Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.5.6.
  1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * Copyright (C) 2017 Oracle.  All Rights Reserved.
  4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
  5 */
  6#include "xfs.h"
  7#include "xfs_fs.h"
  8#include "xfs_shared.h"
  9#include "xfs_format.h"
 10#include "xfs_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_attr.h"
 27#include "xfs_reflink.h"
 28#include "xfs_ag.h"
 29#include "scrub/scrub.h"
 30#include "scrub/common.h"
 31#include "scrub/trace.h"
 32#include "scrub/repair.h"
 33#include "scrub/health.h"
 34
 35/* Common code for the metadata scrubbers. */
 36
 37/*
 38 * Handling operational errors.
 39 *
 40 * The *_process_error() family of functions are used to process error return
 41 * codes from functions called as part of a scrub operation.
 42 *
 43 * If there's no error, we return true to tell the caller that it's ok
 44 * to move on to the next check in its list.
 45 *
 46 * For non-verifier errors (e.g. ENOMEM) we return false to tell the
 47 * caller that something bad happened, and we preserve *error so that
 48 * the caller can return the *error up the stack to userspace.
 49 *
 50 * Verifier errors (EFSBADCRC/EFSCORRUPTED) are recorded by setting
 51 * OFLAG_CORRUPT in sm_flags and the *error is cleared.  In other words,
 52 * we track verifier errors (and failed scrub checks) via OFLAG_CORRUPT,
 53 * not via return codes.  We return false to tell the caller that
 54 * something bad happened.  Since the error has been cleared, the caller
 55 * will (presumably) return that zero and scrubbing will move on to
 56 * whatever's next.
 57 *
 58 * ftrace can be used to record the precise metadata location and the
 59 * approximate code location of the failed operation.
 60 */
 61
 62/* Check for operational errors. */
 63static bool
 64__xchk_process_error(
 65	struct xfs_scrub	*sc,
 66	xfs_agnumber_t		agno,
 67	xfs_agblock_t		bno,
 68	int			*error,
 69	__u32			errflag,
 70	void			*ret_ip)
 71{
 72	switch (*error) {
 73	case 0:
 74		return true;
 75	case -EDEADLOCK:
 76		/* Used to restart an op with deadlock avoidance. */
 77		trace_xchk_deadlock_retry(
 78				sc->ip ? sc->ip : XFS_I(file_inode(sc->file)),
 79				sc->sm, *error);
 80		break;
 81	case -EFSBADCRC:
 82	case -EFSCORRUPTED:
 83		/* Note the badness but don't abort. */
 84		sc->sm->sm_flags |= errflag;
 85		*error = 0;
 86		fallthrough;
 87	default:
 88		trace_xchk_op_error(sc, agno, bno, *error,
 89				ret_ip);
 90		break;
 91	}
 92	return false;
 93}
 94
 95bool
 96xchk_process_error(
 97	struct xfs_scrub	*sc,
 98	xfs_agnumber_t		agno,
 99	xfs_agblock_t		bno,
100	int			*error)
101{
102	return __xchk_process_error(sc, agno, bno, error,
103			XFS_SCRUB_OFLAG_CORRUPT, __return_address);
104}
105
106bool
107xchk_xref_process_error(
108	struct xfs_scrub	*sc,
109	xfs_agnumber_t		agno,
110	xfs_agblock_t		bno,
111	int			*error)
112{
113	return __xchk_process_error(sc, agno, bno, error,
114			XFS_SCRUB_OFLAG_XFAIL, __return_address);
115}
116
117/* Check for operational errors for a file offset. */
118static bool
119__xchk_fblock_process_error(
120	struct xfs_scrub	*sc,
121	int			whichfork,
122	xfs_fileoff_t		offset,
123	int			*error,
124	__u32			errflag,
125	void			*ret_ip)
126{
127	switch (*error) {
128	case 0:
129		return true;
130	case -EDEADLOCK:
131		/* Used to restart an op with deadlock avoidance. */
132		trace_xchk_deadlock_retry(sc->ip, sc->sm, *error);
133		break;
134	case -EFSBADCRC:
135	case -EFSCORRUPTED:
136		/* Note the badness but don't abort. */
137		sc->sm->sm_flags |= errflag;
138		*error = 0;
139		fallthrough;
140	default:
141		trace_xchk_file_op_error(sc, whichfork, offset, *error,
142				ret_ip);
143		break;
144	}
145	return false;
146}
147
148bool
149xchk_fblock_process_error(
150	struct xfs_scrub	*sc,
151	int			whichfork,
152	xfs_fileoff_t		offset,
153	int			*error)
154{
155	return __xchk_fblock_process_error(sc, whichfork, offset, error,
156			XFS_SCRUB_OFLAG_CORRUPT, __return_address);
157}
158
159bool
160xchk_fblock_xref_process_error(
161	struct xfs_scrub	*sc,
162	int			whichfork,
163	xfs_fileoff_t		offset,
164	int			*error)
165{
166	return __xchk_fblock_process_error(sc, whichfork, offset, error,
167			XFS_SCRUB_OFLAG_XFAIL, __return_address);
168}
169
170/*
171 * Handling scrub corruption/optimization/warning checks.
