LU-4838 osd: object iterator in zfs to sync before iteration

The issue with dmu_object_next() is that it doesn't expose
non-committed blocks in the metadnode. Therefore the iterator
doesn't find dnodes allocated in the current (non-committed)
txg which breaks testing and may affect LFSCK. As a short-term
solution, osd_zfs_otable_it_init() ensures the current txg
is committed.

Signed-off-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Change-Id: I639f37ced5790d1e1514c3e24594e360fcd6c1a8
Signed-off-by: Nathaniel Clark <nathaniel.l.clark@intel.com>
Reviewed-on: http://review.whamcloud.com/9838
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Tested-by: Andreas Dilger <andreas.dilger@intel.com>

diff --git a/lustre/osd-zfs/osd_index.c b/lustre/osd-zfs/osd_index.c
index b96e642..cbf391c 100644 (file)
--- a/lustre/osd-zfs/osd_index.c
+++ b/lustre/osd-zfs/osd_index.c
@@ -1360,6 +1360,11 @@ static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
 
        it->mit_dev = dev;
 
+       /* XXX: dmu_object_next() does NOT find dnodes allocated
+        *      in the current non-committed txg, so we force txg
+        *      commit to find all existing dnodes ... */
+       txg_wait_synced(dmu_objset_pool(dev->od_objset.os), 0ULL);
+
        RETURN((struct dt_it *)it);
 }
 
@@ -1425,7 +1430,6 @@ static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
        uchar_t                 *v;
        __u64                    dnode;
        int                      rc, s;
-       ENTRY;
 
        memset(&it->mit_fid, 0, sizeof(it->mit_fid));
 
@@ -1473,7 +1477,7 @@ static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
               it->mit_pos, PFID(&it->mit_fid), rc);
 
 out:
-       RETURN(rc);
+       return rc;
 }
 
 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,

diff --git a/lustre/tests/sanity-lfsck.sh b/lustre/tests/sanity-lfsck.sh
index f1ae69b..49519c4 100644 (file)
--- a/lustre/tests/sanity-lfsck.sh
+++ b/lustre/tests/sanity-lfsck.sh
@@ -844,10 +844,11 @@ test_9a() {
 
        do_facet $SINGLEMDS \
                $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
-       sleep 5
-       STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
-       [ "$STATUS" == "completed" ] ||
-               error "(7) Expect 'completed', but got '$STATUS'"
+
+       wait_update_facet $SINGLEMDS \
+           "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace|\
+           awk '/^status/ { print \\\$2 }'" "completed" 30 ||
+               error "(7) Failed to get expected 'completed'"
 }
 run_test 9a "LFSCK speed control (1)"
 
@@ -1433,7 +1434,8 @@ test_15b() {
 
        mkdir -p $DIR/$tdir
        $LFS setstripe -c 1 -i 0 $DIR/$tdir
-       touch $DIR/$tdir/guard
+       dd if=/dev/zero of=$DIR/$tdir/guard bs=1K count=1
+       cancel_lru_locks osc
 
        echo "Inject failure stub to make the OST-object to back point to"
        echo "other MDT-object"