Line data Source code
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 1576976 : xchk_setup_ag_iallocbt(
33 : struct xfs_scrub *sc)
34 : {
35 1576976 : if (xchk_need_intent_drain(sc))
36 4449 : xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
37 1576976 : 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 2255668584 : 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 2255668584 : struct xfs_inobt_rec_incore frec;
73 2255668584 : struct xfs_btree_cur *cur = sc->sa.fino_cur;
74 2255668584 : bool ffree, fhole;
75 2255668584 : unsigned int frec_idx, fhole_idx;
76 2255668584 : int has_record;
77 2255668584 : int error;
78 :
79 2255668584 : ASSERT(cur->bc_btnum == XFS_BTNUM_FINO);
80 :
81 2255668584 : error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has_record);
82 2256184123 : if (error)
83 : return error;
84 2256184123 : if (!has_record)
85 1758600649 : goto no_record;
86 :
87 497583474 : error = xfs_inobt_get_rec(cur, &frec, &has_record);
88 497583333 : if (!has_record)
89 : return -EFSCORRUPTED;
90 :
91 497583333 : if (frec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
92 480130725 : goto no_record;
93 :
94 : /* There's a finobt record; free and hole status must match. */
95 17452608 : frec_idx = agino - frec.ir_startino;
96 17452608 : ffree = frec.ir_free & (1ULL << frec_idx);
97 17452608 : fhole_idx = frec_idx / XFS_INODES_PER_HOLEMASK_BIT;
98 17452608 : fhole = frec.ir_holemask & (1U << fhole_idx);
99 :
100 17452608 : if (ffree != free)
101 0 : xchk_btree_xref_set_corrupt(sc, cur, 0);
102 17452608 : if (fhole != hole)
103 0 : xchk_btree_xref_set_corrupt(sc, cur, 0);
104 : return 0;
105 :
106 2238731374 : no_record:
107 : /* inobt record is fully allocated */
108 2238731374 : if (irec->ir_free == 0)
109 : return 0;
110 :
111 : /* inobt record is totally unallocated */
112 604036210 : if (irec->ir_free == XFS_INOBT_ALL_FREE)
113 : return 0;
114 :
115 : /* inobt record says this is a hole */
116 604036210 : if (hole)
117 : return 0;
118 :
119 : /* finobt doesn't care about allocated inodes */
120 604036210 : if (!free)
121 : return 0;
122 :
123 0 : xchk_btree_xref_set_corrupt(sc, cur, 0);
124 0 : 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 177065728 : 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 177065728 : xfs_agino_t i;
139 177065728 : unsigned int rec_idx;
140 177065728 : int error;
141 :
142 177065728 : ASSERT(sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT);
143 :
144 177065728 : if (!sc->sa.fino_cur || xchk_skip_xref(sc->sm))
145 2 : return;
146 :
147 177065726 : for (i = agino, rec_idx = agino - irec->ir_startino;
148 2433263847 : i < agino + nr_inodes;
149 2256198121 : i++, rec_idx++) {
150 2256198831 : bool free, hole;
151 2256198831 : unsigned int hole_idx;
152 :
153 2256198831 : free = irec->ir_free & (1ULL << rec_idx);
154 2256198831 : hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
155 2256198831 : hole = irec->ir_holemask & (1U << hole_idx);
156 :
157 2256198831 : error = xchk_inobt_xref_finobt(sc, irec, i, free, hole);
158 2256027659 : 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 22861670 : 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 22861670 : struct xfs_inobt_rec_incore irec;
176 22861670 : struct xfs_btree_cur *cur = sc->sa.ino_cur;
177 22861670 : bool free, hole;
178 22861670 : unsigned int rec_idx, hole_idx;
179 22861670 : int has_record;
180 22861670 : int error;
181 :
182 22861670 : ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
183 :
184 22861670 : error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has_record);
185 22861793 : if (error)
186 : return error;
187 22861793 : if (!has_record)
188 0 : goto no_record;
189 :
190 22861793 : error = xfs_inobt_get_rec(cur, &irec, &has_record);
191 22861677 : if (!has_record)
192 : return -EFSCORRUPTED;
193 :
194 22861677 : if (irec.ir_startino + XFS_INODES_PER_CHUNK <= agino)
195 0 : goto no_record;
196 :
197 : /* There's an inobt record; free and hole status must match. */
198 22861677 : rec_idx = agino - irec.ir_startino;
