xref: /linux/fs/xfs/scrub/agheader.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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_sb.h"
14 #include "xfs_alloc.h"
15 #include "xfs_ialloc.h"
16 #include "xfs_rmap.h"
17 #include "xfs_ag.h"
18 #include "xfs_inode.h"
19 #include "scrub/scrub.h"
20 #include "scrub/common.h"
21 
22 int
23 xchk_setup_agheader(
24 	struct xfs_scrub	*sc)
25 {
26 	if (xchk_need_intent_drain(sc))
27 		xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
28 	return xchk_setup_fs(sc);
29 }
30 
31 /* Superblock */
32 
33 /* Cross-reference with the other btrees. */
34 STATIC void
35 xchk_superblock_xref(
36 	struct xfs_scrub	*sc,
37 	struct xfs_buf		*bp)
38 {
39 	struct xfs_mount	*mp = sc->mp;
40 	xfs_agnumber_t		agno = sc->sm->sm_agno;
41 	xfs_agblock_t		agbno;
42 	int			error;
43 
44 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
45 		return;
46 
47 	agbno = XFS_SB_BLOCK(mp);
48 
49 	error = xchk_ag_init_existing(sc, agno, &sc->sa);
50 	if (!xchk_xref_process_error(sc, agno, agbno, &error))
51 		return;
52 
53 	xchk_xref_is_used_space(sc, agbno, 1);
54 	xchk_xref_is_not_inode_chunk(sc, agbno, 1);
55 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_FS);
56 	xchk_xref_is_not_shared(sc, agbno, 1);
57 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
58 
59 	/* scrub teardown will take care of sc->sa for us */
60 }
61 
62 /*
63  * Scrub the filesystem superblock.
64  *
65  * Note: We do /not/ attempt to check AG 0's superblock.  Mount is
66  * responsible for validating all the geometry information in sb 0, so
67  * if the filesystem is capable of initiating online scrub, then clearly
68  * sb 0 is ok and we can use its information to check everything else.
69  */
70 int
71 xchk_superblock(
72 	struct xfs_scrub	*sc)
73 {
74 	struct xfs_mount	*mp = sc->mp;
75 	struct xfs_buf		*bp;
76 	struct xfs_dsb		*sb;
77 	struct xfs_perag	*pag;
78 	xfs_agnumber_t		agno;
79 	uint32_t		v2_ok;
80 	__be32			features_mask;
81 	int			error;
82 	__be16			vernum_mask;
83 
84 	agno = sc->sm->sm_agno;
85 	if (agno == 0)
86 		return 0;
87 
88 	/*
89 	 * Grab an active reference to the perag structure.  If we can't get
90 	 * it, we're racing with something that's tearing down the AG, so
91 	 * signal that the AG no longer exists.
92 	 */
93 	pag = xfs_perag_get(mp, agno);
94 	if (!pag)
95 		return -ENOENT;
96 
97 	error = xfs_sb_read_secondary(mp, sc->tp, agno, &bp);
98 	/*
99 	 * The superblock verifier can return several different error codes
100 	 * if it thinks the superblock doesn't look right.  For a mount these
101 	 * would all get bounced back to userspace, but if we're here then the
102 	 * fs mounted successfully, which means that this secondary superblock
103 	 * is simply incorrect.  Treat all these codes the same way we treat
104 	 * any corruption.
105 	 */
106 	switch (error) {
107 	case -EINVAL:	/* also -EWRONGFS */
108 	case -ENOSYS:
109 	case -EFBIG:
110 		error = -EFSCORRUPTED;
111 		fallthrough;
112 	default:
113 		break;
114 	}
115 	if (!xchk_process_error(sc, agno, XFS_SB_BLOCK(mp), &error))
116 		goto out_pag;
117 
118 	sb = bp->b_addr;
119 
120 	/*
121 	 * Verify the geometries match.  Fields that are permanently
122 	 * set by mkfs are checked; fields that can be updated later
123 	 * (and are not propagated to backup superblocks) are preen
124 	 * checked.
