xref: /openbmc/linux/fs/xfs/scrub/attr.c (revision e7bae9bb)
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 	struct xfs_inode	*ip)
74 {
75 	int			error;
76 
77 	/*
78 	 * We failed to get memory while checking attrs, so this time try to
79 	 * get all the memory we're ever going to need.  Allocate the buffer
80 	 * without the inode lock held, which means we can sleep.
81 	 */
82 	if (sc->flags & XCHK_TRY_HARDER) {
83 		error = xchk_setup_xattr_buf(sc, XATTR_SIZE_MAX, 0);
84 		if (error)
85 			return error;
86 	}
87 
88 	return xchk_setup_inode_contents(sc, ip, 0);
89 }
90 
91 /* Extended Attributes */
92 
93 struct xchk_xattr {
94 	struct xfs_attr_list_context	context;
95 	struct xfs_scrub		*sc;
96 };
97 
98 /*
99  * Check that an extended attribute key can be looked up by hash.
100  *
101  * We use the XFS attribute list iterator (i.e. xfs_attr_list_ilocked)
102  * to call this function for every attribute key in an inode.  Once
103  * we're here, we load the attribute value to see if any errors happen,
104  * or if we get more or less data than we expected.
105  */
106 static void
107 xchk_xattr_listent(
108 	struct xfs_attr_list_context	*context,
109 	int				flags,
110 	unsigned char			*name,
111 	int				namelen,
112 	int				valuelen)
113 {
114 	struct xchk_xattr		*sx;
115 	struct xfs_da_args		args = { NULL };
116 	int				error = 0;
117 
118 	sx = container_of(context, struct xchk_xattr, context);
119 
120 	if (xchk_should_terminate(sx->sc, &error)) {
121 		context->seen_enough = error;
122 		return;
123 	}
124 
125 	if (flags & XFS_ATTR_INCOMPLETE) {
126 		/* Incomplete attr key, just mark the inode for preening. */
127 		xchk_ino_set_preen(sx->sc, context->dp->i_ino);
128 		return;
129 	}
130 
131 	/* Does this name make sense? */
132 	if (!xfs_attr_namecheck(name, namelen)) {
133 		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK, args.blkno);
134 		return;
135 	}
136 
137 	/*
138 	 * Try to allocate enough memory to extrat the attr value.  If that
139 	 * doesn't work, we overload the seen_enough variable to convey
140 	 * the error message back to the main scrub function.
141 	 */
142 	error = xchk_setup_xattr_buf(sx->sc, valuelen, KM_MAYFAIL);
143 	if (error == -ENOMEM)
144 		error = -EDEADLOCK;
145 	if (error) {
146 		context->seen_enough = error;
147 		return;
148 	}
149 
150 	args.op_flags = XFS_DA_OP_NOTIME;
151 	args.attr_filter = flags & XFS_ATTR_NSP_ONDISK_MASK;
152 	args.geo = context->dp->i_mount->m_attr_geo;
153 	args.whichfork = XFS_ATTR_FORK;
154 	args.dp = context->dp;
155 	args.name = name;
156 	args.namelen = namelen;
157 	args.hashval = xfs_da_hashname(args.name, args.namelen);
158 	args.trans = context->tp;
159 	args.value = xchk_xattr_valuebuf(sx->sc);
160 	args.valuelen = valuelen;
161 
162 	error = xfs_attr_get_ilocked(&args);
163 	/* ENODATA means the hash lookup failed and the attr is bad */
164 	if (error == -ENODATA)
165 		error = -EFSCORRUPTED;
166 	if (!xchk_fblock_process_error(sx->sc, XFS_ATTR_FORK, args.blkno,
167 			&error))
168 		goto fail_xref;
169 	if (args.valuelen != valuelen)
170 		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK,
171 					     args.blkno);
172 fail_xref:
173 	if (sx->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
174 		context->seen_enough = 1;
175 	return;
176 }
177 
178 /*
179  * Mark a range [start, start+len) in this map.  Returns true if the
180  * region was free, and false if there's a conflict or a problem.
