5.15 request: properly backport VFS/XFS syncfs error handling

5.15 request: properly backport VFS/XFS syncfs error handling

[Holger Hoffstätte]

I noticed that our autobot recently backported parts of a series which
fixed XFS syncfs error handling [1]. Unfortunately - due to missing
requirements - it only managed to merge those patches which do not
actually fix anything.

This can be repaired by applying the prerequisites and then the missing
parts of the original series, namely in order:

   9a208ba5c9af fs: remove __sync_filesystem
   70164eb6ccb7 block: remove __sync_blockdev
   1e03a36bdff4 block: simplify the block device syncing code
   5679897eb104 vfs: make sync_filesystem return errors from ->sync_fs
   2d86293c7075 xfs: return errors in xfs_fs_sync_fs

With all that we could also put a cherry on top and merge:

   b97cca3ba909 xfs: only bother with sync_filesystem during readonly remount

but that's just a touchup and not a real bugfix, so probably optional.

thanks,
Holger

[1] https://lore.kernel.org/linux-xfs/164316348940.2600168.17153575889519271710.stgit@magnolia/

Re: 5.15 request: properly backport VFS/XFS syncfs error handling

[Greg KH]

On Fri, Apr 15, 2022 at 02:10:46PM +0200, Holger Hoffstätte wrote:
> 
> I noticed that our autobot recently backported parts of a series which
> fixed XFS syncfs error handling [1]. Unfortunately - due to missing
> requirements - it only managed to merge those patches which do not
> actually fix anything.
> 
> This can be repaired by applying the prerequisites and then the missing
> parts of the original series, namely in order:
> 
>   9a208ba5c9af fs: remove __sync_filesystem
>   70164eb6ccb7 block: remove __sync_blockdev
>   1e03a36bdff4 block: simplify the block device syncing code
>   5679897eb104 vfs: make sync_filesystem return errors from ->sync_fs
>   2d86293c7075 xfs: return errors in xfs_fs_sync_fs
> 
> With all that we could also put a cherry on top and merge:
> 
>   b97cca3ba909 xfs: only bother with sync_filesystem during readonly remount
> 
> but that's just a touchup and not a real bugfix, so probably optional.

Can we get an ack from the XFS developers that this is ok to do?
Without that, we can't apply xfs patches to the stable trees.

thanks,

greg k-h

Re: 5.15 request: properly backport VFS/XFS syncfs error handling

[Darrick J. Wong]

On Mon, Apr 18, 2022 at 01:45:06PM +0200, Greg KH wrote:
> On Fri, Apr 15, 2022 at 02:10:46PM +0200, Holger Hoffstätte wrote:
> > 
> > I noticed that our autobot recently backported parts of a series which
> > fixed XFS syncfs error handling [1]. Unfortunately - due to missing
> > requirements - it only managed to merge those patches which do not
> > actually fix anything.
> > 
> > This can be repaired by applying the prerequisites and then the missing
> > parts of the original series, namely in order:
> > 
> >   9a208ba5c9af fs: remove __sync_filesystem
> >   70164eb6ccb7 block: remove __sync_blockdev
> >   1e03a36bdff4 block: simplify the block device syncing code
> >   5679897eb104 vfs: make sync_filesystem return errors from ->sync_fs
> >   2d86293c7075 xfs: return errors in xfs_fs_sync_fs
> > 
> > With all that we could also put a cherry on top and merge:
> > 
> >   b97cca3ba909 xfs: only bother with sync_filesystem during readonly remount
> > 
> > but that's just a touchup and not a real bugfix, so probably optional.
> 
> Can we get an ack from the XFS developers that this is ok to do?
> Without that, we can't apply xfs patches to the stable trees.

You might consider adding these two other patches:

2719c7160dcf ("vfs: make freeze_super abort when sync_filesystem returns error")
dd5532a4994b ("quota: make dquot_quota_sync return errors from ->sync_fs")

to the pile, but otherwise that looks ok to me.

