xref: /openbmc/linux/fs/xfs/scrub/inode.c (revision b102a46c)
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 {
33 	int			error;
34 
35 	/*
36 	 * Try to get the inode.  If the verifiers fail, we try again
37 	 * in raw mode.
38 	 */
39 	error = xchk_get_inode(sc);
40 	switch (error) {
41 	case 0:
42 		break;
43 	case -EFSCORRUPTED:
44 	case -EFSBADCRC:
45 		return xchk_trans_alloc(sc, 0);
46 	default:
47 		return error;
48 	}
49 
50 	/* Got the inode, lock it and we're ready to go. */
51 	sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
52 	xfs_ilock(sc->ip, sc->ilock_flags);
53 	error = xchk_trans_alloc(sc, 0);
54 	if (error)
55 		goto out;
56 	sc->ilock_flags |= XFS_ILOCK_EXCL;
57 	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
58 
59 out:
60 	/* scrub teardown will unlock and release the inode for us */
61 	return error;
62 }
63 
64 /* Inode core */
65 
66 /* Validate di_extsize hint. */
67 STATIC void
68 xchk_inode_extsize(
69 	struct xfs_scrub	*sc,
70 	struct xfs_dinode	*dip,
71 	xfs_ino_t		ino,
72 	uint16_t		mode,
73 	uint16_t		flags)
74 {
75 	xfs_failaddr_t		fa;
76 	uint32_t		value = be32_to_cpu(dip->di_extsize);
77 
78 	fa = xfs_inode_validate_extsize(sc->mp, value, mode, flags);
79 	if (fa)
80 		xchk_ino_set_corrupt(sc, ino);
81 
82 	/*
83 	 * XFS allows a sysadmin to change the rt extent size when adding a rt
84 	 * section to a filesystem after formatting.  If there are any
85 	 * directories with extszinherit and rtinherit set, the hint could
86 	 * become misaligned with the new rextsize.  The verifier doesn't check
87 	 * this, because we allow rtinherit directories even without an rt
88 	 * device.  Flag this as an administrative warning since we will clean
89 	 * this up eventually.
90 	 */
91 	if ((flags & XFS_DIFLAG_RTINHERIT) &&
92 	    (flags & XFS_DIFLAG_EXTSZINHERIT) &&
93 	    value % sc->mp->m_sb.sb_rextsize > 0)
94 		xchk_ino_set_warning(sc, ino);
95 }
96 
97 /*
98  * Validate di_cowextsize hint.
99  *
100  * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
101  * These functions must be kept in sync with each other.
102  */
103 STATIC void
104 xchk_inode_cowextsize(
105 	struct xfs_scrub	*sc,
106 	struct xfs_dinode	*dip,
107 	xfs_ino_t		ino,
108 	uint16_t		mode,
109 	uint16_t		flags,
110 	uint64_t		flags2)
111 {
112 	xfs_failaddr_t		fa;
113 
114 	fa = xfs_inode_validate_cowextsize(sc->mp,
115 			be32_to_cpu(dip->di_cowextsize), mode, flags,
116 			flags2);
117 	if (fa)
118 		xchk_ino_set_corrupt(sc, ino);
119 }
120 
121 /* Make sure the di_flags make sense for the inode. */
122 STATIC void
123 xchk_inode_flags(
124 	struct xfs_scrub	*sc,
125 	struct xfs_dinode	*dip,
126 	xfs_ino_t		ino,
127 	uint16_t		mode,
128 	uint16_t		flags)
129 {
130 	struct xfs_mount	*mp = sc->mp;
131 
132 	/* di_flags are all taken, last bit cannot be used */
133 	if (flags & ~XFS_DIFLAG_ANY)
134 		goto bad;
135 
136 	/* rt flags require rt device */
137 	if ((flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
138 		goto bad;
139 
140 	/* new rt bitmap flag only valid for rbmino */
141 	if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
142 		goto bad;
143 
144 	/* directory-only flags */
145 	if ((flags & (XFS_DIFLAG_RTINHERIT |
146 		     XFS_DIFLAG_EXTSZINHERIT |
147 		     XFS_DIFLAG_PROJINHERIT |
148 		     XFS_DIFLAG_NOSYMLINKS)) &&
149 	    !S_ISDIR(mode))
150 		goto bad;
151 
152 	/* file-only flags */
153 	if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
