blob: 8033ab9d8f47d51f63e64ac168f3a9ac82d23055 [file] [log] [blame]
Darrick J. Wongdcb660f2017-10-17 21:37:36 -07001/*
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_inode.h"
Darrick J. Wong80e4e122017-10-17 21:37:42 -070033#include "xfs_icache.h"
34#include "xfs_itable.h"
Darrick J. Wongdcb660f2017-10-17 21:37:36 -070035#include "xfs_alloc.h"
36#include "xfs_alloc_btree.h"
37#include "xfs_bmap.h"
38#include "xfs_bmap_btree.h"
39#include "xfs_ialloc.h"
40#include "xfs_ialloc_btree.h"
41#include "xfs_refcount.h"
42#include "xfs_refcount_btree.h"
43#include "xfs_rmap.h"
44#include "xfs_rmap_btree.h"
Darrick J. Wong3daa6642017-10-17 21:37:40 -070045#include "xfs_log.h"
46#include "xfs_trans_priv.h"
Darrick J. Wongdcb660f2017-10-17 21:37:36 -070047#include "scrub/xfs_scrub.h"
48#include "scrub/scrub.h"
49#include "scrub/common.h"
50#include "scrub/trace.h"
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -070051#include "scrub/btree.h"
Darrick J. Wongdcb660f2017-10-17 21:37:36 -070052
53/* Common code for the metadata scrubbers. */
54
Darrick J. Wong4700d222017-10-17 21:37:36 -070055/*
56 * Handling operational errors.
57 *
58 * The *_process_error() family of functions are used to process error return
59 * codes from functions called as part of a scrub operation.
60 *
61 * If there's no error, we return true to tell the caller that it's ok
62 * to move on to the next check in its list.
63 *
64 * For non-verifier errors (e.g. ENOMEM) we return false to tell the
65 * caller that something bad happened, and we preserve *error so that
66 * the caller can return the *error up the stack to userspace.
67 *
68 * Verifier errors (EFSBADCRC/EFSCORRUPTED) are recorded by setting
69 * OFLAG_CORRUPT in sm_flags and the *error is cleared. In other words,
70 * we track verifier errors (and failed scrub checks) via OFLAG_CORRUPT,
71 * not via return codes. We return false to tell the caller that
72 * something bad happened. Since the error has been cleared, the caller
73 * will (presumably) return that zero and scrubbing will move on to
74 * whatever's next.
75 *
76 * ftrace can be used to record the precise metadata location and the
77 * approximate code location of the failed operation.
78 */
79
80/* Check for operational errors. */
Darrick J. Wong64b12562018-01-16 18:52:14 -080081static bool
82__xfs_scrub_process_error(
83 struct xfs_scrub_context *sc,
84 xfs_agnumber_t agno,
85 xfs_agblock_t bno,
86 int *error,
87 __u32 errflag,
88 void *ret_ip)
89{
90 switch (*error) {
91 case 0:
92 return true;
93 case -EDEADLOCK:
94 /* Used to restart an op with deadlock avoidance. */
95 trace_xfs_scrub_deadlock_retry(sc->ip, sc->sm, *error);
96 break;
97 case -EFSBADCRC:
98 case -EFSCORRUPTED:
99 /* Note the badness but don't abort. */
100 sc->sm->sm_flags |= errflag;
101 *error = 0;
102 /* fall through */
103 default:
104 trace_xfs_scrub_op_error(sc, agno, bno, *error,
105 ret_ip);
106 break;
107 }
108 return false;
109}
110
Darrick J. Wong4700d222017-10-17 21:37:36 -0700111bool
112xfs_scrub_process_error(
113 struct xfs_scrub_context *sc,
114 xfs_agnumber_t agno,
115 xfs_agblock_t bno,
116 int *error)
117{
Darrick J. Wong64b12562018-01-16 18:52:14 -0800118 return __xfs_scrub_process_error(sc, agno, bno, error,
119 XFS_SCRUB_OFLAG_CORRUPT, __return_address);
120}
121
122bool
123xfs_scrub_xref_process_error(
124 struct xfs_scrub_context *sc,
125 xfs_agnumber_t agno,
126 xfs_agblock_t bno,
127 int *error)
128{
129 return __xfs_scrub_process_error(sc, agno, bno, error,
130 XFS_SCRUB_OFLAG_XFAIL, __return_address);
Darrick J. Wong4700d222017-10-17 21:37:36 -0700131}
132
133/* Check for operational errors for a file offset. */
Darrick J. Wong64b12562018-01-16 18:52:14 -0800134static bool
135__xfs_scrub_fblock_process_error(
Darrick J. Wong4700d222017-10-17 21:37:36 -0700136 struct xfs_scrub_context *sc,
