1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * Copyright (C) 2017 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
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_defer.h"
13 #include "xfs_btree.h"
14 #include "xfs_bit.h"
15 #include "xfs_log_format.h"
16 #include "xfs_trans.h"
17 #include "xfs_sb.h"
18 #include "xfs_inode.h"
19 #include "xfs_icache.h"
20 #include "xfs_inode_buf.h"
21 #include "xfs_inode_fork.h"
22 #include "xfs_ialloc.h"
23 #include "xfs_da_format.h"
24 #include "xfs_reflink.h"
25 #include "xfs_rmap.h"
26 #include "xfs_bmap.h"
27 #include "xfs_bmap_util.h"
28 #include "scrub/xfs_scrub.h"
29 #include "scrub/scrub.h"
30 #include "scrub/common.h"
31 #include "scrub/btree.h"
32 #include "scrub/trace.h"
33
34 /*
35 * Grab total control of the inode metadata. It doesn't matter here if
36 * the file data is still changing; exclusive access to the metadata is
37 * the goal.
38 */
39 int
xchk_setup_inode(struct xfs_scrub * sc,struct xfs_inode * ip)40 xchk_setup_inode(
41 struct xfs_scrub *sc,
42 struct xfs_inode *ip)
43 {
44 int error;
45
46 /*
47 * Try to get the inode. If the verifiers fail, we try again
48 * in raw mode.
49 */
50 error = xchk_get_inode(sc, ip);
51 switch (error) {
52 case 0:
53 break;
54 case -EFSCORRUPTED:
55 case -EFSBADCRC:
56 return xchk_trans_alloc(sc, 0);
57 default:
58 return error;
59 }
60
61 /* Got the inode, lock it and we're ready to go. */
62 sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
63 xfs_ilock(sc->ip, sc->ilock_flags);
64 error = xchk_trans_alloc(sc, 0);
65 if (error)
66 goto out;
67 sc->ilock_flags |= XFS_ILOCK_EXCL;
68 xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
69
70 out:
71 /* scrub teardown will unlock and release the inode for us */
72 return error;
73 }
74
75 /* Inode core */
76
77 /* Validate di_extsize hint. */
78 STATIC void
xchk_inode_extsize(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino,uint16_t mode,uint16_t flags)79 xchk_inode_extsize(
80 struct xfs_scrub *sc,
81 struct xfs_dinode *dip,
82 xfs_ino_t ino,
83 uint16_t mode,
84 uint16_t flags)
85 {
86 xfs_failaddr_t fa;
87
88 fa = xfs_inode_validate_extsize(sc->mp, be32_to_cpu(dip->di_extsize),
89 mode, flags);
90 if (fa)
91 xchk_ino_set_corrupt(sc, ino);
92 }
93
94 /*
95 * Validate di_cowextsize hint.
96 *
97 * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
98 * These functions must be kept in sync with each other.
