Blob Blame History Raw
autofs-5.0.7 - fix fix wildcard multi map regression

From: Ian Kent <raven@themaw.net>

A recent patch to fix a wildcard multi map mount regression has a
side effect of causing a deadlock at startup when trying to re-connect
to existing mounts.

The patch required the map entry cache write lock be taken so the cache
could be updated. But when starting and trying to re-connect to existing
mounts there's no need to update the cache.
---
 CHANGELOG                |    1 +
 modules/lookup_file.c    |   25 ++++++++++++++++++++-----
 modules/lookup_ldap.c    |   23 +++++++++++++++++++----
 modules/lookup_nisplus.c |   26 +++++++++++++++++++++-----
 modules/lookup_sss.c     |   22 ++++++++++++++++++----
 modules/lookup_yp.c      |   23 +++++++++++++++++++----
 6 files changed, 98 insertions(+), 22 deletions(-)

--- autofs-5.0.7.orig/CHANGELOG
+++ autofs-5.0.7/CHANGELOG
@@ -57,6 +57,7 @@
 - fix a couple of compiler warnings.
 - add after sssd dependency to unit file.
 - fix syncronize handle_mounts() shutdown.
+- fix fix wildcard multi map regression.
 
 25/07/2012 autofs-5.0.7
 =======================
--- autofs-5.0.7.orig/modules/lookup_file.c
+++ autofs-5.0.7/modules/lookup_file.c
@@ -1042,7 +1042,7 @@ int lookup_mount(struct autofs_point *ap
 			return NSS_STATUS_UNAVAIL;
 		}
 
-		cache_writelock(mc);
+		cache_readlock(mc);
 		me = cache_lookup_first(mc);
 		if (me && st.st_mtime <= me->age) {
 			/*
@@ -1084,7 +1084,18 @@ int lookup_mount(struct autofs_point *ap
 		}
 	}
 
-	cache_writelock(mc);
+	/*
+	 * We can't take the writelock for direct mounts. If we're
+	 * starting up or trying to re-connect to an existing direct
+	 * mount we could be iterating through the map entries with
+	 * the readlock held. But we don't need to update the cache
+	 * when we're starting up so just take the readlock in that
+	 * case.
+	 */
+	if (ap->flags & MOUNT_FLAG_REMOUNT)
+		cache_readlock(mc);
+	else
+		cache_writelock(mc);
 do_cache_lookup:
 	me = cache_lookup(mc, key);
 	/*
@@ -1102,10 +1113,11 @@ do_cache_lookup:
 	}
 	if (me && me->mapent) {
 		/*
-		 * Add wildcard match for later validation checks and
-		 * negative cache lookups.
+		 * If this is a lookup add wildcard match for later validation
+		 * checks and negative cache lookups.
 		 */
-		if (ap->type == LKP_INDIRECT && *me->key == '*') {
+		if (!(ap->flags & MOUNT_FLAG_REMOUNT) &&
+		    ap->type == LKP_INDIRECT && *me->key == '*') {
 			ret = cache_update(mc, source, key, me->mapent, me->age);
 			if (!(ret & (CHE_OK | CHE_UPDATED)))
 				me = NULL;
@@ -1130,6 +1142,9 @@ do_cache_lookup:
 	ret = ctxt->parse->parse_mount(ap, key, key_len,
 				       mapent, ctxt->parse->context);
 	if (ret) {
+		/* Don't update negative cache when re-connecting */
+		if (ap->flags & MOUNT_FLAG_REMOUNT)
+			return NSS_STATUS_TRYAGAIN;
 		cache_writelock(mc);
 		cache_update_negative(mc, source, key, ap->negative_timeout);
 		cache_unlock(mc);
--- autofs-5.0.7.orig/modules/lookup_ldap.c
+++ autofs-5.0.7/modules/lookup_ldap.c
@@ -2975,7 +2975,18 @@ int lookup_mount(struct autofs_point *ap
 			return status;
 	}
 