137 int whichfork,
138 xfs_fileoff_t offset,
Darrick J. Wong64b12562018-01-16 18:52:14 -0800139 int *error,
140 __u32 errflag,
141 void *ret_ip)
Darrick J. Wong4700d222017-10-17 21:37:36 -0700142{
143 switch (*error) {
144 case 0:
145 return true;
146 case -EDEADLOCK:
147 /* Used to restart an op with deadlock avoidance. */
148 trace_xfs_scrub_deadlock_retry(sc->ip, sc->sm, *error);
149 break;
150 case -EFSBADCRC:
151 case -EFSCORRUPTED:
152 /* Note the badness but don't abort. */
Darrick J. Wong64b12562018-01-16 18:52:14 -0800153 sc->sm->sm_flags |= errflag;
Darrick J. Wong4700d222017-10-17 21:37:36 -0700154 *error = 0;
155 /* fall through */
156 default:
157 trace_xfs_scrub_file_op_error(sc, whichfork, offset, *error,
Darrick J. Wong64b12562018-01-16 18:52:14 -0800158 ret_ip);
Darrick J. Wong4700d222017-10-17 21:37:36 -0700159 break;
160 }
161 return false;
162}
163
Darrick J. Wong64b12562018-01-16 18:52:14 -0800164bool
165xfs_scrub_fblock_process_error(
166 struct xfs_scrub_context *sc,
167 int whichfork,
168 xfs_fileoff_t offset,
169 int *error)
170{
171 return __xfs_scrub_fblock_process_error(sc, whichfork, offset, error,
172 XFS_SCRUB_OFLAG_CORRUPT, __return_address);
173}
174
175bool
176xfs_scrub_fblock_xref_process_error(
177 struct xfs_scrub_context *sc,
178 int whichfork,
179 xfs_fileoff_t offset,
180 int *error)
181{
182 return __xfs_scrub_fblock_process_error(sc, whichfork, offset, error,
183 XFS_SCRUB_OFLAG_XFAIL, __return_address);
184}
185
Darrick J. Wong4700d222017-10-17 21:37:36 -0700186/*
187 * Handling scrub corruption/optimization/warning checks.
188 *
189 * The *_set_{corrupt,preen,warning}() family of functions are used to
190 * record the presence of metadata that is incorrect (corrupt), could be
191 * optimized somehow (preen), or should be flagged for administrative
192 * review but is not incorrect (warn).
193 *
194 * ftrace can be used to record the precise metadata location and
195 * approximate code location of the failed check.
196 */
197
198/* Record a block which could be optimized. */
199void
200xfs_scrub_block_set_preen(
201 struct xfs_scrub_context *sc,
202 struct xfs_buf *bp)
203{
204 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
205 trace_xfs_scrub_block_preen(sc, bp->b_bn, __return_address);
206}
207
208/*
209 * Record an inode which could be optimized. The trace data will
210 * include the block given by bp if bp is given; otherwise it will use
211 * the block location of the inode record itself.
212 */
213void
214xfs_scrub_ino_set_preen(
215 struct xfs_scrub_context *sc,
Darrick J. Wong0a1e1562017-11-06 11:46:15 -0800216 xfs_ino_t ino,
Darrick J. Wong4700d222017-10-17 21:37:36 -0700217 struct xfs_buf *bp)
218{
219 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
Darrick J. Wong0a1e1562017-11-06 11:46:15 -0800220 trace_xfs_scrub_ino_preen(sc, ino, bp ? bp->b_bn : 0,
Darrick J. Wong4700d222017-10-17 21:37:36 -0700221 __return_address);
222}
223
224/* Record a corrupt block. */
225void
226xfs_scrub_block_set_corrupt(
227 struct xfs_scrub_context *sc,
228 struct xfs_buf *bp)
229{
230 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
231 trace_xfs_scrub_block_error(sc, bp->b_bn, __return_address);
232}
233
Darrick J. Wong64b12562018-01-16 18:52:14 -0800234/* Record a corruption while cross-referencing. */
235void
236xfs_scrub_block_xref_set_corrupt(
237 struct xfs_scrub_context *sc,
238 struct xfs_buf *bp)
239{
240 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
241 trace_xfs_scrub_block_error(sc, bp->b_bn, __return_address);
242}
243
Darrick J. Wong4700d222017-10-17 21:37:36 -0700244/*
245 * Record a corrupt inode. The trace data will include the block given
246 * by bp if bp is given; otherwise it will use the block location of the
247 * inode record itself.
