COW improvements and always_cow support V3

Opencoding this function in the only caller makes it blindly obvious what is going on instead of having to look at two files for that.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/libxfs/xfs_bmap.c | 11 -----------
 fs/xfs/libxfs/xfs_bmap.h |  1 -
 fs/xfs/xfs_aops.c        |  2 +-
 3 files changed, 1 insertion(+), 13 deletions(-)

fs/xfs/libxfs/xfs_bmap.c +0 -11

index 19e921d1586f..f16d42abc500 100644
@@ -3685,17 +3685,6 @@ xfs_trim_extent(
 	}
 }
 
-/* trim extent to within eof */
-void
-xfs_trim_extent_eof(
-	struct xfs_bmbt_irec	*irec,
-	struct xfs_inode	*ip)
-
-{
-	xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
-					      i_size_read(VFS_I(ip))));
-}
-
 /*
  * Trim the returned map to the required bounds
  */

fs/xfs/libxfs/xfs_bmap.h +0 -1

index 488dc8860fd7..f9a925caa70e 100644
@@ -181,7 +181,6 @@ static inline bool xfs_bmap_is_real_extent(struct xfs_bmbt_irec *irec)
 
 void	xfs_trim_extent(struct xfs_bmbt_irec *irec, xfs_fileoff_t bno,
 		xfs_filblks_t len);
-void	xfs_trim_extent_eof(struct xfs_bmbt_irec *, struct xfs_inode *);
 int	xfs_bmap_add_attrfork(struct xfs_inode *ip, int size, int rsvd);
 int	xfs_bmap_set_attrforkoff(struct xfs_inode *ip, int size, int *version);
 void	xfs_bmap_local_to_extents_empty(struct xfs_inode *ip, int whichfork);

fs/xfs/xfs_aops.c +1 -1

index 338b9d9984e0..d7275075878e 100644
@@ -329,7 +329,7 @@ xfs_map_blocks(
 	 * mechanism to protect us from arbitrary extent modifying contexts, not
 	 * just eofblocks.
 	 */
-	xfs_trim_extent_eof(&wpc->imap, ip);
+	xfs_trim_extent(&wpc->imap, 0, XFS_B_TO_FSB(mp, i_size_read(inode)));
 
 	/*
 	 * COW fork blocks can overlap data fork blocks even if the blocks

Signature · 1 line

2.19.1

The io_type field contains what is basically a summary of information from the inode fork and the imap. But we can just as easily use that information directly, simplifying a few bits here and there and improving the trace points.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_aops.c    | 91 ++++++++++++++++++++------------------------
 fs/xfs/xfs_aops.h    | 21 +---------
 fs/xfs/xfs_iomap.c   |  8 ++--
 fs/xfs/xfs_reflink.c |  2 +-
 fs/xfs/xfs_trace.h   | 28 +++++++-------
 5 files changed, 62 insertions(+), 88 deletions(-)

fs/xfs/xfs_aops.c +41 -50

index d7275075878e..8fec6fd4c632 100644
@@ -28,7 +28,7 @@
  */
 struct xfs_writepage_ctx {
 	struct xfs_bmbt_irec    imap;
-	unsigned int		io_type;
+	int			fork;
 	unsigned int		cow_seq;
 	struct xfs_ioend	*ioend;
 };

@@ -255,30 +255,20 @@ xfs_end_io(
 	 */
 	error = blk_status_to_errno(ioend->io_bio->bi_status);
 	if (unlikely(error)) {
-		switch (ioend->io_type) {
-		case XFS_IO_COW:
+		if (ioend->io_fork == XFS_COW_FORK)
 			xfs_reflink_cancel_cow_range(ip, offset, size, true);
-			break;
-		}
-
 		goto done;
 	}
 
 	/*
-	 * Success:  commit the COW or unwritten blocks if needed.
+	 * Success: commit the COW or unwritten blocks if needed.
 	 */
-	switch (ioend->io_type) {
-	case XFS_IO_COW:
+	if (ioend->io_fork == XFS_COW_FORK)
 		error = xfs_reflink_end_cow(ip, offset, size);
-		break;
-	case XFS_IO_UNWRITTEN:
-		/* writeback should never update isize */
+	else if (ioend->io_state == XFS_EXT_UNWRITTEN)
 		error = xfs_iomap_write_unwritten(ip, offset, size, false);
-		break;
-	default:
+	else
 		ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
-		break;
-	}
 
 done:
 	if (ioend->io_append_trans)

@@ -293,7 +283,8 @@ xfs_end_bio(
 	struct xfs_ioend	*ioend = bio->bi_private;
 	struct xfs_mount	*mp = XFS_I(ioend->io_inode)->i_mount;
 
-	if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
+	if (ioend->io_fork == XFS_COW_FORK ||
+	    ioend->io_state == XFS_EXT_UNWRITTEN)
 		queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
 	else if (ioend->io_append_trans)
 		queue_work(mp->m_data_workqueue, &ioend->io_work);

@@ -313,7 +304,6 @@ xfs_map_blocks(
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset), end_fsb;
 	xfs_fileoff_t		cow_fsb = NULLFILEOFF;
 	struct xfs_bmbt_irec	imap;
-	int			whichfork = XFS_DATA_FORK;
 	struct xfs_iext_cursor	icur;
 	bool			imap_valid;
 	int			error = 0;

@@ -350,7 +340,7 @@ xfs_map_blocks(
 		     offset_fsb < wpc->imap.br_startoff + wpc->imap.br_blockcount;
 	if (imap_valid &&
 	    (!xfs_inode_has_cow_data(ip) ||
-	     wpc->io_type == XFS_IO_COW ||
+	     wpc->fork == XFS_COW_FORK ||
 	     wpc->cow_seq == READ_ONCE(ip->i_cowfp->if_seq)))
 		return 0;
 

@@ -382,6 +372,9 @@ xfs_map_blocks(
 	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
 		wpc->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
+
+		wpc->fork = XFS_COW_FORK;
+
 		/*
 		 * Truncate can race with writeback since writeback doesn't
 		 * take the iolock and truncate decreases the file size before

@@ -394,11 +387,13 @@ xfs_map_blocks(
 		 * will kill the contents anyway.
 		 */
 		if (offset > i_size_read(inode)) {
-			wpc->io_type = XFS_IO_HOLE;
+			wpc->imap.br_blockcount = end_fsb - offset_fsb;
+			wpc->imap.br_startoff = offset_fsb;
+			wpc->imap.br_startblock = HOLESTARTBLOCK;
+			wpc->imap.br_state = XFS_EXT_NORM;
 			return 0;
 		}
-		whichfork = XFS_COW_FORK;
-		wpc->io_type = XFS_IO_COW;
+
 		goto allocate_blocks;
 	}
 

@@ -419,12 +414,14 @@ xfs_map_blocks(
 		imap.br_startoff = end_fsb;	/* fake a hole past EOF */
 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
+	wpc->fork = XFS_DATA_FORK;
+
 	if (imap.br_startoff > offset_fsb) {
 		/* landed in a hole or beyond EOF */
 		imap.br_blockcount = imap.br_startoff - offset_fsb;
 		imap.br_startoff = offset_fsb;
 		imap.br_startblock = HOLESTARTBLOCK;
-		wpc->io_type = XFS_IO_HOLE;
+		imap.br_state = XFS_EXT_NORM;
 	} else {
 		/*
 		 * Truncate to the next COW extent if there is one.  This is the

@@ -436,30 +433,23 @@ xfs_map_blocks(
 		    cow_fsb < imap.br_startoff + imap.br_blockcount)
 			imap.br_blockcount = cow_fsb - imap.br_startoff;
 
-		if (isnullstartblock(imap.br_startblock)) {
-			/* got a delalloc extent */
-			wpc->io_type = XFS_IO_DELALLOC;
+		/* got a delalloc extent? */
+		if (isnullstartblock(imap.br_startblock))
 			goto allocate_blocks;
-		}
-
-		if (imap.br_state == XFS_EXT_UNWRITTEN)
-			wpc->io_type = XFS_IO_UNWRITTEN;
-		else
-			wpc->io_type = XFS_IO_OVERWRITE;
 	}
 
 	wpc->imap = imap;
-	trace_xfs_map_blocks_found(ip, offset, count, wpc->io_type, &imap);
+	trace_xfs_map_blocks_found(ip, offset, count, wpc->fork, &imap);
 	return 0;
 allocate_blocks:
-	error = xfs_iomap_write_allocate(ip, whichfork, offset, &imap,
+	error = xfs_iomap_write_allocate(ip, wpc->fork, offset, &imap,
 			&wpc->cow_seq);
 	if (error)
 		return error;
-	ASSERT(whichfork == XFS_COW_FORK || cow_fsb == NULLFILEOFF ||
+	ASSERT(wpc->fork == XFS_COW_FORK || cow_fsb == NULLFILEOFF ||
 	       imap.br_startoff + imap.br_blockcount <= cow_fsb);
 	wpc->imap = imap;
-	trace_xfs_map_blocks_alloc(ip, offset, count, wpc->io_type, &imap);
+	trace_xfs_map_blocks_alloc(ip, offset, count, wpc->fork, &imap);
 	return 0;
 }
 

@@ -484,7 +474,7 @@ xfs_submit_ioend(
 	int			status)
 {
 	/* Convert CoW extents to regular */
-	if (!status && ioend->io_type == XFS_IO_COW) {
+	if (!status && ioend->io_fork == XFS_COW_FORK) {
 		/*
 		 * Yuk. This can do memory allocation, but is not a
 		 * transactional operation so everything is done in GFP_KERNEL

@@ -502,7 +492,8 @@ xfs_submit_ioend(
 
 	/* Reserve log space if we might write beyond the on-disk inode size. */
 	if (!status &&
-	    ioend->io_type != XFS_IO_UNWRITTEN &&
+	    (ioend->io_fork == XFS_COW_FORK ||
+	     ioend->io_state != XFS_EXT_UNWRITTEN) &&
 	    xfs_ioend_is_append(ioend) &&
 	    !ioend->io_append_trans)
 		status = xfs_setfilesize_trans_alloc(ioend);

@@ -531,7 +522,8 @@ xfs_submit_ioend(
 static struct xfs_ioend *
 xfs_alloc_ioend(
 	struct inode		*inode,
-	unsigned int		type,
+	int			fork,
+	xfs_exntst_t		state,
 	xfs_off_t		offset,
 	struct block_device	*bdev,
 	sector_t		sector)

