xref: /linux/fs/xfs/scrub/dir.c (revision 1f2367a39f17bd553a75e179a747f9b257bc9478)
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_defer.h"
13 #include "xfs_btree.h"
14 #include "xfs_bit.h"
15 #include "xfs_log_format.h"
16 #include "xfs_trans.h"
17 #include "xfs_sb.h"
18 #include "xfs_inode.h"
19 #include "xfs_icache.h"
20 #include "xfs_itable.h"
21 #include "xfs_da_format.h"
22 #include "xfs_da_btree.h"
23 #include "xfs_dir2.h"
24 #include "xfs_dir2_priv.h"
25 #include "xfs_ialloc.h"
26 #include "scrub/xfs_scrub.h"
27 #include "scrub/scrub.h"
28 #include "scrub/common.h"
29 #include "scrub/trace.h"
30 #include "scrub/dabtree.h"
31 
32 /* Set us up to scrub directories. */
33 int
34 xchk_setup_directory(
35 	struct xfs_scrub	*sc,
36 	struct xfs_inode	*ip)
37 {
38 	return xchk_setup_inode_contents(sc, ip, 0);
39 }
40 
41 /* Directories */
42 
43 /* Scrub a directory entry. */
44 
45 struct xchk_dir_ctx {
46 	/* VFS fill-directory iterator */
47 	struct dir_context	dir_iter;
48 
49 	struct xfs_scrub	*sc;
50 };
51 
52 /* Check that an inode's mode matches a given DT_ type. */
53 STATIC int
54 xchk_dir_check_ftype(
55 	struct xchk_dir_ctx	*sdc,
56 	xfs_fileoff_t		offset,
57 	xfs_ino_t		inum,
58 	int			dtype)
59 {
60 	struct xfs_mount	*mp = sdc->sc->mp;
61 	struct xfs_inode	*ip;
62 	int			ino_dtype;
63 	int			error = 0;
64 
65 	if (!xfs_sb_version_hasftype(&mp->m_sb)) {
66 		if (dtype != DT_UNKNOWN && dtype != DT_DIR)
67 			xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
68 					offset);
69 		goto out;
70 	}
71 
72 	/*
73 	 * Grab the inode pointed to by the dirent.  We release the
74 	 * inode before we cancel the scrub transaction.  Since we're
75 	 * don't know a priori that releasing the inode won't trigger
76 	 * eofblocks cleanup (which allocates what would be a nested
77 	 * transaction), we can't use DONTCACHE here because DONTCACHE
78 	 * inodes can trigger immediate inactive cleanup of the inode.
79 	 */
80 	error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
81 	if (!xchk_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
82 			&error))
83 		goto out;
84 
85 	/* Convert mode to the DT_* values that dir_emit uses. */
86 	ino_dtype = xfs_dir3_get_dtype(mp,
87 			xfs_mode_to_ftype(VFS_I(ip)->i_mode));
88 	if (ino_dtype != dtype)
89 		xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
90 	xfs_irele(ip);
91 out:
92 	return error;
93 }
94 
95 /*
96  * Scrub a single directory entry.
97  *
98  * We use the VFS directory iterator (i.e. readdir) to call this
99  * function for every directory entry in a directory.  Once we're here,
100  * we check the inode number to make sure it's sane, then we check that
101  * we can look up this filename.  Finally, we check the ftype.
