]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
99d9d8d0 DW |
2 | /* |
3 | * Copyright (C) 2017 Oracle. All Rights Reserved. | |
99d9d8d0 | 4 | * Author: Darrick J. Wong <[email protected]> |
99d9d8d0 DW |
5 | */ |
6 | #include "xfs.h" | |
7 | #include "xfs_fs.h" | |
8 | #include "xfs_shared.h" | |
9 | #include "xfs_format.h" | |
10 | #include "xfs_trans_resv.h" | |
11 | #include "xfs_mount.h" | |
99d9d8d0 DW |
12 | #include "xfs_btree.h" |
13 | #include "xfs_bit.h" | |
14 | #include "xfs_log_format.h" | |
15 | #include "xfs_trans.h" | |
99d9d8d0 | 16 | #include "xfs_inode.h" |
99d9d8d0 | 17 | #include "xfs_alloc.h" |
99d9d8d0 | 18 | #include "xfs_bmap.h" |
99d9d8d0 DW |
19 | #include "xfs_bmap_btree.h" |
20 | #include "xfs_rmap.h" | |
5e777b62 | 21 | #include "xfs_rmap_btree.h" |
99d9d8d0 DW |
22 | #include "scrub/scrub.h" |
23 | #include "scrub/common.h" | |
24 | #include "scrub/btree.h" | |
934933c3 | 25 | #include "xfs_ag.h" |
99d9d8d0 DW |
26 | |
27 | /* Set us up with an inode's bmap. */ | |
28 | int | |
c517b3aa | 29 | xchk_setup_inode_bmap( |
026f57eb | 30 | struct xfs_scrub *sc) |
99d9d8d0 | 31 | { |
032d91f9 | 32 | int error; |
99d9d8d0 | 33 | |
026f57eb | 34 | error = xchk_get_inode(sc); |
99d9d8d0 DW |
35 | if (error) |
36 | goto out; | |
37 | ||
38 | sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; | |
39 | xfs_ilock(sc->ip, sc->ilock_flags); | |
40 | ||
41 | /* | |
42 | * We don't want any ephemeral data fork updates sitting around | |
43 | * while we inspect block mappings, so wait for directio to finish | |
44 | * and flush dirty data if we have delalloc reservations. | |
45 | */ | |
46 | if (S_ISREG(VFS_I(sc->ip)->i_mode) && | |
47 | sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) { | |
eb0efe50 DW |
48 | struct address_space *mapping = VFS_I(sc->ip)->i_mapping; |
49 | ||
99d9d8d0 | 50 | inode_dio_wait(VFS_I(sc->ip)); |
eb0efe50 DW |
51 | |
52 | /* | |
53 | * Try to flush all incore state to disk before we examine the | |
54 | * space mappings for the data fork. Leave accumulated errors | |
55 | * in the mapping for the writer threads to consume. | |
56 | * | |
57 | * On ENOSPC or EIO writeback errors, we continue into the | |
58 | * extent mapping checks because write failures do not | |
59 | * necessarily imply anything about the correctness of the file | |
60 | * metadata. The metadata and the file data could be on | |
61 | * completely separate devices; a media failure might only | |
62 | * affect a subset of the disk, etc. We can handle delalloc | |
63 | * extents in the scrubber, so leaving them in memory is fine. | |
64 | */ | |
65 | error = filemap_fdatawrite(mapping); | |
66 | if (!error) | |
67 | error = filemap_fdatawait_keep_errors(mapping); | |
68 | if (error && (error != -ENOSPC && error != -EIO)) | |
99d9d8d0 DW |
69 | goto out; |
70 | } | |
71 | ||
72 | /* Got the inode, lock it and we're ready to go. */ | |
c517b3aa | 73 | error = xchk_trans_alloc(sc, 0); |
99d9d8d0 DW |
74 | if (error) |
75 | goto out; | |
76 | sc->ilock_flags |= XFS_ILOCK_EXCL; | |
77 | xfs_ilock(sc->ip, XFS_ILOCK_EXCL); | |
78 | ||
79 | out: | |
80 | /* scrub teardown will unlock and release the inode */ | |
81 | return error; | |
82 | } | |
83 | ||
84 | /* | |
85 | * Inode fork block mapping (BMBT) scrubber. | |
86 | * More complex than the others because we have to scrub | |
87 | * all the extents regardless of whether or not the fork | |
88 | * is in btree format. | |
89 | */ | |
90 | ||
c517b3aa | 91 | struct xchk_bmap_info { |
1d8a748a | 92 | struct xfs_scrub *sc; |
032d91f9 DW |
93 | xfs_fileoff_t lastoff; |
94 | bool is_rt; | |
95 | bool is_shared; | |
519e5869 | 96 | bool was_loaded; |
032d91f9 | 97 | int whichfork; |
99d9d8d0 DW |
98 | }; |
