1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2018 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
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"
16 #include "xfs_inode.h"
17 #include "xfs_alloc.h"
18 #include "xfs_alloc_btree.h"
19 #include "xfs_ialloc.h"
20 #include "xfs_ialloc_btree.h"
22 #include "xfs_rmap_btree.h"
23 #include "xfs_refcount_btree.h"
24 #include "xfs_extent_busy.h"
25 #include "xfs_ag_resv.h"
26 #include "xfs_quota.h"
27 #include "scrub/scrub.h"
28 #include "scrub/common.h"
29 #include "scrub/trace.h"
30 #include "scrub/repair.h"
31 #include "scrub/bitmap.h"
34 * Attempt to repair some metadata, if the metadata is corrupt and userspace
35 * told us to fix it. This function returns -EAGAIN to mean "re-run scrub",
36 * and will set *fixed to true if it thinks it repaired anything.
44 trace_xrep_attempt(XFS_I(file_inode(sc->file)), sc->sm, error);
46 xchk_ag_btcur_free(&sc->sa);
48 /* Repair whatever's broken. */
49 ASSERT(sc->ops->repair);
50 error = sc->ops->repair(sc);
51 trace_xrep_done(XFS_I(file_inode(sc->file)), sc->sm, error);
55 * Repair succeeded. Commit the fixes and perform a second
56 * scrub so that we can tell userspace if we fixed the problem.
58 sc->sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT;
59 sc->flags |= XREP_ALREADY_FIXED;
63 /* Tell the caller to try again having grabbed all the locks. */
64 if (!(sc->flags & XCHK_TRY_HARDER)) {
65 sc->flags |= XCHK_TRY_HARDER;
69 * We tried harder but still couldn't grab all the resources
70 * we needed to fix it. The corruption has not been fixed,
71 * so report back to userspace.
80 * Complain about unfixable problems in the filesystem. We don't log
81 * corruptions when IFLAG_REPAIR wasn't set on the assumption that the driver
82 * program is xfs_scrub, which will call back with IFLAG_REPAIR set if the
83 * administrator isn't running xfs_scrub in no-repairs mode.
85 * Use this helper function because _ratelimited silently declares a static
86 * structure to track rate limiting information.
92 xfs_alert_ratelimited(mp,
93 "Corruption not fixed during online repair. Unmount and run xfs_repair.");
97 * Repair probe -- userspace uses this to probe if we're willing to repair a
102 struct xfs_scrub *sc)
106 if (xchk_should_terminate(sc, &error))
113 * Roll a transaction, keeping the AG headers locked and reinitializing
118 struct xfs_scrub *sc)
122 /* Keep the AG header buffers locked so we can keep going. */
124 xfs_trans_bhold(sc->tp, sc->sa.agi_bp);
126 xfs_trans_bhold(sc->tp, sc->sa.agf_bp);
128 xfs_trans_bhold(sc->tp, sc->sa.agfl_bp);
131 * Roll the transaction. We still own the buffer and the buffer lock
132 * regardless of whether or not the roll succeeds. If the roll fails,
133 * the buffers will be released during teardown on our way out of the
134 * kernel. If it succeeds, we join them to the new transaction and
137 error = xfs_trans_roll(&sc->tp);
141 /* Join AG headers to the new transaction. */
143 xfs_trans_bjoin(sc->tp, sc->sa.agi_bp);
145 xfs_trans_bjoin(sc->tp, sc->sa.agf_bp);
147 xfs_trans_bjoin(sc->tp, sc->sa.agfl_bp);
153 * Does the given AG have enough space to rebuild a btree? Neither AG
154 * reservation can be critical, and we must have enough space (factoring
155 * in AG reservations) to construct a whole btree.
159 struct xfs_perag *pag,
160 xfs_extlen_t nr_blocks,
161 enum xfs_ag_resv_type type)
163 return !xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) &&
164 !xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA) &&
165 pag->pagf_freeblks > xfs_ag_resv_needed(pag, type) + nr_blocks;
169 * Figure out how many blocks to reserve for an AG repair. We calculate the
170 * worst case estimate for the number of blocks we'd need to rebuild one of
171 * any type of per-AG btree.
