xref: /linux/fs/xfs/scrub/ialloc.c (revision 3eee9ac24cef892e6883b3669544c6101b70c91e)
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_inode.h"
16 #include "xfs_ialloc.h"
17 #include "xfs_ialloc_btree.h"
18 #include "xfs_icache.h"
19 #include "xfs_rmap.h"
20 #include "scrub/scrub.h"
21 #include "scrub/common.h"
22 #include "scrub/btree.h"
23 #include "scrub/trace.h"
24 #include "xfs_ag.h"
25 
26 /*
27  * Set us up to scrub inode btrees.
28  * If we detect a discrepancy between the inobt and the inode,
29  * try again after forcing logged inode cores out to disk.
30  */
31 int
32 xchk_setup_ag_iallocbt(
33 	struct xfs_scrub	*sc)
34 {
35 	if (xchk_need_intent_drain(sc))
36 		xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
37 	return xchk_setup_ag_btree(sc, sc->flags & XCHK_TRY_HARDER);
38 }
39 
40 /* Inode btree scrubber. */
41 
42 struct xchk_iallocbt {
43 	/* Number of inodes we see while scanning inobt. */
44 	unsigned long long	inodes;
45 
46 	/* Expected next startino, for big block filesystems. */
47 	xfs_agino_t		next_startino;
48 
49 	/* Expected end of the current inode cluster. */
50 	xfs_agino_t		next_cluster_ino;
51 };
52 
53 /*
54  * Does the finobt have a record for this inode with the same hole/free state?
55  * This is a bit complicated because of the following:
56  *
57  * - The finobt need not have a record if all inodes in the inobt record are
58  *   allocated.
59  * - The finobt need not have a record if all inodes in the inobt record are
60  *   free.
61  * - The finobt need not have a record if the inobt record says this is a hole.
62  *   This likely doesn't happen in practice.
63  */
64 STATIC int
65 xchk_inobt_xref_finobt(
66 	struct xfs_scrub	*sc,
67 	struct xfs_inobt_rec_incore *irec,
68 	xfs_agino_t		agino,
69 	bool			free,
70 	bool			hole)
71 {
72 	struct xfs_inobt_rec_incore frec;
73 	struct xfs_btree_cur	*cur = sc->sa.fino_cur;
74 	bool			ffree, fhole;
75 	unsigned int		frec_idx, fhole_idx;
76 	int			has_record;
77 	int			error;
78 
79 	ASSERT(xfs_btree_is_fino(cur->bc_ops));
80 
81 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has_record);
82 	if (error)
83 		return error;
84 	if (!has_record)
85 		goto no_record;
86 
87 	error = xfs_inobt_get_rec(cur, &frec, &has_record);
88 	if (!has_record)
89 		return -EFSCORRUPTED;
90 
91 	if (frec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
92 		goto no_record;
93 
94 	/* There's a finobt record; free and hole status must match. */
95 	frec_idx = agino - frec.ir_startino;
96 	ffree = frec.ir_free & (1ULL << frec_idx);
97 	fhole_idx = frec_idx / XFS_INODES_PER_HOLEMASK_BIT;
98 	fhole = frec.ir_holemask & (1U << fhole_idx);
99 
100 	if (ffree != free)
101 		xchk_btree_xref_set_corrupt(sc, cur, 0);
102 	if (fhole != hole)
103 		xchk_btree_xref_set_corrupt(sc, cur, 0);
104 	return 0;
105 
106 no_record:
107 	/* inobt record is fully allocated */
108 	if (irec->ir_free == 0)
109 		return 0;
110 
111 	/* inobt record is totally unallocated */
112 	if (irec->ir_free == XFS_INOBT_ALL_FREE)
113 		return 0;
114 
115 	/* inobt record says this is a hole */
116 	if (hole)
117 		return 0;
118 
119 	/* finobt doesn't care about allocated inodes */
120 	if (!free)
121 		return 0;
122 
123 	xchk_btree_xref_set_corrupt(sc, cur, 0);
124 	return 0;
125 }
126 
127 /*
128  * Make sure that each inode of this part of an inobt record has the same
129  * sparse and free status as the finobt.
