Re: [PATCH 3/3] xfs: measure all contiguous previous extents for prealloc size

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Tue, May 19, 2020 at 05:54:37AM -0700, Christoph Hellwig wrote:
> The actual logic looks good, but I think the new helper and another
> third set of comment explaining what is going on makes this area even
> more confusing.  What about something like this instead?

This seems reasonable, but the callsite cleanups ought to be a separate
patch from the behavior change.

> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
> index bb590a267a7f9..26f9874361cd3 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -352,22 +352,10 @@ xfs_quota_calc_throttle(
>  }
>  
>  /*
> - * If we are doing a write at the end of the file and there are no allocations
> - * past this one, then extend the allocation out to the file system's write
> - * iosize.
> - *
>   * If we don't have a user specified preallocation size, dynamically increase
>   * the preallocation size as the size of the file grows.  Cap the maximum size
>   * at a single extent or less if the filesystem is near full. The closer the
> - * filesystem is to full, the smaller the maximum prealocation.
> - *
> - * As an exception we don't do any preallocation at all if the file is smaller
> - * than the minimum preallocation and we are using the default dynamic
> - * preallocation scheme, as it is likely this is the only write to the file that
> - * is going to be done.
> - *
> - * We clean up any extra space left over when the file is closed in
> - * xfs_inactive().
> + * filesystem is to full, the smaller the maximum preallocation.
>   */
>  STATIC xfs_fsblock_t
>  xfs_iomap_prealloc_size(
> @@ -380,52 +368,58 @@ xfs_iomap_prealloc_size(
>  	struct xfs_mount	*mp = ip->i_mount;
>  	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
>  	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
> -	struct xfs_bmbt_irec	prev;
> +	struct xfs_iext_cursor	ncur = *icur;
> +	struct xfs_bmbt_irec	prev, got;
>  	int			shift = 0;
>  	int64_t			freesp;
>  	xfs_fsblock_t		qblocks;
>  	int			qshift = 0;
> -	xfs_fsblock_t		alloc_blocks = 0;
> +	xfs_fsblock_t		alloc_blocks;
> +	xfs_extlen_t		plen;
>  
> -	if (offset + count <= XFS_ISIZE(ip))
> -		return 0;
> -
> -	if (!(mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
> -	    (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks)))
> +	/*
> +	 * As an exception we don't do any preallocation at all if the file is
> +	 * smaller than the minimum preallocation and we are using the default
> +	 * dynamic preallocation scheme, as it is likely this is the only write
> +	 * to the file that is going to be done.
> +	 */
> +	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
>  		return 0;
>  
>  	/*
> -	 * If an explicit allocsize is set, the file is small, or we
> -	 * are writing behind a hole, then use the minimum prealloc:
> +	 * Otherwise use the minimum prealloca size for small files, or if we

"preallocation"?

> +	 * are writing right after a hole.
>  	 */
> -	if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) ||
> -	    XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
> -	    !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
> +	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
> +	    !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
>  	    prev.br_startoff + prev.br_blockcount < offset_fsb)
>  		return mp->m_allocsize_blocks;
>  
>  	/*
> -	 * Determine the initial size of the preallocation. We are beyond the
> -	 * current EOF here, but we need to take into account whether this is
> -	 * a sparse write or an extending write when determining the
> -	 * preallocation size.  Hence we need to look up the extent that ends
> -	 * at the current write offset and use the result to determine the
> -	 * preallocation size.
> -	 *
> -	 * If the extent is a hole, then preallocation is essentially disabled.
> -	 * Otherwise we take the size of the preceding data extent as the basis
> -	 * for the preallocation size. If the size of the extent is greater than
> -	 * half the maximum extent length, then use the current offset as the
> -	 * basis. This ensures that for large files the preallocation size
> -	 * always extends to MAXEXTLEN rather than falling short due to things
> -	 * like stripe unit/width alignment of real extents.
> +	 * Take the size of the contiguous preceding data extents as the basis
> +	 * for the preallocation size.  Note that we don't care if the previous
> +	 * extents are written or not.
>  	 */
> -	if (prev.br_blockcount <= (MAXEXTLEN >> 1))
> -		alloc_blocks = prev.br_blockcount << 1;
> -	else
> +	plen = prev.br_blockcount;
> +	while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
> +		if (plen > MAXEXTLEN / 2 ||
> +		    got.br_startoff + got.br_blockcount != prev.br_startoff ||
> +		    got.br_startblock + got.br_blockcount != prev.br_startblock)
> +			break;
> +		plen += got.br_blockcount;
> +		prev = got;
> +	}
> +
> +	/*
> +	 * If the size of the extents is greater than half the maximum extent
> +	 * length, then use the current offset as the basis.  This ensures that
> +	 * for large files the preallocation size always extends to MAXEXTLEN
> +	 * rather than falling short due to things like stripe unit/width
> +	 * alignment of real extents.
> +	 */
> +	alloc_blocks = plen * 2;
> +	if (alloc_blocks > MAXEXTLEN)
>  		alloc_blocks = XFS_B_TO_FSB(mp, offset);
> -	if (!alloc_blocks)
> -		goto check_writeio;
>  	qblocks = alloc_blocks;
>  
>  	/*
> @@ -494,7 +488,6 @@ xfs_iomap_prealloc_size(
>  	 */
>  	while (alloc_blocks && alloc_blocks >= freesp)
>  		alloc_blocks >>= 4;
> -check_writeio:
>  	if (alloc_blocks < mp->m_allocsize_blocks)
>  		alloc_blocks = mp->m_allocsize_blocks;
>  	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
> @@ -961,9 +954,16 @@ xfs_buffered_write_iomap_begin(
>  	if (error)
>  		goto out_unlock;
>  
> -	if (eof) {
> -		prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork, offset,
> -				count, &icur);
> +	if (eof && offset + count > XFS_ISIZE(ip)) {
> +		/*
> +		 * Determine the initial size of the preallocation.
> + 		 * We clean up any extra preallocation when the file is closed.
> +		 */
> +		if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
> +			prealloc_blocks = mp->m_allocsize_blocks;
> +		else
> +			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
> +						offset, count, &icur);

I'm not sure how much we're really gaining from moving the
MOUNT_ALLOCSIZE check out to the caller, but I don't feel all that
passionate about this.

--D

>  		if (prealloc_blocks) {
>  			xfs_extlen_t	align;
>  			xfs_off_t	end_offset;