102  */
103 STATIC int
104 xchk_dir_actor(
105 	struct dir_context	*dir_iter,
106 	const char		*name,
107 	int			namelen,
108 	loff_t			pos,
109 	u64			ino,
110 	unsigned		type)
111 {
112 	struct xfs_mount	*mp;
113 	struct xfs_inode	*ip;
114 	struct xchk_dir_ctx	*sdc;
115 	struct xfs_name		xname;
116 	xfs_ino_t		lookup_ino;
117 	xfs_dablk_t		offset;
118 	int			error = 0;
119 
120 	sdc = container_of(dir_iter, struct xchk_dir_ctx, dir_iter);
121 	ip = sdc->sc->ip;
122 	mp = ip->i_mount;
123 	offset = xfs_dir2_db_to_da(mp->m_dir_geo,
124 			xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
125 
126 	/* Does this inode number make sense? */
127 	if (!xfs_verify_dir_ino(mp, ino)) {
128 		xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
129 		goto out;
130 	}
131 
132 	/* Does this name make sense? */
133 	if (!xfs_dir2_namecheck(name, namelen)) {
134 		xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
135 		goto out;
136 	}
137 
138 	if (!strncmp(".", name, namelen)) {
139 		/* If this is "." then check that the inum matches the dir. */
140 		if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
141 			xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
142 					offset);
143 		if (ino != ip->i_ino)
144 			xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
145 					offset);
146 	} else if (!strncmp("..", name, namelen)) {
147 		/*
148 		 * If this is ".." in the root inode, check that the inum
149 		 * matches this dir.
150 		 */
151 		if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
152 			xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
153 					offset);
154 		if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
155 			xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
156 					offset);
157 	}
158 
159 	/* Verify that we can look up this name by hash. */
160 	xname.name = name;
161 	xname.len = namelen;
162 	xname.type = XFS_DIR3_FT_UNKNOWN;
163 
164 	error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
165 	if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
166 			&error))
167 		goto out;
168 	if (lookup_ino != ino) {
169 		xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
170 		goto out;
171 	}
172 
173 	/* Verify the file type.  This function absorbs error codes. */
174 	error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
175 	if (error)
176 		goto out;
177 out:
178 	/*
179 	 * A negative error code returned here is supposed to cause the
180 	 * dir_emit caller (xfs_readdir) to abort the directory iteration
181 	 * and return zero to xchk_directory.
182 	 */
183 	if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
184 		return -EFSCORRUPTED;
185 	return error;
186 }
187 
188 /* Scrub a directory btree record. */
189 STATIC int
190 xchk_dir_rec(
191 	struct xchk_da_btree		*ds,
192 	int				level,
193 	void				*rec)
194 {
195 	struct xfs_mount		*mp = ds->state->mp;
196 	struct xfs_dir2_leaf_entry	*ent = rec;
197 	struct xfs_inode		*dp = ds->dargs.dp;
198 	struct xfs_dir2_data_entry	*dent;
199 	struct xfs_buf			*bp;
200 	char				*p, *endp;
201 	xfs_ino_t			ino;
202 	xfs_dablk_t			rec_bno;
203 	xfs_dir2_db_t			db;
204 	xfs_dir2_data_aoff_t		off;
205 	xfs_dir2_dataptr_t		ptr;
206 	xfs_dahash_t			calc_hash;
207 	xfs_dahash_t			hash;
208 	unsigned int			tag;
209 	int				error;
210 
211 	/* Check the hash of the entry. */
212 	error = xchk_da_btree_hash(ds, level, &ent->hashval);
213 	if (error)
214 		goto out;
215 
216 	/* Valid hash pointer? */
217 	ptr = be32_to_cpu(ent->address);
218 	if (ptr == 0)
219 		return 0;
220 
221 	/* Find the directory entry's location. */
222 	db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
223 	off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
224 	rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
225 
226 	if (rec_bno >= mp->m_dir_geo->leafblk) {
