xref: /linux/fs/xfs/scrub/inode.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 "xfs_rtbitmap.h"
24 #include "scrub/scrub.h"
25 #include "scrub/common.h"
26 #include "scrub/btree.h"
27 #include "scrub/trace.h"
28 #include "scrub/repair.h"
29 
30 /* Prepare the attached inode for scrubbing. */
31 static inline int
32 xchk_prepare_iscrub(
33 	struct xfs_scrub	*sc)
34 {
35 	int			error;
36 
37 	xchk_ilock(sc, XFS_IOLOCK_EXCL);
38 
39 	error = xchk_trans_alloc(sc, 0);
40 	if (error)
41 		return error;
42 
43 	error = xchk_ino_dqattach(sc);
44 	if (error)
45 		return error;
46 
47 	xchk_ilock(sc, XFS_ILOCK_EXCL);
48 	return 0;
49 }
50 
51 /* Install this scrub-by-handle inode and prepare it for scrubbing. */
52 static inline int
53 xchk_install_handle_iscrub(
54 	struct xfs_scrub	*sc,
55 	struct xfs_inode	*ip)
56 {
57 	int			error;
58 
59 	error = xchk_install_handle_inode(sc, ip);
60 	if (error)
61 		return error;
62 
63 	return xchk_prepare_iscrub(sc);
64 }
65 
66 /*
67  * Grab total control of the inode metadata.  In the best case, we grab the
68  * incore inode and take all locks on it.  If the incore inode cannot be
69  * constructed due to corruption problems, lock the AGI so that we can single
70  * step the loading process to fix everything that can go wrong.
71  */
72 int
73 xchk_setup_inode(
74 	struct xfs_scrub	*sc)
75 {
76 	struct xfs_imap		imap;
77 	struct xfs_inode	*ip;
78 	struct xfs_mount	*mp = sc->mp;
79 	struct xfs_inode	*ip_in = XFS_I(file_inode(sc->file));
80 	struct xfs_buf		*agi_bp;
81 	struct xfs_perag	*pag;
82 	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, sc->sm->sm_ino);
83 	int			error;
84 
85 	if (xchk_need_intent_drain(sc))
86 		xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
87 
88 	/* We want to scan the opened inode, so lock it and exit. */
89 	if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
90 		error = xchk_install_live_inode(sc, ip_in);
91 		if (error)
92 			return error;
93 
94 		return xchk_prepare_iscrub(sc);
95 	}
96 
97 	/* Reject internal metadata files and obviously bad inode numbers. */
98 	if (xfs_internal_inum(mp, sc->sm->sm_ino))
99 		return -ENOENT;
100 	if (!xfs_verify_ino(sc->mp, sc->sm->sm_ino))
101 		return -ENOENT;
102 
103 	/* Try a safe untrusted iget. */
104 	error = xchk_iget_safe(sc, sc->sm->sm_ino, &ip);
105 	if (!error)
106 		return xchk_install_handle_iscrub(sc, ip);
107 	if (error == -ENOENT)
108 		return error;
109 	if (error != -EFSCORRUPTED && error != -EFSBADCRC && error != -EINVAL)
110 		goto out_error;
111 
112 	/*
113 	 * EINVAL with IGET_UNTRUSTED probably means one of several things:
114 	 * userspace gave us an inode number that doesn't correspond to fs
115 	 * space; the inode btree lacks a record for this inode; or there is
116 	 * a record, and it says this inode is free.
117 	 *
118 	 * EFSCORRUPTED/EFSBADCRC could mean that the inode was mappable, but
119 	 * some other metadata corruption (e.g. inode forks) prevented
120 	 * instantiation of the incore inode.  Or it could mean the inobt is
121 	 * corrupt.
122 	 *
123 	 * We want to look up this inode in the inobt directly to distinguish
124 	 * three different scenarios: (1) the inobt says the inode is free,
125 	 * in which case there's nothing to do; (2) the inobt is corrupt so we
126 	 * should flag the corruption and exit to userspace to let it fix the
127 	 * inobt; and (3) the inobt says the inode is allocated, but loading it
128 	 * failed due to corruption.
129 	 *
130 	 * Allocate a transaction and grab the AGI to prevent inobt activity in
131 	 * this AG.  Retry the iget in case someone allocated a new inode after
132 	 * the first iget failed.
