2019-09-04 02:39:25

by Hongzhi, Song

[permalink] [raw]
Subject: Re: Bug Report: Btrfs can't allocate space for delete when block size arounds 512M

Anybody notice this?

--Hongzhi


On 7/17/19 4:34 PM, Hongzhi, Song wrote:
> Hi friends,
>
> *Description:*
>
>
>     One LTP testcase, fs_fill.c, fails on btrfs with kernel error when
> unlink files on Btrfs device:
>
>     "BTRFS warning (device loop0): could not allocate space for a
> delete; will truncate on mount".
>
>
>     I found the loop block device formatted with btrfs roughly rangs
> from 460M to 560M will cause the error.
>
>     256M and 1G all pass.
>
>
>     The fs_fill.c source code:
>
> [https://github.com/linux-test-project/ltp/blob/master/testcases/kernel/fs/fs_fill/fs_fill.c]
>
>
>     The fs_fill.c calls unlink which triggers the error.
>
> [https://github.com/linux-test-project/ltp/blob/e3457e42c1b93f54bb81da746eba314fd34ad40e/testcases/kernel/fs/fs_fill/fs_fill.c#L55]
>
>
> [https://github.com/linux-test-project/ltp/blob/e3457e42c1b93f54bb81da746eba314fd34ad40e/lib/safe_macros.c#L358]
>
>
>
> *Error info:*
>
>     The issue maybe not reproduced everytime but four fifths chance.
>
>     fs_fill.c:53: INFO: Unlinking mntpoint/thread5/file0
>     safe_macros.c:360: BROK: fs_fill.c:55:
> unlink(mntpoint/thread10/file0) failed: ENOSPC
>     safe_macros.c:360: BROK: fs_fill.c:55:
> unlink(mntpoint/thread11/file0) failed: ENOSPC
>     [62477.378848] BTRFS warning (device loop0): could not allocate
> space for a delete; will truncate on mount
>     [62477.378905] BTRFS warning (device loop0): could not allocate
> space for a delete; will truncate on mount
>
>
>
> *Kernel:*
>
>     After v5.2-rc1, qemux86-64
>
>     # make -j40 ARCH=x86_64 CROSS_COMPILE=x86-64-gcc
>     use qemu to bootup kernel
>
>
> *LTP:*
>
>     master branch: I tested on 20190625
>     Reproduce:
>
>     // build Ltp
>     # cd Ltp-source
>     # ./build.sh
>
>     // copy files to qemu
>     # cp runltp testcases/kernel/fs/fs_fill/fs_fill to qemu
>
>     // login to qemu:
>     // adjust block device size to 512M
>     # vi runltp
>     in function: create_block()
>         dd if=/dev/zero of=${TMP}/test.img bs=1024 count=262144
>         --->
>         dd if=/dev/zero of=${TMP}/test.img bs=1024 count=524288
>
>     // execute testcase
>     # runltp -f fs -s fs_fill
>
>
> *Analysis:*
>
>     One new kernel commit contained in v5.2-rc1 introduces the issue.
>
>     commit c8eaeac7b734347c3afba7008b7af62f37b9c140
>     Author: Josef Bacik <[email protected]>
>     Date:   Wed Apr 10 15:56:10 2019 -0400
>
>         btrfs: reserve delalloc metadata differently
>         ...
>
>
> Anyone's reply will be appreciated.
>
> --Hongzhi
>
>
>
>
>
>


2019-09-04 03:46:53

by Hongzhi, Song

[permalink] [raw]
Subject: Re: Bug Report: Btrfs can't allocate space for delete when block size arounds 512M

I remake the test case to be more simply.

And the send the new email.

