xref: /openbmc/linux/fs/xfs/scrub/refcount.c (revision 82e6fdd6)
1 /*
2  * Copyright (C) 2017 Oracle.  All Rights Reserved.
3  *
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version 2
9  * of the License, or (at your option) any later version.
10  *
11  * This program is distributed in the hope that it would be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write the Free Software Foundation,
18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
19  */
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_btree.h"
28 #include "xfs_bit.h"
29 #include "xfs_log_format.h"
30 #include "xfs_trans.h"
31 #include "xfs_sb.h"
32 #include "xfs_alloc.h"
33 #include "xfs_rmap.h"
34 #include "xfs_refcount.h"
35 #include "scrub/xfs_scrub.h"
36 #include "scrub/scrub.h"
37 #include "scrub/common.h"
38 #include "scrub/btree.h"
39 #include "scrub/trace.h"
40 
41 /*
42  * Set us up to scrub reference count btrees.
43  */
44 int
45 xfs_scrub_setup_ag_refcountbt(
46 	struct xfs_scrub_context	*sc,
47 	struct xfs_inode		*ip)
48 {
49 	return xfs_scrub_setup_ag_btree(sc, ip, false);
50 }
51 
52 /* Reference count btree scrubber. */
53 
54 /*
55  * Confirming Reference Counts via Reverse Mappings
56  *
57  * We want to count the reverse mappings overlapping a refcount record
58  * (bno, len, refcount), allowing for the possibility that some of the
59  * overlap may come from smaller adjoining reverse mappings, while some
60  * comes from single extents which overlap the range entirely.  The
61  * outer loop is as follows:
62  *
63  * 1. For all reverse mappings overlapping the refcount extent,
64  *    a. If a given rmap completely overlaps, mark it as seen.
65  *    b. Otherwise, record the fragment (in agbno order) for later
66  *       processing.
67  *
68  * Once we've seen all the rmaps, we know that for all blocks in the
69  * refcount record we want to find $refcount owners and we've already
70  * visited $seen extents that overlap all the blocks.  Therefore, we
71  * need to find ($refcount - $seen) owners for every block in the
72  * extent; call that quantity $target_nr.  Proceed as follows:
73  *
74  * 2. Pull the first $target_nr fragments from the list; all of them
75  *    should start at or before the start of the extent.
76  *    Call this subset of fragments the working set.
77  * 3. Until there are no more unprocessed fragments,
78  *    a. Find the shortest fragments in the set and remove them.
79  *    b. Note the block number of the end of these fragments.
80  *    c. Pull the same number of fragments from the list.  All of these
81  *       fragments should start at the block number recorded in the
82  *       previous step.
83  *    d. Put those fragments in the set.
84  * 4. Check that there are $target_nr fragments remaining in the list,
85  *    and that they all end at or beyond the end of the refcount extent.
86  *
87  * If the refcount is correct, all the check conditions in the algorithm
88  * should always hold true.  If not, the refcount is incorrect.
89  */
90 struct xfs_scrub_refcnt_frag {
91 	struct list_head		list;
92 	struct xfs_rmap_irec		rm;
93 };
94 
95 struct xfs_scrub_refcnt_check {
96 	struct xfs_scrub_context	*sc;
97 	struct list_head		fragments;
98 
99 	/* refcount extent we're examining */
100 	xfs_agblock_t			bno;
101 	xfs_extlen_t			len;
102 	xfs_nlink_t			refcount;
103 
104 	/* number of owners seen */
105 	xfs_nlink_t			seen;
106 };
107 
108 /*
109  * Decide if the given rmap is large enough that we can redeem it
110  * towards refcount verification now, or if it's a fragment, in
111  * which case we'll hang onto it in the hopes that we'll later
112  * discover that we've collected exactly the correct number of
113  * fragments as the refcountbt says we should have.
114  */
115 STATIC int
116 xfs_scrub_refcountbt_rmap_check(
117 	struct xfs_btree_cur		*cur,
118 	struct xfs_rmap_irec		*rec,
119 	void				*priv)
120 {
121 	struct xfs_scrub_refcnt_check	*refchk = priv;
122 	struct xfs_scrub_refcnt_frag	*frag;
123 	xfs_agblock_t			rm_last;
124 	xfs_agblock_t			rc_last;
125 	int				error = 0;
126 
127 	if (xfs_scrub_should_terminate(refchk->sc, &error))
128 		return error;
129 
130 	rm_last = rec->rm_startblock + rec->rm_blockcount - 1;
131 	rc_last = refchk->bno + refchk->len - 1;
132 
133 	/* Confirm that a single-owner refc extent is a CoW stage. */
134 	if (refchk->refcount == 1 && rec->rm_owner != XFS_RMAP_OWN_COW) {
135 		xfs_scrub_btree_xref_set_corrupt(refchk->sc, cur, 0);
136 		return 0;
137 	}
138 
139 	if (rec->rm_startblock <= refchk->bno && rm_last >= rc_last) {
140 		/*
141 		 * The rmap overlaps the refcount record, so we can confirm
142 		 * one refcount owner seen.