248 */
249void
250xfs_scrub_ino_set_corrupt(
251 struct xfs_scrub_context *sc,
252 xfs_ino_t ino,
253 struct xfs_buf *bp)
254{
255 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
256 trace_xfs_scrub_ino_error(sc, ino, bp ? bp->b_bn : 0, __return_address);
257}
258
Darrick J. Wong64b12562018-01-16 18:52:14 -0800259/* Record a corruption while cross-referencing with an inode. */
260void
261xfs_scrub_ino_xref_set_corrupt(
262 struct xfs_scrub_context *sc,
263 xfs_ino_t ino,
264 struct xfs_buf *bp)
265{
266 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
267 trace_xfs_scrub_ino_error(sc, ino, bp ? bp->b_bn : 0, __return_address);
268}
269
Darrick J. Wong4700d222017-10-17 21:37:36 -0700270/* Record corruption in a block indexed by a file fork. */
271void
272xfs_scrub_fblock_set_corrupt(
273 struct xfs_scrub_context *sc,
274 int whichfork,
275 xfs_fileoff_t offset)
276{
277 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
278 trace_xfs_scrub_fblock_error(sc, whichfork, offset, __return_address);
279}
280
Darrick J. Wong64b12562018-01-16 18:52:14 -0800281/* Record a corruption while cross-referencing a fork block. */
282void
283xfs_scrub_fblock_xref_set_corrupt(
284 struct xfs_scrub_context *sc,
285 int whichfork,
286 xfs_fileoff_t offset)
287{
288 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
289 trace_xfs_scrub_fblock_error(sc, whichfork, offset, __return_address);
290}
291
Darrick J. Wong4700d222017-10-17 21:37:36 -0700292/*
293 * Warn about inodes that need administrative review but is not
294 * incorrect.
295 */
296void
297xfs_scrub_ino_set_warning(
298 struct xfs_scrub_context *sc,
Darrick J. Wong0a1e1562017-11-06 11:46:15 -0800299 xfs_ino_t ino,
Darrick J. Wong4700d222017-10-17 21:37:36 -0700300 struct xfs_buf *bp)
301{
302 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
Darrick J. Wong0a1e1562017-11-06 11:46:15 -0800303 trace_xfs_scrub_ino_warning(sc, ino, bp ? bp->b_bn : 0,
Darrick J. Wong4700d222017-10-17 21:37:36 -0700304 __return_address);
305}
306
307/* Warn about a block indexed by a file fork that needs review. */
308void
309xfs_scrub_fblock_set_warning(
310 struct xfs_scrub_context *sc,
311 int whichfork,
312 xfs_fileoff_t offset)
313{
314 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
315 trace_xfs_scrub_fblock_warning(sc, whichfork, offset, __return_address);
316}
317
318/* Signal an incomplete scrub. */
319void
320xfs_scrub_set_incomplete(
321 struct xfs_scrub_context *sc)
322{
323 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_INCOMPLETE;
324 trace_xfs_scrub_incomplete(sc, __return_address);
325}
326
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700327/*
Darrick J. Wongd8526572018-01-16 18:53:08 -0800328 * rmap scrubbing -- compute the number of blocks with a given owner,
329 * at least according to the reverse mapping data.
