
9139 | + if (test_and_clear_bit(DO_BARRIERS, &daemon_flags)) | |
9140 | + process_barriers(); | |
9141 | + | |
9142 | + if (test_and_clear_bit(DO_CALLBACKS, &daemon_flags)) | |
9143 | + process_callbacks(); | |
9144 | + | |
9145 | + if (sm_quorum) { | |
9146 | + if (test_and_clear_bit(DO_RECOVERIES, &daemon_flags)) | |
9147 | + process_recoveries(); | |
9148 | + | |
9149 | + if (test_and_clear_bit(DO_JOINLEAVE, &daemon_flags)) | |
9150 | + process_joinleave(); | |
9151 | + | |
9152 | + if (test_and_clear_bit(DO_MEMBERSHIP, &daemon_flags)) | |
9153 | + process_membership(); | |
9154 | + } | |
9155 | + | |
9156 | + if (!test_bit(DO_RUN, &daemon_flags)) | |
9157 | + break; | |
9158 | + | |
9159 | + current->state = TASK_INTERRUPTIBLE; | |
9160 | + add_wait_queue(&daemon_wait, &wait); | |
9161 | + if (!got_work() && test_bit(DO_RUN, &daemon_flags)) | |
9162 | + schedule(); | |
9163 | + remove_wait_queue(&daemon_wait, &wait); | |
9164 | + current->state = TASK_RUNNING; | |
9165 | + } | |
9166 | + | |
9167 | + complete(&daemon_done); | |
9168 | + return 0; | |
9169 | +} | |
9170 | + | |
9171 | +int start_serviced(void) | |
9172 | +{ | |
9173 | + int error; | |
9174 | + | |
9175 | + error = kernel_thread(serviced, NULL, 0); | |
9176 | + if (error < 0) | |
9177 | + goto out; | |
9178 | + | |
9179 | + error = 0; | |
9180 | + wait_for_completion(&daemon_done); | |
9181 | + | |
9182 | + out: | |
9183 | + return error; | |
9184 | +} | |
9185 | + | |
9186 | +void stop_serviced(void) | |
9187 | +{ | |
9188 | + clear_bit(DO_RUN, &daemon_flags); | |
9189 | + wake_up(&daemon_wait); | |
9190 | + wait_for_completion(&daemon_done); | |
9191 | +} | |
9192 | diff -urN linux-orig/cluster/cman/sm_daemon.h linux-patched/cluster/cman/sm_daemon.h | |
9193 | --- linux-orig/cluster/cman/sm_daemon.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 9194 | +++ linux-patched/cluster/cman/sm_daemon.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 9195 | @@ -0,0 +1,32 @@ |
9196 | +/****************************************************************************** | |
9197 | +******************************************************************************* | |
9198 | +** | |
9199 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9200 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9201 | +** | |
9202 | +** This copyrighted material is made available to anyone wishing to use, | |
9203 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9204 | +** of the GNU General Public License v.2. | |
9205 | +** | |
9206 | +******************************************************************************* | |
9207 | +******************************************************************************/ | |
9208 | + | |
9209 | +#ifndef __SM_DAEMON_DOT_H__ | |
9210 | +#define __SM_DAEMON_DOT_H__ | |
9211 | + | |
9212 | +#define DO_RUN (0) | |
9213 | +#define DO_START_RECOVERY (1) | |
9214 | +#define DO_MESSAGES (2) | |
9215 | +#define DO_BARRIERS (3) | |
9216 | +#define DO_CALLBACKS (4) | |
9217 | +#define DO_JOINLEAVE (5) | |
9218 | +#define DO_RECOVERIES (6) | |
9219 | +#define DO_MEMBERSHIP (7) | |
9220 | +#define DO_RESET (8) | |
9221 | + | |
9222 | +void init_serviced(void); | |
9223 | +void wake_serviced(int do_flag); | |
9224 | +void stop_serviced(void); | |
9225 | +int start_serviced(void); | |
9226 | + | |
9227 | +#endif | |
9228 | diff -urN linux-orig/cluster/cman/sm_internal.h linux-patched/cluster/cman/sm_internal.h | |
9229 | --- linux-orig/cluster/cman/sm_internal.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 9230 | +++ linux-patched/cluster/cman/sm_internal.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 9231 | @@ -0,0 +1,230 @@ |
9232 | +/****************************************************************************** | |
9233 | +******************************************************************************* | |
9234 | +** | |
9235 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9236 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9237 | +** | |
9238 | +** This copyrighted material is made available to anyone wishing to use, | |
9239 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9240 | +** of the GNU General Public License v.2. | |
9241 | +** | |
9242 | +******************************************************************************* | |
9243 | +******************************************************************************/ | |
9244 | + | |
9245 | +#ifndef __SM_INTERNAL_DOT_H__ | |
9246 | +#define __SM_INTERNAL_DOT_H__ | |
9247 | + | |
9248 | +/* | |
9249 | + * Any header files needed by this file should be included before it in sm.h. | |
9250 | + * This file should only be included by sm.h. | |
9251 | + */ | |
9252 | + | |
9253 | +struct sm_group; | |
9254 | +struct sm_sevent; | |
9255 | +struct sm_uevent; | |
9256 | +struct sm_node; | |
9257 | +struct sm_msg; | |
9258 | + | |
9259 | +typedef struct sm_group sm_group_t; | |
9260 | +typedef struct sm_sevent sm_sevent_t; | |
9261 | +typedef struct sm_uevent sm_uevent_t; | |
9262 | +typedef struct sm_node sm_node_t; | |
9263 | +typedef struct sm_msg sm_msg_t; | |
9264 | + | |
9265 | + | |
9266 | +/* | |
9267 | + * Number of seconds to wait before trying again to join or leave an SG | |
9268 | + */ | |
9269 | +#define RETRY_DELAY (2) | |
9270 | + | |
9271 | + | |
9272 | +/* | |
9273 | + * Service Event - what a node uses to join or leave an sg | |
9274 | + */ | |
9275 | + | |
9276 | +/* SE Flags */ | |
9277 | +#define SEFL_CHECK (0) | |
9278 | +#define SEFL_ALLOW_JOIN (1) | |
9279 | +#define SEFL_ALLOW_JSTOP (2) | |
9280 | +#define SEFL_ALLOW_LEAVE (3) | |
9281 | +#define SEFL_ALLOW_LSTOP (4) | |
9282 | +#define SEFL_ALLOW_STARTDONE (5) | |
9283 | +#define SEFL_ALLOW_BARRIER (6) | |
9284 | +#define SEFL_DELAY (7) | |
9285 | +#define SEFL_LEAVE (8) | |
9286 | +#define SEFL_CANCEL (9) | |
9287 | + | |
9288 | +/* SE States */ | |
9289 | +#define SEST_JOIN_BEGIN (1) | |
9290 | +#define SEST_JOIN_ACKWAIT (2) | |
9291 | +#define SEST_JOIN_ACKED (3) | |
9292 | +#define SEST_JSTOP_ACKWAIT (4) | |
9293 | +#define SEST_JSTOP_ACKED (5) | |
9294 | +#define SEST_JSTART_SERVICEWAIT (6) | |
9295 | +#define SEST_JSTART_SERVICEDONE (7) | |
9296 | +#define SEST_BARRIER_WAIT (8) | |
9297 | +#define SEST_BARRIER_DONE (9) | |
9298 | +#define SEST_LEAVE_BEGIN (10) | |
9299 | +#define SEST_LEAVE_ACKWAIT (11) | |
9300 | +#define SEST_LEAVE_ACKED (12) | |
9301 | +#define SEST_LSTOP_ACKWAIT (13) | |
9302 | +#define SEST_LSTOP_ACKED (14) | |
9303 | +#define SEST_LSTART_WAITREMOTE (15) | |
9304 | +#define SEST_LSTART_REMOTEDONE (16) | |
9305 | + | |
9306 | +struct sm_sevent { | |
9307 | + struct list_head se_list; | |
9308 | + unsigned int se_id; | |
9309 | + sm_group_t * se_sg; | |
9310 | + unsigned long se_flags; | |
9311 | + unsigned int se_state; | |
9312 | + | |
9313 | + int se_node_count; | |
9314 | + int se_memb_count; | |
9315 | + int se_reply_count; | |
9316 | + | |
9317 | + uint32_t * se_node_ids; | |
9318 | + char * se_node_status; | |
9319 | + int se_len_ids; /* length of node_ids */ | |
9320 | + int se_len_status; /* length of node_status */ | |
9321 | + | |
9322 | + int se_barrier_status; | |
9323 | + struct timer_list se_restart_timer; | |
9324 | +}; | |
9325 | + | |
9326 | +/* | |
9327 | + * Update Event - what an sg member uses to respond to an sevent | |
9328 | + */ | |
9329 | + | |
9330 | +/* UE Flags */ | |
9331 | +#define UEFL_ALLOW_STARTDONE (0) | |
9332 | +#define UEFL_ALLOW_BARRIER (1) | |
9333 | +#define UEFL_CANCEL (2) | |
9334 | +#define UEFL_LEAVE (3) | |
9335 | +#define UEFL_CHECK (4) | |
9336 | + | |
9337 | +/* UE States */ | |
9338 | +#define UEST_JSTOP (1) | |
9339 | +#define UEST_JSTART_WAITCMD (2) | |
9340 | +#define UEST_JSTART (3) | |
9341 | +#define UEST_JSTART_SERVICEWAIT (4) | |
9342 | +#define UEST_JSTART_SERVICEDONE (5) | |
9343 | +#define UEST_BARRIER_WAIT (6) | |
9344 | +#define UEST_BARRIER_DONE (7) | |
9345 | +#define UEST_LSTOP (8) | |
9346 | +#define UEST_LSTART_WAITCMD (9) | |
9347 | +#define UEST_LSTART (10) | |
9348 | +#define UEST_LSTART_SERVICEWAIT (11) | |
9349 | +#define UEST_LSTART_SERVICEDONE (12) | |
9350 | + | |
9351 | +struct sm_uevent { | |
9352 | + unsigned int ue_state; | |
9353 | + unsigned long ue_flags; | |
9354 | + uint32_t ue_id; | |
9355 | + uint32_t ue_nodeid; | |
9356 | + int ue_num_nodes; | |
9357 | + int ue_barrier_status; | |
9358 | + uint16_t ue_remote_seid; | |
9359 | +}; | |
9360 | + | |
9361 | +/* | |
9362 | + * Service Group | |
9363 | + */ | |
9364 | + | |
9365 | +#define RECOVER_NONE (0) | |
9366 | +#define RECOVER_STOP (1) | |
9367 | +#define RECOVER_START (2) | |
9368 | +#define RECOVER_STARTDONE (3) | |
9369 | +#define RECOVER_BARRIERWAIT (4) | |
9370 | +#define RECOVER_BARRIERDONE (5) | |
9371 | + | |
9372 | +/* SG Flags */ | |
9373 | +#define SGFL_SEVENT (1) | |
9374 | +#define SGFL_UEVENT (2) | |
9375 | +#define SGFL_NEED_RECOVERY (3) | |
9376 | + | |
9377 | +/* SG States */ | |
9378 | +#define SGST_NONE (0) | |
9379 | +#define SGST_JOIN (1) | |
9380 | +#define SGST_RUN (2) | |
9381 | +#define SGST_RECOVER (3) | |
9382 | +#define SGST_UEVENT (4) | |
9383 | + | |
9384 | +struct sm_group { | |
9385 | + struct list_head list; /* list of sg's */ | |
9386 | + uint16_t level; | |
9387 | + uint32_t local_id; | |
9388 | + uint32_t global_id; | |
9389 | + unsigned long flags; | |
9390 | + int state; | |
9391 | + int refcount; /* references from reg/unreg */ | |
9392 | + void * service_data; /* data from the service */ | |
9393 | + struct kcl_service_ops *ops; /* ops from the service */ | |
9394 | + struct completion event_comp; | |
9395 | + | |
9396 | + struct list_head memb; /* Membership List for RC */ | |
9397 | + int memb_count; /* number of nodes in memb */ | |
9398 | + struct list_head joining; /* nodes joining the sg */ | |
9399 | + sm_sevent_t * sevent; | |
9400 | + sm_uevent_t uevent; | |
9401 | + | |
9402 | + int recover_state; | |
9403 | + int recover_stop; | |
9404 | + struct list_head recover_list; /* recovery event list */ | |
9405 | + void * recover_data; | |
9406 | + char recover_barrier[MAX_BARRIER_NAME_LEN]; | |
9407 | + | |
9408 | + int namelen; | |
9409 | + char name[1]; /* must be last field */ | |
9410 | +}; | |
9411 | + | |
9412 | +/* | |
9413 | + * Service Message | |
9414 | + */ | |
9415 | + | |
9416 | +/* SMSG Type */ | |
9417 | +#define SMSG_JOIN_REQ (1) | |
9418 | +#define SMSG_JOIN_REP (2) | |
9419 | +#define SMSG_JSTOP_REQ (3) | |
9420 | +#define SMSG_JSTOP_REP (4) | |
9421 | +#define SMSG_JSTART_CMD (5) | |
9422 | +#define SMSG_LEAVE_REQ (6) | |
9423 | +#define SMSG_LEAVE_REP (7) | |
9424 | +#define SMSG_LSTOP_REQ (8) | |
9425 | +#define SMSG_LSTOP_REP (9) | |
9426 | +#define SMSG_LSTART_CMD (10) | |
9427 | +#define SMSG_LSTART_DONE (11) | |
9428 | +#define SMSG_RECOVER (12) | |
9429 | + | |
9430 | +/* SMSG Status */ | |
9431 | +#define STATUS_POS (1) | |
9432 | +#define STATUS_NEG (2) | |
9433 | +#define STATUS_WAIT (3) | |
9434 | + | |
9435 | +struct sm_msg { | |
9436 | + uint8_t ms_type; | |
9437 | + uint8_t ms_status; | |
9438 | + uint16_t ms_sevent_id; | |
9439 | + uint32_t ms_global_sgid; | |
9440 | + uint32_t ms_global_lastid; | |
9441 | + uint16_t ms_sglevel; | |
9442 | + uint16_t ms_length; | |
9443 | + /* buf of ms_length bytes follows */ | |
9444 | +}; | |
9445 | + | |
9446 | +/* | |
9447 | + * Node structure | |
9448 | + */ | |
9449 | + | |
9450 | +#define SNFL_NEED_RECOVERY (0) | |
9451 | +#define SNFL_CLUSTER_MEMBER (1) | |
9452 | +#define SNFL_LEAVING (2) | |
9453 | + | |
9454 | +struct sm_node { | |
9455 | + struct list_head list; | |
9456 | + uint32_t id; /* node id from cnxman */ | |
9457 | + unsigned long flags; | |
9458 | + int incarnation; /* node incarnation number */ | |
9459 | +}; | |
9460 | + | |
9461 | +#endif /* __SM_INTERNAL_DOT_H__ */ | |
9462 | diff -urN linux-orig/cluster/cman/sm_joinleave.c linux-patched/cluster/cman/sm_joinleave.c | |
9463 | --- linux-orig/cluster/cman/sm_joinleave.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 9464 | +++ linux-patched/cluster/cman/sm_joinleave.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 9465 | @@ -0,0 +1,1286 @@ |
9466 | +/****************************************************************************** | |
9467 | +******************************************************************************* | |
9468 | +** | |
9469 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
9470 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
9471 | +** | |
9472 | +** This copyrighted material is made available to anyone wishing to use, | |
9473 | +** modify, copy, or redistribute it subject to the terms and conditions | |
9474 | +** of the GNU General Public License v.2. | |
9475 | +** | |
9476 | +******************************************************************************* | |
9477 | +******************************************************************************/ | |
9478 | + | |
9479 | +#include "sm.h" | |
9480 | + | |
9481 | +/* | |
9482 | + * Routines used by nodes that are joining or leaving a SG. These "sevent" | |
9483 | + * routines initiate membership changes to a SG. Existing SG members respond | |
9484 | + * using the "uevent" membership update routines. | |
9485 | + */ | |
9486 | + | |
9487 | +extern uint32_t sm_our_nodeid; | |
9488 | +extern struct list_head sm_members; | |
9489 | +static struct list_head new_event; | |
9490 | +static spinlock_t new_event_lock; | |
9491 | +static struct list_head joinleave_events; | |
9492 | + | |
9493 | +void init_joinleave(void) | |
9494 | +{ | |
9495 | + INIT_LIST_HEAD(&new_event); | |
9496 | + spin_lock_init(&new_event_lock); | |
9497 | + INIT_LIST_HEAD(&joinleave_events); | |
9498 | +} | |
9499 | + | |
9500 | +void new_joinleave(sm_sevent_t *sev) | |
9501 | +{ | |
9502 | + spin_lock(&new_event_lock); | |
9503 | + list_add_tail(&sev->se_list, &new_event); | |
9504 | + spin_unlock(&new_event_lock); | |
9505 | + wake_serviced(DO_JOINLEAVE); | |
9506 | +} | |
9507 | + | |
9508 | +sm_sevent_t *find_sevent(unsigned int id) | |
9509 | +{ | |
9510 | + sm_sevent_t *sev; | |
9511 | + | |
9512 | + list_for_each_entry(sev, &joinleave_events, se_list) { | |
9513 | + if (sev->se_id == id) | |
9514 | + return sev; | |
9515 | + } | |
9516 | + return NULL; | |
9517 | +} | |
9518 | + | |
9519 | +static void release_sevent(sm_sevent_t *sev) | |
9520 | +{ | |
9521 | + if (sev->se_len_ids) { | |
9522 | + kfree(sev->se_node_ids); | |
9523 | + sev->se_node_ids = NULL; | |
9524 | + } | |
9525 | + | |
9526 | + if (sev->se_len_status) { | |
9527 | + kfree(sev->se_node_status); | |
9528 | + sev->se_node_status = NULL; | |
9529 | + } | |
9530 | + | |
9531 | + sev->se_node_count = 0; | |
9532 | + sev->se_memb_count = 0; | |
9533 | + sev->se_reply_count = 0; | |
9534 | +} | |
9535 | + | |
9536 | +static int init_sevent(sm_sevent_t *sev) | |
9537 | +{ | |
9538 | + sm_node_t *node; | |
9539 | + int len1, len2, count, cluster_members = 0; | |
9540 | + | |
9541 | + /* clear state from any previous attempt */ | |
9542 | + release_sevent(sev); | |
9543 | + | |
9544 | + list_for_each_entry(node, &sm_members, list) { | |
9545 | + if (test_bit(SNFL_CLUSTER_MEMBER, &node->flags)) | |
9546 | + cluster_members++; | |
9547 | + } | |
9548 | + | |
9549 | + sev->se_node_count = cluster_members; | |
9550 | + sev->se_memb_count = sev->se_sg->memb_count; | |
9551 | + | |
9552 | + /* | |
9553 | + * When joining, we need a node array the size of the entire cluster | |
9554 | + * member list because we get responses from all nodes. When leaving, | |
9555 | + * we only get responses from SG members, so the node array need only | |
9556 | + * be that large. | |
9557 | + */ | |
9558 | + | |
9559 | + if (sev->se_state < SEST_LEAVE_BEGIN) | |
9560 | + count = sev->se_node_count; | |
9561 | + else | |
9562 | + count = sev->se_memb_count; | |
9563 | + | |
9564 | + len1 = count * sizeof(uint32_t); | |
9565 | + sev->se_len_ids = len1; | |
9566 | + | |
9567 | + sev->se_node_ids = (uint32_t *) kmalloc(len1, GFP_KERNEL); | |
9568 | + if (!sev->se_node_ids) | |
9569 | + goto fail; | |
9570 | + | |
9571 | + len2 = count * sizeof (char); | |
9572 | + sev->se_len_status = len2; | |
9573 | + | |
9574 | + sev->se_node_status = (char *) kmalloc(len2, GFP_KERNEL); | |
9575 | + if (!sev->se_node_status) | |
9576 | + goto fail_free; | |
9577 | + | |
9578 | + memset(sev->se_node_status, 0, len2); | |
9579 | + memset(sev->se_node_ids, 0, len1); | |
9580 | + | |
9581 | + return 0; | |
9582 | + | |
9583 | + fail_free: | |
9584 | + kfree(sev->se_node_ids); | |
9585 | + sev->se_node_ids = NULL; | |
9586 | + sev->se_len_ids = 0; | |
9587 | + | |
9588 | + fail: | |
9589 | + return -ENOMEM; | |
9590 | +} | |
9591 | + | |
9592 | +/* Context: timer */ | |
9593 | + | |
9594 | +static void sev_restart(unsigned long data) | |
9595 | +{ | |
9596 | + sm_sevent_t *sev = (sm_sevent_t *) data; | |
9597 | + | |
9598 | + clear_bit(SEFL_DELAY, &sev->se_flags); | |
9599 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
9600 | + wake_serviced(DO_JOINLEAVE); | |
9601 | +} | |
9602 | + | |
9603 | +static void schedule_sev_restart(sm_sevent_t *sev) | |
9604 | +{ | |
9605 | + init_timer(&sev->se_restart_timer); | |
9606 | + sev->se_restart_timer.function = sev_restart; | |
9607 | + sev->se_restart_timer.data = (long) sev; | |
9608 | + mod_timer(&sev->se_restart_timer, jiffies + (RETRY_DELAY * HZ)); | |
9609 | +} | |
9610 | + | |
9611 | +void free_sg_memb(sm_group_t *sg) | |
9612 | +{ | |
9613 | + sm_node_t *node; | |
9614 | + | |
9615 | + while (!list_empty(&sg->memb)) { | |
9616 | + node = list_entry(sg->memb.next, sm_node_t, list); | |
9617 | + list_del(&node->list); | |
9618 | + kfree(node); | |
9619 | + } | |
9620 | + sg->memb_count = 0; | |
9621 | +} | |
9622 | + | |
9623 | +/* | |
9624 | + * 1. First step in joining a SG - send a message to all nodes in the cluster | |
9625 | + * asking to join the named SG. If any nodes are members they will reply with | |
9626 | + * a POS, or a WAIT (wait means try again, only one node can join at a time). | |
9627 | + * If no one knows about this SG, they all send NEG replies which means we form | |
9628 | + * the SG with just ourself as a member. | |
9629 | + */ | |
9630 | + | |
9631 | +static int send_join_notice(sm_sevent_t *sev) | |
9632 | +{ | |
9633 | + sm_group_t *sg = sev->se_sg; | |
9634 | + sm_node_t *node; | |
9635 | + char *msg; | |
9636 | + int i = 0, error, namelen, len = 0; | |
9637 | + | |
9638 | + /* | |
9639 | + * Create node array from member list in which to collect responses. | |
9640 | + */ | |
9641 | + | |
9642 | + error = init_sevent(sev); | |
9643 | + if (error) | |
9644 | + goto out; | |
9645 | + | |
9646 | + list_for_each_entry(node, &sm_members, list) { | |
9647 | + if (test_bit(SNFL_CLUSTER_MEMBER, &node->flags)) | |
9648 | + sev->se_node_ids[i++] = node->id; | |
9649 | + } | |
9650 | + | |
9651 | + /* | |
9652 | + * Create and send a join request message. | |
9653 | + * | |
9654 | + * Other nodes then run process_join_request and reply to us; we | |
9655 | + * collect the responses in process_reply and check them in | |
9656 | + * check_join_notice. | |
9657 | + */ | |
9658 | + | |
9659 | + namelen = sg->namelen; | |
9660 | + msg = create_smsg(sg, SMSG_JOIN_REQ, namelen, &len, sev); | |
9661 | + memcpy(msg + sizeof(sm_msg_t), sg->name, namelen); | |
9662 | + | |
9663 | + error = send_broadcast_message_sev(msg, len, sev); | |
9664 | + | |
9665 | + out: | |
9666 | + return error; | |
9667 | +} | |
9668 | + | |
9669 | +/* | |
9670 | + * 2. Second step in joining a SG - after we collect all replies to our join | |
9671 | + * request, we look at them. If anyone told us to wait, we'll wait a while, go | |
9672 | + * back and start at step 1 again. | |
9673 | + */ | |
9674 | + | |
9675 | +static int check_join_notice(sm_sevent_t *sev) | |
9676 | +{ | |
9677 | + int pos = 0, wait = 0, neg = 0, restart = 0, i, error = 0; | |
9678 | + | |
9679 | + for (i = 0; i < sev->se_node_count; i++) { | |
9680 | + switch (sev->se_node_status[i]) { | |
9681 | + case STATUS_POS: | |
9682 | + /* this node is in the SG and will be in new proposed | |
9683 | + * memb list */ | |
9684 | + pos++; | |
9685 | + break; | |
9686 | + | |
9687 | + case STATUS_WAIT: | |
9688 | + /* this node is in the SG but something else is | |
9689 | + * happening with it at the moment. */ | |
9690 | + wait++; | |
9691 | + break; | |
9692 | + | |
9693 | + case STATUS_NEG: | |
9694 | + /* this node has no record of the SG we're interested | |
9695 | + * in */ | |
9696 | + neg++; | |
9697 | + | |
9698 | + if (sev->se_node_ids[i] == sm_our_nodeid) | |
9699 | + sev->se_node_status[i] = STATUS_POS; | |
9700 | + break; | |
9701 | + | |
9702 | + default: | |
9703 | + /* we didn't get a valid response from this node, | |
9704 | + * restart the entire sev. */ | |
9705 | + restart++; | |
9706 | + break; | |
9707 | + } | |
9708 | + } | |
9709 | + | |
9710 | + if (pos && !wait && !restart) { | |
9711 | + /* all current members of this sg pos'ed our entry */ | |
9712 | + } else if (!pos && !wait && !restart && neg) { | |
9713 | + /* we're the first in the cluster to join this sg */ | |
9714 | + sev->se_sg->global_id = sm_new_global_id(sev->se_sg->level); | |
9715 | + } else | |
9716 | + error = -1; | |
9717 | + | |
9718 | + return error; | |
9719 | +} | |
9720 | + | |
9721 | +/* | |
9722 | + * 3. Third step in joining the SG - tell the nodes that are already members | |
9723 | + * to "stop" the service. We stop them so that everyone can restart with the | |
9724 | + * new member (us!) added. | |
9725 | + */ | |
9726 | + | |
9727 | +static int send_join_stop(sm_sevent_t *sev) | |
9728 | +{ | |
9729 | + sm_group_t *sg = sev->se_sg; | |
9730 | + sm_node_t *node; | |
9731 | + char *msg; | |
9732 | + uint32_t be_count; | |
9733 | + int i, len = 0, error = 0; | |
9734 | + | |
9735 | + /* | |
9736 | + * Form the SG memb list with us in it. | |
9737 | + */ | |
9738 | + | |
9739 | + for (i = 0; i < sev->se_node_count; i++) { | |
9740 | + if (sev->se_node_status[i] != STATUS_POS) | |
9741 | + continue; | |
9742 | + | |
9743 | + node = sm_new_node(sev->se_node_ids[i]); | |
9744 | + if (!node) | |
9745 | + goto fail; | |
9746 | + | |
9747 | + list_add_tail(&node->list, &sg->memb); | |
9748 | + sg->memb_count++; | |
9749 | + } | |
9750 | + | |
9751 | + /* | |
9752 | + * Re-init the node vector in which to collect responses again. | |
9753 | + */ | |
9754 | + | |
9755 | + sev->se_memb_count = sg->memb_count; | |
9756 | + | |
9757 | + memset(sev->se_node_status, 0, sev->se_len_status); | |
9758 | + memset(sev->se_node_ids, 0, sev->se_len_ids); | |
9759 | + i = 0; | |
9760 | + | |
9761 | + list_for_each_entry(node, &sg->memb, list) | |
9762 | + sev->se_node_ids[i++] = node->id; | |
9763 | + | |
9764 | + /* | |
9765 | + * Create and send a stop message. | |
9766 | + * | |
9767 | + * Other nodes then run process_stop_request and process_join_stop and | |
9768 | + * reply to us. They stop the sg we're trying to join if they agree. | |
9769 | + * We collect responses in process_reply and check them in | |
9770 | + * check_join_stop. | |
9771 | + */ | |
9772 | + | |
9773 | + msg = create_smsg(sg, SMSG_JSTOP_REQ, sizeof(uint32_t), &len, sev); | |
9774 | + be_count = cpu_to_be32(sg->memb_count); | |
9775 | + memcpy(msg + sizeof(sm_msg_t), &be_count, sizeof(uint32_t)); | |
9776 | + | |
9777 | + error = send_members_message_sev(sg, msg, len, sev); | |
9778 | + if (error < 0) | |
9779 | + goto fail; | |
9780 | + | |
9781 | + return 0; | |
9782 | + | |
9783 | + fail: | |
9784 | + free_sg_memb(sg); | |
9785 | + return error; | |
9786 | +} | |
9787 | + | |
9788 | +/* | |
9789 | + * 4. Fourth step in joining the SG - after we collect replies to our stop | |
9790 | + * request, we look at them. Everyone sending POS agrees with us joining and | |
9791 | + * has stopped their SG. If some nodes sent NEG, something is wrong and we | |
9792 | + * don't have a good way to address that yet since some nodes may have sent | |
9793 | + * POS. | |
9794 | + * | |
9795 | + * FIXME: even nodes replying with NEG should stop their SG so we can send an | |
9796 | + * abort and have everyone at the same place to start from again. | |
9797 | + */ | |
9798 | + | |
9799 | +static int check_join_stop(sm_sevent_t *sev) | |
9800 | +{ | |
9801 | + sm_group_t *sg = sev->se_sg; | |
9802 | + int i, pos = 0, neg = 0; | |
9803 | + | |
9804 | + for (i = 0; i < sev->se_memb_count; i++) { | |
9805 | + switch (sev->se_node_status[i]) { | |
9806 | + case STATUS_POS: | |
9807 | + pos++; | |
9808 | + break; | |
9809 | + | |
9810 | + case STATUS_NEG: | |
9811 | + log_error(sg, "check_join_stop: neg from nodeid %u " | |
9812 | + "(%d, %d, %u)", sev->se_node_ids[i], | |
9813 | + pos, neg, sev->se_memb_count); | |
9814 | + neg++; | |
9815 | + break; | |
9816 | + | |
9817 | + default: | |
9818 | + log_error(sg, "check_join_stop: unknown status=%u " | |
9819 | + "nodeid=%u", sev->se_node_status[i], | |
9820 | + sev->se_node_ids[i]); | |
9821 | + neg++; | |
9822 | + break; | |
9823 | + } | |
9824 | + } | |
9825 | + | |
9826 | + if (pos == sg->memb_count) | |
9827 | + return 0; | |
9828 | + | |
9829 | + free_sg_memb(sg); | |
9830 | + return -1; | |
9831 | +} | |
9832 | + | |
9833 | +/* | |
9834 | + * 5. Fifth step in joining the SG - everyone has stopped their service and we | |
9835 | + * all now start the service with us, the new member, added to the SG member | |
9836 | + * list. We send start to our own service here and send a message to the other | |
9837 | + * members that they should also start their service. | |
9838 | + */ | |
9839 | + | |
9840 | +static int send_join_start(sm_sevent_t *sev) | |
9841 | +{ | |
9842 | + sm_group_t *sg = sev->se_sg; | |
9843 | + sm_node_t *node; | |
9844 | + uint32_t *memb; | |
9845 | + char *msg; | |
9846 | + int error, count = 0, len = 0; | |
9847 | + | |
9848 | + /* | |
9849 | + * Create a start message and send it. | |
9850 | + */ | |
9851 | + | |
9852 | + msg = create_smsg(sg, SMSG_JSTART_CMD, 0, &len, sev); | |
9853 | + | |
9854 | + error = send_members_message(sg, msg, len); | |
9855 | + if (error < 0) | |
9856 | + goto fail; | |
