Linux Audio

Check our new training course

Loading...
v5.4
  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_bit.h"
 14#include "xfs_log_format.h"
 15#include "xfs_trans.h"
 16#include "xfs_inode.h"
 17#include "xfs_alloc.h"
 18#include "xfs_bmap.h"
 19#include "xfs_bmap_btree.h"
 20#include "xfs_rmap.h"
 21#include "xfs_rmap_btree.h"
 22#include "scrub/scrub.h"
 23#include "scrub/common.h"
 24#include "scrub/btree.h"
 
 25
 26/* Set us up with an inode's bmap. */
 27int
 28xchk_setup_inode_bmap(
 29	struct xfs_scrub	*sc,
 30	struct xfs_inode	*ip)
 31{
 32	int			error;
 33
 34	error = xchk_get_inode(sc, ip);
 35	if (error)
 36		goto out;
 37
 38	sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
 39	xfs_ilock(sc->ip, sc->ilock_flags);
 40
 41	/*
 42	 * We don't want any ephemeral data fork updates sitting around
 43	 * while we inspect block mappings, so wait for directio to finish
 44	 * and flush dirty data if we have delalloc reservations.
 45	 */
 46	if (S_ISREG(VFS_I(sc->ip)->i_mode) &&
 47	    sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) {
 
 
 48		inode_dio_wait(VFS_I(sc->ip));
 49		error = filemap_write_and_wait(VFS_I(sc->ip)->i_mapping);
 50		if (error)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 51			goto out;
 52	}
 53
 54	/* Got the inode, lock it and we're ready to go. */
 55	error = xchk_trans_alloc(sc, 0);
 56	if (error)
 57		goto out;
 58	sc->ilock_flags |= XFS_ILOCK_EXCL;
 59	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
 60
 61out:
 62	/* scrub teardown will unlock and release the inode */
 63	return error;
 64}
 65
 66/*
 67 * Inode fork block mapping (BMBT) scrubber.
 68 * More complex than the others because we have to scrub
 69 * all the extents regardless of whether or not the fork
 70 * is in btree format.
 71 */
 72
 73struct xchk_bmap_info {
 74	struct xfs_scrub	*sc;
 75	xfs_fileoff_t		lastoff;
 76	bool			is_rt;
 77	bool			is_shared;
 78	bool			was_loaded;
 79	int			whichfork;
 80};
 81
 82/* Look for a corresponding rmap for this irec. */
 83static inline bool
 84xchk_bmap_get_rmap(
 85	struct xchk_bmap_info	*info,
 86	struct xfs_bmbt_irec	*irec,
 87	xfs_agblock_t		agbno,
 88	uint64_t		owner,
 89	struct xfs_rmap_irec	*rmap)
 90{
 91	xfs_fileoff_t		offset;
 92	unsigned int		rflags = 0;
 93	int			has_rmap;
 94	int			error;
 95
 96	if (info->whichfork == XFS_ATTR_FORK)
 97		rflags |= XFS_RMAP_ATTR_FORK;
 
