]>
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 | |
2a06705c | 319 | trace_xfs_reflink_cow_alloc(ip, &got); |
3ba020be | 320 | return 0; |
2a06705c | 321 | } |
ef473667 | 322 | |
5eda4300 DW |
323 | /* Convert part of an unwritten CoW extent to a real one. */ |
324 | STATIC int | |
325 | xfs_reflink_convert_cow_extent( | |
326 | struct xfs_inode *ip, | |
327 | struct xfs_bmbt_irec *imap, | |
328 | xfs_fileoff_t offset_fsb, | |
329 | xfs_filblks_t count_fsb, | |
330 | struct xfs_defer_ops *dfops) | |
331 | { | |
4c1a67bd | 332 | xfs_fsblock_t first_block = NULLFSBLOCK; |
5eda4300 DW |
333 | int nimaps = 1; |
334 | ||
335 | if (imap->br_state == XFS_EXT_NORM) | |
336 | return 0; | |
337 | ||
dcf9585a CH |
338 | xfs_trim_extent(imap, offset_fsb, count_fsb); |
339 | trace_xfs_reflink_convert_cow(ip, imap); | |
340 | if (imap->br_blockcount == 0) | |
5eda4300 | 341 | return 0; |
dcf9585a | 342 | return xfs_bmapi_write(NULL, ip, imap->br_startoff, imap->br_blockcount, |
5eda4300 | 343 | XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT, &first_block, |
dcf9585a | 344 | 0, imap, &nimaps, dfops); |
5eda4300 DW |
345 | } |
346 | ||
347 | /* Convert all of the unwritten CoW extents in a file's range to real ones. */ | |
348 | int | |
349 | xfs_reflink_convert_cow( | |
350 | struct xfs_inode *ip, | |
351 | xfs_off_t offset, | |
352 | xfs_off_t count) | |
353 | { | |
5eda4300 | 354 | struct xfs_mount *mp = ip->i_mount; |
5eda4300 DW |
355 | xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); |
356 | xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count); | |
b121459c CH |
357 | xfs_filblks_t count_fsb = end_fsb - offset_fsb; |
358 | struct xfs_bmbt_irec imap; | |
359 | struct xfs_defer_ops dfops; | |
360 | xfs_fsblock_t first_block = NULLFSBLOCK; | |
361 | int nimaps = 1, error = 0; | |
5eda4300 | 362 | |
b121459c | 363 | ASSERT(count != 0); |
5eda4300 | 364 | |
b121459c CH |
365 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
366 | error = xfs_bmapi_write(NULL, ip, offset_fsb, count_fsb, | |
367 | XFS_BMAPI_COWFORK | XFS_BMAPI_CONVERT | | |
368 | XFS_BMAPI_CONVERT_ONLY, &first_block, 0, &imap, &nimaps, | |
369 | &dfops); | |
5eda4300 DW |
370 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
371 | return error; | |
372 | } | |
373 | ||
0613f16c | 374 | /* Allocate all CoW reservations covering a range of blocks in a file. */ |
3c68d44a CH |
375 | int |
376 | xfs_reflink_allocate_cow( | |
0613f16c | 377 | struct xfs_inode *ip, |
3c68d44a CH |
378 | struct xfs_bmbt_irec *imap, |
379 | bool *shared, | |
380 | uint *lockmode) | |
0613f16c DW |
381 | { |
382 | struct xfs_mount *mp = ip->i_mount; | |
3c68d44a CH |
383 | xfs_fileoff_t offset_fsb = imap->br_startoff; |
384 | xfs_filblks_t count_fsb = imap->br_blockcount; | |
385 | struct xfs_bmbt_irec got; | |
0613f16c | 386 | struct xfs_defer_ops dfops; |
3c68d44a | 387 | struct xfs_trans *tp = NULL; |
0613f16c | 388 | xfs_fsblock_t first_block; |
3c68d44a CH |
389 | int nimaps, error = 0; |
390 | bool trimmed; | |
a14234c7 | 391 | xfs_filblks_t resaligned; |
3c68d44a | 392 | xfs_extlen_t resblks = 0; |
b2b1712a | 393 | struct xfs_iext_cursor icur; |
0613f16c | 394 | |
3c68d44a CH |
395 | retry: |
396 | ASSERT(xfs_is_reflink_inode(ip)); | |
397 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED)); | |
0613f16c | 398 | |
a14234c7 CH |
399 | /* |
400 | * Even if the extent is not shared we might have a preallocation for | |
401 | * it in the COW fork. If so use it. | |
402 | */ | |
b2b1712a | 403 | if (xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got) && |
3c68d44a CH |
404 | got.br_startoff <= offset_fsb) { |
405 | *shared = true; | |
406 | ||
a14234c7 CH |
407 | /* If we have a real allocation in the COW fork we're done. */ |
408 | if (!isnullstartblock(got.br_startblock)) { | |
3c68d44a CH |
409 | xfs_trim_extent(&got, offset_fsb, count_fsb); |
410 | *imap = got; | |
411 | goto convert; | |
a14234c7 | 412 | } |
3c68d44a CH |
413 | |
414 | xfs_trim_extent(imap, got.br_startoff, got.br_blockcount); | |
a14234c7 | 415 | } else { |
3c68d44a CH |
416 | error = xfs_reflink_trim_around_shared(ip, imap, shared, &trimmed); |
417 | if (error || !*shared) | |
418 | goto out; | |
419 | } | |
a14234c7 | 420 | |
3c68d44a CH |
421 | if (!tp) { |
422 | resaligned = xfs_aligned_fsb_count(imap->br_startoff, | |
423 | imap->br_blockcount, xfs_get_cowextsz_hint(ip)); | |
424 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); | |
3ba020be | 425 | |
3c68d44a CH |
426 | xfs_iunlock(ip, *lockmode); |
427 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp); | |
428 | *lockmode = XFS_ILOCK_EXCL; | |
429 | xfs_ilock(ip, *lockmode); | |
a14234c7 | 430 | |
3c68d44a CH |
431 | if (error) |
432 | return error; | |
433 | ||
434 | error = xfs_qm_dqattach_locked(ip, 0); | |
435 | if (error) | |
436 | goto out; | |
437 | goto retry; | |
a14234c7 CH |
438 | } |
439 | ||
440 | error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0, | |
441 | XFS_QMOPT_RES_REGBLKS); | |
0613f16c | 442 | if (error) |
3c68d44a | 443 | goto out; |
0613f16c | 444 | |
a14234c7 CH |
445 | xfs_trans_ijoin(tp, ip, 0); |
446 | ||
447 | xfs_defer_init(&dfops, &first_block); | |
448 | nimaps = 1; | |
0613f16c | 449 | |
5eda4300 | 450 | /* Allocate the entire reservation as unwritten blocks. */ |
3c68d44a | 451 | error = xfs_bmapi_write(tp, ip, imap->br_startoff, imap->br_blockcount, |
5eda4300 | 452 | XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, &first_block, |
3c68d44a | 453 | resblks, imap, &nimaps, &dfops); |
0613f16c | 454 | if (error) |
a14234c7 | 455 | goto out_bmap_cancel; |
0613f16c | 456 | |
86d692bf DW |
457 | xfs_inode_set_cowblocks_tag(ip); |
458 | ||
5eda4300 | 459 | /* Finish up. */ |
8ad7c629 | 460 | error = xfs_defer_finish(&tp, &dfops); |
0613f16c | 461 | if (error) |
a14234c7 | 462 | goto out_bmap_cancel; |
0613f16c DW |
463 | |
464 | error = xfs_trans_commit(tp); | |
a14234c7 | 465 | if (error) |
3c68d44a CH |
466 | return error; |
467 | convert: | |
