Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright (C) 2016 Oracle. All Rights Reserved.
3 *
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
5 *
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.
10 *
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.
15 *
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.
19 */
20#include "xfs.h"
21#include "xfs_fs.h"
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"
33#include "xfs_bmap.h"
34#include "xfs_bmap_util.h"
35#include "xfs_error.h"
36#include "xfs_dir2.h"
37#include "xfs_dir2_priv.h"
38#include "xfs_ioctl.h"
39#include "xfs_trace.h"
40#include "xfs_log.h"
41#include "xfs_icache.h"
42#include "xfs_pnfs.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"
48#include "xfs_bit.h"
49#include "xfs_alloc.h"
50#include "xfs_quota_defs.h"
51#include "xfs_quota.h"
52#include "xfs_btree.h"
53#include "xfs_bmap_btree.h"
54#include "xfs_reflink.h"
55#include "xfs_iomap.h"
56#include "xfs_rmap_btree.h"
57#include "xfs_sb.h"
58#include "xfs_ag_resv.h"
59
60/*
61 * Copy on Write of Shared Blocks
62 *
63 * XFS must preserve "the usual" file semantics even when two files share
64 * the same physical blocks. This means that a write to one file must not
65 * alter the blocks in a different file; the way that we'll do that is
66 * through the use of a copy-on-write mechanism. At a high level, that
67 * means that when we want to write to a shared block, we allocate a new
68 * block, write the data to the new block, and if that succeeds we map the
69 * new block into the file.
70 *
71 * XFS provides a "delayed allocation" mechanism that defers the allocation
72 * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
73 * possible. This reduces fragmentation by enabling the filesystem to ask
74 * for bigger chunks less often, which is exactly what we want for CoW.
75 *
76 * The delalloc mechanism begins when the kernel wants to make a block
77 * writable (write_begin or page_mkwrite). If the offset is not mapped, we
78 * create a delalloc mapping, which is a regular in-core extent, but without
79 * a real startblock. (For delalloc mappings, the startblock encodes both
80 * a flag that this is a delalloc mapping, and a worst-case estimate of how
81 * many blocks might be required to put the mapping into the BMBT.) delalloc
82 * mappings are a reservation against the free space in the filesystem;
83 * adjacent mappings can also be combined into fewer larger mappings.
84 *
85 * As an optimization, the CoW extent size hint (cowextsz) creates
86 * outsized aligned delalloc reservations in the hope of landing out of
87 * order nearby CoW writes in a single extent on disk, thereby reducing
88 * fragmentation and improving future performance.
89 *
90 * D: --RRRRRRSSSRRRRRRRR--- (data fork)
91 * C: ------DDDDDDD--------- (CoW fork)
92 *
93 * When dirty pages are being written out (typically in writepage), the
94 * delalloc reservations are converted into unwritten mappings by
95 * allocating blocks and replacing the delalloc mapping with real ones.
96 * A delalloc mapping can be replaced by several unwritten ones if the
97 * free space is fragmented.
98 *
99 * D: --RRRRRRSSSRRRRRRRR---
100 * C: ------UUUUUUU---------
101 *
102 * We want to adapt the delalloc mechanism for copy-on-write, since the
103 * write paths are similar. The first two steps (creating the reservation
104 * and allocating the blocks) are exactly the same as delalloc except that
105 * the mappings must be stored in a separate CoW fork because we do not want
106 * to disturb the mapping in the data fork until we're sure that the write
107 * succeeded. IO completion in this case is the process of removing the old
108 * mapping from the data fork and moving the new mapping from the CoW fork to
109 * the data fork. This will be discussed shortly.
110 *
111 * For now, unaligned directio writes will be bounced back to the page cache.
112 * Block-aligned directio writes will use the same mechanism as buffered
113 * writes.
114 *
115 * Just prior to submitting the actual disk write requests, we convert
116 * the extents representing the range of the file actually being written
117 * (as opposed to extra pieces created for the cowextsize hint) to real
118 * extents. This will become important in the next step:
119 *
120 * D: --RRRRRRSSSRRRRRRRR---
121 * C: ------UUrrUUU---------
122 *
123 * CoW remapping must be done after the data block write completes,
124 * because we don't want to destroy the old data fork map until we're sure
125 * the new block has been written. Since the new mappings are kept in a
126 * separate fork, we can simply iterate these mappings to find the ones
127 * that cover the file blocks that we just CoW'd. For each extent, simply
128 * unmap the corresponding range in the data fork, map the new range into
129 * the data fork, and remove the extent from the CoW fork. Because of
130 * the presence of the cowextsize hint, however, we must be careful
131 * only to remap the blocks that we've actually written out -- we must
132 * never remap delalloc reservations nor CoW staging blocks that have
133 * yet to be written. This corresponds exactly to the real extents in
134 * the CoW fork:
135 *
136 * D: --RRRRRRrrSRRRRRRRR---
137 * C: ------UU--UUU---------
138 *
139 * Since the remapping operation can be applied to an arbitrary file
140 * range, we record the need for the remap step as a flag in the ioend
141 * instead of declaring a new IO type. This is required for direct io
142 * because we only have ioend for the whole dio, and we have to be able to
143 * remember the presence of unwritten blocks and CoW blocks with a single
144 * ioend structure. Better yet, the more ground we can cover with one
145 * ioend, the better.
146 */
147
148/*
149 * Given an AG extent, find the lowest-numbered run of shared blocks
150 * within that range and return the range in fbno/flen. If
151 * find_end_of_shared is true, return the longest contiguous extent of
152 * shared blocks. If there are no shared extents, fbno and flen will
153 * be set to NULLAGBLOCK and 0, respectively.
154 */
155int
156xfs_reflink_find_shared(
157 struct xfs_mount *mp,
158 xfs_agnumber_t agno,
159 xfs_agblock_t agbno,
160 xfs_extlen_t aglen,
161 xfs_agblock_t *fbno,
162 xfs_extlen_t *flen,
163 bool find_end_of_shared)
164{
165 struct xfs_buf *agbp;
166 struct xfs_btree_cur *cur;
167 int error;
168
169 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
170 if (error)
171 return error;
172
173 cur = xfs_refcountbt_init_cursor(mp, NULL, agbp, agno, NULL);
174
175 error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen,
176 find_end_of_shared);
177
178 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
179
180 xfs_buf_relse(agbp);
181 return error;
182}
183
184/*
185 * Trim the mapping to the next block where there's a change in the
186 * shared/unshared status. More specifically, this means that we
187 * find the lowest-numbered extent of shared blocks that coincides with
188 * the given block mapping. If the shared extent overlaps the start of
189 * the mapping, trim the mapping to the end of the shared extent. If
190 * the shared region intersects the mapping, trim the mapping to the
191 * start of the shared extent. If there are no shared regions that
192 * overlap, just return the original extent.