 
 98
 99	/*
100	 * CoW staging extents are owned (on disk) by the refcountbt, so
101	 * their rmaps do not have offsets.
102	 */
103	if (info->whichfork == XFS_COW_FORK)
104		offset = 0;
105	else
106		offset = irec->br_startoff;
107
108	/*
109	 * If the caller thinks this could be a shared bmbt extent (IOWs,
110	 * any data fork extent of a reflink inode) then we have to use the
111	 * range rmap lookup to make sure we get the correct owner/offset.
112	 */
113	if (info->is_shared) {
114		error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno,
115				owner, offset, rflags, rmap, &has_rmap);
116		if (!xchk_should_check_xref(info->sc, &error,
117				&info->sc->sa.rmap_cur))
118			return false;
119		goto out;
120	}
121
122	/*
123	 * Otherwise, use the (faster) regular lookup.
124	 */
125	error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno, 0, owner,
126			offset, rflags, &has_rmap);
127	if (!xchk_should_check_xref(info->sc, &error,
128			&info->sc->sa.rmap_cur))
129		return false;
130	if (!has_rmap)
131		goto out;
132
133	error = xfs_rmap_get_rec(info->sc->sa.rmap_cur, rmap, &has_rmap);
134	if (!xchk_should_check_xref(info->sc, &error,
135			&info->sc->sa.rmap_cur))
136		return false;
137
138out:
139	if (!has_rmap)
140		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
141			irec->br_startoff);
142	return has_rmap;
143}
144
145/* Make sure that we have rmapbt records for this extent. */
146STATIC void
147xchk_bmap_xref_rmap(
148	struct xchk_bmap_info	*info,
149	struct xfs_bmbt_irec	*irec,
150	xfs_agblock_t		agbno)
151{
152	struct xfs_rmap_irec	rmap;
153	unsigned long long	rmap_end;
154	uint64_t		owner;
155
156	if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm))
157		return;
158
159	if (info->whichfork == XFS_COW_FORK)
160		owner = XFS_RMAP_OWN_COW;
161	else
162		owner = info->sc->ip->i_ino;
163
164	/* Find the rmap record for this irec. */
165	if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap))
166		return;
167
168	/* Check the rmap. */
169	rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
170	if (rmap.rm_startblock > agbno ||
171	    agbno + irec->br_blockcount > rmap_end)
172		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
173				irec->br_startoff);
174
175	/*
176	 * Check the logical offsets if applicable.  CoW staging extents
177	 * don't track logical offsets since the mappings only exist in
178	 * memory.
179	 */
180	if (info->whichfork != XFS_COW_FORK) {
181		rmap_end = (unsigned long long)rmap.rm_offset +
182				rmap.rm_blockcount;
183		if (rmap.rm_offset > irec->br_startoff ||
184		    irec->br_startoff + irec->br_blockcount > rmap_end)
185			xchk_fblock_xref_set_corrupt(info->sc,
186					info->whichfork, irec->br_startoff);
187	}
188
189	if (rmap.rm_owner != owner)
190		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
191				irec->br_startoff);
192
193	/*
194	 * Check for discrepancies between the unwritten flag in the irec and
195	 * the rmap.  Note that the (in-memory) CoW fork distinguishes between
196	 * unwritten and written extents, but we don't track that in the rmap
197	 * records because the blocks are owned (on-disk) by the refcountbt,
198	 * which doesn't track unwritten state.
199	 */
200	if (owner != XFS_RMAP_OWN_COW &&
201	    irec->br_state == XFS_EXT_UNWRITTEN &&
202	    !(rmap.rm_flags & XFS_RMAP_UNWRITTEN))
203		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
204				irec->br_startoff);
205
206	if (info->whichfork == XFS_ATTR_FORK &&
207	    !(rmap.rm_flags & XFS_RMAP_ATTR_FORK))
208		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
209				irec->br_startoff);
210	if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK)
211		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
212				irec->br_startoff);
213}
214
215/* Cross-reference a single rtdev extent record. */
216STATIC void
217xchk_bmap_rt_iextent_xref(
218	struct xfs_inode	*ip,
219	struct xchk_bmap_info	*info,
220	struct xfs_bmbt_irec	*irec)
221{
222	xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
223			irec->br_blockcount);
224}
225
226/* Cross-reference a single datadev extent record. */
227STATIC void
228xchk_bmap_iextent_xref(
229	struct xfs_inode	*ip,
230	struct xchk_bmap_info	*info,
231	struct xfs_bmbt_irec	*irec)
232{
233	struct xfs_mount	*mp = info->sc->mp;
234	xfs_agnumber_t		agno;
235	xfs_agblock_t		agbno;
236	xfs_extlen_t		len;
237	int			error;
238
239	agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
240	agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
241	len = irec->br_blockcount;
242
243	error = xchk_ag_init(info->sc, agno, &info->sc->sa);
244	if (!xchk_fblock_process_error(info->sc, info->whichfork,
245			irec->br_startoff, &error))
246		return;
247
248	xchk_xref_is_used_space(info->sc, agbno, len);
249	xchk_xref_is_not_inode_chunk(info->sc, agbno, len);
250	xchk_bmap_xref_rmap(info, irec, agbno);
251	switch (info->whichfork) {
252	case XFS_DATA_FORK:
253		if (xfs_is_reflink_inode(info->sc->ip))
254			break;
255		/* fall through */
256	case XFS_ATTR_FORK:
257		xchk_xref_is_not_shared(info->sc, agbno,
258				irec->br_blockcount);
259		break;
260	case XFS_COW_FORK:
261		xchk_xref_is_cow_staging(info->sc, agbno,
262				irec->br_blockcount);
263		break;
264	}
265
266	xchk_ag_free(info->sc, &info->sc->sa);
267}
268
269/*
270 * Directories and attr forks should never have blocks that can't be addressed
271 * by a xfs_dablk_t.
272 */
273STATIC void
274xchk_bmap_dirattr_extent(
275	struct xfs_inode	*ip,
276	struct xchk_bmap_info	*info,
277	struct xfs_bmbt_irec	*irec)
278{
279	struct xfs_mount	*mp = ip->i_mount;
280	xfs_fileoff_t		off;
281
282	if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK)
283		return;
284
285	if (!xfs_verify_dablk(mp, irec->br_startoff))
286		xchk_fblock_set_corrupt(info->sc, info->whichfork,
287				irec->br_startoff);
288
289	off = irec->br_startoff + irec->br_blockcount - 1;
290	if (!xfs_verify_dablk(mp, off))
291		xchk_fblock_set_corrupt(info->sc, info->whichfork, off);
292}
293
294/* Scrub a single extent record. */
295STATIC int
296xchk_bmap_iextent(
297	struct xfs_inode	*ip,
298	struct xchk_bmap_info	*info,
299	struct xfs_bmbt_irec	*irec)
300{
301	struct xfs_mount	*mp = info->sc->mp;
302	xfs_filblks_t		end;
303	int			error = 0;
304
305	/*
306	 * Check for out-of-order extents.  This record could have come
307	 * from the incore list, for which there is no ordering check.
308	 */
309	if (irec->br_startoff < info->lastoff)
310		xchk_fblock_set_corrupt(info->sc, info->whichfork,
311				irec->br_startoff);
312
 
 
 
