[packages/autofs.git] / autofs-5.0.4-fix-select-fd-limit.patch

autofs-5.0.4 - fix select fd limit

From: Ian Kent <raven@themaw.net>

Using a large number of direct mounts causes autofs to hang. This is
because select(2) is limited to 1024 file handles (usually). To resolve
this we need to convert calls to select(2) to use poll(2).
---

 CHANGELOG                |    1 +
 daemon/spawn.c           |   19 ++++++------
 lib/rpc_subs.c           |   32 ++++++++++++++------
 modules/lookup_program.c |   74 +++++++++++++++++++++++++++-------------------
 4 files changed, 78 insertions(+), 48 deletions(-)


diff --git a/CHANGELOG b/CHANGELOG
index 43f3205..0fb7db5 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -4,6 +4,7 @@
 - fix nested submount expire deadlock.
 - fix negative caching for non-existent map keys.
 - use CLOEXEC flag.
+- fix select(2) fd limit.
 
 4/11/2008 autofs-5.0.4
 -----------------------
diff --git a/daemon/spawn.c b/daemon/spawn.c
index 4ddf46f..e02d926 100644
--- a/daemon/spawn.c
+++ b/daemon/spawn.c
@@ -21,6 +21,7 @@
 #include <sys/types.h>
 #include <dirent.h>
 #include <time.h>
+#include <poll.h>
 #include <sys/wait.h>
 #include <sys/stat.h>
 #include <sys/mount.h>
@@ -89,21 +90,21 @@ void reset_signals(void)
 
 static int timed_read(int pipe, char *buf, size_t len, int time)
 {
-	struct timeval timeout = { 0, 0 };
-	struct timeval *tout = NULL;
-	fd_set wset, rset;
+	struct pollfd pfd[1];
+	int timeout = time;
 	int ret;
 
-	FD_ZERO(&rset);
-	FD_SET(pipe, &rset);
-	wset = rset;
+	pfd[0].fd = pipe;
+	pfd[0].events = POLLIN;
 
 	if (time != -1) {
-		timeout.tv_sec = time;
-		tout = &timeout;
+		if (time >= (INT_MAX - 1)/1000)
+			timeout = INT_MAX - 1;
+		else
+			timeout = time * 1000;
 	}
 
-	ret = select(pipe + 1, &rset, &wset, NULL, tout);
+	ret = poll(pfd, 1, timeout);
 	if (ret <= 0) {
 		if (ret == 0)
 			ret = -ETIMEDOUT;
diff --git a/lib/rpc_subs.c b/lib/rpc_subs.c
index 9ac3657..7b347a7 100644
--- a/lib/rpc_subs.c
+++ b/lib/rpc_subs.c
@@ -31,6 +31,7 @@
 #include <sys/ioctl.h>
 #include <ctype.h>
 #include <pthread.h>
+#include <poll.h>
 
 #include "mount.h"
 #include "rpc_subs.h"
@@ -197,14 +198,15 @@ void rpc_destroy_udp_client(struct conn_info *info)
 /*
  *  Perform a non-blocking connect on the socket fd.
  *
- *  tout contains the timeout.  It will be modified to contain the time
- *  remaining (i.e. time provided - time elasped).
+ *  The input struct timeval always has tv_nsec set to zero,
+ *  we only ever use tv_sec for timeouts.
  */
 static int connect_nb(int fd, struct sockaddr_in *addr, struct timeval *tout)
 {
+	struct pollfd pfd[1];
+	int timeout = tout->tv_sec;
 	int flags, ret;
 	socklen_t len;
-	fd_set wset, rset;
 
 	flags = fcntl(fd, F_GETFL, 0);
 	if (flags < 0)
@@ -229,12 +231,10 @@ static int connect_nb(int fd, struct sockaddr_in *addr, struct timeval *tout)
 	if (ret == 0)
 		goto done;
 
-	/* now wait */
-	FD_ZERO(&rset);
-	FD_SET(fd, &rset);
-	wset = rset;
+	pfd[0].fd = fd;
+	pfd[0].events = POLLOUT;
 
