]>
Commit | Line | Data |
---|---|---|
3993baeb DW |
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" | |
174edb0e | 43 | #include "xfs_btree.h" |
3993baeb DW |
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" | |
3993baeb | 52 | #include "xfs_reflink.h" |
2a06705c | 53 | #include "xfs_iomap.h" |
43caeb18 | 54 | #include "xfs_rmap_btree.h" |
6fa164b8 DW |
55 | #include "xfs_sb.h" |
56 | #include "xfs_ag_resv.h" | |
3993baeb DW |
57 | |
58 | /* | |
59 | * Copy on Write of Shared Blocks | |
60 | * | |
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. | |
68 | * | |
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. | |
73 | * | |
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. | |
82 | * | |
5eda4300 DW |
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. | |
87 | * | |
88 | * D: --RRRRRRSSSRRRRRRRR--- (data fork) | |
89 | * C: ------DDDDDDD--------- (CoW fork) | |
90 | * | |
3993baeb | 91 | * When dirty pages are being written out (typically in writepage), the |
5eda4300 DW |
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. | |
96 | * | |
97 | * D: --RRRRRRSSSRRRRRRRR--- | |
98 | * C: ------UUUUUUU--------- | |
3993baeb DW |
99 | * |
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. | |
108 | * | |
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 | |
111 | * writes. | |
112 | * | |
5eda4300 DW |
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: | |
117 | * | |
118 | * D: --RRRRRRSSSRRRRRRRR--- | |
119 | * C: ------UUrrUUU--------- | |
120 | * | |
3993baeb DW |
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 | |
5eda4300 DW |
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 | |
132 | * the CoW fork: | |
133 | * | |
134 | * D: --RRRRRRrrSRRRRRRRR--- | |
135 | * C: ------UU--UUU--------- | |
3993baeb DW |
136 | * |
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 | |
143 | * ioend, the better. | |
144 | */ | |
2a06705c DW |
145 | |
146 | /* | |
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. | |
152 | */ | |
153 | int | |
154 | xfs_reflink_find_shared( | |
155 | struct xfs_mount *mp, | |
92ff7285 | 156 | struct xfs_trans *tp, |
2a06705c DW |
157 | xfs_agnumber_t agno, |
158 | xfs_agblock_t agbno, | |
159 | xfs_extlen_t aglen, | |
160 | xfs_agblock_t *fbno, | |
161 | xfs_extlen_t *flen, | |
162 | bool find_end_of_shared) | |
163 | { | |
164 | struct xfs_buf *agbp; | |
165 | struct xfs_btree_cur *cur; | |
166 | int error; | |
167 | ||
92ff7285 | 168 | error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp); |
2a06705c DW |
169 | if (error) |
170 | return error; | |
10479e2d DW |
171 | if (!agbp) |
172 | return -ENOMEM; | |
2a06705c | 173 | |
92ff7285 | 174 | cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, NULL); |
2a06705c DW |
175 | |
176 | error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen, | |
177 | find_end_of_shared); | |
178 | ||
179 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
180 | ||
92ff7285 | 181 | xfs_trans_brelse(tp, agbp); |
2a06705c DW |
182 | return error; |
183 | } | |
184 | ||
185 | /* | |
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. | |
194 | */ | |
195 | int | |
196 | xfs_reflink_trim_around_shared( | |
197 | struct xfs_inode *ip, | |
198 | struct xfs_bmbt_irec *irec, | |
199 | bool *shared, | |
200 | bool *trimmed) | |
201 | { | |
202 | xfs_agnumber_t agno; | |
203 | xfs_agblock_t agbno; | |
204 | xfs_extlen_t aglen; | |
205 | xfs_agblock_t fbno; | |
206 | xfs_extlen_t flen; | |
207 | int error = 0; | |
208 | ||
209 | /* Holes, unwritten, and delalloc extents cannot be shared */ | |
9c4f29d3 | 210 | if (!xfs_is_reflink_inode(ip) || !xfs_bmap_is_real_extent(irec)) { |
2a06705c DW |
211 | *shared = false; |
212 | return 0; | |
213 | } | |
214 | ||
215 | trace_xfs_reflink_trim_around_shared(ip, irec); | |
216 | ||
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; | |
220 | ||
92ff7285 | 221 | error = xfs_reflink_find_shared(ip->i_mount, NULL, agno, agbno, |
2a06705c DW |
222 | aglen, &fbno, &flen, true); |
223 | if (error) | |
224 | return error; | |
225 | ||
226 | *shared = *trimmed = false; | |
227 | if (fbno == NULLAGBLOCK) { | |
228 | /* No shared blocks at all. */ | |
229 | return 0; | |
230 | } else if (fbno == agbno) { | |
231 | /* | |
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 | |
235 | * unshared region. | |
236 | */ | |
237 | irec->br_blockcount = flen; | |
238 | *shared = true; | |
239 | if (flen != aglen) | |
240 | *trimmed = true; | |
241 | return 0; | |
242 | } else { | |
243 | /* | |
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. | |
248 | */ | |
249 | irec->br_blockcount = fbno - agbno; | |
250 | *trimmed = true; | |
251 | return 0; | |
252 | } | |
253 | } | |
254 | ||
3ba020be CH |
255 | /* |
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. | |
259 | * | |
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 | |
262 | * operations. | |
263 | */ | |
264 | int | |
265 | xfs_reflink_reserve_cow( | |
2a06705c | 266 | struct xfs_inode *ip, |
3ba020be CH |
267 | struct xfs_bmbt_irec *imap, |
268 | bool *shared) | |
2a06705c | 269 | { |
2755fc44 CH |
270 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
271 | struct xfs_bmbt_irec got; | |
2755fc44 CH |
272 | int error = 0; |
273 | bool eof = false, trimmed; | |
b2b1712a | 274 | struct xfs_iext_cursor icur; |
2a06705c | 275 | |
3ba020be CH |
276 | /* |
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 | |
282 | * tree. | |
283 | */ | |
2755fc44 | 284 | |
b2b1712a | 285 | if (!xfs_iext_lookup_extent(ip, ifp, imap->br_startoff, &icur, &got)) |
2755fc44 | 286 | eof = true; |
3ba020be CH |
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); | |
2a06705c | 290 | |
3ba020be CH |
291 | *shared = true; |
292 | return 0; | |
293 | } | |
2a06705c DW |
294 | |
295 | /* Trim the mapping to the nearest shared extent boundary. */ | |
3ba020be | 296 | error = xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed); |
2a06705c | 297 | if (error) |
3ba020be | 298 | return error; |
2a06705c DW |
299 | |
300 | /* Not shared? Just report the (potentially capped) extent. */ | |
3ba020be CH |
301 | if (!*shared) |
302 | return 0; | |
2a06705c DW |
303 | |