9857 | + | |
9858 | + /* | |
9859 | + * Start the service ourself. The chunk of memory with the member ids | |
9860 | + * must be freed by the service when it is done with it. | |
9861 | + */ | |
9862 | + | |
9863 | + SM_RETRY(memb = kmalloc(sg->memb_count * sizeof(uint32_t), GFP_KERNEL), | |
9864 | + memb); | |
9865 | + | |
9866 | + list_for_each_entry(node, &sg->memb, list) | |
9867 | + memb[count++] = node->id; | |
9868 | + | |
9869 | + set_bit(SEFL_ALLOW_STARTDONE, &sev->se_flags); | |
9870 | + | |
9871 | + sg->ops->start(sg->service_data, memb, count, sev->se_id, | |
9872 | + SERVICE_NODE_JOIN); | |
9873 | + return 0; | |
9874 | + | |
9875 | + fail: | |
9876 | + free_sg_memb(sg); | |
9877 | + return error; | |
9878 | +} | |
9879 | + | |
9880 | +/* | |
9881 | + * 6. Sixth step in joining the SG - once the service has completed its start, | |
9882 | + * it does a kcl_start_done() to signal us that it's done. That gets us here | |
9883 | + * and we do a barrier with all other members which join the barrier when their | |
9884 | + * service is done starting. | |
9885 | + */ | |
9886 | + | |
9887 | +static int startdone_barrier_new(sm_sevent_t *sev) | |
9888 | +{ | |
9889 | + sm_group_t *sg = sev->se_sg; | |
9890 | + char bname[MAX_BARRIER_NAME_LEN]; | |
9891 | + int error; | |
9892 | + | |
9893 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
9894 | + sev->se_barrier_status = -1; | |
9895 | + | |
9896 | + set_bit(SEFL_ALLOW_BARRIER, &sev->se_flags); | |
9897 | + | |
9898 | + /* If we're the only member, skip the barrier */ | |
9899 | + if (sg->memb_count == 1) { | |
9900 | + process_startdone_barrier_new(sg, 0); | |
9901 | + return 0; | |
9902 | + } | |
9903 | + | |
9904 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
9905 | + sg->global_id, sm_our_nodeid, sev->se_id, sg->memb_count); | |
9906 | + | |
9907 | + error = sm_barrier(bname, sg->memb_count, SM_BARRIER_STARTDONE_NEW); | |
9908 | + if (error) | |
9909 | + goto fail; | |
9910 | + | |
9911 | + return 0; | |
9912 | + | |
9913 | + fail: | |
9914 | + clear_bit(SEFL_ALLOW_BARRIER, &sev->se_flags); | |
9915 | + sg->ops->stop(sg->service_data); | |
9916 | + free_sg_memb(sg); | |
9917 | + return error; | |
9918 | +} | |
9919 | + | |
9920 | +/* | |
9921 | + * 7. Seventh step in joining the SG - check that the barrier we joined with | |
9922 | + * all other members returned with a successful status. | |
9923 | + */ | |
9924 | + | |
9925 | +static int check_startdone_barrier_new(sm_sevent_t *sev) | |
9926 | +{ | |
9927 | + sm_group_t *sg = sev->se_sg; | |
9928 | + int error = sev->se_barrier_status; | |
9929 | + | |
9930 | + if (error) { | |
9931 | + sg->ops->stop(sg->service_data); | |
9932 | + free_sg_memb(sg); | |
9933 | + } | |
9934 | + return error; | |
9935 | +} | |
9936 | + | |
9937 | +/* | |
9938 | + * 8. Eigth step in joining the SG - send the service a "finish" indicating | |
9939 | + * that all members have successfully started the service. | |
9940 | + */ | |
9941 | + | |
9942 | +static void do_finish_new(sm_sevent_t *sev) | |
9943 | +{ | |
9944 | + sm_group_t *sg = sev->se_sg; | |
9945 | + | |
9946 | + sg->state = SGST_RUN; | |
9947 | + sg->sevent = NULL; | |
9948 | + clear_bit(SGFL_SEVENT, &sg->flags); | |
9949 | + | |
9950 | + sg->ops->finish(sg->service_data, sev->se_id); | |
9951 | +} | |
9952 | + | |
9953 | +/* | |
9954 | + * 9. Ninth step in joining the SG - it's done so get rid of the sevent stuff | |
9955 | + * and tell the process which initiated the join that it's done. | |
9956 | + */ | |
9957 | + | |
9958 | +static void sevent_done(sm_sevent_t *sev) | |
9959 | +{ | |
9960 | + sm_group_t *sg = sev->se_sg; | |
9961 | + | |
9962 | + list_del(&sev->se_list); | |
9963 | + release_sevent(sev); | |
9964 | + kfree(sev); | |
9965 | + complete(&sg->event_comp); | |
9966 | +} | |
9967 | + | |
9968 | +/* | |
9969 | + * Move through the steps of a join. Summary: | |
9970 | + * | |
9971 | + * 1. Send a join notice to all cluster members. | |
9972 | + * 2. Collect and check replies to the join notice. | |
9973 | + * 3. Send a stop message to all SG members. | |
9974 | + * 4. Collect and check replies to the stop message. | |
9975 | + * 5. Send a start message to all SG members and start service ourself. | |
9976 | + * 6. Use barrier to wait for all nodes to complete the start. | |
9977 | + * 7. Check that all SG members joined the barrier. | |
9978 | + * 8. Send finish to the service indicating that all nodes started it. | |
9979 | + * 9. Clean up sevent and signal completion to the process that started the join | |
9980 | + */ | |
9981 | + | |
9982 | +static void process_join_sevent(sm_sevent_t *sev) | |
9983 | +{ | |
9984 | + int error = 0; | |
9985 | + | |
9986 | + /* | |
9987 | + * We may cancel the current join attempt if another node is also | |
9988 | + * attempting to join or leave. (Only a single node can join or leave | |
9989 | + * at once.) If cancelled, 0ur join attempt will be restarted later. | |
9990 | + */ | |
9991 | + | |
9992 | + if (test_and_clear_bit(SEFL_CANCEL, &sev->se_flags)) { | |
9993 | + error = -1; | |
9994 | + goto cancel; | |
9995 | + } | |
9996 | + | |
9997 | + log_debug(sev->se_sg, "sevent state %u", sev->se_state); | |
9998 | + | |
9999 | + switch (sev->se_state) { | |
10000 | + | |
10001 | + /* | |
10002 | + * An sevent is created in kcl_join_service with a state of | |
10003 | + * JOIN_BEGIN. | |
10004 | + */ | |
10005 | + | |
10006 | + case SEST_JOIN_BEGIN: | |
10007 | + sev->se_state = SEST_JOIN_ACKWAIT; | |
10008 | + error = send_join_notice(sev); | |
10009 | + break; | |
10010 | + | |
10011 | + /* | |
10012 | + * se_state is changed from JOIN_ACKWAIT to JOIN_ACKED in | |
10013 | + * process_reply (when all the replies have been received) | |
10014 | + */ | |
10015 | + | |
10016 | + case SEST_JOIN_ACKED: | |
10017 | + error = check_join_notice(sev); | |
10018 | + if (error) | |
10019 | + break; | |
10020 | + | |
10021 | + sev->se_state = SEST_JSTOP_ACKWAIT; | |
10022 | + error = send_join_stop(sev); | |
10023 | + break; | |
10024 | + | |
10025 | + /* | |
10026 | + * se_state is changed from JSTOP_ACKWAIT to JSTOP_ACKED in | |
10027 | + * proces_reply (when all the replies have been received) | |
10028 | + */ | |
10029 | + | |
10030 | + case SEST_JSTOP_ACKED: | |
10031 | + error = check_join_stop(sev); | |
10032 | + if (error) | |
10033 | + break; | |
10034 | + | |
10035 | + sev->se_state = SEST_JSTART_SERVICEWAIT; | |
10036 | + error = send_join_start(sev); | |
10037 | + break; | |
10038 | + | |
10039 | + /* | |
10040 | + * se_state is changed from JSTART_SERVICEWAIT to | |
10041 | + * JSTART_SERVICEDONE in kcl_start_done | |
10042 | + */ | |
10043 | + | |
10044 | + case SEST_JSTART_SERVICEDONE: | |
10045 | + sev->se_state = SEST_BARRIER_WAIT; | |
10046 | + error = startdone_barrier_new(sev); | |
10047 | + break; | |
10048 | + | |
10049 | + /* | |
10050 | + * se_state is changed from BARRIER_WAIT to BARRIER_DONE in | |
10051 | + * process_startdone_barrier_new | |
10052 | + */ | |
10053 | + | |
10054 | + case SEST_BARRIER_DONE: | |
10055 | + error = check_startdone_barrier_new(sev); | |
10056 | + if (error) | |
10057 | + break; | |
10058 | + | |
10059 | + do_finish_new(sev); | |
10060 | + sevent_done(sev); | |
10061 | + break; | |
10062 | + | |
10063 | + default: | |
10064 | + log_error(sev->se_sg, "no join processing for state %u", | |
10065 | + sev->se_state); | |
10066 | + } | |
10067 | + | |
10068 | + cancel: | |
10069 | + if (error) { | |
10070 | + /* restart the sevent from the beginning */ | |
10071 | + sev->se_state = SEST_JOIN_BEGIN; | |
10072 | + sev->se_sg->global_id = 0; | |
10073 | + set_bit(SEFL_DELAY, &sev->se_flags); | |
10074 | + schedule_sev_restart(sev); | |
10075 | + } | |
10076 | +} | |
10077 | + | |
10078 | +/* | |
10079 | + * 1. First step in leaving an SG - send a message to other SG members asking | |
10080 | + * to leave the SG. Nodes that don't have another active sevent or uevent for | |
10081 | + * this SG will return POS. | |
10082 | + */ | |
10083 | + | |
10084 | +static int send_leave_notice(sm_sevent_t *sev) | |
10085 | +{ | |
10086 | + sm_group_t *sg = sev->se_sg; | |
10087 | + sm_node_t *node; | |
10088 | + char *msg; | |
10089 | + int i = 0, error = -1, len = 0; | |
10090 | + | |
10091 | + /* | |
10092 | + * Create a node array from member list in which to collect responses. | |
10093 | + */ | |
10094 | + | |
10095 | + error = init_sevent(sev); | |
10096 | + if (error) | |
10097 | + goto out; | |
10098 | + | |
10099 | + list_for_each_entry(node, &sg->memb, list) | |
10100 | + sev->se_node_ids[i++] = node->id; | |
10101 | + | |
10102 | + /* | |
10103 | + * Create and send a leave request message. | |
10104 | + */ | |
10105 | + | |
10106 | + msg = create_smsg(sg, SMSG_LEAVE_REQ, 0, &len, sev); | |
10107 | + | |
10108 | + error = send_members_message_sev(sg, msg, len, sev); | |
10109 | + | |
10110 | + out: | |
10111 | + return error; | |
10112 | +} | |
10113 | + | |
10114 | +/* | |
10115 | + * 2. Second step in leaving an SG - after we collect all replies to our leave | |
10116 | + * request, we look at them. If anyone replied with WAIT, we abort our attempt | |
10117 | + * at leaving and try again in a bit. | |
10118 | + */ | |
10119 | + | |
10120 | +static int check_leave_notice(sm_sevent_t *sev) | |
10121 | +{ | |
10122 | + int pos = 0, wait = 0, neg = 0, restart = 0, i; | |
10123 | + | |
10124 | + for (i = 0; i < sev->se_memb_count; i++) { | |
10125 | + switch (sev->se_node_status[i]) { | |
10126 | + case STATUS_POS: | |
10127 | + pos++; | |
10128 | + break; | |
10129 | + | |
10130 | + case STATUS_WAIT: | |
10131 | + wait++; | |
10132 | + break; | |
10133 | + | |
10134 | + case STATUS_NEG: | |
10135 | + neg++; | |
10136 | + break; | |
10137 | + | |
10138 | + default: | |
10139 | + /* we didn't get a valid response from this node, | |
10140 | + * restart the entire sev. */ | |
10141 | + restart++; | |
10142 | + break; | |
10143 | + } | |
10144 | + } | |
10145 | + | |
10146 | + /* all members approve */ | |
10147 | + if (pos && !wait && !restart) | |
10148 | + return 0; | |
10149 | + | |
10150 | + return -1; | |
10151 | +} | |
10152 | + | |
10153 | +/* | |
10154 | + * 3. Third step in leaving the SG - tell the member nodes to "stop" the SG. | |
10155 | + * They must be stopped in order to restart without us as a member. | |
10156 | + */ | |
10157 | + | |
10158 | +static int send_leave_stop(sm_sevent_t *sev) | |
10159 | +{ | |
10160 | + sm_group_t *sg = sev->se_sg; | |
10161 | + char *msg; | |
10162 | + int error, len = 0; | |
10163 | + | |
10164 | + /* | |
10165 | + * Re-init the status vector in which to collect responses. | |
10166 | + */ | |
10167 | + | |
10168 | + memset(sev->se_node_status, 0, sev->se_len_status); | |
10169 | + | |
10170 | + /* | |
10171 | + * Create and send a stop message. | |
10172 | + */ | |
10173 | + | |
10174 | + msg = create_smsg(sg, SMSG_LSTOP_REQ, 0, &len, sev); | |
10175 | + | |
10176 | + error = send_members_message_sev(sg, msg, len, sev); | |
10177 | + if (error < 0) | |
10178 | + goto out; | |
10179 | + | |
10180 | + /* | |
10181 | + * we and all others stop the SG now | |
10182 | + */ | |
10183 | + | |
10184 | + sg->ops->stop(sg->service_data); | |
10185 | + | |
10186 | + out: | |
10187 | + return error; | |
10188 | +} | |
10189 | + | |
10190 | +/* | |
10191 | + * 4. Fourth step in leaving the SG - check the replies to our stop request. | |
10192 | + * Same problem with getting different replies as check_join_stop. | |
10193 | + */ | |
10194 | + | |
10195 | +static int check_leave_stop(sm_sevent_t *sev) | |
10196 | +{ | |
10197 | + sm_group_t *sg = sev->se_sg; | |
10198 | + int i, pos = 0, neg = 0; | |
10199 | + | |
10200 | + for (i = 0; i < sev->se_memb_count; i++) { | |
10201 | + switch (sev->se_node_status[i]) { | |
10202 | + case STATUS_POS: | |
10203 | + pos++; | |
10204 | + break; | |
10205 | + | |
10206 | + case STATUS_NEG: | |
10207 | + log_error(sg, "check_leave_stop: fail from nodeid %u " | |
10208 | + "(%d, %d, %u)", sev->se_node_ids[i], | |
10209 | + pos, neg, sev->se_memb_count); | |
10210 | + neg++; | |
10211 | + break; | |
10212 | + | |
10213 | + default: | |
10214 | + log_error(sg, "check_leave_stop: status %u nodeid %u", | |
10215 | + sev->se_node_status[i], sev->se_node_ids[i]); | |
10216 | + neg++; | |
10217 | + break; | |
10218 | + } | |
10219 | + } | |
10220 | + | |
10221 | + if (pos == sg->memb_count) | |
10222 | + return 0; | |
10223 | + | |
10224 | + return -1; | |
10225 | +} | |
10226 | + | |
10227 | +/* | |
10228 | + * 5. Fifth step in leaving the SG - tell the other SG members to restart the | |
10229 | + * service without us. We, of course, don't start our own stopped service. If | |
10230 | + * we're the last SG member and leaving, we jump right to the next step. | |
10231 | + */ | |
10232 | + | |
10233 | +static int send_leave_start(sm_sevent_t *sev) | |
10234 | +{ | |
10235 | + sm_group_t *sg = sev->se_sg; | |
10236 | + char *msg; | |
10237 | + int error = 0, len = 0; | |
10238 | + | |
10239 | + if (sg->memb_count == 1) { | |
10240 | + sev->se_state = SEST_LSTART_REMOTEDONE; | |
10241 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
10242 | + wake_serviced(DO_JOINLEAVE); | |
10243 | + } else { | |
10244 | + msg = create_smsg(sg, SMSG_LSTART_CMD, 0, &len, sev); | |
10245 | + error = send_members_message(sg, msg, len); | |
10246 | + } | |
10247 | + return error; | |
10248 | +} | |
10249 | + | |
10250 | +/* | |
10251 | + * Move through the steps of a leave. Summary: | |
10252 | + * | |
10253 | + * 1. Send a leave notice to all SG members. | |
10254 | + * 2. Collect and check replies to the leave notice. | |
10255 | + * 3. Send a stop message to all SG members and stop our own SG. | |
10256 | + * 4. Collect and check replies to the stop message. | |
10257 | + * 5. Send a start message to SG members. | |
10258 | + * 6. Clean up sevent and signal completion to the process that | |
10259 | + * started the leave. | |
10260 | + */ | |
10261 | + | |
10262 | +static void process_leave_sevent(sm_sevent_t *sev) | |
10263 | +{ | |
10264 | + int error = 0; | |
10265 | + | |
10266 | + /* | |
10267 | + * We may cancel the current leave attempt if another node is also | |
10268 | + * attempting to join or leave. (Only a single node can join or leave | |
10269 | + * at once.) Our leave attempt will be restarted after being | |
10270 | + * cancelled. | |
10271 | + */ | |
10272 | + | |
10273 | + if (test_and_clear_bit(SEFL_CANCEL, &sev->se_flags)) { | |
10274 | + error = 1; | |
10275 | + goto cancel; | |
10276 | + } | |
10277 | + | |
10278 | + if (test_bit(SGFL_UEVENT, &sev->se_sg->flags)) { | |
10279 | + error = 2; | |
10280 | + goto cancel; | |
10281 | + } | |
10282 | + | |
10283 | + if (!list_empty(&sev->se_sg->joining)) { | |
10284 | + error = 3; | |
10285 | + goto cancel; | |
10286 | + } | |
10287 | + | |
10288 | + log_debug(sev->se_sg, "sevent state %u", sev->se_state); | |
10289 | + | |
10290 | + switch (sev->se_state) { | |
10291 | + | |
10292 | + /* | |
10293 | + * An sevent is created in kcl_leave_service with a state of | |
10294 | + * LEAVE_BEGIN. | |
10295 | + */ | |
10296 | + | |
10297 | + case SEST_LEAVE_BEGIN: | |
10298 | + sev->se_state = SEST_LEAVE_ACKWAIT; | |
10299 | + error = send_leave_notice(sev); | |
10300 | + break; | |
10301 | + | |
10302 | + /* | |
10303 | + * se_state is changed from LEAVE_ACKWAIT to LEAVE_ACKED in | |
10304 | + * process_reply (when all the replies have been received) | |
10305 | + */ | |
10306 | + | |
10307 | + case SEST_LEAVE_ACKED: | |
10308 | + error = check_leave_notice(sev); | |
10309 | + if (error) | |
10310 | + break; | |
10311 | + | |
10312 | + sev->se_state = SEST_LSTOP_ACKWAIT; | |
10313 | + error = send_leave_stop(sev); | |
10314 | + break; | |
10315 | + | |
10316 | + /* | |
10317 | + * se_state is changed from LSTOP_ACKWAIT to LSTOP_ACKED in | |
10318 | + * process_reply | |
10319 | + */ | |
10320 | + | |
10321 | + case SEST_LSTOP_ACKED: | |
10322 | + error = check_leave_stop(sev); | |
10323 | + if (error) | |
10324 | + break; | |
10325 | + | |
10326 | + sev->se_state = SEST_LSTART_WAITREMOTE; | |
10327 | + error = send_leave_start(sev); | |
10328 | + break; | |
10329 | + | |
10330 | + /* | |
10331 | + * se_state is changed from LSTART_WAITREMOTE to | |
10332 | + * LSTART_REMOTEDONE in process_leave_done | |
10333 | + */ | |
10334 | + | |
10335 | + case SEST_LSTART_REMOTEDONE: | |
10336 | + sevent_done(sev); | |
10337 | + break; | |
10338 | + | |
10339 | + default: | |
10340 | + log_error(sev->se_sg, "process_leave_sevent state=%u\n", | |
10341 | + sev->se_state); | |
10342 | + } | |
10343 | + | |
10344 | + cancel: | |
10345 | + if (error) { | |
10346 | + /* restart the sevent from the beginning */ | |
10347 | + sev->se_state = SEST_LEAVE_BEGIN; | |
10348 | + set_bit(SEFL_DELAY, &sev->se_flags); | |
10349 | + schedule_sev_restart(sev); | |
10350 | + } | |
10351 | +} | |
10352 | + | |
10353 | +/* | |
10354 | + * Sevent backout code. Take appropriate steps when a recovery occurs while | |
10355 | + * we're in the midst of an sevent. The recovery may or may not affect the | |
10356 | + * sevent. If it does, it usually means cancelling the sevent and restarting | |
10357 | + * it from the beginning once the recovery processing is done. | |
10358 | + */ | |
10359 | + | |
10360 | +/* | |
10361 | + * If any of the nodes that replied with OK is dead, we give up on the current | |
10362 | + * join attempt and restart. Otherwise, this sevent can continue. | |
10363 | + */ | |
10364 | + | |
10365 | +static int backout_join_acked(sm_sevent_t *sev) | |
10366 | +{ | |
10367 | + sm_node_t *node; | |
10368 | + int i; | |
10369 | + | |
10370 | + for (i = 0; i < sev->se_node_count; i++) { | |
10371 | + if (sev->se_node_status[i] != STATUS_POS) | |
10372 | + continue; | |
10373 | + | |
10374 | + list_for_each_entry(node, &sm_members, list) { | |
10375 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags) && | |
10376 | + (node->id == sev->se_node_ids[i])) | |
10377 | + return TRUE; | |
10378 | + } | |
10379 | + } | |
10380 | + return FALSE; | |
10381 | +} | |
10382 | + | |
10383 | +/* | |
10384 | + * In this state our sg member list exists and mark_affected_sgs() will have | |
10385 | + * set NEED_RECOVERY if any of the nodes in the sg we're joining is dead. We | |
10386 | + * restart the join process if this is the case, otherwise this sevent can | |
10387 | + * continue. | |
10388 | + */ | |
10389 | + | |
10390 | +static int backout_jstop_ackwait(sm_sevent_t *sev) | |
10391 | +{ | |
10392 | + sm_group_t *sg = sev->se_sg; | |
10393 | + | |
10394 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10395 | + return FALSE; | |
10396 | + | |
10397 | + clear_bit(SEFL_ALLOW_JSTOP, &sev->se_flags); | |
10398 | + free_sg_memb(sg); | |
10399 | + return TRUE; | |
10400 | +} | |
10401 | + | |
10402 | +/* | |
10403 | + * Same as previous. | |
10404 | + */ | |
10405 | + | |
10406 | +static int backout_jstop_acked(sm_sevent_t *sev) | |
10407 | +{ | |
10408 | + return backout_jstop_ackwait(sev); | |
10409 | +} | |
10410 | + | |
10411 | +/* | |
10412 | + * If NEED_RECOVERY is set a member of the sg we're joining died while we were | |
10413 | + * starting our service. The recovery process will restart the service on all | |
10414 | + * the prior sg members (not including those that died or us). We will | |
10415 | + * reattempt our join which should be accepted once the nodes are done with | |
10416 | + * recovery. | |
10417 | + */ | |
10418 | + | |
10419 | +static int backout_jstart_servicewait(sm_sevent_t *sev) | |
10420 | +{ | |
10421 | + sm_group_t *sg = sev->se_sg; | |
10422 | + | |
10423 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10424 | + return FALSE; | |
10425 | + | |
10426 | + clear_bit(SEFL_ALLOW_STARTDONE, &sev->se_flags); | |
10427 | + sg->ops->stop(sg->service_data); | |
10428 | + free_sg_memb(sg); | |
10429 | + return TRUE; | |
10430 | +} | |
10431 | + | |
10432 | +/* | |
10433 | + * Same as previous. | |
10434 | + */ | |
10435 | + | |
10436 | +static int backout_jstart_servicedone(sm_sevent_t *sev) | |
10437 | +{ | |
10438 | + return backout_jstart_servicewait(sev); | |
10439 | +} | |
10440 | + | |
10441 | +/* | |
10442 | + * If NEED_RECOVERY is set a member of the sg we're joining died while we were | |
10443 | + * waiting on the "all done" barrier. Stop our service that we just started | |
10444 | + * and cancel the barrier. The recovery process will restart the service on | |
10445 | + * all the prior sg members (not including those that died or us). We will | |
10446 | + * reattempt our join which should be accepted once the nodes are done with | |
10447 | + * recovery. | |
10448 | + */ | |
10449 | + | |
10450 | +static int backout_barrier_wait(sm_sevent_t *sev) | |
10451 | +{ | |
10452 | + sm_group_t *sg = sev->se_sg; | |
10453 | + char bname[MAX_BARRIER_NAME_LEN]; | |
10454 | + | |
10455 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10456 | + return FALSE; | |
10457 | + | |
10458 | + clear_bit(SEFL_ALLOW_BARRIER, &sev->se_flags); | |
10459 | + | |
10460 | + sg->ops->stop(sg->service_data); | |
10461 | + | |
10462 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
10463 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
10464 | + sg->global_id, sm_our_nodeid, sev->se_id, | |
10465 | + sg->memb_count); | |
10466 | + kcl_barrier_cancel(bname); | |
10467 | + | |
10468 | + free_sg_memb(sg); | |
10469 | + return TRUE; | |
10470 | +} | |
10471 | + | |
10472 | +/* | |
10473 | + * If NEED_RECOVERY is set, a member of the sg we just joined has failed. The | |
10474 | + * recovery began after the barrier callback. If the result in the callback is | |
10475 | + * "success" then we are joined, this sevent is finished and we'll process the | |
10476 | + * sg within the forthcoming recovery with the other members. | |
10477 | + * | |
10478 | + * We rely upon cnxman to guarantee that once all nodes have joined a barrier, | |
10479 | + * all nodes will receive the corresponding barrier callback *before any* | |
10480 | + * receive an sm_member_update() due to one of those nodes failing just after | |
10481 | + * joining the barrier. If some nodes receive the sm_member_update() before | |
10482 | + * the barrier callback and others receive the barrier callback before the | |
10483 | + * sm_member_update() then they will disagree as to whether the node joining/ | |
10484 | + * leaving is in/out of the sg. | |
10485 | + */ | |
10486 | + | |
10487 | +static int backout_barrier_done(sm_sevent_t *sev) | |
10488 | +{ | |
10489 | + sm_group_t *sg = sev->se_sg; | |
10490 | + | |
10491 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10492 | + return FALSE; | |
10493 | + | |
10494 | + if (!sev->se_barrier_status) { | |
10495 | + do_finish_new(sev); | |
10496 | + sevent_done(sev); | |
10497 | + return FALSE; | |
10498 | + } else { | |
10499 | + sg->ops->stop(sg->service_data); | |
10500 | + free_sg_memb(sg); | |
10501 | + return TRUE; | |
10502 | + } | |
10503 | +} | |
10504 | + | |
10505 | +/* | |
10506 | + * We've done nothing yet, just restart when recovery is done (if sg is flagged | |
10507 | + * with recovery.) | |
10508 | + */ | |
10509 | + | |
10510 | +static int backout_leave_begin(sm_sevent_t *sev) | |
10511 | +{ | |
10512 | + sm_group_t *sg = sev->se_sg; | |
10513 | + | |
10514 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10515 | + return FALSE; | |
10516 | + | |
10517 | + return TRUE; | |
10518 | +} | |
10519 | + | |
10520 | +/* | |
10521 | + * Ignore any replies to our leave notice and restart when recovery is done (if | |
10522 | + * sg is flagged with recovery.) | |
10523 | + */ | |
10524 | + | |
10525 | +static int backout_leave_ackwait(sm_sevent_t *sev) | |
10526 | +{ | |
10527 | + sm_group_t *sg = sev->se_sg; | |
10528 | + | |
10529 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10530 | + return FALSE; | |
10531 | + | |
10532 | + clear_bit(SEFL_ALLOW_LEAVE, &sev->se_flags); | |
10533 | + | |
10534 | + return TRUE; | |
10535 | +} | |
10536 | + | |
10537 | +/* | |
10538 | + * Same as previous. | |
10539 | + */ | |
10540 | + | |
10541 | +static int backout_leave_acked(sm_sevent_t *sev) | |
10542 | +{ | |
10543 | + return backout_leave_ackwait(sev); | |
10544 | +} | |
10545 | + | |
10546 | +/* | |
10547 | + * Ignore any stop replies. All the members will be stopped anyway to do the | |
10548 | + * recovery. Let that happen and restart our leave when done. | |
10549 | + */ | |
10550 | + | |
10551 | +static int backout_lstop_ackwait(sm_sevent_t *sev) | |
10552 | +{ | |
10553 | + sm_group_t *sg = sev->se_sg; | |
10554 | + | |
10555 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10556 | + return FALSE; | |
10557 | + | |
10558 | + clear_bit(SEFL_ALLOW_LSTOP, &sev->se_flags); | |
10559 | + | |
10560 | + return TRUE; | |
10561 | +} | |
10562 | + | |
10563 | +/* | |
10564 | + * Same as previous. | |
10565 | + */ | |
10566 | + | |
10567 | +static int backout_lstop_acked(sm_sevent_t *sev) | |
10568 | +{ | |
10569 | + return backout_lstop_ackwait(sev); | |
10570 | +} | |
10571 | + | |
10572 | +/* | |
10573 | + * All members will be stopped due to recovery and restarted by recovery | |
10574 | + * processing. That includes us, we have to retry the leave once the recovery | |
10575 | + * is done. | |
10576 | + */ | |
10577 | + | |
10578 | +static int backout_lstart_waitremote(sm_sevent_t *sev) | |
10579 | +{ | |
10580 | + sm_group_t *sg = sev->se_sg; | |
10581 | + | |
10582 | + if (!test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
10583 | + return FALSE; | |
10584 | + | |
10585 | + return TRUE; | |
10586 | +} | |
10587 | + | |
10588 | +/* | |
10589 | + * Reset an sevent to its beginning so it can be restarted. This is necessary | |
10590 | + * when recovery affects an SG while we're trying to join or leave (ie. a node | |
10591 | + * in the SG fails). | |
10592 | + */ | |
10593 | + | |
10594 | +void backout_sevents(void) | |
10595 | +{ | |
10596 | + sm_sevent_t *sev, *safe; | |
10597 | + int delay; | |
10598 | + | |
10599 | + list_for_each_entry_safe(sev, safe, &joinleave_events, se_list) { | |
