Line data Source code
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_btree.h"
13 : #include "xfs_log_format.h"
14 : #include "xfs_trans.h"
15 : #include "xfs_inode.h"
16 : #include "xfs_icache.h"
17 : #include "xfs_alloc.h"
18 : #include "xfs_alloc_btree.h"
19 : #include "xfs_ialloc.h"
20 : #include "xfs_ialloc_btree.h"
21 : #include "xfs_refcount_btree.h"
22 : #include "xfs_rmap.h"
23 : #include "xfs_rmap_btree.h"
24 : #include "xfs_log.h"
25 : #include "xfs_trans_priv.h"
26 : #include "xfs_da_format.h"
27 : #include "xfs_da_btree.h"
28 : #include "xfs_attr.h"
29 : #include "xfs_reflink.h"
30 : #include "xfs_ag.h"
31 : #include "scrub/scrub.h"
32 : #include "scrub/common.h"
33 : #include "scrub/trace.h"
34 : #include "scrub/repair.h"
35 : #include "scrub/health.h"
36 :
37 : /* Common code for the metadata scrubbers. */
38 :
39 : /*
40 : * Handling operational errors.
41 : *
42 : * The *_process_error() family of functions are used to process error return
43 : * codes from functions called as part of a scrub operation.
44 : *
45 : * If there's no error, we return true to tell the caller that it's ok
46 : * to move on to the next check in its list.
47 : *
48 : * For non-verifier errors (e.g. ENOMEM) we return false to tell the
49 : * caller that something bad happened, and we preserve *error so that
50 : * the caller can return the *error up the stack to userspace.
51 : *
52 : * Verifier errors (EFSBADCRC/EFSCORRUPTED) are recorded by setting
53 : * OFLAG_CORRUPT in sm_flags and the *error is cleared. In other words,
54 : * we track verifier errors (and failed scrub checks) via OFLAG_CORRUPT,
55 : * not via return codes. We return false to tell the caller that
56 : * something bad happened. Since the error has been cleared, the caller
57 : * will (presumably) return that zero and scrubbing will move on to
58 : * whatever's next.
59 : *
60 : * ftrace can be used to record the precise metadata location and the
61 : * approximate code location of the failed operation.
62 : */
63 :
64 : /* Check for operational errors. */
65 : static bool
66 453932506 : __xchk_process_error(
67 : struct xfs_scrub *sc,
68 : xfs_agnumber_t agno,
69 : xfs_agblock_t bno,
70 : int *error,
71 : __u32 errflag,
72 : void *ret_ip)
73 : {
74 453932506 : switch (*error) {
75 : case 0:
76 : return true;
77 77654 : case -EDEADLOCK:
78 : case -ECHRNG:
79 : /* Used to restart an op with deadlock avoidance. */
80 155308 : trace_xchk_deadlock_retry(
81 77654 : sc->ip ? sc->ip : XFS_I(file_inode(sc->file)),
82 : sc->sm, *error);
83 77654 : break;
84 11 : case -EFSBADCRC:
85 : case -EFSCORRUPTED:
86 : /* Note the badness but don't abort. */
87 11 : sc->sm->sm_flags |= errflag;
88 11 : *error = 0;
89 30 : fallthrough;
90 30 : default:
91 30 : trace_xchk_op_error(sc, agno, bno, *error,
92 : ret_ip);
93 30 : break;
94 : }
95 : return false;
96 : }
97 :
98 : bool
99 9540052 : xchk_process_error(
100 : struct xfs_scrub *sc,
101 : xfs_agnumber_t agno,
102 : xfs_agblock_t bno,
103 : int *error)
104 : {
105 9540052 : return __xchk_process_error(sc, agno, bno, error,
106 : XFS_SCRUB_OFLAG_CORRUPT, __return_address);
107 : }
108 :
109 : bool
110 444458766 : xchk_xref_process_error(
111 : struct xfs_scrub *sc,
112 : xfs_agnumber_t agno,
113 : xfs_agblock_t bno,
114 : int *error)
115 : {
116 444458766 : return __xchk_process_error(sc, agno, bno, error,
117 : XFS_SCRUB_OFLAG_XFAIL, __return_address);
118 : }
119 :
120 : /* Check for operational errors for a file offset. */
121 : static bool
122 4482598989 : __xchk_fblock_process_error(
123 : struct xfs_scrub *sc,
124 : int whichfork,
125 : xfs_fileoff_t offset,
126 : int *error,
127 : __u32 errflag,
128 : void *ret_ip)
129 : {
130 4482598989 : switch (*error) {
131 : case 0:
132 : return true;
133 0 : case -EDEADLOCK:
134 : case -ECHRNG:
135 : /* Used to restart an op with deadlock avoidance. */
136 0 : trace_xchk_deadlock_retry(sc->ip, sc->sm, *error);
137 0 : break;
138 11 : case -EFSBADCRC:
139 : case -EFSCORRUPTED:
140 : /* Note the badness but don't abort. */
141 11 : sc->sm->sm_flags |= errflag;
142 11 : *error = 0;
143 11 : fallthrough;
144 11 : default:
145 11 : trace_xchk_file_op_error(sc, whichfork, offset, *error,
146 : ret_ip);
147 11 : break;
148 : }
149 : return false;
150 : }
151 :
152 : bool
153 1429527292 : xchk_fblock_process_error(
154 : struct xfs_scrub *sc,
155 : int whichfork,
156 : xfs_fileoff_t offset,
157 : int *error)
158 : {
159 1429876447 : return __xchk_fblock_process_error(sc, whichfork, offset, error,
160 : XFS_SCRUB_OFLAG_CORRUPT, __return_address);
161 : }
162 :
163 : bool
164 3058272528 : xchk_fblock_xref_process_error(
165 : struct xfs_scrub *sc,
166 : int whichfork,
167 : xfs_fileoff_t offset,
168 : int *error)
169 : {
170 3058272528 : return __xchk_fblock_process_error(sc, whichfork, offset, error,
171 : XFS_SCRUB_OFLAG_XFAIL, __return_address);
172 : }
173 :
174 : /*
175 : * Handling scrub corruption/optimization/warning checks.