304 | /* | |
305 | * Fork all the shared blocks from our write offset until the end of | |
306 | * the extent. | |
307 | */ | |
308 | error = xfs_qm_dqattach_locked(ip, 0); | |
309 | if (error) | |
3ba020be CH |
310 | return error; |
311 | ||
3ba020be | 312 | error = xfs_bmapi_reserve_delalloc(ip, XFS_COW_FORK, imap->br_startoff, |
b2b1712a | 313 | imap->br_blockcount, 0, &got, &icur, eof); |
0260d8ff | 314 | if (error == -ENOSPC || error == -EDQUOT) |
3ba020be | 315 | trace_xfs_reflink_cow_enospc(ip, imap); |
0260d8ff | 316 | if (error) |
3ba020be | 317 | return error; |
83104d44 | 318 | |
32eaf569 | 319 | xfs_trim_extent(imap, got.br_startoff, got.br_blockcount); |
2a06705c | 320 | trace_xfs_reflink_cow_alloc(ip, &got); |
3ba020be | 321 | return 0; |
2a06705c | 322 | } |
ef473667 | 323 | |
5eda4300 DW |
324 | /* Convert part of an unwritten CoW extent to a real one. */ |
325 | STATIC int | |
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) | |
332 | { | |
4c1a67bd | 333 | xfs_fsblock_t first_block = NULLFSBLOCK; |
5eda4300 DW |
334 | int nimaps = 1; |
335 | ||
336 | if (imap->br_state == XFS_EXT_NORM) | |
337 | return 0; | |
338 | ||
dcf9585a CH |
339 | xfs_trim_extent(imap, offset_fsb, count_fsb); |
340 | trace_xfs_reflink_convert_cow(ip, imap); | |
341 | if (imap->br_blockcount == 0) | |
5eda4300 | 342 | return 0; |
dcf9585a | 343 | return xfs_bmapi_write(NULL, ip, imap->br_startoff, imap->br_blockcount, |
5eda4300 | 344 | XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, &first_block, |
dcf9585a | 345 | 0, imap, &nimaps, dfops); |
5eda4300 DW |
346 | } |
347 | ||
348 | /* Convert all of the unwritten CoW extents in a file's range to real ones. */ | |
349 | int | |
350 | xfs_reflink_convert_cow( | |
351 | struct xfs_inode *ip, | |
352 | xfs_off_t offset, | |
353 | xfs_off_t count) | |
354 | { | |
5eda4300 | 355 | struct xfs_mount *mp = ip->i_mount; |
5eda4300 DW |
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); | |
b121459c CH |
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; | |
5eda4300 | 363 | |
b121459c | 364 | ASSERT(count != 0); |
5eda4300 | 365 | |
b121459c CH |
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, | |
370 | &dfops); | |
5eda4300 DW |
371 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
372 | return error; | |
373 | } | |
374 | ||
0613f16c | 375 | /* Allocate all CoW reservations covering a range of blocks in a file. */ |
3c68d44a CH |
376 | int |
377 | xfs_reflink_allocate_cow( | |
0613f16c | 378 | struct xfs_inode *ip, |
3c68d44a CH |
379 | struct xfs_bmbt_irec *imap, |
380 | bool *shared, | |
381 | uint *lockmode) | |
0613f16c DW |
382 | { |
383 | struct xfs_mount *mp = ip->i_mount; | |
3c68d44a CH |
384 | xfs_fileoff_t offset_fsb = imap->br_startoff; |
385 | xfs_filblks_t count_fsb = imap->br_blockcount; | |
386 | struct xfs_bmbt_irec got; | |
0613f16c | 387 | struct xfs_defer_ops dfops; |
3c68d44a | 388 | struct xfs_trans *tp = NULL; |
0613f16c | 389 | xfs_fsblock_t first_block; |
3c68d44a CH |
390 | int nimaps, error = 0; |
391 | bool trimmed; | |
a14234c7 | 392 | xfs_filblks_t resaligned; |
3c68d44a | 393 | xfs_extlen_t resblks = 0; |
b2b1712a | 394 | struct xfs_iext_cursor icur; |
0613f16c | 395 | |
3c68d44a CH |
396 | retry: |
397 | ASSERT(xfs_is_reflink_inode(ip)); | |
398 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED)); | |
0613f16c | 399 | |
a14234c7 CH |
400 | /* |
401 | * Even if the extent is not shared we might have a preallocation for | |
402 | * it in the COW fork. If so use it. | |
403 | */ | |
b2b1712a | 404 | if (xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got) && |
3c68d44a CH |
405 | got.br_startoff <= offset_fsb) { |
406 | *shared = true; | |
407 | ||
a14234c7 CH |
408 | /* If we have a real allocation in the COW fork we're done. */ |
409 | if (!isnullstartblock(got.br_startblock)) { | |
3c68d44a CH |
410 | xfs_trim_extent(&got, offset_fsb, count_fsb); |
411 | *imap = got; | |
412 | goto convert; | |
a14234c7 | 413 | } |
3c68d44a CH |
414 | |
415 | xfs_trim_extent(imap, got.br_startoff, got.br_blockcount); | |
a14234c7 | 416 | } else { |
3c68d44a CH |
417 | error = xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed); |
418 | if (error || !*shared) | |
419 | goto out; | |
420 | } | |
a14234c7 | 421 | |
3c68d44a CH |
422 | if (!tp) { |
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); | |
3ba020be | 426 | |
3c68d44a CH |
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); | |
a14234c7 | 431 | |
3c68d44a CH |
432 | if (error) |
433 | return error; | |
434 | ||
435 | error = xfs_qm_dqattach_locked(ip, 0); | |
436 | if (error) | |
437 | goto out; | |
438 | goto retry; | |
a14234c7 CH |
439 | } |
440 | ||
441 | error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0, | |
442 | XFS_QMOPT_RES_REGBLKS); | |
0613f16c | 443 | if (error) |
3c68d44a | 444 | goto out; |
0613f16c | 445 | |
a14234c7 CH |
446 | xfs_trans_ijoin(tp, ip, 0); |
447 | ||
448 | xfs_defer_init(&dfops, &first_block); | |
449 | nimaps = 1; | |
0613f16c | 450 | |
5eda4300 | 451 | /* Allocate the entire reservation as unwritten blocks. */ |
3c68d44a | 452 | error = xfs_bmapi_write(tp, ip, imap->br_startoff, imap->br_blockcount, |
5eda4300 | 453 | XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, &first_block, |
3c68d44a | 454 | resblks, imap, &nimaps, &dfops); |
0613f16c | 455 | if (error) |
a14234c7 | 456 | goto out_bmap_cancel; |
0613f16c | 457 | |
86d692bf DW |
458 | xfs_inode_set_cowblocks_tag(ip); |
459 | ||
5eda4300 | 460 | /* Finish up. */ |
8ad7c629 | 461 | error = xfs_defer_finish(&tp, &dfops); |
0613f16c | 462 | if (error) |
a14234c7 | 463 | goto out_bmap_cancel; |
0613f16c DW |
464 | |
465 | error = xfs_trans_commit(tp); | |
a14234c7 | 466 | if (error) |
3c68d44a CH |
467 | return error; |
468 | convert: | |
469 | return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb, | |
470 | &dfops); | |
a14234c7 | 471 | out_bmap_cancel: |
0613f16c | 472 | xfs_defer_cancel(&dfops); |
a14234c7 CH |
473 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)resblks, 0, |
474 | XFS_QMOPT_RES_REGBLKS); | |
3c68d44a CH |
475 | out: |
476 | if (tp) | |
477 | xfs_trans_cancel(tp); | |