-	ret = select(fd + 1, &rset, &wset, NULL, tout);
+	ret = poll(pfd, 1, timeout);
 	if (ret <= 0) {
 		if (ret == 0)
 			ret = -ETIMEDOUT;
@@ -243,13 +243,27 @@ static int connect_nb(int fd, struct sockaddr_in *addr, struct timeval *tout)
 		goto done;
 	}
 
-	if (FD_ISSET(fd, &rset) || FD_ISSET(fd, &wset)) {
+	if (pfd[0].revents) {
 		int status;
 
 		len = sizeof(ret);
 		status = getsockopt(fd, SOL_SOCKET, SO_ERROR, &ret, &len);
 		if (status < 0) {
+			char buf[MAX_ERR_BUF + 1];
+			char *estr = strerror_r(errno, buf, MAX_ERR_BUF);
+
+			/*
+			 * We assume getsockopt amounts to a read on the
+			 * descriptor and gives us the errno we need for
+			 * the POLLERR revent case.
+			 */
 			ret = -errno;
+
+			/* Unexpected case, log it so we know we got caught */
+			if (pfd[0].revents & POLLNVAL)
+				logerr("unexpected poll(2) error on connect:"
+				       " %s", estr);
+
 			goto done;
 		}
 
diff --git a/modules/lookup_program.c b/modules/lookup_program.c
index 6f4e2a3..f62d3ef 100644
--- a/modules/lookup_program.c
+++ b/modules/lookup_program.c
@@ -24,6 +24,7 @@
 #include <sys/times.h>
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <poll.h>
 
 #define MODULE_LOOKUP
 #include "automount.h"
@@ -113,14 +114,12 @@ int lookup_mount(struct autofs_point *ap, const char *name, int name_len, void *
 	char errbuf[1024], *errp;
 	char ch;
 	int pipefd[2], epipefd[2];
+	struct pollfd pfd[2];
 	pid_t f;
-	int files_left;
 	int status;
-	fd_set readfds, ourfds;
 	enum state { st_space, st_map, st_done } state;
 	int quoted = 0;
 	int ret = 1;
-	int max_fd;
 	int distance;
 	int alloci = 1;
 
@@ -253,30 +252,39 @@ int lookup_mount(struct autofs_point *ap, const char *name, int name_len, void *
 	errp = errbuf;
 	state = st_space;
 
-	FD_ZERO(&ourfds);
-	FD_SET(pipefd[0], &ourfds);
-	FD_SET(epipefd[0], &ourfds);
+	pfd[0].fd = pipefd[0];
+	pfd[0].events = POLLIN;
+	pfd[1].fd = epipefd[0];
+	pfd[1].events = POLLIN;
 
-	max_fd = pipefd[0] > epipefd[0] ? pipefd[0] : epipefd[0];
+	while (1) {
+		int bytes;
 
-	files_left = 2;
+		if (poll(pfd, 2, -1) < 0 && errno != EINTR)
+			break;
+
+		if (pfd[0].fd == -1 && pfd[1].fd == -1)
+			break;
 
-	while (files_left != 0) {
-		readfds = ourfds;
-		if (select(max_fd + 1, &readfds, NULL, NULL, NULL) < 0 && errno != EINTR)
+		if ((pfd[0].revents & (POLLIN|POLLHUP)) == POLLHUP &&
+		    (pfd[1].revents & (POLLIN|POLLHUP)) == POLLHUP)
 			break;
 
 		/* Parse maps from stdout */
-		if (FD_ISSET(pipefd[0], &readfds)) {
-			if (read(pipefd[0], &ch, 1) < 1) {
-				FD_CLR(pipefd[0], &ourfds);
-				files_left--;
+		if (pfd[0].revents) {
+cont:
+			bytes = read(pipefd[0], &ch, 1);
+			if (bytes == 0)
+				goto next;
+			else if (bytes < 0) {
+				pfd[0].fd = -1;
 				state = st_done;
+				goto next;
 			}
 
 			if (!quoted && ch == '\\') {
 				quoted = 1;
-				continue;
+				goto cont;
 			}
 