99 | ||
d852657c DW |
100 | /* Look for a corresponding rmap for this irec. */ |
101 | static inline bool | |
c517b3aa | 102 | xchk_bmap_get_rmap( |
032d91f9 DW |
103 | struct xchk_bmap_info *info, |
104 | struct xfs_bmbt_irec *irec, | |
105 | xfs_agblock_t agbno, | |
106 | uint64_t owner, | |
107 | struct xfs_rmap_irec *rmap) | |
d852657c | 108 | { |
032d91f9 DW |
109 | xfs_fileoff_t offset; |
110 | unsigned int rflags = 0; | |
111 | int has_rmap; | |
112 | int error; | |
d852657c DW |
113 | |
114 | if (info->whichfork == XFS_ATTR_FORK) | |
115 | rflags |= XFS_RMAP_ATTR_FORK; | |
5dda3897 DW |
116 | if (irec->br_state == XFS_EXT_UNWRITTEN) |
117 | rflags |= XFS_RMAP_UNWRITTEN; | |
d852657c DW |
118 | |
119 | /* | |
120 | * CoW staging extents are owned (on disk) by the refcountbt, so | |
121 | * their rmaps do not have offsets. | |
122 | */ | |
123 | if (info->whichfork == XFS_COW_FORK) | |
124 | offset = 0; | |
125 | else | |
126 | offset = irec->br_startoff; | |
127 | ||
128 | /* | |
129 | * If the caller thinks this could be a shared bmbt extent (IOWs, | |
130 | * any data fork extent of a reflink inode) then we have to use the | |
131 | * range rmap lookup to make sure we get the correct owner/offset. | |
132 | */ | |
133 | if (info->is_shared) { | |
134 | error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno, | |
135 | owner, offset, rflags, rmap, &has_rmap); | |
c517b3aa | 136 | if (!xchk_should_check_xref(info->sc, &error, |
d852657c DW |
137 | &info->sc->sa.rmap_cur)) |
138 | return false; | |
139 | goto out; | |
140 | } | |
141 | ||
142 | /* | |
143 | * Otherwise, use the (faster) regular lookup. | |
144 | */ | |
145 | error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno, 0, owner, | |
146 | offset, rflags, &has_rmap); | |
c517b3aa | 147 | if (!xchk_should_check_xref(info->sc, &error, |
d852657c DW |
148 | &info->sc->sa.rmap_cur)) |
149 | return false; | |
150 | if (!has_rmap) | |
151 | goto out; | |
152 | ||
153 | error = xfs_rmap_get_rec(info->sc->sa.rmap_cur, rmap, &has_rmap); | |
c517b3aa | 154 | if (!xchk_should_check_xref(info->sc, &error, |
d852657c DW |
155 | &info->sc->sa.rmap_cur)) |
156 | return false; | |
157 | ||
158 | out: | |
159 | if (!has_rmap) | |
c517b3aa | 160 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
d852657c DW |
161 | irec->br_startoff); |
162 | return has_rmap; | |
163 | } | |
164 | ||
165 | /* Make sure that we have rmapbt records for this extent. */ | |
166 | STATIC void | |
c517b3aa | 167 | xchk_bmap_xref_rmap( |
032d91f9 DW |
168 | struct xchk_bmap_info *info, |
169 | struct xfs_bmbt_irec *irec, | |
170 | xfs_agblock_t agbno) | |
d852657c | 171 | { |
032d91f9 DW |
172 | struct xfs_rmap_irec rmap; |
173 | unsigned long long rmap_end; | |
174 | uint64_t owner; | |
d852657c | 175 | |
c517b3aa | 176 | if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm)) |
d852657c DW |
177 | return; |
178 | ||
179 | if (info->whichfork == XFS_COW_FORK) | |
180 | owner = XFS_RMAP_OWN_COW; | |
181 | else | |
182 | owner = info->sc->ip->i_ino; | |
183 | ||
184 | /* Find the rmap record for this irec. */ | |
c517b3aa | 185 | if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap)) |
d852657c DW |
186 | return; |
187 | ||
188 | /* Check the rmap. */ | |
189 | rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount; | |
190 | if (rmap.rm_startblock > agbno || | |
191 | agbno + irec->br_blockcount > rmap_end) | |
c517b3aa | 192 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
d852657c DW |
193 | irec->br_startoff); |
194 | ||
195 | /* | |
196 | * Check the logical offsets if applicable. CoW staging extents | |
197 | * don't track logical offsets since the mappings only exist in | |
198 | * memory. | |
199 | */ | |