176 : *
177 : * The *_set_{corrupt,preen,warning}() family of functions are used to
178 : * record the presence of metadata that is incorrect (corrupt), could be
179 : * optimized somehow (preen), or should be flagged for administrative
180 : * review but is not incorrect (warn).
181 : *
182 : * ftrace can be used to record the precise metadata location and
183 : * approximate code location of the failed check.
184 : */
185 :
186 : /* Record a block which could be optimized. */
187 : void
188 2705186 : xchk_block_set_preen(
189 : struct xfs_scrub *sc,
190 : struct xfs_buf *bp)
191 : {
192 2705186 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
193 2705186 : trace_xchk_block_preen(sc, xfs_buf_daddr(bp), __return_address);
194 2705314 : }
195 :
196 : /*
197 : * Record an inode which could be optimized. The trace data will
198 : * include the block given by bp if bp is given; otherwise it will use
199 : * the block location of the inode record itself.
200 : */
201 : void
202 3578564 : xchk_ino_set_preen(
203 : struct xfs_scrub *sc,
204 : xfs_ino_t ino)
205 : {
206 3578564 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
207 3578564 : trace_xchk_ino_preen(sc, ino, __return_address);
208 3578516 : }
209 :
210 : /* Record something being wrong with the filesystem primary superblock. */
211 : void
212 0 : xchk_set_corrupt(
213 : struct xfs_scrub *sc)
214 : {
215 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
216 0 : trace_xchk_fs_error(sc, 0, __return_address);
217 0 : }
218 :
219 : /* Record a corrupt block. */
220 : void
221 0 : xchk_block_set_corrupt(
222 : struct xfs_scrub *sc,
223 : struct xfs_buf *bp)
224 : {
225 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
226 0 : trace_xchk_block_error(sc, xfs_buf_daddr(bp), __return_address);
227 0 : }
228 :
229 : /* Record a corruption while cross-referencing. */
230 : void
231 0 : xchk_block_xref_set_corrupt(
232 : struct xfs_scrub *sc,
233 : struct xfs_buf *bp)
234 : {
235 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
236 0 : trace_xchk_block_error(sc, xfs_buf_daddr(bp), __return_address);
237 0 : }
238 :
239 : /*
240 : * Record a corrupt inode. The trace data will include the block given
241 : * by bp if bp is given; otherwise it will use the block location of the
242 : * inode record itself.
243 : */
244 : void
245 0 : xchk_ino_set_corrupt(
246 : struct xfs_scrub *sc,
247 : xfs_ino_t ino)
248 : {
249 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
250 0 : trace_xchk_ino_error(sc, ino, __return_address);
251 0 : }
252 :
253 : /* Record a corruption while cross-referencing with an inode. */
254 : void
255 0 : xchk_ino_xref_set_corrupt(
256 : struct xfs_scrub *sc,
257 : xfs_ino_t ino)
258 : {
259 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
260 0 : trace_xchk_ino_error(sc, ino, __return_address);
261 0 : }
262 :
263 : /* Record corruption in a block indexed by a file fork. */
264 : void
265 22 : xchk_fblock_set_corrupt(
266 : struct xfs_scrub *sc,
267 : int whichfork,
268 : xfs_fileoff_t offset)
269 : {
270 22 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
271 22 : trace_xchk_fblock_error(sc, whichfork, offset, __return_address);
272 22 : }
273 :
274 : /* Record a corruption while cross-referencing a fork block. */
275 : void
276 0 : xchk_fblock_xref_set_corrupt(
277 : struct xfs_scrub *sc,
278 : int whichfork,
279 : xfs_fileoff_t offset)
280 : {
281 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
282 0 : trace_xchk_fblock_error(sc, whichfork, offset, __return_address);
283 0 : }
284 :
285 : /*
286 : * Warn about inodes that need administrative review but is not
287 : * incorrect.
288 : */
289 : void
290 0 : xchk_ino_set_warning(
291 : struct xfs_scrub *sc,
292 : xfs_ino_t ino)
293 : {
294 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
295 0 : trace_xchk_ino_warning(sc, ino, __return_address);
296 0 : }
297 :
298 : /* Warn about a block indexed by a file fork that needs review. */
299 : void
300 147 : xchk_fblock_set_warning(
301 : struct xfs_scrub *sc,
302 : int whichfork,
303 : xfs_fileoff_t offset)
304 : {
305 147 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
306 147 : trace_xchk_fblock_warning(sc, whichfork, offset, __return_address);
307 147 : }
308 :
309 : /* Signal an incomplete scrub. */
310 : void
311 539825 : xchk_set_incomplete(
312 : struct xfs_scrub *sc)
313 : {
314 539825 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_INCOMPLETE;
315 539825 : trace_xchk_incomplete(sc, __return_address);
316 539825 : }
317 :
318 : /*
319 : * rmap scrubbing -- compute the number of blocks with a given owner,
320 : * at least according to the reverse mapping data.
