1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) 2017-2023 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
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"
12 #include "xfs_btree.h"
13 #include "xfs_log_format.h"
14 #include "xfs_trans.h"
15 #include "xfs_ag.h"
16 #include "xfs_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_icache.h"
19 #include "xfs_da_format.h"
20 #include "xfs_reflink.h"
21 #include "xfs_rmap.h"
22 #include "xfs_bmap_util.h"
23 #include "xfs_rtbitmap.h"
24 #include "scrub/scrub.h"
25 #include "scrub/common.h"
26 #include "scrub/btree.h"
27 #include "scrub/trace.h"
28 #include "scrub/repair.h"
29
30 /* Prepare the attached inode for scrubbing. */
31 static inline int
xchk_prepare_iscrub(struct xfs_scrub * sc)32 xchk_prepare_iscrub(
33 struct xfs_scrub *sc)
34 {
35 int error;
36
37 xchk_ilock(sc, XFS_IOLOCK_EXCL);
38
39 error = xchk_trans_alloc(sc, 0);
40 if (error)
41 return error;
42
43 error = xchk_ino_dqattach(sc);
44 if (error)
45 return error;
46
47 xchk_ilock(sc, XFS_ILOCK_EXCL);
48 return 0;
49 }
50
51 /* Install this scrub-by-handle inode and prepare it for scrubbing. */
52 static inline int
xchk_install_handle_iscrub(struct xfs_scrub * sc,struct xfs_inode * ip)53 xchk_install_handle_iscrub(
54 struct xfs_scrub *sc,
55 struct xfs_inode *ip)
56 {
57 int error;
58
59 error = xchk_install_handle_inode(sc, ip);
60 if (error)
61 return error;
62
63 /*
64 * Don't allow scrubbing by handle of any non-directory inode records
65 * in the metadata directory tree. We don't know if any of the scans
66 * launched by this scrubber will end up indirectly trying to lock this
67 * file.
68 *
69 * Scrubbers of inode-rooted metadata files (e.g. quota files) will
70 * attach all the resources needed to scrub the inode and call
71 * xchk_inode directly. Userspace cannot call this directly.
72 */
73 if (xfs_is_metadir_inode(ip) && !S_ISDIR(VFS_I(ip)->i_mode)) {
74 xchk_irele(sc, ip);
75 sc->ip = NULL;
76 return -ENOENT;
77 }
78
79 return xchk_prepare_iscrub(sc);
80 }
81
82 /*
83 * Grab total control of the inode metadata. In the best case, we grab the
84 * incore inode and take all locks on it. If the incore inode cannot be
85 * constructed due to corruption problems, lock the AGI so that we can single
86 * step the loading process to fix everything that can go wrong.
87 */
88 int
xchk_setup_inode(struct xfs_scrub * sc)89 xchk_setup_inode(
90 struct xfs_scrub *sc)
91 {
92 struct xfs_imap imap;
93 struct xfs_inode *ip;
94 struct xfs_mount *mp = sc->mp;
95 struct xfs_inode *ip_in = XFS_I(file_inode(sc->file));
96 struct xfs_buf *agi_bp;
97 struct xfs_perag *pag;
98 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, sc->sm->sm_ino);
99 int error;
100
101 if (xchk_need_intent_drain(sc))
102 xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
103
104 /* We want to scan the opened inode, so lock it and exit. */
105 if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
106 error = xchk_install_live_inode(sc, ip_in);
107 if (error)
108 return error;
109
110 return xchk_prepare_iscrub(sc);
111 }
112
113 /*
114 * On pre-metadir filesystems, reject internal metadata files. For
115 * metadir filesystems, limited scrubbing of any file in the metadata
116 * directory tree by handle is allowed, because that is the only way to
117 * validate the lack of parent pointers in the sb-root metadata inodes.
