xref: /linux/fs/xfs/scrub/refcount.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
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_defer.h"
13 #include "xfs_btree.h"
14 #include "xfs_bit.h"
15 #include "xfs_log_format.h"
16 #include "xfs_trans.h"
17 #include "xfs_sb.h"
18 #include "xfs_alloc.h"
19 #include "xfs_rmap.h"
20 #include "xfs_refcount.h"
21 #include "scrub/xfs_scrub.h"
22 #include "scrub/scrub.h"
23 #include "scrub/common.h"
24 #include "scrub/btree.h"
25 #include "scrub/trace.h"
26 
27 /*
28  * Set us up to scrub reference count btrees.
29  */
30 int
31 xchk_setup_ag_refcountbt(
32 	struct xfs_scrub	*sc,
33 	struct xfs_inode	*ip)
34 {
35 	return xchk_setup_ag_btree(sc, ip, false);
36 }
37 
38 /* Reference count btree scrubber. */
39 
40 /*
41  * Confirming Reference Counts via Reverse Mappings
42  *
43  * We want to count the reverse mappings overlapping a refcount record
44  * (bno, len, refcount), allowing for the possibility that some of the
45  * overlap may come from smaller adjoining reverse mappings, while some
46  * comes from single extents which overlap the range entirely.  The
47  * outer loop is as follows:
48  *
49  * 1. For all reverse mappings overlapping the refcount extent,
50  *    a. If a given rmap completely overlaps, mark it as seen.
51  *    b. Otherwise, record the fragment (in agbno order) for later
52  *       processing.
53  *
54  * Once we've seen all the rmaps, we know that for all blocks in the
55  * refcount record we want to find $refcount owners and we've already
56  * visited $seen extents that overlap all the blocks.  Therefore, we
57  * need to find ($refcount - $seen) owners for every block in the
58  * extent; call that quantity $target_nr.  Proceed as follows:
59  *
60  * 2. Pull the first $target_nr fragments from the list; all of them
61  *    should start at or before the start of the extent.
62  *    Call this subset of fragments the working set.
63  * 3. Until there are no more unprocessed fragments,
64  *    a. Find the shortest fragments in the set and remove them.
65  *    b. Note the block number of the end of these fragments.
66  *    c. Pull the same number of fragments from the list.  All of these
67  *       fragments should start at the block number recorded in the
68  *       previous step.
69  *    d. Put those fragments in the set.
70  * 4. Check that there are $target_nr fragments remaining in the list,
71  *    and that they all end at or beyond the end of the refcount extent.
72  *
73  * If the refcount is correct, all the check conditions in the algorithm
74  * should always hold true.  If not, the refcount is incorrect.
75  */
76 struct xchk_refcnt_frag {
77 	struct list_head	list;
78 	struct xfs_rmap_irec	rm;
79 };
80 
81 struct xchk_refcnt_check {
82 	struct xfs_scrub	*sc;
83 	struct list_head	fragments;
84 
85 	/* refcount extent we're examining */
86 	xfs_agblock_t		bno;
87 	xfs_extlen_t		len;
88 	xfs_nlink_t		refcount;
89 
90 	/* number of owners seen */
91 	xfs_nlink_t		seen;
92 };
93 
94 /*
95  * Decide if the given rmap is large enough that we can redeem it
96  * towards refcount verification now, or if it's a fragment, in
97  * which case we'll hang onto it in the hopes that we'll later
98  * discover that we've collected exactly the correct number of
99  * fragments as the refcountbt says we should have.
100  */
101 STATIC int
102 xchk_refcountbt_rmap_check(
103 	struct xfs_btree_cur		*cur,
104 	struct xfs_rmap_irec		*rec,
105 	void				*priv)
106 {
107 	struct xchk_refcnt_check	*refchk = priv;
108 	struct xchk_refcnt_frag		*frag;
109 	xfs_agblock_t			rm_last;
110 	xfs_agblock_t			rc_last;
111 	int				error = 0;
112 
113 	if (xchk_should_terminate(refchk->sc, &error))
114 		return error;
115 
116 	rm_last = rec->rm_startblock + rec->rm_blockcount - 1;
117 	rc_last = refchk->bno + refchk->len - 1;
118 
119 	/* Confirm that a single-owner refc extent is a CoW stage. */
120 	if (refchk->refcount == 1 && rec->rm_owner != XFS_RMAP_OWN_COW) {
121 		xchk_btree_xref_set_corrupt(refchk->sc, cur, 0);
122 		return 0;
123 	}
124 
125 	if (rec->rm_startblock <= refchk->bno && rm_last >= rc_last) {
126 		/*
127 		 * The rmap overlaps the refcount record, so we can confirm
128 		 * one refcount owner seen.
