xref: /openbmc/linux/fs/xfs/scrub/inode.c (revision 93696d8f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2017-2023 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
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_trans.h"
15 #include "xfs_ag.h"
16 #include "xfs_inode.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_icache.h"
19 #include "xfs_da_format.h"
20 #include "xfs_reflink.h"
21 #include "xfs_rmap.h"
22 #include "xfs_bmap_util.h"
23 #include "scrub/scrub.h"
24 #include "scrub/common.h"
25 #include "scrub/btree.h"
26 #include "scrub/trace.h"
27 
28 /* Prepare the attached inode for scrubbing. */
29 static inline int
30 xchk_prepare_iscrub(
31 	struct xfs_scrub	*sc)
32 {
33 	int			error;
34 
35 	xchk_ilock(sc, XFS_IOLOCK_EXCL);
36 
37 	error = xchk_trans_alloc(sc, 0);
38 	if (error)
39 		return error;
40 
41 	xchk_ilock(sc, XFS_ILOCK_EXCL);
42 	return 0;
43 }
44 
45 /* Install this scrub-by-handle inode and prepare it for scrubbing. */
46 static inline int
47 xchk_install_handle_iscrub(
48 	struct xfs_scrub	*sc,
49 	struct xfs_inode	*ip)
50 {
51 	int			error;
52 
53 	error = xchk_install_handle_inode(sc, ip);
54 	if (error)
55 		return error;
56 
57 	return xchk_prepare_iscrub(sc);
58 }
59 
60 /*
61  * Grab total control of the inode metadata.  In the best case, we grab the
62  * incore inode and take all locks on it.  If the incore inode cannot be
63  * constructed due to corruption problems, lock the AGI so that we can single
64  * step the loading process to fix everything that can go wrong.
65  */
66 int
67 xchk_setup_inode(
68 	struct xfs_scrub	*sc)
69 {
70 	struct xfs_imap		imap;
71 	struct xfs_inode	*ip;
72 	struct xfs_mount	*mp = sc->mp;
73 	struct xfs_inode	*ip_in = XFS_I(file_inode(sc->file));
74 	struct xfs_buf		*agi_bp;
75 	struct xfs_perag	*pag;
76 	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, sc->sm->sm_ino);
77 	int			error;
78 
79 	if (xchk_need_intent_drain(sc))
80 		xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
81 
82 	/* We want to scan the opened inode, so lock it and exit. */
83 	if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
84 		error = xchk_install_live_inode(sc, ip_in);
85 		if (error)
86 			return error;
87 
88 		return xchk_prepare_iscrub(sc);
89 	}
90 
91 	/* Reject internal metadata files and obviously bad inode numbers. */
92 	if (xfs_internal_inum(mp, sc->sm->sm_ino))
93 		return -ENOENT;
94 	if (!xfs_verify_ino(sc->mp, sc->sm->sm_ino))
95 		return -ENOENT;
96 
97 	/* Try a safe untrusted iget. */
98 	error = xchk_iget_safe(sc, sc->sm->sm_ino, &ip);
99 	if (!error)
100 		return xchk_install_handle_iscrub(sc, ip);
101 	if (error == -ENOENT)
102 		return error;
103 	if (error != -EFSCORRUPTED && error != -EFSBADCRC && error != -EINVAL)
104 		goto out_error;
105 
106 	/*
107 	 * EINVAL with IGET_UNTRUSTED probably means one of several things:
108 	 * userspace gave us an inode number that doesn't correspond to fs
109 	 * space; the inode btree lacks a record for this inode; or there is
110 	 * a record, and it says this inode is free.
111 	 *
112 	 * EFSCORRUPTED/EFSBADCRC could mean that the inode was mappable, but
113 	 * some other metadata corruption (e.g. inode forks) prevented
114 	 * instantiation of the incore inode.  Or it could mean the inobt is
115 	 * corrupt.
116 	 *
117 	 * We want to look up this inode in the inobt directly to distinguish
118 	 * three different scenarios: (1) the inobt says the inode is free,
119 	 * in which case there's nothing to do; (2) the inobt is corrupt so we
120 	 * should flag the corruption and exit to userspace to let it fix the
121 	 * inobt; and (3) the inobt says the inode is allocated, but loading it
122 	 * failed due to corruption.