321 : */
322 :
323 : struct xchk_rmap_ownedby_info {
324 : const struct xfs_owner_info *oinfo;
325 : xfs_filblks_t *blocks;
326 : };
327 :
328 : STATIC int
329 3849676514 : xchk_count_rmap_ownedby_irec(
330 : struct xfs_btree_cur *cur,
331 : const struct xfs_rmap_irec *rec,
332 : void *priv)
333 : {
334 3849676514 : struct xchk_rmap_ownedby_info *sroi = priv;
335 3849676514 : bool irec_attr;
336 3849676514 : bool oinfo_attr;
337 :
338 3849676514 : irec_attr = rec->rm_flags & XFS_RMAP_ATTR_FORK;
339 3849676514 : oinfo_attr = sroi->oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK;
340 :
341 3849676514 : if (rec->rm_owner != sroi->oinfo->oi_owner)
342 : return 0;
343 :
344 25024148 : if (XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) || irec_attr == oinfo_attr)
345 25024148 : (*sroi->blocks) += rec->rm_blockcount;
346 :
347 : return 0;
348 : }
349 :
350 : /*
351 : * Calculate the number of blocks the rmap thinks are owned by something.
352 : * The caller should pass us an rmapbt cursor.
353 : */
354 : int
355 2705506 : xchk_count_rmap_ownedby_ag(
356 : struct xfs_scrub *sc,
357 : struct xfs_btree_cur *cur,
358 : const struct xfs_owner_info *oinfo,
359 : xfs_filblks_t *blocks)
360 : {
361 2705506 : struct xchk_rmap_ownedby_info sroi = {
362 : .oinfo = oinfo,
363 : .blocks = blocks,
364 : };
365 :
366 2705506 : *blocks = 0;
367 2705506 : return xfs_rmap_query_all(cur, xchk_count_rmap_ownedby_irec,
368 : &sroi);
369 : }
370 :
371 : /*
372 : * AG scrubbing
373 : *
374 : * These helpers facilitate locking an allocation group's header
375 : * buffers, setting up cursors for all btrees that are present, and
376 : * cleaning everything up once we're through.
377 : */
378 :
379 : /* Decide if we want to return an AG header read failure. */
380 : static inline bool
381 : want_ag_read_header_failure(
382 : struct xfs_scrub *sc,
383 : unsigned int type)
384 : {
385 : /* Return all AG header read failures when scanning btrees. */
386 0 : if (sc->sm->sm_type != XFS_SCRUB_TYPE_AGF &&
387 0 : sc->sm->sm_type != XFS_SCRUB_TYPE_AGFL &&
388 : sc->sm->sm_type != XFS_SCRUB_TYPE_AGI)
389 : return true;
390 : /*
391 : * If we're scanning a given type of AG header, we only want to
392 : * see read failures from that specific header. We'd like the
393 : * other headers to cross-check them, but this isn't required.
394 : */
395 0 : if (sc->sm->sm_type == type)
396 : return true;
397 : return false;
398 : }
399 :
400 : /*
401 : * Grab the AG header buffers for the attached perag structure.
402 : *
403 : * The headers should be released by xchk_ag_free, but as a fail safe we attach
404 : * all the buffers we grab to the scrub transaction so they'll all be freed
405 : * when we cancel it.
406 : */
407 : static inline int
408 673644305 : xchk_perag_read_headers(
409 : struct xfs_scrub *sc,
410 : struct xchk_ag *sa)
411 : {
412 673644305 : int error;
413 :
414 673644305 : error = xfs_ialloc_read_agi(sa->pag, sc->tp, &sa->agi_bp);
415 674237522 : if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGI))
416 : return error;
417 :
418 674237522 : error = xfs_alloc_read_agf(sa->pag, sc->tp, 0, &sa->agf_bp);
419 674239918 : if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGF))
420 0 : return error;
421 :
422 : return 0;
423 : }
424 :
425 : /*
426 : * Grab the AG headers for the attached perag structure and wait for pending
427 : * intents to drain.
428 : */
429 : static int
430 674010136 : xchk_perag_drain_and_lock(
431 : struct xfs_scrub *sc)
432 : {
433 674010136 : struct xchk_ag *sa = &sc->sa;
434 674010136 : int error = 0;
435 :
436 674010136 : ASSERT(sa->pag != NULL);
437 674010136 : ASSERT(sa->agi_bp == NULL);
438 674010136 : ASSERT(sa->agf_bp == NULL);
439 :
440 674058288 : do {
441 674058288 : if (xchk_should_terminate(sc, &error))
442 19 : return error;
443 :
444 673612865 : error = xchk_perag_read_headers(sc, sa);
445 674208311 : if (error)
446 0 : return error;
447 :
448 : /*
449 : * If we've grabbed an inode for scrubbing then we assume that
450 : * holding its ILOCK will suffice to coordinate with any intent
451 : * chains involving this inode.
452 : */
453 674208311 : if (sc->ip)
454 : return 0;
455 :
456 : /*
457 : * Decide if this AG is quiet enough for all metadata to be
458 : * consistent with each other. XFS allows the AG header buffer
459 : * locks to cycle across transaction rolls while processing
460 : * chains of deferred ops, which means that there could be
461 : * other threads in the middle of processing a chain of
462 : * deferred ops. For regular operations we are careful about
463 : * ordering operations to prevent collisions between threads
464 : * (which is why we don't need a per-AG lock), but scrub and
465 : * repair have to serialize against chained operations.
466 : *
467 : * We just locked all the AG headers buffers; now take a look
468 : * to see if there are any intents in progress. If there are,
469 : * drop the AG headers and wait for the intents to drain.
470 : * Since we hold all the AG header locks for the duration of
471 : * the scrub, this is the only time we have to sample the
472 : * intents counter; any threads increasing it after this point
473 : * can't possibly be in the middle of a chain of AG metadata
474 : * updates.
475 : *
476 : * Obviously, this should be slanted against scrub and in favor
477 : * of runtime threads.
