xref: /linux/fs/xfs/scrub/inode.c (revision 4f3c8320c78cdd11c8fdd23c33787407f719322e)
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_btree.h"
13 #include "xfs_log_format.h"
14 #include "xfs_inode.h"
15 #include "xfs_ialloc.h"
16 #include "xfs_da_format.h"
17 #include "xfs_reflink.h"
18 #include "xfs_rmap.h"
19 #include "xfs_bmap_util.h"
20 #include "scrub/scrub.h"
21 #include "scrub/common.h"
22 #include "scrub/btree.h"
23 
24 /*
25  * Grab total control of the inode metadata.  It doesn't matter here if
26  * the file data is still changing; exclusive access to the metadata is
27  * the goal.
28  */
29 int
30 xchk_setup_inode(
31 	struct xfs_scrub	*sc,
32 	struct xfs_inode	*ip)
33 {
34 	int			error;
35 
36 	/*
37 	 * Try to get the inode.  If the verifiers fail, we try again
38 	 * in raw mode.
39 	 */
40 	error = xchk_get_inode(sc, ip);
41 	switch (error) {
42 	case 0:
43 		break;
44 	case -EFSCORRUPTED:
45 	case -EFSBADCRC:
46 		return xchk_trans_alloc(sc, 0);
47 	default:
48 		return error;
49 	}
50 
51 	/* Got the inode, lock it and we're ready to go. */
52 	sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
53 	xfs_ilock(sc->ip, sc->ilock_flags);
54 	error = xchk_trans_alloc(sc, 0);
55 	if (error)
56 		goto out;
57 	sc->ilock_flags |= XFS_ILOCK_EXCL;
58 	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
59 
60 out:
61 	/* scrub teardown will unlock and release the inode for us */
62 	return error;
63 }
64 
65 /* Inode core */
66 
67 /* Validate di_extsize hint. */
68 STATIC void
69 xchk_inode_extsize(
70 	struct xfs_scrub	*sc,
71 	struct xfs_dinode	*dip,
72 	xfs_ino_t		ino,
73 	uint16_t		mode,
74 	uint16_t		flags)
75 {
76 	xfs_failaddr_t		fa;
77 
78 	fa = xfs_inode_validate_extsize(sc->mp, be32_to_cpu(dip->di_extsize),
79 			mode, flags);
80 	if (fa)
81 		xchk_ino_set_corrupt(sc, ino);
82 }
83 
84 /*
85  * Validate di_cowextsize hint.
86  *
87  * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
88  * These functions must be kept in sync with each other.
89  */
90 STATIC void
91 xchk_inode_cowextsize(
92 	struct xfs_scrub	*sc,
93 	struct xfs_dinode	*dip,
94 	xfs_ino_t		ino,
95 	uint16_t		mode,
96 	uint16_t		flags,
97 	uint64_t		flags2)
98 {
99 	xfs_failaddr_t		fa;
100 
101 	fa = xfs_inode_validate_cowextsize(sc->mp,
102 			be32_to_cpu(dip->di_cowextsize), mode, flags,
103 			flags2);
104 	if (fa)
105 		xchk_ino_set_corrupt(sc, ino);
106 }
107 
108 /* Make sure the di_flags make sense for the inode. */
109 STATIC void
110 xchk_inode_flags(
111 	struct xfs_scrub	*sc,
112 	struct xfs_dinode	*dip,
113 	xfs_ino_t		ino,
114 	uint16_t		mode,
115 	uint16_t		flags)
116 {
117 	struct xfs_mount	*mp = sc->mp;
118 
119 	/* di_flags are all taken, last bit cannot be used */
120 	if (flags & ~XFS_DIFLAG_ANY)
121 		goto bad;
122 
123 	/* rt flags require rt device */
124 	if ((flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
125 		goto bad;
126 
127 	/* new rt bitmap flag only valid for rbmino */
128 	if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
129 		goto bad;
130 
131 	/* directory-only flags */
132 	if ((flags & (XFS_DIFLAG_RTINHERIT |
133 		     XFS_DIFLAG_EXTSZINHERIT |
134 		     XFS_DIFLAG_PROJINHERIT |
135 		     XFS_DIFLAG_NOSYMLINKS)) &&
136 	    !S_ISDIR(mode))
137 		goto bad;
138 
139 	/* file-only flags */
140 	if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
141 	    !S_ISREG(mode))
