xref: /openbmc/linux/fs/xfs/scrub/inode.c (revision 5a0bb066)
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 	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 	const xfs_timestamp_t	ts)
203 {
204 	struct timespec64	tv;
205 
206 	tv = xfs_inode_from_disk_ts(ts);
207 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
208 		xchk_ino_set_corrupt(sc, ino);
209 }
210 
211 /* Scrub all the ondisk inode fields. */
212 STATIC void
213 xchk_dinode(
214 	struct xfs_scrub	*sc,
215 	struct xfs_dinode	*dip,
216 	xfs_ino_t		ino)
217 {
218 	struct xfs_mount	*mp = sc->mp;
219 	size_t			fork_recs;
220 	unsigned long long	isize;
221 	uint64_t		flags2;
222 	uint32_t		nextents;
223 	uint16_t		flags;
224 	uint16_t		mode;
225 
226 	flags = be16_to_cpu(dip->di_flags);
227 	if (dip->di_version >= 3)
228 		flags2 = be64_to_cpu(dip->di_flags2);
229 	else
230 		flags2 = 0;
231 
232 	/* di_mode */
233 	mode = be16_to_cpu(dip->di_mode);
234 	switch (mode & S_IFMT) {
235 	case S_IFLNK:
236 	case S_IFREG:
237 	case S_IFDIR:
238 	case S_IFCHR:
239 	case S_IFBLK:
240 	case S_IFIFO:
241 	case S_IFSOCK:
242 		/* mode is recognized */
243 		break;
244 	default:
245 		xchk_ino_set_corrupt(sc, ino);
246 		break;
247 	}
248 
249 	/* v1/v2 fields */
250 	switch (dip->di_version) {
251 	case 1:
252 		/*
253 		 * We autoconvert v1 inodes into v2 inodes on writeout,
254 		 * so just mark this inode for preening.
255 		 */
256 		xchk_ino_set_preen(sc, ino);
257 		break;
258 	case 2:
259 	case 3:
260 		if (dip->di_onlink != 0)
261 			xchk_ino_set_corrupt(sc, ino);
262 
263 		if (dip->di_mode == 0 && sc->ip)
264 			xchk_ino_set_corrupt(sc, ino);
265 
266 		if (dip->di_projid_hi != 0 &&
267 		    !xfs_sb_version_hasprojid32bit(&mp->m_sb))
268 			xchk_ino_set_corrupt(sc, ino);
269 		break;
270 	default:
271 		xchk_ino_set_corrupt(sc, ino);
272 		return;
273 	}
274 
275 	/*
276 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
277 	 * userspace could have created that.
278 	 */
279 	if (dip->di_uid == cpu_to_be32(-1U) ||
280 	    dip->di_gid == cpu_to_be32(-1U))
281 		xchk_ino_set_warning(sc, ino);
282 
283 	/* di_format */
284 	switch (dip->di_format) {
285 	case XFS_DINODE_FMT_DEV:
286 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
287 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
288 			xchk_ino_set_corrupt(sc, ino);
289 		break;
290 	case XFS_DINODE_FMT_LOCAL:
291 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
292 			xchk_ino_set_corrupt(sc, ino);
293 		break;
294 	case XFS_DINODE_FMT_EXTENTS:
295 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
296 			xchk_ino_set_corrupt(sc, ino);
297 		break;
298 	case XFS_DINODE_FMT_BTREE:
299 		if (!S_ISREG(mode) && !S_ISDIR(mode))
300 			xchk_ino_set_corrupt(sc, ino);
301 		break;
302 	case XFS_DINODE_FMT_UUID:
303 	default:
304 		xchk_ino_set_corrupt(sc, ino);
305 		break;
306 	}
307 
308 	/* di_[amc]time.nsec */
309 	xchk_dinode_nsec(sc, ino, dip->di_atime);
310 	xchk_dinode_nsec(sc, ino, dip->di_mtime);
311 	xchk_dinode_nsec(sc, ino, dip->di_ctime);
312 
313 	/*
314 	 * di_size.  xfs_dinode_verify checks for things that screw up
315 	 * the VFS such as the upper bit being set and zero-length
316 	 * symlinks/directories, but we can do more here.
317 	 */
318 	isize = be64_to_cpu(dip->di_size);
319 	if (isize & (1ULL << 63))
320 		xchk_ino_set_corrupt(sc, ino);
321 
322 	/* Devices, fifos, and sockets must have zero size */
323 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
324 		xchk_ino_set_corrupt(sc, ino);
325 
326 	/* Directories can't be larger than the data section size (32G) */
327 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
328 		xchk_ino_set_corrupt(sc, ino);
329 
330 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
331 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
332 		xchk_ino_set_corrupt(sc, ino);
333 
334 	/*
335 	 * Warn if the running kernel can't handle the kinds of offsets
336 	 * needed to deal with the file size.  In other words, if the
337 	 * pagecache can't cache all the blocks in this file due to
338 	 * overly large offsets, flag the inode for admin review.