478 : */
479 9950232 : if (!xfs_perag_intent_busy(sa->pag))
480 : return 0;
481 :
482 152079 : if (sa->agf_bp) {
483 152079 : xfs_trans_brelse(sc->tp, sa->agf_bp);
484 152079 : sa->agf_bp = NULL;
485 : }
486 :
487 152079 : if (sa->agi_bp) {
488 152079 : xfs_trans_brelse(sc->tp, sa->agi_bp);
489 152078 : sa->agi_bp = NULL;
490 : }
491 :
492 152078 : if (!(sc->flags & XCHK_FSGATES_DRAIN))
493 : return -ECHRNG;
494 48152 : error = xfs_perag_intent_drain(sa->pag);
495 48152 : if (error == -ERESTARTSYS)
496 0 : error = -EINTR;
497 48152 : } while (!error);
498 :
499 : return error;
500 : }
501 :
502 : /*
503 : * Grab the per-AG structure, grab all AG header buffers, and wait until there
504 : * aren't any pending intents. Returns -ENOENT if we can't grab the perag
505 : * structure.
506 : */
507 : int
508 673039696 : xchk_ag_read_headers(
509 : struct xfs_scrub *sc,
510 : xfs_agnumber_t agno,
511 : struct xchk_ag *sa)
512 : {
513 673039696 : struct xfs_mount *mp = sc->mp;
514 :
515 673039696 : ASSERT(!sa->pag);
516 673039696 : sa->pag = xfs_perag_get(mp, agno);
517 674242407 : if (!sa->pag)
518 : return -ENOENT;
519 :
520 674242407 : return xchk_perag_drain_and_lock(sc);
521 : }
522 :
523 : /* Release all the AG btree cursors. */
524 : void
525 3166104748 : xchk_ag_btcur_free(
526 : struct xchk_ag *sa)
527 : {
528 3166104748 : if (sa->refc_cur)
529 423302525 : xfs_btree_del_cursor(sa->refc_cur, XFS_BTREE_ERROR);
530 3166556471 : if (sa->rmap_cur)
531 423596534 : xfs_btree_del_cursor(sa->rmap_cur, XFS_BTREE_ERROR);
532 3166494994 : if (sa->fino_cur)
533 673557905 : xfs_btree_del_cursor(sa->fino_cur, XFS_BTREE_ERROR);
534 3167350224 : if (sa->ino_cur)
535 674232541 : xfs_btree_del_cursor(sa->ino_cur, XFS_BTREE_ERROR);
536 3167548439 : if (sa->cnt_cur)
537 674391968 : xfs_btree_del_cursor(sa->cnt_cur, XFS_BTREE_ERROR);
538 3167472135 : if (sa->bno_cur)
539 674296955 : xfs_btree_del_cursor(sa->bno_cur, XFS_BTREE_ERROR);
540 :
541 3167139883 : sa->refc_cur = NULL;
542 3167139883 : sa->rmap_cur = NULL;
543 3167139883 : sa->fino_cur = NULL;
544 3167139883 : sa->ino_cur = NULL;
545 3167139883 : sa->bno_cur = NULL;
546 3167139883 : sa->cnt_cur = NULL;
547 3167139883 : }
548 :
549 : /* Initialize all the btree cursors for an AG. */
550 : void
551 673130519 : xchk_ag_btcur_init(
552 : struct xfs_scrub *sc,
553 : struct xchk_ag *sa)
554 : {
555 673130519 : struct xfs_mount *mp = sc->mp;
556 :
557 1347017620 : if (sa->agf_bp &&
558 673730248 : xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_BNO)) {
559 : /* Set up a bnobt cursor for cross-referencing. */
560 673897239 : sa->bno_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
561 : sa->pag, XFS_BTNUM_BNO);
562 : }
563 :
564 1347886864 : if (sa->agf_bp &&
565 673870447 : xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_CNT)) {
566 : /* Set up a cntbt cursor for cross-referencing. */
567 674310166 : sa->cnt_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
568 : sa->pag, XFS_BTNUM_CNT);
569 : }
570 :
571 : /* Set up a inobt cursor for cross-referencing. */
572 1348442130 : if (sa->agi_bp &&
573 674497631 : xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_INO)) {
574 674250909 : sa->ino_cur = xfs_inobt_init_cursor(sa->pag, sc->tp, sa->agi_bp,
575 : XFS_BTNUM_INO);
576 : }
577 :
578 : /* Set up a finobt cursor for cross-referencing. */
579 1348301526 : if (sa->agi_bp && xfs_has_finobt(mp) &&
580 674101689 : xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_FINO)) {
581 674263410 : sa->fino_cur = xfs_inobt_init_cursor(sa->pag, sc->tp, sa->agi_bp,
582 : XFS_BTNUM_FINO);
583 : }
584 :
585 : /* Set up a rmapbt cursor for cross-referencing. */
586 1097827303 : if (sa->agf_bp && xfs_has_rmapbt(mp) &&
587 423609474 : xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_RMAP)) {
588 423547952 : sa->rmap_cur = xfs_rmapbt_init_cursor(mp, sc->tp, sa->agf_bp,
589 : sa->pag);
590 : }
591 :
592 : /* Set up a refcountbt cursor for cross-referencing. */
593 1097905847 : if (sa->agf_bp && xfs_has_reflink(mp) &&
594 423684154 : xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_REFC)) {
595 423560278 : sa->refc_cur = xfs_refcountbt_init_cursor(mp, sc->tp,
596 : sa->agf_bp, sa->pag);
597 : }
598 674330569 : }
599 :
600 : /* Release the AG header context and btree cursors. */
601 : void
602 2851012226 : xchk_ag_free(
603 : struct xfs_scrub *sc,
604 : struct xchk_ag *sa)
605 : {
606 2851012226 : xchk_ag_btcur_free(sa);
607 2851921935 : if (sa->agf_bp) {
608 673982410 : xfs_trans_brelse(sc->tp, sa->agf_bp);
609 674453610 : sa->agf_bp = NULL;
610 : }
611 2852393135 : if (sa->agi_bp) {
612 674454112 : xfs_trans_brelse(sc->tp, sa->agi_bp);
613 674485466 : sa->agi_bp = NULL;
614 : }
615 2852424489 : if (sa->pag) {
616 674587265 : xfs_perag_put(sa->pag);
617 674594389 : sa->pag = NULL;
618 : }
619 2852431613 : }
620 :
621 : /*
622 : * For scrub, grab the perag structure, the AGI, and the AGF headers, in that
623 : * order. Locking order requires us to get the AGI before the AGF. We use the
624 : * transaction to avoid deadlocking on crosslinked metadata buffers; either the
625 : * caller passes one in (bmap scrub) or we have to create a transaction
626 : * ourselves. Returns ENOENT if the perag struct cannot be grabbed.
