xref: /openbmc/linux/fs/xfs/scrub/inode.c (revision 43ffe817)
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 	/* dax and reflink make no sense, currently */
189 	if ((flags2 & XFS_DIFLAG2_DAX) && (flags2 & XFS_DIFLAG2_REFLINK))
190 		goto bad;
191 
192 	/* no bigtime iflag without the bigtime feature */
193 	if (xfs_dinode_has_bigtime(dip) &&
194 	    !xfs_sb_version_hasbigtime(&mp->m_sb))
195 		goto bad;
196 
197 	return;
198 bad:
199 	xchk_ino_set_corrupt(sc, ino);
200 }
201 
202 static inline void
203 xchk_dinode_nsec(
204 	struct xfs_scrub	*sc,
205 	xfs_ino_t		ino,
206 	struct xfs_dinode	*dip,
207 	const xfs_timestamp_t	ts)
208 {
209 	struct timespec64	tv;
210 
211 	tv = xfs_inode_from_disk_ts(dip, ts);
212 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
213 		xchk_ino_set_corrupt(sc, ino);
214 }
215 
216 /* Scrub all the ondisk inode fields. */
217 STATIC void
218 xchk_dinode(
219 	struct xfs_scrub	*sc,
220 	struct xfs_dinode	*dip,
221 	xfs_ino_t		ino)
222 {
223 	struct xfs_mount	*mp = sc->mp;
224 	size_t			fork_recs;
225 	unsigned long long	isize;
226 	uint64_t		flags2;
227 	uint32_t		nextents;
228 	uint16_t		flags;
229 	uint16_t		mode;
230 
231 	flags = be16_to_cpu(dip->di_flags);
232 	if (dip->di_version >= 3)
233 		flags2 = be64_to_cpu(dip->di_flags2);
234 	else
235 		flags2 = 0;
236 
237 	/* di_mode */
238 	mode = be16_to_cpu(dip->di_mode);
239 	switch (mode & S_IFMT) {
240 	case S_IFLNK:
241 	case S_IFREG:
242 	case S_IFDIR:
243 	case S_IFCHR:
244 	case S_IFBLK:
245 	case S_IFIFO:
246 	case S_IFSOCK:
247 		/* mode is recognized */
248 		break;
249 	default:
250 		xchk_ino_set_corrupt(sc, ino);
251 		break;
252 	}
253 
254 	/* v1/v2 fields */
255 	switch (dip->di_version) {
256 	case 1:
257 		/*
258 		 * We autoconvert v1 inodes into v2 inodes on writeout,
259 		 * so just mark this inode for preening.
260 		 */
261 		xchk_ino_set_preen(sc, ino);
262 		break;
263 	case 2:
264 	case 3:
265 		if (dip->di_onlink != 0)
266 			xchk_ino_set_corrupt(sc, ino);
267 
268 		if (dip->di_mode == 0 && sc->ip)
269 			xchk_ino_set_corrupt(sc, ino);
270 
271 		if (dip->di_projid_hi != 0 &&
272 		    !xfs_sb_version_hasprojid32bit(&mp->m_sb))
273 			xchk_ino_set_corrupt(sc, ino);
274 		break;
275 	default:
276 		xchk_ino_set_corrupt(sc, ino);
277 		return;
278 	}
279 
280 	/*
281 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
282 	 * userspace could have created that.
283 	 */
284 	if (dip->di_uid == cpu_to_be32(-1U) ||
285 	    dip->di_gid == cpu_to_be32(-1U))
286 		xchk_ino_set_warning(sc, ino);
287 
288 	/* di_format */
289 	switch (dip->di_format) {
290 	case XFS_DINODE_FMT_DEV:
291 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
292 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
293 			xchk_ino_set_corrupt(sc, ino);
294 		break;
295 	case XFS_DINODE_FMT_LOCAL:
296 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
297 			xchk_ino_set_corrupt(sc, ino);
298 		break;
299 	case XFS_DINODE_FMT_EXTENTS:
300 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
301 			xchk_ino_set_corrupt(sc, ino);
302 		break;
303 	case XFS_DINODE_FMT_BTREE:
304 		if (!S_ISREG(mode) && !S_ISDIR(mode))
305 			xchk_ino_set_corrupt(sc, ino);
306 		break;
307 	case XFS_DINODE_FMT_UUID:
308 	default:
309 		xchk_ino_set_corrupt(sc, ino);
310 		break;
311 	}
312 
313 	/* di_[amc]time.nsec */
314 	xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
315 	xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
316 	xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
317 
318 	/*
319 	 * di_size.  xfs_dinode_verify checks for things that screw up
320 	 * the VFS such as the upper bit being set and zero-length
321 	 * symlinks/directories, but we can do more here.