227 		xchk_da_set_corrupt(ds, level);
228 		goto out;
229 	}
230 	error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
231 	if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
232 			&error))
233 		goto out;
234 	if (!bp) {
235 		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
236 		goto out;
237 	}
238 	xchk_buffer_recheck(ds->sc, bp);
239 
240 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
241 		goto out_relse;
242 
243 	dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
244 
245 	/* Make sure we got a real directory entry. */
246 	p = (char *)mp->m_dir_inode_ops->data_entry_p(bp->b_addr);
247 	endp = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
248 	if (!endp) {
249 		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
250 		goto out_relse;
251 	}
252 	while (p < endp) {
253 		struct xfs_dir2_data_entry	*dep;
254 		struct xfs_dir2_data_unused	*dup;
255 
256 		dup = (struct xfs_dir2_data_unused *)p;
257 		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
258 			p += be16_to_cpu(dup->length);
259 			continue;
260 		}
261 		dep = (struct xfs_dir2_data_entry *)p;
262 		if (dep == dent)
263 			break;
264 		p += mp->m_dir_inode_ops->data_entsize(dep->namelen);
265 	}
266 	if (p >= endp) {
267 		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
268 		goto out_relse;
269 	}
270 
271 	/* Retrieve the entry, sanity check it, and compare hashes. */
272 	ino = be64_to_cpu(dent->inumber);
273 	hash = be32_to_cpu(ent->hashval);
274 	tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
275 	if (!xfs_verify_dir_ino(mp, ino) || tag != off)
276 		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
277 	if (dent->namelen == 0) {
278 		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
279 		goto out_relse;
280 	}
281 	calc_hash = xfs_da_hashname(dent->name, dent->namelen);
282 	if (calc_hash != hash)
283 		xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
284 
285 out_relse:
286 	xfs_trans_brelse(ds->dargs.trans, bp);
287 out:
288 	return error;
289 }
290 
291 /*
292  * Is this unused entry either in the bestfree or smaller than all of
293  * them?  We've already checked that the bestfrees are sorted longest to
294  * shortest, and that there aren't any bogus entries.
295  */
296 STATIC void
297 xchk_directory_check_free_entry(
298 	struct xfs_scrub		*sc,
299 	xfs_dablk_t			lblk,
300 	struct xfs_dir2_data_free	*bf,
301 	struct xfs_dir2_data_unused	*dup)
302 {
303 	struct xfs_dir2_data_free	*dfp;
304 	unsigned int			dup_length;
305 
306 	dup_length = be16_to_cpu(dup->length);
307 
308 	/* Unused entry is shorter than any of the bestfrees */
309 	if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
310 		return;
311 
312 	for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
313 		if (dup_length == be16_to_cpu(dfp->length))
314 			return;
315 
316 	/* Unused entry should be in the bestfrees but wasn't found. */
317 	xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
318 }
319 
320 /* Check free space info in a directory data block. */
321 STATIC int
322 xchk_directory_data_bestfree(
323 	struct xfs_scrub		*sc,
324 	xfs_dablk_t			lblk,
325 	bool				is_block)
326 {
327 	struct xfs_dir2_data_unused	*dup;
328 	struct xfs_dir2_data_free	*dfp;
329 	struct xfs_buf			*bp;
330 	struct xfs_dir2_data_free	*bf;
331 	struct xfs_mount		*mp = sc->mp;
332 	const struct xfs_dir_ops	*d_ops;
333 	char				*ptr;
334 	char				*endptr;
335 	u16				tag;
336 	unsigned int			nr_bestfrees = 0;
337 	unsigned int			nr_frees = 0;
338 	unsigned int			smallest_bestfree;
339 	int				newlen;
340 	int				offset;
341 	int				error;
342 
343 	d_ops = sc->ip->d_ops;
344 
345 	if (is_block) {
346 		/* dir block format */
347 		if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
348 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