123 	 *
124 	 * Allocate a transaction and grab the AGI to prevent inobt activity in
125 	 * this AG.  Retry the iget in case someone allocated a new inode after
126 	 * the first iget failed.
127 	 */
128 	error = xchk_trans_alloc(sc, 0);
129 	if (error)
130 		goto out_error;
131 
132 	error = xchk_iget_agi(sc, sc->sm->sm_ino, &agi_bp, &ip);
133 	if (error == 0) {
134 		/* Actually got the incore inode, so install it and proceed. */
135 		xchk_trans_cancel(sc);
136 		return xchk_install_handle_iscrub(sc, ip);
137 	}
138 	if (error == -ENOENT)
139 		goto out_gone;
140 	if (error != -EFSCORRUPTED && error != -EFSBADCRC && error != -EINVAL)
141 		goto out_cancel;
142 
143 	/* Ensure that we have protected against inode allocation/freeing. */
144 	if (agi_bp == NULL) {
145 		ASSERT(agi_bp != NULL);
146 		error = -ECANCELED;
147 		goto out_cancel;
148 	}
149 
150 	/*
151 	 * Untrusted iget failed a second time.  Let's try an inobt lookup.
152 	 * If the inobt doesn't think this is an allocated inode then we'll
153 	 * return ENOENT to signal that the check can be skipped.
154 	 *
155 	 * If the lookup signals corruption, we'll mark this inode corrupt and
156 	 * exit to userspace.  There's little chance of fixing anything until
157 	 * the inobt is straightened out, but there's nothing we can do here.
158 	 *
159 	 * If the lookup encounters a runtime error, exit to userspace.
160 	 */
161 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sc->sm->sm_ino));
162 	if (!pag) {
163 		error = -EFSCORRUPTED;
164 		goto out_cancel;
165 	}
166 
167 	error = xfs_imap(pag, sc->tp, sc->sm->sm_ino, &imap,
168 			XFS_IGET_UNTRUSTED);
169 	xfs_perag_put(pag);
170 	if (error == -EINVAL || error == -ENOENT)
171 		goto out_gone;
172 	if (error)
173 		goto out_cancel;
174 
175 	/*
176 	 * The lookup succeeded.  Chances are the ondisk inode is corrupt and
177 	 * preventing iget from reading it.  Retain the scrub transaction and
178 	 * the AGI buffer to prevent anyone from allocating or freeing inodes.
179 	 * This ensures that we preserve the inconsistency between the inobt
180 	 * saying the inode is allocated and the icache being unable to load
181 	 * the inode until we can flag the corruption in xchk_inode.  The
182 	 * scrub function has to note the corruption, since we're not really
183 	 * supposed to do that from the setup function.
184 	 */
185 	return 0;
186 
187 out_cancel:
188 	xchk_trans_cancel(sc);
189 out_error:
190 	trace_xchk_op_error(sc, agno, XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
191 			error, __return_address);
192 	return error;
193 out_gone:
194 	/* The file is gone, so there's nothing to check. */
195 	xchk_trans_cancel(sc);
196 	return -ENOENT;
197 }
198 
199 /* Inode core */
200 
201 /* Validate di_extsize hint. */
202 STATIC void
203 xchk_inode_extsize(
204 	struct xfs_scrub	*sc,
205 	struct xfs_dinode	*dip,
206 	xfs_ino_t		ino,
207 	uint16_t		mode,
208 	uint16_t		flags)
209 {
210 	xfs_failaddr_t		fa;
211 	uint32_t		value = be32_to_cpu(dip->di_extsize);
212 
213 	fa = xfs_inode_validate_extsize(sc->mp, value, mode, flags);
214 	if (fa)
215 		xchk_ino_set_corrupt(sc, ino);
216 
217 	/*
218 	 * XFS allows a sysadmin to change the rt extent size when adding a rt
219 	 * section to a filesystem after formatting.  If there are any
220 	 * directories with extszinherit and rtinherit set, the hint could
221 	 * become misaligned with the new rextsize.  The verifier doesn't check
222 	 * this, because we allow rtinherit directories even without an rt
223 	 * device.  Flag this as an administrative warning since we will clean
224 	 * this up eventually.
225 	 */
226 	if ((flags & XFS_DIFLAG_RTINHERIT) &&
227 	    (flags & XFS_DIFLAG_EXTSZINHERIT) &&
228 	    value % sc->mp->m_sb.sb_rextsize > 0)
229 		xchk_ino_set_warning(sc, ino);
230 }
231 
232 /*
233  * Validate di_cowextsize hint.
