Re: [RFC PATCH v1 06/18] xfs: add iomap async buffered write support

From: Stefan Roesch <shr@fb.com>
To: Dave Chinner <david@fromorbit.com>
Cc: io-uring@vger.kernel.org, kernel-team@fb.com, linux-mm@kvack.org,
	linux-xfs@vger.kernel.org, linux-fsdevel@vger.kernel.org
Subject: Re: [RFC PATCH v1 06/18] xfs: add iomap async buffered write support
Date: Thu, 28 Apr 2022 13:03:49 -0700	[thread overview]
Message-ID: <8b71929a-895d-fdab-468d-541eaa8b4128@fb.com> (raw)
In-Reply-To: <20220426225436.GQ1544202@dread.disaster.area>

On 4/26/22 3:54 PM, Dave Chinner wrote:
> On Tue, Apr 26, 2022 at 10:43:23AM -0700, Stefan Roesch wrote:
>> This adds the async buffered write support to the iomap layer of XFS. If
>> a lock cannot be acquired or additional reads need to be performed, the
>> request will return -EAGAIN in case this is an async buffered write request.
>>
>> Signed-off-by: Stefan Roesch <shr@fb.com>
>> ---
>>  fs/xfs/xfs_iomap.c | 33 +++++++++++++++++++++++++++++++--
>>  1 file changed, 31 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
>> index e552ce541ec2..80b6c48e88af 100644
>> --- a/fs/xfs/xfs_iomap.c
>> +++ b/fs/xfs/xfs_iomap.c
>> @@ -881,18 +881,28 @@ xfs_buffered_write_iomap_begin(
>>  	bool			eof = false, cow_eof = false, shared = false;
>>  	int			allocfork = XFS_DATA_FORK;
>>  	int			error = 0;
>> +	bool			no_wait = (flags & IOMAP_NOWAIT);
>>  
>>  	if (xfs_is_shutdown(mp))
>>  		return -EIO;
>>  
>>  	/* we can't use delayed allocations when using extent size hints */
>> -	if (xfs_get_extsz_hint(ip))
>> +	if (xfs_get_extsz_hint(ip)) {
>> +		if (no_wait)
>> +			return -EAGAIN;
>> +
>>  		return xfs_direct_write_iomap_begin(inode, offset, count,
>>  				flags, iomap, srcmap);
> 
> Why can't we do IOMAP_NOWAIT allocation here?
> xfs_direct_write_iomap_begin() supports IOMAP_NOWAIT semantics just
> fine - that's what the direct IO path uses...

I'll make that change in the next version.
> 
>> +	}
>>  
>>  	ASSERT(!XFS_IS_REALTIME_INODE(ip));
>>  
>> -	xfs_ilock(ip, XFS_ILOCK_EXCL);
>> +	if (no_wait) {
>> +		if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
>> +			return -EAGAIN;
>> +	} else {
>> +		xfs_ilock(ip, XFS_ILOCK_EXCL);
>> +	}
> 
> handled by xfs_ilock_for_iomap().

The helper xfs_ilock_for_iomap cannot be used for this purpose. The function
xfs_ilock_for_iomap tries to deduce the lock mode. For the function xfs_file_buffered_write()
the lock mode must be exclusive. However this is not always the case when the
helper xfs_ilock_for_iomap is used. There are cases where xfs_is_cow_inode(() is not
true.

I'll add a new helper that will be used here.

>>  
>>  	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
>>  	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
>> @@ -902,6 +912,11 @@ xfs_buffered_write_iomap_begin(
>>  
>>  	XFS_STATS_INC(mp, xs_blk_mapw);
>>  
>> +	if (no_wait && xfs_need_iread_extents(XFS_IFORK_PTR(ip, XFS_DATA_FORK))) {
>> +		error = -EAGAIN;
>> +		goto out_unlock;
>> +	}
>> +
> 
> Also handled by xfs_ilock_for_iomap().

I'll remove this check with the next version.

> 
>>  	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
>>  	if (error)
>>  		goto out_unlock;
>> @@ -933,6 +948,10 @@ xfs_buffered_write_iomap_begin(
>>  	if (xfs_is_cow_inode(ip)) {
>>  		if (!ip->i_cowfp) {
>>  			ASSERT(!xfs_is_reflink_inode(ip));
>> +			if (no_wait) {
>> +				error = -EAGAIN;
>> +				goto out_unlock;
>> +			}
>>  			xfs_ifork_init_cow(ip);
>>  		}
> 
> Also handled by xfs_ilock_for_iomap().
> 

I'll remove this check with the next version.

>>  		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
>> @@ -956,6 +975,11 @@ xfs_buffered_write_iomap_begin(
>>  			goto found_imap;
>>  		}
>>  
>> +		if (no_wait) {
>> +			error = -EAGAIN;
>> +			goto out_unlock;
>> +		}
>> +
> 
> Won't get here because this is COW path, and xfs_ilock_for_iomap()
> handles returning -EAGAIN for that case.

I'll remove this check with the next version.

> 
>>  		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
>>  
>>  		/* Trim the mapping to the nearest shared extent boundary. */
>> @@ -993,6 +1017,11 @@ xfs_buffered_write_iomap_begin(
>>  			allocfork = XFS_COW_FORK;
>>  	}
>>  
>> +	if (no_wait) {
>> +		error = -EAGAIN;
>> +		goto out_unlock;
>> +	}
> 
> Why? Delayed allocation doesn't block...

I'll remove this check with the next version.

> 
> Cheers,
> 
> Dave.
> 