1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2022-2023 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_defer.h"
13#include "xfs_btree.h"
14#include "xfs_log_format.h"
15#include "xfs_trans.h"
16#include "xfs_inode.h"
17#include "xfs_inode_fork.h"
18#include "xfs_alloc.h"
19#include "xfs_bmap.h"
20#include "xfs_rmap.h"
21#include "xfs_refcount.h"
22#include "xfs_quota.h"
23#include "xfs_ialloc.h"
24#include "xfs_ag.h"
25#include "xfs_error.h"
26#include "xfs_errortag.h"
27#include "xfs_icache.h"
28#include "xfs_refcount_btree.h"
29#include "scrub/xfs_scrub.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/off_bitmap.h"
36#include "scrub/fsb_bitmap.h"
37#include "scrub/reap.h"
38
39/*
40 * CoW Fork Mapping Repair
41 * =======================
42 *
43 * Although CoW staging extents are owned by incore CoW inode forks, on disk
44 * they are owned by the refcount btree. The ondisk metadata does not record
45 * any ownership information, which limits what we can do to repair the
46 * mappings in the CoW fork. At most, we can replace ifork mappings that lack
47 * an entry in the refcount btree or are described by a reverse mapping record
48 * whose owner is not OWN_COW.
49 *
50 * Replacing extents is also tricky -- we can't touch written CoW fork extents
51 * since they are undergoing writeback, and delalloc extents do not require
52 * repair since they only exist incore. Hence the most we can do is find the
53 * bad parts of unwritten mappings, allocate a replacement set of blocks, and
54 * replace the incore mapping. We use the regular reaping process to unmap
55 * or free the discarded blocks, as appropriate.
56 */
57struct xrep_cow {
58 struct xfs_scrub *sc;
59
60 /* Bitmap of file offset ranges that need replacing. */
61 struct xoff_bitmap bad_fileoffs;
62
63 /* Bitmap of fsblocks that were removed from the CoW fork. */
64 struct xfsb_bitmap old_cowfork_fsblocks;
65
66 /* CoW fork mappings used to scan for bad CoW staging extents. */
67 struct xfs_bmbt_irec irec;
68
69 /* refcount btree block number of irec.br_startblock */
70 unsigned int irec_startbno;
71
72 /* refcount btree block number of the next refcount record we expect */
73 unsigned int next_bno;
74};
75
76/* CoW staging extent. */
77struct xrep_cow_extent {
78 xfs_fsblock_t fsbno;
79 xfs_extlen_t len;
80};
81
82/*
83 * Mark the part of the file range that corresponds to the given physical
84 * space. Caller must ensure that the physical range is within xc->irec.
85 */
86STATIC int
87xrep_cow_mark_file_range(
88 struct xrep_cow *xc,
89 xfs_fsblock_t startblock,
90 xfs_filblks_t blockcount)
91{
92 xfs_fileoff_t startoff;
93
94 startoff = xc->irec.br_startoff +
95 (startblock - xc->irec.br_startblock);
96
97 trace_xrep_cow_mark_file_range(xc->sc->ip, startblock, startoff,
98 blockcount);
99
100 return xoff_bitmap_set(&xc->bad_fileoffs, startoff, blockcount);
101}
102
103/*
104 * Trim @src to fit within the CoW fork mapping being examined, and put the
105 * result in @dst.
106 */
107static inline void
108xrep_cow_trim_refcount(
109 struct xrep_cow *xc,
110 struct xfs_refcount_irec *dst,
111 const struct xfs_refcount_irec *src)
112{
113 unsigned int adj;
114
115 memcpy(dst, src, sizeof(*dst));
116
117 if (dst->rc_startblock < xc->irec_startbno) {
118 adj = xc->irec_startbno - dst->rc_startblock;
119 dst->rc_blockcount -= adj;
120 dst->rc_startblock += adj;
121 }
122
123 if (dst->rc_startblock + dst->rc_blockcount >
124 xc->irec_startbno + xc->irec.br_blockcount) {
125 adj = (dst->rc_startblock + dst->rc_blockcount) -
126 (xc->irec_startbno + xc->irec.br_blockcount);
127 dst->rc_blockcount -= adj;
128 }
129}
130
131/* Mark any shared CoW staging extents. */
132STATIC int
133xrep_cow_mark_shared_staging(
134 struct xfs_btree_cur *cur,
135 const struct xfs_refcount_irec *rec,
136 void *priv)
137{
138 struct xrep_cow *xc = priv;
139 struct xfs_refcount_irec rrec;
140 xfs_fsblock_t fsbno;
141
142 if (!xfs_refcount_check_domain(rec) ||
143 rec->rc_domain != XFS_REFC_DOMAIN_SHARED)
144 return -EFSCORRUPTED;
145
146 xrep_cow_trim_refcount(xc, dst: &rrec, src: rec);
147
148 fsbno = XFS_AGB_TO_FSB(xc->sc->mp, cur->bc_ag.pag->pag_agno,
149 rrec.rc_startblock);
150 return xrep_cow_mark_file_range(xc, fsbno, rrec.rc_blockcount);
151}
152
153/*
154 * Mark any portion of the CoW fork file offset range where there is not a CoW
155 * staging extent record in the refcountbt, and keep a record of where we did
156 * find correct refcountbt records. Staging records are always cleaned out at
157 * mount time, so any two inodes trying to map the same staging area would have
158 * already taken the fs down due to refcount btree verifier errors. Hence this
159 * inode should be the sole creator of the staging extent records ondisk.
