1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_defer.h"
14 #include "xfs_da_format.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_inode.h"
17 #include "xfs_trans.h"
18 #include "xfs_inode_item.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_error.h"
23 #include "xfs_dir2_priv.h"
24 #include "xfs_ioctl.h"
25 #include "xfs_trace.h"
27 #include "xfs_icache.h"
29 #include "xfs_btree.h"
30 #include "xfs_refcount_btree.h"
31 #include "xfs_refcount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_trans_space.h"
35 #include "xfs_alloc.h"
36 #include "xfs_quota_defs.h"
37 #include "xfs_quota.h"
38 #include "xfs_reflink.h"
39 #include "xfs_iomap.h"
40 #include "xfs_rmap_btree.h"
42 #include "xfs_ag_resv.h"
45 * Copy on Write of Shared Blocks
47 * XFS must preserve "the usual" file semantics even when two files share
48 * the same physical blocks. This means that a write to one file must not
49 * alter the blocks in a different file; the way that we'll do that is
50 * through the use of a copy-on-write mechanism. At a high level, that
51 * means that when we want to write to a shared block, we allocate a new
52 * block, write the data to the new block, and if that succeeds we map the
53 * new block into the file.
55 * XFS provides a "delayed allocation" mechanism that defers the allocation
56 * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
57 * possible. This reduces fragmentation by enabling the filesystem to ask
58 * for bigger chunks less often, which is exactly what we want for CoW.
60 * The delalloc mechanism begins when the kernel wants to make a block
61 * writable (write_begin or page_mkwrite). If the offset is not mapped, we
62 * create a delalloc mapping, which is a regular in-core extent, but without
63 * a real startblock. (For delalloc mappings, the startblock encodes both
64 * a flag that this is a delalloc mapping, and a worst-case estimate of how
65 * many blocks might be required to put the mapping into the BMBT.) delalloc
66 * mappings are a reservation against the free space in the filesystem;
67 * adjacent mappings can also be combined into fewer larger mappings.
69 * As an optimization, the CoW extent size hint (cowextsz) creates
70 * outsized aligned delalloc reservations in the hope of landing out of
71 * order nearby CoW writes in a single extent on disk, thereby reducing
72 * fragmentation and improving future performance.
74 * D: --RRRRRRSSSRRRRRRRR--- (data fork)
75 * C: ------DDDDDDD--------- (CoW fork)
77 * When dirty pages are being written out (typically in writepage), the
78 * delalloc reservations are converted into unwritten mappings by
79 * allocating blocks and replacing the delalloc mapping with real ones.
80 * A delalloc mapping can be replaced by several unwritten ones if the
81 * free space is fragmented.
83 * D: --RRRRRRSSSRRRRRRRR---
84 * C: ------UUUUUUU---------
86 * We want to adapt the delalloc mechanism for copy-on-write, since the
87 * write paths are similar. The first two steps (creating the reservation
88 * and allocating the blocks) are exactly the same as delalloc except that
89 * the mappings must be stored in a separate CoW fork because we do not want
90 * to disturb the mapping in the data fork until we're sure that the write
91 * succeeded. IO completion in this case is the process of removing the old
92 * mapping from the data fork and moving the new mapping from the CoW fork to
93 * the data fork. This will be discussed shortly.
95 * For now, unaligned directio writes will be bounced back to the page cache.
96 * Block-aligned directio writes will use the same mechanism as buffered
99 * Just prior to submitting the actual disk write requests, we convert
100 * the extents representing the range of the file actually being written
101 * (as opposed to extra pieces created for the cowextsize hint) to real
102 * extents. This will become important in the next step:
104 * D: --RRRRRRSSSRRRRRRRR---
105 * C: ------UUrrUUU---------
107 * CoW remapping must be done after the data block write completes,
108 * because we don't want to destroy the old data fork map until we're sure
109 * the new block has been written. Since the new mappings are kept in a
110 * separate fork, we can simply iterate these mappings to find the ones
111 * that cover the file blocks that we just CoW'd. For each extent, simply
112 * unmap the corresponding range in the data fork, map the new range into
113 * the data fork, and remove the extent from the CoW fork. Because of
114 * the presence of the cowextsize hint, however, we must be careful
115 * only to remap the blocks that we've actually written out -- we must
116 * never remap delalloc reservations nor CoW staging blocks that have
117 * yet to be written. This corresponds exactly to the real extents in
120 * D: --RRRRRRrrSRRRRRRRR---
121 * C: ------UU--UUU---------
123 * Since the remapping operation can be applied to an arbitrary file
124 * range, we record the need for the remap step as a flag in the ioend
125 * instead of declaring a new IO type. This is required for direct io
126 * because we only have ioend for the whole dio, and we have to be able to
127 * remember the presence of unwritten blocks and CoW blocks with a single
128 * ioend structure. Better yet, the more ground we can cover with one
133 * Given an AG extent, find the lowest-numbered run of shared blocks
134 * within that range and return the range in fbno/flen. If
135 * find_end_of_shared is true, return the longest contiguous extent of
136 * shared blocks. If there are no shared extents, fbno and flen will
137 * be set to NULLAGBLOCK and 0, respectively.