10600 | + | |
10601 | + delay = FALSE; | |
10602 | + | |
10603 | + log_debug(sev->se_sg, "backout sevent state %u", sev->se_state); | |
10604 | + | |
10605 | + switch (sev->se_state) { | |
10606 | + | |
10607 | + /* backout after kcl_join_service and before | |
10608 | + * send_join_notice */ | |
10609 | + case SEST_JOIN_BEGIN: | |
10610 | + break; | |
10611 | + | |
10612 | + /* backout after send_join_notice and before final | |
10613 | + * process_reply */ | |
10614 | + case SEST_JOIN_ACKWAIT: | |
10615 | + clear_bit(SEFL_ALLOW_JOIN, &sev->se_flags); | |
10616 | + sev->se_state = SEST_JOIN_BEGIN; | |
10617 | + schedule_sev_restart(sev); | |
10618 | + break; | |
10619 | + | |
10620 | + /* backout after final process_reply and before | |
10621 | + * check_join_notice */ | |
10622 | + case SEST_JOIN_ACKED: | |
10623 | + delay = backout_join_acked(sev); | |
10624 | + break; | |
10625 | + | |
10626 | + /* backout after send_join_stop and before final | |
10627 | + * process_reply */ | |
10628 | + case SEST_JSTOP_ACKWAIT: | |
10629 | + delay = backout_jstop_ackwait(sev); | |
10630 | + break; | |
10631 | + | |
10632 | + /* backout after final process_reply and before | |
10633 | + * check_join_stop */ | |
10634 | + case SEST_JSTOP_ACKED: | |
10635 | + delay = backout_jstop_acked(sev); | |
10636 | + break; | |
10637 | + | |
10638 | + /* backout after send_join_start and before | |
10639 | + * kcl_start_done */ | |
10640 | + case SEST_JSTART_SERVICEWAIT: | |
10641 | + delay = backout_jstart_servicewait(sev); | |
10642 | + break; | |
10643 | + | |
10644 | + /* backout after kcl_start_done and before | |
10645 | + * startdone_barrier_new */ | |
10646 | + case SEST_JSTART_SERVICEDONE: | |
10647 | + delay = backout_jstart_servicedone(sev); | |
10648 | + break; | |
10649 | + | |
10650 | + /* backout after startdone_barrier_new and before | |
10651 | + * callback_startdone_barrier_new */ | |
10652 | + case SEST_BARRIER_WAIT: | |
10653 | + delay = backout_barrier_wait(sev); | |
10654 | + break; | |
10655 | + | |
10656 | + /* backout after callback_startdone_barrier_new and | |
10657 | + * before check_startdone_barrier_new */ | |
10658 | + case SEST_BARRIER_DONE: | |
10659 | + delay = backout_barrier_done(sev); | |
10660 | + break; | |
10661 | + | |
10662 | + /* backout after kcl_leave_service and before | |
10663 | + * send_leave_notice */ | |
10664 | + case SEST_LEAVE_BEGIN: | |
10665 | + delay = backout_leave_begin(sev); | |
10666 | + break; | |
10667 | + | |
10668 | + /* backout after send_leave_notice and before final | |
10669 | + * process_reply */ | |
10670 | + case SEST_LEAVE_ACKWAIT: | |
10671 | + delay = backout_leave_ackwait(sev); | |
10672 | + break; | |
10673 | + | |
10674 | + /* backout after final process_reply and before | |
10675 | + * check_leave_notice */ | |
10676 | + case SEST_LEAVE_ACKED: | |
10677 | + delay = backout_leave_acked(sev); | |
10678 | + break; | |
10679 | + | |
10680 | + /* backout after send_leave_stop and before final | |
10681 | + * process_reply */ | |
10682 | + case SEST_LSTOP_ACKWAIT: | |
10683 | + delay = backout_lstop_ackwait(sev); | |
10684 | + break; | |
10685 | + | |
10686 | + /* backout after final process_reply and before | |
10687 | + * check_leave_stop */ | |
10688 | + case SEST_LSTOP_ACKED: | |
10689 | + delay = backout_lstop_acked(sev); | |
10690 | + break; | |
10691 | + | |
10692 | + /* backout after send_leave_start and before | |
10693 | + * process_lstart_done */ | |
10694 | + case SEST_LSTART_WAITREMOTE: | |
10695 | + delay = backout_lstart_waitremote(sev); | |
10696 | + break; | |
10697 | + | |
10698 | + /* backout after process_lstart_done and before | |
10699 | + * process_leave_sevent */ | |
10700 | + case SEST_LSTART_REMOTEDONE: | |
10701 | + sevent_done(sev); | |
10702 | + delay = FALSE; | |
10703 | + break; | |
10704 | + | |
10705 | + default: | |
10706 | + log_error(sev->se_sg, "backout_sevents: bad state %d", | |
10707 | + sev->se_state); | |
10708 | + } | |
10709 | + | |
10710 | + if (delay) { | |
10711 | + set_bit(SEFL_DELAY, &sev->se_flags); | |
10712 | + | |
10713 | + if (test_bit(SEFL_LEAVE, &sev->se_flags)) { | |
10714 | + sev->se_state = SEST_LEAVE_BEGIN; | |
10715 | + /* The DELAY flag will be cleared once recovery | |
10716 | + * is done allowing the leave to be retried. */ | |
10717 | + } else { | |
10718 | + sev->se_state = SEST_JOIN_BEGIN; | |
10719 | + /* restart timer function will clear DELAY */ | |
10720 | + schedule_sev_restart(sev); | |
10721 | + } | |
10722 | + } | |
10723 | + } | |
10724 | +} | |
10725 | + | |
10726 | +void process_joinleave(void) | |
10727 | +{ | |
10728 | + sm_sevent_t *sev = NULL, *safe; | |
10729 | + | |
10730 | + spin_lock(&new_event_lock); | |
10731 | + if (!list_empty(&new_event)) { | |
10732 | + sev = list_entry(new_event.next, sm_sevent_t, se_list); | |
10733 | + list_del(&sev->se_list); | |
10734 | + list_add_tail(&sev->se_list, &joinleave_events); | |
10735 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
10736 | + } | |
10737 | + spin_unlock(&new_event_lock); | |
10738 | + | |
10739 | + list_for_each_entry_safe(sev, safe, &joinleave_events, se_list) { | |
10740 | + if (!test_and_clear_bit(SEFL_CHECK, &sev->se_flags)) | |
10741 | + continue; | |
10742 | + | |
10743 | + if (test_bit(SEFL_DELAY, &sev->se_flags)) | |
10744 | + continue; | |
10745 | + | |
10746 | + if (sev->se_state < SEST_LEAVE_BEGIN) | |
10747 | + process_join_sevent(sev); | |
10748 | + else | |
10749 | + process_leave_sevent(sev); | |
10750 | + } | |
10751 | +} | |
10752 | diff -urN linux-orig/cluster/cman/sm_joinleave.h linux-patched/cluster/cman/sm_joinleave.h | |
10753 | --- linux-orig/cluster/cman/sm_joinleave.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 10754 | +++ linux-patched/cluster/cman/sm_joinleave.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 10755 | @@ -0,0 +1,23 @@ |
10756 | +/****************************************************************************** | |
10757 | +******************************************************************************* | |
10758 | +** | |
10759 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
10760 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
10761 | +** | |
10762 | +** This copyrighted material is made available to anyone wishing to use, | |
10763 | +** modify, copy, or redistribute it subject to the terms and conditions | |
10764 | +** of the GNU General Public License v.2. | |
10765 | +** | |
10766 | +******************************************************************************* | |
10767 | +******************************************************************************/ | |
10768 | + | |
10769 | +#ifndef __SM_JOINLEAVE_DOT_H__ | |
10770 | +#define __SM_JOINLEAVE_DOT_H__ | |
10771 | + | |
10772 | +void init_joinleave(void); | |
10773 | +void new_joinleave(sm_sevent_t *sev); | |
10774 | +void process_joinleave(void); | |
10775 | +void backout_sevents(void); | |
10776 | +sm_sevent_t *find_sevent(unsigned int id); | |
10777 | + | |
10778 | +#endif | |
10779 | diff -urN linux-orig/cluster/cman/sm_membership.c linux-patched/cluster/cman/sm_membership.c | |
10780 | --- linux-orig/cluster/cman/sm_membership.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 10781 | +++ linux-patched/cluster/cman/sm_membership.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 10782 | @@ -0,0 +1,696 @@ |
10783 | +/****************************************************************************** | |
10784 | +******************************************************************************* | |
10785 | +** | |
10786 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
10787 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
10788 | +** | |
10789 | +** This copyrighted material is made available to anyone wishing to use, | |
10790 | +** modify, copy, or redistribute it subject to the terms and conditions | |
10791 | +** of the GNU General Public License v.2. | |
10792 | +** | |
10793 | +******************************************************************************* | |
10794 | +******************************************************************************/ | |
10795 | + | |
10796 | +#include "sm.h" | |
10797 | + | |
10798 | +extern struct list_head sm_members; | |
10799 | + | |
10800 | +/* | |
10801 | + * Routines for SG members to handle other nodes joining or leaving the SG. | |
10802 | + * These "uevent" membership update routines are the response to an "sevent" on | |
10803 | + * a joining/leaving node. | |
10804 | + */ | |
10805 | + | |
10806 | +static void del_memb_node(sm_group_t *sg, uint32_t nodeid) | |
10807 | +{ | |
10808 | + sm_node_t *node; | |
10809 | + | |
10810 | + list_for_each_entry(node, &sg->memb, list) { | |
10811 | + if (node->id != nodeid) | |
10812 | + continue; | |
10813 | + list_del(&node->list); | |
10814 | + kfree(node); | |
10815 | + sg->memb_count--; | |
10816 | + log_debug(sg, "del node %u count %d", nodeid, sg->memb_count); | |
10817 | + break; | |
10818 | + } | |
10819 | +} | |
10820 | + | |
10821 | +static void add_memb_node(sm_group_t *sg, sm_node_t *node) | |
10822 | +{ | |
10823 | + list_add_tail(&node->list, &sg->memb); | |
10824 | + sg->memb_count++; | |
10825 | + log_debug(sg, "add node %u count %d", node->id, sg->memb_count); | |
10826 | +} | |
10827 | + | |
10828 | +/* | |
10829 | + * Join 1. The receive end of send_join_stop() from a node requesting to join | |
10830 | + * the SG. We stop the service so it can be restarted with the new node. | |
10831 | + */ | |
10832 | + | |
10833 | +static int process_join_stop(sm_group_t *sg) | |
10834 | +{ | |
10835 | + sm_uevent_t *uev = &sg->uevent; | |
10836 | + sm_node_t *node; | |
10837 | + sm_msg_t reply; | |
10838 | + int error; | |
10839 | + | |
10840 | + if (uev->ue_num_nodes != sg->memb_count + 1) { | |
10841 | + log_error(sg, "process_join_stop: bad num nodes %u %u", | |
10842 | + uev->ue_num_nodes, sg->memb_count); | |
10843 | + return -1; | |
10844 | + } | |
10845 | + | |
10846 | + sm_set_event_id(&uev->ue_id); | |
10847 | + | |
10848 | + node = sm_find_joiner(sg, uev->ue_nodeid); | |
10849 | + SM_ASSERT(node,); | |
10850 | + | |
10851 | + sg->state = SGST_UEVENT; | |
10852 | + sg->ops->stop(sg->service_data); | |
10853 | + | |
10854 | + reply.ms_type = SMSG_JSTOP_REP; | |
10855 | + reply.ms_status = STATUS_POS; | |
10856 | + reply.ms_sevent_id = uev->ue_remote_seid; | |
10857 | + smsg_bswap_out(&reply); | |
10858 | + | |
10859 | + error = send_nodeid_message((char *) &reply, sizeof(reply), | |
10860 | + uev->ue_nodeid); | |
10861 | + if (error < 0) | |
10862 | + return error; | |
10863 | + return 0; | |
10864 | +} | |
10865 | + | |
10866 | +/* | |
10867 | + * Join 2. The receive end of send_join_start() from a node joining the SG. | |
10868 | + * We are re-starting the service with the new member added. | |
10869 | + */ | |
10870 | + | |
10871 | +static int process_join_start(sm_group_t *sg) | |
10872 | +{ | |
10873 | + sm_uevent_t *uev = &sg->uevent; | |
10874 | + sm_node_t *node; | |
10875 | + uint32_t *memb; | |
10876 | + int count = 0; | |
10877 | + | |
10878 | + /* this memory is passed to the service which must free it */ | |
10879 | + SM_RETRY(memb = | |
10880 | + kmalloc((sg->memb_count + 1) * sizeof(uint32_t), GFP_KERNEL), | |
10881 | + memb); | |
10882 | + | |
10883 | + /* transfer joining node from joining list to member list */ | |
10884 | + node = sm_find_joiner(sg, uev->ue_nodeid); | |
10885 | + SM_ASSERT(node, printk("nodeid=%u\n", uev->ue_nodeid);); | |
10886 | + list_del(&node->list); | |
10887 | + add_memb_node(sg, node); | |
10888 | + | |
10889 | + /* the new member list for the service */ | |
10890 | + list_for_each_entry(node, &sg->memb, list) | |
10891 | + memb[count++] = node->id; | |
10892 | + | |
10893 | + set_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
10894 | + | |
10895 | + sg->ops->start(sg->service_data, memb, count, uev->ue_id, | |
10896 | + SERVICE_NODE_JOIN); | |
10897 | + return 0; | |
10898 | +} | |
10899 | + | |
10900 | +/* | |
10901 | + * Join 3. When done starting their local service, every previous SG member | |
10902 | + * calls startdone_barrier() and the new/joining member calls | |
10903 | + * startdone_barrier_new(). The barrier returns when everyone has started | |
10904 | + * their service and joined the barrier. | |
10905 | + */ | |
10906 | + | |
10907 | +static int startdone_barrier(sm_group_t *sg) | |
10908 | +{ | |
10909 | + sm_uevent_t *uev = &sg->uevent; | |
10910 | + char bname[MAX_BARRIER_NAME_LEN]; | |
10911 | + int error; | |
10912 | + | |
10913 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
10914 | + uev->ue_barrier_status = -1; | |
10915 | + | |
10916 | + set_bit(UEFL_ALLOW_BARRIER, &uev->ue_flags); | |
10917 | + | |
10918 | + /* If we're the only member, skip the barrier */ | |
10919 | + if (sg->memb_count == 1) { | |
10920 | + process_startdone_barrier(sg, 0); | |
10921 | + return 0; | |
10922 | + } | |
10923 | + | |
10924 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
10925 | + sg->global_id, uev->ue_nodeid, uev->ue_remote_seid, | |
10926 | + sg->memb_count); | |
10927 | + | |
10928 | + error = sm_barrier(bname, sg->memb_count, SM_BARRIER_STARTDONE); | |
10929 | + | |
10930 | + return error; | |
10931 | +} | |
10932 | + | |
10933 | +/* | |
10934 | + * Join 4. Check that the "all started" barrier returned a successful status. | |
10935 | + * The newly joined member calls check_startdone_barrier_new(). | |
10936 | + */ | |
10937 | + | |
10938 | +static int check_startdone_barrier(sm_group_t *sg) | |
10939 | +{ | |
10940 | + int error = sg->uevent.ue_barrier_status; | |
10941 | + return error; | |
10942 | +} | |
10943 | + | |
10944 | +/* | |
10945 | + * Join 5. Send the service a "finish" indicating that all members have | |
10946 | + * successfully started. The newly joined member calls do_finish_new(). | |
10947 | + */ | |
10948 | + | |
10949 | +static void do_finish(sm_group_t *sg) | |
10950 | +{ | |
10951 | + sg->state = SGST_RUN; | |
10952 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
10953 | + sg->ops->finish(sg->service_data, sg->uevent.ue_id); | |
10954 | +} | |
10955 | + | |
10956 | +/* | |
10957 | + * Join 6. The uevent is done. If this was a uevent for a node leaving the | |
10958 | + * SG, then send a final message to the departed node signalling that the | |
10959 | + * remaining nodes have restarted since it left. | |
10960 | + */ | |
10961 | + | |
10962 | +static void uevent_done(sm_group_t *sg) | |
10963 | +{ | |
10964 | + sm_uevent_t *uev = &sg->uevent; | |
10965 | + sm_msg_t reply; | |
10966 | + | |
10967 | + if (test_bit(UEFL_LEAVE, &uev->ue_flags)) { | |
10968 | + reply.ms_type = SMSG_LSTART_DONE; | |
10969 | + reply.ms_status = STATUS_POS; | |
10970 | + reply.ms_sevent_id = uev->ue_remote_seid; | |
10971 | + smsg_bswap_out(&reply); | |
10972 | + send_nodeid_message((char *) &reply, sizeof(reply), | |
10973 | + uev->ue_nodeid); | |
10974 | + } | |
10975 | + memset(&sg->uevent, 0, sizeof(sm_uevent_t)); | |
10976 | +} | |
10977 | + | |
10978 | +/* | |
10979 | + * Leave 1. The receive end of send_leave_stop() from a node leaving the SG. | |
10980 | + */ | |
10981 | + | |
10982 | +static int process_leave_stop(sm_group_t *sg) | |
10983 | +{ | |
10984 | + sm_uevent_t *uev = &sg->uevent; | |
10985 | + sm_msg_t reply; | |
10986 | + int error; | |
10987 | + | |
10988 | + sm_set_event_id(&uev->ue_id); | |
10989 | + | |
10990 | + sg->state = SGST_UEVENT; | |
10991 | + sg->ops->stop(sg->service_data); | |
10992 | + | |
10993 | + reply.ms_type = SMSG_LSTOP_REP; | |
10994 | + reply.ms_status = STATUS_POS; | |
10995 | + reply.ms_sevent_id = uev->ue_remote_seid; | |
10996 | + smsg_bswap_out(&reply); | |
10997 | + | |
10998 | + error = send_nodeid_message((char *) &reply, sizeof(reply), | |
10999 | + uev->ue_nodeid); | |
11000 | + if (error < 0) | |
11001 | + return error; | |
11002 | + return 0; | |
11003 | +} | |
11004 | + | |
11005 | +/* | |
11006 | + * Leave 2. The receive end of send_leave_start() from a node leaving the SG. | |
11007 | + * We are re-starting the service (without the node that's left naturally.) | |
11008 | + */ | |
11009 | + | |
11010 | +static int process_leave_start(sm_group_t *sg) | |
11011 | +{ | |
11012 | + sm_uevent_t *uev = &sg->uevent; | |
11013 | + sm_node_t *node; | |
11014 | + uint32_t *memb; | |
11015 | + int count = 0; | |
11016 | + | |
11017 | + SM_ASSERT(sg->memb_count > 1, | |
11018 | + printk("memb_count=%u\n", sg->memb_count);); | |
11019 | + | |
11020 | + /* this memory is passed to the service which must free it */ | |
11021 | + SM_RETRY(memb = | |
11022 | + kmalloc((sg->memb_count - 1) * sizeof(uint32_t), GFP_KERNEL), | |
11023 | + memb); | |
11024 | + | |
11025 | + /* remove departed member from sg member list */ | |
11026 | + del_memb_node(sg, uev->ue_nodeid); | |
11027 | + | |
11028 | + /* build member list to pass to service */ | |
11029 | + list_for_each_entry(node, &sg->memb, list) | |
11030 | + memb[count++] = node->id; | |
11031 | + | |
11032 | + /* allow us to accept the start_done callback for this start */ | |
11033 | + set_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11034 | + | |
11035 | + sg->ops->start(sg->service_data, memb, count, uev->ue_id, | |
11036 | + SERVICE_NODE_LEAVE); | |
11037 | + return 0; | |
11038 | +} | |
11039 | + | |
11040 | +/* | |
11041 | + * Move through the steps of another node joining or leaving the SG. | |
11042 | + */ | |
11043 | + | |
11044 | +static void process_one_uevent(sm_group_t *sg) | |
11045 | +{ | |
11046 | + sm_uevent_t *uev = &sg->uevent; | |
11047 | + int error = 0; | |
11048 | + | |
11049 | + log_debug(sg, "uevent state %u node %u", uev->ue_state, uev->ue_nodeid); | |
11050 | + | |
11051 | + switch (uev->ue_state) { | |
11052 | + | |
11053 | + /* | |
11054 | + * a uevent is initialized with state JSTOP in | |
11055 | + * process_stop_request | |
11056 | + */ | |
11057 | + | |
11058 | + case UEST_JSTOP: | |
11059 | + uev->ue_state = UEST_JSTART_WAITCMD; | |
11060 | + error = process_join_stop(sg); | |
11061 | + break; | |
11062 | + | |
11063 | + /* | |
11064 | + * ue_state is changed from JSTART_WAITCMD to JSTART in | |
11065 | + * process_start_request | |
11066 | + */ | |
11067 | + | |
11068 | + case UEST_JSTART: | |
11069 | + uev->ue_state = UEST_JSTART_SERVICEWAIT; | |
11070 | + error = process_join_start(sg); | |
11071 | + break; | |
11072 | + | |
11073 | + /* | |
11074 | + * ue_state is changed from JSTART_SERVICEWAIT to | |
11075 | + * JSTART_SERVICEDONE in kcl_start_done | |
11076 | + */ | |
11077 | + | |
11078 | + case UEST_JSTART_SERVICEDONE: | |
11079 | + uev->ue_state = UEST_BARRIER_WAIT; | |
11080 | + error = startdone_barrier(sg); | |
11081 | + break; | |
11082 | + | |
11083 | + /* | |
11084 | + * ue_state is changed from BARRIER_WAIT to BARRIER_DONE in | |
11085 | + * process_startdone_barrier | |
11086 | + */ | |
11087 | + | |
11088 | + case UEST_BARRIER_DONE: | |
11089 | + error = check_startdone_barrier(sg); | |
11090 | + if (error) | |
11091 | + break; | |
11092 | + | |
11093 | + do_finish(sg); | |
11094 | + uevent_done(sg); | |
11095 | + break; | |
11096 | + | |
11097 | + /* | |
11098 | + * a uevent is initialized with state LSTOP in | |
11099 | + * process_stop_request | |
11100 | + */ | |
11101 | + | |
11102 | + case UEST_LSTOP: | |
11103 | + uev->ue_state = UEST_LSTART_WAITCMD; | |
11104 | + error = process_leave_stop(sg); | |
11105 | + break; | |
11106 | + | |
11107 | + /* | |
11108 | + * a uevent is changed from LSTART_WAITCMD to LSTART in | |
11109 | + * process_start_request | |
11110 | + */ | |
11111 | + | |
11112 | + case UEST_LSTART: | |
11113 | + uev->ue_state = UEST_LSTART_SERVICEWAIT; | |
11114 | + error = process_leave_start(sg); | |
11115 | + break; | |
11116 | + | |
11117 | + /* | |
11118 | + * a uevent is changed from LSTART_SERVICEWAIT to to | |
11119 | + * LSTART_SERVICEDONE in kcl_start_done | |
11120 | + */ | |
11121 | + | |
11122 | + case UEST_LSTART_SERVICEDONE: | |
11123 | + uev->ue_state = UEST_BARRIER_WAIT; | |
11124 | + error = startdone_barrier(sg); | |
11125 | + break; | |
11126 | + | |
11127 | + default: | |
11128 | + error = -1; | |
11129 | + } | |
11130 | + | |
11131 | + /* If we encounter an error during these routines, we do nothing, | |
11132 | + expecting that a node failure related to this sg will cause a | |
11133 | + recovery event to arrive and call cancel_one_uevent(). */ | |
11134 | + | |
11135 | + if (error) | |
11136 | + log_error(sg, "process_one_uevent error %d state %u", | |
11137 | + error, uev->ue_state); | |
11138 | +} | |
11139 | + | |
11140 | +static sm_node_t *failed_memb(sm_group_t *sg, int *count) | |
11141 | +{ | |
11142 | + sm_node_t *node, *sm_node, *failed_uev_node = NULL; | |
11143 | + | |
11144 | + list_for_each_entry(node, &sg->memb, list) { | |
11145 | + | |
11146 | + sm_node = sm_find_member(node->id); | |
11147 | + SM_ASSERT(sm_node, ); | |
11148 | + | |
11149 | + if (test_bit(SNFL_NEED_RECOVERY, &sm_node->flags)) { | |
11150 | + (*count)++; | |
11151 | + if (node->id == sg->uevent.ue_nodeid) | |
11152 | + failed_uev_node = sm_node; | |
11153 | + } | |
11154 | + } | |
11155 | + return failed_uev_node; | |
11156 | +} | |
11157 | + | |
11158 | +static void send_recover_msg(sm_group_t *sg) | |
11159 | +{ | |
11160 | + char *msg; | |
11161 | + int len = 0; | |
11162 | + msg = create_smsg(sg, SMSG_RECOVER, 0, &len, NULL); | |
11163 | + send_members_message(sg, msg, len); | |
11164 | +} | |
11165 | + | |
11166 | +static void cancel_barrier(sm_group_t *sg) | |
11167 | +{ | |
11168 | + sm_uevent_t *uev = &sg->uevent; | |
11169 | + char bname[MAX_BARRIER_NAME_LEN]; | |
11170 | + | |
11171 | + clear_bit(UEFL_ALLOW_BARRIER, &uev->ue_flags); | |
11172 | + | |
11173 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
11174 | + snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.%u.%u", | |
11175 | + sg->global_id, uev->ue_nodeid, uev->ue_remote_seid, | |
11176 | + sg->memb_count); | |
11177 | + kcl_barrier_cancel(bname); | |
11178 | +} | |
11179 | + | |
11180 | +static void cancel_one_uevent(sm_group_t *sg, int *effected) | |
11181 | +{ | |
11182 | + sm_uevent_t *uev = &sg->uevent; | |
11183 | + int failed_count; | |
11184 | + sm_node_t *node, *failed_joiner, *failed_leaver; | |
11185 | + | |
11186 | + log_debug(sg, "cancel uevent state %u node %u", uev->ue_state, | |
11187 | + uev->ue_nodeid); | |
11188 | + | |
11189 | + switch (uev->ue_state) { | |
11190 | + | |
11191 | + case UEST_JSTOP: | |
11192 | + case UEST_JSTART_WAITCMD: | |
11193 | + case UEST_JSTART: | |
11194 | + | |
11195 | + sg->ops->stop(sg->service_data); | |
11196 | + | |
11197 | + failed_count = 0; | |
11198 | + failed_joiner = failed_memb(sg, &failed_count); | |
11199 | + SM_ASSERT(!failed_joiner, ); | |
11200 | + | |
11201 | + node = sm_find_member(uev->ue_nodeid); | |
11202 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11203 | + failed_joiner = node; | |
11204 | + | |
11205 | + if (!failed_count) { | |
11206 | + /* only joining node failed */ | |
11207 | + SM_ASSERT(failed_joiner, ); | |
11208 | + SM_ASSERT(!test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11209 | + set_bit(SGFL_NEED_RECOVERY, &sg->flags); | |
11210 | + (*effected)++; | |
11211 | + /* some nodes may not have gotten a JSTOP message | |
11212 | + in which case this will tell them to begin | |
11213 | + recovery for this sg. */ | |
11214 | + send_recover_msg(sg); | |
11215 | + | |
11216 | + } else { | |
11217 | + /* a member node failed (and possibly joining node, it | |
11218 | + doesn't matter) */ | |
11219 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11220 | + } | |
11221 | + | |
11222 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11223 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11224 | + break; | |
11225 | + | |
11226 | + | |
11227 | + case UEST_JSTART_SERVICEWAIT: | |
11228 | + case UEST_JSTART_SERVICEDONE: | |
11229 | + | |
11230 | + clear_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11231 | + sg->ops->stop(sg->service_data); | |
11232 | + | |
11233 | + failed_count = 0; | |
11234 | + failed_joiner = failed_memb(sg, &failed_count); | |
11235 | + SM_ASSERT(failed_count, ); | |
11236 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11237 | + | |
11238 | + if (failed_count == 1 && failed_joiner) { | |
11239 | + /* only joining node failed */ | |
11240 | + | |
11241 | + } else if (failed_count && failed_joiner) { | |
11242 | + /* joining node and another member failed */ | |
11243 | + | |
11244 | + } else { | |
11245 | + /* other member failed, joining node still alive */ | |
11246 | + SM_ASSERT(!failed_joiner, ); | |
11247 | + del_memb_node(sg, uev->ue_nodeid); | |
11248 | + } | |
11249 | + | |
11250 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11251 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11252 | + break; | |
11253 | + | |
11254 | + | |
11255 | + case UEST_LSTOP: | |
11256 | + case UEST_LSTART_WAITCMD: | |
11257 | + case UEST_LSTART: | |
11258 | + | |
11259 | + sg->ops->stop(sg->service_data); | |
11260 | + | |
11261 | + failed_count = 0; | |
11262 | + failed_leaver = failed_memb(sg, &failed_count); | |
11263 | + SM_ASSERT(failed_count, ); | |
11264 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11265 | + | |
11266 | + if (failed_count == 1 && failed_leaver) { | |
11267 | + /* only leaving node failed */ | |
11268 | + | |
11269 | + } else if (failed_count && failed_leaver) { | |
11270 | + /* leaving node and another member failed */ | |
11271 | + | |
11272 | + } else { | |
11273 | + /* other member failed, leaving node still alive */ | |
11274 | + SM_ASSERT(!failed_leaver, ); | |
11275 | + } | |
11276 | + | |
11277 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11278 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11279 | + break; | |
11280 | + | |
11281 | + | |
11282 | + case UEST_LSTART_SERVICEWAIT: | |
11283 | + case UEST_LSTART_SERVICEDONE: | |
11284 | + | |
11285 | + clear_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags); | |
11286 | + sg->ops->stop(sg->service_data); | |
11287 | + | |
11288 | + failed_count = 0; | |
