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"
26 #include "xfs_ag_resv.h"
27 #include "xfs_quota.h"
29 #include "scrub/scrub.h"
30 #include "scrub/common.h"
31 #include "scrub/trace.h"
32 #include "scrub/repair.h"
33 #include "scrub/bitmap.h"
36 * Attempt to repair some metadata, if the metadata is corrupt and userspace
37 * told us to fix it. This function returns -EAGAIN to mean "re-run scrub",
38 * and will set *fixed to true if it thinks it repaired anything.
46 trace_xrep_attempt(XFS_I(file_inode(sc->file)), sc->sm, error);
48 xchk_ag_btcur_free(&sc->sa);
50 /* Repair whatever's broken. */
51 ASSERT(sc->ops->repair);
52 error = sc->ops->repair(sc);
53 trace_xrep_done(XFS_I(file_inode(sc->file)), sc->sm, error);
57 * Repair succeeded. Commit the fixes and perform a second
58 * scrub so that we can tell userspace if we fixed the problem.
60 sc->sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT;
61 sc->flags |= XREP_ALREADY_FIXED;
65 /* Tell the caller to try again having grabbed all the locks. */
66 if (!(sc->flags & XCHK_TRY_HARDER)) {
67 sc->flags |= XCHK_TRY_HARDER;
71 * We tried harder but still couldn't grab all the resources
72 * we needed to fix it. The corruption has not been fixed,
73 * so report back to userspace.
82 * Complain about unfixable problems in the filesystem. We don't log
83 * corruptions when IFLAG_REPAIR wasn't set on the assumption that the driver
84 * program is xfs_scrub, which will call back with IFLAG_REPAIR set if the
85 * administrator isn't running xfs_scrub in no-repairs mode.
87 * Use this helper function because _ratelimited silently declares a static
88 * structure to track rate limiting information.
94 xfs_alert_ratelimited(mp,
95 "Corruption not fixed during online repair. Unmount and run xfs_repair.");
99 * Repair probe -- userspace uses this to probe if we're willing to repair a
104 struct xfs_scrub *sc)
108 if (xchk_should_terminate(sc, &error))
115 * Roll a transaction, keeping the AG headers locked and reinitializing
120 struct xfs_scrub *sc)
124 /* Keep the AG header buffers locked so we can keep going. */
126 xfs_trans_bhold(sc->tp, sc->sa.agi_bp);
128 xfs_trans_bhold(sc->tp, sc->sa.agf_bp);
130 xfs_trans_bhold(sc->tp, sc->sa.agfl_bp);
133 * Roll the transaction. We still own the buffer and the buffer lock
134 * regardless of whether or not the roll succeeds. If the roll fails,
135 * the buffers will be released during teardown on our way out of the
136 * kernel. If it succeeds, we join them to the new transaction and
139 error = xfs_trans_roll(&sc->tp);
143 /* Join AG headers to the new transaction. */
145 xfs_trans_bjoin(sc->tp, sc->sa.agi_bp);
147 xfs_trans_bjoin(sc->tp, sc->sa.agf_bp);
149 xfs_trans_bjoin(sc->tp, sc->sa.agfl_bp);
155 * Does the given AG have enough space to rebuild a btree? Neither AG
156 * reservation can be critical, and we must have enough space (factoring
157 * in AG reservations) to construct a whole btree.
161 struct xfs_perag *pag,
162 xfs_extlen_t nr_blocks,
163 enum xfs_ag_resv_type type)
165 return !xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) &&
166 !xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA) &&
167 pag->pagf_freeblks > xfs_ag_resv_needed(pag, type) + nr_blocks;
171 * Figure out how many blocks to reserve for an AG repair. We calculate the
172 * worst case estimate for the number of blocks we'd need to rebuild one of
173 * any type of per-AG btree.