330 */
331
332struct xfs_scrub_rmap_ownedby_info {
333 struct xfs_owner_info *oinfo;
334 xfs_filblks_t *blocks;
335};
336
337STATIC int
338xfs_scrub_count_rmap_ownedby_irec(
339 struct xfs_btree_cur *cur,
340 struct xfs_rmap_irec *rec,
341 void *priv)
342{
343 struct xfs_scrub_rmap_ownedby_info *sroi = priv;
344 bool irec_attr;
345 bool oinfo_attr;
346
347 irec_attr = rec->rm_flags & XFS_RMAP_ATTR_FORK;
348 oinfo_attr = sroi->oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK;
349
350 if (rec->rm_owner != sroi->oinfo->oi_owner)
351 return 0;
352
353 if (XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) || irec_attr == oinfo_attr)
354 (*sroi->blocks) += rec->rm_blockcount;
355
356 return 0;
357}
358
359/*
360 * Calculate the number of blocks the rmap thinks are owned by something.
361 * The caller should pass us an rmapbt cursor.
362 */
363int
364xfs_scrub_count_rmap_ownedby_ag(
365 struct xfs_scrub_context *sc,
366 struct xfs_btree_cur *cur,
367 struct xfs_owner_info *oinfo,
368 xfs_filblks_t *blocks)
369{
370 struct xfs_scrub_rmap_ownedby_info sroi;
371
372 sroi.oinfo = oinfo;
373 *blocks = 0;
374 sroi.blocks = blocks;
375
376 return xfs_rmap_query_all(cur, xfs_scrub_count_rmap_ownedby_irec,
377 &sroi);
378}
379
380/*
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700381 * AG scrubbing
382 *
383 * These helpers facilitate locking an allocation group's header
384 * buffers, setting up cursors for all btrees that are present, and
385 * cleaning everything up once we're through.
386 */
387
Darrick J. Wongab9d5dc2017-10-17 21:37:39 -0700388/* Decide if we want to return an AG header read failure. */
389static inline bool
390want_ag_read_header_failure(
391 struct xfs_scrub_context *sc,
392 unsigned int type)
393{
394 /* Return all AG header read failures when scanning btrees. */
395 if (sc->sm->sm_type != XFS_SCRUB_TYPE_AGF &&
Darrick J. Wonga12890a2017-10-17 21:37:39 -0700396 sc->sm->sm_type != XFS_SCRUB_TYPE_AGFL &&
397 sc->sm->sm_type != XFS_SCRUB_TYPE_AGI)
Darrick J. Wongab9d5dc2017-10-17 21:37:39 -0700398 return true;
399 /*
400 * If we're scanning a given type of AG header, we only want to
401 * see read failures from that specific header. We'd like the
402 * other headers to cross-check them, but this isn't required.
403 */
404 if (sc->sm->sm_type == type)
405 return true;
406 return false;
407}
408
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700409/*
410 * Grab all the headers for an AG.
411 *
412 * The headers should be released by xfs_scrub_ag_free, but as a fail
413 * safe we attach all the buffers we grab to the scrub transaction so
414 * they'll all be freed when we cancel it.
415 */
416int
417xfs_scrub_ag_read_headers(
418 struct xfs_scrub_context *sc,
419 xfs_agnumber_t agno,
420 struct xfs_buf **agi,
421 struct xfs_buf **agf,
422 struct xfs_buf **agfl)
423{
424 struct xfs_mount *mp = sc->mp;
425 int error;
426
427 error = xfs_ialloc_read_agi(mp, sc->tp, agno, agi);
Darrick J. Wonga12890a2017-10-17 21:37:39 -0700428 if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGI))
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700429 goto out;
430
431 error = xfs_alloc_read_agf(mp, sc->tp, agno, 0, agf);
Darrick J. Wongab9d5dc2017-10-17 21:37:39 -0700432 if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGF))
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700433 goto out;
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700434
435 error = xfs_alloc_read_agfl(mp, sc->tp, agno, agfl);