160 */
161STATIC int
162xrep_cow_mark_missing_staging(
163 struct xfs_btree_cur *cur,
164 const struct xfs_refcount_irec *rec,
165 void *priv)
166{
167 struct xrep_cow *xc = priv;
168 struct xfs_refcount_irec rrec;
169 int error;
170
171 if (!xfs_refcount_check_domain(rec) ||
172 rec->rc_domain != XFS_REFC_DOMAIN_COW)
173 return -EFSCORRUPTED;
174
175 xrep_cow_trim_refcount(xc, dst: &rrec, src: rec);
176
177 if (xc->next_bno >= rrec.rc_startblock)
178 goto next;
179
180 error = xrep_cow_mark_file_range(xc,
181 XFS_AGB_TO_FSB(xc->sc->mp, cur->bc_ag.pag->pag_agno,
182 xc->next_bno),
183 rrec.rc_startblock - xc->next_bno);
184 if (error)
185 return error;
186
187next:
188 xc->next_bno = rrec.rc_startblock + rrec.rc_blockcount;
189 return 0;
190}
191
192/*
193 * Mark any area that does not correspond to a CoW staging rmap. These are
194 * cross-linked areas that must be avoided.
195 */
196STATIC int
197xrep_cow_mark_missing_staging_rmap(
198 struct xfs_btree_cur *cur,
199 const struct xfs_rmap_irec *rec,
200 void *priv)
201{
202 struct xrep_cow *xc = priv;
203 xfs_fsblock_t fsbno;
204 xfs_agblock_t rec_bno;
205 xfs_extlen_t rec_len;
206 unsigned int adj;
207
208 if (rec->rm_owner == XFS_RMAP_OWN_COW)
209 return 0;
210
211 rec_bno = rec->rm_startblock;
212 rec_len = rec->rm_blockcount;
213 if (rec_bno < xc->irec_startbno) {
214 adj = xc->irec_startbno - rec_bno;
215 rec_len -= adj;
216 rec_bno += adj;
217 }
218
219 if (rec_bno + rec_len > xc->irec_startbno + xc->irec.br_blockcount) {
220 adj = (rec_bno + rec_len) -
221 (xc->irec_startbno + xc->irec.br_blockcount);
222 rec_len -= adj;
223 }
224
225 fsbno = XFS_AGB_TO_FSB(xc->sc->mp, cur->bc_ag.pag->pag_agno, rec_bno);
226 return xrep_cow_mark_file_range(xc, fsbno, rec_len);
227}
228
229/*
230 * Find any part of the CoW fork mapping that isn't a single-owner CoW staging
231 * extent and mark the corresponding part of the file range in the bitmap.