468 | return xfs_reflink_convert_cow_extent(ip, imap, offset_fsb, count_fsb, | |
469 | &dfops); | |
a14234c7 | 470 | out_bmap_cancel: |
0613f16c | 471 | xfs_defer_cancel(&dfops); |
a14234c7 CH |
472 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)resblks, 0, |
473 | XFS_QMOPT_RES_REGBLKS); | |
3c68d44a CH |
474 | out: |
475 | if (tp) | |
476 | xfs_trans_cancel(tp); | |
477 | return error; | |
0613f16c DW |
478 | } |
479 | ||
ef473667 | 480 | /* |
092d5d9d | 481 | * Find the CoW reservation for a given byte offset of a file. |
ef473667 DW |
482 | */ |
483 | bool | |
484 | xfs_reflink_find_cow_mapping( | |
485 | struct xfs_inode *ip, | |
486 | xfs_off_t offset, | |
092d5d9d | 487 | struct xfs_bmbt_irec *imap) |
ef473667 | 488 | { |
092d5d9d CH |
489 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
490 | xfs_fileoff_t offset_fsb; | |
491 | struct xfs_bmbt_irec got; | |
b2b1712a | 492 | struct xfs_iext_cursor icur; |
ef473667 DW |
493 | |
494 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED)); | |
ef473667 | 495 | |
73353f48 DW |
496 | if (!xfs_is_reflink_inode(ip)) |
497 | return false; | |
092d5d9d | 498 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
b2b1712a | 499 | if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got)) |
ef473667 | 500 | return false; |
092d5d9d | 501 | if (got.br_startoff > offset_fsb) |
ef473667 DW |
502 | return false; |
503 | ||
504 | trace_xfs_reflink_find_cow_mapping(ip, offset, 1, XFS_IO_OVERWRITE, | |
092d5d9d CH |
505 | &got); |
506 | *imap = got; | |
ef473667 DW |
507 | return true; |
508 | } | |
509 | ||
510 | /* | |
511 | * Trim an extent to end at the next CoW reservation past offset_fsb. | |
512 | */ | |
86f12ab0 | 513 | void |
ef473667 DW |
514 | xfs_reflink_trim_irec_to_next_cow( |
515 | struct xfs_inode *ip, | |
516 | xfs_fileoff_t offset_fsb, | |
517 | struct xfs_bmbt_irec *imap) | |
518 | { | |
86f12ab0 CH |
519 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
520 | struct xfs_bmbt_irec got; | |
b2b1712a | 521 | struct xfs_iext_cursor icur; |
ef473667 DW |
522 | |
523 | if (!xfs_is_reflink_inode(ip)) | |
86f12ab0 | 524 | return; |
ef473667 DW |
525 | |
526 | /* Find the extent in the CoW fork. */ | |
b2b1712a | 527 | if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got)) |
86f12ab0 | 528 | return; |
ef473667 DW |
529 | |
530 | /* This is the extent before; try sliding up one. */ | |
86f12ab0 | 531 | if (got.br_startoff < offset_fsb) { |
b2b1712a | 532 | if (!xfs_iext_next_extent(ifp, &icur, &got)) |
86f12ab0 | 533 | return; |
ef473667 DW |
534 | } |
535 | ||
86f12ab0 CH |
536 | if (got.br_startoff >= imap->br_startoff + imap->br_blockcount) |
537 | return; | |
ef473667 | 538 | |
86f12ab0 | 539 | imap->br_blockcount = got.br_startoff - imap->br_startoff; |
ef473667 | 540 | trace_xfs_reflink_trim_irec(ip, imap); |
ef473667 | 541 | } |
43caeb18 DW |
542 | |
543 | /* | |
3802a345 CH |
544 | * Cancel CoW reservations for some block range of an inode. |
545 | * | |
546 | * If cancel_real is true this function cancels all COW fork extents for the | |
547 | * inode; if cancel_real is false, real extents are not cleared. | |
43caeb18 DW |
548 | */ |
549 | int | |
550 | xfs_reflink_cancel_cow_blocks( | |
551 | struct xfs_inode *ip, | |
552 | struct xfs_trans **tpp, | |
553 | xfs_fileoff_t offset_fsb, | |
3802a345 CH |
554 | xfs_fileoff_t end_fsb, |
555 | bool cancel_real) | |
43caeb18 | 556 | { |
3e0ee78f | 557 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
df5ab1b5 | 558 | struct xfs_bmbt_irec got, del; |
b2b1712a | 559 | struct xfs_iext_cursor icur; |
43caeb18 DW |
560 | xfs_fsblock_t firstfsb; |
561 | struct xfs_defer_ops dfops; | |
df5ab1b5 | 562 | int error = 0; |
43caeb18 DW |
563 | |
564 | if (!xfs_is_reflink_inode(ip)) | |
565 | return 0; | |
41caabd0 | 566 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) |
3e0ee78f | 567 | return 0; |
43caeb18 | 568 | |
41caabd0 CH |
569 | /* Walk backwards until we're out of the I/O range... */ |
570 | while (got.br_startoff + got.br_blockcount > offset_fsb) { | |
3e0ee78f CH |
571 | del = got; |
572 | xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb); | |
41caabd0 CH |
573 | |
574 | /* Extent delete may have bumped ext forward */ | |
575 | if (!del.br_blockcount) { | |
576 | xfs_iext_prev(ifp, &icur); | |
577 | goto next_extent; | |
578 | } | |
579 | ||
3e0ee78f | 580 | trace_xfs_reflink_cancel_cow(ip, &del); |
43caeb18 | 581 | |
3e0ee78f CH |
582 | if (isnullstartblock(del.br_startblock)) { |
583 | error = xfs_bmap_del_extent_delay(ip, XFS_COW_FORK, | |
b2b1712a | 584 | &icur, &got, &del); |
43caeb18 DW |
585 | if (error) |
586 | break; | |
3802a345 | 587 | } else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) { |
43caeb18 DW |
588 | xfs_trans_ijoin(*tpp, ip, 0); |
589 | xfs_defer_init(&dfops, &firstfsb); | |
590 | ||
174edb0e DW |
591 | /* Free the CoW orphan record. */ |
592 | error = xfs_refcount_free_cow_extent(ip->i_mount, | |
3e0ee78f CH |
593 | &dfops, del.br_startblock, |
594 | del.br_blockcount); | |
174edb0e DW |
595 | if (error) |
596 | break; | |
597 | ||
43caeb18 | 598 | xfs_bmap_add_free(ip->i_mount, &dfops, |
3e0ee78f | 599 | del.br_startblock, del.br_blockcount, |
43caeb18 DW |
600 | NULL); |
601 | ||
602 | /* Update quota accounting */ | |
603 | xfs_trans_mod_dquot_byino(*tpp, ip, XFS_TRANS_DQ_BCOUNT, | |
3e0ee78f | 604 | -(long)del.br_blockcount); |
43caeb18 DW |
605 | |
606 | /* Roll the transaction */ | |
8ad7c629 CH |
607 | xfs_defer_ijoin(&dfops, ip); |
608 | error = xfs_defer_finish(tpp, &dfops); | |
43caeb18 DW |
609 | if (error) { |
610 | xfs_defer_cancel(&dfops); | |
611 | break; | |
612 | } | |
613 | ||
614 | /* Remove the mapping from the CoW fork. */ | |
b2b1712a | 615 | xfs_bmap_del_extent_cow(ip, &icur, &got, &del); |
9d40fba8 DW |
616 | } else { |
617 | /* Didn't do anything, push cursor back. */ | |
618 | xfs_iext_prev(ifp, &icur); | |
43caeb18 | 619 | } |
41caabd0 CH |
620 | next_extent: |
621 | if (!xfs_iext_get_extent(ifp, &icur, &got)) | |