133 	 */
134 	error = xchk_trans_alloc(sc, 0);
135 	if (error)
136 		goto out_error;
137 
138 	error = xchk_iget_agi(sc, sc->sm->sm_ino, &agi_bp, &ip);
139 	if (error == 0) {
140 		/* Actually got the incore inode, so install it and proceed. */
141 		xchk_trans_cancel(sc);
142 		return xchk_install_handle_iscrub(sc, ip);
143 	}
144 	if (error == -ENOENT)
145 		goto out_gone;
146 	if (error != -EFSCORRUPTED && error != -EFSBADCRC && error != -EINVAL)
147 		goto out_cancel;
148 
149 	/* Ensure that we have protected against inode allocation/freeing. */
150 	if (agi_bp == NULL) {
151 		ASSERT(agi_bp != NULL);
152 		error = -ECANCELED;
153 		goto out_cancel;
154 	}
155 
156 	/*
157 	 * Untrusted iget failed a second time.  Let's try an inobt lookup.
158 	 * If the inobt doesn't think this is an allocated inode then we'll
159 	 * return ENOENT to signal that the check can be skipped.
160 	 *
161 	 * If the lookup signals corruption, we'll mark this inode corrupt and
162 	 * exit to userspace.  There's little chance of fixing anything until
163 	 * the inobt is straightened out, but there's nothing we can do here.
164 	 *
165 	 * If the lookup encounters a runtime error, exit to userspace.
166 	 */
167 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sc->sm->sm_ino));
168 	if (!pag) {
169 		error = -EFSCORRUPTED;
170 		goto out_cancel;
171 	}
172 
173 	error = xfs_imap(pag, sc->tp, sc->sm->sm_ino, &imap,
174 			XFS_IGET_UNTRUSTED);
175 	xfs_perag_put(pag);
176 	if (error == -EINVAL || error == -ENOENT)
177 		goto out_gone;
178 	if (error)
179 		goto out_cancel;
180 
181 	/*
182 	 * The lookup succeeded.  Chances are the ondisk inode is corrupt and
183 	 * preventing iget from reading it.  Retain the scrub transaction and
184 	 * the AGI buffer to prevent anyone from allocating or freeing inodes.
185 	 * This ensures that we preserve the inconsistency between the inobt
186 	 * saying the inode is allocated and the icache being unable to load
187 	 * the inode until we can flag the corruption in xchk_inode.  The
188 	 * scrub function has to note the corruption, since we're not really
189 	 * supposed to do that from the setup function.  Save the mapping to
190 	 * make repairs to the ondisk inode buffer.
191 	 */
192 	if (xchk_could_repair(sc))
193 		xrep_setup_inode(sc, &imap);
194 	return 0;
195 
196 out_cancel:
197 	xchk_trans_cancel(sc);
198 out_error:
199 	trace_xchk_op_error(sc, agno, XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
200 			error, __return_address);
201 	return error;
202 out_gone:
203 	/* The file is gone, so there's nothing to check. */
204 	xchk_trans_cancel(sc);
205 	return -ENOENT;
206 }
207 
208 /* Inode core */
209 
210 /* Validate di_extsize hint. */
211 STATIC void
212 xchk_inode_extsize(
213 	struct xfs_scrub	*sc,
214 	struct xfs_dinode	*dip,
215 	xfs_ino_t		ino,
216 	uint16_t		mode,
217 	uint16_t		flags)
218 {
219 	xfs_failaddr_t		fa;
220 	uint32_t		value = be32_to_cpu(dip->di_extsize);
221 
222 	fa = xfs_inode_validate_extsize(sc->mp, value, mode, flags);
223 	if (fa)
224 		xchk_ino_set_corrupt(sc, ino);
225 
226 	/*
227 	 * XFS allows a sysadmin to change the rt extent size when adding a rt
228 	 * section to a filesystem after formatting.  If there are any
229 	 * directories with extszinherit and rtinherit set, the hint could
230 	 * become misaligned with the new rextsize.  The verifier doesn't check
231 	 * this, because we allow rtinherit directories even without an rt
232 	 * device.  Flag this as an administrative warning since we will clean
233 	 * this up eventually.
234 	 */
235 	if ((flags & XFS_DIFLAG_RTINHERIT) &&
236 	    (flags & XFS_DIFLAG_EXTSZINHERIT) &&
237 	    xfs_extlen_to_rtxmod(sc->mp, value) > 0)
238 		xchk_ino_set_warning(sc, ino);
239 }
240 
241 /*
242  * Validate di_cowextsize hint.