125 	 */
126 	if (sb->sb_blocksize != cpu_to_be32(mp->m_sb.sb_blocksize))
127 		xchk_block_set_corrupt(sc, bp);
128 
129 	if (sb->sb_dblocks != cpu_to_be64(mp->m_sb.sb_dblocks))
130 		xchk_block_set_corrupt(sc, bp);
131 
132 	if (sb->sb_rblocks != cpu_to_be64(mp->m_sb.sb_rblocks))
133 		xchk_block_set_corrupt(sc, bp);
134 
135 	if (sb->sb_rextents != cpu_to_be64(mp->m_sb.sb_rextents))
136 		xchk_block_set_corrupt(sc, bp);
137 
138 	if (!uuid_equal(&sb->sb_uuid, &mp->m_sb.sb_uuid))
139 		xchk_block_set_preen(sc, bp);
140 
141 	if (sb->sb_logstart != cpu_to_be64(mp->m_sb.sb_logstart))
142 		xchk_block_set_corrupt(sc, bp);
143 
144 	if (sb->sb_rootino != cpu_to_be64(mp->m_sb.sb_rootino))
145 		xchk_block_set_preen(sc, bp);
146 
147 	if (xfs_has_metadir(sc->mp)) {
148 		if (sb->sb_metadirino != cpu_to_be64(mp->m_sb.sb_metadirino))
149 			xchk_block_set_preen(sc, bp);
150 	} else {
151 		if (sb->sb_rbmino != cpu_to_be64(mp->m_sb.sb_rbmino))
152 			xchk_block_set_preen(sc, bp);
153 
154 		if (sb->sb_rsumino != cpu_to_be64(mp->m_sb.sb_rsumino))
155 			xchk_block_set_preen(sc, bp);
156 	}
157 
158 	if (sb->sb_rextsize != cpu_to_be32(mp->m_sb.sb_rextsize))
159 		xchk_block_set_corrupt(sc, bp);
160 
161 	if (sb->sb_agblocks != cpu_to_be32(mp->m_sb.sb_agblocks))
162 		xchk_block_set_corrupt(sc, bp);
163 
164 	if (sb->sb_agcount != cpu_to_be32(mp->m_sb.sb_agcount))
165 		xchk_block_set_corrupt(sc, bp);
166 
167 	if (sb->sb_rbmblocks != cpu_to_be32(mp->m_sb.sb_rbmblocks))
168 		xchk_block_set_corrupt(sc, bp);
169 
170 	if (sb->sb_logblocks != cpu_to_be32(mp->m_sb.sb_logblocks))
171 		xchk_block_set_corrupt(sc, bp);
172 
173 	/* Check sb_versionnum bits that are set at mkfs time. */
174 	vernum_mask = cpu_to_be16(XFS_SB_VERSION_NUMBITS |
175 				  XFS_SB_VERSION_ALIGNBIT |
176 				  XFS_SB_VERSION_DALIGNBIT |
177 				  XFS_SB_VERSION_SHAREDBIT |
178 				  XFS_SB_VERSION_LOGV2BIT |
179 				  XFS_SB_VERSION_SECTORBIT |
180 				  XFS_SB_VERSION_EXTFLGBIT |
181 				  XFS_SB_VERSION_DIRV2BIT);
182 	if ((sb->sb_versionnum & vernum_mask) !=
183 	    (cpu_to_be16(mp->m_sb.sb_versionnum) & vernum_mask))
184 		xchk_block_set_corrupt(sc, bp);
185 
186 	/* Check sb_versionnum bits that can be set after mkfs time. */
187 	vernum_mask = cpu_to_be16(XFS_SB_VERSION_ATTRBIT |
188 				  XFS_SB_VERSION_NLINKBIT |
189 				  XFS_SB_VERSION_QUOTABIT);
190 	if ((sb->sb_versionnum & vernum_mask) !=
191 	    (cpu_to_be16(mp->m_sb.sb_versionnum) & vernum_mask))
192 		xchk_block_set_preen(sc, bp);
193 
194 	if (sb->sb_sectsize != cpu_to_be16(mp->m_sb.sb_sectsize))
195 		xchk_block_set_corrupt(sc, bp);
196 
197 	if (sb->sb_inodesize != cpu_to_be16(mp->m_sb.sb_inodesize))
198 		xchk_block_set_corrupt(sc, bp);
199 
200 	if (sb->sb_inopblock != cpu_to_be16(mp->m_sb.sb_inopblock))
201 		xchk_block_set_corrupt(sc, bp);
202 
203 	if (memcmp(sb->sb_fname, mp->m_sb.sb_fname, sizeof(sb->sb_fname)))
204 		xchk_block_set_preen(sc, bp);
205 
206 	if (sb->sb_blocklog != mp->m_sb.sb_blocklog)
207 		xchk_block_set_corrupt(sc, bp);
208 
209 	if (sb->sb_sectlog != mp->m_sb.sb_sectlog)
210 		xchk_block_set_corrupt(sc, bp);
211 
212 	if (sb->sb_inodelog != mp->m_sb.sb_inodelog)
213 		xchk_block_set_corrupt(sc, bp);
214 
215 	if (sb->sb_inopblog != mp->m_sb.sb_inopblog)
216 		xchk_block_set_corrupt(sc, bp);
217 
218 	if (sb->sb_agblklog != mp->m_sb.sb_agblklog)
219 		xchk_block_set_corrupt(sc, bp);
220 
221 	if (sb->sb_rextslog != mp->m_sb.sb_rextslog)
222 		xchk_block_set_corrupt(sc, bp);
223 
224 	if (sb->sb_imax_pct != mp->m_sb.sb_imax_pct)
225 		xchk_block_set_preen(sc, bp);
226 
227 	/*
228 	 * Skip the summary counters since we track them in memory anyway.
229 	 * sb_icount, sb_ifree, sb_fdblocks, sb_frexents
230 	 */
231 
232 	if (!xfs_has_metadir(mp)) {
233 		if (sb->sb_uquotino != cpu_to_be64(mp->m_sb.sb_uquotino))
234 			xchk_block_set_preen(sc, bp);
235 
236 		if (sb->sb_gquotino != cpu_to_be64(mp->m_sb.sb_gquotino))
237 			xchk_block_set_preen(sc, bp);
238 	}
239 
240 	/*
241 	 * Skip the quota flags since repair will force quotacheck.
