xref: /linux/fs/xfs/scrub/ialloc.c (revision 546b928da0427b0d6c663cbb992bd7bfa9ac7971)
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_platform.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
xchk_setup_ag_iallocbt(struct xfs_scrub * sc)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
xchk_inobt_xref_finobt(struct xfs_scrub * sc,struct xfs_inobt_rec_incore * irec,xfs_agino_t agino,bool free,bool hole)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 (error)
89 		return error;
90 	if (!has_record)
91 		return -EFSCORRUPTED;
92 
93 	if (frec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
94 		goto no_record;
95 
96 	/* There's a finobt record; free and hole status must match. */
97 	frec_idx = agino - frec.ir_startino;
98 	ffree = frec.ir_free & (1ULL << frec_idx);
99 	fhole_idx = frec_idx / XFS_INODES_PER_HOLEMASK_BIT;
100 	fhole = frec.ir_holemask & (1U << fhole_idx);
101 
102 	if (ffree != free)
103 		xchk_btree_xref_set_corrupt(sc, cur, 0);
104 	if (fhole != hole)
105 		xchk_btree_xref_set_corrupt(sc, cur, 0);
106 	return 0;
107 
108 no_record:
109 	/* inobt record is fully allocated */
110 	if (irec->ir_free == 0)
111 		return 0;
112 
113 	/* inobt record is totally unallocated */
114 	if (irec->ir_free == XFS_INOBT_ALL_FREE)
115 		return 0;
116 
117 	/* inobt record says this is a hole */
118 	if (hole)
119 		return 0;
120 
121 	/* finobt doesn't care about allocated inodes */
122 	if (!free)
123 		return 0;
124 
125 	xchk_btree_xref_set_corrupt(sc, cur, 0);
126 	return 0;
127 }
128 
129 /*
130  * Make sure that each inode of this part of an inobt record has the same
131  * sparse and free status as the finobt.
132  */
133 STATIC void
xchk_inobt_chunk_xref_finobt(struct xfs_scrub * sc,struct xfs_inobt_rec_incore * irec,xfs_agino_t agino,unsigned int nr_inodes)134 xchk_inobt_chunk_xref_finobt(
135 	struct xfs_scrub		*sc,
136 	struct xfs_inobt_rec_incore	*irec,
137 	xfs_agino_t			agino,
138 	unsigned int			nr_inodes)
139 {
140 	xfs_agino_t			i;
141 	unsigned int			rec_idx;
142 	int				error;
143 
144 	ASSERT(sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT);
145 
146 	if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
147 		return;
148 
149 	for (i = agino, rec_idx = agino - irec->ir_startino;
150 	     i < agino + nr_inodes;
151 	     i++, rec_idx++) {
152 		bool			free, hole;
153 		unsigned int		hole_idx;
154 
155 		free = irec->ir_free & (1ULL << rec_idx);
156 		hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
157 		hole = irec->ir_holemask & (1U << hole_idx);
158 
159 		error = xchk_inobt_xref_finobt(sc, irec, i, free, hole);
160 		if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur))
161 			return;
162 	}
163 }
164 
165 /*
166  * Does the inobt have a record for this inode with the same hole/free state?
167  * The inobt must always have a record if there's a finobt record.
168  */
169 STATIC int
xchk_finobt_xref_inobt(struct xfs_scrub * sc,struct xfs_inobt_rec_incore * frec,xfs_agino_t agino,bool ffree,bool fhole)170 xchk_finobt_xref_inobt(
171 	struct xfs_scrub	*sc,
172 	struct xfs_inobt_rec_incore *frec,
173 	xfs_agino_t		agino,
174 	bool			ffree,
175 	bool			fhole)
176 {
177 	struct xfs_inobt_rec_incore irec;
178 	struct xfs_btree_cur	*cur = sc->sa.ino_cur;
179 	bool			free, hole;
180 	unsigned int		rec_idx, hole_idx;
181 	int			has_record;
182 	int			error;
183 
184 	ASSERT(xfs_btree_is_ino(cur->bc_ops));
185 
186 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has_record);
187 	if (error)
188 		return error;
189 	if (!has_record)
190 		goto no_record;
191 
192 	error = xfs_inobt_get_rec(cur, &irec, &has_record);
193 	if (error)
194 		return error;
195 	if (!has_record)
196 		return -EFSCORRUPTED;
197 
198 	if (irec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
199 		goto no_record;
200 
201 	/* There's an inobt record; free and hole status must match. */
202 	rec_idx = agino - irec.ir_startino;
203 	free = irec.ir_free & (1ULL << rec_idx);
204 	hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
205 	hole = irec.ir_holemask & (1U << hole_idx);
206 
207 	if (ffree != free)
208 		xchk_btree_xref_set_corrupt(sc, cur, 0);
209 	if (fhole != hole)
210 		xchk_btree_xref_set_corrupt(sc, cur, 0);
211 	return 0;
212 
213 no_record:
214 	/* finobt should never have a record for which the inobt does not */
215 	xchk_btree_xref_set_corrupt(sc, cur, 0);
216 	return 0;
217 }
218 
219 /*
220  * Make sure that each inode of this part of an finobt record has the same
221  * sparse and free status as the inobt.