172 *
173 * The *_set_{corrupt,preen,warning}() family of functions are used to
174 * record the presence of metadata that is incorrect (corrupt), could be
175 * optimized somehow (preen), or should be flagged for administrative
176 * review but is not incorrect (warn).
177 *
178 * ftrace can be used to record the precise metadata location and
179 * approximate code location of the failed check.
180 */
181
182/* Record a block which could be optimized. */
183void
184xchk_block_set_preen(
185	struct xfs_scrub	*sc,
186	struct xfs_buf		*bp)
187{
188	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
189	trace_xchk_block_preen(sc, bp->b_bn, __return_address);
190}
191
192/*
193 * Record an inode which could be optimized.  The trace data will
194 * include the block given by bp if bp is given; otherwise it will use
195 * the block location of the inode record itself.
196 */
197void
198xchk_ino_set_preen(
199	struct xfs_scrub	*sc,
200	xfs_ino_t		ino)
201{
202	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
203	trace_xchk_ino_preen(sc, ino, __return_address);
204}
205
206/* Record something being wrong with the filesystem primary superblock. */
207void
208xchk_set_corrupt(
209	struct xfs_scrub	*sc)
210{
211	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
212	trace_xchk_fs_error(sc, 0, __return_address);
213}
214
215/* Record a corrupt block. */
216void
217xchk_block_set_corrupt(
218	struct xfs_scrub	*sc,
219	struct xfs_buf		*bp)
220{
221	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
222	trace_xchk_block_error(sc, bp->b_bn, __return_address);
223}
224
225/* Record a corruption while cross-referencing. */
226void
227xchk_block_xref_set_corrupt(
228	struct xfs_scrub	*sc,
229	struct xfs_buf		*bp)
230{
231	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
232	trace_xchk_block_error(sc, bp->b_bn, __return_address);
233}
234
235/*
236 * Record a corrupt inode.  The trace data will include the block given
237 * by bp if bp is given; otherwise it will use the block location of the
238 * inode record itself.
239 */
240void
241xchk_ino_set_corrupt(
242	struct xfs_scrub	*sc,
243	xfs_ino_t		ino)
244{
245	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
246	trace_xchk_ino_error(sc, ino, __return_address);
247}
248
249/* Record a corruption while cross-referencing with an inode. */
250void
251xchk_ino_xref_set_corrupt(
252	struct xfs_scrub	*sc,
253	xfs_ino_t		ino)
254{
255	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
256	trace_xchk_ino_error(sc, ino, __return_address);
257}
258
259/* Record corruption in a block indexed by a file fork. */
260void
261xchk_fblock_set_corrupt(
262	struct xfs_scrub	*sc,
263	int			whichfork,
264	xfs_fileoff_t		offset)
265{
266	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
267	trace_xchk_fblock_error(sc, whichfork, offset, __return_address);
268}
269
270/* Record a corruption while cross-referencing a fork block. */
271void
272xchk_fblock_xref_set_corrupt(
273	struct xfs_scrub	*sc,
274	int			whichfork,
275	xfs_fileoff_t		offset)
276{
277	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
278	trace_xchk_fblock_error(sc, whichfork, offset, __return_address);
279}
280
281/*
282 * Warn about inodes that need administrative review but is not
283 * incorrect.