11289 | + failed_leaver = failed_memb(sg, &failed_count); | |
11290 | + SM_ASSERT(!failed_leaver, ); | |
11291 | + | |
11292 | + node = sm_find_member(uev->ue_nodeid); | |
11293 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11294 | + failed_leaver = node; | |
11295 | + | |
11296 | + if (!failed_count) { | |
11297 | + /* only leaving node failed */ | |
11298 | + SM_ASSERT(failed_leaver, ); | |
11299 | + SM_ASSERT(!test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11300 | + set_bit(SGFL_NEED_RECOVERY, &sg->flags); | |
11301 | + (*effected)++; | |
11302 | + | |
11303 | + } else if (failed_count && failed_leaver) { | |
11304 | + /* leaving node and another member failed */ | |
11305 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11306 | + | |
11307 | + } else { | |
11308 | + /* other member failed, leaving node still alive */ | |
11309 | + SM_ASSERT(failed_count, ); | |
11310 | + SM_ASSERT(!failed_leaver, ); | |
11311 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11312 | + node = sm_new_node(sg->uevent.ue_nodeid); | |
11313 | + add_memb_node(sg, node); | |
11314 | + } | |
11315 | + | |
11316 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11317 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11318 | + break; | |
11319 | + | |
11320 | + | |
11321 | + case UEST_BARRIER_WAIT: | |
11322 | + | |
11323 | + if (test_bit(UEFL_LEAVE, &uev->ue_flags)) | |
11324 | + goto barrier_wait_leave; | |
11325 | + | |
11326 | + sg->ops->stop(sg->service_data); | |
11327 | + cancel_barrier(sg); | |
11328 | + | |
11329 | + barrier_wait_join: | |
11330 | + | |
11331 | + failed_count = 0; | |
11332 | + failed_joiner = failed_memb(sg, &failed_count); | |
11333 | + SM_ASSERT(failed_count, ); | |
11334 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11335 | + | |
11336 | + if (failed_count == 1 && failed_joiner) { | |
11337 | + /* only joining node failed */ | |
11338 | + | |
11339 | + } else if (failed_count && failed_joiner) { | |
11340 | + /* joining node and another member failed */ | |
11341 | + | |
11342 | + } else { | |
11343 | + /* other member failed, joining node still alive */ | |
11344 | + SM_ASSERT(!failed_joiner, ); | |
11345 | + del_memb_node(sg, uev->ue_nodeid); | |
11346 | + } | |
11347 | + | |
11348 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11349 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11350 | + break; | |
11351 | + | |
11352 | + barrier_wait_leave: | |
11353 | + | |
11354 | + failed_count = 0; | |
11355 | + failed_leaver = failed_memb(sg, &failed_count); | |
11356 | + SM_ASSERT(!failed_leaver, ); | |
11357 | + | |
11358 | + node = sm_find_member(uev->ue_nodeid); | |
11359 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11360 | + failed_leaver = node; | |
11361 | + | |
11362 | + if (!failed_count) { | |
11363 | + /* only leaving node failed */ | |
11364 | + SM_ASSERT(failed_leaver, ); | |
11365 | + SM_ASSERT(!test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11366 | + set_bit(SGFL_NEED_RECOVERY, &sg->flags); | |
11367 | + (*effected)++; | |
11368 | + | |
11369 | + } else if (failed_count && failed_leaver) { | |
11370 | + /* leaving node and another member failed */ | |
11371 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11372 | + | |
11373 | + } else { | |
11374 | + /* other member failed, leaving node still alive */ | |
11375 | + SM_ASSERT(failed_count, ); | |
11376 | + SM_ASSERT(!failed_leaver, ); | |
11377 | + SM_ASSERT(test_bit(SGFL_NEED_RECOVERY, &sg->flags), ); | |
11378 | + node = sm_new_node(sg->uevent.ue_nodeid); | |
11379 | + add_memb_node(sg, node); | |
11380 | + } | |
11381 | + | |
11382 | + clear_bit(SGFL_UEVENT, &sg->flags); | |
11383 | + memset(uev, 0, sizeof(sm_uevent_t)); | |
11384 | + break; | |
11385 | + | |
11386 | + | |
11387 | + case UEST_BARRIER_DONE: | |
11388 | + | |
11389 | + if (!uev->ue_barrier_status) { | |
11390 | + do_finish(sg); | |
11391 | + uevent_done(sg); | |
11392 | + break; | |
11393 | + } | |
11394 | + | |
11395 | + if (test_bit(UEFL_LEAVE, &uev->ue_flags)) | |
11396 | + goto barrier_wait_leave; | |
11397 | + else | |
11398 | + goto barrier_wait_join; | |
11399 | + | |
11400 | + | |
11401 | + default: | |
11402 | + log_error(sg, "cancel_one_uevent: state %d", uev->ue_state); | |
11403 | + } | |
11404 | +} | |
11405 | + | |
11406 | +void cancel_uevents(int *effected) | |
11407 | +{ | |
11408 | + sm_group_t *sg; | |
11409 | + sm_node_t *node, *sgnode; | |
11410 | + int i; | |
11411 | + | |
11412 | + list_for_each_entry(node, &sm_members, list) { | |
11413 | + if (!test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11414 | + continue; | |
11415 | + | |
11416 | + /* | |
11417 | + * Clear this dead node from the "interested in joining" list | |
11418 | + * of any SG. The node is added to this list before the uevent | |
11419 | + * begins. | |
11420 | + */ | |
11421 | + | |
11422 | + for (i = 0; i < SG_LEVELS; i++) { | |
11423 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
11424 | + sgnode = sm_find_joiner(sg, node->id); | |
11425 | + if (sgnode) { | |
11426 | + log_debug(sg, "clear joining node %u", | |
11427 | + sgnode->id); | |
11428 | + list_del(&sgnode->list); | |
11429 | + kfree(sgnode); | |
11430 | + } | |
11431 | + } | |
11432 | + } | |
11433 | + } | |
11434 | + | |
11435 | + /* Adjust any uevents in sg's effected by the failed node(s) */ | |
11436 | + | |
11437 | + for (i = 0; i < SG_LEVELS; i++) { | |
11438 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
11439 | + if (!test_bit(SGFL_UEVENT, &sg->flags)) | |
11440 | + continue; | |
11441 | + | |
11442 | + /* We may have some cancelling to do if this sg is | |
11443 | + flagged as having a failed member, or if a joining | |
11444 | + or leaving node has died. */ | |
11445 | + | |
11446 | + if (test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
11447 | + cancel_one_uevent(sg, effected); | |
11448 | + else if (sg->uevent.ue_nodeid) { | |
11449 | + node = sm_find_member(sg->uevent.ue_nodeid); | |
11450 | + SM_ASSERT(node, ); | |
11451 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
11452 | + cancel_one_uevent(sg, effected); | |
11453 | + } | |
11454 | + } | |
11455 | + } | |
11456 | +} | |
11457 | + | |
11458 | +void process_membership(void) | |
11459 | +{ | |
11460 | + sm_group_t *sg; | |
11461 | + int i; | |
11462 | + | |
11463 | + down(&sm_sglock); | |
11464 | + | |
11465 | + for (i = 0; i < SG_LEVELS; i++) { | |
11466 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
11467 | + if (!test_bit(SGFL_UEVENT, &sg->flags)) | |
11468 | + continue; | |
11469 | + | |
11470 | + if (!test_and_clear_bit(UEFL_CHECK, | |
11471 | + &sg->uevent.ue_flags)) | |
11472 | + continue; | |
11473 | + | |
11474 | + process_one_uevent(sg); | |
11475 | + } | |
11476 | + } | |
11477 | + up(&sm_sglock); | |
11478 | +} | |
11479 | diff -urN linux-orig/cluster/cman/sm_membership.h linux-patched/cluster/cman/sm_membership.h | |
11480 | --- linux-orig/cluster/cman/sm_membership.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 11481 | +++ linux-patched/cluster/cman/sm_membership.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 11482 | @@ -0,0 +1,20 @@ |
11483 | +/****************************************************************************** | |
11484 | +******************************************************************************* | |
11485 | +** | |
11486 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
11487 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
11488 | +** | |
11489 | +** This copyrighted material is made available to anyone wishing to use, | |
11490 | +** modify, copy, or redistribute it subject to the terms and conditions | |
11491 | +** of the GNU General Public License v.2. | |
11492 | +** | |
11493 | +******************************************************************************* | |
11494 | +******************************************************************************/ | |
11495 | + | |
11496 | +#ifndef __SM_MEMBERSHIP_DOT_H__ | |
11497 | +#define __SM_MEMBERSHIP_DOT_H__ | |
11498 | + | |
11499 | +void process_membership(void); | |
11500 | +void cancel_uevents(int *effected); | |
11501 | + | |
11502 | +#endif | |
11503 | diff -urN linux-orig/cluster/cman/sm_message.c linux-patched/cluster/cman/sm_message.c | |
11504 | --- linux-orig/cluster/cman/sm_message.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 11505 | +++ linux-patched/cluster/cman/sm_message.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 11506 | @@ -0,0 +1,867 @@ |
11507 | +/****************************************************************************** | |
11508 | +******************************************************************************* | |
11509 | +** | |
11510 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
11511 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
11512 | +** | |
11513 | +** This copyrighted material is made available to anyone wishing to use, | |
11514 | +** modify, copy, or redistribute it subject to the terms and conditions | |
11515 | +** of the GNU General Public License v.2. | |
11516 | +** | |
11517 | +******************************************************************************* | |
11518 | +******************************************************************************/ | |
11519 | + | |
11520 | +#include "sm.h" | |
11521 | + | |
11522 | +#define SMSG_BUF_SIZE (sizeof(sm_msg_t) + MAX_SERVICE_NAME_LEN + 1) | |
11523 | + | |
11524 | +extern struct socket * sm_socket; | |
11525 | +extern uint32_t sm_our_nodeid; | |
11526 | +static uint32_t global_last_id; | |
11527 | +static struct list_head messages; | |
11528 | +static spinlock_t message_lock; | |
11529 | +static char smsg_buf[SMSG_BUF_SIZE]; | |
11530 | + | |
11531 | +int send_nodeid_message(char *msg, int len, uint32_t nodeid); | |
11532 | + | |
11533 | +struct rq_entry { | |
11534 | + struct list_head list; | |
11535 | + char *msg; | |
11536 | + int len; | |
11537 | + uint32_t nodeid; | |
11538 | +}; | |
11539 | +typedef struct rq_entry rq_entry_t; | |
11540 | + | |
11541 | +void init_messages(void) | |
11542 | +{ | |
11543 | + global_last_id = 1; | |
11544 | + INIT_LIST_HEAD(&messages); | |
11545 | + spin_lock_init(&message_lock); | |
11546 | +} | |
11547 | + | |
11548 | +uint32_t sm_new_global_id(int level) | |
11549 | +{ | |
11550 | + uint32_t id = global_last_id++; | |
11551 | + uint8_t l = (uint8_t) level; | |
11552 | + | |
11553 | + if (level > 255) | |
11554 | + return 0; | |
11555 | + | |
11556 | + if (id > 0x00FFFFFF) | |
11557 | + return 0; | |
11558 | + | |
11559 | + id |= (l << 24); | |
11560 | + return id; | |
11561 | +} | |
11562 | + | |
11563 | +static void smsg_copy_in(char *msg, sm_msg_t *smsg) | |
11564 | +{ | |
11565 | + sm_msg_t *in = (sm_msg_t *) msg; | |
11566 | + | |
11567 | + smsg->ms_type = in->ms_type; | |
11568 | + smsg->ms_status = in->ms_status; | |
11569 | + smsg->ms_sevent_id = le16_to_cpu(in->ms_sevent_id); | |
11570 | + smsg->ms_global_sgid = le32_to_cpu(in->ms_global_sgid); | |
11571 | + smsg->ms_global_lastid = le32_to_cpu(in->ms_global_lastid); | |
11572 | + smsg->ms_sglevel = le16_to_cpu(in->ms_sglevel); | |
11573 | + smsg->ms_length = le16_to_cpu(in->ms_length); | |
11574 | +} | |
11575 | + | |
11576 | +/* swapping bytes in place is an easy source of errors - be careful not to | |
11577 | + * access the fields after calling this */ | |
11578 | + | |
11579 | +void smsg_bswap_out(sm_msg_t *smsg) | |
11580 | +{ | |
11581 | + smsg->ms_sevent_id = cpu_to_le16(smsg->ms_sevent_id); | |
11582 | + smsg->ms_global_sgid = cpu_to_le32(smsg->ms_global_sgid); | |
11583 | + smsg->ms_global_lastid = cpu_to_le32(smsg->ms_global_lastid); | |
11584 | + smsg->ms_sglevel = cpu_to_le16(smsg->ms_sglevel); | |
11585 | + smsg->ms_length = cpu_to_le16(smsg->ms_length); | |
11586 | +} | |
11587 | + | |
11588 | +char *create_smsg(sm_group_t *sg, int type, int datalen, int *msglen, | |
11589 | + sm_sevent_t *sev) | |
11590 | +{ | |
11591 | + char *msg; | |
11592 | + sm_msg_t *smsg; | |
11593 | + int fulllen = sizeof(sm_msg_t) + datalen; | |
11594 | + | |
11595 | + msg = smsg_buf; | |
11596 | + memset(smsg_buf, 0, SMSG_BUF_SIZE); | |
11597 | + SM_ASSERT(fulllen <= SMSG_BUF_SIZE,); | |
11598 | + | |
11599 | + smsg = (sm_msg_t *) msg; | |
11600 | + smsg->ms_type = type; | |
11601 | + smsg->ms_global_sgid = sg->global_id; | |
11602 | + smsg->ms_sglevel = sg->level; | |
11603 | + smsg->ms_length = datalen; | |
11604 | + smsg->ms_sevent_id = sev ? sev->se_id : 0; | |
11605 | + | |
11606 | + smsg_bswap_out(smsg); | |
11607 | + *msglen = fulllen; | |
11608 | + return msg; | |
11609 | +} | |
11610 | + | |
11611 | +static unsigned int msgtype_to_flag(int type) | |
11612 | +{ | |
11613 | + unsigned int flag; | |
11614 | + | |
11615 | + switch (type) { | |
11616 | + case SMSG_JOIN_REP: | |
11617 | + case SMSG_JOIN_REQ: | |
11618 | + flag = SEFL_ALLOW_JOIN; | |
11619 | + break; | |
11620 | + | |
11621 | + case SMSG_JSTOP_REP: | |
11622 | + case SMSG_JSTOP_REQ: | |
11623 | + flag = SEFL_ALLOW_JSTOP; | |
11624 | + break; | |
11625 | + | |
11626 | + case SMSG_LEAVE_REP: | |
11627 | + case SMSG_LEAVE_REQ: | |
11628 | + flag = SEFL_ALLOW_LEAVE; | |
11629 | + break; | |
11630 | + | |
11631 | + case SMSG_LSTOP_REP: | |
11632 | + case SMSG_LSTOP_REQ: | |
11633 | + flag = SEFL_ALLOW_LSTOP; | |
11634 | + break; | |
11635 | + | |
11636 | + default: | |
11637 | + SM_ASSERT(0, printk("msgtype_to_flag bad type %d\n", type);); | |
11638 | + } | |
11639 | + return flag; | |
11640 | +} | |
11641 | + | |
11642 | +static int test_allowed_msgtype(sm_sevent_t * sev, int type) | |
11643 | +{ | |
11644 | + unsigned int flag = msgtype_to_flag(type); | |
11645 | + | |
11646 | + return test_bit(flag, &sev->se_flags); | |
11647 | +} | |
11648 | + | |
11649 | +static void clear_allowed_msgtype(sm_sevent_t * sev, int type) | |
11650 | +{ | |
11651 | + unsigned int flag = msgtype_to_flag(type); | |
11652 | + | |
11653 | + clear_bit(flag, &sev->se_flags); | |
11654 | +} | |
11655 | + | |
11656 | +static void set_allowed_msgtype(sm_sevent_t * sev, int type) | |
11657 | +{ | |
11658 | + unsigned int flag = msgtype_to_flag(type); | |
11659 | + | |
11660 | + set_bit(flag, &sev->se_flags); | |
11661 | +} | |
11662 | + | |
11663 | +static int save_global_id(sm_sevent_t * sev, sm_msg_t * smsg) | |
11664 | +{ | |
11665 | + sm_group_t *sg = sev->se_sg; | |
11666 | + | |
11667 | + if (!smsg->ms_global_sgid) { | |
11668 | + log_error(sg, "save_global_id: zero sg id"); | |
11669 | + return -1; | |
11670 | + } | |
11671 | + | |
11672 | + if (!sg->global_id) | |
11673 | + sg->global_id = smsg->ms_global_sgid; | |
11674 | + | |
11675 | + if (sg->global_id != smsg->ms_global_sgid) { | |
11676 | + log_error(sg, "save_global_id: id %x", smsg->ms_global_sgid); | |
11677 | + return -1; | |
11678 | + } | |
11679 | + return 0; | |
11680 | +} | |
11681 | + | |
11682 | +static void save_lastid(sm_msg_t * smsg) | |
11683 | +{ | |
11684 | + uint32_t gid = smsg->ms_global_lastid & 0x00FFFFFF; | |
11685 | + | |
11686 | + /* | |
11687 | + * Keep track of the highst SG id which has been used | |
11688 | + * in the cluster in case we need to choose a new SG id. | |
11689 | + */ | |
11690 | + | |
11691 | + if (gid > global_last_id) | |
11692 | + global_last_id = gid; | |
11693 | +} | |
11694 | + | |
11695 | +static int next_sev_state(int msg_type, int cur_state) | |
11696 | +{ | |
11697 | + int next = 0; | |
11698 | + | |
11699 | + switch (msg_type) { | |
11700 | + case SMSG_JOIN_REP: | |
11701 | + SM_ASSERT(cur_state == SEST_JOIN_ACKWAIT,); | |
11702 | + next = SEST_JOIN_ACKED; | |
11703 | + break; | |
11704 | + | |
11705 | + case SMSG_JSTOP_REP: | |
11706 | + SM_ASSERT(cur_state == SEST_JSTOP_ACKWAIT,); | |
11707 | + next = SEST_JSTOP_ACKED; | |
11708 | + break; | |
11709 | + | |
11710 | + case SMSG_LEAVE_REP: | |
11711 | + SM_ASSERT(cur_state == SEST_LEAVE_ACKWAIT,); | |
11712 | + next = SEST_LEAVE_ACKED; | |
11713 | + break; | |
11714 | + | |
11715 | + case SMSG_LSTOP_REP: | |
11716 | + SM_ASSERT(cur_state == SEST_LSTOP_ACKWAIT,); | |
11717 | + next = SEST_LSTOP_ACKED; | |
11718 | + break; | |
11719 | + } | |
11720 | + return next; | |
11721 | +} | |
11722 | + | |
11723 | +/* | |
11724 | + * Functions in sevent.c send messages to other nodes and then expect replies. | |
11725 | + * This function collects the replies for the sevent messages and moves the | |
11726 | + * sevent to the next stage when all the expected replies have been received. | |
11727 | + */ | |
11728 | + | |
11729 | +static void process_reply(sm_msg_t * smsg, uint32_t nodeid) | |
11730 | +{ | |
11731 | + sm_sevent_t *sev; | |
11732 | + int i, expected, type = smsg->ms_type; | |
11733 | + | |
11734 | + /* | |
11735 | + * Find the relevant sevent. | |
11736 | + */ | |
11737 | + | |
11738 | + sev = find_sevent(smsg->ms_sevent_id); | |
11739 | + if (!sev) { | |
11740 | + log_print("process_reply invalid id=%u nodeid=%u", | |
11741 | + smsg->ms_sevent_id, nodeid); | |
11742 | + goto out; | |
11743 | + } | |
11744 | + | |
11745 | + /* | |
11746 | + * Check if this message type is what this sevent is waiting for. | |
11747 | + */ | |
11748 | + | |
11749 | + if (!test_allowed_msgtype(sev, type)) { | |
11750 | + log_debug(sev->se_sg, "process_reply ignored type=%u nodeid=%u " "id=%u", type, nodeid, sev->se_id); | |
11751 | + goto out; | |
11752 | + } | |
11753 | + | |
11754 | + expected = | |
11755 | + (type == SMSG_JOIN_REP) ? sev->se_node_count : sev->se_memb_count; | |
11756 | + | |
11757 | + SM_ASSERT(expected * sizeof(uint32_t) <= sev->se_len_ids, | |
11758 | + printk("type=%d expected=%d len_ids=%d node_count=%d " | |
11759 | + "memb_count=%d\n", type, expected, sev->se_len_ids, | |
11760 | + sev->se_node_count, sev->se_memb_count);); | |
11761 | + | |
11762 | + SM_ASSERT(expected * sizeof(char) <= sev->se_len_status, | |
11763 | + printk("type=%d expected=%d len_status=%d node_count=%d " | |
11764 | + "memb_count=%d\n", type, expected, sev->se_len_status, | |
11765 | + sev->se_node_count, sev->se_memb_count);); | |
11766 | + | |
11767 | + for (i = 0; i < expected; i++) { | |
11768 | + if (sev->se_node_ids[i] == nodeid) { | |
11769 | + /* | |
11770 | + * Save the status from the replying node | |
11771 | + */ | |
11772 | + | |
11773 | + if (!sev->se_node_status[i]) | |
11774 | + sev->se_node_status[i] = smsg->ms_status; | |
11775 | + else { | |
11776 | + log_error(sev->se_sg, "process_reply duplicate" | |
11777 | + "id=%u nodeid=%u %u/%u", | |
11778 | + sev->se_id, nodeid, | |
11779 | + sev->se_node_status[i], | |
11780 | + smsg->ms_status); | |
11781 | + goto out; | |
11782 | + } | |
11783 | + | |
11784 | + if (type == SMSG_JOIN_REP) { | |
11785 | + save_lastid(smsg); | |
11786 | + | |
11787 | + if (smsg->ms_status == STATUS_POS) | |
11788 | + save_global_id(sev, smsg); | |
11789 | + } | |
11790 | + | |
11791 | + /* | |
11792 | + * Signal sm if we have all replies | |
11793 | + */ | |
11794 | + | |
11795 | + if (++sev->se_reply_count == expected) { | |
11796 | + clear_allowed_msgtype(sev, type); | |
11797 | + sev->se_state = next_sev_state(type, | |
11798 | + sev->se_state); | |
11799 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
11800 | + wake_serviced(DO_JOINLEAVE); | |
11801 | + } | |
11802 | + | |
11803 | + break; | |
11804 | + } | |
11805 | + } | |
11806 | + | |
11807 | + out: | |
11808 | + return; | |
11809 | +} | |
11810 | + | |
11811 | +/* | |
11812 | + * A node wants to join an SG and has run send_join_notice. If we know nothing | |
11813 | + * about the SG , then we have no objection - send back STATUS_POS. If we're a | |
11814 | + * member of the SG, then send back STATUS_POS (go ahead and join) if there's | |
11815 | + * no sevent or uevent of higher priority in progress (only a single join or | |
11816 | + * leave is permitted for the SG at once). If there happens to be a higher | |
11817 | + * priority sevent/uevent in progress, send back STATUS_WAIT to defer the | |
11818 | + * requested join for a bit. | |
11819 | + */ | |
11820 | + | |
11821 | +static void process_join_request(sm_msg_t *smsg, uint32_t nodeid, char *name) | |
11822 | +{ | |
11823 | + sm_group_t *sg = NULL; | |
11824 | + sm_sevent_t *sev = NULL; | |
11825 | + sm_node_t *node; | |
11826 | + int found = FALSE; | |
11827 | + int level = smsg->ms_sglevel; | |
11828 | + sm_msg_t reply; | |
11829 | + | |
11830 | + memset(&reply, 0, sizeof(reply)); | |
11831 | + | |
11832 | + down(&sm_sglock); | |
11833 | + | |
11834 | + if (nodeid == sm_our_nodeid) | |
11835 | + goto next; | |
11836 | + | |
11837 | + /* | |
11838 | + * search SG list for an SG with given name/len | |
11839 | + */ | |
11840 | + | |
11841 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
11842 | + if ((sg->namelen != smsg->ms_length) || | |
11843 | + memcmp(sg->name, name, sg->namelen)) | |
11844 | + continue; | |
11845 | + found = TRUE; | |
11846 | + break; | |
11847 | + } | |
11848 | + | |
11849 | + /* | |
11850 | + * build reply message | |
11851 | + */ | |
11852 | + | |
11853 | + next: | |
11854 | + | |
11855 | + if (!found) { | |
11856 | + reply.ms_type = SMSG_JOIN_REP; | |
11857 | + reply.ms_status = STATUS_NEG; | |
11858 | + reply.ms_global_lastid = global_last_id; | |
11859 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
11860 | + } else { | |
11861 | + reply.ms_type = SMSG_JOIN_REP; | |
11862 | + reply.ms_status = STATUS_POS; | |
11863 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
11864 | + reply.ms_global_sgid = sg->global_id; | |
11865 | + reply.ms_global_lastid = global_last_id; | |
11866 | + | |
11867 | + /* | |
11868 | + * The node trying to join should wait and try again until | |
11869 | + * we're done with recovery. | |
11870 | + */ | |
11871 | + | |
11872 | + if (sg->state == SGST_RECOVER) { | |
11873 | + reply.ms_status = STATUS_WAIT; | |
11874 | + goto send; | |
11875 | + } | |
11876 | + | |
11877 | + /* | |
11878 | + * An sevent node trying to join may have gotten as far as | |
11879 | + * creating a uevent with us and then backed out. That node | |
11880 | + * will retry joining from the beginning so we should not turn | |
11881 | + * them away. If we're handling a uevent for another node, | |
11882 | + * tell the joining node to wait. | |
11883 | + */ | |
11884 | + | |
11885 | + if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
11886 | + if (sg->uevent.ue_nodeid != nodeid) | |
11887 | + reply.ms_status = STATUS_WAIT; | |
11888 | + goto send; | |
11889 | + } | |
11890 | + | |
11891 | + /* | |
11892 | + * We're trying to join or leave the SG at the moment. | |
11893 | + */ | |
11894 | + | |
11895 | + if (test_bit(SGFL_SEVENT, &sg->flags)) { | |
11896 | + sev = sg->sevent; | |
11897 | + | |
11898 | + /* | |
11899 | + * We're trying to leave. Make the join wait until | |
11900 | + * we've left if we're beyond LEAVE_ACKWAIT. | |
11901 | + */ | |
11902 | + | |
11903 | + if (test_bit(SEFL_LEAVE, &sev->se_flags)) { | |
11904 | + if (sev->se_state > SEST_LEAVE_ACKED) | |
11905 | + reply.ms_status = STATUS_WAIT; | |
11906 | + else { | |
11907 | + reply.ms_status = STATUS_POS; | |
11908 | + clear_bit(SEFL_ALLOW_LEAVE, | |
11909 | + &sev->se_flags); | |
11910 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
11911 | + } | |
11912 | + } | |
11913 | + | |
11914 | + /* | |
11915 | + * We're trying to join. Making the other join wait | |
11916 | + * until we're joined if we're beyond JOIN_ACKWAIT or | |
11917 | + * if we have a lower id. (Send NEG to allow the other | |
11918 | + * node to go ahead because we're not in the SG.) | |
11919 | + */ | |
11920 | + | |
11921 | + else { | |
11922 | + if (sev->se_state > SEST_JOIN_ACKED) | |
11923 | + reply.ms_status = STATUS_WAIT; | |
11924 | + else if (sm_our_nodeid < nodeid) | |
11925 | + reply.ms_status = STATUS_WAIT; | |
11926 | + else { | |
11927 | + reply.ms_status = STATUS_NEG; | |
11928 | + clear_bit(SEFL_ALLOW_JOIN, | |
11929 | + &sev->se_flags); | |
11930 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
11931 | + } | |
11932 | + } | |
11933 | + | |
11934 | + if (test_bit(SEFL_CANCEL, &sev->se_flags)) { | |
11935 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
11936 | + wake_serviced(DO_JOINLEAVE); | |
11937 | + } | |
11938 | + goto send; | |
11939 | + } | |
11940 | + | |
11941 | + /* no r,u,s event, stick with STATUS_POS */ | |
11942 | + } | |
11943 | + | |
11944 | + send: | |
11945 | + | |
11946 | + if (reply.ms_status == STATUS_POS) { | |
11947 | + node = sm_find_joiner(sg, nodeid); | |
11948 | + if (!node) { | |
11949 | + node = sm_new_node(nodeid); | |
11950 | + list_add_tail(&node->list, &sg->joining); | |
11951 | + } | |
11952 | + } | |
11953 | + | |
11954 | + up(&sm_sglock); | |
11955 | + smsg_bswap_out(&reply); | |
11956 | + send_nodeid_message((char *) &reply, sizeof(reply), nodeid); | |
11957 | +} | |
11958 | + | |
11959 | +/* | |
11960 | + * Another node wants us to stop a service so it can join or leave the SG. We | |
11961 | + * do this by saving the request info in a uevent and having the sm thread do | |
11962 | + * the processing and then replying. | |
11963 | + */ | |
11964 | + | |
11965 | +static void process_stop_request(sm_msg_t * smsg, uint32_t nodeid, | |
11966 | + uint32_t * msgbuf) | |
11967 | +{ | |
11968 | + sm_group_t *sg; | |
11969 | + sm_uevent_t *uev; | |
11970 | + sm_msg_t reply; | |
11971 | + int type = smsg->ms_type; | |
11972 | + | |
11973 | + if (nodeid == sm_our_nodeid) | |
11974 | + goto agree; | |
11975 | + | |
11976 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
11977 | + if (!sg) { | |
11978 | + log_print("process_stop_request: unknown sg id %x", | |
11979 | + smsg->ms_global_sgid); | |
11980 | + return; | |
11981 | + } | |
11982 | + | |
11983 | + /* | |
11984 | + * We shouldn't get here with uevent already set. | |
11985 | + */ | |
11986 | + | |
11987 | + if (test_and_set_bit(SGFL_UEVENT, &sg->flags)) { | |
11988 | + log_error(sg, "process_stop_request: uevent already set"); | |
11989 | + return; | |
11990 | + } | |
11991 | + | |
11992 | + uev = &sg->uevent; | |
11993 | + uev->ue_nodeid = nodeid; | |
11994 | + uev->ue_remote_seid = smsg->ms_sevent_id; | |