222  */
223 STATIC void
xchk_finobt_chunk_xref_inobt(struct xfs_scrub * sc,struct xfs_inobt_rec_incore * frec,xfs_agino_t agino,unsigned int nr_inodes)224 xchk_finobt_chunk_xref_inobt(
225 	struct xfs_scrub		*sc,
226 	struct xfs_inobt_rec_incore	*frec,
227 	xfs_agino_t			agino,
228 	unsigned int			nr_inodes)
229 {
230 	xfs_agino_t			i;
231 	unsigned int			rec_idx;
232 	int				error;
233 
234 	ASSERT(sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT);
235 
236 	if (!sc->sa.ino_cur || xchk_skip_xref(sc->sm))
237 		return;
238 
239 	for (i = agino, rec_idx = agino - frec->ir_startino;
240 	     i < agino + nr_inodes;
241 	     i++, rec_idx++) {
242 		bool			ffree, fhole;
243 		unsigned int		hole_idx;
244 
245 		ffree = frec->ir_free & (1ULL << rec_idx);
246 		hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
247 		fhole = frec->ir_holemask & (1U << hole_idx);
248 
249 		error = xchk_finobt_xref_inobt(sc, frec, i, ffree, fhole);
250 		if (!xchk_should_check_xref(sc, &error, &sc->sa.ino_cur))
251 			return;
252 	}
253 }
254 
255 /* Is this chunk worth checking and cross-referencing? */
256 STATIC bool
xchk_iallocbt_chunk(struct xchk_btree * bs,struct xfs_inobt_rec_incore * irec,xfs_agino_t agino,unsigned int nr_inodes)257 xchk_iallocbt_chunk(
258 	struct xchk_btree		*bs,
259 	struct xfs_inobt_rec_incore	*irec,
260 	xfs_agino_t			agino,
261 	unsigned int			nr_inodes)
262 {
263 	struct xfs_scrub		*sc = bs->sc;
264 	struct xfs_mount		*mp = bs->cur->bc_mp;
265 	struct xfs_perag		*pag = to_perag(bs->cur->bc_group);
266 	xfs_agblock_t			agbno;
267 	xfs_extlen_t			len;
268 
269 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
270 	len = XFS_B_TO_FSB(mp, nr_inodes * mp->m_sb.sb_inodesize);
271 
272 	if (!xfs_verify_agbext(pag, agbno, len))
273 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
274 
275 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
276 		return false;
277 
278 	xchk_xref_is_used_space(sc, agbno, len);
279 	if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
280 		xchk_inobt_chunk_xref_finobt(sc, irec, agino, nr_inodes);
281 	else
282 		xchk_finobt_chunk_xref_inobt(sc, irec, agino, nr_inodes);
283 	xchk_xref_is_only_owned_by(sc, agbno, len, &XFS_RMAP_OINFO_INODES);
284 	xchk_xref_is_not_shared(sc, agbno, len);
285 	xchk_xref_is_not_cow_staging(sc, agbno, len);
286 	return true;
287 }
288 
289 /*
290  * Check that an inode's allocation status matches ir_free in the inobt
291  * record.  First we try querying the in-core inode state, and if the inode
292  * isn't loaded we examine the on-disk inode directly.
