xref: /openbmc/linux/fs/xfs/scrub/repair.c (revision d7a74cad)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2018-2023 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
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_sb.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"
21 #include "xfs_rmap.h"
22 #include "xfs_rmap_btree.h"
23 #include "xfs_refcount_btree.h"
24 #include "xfs_extent_busy.h"
25 #include "xfs_ag.h"
26 #include "xfs_ag_resv.h"
27 #include "xfs_quota.h"
28 #include "xfs_qm.h"
29 #include "xfs_defer.h"
30 #include "scrub/scrub.h"
31 #include "scrub/common.h"
32 #include "scrub/trace.h"
33 #include "scrub/repair.h"
34 #include "scrub/bitmap.h"
35 #include "scrub/stats.h"
36 
37 /*
38  * Attempt to repair some metadata, if the metadata is corrupt and userspace
39  * told us to fix it.  This function returns -EAGAIN to mean "re-run scrub",
40  * and will set *fixed to true if it thinks it repaired anything.
41  */
42 int
xrep_attempt(struct xfs_scrub * sc,struct xchk_stats_run * run)43 xrep_attempt(
44 	struct xfs_scrub	*sc,
45 	struct xchk_stats_run	*run)
46 {
47 	u64			repair_start;
48 	int			error = 0;
49 
50 	trace_xrep_attempt(XFS_I(file_inode(sc->file)), sc->sm, error);
51 
52 	xchk_ag_btcur_free(&sc->sa);
53 
54 	/* Repair whatever's broken. */
55 	ASSERT(sc->ops->repair);
56 	run->repair_attempted = true;
57 	repair_start = xchk_stats_now();
58 	error = sc->ops->repair(sc);
59 	trace_xrep_done(XFS_I(file_inode(sc->file)), sc->sm, error);
60 	run->repair_ns += xchk_stats_elapsed_ns(repair_start);
61 	switch (error) {
62 	case 0:
63 		/*
64 		 * Repair succeeded.  Commit the fixes and perform a second
65 		 * scrub so that we can tell userspace if we fixed the problem.
66 		 */
67 		sc->sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT;
68 		sc->flags |= XREP_ALREADY_FIXED;
69 		run->repair_succeeded = true;
70 		return -EAGAIN;
71 	case -ECHRNG:
72 		sc->flags |= XCHK_NEED_DRAIN;
73 		run->retries++;
74 		return -EAGAIN;
75 	case -EDEADLOCK:
76 		/* Tell the caller to try again having grabbed all the locks. */
77 		if (!(sc->flags & XCHK_TRY_HARDER)) {
78 			sc->flags |= XCHK_TRY_HARDER;
79 			run->retries++;
80 			return -EAGAIN;
81 		}
82 		/*
83 		 * We tried harder but still couldn't grab all the resources
84 		 * we needed to fix it.  The corruption has not been fixed,
85 		 * so exit to userspace with the scan's output flags unchanged.
86 		 */
87 		return 0;
88 	default:
89 		/*
90 		 * EAGAIN tells the caller to re-scrub, so we cannot return
91 		 * that here.
92 		 */
93 		ASSERT(error != -EAGAIN);
94 		return error;
95 	}
96 }
97 
98 /*
99  * Complain about unfixable problems in the filesystem.  We don't log
100  * corruptions when IFLAG_REPAIR wasn't set on the assumption that the driver
101  * program is xfs_scrub, which will call back with IFLAG_REPAIR set if the
102  * administrator isn't running xfs_scrub in no-repairs mode.
103  *
104  * Use this helper function because _ratelimited silently declares a static
105  * structure to track rate limiting information.
106  */
107 void
xrep_failure(struct xfs_mount * mp)108 xrep_failure(
109 	struct xfs_mount	*mp)
110 {
111 	xfs_alert_ratelimited(mp,
112 "Corruption not fixed during online repair.  Unmount and run xfs_repair.");
113 }
114 
115 /*
116  * Repair probe -- userspace uses this to probe if we're willing to repair a
117  * given mountpoint.
118  */
119 int
xrep_probe(struct xfs_scrub * sc)120 xrep_probe(
121 	struct xfs_scrub	*sc)
122 {
123 	int			error = 0;
124 
125 	if (xchk_should_terminate(sc, &error))
126 		return error;
127 
128 	return 0;
129 }
130 
131 /*
132  * Roll a transaction, keeping the AG headers locked and reinitializing
133  * the btree cursors.