 			switch (state) {
@@ -333,26 +341,32 @@ int lookup_mount(struct autofs_point *ap, const char *name, int name_len, void *
 				/* Eat characters till there's no more output */
 				break;
 			}
+			goto cont;
 		}
 		quoted = 0;
-
+next:
 		/* Deal with stderr */
-		if (FD_ISSET(epipefd[0], &readfds)) {
-			if (read(epipefd[0], &ch, 1) < 1) {
-				FD_CLR(epipefd[0], &ourfds);
-				files_left--;
-			} else if (ch == '\n') {
-				*errp = '\0';
-				if (errbuf[0])
-					logmsg(">> %s", errbuf);
-				errp = errbuf;
-			} else {
-				if (errp >= &errbuf[1023]) {
+		if (pfd[1].revents) {
+			while (1) {
+				bytes = read(epipefd[0], &ch, 1);
+				if (bytes == 0)
+					break;
+				else if (bytes < 0) {
+					pfd[1].fd = -1;
+					break;
+				} else if (ch == '\n') {
 					*errp = '\0';
-					logmsg(">> %s", errbuf);
+					if (errbuf[0])
+						logmsg(">> %s", errbuf);
 					errp = errbuf;
+				} else {
+					if (errp >= &errbuf[1023]) {
+						*errp = '\0';
+						logmsg(">> %s", errbuf);
+						errp = errbuf;
+					}
+					*(errp++) = ch;
 				}
-				*(errp++) = ch;
 			}
 		}
 	}
Commit	Line	Data
e5fd101c PS	1	autofs-5.0.4 - fix select fd limit
	2
	3	From: Ian Kent <raven@themaw.net>
	4
	5	Using a large number of direct mounts causes autofs to hang. This is
	6	because select(2) is limited to 1024 file handles (usually). To resolve
	7	this we need to convert calls to select(2) to use poll(2).
	8	---
	9
	10	CHANGELOG \| 1 +
	11	daemon/spawn.c \| 19 ++++++------
	12	lib/rpc_subs.c \| 32 ++++++++++++++------
	13	modules/lookup_program.c \| 74 +++++++++++++++++++++++++++-------------------
	14	4 files changed, 78 insertions(+), 48 deletions(-)
	15
	16
	17	diff --git a/CHANGELOG b/CHANGELOG
	18	index 43f3205..0fb7db5 100644
	19	--- a/CHANGELOG
	20	+++ b/CHANGELOG
	21	@@ -4,6 +4,7 @@
	22	- fix nested submount expire deadlock.
	23	- fix negative caching for non-existent map keys.
	24	- use CLOEXEC flag.
	25	+- fix select(2) fd limit.
	26
	27	4/11/2008 autofs-5.0.4
	28	-----------------------
	29	diff --git a/daemon/spawn.c b/daemon/spawn.c
	30	index 4ddf46f..e02d926 100644
	31	--- a/daemon/spawn.c
	32	+++ b/daemon/spawn.c
	33	@@ -21,6 +21,7 @@
	34	#include <sys/types.h>
	35	#include <dirent.h>
	36	#include <time.h>
	37	+#include <poll.h>
	38	#include <sys/wait.h>
	39	#include <sys/stat.h>
	40	#include <sys/mount.h>
	41	@@ -89,21 +90,21 @@ void reset_signals(void)
	42
	43	static int timed_read(int pipe, char *buf, size_t len, int time)
	44	{
	45	- struct timeval timeout = { 0, 0 };
	46	- struct timeval *tout = NULL;
	47	- fd_set wset, rset;
	48	+ struct pollfd pfd[1];
	49	+ int timeout = time;
	50	int ret;
	51
	52	- FD_ZERO(&rset);
	53	- FD_SET(pipe, &rset);
	54	- wset = rset;
	55	+ pfd[0].fd = pipe;
	56	+ pfd[0].events = POLLIN;
	57
	58	if (time != -1) {
	59	- timeout.tv_sec = time;
	60	- tout = &timeout;