627 : */
628 : int
629 669245269 : xchk_ag_init(
630 : struct xfs_scrub *sc,
631 : xfs_agnumber_t agno,
632 : struct xchk_ag *sa)
633 : {
634 669245269 : int error;
635 :
636 669245269 : error = xchk_ag_read_headers(sc, agno, sa);
637 670032151 : if (error)
638 : return error;
639 :
640 669994071 : xchk_ag_btcur_init(sc, sa);
641 669994071 : return 0;
642 : }
643 :
644 : /* Per-scrubber setup functions */
645 :
646 : void
647 242 : xchk_trans_cancel(
648 : struct xfs_scrub *sc)
649 : {
650 242 : xfs_trans_cancel(sc->tp);
651 1442 : sc->tp = NULL;
652 0 : }
653 :
654 : /*
655 : * Grab an empty transaction so that we can re-grab locked buffers if
656 : * one of our btrees turns out to be cyclic.
657 : *
658 : * If we're going to repair something, we need to ask for the largest possible
659 : * log reservation so that we can handle the worst case scenario for metadata
660 : * updates while rebuilding a metadata item. We also need to reserve as many
661 : * blocks in the head transaction as we think we're going to need to rebuild
662 : * the metadata object.
663 : */
664 : int
665 2178546011 : xchk_trans_alloc(
666 : struct xfs_scrub *sc,
667 : uint resblks)
668 : {
669 2178546011 : if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
670 317787118 : return xfs_trans_alloc(sc->mp, &M_RES(sc->mp)->tr_itruncate,
671 : resblks, 0, 0, &sc->tp);
672 :
673 1860758893 : return xfs_trans_alloc_empty(sc->mp, &sc->tp);
674 : }
675 :
676 : /* Set us up with a transaction and an empty context. */
677 : int
678 11522512 : xchk_setup_fs(
679 : struct xfs_scrub *sc)
680 : {
681 11522512 : uint resblks;
682 :
683 11522512 : resblks = xrep_calc_ag_resblks(sc);
684 11522823 : return xchk_trans_alloc(sc, resblks);
685 : }
686 :
687 : /* Set us up with AG headers and btree cursors. */
688 : int
689 4591965 : xchk_setup_ag_btree(
690 : struct xfs_scrub *sc,
691 : bool force_log)
692 : {
693 4591965 : struct xfs_mount *mp = sc->mp;
694 4591965 : int error;
695 :
696 : /*
697 : * If the caller asks us to checkpont the log, do so. This
698 : * expensive operation should be performed infrequently and only
699 : * as a last resort. Any caller that sets force_log should
700 : * document why they need to do so.
701 : */
702 4591965 : if (force_log) {
703 0 : error = xchk_checkpoint_log(mp);
704 0 : if (error)
705 : return error;
706 : }
707 :
708 4591965 : error = xchk_setup_fs(sc);
709 4586809 : if (error)
710 : return error;
711 :
712 4585622 : return xchk_ag_init(sc, sc->sm->sm_agno, &sc->sa);
713 : }
714 :
715 : /* Push everything out of the log onto disk. */
716 : int
717 0 : xchk_checkpoint_log(
718 : struct xfs_mount *mp)
719 : {
720 0 : int error;
721 :
722 0 : error = xfs_log_force(mp, XFS_LOG_SYNC);
723 0 : if (error)
724 : return error;
725 0 : xfs_ail_push_all_sync(mp->m_ail);
726 0 : return 0;
727 : }
728 :
729 : /* Verify that an inode is allocated ondisk, then return its cached inode. */
730 : int
731 2250783602 : xchk_iget(
732 : struct xfs_scrub *sc,
733 : xfs_ino_t inum,
734 : struct xfs_inode **ipp)
735 : {
736 2250783602 : return xfs_iget(sc->mp, sc->tp, inum, XFS_IGET_UNTRUSTED, 0, ipp);
737 : }
738 :
739 : /*
740 : * Try to grab an inode in a manner that avoids races with physical inode
741 : * allocation. If we can't, return the locked AGI buffer so that the caller
742 : * can single-step the loading process to see where things went wrong.
743 : * Callers must have a valid scrub transaction.
744 : *
745 : * If the iget succeeds, return 0, a NULL AGI, and the inode.
746 : *
747 : * If the iget fails, return the error, the locked AGI, and a NULL inode. This
748 : * can include -EINVAL and -ENOENT for invalid inode numbers or inodes that are
749 : * no longer allocated; or any other corruption or runtime error.
750 : *
751 : * If the AGI read fails, return the error, a NULL AGI, and NULL inode.
752 : *
753 : * If a fatal signal is pending, return -EINTR, a NULL AGI, and a NULL inode.