176 xrep_calc_ag_resblks(
177 struct xfs_scrub *sc)
179 struct xfs_mount *mp = sc->mp;
180 struct xfs_scrub_metadata *sm = sc->sm;
181 struct xfs_perag *pag;
183 xfs_agino_t icount = NULLAGINO;
184 xfs_extlen_t aglen = NULLAGBLOCK;
185 xfs_extlen_t usedlen;
186 xfs_extlen_t freelen;
187 xfs_extlen_t bnobt_sz;
188 xfs_extlen_t inobt_sz;
189 xfs_extlen_t rmapbt_sz;
190 xfs_extlen_t refcbt_sz;
193 if (!(sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR))
196 pag = xfs_perag_get(mp, sm->sm_agno);
197 if (pag->pagi_init) {
198 /* Use in-core icount if possible. */
199 icount = pag->pagi_count;
201 /* Try to get the actual counters from disk. */
202 error = xfs_ialloc_read_agi(pag, NULL, &bp);
204 icount = pag->pagi_count;
209 /* Now grab the block counters from the AGF. */
210 error = xfs_alloc_read_agf(pag, NULL, 0, &bp);
212 aglen = xfs_ag_block_count(mp, sm->sm_agno);
216 struct xfs_agf *agf = bp->b_addr;
218 aglen = be32_to_cpu(agf->agf_length);
219 freelen = be32_to_cpu(agf->agf_freeblks);
220 usedlen = aglen - freelen;
224 /* If the icount is impossible, make some worst-case assumptions. */
225 if (icount == NULLAGINO ||
226 !xfs_verify_agino(pag, icount)) {
227 icount = pag->agino_max - pag->agino_min + 1;
231 /* If the block counts are impossible, make worst-case assumptions. */
232 if (aglen == NULLAGBLOCK ||
233 aglen != xfs_ag_block_count(mp, sm->sm_agno) ||
235 aglen = xfs_ag_block_count(mp, sm->sm_agno);
240 trace_xrep_calc_ag_resblks(mp, sm->sm_agno, icount, aglen,
244 * Figure out how many blocks we'd need worst case to rebuild
245 * each type of btree. Note that we can only rebuild the
246 * bnobt/cntbt or inobt/finobt as pairs.
248 bnobt_sz = 2 * xfs_allocbt_calc_size(mp, freelen);
249 if (xfs_has_sparseinodes(mp))
250 inobt_sz = xfs_iallocbt_calc_size(mp, icount /
251 XFS_INODES_PER_HOLEMASK_BIT);
253 inobt_sz = xfs_iallocbt_calc_size(mp, icount /
254 XFS_INODES_PER_CHUNK);
255 if (xfs_has_finobt(mp))
257 if (xfs_has_reflink(mp))
258 refcbt_sz = xfs_refcountbt_calc_size(mp, usedlen);
261 if (xfs_has_rmapbt(mp)) {
263 * Guess how many blocks we need to rebuild the rmapbt.
264 * For non-reflink filesystems we can't have more records than
265 * used blocks. However, with reflink it's possible to have
266 * more than one rmap record per AG block. We don't know how
267 * many rmaps there could be in the AG, so we start off with
268 * what we hope is an generous over-estimation.