118 */
119 if (!xfs_has_metadir(mp) && xfs_is_sb_inum(mp, sc->sm->sm_ino))
120 return -ENOENT;
121 /* Reject obviously bad inode numbers. */
122 if (!xfs_verify_ino(sc->mp, sc->sm->sm_ino))
123 return -ENOENT;
124
125 /* Try a safe untrusted iget. */
126 error = xchk_iget_safe(sc, sc->sm->sm_ino, &ip);
127 if (!error)
128 return xchk_install_handle_iscrub(sc, ip);
129 if (error == -ENOENT)
130 return error;
131 if (error != -EFSCORRUPTED && error != -EFSBADCRC && error != -EINVAL)
132 goto out_error;
133
134 /*
135 * EINVAL with IGET_UNTRUSTED probably means one of several things:
136 * userspace gave us an inode number that doesn't correspond to fs
137 * space; the inode btree lacks a record for this inode; or there is
138 * a record, and it says this inode is free.
139 *
140 * EFSCORRUPTED/EFSBADCRC could mean that the inode was mappable, but
141 * some other metadata corruption (e.g. inode forks) prevented
142 * instantiation of the incore inode. Or it could mean the inobt is
143 * corrupt.
144 *
145 * We want to look up this inode in the inobt directly to distinguish
146 * three different scenarios: (1) the inobt says the inode is free,
147 * in which case there's nothing to do; (2) the inobt is corrupt so we
148 * should flag the corruption and exit to userspace to let it fix the
149 * inobt; and (3) the inobt says the inode is allocated, but loading it
150 * failed due to corruption.
151 *
152 * Allocate a transaction and grab the AGI to prevent inobt activity in
153 * this AG. Retry the iget in case someone allocated a new inode after
154 * the first iget failed.
155 */
156 error = xchk_trans_alloc(sc, 0);
157 if (error)
158 goto out_error;
159
160 error = xchk_iget_agi(sc, sc->sm->sm_ino, &agi_bp, &ip);
161 if (error == 0) {
162 /* Actually got the incore inode, so install it and proceed. */
163 xchk_trans_cancel(sc);
164 return xchk_install_handle_iscrub(sc, ip);
165 }
166 if (error == -ENOENT)
167 goto out_gone;
168 if (error != -EFSCORRUPTED && error != -EFSBADCRC && error != -EINVAL)
169 goto out_cancel;
170
171 /* Ensure that we have protected against inode allocation/freeing. */
172 if (agi_bp == NULL) {
173 ASSERT(agi_bp != NULL);
174 error = -ECANCELED;
175 goto out_cancel;
176 }
177
178 /*
179 * Untrusted iget failed a second time. Let's try an inobt lookup.
180 * If the inobt doesn't think this is an allocated inode then we'll
181 * return ENOENT to signal that the check can be skipped.
182 *
183 * If the lookup signals corruption, we'll mark this inode corrupt and
184 * exit to userspace. There's little chance of fixing anything until
185 * the inobt is straightened out, but there's nothing we can do here.
186 *
187 * If the lookup encounters a runtime error, exit to userspace.
188 */
189 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sc->sm->sm_ino));
190 if (!pag) {
191 error = -EFSCORRUPTED;
192 goto out_cancel;
193 }
194
195 error = xfs_imap(pag, sc->tp, sc->sm->sm_ino, &imap,
196 XFS_IGET_UNTRUSTED);
197 xfs_perag_put(pag);
198 if (error == -EINVAL || error == -ENOENT)
199 goto out_gone;
200 if (error)
201 goto out_cancel;
202
203 /*
204 * The lookup succeeded. Chances are the ondisk inode is corrupt and
205 * preventing iget from reading it. Retain the scrub transaction and
206 * the AGI buffer to prevent anyone from allocating or freeing inodes.
207 * This ensures that we preserve the inconsistency between the inobt
208 * saying the inode is allocated and the icache being unable to load
209 * the inode until we can flag the corruption in xchk_inode. The
210 * scrub function has to note the corruption, since we're not really
211 * supposed to do that from the setup function. Save the mapping to
212 * make repairs to the ondisk inode buffer.