130  */
131 STATIC void
132 xchk_inobt_chunk_xref_finobt(
133 	struct xfs_scrub		*sc,
134 	struct xfs_inobt_rec_incore	*irec,
135 	xfs_agino_t			agino,
136 	unsigned int			nr_inodes)
137 {
138 	xfs_agino_t			i;
139 	unsigned int			rec_idx;
140 	int				error;
141 
142 	ASSERT(sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT);
143 
144 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
145 		return;
146 
147 	for (i = agino, rec_idx = agino - irec->ir_startino;
148 	     i < agino + nr_inodes;
149 	     i++, rec_idx++) {
150 		bool			free, hole;
151 		unsigned int		hole_idx;
152 
153 		free = irec->ir_free & (1ULL << rec_idx);
154 		hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
155 		hole = irec->ir_holemask & (1U << hole_idx);
156 
157 		error = xchk_inobt_xref_finobt(sc, irec, i, free, hole);
158 		if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur))
159 			return;
160 	}
161 }
162 
163 /*
164  * Does the inobt have a record for this inode with the same hole/free state?
165  * The inobt must always have a record if there's a finobt record.
166  */
167 STATIC int
168 xchk_finobt_xref_inobt(
169 	struct xfs_scrub	*sc,
170 	struct xfs_inobt_rec_incore *frec,
171 	xfs_agino_t		agino,
172 	bool			ffree,
173 	bool			fhole)
174 {
175 	struct xfs_inobt_rec_incore irec;
176 	struct xfs_btree_cur	*cur = sc->sa.ino_cur;
177 	bool			free, hole;
178 	unsigned int		rec_idx, hole_idx;
179 	int			has_record;
180 	int			error;
181 
182 	ASSERT(xfs_btree_is_ino(cur->bc_ops));
183 
184 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has_record);
185 	if (error)
186 		return error;
187 	if (!has_record)
188 		goto no_record;
189 
190 	error = xfs_inobt_get_rec(cur, &irec, &has_record);
191 	if (!has_record)
192 		return -EFSCORRUPTED;
193 
194 	if (irec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
195 		goto no_record;
196 
197 	/* There's an inobt record; free and hole status must match. */
198 	rec_idx = agino - irec.ir_startino;
199 	free = irec.ir_free & (1ULL << rec_idx);
200 	hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
201 	hole = irec.ir_holemask & (1U << hole_idx);
202 
203 	if (ffree != free)
204 		xchk_btree_xref_set_corrupt(sc, cur, 0);
205 	if (fhole != hole)
206 		xchk_btree_xref_set_corrupt(sc, cur, 0);
207 	return 0;
208 
209 no_record:
210 	/* finobt should never have a record for which the inobt does not */
211 	xchk_btree_xref_set_corrupt(sc, cur, 0);
212 	return 0;
213 }
214 
215 /*
216  * Make sure that each inode of this part of an finobt record has the same
217  * sparse and free status as the inobt.
218  */
219 STATIC void
220 xchk_finobt_chunk_xref_inobt(
221 	struct xfs_scrub		*sc,
222 	struct xfs_inobt_rec_incore	*frec,
223 	xfs_agino_t			agino,
224 	unsigned int			nr_inodes)
225 {
226 	xfs_agino_t			i;
227 	unsigned int			rec_idx;
228 	int				error;
229 
230 	ASSERT(sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT);
231 
232 	if (!sc->sa.ino_cur || xchk_skip_xref(sc->sm))
233 		return;
234 
235 	for (i = agino, rec_idx = agino - frec->ir_startino;
236 	     i < agino + nr_inodes;
237 	     i++, rec_idx++) {
238 		bool			ffree, fhole;
239 		unsigned int		hole_idx;
240 
241 		ffree = frec->ir_free & (1ULL << rec_idx);
242 		hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
243 		fhole = frec->ir_holemask & (1U << hole_idx);
244 
245 		error = xchk_finobt_xref_inobt(sc, frec, i, ffree, fhole);
246 		if (!xchk_should_check_xref(sc, &error, &sc->sa.ino_cur))
247 			return;
248 	}
249 }
250 
251 /* Is this chunk worth checking and cross-referencing? */
252 STATIC bool
253 xchk_iallocbt_chunk(
254 	struct xchk_btree		*bs,
255 	struct xfs_inobt_rec_incore	*irec,
256 	xfs_agino_t			agino,
257 	unsigned int			nr_inodes)