232 */
233STATIC int
234xrep_cow_find_bad(
235 struct xrep_cow *xc)
236{
237 struct xfs_refcount_irec rc_low = { 0 };
238 struct xfs_refcount_irec rc_high = { 0 };
239 struct xfs_rmap_irec rm_low = { 0 };
240 struct xfs_rmap_irec rm_high = { 0 };
241 struct xfs_perag *pag;
242 struct xfs_scrub *sc = xc->sc;
243 xfs_agnumber_t agno;
244 int error;
245
246 agno = XFS_FSB_TO_AGNO(sc->mp, xc->irec.br_startblock);
247 xc->irec_startbno = XFS_FSB_TO_AGBNO(sc->mp, xc->irec.br_startblock);
248
249 pag = xfs_perag_get(sc->mp, agno);
250 if (!pag)
251 return -EFSCORRUPTED;
252
253 error = xrep_ag_init(sc, pag, &sc->sa);
254 if (error)
255 goto out_pag;
256
257 /* Mark any CoW fork extents that are shared. */
258 rc_low.rc_startblock = xc->irec_startbno;
259 rc_high.rc_startblock = xc->irec_startbno + xc->irec.br_blockcount - 1;
260 rc_low.rc_domain = rc_high.rc_domain = XFS_REFC_DOMAIN_SHARED;
261 error = xfs_refcount_query_range(sc->sa.refc_cur, &rc_low, &rc_high,
262 xrep_cow_mark_shared_staging, xc);
263 if (error)
264 goto out_sa;
265
266 /* Make sure there are CoW staging extents for the whole mapping. */
267 rc_low.rc_startblock = xc->irec_startbno;
268 rc_high.rc_startblock = xc->irec_startbno + xc->irec.br_blockcount - 1;
269 rc_low.rc_domain = rc_high.rc_domain = XFS_REFC_DOMAIN_COW;
270 xc->next_bno = xc->irec_startbno;
271 error = xfs_refcount_query_range(sc->sa.refc_cur, &rc_low, &rc_high,
272 xrep_cow_mark_missing_staging, xc);
273 if (error)
274 goto out_sa;
275
276 if (xc->next_bno < xc->irec_startbno + xc->irec.br_blockcount) {
277 error = xrep_cow_mark_file_range(xc,
278 XFS_AGB_TO_FSB(sc->mp, pag->pag_agno,
279 xc->next_bno),
280 xc->irec_startbno + xc->irec.br_blockcount -
281 xc->next_bno);
282 if (error)
283 goto out_sa;
284 }
285
286 /* Mark any area has an rmap that isn't a COW staging extent. */
287 rm_low.rm_startblock = xc->irec_startbno;
288 memset(&rm_high, 0xFF, sizeof(rm_high));
289 rm_high.rm_startblock = xc->irec_startbno + xc->irec.br_blockcount - 1;
290 error = xfs_rmap_query_range(sc->sa.rmap_cur, &rm_low, &rm_high,
291 xrep_cow_mark_missing_staging_rmap, xc);
292 if (error)
293 goto out_sa;
294
295 /*
296 * If userspace is forcing us to rebuild the CoW fork or someone turned
297 * on the debugging knob, replace everything in the CoW fork.
298 */
299 if ((sc->sm->sm_flags & XFS_SCRUB_IFLAG_FORCE_REBUILD) ||
300 XFS_TEST_ERROR(false, sc->mp, XFS_ERRTAG_FORCE_SCRUB_REPAIR)) {
301 error = xrep_cow_mark_file_range(xc, xc->irec.br_startblock,
302 xc->irec.br_blockcount);
303 if (error)
304 return error;
305 }
306
307out_sa:
308 xchk_ag_free(sc, &sc->sa);
309out_pag:
310 xfs_perag_put(pag);
311 return 0;
312}
313
314/*
315 * Allocate a replacement CoW staging extent of up to the given number of
316 * blocks, and fill out the mapping.
317 */
318STATIC int
319xrep_cow_alloc(
320 struct xfs_scrub *sc,
321 xfs_extlen_t maxlen,
322 struct xrep_cow_extent *repl)
323{
324 struct xfs_alloc_arg args = {
325 .tp = sc->tp,
326 .mp = sc->mp,
327 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
328 .minlen = 1,
329 .maxlen = maxlen,
330 .prod = 1,
331 .resv = XFS_AG_RESV_NONE,
332 .datatype = XFS_ALLOC_USERDATA,
333 };
334 int error;
335
336 error = xfs_trans_reserve_more(sc->tp, maxlen, 0);
337 if (error)
338 return error;
339
340 error = xfs_alloc_vextent_start_ag(&args,
341 XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino));
342 if (error)
343 return error;
344 if (args.fsbno == NULLFSBLOCK)
345 return -ENOSPC;
346
347 xfs_refcount_alloc_cow_extent(sc->tp, args.fsbno, args.len);
348
349 repl->fsbno = args.fsbno;
350 repl->len = args.len;
351 return 0;
352}
353
354/*
355 * Look up the current CoW fork mapping so that we only allocate enough to
356 * replace a single mapping. If we don't find a mapping that covers the start
357 * of the file range, or we find a delalloc or written extent, something is
358 * seriously wrong, since we didn't drop the ILOCK.
