Permalink
Browse files

quota: Fix race between dqput() and dquot_scan_active()

Currently last dqput() can race with dquot_scan_active() causing it to
call callback for an already deactivated dquot. The race is as follows:

CPU1					CPU2
  dqput()
    spin_lock(&dq_list_lock);
    if (atomic_read(&dquot->dq_count) > 1) {
     - not taken
    if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
      spin_unlock(&dq_list_lock);
      ->release_dquot(dquot);
        if (atomic_read(&dquot->dq_count) > 1)
         - not taken
					  dquot_scan_active()
					    spin_lock(&dq_list_lock);
					    if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
					     - not taken
					    atomic_inc(&dquot->dq_count);
					    spin_unlock(&dq_list_lock);
        - proceeds to release dquot
					    ret = fn(dquot, priv);
					     - called for inactive dquot

Fix the problem by making sure possible ->release_dquot() is finished by
the time we call the callback and new calls to it will notice reference
dquot_scan_active() has taken and bail out.

CC: stable@vger.kernel.org # >= 2.6.29
Signed-off-by: Jan Kara <jack@suse.cz>
  • Loading branch information...
1 parent 09ebb17 commit 1362f4ea20fa63688ba6026e586d9746ff13a846 @jankara jankara committed Feb 20, 2014
Showing with 11 additions and 3 deletions.
  1. +11 −3 fs/quota/dquot.c
View
14 fs/quota/dquot.c
@@ -581,9 +581,17 @@ int dquot_scan_active(struct super_block *sb,
dqstats_inc(DQST_LOOKUPS);
dqput(old_dquot);
old_dquot = dquot;
- ret = fn(dquot, priv);
- if (ret < 0)
- goto out;
+ /*
+ * ->release_dquot() can be racing with us. Our reference
+ * protects us from new calls to it so just wait for any
+ * outstanding call and recheck the DQ_ACTIVE_B after that.
+ */
+ wait_on_dquot(dquot);
+ if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
+ ret = fn(dquot, priv);
+ if (ret < 0)
+ goto out;
+ }
spin_lock(&dq_list_lock);
/* We are safe to continue now because our dquot could not
* be moved out of the inuse list while we hold the reference */

0 comments on commit 1362f4e

Please sign in to comment.