478 | return error; | |
0613f16c DW |
479 | } |
480 | ||
ef473667 | 481 | /* |
092d5d9d | 482 | * Find the CoW reservation for a given byte offset of a file. |
ef473667 DW |
483 | */ |
484 | bool | |
485 | xfs_reflink_find_cow_mapping( | |
486 | struct xfs_inode *ip, | |
487 | xfs_off_t offset, | |
092d5d9d | 488 | struct xfs_bmbt_irec *imap) |
ef473667 | 489 | { |
092d5d9d CH |
490 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
491 | xfs_fileoff_t offset_fsb; | |
492 | struct xfs_bmbt_irec got; | |
b2b1712a | 493 | struct xfs_iext_cursor icur; |
ef473667 DW |
494 | |
495 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED)); | |
ef473667 | 496 | |
73353f48 DW |
497 | if (!xfs_is_reflink_inode(ip)) |
498 | return false; | |
092d5d9d | 499 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
b2b1712a | 500 | if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got)) |
ef473667 | 501 | return false; |
092d5d9d | 502 | if (got.br_startoff > offset_fsb) |
ef473667 DW |
503 | return false; |
504 | ||
505 | trace_xfs_reflink_find_cow_mapping(ip, offset, 1, XFS_IO_OVERWRITE, | |
092d5d9d CH |
506 | &got); |
507 | *imap = got; | |
ef473667 DW |
508 | return true; |
509 | } | |
510 | ||
511 | /* | |
512 | * Trim an extent to end at the next CoW reservation past offset_fsb. | |
513 | */ | |
86f12ab0 | 514 | void |
ef473667 DW |
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) | |
519 | { | |
86f12ab0 CH |
520 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
521 | struct xfs_bmbt_irec got; | |
b2b1712a | 522 | struct xfs_iext_cursor icur; |
ef473667 DW |
523 | |
524 | if (!xfs_is_reflink_inode(ip)) | |
86f12ab0 | 525 | return; |
ef473667 DW |
526 | |
527 | /* Find the extent in the CoW fork. */ | |
b2b1712a | 528 | if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got)) |
86f12ab0 | 529 | return; |
ef473667 DW |
530 | |
531 | /* This is the extent before; try sliding up one. */ | |
86f12ab0 | 532 | if (got.br_startoff < offset_fsb) { |
b2b1712a | 533 | if (!xfs_iext_next_extent(ifp, &icur, &got)) |
86f12ab0 | 534 | return; |
ef473667 DW |
535 | } |
536 | ||
86f12ab0 CH |
537 | if (got.br_startoff >= imap->br_startoff + imap->br_blockcount) |
538 | return; | |
ef473667 | 539 | |
86f12ab0 | 540 | imap->br_blockcount = got.br_startoff - imap->br_startoff; |
ef473667 | 541 | trace_xfs_reflink_trim_irec(ip, imap); |
ef473667 | 542 | } |
43caeb18 DW |
543 | |
544 | /* | |
3802a345 CH |
545 | * Cancel CoW reservations for some block range of an inode. |
546 | * | |
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. | |
43caeb18 DW |
549 | */ |
550 | int | |
551 | xfs_reflink_cancel_cow_blocks( | |
552 | struct xfs_inode *ip, | |
553 | struct xfs_trans **tpp, | |
554 | xfs_fileoff_t offset_fsb, | |
3802a345 CH |
555 | xfs_fileoff_t end_fsb, |
556 | bool cancel_real) | |
43caeb18 | 557 | { |
3e0ee78f | 558 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
df5ab1b5 | 559 | struct xfs_bmbt_irec got, del; |
b2b1712a | 560 | struct xfs_iext_cursor icur; |
43caeb18 DW |
561 | xfs_fsblock_t firstfsb; |
562 | struct xfs_defer_ops dfops; | |
df5ab1b5 | 563 | int error = 0; |
43caeb18 DW |
564 | |
565 | if (!xfs_is_reflink_inode(ip)) | |
566 | return 0; | |
41caabd0 | 567 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) |
3e0ee78f | 568 | return 0; |
43caeb18 | 569 | |
41caabd0 CH |
570 | /* Walk backwards until we're out of the I/O range... */ |
571 | while (got.br_startoff + got.br_blockcount > offset_fsb) { | |
3e0ee78f CH |
572 | del = got; |
573 | xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb); | |
41caabd0 CH |
574 | |
575 | /* Extent delete may have bumped ext forward */ | |
576 | if (!del.br_blockcount) { | |
577 | xfs_iext_prev(ifp, &icur); | |
578 | goto next_extent; | |
579 | } | |
580 | ||
3e0ee78f | 581 | trace_xfs_reflink_cancel_cow(ip, &del); |
43caeb18 | 582 | |
3e0ee78f CH |
583 | if (isnullstartblock(del.br_startblock)) { |
584 | error = xfs_bmap_del_extent_delay(ip, XFS_COW_FORK, | |
b2b1712a | 585 | &icur, &got, &del); |
43caeb18 DW |
586 | if (error) |
587 | break; | |
3802a345 | 588 | } else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) { |
43caeb18 DW |
589 | xfs_trans_ijoin(*tpp, ip, 0); |
590 | xfs_defer_init(&dfops, &firstfsb); | |
591 | ||
174edb0e DW |
592 | /* Free the CoW orphan record. */ |
593 | error = xfs_refcount_free_cow_extent(ip->i_mount, | |
3e0ee78f CH |
594 | &dfops, del.br_startblock, |
595 | del.br_blockcount); | |
174edb0e DW |
596 | if (error) |
597 | break; | |
598 | ||
43caeb18 | 599 | xfs_bmap_add_free(ip->i_mount, &dfops, |
3e0ee78f | 600 | del.br_startblock, del.br_blockcount, |
43caeb18 DW |
601 | NULL); |
602 | ||
603 | /* Update quota accounting */ | |
604 | xfs_trans_mod_dquot_byino(*tpp, ip, XFS_TRANS_DQ_BCOUNT, | |
3e0ee78f | 605 | -(long)del.br_blockcount); |
43caeb18 DW |
606 | |
607 | /* Roll the transaction */ | |
8ad7c629 CH |
608 | xfs_defer_ijoin(&dfops, ip); |
609 | error = xfs_defer_finish(tpp, &dfops); | |
43caeb18 DW |
610 | if (error) { |
611 | xfs_defer_cancel(&dfops); | |
612 | break; | |
613 | } | |
614 | ||
615 | /* Remove the mapping from the CoW fork. */ | |
b2b1712a | 616 | xfs_bmap_del_extent_cow(ip, &icur, &got, &del); |
9d40fba8 DW |
617 | } else { |
618 | /* Didn't do anything, push cursor back. */ | |
619 | xfs_iext_prev(ifp, &icur); | |
43caeb18 | 620 | } |
41caabd0 CH |
621 | next_extent: |
622 | if (!xfs_iext_get_extent(ifp, &icur, &got)) | |
c17a8ef4 | 623 | break; |
43caeb18 DW |
624 | } |
625 | ||
c17a8ef4 BF |
626 | /* clear tag if cow fork is emptied */ |
627 | if (!ifp->if_bytes) | |
628 | xfs_inode_clear_cowblocks_tag(ip); | |
629 | ||
43caeb18 DW |
630 | return error; |
631 | } | |
632 | ||
633 | /* | |
3802a345 CH |
634 | * Cancel CoW reservations for some byte range of an inode. |
635 | * | |
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. | |
43caeb18 DW |
638 | */ |
639 | int | |
640 | xfs_reflink_cancel_cow_range( | |
641 | struct xfs_inode *ip, | |
642 | xfs_off_t offset, | |
3802a345 CH |
643 | xfs_off_t count, |
644 | bool cancel_real) | |
43caeb18 DW |
645 | { |
646 | struct xfs_trans *tp; | |
647 | xfs_fileoff_t offset_fsb; | |