c17a8ef4 | 622 | break; |
43caeb18 DW |
623 | } |
624 | ||
c17a8ef4 BF |
625 | /* clear tag if cow fork is emptied */ |
626 | if (!ifp->if_bytes) | |
627 | xfs_inode_clear_cowblocks_tag(ip); | |
628 | ||
43caeb18 DW |
629 | return error; |
630 | } | |
631 | ||
632 | /* | |
3802a345 CH |
633 | * Cancel CoW reservations for some byte range of an inode. |
634 | * | |
635 | * If cancel_real is true this function cancels all COW fork extents for the | |
636 | * inode; if cancel_real is false, real extents are not cleared. | |
43caeb18 DW |
637 | */ |
638 | int | |
639 | xfs_reflink_cancel_cow_range( | |
640 | struct xfs_inode *ip, | |
641 | xfs_off_t offset, | |
3802a345 CH |
642 | xfs_off_t count, |
643 | bool cancel_real) | |
43caeb18 DW |
644 | { |
645 | struct xfs_trans *tp; | |
646 | xfs_fileoff_t offset_fsb; | |
647 | xfs_fileoff_t end_fsb; | |
648 | int error; | |
649 | ||
650 | trace_xfs_reflink_cancel_cow_range(ip, offset, count); | |
63646fc5 | 651 | ASSERT(xfs_is_reflink_inode(ip)); |
43caeb18 DW |
652 | |
653 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
654 | if (count == NULLFILEOFF) | |
655 | end_fsb = NULLFILEOFF; | |
656 | else | |
657 | end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count); | |
658 | ||
659 | /* Start a rolling transaction to remove the mappings */ | |
660 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write, | |
661 | 0, 0, 0, &tp); | |
662 | if (error) | |
663 | goto out; | |
664 | ||
665 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
666 | xfs_trans_ijoin(tp, ip, 0); | |
667 | ||
668 | /* Scrape out the old CoW reservations */ | |
3802a345 CH |
669 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb, |
670 | cancel_real); | |
43caeb18 DW |
671 | if (error) |
672 | goto out_cancel; | |
673 | ||
674 | error = xfs_trans_commit(tp); | |
675 | ||
676 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
677 | return error; | |
678 | ||
679 | out_cancel: | |
680 | xfs_trans_cancel(tp); | |
681 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
682 | out: | |
683 | trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_); | |
684 | return error; | |
685 | } | |
686 | ||
687 | /* | |
688 | * Remap parts of a file's data fork after a successful CoW. | |
689 | */ | |
690 | int | |
691 | xfs_reflink_end_cow( | |
692 | struct xfs_inode *ip, | |
693 | xfs_off_t offset, | |
694 | xfs_off_t count) | |
695 | { | |
c1112b6e | 696 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
4ab8671c | 697 | struct xfs_bmbt_irec got, del; |
43caeb18 DW |
698 | struct xfs_trans *tp; |
699 | xfs_fileoff_t offset_fsb; | |
700 | xfs_fileoff_t end_fsb; | |
43caeb18 DW |
701 | xfs_fsblock_t firstfsb; |
702 | struct xfs_defer_ops dfops; | |
4ab8671c | 703 | int error; |
43caeb18 | 704 | unsigned int resblks; |
43caeb18 | 705 | xfs_filblks_t rlen; |
b2b1712a | 706 | struct xfs_iext_cursor icur; |
43caeb18 DW |
707 | |
708 | trace_xfs_reflink_end_cow(ip, offset, count); | |
709 | ||
c1112b6e CH |
710 | /* No COW extents? That's easy! */ |
711 | if (ifp->if_bytes == 0) | |
712 | return 0; | |
713 | ||
43caeb18 DW |
714 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
715 | end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count); | |
43caeb18 | 716 | |
fe0be23e DW |
717 | /* |
718 | * Start a rolling transaction to switch the mappings. We're | |
719 | * unlikely ever to have to remap 16T worth of single-block | |
720 | * extents, so just cap the worst case extent count to 2^32-1. | |
721 | * Stick a warning in just in case, and avoid 64-bit division. | |
722 | */ | |
723 | BUILD_BUG_ON(MAX_RW_COUNT > UINT_MAX); | |
724 | if (end_fsb - offset_fsb > UINT_MAX) { | |
725 | error = -EFSCORRUPTED; | |
726 | xfs_force_shutdown(ip->i_mount, SHUTDOWN_CORRUPT_INCORE); | |
727 | ASSERT(0); | |
728 | goto out; | |
729 | } | |
730 | resblks = XFS_NEXTENTADD_SPACE_RES(ip->i_mount, | |
731 | (unsigned int)(end_fsb - offset_fsb), | |
732 | XFS_DATA_FORK); | |
43caeb18 | 733 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write, |
a192de26 | 734 | resblks, 0, XFS_TRANS_RESERVE, &tp); |
43caeb18 DW |
735 | if (error) |
736 | goto out; | |
737 | ||
738 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
739 | xfs_trans_ijoin(tp, ip, 0); | |
740 | ||
dc56015f CH |
741 | /* |
742 | * In case of racing, overlapping AIO writes no COW extents might be | |
743 | * left by the time I/O completes for the loser of the race. In that | |
744 | * case we are done. | |
745 | */ | |
b2b1712a | 746 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) |
dc56015f | 747 | goto out_cancel; |
43caeb18 | 748 | |
c1112b6e CH |
749 | /* Walk backwards until we're out of the I/O range... */ |
750 | while (got.br_startoff + got.br_blockcount > offset_fsb) { | |
751 | del = got; | |
752 | xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb); | |
753 | ||
b2b1712a | 754 | /* Extent delete may have bumped ext forward */ |
c1112b6e | 755 | if (!del.br_blockcount) { |
b2b1712a | 756 | xfs_iext_prev(ifp, &icur); |
43caeb18 | 757 | goto next_extent; |
c1112b6e CH |
758 | } |
759 | ||
760 | ASSERT(!isnullstartblock(got.br_startblock)); | |
43caeb18 | 761 | |
5eda4300 DW |
762 | /* |
763 | * Don't remap unwritten extents; these are | |
764 | * speculatively preallocated CoW extents that have been | |
765 | * allocated but have not yet been involved in a write. | |
766 | */ | |
767 | if (got.br_state == XFS_EXT_UNWRITTEN) { | |
b2b1712a | 768 | xfs_iext_prev(ifp, &icur); |
5eda4300 DW |
769 | goto next_extent; |
770 | } | |
771 | ||
43caeb18 | 772 | /* Unmap the old blocks in the data fork. */ |
c1112b6e CH |
773 | xfs_defer_init(&dfops, &firstfsb); |
774 | rlen = del.br_blockcount; | |
775 | error = __xfs_bunmapi(tp, ip, del.br_startoff, &rlen, 0, 1, | |
776 | &firstfsb, &dfops); | |
777 | if (error) | |
778 | goto out_defer; | |
43caeb18 | 779 | |
c1112b6e CH |
780 | /* Trim the extent to whatever got unmapped. */ |
781 | if (rlen) { | |
782 | xfs_trim_extent(&del, del.br_startoff + rlen, | |
