xref: /openbmc/linux/fs/xfs/scrub/refcount.c (revision 89df62c3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2017-2023 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_ag.h"
13 #include "xfs_btree.h"
14 #include "xfs_rmap.h"
15 #include "xfs_refcount.h"
16 #include "scrub/scrub.h"
17 #include "scrub/common.h"
18 #include "scrub/btree.h"
19 #include "scrub/trace.h"
20 
21 /*
22  * Set us up to scrub reference count btrees.
23  */
24 int
25 xchk_setup_ag_refcountbt(
26 	struct xfs_scrub	*sc)
27 {
28 	if (xchk_need_intent_drain(sc))
29 		xchk_fsgates_enable(sc, XCHK_FSGATES_DRAIN);
30 	return xchk_setup_ag_btree(sc, false);
31 }
32 
33 /* Reference count btree scrubber. */
34 
35 /*
36  * Confirming Reference Counts via Reverse Mappings
37  *
38  * We want to count the reverse mappings overlapping a refcount record
39  * (bno, len, refcount), allowing for the possibility that some of the
40  * overlap may come from smaller adjoining reverse mappings, while some
41  * comes from single extents which overlap the range entirely.  The
42  * outer loop is as follows:
43  *
44  * 1. For all reverse mappings overlapping the refcount extent,
45  *    a. If a given rmap completely overlaps, mark it as seen.
46  *    b. Otherwise, record the fragment (in agbno order) for later
47  *       processing.
48  *
49  * Once we've seen all the rmaps, we know that for all blocks in the
50  * refcount record we want to find $refcount owners and we've already
51  * visited $seen extents that overlap all the blocks.  Therefore, we
52  * need to find ($refcount - $seen) owners for every block in the
53  * extent; call that quantity $target_nr.  Proceed as follows:
54  *
55  * 2. Pull the first $target_nr fragments from the list; all of them
56  *    should start at or before the start of the extent.
57  *    Call this subset of fragments the working set.
58  * 3. Until there are no more unprocessed fragments,
59  *    a. Find the shortest fragments in the set and remove them.
60  *    b. Note the block number of the end of these fragments.
61  *    c. Pull the same number of fragments from the list.  All of these
62  *       fragments should start at the block number recorded in the
63  *       previous step.
64  *    d. Put those fragments in the set.
65  * 4. Check that there are $target_nr fragments remaining in the list,
66  *    and that they all end at or beyond the end of the refcount extent.
67  *
68  * If the refcount is correct, all the check conditions in the algorithm
69  * should always hold true.  If not, the refcount is incorrect.
70  */
71 struct xchk_refcnt_frag {
72 	struct list_head	list;
73 	struct xfs_rmap_irec	rm;
74 };
75 
76 struct xchk_refcnt_check {
77 	struct xfs_scrub	*sc;
78 	struct list_head	fragments;
79 
80 	/* refcount extent we're examining */
81 	xfs_agblock_t		bno;
82 	xfs_extlen_t		len;
83 	xfs_nlink_t		refcount;
84 
85 	/* number of owners seen */
86 	xfs_nlink_t		seen;
87 };
88 
89 /*
90  * Decide if the given rmap is large enough that we can redeem it
91  * towards refcount verification now, or if it's a fragment, in
92  * which case we'll hang onto it in the hopes that we'll later
93  * discover that we've collected exactly the correct number of
94  * fragments as the refcountbt says we should have.
95  */
96 STATIC int
97 xchk_refcountbt_rmap_check(
98 	struct xfs_btree_cur		*cur,
99 	const struct xfs_rmap_irec	*rec,
100 	void				*priv)
101 {
102 	struct xchk_refcnt_check	*refchk = priv;
103 	struct xchk_refcnt_frag		*frag;
104 	xfs_agblock_t			rm_last;
105 	xfs_agblock_t			rc_last;
106 	int				error = 0;
107 
108 	if (xchk_should_terminate(refchk->sc, &error))
109 		return error;
110 
111 	rm_last = rec->rm_startblock + rec->rm_blockcount - 1;
112 	rc_last = refchk->bno + refchk->len - 1;
113 
114 	/* Confirm that a single-owner refc extent is a CoW stage. */
115 	if (refchk->refcount == 1 && rec->rm_owner != XFS_RMAP_OWN_COW) {
116 		xchk_btree_xref_set_corrupt(refchk->sc, cur, 0);
117 		return 0;
118 	}
119 
120 	if (rec->rm_startblock <= refchk->bno && rm_last >= rc_last) {
121 		/*
122 		 * The rmap overlaps the refcount record, so we can confirm
123 		 * one refcount owner seen.