11995 | + uev->ue_state = (type == SMSG_JSTOP_REQ) ? UEST_JSTOP : UEST_LSTOP; | |
11996 | + | |
11997 | + if (type == SMSG_JSTOP_REQ) | |
11998 | + uev->ue_num_nodes = be32_to_cpu(*msgbuf); | |
11999 | + else | |
12000 | + set_bit(UEFL_LEAVE, &uev->ue_flags); | |
12001 | + | |
12002 | + /* | |
12003 | + * Do process_join_stop() or process_leave_stop(). | |
12004 | + */ | |
12005 | + | |
12006 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
12007 | + wake_serviced(DO_MEMBERSHIP); | |
12008 | + return; | |
12009 | + | |
12010 | + agree: | |
12011 | + reply.ms_status = STATUS_POS; | |
12012 | + reply.ms_type = | |
12013 | + (type == SMSG_JSTOP_REQ) ? SMSG_JSTOP_REP : SMSG_LSTOP_REP; | |
12014 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12015 | + smsg_bswap_out(&reply); | |
12016 | + send_nodeid_message((char *) &reply, sizeof(reply), nodeid); | |
12017 | +} | |
12018 | + | |
12019 | +static void process_start_request(sm_msg_t * smsg, uint32_t nodeid) | |
12020 | +{ | |
12021 | + sm_group_t *sg; | |
12022 | + sm_uevent_t *uev; | |
12023 | + int type = smsg->ms_type; | |
12024 | + | |
12025 | + if (nodeid == sm_our_nodeid) | |
12026 | + return; | |
12027 | + | |
12028 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
12029 | + if (!sg) { | |
12030 | + log_print("process_start_request: unknown sg id %x", | |
12031 | + smsg->ms_global_sgid); | |
12032 | + return; | |
12033 | + } | |
12034 | + | |
12035 | + if (!test_bit(SGFL_UEVENT, &sg->flags)) { | |
12036 | + log_error(sg, "process_start_request: no uevent"); | |
12037 | + return; | |
12038 | + } | |
12039 | + | |
12040 | + uev = &sg->uevent; | |
12041 | + | |
12042 | + if (type == SMSG_JSTART_CMD) | |
12043 | + uev->ue_state = UEST_JSTART; | |
12044 | + else | |
12045 | + uev->ue_state = UEST_LSTART; | |
12046 | + | |
12047 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
12048 | + wake_serviced(DO_MEMBERSHIP); | |
12049 | +} | |
12050 | + | |
12051 | +static void process_leave_request(sm_msg_t * smsg, uint32_t nodeid) | |
12052 | +{ | |
12053 | + sm_group_t *sg; | |
12054 | + sm_node_t *node; | |
12055 | + sm_msg_t reply; | |
12056 | + sm_sevent_t *sev; | |
12057 | + int found = FALSE; | |
12058 | + | |
12059 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
12060 | + if (sg) { | |
12061 | + if (nodeid == sm_our_nodeid) | |
12062 | + found = TRUE; | |
12063 | + else { | |
12064 | + list_for_each_entry(node, &sg->memb, list) { | |
12065 | + if (node->id != nodeid) | |
12066 | + continue; | |
12067 | + set_bit(SNFL_LEAVING, &node->flags); | |
12068 | + found = TRUE; | |
12069 | + break; | |
12070 | + } | |
12071 | + } | |
12072 | + } | |
12073 | + | |
12074 | + if (!found) { | |
12075 | + reply.ms_type = SMSG_LEAVE_REP; | |
12076 | + reply.ms_status = STATUS_NEG; | |
12077 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12078 | + } else { | |
12079 | + reply.ms_type = SMSG_LEAVE_REP; | |
12080 | + reply.ms_status = STATUS_POS; | |
12081 | + reply.ms_sevent_id = smsg->ms_sevent_id; | |
12082 | + | |
12083 | + if (sg->state == SGST_RECOVER) | |
12084 | + reply.ms_status = STATUS_WAIT; | |
12085 | + | |
12086 | + else if (test_bit(SGFL_SEVENT, &sg->flags) && | |
12087 | + nodeid != sm_our_nodeid) { | |
12088 | + sev = sg->sevent; | |
12089 | + | |
12090 | + /* | |
12091 | + * We're trying to join or leave at the moment. If | |
12092 | + * we're past JOIN/LEAVE_ACKWAIT, we make the requestor | |
12093 | + * wait. Otherwise, if joining we'll cancel to let the | |
12094 | + * leave happen first, or if we're leaving allow the | |
12095 | + * lower nodeid to leave first. | |
12096 | + */ | |
12097 | + | |
12098 | + if (test_bit(SEFL_LEAVE, &sev->se_flags)) { | |
12099 | + if (sev->se_state > SEST_LEAVE_ACKWAIT) | |
12100 | + reply.ms_status = STATUS_WAIT; | |
12101 | + else if (sm_our_nodeid < nodeid) | |
12102 | + reply.ms_status = STATUS_WAIT; | |
12103 | + else { | |
12104 | + reply.ms_status = STATUS_POS; | |
12105 | + clear_bit(SEFL_ALLOW_LEAVE, | |
12106 | + &sev->se_flags); | |
12107 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
12108 | + } | |
12109 | + } else { | |
12110 | + if (sev->se_state > SEST_JOIN_ACKWAIT) | |
12111 | + reply.ms_status = STATUS_WAIT; | |
12112 | + else { | |
12113 | + reply.ms_status = STATUS_NEG; | |
12114 | + clear_bit(SEFL_ALLOW_JOIN, | |
12115 | + &sev->se_flags); | |
12116 | + set_bit(SEFL_CANCEL, &sev->se_flags); | |
12117 | + } | |
12118 | + } | |
12119 | + | |
12120 | + if (test_bit(SEFL_CANCEL, &sev->se_flags)) { | |
12121 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
12122 | + wake_serviced(DO_JOINLEAVE); | |
12123 | + } | |
12124 | + } | |
12125 | + | |
12126 | + else if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
12127 | + if (sg->uevent.ue_nodeid != nodeid) | |
12128 | + reply.ms_status = STATUS_WAIT; | |
12129 | + } | |
12130 | + | |
12131 | + } | |
12132 | + | |
12133 | + smsg_bswap_out(&reply); | |
12134 | + send_nodeid_message((char *) &reply, sizeof(reply), nodeid); | |
12135 | +} | |
12136 | + | |
12137 | +/* | |
12138 | + * Each remaining node will send us a done message. We quit when we get the | |
12139 | + * first. The subsequent done messages for the finished sevent get here and | |
12140 | + * are ignored. | |
12141 | + */ | |
12142 | + | |
12143 | +static void process_lstart_done(sm_msg_t *smsg, uint32_t nodeid) | |
12144 | +{ | |
12145 | + sm_sevent_t *sev; | |
12146 | + | |
12147 | + sev = find_sevent(smsg->ms_sevent_id); | |
12148 | + if (!sev) | |
12149 | + return; | |
12150 | + | |
12151 | + if (sev->se_state != SEST_LSTART_WAITREMOTE) | |
12152 | + return; | |
12153 | + | |
12154 | + sev->se_state = SEST_LSTART_REMOTEDONE; | |
12155 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
12156 | + wake_serviced(DO_JOINLEAVE); | |
12157 | +} | |
12158 | + | |
12159 | +/* | |
12160 | + * This function and everything it calls always runs in sm context. | |
12161 | + */ | |
12162 | + | |
12163 | +static void process_message(char *msg, uint32_t nodeid) | |
12164 | +{ | |
12165 | + sm_msg_t smsg; | |
12166 | + | |
12167 | + smsg_copy_in(msg, &smsg); | |
12168 | + | |
12169 | + switch (smsg.ms_type) { | |
12170 | + case SMSG_JOIN_REQ: | |
12171 | + process_join_request(&smsg, nodeid, msg + sizeof(sm_msg_t)); | |
12172 | + break; | |
12173 | + | |
12174 | + case SMSG_JSTOP_REQ: | |
12175 | + process_stop_request(&smsg, nodeid, | |
12176 | + (uint32_t *) (msg + sizeof(sm_msg_t))); | |
12177 | + break; | |
12178 | + | |
12179 | + case SMSG_LEAVE_REQ: | |
12180 | + process_leave_request(&smsg, nodeid); | |
12181 | + break; | |
12182 | + | |
12183 | + case SMSG_LSTOP_REQ: | |
12184 | + process_stop_request(&smsg, nodeid, NULL); | |
12185 | + break; | |
12186 | + | |
12187 | + case SMSG_JSTART_CMD: | |
12188 | + case SMSG_LSTART_CMD: | |
12189 | + process_start_request(&smsg, nodeid); | |
12190 | + break; | |
12191 | + | |
12192 | + case SMSG_LSTART_DONE: | |
12193 | + process_lstart_done(&smsg, nodeid); | |
12194 | + break; | |
12195 | + | |
12196 | + case SMSG_JOIN_REP: | |
12197 | + case SMSG_JSTOP_REP: | |
12198 | + case SMSG_LEAVE_REP: | |
12199 | + case SMSG_LSTOP_REP: | |
12200 | + process_reply(&smsg, nodeid); | |
12201 | + break; | |
12202 | + | |
12203 | + case SMSG_RECOVER: | |
12204 | + process_recover_msg(&smsg, nodeid); | |
12205 | + break; | |
12206 | + | |
12207 | + default: | |
12208 | + log_print("process_message: unknown type %u nodeid %u", | |
12209 | + smsg.ms_type, nodeid); | |
12210 | + } | |
12211 | +} | |
12212 | + | |
12213 | +/* | |
12214 | + * Always called from sm context. | |
12215 | + */ | |
12216 | + | |
12217 | +void process_messages(void) | |
12218 | +{ | |
12219 | + rq_entry_t *re; | |
12220 | + | |
12221 | + while (1) { | |
12222 | + re = NULL; | |
12223 | + | |
12224 | + spin_lock(&message_lock); | |
12225 | + if (!list_empty(&messages)) { | |
12226 | + re = list_entry(messages.next, rq_entry_t, list); | |
12227 | + list_del(&re->list); | |
12228 | + } | |
12229 | + spin_unlock(&message_lock); | |
12230 | + | |
12231 | + if (!re) | |
12232 | + break; | |
12233 | + process_message(re->msg, re->nodeid); | |
12234 | + kfree(re->msg); | |
12235 | + kfree(re); | |
12236 | + schedule(); | |
12237 | + } | |
12238 | +} | |
12239 | + | |
12240 | +/* | |
12241 | + * Context: cnxman and sm | |
12242 | + */ | |
12243 | + | |
12244 | +static int add_to_recvqueue(char *msg, int len, uint32_t nodeid) | |
12245 | +{ | |
12246 | + rq_entry_t *re; | |
12247 | + | |
12248 | + SM_RETRY(re = (rq_entry_t *) kmalloc(sizeof(rq_entry_t), GFP_KERNEL), | |
12249 | + re); | |
12250 | + SM_RETRY(re->msg = (char *) kmalloc(len, GFP_KERNEL), re->msg); | |
12251 | + | |
12252 | + memcpy(re->msg, msg, len); | |
12253 | + re->len = len; | |
12254 | + re->nodeid = nodeid; | |
12255 | + | |
12256 | + spin_lock(&message_lock); | |
12257 | + list_add_tail(&re->list, &messages); | |
12258 | + spin_unlock(&message_lock); | |
12259 | + | |
12260 | + wake_serviced(DO_MESSAGES); | |
12261 | + return 0; | |
12262 | +} | |
12263 | + | |
12264 | +/* | |
12265 | + * Context: cnxman | |
12266 | + * Called by cnxman when a service manager message arrives. | |
12267 | + */ | |
12268 | + | |
12269 | +int sm_cluster_message(char *msg, int len, char *addr, int addr_len, | |
12270 | + unsigned int node_id) | |
12271 | +{ | |
12272 | + struct kcl_cluster_node kclnode; | |
12273 | + uint32_t nodeid = 0; | |
12274 | + int error = 0; | |
12275 | + | |
12276 | + if (!node_id) { | |
12277 | + error = kcl_get_node_by_addr(addr, addr_len, &kclnode); | |
12278 | + if (error) | |
12279 | + return error; | |
12280 | + nodeid = kclnode.node_id; | |
12281 | + } else | |
12282 | + nodeid = node_id; | |
12283 | + | |
12284 | + return add_to_recvqueue(msg, len, nodeid); | |
12285 | +} | |
12286 | + | |
12287 | +/* | |
12288 | + * These send routines are used by sm and are always called from sm context. | |
12289 | + */ | |
12290 | + | |
12291 | +int send_nodeid_message(char *msg, int len, uint32_t nodeid) | |
12292 | +{ | |
12293 | + int error = 0; | |
12294 | + struct sockaddr_cl saddr; | |
12295 | + | |
12296 | + if (nodeid == sm_our_nodeid) { | |
12297 | + add_to_recvqueue(msg, len, nodeid); | |
12298 | + goto out; | |
12299 | + } | |
12300 | + | |
12301 | + saddr.scl_family = AF_CLUSTER; | |
12302 | + saddr.scl_port = CLUSTER_PORT_SERVICES; | |
12303 | + saddr.scl_nodeid = nodeid; | |
12304 | + error = kcl_sendmsg(sm_socket, msg, len, &saddr, | |
12305 | + sizeof(saddr), 0); | |
12306 | + if (error > 0) | |
12307 | + error = 0; | |
12308 | + | |
12309 | + if (error) | |
12310 | + log_print("send_nodeid_message error %d to %u", error, nodeid); | |
12311 | + out: | |
12312 | + return error; | |
12313 | +} | |
12314 | + | |
12315 | +int send_broadcast_message(char *msg, int len) | |
12316 | +{ | |
12317 | + int error; | |
12318 | + | |
12319 | + error = kcl_sendmsg(sm_socket, msg, len, NULL, 0, 0); | |
12320 | + if (error > 0) | |
12321 | + error = 0; | |
12322 | + | |
12323 | + add_to_recvqueue(msg, len, sm_our_nodeid); | |
12324 | + | |
12325 | + if (error) | |
12326 | + log_print("send_broadcast_message error %d", error); | |
12327 | + | |
12328 | + return error; | |
12329 | +} | |
12330 | + | |
12331 | +int send_members_message(sm_group_t *sg, char *msg, int len) | |
12332 | +{ | |
12333 | + sm_node_t *node; | |
12334 | + int error = 0; | |
12335 | + | |
12336 | + list_for_each_entry(node, &sg->memb, list) { | |
12337 | + error = send_nodeid_message(msg, len, node->id); | |
12338 | + if (error < 0) | |
12339 | + break; | |
12340 | + } | |
12341 | + return error; | |
12342 | +} | |
12343 | + | |
12344 | +int send_members_message_sev(sm_group_t *sg, char *msg, int len, | |
12345 | + sm_sevent_t * sev) | |
12346 | +{ | |
12347 | + int error; | |
12348 | + sm_msg_t *smsg = (sm_msg_t *) msg; | |
12349 | + | |
12350 | + set_allowed_msgtype(sev, smsg->ms_type); | |
12351 | + sev->se_reply_count = 0; | |
12352 | + | |
12353 | + error = send_members_message(sg, msg, len); | |
12354 | + if (error < 0) | |
12355 | + clear_allowed_msgtype(sev, smsg->ms_type); | |
12356 | + | |
12357 | + return error; | |
12358 | +} | |
12359 | + | |
12360 | +int send_broadcast_message_sev(char *msg, int len, sm_sevent_t * sev) | |
12361 | +{ | |
12362 | + int error; | |
12363 | + sm_msg_t *smsg = (sm_msg_t *) msg; | |
12364 | + | |
12365 | + set_allowed_msgtype(sev, smsg->ms_type); | |
12366 | + sev->se_reply_count = 0; | |
12367 | + | |
12368 | + error = send_broadcast_message(msg, len); | |
12369 | + if (error < 0) | |
12370 | + clear_allowed_msgtype(sev, smsg->ms_type); | |
12371 | + | |
12372 | + return error; | |
12373 | +} | |
12374 | diff -urN linux-orig/cluster/cman/sm_message.h linux-patched/cluster/cman/sm_message.h | |
12375 | --- linux-orig/cluster/cman/sm_message.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 12376 | +++ linux-patched/cluster/cman/sm_message.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 12377 | @@ -0,0 +1,34 @@ |
12378 | +/****************************************************************************** | |
12379 | +******************************************************************************* | |
12380 | +** | |
12381 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
12382 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
12383 | +** | |
12384 | +** This copyrighted material is made available to anyone wishing to use, | |
12385 | +** modify, copy, or redistribute it subject to the terms and conditions | |
12386 | +** of the GNU General Public License v.2. | |
12387 | +** | |
12388 | +******************************************************************************* | |
12389 | +******************************************************************************/ | |
12390 | + | |
12391 | +#ifndef __SM_MESSAGE_DOT_H__ | |
12392 | +#define __SM_MESSAGE_DOT_H__ | |
12393 | + | |
12394 | +void init_messages(void); | |
12395 | +uint32_t sm_new_global_id(int level); | |
12396 | +void smsg_bswap_out(sm_msg_t * smsg); | |
12397 | +char *create_smsg(sm_group_t *sg, int type, int datalen, int *msglen, | |
12398 | + sm_sevent_t *sev); | |
12399 | +void process_messages(void); | |
12400 | +int sm_cluster_message(char *msg, int len, char *addr, int addr_len, | |
12401 | + unsigned int node_id); | |
12402 | +int send_nodeid_message(char *msg, int len, uint32_t nodeid); | |
12403 | +int send_broadcast_message(char *msg, int len); | |
12404 | +int send_broadcast_message_sev(char *msg, int len, sm_sevent_t * sev); | |
12405 | +int send_members_message(sm_group_t *sg, char *msg, int len); | |
12406 | +int send_members_message_sev(sm_group_t *sg, char *msg, int len, | |
12407 | + sm_sevent_t * sev); | |
12408 | +int sm_cluster_message(char *msg, int len, char *addr, int addr_len, | |
12409 | + unsigned int node_id); | |
12410 | + | |
12411 | +#endif | |
12412 | diff -urN linux-orig/cluster/cman/sm_misc.c linux-patched/cluster/cman/sm_misc.c | |
12413 | --- linux-orig/cluster/cman/sm_misc.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 12414 | +++ linux-patched/cluster/cman/sm_misc.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 12415 | @@ -0,0 +1,369 @@ |
12416 | +/****************************************************************************** | |
12417 | +******************************************************************************* | |
12418 | +** | |
12419 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
12420 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
12421 | +** | |
12422 | +** This copyrighted material is made available to anyone wishing to use, | |
12423 | +** modify, copy, or redistribute it subject to the terms and conditions | |
12424 | +** of the GNU General Public License v.2. | |
12425 | +** | |
12426 | +******************************************************************************* | |
12427 | +******************************************************************************/ | |
12428 | + | |
12429 | +#include "sm.h" | |
12430 | +#include "config.h" | |
12431 | + | |
12432 | +#define MAX_DEBUG_MSG_LEN (40) | |
12433 | + | |
12434 | +extern struct list_head sm_members; | |
12435 | +static uint32_t local_ids; | |
12436 | +static uint32_t event_id; | |
12437 | +static spinlock_t event_id_lock; | |
12438 | +static char * debug_buf; | |
12439 | +static unsigned int debug_size; | |
12440 | +static unsigned int debug_point; | |
12441 | +static int debug_wrap; | |
12442 | +static spinlock_t debug_lock; | |
12443 | + | |
12444 | + | |
12445 | +void init_sm_misc(void) | |
12446 | +{ | |
12447 | + local_ids = 1; | |
12448 | + event_id = 1; | |
12449 | + spin_lock_init(&event_id_lock); | |
12450 | + debug_buf = NULL; | |
12451 | + debug_size = 0; | |
12452 | + debug_point = 0; | |
12453 | + debug_wrap = 0; | |
12454 | + spin_lock_init(&debug_lock); | |
12455 | + | |
12456 | + sm_debug_setup(cman_config.sm_debug_size); | |
12457 | +} | |
12458 | + | |
12459 | +sm_node_t *sm_new_node(uint32_t nodeid) | |
12460 | +{ | |
12461 | + struct kcl_cluster_node kclnode; | |
12462 | + sm_node_t *node; | |
12463 | + int error; | |
12464 | + | |
12465 | + error = kcl_get_node_by_nodeid(nodeid, &kclnode); | |
12466 | + SM_ASSERT(!error,); | |
12467 | + | |
12468 | + SM_RETRY(node = (sm_node_t *) kmalloc(sizeof(sm_node_t), GFP_KERNEL), | |
12469 | + node); | |
12470 | + | |
12471 | + memset(node, 0, sizeof(sm_node_t)); | |
12472 | + node->id = nodeid; | |
12473 | + node->incarnation = kclnode.incarnation; | |
12474 | + return node; | |
12475 | +} | |
12476 | + | |
12477 | +sm_node_t *sm_find_joiner(sm_group_t *sg, uint32_t nodeid) | |
12478 | +{ | |
12479 | + sm_node_t *node; | |
12480 | + | |
12481 | + list_for_each_entry(node, &sg->joining, list) { | |
12482 | + if (node->id == nodeid) | |
12483 | + return node; | |
12484 | + } | |
12485 | + return NULL; | |
12486 | +} | |
12487 | + | |
12488 | +sm_node_t *sm_find_member(uint32_t nodeid) | |
12489 | +{ | |
12490 | + sm_node_t *node; | |
12491 | + | |
12492 | + list_for_each_entry(node, &sm_members, list) { | |
12493 | + if (node->id == nodeid) | |
12494 | + return node; | |
12495 | + } | |
12496 | + return NULL; | |
12497 | +} | |
12498 | + | |
12499 | +uint32_t sm_new_local_id(int level) | |
12500 | +{ | |
12501 | + uint32_t id = local_ids++; | |
12502 | + uint8_t l = (uint8_t) level; | |
12503 | + | |
12504 | + if (level > 0xFF) | |
12505 | + return 0; | |
12506 | + | |
12507 | + if (id > 0x00FFFFFF) | |
12508 | + return 0; | |
12509 | + | |
12510 | + id |= (l << 24); | |
12511 | + return id; | |
12512 | +} | |
12513 | + | |
12514 | +int sm_id_to_level(uint32_t id) | |
12515 | +{ | |
12516 | + uint8_t l = (id & 0xFF000000) >> 24; | |
12517 | + | |
12518 | + return (int) l; | |
12519 | +} | |
12520 | + | |
12521 | +void sm_set_event_id(int *id) | |
12522 | +{ | |
12523 | + spin_lock(&event_id_lock); | |
12524 | + *id = event_id++; | |
12525 | + spin_unlock(&event_id_lock); | |
12526 | +} | |
12527 | + | |
12528 | +sm_group_t *sm_local_id_to_sg(int id) | |
12529 | +{ | |
12530 | + sm_group_t *sg; | |
12531 | + int level = sm_id_to_level(id); | |
12532 | + int found = FALSE; | |
12533 | + | |
12534 | + down(&sm_sglock); | |
12535 | + | |
12536 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
12537 | + if (sg->local_id == id) { | |
12538 | + found = TRUE; | |
12539 | + break; | |
12540 | + } | |
12541 | + } | |
12542 | + up(&sm_sglock); | |
12543 | + if (!found) | |
12544 | + sg = NULL; | |
12545 | + return sg; | |
12546 | +} | |
12547 | + | |
12548 | +sm_group_t *sm_global_id_to_sg(int id) | |
12549 | +{ | |
12550 | + sm_group_t *sg; | |
12551 | + int level = sm_id_to_level(id); | |
12552 | + int found = FALSE; | |
12553 | + | |
12554 | + down(&sm_sglock); | |
12555 | + | |
12556 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
12557 | + if (sg->global_id == id) { | |
12558 | + found = TRUE; | |
12559 | + break; | |
12560 | + } | |
12561 | + } | |
12562 | + up(&sm_sglock); | |
12563 | + if (!found) | |
12564 | + sg = NULL; | |
12565 | + return sg; | |
12566 | +} | |
12567 | + | |
12568 | +void sm_debug_log(sm_group_t *sg, const char *fmt, ...) | |
12569 | +{ | |
12570 | + va_list va; | |
12571 | + int i, n, size, len; | |
12572 | + char buf[MAX_DEBUG_MSG_LEN+1]; | |
12573 | + | |
12574 | + spin_lock(&debug_lock); | |
12575 | + | |
12576 | + if (!debug_buf) | |
12577 | + goto out; | |
12578 | + | |
12579 | + size = MAX_DEBUG_MSG_LEN; | |
12580 | + memset(buf, 0, size+1); | |
12581 | + | |
12582 | + n = snprintf(buf, size, "%08x ", sg->global_id); | |
12583 | + size -= n; | |
12584 | + | |
12585 | + va_start(va, fmt); | |
12586 | + vsnprintf(buf+n, size, fmt, va); | |
12587 | + va_end(va); | |
12588 | + | |
12589 | + len = strlen(buf); | |
12590 | + if (len > MAX_DEBUG_MSG_LEN-1) | |
12591 | + len = MAX_DEBUG_MSG_LEN-1; | |
12592 | + buf[len] = '\n'; | |
12593 | + buf[len+1] = '\0'; | |
12594 | + | |
12595 | + for (i = 0; i < strlen(buf); i++) { | |
12596 | + debug_buf[debug_point++] = buf[i]; | |
12597 | + | |
12598 | + if (debug_point == debug_size) { | |
12599 | + debug_point = 0; | |
12600 | + debug_wrap = 1; | |
12601 | + } | |
12602 | + } | |
12603 | + out: | |
12604 | + spin_unlock(&debug_lock); | |
12605 | +} | |
12606 | + | |
12607 | +void sm_debug_setup(int size) | |
12608 | +{ | |
12609 | + char *b = kmalloc(size, GFP_KERNEL); | |
12610 | + | |
12611 | + spin_lock(&debug_lock); | |
12612 | + if (debug_buf) | |
12613 | + kfree(debug_buf); | |
12614 | + | |
12615 | + if (size > PAGE_SIZE) | |
12616 | + size = PAGE_SIZE; | |
12617 | + debug_size = size; | |
12618 | + debug_point = 0; | |
12619 | + debug_wrap = 0; | |
12620 | + debug_buf = b; | |
12621 | + memset(debug_buf, 0, debug_size); | |
12622 | + spin_unlock(&debug_lock); | |
12623 | +} | |
12624 | + | |
12625 | +#ifdef CONFIG_PROC_FS | |
12626 | + | |
12627 | +int sm_debug_info(char *b, char **start, off_t offset, int length) | |
12628 | +{ | |
12629 | + int i, n = 0; | |
12630 | + | |
12631 | + spin_lock(&debug_lock); | |
12632 | + | |
12633 | + if (debug_wrap) { | |
12634 | + for (i = debug_point; i < debug_size; i++) | |
12635 | + n += sprintf(b + n, "%c", debug_buf[i]); | |
12636 | + } | |
12637 | + for (i = 0; i < debug_point; i++) | |
12638 | + n += sprintf(b + n, "%c", debug_buf[i]); | |
12639 | + | |
12640 | + spin_unlock(&debug_lock); | |
12641 | + | |
12642 | + return n; | |
12643 | +} | |
12644 | + | |
12645 | +int sm_procdata(char *b, char **start, off_t offset, int length) | |
12646 | +{ | |
12647 | + sm_group_t *sg; | |
12648 | + sm_node_t *node; | |
12649 | + int n = 0, level, i; | |
12650 | + | |
12651 | + n += sprintf(b + n, "\n"); | |
12652 | + | |
12653 | + /* | |
12654 | + * Header | |
12655 | + */ | |
12656 | + | |
12657 | + n += sprintf(b + n, | |
12658 | + "Service Name GID LID State Code\n"); | |
12659 | + | |
12660 | + down(&sm_sglock); | |
12661 | + | |
12662 | + for (level = 0; level < SG_LEVELS; level++) { | |
12663 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
12664 | + | |
12665 | + /* | |
12666 | + * Cluster Service | |
12667 | + */ | |
12668 | + | |
12669 | + switch (level) { | |
12670 | + case SERVICE_LEVEL_FENCE: | |
12671 | + n += sprintf(b + n, "Fence Domain: "); | |
12672 | + break; | |
12673 | + case SERVICE_LEVEL_GDLM: | |
12674 | + n += sprintf(b + n, "DLM Lock Space: "); | |
12675 | + break; | |
12676 | + case SERVICE_LEVEL_GFS: | |
12677 | + n += sprintf(b + n, "GFS Mount Group: "); | |
12678 | + break; | |
12679 | + case SERVICE_LEVEL_USER: | |
12680 | + n += sprintf(b + n, "User: "); | |
12681 | + break; | |
12682 | + } | |
12683 | + | |
12684 | + /* | |
12685 | + * Name | |
12686 | + */ | |
12687 | + | |
12688 | + n += sprintf(b + n, "\""); | |
12689 | + for (i = 0; i < sg->namelen; i++) | |
12690 | + n += sprintf(b + n, "%c", sg->name[i]); | |
12691 | + n += sprintf(b + n, "\""); | |
12692 | + | |
12693 | + for (; i < MAX_SERVICE_NAME_LEN-1; i++) | |
12694 | + n += sprintf(b + n, " "); | |
12695 | + | |
12696 | + /* | |
12697 | + * GID LID (sans level from top byte) | |
12698 | + */ | |
12699 | + | |
12700 | + n += sprintf(b + n, "%3u %3u ", | |
12701 | + (sg->global_id & 0x00FFFFFF), | |
12702 | + (sg->local_id & 0x00FFFFFF)); | |
12703 | + | |
12704 | + /* | |
12705 | + * State | |
12706 | + */ | |
12707 | + | |
12708 | + switch (sg->state) { | |
12709 | + case SGST_NONE: | |
12710 | + n += sprintf(b + n, "none "); | |
12711 | + break; | |
12712 | + case SGST_JOIN: | |
12713 | + n += sprintf(b + n, "join "); | |
12714 | + break; | |
12715 | + case SGST_RUN: | |
12716 | + n += sprintf(b + n, "run "); | |
12717 | + break; | |
12718 | + case SGST_RECOVER: | |
12719 | + n += sprintf(b + n, "recover %u ", | |
12720 | + sg->recover_state); | |
12721 | + break; | |
12722 | + case SGST_UEVENT: | |
12723 | + n += sprintf(b + n, "update "); | |
12724 | + break; | |
12725 | + } | |
12726 | + | |
12727 | + /* | |
12728 | + * Code | |
12729 | + */ | |
12730 | + | |
12731 | + if (test_bit(SGFL_SEVENT, &sg->flags)) | |
12732 | + n += sprintf(b + n, "S"); | |
12733 | + if (test_bit(SGFL_UEVENT, &sg->flags)) | |
12734 | + n += sprintf(b + n, "U"); | |
12735 | + if (test_bit(SGFL_NEED_RECOVERY, &sg->flags)) | |
12736 | + n += sprintf(b + n, "N"); | |
12737 | + | |
12738 | + n += sprintf(b + n, "-"); | |
12739 | + | |
12740 | + if (test_bit(SGFL_SEVENT, &sg->flags) | |
12741 | + && sg->sevent) { | |
12742 | + n += sprintf(b + n, "%u,%lx,%u", | |
12743 | + sg->sevent->se_state, | |
12744 | + sg->sevent->se_flags, | |
12745 | + sg->sevent->se_reply_count); | |
12746 | + } | |
12747 | + | |
12748 | + if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
12749 | + n += sprintf(b + n, "%u,%lx,%u", | |
12750 | + sg->uevent.ue_state, | |
12751 | + sg->uevent.ue_flags, | |
12752 | + sg->uevent.ue_nodeid); | |
12753 | + } | |
12754 | + | |
12755 | + n += sprintf(b + n, "\n"); | |
12756 | + | |
12757 | + /* | |
12758 | + * node list | |
12759 | + */ | |
12760 | + | |
12761 | + i = 0; | |
12762 | + | |
12763 | + n += sprintf(b + n, "["); | |
12764 | + | |
12765 | + list_for_each_entry(node, &sg->memb, list) { | |
12766 | + if (i && !(i % 24)) | |
12767 | + n += sprintf(b + n, "\n"); | |
