xref: /openbmc/linux/fs/xfs/scrub/inode.c (revision 37744fee)
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 /* Scrub all the ondisk inode fields. */
199 STATIC void
200 xchk_dinode(
201 	struct xfs_scrub	*sc,
202 	struct xfs_dinode	*dip,
203 	xfs_ino_t		ino)
204 {
205 	struct xfs_mount	*mp = sc->mp;
206 	size_t			fork_recs;
207 	unsigned long long	isize;
208 	uint64_t		flags2;
209 	uint32_t		nextents;
210 	uint16_t		flags;
211 	uint16_t		mode;
212 
213 	flags = be16_to_cpu(dip->di_flags);
214 	if (dip->di_version >= 3)
215 		flags2 = be64_to_cpu(dip->di_flags2);
216 	else
217 		flags2 = 0;
218 
219 	/* di_mode */
220 	mode = be16_to_cpu(dip->di_mode);
221 	switch (mode & S_IFMT) {
222 	case S_IFLNK:
223 	case S_IFREG:
224 	case S_IFDIR:
225 	case S_IFCHR:
226 	case S_IFBLK:
227 	case S_IFIFO:
228 	case S_IFSOCK:
229 		/* mode is recognized */
230 		break;
231 	default:
232 		xchk_ino_set_corrupt(sc, ino);
233 		break;
234 	}
235 
236 	/* v1/v2 fields */
237 	switch (dip->di_version) {
238 	case 1:
239 		/*
240 		 * We autoconvert v1 inodes into v2 inodes on writeout,
241 		 * so just mark this inode for preening.
242 		 */
243 		xchk_ino_set_preen(sc, ino);
244 		break;
245 	case 2:
246 	case 3:
247 		if (dip->di_onlink != 0)
248 			xchk_ino_set_corrupt(sc, ino);
249 
250 		if (dip->di_mode == 0 && sc->ip)
251 			xchk_ino_set_corrupt(sc, ino);
252 
253 		if (dip->di_projid_hi != 0 &&
254 		    !xfs_sb_version_hasprojid32bit(&mp->m_sb))
255 			xchk_ino_set_corrupt(sc, ino);
256 		break;
257 	default:
258 		xchk_ino_set_corrupt(sc, ino);
259 		return;
260 	}
261 
262 	/*
263 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
264 	 * userspace could have created that.
265 	 */
266 	if (dip->di_uid == cpu_to_be32(-1U) ||
267 	    dip->di_gid == cpu_to_be32(-1U))
268 		xchk_ino_set_warning(sc, ino);
269 
270 	/* di_format */
271 	switch (dip->di_format) {
272 	case XFS_DINODE_FMT_DEV:
273 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
274 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
275 			xchk_ino_set_corrupt(sc, ino);
276 		break;
277 	case XFS_DINODE_FMT_LOCAL:
278 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
279 			xchk_ino_set_corrupt(sc, ino);
280 		break;
281 	case XFS_DINODE_FMT_EXTENTS:
282 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
283 			xchk_ino_set_corrupt(sc, ino);
284 		break;
285 	case XFS_DINODE_FMT_BTREE:
286 		if (!S_ISREG(mode) && !S_ISDIR(mode))
287 			xchk_ino_set_corrupt(sc, ino);
288 		break;
289 	case XFS_DINODE_FMT_UUID:
290 	default:
291 		xchk_ino_set_corrupt(sc, ino);
292 		break;
293 	}
294 
295 	/* di_[amc]time.nsec */
296 	if (be32_to_cpu(dip->di_atime.t_nsec) >= NSEC_PER_SEC)
297 		xchk_ino_set_corrupt(sc, ino);
298 	if (be32_to_cpu(dip->di_mtime.t_nsec) >= NSEC_PER_SEC)
299 		xchk_ino_set_corrupt(sc, ino);
300 	if (be32_to_cpu(dip->di_ctime.t_nsec) >= NSEC_PER_SEC)
301 		xchk_ino_set_corrupt(sc, ino);
302 
303 	/*
304 	 * di_size.  xfs_dinode_verify checks for things that screw up
305 	 * the VFS such as the upper bit being set and zero-length
306 	 * symlinks/directories, but we can do more here.
