xref: /linux/fs/xfs/scrub/attr.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
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_inode.h"
14 #include "xfs_da_format.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_attr.h"
17 #include "xfs_attr_leaf.h"
18 #include "scrub/scrub.h"
19 #include "scrub/common.h"
20 #include "scrub/dabtree.h"
21 #include "scrub/attr.h"
22 
23 /*
24  * Allocate enough memory to hold an attr value and attr block bitmaps,
25  * reallocating the buffer if necessary.  Buffer contents are not preserved
26  * across a reallocation.
27  */
28 int
29 xchk_setup_xattr_buf(
30 	struct xfs_scrub	*sc,
31 	size_t			value_size,
32 	xfs_km_flags_t		flags)
33 {
34 	size_t			sz;
35 	struct xchk_xattr_buf	*ab = sc->buf;
36 
37 	/*
38 	 * We need enough space to read an xattr value from the file or enough
39 	 * space to hold three copies of the xattr free space bitmap.  We don't
40 	 * need the buffer space for both purposes at the same time.
41 	 */
42 	sz = 3 * sizeof(long) * BITS_TO_LONGS(sc->mp->m_attr_geo->blksize);
43 	sz = max_t(size_t, sz, value_size);
44 
45 	/*
46 	 * If there's already a buffer, figure out if we need to reallocate it
47 	 * to accommodate a larger size.
48 	 */
49 	if (ab) {
50 		if (sz <= ab->sz)
51 			return 0;
52 		kmem_free(ab);
53 		sc->buf = NULL;
54 	}
55 
56 	/*
57 	 * Don't zero the buffer upon allocation to avoid runtime overhead.
58 	 * All users must be careful never to read uninitialized contents.
59 	 */
60 	ab = kmem_alloc_large(sizeof(*ab) + sz, flags);
61 	if (!ab)
62 		return -ENOMEM;
63 
64 	ab->sz = sz;
65 	sc->buf = ab;
66 	return 0;
67 }
68 
69 /* Set us up to scrub an inode's extended attributes. */
70 int
71 xchk_setup_xattr(
72 	struct xfs_scrub	*sc)
73 {
74 	int			error;
75 
76 	/*
77 	 * We failed to get memory while checking attrs, so this time try to
78 	 * get all the memory we're ever going to need.  Allocate the buffer
79 	 * without the inode lock held, which means we can sleep.
80 	 */
81 	if (sc->flags & XCHK_TRY_HARDER) {
82 		error = xchk_setup_xattr_buf(sc, XATTR_SIZE_MAX, 0);
83 		if (error)
84 			return error;
85 	}
86 
87 	return xchk_setup_inode_contents(sc, 0);
88 }
89 
90 /* Extended Attributes */
91 
92 struct xchk_xattr {
93 	struct xfs_attr_list_context	context;
94 	struct xfs_scrub		*sc;
95 };
96 
97 /*
98  * Check that an extended attribute key can be looked up by hash.
99  *
100  * We use the XFS attribute list iterator (i.e. xfs_attr_list_ilocked)
101  * to call this function for every attribute key in an inode.  Once
102  * we're here, we load the attribute value to see if any errors happen,
103  * or if we get more or less data than we expected.
104  */
105 static void
106 xchk_xattr_listent(
107 	struct xfs_attr_list_context	*context,
108 	int				flags,
109 	unsigned char			*name,
110 	int				namelen,
111 	int				valuelen)
112 {
113 	struct xchk_xattr		*sx;
114 	struct xfs_da_args		args = { NULL };
115 	int				error = 0;
116 
117 	sx = container_of(context, struct xchk_xattr, context);
118 
119 	if (xchk_should_terminate(sx->sc, &error)) {
120 		context->seen_enough = error;
121 		return;
122 	}
123 
124 	if (flags & XFS_ATTR_INCOMPLETE) {
125 		/* Incomplete attr key, just mark the inode for preening. */
126 		xchk_ino_set_preen(sx->sc, context->dp->i_ino);
127 		return;
128 	}
129 
130 	/* Does this name make sense? */
131 	if (!xfs_attr_namecheck(name, namelen)) {
132 		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK, args.blkno);
133 		return;
134 	}
135 
136 	/*
137 	 * Try to allocate enough memory to extrat the attr value.  If that
138 	 * doesn't work, we overload the seen_enough variable to convey
139 	 * the error message back to the main scrub function.
