2 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version 2
9 * of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it would be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
32 #include "xfs_inode_item.h"
34 #include "xfs_bmap_util.h"
35 #include "xfs_error.h"
37 #include "xfs_dir2_priv.h"
38 #include "xfs_ioctl.h"
39 #include "xfs_trace.h"
41 #include "xfs_icache.h"
43 #include "xfs_btree.h"
44 #include "xfs_refcount_btree.h"
45 #include "xfs_refcount.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_trans_space.h"
49 #include "xfs_alloc.h"
50 #include "xfs_quota_defs.h"
51 #include "xfs_quota.h"
52 #include "xfs_btree.h"
53 #include "xfs_bmap_btree.h"
54 #include "xfs_reflink.h"
55 #include "xfs_iomap.h"
56 #include "xfs_rmap_btree.h"
58 #include "xfs_ag_resv.h"
61 * Copy on Write of Shared Blocks
63 * XFS must preserve "the usual" file semantics even when two files share
64 * the same physical blocks. This means that a write to one file must not
65 * alter the blocks in a different file; the way that we'll do that is
66 * through the use of a copy-on-write mechanism. At a high level, that
67 * means that when we want to write to a shared block, we allocate a new
68 * block, write the data to the new block, and if that succeeds we map the
69 * new block into the file.
71 * XFS provides a "delayed allocation" mechanism that defers the allocation
72 * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
73 * possible. This reduces fragmentation by enabling the filesystem to ask
74 * for bigger chunks less often, which is exactly what we want for CoW.
76 * The delalloc mechanism begins when the kernel wants to make a block
77 * writable (write_begin or page_mkwrite). If the offset is not mapped, we
78 * create a delalloc mapping, which is a regular in-core extent, but without
79 * a real startblock. (For delalloc mappings, the startblock encodes both
80 * a flag that this is a delalloc mapping, and a worst-case estimate of how
81 * many blocks might be required to put the mapping into the BMBT.) delalloc
82 * mappings are a reservation against the free space in the filesystem;
83 * adjacent mappings can also be combined into fewer larger mappings.
85 * When dirty pages are being written out (typically in writepage), the
86 * delalloc reservations are converted into real mappings by allocating
87 * blocks and replacing the delalloc mapping with real ones. A delalloc
88 * mapping can be replaced by several real ones if the free space is
91 * We want to adapt the delalloc mechanism for copy-on-write, since the
92 * write paths are similar. The first two steps (creating the reservation
93 * and allocating the blocks) are exactly the same as delalloc except that
94 * the mappings must be stored in a separate CoW fork because we do not want
95 * to disturb the mapping in the data fork until we're sure that the write
96 * succeeded. IO completion in this case is the process of removing the old
97 * mapping from the data fork and moving the new mapping from the CoW fork to
98 * the data fork. This will be discussed shortly.
100 * For now, unaligned directio writes will be bounced back to the page cache.
101 * Block-aligned directio writes will use the same mechanism as buffered
104 * CoW remapping must be done after the data block write completes,
105 * because we don't want to destroy the old data fork map until we're sure
106 * the new block has been written. Since the new mappings are kept in a
107 * separate fork, we can simply iterate these mappings to find the ones
108 * that cover the file blocks that we just CoW'd. For each extent, simply
109 * unmap the corresponding range in the data fork, map the new range into
110 * the data fork, and remove the extent from the CoW fork.
112 * Since the remapping operation can be applied to an arbitrary file
113 * range, we record the need for the remap step as a flag in the ioend
114 * instead of declaring a new IO type. This is required for direct io
115 * because we only have ioend for the whole dio, and we have to be able to
116 * remember the presence of unwritten blocks and CoW blocks with a single
117 * ioend structure. Better yet, the more ground we can cover with one
122 * Given an AG extent, find the lowest-numbered run of shared blocks
123 * within that range and return the range in fbno/flen. If
124 * find_end_of_shared is true, return the longest contiguous extent of
125 * shared blocks. If there are no shared extents, fbno and flen will
126 * be set to NULLAGBLOCK and 0, respectively.
129 xfs_reflink_find_shared(
130 struct xfs_mount *mp,
136 bool find_end_of_shared)
138 struct xfs_buf *agbp;
139 struct xfs_btree_cur *cur;
142 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
146 cur = xfs_refcountbt_init_cursor(mp, NULL, agbp, agno, NULL);
148 error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen,
151 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
158 * Trim the mapping to the next block where there's a change in the
159 * shared/unshared status. More specifically, this means that we
160 * find the lowest-numbered extent of shared blocks that coincides with
161 * the given block mapping. If the shared extent overlaps the start of
162 * the mapping, trim the mapping to the end of the shared extent. If
163 * the shared region intersects the mapping, trim the mapping to the
164 * start of the shared extent. If there are no shared regions that
165 * overlap, just return the original extent.