234  *
235  * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
236  * These functions must be kept in sync with each other.
237  */
238 STATIC void
239 xchk_inode_cowextsize(
240 	struct xfs_scrub	*sc,
241 	struct xfs_dinode	*dip,
242 	xfs_ino_t		ino,
243 	uint16_t		mode,
244 	uint16_t		flags,
245 	uint64_t		flags2)
246 {
247 	xfs_failaddr_t		fa;
248 
249 	fa = xfs_inode_validate_cowextsize(sc->mp,
250 			be32_to_cpu(dip->di_cowextsize), mode, flags,
251 			flags2);
252 	if (fa)
253 		xchk_ino_set_corrupt(sc, ino);
254 }
255 
256 /* Make sure the di_flags make sense for the inode. */
257 STATIC void
258 xchk_inode_flags(
259 	struct xfs_scrub	*sc,
260 	struct xfs_dinode	*dip,
261 	xfs_ino_t		ino,
262 	uint16_t		mode,
263 	uint16_t		flags)
264 {
265 	struct xfs_mount	*mp = sc->mp;
266 
267 	/* di_flags are all taken, last bit cannot be used */
268 	if (flags & ~XFS_DIFLAG_ANY)
269 		goto bad;
270 
271 	/* rt flags require rt device */
272 	if ((flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
273 		goto bad;
274 
275 	/* new rt bitmap flag only valid for rbmino */
276 	if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
277 		goto bad;
278 
279 	/* directory-only flags */
280 	if ((flags & (XFS_DIFLAG_RTINHERIT |
281 		     XFS_DIFLAG_EXTSZINHERIT |
282 		     XFS_DIFLAG_PROJINHERIT |
283 		     XFS_DIFLAG_NOSYMLINKS)) &&
284 	    !S_ISDIR(mode))
285 		goto bad;
286 
287 	/* file-only flags */
288 	if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
289 	    !S_ISREG(mode))
290 		goto bad;
291 
292 	/* filestreams and rt make no sense */
293 	if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
294 		goto bad;
295 
296 	return;
297 bad:
298 	xchk_ino_set_corrupt(sc, ino);
299 }
300 
301 /* Make sure the di_flags2 make sense for the inode. */
302 STATIC void
303 xchk_inode_flags2(
304 	struct xfs_scrub	*sc,
305 	struct xfs_dinode	*dip,
306 	xfs_ino_t		ino,
307 	uint16_t		mode,
308 	uint16_t		flags,
309 	uint64_t		flags2)
310 {
311 	struct xfs_mount	*mp = sc->mp;
312 
313 	/* Unknown di_flags2 could be from a future kernel */
314 	if (flags2 & ~XFS_DIFLAG2_ANY)
315 		xchk_ino_set_warning(sc, ino);
316 
317 	/* reflink flag requires reflink feature */
318 	if ((flags2 & XFS_DIFLAG2_REFLINK) &&
319 	    !xfs_has_reflink(mp))
320 		goto bad;
321 
322 	/* cowextsize flag is checked w.r.t. mode separately */
323 
324 	/* file/dir-only flags */
325 	if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
326 		goto bad;
327 
328 	/* file-only flags */
329 	if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
330 		goto bad;
331 
332 	/* realtime and reflink make no sense, currently */
333 	if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK))
334 		goto bad;
335 
336 	/* no bigtime iflag without the bigtime feature */
337 	if (xfs_dinode_has_bigtime(dip) && !xfs_has_bigtime(mp))
338 		goto bad;
339 
340 	/* no large extent counts without the filesystem feature */
341 	if ((flags2 & XFS_DIFLAG2_NREXT64) && !xfs_has_large_extent_counts(mp))
342 		goto bad;
343 
344 	return;
345 bad:
346 	xchk_ino_set_corrupt(sc, ino);
347 }
348 
349 static inline void
350 xchk_dinode_nsec(
351 	struct xfs_scrub	*sc,
352 	xfs_ino_t		ino,
353 	struct xfs_dinode	*dip,
354 	const xfs_timestamp_t	ts)
355 {
356 	struct timespec64	tv;
357 
358 	tv = xfs_inode_from_disk_ts(dip, ts);