 
313	xchk_bmap_dirattr_extent(ip, info, irec);
314
315	/* There should never be a "hole" extent in either extent list. */
316	if (irec->br_startblock == HOLESTARTBLOCK)
317		xchk_fblock_set_corrupt(info->sc, info->whichfork,
318				irec->br_startoff);
319
320	/*
321	 * Check for delalloc extents.  We never iterate the ones in the
322	 * in-core extent scan, and we should never see these in the bmbt.
323	 */
324	if (isnullstartblock(irec->br_startblock))
325		xchk_fblock_set_corrupt(info->sc, info->whichfork,
326				irec->br_startoff);
327
328	/* Make sure the extent points to a valid place. */
329	if (irec->br_blockcount > MAXEXTLEN)
330		xchk_fblock_set_corrupt(info->sc, info->whichfork,
331				irec->br_startoff);
332	if (irec->br_startblock + irec->br_blockcount <= irec->br_startblock)
333		xchk_fblock_set_corrupt(info->sc, info->whichfork,
334				irec->br_startoff);
335	end = irec->br_startblock + irec->br_blockcount - 1;
336	if (info->is_rt &&
337	    (!xfs_verify_rtbno(mp, irec->br_startblock) ||
338	     !xfs_verify_rtbno(mp, end)))
339		xchk_fblock_set_corrupt(info->sc, info->whichfork,
340				irec->br_startoff);
341	if (!info->is_rt &&
342	    (!xfs_verify_fsbno(mp, irec->br_startblock) ||
343	     !xfs_verify_fsbno(mp, end) ||
344	     XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
345				XFS_FSB_TO_AGNO(mp, end)))
346		xchk_fblock_set_corrupt(info->sc, info->whichfork,
347				irec->br_startoff);
348
349	/* We don't allow unwritten extents on attr forks. */
350	if (irec->br_state == XFS_EXT_UNWRITTEN &&
351	    info->whichfork == XFS_ATTR_FORK)
352		xchk_fblock_set_corrupt(info->sc, info->whichfork,
353				irec->br_startoff);
354
355	if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
356		return 0;
357
358	if (info->is_rt)
359		xchk_bmap_rt_iextent_xref(ip, info, irec);
360	else
361		xchk_bmap_iextent_xref(ip, info, irec);
362
363	info->lastoff = irec->br_startoff + irec->br_blockcount;
364	return error;
365}
366
367/* Scrub a bmbt record. */
368STATIC int
369xchk_bmapbt_rec(
370	struct xchk_btree	*bs,
371	union xfs_btree_rec	*rec)
372{
373	struct xfs_bmbt_irec	irec;
374	struct xfs_bmbt_irec	iext_irec;
375	struct xfs_iext_cursor	icur;
376	struct xchk_bmap_info	*info = bs->private;
377	struct xfs_inode	*ip = bs->cur->bc_private.b.ip;
378	struct xfs_buf		*bp = NULL;
379	struct xfs_btree_block	*block;
380	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, info->whichfork);
381	uint64_t		owner;
382	int			i;
383
384	/*
385	 * Check the owners of the btree blocks up to the level below
386	 * the root since the verifiers don't do that.
387	 */
388	if (xfs_sb_version_hascrc(&bs->cur->bc_mp->m_sb) &&
389	    bs->cur->bc_ptrs[0] == 1) {
390		for (i = 0; i < bs->cur->bc_nlevels - 1; i++) {
391			block = xfs_btree_get_block(bs->cur, i, &bp);
392			owner = be64_to_cpu(block->bb_u.l.bb_owner);
393			if (owner != ip->i_ino)
394				xchk_fblock_set_corrupt(bs->sc,
395						info->whichfork, 0);
396		}
397	}
398
399	/*
400	 * Check that the incore extent tree contains an extent that matches
401	 * this one exactly.  We validate those cached bmaps later, so we don't
402	 * need to check them here.  If the incore extent tree was just loaded
403	 * from disk by the scrubber, we assume that its contents match what's
404	 * on disk (we still hold the ILOCK) and skip the equivalence check.
405	 */
406	if (!info->was_loaded)
407		return 0;
408
409	xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
410	if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur,
411				&iext_irec) ||
412	    irec.br_startoff != iext_irec.br_startoff ||
413	    irec.br_startblock != iext_irec.br_startblock ||
414	    irec.br_blockcount != iext_irec.br_blockcount ||
415	    irec.br_state != iext_irec.br_state)
416		xchk_fblock_set_corrupt(bs->sc, info->whichfork,
417				irec.br_startoff);
418	return 0;
419}
420
421/* Scan the btree records. */
422STATIC int
423xchk_bmap_btree(
424	struct xfs_scrub	*sc,
425	int			whichfork,
426	struct xchk_bmap_info	*info)
427{
428	struct xfs_owner_info	oinfo;
429	struct xfs_ifork	*ifp = XFS_IFORK_PTR(sc->ip, whichfork);
430	struct xfs_mount	*mp = sc->mp;
431	struct xfs_inode	*ip = sc->ip;
432	struct xfs_btree_cur	*cur;
433	int			error;
434
435	/* Load the incore bmap cache if it's not loaded. */
436	info->was_loaded = ifp->if_flags & XFS_IFEXTENTS;
437	if (!info->was_loaded) {
438		error = xfs_iread_extents(sc->tp, ip, whichfork);
439		if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
440			goto out;
441	}
442
443	/* Check the btree structure. */
444	cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
445	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
446	error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
447	xfs_btree_del_cursor(cur, error);
448out:
449	return error;
450}
451
452struct xchk_bmap_check_rmap_info {
453	struct xfs_scrub	*sc;
454	int			whichfork;
455	struct xfs_iext_cursor	icur;
456};
457
458/* Can we find bmaps that fit this rmap? */
459STATIC int
460xchk_bmap_check_rmap(
461	struct xfs_btree_cur		*cur,
462	struct xfs_rmap_irec		*rec,
463	void				*priv)
464{
465	struct xfs_bmbt_irec		irec;
466	struct xchk_bmap_check_rmap_info	*sbcri = priv;
467	struct xfs_ifork		*ifp;
468	struct xfs_scrub		*sc = sbcri->sc;
469	bool				have_map;
470
471	/* Is this even the right fork? */
472	if (rec->rm_owner != sc->ip->i_ino)
473		return 0;
474	if ((sbcri->whichfork == XFS_ATTR_FORK) ^
475	    !!(rec->rm_flags & XFS_RMAP_ATTR_FORK))
476		return 0;
477	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
478		return 0;
479
480	/* Now look up the bmbt record. */
481	ifp = XFS_IFORK_PTR(sc->ip, sbcri->whichfork);
482	if (!ifp) {
483		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
484				rec->rm_offset);
485		goto out;
486	}
487	have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset,
488			&sbcri->icur, &irec);
489	if (!have_map)
490		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
491				rec->rm_offset);
492	/*
493	 * bmap extent record lengths are constrained to 2^21 blocks in length
494	 * because of space constraints in the on-disk metadata structure.
495	 * However, rmap extent record lengths are constrained only by AG
496	 * length, so we have to loop through the bmbt to make sure that the
497	 * entire rmap is covered by bmbt records.
498	 */
499	while (have_map) {
500		if (irec.br_startoff != rec->rm_offset)
501			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
502					rec->rm_offset);
503		if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp,
504				cur->bc_private.a.agno, rec->rm_startblock))
505			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
506					rec->rm_offset);
507		if (irec.br_blockcount > rec->rm_blockcount)
508			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
509					rec->rm_offset);
510		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
511			break;
512		rec->rm_startblock += irec.br_blockcount;
513		rec->rm_offset += irec.br_blockcount;
514		rec->rm_blockcount -= irec.br_blockcount;
515		if (rec->rm_blockcount == 0)
516			break;
517		have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec);
518		if (!have_map)
519			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
520					rec->rm_offset);
521	}
522
523out:
524	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
525		return -ECANCELED;
526	return 0;
527}
528
529/* Make sure each rmap has a corresponding bmbt entry. */
530STATIC int
531xchk_bmap_check_ag_rmaps(
532	struct xfs_scrub		*sc,
533	int				whichfork,
534	xfs_agnumber_t			agno)
535{
536	struct xchk_bmap_check_rmap_info	sbcri;
537	struct xfs_btree_cur		*cur;
538	struct xfs_buf			*agf;
539	int				error;
540
541	error = xfs_alloc_read_agf(sc->mp, sc->tp, agno, 0, &agf);
542	if (error)
543		return error;
544
545	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, agno);
546	if (!cur) {
547		error = -ENOMEM;
548		goto out_agf;
549	}
550
551	sbcri.sc = sc;
552	sbcri.whichfork = whichfork;
553	error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri);
554	if (error == -ECANCELED)
555		error = 0;
556
557	xfs_btree_del_cursor(cur, error);
558out_agf:
559	xfs_trans_brelse(sc->tp, agf);
560	return error;
561}
562
563/* Make sure each rmap has a corresponding bmbt entry. */
564STATIC int
565xchk_bmap_check_rmaps(
566	struct xfs_scrub	*sc,
567	int			whichfork)
568{
569	loff_t			size;
 