284 */
285void
286xchk_ino_set_warning(
287	struct xfs_scrub	*sc,
288	xfs_ino_t		ino)
289{
290	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
291	trace_xchk_ino_warning(sc, ino, __return_address);
292}
293
294/* Warn about a block indexed by a file fork that needs review. */
295void
296xchk_fblock_set_warning(
297	struct xfs_scrub	*sc,
298	int			whichfork,
299	xfs_fileoff_t		offset)
300{
301	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
302	trace_xchk_fblock_warning(sc, whichfork, offset, __return_address);
303}
304
305/* Signal an incomplete scrub. */
306void
307xchk_set_incomplete(
308	struct xfs_scrub	*sc)
309{
310	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_INCOMPLETE;
311	trace_xchk_incomplete(sc, __return_address);
312}
313
314/*
315 * rmap scrubbing -- compute the number of blocks with a given owner,
316 * at least according to the reverse mapping data.
317 */
318
319struct xchk_rmap_ownedby_info {
320	const struct xfs_owner_info	*oinfo;
321	xfs_filblks_t			*blocks;
322};
323
324STATIC int
325xchk_count_rmap_ownedby_irec(
326	struct xfs_btree_cur		*cur,
327	struct xfs_rmap_irec		*rec,
328	void				*priv)
329{
330	struct xchk_rmap_ownedby_info	*sroi = priv;
331	bool				irec_attr;
332	bool				oinfo_attr;
333
334	irec_attr = rec->rm_flags & XFS_RMAP_ATTR_FORK;
335	oinfo_attr = sroi->oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK;
336
337	if (rec->rm_owner != sroi->oinfo->oi_owner)
338		return 0;
339
340	if (XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) || irec_attr == oinfo_attr)
341		(*sroi->blocks) += rec->rm_blockcount;
342
343	return 0;
344}
345
346/*
347 * Calculate the number of blocks the rmap thinks are owned by something.
348 * The caller should pass us an rmapbt cursor.
349 */
350int
351xchk_count_rmap_ownedby_ag(
352	struct xfs_scrub		*sc,
353	struct xfs_btree_cur		*cur,
354	const struct xfs_owner_info	*oinfo,
355	xfs_filblks_t			*blocks)
356{
357	struct xchk_rmap_ownedby_info	sroi = {
358		.oinfo			= oinfo,
359		.blocks			= blocks,
360	};
361
362	*blocks = 0;
363	return xfs_rmap_query_all(cur, xchk_count_rmap_ownedby_irec,
364			&sroi);
365}
366
367/*
368 * AG scrubbing
369 *
370 * These helpers facilitate locking an allocation group's header
371 * buffers, setting up cursors for all btrees that are present, and
372 * cleaning everything up once we're through.
373 */
374
375/* Decide if we want to return an AG header read failure. */
376static inline bool
377want_ag_read_header_failure(
378	struct xfs_scrub	*sc,
379	unsigned int		type)
380{
381	/* Return all AG header read failures when scanning btrees. */
382	if (sc->sm->sm_type != XFS_SCRUB_TYPE_AGF &&
383	    sc->sm->sm_type != XFS_SCRUB_TYPE_AGFL &&
384	    sc->sm->sm_type != XFS_SCRUB_TYPE_AGI)
385		return true;
386	/*
387	 * If we're scanning a given type of AG header, we only want to
388	 * see read failures from that specific header.  We'd like the
389	 * other headers to cross-check them, but this isn't required.
390	 */
391	if (sc->sm->sm_type == type)
392		return true;
393	return false;
394}
395
396/*
397 * Grab all the headers for an AG.
398 *
399 * The headers should be released by xchk_ag_free, but as a fail
400 * safe we attach all the buffers we grab to the scrub transaction so
401 * they'll all be freed when we cancel it.