193 */
194int
195xfs_reflink_trim_around_shared(
196 struct xfs_inode *ip,
197 struct xfs_bmbt_irec *irec,
198 bool *shared,
199 bool *trimmed)
200{
201 xfs_agnumber_t agno;
202 xfs_agblock_t agbno;
203 xfs_extlen_t aglen;
204 xfs_agblock_t fbno;
205 xfs_extlen_t flen;
206 int error = 0;
207
208 /* Holes, unwritten, and delalloc extents cannot be shared */
209 if (!xfs_is_reflink_inode(ip) ||
210 ISUNWRITTEN(irec) ||
211 irec->br_startblock == HOLESTARTBLOCK ||
212 irec->br_startblock == DELAYSTARTBLOCK ||
213 isnullstartblock(irec->br_startblock)) {
214 *shared = false;
215 return 0;
216 }
217
218 trace_xfs_reflink_trim_around_shared(ip, irec);
219
220 agno = XFS_FSB_TO_AGNO(ip->i_mount, irec->br_startblock);
221 agbno = XFS_FSB_TO_AGBNO(ip->i_mount, irec->br_startblock);
222 aglen = irec->br_blockcount;
223
224 error = xfs_reflink_find_shared(ip->i_mount, agno, agbno,
225 aglen, &fbno, &flen, true);
226 if (error)
227 return error;
228
229 *shared = *trimmed = false;
230 if (fbno == NULLAGBLOCK) {
231 /* No shared blocks at all. */
232 return 0;
233 } else if (fbno == agbno) {
234 /*
235 * The start of this extent is shared. Truncate the
236 * mapping at the end of the shared region so that a
237 * subsequent iteration starts at the start of the
238 * unshared region.
239 */
240 irec->br_blockcount = flen;
241 *shared = true;
242 if (flen != aglen)
243 *trimmed = true;
244 return 0;
245 } else {
246 /*
247 * There's a shared extent midway through this extent.
248 * Truncate the mapping at the start of the shared
249 * extent so that a subsequent iteration starts at the
250 * start of the shared region.
251 */
252 irec->br_blockcount = fbno - agbno;
253 *trimmed = true;
254 return 0;
255 }
256}
257
258/*
259 * Trim the passed in imap to the next shared/unshared extent boundary, and
260 * if imap->br_startoff points to a shared extent reserve space for it in the
261 * COW fork. In this case *shared is set to true, else to false.
262 *
263 * Note that imap will always contain the block numbers for the existing blocks
264 * in the data fork, as the upper layers need them for read-modify-write
265 * operations.
266 */
267int
268xfs_reflink_reserve_cow(
269 struct xfs_inode *ip,
270 struct xfs_bmbt_irec *imap,
271 bool *shared)
272{
273 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
274 struct xfs_bmbt_irec got;
275 int error = 0;
276 bool eof = false, trimmed;
277 xfs_extnum_t idx;
278
279 /*
280 * Search the COW fork extent list first. This serves two purposes:
281 * first this implement the speculative preallocation using cowextisze,
282 * so that we also unshared block adjacent to shared blocks instead
283 * of just the shared blocks themselves. Second the lookup in the
284 * extent list is generally faster than going out to the shared extent
285 * tree.
286 */
287
288 if (!xfs_iext_lookup_extent(ip, ifp, imap->br_startoff, &idx, &got))
289 eof = true;
290 if (!eof && got.br_startoff <= imap->br_startoff) {
291 trace_xfs_reflink_cow_found(ip, imap);
292 xfs_trim_extent(imap, got.br_startoff, got.br_blockcount);
293
294 *shared = true;
295 return 0;
296 }
297
298 /* Trim the mapping to the nearest shared extent boundary. */
299 error = xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed);
300 if (error)
301 return error;
302
303 /* Not shared? Just report the (potentially capped) extent. */
304 if (!*shared)
305 return 0;
306
307 /*
308 * Fork all the shared blocks from our write offset until the end of
309 * the extent.
310 */
311 error = xfs_qm_dqattach_locked(ip, 0);
312 if (error)
313 return error;
314
315 error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, imap->br_startoff,
316 imap->br_blockcount, 0, &got, &idx, eof);
317 if (error == -ENOSPC || error == -EDQUOT)
318 trace_xfs_reflink_cow_enospc(ip, imap);
319 if (error)
320 return error;
321
322 trace_xfs_reflink_cow_alloc(ip, &got);
323 return 0;
324}
325
326/* Convert part of an unwritten CoW extent to a real one. */
327STATIC int
328xfs_reflink_convert_cow_extent(
329 struct xfs_inode *ip,
330 struct xfs_bmbt_irec *imap,
331 xfs_fileoff_t offset_fsb,
332 xfs_filblks_t count_fsb,
333 struct xfs_defer_ops *dfops)
334{
335 struct xfs_bmbt_irec irec = *imap;
336 xfs_fsblock_t first_block;
337 int nimaps = 1;
338
339 if (imap->br_state == XFS_EXT_NORM)
340 return 0;
341
342 xfs_trim_extent(&irec, offset_fsb, count_fsb);
343 trace_xfs_reflink_convert_cow(ip, &irec);
344 if (irec.br_blockcount == 0)
345 return 0;
346 return xfs_bmapi_write(NULL, ip, irec.br_startoff, irec.br_blockcount,
347 XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, &first_block,
348 0, &irec, &nimaps, dfops);
349}
350
351/* Convert all of the unwritten CoW extents in a file's range to real ones. */
352int
353xfs_reflink_convert_cow(
354 struct xfs_inode *ip,
355 xfs_off_t offset,
356 xfs_off_t count)
357{
358 struct xfs_bmbt_irec got;
359 struct xfs_defer_ops dfops;
360 struct xfs_mount *mp = ip->i_mount;
361 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
362 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
363 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
364 xfs_extnum_t idx;
365 bool found;
366 int error = 0;
367