129 		 */
130 		refchk->seen++;
131 	} else {
132 		/*
133 		 * This rmap covers only part of the refcount record, so
134 		 * save the fragment for later processing.  If the rmapbt
135 		 * is healthy each rmap_irec we see will be in agbno order
136 		 * so we don't need insertion sort here.
137 		 */
138 		frag = kmem_alloc(sizeof(struct xchk_refcnt_frag),
139 				KM_MAYFAIL);
140 		if (!frag)
141 			return -ENOMEM;
142 		memcpy(&frag->rm, rec, sizeof(frag->rm));
143 		list_add_tail(&frag->list, &refchk->fragments);
144 	}
145 
146 	return 0;
147 }
148 
149 /*
150  * Given a bunch of rmap fragments, iterate through them, keeping
151  * a running tally of the refcount.  If this ever deviates from
152  * what we expect (which is the refcountbt's refcount minus the
153  * number of extents that totally covered the refcountbt extent),
154  * we have a refcountbt error.
155  */
156 STATIC void
157 xchk_refcountbt_process_rmap_fragments(
158 	struct xchk_refcnt_check	*refchk)
159 {
160 	struct list_head		worklist;
161 	struct xchk_refcnt_frag		*frag;
162 	struct xchk_refcnt_frag		*n;
163 	xfs_agblock_t			bno;
164 	xfs_agblock_t			rbno;
165 	xfs_agblock_t			next_rbno;
166 	xfs_nlink_t			nr;
167 	xfs_nlink_t			target_nr;
168 
169 	target_nr = refchk->refcount - refchk->seen;
170 	if (target_nr == 0)
171 		return;
172 
173 	/*
174 	 * There are (refchk->rc.rc_refcount - refchk->nr refcount)
175 	 * references we haven't found yet.  Pull that many off the
176 	 * fragment list and figure out where the smallest rmap ends
177 	 * (and therefore the next rmap should start).  All the rmaps
178 	 * we pull off should start at or before the beginning of the
179 	 * refcount record's range.
180 	 */
181 	INIT_LIST_HEAD(&worklist);
182 	rbno = NULLAGBLOCK;
183 	nr = 1;
184 
185 	/* Make sure the fragments actually /are/ in agbno order. */
186 	bno = 0;
187 	list_for_each_entry(frag, &refchk->fragments, list) {
188 		if (frag->rm.rm_startblock < bno)
189 			goto done;
190 		bno = frag->rm.rm_startblock;
191 	}
192 
193 	/*
194 	 * Find all the rmaps that start at or before the refc extent,
195 	 * and put them on the worklist.
196 	 */
197 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
198 		if (frag->rm.rm_startblock > refchk->bno)
199 			goto done;
200 		bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
201 		if (bno < rbno)
202 			rbno = bno;
203 		list_move_tail(&frag->list, &worklist);
204 		if (nr == target_nr)
205 			break;
206 		nr++;
207 	}
208 
209 	/*
210 	 * We should have found exactly $target_nr rmap fragments starting
211 	 * at or before the refcount extent.
212 	 */
213 	if (nr != target_nr)
214 		goto done;
215 
216 	while (!list_empty(&refchk->fragments)) {
217 		/* Discard any fragments ending at rbno from the worklist. */
218 		nr = 0;
219 		next_rbno = NULLAGBLOCK;
220 		list_for_each_entry_safe(frag, n, &worklist, list) {
221 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
222 			if (bno != rbno) {
223 				if (bno < next_rbno)
224 					next_rbno = bno;
225 				continue;
226 			}
227 			list_del(&frag->list);
228 			kmem_free(frag);
229 			nr++;
230 		}
231 
232 		/* Try to add nr rmaps starting at rbno to the worklist. */
233 		list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
234 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
235 			if (frag->rm.rm_startblock != rbno)
236 				goto done;
237 			list_move_tail(&frag->list, &worklist);
238 			if (next_rbno > bno)
239 				next_rbno = bno;
240 			nr--;
241 			if (nr == 0)
242 				break;
243 		}
244 
245 		/*
246 		 * If we get here and nr > 0, this means that we added fewer
247 		 * items to the worklist than we discarded because the fragment
248 		 * list ran out of items.  Therefore, we cannot maintain the
249 		 * required refcount.  Something is wrong, so we're done.