140 xfs_reflink_find_shared(
141 struct xfs_mount *mp,
142 struct xfs_trans *tp,
148 bool find_end_of_shared)
150 struct xfs_buf *agbp;
151 struct xfs_btree_cur *cur;
154 error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
160 cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno);
162 error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen,
165 xfs_btree_del_cursor(cur, error);
167 xfs_trans_brelse(tp, agbp);
172 * Trim the mapping to the next block where there's a change in the
173 * shared/unshared status. More specifically, this means that we
174 * find the lowest-numbered extent of shared blocks that coincides with
175 * the given block mapping. If the shared extent overlaps the start of
176 * the mapping, trim the mapping to the end of the shared extent. If
177 * the shared region intersects the mapping, trim the mapping to the
178 * start of the shared extent. If there are no shared regions that
179 * overlap, just return the original extent.
182 xfs_reflink_trim_around_shared(
183 struct xfs_inode *ip,
184 struct xfs_bmbt_irec *irec,
195 /* Holes, unwritten, and delalloc extents cannot be shared */
196 if (!xfs_is_reflink_inode(ip) || !xfs_bmap_is_real_extent(irec)) {
201 trace_xfs_reflink_trim_around_shared(ip, irec);
203 agno = XFS_FSB_TO_AGNO(ip->i_mount, irec->br_startblock);
204 agbno = XFS_FSB_TO_AGBNO(ip->i_mount, irec->br_startblock);
205 aglen = irec->br_blockcount;
207 error = xfs_reflink_find_shared(ip->i_mount, NULL, agno, agbno,
208 aglen, &fbno, &flen, true);
212 *shared = *trimmed = false;
213 if (fbno == NULLAGBLOCK) {
214 /* No shared blocks at all. */
216 } else if (fbno == agbno) {
218 * The start of this extent is shared. Truncate the
219 * mapping at the end of the shared region so that a
220 * subsequent iteration starts at the start of the
223 irec->br_blockcount = flen;
230 * There's a shared extent midway through this extent.
231 * Truncate the mapping at the start of the shared
232 * extent so that a subsequent iteration starts at the
233 * start of the shared region.
235 irec->br_blockcount = fbno - agbno;
242 * Trim the passed in imap to the next shared/unshared extent boundary, and
243 * if imap->br_startoff points to a shared extent reserve space for it in the
244 * COW fork. In this case *shared is set to true, else to false.
246 * Note that imap will always contain the block numbers for the existing blocks
247 * in the data fork, as the upper layers need them for read-modify-write
251 xfs_reflink_reserve_cow(
252 struct xfs_inode *ip,
253 struct xfs_bmbt_irec *imap,
256 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
257 struct xfs_bmbt_irec got;
259 bool eof = false, trimmed;
260 struct xfs_iext_cursor icur;
263 * Search the COW fork extent list first. This serves two purposes:
264 * first this implement the speculative preallocation using cowextisze,
265 * so that we also unshared block adjacent to shared blocks instead
266 * of just the shared blocks themselves. Second the lookup in the
267 * extent list is generally faster than going out to the shared extent
271 if (!xfs_iext_lookup_extent(ip, ifp, imap->br_startoff, &icur, &got))
273 if (!eof && got.br_startoff <= imap->br_startoff) {
274 trace_xfs_reflink_cow_found(ip, imap);
275 xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
281 /* Trim the mapping to the nearest shared extent boundary. */
282 error = xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed);
286 /* Not shared? Just report the (potentially capped) extent. */
291 * Fork all the shared blocks from our write offset until the end of
294 error = xfs_qm_dqattach_locked(ip, false);
298 error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, imap->br_startoff,
299 imap->br_blockcount, 0, &got, &icur, eof);
300 if (error == -ENOSPC || error == -EDQUOT)
301 trace_xfs_reflink_cow_enospc(ip, imap);
305 trace_xfs_reflink_cow_alloc(ip, &got);
309 /* Convert part of an unwritten CoW extent to a real one. */
311 xfs_reflink_convert_cow_extent(
312 struct xfs_inode *ip,
313 struct xfs_bmbt_irec *imap,
314 xfs_fileoff_t offset_fsb,
315 xfs_filblks_t count_fsb)
319 if (imap->br_state == XFS_EXT_NORM)
322 xfs_trim_extent(imap, offset_fsb, count_fsb);
323 trace_xfs_reflink_convert_cow(ip, imap);
324 if (imap->br_blockcount == 0)
326 return xfs_bmapi_write(NULL, ip, imap->br_startoff, imap->br_blockcount,
327 XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, 0, imap,
331 /* Convert all of the unwritten CoW extents in a file's range to real ones. */
333 xfs_reflink_convert_cow(
334 struct xfs_inode *ip,
338 struct xfs_mount *mp = ip->i_mount;
339 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
340 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
341 xfs_filblks_t count_fsb = end_fsb - offset_fsb;
342 struct xfs_bmbt_irec imap;
343 int nimaps = 1, error = 0;
347 xfs_ilock(ip, XFS_ILOCK_EXCL);
348 error = xfs_bmapi_write(NULL, ip, offset_fsb, count_fsb,
349 XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT |
350 XFS_BMAPI_CONVERT_ONLY, 0, &imap, &nimaps);
351 xfs_iunlock(ip, XFS_ILOCK_EXCL);
356 * Find the extent that maps the given range in the COW fork. Even if the extent
357 * is not shared we might have a preallocation for it in the COW fork. If so we
358 * use it that rather than trigger a new allocation.
361 xfs_find_trim_cow_extent(
362 struct xfs_inode *ip,
363 struct xfs_bmbt_irec *imap,
367 xfs_fileoff_t offset_fsb = imap->br_startoff;
368 xfs_filblks_t count_fsb = imap->br_blockcount;
369 struct xfs_iext_cursor icur;
370 struct xfs_bmbt_irec got;
376 * If we don't find an overlapping extent, trim the range we need to
377 * allocate to fit the hole we found.
379 if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got) ||
380 got.br_startoff > offset_fsb)
381 return xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed);
384 if (isnullstartblock(got.br_startblock)) {
385 xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
389 /* real extent found - no need to allocate */
390 xfs_trim_extent(&got, offset_fsb, count_fsb);
396 /* Allocate all CoW reservations covering a range of blocks in a file. */
398 xfs_reflink_allocate_cow(
399 struct xfs_inode *ip,
400 struct xfs_bmbt_irec *imap,
404 struct xfs_mount *mp = ip->i_mount;
405 xfs_fileoff_t offset_fsb = imap->br_startoff;
406 xfs_filblks_t count_fsb = imap->br_blockcount;
407 struct xfs_trans *tp;
408 int nimaps, error = 0;
410 xfs_filblks_t resaligned;
411 xfs_extlen_t resblks = 0;
413 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
414 ASSERT(xfs_is_reflink_inode(ip));
416 error = xfs_find_trim_cow_extent(ip, imap, shared, &found);
417 if (error || !*shared)
422 resaligned = xfs_aligned_fsb_count(imap->br_startoff,
423 imap->br_blockcount, xfs_get_cowextsz_hint(ip));
424 resblks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
426 xfs_iunlock(ip, *lockmode);
427 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
428 *lockmode = XFS_ILOCK_EXCL;
429 xfs_ilock(ip, *lockmode);
434 error = xfs_qm_dqattach_locked(ip, false);
436 goto out_trans_cancel;
439 * Check for an overlapping extent again now that we dropped the ilock.
441 error = xfs_find_trim_cow_extent(ip, imap, shared, &found);
442 if (error || !*shared)
443 goto out_trans_cancel;
445 xfs_trans_cancel(tp);
449 error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0,
450 XFS_QMOPT_RES_REGBLKS);
452 goto out_trans_cancel;
454 xfs_trans_ijoin(tp, ip, 0);
456 /* Allocate the entire reservation as unwritten blocks. */
458 error = xfs_bmapi_write(tp, ip, imap->br_startoff, imap->br_blockcount,
459 XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC,
460 resblks, imap, &nimaps);
464 xfs_inode_set_cowblocks_tag(ip);
465 error = xfs_trans_commit(tp);
470 * Allocation succeeded but the requested range was not even partially
471 * satisfied? Bail out!
476 return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb);
479 xfs_trans_unreserve_quota_nblks(tp, ip, (long)resblks, 0,
480 XFS_QMOPT_RES_REGBLKS);
482 xfs_trans_cancel(tp);
487 * Cancel CoW reservations for some block range of an inode.
489 * If cancel_real is true this function cancels all COW fork extents for the
490 * inode; if cancel_real is false, real extents are not cleared.
492 * Caller must have already joined the inode to the current transaction. The
493 * inode will be joined to the transaction returned to the caller.
496 xfs_reflink_cancel_cow_blocks(
497 struct xfs_inode *ip,
498 struct xfs_trans **tpp,
499 xfs_fileoff_t offset_fsb,
500 xfs_fileoff_t end_fsb,
503 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
504 struct xfs_bmbt_irec got, del;
505 struct xfs_iext_cursor icur;
508 if (!xfs_inode_has_cow_data(ip))
510 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
513 /* Walk backwards until we're out of the I/O range... */
514 while (got.br_startoff + got.br_blockcount > offset_fsb) {
516 xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
518 /* Extent delete may have bumped ext forward */
519 if (!del.br_blockcount) {
520 xfs_iext_prev(ifp, &icur);
524 trace_xfs_reflink_cancel_cow(ip, &del);
526 if (isnullstartblock(del.br_startblock)) {
527 error = xfs_bmap_del_extent_delay(ip, XFS_COW_FORK,
531 } else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) {
532 ASSERT((*tpp)->t_firstblock == NULLFSBLOCK);
534 /* Free the CoW orphan record. */
535 error = xfs_refcount_free_cow_extent(*tpp,
536 del.br_startblock, del.br_blockcount);
540 xfs_bmap_add_free(*tpp, del.br_startblock,
541 del.br_blockcount, NULL);
543 /* Roll the transaction */
544 error = xfs_defer_finish(tpp);
548 /* Remove the mapping from the CoW fork. */
549 xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
551 /* Remove the quota reservation */
552 error = xfs_trans_reserve_quota_nblks(NULL, ip,
553 -(long)del.br_blockcount, 0,
554 XFS_QMOPT_RES_REGBLKS);
558 /* Didn't do anything, push cursor back. */
559 xfs_iext_prev(ifp, &icur);
562 if (!xfs_iext_get_extent(ifp, &icur, &got))
566 /* clear tag if cow fork is emptied */
568 xfs_inode_clear_cowblocks_tag(ip);
573 * Cancel CoW reservations for some byte range of an inode.
575 * If cancel_real is true this function cancels all COW fork extents for the
576 * inode; if cancel_real is false, real extents are not cleared.
579 xfs_reflink_cancel_cow_range(
580 struct xfs_inode *ip,
585 struct xfs_trans *tp;
586 xfs_fileoff_t offset_fsb;
587 xfs_fileoff_t end_fsb;
590 trace_xfs_reflink_cancel_cow_range(ip, offset, count);
591 ASSERT(xfs_is_reflink_inode(ip));
593 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
594 if (count == NULLFILEOFF)
595 end_fsb = NULLFILEOFF;
597 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
599 /* Start a rolling transaction to remove the mappings */
600 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
601 0, 0, XFS_TRANS_NOFS, &tp);
605 xfs_ilock(ip, XFS_ILOCK_EXCL);
606 xfs_trans_ijoin(tp, ip, 0);
608 /* Scrape out the old CoW reservations */
609 error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb,
614 error = xfs_trans_commit(tp);
616 xfs_iunlock(ip, XFS_ILOCK_EXCL);
620 xfs_trans_cancel(tp);
621 xfs_iunlock(ip, XFS_ILOCK_EXCL);
623 trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_);
628 * Remap parts of a file's data fork after a successful CoW.
632 struct xfs_inode *ip,
636 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
637 struct xfs_bmbt_irec got, del;
638 struct xfs_trans *tp;
639 xfs_fileoff_t offset_fsb;
640 xfs_fileoff_t end_fsb;
642 unsigned int resblks;
644 struct xfs_iext_cursor icur;
646 trace_xfs_reflink_end_cow(ip, offset, count);
648 /* No COW extents? That's easy! */
649 if (ifp->if_bytes == 0)
652 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
653 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
656 * Start a rolling transaction to switch the mappings. We're
657 * unlikely ever to have to remap 16T worth of single-block
658 * extents, so just cap the worst case extent count to 2^32-1.
659 * Stick a warning in just in case, and avoid 64-bit division.
661 BUILD_BUG_ON(MAX_RW_COUNT > UINT_MAX);
662 if (end_fsb - offset_fsb > UINT_MAX) {
663 error = -EFSCORRUPTED;
664 xfs_force_shutdown(ip->i_mount, SHUTDOWN_CORRUPT_INCORE);
668 resblks = XFS_NEXTENTADD_SPACE_RES(ip->i_mount,
669 (unsigned int)(end_fsb - offset_fsb),
671 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
672 resblks, 0, XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
676 xfs_ilock(ip, XFS_ILOCK_EXCL);
677 xfs_trans_ijoin(tp, ip, 0);
680 * In case of racing, overlapping AIO writes no COW extents might be
681 * left by the time I/O completes for the loser of the race. In that
684 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
687 /* Walk backwards until we're out of the I/O range... */
688 while (got.br_startoff + got.br_blockcount > offset_fsb) {
690 xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
692 /* Extent delete may have bumped ext forward */
693 if (!del.br_blockcount)
697 * Only remap real extent that contain data. With AIO
698 * speculatively preallocations can leak into the range we
699 * are called upon, and we need to skip them.
701 if (!xfs_bmap_is_real_extent(&got))
704 /* Unmap the old blocks in the data fork. */
705 ASSERT(tp->t_firstblock == NULLFSBLOCK);
706 rlen = del.br_blockcount;
707 error = __xfs_bunmapi(tp, ip, del.br_startoff, &rlen, 0, 1);
711 /* Trim the extent to whatever got unmapped. */
713 xfs_trim_extent(&del, del.br_startoff + rlen,
714 del.br_blockcount - rlen);
716 trace_xfs_reflink_cow_remap(ip, &del);
718 /* Free the CoW orphan record. */
719 error = xfs_refcount_free_cow_extent(tp, del.br_startblock,
724 /* Map the new blocks into the data fork. */
725 error = xfs_bmap_map_extent(tp, ip, &del);
729 /* Charge this new data fork mapping to the on-disk quota. */
730 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_DELBCOUNT,
731 (long)del.br_blockcount);
733 /* Remove the mapping from the CoW fork. */
734 xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
736 error = xfs_defer_finish(&tp);
739 if (!xfs_iext_get_extent(ifp, &icur, &got))
743 if (!xfs_iext_prev_extent(ifp, &icur, &got))
747 error = xfs_trans_commit(tp);
748 xfs_iunlock(ip, XFS_ILOCK_EXCL);
754 xfs_trans_cancel(tp);
755 xfs_iunlock(ip, XFS_ILOCK_EXCL);
757 trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
762 * Free leftover CoW reservations that didn't get cleaned out.
765 xfs_reflink_recover_cow(
766 struct xfs_mount *mp)
771 if (!xfs_sb_version_hasreflink(&mp->m_sb))
774 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
775 error = xfs_refcount_recover_cow_leftovers(mp, agno);
784 * Reflinking (Block) Ranges of Two Files Together
786 * First, ensure that the reflink flag is set on both inodes. The flag is an
787 * optimization to avoid unnecessary refcount btree lookups in the write path.
789 * Now we can iteratively remap the range of extents (and holes) in src to the
790 * corresponding ranges in dest. Let drange and srange denote the ranges of
791 * logical blocks in dest and src touched by the reflink operation.
793 * While the length of drange is greater than zero,
794 * - Read src's bmbt at the start of srange ("imap")
795 * - If imap doesn't exist, make imap appear to start at the end of srange
797 * - If imap starts before srange, advance imap to start at srange.
798 * - If imap goes beyond srange, truncate imap to end at the end of srange.
799 * - Punch (imap start - srange start + imap len) blocks from dest at
800 * offset (drange start).
801 * - If imap points to a real range of pblks,
802 * > Increase the refcount of the imap's pblks
803 * > Map imap's pblks into dest at the offset
804 * (drange start + imap start - srange start)
805 * - Advance drange and srange by (imap start - srange start + imap len)
807 * Finally, if the reflink made dest longer, update both the in-core and
808 * on-disk file sizes.
810 * ASCII Art Demonstration:
812 * Let's say we want to reflink this source file:
814 * ----SSSSSSS-SSSSS----SSSSSS (src file)
815 * <-------------------->
817 * into this destination file:
819 * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
820 * <-------------------->
821 * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
822 * Observe that the range has different logical offsets in either file.
824 * Consider that the first extent in the source file doesn't line up with our
825 * reflink range. Unmapping and remapping are separate operations, so we can
826 * unmap more blocks from the destination file than we remap.
828 * ----SSSSSSS-SSSSS----SSSSSS
830 * --DDDDD---------DDDDD--DDD
833 * Now remap the source extent into the destination file:
835 * ----SSSSSSS-SSSSS----SSSSSS
837 * --DDDDD--SSSSSSSDDDDD--DDD
840 * Do likewise with the second hole and extent in our range. Holes in the
841 * unmap range don't affect our operation.
843 * ----SSSSSSS-SSSSS----SSSSSS
845 * --DDDDD--SSSSSSS-SSSSS-DDD
848 * Finally, unmap and remap part of the third extent. This will increase the
849 * size of the destination file.
851 * ----SSSSSSS-SSSSS----SSSSSS
853 * --DDDDD--SSSSSSS-SSSSS----SSS
856 * Once we update the destination file's i_size, we're done.
860 * Ensure the reflink bit is set in both inodes.
863 xfs_reflink_set_inode_flag(
864 struct xfs_inode *src,
865 struct xfs_inode *dest)
867 struct xfs_mount *mp = src->i_mount;
869 struct xfs_trans *tp;
871 if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest))
874 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
878 /* Lock both files against IO */
879 if (src->i_ino == dest->i_ino)
880 xfs_ilock(src, XFS_ILOCK_EXCL);
882 xfs_lock_two_inodes(src, XFS_ILOCK_EXCL, dest, XFS_ILOCK_EXCL);
884 if (!xfs_is_reflink_inode(src)) {
885 trace_xfs_reflink_set_inode_flag(src);
886 xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL);
887 src->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
888 xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
889 xfs_ifork_init_cow(src);
891 xfs_iunlock(src, XFS_ILOCK_EXCL);
893 if (src->i_ino == dest->i_ino)
896 if (!xfs_is_reflink_inode(dest)) {
897 trace_xfs_reflink_set_inode_flag(dest);
898 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
899 dest->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
900 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
901 xfs_ifork_init_cow(dest);
903 xfs_iunlock(dest, XFS_ILOCK_EXCL);
906 error = xfs_trans_commit(tp);
912 trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_);
917 * Update destination inode size & cowextsize hint, if necessary.
920 xfs_reflink_update_dest(
921 struct xfs_inode *dest,
923 xfs_extlen_t cowextsize,
924 unsigned int remap_flags)
926 struct xfs_mount *mp = dest->i_mount;
927 struct xfs_trans *tp;
930 if ((remap_flags & REMAP_FILE_DEDUP) &&
931 newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0)
934 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
938 xfs_ilock(dest, XFS_ILOCK_EXCL);
939 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
941 if (newlen > i_size_read(VFS_I(dest))) {
942 trace_xfs_reflink_update_inode_size(dest, newlen);
943 i_size_write(VFS_I(dest), newlen);
944 dest->i_d.di_size = newlen;
948 dest->i_d.di_cowextsize = cowextsize;
949 dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
952 if (!(remap_flags & REMAP_FILE_DEDUP)) {
953 xfs_trans_ichgtime(tp, dest,
954 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
956 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
958 error = xfs_trans_commit(tp);
964 trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_);
969 * Do we have enough reserve in this AG to handle a reflink? The refcount
970 * btree already reserved all the space it needs, but the rmap btree can grow
971 * infinitely, so we won't allow more reflinks when the AG is down to the
975 xfs_reflink_ag_has_free_space(
976 struct xfs_mount *mp,
979 struct xfs_perag *pag;
982 if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
985 pag = xfs_perag_get(mp, agno);
986 if (xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) ||
987 xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA))
994 * Unmap a range of blocks from a file, then map other blocks into the hole.
995 * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount).
996 * The extent irec is mapped into dest at irec->br_startoff.
999 xfs_reflink_remap_extent(
1000 struct xfs_inode *ip,
1001 struct xfs_bmbt_irec *irec,
1002 xfs_fileoff_t destoff,
1003 xfs_off_t new_isize)
1005 struct xfs_mount *mp = ip->i_mount;
1006 bool real_extent = xfs_bmap_is_real_extent(irec);
1007 struct xfs_trans *tp;
1008 unsigned int resblks;
1009 struct xfs_bmbt_irec uirec;
1011 xfs_filblks_t unmap_len;
1015 unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
1016 trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
1018 /* No reflinking if we're low on space */
1020 error = xfs_reflink_ag_has_free_space(mp,
1021 XFS_FSB_TO_AGNO(mp, irec->br_startblock));
1026 /* Start a rolling transaction to switch the mappings */
1027 resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
1028 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1032 xfs_ilock(ip, XFS_ILOCK_EXCL);
1033 xfs_trans_ijoin(tp, ip, 0);
1035 /* If we're not just clearing space, then do we have enough quota? */
1037 error = xfs_trans_reserve_quota_nblks(tp, ip,
1038 irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS);
1043 trace_xfs_reflink_remap(ip, irec->br_startoff,
1044 irec->br_blockcount, irec->br_startblock);
1046 /* Unmap the old blocks in the data fork. */
1049 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1050 error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1);
1055 * Trim the extent to whatever got unmapped.
1056 * Remember, bunmapi works backwards.
1058 uirec.br_startblock = irec->br_startblock + rlen;
1059 uirec.br_startoff = irec->br_startoff + rlen;
1060 uirec.br_blockcount = unmap_len - rlen;
1063 /* If this isn't a real mapping, we're done. */
1064 if (!real_extent || uirec.br_blockcount == 0)
1067 trace_xfs_reflink_remap(ip, uirec.br_startoff,
1068 uirec.br_blockcount, uirec.br_startblock);
1070 /* Update the refcount tree */
1071 error = xfs_refcount_increase_extent(tp, &uirec);
1075 /* Map the new blocks into the data fork. */
1076 error = xfs_bmap_map_extent(tp, ip, &uirec);
1080 /* Update quota accounting. */
1081 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
1082 uirec.br_blockcount);
1084 /* Update dest isize if needed. */
1085 newlen = XFS_FSB_TO_B(mp,
1086 uirec.br_startoff + uirec.br_blockcount);
1087 newlen = min_t(xfs_off_t, newlen, new_isize);
1088 if (newlen > i_size_read(VFS_I(ip))) {
1089 trace_xfs_reflink_update_inode_size(ip, newlen);
1090 i_size_write(VFS_I(ip), newlen);
1091 ip->i_d.di_size = newlen;
1092 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1096 /* Process all the deferred stuff. */
1097 error = xfs_defer_finish(&tp);
1102 error = xfs_trans_commit(tp);
1103 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1109 xfs_trans_cancel(tp);
1110 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1112 trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
1117 * Iteratively remap one file's extents (and holes) to another's.
1120 xfs_reflink_remap_blocks(
1121 struct xfs_inode *src,
1123 struct xfs_inode *dest,
1128 struct xfs_bmbt_irec imap;
1129 xfs_fileoff_t srcoff;
1130 xfs_fileoff_t destoff;
1132 xfs_filblks_t range_len;
1133 xfs_filblks_t remapped_len = 0;
1134 xfs_off_t new_isize = pos_out + remap_len;
1138 destoff = XFS_B_TO_FSBT(src->i_mount, pos_out);
1139 srcoff = XFS_B_TO_FSBT(src->i_mount, pos_in);
1140 len = XFS_B_TO_FSB(src->i_mount, remap_len);
1142 /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
1146 trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
1149 /* Read extent from the source file */
1151 lock_mode = xfs_ilock_data_map_shared(src);
1152 error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0);
1153 xfs_iunlock(src, lock_mode);
1156 ASSERT(nimaps == 1);
1158 trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
1161 /* Translate imap into the destination file. */
1162 range_len = imap.br_startoff + imap.br_blockcount - srcoff;
1163 imap.br_startoff += destoff - srcoff;
1165 /* Clear dest from destoff to the end of imap and map it in. */
1166 error = xfs_reflink_remap_extent(dest, &imap, destoff,
1171 if (fatal_signal_pending(current)) {
1176 /* Advance drange/srange */
1177 srcoff += range_len;
1178 destoff += range_len;
1180 remapped_len += range_len;
1184 trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_);
1185 *remapped = min_t(loff_t, remap_len,
1186 XFS_FSB_TO_B(src->i_mount, remapped_len));
1191 * Grab the exclusive iolock for a data copy from src to dest, making
1192 * sure to abide vfs locking order (lowest pointer value goes first) and
1193 * breaking the pnfs layout leases on dest before proceeding. The loop
1194 * is needed because we cannot call the blocking break_layout() with the
1195 * src iolock held, and therefore have to back out both locks.
1198 xfs_iolock_two_inodes_and_break_layout(
1206 inode_lock_shared(src);
1207 inode_lock_nested(dest, I_MUTEX_NONDIR2);
1213 error = break_layout(dest, false);
1214 if (error == -EWOULDBLOCK) {
1217 inode_unlock_shared(src);
1218 error = break_layout(dest, true);
1226 inode_unlock_shared(src);
1230 inode_lock_shared_nested(src, I_MUTEX_NONDIR2);
1234 /* Unlock both inodes after they've been prepped for a range clone. */
1236 xfs_reflink_remap_unlock(
1237 struct file *file_in,
1238 struct file *file_out)
1240 struct inode *inode_in = file_inode(file_in);
1241 struct xfs_inode *src = XFS_I(inode_in);
1242 struct inode *inode_out = file_inode(file_out);
1243 struct xfs_inode *dest = XFS_I(inode_out);
1244 bool same_inode = (inode_in == inode_out);
1246 xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
1248 xfs_iunlock(src, XFS_MMAPLOCK_SHARED);
1249 inode_unlock(inode_out);
1251 inode_unlock_shared(inode_in);
1255 * If we're reflinking to a point past the destination file's EOF, we must
1256 * zero any speculative post-EOF preallocations that sit between the old EOF
1257 * and the destination file offset.
