1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2017-2023 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_icache.h"
17 #include "xfs_alloc.h"
18 #include "xfs_alloc_btree.h"
19 #include "xfs_ialloc.h"
20 #include "xfs_ialloc_btree.h"
21 #include "xfs_refcount_btree.h"
23 #include "xfs_rmap_btree.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_da_format.h"
27 #include "xfs_da_btree.h"
29 #include "xfs_reflink.h"
31 #include "scrub/scrub.h"
32 #include "scrub/common.h"
33 #include "scrub/trace.h"
34 #include "scrub/repair.h"
35 #include "scrub/health.h"
37 /* Common code for the metadata scrubbers. */
40 * Handling operational errors.
42 * The *_process_error() family of functions are used to process error return
43 * codes from functions called as part of a scrub operation.
45 * If there's no error, we return true to tell the caller that it's ok
46 * to move on to the next check in its list.
48 * For non-verifier errors (e.g. ENOMEM) we return false to tell the
49 * caller that something bad happened, and we preserve *error so that
50 * the caller can return the *error up the stack to userspace.
52 * Verifier errors (EFSBADCRC/EFSCORRUPTED) are recorded by setting
53 * OFLAG_CORRUPT in sm_flags and the *error is cleared. In other words,
54 * we track verifier errors (and failed scrub checks) via OFLAG_CORRUPT,
55 * not via return codes. We return false to tell the caller that
56 * something bad happened. Since the error has been cleared, the caller
57 * will (presumably) return that zero and scrubbing will move on to
60 * ftrace can be used to record the precise metadata location and the
61 * approximate code location of the failed operation.
64 /* Check for operational errors. */
79 /* Used to restart an op with deadlock avoidance. */
80 trace_xchk_deadlock_retry(
81 sc->ip ? sc->ip : XFS_I(file_inode(sc->file)),
86 /* Note the badness but don't abort. */
87 sc->sm->sm_flags |= errflag;
91 trace_xchk_op_error(sc, agno, bno, *error,
100 struct xfs_scrub *sc,
105 return __xchk_process_error(sc, agno, bno, error,
106 XFS_SCRUB_OFLAG_CORRUPT, __return_address);
110 xchk_xref_process_error(
111 struct xfs_scrub *sc,
116 return __xchk_process_error(sc, agno, bno, error,
117 XFS_SCRUB_OFLAG_XFAIL, __return_address);
120 /* Check for operational errors for a file offset. */
122 __xchk_fblock_process_error(
123 struct xfs_scrub *sc,
125 xfs_fileoff_t offset,
135 /* Used to restart an op with deadlock avoidance. */
136 trace_xchk_deadlock_retry(sc->ip, sc->sm, *error);
140 /* Note the badness but don't abort. */
141 sc->sm->sm_flags |= errflag;
145 trace_xchk_file_op_error(sc, whichfork, offset, *error,
153 xchk_fblock_process_error(
154 struct xfs_scrub *sc,
156 xfs_fileoff_t offset,
159 return __xchk_fblock_process_error(sc, whichfork, offset, error,
160 XFS_SCRUB_OFLAG_CORRUPT, __return_address);
164 xchk_fblock_xref_process_error(
165 struct xfs_scrub *sc,
167 xfs_fileoff_t offset,
170 return __xchk_fblock_process_error(sc, whichfork, offset, error,
171 XFS_SCRUB_OFLAG_XFAIL, __return_address);
175 * Handling scrub corruption/optimization/warning checks.
177 * The *_set_{corrupt,preen,warning}() family of functions are used to
178 * record the presence of metadata that is incorrect (corrupt), could be
179 * optimized somehow (preen), or should be flagged for administrative
180 * review but is not incorrect (warn).
182 * ftrace can be used to record the precise metadata location and
183 * approximate code location of the failed check.
186 /* Record a block which could be optimized. */
188 xchk_block_set_preen(
189 struct xfs_scrub *sc,
192 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
193 trace_xchk_block_preen(sc, xfs_buf_daddr(bp), __return_address);
197 * Record an inode which could be optimized. The trace data will
198 * include the block given by bp if bp is given; otherwise it will use
199 * the block location of the inode record itself.