154 	    !S_ISREG(mode))
155 		goto bad;
156 
157 	/* filestreams and rt make no sense */
158 	if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
159 		goto bad;
160 
161 	return;
162 bad:
163 	xchk_ino_set_corrupt(sc, ino);
164 }
165 
166 /* Make sure the di_flags2 make sense for the inode. */
167 STATIC void
168 xchk_inode_flags2(
169 	struct xfs_scrub	*sc,
170 	struct xfs_dinode	*dip,
171 	xfs_ino_t		ino,
172 	uint16_t		mode,
173 	uint16_t		flags,
174 	uint64_t		flags2)
175 {
176 	struct xfs_mount	*mp = sc->mp;
177 
178 	/* Unknown di_flags2 could be from a future kernel */
179 	if (flags2 & ~XFS_DIFLAG2_ANY)
180 		xchk_ino_set_warning(sc, ino);
181 
182 	/* reflink flag requires reflink feature */
183 	if ((flags2 & XFS_DIFLAG2_REFLINK) &&
184 	    !xfs_sb_version_hasreflink(&mp->m_sb))
185 		goto bad;
186 
187 	/* cowextsize flag is checked w.r.t. mode separately */
188 
189 	/* file/dir-only flags */
190 	if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
191 		goto bad;
192 
193 	/* file-only flags */
194 	if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
195 		goto bad;
196 
197 	/* realtime and reflink make no sense, currently */
198 	if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK))
199 		goto bad;
200 
201 	/* no bigtime iflag without the bigtime feature */
202 	if (xfs_dinode_has_bigtime(dip) &&
203 	    !xfs_sb_version_hasbigtime(&mp->m_sb))
204 		goto bad;
205 
206 	return;
207 bad:
208 	xchk_ino_set_corrupt(sc, ino);
209 }
210 
211 static inline void
212 xchk_dinode_nsec(
213 	struct xfs_scrub	*sc,
214 	xfs_ino_t		ino,
215 	struct xfs_dinode	*dip,
216 	const xfs_timestamp_t	ts)
217 {
218 	struct timespec64	tv;
219 
220 	tv = xfs_inode_from_disk_ts(dip, ts);
221 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
222 		xchk_ino_set_corrupt(sc, ino);
223 }
224 
225 /* Scrub all the ondisk inode fields. */
226 STATIC void
227 xchk_dinode(
228 	struct xfs_scrub	*sc,
229 	struct xfs_dinode	*dip,
230 	xfs_ino_t		ino)
231 {
232 	struct xfs_mount	*mp = sc->mp;
233 	size_t			fork_recs;
234 	unsigned long long	isize;
235 	uint64_t		flags2;
236 	uint32_t		nextents;
237 	uint16_t		flags;
238 	uint16_t		mode;
239 
240 	flags = be16_to_cpu(dip->di_flags);
241 	if (dip->di_version >= 3)
242 		flags2 = be64_to_cpu(dip->di_flags2);
243 	else
244 		flags2 = 0;
245 
246 	/* di_mode */
247 	mode = be16_to_cpu(dip->di_mode);
248 	switch (mode & S_IFMT) {
249 	case S_IFLNK:
250 	case S_IFREG:
251 	case S_IFDIR:
252 	case S_IFCHR:
253 	case S_IFBLK:
254 	case S_IFIFO:
255 	case S_IFSOCK:
256 		/* mode is recognized */
257 		break;
258 	default:
259 		xchk_ino_set_corrupt(sc, ino);
260 		break;
261 	}
262 
263 	/* v1/v2 fields */
264 	switch (dip->di_version) {
265 	case 1:
266 		/*
267 		 * We autoconvert v1 inodes into v2 inodes on writeout,
268 		 * so just mark this inode for preening.
269 		 */
270 		xchk_ino_set_preen(sc, ino);
271 		break;
272 	case 2:
273 	case 3:
274 		if (dip->di_onlink != 0)
275 			xchk_ino_set_corrupt(sc, ino);
276 
277 		if (dip->di_mode == 0 && sc->ip)
278 			xchk_ino_set_corrupt(sc, ino);
279 
280 		if (dip->di_projid_hi != 0 &&
281 		    !xfs_sb_version_hasprojid32bit(&mp->m_sb))
282 			xchk_ino_set_corrupt(sc, ino);
283 		break;
284 	default:
285 		xchk_ino_set_corrupt(sc, ino);
286 		return;
287 	}
288 
289 	/*
290 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
291 	 * userspace could have created that.