174 xrep_calc_ag_resblks(
175 struct xfs_scrub *sc)
177 struct xfs_mount *mp = sc->mp;
178 struct xfs_scrub_metadata *sm = sc->sm;
179 struct xfs_perag *pag;
181 xfs_agino_t icount = NULLAGINO;
182 xfs_extlen_t aglen = NULLAGBLOCK;
183 xfs_extlen_t usedlen;
184 xfs_extlen_t freelen;
185 xfs_extlen_t bnobt_sz;
186 xfs_extlen_t inobt_sz;
187 xfs_extlen_t rmapbt_sz;
188 xfs_extlen_t refcbt_sz;
191 if (!(sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR))
194 pag = xfs_perag_get(mp, sm->sm_agno);
195 if (pag->pagi_init) {
196 /* Use in-core icount if possible. */
197 icount = pag->pagi_count;
199 /* Try to get the actual counters from disk. */
200 error = xfs_ialloc_read_agi(mp, NULL, sm->sm_agno, &bp);
202 icount = pag->pagi_count;
207 /* Now grab the block counters from the AGF. */
208 error = xfs_alloc_read_agf(mp, NULL, sm->sm_agno, 0, &bp);
210 aglen = xfs_ag_block_count(mp, sm->sm_agno);
214 struct xfs_agf *agf = bp->b_addr;
216 aglen = be32_to_cpu(agf->agf_length);
217 freelen = be32_to_cpu(agf->agf_freeblks);
218 usedlen = aglen - freelen;
223 /* If the icount is impossible, make some worst-case assumptions. */
224 if (icount == NULLAGINO ||
225 !xfs_verify_agino(mp, sm->sm_agno, icount)) {
226 xfs_agino_t first, last;
228 xfs_agino_range(mp, sm->sm_agno, &first, &last);
229 icount = last - first + 1;
232 /* If the block counts are impossible, make worst-case assumptions. */
233 if (aglen == NULLAGBLOCK ||
234 aglen != xfs_ag_block_count(mp, sm->sm_agno) ||
236 aglen = xfs_ag_block_count(mp, sm->sm_agno);
241 trace_xrep_calc_ag_resblks(mp, sm->sm_agno, icount, aglen,
245 * Figure out how many blocks we'd need worst case to rebuild
246 * each type of btree. Note that we can only rebuild the
247 * bnobt/cntbt or inobt/finobt as pairs.
249 bnobt_sz = 2 * xfs_allocbt_calc_size(mp, freelen);
250 if (xfs_sb_version_hassparseinodes(&mp->m_sb))
251 inobt_sz = xfs_iallocbt_calc_size(mp, icount /
252 XFS_INODES_PER_HOLEMASK_BIT);
254 inobt_sz = xfs_iallocbt_calc_size(mp, icount /
255 XFS_INODES_PER_CHUNK);
256 if (xfs_sb_version_hasfinobt(&mp->m_sb))
258 if (xfs_sb_version_hasreflink(&mp->m_sb))
259 refcbt_sz = xfs_refcountbt_calc_size(mp, usedlen);
262 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
264 * Guess how many blocks we need to rebuild the rmapbt.
265 * For non-reflink filesystems we can't have more records than
266 * used blocks. However, with reflink it's possible to have
267 * more than one rmap record per AG block. We don't know how
268 * many rmaps there could be in the AG, so we start off with
269 * what we hope is an generous over-estimation.