243  *
244  * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
245  * These functions must be kept in sync with each other.
246  */
247 STATIC void
248 xchk_inode_cowextsize(
249 	struct xfs_scrub	*sc,
250 	struct xfs_dinode	*dip,
251 	xfs_ino_t		ino,
252 	uint16_t		mode,
253 	uint16_t		flags,
254 	uint64_t		flags2)
255 {
256 	xfs_failaddr_t		fa;
257 
258 	fa = xfs_inode_validate_cowextsize(sc->mp,
259 			be32_to_cpu(dip->di_cowextsize), mode, flags,
260 			flags2);
261 	if (fa)
262 		xchk_ino_set_corrupt(sc, ino);
263 }
264 
265 /* Make sure the di_flags make sense for the inode. */
266 STATIC void
267 xchk_inode_flags(
268 	struct xfs_scrub	*sc,
269 	struct xfs_dinode	*dip,
270 	xfs_ino_t		ino,
271 	uint16_t		mode,
272 	uint16_t		flags)
273 {
274 	struct xfs_mount	*mp = sc->mp;
275 
276 	/* di_flags are all taken, last bit cannot be used */
277 	if (flags & ~XFS_DIFLAG_ANY)
278 		goto bad;
279 
280 	/* rt flags require rt device */
281 	if ((flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
282 		goto bad;
283 
284 	/* new rt bitmap flag only valid for rbmino */
285 	if ((flags & XFS_DIFLAG_NEWRTBM) && ino != mp->m_sb.sb_rbmino)
286 		goto bad;
287 
288 	/* directory-only flags */
289 	if ((flags & (XFS_DIFLAG_RTINHERIT |
290 		     XFS_DIFLAG_EXTSZINHERIT |
291 		     XFS_DIFLAG_PROJINHERIT |
292 		     XFS_DIFLAG_NOSYMLINKS)) &&
293 	    !S_ISDIR(mode))
294 		goto bad;
295 
296 	/* file-only flags */
297 	if ((flags & (XFS_DIFLAG_REALTIME | FS_XFLAG_EXTSIZE)) &&
298 	    !S_ISREG(mode))
299 		goto bad;
300 
301 	/* filestreams and rt make no sense */
302 	if ((flags & XFS_DIFLAG_FILESTREAM) && (flags & XFS_DIFLAG_REALTIME))
303 		goto bad;
304 
305 	return;
306 bad:
307 	xchk_ino_set_corrupt(sc, ino);
308 }
309 
310 /* Make sure the di_flags2 make sense for the inode. */
311 STATIC void
312 xchk_inode_flags2(
313 	struct xfs_scrub	*sc,
314 	struct xfs_dinode	*dip,
315 	xfs_ino_t		ino,
316 	uint16_t		mode,
317 	uint16_t		flags,
318 	uint64_t		flags2)
319 {
320 	struct xfs_mount	*mp = sc->mp;
321 
322 	/* Unknown di_flags2 could be from a future kernel */
323 	if (flags2 & ~XFS_DIFLAG2_ANY)
324 		xchk_ino_set_warning(sc, ino);
325 
326 	/* reflink flag requires reflink feature */
327 	if ((flags2 & XFS_DIFLAG2_REFLINK) &&
328 	    !xfs_has_reflink(mp))
329 		goto bad;
330 
331 	/* cowextsize flag is checked w.r.t. mode separately */
332 
333 	/* file/dir-only flags */
334 	if ((flags2 & XFS_DIFLAG2_DAX) && !(S_ISREG(mode) || S_ISDIR(mode)))
335 		goto bad;
336 
337 	/* file-only flags */
338 	if ((flags2 & XFS_DIFLAG2_REFLINK) && !S_ISREG(mode))
339 		goto bad;
340 
341 	/* realtime and reflink make no sense, currently */
342 	if ((flags & XFS_DIFLAG_REALTIME) && (flags2 & XFS_DIFLAG2_REFLINK))
343 		goto bad;
