Re: [PATCH v3 12/14] xfs: Unmap blocks according to forcealign

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

On 06/08/2024 21:14, Darrick J. Wong wrote:
> On Thu, Aug 01, 2024 at 04:30:55PM +0000, John Garry wrote:
>> For when forcealign is enabled, blocks in an inode need to be unmapped
>> according to extent alignment, like what is already done for rtvol.
>>
>> Change variable isrt in __xfs_bunmapi() to a bool, as that is really what
>> it is.
>>
>> Signed-off-by: John Garry <john.g.garry@oracle.com>
>> ---
>>   fs/xfs/libxfs/xfs_bmap.c | 48 +++++++++++++++++++++++++++++-----------
>>   fs/xfs/xfs_inode.c       | 16 ++++++++++++++
>>   fs/xfs/xfs_inode.h       |  2 ++
>>   3 files changed, 53 insertions(+), 13 deletions(-)
>>
>> diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c
>> index 0c3df8c71c6d..d6ae344a17fc 100644
>> --- a/fs/xfs/libxfs/xfs_bmap.c
>> +++ b/fs/xfs/libxfs/xfs_bmap.c
>> @@ -5409,6 +5409,25 @@ xfs_bmap_del_extent_real(
>>   	return 0;
>>   }
>>   
>> +static xfs_extlen_t
>> +xfs_bunmapi_align(
>> +	struct xfs_inode	*ip,
>> +	xfs_fsblock_t		fsbno,
>> +	xfs_extlen_t *off)
>> +{
>> +	struct xfs_mount	*mp = ip->i_mount;
>> +
>> +	if (XFS_IS_REALTIME_INODE(ip))
>> +		return xfs_inode_alloc_fsbsize_align(ip, fsbno, off);
>> +	/*
>> +	 * The agbno for the fsbno is aligned to extsize, but the fsbno itself
>> +	 * is not necessarily aligned (to extsize), so use agbno to determine
>> +	 * mod+offset to the alloc unit boundary.
>> +	 */
>> +	return xfs_inode_alloc_fsbsize_align(ip, XFS_FSB_TO_AGBNO(mp, fsbno),
>> +					off);
>> +}
>> +
>>   /*
>>    * Unmap (remove) blocks from a file.
>>    * If nexts is nonzero then the number of extents to remove is limited to
>> @@ -5430,7 +5449,8 @@ __xfs_bunmapi(
>>   	xfs_extnum_t		extno;		/* extent number in list */
>>   	struct xfs_bmbt_irec	got;		/* current extent record */
>>   	struct xfs_ifork	*ifp;		/* inode fork pointer */
>> -	int			isrt;		/* freeing in rt area */
>> +	bool			isrt;		/* freeing in rt area */
>> +	bool			isforcealign;	/* forcealign inode */
>>   	int			logflags;	/* transaction logging flags */
>>   	xfs_extlen_t		mod;		/* rt extent offset */
>>   	struct xfs_mount	*mp = ip->i_mount;
>> @@ -5468,6 +5488,8 @@ __xfs_bunmapi(
>>   	}
>>   	XFS_STATS_INC(mp, xs_blk_unmap);
>>   	isrt = xfs_ifork_is_realtime(ip, whichfork);
>> +	isforcealign = (whichfork != XFS_ATTR_FORK) &&
>> +			xfs_inode_has_forcealign(ip);
>>   	end = start + len;
>>   
>>   	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
>> @@ -5486,6 +5508,8 @@ __xfs_bunmapi(
>>   	extno = 0;
>>   	while (end != (xfs_fileoff_t)-1 && end >= start &&
>>   	       (nexts == 0 || extno < nexts)) {
>> +		xfs_extlen_t off;
> 
> I got really confused because I thought this was a file block offset and
> only after more digging realized that this is a sometimes dummy
> adjustment variable.

yeah, I put it here to use the common helper in both callsites

> 
>> +
>>   		/*
>>   		 * Is the found extent after a hole in which end lives?
>>   		 * Just back up to the previous extent, if so.
>> @@ -5519,18 +5543,18 @@ __xfs_bunmapi(
>>   		if (del.br_startoff + del.br_blockcount > end + 1)
>>   			del.br_blockcount = end + 1 - del.br_startoff;
>>   
>> -		if (!isrt || (flags & XFS_BMAPI_REMAP))
>> +		if ((!isrt && !isforcealign) || (flags & XFS_BMAPI_REMAP))
>>   			goto delete;
>>   
>> -		mod = xfs_rtb_to_rtxoff(mp,
>> -				del.br_startblock + del.br_blockcount);
>> +		mod = xfs_bunmapi_align(ip,
>> +				del.br_startblock + del.br_blockcount, &off);
>>   		if (mod) {
> 
> Oof.  I don't like how this loop body has the rtx adjustment code
> inlined into it.  We only use the isrt flag for the one test above.
> I tried hoisting this into something less gross involving separate
> adjustment functions but then you have to pass in so many outer
> variables that it becomes a mess.
> 
> The best I can come up with for now is:
> 
> 	unsigned int		alloc_fsb = xfs_inode_alloc_fsbsize(ip);
> 	/* no more isrt/isforcealign bools */
> 
> ...
> 
> 		if (alloc_fsb == 1 || (flags & XFS_BMAPI_REMAP))
> 			goto delete;

ok, good

> 
> 		mod = do_div(del.br_startblock + del.br_blockcount,
> 				alloc_fsb);

Note that xfs_bunmapi_align() uses agbno for !rt

> 		if (mod) {
> 
>>   			/*
>> -			 * Realtime extent not lined up at the end.
>> +			 * Not aligned to allocation unit on the end.
>>   			 * The extent could have been split into written
>>   			 * and unwritten pieces, or we could just be
>>   			 * unmapping part of it.  But we can't really
>> -			 * get rid of part of a realtime extent.
>> +			 * get rid of part of an extent.
>>   			 */
>>   			if (del.br_state == XFS_EXT_UNWRITTEN) {
>>   				/*
>> @@ -5554,7 +5578,7 @@ __xfs_bunmapi(
>>   			ASSERT(del.br_state == XFS_EXT_NORM);
>>   			ASSERT(tp->t_blk_res > 0);
>>   			/*
>> -			 * If this spans a realtime extent boundary,
>> +			 * If this spans an extent boundary,
>>   			 * chop it back to the start of the one we end at.
>>   			 */
>>   			if (del.br_blockcount > mod) {
>> @@ -5571,14 +5595,12 @@ __xfs_bunmapi(
>>   			goto nodelete;
>>   		}
>>   
>> -		mod = xfs_rtb_to_rtxoff(mp, del.br_startblock);
>> +		mod = xfs_bunmapi_align(ip, del.br_startblock, &off);
>>   		if (mod) {
>> -			xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
> 
> 		mod = do_div(del.br_startblock, alloc_fsb);
> 		if (mod) {
> 			xfs_extlen_t off = alloc_fsb - mod;
> 
> At least then you don't need this weird xfs_inode_alloc_fsbsize_align
> that passes back two xfs_extlen_t arguments.

Sure, but same point about xfs_bunmapi_align() using agbno. I suppose I 
can just make the change to not have xfs_bunmapi_align() be passed the 
off address pointer, and do the calcation here for off here (as you 
suggest).

Thanks,
John