270 if (xfs_has_reflink(mp))
271 rmapbt_sz = xfs_rmapbt_calc_size(mp,
272 (unsigned long long)aglen * 2);
274 rmapbt_sz = xfs_rmapbt_calc_size(mp, usedlen);
279 trace_xrep_calc_ag_resblks_btsize(mp, sm->sm_agno, bnobt_sz,
280 inobt_sz, rmapbt_sz, refcbt_sz);
282 return max(max(bnobt_sz, inobt_sz), max(rmapbt_sz, refcbt_sz));
285 /* Allocate a block in an AG. */
288 struct xfs_scrub *sc,
289 const struct xfs_owner_info *oinfo,
290 xfs_fsblock_t *fsbno,
291 enum xfs_ag_resv_type resv)
293 struct xfs_alloc_arg args = {0};
298 case XFS_AG_RESV_AGFL:
299 case XFS_AG_RESV_RMAPBT:
300 error = xfs_alloc_get_freelist(sc->sa.pag, sc->tp,
301 sc->sa.agf_bp, &bno, 1);
304 if (bno == NULLAGBLOCK)
306 xfs_extent_busy_reuse(sc->mp, sc->sa.pag, bno, 1, false);
307 *fsbno = XFS_AGB_TO_FSB(sc->mp, sc->sa.pag->pag_agno, bno);
308 if (resv == XFS_AG_RESV_RMAPBT)
309 xfs_ag_resv_rmapbt_alloc(sc->mp, sc->sa.pag->pag_agno);
318 args.fsbno = XFS_AGB_TO_FSB(args.mp, sc->sa.pag->pag_agno, 0);
322 args.type = XFS_ALLOCTYPE_THIS_AG;
325 error = xfs_alloc_vextent(&args);
328 if (args.fsbno == NULLFSBLOCK)
330 ASSERT(args.len == 1);
336 /* Initialize a new AG btree root block with zero entries. */
339 struct xfs_scrub *sc,
341 struct xfs_buf **bpp,
343 const struct xfs_buf_ops *ops)
345 struct xfs_trans *tp = sc->tp;
346 struct xfs_mount *mp = sc->mp;
350 trace_xrep_init_btblock(mp, XFS_FSB_TO_AGNO(mp, fsb),
351 XFS_FSB_TO_AGBNO(mp, fsb), btnum);
353 ASSERT(XFS_FSB_TO_AGNO(mp, fsb) == sc->sa.pag->pag_agno);
354 error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
355 XFS_FSB_TO_DADDR(mp, fsb), XFS_FSB_TO_BB(mp, 1), 0,
359 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
360 xfs_btree_init_block(mp, bp, btnum, 0, 0, sc->sa.pag->pag_agno);
361 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_BTREE_BUF);
362 xfs_trans_log_buf(tp, bp, 0, BBTOB(bp->b_length) - 1);
370 * Reconstructing per-AG Btrees
372 * When a space btree is corrupt, we don't bother trying to fix it. Instead,
373 * we scan secondary space metadata to derive the records that should be in
374 * the damaged btree, initialize a fresh btree root, and insert the records.
375 * Note that for rebuilding the rmapbt we scan all the primary data to
376 * generate the new records.
378 * However, that leaves the matter of removing all the metadata describing the
379 * old broken structure. For primary metadata we use the rmap data to collect
380 * every extent with a matching rmap owner (bitmap); we then iterate all other
381 * metadata structures with the same rmap owner to collect the extents that
382 * cannot be removed (sublist). We then subtract sublist from bitmap to
383 * derive the blocks that were used by the old btree. These blocks can be
386 * For rmapbt reconstructions we must use different tactics for extent
387 * collection. First we iterate all primary metadata (this excludes the old
388 * rmapbt, obviously) to generate new rmap records. The gaps in the rmap
389 * records are collected as bitmap. The bnobt records are collected as
390 * sublist. As with the other btrees we subtract sublist from bitmap, and the
391 * result (since the rmapbt lives in the free space) are the blocks from the
394 * Disposal of Blocks from Old per-AG Btrees
396 * Now that we've constructed a new btree to replace the damaged one, we want
397 * to dispose of the blocks that (we think) the old btree was using.
398 * Previously, we used the rmapbt to collect the extents (bitmap) with the
399 * rmap owner corresponding to the tree we rebuilt, collected extents for any
400 * blocks with the same rmap owner that are owned by another data structure
401 * (sublist), and subtracted sublist from bitmap. In theory the extents
402 * remaining in bitmap are the old btree's blocks.
404 * Unfortunately, it's possible that the btree was crosslinked with other
405 * blocks on disk. The rmap data can tell us if there are multiple owners, so
406 * if the rmapbt says there is an owner of this block other than @oinfo, then
407 * the block is crosslinked. Remove the reverse mapping and continue.