783 | del.br_blockcount - rlen); | |
784 | } | |
785 | trace_xfs_reflink_cow_remap(ip, &del); | |
174edb0e | 786 | |
c1112b6e CH |
787 | /* Free the CoW orphan record. */ |
788 | error = xfs_refcount_free_cow_extent(tp->t_mountp, &dfops, | |
789 | del.br_startblock, del.br_blockcount); | |
790 | if (error) | |
791 | goto out_defer; | |
43caeb18 | 792 | |
c1112b6e CH |
793 | /* Map the new blocks into the data fork. */ |
794 | error = xfs_bmap_map_extent(tp->t_mountp, &dfops, ip, &del); | |
795 | if (error) | |
796 | goto out_defer; | |
43caeb18 | 797 | |
c1112b6e | 798 | /* Remove the mapping from the CoW fork. */ |
b2b1712a | 799 | xfs_bmap_del_extent_cow(ip, &icur, &got, &del); |
c1112b6e | 800 | |
8ad7c629 CH |
801 | xfs_defer_ijoin(&dfops, ip); |
802 | error = xfs_defer_finish(&tp, &dfops); | |
c1112b6e CH |
803 | if (error) |
804 | goto out_defer; | |
43caeb18 | 805 | next_extent: |
b2b1712a | 806 | if (!xfs_iext_get_extent(ifp, &icur, &got)) |
c1112b6e | 807 | break; |
43caeb18 DW |
808 | } |
809 | ||
810 | error = xfs_trans_commit(tp); | |
811 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
812 | if (error) | |
813 | goto out; | |
814 | return 0; | |
815 | ||
816 | out_defer: | |
817 | xfs_defer_cancel(&dfops); | |
e12199f8 | 818 | out_cancel: |
43caeb18 DW |
819 | xfs_trans_cancel(tp); |
820 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
821 | out: | |
822 | trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_); | |
823 | return error; | |
824 | } | |
174edb0e DW |
825 | |
826 | /* | |
827 | * Free leftover CoW reservations that didn't get cleaned out. | |
828 | */ | |
829 | int | |
830 | xfs_reflink_recover_cow( | |
831 | struct xfs_mount *mp) | |
832 | { | |
833 | xfs_agnumber_t agno; | |
834 | int error = 0; | |
835 | ||
836 | if (!xfs_sb_version_hasreflink(&mp->m_sb)) | |
837 | return 0; | |
838 | ||
839 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
840 | error = xfs_refcount_recover_cow_leftovers(mp, agno); | |
841 | if (error) | |
842 | break; | |
843 | } | |
844 | ||
845 | return error; | |
846 | } | |
862bb360 DW |
847 | |
848 | /* | |
849 | * Reflinking (Block) Ranges of Two Files Together | |
850 | * | |
851 | * First, ensure that the reflink flag is set on both inodes. The flag is an | |
852 | * optimization to avoid unnecessary refcount btree lookups in the write path. | |
853 | * | |
854 | * Now we can iteratively remap the range of extents (and holes) in src to the | |
855 | * corresponding ranges in dest. Let drange and srange denote the ranges of | |
856 | * logical blocks in dest and src touched by the reflink operation. | |
857 | * | |
858 | * While the length of drange is greater than zero, | |
859 | * - Read src's bmbt at the start of srange ("imap") | |
860 | * - If imap doesn't exist, make imap appear to start at the end of srange | |
861 | * with zero length. | |
862 | * - If imap starts before srange, advance imap to start at srange. | |
863 | * - If imap goes beyond srange, truncate imap to end at the end of srange. | |
864 | * - Punch (imap start - srange start + imap len) blocks from dest at | |
865 | * offset (drange start). | |
866 | * - If imap points to a real range of pblks, | |
867 | * > Increase the refcount of the imap's pblks | |
868 | * > Map imap's pblks into dest at the offset | |
869 | * (drange start + imap start - srange start) | |
870 | * - Advance drange and srange by (imap start - srange start + imap len) | |
871 | * | |
872 | * Finally, if the reflink made dest longer, update both the in-core and | |
873 | * on-disk file sizes. | |
874 | * | |
875 | * ASCII Art Demonstration: | |
876 | * | |
877 | * Let's say we want to reflink this source file: | |
878 | * | |
879 | * ----SSSSSSS-SSSSS----SSSSSS (src file) | |
880 | * <--------------------> | |
881 | * | |
882 | * into this destination file: | |
883 | * | |
884 | * --DDDDDDDDDDDDDDDDDDD--DDD (dest file) | |
885 | * <--------------------> | |
886 | * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest. | |
887 | * Observe that the range has different logical offsets in either file. | |
888 | * | |
889 | * Consider that the first extent in the source file doesn't line up with our | |
890 | * reflink range. Unmapping and remapping are separate operations, so we can | |
891 | * unmap more blocks from the destination file than we remap. | |
892 | * | |
893 | * ----SSSSSSS-SSSSS----SSSSSS | |
894 | * <-------> | |
895 | * --DDDDD---------DDDDD--DDD | |
896 | * <-------> | |
897 | * | |
898 | * Now remap the source extent into the destination file: | |
899 | * | |
900 | * ----SSSSSSS-SSSSS----SSSSSS | |
901 | * <-------> | |
902 | * --DDDDD--SSSSSSSDDDDD--DDD | |
903 | * <-------> | |
904 | * | |
905 | * Do likewise with the second hole and extent in our range. Holes in the | |
906 | * unmap range don't affect our operation. | |
907 | * | |
908 | * ----SSSSSSS-SSSSS----SSSSSS | |
909 | * <----> | |
910 | * --DDDDD--SSSSSSS-SSSSS-DDD | |
911 | * <----> | |
912 | * | |
913 | * Finally, unmap and remap part of the third extent. This will increase the | |
914 | * size of the destination file. | |
915 | * | |
916 | * ----SSSSSSS-SSSSS----SSSSSS | |
917 | * <-----> | |
918 | * --DDDDD--SSSSSSS-SSSSS----SSS | |
919 | * <-----> | |
920 | * | |
921 | * Once we update the destination file's i_size, we're done. | |
922 | */ | |
923 | ||
924 | /* | |
925 | * Ensure the reflink bit is set in both inodes. | |
926 | */ | |
927 | STATIC int | |
928 | xfs_reflink_set_inode_flag( | |
929 | struct xfs_inode *src, | |
930 | struct xfs_inode *dest) | |
931 | { | |
932 | struct xfs_mount *mp = src->i_mount; | |
933 | int error; | |
934 | struct xfs_trans *tp; | |
935 | ||
936 | if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest)) | |
937 | return 0; | |
938 | ||
939 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); | |
940 | if (error) | |
941 | goto out_error; | |
942 | ||
943 | /* Lock both files against IO */ | |
944 | if (src->i_ino == dest->i_ino) | |
945 | xfs_ilock(src, XFS_ILOCK_EXCL); | |
