Fix empty xattr dir causing lockup

During zfs_rmnode on a xattr dir, if the system crash just after
dmu_free_long_range, we would get empty xattr dir in delete queue. This would
cause blkid=0 be passed into zap_get_leaf_byblk when doing zfs_purgedir during
mount, and would try to do rw_enter on a wrong structure and cause system
lockup.

We fix this by returning ENOENT when blkid is zero in zap_get_leaf_byblk.

Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4114
Closes #4052
Closes #4006
Closes #3018
Closes #2861

diff --git a/module/zfs/zap.c b/module/zfs/zap.c
index 4640fb98cc6952c715818e6fced4f39ba559bde5..cf0b92c7d8fe4a376ba395e0cbe590088c1a5293 100644 (file)
--- a/module/zfs/zap.c
+++ b/module/zfs/zap.c
@@ -504,6 +504,16 @@ zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
 
        ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 
+       /*
+        * If system crashed just after dmu_free_long_range in zfs_rmnode, we
+        * would be left with an empty xattr dir in delete queue. blkid=0
+        * would be passed in when doing zfs_purgedir. If that's the case we
+        * should just return immediately. The underlying objects should
+        * already be freed, so this should be perfectly fine.
+        */
+       if (blkid == 0)
+               return (ENOENT);
+
        err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
            blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
        if (err)