402 */
403int
404xchk_ag_read_headers(
405	struct xfs_scrub	*sc,
406	xfs_agnumber_t		agno,
407	struct xchk_ag		*sa)
408{
409	struct xfs_mount	*mp = sc->mp;
410	int			error;
411
412	sa->agno = agno;
413
414	error = xfs_ialloc_read_agi(mp, sc->tp, agno, &sa->agi_bp);
415	if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGI))
416		goto out;
417
418	error = xfs_alloc_read_agf(mp, sc->tp, agno, 0, &sa->agf_bp);
419	if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGF))
420		goto out;
421
422	error = xfs_alloc_read_agfl(mp, sc->tp, agno, &sa->agfl_bp);
423	if (error && want_ag_read_header_failure(sc, XFS_SCRUB_TYPE_AGFL))
424		goto out;
425	error = 0;
426out:
427	return error;
428}
429
430/* Release all the AG btree cursors. */
431void
432xchk_ag_btcur_free(
433	struct xchk_ag		*sa)
434{
435	if (sa->refc_cur)
436		xfs_btree_del_cursor(sa->refc_cur, XFS_BTREE_ERROR);
437	if (sa->rmap_cur)
438		xfs_btree_del_cursor(sa->rmap_cur, XFS_BTREE_ERROR);
439	if (sa->fino_cur)
440		xfs_btree_del_cursor(sa->fino_cur, XFS_BTREE_ERROR);
441	if (sa->ino_cur)
442		xfs_btree_del_cursor(sa->ino_cur, XFS_BTREE_ERROR);
443	if (sa->cnt_cur)
444		xfs_btree_del_cursor(sa->cnt_cur, XFS_BTREE_ERROR);
445	if (sa->bno_cur)
446		xfs_btree_del_cursor(sa->bno_cur, XFS_BTREE_ERROR);
447
448	sa->refc_cur = NULL;
449	sa->rmap_cur = NULL;
450	sa->fino_cur = NULL;
451	sa->ino_cur = NULL;
452	sa->bno_cur = NULL;
453	sa->cnt_cur = NULL;
454}
455
456/* Initialize all the btree cursors for an AG. */
457void
458xchk_ag_btcur_init(
459	struct xfs_scrub	*sc,
460	struct xchk_ag		*sa)
461{
462	struct xfs_mount	*mp = sc->mp;
463
464	xchk_perag_get(sc->mp, sa);
465	if (sa->agf_bp &&
466	    xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_BNO)) {
467		/* Set up a bnobt cursor for cross-referencing. */
468		sa->bno_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
469				sa->pag, XFS_BTNUM_BNO);
470	}
471
472	if (sa->agf_bp &&
473	    xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_CNT)) {
474		/* Set up a cntbt cursor for cross-referencing. */
475		sa->cnt_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
476				sa->pag, XFS_BTNUM_CNT);
477	}
478
479	/* Set up a inobt cursor for cross-referencing. */
480	if (sa->agi_bp &&
481	    xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_INO)) {
482		sa->ino_cur = xfs_inobt_init_cursor(mp, sc->tp, sa->agi_bp,
483				sa->pag, XFS_BTNUM_INO);
484	}
485
486	/* Set up a finobt cursor for cross-referencing. */
487	if (sa->agi_bp && xfs_sb_version_hasfinobt(&mp->m_sb) &&
488	    xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_FINO)) {
489		sa->fino_cur = xfs_inobt_init_cursor(mp, sc->tp, sa->agi_bp,
490				sa->pag, XFS_BTNUM_FINO);
491	}
492
493	/* Set up a rmapbt cursor for cross-referencing. */
494	if (sa->agf_bp && xfs_sb_version_hasrmapbt(&mp->m_sb) &&
495	    xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_RMAP)) {
496		sa->rmap_cur = xfs_rmapbt_init_cursor(mp, sc->tp, sa->agf_bp,
497				sa->pag);
498	}
499
500	/* Set up a refcountbt cursor for cross-referencing. */
501	if (sa->agf_bp && xfs_sb_version_hasreflink(&mp->m_sb) &&
502	    xchk_ag_btree_healthy_enough(sc, sa->pag, XFS_BTNUM_REFC)) {
503		sa->refc_cur = xfs_refcountbt_init_cursor(mp, sc->tp,
504				sa->agf_bp, sa->pag);
505	}
506}
507
508/* Release the AG header context and btree cursors. */
509void
510xchk_ag_free(
511	struct xfs_scrub	*sc,
512	struct xchk_ag		*sa)
513{
514	xchk_ag_btcur_free(sa);