1260 xfs_reflink_zero_posteof(
1261 struct xfs_inode *ip,
1264 loff_t isize = i_size_read(VFS_I(ip));
1269 trace_xfs_zero_eof(ip, isize, pos - isize);
1270 return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
1275 * Prepare two files for range cloning. Upon a successful return both inodes
1276 * will have the iolock and mmaplock held, the page cache of the out file will
1277 * be truncated, and any leases on the out file will have been broken. This
1278 * function borrows heavily from xfs_file_aio_write_checks.
1280 * The VFS allows partial EOF blocks to "match" for dedupe even though it hasn't
1281 * checked that the bytes beyond EOF physically match. Hence we cannot use the
1282 * EOF block in the source dedupe range because it's not a complete block match,
1283 * hence can introduce a corruption into the file that has it's block replaced.
1285 * In similar fashion, the VFS file cloning also allows partial EOF blocks to be
1286 * "block aligned" for the purposes of cloning entire files. However, if the
1287 * source file range includes the EOF block and it lands within the existing EOF
1288 * of the destination file, then we can expose stale data from beyond the source
1289 * file EOF in the destination file.
1291 * XFS doesn't support partial block sharing, so in both cases we have check
1292 * these cases ourselves. For dedupe, we can simply round the length to dedupe
1293 * down to the previous whole block and ignore the partial EOF block. While this
1294 * means we can't dedupe the last block of a file, this is an acceptible
1295 * tradeoff for simplicity on implementation.
1297 * For cloning, we want to share the partial EOF block if it is also the new EOF
1298 * block of the destination file. If the partial EOF block lies inside the
1299 * existing destination EOF, then we have to abort the clone to avoid exposing
1300 * stale data in the destination file. Hence we reject these clone attempts with
1301 * -EINVAL in this case.
1304 xfs_reflink_remap_prep(
1305 struct file *file_in,
1307 struct file *file_out,
1310 unsigned int remap_flags)
1312 struct inode *inode_in = file_inode(file_in);
1313 struct xfs_inode *src = XFS_I(inode_in);
1314 struct inode *inode_out = file_inode(file_out);
1315 struct xfs_inode *dest = XFS_I(inode_out);
1316 bool same_inode = (inode_in == inode_out);
1319 /* Lock both files against IO */
1320 ret = xfs_iolock_two_inodes_and_break_layout(inode_in, inode_out);
1324 xfs_ilock(src, XFS_MMAPLOCK_EXCL);
1326 xfs_lock_two_inodes(src, XFS_MMAPLOCK_SHARED, dest,
1329 /* Check file eligibility and prepare for block sharing. */
1331 /* Don't reflink realtime inodes */
1332 if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
1335 /* Don't share DAX file data for now. */
1336 if (IS_DAX(inode_in) || IS_DAX(inode_out))
1339 ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
1341 if (ret < 0 || *len == 0)
1344 /* Attach dquots to dest inode before changing block map */
1345 ret = xfs_qm_dqattach(dest);
1350 * Zero existing post-eof speculative preallocations in the destination
1353 ret = xfs_reflink_zero_posteof(dest, pos_out);
1357 /* Set flags and remap blocks. */
1358 ret = xfs_reflink_set_inode_flag(src, dest);
1362 /* Zap any page cache for the destination file's range. */
1363 truncate_inode_pages_range(&inode_out->i_data,
1364 round_down(pos_out, PAGE_SIZE),
1365 round_up(pos_out + *len, PAGE_SIZE) - 1);
1369 xfs_reflink_remap_unlock(file_in, file_out);
1374 * Link a range of blocks from one file to another.
1377 xfs_reflink_remap_range(
1378 struct file *file_in,
1380 struct file *file_out,
1383 unsigned int remap_flags)
1385 struct inode *inode_in = file_inode(file_in);
1386 struct xfs_inode *src = XFS_I(inode_in);
1387 struct inode *inode_out = file_inode(file_out);
1388 struct xfs_inode *dest = XFS_I(inode_out);
1389 struct xfs_mount *mp = src->i_mount;
1390 loff_t remapped = 0;
1391 xfs_extlen_t cowextsize;
1394 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1397 if (XFS_FORCED_SHUTDOWN(mp))
1400 /* Prepare and then clone file data. */
1401 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1403 if (ret < 0 || len == 0)
1406 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1408 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1414 * Carry the cowextsize hint from src to dest if we're sharing the
1415 * entire source file to the entire destination file, the source file
1416 * has a cowextsize hint, and the destination file does not.
1419 if (pos_in == 0 && len == i_size_read(inode_in) &&
1420 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1421 pos_out == 0 && len >= i_size_read(inode_out) &&
1422 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1423 cowextsize = src->i_d.di_cowextsize;
1425 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1429 xfs_reflink_remap_unlock(file_in, file_out);
1431 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1432 return remapped > 0 ? remapped : ret;
1436 * The user wants to preemptively CoW all shared blocks in this file,
1437 * which enables us to turn off the reflink flag. Iterate all
1438 * extents which are not prealloc/delalloc to see which ranges are
1439 * mentioned in the refcount tree, then read those blocks into the
1440 * pagecache, dirty them, fsync them back out, and then we can update
1441 * the inode flag. What happens if we run out of memory? :)
1444 xfs_reflink_dirty_extents(
1445 struct xfs_inode *ip,
1450 struct xfs_mount *mp = ip->i_mount;
1451 xfs_agnumber_t agno;
1452 xfs_agblock_t agbno;
1458 struct xfs_bmbt_irec map[2];
1462 while (end - fbno > 0) {
1465 * Look for extents in the file. Skip holes, delalloc, or
1466 * unwritten extents; they can't be reflinked.