140 	 */
141 	error = xchk_setup_xattr_buf(sx->sc, valuelen, KM_MAYFAIL);
142 	if (error == -ENOMEM)
143 		error = -EDEADLOCK;
144 	if (error) {
145 		context->seen_enough = error;
146 		return;
147 	}
148 
149 	args.op_flags = XFS_DA_OP_NOTIME;
150 	args.attr_filter = flags & XFS_ATTR_NSP_ONDISK_MASK;
151 	args.geo = context->dp->i_mount->m_attr_geo;
152 	args.whichfork = XFS_ATTR_FORK;
153 	args.dp = context->dp;
154 	args.name = name;
155 	args.namelen = namelen;
156 	args.hashval = xfs_da_hashname(args.name, args.namelen);
157 	args.trans = context->tp;
158 	args.value = xchk_xattr_valuebuf(sx->sc);
159 	args.valuelen = valuelen;
160 
161 	error = xfs_attr_get_ilocked(&args);
162 	/* ENODATA means the hash lookup failed and the attr is bad */
163 	if (error == -ENODATA)
164 		error = -EFSCORRUPTED;
165 	if (!xchk_fblock_process_error(sx->sc, XFS_ATTR_FORK, args.blkno,
166 			&error))
167 		goto fail_xref;
168 	if (args.valuelen != valuelen)
169 		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK,
170 					     args.blkno);
171 fail_xref:
172 	if (sx->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
173 		context->seen_enough = 1;
174 	return;
175 }
176 
177 /*
178  * Mark a range [start, start+len) in this map.  Returns true if the
179  * region was free, and false if there's a conflict or a problem.
180  *
181  * Within a char, the lowest bit of the char represents the byte with
182  * the smallest address
183  */
184 STATIC bool
185 xchk_xattr_set_map(
186 	struct xfs_scrub	*sc,
187 	unsigned long		*map,
188 	unsigned int		start,
189 	unsigned int		len)
190 {
191 	unsigned int		mapsize = sc->mp->m_attr_geo->blksize;
192 	bool			ret = true;
193 
194 	if (start >= mapsize)
195 		return false;
196 	if (start + len > mapsize) {
197 		len = mapsize - start;
198 		ret = false;
199 	}
200 
201 	if (find_next_bit(map, mapsize, start) < start + len)
202 		ret = false;
203 	bitmap_set(map, start, len);
204 
205 	return ret;
206 }
207 
208 /*
209  * Check the leaf freemap from the usage bitmap.  Returns false if the
210  * attr freemap has problems or points to used space.
211  */
212 STATIC bool
213 xchk_xattr_check_freemap(
214 	struct xfs_scrub		*sc,
215 	unsigned long			*map,
216 	struct xfs_attr3_icleaf_hdr	*leafhdr)
217 {
218 	unsigned long			*freemap = xchk_xattr_freemap(sc);
219 	unsigned long			*dstmap = xchk_xattr_dstmap(sc);
220 	unsigned int			mapsize = sc->mp->m_attr_geo->blksize;
221 	int				i;
222 
223 	/* Construct bitmap of freemap contents. */
224 	bitmap_zero(freemap, mapsize);
225 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
226 		if (!xchk_xattr_set_map(sc, freemap,
227 				leafhdr->freemap[i].base,
228 				leafhdr->freemap[i].size))
229 			return false;
230 	}
231 
232 	/* Look for bits that are set in freemap and are marked in use. */
233 	return bitmap_and(dstmap, freemap, map, mapsize) == 0;
234 }
235 
236 /*
237  * Check this leaf entry's relations to everything else.