570	xfs_agnumber_t		agno;
 
571	int			error;
572
573	if (!xfs_sb_version_hasrmapbt(&sc->mp->m_sb) ||
574	    whichfork == XFS_COW_FORK ||
575	    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
576		return 0;
577
578	/* Don't support realtime rmap checks yet. */
579	if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK)
580		return 0;
581
 
 
582	/*
583	 * Only do this for complex maps that are in btree format, or for
584	 * situations where we would seem to have a size but zero extents.
585	 * The inode repair code can zap broken iforks, which means we have
586	 * to flag this bmap as corrupt if there are rmaps that need to be
587	 * reattached.
588	 */
589	switch (whichfork) {
590	case XFS_DATA_FORK:
591		size = i_size_read(VFS_I(sc->ip));
592		break;
593	case XFS_ATTR_FORK:
594		size = XFS_IFORK_Q(sc->ip);
595		break;
596	default:
597		size = 0;
598		break;
599	}
600	if (XFS_IFORK_FORMAT(sc->ip, whichfork) != XFS_DINODE_FMT_BTREE &&
601	    (size == 0 || XFS_IFORK_NEXTENTS(sc->ip, whichfork) > 0))
602		return 0;
603
604	for (agno = 0; agno < sc->mp->m_sb.sb_agcount; agno++) {
605		error = xchk_bmap_check_ag_rmaps(sc, whichfork, agno);
606		if (error)
607			return error;
608		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
609			break;
610	}
611
612	return 0;
 
613}
614
615/*
616 * Scrub an inode fork's block mappings.
617 *
618 * First we scan every record in every btree block, if applicable.
619 * Then we unconditionally scan the incore extent cache.
620 */
621STATIC int
622xchk_bmap(
623	struct xfs_scrub	*sc,
624	int			whichfork)
625{
626	struct xfs_bmbt_irec	irec;
627	struct xchk_bmap_info	info = { NULL };
628	struct xfs_mount	*mp = sc->mp;
629	struct xfs_inode	*ip = sc->ip;
630	struct xfs_ifork	*ifp;
631	xfs_fileoff_t		endoff;
632	struct xfs_iext_cursor	icur;
633	int			error = 0;
634
635	ifp = XFS_IFORK_PTR(ip, whichfork);
 
 
636
637	info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
638	info.whichfork = whichfork;
639	info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
640	info.sc = sc;
641
642	switch (whichfork) {
643	case XFS_COW_FORK:
644		/* Non-existent CoW forks are ignorable. */
645		if (!ifp)
646			goto out;
647		/* No CoW forks on non-reflink inodes/filesystems. */
648		if (!xfs_is_reflink_inode(ip)) {
649			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
650			goto out;
651		}
652		break;
653	case XFS_ATTR_FORK:
654		if (!ifp)
655			goto out_check_rmap;
656		if (!xfs_sb_version_hasattr(&mp->m_sb) &&
657		    !xfs_sb_version_hasattr2(&mp->m_sb))
658			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
659		break;
660	default:
661		ASSERT(whichfork == XFS_DATA_FORK);
662		break;
663	}
664
665	/* Check the fork values */
666	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
667	case XFS_DINODE_FMT_UUID:
668	case XFS_DINODE_FMT_DEV:
669	case XFS_DINODE_FMT_LOCAL:
670		/* No mappings to check. */
671		goto out;
672	case XFS_DINODE_FMT_EXTENTS:
673		if (!(ifp->if_flags & XFS_IFEXTENTS)) {
674			xchk_fblock_set_corrupt(sc, whichfork, 0);
675			goto out;
676		}
677		break;
678	case XFS_DINODE_FMT_BTREE:
679		if (whichfork == XFS_COW_FORK) {
680			xchk_fblock_set_corrupt(sc, whichfork, 0);
681			goto out;
682		}
683
684		error = xchk_bmap_btree(sc, whichfork, &info);
685		if (error)
686			goto out;
687		break;
688	default:
689		xchk_fblock_set_corrupt(sc, whichfork, 0);
690		goto out;
691	}
692
693	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
694		goto out;
695
696	/* Find the offset of the last extent in the mapping. */
697	error = xfs_bmap_last_offset(ip, &endoff, whichfork);
698	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
699		goto out;
700
701	/* Scrub extent records. */
702	info.lastoff = 0;
703	ifp = XFS_IFORK_PTR(ip, whichfork);
704	for_each_xfs_iext(ifp, &icur, &irec) {
705		if (xchk_should_terminate(sc, &error) ||
706		    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
707			goto out;
708		if (isnullstartblock(irec.br_startblock))
709			continue;
710		if (irec.br_startoff >= endoff) {
711			xchk_fblock_set_corrupt(sc, whichfork,
712					irec.br_startoff);
713			goto out;
714		}
715		error = xchk_bmap_iextent(ip, &info, &irec);
716		if (error)
717			goto out;
718	}
719
720out_check_rmap:
721	error = xchk_bmap_check_rmaps(sc, whichfork);
722	if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error))
723		goto out;
724out:
725	return error;
726}
727
728/* Scrub an inode's data fork. */
729int
730xchk_bmap_data(
731	struct xfs_scrub	*sc)
732{
733	return xchk_bmap(sc, XFS_DATA_FORK);
734}
735
736/* Scrub an inode's attr fork. */
737int
738xchk_bmap_attr(
739	struct xfs_scrub	*sc)
740{
741	return xchk_bmap(sc, XFS_ATTR_FORK);
742}
743
744/* Scrub an inode's CoW fork. */
745int
746xchk_bmap_cow(
747	struct xfs_scrub	*sc)
748{
749	if (!xfs_is_reflink_inode(sc->ip))
750		return -ENOENT;
751
752	return xchk_bmap(sc, XFS_COW_FORK);
753}
v5.14.15
  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_bit.h"
 14#include "xfs_log_format.h"
 15#include "xfs_trans.h"
 16#include "xfs_inode.h"
 17#include "xfs_alloc.h"
 18#include "xfs_bmap.h"
 19#include "xfs_bmap_btree.h"
 20#include "xfs_rmap.h"
 21#include "xfs_rmap_btree.h"
 22#include "scrub/scrub.h"
 23#include "scrub/common.h"
 24#include "scrub/btree.h"
 25#include "xfs_ag.h"
 26
 27/* Set us up with an inode's bmap. */
 28int
 29xchk_setup_inode_bmap(
 30	struct xfs_scrub	*sc)
 