@@ -545,7 +537,8 @@ xfs_alloc_ioend(
 
 	ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
 	INIT_LIST_HEAD(&ioend->io_list);
-	ioend->io_type = type;
+	ioend->io_fork = fork;
+	ioend->io_state = state;
 	ioend->io_inode = inode;
 	ioend->io_size = 0;
 	ioend->io_offset = offset;

@@ -606,13 +599,15 @@ xfs_add_to_ioend(
 	sector = xfs_fsb_to_db(ip, wpc->imap.br_startblock) +
 		((offset - XFS_FSB_TO_B(mp, wpc->imap.br_startoff)) >> 9);
 
-	if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
+	if (!wpc->ioend ||
+	    wpc->fork != wpc->ioend->io_fork ||
+	    wpc->imap.br_state != wpc->ioend->io_state ||
 	    sector != bio_end_sector(wpc->ioend->io_bio) ||
 	    offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
 		if (wpc->ioend)
 			list_add(&wpc->ioend->io_list, iolist);
-		wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset,
-				bdev, sector);
+		wpc->ioend = xfs_alloc_ioend(inode, wpc->fork,
+				wpc->imap.br_state, offset, bdev, sector);
 	}
 
 	if (!__bio_try_merge_page(wpc->ioend->io_bio, page, len, poff)) {

@@ -721,7 +716,7 @@ xfs_writepage_map(
 		error = xfs_map_blocks(wpc, inode, file_offset);
 		if (error)
 			break;
-		if (wpc->io_type == XFS_IO_HOLE)
+		if (wpc->imap.br_startblock == HOLESTARTBLOCK)
 			continue;
 		xfs_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
 				 &submit_list);

@@ -916,9 +911,7 @@ xfs_vm_writepage(
 	struct page		*page,
 	struct writeback_control *wbc)
 {
-	struct xfs_writepage_ctx wpc = {
-		.io_type = XFS_IO_HOLE,
-	};
+	struct xfs_writepage_ctx wpc = { };
 	int			ret;
 
 	ret = xfs_do_writepage(page, wbc, &wpc);

@@ -932,9 +925,7 @@ xfs_vm_writepages(
 	struct address_space	*mapping,
 	struct writeback_control *wbc)
 {
-	struct xfs_writepage_ctx wpc = {
-		.io_type = XFS_IO_HOLE,
-	};
+	struct xfs_writepage_ctx wpc = { };
 	int			ret;
 
 	xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);

fs/xfs/xfs_aops.h +2 -19

index 494b4338446e..6c2615b83c5d 100644
@@ -8,30 +8,13 @@
 
 extern struct bio_set xfs_ioend_bioset;
 
-/*
- * Types of I/O for bmap clustering and I/O completion tracking.
- */
-enum {
-	XFS_IO_HOLE,		/* covers region without any block allocation */
-	XFS_IO_DELALLOC,	/* covers delalloc region */
-	XFS_IO_UNWRITTEN,	/* covers allocated but uninitialized data */
-	XFS_IO_OVERWRITE,	/* covers already allocated extent */
-	XFS_IO_COW,		/* covers copy-on-write extent */
-};
-
-#define XFS_IO_TYPES \
-	{ XFS_IO_HOLE,			"hole" },	\
-	{ XFS_IO_DELALLOC,		"delalloc" },	\
-	{ XFS_IO_UNWRITTEN,		"unwritten" },	\
-	{ XFS_IO_OVERWRITE,		"overwrite" },	\
-	{ XFS_IO_COW,			"CoW" }
-
 /*
  * Structure for buffered I/O completions.
  */
 struct xfs_ioend {
 	struct list_head	io_list;	/* next ioend in chain */
-	unsigned int		io_type;	/* delalloc / unwritten */
+	int			io_fork;	/* inode fork written back */
+	xfs_exntst_t		io_state;	/* extent state */
 	struct inode		*io_inode;	/* file being written to */
 	size_t			io_size;	/* size of the extent */
 	xfs_off_t		io_offset;	/* offset in the file */

fs/xfs/xfs_iomap.c +4 -4

index 27c93b5f029d..32a7c169e096 100644
@@ -575,7 +575,7 @@ xfs_file_iomap_begin_delay(
 				goto out_unlock;
 		}
 
-		trace_xfs_iomap_found(ip, offset, count, 0, &got);
+		trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK, &got);
 		goto done;
 	}
 

@@ -647,7 +647,7 @@ xfs_file_iomap_begin_delay(
 	 * them out if the write happens to fail.
 	 */
 	iomap->flags |= IOMAP_F_NEW;
-	trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
+	trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &got);
 done:
 	if (isnullstartblock(got.br_startblock))
 		got.br_startblock = DELAYSTARTBLOCK;

@@ -1139,7 +1139,7 @@ xfs_file_iomap_begin(
 		return error;
 
 	iomap->flags |= IOMAP_F_NEW;
-	trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
+	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
 
 out_finish:
 	if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields

@@ -1155,7 +1155,7 @@ xfs_file_iomap_begin(
 out_found:
 	ASSERT(nimaps);
 	xfs_iunlock(ip, lockmode);
-	trace_xfs_iomap_found(ip, offset, length, 0, &imap);
+	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
 	goto out_finish;
 
 out_unlock:

fs/xfs/xfs_reflink.c +1 -1

index 322a852ce284..a8c32632090c 100644
@@ -1148,7 +1148,7 @@ xfs_reflink_remap_blocks(
 			break;
 		ASSERT(nimaps == 1);
 
-		trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
+		trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_DATA_FORK,
 				&imap);
 
 		/* Translate imap into the destination file. */

fs/xfs/xfs_trace.h +14 -14

index 8a6532aae779..870865913bd8 100644
@@ -1210,15 +1210,15 @@ DEFINE_READPAGE_EVENT(xfs_vm_readpages);
 
 DECLARE_EVENT_CLASS(xfs_imap_class,
 	TP_PROTO(struct xfs_inode *ip, xfs_off_t offset, ssize_t count,
-		 int type, struct xfs_bmbt_irec *irec),
-	TP_ARGS(ip, offset, count, type, irec),
+		 int whichfork, struct xfs_bmbt_irec *irec),
+	TP_ARGS(ip, offset, count, whichfork, irec),
 	TP_STRUCT__entry(
 		__field(dev_t, dev)
 		__field(xfs_ino_t, ino)
 		__field(loff_t, size)
 		__field(loff_t, offset)
 		__field(size_t, count)
-		__field(int, type)
+		__field(int, whichfork)
 		__field(xfs_fileoff_t, startoff)
 		__field(xfs_fsblock_t, startblock)
 		__field(xfs_filblks_t, blockcount)

@@ -1229,33 +1229,33 @@ DECLARE_EVENT_CLASS(xfs_imap_class,
 		__entry->size = ip->i_d.di_size;
 		__entry->offset = offset;
 		__entry->count = count;
-		__entry->type = type;
+		__entry->whichfork = whichfork;
 		__entry->startoff = irec ? irec->br_startoff : 0;
 		__entry->startblock = irec ? irec->br_startblock : 0;
 		__entry->blockcount = irec ? irec->br_blockcount : 0;
 	),
 	TP_printk("dev %d:%d ino 0x%llx size 0x%llx offset 0x%llx count %zd "
-		  "type %s startoff 0x%llx startblock %lld blockcount 0x%llx",
+		  "fork %s startoff 0x%llx startblock %lld blockcount 0x%llx",
 		  MAJOR(__entry->dev), MINOR(__entry->dev),
 		  __entry->ino,
 		  __entry->size,
 		  __entry->offset,
 		  __entry->count,
-		  __print_symbolic(__entry->type, XFS_IO_TYPES),
+		  __entry->whichfork == XFS_COW_FORK ? "cow" : "data",
 		  __entry->startoff,
 		  (int64_t)__entry->startblock,
 		  __entry->blockcount)
 )
 
-#define DEFINE_IOMAP_EVENT(name)	\
+#define DEFINE_IMAP_EVENT(name)	\
 DEFINE_EVENT(xfs_imap_class, name,	\
 	TP_PROTO(struct xfs_inode *ip, xfs_off_t offset, ssize_t count,	\
-		 int type, struct xfs_bmbt_irec *irec),		\
-	TP_ARGS(ip, offset, count, type, irec))
-DEFINE_IOMAP_EVENT(xfs_map_blocks_found);
-DEFINE_IOMAP_EVENT(xfs_map_blocks_alloc);
-DEFINE_IOMAP_EVENT(xfs_iomap_alloc);
-DEFINE_IOMAP_EVENT(xfs_iomap_found);
+		 int whichfork, struct xfs_bmbt_irec *irec),		\
+	TP_ARGS(ip, offset, count, whichfork, irec))
+DEFINE_IMAP_EVENT(xfs_map_blocks_found);
+DEFINE_IMAP_EVENT(xfs_map_blocks_alloc);
+DEFINE_IMAP_EVENT(xfs_iomap_alloc);
+DEFINE_IMAP_EVENT(xfs_iomap_found);
 
 DECLARE_EVENT_CLASS(xfs_simple_io_class,
 	TP_PROTO(struct xfs_inode *ip, xfs_off_t offset, ssize_t count),

@@ -3055,7 +3055,7 @@ DEFINE_EVENT(xfs_inode_irec_class, name, \
 DEFINE_INODE_EVENT(xfs_reflink_set_inode_flag);
 DEFINE_INODE_EVENT(xfs_reflink_unset_inode_flag);
 DEFINE_ITRUNC_EVENT(xfs_reflink_update_inode_size);
-DEFINE_IOMAP_EVENT(xfs_reflink_remap_imap);
+DEFINE_IMAP_EVENT(xfs_reflink_remap_imap);
 TRACE_EVENT(xfs_reflink_remap_blocks_loop,
 	TP_PROTO(struct xfs_inode *src, xfs_fileoff_t soffset,
 		 xfs_filblks_t len, struct xfs_inode *dest,

Signature · 1 line

2.19.1

We already ensure all data fits into s_maxbytes in the write / fault path. The only reason we have them here is that they were copy and pasted from xfs_bmapi_read when we stopped using that function.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_aops.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

fs/xfs/xfs_aops.c +2 -6

index 8fec6fd4c632..5b6fab283316 100644
@@ -301,7 +301,8 @@ xfs_map_blocks(
 	struct xfs_inode	*ip = XFS_I(inode);
 	struct xfs_mount	*mp = ip->i_mount;
 	ssize_t			count = i_blocksize(inode);
-	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset), end_fsb;
+	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
+	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
 	xfs_fileoff_t		cow_fsb = NULLFILEOFF;
 	struct xfs_bmbt_irec	imap;
 	struct xfs_iext_cursor	icur;