946 | else | |
947 | xfs_lock_two_inodes(src, dest, XFS_ILOCK_EXCL); | |
948 | ||
949 | if (!xfs_is_reflink_inode(src)) { | |
950 | trace_xfs_reflink_set_inode_flag(src); | |
951 | xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL); | |
952 | src->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK; | |
953 | xfs_trans_log_inode(tp, src, XFS_ILOG_CORE); | |
954 | xfs_ifork_init_cow(src); | |
955 | } else | |
956 | xfs_iunlock(src, XFS_ILOCK_EXCL); | |
957 | ||
958 | if (src->i_ino == dest->i_ino) | |
959 | goto commit_flags; | |
960 | ||
961 | if (!xfs_is_reflink_inode(dest)) { | |
962 | trace_xfs_reflink_set_inode_flag(dest); | |
963 | xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL); | |
964 | dest->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK; | |
965 | xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE); | |
966 | xfs_ifork_init_cow(dest); | |
967 | } else | |
968 | xfs_iunlock(dest, XFS_ILOCK_EXCL); | |
969 | ||
970 | commit_flags: | |
971 | error = xfs_trans_commit(tp); | |
972 | if (error) | |
973 | goto out_error; | |
974 | return error; | |
975 | ||
976 | out_error: | |
977 | trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_); | |
978 | return error; | |
979 | } | |
980 | ||
981 | /* | |
f7ca3522 | 982 | * Update destination inode size & cowextsize hint, if necessary. |
862bb360 DW |
983 | */ |
984 | STATIC int | |
985 | xfs_reflink_update_dest( | |
986 | struct xfs_inode *dest, | |
f7ca3522 | 987 | xfs_off_t newlen, |
c5ecb423 CH |
988 | xfs_extlen_t cowextsize, |
989 | bool is_dedupe) | |
862bb360 DW |
990 | { |
991 | struct xfs_mount *mp = dest->i_mount; | |
992 | struct xfs_trans *tp; | |
993 | int error; | |
994 | ||
c5ecb423 | 995 | if (is_dedupe && newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0) |
862bb360 DW |
996 | return 0; |
997 | ||
998 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); | |
999 | if (error) | |
1000 | goto out_error; | |
1001 | ||
1002 | xfs_ilock(dest, XFS_ILOCK_EXCL); | |
1003 | xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL); | |
1004 | ||
f7ca3522 DW |
1005 | if (newlen > i_size_read(VFS_I(dest))) { |
1006 | trace_xfs_reflink_update_inode_size(dest, newlen); | |
1007 | i_size_write(VFS_I(dest), newlen); | |
1008 | dest->i_d.di_size = newlen; | |
1009 | } | |
1010 | ||
1011 | if (cowextsize) { | |
1012 | dest->i_d.di_cowextsize = cowextsize; | |
1013 | dest->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; | |
1014 | } | |
1015 | ||
c5ecb423 CH |
1016 | if (!is_dedupe) { |
1017 | xfs_trans_ichgtime(tp, dest, | |
1018 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1019 | } | |
862bb360 DW |
1020 | xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE); |
1021 | ||
1022 | error = xfs_trans_commit(tp); | |
1023 | if (error) | |
1024 | goto out_error; | |
1025 | return error; | |
1026 | ||
1027 | out_error: | |
1028 | trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_); | |
1029 | return error; | |
1030 | } | |
1031 | ||
6fa164b8 DW |
1032 | /* |
1033 | * Do we have enough reserve in this AG to handle a reflink? The refcount | |
1034 | * btree already reserved all the space it needs, but the rmap btree can grow | |
1035 | * infinitely, so we won't allow more reflinks when the AG is down to the | |
1036 | * btree reserves. | |
1037 | */ | |
1038 | static int | |
1039 | xfs_reflink_ag_has_free_space( | |
1040 | struct xfs_mount *mp, | |
1041 | xfs_agnumber_t agno) | |
1042 | { | |
1043 | struct xfs_perag *pag; | |
1044 | int error = 0; | |
1045 | ||
1046 | if (!xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
1047 | return 0; | |
1048 | ||
1049 | pag = xfs_perag_get(mp, agno); | |
1050 | if (xfs_ag_resv_critical(pag, XFS_AG_RESV_AGFL) || | |
1051 | xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA)) | |
1052 | error = -ENOSPC; | |
1053 | xfs_perag_put(pag); | |
1054 | return error; | |
1055 | } | |
1056 | ||
862bb360 DW |
1057 | /* |
1058 | * Unmap a range of blocks from a file, then map other blocks into the hole. | |
1059 | * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount). | |
1060 | * The extent irec is mapped into dest at irec->br_startoff. | |
1061 | */ | |
1062 | STATIC int | |
1063 | xfs_reflink_remap_extent( | |
1064 | struct xfs_inode *ip, | |
1065 | struct xfs_bmbt_irec *irec, | |
1066 | xfs_fileoff_t destoff, | |
1067 | xfs_off_t new_isize) | |
1068 | { | |
1069 | struct xfs_mount *mp = ip->i_mount; | |
9c4f29d3 | 1070 | bool real_extent = xfs_bmap_is_real_extent(irec); |
862bb360 DW |
1071 | struct xfs_trans *tp; |
1072 | xfs_fsblock_t firstfsb; | |
1073 | unsigned int resblks; | |
1074 | struct xfs_defer_ops dfops; | |
1075 | struct xfs_bmbt_irec uirec; | |
862bb360 DW |
1076 | xfs_filblks_t rlen; |
1077 | xfs_filblks_t unmap_len; | |
1078 | xfs_off_t newlen; | |
1079 | int error; | |
1080 | ||
1081 | unmap_len = irec->br_startoff + irec->br_blockcount - destoff; | |
1082 | trace_xfs_reflink_punch_range(ip, destoff, unmap_len); | |
1083 | ||
6fa164b8 DW |
1084 | /* No reflinking if we're low on space */ |
1085 | if (real_extent) { | |
1086 | error = xfs_reflink_ag_has_free_space(mp, | |
1087 | XFS_FSB_TO_AGNO(mp, irec->br_startblock)); | |
1088 | if (error) | |
1089 | goto out; | |
1090 | } | |
1091 | ||
862bb360 DW |
1092 | /* Start a rolling transaction to switch the mappings */ |
1093 | resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK); | |
1094 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp); | |
1095 | if (error) | |
1096 | goto out; | |
1097 | ||
1098 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1099 | xfs_trans_ijoin(tp, ip, 0); | |
1100 | ||
1101 | /* If we're not just clearing space, then do we have enough quota? */ | |
1102 | if (real_extent) { | |
1103 | error = xfs_trans_reserve_quota_nblks(tp, ip, | |
1104 | irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS); | |
1105 | if (error) | |
1106 | goto out_cancel; | |
1107 | } | |
1108 | ||
1109 | trace_xfs_reflink_remap(ip, irec->br_startoff, | |