409 * If there is one rmap record, we can free the block, which removes the
410 * reverse mapping but doesn't add the block to the free space. Our repair
411 * strategy is to hope the other metadata objects crosslinked on this block
412 * will be rebuilt (atop different blocks), thereby removing all the cross
415 * If there are no rmap records at all, we also free the block. If the btree
416 * being rebuilt lives in the free space (bnobt/cntbt/rmapbt) then there isn't
417 * supposed to be a rmap record and everything is ok. For other btrees there
418 * had to have been an rmap entry for the block to have ended up on @bitmap,
419 * so if it's gone now there's something wrong and the fs will shut down.
421 * Note: If there are multiple rmap records with only the same rmap owner as
422 * the btree we're trying to rebuild and the block is indeed owned by another
423 * data structure with the same rmap owner, then the block will be in sublist
424 * and therefore doesn't need disposal. If there are multiple rmap records
425 * with only the same rmap owner but the block is not owned by something with
426 * the same rmap owner, the block will be freed.
428 * The caller is responsible for locking the AG headers for the entire rebuild
429 * operation so that nothing else can sneak in and change the AG state while
430 * we're not looking. We also assume that the caller already invalidated any
431 * buffers associated with @bitmap.
435 * Invalidate buffers for per-AG btree blocks we're dumping. This function
436 * is not intended for use with file data repairs; we have bunmapi for that.
439 xrep_invalidate_blocks(
440 struct xfs_scrub *sc,
441 struct xbitmap *bitmap)
443 struct xbitmap_range *bmr;
444 struct xbitmap_range *n;
449 * For each block in each extent, see if there's an incore buffer for
450 * exactly that block; if so, invalidate it. The buffer cache only
451 * lets us look for one buffer at a time, so we have to look one block
452 * at a time. Avoid invalidating AG headers and post-EOFS blocks
453 * because we never own those; and if we can't TRYLOCK the buffer we
454 * assume it's owned by someone else.
456 for_each_xbitmap_block(fsbno, bmr, n, bitmap) {
457 /* Skip AG headers and post-EOFS blocks */
458 if (!xfs_verify_fsbno(sc->mp, fsbno))
460 bp = xfs_buf_incore(sc->mp->m_ddev_targp,
461 XFS_FSB_TO_DADDR(sc->mp, fsbno),
462 XFS_FSB_TO_BB(sc->mp, 1), XBF_TRYLOCK);
464 xfs_trans_bjoin(sc->tp, bp);
465 xfs_trans_binval(sc->tp, bp);
472 /* Ensure the freelist is the correct size. */
475 struct xfs_scrub *sc,
478 struct xfs_alloc_arg args = {0};
482 args.agno = sc->sa.pag->pag_agno;
484 args.pag = sc->sa.pag;
486 return xfs_alloc_fix_freelist(&args,
487 can_shrink ? 0 : XFS_ALLOC_FLAG_NOSHRINK);
491 * Put a block back on the AGFL.
495 struct xfs_scrub *sc,
500 /* Make sure there's space on the freelist. */
501 error = xrep_fix_freelist(sc, true);
506 * Since we're "freeing" a lost block onto the AGFL, we have to
507 * create an rmap for the block prior to merging it or else other
510 error = xfs_rmap_alloc(sc->tp, sc->sa.agf_bp, sc->sa.pag, agbno, 1,
515 /* Put the block on the AGFL. */
516 error = xfs_alloc_put_freelist(sc->sa.pag, sc->tp, sc->sa.agf_bp,
517 sc->sa.agfl_bp, agbno, 0);
520 xfs_extent_busy_insert(sc->tp, sc->sa.pag, agbno, 1,
521 XFS_EXTENT_BUSY_SKIP_DISCARD);
526 /* Dispose of a single block. */
529 struct xfs_scrub *sc,
531 const struct xfs_owner_info *oinfo,
532 enum xfs_ag_resv_type resv)
534 struct xfs_btree_cur *cur;
535 struct xfs_buf *agf_bp = NULL;
541 agno = XFS_FSB_TO_AGNO(sc->mp, fsbno);
542 agbno = XFS_FSB_TO_AGBNO(sc->mp, fsbno);