181  *
182  * Within a char, the lowest bit of the char represents the byte with
183  * the smallest address
184  */
185 STATIC bool
186 xchk_xattr_set_map(
187 	struct xfs_scrub	*sc,
188 	unsigned long		*map,
189 	unsigned int		start,
190 	unsigned int		len)
191 {
192 	unsigned int		mapsize = sc->mp->m_attr_geo->blksize;
193 	bool			ret = true;
194 
195 	if (start >= mapsize)
196 		return false;
197 	if (start + len > mapsize) {
198 		len = mapsize - start;
199 		ret = false;
200 	}
201 
202 	if (find_next_bit(map, mapsize, start) < start + len)
203 		ret = false;
204 	bitmap_set(map, start, len);
205 
206 	return ret;
207 }
208 
209 /*
210  * Check the leaf freemap from the usage bitmap.  Returns false if the
211  * attr freemap has problems or points to used space.
212  */
213 STATIC bool
214 xchk_xattr_check_freemap(
215 	struct xfs_scrub		*sc,
216 	unsigned long			*map,
217 	struct xfs_attr3_icleaf_hdr	*leafhdr)
218 {
219 	unsigned long			*freemap = xchk_xattr_freemap(sc);
220 	unsigned long			*dstmap = xchk_xattr_dstmap(sc);
221 	unsigned int			mapsize = sc->mp->m_attr_geo->blksize;
222 	int				i;
223 
224 	/* Construct bitmap of freemap contents. */
225 	bitmap_zero(freemap, mapsize);
226 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
227 		if (!xchk_xattr_set_map(sc, freemap,
228 				leafhdr->freemap[i].base,
229 				leafhdr->freemap[i].size))
230 			return false;
231 	}
232 
233 	/* Look for bits that are set in freemap and are marked in use. */
234 	return bitmap_and(dstmap, freemap, map, mapsize) == 0;
235 }
236 
237 /*
238  * Check this leaf entry's relations to everything else.
239  * Returns the number of bytes used for the name/value data.
240  */
241 STATIC void
242 xchk_xattr_entry(
243 	struct xchk_da_btree		*ds,
244 	int				level,
245 	char				*buf_end,
246 	struct xfs_attr_leafblock	*leaf,
247 	struct xfs_attr3_icleaf_hdr	*leafhdr,
248 	struct xfs_attr_leaf_entry	*ent,
249 	int				idx,
250 	unsigned int			*usedbytes,
251 	__u32				*last_hashval)
252 {
253 	struct xfs_mount		*mp = ds->state->mp;
254 	unsigned long			*usedmap = xchk_xattr_usedmap(ds->sc);
255 	char				*name_end;
256 	struct xfs_attr_leaf_name_local	*lentry;
257 	struct xfs_attr_leaf_name_remote *rentry;
258 	unsigned int			nameidx;
259 	unsigned int			namesize;
260 
261 	if (ent->pad2 != 0)
262 		xchk_da_set_corrupt(ds, level);
263 
264 	/* Hash values in order? */
265 	if (be32_to_cpu(ent->hashval) < *last_hashval)
266 		xchk_da_set_corrupt(ds, level);
267 	*last_hashval = be32_to_cpu(ent->hashval);
268 
269 	nameidx = be16_to_cpu(ent->nameidx);
270 	if (nameidx < leafhdr->firstused ||
271 	    nameidx >= mp->m_attr_geo->blksize) {
272 		xchk_da_set_corrupt(ds, level);
273 		return;
274 	}
275 
276 	/* Check the name information. */
277 	if (ent->flags & XFS_ATTR_LOCAL) {
278 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
279 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
280 				be16_to_cpu(lentry->valuelen));
281 		name_end = (char *)lentry + namesize;
282 		if (lentry->namelen == 0)
283 			xchk_da_set_corrupt(ds, level);
284 	} else {
285 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
286 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
287 		name_end = (char *)rentry + namesize;
288 		if (rentry->namelen == 0 || rentry->valueblk == 0)
289 			xchk_da_set_corrupt(ds, level);
290 	}
291 	if (name_end > buf_end)
292 		xchk_da_set_corrupt(ds, level);
293 
294 	if (!xchk_xattr_set_map(ds->sc, usedmap, nameidx, namesize))
295 		xchk_da_set_corrupt(ds, level);
296 	if (!(ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
297 		*usedbytes += namesize;