213 */
214 if (xchk_could_repair(sc))
215 xrep_setup_inode(sc, &imap);
216 return 0;
217
218 out_cancel:
219 xchk_trans_cancel(sc);
220 out_error:
221 trace_xchk_op_error(sc, agno, XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
222 error, __return_address);
223 return error;
224 out_gone:
225 /* The file is gone, so there's nothing to check. */
226 xchk_trans_cancel(sc);
227 return -ENOENT;
228 }
229
230 /* Inode core */
231
232 /* Validate di_extsize hint. */
233 STATIC void
xchk_inode_extsize(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino,uint16_t mode,uint16_t flags)234 xchk_inode_extsize(
235 struct xfs_scrub *sc,
236 struct xfs_dinode *dip,
237 xfs_ino_t ino,
238 uint16_t mode,
239 uint16_t flags)
240 {
241 xfs_failaddr_t fa;
242 uint32_t value = be32_to_cpu(dip->di_extsize);
243
244 fa = xfs_inode_validate_extsize(sc->mp, value, mode, flags);
245 if (fa)
246 xchk_ino_set_corrupt(sc, ino);
247
248 /*
249 * XFS allows a sysadmin to change the rt extent size when adding a rt
250 * section to a filesystem after formatting. If there are any
251 * directories with extszinherit and rtinherit set, the hint could
252 * become misaligned with the new rextsize. The verifier doesn't check
253 * this, because we allow rtinherit directories even without an rt
254 * device. Flag this as an administrative warning since we will clean
255 * this up eventually.
256 */
257 if ((flags & XFS_DIFLAG_RTINHERIT) &&
258 (flags & XFS_DIFLAG_EXTSZINHERIT) &&
259 xfs_extlen_to_rtxmod(sc->mp, value) > 0)
260 xchk_ino_set_warning(sc, ino);
261 }
262
263 /* Validate di_cowextsize hint. */
264 STATIC void
xchk_inode_cowextsize(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino,uint16_t mode,uint16_t flags,uint64_t flags2)265 xchk_inode_cowextsize(
266 struct xfs_scrub *sc,
267 struct xfs_dinode *dip,
268 xfs_ino_t ino,
269 uint16_t mode,
270 uint16_t flags,
271 uint64_t flags2)
272 {
273 xfs_failaddr_t fa;
274 uint32_t value = be32_to_cpu(dip->di_cowextsize);
275
276 fa = xfs_inode_validate_cowextsize(sc->mp, value, mode, flags, flags2);
277 if (fa)
278 xchk_ino_set_corrupt(sc, ino);
279
280 /*
281 * XFS allows a sysadmin to change the rt extent size when adding a rt
282 * section to a filesystem after formatting. If there are any
283 * directories with cowextsize and rtinherit set, the hint could become
284 * misaligned with the new rextsize. The verifier doesn't check this,
285 * because we allow rtinherit directories even without an rt device.
286 * Flag this as an administrative warning since we will clean this up
287 * eventually.
288 */
289 if ((flags & XFS_DIFLAG_RTINHERIT) &&
290 (flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
291 value % sc->mp->m_sb.sb_rextsize > 0)
292 xchk_ino_set_warning(sc, ino);
293 }
294
295 /* Make sure the di_flags make sense for the inode. */
296 STATIC void
xchk_inode_flags(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino,uint16_t mode,uint16_t flags)297 xchk_inode_flags(
298 struct xfs_scrub *sc,
299 struct xfs_dinode *dip,
300 xfs_ino_t ino,
301 uint16_t mode,
302 uint16_t flags)
303 {
304 struct xfs_mount *mp = sc->mp;
305
306 /* di_flags are all taken, last bit cannot be used */
307 if (flags & ~XFS_DIFLAG_ANY)
308 goto bad;
309
310 /* rt flags require rt device */
311 if ((flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
312 goto bad;
313
314 /* new rt bitmap flag only valid for rbmino */
315 if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
316 goto bad;
317