200 | if (info->whichfork != XFS_COW_FORK) { | |
201 | rmap_end = (unsigned long long)rmap.rm_offset + | |
202 | rmap.rm_blockcount; | |
203 | if (rmap.rm_offset > irec->br_startoff || | |
204 | irec->br_startoff + irec->br_blockcount > rmap_end) | |
c517b3aa | 205 | xchk_fblock_xref_set_corrupt(info->sc, |
d852657c DW |
206 | info->whichfork, irec->br_startoff); |
207 | } | |
208 | ||
209 | if (rmap.rm_owner != owner) | |
c517b3aa | 210 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
d852657c DW |
211 | irec->br_startoff); |
212 | ||
213 | /* | |
214 | * Check for discrepancies between the unwritten flag in the irec and | |
215 | * the rmap. Note that the (in-memory) CoW fork distinguishes between | |
216 | * unwritten and written extents, but we don't track that in the rmap | |
217 | * records because the blocks are owned (on-disk) by the refcountbt, | |
218 | * which doesn't track unwritten state. | |
219 | */ | |
220 | if (owner != XFS_RMAP_OWN_COW && | |
498fe261 DW |
221 | !!(irec->br_state == XFS_EXT_UNWRITTEN) != |
222 | !!(rmap.rm_flags & XFS_RMAP_UNWRITTEN)) | |
c517b3aa | 223 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
d852657c DW |
224 | irec->br_startoff); |
225 | ||
498fe261 DW |
226 | if (!!(info->whichfork == XFS_ATTR_FORK) != |
227 | !!(rmap.rm_flags & XFS_RMAP_ATTR_FORK)) | |
c517b3aa | 228 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
d852657c DW |
229 | irec->br_startoff); |
230 | if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK) | |
c517b3aa | 231 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
d852657c DW |
232 | irec->br_startoff); |
233 | } | |
234 | ||
166d7641 DW |
235 | /* Cross-reference a single rtdev extent record. */ |
236 | STATIC void | |
519e5869 | 237 | xchk_bmap_rt_iextent_xref( |
032d91f9 | 238 | struct xfs_inode *ip, |
519e5869 | 239 | struct xchk_bmap_info *info, |
032d91f9 | 240 | struct xfs_bmbt_irec *irec) |
166d7641 | 241 | { |
c517b3aa | 242 | xchk_xref_is_used_rt_space(info->sc, irec->br_startblock, |
46d9bfb5 | 243 | irec->br_blockcount); |
166d7641 DW |
244 | } |
245 | ||
246 | /* Cross-reference a single datadev extent record. */ | |
247 | STATIC void | |
519e5869 | 248 | xchk_bmap_iextent_xref( |
032d91f9 | 249 | struct xfs_inode *ip, |
519e5869 | 250 | struct xchk_bmap_info *info, |
032d91f9 | 251 | struct xfs_bmbt_irec *irec) |
166d7641 | 252 | { |
032d91f9 DW |
253 | struct xfs_mount *mp = info->sc->mp; |
254 | xfs_agnumber_t agno; | |
255 | xfs_agblock_t agbno; | |
256 | xfs_extlen_t len; | |
257 | int error; | |
52dc4b44 | 258 | |
52dc4b44 DW |
259 | agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock); |
260 | agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock); | |
261 | len = irec->br_blockcount; | |
262 | ||
48c6615c | 263 | error = xchk_ag_init_existing(info->sc, agno, &info->sc->sa); |
c517b3aa | 264 | if (!xchk_fblock_process_error(info->sc, info->whichfork, |
52dc4b44 | 265 | irec->br_startoff, &error)) |
61e0d0cc | 266 | goto out_free; |
52dc4b44 | 267 | |
c517b3aa DW |
268 | xchk_xref_is_used_space(info->sc, agbno, len); |
269 | xchk_xref_is_not_inode_chunk(info->sc, agbno, len); | |
270 | xchk_bmap_xref_rmap(info, irec, agbno); | |
f6d5fc21 DW |
271 | switch (info->whichfork) { |
272 | case XFS_DATA_FORK: | |
273 | if (xfs_is_reflink_inode(info->sc->ip)) | |
274 | break; | |
53004ee7 | 275 | fallthrough; |
f6d5fc21 | 276 | case XFS_ATTR_FORK: |
c517b3aa | 277 | xchk_xref_is_not_shared(info->sc, agbno, |
f6d5fc21 DW |
278 | irec->br_blockcount); |
279 | break; | |
280 | case XFS_COW_FORK: | |
c517b3aa | 281 | xchk_xref_is_cow_staging(info->sc, agbno, |
f6d5fc21 DW |
282 | irec->br_blockcount); |
283 | break; | |
284 | } | |
52dc4b44 | 285 | |
61e0d0cc | 286 | out_free: |