344 
345 	/* no bigtime iflag without the bigtime feature */
346 	if (xfs_dinode_has_bigtime(dip) && !xfs_has_bigtime(mp))
347 		goto bad;
348 
349 	/* no large extent counts without the filesystem feature */
350 	if ((flags2 & XFS_DIFLAG2_NREXT64) && !xfs_has_large_extent_counts(mp))
351 		goto bad;
352 
353 	return;
354 bad:
355 	xchk_ino_set_corrupt(sc, ino);
356 }
357 
358 static inline void
359 xchk_dinode_nsec(
360 	struct xfs_scrub	*sc,
361 	xfs_ino_t		ino,
362 	struct xfs_dinode	*dip,
363 	const xfs_timestamp_t	ts)
364 {
365 	struct timespec64	tv;
366 
367 	tv = xfs_inode_from_disk_ts(dip, ts);
368 	if (tv.tv_nsec < 0 || tv.tv_nsec >= NSEC_PER_SEC)
369 		xchk_ino_set_corrupt(sc, ino);
370 }
371 
372 /* Scrub all the ondisk inode fields. */
373 STATIC void
374 xchk_dinode(
375 	struct xfs_scrub	*sc,
376 	struct xfs_dinode	*dip,
377 	xfs_ino_t		ino)
378 {
379 	struct xfs_mount	*mp = sc->mp;
380 	size_t			fork_recs;
381 	unsigned long long	isize;
382 	uint64_t		flags2;
383 	xfs_extnum_t		nextents;
384 	xfs_extnum_t		naextents;
385 	prid_t			prid;
386 	uint16_t		flags;
387 	uint16_t		mode;
388 
389 	flags = be16_to_cpu(dip->di_flags);
390 	if (dip->di_version >= 3)
391 		flags2 = be64_to_cpu(dip->di_flags2);
392 	else
393 		flags2 = 0;
394 
395 	/* di_mode */
396 	mode = be16_to_cpu(dip->di_mode);
397 	switch (mode & S_IFMT) {
398 	case S_IFLNK:
399 	case S_IFREG:
400 	case S_IFDIR:
401 	case S_IFCHR:
402 	case S_IFBLK:
403 	case S_IFIFO:
404 	case S_IFSOCK:
405 		/* mode is recognized */
406 		break;
407 	default:
408 		xchk_ino_set_corrupt(sc, ino);
409 		break;
410 	}
411 
412 	/* v1/v2 fields */
413 	switch (dip->di_version) {
414 	case 1:
415 		/*
416 		 * We autoconvert v1 inodes into v2 inodes on writeout,
417 		 * so just mark this inode for preening.
418 		 */
419 		xchk_ino_set_preen(sc, ino);
420 		prid = 0;
421 		break;
422 	case 2:
423 	case 3:
424 		if (dip->di_onlink != 0)
425 			xchk_ino_set_corrupt(sc, ino);
426 
427 		if (dip->di_mode == 0 && sc->ip)
428 			xchk_ino_set_corrupt(sc, ino);
429 
430 		if (dip->di_projid_hi != 0 &&
431 		    !xfs_has_projid32(mp))
432 			xchk_ino_set_corrupt(sc, ino);
433 
434 		prid = be16_to_cpu(dip->di_projid_lo);
435 		break;
436 	default:
437 		xchk_ino_set_corrupt(sc, ino);
438 		return;
439 	}
440 
441 	if (xfs_has_projid32(mp))
442 		prid |= (prid_t)be16_to_cpu(dip->di_projid_hi) << 16;
443 
444 	/*
445 	 * di_uid/di_gid -- -1 isn't invalid, but there's no way that
446 	 * userspace could have created that.
447 	 */
448 	if (dip->di_uid == cpu_to_be32(-1U) ||
449 	    dip->di_gid == cpu_to_be32(-1U))
450 		xchk_ino_set_warning(sc, ino);
451 
452 	/*
453 	 * project id of -1 isn't supposed to be valid, but the kernel didn't
454 	 * always validate that.
