xref: /openbmc/linux/fs/xfs/scrub/attr.c (revision 03638e62)
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, KM_SLEEP);
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_int_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.flags = ATTR_KERNOTIME;
151 	if (flags & XFS_ATTR_ROOT)
152 		args.flags |= ATTR_ROOT;
153 	else if (flags & XFS_ATTR_SECURE)
154 		args.flags |= ATTR_SECURE;
155 	args.geo = context->dp->i_mount->m_attr_geo;
156 	args.whichfork = XFS_ATTR_FORK;
157 	args.dp = context->dp;
158 	args.name = name;
159 	args.namelen = namelen;
160 	args.hashval = xfs_da_hashname(args.name, args.namelen);
161 	args.trans = context->tp;
162 	args.value = xchk_xattr_valuebuf(sx->sc);
163 	args.valuelen = valuelen;
164 
165 	error = xfs_attr_get_ilocked(context->dp, &args);
166 	if (error == -EEXIST)
167 		error = 0;
168 	if (!xchk_fblock_process_error(sx->sc, XFS_ATTR_FORK, args.blkno,
169 			&error))
170 		goto fail_xref;
171 	if (args.valuelen != valuelen)
172 		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK,
173 					     args.blkno);
174 fail_xref:
175 	if (sx->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
176 		context->seen_enough = XFS_ITER_ABORT;
177 	return;
178 }
179 
180 /*
181  * Mark a range [start, start+len) in this map.  Returns true if the
182  * region was free, and false if there's a conflict or a problem.
183  *
184  * Within a char, the lowest bit of the char represents the byte with
185  * the smallest address
186  */
187 STATIC bool
188 xchk_xattr_set_map(
189 	struct xfs_scrub	*sc,
190 	unsigned long		*map,
191 	unsigned int		start,
192 	unsigned int		len)
193 {
194 	unsigned int		mapsize = sc->mp->m_attr_geo->blksize;
195 	bool			ret = true;
196 
197 	if (start >= mapsize)
198 		return false;
199 	if (start + len > mapsize) {
200 		len = mapsize - start;
201 		ret = false;
202 	}
203 
204 	if (find_next_bit(map, mapsize, start) < start + len)
205 		ret = false;
206 	bitmap_set(map, start, len);
207 
208 	return ret;
209 }
210 
211 /*
212  * Check the leaf freemap from the usage bitmap.  Returns false if the
213  * attr freemap has problems or points to used space.
214  */
215 STATIC bool
216 xchk_xattr_check_freemap(
217 	struct xfs_scrub		*sc,
218 	unsigned long			*map,
219 	struct xfs_attr3_icleaf_hdr	*leafhdr)
220 {
221 	unsigned long			*freemap = xchk_xattr_freemap(sc);
222 	unsigned long			*dstmap = xchk_xattr_dstmap(sc);
223 	unsigned int			mapsize = sc->mp->m_attr_geo->blksize;
224 	int				i;
225 
226 	/* Construct bitmap of freemap contents. */
227 	bitmap_zero(freemap, mapsize);
228 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
229 		if (!xchk_xattr_set_map(sc, freemap,
230 				leafhdr->freemap[i].base,
231 				leafhdr->freemap[i].size))
232 			return false;
233 	}
234 
235 	/* Look for bits that are set in freemap and are marked in use. */
236 	return bitmap_and(dstmap, freemap, map, mapsize) == 0;
237 }
238 
239 /*
240  * Check this leaf entry's relations to everything else.
241  * Returns the number of bytes used for the name/value data.