368 xfs_ilock(ip, XFS_ILOCK_EXCL);
369
370 /* Convert all the extents to real from unwritten. */
371 for (found = xfs_iext_lookup_extent(ip, ifp, offset_fsb, &idx, &got);
372 found && got.br_startoff < end_fsb;
373 found = xfs_iext_get_extent(ifp, ++idx, &got)) {
374 error = xfs_reflink_convert_cow_extent(ip, &got, offset_fsb,
375 end_fsb - offset_fsb, &dfops);
376 if (error)
377 break;
378 }
379
380 /* Finish up. */
381 xfs_iunlock(ip, XFS_ILOCK_EXCL);
382 return error;
383}
384
385/* Allocate all CoW reservations covering a range of blocks in a file. */
386static int
387__xfs_reflink_allocate_cow(
388 struct xfs_inode *ip,
389 xfs_fileoff_t *offset_fsb,
390 xfs_fileoff_t end_fsb)
391{
392 struct xfs_mount *mp = ip->i_mount;
393 struct xfs_bmbt_irec imap;
394 struct xfs_defer_ops dfops;
395 struct xfs_trans *tp;
396 xfs_fsblock_t first_block;
397 int nimaps = 1, error;
398 bool shared;
399
400 xfs_defer_init(&dfops, &first_block);
401
402 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
403 XFS_TRANS_RESERVE, &tp);
404 if (error)
405 return error;
406
407 xfs_ilock(ip, XFS_ILOCK_EXCL);
408
409 /* Read extent from the source file. */
410 nimaps = 1;
411 error = xfs_bmapi_read(ip, *offset_fsb, end_fsb - *offset_fsb,
412 &imap, &nimaps, 0);
413 if (error)
414 goto out_unlock;
415 ASSERT(nimaps == 1);
416
417 /* Make sure there's a CoW reservation for it. */
418 error = xfs_reflink_reserve_cow(ip, &imap, &shared);
419 if (error)
420 goto out_trans_cancel;
421
422 if (!shared) {
423 *offset_fsb = imap.br_startoff + imap.br_blockcount;
424 goto out_trans_cancel;
425 }
426
427 /* Allocate the entire reservation as unwritten blocks. */
428 xfs_trans_ijoin(tp, ip, 0);
429 error = xfs_bmapi_write(tp, ip, imap.br_startoff, imap.br_blockcount,
430 XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, &first_block,
431 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK),
432 &imap, &nimaps, &dfops);
433 if (error)
434 goto out_trans_cancel;
435
436 /* Finish up. */
437 error = xfs_defer_finish(&tp, &dfops, NULL);
438 if (error)
439 goto out_trans_cancel;
440
441 error = xfs_trans_commit(tp);
442
443 *offset_fsb = imap.br_startoff + imap.br_blockcount;
444out_unlock:
445 xfs_iunlock(ip, XFS_ILOCK_EXCL);
446 return error;
447out_trans_cancel:
448 xfs_defer_cancel(&dfops);
449 xfs_trans_cancel(tp);
450 goto out_unlock;
451}
452
453/* Allocate all CoW reservations covering a part of a file. */
454int
455xfs_reflink_allocate_cow_range(
456 struct xfs_inode *ip,
457 xfs_off_t offset,
458 xfs_off_t count)
459{
460 struct xfs_mount *mp = ip->i_mount;
461 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
462 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
463 int error;
464
465 ASSERT(xfs_is_reflink_inode(ip));
466
467 trace_xfs_reflink_allocate_cow_range(ip, offset, count);
468
469 /*
470 * Make sure that the dquots are there.
471 */
472 error = xfs_qm_dqattach(ip, 0);
473 if (error)
474 return error;
475
476 while (offset_fsb < end_fsb) {
477 error = __xfs_reflink_allocate_cow(ip, &offset_fsb, end_fsb);
478 if (error) {
479 trace_xfs_reflink_allocate_cow_range_error(ip, error,
480 _RET_IP_);
481 return error;
482 }
483 }
484
485 /* Convert the CoW extents to regular. */
486 return xfs_reflink_convert_cow(ip, offset, count);
487}
488
489/*
490 * Find the CoW reservation for a given byte offset of a file.
491 */
492bool
493xfs_reflink_find_cow_mapping(
494 struct xfs_inode *ip,
495 xfs_off_t offset,
496 struct xfs_bmbt_irec *imap)
497{
498 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
499 xfs_fileoff_t offset_fsb;
500 struct xfs_bmbt_irec got;
501 xfs_extnum_t idx;
502
503 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
504 ASSERT(xfs_is_reflink_inode(ip));
505
506 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
507 if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &idx, &got))
508 return false;
509 if (got.br_startoff > offset_fsb)
510 return false;
511
512 trace_xfs_reflink_find_cow_mapping(ip, offset, 1, XFS_IO_OVERWRITE,
513 &got);
514 *imap = got;
515 return true;
516}
517
518/*
519 * Trim an extent to end at the next CoW reservation past offset_fsb.
520 */
521void
522xfs_reflink_trim_irec_to_next_cow(
523 struct xfs_inode *ip,
524 xfs_fileoff_t offset_fsb,
525 struct xfs_bmbt_irec *imap)
526{
527 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
528 struct xfs_bmbt_irec got;
529 xfs_extnum_t idx;
530
531 if (!xfs_is_reflink_inode(ip))
532 return;
533
534 /* Find the extent in the CoW fork. */
535 if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &idx, &got))
536 return;
537
538 /* This is the extent before; try sliding up one. */
539 if (got.br_startoff < offset_fsb) {
540 if (!xfs_iext_get_extent(ifp, idx + 1, &got))
541 return;
542 }
543
544 if (got.br_startoff >= imap->br_startoff + imap->br_blockcount)
545 return;
546
547 imap->br_blockcount = got.br_startoff - imap->br_startoff;
548 trace_xfs_reflink_trim_irec(ip, imap);
549}
550
551/*
552 * Cancel CoW reservations for some block range of an inode.
553 *
554 * If cancel_real is true this function cancels all COW fork extents for the
555 * inode; if cancel_real is false, real extents are not cleared.