349 		error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
350 	} else {
351 		/* dir data format */
352 		error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
353 	}
354 	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
355 		goto out;
356 	xchk_buffer_recheck(sc, bp);
357 
358 	/* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
359 
360 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
361 		goto out_buf;
362 
363 	/* Do the bestfrees correspond to actual free space? */
364 	bf = d_ops->data_bestfree_p(bp->b_addr);
365 	smallest_bestfree = UINT_MAX;
366 	for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
367 		offset = be16_to_cpu(dfp->offset);
368 		if (offset == 0)
369 			continue;
370 		if (offset >= mp->m_dir_geo->blksize) {
371 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
372 			goto out_buf;
373 		}
374 		dup = (struct xfs_dir2_data_unused *)(bp->b_addr + offset);
375 		tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
376 
377 		/* bestfree doesn't match the entry it points at? */
378 		if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
379 		    be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
380 		    tag != ((char *)dup - (char *)bp->b_addr)) {
381 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
382 			goto out_buf;
383 		}
384 
385 		/* bestfree records should be ordered largest to smallest */
386 		if (smallest_bestfree < be16_to_cpu(dfp->length)) {
387 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
388 			goto out_buf;
389 		}
390 
391 		smallest_bestfree = be16_to_cpu(dfp->length);
392 		nr_bestfrees++;
393 	}
394 
395 	/* Make sure the bestfrees are actually the best free spaces. */
396 	ptr = (char *)d_ops->data_entry_p(bp->b_addr);
397 	endptr = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
398 
399 	/* Iterate the entries, stopping when we hit or go past the end. */
400 	while (ptr < endptr) {
401 		dup = (struct xfs_dir2_data_unused *)ptr;
402 		/* Skip real entries */
403 		if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
404 			struct xfs_dir2_data_entry	*dep;
405 
406 			dep = (struct xfs_dir2_data_entry *)ptr;
407 			newlen = d_ops->data_entsize(dep->namelen);
408 			if (newlen <= 0) {
409 				xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
410 						lblk);
411 				goto out_buf;
412 			}
413 			ptr += newlen;
414 			continue;
415 		}
416 
417 		/* Spot check this free entry */
418 		tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
419 		if (tag != ((char *)dup - (char *)bp->b_addr)) {
420 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
421 			goto out_buf;
422 		}
423 
424 		/*
425 		 * Either this entry is a bestfree or it's smaller than
426 		 * any of the bestfrees.
427 		 */
428 		xchk_directory_check_free_entry(sc, lblk, bf, dup);
429 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
430 			goto out_buf;
431 
432 		/* Move on. */
433 		newlen = be16_to_cpu(dup->length);
434 		if (newlen <= 0) {
435 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
436 			goto out_buf;
437 		}
438 		ptr += newlen;
439 		if (ptr <= endptr)
440 			nr_frees++;
441 	}
442 
443 	/* We're required to fill all the space. */
444 	if (ptr != endptr)
445 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
446 
447 	/* Did we see at least as many free slots as there are bestfrees? */
448 	if (nr_frees < nr_bestfrees)
449 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
450 out_buf:
451 	xfs_trans_brelse(sc->tp, bp);
452 out:
453 	return error;
454 }
455 
456 /*
457  * Does the free space length in the free space index block ($len) match
458  * the longest length in the directory data block's bestfree array?
459  * Assume that we've already checked that the data block's bestfree
460  * array is in order.