142 		goto bad;
143 
144 	/* filestreams and rt make no sense */
145 	if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
146 		goto bad;
147 
148 	return;
149 bad:
150 	xchk_ino_set_corrupt(sc, ino);
151 }
152 
153 /* Make sure the di_flags2 make sense for the inode. */
154 STATIC void
155 xchk_inode_flags2(
156 	struct xfs_scrub	*sc,
157 	struct xfs_dinode	*dip,
158 	xfs_ino_t		ino,
159 	uint16_t		mode,
160 	uint16_t		flags,
161 	uint64_t		flags2)
162 {
163 	struct xfs_mount	*mp = sc->mp;
164 
165 	/* Unknown di_flags2 could be from a future kernel */
166 	if (flags2 & ~XFS_DIFLAG2_ANY)
167 		xchk_ino_set_warning(sc, ino);
168 
169 	/* reflink flag requires reflink feature */
170 	if ((flags2 & XFS_DIFLAG2_REFLINK) &&
171 	    !xfs_sb_version_hasreflink(&mp->m_sb))
172 		goto bad;
173 
174 	/* cowextsize flag is checked w.r.t. mode separately */
175 
176 	/* file/dir-only flags */
177 	if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
178 		goto bad;
179 
180 	/* file-only flags */
181 	if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
182 		goto bad;
183 
184 	/* realtime and reflink make no sense, currently */
185 	if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK))
186 		goto bad;
187 
188 	/* no bigtime iflag without the bigtime feature */
189 	if (xfs_dinode_has_bigtime(dip) &&
190 	    !xfs_sb_version_hasbigtime(&mp->m_sb))
191 		goto bad;
192 
193 	return;
194 bad:
195 	xchk_ino_set_corrupt(sc, ino);
196 }
197 
198 static inline void
199 xchk_dinode_nsec(
200 	struct xfs_scrub	*sc,
201 	xfs_ino_t		ino,
202 	struct xfs_dinode	*dip,
203 	const xfs_timestamp_t	ts)
204 {
205 	struct timespec64	tv;
206 
207 	tv = xfs_inode_from_disk_ts(dip, ts);
208 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
209 		xchk_ino_set_corrupt(sc, ino);
210 }
211 
212 /* Scrub all the ondisk inode fields. */
213 STATIC void
214 xchk_dinode(
215 	struct xfs_scrub	*sc,
216 	struct xfs_dinode	*dip,
217 	xfs_ino_t		ino)
218 {
219 	struct xfs_mount	*mp = sc->mp;
220 	size_t			fork_recs;
221 	unsigned long long	isize;
222 	uint64_t		flags2;
223 	uint32_t		nextents;
224 	uint16_t		flags;
225 	uint16_t		mode;
226 
227 	flags = be16_to_cpu(dip->di_flags);
228 	if (dip->di_version >= 3)
229 		flags2 = be64_to_cpu(dip->di_flags2);
230 	else
231 		flags2 = 0;
232 
233 	/* di_mode */
234 	mode = be16_to_cpu(dip->di_mode);
235 	switch (mode & S_IFMT) {
236 	case S_IFLNK:
237 	case S_IFREG:
238 	case S_IFDIR:
239 	case S_IFCHR:
240 	case S_IFBLK:
241 	case S_IFIFO:
242 	case S_IFSOCK:
243 		/* mode is recognized */
244 		break;
245 	default:
246 		xchk_ino_set_corrupt(sc, ino);
247 		break;
248 	}
249 
250 	/* v1/v2 fields */
251 	switch (dip->di_version) {
252 	case 1:
253 		/*
254 		 * We autoconvert v1 inodes into v2 inodes on writeout,
255 		 * so just mark this inode for preening.
256 		 */
257 		xchk_ino_set_preen(sc, ino);
258 		break;
259 	case 2:
260 	case 3:
261 		if (dip->di_onlink != 0)
262 			xchk_ino_set_corrupt(sc, ino);
263 
264 		if (dip->di_mode == 0 && sc->ip)
265 			xchk_ino_set_corrupt(sc, ino);
266 
267 		if (dip->di_projid_hi != 0 &&
268 		    !xfs_sb_version_hasprojid32bit(&mp->m_sb))
269 			xchk_ino_set_corrupt(sc, ino);
270 		break;
271 	default:
272 		xchk_ino_set_corrupt(sc, ino);
273 		return;
274 	}
275 
276 	/*
277 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
278 	 * userspace could have created that.
