xref: /linux/fs/xfs/scrub/refcount.c (revision 8a79db5e83a5d52c74e6f3c40d6f312cf899213e)
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_btree.h"
11 #include "xfs_rmap.h"
12 #include "xfs_refcount.h"
13 #include "scrub/scrub.h"
14 #include "scrub/common.h"
15 #include "scrub/btree.h"
16 
17 /*
18  * Set us up to scrub reference count btrees.
19  */
20 int
21 xchk_setup_ag_refcountbt(
22 	struct xfs_scrub	*sc,
23 	struct xfs_inode	*ip)
24 {
25 	return xchk_setup_ag_btree(sc, ip, false);
26 }
27 
28 /* Reference count btree scrubber. */
29 
30 /*
31  * Confirming Reference Counts via Reverse Mappings
32  *
33  * We want to count the reverse mappings overlapping a refcount record
34  * (bno, len, refcount), allowing for the possibility that some of the
35  * overlap may come from smaller adjoining reverse mappings, while some
36  * comes from single extents which overlap the range entirely.  The
37  * outer loop is as follows:
38  *
39  * 1. For all reverse mappings overlapping the refcount extent,
40  *    a. If a given rmap completely overlaps, mark it as seen.
41  *    b. Otherwise, record the fragment (in agbno order) for later
42  *       processing.
43  *
44  * Once we've seen all the rmaps, we know that for all blocks in the
45  * refcount record we want to find $refcount owners and we've already
46  * visited $seen extents that overlap all the blocks.  Therefore, we
47  * need to find ($refcount - $seen) owners for every block in the
48  * extent; call that quantity $target_nr.  Proceed as follows:
49  *
50  * 2. Pull the first $target_nr fragments from the list; all of them
51  *    should start at or before the start of the extent.
52  *    Call this subset of fragments the working set.
53  * 3. Until there are no more unprocessed fragments,
54  *    a. Find the shortest fragments in the set and remove them.
55  *    b. Note the block number of the end of these fragments.
56  *    c. Pull the same number of fragments from the list.  All of these
57  *       fragments should start at the block number recorded in the
58  *       previous step.
59  *    d. Put those fragments in the set.
60  * 4. Check that there are $target_nr fragments remaining in the list,
61  *    and that they all end at or beyond the end of the refcount extent.
62  *
63  * If the refcount is correct, all the check conditions in the algorithm
64  * should always hold true.  If not, the refcount is incorrect.
65  */
66 struct xchk_refcnt_frag {
67 	struct list_head	list;
68 	struct xfs_rmap_irec	rm;
69 };
70 
71 struct xchk_refcnt_check {
72 	struct xfs_scrub	*sc;
73 	struct list_head	fragments;
74 
75 	/* refcount extent we're examining */
76 	xfs_agblock_t		bno;
77 	xfs_extlen_t		len;
78 	xfs_nlink_t		refcount;
79 
80 	/* number of owners seen */
81 	xfs_nlink_t		seen;
82 };
83 
84 /*
85  * Decide if the given rmap is large enough that we can redeem it
86  * towards refcount verification now, or if it's a fragment, in
87  * which case we'll hang onto it in the hopes that we'll later
88  * discover that we've collected exactly the correct number of
89  * fragments as the refcountbt says we should have.
90  */
91 STATIC int
92 xchk_refcountbt_rmap_check(
93 	struct xfs_btree_cur		*cur,
94 	struct xfs_rmap_irec		*rec,
95 	void				*priv)
96 {
97 	struct xchk_refcnt_check	*refchk = priv;
98 	struct xchk_refcnt_frag		*frag;
99 	xfs_agblock_t			rm_last;
100 	xfs_agblock_t			rc_last;
101 	int				error = 0;
102 
103 	if (xchk_should_terminate(refchk->sc, &error))
104 		return error;
105 
106 	rm_last = rec->rm_startblock + rec->rm_blockcount - 1;
107 	rc_last = refchk->bno + refchk->len - 1;
108 
109 	/* Confirm that a single-owner refc extent is a CoW stage. */
110 	if (refchk->refcount == 1 && rec->rm_owner != XFS_RMAP_OWN_COW) {
111 		xchk_btree_xref_set_corrupt(refchk->sc, cur, 0);
112 		return 0;
113 	}
114 
115 	if (rec->rm_startblock <= refchk->bno && rm_last >= rc_last) {
116 		/*
117 		 * The rmap overlaps the refcount record, so we can confirm
118 		 * one refcount owner seen.