556 */
557int
558xfs_reflink_cancel_cow_blocks(
559 struct xfs_inode *ip,
560 struct xfs_trans **tpp,
561 xfs_fileoff_t offset_fsb,
562 xfs_fileoff_t end_fsb,
563 bool cancel_real)
564{
565 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
566 struct xfs_bmbt_irec got, del;
567 xfs_extnum_t idx;
568 xfs_fsblock_t firstfsb;
569 struct xfs_defer_ops dfops;
570 int error = 0;
571
572 if (!xfs_is_reflink_inode(ip))
573 return 0;
574 if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &idx, &got))
575 return 0;
576
577 while (got.br_startoff < end_fsb) {
578 del = got;
579 xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
580 trace_xfs_reflink_cancel_cow(ip, &del);
581
582 if (isnullstartblock(del.br_startblock)) {
583 error = xfs_bmap_del_extent_delay(ip, XFS_COW_FORK,
584 &idx, &got, &del);
585 if (error)
586 break;
587 } else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) {
588 xfs_trans_ijoin(*tpp, ip, 0);
589 xfs_defer_init(&dfops, &firstfsb);
590
591 /* Free the CoW orphan record. */
592 error = xfs_refcount_free_cow_extent(ip->i_mount,
593 &dfops, del.br_startblock,
594 del.br_blockcount);
595 if (error)
596 break;
597
598 xfs_bmap_add_free(ip->i_mount, &dfops,
599 del.br_startblock, del.br_blockcount,
600 NULL);
601
602 /* Update quota accounting */
603 xfs_trans_mod_dquot_byino(*tpp, ip, XFS_TRANS_DQ_BCOUNT,
604 -(long)del.br_blockcount);
605
606 /* Roll the transaction */
607 error = xfs_defer_finish(tpp, &dfops, ip);
608 if (error) {
609 xfs_defer_cancel(&dfops);
610 break;
611 }
612
613 /* Remove the mapping from the CoW fork. */
614 xfs_bmap_del_extent_cow(ip, &idx, &got, &del);
615 }
616
617 if (!xfs_iext_get_extent(ifp, ++idx, &got))
618 break;
619 }
620
621 /* clear tag if cow fork is emptied */
622 if (!ifp->if_bytes)
623 xfs_inode_clear_cowblocks_tag(ip);
624
625 return error;
626}
627
628/*
629 * Cancel CoW reservations for some byte range of an inode.
630 *
631 * If cancel_real is true this function cancels all COW fork extents for the
632 * inode; if cancel_real is false, real extents are not cleared.
633 */
634int
635xfs_reflink_cancel_cow_range(
636 struct xfs_inode *ip,
637 xfs_off_t offset,
638 xfs_off_t count,
639 bool cancel_real)
640{
641 struct xfs_trans *tp;
642 xfs_fileoff_t offset_fsb;
643 xfs_fileoff_t end_fsb;
644 int error;
645
646 trace_xfs_reflink_cancel_cow_range(ip, offset, count);
647 ASSERT(xfs_is_reflink_inode(ip));
648
649 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
650 if (count == NULLFILEOFF)
651 end_fsb = NULLFILEOFF;
652 else
653 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
654
655 /* Start a rolling transaction to remove the mappings */
656 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
657 0, 0, 0, &tp);
658 if (error)
659 goto out;
660
661 xfs_ilock(ip, XFS_ILOCK_EXCL);
662 xfs_trans_ijoin(tp, ip, 0);
663
664 /* Scrape out the old CoW reservations */
665 error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb,
666 cancel_real);
667 if (error)
668 goto out_cancel;
669
670 error = xfs_trans_commit(tp);
671
672 xfs_iunlock(ip, XFS_ILOCK_EXCL);
673 return error;
674
675out_cancel:
676 xfs_trans_cancel(tp);
677 xfs_iunlock(ip, XFS_ILOCK_EXCL);
678out:
679 trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_);
680 return error;
681}
682
683/*
684 * Remap parts of a file's data fork after a successful CoW.
685 */
686int
687xfs_reflink_end_cow(
688 struct xfs_inode *ip,
689 xfs_off_t offset,
690 xfs_off_t count)
691{
692 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
693 struct xfs_bmbt_irec got, del;
694 struct xfs_trans *tp;
695 xfs_fileoff_t offset_fsb;
696 xfs_fileoff_t end_fsb;
697 xfs_fsblock_t firstfsb;
698 struct xfs_defer_ops dfops;
699 int error;
700 unsigned int resblks;
701 xfs_filblks_t rlen;
702 xfs_extnum_t idx;
703
704 trace_xfs_reflink_end_cow(ip, offset, count);
705
706 /* No COW extents? That's easy! */
707 if (ifp->if_bytes == 0)
708 return 0;
709
710 offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
711 end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
712
713 /* Start a rolling transaction to switch the mappings */
714 resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
715 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
716 resblks, 0, 0, &tp);
717 if (error)
718 goto out;
719
720 xfs_ilock(ip, XFS_ILOCK_EXCL);
721 xfs_trans_ijoin(tp, ip, 0);
722
723 /* If there is a hole at end_fsb - 1 go to the previous extent */
724 if (!xfs_iext_lookup_extent(ip, ifp, end_fsb - 1, &idx, &got) ||
725 got.br_startoff > end_fsb) {
726 ASSERT(idx > 0);
727 xfs_iext_get_extent(ifp, --idx, &got);
728 }
729
730 /* Walk backwards until we're out of the I/O range... */
731 while (got.br_startoff + got.br_blockcount > offset_fsb) {
732 del = got;
733 xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
734
735 /* Extent delete may have bumped idx forward */
736 if (!del.br_blockcount) {
737 idx--;
738 goto next_extent;
739 }
740
741 ASSERT(!isnullstartblock(got.br_startblock));
742
743 /*
744 * Don't remap unwritten extents; these are
745 * speculatively preallocated CoW extents that have been
746 * allocated but have not yet been involved in a write.
747 */
748 if (got.br_state == XFS_EXT_UNWRITTEN) {
749 idx--;
750 goto next_extent;
751 }
752
753 /* Unmap the old blocks in the data fork. */
754 xfs_defer_init(&dfops, &firstfsb);
755 rlen = del.br_blockcount;
756 error = __xfs_bunmapi(tp, ip, del.br_startoff, &rlen, 0, 1,
757 &firstfsb, &dfops);
758 if (error)
759 goto out_defer;
760
761 /* Trim the extent to whatever got unmapped. */
762 if (rlen) {
763 xfs_trim_extent(&del, del.br_startoff + rlen,
764 del.br_blockcount - rlen);
765 }
766 trace_xfs_reflink_cow_remap(ip, &del);
767
768 /* Free the CoW orphan record. */
769 error = xfs_refcount_free_cow_extent(tp->t_mountp, &dfops,
770 del.br_startblock, del.br_blockcount);
771 if (error)
772 goto out_defer;
773
774 /* Map the new blocks into the data fork. */
775 error = xfs_bmap_map_extent(tp->t_mountp, &dfops, ip, &del);
776 if (error)
777 goto out_defer;
778
779 /* Remove the mapping from the CoW fork. */
780 xfs_bmap_del_extent_cow(ip, &idx, &got, &del);
781
782 error = xfs_defer_finish(&tp, &dfops, ip);
783 if (error)
784 goto out_defer;
785next_extent:
786 if (!xfs_iext_get_extent(ifp, idx, &got))
787 break;
788 }
789
790 error = xfs_trans_commit(tp);
791 xfs_iunlock(ip, XFS_ILOCK_EXCL);
792 if (error)
793 goto out;
794 return 0;
795
796out_defer:
797 xfs_defer_cancel(&dfops);
798 xfs_trans_cancel(tp);
799 xfs_iunlock(ip, XFS_ILOCK_EXCL);
800out:
801 trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
802 return error;
803}
804
805/*
806 * Free leftover CoW reservations that didn't get cleaned out.