Darrick J. Wongab9d5dc2017-10-17 21:37:39 -0700436 if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGFL))
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700437 goto out;
Darrick J. Wong5a0f4332018-01-08 10:49:02 -0800438 error = 0;
Darrick J. Wongb6c1beb2017-10-17 21:37:38 -0700439out:
440 return error;
441}
442
443/* Release all the AG btree cursors. */
444void
445xfs_scrub_ag_btcur_free(
446 struct xfs_scrub_ag *sa)
447{
448 if (sa->refc_cur)
449 xfs_btree_del_cursor(sa->refc_cur, XFS_BTREE_ERROR);
450 if (sa->rmap_cur)
451 xfs_btree_del_cursor(sa->rmap_cur, XFS_BTREE_ERROR);
452 if (sa->fino_cur)
453 xfs_btree_del_cursor(sa->fino_cur, XFS_BTREE_ERROR);
454 if (sa->ino_cur)
455 xfs_btree_del_cursor(sa->ino_cur, XFS_BTREE_ERROR);
456 if (sa->cnt_cur)
457 xfs_btree_del_cursor(sa->cnt_cur, XFS_BTREE_ERROR);
458 if (sa->bno_cur)
459 xfs_btree_del_cursor(sa->bno_cur, XFS_BTREE_ERROR);
460
461 sa->refc_cur = NULL;
462 sa->rmap_cur = NULL;
463 sa->fino_cur = NULL;
464 sa->ino_cur = NULL;
465 sa->bno_cur = NULL;
466 sa->cnt_cur = NULL;
467}
468
469/* Initialize all the btree cursors for an AG. */
470int
471xfs_scrub_ag_btcur_init(
472 struct xfs_scrub_context *sc,
473 struct xfs_scrub_ag *sa)
474{
475 struct xfs_mount *mp = sc->mp;
476 xfs_agnumber_t agno = sa->agno;
477
478 if (sa->agf_bp) {
479 /* Set up a bnobt cursor for cross-referencing. */
480 sa->bno_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
481 agno, XFS_BTNUM_BNO);
482 if (!sa->bno_cur)
483 goto err;
484
485 /* Set up a cntbt cursor for cross-referencing. */
486 sa->cnt_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
487 agno, XFS_BTNUM_CNT);
488 if (!sa->cnt_cur)
489 goto err;
490 }
491
492 /* Set up a inobt cursor for cross-referencing. */
493 if (sa->agi_bp) {
494 sa->ino_cur = xfs_inobt_init_cursor(mp, sc->tp, sa->agi_bp,
495 agno, XFS_BTNUM_INO);
496 if (!sa->ino_cur)
497 goto err;
498 }
499
500 /* Set up a finobt cursor for cross-referencing. */
501 if (sa->agi_bp && xfs_sb_version_hasfinobt(&mp->m_sb)) {
502 sa->fino_cur = xfs_inobt_init_cursor(mp, sc->tp, sa->agi_bp,
503 agno, XFS_BTNUM_FINO);
504 if (!sa->fino_cur)
505 goto err;
506 }
507
508 /* Set up a rmapbt cursor for cross-referencing. */
509 if (sa->agf_bp && xfs_sb_version_hasrmapbt(&mp->m_sb)) {
510 sa->rmap_cur = xfs_rmapbt_init_cursor(mp, sc->tp, sa->agf_bp,
511 agno);
512 if (!sa->rmap_cur)
513 goto err;
514 }
515
516 /* Set up a refcountbt cursor for cross-referencing. */
517 if (sa->agf_bp && xfs_sb_version_hasreflink(&mp->m_sb)) {
518 sa->refc_cur = xfs_refcountbt_init_cursor(mp, sc->tp,
519 sa->agf_bp, agno, NULL);
520 if (!sa->refc_cur)
521 goto err;
522 }
523
524 return 0;
525err:
526 return -ENOMEM;
527}
528
529/* Release the AG header context and btree cursors. */
530void
531xfs_scrub_ag_free(
532 struct xfs_scrub_context *sc,
533 struct xfs_scrub_ag *sa)
534{
535 xfs_scrub_ag_btcur_free(sa);
536 if (sa->agfl_bp) {
537 xfs_trans_brelse(sc->tp, sa->agfl_bp);
538 sa->agfl_bp = NULL;
539 }
540 if (sa->agf_bp) {
541 xfs_trans_brelse(sc->tp, sa->agf_bp);
542 sa->agf_bp = NULL;
543 }
544 if (sa->agi_bp) {
545 xfs_trans_brelse(sc->tp, sa->agi_bp);
546 sa->agi_bp = NULL;
547 }
548 sa->agno = NULLAGNUMBER;
549}
550
551/*
552 * For scrub, grab the AGI and the AGF headers, in that order. Locking
553 * order requires us to get the AGI before the AGF. We use the
554 * transaction to avoid deadlocking on crosslinked metadata buffers;
555 * either the caller passes one in (bmap scrub) or we have to create a
556 * transaction ourselves.
