1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2017-2023 Oracle.  All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_trans_resv.h"
11#include "xfs_mount.h"
12#include "xfs_log_format.h"
13#include "xfs_trans.h"
14#include "xfs_inode.h"
15#include "xfs_icache.h"
16#include "xfs_dir2.h"
17#include "xfs_dir2_priv.h"
18#include "xfs_health.h"
19#include "scrub/scrub.h"
20#include "scrub/common.h"
21#include "scrub/dabtree.h"
22#include "scrub/readdir.h"
23#include "scrub/health.h"
24
25/* Set us up to scrub directories. */
26int
27xchk_setup_directory(
28	struct xfs_scrub	*sc)
29{
30	return xchk_setup_inode_contents(sc, 0);
31}
32
33/* Directories */
34
35/* Scrub a directory entry. */
36
37/* Check that an inode's mode matches a given XFS_DIR3_FT_* type. */
38STATIC void
39xchk_dir_check_ftype(
40	struct xfs_scrub	*sc,
41	xfs_fileoff_t		offset,
42	struct xfs_inode	*ip,
43	int			ftype)
44{
45	struct xfs_mount	*mp = sc->mp;
46
47	if (!xfs_has_ftype(mp)) {
48		if (ftype != XFS_DIR3_FT_UNKNOWN && ftype != XFS_DIR3_FT_DIR)
49			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
50		return;
51	}
52
53	if (xfs_mode_to_ftype(VFS_I(ip)->i_mode) != ftype)
54		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
55}
56
57/*
58 * Scrub a single directory entry.
59 *
60 * Check the inode number to make sure it's sane, then we check that we can
61 * look up this filename.  Finally, we check the ftype.
62 */
63STATIC int
64xchk_dir_actor(
65	struct xfs_scrub	*sc,
66	struct xfs_inode	*dp,
67	xfs_dir2_dataptr_t	dapos,
68	const struct xfs_name	*name,
69	xfs_ino_t		ino,
70	void			*priv)
71{
72	struct xfs_mount	*mp = dp->i_mount;
73	struct xfs_inode	*ip;
74	xfs_ino_t		lookup_ino;
75	xfs_dablk_t		offset;
76	int			error = 0;
77
78	offset = xfs_dir2_db_to_da(mp->m_dir_geo,
79			xfs_dir2_dataptr_to_db(mp->m_dir_geo, dapos));
80
81	if (xchk_should_terminate(sc, &error))
82		return error;
83
84	/* Does this inode number make sense? */
85	if (!xfs_verify_dir_ino(mp, ino)) {
86		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
87		return -ECANCELED;
88	}
89
90	/* Does this name make sense? */
91	if (!xfs_dir2_namecheck(name->name, name->len)) {
92		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
93		return -ECANCELED;
94	}
95
96	if (xfs_dir2_samename(name, &xfs_name_dot)) {
97		/* If this is "." then check that the inum matches the dir. */
98		if (ino != dp->i_ino)
99			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
100	} else if (xfs_dir2_samename(name, &xfs_name_dotdot)) {
101		/*
102		 * If this is ".." in the root inode, check that the inum
103		 * matches this dir.
104		 */
105		if (dp->i_ino == mp->m_sb.sb_rootino && ino != dp->i_ino)
106			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
107	}
108
109	/* Verify that we can look up this name by hash. */
110	error = xchk_dir_lookup(sc, dp, name, &lookup_ino);
111	/* ENOENT means the hash lookup failed and the dir is corrupt */
112	if (error == -ENOENT)
113		error = -EFSCORRUPTED;
114	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, offset, &error))
115		goto out;
116	if (lookup_ino != ino) {
117		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, offset);
118		return -ECANCELED;
119	}
120
121	/*
122	 * Grab the inode pointed to by the dirent.  We release the inode
123	 * before we cancel the scrub transaction.
124	 *
125	 * If _iget returns -EINVAL or -ENOENT then the child inode number is
126	 * garbage and the directory is corrupt.  If the _iget returns
127	 * -EFSCORRUPTED or -EFSBADCRC then the child is corrupt which is a
128	 *  cross referencing error.  Any other error is an operational error.