203 struct xfs_scrub *sc,
206 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
207 trace_xchk_ino_preen(sc, ino, __return_address);
210 /* Record something being wrong with the filesystem primary superblock. */
213 struct xfs_scrub *sc)
215 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
216 trace_xchk_fs_error(sc, 0, __return_address);
219 /* Record a corrupt block. */
221 xchk_block_set_corrupt(
222 struct xfs_scrub *sc,
225 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
226 trace_xchk_block_error(sc, xfs_buf_daddr(bp), __return_address);
229 /* Record a corruption while cross-referencing. */
231 xchk_block_xref_set_corrupt(
232 struct xfs_scrub *sc,
235 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
236 trace_xchk_block_error(sc, xfs_buf_daddr(bp), __return_address);
240 * Record a corrupt inode. The trace data will include the block given
241 * by bp if bp is given; otherwise it will use the block location of the
242 * inode record itself.
245 xchk_ino_set_corrupt(
246 struct xfs_scrub *sc,
249 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
250 trace_xchk_ino_error(sc, ino, __return_address);
253 /* Record a corruption while cross-referencing with an inode. */
255 xchk_ino_xref_set_corrupt(
256 struct xfs_scrub *sc,
259 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
260 trace_xchk_ino_error(sc, ino, __return_address);
263 /* Record corruption in a block indexed by a file fork. */
265 xchk_fblock_set_corrupt(
266 struct xfs_scrub *sc,
268 xfs_fileoff_t offset)
270 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
271 trace_xchk_fblock_error(sc, whichfork, offset, __return_address);
274 /* Record a corruption while cross-referencing a fork block. */
276 xchk_fblock_xref_set_corrupt(
277 struct xfs_scrub *sc,
279 xfs_fileoff_t offset)
281 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
282 trace_xchk_fblock_error(sc, whichfork, offset, __return_address);
286 * Warn about inodes that need administrative review but is not
290 xchk_ino_set_warning(
291 struct xfs_scrub *sc,
294 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
295 trace_xchk_ino_warning(sc, ino, __return_address);
298 /* Warn about a block indexed by a file fork that needs review. */
300 xchk_fblock_set_warning(
301 struct xfs_scrub *sc,
303 xfs_fileoff_t offset)
305 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
306 trace_xchk_fblock_warning(sc, whichfork, offset, __return_address);
309 /* Signal an incomplete scrub. */
312 struct xfs_scrub *sc)
314 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_INCOMPLETE;
315 trace_xchk_incomplete(sc, __return_address);
319 * rmap scrubbing -- compute the number of blocks with a given owner,
320 * at least according to the reverse mapping data.
323 struct xchk_rmap_ownedby_info {
324 const struct xfs_owner_info *oinfo;
325 xfs_filblks_t *blocks;
329 xchk_count_rmap_ownedby_irec(
330 struct xfs_btree_cur *cur,
331 const struct xfs_rmap_irec *rec,
334 struct xchk_rmap_ownedby_info *sroi = priv;
338 irec_attr = rec->rm_flags & XFS_RMAP_ATTR_FORK;
339 oinfo_attr = sroi->oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK;
341 if (rec->rm_owner != sroi->oinfo->oi_owner)
344 if (XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) || irec_attr == oinfo_attr)
345 (*sroi->blocks) += rec->rm_blockcount;
351 * Calculate the number of blocks the rmap thinks are owned by something.
352 * The caller should pass us an rmapbt cursor.
355 xchk_count_rmap_ownedby_ag(
356 struct xfs_scrub *sc,
357 struct xfs_btree_cur *cur,
358 const struct xfs_owner_info *oinfo,
359 xfs_filblks_t *blocks)
361 struct xchk_rmap_ownedby_info sroi = {
367 return xfs_rmap_query_all(cur, xchk_count_rmap_ownedby_irec,
374 * These helpers facilitate locking an allocation group's header
375 * buffers, setting up cursors for all btrees that are present, and
376 * cleaning everything up once we're through.