	61	+ if (time >= (INT_MAX - 1)/1000)
	62	+ timeout = INT_MAX - 1;
	63	+ else
	64	+ timeout = time * 1000;
65	}
66
67	- ret = select(pipe + 1, &rset, &wset, NULL, tout);
68	+ ret = poll(pfd, 1, timeout);
69	if (ret <= 0) {
70	if (ret == 0)
71	ret = -ETIMEDOUT;
72	diff --git a/lib/rpc_subs.c b/lib/rpc_subs.c
73	index 9ac3657..7b347a7 100644
74	--- a/lib/rpc_subs.c
75	+++ b/lib/rpc_subs.c
76	@@ -31,6 +31,7 @@
77	#include <sys/ioctl.h>
78	#include <ctype.h>
79	#include <pthread.h>
80	+#include <poll.h>
81
82	#include "mount.h"
83	#include "rpc_subs.h"
84	@@ -197,14 +198,15 @@ void rpc_destroy_udp_client(struct conn_info *info)
85	/*
86	* Perform a non-blocking connect on the socket fd.
87	*
88	- * tout contains the timeout. It will be modified to contain the time
89	- * remaining (i.e. time provided - time elasped).
90	+ * The input struct timeval always has tv_nsec set to zero,
91	+ * we only ever use tv_sec for timeouts.
92	*/
93	static int connect_nb(int fd, struct sockaddr_in addr, struct timeval tout)
94	{
95	+ struct pollfd pfd[1];
96	+ int timeout = tout->tv_sec;
97	int flags, ret;
98	socklen_t len;
99	- fd_set wset, rset;
100
101	flags = fcntl(fd, F_GETFL, 0);
102	if (flags < 0)
103	@@ -229,12 +231,10 @@ static int connect_nb(int fd, struct sockaddr_in addr, struct timeval tout)
104	if (ret == 0)
105	goto done;
106
107	- /* now wait */
108	- FD_ZERO(&rset);
109	- FD_SET(fd, &rset);
110	- wset = rset;
111	+ pfd[0].fd = fd;
112	+ pfd[0].events = POLLOUT;
113
114	- ret = select(fd + 1, &rset, &wset, NULL, tout);
115	+ ret = poll(pfd, 1, timeout);
116	if (ret <= 0) {
117	if (ret == 0)
118	ret = -ETIMEDOUT;
119	@@ -243,13 +243,27 @@ static int connect_nb(int fd, struct sockaddr_in addr, struct timeval tout)
120	goto done;
121	}
122
123	- if (FD_ISSET(fd, &rset) \|\| FD_ISSET(fd, &wset)) {
124	+ if (pfd[0].revents) {
125	int status;
126
127	len = sizeof(ret);
128	status = getsockopt(fd, SOL_SOCKET, SO_ERROR, &ret, &len);
129	if (status < 0) {
130	+ char buf[MAX_ERR_BUF + 1];
131	+ char *estr = strerror_r(errno, buf, MAX_ERR_BUF);
132	+
133	+ /*
134	+ * We assume getsockopt amounts to a read on the
135	+ * descriptor and gives us the errno we need for
136	+ * the POLLERR revent case.
137	+ */
138	ret = -errno;
139	+
140	+ /* Unexpected case, log it so we know we got caught */
141	+ if (pfd[0].revents & POLLNVAL)
142	+ logerr("unexpected poll(2) error on connect:"
143	+ " %s", estr);
144	+
145	goto done;
146	}
147
148	diff --git a/modules/lookup_program.c b/modules/lookup_program.c
149	index 6f4e2a3..f62d3ef 100644
150	--- a/modules/lookup_program.c
151	+++ b/modules/lookup_program.c
152	@@ -24,6 +24,7 @@
153	#include <sys/times.h>
154	#include <sys/types.h>
155	#include <sys/wait.h>
156	+#include <poll.h>
157
158	#define MODULE_LOOKUP
159	#include "automount.h"
160	@@ -113,14 +114,12 @@ int lookup_mount(struct autofs_point ap, const char name, int name_len, void *
