summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rwxr-xr-xrun-tests.sh1
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-utils.c5
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-volume-ops.c22
3 files changed, 27 insertions, 1 deletions
diff --git a/run-tests.sh b/run-tests.sh
index b0a3c65..2b934f9 100755
--- a/run-tests.sh
+++ b/run-tests.sh
@@ -202,7 +202,6 @@ function is_bad_test ()
./tests/bugs/snapshot/bug-1109889.t \
./tests/bugs/distribute/bug-1066798.t \
./tests/bugs/glusterd/bug-1238706-daemons-stop-on-peer-cleanup.t \
- ./tests/bugs/glusterd/bug-948686.t \
./tests/geo-rep/georep-basic-dr-rsync.t \
./tests/geo-rep/georep-basic-dr-tarssh.t \
./tests/bugs/replicate/bug-1221481-allow-fops-on-dir-split-brain.t \
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index 8f8b3b3..a529fd3 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -3878,9 +3878,14 @@ glusterd_import_friend_volume (dict_t *peer_data, size_t count)
ret = glusterd_volinfo_find (new_volinfo->volname, &old_volinfo);
if (0 == ret) {
+ /* Ref count the old_volinfo such that deleting it doesn't crash
+ * if its been already in use by other thread
+ */
+ glusterd_volinfo_ref (old_volinfo);
(void) gd_check_and_update_rebalance_info (old_volinfo,
new_volinfo);
(void) glusterd_delete_stale_volume (old_volinfo, new_volinfo);
+ glusterd_volinfo_unref (old_volinfo);
}
if (glusterd_is_volume_started (new_volinfo)) {
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-ops.c b/xlators/mgmt/glusterd/src/glusterd-volume-ops.c
index 75f11e7..d997d60 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-ops.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-ops.c
@@ -1450,6 +1450,14 @@ glusterd_op_stage_start_volume (dict_t *dict, char **op_errstr,
volname);
goto out;
}
+ /* This is an incremental approach to have all the volinfo objects ref
+ * count. The first attempt is made in volume start transaction to
+ * ensure it doesn't race with import volume where stale volume is
+ * deleted. There are multiple instances of GlusterD crashing in
+ * bug-948686.t because of this. Once this approach is full proof, all
+ * other volinfo objects will be refcounted.
+ */
+ glusterd_volinfo_ref (volinfo);
ret = glusterd_validate_volume_id (dict, volinfo);
if (ret)
@@ -1580,6 +1588,9 @@ glusterd_op_stage_start_volume (dict_t *dict, char **op_errstr,
volinfo->caps = caps;
ret = 0;
out:
+ if (volinfo)
+ glusterd_volinfo_unref (volinfo);
+
if (ret && (msg[0] != '\0')) {
gf_msg (this->name, GF_LOG_ERROR, 0,
GD_MSG_OP_STAGE_START_VOL_FAIL, "%s", msg);
@@ -2463,6 +2474,14 @@ glusterd_op_start_volume (dict_t *dict, char **op_errstr)
volname);
goto out;
}
+ /* This is an incremental approach to have all the volinfo objects ref
+ * count. The first attempt is made in volume start transaction to
+ * ensure it doesn't race with import volume where stale volume is
+ * deleted. There are multiple instances of GlusterD crashing in
+ * bug-948686.t because of this. Once this approach is full proof, all
+ * other volinfo objects will be refcounted.
+ */
+ glusterd_volinfo_ref (volinfo);
/* A bricks mount dir is required only by snapshots which were
* introduced in gluster-3.6.0
@@ -2537,6 +2556,9 @@ glusterd_op_start_volume (dict_t *dict, char **op_errstr)
ret = glusterd_svcs_manager (volinfo);
out:
+ if (!volinfo)
+ glusterd_volinfo_unref (volinfo);
+
gf_msg_trace (this->name, 0, "returning %d ", ret);
return ret;
}