379 /* Decide if we want to return an AG header read failure. */
381 want_ag_read_header_failure(
382 struct xfs_scrub *sc,
385 /* Return all AG header read failures when scanning btrees. */
386 if (sc->sm->sm_type != XFS_SCRUB_TYPE_AGF &&
387 sc->sm->sm_type != XFS_SCRUB_TYPE_AGFL &&
388 sc->sm->sm_type != XFS_SCRUB_TYPE_AGI)
391 * If we're scanning a given type of AG header, we only want to
392 * see read failures from that specific header. We'd like the
393 * other headers to cross-check them, but this isn't required.
395 if (sc->sm->sm_type == type)
401 * Grab the AG header buffers for the attached perag structure.
403 * The headers should be released by xchk_ag_free, but as a fail safe we attach
404 * all the buffers we grab to the scrub transaction so they'll all be freed
408 xchk_perag_read_headers(
409 struct xfs_scrub *sc,
414 error = xfs_ialloc_read_agi(sa->pag, sc->tp, &sa->agi_bp);
415 if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGI))
418 error = xfs_alloc_read_agf(sa->pag, sc->tp, 0, &sa->agf_bp);
419 if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGF))
426 * Grab the AG headers for the attached perag structure and wait for pending
430 xchk_perag_drain_and_lock(
431 struct xfs_scrub *sc)
433 struct xchk_ag *sa = &sc->sa;
436 ASSERT(sa->pag != NULL);
437 ASSERT(sa->agi_bp == NULL);
438 ASSERT(sa->agf_bp == NULL);
441 if (xchk_should_terminate(sc, &error))
444 error = xchk_perag_read_headers(sc, sa);
449 * If we've grabbed an inode for scrubbing then we assume that
450 * holding its ILOCK will suffice to coordinate with any intent
451 * chains involving this inode.
457 * Decide if this AG is quiet enough for all metadata to be
458 * consistent with each other. XFS allows the AG header buffer
459 * locks to cycle across transaction rolls while processing
460 * chains of deferred ops, which means that there could be
461 * other threads in the middle of processing a chain of
462 * deferred ops. For regular operations we are careful about
463 * ordering operations to prevent collisions between threads
464 * (which is why we don't need a per-AG lock), but scrub and
465 * repair have to serialize against chained operations.
467 * We just locked all the AG headers buffers; now take a look
468 * to see if there are any intents in progress. If there are,
469 * drop the AG headers and wait for the intents to drain.
470 * Since we hold all the AG header locks for the duration of
471 * the scrub, this is the only time we have to sample the
472 * intents counter; any threads increasing it after this point
473 * can't possibly be in the middle of a chain of AG metadata
476 * Obviously, this should be slanted against scrub and in favor
477 * of runtime threads.
479 if (!xfs_perag_intent_busy(sa->pag))
483 xfs_trans_brelse(sc->tp, sa->agf_bp);
488 xfs_trans_brelse(sc->tp, sa->agi_bp);
492 if (!(sc->flags & XCHK_FSGATES_DRAIN))
494 error = xfs_perag_intent_drain(sa->pag);
495 if (error == -ERESTARTSYS)
503 * Grab the per-AG structure, grab all AG header buffers, and wait until there
504 * aren't any pending intents. Returns -ENOENT if we can't grab the perag
508 xchk_ag_read_headers(
509 struct xfs_scrub *sc,
513 struct xfs_mount *mp = sc->mp;
516 sa->pag = xfs_perag_get(mp, agno);
520 return xchk_perag_drain_and_lock(sc);
523 /* Release all the AG btree cursors. */
529 xfs_btree_del_cursor(sa->refc_cur, XFS_BTREE_ERROR);
531 xfs_btree_del_cursor(sa->rmap_cur, XFS_BTREE_ERROR);
533 xfs_btree_del_cursor(sa->fino_cur, XFS_BTREE_ERROR);
535 xfs_btree_del_cursor(sa->ino_cur, XFS_BTREE_ERROR);
