diff options
| author | Anand Avati <avati@gluster.com> | 2010-01-23 05:14:31 +0000 |
|---|---|---|
| committer | Anand V. Avati <avati@dev.gluster.com> | 2010-01-23 02:58:34 -0800 |
| commit | a0b148ea4e2a0163548eeb89b7580be4adbb8070 (patch) | |
| tree | 54ecf54695c9a891bb13d9542bf80fe443b4c725 /xlators/protocol/server/src/server-protocol.c | |
| parent | 9c53d5daf403f6fbfde76dec121295a4c156b32e (diff) | |
Server backend storage hang should not cause the mount point to hang.
Submitted-by: Krishna Srinivas <krishna@gluster.com>
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 <krishna@gluster.com>
Signed-off-by: Anand V. Avati <avati@blackhole.gluster.com>
Signed-off-by: Anand V. Avati <avati@dev.gluster.com>
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
Diffstat (limited to 'xlators/protocol/server/src/server-protocol.c')
| -rw-r--r-- | xlators/protocol/server/src/server-protocol.c | 39 |
1 files changed, 37 insertions, 2 deletions
diff --git a/xlators/protocol/server/src/server-protocol.c b/xlators/protocol/server/src/server-protocol.c index 0528699ac..a7f6294f3 100644 --- a/xlators/protocol/server/src/server-protocol.c +++ b/xlators/protocol/server/src/server-protocol.c @@ -5654,6 +5654,8 @@ mop_setvolume (call_frame_t *frame, xlator_t *bound_xl, char *volfile_key = NULL; uint32_t checksum = 0; int32_t lru_limit = 1024; + xlator_list_t *xltrav = NULL; + int subvol_idx = 0; params = dict_new (); reply = dict_new (); @@ -5695,7 +5697,7 @@ mop_setvolume (call_frame_t *frame, xlator_t *bound_xl, } - conn = server_connection_get (frame->this, process_uuid); + conn = server_connection_get (frame->this, process_uuid, trans); if (trans->xl_private != conn) trans->xl_private = conn; @@ -5862,6 +5864,22 @@ mop_setvolume (call_frame_t *frame, xlator_t *bound_xl, ret = dict_set_uint64 (reply, "transport-ptr", ((uint64_t) (long) trans)); + xltrav = frame->this->children; + while (xltrav) { + if (xltrav->xlator == xl) + break; + xltrav = xltrav->next; + subvol_idx++; + } + + if (conf->subvol_list[subvol_idx] == 0) { + gf_log (xl->name, GF_LOG_DEBUG, + "subvolume %d down (filesystem not accesible), failed to setvolume", subvol_idx); + op_ret = -1; + op_errno = ENOTCONN; + goto fail; + } + fail: dict_len = dict_serialized_length (reply); if (dict_len < 0) { @@ -6550,6 +6568,8 @@ init (xlator_t *this) server_conf_t *conf = NULL; data_t *data = NULL; data_t *trace = NULL; + int i = 0; + xlator_list_t *xltrav = NULL; if (this->children == NULL) { gf_log (this->name, GF_LOG_ERROR, @@ -6635,6 +6655,15 @@ init (xlator_t *this) } } + xltrav = this->children; + + while (xltrav) { + i++; + xltrav = xltrav->next; + } + + conf->subvol_list = calloc (i, sizeof (char)); + #ifndef GF_DARWIN_HOST_OS { struct rlimit lim; @@ -6736,6 +6765,12 @@ notify (xlator_t *this, int32_t event, void *data, ...) } switch (event) { + case GF_EVENT_CHILD_DOWN: + server_child_down (this, data); + break; + case GF_EVENT_CHILD_UP: + server_child_up (this, data); + break; case GF_EVENT_POLLIN: ret = protocol_server_pollin (this, trans); break; @@ -6756,7 +6791,7 @@ notify (xlator_t *this, int32_t event, void *data, ...) * FIXME: shouldn't we check for return value? * what should be done if cleanup fails? */ - server_connection_cleanup (this, trans->xl_private); + server_connection_cleanup (this, trans->xl_private, trans); } } break; |