@@ -356,11 +357,6 @@ xfs_map_blocks(
 	xfs_ilock(ip, XFS_ILOCK_SHARED);
 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
 	       (ip->i_df.if_flags & XFS_IFEXTENTS));
-	ASSERT(offset <= mp->m_super->s_maxbytes);
-
-	if (offset > mp->m_super->s_maxbytes - count)
-		count = mp->m_super->s_maxbytes - offset;
-	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 
 	/*
 	 * Check if this is offset is covered by a COW extents, and if yes use

Signature · 1 line

2.19.1

We currently try to handle the races with truncate and COW to data fork conversion rather ad-hoc in a few places in the writeback path:

- xfs_map_blocks contains an i_size check for the COW fork only, and returns an imap containing a hole to make the writeback code skip the rest of the page
- xfs_iomap_write_allocate does another i_size check under ilock, and does an extent tree lookup to find the last extent to skip everthing beyond that, returning -EAGAIN if either is invalid to make the writeback code exit early
- xfs_bmapi_write can ignore holes for delalloc conversions, but only does so if called for the COW fork

Replace this with a coherent scheme:

- check i_size first in xfs_map_blocks, and skip any processing if we already are beyond i_size by presenting a hole until the end of the file to the caller
- in xfs_iomap_write_allocate check i_size again, and return -EAGAIN if we are beyond it now that we've taken ilock.
- Skip holes for all delalloc conversion in xfs_bmapi_write instead of doing a separate lookup before calling it
- in xfs_map_blocks retry the case where xfs_iomap_write_allocate could not perform a conversion one single time if we were on a COW fork to handle the race where an extent moved from the COW to the data fork, and else return a hole to skip writeback as we must have races with writeback

Overall this greatly simplifies the code, makes it more robust and also handles the COW to data fork race properly that we did not handle previosuly.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/libxfs/xfs_bmap.c |  27 ++++-----
 fs/xfs/xfs_aops.c        |  61 +++++++++++++-------
 fs/xfs/xfs_iomap.c       | 121 ++++++++++++---------------------------
 3 files changed, 87 insertions(+), 122 deletions(-)

fs/xfs/libxfs/xfs_bmap.c +10 -17

index f16d42abc500..1992ed8a60b0 100644
@@ -4305,28 +4305,21 @@ xfs_bmapi_write(
 		/* in hole or beyond EOF? */
 		if (eof || bma.got.br_startoff > bno) {
 			/*
-			 * CoW fork conversions should /never/ hit EOF or
-			 * holes.  There should always be something for us
-			 * to work on.
+			 * It is possible that the extents have changed since
+			 * we did the read call as we dropped the ilock for a
+			 * while.  We have to be careful about truncates or hole
+			 * punchs here - we are not allowed to allocate
+			 * non-delalloc blocks here.
+			 *
+			 * The only protection against truncation is the pages
+			 * for the range we are being asked to convert are
+			 * locked and hence a truncate will block on them
+			 * first.
 			 */
 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
 			         (flags & XFS_BMAPI_COWFORK)));
 
 			if (flags & XFS_BMAPI_DELALLOC) {
-				/*
-				 * For the COW fork we can reasonably get a
-				 * request for converting an extent that races
-				 * with other threads already having converted
-				 * part of it, as there converting COW to
-				 * regular blocks is not protected using the
-				 * IOLOCK.
-				 */
-				ASSERT(flags & XFS_BMAPI_COWFORK);
-				if (!(flags & XFS_BMAPI_COWFORK)) {
-					error = -EIO;
-					goto error0;
-				}
-
 				if (eof || bno >= end)
 					break;
 			} else {

fs/xfs/xfs_aops.c +1 -0

index 5b6fab283316..124b8de37115 100644
@@ -300,6 +300,7 @@ xfs_map_blocks(
 {
 	struct xfs_inode	*ip = XFS_I(inode);
 	struct xfs_mount	*mp = ip->i_mount;
+	loff_t			isize = i_size_read(inode);
 	ssize_t			count = i_blocksize(inode);
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);

fs/xfs/xfs_aops.c +23 -22

+	 * under the ilock later before doing delalloc conversions.
+	 */
+	if (offset > isize)
+		goto eof;
 
 	/*
 	 * We have to make sure the cached mapping is within EOF to protect

@@ -320,7 +330,7 @@ xfs_map_blocks(
 	 * mechanism to protect us from arbitrary extent modifying contexts, not
 	 * just eofblocks.
 	 */
-	xfs_trim_extent(&wpc->imap, 0, XFS_B_TO_FSB(mp, i_size_read(inode)));
+	xfs_trim_extent(&wpc->imap, 0, XFS_B_TO_FSB(mp, isize));
 
 	/*
 	 * COW fork blocks can overlap data fork blocks even if the blocks

@@ -354,6 +364,7 @@ xfs_map_blocks(
 	 * into real extents.  If we return without a valid map, it means we
 	 * landed in a hole and we skip the block.
 	 */
+retry:
 	xfs_ilock(ip, XFS_ILOCK_SHARED);
 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
 	       (ip->i_df.if_flags & XFS_IFEXTENTS));

@@ -370,26 +381,6 @@ xfs_map_blocks(
 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 		wpc->fork = XFS_COW_FORK;
-
-		/*
-		 * Truncate can race with writeback since writeback doesn't
-		 * take the iolock and truncate decreases the file size before
-		 * it starts truncating the pages between new_size and old_size.
-		 * Therefore, we can end up in the situation where writeback
-		 * gets a CoW fork mapping but the truncate makes the mapping
-		 * invalid and we end up in here trying to get a new mapping.
-		 * bail out here so that we simply never get a valid mapping
-		 * and so we drop the write altogether.  The page truncation
-		 * will kill the contents anyway.
-		 */
-		if (offset > i_size_read(inode)) {
-			wpc->imap.br_blockcount = end_fsb - offset_fsb;
-			wpc->imap.br_startoff = offset_fsb;
-			wpc->imap.br_startblock = HOLESTARTBLOCK;
-			wpc->imap.br_state = XFS_EXT_NORM;
-			return 0;
-		}
-
 		goto allocate_blocks;
 	}
 

@@ -440,13 +431,39 @@ xfs_map_blocks(
 allocate_blocks:
 	error = xfs_iomap_write_allocate(ip, wpc->fork, offset, &imap,
 			&wpc->cow_seq);
-	if (error)
+	if (error) {
+		if (error == -EAGAIN)
+			goto truncate_race;
 		return error;
+	}
 	ASSERT(wpc->fork == XFS_COW_FORK || cow_fsb == NULLFILEOFF ||
 	       imap.br_startoff + imap.br_blockcount <= cow_fsb);
 	wpc->imap = imap;
 	trace_xfs_map_blocks_alloc(ip, offset, count, wpc->fork, &imap);
 	return 0;
+
+truncate_race:
+	/*
+	 * If we failed to find the extent in the COW fork we might have raced
+	 * with a COW to data fork conversion or truncate.  Restart the lookup
+	 * to catch the extent in the data fork for the former case, but prevent
+	 * additional retries to avoid looping forever for the latter case.
+	 */
+	if (wpc->fork == XFS_COW_FORK && !retries++) {
+		imap_valid = false;
+		goto retry;
+	}
+eof:

fs/xfs/xfs_aops.c +0 -0

 }
 
 /*

fs/xfs/xfs_iomap.c +1 -4

index 32a7c169e096..6acfed2ae858 100644
@@ -685,14 +685,13 @@ xfs_iomap_write_allocate(
 {
 	xfs_mount_t	*mp = ip->i_mount;
 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
-	xfs_fileoff_t	offset_fsb, last_block;
+	xfs_fileoff_t	offset_fsb;
 	xfs_fileoff_t	end_fsb, map_start_fsb;
 	xfs_filblks_t	count_fsb;
 	xfs_trans_t	*tp;
 	int		nimaps;
 	int		error = 0;
 	int		flags = XFS_BMAPI_DELALLOC;
-	int		nres;
 
 	if (whichfork == XFS_COW_FORK)
 		flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;

@@ -712,95 +711,51 @@ xfs_iomap_write_allocate(
 
 	while (count_fsb != 0) {
 		/*
-		 * Set up a transaction with which to allocate the
-		 * backing store for the file.  Do allocations in a

fs/xfs/xfs_iomap.c +34 -73

-			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
-			/*
-			 * We have already reserved space for the extent and any
-			 * indirect blocks when creating the delalloc extent,
-			 * there is no need to reserve space in this transaction
-			 * again.
-			 */
-			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
-					0, XFS_TRANS_RESERVE, &tp);
-			if (error)
-				return error;
+		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
+				0, XFS_TRANS_RESERVE, &tp);
+		if (error)
+			return error;
 
-			xfs_ilock(ip, XFS_ILOCK_EXCL);
-			xfs_trans_ijoin(tp, ip, 0);
+		xfs_ilock(ip, XFS_ILOCK_EXCL);
 
-			/*
-			 * it is possible that the extents have changed since
-			 * we did the read call as we dropped the ilock for a
-			 * while. We have to be careful about truncates or hole
-			 * punchs here - we are not allowed to allocate
-			 * non-delalloc blocks here.
-			 *
-			 * The only protection against truncation is the pages
-			 * for the range we are being asked to convert are
-			 * locked and hence a truncate will block on them
-			 * first.
-			 *
-			 * As a result, if we go beyond the range we really
-			 * need and hit an delalloc extent boundary followed by
-			 * a hole while we have excess blocks in the map, we
-			 * will fill the hole incorrectly and overrun the
-			 * transaction reservation.
-			 *
-			 * Using a single map prevents this as we are forced to
-			 * check each map we look for overlap with the desired
-			 * range and abort as soon as we find it. Also, given
-			 * that we only return a single map, having one beyond
-			 * what we can return is probably a bit silly.
-			 *
-			 * We also need to check that we don't go beyond EOF;
-			 * this is a truncate optimisation as a truncate sets
-			 * the new file size before block on the pages we
-			 * currently have locked under writeback. Because they
-			 * are about to be tossed, we don't need to write them
-			 * back....
-			 */
-			nimaps = 1;
-			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
-			error = xfs_bmap_last_offset(ip, &last_block,
-							XFS_DATA_FORK);
-			if (error)
+		/*
+		 * We need to check that we don't go beyond EOF; this is a
+		 * truncate optimisation as a truncate sets the new file size
+		 * before block on the pages we currently have locked under
+		 * writeback.  Because they are about to be tossed, we don't
+		 * need to write them back....
+		 */
+		end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
+		if (map_start_fsb + count_fsb > end_fsb) {
+			count_fsb = end_fsb - map_start_fsb;
+			if (count_fsb == 0) {
+				error = -EAGAIN;
 				goto trans_cancel;
-
-			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
-			if ((map_start_fsb + count_fsb) > last_block) {
-				count_fsb = last_block - map_start_fsb;
-				if (count_fsb == 0) {
-					error = -EAGAIN;
-					goto trans_cancel;
-				}
 			}
+		}
 