537 xfs_btree_del_cursor(sa->cnt_cur, XFS_BTREE_ERROR);
539 xfs_btree_del_cursor(sa->bno_cur, XFS_BTREE_ERROR);
549 /* Initialize all the btree cursors for an AG. */
552 struct xfs_scrub *sc,
555 struct xfs_mount *mp = sc->mp;
558 xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_BNO)) {
559 /* Set up a bnobt cursor for cross-referencing. */
560 sa->bno_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
561 sa->pag, XFS_BTNUM_BNO);
565 xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_CNT)) {
566 /* Set up a cntbt cursor for cross-referencing. */
567 sa->cnt_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
568 sa->pag, XFS_BTNUM_CNT);
571 /* Set up a inobt cursor for cross-referencing. */
573 xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_INO)) {
574 sa->ino_cur = xfs_inobt_init_cursor(sa->pag, sc->tp, sa->agi_bp,
578 /* Set up a finobt cursor for cross-referencing. */
579 if (sa->agi_bp && xfs_has_finobt(mp) &&
580 xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_FINO)) {
581 sa->fino_cur = xfs_inobt_init_cursor(sa->pag, sc->tp, sa->agi_bp,
585 /* Set up a rmapbt cursor for cross-referencing. */
586 if (sa->agf_bp && xfs_has_rmapbt(mp) &&
587 xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_RMAP)) {
588 sa->rmap_cur = xfs_rmapbt_init_cursor(mp, sc->tp, sa->agf_bp,
592 /* Set up a refcountbt cursor for cross-referencing. */
593 if (sa->agf_bp && xfs_has_reflink(mp) &&
594 xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_REFC)) {
595 sa->refc_cur = xfs_refcountbt_init_cursor(mp, sc->tp,
596 sa->agf_bp, sa->pag);
600 /* Release the AG header context and btree cursors. */
603 struct xfs_scrub *sc,
606 xchk_ag_btcur_free(sa);
608 xfs_trans_brelse(sc->tp, sa->agf_bp);
612 xfs_trans_brelse(sc->tp, sa->agi_bp);
616 xfs_perag_put(sa->pag);
622 * For scrub, grab the perag structure, the AGI, and the AGF headers, in that
623 * order. Locking order requires us to get the AGI before the AGF. We use the
624 * transaction to avoid deadlocking on crosslinked metadata buffers; either the
625 * caller passes one in (bmap scrub) or we have to create a transaction
626 * ourselves. Returns ENOENT if the perag struct cannot be grabbed.
630 struct xfs_scrub *sc,
636 error = xchk_ag_read_headers(sc, agno, sa);
640 xchk_ag_btcur_init(sc, sa);
644 /* Per-scrubber setup functions */
648 struct xfs_scrub *sc)
650 xfs_trans_cancel(sc->tp);
655 * Grab an empty transaction so that we can re-grab locked buffers if
656 * one of our btrees turns out to be cyclic.
658 * If we're going to repair something, we need to ask for the largest possible
659 * log reservation so that we can handle the worst case scenario for metadata
660 * updates while rebuilding a metadata item. We also need to reserve as many
661 * blocks in the head transaction as we think we're going to need to rebuild
662 * the metadata object.
666 struct xfs_scrub *sc,
669 if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
670 return xfs_trans_alloc(sc->mp, &M_RES(sc->mp)->tr_itruncate,
671 resblks, 0, 0, &sc->tp);
673 return xfs_trans_alloc_empty(sc->mp, &sc->tp);
676 /* Set us up with a transaction and an empty context. */
679 struct xfs_scrub *sc)
683 resblks = xrep_calc_ag_resblks(sc);
684 return xchk_trans_alloc(sc, resblks);
687 /* Set us up with AG headers and btree cursors. */
690 struct xfs_scrub *sc,
693 struct xfs_mount *mp = sc->mp;
697 * If the caller asks us to checkpont the log, do so. This
698 * expensive operation should be performed infrequently and only
699 * as a last resort. Any caller that sets force_log should
700 * document why they need to do so.