250 		 */
251 		if (nr)
252 			goto done;
253 
254 		rbno = next_rbno;
255 	}
256 
257 	/*
258 	 * Make sure the last extent we processed ends at or beyond
259 	 * the end of the refcount extent.
260 	 */
261 	if (rbno < refchk->bno + refchk->len)
262 		goto done;
263 
264 	/* Actually record us having seen the remaining refcount. */
265 	refchk->seen = refchk->refcount;
266 done:
267 	/* Delete fragments and work list. */
268 	list_for_each_entry_safe(frag, n, &worklist, list) {
269 		list_del(&frag->list);
270 		kmem_free(frag);
271 	}
272 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
273 		list_del(&frag->list);
274 		kmem_free(frag);
275 	}
276 }
277 
278 /* Use the rmap entries covering this extent to verify the refcount. */
279 STATIC void
280 xchk_refcountbt_xref_rmap(
281 	struct xfs_scrub		*sc,
282 	xfs_agblock_t			bno,
283 	xfs_extlen_t			len,
284 	xfs_nlink_t			refcount)
285 {
286 	struct xchk_refcnt_check	refchk = {
287 		.sc = sc,
288 		.bno = bno,
289 		.len = len,
290 		.refcount = refcount,
291 		.seen = 0,
292 	};
293 	struct xfs_rmap_irec		low;
294 	struct xfs_rmap_irec		high;
295 	struct xchk_refcnt_frag		*frag;
296 	struct xchk_refcnt_frag		*n;
297 	int				error;
298 
299 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
300 		return;
301 
302 	/* Cross-reference with the rmapbt to confirm the refcount. */
303 	memset(&low, 0, sizeof(low));
304 	low.rm_startblock = bno;
305 	memset(&high, 0xFF, sizeof(high));
306 	high.rm_startblock = bno + len - 1;
307 
308 	INIT_LIST_HEAD(&refchk.fragments);
309 	error = xfs_rmap_query_range(sc->sa.rmap_cur, &low, &high,
310 			&xchk_refcountbt_rmap_check, &refchk);
311 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
312 		goto out_free;
313 
314 	xchk_refcountbt_process_rmap_fragments(&refchk);
315 	if (refcount != refchk.seen)
316 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
317 
318 out_free:
319 	list_for_each_entry_safe(frag, n, &refchk.fragments, list) {
320 		list_del(&frag->list);
321 		kmem_free(frag);
322 	}
323 }
324 
325 /* Cross-reference with the other btrees. */
326 STATIC void
327 xchk_refcountbt_xref(
328 	struct xfs_scrub	*sc,
329 	xfs_agblock_t		agbno,
330 	xfs_extlen_t		len,
331 	xfs_nlink_t		refcount)
332 {
333 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
334 		return;
335 
336 	xchk_xref_is_used_space(sc, agbno, len);
337 	xchk_xref_is_not_inode_chunk(sc, agbno, len);
338 	xchk_refcountbt_xref_rmap(sc, agbno, len, refcount);
339 }
340 
341 /* Scrub a refcountbt record. */
342 STATIC int
343 xchk_refcountbt_rec(
344 	struct xchk_btree	*bs,
345 	union xfs_btree_rec	*rec)
346 {
347 	struct xfs_mount	*mp = bs->cur->bc_mp;
348 	xfs_agblock_t		*cow_blocks = bs->private;
349 	xfs_agnumber_t		agno = bs->cur->bc_private.a.agno;
350 	xfs_agblock_t		bno;
351 	xfs_extlen_t		len;
352 	xfs_nlink_t		refcount;
353 	bool			has_cowflag;
354 	int			error = 0;
355 
356 	bno = be32_to_cpu(rec->refc.rc_startblock);
357 	len = be32_to_cpu(rec->refc.rc_blockcount);
358 	refcount = be32_to_cpu(rec->refc.rc_refcount);
359 
360 	/* Only CoW records can have refcount == 1. */
361 	has_cowflag = (bno & XFS_REFC_COW_START);
362 	if ((refcount == 1 && !has_cowflag) || (refcount != 1 && has_cowflag))
363 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
364 	if (has_cowflag)
365 		(*cow_blocks) += len;
366 
367 	/* Check the extent. */
368 	bno &= ~XFS_REFC_COW_START;
369 	if (bno + len <= bno ||
370 	    !xfs_verify_agbno(mp, agno, bno) ||
371 	    !xfs_verify_agbno(mp, agno, bno + len - 1))
372 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
373 
374 	if (refcount == 0)