461  */
462 STATIC void
463 xchk_directory_check_freesp(
464 	struct xfs_scrub		*sc,
465 	xfs_dablk_t			lblk,
466 	struct xfs_buf			*dbp,
467 	unsigned int			len)
468 {
469 	struct xfs_dir2_data_free	*dfp;
470 
471 	dfp = sc->ip->d_ops->data_bestfree_p(dbp->b_addr);
472 
473 	if (len != be16_to_cpu(dfp->length))
474 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
475 
476 	if (len > 0 && be16_to_cpu(dfp->offset) == 0)
477 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
478 }
479 
480 /* Check free space info in a directory leaf1 block. */
481 STATIC int
482 xchk_directory_leaf1_bestfree(
483 	struct xfs_scrub		*sc,
484 	struct xfs_da_args		*args,
485 	xfs_dablk_t			lblk)
486 {
487 	struct xfs_dir3_icleaf_hdr	leafhdr;
488 	struct xfs_dir2_leaf_entry	*ents;
489 	struct xfs_dir2_leaf_tail	*ltp;
490 	struct xfs_dir2_leaf		*leaf;
491 	struct xfs_buf			*dbp;
492 	struct xfs_buf			*bp;
493 	const struct xfs_dir_ops	*d_ops = sc->ip->d_ops;
494 	struct xfs_da_geometry		*geo = sc->mp->m_dir_geo;
495 	__be16				*bestp;
496 	__u16				best;
497 	__u32				hash;
498 	__u32				lasthash = 0;
499 	__u32				bestcount;
500 	unsigned int			stale = 0;
501 	int				i;
502 	int				error;
503 
504 	/* Read the free space block. */
505 	error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
506 	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
507 		goto out;
508 	xchk_buffer_recheck(sc, bp);
509 
510 	leaf = bp->b_addr;
511 	d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
512 	ents = d_ops->leaf_ents_p(leaf);
513 	ltp = xfs_dir2_leaf_tail_p(geo, leaf);
514 	bestcount = be32_to_cpu(ltp->bestcount);
515 	bestp = xfs_dir2_leaf_bests_p(ltp);
516 
517 	if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
518 		struct xfs_dir3_leaf_hdr	*hdr3 = bp->b_addr;
519 
520 		if (hdr3->pad != cpu_to_be32(0))
521 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
522 	}
523 
524 	/*
525 	 * There should be as many bestfree slots as there are dir data
526 	 * blocks that can fit under i_size.
527 	 */
528 	if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
529 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
530 		goto out;
531 	}
532 
533 	/* Is the leaf count even remotely sane? */
534 	if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
535 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
536 		goto out;
537 	}
538 
539 	/* Leaves and bests don't overlap in leaf format. */
540 	if ((char *)&ents[leafhdr.count] > (char *)bestp) {
541 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
542 		goto out;
543 	}
544 
545 	/* Check hash value order, count stale entries.  */
546 	for (i = 0; i < leafhdr.count; i++) {
547 		hash = be32_to_cpu(ents[i].hashval);
548 		if (i > 0 && lasthash > hash)
549 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
550 		lasthash = hash;
551 		if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
552 			stale++;
553 	}
554 	if (leafhdr.stale != stale)
555 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
556 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
557 		goto out;
558 
559 	/* Check all the bestfree entries. */
560 	for (i = 0; i < bestcount; i++, bestp++) {
561 		best = be16_to_cpu(*bestp);
562 		if (best == NULLDATAOFF)
563 			continue;
564 		error = xfs_dir3_data_read(sc->tp, sc->ip,
565 				i * args->geo->fsbcount, -1, &dbp);
566 		if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
567 				&error))
568 			break;
569 		xchk_directory_check_freesp(sc, lblk, dbp, best);
570 		xfs_trans_brelse(sc->tp, dbp);
571 		if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
572 			goto out;
573 	}
574 out:
575 	return error;
576 }
577 
578 /* Check free space info in a directory freespace block. */
579 STATIC int
580 xchk_directory_free_bestfree(
581 	struct xfs_scrub		*sc,
582 	struct xfs_da_args		*args,
583 	xfs_dablk_t			lblk)
584 {
585 	struct xfs_dir3_icfree_hdr	freehdr;
586 	struct xfs_buf			*dbp;
587 	struct xfs_buf			*bp;
588 	__be16				*bestp;
589 	__u16				best;
590 	unsigned int			stale = 0;
591 	int				i;
592 	int				error;
593 
594 	/* Read the free space block */
595 	error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
596 	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
597 		goto out;
598 	xchk_buffer_recheck(sc, bp);
599 
600 	if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
601 		struct xfs_dir3_free_hdr	*hdr3 = bp->b_addr;
602 
