LU-14513 osd: release o_guard before quota acquisition

[fs/lustre-release.git] / lustre / osd-zfs / osd_object.c

diff --git a/lustre/osd-zfs/osd_object.c b/lustre/osd-zfs/osd_object.c
index 49f887e..464e0d6 100644 (file)
--- a/lustre/osd-zfs/osd_object.c
+++ b/lustre/osd-zfs/osd_object.c
@@ -1131,7 +1131,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 
        down_read(&obj->oo_guard);
        if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
-               GOTO(out, rc = 0);
+               GOTO(out_sem, rc = 0);
 
        LASSERT(obj->oo_sa_hdl != NULL);
        LASSERT(oh->ot_tx != NULL);
@@ -1153,7 +1153,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
        if (!found)
                dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
        if (oh->ot_tx->tx_err != 0)
-               GOTO(out, rc = -oh->ot_tx->tx_err);
+               GOTO(out_sem, rc = -oh->ot_tx->tx_err);
 
        if (attr && attr->la_valid & LA_FLAGS) {
                /* LMA is usually a part of bonus, no need to declare
@@ -1169,6 +1169,9 @@ static int osd_declare_attr_set(const struct lu_env *env,
                       osd->od_svname,
                       attr->la_uid, attr->la_gid, bspace, blksize);
        }
+       /* to preserve locking order - qsd_transfer() may need to flush
+        * currently running transaction when we're out of quota. */
+       up_read(&obj->oo_guard);
 
        if (attr && attr->la_valid & LA_UID) {
                /* quota enforcement for user */
@@ -1229,6 +1232,8 @@ static int osd_declare_attr_set(const struct lu_env *env,
        }
 #endif
 out:
+       RETURN(rc);
+out_sem:
        up_read(&obj->oo_guard);
        RETURN(rc);
 }