129	 */
130	error = xchk_iget(sc, ino, &ip);
131	if (error == -EINVAL || error == -ENOENT) {
132		error = -EFSCORRUPTED;
133		xchk_fblock_process_error(sc, XFS_DATA_FORK, 0, &error);
134		goto out;
135	}
136	if (!xchk_fblock_xref_process_error(sc, XFS_DATA_FORK, offset, &error))
137		goto out;
138
139	xchk_dir_check_ftype(sc, offset, ip, name->type);
140	xchk_irele(sc, ip);
141out:
142	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
143		return -ECANCELED;
144	return error;
145}
146
147/* Scrub a directory btree record. */
148STATIC int
149xchk_dir_rec(
150	struct xchk_da_btree		*ds,
151	int				level)
152{
153	struct xfs_name			dname = { };
154	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
155	struct xfs_mount		*mp = ds->state->mp;
156	struct xfs_inode		*dp = ds->dargs.dp;
157	struct xfs_da_geometry		*geo = mp->m_dir_geo;
158	struct xfs_dir2_data_entry	*dent;
159	struct xfs_buf			*bp;
160	struct xfs_dir2_leaf_entry	*ent;
161	unsigned int			end;
162	unsigned int			iter_off;
163	xfs_ino_t			ino;
164	xfs_dablk_t			rec_bno;
165	xfs_dir2_db_t			db;
166	xfs_dir2_data_aoff_t		off;
167	xfs_dir2_dataptr_t		ptr;
168	xfs_dahash_t			calc_hash;
169	xfs_dahash_t			hash;
170	struct xfs_dir3_icleaf_hdr	hdr;
171	unsigned int			tag;
172	int				error;
173
174	ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
175	       blk->magic == XFS_DIR2_LEAFN_MAGIC);
176
177	xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
178	ent = hdr.ents + blk->index;
179
180	/* Check the hash of the entry. */
181	error = xchk_da_btree_hash(ds, level, &ent->hashval);
182	if (error)
183		goto out;
184
185	/* Valid hash pointer? */
186	ptr = be32_to_cpu(ent->address);
187	if (ptr == 0)
188		return 0;
189
190	/* Find the directory entry's location. */
191	db = xfs_dir2_dataptr_to_db(geo, ptr);
192	off = xfs_dir2_dataptr_to_off(geo, ptr);
193	rec_bno = xfs_dir2_db_to_da(geo, db);
194
195	if (rec_bno >= geo->leafblk) {
196		xchk_da_set_corrupt(ds, level);
197		goto out;
198	}
199	error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno,
200			XFS_DABUF_MAP_HOLE_OK, &bp);
201	if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
202			&error))
203		goto out;
204	if (!bp) {
205		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
206		goto out;
207	}
208	xchk_buffer_recheck(ds->sc, bp);
209
210	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
211		goto out_relse;
212
213	dent = bp->b_addr + off;
214
215	/* Make sure we got a real directory entry. */
216	iter_off = geo->data_entry_offset;
217	end = xfs_dir3_data_end_offset(geo, bp->b_addr);
218	if (!end) {
219		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
220		goto out_relse;
221	}
222	for (;;) {
223		struct xfs_dir2_data_entry	*dep = bp->b_addr + iter_off;
224		struct xfs_dir2_data_unused	*dup = bp->b_addr + iter_off;
225
226		if (iter_off >= end) {
227			xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
228			goto out_relse;
229		}
230
231		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
232			iter_off += be16_to_cpu(dup->length);
233			continue;
234		}
235		if (dep == dent)
236			break;
237		iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
238	}
239
240	/* Retrieve the entry, sanity check it, and compare hashes. */
241	ino = be64_to_cpu(dent->inumber);
242	hash = be32_to_cpu(ent->hashval);
243	tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
244	if (!xfs_verify_dir_ino(mp, ino) || tag != off)
245		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
246	if (dent->namelen == 0) {
247		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
248		goto out_relse;
249	}
250
251	/* Does the directory hash match? */
252	dname.name = dent->name;
253	dname.len = dent->namelen;
254	calc_hash = xfs_dir2_hashname(mp, &dname);
255	if (calc_hash != hash)
256		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
257
258out_relse:
259	xfs_trans_brelse(ds->dargs.trans, bp);
260out:
261	return error;
262}
263
264/*
265 * Is this unused entry either in the bestfree or smaller than all of
266 * them?  We've already checked that the bestfrees are sorted longest to
267 * shortest, and that there aren't any bogus entries.
