xref: /openbmc/linux/fs/xfs/scrub/inode.c (revision 6c9111bc)
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 | XFS_DIFLAG_RTINHERIT)) &&
125 	    !mp->m_rtdev_targp)
126 		goto bad;
127 
128 	/* new rt bitmap flag only valid for rbmino */
129 	if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
130 		goto bad;
131 
132 	/* directory-only flags */
133 	if ((flags & (XFS_DIFLAG_RTINHERIT |
134 		     XFS_DIFLAG_EXTSZINHERIT |
135 		     XFS_DIFLAG_PROJINHERIT |
136 		     XFS_DIFLAG_NOSYMLINKS)) &&
137 	    !S_ISDIR(mode))
138 		goto bad;
139 
140 	/* file-only flags */
141 	if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
142 	    !S_ISREG(mode))
143 		goto bad;
144 
145 	/* filestreams and rt make no sense */
146 	if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
147 		goto bad;
148 
149 	return;
150 bad:
151 	xchk_ino_set_corrupt(sc, ino);
152 }
153 
154 /* Make sure the di_flags2 make sense for the inode. */
155 STATIC void
156 xchk_inode_flags2(
157 	struct xfs_scrub	*sc,
158 	struct xfs_dinode	*dip,
159 	xfs_ino_t		ino,
160 	uint16_t		mode,
161 	uint16_t		flags,
162 	uint64_t		flags2)
163 {
164 	struct xfs_mount	*mp = sc->mp;
165 
166 	/* Unknown di_flags2 could be from a future kernel */
167 	if (flags2 & ~XFS_DIFLAG2_ANY)
168 		xchk_ino_set_warning(sc, ino);
169 
170 	/* reflink flag requires reflink feature */
171 	if ((flags2 & XFS_DIFLAG2_REFLINK) &&
172 	    !xfs_sb_version_hasreflink(&mp->m_sb))
173 		goto bad;
174 
175 	/* cowextsize flag is checked w.r.t. mode separately */
176 
177 	/* file/dir-only flags */
178 	if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
179 		goto bad;
180 
181 	/* file-only flags */
182 	if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
183 		goto bad;
184 
185 	/* realtime and reflink make no sense, currently */
186 	if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK))
187 		goto bad;
188 
189 	/* dax and reflink make no sense, currently */
190 	if ((flags2 & XFS_DIFLAG2_DAX) && (flags2 & XFS_DIFLAG2_REFLINK))
191 		goto bad;
192 
193 	/* no bigtime iflag without the bigtime feature */
194 	if (xfs_dinode_has_bigtime(dip) &&
195 	    !xfs_sb_version_hasbigtime(&mp->m_sb))
196 		goto bad;
197 
198 	return;
199 bad:
200 	xchk_ino_set_corrupt(sc, ino);
201 }
202 
203 static inline void
204 xchk_dinode_nsec(
205 	struct xfs_scrub	*sc,
206 	xfs_ino_t		ino,
207 	struct xfs_dinode	*dip,
208 	const xfs_timestamp_t	ts)
209 {
210 	struct timespec64	tv;
211 
212 	tv = xfs_inode_from_disk_ts(dip, ts);
213 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
214 		xchk_ino_set_corrupt(sc, ino);
215 }
216 
217 /* Scrub all the ondisk inode fields. */
218 STATIC void
219 xchk_dinode(
220 	struct xfs_scrub	*sc,
221 	struct xfs_dinode	*dip,
222 	xfs_ino_t		ino)
223 {
224 	struct xfs_mount	*mp = sc->mp;
225 	size_t			fork_recs;
226 	unsigned long long	isize;
227 	uint64_t		flags2;
228 	uint32_t		nextents;
229 	uint16_t		flags;
230 	uint16_t		mode;
231 
232 	flags = be16_to_cpu(dip->di_flags);
233 	if (dip->di_version >= 3)
234 		flags2 = be64_to_cpu(dip->di_flags2);
235 	else
236 		flags2 = 0;
237 
238 	/* di_mode */
239 	mode = be16_to_cpu(dip->di_mode);
240 	switch (mode & S_IFMT) {
241 	case S_IFLNK:
242 	case S_IFREG:
243 	case S_IFDIR:
244 	case S_IFCHR:
245 	case S_IFBLK:
246 	case S_IFIFO:
247 	case S_IFSOCK:
248 		/* mode is recognized */
249 		break;
250 	default:
251 		xchk_ino_set_corrupt(sc, ino);
252 		break;
253 	}
254 
255 	/* v1/v2 fields */
256 	switch (dip->di_version) {
257 	case 1:
258 		/*
259 		 * We autoconvert v1 inodes into v2 inodes on writeout,
260 		 * so just mark this inode for preening.