307 	 */
308 	isize = be64_to_cpu(dip->di_size);
309 	if (isize & (1ULL << 63))
310 		xchk_ino_set_corrupt(sc, ino);
311 
312 	/* Devices, fifos, and sockets must have zero size */
313 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
314 		xchk_ino_set_corrupt(sc, ino);
315 
316 	/* Directories can't be larger than the data section size (32G) */
317 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
318 		xchk_ino_set_corrupt(sc, ino);
319 
320 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
321 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
322 		xchk_ino_set_corrupt(sc, ino);
323 
324 	/*
325 	 * Warn if the running kernel can't handle the kinds of offsets
326 	 * needed to deal with the file size.  In other words, if the
327 	 * pagecache can't cache all the blocks in this file due to
328 	 * overly large offsets, flag the inode for admin review.
329 	 */
330 	if (isize >= mp->m_super->s_maxbytes)
331 		xchk_ino_set_warning(sc, ino);
332 
333 	/* di_nblocks */
334 	if (flags2 & XFS_DIFLAG2_REFLINK) {
335 		; /* nblocks can exceed dblocks */
336 	} else if (flags & XFS_DIFLAG_REALTIME) {
337 		/*
338 		 * nblocks is the sum of data extents (in the rtdev),
339 		 * attr extents (in the datadev), and both forks' bmbt
340 		 * blocks (in the datadev).  This clumsy check is the
341 		 * best we can do without cross-referencing with the
342 		 * inode forks.
343 		 */
344 		if (be64_to_cpu(dip->di_nblocks) >=
345 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
346 			xchk_ino_set_corrupt(sc, ino);
347 	} else {
348 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
349 			xchk_ino_set_corrupt(sc, ino);
350 	}
351 
352 	xchk_inode_flags(sc, dip, ino, mode, flags);
353 
354 	xchk_inode_extsize(sc, dip, ino, mode, flags);
355 
356 	/* di_nextents */
357 	nextents = be32_to_cpu(dip->di_nextents);
358 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
359 	switch (dip->di_format) {
360 	case XFS_DINODE_FMT_EXTENTS:
361 		if (nextents > fork_recs)
362 			xchk_ino_set_corrupt(sc, ino);
363 		break;
364 	case XFS_DINODE_FMT_BTREE:
365 		if (nextents <= fork_recs)
366 			xchk_ino_set_corrupt(sc, ino);
367 		break;
368 	default:
369 		if (nextents != 0)
370 			xchk_ino_set_corrupt(sc, ino);
371 		break;
372 	}
373 
374 	/* di_forkoff */
375 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
376 		xchk_ino_set_corrupt(sc, ino);
377 	if (dip->di_anextents != 0 && dip->di_forkoff == 0)
378 		xchk_ino_set_corrupt(sc, ino);
379 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
380 		xchk_ino_set_corrupt(sc, ino);
381 
382 	/* di_aformat */
383 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
384 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
385 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
386 		xchk_ino_set_corrupt(sc, ino);
387 
388 	/* di_anextents */
389 	nextents = be16_to_cpu(dip->di_anextents);
390 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
391 	switch (dip->di_aformat) {
392 	case XFS_DINODE_FMT_EXTENTS:
393 		if (nextents > fork_recs)
394 			xchk_ino_set_corrupt(sc, ino);
395 		break;
396 	case XFS_DINODE_FMT_BTREE:
397 		if (nextents <= fork_recs)
398 			xchk_ino_set_corrupt(sc, ino);
399 		break;
400 	default:
401 		if (nextents != 0)
402 			xchk_ino_set_corrupt(sc, ino);
403 	}
404 
405 	if (dip->di_version >= 3) {
406 		if (be32_to_cpu(dip->di_crtime.t_nsec) >= NSEC_PER_SEC)
407 			xchk_ino_set_corrupt(sc, ino);
408 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
409 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
410 				flags2);
411 	}
412 }
413 
414 /*
415  * Make sure the finobt doesn't think this inode is free.
416  * We don't have to check the inobt ourselves because we got the inode via
417  * IGET_UNTRUSTED, which checks the inobt for us.
418  */
419 static void
420 xchk_inode_xref_finobt(
421 	struct xfs_scrub		*sc,
422 	xfs_ino_t			ino)
423 {
424 	struct xfs_inobt_rec_incore	rec;
425 	xfs_agino_t			agino;
426 	int				has_record;
427 	int				error;
428 
429 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
430 		return;
431 
432 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
433 
434 	/*
435 	 * Try to get the finobt record.  If we can't get it, then we're
436 	 * in good shape.