242  */
243 STATIC void
244 xchk_xattr_entry(
245 	struct xchk_da_btree		*ds,
246 	int				level,
247 	char				*buf_end,
248 	struct xfs_attr_leafblock	*leaf,
249 	struct xfs_attr3_icleaf_hdr	*leafhdr,
250 	struct xfs_attr_leaf_entry	*ent,
251 	int				idx,
252 	unsigned int			*usedbytes,
253 	__u32				*last_hashval)
254 {
255 	struct xfs_mount		*mp = ds->state->mp;
256 	unsigned long			*usedmap = xchk_xattr_usedmap(ds->sc);
257 	char				*name_end;
258 	struct xfs_attr_leaf_name_local	*lentry;
259 	struct xfs_attr_leaf_name_remote *rentry;
260 	unsigned int			nameidx;
261 	unsigned int			namesize;
262 
263 	if (ent->pad2 != 0)
264 		xchk_da_set_corrupt(ds, level);
265 
266 	/* Hash values in order? */
267 	if (be32_to_cpu(ent->hashval) < *last_hashval)
268 		xchk_da_set_corrupt(ds, level);
269 	*last_hashval = be32_to_cpu(ent->hashval);
270 
271 	nameidx = be16_to_cpu(ent->nameidx);
272 	if (nameidx < leafhdr->firstused ||
273 	    nameidx >= mp->m_attr_geo->blksize) {
274 		xchk_da_set_corrupt(ds, level);
275 		return;
276 	}
277 
278 	/* Check the name information. */
279 	if (ent->flags & XFS_ATTR_LOCAL) {
280 		lentry = xfs_attr3_leaf_name_local(leaf, idx);
281 		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
282 				be16_to_cpu(lentry->valuelen));
283 		name_end = (char *)lentry + namesize;
284 		if (lentry->namelen == 0)
285 			xchk_da_set_corrupt(ds, level);
286 	} else {
287 		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
288 		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
289 		name_end = (char *)rentry + namesize;
290 		if (rentry->namelen == 0 || rentry->valueblk == 0)
291 			xchk_da_set_corrupt(ds, level);
292 	}
293 	if (name_end > buf_end)
294 		xchk_da_set_corrupt(ds, level);
295 
296 	if (!xchk_xattr_set_map(ds->sc, usedmap, nameidx, namesize))
297 		xchk_da_set_corrupt(ds, level);
298 	if (!(ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
299 		*usedbytes += namesize;
300 }
301 
302 /* Scrub an attribute leaf. */
303 STATIC int
304 xchk_xattr_block(
305 	struct xchk_da_btree		*ds,
306 	int				level)
307 {
308 	struct xfs_attr3_icleaf_hdr	leafhdr;
309 	struct xfs_mount		*mp = ds->state->mp;
310 	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
311 	struct xfs_buf			*bp = blk->bp;
312 	xfs_dablk_t			*last_checked = ds->private;
313 	struct xfs_attr_leafblock	*leaf = bp->b_addr;
314 	struct xfs_attr_leaf_entry	*ent;
315 	struct xfs_attr_leaf_entry	*entries;
316 	unsigned long			*usedmap;
317 	char				*buf_end;
318 	size_t				off;
319 	__u32				last_hashval = 0;
320 	unsigned int			usedbytes = 0;
321 	unsigned int			hdrsize;
322 	int				i;
323 	int				error;
324 
325 	if (*last_checked == blk->blkno)
326 		return 0;
327 
328 	/* Allocate memory for block usage checking. */
329 	error = xchk_setup_xattr_buf(ds->sc, 0, KM_MAYFAIL);
330 	if (error == -ENOMEM)
331 		return -EDEADLOCK;
332 	if (error)
333 		return error;
334 	usedmap = xchk_xattr_usedmap(ds->sc);
335 
336 	*last_checked = blk->blkno;
337 	bitmap_zero(usedmap, mp->m_attr_geo->blksize);
338 
339 	/* Check all the padding. */
340 	if (xfs_sb_version_hascrc(&ds->sc->mp->m_sb)) {
341 		struct xfs_attr3_leafblock	*leaf = bp->b_addr;
342 
343 		if (leaf->hdr.pad1 != 0 || leaf->hdr.pad2 != 0 ||
344 		    leaf->hdr.info.hdr.pad != 0)
345 			xchk_da_set_corrupt(ds, level);
346 	} else {
347 		if (leaf->hdr.pad1 != 0 || leaf->hdr.info.pad != 0)
348 			xchk_da_set_corrupt(ds, level);
349 	}
350 
351 	/* Check the leaf header */
352 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
353 	hdrsize = xfs_attr3_leaf_hdr_size(leaf);
354 
355 	if (leafhdr.usedbytes > mp->m_attr_geo->blksize)
356 		xchk_da_set_corrupt(ds, level);
357 	if (leafhdr.firstused > mp->m_attr_geo->blksize)
358 		xchk_da_set_corrupt(ds, level);
359 	if (leafhdr.firstused < hdrsize)
360 		xchk_da_set_corrupt(ds, level);
361 	if (!xchk_xattr_set_map(ds->sc, usedmap, 0, hdrsize))
362 		xchk_da_set_corrupt(ds, level);
363 
364 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
365 		goto out;
366 
367 	entries = xfs_attr3_leaf_entryp(leaf);
368 	if ((char *)&entries[leafhdr.count] > (char *)leaf + leafhdr.firstused)
369 		xchk_da_set_corrupt(ds, level);
370 
371 	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
