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 struct xfs_perag *pag = bs->cur->bc_ag.pag;
110 bno = XFS_AGINO_TO_AGBNO(mp, agino);
112 if (!xfs_verify_agbext(pag, bno, len))
113 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
115 xchk_iallocbt_chunk_xref(bs->sc, irec, agino, bno, len);
120 /* Count the number of free inodes. */
122 xchk_iallocbt_freecount(
123 xfs_inofree_t freemask)
125 BUILD_BUG_ON(sizeof(freemask) != sizeof(__u64));
126 return hweight64(freemask);
130 * Check that an inode's allocation status matches ir_free in the inobt
131 * record. First we try querying the in-core inode state, and if the inode
132 * isn't loaded we examine the on-disk inode directly.
134 * Since there can be 1:M and M:1 mappings between inobt records and inode
135 * clusters, we pass in the inode location information as an inobt record;
136 * the index of an inode cluster within the inobt record (as well as the
137 * cluster buffer itself); and the index of the inode within the cluster.
139 * @irec is the inobt record.
140 * @irec_ino is the inode offset from the start of the record.
141 * @dip is the on-disk inode.
144 xchk_iallocbt_check_cluster_ifree(
145 struct xchk_btree *bs,
146 struct xfs_inobt_rec_incore *irec,
147 unsigned int irec_ino,
148 struct xfs_dinode *dip)
150 struct xfs_mount *mp = bs->cur->bc_mp;
158 if (xchk_should_terminate(bs->sc, &error))
162 * Given an inobt record and the offset of an inode from the start of
163 * the record, compute which fs inode we're talking about.
165 agino = irec->ir_startino + irec_ino;
166 fsino = XFS_AGINO_TO_INO(mp, bs->cur->bc_ag.pag->pag_agno, agino);
167 irec_free = (irec->ir_free & XFS_INOBT_MASK(irec_ino));
169 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
170 (dip->di_version >= 3 && be64_to_cpu(dip->di_ino) != fsino)) {
171 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
175 error = xfs_icache_inode_is_allocated(mp, bs->cur->bc_tp, fsino,
177 if (error == -ENODATA) {
178 /* Not cached, just read the disk buffer */
179 freemask_ok = irec_free ^ !!(dip->di_mode);
180 if (!(bs->sc->flags & XCHK_TRY_HARDER) && !freemask_ok)
182 } else if (error < 0) {
184 * Inode is only half assembled, or there was an IO error,
185 * or the verifier failed, so don't bother trying to check.
186 * The inode scrubber can deal with this.
190 /* Inode is all there. */
191 freemask_ok = irec_free ^ ino_inuse;
194 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
200 * Check that the holemask and freemask of a hypothetical inode cluster match
201 * what's actually on disk. If sparse inodes are enabled, the cluster does
202 * not actually have to map to inodes if the corresponding holemask bit is set.
204 * @cluster_base is the first inode in the cluster within the @irec.
207 xchk_iallocbt_check_cluster(
208 struct xchk_btree *bs,
209 struct xfs_inobt_rec_incore *irec,
210 unsigned int cluster_base)
212 struct xfs_imap imap;
213 struct xfs_mount *mp = bs->cur->bc_mp;
214 struct xfs_buf *cluster_bp;
215 unsigned int nr_inodes;
216 xfs_agnumber_t agno = bs->cur->bc_ag.pag->pag_agno;
218 unsigned int cluster_index;
219 uint16_t cluster_mask = 0;
220 uint16_t ir_holemask;
223 nr_inodes = min_t(unsigned int, XFS_INODES_PER_CHUNK,
224 M_IGEO(mp)->inodes_per_cluster);
226 /* Map this inode cluster */
227 agbno = XFS_AGINO_TO_AGBNO(mp, irec->ir_startino + cluster_base);
229 /* Compute a bitmask for this cluster that can be used for holemask. */
230 for (cluster_index = 0;
231 cluster_index < nr_inodes;
232 cluster_index += XFS_INODES_PER_HOLEMASK_BIT)
233 cluster_mask |= XFS_INOBT_MASK((cluster_base + cluster_index) /
234 XFS_INODES_PER_HOLEMASK_BIT);
237 * Map the first inode of this cluster to a buffer and offset.