298 }
299 
300 /* Scrub an attribute leaf. */
301 STATIC int
302 xchk_xattr_block(
303 	struct xchk_da_btree		*ds,
304 	int				level)
305 {
306 	struct xfs_attr3_icleaf_hdr	leafhdr;
307 	struct xfs_mount		*mp = ds->state->mp;
308 	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
309 	struct xfs_buf			*bp = blk->bp;
310 	xfs_dablk_t			*last_checked = ds->private;
311 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
312 	struct xfs_attr_leaf_entry	*ent;
313 	struct xfs_attr_leaf_entry	*entries;
314 	unsigned long			*usedmap;
315 	char				*buf_end;
316 	size_t				off;
317 	__u32				last_hashval = 0;
318 	unsigned int			usedbytes = 0;
319 	unsigned int			hdrsize;
320 	int				i;
321 	int				error;
322 
323 	if (*last_checked == blk->blkno)
324 		return 0;
325 
326 	/* Allocate memory for block usage checking. */
327 	error = xchk_setup_xattr_buf(ds->sc, 0, KM_MAYFAIL);
328 	if (error == -ENOMEM)
329 		return -EDEADLOCK;
330 	if (error)
331 		return error;
332 	usedmap = xchk_xattr_usedmap(ds->sc);
333 
334 	*last_checked = blk->blkno;
335 	bitmap_zero(usedmap, mp->m_attr_geo->blksize);
336 
337 	/* Check all the padding. */
338 	if (xfs_sb_version_hascrc(&ds->sc->mp->m_sb)) {
339 		struct xfs_attr3_leafblock	*leaf = bp->b_addr;
340 
341 		if (leaf->hdr.pad1 != 0 || leaf->hdr.pad2 != 0 ||
342 		    leaf->hdr.info.hdr.pad != 0)
343 			xchk_da_set_corrupt(ds, level);
344 	} else {
345 		if (leaf->hdr.pad1 != 0 || leaf->hdr.info.pad != 0)
346 			xchk_da_set_corrupt(ds, level);
347 	}
348 
349 	/* Check the leaf header */
350 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
351 	hdrsize = xfs_attr3_leaf_hdr_size(leaf);
352 
353 	if (leafhdr.usedbytes > mp->m_attr_geo->blksize)
354 		xchk_da_set_corrupt(ds, level);
355 	if (leafhdr.firstused > mp->m_attr_geo->blksize)
356 		xchk_da_set_corrupt(ds, level);
357 	if (leafhdr.firstused < hdrsize)
358 		xchk_da_set_corrupt(ds, level);
359 	if (!xchk_xattr_set_map(ds->sc, usedmap, 0, hdrsize))
360 		xchk_da_set_corrupt(ds, level);
361 
362 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
363 		goto out;
364 
365 	entries = xfs_attr3_leaf_entryp(leaf);
366 	if ((char *)&entries[leafhdr.count] > (char *)leaf + leafhdr.firstused)
367 		xchk_da_set_corrupt(ds, level);
368 
369 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
370 	for (i = 0, ent = entries; i < leafhdr.count; ent++, i++) {
371 		/* Mark the leaf entry itself. */
372 		off = (char *)ent - (char *)leaf;
373 		if (!xchk_xattr_set_map(ds->sc, usedmap, off,
374 				sizeof(xfs_attr_leaf_entry_t))) {
375 			xchk_da_set_corrupt(ds, level);
376 			goto out;
377 		}
378 
379 		/* Check the entry and nameval. */
380 		xchk_xattr_entry(ds, level, buf_end, leaf, &leafhdr,
381 				ent, i, &usedbytes, &last_hashval);
382 
383 		if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
384 			goto out;
385 	}
386 
387 	if (!xchk_xattr_check_freemap(ds->sc, usedmap, &leafhdr))
388 		xchk_da_set_corrupt(ds, level);
389 
390 	if (leafhdr.usedbytes != usedbytes)
391 		xchk_da_set_corrupt(ds, level);
392 
393 out:
394 	return 0;
395 }
396 
397 /* Scrub a attribute btree record. */
398 STATIC int
399 xchk_xattr_rec(
400 	struct xchk_da_btree		*ds,
401 	int				level)
402 {
403 	struct xfs_mount		*mp = ds->state->mp;
404 	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
405 	struct xfs_attr_leaf_name_local	*lentry;
406 	struct xfs_attr_leaf_name_remote	*rentry;
407 	struct xfs_buf			*bp;
408 	struct xfs_attr_leaf_entry	*ent;
409 	xfs_dahash_t			calc_hash;
410 	xfs_dahash_t			hash;
411 	int				nameidx;
412 	int				hdrsize;
413 	unsigned int			badflags;
