xfs
[Top] [All Lists]

[PATCH] quota: Fix race between dqput() and dquot_scan_active()

To: linux-fsdevel@xxxxxxxxxxxxxxx
Subject: [PATCH] quota: Fix race between dqput() and dquot_scan_active()
From: Jan Kara <jack@xxxxxxx>
Date: Fri, 10 Oct 2014 16:23:32 +0200
Cc: linux-ext4@xxxxxxxxxxxxxxx, Dave Chinner <david@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx, cluster-devel@xxxxxxxxxx, Steven Whitehouse <swhiteho@xxxxxxxxxx>, Mark Fasheh <mfasheh@xxxxxxxx>, Joel Becker <jlbec@xxxxxxxxxxxx>, ocfs2-devel@xxxxxxxxxxxxxx, reiserfs-devel@xxxxxxxxxxxxxxx, Jeff Mahoney <jeffm@xxxxxxx>, Dave Kleikamp <shaggy@xxxxxxxxxx>, jfs-discussion@xxxxxxxxxxxxxxxxxxxxx, tytso@xxxxxxx, viro@xxxxxxxxxxxxxxxxxx, Jan Kara <jack@xxxxxxx>
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <1412951028-4085-1-git-send-email-jack@xxxxxxx>
References: <1412951028-4085-1-git-send-email-jack@xxxxxxx>
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.

Signed-off-by: Jan Kara <jack@xxxxxxx>
---
 fs/quota/dquot.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

  This is the last patch needed to make ocfs2 quotas rock solid in my testing.
I will carry it in my tree and push it to Linus soon.

diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
index 831d49a4111f..cfc8dcc16043 100644
--- a/fs/quota/dquot.c
+++ b/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 */
-- 
1.8.1.4

<Prev in Thread] Current Thread [Next in Thread>