143 		 */
144 		refchk->seen++;
145 	} else {
146 		/*
147 		 * This rmap covers only part of the refcount record, so
148 		 * save the fragment for later processing.  If the rmapbt
149 		 * is healthy each rmap_irec we see will be in agbno order
150 		 * so we don't need insertion sort here.
151 		 */
152 		frag = kmem_alloc(sizeof(struct xfs_scrub_refcnt_frag),
153 				KM_MAYFAIL | KM_NOFS);
154 		if (!frag)
155 			return -ENOMEM;
156 		memcpy(&frag->rm, rec, sizeof(frag->rm));
157 		list_add_tail(&frag->list, &refchk->fragments);
158 	}
159 
160 	return 0;
161 }
162 
163 /*
164  * Given a bunch of rmap fragments, iterate through them, keeping
165  * a running tally of the refcount.  If this ever deviates from
166  * what we expect (which is the refcountbt's refcount minus the
167  * number of extents that totally covered the refcountbt extent),
168  * we have a refcountbt error.
169  */
170 STATIC void
171 xfs_scrub_refcountbt_process_rmap_fragments(
172 	struct xfs_scrub_refcnt_check	*refchk)
173 {
174 	struct list_head		worklist;
175 	struct xfs_scrub_refcnt_frag	*frag;
176 	struct xfs_scrub_refcnt_frag	*n;
177 	xfs_agblock_t			bno;
178 	xfs_agblock_t			rbno;
179 	xfs_agblock_t			next_rbno;
180 	xfs_nlink_t			nr;
181 	xfs_nlink_t			target_nr;
182 
183 	target_nr = refchk->refcount - refchk->seen;
184 	if (target_nr == 0)
185 		return;
186 
187 	/*
188 	 * There are (refchk->rc.rc_refcount - refchk->nr refcount)
189 	 * references we haven't found yet.  Pull that many off the
190 	 * fragment list and figure out where the smallest rmap ends
191 	 * (and therefore the next rmap should start).  All the rmaps
192 	 * we pull off should start at or before the beginning of the
193 	 * refcount record's range.
194 	 */
195 	INIT_LIST_HEAD(&worklist);
196 	rbno = NULLAGBLOCK;
197 	nr = 1;
198 
199 	/* Make sure the fragments actually /are/ in agbno order. */
200 	bno = 0;
201 	list_for_each_entry(frag, &refchk->fragments, list) {
202 		if (frag->rm.rm_startblock < bno)
203 			goto done;
204 		bno = frag->rm.rm_startblock;
205 	}
206 
207 	/*
208 	 * Find all the rmaps that start at or before the refc extent,
209 	 * and put them on the worklist.
210 	 */
211 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
212 		if (frag->rm.rm_startblock > refchk->bno)
213 			goto done;
214 		bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
215 		if (bno < rbno)
216 			rbno = bno;
217 		list_move_tail(&frag->list, &worklist);
218 		if (nr == target_nr)
219 			break;
220 		nr++;
221 	}
222 
223 	/*
224 	 * We should have found exactly $target_nr rmap fragments starting
225 	 * at or before the refcount extent.
226 	 */
227 	if (nr != target_nr)
228 		goto done;
229 
230 	while (!list_empty(&refchk->fragments)) {
231 		/* Discard any fragments ending at rbno from the worklist. */
232 		nr = 0;
233 		next_rbno = NULLAGBLOCK;
234 		list_for_each_entry_safe(frag, n, &worklist, list) {
235 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
236 			if (bno != rbno) {
237 				if (bno < next_rbno)
238 					next_rbno = bno;
239 				continue;
240 			}
241 			list_del(&frag->list);
242 			kmem_free(frag);
243 			nr++;
244 		}
245 
246 		/* Try to add nr rmaps starting at rbno to the worklist. */
247 		list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
248 			bno = frag->rm.rm_startblock + frag->rm.rm_blockcount;
249 			if (frag->rm.rm_startblock != rbno)
250 				goto done;
251 			list_move_tail(&frag->list, &worklist);
252 			if (next_rbno > bno)
253 				next_rbno = bno;
254 			nr--;
255 			if (nr == 0)
256 				break;
257 		}
258 
259 		/*
260 		 * If we get here and nr > 0, this means that we added fewer
261 		 * items to the worklist than we discarded because the fragment
262 		 * list ran out of items.  Therefore, we cannot maintain the
263 		 * required refcount.  Something is wrong, so we're done.