359 */
360static inline int
361xrep_cow_find_mapping(
362 struct xrep_cow *xc,
363 struct xfs_iext_cursor *icur,
364 xfs_fileoff_t startoff,
365 struct xfs_bmbt_irec *got)
366{
367 struct xfs_inode *ip = xc->sc->ip;
368 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
369
370 if (!xfs_iext_lookup_extent(ip, ifp, startoff, icur, got))
371 goto bad;
372
373 if (got->br_startoff > startoff)
374 goto bad;
375
376 if (got->br_blockcount == 0)
377 goto bad;
378
379 if (isnullstartblock(got->br_startblock))
380 goto bad;
381
382 if (xfs_bmap_is_written_extent(got))
383 goto bad;
384
385 return 0;
386bad:
387 ASSERT(0);
388 return -EFSCORRUPTED;
389}
390
391#define REPLACE_LEFT_SIDE (1U << 0)
392#define REPLACE_RIGHT_SIDE (1U << 1)
393
394/*
395 * Given a CoW fork mapping @got and a replacement mapping @repl, remap the
396 * beginning of @got with the space described by @rep.
397 */
398static inline void
399xrep_cow_replace_mapping(
400 struct xfs_inode *ip,
401 struct xfs_iext_cursor *icur,
402 const struct xfs_bmbt_irec *got,
403 const struct xrep_cow_extent *repl)
404{
405 struct xfs_bmbt_irec new = *got; /* struct copy */
406
407 ASSERT(repl->len > 0);
408 ASSERT(!isnullstartblock(got->br_startblock));
409
410 trace_xrep_cow_replace_mapping(ip, got, repl->fsbno, repl->len);
411
412 if (got->br_blockcount == repl->len) {
413 /*
414 * The new extent is a complete replacement for the existing
415 * extent. Update the COW fork record.
416 */
417 new.br_startblock = repl->fsbno;
418 xfs_iext_update_extent(ip, BMAP_COWFORK, icur, &new);
419 return;
420 }
421
422 /*
423 * The new extent can replace the beginning of the COW fork record.
424 * Move the left side of @got upwards, then insert the new record.
425 */
426 new.br_startoff += repl->len;
427 new.br_startblock += repl->len;
428 new.br_blockcount -= repl->len;
429 xfs_iext_update_extent(ip, BMAP_COWFORK, icur, &new);
430
431 new.br_startoff = got->br_startoff;
432 new.br_startblock = repl->fsbno;
433 new.br_blockcount = repl->len;
434 xfs_iext_insert(ip, icur, &new, BMAP_COWFORK);
435}
436
437/*
438 * Replace the unwritten CoW staging extent backing the given file range with a
439 * new space extent that isn't as problematic.
440 */
441STATIC int
442xrep_cow_replace_range(
443 struct xrep_cow *xc,
444 xfs_fileoff_t startoff,
445 xfs_extlen_t *blockcount)
446{
447 struct xfs_iext_cursor icur;
448 struct xrep_cow_extent repl;
449 struct xfs_bmbt_irec got;
450 struct xfs_scrub *sc = xc->sc;
451 xfs_fileoff_t nextoff;
452 xfs_extlen_t alloc_len;
453 int error;
454
455 /*
456 * Put the existing CoW fork mapping in @got. If @got ends before
457 * @rep, truncate @rep so we only replace one extent mapping at a time.
458 */
459 error = xrep_cow_find_mapping(xc, &icur, startoff, &got);
460 if (error)
461 return error;
462 nextoff = min(startoff + *blockcount,
463 got.br_startoff + got.br_blockcount);
464
465 /*
466 * Allocate a replacement extent. If we don't fill all the blocks,
467 * shorten the quantity that will be deleted in this step.
468 */
469 alloc_len = min_t(xfs_fileoff_t, XFS_MAX_BMBT_EXTLEN,
470 nextoff - startoff);
471 error = xrep_cow_alloc(sc, alloc_len, &repl);
472 if (error)
473 return error;
474
475 /*
476 * Replace the old mapping with the new one, and commit the metadata
477 * changes made so far.
