Re: [PATCH v3 07/14] xfs: Introduce FORCEALIGN inode flag

[John Garry <john.g.garry@oracle.com>]

On 06/08/2024 20:02, Darrick J. Wong wrote:
> On Thu, Aug 01, 2024 at 04:30:50PM +0000, John Garry wrote:
>> From: "Darrick J. Wong" <djwong@kernel.org>
>>
>> Add a new inode flag to require that all file data extent mappings must
>> be aligned (both the file offset range and the allocated space itself)
>> to the extent size hint.  Having a separate COW extent size hint is no
>> longer allowed.
>>
>> The goal here is to enable sysadmins and users to mandate that all space
>> mappings in a file must have a startoff/blockcount that are aligned to
>> (say) a 2MB alignment and that the startblock/blockcount will follow the
>> same alignment.
>>
>> Allocated space will be aligned to start of the AG, and not necessarily
>> aligned with disk blocks. The upcoming atomic writes feature will rely and
>> forcealign and will also require allocated space will also be aligned to
>> disk blocks.
>>
>> reflink will not be supported for forcealign yet, so disallow a mount under
>> this condition. This is because we have the limitation of pageache
>> writeback not knowing how to writeback an entire allocation unut, so
>> reject a mount with relink.
>>
>> RT vol will not be supported for forcealign yet, so disallow a mount under
>> this condition. It will be possible to support RT vol and forcealign in
>> future. For this, the inode extsize must be a multiple of rtextsize - this
>> is enforced already in xfs_ioctl_setattr_check_extsize() and
>> xfs_inode_validate_extsize().
>>
>> Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
>> Co-developed-by: John Garry <john.g.garry@oracle.com>
>> [jpg: many changes from orig, including forcealign inode verification
>>   rework, ioctl setattr rework disallow reflink a forcealign inode,
>>   disallow mount for forcealign + reflink or rt]
>> Signed-off-by: John Garry <john.g.garry@oracle.com>
> 
> This patch looks ready to me but as I'm the original author I cannot add
> a RVB tag.  Someone else needs to add that -- frankly, John is the best
> candidate because he grabbed my patch into his tree and actually
> modified it to do what he wants, which means he's the most familiar with
> it.

I thought my review would be implied since I noted how I appended it, above.

Anyway,

Reviewed-by: John Garry <john.g.garry@oracle.com>

I am hoping that Dave and Christoph will give some formal ack/review 
when they get a chance.

BTW, at what stage do we give XFS_SB_FEAT_RO_COMPAT_FORCEALIGN a more 
proper value? So far it has the experimental dev value of 1 << 30, below.

Thanks!


>>
>> diff --git a/fs/xfs/libxfs/xfs_format.h b/fs/xfs/libxfs/xfs_format.h
>> index e1bfee0c3b1a..95f5259c4255 100644
>> --- a/fs/xfs/libxfs/xfs_format.h
>> +++ b/fs/xfs/libxfs/xfs_format.h
>> @@ -352,6 +352,7 @@ xfs_sb_has_compat_feature(
>>   #define XFS_SB_FEAT_RO_COMPAT_RMAPBT   (1 << 1)		/* reverse map btree */
>>   #define XFS_SB_FEAT_RO_COMPAT_REFLINK  (1 << 2)		/* reflinked files */
>>   #define XFS_SB_FEAT_RO_COMPAT_INOBTCNT (1 << 3)		/* inobt block counts */
>> +#define XFS_SB_FEAT_RO_COMPAT_FORCEALIGN (1 << 30)	/* aligned file data extents */
>>   #define XFS_SB_FEAT_RO_COMPAT_ALL \
>>   		(XFS_SB_FEAT_RO_COMPAT_FINOBT | \
>>   		 XFS_SB_FEAT_RO_COMPAT_RMAPBT | \