648 | xfs_fileoff_t end_fsb; | |
649 | int error; | |
650 | ||
651 | trace_xfs_reflink_cancel_cow_range(ip, offset, count); | |
63646fc5 | 652 | ASSERT(xfs_is_reflink_inode(ip)); |
43caeb18 DW |
653 | |
654 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
655 | if (count == NULLFILEOFF) | |
656 | end_fsb = NULLFILEOFF; | |
657 | else | |
658 | end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count); | |
659 | ||
660 | /* Start a rolling transaction to remove the mappings */ | |
661 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write, | |
662 | 0, 0, 0, &tp); | |
663 | if (error) | |
664 | goto out; | |
665 | ||
666 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
667 | xfs_trans_ijoin(tp, ip, 0); | |
668 | ||
669 | /* Scrape out the old CoW reservations */ | |
3802a345 CH |
670 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb, |
671 | cancel_real); | |
43caeb18 DW |
672 | if (error) |
673 | goto out_cancel; | |
674 | ||
675 | error = xfs_trans_commit(tp); | |
676 | ||
677 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
678 | return error; | |
679 | ||
680 | out_cancel: | |
681 | xfs_trans_cancel(tp); | |
682 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
683 | out: | |
684 | trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_); | |
685 | return error; | |
686 | } | |
687 | ||
688 | /* | |
689 | * Remap parts of a file's data fork after a successful CoW. | |
690 | */ | |
691 | int | |
692 | xfs_reflink_end_cow( | |
693 | struct xfs_inode *ip, | |
694 | xfs_off_t offset, | |
695 | xfs_off_t count) | |
696 | { | |
c1112b6e | 697 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
4ab8671c | 698 | struct xfs_bmbt_irec got, del; |
43caeb18 DW |
699 | struct xfs_trans *tp; |
700 | xfs_fileoff_t offset_fsb; | |
701 | xfs_fileoff_t end_fsb; | |
43caeb18 DW |
702 | xfs_fsblock_t firstfsb; |
703 | struct xfs_defer_ops dfops; | |
4ab8671c | 704 | int error; |
43caeb18 | 705 | unsigned int resblks; |
43caeb18 | 706 | xfs_filblks_t rlen; |
b2b1712a | 707 | struct xfs_iext_cursor icur; |
43caeb18 DW |
708 | |
709 | trace_xfs_reflink_end_cow(ip, offset, count); | |
710 | ||
c1112b6e CH |
711 | /* No COW extents? That's easy! */ |
712 | if (ifp->if_bytes == 0) | |
713 | return 0; | |
714 | ||
43caeb18 DW |
715 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
716 | end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count); | |
43caeb18 | 717 | |
fe0be23e DW |
718 | /* |
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. | |
723 | */ | |
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); | |
728 | ASSERT(0); | |
729 | goto out; | |
730 | } | |
731 | resblks = XFS_NEXTENTADD_SPACE_RES(ip->i_mount, | |
732 | (unsigned int)(end_fsb - offset_fsb), | |
733 | XFS_DATA_FORK); | |
43caeb18 | 734 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write, |
a192de26 | 735 | resblks, 0, XFS_TRANS_RESERVE, &tp); |
43caeb18 DW |
736 | if (error) |
737 | goto out; | |
738 | ||
739 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
740 | xfs_trans_ijoin(tp, ip, 0); | |
741 | ||
dc56015f CH |
742 | /* |
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 | |
745 | * case we are done. | |
746 | */ | |
b2b1712a | 747 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) |
dc56015f | 748 | goto out_cancel; |
43caeb18 | 749 | |
c1112b6e CH |
750 | /* Walk backwards until we're out of the I/O range... */ |
751 | while (got.br_startoff + got.br_blockcount > offset_fsb) { | |
752 | del = got; | |
753 | xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb); | |
754 | ||
b2b1712a | 755 | /* Extent delete may have bumped ext forward */ |
c1112b6e | 756 | if (!del.br_blockcount) { |
b2b1712a | 757 | xfs_iext_prev(ifp, &icur); |
43caeb18 | 758 | goto next_extent; |
c1112b6e CH |
759 | } |
760 | ||
761 | ASSERT(!isnullstartblock(got.br_startblock)); | |
43caeb18 | 762 | |
5eda4300 DW |
763 | /* |
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. | |
767 | */ | |
768 | if (got.br_state == XFS_EXT_UNWRITTEN) { | |
b2b1712a | 769 | xfs_iext_prev(ifp, &icur); |
5eda4300 DW |
770 | goto next_extent; |
771 | } | |
772 | ||
43caeb18 | 773 | /* Unmap the old blocks in the data fork. */ |
c1112b6e CH |
774 | xfs_defer_init(&dfops, &firstfsb); |
775 | rlen = del.br_blockcount; | |
776 | error = __xfs_bunmapi(tp, ip, del.br_startoff, &rlen, 0, 1, | |
777 | &firstfsb, &dfops); | |
778 | if (error) | |
779 | goto out_defer; | |
43caeb18 | 780 | |
c1112b6e CH |
781 | /* Trim the extent to whatever got unmapped. */ |
782 | if (rlen) { | |
783 | xfs_trim_extent(&del, del.br_startoff + rlen, | |
784 | del.br_blockcount - rlen); | |
785 | } | |
786 | trace_xfs_reflink_cow_remap(ip, &del); | |
174edb0e | 787 | |
c1112b6e CH |
788 | /* Free the CoW orphan record. */ |
789 | error = xfs_refcount_free_cow_extent(tp->t_mountp, &dfops, | |
790 | del.br_startblock, del.br_blockcount); | |
791 | if (error) | |
792 | goto out_defer; | |
43caeb18 | 793 | |
c1112b6e CH |
794 | /* Map the new blocks into the data fork. */ |
795 | error = xfs_bmap_map_extent(tp->t_mountp, &dfops, ip, &del); | |
796 | if (error) | |
797 | goto out_defer; | |
43caeb18 | 798 | |
c1112b6e | 799 | /* Remove the mapping from the CoW fork. */ |
b2b1712a | 800 | xfs_bmap_del_extent_cow(ip, &icur, &got, &del); |
c1112b6e | 801 | |
8ad7c629 CH |
802 | xfs_defer_ijoin(&dfops, ip); |
803 | error = xfs_defer_finish(&tp, &dfops); | |
c1112b6e CH |
804 | if (error) |
805 | goto out_defer; | |
43caeb18 | 806 | next_extent: |
b2b1712a | 807 | if (!xfs_iext_get_extent(ifp, &icur, &got)) |
c1112b6e | 808 | break; |
43caeb18 DW |
809 | } |
810 | ||
811 | error = xfs_trans_commit(tp); | |
812 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
813 | if (error) | |
814 | goto out; | |
815 | return 0; | |
816 | ||
817 | out_defer: | |
818 | xfs_defer_cancel(&dfops); | |
e12199f8 | 819 | out_cancel: |
43caeb18 DW |
820 | xfs_trans_cancel(tp); |
821 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
822 | out: | |
823 | trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_); | |
824 | return error; | |
825 | } | |
174edb0e DW |
826 | |
827 | /* | |
828 | * Free leftover CoW reservations that didn't get cleaned out. | |