c517b3aa | 287 | xchk_ag_free(info->sc, &info->sc->sa); |
166d7641 DW |
288 | } |
289 | ||
f8c1d702 DW |
290 | /* |
291 | * Directories and attr forks should never have blocks that can't be addressed | |
292 | * by a xfs_dablk_t. | |
293 | */ | |
294 | STATIC void | |
295 | xchk_bmap_dirattr_extent( | |
296 | struct xfs_inode *ip, | |
297 | struct xchk_bmap_info *info, | |
298 | struct xfs_bmbt_irec *irec) | |
299 | { | |
300 | struct xfs_mount *mp = ip->i_mount; | |
301 | xfs_fileoff_t off; | |
302 | ||
303 | if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK) | |
304 | return; | |
305 | ||
306 | if (!xfs_verify_dablk(mp, irec->br_startoff)) | |
307 | xchk_fblock_set_corrupt(info->sc, info->whichfork, | |
308 | irec->br_startoff); | |
309 | ||
310 | off = irec->br_startoff + irec->br_blockcount - 1; | |
311 | if (!xfs_verify_dablk(mp, off)) | |
312 | xchk_fblock_set_corrupt(info->sc, info->whichfork, off); | |
313 | } | |
314 | ||
99d9d8d0 DW |
315 | /* Scrub a single extent record. */ |
316 | STATIC int | |
519e5869 | 317 | xchk_bmap_iextent( |
032d91f9 | 318 | struct xfs_inode *ip, |
032d91f9 DW |
319 | struct xchk_bmap_info *info, |
320 | struct xfs_bmbt_irec *irec) | |
99d9d8d0 | 321 | { |
032d91f9 | 322 | struct xfs_mount *mp = info->sc->mp; |
032d91f9 | 323 | int error = 0; |
99d9d8d0 | 324 | |
99d9d8d0 DW |
325 | /* |
326 | * Check for out-of-order extents. This record could have come | |
327 | * from the incore list, for which there is no ordering check. | |
328 | */ | |
329 | if (irec->br_startoff < info->lastoff) | |
c517b3aa | 330 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
99d9d8d0 DW |
331 | irec->br_startoff); |
332 | ||
33005fd0 DW |
333 | if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount)) |
334 | xchk_fblock_set_corrupt(info->sc, info->whichfork, | |
335 | irec->br_startoff); | |
336 | ||
f8c1d702 DW |
337 | xchk_bmap_dirattr_extent(ip, info, irec); |
338 | ||
99d9d8d0 DW |
339 | /* There should never be a "hole" extent in either extent list. */ |
340 | if (irec->br_startblock == HOLESTARTBLOCK) | |
c517b3aa | 341 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
99d9d8d0 DW |
342 | irec->br_startoff); |
343 | ||
344 | /* | |
345 | * Check for delalloc extents. We never iterate the ones in the | |
346 | * in-core extent scan, and we should never see these in the bmbt. | |
347 | */ | |
348 | if (isnullstartblock(irec->br_startblock)) | |
c517b3aa | 349 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
99d9d8d0 DW |
350 | irec->br_startoff); |
351 | ||
352 | /* Make sure the extent points to a valid place. */ | |
a5f460b1 | 353 | if (irec->br_blockcount > MAXEXTLEN) |
c517b3aa | 354 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
a5f460b1 | 355 | irec->br_startoff); |
99d9d8d0 | 356 | if (info->is_rt && |
18695ad4 | 357 | !xfs_verify_rtext(mp, irec->br_startblock, irec->br_blockcount)) |
c517b3aa | 358 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
99d9d8d0 DW |
359 | irec->br_startoff); |
360 | if (!info->is_rt && | |
67457eb0 | 361 | !xfs_verify_fsbext(mp, irec->br_startblock, irec->br_blockcount)) |
c517b3aa | 362 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
99d9d8d0 DW |
363 | irec->br_startoff); |
364 | ||
365 | /* We don't allow unwritten extents on attr forks. */ | |
366 | if (irec->br_state == XFS_EXT_UNWRITTEN && | |
367 | info->whichfork == XFS_ATTR_FORK) | |
c517b3aa | 368 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
99d9d8d0 DW |
369 | irec->br_startoff); |
370 | ||
519e5869 DW |
371 | if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
372 | return 0; | |
373 | ||
166d7641 | 374 | if (info->is_rt) |
519e5869 | 375 | xchk_bmap_rt_iextent_xref(ip, info, irec); |