99 */
100 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)101 xchk_inode_cowextsize(
102 struct xfs_scrub *sc,
103 struct xfs_dinode *dip,
104 xfs_ino_t ino,
105 uint16_t mode,
106 uint16_t flags,
107 uint64_t flags2)
108 {
109 xfs_failaddr_t fa;
110
111 fa = xfs_inode_validate_cowextsize(sc->mp,
112 be32_to_cpu(dip->di_cowextsize), mode, flags,
113 flags2);
114 if (fa)
115 xchk_ino_set_corrupt(sc, ino);
116 }
117
118 /* Make sure the di_flags make sense for the inode. */
119 STATIC void
xchk_inode_flags(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino,uint16_t mode,uint16_t flags)120 xchk_inode_flags(
121 struct xfs_scrub *sc,
122 struct xfs_dinode *dip,
123 xfs_ino_t ino,
124 uint16_t mode,
125 uint16_t flags)
126 {
127 struct xfs_mount *mp = sc->mp;
128
129 /* di_flags are all taken, last bit cannot be used */
130 if (flags & ~XFS_DIFLAG_ANY)
131 goto bad;
132
133 /* rt flags require rt device */
134 if ((flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
135 goto bad;
136
137 /* new rt bitmap flag only valid for rbmino */
138 if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
139 goto bad;
140
141 /* directory-only flags */
142 if ((flags & (XFS_DIFLAG_RTINHERIT |
143 XFS_DIFLAG_EXTSZINHERIT |
144 XFS_DIFLAG_PROJINHERIT |
145 XFS_DIFLAG_NOSYMLINKS)) &&
146 !S_ISDIR(mode))
147 goto bad;
148
149 /* file-only flags */
150 if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
151 !S_ISREG(mode))
152 goto bad;
153
154 /* filestreams and rt make no sense */
155 if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
156 goto bad;
157
158 return;
159 bad:
160 xchk_ino_set_corrupt(sc, ino);
161 }
162
163 /* Make sure the di_flags2 make sense for the inode. */
164 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)165 xchk_inode_flags2(
166 struct xfs_scrub *sc,
167 struct xfs_dinode *dip,
168 xfs_ino_t ino,
169 uint16_t mode,
170 uint16_t flags,
171 uint64_t flags2)
172 {
173 struct xfs_mount *mp = sc->mp;
174
175 /* Unknown di_flags2 could be from a future kernel */
176 if (flags2 & ~XFS_DIFLAG2_ANY)
177 xchk_ino_set_warning(sc, ino);
178
179 /* reflink flag requires reflink feature */
180 if ((flags2 & XFS_DIFLAG2_REFLINK) &&
181 !xfs_sb_version_hasreflink(&mp->m_sb))
182 goto bad;
183
184 /* cowextsize flag is checked w.r.t. mode separately */
185
186 /* file/dir-only flags */
187 if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
188 goto bad;
189
190 /* file-only flags */
191 if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
192 goto bad;
193
194 /* realtime and reflink make no sense, currently */
195 if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK))
196 goto bad;
197
198 /* dax and reflink make no sense, currently */
199 if ((flags2 & XFS_DIFLAG2_DAX) && (flags2 & XFS_DIFLAG2_REFLINK))
200 goto bad;
201
202 return;
203 bad:
204 xchk_ino_set_corrupt(sc, ino);
205 }
206
207 /* Scrub all the ondisk inode fields. */
208 STATIC void
xchk_dinode(struct xfs_scrub * sc,struct xfs_dinode * dip,xfs_ino_t ino)209 xchk_dinode(
210 struct xfs_scrub *sc,
211 struct xfs_dinode *dip,
212 xfs_ino_t ino)
213 {
214 struct xfs_mount *mp = sc->mp;
215 size_t fork_recs;
216 unsigned long long isize;
217 uint64_t flags2;
218 uint32_t nextents;
219 uint16_t flags;
220 uint16_t mode;
221
222 flags = be16_to_cpu(dip->di_flags);
223 if (dip->di_version >= 3)
224 flags2 = be64_to_cpu(dip->di_flags2);
225 else
226 flags2 = 0;
227
228 /* di_mode */
229 mode = be16_to_cpu(dip->di_mode);
230 switch (mode & S_IFMT) {
231 case S_IFLNK:
232 case S_IFREG:
233 case S_IFDIR:
234 case S_IFCHR:
235 case S_IFBLK:
236 case S_IFIFO:
237 case S_IFSOCK:
238 /* mode is recognized */
239 break;
240 default:
241 xchk_ino_set_corrupt(sc, ino);
242 break;
243 }
244
245 /* v1/v2 fields */
246 switch (dip->di_version) {
247 case 1:
248 /*
249 * We autoconvert v1 inodes into v2 inodes on writeout,
250 * so just mark this inode for preening.