261 		 */
262 		xchk_ino_set_preen(sc, ino);
263 		break;
264 	case 2:
265 	case 3:
266 		if (dip->di_onlink != 0)
267 			xchk_ino_set_corrupt(sc, ino);
268 
269 		if (dip->di_mode == 0 && sc->ip)
270 			xchk_ino_set_corrupt(sc, ino);
271 
272 		if (dip->di_projid_hi != 0 &&
273 		    !xfs_sb_version_hasprojid32bit(&mp->m_sb))
274 			xchk_ino_set_corrupt(sc, ino);
275 		break;
276 	default:
277 		xchk_ino_set_corrupt(sc, ino);
278 		return;
279 	}
280 
281 	/*
282 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
283 	 * userspace could have created that.
284 	 */
285 	if (dip->di_uid == cpu_to_be32(-1U) ||
286 	    dip->di_gid == cpu_to_be32(-1U))
287 		xchk_ino_set_warning(sc, ino);
288 
289 	/* di_format */
290 	switch (dip->di_format) {
291 	case XFS_DINODE_FMT_DEV:
292 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
293 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
294 			xchk_ino_set_corrupt(sc, ino);
295 		break;
296 	case XFS_DINODE_FMT_LOCAL:
297 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
298 			xchk_ino_set_corrupt(sc, ino);
299 		break;
300 	case XFS_DINODE_FMT_EXTENTS:
301 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
302 			xchk_ino_set_corrupt(sc, ino);
303 		break;
304 	case XFS_DINODE_FMT_BTREE:
305 		if (!S_ISREG(mode) && !S_ISDIR(mode))
306 			xchk_ino_set_corrupt(sc, ino);
307 		break;
308 	case XFS_DINODE_FMT_UUID:
309 	default:
310 		xchk_ino_set_corrupt(sc, ino);
311 		break;
312 	}
313 
314 	/* di_[amc]time.nsec */
315 	xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
316 	xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
317 	xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
318 
319 	/*
320 	 * di_size.  xfs_dinode_verify checks for things that screw up
321 	 * the VFS such as the upper bit being set and zero-length
322 	 * symlinks/directories, but we can do more here.
323 	 */
324 	isize = be64_to_cpu(dip->di_size);
325 	if (isize & (1ULL << 63))
326 		xchk_ino_set_corrupt(sc, ino);
327 
328 	/* Devices, fifos, and sockets must have zero size */
329 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
330 		xchk_ino_set_corrupt(sc, ino);
331 
332 	/* Directories can't be larger than the data section size (32G) */
333 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
334 		xchk_ino_set_corrupt(sc, ino);
335 
336 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
337 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
338 		xchk_ino_set_corrupt(sc, ino);
339 
340 	/*
341 	 * Warn if the running kernel can't handle the kinds of offsets
342 	 * needed to deal with the file size.  In other words, if the
343 	 * pagecache can't cache all the blocks in this file due to
344 	 * overly large offsets, flag the inode for admin review.