-			/*
-			 * From this point onwards we overwrite the imap
-			 * pointer that the caller gave to us.
-			 */
-			error = xfs_bmapi_write(tp, ip, map_start_fsb,
-						count_fsb, flags, nres, imap,
-						&nimaps);
-			if (error)
-				goto trans_cancel;
+		nimaps = 1;
+		xfs_trans_ijoin(tp, ip, 0);
+		error = xfs_bmapi_write(tp, ip, map_start_fsb, count_fsb, flags,
+				XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK),
+				imap, &nimaps);
+		if (error)
+			goto trans_cancel;
 
-			error = xfs_trans_commit(tp);
-			if (error)
-				goto error0;
+		error = xfs_trans_commit(tp);
+		if (error)
+			goto error0;
 
-			if (whichfork == XFS_COW_FORK)
-				*cow_seq = READ_ONCE(ifp->if_seq);
-			xfs_iunlock(ip, XFS_ILOCK_EXCL);
-		}
+		if (whichfork == XFS_COW_FORK)
+			*cow_seq = READ_ONCE(ifp->if_seq);
+		xfs_iunlock(ip, XFS_ILOCK_EXCL);
+
+		if (nimaps == 0)
+			return -EAGAIN;
 
 		/*
 		 * See if we were able to allocate an extent that

Signature · 1 line

2.19.1

Quoted 6 lines

fs/xfs/xfs_aops.c +6 -0

+eof:
+	/*
+	 * If we raced with truncate there might be no data left at this offset.
+	 * In that case we need to return a hole so that the writeback code
+	 * skips writeback for the rest of the file.
+	 */

fs/xfs/xfs_aops.c xfs_map_blocks( · line 466

Quoted 5 lines

fs/xfs/xfs_aops.c +5 -0

+	wpc->imap.br_startoff = offset_fsb;
+	wpc->imap.br_blockcount = end_fsb - offset_fsb;
+	wpc->imap.br_startblock = HOLESTARTBLOCK;
+	wpc->imap.br_state = XFS_EXT_NORM;
+	return 0;

This function has become rather spaghetti-like. Any way we can clean this up reasonably?

fs/xfs/xfs_iomap.c xfs_iomap_write_allocate( · line 723

Quoted 2 lines

fs/xfs/xfs_iomap.c +0 -2

-		nimaps = 0;
-		while (nimaps == 0) {

This removal of the nimaps == 0 loop bothers me: why is doing so safe?

I see that we can return from xfs_bmapi_write with nimaps == 0 if something is trying to punch or truncate the range that we're writing back, but it also seems to me that bmapi_write can return zero mappings because xfs_bmapi_allocate() didn't find any blocks. I /think/ that's impossible because we're converting delalloc reservations and so we should never run out of space, right?

Anyway, when _write_allocate gets zero mappings, it'll return -EAGAIN to xfs_map_blocks, which will retry once to cover the case of racing with cow -> data fork remapping but otherwise it won't bother? And that's why it's fine that only to loop once?

Am I reasoning this correctly?

Move checking for invalid zero blocks and setting of various iomap flags into this helper. Also make it deal with "raw" delalloc extents to avoid clutter in the callers.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_iomap.c | 84 +++++++++++++++++++++-------------------------
 fs/xfs/xfs_iomap.h |  4 +--
 fs/xfs/xfs_pnfs.c  |  2 +-
 3 files changed, 41 insertions(+), 49 deletions(-)

fs/xfs/xfs_iomap.c +38 -46

index 6acfed2ae858..9f1fd224bb06 100644
@@ -35,18 +35,40 @@
 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
 						<< mp->m_writeio_log)
 
-void
+static int
+xfs_alert_fsblock_zero(
+	xfs_inode_t	*ip,
+	xfs_bmbt_irec_t	*imap)
+{
+	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
+			"Access to block zero in inode %llu "
+			"start_block: %llx start_off: %llx "
+			"blkcnt: %llx extent-state: %x",
+		(unsigned long long)ip->i_ino,
+		(unsigned long long)imap->br_startblock,
+		(unsigned long long)imap->br_startoff,
+		(unsigned long long)imap->br_blockcount,
+		imap->br_state);
+	return -EFSCORRUPTED;
+}
+
+int
 xfs_bmbt_to_iomap(
 	struct xfs_inode	*ip,
 	struct iomap		*iomap,
-	struct xfs_bmbt_irec	*imap)
+	struct xfs_bmbt_irec	*imap,
+	bool			shared)
 {
 	struct xfs_mount	*mp = ip->i_mount;
 
+	if (unlikely(!imap->br_startblock && !XFS_IS_REALTIME_INODE(ip)))
+		return xfs_alert_fsblock_zero(ip, imap);
+
 	if (imap->br_startblock == HOLESTARTBLOCK) {
 		iomap->addr = IOMAP_NULL_ADDR;
 		iomap->type = IOMAP_HOLE;
-	} else if (imap->br_startblock == DELAYSTARTBLOCK) {
+	} else if (imap->br_startblock == DELAYSTARTBLOCK ||
+		   isnullstartblock(imap->br_startblock)) {
 		iomap->addr = IOMAP_NULL_ADDR;
 		iomap->type = IOMAP_DELALLOC;
 	} else {

@@ -60,6 +82,13 @@ xfs_bmbt_to_iomap(
 	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
 	iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
 	iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
+
+	if (xfs_ipincount(ip) &&
+	    (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
+		iomap->flags |= IOMAP_F_DIRTY;
+	if (shared)
+		iomap->flags |= IOMAP_F_SHARED;
+	return 0;
 }
 
 static void

@@ -138,23 +167,6 @@ xfs_iomap_eof_align_last_fsb(
 	return 0;
 }
 
-STATIC int
-xfs_alert_fsblock_zero(
-	xfs_inode_t	*ip,
-	xfs_bmbt_irec_t	*imap)
-{
-	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
-			"Access to block zero in inode %llu "
-			"start_block: %llx start_off: %llx "
-			"blkcnt: %llx extent-state: %x",
-		(unsigned long long)ip->i_ino,
-		(unsigned long long)imap->br_startblock,
-		(unsigned long long)imap->br_startoff,
-		(unsigned long long)imap->br_blockcount,
-		imap->br_state);
-	return -EFSCORRUPTED;
-}
-
 int
 xfs_iomap_write_direct(
 	xfs_inode_t	*ip,

@@ -649,17 +661,7 @@ xfs_file_iomap_begin_delay(
 	iomap->flags |= IOMAP_F_NEW;
 	trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &got);
 done:
-	if (isnullstartblock(got.br_startblock))
-		got.br_startblock = DELAYSTARTBLOCK;
-
-	if (!got.br_startblock) {
-		error = xfs_alert_fsblock_zero(ip, &got);
-		if (error)
-			goto out_unlock;
-	}
-
-	xfs_bmbt_to_iomap(ip, iomap, &got);
-
+	error = xfs_bmbt_to_iomap(ip, iomap, &got, false);
 out_unlock:
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;

@@ -1097,15 +1099,7 @@ xfs_file_iomap_begin(
 	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
 
 out_finish:
-	if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields
-				& ~XFS_ILOG_TIMESTAMP))
-		iomap->flags |= IOMAP_F_DIRTY;
-
-	xfs_bmbt_to_iomap(ip, iomap, &imap);
-
-	if (shared)
-		iomap->flags |= IOMAP_F_SHARED;
-	return 0;
+	return xfs_bmbt_to_iomap(ip, iomap, &imap, shared);
 
 out_found:
 	ASSERT(nimaps);

@@ -1228,12 +1222,10 @@ xfs_xattr_iomap_begin(
 out_unlock:
 	xfs_iunlock(ip, lockmode);
 
-	if (!error) {
-		ASSERT(nimaps);
-		xfs_bmbt_to_iomap(ip, iomap, &imap);
-	}
-
-	return error;
+	if (error)
+		return error;
+	ASSERT(nimaps);
+	return xfs_bmbt_to_iomap(ip, iomap, &imap, false);
 }
 
 const struct iomap_ops xfs_xattr_iomap_ops = {

fs/xfs/xfs_iomap.h +2 -2

index c6170548831b..ed27e41b687c 100644
@@ -17,8 +17,8 @@ int xfs_iomap_write_allocate(struct xfs_inode *, int, xfs_off_t,
 			struct xfs_bmbt_irec *, unsigned int *);
 int xfs_iomap_write_unwritten(struct xfs_inode *, xfs_off_t, xfs_off_t, bool);
 
-void xfs_bmbt_to_iomap(struct xfs_inode *, struct iomap *,
-		struct xfs_bmbt_irec *);
+int xfs_bmbt_to_iomap(struct xfs_inode *, struct iomap *,
+		struct xfs_bmbt_irec *, bool shared);
 xfs_extlen_t xfs_eof_alignment(struct xfs_inode *ip, xfs_extlen_t extsize);
 
 static inline xfs_filblks_t

fs/xfs/xfs_pnfs.c +1 -1

index f44c3599527d..bde2c9f56a46 100644
@@ -185,7 +185,7 @@ xfs_fs_map_blocks(
 	}
 	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 
-	xfs_bmbt_to_iomap(ip, iomap, &imap);
+	error = xfs_bmbt_to_iomap(ip, iomap, &imap, false);
 	*device_generation = mp->m_generation;
 	return error;
 out_unlock:

Signature · 1 line

2.19.1

While using delalloc for extsize hints is generally a good idea, the current code that does so only for COW doesn't help us much and creates a lot of special cases. Switch it to use real allocations like we do for direct I/O.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_iomap.c   | 28 +++++++++++++++++-----------
 fs/xfs/xfs_reflink.c |  5 ++++-
 fs/xfs/xfs_reflink.h |  5 ++---
 3 files changed, 23 insertions(+), 15 deletions(-)

fs/xfs/xfs_iomap.c +17 -11

index 9f1fd224bb06..d851abac16a9 100644
@@ -1039,22 +1039,28 @@ xfs_file_iomap_begin(
 	 * been done up front, so we don't need to do them here.
 	 */
 	if (xfs_is_reflink_inode(ip)) {
+		struct xfs_bmbt_irec	orig = imap;
+
 		/* if zeroing doesn't need COW allocation, then we are done. */
 		if ((flags & IOMAP_ZERO) &&
 		    !needs_cow_for_zeroing(&imap, nimaps))
 			goto out_found;
 
-		if (flags & IOMAP_DIRECT) {
-			/* may drop and re-acquire the ilock */
-			error = xfs_reflink_allocate_cow(ip, &imap, &shared,
-					&lockmode);
-			if (error)
-				goto out_unlock;
-		} else {
-			error = xfs_reflink_reserve_cow(ip, &imap);
-			if (error)
-				goto out_unlock;
-		}
+		error = xfs_reflink_allocate_cow(ip, &imap, &shared, &lockmode,
+						 flags);
+		if (error)
+			goto out_unlock;
+
+		/*
+		 * For buffered writes we need to report the address of the
+		 * previous block (if there was any) so that the higher level
+		 * write code can perform read-modify-write operations.  For
+		 * direct I/O code, which must be block aligned we need to
+		 * report the newly allocated address.
+		 */
+		if (!(flags & IOMAP_DIRECT) &&
+		    orig.br_startblock != HOLESTARTBLOCK)
+			imap = orig;
 
 		end_fsb = imap.br_startoff + imap.br_blockcount;
 		length = XFS_FSB_TO_B(mp, end_fsb) - offset;

fs/xfs/xfs_reflink.c +0 -0

index a8c32632090c..bdbaff1b3fb7 100644

fs/xfs/xfs_reflink.c +1 -0

+		return 0;
 	return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb);
 
 out_unreserve:

fs/xfs/xfs_reflink.h +2 -3

index 6d73daef1f13..d76fc520cac8 100644
@@ -12,10 +12,9 @@ extern int xfs_reflink_find_shared(struct xfs_mount *mp, struct xfs_trans *tp,
 extern int xfs_reflink_trim_around_shared(struct xfs_inode *ip,
 		struct xfs_bmbt_irec *irec, bool *shared);
 
-extern int xfs_reflink_reserve_cow(struct xfs_inode *ip,
-		struct xfs_bmbt_irec *imap);
 extern int xfs_reflink_allocate_cow(struct xfs_inode *ip,
-		struct xfs_bmbt_irec *imap, bool *shared, uint *lockmode);
+		struct xfs_bmbt_irec *imap, bool *shared, uint *lockmode,
+		unsigned flags);
 extern int xfs_reflink_convert_cow(struct xfs_inode *ip, xfs_off_t offset,
 		xfs_off_t count);
 

Signature · 1 line

2.19.1

fs/xfs/xfs_reflink.c xfs_reflink_allocate_cow( · line 401

Quoted 2 lines

fs/xfs/xfs_reflink.c +2 -0

+	uint			*lockmode,
+	unsigned		flags)

I'm not thrilled with passing iomap flags into the reflink code here...

...because I feel that it's easy to miss the subtlety here that for buffered writes we don't care if the cow extent is unwritten or written, but for directio we very /much/ care that the cow extent is written, because we're writing to it immediately. Can this grow a comment to reinforce why we skip the conversion?

Also, can we call this 'iomap_flags' to make it clearer which flags we're talking about?

Quoted 4 lines · Darrick J. Wong

fs/xfs/xfs_reflink.c xfs_reflink_allocate_cow( · line 401

Quoted 2 lines

fs/xfs/xfs_reflink.c +2 -0

+	uint			*lockmode,
+	unsigned		flags)

I'm not thrilled with passing iomap flags into the reflink code here...

So what is the alternative? A non-descriptive bool argument?

Quoted 5 lines

...because I feel that it's easy to miss the subtlety here that for buffered writes we don't care if the cow extent is unwritten or written, but for directio we very /much/ care that the cow extent is written, because we're writing to it immediately. Can this grow a comment to reinforce why we skip the conversion?

Sure.

Quoted 2 lines

Also, can we call this 'iomap_flags' to make it clearer which flags we're talking about?

Sure.

This only matters if we want to write data through the COW fork that is not actually an overwrite of existing data. Reasons for that are speculative COW fork allocations using the cowextsize, or a mode where we always write through the COW fork. Currently both can't actually happen, but I plan to enable them.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_aops.c | 31 ++++++++++++++++---------------
 1 file changed, 16 insertions(+), 15 deletions(-)

fs/xfs/xfs_aops.c +16 -15

index 124b8de37115..7d95a84064e7 100644
@@ -403,28 +403,29 @@ xfs_map_blocks(
 
 	wpc->fork = XFS_DATA_FORK;
 
+	/* landed in a hole or beyond EOF? */
 	if (imap.br_startoff > offset_fsb) {
-		/* landed in a hole or beyond EOF */
 		imap.br_blockcount = imap.br_startoff - offset_fsb;
 		imap.br_startoff = offset_fsb;
 		imap.br_startblock = HOLESTARTBLOCK;
 		imap.br_state = XFS_EXT_NORM;
-	} else {
-		/*
-		 * Truncate to the next COW extent if there is one.  This is the
-		 * only opportunity to do this because we can skip COW fork
-		 * lookups for the subsequent blocks in the mapping; however,
-		 * the requirement to treat the COW range separately remains.
-		 */
-		if (cow_fsb != NULLFILEOFF &&
-		    cow_fsb < imap.br_startoff + imap.br_blockcount)
-			imap.br_blockcount = cow_fsb - imap.br_startoff;
-
-		/* got a delalloc extent? */
-		if (isnullstartblock(imap.br_startblock))
-			goto allocate_blocks;
 	}
 
+	/*
+	 * Truncate to the next COW extent if there is one.  This is the only
+	 * opportunity to do this because we can skip COW fork lookups for the
+	 * subsequent blocks in the mapping; however, the requirement to treat
+	 * the COW range separately remains.
+	 */
+	if (cow_fsb != NULLFILEOFF &&
+	    cow_fsb < imap.br_startoff + imap.br_blockcount)
+		imap.br_blockcount = cow_fsb - imap.br_startoff;
+
+	/* got a delalloc extent? */
+	if (imap.br_startblock != HOLESTARTBLOCK &&
+	    isnullstartblock(imap.br_startblock))
+		goto allocate_blocks;
+
 	wpc->imap = imap;
 	trace_xfs_map_blocks_found(ip, offset, count, wpc->fork, &imap);
 	return 0;

Signature · 1 line

2.19.1

Besides simplifying the code a bit this allows to actually implement the behavior of using COW preallocation for non-COW data mentioned in the current comments.

Note that this breaks the current version of xfs/420, but that is because the test is broken. A separate fix will be sent for it.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_iomap.c   | 132 ++++++++++++++++++++++++++++++-------------
 fs/xfs/xfs_reflink.c |  67 ----------------------
 fs/xfs/xfs_trace.h   |   1 -
 3 files changed, 93 insertions(+), 107 deletions(-)

fs/xfs/xfs_iomap.c +14 -8

index d851abac16a9..d19f99e5476a 100644
@@ -534,15 +534,16 @@ xfs_file_iomap_begin_delay(
 {
 	struct xfs_inode	*ip = XFS_I(inode);
 	struct xfs_mount	*mp = ip->i_mount;
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	xfs_fileoff_t		maxbytes_fsb =
 		XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 	xfs_fileoff_t		end_fsb;
-	int			error = 0, eof = 0;
-	struct xfs_bmbt_irec	got;
-	struct xfs_iext_cursor	icur;
+	struct xfs_bmbt_irec	imap, cmap;
+	struct xfs_iext_cursor	icur, ccur;
 	xfs_fsblock_t		prealloc_blocks = 0;
+	bool			eof = false, cow_eof = false, shared;
+	int			whichfork = XFS_DATA_FORK;
+	int			error = 0;
 
 	ASSERT(!XFS_IS_REALTIME_INODE(ip));
 	ASSERT(!xfs_get_extsz_hint(ip));

@@ -560,7 +561,7 @@ xfs_file_iomap_begin_delay(
 
 	XFS_STATS_INC(mp, xs_blk_mapw);
 
-	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
+	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
 		if (error)
 			goto out_unlock;

@@ -568,51 +569,92 @@ xfs_file_iomap_begin_delay(
 
 	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 
-	eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
+	/*
+	 * Search the data fork fork first to look up our source mapping.  We
+	 * always need the data fork map, as we have to return it to the
+	 * iomap code so that the higher level write code can read data in to
+	 * perform read-modify-write cycles for unaligned writes.
+	 */
+	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
 	if (eof)
-		got.br_startoff = end_fsb; /* fake hole until the end */
+		imap.br_startoff = end_fsb; /* fake hole until the end */
 
-	if (got.br_startoff <= offset_fsb) {

fs/xfs/xfs_iomap.c +57 -29

+	 * block adjacent to shared blocks instead of just the shared blocks
+	 * themselves.  Second the lookup in the extent list is generally faster
+	 * than going out to the shared extent tree.
+	 */
+	if (xfs_is_reflink_inode(ip)) {
+		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
+				&ccur, &cmap);
+		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
+			trace_xfs_reflink_cow_found(ip, &cmap);
+			whichfork = XFS_COW_FORK;
+			goto done;
+		}
+	}
+
+	if (imap.br_startoff <= offset_fsb) {
 		/*
 		 * For reflink files we may need a delalloc reservation when
 		 * overwriting shared extents.   This includes zeroing of
 		 * existing extents that contain data.
 		 */
-		if (xfs_is_reflink_inode(ip) &&
-		    ((flags & IOMAP_WRITE) ||
-		     got.br_state != XFS_EXT_UNWRITTEN)) {
-			xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
-			error = xfs_reflink_reserve_cow(ip, &got);
-			if (error)
-				goto out_unlock;
+		if (!xfs_is_reflink_inode(ip) ||
+		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
+			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
+					&imap);
+			goto done;
 		}
 