119 		 */
120 		refchk->seen++;
121 	} else {
122 		/*
123 		 * This rmap covers only part of the refcount record, so
124 		 * save the fragment for later processing.  If the rmapbt
125 		 * is healthy each rmap_irec we see will be in agbno order
126 		 * so we don't need insertion sort here.
127 		 */
128 		frag = kmem_alloc(sizeof(struct xchk_refcnt_frag),
129 				KM_MAYFAIL);
130 		if (!frag)
131 			return -ENOMEM;
132 		memcpy(&frag->rm, rec, sizeof(frag->rm));
133 		list_add_tail(&frag->list, &refchk->fragments);
134 	}
135 
136 	return 0;
137 }
138 
139 /*
140  * Given a bunch of rmap fragments, iterate through them, keeping
141  * a running tally of the refcount.  If this ever deviates from
142  * what we expect (which is the refcountbt's refcount minus the
143  * number of extents that totally covered the refcountbt extent),
144  * we have a refcountbt error.
145  */
146 STATIC void
147 xchk_refcountbt_process_rmap_fragments(
148 	struct xchk_refcnt_check	*refchk)
149 {
150 	struct list_head		worklist;
151 	struct xchk_refcnt_frag		*frag;
152 	struct xchk_refcnt_frag		*n;
153 	xfs_agblock_t			bno;
154 	xfs_agblock_t			rbno;
155 	xfs_agblock_t			next_rbno;
156 	xfs_nlink_t			nr;
157 	xfs_nlink_t			target_nr;
158 
159 	target_nr = refchk->refcount - refchk->seen;
160 	if (target_nr == 0)
161 		return;
162 
163 	/*
164 	 * There are (refchk->rc.rc_refcount - refchk->nr refcount)
165 	 * references we haven't found yet.  Pull that many off the
166 	 * fragment list and figure out where the smallest rmap ends
167 	 * (and therefore the next rmap should start).  All the rmaps
168 	 * we pull off should start at or before the beginning of the
169 	 * refcount record's range.
170 	 */
171 	INIT_LIST_HEAD(&worklist);
172 	rbno = NULLAGBLOCK;
173 	nr = 1;
174 
175 	/* Make sure the fragments actually /are/ in agbno order. */
176 	bno = 0;
177 	list_for_each_entry(frag, &refchk->fragments, list) {
178 		if (frag->rm.rm_startblock < bno)
179 			goto done;
180 		bno = frag->rm.rm_startblock;
181 	}
182 
183 	/*
184 	 * Find all the rmaps that start at or before the refc extent,
185 	 * and put them on the worklist.
186 	 */
187 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
188 		if (frag->rm.rm_startblock > refchk->bno)
189 			goto done;
190 		bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
191 		if (bno < rbno)
192 			rbno = bno;
193 		list_move_tail(&frag->list, &worklist);
194 		if (nr == target_nr)
195 			break;
196 		nr++;
197 	}
198 
199 	/*
200 	 * We should have found exactly $target_nr rmap fragments starting
201 	 * at or before the refcount extent.
202 	 */
203 	if (nr != target_nr)
204 		goto done;
205 
206 	while (!list_empty(&refchk->fragments)) {
207 		/* Discard any fragments ending at rbno from the worklist. */
208 		nr = 0;
209 		next_rbno = NULLAGBLOCK;
210 		list_for_each_entry_safe(frag, n, &worklist, list) {
211 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
212 			if (bno != rbno) {
213 				if (bno < next_rbno)
214 					next_rbno = bno;
215 				continue;
216 			}
217 			list_del(&frag->list);
218 			kmem_free(frag);
219 			nr++;
220 		}
221 
222 		/* Try to add nr rmaps starting at rbno to the worklist. */
223 		list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
224 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
225 			if (frag->rm.rm_startblock != rbno)
226 				goto done;
227 			list_move_tail(&frag->list, &worklist);
228 			if (next_rbno > bno)
229 				next_rbno = bno;
230 			nr--;
231 			if (nr == 0)
232 				break;
233 		}
234 
235 		/*
236 		 * If we get here and nr > 0, this means that we added fewer
237 		 * items to the worklist than we discarded because the fragment
238 		 * list ran out of items.  Therefore, we cannot maintain the
239 		 * required refcount.  Something is wrong, so we're done.
