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"
14 #include "xfs_log_format.h"
15 #include "xfs_trans.h"
16 #include "xfs_inode.h"
17 #include "xfs_alloc.h"
19 #include "xfs_bmap_btree.h"
21 #include "xfs_rmap_btree.h"
22 #include "scrub/scrub.h"
23 #include "scrub/common.h"
24 #include "scrub/btree.h"
26 /* Set us up with an inode's bmap. */
28 xchk_setup_inode_bmap(
34 error = xchk_get_inode(sc, ip);
38 sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
39 xfs_ilock(sc->ip, sc->ilock_flags);
42 * We don't want any ephemeral data fork updates sitting around
43 * while we inspect block mappings, so wait for directio to finish
44 * and flush dirty data if we have delalloc reservations.
46 if (S_ISREG(VFS_I(sc->ip)->i_mode) &&
47 sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) {
48 struct address_space *mapping = VFS_I(sc->ip)->i_mapping;
50 inode_dio_wait(VFS_I(sc->ip));
53 * Try to flush all incore state to disk before we examine the
54 * space mappings for the data fork. Leave accumulated errors
55 * in the mapping for the writer threads to consume.
57 * On ENOSPC or EIO writeback errors, we continue into the
58 * extent mapping checks because write failures do not
59 * necessarily imply anything about the correctness of the file
60 * metadata. The metadata and the file data could be on
61 * completely separate devices; a media failure might only
62 * affect a subset of the disk, etc. We can handle delalloc
63 * extents in the scrubber, so leaving them in memory is fine.
65 error = filemap_fdatawrite(mapping);
67 error = filemap_fdatawait_keep_errors(mapping);
68 if (error && (error != -ENOSPC && error != -EIO))
72 /* Got the inode, lock it and we're ready to go. */
73 error = xchk_trans_alloc(sc, 0);
76 sc->ilock_flags |= XFS_ILOCK_EXCL;
77 xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
80 /* scrub teardown will unlock and release the inode */
85 * Inode fork block mapping (BMBT) scrubber.
86 * More complex than the others because we have to scrub
87 * all the extents regardless of whether or not the fork
91 struct xchk_bmap_info {
93 xfs_fileoff_t lastoff;
100 /* Look for a corresponding rmap for this irec. */
103 struct xchk_bmap_info *info,
104 struct xfs_bmbt_irec *irec,
107 struct xfs_rmap_irec *rmap)
109 xfs_fileoff_t offset;
110 unsigned int rflags = 0;
114 if (info->whichfork == XFS_ATTR_FORK)
115 rflags |= XFS_RMAP_ATTR_FORK;
118 * CoW staging extents are owned (on disk) by the refcountbt, so
119 * their rmaps do not have offsets.
121 if (info->whichfork == XFS_COW_FORK)
124 offset = irec->br_startoff;
127 * If the caller thinks this could be a shared bmbt extent (IOWs,
128 * any data fork extent of a reflink inode) then we have to use the
129 * range rmap lookup to make sure we get the correct owner/offset.
131 if (info->is_shared) {
132 error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno,
133 owner, offset, rflags, rmap, &has_rmap);
134 if (!xchk_should_check_xref(info->sc, &error,
135 &info->sc->sa.rmap_cur))
141 * Otherwise, use the (faster) regular lookup.