557 */
558int
559xfs_scrub_ag_init(
560 struct xfs_scrub_context *sc,
561 xfs_agnumber_t agno,
562 struct xfs_scrub_ag *sa)
563{
564 int error;
565
566 sa->agno = agno;
567 error = xfs_scrub_ag_read_headers(sc, agno, &sa->agi_bp,
568 &sa->agf_bp, &sa->agfl_bp);
569 if (error)
570 return error;
571
572 return xfs_scrub_ag_btcur_init(sc, sa);
573}
574
Darrick J. Wongdcb660f2017-10-17 21:37:36 -0700575/* Per-scrubber setup functions */
576
577/* Set us up with a transaction and an empty context. */
578int
579xfs_scrub_setup_fs(
580 struct xfs_scrub_context *sc,
581 struct xfs_inode *ip)
582{
583 return xfs_scrub_trans_alloc(sc->sm, sc->mp, &sc->tp);
584}
Darrick J. Wongefa7a992017-10-17 21:37:40 -0700585
586/* Set us up with AG headers and btree cursors. */
587int
588xfs_scrub_setup_ag_btree(
589 struct xfs_scrub_context *sc,
590 struct xfs_inode *ip,
591 bool force_log)
592{
Darrick J. Wong3daa6642017-10-17 21:37:40 -0700593 struct xfs_mount *mp = sc->mp;
Darrick J. Wongefa7a992017-10-17 21:37:40 -0700594 int error;
595
Darrick J. Wong3daa6642017-10-17 21:37:40 -0700596 /*
597 * If the caller asks us to checkpont the log, do so. This
598 * expensive operation should be performed infrequently and only
599 * as a last resort. Any caller that sets force_log should
600 * document why they need to do so.
601 */
602 if (force_log) {
603 error = xfs_scrub_checkpoint_log(mp);
604 if (error)
605 return error;
606 }
607
Eric Sandeen8e630832018-01-08 10:41:34 -0800608 error = xfs_scrub_setup_fs(sc, ip);
Darrick J. Wongefa7a992017-10-17 21:37:40 -0700609 if (error)
610 return error;
611
612 return xfs_scrub_ag_init(sc, sc->sm->sm_agno, &sc->sa);
613}
Darrick J. Wong3daa6642017-10-17 21:37:40 -0700614
615/* Push everything out of the log onto disk. */
616int
617xfs_scrub_checkpoint_log(
618 struct xfs_mount *mp)
619{
620 int error;
621
622 error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
623 if (error)
624 return error;
625 xfs_ail_push_all_sync(mp->m_ail);
626 return 0;
627}
Darrick J. Wong80e4e122017-10-17 21:37:42 -0700628
629/*
630 * Given an inode and the scrub control structure, grab either the
631 * inode referenced in the control structure or the inode passed in.
632 * The inode is not locked.
633 */
634int
635xfs_scrub_get_inode(
636 struct xfs_scrub_context *sc,
637 struct xfs_inode *ip_in)
638{
Darrick J. Wongd658e722018-01-08 10:49:04 -0800639 struct xfs_imap imap;
Darrick J. Wong80e4e122017-10-17 21:37:42 -0700640 struct xfs_mount *mp = sc->mp;
641 struct xfs_inode *ip = NULL;
642 int error;
643
Darrick J. Wong80e4e122017-10-17 21:37:42 -0700644 /* We want to scan the inode we already had opened. */
645 if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
646 sc->ip = ip_in;
647 return 0;
648 }
649
650 /* Look up the inode, see if the generation number matches. */
651 if (xfs_internal_inum(mp, sc->sm->sm_ino))
652 return -ENOENT;
653 error = xfs_iget(mp, NULL, sc->sm->sm_ino,
654 XFS_IGET_UNTRUSTED | XFS_IGET_DONTCACHE, 0, &ip);
Darrick J. Wongd658e722018-01-08 10:49:04 -0800655 switch (error) {
656 case -ENOENT:
657 /* Inode doesn't exist, just bail out. */
658 return error;
659 case 0:
660 /* Got an inode, continue. */
661 break;
662 case -EINVAL:
663 /*
664 * -EINVAL with IGET_UNTRUSTED could mean one of several
665 * things: userspace gave us an inode number that doesn't
666 * correspond to fs space, or doesn't have an inobt entry;
667 * or it could simply mean that the inode buffer failed the
668 * read verifiers.