345 	 */
346 	if (isize >= mp->m_super->s_maxbytes)
347 		xchk_ino_set_warning(sc, ino);
348 
349 	/* di_nblocks */
350 	if (flags2 & XFS_DIFLAG2_REFLINK) {
351 		; /* nblocks can exceed dblocks */
352 	} else if (flags & XFS_DIFLAG_REALTIME) {
353 		/*
354 		 * nblocks is the sum of data extents (in the rtdev),
355 		 * attr extents (in the datadev), and both forks' bmbt
356 		 * blocks (in the datadev).  This clumsy check is the
357 		 * best we can do without cross-referencing with the
358 		 * inode forks.
359 		 */
360 		if (be64_to_cpu(dip->di_nblocks) >=
361 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
362 			xchk_ino_set_corrupt(sc, ino);
363 	} else {
364 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
365 			xchk_ino_set_corrupt(sc, ino);
366 	}
367 
368 	xchk_inode_flags(sc, dip, ino, mode, flags);
369 
370 	xchk_inode_extsize(sc, dip, ino, mode, flags);
371 
372 	/* di_nextents */
373 	nextents = be32_to_cpu(dip->di_nextents);
374 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
375 	switch (dip->di_format) {
376 	case XFS_DINODE_FMT_EXTENTS:
377 		if (nextents > fork_recs)
378 			xchk_ino_set_corrupt(sc, ino);
379 		break;
380 	case XFS_DINODE_FMT_BTREE:
381 		if (nextents <= fork_recs)
382 			xchk_ino_set_corrupt(sc, ino);
383 		break;
384 	default:
385 		if (nextents != 0)
386 			xchk_ino_set_corrupt(sc, ino);
387 		break;
388 	}
389 
390 	/* di_forkoff */
391 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
392 		xchk_ino_set_corrupt(sc, ino);
393 	if (dip->di_anextents != 0 && dip->di_forkoff == 0)
394 		xchk_ino_set_corrupt(sc, ino);
395 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
396 		xchk_ino_set_corrupt(sc, ino);
397 
398 	/* di_aformat */
399 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
400 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
401 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
402 		xchk_ino_set_corrupt(sc, ino);
403 
404 	/* di_anextents */
405 	nextents = be16_to_cpu(dip->di_anextents);
406 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
407 	switch (dip->di_aformat) {
408 	case XFS_DINODE_FMT_EXTENTS:
409 		if (nextents > fork_recs)
410 			xchk_ino_set_corrupt(sc, ino);
411 		break;
412 	case XFS_DINODE_FMT_BTREE:
413 		if (nextents <= fork_recs)
414 			xchk_ino_set_corrupt(sc, ino);
415 		break;
416 	default:
417 		if (nextents != 0)
418 			xchk_ino_set_corrupt(sc, ino);
419 	}
420 
421 	if (dip->di_version >= 3) {
422 		xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
423 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
424 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
425 				flags2);
426 	}
427 }
428 
429 /*
430  * Make sure the finobt doesn't think this inode is free.
431  * We don't have to check the inobt ourselves because we got the inode via
432  * IGET_UNTRUSTED, which checks the inobt for us.
433  */
434 static void
435 xchk_inode_xref_finobt(
436 	struct xfs_scrub		*sc,
437 	xfs_ino_t			ino)
438 {
439 	struct xfs_inobt_rec_incore	rec;
440 	xfs_agino_t			agino;
441 	int				has_record;
442 	int				error;
443 
444 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
445 		return;
446 
447 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
448 
449 	/*
450 	 * Try to get the finobt record.  If we can't get it, then we're
451 	 * in good shape.
452 	 */
453 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
454 			&has_record);
455 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
456 	    !has_record)
457 		return;
458 
459 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
460 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
461 	    !has_record)
462 		return;
463 
464 	/*
465 	 * Otherwise, make sure this record either doesn't cover this inode,
466 	 * or that it does but it's marked present.