238  * Returns the number of bytes used for the name/value data.
239  */
240 STATIC void
241 xchk_xattr_entry(
242 	struct xchk_da_btree		*ds,
243 	int				level,
244 	char				*buf_end,
245 	struct xfs_attr_leafblock	*leaf,
246 	struct xfs_attr3_icleaf_hdr	*leafhdr,
247 	struct xfs_attr_leaf_entry	*ent,
248 	int				idx,
249 	unsigned int			*usedbytes,
250 	__u32				*last_hashval)
251 {
252 	struct xfs_mount		*mp = ds->state->mp;
253 	unsigned long			*usedmap = xchk_xattr_usedmap(ds->sc);
254 	char				*name_end;
255 	struct xfs_attr_leaf_name_local	*lentry;
256 	struct xfs_attr_leaf_name_remote *rentry;
257 	unsigned int			nameidx;
258 	unsigned int			namesize;
259 
260 	if (ent->pad2 != 0)
261 		xchk_da_set_corrupt(ds, level);
262 
263 	/* Hash values in order? */
264 	if (be32_to_cpu(ent->hashval) < *last_hashval)
265 		xchk_da_set_corrupt(ds, level);
266 	*last_hashval = be32_to_cpu(ent->hashval);
267 
268 	nameidx = be16_to_cpu(ent->nameidx);
269 	if (nameidx < leafhdr->firstused ||
270 	    nameidx >= mp->m_attr_geo->blksize) {
271 		xchk_da_set_corrupt(ds, level);
272 		return;
273 	}
274 
275 	/* Check the name information. */
276 	if (ent->flags & XFS_ATTR_LOCAL) {
277 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
278 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
279 				be16_to_cpu(lentry->valuelen));
280 		name_end = (char *)lentry + namesize;
281 		if (lentry->namelen == 0)
282 			xchk_da_set_corrupt(ds, level);
283 	} else {
284 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
285 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
286 		name_end = (char *)rentry + namesize;
287 		if (rentry->namelen == 0 || rentry->valueblk == 0)
288 			xchk_da_set_corrupt(ds, level);
289 	}
290 	if (name_end > buf_end)
291 		xchk_da_set_corrupt(ds, level);
292 
293 	if (!xchk_xattr_set_map(ds->sc, usedmap, nameidx, namesize))
294 		xchk_da_set_corrupt(ds, level);
295 	if (!(ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
296 		*usedbytes += namesize;
297 }
298 
299 /* Scrub an attribute leaf. */
300 STATIC int
301 xchk_xattr_block(
302 	struct xchk_da_btree		*ds,
303 	int				level)
304 {
305 	struct xfs_attr3_icleaf_hdr	leafhdr;
306 	struct xfs_mount		*mp = ds->state->mp;
307 	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
308 	struct xfs_buf			*bp = blk->bp;
309 	xfs_dablk_t			*last_checked = ds->private;
310 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
311 	struct xfs_attr_leaf_entry	*ent;
312 	struct xfs_attr_leaf_entry	*entries;
313 	unsigned long			*usedmap;
314 	char				*buf_end;
315 	size_t				off;
316 	__u32				last_hashval = 0;
317 	unsigned int			usedbytes = 0;
318 	unsigned int			hdrsize;
319 	int				i;
320 	int				error;
321 
322 	if (*last_checked == blk->blkno)
323 		return 0;
324 
325 	/* Allocate memory for block usage checking. */
326 	error = xchk_setup_xattr_buf(ds->sc, 0, KM_MAYFAIL);
327 	if (error == -ENOMEM)
328 		return -EDEADLOCK;
329 	if (error)
330 		return error;
331 	usedmap = xchk_xattr_usedmap(ds->sc);
332 
333 	*last_checked = blk->blkno;