455 	 */
456 	if (prid == -1U)
457 		xchk_ino_set_warning(sc, ino);
458 
459 	/* di_format */
460 	switch (dip->di_format) {
461 	case XFS_DINODE_FMT_DEV:
462 		if (!S_ISCHR(mode) && !S_ISBLK(mode) &&
463 		    !S_ISFIFO(mode) && !S_ISSOCK(mode))
464 			xchk_ino_set_corrupt(sc, ino);
465 		break;
466 	case XFS_DINODE_FMT_LOCAL:
467 		if (!S_ISDIR(mode) && !S_ISLNK(mode))
468 			xchk_ino_set_corrupt(sc, ino);
469 		break;
470 	case XFS_DINODE_FMT_EXTENTS:
471 		if (!S_ISREG(mode) && !S_ISDIR(mode) && !S_ISLNK(mode))
472 			xchk_ino_set_corrupt(sc, ino);
473 		break;
474 	case XFS_DINODE_FMT_BTREE:
475 		if (!S_ISREG(mode) && !S_ISDIR(mode))
476 			xchk_ino_set_corrupt(sc, ino);
477 		break;
478 	case XFS_DINODE_FMT_UUID:
479 	default:
480 		xchk_ino_set_corrupt(sc, ino);
481 		break;
482 	}
483 
484 	/* di_[amc]time.nsec */
485 	xchk_dinode_nsec(sc, ino, dip, dip->di_atime);
486 	xchk_dinode_nsec(sc, ino, dip, dip->di_mtime);
487 	xchk_dinode_nsec(sc, ino, dip, dip->di_ctime);
488 
489 	/*
490 	 * di_size.  xfs_dinode_verify checks for things that screw up
491 	 * the VFS such as the upper bit being set and zero-length
492 	 * symlinks/directories, but we can do more here.
493 	 */
494 	isize = be64_to_cpu(dip->di_size);
495 	if (isize & (1ULL << 63))
496 		xchk_ino_set_corrupt(sc, ino);
497 
498 	/* Devices, fifos, and sockets must have zero size */
499 	if (!S_ISDIR(mode) && !S_ISREG(mode) && !S_ISLNK(mode) && isize != 0)
500 		xchk_ino_set_corrupt(sc, ino);
501 
502 	/* Directories can't be larger than the data section size (32G) */
503 	if (S_ISDIR(mode) && (isize == 0 || isize >= XFS_DIR2_SPACE_SIZE))
504 		xchk_ino_set_corrupt(sc, ino);
505 
506 	/* Symlinks can't be larger than SYMLINK_MAXLEN */
507 	if (S_ISLNK(mode) && (isize == 0 || isize >= XFS_SYMLINK_MAXLEN))
508 		xchk_ino_set_corrupt(sc, ino);
509 
510 	/*
511 	 * Warn if the running kernel can't handle the kinds of offsets
512 	 * needed to deal with the file size.  In other words, if the
513 	 * pagecache can't cache all the blocks in this file due to
514 	 * overly large offsets, flag the inode for admin review.
515 	 */
516 	if (isize > mp->m_super->s_maxbytes)
517 		xchk_ino_set_warning(sc, ino);
518 
519 	/* di_nblocks */
520 	if (flags2 & XFS_DIFLAG2_REFLINK) {
521 		; /* nblocks can exceed dblocks */
522 	} else if (flags & XFS_DIFLAG_REALTIME) {
523 		/*
524 		 * nblocks is the sum of data extents (in the rtdev),
525 		 * attr extents (in the datadev), and both forks' bmbt
526 		 * blocks (in the datadev).  This clumsy check is the
527 		 * best we can do without cross-referencing with the
528 		 * inode forks.
529 		 */
530 		if (be64_to_cpu(dip->di_nblocks) >=
531 		    mp->m_sb.sb_dblocks + mp->m_sb.sb_rblocks)
532 			xchk_ino_set_corrupt(sc, ino);
533 	} else {
534 		if (be64_to_cpu(dip->di_nblocks) >= mp->m_sb.sb_dblocks)
535 			xchk_ino_set_corrupt(sc, ino);
536 	}
537 
538 	xchk_inode_flags(sc, dip, ino, mode, flags);
539 
540 	xchk_inode_extsize(sc, dip, ino, mode, flags);