124 		 */
125 		refchk->seen++;
126 	} else {
127 		/*
128 		 * This rmap covers only part of the refcount record, so
129 		 * save the fragment for later processing.  If the rmapbt
130 		 * is healthy each rmap_irec we see will be in agbno order
131 		 * so we don't need insertion sort here.
132 		 */
133 		frag = kmalloc(sizeof(struct xchk_refcnt_frag),
134 				XCHK_GFP_FLAGS);
135 		if (!frag)
136 			return -ENOMEM;
137 		memcpy(&frag->rm, rec, sizeof(frag->rm));
138 		list_add_tail(&frag->list, &refchk->fragments);
139 	}
140 
141 	return 0;
142 }
143 
144 /*
145  * Given a bunch of rmap fragments, iterate through them, keeping
146  * a running tally of the refcount.  If this ever deviates from
147  * what we expect (which is the refcountbt's refcount minus the
148  * number of extents that totally covered the refcountbt extent),
149  * we have a refcountbt error.
150  */
151 STATIC void
152 xchk_refcountbt_process_rmap_fragments(
153 	struct xchk_refcnt_check	*refchk)
154 {
155 	struct list_head		worklist;
156 	struct xchk_refcnt_frag		*frag;
157 	struct xchk_refcnt_frag		*n;
158 	xfs_agblock_t			bno;
159 	xfs_agblock_t			rbno;
160 	xfs_agblock_t			next_rbno;
161 	xfs_nlink_t			nr;
162 	xfs_nlink_t			target_nr;
163 
164 	target_nr = refchk->refcount - refchk->seen;
165 	if (target_nr == 0)
166 		return;
167 
168 	/*
169 	 * There are (refchk->rc.rc_refcount - refchk->nr refcount)
170 	 * references we haven't found yet.  Pull that many off the
171 	 * fragment list and figure out where the smallest rmap ends
172 	 * (and therefore the next rmap should start).  All the rmaps
173 	 * we pull off should start at or before the beginning of the
174 	 * refcount record's range.
175 	 */
176 	INIT_LIST_HEAD(&worklist);
177 	rbno = NULLAGBLOCK;
178 
179 	/* Make sure the fragments actually /are/ in agbno order. */
180 	bno = 0;
181 	list_for_each_entry(frag, &refchk->fragments, list) {
182 		if (frag->rm.rm_startblock < bno)
183 			goto done;
184 		bno = frag->rm.rm_startblock;
185 	}
186 
187 	/*
188 	 * Find all the rmaps that start at or before the refc extent,
189 	 * and put them on the worklist.
190 	 */
191 	nr = 0;
192 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
193 		if (frag->rm.rm_startblock > refchk->bno || nr > target_nr)
194 			break;
195 		bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
196 		if (bno < rbno)
197 			rbno = bno;
198 		list_move_tail(&frag->list, &worklist);
199 		nr++;
200 	}
201 
202 	/*
203 	 * We should have found exactly $target_nr rmap fragments starting
204 	 * at or before the refcount extent.
205 	 */
206 	if (nr != target_nr)
207 		goto done;
208 
209 	while (!list_empty(&refchk->fragments)) {
210 		/* Discard any fragments ending at rbno from the worklist. */
211 		nr = 0;
212 		next_rbno = NULLAGBLOCK;
213 		list_for_each_entry_safe(frag, n, &worklist, list) {
214 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
215 			if (bno != rbno) {
216 				if (bno < next_rbno)
217 					next_rbno = bno;
218 				continue;
219 			}
220 			list_del(&frag->list);
221 			kfree(frag);
222 			nr++;
223 		}
224 
225 		/* Try to add nr rmaps starting at rbno to the worklist. */
226 		list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
227 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
228 			if (frag->rm.rm_startblock != rbno)
229 				goto done;
230 			list_move_tail(&frag->list, &worklist);
231 			if (next_rbno > bno)
232 				next_rbno = bno;
233 			nr--;
234 			if (nr == 0)
235 				break;
236 		}
237 
238 		/*
239 		 * If we get here and nr > 0, this means that we added fewer
240 		 * items to the worklist than we discarded because the fragment
241 		 * list ran out of items.  Therefore, we cannot maintain the
242 		 * required refcount.  Something is wrong, so we're done.