12768 | + | |
12769 | + if (i) | |
12770 | + n += sprintf(b + n, " "); | |
12771 | + | |
12772 | + n += sprintf(b + n, "%u", node->id); | |
12773 | + i++; | |
12774 | + } | |
12775 | + | |
12776 | + n += sprintf(b + n, "]\n\n"); | |
12777 | + } | |
12778 | + } | |
12779 | + | |
12780 | + up(&sm_sglock); | |
12781 | + | |
12782 | + return n; | |
12783 | +} | |
12784 | +#endif | |
12785 | diff -urN linux-orig/cluster/cman/sm_misc.h linux-patched/cluster/cman/sm_misc.h | |
12786 | --- linux-orig/cluster/cman/sm_misc.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 12787 | +++ linux-patched/cluster/cman/sm_misc.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 12788 | @@ -0,0 +1,29 @@ |
12789 | +/****************************************************************************** | |
12790 | +******************************************************************************* | |
12791 | +** | |
12792 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
12793 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
12794 | +** | |
12795 | +** This copyrighted material is made available to anyone wishing to use, | |
12796 | +** modify, copy, or redistribute it subject to the terms and conditions | |
12797 | +** of the GNU General Public License v.2. | |
12798 | +** | |
12799 | +******************************************************************************* | |
12800 | +******************************************************************************/ | |
12801 | + | |
12802 | +#ifndef __SM_MISC_DOT_H__ | |
12803 | +#define __SM_MISC_DOT_H__ | |
12804 | + | |
12805 | +void init_sm_misc(void); | |
12806 | +sm_node_t *sm_new_node(uint32_t nodeid); | |
12807 | +sm_node_t *sm_find_joiner(sm_group_t *sg, uint32_t nodeid); | |
12808 | +sm_node_t *sm_find_member(uint32_t nodeid); | |
12809 | +uint32_t sm_new_local_id(int level); | |
12810 | +int sm_id_to_level(uint32_t id); | |
12811 | +void sm_set_event_id(int *id); | |
12812 | +sm_group_t *sm_local_id_to_sg(int id); | |
12813 | +sm_group_t *sm_global_id_to_sg(int id); | |
12814 | +void sm_debug_log(sm_group_t *sg, const char *fmt, ...); | |
12815 | +void sm_debug_setup(int size); | |
12816 | + | |
12817 | +#endif | |
12818 | diff -urN linux-orig/cluster/cman/sm_recover.c linux-patched/cluster/cman/sm_recover.c | |
12819 | --- linux-orig/cluster/cman/sm_recover.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 12820 | +++ linux-patched/cluster/cman/sm_recover.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 12821 | @@ -0,0 +1,522 @@ |
12822 | +/****************************************************************************** | |
12823 | +******************************************************************************* | |
12824 | +** | |
12825 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
12826 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
12827 | +** | |
12828 | +** This copyrighted material is made available to anyone wishing to use, | |
12829 | +** modify, copy, or redistribute it subject to the terms and conditions | |
12830 | +** of the GNU General Public License v.2. | |
12831 | +** | |
12832 | +******************************************************************************* | |
12833 | +******************************************************************************/ | |
12834 | + | |
12835 | +#include "sm.h" | |
12836 | +#include "config.h" | |
12837 | + | |
12838 | +/* | |
12839 | + * A collection of sg's which need to be recovered due to a failed member. | |
12840 | + * These sg's are recovered in order of level. An sg subject to cascading | |
12841 | + * failures is moved from one of these structs to a newer one. | |
12842 | + */ | |
12843 | + | |
12844 | +struct recover { | |
12845 | + struct list_head list; /* list of current re's */ | |
12846 | + struct list_head sgs[SG_LEVELS]; /* lists of sg's by level */ | |
12847 | + int event_id; /* event id */ | |
12848 | + int cur_level; | |
12849 | +}; | |
12850 | +typedef struct recover recover_t; | |
12851 | + | |
12852 | + | |
12853 | +extern uint32_t * sm_new_nodeids; | |
12854 | +extern int sm_quorum, sm_quorum_next; | |
12855 | +extern uint32_t sm_our_nodeid; | |
12856 | +extern struct list_head sm_members; | |
12857 | +extern int sm_member_count; | |
12858 | +static struct list_head recoveries; | |
12859 | + | |
12860 | + | |
12861 | +void init_recovery(void) | |
12862 | +{ | |
12863 | + INIT_LIST_HEAD(&recoveries); | |
12864 | +} | |
12865 | + | |
12866 | +/* | |
12867 | + * This is the first thing called when a change is announced in cluster | |
12868 | + * membership. Nodes are marked as being a CLUSTER_MEMBER or not. SM adds new | |
12869 | + * nodes to its sm_members list which it's not seen before. Nodes which were | |
12870 | + * alive but are now gone are marked as "need recovery". | |
12871 | + * | |
12872 | + * The "need recovery" status of nodes is propagated to the node's SG's in | |
12873 | + * mark_effected_sgs. The effected SG's are themselves marked as needing | |
12874 | + * recovery and in new_recovery the dead nodes are removed from the SG's | |
12875 | + * individual member lists. The "need recovery" status of nodes is cleared in | |
12876 | + * adjust_members_done(). | |
12877 | + */ | |
12878 | + | |
12879 | +static int adjust_members(void) | |
12880 | +{ | |
12881 | + sm_node_t *node; | |
12882 | + struct kcl_cluster_node knode; | |
12883 | + int i, error, num_nodes, sub = 0, add = 0, found; | |
12884 | + | |
12885 | + /* | |
12886 | + * Get list of current members from cnxman | |
12887 | + */ | |
12888 | + | |
12889 | + memset(sm_new_nodeids, 0, cman_config.max_nodes * sizeof(uint32_t)); | |
12890 | + num_nodes = kcl_get_member_ids(sm_new_nodeids, cman_config.max_nodes); | |
12891 | + | |
12892 | + /* | |
12893 | + * Determine who's gone | |
12894 | + */ | |
12895 | + | |
12896 | + list_for_each_entry(node, &sm_members, list) { | |
12897 | + found = FALSE; | |
12898 | + for (i = 0; i < num_nodes; i++) { | |
12899 | + if (node->id == sm_new_nodeids[i]) { | |
12900 | + found = TRUE; | |
12901 | + sm_new_nodeids[i] = 0; | |
12902 | + break; | |
12903 | + } | |
12904 | + } | |
12905 | + | |
12906 | + if (found) { | |
12907 | + error = kcl_get_node_by_nodeid(node->id, &knode); | |
12908 | + SM_ASSERT(!error, printk("error=%d\n", error);); | |
12909 | + | |
12910 | + if (!test_bit(SNFL_CLUSTER_MEMBER, &node->flags)) { | |
12911 | + /* former member is back */ | |
12912 | + set_bit(SNFL_CLUSTER_MEMBER, &node->flags); | |
12913 | + node->incarnation = knode.incarnation; | |
12914 | + add++; | |
12915 | + } else { | |
12916 | + /* current member is still alive - if the | |
12917 | + * incarnation number is different it died and | |
12918 | + * returned between checks */ | |
12919 | + if (node->incarnation != knode.incarnation) { | |
12920 | + set_bit(SNFL_NEED_RECOVERY, | |
12921 | + &node->flags); | |
12922 | + node->incarnation = knode.incarnation; | |
12923 | + sub++; | |
12924 | + } | |
12925 | + } | |
12926 | + } else { | |
12927 | + /* current member has died */ | |
12928 | + if (test_and_clear_bit(SNFL_CLUSTER_MEMBER, | |
12929 | + &node->flags)) { | |
12930 | + set_bit(SNFL_NEED_RECOVERY, &node->flags); | |
12931 | + sub++; | |
12932 | + } | |
12933 | + } | |
12934 | + } | |
12935 | + | |
12936 | + /* | |
12937 | + * Look for new nodes | |
12938 | + */ | |
12939 | + | |
12940 | + for (i = 0; i < num_nodes; i++) { | |
12941 | + if (sm_new_nodeids[i]) { | |
12942 | + node = sm_new_node(sm_new_nodeids[i]); | |
12943 | + set_bit(SNFL_CLUSTER_MEMBER, &node->flags); | |
12944 | + add++; | |
12945 | + list_add_tail(&node->list, &sm_members); | |
12946 | + sm_member_count++; | |
12947 | + } | |
12948 | + } | |
12949 | + | |
12950 | + /* | |
12951 | + * Get our own nodeid | |
12952 | + */ | |
12953 | + | |
12954 | + if (!sm_our_nodeid) { | |
12955 | + list_for_each_entry(node, &sm_members, list) { | |
12956 | + error = kcl_get_node_by_nodeid(node->id, &knode); | |
12957 | + SM_ASSERT(!error, printk("error=%d\n", error);); | |
12958 | + | |
12959 | + if (knode.us) { | |
12960 | + sm_our_nodeid = knode.node_id; | |
12961 | + break; | |
12962 | + } | |
12963 | + } | |
12964 | + } | |
12965 | + | |
12966 | + return sub; | |
12967 | +} | |
12968 | + | |
12969 | +/* | |
12970 | + * Given some number of dead nodes, flag SG's the dead nodes were part of. | |
12971 | + * This requires a number of loops because each node structure does not keep a | |
12972 | + * list of SG's it's in. | |
12973 | + */ | |
12974 | + | |
12975 | +static int mark_effected_sgs(void) | |
12976 | +{ | |
12977 | + sm_group_t *sg; | |
12978 | + sm_node_t *node, *sgnode; | |
12979 | + uint32_t dead_id; | |
12980 | + int i, effected = 0; | |
12981 | + | |
12982 | + down(&sm_sglock); | |
12983 | + | |
12984 | + list_for_each_entry(node, &sm_members, list) { | |
12985 | + if (!test_bit(SNFL_NEED_RECOVERY, &node->flags)) | |
12986 | + continue; | |
12987 | + | |
12988 | + dead_id = node->id; | |
12989 | + | |
12990 | + for (i = 0; i < SG_LEVELS; i++) { | |
12991 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
12992 | + /* check if dead node is among sg's members */ | |
12993 | + list_for_each_entry(sgnode, &sg->memb, list) { | |
12994 | + if (sgnode->id == dead_id) { | |
12995 | + set_bit(SGFL_NEED_RECOVERY, | |
12996 | + &sg->flags); | |
12997 | + effected++; | |
12998 | + break; | |
12999 | + } | |
13000 | + } | |
13001 | + } | |
13002 | + } | |
13003 | + } | |
13004 | + up(&sm_sglock); | |
13005 | + | |
13006 | + return effected; | |
13007 | +} | |
13008 | + | |
13009 | +static recover_t *alloc_recover(void) | |
13010 | +{ | |
13011 | + recover_t *rev; | |
13012 | + int i; | |
13013 | + | |
13014 | + SM_RETRY(rev = kmalloc(sizeof(recover_t), GFP_KERNEL), rev); | |
13015 | + | |
13016 | + memset(rev, 0, sizeof(recover_t)); | |
13017 | + | |
13018 | + sm_set_event_id(&rev->event_id); | |
13019 | + | |
13020 | + for (i = 0; i < SG_LEVELS; i++) { | |
13021 | + INIT_LIST_HEAD(&rev->sgs[i]); | |
13022 | + } | |
13023 | + | |
13024 | + return rev; | |
13025 | +} | |
13026 | + | |
13027 | +/* | |
13028 | + * An in-progress revent re-start for an SG is interrupted by another node | |
13029 | + * failure in the SG. Cancel an outstanding barrier if there is one. The SG | |
13030 | + * will be moved to the new revent and re-started as part of that. | |
13031 | + */ | |
13032 | + | |
13033 | +static void cancel_prev_recovery(sm_group_t *sg) | |
13034 | +{ | |
13035 | + int error; | |
13036 | + | |
13037 | + if (sg->recover_state == RECOVER_BARRIERWAIT) { | |
13038 | + error = kcl_barrier_cancel(sg->recover_barrier); | |
13039 | + if (error) | |
13040 | + log_error(sg, "cancel_prev_recovery: error %d", error); | |
13041 | + } | |
13042 | +} | |
13043 | + | |
13044 | +static void pre_recover_sg(sm_group_t *sg, recover_t *rev) | |
13045 | +{ | |
13046 | + if (sg->state == SGST_RECOVER) { | |
13047 | + cancel_prev_recovery(sg); | |
13048 | + list_del(&sg->recover_list); | |
13049 | + } | |
13050 | + | |
13051 | + sg->ops->stop(sg->service_data); | |
13052 | + sg->state = SGST_RECOVER; | |
13053 | + sg->recover_state = RECOVER_NONE; | |
13054 | + sg->recover_data = rev; | |
13055 | + list_add(&sg->recover_list, &rev->sgs[sg->level]); | |
13056 | +} | |
13057 | + | |
13058 | +/* | |
13059 | + * When adjust_members finds that some nodes are dead and mark_effected_sgs | |
13060 | + * finds that some SG's are effected by departed nodes, this is called to | |
13061 | + * collect together the SG's which need to be recovered. An revent (recovery | |
13062 | + * event) is the group of effected SG's. | |
13063 | + */ | |
13064 | + | |
13065 | +static int new_recovery(void) | |
13066 | +{ | |
13067 | + sm_group_t *sg; | |
13068 | + recover_t *rev; | |
13069 | + sm_node_t *node, *sgnode, *safe; | |
13070 | + int i; | |
13071 | + | |
13072 | + rev = alloc_recover(); | |
13073 | + list_add_tail(&rev->list, &recoveries); | |
13074 | + | |
13075 | + down(&sm_sglock); | |
13076 | + | |
13077 | + /* | |
13078 | + * Stop effected SG's and add them to the rev | |
13079 | + */ | |
13080 | + | |
13081 | + for (i = 0; i < SG_LEVELS; i++) { | |
13082 | + list_for_each_entry(sg, &sm_sg[i], list) { | |
13083 | + if (test_and_clear_bit(SGFL_NEED_RECOVERY, &sg->flags)){ | |
13084 | + if (sg->state == SGST_JOIN) | |
13085 | + continue; | |
13086 | + pre_recover_sg(sg, rev); | |
13087 | + } | |
13088 | + } | |
13089 | + } | |
13090 | + | |
13091 | + /* | |
13092 | + * For an SG needing recovery, remove dead nodes from sg->memb list | |
13093 | + */ | |
13094 | + | |
13095 | + for (i = 0; i < SG_LEVELS; i++) { | |
13096 | + list_for_each_entry(sg, &rev->sgs[i], recover_list) { | |
13097 | + | |
13098 | + /* Remove dead members from SG's member list */ | |
13099 | + list_for_each_entry_safe(sgnode, safe, &sg->memb, list){ | |
13100 | + | |
13101 | + node = sm_find_member(sgnode->id); | |
13102 | + SM_ASSERT(node, printk("id %u\n", sgnode->id);); | |
13103 | + | |
13104 | + if (test_bit(SNFL_NEED_RECOVERY, &node->flags)){ | |
13105 | + list_del(&sgnode->list); | |
13106 | + kfree(sgnode); | |
13107 | + sg->memb_count--; | |
13108 | + log_debug(sg, "remove node %u count %d", | |
13109 | + sgnode->id, sg->memb_count); | |
13110 | + } | |
13111 | + } | |
13112 | + } | |
13113 | + } | |
13114 | + | |
13115 | + up(&sm_sglock); | |
13116 | + rev->cur_level = 0; | |
13117 | + return 0; | |
13118 | +} | |
13119 | + | |
13120 | +/* | |
13121 | + * The NEED_RECOVERY bit on MML nodes is set in adjust_members() and is used in | |
13122 | + * mark_effected_sgs() and add_revent(). After that, we're done using the bit | |
13123 | + * and we clear it here. | |
13124 | + */ | |
13125 | + | |
13126 | +static void adjust_members_done(void) | |
13127 | +{ | |
13128 | + sm_node_t *node; | |
13129 | + | |
13130 | + list_for_each_entry(node, &sm_members, list) | |
13131 | + clear_bit(SNFL_NEED_RECOVERY, &node->flags); | |
13132 | +} | |
13133 | + | |
13134 | +/* | |
13135 | + * Start the service of the given SG. The service must be given an array of | |
13136 | + * nodeids specifying the new sg membership. The service is responsible to | |
13137 | + * free this chunk of memory when done with it. | |
13138 | + */ | |
13139 | + | |
13140 | +static void start_sg(sm_group_t *sg, uint32_t event_id) | |
13141 | +{ | |
13142 | + sm_node_t *node; | |
13143 | + uint32_t *memb; | |
13144 | + int count = 0; | |
13145 | + | |
13146 | + SM_RETRY(memb = kmalloc(sg->memb_count * sizeof(uint32_t), GFP_KERNEL), | |
13147 | + memb); | |
13148 | + | |
13149 | + list_for_each_entry(node, &sg->memb, list) | |
13150 | + memb[count++] = node->id; | |
13151 | + | |
13152 | + sg->ops->start(sg->service_data, memb, count, event_id, | |
13153 | + SERVICE_NODE_FAILED); | |
13154 | +} | |
13155 | + | |
13156 | +static void recovery_barrier(sm_group_t *sg) | |
13157 | +{ | |
13158 | + char bname[MAX_BARRIER_NAME_LEN]; | |
13159 | + int error, len; | |
13160 | + | |
13161 | + memset(bname, 0, MAX_BARRIER_NAME_LEN); | |
13162 | + | |
13163 | + /* bypass the barrier if we're the only member */ | |
13164 | + if (sg->memb_count == 1) { | |
13165 | + process_recovery_barrier(sg, 0); | |
13166 | + return; | |
13167 | + } | |
13168 | + | |
13169 | + len = snprintf(bname, MAX_BARRIER_NAME_LEN, "sm.%u.%u.RECOV.%u", | |
13170 | + sg->global_id, sg->recover_stop, sg->memb_count); | |
13171 | + | |
13172 | + /* We save this barrier name so we can cancel it if needed. */ | |
13173 | + memset(sg->recover_barrier, 0, MAX_BARRIER_NAME_LEN); | |
13174 | + memcpy(sg->recover_barrier, bname, len); | |
13175 | + | |
13176 | + error = sm_barrier(bname, sg->memb_count, SM_BARRIER_RECOVERY); | |
13177 | + if (error) | |
13178 | + log_error(sg, "recovery_barrier error %d: %s", error, bname); | |
13179 | +} | |
13180 | + | |
13181 | +static void recover_sg(sm_group_t *sg, int event_id) | |
13182 | +{ | |
13183 | + log_debug(sg, "recover state %d", sg->recover_state); | |
13184 | + | |
13185 | + switch (sg->recover_state) { | |
13186 | + | |
13187 | + case RECOVER_NONE: | |
13188 | + /* must wait for recovery to stop sg on all nodes */ | |
13189 | + sg->recover_state = RECOVER_BARRIERWAIT; | |
13190 | + sg->recover_stop = 0; | |
13191 | + recovery_barrier(sg); | |
13192 | + break; | |
13193 | + | |
13194 | + case RECOVER_BARRIERWAIT: | |
13195 | + break; | |
13196 | + | |
13197 | + case RECOVER_STOP: | |
13198 | + /* barrier callback sets state STOP */ | |
13199 | + sg->recover_stop = 1; | |
13200 | + sg->recover_state = RECOVER_START; | |
13201 | + start_sg(sg, event_id); | |
13202 | + break; | |
13203 | + | |
13204 | + case RECOVER_START: | |
13205 | + break; | |
13206 | + | |
13207 | + case RECOVER_STARTDONE: | |
13208 | + /* service callback sets state STARTDONE */ | |
13209 | + sg->recover_state = RECOVER_BARRIERWAIT; | |
13210 | + recovery_barrier(sg); | |
13211 | + break; | |
13212 | + | |
13213 | + case RECOVER_BARRIERDONE: | |
13214 | + /* barrier callback sets state BARRIERDONE */ | |
13215 | + sg->ops->finish(sg->service_data, event_id); | |
13216 | + list_del(&sg->recover_list); | |
13217 | + sg->recover_state = RECOVER_NONE; | |
13218 | + sg->state = SGST_RUN; | |
13219 | + | |
13220 | + /* Continue a previous, interrupted attempt to leave the sg */ | |
13221 | + if (sg->sevent) { | |
13222 | + clear_bit(SEFL_DELAY, &sg->sevent->se_flags); | |
13223 | + set_bit(SEFL_CHECK, &sg->sevent->se_flags); | |
13224 | + wake_serviced(DO_JOINLEAVE); | |
13225 | + } | |
13226 | + break; | |
13227 | + | |
13228 | + default: | |
13229 | + log_error(sg, "invalid recover_state %u", sg->recover_state); | |
13230 | + } | |
13231 | +} | |
13232 | + | |
13233 | +static void recover_level(recover_t *rev, int level) | |
13234 | +{ | |
13235 | + sm_group_t *sg, *safe; | |
13236 | + | |
13237 | + list_for_each_entry_safe(sg, safe, &rev->sgs[level], recover_list) | |
13238 | + recover_sg(sg, rev->event_id); | |
13239 | +} | |
13240 | + | |
13241 | +static void recover_levels(recover_t *rev) | |
13242 | +{ | |
13243 | + for (;;) { | |
13244 | + recover_level(rev, rev->cur_level); | |
13245 | + | |
13246 | + if (list_empty(&rev->sgs[rev->cur_level])) { | |
13247 | + if (rev->cur_level == SG_LEVELS - 1) { | |
13248 | + list_del(&rev->list); | |
13249 | + kfree(rev); | |
13250 | + return; | |
13251 | + } | |
13252 | + rev->cur_level++; | |
13253 | + continue; | |
13254 | + } | |
13255 | + break; | |
13256 | + } | |
13257 | +} | |
13258 | + | |
13259 | +/* | |
13260 | + * Called by SM thread when the cluster is quorate. It restarts | |
13261 | + * SG's that were stopped in new_recovery() due to a member death. | |
13262 | + * It waits for all SG's at level N to complete restart before | |
13263 | + * restarting SG's at level N+1. | |
13264 | + */ | |
13265 | + | |
13266 | +void process_recoveries(void) | |
13267 | +{ | |
13268 | + recover_t *rev, *safe; | |
13269 | + | |
13270 | + down(&sm_sglock); | |
13271 | + list_for_each_entry_safe(rev, safe, &recoveries, list) | |
13272 | + recover_levels(rev); | |
13273 | + up(&sm_sglock); | |
13274 | +} | |
13275 | + | |
13276 | +/* | |
13277 | + * The cnxman membership has changed. Check if there's still quorum and | |
13278 | + * whether any nodes have died. If nodes have died, initiate recovery on any | |
13279 | + * SG's they were in. This begins immediately if the cluster remains quorate; | |
13280 | + * if not this waits until the cluster regains quorum. | |
13281 | + */ | |
13282 | + | |
13283 | +void process_nodechange(void) | |
13284 | +{ | |
13285 | + int gone, effected; | |
13286 | + | |
13287 | + if ((sm_quorum = sm_quorum_next)) | |
13288 | + wake_serviced(DO_RUN); | |
13289 | + | |
13290 | + gone = adjust_members(); | |
13291 | + if (gone > 0) { | |
13292 | + effected = mark_effected_sgs(); | |
13293 | + | |
13294 | + backout_sevents(); | |
13295 | + cancel_uevents(&effected); | |
13296 | + | |
13297 | + if (effected > 0) { | |
13298 | + new_recovery(); | |
13299 | + wake_serviced(DO_RECOVERIES); | |
13300 | + } | |
13301 | + } | |
13302 | + adjust_members_done(); | |
13303 | +} | |
13304 | + | |
13305 | +int check_recovery(sm_group_t *sg, int event_id) | |
13306 | +{ | |
13307 | + if (sg->state == SGST_RECOVER) { | |
13308 | + recover_t *rev = (recover_t *) sg->recover_data; | |
13309 | + if (rev && rev->event_id == event_id) | |
13310 | + return 1; | |
13311 | + } | |
13312 | + return 0; | |
13313 | +} | |
13314 | + | |
13315 | +void process_recover_msg(sm_msg_t *smsg, uint32_t nodeid) | |
13316 | +{ | |
13317 | + sm_group_t *sg; | |
13318 | + recover_t *rev; | |
13319 | + | |
13320 | + sg = sm_global_id_to_sg(smsg->ms_global_sgid); | |
13321 | + if (!sg) { | |
13322 | + log_print("process_recover_msg: unknown sg id %x", | |
13323 | + smsg->ms_global_sgid); | |
13324 | + return; | |
13325 | + } | |
13326 | + | |
13327 | + /* we already know about the recovery and can ignore the msg */ | |
13328 | + if (sg->state == SGST_RECOVER) | |
13329 | + return; | |
13330 | + | |
13331 | + if (test_bit(SGFL_UEVENT, &sg->flags)) { | |
13332 | + /* we will initiate recovery on our own if we know about the | |
13333 | + uevent so we can ignore this */ | |
13334 | + log_debug(sg, "process_recover_msg: ignore from %u", nodeid); | |
13335 | + return; | |
13336 | + } | |
13337 | + | |
13338 | + log_debug(sg, "recovery initiated by msg from %u", nodeid); | |
13339 | + rev = alloc_recover(); | |
13340 | + list_add_tail(&rev->list, &recoveries); | |
13341 | + pre_recover_sg(sg, rev); | |
13342 | + wake_serviced(DO_RECOVERIES); | |
13343 | +} | |
13344 | diff -urN linux-orig/cluster/cman/sm_recover.h linux-patched/cluster/cman/sm_recover.h | |
13345 | --- linux-orig/cluster/cman/sm_recover.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 13346 | +++ linux-patched/cluster/cman/sm_recover.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 13347 | @@ -0,0 +1,23 @@ |
13348 | +/****************************************************************************** | |
13349 | +******************************************************************************* | |
13350 | +** | |
13351 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13352 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13353 | +** | |
13354 | +** This copyrighted material is made available to anyone wishing to use, | |
13355 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13356 | +** of the GNU General Public License v.2. | |
13357 | +** | |
13358 | +******************************************************************************* | |
13359 | +******************************************************************************/ | |
13360 | + | |
13361 | +#ifndef __SM_RECOVER_DOT_H__ | |
13362 | +#define __SM_RECOVER_DOT_H__ | |
13363 | + | |
13364 | +void init_recovery(void); | |
13365 | +void process_recoveries(void); | |
13366 | +void process_nodechange(void); | |
13367 | +int check_recovery(sm_group_t *sg, int event_id); | |
13368 | +void process_recover_msg(sm_msg_t *smsg, uint32_t nodeid); | |
13369 | + | |
13370 | +#endif | |
13371 | diff -urN linux-orig/cluster/cman/sm_services.c linux-patched/cluster/cman/sm_services.c | |
13372 | --- linux-orig/cluster/cman/sm_services.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 13373 | +++ linux-patched/cluster/cman/sm_services.c 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 13374 | @@ -0,0 +1,418 @@ |
13375 | +/****************************************************************************** | |
13376 | +******************************************************************************* | |
13377 | +** | |
13378 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13379 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13380 | +** | |
13381 | +** This copyrighted material is made available to anyone wishing to use, | |
13382 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13383 | +** of the GNU General Public License v.2. | |
13384 | +** | |
13385 | +******************************************************************************* | |
13386 | +******************************************************************************/ | |
13387 | + | |
13388 | +#include "sm.h" | |
13389 | + | |
13390 | +static struct list_head callbacks; | |
13391 | +static spinlock_t callback_lock; | |
13392 | +static struct list_head sg_registered[SG_LEVELS]; | |
13393 | + | |
13394 | +/* | |
13395 | + * These are the functions to register, join, leave, unregister, callback | |
13396 | + * with/to the sm. | |
13397 | + */ | |
13398 | + | |
13399 | +struct sc_entry { | |
13400 | + struct list_head list; | |
13401 | + uint32_t local_id; | |
13402 | + int event_id; | |
13403 | +}; | |
13404 | +typedef struct sc_entry sc_entry_t; | |
13405 | + | |
13406 | +void init_services(void) | |
13407 | +{ | |
13408 | + int i; | |
13409 | + | |
13410 | + INIT_LIST_HEAD(&callbacks); | |
13411 | + spin_lock_init(&callback_lock); | |
13412 | + | |
13413 | + for (i = 0; i < SG_LEVELS; i++) { | |
13414 | + INIT_LIST_HEAD(&sm_sg[i]); | |
13415 | + INIT_LIST_HEAD(&sg_registered[i]); | |
13416 | + } | |
13417 | + init_MUTEX(&sm_sglock); | |
13418 | +} | |
13419 | + | |
13420 | +/* Context: service */ | |
13421 | + | |
13422 | +int kcl_register_service(char *name, int namelen, int level, | |
13423 | + struct kcl_service_ops *ops, int unique, | |
13424 | + void *servicedata, uint32_t *service_id) | |
13425 | +{ | |
13426 | + sm_group_t *sg; | |
13427 | + int found = FALSE; | |
13428 | + int error = -EINVAL; | |
13429 | + | |
13430 | + if (level > SG_LEVELS - 1) | |
13431 | + goto fail; | |
13432 | + | |
13433 | + if (namelen > MAX_SERVICE_NAME_LEN) | |
13434 | + goto fail; | |
13435 | + | |
13436 | + error = kcl_addref_cluster(); | |
13437 | + if (error) | |
13438 | + goto fail; | |
13439 | + | |
13440 | + down(&sm_sglock); | |
13441 | + | |
13442 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
13443 | + if ((sg->namelen == namelen) && | |
13444 | + (!strncmp(sg->name, name, namelen))) { | |
13445 | + found = TRUE; | |
13446 | + goto next; | |
13447 | + } | |
13448 | + } | |
13449 | + | |
13450 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13451 | + if ((sg->namelen == namelen) && | |