199 22861677 : free = irec.ir_free & (1ULL << rec_idx);
200 22861677 : hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
201 22861677 : hole = irec.ir_holemask & (1U << hole_idx);
202 :
203 22861677 : if (ffree != free)
204 0 : xchk_btree_xref_set_corrupt(sc, cur, 0);
205 22861677 : if (fhole != hole)
206 0 : xchk_btree_xref_set_corrupt(sc, cur, 0);
207 : return 0;
208 :
209 0 : no_record:
210 : /* finobt should never have a record for which the inobt does not */
211 0 : xchk_btree_xref_set_corrupt(sc, cur, 0);
212 0 : 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 808523 : 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 808523 : xfs_agino_t i;
227 808523 : unsigned int rec_idx;
228 808523 : int error;
229 :
230 808523 : ASSERT(sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT);
231 :
232 808523 : if (!sc->sa.ino_cur || xchk_skip_xref(sc->sm))
233 0 : return;
234 :
235 808523 : for (i = agino, rec_idx = agino - frec->ir_startino;
236 23670284 : i < agino + nr_inodes;
237 22861761 : i++, rec_idx++) {
238 22861761 : bool ffree, fhole;
239 22861761 : unsigned int hole_idx;
240 :
241 22861761 : ffree = frec->ir_free & (1ULL << rec_idx);
242 22861761 : hole_idx = rec_idx / XFS_INODES_PER_HOLEMASK_BIT;
243 22861761 : fhole = frec->ir_holemask & (1U << hole_idx);
244 :
245 22861761 : error = xchk_finobt_xref_inobt(sc, frec, i, ffree, fhole);
246 22861769 : 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 177871060 : 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 177871060 : struct xfs_scrub *sc = bs->sc;
260 177871060 : struct xfs_mount *mp = bs->cur->bc_mp;
261 177871060 : struct xfs_perag *pag = bs->cur->bc_ag.pag;
262 177871060 : xfs_agblock_t agbno;
263 177871060 : xfs_extlen_t len;
264 :
265 177871060 : agbno = XFS_AGINO_TO_AGBNO(mp, agino);
266 177871060 : len = XFS_B_TO_FSB(mp, nr_inodes * mp->m_sb.sb_inodesize);
267 :
268 177871060 : if (!xfs_verify_agbext(pag, agbno, len))
269 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
270 :
271 177871060 : if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
272 : return false;
273 :
274 177873141 : xchk_xref_is_used_space(sc, agbno, len);
275 177873591 : if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
276 177065068 : xchk_inobt_chunk_xref_finobt(sc, irec, agino, nr_inodes);
277 : else
278 808523 : xchk_finobt_chunk_xref_inobt(sc, irec, agino, nr_inodes);
279 177872529 : xchk_xref_is_only_owned_by(sc, agbno, len, &XFS_RMAP_OINFO_INODES);
280 177874334 : xchk_xref_is_not_shared(sc, agbno, len);
281 177874740 : xchk_xref_is_not_cow_staging(sc, agbno, len);
282 177874740 : 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 2278623899 : 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 2278623899 : struct xfs_mount *mp = bs->cur->bc_mp;
307 2278623899 : xfs_ino_t fsino;
308 2278623899 : xfs_agino_t agino;
309 2278623899 : bool irec_free;
310 2278623899 : bool ino_inuse;
311 2278623899 : bool freemask_ok;
312 2278623899 : int error = 0;
313 :
314 2278623899 : if (xchk_should_terminate(bs->sc, &error))
315 0 : 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 2278607044 : agino = irec->ir_startino + irec_ino;
322 2278607044 : fsino = XFS_AGINO_TO_INO(mp, bs->cur->bc_ag.pag->pag_agno, agino);
323 2278607044 : irec_free = (irec->ir_free & XFS_INOBT_MASK(irec_ino));
324 :
325 2278607044 : if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
326 2277945378 : (dip->di_version >= 3 && be64_to_cpu(dip->di_ino) != fsino)) {
327 661666 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
328 0 : goto out;
329 : }
330 :
331 2277945378 : error = xfs_icache_inode_is_allocated(mp, bs->cur->bc_tp, fsino,
332 : &ino_inuse);
333 2279081921 : if (error == -ENODATA) {
334 : /* Not cached, just read the disk buffer */
335 29627034 : freemask_ok = irec_free ^ !!(dip->di_mode);
336 29627034 : if (!(bs->sc->flags & XCHK_TRY_HARDER) && !freemask_ok)
337 : return -EDEADLOCK;
338 2249454887 : } else if (error < 0) {
339 : /*
340 : * Inode is only half assembled, or there was an IO error,
341 : * or the verifier failed, so don't bother trying to check.