703 error = xchk_checkpoint_log(mp);
708 error = xchk_setup_fs(sc);
712 return xchk_ag_init(sc, sc->sm->sm_agno, &sc->sa);
715 /* Push everything out of the log onto disk. */
718 struct xfs_mount *mp)
722 error = xfs_log_force(mp, XFS_LOG_SYNC);
725 xfs_ail_push_all_sync(mp->m_ail);
729 /* Verify that an inode is allocated ondisk, then return its cached inode. */
732 struct xfs_scrub *sc,
734 struct xfs_inode **ipp)
736 return xfs_iget(sc->mp, sc->tp, inum, XFS_IGET_UNTRUSTED, 0, ipp);
740 * Try to grab an inode in a manner that avoids races with physical inode
741 * allocation. If we can't, return the locked AGI buffer so that the caller
742 * can single-step the loading process to see where things went wrong.
743 * Callers must have a valid scrub transaction.
745 * If the iget succeeds, return 0, a NULL AGI, and the inode.
747 * If the iget fails, return the error, the locked AGI, and a NULL inode. This
748 * can include -EINVAL and -ENOENT for invalid inode numbers or inodes that are
749 * no longer allocated; or any other corruption or runtime error.
751 * If the AGI read fails, return the error, a NULL AGI, and NULL inode.
753 * If a fatal signal is pending, return -EINTR, a NULL AGI, and a NULL inode.
757 struct xfs_scrub *sc,
759 struct xfs_buf **agi_bpp,
760 struct xfs_inode **ipp)
762 struct xfs_mount *mp = sc->mp;
763 struct xfs_trans *tp = sc->tp;
764 struct xfs_perag *pag;
767 ASSERT(sc->tp != NULL);
774 if (xchk_should_terminate(sc, &error))
778 * Attach the AGI buffer to the scrub transaction to avoid deadlocks
779 * in the iget cache miss path.
781 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
782 error = xfs_ialloc_read_agi(pag, tp, agi_bpp);
787 error = xfs_iget(mp, tp, inum,
788 XFS_IGET_NORETRY | XFS_IGET_UNTRUSTED, 0, ipp);
789 if (error == -EAGAIN) {
791 * The inode may be in core but temporarily unavailable and may
792 * require the AGI buffer before it can be returned. Drop the
793 * AGI buffer and retry the lookup.
795 * Incore lookup will fail with EAGAIN on a cache hit if the
796 * inode is queued to the inactivation list. The inactivation
797 * worker may remove the inode from the unlinked list and hence
800 * Hence xchk_iget_agi() needs to drop the AGI lock on EAGAIN
801 * to allow inodegc to make progress and move the inode to
802 * IRECLAIMABLE state where xfs_iget will be able to return it
803 * again if it can lock the inode.
805 xfs_trans_brelse(tp, *agi_bpp);
812 /* We got the inode, so we can release the AGI. */
813 ASSERT(*ipp != NULL);
814 xfs_trans_brelse(tp, *agi_bpp);
819 /* Install an inode that we opened by handle for scrubbing. */
821 xchk_install_handle_inode(
822 struct xfs_scrub *sc,
823 struct xfs_inode *ip)
825 if (VFS_I(ip)->i_generation != sc->sm->sm_gen) {
835 * In preparation to scrub metadata structures that hang off of an inode,
836 * grab either the inode referenced in the scrub control structure or the
837 * inode passed in. If the inumber does not reference an allocated inode
838 * record, the function returns ENOENT to end the scrub early. The inode
842 xchk_iget_for_scrubbing(
843 struct xfs_scrub *sc)
845 struct xfs_imap imap;
846 struct xfs_mount *mp = sc->mp;
847 struct xfs_perag *pag;
848 struct xfs_buf *agi_bp;
849 struct xfs_inode *ip_in = XFS_I(file_inode(sc->file));
850 struct xfs_inode *ip = NULL;
851 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, sc->sm->sm_ino);
854 ASSERT(sc->tp == NULL);
856 /* We want to scan the inode we already had opened. */
857 if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
862 /* Reject internal metadata files and obviously bad inode numbers. */
863 if (xfs_internal_inum(mp, sc->sm->sm_ino))
865 if (!xfs_verify_ino(sc->mp, sc->sm->sm_ino))
868 /* Try a regular untrusted iget. */
869 error = xchk_iget(sc, sc->sm->sm_ino, &ip);
871 return xchk_install_handle_inode(sc, ip);
872 if (error == -ENOENT)
874 if (error != -EINVAL)
878 * EINVAL with IGET_UNTRUSTED probably means one of several things:
879 * userspace gave us an inode number that doesn't correspond to fs
880 * space; the inode btree lacks a record for this inode; or there is a
881 * record, and it says this inode is free.