268 */
269STATIC void
270xchk_directory_check_free_entry(
271	struct xfs_scrub		*sc,
272	xfs_dablk_t			lblk,
273	struct xfs_dir2_data_free	*bf,
274	struct xfs_dir2_data_unused	*dup)
275{
276	struct xfs_dir2_data_free	*dfp;
277	unsigned int			dup_length;
278
279	dup_length = be16_to_cpu(dup->length);
280
281	/* Unused entry is shorter than any of the bestfrees */
282	if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
283		return;
284
285	for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
286		if (dup_length == be16_to_cpu(dfp->length))
287			return;
288
289	/* Unused entry should be in the bestfrees but wasn't found. */
290	xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
291}
292
293/* Check free space info in a directory data block. */
294STATIC int
295xchk_directory_data_bestfree(
296	struct xfs_scrub		*sc,
297	xfs_dablk_t			lblk,
298	bool				is_block)
299{
300	struct xfs_dir2_data_unused	*dup;
301	struct xfs_dir2_data_free	*dfp;
302	struct xfs_buf			*bp;
303	struct xfs_dir2_data_free	*bf;
304	struct xfs_mount		*mp = sc->mp;
305	u16				tag;
306	unsigned int			nr_bestfrees = 0;
307	unsigned int			nr_frees = 0;
308	unsigned int			smallest_bestfree;
309	int				newlen;
310	unsigned int			offset;
311	unsigned int			end;
312	int				error;
313
314	if (is_block) {
315		/* dir block format */
316		if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
317			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
318		error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
319	} else {
320		/* dir data format */
321		error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, 0, &bp);
322	}
323	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
324		goto out;
325	xchk_buffer_recheck(sc, bp);
326
327	/* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
328
329	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
330		goto out_buf;
331
332	/* Do the bestfrees correspond to actual free space? */
333	bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
334	smallest_bestfree = UINT_MAX;
335	for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
336		offset = be16_to_cpu(dfp->offset);
337		if (offset == 0)
338			continue;
339		if (offset >= mp->m_dir_geo->blksize) {
340			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
341			goto out_buf;
342		}
343		dup = bp->b_addr + offset;
344		tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
345
346		/* bestfree doesn't match the entry it points at? */
347		if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
348		    be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
349		    tag != offset) {
350			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
351			goto out_buf;
352		}
353
354		/* bestfree records should be ordered largest to smallest */
355		if (smallest_bestfree < be16_to_cpu(dfp->length)) {
356			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
357			goto out_buf;
358		}
359
360		smallest_bestfree = be16_to_cpu(dfp->length);
361		nr_bestfrees++;
362	}
363
364	/* Make sure the bestfrees are actually the best free spaces. */
365	offset = mp->m_dir_geo->data_entry_offset;
366	end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
367
368	/* Iterate the entries, stopping when we hit or go past the end. */
369	while (offset < end) {
370		dup = bp->b_addr + offset;
371
372		/* Skip real entries */
373		if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
374			struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
375
376			newlen = xfs_dir2_data_entsize(mp, dep->namelen);
377			if (newlen <= 0) {
378				xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
379						lblk);
380				goto out_buf;
381			}
382			offset += newlen;
383			continue;
384		}
385
386		/* Spot check this free entry */
387		tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
388		if (tag != offset) {
389			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
390			goto out_buf;
391		}
392
393		/*
394		 * Either this entry is a bestfree or it's smaller than
395		 * any of the bestfrees.
396		 */
397		xchk_directory_check_free_entry(sc, lblk, bf, dup);
398		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
399			goto out_buf;
400
401		/* Move on. */
402		newlen = be16_to_cpu(dup->length);
403		if (newlen <= 0) {
404			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
405			goto out_buf;
406		}
407		offset += newlen;
408		if (offset <= end)
409			nr_frees++;
410	}
411
412	/* We're required to fill all the space. */
413	if (offset != end)
414		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
415
416	/* Did we see at least as many free slots as there are bestfrees? */
417	if (nr_frees < nr_bestfrees)
418		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
419out_buf:
420	xfs_trans_brelse(sc->tp, bp);
421out:
422	return error;
423}
424
425/*
426 * Does the free space length in the free space index block ($len) match
427 * the longest length in the directory data block's bestfree array?