134  */
135 int
xrep_roll_ag_trans(struct xfs_scrub * sc)136 xrep_roll_ag_trans(
137 	struct xfs_scrub	*sc)
138 {
139 	int			error;
140 
141 	/*
142 	 * Keep the AG header buffers locked while we roll the transaction.
143 	 * Ensure that both AG buffers are dirty and held when we roll the
144 	 * transaction so that they move forward in the log without losing the
145 	 * bli (and hence the bli type) when the transaction commits.
146 	 *
147 	 * Normal code would never hold clean buffers across a roll, but repair
148 	 * needs both buffers to maintain a total lock on the AG.
149 	 */
150 	if (sc->sa.agi_bp) {
151 		xfs_ialloc_log_agi(sc->tp, sc->sa.agi_bp, XFS_AGI_MAGICNUM);
152 		xfs_trans_bhold(sc->tp, sc->sa.agi_bp);
153 	}
154 
155 	if (sc->sa.agf_bp) {
156 		xfs_alloc_log_agf(sc->tp, sc->sa.agf_bp, XFS_AGF_MAGICNUM);
157 		xfs_trans_bhold(sc->tp, sc->sa.agf_bp);
158 	}
159 
160 	/*
161 	 * Roll the transaction.  We still hold the AG header buffers locked
162 	 * regardless of whether or not that succeeds.  On failure, the buffers
163 	 * will be released during teardown on our way out of the kernel.  If
164 	 * successful, join the buffers to the new transaction and move on.
165 	 */
166 	error = xfs_trans_roll(&sc->tp);
167 	if (error)
168 		return error;
169 
170 	/* Join the AG headers to the new transaction. */
171 	if (sc->sa.agi_bp)
172 		xfs_trans_bjoin(sc->tp, sc->sa.agi_bp);
173 	if (sc->sa.agf_bp)
174 		xfs_trans_bjoin(sc->tp, sc->sa.agf_bp);
175 
176 	return 0;
177 }
178 
179 /* Finish all deferred work attached to the repair transaction. */
180 int
xrep_defer_finish(struct xfs_scrub * sc)181 xrep_defer_finish(
182 	struct xfs_scrub	*sc)
183 {
184 	int			error;
185 
186 	/*
187 	 * Keep the AG header buffers locked while we complete deferred work
188 	 * items.  Ensure that both AG buffers are dirty and held when we roll
189 	 * the transaction so that they move forward in the log without losing
190 	 * the bli (and hence the bli type) when the transaction commits.
191 	 *
192 	 * Normal code would never hold clean buffers across a roll, but repair
193 	 * needs both buffers to maintain a total lock on the AG.
194 	 */
195 	if (sc->sa.agi_bp) {
196 		xfs_ialloc_log_agi(sc->tp, sc->sa.agi_bp, XFS_AGI_MAGICNUM);
197 		xfs_trans_bhold(sc->tp, sc->sa.agi_bp);
198 	}
199 
200 	if (sc->sa.agf_bp) {
201 		xfs_alloc_log_agf(sc->tp, sc->sa.agf_bp, XFS_AGF_MAGICNUM);
202 		xfs_trans_bhold(sc->tp, sc->sa.agf_bp);
203 	}
204 
205 	/*
206 	 * Finish all deferred work items.  We still hold the AG header buffers
207 	 * locked regardless of whether or not that succeeds.  On failure, the
208 	 * buffers will be released during teardown on our way out of the
209 	 * kernel.  If successful, join the buffers to the new transaction
210 	 * and move on.
211 	 */
212 	error = xfs_defer_finish(&sc->tp);
213 	if (error)
214 		return error;
215 
216 	/*
217 	 * Release the hold that we set above because defer_finish won't do
218 	 * that for us.  The defer roll code redirties held buffers after each
219 	 * roll, so the AG header buffers should be ready for logging.
220 	 */
221 	if (sc->sa.agi_bp)
222 		xfs_trans_bhold_release(sc->tp, sc->sa.agi_bp);
223 	if (sc->sa.agf_bp)
224 		xfs_trans_bhold_release(sc->tp, sc->sa.agf_bp);
225 
226 	return 0;
227 }
228 
229 /*
230  * Does the given AG have enough space to rebuild a btree?  Neither AG
231  * reservation can be critical, and we must have enough space (factoring
232  * in AG reservations) to construct a whole btree.