 31{
 32	int			error;
 33
 34	error = xchk_get_inode(sc);
 35	if (error)
 36		goto out;
 37
 38	sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
 39	xfs_ilock(sc->ip, sc->ilock_flags);
 40
 41	/*
 42	 * We don't want any ephemeral data fork updates sitting around
 43	 * while we inspect block mappings, so wait for directio to finish
 44	 * and flush dirty data if we have delalloc reservations.
 45	 */
 46	if (S_ISREG(VFS_I(sc->ip)->i_mode) &&
 47	    sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) {
 48		struct address_space	*mapping = VFS_I(sc->ip)->i_mapping;
 49
 50		inode_dio_wait(VFS_I(sc->ip));
 51
 52		/*
 53		 * Try to flush all incore state to disk before we examine the
 54		 * space mappings for the data fork.  Leave accumulated errors
 55		 * in the mapping for the writer threads to consume.
 56		 *
 57		 * On ENOSPC or EIO writeback errors, we continue into the
 58		 * extent mapping checks because write failures do not
 59		 * necessarily imply anything about the correctness of the file
 60		 * metadata.  The metadata and the file data could be on
 61		 * completely separate devices; a media failure might only
 62		 * affect a subset of the disk, etc.  We can handle delalloc
 63		 * extents in the scrubber, so leaving them in memory is fine.
 64		 */
 65		error = filemap_fdatawrite(mapping);
 66		if (!error)
 67			error = filemap_fdatawait_keep_errors(mapping);
 68		if (error && (error != -ENOSPC && error != -EIO))
 69			goto out;
 70	}
 71
 72	/* Got the inode, lock it and we're ready to go. */
 73	error = xchk_trans_alloc(sc, 0);
 74	if (error)
 75		goto out;
 76	sc->ilock_flags |= XFS_ILOCK_EXCL;
 77	xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
 78
 79out:
 80	/* scrub teardown will unlock and release the inode */
 81	return error;
 82}
 83
 84/*
 85 * Inode fork block mapping (BMBT) scrubber.
 86 * More complex than the others because we have to scrub
 87 * all the extents regardless of whether or not the fork
 88 * is in btree format.
 89 */
 90
 91struct xchk_bmap_info {
 92	struct xfs_scrub	*sc;
 93	xfs_fileoff_t		lastoff;
 94	bool			is_rt;
 95	bool			is_shared;
 96	bool			was_loaded;
 97	int			whichfork;
 98};
 99
100/* Look for a corresponding rmap for this irec. */
101static inline bool
102xchk_bmap_get_rmap(
103	struct xchk_bmap_info	*info,
104	struct xfs_bmbt_irec	*irec,
105	xfs_agblock_t		agbno,
106	uint64_t		owner,
107	struct xfs_rmap_irec	*rmap)
108{
109	xfs_fileoff_t		offset;
110	unsigned int		rflags = 0;
111	int			has_rmap;
112	int			error;
113
114	if (info->whichfork == XFS_ATTR_FORK)
115		rflags |= XFS_RMAP_ATTR_FORK;
116	if (irec->br_state == XFS_EXT_UNWRITTEN)
117		rflags |= XFS_RMAP_UNWRITTEN;
118
119	/*
120	 * CoW staging extents are owned (on disk) by the refcountbt, so
121	 * their rmaps do not have offsets.
122	 */
123	if (info->whichfork == XFS_COW_FORK)
124		offset = 0;
125	else
126		offset = irec->br_startoff;
127
128	/*
129	 * If the caller thinks this could be a shared bmbt extent (IOWs,
130	 * any data fork extent of a reflink inode) then we have to use the
131	 * range rmap lookup to make sure we get the correct owner/offset.
132	 */
133	if (info->is_shared) {
134		error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno,
135				owner, offset, rflags, rmap, &has_rmap);
136		if (!xchk_should_check_xref(info->sc, &error,
137				&info->sc->sa.rmap_cur))
138			return false;
139		goto out;
140	}
141
142	/*
143	 * Otherwise, use the (faster) regular lookup.
144	 */
145	error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno, 0, owner,
146			offset, rflags, &has_rmap);
147	if (!xchk_should_check_xref(info->sc, &error,
148			&info->sc->sa.rmap_cur))
149		return false;
150	if (!has_rmap)
151		goto out;
152
153	error = xfs_rmap_get_rec(info->sc->sa.rmap_cur, rmap, &has_rmap);
154	if (!xchk_should_check_xref(info->sc, &error,
155			&info->sc->sa.rmap_cur))
156		return false;
157
158out:
159	if (!has_rmap)
160		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
161			irec->br_startoff);
162	return has_rmap;
163}
164
165/* Make sure that we have rmapbt records for this extent. */
166STATIC void
167xchk_bmap_xref_rmap(
168	struct xchk_bmap_info	*info,
169	struct xfs_bmbt_irec	*irec,
170	xfs_agblock_t		agbno)
171{
172	struct xfs_rmap_irec	rmap;
173	unsigned long long	rmap_end;
174	uint64_t		owner;
175
176	if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm))
177		return;
178
179	if (info->whichfork == XFS_COW_FORK)
180		owner = XFS_RMAP_OWN_COW;
181	else
182		owner = info->sc->ip->i_ino;
183
184	/* Find the rmap record for this irec. */
185	if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap))
186		return;
187
188	/* Check the rmap. */
189	rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount;
190	if (rmap.rm_startblock > agbno ||
191	    agbno + irec->br_blockcount > rmap_end)
192		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
193				irec->br_startoff);
194
195	/*
196	 * Check the logical offsets if applicable.  CoW staging extents
197	 * don't track logical offsets since the mappings only exist in
198	 * memory.
199	 */
200	if (info->whichfork != XFS_COW_FORK) {
201		rmap_end = (unsigned long long)rmap.rm_offset +
202				rmap.rm_blockcount;
203		if (rmap.rm_offset > irec->br_startoff ||
204		    irec->br_startoff + irec->br_blockcount > rmap_end)
205			xchk_fblock_xref_set_corrupt(info->sc,
206					info->whichfork, irec->br_startoff);
207	}
208
209	if (rmap.rm_owner != owner)
210		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
211				irec->br_startoff);
212
213	/*
214	 * Check for discrepancies between the unwritten flag in the irec and
215	 * the rmap.  Note that the (in-memory) CoW fork distinguishes between
216	 * unwritten and written extents, but we don't track that in the rmap
217	 * records because the blocks are owned (on-disk) by the refcountbt,
218	 * which doesn't track unwritten state.
219	 */
220	if (owner != XFS_RMAP_OWN_COW &&
221	    !!(irec->br_state == XFS_EXT_UNWRITTEN) !=
222	    !!(rmap.rm_flags & XFS_RMAP_UNWRITTEN))
223		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
224				irec->br_startoff);
225
226	if (!!(info->whichfork == XFS_ATTR_FORK) !=
227	    !!(rmap.rm_flags & XFS_RMAP_ATTR_FORK))
228		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
229				irec->br_startoff);
230	if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK)
231		xchk_fblock_xref_set_corrupt(info->sc, info->whichfork,
232				irec->br_startoff);
233}
234
235/* Cross-reference a single rtdev extent record. */
236STATIC void
237xchk_bmap_rt_iextent_xref(
238	struct xfs_inode	*ip,
239	struct xchk_bmap_info	*info,
240	struct xfs_bmbt_irec	*irec)
241{
242	xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
243			irec->br_blockcount);
244}
245
246/* Cross-reference a single datadev extent record. */
247STATIC void
248xchk_bmap_iextent_xref(
249	struct xfs_inode	*ip,
250	struct xchk_bmap_info	*info,
251	struct xfs_bmbt_irec	*irec)
252{
253	struct xfs_mount	*mp = info->sc->mp;
254	xfs_agnumber_t		agno;
255	xfs_agblock_t		agbno;
256	xfs_extlen_t		len;
257	int			error;
258
259	agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
260	agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
261	len = irec->br_blockcount;
262
263	error = xchk_ag_init(info->sc, agno, &info->sc->sa);
264	if (!xchk_fblock_process_error(info->sc, info->whichfork,
265			irec->br_startoff, &error))
266		return;
267
268	xchk_xref_is_used_space(info->sc, agbno, len);
269	xchk_xref_is_not_inode_chunk(info->sc, agbno, len);
270	xchk_bmap_xref_rmap(info, irec, agbno);
271	switch (info->whichfork) {
272	case XFS_DATA_FORK:
273		if (xfs_is_reflink_inode(info->sc->ip))
274			break;
275		fallthrough;
276	case XFS_ATTR_FORK:
277		xchk_xref_is_not_shared(info->sc, agbno,
278				irec->br_blockcount);
279		break;
280	case XFS_COW_FORK:
281		xchk_xref_is_cow_staging(info->sc, agbno,
282				irec->br_blockcount);
283		break;
284	}
285
286	xchk_ag_free(info->sc, &info->sc->sa);
287}
288
289/*
290 * Directories and attr forks should never have blocks that can't be addressed
291 * by a xfs_dablk_t.
292 */
293STATIC void
294xchk_bmap_dirattr_extent(
295	struct xfs_inode	*ip,
296	struct xchk_bmap_info	*info,
297	struct xfs_bmbt_irec	*irec)
298{
299	struct xfs_mount	*mp = ip->i_mount;
300	xfs_fileoff_t		off;
301
302	if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK)
303		return;
304
305	if (!xfs_verify_dablk(mp, irec->br_startoff))
306		xchk_fblock_set_corrupt(info->sc, info->whichfork,
307				irec->br_startoff);
308
309	off = irec->br_startoff + irec->br_blockcount - 1;
310	if (!xfs_verify_dablk(mp, off))
311		xchk_fblock_set_corrupt(info->sc, info->whichfork, off);
312}
313
314/* Scrub a single extent record. */
315STATIC int
316xchk_bmap_iextent(
317	struct xfs_inode	*ip,
318	struct xchk_bmap_info	*info,
319	struct xfs_bmbt_irec	*irec)
320{
321	struct xfs_mount	*mp = info->sc->mp;
 