342 : * The inode scrubber can deal with this.
343 : */
344 374270 : goto out;
345 : } else {
346 : /* Inode is all there. */
347 2249080617 : freemask_ok = irec_free ^ ino_inuse;
348 : }
349 2278707651 : if (!freemask_ok)
350 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
351 2278707651 : out:
352 : return 0;
353 : }
354 :
355 : /*
356 : * Check that the holemask and freemask of a hypothetical inode cluster match
357 : * what's actually on disk. If sparse inodes are enabled, the cluster does
358 : * not actually have to map to inodes if the corresponding holemask bit is set.
359 : *
360 : * @cluster_base is the first inode in the cluster within the @irec.
361 : */
362 : STATIC int
363 90200567 : xchk_iallocbt_check_cluster(
364 : struct xchk_btree *bs,
365 : struct xfs_inobt_rec_incore *irec,
366 : unsigned int cluster_base)
367 : {
368 90200567 : struct xfs_imap imap;
369 90200567 : struct xfs_mount *mp = bs->cur->bc_mp;
370 90200567 : struct xfs_buf *cluster_bp;
371 90200567 : unsigned int nr_inodes;
372 90200567 : xfs_agnumber_t agno = bs->cur->bc_ag.pag->pag_agno;
373 90200567 : xfs_agblock_t agbno;
374 90200567 : unsigned int cluster_index;
375 90200567 : uint16_t cluster_mask = 0;
376 90200567 : uint16_t ir_holemask;
377 90200567 : int error = 0;
378 :
379 90200567 : nr_inodes = min_t(unsigned int, XFS_INODES_PER_CHUNK,
380 : M_IGEO(mp)->inodes_per_cluster);
381 :
382 : /* Map this inode cluster */
383 90200567 : agbno = XFS_AGINO_TO_AGBNO(mp, irec->ir_startino + cluster_base);
384 :
385 : /* Compute a bitmask for this cluster that can be used for holemask. */
386 90200567 : for (cluster_index = 0;
387 811800999 : cluster_index < nr_inodes;
388 721600432 : cluster_index += XFS_INODES_PER_HOLEMASK_BIT)
389 721600432 : cluster_mask |= XFS_INOBT_MASK((cluster_base + cluster_index) /
390 : XFS_INODES_PER_HOLEMASK_BIT);
391 :
392 : /*
393 : * Map the first inode of this cluster to a buffer and offset.
394 : * Be careful about inobt records that don't align with the start of
395 : * the inode buffer when block sizes are large enough to hold multiple
396 : * inode chunks. When this happens, cluster_base will be zero but
397 : * ir_startino can be large enough to make im_boffset nonzero.
