Patchworkβ [Bug,14354] Bad corruption with 2.6.32-rc1 and upwards

login
register
about
Submitterbugzilla-daemon@bugzilla.kernel.org
Date2009-10-27 06:39:50
Message ID<200910270639.n9R6dosP018078@demeter.kernel.org>
Downloadmbox | patch
Permalink/patch/36967/
State Superseded
Headersshow

Comments

bugzilla-daemon@bugzilla.kernel.org - 2009-10-27 06:39:50
http://bugzilla.kernel.org/show_bug.cgi?id=14354





--- Comment #122 from Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>  2009-10-27 06:39:41 ---
Can you test with the below patch

Commit 487caeef9fc08c0565e082c40a8aaf58dad92bbb fixed a deadlock
between get_block and truncate by dropping the i_data_sem during
truncate. But i guess we should make sure we drop any preallocation
that we did as a part of the block allocation. Otherwise we definitely
can get multiple claimed blocks. But this can happen only if we have
block allocation and truncate happening on the same file simultaneously.

With delayed allocation i guess it is possible that pdflush tries to do
block allocation while application does a truncate

Patch

diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 10539e3..14539b5 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -113,6 +113,7 @@  static int ext4_ext_truncate_extend_restart(handle_t
*handle,
      */
     ext4_ext_invalidate_cache(inode);

+    ext4_discard_preallocations(inode);
     return err;
 }