515	if (sa->agfl_bp) {
516		xfs_trans_brelse(sc->tp, sa->agfl_bp);
517		sa->agfl_bp = NULL;
518	}
519	if (sa->agf_bp) {
520		xfs_trans_brelse(sc->tp, sa->agf_bp);
521		sa->agf_bp = NULL;
522	}
523	if (sa->agi_bp) {
524		xfs_trans_brelse(sc->tp, sa->agi_bp);
525		sa->agi_bp = NULL;
526	}
527	if (sa->pag) {
528		xfs_perag_put(sa->pag);
529		sa->pag = NULL;
530	}
531	sa->agno = NULLAGNUMBER;
532}
533
534/*
535 * For scrub, grab the AGI and the AGF headers, in that order.  Locking
536 * order requires us to get the AGI before the AGF.  We use the
537 * transaction to avoid deadlocking on crosslinked metadata buffers;
538 * either the caller passes one in (bmap scrub) or we have to create a
539 * transaction ourselves.
540 */
541int
542xchk_ag_init(
543	struct xfs_scrub	*sc,
544	xfs_agnumber_t		agno,
545	struct xchk_ag		*sa)
546{
547	int			error;
548
549	error = xchk_ag_read_headers(sc, agno, sa);
550	if (error)
551		return error;
552
553	xchk_ag_btcur_init(sc, sa);
554	return 0;
555}
556
557/*
558 * Grab the per-ag structure if we haven't already gotten it.  Teardown of the
559 * xchk_ag will release it for us.
560 */
561void
562xchk_perag_get(
563	struct xfs_mount	*mp,
564	struct xchk_ag		*sa)
565{
566	if (!sa->pag)
567		sa->pag = xfs_perag_get(mp, sa->agno);
568}
569
570/* Per-scrubber setup functions */
571
572/*
573 * Grab an empty transaction so that we can re-grab locked buffers if
574 * one of our btrees turns out to be cyclic.
575 *
576 * If we're going to repair something, we need to ask for the largest possible
577 * log reservation so that we can handle the worst case scenario for metadata
578 * updates while rebuilding a metadata item.  We also need to reserve as many
579 * blocks in the head transaction as we think we're going to need to rebuild
580 * the metadata object.
581 */
582int
583xchk_trans_alloc(
584	struct xfs_scrub	*sc,
585	uint			resblks)
586{
587	if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
588		return xfs_trans_alloc(sc->mp, &M_RES(sc->mp)->tr_itruncate,
589				resblks, 0, 0, &sc->tp);
590
591	return xfs_trans_alloc_empty(sc->mp, &sc->tp);
592}
593
594/* Set us up with a transaction and an empty context. */
595int
596xchk_setup_fs(
597	struct xfs_scrub	*sc)
598{
599	uint			resblks;
600
601	resblks = xrep_calc_ag_resblks(sc);
602	return xchk_trans_alloc(sc, resblks);
603}
604
605/* Set us up with AG headers and btree cursors. */
606int
607xchk_setup_ag_btree(
608	struct xfs_scrub	*sc,
609	bool			force_log)
610{
611	struct xfs_mount	*mp = sc->mp;
612	int			error;
613
614	/*
615	 * If the caller asks us to checkpont the log, do so.  This
616	 * expensive operation should be performed infrequently and only
617	 * as a last resort.  Any caller that sets force_log should
618	 * document why they need to do so.
619	 */
620	if (force_log) {
621		error = xchk_checkpoint_log(mp);
622		if (error)
623			return error;
624	}
625
626	error = xchk_setup_fs(sc);
627	if (error)
628		return error;
629
630	return xchk_ag_init(sc, sc->sm->sm_agno, &sc->sa);
631}
632
633/* Push everything out of the log onto disk. */
634int
635xchk_checkpoint_log(
636	struct xfs_mount	*mp)
637{
638	int			error;
639
640	error = xfs_log_force(mp, XFS_LOG_SYNC);
641	if (error)
642		return error;
643	xfs_ail_push_all_sync(mp->m_ail);
644	return 0;
645}
646
647/*
648 * Given an inode and the scrub control structure, grab either the
649 * inode referenced in the control structure or the inode passed in.