1468 error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0);
1473 if (!xfs_bmap_is_real_extent(&map[0]))
1477 while (map[1].br_blockcount) {
1478 agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock);
1479 agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock);
1480 aglen = map[1].br_blockcount;
1482 error = xfs_reflink_find_shared(mp, NULL, agno, agbno,
1483 aglen, &rbno, &rlen, true);
1486 if (rbno == NULLAGBLOCK)
1489 /* Dirty the pages */
1490 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1491 fpos = XFS_FSB_TO_B(mp, map[1].br_startoff +
1493 flen = XFS_FSB_TO_B(mp, rlen);
1494 if (fpos + flen > isize)
1495 flen = isize - fpos;
1496 error = iomap_file_dirty(VFS_I(ip), fpos, flen,
1498 xfs_ilock(ip, XFS_ILOCK_EXCL);
1502 map[1].br_blockcount -= (rbno - agbno + rlen);
1503 map[1].br_startoff += (rbno - agbno + rlen);
1504 map[1].br_startblock += (rbno - agbno + rlen);
1508 fbno = map[0].br_startoff + map[0].br_blockcount;
1514 /* Does this inode need the reflink flag? */
1516 xfs_reflink_inode_has_shared_extents(
1517 struct xfs_trans *tp,
1518 struct xfs_inode *ip,
1521 struct xfs_bmbt_irec got;
1522 struct xfs_mount *mp = ip->i_mount;
1523 struct xfs_ifork *ifp;
1524 xfs_agnumber_t agno;
1525 xfs_agblock_t agbno;
1529 struct xfs_iext_cursor icur;
1533 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1534 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1535 error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1540 *has_shared = false;
1541 found = xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got);
1543 if (isnullstartblock(got.br_startblock) ||
1544 got.br_state != XFS_EXT_NORM)
1546 agno = XFS_FSB_TO_AGNO(mp, got.br_startblock);
1547 agbno = XFS_FSB_TO_AGBNO(mp, got.br_startblock);
1548 aglen = got.br_blockcount;
1550 error = xfs_reflink_find_shared(mp, tp, agno, agbno, aglen,
1551 &rbno, &rlen, false);
1554 /* Is there still a shared block here? */
1555 if (rbno != NULLAGBLOCK) {
1560 found = xfs_iext_next_extent(ifp, &icur, &got);
1567 * Clear the inode reflink flag if there are no shared extents.
1569 * The caller is responsible for joining the inode to the transaction passed in.
1570 * The inode will be joined to the transaction that is returned to the caller.
1573 xfs_reflink_clear_inode_flag(
1574 struct xfs_inode *ip,
1575 struct xfs_trans **tpp)
1580 ASSERT(xfs_is_reflink_inode(ip));
1582 error = xfs_reflink_inode_has_shared_extents(*tpp, ip, &needs_flag);
1583 if (error || needs_flag)
1587 * We didn't find any shared blocks so turn off the reflink flag.
1588 * First, get rid of any leftover CoW mappings.
1590 error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF, true);
1594 /* Clear the inode flag. */
1595 trace_xfs_reflink_unset_inode_flag(ip);
1596 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1597 xfs_inode_clear_cowblocks_tag(ip);
1598 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1604 * Clear the inode reflink flag if there are no shared extents and the size
1608 xfs_reflink_try_clear_inode_flag(
1609 struct xfs_inode *ip)
1611 struct xfs_mount *mp = ip->i_mount;
1612 struct xfs_trans *tp;
1615 /* Start a rolling transaction to remove the mappings */
1616 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1620 xfs_ilock(ip, XFS_ILOCK_EXCL);
1621 xfs_trans_ijoin(tp, ip, 0);
1623 error = xfs_reflink_clear_inode_flag(ip, &tp);
1627 error = xfs_trans_commit(tp);
1631 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1634 xfs_trans_cancel(tp);
1636 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1641 * Pre-COW all shared blocks within a given byte range of a file and turn off
1642 * the reflink flag if we unshare all of the file's blocks.
1645 xfs_reflink_unshare(
1646 struct xfs_inode *ip,
1650 struct xfs_mount *mp = ip->i_mount;
1656 if (!xfs_is_reflink_inode(ip))
1659 trace_xfs_reflink_unshare(ip, offset, len);
1661 inode_dio_wait(VFS_I(ip));
1663 /* Try to CoW the selected ranges */
1664 xfs_ilock(ip, XFS_ILOCK_EXCL);
1665 fbno = XFS_B_TO_FSBT(mp, offset);
1666 isize = i_size_read(VFS_I(ip));
1667 end = XFS_B_TO_FSB(mp, offset + len);
1668 error = xfs_reflink_dirty_extents(ip, fbno, end, isize);
1671 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1673 /* Wait for the IO to finish */
1674 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1678 /* Turn off the reflink flag if possible. */
1679 error = xfs_reflink_try_clear_inode_flag(ip);
1686 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1688 trace_xfs_reflink_unshare_error(ip, error, _RET_IP_);