414 	int				error;
415 
416 	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
417 
418 	ent = xfs_attr3_leaf_entryp(blk->bp->b_addr) + blk->index;
419 
420 	/* Check the whole block, if necessary. */
421 	error = xchk_xattr_block(ds, level);
422 	if (error)
423 		goto out;
424 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
425 		goto out;
426 
427 	/* Check the hash of the entry. */
428 	error = xchk_da_btree_hash(ds, level, &ent->hashval);
429 	if (error)
430 		goto out;
431 
432 	/* Find the attr entry's location. */
433 	bp = blk->bp;
434 	hdrsize = xfs_attr3_leaf_hdr_size(bp->b_addr);
435 	nameidx = be16_to_cpu(ent->nameidx);
436 	if (nameidx < hdrsize || nameidx >= mp->m_attr_geo->blksize) {
437 		xchk_da_set_corrupt(ds, level);
438 		goto out;
439 	}
440 
441 	/* Retrieve the entry and check it. */
442 	hash = be32_to_cpu(ent->hashval);
443 	badflags = ~(XFS_ATTR_LOCAL | XFS_ATTR_ROOT | XFS_ATTR_SECURE |
444 			XFS_ATTR_INCOMPLETE);
445 	if ((ent->flags & badflags) != 0)
446 		xchk_da_set_corrupt(ds, level);
447 	if (ent->flags & XFS_ATTR_LOCAL) {
448 		lentry = (struct xfs_attr_leaf_name_local *)
449 				(((char *)bp->b_addr) + nameidx);
450 		if (lentry->namelen <= 0) {
451 			xchk_da_set_corrupt(ds, level);
452 			goto out;
453 		}
454 		calc_hash = xfs_da_hashname(lentry->nameval, lentry->namelen);
455 	} else {
456 		rentry = (struct xfs_attr_leaf_name_remote *)
457 				(((char *)bp->b_addr) + nameidx);
458 		if (rentry->namelen <= 0) {
459 			xchk_da_set_corrupt(ds, level);
460 			goto out;
461 		}
462 		calc_hash = xfs_da_hashname(rentry->name, rentry->namelen);
463 	}
464 	if (calc_hash != hash)
465 		xchk_da_set_corrupt(ds, level);
466 
467 out:
468 	return error;
469 }
470 
471 /* Scrub the extended attribute metadata. */
472 int
473 xchk_xattr(
474 	struct xfs_scrub		*sc)
475 {
476 	struct xchk_xattr		sx;
477 	xfs_dablk_t			last_checked = -1U;
478 	int				error = 0;
479 
480 	if (!xfs_inode_hasattr(sc->ip))
481 		return -ENOENT;
482 
483 	memset(&sx, 0, sizeof(sx));
484 	/* Check attribute tree structure */
485 	error = xchk_da_btree(sc, XFS_ATTR_FORK, xchk_xattr_rec,
486 			&last_checked);
487 	if (error)
488 		goto out;
489 
490 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
491 		goto out;
492 
493 	/* Check that every attr key can also be looked up by hash. */
494 	sx.context.dp = sc->ip;
495 	sx.context.resynch = 1;
496 	sx.context.put_listent = xchk_xattr_listent;
497 	sx.context.tp = sc->tp;
498 	sx.context.allow_incomplete = true;
499 	sx.sc = sc;
500 
501 	/*
502 	 * Look up every xattr in this file by name.
503 	 *
504 	 * Use the backend implementation of xfs_attr_list to call
505 	 * xchk_xattr_listent on every attribute key in this inode.
506 	 * In other words, we use the same iterator/callback mechanism
507 	 * that listattr uses to scrub extended attributes, though in our
508 	 * _listent function, we check the value of the attribute.
509 	 *
510 	 * The VFS only locks i_rwsem when modifying attrs, so keep all
511 	 * three locks held because that's the only way to ensure we're
512 	 * the only thread poking into the da btree.  We traverse the da
513 	 * btree while holding a leaf buffer locked for the xattr name
514 	 * iteration, which doesn't really follow the usual buffer
515 	 * locking order.
516 	 */
517 	error = xfs_attr_list_ilocked(&sx.context);
518 	if (!xchk_fblock_process_error(sc, XFS_ATTR_FORK, 0, &error))
519 		goto out;
520 
521 	/* Did our listent function try to return any errors? */
522 	if (sx.context.seen_enough < 0)
523 		error = sx.context.seen_enough;
524 out:
525 	return error;
526 }
527