240 		 */
241 		if (nr)
242 			goto done;
243 
244 		rbno = next_rbno;
245 	}
246 
247 	/*
248 	 * Make sure the last extent we processed ends at or beyond
249 	 * the end of the refcount extent.
250 	 */
251 	if (rbno < refchk->bno + refchk->len)
252 		goto done;
253 
254 	/* Actually record us having seen the remaining refcount. */
255 	refchk->seen = refchk->refcount;
256 done:
257 	/* Delete fragments and work list. */
258 	list_for_each_entry_safe(frag, n, &worklist, list) {
259 		list_del(&frag->list);
260 		kmem_free(frag);
261 	}
262 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
263 		list_del(&frag->list);
264 		kmem_free(frag);
265 	}
266 }
267 
268 /* Use the rmap entries covering this extent to verify the refcount. */
269 STATIC void
270 xchk_refcountbt_xref_rmap(
271 	struct xfs_scrub		*sc,
272 	xfs_agblock_t			bno,
273 	xfs_extlen_t			len,
274 	xfs_nlink_t			refcount)
275 {
276 	struct xchk_refcnt_check	refchk = {
277 		.sc = sc,
278 		.bno = bno,
279 		.len = len,
280 		.refcount = refcount,
281 		.seen = 0,
282 	};
283 	struct xfs_rmap_irec		low;
284 	struct xfs_rmap_irec		high;
285 	struct xchk_refcnt_frag		*frag;
286 	struct xchk_refcnt_frag		*n;
287 	int				error;
288 
289 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
290 		return;
291 
292 	/* Cross-reference with the rmapbt to confirm the refcount. */
293 	memset(&low, 0, sizeof(low));
294 	low.rm_startblock = bno;
295 	memset(&high, 0xFF, sizeof(high));
296 	high.rm_startblock = bno + len - 1;
297 
298 	INIT_LIST_HEAD(&refchk.fragments);
299 	error = xfs_rmap_query_range(sc->sa.rmap_cur, &low, &high,
300 			&xchk_refcountbt_rmap_check, &refchk);
301 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
302 		goto out_free;
303 
304 	xchk_refcountbt_process_rmap_fragments(&refchk);
305 	if (refcount != refchk.seen)
306 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
307 
308 out_free:
309 	list_for_each_entry_safe(frag, n, &refchk.fragments, list) {
310 		list_del(&frag->list);
311 		kmem_free(frag);
312 	}
313 }
314 
315 /* Cross-reference with the other btrees. */
316 STATIC void
317 xchk_refcountbt_xref(
318 	struct xfs_scrub	*sc,
319 	xfs_agblock_t		agbno,
320 	xfs_extlen_t		len,
321 	xfs_nlink_t		refcount)
322 {
323 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
324 		return;
325 
326 	xchk_xref_is_used_space(sc, agbno, len);
327 	xchk_xref_is_not_inode_chunk(sc, agbno, len);
328 	xchk_refcountbt_xref_rmap(sc, agbno, len, refcount);
329 }
330 
331 /* Scrub a refcountbt record. */
332 STATIC int
333 xchk_refcountbt_rec(
334 	struct xchk_btree	*bs,
335 	union xfs_btree_rec	*rec)
336 {
337 	struct xfs_mount	*mp = bs->cur->bc_mp;
338 	xfs_agblock_t		*cow_blocks = bs->private;
339 	xfs_agnumber_t		agno = bs->cur->bc_private.a.agno;
340 	xfs_agblock_t		bno;
341 	xfs_extlen_t		len;
342 	xfs_nlink_t		refcount;
343 	bool			has_cowflag;
344 
345 	bno = be32_to_cpu(rec->refc.rc_startblock);
346 	len = be32_to_cpu(rec->refc.rc_blockcount);
347 	refcount = be32_to_cpu(rec->refc.rc_refcount);
348 
349 	/* Only CoW records can have refcount == 1. */
350 	has_cowflag = (bno & XFS_REFC_COW_START);
351 	if ((refcount == 1 && !has_cowflag) || (refcount != 1 && has_cowflag))
352 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
353 	if (has_cowflag)
354 		(*cow_blocks) += len;
355 
356 	/* Check the extent. */
357 	bno &= ~XFS_REFC_COW_START;
358 	if (bno + len <= bno ||
359 	    !xfs_verify_agbno(mp, agno, bno) ||
360 	    !xfs_verify_agbno(mp, agno, bno + len - 1))
361 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
362 
363 	if (refcount == 0)
364 		xchk_btree_set_corrupt(bs->sc, bs->cur, 0);