168 xfs_reflink_trim_around_shared(
169 struct xfs_inode *ip,
170 struct xfs_bmbt_irec *irec,
181 /* Holes, unwritten, and delalloc extents cannot be shared */
182 if (!xfs_is_reflink_inode(ip) ||
184 irec->br_startblock == HOLESTARTBLOCK ||
185 irec->br_startblock == DELAYSTARTBLOCK ||
186 isnullstartblock(irec->br_startblock)) {
191 trace_xfs_reflink_trim_around_shared(ip, irec);
193 agno = XFS_FSB_TO_AGNO(ip->i_mount, irec->br_startblock);
194 agbno = XFS_FSB_TO_AGBNO(ip->i_mount, irec->br_startblock);
195 aglen = irec->br_blockcount;
197 error = xfs_reflink_find_shared(ip->i_mount, agno, agbno,
198 aglen, &fbno, &flen, true);
202 *shared = *trimmed = false;
203 if (fbno == NULLAGBLOCK) {
204 /* No shared blocks at all. */
206 } else if (fbno == agbno) {
208 * The start of this extent is shared. Truncate the
209 * mapping at the end of the shared region so that a
210 * subsequent iteration starts at the start of the
213 irec->br_blockcount = flen;
220 * There's a shared extent midway through this extent.
221 * Truncate the mapping at the start of the shared
222 * extent so that a subsequent iteration starts at the
223 * start of the shared region.
225 irec->br_blockcount = fbno - agbno;
231 /* Create a CoW reservation for a range of blocks within a file. */
233 __xfs_reflink_reserve_cow(
234 struct xfs_inode *ip,
235 xfs_fileoff_t *offset_fsb,
236 xfs_fileoff_t end_fsb,
239 struct xfs_bmbt_irec got, prev, imap;
240 xfs_fileoff_t orig_end_fsb;
241 int nimaps, eof = 0, error = 0;
242 bool shared = false, trimmed = false;
246 /* Already reserved? Skip the refcount btree access. */
247 xfs_bmap_search_extents(ip, *offset_fsb, XFS_COW_FORK, &eof, &idx,
249 if (!eof && got.br_startoff <= *offset_fsb) {
250 end_fsb = orig_end_fsb = got.br_startoff + got.br_blockcount;
251 trace_xfs_reflink_cow_found(ip, &got);
255 /* Read extent from the source file. */
257 error = xfs_bmapi_read(ip, *offset_fsb, end_fsb - *offset_fsb,
263 /* Trim the mapping to the nearest shared extent boundary. */
264 error = xfs_reflink_trim_around_shared(ip, &imap, &shared, &trimmed);
268 end_fsb = orig_end_fsb = imap.br_startoff + imap.br_blockcount;
270 /* Not shared? Just report the (potentially capped) extent. */
277 * Fork all the shared blocks from our write offset until the end of
280 error = xfs_qm_dqattach_locked(ip, 0);
284 align = xfs_eof_alignment(ip, xfs_get_cowextsz_hint(ip));
286 end_fsb = roundup_64(end_fsb, align);
289 error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, *offset_fsb,
290 end_fsb - *offset_fsb, &got,
297 /* retry without any preallocation */
298 trace_xfs_reflink_cow_enospc(ip, &imap);
299 if (end_fsb != orig_end_fsb) {
300 end_fsb = orig_end_fsb;
308 if (end_fsb != orig_end_fsb)
309 xfs_inode_set_cowblocks_tag(ip);
311 trace_xfs_reflink_cow_alloc(ip, &got);
313 *offset_fsb = end_fsb;
318 /* Create a CoW reservation for part of a file. */
320 xfs_reflink_reserve_cow_range(
321 struct xfs_inode *ip,
325 struct xfs_mount *mp = ip->i_mount;
326 xfs_fileoff_t offset_fsb, end_fsb;
327 bool skipped = false;
330 trace_xfs_reflink_reserve_cow_range(ip, offset, count);
332 offset_fsb = XFS_B_TO_FSBT(mp, offset);
333 end_fsb = XFS_B_TO_FSB(mp, offset + count);
335 xfs_ilock(ip, XFS_ILOCK_EXCL);
336 while (offset_fsb < end_fsb) {
337 error = __xfs_reflink_reserve_cow(ip, &offset_fsb, end_fsb,
340 trace_xfs_reflink_reserve_cow_range_error(ip, error,
345 xfs_iunlock(ip, XFS_ILOCK_EXCL);
350 /* Allocate all CoW reservations covering a range of blocks in a file. */
352 __xfs_reflink_allocate_cow(
353 struct xfs_inode *ip,
354 xfs_fileoff_t *offset_fsb,
355 xfs_fileoff_t end_fsb)
357 struct xfs_mount *mp = ip->i_mount;
358 struct xfs_bmbt_irec imap;
359 struct xfs_defer_ops dfops;
360 struct xfs_trans *tp;
361 xfs_fsblock_t first_block;
362 xfs_fileoff_t next_fsb;
363 int nimaps = 1, error;
364 bool skipped = false;
366 xfs_defer_init(&dfops, &first_block);
368 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
369 XFS_TRANS_RESERVE, &tp);
373 xfs_ilock(ip, XFS_ILOCK_EXCL);
375 next_fsb = *offset_fsb;
376 error = __xfs_reflink_reserve_cow(ip, &next_fsb, end_fsb, &skipped);
378 goto out_trans_cancel;
381 *offset_fsb = next_fsb;
382 goto out_trans_cancel;
385 xfs_trans_ijoin(tp, ip, 0);
386 error = xfs_bmapi_write(tp, ip, *offset_fsb, next_fsb - *offset_fsb,
387 XFS_BMAPI_COWFORK, &first_block,
388 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK),
389 &imap, &nimaps, &dfops);
391 goto out_trans_cancel;
393 /* We might not have been able to map the whole delalloc extent */
394 *offset_fsb = min(*offset_fsb + imap.br_blockcount, next_fsb);
396 error = xfs_defer_finish(&tp, &dfops, NULL);
398 goto out_trans_cancel;
400 error = xfs_trans_commit(tp);
403 xfs_iunlock(ip, XFS_ILOCK_EXCL);
406 xfs_defer_cancel(&dfops);
407 xfs_trans_cancel(tp);
411 /* Allocate all CoW reservations covering a part of a file. */
413 xfs_reflink_allocate_cow_range(
414 struct xfs_inode *ip,
418 struct xfs_mount *mp = ip->i_mount;
419 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
420 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
423 ASSERT(xfs_is_reflink_inode(ip));
425 trace_xfs_reflink_allocate_cow_range(ip, offset, count);
428 * Make sure that the dquots are there.