258 {
259 	struct xfs_scrub		*sc = bs->sc;
260 	struct xfs_mount		*mp = bs->cur->bc_mp;
261 	struct xfs_perag		*pag = bs->cur->bc_ag.pag;
262 	xfs_agblock_t			agbno;
263 	xfs_extlen_t			len;
264 
265 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
266 	len = XFS_B_TO_FSB(mp, nr_inodes * mp->m_sb.sb_inodesize);
267 
268 	if (!xfs_verify_agbext(pag, agbno, len))
269 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
270 
271 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
272 		return false;
273 
274 	xchk_xref_is_used_space(sc, agbno, len);
275 	if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
276 		xchk_inobt_chunk_xref_finobt(sc, irec, agino, nr_inodes);
277 	else
278 		xchk_finobt_chunk_xref_inobt(sc, irec, agino, nr_inodes);
279 	xchk_xref_is_only_owned_by(sc, agbno, len, &XFS_RMAP_OINFO_INODES);
280 	xchk_xref_is_not_shared(sc, agbno, len);
281 	xchk_xref_is_not_cow_staging(sc, agbno, len);
282 	return true;
283 }
284 
285 /*
286  * Check that an inode's allocation status matches ir_free in the inobt
287  * record.  First we try querying the in-core inode state, and if the inode
288  * isn't loaded we examine the on-disk inode directly.
289  *
290  * Since there can be 1:M and M:1 mappings between inobt records and inode
291  * clusters, we pass in the inode location information as an inobt record;
292  * the index of an inode cluster within the inobt record (as well as the
293  * cluster buffer itself); and the index of the inode within the cluster.
294  *
295  * @irec is the inobt record.
296  * @irec_ino is the inode offset from the start of the record.
297  * @dip is the on-disk inode.
298  */
299 STATIC int
300 xchk_iallocbt_check_cluster_ifree(
301 	struct xchk_btree		*bs,
302 	struct xfs_inobt_rec_incore	*irec,
303 	unsigned int			irec_ino,
304 	struct xfs_dinode		*dip)
305 {
306 	struct xfs_mount		*mp = bs->cur->bc_mp;
307 	xfs_ino_t			fsino;
308 	xfs_agino_t			agino;
309 	bool				irec_free;
310 	bool				ino_inuse;
311 	bool				freemask_ok;
312 	int				error = 0;
313 
314 	if (xchk_should_terminate(bs->sc, &error))
315 		return error;
316 
317 	/*
318 	 * Given an inobt record and the offset of an inode from the start of
319 	 * the record, compute which fs inode we're talking about.
320 	 */
321 	agino = irec->ir_startino + irec_ino;
322 	fsino = XFS_AGINO_TO_INO(mp, bs->cur->bc_ag.pag->pag_agno, agino);
323 	irec_free = (irec->ir_free & XFS_INOBT_MASK(irec_ino));
324 
325 	if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
326 	    (dip->di_version >= 3 && be64_to_cpu(dip->di_ino) != fsino)) {
327 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
328 		goto out;
329 	}
330 
331 	error = xchk_inode_is_allocated(bs->sc, agino, &ino_inuse);
332 	if (error == -ENODATA) {
333 		/* Not cached, just read the disk buffer */
334 		freemask_ok = irec_free ^ !!(dip->di_mode);
335 		if (!(bs->sc->flags & XCHK_TRY_HARDER) && !freemask_ok)
336 			return -EDEADLOCK;
337 	} else if (error < 0) {
338 		/*
339 		 * Inode is only half assembled, or there was an IO error,
340 		 * or the verifier failed, so don't bother trying to check.
341 		 * The inode scrubber can deal with this.
342 		 */
343 		goto out;
344 	} else {
345 		/* Inode is all there. */
346 		freemask_ok = irec_free ^ ino_inuse;
347 	}
348 	if (!freemask_ok)
349 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
350 out:
351 	return 0;
352 }
353 
354 /*
355  * Check that the holemask and freemask of a hypothetical inode cluster match
356  * what's actually on disk.  If sparse inodes are enabled, the cluster does
357  * not actually have to map to inodes if the corresponding holemask bit is set.
358  *
359  * @cluster_base is the first inode in the cluster within the @irec.
