From f01f2174bcc27c06c8cd5e64cb6e63452fe6c07b Mon Sep 17 00:00:00 2001
From: Nathaniel Clark <nathaniel.l.clark@intel.com>
Date: Mon, 16 Apr 2018 17:10:49 -0400
Subject: [PATCH] LU-10843 mgs: allow snapshot after MGS remount
If the MGS is unmounted/mounted without the MDS reconnecting,
the fsdb FSNAME-barrier would not be created.
This change allows mgs_barrier_freeze (called from snapshot_create)
to attempt the create the required fsdb.
This adds a test to sanity-lsnapshot.sh for this issue.
Test-Parameters: testlist=sanity-lsnapshot mgtfilesystemtype=zfs ostfilesystemtype=zfs mdtfilesystemtype=zfs combinedmdsmgs=false ostcount=2 standalonemgs=true
Lustre-change: https://review.whamcloud.com/31779
Lustre-commit: 2d63d5bc74a5b2477d795bf9e9b0eba94f7fc19f
Signed-off-by: Nathaniel Clark <nathaniel.l.clark@intel.com>
Change-Id: I2432cc0bdaddb07f024744065ca2ced77288fd7b
Reviewed-by: Fan Yong <fan.yong@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Jian Yu <jian.yu@intel.com>
Reviewed-by: Mike Pershin <mike.pershin@intel.com>
Signed-off-by: Minh Diep <minh.diep@intel.com>
Reviewed-on: https://review.whamcloud.com/32519
Reviewed-by: Nathaniel Clark <nclark@whamcloud.com>
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Jian Yu <yujian@whamcloud.com>
Reviewed-by: John L. Hammond <jhammond@whamcloud.com>
---
lustre/mgs/mgs_barrier.c | 83 +++++++++++++++++++++++++++-----
lustre/mgs/mgs_internal.h | 3 ++
lustre/mgs/mgs_llog.c | 21 ++++++--
lustre/tests/sanity-lsnapshot.sh | 35 ++++++++++++++
4 files changed, 125 insertions(+), 17 deletions(-)
diff --git a/lustre/mgs/mgs_barrier.c b/lustre/mgs/mgs_barrier.c
index e661063c51..7da6fb9854 100644
--- a/lustre/mgs/mgs_barrier.c
+++ b/lustre/mgs/mgs_barrier.c
@@ -304,17 +304,36 @@ static int mgs_barrier_freeze(const struct lu_env *env,
down_write(&mgs->mgs_barrier_rwsem);
mutex_lock(&mgs->mgs_mutex);
- fsdb = mgs_find_fsdb(mgs, name);
- if (!fsdb) {
+ rc = mgs_find_or_make_fsdb_nolock(env, mgs, name, &fsdb);
+ if (rc) {
mutex_unlock(&mgs->mgs_mutex);
up_write(&mgs->mgs_barrier_rwsem);
-
- RETURN(-ENODEV);
+ RETURN(rc);
}
- if (unlikely(fsdb->fsdb_mdt_count == 0))
+ if (unlikely(fsdb->fsdb_mdt_count == 0)) {
mgs_barrier_bitmap_setup(mgs, fsdb, bc->bc_name);
+ /* fsdb was just created, ensure that fsdb_barrier_disabled is
+ * set correctly */
+ if (fsdb->fsdb_mdt_count > 0) {
+ struct obd_export *exp;
+ struct obd_device *mgs_obd = mgs->mgs_obd;
+
+ spin_lock(&mgs_obd->obd_dev_lock);
+ list_for_each_entry(exp, &mgs_obd->obd_exports,
+ exp_obd_chain) {
+ __u64 flags = exp_connect_flags(exp);
+ if (!!(flags & OBD_CONNECT_MDS_MDS) &&
+ !(flags & OBD_CONNECT_BARRIER)) {
+ fsdb->fsdb_barrier_disabled = 1;
+ break;
+ }
+ }
+ spin_unlock(&mgs_obd->obd_dev_lock);
+ }
+ }
+
mutex_lock(&fsdb->fsdb_mutex);