318 /* directory-only flags */
319 if ((flags & (XFS_DIFLAG_RTINHERIT |
320 XFS_DIFLAG_EXTSZINHERIT |
321 XFS_DIFLAG_PROJINHERIT |
322 XFS_DIFLAG_NOSYMLINKS)) &&
323 !S_ISDIR(mode))
324 goto bad;
325
326 /* file-only flags */
327 if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
328 !S_ISREG(mode))
329 goto bad;
330
331 /* filestreams and rt make no sense */
332 if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
333 goto bad;
334
335 return;
336 bad:
337 xchk_ino_set_corrupt(sc, ino);
338 }
339
340 /* Make sure the di_flags2 make sense for the inode. */
341 STATIC void
xchk_inode_flags2(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino,uint16_t mode,uint16_t flags,uint64_t flags2)342 xchk_inode_flags2(
343 struct xfs_scrub *sc,
344 struct xfs_dinode *dip,
345 xfs_ino_t ino,
346 uint16_t mode,
347 uint16_t flags,
348 uint64_t flags2)
349 {
350 struct xfs_mount *mp = sc->mp;
351
352 /* Unknown di_flags2 could be from a future kernel */
353 if (flags2 & ~XFS_DIFLAG2_ANY)
354 xchk_ino_set_warning(sc, ino);
355
356 /* reflink flag requires reflink feature */
357 if ((flags2 & XFS_DIFLAG2_REFLINK) &&
358 !xfs_has_reflink(mp))
359 goto bad;
360
361 /* cowextsize flag is checked w.r.t. mode separately */
362
363 /* file/dir-only flags */
364 if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
365 goto bad;
366
367 /* file-only flags */
368 if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
369 goto bad;
370
371 /* realtime and reflink don't always go together */
372 if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK) &&
373 !xfs_has_rtreflink(mp))
374 goto bad;
375
376 /* no bigtime iflag without the bigtime feature */
377 if (xfs_dinode_has_bigtime(dip) && !xfs_has_bigtime(mp))
378 goto bad;
379
380 /* no large extent counts without the filesystem feature */
381 if ((flags2 & XFS_DIFLAG2_NREXT64) && !xfs_has_large_extent_counts(mp))
382 goto bad;
383
384 return;
385 bad:
386 xchk_ino_set_corrupt(sc, ino);
387 }
388
389 static inline void
xchk_dinode_nsec(struct xfs_scrub * sc,xfs_ino_t ino,struct xfs_dinode * dip,const xfs_timestamp_t ts)390 xchk_dinode_nsec(
391 struct xfs_scrub *sc,
392 xfs_ino_t ino,
393 struct xfs_dinode *dip,
394 const xfs_timestamp_t ts)
395 {
396 struct timespec64 tv;
397
398 tv = xfs_inode_from_disk_ts(dip, ts);
399 if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
400 xchk_ino_set_corrupt(sc, ino);
401 }
402
403 /* Scrub all the ondisk inode fields. */
404 STATIC void
xchk_dinode(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino)405 xchk_dinode(
406 struct xfs_scrub *sc,
407 struct xfs_dinode *dip,
408 xfs_ino_t ino)
409 {
410 struct xfs_mount *mp = sc->mp;
411 size_t fork_recs;
412 unsigned long long isize;
413 uint64_t flags2;
414 xfs_extnum_t nextents;
415 xfs_extnum_t naextents;
416 prid_t prid;
417 uint16_t flags;
418 uint16_t mode;
419
420 flags = be16_to_cpu(dip->di_flags);
421 if (dip->di_version >= 3)
422 flags2 = be64_to_cpu(dip->di_flags2);
423 else
424 flags2 = 0;
425
426 /* di_mode */
427 mode = be16_to_cpu(dip->di_mode);
428 switch (mode & S_IFMT) {
429 case S_IFLNK:
430 case S_IFREG:
431 case S_IFDIR:
432 case S_IFCHR:
433 case S_IFBLK:
434 case S_IFIFO:
435 case S_IFSOCK:
436 /* mode is recognized */
437 break;
438 default:
439 xchk_ino_set_corrupt(sc, ino);
440 break;
441 }
442
443 /* v1/v2 fields */
444 switch (dip->di_version) {
445 case 1:
446 /*
447 * We autoconvert v1 inodes into v2 inodes on writeout,
448 * so just mark this inode for preening.
