1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2017 Oracle. All Rights Reserved.
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"
20 #include "scrub/scrub.h"
21 #include "scrub/common.h"
22 #include "scrub/btree.h"
23 #include "scrub/trace.h"
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.
32 xchk_setup_ag_iallocbt(
35 return xchk_setup_ag_btree(sc, sc->flags & XCHK_TRY_HARDER);
38 /* Inode btree scrubber. */
40 struct xchk_iallocbt {
41 /* Number of inodes we see while scanning inobt. */
42 unsigned long long inodes;
44 /* Expected next startino, for big block filesystems. */
45 xfs_agino_t next_startino;
47 /* Expected end of the current inode cluster. */
48 xfs_agino_t next_cluster_ino;
52 * If we're checking the finobt, cross-reference with the inobt.
53 * Otherwise we're checking the inobt; if there is an finobt, make sure
54 * we have a record or not depending on freecount.
57 xchk_iallocbt_chunk_xref_other(
59 struct xfs_inobt_rec_incore *irec,
62 struct xfs_btree_cur **pcur;
66 if (sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT)
67 pcur = &sc->sa.ino_cur;
69 pcur = &sc->sa.fino_cur;
72 error = xfs_ialloc_has_inode_record(*pcur, agino, agino, &has_irec);
73 if (!xchk_should_check_xref(sc, &error, pcur))
75 if (((irec->ir_freecount > 0 && !has_irec) ||
76 (irec->ir_freecount == 0 && has_irec)))
77 xchk_btree_xref_set_corrupt(sc, *pcur, 0);
80 /* Cross-reference with the other btrees. */
82 xchk_iallocbt_chunk_xref(
84 struct xfs_inobt_rec_incore *irec,
89 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
92 xchk_xref_is_used_space(sc, agbno, len);
93 xchk_iallocbt_chunk_xref_other(sc, irec, agino);
94 xchk_xref_is_owned_by(sc, agbno, len, &XFS_RMAP_OINFO_INODES);
95 xchk_xref_is_not_shared(sc, agbno, len);
98 /* Is this chunk worth checking? */
101 struct xchk_btree *bs,
102 struct xfs_inobt_rec_incore *irec,
106 struct xfs_mount *mp = bs->cur->bc_mp;
107 xfs_agnumber_t agno = bs->cur->bc_ag.pag->pag_agno;
110 bno = XFS_AGINO_TO_AGBNO(mp, agino);
111 if (bno + len <= bno ||
112 !xfs_verify_agbno(mp, agno, bno) ||
113 !xfs_verify_agbno(mp, agno, bno + len - 1))
114 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
116 xchk_iallocbt_chunk_xref(bs->sc, irec, agino, bno, len);
121 /* Count the number of free inodes. */
123 xchk_iallocbt_freecount(
124 xfs_inofree_t freemask)
126 BUILD_BUG_ON(sizeof(freemask) != sizeof(__u64));
127 return hweight64(freemask);
131 * Check that an inode's allocation status matches ir_free in the inobt
132 * record. First we try querying the in-core inode state, and if the inode
133 * isn't loaded we examine the on-disk inode directly.
135 * Since there can be 1:M and M:1 mappings between inobt records and inode
136 * clusters, we pass in the inode location information as an inobt record;
137 * the index of an inode cluster within the inobt record (as well as the
138 * cluster buffer itself); and the index of the inode within the cluster.
140 * @irec is the inobt record.
141 * @irec_ino is the inode offset from the start of the record.
142 * @dip is the on-disk inode.
145 xchk_iallocbt_check_cluster_ifree(
146 struct xchk_btree *bs,
147 struct xfs_inobt_rec_incore *irec,
148 unsigned int irec_ino,
149 struct xfs_dinode *dip)
151 struct xfs_mount *mp = bs->cur->bc_mp;
159 if (xchk_should_terminate(bs->sc, &error))
163 * Given an inobt record and the offset of an inode from the start of
164 * the record, compute which fs inode we're talking about.
166 agino = irec->ir_startino + irec_ino;
167 fsino = XFS_AGINO_TO_INO(mp, bs->cur->bc_ag.pag->pag_agno, agino);
168 irec_free = (irec->ir_free & XFS_INOBT_MASK(irec_ino));
170 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
171 (dip->di_version >= 3 && be64_to_cpu(dip->di_ino) != fsino)) {
172 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
176 error = xfs_icache_inode_is_allocated(mp, bs->cur->bc_tp, fsino,
178 if (error == -ENODATA) {
179 /* Not cached, just read the disk buffer */
180 freemask_ok = irec_free ^ !!(dip->di_mode);
181 if (!(bs->sc->flags & XCHK_TRY_HARDER) && !freemask_ok)
183 } else if (error < 0) {
185 * Inode is only half assembled, or there was an IO error,
186 * or the verifier failed, so don't bother trying to check.