829 | */ | |
830 | int | |
831 | xfs_reflink_recover_cow( | |
832 | struct xfs_mount *mp) | |
833 | { | |
834 | xfs_agnumber_t agno; | |
835 | int error = 0; | |
836 | ||
837 | if (!xfs_sb_version_hasreflink(&mp->m_sb)) | |
838 | return 0; | |
839 | ||
840 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
841 | error = xfs_refcount_recover_cow_leftovers(mp, agno); | |
842 | if (error) | |
843 | break; | |
844 | } | |
845 | ||
846 | return error; | |
847 | } | |
862bb360 DW |
848 | |
849 | /* | |
850 | * Reflinking (Block) Ranges of Two Files Together | |
851 | * | |
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. | |
854 | * | |
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. | |
858 | * | |
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 | |
862 | * with zero length. | |
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) | |
872 | * | |
873 | * Finally, if the reflink made dest longer, update both the in-core and | |
874 | * on-disk file sizes. | |
875 | * | |
876 | * ASCII Art Demonstration: | |
877 | * | |
878 | * Let's say we want to reflink this source file: | |
879 | * | |
880 | * ----SSSSSSS-SSSSS----SSSSSS (src file) | |
881 | * <--------------------> | |
882 | * | |
883 | * into this destination file: | |
884 | * | |
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. | |
889 | * | |
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. | |
893 | * | |
894 | * ----SSSSSSS-SSSSS----SSSSSS | |
895 | * <-------> | |
896 | * --DDDDD---------DDDDD--DDD | |
897 | * <-------> | |
898 | * | |
899 | * Now remap the source extent into the destination file: | |
900 | * | |
901 | * ----SSSSSSS-SSSSS----SSSSSS | |
902 | * <-------> | |
903 | * --DDDDD--SSSSSSSDDDDD--DDD | |
904 | * <-------> | |
905 | * | |
906 | * Do likewise with the second hole and extent in our range. Holes in the | |
907 | * unmap range don't affect our operation. | |
908 | * | |
909 | * ----SSSSSSS-SSSSS----SSSSSS | |
910 | * <----> | |
911 | * --DDDDD--SSSSSSS-SSSSS-DDD | |
912 | * <----> | |
913 | * | |
914 | * Finally, unmap and remap part of the third extent. This will increase the | |
915 | * size of the destination file. | |
916 | * | |
917 | * ----SSSSSSS-SSSSS----SSSSSS | |
918 | * <-----> | |
919 | * --DDDDD--SSSSSSS-SSSSS----SSS | |
920 | * <-----> | |
921 | * | |
922 | * Once we update the destination file's i_size, we're done. | |
923 | */ | |
924 | ||
925 | /* | |
926 | * Ensure the reflink bit is set in both inodes. | |
927 | */ | |
928 | STATIC int | |
929 | xfs_reflink_set_inode_flag( | |
930 | struct xfs_inode *src, | |
931 | struct xfs_inode *dest) | |
932 | { | |
933 | struct xfs_mount *mp = src->i_mount; | |
934 | int error; | |
935 | struct xfs_trans *tp; | |
936 | ||
937 | if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest)) | |
938 | return 0; | |
939 | ||
940 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); | |
941 | if (error) | |
942 | goto out_error; | |
943 | ||
944 | /* Lock both files against IO */ | |
945 | if (src->i_ino == dest->i_ino) | |
946 | xfs_ilock(src, XFS_ILOCK_EXCL); | |
947 | else | |
948 | xfs_lock_two_inodes(src, dest, XFS_ILOCK_EXCL); | |
949 | ||
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); | |
956 | } else | |
957 | xfs_iunlock(src, XFS_ILOCK_EXCL); | |
958 | ||
959 | if (src->i_ino == dest->i_ino) | |
960 | goto commit_flags; | |
961 | ||
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); | |
968 | } else | |
969 | xfs_iunlock(dest, XFS_ILOCK_EXCL); | |
970 | ||
971 | commit_flags: | |
972 | error = xfs_trans_commit(tp); | |
973 | if (error) | |
974 | goto out_error; | |
975 | return error; | |
976 | ||
977 | out_error: | |
978 | trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_); | |
979 | return error; | |
980 | } | |
981 | ||
982 | /* | |
f7ca3522 | 983 | * Update destination inode size & cowextsize hint, if necessary. |
862bb360 DW |
984 | */ |
985 | STATIC int | |
986 | xfs_reflink_update_dest( | |
987 | struct xfs_inode *dest, | |
f7ca3522 | 988 | xfs_off_t newlen, |
c5ecb423 CH |
989 | xfs_extlen_t cowextsize, |
990 | bool is_dedupe) | |
862bb360 DW |
991 | { |
992 | struct xfs_mount *mp = dest->i_mount; | |
993 | struct xfs_trans *tp; | |
994 | int error; | |
995 | ||
c5ecb423 | 996 | if (is_dedupe && newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0) |
862bb360 DW |
997 | return 0; |
998 | ||
999 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); | |
1000 | if (error) | |
1001 | goto out_error; | |
1002 | ||
1003 | xfs_ilock(dest, XFS_ILOCK_EXCL); | |
1004 | xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL); | |
1005 | ||
f7ca3522 DW |
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; | |
1010 | } | |
1011 | ||
1012 | if (cowextsize) { | |
1013 | dest->i_d.di_cowextsize = cowextsize; | |
1014 | dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; | |
1015 | } | |
1016 | ||
c5ecb423 CH |
1017 | if (!is_dedupe) { |
1018 | xfs_trans_ichgtime(tp, dest, | |
1019 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1020 | } | |
862bb360 DW |
1021 | xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE); |
1022 | ||
1023 | error = xfs_trans_commit(tp); | |
1024 | if (error) | |
1025 | goto out_error; | |
1026 | return error; | |
1027 | ||
1028 | out_error: | |
1029 | trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_); | |
1030 | return error; | |
1031 | } | |
1032 | ||
6fa164b8 DW |
1033 | /* |
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 | |
1037 | * btree reserves. | |
1038 | */ | |
1039 | static int | |
1040 | xfs_reflink_ag_has_free_space( | |
1041 | struct xfs_mount *mp, | |
1042 | xfs_agnumber_t agno) | |
1043 | { | |
1044 | struct xfs_perag *pag; | |
1045 | int error = 0; | |
1046 | ||
1047 | if (!xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
1048 | return 0; | |
1049 | ||
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)) | |
1053 | error = -ENOSPC; | |
1054 | xfs_perag_put(pag); | |
1055 | return error; | |
1056 | } | |
1057 | ||
862bb360 DW |
1058 | /* |
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. | |
1062 | */ | |
1063 | STATIC int | |
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) | |