271 if (xfs_sb_version_hasreflink(&mp->m_sb))
272 rmapbt_sz = xfs_rmapbt_calc_size(mp,
273 (unsigned long long)aglen * 2);
275 rmapbt_sz = xfs_rmapbt_calc_size(mp, usedlen);
280 trace_xrep_calc_ag_resblks_btsize(mp, sm->sm_agno, bnobt_sz,
281 inobt_sz, rmapbt_sz, refcbt_sz);
283 return max(max(bnobt_sz, inobt_sz), max(rmapbt_sz, refcbt_sz));
286 /* Allocate a block in an AG. */
289 struct xfs_scrub *sc,
290 const struct xfs_owner_info *oinfo,
291 xfs_fsblock_t *fsbno,
292 enum xfs_ag_resv_type resv)
294 struct xfs_alloc_arg args = {0};
299 case XFS_AG_RESV_AGFL:
300 case XFS_AG_RESV_RMAPBT:
301 error = xfs_alloc_get_freelist(sc->tp, sc->sa.agf_bp, &bno, 1);
304 if (bno == NULLAGBLOCK)
306 xfs_extent_busy_reuse(sc->mp, sc->sa.agno, bno,
308 *fsbno = XFS_AGB_TO_FSB(sc->mp, sc->sa.agno, bno);
309 if (resv == XFS_AG_RESV_RMAPBT)
310 xfs_ag_resv_rmapbt_alloc(sc->mp, sc->sa.agno);
319 args.fsbno = XFS_AGB_TO_FSB(args.mp, sc->sa.agno, 0);
323 args.type = XFS_ALLOCTYPE_THIS_AG;
326 error = xfs_alloc_vextent(&args);
329 if (args.fsbno == NULLFSBLOCK)
331 ASSERT(args.len == 1);
337 /* Initialize a new AG btree root block with zero entries. */
340 struct xfs_scrub *sc,
342 struct xfs_buf **bpp,
344 const struct xfs_buf_ops *ops)
346 struct xfs_trans *tp = sc->tp;
347 struct xfs_mount *mp = sc->mp;
351 trace_xrep_init_btblock(mp, XFS_FSB_TO_AGNO(mp, fsb),
352 XFS_FSB_TO_AGBNO(mp, fsb), btnum);
354 ASSERT(XFS_FSB_TO_AGNO(mp, fsb) == sc->sa.agno);
355 error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
356 XFS_FSB_TO_DADDR(mp, fsb), XFS_FSB_TO_BB(mp, 1), 0,
360 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
361 xfs_btree_init_block(mp, bp, btnum, 0, 0, sc->sa.agno);
362 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_BTREE_BUF);
363 xfs_trans_log_buf(tp, bp, 0, BBTOB(bp->b_length) - 1);
371 * Reconstructing per-AG Btrees
373 * When a space btree is corrupt, we don't bother trying to fix it. Instead,
374 * we scan secondary space metadata to derive the records that should be in
375 * the damaged btree, initialize a fresh btree root, and insert the records.
376 * Note that for rebuilding the rmapbt we scan all the primary data to
377 * generate the new records.
379 * However, that leaves the matter of removing all the metadata describing the
380 * old broken structure. For primary metadata we use the rmap data to collect
381 * every extent with a matching rmap owner (bitmap); we then iterate all other
382 * metadata structures with the same rmap owner to collect the extents that
383 * cannot be removed (sublist). We then subtract sublist from bitmap to
384 * derive the blocks that were used by the old btree. These blocks can be
387 * For rmapbt reconstructions we must use different tactics for extent
388 * collection. First we iterate all primary metadata (this excludes the old
389 * rmapbt, obviously) to generate new rmap records. The gaps in the rmap
390 * records are collected as bitmap. The bnobt records are collected as
391 * sublist. As with the other btrees we subtract sublist from bitmap, and the
392 * result (since the rmapbt lives in the free space) are the blocks from the
395 * Disposal of Blocks from Old per-AG Btrees
397 * Now that we've constructed a new btree to replace the damaged one, we want
398 * to dispose of the blocks that (we think) the old btree was using.
399 * Previously, we used the rmapbt to collect the extents (bitmap) with the
400 * rmap owner corresponding to the tree we rebuilt, collected extents for any
401 * blocks with the same rmap owner that are owned by another data structure
402 * (sublist), and subtracted sublist from bitmap. In theory the extents
403 * remaining in bitmap are the old btree's blocks.
405 * Unfortunately, it's possible that the btree was crosslinked with other
406 * blocks on disk. The rmap data can tell us if there are multiple owners, so
407 * if the rmapbt says there is an owner of this block other than @oinfo, then
408 * the block is crosslinked. Remove the reverse mapping and continue.