807 */
808int
809xfs_reflink_recover_cow(
810 struct xfs_mount *mp)
811{
812 xfs_agnumber_t agno;
813 int error = 0;
814
815 if (!xfs_sb_version_hasreflink(&mp->m_sb))
816 return 0;
817
818 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
819 error = xfs_refcount_recover_cow_leftovers(mp, agno);
820 if (error)
821 break;
822 }
823
824 return error;
825}
826
827/*
828 * Reflinking (Block) Ranges of Two Files Together
829 *
830 * First, ensure that the reflink flag is set on both inodes. The flag is an
831 * optimization to avoid unnecessary refcount btree lookups in the write path.
832 *
833 * Now we can iteratively remap the range of extents (and holes) in src to the
834 * corresponding ranges in dest. Let drange and srange denote the ranges of
835 * logical blocks in dest and src touched by the reflink operation.
836 *
837 * While the length of drange is greater than zero,
838 * - Read src's bmbt at the start of srange ("imap")
839 * - If imap doesn't exist, make imap appear to start at the end of srange
840 * with zero length.
841 * - If imap starts before srange, advance imap to start at srange.
842 * - If imap goes beyond srange, truncate imap to end at the end of srange.
843 * - Punch (imap start - srange start + imap len) blocks from dest at
844 * offset (drange start).
845 * - If imap points to a real range of pblks,
846 * > Increase the refcount of the imap's pblks
847 * > Map imap's pblks into dest at the offset
848 * (drange start + imap start - srange start)
849 * - Advance drange and srange by (imap start - srange start + imap len)
850 *
851 * Finally, if the reflink made dest longer, update both the in-core and
852 * on-disk file sizes.
853 *
854 * ASCII Art Demonstration:
855 *
856 * Let's say we want to reflink this source file:
857 *
858 * ----SSSSSSS-SSSSS----SSSSSS (src file)
859 * <-------------------->
860 *
861 * into this destination file:
862 *
863 * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
864 * <-------------------->
865 * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
866 * Observe that the range has different logical offsets in either file.
867 *
868 * Consider that the first extent in the source file doesn't line up with our
869 * reflink range. Unmapping and remapping are separate operations, so we can
870 * unmap more blocks from the destination file than we remap.
871 *
872 * ----SSSSSSS-SSSSS----SSSSSS
873 * <------->
874 * --DDDDD---------DDDDD--DDD
875 * <------->
876 *
877 * Now remap the source extent into the destination file:
878 *
879 * ----SSSSSSS-SSSSS----SSSSSS
880 * <------->
881 * --DDDDD--SSSSSSSDDDDD--DDD
882 * <------->
883 *
884 * Do likewise with the second hole and extent in our range. Holes in the
885 * unmap range don't affect our operation.
886 *
887 * ----SSSSSSS-SSSSS----SSSSSS
888 * <---->
889 * --DDDDD--SSSSSSS-SSSSS-DDD
890 * <---->
891 *
892 * Finally, unmap and remap part of the third extent. This will increase the
893 * size of the destination file.
894 *
895 * ----SSSSSSS-SSSSS----SSSSSS
896 * <----->
897 * --DDDDD--SSSSSSS-SSSSS----SSS
898 * <----->
899 *
900 * Once we update the destination file's i_size, we're done.
901 */
902
903/*
904 * Ensure the reflink bit is set in both inodes.
905 */
906STATIC int
907xfs_reflink_set_inode_flag(
908 struct xfs_inode *src,
909 struct xfs_inode *dest)
910{
911 struct xfs_mount *mp = src->i_mount;
912 int error;
913 struct xfs_trans *tp;
914
915 if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest))
916 return 0;
917
918 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
919 if (error)
920 goto out_error;
921
922 /* Lock both files against IO */
923 if (src->i_ino == dest->i_ino)
924 xfs_ilock(src, XFS_ILOCK_EXCL);
925 else
926 xfs_lock_two_inodes(src, dest, XFS_ILOCK_EXCL);
927
928 if (!xfs_is_reflink_inode(src)) {
929 trace_xfs_reflink_set_inode_flag(src);
930 xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL);
931 src->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
932 xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
933 xfs_ifork_init_cow(src);
934 } else
935 xfs_iunlock(src, XFS_ILOCK_EXCL);
936
937 if (src->i_ino == dest->i_ino)
938 goto commit_flags;
939
940 if (!xfs_is_reflink_inode(dest)) {
941 trace_xfs_reflink_set_inode_flag(dest);
942 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
943 dest->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
944 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
945 xfs_ifork_init_cow(dest);
946 } else
947 xfs_iunlock(dest, XFS_ILOCK_EXCL);
948
949commit_flags:
950 error = xfs_trans_commit(tp);
951 if (error)
952 goto out_error;
953 return error;
954
955out_error:
956 trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_);
957 return error;
958}
959
960/*
961 * Update destination inode size & cowextsize hint, if necessary.
962 */
963STATIC int
964xfs_reflink_update_dest(
965 struct xfs_inode *dest,
966 xfs_off_t newlen,
967 xfs_extlen_t cowextsize,
968 bool is_dedupe)
969{
970 struct xfs_mount *mp = dest->i_mount;
971 struct xfs_trans *tp;
972 int error;
973
974 if (is_dedupe && newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0)
975 return 0;
976
977 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
978 if (error)
979 goto out_error;
980
981 xfs_ilock(dest, XFS_ILOCK_EXCL);
982 xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
983
984 if (newlen > i_size_read(VFS_I(dest))) {
985 trace_xfs_reflink_update_inode_size(dest, newlen);
986 i_size_write(VFS_I(dest), newlen);
987 dest->i_d.di_size = newlen;
988 }
989
990 if (cowextsize) {
991 dest->i_d.di_cowextsize = cowextsize;
992 dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
993 }
994
995 if (!is_dedupe) {
996 xfs_trans_ichgtime(tp, dest,
997 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
998 }
999 xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
1000
1001 error = xfs_trans_commit(tp);
1002 if (error)
1003 goto out_error;
1004 return error;
1005
1006out_error:
1007 trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_);
1008 return error;
1009}
1010
1011/*
1012 * Do we have enough reserve in this AG to handle a reflink? The refcount
1013 * btree already reserved all the space it needs, but the rmap btree can grow
1014 * infinitely, so we won't allow more reflinks when the AG is down to the
1015 * btree reserves.