166d7641 | 376 | else |
519e5869 | 377 | xchk_bmap_iextent_xref(ip, info, irec); |
166d7641 | 378 | |
99d9d8d0 DW |
379 | info->lastoff = irec->br_startoff + irec->br_blockcount; |
380 | return error; | |
381 | } | |
382 | ||
383 | /* Scrub a bmbt record. */ | |
384 | STATIC int | |
c517b3aa | 385 | xchk_bmapbt_rec( |
032d91f9 | 386 | struct xchk_btree *bs, |
22ece4e8 | 387 | const union xfs_btree_rec *rec) |
99d9d8d0 | 388 | { |
032d91f9 | 389 | struct xfs_bmbt_irec irec; |
519e5869 DW |
390 | struct xfs_bmbt_irec iext_irec; |
391 | struct xfs_iext_cursor icur; | |
032d91f9 | 392 | struct xchk_bmap_info *info = bs->private; |
92219c29 | 393 | struct xfs_inode *ip = bs->cur->bc_ino.ip; |
032d91f9 DW |
394 | struct xfs_buf *bp = NULL; |
395 | struct xfs_btree_block *block; | |
519e5869 | 396 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, info->whichfork); |
032d91f9 DW |
397 | uint64_t owner; |
398 | int i; | |
99d9d8d0 DW |
399 | |
400 | /* | |
401 | * Check the owners of the btree blocks up to the level below | |
402 | * the root since the verifiers don't do that. | |
403 | */ | |
38c26bfd | 404 | if (xfs_has_crc(bs->cur->bc_mp) && |
99d9d8d0 DW |
405 | bs->cur->bc_ptrs[0] == 1) { |
406 | for (i = 0; i < bs->cur->bc_nlevels - 1; i++) { | |
407 | block = xfs_btree_get_block(bs->cur, i, &bp); | |
408 | owner = be64_to_cpu(block->bb_u.l.bb_owner); | |
409 | if (owner != ip->i_ino) | |
c517b3aa | 410 | xchk_fblock_set_corrupt(bs->sc, |
99d9d8d0 DW |
411 | info->whichfork, 0); |
412 | } | |
413 | } | |
414 | ||
519e5869 DW |
415 | /* |
416 | * Check that the incore extent tree contains an extent that matches | |
417 | * this one exactly. We validate those cached bmaps later, so we don't | |
418 | * need to check them here. If the incore extent tree was just loaded | |
419 | * from disk by the scrubber, we assume that its contents match what's | |
420 | * on disk (we still hold the ILOCK) and skip the equivalence check. | |
421 | */ | |
422 | if (!info->was_loaded) | |
423 | return 0; | |
424 | ||
6bdcf26a | 425 | xfs_bmbt_disk_get_all(&rec->bmbt, &irec); |
519e5869 DW |
426 | if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur, |
427 | &iext_irec) || | |
428 | irec.br_startoff != iext_irec.br_startoff || | |
429 | irec.br_startblock != iext_irec.br_startblock || | |
430 | irec.br_blockcount != iext_irec.br_blockcount || | |
431 | irec.br_state != iext_irec.br_state) | |
432 | xchk_fblock_set_corrupt(bs->sc, info->whichfork, | |
433 | irec.br_startoff); | |
434 | return 0; | |
99d9d8d0 DW |
435 | } |
436 | ||
437 | /* Scan the btree records. */ | |
438 | STATIC int | |
c517b3aa | 439 | xchk_bmap_btree( |
1d8a748a | 440 | struct xfs_scrub *sc, |
032d91f9 DW |
441 | int whichfork, |
442 | struct xchk_bmap_info *info) | |
99d9d8d0 | 443 | { |
032d91f9 | 444 | struct xfs_owner_info oinfo; |
519e5869 | 445 | struct xfs_ifork *ifp = XFS_IFORK_PTR(sc->ip, whichfork); |
032d91f9 DW |
446 | struct xfs_mount *mp = sc->mp; |
447 | struct xfs_inode *ip = sc->ip; | |
448 | struct xfs_btree_cur *cur; | |
449 | int error; | |
99d9d8d0 | 450 | |
519e5869 | 451 | /* Load the incore bmap cache if it's not loaded. */ |
b2197a36 | 452 | info->was_loaded = !xfs_need_iread_extents(ifp); |
862a804a CH |
453 | |
454 | error = xfs_iread_extents(sc->tp, ip, whichfork); | |
455 | if (!xchk_fblock_process_error(sc, whichfork, 0, &error)) | |
456 | goto out; | |
519e5869 DW |
457 | |
458 | /* Check the btree structure. */ | |
99d9d8d0 DW |
459 | cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork); |
460 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); | |
c517b3aa | 461 | error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info); |
0b04b6b8 | 462 | xfs_btree_del_cursor(cur, error); |