187 * The inode scrubber can deal with this.
191 /* Inode is all there. */
192 freemask_ok = irec_free ^ ino_inuse;
195 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
201 * Check that the holemask and freemask of a hypothetical inode cluster match
202 * what's actually on disk. If sparse inodes are enabled, the cluster does
203 * not actually have to map to inodes if the corresponding holemask bit is set.
205 * @cluster_base is the first inode in the cluster within the @irec.
208 xchk_iallocbt_check_cluster(
209 struct xchk_btree *bs,
210 struct xfs_inobt_rec_incore *irec,
211 unsigned int cluster_base)
213 struct xfs_imap imap;
214 struct xfs_mount *mp = bs->cur->bc_mp;
215 struct xfs_buf *cluster_bp;
216 unsigned int nr_inodes;
217 xfs_agnumber_t agno = bs->cur->bc_ag.pag->pag_agno;
219 unsigned int cluster_index;
220 uint16_t cluster_mask = 0;
221 uint16_t ir_holemask;
224 nr_inodes = min_t(unsigned int, XFS_INODES_PER_CHUNK,
225 M_IGEO(mp)->inodes_per_cluster);
227 /* Map this inode cluster */
228 agbno = XFS_AGINO_TO_AGBNO(mp, irec->ir_startino + cluster_base);
230 /* Compute a bitmask for this cluster that can be used for holemask. */
231 for (cluster_index = 0;
232 cluster_index < nr_inodes;
233 cluster_index += XFS_INODES_PER_HOLEMASK_BIT)
234 cluster_mask |= XFS_INOBT_MASK((cluster_base + cluster_index) /
235 XFS_INODES_PER_HOLEMASK_BIT);
238 * Map the first inode of this cluster to a buffer and offset.
239 * Be careful about inobt records that don't align with the start of
240 * the inode buffer when block sizes are large enough to hold multiple
241 * inode chunks. When this happens, cluster_base will be zero but
242 * ir_startino can be large enough to make im_boffset nonzero.
244 ir_holemask = (irec->ir_holemask & cluster_mask);
245 imap.im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
246 imap.im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
247 imap.im_boffset = XFS_INO_TO_OFFSET(mp, irec->ir_startino) <<
248 mp->m_sb.sb_inodelog;
250 if (imap.im_boffset != 0 && cluster_base != 0) {
251 ASSERT(imap.im_boffset == 0 || cluster_base == 0);
252 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
256 trace_xchk_iallocbt_check_cluster(mp, agno, irec->ir_startino,
257 imap.im_blkno, imap.im_len, cluster_base, nr_inodes,
258 cluster_mask, ir_holemask,
259 XFS_INO_TO_OFFSET(mp, irec->ir_startino +
262 /* The whole cluster must be a hole or not a hole. */
263 if (ir_holemask != cluster_mask && ir_holemask != 0) {
264 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
268 /* If any part of this is a hole, skip it. */
270 xchk_xref_is_not_owned_by(bs->sc, agbno,
271 M_IGEO(mp)->blocks_per_cluster,
272 &XFS_RMAP_OINFO_INODES);
276 xchk_xref_is_owned_by(bs->sc, agbno, M_IGEO(mp)->blocks_per_cluster,
277 &XFS_RMAP_OINFO_INODES);
279 /* Grab the inode cluster buffer. */
280 error = xfs_imap_to_bp(mp, bs->cur->bc_tp, &imap, &cluster_bp);
281 if (!xchk_btree_xref_process_error(bs->sc, bs->cur, 0, &error))
284 /* Check free status of each inode within this cluster. */
285 for (cluster_index = 0; cluster_index < nr_inodes; cluster_index++) {
286 struct xfs_dinode *dip;
288 if (imap.im_boffset >= BBTOB(cluster_bp->b_length)) {
289 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
293 dip = xfs_buf_offset(cluster_bp, imap.im_boffset);
294 error = xchk_iallocbt_check_cluster_ifree(bs, irec,
295 cluster_base + cluster_index, dip);
298 imap.im_boffset += mp->m_sb.sb_inodesize;
301 xfs_trans_brelse(bs->cur->bc_tp, cluster_bp);
306 * For all the inode clusters that could map to this inobt record, make sure
307 * that the holemask makes sense and that the allocation status of each inode
308 * matches the freemask.
311 xchk_iallocbt_check_clusters(
312 struct xchk_btree *bs,
313 struct xfs_inobt_rec_incore *irec)
315 unsigned int cluster_base;
319 * For the common case where this inobt record maps to multiple inode
320 * clusters this will call _check_cluster for each cluster.
322 * For the case that multiple inobt records map to a single cluster,
323 * this will call _check_cluster once.