543 ASSERT(agno == sc->sa.pag->pag_agno);
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->sa.pag, sc->tp, 0, &agf_bp);
555 agf_bp = sc->sa.agf_bp;
557 cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf_bp, sc->sa.pag);
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, sc->sa.pag, agbno,
581 else if (resv == XFS_AG_RESV_AGFL)
582 error = xrep_put_freelist(sc, agbno);
584 error = xfs_free_extent(sc->tp, fsbno, 1, oinfo, resv);
585 if (agf_bp != sc->sa.agf_bp)
586 xfs_trans_brelse(sc->tp, agf_bp);
591 return xfs_trans_roll_inode(&sc->tp, sc->ip);
592 return xrep_roll_ag_trans(sc);
595 if (agf_bp != sc->sa.agf_bp)
596 xfs_trans_brelse(sc->tp, agf_bp);
600 /* Dispose of every block of every extent in the bitmap. */
603 struct xfs_scrub *sc,
604 struct xbitmap *bitmap,
605 const struct xfs_owner_info *oinfo,
606 enum xfs_ag_resv_type type)
608 struct xbitmap_range *bmr;
609 struct xbitmap_range *n;
613 ASSERT(xfs_has_rmapbt(sc->mp));
615 for_each_xbitmap_block(fsbno, bmr, n, bitmap) {
616 ASSERT(sc->ip != NULL ||
617 XFS_FSB_TO_AGNO(sc->mp, fsbno) == sc->sa.pag->pag_agno);
618 trace_xrep_dispose_btree_extent(sc->mp,
619 XFS_FSB_TO_AGNO(sc->mp, fsbno),
620 XFS_FSB_TO_AGBNO(sc->mp, fsbno), 1);
622 error = xrep_reap_block(sc, fsbno, oinfo, type);
631 * Finding per-AG Btree Roots for AGF/AGI Reconstruction
633 * If the AGF or AGI become slightly corrupted, it may be necessary to rebuild
634 * the AG headers by using the rmap data to rummage through the AG looking for
635 * btree roots. This is not guaranteed to work if the AG is heavily damaged
636 * or the rmap data are corrupt.
638 * Callers of xrep_find_ag_btree_roots must lock the AGF and AGFL
639 * buffers if the AGF is being rebuilt; or the AGF and AGI buffers if the
640 * AGI is being rebuilt. It must maintain these locks until it's safe for
641 * other threads to change the btrees' shapes. The caller provides
642 * information about the btrees to look for by passing in an array of
643 * xrep_find_ag_btree with the (rmap owner, buf_ops, magic) fields set.
644 * The (root, height) fields will be set on return if anything is found. The
645 * last element of the array should have a NULL buf_ops to mark the end of the
648 * For every rmapbt record matching any of the rmap owners in btree_info,
649 * read each block referenced by the rmap record. If the block is a btree
650 * block from this filesystem matching any of the magic numbers and has a
651 * level higher than what we've already seen, remember the block and the
652 * height of the tree required to have such a block. When the call completes,
653 * we return the highest block we've found for each btree description; those
654 * should be the roots.
657 struct xrep_findroot {
658 struct xfs_scrub *sc;
659 struct xfs_buf *agfl_bp;
661 struct xrep_find_ag_btree *btree_info;
664 /* See if our block is in the AGFL. */
666 xrep_findroot_agfl_walk(
667 struct xfs_mount *mp,
671 xfs_agblock_t *agbno = priv;
673 return (*agbno == bno) ? -ECANCELED : 0;
676 /* Does this block match the btree information passed in? */
679 struct xrep_findroot *ri,
680 struct xrep_find_ag_btree *fab,
683 bool *done_with_block)
685 struct xfs_mount *mp = ri->sc->mp;
687 struct xfs_btree_block *btblock;
692 daddr = XFS_AGB_TO_DADDR(mp, ri->sc->sa.pag->pag_agno, agbno);
695 * Blocks in the AGFL have stale contents that might just happen to
696 * have a matching magic and uuid. We don't want to pull these blocks
697 * in as part of a tree root, so we have to filter out the AGFL stuff
698 * here. If the AGFL looks insane we'll just refuse to repair.