754 : */
755 : int
756 1442 : xchk_iget_agi(
757 : struct xfs_scrub *sc,
758 : xfs_ino_t inum,
759 : struct xfs_buf **agi_bpp,
760 : struct xfs_inode **ipp)
761 : {
762 1442 : struct xfs_mount *mp = sc->mp;
763 1442 : struct xfs_trans *tp = sc->tp;
764 1442 : struct xfs_perag *pag;
765 1442 : int error;
766 :
767 1442 : ASSERT(sc->tp != NULL);
768 :
769 1442 : again:
770 1442 : *agi_bpp = NULL;
771 1442 : *ipp = NULL;
772 1442 : error = 0;
773 :
774 1442 : if (xchk_should_terminate(sc, &error))
775 0 : return error;
776 :
777 : /*
778 : * Attach the AGI buffer to the scrub transaction to avoid deadlocks
779 : * in the iget cache miss path.
780 : */
781 1442 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
782 1442 : error = xfs_ialloc_read_agi(pag, tp, agi_bpp);
783 1442 : xfs_perag_put(pag);
784 1442 : if (error)
785 0 : return error;
786 :
787 1442 : error = xfs_iget(mp, tp, inum,
788 : XFS_IGET_NORETRY | XFS_IGET_UNTRUSTED, 0, ipp);
789 1442 : if (error == -EAGAIN) {
790 : /*
791 : * The inode may be in core but temporarily unavailable and may
792 : * require the AGI buffer before it can be returned. Drop the
793 : * AGI buffer and retry the lookup.
794 : *
795 : * Incore lookup will fail with EAGAIN on a cache hit if the
796 : * inode is queued to the inactivation list. The inactivation
797 : * worker may remove the inode from the unlinked list and hence
798 : * needs the AGI.
799 : *
800 : * Hence xchk_iget_agi() needs to drop the AGI lock on EAGAIN
801 : * to allow inodegc to make progress and move the inode to
802 : * IRECLAIMABLE state where xfs_iget will be able to return it
803 : * again if it can lock the inode.
804 : */
805 0 : xfs_trans_brelse(tp, *agi_bpp);
806 0 : delay(1);
807 0 : goto again;
808 : }
809 1442 : if (error)
810 : return error;
811 :
812 : /* We got the inode, so we can release the AGI. */
813 0 : ASSERT(*ipp != NULL);
814 0 : xfs_trans_brelse(tp, *agi_bpp);
815 0 : *agi_bpp = NULL;
816 0 : return 0;
817 : }
818 :
819 : /* Install an inode that we opened by handle for scrubbing. */
820 : int
821 1470678913 : xchk_install_handle_inode(
822 : struct xfs_scrub *sc,
823 : struct xfs_inode *ip)
824 : {
825 1470678913 : if (VFS_I(ip)->i_generation != sc->sm->sm_gen) {
826 1396365 : xchk_irele(sc, ip);
827 1396365 : return -ENOENT;
828 : }
829 :
830 1469282548 : sc->ip = ip;
831 1469282548 : return 0;
832 : }
833 :
834 : /*
835 : * In preparation to scrub metadata structures that hang off of an inode,
836 : * grab either the inode referenced in the scrub control structure or the
837 : * inode passed in. If the inumber does not reference an allocated inode
838 : * record, the function returns ENOENT to end the scrub early. The inode
839 : * is not locked.
840 : */
841 : int
842 1799165910 : xchk_iget_for_scrubbing(
843 : struct xfs_scrub *sc)
844 : {
845 1799165910 : struct xfs_imap imap;
846 1799165910 : struct xfs_mount *mp = sc->mp;
847 1799165910 : struct xfs_perag *pag;
848 1799165910 : struct xfs_buf *agi_bp;
849 1799165910 : struct xfs_inode *ip_in = XFS_I(file_inode(sc->file));
850 1799165910 : struct xfs_inode *ip = NULL;
851 1799165910 : xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, sc->sm->sm_ino);
852 1799165910 : int error;
853 :
854 1799165910 : ASSERT(sc->tp == NULL);
855 :
856 : /* We want to scan the inode we already had opened. */
857 1799165910 : if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
858 583974585 : sc->ip = ip_in;
859 583974585 : return 0;
860 : }
861 :
862 : /* Reject internal metadata files and obviously bad inode numbers. */
863 1215191325 : if (xfs_internal_inum(mp, sc->sm->sm_ino))
864 : return -ENOENT;
865 1210913530 : if (!xfs_verify_ino(sc->mp, sc->sm->sm_ino))
866 : return -ENOENT;
867 :
868 : /* Try a regular untrusted iget. */
869 1212023866 : error = xchk_iget(sc, sc->sm->sm_ino, &ip);
870 1213879909 : if (!error)
871 1208622719 : return xchk_install_handle_inode(sc, ip);
872 5257190 : if (error == -ENOENT)
873 : return error;
874 1200 : if (error != -EINVAL)
875 0 : goto out_error;
876 :
877 : /*
878 : * EINVAL with IGET_UNTRUSTED probably means one of several things:
879 : * userspace gave us an inode number that doesn't correspond to fs
880 : * space; the inode btree lacks a record for this inode; or there is a
881 : * record, and it says this inode is free.
882 : *
883 : * We want to look up this inode in the inobt to distinguish two
884 : * scenarios: (1) the inobt says the inode is free, in which case
885 : * there's nothing to do; and (2) the inobt says the inode is
886 : * allocated, but loading it failed due to corruption.
887 : *
888 : * Allocate a transaction and grab the AGI to prevent inobt activity
889 : * in this AG. Retry the iget in case someone allocated a new inode
890 : * after the first iget failed.
