From a0b148ea4e2a0163548eeb89b7580be4adbb8070 Mon Sep 17 00:00:00 2001 From: Anand Avati Date: Sat, 23 Jan 2010 05:14:31 +0000 Subject: Server backend storage hang should not cause the mount point to hang. Submitted-by: Krishna Srinivas NOTE: fixed compilation issues in posix.c introduced while merging storage/posix polls for FS/kernel being functional by issuing statvfs() call. In case statvfs expires the timer, storage/posix will send CHILD_DOWN to upper translator. Ultimately this will cause protocol/server to disconnect all clients connected and also cleans up the data structures. Hence if soft lockup or other kernel bug causes backend FS to hang, the clients will not be hung. Signed-off-by: Krishna Srinivas Signed-off-by: Anand V. Avati Signed-off-by: Anand V. Avati BUG: 272 (Server backend storage hang should not cause the mount point to hang) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=272 --- xlators/protocol/server/src/server-helpers.c | 112 ++++++++++++++++++++++++++- 1 file changed, 110 insertions(+), 2 deletions(-) (limited to 'xlators/protocol/server/src/server-helpers.c') diff --git a/xlators/protocol/server/src/server-helpers.c b/xlators/protocol/server/src/server-helpers.c index 88dada1516e..0e22f0ba5fa 100644 --- a/xlators/protocol/server/src/server-helpers.c +++ b/xlators/protocol/server/src/server-helpers.c @@ -645,13 +645,14 @@ out: int -server_connection_cleanup (xlator_t *this, server_connection_t *conn) +server_connection_cleanup (xlator_t *this, server_connection_t *conn, transport_t *trans) { char do_cleanup = 0; struct _lock_table *ltable = NULL; fdentry_t *fdentries = NULL; uint32_t fd_count = 0; int ret = 0; + int i = 0; if (conn == NULL) { goto out; @@ -659,6 +660,12 @@ server_connection_cleanup (xlator_t *this, server_connection_t *conn) pthread_mutex_lock (&conn->lock); { + for (i = 0; i < TRANSPORTS_PER_SERVER_CONN; i++) { + if (conn->transports[i] == trans) { + conn->transports[i] = NULL; + transport_unref (trans); + } + } conn->active_transports--; if (conn->active_transports == 0) { if (conn->ltable) { @@ -850,11 +857,12 @@ out: server_connection_t * -server_connection_get (xlator_t *this, const char *id) +server_connection_get (xlator_t *this, const char *id, transport_t *trans) { server_connection_t *conn = NULL; server_connection_t *trav = NULL; server_conf_t *conf = NULL; + int i = 0; conf = this->private; @@ -878,10 +886,30 @@ server_connection_get (xlator_t *this, const char *id) list_add (&conn->list, &conf->conns); } + if (conn->active_transports == TRANSPORTS_PER_SERVER_CONN) { + gf_log (this->name, GF_LOG_DEBUG, + "Maximum number of connections allowed is %d", + TRANSPORTS_PER_SERVER_CONN); + goto unlock; + } + + for (i = 0; i < TRANSPORTS_PER_SERVER_CONN; i++) { + if (!conn->transports[i]) + break; + } + + if (i == TRANSPORTS_PER_SERVER_CONN) { + gf_log (this->name, GF_LOG_DEBUG, + "Could not find a vacant slot"); + goto unlock; + } + + conn->transports[i] = transport_ref (trans); conn->ref++; conn->active_transports++; } +unlock: pthread_mutex_unlock (&conf->mutex); return conn; @@ -918,3 +946,83 @@ server_connection_put (xlator_t *this, server_connection_t *conn) out: return; } + +void +server_child_down (xlator_t *this, xlator_t *bound_xl) +{ + server_conf_t *conf = NULL; + server_connection_t *trav = NULL; + transport_t *trans = NULL; + int subvol_idx = 0; + int i = 0; + xlator_list_t *xltrav = NULL; + + conf = this->private; + + if (conf == NULL) + return; + + xltrav = this->children; + + while (xltrav) { + if (xltrav->xlator == bound_xl) + break; + xltrav = xltrav->next; + subvol_idx++; + } + gf_log (this->name, GF_LOG_DEBUG, + "subvolume %s(%d) went down", bound_xl->name, subvol_idx); + + conf->subvol_list[subvol_idx] = 0; + + pthread_mutex_lock (&conf->mutex); + { + if (!list_empty(&conf->conns)) { + list_for_each_entry (trav, &conf->conns, list) { + if (bound_xl == trav->bound_xl) { + gf_log (this->name, GF_LOG_DEBUG, + "disonnecting conn=%p", trav); + for (i = 0; i < TRANSPORTS_PER_SERVER_CONN; i++) + { + trans = trav->transports[i]; + if (trans == NULL) + continue; + gf_log (this->name, GF_LOG_DEBUG, + "disconnecting %p(%d)", + trans, i); + transport_disconnect (trans); + } + } + } + } + } + pthread_mutex_unlock (&conf->mutex); +} + +void +server_child_up (xlator_t *this, xlator_t *bound_xl) +{ + server_conf_t *conf = NULL; + int subvol_idx = 0; + xlator_list_t *xltrav = NULL; + + conf = this->private; + + if (conf == NULL) + return; + + xltrav = this->children; + + while (xltrav) { + if (bound_xl == xltrav->xlator) { + break; + } + subvol_idx++; + xltrav = xltrav->next; + } + + gf_log (this->name, GF_LOG_DEBUG, + "subvolume %s(%d) came up", bound_xl->name, subvol_idx); + + conf->subvol_list[subvol_idx] = 1; +} -- cgit