Thanks,

--Hongzhi


On 9/4/19 10:36 AM, Hongzhi, Song wrote:
> Anybody notice this?
>
> --Hongzhi
>
>
> On 7/17/19 4:34 PM, Hongzhi, Song wrote:
>> Hi friends,
>>
>> *Description:*
>>
>>
>>     One LTP testcase, fs_fill.c, fails on btrfs with kernel error
>> when unlink files on Btrfs device:
>>
>>     "BTRFS warning (device loop0): could not allocate space for a
>> delete; will truncate on mount".
>>
>>
>>     I found the loop block device formatted with btrfs roughly rangs
>> from 460M to 560M will cause the error.
>>
>>     256M and 1G all pass.
>>
>>
>>     The fs_fill.c source code:
>>
>> [https://github.com/linux-test-project/ltp/blob/master/testcases/kernel/fs/fs_fill/fs_fill.c]
>>
>>
>>     The fs_fill.c calls unlink which triggers the error.
>>
>> [https://github.com/linux-test-project/ltp/blob/e3457e42c1b93f54bb81da746eba314fd34ad40e/testcases/kernel/fs/fs_fill/fs_fill.c#L55]
>>
>>
>> [https://github.com/linux-test-project/ltp/blob/e3457e42c1b93f54bb81da746eba314fd34ad40e/lib/safe_macros.c#L358]
>>
>>
>>
>> *Error info:*
>>
>>     The issue maybe not reproduced everytime but four fifths chance.
>>
>>     fs_fill.c:53: INFO: Unlinking mntpoint/thread5/file0
>>     safe_macros.c:360: BROK: fs_fill.c:55:
>> unlink(mntpoint/thread10/file0) failed: ENOSPC
>>     safe_macros.c:360: BROK: fs_fill.c:55:
>> unlink(mntpoint/thread11/file0) failed: ENOSPC
>>     [62477.378848] BTRFS warning (device loop0): could not allocate
>> space for a delete; will truncate on mount
>>     [62477.378905] BTRFS warning (device loop0): could not allocate
>> space for a delete; will truncate on mount
>>
>>
>>
>> *Kernel:*
>>
>>     After v5.2-rc1, qemux86-64
>>
>>     # make -j40 ARCH=x86_64 CROSS_COMPILE=x86-64-gcc
>>     use qemu to bootup kernel
>>
>>
>> *LTP:*
>>
>>     master branch: I tested on 20190625
>>     Reproduce:
>>
>>     // build Ltp
>>     # cd Ltp-source
>>     # ./build.sh
>>
>>     // copy files to qemu
>>     # cp runltp testcases/kernel/fs/fs_fill/fs_fill to qemu
>>
>>     // login to qemu:
>>     // adjust block device size to 512M
>>     # vi runltp
>>     in function: create_block()
>>         dd if=/dev/zero of=${TMP}/test.img bs=1024 count=262144
>>         --->
>>         dd if=/dev/zero of=${TMP}/test.img bs=1024 count=524288
>>
>>     // execute testcase
>>     # runltp -f fs -s fs_fill
>>
>>
>> *Analysis:*
>>
>>     One new kernel commit contained in v5.2-rc1 introduces the issue.
>>
>>     commit c8eaeac7b734347c3afba7008b7af62f37b9c140
>>     Author: Josef Bacik <[email protected]>
>>     Date:   Wed Apr 10 15:56:10 2019 -0400
>>
>>         btrfs: reserve delalloc metadata differently
>>         ...
>>
>>
>> Anyone's reply will be appreciated.
>>
>> --Hongzhi
>>
>>
>>
>>
>>
>>
>

2019-09-04 08:04:22

by Nikolay Borisov

[permalink] [raw]
Subject: Re: Bug Report: Btrfs can't allocate space for delete when block size arounds 512M



On 4.09.19 г. 5:36 ч., Hongzhi, Song wrote:
> Anybody notice this?
>

There were multiple fixes from Josef recently improving btrfs enospc
handling with tiny filesystems (which is generally not the targeted use
case of btrfs). The code lives in
https://github.com/kdave/btrfs-devel/commits/misc-next should you want
to test it. Otherwise re-test after next merge windows when those
patches are supposed to be merged for 5.4