242 	 * sb_qflags
243 	 */
244 
245 	if (sb->sb_flags != mp->m_sb.sb_flags)
246 		xchk_block_set_corrupt(sc, bp);
247 
248 	if (sb->sb_shared_vn != mp->m_sb.sb_shared_vn)
249 		xchk_block_set_corrupt(sc, bp);
250 
251 	if (sb->sb_inoalignmt != cpu_to_be32(mp->m_sb.sb_inoalignmt))
252 		xchk_block_set_corrupt(sc, bp);
253 
254 	if (sb->sb_unit != cpu_to_be32(mp->m_sb.sb_unit))
255 		xchk_block_set_preen(sc, bp);
256 
257 	if (sb->sb_width != cpu_to_be32(mp->m_sb.sb_width))
258 		xchk_block_set_preen(sc, bp);
259 
260 	if (sb->sb_dirblklog != mp->m_sb.sb_dirblklog)
261 		xchk_block_set_corrupt(sc, bp);
262 
263 	if (sb->sb_logsectlog != mp->m_sb.sb_logsectlog)
264 		xchk_block_set_corrupt(sc, bp);
265 
266 	if (sb->sb_logsectsize != cpu_to_be16(mp->m_sb.sb_logsectsize))
267 		xchk_block_set_corrupt(sc, bp);
268 
269 	if (sb->sb_logsunit != cpu_to_be32(mp->m_sb.sb_logsunit))
270 		xchk_block_set_corrupt(sc, bp);
271 
272 	/* Do we see any invalid bits in sb_features2? */
273 	if (!xfs_sb_version_hasmorebits(&mp->m_sb)) {
274 		if (sb->sb_features2 != 0)
275 			xchk_block_set_corrupt(sc, bp);
276 	} else {
277 		v2_ok = XFS_SB_VERSION2_OKBITS;
278 		if (xfs_sb_is_v5(&mp->m_sb))
279 			v2_ok |= XFS_SB_VERSION2_CRCBIT;
280 
281 		if (!!(sb->sb_features2 & cpu_to_be32(~v2_ok)))
282 			xchk_block_set_corrupt(sc, bp);
283 
284 		if (xfs_has_metadir(mp)) {
285 			if (sb->sb_rgblklog != mp->m_sb.sb_rgblklog)
286 				xchk_block_set_corrupt(sc, bp);
287 			if (memchr_inv(sb->sb_pad, 0, sizeof(sb->sb_pad)))
288 				xchk_block_set_preen(sc, bp);
289 		} else {
290 			if (sb->sb_features2 != sb->sb_bad_features2)
291 				xchk_block_set_preen(sc, bp);
292 		}
293 	}
294 
295 	/* Check sb_features2 flags that are set at mkfs time. */
296 	features_mask = cpu_to_be32(XFS_SB_VERSION2_LAZYSBCOUNTBIT |
297 				    XFS_SB_VERSION2_PROJID32BIT |
298 				    XFS_SB_VERSION2_CRCBIT |
299 				    XFS_SB_VERSION2_FTYPE);
300 	if ((sb->sb_features2 & features_mask) !=
301 	    (cpu_to_be32(mp->m_sb.sb_features2) & features_mask))
302 		xchk_block_set_corrupt(sc, bp);
303 
304 	/* Check sb_features2 flags that can be set after mkfs time. */
305 	features_mask = cpu_to_be32(XFS_SB_VERSION2_ATTR2BIT);
306 	if ((sb->sb_features2 & features_mask) !=
307 	    (cpu_to_be32(mp->m_sb.sb_features2) & features_mask))
308 		xchk_block_set_preen(sc, bp);
309 
310 	if (!xfs_has_crc(mp)) {
311 		/* all v5 fields must be zero */
312 		if (memchr_inv(&sb->sb_features_compat, 0,
313 				sizeof(struct xfs_dsb) -
314 				offsetof(struct xfs_dsb, sb_features_compat)))
315 			xchk_block_set_corrupt(sc, bp);
316 	} else {
317 		/* compat features must match */
318 		if (sb->sb_features_compat !=
319 				cpu_to_be32(mp->m_sb.sb_features_compat))
320 			xchk_block_set_corrupt(sc, bp);
321 
322 		/* ro compat features must match */
323 		if (sb->sb_features_ro_compat !=
324 				cpu_to_be32(mp->m_sb.sb_features_ro_compat))
325 			xchk_block_set_corrupt(sc, bp);
326 
327 		/*
328 		 * NEEDSREPAIR is ignored on a secondary super, so we should
329 		 * clear it when we find it, though it's not a corruption.
330 		 */
331 		features_mask = cpu_to_be32(XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR);
332 		if ((cpu_to_be32(mp->m_sb.sb_features_incompat) ^
333 				sb->sb_features_incompat) & features_mask)
334 			xchk_block_set_preen(sc, bp);
335 
336 		/* all other incompat features must match */
337 		if ((cpu_to_be32(mp->m_sb.sb_features_incompat) ^
338 				sb->sb_features_incompat) & ~features_mask)
339 			xchk_block_set_corrupt(sc, bp);
340 
341 		/*
342 		 * log incompat features protect newer log record types from
343 		 * older log recovery code.  Log recovery doesn't check the
344 		 * secondary supers, so we can clear these if needed.