13452 | + (!strncmp(sg->name, name, namelen))) { | |
13453 | + found = TRUE; | |
13454 | + goto next; | |
13455 | + } | |
13456 | + } | |
13457 | + | |
13458 | + next: | |
13459 | + | |
13460 | + if (found && unique) { | |
13461 | + error = -EEXIST; | |
13462 | + goto fail_unlock; | |
13463 | + } | |
13464 | + | |
13465 | + if (found) { | |
13466 | + sg->refcount++; | |
13467 | + goto out; | |
13468 | + } | |
13469 | + | |
13470 | + sg = (sm_group_t *) kmalloc(sizeof(sm_group_t) + namelen, GFP_KERNEL); | |
13471 | + if (!sg) { | |
13472 | + error = -ENOMEM; | |
13473 | + goto fail_unlock; | |
13474 | + } | |
13475 | + memset(sg, 0, sizeof(sm_group_t) + namelen); | |
13476 | + | |
13477 | + sg->refcount = 1; | |
13478 | + sg->service_data = servicedata; | |
13479 | + sg->ops = ops; | |
13480 | + sg->level = level; | |
13481 | + sg->namelen = namelen; | |
13482 | + memcpy(sg->name, name, namelen); | |
13483 | + sg->local_id = sm_new_local_id(level); | |
13484 | + sg->state = SGST_NONE; | |
13485 | + INIT_LIST_HEAD(&sg->memb); | |
13486 | + INIT_LIST_HEAD(&sg->joining); | |
13487 | + init_completion(&sg->event_comp); | |
13488 | + | |
13489 | + list_add_tail(&sg->list, &sg_registered[level]); | |
13490 | + | |
13491 | + out: | |
13492 | + *service_id = sg->local_id; | |
13493 | + up(&sm_sglock); | |
13494 | + return 0; | |
13495 | + | |
13496 | + fail_unlock: | |
13497 | + up(&sm_sglock); | |
13498 | + kcl_releaseref_cluster(); | |
13499 | + fail: | |
13500 | + return error; | |
13501 | +} | |
13502 | + | |
13503 | +/* Context: service */ | |
13504 | + | |
13505 | +void kcl_unregister_service(uint32_t local_id) | |
13506 | +{ | |
13507 | + sm_group_t *sg; | |
13508 | + int level = sm_id_to_level(local_id); | |
13509 | + | |
13510 | + down(&sm_sglock); | |
13511 | + | |
13512 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13513 | + if (sg->local_id == local_id) { | |
13514 | + SM_ASSERT(sg->refcount,); | |
13515 | + sg->refcount--; | |
13516 | + | |
13517 | + if (!sg->refcount) { | |
13518 | + list_del(&sg->list); | |
13519 | + kfree(sg); | |
13520 | + } | |
13521 | + kcl_releaseref_cluster(); | |
13522 | + break; | |
13523 | + } | |
13524 | + } | |
13525 | + up(&sm_sglock); | |
13526 | +} | |
13527 | + | |
13528 | +/* Context: service */ | |
13529 | + | |
13530 | +int kcl_join_service(uint32_t local_id) | |
13531 | +{ | |
13532 | + sm_group_t *sg; | |
13533 | + sm_sevent_t *sev; | |
13534 | + int level = sm_id_to_level(local_id); | |
13535 | + int error, found = FALSE; | |
13536 | + | |
13537 | + down(&sm_sglock); | |
13538 | + | |
13539 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13540 | + if (sg->local_id == local_id) { | |
13541 | + found = TRUE; | |
13542 | + break; | |
13543 | + } | |
13544 | + } | |
13545 | + | |
13546 | + if (!found) { | |
13547 | + up(&sm_sglock); | |
13548 | + error = -ENOENT; | |
13549 | + goto out; | |
13550 | + } | |
13551 | + | |
13552 | + if (sg->state != SGST_NONE) { | |
13553 | + up(&sm_sglock); | |
13554 | + error = -EINVAL; | |
13555 | + goto out; | |
13556 | + } | |
13557 | + | |
13558 | + sg->state = SGST_JOIN; | |
13559 | + set_bit(SGFL_SEVENT, &sg->flags); | |
13560 | + list_del(&sg->list); | |
13561 | + list_add_tail(&sg->list, &sm_sg[sg->level]); | |
13562 | + | |
13563 | + up(&sm_sglock); | |
13564 | + | |
13565 | + /* | |
13566 | + * The join is a service event which will be processed asynchronously. | |
13567 | + */ | |
13568 | + | |
13569 | + sev = kmalloc(sizeof(sm_sevent_t), GFP_KERNEL); | |
13570 | + if (!sev) { | |
13571 | + error = -ENOMEM; | |
13572 | + goto out; | |
13573 | + } | |
13574 | + | |
13575 | + memset(sev, 0, sizeof (sm_sevent_t)); | |
13576 | + sev->se_state = SEST_JOIN_BEGIN; | |
13577 | + sev->se_sg = sg; | |
13578 | + sg->sevent = sev; | |
13579 | + sm_set_event_id(&sev->se_id); | |
13580 | + | |
13581 | + new_joinleave(sev); | |
13582 | + wait_for_completion(&sg->event_comp); | |
13583 | + error = 0; | |
13584 | + | |
13585 | + out: | |
13586 | + return error; | |
13587 | +} | |
13588 | + | |
13589 | +/* Context: service */ | |
13590 | + | |
13591 | +int kcl_leave_service(uint32_t local_id) | |
13592 | +{ | |
13593 | + sm_group_t *sg = NULL; | |
13594 | + sm_sevent_t *sev; | |
13595 | + int error; | |
13596 | + | |
13597 | + error = -ENOENT; | |
13598 | + sg = sm_local_id_to_sg(local_id); | |
13599 | + if (!sg) | |
13600 | + goto out; | |
13601 | + | |
13602 | + /* sg was never joined */ | |
13603 | + error = -EINVAL; | |
13604 | + if (sg->state == SGST_NONE) | |
13605 | + goto out; | |
13606 | + | |
13607 | + /* may still be joining */ | |
13608 | + error = -EBUSY; | |
13609 | + if (test_and_set_bit(SGFL_SEVENT, &sg->flags)) | |
13610 | + goto out; | |
13611 | + | |
13612 | + error = -ENOMEM; | |
13613 | + sev = kmalloc(sizeof(sm_sevent_t), GFP_KERNEL); | |
13614 | + if (!sev) | |
13615 | + goto out; | |
13616 | + | |
13617 | + memset(sev, 0, sizeof (sm_sevent_t)); | |
13618 | + sev->se_state = SEST_LEAVE_BEGIN; | |
13619 | + set_bit(SEFL_LEAVE, &sev->se_flags); | |
13620 | + sev->se_sg = sg; | |
13621 | + sg->sevent = sev; | |
13622 | + sm_set_event_id(&sev->se_id); | |
13623 | + | |
13624 | + new_joinleave(sev); | |
13625 | + wait_for_completion(&sg->event_comp); | |
13626 | + error = 0; | |
13627 | + | |
13628 | + down(&sm_sglock); | |
13629 | + list_del(&sg->list); | |
13630 | + list_add_tail(&sg->list, &sg_registered[sg->level]); | |
13631 | + up(&sm_sglock); | |
13632 | + | |
13633 | + out: | |
13634 | + return error; | |
13635 | +} | |
13636 | + | |
13637 | +static void process_callback(uint32_t local_id, int event_id) | |
13638 | +{ | |
13639 | + sm_group_t *sg; | |
13640 | + sm_sevent_t *sev; | |
13641 | + sm_uevent_t *uev; | |
13642 | + | |
13643 | + sg = sm_local_id_to_sg(local_id); | |
13644 | + if (!sg) | |
13645 | + return; | |
13646 | + | |
13647 | + if (sg->state == SGST_RECOVER) { | |
13648 | + if (!check_recovery(sg, event_id)) { | |
13649 | + log_error(sg, "process_callback invalid recover " | |
13650 | + "event id %d", event_id); | |
13651 | + return; | |
13652 | + } | |
13653 | + | |
13654 | + if (sg->recover_state == RECOVER_START) | |
13655 | + sg->recover_state = RECOVER_STARTDONE; | |
13656 | + else | |
13657 | + log_error(sg, "process_callback recover state %u", | |
13658 | + sg->recover_state); | |
13659 | + wake_serviced(DO_RECOVERIES); | |
13660 | + } | |
13661 | + | |
13662 | + else if (test_bit(SGFL_SEVENT, &sg->flags) && sg->sevent && | |
13663 | + (sg->sevent->se_id == event_id)) { | |
13664 | + sev = sg->sevent; | |
13665 | + | |
13666 | + if (test_and_clear_bit(SEFL_ALLOW_STARTDONE, &sev->se_flags) && | |
13667 | + (sev->se_state == SEST_JSTART_SERVICEWAIT)) | |
13668 | + sev->se_state = SEST_JSTART_SERVICEDONE; | |
13669 | + | |
13670 | + set_bit(SEFL_CHECK, &sev->se_flags); | |
13671 | + wake_serviced(DO_JOINLEAVE); | |
13672 | + } | |
13673 | + | |
13674 | + else if (test_bit(SGFL_UEVENT, &sg->flags) && | |
13675 | + (sg->uevent.ue_id == event_id)) { | |
13676 | + uev = &sg->uevent; | |
13677 | + | |
13678 | + if (test_and_clear_bit(UEFL_ALLOW_STARTDONE, &uev->ue_flags)) { | |
13679 | + if (uev->ue_state == UEST_JSTART_SERVICEWAIT) | |
13680 | + uev->ue_state = UEST_JSTART_SERVICEDONE; | |
13681 | + else if (uev->ue_state == UEST_LSTART_SERVICEWAIT) | |
13682 | + uev->ue_state = UEST_LSTART_SERVICEDONE; | |
13683 | + } | |
13684 | + set_bit(UEFL_CHECK, &uev->ue_flags); | |
13685 | + wake_serviced(DO_MEMBERSHIP); | |
13686 | + } | |
13687 | + | |
13688 | + else | |
13689 | + log_error(sg, "ignoring service callback id=%x event=%u", | |
13690 | + local_id, event_id); | |
13691 | +} | |
13692 | + | |
13693 | +void process_callbacks(void) | |
13694 | +{ | |
13695 | + sc_entry_t *se; | |
13696 | + | |
13697 | + while (1) { | |
13698 | + se = NULL; | |
13699 | + | |
13700 | + spin_lock(&callback_lock); | |
13701 | + if (!list_empty(&callbacks)) { | |
13702 | + se = list_entry(callbacks.next, sc_entry_t, list); | |
13703 | + list_del(&se->list); | |
13704 | + } | |
13705 | + spin_unlock(&callback_lock); | |
13706 | + | |
13707 | + if (!se) | |
13708 | + break; | |
13709 | + process_callback(se->local_id, se->event_id); | |
13710 | + kfree(se); | |
13711 | + schedule(); | |
13712 | + } | |
13713 | +} | |
13714 | + | |
13715 | +/* Context: service */ | |
13716 | + | |
13717 | +void kcl_start_done(uint32_t local_id, int event_id) | |
13718 | +{ | |
13719 | + sc_entry_t *se; | |
13720 | + | |
13721 | + SM_RETRY(se = kmalloc(sizeof(sc_entry_t), GFP_KERNEL), se); | |
13722 | + | |
13723 | + se->local_id = local_id; | |
13724 | + se->event_id = event_id; | |
13725 | + | |
13726 | + spin_lock(&callback_lock); | |
13727 | + list_add_tail(&se->list, &callbacks); | |
13728 | + spin_unlock(&callback_lock); | |
13729 | + | |
13730 | + wake_serviced(DO_CALLBACKS); | |
13731 | +} | |
13732 | + | |
13733 | +/* Context: service */ | |
13734 | + | |
13735 | +void kcl_global_service_id(uint32_t local_id, uint32_t *global_id) | |
13736 | +{ | |
13737 | + sm_group_t *sg = sm_local_id_to_sg(local_id); | |
13738 | + | |
13739 | + if (!sg) | |
13740 | + log_print("kcl_global_service_id: can't find %x", local_id); | |
13741 | + else | |
13742 | + *global_id = sg->global_id; | |
13743 | +} | |
13744 | + | |
13745 | +static void copy_to_service(sm_group_t *sg, struct kcl_service *s) | |
13746 | +{ | |
13747 | + s->level = sg->level; | |
13748 | + s->local_id = sg->local_id; | |
13749 | + s->global_id = sg->global_id; | |
13750 | + s->node_count = sg->memb_count; | |
13751 | + strcpy(s->name, sg->name); | |
13752 | +} | |
13753 | + | |
13754 | +int kcl_get_services(struct list_head *head, int level) | |
13755 | +{ | |
13756 | + sm_group_t *sg; | |
13757 | + struct kcl_service *s; | |
13758 | + int error = -ENOMEM, count = 0; | |
13759 | + | |
13760 | + down(&sm_sglock); | |
13761 | + | |
13762 | + list_for_each_entry(sg, &sg_registered[level], list) { | |
13763 | + if (head) { | |
13764 | + s = kmalloc(sizeof(struct kcl_service), GFP_KERNEL); | |
13765 | + if (!s) | |
13766 | + goto out; | |
13767 | + copy_to_service(sg, s); | |
13768 | + list_add(&s->list, head); | |
13769 | + } | |
13770 | + count++; | |
13771 | + } | |
13772 | + | |
13773 | + list_for_each_entry(sg, &sm_sg[level], list) { | |
13774 | + if (head) { | |
13775 | + s = kmalloc(sizeof(struct kcl_service), GFP_KERNEL); | |
13776 | + if (!s) | |
13777 | + goto out; | |
13778 | + copy_to_service(sg, s); | |
13779 | + list_add(&s->list, head); | |
13780 | + } | |
13781 | + count++; | |
13782 | + } | |
13783 | + | |
13784 | + error = count; | |
13785 | + out: | |
13786 | + up(&sm_sglock); | |
13787 | + return error; | |
13788 | +} | |
13789 | + | |
13790 | +/* These three global variables listed in extern form in sm.h. */ | |
13791 | +struct list_head sm_sg[SG_LEVELS]; | |
13792 | +struct semaphore sm_sglock; | |
13793 | diff -urN linux-orig/cluster/cman/sm_services.h linux-patched/cluster/cman/sm_services.h | |
13794 | --- linux-orig/cluster/cman/sm_services.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 13795 | +++ linux-patched/cluster/cman/sm_services.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 13796 | @@ -0,0 +1,20 @@ |
13797 | +/****************************************************************************** | |
13798 | +******************************************************************************* | |
13799 | +** | |
13800 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13801 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13802 | +** | |
13803 | +** This copyrighted material is made available to anyone wishing to use, | |
13804 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13805 | +** of the GNU General Public License v.2. | |
13806 | +** | |
13807 | +******************************************************************************* | |
13808 | +******************************************************************************/ | |
13809 | + | |
13810 | +#ifndef __SM_SERVICES_DOT_H__ | |
13811 | +#define __SM_SERVICES_DOT_H__ | |
13812 | + | |
13813 | +void init_services(void); | |
13814 | +void process_callbacks(void); | |
13815 | + | |
13816 | +#endif | |
13817 | diff -urN linux-orig/cluster/cman/sm_user.c linux-patched/cluster/cman/sm_user.c | |
13818 | --- linux-orig/cluster/cman/sm_user.c 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b AM |
13819 | +++ linux-patched/cluster/cman/sm_user.c 2004-06-29 20:07:51.000000000 +0800 |
13820 | @@ -0,0 +1,569 @@ | |
4bf12011 | 13821 | +/****************************************************************************** |
13822 | +******************************************************************************* | |
13823 | +** | |
13824 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
13825 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
13826 | +** | |
13827 | +** This copyrighted material is made available to anyone wishing to use, | |
13828 | +** modify, copy, or redistribute it subject to the terms and conditions | |
13829 | +** of the GNU General Public License v.2. | |
13830 | +** | |
13831 | +******************************************************************************* | |
13832 | +******************************************************************************/ | |
13833 | + | |
13834 | +#include "sm.h" | |
13835 | +#include "cnxman-private.h" | |
13836 | + | |
13837 | +void copy_to_usernode(struct cluster_node *node, struct cl_cluster_node *unode); | |
13838 | + | |
13839 | +#define UST_REGISTER 1 | |
13840 | +#define UST_UNREGISTER 2 | |
13841 | +#define UST_JOIN 3 | |
13842 | +#define UST_LEAVE 4 | |
13843 | +#define UST_JOINED 5 | |
13844 | + | |
13845 | +struct event { | |
13846 | + struct list_head list; | |
13847 | + service_event_t type; | |
13848 | + service_start_t start_type; | |
13849 | + unsigned int event_id; | |
13850 | + unsigned int last_stop; | |
13851 | + unsigned int last_start; | |
13852 | + unsigned int last_finish; | |
13853 | + unsigned int node_count; | |
13854 | + uint32_t * nodeids; | |
13855 | +}; | |
13856 | +typedef struct event event_t; | |
13857 | + | |
13858 | +struct user_service { | |
13859 | + uint32_t local_id; | |
13860 | + pid_t pid; | |
13861 | + int signal; | |
13862 | + struct socket * sock; | |
13863 | + uint8_t state; | |
13864 | + uint8_t async; | |
13865 | + struct semaphore lock; | |
13866 | + struct list_head events; | |
13867 | + spinlock_t event_lock; | |
13868 | + unsigned int last_stop; | |
13869 | + unsigned int last_start; | |
13870 | + unsigned int last_finish; | |
13871 | + unsigned int need_startdone; | |
13872 | + unsigned int node_count; | |
13873 | + uint32_t * nodeids; | |
13874 | + int name_len; | |
13875 | + char name[MAX_SERVICE_NAME_LEN]; | |
13876 | +}; | |
13877 | +typedef struct user_service user_service_t; | |
13878 | + | |
13879 | + | |
13880 | +static void add_event(user_service_t *us, event_t *ev) | |
13881 | +{ | |
13882 | + spin_lock(&us->event_lock); | |
13883 | + list_add_tail(&ev->list, &us->events); | |
13884 | + | |
13885 | + switch(ev->type) { | |
13886 | + case SERVICE_EVENT_STOP: | |
13887 | + us->last_stop = us->last_start; | |
13888 | + break; | |
13889 | + case SERVICE_EVENT_START: | |
13890 | + us->last_start = ev->event_id; | |
13891 | + break; | |
13892 | + case SERVICE_EVENT_FINISH: | |
13893 | + us->last_finish = ev->event_id; | |
13894 | + break; | |
13895 | + case SERVICE_EVENT_LEAVEDONE: | |
13896 | + break; | |
13897 | + } | |
13898 | + spin_unlock(&us->event_lock); | |
13899 | +} | |
13900 | + | |
13901 | +static event_t *get_event(user_service_t *us) | |
13902 | +{ | |
13903 | + event_t *ev = NULL; | |
13904 | + | |
13905 | + spin_lock(&us->event_lock); | |
13906 | + if (!list_empty(&us->events)) { | |
13907 | + ev = list_entry(us->events.next, event_t, list); | |
13908 | + ev->last_stop = us->last_stop; | |
13909 | + ev->last_start = us->last_start; | |
13910 | + ev->last_finish = us->last_finish; | |
13911 | + } | |
13912 | + spin_unlock(&us->event_lock); | |
13913 | + return ev; | |
13914 | +} | |
13915 | + | |
13916 | +static void del_event(user_service_t *us, event_t *ev) | |
13917 | +{ | |
13918 | + spin_lock(&us->event_lock); | |
13919 | + list_del(&ev->list); | |
13920 | + spin_unlock(&us->event_lock); | |
13921 | +} | |
13922 | + | |
13923 | +static event_t *alloc_event(void) | |
13924 | +{ | |
13925 | + event_t *ev; | |
13926 | + SM_RETRY(ev = (event_t *) kmalloc(sizeof(event_t), GFP_KERNEL), ev); | |
13927 | + memset(ev, 0, sizeof(event_t)); | |
13928 | + return ev; | |
13929 | +} | |
13930 | + | |
13931 | +/* us->lock must be held before calling */ | |
13932 | +static void user_notify(user_service_t *us) | |
13933 | +{ | |
13934 | + if (us->sock) | |
13935 | + queue_oob_skb(us->sock, CLUSTER_OOB_MSG_SERVICEEVENT); | |
13936 | + if (us->pid && us->signal) | |
13937 | + kill_proc(us->pid, us->signal, 0); | |
13938 | +} | |
13939 | + | |
13940 | +static service_start_t start_type(int type) | |
13941 | +{ | |
13942 | + switch (type) { | |
13943 | + case SERVICE_NODE_FAILED: | |
13944 | + return SERVICE_START_FAILED; | |
13945 | + case SERVICE_NODE_JOIN: | |
13946 | + return SERVICE_START_JOIN; | |
13947 | + case SERVICE_NODE_LEAVE: | |
13948 | + return SERVICE_START_LEAVE; | |
13949 | + } | |
13950 | + return 0; | |
13951 | +} | |
13952 | + | |
13953 | +static int user_stop(void *servicedata) | |
13954 | +{ | |
13955 | + user_service_t *us = (user_service_t *) servicedata; | |
13956 | + event_t *ev; | |
13957 | + | |
13958 | + down(&us->lock); | |
13959 | + if (!us->sock) | |
13960 | + goto out; | |
13961 | + | |
13962 | + ev = alloc_event(); | |
13963 | + ev->type = SERVICE_EVENT_STOP; | |
13964 | + | |
13965 | + add_event(us, ev); | |
13966 | + user_notify(us); | |
13967 | + out: | |
13968 | + up(&us->lock); | |
13969 | + return 0; | |
13970 | +} | |
13971 | + | |
13972 | +static int user_start(void *servicedata, uint32_t *nodeids, int count, | |
13973 | + int event_id, int type) | |
13974 | +{ | |
13975 | + user_service_t *us = (user_service_t *) servicedata; | |
13976 | + event_t *ev; | |
13977 | + | |
13978 | + down(&us->lock); | |
13979 | + if (!us->sock) { | |
13980 | + kcl_start_done(us->local_id, event_id); | |
13981 | + goto out; | |
13982 | + } | |
13983 | + | |
13984 | + us->need_startdone = event_id; | |
13985 | + | |
13986 | + ev = alloc_event(); | |
13987 | + ev->type = SERVICE_EVENT_START; | |
13988 | + ev->node_count = count; | |
13989 | + ev->start_type = start_type(type); | |
13990 | + ev->event_id = event_id; | |
13991 | + ev->nodeids = nodeids; | |
13992 | + | |
13993 | + add_event(us, ev); | |
13994 | + user_notify(us); | |
13995 | + out: | |
13996 | + up(&us->lock); | |
13997 | + return 0; | |
13998 | +} | |
13999 | + | |
14000 | +static void user_finish(void *servicedata, int event_id) | |
14001 | +{ | |
14002 | + user_service_t *us = (user_service_t *) servicedata; | |
14003 | + event_t *ev; | |
14004 | + | |
14005 | + down(&us->lock); | |
14006 | + if (!us->sock) | |
14007 | + goto out; | |
14008 | + | |
14009 | + ev = alloc_event(); | |
14010 | + ev->type = SERVICE_EVENT_FINISH; | |
14011 | + ev->event_id = event_id; | |
14012 | + | |
14013 | + add_event(us, ev); | |
14014 | + user_notify(us); | |
14015 | + out: | |
14016 | + up(&us->lock); | |
14017 | +} | |
14018 | + | |
14019 | +struct kcl_service_ops user_service_ops = { | |
14020 | + .stop = user_stop, | |
14021 | + .start = user_start, | |
14022 | + .finish = user_finish | |
14023 | +}; | |
14024 | + | |
5cdbd17b | 14025 | +static int user_register(char *u_name, user_service_t **us_data) |
4bf12011 | 14026 | +{ |
14027 | + user_service_t *us; | |
5cdbd17b AM |
14028 | + char name[MAX_SERVICE_NAME_LEN+1]; |
14029 | + int len, error; | |
14030 | + | |
14031 | + memset(name, 0, MAX_SERVICE_NAME_LEN+1); | |
14032 | + | |
14033 | + if (copy_from_user(&name, u_name, MAX_SERVICE_NAME_LEN)) | |
14034 | + return -EFAULT; | |
4bf12011 | 14035 | + |
5cdbd17b AM |
14036 | + len = strlen(name); |
14037 | + if (len > MAX_SERVICE_NAME_LEN) | |
4bf12011 | 14038 | + return -ENAMETOOLONG; |
14039 | + if (!len) | |
14040 | + return -EINVAL; | |
14041 | + | |
14042 | + us = kmalloc(sizeof(user_service_t), GFP_KERNEL); | |
14043 | + if (!us) | |
14044 | + return -ENOMEM; | |
14045 | + memset(us, 0, sizeof(user_service_t)); | |
14046 | + us->nodeids = NULL; | |
14047 | + INIT_LIST_HEAD(&us->events); | |
14048 | + spin_lock_init(&us->event_lock); | |
14049 | + init_MUTEX(&us->lock); | |
14050 | + us->name_len = len; | |
14051 | + memcpy(us->name, name, len); | |
14052 | + | |
14053 | + error = kcl_register_service(name, len, SERVICE_LEVEL_USER, | |
14054 | + &user_service_ops, TRUE, (void *) us, | |
14055 | + &us->local_id); | |
14056 | + if (error) { | |
14057 | + kfree(us); | |
14058 | + us = NULL; | |
14059 | + } | |
14060 | + *us_data = us; | |
14061 | + return error; | |
14062 | +} | |
14063 | + | |
14064 | +static void user_unregister(user_service_t *us) | |
14065 | +{ | |
14066 | + event_t *ev; | |
14067 | + | |
14068 | + kcl_unregister_service(us->local_id); | |
14069 | + | |
14070 | + if (us->nodeids) | |
14071 | + kfree(us->nodeids); | |
14072 | + | |
14073 | + while ((ev = get_event(us))) { | |
14074 | + del_event(us, ev); | |
14075 | + if (ev->nodeids) | |
14076 | + kfree(ev->nodeids); | |
14077 | + kfree(ev); | |
14078 | + } | |
14079 | +} | |
14080 | + | |
14081 | +static int user_join_async(void *arg) | |
14082 | +{ | |
14083 | + user_service_t *us = arg; | |
14084 | + int user_gone = 0; | |
14085 | + | |
14086 | + daemonize("cman_userjoin"); | |
14087 | + | |
14088 | + kcl_join_service(us->local_id); | |
14089 | + | |
14090 | + down(&us->lock); | |
14091 | + us->state = UST_JOINED; | |
14092 | + us->async = 0; | |
14093 | + if (!us->sock) { | |
14094 | + if (us->need_startdone) | |
14095 | + kcl_start_done(us->local_id, us->need_startdone); | |
14096 | + user_gone = 1; | |
14097 | + } | |
14098 | + up(&us->lock); | |
14099 | + | |
14100 | + if (user_gone) { | |
14101 | + kcl_leave_service(us->local_id); | |
14102 | + user_unregister(us); | |
14103 | + kfree(us); | |
14104 | + } | |
14105 | + return 0; | |
14106 | +} | |
14107 | + | |
14108 | +static int user_leave_async(void *arg) | |
14109 | +{ | |
14110 | + user_service_t *us = arg; | |
14111 | + | |
14112 | + daemonize("cman_userleave"); | |
14113 | + | |
14114 | + kcl_leave_service(us->local_id); | |
14115 | + | |
14116 | + down(&us->lock); | |
14117 | + us->async = 0; | |
14118 | + if (!us->sock) { | |
14119 | + user_unregister(us); | |
14120 | + kfree(us); | |
14121 | + } else { | |
14122 | + event_t *ev = alloc_event(); | |
14123 | + ev->type = SERVICE_EVENT_LEAVEDONE; | |
14124 | + add_event(us, ev); | |
14125 | + user_notify(us); | |
14126 | + up(&us->lock); | |
14127 | + } | |
14128 | + | |
14129 | + return 0; | |
14130 | +} | |
14131 | + | |
14132 | +static int user_join(user_service_t *us, int wait) | |
14133 | +{ | |
14134 | + int error = 0; | |
14135 | + | |
14136 | + if (wait) { | |
14137 | + error = kcl_join_service(us->local_id); | |
14138 | + us->state = UST_JOINED; | |
14139 | + } | |
14140 | + else { | |
14141 | + us->async = 1; | |
14142 | + kernel_thread(user_join_async, us, 0); | |
14143 | + } | |
14144 | + | |
14145 | + return error; | |
14146 | +} | |
14147 | + | |
14148 | +static void user_leave(user_service_t *us, int wait) | |
14149 | +{ | |
14150 | + if (wait) | |
14151 | + kcl_leave_service(us->local_id); | |
14152 | + else { | |
14153 | + us->async = 1; | |
14154 | + kernel_thread(user_leave_async, us, 0); | |
14155 | + } | |
14156 | +} | |
14157 | + | |
14158 | +static int user_start_done(user_service_t *us, unsigned int event_id) | |
14159 | +{ | |
14160 | + if (!us->need_startdone) | |
14161 | + return -EINVAL; | |
14162 | + if (us->need_startdone == event_id) | |
14163 | + us->need_startdone = 0; | |
14164 | + kcl_start_done(us->local_id, event_id); | |
14165 | + return 0; | |
14166 | +} | |
14167 | + | |
14168 | +static void user_set_signal(user_service_t *us, int signal) | |
14169 | +{ | |
14170 | + us->pid = current->pid; | |
14171 | + us->signal = signal; | |
14172 | +} | |
14173 | + | |
14174 | +static int user_get_event(user_service_t *us, | |
14175 | + struct cl_service_event *user_event) | |
14176 | +{ | |
14177 | + event_t *ev; | |
14178 | + struct cl_service_event event; | |
14179 | + | |
14180 | + ev = get_event(us); | |
14181 | + if (!ev) | |
14182 | + return 0; | |
14183 | + | |
14184 | + event.type = ev->type; | |
14185 | + event.start_type = ev->start_type; | |
14186 | + event.event_id = ev->event_id; | |
14187 | + event.last_stop = ev->last_stop; | |
14188 | + event.last_start = ev->last_start; | |
14189 | + event.last_finish = ev->last_finish; | |
14190 | + event.node_count = ev->node_count; | |
14191 | + | |
14192 | + if (copy_to_user(user_event, &event, sizeof(struct cl_service_event))) | |
14193 | + return -EFAULT; | |
14194 | + | |
14195 | + del_event(us, ev); | |
14196 | + | |
14197 | + if (ev->type == SERVICE_EVENT_START) { | |
14198 | + if (us->nodeids) | |
14199 | + kfree(us->nodeids); | |
14200 | + us->nodeids = ev->nodeids; | |
14201 | + us->node_count = ev->node_count; | |
14202 | + } | |
14203 | + | |
14204 | + kfree(ev); | |
14205 | + return 1; | |
14206 | +} | |
14207 | + | |
14208 | +static int user_get_members(user_service_t *us, | |
14209 | + struct cl_cluster_nodelist *u_nodelist) | |
14210 | +{ | |
14211 | + struct cl_cluster_nodelist user_nodelist; | |
14212 | + struct cl_cluster_node user_node, *u_node; | |
14213 | + struct cluster_node *node; | |
14214 | + unsigned int i; | |
14215 | + int num_nodes = 0; | |
14216 | + | |
14217 | + if (!u_nodelist) | |
14218 | + return us->node_count; | |
14219 | + | |
14220 | + if (copy_from_user(&user_nodelist, (void __user *) u_nodelist, | |
14221 | + sizeof(struct cl_cluster_nodelist))) | |
14222 | + return -EFAULT; | |
14223 | + | |
14224 | + if (user_nodelist.max_members < us->node_count) | |
14225 | + return -E2BIG; | |
14226 | + | |
14227 | + u_node = user_nodelist.nodes; | |
14228 | + | |
14229 | + for (i = 0; i < us->node_count; i++) { | |