238 * Be careful about inobt records that don't align with the start of
239 * the inode buffer when block sizes are large enough to hold multiple
240 * inode chunks. When this happens, cluster_base will be zero but
241 * ir_startino can be large enough to make im_boffset nonzero.
243 ir_holemask = (irec->ir_holemask & cluster_mask);
244 imap.im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
245 imap.im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
246 imap.im_boffset = XFS_INO_TO_OFFSET(mp, irec->ir_startino) <<
247 mp->m_sb.sb_inodelog;
249 if (imap.im_boffset != 0 && cluster_base != 0) {
250 ASSERT(imap.im_boffset == 0 || cluster_base == 0);
251 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
255 trace_xchk_iallocbt_check_cluster(mp, agno, irec->ir_startino,
256 imap.im_blkno, imap.im_len, cluster_base, nr_inodes,
257 cluster_mask, ir_holemask,
258 XFS_INO_TO_OFFSET(mp, irec->ir_startino +
261 /* The whole cluster must be a hole or not a hole. */
262 if (ir_holemask != cluster_mask && ir_holemask != 0) {
263 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
267 /* If any part of this is a hole, skip it. */
269 xchk_xref_is_not_owned_by(bs->sc, agbno,
270 M_IGEO(mp)->blocks_per_cluster,
271 &XFS_RMAP_OINFO_INODES);
275 xchk_xref_is_owned_by(bs->sc, agbno, M_IGEO(mp)->blocks_per_cluster,
276 &XFS_RMAP_OINFO_INODES);
278 /* Grab the inode cluster buffer. */
279 error = xfs_imap_to_bp(mp, bs->cur->bc_tp, &imap, &cluster_bp);
280 if (!xchk_btree_xref_process_error(bs->sc, bs->cur, 0, &error))
283 /* Check free status of each inode within this cluster. */
284 for (cluster_index = 0; cluster_index < nr_inodes; cluster_index++) {
285 struct xfs_dinode *dip;
287 if (imap.im_boffset >= BBTOB(cluster_bp->b_length)) {
288 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
292 dip = xfs_buf_offset(cluster_bp, imap.im_boffset);
293 error = xchk_iallocbt_check_cluster_ifree(bs, irec,
294 cluster_base + cluster_index, dip);
297 imap.im_boffset += mp->m_sb.sb_inodesize;
300 xfs_trans_brelse(bs->cur->bc_tp, cluster_bp);
305 * For all the inode clusters that could map to this inobt record, make sure
306 * that the holemask makes sense and that the allocation status of each inode
307 * matches the freemask.
310 xchk_iallocbt_check_clusters(
311 struct xchk_btree *bs,
312 struct xfs_inobt_rec_incore *irec)
314 unsigned int cluster_base;
318 * For the common case where this inobt record maps to multiple inode
319 * clusters this will call _check_cluster for each cluster.
321 * For the case that multiple inobt records map to a single cluster,
322 * this will call _check_cluster once.
324 for (cluster_base = 0;
325 cluster_base < XFS_INODES_PER_CHUNK;
326 cluster_base += M_IGEO(bs->sc->mp)->inodes_per_cluster) {
327 error = xchk_iallocbt_check_cluster(bs, irec, cluster_base);
336 * Make sure this inode btree record is aligned properly. Because a fs block
337 * contains multiple inodes, we check that the inobt record is aligned to the
338 * correct inode, not just the correct block on disk. This results in a finer
339 * grained corruption check.
342 xchk_iallocbt_rec_alignment(
343 struct xchk_btree *bs,
344 struct xfs_inobt_rec_incore *irec)
346 struct xfs_mount *mp = bs->sc->mp;
347 struct xchk_iallocbt *iabt = bs->private;
348 struct xfs_ino_geometry *igeo = M_IGEO(mp);
351 * finobt records have different positioning requirements than inobt
352 * records: each finobt record must have a corresponding inobt record.
353 * That is checked in the xref function, so for now we only catch the
354 * obvious case where the record isn't at all aligned properly.
356 * Note that if a fs block contains more than a single chunk of inodes,
357 * we will have finobt records only for those chunks containing free
358 * inodes, and therefore expect chunk alignment of finobt records.
359 * Otherwise, we expect that the finobt record is aligned to the
360 * cluster alignment as told by the superblock.