-		trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK, &got);
-		goto done;
-	}
+		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
 
-	if (flags & IOMAP_ZERO) {
-		xfs_hole_to_iomap(ip, iomap, offset_fsb, got.br_startoff);
-		goto out_unlock;
+		/* Trim the mapping to the nearest shared extent boundary. */
+		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
+		if (error)
+			goto out_unlock;
+
+		/* Not shared?  Just report the (potentially capped) extent. */
+		if (!shared) {
+			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
+					&imap);
+			goto done;
+		}
+
+		/*
+		 * Fork all the shared blocks from our write offset until the
+		 * end of the extent.
+		 */
+		whichfork = XFS_COW_FORK;
+		end_fsb = imap.br_startoff + imap.br_blockcount;
+	} else {
+		/*
+		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
+		 * pages to keep the chunks of work done where somewhat
+		 * symmetric with the work writeback does.  This is a completely
+		 * arbitrary number pulled out of thin air.
+		 *
+		 * Note that the values needs to be less than 32-bits wide until
+		 * the lower level functions are updated.
+		 */
+		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
+		end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 	}
 
 	error = xfs_qm_dqattach_locked(ip, false);
 	if (error)
 		goto out_unlock;
 
-	/*
-	 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
-	 * to keep the chunks of work done where somewhat symmetric with the
-	 * work writeback does. This is a completely arbitrary number pulled
-	 * out of thin air as a best guess for initial testing.
-	 *
-	 * Note that the values needs to be less than 32-bits wide until
-	 * the lower level functions are updated.
-	 */
-	count = min_t(loff_t, count, 1024 * PAGE_SIZE);
-	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
-
-	if (eof) {
+	if (eof && whichfork == XFS_DATA_FORK) {
 		prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
 				&icur);
 		if (prealloc_blocks) {

@@ -635,9 +677,11 @@ xfs_file_iomap_begin_delay(
 	}
 
 retry:
-	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
-			end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
-			eof);
+	error = xfs_bmapi_reserve_delalloc(ip, whichfork, offset_fsb,
+			end_fsb - offset_fsb, prealloc_blocks,
+			whichfork == XFS_DATA_FORK ? &imap : &cmap,
+			whichfork == XFS_DATA_FORK ? &icur : &ccur,
+			whichfork == XFS_DATA_FORK ? eof : cow_eof);
 	switch (error) {
 	case 0:
 		break;

fs/xfs/xfs_iomap.c +11 -1

 done:
-	error = xfs_bmbt_to_iomap(ip, iomap, &got, false);
+	if (whichfork == XFS_COW_FORK) {
+		if (imap.br_startoff > offset_fsb) {
+			xfs_trim_extent(&cmap, offset_fsb,
+					imap.br_startoff - offset_fsb);
+			error = xfs_bmbt_to_iomap(ip, iomap, &cmap, false);
+			goto out_unlock;
+		}
+		/* ensure we only report blocks we have a reservation for */
+		xfs_trim_extent(&imap, cmap.br_startoff, cmap.br_blockcount);
+	}
+	error = xfs_bmbt_to_iomap(ip, iomap, &imap, false);
 out_unlock:
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;

fs/xfs/xfs_reflink.c +0 -67

index bdbaff1b3fb7..d59b556d42cb 100644
@@ -234,73 +234,6 @@ xfs_reflink_trim_around_shared(
 	}
 }
 
-/*
- * Trim the passed in imap to the next shared/unshared extent boundary, and
- * if imap->br_startoff points to a shared extent reserve space for it in the
- * COW fork.
- *
- * Note that imap will always contain the block numbers for the existing blocks
- * in the data fork, as the upper layers need them for read-modify-write
- * operations.
- */
-int
-xfs_reflink_reserve_cow(
-	struct xfs_inode	*ip,
-	struct xfs_bmbt_irec	*imap)
-{
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
-	struct xfs_bmbt_irec	got;
-	int			error = 0;
-	bool			eof = false;
-	struct xfs_iext_cursor	icur;
-	bool			shared;
-
-	/*
-	 * Search the COW fork extent list first.  This serves two purposes:
-	 * first this implement the speculative preallocation using cowextisze,
-	 * so that we also unshared block adjacent to shared blocks instead
-	 * of just the shared blocks themselves.  Second the lookup in the
-	 * extent list is generally faster than going out to the shared extent
-	 * tree.
-	 */
-
-	if (!xfs_iext_lookup_extent(ip, ifp, imap->br_startoff, &icur, &got))
-		eof = true;
-	if (!eof && got.br_startoff <= imap->br_startoff) {
-		trace_xfs_reflink_cow_found(ip, imap);
-		xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
-		return 0;
-	}
-
-	/* Trim the mapping to the nearest shared extent boundary. */
-	error = xfs_reflink_trim_around_shared(ip, imap, &shared);
-	if (error)
-		return error;
-
-	/* Not shared?  Just report the (potentially capped) extent. */
-	if (!shared)
-		return 0;
-
-	/*
-	 * Fork all the shared blocks from our write offset until the end of
-	 * the extent.
-	 */
-	error = xfs_qm_dqattach_locked(ip, false);
-	if (error)
-		return error;
-
-	error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, imap->br_startoff,
-			imap->br_blockcount, 0, &got, &icur, eof);
-	if (error == -ENOSPC || error == -EDQUOT)
-		trace_xfs_reflink_cow_enospc(ip, imap);
-	if (error)
-		return error;
-
-	xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
-	trace_xfs_reflink_cow_alloc(ip, &got);
-	return 0;
-}
-
 /* Convert part of an unwritten CoW extent to a real one. */
 STATIC int
 xfs_reflink_convert_cow_extent(

fs/xfs/xfs_trace.h +0 -1

index 870865913bd8..36e74fc90700 100644
@@ -3179,7 +3179,6 @@ DEFINE_INODE_ERROR_EVENT(xfs_reflink_unshare_error);
 
 /* copy on write */
 DEFINE_INODE_IREC_EVENT(xfs_reflink_trim_around_shared);
-DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_alloc);
 DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_found);
 DEFINE_INODE_IREC_EVENT(xfs_reflink_cow_enospc);
 DEFINE_INODE_IREC_EVENT(xfs_reflink_convert_cow);

Signature · 1 line

2.19.1

Quoted 6 lines · Christoph Hellwig

Besides simplifying the code a bit this allows to actually implement the behavior of using COW preallocation for non-COW data mentioned in the current comments.

Note that this breaks the current version of xfs/420, but that is because the test is broken. A separate fix will be sent for it.

(Still waiting for this...)

Quoted 59 lines

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_iomap.c   | 132 ++++++++++++++++++++++++++++++-------------
 fs/xfs/xfs_reflink.c |  67 ----------------------
 fs/xfs/xfs_trace.h   |   1 -
 3 files changed, 93 insertions(+), 107 deletions(-)

fs/xfs/xfs_iomap.c +14 -8

index d851abac16a9..d19f99e5476a 100644
@@ -534,15 +534,16 @@ xfs_file_iomap_begin_delay(
 {
 	struct xfs_inode	*ip = XFS_I(inode);
 	struct xfs_mount	*mp = ip->i_mount;
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	xfs_fileoff_t		maxbytes_fsb =
 		XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 	xfs_fileoff_t		end_fsb;
-	int			error = 0, eof = 0;
-	struct xfs_bmbt_irec	got;
-	struct xfs_iext_cursor	icur;
+	struct xfs_bmbt_irec	imap, cmap;
+	struct xfs_iext_cursor	icur, ccur;
 	xfs_fsblock_t		prealloc_blocks = 0;
+	bool			eof = false, cow_eof = false, shared;
+	int			whichfork = XFS_DATA_FORK;
+	int			error = 0;
 
 	ASSERT(!XFS_IS_REALTIME_INODE(ip));
 	ASSERT(!xfs_get_extsz_hint(ip));

@@ -560,7 +561,7 @@ xfs_file_iomap_begin_delay(
 
 	XFS_STATS_INC(mp, xs_blk_mapw);
 
-	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
+	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
 		if (error)
 			goto out_unlock;

@@ -568,51 +569,92 @@ xfs_file_iomap_begin_delay(
 
 	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 
-	eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
+	/*
+	 * Search the data fork fork first to look up our source mapping.  We
+	 * always need the data fork map, as we have to return it to the
+	 * iomap code so that the higher level write code can read data in to
+	 * perform read-modify-write cycles for unaligned writes.
+	 */
+	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
 	if (eof)
-		got.br_startoff = end_fsb; /* fake hole until the end */
+		imap.br_startoff = end_fsb; /* fake hole until the end */
 
-	if (got.br_startoff <= offset_fsb) {

fs/xfs/xfs_iomap.c xfs_file_iomap_begin_delay( · line 591

Quoted 10 lines

fs/xfs/xfs_iomap.c +10 -0

+	/* We never need to allocate blocks for zeroing a hole. */
+	if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
+		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
+		goto out_unlock;
+	}
+
+	/*
+	 * Search the COW fork extent list even if we did not find a data fork
+	 * extent.  This serves two purposes: first this implements the
+	 * speculative preallocation using cowextisze, so that we also unshare

"cowextsize"

Quoted 111 lines

fs/xfs/xfs_iomap.c +57 -29

+	 * block adjacent to shared blocks instead of just the shared blocks
+	 * themselves.  Second the lookup in the extent list is generally faster
+	 * than going out to the shared extent tree.
+	 */
+	if (xfs_is_reflink_inode(ip)) {
+		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
+				&ccur, &cmap);
+		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
+			trace_xfs_reflink_cow_found(ip, &cmap);
+			whichfork = XFS_COW_FORK;
+			goto done;
+		}
+	}
+
+	if (imap.br_startoff <= offset_fsb) {
 		/*
 		 * For reflink files we may need a delalloc reservation when
 		 * overwriting shared extents.   This includes zeroing of
 		 * existing extents that contain data.
 		 */
-		if (xfs_is_reflink_inode(ip) &&
-		    ((flags & IOMAP_WRITE) ||
-		     got.br_state != XFS_EXT_UNWRITTEN)) {
-			xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
-			error = xfs_reflink_reserve_cow(ip, &got);
-			if (error)
-				goto out_unlock;
+		if (!xfs_is_reflink_inode(ip) ||
+		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
+			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
+					&imap);
+			goto done;
 		}
 
-		trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK, &got);
-		goto done;
-	}
+		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
 