360  */
361 STATIC int
362 xchk_iallocbt_check_cluster(
363 	struct xchk_btree		*bs,
364 	struct xfs_inobt_rec_incore	*irec,
365 	unsigned int			cluster_base)
366 {
367 	struct xfs_imap			imap;
368 	struct xfs_mount		*mp = bs->cur->bc_mp;
369 	struct xfs_buf			*cluster_bp;
370 	unsigned int			nr_inodes;
371 	xfs_agnumber_t			agno = bs->cur->bc_ag.pag->pag_agno;
372 	xfs_agblock_t			agbno;
373 	unsigned int			cluster_index;
374 	uint16_t			cluster_mask = 0;
375 	uint16_t			ir_holemask;
376 	int				error = 0;
377 
378 	nr_inodes = min_t(unsigned int, XFS_INODES_PER_CHUNK,
379 			M_IGEO(mp)->inodes_per_cluster);
380 
381 	/* Map this inode cluster */
382 	agbno = XFS_AGINO_TO_AGBNO(mp, irec->ir_startino + cluster_base);
383 
384 	/* Compute a bitmask for this cluster that can be used for holemask. */
385 	for (cluster_index = 0;
386 	     cluster_index < nr_inodes;
387 	     cluster_index += XFS_INODES_PER_HOLEMASK_BIT)
388 		cluster_mask |= XFS_INOBT_MASK((cluster_base + cluster_index) /
389 				XFS_INODES_PER_HOLEMASK_BIT);
390 
391 	/*
392 	 * Map the first inode of this cluster to a buffer and offset.
393 	 * Be careful about inobt records that don't align with the start of
394 	 * the inode buffer when block sizes are large enough to hold multiple
395 	 * inode chunks.  When this happens, cluster_base will be zero but
396 	 * ir_startino can be large enough to make im_boffset nonzero.
397 	 */
398 	ir_holemask = (irec->ir_holemask & cluster_mask);
399 	imap.im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
400 	imap.im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
401 	imap.im_boffset = XFS_INO_TO_OFFSET(mp, irec->ir_startino) <<
402 			mp->m_sb.sb_inodelog;
403 
404 	if (imap.im_boffset != 0 && cluster_base != 0) {
405 		ASSERT(imap.im_boffset == 0 || cluster_base == 0);
406 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
407 		return 0;
408 	}
409 
410 	trace_xchk_iallocbt_check_cluster(mp, agno, irec->ir_startino,
411 			imap.im_blkno, imap.im_len, cluster_base, nr_inodes,
412 			cluster_mask, ir_holemask,
413 			XFS_INO_TO_OFFSET(mp, irec->ir_startino +
414 					  cluster_base));
415 
416 	/* The whole cluster must be a hole or not a hole. */
417 	if (ir_holemask != cluster_mask && ir_holemask != 0) {
418 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
419 		return 0;
420 	}
421 
422 	/* If any part of this is a hole, skip it. */
423 	if (ir_holemask) {
424 		xchk_xref_is_not_owned_by(bs->sc, agbno,
425 				M_IGEO(mp)->blocks_per_cluster,
426 				&XFS_RMAP_OINFO_INODES);
427 		return 0;
428 	}
429 
430 	xchk_xref_is_only_owned_by(bs->sc, agbno, M_IGEO(mp)->blocks_per_cluster,
431 			&XFS_RMAP_OINFO_INODES);
432 
433 	/* Grab the inode cluster buffer. */
434 	error = xfs_imap_to_bp(mp, bs->cur->bc_tp, &imap, &cluster_bp);
435 	if (!xchk_btree_xref_process_error(bs->sc, bs->cur, 0, &error))
436 		return error;
437 
438 	/* Check free status of each inode within this cluster. */
439 	for (cluster_index = 0; cluster_index < nr_inodes; cluster_index++) {
440 		struct xfs_dinode	*dip;
441 
442 		if (imap.im_boffset >= BBTOB(cluster_bp->b_length)) {
443 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
444 			break;
445 		}
446 
447 		dip = xfs_buf_offset(cluster_bp, imap.im_boffset);
448 		error = xchk_iallocbt_check_cluster_ifree(bs, irec,
449 				cluster_base + cluster_index, dip);
450 		if (error)
451 			break;
452 		imap.im_boffset += mp->m_sb.sb_inodesize;
453 	}
454 
455 	xfs_trans_brelse(bs->cur->bc_tp, cluster_bp);
456 	return error;
457 }
458 
459 /*
460  * For all the inode clusters that could map to this inobt record, make sure
461  * that the holemask makes sense and that the allocation status of each inode
462  * matches the freemask.