410 * If there is one rmap record, we can free the block, which removes the
411 * reverse mapping but doesn't add the block to the free space. Our repair
412 * strategy is to hope the other metadata objects crosslinked on this block
413 * will be rebuilt (atop different blocks), thereby removing all the cross
416 * If there are no rmap records at all, we also free the block. If the btree
417 * being rebuilt lives in the free space (bnobt/cntbt/rmapbt) then there isn't
418 * supposed to be a rmap record and everything is ok. For other btrees there
419 * had to have been an rmap entry for the block to have ended up on @bitmap,
420 * so if it's gone now there's something wrong and the fs will shut down.
422 * Note: If there are multiple rmap records with only the same rmap owner as
423 * the btree we're trying to rebuild and the block is indeed owned by another
424 * data structure with the same rmap owner, then the block will be in sublist
425 * and therefore doesn't need disposal. If there are multiple rmap records
426 * with only the same rmap owner but the block is not owned by something with
427 * the same rmap owner, the block will be freed.
429 * The caller is responsible for locking the AG headers for the entire rebuild
430 * operation so that nothing else can sneak in and change the AG state while
431 * we're not looking. We also assume that the caller already invalidated any
432 * buffers associated with @bitmap.
436 * Invalidate buffers for per-AG btree blocks we're dumping. This function
437 * is not intended for use with file data repairs; we have bunmapi for that.
440 xrep_invalidate_blocks(
441 struct xfs_scrub *sc,
442 struct xbitmap *bitmap)
444 struct xbitmap_range *bmr;
445 struct xbitmap_range *n;
450 * For each block in each extent, see if there's an incore buffer for
451 * exactly that block; if so, invalidate it. The buffer cache only
452 * lets us look for one buffer at a time, so we have to look one block
453 * at a time. Avoid invalidating AG headers and post-EOFS blocks
454 * because we never own those; and if we can't TRYLOCK the buffer we
455 * assume it's owned by someone else.
457 for_each_xbitmap_block(fsbno, bmr, n, bitmap) {
458 /* Skip AG headers and post-EOFS blocks */
459 if (!xfs_verify_fsbno(sc->mp, fsbno))
461 bp = xfs_buf_incore(sc->mp->m_ddev_targp,
462 XFS_FSB_TO_DADDR(sc->mp, fsbno),
463 XFS_FSB_TO_BB(sc->mp, 1), XBF_TRYLOCK);
465 xfs_trans_bjoin(sc->tp, bp);
466 xfs_trans_binval(sc->tp, bp);
473 /* Ensure the freelist is the correct size. */
476 struct xfs_scrub *sc,
479 struct xfs_alloc_arg args = {0};
483 args.agno = sc->sa.agno;
485 args.pag = sc->sa.pag;
487 return xfs_alloc_fix_freelist(&args,
488 can_shrink ? 0 : XFS_ALLOC_FLAG_NOSHRINK);
492 * Put a block back on the AGFL.
496 struct xfs_scrub *sc,
501 /* Make sure there's space on the freelist. */
502 error = xrep_fix_freelist(sc, true);
507 * Since we're "freeing" a lost block onto the AGFL, we have to
508 * create an rmap for the block prior to merging it or else other
511 error = xfs_rmap_alloc(sc->tp, sc->sa.agf_bp, sc->sa.agno, agbno, 1,
516 /* Put the block on the AGFL. */
517 error = xfs_alloc_put_freelist(sc->tp, sc->sa.agf_bp, sc->sa.agfl_bp,
521 xfs_extent_busy_insert(sc->tp, sc->sa.agno, agbno, 1,
522 XFS_EXTENT_BUSY_SKIP_DISCARD);
527 /* Dispose of a single block. */
530 struct xfs_scrub *sc,
532 const struct xfs_owner_info *oinfo,
533 enum xfs_ag_resv_type resv)
535 struct xfs_btree_cur *cur;
536 struct xfs_buf *agf_bp = NULL;
542 agno = XFS_FSB_TO_AGNO(sc->mp, fsbno);
543 agbno = XFS_FSB_TO_AGBNO(sc->mp, fsbno);
546 * If we are repairing per-inode metadata, we need to read in the AGF
547 * buffer. Otherwise, we're repairing a per-AG structure, so reuse
548 * the AGF buffer that the setup functions already grabbed.