322	int			error = 0;
323
324	/*
325	 * Check for out-of-order extents.  This record could have come
326	 * from the incore list, for which there is no ordering check.
327	 */
328	if (irec->br_startoff < info->lastoff)
329		xchk_fblock_set_corrupt(info->sc, info->whichfork,
330				irec->br_startoff);
331
332	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
333		xchk_fblock_set_corrupt(info->sc, info->whichfork,
334				irec->br_startoff);
335
336	xchk_bmap_dirattr_extent(ip, info, irec);
337
338	/* There should never be a "hole" extent in either extent list. */
339	if (irec->br_startblock == HOLESTARTBLOCK)
340		xchk_fblock_set_corrupt(info->sc, info->whichfork,
341				irec->br_startoff);
342
343	/*
344	 * Check for delalloc extents.  We never iterate the ones in the
345	 * in-core extent scan, and we should never see these in the bmbt.
346	 */
347	if (isnullstartblock(irec->br_startblock))
348		xchk_fblock_set_corrupt(info->sc, info->whichfork,
349				irec->br_startoff);
350
351	/* Make sure the extent points to a valid place. */
352	if (irec->br_blockcount > MAXEXTLEN)
353		xchk_fblock_set_corrupt(info->sc, info->whichfork,
354				irec->br_startoff);
 
 
 
 
355	if (info->is_rt &&
356	    !xfs_verify_rtext(mp, irec->br_startblock, irec->br_blockcount))
 
357		xchk_fblock_set_corrupt(info->sc, info->whichfork,
358				irec->br_startoff);
359	if (!info->is_rt &&
360	    !xfs_verify_fsbext(mp, irec->br_startblock, irec->br_blockcount))
 
 
 