293  *
294  * Since there can be 1:M and M:1 mappings between inobt records and inode
295  * clusters, we pass in the inode location information as an inobt record;
296  * the index of an inode cluster within the inobt record (as well as the
297  * cluster buffer itself); and the index of the inode within the cluster.
298  *
299  * @irec is the inobt record.
300  * @irec_ino is the inode offset from the start of the record.
301  * @dip is the on-disk inode.
302  */
303 STATIC int
xchk_iallocbt_check_cluster_ifree(struct xchk_btree * bs,struct xfs_inobt_rec_incore * irec,unsigned int irec_ino,struct xfs_dinode * dip)304 xchk_iallocbt_check_cluster_ifree(
305 	struct xchk_btree		*bs,
306 	struct xfs_inobt_rec_incore	*irec,
307 	unsigned int			irec_ino,
308 	struct xfs_dinode		*dip)
309 {
310 	xfs_ino_t			fsino;
311 	xfs_agino_t			agino;
312 	bool				irec_free;
313 	bool				ino_inuse;
314 	bool				freemask_ok;
315 	int				error = 0;
316 
317 	if (xchk_should_terminate(bs->sc, &error))
318 		return error;
319 
320 	/*
321 	 * Given an inobt record and the offset of an inode from the start of
322 	 * the record, compute which fs inode we're talking about.
323 	 */
324 	agino = irec->ir_startino + irec_ino;
325 	fsino = xfs_agino_to_ino(to_perag(bs->cur->bc_group), agino);
326 	irec_free = (irec->ir_free & XFS_INOBT_MASK(irec_ino));
327 
328 	if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
329 	    (dip->di_version >= 3 && be64_to_cpu(dip->di_ino) != fsino)) {
330 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
331 		goto out;
332 	}
333 
334 	error = xchk_inode_is_allocated(bs->sc, agino, &ino_inuse);
335 	if (error == -ENODATA) {
336 		/* Not cached, just read the disk buffer */
337 		freemask_ok = irec_free ^ !!(dip->di_mode);
338 		if (!(bs->sc->flags & XCHK_TRY_HARDER) && !freemask_ok)
339 			return -EDEADLOCK;
340 	} else if (error < 0) {
341 		/*
342 		 * Inode is only half assembled, or there was an IO error,
343 		 * or the verifier failed, so don't bother trying to check.
344 		 * The inode scrubber can deal with this.
345 		 */
346 		goto out;
347 	} else {
348 		/* Inode is all there. */
349 		freemask_ok = irec_free ^ ino_inuse;
350 	}
351 	if (!freemask_ok)
352 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
353 out:
354 	return 0;
355 }
356 
357 /*
358  * Check that the holemask and freemask of a hypothetical inode cluster match
359  * what's actually on disk.  If sparse inodes are enabled, the cluster does
360  * not actually have to map to inodes if the corresponding holemask bit is set.
361  *
362  * @cluster_base is the first inode in the cluster within the @irec.
363  */
364 STATIC int
xchk_iallocbt_check_cluster(struct xchk_btree * bs,struct xfs_inobt_rec_incore * irec,unsigned int cluster_base)365 xchk_iallocbt_check_cluster(
366 	struct xchk_btree		*bs,
367 	struct xfs_inobt_rec_incore	*irec,
368 	unsigned int			cluster_base)
369 {
370 	struct xfs_perag		*pag = to_perag(bs->cur->bc_group);
371 	struct xfs_imap			imap;
372 	struct xfs_mount		*mp = bs->cur->bc_mp;
373 	struct xfs_buf			*cluster_bp;
374 	unsigned int			nr_inodes;
375 	xfs_agblock_t			agbno;
376 	unsigned int			cluster_index;
377 	uint16_t			cluster_mask = 0;
378 	uint16_t			ir_holemask;
379 	int				error = 0;
380 
381 	nr_inodes = min_t(unsigned int, XFS_INODES_PER_CHUNK,
382 			M_IGEO(mp)->inodes_per_cluster);
383 
384 	/* Map this inode cluster */
385 	agbno = XFS_AGINO_TO_AGBNO(mp, irec->ir_startino + cluster_base);
386 
387 	/* Compute a bitmask for this cluster that can be used for holemask. */
388 	for (cluster_index = 0;
389 	     cluster_index < nr_inodes;
390 	     cluster_index += XFS_INODES_PER_HOLEMASK_BIT)
391 		cluster_mask |= XFS_INOBT_MASK((cluster_base + cluster_index) /
392 				XFS_INODES_PER_HOLEMASK_BIT);
393 
394 	/*
395 	 * Map the first inode of this cluster to a buffer and offset.
