Date:	Wed, 01 Jun 2011 10:27:56 -0500
From:	Eric Sandeen <sandeen@...hat.com>
To:	Amir Goldstein <amir73il@...il.com>
CC:	sergey ivanov <sergey57@...il.com>, XFS <xfs@....sgi.com>,
	Ext4 Developers List <linux-ext4@...r.kernel.org>,
	linux-fsdevel <linux-fsdevel@...r.kernel.org>
Subject: Re: [PATCH] xfstests: add support for ext4dev FSTYP

On 6/1/11 7:37 AM, Amir Goldstein wrote:
> On Wed, Jun 1, 2011 at 2:48 PM, sergey ivanov <sergey57@...il.com> wrote:
>> On Wed, Jun 1, 2011 at 2:37 AM, Amir Goldstein <amir73il@...il.com> wrote:
>>>
>>> On Wed, Jun 1, 2011 at 8:34 AM, Eric Sandeen <sandeen@...hat.com> wrote:
>>>> On 6/1/11 12:22 AM, Eric Sandeen wrote:
>>>>> On 5/31/11 11:56 PM, Amir Goldstein wrote:
>>>>>> On Wed, Jun 1, 2011 at 7:37 AM, Eric Sandeen <sandeen@...hat.com>
>>>>>> wrote:
>>>>>>> On 5/31/11 10:13 PM, Amir Goldstein wrote:
>>>>>>>> From: Amir Goldstein <amir73il@...rs.sf.net>
>>>>>>>>
>>>>>>>> blkid knows to identify the ext4dev FSTYP of a partition that was
>>>>>>>> formatted with mkfs.ext4dev.
>>>>>>>> quota tools and various util-linux utils are also aware of ext4dev,
>>>>>>>> so ext4dev shares the same capabilities as ext4.
>>>>>>>>
>>>>>>>> While testing on Fedora 15, we encoutered a buggy fsck utility,
>>>>>>>> which
>>>>>>>> invokes fsck.ext4, even though it was called with -t ext4dev
>>>>>>>> argument.
>>>>>>>> In our setup fsck.ext4dev knows about new fs features that fsck.ext4
>>>>>>>> doesn't know, so the generic_fs_check fails.
>>>>>>>> Since we have no real use of the extra capabilities provided by fsck
>>>>>>>> util,
>>>>>>>> we decided to invoke fsck.$FSTYP directly to avoid this issue.
>>>>>>>
>>>>>>> Adding ext4dev to every case seems harmless enough.  TBH I thought I
>>>>>>> had
>>>>>>> it there already but I guess not.
>>>>>>>
>>>>>>> I'm less certain of the change from fsck -t $FSTYP to fsck.$FSTYP
>>>>>>>
>>>>>>> What issue are you avoiding?  wouldn't fsck -t ext4dev invoke
>>>>>>> fsck.ext4dev anyway?
>>>>>>>
>>>>>>> It seems like it should be harmless, but I don't understand how it
>>>>>>> helps you.
>>>>>>>
>>>>>>
>>>>>> As I wrote in the patch description, the fsck utility in Fedora 15
>>>>>> invokes
>>>>>> fsck.ext4 for some reason when calling fsck -t ext4dev.
>>>>>
>>>>> Oh, right.
>>>>>
>>>>>> this fails because fsck.ext4 doesn't know the snapshot feature.
>>>>>> I didn't debug fsck utility for that. it seemed pointless.
>>>>>
>>>>> Did you file a bug with Fedora?  I'd rather fix the root cause than
>>>>> work around it...
>>>>> Feel free to cc: me on the bug.
>>>
>>> No, I didn't file a bug.
>>> In any case, it was Sergey, who tested and reported the problem on F15.
>>>
>>> Would you agree to fix the problem in xfstests now, so that F15 users can
>>> test ext4dev and fix the bug in fsck regardless?
>>>
>>>>
>>>> RHEL6 does the same; mkfs.ext4dev then fsck -t ext4dev invokes
>>>> fsck.ext4; but this
>>>> is because blkid identifies it as ext4, not ext4dev, despite the test_fs
>>>> flag being set.
>>>>
>>>> ISTR this is due to some tortured logic about when ext4dev isn't
>>>> ext4dev, but
>>>> I don't remember the details... I don't know if this is the same
>>>> situation
>>>> you're seeing; just to double check - does blkid correctly identify it
>>>> as ext4dev