345 		 */
346 		if (sb->sb_features_log_incompat)
347 			xchk_block_set_preen(sc, bp);
348 
349 		/* Don't care about sb_crc */
350 
351 		if (sb->sb_spino_align != cpu_to_be32(mp->m_sb.sb_spino_align))
352 			xchk_block_set_corrupt(sc, bp);
353 
354 		if (!xfs_has_metadir(mp)) {
355 			if (sb->sb_pquotino != cpu_to_be64(mp->m_sb.sb_pquotino))
356 				xchk_block_set_preen(sc, bp);
357 		}
358 
359 		/* Don't care about sb_lsn */
360 	}
361 
362 	if (xfs_has_metauuid(mp)) {
363 		/* The metadata UUID must be the same for all supers */
364 		if (!uuid_equal(&sb->sb_meta_uuid, &mp->m_sb.sb_meta_uuid))
365 			xchk_block_set_corrupt(sc, bp);
366 	}
367 
368 	if (xfs_has_metadir(mp)) {
369 		if (sb->sb_rgcount != cpu_to_be32(mp->m_sb.sb_rgcount))
370 			xchk_block_set_corrupt(sc, bp);
371 
372 		if (sb->sb_rgextents != cpu_to_be32(mp->m_sb.sb_rgextents))
373 			xchk_block_set_corrupt(sc, bp);
374 	}
375 
376 	/* Everything else must be zero. */
377 	if (memchr_inv(sb + 1, 0,
378 			BBTOB(bp->b_length) - sizeof(struct xfs_dsb)))
379 		xchk_block_set_corrupt(sc, bp);
380 
381 	xchk_superblock_xref(sc, bp);
382 out_pag:
383 	xfs_perag_put(pag);
384 	return error;
385 }
386 
387 /* AGF */
388 
389 /* Tally freespace record lengths. */
390 STATIC int
391 xchk_agf_record_bno_lengths(
392 	struct xfs_btree_cur		*cur,
393 	const struct xfs_alloc_rec_incore *rec,
394 	void				*priv)
395 {
396 	xfs_extlen_t			*blocks = priv;
397 
398 	(*blocks) += rec->ar_blockcount;
399 	return 0;
400 }
401 
402 /* Check agf_freeblks */
403 static inline void
404 xchk_agf_xref_freeblks(
405 	struct xfs_scrub	*sc)
406 {
407 	struct xfs_agf		*agf = sc->sa.agf_bp->b_addr;
408 	xfs_extlen_t		blocks = 0;
409 	int			error;
410 
411 	if (!sc->sa.bno_cur)
412 		return;
413 
414 	error = xfs_alloc_query_all(sc->sa.bno_cur,
415 			xchk_agf_record_bno_lengths, &blocks);
416 	if (!xchk_should_check_xref(sc, &error, &sc->sa.bno_cur))
417 		return;
418 	if (blocks != be32_to_cpu(agf->agf_freeblks))
419 		xchk_block_xref_set_corrupt(sc, sc->sa.agf_bp);
420 }
421 
422 /* Cross reference the AGF with the cntbt (freespace by length btree) */
423 static inline void
424 xchk_agf_xref_cntbt(
425 	struct xfs_scrub	*sc)
426 {
427 	struct xfs_agf		*agf = sc->sa.agf_bp->b_addr;
428 	xfs_agblock_t		agbno;
429 	xfs_extlen_t		blocks;
430 	int			have;
431 	int			error;
432 
433 	if (!sc->sa.cnt_cur)
434 		return;
435 
436 	/* Any freespace at all? */
437 	error = xfs_alloc_lookup_le(sc->sa.cnt_cur, 0, -1U, &have);
438 	if (!xchk_should_check_xref(sc, &error, &sc->sa.cnt_cur))
439 		return;
440 	if (!have) {
441 		if (agf->agf_freeblks != cpu_to_be32(0))
442 			xchk_block_xref_set_corrupt(sc, sc->sa.agf_bp);
443 		return;
444 	}
445 
446 	/* Check agf_longest */
447 	error = xfs_alloc_get_rec(sc->sa.cnt_cur, &agbno, &blocks, &have);
448 	if (!xchk_should_check_xref(sc, &error, &sc->sa.cnt_cur))
449 		return;
450 	if (!have || blocks != be32_to_cpu(agf->agf_longest))
451 		xchk_block_xref_set_corrupt(sc, sc->sa.agf_bp);
452 }
453 
454 /* Check the btree block counts in the AGF against the btrees. */
455 STATIC void
456 xchk_agf_xref_btreeblks(
457 	struct xfs_scrub	*sc)
458 {
459 	struct xfs_agf		*agf = sc->sa.agf_bp->b_addr;
460 	struct xfs_mount	*mp = sc->mp;
461 	xfs_agblock_t		blocks;
462 	xfs_agblock_t		btreeblks;
463 	int			error;
464 
465 	/* agf_btreeblks didn't exist before lazysbcount */
466 	if (!xfs_has_lazysbcount(sc->mp))
467 		return;
468 
469 	/* Check agf_rmap_blocks; set up for agf_btreeblks check */
470 	if (sc->sa.rmap_cur) {
471 		error = xfs_btree_count_blocks(sc->sa.rmap_cur, &blocks);
472 		if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
473 			return;
474 		btreeblks = blocks - 1;
475 		if (blocks != be32_to_cpu(agf->agf_rmap_blocks))
476 			xchk_block_xref_set_corrupt(sc, sc->sa.agf_bp);
477 	} else {
478 		btreeblks = 0;
479 	}
480 
481 	/*
482 	 * No rmap cursor; we can't xref if we have the rmapbt feature.