325 for (cluster_base = 0;
326 cluster_base < XFS_INODES_PER_CHUNK;
327 cluster_base += M_IGEO(bs->sc->mp)->inodes_per_cluster) {
328 error = xchk_iallocbt_check_cluster(bs, irec, cluster_base);
337 * Make sure this inode btree record is aligned properly. Because a fs block
338 * contains multiple inodes, we check that the inobt record is aligned to the
339 * correct inode, not just the correct block on disk. This results in a finer
340 * grained corruption check.
343 xchk_iallocbt_rec_alignment(
344 struct xchk_btree *bs,
345 struct xfs_inobt_rec_incore *irec)
347 struct xfs_mount *mp = bs->sc->mp;
348 struct xchk_iallocbt *iabt = bs->private;
349 struct xfs_ino_geometry *igeo = M_IGEO(mp);
352 * finobt records have different positioning requirements than inobt
353 * records: each finobt record must have a corresponding inobt record.
354 * That is checked in the xref function, so for now we only catch the
355 * obvious case where the record isn't at all aligned properly.
357 * Note that if a fs block contains more than a single chunk of inodes,
358 * we will have finobt records only for those chunks containing free
359 * inodes, and therefore expect chunk alignment of finobt records.
360 * Otherwise, we expect that the finobt record is aligned to the
361 * cluster alignment as told by the superblock.
363 if (bs->cur->bc_btnum == XFS_BTNUM_FINO) {
366 imask = min_t(unsigned int, XFS_INODES_PER_CHUNK,
367 igeo->cluster_align_inodes) - 1;
368 if (irec->ir_startino & imask)
369 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
373 if (iabt->next_startino != NULLAGINO) {
375 * We're midway through a cluster of inodes that is mapped by
376 * multiple inobt records. Did we get the record for the next
377 * irec in the sequence?
379 if (irec->ir_startino != iabt->next_startino) {
380 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
384 iabt->next_startino += XFS_INODES_PER_CHUNK;
386 /* Are we done with the cluster? */
387 if (iabt->next_startino >= iabt->next_cluster_ino) {
388 iabt->next_startino = NULLAGINO;
389 iabt->next_cluster_ino = NULLAGINO;
394 /* inobt records must be aligned to cluster and inoalignmnt size. */
395 if (irec->ir_startino & (igeo->cluster_align_inodes - 1)) {
396 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
400 if (irec->ir_startino & (igeo->inodes_per_cluster - 1)) {
401 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
405 if (igeo->inodes_per_cluster <= XFS_INODES_PER_CHUNK)
409 * If this is the start of an inode cluster that can be mapped by
410 * multiple inobt records, the next inobt record must follow exactly
413 iabt->next_startino = irec->ir_startino + XFS_INODES_PER_CHUNK;
414 iabt->next_cluster_ino = irec->ir_startino + igeo->inodes_per_cluster;
417 /* Scrub an inobt/finobt record. */
420 struct xchk_btree *bs,
421 const union xfs_btree_rec *rec)
423 struct xfs_mount *mp = bs->cur->bc_mp;
424 struct xchk_iallocbt *iabt = bs->private;
425 struct xfs_inobt_rec_incore irec;
427 xfs_agnumber_t agno = bs->cur->bc_ag.pag->pag_agno;
433 unsigned int real_freecount;
436 xfs_inobt_btrec_to_irec(mp, rec, &irec);
438 if (irec.ir_count > XFS_INODES_PER_CHUNK ||
439 irec.ir_freecount > XFS_INODES_PER_CHUNK)
440 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
442 real_freecount = irec.ir_freecount +
443 (XFS_INODES_PER_CHUNK - irec.ir_count);
444 if (real_freecount != xchk_iallocbt_freecount(irec.ir_free))
445 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
447 agino = irec.ir_startino;
448 /* Record has to be properly aligned within the AG. */
449 if (!xfs_verify_agino(mp, agno, agino) ||
450 !xfs_verify_agino(mp, agno, agino + XFS_INODES_PER_CHUNK - 1)) {
451 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
455 xchk_iallocbt_rec_alignment(bs, &irec);
456 if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
459 iabt->inodes += irec.ir_count;
461 /* Handle non-sparse inodes */
462 if (!xfs_inobt_issparse(irec.ir_holemask)) {
463 len = XFS_B_TO_FSB(mp,
464 XFS_INODES_PER_CHUNK * mp->m_sb.sb_inodesize);
465 if (irec.ir_count != XFS_INODES_PER_CHUNK)
466 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
468 if (!xchk_iallocbt_chunk(bs, &irec, agino, len))
473 /* Check each chunk of a sparse inode cluster. */
474 holemask = irec.ir_holemask;
476 len = XFS_B_TO_FSB(mp,
477 XFS_INODES_PER_HOLEMASK_BIT * mp->m_sb.sb_inodesize);
478 holes = ~xfs_inobt_irec_to_allocmask(&irec);
479 if ((holes & irec.ir_free) != holes ||
480 irec.ir_freecount > irec.ir_count)
481 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
483 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
485 holecount += XFS_INODES_PER_HOLEMASK_BIT;
486 else if (!xchk_iallocbt_chunk(bs, &irec, agino, len))
489 agino += XFS_INODES_PER_HOLEMASK_BIT;
492 if (holecount > XFS_INODES_PER_CHUNK ||
493 holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
494 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
497 error = xchk_iallocbt_check_clusters(bs, &irec);
506 * Make sure the inode btrees are as large as the rmap thinks they are.