233  */
234 bool
xrep_ag_has_space(struct xfs_perag * pag,xfs_extlen_t nr_blocks,enum xfs_ag_resv_type type)235 xrep_ag_has_space(
236 	struct xfs_perag	*pag,
237 	xfs_extlen_t		nr_blocks,
238 	enum xfs_ag_resv_type	type)
239 {
240 	return  !xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) &&
241 		!xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA) &&
242 		pag->pagf_freeblks > xfs_ag_resv_needed(pag, type) + nr_blocks;
243 }
244 
245 /*
246  * Figure out how many blocks to reserve for an AG repair.  We calculate the
247  * worst case estimate for the number of blocks we'd need to rebuild one of
248  * any type of per-AG btree.
249  */
250 xfs_extlen_t
xrep_calc_ag_resblks(struct xfs_scrub * sc)251 xrep_calc_ag_resblks(
252 	struct xfs_scrub		*sc)
253 {
254 	struct xfs_mount		*mp = sc->mp;
255 	struct xfs_scrub_metadata	*sm = sc->sm;
256 	struct xfs_perag		*pag;
257 	struct xfs_buf			*bp;
258 	xfs_agino_t			icount = NULLAGINO;
259 	xfs_extlen_t			aglen = NULLAGBLOCK;
260 	xfs_extlen_t			usedlen;
261 	xfs_extlen_t			freelen;
262 	xfs_extlen_t			bnobt_sz;
263 	xfs_extlen_t			inobt_sz;
264 	xfs_extlen_t			rmapbt_sz;
265 	xfs_extlen_t			refcbt_sz;
266 	int				error;
267 
268 	if (!(sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR))
269 		return 0;
270 
271 	pag = xfs_perag_get(mp, sm->sm_agno);
272 	if (xfs_perag_initialised_agi(pag)) {
273 		/* Use in-core icount if possible. */
274 		icount = pag->pagi_count;
275 	} else {
276 		/* Try to get the actual counters from disk. */
277 		error = xfs_ialloc_read_agi(pag, NULL, &bp);
278 		if (!error) {
279 			icount = pag->pagi_count;
280 			xfs_buf_relse(bp);
281 		}
282 	}
283 
284 	/* Now grab the block counters from the AGF. */
285 	error = xfs_alloc_read_agf(pag, NULL, 0, &bp);
286 	if (error) {
287 		aglen = pag->block_count;
288 		freelen = aglen;
289 		usedlen = aglen;
290 	} else {
291 		struct xfs_agf	*agf = bp->b_addr;
292 
293 		aglen = be32_to_cpu(agf->agf_length);
294 		freelen = be32_to_cpu(agf->agf_freeblks);
295 		usedlen = aglen - freelen;
296 		xfs_buf_relse(bp);
297 	}
298 
299 	/* If the icount is impossible, make some worst-case assumptions. */
300 	if (icount == NULLAGINO ||
301 	    !xfs_verify_agino(pag, icount)) {
302 		icount = pag->agino_max - pag->agino_min + 1;
303 	}
304 
305 	/* If the block counts are impossible, make worst-case assumptions. */
306 	if (aglen == NULLAGBLOCK ||
307 	    aglen != pag->block_count ||
308 	    freelen >= aglen) {
309 		aglen = pag->block_count;
310 		freelen = aglen;
311 		usedlen = aglen;
312 	}
313 	xfs_perag_put(pag);
314 
315 	trace_xrep_calc_ag_resblks(mp, sm->sm_agno, icount, aglen,
316 			freelen, usedlen);
317 
318 	/*
319 	 * Figure out how many blocks we'd need worst case to rebuild
320 	 * each type of btree.  Note that we can only rebuild the
321 	 * bnobt/cntbt or inobt/finobt as pairs.
322 	 */
323 	bnobt_sz = 2 * xfs_allocbt_calc_size(mp, freelen);
324 	if (xfs_has_sparseinodes(mp))
325 		inobt_sz = xfs_iallocbt_calc_size(mp, icount /
326 				XFS_INODES_PER_HOLEMASK_BIT);
327 	else
328 		inobt_sz = xfs_iallocbt_calc_size(mp, icount /
329 				XFS_INODES_PER_CHUNK);
330 	if (xfs_has_finobt(mp))
331 		inobt_sz *= 2;
332 	if (xfs_has_reflink(mp))
333 		refcbt_sz = xfs_refcountbt_calc_size(mp, usedlen);
334 	else
335 		refcbt_sz = 0;
336 	if (xfs_has_rmapbt(mp)) {
337 		/*
338 		 * Guess how many blocks we need to rebuild the rmapbt.