430 error = xfs_qm_dqattach(ip, 0);
434 while (offset_fsb < end_fsb) {
435 error = __xfs_reflink_allocate_cow(ip, &offset_fsb, end_fsb);
437 trace_xfs_reflink_allocate_cow_range_error(ip, error,
447 * Find the CoW reservation (and whether or not it needs block allocation)
448 * for a given byte offset of a file.
451 xfs_reflink_find_cow_mapping(
452 struct xfs_inode *ip,
454 struct xfs_bmbt_irec *imap,
457 struct xfs_bmbt_irec irec;
458 struct xfs_ifork *ifp;
459 struct xfs_bmbt_rec_host *gotp;
463 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
464 ASSERT(xfs_is_reflink_inode(ip));
466 /* Find the extent in the CoW fork. */
467 ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
468 bno = XFS_B_TO_FSBT(ip->i_mount, offset);
469 gotp = xfs_iext_bno_to_ext(ifp, bno, &idx);
473 xfs_bmbt_get_all(gotp, &irec);
474 if (bno >= irec.br_startoff + irec.br_blockcount ||
475 bno < irec.br_startoff)
478 trace_xfs_reflink_find_cow_mapping(ip, offset, 1, XFS_IO_OVERWRITE,
481 /* If it's still delalloc, we must allocate later. */
483 *need_alloc = !!(isnullstartblock(irec.br_startblock));
489 * Trim an extent to end at the next CoW reservation past offset_fsb.
492 xfs_reflink_trim_irec_to_next_cow(
493 struct xfs_inode *ip,
494 xfs_fileoff_t offset_fsb,
495 struct xfs_bmbt_irec *imap)
497 struct xfs_bmbt_irec irec;
498 struct xfs_ifork *ifp;
499 struct xfs_bmbt_rec_host *gotp;
502 if (!xfs_is_reflink_inode(ip))
505 /* Find the extent in the CoW fork. */
506 ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
507 gotp = xfs_iext_bno_to_ext(ifp, offset_fsb, &idx);
510 xfs_bmbt_get_all(gotp, &irec);
512 /* This is the extent before; try sliding up one. */
513 if (irec.br_startoff < offset_fsb) {
515 if (idx >= ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
517 gotp = xfs_iext_get_ext(ifp, idx);
518 xfs_bmbt_get_all(gotp, &irec);
521 if (irec.br_startoff >= imap->br_startoff + imap->br_blockcount)
524 imap->br_blockcount = irec.br_startoff - imap->br_startoff;
525 trace_xfs_reflink_trim_irec(ip, imap);
531 * Cancel all pending CoW reservations for some block range of an inode.
534 xfs_reflink_cancel_cow_blocks(
535 struct xfs_inode *ip,
536 struct xfs_trans **tpp,
537 xfs_fileoff_t offset_fsb,
538 xfs_fileoff_t end_fsb)
540 struct xfs_bmbt_irec irec;
541 xfs_filblks_t count_fsb;
542 xfs_fsblock_t firstfsb;
543 struct xfs_defer_ops dfops;
547 if (!xfs_is_reflink_inode(ip))
550 /* Go find the old extent in the CoW fork. */
551 while (offset_fsb < end_fsb) {
553 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
554 error = xfs_bmapi_read(ip, offset_fsb, count_fsb, &irec,
555 &nimaps, XFS_BMAPI_COWFORK);
560 trace_xfs_reflink_cancel_cow(ip, &irec);
562 if (irec.br_startblock == DELAYSTARTBLOCK) {
563 /* Free a delayed allocation. */
564 xfs_mod_fdblocks(ip->i_mount, irec.br_blockcount,
566 ip->i_delayed_blks -= irec.br_blockcount;
568 /* Remove the mapping from the CoW fork. */
569 error = xfs_bunmapi_cow(ip, &irec);
572 } else if (irec.br_startblock == HOLESTARTBLOCK) {
575 xfs_trans_ijoin(*tpp, ip, 0);
576 xfs_defer_init(&dfops, &firstfsb);
578 /* Free the CoW orphan record. */
579 error = xfs_refcount_free_cow_extent(ip->i_mount,
580 &dfops, irec.br_startblock,
585 xfs_bmap_add_free(ip->i_mount, &dfops,
586 irec.br_startblock, irec.br_blockcount,
589 /* Update quota accounting */
590 xfs_trans_mod_dquot_byino(*tpp, ip, XFS_TRANS_DQ_BCOUNT,
591 -(long)irec.br_blockcount);
593 /* Roll the transaction */
594 error = xfs_defer_finish(tpp, &dfops, ip);
596 xfs_defer_cancel(&dfops);
600 /* Remove the mapping from the CoW fork. */
601 error = xfs_bunmapi_cow(ip, &irec);
607 offset_fsb = irec.br_startoff + irec.br_blockcount;
614 * Cancel all pending CoW reservations for some byte range of an inode.
