lkml.org 
[lkml]   [2014]   [Aug]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [btrfs] 8d875f95: xfstests.generic.226.fail


On 08/20/2014 06:52 AM, Miao Xie wrote:
> On Tue, 19 Aug 2014 10:58:09 -0400, Chris Mason wrote:
>> On 08/19/2014 10:23 AM, David Sterba wrote:
>>> On Tue, Aug 19, 2014 at 07:58:20PM +0800, Fengguang Wu wrote:
>>>> We noticed an xfstests failure on commit
>>>>
>>>> 8d875f95da43c6a8f18f77869f2ef26e9594fecc ("btrfs: disable strict file flushes for renames and truncates")
>>>>
>>>> It's 100% reproducible in the 5 test runs.
>>>
>>> Same here, different mkfs configurations.
>>>
>>> generic/226 28s ... [16:11:52] [16:12:55] - output mismatch (see /root/xfstests/results//generic/226.out.bad)
>>> --- tests/generic/226.out 2013-05-29 17:16:03.000000000 +0200
>>> +++ /root/xfstests/results//generic/226.out.bad 2014-08-19 16:12:55.000000000 +0200
>>> @@ -1,6 +1,8 @@
>>> QA output created by 226
>>> --> mkfs 256m filesystem
>>> --> 16 buffered 64m writes in a loop
>>> -1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
>>> +1 2 3 4 pwrite64: No space left on device
>>> +5 6 7 8 9 10 11 12 pwrite64: No space left on device
>>> +13 14 15 16
>>>
>>> enospc on a small filesystem (256M)
>>
>> I'm calling filemap flush more often, but otherwise everything else is
>> the same. I'll take a look.
>
> I found the problem is caused by the following function:
>
> int btrfs_release_file(struct inode *inode, struct file *filp)
> {
> ...
> filemap_flush(inode->i_mapping);
> return 0;
> }
>
> I don't think we need flush file at most situation. Ext4 flushes the file only
> after someone truncate the file to be zero-length, I don't know the real reason
> why ext4 flush the file only after the file is truncated, someone said it is to
> reduce the risk that the users find a zero-length file after a crash, which happens
> after truncate-write-close process.
>
> If we change btrfs_release_file by ext4's implementation, both the failure of
> xfstests's generic/226 and performance regression can be fixed.
>

You're completely right, my original had more checks here and I stripped
them out by accident. Fixing, thanks!

-chris



\
 
 \ /
  Last update: 2014-08-20 16:21    [W:0.047 / U:0.116 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site