339 		 * For non-reflink filesystems we can't have more records than
340 		 * used blocks.  However, with reflink it's possible to have
341 		 * more than one rmap record per AG block.  We don't know how
342 		 * many rmaps there could be in the AG, so we start off with
343 		 * what we hope is an generous over-estimation.
344 		 */
345 		if (xfs_has_reflink(mp))
346 			rmapbt_sz = xfs_rmapbt_calc_size(mp,
347 					(unsigned long long)aglen * 2);
348 		else
349 			rmapbt_sz = xfs_rmapbt_calc_size(mp, usedlen);
350 	} else {
351 		rmapbt_sz = 0;
352 	}
353 
354 	trace_xrep_calc_ag_resblks_btsize(mp, sm->sm_agno, bnobt_sz,
355 			inobt_sz, rmapbt_sz, refcbt_sz);
356 
357 	return max(max(bnobt_sz, inobt_sz), max(rmapbt_sz, refcbt_sz));
358 }
359 
360 /*
361  * Reconstructing per-AG Btrees
362  *
363  * When a space btree is corrupt, we don't bother trying to fix it.  Instead,
364  * we scan secondary space metadata to derive the records that should be in
365  * the damaged btree, initialize a fresh btree root, and insert the records.
366  * Note that for rebuilding the rmapbt we scan all the primary data to
367  * generate the new records.
368  *
369  * However, that leaves the matter of removing all the metadata describing the
370  * old broken structure.  For primary metadata we use the rmap data to collect
371  * every extent with a matching rmap owner (bitmap); we then iterate all other
372  * metadata structures with the same rmap owner to collect the extents that
373  * cannot be removed (sublist).  We then subtract sublist from bitmap to
374  * derive the blocks that were used by the old btree.  These blocks can be
375  * reaped.
376  *
377  * For rmapbt reconstructions we must use different tactics for extent
378  * collection.  First we iterate all primary metadata (this excludes the old
379  * rmapbt, obviously) to generate new rmap records.  The gaps in the rmap
380  * records are collected as bitmap.  The bnobt records are collected as
381  * sublist.  As with the other btrees we subtract sublist from bitmap, and the
382  * result (since the rmapbt lives in the free space) are the blocks from the
383  * old rmapbt.
384  */
385 
386 /* Ensure the freelist is the correct size. */
387 int
xrep_fix_freelist(struct xfs_scrub * sc,bool can_shrink)388 xrep_fix_freelist(
389 	struct xfs_scrub	*sc,
390 	bool			can_shrink)
391 {
392 	struct xfs_alloc_arg	args = {0};
393 
394 	args.mp = sc->mp;
395 	args.tp = sc->tp;
396 	args.agno = sc->sa.pag->pag_agno;
397 	args.alignment = 1;
398 	args.pag = sc->sa.pag;
399 
400 	return xfs_alloc_fix_freelist(&args,
401 			can_shrink ? 0 : XFS_ALLOC_FLAG_NOSHRINK);
402 }
403 
404 /*
405  * Finding per-AG Btree Roots for AGF/AGI Reconstruction
406  *
407  * If the AGF or AGI become slightly corrupted, it may be necessary to rebuild
408  * the AG headers by using the rmap data to rummage through the AG looking for
409  * btree roots.  This is not guaranteed to work if the AG is heavily damaged
410  * or the rmap data are corrupt.
411  *
412  * Callers of xrep_find_ag_btree_roots must lock the AGF and AGFL
413  * buffers if the AGF is being rebuilt; or the AGF and AGI buffers if the
414  * AGI is being rebuilt.  It must maintain these locks until it's safe for
415  * other threads to change the btrees' shapes.  The caller provides
416  * information about the btrees to look for by passing in an array of
417  * xrep_find_ag_btree with the (rmap owner, buf_ops, magic) fields set.
418  * The (root, height) fields will be set on return if anything is found.  The
419  * last element of the array should have a NULL buf_ops to mark the end of the
420  * array.
421  *
422  * For every rmapbt record matching any of the rmap owners in btree_info,
423  * read each block referenced by the rmap record.  If the block is a btree
424  * block from this filesystem matching any of the magic numbers and has a
425  * level higher than what we've already seen, remember the block and the
426  * height of the tree required to have such a block.  When the call completes,
427  * we return the highest block we've found for each btree description; those
428  * should be the roots.