1110 | irec->br_blockcount, irec->br_startblock); | |
1111 | ||
1112 | /* Unmap the old blocks in the data fork. */ | |
1113 | rlen = unmap_len; | |
1114 | while (rlen) { | |
1115 | xfs_defer_init(&dfops, &firstfsb); | |
1116 | error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1, | |
1117 | &firstfsb, &dfops); | |
1118 | if (error) | |
1119 | goto out_defer; | |
1120 | ||
1121 | /* | |
1122 | * Trim the extent to whatever got unmapped. | |
1123 | * Remember, bunmapi works backwards. | |
1124 | */ | |
1125 | uirec.br_startblock = irec->br_startblock + rlen; | |
1126 | uirec.br_startoff = irec->br_startoff + rlen; | |
1127 | uirec.br_blockcount = unmap_len - rlen; | |
1128 | unmap_len = rlen; | |
1129 | ||
1130 | /* If this isn't a real mapping, we're done. */ | |
1131 | if (!real_extent || uirec.br_blockcount == 0) | |
1132 | goto next_extent; | |
1133 | ||
1134 | trace_xfs_reflink_remap(ip, uirec.br_startoff, | |
1135 | uirec.br_blockcount, uirec.br_startblock); | |
1136 | ||
1137 | /* Update the refcount tree */ | |
1138 | error = xfs_refcount_increase_extent(mp, &dfops, &uirec); | |
1139 | if (error) | |
1140 | goto out_defer; | |
1141 | ||
1142 | /* Map the new blocks into the data fork. */ | |
1143 | error = xfs_bmap_map_extent(mp, &dfops, ip, &uirec); | |
1144 | if (error) | |
1145 | goto out_defer; | |
1146 | ||
1147 | /* Update quota accounting. */ | |
1148 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, | |
1149 | uirec.br_blockcount); | |
1150 | ||
1151 | /* Update dest isize if needed. */ | |
1152 | newlen = XFS_FSB_TO_B(mp, | |
1153 | uirec.br_startoff + uirec.br_blockcount); | |
1154 | newlen = min_t(xfs_off_t, newlen, new_isize); | |
1155 | if (newlen > i_size_read(VFS_I(ip))) { | |
1156 | trace_xfs_reflink_update_inode_size(ip, newlen); | |
1157 | i_size_write(VFS_I(ip), newlen); | |
1158 | ip->i_d.di_size = newlen; | |
1159 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1160 | } | |
1161 | ||
1162 | next_extent: | |
1163 | /* Process all the deferred stuff. */ | |
8ad7c629 CH |
1164 | xfs_defer_ijoin(&dfops, ip); |
1165 | error = xfs_defer_finish(&tp, &dfops); | |
862bb360 DW |
1166 | if (error) |
1167 | goto out_defer; | |
1168 | } | |
1169 | ||
1170 | error = xfs_trans_commit(tp); | |
1171 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1172 | if (error) | |
1173 | goto out; | |
1174 | return 0; | |
1175 | ||
1176 | out_defer: | |
1177 | xfs_defer_cancel(&dfops); | |
1178 | out_cancel: | |
1179 | xfs_trans_cancel(tp); | |
1180 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1181 | out: | |
1182 | trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_); | |
1183 | return error; | |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * Iteratively remap one file's extents (and holes) to another's. | |
1188 | */ | |
1189 | STATIC int | |
1190 | xfs_reflink_remap_blocks( | |
1191 | struct xfs_inode *src, | |
1192 | xfs_fileoff_t srcoff, | |
1193 | struct xfs_inode *dest, | |
1194 | xfs_fileoff_t destoff, | |
1195 | xfs_filblks_t len, | |
1196 | xfs_off_t new_isize) | |
1197 | { | |
1198 | struct xfs_bmbt_irec imap; | |
1199 | int nimaps; | |
1200 | int error = 0; | |
1201 | xfs_filblks_t range_len; | |
1202 | ||
1203 | /* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */ | |
1204 | while (len) { | |
1205 | trace_xfs_reflink_remap_blocks_loop(src, srcoff, len, | |
1206 | dest, destoff); | |
1207 | /* Read extent from the source file */ | |
1208 | nimaps = 1; | |
1209 | xfs_ilock(src, XFS_ILOCK_EXCL); | |
1210 | error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0); | |
1211 | xfs_iunlock(src, XFS_ILOCK_EXCL); | |
1212 | if (error) | |
1213 | goto err; | |
1214 | ASSERT(nimaps == 1); | |
1215 | ||
1216 | trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_IO_OVERWRITE, | |
1217 | &imap); | |
1218 | ||
1219 | /* Translate imap into the destination file. */ | |
1220 | range_len = imap.br_startoff + imap.br_blockcount - srcoff; | |
1221 | imap.br_startoff += destoff - srcoff; | |
1222 | ||
1223 | /* Clear dest from destoff to the end of imap and map it in. */ | |
1224 | error = xfs_reflink_remap_extent(dest, &imap, destoff, | |
1225 | new_isize); | |
1226 | if (error) | |
1227 | goto err; | |
1228 | ||
1229 | if (fatal_signal_pending(current)) { | |
1230 | error = -EINTR; | |
1231 | goto err; | |
1232 | } | |
1233 | ||
1234 | /* Advance drange/srange */ | |
1235 | srcoff += range_len; | |
1236 | destoff += range_len; | |
1237 | len -= range_len; | |
1238 | } | |
1239 | ||
1240 | return 0; | |
1241 | ||
1242 | err: | |
1243 | trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_); | |
1244 | return error; | |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * Link a range of blocks from one file to another. | |
1249 | */ | |
1250 | int | |
1251 | xfs_reflink_remap_range( | |
5faaf4fa CH |
1252 | struct file *file_in, |
1253 | loff_t pos_in, | |
1254 | struct file *file_out, | |
1255 | loff_t pos_out, | |
1256 | u64 len, | |
1257 | bool is_dedupe) | |
862bb360 | 1258 | { |
5faaf4fa CH |
1259 | struct inode *inode_in = file_inode(file_in); |
1260 | struct xfs_inode *src = XFS_I(inode_in); | |
1261 | struct inode *inode_out = file_inode(file_out); | |
1262 | struct xfs_inode *dest = XFS_I(inode_out); | |
862bb360 | 1263 | struct xfs_mount *mp = src->i_mount; |
5faaf4fa | 1264 | bool same_inode = (inode_in == inode_out); |
862bb360 DW |
1265 | xfs_fileoff_t sfsbno, dfsbno; |
1266 | xfs_filblks_t fsblen; | |
f7ca3522 | 1267 | xfs_extlen_t cowextsize; |
5faaf4fa | 1268 | ssize_t ret; |
862bb360 DW |
1269 | |
1270 | if (!xfs_sb_version_hasreflink(&mp->m_sb)) | |
1271 | return -EOPNOTSUPP; | |
1272 | ||
1273 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1274 | return -EIO; | |
1275 | ||
5faaf4fa | 1276 | /* Lock both files against IO */ |
65523218 CH |
1277 | lock_two_nondirectories(inode_in, inode_out); |
1278 | if (same_inode) | |
5faaf4fa | 1279 | xfs_ilock(src, XFS_MMAPLOCK_EXCL); |
65523218 | 1280 | else |
5faaf4fa | 1281 | xfs_lock_two_inodes(src, dest, XFS_MMAPLOCK_EXCL); |