396 	 * Be careful about inobt records that don't align with the start of
397 	 * the inode buffer when block sizes are large enough to hold multiple
398 	 * inode chunks.  When this happens, cluster_base will be zero but
399 	 * ir_startino can be large enough to make im_boffset nonzero.
400 	 */
401 	ir_holemask = (irec->ir_holemask & cluster_mask);
402 	imap.im_agbno = agbno;
403 	imap.im_boffset = XFS_INO_TO_OFFSET(mp, irec->ir_startino) <<
404 			mp->m_sb.sb_inodelog;
405 
406 	if (imap.im_boffset != 0 && cluster_base != 0) {
407 		ASSERT(imap.im_boffset == 0 || cluster_base == 0);
408 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
409 		return 0;
410 	}
411 
412 	trace_xchk_iallocbt_check_cluster(pag, irec->ir_startino, imap.im_agbno,
413 			XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster),
414 			cluster_base, nr_inodes, cluster_mask, ir_holemask,
415 			XFS_INO_TO_OFFSET(mp, irec->ir_startino +
416 					  cluster_base));
417 
418 	/* The whole cluster must be a hole or not a hole. */
419 	if (ir_holemask != cluster_mask && ir_holemask != 0) {
420 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
421 		return 0;
422 	}
423 
424 	/* If any part of this is a hole, skip it. */
425 	if (ir_holemask) {
426 		xchk_xref_is_not_owned_by(bs->sc, agbno,
427 				M_IGEO(mp)->blocks_per_cluster,
428 				&XFS_RMAP_OINFO_INODES);
429 		return 0;
430 	}
431 
432 	xchk_xref_is_only_owned_by(bs->sc, agbno, M_IGEO(mp)->blocks_per_cluster,
433 			&XFS_RMAP_OINFO_INODES);
434 
435 	/* Grab the inode cluster buffer. */
436 	error = xfs_read_icluster(pag, bs->cur->bc_tp, imap.im_agbno,
437 			&cluster_bp);
438 	if (!xchk_btree_xref_process_error(bs->sc, bs->cur, 0, &error))
439 		return error;
440 
441 	/* Check free status of each inode within this cluster. */
442 	for (cluster_index = 0; cluster_index < nr_inodes; cluster_index++) {
443 		struct xfs_dinode	*dip;
444 
445 		if (imap.im_boffset >= BBTOB(cluster_bp->b_length)) {
446 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
447 			break;
448 		}
449 
450 		dip = xfs_buf_offset(cluster_bp, imap.im_boffset);
451 		error = xchk_iallocbt_check_cluster_ifree(bs, irec,
452 				cluster_base + cluster_index, dip);
453 		if (error)
454 			break;
455 		imap.im_boffset += mp->m_sb.sb_inodesize;
456 	}
457 
458 	xfs_trans_brelse(bs->cur->bc_tp, cluster_bp);
459 	return error;
460 }
461 
462 /*
463  * For all the inode clusters that could map to this inobt record, make sure
464  * that the holemask makes sense and that the allocation status of each inode
465  * matches the freemask.
466  */
467 STATIC int
xchk_iallocbt_check_clusters(struct xchk_btree * bs,struct xfs_inobt_rec_incore * irec)468 xchk_iallocbt_check_clusters(
469 	struct xchk_btree		*bs,
470 	struct xfs_inobt_rec_incore	*irec)
471 {
472 	unsigned int			cluster_base;
473 	int				error = 0;
474 
475 	/*
476 	 * For the common case where this inobt record maps to multiple inode
477 	 * clusters this will call _check_cluster for each cluster.
478 	 *
479 	 * For the case that multiple inobt records map to a single cluster,
480 	 * this will call _check_cluster once.
