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