519e5869 | 463 | out: |
99d9d8d0 DW |
464 | return error; |
465 | } | |
466 | ||
c517b3aa | 467 | struct xchk_bmap_check_rmap_info { |
1d8a748a | 468 | struct xfs_scrub *sc; |
032d91f9 DW |
469 | int whichfork; |
470 | struct xfs_iext_cursor icur; | |
5e777b62 DW |
471 | }; |
472 | ||
473 | /* Can we find bmaps that fit this rmap? */ | |
474 | STATIC int | |
c517b3aa | 475 | xchk_bmap_check_rmap( |
5e777b62 | 476 | struct xfs_btree_cur *cur, |
159eb69d | 477 | const struct xfs_rmap_irec *rec, |
5e777b62 DW |
478 | void *priv) |
479 | { | |
480 | struct xfs_bmbt_irec irec; | |
159eb69d | 481 | struct xfs_rmap_irec check_rec; |
c517b3aa | 482 | struct xchk_bmap_check_rmap_info *sbcri = priv; |
5e777b62 | 483 | struct xfs_ifork *ifp; |
032d91f9 | 484 | struct xfs_scrub *sc = sbcri->sc; |
5e777b62 DW |
485 | bool have_map; |
486 | ||
487 | /* Is this even the right fork? */ | |
488 | if (rec->rm_owner != sc->ip->i_ino) | |
489 | return 0; | |
490 | if ((sbcri->whichfork == XFS_ATTR_FORK) ^ | |
491 | !!(rec->rm_flags & XFS_RMAP_ATTR_FORK)) | |
492 | return 0; | |
493 | if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK) | |
494 | return 0; | |
495 | ||
496 | /* Now look up the bmbt record. */ | |
497 | ifp = XFS_IFORK_PTR(sc->ip, sbcri->whichfork); | |
498 | if (!ifp) { | |
c517b3aa | 499 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
5e777b62 DW |
500 | rec->rm_offset); |
501 | goto out; | |
502 | } | |
503 | have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset, | |
504 | &sbcri->icur, &irec); | |
505 | if (!have_map) | |
c517b3aa | 506 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
5e777b62 DW |
507 | rec->rm_offset); |
508 | /* | |
509 | * bmap extent record lengths are constrained to 2^21 blocks in length | |
510 | * because of space constraints in the on-disk metadata structure. | |
511 | * However, rmap extent record lengths are constrained only by AG | |
512 | * length, so we have to loop through the bmbt to make sure that the | |
513 | * entire rmap is covered by bmbt records. | |
514 | */ | |
159eb69d | 515 | check_rec = *rec; |
5e777b62 | 516 | while (have_map) { |
159eb69d | 517 | if (irec.br_startoff != check_rec.rm_offset) |
c517b3aa | 518 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
159eb69d | 519 | check_rec.rm_offset); |
5e777b62 | 520 | if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp, |
159eb69d DW |
521 | cur->bc_ag.pag->pag_agno, |
522 | check_rec.rm_startblock)) | |
c517b3aa | 523 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
159eb69d DW |
524 | check_rec.rm_offset); |
525 | if (irec.br_blockcount > check_rec.rm_blockcount) | |
c517b3aa | 526 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
159eb69d | 527 | check_rec.rm_offset); |
5e777b62 DW |
528 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
529 | break; | |
159eb69d DW |
530 | check_rec.rm_startblock += irec.br_blockcount; |
531 | check_rec.rm_offset += irec.br_blockcount; | |
532 | check_rec.rm_blockcount -= irec.br_blockcount; | |
533 | if (check_rec.rm_blockcount == 0) | |
5e777b62 DW |
534 | break; |
535 | have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec); | |
536 | if (!have_map) | |
c517b3aa | 537 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
159eb69d | 538 | check_rec.rm_offset); |
5e777b62 DW |
539 | } |
540 | ||
541 | out: | |
542 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) | |
e7ee96df | 543 | return -ECANCELED; |
5e777b62 DW |
544 | return 0; |
545 | } | |
546 | ||
547 | /* Make sure each rmap has a corresponding bmbt entry. */ | |
548 | STATIC int | |
c517b3aa | 549 | xchk_bmap_check_ag_rmaps( |
032d91f9 | 550 | struct xfs_scrub *sc, |
5e777b62 | 551 | int whichfork, |
a81a0621 | 552 | struct xfs_perag *pag) |