481 	 */
482 	for (cluster_base = 0;
483 	     cluster_base < XFS_INODES_PER_CHUNK;
484 	     cluster_base += M_IGEO(bs->sc->mp)->inodes_per_cluster) {
485 		error = xchk_iallocbt_check_cluster(bs, irec, cluster_base);
486 		if (error)
487 			break;
488 	}
489 
490 	return error;
491 }
492 
493 /*
494  * Make sure this inode btree record is aligned properly.  Because a fs block
495  * contains multiple inodes, we check that the inobt record is aligned to the
496  * correct inode, not just the correct block on disk.  This results in a finer
497  * grained corruption check.
498  */
499 STATIC void
xchk_iallocbt_rec_alignment(struct xchk_btree * bs,struct xfs_inobt_rec_incore * irec)500 xchk_iallocbt_rec_alignment(
501 	struct xchk_btree		*bs,
502 	struct xfs_inobt_rec_incore	*irec)
503 {
504 	struct xfs_mount		*mp = bs->sc->mp;
505 	struct xchk_iallocbt		*iabt = bs->private;
506 	struct xfs_ino_geometry		*igeo = M_IGEO(mp);
507 
508 	/*
509 	 * finobt records have different positioning requirements than inobt
510 	 * records: each finobt record must have a corresponding inobt record.
511 	 * That is checked in the xref function, so for now we only catch the
512 	 * obvious case where the record isn't at all aligned properly.
513 	 *
514 	 * Note that if a fs block contains more than a single chunk of inodes,
515 	 * we will have finobt records only for those chunks containing free
516 	 * inodes, and therefore expect chunk alignment of finobt records.
517 	 * Otherwise, we expect that the finobt record is aligned to the
518 	 * cluster alignment as told by the superblock.
519 	 */
520 	if (xfs_btree_is_fino(bs->cur->bc_ops)) {
521 		unsigned int	imask;
522 
523 		imask = min_t(unsigned int, XFS_INODES_PER_CHUNK,
524 				igeo->cluster_align_inodes) - 1;
525 		if (irec->ir_startino & imask)
526 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
527 		return;
528 	}
529 
530 	if (iabt->next_startino != NULLAGINO) {
531 		/*
532 		 * We're midway through a cluster of inodes that is mapped by
533 		 * multiple inobt records.  Did we get the record for the next
534 		 * irec in the sequence?
535 		 */
536 		if (irec->ir_startino != iabt->next_startino) {
537 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
538 			return;
539 		}
540 
541 		iabt->next_startino += XFS_INODES_PER_CHUNK;
542 
543 		/* Are we done with the cluster? */
544 		if (iabt->next_startino >= iabt->next_cluster_ino) {
545 			iabt->next_startino = NULLAGINO;
546 			iabt->next_cluster_ino = NULLAGINO;
547 		}
548 		return;
549 	}
550 
551 	/* inobt records must be aligned to cluster and inoalignmnt size. */
552 	if (irec->ir_startino & (igeo->cluster_align_inodes - 1)) {
553 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
554 		return;
555 	}
556 
557 	if (irec->ir_startino & (igeo->inodes_per_cluster - 1)) {
558 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
559 		return;
560 	}
561 
562 	if (igeo->inodes_per_cluster <= XFS_INODES_PER_CHUNK)
563 		return;
564 
565 	/*
566 	 * If this is the start of an inode cluster that can be mapped by
567 	 * multiple inobt records, the next inobt record must follow exactly
568 	 * after this one.