322 	 */
323 	isize = be64_to_cpu(dip->di_size);
324 	if (isize & (1ULL << 63))
325 		xchk_ino_set_corrupt(sc, ino);
326 
327 	/* Devices, fifos, and sockets must have zero size */
328 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
329 		xchk_ino_set_corrupt(sc, ino);
330 
331 	/* Directories can't be larger than the data section size (32G) */
332 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
333 		xchk_ino_set_corrupt(sc, ino);
334 
335 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
336 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
337 		xchk_ino_set_corrupt(sc, ino);
338 
339 	/*
340 	 * Warn if the running kernel can't handle the kinds of offsets
341 	 * needed to deal with the file size.  In other words, if the
342 	 * pagecache can't cache all the blocks in this file due to
343 	 * overly large offsets, flag the inode for admin review.
344 	 */
345 	if (isize >= mp->m_super->s_maxbytes)
346 		xchk_ino_set_warning(sc, ino);
347 
348 	/* di_nblocks */
349 	if (flags2 & XFS_DIFLAG2_REFLINK) {
350 		; /* nblocks can exceed dblocks */
351 	} else if (flags & XFS_DIFLAG_REALTIME) {
352 		/*
353 		 * nblocks is the sum of data extents (in the rtdev),
354 		 * attr extents (in the datadev), and both forks' bmbt
355 		 * blocks (in the datadev).  This clumsy check is the
356 		 * best we can do without cross-referencing with the
357 		 * inode forks.
358 		 */
359 		if (be64_to_cpu(dip->di_nblocks) >=
360 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
361 			xchk_ino_set_corrupt(sc, ino);
362 	} else {
363 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
364 			xchk_ino_set_corrupt(sc, ino);
365 	}
366 
367 	xchk_inode_flags(sc, dip, ino, mode, flags);
368 
369 	xchk_inode_extsize(sc, dip, ino, mode, flags);
370 
371 	/* di_nextents */
372 	nextents = be32_to_cpu(dip->di_nextents);
373 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
374 	switch (dip->di_format) {
375 	case XFS_DINODE_FMT_EXTENTS:
376 		if (nextents > fork_recs)
377 			xchk_ino_set_corrupt(sc, ino);
378 		break;
379 	case XFS_DINODE_FMT_BTREE:
380 		if (nextents <= fork_recs)
381 			xchk_ino_set_corrupt(sc, ino);
382 		break;
383 	default:
384 		if (nextents != 0)
385 			xchk_ino_set_corrupt(sc, ino);
386 		break;
387 	}
388 
389 	/* di_forkoff */
390 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
391 		xchk_ino_set_corrupt(sc, ino);
392 	if (dip->di_anextents != 0 && dip->di_forkoff == 0)
393 		xchk_ino_set_corrupt(sc, ino);
394 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
395 		xchk_ino_set_corrupt(sc, ino);
396 
397 	/* di_aformat */
398 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
399 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
400 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
401 		xchk_ino_set_corrupt(sc, ino);
402 
403 	/* di_anextents */
404 	nextents = be16_to_cpu(dip->di_anextents);
405 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
406 	switch (dip->di_aformat) {
407 	case XFS_DINODE_FMT_EXTENTS:
408 		if (nextents > fork_recs)
409 			xchk_ino_set_corrupt(sc, ino);
410 		break;
411 	case XFS_DINODE_FMT_BTREE:
412 		if (nextents <= fork_recs)
413 			xchk_ino_set_corrupt(sc, ino);
414 		break;
415 	default:
416 		if (nextents != 0)
417 			xchk_ino_set_corrupt(sc, ino);
418 	}
419 
420 	if (dip->di_version >= 3) {
421 		xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
422 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
423 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
424 				flags2);
425 	}
426 }
427 
428 /*
429  * Make sure the finobt doesn't think this inode is free.
430  * We don't have to check the inobt ourselves because we got the inode via
431  * IGET_UNTRUSTED, which checks the inobt for us.