617 xfs_reflink_cancel_cow_range(
618 struct xfs_inode *ip,
622 struct xfs_trans *tp;
623 xfs_fileoff_t offset_fsb;
624 xfs_fileoff_t end_fsb;
627 trace_xfs_reflink_cancel_cow_range(ip, offset, count);
628 ASSERT(xfs_is_reflink_inode(ip));
630 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
631 if (count == NULLFILEOFF)
632 end_fsb = NULLFILEOFF;
634 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
636 /* Start a rolling transaction to remove the mappings */
637 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
642 xfs_ilock(ip, XFS_ILOCK_EXCL);
643 xfs_trans_ijoin(tp, ip, 0);
645 /* Scrape out the old CoW reservations */
646 error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb);
650 error = xfs_trans_commit(tp);
652 xfs_iunlock(ip, XFS_ILOCK_EXCL);
656 xfs_trans_cancel(tp);
657 xfs_iunlock(ip, XFS_ILOCK_EXCL);
659 trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_);
664 * Remap parts of a file's data fork after a successful CoW.
668 struct xfs_inode *ip,
672 struct xfs_bmbt_irec irec;
673 struct xfs_bmbt_irec uirec;
674 struct xfs_trans *tp;
675 xfs_fileoff_t offset_fsb;
676 xfs_fileoff_t end_fsb;
677 xfs_filblks_t count_fsb;
678 xfs_fsblock_t firstfsb;
679 struct xfs_defer_ops dfops;
681 unsigned int resblks;
686 trace_xfs_reflink_end_cow(ip, offset, count);
688 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
689 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
690 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
692 /* Start a rolling transaction to switch the mappings */
693 resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
694 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
699 xfs_ilock(ip, XFS_ILOCK_EXCL);
700 xfs_trans_ijoin(tp, ip, 0);
702 /* Go find the old extent in the CoW fork. */
703 while (offset_fsb < end_fsb) {
704 /* Read extent from the source file */
706 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
707 error = xfs_bmapi_read(ip, offset_fsb, count_fsb, &irec,
708 &nimaps, XFS_BMAPI_COWFORK);
713 ASSERT(irec.br_startblock != DELAYSTARTBLOCK);
714 trace_xfs_reflink_cow_remap(ip, &irec);
717 * We can have a hole in the CoW fork if part of a directio
718 * write is CoW but part of it isn't.
720 rlen = ilen = irec.br_blockcount;
721 if (irec.br_startblock == HOLESTARTBLOCK)
724 /* Unmap the old blocks in the data fork. */
726 xfs_defer_init(&dfops, &firstfsb);
727 error = __xfs_bunmapi(tp, ip, irec.br_startoff,
728 &rlen, 0, 1, &firstfsb, &dfops);
733 * Trim the extent to whatever got unmapped.
734 * Remember, bunmapi works backwards.
736 uirec.br_startblock = irec.br_startblock + rlen;
737 uirec.br_startoff = irec.br_startoff + rlen;
738 uirec.br_blockcount = irec.br_blockcount - rlen;
739 irec.br_blockcount = rlen;
740 trace_xfs_reflink_cow_remap_piece(ip, &uirec);
742 /* Free the CoW orphan record. */
743 error = xfs_refcount_free_cow_extent(tp->t_mountp,
744 &dfops, uirec.br_startblock,
745 uirec.br_blockcount);
749 /* Map the new blocks into the data fork. */
750 error = xfs_bmap_map_extent(tp->t_mountp, &dfops,
755 /* Remove the mapping from the CoW fork. */
756 error = xfs_bunmapi_cow(ip, &uirec);
760 error = xfs_defer_finish(&tp, &dfops, ip);
767 offset_fsb = irec.br_startoff + ilen;
770 error = xfs_trans_commit(tp);
771 xfs_iunlock(ip, XFS_ILOCK_EXCL);
777 xfs_defer_cancel(&dfops);
779 xfs_trans_cancel(tp);
780 xfs_iunlock(ip, XFS_ILOCK_EXCL);
782 trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
787 * Free leftover CoW reservations that didn't get cleaned out.
790 xfs_reflink_recover_cow(
791 struct xfs_mount *mp)
796 if (!xfs_sb_version_hasreflink(&mp->m_sb))
799 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
800 error = xfs_refcount_recover_cow_leftovers(mp, agno);
809 * Reflinking (Block) Ranges of Two Files Together
811 * First, ensure that the reflink flag is set on both inodes. The flag is an
812 * optimization to avoid unnecessary refcount btree lookups in the write path.
814 * Now we can iteratively remap the range of extents (and holes) in src to the
815 * corresponding ranges in dest. Let drange and srange denote the ranges of
816 * logical blocks in dest and src touched by the reflink operation.
818 * While the length of drange is greater than zero,
819 * - Read src's bmbt at the start of srange ("imap")
820 * - If imap doesn't exist, make imap appear to start at the end of srange
822 * - If imap starts before srange, advance imap to start at srange.
823 * - If imap goes beyond srange, truncate imap to end at the end of srange.
824 * - Punch (imap start - srange start + imap len) blocks from dest at
825 * offset (drange start).
