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