569 	 */
570 	iabt->next_startino = irec->ir_startino + XFS_INODES_PER_CHUNK;
571 	iabt->next_cluster_ino = irec->ir_startino + igeo->inodes_per_cluster;
572 }
573 
574 /* Scrub an inobt/finobt record. */
575 STATIC int
xchk_iallocbt_rec(struct xchk_btree * bs,const union xfs_btree_rec * rec)576 xchk_iallocbt_rec(
577 	struct xchk_btree		*bs,
578 	const union xfs_btree_rec	*rec)
579 {
580 	struct xfs_mount		*mp = bs->cur->bc_mp;
581 	struct xchk_iallocbt		*iabt = bs->private;
582 	struct xfs_inobt_rec_incore	irec;
583 	uint64_t			holes;
584 	xfs_agino_t			agino;
585 	int				holecount;
586 	int				i;
587 	int				error = 0;
588 	uint16_t			holemask;
589 
590 	xfs_inobt_btrec_to_irec(mp, rec, &irec);
591 	if (xfs_inobt_check_irec(to_perag(bs->cur->bc_group), &irec) != NULL) {
592 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
593 		return 0;
594 	}
595 
596 	agino = irec.ir_startino;
597 
598 	xchk_iallocbt_rec_alignment(bs, &irec);
599 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
600 		goto out;
601 
602 	iabt->inodes += irec.ir_count;
603 
604 	/* Handle non-sparse inodes */
605 	if (!xfs_inobt_issparse(irec.ir_holemask)) {
606 		if (irec.ir_count != XFS_INODES_PER_CHUNK)
607 			xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
608 
609 		if (!xchk_iallocbt_chunk(bs, &irec, agino,
610 					XFS_INODES_PER_CHUNK))
611 			goto out;
612 		goto check_clusters;
613 	}
614 
615 	/* Check each chunk of a sparse inode cluster. */
616 	holemask = irec.ir_holemask;
617 	holecount = 0;
618 	holes = ~xfs_inobt_irec_to_allocmask(&irec);
619 	if ((holes & irec.ir_free) != holes ||
620 	    irec.ir_freecount > irec.ir_count)
621 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
622 
623 	for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
624 		if (holemask & 1)
625 			holecount += XFS_INODES_PER_HOLEMASK_BIT;
626 		else if (!xchk_iallocbt_chunk(bs, &irec, agino,
627 					XFS_INODES_PER_HOLEMASK_BIT))
628 			goto out;
629 		holemask >>= 1;
630 		agino += XFS_INODES_PER_HOLEMASK_BIT;
631 	}
632 
633 	if (holecount > XFS_INODES_PER_CHUNK ||
634 	    holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
635 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
636 
637 check_clusters:
638 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
639 		goto out;
640 
641 	error = xchk_iallocbt_check_clusters(bs, &irec);
642 	if (error)
643 		goto out;
644 
645 out:
646 	return error;
647 }
648 
649 /*
650  * Make sure the inode btrees are as large as the rmap thinks they are.
651  * Don't bother if we're missing btree cursors, as we're already corrupt.
652  */
653 STATIC void
xchk_iallocbt_xref_rmap_btreeblks(struct xfs_scrub * sc)654 xchk_iallocbt_xref_rmap_btreeblks(
655 	struct xfs_scrub	*sc)
656 {
657 	xfs_filblks_t		blocks;
658 	xfs_filblks_t		inobt_blocks = 0;
659 	xfs_filblks_t		finobt_blocks = 0;
660 	int			error;
661 
662 	if (!sc->sa.ino_cur || !sc->sa.rmap_cur ||
663 	    (xfs_has_finobt(sc->mp) && !sc->sa.fino_cur) ||
664 	    xchk_skip_xref(sc->sm))
665 		return;
666 
667 	/* Check that we saw as many inobt blocks as the rmap says. */
668 	error = xfs_btree_count_blocks(sc->sa.ino_cur, &inobt_blocks);
669 	if (!xchk_process_error(sc, 0, 0, &error))
670 		return;
671 
672 	if (sc->sa.fino_cur) {
673 		error = xfs_btree_count_blocks(sc->sa.fino_cur, &finobt_blocks);
674 		if (!xchk_process_error(sc, 0, 0, &error))
675 			return;
676 	}
677 
678 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
679 			&XFS_RMAP_OINFO_INOBT, &blocks);
680 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
681 		return;
682 	if (blocks != inobt_blocks + finobt_blocks)
683 		xchk_btree_set_corrupt(sc, sc->sa.ino_cur, 0);
684 }
685 
686 /*
687  * Make sure that the inobt records point to the same number of blocks as
688  * the rmap says are owned by inodes.