279 	 */
280 	if (dip->di_uid == cpu_to_be32(-1U) ||
281 	    dip->di_gid == cpu_to_be32(-1U))
282 		xchk_ino_set_warning(sc, ino);
283 
284 	/* di_format */
285 	switch (dip->di_format) {
286 	case XFS_DINODE_FMT_DEV:
287 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
288 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
289 			xchk_ino_set_corrupt(sc, ino);
290 		break;
291 	case XFS_DINODE_FMT_LOCAL:
292 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
293 			xchk_ino_set_corrupt(sc, ino);
294 		break;
295 	case XFS_DINODE_FMT_EXTENTS:
296 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
297 			xchk_ino_set_corrupt(sc, ino);
298 		break;
299 	case XFS_DINODE_FMT_BTREE:
300 		if (!S_ISREG(mode) && !S_ISDIR(mode))
301 			xchk_ino_set_corrupt(sc, ino);
302 		break;
303 	case XFS_DINODE_FMT_UUID:
304 	default:
305 		xchk_ino_set_corrupt(sc, ino);
306 		break;
307 	}
308 
309 	/* di_[amc]time.nsec */
310 	xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
311 	xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
312 	xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
313 
314 	/*
315 	 * di_size.  xfs_dinode_verify checks for things that screw up
316 	 * the VFS such as the upper bit being set and zero-length
317 	 * symlinks/directories, but we can do more here.
318 	 */
319 	isize = be64_to_cpu(dip->di_size);
320 	if (isize & (1ULL << 63))
321 		xchk_ino_set_corrupt(sc, ino);
322 
323 	/* Devices, fifos, and sockets must have zero size */
324 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
325 		xchk_ino_set_corrupt(sc, ino);
326 
327 	/* Directories can't be larger than the data section size (32G) */
328 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
329 		xchk_ino_set_corrupt(sc, ino);
330 
331 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
332 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
333 		xchk_ino_set_corrupt(sc, ino);
334 
335 	/*
336 	 * Warn if the running kernel can't handle the kinds of offsets
337 	 * needed to deal with the file size.  In other words, if the
338 	 * pagecache can't cache all the blocks in this file due to
339 	 * overly large offsets, flag the inode for admin review.
340 	 */
341 	if (isize >= mp->m_super->s_maxbytes)
342 		xchk_ino_set_warning(sc, ino);
343 
344 	/* di_nblocks */
345 	if (flags2 & XFS_DIFLAG2_REFLINK) {
346 		; /* nblocks can exceed dblocks */
347 	} else if (flags & XFS_DIFLAG_REALTIME) {
348 		/*
349 		 * nblocks is the sum of data extents (in the rtdev),
350 		 * attr extents (in the datadev), and both forks' bmbt
351 		 * blocks (in the datadev).  This clumsy check is the
352 		 * best we can do without cross-referencing with the
353 		 * inode forks.
354 		 */
355 		if (be64_to_cpu(dip->di_nblocks) >=
356 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
357 			xchk_ino_set_corrupt(sc, ino);
358 	} else {
359 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
360 			xchk_ino_set_corrupt(sc, ino);
361 	}
362 
363 	xchk_inode_flags(sc, dip, ino, mode, flags);
364 
365 	xchk_inode_extsize(sc, dip, ino, mode, flags);
366 
367 	/* di_nextents */
368 	nextents = be32_to_cpu(dip->di_nextents);
369 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
370 	switch (dip->di_format) {
371 	case XFS_DINODE_FMT_EXTENTS:
372 		if (nextents > fork_recs)
373 			xchk_ino_set_corrupt(sc, ino);
374 		break;
375 	case XFS_DINODE_FMT_BTREE:
376 		if (nextents <= fork_recs)
377 			xchk_ino_set_corrupt(sc, ino);
378 		break;
379 	default:
380 		if (nextents != 0)
381 			xchk_ino_set_corrupt(sc, ino);
382 		break;
383 	}
384 
385 	/* di_forkoff */
386 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
387 		xchk_ino_set_corrupt(sc, ino);
388 	if (dip->di_anextents != 0 && dip->di_forkoff == 0)
389 		xchk_ino_set_corrupt(sc, ino);
390 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
391 		xchk_ino_set_corrupt(sc, ino);
392 
393 	/* di_aformat */
394 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
395 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
396 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
397 		xchk_ino_set_corrupt(sc, ino);
398 
399 	/* di_anextents */
400 	nextents = be16_to_cpu(dip->di_anextents);
401 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
402 	switch (dip->di_aformat) {
403 	case XFS_DINODE_FMT_EXTENTS:
404 		if (nextents > fork_recs)
405 			xchk_ino_set_corrupt(sc, ino);
406 		break;
407 	case XFS_DINODE_FMT_BTREE:
408 		if (nextents <= fork_recs)
409 			xchk_ino_set_corrupt(sc, ino);
410 		break;
411 	default:
412 		if (nextents != 0)
413 			xchk_ino_set_corrupt(sc, ino);
414 	}
415 
416 	if (dip->di_version >= 3) {
417 		xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
418 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
419 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
420 				flags2);
421 	}
422 }
423 
424 /*
425  * Make sure the finobt doesn't think this inode is free.