>>>> on F15?
>>>
>>>
>>> For me (on Ubuntu) blkid identifies ext4dev, but maybe the tortured logic
>>> finds unknown features a justification for declaring ext4dev?
>>> Segrey, can you answer the question for F15?
>>> Did you set FSTYP to ext4dev manually or did blkid identified it for you?
>>
>> Amir, Fedora's fsck behaves differently on ext4dev and next4. When I made
>> next4 partition on Fedora-15 by
>> {{{
>> mkfs  -t next4 /dev/sdd9
>> }}}
>> it requires "export FSTYP=next4" to be defined and exported in local.conf.
>> It mounts partition with
>> {{{
>> mount -t next4 /dev/sdd9 /mnt/sdd9
>> }}}
>> But on attempt to fsck it calls wrong fsck, as you reported:
>> {{{
>> [seriv@...bra xfstests]$ sudo fsck -t next4 /dev/sdd9
>> fsck from util-linux 2.19.1
>> e2fsck 1.41.14 (22-Dec-2010)
>> /dev/sdd9 has unsupported feature(s): FEATURE_C7 FEATURE_C8 FEATURE_R7
>> e2fsck: Get a newer version of e2fsck!
>> }}}
>> And the problem can be easily traced to this wrong call:
>> {{{
>> [seriv@...bra xfstests]$ sudo strace -o /var/log/fsck.strace fsck -t next4
>> /dev/sdd9
>> fsck from util-linux 2.19.1
>> e2fsck 1.41.14 (22-Dec-2010)
>> /dev/sdd9 has unsupported feature(s): FEATURE_C7 FEATURE_C8 FEATURE_R7
>> e2fsck: Get a newer version of e2fsck!
>> [seriv@...bra xfstests]$ grep sbin /var/log/fsck.strace
>> execve("/sbin/fsck", ["fsck", "-t", "next4", "/dev/sdd9"], [/* 17 vars */])
>> = 0
>> stat("/sbin/fsck.ext4", {st_mode=S_IFREG|0755, st_size=194280, ...}) = 0
>> }}}
>> But it's not the case for ext4dev, with it I even don't need "-t ext4dev",
>> it is recognized by blkid:
>> {{{
>> [seriv@...bra xfstests]$ sudo mkfs -t ext4dev /dev/sdd9
>> mke2fs 1.41.14-next3-1.0.13-7 (24-May-2011)
>> [skip]
>> This filesystem will be automatically checked every 27 mounts or
>> 180 days, whichever comes first.  Use tune2fs -c or -i to override.
>> [seriv@...bra xfstests]$ sudo mount /dev/sdd9 /mnt/sdd9
>> [seriv@...bra xfstests]$ mount | grep sdd9
>> /dev/sdd9 on /mnt/sdd9 type ext4dev (rw)
>> [seriv@...bra xfstests]$ sudo umount /dev/sdd9
>> [seriv@...bra xfstests]$ sudo fsck /dev/sdd9
>> fsck from util-linux 2.19.1
>> e2fsck 1.41.14-next3-1.0.13-7 (24-May-2011)
>> The test_fs flag is set (and ext4 is available).  Clear<y>?
>> }}}
>> and at this point in another session
>> {{{
>> [seriv@...bra xfstests]$ pgrep -f -l fsck
>> 4365 sudo fsck /dev/sdd9
>> 4366 fsck /dev/sdd9
>> 4367 fsck.ext4dev /dev/sdd9
>> }}}
>> So I'd like this patch to be applied to xfstests to be able to use it on the
>> different filesystems, - if not, then why have this FSTYP and not rely upon
>> blkid
>> --
> 
> To make a long story a bit shorter, there was no intention to set
> FSTYP manually,
> that was only a temporary hack to make next4 clone work, but since ext4dev is
> identified by blkid and respected by fsck/mkfs/mount, we are going to
> work with it
> and not with next4.
> 
> So I understood the problem incorrectly and the patch to xfstests doesn't need
> to change fsck -t $FSTYP to fsck.$FSTYP.
> I will resend the patch without this change.

Thanks, that makes sense to me.

Just wanted to make sure we weren't masking some other problem by working around it in xfstests ...

-Eric

> Amir.

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists