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_reflink.h"
53 #include "xfs_iomap.h"
54 #include "xfs_rmap_btree.h"
56 #include "xfs_ag_resv.h"
59 * Copy on Write of Shared Blocks
61 * XFS must preserve "the usual" file semantics even when two files share
62 * the same physical blocks. This means that a write to one file must not
63 * alter the blocks in a different file; the way that we'll do that is
64 * through the use of a copy-on-write mechanism. At a high level, that
65 * means that when we want to write to a shared block, we allocate a new
66 * block, write the data to the new block, and if that succeeds we map the
67 * new block into the file.
69 * XFS provides a "delayed allocation" mechanism that defers the allocation
70 * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
71 * possible. This reduces fragmentation by enabling the filesystem to ask
72 * for bigger chunks less often, which is exactly what we want for CoW.
74 * The delalloc mechanism begins when the kernel wants to make a block
75 * writable (write_begin or page_mkwrite). If the offset is not mapped, we
76 * create a delalloc mapping, which is a regular in-core extent, but without
77 * a real startblock. (For delalloc mappings, the startblock encodes both
78 * a flag that this is a delalloc mapping, and a worst-case estimate of how
79 * many blocks might be required to put the mapping into the BMBT.) delalloc
80 * mappings are a reservation against the free space in the filesystem;
81 * adjacent mappings can also be combined into fewer larger mappings.
83 * As an optimization, the CoW extent size hint (cowextsz) creates
84 * outsized aligned delalloc reservations in the hope of landing out of
85 * order nearby CoW writes in a single extent on disk, thereby reducing
86 * fragmentation and improving future performance.
88 * D: --RRRRRRSSSRRRRRRRR--- (data fork)
89 * C: ------DDDDDDD--------- (CoW fork)
91 * When dirty pages are being written out (typically in writepage), the
92 * delalloc reservations are converted into unwritten mappings by
93 * allocating blocks and replacing the delalloc mapping with real ones.
94 * A delalloc mapping can be replaced by several unwritten ones if the
95 * free space is fragmented.
97 * D: --RRRRRRSSSRRRRRRRR---
98 * C: ------UUUUUUU---------
100 * We want to adapt the delalloc mechanism for copy-on-write, since the
101 * write paths are similar. The first two steps (creating the reservation
102 * and allocating the blocks) are exactly the same as delalloc except that
103 * the mappings must be stored in a separate CoW fork because we do not want
104 * to disturb the mapping in the data fork until we're sure that the write
105 * succeeded. IO completion in this case is the process of removing the old
106 * mapping from the data fork and moving the new mapping from the CoW fork to
107 * the data fork. This will be discussed shortly.
109 * For now, unaligned directio writes will be bounced back to the page cache.
110 * Block-aligned directio writes will use the same mechanism as buffered
113 * Just prior to submitting the actual disk write requests, we convert
114 * the extents representing the range of the file actually being written
115 * (as opposed to extra pieces created for the cowextsize hint) to real
116 * extents. This will become important in the next step:
118 * D: --RRRRRRSSSRRRRRRRR---
119 * C: ------UUrrUUU---------
121 * CoW remapping must be done after the data block write completes,
122 * because we don't want to destroy the old data fork map until we're sure
123 * the new block has been written. Since the new mappings are kept in a
124 * separate fork, we can simply iterate these mappings to find the ones
125 * that cover the file blocks that we just CoW'd. For each extent, simply
126 * unmap the corresponding range in the data fork, map the new range into
127 * the data fork, and remove the extent from the CoW fork. Because of
128 * the presence of the cowextsize hint, however, we must be careful
129 * only to remap the blocks that we've actually written out -- we must
130 * never remap delalloc reservations nor CoW staging blocks that have
131 * yet to be written. This corresponds exactly to the real extents in
134 * D: --RRRRRRrrSRRRRRRRR---
135 * C: ------UU--UUU---------
137 * Since the remapping operation can be applied to an arbitrary file
138 * range, we record the need for the remap step as a flag in the ioend
139 * instead of declaring a new IO type. This is required for direct io
140 * because we only have ioend for the whole dio, and we have to be able to
141 * remember the presence of unwritten blocks and CoW blocks with a single
142 * ioend structure. Better yet, the more ground we can cover with one
147 * Given an AG extent, find the lowest-numbered run of shared blocks
148 * within that range and return the range in fbno/flen. If
149 * find_end_of_shared is true, return the longest contiguous extent of
150 * shared blocks. If there are no shared extents, fbno and flen will
151 * be set to NULLAGBLOCK and 0, respectively.
154 xfs_reflink_find_shared(
155 struct xfs_mount
*mp
,
156 struct xfs_trans
*tp
,
162 bool find_end_of_shared
)
164 struct xfs_buf
*agbp
;
165 struct xfs_btree_cur
*cur
;
168 error
= xfs_alloc_read_agf(mp
, tp
, agno
, 0, &agbp
);
174 cur
= xfs_refcountbt_init_cursor(mp
, tp
, agbp
, agno
, NULL
);
176 error
= xfs_refcount_find_shared(cur
, agbno
, aglen
, fbno
, flen
,
179 xfs_btree_del_cursor(cur
, error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
181 xfs_trans_brelse(tp
, agbp
);
186 * Trim the mapping to the next block where there's a change in the
187 * shared/unshared status. More specifically, this means that we
188 * find the lowest-numbered extent of shared blocks that coincides with
189 * the given block mapping. If the shared extent overlaps the start of
190 * the mapping, trim the mapping to the end of the shared extent. If
191 * the shared region intersects the mapping, trim the mapping to the
192 * start of the shared extent. If there are no shared regions that
193 * overlap, just return the original extent.
196 xfs_reflink_trim_around_shared(
197 struct xfs_inode
*ip
,
198 struct xfs_bmbt_irec
*irec
,
209 /* Holes, unwritten, and delalloc extents cannot be shared */
210 if (!xfs_is_reflink_inode(ip
) || !xfs_bmap_is_real_extent(irec
)) {
215 trace_xfs_reflink_trim_around_shared(ip
, irec
);
217 agno
= XFS_FSB_TO_AGNO(ip
->i_mount
, irec
->br_startblock
);
218 agbno
= XFS_FSB_TO_AGBNO(ip
->i_mount
, irec
->br_startblock
);
219 aglen
= irec
->br_blockcount
;
221 error
= xfs_reflink_find_shared(ip
->i_mount
, NULL
, agno
, agbno
,
222 aglen
, &fbno
, &flen
, true);
226 *shared
= *trimmed
= false;
227 if (fbno
== NULLAGBLOCK
) {
228 /* No shared blocks at all. */
230 } else if (fbno
== agbno
) {
232 * The start of this extent is shared. Truncate the
233 * mapping at the end of the shared region so that a
234 * subsequent iteration starts at the start of the
237 irec
->br_blockcount
= flen
;
244 * There's a shared extent midway through this extent.
245 * Truncate the mapping at the start of the shared
246 * extent so that a subsequent iteration starts at the
247 * start of the shared region.
249 irec
->br_blockcount
= fbno
- agbno
;
256 * Trim the passed in imap to the next shared/unshared extent boundary, and
257 * if imap->br_startoff points to a shared extent reserve space for it in the
258 * COW fork. In this case *shared is set to true, else to false.
260 * Note that imap will always contain the block numbers for the existing blocks
261 * in the data fork, as the upper layers need them for read-modify-write
265 xfs_reflink_reserve_cow(
266 struct xfs_inode
*ip
,
267 struct xfs_bmbt_irec
*imap
,
270 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
271 struct xfs_bmbt_irec got
;
273 bool eof
= false, trimmed
;
274 struct xfs_iext_cursor icur
;
277 * Search the COW fork extent list first. This serves two purposes:
278 * first this implement the speculative preallocation using cowextisze,
279 * so that we also unshared block adjacent to shared blocks instead
280 * of just the shared blocks themselves. Second the lookup in the
281 * extent list is generally faster than going out to the shared extent
285 if (!xfs_iext_lookup_extent(ip
, ifp
, imap
->br_startoff
, &icur
, &got
))
287 if (!eof
&& got
.br_startoff
<= imap
->br_startoff
) {
288 trace_xfs_reflink_cow_found(ip
, imap
);
289 xfs_trim_extent(imap
, got
.br_startoff
, got
.br_blockcount
);
295 /* Trim the mapping to the nearest shared extent boundary. */
296 error
= xfs_reflink_trim_around_shared(ip
, imap
, shared
, &trimmed
);
300 /* Not shared? Just report the (potentially capped) extent. */
305 * Fork all the shared blocks from our write offset until the end of
308 error
= xfs_qm_dqattach_locked(ip
, 0);
312 error
= xfs_bmapi_reserve_delalloc(ip
, XFS_COW_FORK
, imap
->br_startoff
,
313 imap
->br_blockcount
, 0, &got
, &icur
, eof
);
314 if (error
== -ENOSPC
|| error
== -EDQUOT
)
315 trace_xfs_reflink_cow_enospc(ip
, imap
);
319 xfs_trim_extent(imap
, got
.br_startoff
, got
.br_blockcount
);
320 trace_xfs_reflink_cow_alloc(ip
, &got
);
324 /* Convert part of an unwritten CoW extent to a real one. */
326 xfs_reflink_convert_cow_extent(
327 struct xfs_inode
*ip
,
328 struct xfs_bmbt_irec
*imap
,
329 xfs_fileoff_t offset_fsb
,
330 xfs_filblks_t count_fsb
,
331 struct xfs_defer_ops
*dfops
)
333 xfs_fsblock_t first_block
= NULLFSBLOCK
;
336 if (imap
->br_state
== XFS_EXT_NORM
)
339 xfs_trim_extent(imap
, offset_fsb
, count_fsb
);
340 trace_xfs_reflink_convert_cow(ip
, imap
);
341 if (imap
->br_blockcount
== 0)
343 return xfs_bmapi_write(NULL
, ip
, imap
->br_startoff
, imap
->br_blockcount
,
344 XFS_BMAPI_COWFORK
| XFS_BMAPI_CONVERT
, &first_block
,
345 0, imap
, &nimaps
, dfops
);
348 /* Convert all of the unwritten CoW extents in a file's range to real ones. */
350 xfs_reflink_convert_cow(
351 struct xfs_inode
*ip
,
355 struct xfs_mount
*mp
= ip
->i_mount
;
356 xfs_fileoff_t offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
357 xfs_fileoff_t end_fsb
= XFS_B_TO_FSB(mp
, offset
+ count
);
358 xfs_filblks_t count_fsb
= end_fsb
- offset_fsb
;
359 struct xfs_bmbt_irec imap
;
360 struct xfs_defer_ops dfops
;
361 xfs_fsblock_t first_block
= NULLFSBLOCK
;
362 int nimaps
= 1, error
= 0;
366 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
367 error
= xfs_bmapi_write(NULL
, ip
, offset_fsb
, count_fsb
,
368 XFS_BMAPI_COWFORK
| XFS_BMAPI_CONVERT
|
369 XFS_BMAPI_CONVERT_ONLY
, &first_block
, 0, &imap
, &nimaps
,
371 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
375 /* Allocate all CoW reservations covering a range of blocks in a file. */
377 xfs_reflink_allocate_cow(
378 struct xfs_inode
*ip
,
379 struct xfs_bmbt_irec
*imap
,
383 struct xfs_mount
*mp
= ip
->i_mount
;
384 xfs_fileoff_t offset_fsb
= imap
->br_startoff
;
385 xfs_filblks_t count_fsb
= imap
->br_blockcount
;
386 struct xfs_bmbt_irec got
;
387 struct xfs_defer_ops dfops
;
388 struct xfs_trans
*tp
= NULL
;
389 xfs_fsblock_t first_block
;
390 int nimaps
, error
= 0;
392 xfs_filblks_t resaligned
;
393 xfs_extlen_t resblks
= 0;
394 struct xfs_iext_cursor icur
;
397 ASSERT(xfs_is_reflink_inode(ip
));
398 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
| XFS_ILOCK_SHARED
));
401 * Even if the extent is not shared we might have a preallocation for
402 * it in the COW fork. If so use it.
404 if (xfs_iext_lookup_extent(ip
, ip
->i_cowfp
, offset_fsb
, &icur
, &got
) &&
405 got
.br_startoff
<= offset_fsb
) {
408 /* If we have a real allocation in the COW fork we're done. */
409 if (!isnullstartblock(got
.br_startblock
)) {
410 xfs_trim_extent(&got
, offset_fsb
, count_fsb
);
415 xfs_trim_extent(imap
, got
.br_startoff
, got
.br_blockcount
);
417 error
= xfs_reflink_trim_around_shared(ip
, imap
, shared
, &trimmed
);
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
, 0);
441 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, resblks
, 0,
442 XFS_QMOPT_RES_REGBLKS
);
446 xfs_trans_ijoin(tp
, ip
, 0);
448 xfs_defer_init(&dfops
, &first_block
);
451 /* Allocate the entire reservation as unwritten blocks. */
452 error
= xfs_bmapi_write(tp
, ip
, imap
->br_startoff
, imap
->br_blockcount
,
453 XFS_BMAPI_COWFORK
| XFS_BMAPI_PREALLOC
, &first_block
,
454 resblks
, imap
, &nimaps
, &dfops
);
456 goto out_bmap_cancel
;
458 xfs_inode_set_cowblocks_tag(ip
);
461 error
= xfs_defer_finish(&tp
, &dfops
);
463 goto out_bmap_cancel
;
465 error
= xfs_trans_commit(tp
);
469 return xfs_reflink_convert_cow_extent(ip
, imap
, offset_fsb
, count_fsb
,
472 xfs_defer_cancel(&dfops
);
473 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)resblks
, 0,
474 XFS_QMOPT_RES_REGBLKS
);
477 xfs_trans_cancel(tp
);
482 * Find the CoW reservation for a given byte offset of a file.
485 xfs_reflink_find_cow_mapping(
486 struct xfs_inode
*ip
,
488 struct xfs_bmbt_irec
*imap
)
490 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
491 xfs_fileoff_t offset_fsb
;
492 struct xfs_bmbt_irec got
;
493 struct xfs_iext_cursor icur
;
495 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
| XFS_ILOCK_SHARED
));
497 if (!xfs_is_reflink_inode(ip
))
499 offset_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
500 if (!xfs_iext_lookup_extent(ip
, ifp
, offset_fsb
, &icur
, &got
))
502 if (got
.br_startoff
> offset_fsb
)
505 trace_xfs_reflink_find_cow_mapping(ip
, offset
, 1, XFS_IO_OVERWRITE
,
512 * Trim an extent to end at the next CoW reservation past offset_fsb.
515 xfs_reflink_trim_irec_to_next_cow(
516 struct xfs_inode
*ip
,
517 xfs_fileoff_t offset_fsb
,
518 struct xfs_bmbt_irec
*imap
)
520 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
521 struct xfs_bmbt_irec got
;
522 struct xfs_iext_cursor icur
;
524 if (!xfs_is_reflink_inode(ip
))
527 /* Find the extent in the CoW fork. */
528 if (!xfs_iext_lookup_extent(ip
, ifp
, offset_fsb
, &icur
, &got
))
531 /* This is the extent before; try sliding up one. */
532 if (got
.br_startoff
< offset_fsb
) {
533 if (!xfs_iext_next_extent(ifp
, &icur
, &got
))
537 if (got
.br_startoff
>= imap
->br_startoff
+ imap
->br_blockcount
)
540 imap
->br_blockcount
= got
.br_startoff
- imap
->br_startoff
;
541 trace_xfs_reflink_trim_irec(ip
, imap
);
545 * Cancel CoW reservations for some block range of an inode.
547 * If cancel_real is true this function cancels all COW fork extents for the
548 * inode; if cancel_real is false, real extents are not cleared.
551 xfs_reflink_cancel_cow_blocks(
552 struct xfs_inode
*ip
,
553 struct xfs_trans
**tpp
,
554 xfs_fileoff_t offset_fsb
,
555 xfs_fileoff_t end_fsb
,
558 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
559 struct xfs_bmbt_irec got
, del
;
560 struct xfs_iext_cursor icur
;
561 xfs_fsblock_t firstfsb
;
562 struct xfs_defer_ops dfops
;
565 if (!xfs_is_reflink_inode(ip
))
567 if (!xfs_iext_lookup_extent_before(ip
, ifp
, &end_fsb
, &icur
, &got
))
570 /* Walk backwards until we're out of the I/O range... */
571 while (got
.br_startoff
+ got
.br_blockcount
> offset_fsb
) {
573 xfs_trim_extent(&del
, offset_fsb
, end_fsb
- offset_fsb
);
575 /* Extent delete may have bumped ext forward */
576 if (!del
.br_blockcount
) {
577 xfs_iext_prev(ifp
, &icur
);
581 trace_xfs_reflink_cancel_cow(ip
, &del
);
583 if (isnullstartblock(del
.br_startblock
)) {
584 error
= xfs_bmap_del_extent_delay(ip
, XFS_COW_FORK
,
588 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
|| cancel_real
) {
589 xfs_trans_ijoin(*tpp
, ip
, 0);
590 xfs_defer_init(&dfops
, &firstfsb
);
592 /* Free the CoW orphan record. */
593 error
= xfs_refcount_free_cow_extent(ip
->i_mount
,
594 &dfops
, del
.br_startblock
,
599 xfs_bmap_add_free(ip
->i_mount
, &dfops
,
600 del
.br_startblock
, del
.br_blockcount
,
603 /* Update quota accounting */
604 xfs_trans_mod_dquot_byino(*tpp
, ip
, XFS_TRANS_DQ_BCOUNT
,
605 -(long)del
.br_blockcount
);
607 /* Roll the transaction */
608 xfs_defer_ijoin(&dfops
, ip
);
609 error
= xfs_defer_finish(tpp
, &dfops
);
611 xfs_defer_cancel(&dfops
);
615 /* Remove the mapping from the CoW fork. */
616 xfs_bmap_del_extent_cow(ip
, &icur
, &got
, &del
);
618 /* Didn't do anything, push cursor back. */
619 xfs_iext_prev(ifp
, &icur
);
622 if (!xfs_iext_get_extent(ifp
, &icur
, &got
))
626 /* clear tag if cow fork is emptied */
628 xfs_inode_clear_cowblocks_tag(ip
);
634 * Cancel CoW reservations for some byte range of an inode.
636 * If cancel_real is true this function cancels all COW fork extents for the
637 * inode; if cancel_real is false, real extents are not cleared.
640 xfs_reflink_cancel_cow_range(
641 struct xfs_inode
*ip
,
646 struct xfs_trans
*tp
;
647 xfs_fileoff_t offset_fsb
;
648 xfs_fileoff_t end_fsb
;
651 trace_xfs_reflink_cancel_cow_range(ip
, offset
, count
);
652 ASSERT(xfs_is_reflink_inode(ip
));
654 offset_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
655 if (count
== NULLFILEOFF
)
656 end_fsb
= NULLFILEOFF
;
658 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, offset
+ count
);
660 /* Start a rolling transaction to remove the mappings */
661 error
= xfs_trans_alloc(ip
->i_mount
, &M_RES(ip
->i_mount
)->tr_write
,
666 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
667 xfs_trans_ijoin(tp
, ip
, 0);
669 /* Scrape out the old CoW reservations */
670 error
= xfs_reflink_cancel_cow_blocks(ip
, &tp
, offset_fsb
, end_fsb
,
675 error
= xfs_trans_commit(tp
);
677 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
681 xfs_trans_cancel(tp
);
682 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
684 trace_xfs_reflink_cancel_cow_range_error(ip
, error
, _RET_IP_
);
689 * Remap parts of a file's data fork after a successful CoW.
693 struct xfs_inode
*ip
,
697 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
698 struct xfs_bmbt_irec got
, del
;
699 struct xfs_trans
*tp
;
700 xfs_fileoff_t offset_fsb
;
701 xfs_fileoff_t end_fsb
;
702 xfs_fsblock_t firstfsb
;
703 struct xfs_defer_ops dfops
;
705 unsigned int resblks
;
707 struct xfs_iext_cursor icur
;
709 trace_xfs_reflink_end_cow(ip
, offset
, count
);
711 /* No COW extents? That's easy! */
712 if (ifp
->if_bytes
== 0)
715 offset_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
716 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, offset
+ count
);
719 * Start a rolling transaction to switch the mappings. We're
720 * unlikely ever to have to remap 16T worth of single-block
721 * extents, so just cap the worst case extent count to 2^32-1.
722 * Stick a warning in just in case, and avoid 64-bit division.
724 BUILD_BUG_ON(MAX_RW_COUNT
> UINT_MAX
);
725 if (end_fsb
- offset_fsb
> UINT_MAX
) {
726 error
= -EFSCORRUPTED
;
727 xfs_force_shutdown(ip
->i_mount
, SHUTDOWN_CORRUPT_INCORE
);
731 resblks
= XFS_NEXTENTADD_SPACE_RES(ip
->i_mount
,
732 (unsigned int)(end_fsb
- offset_fsb
),
734 error
= xfs_trans_alloc(ip
->i_mount
, &M_RES(ip
->i_mount
)->tr_write
,
735 resblks
, 0, XFS_TRANS_RESERVE
, &tp
);
739 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
740 xfs_trans_ijoin(tp
, ip
, 0);
743 * In case of racing, overlapping AIO writes no COW extents might be
744 * left by the time I/O completes for the loser of the race. In that
747 if (!xfs_iext_lookup_extent_before(ip
, ifp
, &end_fsb
, &icur
, &got
))
750 /* Walk backwards until we're out of the I/O range... */
751 while (got
.br_startoff
+ got
.br_blockcount
> offset_fsb
) {
753 xfs_trim_extent(&del
, offset_fsb
, end_fsb
- offset_fsb
);
755 /* Extent delete may have bumped ext forward */
756 if (!del
.br_blockcount
) {
757 xfs_iext_prev(ifp
, &icur
);
761 ASSERT(!isnullstartblock(got
.br_startblock
));
764 * Don't remap unwritten extents; these are
765 * speculatively preallocated CoW extents that have been
766 * allocated but have not yet been involved in a write.
768 if (got
.br_state
== XFS_EXT_UNWRITTEN
) {
769 xfs_iext_prev(ifp
, &icur
);
773 /* Unmap the old blocks in the data fork. */
774 xfs_defer_init(&dfops
, &firstfsb
);
775 rlen
= del
.br_blockcount
;
776 error
= __xfs_bunmapi(tp
, ip
, del
.br_startoff
, &rlen
, 0, 1,
781 /* Trim the extent to whatever got unmapped. */
783 xfs_trim_extent(&del
, del
.br_startoff
+ rlen
,
784 del
.br_blockcount
- rlen
);
786 trace_xfs_reflink_cow_remap(ip
, &del
);
788 /* Free the CoW orphan record. */
789 error
= xfs_refcount_free_cow_extent(tp
->t_mountp
, &dfops
,
790 del
.br_startblock
, del
.br_blockcount
);
794 /* Map the new blocks into the data fork. */
795 error
= xfs_bmap_map_extent(tp
->t_mountp
, &dfops
, ip
, &del
);
799 /* Remove the mapping from the CoW fork. */
800 xfs_bmap_del_extent_cow(ip
, &icur
, &got
, &del
);
802 xfs_defer_ijoin(&dfops
, ip
);
803 error
= xfs_defer_finish(&tp
, &dfops
);
807 if (!xfs_iext_get_extent(ifp
, &icur
, &got
))
811 error
= xfs_trans_commit(tp
);
812 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
818 xfs_defer_cancel(&dfops
);
820 xfs_trans_cancel(tp
);
821 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
823 trace_xfs_reflink_end_cow_error(ip
, error
, _RET_IP_
);
828 * Free leftover CoW reservations that didn't get cleaned out.
831 xfs_reflink_recover_cow(
832 struct xfs_mount
*mp
)
837 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
840 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
841 error
= xfs_refcount_recover_cow_leftovers(mp
, agno
);
850 * Reflinking (Block) Ranges of Two Files Together
852 * First, ensure that the reflink flag is set on both inodes. The flag is an
853 * optimization to avoid unnecessary refcount btree lookups in the write path.
855 * Now we can iteratively remap the range of extents (and holes) in src to the
856 * corresponding ranges in dest. Let drange and srange denote the ranges of
857 * logical blocks in dest and src touched by the reflink operation.
859 * While the length of drange is greater than zero,
860 * - Read src's bmbt at the start of srange ("imap")
861 * - If imap doesn't exist, make imap appear to start at the end of srange
863 * - If imap starts before srange, advance imap to start at srange.
864 * - If imap goes beyond srange, truncate imap to end at the end of srange.
865 * - Punch (imap start - srange start + imap len) blocks from dest at
866 * offset (drange start).
867 * - If imap points to a real range of pblks,
868 * > Increase the refcount of the imap's pblks
869 * > Map imap's pblks into dest at the offset
870 * (drange start + imap start - srange start)
871 * - Advance drange and srange by (imap start - srange start + imap len)
873 * Finally, if the reflink made dest longer, update both the in-core and
874 * on-disk file sizes.
876 * ASCII Art Demonstration:
878 * Let's say we want to reflink this source file:
880 * ----SSSSSSS-SSSSS----SSSSSS (src file)
881 * <-------------------->
883 * into this destination file:
885 * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
886 * <-------------------->
887 * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
888 * Observe that the range has different logical offsets in either file.
890 * Consider that the first extent in the source file doesn't line up with our
891 * reflink range. Unmapping and remapping are separate operations, so we can
892 * unmap more blocks from the destination file than we remap.
894 * ----SSSSSSS-SSSSS----SSSSSS
896 * --DDDDD---------DDDDD--DDD
899 * Now remap the source extent into the destination file:
901 * ----SSSSSSS-SSSSS----SSSSSS
903 * --DDDDD--SSSSSSSDDDDD--DDD
906 * Do likewise with the second hole and extent in our range. Holes in the
907 * unmap range don't affect our operation.
909 * ----SSSSSSS-SSSSS----SSSSSS
911 * --DDDDD--SSSSSSS-SSSSS-DDD
914 * Finally, unmap and remap part of the third extent. This will increase the
915 * size of the destination file.
917 * ----SSSSSSS-SSSSS----SSSSSS
919 * --DDDDD--SSSSSSS-SSSSS----SSS
922 * Once we update the destination file's i_size, we're done.
926 * Ensure the reflink bit is set in both inodes.
929 xfs_reflink_set_inode_flag(
930 struct xfs_inode
*src
,
931 struct xfs_inode
*dest
)
933 struct xfs_mount
*mp
= src
->i_mount
;
935 struct xfs_trans
*tp
;
937 if (xfs_is_reflink_inode(src
) && xfs_is_reflink_inode(dest
))
940 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ichange
, 0, 0, 0, &tp
);
944 /* Lock both files against IO */
945 if (src
->i_ino
== dest
->i_ino
)
946 xfs_ilock(src
, XFS_ILOCK_EXCL
);
948 xfs_lock_two_inodes(src
, dest
, XFS_ILOCK_EXCL
);
950 if (!xfs_is_reflink_inode(src
)) {
951 trace_xfs_reflink_set_inode_flag(src
);
952 xfs_trans_ijoin(tp
, src
, XFS_ILOCK_EXCL
);
953 src
->i_d
.di_flags2
|= XFS_DIFLAG2_REFLINK
;
954 xfs_trans_log_inode(tp
, src
, XFS_ILOG_CORE
);
955 xfs_ifork_init_cow(src
);
957 xfs_iunlock(src
, XFS_ILOCK_EXCL
);
959 if (src
->i_ino
== dest
->i_ino
)
962 if (!xfs_is_reflink_inode(dest
)) {
963 trace_xfs_reflink_set_inode_flag(dest
);
964 xfs_trans_ijoin(tp
, dest
, XFS_ILOCK_EXCL
);
965 dest
->i_d
.di_flags2
|= XFS_DIFLAG2_REFLINK
;
966 xfs_trans_log_inode(tp
, dest
, XFS_ILOG_CORE
);
967 xfs_ifork_init_cow(dest
);
969 xfs_iunlock(dest
, XFS_ILOCK_EXCL
);
972 error
= xfs_trans_commit(tp
);
978 trace_xfs_reflink_set_inode_flag_error(dest
, error
, _RET_IP_
);
983 * Update destination inode size & cowextsize hint, if necessary.
986 xfs_reflink_update_dest(
987 struct xfs_inode
*dest
,
989 xfs_extlen_t cowextsize
,
992 struct xfs_mount
*mp
= dest
->i_mount
;
993 struct xfs_trans
*tp
;
996 if (is_dedupe
&& newlen
<= i_size_read(VFS_I(dest
)) && cowextsize
== 0)
999 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ichange
, 0, 0, 0, &tp
);
1003 xfs_ilock(dest
, XFS_ILOCK_EXCL
);
1004 xfs_trans_ijoin(tp
, dest
, XFS_ILOCK_EXCL
);
1006 if (newlen
> i_size_read(VFS_I(dest
))) {
1007 trace_xfs_reflink_update_inode_size(dest
, newlen
);
1008 i_size_write(VFS_I(dest
), newlen
);
1009 dest
->i_d
.di_size
= newlen
;
1013 dest
->i_d
.di_cowextsize
= cowextsize
;
1014 dest
->i_d
.di_flags2
|= XFS_DIFLAG2_COWEXTSIZE
;
1018 xfs_trans_ichgtime(tp
, dest
,
1019 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1021 xfs_trans_log_inode(tp
, dest
, XFS_ILOG_CORE
);
1023 error
= xfs_trans_commit(tp
);
1029 trace_xfs_reflink_update_inode_size_error(dest
, error
, _RET_IP_
);
1034 * Do we have enough reserve in this AG to handle a reflink? The refcount
1035 * btree already reserved all the space it needs, but the rmap btree can grow
1036 * infinitely, so we won't allow more reflinks when the AG is down to the
1040 xfs_reflink_ag_has_free_space(
1041 struct xfs_mount
*mp
,
1042 xfs_agnumber_t agno
)
1044 struct xfs_perag
*pag
;
1047 if (!xfs_sb_version_hasrmapbt(&mp
->m_sb
))
1050 pag
= xfs_perag_get(mp
, agno
);
1051 if (xfs_ag_resv_critical(pag
, XFS_AG_RESV_AGFL
) ||
1052 xfs_ag_resv_critical(pag
, XFS_AG_RESV_METADATA
))
1059 * Unmap a range of blocks from a file, then map other blocks into the hole.
1060 * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount).
1061 * The extent irec is mapped into dest at irec->br_startoff.
1064 xfs_reflink_remap_extent(
1065 struct xfs_inode
*ip
,
1066 struct xfs_bmbt_irec
*irec
,
1067 xfs_fileoff_t destoff
,
1068 xfs_off_t new_isize
)
1070 struct xfs_mount
*mp
= ip
->i_mount
;
1071 bool real_extent
= xfs_bmap_is_real_extent(irec
);
1072 struct xfs_trans
*tp
;
1073 xfs_fsblock_t firstfsb
;
1074 unsigned int resblks
;
1075 struct xfs_defer_ops dfops
;
1076 struct xfs_bmbt_irec uirec
;
1078 xfs_filblks_t unmap_len
;
1082 unmap_len
= irec
->br_startoff
+ irec
->br_blockcount
- destoff
;
1083 trace_xfs_reflink_punch_range(ip
, destoff
, unmap_len
);
1085 /* No reflinking if we're low on space */
1087 error
= xfs_reflink_ag_has_free_space(mp
,
1088 XFS_FSB_TO_AGNO(mp
, irec
->br_startblock
));
1093 /* Start a rolling transaction to switch the mappings */
1094 resblks
= XFS_EXTENTADD_SPACE_RES(ip
->i_mount
, XFS_DATA_FORK
);
1095 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, resblks
, 0, 0, &tp
);
1099 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1100 xfs_trans_ijoin(tp
, ip
, 0);
1102 /* If we're not just clearing space, then do we have enough quota? */
1104 error
= xfs_trans_reserve_quota_nblks(tp
, ip
,
1105 irec
->br_blockcount
, 0, XFS_QMOPT_RES_REGBLKS
);
1110 trace_xfs_reflink_remap(ip
, irec
->br_startoff
,
1111 irec
->br_blockcount
, irec
->br_startblock
);
1113 /* Unmap the old blocks in the data fork. */
1116 xfs_defer_init(&dfops
, &firstfsb
);
1117 error
= __xfs_bunmapi(tp
, ip
, destoff
, &rlen
, 0, 1,
1123 * Trim the extent to whatever got unmapped.
1124 * Remember, bunmapi works backwards.
1126 uirec
.br_startblock
= irec
->br_startblock
+ rlen
;
1127 uirec
.br_startoff
= irec
->br_startoff
+ rlen
;
1128 uirec
.br_blockcount
= unmap_len
- rlen
;
1131 /* If this isn't a real mapping, we're done. */
1132 if (!real_extent
|| uirec
.br_blockcount
== 0)
1135 trace_xfs_reflink_remap(ip
, uirec
.br_startoff
,
1136 uirec
.br_blockcount
, uirec
.br_startblock
);
1138 /* Update the refcount tree */
1139 error
= xfs_refcount_increase_extent(mp
, &dfops
, &uirec
);
1143 /* Map the new blocks into the data fork. */
1144 error
= xfs_bmap_map_extent(mp
, &dfops
, ip
, &uirec
);
1148 /* Update quota accounting. */
1149 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
,
1150 uirec
.br_blockcount
);
1152 /* Update dest isize if needed. */
1153 newlen
= XFS_FSB_TO_B(mp
,
1154 uirec
.br_startoff
+ uirec
.br_blockcount
);
1155 newlen
= min_t(xfs_off_t
, newlen
, new_isize
);
1156 if (newlen
> i_size_read(VFS_I(ip
))) {
1157 trace_xfs_reflink_update_inode_size(ip
, newlen
);
1158 i_size_write(VFS_I(ip
), newlen
);
1159 ip
->i_d
.di_size
= newlen
;
1160 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1164 /* Process all the deferred stuff. */
1165 xfs_defer_ijoin(&dfops
, ip
);
1166 error
= xfs_defer_finish(&tp
, &dfops
);
1171 error
= xfs_trans_commit(tp
);
1172 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1178 xfs_defer_cancel(&dfops
);
1180 xfs_trans_cancel(tp
);
1181 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1183 trace_xfs_reflink_remap_extent_error(ip
, error
, _RET_IP_
);
1188 * Iteratively remap one file's extents (and holes) to another's.
1191 xfs_reflink_remap_blocks(
1192 struct xfs_inode
*src
,
1193 xfs_fileoff_t srcoff
,
1194 struct xfs_inode
*dest
,
1195 xfs_fileoff_t destoff
,
1197 xfs_off_t new_isize
)
1199 struct xfs_bmbt_irec imap
;
1202 xfs_filblks_t range_len
;
1204 /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
1206 trace_xfs_reflink_remap_blocks_loop(src
, srcoff
, len
,
1208 /* Read extent from the source file */
1210 xfs_ilock(src
, XFS_ILOCK_EXCL
);
1211 error
= xfs_bmapi_read(src
, srcoff
, len
, &imap
, &nimaps
, 0);
1212 xfs_iunlock(src
, XFS_ILOCK_EXCL
);
1215 ASSERT(nimaps
== 1);
1217 trace_xfs_reflink_remap_imap(src
, srcoff
, len
, XFS_IO_OVERWRITE
,
1220 /* Translate imap into the destination file. */
1221 range_len
= imap
.br_startoff
+ imap
.br_blockcount
- srcoff
;
1222 imap
.br_startoff
+= destoff
- srcoff
;
1224 /* Clear dest from destoff to the end of imap and map it in. */
1225 error
= xfs_reflink_remap_extent(dest
, &imap
, destoff
,
1230 if (fatal_signal_pending(current
)) {
1235 /* Advance drange/srange */
1236 srcoff
+= range_len
;
1237 destoff
+= range_len
;
1244 trace_xfs_reflink_remap_blocks_error(dest
, error
, _RET_IP_
);
1249 * Link a range of blocks from one file to another.
1252 xfs_reflink_remap_range(
1253 struct file
*file_in
,
1255 struct file
*file_out
,
1260 struct inode
*inode_in
= file_inode(file_in
);
1261 struct xfs_inode
*src
= XFS_I(inode_in
);
1262 struct inode
*inode_out
= file_inode(file_out
);
1263 struct xfs_inode
*dest
= XFS_I(inode_out
);
1264 struct xfs_mount
*mp
= src
->i_mount
;
1265 bool same_inode
= (inode_in
== inode_out
);
1266 xfs_fileoff_t sfsbno
, dfsbno
;
1267 xfs_filblks_t fsblen
;
1268 xfs_extlen_t cowextsize
;
1271 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
1274 if (XFS_FORCED_SHUTDOWN(mp
))
1277 /* Lock both files against IO */
1278 lock_two_nondirectories(inode_in
, inode_out
);
1280 xfs_ilock(src
, XFS_MMAPLOCK_EXCL
);
1282 xfs_lock_two_inodes(src
, dest
, XFS_MMAPLOCK_EXCL
);
1284 /* Check file eligibility and prepare for block sharing. */
1286 /* Don't reflink realtime inodes */
1287 if (XFS_IS_REALTIME_INODE(src
) || XFS_IS_REALTIME_INODE(dest
))
1290 /* Don't share DAX file data for now. */
1291 if (IS_DAX(inode_in
) || IS_DAX(inode_out
))
1294 ret
= vfs_clone_file_prep_inodes(inode_in
, pos_in
, inode_out
, pos_out
,
1299 trace_xfs_reflink_remap_range(src
, pos_in
, len
, dest
, pos_out
);
1302 * Clear out post-eof preallocations because we don't have page cache
1303 * backing the delayed allocations and they'll never get freed on
1306 if (xfs_can_free_eofblocks(dest
, true)) {
1307 ret
= xfs_free_eofblocks(dest
);
1312 /* Set flags and remap blocks. */
1313 ret
= xfs_reflink_set_inode_flag(src
, dest
);
1317 dfsbno
= XFS_B_TO_FSBT(mp
, pos_out
);
1318 sfsbno
= XFS_B_TO_FSBT(mp
, pos_in
);
1319 fsblen
= XFS_B_TO_FSB(mp
, len
);
1320 ret
= xfs_reflink_remap_blocks(src
, sfsbno
, dest
, dfsbno
, fsblen
,
1326 * If pos_out > EOF, we may have dirtied blocks between EOF and
1327 * pos_out. In that case, we need to extend the flush and unmap to cover
1328 * from EOF to the end of the copy length.
1330 if (pos_out
> XFS_ISIZE(dest
)) {
1331 loff_t flen
= len
+ (pos_out
- XFS_ISIZE(dest
));
1332 ret
= xfs_flush_unmap_range(dest
, XFS_ISIZE(dest
), flen
);
1334 ret
= xfs_flush_unmap_range(dest
, pos_out
, len
);
1340 * Carry the cowextsize hint from src to dest if we're sharing the
1341 * entire source file to the entire destination file, the source file
1342 * has a cowextsize hint, and the destination file does not.
1345 if (pos_in
== 0 && len
== i_size_read(inode_in
) &&
1346 (src
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
) &&
1347 pos_out
== 0 && len
>= i_size_read(inode_out
) &&
1348 !(dest
->i_d
.di_flags2
& XFS_DIFLAG2_COWEXTSIZE
))
1349 cowextsize
= src
->i_d
.di_cowextsize
;
1351 ret
= xfs_reflink_update_dest(dest
, pos_out
+ len
, cowextsize
,
1355 xfs_iunlock(src
, XFS_MMAPLOCK_EXCL
);
1357 xfs_iunlock(dest
, XFS_MMAPLOCK_EXCL
);
1358 unlock_two_nondirectories(inode_in
, inode_out
);
1360 trace_xfs_reflink_remap_range_error(dest
, ret
, _RET_IP_
);
1365 * The user wants to preemptively CoW all shared blocks in this file,
1366 * which enables us to turn off the reflink flag. Iterate all
1367 * extents which are not prealloc/delalloc to see which ranges are
1368 * mentioned in the refcount tree, then read those blocks into the
1369 * pagecache, dirty them, fsync them back out, and then we can update
1370 * the inode flag. What happens if we run out of memory? :)
1373 xfs_reflink_dirty_extents(
1374 struct xfs_inode
*ip
,
1379 struct xfs_mount
*mp
= ip
->i_mount
;
1380 xfs_agnumber_t agno
;
1381 xfs_agblock_t agbno
;
1387 struct xfs_bmbt_irec map
[2];
1391 while (end
- fbno
> 0) {
1394 * Look for extents in the file. Skip holes, delalloc, or
1395 * unwritten extents; they can't be reflinked.
1397 error
= xfs_bmapi_read(ip
, fbno
, end
- fbno
, map
, &nmaps
, 0);
1402 if (!xfs_bmap_is_real_extent(&map
[0]))
1406 while (map
[1].br_blockcount
) {
1407 agno
= XFS_FSB_TO_AGNO(mp
, map
[1].br_startblock
);
1408 agbno
= XFS_FSB_TO_AGBNO(mp
, map
[1].br_startblock
);
1409 aglen
= map
[1].br_blockcount
;
1411 error
= xfs_reflink_find_shared(mp
, NULL
, agno
, agbno
,
1412 aglen
, &rbno
, &rlen
, true);
1415 if (rbno
== NULLAGBLOCK
)
1418 /* Dirty the pages */
1419 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1420 fpos
= XFS_FSB_TO_B(mp
, map
[1].br_startoff
+
1422 flen
= XFS_FSB_TO_B(mp
, rlen
);
1423 if (fpos
+ flen
> isize
)
1424 flen
= isize
- fpos
;
1425 error
= iomap_file_dirty(VFS_I(ip
), fpos
, flen
,
1427 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1431 map
[1].br_blockcount
-= (rbno
- agbno
+ rlen
);
1432 map
[1].br_startoff
+= (rbno
- agbno
+ rlen
);
1433 map
[1].br_startblock
+= (rbno
- agbno
+ rlen
);
1437 fbno
= map
[0].br_startoff
+ map
[0].br_blockcount
;
1443 /* Does this inode need the reflink flag? */
1445 xfs_reflink_inode_has_shared_extents(
1446 struct xfs_trans
*tp
,
1447 struct xfs_inode
*ip
,
1450 struct xfs_bmbt_irec got
;
1451 struct xfs_mount
*mp
= ip
->i_mount
;
1452 struct xfs_ifork
*ifp
;
1453 xfs_agnumber_t agno
;
1454 xfs_agblock_t agbno
;
1458 struct xfs_iext_cursor icur
;
1462 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1463 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1464 error
= xfs_iread_extents(tp
, ip
, XFS_DATA_FORK
);
1469 *has_shared
= false;
1470 found
= xfs_iext_lookup_extent(ip
, ifp
, 0, &icur
, &got
);
1472 if (isnullstartblock(got
.br_startblock
) ||
1473 got
.br_state
!= XFS_EXT_NORM
)
1475 agno
= XFS_FSB_TO_AGNO(mp
, got
.br_startblock
);
1476 agbno
= XFS_FSB_TO_AGBNO(mp
, got
.br_startblock
);
1477 aglen
= got
.br_blockcount
;
1479 error
= xfs_reflink_find_shared(mp
, tp
, agno
, agbno
, aglen
,
1480 &rbno
, &rlen
, false);
1483 /* Is there still a shared block here? */
1484 if (rbno
!= NULLAGBLOCK
) {
1489 found
= xfs_iext_next_extent(ifp
, &icur
, &got
);
1495 /* Clear the inode reflink flag if there are no shared extents. */
1497 xfs_reflink_clear_inode_flag(
1498 struct xfs_inode
*ip
,
1499 struct xfs_trans
**tpp
)
1504 ASSERT(xfs_is_reflink_inode(ip
));
1506 error
= xfs_reflink_inode_has_shared_extents(*tpp
, ip
, &needs_flag
);
1507 if (error
|| needs_flag
)
1511 * We didn't find any shared blocks so turn off the reflink flag.
1512 * First, get rid of any leftover CoW mappings.
1514 error
= xfs_reflink_cancel_cow_blocks(ip
, tpp
, 0, NULLFILEOFF
, true);
1518 /* Clear the inode flag. */
1519 trace_xfs_reflink_unset_inode_flag(ip
);
1520 ip
->i_d
.di_flags2
&= ~XFS_DIFLAG2_REFLINK
;
1521 xfs_inode_clear_cowblocks_tag(ip
);
1522 xfs_trans_ijoin(*tpp
, ip
, 0);
1523 xfs_trans_log_inode(*tpp
, ip
, XFS_ILOG_CORE
);
1529 * Clear the inode reflink flag if there are no shared extents and the size
1533 xfs_reflink_try_clear_inode_flag(
1534 struct xfs_inode
*ip
)
1536 struct xfs_mount
*mp
= ip
->i_mount
;
1537 struct xfs_trans
*tp
;
1540 /* Start a rolling transaction to remove the mappings */
1541 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, 0, 0, 0, &tp
);
1545 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1546 xfs_trans_ijoin(tp
, ip
, 0);
1548 error
= xfs_reflink_clear_inode_flag(ip
, &tp
);
1552 error
= xfs_trans_commit(tp
);
1556 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1559 xfs_trans_cancel(tp
);
1561 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1566 * Pre-COW all shared blocks within a given byte range of a file and turn off
1567 * the reflink flag if we unshare all of the file's blocks.
1570 xfs_reflink_unshare(
1571 struct xfs_inode
*ip
,
1575 struct xfs_mount
*mp
= ip
->i_mount
;
1581 if (!xfs_is_reflink_inode(ip
))
1584 trace_xfs_reflink_unshare(ip
, offset
, len
);
1586 inode_dio_wait(VFS_I(ip
));
1588 /* Try to CoW the selected ranges */
1589 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1590 fbno
= XFS_B_TO_FSBT(mp
, offset
);
1591 isize
= i_size_read(VFS_I(ip
));
1592 end
= XFS_B_TO_FSB(mp
, offset
+ len
);
1593 error
= xfs_reflink_dirty_extents(ip
, fbno
, end
, isize
);
1596 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1598 /* Wait for the IO to finish */
1599 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
1603 /* Turn off the reflink flag if possible. */
1604 error
= xfs_reflink_try_clear_inode_flag(ip
);
1611 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1613 trace_xfs_reflink_unshare_error(ip
, error
, _RET_IP_
);