5faaf4fa | 1282 | |
876bec6f | 1283 | /* Check file eligibility and prepare for block sharing. */ |
5faaf4fa | 1284 | ret = -EINVAL; |
862bb360 DW |
1285 | /* Don't reflink realtime inodes */ |
1286 | if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest)) | |
5faaf4fa CH |
1287 | goto out_unlock; |
1288 | ||
1289 | /* Don't share DAX file data for now. */ | |
1290 | if (IS_DAX(inode_in) || IS_DAX(inode_out)) | |
1291 | goto out_unlock; | |
1292 | ||
876bec6f DW |
1293 | ret = vfs_clone_file_prep_inodes(inode_in, pos_in, inode_out, pos_out, |
1294 | &len, is_dedupe); | |
22725ce4 | 1295 | if (ret <= 0) |
5faaf4fa CH |
1296 | goto out_unlock; |
1297 | ||
1298 | trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out); | |
862bb360 | 1299 | |
5c989a0e DW |
1300 | /* |
1301 | * Clear out post-eof preallocations because we don't have page cache | |
1302 | * backing the delayed allocations and they'll never get freed on | |
1303 | * their own. | |
1304 | */ | |
1305 | if (xfs_can_free_eofblocks(dest, true)) { | |
1306 | ret = xfs_free_eofblocks(dest); | |
1307 | if (ret) | |
1308 | goto out_unlock; | |
1309 | } | |
1310 | ||
876bec6f | 1311 | /* Set flags and remap blocks. */ |
5faaf4fa CH |
1312 | ret = xfs_reflink_set_inode_flag(src, dest); |
1313 | if (ret) | |
1314 | goto out_unlock; | |
862bb360 | 1315 | |
5faaf4fa CH |
1316 | dfsbno = XFS_B_TO_FSBT(mp, pos_out); |
1317 | sfsbno = XFS_B_TO_FSBT(mp, pos_in); | |
862bb360 | 1318 | fsblen = XFS_B_TO_FSB(mp, len); |
5faaf4fa CH |
1319 | ret = xfs_reflink_remap_blocks(src, sfsbno, dest, dfsbno, fsblen, |
1320 | pos_out + len); | |
1321 | if (ret) | |
1322 | goto out_unlock; | |
862bb360 | 1323 | |
876bec6f DW |
1324 | /* Zap any page cache for the destination file's range. */ |
1325 | truncate_inode_pages_range(&inode_out->i_data, pos_out, | |
1326 | PAGE_ALIGN(pos_out + len) - 1); | |
1327 | ||
f7ca3522 DW |
1328 | /* |
1329 | * Carry the cowextsize hint from src to dest if we're sharing the | |
1330 | * entire source file to the entire destination file, the source file | |
1331 | * has a cowextsize hint, and the destination file does not. | |
1332 | */ | |
1333 | cowextsize = 0; | |
5faaf4fa | 1334 | if (pos_in == 0 && len == i_size_read(inode_in) && |
f7ca3522 | 1335 | (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) && |
5faaf4fa | 1336 | pos_out == 0 && len >= i_size_read(inode_out) && |
f7ca3522 DW |
1337 | !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)) |
1338 | cowextsize = src->i_d.di_cowextsize; | |
1339 | ||
c5ecb423 CH |
1340 | ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize, |
1341 | is_dedupe); | |
862bb360 | 1342 | |
5faaf4fa CH |
1343 | out_unlock: |
1344 | xfs_iunlock(src, XFS_MMAPLOCK_EXCL); | |
65523218 | 1345 | if (!same_inode) |
5faaf4fa | 1346 | xfs_iunlock(dest, XFS_MMAPLOCK_EXCL); |
65523218 | 1347 | unlock_two_nondirectories(inode_in, inode_out); |
5faaf4fa CH |
1348 | if (ret) |
1349 | trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_); | |
1350 | return ret; | |
862bb360 | 1351 | } |
98cc2db5 DW |
1352 | |
1353 | /* | |
1354 | * The user wants to preemptively CoW all shared blocks in this file, | |
1355 | * which enables us to turn off the reflink flag. Iterate all | |
1356 | * extents which are not prealloc/delalloc to see which ranges are | |
1357 | * mentioned in the refcount tree, then read those blocks into the | |
1358 | * pagecache, dirty them, fsync them back out, and then we can update | |
1359 | * the inode flag. What happens if we run out of memory? :) | |
1360 | */ | |
1361 | STATIC int | |
1362 | xfs_reflink_dirty_extents( | |
1363 | struct xfs_inode *ip, | |
1364 | xfs_fileoff_t fbno, | |
1365 | xfs_filblks_t end, | |
1366 | xfs_off_t isize) | |
1367 | { | |
1368 | struct xfs_mount *mp = ip->i_mount; | |
1369 | xfs_agnumber_t agno; | |
1370 | xfs_agblock_t agbno; | |
1371 | xfs_extlen_t aglen; | |
1372 | xfs_agblock_t rbno; | |
1373 | xfs_extlen_t rlen; | |
1374 | xfs_off_t fpos; | |
1375 | xfs_off_t flen; | |
1376 | struct xfs_bmbt_irec map[2]; | |
1377 | int nmaps; | |
9780643c | 1378 | int error = 0; |
98cc2db5 DW |
1379 | |
1380 | while (end - fbno > 0) { | |
1381 | nmaps = 1; | |
1382 | /* | |
1383 | * Look for extents in the file. Skip holes, delalloc, or | |
1384 | * unwritten extents; they can't be reflinked. | |
1385 | */ | |
1386 | error = xfs_bmapi_read(ip, fbno, end - fbno, map, &nmaps, 0); | |
1387 | if (error) | |
1388 | goto out; | |
1389 | if (nmaps == 0) | |
1390 | break; | |
9c4f29d3 | 1391 | if (!xfs_bmap_is_real_extent(&map[0])) |
98cc2db5 DW |
1392 | goto next; |
1393 | ||
1394 | map[1] = map[0]; | |
1395 | while (map[1].br_blockcount) { | |
1396 | agno = XFS_FSB_TO_AGNO(mp, map[1].br_startblock); | |
1397 | agbno = XFS_FSB_TO_AGBNO(mp, map[1].br_startblock); | |
1398 | aglen = map[1].br_blockcount; | |
1399 | ||
92ff7285 DW |
1400 | error = xfs_reflink_find_shared(mp, NULL, agno, agbno, |
1401 | aglen, &rbno, &rlen, true); | |
98cc2db5 DW |
1402 | if (error) |
1403 | goto out; | |
1404 | if (rbno == NULLAGBLOCK) | |
1405 | break; | |
1406 | ||
1407 | /* Dirty the pages */ | |
1408 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1409 | fpos = XFS_FSB_TO_B(mp, map[1].br_startoff + | |
1410 | (rbno - agbno)); | |
1411 | flen = XFS_FSB_TO_B(mp, rlen); | |
1412 | if (fpos + flen > isize) | |
1413 | flen = isize - fpos; | |
1414 | error = iomap_file_dirty(VFS_I(ip), fpos, flen, | |
1415 | &xfs_iomap_ops); | |
1416 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1417 | if (error) | |
1418 | goto out; | |
1419 | ||
1420 | map[1].br_blockcount -= (rbno - agbno + rlen); | |
1421 | map[1].br_startoff += (rbno - agbno + rlen); | |
1422 | map[1].br_startblock += (rbno - agbno + rlen); | |
1423 | } | |
1424 | ||
1425 | next: | |
1426 | fbno = map[0].br_startoff + map[0].br_blockcount; | |
1427 | } | |
1428 | out: | |
1429 | return error; | |
1430 | } | |
1431 | ||
ea7cdd7b | 1432 | /* Does this inode need the reflink flag? */ |