428 * Assume that we've already checked that the data block's bestfree
429 * array is in order.
430 */
431STATIC void
432xchk_directory_check_freesp(
433	struct xfs_scrub		*sc,
434	xfs_dablk_t			lblk,
435	struct xfs_buf			*dbp,
436	unsigned int			len)
437{
438	struct xfs_dir2_data_free	*dfp;
439
440	dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
441
442	if (len != be16_to_cpu(dfp->length))
443		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
444
445	if (len > 0 && be16_to_cpu(dfp->offset) == 0)
446		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
447}
448
449/* Check free space info in a directory leaf1 block. */
450STATIC int
451xchk_directory_leaf1_bestfree(
452	struct xfs_scrub		*sc,
453	struct xfs_da_args		*args,
454	xfs_dir2_db_t			last_data_db,
455	xfs_dablk_t			lblk)
456{
457	struct xfs_dir3_icleaf_hdr	leafhdr;
458	struct xfs_dir2_leaf_tail	*ltp;
459	struct xfs_dir2_leaf		*leaf;
460	struct xfs_buf			*dbp;
461	struct xfs_buf			*bp;
462	struct xfs_da_geometry		*geo = sc->mp->m_dir_geo;
463	__be16				*bestp;
464	__u16				best;
465	__u32				hash;
466	__u32				lasthash = 0;
467	__u32				bestcount;
468	unsigned int			stale = 0;
469	int				i;
470	int				error;
471
472	/* Read the free space block. */
473	error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, &bp);
474	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
475		return error;
476	xchk_buffer_recheck(sc, bp);
477
478	leaf = bp->b_addr;
479	xfs_dir2_leaf_hdr_from_disk(sc->ip->i_mount, &leafhdr, leaf);
480	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
481	bestcount = be32_to_cpu(ltp->bestcount);
482	bestp = xfs_dir2_leaf_bests_p(ltp);
483
484	if (xfs_has_crc(sc->mp)) {
485		struct xfs_dir3_leaf_hdr	*hdr3 = bp->b_addr;
486
487		if (hdr3->pad != cpu_to_be32(0))
488			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
489	}
490
491	/*
492	 * There must be enough bestfree slots to cover all the directory data
493	 * blocks that we scanned.  It is possible for there to be a hole
494	 * between the last data block and i_disk_size.  This seems like an
495	 * oversight to the scrub author, but as we have been writing out
496	 * directories like this (and xfs_repair doesn't mind them) for years,
497	 * that's what we have to check.
498	 */
499	if (bestcount != last_data_db + 1) {
500		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
501		goto out;
502	}
503
504	/* Is the leaf count even remotely sane? */
505	if (leafhdr.count > geo->leaf_max_ents) {
506		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
507		goto out;
508	}
509
510	/* Leaves and bests don't overlap in leaf format. */
511	if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
512		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
513		goto out;
514	}
515
516	/* Check hash value order, count stale entries.  */
517	for (i = 0; i < leafhdr.count; i++) {
518		hash = be32_to_cpu(leafhdr.ents[i].hashval);
519		if (i > 0 && lasthash > hash)
520			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
521		lasthash = hash;
522		if (leafhdr.ents[i].address ==
523		    cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
524			stale++;
525	}
526	if (leafhdr.stale != stale)
527		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
528	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
529		goto out;
530
531	/* Check all the bestfree entries. */
532	for (i = 0; i < bestcount; i++, bestp++) {
533		best = be16_to_cpu(*bestp);
534		error = xfs_dir3_data_read(sc->tp, sc->ip,
535				xfs_dir2_db_to_da(args->geo, i),
536				XFS_DABUF_MAP_HOLE_OK,
537				&dbp);
538		if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
539				&error))
540			break;
541
542		if (!dbp) {
543			if (best != NULLDATAOFF) {
544				xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