362 if (bs->cur->bc_btnum == XFS_BTNUM_FINO) {
365 imask = min_t(unsigned int, XFS_INODES_PER_CHUNK,
366 igeo->cluster_align_inodes) - 1;
367 if (irec->ir_startino & imask)
368 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
372 if (iabt->next_startino != NULLAGINO) {
374 * We're midway through a cluster of inodes that is mapped by
375 * multiple inobt records. Did we get the record for the next
376 * irec in the sequence?
378 if (irec->ir_startino != iabt->next_startino) {
379 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
383 iabt->next_startino += XFS_INODES_PER_CHUNK;
385 /* Are we done with the cluster? */
386 if (iabt->next_startino >= iabt->next_cluster_ino) {
387 iabt->next_startino = NULLAGINO;
388 iabt->next_cluster_ino = NULLAGINO;
393 /* inobt records must be aligned to cluster and inoalignmnt size. */
394 if (irec->ir_startino & (igeo->cluster_align_inodes - 1)) {
395 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
399 if (irec->ir_startino & (igeo->inodes_per_cluster - 1)) {
400 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
404 if (igeo->inodes_per_cluster <= XFS_INODES_PER_CHUNK)
408 * If this is the start of an inode cluster that can be mapped by
409 * multiple inobt records, the next inobt record must follow exactly
412 iabt->next_startino = irec->ir_startino + XFS_INODES_PER_CHUNK;
413 iabt->next_cluster_ino = irec->ir_startino + igeo->inodes_per_cluster;
416 /* Scrub an inobt/finobt record. */
419 struct xchk_btree *bs,
420 const union xfs_btree_rec *rec)
422 struct xfs_mount *mp = bs->cur->bc_mp;
423 struct xfs_perag *pag = bs->cur->bc_ag.pag;
424 struct xchk_iallocbt *iabt = bs->private;
425 struct xfs_inobt_rec_incore irec;
432 unsigned int real_freecount;
435 xfs_inobt_btrec_to_irec(mp, rec, &irec);
437 if (irec.ir_count > XFS_INODES_PER_CHUNK ||
438 irec.ir_freecount > XFS_INODES_PER_CHUNK)
439 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
441 real_freecount = irec.ir_freecount +
442 (XFS_INODES_PER_CHUNK - irec.ir_count);
443 if (real_freecount != xchk_iallocbt_freecount(irec.ir_free))
444 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
446 agino = irec.ir_startino;
447 /* Record has to be properly aligned within the AG. */
448 if (!xfs_verify_agino(pag, agino) ||
449 !xfs_verify_agino(pag, agino + XFS_INODES_PER_CHUNK - 1)) {
450 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
454 xchk_iallocbt_rec_alignment(bs, &irec);
455 if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
458 iabt->inodes += irec.ir_count;
460 /* Handle non-sparse inodes */
461 if (!xfs_inobt_issparse(irec.ir_holemask)) {
462 len = XFS_B_TO_FSB(mp,
463 XFS_INODES_PER_CHUNK * mp->m_sb.sb_inodesize);
464 if (irec.ir_count != XFS_INODES_PER_CHUNK)
465 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
467 if (!xchk_iallocbt_chunk(bs, &irec, agino, len))
472 /* Check each chunk of a sparse inode cluster. */
473 holemask = irec.ir_holemask;
475 len = XFS_B_TO_FSB(mp,
476 XFS_INODES_PER_HOLEMASK_BIT * mp->m_sb.sb_inodesize);
477 holes = ~xfs_inobt_irec_to_allocmask(&irec);
478 if ((holes & irec.ir_free) != holes ||
479 irec.ir_freecount > irec.ir_count)
480 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
482 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
484 holecount += XFS_INODES_PER_HOLEMASK_BIT;
485 else if (!xchk_iallocbt_chunk(bs, &irec, agino, len))
488 agino += XFS_INODES_PER_HOLEMASK_BIT;
491 if (holecount > XFS_INODES_PER_CHUNK ||
492 holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
493 xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
496 error = xchk_iallocbt_check_clusters(bs, &irec);
505 * Make sure the inode btrees are as large as the rmap thinks they are.
506 * Don't bother if we're missing btree cursors, as we're already corrupt.