700 if (owner == XFS_RMAP_OWN_AG) {
701 error = xfs_agfl_walk(mp, ri->agf, ri->agfl_bp,
702 xrep_findroot_agfl_walk, &agbno);
703 if (error == -ECANCELED)
710 * Read the buffer into memory so that we can see if it's a match for
711 * our btree type. We have no clue if it is beforehand, and we want to
712 * avoid xfs_trans_read_buf's behavior of dumping the DONE state (which
713 * will cause needless disk reads in subsequent calls to this function)
714 * and logging metadata verifier failures.
716 * Therefore, pass in NULL buffer ops. If the buffer was already in
717 * memory from some other caller it will already have b_ops assigned.
718 * If it was in memory from a previous unsuccessful findroot_block
719 * call, the buffer won't have b_ops but it should be clean and ready
720 * for us to try to verify if the read call succeeds. The same applies
721 * if the buffer wasn't in memory at all.
723 * Note: If we never match a btree type with this buffer, it will be
724 * left in memory with NULL b_ops. This shouldn't be a problem unless
725 * the buffer gets written.
727 error = xfs_trans_read_buf(mp, ri->sc->tp, mp->m_ddev_targp, daddr,
728 mp->m_bsize, 0, &bp, NULL);
732 /* Ensure the block magic matches the btree type we're looking for. */
733 btblock = XFS_BUF_TO_BLOCK(bp);
734 ASSERT(fab->buf_ops->magic[1] != 0);
735 if (btblock->bb_magic != fab->buf_ops->magic[1])
739 * If the buffer already has ops applied and they're not the ones for
740 * this btree type, we know this block doesn't match the btree and we
743 * If the buffer ops match ours, someone else has already validated
744 * the block for us, so we can move on to checking if this is a root
747 * If the buffer does not have ops, nobody has successfully validated
748 * the contents and the buffer cannot be dirty. If the magic, uuid,
749 * and structure match this btree type then we'll move on to checking
750 * if it's a root block candidate. If there is no match, bail out.
753 if (bp->b_ops != fab->buf_ops)
756 ASSERT(!xfs_trans_buf_is_dirty(bp));
757 if (!uuid_equal(&btblock->bb_u.s.bb_uuid,
758 &mp->m_sb.sb_meta_uuid))
761 * Read verifiers can reference b_ops, so we set the pointer
762 * here. If the verifier fails we'll reset the buffer state
763 * to what it was before we touched the buffer.
765 bp->b_ops = fab->buf_ops;
766 fab->buf_ops->verify_read(bp);
774 * Some read verifiers will (re)set b_ops, so we must be
775 * careful not to change b_ops after running the verifier.
780 * This block passes the magic/uuid and verifier tests for this btree
781 * type. We don't need the caller to try the other tree types.
783 *done_with_block = true;
786 * Compare this btree block's level to the height of the current
787 * candidate root block.
789 * If the level matches the root we found previously, throw away both
790 * blocks because there can't be two candidate roots.
792 * If level is lower in the tree than the root we found previously,
795 block_level = xfs_btree_get_level(btblock);
796 if (block_level + 1 == fab->height) {
797 fab->root = NULLAGBLOCK;
799 } else if (block_level < fab->height) {
804 * This is the highest block in the tree that we've found so far.