432  */
433 static void
434 xchk_inode_xref_finobt(
435 	struct xfs_scrub		*sc,
436 	xfs_ino_t			ino)
437 {
438 	struct xfs_inobt_rec_incore	rec;
439 	xfs_agino_t			agino;
440 	int				has_record;
441 	int				error;
442 
443 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
444 		return;
445 
446 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
447 
448 	/*
449 	 * Try to get the finobt record.  If we can't get it, then we're
450 	 * in good shape.
451 	 */
452 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
453 			&has_record);
454 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
455 	    !has_record)
456 		return;
457 
458 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
459 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
460 	    !has_record)
461 		return;
462 
463 	/*
464 	 * Otherwise, make sure this record either doesn't cover this inode,
465 	 * or that it does but it's marked present.
466 	 */
467 	if (rec.ir_startino > agino ||
468 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
469 		return;
470 
471 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
472 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
473 }
474 
475 /* Cross reference the inode fields with the forks. */
476 STATIC void
477 xchk_inode_xref_bmap(
478 	struct xfs_scrub	*sc,
479 	struct xfs_dinode	*dip)
480 {
481 	xfs_extnum_t		nextents;
482 	xfs_filblks_t		count;
483 	xfs_filblks_t		acount;
484 	int			error;
485 
486 	if (xchk_skip_xref(sc->sm))
487 		return;
488 
489 	/* Walk all the extents to check nextents/naextents/nblocks. */
490 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
491 			&nextents, &count);
492 	if (!xchk_should_check_xref(sc, &error, NULL))
493 		return;
494 	if (nextents < be32_to_cpu(dip->di_nextents))
495 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
496 
497 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
498 			&nextents, &acount);
499 	if (!xchk_should_check_xref(sc, &error, NULL))
500 		return;
501 	if (nextents != be16_to_cpu(dip->di_anextents))
502 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
503 
504 	/* Check nblocks against the inode. */
505 	if (count + acount != be64_to_cpu(dip->di_nblocks))
506 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
507 }
508 
509 /* Cross-reference with the other btrees. */
510 STATIC void
511 xchk_inode_xref(
512 	struct xfs_scrub	*sc,
513 	xfs_ino_t		ino,
514 	struct xfs_dinode	*dip)
515 {
516 	xfs_agnumber_t		agno;
517 	xfs_agblock_t		agbno;
518 	int			error;
519 
520 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
521 		return;
522 
523 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
524 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
525 
526 	error = xchk_ag_init(sc, agno, &sc->sa);
527 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
528 		return;
529 
530 	xchk_xref_is_used_space(sc, agbno, 1);
531 	xchk_inode_xref_finobt(sc, ino);
532 	xchk_xref_is_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
533 	xchk_xref_is_not_shared(sc, agbno, 1);
534 	xchk_inode_xref_bmap(sc, dip);
535 
536 	xchk_ag_free(sc, &sc->sa);
537 }
538 
539 /*
540  * If the reflink iflag disagrees with a scan for shared data fork extents,
541  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
542  * any shared extents).  We already checked for reflink iflag set on a non
543  * reflink filesystem.
544  */
545 static void
546 xchk_inode_check_reflink_iflag(
547 	struct xfs_scrub	*sc,
548 	xfs_ino_t		ino)
549 {
550 	struct xfs_mount	*mp = sc->mp;
551 	bool			has_shared;
552 	int			error;
553 
554 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
555 		return;
556 
557 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
558 			&has_shared);
559 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
560 			XFS_INO_TO_AGBNO(mp, ino), &error))
561 		return;
562 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
563 		xchk_ino_set_preen(sc, ino);
564 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
565 		xchk_ino_set_corrupt(sc, ino);
566 }
567 
568 /* Scrub an inode. */
569 int
570 xchk_inode(
571 	struct xfs_scrub	*sc)
572 {
573 	struct xfs_dinode	di;
574 	int			error = 0;
575 
576 	/*
577 	 * If sc->ip is NULL, that means that the setup function called
578 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
579 	 * and a NULL inode, so flag the corruption error and return.
580 	 */
581 	if (!sc->ip) {
582 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
583 		return 0;
584 	}
585 
586 	/* Scrub the inode core. */
587 	xfs_inode_to_disk(sc->ip, &di, 0);
588 	xchk_dinode(sc, &di, sc->ip->i_ino);
589 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
590 		goto out;
591 
592 	/*
593 	 * Look for discrepancies between file's data blocks and the reflink
594 	 * iflag.  We already checked the iflag against the file mode when
595 	 * we scrubbed the dinode.
596 	 */
597 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
598 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
599 
600 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
601 out:
602 	return error;
603 }
604