375 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
376 
377 	xchk_refcountbt_xref(bs->sc, bno, len, refcount);
378 
379 	return error;
380 }
381 
382 /* Make sure we have as many refc blocks as the rmap says. */
383 STATIC void
384 xchk_refcount_xref_rmap(
385 	struct xfs_scrub	*sc,
386 	xfs_filblks_t		cow_blocks)
387 {
388 	xfs_extlen_t		refcbt_blocks = 0;
389 	xfs_filblks_t		blocks;
390 	int			error;
391 
392 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
393 		return;
394 
395 	/* Check that we saw as many refcbt blocks as the rmap knows about. */
396 	error = xfs_btree_count_blocks(sc->sa.refc_cur, &refcbt_blocks);
397 	if (!xchk_btree_process_error(sc, sc->sa.refc_cur, 0, &error))
398 		return;
399 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
400 			&XFS_RMAP_OINFO_REFC, &blocks);
401 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
402 		return;
403 	if (blocks != refcbt_blocks)
404 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
405 
406 	/* Check that we saw as many cow blocks as the rmap knows about. */
407 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
408 			&XFS_RMAP_OINFO_COW, &blocks);
409 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
410 		return;
411 	if (blocks != cow_blocks)
412 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
413 }
414 
415 /* Scrub the refcount btree for some AG. */
416 int
417 xchk_refcountbt(
418 	struct xfs_scrub	*sc)
419 {
420 	xfs_agblock_t		cow_blocks = 0;
421 	int			error;
422 
423 	error = xchk_btree(sc, sc->sa.refc_cur, xchk_refcountbt_rec,
424 			&XFS_RMAP_OINFO_REFC, &cow_blocks);
425 	if (error)
426 		return error;
427 
428 	xchk_refcount_xref_rmap(sc, cow_blocks);
429 
430 	return 0;
431 }
432 
433 /* xref check that a cow staging extent is marked in the refcountbt. */
434 void
435 xchk_xref_is_cow_staging(
436 	struct xfs_scrub		*sc,
437 	xfs_agblock_t			agbno,
438 	xfs_extlen_t			len)
439 {
440 	struct xfs_refcount_irec	rc;
441 	bool				has_cowflag;
442 	int				has_refcount;
443 	int				error;
444 
445 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
446 		return;
447 
448 	/* Find the CoW staging extent. */
449 	error = xfs_refcount_lookup_le(sc->sa.refc_cur,
450 			agbno + XFS_REFC_COW_START, &has_refcount);
451 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
452 		return;
453 	if (!has_refcount) {
454 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
455 		return;
456 	}
457 
458 	error = xfs_refcount_get_rec(sc->sa.refc_cur, &rc, &has_refcount);
459 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
460 		return;
461 	if (!has_refcount) {
462 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
463 		return;
464 	}
465 
466 	/* CoW flag must be set, refcount must be 1. */
467 	has_cowflag = (rc.rc_startblock & XFS_REFC_COW_START);
468 	if (!has_cowflag || rc.rc_refcount != 1)
469 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
470 
471 	/* Must be at least as long as what was passed in */
472 	if (rc.rc_blockcount < len)
473 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
474 }
475 
476 /*
477  * xref check that the extent is not shared.  Only file data blocks
478  * can have multiple owners.
479  */
480 void
481 xchk_xref_is_not_shared(
482 	struct xfs_scrub	*sc,
483 	xfs_agblock_t		agbno,
484 	xfs_extlen_t		len)
485 {
486 	bool			shared;
487 	int			error;
488 
489 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
490 		return;
491 
492 	error = xfs_refcount_has_record(sc->sa.refc_cur, agbno, len, &shared);
493 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
494 		return;
495 	if (shared)
496 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
497 }
498