1016 */
1017static int
1018xfs_reflink_ag_has_free_space(
1019 struct xfs_mount *mp,
1020 xfs_agnumber_t agno)
1021{
1022 struct xfs_perag *pag;
1023 int error = 0;
1024
1025 if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
1026 return 0;
1027
1028 pag = xfs_perag_get(mp, agno);
1029 if (xfs_ag_resv_critical(pag, XFS_AG_RESV_AGFL) ||
1030 xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA))
1031 error = -ENOSPC;
1032 xfs_perag_put(pag);
1033 return error;
1034}
1035
1036/*
1037 * Unmap a range of blocks from a file, then map other blocks into the hole.
1038 * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount).
1039 * The extent irec is mapped into dest at irec->br_startoff.
1040 */
1041STATIC int
1042xfs_reflink_remap_extent(
1043 struct xfs_inode *ip,
1044 struct xfs_bmbt_irec *irec,
1045 xfs_fileoff_t destoff,
1046 xfs_off_t new_isize)
1047{
1048 struct xfs_mount *mp = ip->i_mount;
1049 struct xfs_trans *tp;
1050 xfs_fsblock_t firstfsb;
1051 unsigned int resblks;
1052 struct xfs_defer_ops dfops;
1053 struct xfs_bmbt_irec uirec;
1054 bool real_extent;
1055 xfs_filblks_t rlen;
1056 xfs_filblks_t unmap_len;
1057 xfs_off_t newlen;
1058 int error;
1059
1060 unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
1061 trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
1062
1063 /* Only remap normal extents. */
1064 real_extent = (irec->br_startblock != HOLESTARTBLOCK &&
1065 irec->br_startblock != DELAYSTARTBLOCK &&
1066 !ISUNWRITTEN(irec));
1067
1068 /* No reflinking if we're low on space */
1069 if (real_extent) {
1070 error = xfs_reflink_ag_has_free_space(mp,
1071 XFS_FSB_TO_AGNO(mp, irec->br_startblock));
1072 if (error)
1073 goto out;
1074 }
1075
1076 /* Start a rolling transaction to switch the mappings */
1077 resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
1078 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1079 if (error)
1080 goto out;
1081
1082 xfs_ilock(ip, XFS_ILOCK_EXCL);
1083 xfs_trans_ijoin(tp, ip, 0);
1084
1085 /* If we're not just clearing space, then do we have enough quota? */
1086 if (real_extent) {
1087 error = xfs_trans_reserve_quota_nblks(tp, ip,
1088 irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS);
1089 if (error)
1090 goto out_cancel;
1091 }
1092
1093 trace_xfs_reflink_remap(ip, irec->br_startoff,
1094 irec->br_blockcount, irec->br_startblock);
1095
1096 /* Unmap the old blocks in the data fork. */
1097 rlen = unmap_len;
1098 while (rlen) {
1099 xfs_defer_init(&dfops, &firstfsb);
1100 error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1,
1101 &firstfsb, &dfops);
1102 if (error)
1103 goto out_defer;
1104
1105 /*
1106 * Trim the extent to whatever got unmapped.
1107 * Remember, bunmapi works backwards.
1108 */
1109 uirec.br_startblock = irec->br_startblock + rlen;
1110 uirec.br_startoff = irec->br_startoff + rlen;
1111 uirec.br_blockcount = unmap_len - rlen;
1112 unmap_len = rlen;
1113
1114 /* If this isn't a real mapping, we're done. */
1115 if (!real_extent || uirec.br_blockcount == 0)
1116 goto next_extent;
1117
1118 trace_xfs_reflink_remap(ip, uirec.br_startoff,
1119 uirec.br_blockcount, uirec.br_startblock);
1120
1121 /* Update the refcount tree */
1122 error = xfs_refcount_increase_extent(mp, &dfops, &uirec);
1123 if (error)
1124 goto out_defer;
1125
1126 /* Map the new blocks into the data fork. */
1127 error = xfs_bmap_map_extent(mp, &dfops, ip, &uirec);
1128 if (error)
1129 goto out_defer;
1130
1131 /* Update quota accounting. */
1132 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
1133 uirec.br_blockcount);
1134
1135 /* Update dest isize if needed. */
1136 newlen = XFS_FSB_TO_B(mp,
1137 uirec.br_startoff + uirec.br_blockcount);
1138 newlen = min_t(xfs_off_t, newlen, new_isize);
1139 if (newlen > i_size_read(VFS_I(ip))) {
1140 trace_xfs_reflink_update_inode_size(ip, newlen);
1141 i_size_write(VFS_I(ip), newlen);
1142 ip->i_d.di_size = newlen;
1143 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1144 }
1145
1146next_extent:
1147 /* Process all the deferred stuff. */
1148 error = xfs_defer_finish(&tp, &dfops, ip);
1149 if (error)
1150 goto out_defer;
1151 }
1152
1153 error = xfs_trans_commit(tp);
1154 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1155 if (error)
1156 goto out;
1157 return 0;
1158
1159out_defer:
1160 xfs_defer_cancel(&dfops);
1161out_cancel:
1162 xfs_trans_cancel(tp);
1163 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1164out:
1165 trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
1166 return error;
1167}
1168
1169/*
1170 * Iteratively remap one file's extents (and holes) to another's.