264 		 */
265 		if (nr)
266 			goto done;
267 
268 		rbno = next_rbno;
269 	}
270 
271 	/*
272 	 * Make sure the last extent we processed ends at or beyond
273 	 * the end of the refcount extent.
274 	 */
275 	if (rbno < refchk->bno + refchk->len)
276 		goto done;
277 
278 	/* Actually record us having seen the remaining refcount. */
279 	refchk->seen = refchk->refcount;
280 done:
281 	/* Delete fragments and work list. */
282 	list_for_each_entry_safe(frag, n, &worklist, list) {
283 		list_del(&frag->list);
284 		kmem_free(frag);
285 	}
286 	list_for_each_entry_safe(frag, n, &refchk->fragments, list) {
287 		list_del(&frag->list);
288 		kmem_free(frag);
289 	}
290 }
291 
292 /* Use the rmap entries covering this extent to verify the refcount. */
293 STATIC void
294 xfs_scrub_refcountbt_xref_rmap(
295 	struct xfs_scrub_context	*sc,
296 	xfs_agblock_t			bno,
297 	xfs_extlen_t			len,
298 	xfs_nlink_t			refcount)
299 {
300 	struct xfs_scrub_refcnt_check	refchk = {
301 		.sc = sc,
302 		.bno = bno,
303 		.len = len,
304 		.refcount = refcount,
305 		.seen = 0,
306 	};
307 	struct xfs_rmap_irec		low;
308 	struct xfs_rmap_irec		high;
309 	struct xfs_scrub_refcnt_frag	*frag;
310 	struct xfs_scrub_refcnt_frag	*n;
311 	int				error;
312 
313 	if (!sc->sa.rmap_cur)
314 		return;
315 
316 	/* Cross-reference with the rmapbt to confirm the refcount. */
317 	memset(&low, 0, sizeof(low));
318 	low.rm_startblock = bno;
319 	memset(&high, 0xFF, sizeof(high));
320 	high.rm_startblock = bno + len - 1;
321 
322 	INIT_LIST_HEAD(&refchk.fragments);
323 	error = xfs_rmap_query_range(sc->sa.rmap_cur, &low, &high,
324 			&xfs_scrub_refcountbt_rmap_check, &refchk);
325 	if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.rmap_cur))
326 		goto out_free;
327 
328 	xfs_scrub_refcountbt_process_rmap_fragments(&refchk);
329 	if (refcount != refchk.seen)
330 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
331 
332 out_free:
333 	list_for_each_entry_safe(frag, n, &refchk.fragments, list) {
334 		list_del(&frag->list);
335 		kmem_free(frag);
336 	}
337 }
338 
339 /* Cross-reference with the other btrees. */
340 STATIC void
341 xfs_scrub_refcountbt_xref(
342 	struct xfs_scrub_context	*sc,
343 	xfs_agblock_t			agbno,
344 	xfs_extlen_t			len,
345 	xfs_nlink_t			refcount)
346 {
347 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
348 		return;
349 
350 	xfs_scrub_xref_is_used_space(sc, agbno, len);
351 	xfs_scrub_xref_is_not_inode_chunk(sc, agbno, len);
352 	xfs_scrub_refcountbt_xref_rmap(sc, agbno, len, refcount);
353 }
354 
355 /* Scrub a refcountbt record. */
356 STATIC int
357 xfs_scrub_refcountbt_rec(
358 	struct xfs_scrub_btree		*bs,
359 	union xfs_btree_rec		*rec)
360 {
361 	struct xfs_mount		*mp = bs->cur->bc_mp;
362 	xfs_agblock_t			*cow_blocks = bs->private;
363 	xfs_agnumber_t			agno = bs->cur->bc_private.a.agno;
364 	xfs_agblock_t			bno;
365 	xfs_extlen_t			len;
366 	xfs_nlink_t			refcount;
367 	bool				has_cowflag;
368 	int				error = 0;
369 
370 	bno = be32_to_cpu(rec->refc.rc_startblock);
371 	len = be32_to_cpu(rec->refc.rc_blockcount);
372 	refcount = be32_to_cpu(rec->refc.rc_refcount);
373 
374 	/* Only CoW records can have refcount == 1. */
375 	has_cowflag = (bno & XFS_REFC_COW_START);
376 	if ((refcount == 1 && !has_cowflag) || (refcount != 1 && has_cowflag))
377 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
378 	if (has_cowflag)
379 		(*cow_blocks) += len;
380 
381 	/* Check the extent. */
382 	bno &= ~XFS_REFC_COW_START;
383 	if (bno + len <= bno ||
384 	    !xfs_verify_agbno(mp, agno, bno) ||
385 	    !xfs_verify_agbno(mp, agno, bno + len - 1))
386 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
387 
388 	if (refcount == 0)
389 		xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
390 
391 	xfs_scrub_refcountbt_xref(bs->sc, bno, len, refcount);