5e777b62 | 553 | { |
c517b3aa | 554 | struct xchk_bmap_check_rmap_info sbcri; |
5e777b62 DW |
555 | struct xfs_btree_cur *cur; |
556 | struct xfs_buf *agf; | |
557 | int error; | |
558 | ||
a81a0621 | 559 | error = xfs_alloc_read_agf(sc->mp, sc->tp, pag->pag_agno, 0, &agf); |
5e777b62 DW |
560 | if (error) |
561 | return error; | |
562 | ||
a81a0621 | 563 | cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, pag); |
5e777b62 DW |
564 | |
565 | sbcri.sc = sc; | |
566 | sbcri.whichfork = whichfork; | |
c517b3aa | 567 | error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri); |
e7ee96df | 568 | if (error == -ECANCELED) |
5e777b62 DW |
569 | error = 0; |
570 | ||
0b04b6b8 | 571 | xfs_btree_del_cursor(cur, error); |
5e777b62 DW |
572 | xfs_trans_brelse(sc->tp, agf); |
573 | return error; | |
574 | } | |
575 | ||
576 | /* Make sure each rmap has a corresponding bmbt entry. */ | |
577 | STATIC int | |
c517b3aa | 578 | xchk_bmap_check_rmaps( |
1d8a748a | 579 | struct xfs_scrub *sc, |
032d91f9 | 580 | int whichfork) |
5e777b62 | 581 | { |
daf83964 | 582 | struct xfs_ifork *ifp = XFS_IFORK_PTR(sc->ip, whichfork); |
934933c3 | 583 | struct xfs_perag *pag; |
032d91f9 | 584 | xfs_agnumber_t agno; |
5fd68bdb | 585 | bool zero_size; |
032d91f9 | 586 | int error; |
5e777b62 | 587 | |
38c26bfd | 588 | if (!xfs_has_rmapbt(sc->mp) || |
5e777b62 DW |
589 | whichfork == XFS_COW_FORK || |
590 | (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) | |
591 | return 0; | |
592 | ||
593 | /* Don't support realtime rmap checks yet. */ | |
594 | if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK) | |
595 | return 0; | |
596 | ||
5fd68bdb DW |
597 | ASSERT(XFS_IFORK_PTR(sc->ip, whichfork) != NULL); |
598 | ||
5e777b62 DW |
599 | /* |
600 | * Only do this for complex maps that are in btree format, or for | |
601 | * situations where we would seem to have a size but zero extents. | |
602 | * The inode repair code can zap broken iforks, which means we have | |
603 | * to flag this bmap as corrupt if there are rmaps that need to be | |
604 | * reattached. | |
605 | */ | |
f7e67b20 | 606 | |
5fd68bdb DW |
607 | if (whichfork == XFS_DATA_FORK) |
608 | zero_size = i_size_read(VFS_I(sc->ip)) == 0; | |
609 | else | |
610 | zero_size = false; | |
611 | ||
f7e67b20 | 612 | if (ifp->if_format != XFS_DINODE_FMT_BTREE && |
daf83964 | 613 | (zero_size || ifp->if_nextents > 0)) |
5e777b62 DW |
614 | return 0; |
615 | ||
934933c3 | 616 | for_each_perag(sc->mp, agno, pag) { |
a81a0621 | 617 | error = xchk_bmap_check_ag_rmaps(sc, whichfork, pag); |
5e777b62 | 618 | if (error) |
934933c3 | 619 | break; |
5e777b62 DW |
620 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
621 | break; | |
622 | } | |
934933c3 DC |
623 | if (pag) |
624 | xfs_perag_put(pag); | |
625 | return error; | |
5e777b62 DW |
626 | } |
627 | ||
99d9d8d0 DW |
628 | /* |
629 | * Scrub an inode fork's block mappings. | |
630 | * | |
631 | * First we scan every record in every btree block, if applicable. | |
632 | * Then we unconditionally scan the incore extent cache. | |
633 | */ | |
634 | STATIC int | |
c517b3aa | 635 | xchk_bmap( |
1d8a748a | 636 | struct xfs_scrub *sc, |
032d91f9 | 637 | int whichfork) |
99d9d8d0 | 638 | { |
032d91f9 DW |
639 | struct xfs_bmbt_irec irec; |
640 | struct xchk_bmap_info info = { NULL }; | |
641 | struct xfs_mount *mp = sc->mp; | |
642 | struct xfs_inode *ip = sc->ip; | |
5fd68bdb | 643 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
032d91f9 DW |
644 | xfs_fileoff_t endoff; |
645 | struct xfs_iext_cursor icur; | |
646 | int error = 0; | |
99d9d8d0 | 647 | |
5fd68bdb DW |
648 | /* Non-existent forks can be ignored. */ |
649 | if (!ifp) | |
650 | goto out; | |
99d9d8d0 DW |