463  */
464 STATIC int
465 xchk_iallocbt_check_clusters(
466 	struct xchk_btree		*bs,
467 	struct xfs_inobt_rec_incore	*irec)
468 {
469 	unsigned int			cluster_base;
470 	int				error = 0;
471 
472 	/*
473 	 * For the common case where this inobt record maps to multiple inode
474 	 * clusters this will call _check_cluster for each cluster.
475 	 *
476 	 * For the case that multiple inobt records map to a single cluster,
477 	 * this will call _check_cluster once.
478 	 */
479 	for (cluster_base = 0;
480 	     cluster_base < XFS_INODES_PER_CHUNK;
481 	     cluster_base += M_IGEO(bs->sc->mp)->inodes_per_cluster) {
482 		error = xchk_iallocbt_check_cluster(bs, irec, cluster_base);
483 		if (error)
484 			break;
485 	}
486 
487 	return error;
488 }
489 
490 /*
491  * Make sure this inode btree record is aligned properly.  Because a fs block
492  * contains multiple inodes, we check that the inobt record is aligned to the
493  * correct inode, not just the correct block on disk.  This results in a finer
494  * grained corruption check.
495  */
496 STATIC void
497 xchk_iallocbt_rec_alignment(
498 	struct xchk_btree		*bs,
499 	struct xfs_inobt_rec_incore	*irec)
500 {
501 	struct xfs_mount		*mp = bs->sc->mp;
502 	struct xchk_iallocbt		*iabt = bs->private;
503 	struct xfs_ino_geometry		*igeo = M_IGEO(mp);
504 
505 	/*
506 	 * finobt records have different positioning requirements than inobt
507 	 * records: each finobt record must have a corresponding inobt record.
508 	 * That is checked in the xref function, so for now we only catch the
509 	 * obvious case where the record isn't at all aligned properly.
510 	 *
511 	 * Note that if a fs block contains more than a single chunk of inodes,
512 	 * we will have finobt records only for those chunks containing free
513 	 * inodes, and therefore expect chunk alignment of finobt records.
514 	 * Otherwise, we expect that the finobt record is aligned to the
515 	 * cluster alignment as told by the superblock.
516 	 */
517 	if (xfs_btree_is_fino(bs->cur->bc_ops)) {
518 		unsigned int	imask;
519 
520 		imask = min_t(unsigned int, XFS_INODES_PER_CHUNK,
521 				igeo->cluster_align_inodes) - 1;
522 		if (irec->ir_startino & imask)
523 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
524 		return;
525 	}
526 
527 	if (iabt->next_startino != NULLAGINO) {
528 		/*
529 		 * We're midway through a cluster of inodes that is mapped by
530 		 * multiple inobt records.  Did we get the record for the next
531 		 * irec in the sequence?
532 		 */
533 		if (irec->ir_startino != iabt->next_startino) {
534 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
535 			return;
536 		}
537 
538 		iabt->next_startino += XFS_INODES_PER_CHUNK;
539 
540 		/* Are we done with the cluster? */
541 		if (iabt->next_startino >= iabt->next_cluster_ino) {
542 			iabt->next_startino = NULLAGINO;
543 			iabt->next_cluster_ino = NULLAGINO;
544 		}
545 		return;
546 	}
547 
548 	/* inobt records must be aligned to cluster and inoalignmnt size. */
549 	if (irec->ir_startino & (igeo->cluster_align_inodes - 1)) {
550 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
551 		return;
552 	}
553 
554 	if (irec->ir_startino & (igeo->inodes_per_cluster - 1)) {
555 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
556 		return;
557 	}
558 
559 	if (igeo->inodes_per_cluster <= XFS_INODES_PER_CHUNK)
560 		return;
561 
562 	/*
563 	 * If this is the start of an inode cluster that can be mapped by
564 	 * multiple inobt records, the next inobt record must follow exactly
565 	 * after this one.