689  */
690 STATIC void
xchk_iallocbt_xref_rmap_inodes(struct xfs_scrub * sc,unsigned long long inodes)691 xchk_iallocbt_xref_rmap_inodes(
692 	struct xfs_scrub	*sc,
693 	unsigned long long	inodes)
694 {
695 	xfs_filblks_t		blocks;
696 	xfs_filblks_t		inode_blocks;
697 	int			error;
698 
699 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
700 		return;
701 
702 	/* Check that we saw as many inode blocks as the rmap knows about. */
703 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
704 			&XFS_RMAP_OINFO_INODES, &blocks);
705 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
706 		return;
707 	inode_blocks = XFS_B_TO_FSB(sc->mp, inodes * sc->mp->m_sb.sb_inodesize);
708 	if (blocks != inode_blocks)
709 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
710 }
711 
712 /* Scrub one of the inode btrees for some AG. */
713 int
xchk_iallocbt(struct xfs_scrub * sc)714 xchk_iallocbt(
715 	struct xfs_scrub	*sc)
716 {
717 	struct xfs_btree_cur	*cur;
718 	struct xchk_iallocbt	iabt = {
719 		.inodes		= 0,
720 		.next_startino	= NULLAGINO,
721 		.next_cluster_ino = NULLAGINO,
722 	};
723 	int			error;
724 
725 	switch (sc->sm->sm_type) {
726 	case XFS_SCRUB_TYPE_INOBT:
727 		cur = sc->sa.ino_cur;
728 		break;
729 	case XFS_SCRUB_TYPE_FINOBT:
730 		cur = sc->sa.fino_cur;
731 		break;
732 	default:
733 		ASSERT(0);
734 		return -EIO;
735 	}
736 
737 	error = xchk_btree(sc, cur, xchk_iallocbt_rec, &XFS_RMAP_OINFO_INOBT,
738 			&iabt);
739 	if (error)
740 		return error;
741 
742 	xchk_iallocbt_xref_rmap_btreeblks(sc);
743 
744 	/*
745 	 * If we're scrubbing the inode btree, inode_blocks is the number of
746 	 * blocks pointed to by all the inode chunk records.  Therefore, we
747 	 * should compare to the number of inode chunk blocks that the rmap
748 	 * knows about.  We can't do this for the finobt since it only points
749 	 * to inode chunks with free inodes.
750 	 */
751 	if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
752 		xchk_iallocbt_xref_rmap_inodes(sc, iabt.inodes);
753 	return error;
754 }
755 
756 /* See if an inode btree has (or doesn't have) an inode chunk record. */
757 static inline void
xchk_xref_inode_check(struct xfs_scrub * sc,xfs_agblock_t agbno,xfs_extlen_t len,struct xfs_btree_cur ** icur,enum xbtree_recpacking expected)758 xchk_xref_inode_check(
759 	struct xfs_scrub	*sc,
760 	xfs_agblock_t		agbno,
761 	xfs_extlen_t		len,
762 	struct xfs_btree_cur	**icur,
763 	enum xbtree_recpacking	expected)
764 {
765 	enum xbtree_recpacking	outcome;
766 	int			error;
767 
768 	if (!(*icur) || xchk_skip_xref(sc->sm))
769 		return;
770 
771 	error = xfs_ialloc_has_inodes_at_extent(*icur, agbno, len, &outcome);
772 	if (!xchk_should_check_xref(sc, &error, icur))
773 		return;
774 	if (outcome != expected)
775 		xchk_btree_xref_set_corrupt(sc, *icur, 0);
776 }
777 
778 /* xref check that the extent is not covered by inodes */
779 void
xchk_xref_is_not_inode_chunk(struct xfs_scrub * sc,xfs_agblock_t agbno,xfs_extlen_t len)780 xchk_xref_is_not_inode_chunk(
781 	struct xfs_scrub	*sc,
782 	xfs_agblock_t		agbno,
783 	xfs_extlen_t		len)
784 {
785 	xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur,
786 			XBTREE_RECPACKING_EMPTY);
787 	xchk_xref_inode_check(sc, agbno, len, &sc->sa.fino_cur,
788 			XBTREE_RECPACKING_EMPTY);
789 }
790 
791 /* xref check that the extent is covered by inodes */
792 void
xchk_xref_is_inode_chunk(struct xfs_scrub * sc,xfs_agblock_t agbno,xfs_extlen_t len)793 xchk_xref_is_inode_chunk(
794 	struct xfs_scrub	*sc,
795 	xfs_agblock_t		agbno,
796 	xfs_extlen_t		len)
797 {
798 	xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur,
799 			XBTREE_RECPACKING_FULL);
800 }
801