826 * - If imap points to a real range of pblks,
827 * > Increase the refcount of the imap's pblks
828 * > Map imap's pblks into dest at the offset
829 * (drange start + imap start - srange start)
830 * - Advance drange and srange by (imap start - srange start + imap len)
832 * Finally, if the reflink made dest longer, update both the in-core and
833 * on-disk file sizes.
835 * ASCII Art Demonstration:
837 * Let's say we want to reflink this source file:
839 * ----SSSSSSS-SSSSS----SSSSSS (src file)
840 * <-------------------->
842 * into this destination file:
844 * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
845 * <-------------------->
846 * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
847 * Observe that the range has different logical offsets in either file.
849 * Consider that the first extent in the source file doesn't line up with our
850 * reflink range. Unmapping and remapping are separate operations, so we can
851 * unmap more blocks from the destination file than we remap.
853 * ----SSSSSSS-SSSSS----SSSSSS
855 * --DDDDD---------DDDDD--DDD
858 * Now remap the source extent into the destination file:
860 * ----SSSSSSS-SSSSS----SSSSSS
862 * --DDDDD--SSSSSSSDDDDD--DDD
865 * Do likewise with the second hole and extent in our range. Holes in the
866 * unmap range don't affect our operation.
868 * ----SSSSSSS-SSSSS----SSSSSS
870 * --DDDDD--SSSSSSS-SSSSS-DDD
873 * Finally, unmap and remap part of the third extent. This will increase the
874 * size of the destination file.
876 * ----SSSSSSS-SSSSS----SSSSSS
878 * --DDDDD--SSSSSSS-SSSSS----SSS
881 * Once we update the destination file's i_size, we're done.
885 * Ensure the reflink bit is set in both inodes.
888 xfs_reflink_set_inode_flag(
889 struct xfs_inode *src,
890 struct xfs_inode *dest)
892 struct xfs_mount *mp = src->i_mount;
894 struct xfs_trans *tp;
896 if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest))
899 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
903 /* Lock both files against IO */
904 if (src->i_ino == dest->i_ino)
905 xfs_ilock(src, XFS_ILOCK_EXCL);
907 xfs_lock_two_inodes(src, dest, XFS_ILOCK_EXCL);
909 if (!xfs_is_reflink_inode(src)) {
910 trace_xfs_reflink_set_inode_flag(src);
911 xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL);
912 src->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
913 xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
914 xfs_ifork_init_cow(src);
916 xfs_iunlock(src, XFS_ILOCK_EXCL);
918 if (src->i_ino == dest->i_ino)
921 if (!xfs_is_reflink_inode(dest)) {
922 trace_xfs_reflink_set_inode_flag(dest);
923 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
924 dest->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
925 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
926 xfs_ifork_init_cow(dest);
928 xfs_iunlock(dest, XFS_ILOCK_EXCL);
931 error = xfs_trans_commit(tp);
937 trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_);
942 * Update destination inode size & cowextsize hint, if necessary.
945 xfs_reflink_update_dest(
946 struct xfs_inode *dest,
948 xfs_extlen_t cowextsize)
950 struct xfs_mount *mp = dest->i_mount;
951 struct xfs_trans *tp;
954 if (newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0)
957 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
961 xfs_ilock(dest, XFS_ILOCK_EXCL);
962 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
964 if (newlen > i_size_read(VFS_I(dest))) {
965 trace_xfs_reflink_update_inode_size(dest, newlen);
966 i_size_write(VFS_I(dest), newlen);
967 dest->i_d.di_size = newlen;
971 dest->i_d.di_cowextsize = cowextsize;
972 dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
975 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
977 error = xfs_trans_commit(tp);
983 trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_);
988 * Do we have enough reserve in this AG to handle a reflink? The refcount
989 * btree already reserved all the space it needs, but the rmap btree can grow
990 * infinitely, so we won't allow more reflinks when the AG is down to the
994 xfs_reflink_ag_has_free_space(
995 struct xfs_mount *mp,
998 struct xfs_perag *pag;
1001 if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
1004 pag = xfs_perag_get(mp, agno);
1005 if (xfs_ag_resv_critical(pag, XFS_AG_RESV_AGFL) ||
1006 xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA))
1013 * Unmap a range of blocks from a file, then map other blocks into the hole.
1014 * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount).
1015 * The extent irec is mapped into dest at irec->br_startoff.
1018 xfs_reflink_remap_extent(
1019 struct xfs_inode *ip,
1020 struct xfs_bmbt_irec *irec,
1021 xfs_fileoff_t destoff,
1022 xfs_off_t new_isize)
1024 struct xfs_mount *mp = ip->i_mount;
1025 struct xfs_trans *tp;
1026 xfs_fsblock_t firstfsb;
1027 unsigned int resblks;
1028 struct xfs_defer_ops dfops;
1029 struct xfs_bmbt_irec uirec;
1032 xfs_filblks_t unmap_len;
1036 unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
1037 trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
1039 /* Only remap normal extents. */
1040 real_extent = (irec->br_startblock != HOLESTARTBLOCK &&
1041 irec->br_startblock != DELAYSTARTBLOCK &&
1042 !ISUNWRITTEN(irec));
1044 /* No reflinking if we're low on space */
1046 error = xfs_reflink_ag_has_free_space(mp,
1047 XFS_FSB_TO_AGNO(mp, irec->br_startblock));
1052 /* Start a rolling transaction to switch the mappings */
1053 resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
1054 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1058 xfs_ilock(ip, XFS_ILOCK_EXCL);
1059 xfs_trans_ijoin(tp, ip, 0);
1061 /* If we're not just clearing space, then do we have enough quota? */
1063 error = xfs_trans_reserve_quota_nblks(tp, ip,
1064 irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS);
1069 trace_xfs_reflink_remap(ip, irec->br_startoff,
1070 irec->br_blockcount, irec->br_startblock);
1072 /* Unmap the old blocks in the data fork. */
1075 xfs_defer_init(&dfops, &firstfsb);
1076 error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1,
1082 * Trim the extent to whatever got unmapped.