14230 | + node = find_node_by_nodeid(us->nodeids[i]); | |
14231 | + if (!node) | |
14232 | + continue; | |
14233 | + | |
14234 | + copy_to_usernode(node, &user_node); | |
14235 | + if (copy_to_user(u_node, &user_node, | |
14236 | + sizeof(struct cl_cluster_node))) | |
14237 | + return -EFAULT; | |
14238 | + | |
14239 | + u_node++; | |
14240 | + num_nodes++; | |
14241 | + } | |
14242 | + return num_nodes; | |
14243 | +} | |
14244 | + | |
14245 | +static int user_global_id(user_service_t *us, uint32_t *id) | |
14246 | +{ | |
14247 | + uint32_t gid = 0; | |
14248 | + | |
14249 | + if (us->state != UST_JOINED) | |
14250 | + return -EINVAL; | |
14251 | + | |
14252 | + kcl_global_service_id(us->local_id, &gid); | |
14253 | + | |
14254 | + if (copy_to_user(id, &gid, sizeof(uint32_t))) | |
14255 | + return -EFAULT; | |
14256 | + return 0; | |
14257 | +} | |
14258 | + | |
14259 | +static int user_set_level(user_service_t *us, int level) | |
14260 | +{ | |
14261 | + int prev_id = us->local_id; | |
14262 | + int error; | |
14263 | + | |
14264 | + if (us->state != UST_REGISTER) | |
14265 | + return -EINVAL; | |
14266 | + | |
14267 | + error = kcl_register_service(us->name, us->name_len, level, | |
14268 | + &user_service_ops, TRUE, (void *) us, | |
14269 | + &us->local_id); | |
14270 | + if (error) | |
14271 | + return error; | |
14272 | + | |
14273 | + kcl_unregister_service(prev_id); | |
14274 | + return 0; | |
14275 | +} | |
14276 | + | |
14277 | +int sm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) | |
14278 | +{ | |
14279 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
14280 | + user_service_t *us = c->service_data; | |
14281 | + int error = 0; | |
14282 | + | |
14283 | + if (!us && cmd != SIOCCLUSTER_SERVICE_REGISTER) | |
14284 | + return -EINVAL; | |
14285 | + | |
14286 | + switch (cmd) { | |
14287 | + case SIOCCLUSTER_SERVICE_REGISTER: | |
14288 | + error = user_register((char *) arg, &us); | |
14289 | + if (!error) { | |
14290 | + us->state = UST_REGISTER; | |
14291 | + us->sock = sock; | |
14292 | + c->service_data = us; | |
14293 | + } | |
14294 | + break; | |
14295 | + | |
14296 | + case SIOCCLUSTER_SERVICE_UNREGISTER: | |
14297 | + down(&us->lock); | |
14298 | + us->state = UST_UNREGISTER; | |
14299 | + user_unregister(us); | |
14300 | + up(&us->lock); | |
14301 | + break; | |
14302 | + | |
14303 | + case SIOCCLUSTER_SERVICE_JOIN: | |
14304 | + us->state = UST_JOIN; | |
14305 | + user_join(us, 0); | |
14306 | + break; | |
14307 | + | |
14308 | + case SIOCCLUSTER_SERVICE_LEAVE: | |
14309 | + down(&us->lock); | |
14310 | + if (us->state != UST_JOINED) { | |
14311 | + error = -EBUSY; | |
14312 | + up(&us->lock); | |
14313 | + } else { | |
14314 | + us->state = UST_LEAVE; | |
14315 | + up(&us->lock); | |
14316 | + user_leave(us, 0); | |
14317 | + } | |
14318 | + break; | |
14319 | + | |
14320 | + case SIOCCLUSTER_SERVICE_SETSIGNAL: | |
14321 | + user_set_signal(us, (int) arg); | |
14322 | + break; | |
14323 | + | |
14324 | + case SIOCCLUSTER_SERVICE_STARTDONE: | |
14325 | + error = user_start_done(us, (unsigned int) arg); | |
14326 | + break; | |
14327 | + | |
14328 | + case SIOCCLUSTER_SERVICE_GETEVENT: | |
14329 | + error = user_get_event(us, (struct cl_service_event *) arg); | |
14330 | + break; | |
14331 | + | |
14332 | + case SIOCCLUSTER_SERVICE_GETMEMBERS: | |
14333 | + error = user_get_members(us, (struct cl_cluster_nodelist *)arg); | |
14334 | + break; | |
14335 | + | |
14336 | + case SIOCCLUSTER_SERVICE_GLOBALID: | |
14337 | + error = user_global_id(us, (uint32_t *) arg); | |
14338 | + break; | |
14339 | + | |
14340 | + case SIOCCLUSTER_SERVICE_SETLEVEL: | |
14341 | + error = user_set_level(us, (int) arg); | |
14342 | + break; | |
14343 | + | |
14344 | + default: | |
14345 | + error = -EINVAL; | |
14346 | + } | |
14347 | + | |
14348 | + return error; | |
14349 | +} | |
14350 | + | |
14351 | +void sm_sock_release(struct socket *sock) | |
14352 | +{ | |
14353 | + struct cluster_sock *c = cluster_sk(sock->sk); | |
14354 | + user_service_t *us = c->service_data; | |
14355 | + int state; | |
14356 | + | |
14357 | + if (!us) | |
14358 | + return; | |
14359 | + | |
14360 | + down(&us->lock); | |
14361 | + us->sock = NULL; | |
14362 | + c->service_data = NULL; | |
14363 | + | |
14364 | + if (us->need_startdone) | |
14365 | + kcl_start_done(us->local_id, us->need_startdone); | |
14366 | + | |
14367 | + if (us->async) { | |
14368 | + /* async thread will clean up before exiting */ | |
14369 | + up(&us->lock); | |
14370 | + return; | |
14371 | + } | |
14372 | + state = us->state; | |
14373 | + up(&us->lock); | |
14374 | + | |
14375 | + switch (state) { | |
14376 | + case UST_JOIN: | |
14377 | + break; | |
14378 | + case UST_JOINED: | |
14379 | + user_leave(us, 1); | |
14380 | + /* fall through */ | |
14381 | + case UST_LEAVE: | |
14382 | + case UST_REGISTER: | |
14383 | + user_unregister(us); | |
14384 | + /* fall through */ | |
14385 | + case UST_UNREGISTER: | |
14386 | + kfree(us); | |
14387 | + break; | |
14388 | + } | |
14389 | +} | |
14390 | diff -urN linux-orig/cluster/cman/sm_user.h linux-patched/cluster/cman/sm_user.h | |
14391 | --- linux-orig/cluster/cman/sm_user.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 14392 | +++ linux-patched/cluster/cman/sm_user.h 2004-06-29 20:07:51.000000000 +0800 |
4bf12011 | 14393 | @@ -0,0 +1,21 @@ |
14394 | +/****************************************************************************** | |
14395 | +******************************************************************************* | |
14396 | +** | |
14397 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14398 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14399 | +** | |
14400 | +** This copyrighted material is made available to anyone wishing to use, | |
14401 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14402 | +** of the GNU General Public License v.2. | |
14403 | +** | |
14404 | +******************************************************************************* | |
14405 | +******************************************************************************/ | |
14406 | + | |
14407 | +#ifndef __SM_USER_DOT_H__ | |
14408 | +#define __SM_USER_DOT_H__ | |
14409 | + | |
14410 | +int sm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg); | |
14411 | +void sm_sock_release(struct socket *sock); | |
14412 | +void sm_sock_bind(struct socket *sock); | |
14413 | + | |
14414 | +#endif | |
14415 | diff -urN linux-orig/include/cluster/cnxman-socket.h linux-patched/include/cluster/cnxman-socket.h | |
14416 | --- linux-orig/include/cluster/cnxman-socket.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 14417 | +++ linux-patched/include/cluster/cnxman-socket.h 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 14418 | @@ -0,0 +1,226 @@ |
14419 | +/****************************************************************************** | |
14420 | +******************************************************************************* | |
14421 | +** | |
14422 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14423 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14424 | +** | |
14425 | +** This copyrighted material is made available to anyone wishing to use, | |
14426 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14427 | +** of the GNU General Public License v.2. | |
14428 | +** | |
14429 | +******************************************************************************* | |
14430 | +******************************************************************************/ | |
14431 | + | |
14432 | +/* CMAN socket interface header, | |
14433 | + may be include by user or kernel code */ | |
14434 | + | |
14435 | +#ifndef __CNXMAN_SOCKET_H | |
14436 | +#define __CNXMAN_SOCKET_H | |
14437 | + | |
14438 | +/* Just made these up but the address family must be less than 32 (NPROTO) */ | |
14439 | +#define AF_CLUSTER 31 | |
14440 | +#define PF_CLUSTER AF_CLUSTER | |
14441 | + | |
14442 | +/* Protocol(socket) types */ | |
14443 | +#define CLPROTO_MASTER 2 | |
14444 | +#define CLPROTO_CLIENT 3 | |
14445 | + | |
14446 | +/* Setsockopt -- maybe should be ioctls?? */ | |
14447 | +#define CLU_SET_MULTICAST 100 | |
14448 | +#define CLU_JOIN_CLUSTER 101 | |
14449 | +#define CLU_LEAVE_CLUSTER 102 | |
14450 | +#define CLU_SET_RCVONLY 103 | |
14451 | +#define CLU_SET_UNICAST 104 | |
14452 | +#define KCL_SET_MULTICAST 105 | |
14453 | +#define KCL_SET_RCVONLY 106 | |
14454 | +#define KCL_SET_UNICAST 107 | |
14455 | +#define KCL_SET_NODENAME 108 | |
14456 | +#define CLU_SET_NODENAME 109 | |
14457 | + | |
14458 | +/* ioctls -- should register these properly */ | |
14459 | +#define SIOCCLUSTER_NOTIFY _IOW('x', 0x01, int) | |
14460 | +#define SIOCCLUSTER_REMOVENOTIFY _IO( 'x', 0x02) | |
14461 | +#define SIOCCLUSTER_GETMEMBERS _IOR('x', 0x03, struct cl_cluster_nodelist) | |
14462 | +#define SIOCCLUSTER_SETEXPECTED_VOTES _IOW('x', 0x04, int) | |
14463 | +#define SIOCCLUSTER_ISQUORATE _IO( 'x', 0x05) | |
14464 | +#define SIOCCLUSTER_ISLISTENING _IOW('x', 0x06, struct cl_listen_request) | |
14465 | +#define SIOCCLUSTER_GETALLMEMBERS _IOR('x', 0x07, struct cl_cluster_nodelist) | |
14466 | +#define SIOCCLUSTER_SET_VOTES _IOW('x', 0x08, int) | |
14467 | +#define SIOCCLUSTER_GET_VERSION _IOR('x', 0x09, struct cl_version) | |
14468 | +#define SIOCCLUSTER_SET_VERSION _IOW('x', 0x0a, struct cl_version) | |
14469 | +#define SIOCCLUSTER_ISACTIVE _IO( 'x', 0x0b) | |
14470 | +#define SIOCCLUSTER_KILLNODE _IOW('x', 0x0c, int) | |
14471 | +#define SIOCCLUSTER_GET_JOINCOUNT _IO( 'x', 0x0d) | |
14472 | +#define SIOCCLUSTER_SERVICE_REGISTER _IOW('x', 0x0e, char) | |
14473 | +#define SIOCCLUSTER_SERVICE_UNREGISTER _IO('x', 0x0f) | |
14474 | +#define SIOCCLUSTER_SERVICE_JOIN _IO( 'x', 0x10) | |
14475 | +#define SIOCCLUSTER_SERVICE_LEAVE _IO( 'x', 0x20) | |
14476 | +#define SIOCCLUSTER_SERVICE_SETSIGNAL _IOW('x', 0x30, int) | |
14477 | +#define SIOCCLUSTER_SERVICE_STARTDONE _IOW('x', 0x40, unsigned int) | |
14478 | +#define SIOCCLUSTER_SERVICE_GETEVENT _IOR('x', 0x50, struct cl_service_event) | |
14479 | +#define SIOCCLUSTER_SERVICE_GETMEMBERS _IOR('x', 0x60, struct cl_cluster_nodelist) | |
14480 | +#define SIOCCLUSTER_SERVICE_GLOBALID _IOR('x', 0x70, uint32_t) | |
14481 | +#define SIOCCLUSTER_SERVICE_SETLEVEL _IOR('x', 0x80, int) | |
14482 | +#define SIOCCLUSTER_GETNODE _IOWR('x', 0x90, struct cl_cluster_node) | |
14483 | +#define SIOCCLUSTER_BARRIER _IOW('x', 0x0a0, struct cl_barrier_info) | |
14484 | + | |
14485 | +/* Maximum size of a cluster message */ | |
14486 | +#define MAX_CLUSTER_MESSAGE 1500 | |
14487 | +#define MAX_CLUSTER_MEMBER_NAME_LEN 255 | |
14488 | +#define MAX_BARRIER_NAME_LEN 33 | |
14489 | +#define MAX_SA_ADDR_LEN 12 | |
14490 | +#define MAX_CLUSTER_NAME_LEN 16 | |
14491 | + | |
14492 | +/* Well-known cluster port numbers */ | |
14493 | +#define CLUSTER_PORT_MEMBERSHIP 1 /* Mustn't block during cluster | |
14494 | + * transitions! */ | |
14495 | +#define CLUSTER_PORT_SERVICES 2 | |
14496 | +#define CLUSTER_PORT_SYSMAN 10 /* Remote execution daemon */ | |
14497 | +#define CLUSTER_PORT_CLVMD 11 /* Cluster LVM daemon */ | |
14498 | +#define CLUSTER_PORT_SLM 12 /* LVM SLM (simple lock manager) */ | |
14499 | + | |
14500 | +/* Port numbers above this will be blocked when the cluster is inquorate or in | |
14501 | + * transition */ | |
14502 | +#define HIGH_PROTECTED_PORT 9 | |
14503 | + | |
14504 | +/* Reasons for leaving the cluster */ | |
14505 | +#define CLUSTER_LEAVEFLAG_DOWN 0 /* Normal shutdown */ | |
14506 | +#define CLUSTER_LEAVEFLAG_KILLED 1 | |
14507 | +#define CLUSTER_LEAVEFLAG_PANIC 2 | |
14508 | +#define CLUSTER_LEAVEFLAG_REMOVED 3 /* This one can reduce quorum */ | |
14509 | +#define CLUSTER_LEAVEFLAG_REJECTED 4 /* Not allowed into the cluster in the | |
14510 | + * first place */ | |
14511 | +#define CLUSTER_LEAVEFLAG_INCONSISTENT 5 /* Our view of the cluster is | |
14512 | + * in a minority */ | |
14513 | +#define CLUSTER_LEAVEFLAG_DEAD 6 /* Discovered to be dead */ | |
14514 | +#define CLUSTER_LEAVEFLAG_FORCE 0x10 /* Forced by command-line */ | |
14515 | + | |
14516 | +/* OOB messages sent to a local socket */ | |
14517 | +#define CLUSTER_OOB_MSG_PORTCLOSED 1 | |
14518 | +#define CLUSTER_OOB_MSG_STATECHANGE 2 | |
14519 | +#define CLUSTER_OOB_MSG_SERVICEEVENT 3 | |
14520 | + | |
14521 | +/* Sendmsg flags, these are above the normal sendmsg flags so they don't | |
14522 | + * interfere */ | |
14523 | +#define MSG_NOACK 0x010000 /* Don't need an ACK for this message */ | |
14524 | +#define MSG_QUEUE 0x020000 /* Queue the message for sending later */ | |
14525 | +#define MSG_MULTICAST 0x080000 /* Message was sent to all nodes in the cluster | |
14526 | + */ | |
14527 | +#define MSG_ALLINT 0x100000 /* Send out of all interfaces */ | |
14528 | + | |
14529 | +typedef enum { NODESTATE_REMOTEMEMBER, NODESTATE_JOINING, NODESTATE_MEMBER, | |
14530 | + NODESTATE_DEAD } nodestate_t; | |
14531 | + | |
14532 | + | |
14533 | +struct sockaddr_cl { | |
14534 | + unsigned short scl_family; | |
14535 | + unsigned char scl_flags; | |
14536 | + unsigned char scl_port; | |
14537 | + int scl_nodeid; | |
14538 | +}; | |
14539 | + | |
14540 | +/* This is how we pass the multicast socket into kernel space. addr is the | |
14541 | + * multicast address to use in the address family of the socket (eg for UDP it | |
14542 | + * might be 255.255.255.0) */ | |
14543 | +struct cl_multicast_sock { | |
14544 | + int fd; /* FD of master socket to do multicast on */ | |
14545 | + int number; /* Socket number, to match up recvonly & bcast | |
14546 | + * sockets */ | |
14547 | +}; | |
14548 | + | |
14549 | +/* Cluster configuration info passed when we join the cluster */ | |
14550 | +struct cl_join_cluster_info { | |
14551 | + unsigned char votes; | |
14552 | + unsigned int expected_votes; | |
14553 | + unsigned int two_node; | |
14554 | + unsigned int config_version; | |
14555 | + | |
14556 | + char cluster_name[17]; | |
14557 | +}; | |
14558 | + | |
14559 | + | |
14560 | +/* This is the structure, per node, returned from the membership ioctl */ | |
14561 | +struct cl_cluster_node { | |
14562 | + unsigned int size; | |
14563 | + unsigned int node_id; | |
14564 | + unsigned int us; | |
14565 | + unsigned int leave_reason; | |
14566 | + unsigned int incarnation; | |
14567 | + nodestate_t state; | |
14568 | + char name[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
14569 | + unsigned char votes; | |
14570 | +}; | |
14571 | + | |
14572 | +/* The struct passed to the membership ioctls */ | |
14573 | +struct cl_cluster_nodelist { | |
14574 | + uint32_t max_members; | |
14575 | + struct cl_cluster_node *nodes; | |
14576 | +}; | |
14577 | + | |
14578 | +/* Structure passed to SIOCCLUSTER_ISLISTENING */ | |
14579 | +struct cl_listen_request { | |
14580 | + unsigned char port; | |
14581 | + int nodeid; | |
14582 | +}; | |
14583 | + | |
14584 | +/* A Cluster PORTCLOSED message - received by a local user as an OOB message */ | |
14585 | +struct cl_portclosed_oob { | |
14586 | + unsigned char cmd; /* CLUSTER_OOB_MSG_PORTCLOSED */ | |
14587 | + unsigned char port; | |
14588 | +}; | |
14589 | + | |
14590 | +/* Get all version numbers or set the config version */ | |
14591 | +struct cl_version { | |
14592 | + unsigned int major; | |
14593 | + unsigned int minor; | |
14594 | + unsigned int patch; | |
14595 | + unsigned int config; | |
14596 | +}; | |
14597 | + | |
14598 | +/* structure passed to barrier ioctls */ | |
14599 | +struct cl_barrier_info { | |
14600 | + char cmd; | |
14601 | + char name[MAX_BARRIER_NAME_LEN]; | |
14602 | + unsigned int flags; | |
14603 | + unsigned long arg; | |
14604 | +}; | |
14605 | + | |
14606 | +typedef enum { SERVICE_EVENT_STOP, SERVICE_EVENT_START, SERVICE_EVENT_FINISH, | |
14607 | + SERVICE_EVENT_LEAVEDONE } service_event_t; | |
14608 | + | |
14609 | +typedef enum { SERVICE_START_FAILED, SERVICE_START_JOIN, SERVICE_START_LEAVE } | |
14610 | + service_start_t; | |
14611 | + | |
14612 | +struct cl_service_event { | |
14613 | + service_event_t type; | |
14614 | + service_start_t start_type; | |
14615 | + unsigned int event_id; | |
14616 | + unsigned int last_stop; | |
14617 | + unsigned int last_start; | |
14618 | + unsigned int last_finish; | |
14619 | + unsigned int node_count; | |
14620 | +}; | |
14621 | + | |
14622 | + | |
14623 | +/* Commands to the barrier ioctl */ | |
14624 | +#define BARRIER_IOCTL_REGISTER 1 | |
14625 | +#define BARRIER_IOCTL_CHANGE 2 | |
14626 | +#define BARRIER_IOCTL_DELETE 3 | |
14627 | +#define BARRIER_IOCTL_WAIT 4 | |
14628 | + | |
14629 | +/* Attributes of a barrier - bitmask */ | |
14630 | +#define BARRIER_ATTR_AUTODELETE 1 | |
14631 | +#define BARRIER_ATTR_MULTISTEP 2 | |
14632 | +#define BARRIER_ATTR_MANUAL 4 | |
14633 | +#define BARRIER_ATTR_ENABLED 8 | |
14634 | +#define BARRIER_ATTR_CALLBACK 16 | |
14635 | + | |
14636 | +/* Attribute setting commands */ | |
14637 | +#define BARRIER_SETATTR_AUTODELETE 1 | |
14638 | +#define BARRIER_SETATTR_MULTISTEP 2 | |
14639 | +#define BARRIER_SETATTR_ENABLED 3 | |
14640 | +#define BARRIER_SETATTR_NODES 4 | |
14641 | +#define BARRIER_SETATTR_CALLBACK 5 | |
14642 | +#define BARRIER_SETATTR_TIMEOUT 6 | |
14643 | + | |
14644 | +#endif | |
14645 | diff -urN linux-orig/include/cluster/cnxman.h linux-patched/include/cluster/cnxman.h | |
14646 | --- linux-orig/include/cluster/cnxman.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 14647 | +++ linux-patched/include/cluster/cnxman.h 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 14648 | @@ -0,0 +1,87 @@ |
14649 | +/****************************************************************************** | |
14650 | +******************************************************************************* | |
14651 | +** | |
14652 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14653 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14654 | +** | |
14655 | +** This copyrighted material is made available to anyone wishing to use, | |
14656 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14657 | +** of the GNU General Public License v.2. | |
14658 | +** | |
14659 | +******************************************************************************* | |
14660 | +******************************************************************************/ | |
14661 | + | |
14662 | +#ifndef __CNXMAN_H | |
14663 | +#define __CNXMAN_H | |
14664 | + | |
14665 | +#include "linux/in6.h" | |
14666 | +#include "cluster/cnxman-socket.h" | |
14667 | + | |
14668 | +/* In-kernel API */ | |
14669 | + | |
14670 | +/* This is the structure, per node, returned from the membership request */ | |
14671 | +struct kcl_cluster_node { | |
14672 | + unsigned int size; | |
14673 | + unsigned int node_id; | |
14674 | + unsigned int us; | |
14675 | + unsigned int leave_reason; | |
14676 | + unsigned int incarnation; | |
14677 | + nodestate_t state; | |
14678 | + struct list_head list; | |
14679 | + char name[MAX_CLUSTER_MEMBER_NAME_LEN]; | |
14680 | + unsigned char votes; | |
14681 | +}; | |
14682 | + | |
14683 | +struct cluster_node_addr { | |
14684 | + struct list_head list; | |
14685 | + unsigned char addr[sizeof(struct sockaddr_in6)];/* A large sockaddr */ | |
14686 | + int addr_len; | |
14687 | +}; | |
14688 | + | |
14689 | + | |
14690 | +/* Reasons for a kernel membership callback */ | |
14691 | +typedef enum { CLUSTER_RECONFIG, DIED, LEAVING, NEWNODE } kcl_callback_reason; | |
14692 | + | |
14693 | +/* Kernel version of above, the void *sock is a struct socket */ | |
14694 | +struct kcl_multicast_sock { | |
14695 | + void *sock; | |
14696 | + int number; /* Socket number, to match up recvonly & bcast | |
14697 | + * sockets */ | |
14698 | +}; | |
14699 | + | |
14700 | +extern int kcl_sendmsg(struct socket *sock, void *buf, int size, | |
14701 | + struct sockaddr_cl *caddr, int addr_len, | |
14702 | + unsigned int flags); | |
14703 | +extern int kcl_register_read_callback(struct socket *sock, | |
14704 | + int (*routine) (char *, int, char *, int, | |
14705 | + unsigned int)); | |
14706 | +extern int kcl_add_callback(void (*callback) (kcl_callback_reason, long)); | |
14707 | +extern int kcl_remove_callback(void (*callback) (kcl_callback_reason, long)); | |
14708 | +extern int kcl_get_members(struct list_head *list); | |
14709 | +extern int kcl_get_member_ids(uint32_t * idbuf, int size); | |
14710 | +extern int kcl_get_all_members(struct list_head *list); | |
14711 | +extern int kcl_get_node_by_addr(unsigned char *addr, int addr_len, | |
14712 | + struct kcl_cluster_node *n); | |
14713 | +extern int kcl_get_node_by_name(unsigned char *name, | |
14714 | + struct kcl_cluster_node *n); | |
14715 | +extern int kcl_get_node_by_nodeid(int nodeid, struct kcl_cluster_node *n); | |
14716 | +extern int kcl_is_quorate(void); | |
14717 | +extern int kcl_addref_cluster(void); | |
14718 | +extern int kcl_releaseref_cluster(void); | |
14719 | +extern int kcl_cluster_name(char **cname); | |
14720 | +extern int kcl_get_current_interface(void); | |
14721 | +extern struct list_head *kcl_get_node_addresses(int nodeid); | |
14722 | + | |
14723 | +extern int kcl_barrier_register(char *name, unsigned int flags, | |
14724 | + unsigned int nodes); | |
14725 | +extern int kcl_barrier_setattr(char *name, unsigned int attr, | |
14726 | + unsigned long arg); | |
14727 | +extern int kcl_barrier_delete(char *name); | |
14728 | +extern int kcl_barrier_wait(char *name); | |
14729 | +extern int kcl_barrier_cancel(char *name); | |
14730 | + | |
14731 | +extern int kcl_register_quorum_device(char *name, int votes); | |
14732 | +extern int kcl_unregister_quorum_device(void); | |
14733 | +extern int kcl_quorum_device_available(int yesno); | |
14734 | + | |
14735 | +#endif | |
14736 | diff -urN linux-orig/include/cluster/service.h linux-patched/include/cluster/service.h | |
14737 | --- linux-orig/include/cluster/service.h 1970-01-01 07:30:00.000000000 +0730 | |
5cdbd17b | 14738 | +++ linux-patched/include/cluster/service.h 2004-06-29 20:07:50.000000000 +0800 |
4bf12011 | 14739 | @@ -0,0 +1,102 @@ |
14740 | +/****************************************************************************** | |
14741 | +******************************************************************************* | |
14742 | +** | |
14743 | +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
14744 | +** Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
14745 | +** | |
14746 | +** This copyrighted material is made available to anyone wishing to use, | |
14747 | +** modify, copy, or redistribute it subject to the terms and conditions | |
14748 | +** of the GNU General Public License v.2. | |
14749 | +** | |
14750 | +******************************************************************************* | |
14751 | +******************************************************************************/ | |
14752 | + | |
14753 | +#ifndef __SERVICE_DOT_H__ | |
14754 | +#define __SERVICE_DOT_H__ | |
14755 | + | |
14756 | +/* | |
14757 | + * Interface between service manager and services | |
14758 | + */ | |
14759 | + | |
14760 | +/* | |
14761 | + * Service levels are started in order from lowest, so level 0 is started on | |
14762 | + * all nodes before level 1 is started. | |
14763 | + */ | |
14764 | + | |
14765 | +#define SERVICE_LEVEL_FENCE (0) | |
14766 | +#define SERVICE_LEVEL_GDLM (1) | |
14767 | +#define SERVICE_LEVEL_GFS (2) | |
14768 | +#define SERVICE_LEVEL_USER (3) | |
14769 | + | |
14770 | +#define MAX_SERVICE_NAME_LEN (33) | |
14771 | + | |
14772 | +/* | |
14773 | + * The type of start a service receives. The start (and preceding stop) may be | |
14774 | + * due to a node joining or leaving the SG or due to a node having failed. | |
14775 | + */ | |
14776 | + | |
14777 | +#define SERVICE_NODE_FAILED (1) | |
14778 | +#define SERVICE_NODE_JOIN (2) | |
14779 | +#define SERVICE_NODE_LEAVE (3) | |
14780 | + | |
14781 | + | |
14782 | +struct kcl_service { | |
14783 | + struct list_head list; | |
14784 | + uint16_t level; | |
14785 | + uint32_t local_id; | |
14786 | + uint32_t global_id; | |
14787 | + int node_count; | |
14788 | + char name[MAX_SERVICE_NAME_LEN]; | |
14789 | +}; | |
14790 | + | |
14791 | +int kcl_get_services(struct list_head *list, int level); | |
14792 | + | |
14793 | + | |
14794 | +/* | |
14795 | + * These routines which run in CMAN context must return quickly and cannot | |
14796 | + * block. | |
14797 | + */ | |
14798 | + | |
14799 | +struct kcl_service_ops { | |
14800 | + int (*stop) (void *servicedata); | |
14801 | + int (*start) (void *servicedata, uint32_t *nodeids, int count, | |
14802 | + int event_id, int type); | |
14803 | + void (*finish) (void *servicedata, int event_id); | |
14804 | +}; | |
14805 | + | |
14806 | +/* | |
14807 | + * Register will cause CMAN to create a Service Group (SG) for the named | |
14808 | + * instance of the service. A local ID is returned which is used to join, | |
14809 | + * leave and unregister the service. | |
14810 | + */ | |
14811 | + | |
14812 | +int kcl_register_service(char *name, int namelen, int level, | |
14813 | + struct kcl_service_ops *ops, int unique, | |
14814 | + void *servicedata, uint32_t *local_id); | |
14815 | + | |
14816 | +void kcl_unregister_service(uint32_t local_id); | |
14817 | + | |
14818 | +/* | |
14819 | + * Once a service is joined it will be managed by CMAN and receive start, stop, | |
14820 | + * and finish calls. After leave is called the service is no longer managed by | |
14821 | + * CMAN. The first start for a service may arrive before kcl_join_service() | |
14822 | + * returns. | |
14823 | + */ | |
14824 | + | |
14825 | +int kcl_join_service(uint32_t local_id); | |
14826 | +int kcl_leave_service(uint32_t local_id); | |
14827 | + | |
14828 | +/* | |
14829 | + * After a service is started, it can ask for its cluster-wide unique ID. | |
14830 | + */ | |
14831 | + | |
14832 | +void kcl_global_service_id(uint32_t local_id, uint32_t * global_id); | |
14833 | + | |
14834 | +/* | |
14835 | + * Called by a service when it's done with a start(). Cannot be called from | |
14836 | + * the start function. | |
14837 | + */ | |
14838 | + | |
14839 | +void kcl_start_done(uint32_t local_id, int event_id); | |
14840 | + | |
14841 | +#endif |