883 * We want to look up this inode in the inobt to distinguish two
884 * scenarios: (1) the inobt says the inode is free, in which case
885 * there's nothing to do; and (2) the inobt says the inode is
886 * allocated, but loading it failed due to corruption.
888 * Allocate a transaction and grab the AGI to prevent inobt activity
889 * in this AG. Retry the iget in case someone allocated a new inode
890 * after the first iget failed.
892 error = xchk_trans_alloc(sc, 0);
896 error = xchk_iget_agi(sc, sc->sm->sm_ino, &agi_bp, &ip);
898 /* Actually got the inode, so install it. */
899 xchk_trans_cancel(sc);
900 return xchk_install_handle_inode(sc, ip);
902 if (error == -ENOENT)
904 if (error != -EINVAL)
907 /* Ensure that we have protected against inode allocation/freeing. */
908 if (agi_bp == NULL) {
909 ASSERT(agi_bp != NULL);
915 * Untrusted iget failed a second time. Let's try an inobt lookup.
916 * If the inobt thinks this the inode neither can exist inside the
917 * filesystem nor is allocated, return ENOENT to signal that the check
920 * If the lookup returns corruption, we'll mark this inode corrupt and
921 * exit to userspace. There's little chance of fixing anything until
922 * the inobt is straightened out, but there's nothing we can do here.
924 * If the lookup encounters any other error, exit to userspace.
926 * If the lookup succeeds, something else must be very wrong in the fs
927 * such that setting up the incore inode failed in some strange way.
928 * Treat those as corruptions.
930 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sc->sm->sm_ino));
932 error = -EFSCORRUPTED;
936 error = xfs_imap(pag, sc->tp, sc->sm->sm_ino, &imap,
939 if (error == -EINVAL || error == -ENOENT)
942 error = -EFSCORRUPTED;
945 xchk_trans_cancel(sc);
947 trace_xchk_op_error(sc, agno, XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
948 error, __return_address);
951 /* The file is gone, so there's nothing to check. */
952 xchk_trans_cancel(sc);
956 /* Release an inode, possibly dropping it in the process. */
959 struct xfs_scrub *sc,
960 struct xfs_inode *ip)
962 if (current->journal_info != NULL) {
963 ASSERT(current->journal_info == sc->tp);
966 * If we are in a transaction, we /cannot/ drop the inode
967 * ourselves, because the VFS will trigger writeback, which
968 * can require a transaction. Clear DONTCACHE to force the
969 * inode to the LRU, where someone else can take care of
972 * Note that when we grabbed our reference to the inode, it
973 * could have had an active ref and DONTCACHE set if a sysadmin
974 * is trying to coerce a change in file access mode. icache
975 * hits do not clear DONTCACHE, so we must do it here.
977 spin_lock(&VFS_I(ip)->i_lock);
978 VFS_I(ip)->i_state &= ~I_DONTCACHE;
979 spin_unlock(&VFS_I(ip)->i_lock);
980 } else if (atomic_read(&VFS_I(ip)->i_count) == 1) {
982 * If this is the last reference to the inode and the caller
983 * permits it, set DONTCACHE to avoid thrashing.
985 d_mark_dontcache(VFS_I(ip));
992 * Set us up to scrub metadata mapped by a file's fork. Callers must not use
993 * this to operate on user-accessible regular file data because the MMAPLOCK is
997 xchk_setup_inode_contents(
998 struct xfs_scrub *sc,
999 unsigned int resblks)
1003 error = xchk_iget_for_scrubbing(sc);
1007 /* Lock the inode so the VFS cannot touch this file. */
1008 sc->ilock_flags = XFS_IOLOCK_EXCL;
1009 xfs_ilock(sc->ip, sc->ilock_flags);
1011 error = xchk_trans_alloc(sc, resblks);
1014 sc->ilock_flags |= XFS_ILOCK_EXCL;
1015 xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
1018 /* scrub teardown will unlock and release the inode for us */
1023 * Predicate that decides if we need to evaluate the cross-reference check.