429  */
430 
431 struct xrep_findroot {
432 	struct xfs_scrub		*sc;
433 	struct xfs_buf			*agfl_bp;
434 	struct xfs_agf			*agf;
435 	struct xrep_find_ag_btree	*btree_info;
436 };
437 
438 /* See if our block is in the AGFL. */
439 STATIC int
xrep_findroot_agfl_walk(struct xfs_mount * mp,xfs_agblock_t bno,void * priv)440 xrep_findroot_agfl_walk(
441 	struct xfs_mount	*mp,
442 	xfs_agblock_t		bno,
443 	void			*priv)
444 {
445 	xfs_agblock_t		*agbno = priv;
446 
447 	return (*agbno == bno) ? -ECANCELED : 0;
448 }
449 
450 /* Does this block match the btree information passed in? */
451 STATIC int
xrep_findroot_block(struct xrep_findroot * ri,struct xrep_find_ag_btree * fab,uint64_t owner,xfs_agblock_t agbno,bool * done_with_block)452 xrep_findroot_block(
453 	struct xrep_findroot		*ri,
454 	struct xrep_find_ag_btree	*fab,
455 	uint64_t			owner,
456 	xfs_agblock_t			agbno,
457 	bool				*done_with_block)
458 {
459 	struct xfs_mount		*mp = ri->sc->mp;
460 	struct xfs_buf			*bp;
461 	struct xfs_btree_block		*btblock;
462 	xfs_daddr_t			daddr;
463 	int				block_level;
464 	int				error = 0;
465 
466 	daddr = XFS_AGB_TO_DADDR(mp, ri->sc->sa.pag->pag_agno, agbno);
467 
468 	/*
469 	 * Blocks in the AGFL have stale contents that might just happen to
470 	 * have a matching magic and uuid.  We don't want to pull these blocks
471 	 * in as part of a tree root, so we have to filter out the AGFL stuff
472 	 * here.  If the AGFL looks insane we'll just refuse to repair.
473 	 */
474 	if (owner == XFS_RMAP_OWN_AG) {
475 		error = xfs_agfl_walk(mp, ri->agf, ri->agfl_bp,
476 				xrep_findroot_agfl_walk, &agbno);
477 		if (error == -ECANCELED)
478 			return 0;
479 		if (error)
480 			return error;
481 	}
482 
483 	/*
484 	 * Read the buffer into memory so that we can see if it's a match for
485 	 * our btree type.  We have no clue if it is beforehand, and we want to
486 	 * avoid xfs_trans_read_buf's behavior of dumping the DONE state (which
487 	 * will cause needless disk reads in subsequent calls to this function)
488 	 * and logging metadata verifier failures.
489 	 *
490 	 * Therefore, pass in NULL buffer ops.  If the buffer was already in
491 	 * memory from some other caller it will already have b_ops assigned.
492 	 * If it was in memory from a previous unsuccessful findroot_block
493 	 * call, the buffer won't have b_ops but it should be clean and ready
494 	 * for us to try to verify if the read call succeeds.  The same applies
495 	 * if the buffer wasn't in memory at all.
496 	 *
497 	 * Note: If we never match a btree type with this buffer, it will be
498 	 * left in memory with NULL b_ops.  This shouldn't be a problem unless
499 	 * the buffer gets written.
500 	 */
501 	error = xfs_trans_read_buf(mp, ri->sc->tp, mp->m_ddev_targp, daddr,
502 			mp->m_bsize, 0, &bp, NULL);
503 	if (error)
504 		return error;
505 
506 	/* Ensure the block magic matches the btree type we're looking for. */
507 	btblock = XFS_BUF_TO_BLOCK(bp);
508 	ASSERT(fab->buf_ops->magic[1] != 0);
509 	if (btblock->bb_magic != fab->buf_ops->magic[1])
510 		goto out;
511 
512 	/*
513 	 * If the buffer already has ops applied and they're not the ones for
514 	 * this btree type, we know this block doesn't match the btree and we
515 	 * can bail out.
516 	 *
517 	 * If the buffer ops match ours, someone else has already validated
518 	 * the block for us, so we can move on to checking if this is a root
519 	 * block candidate.