551 error = xfs_alloc_read_agf(sc->mp, sc->tp, agno, 0, &agf_bp);
555 agf_bp = sc->sa.agf_bp;
557 cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf_bp, agno);
559 /* Can we find any other rmappings? */
560 error = xfs_rmap_has_other_keys(cur, agbno, 1, oinfo, &has_other_rmap);
561 xfs_btree_del_cursor(cur, error);
566 * If there are other rmappings, this block is cross linked and must
567 * not be freed. Remove the reverse mapping and move on. Otherwise,
568 * we were the only owner of the block, so free the extent, which will
569 * also remove the rmap.
571 * XXX: XFS doesn't support detecting the case where a single block
572 * metadata structure is crosslinked with a multi-block structure
573 * because the buffer cache doesn't detect aliasing problems, so we
574 * can't fix 100% of crosslinking problems (yet). The verifiers will
575 * blow on writeout, the filesystem will shut down, and the admin gets
579 error = xfs_rmap_free(sc->tp, agf_bp, agno, agbno, 1, oinfo);
580 else if (resv == XFS_AG_RESV_AGFL)
581 error = xrep_put_freelist(sc, agbno);
583 error = xfs_free_extent(sc->tp, fsbno, 1, oinfo, resv);
584 if (agf_bp != sc->sa.agf_bp)
585 xfs_trans_brelse(sc->tp, agf_bp);
590 return xfs_trans_roll_inode(&sc->tp, sc->ip);
591 return xrep_roll_ag_trans(sc);
594 if (agf_bp != sc->sa.agf_bp)
595 xfs_trans_brelse(sc->tp, agf_bp);
599 /* Dispose of every block of every extent in the bitmap. */
602 struct xfs_scrub *sc,
603 struct xbitmap *bitmap,
604 const struct xfs_owner_info *oinfo,
605 enum xfs_ag_resv_type type)
607 struct xbitmap_range *bmr;
608 struct xbitmap_range *n;
612 ASSERT(xfs_sb_version_hasrmapbt(&sc->mp->m_sb));
614 for_each_xbitmap_block(fsbno, bmr, n, bitmap) {
615 ASSERT(sc->ip != NULL ||
616 XFS_FSB_TO_AGNO(sc->mp, fsbno) == sc->sa.agno);
617 trace_xrep_dispose_btree_extent(sc->mp,
618 XFS_FSB_TO_AGNO(sc->mp, fsbno),
619 XFS_FSB_TO_AGBNO(sc->mp, fsbno), 1);
621 error = xrep_reap_block(sc, fsbno, oinfo, type);
630 * Finding per-AG Btree Roots for AGF/AGI Reconstruction
632 * If the AGF or AGI become slightly corrupted, it may be necessary to rebuild
633 * the AG headers by using the rmap data to rummage through the AG looking for
634 * btree roots. This is not guaranteed to work if the AG is heavily damaged
635 * or the rmap data are corrupt.
637 * Callers of xrep_find_ag_btree_roots must lock the AGF and AGFL
638 * buffers if the AGF is being rebuilt; or the AGF and AGI buffers if the
639 * AGI is being rebuilt. It must maintain these locks until it's safe for
640 * other threads to change the btrees' shapes. The caller provides
641 * information about the btrees to look for by passing in an array of
642 * xrep_find_ag_btree with the (rmap owner, buf_ops, magic) fields set.
643 * The (root, height) fields will be set on return if anything is found. The
644 * last element of the array should have a NULL buf_ops to mark the end of the
647 * For every rmapbt record matching any of the rmap owners in btree_info,
648 * read each block referenced by the rmap record. If the block is a btree
649 * block from this filesystem matching any of the magic numbers and has a
650 * level higher than what we've already seen, remember the block and the
651 * height of the tree required to have such a block. When the call completes,
652 * we return the highest block we've found for each btree description; those
653 * should be the roots.