467 	 */
468 	if (rec.ir_startino > agino ||
469 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
470 		return;
471 
472 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
473 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
474 }
475 
476 /* Cross reference the inode fields with the forks. */
477 STATIC void
478 xchk_inode_xref_bmap(
479 	struct xfs_scrub	*sc,
480 	struct xfs_dinode	*dip)
481 {
482 	xfs_extnum_t		nextents;
483 	xfs_filblks_t		count;
484 	xfs_filblks_t		acount;
485 	int			error;
486 
487 	if (xchk_skip_xref(sc->sm))
488 		return;
489 
490 	/* Walk all the extents to check nextents/naextents/nblocks. */
491 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
492 			&nextents, &count);
493 	if (!xchk_should_check_xref(sc, &error, NULL))
494 		return;
495 	if (nextents < be32_to_cpu(dip->di_nextents))
496 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
497 
498 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
499 			&nextents, &acount);
500 	if (!xchk_should_check_xref(sc, &error, NULL))
501 		return;
502 	if (nextents != be16_to_cpu(dip->di_anextents))
503 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
504 
505 	/* Check nblocks against the inode. */
506 	if (count + acount != be64_to_cpu(dip->di_nblocks))
507 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
508 }
509 
510 /* Cross-reference with the other btrees. */
511 STATIC void
512 xchk_inode_xref(
513 	struct xfs_scrub	*sc,
514 	xfs_ino_t		ino,
515 	struct xfs_dinode	*dip)
516 {
517 	xfs_agnumber_t		agno;
518 	xfs_agblock_t		agbno;
519 	int			error;
520 
521 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
522 		return;
523 
524 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
525 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
526 
527 	error = xchk_ag_init(sc, agno, &sc->sa);
528 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
529 		return;
530 
531 	xchk_xref_is_used_space(sc, agbno, 1);
532 	xchk_inode_xref_finobt(sc, ino);
533 	xchk_xref_is_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
534 	xchk_xref_is_not_shared(sc, agbno, 1);
535 	xchk_inode_xref_bmap(sc, dip);
536 
537 	xchk_ag_free(sc, &sc->sa);
538 }
539 
540 /*
541  * If the reflink iflag disagrees with a scan for shared data fork extents,
542  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
543  * any shared extents).  We already checked for reflink iflag set on a non
544  * reflink filesystem.
545  */
546 static void
547 xchk_inode_check_reflink_iflag(
548 	struct xfs_scrub	*sc,
549 	xfs_ino_t		ino)
550 {
551 	struct xfs_mount	*mp = sc->mp;
552 	bool			has_shared;
553 	int			error;
554 
555 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
556 		return;
557 
558 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
559 			&has_shared);
560 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
561 			XFS_INO_TO_AGBNO(mp, ino), &error))
562 		return;
563 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
564 		xchk_ino_set_preen(sc, ino);
565 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
566 		xchk_ino_set_corrupt(sc, ino);
567 }
568 
569 /* Scrub an inode. */
570 int
571 xchk_inode(
572 	struct xfs_scrub	*sc)
573 {
574 	struct xfs_dinode	di;
575 	int			error = 0;
576 
577 	/*
578 	 * If sc->ip is NULL, that means that the setup function called
579 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
580 	 * and a NULL inode, so flag the corruption error and return.
581 	 */
582 	if (!sc->ip) {
583 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
584 		return 0;
585 	}
586 
587 	/* Scrub the inode core. */
588 	xfs_inode_to_disk(sc->ip, &di, 0);
589 	xchk_dinode(sc, &di, sc->ip->i_ino);
590 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
591 		goto out;
592 
593 	/*
594 	 * Look for discrepancies between file's data blocks and the reflink
595 	 * iflag.  We already checked the iflag against the file mode when
596 	 * we scrubbed the dinode.
597 	 */
598 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
599 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
600 
601 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
602 out:
603 	return error;
604 }
605