541 
542 	nextents = xfs_dfork_data_extents(dip);
543 	naextents = xfs_dfork_attr_extents(dip);
544 
545 	/* di_nextents */
546 	fork_recs =  XFS_DFORK_DSIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
547 	switch (dip->di_format) {
548 	case XFS_DINODE_FMT_EXTENTS:
549 		if (nextents > fork_recs)
550 			xchk_ino_set_corrupt(sc, ino);
551 		break;
552 	case XFS_DINODE_FMT_BTREE:
553 		if (nextents <= fork_recs)
554 			xchk_ino_set_corrupt(sc, ino);
555 		break;
556 	default:
557 		if (nextents != 0)
558 			xchk_ino_set_corrupt(sc, ino);
559 		break;
560 	}
561 
562 	/* di_forkoff */
563 	if (XFS_DFORK_BOFF(dip) >= mp->m_sb.sb_inodesize)
564 		xchk_ino_set_corrupt(sc, ino);
565 	if (naextents != 0 && dip->di_forkoff == 0)
566 		xchk_ino_set_corrupt(sc, ino);
567 	if (dip->di_forkoff == 0 && dip->di_aformat != XFS_DINODE_FMT_EXTENTS)
568 		xchk_ino_set_corrupt(sc, ino);
569 
570 	/* di_aformat */
571 	if (dip->di_aformat != XFS_DINODE_FMT_LOCAL &&
572 	    dip->di_aformat != XFS_DINODE_FMT_EXTENTS &&
573 	    dip->di_aformat != XFS_DINODE_FMT_BTREE)
574 		xchk_ino_set_corrupt(sc, ino);
575 
576 	/* di_anextents */
577 	fork_recs =  XFS_DFORK_ASIZE(dip, mp) / sizeof(struct xfs_bmbt_rec);
578 	switch (dip->di_aformat) {
579 	case XFS_DINODE_FMT_EXTENTS:
580 		if (naextents > fork_recs)
581 			xchk_ino_set_corrupt(sc, ino);
582 		break;
583 	case XFS_DINODE_FMT_BTREE:
584 		if (naextents <= fork_recs)
585 			xchk_ino_set_corrupt(sc, ino);
586 		break;
587 	default:
588 		if (naextents != 0)
589 			xchk_ino_set_corrupt(sc, ino);
590 	}
591 
592 	if (dip->di_version >= 3) {
593 		xchk_dinode_nsec(sc, ino, dip, dip->di_crtime);
594 		xchk_inode_flags2(sc, dip, ino, mode, flags, flags2);
595 		xchk_inode_cowextsize(sc, dip, ino, mode, flags,
596 				flags2);
597 	}
598 }
599 
600 /*
601  * Make sure the finobt doesn't think this inode is free.
602  * We don't have to check the inobt ourselves because we got the inode via
603  * IGET_UNTRUSTED, which checks the inobt for us.
604  */
605 static void
606 xchk_inode_xref_finobt(
607 	struct xfs_scrub		*sc,
608 	xfs_ino_t			ino)
609 {
610 	struct xfs_inobt_rec_incore	rec;
611 	xfs_agino_t			agino;
612 	int				has_record;
613 	int				error;
614 
615 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
616 		return;
617 
618 	agino = XFS_INO_TO_AGINO(sc->mp, ino);
619 
620 	/*
621 	 * Try to get the finobt record.  If we can't get it, then we're
622 	 * in good shape.
623 	 */
624 	error = xfs_inobt_lookup(sc->sa.fino_cur, agino, XFS_LOOKUP_LE,
625 			&has_record);
626 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
627 	    !has_record)
628 		return;
629 
630 	error = xfs_inobt_get_rec(sc->sa.fino_cur, &rec, &has_record);
631 	if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur) ||
632 	    !has_record)
633 		return;
634 
635 	/*
636 	 * Otherwise, make sure this record either doesn't cover this inode,
637 	 * or that it does but it's marked present.