509 xchk_iallocbt_xref_rmap_btreeblks(
510 struct xfs_scrub *sc,
513 xfs_filblks_t blocks;
514 xfs_extlen_t inobt_blocks = 0;
515 xfs_extlen_t finobt_blocks = 0;
518 if (!sc->sa.ino_cur || !sc->sa.rmap_cur ||
519 (xfs_has_finobt(sc->mp) && !sc->sa.fino_cur) ||
520 xchk_skip_xref(sc->sm))
523 /* Check that we saw as many inobt blocks as the rmap says. */
524 error = xfs_btree_count_blocks(sc->sa.ino_cur, &inobt_blocks);
525 if (!xchk_process_error(sc, 0, 0, &error))
528 if (sc->sa.fino_cur) {
529 error = xfs_btree_count_blocks(sc->sa.fino_cur, &finobt_blocks);
530 if (!xchk_process_error(sc, 0, 0, &error))
534 error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
535 &XFS_RMAP_OINFO_INOBT, &blocks);
536 if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
538 if (blocks != inobt_blocks + finobt_blocks)
539 xchk_btree_set_corrupt(sc, sc->sa.ino_cur, 0);
543 * Make sure that the inobt records point to the same number of blocks as
544 * the rmap says are owned by inodes.
547 xchk_iallocbt_xref_rmap_inodes(
548 struct xfs_scrub *sc,
550 unsigned long long inodes)
552 xfs_filblks_t blocks;
553 xfs_filblks_t inode_blocks;
556 if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
559 /* Check that we saw as many inode blocks as the rmap knows about. */
560 error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
561 &XFS_RMAP_OINFO_INODES, &blocks);
562 if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
564 inode_blocks = XFS_B_TO_FSB(sc->mp, inodes * sc->mp->m_sb.sb_inodesize);
565 if (blocks != inode_blocks)
566 xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
569 /* Scrub the inode btrees for some AG. */
572 struct xfs_scrub *sc,
575 struct xfs_btree_cur *cur;
576 struct xchk_iallocbt iabt = {
578 .next_startino = NULLAGINO,
579 .next_cluster_ino = NULLAGINO,
583 cur = which == XFS_BTNUM_INO ? sc->sa.ino_cur : sc->sa.fino_cur;
584 error = xchk_btree(sc, cur, xchk_iallocbt_rec, &XFS_RMAP_OINFO_INOBT,
589 xchk_iallocbt_xref_rmap_btreeblks(sc, which);
592 * If we're scrubbing the inode btree, inode_blocks is the number of
593 * blocks pointed to by all the inode chunk records. Therefore, we
594 * should compare to the number of inode chunk blocks that the rmap
595 * knows about. We can't do this for the finobt since it only points
596 * to inode chunks with free inodes.
598 if (which == XFS_BTNUM_INO)
599 xchk_iallocbt_xref_rmap_inodes(sc, which, iabt.inodes);
606 struct xfs_scrub *sc)
608 return xchk_iallocbt(sc, XFS_BTNUM_INO);
613 struct xfs_scrub *sc)
615 return xchk_iallocbt(sc, XFS_BTNUM_FINO);
618 /* See if an inode btree has (or doesn't have) an inode chunk record. */
620 xchk_xref_inode_check(
621 struct xfs_scrub *sc,
624 struct xfs_btree_cur **icur,
625 bool should_have_inodes)
630 if (!(*icur) || xchk_skip_xref(sc->sm))
633 error = xfs_ialloc_has_inodes_at_extent(*icur, agbno, len, &has_inodes);
634 if (!xchk_should_check_xref(sc, &error, icur))
636 if (has_inodes != should_have_inodes)
637 xchk_btree_xref_set_corrupt(sc, *icur, 0);
640 /* xref check that the extent is not covered by inodes */
642 xchk_xref_is_not_inode_chunk(
643 struct xfs_scrub *sc,
647 xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur, false);
648 xchk_xref_inode_check(sc, agbno, len, &sc->sa.fino_cur, false);
651 /* xref check that the extent is covered by inodes */
653 xchk_xref_is_inode_chunk(
654 struct xfs_scrub *sc,
658 xchk_xref_inode_check(sc, agbno, len, &sc->sa.ino_cur, true);