603 		if (hdr3->pad != cpu_to_be32(0))
604 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
605 	}
606 
607 	/* Check all the entries. */
608 	sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
609 	bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
610 	for (i = 0; i < freehdr.nvalid; i++, bestp++) {
611 		best = be16_to_cpu(*bestp);
612 		if (best == NULLDATAOFF) {
613 			stale++;
614 			continue;
615 		}
616 		error = xfs_dir3_data_read(sc->tp, sc->ip,
617 				(freehdr.firstdb + i) * args->geo->fsbcount,
618 				-1, &dbp);
619 		if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
620 				&error))
621 			break;
622 		xchk_directory_check_freesp(sc, lblk, dbp, best);
623 		xfs_trans_brelse(sc->tp, dbp);
624 	}
625 
626 	if (freehdr.nused + stale != freehdr.nvalid)
627 		xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
628 out:
629 	return error;
630 }
631 
632 /* Check free space information in directories. */
633 STATIC int
634 xchk_directory_blocks(
635 	struct xfs_scrub	*sc)
636 {
637 	struct xfs_bmbt_irec	got;
638 	struct xfs_da_args	args;
639 	struct xfs_ifork	*ifp;
640 	struct xfs_mount	*mp = sc->mp;
641 	xfs_fileoff_t		leaf_lblk;
642 	xfs_fileoff_t		free_lblk;
643 	xfs_fileoff_t		lblk;
644 	struct xfs_iext_cursor	icur;
645 	xfs_dablk_t		dabno;
646 	bool			found;
647 	int			is_block = 0;
648 	int			error;
649 
650 	/* Ignore local format directories. */
651 	if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
652 	    sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
653 		return 0;
654 
655 	ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
656 	lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
657 	leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
658 	free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
659 
660 	/* Is this a block dir? */
661 	args.dp = sc->ip;
662 	args.geo = mp->m_dir_geo;
663 	args.trans = sc->tp;
664 	error = xfs_dir2_isblock(&args, &is_block);
665 	if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
666 		goto out;
667 
668 	/* Iterate all the data extents in the directory... */
669 	found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
670 	while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
671 		/* Block directories only have a single block at offset 0. */
672 		if (is_block &&
673 		    (got.br_startoff > 0 ||
674 		     got.br_blockcount != args.geo->fsbcount)) {
675 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
676 					got.br_startoff);
677 			break;
678 		}
679 
680 		/* No more data blocks... */
681 		if (got.br_startoff >= leaf_lblk)
682 			break;
683 
684 		/*
685 		 * Check each data block's bestfree data.
686 		 *
687 		 * Iterate all the fsbcount-aligned block offsets in
688 		 * this directory.  The directory block reading code is
689 		 * smart enough to do its own bmap lookups to handle
690 		 * discontiguous directory blocks.  When we're done
691 		 * with the extent record, re-query the bmap at the
692 		 * next fsbcount-aligned offset to avoid redundant
693 		 * block checks.
694 		 */
695 		for (lblk = roundup((xfs_dablk_t)got.br_startoff,
696 				args.geo->fsbcount);
697 		     lblk < got.br_startoff + got.br_blockcount;
698 		     lblk += args.geo->fsbcount) {
699 			error = xchk_directory_data_bestfree(sc, lblk,
700 					is_block);
701 			if (error)
702 				goto out;
703 		}
704 		dabno = got.br_startoff + got.br_blockcount;
705 		lblk = roundup(dabno, args.geo->fsbcount);
706 		found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
707 	}
708 
709 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
710 		goto out;
711 
712 	/* Look for a leaf1 block, which has free info. */
713 	if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
714 	    got.br_startoff == leaf_lblk &&
715 	    got.br_blockcount == args.geo->fsbcount &&
716 	    !xfs_iext_next_extent(ifp, &icur, &got)) {
717 		if (is_block) {
718 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
719 			goto out;
720 		}
721 		error = xchk_directory_leaf1_bestfree(sc, &args,
722 				leaf_lblk);
723 		if (error)
724 			goto out;
725 	}
726 
727 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
728 		goto out;
729 
730 	/* Scan for free blocks */
731 	lblk = free_lblk;
732 	found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
733 	while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
734 		/*
735 		 * Dirs can't have blocks mapped above 2^32.