483 	 * We also can't do it if we're missing the free space btree cursors.
484 	 */
485 	if ((xfs_has_rmapbt(mp) && !sc->sa.rmap_cur) ||
486 	    !sc->sa.bno_cur || !sc->sa.cnt_cur)
487 		return;
488 
489 	/* Check agf_btreeblks */
490 	error = xfs_btree_count_blocks(sc->sa.bno_cur, &blocks);
491 	if (!xchk_should_check_xref(sc, &error, &sc->sa.bno_cur))
492 		return;
493 	btreeblks += blocks - 1;
494 
495 	error = xfs_btree_count_blocks(sc->sa.cnt_cur, &blocks);
496 	if (!xchk_should_check_xref(sc, &error, &sc->sa.cnt_cur))
497 		return;
498 	btreeblks += blocks - 1;
499 
500 	if (btreeblks != be32_to_cpu(agf->agf_btreeblks))
501 		xchk_block_xref_set_corrupt(sc, sc->sa.agf_bp);
502 }
503 
504 /* Check agf_refcount_blocks against tree size */
505 static inline void
506 xchk_agf_xref_refcblks(
507 	struct xfs_scrub	*sc)
508 {
509 	struct xfs_agf		*agf = sc->sa.agf_bp->b_addr;
510 	xfs_agblock_t		blocks;
511 	int			error;
512 
513 	if (!sc->sa.refc_cur)
514 		return;
515 
516 	error = xfs_btree_count_blocks(sc->sa.refc_cur, &blocks);
517 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
518 		return;
519 	if (blocks != be32_to_cpu(agf->agf_refcount_blocks))
520 		xchk_block_xref_set_corrupt(sc, sc->sa.agf_bp);
521 }
522 
523 /* Cross-reference with the other btrees. */
524 STATIC void
525 xchk_agf_xref(
526 	struct xfs_scrub	*sc)
527 {
528 	struct xfs_mount	*mp = sc->mp;
529 	xfs_agblock_t		agbno;
530 
531 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
532 		return;
533 
534 	agbno = XFS_AGF_BLOCK(mp);
535 
536 	xchk_ag_btcur_init(sc, &sc->sa);
537 
538 	xchk_xref_is_used_space(sc, agbno, 1);
539 	xchk_agf_xref_freeblks(sc);
540 	xchk_agf_xref_cntbt(sc);
541 	xchk_xref_is_not_inode_chunk(sc, agbno, 1);
542 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_FS);
543 	xchk_agf_xref_btreeblks(sc);
544 	xchk_xref_is_not_shared(sc, agbno, 1);
545 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
546 	xchk_agf_xref_refcblks(sc);
547 
548 	/* scrub teardown will take care of sc->sa for us */
549 }
550 
551 /* Scrub the AGF. */
552 int
553 xchk_agf(
554 	struct xfs_scrub	*sc)
555 {
556 	struct xfs_mount	*mp = sc->mp;
557 	struct xfs_agf		*agf;
558 	struct xfs_perag	*pag;
559 	xfs_agnumber_t		agno = sc->sm->sm_agno;
560 	xfs_agblock_t		agbno;
561 	xfs_agblock_t		eoag;
562 	xfs_agblock_t		agfl_first;
563 	xfs_agblock_t		agfl_last;
564 	xfs_agblock_t		agfl_count;
565 	xfs_agblock_t		fl_count;
566 	int			level;
567 	int			error = 0;
568 
569 	error = xchk_ag_read_headers(sc, agno, &sc->sa);
570 	if (!xchk_process_error(sc, agno, XFS_AGF_BLOCK(sc->mp), &error))
571 		goto out;
572 	xchk_buffer_recheck(sc, sc->sa.agf_bp);
573 
574 	agf = sc->sa.agf_bp->b_addr;
575 	pag = sc->sa.pag;
576 
577 	/* Check the AG length */
578 	eoag = be32_to_cpu(agf->agf_length);
579 	if (eoag != pag_group(pag)->xg_block_count)
580 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
581 
582 	/* Check the AGF btree roots and levels */
583 	agbno = be32_to_cpu(agf->agf_bno_root);
584 	if (!xfs_verify_agbno(pag, agbno))
585 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
586 
587 	agbno = be32_to_cpu(agf->agf_cnt_root);
588 	if (!xfs_verify_agbno(pag, agbno))
589 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
590 
591 	level = be32_to_cpu(agf->agf_bno_level);
592 	if (level <= 0 || level > mp->m_alloc_maxlevels)
593 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
594 
595 	level = be32_to_cpu(agf->agf_cnt_level);
596 	if (level <= 0 || level > mp->m_alloc_maxlevels)
597 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
598 
599 	if (xfs_has_rmapbt(mp)) {
600 		agbno = be32_to_cpu(agf->agf_rmap_root);
601 		if (!xfs_verify_agbno(pag, agbno))
602 			xchk_block_set_corrupt(sc, sc->sa.agf_bp);
603 
604 		level = be32_to_cpu(agf->agf_rmap_level);