805 * Update the btree height to reflect what we've learned from this
808 fab->height = block_level + 1;
811 * If this block doesn't have sibling pointers, then it's the new root
812 * block candidate. Otherwise, the root will be found farther up the
815 if (btblock->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) &&
816 btblock->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
819 fab->root = NULLAGBLOCK;
821 trace_xrep_findroot_block(mp, ri->sc->sa.pag->pag_agno, agbno,
822 be32_to_cpu(btblock->bb_magic), fab->height - 1);
824 xfs_trans_brelse(ri->sc->tp, bp);
829 * Do any of the blocks in this rmap record match one of the btrees we're
834 struct xfs_btree_cur *cur,
835 const struct xfs_rmap_irec *rec,
838 struct xrep_findroot *ri = priv;
839 struct xrep_find_ag_btree *fab;
844 /* Ignore anything that isn't AG metadata. */
845 if (!XFS_RMAP_NON_INODE_OWNER(rec->rm_owner))
848 /* Otherwise scan each block + btree type. */
849 for (b = 0; b < rec->rm_blockcount; b++) {
851 for (fab = ri->btree_info; fab->buf_ops; fab++) {
852 if (rec->rm_owner != fab->rmap_owner)
854 error = xrep_findroot_block(ri, fab,
855 rec->rm_owner, rec->rm_startblock + b,
867 /* Find the roots of the per-AG btrees described in btree_info. */
869 xrep_find_ag_btree_roots(
870 struct xfs_scrub *sc,
871 struct xfs_buf *agf_bp,
872 struct xrep_find_ag_btree *btree_info,
873 struct xfs_buf *agfl_bp)
875 struct xfs_mount *mp = sc->mp;
876 struct xrep_findroot ri;
877 struct xrep_find_ag_btree *fab;
878 struct xfs_btree_cur *cur;
881 ASSERT(xfs_buf_islocked(agf_bp));
882 ASSERT(agfl_bp == NULL || xfs_buf_islocked(agfl_bp));
885 ri.btree_info = btree_info;
886 ri.agf = agf_bp->b_addr;
887 ri.agfl_bp = agfl_bp;
888 for (fab = btree_info; fab->buf_ops; fab++) {
889 ASSERT(agfl_bp || fab->rmap_owner != XFS_RMAP_OWN_AG);
890 ASSERT(XFS_RMAP_NON_INODE_OWNER(fab->rmap_owner));
891 fab->root = NULLAGBLOCK;
895 cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag);
896 error = xfs_rmap_query_all(cur, xrep_findroot_rmap, &ri);
897 xfs_btree_del_cursor(cur, error);
902 /* Force a quotacheck the next time we mount. */
904 xrep_force_quotacheck(
905 struct xfs_scrub *sc,
910 flag = xfs_quota_chkd_flag(type);
911 if (!(flag & sc->mp->m_qflags))
914 mutex_lock(&sc->mp->m_quotainfo->qi_quotaofflock);
915 sc->mp->m_qflags &= ~flag;
916 spin_lock(&sc->mp->m_sb_lock);
917 sc->mp->m_sb.sb_qflags &= ~flag;
918 spin_unlock(&sc->mp->m_sb_lock);
920 mutex_unlock(&sc->mp->m_quotainfo->qi_quotaofflock);
924 * Attach dquots to this inode, or schedule quotacheck to fix them.
926 * This function ensures that the appropriate dquots are attached to an inode.
927 * We cannot allow the dquot code to allocate an on-disk dquot block here
928 * because we're already in transaction context with the inode locked. The
929 * on-disk dquot should already exist anyway. If the quota code signals
930 * corruption or missing quota information, schedule quotacheck, which will
931 * repair corruptions in the quota metadata.
935 struct xfs_scrub *sc)
939 error = xfs_qm_dqattach_locked(sc->ip, false);
944 xfs_err_ratelimited(sc->mp,
945 "inode %llu repair encountered quota error %d, quotacheck forced.",
946 (unsigned long long)sc->ip->i_ino, error);
947 if (XFS_IS_UQUOTA_ON(sc->mp) && !sc->ip->i_udquot)
948 xrep_force_quotacheck(sc, XFS_DQTYPE_USER);
949 if (XFS_IS_GQUOTA_ON(sc->mp) && !sc->ip->i_gdquot)
950 xrep_force_quotacheck(sc, XFS_DQTYPE_GROUP);
951 if (XFS_IS_PQUOTA_ON(sc->mp) && !sc->ip->i_pdquot)
952 xrep_force_quotacheck(sc, XFS_DQTYPE_PROJ);