449 */
450 xchk_ino_set_preen(sc, ino);
451 prid = 0;
452 break;
453 case 2:
454 case 3:
455 if (xfs_dinode_is_metadir(dip)) {
456 if (be16_to_cpu(dip->di_metatype) >= XFS_METAFILE_MAX)
457 xchk_ino_set_corrupt(sc, ino);
458 } else {
459 if (dip->di_metatype != 0)
460 xchk_ino_set_corrupt(sc, ino);
461 }
462
463 if (dip->di_mode == 0 && sc->ip)
464 xchk_ino_set_corrupt(sc, ino);
465
466 if (dip->di_projid_hi != 0 &&
467 !xfs_has_projid32(mp))
468 xchk_ino_set_corrupt(sc, ino);
469
470 prid = be16_to_cpu(dip->di_projid_lo);
471 break;
472 default:
473 xchk_ino_set_corrupt(sc, ino);
474 return;
475 }
476
477 if (xfs_has_projid32(mp))
478 prid |= (prid_t)be16_to_cpu(dip->di_projid_hi) << 16;
479
480 /*
481 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
482 * userspace could have created that.
483 */
484 if (dip->di_uid == cpu_to_be32(-1U) ||
485 dip->di_gid == cpu_to_be32(-1U))
486 xchk_ino_set_warning(sc, ino);
487
488 /*
489 * project id of -1 isn't supposed to be valid, but the kernel didn't
490 * always validate that.
491 */
492 if (prid == -1U)
493 xchk_ino_set_warning(sc, ino);
494
495 /* di_format */
496 switch (dip->di_format) {
497 case XFS_DINODE_FMT_DEV:
498 if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
499 !S_ISFIFO(mode) && !S_ISSOCK(mode))
500 xchk_ino_set_corrupt(sc, ino);
501 break;
502 case XFS_DINODE_FMT_LOCAL:
503 if (!S_ISDIR(mode) && !S_ISLNK(mode))
504 xchk_ino_set_corrupt(sc, ino);
505 break;
506 case XFS_DINODE_FMT_EXTENTS:
507 if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
508 xchk_ino_set_corrupt(sc, ino);
509 break;
510 case XFS_DINODE_FMT_BTREE:
511 if (!S_ISREG(mode) && !S_ISDIR(mode))
512 xchk_ino_set_corrupt(sc, ino);
513 break;
514 case XFS_DINODE_FMT_META_BTREE:
515 if (!S_ISREG(mode))
516 xchk_ino_set_corrupt(sc, ino);
517 break;
518 case XFS_DINODE_FMT_UUID:
519 default:
520 xchk_ino_set_corrupt(sc, ino);
521 break;
522 }
523
524 /* di_[amc]time.nsec */
525 xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
526 xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
527 xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
528
529 /*
530 * di_size. xfs_dinode_verify checks for things that screw up
531 * the VFS such as the upper bit being set and zero-length
532 * symlinks/directories, but we can do more here.
533 */
534 isize = be64_to_cpu(dip->di_size);
535 if (isize & (1ULL << 63))
536 xchk_ino_set_corrupt(sc, ino);
537
538 /* Devices, fifos, and sockets must have zero size */
539 if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
540 xchk_ino_set_corrupt(sc, ino);
541
542 /* Directories can't be larger than the data section size (32G) */
543 if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
544 xchk_ino_set_corrupt(sc, ino);
545
546 /* Symlinks can't be larger than SYMLINK_MAXLEN */
547 if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
548 xchk_ino_set_corrupt(sc, ino);
549
550 /*
551 * Warn if the running kernel can't handle the kinds of offsets
552 * needed to deal with the file size. In other words, if the
553 * pagecache can't cache all the blocks in this file due to
554 * overly large offsets, flag the inode for admin review.
555 */
556 if (isize > mp->m_super->s_maxbytes)
557 xchk_ino_set_warning(sc, ino);
558
559 /* di_nblocks */
560 if (flags2 & XFS_DIFLAG2_REFLINK) {
561 ; /* nblocks can exceed dblocks */
562 } else if (flags & XFS_DIFLAG_REALTIME) {
563 /*
564 * nblocks is the sum of data extents (in the rtdev),
565 * attr extents (in the datadev), and both forks' bmbt
566 * blocks (in the datadev). This clumsy check is the
567 * best we can do without cross-referencing with the
568 * inode forks.
