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 xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
306 trace_xfs_reflink_cow_alloc(ip, &got);
310 /* Convert part of an unwritten CoW extent to a real one. */
312 xfs_reflink_convert_cow_extent(
313 struct xfs_inode *ip,
314 struct xfs_bmbt_irec *imap,
315 xfs_fileoff_t offset_fsb,
316 xfs_filblks_t count_fsb)
320 if (imap->br_state == XFS_EXT_NORM)
323 xfs_trim_extent(imap, offset_fsb, count_fsb);
324 trace_xfs_reflink_convert_cow(ip, imap);
325 if (imap->br_blockcount == 0)
327 return xfs_bmapi_write(NULL, ip, imap->br_startoff, imap->br_blockcount,
328 XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, 0, imap,
332 /* Convert all of the unwritten CoW extents in a file's range to real ones. */
334 xfs_reflink_convert_cow(
335 struct xfs_inode *ip,
339 struct xfs_mount *mp = ip->i_mount;
340 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
341 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
342 xfs_filblks_t count_fsb = end_fsb - offset_fsb;
343 struct xfs_bmbt_irec imap;
344 int nimaps = 1, error = 0;
348 xfs_ilock(ip, XFS_ILOCK_EXCL);
349 error = xfs_bmapi_write(NULL, ip, offset_fsb, count_fsb,
350 XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT |
351 XFS_BMAPI_CONVERT_ONLY, 0, &imap, &nimaps);
352 xfs_iunlock(ip, XFS_ILOCK_EXCL);
357 * Find the extent that maps the given range in the COW fork. Even if the extent
358 * is not shared we might have a preallocation for it in the COW fork. If so we
359 * use it that rather than trigger a new allocation.
362 xfs_find_trim_cow_extent(
363 struct xfs_inode *ip,
364 struct xfs_bmbt_irec *imap,
368 xfs_fileoff_t offset_fsb = imap->br_startoff;
369 xfs_filblks_t count_fsb = imap->br_blockcount;
370 struct xfs_iext_cursor icur;
371 struct xfs_bmbt_irec got;
377 * If we don't find an overlapping extent, trim the range we need to
378 * allocate to fit the hole we found.
380 if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got) ||
381 got.br_startoff > offset_fsb)
382 return xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed);
385 if (isnullstartblock(got.br_startblock)) {
386 xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
390 /* real extent found - no need to allocate */
391 xfs_trim_extent(&got, offset_fsb, count_fsb);
397 /* Allocate all CoW reservations covering a range of blocks in a file. */
399 xfs_reflink_allocate_cow(
400 struct xfs_inode *ip,
401 struct xfs_bmbt_irec *imap,
405 struct xfs_mount *mp = ip->i_mount;
406 xfs_fileoff_t offset_fsb = imap->br_startoff;
407 xfs_filblks_t count_fsb = imap->br_blockcount;
408 struct xfs_trans *tp;
409 int nimaps, error = 0;
411 xfs_filblks_t resaligned;
412 xfs_extlen_t resblks = 0;
414 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
415 ASSERT(xfs_is_reflink_inode(ip));
417 error = xfs_find_trim_cow_extent(ip, imap, shared, &found);
418 if (error || !*shared)
423 resaligned = xfs_aligned_fsb_count(imap->br_startoff,
424 imap->br_blockcount, xfs_get_cowextsz_hint(ip));
425 resblks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
427 xfs_iunlock(ip, *lockmode);
428 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
429 *lockmode = XFS_ILOCK_EXCL;
430 xfs_ilock(ip, *lockmode);
435 error = xfs_qm_dqattach_locked(ip, false);
437 goto out_trans_cancel;
440 * Check for an overlapping extent again now that we dropped the ilock.
442 error = xfs_find_trim_cow_extent(ip, imap, shared, &found);
443 if (error || !*shared)
444 goto out_trans_cancel;
446 xfs_trans_cancel(tp);
450 error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0,
451 XFS_QMOPT_RES_REGBLKS);
453 goto out_trans_cancel;
455 xfs_trans_ijoin(tp, ip, 0);
457 /* Allocate the entire reservation as unwritten blocks. */
459 error = xfs_bmapi_write(tp, ip, imap->br_startoff, imap->br_blockcount,
460 XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC,
461 resblks, imap, &nimaps);
465 xfs_inode_set_cowblocks_tag(ip);
466 error = xfs_trans_commit(tp);
471 * Allocation succeeded but the requested range was not even partially
472 * satisfied? Bail out!
477 return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb);
480 xfs_trans_unreserve_quota_nblks(tp, ip, (long)resblks, 0,
481 XFS_QMOPT_RES_REGBLKS);
483 xfs_trans_cancel(tp);
488 * Cancel CoW reservations for some block range of an inode.
490 * If cancel_real is true this function cancels all COW fork extents for the
491 * inode; if cancel_real is false, real extents are not cleared.
493 * Caller must have already joined the inode to the current transaction. The
494 * inode will be joined to the transaction returned to the caller.
497 xfs_reflink_cancel_cow_blocks(
498 struct xfs_inode *ip,
499 struct xfs_trans **tpp,
500 xfs_fileoff_t offset_fsb,
501 xfs_fileoff_t end_fsb,
504 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
505 struct xfs_bmbt_irec got, del;
506 struct xfs_iext_cursor icur;
509 if (!xfs_inode_has_cow_data(ip))
511 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
514 /* Walk backwards until we're out of the I/O range... */
515 while (got.br_startoff + got.br_blockcount > offset_fsb) {
517 xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
519 /* Extent delete may have bumped ext forward */
520 if (!del.br_blockcount) {
521 xfs_iext_prev(ifp, &icur);
525 trace_xfs_reflink_cancel_cow(ip, &del);
527 if (isnullstartblock(del.br_startblock)) {
528 error = xfs_bmap_del_extent_delay(ip, XFS_COW_FORK,
532 } else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) {
533 ASSERT((*tpp)->t_firstblock == NULLFSBLOCK);
535 /* Free the CoW orphan record. */
536 error = xfs_refcount_free_cow_extent(*tpp,
537 del.br_startblock, del.br_blockcount);
541 xfs_bmap_add_free(*tpp, del.br_startblock,
542 del.br_blockcount, NULL);
544 /* Roll the transaction */
545 error = xfs_defer_finish(tpp);
549 /* Remove the mapping from the CoW fork. */
550 xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
552 /* Remove the quota reservation */
553 error = xfs_trans_reserve_quota_nblks(NULL, ip,
554 -(long)del.br_blockcount, 0,
555 XFS_QMOPT_RES_REGBLKS);
559 /* Didn't do anything, push cursor back. */
560 xfs_iext_prev(ifp, &icur);
563 if (!xfs_iext_get_extent(ifp, &icur, &got))
567 /* clear tag if cow fork is emptied */
569 xfs_inode_clear_cowblocks_tag(ip);
574 * Cancel CoW reservations for some byte range of an inode.
576 * If cancel_real is true this function cancels all COW fork extents for the
577 * inode; if cancel_real is false, real extents are not cleared.
580 xfs_reflink_cancel_cow_range(
581 struct xfs_inode *ip,
586 struct xfs_trans *tp;
587 xfs_fileoff_t offset_fsb;
588 xfs_fileoff_t end_fsb;
591 trace_xfs_reflink_cancel_cow_range(ip, offset, count);
592 ASSERT(xfs_is_reflink_inode(ip));
594 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
595 if (count == NULLFILEOFF)
596 end_fsb = NULLFILEOFF;
598 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
600 /* Start a rolling transaction to remove the mappings */
601 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
602 0, 0, XFS_TRANS_NOFS, &tp);
606 xfs_ilock(ip, XFS_ILOCK_EXCL);
607 xfs_trans_ijoin(tp, ip, 0);
609 /* Scrape out the old CoW reservations */
610 error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb,
615 error = xfs_trans_commit(tp);
617 xfs_iunlock(ip, XFS_ILOCK_EXCL);
621 xfs_trans_cancel(tp);
622 xfs_iunlock(ip, XFS_ILOCK_EXCL);
624 trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_);
629 * Remap parts of a file's data fork after a successful CoW.
633 struct xfs_inode *ip,
637 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
638 struct xfs_bmbt_irec got, del;
639 struct xfs_trans *tp;
640 xfs_fileoff_t offset_fsb;
641 xfs_fileoff_t end_fsb;
643 unsigned int resblks;
645 struct xfs_iext_cursor icur;
647 trace_xfs_reflink_end_cow(ip, offset, count);
649 /* No COW extents? That's easy! */
650 if (ifp->if_bytes == 0)
653 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
654 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
657 * Start a rolling transaction to switch the mappings. We're
658 * unlikely ever to have to remap 16T worth of single-block
659 * extents, so just cap the worst case extent count to 2^32-1.
660 * Stick a warning in just in case, and avoid 64-bit division.
662 BUILD_BUG_ON(MAX_RW_COUNT > UINT_MAX);
663 if (end_fsb - offset_fsb > UINT_MAX) {
664 error = -EFSCORRUPTED;
665 xfs_force_shutdown(ip->i_mount, SHUTDOWN_CORRUPT_INCORE);
669 resblks = XFS_NEXTENTADD_SPACE_RES(ip->i_mount,
670 (unsigned int)(end_fsb - offset_fsb),
672 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
673 resblks, 0, XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
677 xfs_ilock(ip, XFS_ILOCK_EXCL);
678 xfs_trans_ijoin(tp, ip, 0);
681 * In case of racing, overlapping AIO writes no COW extents might be
682 * left by the time I/O completes for the loser of the race. In that
685 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
688 /* Walk backwards until we're out of the I/O range... */
689 while (got.br_startoff + got.br_blockcount > offset_fsb) {
691 xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
693 /* Extent delete may have bumped ext forward */
694 if (!del.br_blockcount)
698 * Only remap real extent that contain data. With AIO
699 * speculatively preallocations can leak into the range we
700 * are called upon, and we need to skip them.
702 if (!xfs_bmap_is_real_extent(&got))
705 /* Unmap the old blocks in the data fork. */
706 ASSERT(tp->t_firstblock == NULLFSBLOCK);
707 rlen = del.br_blockcount;
708 error = __xfs_bunmapi(tp, ip, del.br_startoff, &rlen, 0, 1);
712 /* Trim the extent to whatever got unmapped. */
714 xfs_trim_extent(&del, del.br_startoff + rlen,
715 del.br_blockcount - rlen);
717 trace_xfs_reflink_cow_remap(ip, &del);
719 /* Free the CoW orphan record. */
720 error = xfs_refcount_free_cow_extent(tp, del.br_startblock,
725 /* Map the new blocks into the data fork. */
726 error = xfs_bmap_map_extent(tp, ip, &del);
730 /* Charge this new data fork mapping to the on-disk quota. */
731 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_DELBCOUNT,
732 (long)del.br_blockcount);
734 /* Remove the mapping from the CoW fork. */
735 xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
737 error = xfs_defer_finish(&tp);
740 if (!xfs_iext_get_extent(ifp, &icur, &got))
744 if (!xfs_iext_prev_extent(ifp, &icur, &got))
748 error = xfs_trans_commit(tp);
749 xfs_iunlock(ip, XFS_ILOCK_EXCL);
755 xfs_trans_cancel(tp);
756 xfs_iunlock(ip, XFS_ILOCK_EXCL);
758 trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
763 * Free leftover CoW reservations that didn't get cleaned out.
766 xfs_reflink_recover_cow(
767 struct xfs_mount *mp)
772 if (!xfs_sb_version_hasreflink(&mp->m_sb))
775 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
776 error = xfs_refcount_recover_cow_leftovers(mp, agno);
785 * Reflinking (Block) Ranges of Two Files Together
787 * First, ensure that the reflink flag is set on both inodes. The flag is an
788 * optimization to avoid unnecessary refcount btree lookups in the write path.
790 * Now we can iteratively remap the range of extents (and holes) in src to the
791 * corresponding ranges in dest. Let drange and srange denote the ranges of
792 * logical blocks in dest and src touched by the reflink operation.
794 * While the length of drange is greater than zero,
795 * - Read src's bmbt at the start of srange ("imap")
796 * - If imap doesn't exist, make imap appear to start at the end of srange
798 * - If imap starts before srange, advance imap to start at srange.
799 * - If imap goes beyond srange, truncate imap to end at the end of srange.
800 * - Punch (imap start - srange start + imap len) blocks from dest at
801 * offset (drange start).
802 * - If imap points to a real range of pblks,
803 * > Increase the refcount of the imap's pblks
804 * > Map imap's pblks into dest at the offset
805 * (drange start + imap start - srange start)
806 * - Advance drange and srange by (imap start - srange start + imap len)
808 * Finally, if the reflink made dest longer, update both the in-core and
809 * on-disk file sizes.
811 * ASCII Art Demonstration:
813 * Let's say we want to reflink this source file:
815 * ----SSSSSSS-SSSSS----SSSSSS (src file)
816 * <-------------------->
818 * into this destination file:
820 * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
821 * <-------------------->
822 * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
823 * Observe that the range has different logical offsets in either file.
825 * Consider that the first extent in the source file doesn't line up with our
826 * reflink range. Unmapping and remapping are separate operations, so we can
827 * unmap more blocks from the destination file than we remap.
829 * ----SSSSSSS-SSSSS----SSSSSS
831 * --DDDDD---------DDDDD--DDD
834 * Now remap the source extent into the destination file:
836 * ----SSSSSSS-SSSSS----SSSSSS
838 * --DDDDD--SSSSSSSDDDDD--DDD
841 * Do likewise with the second hole and extent in our range. Holes in the
842 * unmap range don't affect our operation.
844 * ----SSSSSSS-SSSSS----SSSSSS
846 * --DDDDD--SSSSSSS-SSSSS-DDD
849 * Finally, unmap and remap part of the third extent. This will increase the
850 * size of the destination file.
852 * ----SSSSSSS-SSSSS----SSSSSS
854 * --DDDDD--SSSSSSS-SSSSS----SSS
857 * Once we update the destination file's i_size, we're done.
861 * Ensure the reflink bit is set in both inodes.
864 xfs_reflink_set_inode_flag(
865 struct xfs_inode *src,
866 struct xfs_inode *dest)
868 struct xfs_mount *mp = src->i_mount;
870 struct xfs_trans *tp;
872 if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest))
875 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
879 /* Lock both files against IO */
880 if (src->i_ino == dest->i_ino)
881 xfs_ilock(src, XFS_ILOCK_EXCL);
883 xfs_lock_two_inodes(src, XFS_ILOCK_EXCL, dest, XFS_ILOCK_EXCL);
885 if (!xfs_is_reflink_inode(src)) {
886 trace_xfs_reflink_set_inode_flag(src);
887 xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL);
888 src->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
889 xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
890 xfs_ifork_init_cow(src);
892 xfs_iunlock(src, XFS_ILOCK_EXCL);
894 if (src->i_ino == dest->i_ino)
897 if (!xfs_is_reflink_inode(dest)) {
898 trace_xfs_reflink_set_inode_flag(dest);
899 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
900 dest->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
901 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
902 xfs_ifork_init_cow(dest);
904 xfs_iunlock(dest, XFS_ILOCK_EXCL);
907 error = xfs_trans_commit(tp);
913 trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_);
918 * Update destination inode size & cowextsize hint, if necessary.
921 xfs_reflink_update_dest(
922 struct xfs_inode *dest,
924 xfs_extlen_t cowextsize,
927 struct xfs_mount *mp = dest->i_mount;
928 struct xfs_trans *tp;
931 if (is_dedupe && 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;
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;
1016 unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
1017 trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
1019 /* No reflinking if we're low on space */
1021 error = xfs_reflink_ag_has_free_space(mp,
1022 XFS_FSB_TO_AGNO(mp, irec->br_startblock));
1027 /* Start a rolling transaction to switch the mappings */
1028 resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
1029 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1033 xfs_ilock(ip, XFS_ILOCK_EXCL);
1034 xfs_trans_ijoin(tp, ip, 0);
1037 * Reserve quota for this operation. We don't know if the first unmap
1038 * in the dest file will cause a bmap btree split, so we always reserve
1039 * at least enough blocks for that split. If the extent being mapped
1040 * in is written, we need to reserve quota for that too.
1042 qres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
1044 qres += irec->br_blockcount;
1045 error = xfs_trans_reserve_quota_nblks(tp, ip, qres, 0,
1046 XFS_QMOPT_RES_REGBLKS);
1050 trace_xfs_reflink_remap(ip, irec->br_startoff,
1051 irec->br_blockcount, irec->br_startblock);
1053 /* Unmap the old blocks in the data fork. */
1056 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1057 error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1);
1062 * Trim the extent to whatever got unmapped.
1063 * Remember, bunmapi works backwards.
1065 uirec.br_startblock = irec->br_startblock + rlen;
1066 uirec.br_startoff = irec->br_startoff + rlen;
1067 uirec.br_blockcount = unmap_len - rlen;
1068 uirec.br_state = irec->br_state;
1071 /* If this isn't a real mapping, we're done. */
1072 if (!real_extent || uirec.br_blockcount == 0)
1075 trace_xfs_reflink_remap(ip, uirec.br_startoff,
1076 uirec.br_blockcount, uirec.br_startblock);
1078 /* Update the refcount tree */
1079 error = xfs_refcount_increase_extent(tp, &uirec);
1083 /* Map the new blocks into the data fork. */
1084 error = xfs_bmap_map_extent(tp, ip, &uirec);
1088 /* Update quota accounting. */
1089 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
1090 uirec.br_blockcount);
1092 /* Update dest isize if needed. */
1093 newlen = XFS_FSB_TO_B(mp,
1094 uirec.br_startoff + uirec.br_blockcount);
1095 newlen = min_t(xfs_off_t, newlen, new_isize);
1096 if (newlen > i_size_read(VFS_I(ip))) {
1097 trace_xfs_reflink_update_inode_size(ip, newlen);
1098 i_size_write(VFS_I(ip), newlen);
1099 ip->i_d.di_size = newlen;
1100 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1104 /* Process all the deferred stuff. */
1105 error = xfs_defer_finish(&tp);
1110 error = xfs_trans_commit(tp);
1111 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1117 xfs_trans_cancel(tp);
1118 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1120 trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
1125 * Iteratively remap one file's extents (and holes) to another's.
1128 xfs_reflink_remap_blocks(
1129 struct xfs_inode *src,
1130 xfs_fileoff_t srcoff,
1131 struct xfs_inode *dest,
1132 xfs_fileoff_t destoff,
1134 xfs_off_t new_isize)
1136 struct xfs_bmbt_irec imap;
1139 xfs_filblks_t range_len;
1141 /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
1145 trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
1148 /* Read extent from the source file */
1150 lock_mode = xfs_ilock_data_map_shared(src);
1151 error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0);
1152 xfs_iunlock(src, lock_mode);
1155 ASSERT(nimaps == 1);
1157 trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
1160 /* Translate imap into the destination file. */
1161 range_len = imap.br_startoff + imap.br_blockcount - srcoff;
1162 imap.br_startoff += destoff - srcoff;
1164 /* Clear dest from destoff to the end of imap and map it in. */
1165 error = xfs_reflink_remap_extent(dest, &imap, destoff,
1170 if (fatal_signal_pending(current)) {
1175 /* Advance drange/srange */
1176 srcoff += range_len;
1177 destoff += range_len;
1184 trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_);
1189 * Grab the exclusive iolock for a data copy from src to dest, making
1190 * sure to abide vfs locking order (lowest pointer value goes first) and
1191 * breaking the pnfs layout leases on dest before proceeding. The loop
1192 * is needed because we cannot call the blocking break_layout() with the
1193 * src iolock held, and therefore have to back out both locks.
1196 xfs_iolock_two_inodes_and_break_layout(
1204 inode_lock_shared(src);
1205 inode_lock_nested(dest, I_MUTEX_NONDIR2);
1211 error = break_layout(dest, false);
1212 if (error == -EWOULDBLOCK) {
1215 inode_unlock_shared(src);
1216 error = break_layout(dest, true);
1224 inode_unlock_shared(src);
1228 inode_lock_shared_nested(src, I_MUTEX_NONDIR2);
1232 /* Unlock both inodes after they've been prepped for a range clone. */
1234 xfs_reflink_remap_unlock(
1235 struct file *file_in,
1236 struct file *file_out)
1238 struct inode *inode_in = file_inode(file_in);
1239 struct xfs_inode *src = XFS_I(inode_in);
1240 struct inode *inode_out = file_inode(file_out);
1241 struct xfs_inode *dest = XFS_I(inode_out);
1242 bool same_inode = (inode_in == inode_out);
1244 xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
1246 xfs_iunlock(src, XFS_MMAPLOCK_SHARED);
1247 inode_unlock(inode_out);
1249 inode_unlock_shared(inode_in);
1253 * If we're reflinking to a point past the destination file's EOF, we must
1254 * zero any speculative post-EOF preallocations that sit between the old EOF
1255 * and the destination file offset.
1258 xfs_reflink_zero_posteof(
1259 struct xfs_inode *ip,
1262 loff_t isize = i_size_read(VFS_I(ip));
1267 trace_xfs_zero_eof(ip, isize, pos - isize);
1268 return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
1273 * Prepare two files for range cloning. Upon a successful return both inodes
1274 * will have the iolock and mmaplock held, the page cache of the out file will
1275 * be truncated, and any leases on the out file will have been broken. This
1276 * function borrows heavily from xfs_file_aio_write_checks.
1278 * The VFS allows partial EOF blocks to "match" for dedupe even though it hasn't
1279 * checked that the bytes beyond EOF physically match. Hence we cannot use the
1280 * EOF block in the source dedupe range because it's not a complete block match,
1281 * hence can introduce a corruption into the file that has it's block replaced.
1283 * In similar fashion, the VFS file cloning also allows partial EOF blocks to be
1284 * "block aligned" for the purposes of cloning entire files. However, if the
1285 * source file range includes the EOF block and it lands within the existing EOF
1286 * of the destination file, then we can expose stale data from beyond the source
1287 * file EOF in the destination file.
1289 * XFS doesn't support partial block sharing, so in both cases we have check
1290 * these cases ourselves. For dedupe, we can simply round the length to dedupe
1291 * down to the previous whole block and ignore the partial EOF block. While this
1292 * means we can't dedupe the last block of a file, this is an acceptible
1293 * tradeoff for simplicity on implementation.
1295 * For cloning, we want to share the partial EOF block if it is also the new EOF
1296 * block of the destination file. If the partial EOF block lies inside the
1297 * existing destination EOF, then we have to abort the clone to avoid exposing
1298 * stale data in the destination file. Hence we reject these clone attempts with
1299 * -EINVAL in this case.
1302 xfs_reflink_remap_prep(
1303 struct file *file_in,
1305 struct file *file_out,
1310 struct inode *inode_in = file_inode(file_in);
1311 struct xfs_inode *src = XFS_I(inode_in);
1312 struct inode *inode_out = file_inode(file_out);
1313 struct xfs_inode *dest = XFS_I(inode_out);
1314 bool same_inode = (inode_in == inode_out);
1315 u64 blkmask = i_blocksize(inode_in) - 1;
1318 /* Lock both files against IO */
1319 ret = xfs_iolock_two_inodes_and_break_layout(inode_in, inode_out);
1323 xfs_ilock(src, XFS_MMAPLOCK_EXCL);
1325 xfs_lock_two_inodes(src, XFS_MMAPLOCK_SHARED, dest,
1328 /* Check file eligibility and prepare for block sharing. */
1330 /* Don't reflink realtime inodes */
1331 if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
1334 /* Don't share DAX file data for now. */
1335 if (IS_DAX(inode_in) || IS_DAX(inode_out))
1338 ret = vfs_clone_file_prep_inodes(inode_in, pos_in, inode_out, pos_out,
1344 * If the dedupe data matches, chop off the partial EOF block
1345 * from the source file so we don't try to dedupe the partial
1350 } else if (*len & blkmask) {
1352 * The user is attempting to share a partial EOF block,
1353 * if it's inside the destination EOF then reject it.
1355 if (pos_out + *len < i_size_read(inode_out)) {
1361 /* Attach dquots to dest inode before changing block map */
1362 ret = xfs_qm_dqattach(dest);
1367 * Zero existing post-eof speculative preallocations in the destination
1370 ret = xfs_reflink_zero_posteof(dest, pos_out);
1374 /* Set flags and remap blocks. */
1375 ret = xfs_reflink_set_inode_flag(src, dest);
1380 * If pos_out > EOF, we may have dirtied blocks between EOF and
1381 * pos_out. In that case, we need to extend the flush and unmap to cover
1382 * from EOF to the end of the copy length.
1384 if (pos_out > XFS_ISIZE(dest)) {
1385 loff_t flen = *len + (pos_out - XFS_ISIZE(dest));
1386 ret = xfs_flush_unmap_range(dest, XFS_ISIZE(dest), flen);
1388 ret = xfs_flush_unmap_range(dest, pos_out, *len);
1393 /* If we're altering the file contents... */
1396 * ...update the timestamps (which will grab the ilock again
1397 * from xfs_fs_dirty_inode, so we have to call it before we
1400 if (!(file_out->f_mode & FMODE_NOCMTIME)) {
1401 ret = file_update_time(file_out);
1407 * ...clear the security bits if the process is not being run
1408 * by root. This keeps people from modifying setuid and setgid
1411 ret = file_remove_privs(file_out);
1418 xfs_reflink_remap_unlock(file_in, file_out);
1423 * Link a range of blocks from one file to another.
1426 xfs_reflink_remap_range(
1427 struct file *file_in,
1429 struct file *file_out,
1434 struct inode *inode_in = file_inode(file_in);
1435 struct xfs_inode *src = XFS_I(inode_in);
1436 struct inode *inode_out = file_inode(file_out);
1437 struct xfs_inode *dest = XFS_I(inode_out);
1438 struct xfs_mount *mp = src->i_mount;
1439 xfs_fileoff_t sfsbno, dfsbno;
1440 xfs_filblks_t fsblen;
1441 xfs_extlen_t cowextsize;
1444 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1447 if (XFS_FORCED_SHUTDOWN(mp))
1450 /* Prepare and then clone file data. */
1451 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1456 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1458 dfsbno = XFS_B_TO_FSBT(mp, pos_out);
1459 sfsbno = XFS_B_TO_FSBT(mp, pos_in);
1460 fsblen = XFS_B_TO_FSB(mp, len);
1461 ret = xfs_reflink_remap_blocks(src, sfsbno, dest, dfsbno, fsblen,
1467 * Carry the cowextsize hint from src to dest if we're sharing the
1468 * entire source file to the entire destination file, the source file
1469 * has a cowextsize hint, and the destination file does not.
1472 if (pos_in == 0 && len == i_size_read(inode_in) &&
1473 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1474 pos_out == 0 && len >= i_size_read(inode_out) &&
1475 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1476 cowextsize = src->i_d.di_cowextsize;
1478 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1482 xfs_reflink_remap_unlock(file_in, file_out);
1484 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1489 * The user wants to preemptively CoW all shared blocks in this file,
1490 * which enables us to turn off the reflink flag. Iterate all
1491 * extents which are not prealloc/delalloc to see which ranges are
1492 * mentioned in the refcount tree, then read those blocks into the
1493 * pagecache, dirty them, fsync them back out, and then we can update
1494 * the inode flag. What happens if we run out of memory? :)
1497 xfs_reflink_dirty_extents(
1498 struct xfs_inode *ip,
1503 struct xfs_mount *mp = ip->i_mount;
1504 xfs_agnumber_t agno;
1505 xfs_agblock_t agbno;
1511 struct xfs_bmbt_irec map[2];
1515 while (end - fbno > 0) {
1518 * Look for extents in the file. Skip holes, delalloc, or
1519 * unwritten extents; they can't be reflinked.
1521 error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0);
1526 if (!xfs_bmap_is_real_extent(&map[0]))
1530 while (map[1].br_blockcount) {
1531 agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock);
1532 agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock);
1533 aglen = map[1].br_blockcount;
1535 error = xfs_reflink_find_shared(mp, NULL, agno, agbno,
1536 aglen, &rbno, &rlen, true);
1539 if (rbno == NULLAGBLOCK)
1542 /* Dirty the pages */
1543 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1544 fpos = XFS_FSB_TO_B(mp, map[1].br_startoff +
1546 flen = XFS_FSB_TO_B(mp, rlen);
1547 if (fpos + flen > isize)
1548 flen = isize - fpos;
1549 error = iomap_file_dirty(VFS_I(ip), fpos, flen,
1551 xfs_ilock(ip, XFS_ILOCK_EXCL);
1555 map[1].br_blockcount -= (rbno - agbno + rlen);
1556 map[1].br_startoff += (rbno - agbno + rlen);
1557 map[1].br_startblock += (rbno - agbno + rlen);
1561 fbno = map[0].br_startoff + map[0].br_blockcount;
1567 /* Does this inode need the reflink flag? */
1569 xfs_reflink_inode_has_shared_extents(
1570 struct xfs_trans *tp,
1571 struct xfs_inode *ip,
1574 struct xfs_bmbt_irec got;
1575 struct xfs_mount *mp = ip->i_mount;
1576 struct xfs_ifork *ifp;
1577 xfs_agnumber_t agno;
1578 xfs_agblock_t agbno;
1582 struct xfs_iext_cursor icur;
1586 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1587 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1588 error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1593 *has_shared = false;
1594 found = xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got);
1596 if (isnullstartblock(got.br_startblock) ||
1597 got.br_state != XFS_EXT_NORM)
1599 agno = XFS_FSB_TO_AGNO(mp, got.br_startblock);
1600 agbno = XFS_FSB_TO_AGBNO(mp, got.br_startblock);
1601 aglen = got.br_blockcount;
1603 error = xfs_reflink_find_shared(mp, tp, agno, agbno, aglen,
1604 &rbno, &rlen, false);
1607 /* Is there still a shared block here? */
1608 if (rbno != NULLAGBLOCK) {
1613 found = xfs_iext_next_extent(ifp, &icur, &got);
1620 * Clear the inode reflink flag if there are no shared extents.
1622 * The caller is responsible for joining the inode to the transaction passed in.
1623 * The inode will be joined to the transaction that is returned to the caller.
1626 xfs_reflink_clear_inode_flag(
1627 struct xfs_inode *ip,
1628 struct xfs_trans **tpp)
1633 ASSERT(xfs_is_reflink_inode(ip));
1635 error = xfs_reflink_inode_has_shared_extents(*tpp, ip, &needs_flag);
1636 if (error || needs_flag)
1640 * We didn't find any shared blocks so turn off the reflink flag.
1641 * First, get rid of any leftover CoW mappings.
1643 error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF, true);
1647 /* Clear the inode flag. */
1648 trace_xfs_reflink_unset_inode_flag(ip);
1649 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1650 xfs_inode_clear_cowblocks_tag(ip);
1651 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1657 * Clear the inode reflink flag if there are no shared extents and the size
1661 xfs_reflink_try_clear_inode_flag(
1662 struct xfs_inode *ip)
1664 struct xfs_mount *mp = ip->i_mount;
1665 struct xfs_trans *tp;
1668 /* Start a rolling transaction to remove the mappings */
1669 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1673 xfs_ilock(ip, XFS_ILOCK_EXCL);
1674 xfs_trans_ijoin(tp, ip, 0);
1676 error = xfs_reflink_clear_inode_flag(ip, &tp);
1680 error = xfs_trans_commit(tp);
1684 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1687 xfs_trans_cancel(tp);
1689 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1694 * Pre-COW all shared blocks within a given byte range of a file and turn off
1695 * the reflink flag if we unshare all of the file's blocks.
1698 xfs_reflink_unshare(
1699 struct xfs_inode *ip,
1703 struct xfs_mount *mp = ip->i_mount;
1709 if (!xfs_is_reflink_inode(ip))
1712 trace_xfs_reflink_unshare(ip, offset, len);
1714 inode_dio_wait(VFS_I(ip));
1716 /* Try to CoW the selected ranges */
1717 xfs_ilock(ip, XFS_ILOCK_EXCL);
1718 fbno = XFS_B_TO_FSBT(mp, offset);
1719 isize = i_size_read(VFS_I(ip));
1720 end = XFS_B_TO_FSB(mp, offset + len);
1721 error = xfs_reflink_dirty_extents(ip, fbno, end, isize);
1724 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1726 /* Wait for the IO to finish */
1727 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1731 /* Turn off the reflink flag if possible. */
1732 error = xfs_reflink_try_clear_inode_flag(ip);
1739 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1741 trace_xfs_reflink_unshare_error(ip, error, _RET_IP_);