243 		 */
244 		if (nr)
245 			goto done;
246 
247 		rbno = next_rbno;
248 	}
249 
250 	/*
251 	 * Make sure the last extent we processed ends at or beyond
252 	 * the end of the refcount extent.
253 	 */
254 	if (rbno < refchk->bno + refchk->len)
255 		goto done;
256 
257 	/* Actually record us having seen the remaining refcount. */
258 	refchk->seen = refchk->refcount;
259 done:
260 	/* Delete fragments and work list. */
261 	list_for_each_entry_safe(frag, n, &worklist, list) {
262 		list_del(&frag->list);
263 		kfree(frag);
264 	}
265 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
266 		list_del(&frag->list);
267 		kfree(frag);
268 	}
269 }
270 
271 /* Use the rmap entries covering this extent to verify the refcount. */
272 STATIC void
273 xchk_refcountbt_xref_rmap(
274 	struct xfs_scrub		*sc,
275 	const struct xfs_refcount_irec	*irec)
276 {
277 	struct xchk_refcnt_check	refchk = {
278 		.sc			= sc,
279 		.bno			= irec->rc_startblock,
280 		.len			= irec->rc_blockcount,
281 		.refcount		= irec->rc_refcount,
282 		.seen = 0,
283 	};
284 	struct xfs_rmap_irec		low;
285 	struct xfs_rmap_irec		high;
286 	struct xchk_refcnt_frag		*frag;
287 	struct xchk_refcnt_frag		*n;
288 	int				error;
289 
290 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
291 		return;
292 
293 	/* Cross-reference with the rmapbt to confirm the refcount. */
294 	memset(&low, 0, sizeof(low));
295 	low.rm_startblock = irec->rc_startblock;
296 	memset(&high, 0xFF, sizeof(high));
297 	high.rm_startblock = irec->rc_startblock + irec->rc_blockcount - 1;
298 
299 	INIT_LIST_HEAD(&refchk.fragments);
300 	error = xfs_rmap_query_range(sc->sa.rmap_cur, &low, &high,
301 			&xchk_refcountbt_rmap_check, &refchk);
302 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
303 		goto out_free;
304 
305 	xchk_refcountbt_process_rmap_fragments(&refchk);
306 	if (irec->rc_refcount != refchk.seen) {
307 		trace_xchk_refcount_incorrect(sc->sa.pag, irec, refchk.seen);
308 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
309 	}
310 
311 out_free:
312 	list_for_each_entry_safe(frag, n, &refchk.fragments, list) {
313 		list_del(&frag->list);
314 		kfree(frag);
315 	}
316 }
317 
318 /* Cross-reference with the other btrees. */
319 STATIC void
320 xchk_refcountbt_xref(
321 	struct xfs_scrub		*sc,
322 	const struct xfs_refcount_irec	*irec)
323 {
324 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
325 		return;
326 
327 	xchk_xref_is_used_space(sc, irec->rc_startblock, irec->rc_blockcount);
328 	xchk_xref_is_not_inode_chunk(sc, irec->rc_startblock,
329 			irec->rc_blockcount);
330 	xchk_refcountbt_xref_rmap(sc, irec);
331 }
332 
333 struct xchk_refcbt_records {
334 	/* Previous refcount record. */
335 	struct xfs_refcount_irec prev_rec;
336 
337 	/* The next AG block where we aren't expecting shared extents. */
338 	xfs_agblock_t		next_unshared_agbno;
339 
340 	/* Number of CoW blocks we expect. */
341 	xfs_agblock_t		cow_blocks;
342 
343 	/* Was the last record a shared or CoW staging extent? */
344 	enum xfs_refc_domain	prev_domain;
345 };
346 
347 STATIC int
348 xchk_refcountbt_rmap_check_gap(
349 	struct xfs_btree_cur		*cur,
350 	const struct xfs_rmap_irec	*rec,
351 	void				*priv)
352 {
353 	xfs_agblock_t			*next_bno = priv;
354 
355 	if (*next_bno != NULLAGBLOCK && rec->rm_startblock < *next_bno)
356 		return -ECANCELED;
357 
358 	*next_bno = rec->rm_startblock + rec->rm_blockcount;
359 	return 0;
360 }
361 
362 /*
363  * Make sure that a gap in the reference count records does not correspond to
364  * overlapping records (i.e. shared extents) in the reverse mappings.