398 : */
399 90200567 : ir_holemask = (irec->ir_holemask & cluster_mask);
400 90200567 : imap.im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
401 90200567 : imap.im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
402 90200567 : imap.im_boffset = XFS_INO_TO_OFFSET(mp, irec->ir_startino) <<
403 90200567 : mp->m_sb.sb_inodelog;
404 :
405 90200567 : if (imap.im_boffset != 0 && cluster_base != 0) {
406 0 : ASSERT(imap.im_boffset == 0 || cluster_base == 0);
407 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
408 0 : return 0;
409 : }
410 :
411 180401134 : trace_xchk_iallocbt_check_cluster(mp, agno, irec->ir_startino,
412 : imap.im_blkno, imap.im_len, cluster_base, nr_inodes,
413 : cluster_mask, ir_holemask,
414 90200567 : XFS_INO_TO_OFFSET(mp, irec->ir_startino +
415 : cluster_base));
416 :
417 : /* The whole cluster must be a hole or not a hole. */
418 90202667 : if (ir_holemask != cluster_mask && ir_holemask != 0) {
419 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
420 0 : return 0;
421 : }
422 :
423 : /* If any part of this is a hole, skip it. */
424 90202667 : if (ir_holemask) {
425 18967611 : xchk_xref_is_not_owned_by(bs->sc, agbno,
426 : M_IGEO(mp)->blocks_per_cluster,
427 : &XFS_RMAP_OINFO_INODES);
428 18967611 : return 0;
429 : }
430 :
431 71235056 : xchk_xref_is_only_owned_by(bs->sc, agbno, M_IGEO(mp)->blocks_per_cluster,
432 : &XFS_RMAP_OINFO_INODES);
433 :
434 : /* Grab the inode cluster buffer. */
435 71235037 : error = xfs_imap_to_bp(mp, bs->cur->bc_tp, &imap, &cluster_bp);
436 71236805 : if (!xchk_btree_xref_process_error(bs->sc, bs->cur, 0, &error))
437 0 : return error;
438 :
439 : /* Check free status of each inode within this cluster. */
440 2350313777 : for (cluster_index = 0; cluster_index < nr_inodes; cluster_index++) {
441 2279077619 : struct xfs_dinode *dip;
442 :
443 2279077619 : if (imap.im_boffset >= BBTOB(cluster_bp->b_length)) {
444 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
445 0 : break;
446 : }
447 :
448 2279077619 : dip = xfs_buf_offset(cluster_bp, imap.im_boffset);
449 2278964227 : error = xchk_iallocbt_check_cluster_ifree(bs, irec,
450 : cluster_base + cluster_index, dip);
451 2279076923 : if (error)
452 : break;
453 2279076923 : imap.im_boffset += mp->m_sb.sb_inodesize;
454 : }
455 :
456 71236158 : xfs_trans_brelse(bs->cur->bc_tp, cluster_bp);
457 71236992 : return error;
458 : }
459 :
460 : /*
461 : * For all the inode clusters that could map to this inobt record, make sure
462 : * that the holemask makes sense and that the allocation status of each inode
463 : * matches the freemask.
464 : */
465 : STATIC int
466 45102271 : xchk_iallocbt_check_clusters(
467 : struct xchk_btree *bs,
468 : struct xfs_inobt_rec_incore *irec)
469 : {
470 45102271 : unsigned int cluster_base;
471 45102271 : int error = 0;
472 :
473 : /*
474 : * For the common case where this inobt record maps to multiple inode
475 : * clusters this will call _check_cluster for each cluster.
476 : *
477 : * For the case that multiple inobt records map to a single cluster,
478 : * this will call _check_cluster once.
479 : */
480 45102271 : for (cluster_base = 0;
481 135304784 : cluster_base < XFS_INODES_PER_CHUNK;
482 90202513 : cluster_base += M_IGEO(bs->sc->mp)->inodes_per_cluster) {
483 90202386 : error = xchk_iallocbt_check_cluster(bs, irec, cluster_base);
484 90202513 : if (error)
485 : break;
486 : }
487 :
488 45102398 : return error;
489 : }
490 :
491 : /*
492 : * Make sure this inode btree record is aligned properly. Because a fs block
493 : * contains multiple inodes, we check that the inobt record is aligned to the
494 : * correct inode, not just the correct block on disk. This results in a finer
495 : * grained corruption check.
496 : */
497 : STATIC void
498 45102306 : xchk_iallocbt_rec_alignment(
499 : struct xchk_btree *bs,
500 : struct xfs_inobt_rec_incore *irec)
501 : {
502 45102306 : struct xfs_mount *mp = bs->sc->mp;
503 45102306 : struct xchk_iallocbt *iabt = bs->private;
504 45102306 : struct xfs_ino_geometry *igeo = M_IGEO(mp);
505 :
506 : /*
507 : * finobt records have different positioning requirements than inobt
508 : * records: each finobt record must have a corresponding inobt record.