-	cache_writelock(mc);
+	/*
+	 * We can't take the writelock for direct mounts. If we're
+	 * starting up or trying to re-connect to an existing direct
+	 * mount we could be iterating through the map entries with
+	 * the readlock held. But we don't need to update the cache
+	 * when we're starting up so just take the readlock in that
+	 * case.
+	 */
+	if (ap->flags & MOUNT_FLAG_REMOUNT)
+		cache_readlock(mc);
+	else
+		cache_writelock(mc);
 	me = cache_lookup(mc, key);
 	/* Stale mapent => check for entry in alternate source or wildcard */
 	if (me && !me->mapent) {
@@ -2987,10 +2998,11 @@ int lookup_mount(struct autofs_point *ap
 	}
 	if (me && me->mapent) {
 		/*
-		 * Add wildcard match for later validation checks and
-		 * negative cache lookups.
+		 * If this is a lookup add wildcard match for later validation
+		 * checks and negative cache lookups.
 		 */
-		if (ap->type == LKP_INDIRECT && *me->key == '*') {
+		if (!(ap->flags & MOUNT_FLAG_REMOUNT) &&
+		    ap->type == LKP_INDIRECT && *me->key == '*') {
 			ret = cache_update(mc, source, key, me->mapent, me->age);
 			if (!(ret & (CHE_OK | CHE_UPDATED)))
 				me = NULL;
@@ -3012,6 +3024,9 @@ int lookup_mount(struct autofs_point *ap
 	ret = ctxt->parse->parse_mount(ap, key, key_len,
 				       mapent, ctxt->parse->context);
 	if (ret) {
+		/* Don't update negative cache when re-connecting */
+		if (ap->flags & MOUNT_FLAG_REMOUNT)
+			return NSS_STATUS_TRYAGAIN;
 		cache_writelock(mc);
 		cache_update_negative(mc, source, key, ap->negative_timeout);
 		cache_unlock(mc);
--- autofs-5.0.7.orig/modules/lookup_nisplus.c
+++ autofs-5.0.7/modules/lookup_nisplus.c
@@ -561,7 +561,18 @@ int lookup_mount(struct autofs_point *ap
 			return status;
 	}
 
-	cache_writelock(mc);
+	/*
+	 * We can't take the writelock for direct mounts. If we're
+	 * starting up or trying to re-connect to an existing direct
+	 * mount we could be iterating through the map entries with
+	 * the readlock held. But we don't need to update the cache
+	 * when we're starting up so just take the readlock in that
+	 * case.
+	 */
+	if (ap->flags & MOUNT_FLAG_REMOUNT)
+		cache_readlock(mc);
+	else
+		cache_writelock(mc);
 	me = cache_lookup(mc, key);
 	/* Stale mapent => check for entry in alternate source or wildcard */
 	if (me && !me->mapent) {
@@ -573,10 +584,11 @@ int lookup_mount(struct autofs_point *ap
 	}
 	if (me && me->mapent) {
 		/*
-		 * Add wildcard match for later validation checks and
-		 * negative cache lookups.
+		 * If this is a lookup add wildcard match for later validation
+		 * checks and negative cache lookups.
 		 */
-		if (ap->type == LKP_INDIRECT && *me->key == '*') {
+		if (!(ap->flags & MOUNT_FLAG_REMOUNT) &&
+		    ap->type == LKP_INDIRECT && *me->key == '*') {
 			ret = cache_update(mc, source, key, me->mapent, me->age);
 			if (!(ret & (CHE_OK | CHE_UPDATED)))
 				me = NULL;
@@ -603,6 +615,11 @@ int lookup_mount(struct autofs_point *ap
 		time_t now = time(NULL);
 		int rv = CHE_OK;
 
+		free(mapent);
+
+		/* Don't update negative cache when re-connecting */
+		if (ap->flags & MOUNT_FLAG_REMOUNT)
+			return NSS_STATUS_TRYAGAIN;
 		cache_writelock(mc);
 		me = cache_lookup_distinct(mc, key);
 		if (!me)
@@ -612,7 +629,6 @@ int lookup_mount(struct autofs_point *ap
 			me->status = time(NULL) + ap->negative_timeout;
 		}
 		cache_unlock(mc);
-		free(mapent);
 		return NSS_STATUS_TRYAGAIN;
 	}
 	free(mapent);
--- autofs-5.0.7.orig/modules/lookup_sss.c
+++ autofs-5.0.7/modules/lookup_sss.c
@@ -635,7 +635,17 @@ int lookup_mount(struct autofs_point *ap
 			return status;
 	}
 