638 	 */
639 	if (rec.ir_startino > agino ||
640 	    rec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
641 		return;
642 
643 	if (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))
644 		xchk_btree_xref_set_corrupt(sc, sc->sa.fino_cur, 0);
645 }
646 
647 /* Cross reference the inode fields with the forks. */
648 STATIC void
649 xchk_inode_xref_bmap(
650 	struct xfs_scrub	*sc,
651 	struct xfs_dinode	*dip)
652 {
653 	xfs_extnum_t		nextents;
654 	xfs_filblks_t		count;
655 	xfs_filblks_t		acount;
656 	int			error;
657 
658 	if (xchk_skip_xref(sc->sm))
659 		return;
660 
661 	/* Walk all the extents to check nextents/naextents/nblocks. */
662 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,
663 			&nextents, &count);
664 	if (!xchk_should_check_xref(sc, &error, NULL))
665 		return;
666 	if (nextents < xfs_dfork_data_extents(dip))
667 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
668 
669 	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,
670 			&nextents, &acount);
671 	if (!xchk_should_check_xref(sc, &error, NULL))
672 		return;
673 	if (nextents != xfs_dfork_attr_extents(dip))
674 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
675 
676 	/* Check nblocks against the inode. */
677 	if (count + acount != be64_to_cpu(dip->di_nblocks))
678 		xchk_ino_xref_set_corrupt(sc, sc->ip->i_ino);
679 }
680 
681 /* Cross-reference with the other btrees. */
682 STATIC void
683 xchk_inode_xref(
684 	struct xfs_scrub	*sc,
685 	xfs_ino_t		ino,
686 	struct xfs_dinode	*dip)
687 {
688 	xfs_agnumber_t		agno;
689 	xfs_agblock_t		agbno;
690 	int			error;
691 
692 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
693 		return;
694 
695 	agno = XFS_INO_TO_AGNO(sc->mp, ino);
696 	agbno = XFS_INO_TO_AGBNO(sc->mp, ino);
697 
698 	error = xchk_ag_init_existing(sc, agno, &sc->sa);
699 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
700 		goto out_free;
701 
702 	xchk_xref_is_used_space(sc, agbno, 1);
703 	xchk_inode_xref_finobt(sc, ino);
704 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_INODES);
705 	xchk_xref_is_not_shared(sc, agbno, 1);
706 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
707 	xchk_inode_xref_bmap(sc, dip);
708 
709 out_free:
710 	xchk_ag_free(sc, &sc->sa);
711 }
712 
713 /*
714  * If the reflink iflag disagrees with a scan for shared data fork extents,
715  * either flag an error (shared extents w/ no flag) or a preen (flag set w/o
716  * any shared extents).  We already checked for reflink iflag set on a non
717  * reflink filesystem.
718  */
719 static void
720 xchk_inode_check_reflink_iflag(
721 	struct xfs_scrub	*sc,
722 	xfs_ino_t		ino)
723 {
724 	struct xfs_mount	*mp = sc->mp;
725 	bool			has_shared;
726 	int			error;
727 
728 	if (!xfs_has_reflink(mp))
729 		return;
730 
731 	error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
732 			&has_shared);
733 	if (!xchk_xref_process_error(sc, XFS_INO_TO_AGNO(mp, ino),
734 			XFS_INO_TO_AGBNO(mp, ino), &error))
735 		return;
736 	if (xfs_is_reflink_inode(sc->ip) && !has_shared)
737 		xchk_ino_set_preen(sc, ino);
738 	else if (!xfs_is_reflink_inode(sc->ip) && has_shared)
739 		xchk_ino_set_corrupt(sc, ino);
740 }
741 
742 /*
743  * If this inode has zero link count, it must be on the unlinked list.  If
744  * it has nonzero link count, it must not be on the unlinked list.
745  */
746 STATIC void
747 xchk_inode_check_unlinked(
748 	struct xfs_scrub	*sc)
749 {
750 	if (VFS_I(sc->ip)->i_nlink == 0) {
751 		if (!xfs_inode_on_unlinked_list(sc->ip))
752 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
753 	} else {
754 		if (xfs_inode_on_unlinked_list(sc->ip))
755 			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
756 	}
757 }
758 
759 /* Scrub an inode. */
760 int
761 xchk_inode(
762 	struct xfs_scrub	*sc)
763 {
764 	struct xfs_dinode	di;
765 	int			error = 0;
766 
767 	/*
768 	 * If sc->ip is NULL, that means that the setup function called
769 	 * xfs_iget to look up the inode.  xfs_iget returned a EFSCORRUPTED
770 	 * and a NULL inode, so flag the corruption error and return.
771 	 */
772 	if (!sc->ip) {
773 		xchk_ino_set_corrupt(sc, sc->sm->sm_ino);
774 		return 0;
775 	}
776 
777 	/* Scrub the inode core. */
778 	xfs_inode_to_disk(sc->ip, &di, 0);
779 	xchk_dinode(sc, &di, sc->ip->i_ino);
780 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
781 		goto out;
782 
783 	/*
784 	 * Look for discrepancies between file's data blocks and the reflink
785 	 * iflag.  We already checked the iflag against the file mode when
786 	 * we scrubbed the dinode.
787 	 */
788 	if (S_ISREG(VFS_I(sc->ip)->i_mode))
789 		xchk_inode_check_reflink_iflag(sc, sc->ip->i_ino);
790 
791 	xchk_inode_check_unlinked(sc);
792 
793 	xchk_inode_xref(sc, sc->ip->i_ino, &di);
794 out:
795 	return error;
796 }
797