437 	 */
438 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
439 			&has_record);
440 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
441 	    !has_record)
442 		return;
443 
444 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
445 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
446 	    !has_record)
447 		return;
448 
449 	/*
450 	 * Otherwise, make sure this record either doesn't cover this inode,
451 	 * or that it does but it's marked present.
452 	 */
453 	if (rec.ir_startino > agino ||
454 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
455 		return;
456 
457 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
458 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
459 }
460 
461 /* Cross reference the inode fields with the forks. */
462 STATIC void
463 xchk_inode_xref_bmap(
464 	struct xfs_scrub	*sc,
465 	struct xfs_dinode	*dip)
466 {
467 	xfs_extnum_t		nextents;
468 	xfs_filblks_t		count;
469 	xfs_filblks_t		acount;
470 	int			error;
471 
472 	if (xchk_skip_xref(sc->sm))
473 		return;
474 
475 	/* Walk all the extents to check nextents/naextents/nblocks. */
476 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
477 			&nextents, &count);
478 	if (!xchk_should_check_xref(sc, &error, NULL))
479 		return;
480 	if (nextents < be32_to_cpu(dip->di_nextents))
481 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
482 
483 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
484 			&nextents, &acount);
485 	if (!xchk_should_check_xref(sc, &error, NULL))
486 		return;
487 	if (nextents != be16_to_cpu(dip->di_anextents))
488 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
489 
490 	/* Check nblocks against the inode. */
491 	if (count + acount != be64_to_cpu(dip->di_nblocks))
492 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
493 }
494 
495 /* Cross-reference with the other btrees. */
496 STATIC void
497 xchk_inode_xref(
498 	struct xfs_scrub	*sc,
499 	xfs_ino_t		ino,
500 	struct xfs_dinode	*dip)
501 {
502 	xfs_agnumber_t		agno;
503 	xfs_agblock_t		agbno;
504 	int			error;
505 
506 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
507 		return;
508 
509 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
510 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
511 
512 	error = xchk_ag_init(sc, agno, &sc->sa);
513 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
514 		return;
515 
516 	xchk_xref_is_used_space(sc, agbno, 1);
517 	xchk_inode_xref_finobt(sc, ino);
518 	xchk_xref_is_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
519 	xchk_xref_is_not_shared(sc, agbno, 1);
520 	xchk_inode_xref_bmap(sc, dip);
521 
522 	xchk_ag_free(sc, &sc->sa);
523 }
524 
525 /*
526  * If the reflink iflag disagrees with a scan for shared data fork extents,
527  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
528  * any shared extents).  We already checked for reflink iflag set on a non
529  * reflink filesystem.
530  */
531 static void
532 xchk_inode_check_reflink_iflag(
533 	struct xfs_scrub	*sc,
534 	xfs_ino_t		ino)
535 {
536 	struct xfs_mount	*mp = sc->mp;
537 	bool			has_shared;
538 	int			error;
539 
540 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
541 		return;
542 
543 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
544 			&has_shared);
545 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
546 			XFS_INO_TO_AGBNO(mp, ino), &error))
547 		return;
548 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
549 		xchk_ino_set_preen(sc, ino);
550 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
551 		xchk_ino_set_corrupt(sc, ino);
552 }
553 
554 /* Scrub an inode. */
555 int
556 xchk_inode(
557 	struct xfs_scrub	*sc)
558 {
559 	struct xfs_dinode	di;
560 	int			error = 0;
561 
562 	/*
563 	 * If sc->ip is NULL, that means that the setup function called
564 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
565 	 * and a NULL inode, so flag the corruption error and return.
566 	 */
567 	if (!sc->ip) {
568 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
569 		return 0;
570 	}
571 
572 	/* Scrub the inode core. */
573 	xfs_inode_to_disk(sc->ip, &di, 0);
574 	xchk_dinode(sc, &di, sc->ip->i_ino);
575 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
576 		goto out;
577 
578 	/*
579 	 * Look for discrepancies between file's data blocks and the reflink
580 	 * iflag.  We already checked the iflag against the file mode when
581 	 * we scrubbed the dinode.
582 	 */
583 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
584 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
585 
586 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
587 out:
588 	return error;
589 }
590