520 	 *
521 	 * If the buffer does not have ops, nobody has successfully validated
522 	 * the contents and the buffer cannot be dirty.  If the magic, uuid,
523 	 * and structure match this btree type then we'll move on to checking
524 	 * if it's a root block candidate.  If there is no match, bail out.
525 	 */
526 	if (bp->b_ops) {
527 		if (bp->b_ops != fab->buf_ops)
528 			goto out;
529 	} else {
530 		ASSERT(!xfs_trans_buf_is_dirty(bp));
531 		if (!uuid_equal(&btblock->bb_u.s.bb_uuid,
532 				&mp->m_sb.sb_meta_uuid))
533 			goto out;
534 		/*
535 		 * Read verifiers can reference b_ops, so we set the pointer
536 		 * here.  If the verifier fails we'll reset the buffer state
537 		 * to what it was before we touched the buffer.
538 		 */
539 		bp->b_ops = fab->buf_ops;
540 		fab->buf_ops->verify_read(bp);
541 		if (bp->b_error) {
542 			bp->b_ops = NULL;
543 			bp->b_error = 0;
544 			goto out;
545 		}
546 
547 		/*
548 		 * Some read verifiers will (re)set b_ops, so we must be
549 		 * careful not to change b_ops after running the verifier.
550 		 */
551 	}
552 
553 	/*
554 	 * This block passes the magic/uuid and verifier tests for this btree
555 	 * type.  We don't need the caller to try the other tree types.
556 	 */
557 	*done_with_block = true;
558 
559 	/*
560 	 * Compare this btree block's level to the height of the current
561 	 * candidate root block.
562 	 *
563 	 * If the level matches the root we found previously, throw away both
564 	 * blocks because there can't be two candidate roots.
565 	 *
566 	 * If level is lower in the tree than the root we found previously,
567 	 * ignore this block.
568 	 */
569 	block_level = xfs_btree_get_level(btblock);
570 	if (block_level + 1 == fab->height) {
571 		fab->root = NULLAGBLOCK;
572 		goto out;
573 	} else if (block_level < fab->height) {
574 		goto out;
575 	}
576 
577 	/*
578 	 * This is the highest block in the tree that we've found so far.
579 	 * Update the btree height to reflect what we've learned from this
580 	 * block.
581 	 */
582 	fab->height = block_level + 1;
583 
584 	/*
585 	 * If this block doesn't have sibling pointers, then it's the new root
586 	 * block candidate.  Otherwise, the root will be found farther up the
587 	 * tree.
588 	 */
589 	if (btblock->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) &&
590 	    btblock->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
591 		fab->root = agbno;
592 	else
593 		fab->root = NULLAGBLOCK;
594 
595 	trace_xrep_findroot_block(mp, ri->sc->sa.pag->pag_agno, agbno,
596 			be32_to_cpu(btblock->bb_magic), fab->height - 1);
597 out:
598 	xfs_trans_brelse(ri->sc->tp, bp);
599 	return error;
600 }
601 
602 /*
603  * Do any of the blocks in this rmap record match one of the btrees we're
604  * looking for?
605  */
606 STATIC int
xrep_findroot_rmap(struct xfs_btree_cur * cur,const struct xfs_rmap_irec * rec,void * priv)607 xrep_findroot_rmap(
608 	struct xfs_btree_cur		*cur,
609 	const struct xfs_rmap_irec	*rec,
610 	void				*priv)
611 {
612 	struct xrep_findroot		*ri = priv;
613 	struct xrep_find_ag_btree	*fab;
614 	xfs_agblock_t			b;
615 	bool				done;
616 	int				error = 0;
617 
618 	/* Ignore anything that isn't AG metadata. */
619 	if (!XFS_RMAP_NON_INODE_OWNER(rec->rm_owner))
620 		return 0;
621 
622 	/* Otherwise scan each block + btree type. */
623 	for (b = 0; b < rec->rm_blockcount; b++) {
624 		done = false;
625 		for (fab = ri->btree_info; fab->buf_ops; fab++) {
626 			if (rec->rm_owner != fab->rmap_owner)