251 */
252 xchk_ino_set_preen(sc, ino);
253 break;
254 case 2:
255 case 3:
256 if (dip->di_onlink != 0)
257 xchk_ino_set_corrupt(sc, ino);
258
259 if (dip->di_mode == 0 && sc->ip)
260 xchk_ino_set_corrupt(sc, ino);
261
262 if (dip->di_projid_hi != 0 &&
263 !xfs_sb_version_hasprojid32bit(&mp->m_sb))
264 xchk_ino_set_corrupt(sc, ino);
265 break;
266 default:
267 xchk_ino_set_corrupt(sc, ino);
268 return;
269 }
270
271 /*
272 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
273 * userspace could have created that.
274 */
275 if (dip->di_uid == cpu_to_be32(-1U) ||
276 dip->di_gid == cpu_to_be32(-1U))
277 xchk_ino_set_warning(sc, ino);
278
279 /* di_format */
280 switch (dip->di_format) {
281 case XFS_DINODE_FMT_DEV:
282 if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
283 !S_ISFIFO(mode) && !S_ISSOCK(mode))
284 xchk_ino_set_corrupt(sc, ino);
285 break;
286 case XFS_DINODE_FMT_LOCAL:
287 if (!S_ISDIR(mode) && !S_ISLNK(mode))
288 xchk_ino_set_corrupt(sc, ino);
289 break;
290 case XFS_DINODE_FMT_EXTENTS:
291 if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
292 xchk_ino_set_corrupt(sc, ino);
293 break;
294 case XFS_DINODE_FMT_BTREE:
295 if (!S_ISREG(mode) && !S_ISDIR(mode))
296 xchk_ino_set_corrupt(sc, ino);
297 break;
298 case XFS_DINODE_FMT_UUID:
299 default:
300 xchk_ino_set_corrupt(sc, ino);
301 break;
302 }
303
304 /* di_[amc]time.nsec */
305 if (be32_to_cpu(dip->di_atime.t_nsec) >= NSEC_PER_SEC)
306 xchk_ino_set_corrupt(sc, ino);
307 if (be32_to_cpu(dip->di_mtime.t_nsec) >= NSEC_PER_SEC)
308 xchk_ino_set_corrupt(sc, ino);
309 if (be32_to_cpu(dip->di_ctime.t_nsec) >= NSEC_PER_SEC)
310 xchk_ino_set_corrupt(sc, ino);
311
312 /*
313 * di_size. xfs_dinode_verify checks for things that screw up
314 * the VFS such as the upper bit being set and zero-length
315 * symlinks/directories, but we can do more here.
316 */
317 isize = be64_to_cpu(dip->di_size);
318 if (isize & (1ULL << 63))
319 xchk_ino_set_corrupt(sc, ino);
320
321 /* Devices, fifos, and sockets must have zero size */
322 if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
323 xchk_ino_set_corrupt(sc, ino);
324
325 /* Directories can't be larger than the data section size (32G) */
326 if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
327 xchk_ino_set_corrupt(sc, ino);
328
329 /* Symlinks can't be larger than SYMLINK_MAXLEN */
330 if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
331 xchk_ino_set_corrupt(sc, ino);
332
333 /*
334 * Warn if the running kernel can't handle the kinds of offsets
335 * needed to deal with the file size. In other words, if the
336 * pagecache can't cache all the blocks in this file due to
337 * overly large offsets, flag the inode for admin review.
338 */
339 if (isize >= mp->m_super->s_maxbytes)
340 xchk_ino_set_warning(sc, ino);
341
342 /* di_nblocks */
343 if (flags2 & XFS_DIFLAG2_REFLINK) {
344 ; /* nblocks can exceed dblocks */
345 } else if (flags & XFS_DIFLAG_REALTIME) {
346 /*
347 * nblocks is the sum of data extents (in the rtdev),
348 * attr extents (in the datadev), and both forks' bmbt
349 * blocks (in the datadev). This clumsy check is the
350 * best we can do without cross-referencing with the
351 * inode forks.