605 		if (level <= 0 || level > mp->m_rmap_maxlevels)
606 			xchk_block_set_corrupt(sc, sc->sa.agf_bp);
607 	}
608 
609 	if (xfs_has_reflink(mp)) {
610 		agbno = be32_to_cpu(agf->agf_refcount_root);
611 		if (!xfs_verify_agbno(pag, agbno))
612 			xchk_block_set_corrupt(sc, sc->sa.agf_bp);
613 
614 		level = be32_to_cpu(agf->agf_refcount_level);
615 		if (level <= 0 || level > mp->m_refc_maxlevels)
616 			xchk_block_set_corrupt(sc, sc->sa.agf_bp);
617 	}
618 
619 	/* Check the AGFL counters */
620 	agfl_first = be32_to_cpu(agf->agf_flfirst);
621 	agfl_last = be32_to_cpu(agf->agf_fllast);
622 	agfl_count = be32_to_cpu(agf->agf_flcount);
623 	if (agfl_last > agfl_first)
624 		fl_count = agfl_last - agfl_first + 1;
625 	else
626 		fl_count = xfs_agfl_size(mp) - agfl_first + agfl_last + 1;
627 	if (agfl_count != 0 && fl_count != agfl_count)
628 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
629 
630 	/* Do the incore counters match? */
631 	if (pag->pagf_freeblks != be32_to_cpu(agf->agf_freeblks))
632 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
633 	if (pag->pagf_flcount != be32_to_cpu(agf->agf_flcount))
634 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
635 	if (xfs_has_lazysbcount(sc->mp) &&
636 	    pag->pagf_btreeblks != be32_to_cpu(agf->agf_btreeblks))
637 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
638 
639 	xchk_agf_xref(sc);
640 out:
641 	return error;
642 }
643 
644 /* AGFL */
645 
646 struct xchk_agfl_info {
647 	/* Number of AGFL entries that the AGF claims are in use. */
648 	unsigned int		agflcount;
649 
650 	/* Number of AGFL entries that we found. */
651 	unsigned int		nr_entries;
652 
653 	/* Buffer to hold AGFL entries for extent checking. */
654 	xfs_agblock_t		*entries;
655 
656 	struct xfs_buf		*agfl_bp;
657 	struct xfs_scrub	*sc;
658 };
659 
660 /* Cross-reference with the other btrees. */
661 STATIC void
662 xchk_agfl_block_xref(
663 	struct xfs_scrub	*sc,
664 	xfs_agblock_t		agbno)
665 {
666 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
667 		return;
668 
669 	xchk_xref_is_used_space(sc, agbno, 1);
670 	xchk_xref_is_not_inode_chunk(sc, agbno, 1);
671 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_AG);
672 	xchk_xref_is_not_shared(sc, agbno, 1);
673 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
674 }
675 
676 /* Scrub an AGFL block. */
677 STATIC int
678 xchk_agfl_block(
679 	struct xfs_mount	*mp,
680 	xfs_agblock_t		agbno,
681 	void			*priv)
682 {
683 	struct xchk_agfl_info	*sai = priv;
684 	struct xfs_scrub	*sc = sai->sc;
685 
686 	if (xfs_verify_agbno(sc->sa.pag, agbno) &&
687 	    sai->nr_entries < sai->agflcount)
688 		sai->entries[sai->nr_entries++] = agbno;
689 	else
690 		xchk_block_set_corrupt(sc, sai->agfl_bp);
691 
692 	xchk_agfl_block_xref(sc, agbno);
693 
694 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
695 		return -ECANCELED;
696 
697 	return 0;
698 }
699 
700 static int
701 xchk_agblock_cmp(
702 	const void		*pa,
703 	const void		*pb)
704 {
705 	const xfs_agblock_t	*a = pa;
706 	const xfs_agblock_t	*b = pb;
707 
708 	return (int)*a - (int)*b;
709 }
710 
711 /* Cross-reference with the other btrees. */
712 STATIC void
713 xchk_agfl_xref(
714 	struct xfs_scrub	*sc)
715 {
716 	struct xfs_mount	*mp = sc->mp;
717 	xfs_agblock_t		agbno;
718 
719 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
720 		return;
721 
722 	agbno = XFS_AGFL_BLOCK(mp);
723 
724 	xchk_ag_btcur_init(sc, &sc->sa);
725 
726 	xchk_xref_is_used_space(sc, agbno, 1);
727 	xchk_xref_is_not_inode_chunk(sc, agbno, 1);
728 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_FS);
729 	xchk_xref_is_not_shared(sc, agbno, 1);
730 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
731 
732 	/*
733 	 * Scrub teardown will take care of sc->sa for us.  Leave sc->sa
734 	 * active so that the agfl block xref can use it too.