334 	bitmap_zero(usedmap, mp->m_attr_geo->blksize);
335 
336 	/* Check all the padding. */
337 	if (xfs_sb_version_hascrc(&ds->sc->mp->m_sb)) {
338 		struct xfs_attr3_leafblock	*leaf = bp->b_addr;
339 
340 		if (leaf->hdr.pad1 != 0 || leaf->hdr.pad2 != 0 ||
341 		    leaf->hdr.info.hdr.pad != 0)
342 			xchk_da_set_corrupt(ds, level);
343 	} else {
344 		if (leaf->hdr.pad1 != 0 || leaf->hdr.info.pad != 0)
345 			xchk_da_set_corrupt(ds, level);
346 	}
347 
348 	/* Check the leaf header */
349 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
350 	hdrsize = xfs_attr3_leaf_hdr_size(leaf);
351 
352 	if (leafhdr.usedbytes > mp->m_attr_geo->blksize)
353 		xchk_da_set_corrupt(ds, level);
354 	if (leafhdr.firstused > mp->m_attr_geo->blksize)
355 		xchk_da_set_corrupt(ds, level);
356 	if (leafhdr.firstused < hdrsize)
357 		xchk_da_set_corrupt(ds, level);
358 	if (!xchk_xattr_set_map(ds->sc, usedmap, 0, hdrsize))
359 		xchk_da_set_corrupt(ds, level);
360 
361 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
362 		goto out;
363 
364 	entries = xfs_attr3_leaf_entryp(leaf);
365 	if ((char *)&entries[leafhdr.count] > (char *)leaf + leafhdr.firstused)
366 		xchk_da_set_corrupt(ds, level);
367 
368 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
369 	for (i = 0, ent = entries; i < leafhdr.count; ent++, i++) {
370 		/* Mark the leaf entry itself. */
371 		off = (char *)ent - (char *)leaf;
372 		if (!xchk_xattr_set_map(ds->sc, usedmap, off,
373 				sizeof(xfs_attr_leaf_entry_t))) {
374 			xchk_da_set_corrupt(ds, level);
375 			goto out;
376 		}
377 
378 		/* Check the entry and nameval. */
379 		xchk_xattr_entry(ds, level, buf_end, leaf, &leafhdr,
380 				ent, i, &usedbytes, &last_hashval);
381 
382 		if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
383 			goto out;
384 	}
385 
386 	if (!xchk_xattr_check_freemap(ds->sc, usedmap, &leafhdr))
387 		xchk_da_set_corrupt(ds, level);
388 
389 	if (leafhdr.usedbytes != usedbytes)
390 		xchk_da_set_corrupt(ds, level);
391 
392 out:
393 	return 0;
394 }
395 
396 /* Scrub a attribute btree record. */
397 STATIC int
398 xchk_xattr_rec(
399 	struct xchk_da_btree		*ds,
400 	int				level)
401 {
402 	struct xfs_mount		*mp = ds->state->mp;
403 	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
404 	struct xfs_attr_leaf_name_local	*lentry;
405 	struct xfs_attr_leaf_name_remote	*rentry;
406 	struct xfs_buf			*bp;
407 	struct xfs_attr_leaf_entry	*ent;
408 	xfs_dahash_t			calc_hash;
409 	xfs_dahash_t			hash;
410 	int				nameidx;
411 	int				hdrsize;
412 	unsigned int			badflags;
413 	int				error;
414 
415 	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
416 
417 	ent = xfs_attr3_leaf_entryp(blk->bp->b_addr) + blk->index;
418 
419 	/* Check the whole block, if necessary. */
420 	error = xchk_xattr_block(ds, level);
421 	if (error)
422 		goto out;
423 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
424 		goto out;
425 
426 	/* Check the hash of the entry. */
427 	error = xchk_da_btree_hash(ds, level, &ent->hashval);
428 	if (error)
429 		goto out;
430 
431 	/* Find the attr entry's location. */