352 */
353 if (be64_to_cpu(dip->di_nblocks) >=
354 mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
355 xchk_ino_set_corrupt(sc, ino);
356 } else {
357 if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
358 xchk_ino_set_corrupt(sc, ino);
359 }
360
361 xchk_inode_flags(sc, dip, ino, mode, flags);
362
363 xchk_inode_extsize(sc, dip, ino, mode, flags);
364
365 /* di_nextents */
366 nextents = be32_to_cpu(dip->di_nextents);
367 fork_recs = XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
368 switch (dip->di_format) {
369 case XFS_DINODE_FMT_EXTENTS:
370 if (nextents > fork_recs)
371 xchk_ino_set_corrupt(sc, ino);
372 break;
373 case XFS_DINODE_FMT_BTREE:
374 if (nextents <= fork_recs)
375 xchk_ino_set_corrupt(sc, ino);
376 break;
377 default:
378 if (nextents != 0)
379 xchk_ino_set_corrupt(sc, ino);
380 break;
381 }
382
383 /* di_forkoff */
384 if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
385 xchk_ino_set_corrupt(sc, ino);
386 if (dip->di_anextents != 0 && dip->di_forkoff == 0)
387 xchk_ino_set_corrupt(sc, ino);
388 if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
389 xchk_ino_set_corrupt(sc, ino);
390
391 /* di_aformat */
392 if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
393 dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
394 dip->di_aformat != XFS_DINODE_FMT_BTREE)
395 xchk_ino_set_corrupt(sc, ino);
396
397 /* di_anextents */
398 nextents = be16_to_cpu(dip->di_anextents);
399 fork_recs = XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
400 switch (dip->di_aformat) {
401 case XFS_DINODE_FMT_EXTENTS:
402 if (nextents > fork_recs)
403 xchk_ino_set_corrupt(sc, ino);
404 break;
405 case XFS_DINODE_FMT_BTREE:
406 if (nextents <= fork_recs)
407 xchk_ino_set_corrupt(sc, ino);
408 break;
409 default:
410 if (nextents != 0)
411 xchk_ino_set_corrupt(sc, ino);
412 }
413
414 if (dip->di_version >= 3) {
415 if (be32_to_cpu(dip->di_crtime.t_nsec) >= NSEC_PER_SEC)
416 xchk_ino_set_corrupt(sc, ino);
417 xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
418 xchk_inode_cowextsize(sc, dip, ino, mode, flags,
419 flags2);
420 }
421 }
422
423 /*
424 * Make sure the finobt doesn't think this inode is free.
425 * We don't have to check the inobt ourselves because we got the inode via
426 * IGET_UNTRUSTED, which checks the inobt for us.
427 */
428 static void
xchk_inode_xref_finobt(struct xfs_scrub * sc,xfs_ino_t ino)429 xchk_inode_xref_finobt(
430 struct xfs_scrub *sc,
431 xfs_ino_t ino)
432 {
433 struct xfs_inobt_rec_incore rec;
434 xfs_agino_t agino;
435 int has_record;
436 int error;
437
438 if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
439 return;
440
441 agino = XFS_INO_TO_AGINO(sc->mp, ino);
442
443 /*
444 * Try to get the finobt record. If we can't get it, then we're
445 * in good shape.
446 */
447 error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
448 &has_record);
449 if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
450 !has_record)
451 return;
452
453 error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
454 if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
455 !has_record)
456 return;
457
458 /*
459 * Otherwise, make sure this record either doesn't cover this inode,
460 * or that it does but it's marked present.