mutex_unlock(&mgs->mgs_mutex);
@@ -445,17 +464,36 @@ static int mgs_barrier_thaw(const struct lu_env *env,
down_write(&mgs->mgs_barrier_rwsem);
mutex_lock(&mgs->mgs_mutex);
- fsdb = mgs_find_fsdb(mgs, name);
- if (!fsdb) {
+ rc = mgs_find_or_make_fsdb_nolock(env, mgs, name, &fsdb);
+ if (rc) {
mutex_unlock(&mgs->mgs_mutex);
up_write(&mgs->mgs_barrier_rwsem);
-
- RETURN(-ENODEV);
+ RETURN(rc);
}
- if (unlikely(fsdb->fsdb_mdt_count == 0))
+ if (unlikely(fsdb->fsdb_mdt_count == 0)) {
mgs_barrier_bitmap_setup(mgs, fsdb, bc->bc_name);
+ /* fsdb was just created, ensure that fsdb_barrier_disabled is
+ * set correctly */
+ if (fsdb->fsdb_mdt_count > 0) {
+ struct obd_export *exp;
+ struct obd_device *mgs_obd = mgs->mgs_obd;
+
+ spin_lock(&mgs_obd->obd_dev_lock);
+ list_for_each_entry(exp, &mgs_obd->obd_exports,
+ exp_obd_chain) {
+ __u64 flags = exp_connect_flags(exp);
+ if (!!(flags & OBD_CONNECT_MDS_MDS) &&
+ !(flags & OBD_CONNECT_BARRIER)) {
+ fsdb->fsdb_barrier_disabled = 1;
+ break;
+ }
+ }
+ spin_unlock(&mgs_obd->obd_dev_lock);
+ }
+ }
+
mutex_lock(&fsdb->fsdb_mutex);
mutex_unlock(&mgs->mgs_mutex);
@@ -596,13 +634,32 @@ static int mgs_barrier_rescan(const struct lu_env *env,
snprintf(name, sizeof(mgs_env_info(env)->mgi_fsname) - 1, "%s-%s",
bc->bc_name, BARRIER_FILENAME);
- b_fsdb = mgs_find_fsdb(mgs, name);
- if (!b_fsdb) {
+ rc = mgs_find_or_make_fsdb_nolock(env, mgs, name, &b_fsdb);
+ if (rc) {
mutex_unlock(&mgs->mgs_mutex);
up_write(&mgs->mgs_barrier_rwsem);
mgs_put_fsdb(mgs, c_fsdb);
+ RETURN(rc);
+ }
- RETURN(-ENODEV);
+ if (unlikely(b_fsdb->fsdb_mdt_count == 0 &&
+ c_fsdb->fsdb_mdt_count > 0)) {
+ /* fsdb was just created, ensure that fsdb_barrier_disabled is
+ * set correctly */
+ struct obd_export *exp;
+ struct obd_device *mgs_obd = mgs->mgs_obd;
+
+ spin_lock(&mgs_obd->obd_dev_lock);
+ list_for_each_entry(exp, &mgs_obd->obd_exports,
+ exp_obd_chain) {
+ __u64 flags = exp_connect_flags(exp);
+ if (!!(flags & OBD_CONNECT_MDS_MDS) &&
+ !(flags & OBD_CONNECT_BARRIER)) {
+ b_fsdb->fsdb_barrier_disabled = 1;
+ break;
+ }
+ }
+ spin_unlock(&mgs_obd->obd_dev_lock);
}
mutex_lock(&b_fsdb->fsdb_mutex);
diff --git a/lustre/mgs/mgs_internal.h b/lustre/mgs/mgs_internal.h
index 28aa7db0d5..5eb9ea7f15 100644
--- a/lustre/mgs/mgs_internal.h
+++ b/lustre/mgs/mgs_internal.h
@@ -201,6 +201,9 @@ int mgs_params_fsdb_setup(const struct lu_env *env, struct mgs_device *mgs);
int mgs_params_fsdb_cleanup(const struct lu_env *env, struct mgs_device *mgs);
int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs,
char *name, struct fs_db **dbh);
+int mgs_find_or_make_fsdb_nolock(const struct lu_env *env,
+ struct mgs_device *mgs, char *name,
+ struct fs_db **dbh);
struct fs_db *mgs_find_fsdb(struct mgs_device *mgs, const char *fsname);
void mgs_put_fsdb(struct mgs_device *mgs, struct fs_db *fsdb);