566 	 */
567 	iabt->next_startino = irec->ir_startino + XFS_INODES_PER_CHUNK;
568 	iabt->next_cluster_ino = irec->ir_startino + igeo->inodes_per_cluster;
569 }
570 
571 /* Scrub an inobt/finobt record. */
572 STATIC int
573 xchk_iallocbt_rec(
574 	struct xchk_btree		*bs,
575 	const union xfs_btree_rec	*rec)
576 {
577 	struct xfs_mount		*mp = bs->cur->bc_mp;
578 	struct xchk_iallocbt		*iabt = bs->private;
579 	struct xfs_inobt_rec_incore	irec;
580 	uint64_t			holes;
581 	xfs_agino_t			agino;
582 	int				holecount;
583 	int				i;
584 	int				error = 0;
585 	uint16_t			holemask;
586 
587 	xfs_inobt_btrec_to_irec(mp, rec, &irec);
588 	if (xfs_inobt_check_irec(bs->cur->bc_ag.pag, &irec) != NULL) {
589 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
590 		return 0;
591 	}
592 
593 	agino = irec.ir_startino;
594 
595 	xchk_iallocbt_rec_alignment(bs, &irec);
596 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
597 		goto out;
598 
599 	iabt->inodes += irec.ir_count;
600 
601 	/* Handle non-sparse inodes */
602 	if (!xfs_inobt_issparse(irec.ir_holemask)) {
603 		if (irec.ir_count != XFS_INODES_PER_CHUNK)
604 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
605 
606 		if (!xchk_iallocbt_chunk(bs, &irec, agino,
607 					XFS_INODES_PER_CHUNK))
608 			goto out;
609 		goto check_clusters;
610 	}
611 
612 	/* Check each chunk of a sparse inode cluster. */
613 	holemask = irec.ir_holemask;
614 	holecount = 0;
615 	holes = ~xfs_inobt_irec_to_allocmask(&irec);
616 	if ((holes & irec.ir_free) != holes ||
617 	    irec.ir_freecount > irec.ir_count)
618 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
619 
620 	for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
621 		if (holemask & 1)
622 			holecount += XFS_INODES_PER_HOLEMASK_BIT;
623 		else if (!xchk_iallocbt_chunk(bs, &irec, agino,
624 					XFS_INODES_PER_HOLEMASK_BIT))
625 			goto out;
626 		holemask >>= 1;
627 		agino += XFS_INODES_PER_HOLEMASK_BIT;
628 	}
629 
630 	if (holecount > XFS_INODES_PER_CHUNK ||
631 	    holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
632 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
633 
634 check_clusters:
635 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
636 		goto out;
637 
638 	error = xchk_iallocbt_check_clusters(bs, &irec);
639 	if (error)
640 		goto out;
641 
642 out:
643 	return error;
644 }
645 
646 /*
647  * Make sure the inode btrees are as large as the rmap thinks they are.
648  * Don't bother if we're missing btree cursors, as we're already corrupt.
649  */
650 STATIC void
651 xchk_iallocbt_xref_rmap_btreeblks(
652 	struct xfs_scrub	*sc)
653 {
654 	xfs_filblks_t		blocks;
655 	xfs_extlen_t		inobt_blocks = 0;
656 	xfs_extlen_t		finobt_blocks = 0;
657 	int			error;
658 
659 	if (!sc->sa.ino_cur || !sc->sa.rmap_cur ||
660 	    (xfs_has_finobt(sc->mp) && !sc->sa.fino_cur) ||
661 	    xchk_skip_xref(sc->sm))
662 		return;
663 
664 	/* Check that we saw as many inobt blocks as the rmap says. */
665 	error = xfs_btree_count_blocks(sc->sa.ino_cur, &inobt_blocks);
666 	if (!xchk_process_error(sc, 0, 0, &error))
667 		return;
668 
669 	if (sc->sa.fino_cur) {
670 		error = xfs_btree_count_blocks(sc->sa.fino_cur, &finobt_blocks);
671 		if (!xchk_process_error(sc, 0, 0, &error))
672 			return;
673 	}
674 
675 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
676 			&XFS_RMAP_OINFO_INOBT, &blocks);
677 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
678 		return;
679 	if (blocks != inobt_blocks + finobt_blocks)
680 		xchk_btree_set_corrupt(sc, sc->sa.ino_cur, 0);
681 }
682 
683 /*
684  * Make sure that the inobt records point to the same number of blocks as
685  * the rmap says are owned by inodes.