1083 * Remember, bunmapi works backwards.
1085 uirec.br_startblock = irec->br_startblock + rlen;
1086 uirec.br_startoff = irec->br_startoff + rlen;
1087 uirec.br_blockcount = unmap_len - rlen;
1090 /* If this isn't a real mapping, we're done. */
1091 if (!real_extent || uirec.br_blockcount == 0)
1094 trace_xfs_reflink_remap(ip, uirec.br_startoff,
1095 uirec.br_blockcount, uirec.br_startblock);
1097 /* Update the refcount tree */
1098 error = xfs_refcount_increase_extent(mp, &dfops, &uirec);
1102 /* Map the new blocks into the data fork. */
1103 error = xfs_bmap_map_extent(mp, &dfops, ip, &uirec);
1107 /* Update quota accounting. */
1108 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
1109 uirec.br_blockcount);
1111 /* Update dest isize if needed. */
1112 newlen = XFS_FSB_TO_B(mp,
1113 uirec.br_startoff + uirec.br_blockcount);
1114 newlen = min_t(xfs_off_t, newlen, new_isize);
1115 if (newlen > i_size_read(VFS_I(ip))) {
1116 trace_xfs_reflink_update_inode_size(ip, newlen);
1117 i_size_write(VFS_I(ip), newlen);
1118 ip->i_d.di_size = newlen;
1119 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1123 /* Process all the deferred stuff. */
1124 error = xfs_defer_finish(&tp, &dfops, ip);
1129 error = xfs_trans_commit(tp);
1130 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1136 xfs_defer_cancel(&dfops);
1138 xfs_trans_cancel(tp);
1139 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1141 trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
1146 * Iteratively remap one file's extents (and holes) to another's.
1149 xfs_reflink_remap_blocks(
1150 struct xfs_inode *src,
1151 xfs_fileoff_t srcoff,
1152 struct xfs_inode *dest,
1153 xfs_fileoff_t destoff,
1155 xfs_off_t new_isize)
1157 struct xfs_bmbt_irec imap;
1160 xfs_filblks_t range_len;
1162 /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
1164 trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
1166 /* Read extent from the source file */
1168 xfs_ilock(src, XFS_ILOCK_EXCL);
1169 error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0);
1170 xfs_iunlock(src, XFS_ILOCK_EXCL);
1173 ASSERT(nimaps == 1);
1175 trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
1178 /* Translate imap into the destination file. */
1179 range_len = imap.br_startoff + imap.br_blockcount - srcoff;
1180 imap.br_startoff += destoff - srcoff;
1182 /* Clear dest from destoff to the end of imap and map it in. */
1183 error = xfs_reflink_remap_extent(dest, &imap, destoff,
1188 if (fatal_signal_pending(current)) {
1193 /* Advance drange/srange */
1194 srcoff += range_len;
1195 destoff += range_len;
1202 trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_);
1207 * Read a page's worth of file data into the page cache. Return the page
1210 static struct page *
1212 struct inode *inode,
1215 struct address_space *mapping;
1219 n = offset >> PAGE_SHIFT;
1220 mapping = inode->i_mapping;
1221 page = read_mapping_page(mapping, n, NULL);
1224 if (!PageUptodate(page)) {
1226 return ERR_PTR(-EIO);
1233 * Compare extents of two files to see if they are the same.
1236 xfs_compare_extents(
1245 xfs_off_t dest_poff;
1248 struct page *src_page;
1249 struct page *dest_page;
1257 src_poff = srcoff & (PAGE_SIZE - 1);
1258 dest_poff = destoff & (PAGE_SIZE - 1);
1259 cmp_len = min(PAGE_SIZE - src_poff,
1260 PAGE_SIZE - dest_poff);
1261 cmp_len = min(cmp_len, len);
1262 ASSERT(cmp_len > 0);
1264 trace_xfs_reflink_compare_extents(XFS_I(src), srcoff, cmp_len,
1265 XFS_I(dest), destoff);
1267 src_page = xfs_get_page(src, srcoff);
1268 if (IS_ERR(src_page)) {
1269 error = PTR_ERR(src_page);
1272 dest_page = xfs_get_page(dest, destoff);
1273 if (IS_ERR(dest_page)) {
1274 error = PTR_ERR(dest_page);
1275 unlock_page(src_page);
1279 src_addr = kmap_atomic(src_page);
1280 dest_addr = kmap_atomic(dest_page);
1282 flush_dcache_page(src_page);
1283 flush_dcache_page(dest_page);
1285 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1288 kunmap_atomic(dest_addr);
1289 kunmap_atomic(src_addr);
1290 unlock_page(dest_page);
1291 unlock_page(src_page);
1292 put_page(dest_page);
1307 trace_xfs_reflink_compare_extents_error(XFS_I(dest), error, _RET_IP_);
1312 * Link a range of blocks from one file to another.