361		xchk_fblock_set_corrupt(info->sc, info->whichfork,
362				irec->br_startoff);
363
364	/* We don't allow unwritten extents on attr forks. */
365	if (irec->br_state == XFS_EXT_UNWRITTEN &&
366	    info->whichfork == XFS_ATTR_FORK)
367		xchk_fblock_set_corrupt(info->sc, info->whichfork,
368				irec->br_startoff);
369
370	if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
371		return 0;
372
373	if (info->is_rt)
374		xchk_bmap_rt_iextent_xref(ip, info, irec);
375	else
376		xchk_bmap_iextent_xref(ip, info, irec);
377
378	info->lastoff = irec->br_startoff + irec->br_blockcount;
379	return error;
380}
381
382/* Scrub a bmbt record. */
383STATIC int
384xchk_bmapbt_rec(
385	struct xchk_btree	*bs,
386	union xfs_btree_rec	*rec)
387{
388	struct xfs_bmbt_irec	irec;
389	struct xfs_bmbt_irec	iext_irec;
390	struct xfs_iext_cursor	icur;
391	struct xchk_bmap_info	*info = bs->private;
392	struct xfs_inode	*ip = bs->cur->bc_ino.ip;
393	struct xfs_buf		*bp = NULL;
394	struct xfs_btree_block	*block;
395	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, info->whichfork);
396	uint64_t		owner;
397	int			i;
398
399	/*
400	 * Check the owners of the btree blocks up to the level below
401	 * the root since the verifiers don't do that.
402	 */
403	if (xfs_sb_version_hascrc(&bs->cur->bc_mp->m_sb) &&
404	    bs->cur->bc_ptrs[0] == 1) {
405		for (i = 0; i < bs->cur->bc_nlevels - 1; i++) {
406			block = xfs_btree_get_block(bs->cur, i, &bp);
407			owner = be64_to_cpu(block->bb_u.l.bb_owner);
408			if (owner != ip->i_ino)
409				xchk_fblock_set_corrupt(bs->sc,
410						info->whichfork, 0);
411		}
412	}
413
414	/*
415	 * Check that the incore extent tree contains an extent that matches
416	 * this one exactly.  We validate those cached bmaps later, so we don't
417	 * need to check them here.  If the incore extent tree was just loaded
418	 * from disk by the scrubber, we assume that its contents match what's
419	 * on disk (we still hold the ILOCK) and skip the equivalence check.
420	 */
421	if (!info->was_loaded)
422		return 0;
423
424	xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
425	if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur,
426				&iext_irec) ||
427	    irec.br_startoff != iext_irec.br_startoff ||
428	    irec.br_startblock != iext_irec.br_startblock ||
429	    irec.br_blockcount != iext_irec.br_blockcount ||
430	    irec.br_state != iext_irec.br_state)
431		xchk_fblock_set_corrupt(bs->sc, info->whichfork,
432				irec.br_startoff);
433	return 0;
434}
435
436/* Scan the btree records. */
437STATIC int
438xchk_bmap_btree(
439	struct xfs_scrub	*sc,
440	int			whichfork,
441	struct xchk_bmap_info	*info)
442{
443	struct xfs_owner_info	oinfo;
444	struct xfs_ifork	*ifp = XFS_IFORK_PTR(sc->ip, whichfork);
445	struct xfs_mount	*mp = sc->mp;
446	struct xfs_inode	*ip = sc->ip;
447	struct xfs_btree_cur	*cur;
448	int			error;
449
450	/* Load the incore bmap cache if it's not loaded. */
451	info->was_loaded = !xfs_need_iread_extents(ifp);
452
453	error = xfs_iread_extents(sc->tp, ip, whichfork);
454	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
455		goto out;
 
456
457	/* Check the btree structure. */
458	cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
459	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
460	error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
461	xfs_btree_del_cursor(cur, error);
462out:
463	return error;
464}
465
466struct xchk_bmap_check_rmap_info {
467	struct xfs_scrub	*sc;
468	int			whichfork;
469	struct xfs_iext_cursor	icur;
470};
471
472/* Can we find bmaps that fit this rmap? */
473STATIC int
474xchk_bmap_check_rmap(
475	struct xfs_btree_cur		*cur,
476	struct xfs_rmap_irec		*rec,
477	void				*priv)
478{
479	struct xfs_bmbt_irec		irec;
480	struct xchk_bmap_check_rmap_info	*sbcri = priv;
481	struct xfs_ifork		*ifp;
482	struct xfs_scrub		*sc = sbcri->sc;
483	bool				have_map;
484
485	/* Is this even the right fork? */
486	if (rec->rm_owner != sc->ip->i_ino)
487		return 0;
488	if ((sbcri->whichfork == XFS_ATTR_FORK) ^
489	    !!(rec->rm_flags & XFS_RMAP_ATTR_FORK))
490		return 0;
491	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
492		return 0;
493
494	/* Now look up the bmbt record. */
495	ifp = XFS_IFORK_PTR(sc->ip, sbcri->whichfork);
496	if (!ifp) {
497		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
498				rec->rm_offset);
499		goto out;
500	}
501	have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset,
502			&sbcri->icur, &irec);
503	if (!have_map)
504		xchk_fblock_set_corrupt(sc, sbcri->whichfork,
505				rec->rm_offset);
506	/*
507	 * bmap extent record lengths are constrained to 2^21 blocks in length
508	 * because of space constraints in the on-disk metadata structure.
509	 * However, rmap extent record lengths are constrained only by AG
510	 * length, so we have to loop through the bmbt to make sure that the
511	 * entire rmap is covered by bmbt records.
512	 */
513	while (have_map) {
514		if (irec.br_startoff != rec->rm_offset)
515			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
516					rec->rm_offset);
517		if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp,
518				cur->bc_ag.pag->pag_agno, rec->rm_startblock))
519			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
520					rec->rm_offset);
521		if (irec.br_blockcount > rec->rm_blockcount)
522			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
523					rec->rm_offset);
524		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
525			break;
526		rec->rm_startblock += irec.br_blockcount;
527		rec->rm_offset += irec.br_blockcount;
528		rec->rm_blockcount -= irec.br_blockcount;
529		if (rec->rm_blockcount == 0)
530			break;
531		have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec);
532		if (!have_map)
533			xchk_fblock_set_corrupt(sc, sbcri->whichfork,
534					rec->rm_offset);
535	}
536
537out:
538	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
539		return -ECANCELED;
540	return 0;
541}
542
543/* Make sure each rmap has a corresponding bmbt entry. */
544STATIC int
545xchk_bmap_check_ag_rmaps(
546	struct xfs_scrub		*sc,
547	int				whichfork,
548	struct xfs_perag		*pag)
549{
550	struct xchk_bmap_check_rmap_info	sbcri;
551	struct xfs_btree_cur		*cur;
552	struct xfs_buf			*agf;
553	int				error;
554
555	error = xfs_alloc_read_agf(sc->mp, sc->tp, pag->pag_agno, 0, &agf);
556	if (error)
557		return error;
558
559	cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, pag);
 