686  */
687 STATIC void
688 xchk_iallocbt_xref_rmap_inodes(
689 	struct xfs_scrub	*sc,
690 	unsigned long long	inodes)
691 {
692 	xfs_filblks_t		blocks;
693 	xfs_filblks_t		inode_blocks;
694 	int			error;
695 
696 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
697 		return;
698 
699 	/* Check that we saw as many inode blocks as the rmap knows about. */
700 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
701 			&XFS_RMAP_OINFO_INODES, &blocks);
702 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
703 		return;
704 	inode_blocks = XFS_B_TO_FSB(sc->mp, inodes * sc->mp->m_sb.sb_inodesize);
705 	if (blocks != inode_blocks)
706 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
707 }
708 
709 /* Scrub one of the inode btrees for some AG. */
710 int
711 xchk_iallocbt(
712 	struct xfs_scrub	*sc)
713 {
714 	struct xfs_btree_cur	*cur;
715 	struct xchk_iallocbt	iabt = {
716 		.inodes		= 0,
717 		.next_startino	= NULLAGINO,
718 		.next_cluster_ino = NULLAGINO,
719 	};
720 	int			error;
721 
722 	switch (sc->sm->sm_type) {
723 	case XFS_SCRUB_TYPE_INOBT:
724 		cur = sc->sa.ino_cur;
725 		break;
726 	case XFS_SCRUB_TYPE_FINOBT:
727 		cur = sc->sa.fino_cur;
728 		break;
729 	default:
730 		ASSERT(0);
731 		return -EIO;
732 	}
733 
734 	error = xchk_btree(sc, cur, xchk_iallocbt_rec, &XFS_RMAP_OINFO_INOBT,
735 			&iabt);
736 	if (error)
737 		return error;
738 
739 	xchk_iallocbt_xref_rmap_btreeblks(sc);
740 
741 	/*
742 	 * If we're scrubbing the inode btree, inode_blocks is the number of
743 	 * blocks pointed to by all the inode chunk records.  Therefore, we
744 	 * should compare to the number of inode chunk blocks that the rmap
745 	 * knows about.  We can't do this for the finobt since it only points
746 	 * to inode chunks with free inodes.
747 	 */
748 	if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
749 		xchk_iallocbt_xref_rmap_inodes(sc, iabt.inodes);
750 	return error;
751 }
752 
753 /* See if an inode btree has (or doesn't have) an inode chunk record. */
754 static inline void
755 xchk_xref_inode_check(
756 	struct xfs_scrub	*sc,
757 	xfs_agblock_t		agbno,
758 	xfs_extlen_t		len,
759 	struct xfs_btree_cur	**icur,
760 	enum xbtree_recpacking	expected)
761 {
762 	enum xbtree_recpacking	outcome;
763 	int			error;
764 
765 	if (!(*icur) || xchk_skip_xref(sc->sm))
766 		return;
767 
768 	error = xfs_ialloc_has_inodes_at_extent(*icur, agbno, len, &outcome);
769 	if (!xchk_should_check_xref(sc, &error, icur))
770 		return;
771 	if (outcome != expected)
772 		xchk_btree_xref_set_corrupt(sc, *icur, 0);
773 }
774 
775 /* xref check that the extent is not covered by inodes */
776 void
777 xchk_xref_is_not_inode_chunk(
778 	struct xfs_scrub	*sc,
779 	xfs_agblock_t		agbno,
780 	xfs_extlen_t		len)
781 {
782 	xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur,
783 			XBTREE_RECPACKING_EMPTY);
784 	xchk_xref_inode_check(sc, agbno, len, &sc->sa.fino_cur,
785 			XBTREE_RECPACKING_EMPTY);
786 }
787 
788 /* xref check that the extent is covered by inodes */
789 void
790 xchk_xref_is_inode_chunk(
791 	struct xfs_scrub	*sc,
792 	xfs_agblock_t		agbno,
793 	xfs_extlen_t		len)
794 {
795 	xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur,
796 			XBTREE_RECPACKING_FULL);
797 }
798