651 | |
652 | info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip); | |
653 | info.whichfork = whichfork; | |
654 | info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip); | |
655 | info.sc = sc; | |
656 | ||
657 | switch (whichfork) { | |
658 | case XFS_COW_FORK: | |
99d9d8d0 DW |
659 | /* No CoW forks on non-reflink inodes/filesystems. */ |
660 | if (!xfs_is_reflink_inode(ip)) { | |
c517b3aa | 661 | xchk_ino_set_corrupt(sc, sc->ip->i_ino); |
99d9d8d0 DW |
662 | goto out; |
663 | } | |
664 | break; | |
665 | case XFS_ATTR_FORK: | |
ebd9027d | 666 | if (!xfs_has_attr(mp) && !xfs_has_attr2(mp)) |
c517b3aa | 667 | xchk_ino_set_corrupt(sc, sc->ip->i_ino); |
99d9d8d0 DW |
668 | break; |
669 | default: | |
670 | ASSERT(whichfork == XFS_DATA_FORK); | |
671 | break; | |
672 | } | |
673 | ||
674 | /* Check the fork values */ | |
f7e67b20 | 675 | switch (ifp->if_format) { |
99d9d8d0 DW |
676 | case XFS_DINODE_FMT_UUID: |
677 | case XFS_DINODE_FMT_DEV: | |
678 | case XFS_DINODE_FMT_LOCAL: | |
679 | /* No mappings to check. */ | |
680 | goto out; | |
681 | case XFS_DINODE_FMT_EXTENTS: | |
99d9d8d0 DW |
682 | break; |
683 | case XFS_DINODE_FMT_BTREE: | |
684 | if (whichfork == XFS_COW_FORK) { | |
c517b3aa | 685 | xchk_fblock_set_corrupt(sc, whichfork, 0); |
99d9d8d0 DW |
686 | goto out; |
687 | } | |
688 | ||
c517b3aa | 689 | error = xchk_bmap_btree(sc, whichfork, &info); |
99d9d8d0 DW |
690 | if (error) |
691 | goto out; | |
692 | break; | |
693 | default: | |
c517b3aa | 694 | xchk_fblock_set_corrupt(sc, whichfork, 0); |
99d9d8d0 DW |
695 | goto out; |
696 | } | |
697 | ||
698 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) | |
699 | goto out; | |
700 | ||
99d9d8d0 DW |
701 | /* Find the offset of the last extent in the mapping. */ |
702 | error = xfs_bmap_last_offset(ip, &endoff, whichfork); | |
c517b3aa | 703 | if (!xchk_fblock_process_error(sc, whichfork, 0, &error)) |
99d9d8d0 DW |
704 | goto out; |
705 | ||
706 | /* Scrub extent records. */ | |
707 | info.lastoff = 0; | |
708 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2b9e9b57 | 709 | for_each_xfs_iext(ifp, &icur, &irec) { |
c517b3aa | 710 | if (xchk_should_terminate(sc, &error) || |
8bc763c2 | 711 | (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) |
519e5869 | 712 | goto out; |
99d9d8d0 DW |
713 | if (isnullstartblock(irec.br_startblock)) |
714 | continue; | |
715 | if (irec.br_startoff >= endoff) { | |
c517b3aa | 716 | xchk_fblock_set_corrupt(sc, whichfork, |
99d9d8d0 DW |
717 | irec.br_startoff); |
718 | goto out; | |
719 | } | |
519e5869 | 720 | error = xchk_bmap_iextent(ip, &info, &irec); |
99d9d8d0 DW |
721 | if (error) |
722 | goto out; | |
723 | } | |
724 | ||
c517b3aa DW |
725 | error = xchk_bmap_check_rmaps(sc, whichfork); |
726 | if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error)) | |
5e777b62 | 727 | goto out; |
99d9d8d0 DW |
728 | out: |
729 | return error; | |
730 | } | |
731 | ||
732 | /* Scrub an inode's data fork. */ | |
733 | int | |
c517b3aa | 734 | xchk_bmap_data( |
1d8a748a | 735 | struct xfs_scrub *sc) |
99d9d8d0 | 736 | { |
c517b3aa | 737 | return xchk_bmap(sc, XFS_DATA_FORK); |
99d9d8d0 DW |
738 | } |
739 | ||
740 | /* Scrub an inode's attr fork. */ | |
741 | int | |
c517b3aa | 742 | xchk_bmap_attr( |
1d8a748a | 743 | struct xfs_scrub *sc) |
99d9d8d0 | 744 | { |
c517b3aa | 745 | return xchk_bmap(sc, XFS_ATTR_FORK); |
99d9d8d0 DW |
746 | } |
747 | ||
748 | /* Scrub an inode's CoW fork. */ | |
749 | int | |
c517b3aa | 750 | xchk_bmap_cow( |
1d8a748a | 751 | struct xfs_scrub *sc) |
99d9d8d0 DW |
752 | { |
753 | if (!xfs_is_reflink_inode(sc->ip)) | |
754 | return -ENOENT; | |
755 | ||
c517b3aa | 756 | return xchk_bmap(sc, XFS_COW_FORK); |
99d9d8d0 | 757 | } |