669 *
670 * Try just the inode mapping lookup -- if it succeeds, then
671 * the inode buffer verifier failed and something needs fixing.
672 * Otherwise, we really couldn't find it so tell userspace
673 * that it no longer exists.
674 */
675 error = xfs_imap(sc->mp, sc->tp, sc->sm->sm_ino, &imap,
676 XFS_IGET_UNTRUSTED | XFS_IGET_DONTCACHE);
677 if (error)
678 return -ENOENT;
679 error = -EFSCORRUPTED;
680 /* fall through */
681 default:
Darrick J. Wong80e4e122017-10-17 21:37:42 -0700682 trace_xfs_scrub_op_error(sc,
683 XFS_INO_TO_AGNO(mp, sc->sm->sm_ino),
684 XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
685 error, __return_address);
686 return error;
687 }
688 if (VFS_I(ip)->i_generation != sc->sm->sm_gen) {
689 iput(VFS_I(ip));
690 return -ENOENT;
691 }
692
693 sc->ip = ip;
694 return 0;
695}
Darrick J. Wonga5c46e52017-10-17 21:37:44 -0700696
697/* Set us up to scrub a file's contents. */
698int
699xfs_scrub_setup_inode_contents(
700 struct xfs_scrub_context *sc,
701 struct xfs_inode *ip,
702 unsigned int resblks)
703{
704 struct xfs_mount *mp = sc->mp;
705 int error;
706
707 error = xfs_scrub_get_inode(sc, ip);
708 if (error)
709 return error;
710
711 /* Got the inode, lock it and we're ready to go. */
712 sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
713 xfs_ilock(sc->ip, sc->ilock_flags);
714 error = xfs_scrub_trans_alloc(sc->sm, mp, &sc->tp);
715 if (error)
716 goto out;
717 sc->ilock_flags |= XFS_ILOCK_EXCL;
718 xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
719
720out:
721 /* scrub teardown will unlock and release the inode for us */
722 return error;
723}
Darrick J. Wong64b12562018-01-16 18:52:14 -0800724
725/*
726 * Predicate that decides if we need to evaluate the cross-reference check.
727 * If there was an error accessing the cross-reference btree, just delete
728 * the cursor and skip the check.
729 */
730bool
731xfs_scrub_should_check_xref(
732 struct xfs_scrub_context *sc,
733 int *error,
734 struct xfs_btree_cur **curpp)
735{
736 if (*error == 0)
737 return true;
738
739 if (curpp) {
740 /* If we've already given up on xref, just bail out. */
741 if (!*curpp)
742 return false;
743
744 /* xref error, delete cursor and bail out. */
745 xfs_btree_del_cursor(*curpp, XFS_BTREE_ERROR);
746 *curpp = NULL;
747 }
748
749 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
750 trace_xfs_scrub_xref_error(sc, *error, __return_address);
751
752 /*
753 * Errors encountered during cross-referencing with another
754 * data structure should not cause this scrubber to abort.
755 */
756 *error = 0;
757 return false;
758}
Darrick J. Wongcf1b0b82018-01-16 18:53:11 -0800759
760/* Run the structure verifiers on in-memory buffers to detect bad memory. */
761void
762xfs_scrub_buffer_recheck(
763 struct xfs_scrub_context *sc,
764 struct xfs_buf *bp)
765{
766 xfs_failaddr_t fa;
767
768 if (bp->b_ops == NULL) {
769 xfs_scrub_block_set_corrupt(sc, bp);
770 return;
771 }
772 if (bp->b_ops->verify_struct == NULL) {
773 xfs_scrub_set_incomplete(sc);
774 return;
775 }
776 fa = bp->b_ops->verify_struct(bp);
777 if (!fa)
778 return;
779 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
780 trace_xfs_scrub_block_error(sc, bp->b_bn, fa);
781}