161	char errbuf[1024], *errp;
162	char ch;
163	int pipefd[2], epipefd[2];
164	+ struct pollfd pfd[2];
165	pid_t f;
166	- int files_left;
167	int status;
168	- fd_set readfds, ourfds;
169	enum state { st_space, st_map, st_done } state;
170	int quoted = 0;
171	int ret = 1;
172	- int max_fd;
173	int distance;
174	int alloci = 1;
175
176	@@ -253,30 +252,39 @@ int lookup_mount(struct autofs_point ap, const char name, int name_len, void *
177	errp = errbuf;
178	state = st_space;
179
180	- FD_ZERO(&ourfds);
181	- FD_SET(pipefd[0], &ourfds);
182	- FD_SET(epipefd[0], &ourfds);
183	+ pfd[0].fd = pipefd[0];
184	+ pfd[0].events = POLLIN;
185	+ pfd[1].fd = epipefd[0];
186	+ pfd[1].events = POLLIN;
187
188	- max_fd = pipefd[0] > epipefd[0] ? pipefd[0] : epipefd[0];
189	+ while (1) {
190	+ int bytes;
191
192	- files_left = 2;
193	+ if (poll(pfd, 2, -1) < 0 && errno != EINTR)
194	+ break;
195	+
196	+ if (pfd[0].fd == -1 && pfd[1].fd == -1)
197	+ break;
198
199	- while (files_left != 0) {
200	- readfds = ourfds;
201	- if (select(max_fd + 1, &readfds, NULL, NULL, NULL) < 0 && errno != EINTR)
202	+ if ((pfd[0].revents & (POLLIN\|POLLHUP)) == POLLHUP &&
203	+ (pfd[1].revents & (POLLIN\|POLLHUP)) == POLLHUP)
204	break;
205
206	/* Parse maps from stdout */
207	- if (FD_ISSET(pipefd[0], &readfds)) {
208	- if (read(pipefd[0], &ch, 1) < 1) {
209	- FD_CLR(pipefd[0], &ourfds);
210	- files_left--;
211	+ if (pfd[0].revents) {
212	+cont:
213	+ bytes = read(pipefd[0], &ch, 1);
214	+ if (bytes == 0)
215	+ goto next;
216	+ else if (bytes < 0) {
217	+ pfd[0].fd = -1;
218	state = st_done;
219	+ goto next;
220	}
221
222	if (!quoted && ch == '\\') {
223	quoted = 1;
224	- continue;
225	+ goto cont;
226	}
227
228	switch (state) {
229	@@ -333,26 +341,32 @@ int lookup_mount(struct autofs_point ap, const char name, int name_len, void *
230	/* Eat characters till there's no more output */
231	break;
232	}
233	+ goto cont;
234	}
235	quoted = 0;
236	-
237	+next:
238	/* Deal with stderr */
239	- if (FD_ISSET(epipefd[0], &readfds)) {
240	- if (read(epipefd[0], &ch, 1) < 1) {
241	- FD_CLR(epipefd[0], &ourfds);
242	- files_left--;
243	- } else if (ch == '\n') {
244	- *errp = '\0';
245	- if (errbuf[0])
246	- logmsg(">> %s", errbuf);
247	- errp = errbuf;
248	- } else {
249	- if (errp >= &errbuf[1023]) {
250	+ if (pfd[1].revents) {
251	+ while (1) {
252	+ bytes = read(epipefd[0], &ch, 1);
253	+ if (bytes == 0)
254	+ break;
255	+ else if (bytes < 0) {
256	+ pfd[1].fd = -1;
257	+ break;
258	+ } else if (ch == '\n') {
259	*errp = '\0';
260	- logmsg(">> %s", errbuf);
261	+ if (errbuf[0])
262	+ logmsg(">> %s", errbuf);
263	errp = errbuf;
264	+ } else {
265	+ if (errp >= &errbuf[1023]) {
266	+ *errp = '\0';
267	+ logmsg(">> %s", errbuf);
268	+ errp = errbuf;
269	+ }
270	+ *(errp++) = ch;
271	}
272	- *(errp++) = ch;
273	}
274	}
275	}