339 	 */
340 	if (isize >= mp->m_super->s_maxbytes)
341 		xchk_ino_set_warning(sc, ino);
342 
343 	/* di_nblocks */
344 	if (flags2 & XFS_DIFLAG2_REFLINK) {
345 		; /* nblocks can exceed dblocks */
346 	} else if (flags & XFS_DIFLAG_REALTIME) {
347 		/*
348 		 * nblocks is the sum of data extents (in the rtdev),
349 		 * attr extents (in the datadev), and both forks' bmbt
350 		 * blocks (in the datadev).  This clumsy check is the
351 		 * best we can do without cross-referencing with the
352 		 * inode forks.
353 		 */
354 		if (be64_to_cpu(dip->di_nblocks) >=
355 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
356 			xchk_ino_set_corrupt(sc, ino);
357 	} else {
358 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
359 			xchk_ino_set_corrupt(sc, ino);
360 	}
361 
362 	xchk_inode_flags(sc, dip, ino, mode, flags);
363 
364 	xchk_inode_extsize(sc, dip, ino, mode, flags);
365 
366 	/* di_nextents */
367 	nextents = be32_to_cpu(dip->di_nextents);
368 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
369 	switch (dip->di_format) {
370 	case XFS_DINODE_FMT_EXTENTS:
371 		if (nextents > fork_recs)
372 			xchk_ino_set_corrupt(sc, ino);
373 		break;
374 	case XFS_DINODE_FMT_BTREE:
375 		if (nextents <= fork_recs)
376 			xchk_ino_set_corrupt(sc, ino);
377 		break;
378 	default:
379 		if (nextents != 0)
380 			xchk_ino_set_corrupt(sc, ino);
381 		break;
382 	}
383 
384 	/* di_forkoff */
385 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
386 		xchk_ino_set_corrupt(sc, ino);
387 	if (dip->di_anextents != 0 && dip->di_forkoff == 0)
388 		xchk_ino_set_corrupt(sc, ino);
389 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
390 		xchk_ino_set_corrupt(sc, ino);
391 
392 	/* di_aformat */
393 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
394 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
395 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
396 		xchk_ino_set_corrupt(sc, ino);
397 
398 	/* di_anextents */
399 	nextents = be16_to_cpu(dip->di_anextents);
400 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
401 	switch (dip->di_aformat) {
402 	case XFS_DINODE_FMT_EXTENTS:
403 		if (nextents > fork_recs)
404 			xchk_ino_set_corrupt(sc, ino);
405 		break;
406 	case XFS_DINODE_FMT_BTREE:
407 		if (nextents <= fork_recs)
408 			xchk_ino_set_corrupt(sc, ino);
409 		break;
410 	default:
411 		if (nextents != 0)
412 			xchk_ino_set_corrupt(sc, ino);
413 	}
414 
415 	if (dip->di_version >= 3) {
416 		xchk_dinode_nsec(sc, ino, dip->di_crtime);
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
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
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
506 xchk_inode_xref(
507 	struct xfs_scrub	*sc,
508 	xfs_ino_t		ino,
509 	struct xfs_dinode	*dip)
510 {
511 	xfs_agnumber_t		agno;
512 	xfs_agblock_t		agbno;
513 	int			error;
514 
515 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
516 		return;
517 
518 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
519 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
520 
521 	error = xchk_ag_init(sc, agno, &sc->sa);
522 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
523 		return;
524 
525 	xchk_xref_is_used_space(sc, agbno, 1);
526 	xchk_inode_xref_finobt(sc, ino);
527 	xchk_xref_is_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
528 	xchk_xref_is_not_shared(sc, agbno, 1);
529 	xchk_inode_xref_bmap(sc, dip);
530 
531 	xchk_ag_free(sc, &sc->sa);
532 }
533 
534 /*
535  * If the reflink iflag disagrees with a scan for shared data fork extents,
536  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
537  * any shared extents).  We already checked for reflink iflag set on a non
538  * reflink filesystem.
539  */
540 static void
541 xchk_inode_check_reflink_iflag(
542 	struct xfs_scrub	*sc,
543 	xfs_ino_t		ino)
544 {
545 	struct xfs_mount	*mp = sc->mp;
546 	bool			has_shared;
547 	int			error;
548 
549 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
550 		return;
551 
552 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
553 			&has_shared);
554 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
555 			XFS_INO_TO_AGBNO(mp, ino), &error))
556 		return;
557 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
558 		xchk_ino_set_preen(sc, ino);
559 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
560 		xchk_ino_set_corrupt(sc, ino);
561 }
562 
563 /* Scrub an inode. */
564 int
565 xchk_inode(
566 	struct xfs_scrub	*sc)
567 {
568 	struct xfs_dinode	di;
569 	int			error = 0;
570 
571 	/*
572 	 * If sc->ip is NULL, that means that the setup function called
573 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
574 	 * and a NULL inode, so flag the corruption error and return.
575 	 */
576 	if (!sc->ip) {
577 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
578 		return 0;
579 	}
580 
581 	/* Scrub the inode core. */
582 	xfs_inode_to_disk(sc->ip, &di, 0);
583 	xchk_dinode(sc, &di, sc->ip->i_ino);
584 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
585 		goto out;
586 
587 	/*
588 	 * Look for discrepancies between file's data blocks and the reflink
589 	 * iflag.  We already checked the iflag against the file mode when
590 	 * we scrubbed the dinode.
591 	 */
592 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
593 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
594 
595 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
596 out:
597 	return error;
598 }
599