509 : * That is checked in the xref function, so for now we only catch the
510 : * obvious case where the record isn't at all aligned properly.
511 : *
512 : * Note that if a fs block contains more than a single chunk of inodes,
513 : * we will have finobt records only for those chunks containing free
514 : * inodes, and therefore expect chunk alignment of finobt records.
515 : * Otherwise, we expect that the finobt record is aligned to the
516 : * cluster alignment as told by the superblock.
517 : */
518 45102306 : if (bs->cur->bc_btnum == XFS_BTNUM_FINO) {
519 387304 : unsigned int imask;
520 :
521 387304 : imask = min_t(unsigned int, XFS_INODES_PER_CHUNK,
522 : igeo->cluster_align_inodes) - 1;
523 387304 : if (irec->ir_startino & imask)
524 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
525 387304 : return;
526 : }
527 :
528 44715002 : if (iabt->next_startino != NULLAGINO) {
529 : /*
530 : * We're midway through a cluster of inodes that is mapped by
531 : * multiple inobt records. Did we get the record for the next
532 : * irec in the sequence?
533 : */
534 2 : if (irec->ir_startino != iabt->next_startino) {
535 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
536 0 : return;
537 : }
538 :
539 2 : iabt->next_startino += XFS_INODES_PER_CHUNK;
540 :
541 : /* Are we done with the cluster? */
542 2 : if (iabt->next_startino >= iabt->next_cluster_ino) {
543 2 : iabt->next_startino = NULLAGINO;
544 2 : iabt->next_cluster_ino = NULLAGINO;
545 : }
546 2 : return;
547 : }
548 :
549 : /* inobt records must be aligned to cluster and inoalignmnt size. */
550 44715000 : if (irec->ir_startino & (igeo->cluster_align_inodes - 1)) {
551 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
552 0 : return;
553 : }
554 :
555 44715000 : if (irec->ir_startino & (igeo->inodes_per_cluster - 1)) {
556 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
557 0 : return;
558 : }
559 :
560 44715000 : if (igeo->inodes_per_cluster <= XFS_INODES_PER_CHUNK)
561 : return;
562 :
563 : /*
564 : * If this is the start of an inode cluster that can be mapped by
565 : * multiple inobt records, the next inobt record must follow exactly
566 : * after this one.
567 : */
568 2 : iabt->next_startino = irec->ir_startino + XFS_INODES_PER_CHUNK;
569 2 : iabt->next_cluster_ino = irec->ir_startino + igeo->inodes_per_cluster;
570 : }
571 :
572 : /* Scrub an inobt/finobt record. */
573 : STATIC int
574 45102187 : xchk_iallocbt_rec(
575 : struct xchk_btree *bs,
576 : const union xfs_btree_rec *rec)
577 : {
578 45102187 : struct xfs_mount *mp = bs->cur->bc_mp;
579 45102187 : struct xchk_iallocbt *iabt = bs->private;
580 45102187 : struct xfs_inobt_rec_incore irec;
581 45102187 : uint64_t holes;
582 45102187 : xfs_agino_t agino;
583 45102187 : int holecount;
584 45102187 : int i;
585 45102187 : int error = 0;
586 45102187 : uint16_t holemask;
587 :
588 45102187 : xfs_inobt_btrec_to_irec(mp, rec, &irec);
589 45102320 : if (xfs_inobt_check_irec(bs->cur, &irec) != NULL) {
590 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
591 0 : return 0;
592 : }
593 :
594 45102303 : agino = irec.ir_startino;
595 :
596 45102303 : xchk_iallocbt_rec_alignment(bs, &irec);
597 45102408 : if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