1069 | { | |
1070 | struct xfs_mount *mp = ip->i_mount; | |
9c4f29d3 | 1071 | bool real_extent = xfs_bmap_is_real_extent(irec); |
862bb360 DW |
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; | |
862bb360 DW |
1077 | xfs_filblks_t rlen; |
1078 | xfs_filblks_t unmap_len; | |
1079 | xfs_off_t newlen; | |
1080 | int error; | |
1081 | ||
1082 | unmap_len = irec->br_startoff + irec->br_blockcount - destoff; | |
1083 | trace_xfs_reflink_punch_range(ip, destoff, unmap_len); | |
1084 | ||
6fa164b8 DW |
1085 | /* No reflinking if we're low on space */ |
1086 | if (real_extent) { | |
1087 | error = xfs_reflink_ag_has_free_space(mp, | |
1088 | XFS_FSB_TO_AGNO(mp, irec->br_startblock)); | |
1089 | if (error) | |
1090 | goto out; | |
1091 | } | |
1092 | ||
862bb360 DW |
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); | |
1096 | if (error) | |
1097 | goto out; | |
1098 | ||
1099 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1100 | xfs_trans_ijoin(tp, ip, 0); | |
1101 | ||
1102 | /* If we're not just clearing space, then do we have enough quota? */ | |
1103 | if (real_extent) { | |
1104 | error = xfs_trans_reserve_quota_nblks(tp, ip, | |
1105 | irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS); | |
1106 | if (error) | |
1107 | goto out_cancel; | |
1108 | } | |
1109 | ||
1110 | trace_xfs_reflink_remap(ip, irec->br_startoff, | |
1111 | irec->br_blockcount, irec->br_startblock); | |
1112 | ||
1113 | /* Unmap the old blocks in the data fork. */ | |
1114 | rlen = unmap_len; | |
1115 | while (rlen) { | |
1116 | xfs_defer_init(&dfops, &firstfsb); | |
1117 | error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1, | |
1118 | &firstfsb, &dfops); | |
1119 | if (error) | |
1120 | goto out_defer; | |
1121 | ||
1122 | /* | |
1123 | * Trim the extent to whatever got unmapped. | |
1124 | * Remember, bunmapi works backwards. | |
1125 | */ | |
1126 | uirec.br_startblock = irec->br_startblock + rlen; | |
1127 | uirec.br_startoff = irec->br_startoff + rlen; | |
1128 | uirec.br_blockcount = unmap_len - rlen; | |
1129 | unmap_len = rlen; | |
1130 | ||
1131 | /* If this isn't a real mapping, we're done. */ | |
1132 | if (!real_extent || uirec.br_blockcount == 0) | |
1133 | goto next_extent; | |
1134 | ||
1135 | trace_xfs_reflink_remap(ip, uirec.br_startoff, | |
1136 | uirec.br_blockcount, uirec.br_startblock); | |
1137 | ||
1138 | /* Update the refcount tree */ | |
1139 | error = xfs_refcount_increase_extent(mp, &dfops, &uirec); | |
1140 | if (error) | |
1141 | goto out_defer; | |
1142 | ||
1143 | /* Map the new blocks into the data fork. */ | |
1144 | error = xfs_bmap_map_extent(mp, &dfops, ip, &uirec); | |
1145 | if (error) | |
1146 | goto out_defer; | |
1147 | ||
1148 | /* Update quota accounting. */ | |
1149 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, | |
1150 | uirec.br_blockcount); | |
1151 | ||
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); | |
1161 | } | |
1162 | ||
1163 | next_extent: | |
1164 | /* Process all the deferred stuff. */ | |
8ad7c629 CH |
1165 | xfs_defer_ijoin(&dfops, ip); |
1166 | error = xfs_defer_finish(&tp, &dfops); | |
862bb360 DW |
1167 | if (error) |
1168 | goto out_defer; | |
1169 | } | |
1170 | ||
1171 | error = xfs_trans_commit(tp); | |
1172 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1173 | if (error) | |
1174 | goto out; | |
1175 | return 0; | |
1176 | ||
1177 | out_defer: | |
1178 | xfs_defer_cancel(&dfops); | |
1179 | out_cancel: | |
1180 | xfs_trans_cancel(tp); | |
1181 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1182 | out: | |
1183 | trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_); | |
1184 | return error; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * Iteratively remap one file's extents (and holes) to another's. | |
1189 | */ | |
1190 | STATIC int | |
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, | |
1196 | xfs_filblks_t len, | |
1197 | xfs_off_t new_isize) | |
1198 | { | |
1199 | struct xfs_bmbt_irec imap; | |
1200 | int nimaps; | |
1201 | int error = 0; | |
1202 | xfs_filblks_t range_len; | |
1203 | ||
1204 | /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */ | |
1205 | while (len) { | |
1206 | trace_xfs_reflink_remap_blocks_loop(src, srcoff, len, | |
1207 | dest, destoff); | |
1208 | /* Read extent from the source file */ | |
1209 | nimaps = 1; | |
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); | |
1213 | if (error) | |
1214 | goto err; | |
1215 | ASSERT(nimaps == 1); | |
1216 | ||
1217 | trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE, | |
1218 | &imap); | |
1219 | ||
1220 | /* Translate imap into the destination file. */ | |
1221 | range_len = imap.br_startoff + imap.br_blockcount - srcoff; | |
1222 | imap.br_startoff += destoff - srcoff; | |
1223 | ||
1224 | /* Clear dest from destoff to the end of imap and map it in. */ | |
1225 | error = xfs_reflink_remap_extent(dest, &imap, destoff, | |
1226 | new_isize); | |
1227 | if (error) | |
1228 | goto err; | |
1229 | ||
1230 | if (fatal_signal_pending(current)) { | |
1231 | error = -EINTR; | |
1232 | goto err; | |
1233 | } | |
1234 | ||
1235 | /* Advance drange/srange */ | |
1236 | srcoff += range_len; | |
1237 | destoff += range_len; | |
1238 | len -= range_len; | |
1239 | } | |
1240 | ||
1241 | return 0; | |
1242 | ||
1243 | err: | |
1244 | trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_); | |
1245 | return error; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Link a range of blocks from one file to another. | |
1250 | */ | |
1251 | int | |
1252 | xfs_reflink_remap_range( | |
5faaf4fa CH |
1253 | struct file *file_in, |
1254 | loff_t pos_in, | |
1255 | struct file *file_out, | |
1256 | loff_t pos_out, | |
1257 | u64 len, | |
1258 | bool is_dedupe) | |
862bb360 | 1259 | { |
5faaf4fa CH |
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); | |
862bb360 | 1264 | struct xfs_mount *mp = src->i_mount; |
5faaf4fa | 1265 | bool same_inode = (inode_in == inode_out); |
862bb360 DW |
1266 | xfs_fileoff_t sfsbno, dfsbno; |
1267 | xfs_filblks_t fsblen; | |
f7ca3522 | 1268 | xfs_extlen_t cowextsize; |
5faaf4fa | 1269 | ssize_t ret; |
862bb360 DW |
1270 | |