 
 
 
560
561	sbcri.sc = sc;
562	sbcri.whichfork = whichfork;
563	error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri);
564	if (error == -ECANCELED)
565		error = 0;
566
567	xfs_btree_del_cursor(cur, error);
 
568	xfs_trans_brelse(sc->tp, agf);
569	return error;
570}
571
572/* Make sure each rmap has a corresponding bmbt entry. */
573STATIC int
574xchk_bmap_check_rmaps(
575	struct xfs_scrub	*sc,
576	int			whichfork)
577{
578	struct xfs_ifork	*ifp = XFS_IFORK_PTR(sc->ip, whichfork);
579	struct xfs_perag	*pag;
580	xfs_agnumber_t		agno;
581	bool			zero_size;
582	int			error;
583
584	if (!xfs_sb_version_hasrmapbt(&sc->mp->m_sb) ||
585	    whichfork == XFS_COW_FORK ||
586	    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
587		return 0;
588
589	/* Don't support realtime rmap checks yet. */
590	if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK)
591		return 0;
592
593	ASSERT(XFS_IFORK_PTR(sc->ip, whichfork) != NULL);
594
595	/*
596	 * Only do this for complex maps that are in btree format, or for
597	 * situations where we would seem to have a size but zero extents.
598	 * The inode repair code can zap broken iforks, which means we have
599	 * to flag this bmap as corrupt if there are rmaps that need to be
600	 * reattached.
601	 */
602
603	if (whichfork == XFS_DATA_FORK)
604		zero_size = i_size_read(VFS_I(sc->ip)) == 0;
605	else
606		zero_size = false;
607
608	if (ifp->if_format != XFS_DINODE_FMT_BTREE &&
609	    (zero_size || ifp->if_nextents > 0))
 
 
 
 
 
610		return 0;
611
612	for_each_perag(sc->mp, agno, pag) {
613		error = xchk_bmap_check_ag_rmaps(sc, whichfork, pag);
614		if (error)
615			break;
616		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
617			break;
618	}
619	if (pag)
620		xfs_perag_put(pag);
621	return error;
622}
623
624/*
625 * Scrub an inode fork's block mappings.
626 *
627 * First we scan every record in every btree block, if applicable.
628 * Then we unconditionally scan the incore extent cache.
629 */
630STATIC int
631xchk_bmap(
632	struct xfs_scrub	*sc,
633	int			whichfork)
634{
635	struct xfs_bmbt_irec	irec;
636	struct xchk_bmap_info	info = { NULL };
637	struct xfs_mount	*mp = sc->mp;
638	struct xfs_inode	*ip = sc->ip;
639	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
640	xfs_fileoff_t		endoff;
641	struct xfs_iext_cursor	icur;
642	int			error = 0;
643
644	/* Non-existent forks can be ignored. */
645	if (!ifp)
646		goto out;
647
648	info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
649	info.whichfork = whichfork;
650	info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
651	info.sc = sc;
652
653	switch (whichfork) {
654	case XFS_COW_FORK:
 
 
 
655		/* No CoW forks on non-reflink inodes/filesystems. */
656		if (!xfs_is_reflink_inode(ip)) {
657			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
658			goto out;
659		}
660		break;
661	case XFS_ATTR_FORK:
 
 
662		if (!xfs_sb_version_hasattr(&mp->m_sb) &&
663		    !xfs_sb_version_hasattr2(&mp->m_sb))
664			xchk_ino_set_corrupt(sc, sc->ip->i_ino);
665		break;
666	default:
667		ASSERT(whichfork == XFS_DATA_FORK);
668		break;
669	}
670
671	/* Check the fork values */
672	switch (ifp->if_format) {
673	case XFS_DINODE_FMT_UUID:
674	case XFS_DINODE_FMT_DEV:
675	case XFS_DINODE_FMT_LOCAL:
676		/* No mappings to check. */
677		goto out;
678	case XFS_DINODE_FMT_EXTENTS:
 
 
 
 
679		break;
680	case XFS_DINODE_FMT_BTREE:
681		if (whichfork == XFS_COW_FORK) {
682			xchk_fblock_set_corrupt(sc, whichfork, 0);
683			goto out;
684		}
685
686		error = xchk_bmap_btree(sc, whichfork, &info);
687		if (error)
688			goto out;
689		break;
690	default:
691		xchk_fblock_set_corrupt(sc, whichfork, 0);
692		goto out;
693	}
694
695	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
696		goto out;
697
698	/* Find the offset of the last extent in the mapping. */
699	error = xfs_bmap_last_offset(ip, &endoff, whichfork);
700	if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
701		goto out;
702
703	/* Scrub extent records. */
704	info.lastoff = 0;
705	ifp = XFS_IFORK_PTR(ip, whichfork);
706	for_each_xfs_iext(ifp, &icur, &irec) {
707		if (xchk_should_terminate(sc, &error) ||
708		    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
709			goto out;
710		if (isnullstartblock(irec.br_startblock))
711			continue;
712		if (irec.br_startoff >= endoff) {
713			xchk_fblock_set_corrupt(sc, whichfork,
714					irec.br_startoff);
715			goto out;
716		}
717		error = xchk_bmap_iextent(ip, &info, &irec);
718		if (error)
719			goto out;
720	}
721
 
722	error = xchk_bmap_check_rmaps(sc, whichfork);
723	if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error))
724		goto out;
725out:
726	return error;
727}
728
729/* Scrub an inode's data fork. */
730int
731xchk_bmap_data(
732	struct xfs_scrub	*sc)
733{
734	return xchk_bmap(sc, XFS_DATA_FORK);
735}
736
737/* Scrub an inode's attr fork. */
738int
739xchk_bmap_attr(
740	struct xfs_scrub	*sc)
741{
742	return xchk_bmap(sc, XFS_ATTR_FORK);
743}
744
745/* Scrub an inode's CoW fork. */
746int
747xchk_bmap_cow(
748	struct xfs_scrub	*sc)
749{
750	if (!xfs_is_reflink_inode(sc->ip))
751		return -ENOENT;
752
753	return xchk_bmap(sc, XFS_COW_FORK);
754}