627 				continue;
628 			error = xrep_findroot_block(ri, fab,
629 					rec->rm_owner, rec->rm_startblock + b,
630 					&done);
631 			if (error)
632 				return error;
633 			if (done)
634 				break;
635 		}
636 	}
637 
638 	return 0;
639 }
640 
641 /* Find the roots of the per-AG btrees described in btree_info. */
642 int
xrep_find_ag_btree_roots(struct xfs_scrub * sc,struct xfs_buf * agf_bp,struct xrep_find_ag_btree * btree_info,struct xfs_buf * agfl_bp)643 xrep_find_ag_btree_roots(
644 	struct xfs_scrub		*sc,
645 	struct xfs_buf			*agf_bp,
646 	struct xrep_find_ag_btree	*btree_info,
647 	struct xfs_buf			*agfl_bp)
648 {
649 	struct xfs_mount		*mp = sc->mp;
650 	struct xrep_findroot		ri;
651 	struct xrep_find_ag_btree	*fab;
652 	struct xfs_btree_cur		*cur;
653 	int				error;
654 
655 	ASSERT(xfs_buf_islocked(agf_bp));
656 	ASSERT(agfl_bp == NULL || xfs_buf_islocked(agfl_bp));
657 
658 	ri.sc = sc;
659 	ri.btree_info = btree_info;
660 	ri.agf = agf_bp->b_addr;
661 	ri.agfl_bp = agfl_bp;
662 	for (fab = btree_info; fab->buf_ops; fab++) {
663 		ASSERT(agfl_bp || fab->rmap_owner != XFS_RMAP_OWN_AG);
664 		ASSERT(XFS_RMAP_NON_INODE_OWNER(fab->rmap_owner));
665 		fab->root = NULLAGBLOCK;
666 		fab->height = 0;
667 	}
668 
669 	cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag);
670 	error = xfs_rmap_query_all(cur, xrep_findroot_rmap, &ri);
671 	xfs_btree_del_cursor(cur, error);
672 
673 	return error;
674 }
675 
676 /* Force a quotacheck the next time we mount. */
677 void
xrep_force_quotacheck(struct xfs_scrub * sc,xfs_dqtype_t type)678 xrep_force_quotacheck(
679 	struct xfs_scrub	*sc,
680 	xfs_dqtype_t		type)
681 {
682 	uint			flag;
683 
684 	flag = xfs_quota_chkd_flag(type);
685 	if (!(flag & sc->mp->m_qflags))
686 		return;
687 
688 	mutex_lock(&sc->mp->m_quotainfo->qi_quotaofflock);
689 	sc->mp->m_qflags &= ~flag;
690 	spin_lock(&sc->mp->m_sb_lock);
691 	sc->mp->m_sb.sb_qflags &= ~flag;
692 	spin_unlock(&sc->mp->m_sb_lock);
693 	xfs_log_sb(sc->tp);
694 	mutex_unlock(&sc->mp->m_quotainfo->qi_quotaofflock);
695 }
696 
697 /*
698  * Attach dquots to this inode, or schedule quotacheck to fix them.
699  *
700  * This function ensures that the appropriate dquots are attached to an inode.
701  * We cannot allow the dquot code to allocate an on-disk dquot block here
702  * because we're already in transaction context with the inode locked.  The
703  * on-disk dquot should already exist anyway.  If the quota code signals
704  * corruption or missing quota information, schedule quotacheck, which will
705  * repair corruptions in the quota metadata.
706  */
707 int
xrep_ino_dqattach(struct xfs_scrub * sc)708 xrep_ino_dqattach(
709 	struct xfs_scrub	*sc)
710 {
711 	int			error;
712 
713 	error = xfs_qm_dqattach_locked(sc->ip, false);
714 	switch (error) {
715 	case -EFSBADCRC:
716 	case -EFSCORRUPTED:
717 	case -ENOENT:
718 		xfs_err_ratelimited(sc->mp,
719 "inode %llu repair encountered quota error %d, quotacheck forced.",
720 				(unsigned long long)sc->ip->i_ino, error);
721 		if (XFS_IS_UQUOTA_ON(sc->mp) && !sc->ip->i_udquot)
722 			xrep_force_quotacheck(sc, XFS_DQTYPE_USER);
723 		if (XFS_IS_GQUOTA_ON(sc->mp) && !sc->ip->i_gdquot)
724 			xrep_force_quotacheck(sc, XFS_DQTYPE_GROUP);
725 		if (XFS_IS_PQUOTA_ON(sc->mp) && !sc->ip->i_pdquot)
726 			xrep_force_quotacheck(sc, XFS_DQTYPE_PROJ);
727 		fallthrough;
728 	case -ESRCH:
729 		error = 0;
730 		break;
731 	default:
732 		break;
733 	}
734 
735 	return error;
736 }
737