735 	 */
736 }
737 
738 /* Scrub the AGFL. */
739 int
740 xchk_agfl(
741 	struct xfs_scrub	*sc)
742 {
743 	struct xchk_agfl_info	sai = {
744 		.sc		= sc,
745 	};
746 	struct xfs_agf		*agf;
747 	xfs_agnumber_t		agno = sc->sm->sm_agno;
748 	unsigned int		i;
749 	int			error;
750 
751 	/* Lock the AGF and AGI so that nobody can touch this AG. */
752 	error = xchk_ag_read_headers(sc, agno, &sc->sa);
753 	if (!xchk_process_error(sc, agno, XFS_AGFL_BLOCK(sc->mp), &error))
754 		return error;
755 	if (!sc->sa.agf_bp)
756 		return -EFSCORRUPTED;
757 
758 	/* Try to read the AGFL, and verify its structure if we get it. */
759 	error = xfs_alloc_read_agfl(sc->sa.pag, sc->tp, &sai.agfl_bp);
760 	if (!xchk_process_error(sc, agno, XFS_AGFL_BLOCK(sc->mp), &error))
761 		return error;
762 	xchk_buffer_recheck(sc, sai.agfl_bp);
763 
764 	xchk_agfl_xref(sc);
765 
766 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
767 		goto out;
768 
769 	/* Allocate buffer to ensure uniqueness of AGFL entries. */
770 	agf = sc->sa.agf_bp->b_addr;
771 	sai.agflcount = be32_to_cpu(agf->agf_flcount);
772 	if (sai.agflcount > xfs_agfl_size(sc->mp)) {
773 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
774 		goto out;
775 	}
776 	sai.entries = kvcalloc(sai.agflcount, sizeof(xfs_agblock_t),
777 			       XCHK_GFP_FLAGS);
778 	if (!sai.entries) {
779 		error = -ENOMEM;
780 		goto out;
781 	}
782 
783 	/* Check the blocks in the AGFL. */
784 	error = xfs_agfl_walk(sc->mp, sc->sa.agf_bp->b_addr, sai.agfl_bp,
785 			xchk_agfl_block, &sai);
786 	if (error == -ECANCELED) {
787 		error = 0;
788 		goto out_free;
789 	}
790 	if (error)
791 		goto out_free;
792 
793 	if (sai.agflcount != sai.nr_entries) {
794 		xchk_block_set_corrupt(sc, sc->sa.agf_bp);
795 		goto out_free;
796 	}
797 
798 	/* Sort entries, check for duplicates. */
799 	sort(sai.entries, sai.nr_entries, sizeof(sai.entries[0]),
800 			xchk_agblock_cmp, NULL);
801 	for (i = 1; i < sai.nr_entries; i++) {
802 		if (sai.entries[i] == sai.entries[i - 1]) {
803 			xchk_block_set_corrupt(sc, sc->sa.agf_bp);
804 			break;
805 		}
806 	}
807 
808 out_free:
809 	kvfree(sai.entries);
810 out:
811 	return error;
812 }
813 
814 /* AGI */
815 
816 /* Check agi_count/agi_freecount */
817 static inline void
818 xchk_agi_xref_icounts(
819 	struct xfs_scrub	*sc)
820 {
821 	struct xfs_agi		*agi = sc->sa.agi_bp->b_addr;
822 	xfs_agino_t		icount;
823 	xfs_agino_t		freecount;
824 	int			error;
825 
826 	if (!sc->sa.ino_cur)
827 		return;
828 
829 	error = xfs_ialloc_count_inodes(sc->sa.ino_cur, &icount, &freecount);
830 	if (!xchk_should_check_xref(sc, &error, &sc->sa.ino_cur))
831 		return;
832 	if (be32_to_cpu(agi->agi_count) != icount ||
833 	    be32_to_cpu(agi->agi_freecount) != freecount)
834 		xchk_block_xref_set_corrupt(sc, sc->sa.agi_bp);
835 }
836 
837 /* Check agi_[fi]blocks against tree size */
838 static inline void
839 xchk_agi_xref_fiblocks(
840 	struct xfs_scrub	*sc)
841 {
842 	struct xfs_agi		*agi = sc->sa.agi_bp->b_addr;
843 	xfs_agblock_t		blocks;
844 	int			error = 0;
845 
846 	if (!xfs_has_inobtcounts(sc->mp))
847 		return;
848 
849 	if (sc->sa.ino_cur) {
850 		error = xfs_btree_count_blocks(sc->sa.ino_cur, &blocks);
851 		if (!xchk_should_check_xref(sc, &error, &sc->sa.ino_cur))
852 			return;
853 		if (blocks != be32_to_cpu(agi->agi_iblocks))
854 			xchk_block_xref_set_corrupt(sc, sc->sa.agi_bp);
855 	}
856 
857 	if (sc->sa.fino_cur) {
858 		error = xfs_btree_count_blocks(sc->sa.fino_cur, &blocks);
859 		if (!xchk_should_check_xref(sc, &error, &sc->sa.fino_cur))
860 			return;
861 		if (blocks != be32_to_cpu(agi->agi_fblocks))
862 			xchk_block_xref_set_corrupt(sc, sc->sa.agi_bp);
863 	}
864 }
865 
866 /* Cross-reference with the other btrees. */
867 STATIC void
868 xchk_agi_xref(
869 	struct xfs_scrub	*sc)
870 {
871 	struct xfs_mount	*mp = sc->mp;
872 	xfs_agblock_t		agbno;
873 
874 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
875 		return;
876 
877 	agbno = XFS_AGI_BLOCK(mp);
878 
879 	xchk_ag_btcur_init(sc, &sc->sa);
880 
881 	xchk_xref_is_used_space(sc, agbno, 1);
882 	xchk_xref_is_not_inode_chunk(sc, agbno, 1);
883 	xchk_agi_xref_icounts(sc);
884 	xchk_xref_is_only_owned_by(sc, agbno, 1, &XFS_RMAP_OINFO_FS);
885 	xchk_xref_is_not_shared(sc, agbno, 1);
886 	xchk_xref_is_not_cow_staging(sc, agbno, 1);
887 	xchk_agi_xref_fiblocks(sc);
888 
889 	/* scrub teardown will take care of sc->sa for us */
890 }
891 
892 /*
893  * Check the unlinked buckets for links to bad inodes.  We hold the AGI, so
894  * there cannot be any threads updating unlinked list pointers in this AG.