1315 xfs_reflink_remap_range(
1316 struct file *file_in,
1318 struct file *file_out,
1323 struct inode *inode_in = file_inode(file_in);
1324 struct xfs_inode *src = XFS_I(inode_in);
1325 struct inode *inode_out = file_inode(file_out);
1326 struct xfs_inode *dest = XFS_I(inode_out);
1327 struct xfs_mount *mp = src->i_mount;
1328 loff_t bs = inode_out->i_sb->s_blocksize;
1329 bool same_inode = (inode_in == inode_out);
1330 xfs_fileoff_t sfsbno, dfsbno;
1331 xfs_filblks_t fsblen;
1332 xfs_extlen_t cowextsize;
1337 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1340 if (XFS_FORCED_SHUTDOWN(mp))
1343 /* Lock both files against IO */
1345 xfs_ilock(src, XFS_IOLOCK_EXCL);
1346 xfs_ilock(src, XFS_MMAPLOCK_EXCL);
1348 xfs_lock_two_inodes(src, dest, XFS_IOLOCK_EXCL);
1349 xfs_lock_two_inodes(src, dest, XFS_MMAPLOCK_EXCL);
1352 /* Don't touch certain kinds of inodes */
1354 if (IS_IMMUTABLE(inode_out))
1358 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1362 /* Don't reflink dirs, pipes, sockets... */
1364 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1367 if (S_ISFIFO(inode_in->i_mode) || S_ISFIFO(inode_out->i_mode))
1369 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1372 /* Don't reflink realtime inodes */
1373 if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
1376 /* Don't share DAX file data for now. */
1377 if (IS_DAX(inode_in) || IS_DAX(inode_out))
1380 /* Are we going all the way to the end? */
1381 isize = i_size_read(inode_in);
1388 len = isize - pos_in;
1390 /* Ensure offsets don't wrap and the input is inside i_size */
1391 if (pos_in + len < pos_in || pos_out + len < pos_out ||
1392 pos_in + len > isize)
1395 /* Don't allow dedupe past EOF in the dest file */
1399 disize = i_size_read(inode_out);
1400 if (pos_out >= disize || pos_out + len > disize)
1404 /* If we're linking to EOF, continue to the block boundary. */
1405 if (pos_in + len == isize)
1406 blen = ALIGN(isize, bs) - pos_in;
1410 /* Only reflink if we're aligned to block boundaries */
1411 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1412 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1415 /* Don't allow overlapped reflink within the same file */
1417 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1421 /* Wait for the completion of any pending IOs on both files */
1422 inode_dio_wait(inode_in);
1424 inode_dio_wait(inode_out);
1426 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1427 pos_in, pos_in + len - 1);
1431 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1432 pos_out, pos_out + len - 1);
1436 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1439 * Check that the extents are the same.
1442 bool is_same = false;
1444 ret = xfs_compare_extents(inode_in, pos_in, inode_out, pos_out,
1454 ret = xfs_reflink_set_inode_flag(src, dest);
1459 * Invalidate the page cache so that we can clear any CoW mappings
1460 * in the destination file.
1462 truncate_inode_pages_range(&inode_out->i_data, pos_out,
1463 PAGE_ALIGN(pos_out + len) - 1);
1465 dfsbno = XFS_B_TO_FSBT(mp, pos_out);
1466 sfsbno = XFS_B_TO_FSBT(mp, pos_in);
1467 fsblen = XFS_B_TO_FSB(mp, len);
1468 ret = xfs_reflink_remap_blocks(src, sfsbno, dest, dfsbno, fsblen,
1474 * Carry the cowextsize hint from src to dest if we're sharing the
1475 * entire source file to the entire destination file, the source file
1476 * has a cowextsize hint, and the destination file does not.
1479 if (pos_in == 0 && len == i_size_read(inode_in) &&
1480 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1481 pos_out == 0 && len >= i_size_read(inode_out) &&
1482 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1483 cowextsize = src->i_d.di_cowextsize;
1485 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize);
1488 xfs_iunlock(src, XFS_MMAPLOCK_EXCL);
1489 xfs_iunlock(src, XFS_IOLOCK_EXCL);
1490 if (src->i_ino != dest->i_ino) {
1491 xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
1492 xfs_iunlock(dest, XFS_IOLOCK_EXCL);
1495 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1500 * The user wants to preemptively CoW all shared blocks in this file,
1501 * which enables us to turn off the reflink flag. Iterate all
1502 * extents which are not prealloc/delalloc to see which ranges are
1503 * mentioned in the refcount tree, then read those blocks into the
1504 * pagecache, dirty them, fsync them back out, and then we can update
1505 * the inode flag. What happens if we run out of memory? :)
1508 xfs_reflink_dirty_extents(
1509 struct xfs_inode *ip,
1514 struct xfs_mount *mp = ip->i_mount;
1515 xfs_agnumber_t agno;
1516 xfs_agblock_t agbno;
1522 struct xfs_bmbt_irec map[2];
1526 while (end - fbno > 0) {
1529 * Look for extents in the file. Skip holes, delalloc, or
1530 * unwritten extents; they can't be reflinked.