-	if (flags & IOMAP_ZERO) {
-		xfs_hole_to_iomap(ip, iomap, offset_fsb, got.br_startoff);
-		goto out_unlock;
+		/* Trim the mapping to the nearest shared extent boundary. */
+		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
+		if (error)
+			goto out_unlock;
+
+		/* Not shared?  Just report the (potentially capped) extent. */
+		if (!shared) {
+			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
+					&imap);
+			goto done;
+		}
+
+		/*
+		 * Fork all the shared blocks from our write offset until the
+		 * end of the extent.
+		 */
+		whichfork = XFS_COW_FORK;
+		end_fsb = imap.br_startoff + imap.br_blockcount;
+	} else {
+		/*
+		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
+		 * pages to keep the chunks of work done where somewhat
+		 * symmetric with the work writeback does.  This is a completely
+		 * arbitrary number pulled out of thin air.
+		 *
+		 * Note that the values needs to be less than 32-bits wide until
+		 * the lower level functions are updated.
+		 */
+		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
+		end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 	}
 
 	error = xfs_qm_dqattach_locked(ip, false);
 	if (error)
 		goto out_unlock;
 
-	/*
-	 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
-	 * to keep the chunks of work done where somewhat symmetric with the
-	 * work writeback does. This is a completely arbitrary number pulled
-	 * out of thin air as a best guess for initial testing.
-	 *
-	 * Note that the values needs to be less than 32-bits wide until
-	 * the lower level functions are updated.
-	 */
-	count = min_t(loff_t, count, 1024 * PAGE_SIZE);
-	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
-
-	if (eof) {
+	if (eof && whichfork == XFS_DATA_FORK) {
 		prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
 				&icur);
 		if (prealloc_blocks) {

@@ -635,9 +677,11 @@ xfs_file_iomap_begin_delay(
 	}
 
 retry:
-	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
-			end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
-			eof);
+	error = xfs_bmapi_reserve_delalloc(ip, whichfork, offset_fsb,
+			end_fsb - offset_fsb, prealloc_blocks,
+			whichfork == XFS_DATA_FORK ? &imap : &cmap,
+			whichfork == XFS_DATA_FORK ? &icur : &ccur,
+			whichfork == XFS_DATA_FORK ? eof : cow_eof);
 	switch (error) {
 	case 0:
 		break;

@@ -659,9 +703,19 @@ xfs_file_iomap_begin_delay(

fs/xfs/xfs_iomap.c xfs_file_iomap_begin_delay( · line 706

Quoted 5 lines

fs/xfs/xfs_iomap.c +1 -1

 	 * them out if the write happens to fail.
 	 */
 	iomap->flags |= IOMAP_F_NEW;
-	trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &got);
+	trace_xfs_iomap_alloc(ip, offset, count, whichfork, &imap);

I'm confused by this, if whichfork == COW then won't ftrace report results from the wrong fork?

Quoted 9 lines · Darrick J. Wong

Quoted 6 lines · Christoph Hellwig

Besides simplifying the code a bit this allows to actually implement the behavior of using COW preallocation for non-COW data mentioned in the current comments.

Note that this breaks the current version of xfs/420, but that is because the test is broken. A separate fix will be sent for it.

(Still waiting for this...)

It was sent quite a while ago, but I need to resend it..

Quoted 183 lines

Quoted 59 lines

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_iomap.c   | 132 ++++++++++++++++++++++++++++++-------------
 fs/xfs/xfs_reflink.c |  67 ----------------------
 fs/xfs/xfs_trace.h   |   1 -
 3 files changed, 93 insertions(+), 107 deletions(-)

fs/xfs/xfs_iomap.c +14 -8

index d851abac16a9..d19f99e5476a 100644
@@ -534,15 +534,16 @@ xfs_file_iomap_begin_delay(
 {
 	struct xfs_inode	*ip = XFS_I(inode);
 	struct xfs_mount	*mp = ip->i_mount;
-	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	xfs_fileoff_t		maxbytes_fsb =
 		XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 	xfs_fileoff_t		end_fsb;
-	int			error = 0, eof = 0;
-	struct xfs_bmbt_irec	got;
-	struct xfs_iext_cursor	icur;
+	struct xfs_bmbt_irec	imap, cmap;
+	struct xfs_iext_cursor	icur, ccur;
 	xfs_fsblock_t		prealloc_blocks = 0;
+	bool			eof = false, cow_eof = false, shared;
+	int			whichfork = XFS_DATA_FORK;
+	int			error = 0;
 
 	ASSERT(!XFS_IS_REALTIME_INODE(ip));
 	ASSERT(!xfs_get_extsz_hint(ip));

@@ -560,7 +561,7 @@ xfs_file_iomap_begin_delay(
 
 	XFS_STATS_INC(mp, xs_blk_mapw);
 
-	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
+	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
 		if (error)
 			goto out_unlock;

@@ -568,51 +569,92 @@ xfs_file_iomap_begin_delay(
 
 	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 
-	eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
+	/*
+	 * Search the data fork fork first to look up our source mapping.  We
+	 * always need the data fork map, as we have to return it to the
+	 * iomap code so that the higher level write code can read data in to
+	 * perform read-modify-write cycles for unaligned writes.
+	 */
+	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
 	if (eof)
-		got.br_startoff = end_fsb; /* fake hole until the end */
+		imap.br_startoff = end_fsb; /* fake hole until the end */
 
-	if (got.br_startoff <= offset_fsb) {

fs/xfs/xfs_iomap.c xfs_file_iomap_begin_delay( · line 591

Quoted 10 lines

fs/xfs/xfs_iomap.c +10 -0

+	/* We never need to allocate blocks for zeroing a hole. */
+	if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
+		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
+		goto out_unlock;
+	}
+
+	/*
+	 * Search the COW fork extent list even if we did not find a data fork
+	 * extent.  This serves two purposes: first this implements the
+	 * speculative preallocation using cowextisze, so that we also unshare

"cowextsize"

Quoted 111 lines

fs/xfs/xfs_iomap.c +57 -29

+	 * block adjacent to shared blocks instead of just the shared blocks
+	 * themselves.  Second the lookup in the extent list is generally faster
+	 * than going out to the shared extent tree.
+	 */
+	if (xfs_is_reflink_inode(ip)) {
+		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
+				&ccur, &cmap);
+		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
+			trace_xfs_reflink_cow_found(ip, &cmap);
+			whichfork = XFS_COW_FORK;
+			goto done;
+		}
+	}
+
+	if (imap.br_startoff <= offset_fsb) {
 		/*
 		 * For reflink files we may need a delalloc reservation when
 		 * overwriting shared extents.   This includes zeroing of
 		 * existing extents that contain data.
 		 */
-		if (xfs_is_reflink_inode(ip) &&
-		    ((flags & IOMAP_WRITE) ||
-		     got.br_state != XFS_EXT_UNWRITTEN)) {
-			xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
-			error = xfs_reflink_reserve_cow(ip, &got);
-			if (error)
-				goto out_unlock;
+		if (!xfs_is_reflink_inode(ip) ||
+		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
+			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
+					&imap);
+			goto done;
 		}
 
-		trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK, &got);
-		goto done;
-	}
+		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
 
-	if (flags & IOMAP_ZERO) {
-		xfs_hole_to_iomap(ip, iomap, offset_fsb, got.br_startoff);
-		goto out_unlock;
+		/* Trim the mapping to the nearest shared extent boundary. */
+		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
+		if (error)
+			goto out_unlock;
+
+		/* Not shared?  Just report the (potentially capped) extent. */
+		if (!shared) {
+			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
+					&imap);
+			goto done;
+		}
+
+		/*
+		 * Fork all the shared blocks from our write offset until the
+		 * end of the extent.
+		 */
+		whichfork = XFS_COW_FORK;
+		end_fsb = imap.br_startoff + imap.br_blockcount;
+	} else {
+		/*
+		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
+		 * pages to keep the chunks of work done where somewhat
+		 * symmetric with the work writeback does.  This is a completely
+		 * arbitrary number pulled out of thin air.
+		 *
+		 * Note that the values needs to be less than 32-bits wide until
+		 * the lower level functions are updated.
+		 */
+		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
+		end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 	}
 
 	error = xfs_qm_dqattach_locked(ip, false);
 	if (error)
 		goto out_unlock;
 
-	/*
-	 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
-	 * to keep the chunks of work done where somewhat symmetric with the
-	 * work writeback does. This is a completely arbitrary number pulled
-	 * out of thin air as a best guess for initial testing.
-	 *
-	 * Note that the values needs to be less than 32-bits wide until
-	 * the lower level functions are updated.
-	 */
-	count = min_t(loff_t, count, 1024 * PAGE_SIZE);
-	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
-
-	if (eof) {
+	if (eof && whichfork == XFS_DATA_FORK) {
 		prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
 				&icur);
 		if (prealloc_blocks) {

@@ -635,9 +677,11 @@ xfs_file_iomap_begin_delay(
 	}
 
 retry:
-	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
-			end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
-			eof);
+	error = xfs_bmapi_reserve_delalloc(ip, whichfork, offset_fsb,
+			end_fsb - offset_fsb, prealloc_blocks,
+			whichfork == XFS_DATA_FORK ? &imap : &cmap,
+			whichfork == XFS_DATA_FORK ? &icur : &ccur,
+			whichfork == XFS_DATA_FORK ? eof : cow_eof);
 	switch (error) {
 	case 0:
 		break;

@@ -659,9 +703,19 @@ xfs_file_iomap_begin_delay(

Quoted 8 lines

fs/xfs/xfs_iomap.c xfs_file_iomap_begin_delay( · line 706

Quoted 5 lines

fs/xfs/xfs_iomap.c +1 -1

 	 * them out if the write happens to fail.
 	 */
 	iomap->flags |= IOMAP_F_NEW;
-	trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &got);
+	trace_xfs_iomap_alloc(ip, offset, count, whichfork, &imap);

I'm confused by this, if whichfork == COW then won't ftrace report results from the wrong fork?

The question is what the "right" fork to trace is as both matter here. At least we now clearly tell you which fork the trace belongs too.