359 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
360 		xchk_ino_set_corrupt(sc, ino);
361 }
362 
363 /* Scrub all the ondisk inode fields. */
364 STATIC void
365 xchk_dinode(
366 	struct xfs_scrub	*sc,
367 	struct xfs_dinode	*dip,
368 	xfs_ino_t		ino)
369 {
370 	struct xfs_mount	*mp = sc->mp;
371 	size_t			fork_recs;
372 	unsigned long long	isize;
373 	uint64_t		flags2;
374 	xfs_extnum_t		nextents;
375 	xfs_extnum_t		naextents;
376 	prid_t			prid;
377 	uint16_t		flags;
378 	uint16_t		mode;
379 
380 	flags = be16_to_cpu(dip->di_flags);
381 	if (dip->di_version >= 3)
382 		flags2 = be64_to_cpu(dip->di_flags2);
383 	else
384 		flags2 = 0;
385 
386 	/* di_mode */
387 	mode = be16_to_cpu(dip->di_mode);
388 	switch (mode & S_IFMT) {
389 	case S_IFLNK:
390 	case S_IFREG:
391 	case S_IFDIR:
392 	case S_IFCHR:
393 	case S_IFBLK:
394 	case S_IFIFO:
395 	case S_IFSOCK:
396 		/* mode is recognized */
397 		break;
398 	default:
399 		xchk_ino_set_corrupt(sc, ino);
400 		break;
401 	}
402 
403 	/* v1/v2 fields */
404 	switch (dip->di_version) {
405 	case 1:
406 		/*
407 		 * We autoconvert v1 inodes into v2 inodes on writeout,
408 		 * so just mark this inode for preening.
409 		 */
410 		xchk_ino_set_preen(sc, ino);
411 		prid = 0;
412 		break;
413 	case 2:
414 	case 3:
415 		if (dip->di_onlink != 0)
416 			xchk_ino_set_corrupt(sc, ino);
417 
418 		if (dip->di_mode == 0 && sc->ip)
419 			xchk_ino_set_corrupt(sc, ino);
420 
421 		if (dip->di_projid_hi != 0 &&
422 		    !xfs_has_projid32(mp))
423 			xchk_ino_set_corrupt(sc, ino);
424 
425 		prid = be16_to_cpu(dip->di_projid_lo);
426 		break;
427 	default:
428 		xchk_ino_set_corrupt(sc, ino);
429 		return;
430 	}
431 
432 	if (xfs_has_projid32(mp))
433 		prid |= (prid_t)be16_to_cpu(dip->di_projid_hi) << 16;
434 
435 	/*
436 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
437 	 * userspace could have created that.
438 	 */
439 	if (dip->di_uid == cpu_to_be32(-1U) ||
440 	    dip->di_gid == cpu_to_be32(-1U))
441 		xchk_ino_set_warning(sc, ino);
442 
443 	/*
444 	 * project id of -1 isn't supposed to be valid, but the kernel didn't
445 	 * always validate that.
446 	 */
447 	if (prid == -1U)
448 		xchk_ino_set_warning(sc, ino);
449 
450 	/* di_format */
451 	switch (dip->di_format) {
452 	case XFS_DINODE_FMT_DEV:
453 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
454 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
455 			xchk_ino_set_corrupt(sc, ino);
456 		break;
457 	case XFS_DINODE_FMT_LOCAL:
458 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
459 			xchk_ino_set_corrupt(sc, ino);
460 		break;
461 	case XFS_DINODE_FMT_EXTENTS:
462 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
463 			xchk_ino_set_corrupt(sc, ino);
464 		break;
465 	case XFS_DINODE_FMT_BTREE:
466 		if (!S_ISREG(mode) && !S_ISDIR(mode))
467 			xchk_ino_set_corrupt(sc, ino);
468 		break;
469 	case XFS_DINODE_FMT_UUID:
470 	default:
471 		xchk_ino_set_corrupt(sc, ino);
472 		break;
473 	}
474 
475 	/* di_[amc]time.nsec */
476 	xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
477 	xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
478 	xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
479 
480 	/*
481 	 * di_size.  xfs_dinode_verify checks for things that screw up
482 	 * the VFS such as the upper bit being set and zero-length
483 	 * symlinks/directories, but we can do more here.