507 * Don't bother if we're missing btree cursors, as we're already corrupt.
510 xchk_iallocbt_xref_rmap_btreeblks(
511 struct xfs_scrub *sc,
514 xfs_filblks_t blocks;
515 xfs_extlen_t inobt_blocks = 0;
516 xfs_extlen_t finobt_blocks = 0;
519 if (!sc->sa.ino_cur || !sc->sa.rmap_cur ||
520 (xfs_has_finobt(sc->mp) && !sc->sa.fino_cur) ||
521 xchk_skip_xref(sc->sm))
524 /* Check that we saw as many inobt blocks as the rmap says. */
525 error = xfs_btree_count_blocks(sc->sa.ino_cur, &inobt_blocks);
526 if (!xchk_process_error(sc, 0, 0, &error))
529 if (sc->sa.fino_cur) {
530 error = xfs_btree_count_blocks(sc->sa.fino_cur, &finobt_blocks);
531 if (!xchk_process_error(sc, 0, 0, &error))
535 error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
536 &XFS_RMAP_OINFO_INOBT, &blocks);
537 if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
539 if (blocks != inobt_blocks + finobt_blocks)
540 xchk_btree_set_corrupt(sc, sc->sa.ino_cur, 0);
544 * Make sure that the inobt records point to the same number of blocks as
545 * the rmap says are owned by inodes.
548 xchk_iallocbt_xref_rmap_inodes(
549 struct xfs_scrub *sc,
551 unsigned long long inodes)
553 xfs_filblks_t blocks;
554 xfs_filblks_t inode_blocks;
557 if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
560 /* Check that we saw as many inode blocks as the rmap knows about. */
561 error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
562 &XFS_RMAP_OINFO_INODES, &blocks);
563 if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
565 inode_blocks = XFS_B_TO_FSB(sc->mp, inodes * sc->mp->m_sb.sb_inodesize);
566 if (blocks != inode_blocks)
567 xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
570 /* Scrub the inode btrees for some AG. */
573 struct xfs_scrub *sc,
576 struct xfs_btree_cur *cur;
577 struct xchk_iallocbt iabt = {
579 .next_startino = NULLAGINO,
580 .next_cluster_ino = NULLAGINO,
584 cur = which == XFS_BTNUM_INO ? sc->sa.ino_cur : sc->sa.fino_cur;
585 error = xchk_btree(sc, cur, xchk_iallocbt_rec, &XFS_RMAP_OINFO_INOBT,
590 xchk_iallocbt_xref_rmap_btreeblks(sc, which);
593 * If we're scrubbing the inode btree, inode_blocks is the number of
594 * blocks pointed to by all the inode chunk records. Therefore, we
595 * should compare to the number of inode chunk blocks that the rmap
596 * knows about. We can't do this for the finobt since it only points
597 * to inode chunks with free inodes.
599 if (which == XFS_BTNUM_INO)
600 xchk_iallocbt_xref_rmap_inodes(sc, which, iabt.inodes);
607 struct xfs_scrub *sc)
609 return xchk_iallocbt(sc, XFS_BTNUM_INO);
614 struct xfs_scrub *sc)
616 return xchk_iallocbt(sc, XFS_BTNUM_FINO);
619 /* See if an inode btree has (or doesn't have) an inode chunk record. */
621 xchk_xref_inode_check(
622 struct xfs_scrub *sc,
625 struct xfs_btree_cur **icur,
626 bool should_have_inodes)
631 if (!(*icur) || xchk_skip_xref(sc->sm))
634 error = xfs_ialloc_has_inodes_at_extent(*icur, agbno, len, &has_inodes);
635 if (!xchk_should_check_xref(sc, &error, icur))
637 if (has_inodes != should_have_inodes)
638 xchk_btree_xref_set_corrupt(sc, *icur, 0);
641 /* xref check that the extent is not covered by inodes */
643 xchk_xref_is_not_inode_chunk(
644 struct xfs_scrub *sc,
648 xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur, false);
649 xchk_xref_inode_check(sc, agbno, len, &sc->sa.fino_cur, false);
652 /* xref check that the extent is covered by inodes */
654 xchk_xref_is_inode_chunk(
655 struct xfs_scrub *sc,
659 xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur, true);