143 error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno, 0, owner,
144 offset, rflags, &has_rmap);
145 if (!xchk_should_check_xref(info->sc, &error,
146 &info->sc->sa.rmap_cur))
151 error = xfs_rmap_get_rec(info->sc->sa.rmap_cur, rmap, &has_rmap);
152 if (!xchk_should_check_xref(info->sc, &error,
153 &info->sc->sa.rmap_cur))
158 xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
163 /* Make sure that we have rmapbt records for this extent. */
166 struct xchk_bmap_info *info,
167 struct xfs_bmbt_irec *irec,
170 struct xfs_rmap_irec rmap;
171 unsigned long long rmap_end;
174 if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm))
177 if (info->whichfork == XFS_COW_FORK)
178 owner = XFS_RMAP_OWN_COW;
180 owner = info->sc->ip->i_ino;
182 /* Find the rmap record for this irec. */
183 if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap))
186 /* Check the rmap. */
187 rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
188 if (rmap.rm_startblock > agbno ||
189 agbno + irec->br_blockcount > rmap_end)
190 xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
194 * Check the logical offsets if applicable. CoW staging extents
195 * don't track logical offsets since the mappings only exist in
198 if (info->whichfork != XFS_COW_FORK) {
199 rmap_end = (unsigned long long)rmap.rm_offset +
201 if (rmap.rm_offset > irec->br_startoff ||
202 irec->br_startoff + irec->br_blockcount > rmap_end)
203 xchk_fblock_xref_set_corrupt(info->sc,
204 info->whichfork, irec->br_startoff);
207 if (rmap.rm_owner != owner)
208 xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
212 * Check for discrepancies between the unwritten flag in the irec and
213 * the rmap. Note that the (in-memory) CoW fork distinguishes between
214 * unwritten and written extents, but we don't track that in the rmap
215 * records because the blocks are owned (on-disk) by the refcountbt,
216 * which doesn't track unwritten state.
218 if (owner != XFS_RMAP_OWN_COW &&
219 irec->br_state == XFS_EXT_UNWRITTEN &&
220 !(rmap.rm_flags & XFS_RMAP_UNWRITTEN))
221 xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
224 if (info->whichfork == XFS_ATTR_FORK &&
225 !(rmap.rm_flags & XFS_RMAP_ATTR_FORK))
226 xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
228 if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK)
229 xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
233 /* Cross-reference a single rtdev extent record. */
235 xchk_bmap_rt_iextent_xref(
236 struct xfs_inode *ip,
237 struct xchk_bmap_info *info,
238 struct xfs_bmbt_irec *irec)
240 xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
241 irec->br_blockcount);
244 /* Cross-reference a single datadev extent record. */
246 xchk_bmap_iextent_xref(
247 struct xfs_inode *ip,
248 struct xchk_bmap_info *info,
249 struct xfs_bmbt_irec *irec)
251 struct xfs_mount *mp = info->sc->mp;
257 agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
258 agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
259 len = irec->br_blockcount;
261 error = xchk_ag_init(info->sc, agno, &info->sc->sa);
262 if (!xchk_fblock_process_error(info->sc, info->whichfork,
263 irec->br_startoff, &error))
266 xchk_xref_is_used_space(info->sc, agbno, len);
267 xchk_xref_is_not_inode_chunk(info->sc, agbno, len);
268 xchk_bmap_xref_rmap(info, irec, agbno);
269 switch (info->whichfork) {
271 if (xfs_is_reflink_inode(info->sc->ip))
275 xchk_xref_is_not_shared(info->sc, agbno,
276 irec->br_blockcount);
279 xchk_xref_is_cow_staging(info->sc, agbno,
280 irec->br_blockcount);
284 xchk_ag_free(info->sc, &info->sc->sa);
288 * Directories and attr forks should never have blocks that can't be addressed
292 xchk_bmap_dirattr_extent(
293 struct xfs_inode *ip,
294 struct xchk_bmap_info *info,
295 struct xfs_bmbt_irec *irec)
297 struct xfs_mount *mp = ip->i_mount;
300 if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK)
303 if (!xfs_verify_dablk(mp, irec->br_startoff))
304 xchk_fblock_set_corrupt(info->sc, info->whichfork,
307 off = irec->br_startoff + irec->br_blockcount - 1;
308 if (!xfs_verify_dablk(mp, off))
309 xchk_fblock_set_corrupt(info->sc, info->whichfork, off);
312 /* Scrub a single extent record. */
315 struct xfs_inode *ip,
316 struct xchk_bmap_info *info,
317 struct xfs_bmbt_irec *irec)
319 struct xfs_mount *mp = info->sc->mp;
324 * Check for out-of-order extents. This record could have come
325 * from the incore list, for which there is no ordering check.