int mgs_get_fsdb_srpc_from_llog(const struct lu_env *env,
diff --git a/lustre/mgs/mgs_llog.c b/lustre/mgs/mgs_llog.c
index 350ab003e8..d9d110c982 100644
--- a/lustre/mgs/mgs_llog.c
+++ b/lustre/mgs/mgs_llog.c
@@ -541,14 +541,15 @@ int mgs_cleanup_fsdb_list(struct mgs_device *mgs)
return 0;
}
-int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs,
- char *name, struct fs_db **dbh)
+/* The caller must hold mgs->mgs_mutex. */
+int mgs_find_or_make_fsdb_nolock(const struct lu_env *env,
+ struct mgs_device *mgs,
+ char *name, struct fs_db **dbh)
{
struct fs_db *fsdb;
int rc = 0;
ENTRY;
- mutex_lock(&mgs->mgs_mutex);
fsdb = mgs_find_fsdb(mgs, name);
if (!fsdb) {
fsdb = mgs_new_fsdb(env, mgs, name);
@@ -557,7 +558,6 @@ int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs,
CDEBUG(D_MGS, "Created new db: rc = %d\n", rc);
}
- mutex_unlock(&mgs->mgs_mutex);
if (!rc)
*dbh = fsdb;
@@ -565,6 +565,19 @@ int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs,
RETURN(rc);
}
+int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs,
+ char *name, struct fs_db **dbh)
+{
+ int rc;
+ ENTRY;
+
+ mutex_lock(&mgs->mgs_mutex);
+ rc = mgs_find_or_make_fsdb_nolock(env, mgs, name, dbh);
+ mutex_unlock(&mgs->mgs_mutex);
+
+ RETURN(rc);
+}
+
/* 1 = index in use
0 = index unused
-1= empty client log */
diff --git a/lustre/tests/sanity-lsnapshot.sh b/lustre/tests/sanity-lsnapshot.sh
index 26fd232f5e..8126799476 100755
--- a/lustre/tests/sanity-lsnapshot.sh
+++ b/lustre/tests/sanity-lsnapshot.sh
@@ -330,6 +330,41 @@ test_3b() {
}
run_test 3b "modify snapshot without original filesystem mounted"
+test_4() { # LU-10843
+ combined_mgs_mds && skip "Combined MGS/MDS" && return
+
+ local rcmd="$LCTL get_param ldlm.namespaces.MGS.resource_count"
+
+ local exports=$(do_facet mgs "$LCTL get_param -n mgs.MGS.num_exports")
+ local rcount=$(do_facet mgs $rcmd)
+
+ echo "Remount MGT"
+ stop mgs || error "stop mgs failed"
+ start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "start mgs failed"
+
+ echo "Wait for all reconnects"
+ local CMD="$LCTL get_param -n mgs.MGS.num_exports"
+ wait_update_facet mgs "$CMD" $exports ||
+ lss_err "(1) failed to export from mgs"
+
+ wait_update_facet mgs "$rcmd" $rcount ||
+ lss_err "(2) failed to reconnect mds"
+
+ echo "Create lss_4_0"
+ lsnapshot_create -n lss_4_0 -c "'It is test_4'" ||
+ lss_err "(3) Fail to create lss_4_0"
+
+ echo "List lss_4_0"
+ lsnapshot_list -n lss_4_0 ||
+ lss_err "(4) Fail to list lss_4_0"
+
+ echo "Destroy lss_4_0"
+ lsnapshot_destroy -n lss_4_0 ||
+ lss_err "(5) Fail to destroy lss_4_0"
+
+}
+run_test 4 "create/delete snapshot after MGS remount"
+
lss_cleanup
do_facet mgs $LCTL set_param debug=-snapshot
do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param debug=-snapshot
--
GitLab