98cc2db5 | 1433 | int |
ea7cdd7b DW |
1434 | xfs_reflink_inode_has_shared_extents( |
1435 | struct xfs_trans *tp, | |
1436 | struct xfs_inode *ip, | |
1437 | bool *has_shared) | |
98cc2db5 | 1438 | { |
ea7cdd7b DW |
1439 | struct xfs_bmbt_irec got; |
1440 | struct xfs_mount *mp = ip->i_mount; | |
1441 | struct xfs_ifork *ifp; | |
1442 | xfs_agnumber_t agno; | |
1443 | xfs_agblock_t agbno; | |
1444 | xfs_extlen_t aglen; | |
1445 | xfs_agblock_t rbno; | |
1446 | xfs_extlen_t rlen; | |
b2b1712a | 1447 | struct xfs_iext_cursor icur; |
ea7cdd7b DW |
1448 | bool found; |
1449 | int error; | |
98cc2db5 | 1450 | |
ea7cdd7b DW |
1451 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); |
1452 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1453 | error = xfs_iread_extents(tp, ip, XFS_DATA_FORK); | |
98cc2db5 DW |
1454 | if (error) |
1455 | return error; | |
ea7cdd7b | 1456 | } |
98cc2db5 | 1457 | |
ea7cdd7b | 1458 | *has_shared = false; |
b2b1712a | 1459 | found = xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got); |
ea7cdd7b DW |
1460 | while (found) { |
1461 | if (isnullstartblock(got.br_startblock) || | |
1462 | got.br_state != XFS_EXT_NORM) | |
1463 | goto next; | |
1464 | agno = XFS_FSB_TO_AGNO(mp, got.br_startblock); | |
1465 | agbno = XFS_FSB_TO_AGBNO(mp, got.br_startblock); | |
1466 | aglen = got.br_blockcount; | |
98cc2db5 | 1467 | |
ea7cdd7b | 1468 | error = xfs_reflink_find_shared(mp, tp, agno, agbno, aglen, |
024adf48 DW |
1469 | &rbno, &rlen, false); |
1470 | if (error) | |
1471 | return error; | |
1472 | /* Is there still a shared block here? */ | |
ea7cdd7b DW |
1473 | if (rbno != NULLAGBLOCK) { |
1474 | *has_shared = true; | |
024adf48 | 1475 | return 0; |
ea7cdd7b | 1476 | } |
98cc2db5 | 1477 | next: |
b2b1712a | 1478 | found = xfs_iext_next_extent(ifp, &icur, &got); |
98cc2db5 DW |
1479 | } |
1480 | ||
ea7cdd7b DW |
1481 | return 0; |
1482 | } | |
1483 | ||
1484 | /* Clear the inode reflink flag if there are no shared extents. */ | |
1485 | int | |
1486 | xfs_reflink_clear_inode_flag( | |
1487 | struct xfs_inode *ip, | |
1488 | struct xfs_trans **tpp) | |
1489 | { | |
1490 | bool needs_flag; | |
1491 | int error = 0; | |
1492 | ||
1493 | ASSERT(xfs_is_reflink_inode(ip)); | |
1494 | ||
1495 | error = xfs_reflink_inode_has_shared_extents(*tpp, ip, &needs_flag); | |
1496 | if (error || needs_flag) | |
1497 | return error; | |
1498 | ||
98cc2db5 DW |
1499 | /* |
1500 | * We didn't find any shared blocks so turn off the reflink flag. | |
1501 | * First, get rid of any leftover CoW mappings. | |
1502 | */ | |
3802a345 | 1503 | error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, NULLFILEOFF, true); |
98cc2db5 DW |
1504 | if (error) |
1505 | return error; | |
1506 | ||
1507 | /* Clear the inode flag. */ | |
1508 | trace_xfs_reflink_unset_inode_flag(ip); | |
1509 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
83104d44 | 1510 | xfs_inode_clear_cowblocks_tag(ip); |
98cc2db5 DW |
1511 | xfs_trans_ijoin(*tpp, ip, 0); |
1512 | xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE); | |
1513 | ||
1514 | return error; | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * Clear the inode reflink flag if there are no shared extents and the size | |
1519 | * hasn't changed. | |
1520 | */ | |
1521 | STATIC int | |
1522 | xfs_reflink_try_clear_inode_flag( | |
97a1b87e | 1523 | struct xfs_inode *ip) |
98cc2db5 DW |
1524 | { |
1525 | struct xfs_mount *mp = ip->i_mount; | |
1526 | struct xfs_trans *tp; | |
1527 | int error = 0; | |
1528 | ||
1529 | /* Start a rolling transaction to remove the mappings */ | |
1530 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp); | |
1531 | if (error) | |
1532 | return error; | |
1533 | ||
1534 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1535 | xfs_trans_ijoin(tp, ip, 0); | |
1536 | ||
98cc2db5 DW |
1537 | error = xfs_reflink_clear_inode_flag(ip, &tp); |
1538 | if (error) | |
1539 | goto cancel; | |
1540 | ||
1541 | error = xfs_trans_commit(tp); | |
1542 | if (error) | |
1543 | goto out; | |
1544 | ||
1545 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1546 | return 0; | |
1547 | cancel: | |
1548 | xfs_trans_cancel(tp); | |
1549 | out: | |
1550 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1551 | return error; | |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * Pre-COW all shared blocks within a given byte range of a file and turn off | |
1556 | * the reflink flag if we unshare all of the file's blocks. | |
1557 | */ | |
1558 | int | |
1559 | xfs_reflink_unshare( | |
1560 | struct xfs_inode *ip, | |
1561 | xfs_off_t offset, | |
1562 | xfs_off_t len) | |
1563 | { | |
1564 | struct xfs_mount *mp = ip->i_mount; | |
1565 | xfs_fileoff_t fbno; | |
1566 | xfs_filblks_t end; | |
1567 | xfs_off_t isize; | |
1568 | int error; | |
1569 | ||
1570 | if (!xfs_is_reflink_inode(ip)) | |
1571 | return 0; | |
1572 | ||
1573 | trace_xfs_reflink_unshare(ip, offset, len); | |
1574 | ||
1575 | inode_dio_wait(VFS_I(ip)); | |
1576 | ||
1577 | /* Try to CoW the selected ranges */ | |
1578 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
97a1b87e | 1579 | fbno = XFS_B_TO_FSBT(mp, offset); |
98cc2db5 DW |
1580 | isize = i_size_read(VFS_I(ip)); |
1581 | end = XFS_B_TO_FSB(mp, offset + len); | |
1582 | error = xfs_reflink_dirty_extents(ip, fbno, end, isize); | |
1583 | if (error) | |
1584 | goto out_unlock; | |
1585 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1586 | ||
1587 | /* Wait for the IO to finish */ | |
1588 | error = filemap_write_and_wait(VFS_I(ip)->i_mapping); | |
1589 | if (error) | |
1590 | goto out; | |
1591 | ||
97a1b87e DW |
1592 | /* Turn off the reflink flag if possible. */ |
1593 | error = xfs_reflink_try_clear_inode_flag(ip); | |
1594 | if (error) | |
1595 | goto out; | |
98cc2db5 DW |
1596 | |
1597 | return 0; | |
1598 | ||
1599 | out_unlock: | |
1600 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1601 | out: | |
1602 | trace_xfs_reflink_unshare_error(ip, error, _RET_IP_); | |
1603 | return error; | |
1604 | } |