regression in 4.5-rc5 - generic/324 failure

regression in 4.5-rc5 - generic/324 failure

[Eric Whitney]

I'm seeing generic/324 fail consistently when run with xfstest-bld's test
appliance in the 1k block size test case on x86_64.  This happened first on my
4.5-rc5 regression run, and it repeated on -rc6 and -rc7.

Reverting a commit introduced in 4.5-rc5 inhibits the failure:
"ext4: don't read blocks from disk after extents being swapped" (bcff24887d00)

And here's the failed test output:

generic/324 15s ...	[22:10:20][    3.304758] run fstests generic/324 at 2016-03-09 22:10:20
 [22:10:21] [failed, exit status 1] - output mismatch (see /results/results-1k/generic/324.out.bad)
    --- tests/generic/324.out	2015-10-25 10:50:34.000000000 +0000
    +++ /results/results-1k/generic/324.out.bad	2016-03-09 22:10:21.753937505 +0000
    @@ -22,2509 +22,5 @@
     XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
     Before: in_range(10, 40)
     After: in_range(1, -1)
    -Defragment file with 250 * 2 fragments
    -wrote 1234/1234 bytes at offset 0
    -XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
    -wrote 1234/1234 bytes at offset 123400
    ...
    (Run 'diff -u tests/generic/324.out /results/results-1k/generic/324.out.bad'  to see the entire diff)
diff: standard output: Broken pipe
Ran: generic/324
Failures: generic/324
Failed 1 of 1 tests

Thanks,
Eric

Re: regression in 4.5-rc5 - generic/324 failure

[Theodore Ts'o]

On Wed, Mar 09, 2016 at 05:17:46PM -0500, Eric Whitney wrote:
> I'm seeing generic/324 fail consistently when run with xfstest-bld's test
> appliance in the 1k block size test case on x86_64.  This happened first on my
> 4.5-rc5 regression run, and it repeated on -rc6 and -rc7.
> 
> Reverting a commit introduced in 4.5-rc5 inhibits the failure:
> "ext4: don't read blocks from disk after extents being swapped" (bcff24887d00)

Yeah, sorry, I didn't recognize the failure you described on the
phone.  This has been fixed in the ext4 dev tree:

commit 87f9a031af48defee9f34c6aaf06d6f1988c244d
Author: Eryu Guan <guaneryu@gmail.com>
Date:   Sun Feb 21 18:38:44 2016 -0500

    ext4: iterate over buffer heads correctly in move_extent_per_page()
    
    In commit bcff24887d00 ("ext4: don't read blocks from disk after extents
    being swapped") bh is not updated correctly in the for loop and wrong
    data has been written to disk. generic/324 catches this on sub-page
    block size ext4.
    
    Fixes: bcff24887d00 ("ext4: don't read blocks from disk after extentsbeing swapped")
    Signed-off-by: Eryu Guan <guaneryu@gmail.com>
    Signed-off-by: Theodore Ts'o <tytso@mit.edu>
    Cc: stable@vger.kernel.org

I should have realized this was a 4.5 regression that should have been
pushed to Linus directly.  It's awfully late, and I'm not sure there's
going to be a -rc8, but I'll see if I can get this to Linus this cycle
or not.

						- Ted