891 : */
892 1200 : error = xchk_trans_alloc(sc, 0);
893 1200 : if (error)
894 0 : goto out_error;
895 :
896 1200 : error = xchk_iget_agi(sc, sc->sm->sm_ino, &agi_bp, &ip);
897 1200 : if (error == 0) {
898 : /* Actually got the inode, so install it. */
899 0 : xchk_trans_cancel(sc);
900 0 : return xchk_install_handle_inode(sc, ip);
901 : }
902 1200 : if (error == -ENOENT)
903 0 : goto out_gone;
904 1200 : if (error != -EINVAL)
905 0 : goto out_cancel;
906 :
907 : /* Ensure that we have protected against inode allocation/freeing. */
908 1200 : if (agi_bp == NULL) {
909 0 : ASSERT(agi_bp != NULL);
910 0 : error = -ECANCELED;
911 0 : goto out_cancel;
912 : }
913 :
914 : /*
915 : * Untrusted iget failed a second time. Let's try an inobt lookup.
916 : * If the inobt thinks this the inode neither can exist inside the
917 : * filesystem nor is allocated, return ENOENT to signal that the check
918 : * can be skipped.
919 : *
920 : * If the lookup returns corruption, we'll mark this inode corrupt and
921 : * exit to userspace. There's little chance of fixing anything until
922 : * the inobt is straightened out, but there's nothing we can do here.
923 : *
924 : * If the lookup encounters any other error, exit to userspace.
925 : *
926 : * If the lookup succeeds, something else must be very wrong in the fs
927 : * such that setting up the incore inode failed in some strange way.
928 : * Treat those as corruptions.
929 : */
930 1200 : pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sc->sm->sm_ino));
931 1200 : if (!pag) {
932 0 : error = -EFSCORRUPTED;
933 0 : goto out_cancel;
934 : }
935 :
936 1200 : error = xfs_imap(pag, sc->tp, sc->sm->sm_ino, &imap,
937 : XFS_IGET_UNTRUSTED);
938 1200 : xfs_perag_put(pag);
939 1200 : if (error == -EINVAL || error == -ENOENT)
940 1200 : goto out_gone;
941 0 : if (!error)
942 0 : error = -EFSCORRUPTED;
943 :
944 0 : out_cancel:
945 0 : xchk_trans_cancel(sc);
946 0 : out_error:
947 0 : trace_xchk_op_error(sc, agno, XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
948 : error, __return_address);
949 0 : return error;
950 1200 : out_gone:
951 : /* The file is gone, so there's nothing to check. */
952 1200 : xchk_trans_cancel(sc);
953 1200 : return -ENOENT;
954 : }
955 :
956 : /* Release an inode, possibly dropping it in the process. */
957 : void
958 2246534792 : xchk_irele(
959 : struct xfs_scrub *sc,
960 : struct xfs_inode *ip)
961 : {
962 2246534792 : if (current->journal_info != NULL) {
963 778377742 : ASSERT(current->journal_info == sc->tp);
964 :
965 : /*
966 : * If we are in a transaction, we /cannot/ drop the inode
967 : * ourselves, because the VFS will trigger writeback, which
968 : * can require a transaction. Clear DONTCACHE to force the
969 : * inode to the LRU, where someone else can take care of
970 : * dropping it.
971 : *
972 : * Note that when we grabbed our reference to the inode, it
973 : * could have had an active ref and DONTCACHE set if a sysadmin
974 : * is trying to coerce a change in file access mode. icache
975 : * hits do not clear DONTCACHE, so we must do it here.
976 : */
977 778377742 : spin_lock(&VFS_I(ip)->i_lock);
978 779645531 : VFS_I(ip)->i_state &= ~I_DONTCACHE;
979 779645531 : spin_unlock(&VFS_I(ip)->i_lock);
980 1468157050 : } else if (atomic_read(&VFS_I(ip)->i_count) == 1) {
981 : /*
982 : * If this is the last reference to the inode and the caller
983 : * permits it, set DONTCACHE to avoid thrashing.
984 : */
985 237762730 : d_mark_dontcache(VFS_I(ip));
986 : }
987 :
988 2248503081 : xfs_irele(ip);
989 2247097840 : }
990 :
991 : /*
992 : * Set us up to scrub metadata mapped by a file's fork. Callers must not use
993 : * this to operate on user-accessible regular file data because the MMAPLOCK is
994 : * not taken.
995 : */
996 : int
997 744947763 : xchk_setup_inode_contents(
998 : struct xfs_scrub *sc,
999 : unsigned int resblks)
1000 : {
1001 744947763 : int error;
1002 :
1003 744947763 : error = xchk_iget_for_scrubbing(sc);
1004 745617689 : if (error)
1005 : return error;
1006 :
1007 : /* Lock the inode so the VFS cannot touch this file. */
1008 741993934 : sc->ilock_flags = XFS_IOLOCK_EXCL;
1009 741993934 : xfs_ilock(sc->ip, sc->ilock_flags);
1010 :
1011 741669458 : error = xchk_trans_alloc(sc, resblks);
1012 741538622 : if (error)
1013 0 : goto out;
1014 741538622 : sc->ilock_flags |= XFS_ILOCK_EXCL;
1015 741538622 : xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
1016 :
1017 : out:
1018 : /* scrub teardown will unlock and release the inode for us */
1019 : return error;
1020 : }
1021 :
1022 : /*
1023 : * Predicate that decides if we need to evaluate the cross-reference check.
1024 : * If there was an error accessing the cross-reference btree, just delete
1025 : * the cursor and skip the check.