569 */
570 if (be64_to_cpu(dip->di_nblocks) >=
571 mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
572 xchk_ino_set_corrupt(sc, ino);
573 } else {
574 if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
575 xchk_ino_set_corrupt(sc, ino);
576 }
577
578 xchk_inode_flags(sc, dip, ino, mode, flags);
579
580 xchk_inode_extsize(sc, dip, ino, mode, flags);
581
582 nextents = xfs_dfork_data_extents(dip);
583 naextents = xfs_dfork_attr_extents(dip);
584
585 /* di_nextents */
586 fork_recs = XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
587 switch (dip->di_format) {
588 case XFS_DINODE_FMT_EXTENTS:
589 if (nextents > fork_recs)
590 xchk_ino_set_corrupt(sc, ino);
591 break;
592 case XFS_DINODE_FMT_BTREE:
593 if (nextents <= fork_recs)
594 xchk_ino_set_corrupt(sc, ino);
595 break;
596 default:
597 if (nextents != 0)
598 xchk_ino_set_corrupt(sc, ino);
599 break;
600 }
601
602 /* di_forkoff */
603 if (XFS_DFORK_BOFF(dip) >= mp->m_sb.sb_inodesize)
604 xchk_ino_set_corrupt(sc, ino);
605 if (naextents != 0 && dip->di_forkoff == 0)
606 xchk_ino_set_corrupt(sc, ino);
607 if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
608 xchk_ino_set_corrupt(sc, ino);
609
610 /* di_aformat */
611 if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
612 dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
613 dip->di_aformat != XFS_DINODE_FMT_BTREE)
614 xchk_ino_set_corrupt(sc, ino);
615
616 /* di_anextents */
617 fork_recs = XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
618 switch (dip->di_aformat) {
619 case XFS_DINODE_FMT_EXTENTS:
620 if (naextents > fork_recs)
621 xchk_ino_set_corrupt(sc, ino);
622 break;
623 case XFS_DINODE_FMT_BTREE:
624 if (naextents <= fork_recs)
625 xchk_ino_set_corrupt(sc, ino);
626 break;
627 default:
628 if (naextents != 0)
629 xchk_ino_set_corrupt(sc, ino);
630 }
631
632 if (dip->di_version >= 3) {
633 xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
634 xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
635 xchk_inode_cowextsize(sc, dip, ino, mode, flags,
636 flags2);
637 }
638 }
639
640 /*
641 * Make sure the finobt doesn't think this inode is free.
642 * We don't have to check the inobt ourselves because we got the inode via
643 * IGET_UNTRUSTED, which checks the inobt for us.
644 */
645 static void
xchk_inode_xref_finobt(struct xfs_scrub * sc,xfs_ino_t ino)646 xchk_inode_xref_finobt(
647 struct xfs_scrub *sc,
648 xfs_ino_t ino)
649 {
650 struct xfs_inobt_rec_incore rec;
651 xfs_agino_t agino;
652 int has_record;
653 int error;
654
655 if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
656 return;
657
658 agino = XFS_INO_TO_AGINO(sc->mp, ino);
659
660 /*
661 * Try to get the finobt record. If we can't get it, then we're
662 * in good shape.
663 */
664 error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
665 &has_record);
666 if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
667 !has_record)
668 return;
669
670 error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
671 if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
672 !has_record)
673 return;
674
675 /*
676 * Otherwise, make sure this record either doesn't cover this inode,
677 * or that it does but it's marked present.