1271 | if (!xfs_sb_version_hasreflink(&mp->m_sb)) | |
1272 | return -EOPNOTSUPP; | |
1273 | ||
1274 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1275 | return -EIO; | |
1276 | ||
5faaf4fa | 1277 | /* Lock both files against IO */ |
65523218 CH |
1278 | lock_two_nondirectories(inode_in, inode_out); |
1279 | if (same_inode) | |
5faaf4fa | 1280 | xfs_ilock(src, XFS_MMAPLOCK_EXCL); |
65523218 | 1281 | else |
5faaf4fa | 1282 | xfs_lock_two_inodes(src, dest, XFS_MMAPLOCK_EXCL); |
5faaf4fa | 1283 | |
876bec6f | 1284 | /* Check file eligibility and prepare for block sharing. */ |
5faaf4fa | 1285 | ret = -EINVAL; |
862bb360 DW |
1286 | /* Don't reflink realtime inodes */ |
1287 | if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest)) | |
5faaf4fa CH |
1288 | goto out_unlock; |
1289 | ||
1290 | /* Don't share DAX file data for now. */ | |
1291 | if (IS_DAX(inode_in) || IS_DAX(inode_out)) | |
1292 | goto out_unlock; | |
1293 | ||
876bec6f DW |
1294 | ret = vfs_clone_file_prep_inodes(inode_in, pos_in, inode_out, pos_out, |
1295 | &len, is_dedupe); | |
22725ce4 | 1296 | if (ret <= 0) |
5faaf4fa CH |
1297 | goto out_unlock; |
1298 | ||
1299 | trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out); | |
862bb360 | 1300 | |
5c989a0e DW |
1301 | /* |
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 | |
1304 | * their own. | |
1305 | */ | |
1306 | if (xfs_can_free_eofblocks(dest, true)) { | |
1307 | ret = xfs_free_eofblocks(dest); | |
1308 | if (ret) | |
1309 | goto out_unlock; | |
1310 | } | |
1311 | ||
876bec6f | 1312 | /* Set flags and remap blocks. */ |
5faaf4fa CH |
1313 | ret = xfs_reflink_set_inode_flag(src, dest); |
1314 | if (ret) | |
1315 | goto out_unlock; | |
862bb360 | 1316 | |
5faaf4fa CH |
1317 | dfsbno = XFS_B_TO_FSBT(mp, pos_out); |
1318 | sfsbno = XFS_B_TO_FSBT(mp, pos_in); | |
862bb360 | 1319 | fsblen = XFS_B_TO_FSB(mp, len); |
5faaf4fa CH |
1320 | ret = xfs_reflink_remap_blocks(src, sfsbno, dest, dfsbno, fsblen, |
1321 | pos_out + len); | |
1322 | if (ret) | |
1323 | goto out_unlock; | |
862bb360 | 1324 | |
876bec6f DW |
1325 | /* Zap any page cache for the destination file's range. */ |
1326 | truncate_inode_pages_range(&inode_out->i_data, pos_out, | |
1327 | PAGE_ALIGN(pos_out + len) - 1); | |
1328 | ||
f7ca3522 DW |
1329 | /* |
1330 | * Carry the cowextsize hint from src to dest if we're sharing the | |
1331 | * entire source file to the entire destination file, the source file | |
1332 | * has a cowextsize hint, and the destination file does not. | |
1333 | */ | |
1334 | cowextsize = 0; | |
5faaf4fa | 1335 | if (pos_in == 0 && len == i_size_read(inode_in) && |
f7ca3522 | 1336 | (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) && |
5faaf4fa | 1337 | pos_out == 0 && len >= i_size_read(inode_out) && |
f7ca3522 DW |
1338 | !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)) |
1339 | cowextsize = src->i_d.di_cowextsize; | |
1340 | ||
c5ecb423 CH |
1341 | ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize, |
1342 | is_dedupe); | |
862bb360 | 1343 | |
5faaf4fa CH |
1344 | out_unlock: |
1345 | xfs_iunlock(src, XFS_MMAPLOCK_EXCL); | |
65523218 | 1346 | if (!same_inode) |
5faaf4fa | 1347 | xfs_iunlock(dest, XFS_MMAPLOCK_EXCL); |
65523218 | 1348 | unlock_two_nondirectories(inode_in, inode_out); |
5faaf4fa CH |
1349 | if (ret) |
1350 | trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_); | |
1351 | return ret; | |
862bb360 | 1352 | } |
98cc2db5 DW |
1353 | |
1354 | /* | |
1355 | * The user wants to preemptively CoW all shared blocks in this file, | |
1356 | * which enables us to turn off the reflink flag. Iterate all | |
1357 | * extents which are not prealloc/delalloc to see which ranges are | |
1358 | * mentioned in the refcount tree, then read those blocks into the | |
1359 | * pagecache, dirty them, fsync them back out, and then we can update | |
1360 | * the inode flag. What happens if we run out of memory? :) | |
1361 | */ | |
1362 | STATIC int | |
1363 | xfs_reflink_dirty_extents( | |
1364 | struct xfs_inode *ip, | |
1365 | xfs_fileoff_t fbno, | |
1366 | xfs_filblks_t end, | |
1367 | xfs_off_t isize) | |
1368 | { | |
1369 | struct xfs_mount *mp = ip->i_mount; | |
1370 | xfs_agnumber_t agno; | |
1371 | xfs_agblock_t agbno; | |
1372 | xfs_extlen_t aglen; | |
1373 | xfs_agblock_t rbno; | |
1374 | xfs_extlen_t rlen; | |
1375 | xfs_off_t fpos; | |
1376 | xfs_off_t flen; | |
1377 | struct xfs_bmbt_irec map[2]; | |
1378 | int nmaps; | |
9780643c | 1379 | int error = 0; |
98cc2db5 DW |
1380 | |
1381 | while (end - fbno > 0) { | |
1382 | nmaps = 1; | |
1383 | /* | |
1384 | * Look for extents in the file. Skip holes, delalloc, or | |
1385 | * unwritten extents; they can't be reflinked. | |
1386 | */ | |
1387 | error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0); | |
1388 | if (error) | |
1389 | goto out; | |
1390 | if (nmaps == 0) | |
1391 | break; | |
9c4f29d3 | 1392 | if (!xfs_bmap_is_real_extent(&map[0])) |
98cc2db5 DW |
1393 | goto next; |
1394 | ||
1395 | map[1] = map[0]; | |
1396 | while (map[1].br_blockcount) { | |
1397 | agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock); | |
1398 | agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock); | |
1399 | aglen = map[1].br_blockcount; | |
1400 | ||
92ff7285 DW |
1401 | error = xfs_reflink_find_shared(mp, NULL, agno, agbno, |
1402 | aglen, &rbno, &rlen, true); | |
98cc2db5 DW |
1403 | if (error) |
1404 | goto out; | |
1405 | if (rbno == NULLAGBLOCK) | |
1406 | break; | |
1407 | ||
1408 | /* Dirty the pages */ | |
1409 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1410 | fpos = XFS_FSB_TO_B(mp, map[1].br_startoff + | |
1411 | (rbno - agbno)); | |
1412 | flen = XFS_FSB_TO_B(mp, rlen); | |
1413 | if (fpos + flen > isize) | |
1414 | flen = isize - fpos; | |
1415 | error = iomap_file_dirty(VFS_I(ip), fpos, flen, | |
1416 | &xfs_iomap_ops); | |
1417 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1418 | if (error) | |
1419 | goto out; | |
1420 | ||
1421 | map[1].br_blockcount -= (rbno - agbno + rlen); | |
1422 | map[1].br_startoff += (rbno - agbno + rlen); | |
1423 | map[1].br_startblock += (rbno - agbno + rlen); | |
1424 | } | |
1425 | ||
1426 | next: | |