365  */
366 static inline void
367 xchk_refcountbt_xref_gaps(
368 	struct xfs_scrub	*sc,
369 	struct xchk_refcbt_records *rrc,
370 	xfs_agblock_t		bno)
371 {
372 	struct xfs_rmap_irec	low;
373 	struct xfs_rmap_irec	high;
374 	xfs_agblock_t		next_bno = NULLAGBLOCK;
375 	int			error;
376 
377 	if (bno <= rrc->next_unshared_agbno || !sc->sa.rmap_cur ||
378             xchk_skip_xref(sc->sm))
379 		return;
380 
381 	memset(&low, 0, sizeof(low));
382 	low.rm_startblock = rrc->next_unshared_agbno;
383 	memset(&high, 0xFF, sizeof(high));
384 	high.rm_startblock = bno - 1;
385 
386 	error = xfs_rmap_query_range(sc->sa.rmap_cur, &low, &high,
387 			xchk_refcountbt_rmap_check_gap, &next_bno);
388 	if (error == -ECANCELED)
389 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
390 	else
391 		xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur);
392 }
393 
394 static inline bool
395 xchk_refcount_mergeable(
396 	struct xchk_refcbt_records	*rrc,
397 	const struct xfs_refcount_irec	*r2)
398 {
399 	const struct xfs_refcount_irec	*r1 = &rrc->prev_rec;
400 
401 	/* Ignore if prev_rec is not yet initialized. */
402 	if (r1->rc_blockcount > 0)
403 		return false;
404 
405 	if (r1->rc_domain != r2->rc_domain)
406 		return false;
407 	if (r1->rc_startblock + r1->rc_blockcount != r2->rc_startblock)
408 		return false;
409 	if (r1->rc_refcount != r2->rc_refcount)
410 		return false;
411 	if ((unsigned long long)r1->rc_blockcount + r2->rc_blockcount >
412 			MAXREFCEXTLEN)
413 		return false;
414 
415 	return true;
416 }
417 
418 /* Flag failures for records that could be merged. */
419 STATIC void
420 xchk_refcountbt_check_mergeable(
421 	struct xchk_btree		*bs,
422 	struct xchk_refcbt_records	*rrc,
423 	const struct xfs_refcount_irec	*irec)
424 {
425 	if (bs->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
426 		return;
427 
428 	if (xchk_refcount_mergeable(rrc, irec))
429 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
430 
431 	memcpy(&rrc->prev_rec, irec, sizeof(struct xfs_refcount_irec));
432 }
433 
434 /* Scrub a refcountbt record. */
435 STATIC int
436 xchk_refcountbt_rec(
437 	struct xchk_btree	*bs,
438 	const union xfs_btree_rec *rec)
439 {
440 	struct xfs_refcount_irec irec;
441 	struct xchk_refcbt_records *rrc = bs->private;
442 
443 	xfs_refcount_btrec_to_irec(rec, &irec);
444 	if (xfs_refcount_check_irec(bs->cur, &irec) != NULL) {
445 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
446 		return 0;
447 	}
448 
449 	if (irec.rc_domain == XFS_REFC_DOMAIN_COW)
450 		rrc->cow_blocks += irec.rc_blockcount;
451 
452 	/* Shared records always come before CoW records. */
453 	if (irec.rc_domain == XFS_REFC_DOMAIN_SHARED &&
454 	    rrc->prev_domain == XFS_REFC_DOMAIN_COW)
455 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
456 	rrc->prev_domain = irec.rc_domain;
457 
458 	xchk_refcountbt_check_mergeable(bs, rrc, &irec);
459 	xchk_refcountbt_xref(bs->sc, &irec);
460 
461 	/*
462 	 * If this is a record for a shared extent, check that all blocks
463 	 * between the previous record and this one have at most one reverse
464 	 * mapping.