327 if (irec->br_startoff < info->lastoff)
328 xchk_fblock_set_corrupt(info->sc, info->whichfork,
331 xchk_bmap_dirattr_extent(ip, info, irec);
333 /* There should never be a "hole" extent in either extent list. */
334 if (irec->br_startblock == HOLESTARTBLOCK)
335 xchk_fblock_set_corrupt(info->sc, info->whichfork,
339 * Check for delalloc extents. We never iterate the ones in the
340 * in-core extent scan, and we should never see these in the bmbt.
342 if (isnullstartblock(irec->br_startblock))
343 xchk_fblock_set_corrupt(info->sc, info->whichfork,
346 /* Make sure the extent points to a valid place. */
347 if (irec->br_blockcount > MAXEXTLEN)
348 xchk_fblock_set_corrupt(info->sc, info->whichfork,
350 if (irec->br_startblock + irec->br_blockcount <= irec->br_startblock)
351 xchk_fblock_set_corrupt(info->sc, info->whichfork,
353 end = irec->br_startblock + irec->br_blockcount - 1;
355 (!xfs_verify_rtbno(mp, irec->br_startblock) ||
356 !xfs_verify_rtbno(mp, end)))
357 xchk_fblock_set_corrupt(info->sc, info->whichfork,
360 (!xfs_verify_fsbno(mp, irec->br_startblock) ||
361 !xfs_verify_fsbno(mp, end) ||
362 XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
363 XFS_FSB_TO_AGNO(mp, end)))
364 xchk_fblock_set_corrupt(info->sc, info->whichfork,
367 /* We don't allow unwritten extents on attr forks. */
368 if (irec->br_state == XFS_EXT_UNWRITTEN &&
369 info->whichfork == XFS_ATTR_FORK)
370 xchk_fblock_set_corrupt(info->sc, info->whichfork,
373 if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
377 xchk_bmap_rt_iextent_xref(ip, info, irec);
379 xchk_bmap_iextent_xref(ip, info, irec);
381 info->lastoff = irec->br_startoff + irec->br_blockcount;
385 /* Scrub a bmbt record. */
388 struct xchk_btree *bs,
389 union xfs_btree_rec *rec)
391 struct xfs_bmbt_irec irec;
392 struct xfs_bmbt_irec iext_irec;
393 struct xfs_iext_cursor icur;
394 struct xchk_bmap_info *info = bs->private;
395 struct xfs_inode *ip = bs->cur->bc_ino.ip;
396 struct xfs_buf *bp = NULL;
397 struct xfs_btree_block *block;
398 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, info->whichfork);
403 * Check the owners of the btree blocks up to the level below
404 * the root since the verifiers don't do that.
406 if (xfs_sb_version_hascrc(&bs->cur->bc_mp->m_sb) &&
407 bs->cur->bc_ptrs[0] == 1) {
408 for (i = 0; i < bs->cur->bc_nlevels - 1; i++) {
409 block = xfs_btree_get_block(bs->cur, i, &bp);
410 owner = be64_to_cpu(block->bb_u.l.bb_owner);
411 if (owner != ip->i_ino)
412 xchk_fblock_set_corrupt(bs->sc,
418 * Check that the incore extent tree contains an extent that matches
419 * this one exactly. We validate those cached bmaps later, so we don't
420 * need to check them here. If the incore extent tree was just loaded
421 * from disk by the scrubber, we assume that its contents match what's
422 * on disk (we still hold the ILOCK) and skip the equivalence check.