-	cache_readlock(mc);
+	/*
+	 * We can't take the writelock for direct mounts. If we're
+	 * starting up or trying to re-connect to an existing direct
+	 * mount we could be iterating through the map entries with
+	 * the readlock held. But we don't need to update the cache
+	 * when we're starting up so just take the readlock in that
+	 */
+	if (ap->flags & MOUNT_FLAG_REMOUNT)
+		cache_writelock(mc);
+	else
+		cache_readlock(mc);
 	me = cache_lookup(mc, key);
 	/* Stale mapent => check for entry in alternate source or wildcard */
 	if (me && !me->mapent) {
@@ -647,10 +657,11 @@ int lookup_mount(struct autofs_point *ap
 	}
 	if (me && me->mapent) {
 		/*
-		 * Add wildcard match for later validation checks and
-		 * negative cache lookups.
+		 * If this is a lookup add wildcard match for later validation
+		 * checks and negative cache lookups.
 		 */
-		if (ap->type == LKP_INDIRECT && *me->key == '*') {
+		if (ap->type == LKP_INDIRECT && *me->key == '*' &&
+		   !(ap->flags & MOUNT_FLAG_REMOUNT)) {
 			ret = cache_update(mc, source, key, me->mapent, me->age);
 			if (!(ret & (CHE_OK | CHE_UPDATED)))
 				me = NULL;
@@ -672,6 +683,9 @@ int lookup_mount(struct autofs_point *ap
 	ret = ctxt->parse->parse_mount(ap, key, key_len,
 				       mapent, ctxt->parse->context);
 	if (ret) {
+		/* Don't update negative cache when re-connecting */
+		if (ap->flags & MOUNT_FLAG_REMOUNT)
+			return NSS_STATUS_TRYAGAIN;
 		cache_writelock(mc);
 		cache_update_negative(mc, source, key, ap->negative_timeout);
 		cache_unlock(mc);
--- autofs-5.0.7.orig/modules/lookup_yp.c
+++ autofs-5.0.7/modules/lookup_yp.c
@@ -662,7 +662,18 @@ int lookup_mount(struct autofs_point *ap
 			return status;
 	}
 
-	cache_writelock(mc);
+	/*
+	 * We can't take the writelock for direct mounts. If we're
+	 * starting up or trying to re-connect to an existing direct
+	 * mount we could be iterating through the map entries with
+	 * the readlock held. But we don't need to update the cache
+	 * when we're starting up so just take the readlock in that
+	 * case.
+	 */
+	if (ap->flags & MOUNT_FLAG_REMOUNT)
+		cache_readlock(mc);
+	else
+		cache_writelock(mc);
 	me = cache_lookup(mc, key);
 	/* Stale mapent => check for entry in alternate source or wildcard */
 	if (me && !me->mapent) {
@@ -674,10 +685,11 @@ int lookup_mount(struct autofs_point *ap
 	}
 	if (me && me->mapent) {
 		/*
-		 * Add wildcard match for later validation checks and
-		 * negative cache lookups.
+		 * If this is a lookup add wildcard match for later validation
+		 * checks and negative cache lookups.
 		 */
-		if (ap->type == LKP_INDIRECT && *me->key == '*') {
+		if (ap->type == LKP_INDIRECT && *me->key == '*' &&
+		   !(ap->flags & MOUNT_FLAG_REMOUNT)) {
 			ret = cache_update(mc, source, key, me->mapent, me->age);
 			if (!(ret & (CHE_OK | CHE_UPDATED)))
 				me = NULL;
@@ -698,6 +710,9 @@ int lookup_mount(struct autofs_point *ap
 		ret = ctxt->parse->parse_mount(ap, key, key_len,
 					       mapent, ctxt->parse->context);
 		if (ret) {
+			/* Don't update negative cache when re-connecting */
+			if (ap->flags & MOUNT_FLAG_REMOUNT)
+				return NSS_STATUS_TRYAGAIN;
 			cache_writelock(mc);
 			cache_update_negative(mc, source, key, ap->negative_timeout);
 			cache_unlock(mc);