461 */
462 if (rec.ir_startino > agino ||
463 rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
464 return;
465
466 if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
467 xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
468 }
469
470 /* Cross reference the inode fields with the forks. */
471 STATIC void
xchk_inode_xref_bmap(struct xfs_scrub * sc,struct xfs_dinode * dip)472 xchk_inode_xref_bmap(
473 struct xfs_scrub *sc,
474 struct xfs_dinode *dip)
475 {
476 xfs_extnum_t nextents;
477 xfs_filblks_t count;
478 xfs_filblks_t acount;
479 int error;
480
481 if (xchk_skip_xref(sc->sm))
482 return;
483
484 /* Walk all the extents to check nextents/naextents/nblocks. */
485 error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
486 &nextents, &count);
487 if (!xchk_should_check_xref(sc, &error, NULL))
488 return;
489 if (nextents < be32_to_cpu(dip->di_nextents))
490 xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
491
492 error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
493 &nextents, &acount);
494 if (!xchk_should_check_xref(sc, &error, NULL))
495 return;
496 if (nextents != be16_to_cpu(dip->di_anextents))
497 xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
498
499 /* Check nblocks against the inode. */
500 if (count + acount != be64_to_cpu(dip->di_nblocks))
501 xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
502 }
503
504 /* Cross-reference with the other btrees. */
505 STATIC void
xchk_inode_xref(struct xfs_scrub * sc,xfs_ino_t ino,struct xfs_dinode * dip)506 xchk_inode_xref(
507 struct xfs_scrub *sc,
508 xfs_ino_t ino,
509 struct xfs_dinode *dip)
510 {
511 struct xfs_owner_info oinfo;
512 xfs_agnumber_t agno;
513 xfs_agblock_t agbno;
514 int error;
515
516 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
517 return;
518
519 agno = XFS_INO_TO_AGNO(sc->mp, ino);
520 agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
521
522 error = xchk_ag_init(sc, agno, &sc->sa);
523 if (!xchk_xref_process_error(sc, agno, agbno, &error))
524 return;
525
526 xchk_xref_is_used_space(sc, agbno, 1);
527 xchk_inode_xref_finobt(sc, ino);
528 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES);
529 xchk_xref_is_owned_by(sc, agbno, 1, &oinfo);
530 xchk_xref_is_not_shared(sc, agbno, 1);
531 xchk_inode_xref_bmap(sc, dip);
532
533 xchk_ag_free(sc, &sc->sa);
534 }
535
536 /*
537 * If the reflink iflag disagrees with a scan for shared data fork extents,
538 * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
539 * any shared extents). We already checked for reflink iflag set on a non
540 * reflink filesystem.
541 */
542 static void
xchk_inode_check_reflink_iflag(struct xfs_scrub * sc,xfs_ino_t ino)543 xchk_inode_check_reflink_iflag(
544 struct xfs_scrub *sc,
545 xfs_ino_t ino)
546 {
547 struct xfs_mount *mp = sc->mp;
548 bool has_shared;
549 int error;
550
551 if (!xfs_sb_version_hasreflink(&mp->m_sb))
552 return;
553
554 error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
555 &has_shared);
556 if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
557 XFS_INO_TO_AGBNO(mp, ino), &error))
558 return;
559 if (xfs_is_reflink_inode(sc->ip) && !has_shared)
560 xchk_ino_set_preen(sc, ino);
561 else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
562 xchk_ino_set_corrupt(sc, ino);
563 }
564
565 /* Scrub an inode. */
566 int
xchk_inode(struct xfs_scrub * sc)567 xchk_inode(
568 struct xfs_scrub *sc)
569 {
570 struct xfs_dinode di;
571 int error = 0;
572
573 /*
574 * If sc->ip is NULL, that means that the setup function called
575 * xfs_iget to look up the inode. xfs_iget returned a EFSCORRUPTED
576 * and a NULL inode, so flag the corruption error and return.
577 */
578 if (!sc->ip) {
579 xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
580 return 0;
581 }
582
583 /* Scrub the inode core. */
584 xfs_inode_to_disk(sc->ip, &di, 0);
585 xchk_dinode(sc, &di, sc->ip->i_ino);
586 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
587 goto out;
588
589 /*
590 * Look for discrepancies between file's data blocks and the reflink
591 * iflag. We already checked the iflag against the file mode when
592 * we scrubbed the dinode.
593 */
594 if (S_ISREG(VFS_I(sc->ip)->i_mode))
595 xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
596
597 xchk_inode_xref(sc, sc->ip->i_ino, &di);
598 out:
599 return error;
600 }
601