895  */
896 STATIC void
897 xchk_iunlink(
898 	struct xfs_scrub	*sc,
899 	struct xfs_agi		*agi)
900 {
901 	unsigned int		i;
902 	struct xfs_inode	*ip;
903 
904 	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
905 		xfs_agino_t	agino = be32_to_cpu(agi->agi_unlinked[i]);
906 
907 		while (agino != NULLAGINO) {
908 			if (agino % XFS_AGI_UNLINKED_BUCKETS != i) {
909 				xchk_block_set_corrupt(sc, sc->sa.agi_bp);
910 				return;
911 			}
912 
913 			ip = xfs_iunlink_lookup(sc->sa.pag, agino);
914 			if (!ip) {
915 				xchk_block_set_corrupt(sc, sc->sa.agi_bp);
916 				return;
917 			}
918 
919 			if (!xfs_inode_on_unlinked_list(ip)) {
920 				xchk_block_set_corrupt(sc, sc->sa.agi_bp);
921 				return;
922 			}
923 
924 			agino = ip->i_next_unlinked;
925 		}
926 	}
927 }
928 
929 /* Scrub the AGI. */
930 int
931 xchk_agi(
932 	struct xfs_scrub	*sc)
933 {
934 	struct xfs_mount	*mp = sc->mp;
935 	struct xfs_agi		*agi;
936 	struct xfs_perag	*pag;
937 	struct xfs_ino_geometry	*igeo = M_IGEO(sc->mp);
938 	xfs_agnumber_t		agno = sc->sm->sm_agno;
939 	xfs_agblock_t		agbno;
940 	xfs_agblock_t		eoag;
941 	xfs_agino_t		agino;
942 	xfs_agino_t		first_agino;
943 	xfs_agino_t		last_agino;
944 	xfs_agino_t		icount;
945 	int			i;
946 	int			level;
947 	int			error = 0;
948 
949 	error = xchk_ag_read_headers(sc, agno, &sc->sa);
950 	if (!xchk_process_error(sc, agno, XFS_AGI_BLOCK(sc->mp), &error))
951 		goto out;
952 	xchk_buffer_recheck(sc, sc->sa.agi_bp);
953 
954 	agi = sc->sa.agi_bp->b_addr;
955 	pag = sc->sa.pag;
956 
957 	/* Check the AG length */
958 	eoag = be32_to_cpu(agi->agi_length);
959 	if (eoag != pag_group(pag)->xg_block_count)
960 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
961 
962 	/* Check btree roots and levels */
963 	agbno = be32_to_cpu(agi->agi_root);
964 	if (!xfs_verify_agbno(pag, agbno))
965 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
966 
967 	level = be32_to_cpu(agi->agi_level);
968 	if (level <= 0 || level > igeo->inobt_maxlevels)
969 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
970 
971 	if (xfs_has_finobt(mp)) {
972 		agbno = be32_to_cpu(agi->agi_free_root);
973 		if (!xfs_verify_agbno(pag, agbno))
974 			xchk_block_set_corrupt(sc, sc->sa.agi_bp);
975 
976 		level = be32_to_cpu(agi->agi_free_level);
977 		if (level <= 0 || level > igeo->inobt_maxlevels)
978 			xchk_block_set_corrupt(sc, sc->sa.agi_bp);
979 	}
980 
981 	/* Check inode counters */
982 	xfs_agino_range(mp, agno, &first_agino, &last_agino);
983 	icount = be32_to_cpu(agi->agi_count);
984 	if (icount > last_agino - first_agino + 1 ||
985 	    icount < be32_to_cpu(agi->agi_freecount))
986 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
987 
988 	/* Check inode pointers */
989 	agino = be32_to_cpu(agi->agi_newino);
990 	if (!xfs_verify_agino_or_null(pag, agino))
991 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
992 
993 	agino = be32_to_cpu(agi->agi_dirino);
994 	if (!xfs_verify_agino_or_null(pag, agino))
995 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
996 
997 	/* Check unlinked inode buckets */
998 	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
999 		agino = be32_to_cpu(agi->agi_unlinked[i]);
1000 		if (!xfs_verify_agino_or_null(pag, agino))
1001 			xchk_block_set_corrupt(sc, sc->sa.agi_bp);
1002 	}
1003 
1004 	if (agi->agi_pad32 != cpu_to_be32(0))
1005 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
1006 
1007 	/* Do the incore counters match? */
1008 	if (pag->pagi_count != be32_to_cpu(agi->agi_count))
1009 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
1010 	if (pag->pagi_freecount != be32_to_cpu(agi->agi_freecount))
1011 		xchk_block_set_corrupt(sc, sc->sa.agi_bp);
1012 
1013 	xchk_iunlink(sc, agi);
1014 
1015 	xchk_agi_xref(sc);
1016 out:
1017 	return error;
1018 }
1019