1024 * If there was an error accessing the cross-reference btree, just delete
1025 * the cursor and skip the check.
1028 xchk_should_check_xref(
1029 struct xfs_scrub *sc,
1031 struct xfs_btree_cur **curpp)
1033 /* No point in xref if we already know we're corrupt. */
1034 if (xchk_skip_xref(sc->sm))
1041 /* If we've already given up on xref, just bail out. */
1045 /* xref error, delete cursor and bail out. */
1046 xfs_btree_del_cursor(*curpp, XFS_BTREE_ERROR);
1050 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
1051 trace_xchk_xref_error(sc, *error, __return_address);
1054 * Errors encountered during cross-referencing with another
1055 * data structure should not cause this scrubber to abort.
1061 /* Run the structure verifiers on in-memory buffers to detect bad memory. */
1063 xchk_buffer_recheck(
1064 struct xfs_scrub *sc,
1069 if (bp->b_ops == NULL) {
1070 xchk_block_set_corrupt(sc, bp);
1073 if (bp->b_ops->verify_struct == NULL) {
1074 xchk_set_incomplete(sc);
1077 fa = bp->b_ops->verify_struct(bp);
1080 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
1081 trace_xchk_block_error(sc, xfs_buf_daddr(bp), fa);
1085 xchk_metadata_inode_subtype(
1086 struct xfs_scrub *sc,
1087 unsigned int scrub_type)
1089 __u32 smtype = sc->sm->sm_type;
1092 sc->sm->sm_type = scrub_type;
1094 switch (scrub_type) {
1095 case XFS_SCRUB_TYPE_INODE:
1096 error = xchk_inode(sc);
1098 case XFS_SCRUB_TYPE_BMBTD:
1099 error = xchk_bmap_data(sc);
1103 error = -EFSCORRUPTED;
1107 sc->sm->sm_type = smtype;
1112 * Scrub the attr/data forks of a metadata inode. The metadata inode must be
1113 * pointed to by sc->ip and the ILOCK must be held.
1116 xchk_metadata_inode_forks(
1117 struct xfs_scrub *sc)
1122 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
1125 /* Check the inode record. */
1126 error = xchk_metadata_inode_subtype(sc, XFS_SCRUB_TYPE_INODE);
1127 if (error || (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
1130 /* Metadata inodes don't live on the rt device. */
1131 if (sc->ip->i_diflags & XFS_DIFLAG_REALTIME) {
1132 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1136 /* They should never participate in reflink. */
1137 if (xfs_is_reflink_inode(sc->ip)) {
1138 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1142 /* They also should never have extended attributes. */
1143 if (xfs_inode_hasattr(sc->ip)) {
1144 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1148 /* Invoke the data fork scrubber. */
1149 error = xchk_metadata_inode_subtype(sc, XFS_SCRUB_TYPE_BMBTD);
1150 if (error || (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
1153 /* Look for incorrect shared blocks. */
1154 if (xfs_has_reflink(sc->mp)) {
1155 error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
1157 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
1161 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1168 * Enable filesystem hooks (i.e. runtime code patching) before starting a scrub
1169 * operation. Callers must not hold any locks that intersect with the CPU
1170 * hotplug lock (e.g. writeback locks) because code patching must halt the CPUs
1171 * to change kernel code.
1174 xchk_fsgates_enable(
1175 struct xfs_scrub *sc,
1176 unsigned int scrub_fsgates)
1178 ASSERT(!(scrub_fsgates & ~XCHK_FSGATES_ALL));
1179 ASSERT(!(sc->flags & scrub_fsgates));
1181 trace_xchk_fsgates_enable(sc, scrub_fsgates);
1183 if (scrub_fsgates & XCHK_FSGATES_DRAIN)
1184 xfs_drain_wait_enable();
1186 sc->flags |= scrub_fsgates;