432 	bp = blk->bp;
433 	hdrsize = xfs_attr3_leaf_hdr_size(bp->b_addr);
434 	nameidx = be16_to_cpu(ent->nameidx);
435 	if (nameidx < hdrsize || nameidx >= mp->m_attr_geo->blksize) {
436 		xchk_da_set_corrupt(ds, level);
437 		goto out;
438 	}
439 
440 	/* Retrieve the entry and check it. */
441 	hash = be32_to_cpu(ent->hashval);
442 	badflags = ~(XFS_ATTR_LOCAL | XFS_ATTR_ROOT | XFS_ATTR_SECURE |
443 			XFS_ATTR_INCOMPLETE);
444 	if ((ent->flags & badflags) != 0)
445 		xchk_da_set_corrupt(ds, level);
446 	if (ent->flags & XFS_ATTR_LOCAL) {
447 		lentry = (struct xfs_attr_leaf_name_local *)
448 				(((char *)bp->b_addr) + nameidx);
449 		if (lentry->namelen <= 0) {
450 			xchk_da_set_corrupt(ds, level);
451 			goto out;
452 		}
453 		calc_hash = xfs_da_hashname(lentry->nameval, lentry->namelen);
454 	} else {
455 		rentry = (struct xfs_attr_leaf_name_remote *)
456 				(((char *)bp->b_addr) + nameidx);
457 		if (rentry->namelen <= 0) {
458 			xchk_da_set_corrupt(ds, level);
459 			goto out;
460 		}
461 		calc_hash = xfs_da_hashname(rentry->name, rentry->namelen);
462 	}
463 	if (calc_hash != hash)
464 		xchk_da_set_corrupt(ds, level);
465 
466 out:
467 	return error;
468 }
469 
470 /* Scrub the extended attribute metadata. */
471 int
472 xchk_xattr(
473 	struct xfs_scrub		*sc)
474 {
475 	struct xchk_xattr		sx;
476 	xfs_dablk_t			last_checked = -1U;
477 	int				error = 0;
478 
479 	if (!xfs_inode_hasattr(sc->ip))
480 		return -ENOENT;
481 
482 	memset(&sx, 0, sizeof(sx));
483 	/* Check attribute tree structure */
484 	error = xchk_da_btree(sc, XFS_ATTR_FORK, xchk_xattr_rec,
485 			&last_checked);
486 	if (error)
487 		goto out;
488 
489 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
490 		goto out;
491 
492 	/* Check that every attr key can also be looked up by hash. */
493 	sx.context.dp = sc->ip;
494 	sx.context.resynch = 1;
495 	sx.context.put_listent = xchk_xattr_listent;
496 	sx.context.tp = sc->tp;
497 	sx.context.allow_incomplete = true;
498 	sx.sc = sc;
499 
500 	/*
501 	 * Look up every xattr in this file by name.
502 	 *
503 	 * Use the backend implementation of xfs_attr_list to call
504 	 * xchk_xattr_listent on every attribute key in this inode.
505 	 * In other words, we use the same iterator/callback mechanism
506 	 * that listattr uses to scrub extended attributes, though in our
507 	 * _listent function, we check the value of the attribute.
508 	 *
509 	 * The VFS only locks i_rwsem when modifying attrs, so keep all
510 	 * three locks held because that's the only way to ensure we're
511 	 * the only thread poking into the da btree.  We traverse the da
512 	 * btree while holding a leaf buffer locked for the xattr name
513 	 * iteration, which doesn't really follow the usual buffer
514 	 * locking order.
515 	 */
516 	error = xfs_attr_list_ilocked(&sx.context);
517 	if (!xchk_fblock_process_error(sc, XFS_ATTR_FORK, 0, &error))
518 		goto out;
519 
520 	/* Did our listent function try to return any errors? */
521 	if (sx.context.seen_enough < 0)
522 		error = sx.context.seen_enough;
523 out:
524 	return error;
525 }
526