598 0 : goto out;
599 :
600 45102408 : iabt->inodes += irec.ir_count;
601 :
602 : /* Handle non-sparse inodes */
603 45102408 : if (!xfs_inobt_issparse(irec.ir_holemask)) {
604 26134674 : if (irec.ir_count != XFS_INODES_PER_CHUNK)
605 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
606 :
607 26134674 : if (!xchk_iallocbt_chunk(bs, &irec, agino,
608 : XFS_INODES_PER_CHUNK))
609 0 : goto out;
610 26134642 : goto check_clusters;
611 : }
612 :
613 : /* Check each chunk of a sparse inode cluster. */
614 18967734 : holemask = irec.ir_holemask;
615 18967734 : holecount = 0;
616 18967734 : holes = ~xfs_inobt_irec_to_allocmask(&irec);
617 18967729 : if ((holes & irec.ir_free) != holes ||
618 18967729 : irec.ir_freecount > irec.ir_count)
619 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
620 :
621 322447470 : for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
622 303479760 : if (holemask & 1)
623 151741408 : holecount += XFS_INODES_PER_HOLEMASK_BIT;
624 151738352 : else if (!xchk_iallocbt_chunk(bs, &irec, agino,
625 : XFS_INODES_PER_HOLEMASK_BIT))
626 0 : goto out;
627 303479739 : holemask >>= 1;
628 303479739 : agino += XFS_INODES_PER_HOLEMASK_BIT;
629 : }
630 :
631 18967710 : if (holecount > XFS_INODES_PER_CHUNK ||
632 18967710 : holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
633 0 : xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
634 :
635 18967719 : check_clusters:
636 45102307 : if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
637 0 : goto out;
638 :
639 45102307 : error = xchk_iallocbt_check_clusters(bs, &irec);
640 45102341 : if (error)
641 0 : goto out;
642 :
643 45102341 : out:
644 : return error;
645 : }
646 :
647 : /*
648 : * Make sure the inode btrees are as large as the rmap thinks they are.
649 : * Don't bother if we're missing btree cursors, as we're already corrupt.
650 : */
651 : STATIC void
652 1572529 : xchk_iallocbt_xref_rmap_btreeblks(
653 : struct xfs_scrub *sc,
654 : int which)
655 : {
656 1572529 : xfs_filblks_t blocks;
657 1572529 : xfs_extlen_t inobt_blocks = 0;
658 1572529 : xfs_extlen_t finobt_blocks = 0;
659 1572529 : int error;
660 :
661 1572529 : if (!sc->sa.ino_cur || !sc->sa.rmap_cur ||
662 1000283 : (xfs_has_finobt(sc->mp) && !sc->sa.fino_cur) ||
663 1000283 : xchk_skip_xref(sc->sm))
664 572246 : return;
665 :
666 : /* Check that we saw as many inobt blocks as the rmap says. */
667 1000283 : error = xfs_btree_count_blocks(sc->sa.ino_cur, &inobt_blocks);
668 1000273 : if (!xchk_process_error(sc, 0, 0, &error))
669 : return;
670 :
671 1000277 : if (sc->sa.fino_cur) {
672 1000277 : error = xfs_btree_count_blocks(sc->sa.fino_cur, &finobt_blocks);
673 1000275 : if (!xchk_process_error(sc, 0, 0, &error))
674 : return;
675 : }
676 :
677 1000286 : error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
678 : &XFS_RMAP_OINFO_INOBT, &blocks);
679 1000295 : if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
680 : return;
681 1000294 : if (blocks != inobt_blocks + finobt_blocks)
682 0 : xchk_btree_set_corrupt(sc, sc->sa.ino_cur, 0);
683 : }
684 :
685 : /*
686 : * Make sure that the inobt records point to the same number of blocks as
687 : * the rmap says are owned by inodes.