650 * The inode is not locked.
651 */
652int
653xchk_get_inode(
654	struct xfs_scrub	*sc)
655{
656	struct xfs_imap		imap;
657	struct xfs_mount	*mp = sc->mp;
658	struct xfs_inode	*ip_in = XFS_I(file_inode(sc->file));
659	struct xfs_inode	*ip = NULL;
660	int			error;
661
662	/* We want to scan the inode we already had opened. */
663	if (sc->sm->sm_ino == 0 || sc->sm->sm_ino == ip_in->i_ino) {
664		sc->ip = ip_in;
665		return 0;
666	}
667
668	/* Look up the inode, see if the generation number matches. */
669	if (xfs_internal_inum(mp, sc->sm->sm_ino))
670		return -ENOENT;
671	error = xfs_iget(mp, NULL, sc->sm->sm_ino,
672			XFS_IGET_UNTRUSTED | XFS_IGET_DONTCACHE, 0, &ip);
673	switch (error) {
674	case -ENOENT:
675		/* Inode doesn't exist, just bail out. */
676		return error;
677	case 0:
678		/* Got an inode, continue. */
679		break;
680	case -EINVAL:
681		/*
682		 * -EINVAL with IGET_UNTRUSTED could mean one of several
683		 * things: userspace gave us an inode number that doesn't
684		 * correspond to fs space, or doesn't have an inobt entry;
685		 * or it could simply mean that the inode buffer failed the
686		 * read verifiers.
687		 *
688		 * Try just the inode mapping lookup -- if it succeeds, then
689		 * the inode buffer verifier failed and something needs fixing.
690		 * Otherwise, we really couldn't find it so tell userspace
691		 * that it no longer exists.
692		 */
693		error = xfs_imap(sc->mp, sc->tp, sc->sm->sm_ino, &imap,
694				XFS_IGET_UNTRUSTED | XFS_IGET_DONTCACHE);
695		if (error)
696			return -ENOENT;
697		error = -EFSCORRUPTED;
698		fallthrough;
699	default:
700		trace_xchk_op_error(sc,
701				XFS_INO_TO_AGNO(mp, sc->sm->sm_ino),
702				XFS_INO_TO_AGBNO(mp, sc->sm->sm_ino),
703				error, __return_address);
704		return error;
705	}
706	if (VFS_I(ip)->i_generation != sc->sm->sm_gen) {
707		xfs_irele(ip);
708		return -ENOENT;
709	}
710
711	sc->ip = ip;
712	return 0;
713}
714
715/* Set us up to scrub a file's contents. */
716int
717xchk_setup_inode_contents(
718	struct xfs_scrub	*sc,
719	unsigned int		resblks)
720{
721	int			error;
722
723	error = xchk_get_inode(sc);
724	if (error)
725		return error;
726
727	/* Got the inode, lock it and we're ready to go. */
728	sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
729	xfs_ilock(sc->ip, sc->ilock_flags);
730	error = xchk_trans_alloc(sc, resblks);
731	if (error)
732		goto out;
733	sc->ilock_flags |= XFS_ILOCK_EXCL;
734	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
735
736out:
737	/* scrub teardown will unlock and release the inode for us */
738	return error;
739}
740
741/*
742 * Predicate that decides if we need to evaluate the cross-reference check.
743 * If there was an error accessing the cross-reference btree, just delete
744 * the cursor and skip the check.