656 struct xrep_findroot {
657 struct xfs_scrub *sc;
658 struct xfs_buf *agfl_bp;
660 struct xrep_find_ag_btree *btree_info;
663 /* See if our block is in the AGFL. */
665 xrep_findroot_agfl_walk(
666 struct xfs_mount *mp,
670 xfs_agblock_t *agbno = priv;
672 return (*agbno == bno) ? -ECANCELED : 0;
675 /* Does this block match the btree information passed in? */
678 struct xrep_findroot *ri,
679 struct xrep_find_ag_btree *fab,
682 bool *done_with_block)
684 struct xfs_mount *mp = ri->sc->mp;
686 struct xfs_btree_block *btblock;
691 daddr = XFS_AGB_TO_DADDR(mp, ri->sc->sa.agno, agbno);
694 * Blocks in the AGFL have stale contents that might just happen to
695 * have a matching magic and uuid. We don't want to pull these blocks
696 * in as part of a tree root, so we have to filter out the AGFL stuff
697 * here. If the AGFL looks insane we'll just refuse to repair.
699 if (owner == XFS_RMAP_OWN_AG) {
700 error = xfs_agfl_walk(mp, ri->agf, ri->agfl_bp,
701 xrep_findroot_agfl_walk, &agbno);
702 if (error == -ECANCELED)
709 * Read the buffer into memory so that we can see if it's a match for
710 * our btree type. We have no clue if it is beforehand, and we want to
711 * avoid xfs_trans_read_buf's behavior of dumping the DONE state (which
712 * will cause needless disk reads in subsequent calls to this function)
713 * and logging metadata verifier failures.
715 * Therefore, pass in NULL buffer ops. If the buffer was already in
716 * memory from some other caller it will already have b_ops assigned.
717 * If it was in memory from a previous unsuccessful findroot_block
718 * call, the buffer won't have b_ops but it should be clean and ready
719 * for us to try to verify if the read call succeeds. The same applies
720 * if the buffer wasn't in memory at all.
722 * Note: If we never match a btree type with this buffer, it will be
723 * left in memory with NULL b_ops. This shouldn't be a problem unless
724 * the buffer gets written.
726 error = xfs_trans_read_buf(mp, ri->sc->tp, mp->m_ddev_targp, daddr,
727 mp->m_bsize, 0, &bp, NULL);
731 /* Ensure the block magic matches the btree type we're looking for. */
732 btblock = XFS_BUF_TO_BLOCK(bp);
733 ASSERT(fab->buf_ops->magic[1] != 0);
734 if (btblock->bb_magic != fab->buf_ops->magic[1])
738 * If the buffer already has ops applied and they're not the ones for
739 * this btree type, we know this block doesn't match the btree and we
742 * If the buffer ops match ours, someone else has already validated
743 * the block for us, so we can move on to checking if this is a root
746 * If the buffer does not have ops, nobody has successfully validated
747 * the contents and the buffer cannot be dirty. If the magic, uuid,
748 * and structure match this btree type then we'll move on to checking
749 * if it's a root block candidate. If there is no match, bail out.
752 if (bp->b_ops != fab->buf_ops)
755 ASSERT(!xfs_trans_buf_is_dirty(bp));
756 if (!uuid_equal(&btblock->bb_u.s.bb_uuid,
757 &mp->m_sb.sb_meta_uuid))
760 * Read verifiers can reference b_ops, so we set the pointer
761 * here. If the verifier fails we'll reset the buffer state
762 * to what it was before we touched the buffer.
764 bp->b_ops = fab->buf_ops;
765 fab->buf_ops->verify_read(bp);
773 * Some read verifiers will (re)set b_ops, so we must be
774 * careful not to change b_ops after running the verifier.
779 * This block passes the magic/uuid and verifier tests for this btree
780 * type. We don't need the caller to try the other tree types.
782 *done_with_block = true;
785 * Compare this btree block's level to the height of the current
786 * candidate root block.
788 * If the level matches the root we found previously, throw away both
789 * blocks because there can't be two candidate roots.
791 * If level is lower in the tree than the root we found previously,
794 block_level = xfs_btree_get_level(btblock);
795 if (block_level + 1 == fab->height) {
796 fab->root = NULLAGBLOCK;
798 } else if (block_level < fab->height) {
803 * This is the highest block in the tree that we've found so far.