292 	 */
293 	if (dip->di_uid == cpu_to_be32(-1U) ||
294 	    dip->di_gid == cpu_to_be32(-1U))
295 		xchk_ino_set_warning(sc, ino);
296 
297 	/* di_format */
298 	switch (dip->di_format) {
299 	case XFS_DINODE_FMT_DEV:
300 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
301 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
302 			xchk_ino_set_corrupt(sc, ino);
303 		break;
304 	case XFS_DINODE_FMT_LOCAL:
305 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
306 			xchk_ino_set_corrupt(sc, ino);
307 		break;
308 	case XFS_DINODE_FMT_EXTENTS:
309 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
310 			xchk_ino_set_corrupt(sc, ino);
311 		break;
312 	case XFS_DINODE_FMT_BTREE:
313 		if (!S_ISREG(mode) && !S_ISDIR(mode))
314 			xchk_ino_set_corrupt(sc, ino);
315 		break;
316 	case XFS_DINODE_FMT_UUID:
317 	default:
318 		xchk_ino_set_corrupt(sc, ino);
319 		break;
320 	}
321 
322 	/* di_[amc]time.nsec */
323 	xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
324 	xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
325 	xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
326 
327 	/*
328 	 * di_size.  xfs_dinode_verify checks for things that screw up
329 	 * the VFS such as the upper bit being set and zero-length
330 	 * symlinks/directories, but we can do more here.
331 	 */
332 	isize = be64_to_cpu(dip->di_size);
333 	if (isize & (1ULL << 63))
334 		xchk_ino_set_corrupt(sc, ino);
335 
336 	/* Devices, fifos, and sockets must have zero size */
337 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
338 		xchk_ino_set_corrupt(sc, ino);
339 
340 	/* Directories can't be larger than the data section size (32G) */
341 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
342 		xchk_ino_set_corrupt(sc, ino);
343 
344 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
345 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
346 		xchk_ino_set_corrupt(sc, ino);
347 
348 	/*
349 	 * Warn if the running kernel can't handle the kinds of offsets
350 	 * needed to deal with the file size.  In other words, if the
351 	 * pagecache can't cache all the blocks in this file due to
352 	 * overly large offsets, flag the inode for admin review.
353 	 */
354 	if (isize >= mp->m_super->s_maxbytes)
355 		xchk_ino_set_warning(sc, ino);
356 
357 	/* di_nblocks */
358 	if (flags2 & XFS_DIFLAG2_REFLINK) {
359 		; /* nblocks can exceed dblocks */
360 	} else if (flags & XFS_DIFLAG_REALTIME) {
361 		/*
362 		 * nblocks is the sum of data extents (in the rtdev),
363 		 * attr extents (in the datadev), and both forks' bmbt
364 		 * blocks (in the datadev).  This clumsy check is the
365 		 * best we can do without cross-referencing with the
366 		 * inode forks.
367 		 */
368 		if (be64_to_cpu(dip->di_nblocks) >=
369 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
370 			xchk_ino_set_corrupt(sc, ino);
371 	} else {
372 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
373 			xchk_ino_set_corrupt(sc, ino);
374 	}
375 
376 	xchk_inode_flags(sc, dip, ino, mode, flags);
377 
378 	xchk_inode_extsize(sc, dip, ino, mode, flags);
379 
380 	/* di_nextents */
381 	nextents = be32_to_cpu(dip->di_nextents);
382 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
383 	switch (dip->di_format) {
384 	case XFS_DINODE_FMT_EXTENTS:
385 		if (nextents > fork_recs)
386 			xchk_ino_set_corrupt(sc, ino);
387 		break;
388 	case XFS_DINODE_FMT_BTREE:
389 		if (nextents <= fork_recs)
390 			xchk_ino_set_corrupt(sc, ino);
391 		break;
392 	default:
393 		if (nextents != 0)
394 			xchk_ino_set_corrupt(sc, ino);
395 		break;
396 	}
397 
398 	/* di_forkoff */
399 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
400 		xchk_ino_set_corrupt(sc, ino);
401 	if (dip->di_anextents != 0 && dip->di_forkoff == 0)
402 		xchk_ino_set_corrupt(sc, ino);
403 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
404 		xchk_ino_set_corrupt(sc, ino);
405 
406 	/* di_aformat */
407 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
408 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
409 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
410 		xchk_ino_set_corrupt(sc, ino);
411 
412 	/* di_anextents */
413 	nextents = be16_to_cpu(dip->di_anextents);
414 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
415 	switch (dip->di_aformat) {
416 	case XFS_DINODE_FMT_EXTENTS:
417 		if (nextents > fork_recs)
418 			xchk_ino_set_corrupt(sc, ino);
419 		break;
420 	case XFS_DINODE_FMT_BTREE:
421 		if (nextents <= fork_recs)
422 			xchk_ino_set_corrupt(sc, ino);
423 		break;
424 	default:
425 		if (nextents != 0)
426 			xchk_ino_set_corrupt(sc, ino);
427 	}
428 
429 	if (dip->di_version >= 3) {
430 		xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
431 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
432 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
433 				flags2);
434 	}
435 }
436 
437 /*
438  * Make sure the finobt doesn't think this inode is free.
439  * We don't have to check the inobt ourselves because we got the inode via
440  * IGET_UNTRUSTED, which checks the inobt for us.