545						lblk);
546				break;
547			}
548			continue;
549		}
550
551		if (best == NULLDATAOFF)
552			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
553		else
554			xchk_directory_check_freesp(sc, lblk, dbp, best);
555		xfs_trans_brelse(sc->tp, dbp);
556		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
557			break;
558	}
559out:
560	xfs_trans_brelse(sc->tp, bp);
561	return error;
562}
563
564/* Check free space info in a directory freespace block. */
565STATIC int
566xchk_directory_free_bestfree(
567	struct xfs_scrub		*sc,
568	struct xfs_da_args		*args,
569	xfs_dablk_t			lblk)
570{
571	struct xfs_dir3_icfree_hdr	freehdr;
572	struct xfs_buf			*dbp;
573	struct xfs_buf			*bp;
574	__u16				best;
575	unsigned int			stale = 0;
576	int				i;
577	int				error;
578
579	/* Read the free space block */
580	error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
581	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
582		return error;
583	xchk_buffer_recheck(sc, bp);
584
585	if (xfs_has_crc(sc->mp)) {
586		struct xfs_dir3_free_hdr	*hdr3 = bp->b_addr;
587
588		if (hdr3->pad != cpu_to_be32(0))
589			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
590	}
591
592	/* Check all the entries. */
593	xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
594	for (i = 0; i < freehdr.nvalid; i++) {
595		best = be16_to_cpu(freehdr.bests[i]);
596		if (best == NULLDATAOFF) {
597			stale++;
598			continue;
599		}
600		error = xfs_dir3_data_read(sc->tp, sc->ip,
601				(freehdr.firstdb + i) * args->geo->fsbcount,
602				0, &dbp);
603		if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
604				&error))
605			goto out;
606		xchk_directory_check_freesp(sc, lblk, dbp, best);
607		xfs_trans_brelse(sc->tp, dbp);
608	}
609
610	if (freehdr.nused + stale != freehdr.nvalid)
611		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
612out:
613	xfs_trans_brelse(sc->tp, bp);
614	return error;
615}
616
617/* Check free space information in directories. */
618STATIC int
619xchk_directory_blocks(
620	struct xfs_scrub	*sc)
621{
622	struct xfs_bmbt_irec	got;
623	struct xfs_da_args	args = {
624		.dp		= sc ->ip,
625		.whichfork	= XFS_DATA_FORK,
626		.geo		= sc->mp->m_dir_geo,
627		.trans		= sc->tp,
628	};
629	struct xfs_ifork	*ifp = xfs_ifork_ptr(sc->ip, XFS_DATA_FORK);
630	struct xfs_mount	*mp = sc->mp;
631	xfs_fileoff_t		leaf_lblk;
632	xfs_fileoff_t		free_lblk;
633	xfs_fileoff_t		lblk;
634	struct xfs_iext_cursor	icur;
635	xfs_dablk_t		dabno;
636	xfs_dir2_db_t		last_data_db = 0;
637	bool			found;
638	bool			is_block = false;
639	int			error;
640
641	/* Ignore local format directories. */
642	if (ifp->if_format != XFS_DINODE_FMT_EXTENTS &&
643	    ifp->if_format != XFS_DINODE_FMT_BTREE)
644		return 0;
645
646	lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
647	leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
648	free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
649
650	/* Is this a block dir? */
651	error = xfs_dir2_isblock(&args, &is_block);
652	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
653		goto out;
654
655	/* Iterate all the data extents in the directory... */
656	found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
657	while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
658		/* No more data blocks... */
659		if (got.br_startoff >= leaf_lblk)
660			break;
661
662		/*
663		 * Check each data block's bestfree data.
664		 *
665		 * Iterate all the fsbcount-aligned block offsets in
666		 * this directory.  The directory block reading code is
667		 * smart enough to do its own bmap lookups to handle
668		 * discontiguous directory blocks.  When we're done
669		 * with the extent record, re-query the bmap at the
670		 * next fsbcount-aligned offset to avoid redundant
671		 * block checks.