478 */
479 xrep_cow_replace_mapping(ip: sc->ip, icur: &icur, got: &got, repl: &repl);
480
481 xfs_inode_set_cowblocks_tag(sc->ip);
482 error = xfs_defer_finish(&sc->tp);
483 if (error)
484 return error;
485
486 /* Note the old CoW staging extents; we'll reap them all later. */
487 error = xfsb_bitmap_set(&xc->old_cowfork_fsblocks, got.br_startblock,
488 repl.len);
489 if (error)
490 return error;
491
492 *blockcount = repl.len;
493 return 0;
494}
495
496/*
497 * Replace a bad part of an unwritten CoW staging extent with a fresh delalloc
498 * reservation.
499 */
500STATIC int
501xrep_cow_replace(
502 uint64_t startoff,
503 uint64_t blockcount,
504 void *priv)
505{
506 struct xrep_cow *xc = priv;
507 int error = 0;
508
509 while (blockcount > 0) {
510 xfs_extlen_t len = min_t(xfs_filblks_t, blockcount,
511 XFS_MAX_BMBT_EXTLEN);
512
513 error = xrep_cow_replace_range(xc, startoff, &len);
514 if (error)
515 break;
516
517 blockcount -= len;
518 startoff += len;
519 }
520
521 return error;
522}
523
524/*
525 * Repair an inode's CoW fork. The CoW fork is an in-core structure, so
526 * there's no btree to rebuid. Instead, we replace any mappings that are
527 * cross-linked or lack ondisk CoW fork records in the refcount btree.
528 */
529int
530xrep_bmap_cow(
531 struct xfs_scrub *sc)
532{
533 struct xrep_cow *xc;
534 struct xfs_iext_cursor icur;
535 struct xfs_ifork *ifp = xfs_ifork_ptr(sc->ip, XFS_COW_FORK);
536 int error;
537
538 if (!xfs_has_rmapbt(sc->mp) || !xfs_has_reflink(sc->mp))
539 return -EOPNOTSUPP;
540
541 if (!ifp)
542 return 0;
543
544 /* realtime files aren't supported yet */
545 if (XFS_IS_REALTIME_INODE(sc->ip))
546 return -EOPNOTSUPP;
547
548 /*
549 * If we're somehow not in extents format, then reinitialize it to
550 * an empty extent mapping fork and exit.
551 */
552 if (ifp->if_format != XFS_DINODE_FMT_EXTENTS) {
553 ifp->if_format = XFS_DINODE_FMT_EXTENTS;
554 ifp->if_nextents = 0;
555 return 0;
556 }
557
558 xc = kzalloc(sizeof(struct xrep_cow), XCHK_GFP_FLAGS);
559 if (!xc)
560 return -ENOMEM;
561
562 xfs_trans_ijoin(sc->tp, sc->ip, 0);
563
564 xc->sc = sc;
565 xoff_bitmap_init(&xc->bad_fileoffs);
566 xfsb_bitmap_init(&xc->old_cowfork_fsblocks);
567
568 for_each_xfs_iext(ifp, &icur, &xc->irec) {
569 if (xchk_should_terminate(sc, &error))
570 goto out_bitmap;
571
572 /*
573 * delalloc reservations only exist incore, so there is no
574 * ondisk metadata that we can examine. Hence we leave them
575 * alone.
576 */
577 if (isnullstartblock(xc->irec.br_startblock))
578 continue;
579
580 /*
581 * COW fork extents are only in the written state if writeback
582 * is actively writing to disk. We cannot restart the write
583 * at a different disk address since we've already issued the
584 * IO, so we leave these alone and hope for the best.
585 */
586 if (xfs_bmap_is_written_extent(&xc->irec))
587 continue;
588
589 error = xrep_cow_find_bad(xc);
590 if (error)
591 goto out_bitmap;
592 }
593
594 /* Replace any bad unwritten mappings with fresh reservations. */
595 error = xoff_bitmap_walk(&xc->bad_fileoffs, xrep_cow_replace, xc);
596 if (error)
597 goto out_bitmap;
598
599 /*
600 * Reap as many of the old CoW blocks as we can. They are owned ondisk
601 * by the refcount btree, not the inode, so it is correct to treat them
602 * like inode metadata.
603 */
604 error = xrep_reap_fsblocks(sc, &xc->old_cowfork_fsblocks,
605 &XFS_RMAP_OINFO_COW);
606 if (error)
607 goto out_bitmap;
608
609out_bitmap:
610 xfsb_bitmap_destroy(&xc->old_cowfork_fsblocks);
611 xoff_bitmap_destroy(&xc->bad_fileoffs);
612 kfree(xc);
613 return error;
614}
615

source code of linux/fs/xfs/scrub/cow_repair.c