392 
393 	return error;
394 }
395 
396 /* Make sure we have as many refc blocks as the rmap says. */
397 STATIC void
398 xfs_scrub_refcount_xref_rmap(
399 	struct xfs_scrub_context	*sc,
400 	struct xfs_owner_info		*oinfo,
401 	xfs_filblks_t			cow_blocks)
402 {
403 	xfs_extlen_t			refcbt_blocks = 0;
404 	xfs_filblks_t			blocks;
405 	int				error;
406 
407 	if (!sc->sa.rmap_cur)
408 		return;
409 
410 	/* Check that we saw as many refcbt blocks as the rmap knows about. */
411 	error = xfs_btree_count_blocks(sc->sa.refc_cur, &refcbt_blocks);
412 	if (!xfs_scrub_btree_process_error(sc, sc->sa.refc_cur, 0, &error))
413 		return;
414 	error = xfs_scrub_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur, oinfo,
415 			&blocks);
416 	if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.rmap_cur))
417 		return;
418 	if (blocks != refcbt_blocks)
419 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
420 
421 	/* Check that we saw as many cow blocks as the rmap knows about. */
422 	xfs_rmap_ag_owner(oinfo, XFS_RMAP_OWN_COW);
423 	error = xfs_scrub_count_rmap_ownedby_ag(sc, sc->sa.rmap_cur, oinfo,
424 			&blocks);
425 	if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.rmap_cur))
426 		return;
427 	if (blocks != cow_blocks)
428 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.rmap_cur, 0);
429 }
430 
431 /* Scrub the refcount btree for some AG. */
432 int
433 xfs_scrub_refcountbt(
434 	struct xfs_scrub_context	*sc)
435 {
436 	struct xfs_owner_info		oinfo;
437 	xfs_agblock_t			cow_blocks = 0;
438 	int				error;
439 
440 	xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_REFC);
441 	error = xfs_scrub_btree(sc, sc->sa.refc_cur, xfs_scrub_refcountbt_rec,
442 			&oinfo, &cow_blocks);
443 	if (error)
444 		return error;
445 
446 	xfs_scrub_refcount_xref_rmap(sc, &oinfo, cow_blocks);
447 
448 	return 0;
449 }
450 
451 /* xref check that a cow staging extent is marked in the refcountbt. */
452 void
453 xfs_scrub_xref_is_cow_staging(
454 	struct xfs_scrub_context	*sc,
455 	xfs_agblock_t			agbno,
456 	xfs_extlen_t			len)
457 {
458 	struct xfs_refcount_irec	rc;
459 	bool				has_cowflag;
460 	int				has_refcount;
461 	int				error;
462 
463 	if (!sc->sa.refc_cur)
464 		return;
465 
466 	/* Find the CoW staging extent. */
467 	error = xfs_refcount_lookup_le(sc->sa.refc_cur,
468 			agbno + XFS_REFC_COW_START, &has_refcount);
469 	if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.refc_cur))
470 		return;
471 	if (!has_refcount) {
472 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
473 		return;
474 	}
475 
476 	error = xfs_refcount_get_rec(sc->sa.refc_cur, &rc, &has_refcount);
477 	if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.refc_cur))
478 		return;
479 	if (!has_refcount) {
480 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
481 		return;
482 	}
483 
484 	/* CoW flag must be set, refcount must be 1. */
485 	has_cowflag = (rc.rc_startblock & XFS_REFC_COW_START);
486 	if (!has_cowflag || rc.rc_refcount != 1)
487 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
488 
489 	/* Must be at least as long as what was passed in */
490 	if (rc.rc_blockcount < len)
491 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
492 }
493 
494 /*
495  * xref check that the extent is not shared.  Only file data blocks
496  * can have multiple owners.
497  */
498 void
499 xfs_scrub_xref_is_not_shared(
500 	struct xfs_scrub_context	*sc,
501 	xfs_agblock_t			agbno,
502 	xfs_extlen_t			len)
503 {
504 	bool				shared;
505 	int				error;
506 
507 	if (!sc->sa.refc_cur)
508 		return;
509 
510 	error = xfs_refcount_has_record(sc->sa.refc_cur, agbno, len, &shared);
511 	if (!xfs_scrub_should_check_xref(sc, &error, &sc->sa.refc_cur))
512 		return;
513 	if (shared)
514 		xfs_scrub_btree_xref_set_corrupt(sc, sc->sa.refc_cur, 0);
515 }
516