1026 : */
1027 : bool
1028 11531060536 : xchk_should_check_xref(
1029 : struct xfs_scrub *sc,
1030 : int *error,
1031 : struct xfs_btree_cur **curpp)
1032 : {
1033 : /* No point in xref if we already know we're corrupt. */
1034 11531060536 : if (xchk_skip_xref(sc->sm))
1035 : return false;
1036 :
1037 11531060536 : if (*error == 0)
1038 : return true;
1039 :
1040 0 : if (curpp) {
1041 : /* If we've already given up on xref, just bail out. */
1042 0 : if (!*curpp)
1043 : return false;
1044 :
1045 : /* xref error, delete cursor and bail out. */
1046 0 : xfs_btree_del_cursor(*curpp, XFS_BTREE_ERROR);
1047 0 : *curpp = NULL;
1048 : }
1049 :
1050 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
1051 0 : trace_xchk_xref_error(sc, *error, __return_address);
1052 :
1053 : /*
1054 : * Errors encountered during cross-referencing with another
1055 : * data structure should not cause this scrubber to abort.
1056 : */
1057 0 : *error = 0;
1058 0 : return false;
1059 : }
1060 :
1061 : /* Run the structure verifiers on in-memory buffers to detect bad memory. */
1062 : void
1063 150553504 : xchk_buffer_recheck(
1064 : struct xfs_scrub *sc,
1065 : struct xfs_buf *bp)
1066 : {
1067 150553504 : xfs_failaddr_t fa;
1068 :
1069 150553504 : if (bp->b_ops == NULL) {
1070 0 : xchk_block_set_corrupt(sc, bp);
1071 0 : return;
1072 : }
1073 150553504 : if (bp->b_ops->verify_struct == NULL) {
1074 0 : xchk_set_incomplete(sc);
1075 0 : return;
1076 : }
1077 150553504 : fa = bp->b_ops->verify_struct(bp);
1078 151228902 : if (!fa)
1079 : return;
1080 0 : sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
1081 0 : trace_xchk_block_error(sc, xfs_buf_daddr(bp), fa);
1082 : }
1083 :
1084 : static inline int
1085 1957300 : xchk_metadata_inode_subtype(
1086 : struct xfs_scrub *sc,
1087 : unsigned int scrub_type)
1088 : {
1089 1957300 : __u32 smtype = sc->sm->sm_type;
1090 1957300 : int error;
1091 :
1092 1957300 : sc->sm->sm_type = scrub_type;
1093 :
1094 1957300 : switch (scrub_type) {
1095 978650 : case XFS_SCRUB_TYPE_INODE:
1096 978650 : error = xchk_inode(sc);
1097 978650 : break;
1098 978650 : case XFS_SCRUB_TYPE_BMBTD:
1099 978650 : error = xchk_bmap_data(sc);
1100 978650 : break;
1101 0 : default:
1102 0 : ASSERT(0);
1103 0 : error = -EFSCORRUPTED;
1104 0 : break;
1105 : }
1106 :
1107 1957300 : sc->sm->sm_type = smtype;
1108 1957300 : return error;
1109 : }
1110 :
1111 : /*
1112 : * Scrub the attr/data forks of a metadata inode. The metadata inode must be
1113 : * pointed to by sc->ip and the ILOCK must be held.
1114 : */
1115 : int
1116 978650 : xchk_metadata_inode_forks(
1117 : struct xfs_scrub *sc)
1118 : {
1119 978650 : bool shared;
1120 978650 : int error;
1121 :
1122 978650 : if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
1123 : return 0;
1124 :
1125 : /* Check the inode record. */
1126 978650 : error = xchk_metadata_inode_subtype(sc, XFS_SCRUB_TYPE_INODE);
1127 978650 : if (error || (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
1128 : return error;
1129 :
1130 : /* Metadata inodes don't live on the rt device. */
1131 978650 : if (sc->ip->i_diflags & XFS_DIFLAG_REALTIME) {
1132 0 : xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1133 0 : return 0;
1134 : }
1135 :
1136 : /* They should never participate in reflink. */
1137 978650 : if (xfs_is_reflink_inode(sc->ip)) {
1138 0 : xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1139 0 : return 0;
1140 : }
1141 :
1142 : /* They also should never have extended attributes. */
1143 978650 : if (xfs_inode_hasattr(sc->ip)) {
1144 0 : xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1145 0 : return 0;
1146 : }
1147 :
1148 : /* Invoke the data fork scrubber. */
1149 978650 : error = xchk_metadata_inode_subtype(sc, XFS_SCRUB_TYPE_BMBTD);
1150 978650 : if (error || (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
1151 : return error;
1152 :
1153 : /* Look for incorrect shared blocks. */
1154 978649 : if (xfs_has_reflink(sc->mp)) {
1155 349155 : error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
1156 : &shared);
1157 698310 : if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
1158 : &error))
1159 0 : return error;
1160 349155 : if (shared)
1161 0 : xchk_ino_set_corrupt(sc, sc->ip->i_ino);
1162 : }
1163 :
1164 : return 0;
1165 : }
1166 :
1167 : /*
1168 : * Enable filesystem hooks (i.e. runtime code patching) before starting a scrub
1169 : * operation. Callers must not hold any locks that intersect with the CPU
1170 : * hotplug lock (e.g. writeback locks) because code patching must halt the CPUs
1171 : * to change kernel code.
1172 : */
1173 : void
1174 103673 : xchk_fsgates_enable(
1175 : struct xfs_scrub *sc,
1176 : unsigned int scrub_fsgates)
1177 : {
1178 103673 : ASSERT(!(scrub_fsgates & ~XCHK_FSGATES_ALL));
1179 103673 : ASSERT(!(sc->flags & scrub_fsgates));
1180 :
1181 103673 : trace_xchk_fsgates_enable(sc, scrub_fsgates);
1182 :
1183 103673 : if (scrub_fsgates & XCHK_FSGATES_DRAIN)
1184 103673 : xfs_drain_wait_enable();
1185 :
1186 103677 : sc->flags |= scrub_fsgates;
1187 103677 : }
|