484 	 */
485 	isize = be64_to_cpu(dip->di_size);
486 	if (isize & (1ULL << 63))
487 		xchk_ino_set_corrupt(sc, ino);
488 
489 	/* Devices, fifos, and sockets must have zero size */
490 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
491 		xchk_ino_set_corrupt(sc, ino);
492 
493 	/* Directories can't be larger than the data section size (32G) */
494 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
495 		xchk_ino_set_corrupt(sc, ino);
496 
497 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
498 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
499 		xchk_ino_set_corrupt(sc, ino);
500 
501 	/*
502 	 * Warn if the running kernel can't handle the kinds of offsets
503 	 * needed to deal with the file size.  In other words, if the
504 	 * pagecache can't cache all the blocks in this file due to
505 	 * overly large offsets, flag the inode for admin review.
506 	 */
507 	if (isize > mp->m_super->s_maxbytes)
508 		xchk_ino_set_warning(sc, ino);
509 
510 	/* di_nblocks */
511 	if (flags2 & XFS_DIFLAG2_REFLINK) {
512 		; /* nblocks can exceed dblocks */
513 	} else if (flags & XFS_DIFLAG_REALTIME) {
514 		/*
515 		 * nblocks is the sum of data extents (in the rtdev),
516 		 * attr extents (in the datadev), and both forks' bmbt
517 		 * blocks (in the datadev).  This clumsy check is the
518 		 * best we can do without cross-referencing with the
519 		 * inode forks.
520 		 */
521 		if (be64_to_cpu(dip->di_nblocks) >=
522 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
523 			xchk_ino_set_corrupt(sc, ino);
524 	} else {
525 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
526 			xchk_ino_set_corrupt(sc, ino);
527 	}
528 
529 	xchk_inode_flags(sc, dip, ino, mode, flags);
530 
531 	xchk_inode_extsize(sc, dip, ino, mode, flags);
532 
533 	nextents = xfs_dfork_data_extents(dip);
534 	naextents = xfs_dfork_attr_extents(dip);
535 
536 	/* di_nextents */
537 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
538 	switch (dip->di_format) {
539 	case XFS_DINODE_FMT_EXTENTS:
540 		if (nextents > fork_recs)
541 			xchk_ino_set_corrupt(sc, ino);
542 		break;
543 	case XFS_DINODE_FMT_BTREE:
544 		if (nextents <= fork_recs)
545 			xchk_ino_set_corrupt(sc, ino);
546 		break;
547 	default:
548 		if (nextents != 0)
549 			xchk_ino_set_corrupt(sc, ino);
550 		break;
551 	}
552 
553 	/* di_forkoff */
554 	if (XFS_DFORK_APTR(dip) >= (char *)dip + mp->m_sb.sb_inodesize)
555 		xchk_ino_set_corrupt(sc, ino);
556 	if (naextents != 0 && dip->di_forkoff == 0)
557 		xchk_ino_set_corrupt(sc, ino);
558 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
559 		xchk_ino_set_corrupt(sc, ino);
560 
561 	/* di_aformat */
562 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
563 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
564 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
565 		xchk_ino_set_corrupt(sc, ino);
566 
567 	/* di_anextents */
568 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
569 	switch (dip->di_aformat) {
570 	case XFS_DINODE_FMT_EXTENTS:
571 		if (naextents > fork_recs)
572 			xchk_ino_set_corrupt(sc, ino);
573 		break;
574 	case XFS_DINODE_FMT_BTREE:
575 		if (naextents <= fork_recs)
576 			xchk_ino_set_corrupt(sc, ino);
577 		break;
578 	default:
579 		if (naextents != 0)
580 			xchk_ino_set_corrupt(sc, ino);
581 	}
582 
583 	if (dip->di_version >= 3) {
584 		xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
585 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
586 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
587 				flags2);
588 	}
589 }
590 
591 /*
592  * Make sure the finobt doesn't think this inode is free.
593  * We don't have to check the inobt ourselves because we got the inode via
594  * IGET_UNTRUSTED, which checks the inobt for us.
595  */
596 static void
597 xchk_inode_xref_finobt(
598 	struct xfs_scrub		*sc,
599 	xfs_ino_t			ino)
600 {
601 	struct xfs_inobt_rec_incore	rec;
602 	xfs_agino_t			agino;
603 	int				has_record;
604 	int				error;
605 
606 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
607 		return;
608 
609 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
610 
611 	/*
612 	 * Try to get the finobt record.  If we can't get it, then we're
613 	 * in good shape.