426  * We don't have to check the inobt ourselves because we got the inode via
427  * IGET_UNTRUSTED, which checks the inobt for us.
428  */
429 static void
430 xchk_inode_xref_finobt(
431 	struct xfs_scrub		*sc,
432 	xfs_ino_t			ino)
433 {
434 	struct xfs_inobt_rec_incore	rec;
435 	xfs_agino_t			agino;
436 	int				has_record;
437 	int				error;
438 
439 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
440 		return;
441 
442 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
443 
444 	/*
445 	 * Try to get the finobt record.  If we can't get it, then we're
446 	 * in good shape.
447 	 */
448 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
449 			&has_record);
450 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
451 	    !has_record)
452 		return;
453 
454 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
455 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
456 	    !has_record)
457 		return;
458 
459 	/*
460 	 * Otherwise, make sure this record either doesn't cover this inode,
461 	 * or that it does but it's marked present.
462 	 */
463 	if (rec.ir_startino > agino ||
464 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
465 		return;
466 
467 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
468 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
469 }
470 
471 /* Cross reference the inode fields with the forks. */
472 STATIC void
473 xchk_inode_xref_bmap(
474 	struct xfs_scrub	*sc,
475 	struct xfs_dinode	*dip)
476 {
477 	xfs_extnum_t		nextents;
478 	xfs_filblks_t		count;
479 	xfs_filblks_t		acount;
480 	int			error;
481 
482 	if (xchk_skip_xref(sc->sm))
483 		return;
484 
485 	/* Walk all the extents to check nextents/naextents/nblocks. */
486 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
487 			&nextents, &count);
488 	if (!xchk_should_check_xref(sc, &error, NULL))
489 		return;
490 	if (nextents < be32_to_cpu(dip->di_nextents))
491 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
492 
493 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
494 			&nextents, &acount);
495 	if (!xchk_should_check_xref(sc, &error, NULL))
496 		return;
497 	if (nextents != be16_to_cpu(dip->di_anextents))
498 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
499 
500 	/* Check nblocks against the inode. */
501 	if (count + acount != be64_to_cpu(dip->di_nblocks))
502 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
503 }
504 
505 /* Cross-reference with the other btrees. */
506 STATIC void
507 xchk_inode_xref(
508 	struct xfs_scrub	*sc,
509 	xfs_ino_t		ino,
510 	struct xfs_dinode	*dip)
511 {
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 	xchk_xref_is_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
529 	xchk_xref_is_not_shared(sc, agbno, 1);
530 	xchk_inode_xref_bmap(sc, dip);
531 
532 	xchk_ag_free(sc, &sc->sa);
533 }
534 
535 /*
536  * If the reflink iflag disagrees with a scan for shared data fork extents,
537  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
538  * any shared extents).  We already checked for reflink iflag set on a non
539  * reflink filesystem.
540  */
541 static void
542 xchk_inode_check_reflink_iflag(
543 	struct xfs_scrub	*sc,
544 	xfs_ino_t		ino)
545 {
546 	struct xfs_mount	*mp = sc->mp;
547 	bool			has_shared;
548 	int			error;
549 
550 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
551 		return;
552 
553 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
554 			&has_shared);
555 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
556 			XFS_INO_TO_AGBNO(mp, ino), &error))
557 		return;
558 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
559 		xchk_ino_set_preen(sc, ino);
560 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
561 		xchk_ino_set_corrupt(sc, ino);
562 }
563 
564 /* Scrub an inode. */
565 int
566 xchk_inode(
567 	struct xfs_scrub	*sc)
568 {
569 	struct xfs_dinode	di;
570 	int			error = 0;
571 
572 	/*
573 	 * If sc->ip is NULL, that means that the setup function called
574 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
575 	 * and a NULL inode, so flag the corruption error and return.
576 	 */
577 	if (!sc->ip) {
578 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
579 		return 0;
580 	}
581 
582 	/* Scrub the inode core. */
583 	xfs_inode_to_disk(sc->ip, &di, 0);
584 	xchk_dinode(sc, &di, sc->ip->i_ino);
585 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
586 		goto out;
587 
588 	/*
589 	 * Look for discrepancies between file's data blocks and the reflink
590 	 * iflag.  We already checked the iflag against the file mode when
591 	 * we scrubbed the dinode.
592 	 */
593 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
594 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
595 
596 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
597 out:
598 	return error;
599 }
600