424 if (!info->was_loaded)
427 xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
428 if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur,
430 irec.br_startoff != iext_irec.br_startoff ||
431 irec.br_startblock != iext_irec.br_startblock ||
432 irec.br_blockcount != iext_irec.br_blockcount ||
433 irec.br_state != iext_irec.br_state)
434 xchk_fblock_set_corrupt(bs->sc, info->whichfork,
439 /* Scan the btree records. */
442 struct xfs_scrub *sc,
444 struct xchk_bmap_info *info)
446 struct xfs_owner_info oinfo;
447 struct xfs_ifork *ifp = XFS_IFORK_PTR(sc->ip, whichfork);
448 struct xfs_mount *mp = sc->mp;
449 struct xfs_inode *ip = sc->ip;
450 struct xfs_btree_cur *cur;
453 /* Load the incore bmap cache if it's not loaded. */
454 info->was_loaded = ifp->if_flags & XFS_IFEXTENTS;
455 if (!info->was_loaded) {
456 error = xfs_iread_extents(sc->tp, ip, whichfork);
457 if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
461 /* Check the btree structure. */
462 cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
463 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
464 error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
465 xfs_btree_del_cursor(cur, error);
470 struct xchk_bmap_check_rmap_info {
471 struct xfs_scrub *sc;
473 struct xfs_iext_cursor icur;
476 /* Can we find bmaps that fit this rmap? */
478 xchk_bmap_check_rmap(
479 struct xfs_btree_cur *cur,
480 struct xfs_rmap_irec *rec,
483 struct xfs_bmbt_irec irec;
484 struct xchk_bmap_check_rmap_info *sbcri = priv;
485 struct xfs_ifork *ifp;
486 struct xfs_scrub *sc = sbcri->sc;
489 /* Is this even the right fork? */
490 if (rec->rm_owner != sc->ip->i_ino)
492 if ((sbcri->whichfork == XFS_ATTR_FORK) ^
493 !!(rec->rm_flags & XFS_RMAP_ATTR_FORK))
495 if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
498 /* Now look up the bmbt record. */
499 ifp = XFS_IFORK_PTR(sc->ip, sbcri->whichfork);
501 xchk_fblock_set_corrupt(sc, sbcri->whichfork,
505 have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset,
506 &sbcri->icur, &irec);
508 xchk_fblock_set_corrupt(sc, sbcri->whichfork,
511 * bmap extent record lengths are constrained to 2^21 blocks in length
512 * because of space constraints in the on-disk metadata structure.
513 * However, rmap extent record lengths are constrained only by AG
514 * length, so we have to loop through the bmbt to make sure that the
515 * entire rmap is covered by bmbt records.
518 if (irec.br_startoff != rec->rm_offset)
519 xchk_fblock_set_corrupt(sc, sbcri->whichfork,
521 if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp,
522 cur->bc_ag.agno, rec->rm_startblock))
523 xchk_fblock_set_corrupt(sc, sbcri->whichfork,
525 if (irec.br_blockcount > rec->rm_blockcount)
526 xchk_fblock_set_corrupt(sc, sbcri->whichfork,
528 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
530 rec->rm_startblock += irec.br_blockcount;
531 rec->rm_offset += irec.br_blockcount;
532 rec->rm_blockcount -= irec.br_blockcount;
533 if (rec->rm_blockcount == 0)
535 have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec);
537 xchk_fblock_set_corrupt(sc, sbcri->whichfork,
542 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
547 /* Make sure each rmap has a corresponding bmbt entry. */
549 xchk_bmap_check_ag_rmaps(
550 struct xfs_scrub *sc,
554 struct xchk_bmap_check_rmap_info sbcri;
555 struct xfs_btree_cur *cur;
559 error = xfs_alloc_read_agf(sc->mp, sc->tp, agno, 0, &agf);
563 cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, agno);
570 sbcri.whichfork = whichfork;
571 error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri);
572 if (error == -ECANCELED)
575 xfs_btree_del_cursor(cur, error);
577 xfs_trans_brelse(sc->tp, agf);
581 /* Make sure each rmap has a corresponding bmbt entry. */
583 xchk_bmap_check_rmaps(
584 struct xfs_scrub *sc,
587 struct xfs_ifork *ifp = XFS_IFORK_PTR(sc->ip, whichfork);
592 if (!xfs_sb_version_hasrmapbt(&sc->mp->m_sb) ||
593 whichfork == XFS_COW_FORK ||
594 (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
597 /* Don't support realtime rmap checks yet. */
598 if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK)
601 ASSERT(XFS_IFORK_PTR(sc->ip, whichfork) != NULL);
604 * Only do this for complex maps that are in btree format, or for
605 * situations where we would seem to have a size but zero extents.