--D

> thanks,
> 
> greg k-h

Re: 5.15 request: properly backport VFS/XFS syncfs error handling

[Holger Hoffstätte]

On 2022-04-18 22:31, Darrick J. Wong wrote:
> On Mon, Apr 18, 2022 at 01:45:06PM +0200, Greg KH wrote:
>> On Fri, Apr 15, 2022 at 02:10:46PM +0200, Holger Hoffstätte wrote:
>>>
>>> I noticed that our autobot recently backported parts of a series which
>>> fixed XFS syncfs error handling [1]. Unfortunately - due to missing
>>> requirements - it only managed to merge those patches which do not
>>> actually fix anything.
>>>
>>> This can be repaired by applying the prerequisites and then the missing
>>> parts of the original series, namely in order:
>>>
>>>    9a208ba5c9af fs: remove __sync_filesystem
>>>    70164eb6ccb7 block: remove __sync_blockdev
>>>    1e03a36bdff4 block: simplify the block device syncing code
>>>    5679897eb104 vfs: make sync_filesystem return errors from ->sync_fs
>>>    2d86293c7075 xfs: return errors in xfs_fs_sync_fs
>>>
>>> With all that we could also put a cherry on top and merge:
>>>
>>>    b97cca3ba909 xfs: only bother with sync_filesystem during readonly remount
>>>
>>> but that's just a touchup and not a real bugfix, so probably optional.
>>
>> Can we get an ack from the XFS developers that this is ok to do?
>> Without that, we can't apply xfs patches to the stable trees.
> 
> You might consider adding these two other patches:
> 
> 2719c7160dcf ("vfs: make freeze_super abort when sync_filesystem returns error")
> dd5532a4994b ("quota: make dquot_quota_sync return errors from ->sync_fs")
> 
> to the pile, but otherwise that looks ok to me.
> 
> --D

Those two are the ones from the series that were already merged into 5.15.x,
some time back in February. Looks like we have them all.

cheers
Holger

Re: 5.15 request: properly backport VFS/XFS syncfs error handling

[Sasha Levin]

On Mon, Apr 18, 2022 at 11:52:47PM +0200, Holger Hoffstätte wrote:
>On 2022-04-18 22:31, Darrick J. Wong wrote:
>>On Mon, Apr 18, 2022 at 01:45:06PM +0200, Greg KH wrote:
>>>On Fri, Apr 15, 2022 at 02:10:46PM +0200, Holger Hoffstätte wrote:
>>>>
>>>>I noticed that our autobot recently backported parts of a series which
>>>>fixed XFS syncfs error handling [1]. Unfortunately - due to missing
>>>>requirements - it only managed to merge those patches which do not
>>>>actually fix anything.
>>>>
>>>>This can be repaired by applying the prerequisites and then the missing
>>>>parts of the original series, namely in order:
>>>>
>>>>   9a208ba5c9af fs: remove __sync_filesystem
>>>>   70164eb6ccb7 block: remove __sync_blockdev
>>>>   1e03a36bdff4 block: simplify the block device syncing code
>>>>   5679897eb104 vfs: make sync_filesystem return errors from ->sync_fs
>>>>   2d86293c7075 xfs: return errors in xfs_fs_sync_fs
>>>>
>>>>With all that we could also put a cherry on top and merge:
>>>>
>>>>   b97cca3ba909 xfs: only bother with sync_filesystem during readonly remount
>>>>
>>>>but that's just a touchup and not a real bugfix, so probably optional.
>>>
>>>Can we get an ack from the XFS developers that this is ok to do?
>>>Without that, we can't apply xfs patches to the stable trees.
>>
>>You might consider adding these two other patches:
>>
>>2719c7160dcf ("vfs: make freeze_super abort when sync_filesystem returns error")
>>dd5532a4994b ("quota: make dquot_quota_sync return errors from ->sync_fs")
>>
>>to the pile, but otherwise that looks ok to me.
>>
>>--D
>
>Those two are the ones from the series that were already merged into 5.15.x,
>some time back in February. Looks like we have them all.

Queued up, thanks.

-- 
Thanks,
Sasha