1171 */
1172STATIC int
1173xfs_reflink_remap_blocks(
1174 struct xfs_inode *src,
1175 xfs_fileoff_t srcoff,
1176 struct xfs_inode *dest,
1177 xfs_fileoff_t destoff,
1178 xfs_filblks_t len,
1179 xfs_off_t new_isize)
1180{
1181 struct xfs_bmbt_irec imap;
1182 int nimaps;
1183 int error = 0;
1184 xfs_filblks_t range_len;
1185
1186 /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
1187 while (len) {
1188 trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
1189 dest, destoff);
1190 /* Read extent from the source file */
1191 nimaps = 1;
1192 xfs_ilock(src, XFS_ILOCK_EXCL);
1193 error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0);
1194 xfs_iunlock(src, XFS_ILOCK_EXCL);
1195 if (error)
1196 goto err;
1197 ASSERT(nimaps == 1);
1198
1199 trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE,
1200 &imap);
1201
1202 /* Translate imap into the destination file. */
1203 range_len = imap.br_startoff + imap.br_blockcount - srcoff;
1204 imap.br_startoff += destoff - srcoff;
1205
1206 /* Clear dest from destoff to the end of imap and map it in. */
1207 error = xfs_reflink_remap_extent(dest, &imap, destoff,
1208 new_isize);
1209 if (error)
1210 goto err;
1211
1212 if (fatal_signal_pending(current)) {
1213 error = -EINTR;
1214 goto err;
1215 }
1216
1217 /* Advance drange/srange */
1218 srcoff += range_len;
1219 destoff += range_len;
1220 len -= range_len;
1221 }
1222
1223 return 0;
1224
1225err:
1226 trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_);
1227 return error;
1228}
1229
1230/*
1231 * Link a range of blocks from one file to another.
1232 */
1233int
1234xfs_reflink_remap_range(
1235 struct file *file_in,
1236 loff_t pos_in,
1237 struct file *file_out,
1238 loff_t pos_out,
1239 u64 len,
1240 bool is_dedupe)
1241{
1242 struct inode *inode_in = file_inode(file_in);
1243 struct xfs_inode *src = XFS_I(inode_in);
1244 struct inode *inode_out = file_inode(file_out);
1245 struct xfs_inode *dest = XFS_I(inode_out);
1246 struct xfs_mount *mp = src->i_mount;
1247 bool same_inode = (inode_in == inode_out);
1248 xfs_fileoff_t sfsbno, dfsbno;
1249 xfs_filblks_t fsblen;
1250 xfs_extlen_t cowextsize;
1251 ssize_t ret;
1252
1253 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1254 return -EOPNOTSUPP;
1255
1256 if (XFS_FORCED_SHUTDOWN(mp))
1257 return -EIO;
1258
1259 /* Lock both files against IO */
1260 lock_two_nondirectories(inode_in, inode_out);
1261 if (same_inode)
1262 xfs_ilock(src, XFS_MMAPLOCK_EXCL);
1263 else
1264 xfs_lock_two_inodes(src, dest, XFS_MMAPLOCK_EXCL);
1265
1266 /* Check file eligibility and prepare for block sharing. */
1267 ret = -EINVAL;
1268 /* Don't reflink realtime inodes */
1269 if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
1270 goto out_unlock;
1271
1272 /* Don't share DAX file data for now. */
1273 if (IS_DAX(inode_in) || IS_DAX(inode_out))
1274 goto out_unlock;
1275
1276 ret = vfs_clone_file_prep_inodes(inode_in, pos_in, inode_out, pos_out,
1277 &len, is_dedupe);
1278 if (ret <= 0)
1279 goto out_unlock;
1280
1281 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1282
1283 /* Set flags and remap blocks. */
1284 ret = xfs_reflink_set_inode_flag(src, dest);
1285 if (ret)
1286 goto out_unlock;
1287
1288 dfsbno = XFS_B_TO_FSBT(mp, pos_out);
1289 sfsbno = XFS_B_TO_FSBT(mp, pos_in);
1290 fsblen = XFS_B_TO_FSB(mp, len);
1291 ret = xfs_reflink_remap_blocks(src, sfsbno, dest, dfsbno, fsblen,
1292 pos_out + len);
1293 if (ret)
1294 goto out_unlock;
1295
1296 /* Zap any page cache for the destination file's range. */
1297 truncate_inode_pages_range(&inode_out->i_data, pos_out,
1298 PAGE_ALIGN(pos_out + len) - 1);
1299
1300 /*
1301 * Carry the cowextsize hint from src to dest if we're sharing the
1302 * entire source file to the entire destination file, the source file
1303 * has a cowextsize hint, and the destination file does not.
1304 */
1305 cowextsize = 0;
1306 if (pos_in == 0 && len == i_size_read(inode_in) &&
1307 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1308 pos_out == 0 && len >= i_size_read(inode_out) &&
1309 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1310 cowextsize = src->i_d.di_cowextsize;
1311
1312 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1313 is_dedupe);
1314
1315out_unlock:
1316 xfs_iunlock(src, XFS_MMAPLOCK_EXCL);
1317 if (!same_inode)
1318 xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
1319 unlock_two_nondirectories(inode_in, inode_out);
1320 if (ret)
1321 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1322 return ret;
1323}
1324
1325/*
1326 * The user wants to preemptively CoW all shared blocks in this file,
1327 * which enables us to turn off the reflink flag. Iterate all
1328 * extents which are not prealloc/delalloc to see which ranges are
1329 * mentioned in the refcount tree, then read those blocks into the
1330 * pagecache, dirty them, fsync them back out, and then we can update
1331 * the inode flag. What happens if we run out of memory? :)
1332 */
1333STATIC int
1334xfs_reflink_dirty_extents(
1335 struct xfs_inode *ip,
1336 xfs_fileoff_t fbno,
1337 xfs_filblks_t end,
1338 xfs_off_t isize)
1339{
1340 struct xfs_mount *mp = ip->i_mount;
1341 xfs_agnumber_t agno;
1342 xfs_agblock_t agbno;
1343 xfs_extlen_t aglen;
1344 xfs_agblock_t rbno;
1345 xfs_extlen_t rlen;
1346 xfs_off_t fpos;
1347 xfs_off_t flen;
1348 struct xfs_bmbt_irec map[2];
1349 int nmaps;
1350 int error = 0;
1351
1352 while (end - fbno > 0) {
1353 nmaps = 1;
1354 /*
1355 * Look for extents in the file. Skip holes, delalloc, or
1356 * unwritten extents; they can't be reflinked.