441  */
442 static void
443 xchk_inode_xref_finobt(
444 	struct xfs_scrub		*sc,
445 	xfs_ino_t			ino)
446 {
447 	struct xfs_inobt_rec_incore	rec;
448 	xfs_agino_t			agino;
449 	int				has_record;
450 	int				error;
451 
452 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
453 		return;
454 
455 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
456 
457 	/*
458 	 * Try to get the finobt record.  If we can't get it, then we're
459 	 * in good shape.
460 	 */
461 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
462 			&has_record);
463 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
464 	    !has_record)
465 		return;
466 
467 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
468 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
469 	    !has_record)
470 		return;
471 
472 	/*
473 	 * Otherwise, make sure this record either doesn't cover this inode,
474 	 * or that it does but it's marked present.
475 	 */
476 	if (rec.ir_startino > agino ||
477 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
478 		return;
479 
480 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
481 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
482 }
483 
484 /* Cross reference the inode fields with the forks. */
485 STATIC void
486 xchk_inode_xref_bmap(
487 	struct xfs_scrub	*sc,
488 	struct xfs_dinode	*dip)
489 {
490 	xfs_extnum_t		nextents;
491 	xfs_filblks_t		count;
492 	xfs_filblks_t		acount;
493 	int			error;
494 
495 	if (xchk_skip_xref(sc->sm))
496 		return;
497 
498 	/* Walk all the extents to check nextents/naextents/nblocks. */
499 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
500 			&nextents, &count);
501 	if (!xchk_should_check_xref(sc, &error, NULL))
502 		return;
503 	if (nextents < be32_to_cpu(dip->di_nextents))
504 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
505 
506 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
507 			&nextents, &acount);
508 	if (!xchk_should_check_xref(sc, &error, NULL))
509 		return;
510 	if (nextents != be16_to_cpu(dip->di_anextents))
511 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
512 
513 	/* Check nblocks against the inode. */
514 	if (count + acount != be64_to_cpu(dip->di_nblocks))
515 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
516 }
517 
518 /* Cross-reference with the other btrees. */
519 STATIC void
520 xchk_inode_xref(
521 	struct xfs_scrub	*sc,
522 	xfs_ino_t		ino,
523 	struct xfs_dinode	*dip)
524 {
525 	xfs_agnumber_t		agno;
526 	xfs_agblock_t		agbno;
527 	int			error;
528 
529 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
530 		return;
531 
532 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
533 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
534 
535 	error = xchk_ag_init(sc, agno, &sc->sa);
536 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
537 		return;
538 
539 	xchk_xref_is_used_space(sc, agbno, 1);
540 	xchk_inode_xref_finobt(sc, ino);
541 	xchk_xref_is_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
542 	xchk_xref_is_not_shared(sc, agbno, 1);
543 	xchk_inode_xref_bmap(sc, dip);
544 
545 	xchk_ag_free(sc, &sc->sa);
546 }
547 
548 /*
549  * If the reflink iflag disagrees with a scan for shared data fork extents,
550  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
551  * any shared extents).  We already checked for reflink iflag set on a non
552  * reflink filesystem.
553  */
554 static void
555 xchk_inode_check_reflink_iflag(
556 	struct xfs_scrub	*sc,
557 	xfs_ino_t		ino)
558 {
559 	struct xfs_mount	*mp = sc->mp;
560 	bool			has_shared;
561 	int			error;
562 
563 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
564 		return;
565 
566 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
567 			&has_shared);
568 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
569 			XFS_INO_TO_AGBNO(mp, ino), &error))
570 		return;
571 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
572 		xchk_ino_set_preen(sc, ino);
573 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
574 		xchk_ino_set_corrupt(sc, ino);
575 }
576 
577 /* Scrub an inode. */
578 int
579 xchk_inode(
580 	struct xfs_scrub	*sc)
581 {
582 	struct xfs_dinode	di;
583 	int			error = 0;
584 
585 	/*
586 	 * If sc->ip is NULL, that means that the setup function called
587 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
588 	 * and a NULL inode, so flag the corruption error and return.
589 	 */
590 	if (!sc->ip) {
591 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
592 		return 0;
593 	}
594 
595 	/* Scrub the inode core. */
596 	xfs_inode_to_disk(sc->ip, &di, 0);
597 	xchk_dinode(sc, &di, sc->ip->i_ino);
598 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
599 		goto out;
600 
601 	/*
602 	 * Look for discrepancies between file's data blocks and the reflink
603 	 * iflag.  We already checked the iflag against the file mode when
604 	 * we scrubbed the dinode.
605 	 */
606 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
607 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
608 
609 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
610 out:
611 	return error;
612 }
613