804 * Update the btree height to reflect what we've learned from this
807 fab->height = block_level + 1;
810 * If this block doesn't have sibling pointers, then it's the new root
811 * block candidate. Otherwise, the root will be found farther up the
814 if (btblock->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) &&
815 btblock->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
818 fab->root = NULLAGBLOCK;
820 trace_xrep_findroot_block(mp, ri->sc->sa.agno, agbno,
821 be32_to_cpu(btblock->bb_magic), fab->height - 1);
823 xfs_trans_brelse(ri->sc->tp, bp);
828 * Do any of the blocks in this rmap record match one of the btrees we're
833 struct xfs_btree_cur *cur,
834 struct xfs_rmap_irec *rec,
837 struct xrep_findroot *ri = priv;
838 struct xrep_find_ag_btree *fab;
843 /* Ignore anything that isn't AG metadata. */
844 if (!XFS_RMAP_NON_INODE_OWNER(rec->rm_owner))
847 /* Otherwise scan each block + btree type. */
848 for (b = 0; b < rec->rm_blockcount; b++) {
850 for (fab = ri->btree_info; fab->buf_ops; fab++) {
851 if (rec->rm_owner != fab->rmap_owner)
853 error = xrep_findroot_block(ri, fab,
854 rec->rm_owner, rec->rm_startblock + b,
866 /* Find the roots of the per-AG btrees described in btree_info. */
868 xrep_find_ag_btree_roots(
869 struct xfs_scrub *sc,
870 struct xfs_buf *agf_bp,
871 struct xrep_find_ag_btree *btree_info,
872 struct xfs_buf *agfl_bp)
874 struct xfs_mount *mp = sc->mp;
875 struct xrep_findroot ri;
876 struct xrep_find_ag_btree *fab;
877 struct xfs_btree_cur *cur;
880 ASSERT(xfs_buf_islocked(agf_bp));
881 ASSERT(agfl_bp == NULL || xfs_buf_islocked(agfl_bp));
884 ri.btree_info = btree_info;
885 ri.agf = agf_bp->b_addr;
886 ri.agfl_bp = agfl_bp;
887 for (fab = btree_info; fab->buf_ops; fab++) {
888 ASSERT(agfl_bp || fab->rmap_owner != XFS_RMAP_OWN_AG);
889 ASSERT(XFS_RMAP_NON_INODE_OWNER(fab->rmap_owner));
890 fab->root = NULLAGBLOCK;
894 cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.agno);
895 error = xfs_rmap_query_all(cur, xrep_findroot_rmap, &ri);
896 xfs_btree_del_cursor(cur, error);
901 /* Force a quotacheck the next time we mount. */
903 xrep_force_quotacheck(
904 struct xfs_scrub *sc,
909 flag = xfs_quota_chkd_flag(type);
910 if (!(flag & sc->mp->m_qflags))
913 sc->mp->m_qflags &= ~flag;
914 spin_lock(&sc->mp->m_sb_lock);
915 sc->mp->m_sb.sb_qflags &= ~flag;
916 spin_unlock(&sc->mp->m_sb_lock);
921 * Attach dquots to this inode, or schedule quotacheck to fix them.
923 * This function ensures that the appropriate dquots are attached to an inode.
924 * We cannot allow the dquot code to allocate an on-disk dquot block here
925 * because we're already in transaction context with the inode locked. The
926 * on-disk dquot should already exist anyway. If the quota code signals
927 * corruption or missing quota information, schedule quotacheck, which will
928 * repair corruptions in the quota metadata.
932 struct xfs_scrub *sc)
936 error = xfs_qm_dqattach_locked(sc->ip, false);
941 xfs_err_ratelimited(sc->mp,
942 "inode %llu repair encountered quota error %d, quotacheck forced.",
943 (unsigned long long)sc->ip->i_ino, error);
944 if (XFS_IS_UQUOTA_ON(sc->mp) && !sc->ip->i_udquot)
945 xrep_force_quotacheck(sc, XFS_DQTYPE_USER);
946 if (XFS_IS_GQUOTA_ON(sc->mp) && !sc->ip->i_gdquot)
947 xrep_force_quotacheck(sc, XFS_DQTYPE_GROUP);
948 if (XFS_IS_PQUOTA_ON(sc->mp) && !sc->ip->i_pdquot)
949 xrep_force_quotacheck(sc, XFS_DQTYPE_PROJ);