365 
366 	xchk_refcountbt_xref(bs->sc, bno, len, refcount);
367 
368 	return 0;
369 }
370 
371 /* Make sure we have as many refc blocks as the rmap says. */
372 STATIC void
373 xchk_refcount_xref_rmap(
374 	struct xfs_scrub	*sc,
375 	xfs_filblks_t		cow_blocks)
376 {
377 	xfs_extlen_t		refcbt_blocks = 0;
378 	xfs_filblks_t		blocks;
379 	int			error;
380 
381 	if (!sc->sa.rmap_cur || xchk_skip_xref(sc->sm))
382 		return;
383 
384 	/* Check that we saw as many refcbt blocks as the rmap knows about. */
385 	error = xfs_btree_count_blocks(sc->sa.refc_cur, &refcbt_blocks);
386 	if (!xchk_btree_process_error(sc, sc->sa.refc_cur, 0, &error))
387 		return;
388 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
389 			&XFS_RMAP_OINFO_REFC, &blocks);
390 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
391 		return;
392 	if (blocks != refcbt_blocks)
393 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
394 
395 	/* Check that we saw as many cow blocks as the rmap knows about. */
396 	error = xchk_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur,
397 			&XFS_RMAP_OINFO_COW, &blocks);
398 	if (!xchk_should_check_xref(sc, &error, &sc->sa.rmap_cur))
399 		return;
400 	if (blocks != cow_blocks)
401 		xchk_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
402 }
403 
404 /* Scrub the refcount btree for some AG. */
405 int
406 xchk_refcountbt(
407 	struct xfs_scrub	*sc)
408 {
409 	xfs_agblock_t		cow_blocks = 0;
410 	int			error;
411 
412 	error = xchk_btree(sc, sc->sa.refc_cur, xchk_refcountbt_rec,
413 			&XFS_RMAP_OINFO_REFC, &cow_blocks);
414 	if (error)
415 		return error;
416 
417 	xchk_refcount_xref_rmap(sc, cow_blocks);
418 
419 	return 0;
420 }
421 
422 /* xref check that a cow staging extent is marked in the refcountbt. */
423 void
424 xchk_xref_is_cow_staging(
425 	struct xfs_scrub		*sc,
426 	xfs_agblock_t			agbno,
427 	xfs_extlen_t			len)
428 {
429 	struct xfs_refcount_irec	rc;
430 	bool				has_cowflag;
431 	int				has_refcount;
432 	int				error;
433 
434 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
435 		return;
436 
437 	/* Find the CoW staging extent. */
438 	error = xfs_refcount_lookup_le(sc->sa.refc_cur,
439 			agbno + XFS_REFC_COW_START, &has_refcount);
440 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
441 		return;
442 	if (!has_refcount) {
443 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
444 		return;
445 	}
446 
447 	error = xfs_refcount_get_rec(sc->sa.refc_cur, &rc, &has_refcount);
448 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
449 		return;
450 	if (!has_refcount) {
451 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
452 		return;
453 	}
454 
455 	/* CoW flag must be set, refcount must be 1. */
456 	has_cowflag = (rc.rc_startblock & XFS_REFC_COW_START);
457 	if (!has_cowflag || rc.rc_refcount != 1)
458 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
459 
460 	/* Must be at least as long as what was passed in */
461 	if (rc.rc_blockcount < len)
462 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
463 }
464 
465 /*
466  * xref check that the extent is not shared.  Only file data blocks
467  * can have multiple owners.
468  */
469 void
470 xchk_xref_is_not_shared(
471 	struct xfs_scrub	*sc,
472 	xfs_agblock_t		agbno,
473 	xfs_extlen_t		len)
474 {
475 	bool			shared;
476 	int			error;
477 
478 	if (!sc->sa.refc_cur || xchk_skip_xref(sc->sm))
479 		return;
480 
481 	error = xfs_refcount_has_record(sc->sa.refc_cur, agbno, len, &shared);
482 	if (!xchk_should_check_xref(sc, &error, &sc->sa.refc_cur))
483 		return;
484 	if (shared)
485 		xchk_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
486 }
487