No user of it in the iomap code at the moment, but we should not actively report wrong information if we can trivially get it right.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_iomap.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

fs/xfs/xfs_iomap.c +1 -1

index d19f99e5476a..bbc5d2e06b06 100644
@@ -541,7 +541,7 @@ xfs_file_iomap_begin_delay(
 	struct xfs_bmbt_irec	imap, cmap;
 	struct xfs_iext_cursor	icur, ccur;
 	xfs_fsblock_t		prealloc_blocks = 0;
-	bool			eof = false, cow_eof = false, shared;
+	bool			eof = false, cow_eof = false, shared = false;
 	int			whichfork = XFS_DATA_FORK;
 	int			error = 0;
 

fs/xfs/xfs_iomap.c +2 -1

 			goto out_unlock;
 		}
 		/* ensure we only report blocks we have a reservation for */
 		xfs_trim_extent(&imap, cmap.br_startoff, cmap.br_blockcount);
+		shared = true;
 	}
-	error = xfs_bmbt_to_iomap(ip, iomap, &imap, false);
+	error = xfs_bmbt_to_iomap(ip, iomap, &imap, shared);
 out_unlock:
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;

Signature · 1 line

2.19.1

Quoted 9 lines

fs/xfs/xfs_iomap.c +1 -1

@@ -541,7 +541,7 @@ xfs_file_iomap_begin_delay(
 	struct xfs_bmbt_irec	imap, cmap;
 	struct xfs_iext_cursor	icur, ccur;
 	xfs_fsblock_t		prealloc_blocks = 0;
-	bool			eof = false, cow_eof = false, shared;
+	bool			eof = false, cow_eof = false, shared = false;
 	int			whichfork = XFS_DATA_FORK;
 	int			error = 0;
 

fs/xfs/xfs_iomap.c xfs_file_iomap_begin_delay( · line 712

Quoted 6 lines

fs/xfs/xfs_iomap.c +1 -1

@@ -709,13 +709,14 @@ xfs_file_iomap_begin_delay(
 		if (imap.br_startoff > offset_fsb) {
 			xfs_trim_extent(&cmap, offset_fsb,
 					imap.br_startoff - offset_fsb);
-			error = xfs_bmbt_to_iomap(ip, iomap, &cmap, false);
+			error = xfs_bmbt_to_iomap(ip, iomap, &cmap, true);

Does this belong in the previous patch? (Maybe all of it?)

If we have racing buffered and direct I/O COW fork extents under writeback can have been moved to the data fork by the time we call xfs_reflink_convert_cow from xfs_submit_ioend. This would be mostly harmless as the block numbers don't change by this move, except for the fact that xfs_bmapi_write will crash or trigger asserts when not finding existing extents, even despite trying to paper over this with the XFS_BMAPI_CONVERT_ONLY flag.

Instead of special casing non-transaction conversions in the already way too complicated xfs_bmapi_write just add a new helper for the much simpler non-transactional COW fork case, which simplify ignores not found extents.

Signed-off-by Christoph Hellwig <[email protected]>

---
 fs/xfs/libxfs/xfs_bmap.c | 12 ++------
 fs/xfs/libxfs/xfs_bmap.h |  8 +++---
 fs/xfs/xfs_reflink.c     | 61 +++++++++++++++++++++++++---------------
 3 files changed, 45 insertions(+), 36 deletions(-)

fs/xfs/libxfs/xfs_bmap.c +3 -9

index 1992ed8a60b0..fbed7ed34a7f 100644
@@ -2029,7 +2029,7 @@ xfs_bmap_add_extent_delay_real(
 /*
  * Convert an unwritten allocation to a real allocation or vice versa.
  */
-STATIC int				/* error */
+int					/* error */
 xfs_bmap_add_extent_unwritten_real(
 	struct xfs_trans	*tp,
 	xfs_inode_t		*ip,	/* incore inode pointer */

@@ -4236,9 +4236,7 @@ xfs_bmapi_write(
 
 	ASSERT(*nmap >= 1);
 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
-	ASSERT(tp != NULL ||
-	       (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
-			(XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
+	ASSERT(tp != NULL);
 	ASSERT(len > 0);
 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));

@@ -4316,9 +4314,6 @@ xfs_bmapi_write(
 			 * locked and hence a truncate will block on them
 			 * first.
 			 */
-			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
-			         (flags & XFS_BMAPI_COWFORK)));
-
 			if (flags & XFS_BMAPI_DELALLOC) {
 				if (eof || bno >= end)
 					break;

@@ -4333,8 +4328,7 @@ xfs_bmapi_write(
 		 * First, deal with the hole before the allocated space
 		 * that we found, if any.
 		 */
-		if ((need_alloc || wasdelay) &&
-		    !(flags & XFS_BMAPI_CONVERT_ONLY)) {
+		if (need_alloc || wasdelay) {
 			bma.eof = eof;
 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
 			bma.wasdel = wasdelay;

fs/xfs/libxfs/xfs_bmap.h +4 -4

index f9a925caa70e..ee3848680684 100644
@@ -98,9 +98,6 @@ struct xfs_extent_free_item
 /* Only convert delalloc space, don't allocate entirely new extents */
 #define XFS_BMAPI_DELALLOC	0x400
 
-/* Only convert unwritten extents, don't allocate new blocks */
-#define XFS_BMAPI_CONVERT_ONLY	0x800
-
 /* Skip online discard of freed extents */
 #define XFS_BMAPI_NODISCARD	0x1000
 

@@ -118,7 +115,6 @@ struct xfs_extent_free_item
 	{ XFS_BMAPI_REMAP,	"REMAP" }, \
 	{ XFS_BMAPI_COWFORK,	"COWFORK" }, \
 	{ XFS_BMAPI_DELALLOC,	"DELALLOC" }, \
-	{ XFS_BMAPI_CONVERT_ONLY, "CONVERT_ONLY" }, \
 	{ XFS_BMAPI_NODISCARD,	"NODISCARD" }, \
 	{ XFS_BMAPI_NORMAP,	"NORMAP" }
 

@@ -227,6 +223,10 @@ int	xfs_bmapi_reserve_delalloc(struct xfs_inode *ip, int whichfork,
 		xfs_fileoff_t off, xfs_filblks_t len, xfs_filblks_t prealloc,
 		struct xfs_bmbt_irec *got, struct xfs_iext_cursor *cur,
 		int eof);
+int	xfs_bmap_add_extent_unwritten_real(struct xfs_trans *tp,
+		struct xfs_inode *ip, int whichfork,
+		struct xfs_iext_cursor *icur, struct xfs_btree_cur **curp,
+		struct xfs_bmbt_irec *new, int *logflagsp);
 
 static inline void
 xfs_bmap_add_free(

fs/xfs/xfs_reflink.c +32 -17

index d59b556d42cb..0cf13cb1b2fe 100644
@@ -234,26 +234,42 @@ xfs_reflink_trim_around_shared(
 	}
 }
 
-/* Convert part of an unwritten CoW extent to a real one. */
-STATIC int
-xfs_reflink_convert_cow_extent(
-	struct xfs_inode		*ip,
-	struct xfs_bmbt_irec		*imap,
-	xfs_fileoff_t			offset_fsb,
-	xfs_filblks_t			count_fsb)
+static int
+xfs_reflink_convert_cow_locked(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		offset_fsb,
+	xfs_filblks_t		count_fsb)
 {
-	int				nimaps = 1;
+	struct xfs_iext_cursor	icur;
+	struct xfs_bmbt_irec	got;
+	struct xfs_btree_cur	*dummy_cur = NULL;
+	int			dummy_logflags;
+	int			error;
 
-	if (imap->br_state == XFS_EXT_NORM)
+	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got))
 		return 0;
 
-	xfs_trim_extent(imap, offset_fsb, count_fsb);
-	trace_xfs_reflink_convert_cow(ip, imap);
-	if (imap->br_blockcount == 0)
-		return 0;
-	return xfs_bmapi_write(NULL, ip, imap->br_startoff, imap->br_blockcount,
-			XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, 0, imap,
-			&nimaps);
+	do {
+		if (got.br_startoff >= offset_fsb + count_fsb)
+			break;
+		if (got.br_state == XFS_EXT_NORM)
+			continue;
+		if (WARN_ON_ONCE(isnullstartblock(got.br_startblock)))
+			return -EIO;
+
+		xfs_trim_extent(&got, offset_fsb, count_fsb);
+		if (!got.br_blockcount)
+			continue;
+
+		got.br_state = XFS_EXT_NORM;
+		error = xfs_bmap_add_extent_unwritten_real(NULL, ip,
+				XFS_COW_FORK, &icur, &dummy_cur, &got,
+				&dummy_logflags);
+		if (error)
+			return error;
+	} while (xfs_iext_next_extent(ip->i_cowfp, &icur, &got));
+
+	return error;
 }
 
 /* Convert all of the unwritten CoW extents in a file's range to real ones. */

@@ -267,15 +283,12 @@ xfs_reflink_convert_cow(
 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
 	xfs_filblks_t		count_fsb = end_fsb - offset_fsb;
-	struct xfs_bmbt_irec	imap;

fs/xfs/xfs_reflink.c +4 -2

 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 	return error;
 }

@@ -405,9 +418,11 @@ xfs_reflink_allocate_cow(
 	if (nimaps == 0)
 		return -ENOSPC;
 convert:
-	if (!(flags & IOMAP_DIRECT))
+	xfs_trim_extent(imap, offset_fsb, count_fsb);
+	if (!(flags & IOMAP_DIRECT) || imap->br_state == XFS_EXT_NORM)
 		return 0;
-	return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb);
+	trace_xfs_reflink_convert_cow(ip, imap);
+	return xfs_reflink_convert_cow_locked(ip, offset_fsb, count_fsb);
 
 out_unreserve:
 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)resblks, 0,

Signature · 1 line

2.19.1

Quoted 8 lines

fs/xfs/xfs_reflink.c +1 -2

 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
 	xfs_filblks_t		count_fsb = end_fsb - offset_fsb;
-	struct xfs_bmbt_irec	imap;
-	int			nimaps = 1, error = 0;
+	int			error;
 
 	ASSERT(count != 0);

fs/xfs/xfs_reflink.c xfs_reflink_convert_cow( · line 291

Quoted 5 lines

fs/xfs/xfs_reflink.c +1 -3

 	xfs_ilock(ip, XFS_ILOCK_EXCL);
-	error = xfs_bmapi_write(NULL, ip, offset_fsb, count_fsb,
-			XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT |
-			XFS_BMAPI_CONVERT_ONLY, 0, &imap, &nimaps);
+	error = xfs_reflink_convert_cow_locked(ip, offset_fsb, count_fsb);

At this point you might as well convert the one remaining caller of xfs_reflink_convert_cow to take and drop the ILOCK around the reflink_convert_cow call...