1532 error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0);
1537 if (map[0].br_startblock == HOLESTARTBLOCK ||
1538 map[0].br_startblock == DELAYSTARTBLOCK ||
1539 ISUNWRITTEN(&map[0]))
1543 while (map[1].br_blockcount) {
1544 agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock);
1545 agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock);
1546 aglen = map[1].br_blockcount;
1548 error = xfs_reflink_find_shared(mp, agno, agbno, aglen,
1549 &rbno, &rlen, true);
1552 if (rbno == NULLAGBLOCK)
1555 /* Dirty the pages */
1556 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1557 fpos = XFS_FSB_TO_B(mp, map[1].br_startoff +
1559 flen = XFS_FSB_TO_B(mp, rlen);
1560 if (fpos + flen > isize)
1561 flen = isize - fpos;
1562 error = iomap_file_dirty(VFS_I(ip), fpos, flen,
1564 xfs_ilock(ip, XFS_ILOCK_EXCL);
1568 map[1].br_blockcount -= (rbno - agbno + rlen);
1569 map[1].br_startoff += (rbno - agbno + rlen);
1570 map[1].br_startblock += (rbno - agbno + rlen);
1574 fbno = map[0].br_startoff + map[0].br_blockcount;
1580 /* Clear the inode reflink flag if there are no shared extents. */
1582 xfs_reflink_clear_inode_flag(
1583 struct xfs_inode *ip,
1584 struct xfs_trans **tpp)
1586 struct xfs_mount *mp = ip->i_mount;
1589 xfs_agnumber_t agno;
1590 xfs_agblock_t agbno;
1594 struct xfs_bmbt_irec map;
1598 ASSERT(xfs_is_reflink_inode(ip));
1601 end = XFS_B_TO_FSB(mp, i_size_read(VFS_I(ip)));
1602 while (end - fbno > 0) {
1605 * Look for extents in the file. Skip holes, delalloc, or
1606 * unwritten extents; they can't be reflinked.
1608 error = xfs_bmapi_read(ip, fbno, end - fbno, &map, &nmaps, 0);
1613 if (map.br_startblock == HOLESTARTBLOCK ||
1614 map.br_startblock == DELAYSTARTBLOCK ||
1618 agno = XFS_FSB_TO_AGNO(mp, map.br_startblock);
1619 agbno = XFS_FSB_TO_AGBNO(mp, map.br_startblock);
1620 aglen = map.br_blockcount;
1622 error = xfs_reflink_find_shared(mp, agno, agbno, aglen,
1623 &rbno, &rlen, false);
1626 /* Is there still a shared block here? */
1627 if (rbno != NULLAGBLOCK)
1630 fbno = map.br_startoff + map.br_blockcount;
1634 * We didn't find any shared blocks so turn off the reflink flag.
1635 * First, get rid of any leftover CoW mappings.
1637 error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF);
1641 /* Clear the inode flag. */
1642 trace_xfs_reflink_unset_inode_flag(ip);
1643 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1644 xfs_inode_clear_cowblocks_tag(ip);
1645 xfs_trans_ijoin(*tpp, ip, 0);
1646 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1652 * Clear the inode reflink flag if there are no shared extents and the size
1656 xfs_reflink_try_clear_inode_flag(
1657 struct xfs_inode *ip)
1659 struct xfs_mount *mp = ip->i_mount;
1660 struct xfs_trans *tp;
1663 /* Start a rolling transaction to remove the mappings */
1664 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1668 xfs_ilock(ip, XFS_ILOCK_EXCL);
1669 xfs_trans_ijoin(tp, ip, 0);
1671 error = xfs_reflink_clear_inode_flag(ip, &tp);
1675 error = xfs_trans_commit(tp);
1679 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1682 xfs_trans_cancel(tp);
1684 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1689 * Pre-COW all shared blocks within a given byte range of a file and turn off
1690 * the reflink flag if we unshare all of the file's blocks.
1693 xfs_reflink_unshare(
1694 struct xfs_inode *ip,
1698 struct xfs_mount *mp = ip->i_mount;
1704 if (!xfs_is_reflink_inode(ip))
1707 trace_xfs_reflink_unshare(ip, offset, len);
1709 inode_dio_wait(VFS_I(ip));
1711 /* Try to CoW the selected ranges */
1712 xfs_ilock(ip, XFS_ILOCK_EXCL);
1713 fbno = XFS_B_TO_FSBT(mp, offset);
1714 isize = i_size_read(VFS_I(ip));
1715 end = XFS_B_TO_FSB(mp, offset + len);
1716 error = xfs_reflink_dirty_extents(ip, fbno, end, isize);
1719 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1721 /* Wait for the IO to finish */
1722 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1726 /* Turn off the reflink flag if possible. */
1727 error = xfs_reflink_try_clear_inode_flag(ip);
1734 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1736 trace_xfs_reflink_unshare_error(ip, error, _RET_IP_);
1741 * Does this inode have any real CoW reservations?
1744 xfs_reflink_has_real_cow_blocks(
1745 struct xfs_inode *ip)
1747 struct xfs_bmbt_irec irec;
1748 struct xfs_ifork *ifp;
1749 struct xfs_bmbt_rec_host *gotp;
1752 if (!xfs_is_reflink_inode(ip))
1755 /* Go find the old extent in the CoW fork. */
1756 ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1757 gotp = xfs_iext_bno_to_ext(ifp, 0, &idx);
1759 xfs_bmbt_get_all(gotp, &irec);
1761 if (!isnullstartblock(irec.br_startblock))
1766 if (idx >= ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
1768 gotp = xfs_iext_get_ext(ifp, idx);