465 	 */
466 	if (irec.rc_domain == XFS_REFC_DOMAIN_SHARED) {
467 		xchk_refcountbt_xref_gaps(bs->sc, rrc, irec.rc_startblock);
468 		rrc->next_unshared_agbno = irec.rc_startblock +
469 					   irec.rc_blockcount;
470 	}
471 
472 	return 0;
473 }
474 
475 /* Make sure we have as many refc blocks as the rmap says. */
476 STATIC void
477 xchk_refcount_xref_rmap(
478 	struct xfs_scrub	*sc,
479 	xfs_filblks_t		cow_blocks)
480 {
481 	xfs_extlen_t		refcbt_blocks = 0;
482 	xfs_filblks_t		blocks;
483 	int			error;
484 
485 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
486 		return;
487 
488 	/* Check that we saw as many refcbt blocks as the rmap knows about. */
489 	error = xfs_btree_count_blocks(sc->sa.refc_cur, &refcbt_blocks);
490 	if (!xchk_btree_process_error(sc, sc->sa.refc_cur, 0, &error))
491 		return;
492 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
493 			&XFS_RMAP_OINFO_REFC, &blocks);
494 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
495 		return;
496 	if (blocks != refcbt_blocks)
497 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
498 
499 	/* Check that we saw as many cow blocks as the rmap knows about. */
500 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
501 			&XFS_RMAP_OINFO_COW, &blocks);
502 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
503 		return;
504 	if (blocks != cow_blocks)
505 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
506 }
507 
508 /* Scrub the refcount btree for some AG. */
509 int
510 xchk_refcountbt(
511 	struct xfs_scrub	*sc)
512 {
513 	struct xchk_refcbt_records rrc = {
514 		.cow_blocks		= 0,
515 		.next_unshared_agbno	= 0,
516 		.prev_domain		= XFS_REFC_DOMAIN_SHARED,
517 	};
518 	int			error;
519 
520 	error = xchk_btree(sc, sc->sa.refc_cur, xchk_refcountbt_rec,
521 			&XFS_RMAP_OINFO_REFC, &rrc);
522 	if (error)
523 		return error;
524 
525 	/*
526 	 * Check that all blocks between the last refcount > 1 record and the
527 	 * end of the AG have at most one reverse mapping.
528 	 */
529 	xchk_refcountbt_xref_gaps(sc, &rrc, sc->mp->m_sb.sb_agblocks);
530 
531 	xchk_refcount_xref_rmap(sc, rrc.cow_blocks);
532 
533 	return 0;
534 }
535 
536 /* xref check that a cow staging extent is marked in the refcountbt. */
537 void
538 xchk_xref_is_cow_staging(
539 	struct xfs_scrub		*sc,
540 	xfs_agblock_t			agbno,
541 	xfs_extlen_t			len)
542 {
543 	struct xfs_refcount_irec	rc;
544 	int				has_refcount;
545 	int				error;
546 
547 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
548 		return;
549 
550 	/* Find the CoW staging extent. */
551 	error = xfs_refcount_lookup_le(sc->sa.refc_cur, XFS_REFC_DOMAIN_COW,
552 			agbno, &has_refcount);
553 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
554 		return;
555 	if (!has_refcount) {
556 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
557 		return;
558 	}
559 
560 	error = xfs_refcount_get_rec(sc->sa.refc_cur, &rc, &has_refcount);
561 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
562 		return;
563 	if (!has_refcount) {
564 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
565 		return;
566 	}
567 
568 	/* CoW lookup returned a shared extent record? */
569 	if (rc.rc_domain != XFS_REFC_DOMAIN_COW)
570 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
571 
572 	/* Must be at least as long as what was passed in */
573 	if (rc.rc_blockcount < len)
574 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
575 }
576 
577 /*
578  * xref check that the extent is not shared.  Only file data blocks
579  * can have multiple owners.
580  */
581 void
582 xchk_xref_is_not_shared(
583 	struct xfs_scrub	*sc,
584 	xfs_agblock_t		agbno,
585 	xfs_extlen_t		len)
586 {
587 	enum xbtree_recpacking	outcome;
588 	int			error;
589 
590 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
591 		return;
592 
593 	error = xfs_refcount_has_records(sc->sa.refc_cur,
594 			XFS_REFC_DOMAIN_SHARED, agbno, len, &outcome);
595 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
596 		return;
597 	if (outcome != XBTREE_RECPACKING_EMPTY)
598 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
599 }
600 
601 /* xref check that the extent is not being used for CoW staging. */
602 void
603 xchk_xref_is_not_cow_staging(
604 	struct xfs_scrub	*sc,
605 	xfs_agblock_t		agbno,
606 	xfs_extlen_t		len)
607 {
608 	enum xbtree_recpacking	outcome;
609 	int			error;
610 
611 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
612 		return;
613 
614 	error = xfs_refcount_has_records(sc->sa.refc_cur, XFS_REFC_DOMAIN_COW,
615 			agbno, len, &outcome);
616 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
617 		return;
618 	if (outcome != XBTREE_RECPACKING_EMPTY)
619 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
620 }
621