678 */
679 if (rec.ir_startino > agino ||
680 rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
681 return;
682
683 if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
684 xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
685 }
686
687 /* Cross reference the inode fields with the forks. */
688 STATIC void
xchk_inode_xref_bmap(struct xfs_scrub * sc,struct xfs_dinode * dip)689 xchk_inode_xref_bmap(
690 struct xfs_scrub *sc,
691 struct xfs_dinode *dip)
692 {
693 xfs_extnum_t nextents;
694 xfs_filblks_t count;
695 xfs_filblks_t acount;
696 int error;
697
698 if (xchk_skip_xref(sc->sm))
699 return;
700
701 /* Walk all the extents to check nextents/naextents/nblocks. */
702 error = xchk_inode_count_blocks(sc, XFS_DATA_FORK, &nextents, &count);
703 if (!xchk_should_check_xref(sc, &error, NULL))
704 return;
705 if (nextents < xfs_dfork_data_extents(dip))
706 xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
707
708 error = xchk_inode_count_blocks(sc, XFS_ATTR_FORK, &nextents, &acount);
709 if (!xchk_should_check_xref(sc, &error, NULL))
710 return;
711 if (nextents != xfs_dfork_attr_extents(dip))
712 xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
713
714 /* Check nblocks against the inode. */
715 if (count + acount != be64_to_cpu(dip->di_nblocks))
716 xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
717 }
718
719 /* Cross-reference with the other btrees. */
720 STATIC void
xchk_inode_xref(struct xfs_scrub * sc,xfs_ino_t ino,struct xfs_dinode * dip)721 xchk_inode_xref(
722 struct xfs_scrub *sc,
723 xfs_ino_t ino,
724 struct xfs_dinode *dip)
725 {
726 xfs_agnumber_t agno;
727 xfs_agblock_t agbno;
728 int error;
729
730 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
731 return;
732
733 agno = XFS_INO_TO_AGNO(sc->mp, ino);
734 agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
735
736 error = xchk_ag_init_existing(sc, agno, &sc->sa);
737 if (!xchk_xref_process_error(sc, agno, agbno, &error))
738 goto out_free;
739
740 xchk_xref_is_used_space(sc, agbno, 1);
741 xchk_inode_xref_finobt(sc, ino);
742 xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
743 xchk_xref_is_not_shared(sc, agbno, 1);
744 xchk_xref_is_not_cow_staging(sc, agbno, 1);
745 xchk_inode_xref_bmap(sc, dip);
746
747 out_free:
748 xchk_ag_free(sc, &sc->sa);
749 }
750
751 /*
752 * If the reflink iflag disagrees with a scan for shared data fork extents,
753 * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
754 * any shared extents). We already checked for reflink iflag set on a non
755 * reflink filesystem.
756 */
757 static void
xchk_inode_check_reflink_iflag(struct xfs_scrub * sc,xfs_ino_t ino)758 xchk_inode_check_reflink_iflag(
759 struct xfs_scrub *sc,
760 xfs_ino_t ino)
761 {
762 struct xfs_mount *mp = sc->mp;
763 bool has_shared;
764 int error;
765
766 if (!xfs_has_reflink(mp))
767 return;
768
769 error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
770 &has_shared);
771 if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
772 XFS_INO_TO_AGBNO(mp, ino), &error))
773 return;
774 if (xfs_is_reflink_inode(sc->ip) && !has_shared)
775 xchk_ino_set_preen(sc, ino);
776 else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
777 xchk_ino_set_corrupt(sc, ino);
778 }
779
780 /*
781 * If this inode has zero link count, it must be on the unlinked list. If
782 * it has nonzero link count, it must not be on the unlinked list.
783 */
784 STATIC void
xchk_inode_check_unlinked(struct xfs_scrub * sc)785 xchk_inode_check_unlinked(
786 struct xfs_scrub *sc)
787 {
788 if (VFS_I(sc->ip)->i_nlink == 0) {
789 if (!xfs_inode_on_unlinked_list(sc->ip))
790 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
791 } else {
792 if (xfs_inode_on_unlinked_list(sc->ip))
793 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
794 }
795 }
796
797 /* Scrub an inode. */
798 int
xchk_inode(struct xfs_scrub * sc)799 xchk_inode(
800 struct xfs_scrub *sc)
801 {
802 struct xfs_dinode di;
803 int error = 0;
804
805 /*
806 * If sc->ip is NULL, that means that the setup function called
807 * xfs_iget to look up the inode. xfs_iget returned a EFSCORRUPTED
808 * and a NULL inode, so flag the corruption error and return.
809 */
810 if (!sc->ip) {
811 xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
812 return 0;
813 }
814
815 /* Scrub the inode core. */
816 xfs_inode_to_disk(sc->ip, &di, 0);
817 xchk_dinode(sc, &di, sc->ip->i_ino);
818 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
819 goto out;
820
821 /*
822 * Look for discrepancies between file's data blocks and the reflink
823 * iflag. We already checked the iflag against the file mode when
824 * we scrubbed the dinode.
825 */
826 if (S_ISREG(VFS_I(sc->ip)->i_mode))
827 xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
828
829 xchk_inode_check_unlinked(sc);
830
831 xchk_inode_xref(sc, sc->ip->i_ino, &di);
832 out:
833 return error;
834 }
835