606 * The inode repair code can zap broken iforks, which means we have
607 * to flag this bmap as corrupt if there are rmaps that need to be
611 if (whichfork == XFS_DATA_FORK)
612 zero_size = i_size_read(VFS_I(sc->ip)) == 0;
616 if (ifp->if_format != XFS_DINODE_FMT_BTREE &&
617 (zero_size || ifp->if_nextents > 0))
620 for (agno = 0; agno < sc->mp->m_sb.sb_agcount; agno++) {
621 error = xchk_bmap_check_ag_rmaps(sc, whichfork, agno);
624 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
632 * Scrub an inode fork's block mappings.
634 * First we scan every record in every btree block, if applicable.
635 * Then we unconditionally scan the incore extent cache.
639 struct xfs_scrub *sc,
642 struct xfs_bmbt_irec irec;
643 struct xchk_bmap_info info = { NULL };
644 struct xfs_mount *mp = sc->mp;
645 struct xfs_inode *ip = sc->ip;
646 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
647 xfs_fileoff_t endoff;
648 struct xfs_iext_cursor icur;
651 /* Non-existent forks can be ignored. */
655 info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
656 info.whichfork = whichfork;
657 info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
662 /* No CoW forks on non-reflink inodes/filesystems. */
663 if (!xfs_is_reflink_inode(ip)) {
664 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
669 if (!xfs_sb_version_hasattr(&mp->m_sb) &&
670 !xfs_sb_version_hasattr2(&mp->m_sb))
671 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
674 ASSERT(whichfork == XFS_DATA_FORK);
678 /* Check the fork values */
679 switch (ifp->if_format) {
680 case XFS_DINODE_FMT_UUID:
681 case XFS_DINODE_FMT_DEV:
682 case XFS_DINODE_FMT_LOCAL:
683 /* No mappings to check. */
685 case XFS_DINODE_FMT_EXTENTS:
686 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
687 xchk_fblock_set_corrupt(sc, whichfork, 0);
691 case XFS_DINODE_FMT_BTREE:
692 if (whichfork == XFS_COW_FORK) {
693 xchk_fblock_set_corrupt(sc, whichfork, 0);
697 error = xchk_bmap_btree(sc, whichfork, &info);
702 xchk_fblock_set_corrupt(sc, whichfork, 0);
706 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
709 /* Find the offset of the last extent in the mapping. */
710 error = xfs_bmap_last_offset(ip, &endoff, whichfork);
711 if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
714 /* Scrub extent records. */
716 ifp = XFS_IFORK_PTR(ip, whichfork);
717 for_each_xfs_iext(ifp, &icur, &irec) {
718 if (xchk_should_terminate(sc, &error) ||
719 (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
721 if (isnullstartblock(irec.br_startblock))
723 if (irec.br_startoff >= endoff) {
724 xchk_fblock_set_corrupt(sc, whichfork,
728 error = xchk_bmap_iextent(ip, &info, &irec);
733 error = xchk_bmap_check_rmaps(sc, whichfork);
734 if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error))
740 /* Scrub an inode's data fork. */
743 struct xfs_scrub *sc)
745 return xchk_bmap(sc, XFS_DATA_FORK);
748 /* Scrub an inode's attr fork. */
751 struct xfs_scrub *sc)
753 return xchk_bmap(sc, XFS_ATTR_FORK);
756 /* Scrub an inode's CoW fork. */
759 struct xfs_scrub *sc)
761 if (!xfs_is_reflink_inode(sc->ip))
764 return xchk_bmap(sc, XFS_COW_FORK);