745 */
746bool
747xchk_should_check_xref(
748	struct xfs_scrub	*sc,
749	int			*error,
750	struct xfs_btree_cur	**curpp)
751{
752	/* No point in xref if we already know we're corrupt. */
753	if (xchk_skip_xref(sc->sm))
754		return false;
755
756	if (*error == 0)
757		return true;
758
759	if (curpp) {
760		/* If we've already given up on xref, just bail out. */
761		if (!*curpp)
762			return false;
763
764		/* xref error, delete cursor and bail out. */
765		xfs_btree_del_cursor(*curpp, XFS_BTREE_ERROR);
766		*curpp = NULL;
767	}
768
769	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
770	trace_xchk_xref_error(sc, *error, __return_address);
771
772	/*
773	 * Errors encountered during cross-referencing with another
774	 * data structure should not cause this scrubber to abort.
775	 */
776	*error = 0;
777	return false;
778}
779
780/* Run the structure verifiers on in-memory buffers to detect bad memory. */
781void
782xchk_buffer_recheck(
783	struct xfs_scrub	*sc,
784	struct xfs_buf		*bp)
785{
786	xfs_failaddr_t		fa;
787
788	if (bp->b_ops == NULL) {
789		xchk_block_set_corrupt(sc, bp);
790		return;
791	}
792	if (bp->b_ops->verify_struct == NULL) {
793		xchk_set_incomplete(sc);
794		return;
795	}
796	fa = bp->b_ops->verify_struct(bp);
797	if (!fa)
798		return;
799	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
800	trace_xchk_block_error(sc, bp->b_bn, fa);
801}
802
803/*
804 * Scrub the attr/data forks of a metadata inode.  The metadata inode must be
805 * pointed to by sc->ip and the ILOCK must be held.
806 */
807int
808xchk_metadata_inode_forks(
809	struct xfs_scrub	*sc)
810{
811	__u32			smtype;
812	bool			shared;
813	int			error;
814
815	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
816		return 0;
817
818	/* Metadata inodes don't live on the rt device. */
819	if (sc->ip->i_diflags & XFS_DIFLAG_REALTIME) {
820		xchk_ino_set_corrupt(sc, sc->ip->i_ino);
821		return 0;
822	}
823
824	/* They should never participate in reflink. */
825	if (xfs_is_reflink_inode(sc->ip)) {
826		xchk_ino_set_corrupt(sc, sc->ip->i_ino);
827		return 0;
828	}
829
830	/* They also should never have extended attributes. */
831	if (xfs_inode_hasattr(sc->ip)) {
832		xchk_ino_set_corrupt(sc, sc->ip->i_ino);
833		return 0;
834	}
835
836	/* Invoke the data fork scrubber. */
837	smtype = sc->sm->sm_type;
838	sc->sm->sm_type = XFS_SCRUB_TYPE_BMBTD;
839	error = xchk_bmap_data(sc);
840	sc->sm->sm_type = smtype;
841	if (error || (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
842		return error;
843
844	/* Look for incorrect shared blocks. */
845	if (xfs_sb_version_hasreflink(&sc->mp->m_sb)) {
846		error = xfs_reflink_inode_has_shared_extents(sc->tp, sc->ip,
847				&shared);
848		if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
849				&error))
850			return error;
851		if (shared)
852			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
853	}
854
855	return error;
856}
857
858/*
859 * Try to lock an inode in violation of the usual locking order rules.  For
860 * example, trying to get the IOLOCK while in transaction context, or just
861 * plain breaking AG-order or inode-order inode locking rules.  Either way,
862 * the only way to avoid an ABBA deadlock is to use trylock and back off if
863 * we can't.
864 */
865int
866xchk_ilock_inverted(
867	struct xfs_inode	*ip,
868	uint			lock_mode)
869{
870	int			i;
871
872	for (i = 0; i < 20; i++) {
873		if (xfs_ilock_nowait(ip, lock_mode))
874			return 0;
875		delay(1);
876	}
877	return -EDEADLOCK;
878}
879
880/* Pause background reaping of resources. */
881void
882xchk_stop_reaping(
883	struct xfs_scrub	*sc)
884{
885	sc->flags |= XCHK_REAPING_DISABLED;
886	xfs_blockgc_stop(sc->mp);
887}
888
889/* Restart background reaping of resources. */
890void
891xchk_start_reaping(
892	struct xfs_scrub	*sc)
893{
894	xfs_blockgc_start(sc->mp);
895	sc->flags &= ~XCHK_REAPING_DISABLED;
896}