1357 */
1358 error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0);
1359 if (error)
1360 goto out;
1361 if (nmaps == 0)
1362 break;
1363 if (map[0].br_startblock == HOLESTARTBLOCK ||
1364 map[0].br_startblock == DELAYSTARTBLOCK ||
1365 ISUNWRITTEN(&map[0]))
1366 goto next;
1367
1368 map[1] = map[0];
1369 while (map[1].br_blockcount) {
1370 agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock);
1371 agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock);
1372 aglen = map[1].br_blockcount;
1373
1374 error = xfs_reflink_find_shared(mp, agno, agbno, aglen,
1375 &rbno, &rlen, true);
1376 if (error)
1377 goto out;
1378 if (rbno == NULLAGBLOCK)
1379 break;
1380
1381 /* Dirty the pages */
1382 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1383 fpos = XFS_FSB_TO_B(mp, map[1].br_startoff +
1384 (rbno - agbno));
1385 flen = XFS_FSB_TO_B(mp, rlen);
1386 if (fpos + flen > isize)
1387 flen = isize - fpos;
1388 error = iomap_file_dirty(VFS_I(ip), fpos, flen,
1389 &xfs_iomap_ops);
1390 xfs_ilock(ip, XFS_ILOCK_EXCL);
1391 if (error)
1392 goto out;
1393
1394 map[1].br_blockcount -= (rbno - agbno + rlen);
1395 map[1].br_startoff += (rbno - agbno + rlen);
1396 map[1].br_startblock += (rbno - agbno + rlen);
1397 }
1398
1399next:
1400 fbno = map[0].br_startoff + map[0].br_blockcount;
1401 }
1402out:
1403 return error;
1404}
1405
1406/* Clear the inode reflink flag if there are no shared extents. */
1407int
1408xfs_reflink_clear_inode_flag(
1409 struct xfs_inode *ip,
1410 struct xfs_trans **tpp)
1411{
1412 struct xfs_mount *mp = ip->i_mount;
1413 xfs_fileoff_t fbno;
1414 xfs_filblks_t end;
1415 xfs_agnumber_t agno;
1416 xfs_agblock_t agbno;
1417 xfs_extlen_t aglen;
1418 xfs_agblock_t rbno;
1419 xfs_extlen_t rlen;
1420 struct xfs_bmbt_irec map;
1421 int nmaps;
1422 int error = 0;
1423
1424 ASSERT(xfs_is_reflink_inode(ip));
1425
1426 fbno = 0;
1427 end = XFS_B_TO_FSB(mp, i_size_read(VFS_I(ip)));
1428 while (end - fbno > 0) {
1429 nmaps = 1;
1430 /*
1431 * Look for extents in the file. Skip holes, delalloc, or
1432 * unwritten extents; they can't be reflinked.
1433 */
1434 error = xfs_bmapi_read(ip, fbno, end - fbno, &map, &nmaps, 0);
1435 if (error)
1436 return error;
1437 if (nmaps == 0)
1438 break;
1439 if (map.br_startblock == HOLESTARTBLOCK ||
1440 map.br_startblock == DELAYSTARTBLOCK ||
1441 ISUNWRITTEN(&map))
1442 goto next;
1443
1444 agno = XFS_FSB_TO_AGNO(mp, map.br_startblock);
1445 agbno = XFS_FSB_TO_AGBNO(mp, map.br_startblock);
1446 aglen = map.br_blockcount;
1447
1448 error = xfs_reflink_find_shared(mp, agno, agbno, aglen,
1449 &rbno, &rlen, false);
1450 if (error)
1451 return error;
1452 /* Is there still a shared block here? */
1453 if (rbno != NULLAGBLOCK)
1454 return 0;
1455next:
1456 fbno = map.br_startoff + map.br_blockcount;
1457 }
1458
1459 /*
1460 * We didn't find any shared blocks so turn off the reflink flag.
1461 * First, get rid of any leftover CoW mappings.
1462 */
1463 error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF, true);
1464 if (error)
1465 return error;
1466
1467 /* Clear the inode flag. */
1468 trace_xfs_reflink_unset_inode_flag(ip);
1469 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1470 xfs_inode_clear_cowblocks_tag(ip);
1471 xfs_trans_ijoin(*tpp, ip, 0);
1472 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1473
1474 return error;
1475}
1476
1477/*
1478 * Clear the inode reflink flag if there are no shared extents and the size
1479 * hasn't changed.
1480 */
1481STATIC int
1482xfs_reflink_try_clear_inode_flag(
1483 struct xfs_inode *ip)
1484{
1485 struct xfs_mount *mp = ip->i_mount;
1486 struct xfs_trans *tp;
1487 int error = 0;
1488
1489 /* Start a rolling transaction to remove the mappings */
1490 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1491 if (error)
1492 return error;
1493
1494 xfs_ilock(ip, XFS_ILOCK_EXCL);
1495 xfs_trans_ijoin(tp, ip, 0);
1496
1497 error = xfs_reflink_clear_inode_flag(ip, &tp);
1498 if (error)
1499 goto cancel;
1500
1501 error = xfs_trans_commit(tp);
1502 if (error)
1503 goto out;
1504
1505 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1506 return 0;
1507cancel:
1508 xfs_trans_cancel(tp);
1509out:
1510 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1511 return error;
1512}
1513
1514/*
1515 * Pre-COW all shared blocks within a given byte range of a file and turn off
1516 * the reflink flag if we unshare all of the file's blocks.
1517 */
1518int
1519xfs_reflink_unshare(
1520 struct xfs_inode *ip,
1521 xfs_off_t offset,
1522 xfs_off_t len)
1523{
1524 struct xfs_mount *mp = ip->i_mount;
1525 xfs_fileoff_t fbno;
1526 xfs_filblks_t end;
1527 xfs_off_t isize;
1528 int error;
1529
1530 if (!xfs_is_reflink_inode(ip))
1531 return 0;
1532
1533 trace_xfs_reflink_unshare(ip, offset, len);
1534
1535 inode_dio_wait(VFS_I(ip));
1536
1537 /* Try to CoW the selected ranges */
1538 xfs_ilock(ip, XFS_ILOCK_EXCL);
1539 fbno = XFS_B_TO_FSBT(mp, offset);
1540 isize = i_size_read(VFS_I(ip));
1541 end = XFS_B_TO_FSB(mp, offset + len);
1542 error = xfs_reflink_dirty_extents(ip, fbno, end, isize);
1543 if (error)
1544 goto out_unlock;
1545 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1546
1547 /* Wait for the IO to finish */
1548 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1549 if (error)
1550 goto out;
1551
1552 /* Turn off the reflink flag if possible. */
1553 error = xfs_reflink_try_clear_inode_flag(ip);
1554 if (error)
1555 goto out;
1556
1557 return 0;
1558
1559out_unlock:
1560 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1561out:
1562 trace_xfs_reflink_unshare_error(ip, error, _RET_IP_);
1563 return error;
1564}