1427 | fbno = map[0].br_startoff + map[0].br_blockcount; | |
1428 | } | |
1429 | out: | |
1430 | return error; | |
1431 | } | |
1432 | ||
ea7cdd7b | 1433 | /* Does this inode need the reflink flag? */ |
98cc2db5 | 1434 | int |
ea7cdd7b DW |
1435 | xfs_reflink_inode_has_shared_extents( |
1436 | struct xfs_trans *tp, | |
1437 | struct xfs_inode *ip, | |
1438 | bool *has_shared) | |
98cc2db5 | 1439 | { |
ea7cdd7b DW |
1440 | struct xfs_bmbt_irec got; |
1441 | struct xfs_mount *mp = ip->i_mount; | |
1442 | struct xfs_ifork *ifp; | |
1443 | xfs_agnumber_t agno; | |
1444 | xfs_agblock_t agbno; | |
1445 | xfs_extlen_t aglen; | |
1446 | xfs_agblock_t rbno; | |
1447 | xfs_extlen_t rlen; | |
b2b1712a | 1448 | struct xfs_iext_cursor icur; |
ea7cdd7b DW |
1449 | bool found; |
1450 | int error; | |
98cc2db5 | 1451 | |
ea7cdd7b DW |
1452 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); |
1453 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1454 | error = xfs_iread_extents(tp, ip, XFS_DATA_FORK); | |
98cc2db5 DW |
1455 | if (error) |
1456 | return error; | |
ea7cdd7b | 1457 | } |
98cc2db5 | 1458 | |
ea7cdd7b | 1459 | *has_shared = false; |
b2b1712a | 1460 | found = xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got); |
ea7cdd7b DW |
1461 | while (found) { |
1462 | if (isnullstartblock(got.br_startblock) || | |
1463 | got.br_state != XFS_EXT_NORM) | |
1464 | goto next; | |
1465 | agno = XFS_FSB_TO_AGNO(mp, got.br_startblock); | |
1466 | agbno = XFS_FSB_TO_AGBNO(mp, got.br_startblock); | |
1467 | aglen = got.br_blockcount; | |
98cc2db5 | 1468 | |
ea7cdd7b | 1469 | error = xfs_reflink_find_shared(mp, tp, agno, agbno, aglen, |
024adf48 DW |
1470 | &rbno, &rlen, false); |
1471 | if (error) | |
1472 | return error; | |
1473 | /* Is there still a shared block here? */ | |
ea7cdd7b DW |
1474 | if (rbno != NULLAGBLOCK) { |
1475 | *has_shared = true; | |
024adf48 | 1476 | return 0; |
ea7cdd7b | 1477 | } |
98cc2db5 | 1478 | next: |
b2b1712a | 1479 | found = xfs_iext_next_extent(ifp, &icur, &got); |
98cc2db5 DW |
1480 | } |
1481 | ||
ea7cdd7b DW |
1482 | return 0; |
1483 | } | |
1484 | ||
1485 | /* Clear the inode reflink flag if there are no shared extents. */ | |
1486 | int | |
1487 | xfs_reflink_clear_inode_flag( | |
1488 | struct xfs_inode *ip, | |
1489 | struct xfs_trans **tpp) | |
1490 | { | |
1491 | bool needs_flag; | |
1492 | int error = 0; | |
1493 | ||
1494 | ASSERT(xfs_is_reflink_inode(ip)); | |
1495 | ||
1496 | error = xfs_reflink_inode_has_shared_extents(*tpp, ip, &needs_flag); | |
1497 | if (error || needs_flag) | |
1498 | return error; | |
1499 | ||
98cc2db5 DW |
1500 | /* |
1501 | * We didn't find any shared blocks so turn off the reflink flag. | |
1502 | * First, get rid of any leftover CoW mappings. | |
1503 | */ | |
3802a345 | 1504 | error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF, true); |
98cc2db5 DW |
1505 | if (error) |
1506 | return error; | |
1507 | ||
1508 | /* Clear the inode flag. */ | |
1509 | trace_xfs_reflink_unset_inode_flag(ip); | |
1510 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
83104d44 | 1511 | xfs_inode_clear_cowblocks_tag(ip); |
98cc2db5 DW |
1512 | xfs_trans_ijoin(*tpp, ip, 0); |
1513 | xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE); | |
1514 | ||
1515 | return error; | |
1516 | } | |
1517 | ||
1518 | /* | |
1519 | * Clear the inode reflink flag if there are no shared extents and the size | |
1520 | * hasn't changed. | |
1521 | */ | |
1522 | STATIC int | |
1523 | xfs_reflink_try_clear_inode_flag( | |
97a1b87e | 1524 | struct xfs_inode *ip) |
98cc2db5 DW |
1525 | { |
1526 | struct xfs_mount *mp = ip->i_mount; | |
1527 | struct xfs_trans *tp; | |
1528 | int error = 0; | |
1529 | ||
1530 | /* Start a rolling transaction to remove the mappings */ | |
1531 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp); | |
1532 | if (error) | |
1533 | return error; | |
1534 | ||
1535 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1536 | xfs_trans_ijoin(tp, ip, 0); | |
1537 | ||
98cc2db5 DW |
1538 | error = xfs_reflink_clear_inode_flag(ip, &tp); |
1539 | if (error) | |
1540 | goto cancel; | |
1541 | ||
1542 | error = xfs_trans_commit(tp); | |
1543 | if (error) | |
1544 | goto out; | |
1545 | ||
1546 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1547 | return 0; | |
1548 | cancel: | |
1549 | xfs_trans_cancel(tp); | |
1550 | out: | |
1551 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1552 | return error; | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * Pre-COW all shared blocks within a given byte range of a file and turn off | |
1557 | * the reflink flag if we unshare all of the file's blocks. | |
1558 | */ | |
1559 | int | |
1560 | xfs_reflink_unshare( | |
1561 | struct xfs_inode *ip, | |
1562 | xfs_off_t offset, | |
1563 | xfs_off_t len) | |
1564 | { | |
1565 | struct xfs_mount *mp = ip->i_mount; | |
1566 | xfs_fileoff_t fbno; | |
1567 | xfs_filblks_t end; | |
1568 | xfs_off_t isize; | |
1569 | int error; | |
1570 | ||
1571 | if (!xfs_is_reflink_inode(ip)) | |
1572 | return 0; | |
1573 | ||
1574 | trace_xfs_reflink_unshare(ip, offset, len); | |
1575 | ||
1576 | inode_dio_wait(VFS_I(ip)); | |
1577 | ||
1578 | /* Try to CoW the selected ranges */ | |
1579 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
97a1b87e | 1580 | fbno = XFS_B_TO_FSBT(mp, offset); |
98cc2db5 DW |
1581 | isize = i_size_read(VFS_I(ip)); |
1582 | end = XFS_B_TO_FSB(mp, offset + len); | |
1583 | error = xfs_reflink_dirty_extents(ip, fbno, end, isize); | |
1584 | if (error) | |
1585 | goto out_unlock; | |
1586 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1587 | ||
1588 | /* Wait for the IO to finish */ | |
1589 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); | |
1590 | if (error) | |
1591 | goto out; | |
1592 | ||
97a1b87e DW |
1593 | /* Turn off the reflink flag if possible. */ |
1594 | error = xfs_reflink_try_clear_inode_flag(ip); | |
1595 | if (error) | |
1596 | goto out; | |
98cc2db5 DW |
1597 | |
1598 | return 0; | |
1599 | ||
1600 | out_unlock: | |
1601 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1602 | out: | |
1603 | trace_xfs_reflink_unshare_error(ip, error, _RET_IP_); | |
1604 | return error; | |
1605 | } |