736 		 * Single-block dirs shouldn't even be here.
737 		 */
738 		lblk = got.br_startoff;
739 		if (lblk & ~0xFFFFFFFFULL) {
740 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
741 			goto out;
742 		}
743 		if (is_block) {
744 			xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
745 			goto out;
746 		}
747 
748 		/*
749 		 * Check each dir free block's bestfree data.
750 		 *
751 		 * Iterate all the fsbcount-aligned block offsets in
752 		 * this directory.  The directory block reading code is
753 		 * smart enough to do its own bmap lookups to handle
754 		 * discontiguous directory blocks.  When we're done
755 		 * with the extent record, re-query the bmap at the
756 		 * next fsbcount-aligned offset to avoid redundant
757 		 * block checks.
758 		 */
759 		for (lblk = roundup((xfs_dablk_t)got.br_startoff,
760 				args.geo->fsbcount);
761 		     lblk < got.br_startoff + got.br_blockcount;
762 		     lblk += args.geo->fsbcount) {
763 			error = xchk_directory_free_bestfree(sc, &args,
764 					lblk);
765 			if (error)
766 				goto out;
767 		}
768 		dabno = got.br_startoff + got.br_blockcount;
769 		lblk = roundup(dabno, args.geo->fsbcount);
770 		found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
771 	}
772 out:
773 	return error;
774 }
775 
776 /* Scrub a whole directory. */
777 int
778 xchk_directory(
779 	struct xfs_scrub	*sc)
780 {
781 	struct xchk_dir_ctx	sdc = {
782 		.dir_iter.actor = xchk_dir_actor,
783 		.dir_iter.pos = 0,
784 		.sc = sc,
785 	};
786 	size_t			bufsize;
787 	loff_t			oldpos;
788 	int			error = 0;
789 
790 	if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
791 		return -ENOENT;
792 
793 	/* Plausible size? */
794 	if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
795 		xchk_ino_set_corrupt(sc, sc->ip->i_ino);
796 		goto out;
797 	}
798 
799 	/* Check directory tree structure */
800 	error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
801 	if (error)
802 		return error;
803 
804 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
805 		return error;
806 
807 	/* Check the freespace. */
808 	error = xchk_directory_blocks(sc);
809 	if (error)
810 		return error;
811 
812 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
813 		return error;
814 
815 	/*
816 	 * Check that every dirent we see can also be looked up by hash.
817 	 * Userspace usually asks for a 32k buffer, so we will too.
818 	 */
819 	bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
820 			sc->ip->i_d.di_size);
821 
822 	/*
823 	 * Look up every name in this directory by hash.
824 	 *
825 	 * Use the xfs_readdir function to call xchk_dir_actor on
826 	 * every directory entry in this directory.  In _actor, we check
827 	 * the name, inode number, and ftype (if applicable) of the
828 	 * entry.  xfs_readdir uses the VFS filldir functions to provide
829 	 * iteration context.
830 	 *
831 	 * The VFS grabs a read or write lock via i_rwsem before it reads
832 	 * or writes to a directory.  If we've gotten this far we've
833 	 * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
834 	 * getting a write lock on i_rwsem.  Therefore, it is safe for us
835 	 * to drop the ILOCK here in order to reuse the _readdir and
836 	 * _dir_lookup routines, which do their own ILOCK locking.
837 	 */
838 	oldpos = 0;
839 	sc->ilock_flags &= ~XFS_ILOCK_EXCL;
840 	xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
841 	while (true) {
842 		error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
843 		if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
844 				&error))
845 			goto out;
846 		if (oldpos == sdc.dir_iter.pos)
847 			break;
848 		oldpos = sdc.dir_iter.pos;
849 	}
850 
851 out:
852 	return error;
853 }
854