688 : */
689 : STATIC void
690 758760 : xchk_iallocbt_xref_rmap_inodes(
691 : struct xfs_scrub *sc,
692 : int which,
693 : unsigned long long inodes)
694 : {
695 758760 : xfs_filblks_t blocks;
696 758760 : xfs_filblks_t inode_blocks;
697 758760 : int error;
698 :
699 758760 : if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
700 275795 : return;
701 :
702 : /* Check that we saw as many inode blocks as the rmap knows about. */
703 482965 : error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
704 : &XFS_RMAP_OINFO_INODES, &blocks);
705 482965 : if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
706 : return;
707 482965 : inode_blocks = XFS_B_TO_FSB(sc->mp, inodes * sc->mp->m_sb.sb_inodesize);
708 482965 : if (blocks != inode_blocks)
709 0 : xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
710 : }
711 :
712 : /* Scrub the inode btrees for some AG. */
713 : STATIC int
714 1572556 : xchk_iallocbt(
715 : struct xfs_scrub *sc,
716 : xfs_btnum_t which)
717 : {
718 1572556 : struct xfs_btree_cur *cur;
719 1572556 : struct xchk_iallocbt iabt = {
720 : .inodes = 0,
721 : .next_startino = NULLAGINO,
722 : .next_cluster_ino = NULLAGINO,
723 : };
724 1572556 : int error;
725 :
726 1572556 : cur = which == XFS_BTNUM_INO ? sc->sa.ino_cur : sc->sa.fino_cur;
727 1572556 : error = xchk_btree(sc, cur, xchk_iallocbt_rec, &XFS_RMAP_OINFO_INOBT,
728 : &iabt);
729 1572534 : if (error)
730 : return error;
731 :
732 1572529 : xchk_iallocbt_xref_rmap_btreeblks(sc, which);
733 :
734 : /*
735 : * If we're scrubbing the inode btree, inode_blocks is the number of
736 : * blocks pointed to by all the inode chunk records. Therefore, we
737 : * should compare to the number of inode chunk blocks that the rmap
738 : * knows about. We can't do this for the finobt since it only points
739 : * to inode chunks with free inodes.
740 : */
741 1572551 : if (which == XFS_BTNUM_INO)
742 758772 : xchk_iallocbt_xref_rmap_inodes(sc, which, iabt.inodes);
743 :
744 : return error;
745 : }
746 :
747 : int
748 758776 : xchk_inobt(
749 : struct xfs_scrub *sc)
750 : {
751 758776 : return xchk_iallocbt(sc, XFS_BTNUM_INO);
752 : }
753 :
754 : int
755 813786 : xchk_finobt(
756 : struct xfs_scrub *sc)
757 : {
758 813786 : return xchk_iallocbt(sc, XFS_BTNUM_FINO);
759 : }
760 :
761 : /* See if an inode btree has (or doesn't have) an inode chunk record. */
762 : static inline void
763 2109151352 : xchk_xref_inode_check(
764 : struct xfs_scrub *sc,
765 : xfs_agblock_t agbno,
766 : xfs_extlen_t len,
767 : struct xfs_btree_cur **icur,
768 : enum xbtree_recpacking expected)
769 : {
770 2109151352 : enum xbtree_recpacking outcome;
771 2109151352 : int error;
772 :
773 2109151352 : if (!(*icur) || xchk_skip_xref(sc->sm))
774 38 : return;
775 :
776 2109151314 : error = xfs_ialloc_has_inodes_at_extent(*icur, agbno, len, &outcome);
777 2109151355 : if (!xchk_should_check_xref(sc, &error, icur))
778 : return;
779 2109161354 : if (outcome != expected)
780 0 : xchk_btree_xref_set_corrupt(sc, *icur, 0);
781 : }
782 :
783 : /* xref check that the extent is not covered by inodes */
784 : void
785 1052290672 : xchk_xref_is_not_inode_chunk(
786 : struct xfs_scrub *sc,
787 : xfs_agblock_t agbno,
788 : xfs_extlen_t len)
789 : {
790 1052290672 : xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur,
791 : XBTREE_RECPACKING_EMPTY);
792 1052299103 : xchk_xref_inode_check(sc, agbno, len, &sc->sa.fino_cur,
793 : XBTREE_RECPACKING_EMPTY);
794 1052210218 : }
795 :
796 : /* xref check that the extent is covered by inodes */
797 : void
798 4880139 : xchk_xref_is_inode_chunk(
799 : struct xfs_scrub *sc,
800 : xfs_agblock_t agbno,
801 : xfs_extlen_t len)
802 : {
803 4880139 : xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur,
804 : XBTREE_RECPACKING_FULL);
805 4880143 : }
|