672		 */
673		for (lblk = roundup((xfs_dablk_t)got.br_startoff,
674				args.geo->fsbcount);
675		     lblk < got.br_startoff + got.br_blockcount;
676		     lblk += args.geo->fsbcount) {
677			last_data_db = xfs_dir2_da_to_db(args.geo, lblk);
678			error = xchk_directory_data_bestfree(sc, lblk,
679					is_block);
680			if (error)
681				goto out;
682		}
683		dabno = got.br_startoff + got.br_blockcount;
684		lblk = roundup(dabno, args.geo->fsbcount);
685		found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
686	}
687
688	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
689		goto out;
690
691	/* Look for a leaf1 block, which has free info. */
692	if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
693	    got.br_startoff == leaf_lblk &&
694	    got.br_blockcount == args.geo->fsbcount &&
695	    !xfs_iext_next_extent(ifp, &icur, &got)) {
696		if (is_block) {
697			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
698			goto out;
699		}
700		error = xchk_directory_leaf1_bestfree(sc, &args, last_data_db,
701				leaf_lblk);
702		if (error)
703			goto out;
704	}
705
706	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
707		goto out;
708
709	/* Scan for free blocks */
710	lblk = free_lblk;
711	found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
712	while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
713		/*
714		 * Dirs can't have blocks mapped above 2^32.
715		 * Single-block dirs shouldn't even be here.
716		 */
717		lblk = got.br_startoff;
718		if (lblk & ~0xFFFFFFFFULL) {
719			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
720			goto out;
721		}
722		if (is_block) {
723			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
724			goto out;
725		}
726
727		/*
728		 * Check each dir free block's bestfree data.
729		 *
730		 * Iterate all the fsbcount-aligned block offsets in
731		 * this directory.  The directory block reading code is
732		 * smart enough to do its own bmap lookups to handle
733		 * discontiguous directory blocks.  When we're done
734		 * with the extent record, re-query the bmap at the
735		 * next fsbcount-aligned offset to avoid redundant
736		 * block checks.
737		 */
738		for (lblk = roundup((xfs_dablk_t)got.br_startoff,
739				args.geo->fsbcount);
740		     lblk < got.br_startoff + got.br_blockcount;
741		     lblk += args.geo->fsbcount) {
742			error = xchk_directory_free_bestfree(sc, &args,
743					lblk);
744			if (error)
745				goto out;
746		}
747		dabno = got.br_startoff + got.br_blockcount;
748		lblk = roundup(dabno, args.geo->fsbcount);
749		found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
750	}
751out:
752	return error;
753}
754
755/* Scrub a whole directory. */
756int
757xchk_directory(
758	struct xfs_scrub	*sc)
759{
760	int			error;
761
762	if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
763		return -ENOENT;
764
765	if (xchk_file_looks_zapped(sc, XFS_SICK_INO_DIR_ZAPPED)) {
766		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
767		return 0;
768	}
769
770	/* Plausible size? */
771	if (sc->ip->i_disk_size < xfs_dir2_sf_hdr_size(0)) {
772		xchk_ino_set_corrupt(sc, sc->ip->i_ino);
773		return 0;
774	}
775
776	/* Check directory tree structure */
777	error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
778	if (error)
779		return error;
780
781	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
782		return 0;
783
784	/* Check the freespace. */
785	error = xchk_directory_blocks(sc);
786	if (error)
787		return error;
788
789	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
790		return 0;
791
792	/* Look up every name in this directory by hash. */
793	error = xchk_dir_walk(sc, sc->ip, xchk_dir_actor, NULL);
794	if (error && error != -ECANCELED)
795		return error;
796
797	/* If the dir is clean, it is clearly not zapped. */
798	xchk_mark_healthy_if_clean(sc, XFS_SICK_INO_DIR_ZAPPED);
799	return 0;
800}
801
802/*
803 * Decide if this directory has been zapped to satisfy the inode and ifork
804 * verifiers.  Checking and repairing should be postponed until the directory
805 * is fixed.
806 */
807bool
808xchk_dir_looks_zapped(
809	struct xfs_inode	*dp)
810{
811	/* Repair zapped this dir's data fork a short time ago */
812	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
813		return true;
814
815	/*
816	 * If the dinode repair found a bad data fork, it will reset the fork
817	 * to extents format with zero records and wait for the bmapbtd
818	 * scrubber to reconstruct the block mappings.  Directories always
819	 * contain some content, so this is a clear sign of a zapped directory.
820	 * The state checked by xfs_ifork_zapped is not persisted, so this is
821	 * the secondary strategy if repairs are interrupted by a crash or an
822	 * unmount.
823	 */
824	return dp->i_df.if_format == XFS_DINODE_FMT_EXTENTS &&
825	       dp->i_df.if_nextents == 0;
826}
827