614 	 */
615 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
616 			&has_record);
617 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
618 	    !has_record)
619 		return;
620 
621 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
622 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
623 	    !has_record)
624 		return;
625 
626 	/*
627 	 * Otherwise, make sure this record either doesn't cover this inode,
628 	 * or that it does but it's marked present.
629 	 */
630 	if (rec.ir_startino > agino ||
631 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
632 		return;
633 
634 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
635 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
636 }
637 
638 /* Cross reference the inode fields with the forks. */
639 STATIC void
640 xchk_inode_xref_bmap(
641 	struct xfs_scrub	*sc,
642 	struct xfs_dinode	*dip)
643 {
644 	xfs_extnum_t		nextents;
645 	xfs_filblks_t		count;
646 	xfs_filblks_t		acount;
647 	int			error;
648 
649 	if (xchk_skip_xref(sc->sm))
650 		return;
651 
652 	/* Walk all the extents to check nextents/naextents/nblocks. */
653 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
654 			&nextents, &count);
655 	if (!xchk_should_check_xref(sc, &error, NULL))
656 		return;
657 	if (nextents < xfs_dfork_data_extents(dip))
658 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
659 
660 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
661 			&nextents, &acount);
662 	if (!xchk_should_check_xref(sc, &error, NULL))
663 		return;
664 	if (nextents != xfs_dfork_attr_extents(dip))
665 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
666 
667 	/* Check nblocks against the inode. */
668 	if (count + acount != be64_to_cpu(dip->di_nblocks))
669 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
670 }
671 
672 /* Cross-reference with the other btrees. */
673 STATIC void
674 xchk_inode_xref(
675 	struct xfs_scrub	*sc,
676 	xfs_ino_t		ino,
677 	struct xfs_dinode	*dip)
678 {
679 	xfs_agnumber_t		agno;
680 	xfs_agblock_t		agbno;
681 	int			error;
682 
683 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
684 		return;
685 
686 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
687 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
688 
689 	error = xchk_ag_init_existing(sc, agno, &sc->sa);
690 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
691 		goto out_free;
692 
693 	xchk_xref_is_used_space(sc, agbno, 1);
694 	xchk_inode_xref_finobt(sc, ino);
695 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
696 	xchk_xref_is_not_shared(sc, agbno, 1);
697 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
698 	xchk_inode_xref_bmap(sc, dip);
699 
700 out_free:
701 	xchk_ag_free(sc, &sc->sa);
702 }
703 
704 /*
705  * If the reflink iflag disagrees with a scan for shared data fork extents,
706  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
707  * any shared extents).  We already checked for reflink iflag set on a non
708  * reflink filesystem.
709  */
710 static void
711 xchk_inode_check_reflink_iflag(
712 	struct xfs_scrub	*sc,
713 	xfs_ino_t		ino)
714 {
715 	struct xfs_mount	*mp = sc->mp;
716 	bool			has_shared;
717 	int			error;
718 
719 	if (!xfs_has_reflink(mp))
720 		return;
721 
722 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
723 			&has_shared);
724 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
725 			XFS_INO_TO_AGBNO(mp, ino), &error))
726 		return;
727 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
728 		xchk_ino_set_preen(sc, ino);
729 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
730 		xchk_ino_set_corrupt(sc, ino);
731 }
732 
733 /* Scrub an inode. */
734 int
735 xchk_inode(
736 	struct xfs_scrub	*sc)
737 {
738 	struct xfs_dinode	di;
739 	int			error = 0;
740 
741 	/*
742 	 * If sc->ip is NULL, that means that the setup function called
743 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
744 	 * and a NULL inode, so flag the corruption error and return.
745 	 */
746 	if (!sc->ip) {
747 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
748 		return 0;
749 	}
750 
751 	/* Scrub the inode core. */
752 	xfs_inode_to_disk(sc->ip, &di, 0);
753 	xchk_dinode(sc, &di, sc->ip->i_ino);
754 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
755 		goto out;
756 
757 	/*
758 	 * Look for discrepancies between file's data blocks and the reflink
759 	 * iflag.  We already checked the iflag against the file mode when
760 	 * we scrubbed the dinode.
761 	 */
762 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
763 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
764 
765 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
766 out:
767 	return error;
768 }
769