372 	for (i = 0, ent = entries; i < leafhdr.count; ent++, i++) {
373 		/* Mark the leaf entry itself. */
374 		off = (char *)ent - (char *)leaf;
375 		if (!xchk_xattr_set_map(ds->sc, usedmap, off,
376 				sizeof(xfs_attr_leaf_entry_t))) {
377 			xchk_da_set_corrupt(ds, level);
378 			goto out;
379 		}
380 
381 		/* Check the entry and nameval. */
382 		xchk_xattr_entry(ds, level, buf_end, leaf, &leafhdr,
383 				ent, i, &usedbytes, &last_hashval);
384 
385 		if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
386 			goto out;
387 	}
388 
389 	if (!xchk_xattr_check_freemap(ds->sc, usedmap, &leafhdr))
390 		xchk_da_set_corrupt(ds, level);
391 
392 	if (leafhdr.usedbytes != usedbytes)
393 		xchk_da_set_corrupt(ds, level);
394 
395 out:
396 	return 0;
397 }
398 
399 /* Scrub a attribute btree record. */
400 STATIC int
401 xchk_xattr_rec(
402 	struct xchk_da_btree		*ds,
403 	int				level,
404 	void				*rec)
405 {
406 	struct xfs_mount		*mp = ds->state->mp;
407 	struct xfs_attr_leaf_entry	*ent = rec;
408 	struct xfs_da_state_blk		*blk;
409 	struct xfs_attr_leaf_name_local	*lentry;
410 	struct xfs_attr_leaf_name_remote	*rentry;
411 	struct xfs_buf			*bp;
412 	xfs_dahash_t			calc_hash;
413 	xfs_dahash_t			hash;
414 	int				nameidx;
415 	int				hdrsize;
416 	unsigned int			badflags;
417 	int				error;
418 
419 	blk = &ds->state->path.blk[level];
420 
421 	/* Check the whole block, if necessary. */
422 	error = xchk_xattr_block(ds, level);
423 	if (error)
424 		goto out;
425 	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
426 		goto out;
427 
428 	/* Check the hash of the entry. */
429 	error = xchk_da_btree_hash(ds, level, &ent->hashval);
430 	if (error)
431 		goto out;
432 
433 	/* Find the attr entry's location. */
434 	bp = blk->bp;
435 	hdrsize = xfs_attr3_leaf_hdr_size(bp->b_addr);
436 	nameidx = be16_to_cpu(ent->nameidx);
437 	if (nameidx < hdrsize || nameidx >= mp->m_attr_geo->blksize) {
438 		xchk_da_set_corrupt(ds, level);
439 		goto out;
440 	}
441 
442 	/* Retrieve the entry and check it. */
443 	hash = be32_to_cpu(ent->hashval);
444 	badflags = ~(XFS_ATTR_LOCAL | XFS_ATTR_ROOT | XFS_ATTR_SECURE |
445 			XFS_ATTR_INCOMPLETE);
446 	if ((ent->flags & badflags) != 0)
447 		xchk_da_set_corrupt(ds, level);
448 	if (ent->flags & XFS_ATTR_LOCAL) {
449 		lentry = (struct xfs_attr_leaf_name_local *)
450 				(((char *)bp->b_addr) + nameidx);
451 		if (lentry->namelen <= 0) {
452 			xchk_da_set_corrupt(ds, level);
453 			goto out;
454 		}
455 		calc_hash = xfs_da_hashname(lentry->nameval, lentry->namelen);
456 	} else {
457 		rentry = (struct xfs_attr_leaf_name_remote *)
458 				(((char *)bp->b_addr) + nameidx);
459 		if (rentry->namelen <= 0) {
460 			xchk_da_set_corrupt(ds, level);
461 			goto out;
462 		}
463 		calc_hash = xfs_da_hashname(rentry->name, rentry->namelen);
464 	}
465 	if (calc_hash != hash)
466 		xchk_da_set_corrupt(ds, level);
467 
468 out:
469 	return error;
470 }
471 
472 /* Scrub the extended attribute metadata. */
473 int
474 xchk_xattr(
475 	struct xfs_scrub		*sc)
476 {
477 	struct xchk_xattr		sx;
478 	struct attrlist_cursor_kern	cursor = { 0 };
479 	xfs_dablk_t			last_checked = -1U;
480 	int				error = 0;
481 
482 	if (!xfs_inode_hasattr(sc->ip))
483 		return -ENOENT;
484 
485 	memset(&sx, 0, sizeof(sx));
486 	/* Check attribute tree structure */
487 	error = xchk_da_btree(sc, XFS_ATTR_FORK, xchk_xattr_rec,
488 			&last_checked);
489 	if (error)
490 		goto out;
491 
492 	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
493 		goto out;
494 
495 	/* Check that every attr key can also be looked up by hash. */
496 	sx.context.dp = sc->ip;
497 	sx.context.cursor = &cursor;
498 	sx.context.resynch = 1;
499 	sx.context.put_listent = xchk_xattr_listent;
500 	sx.context.tp = sc->tp;
501 	sx.context.flags = ATTR_INCOMPLETE;
502 	sx.sc = sc;
503 
504 	/*
505 	 * Look up every xattr in this file by name.
506 	 *
507 	 * Use the backend implementation of xfs_attr_list to call
508 	 * xchk_xattr_listent on every attribute key in this inode.
509 	 * In other words, we use the same iterator/callback mechanism
510 	 * that listattr uses to scrub extended attributes, though in our
511 	 * _listent function, we check the value of the attribute.
512 	 *
513 	 * The VFS only locks i_rwsem when modifying attrs, so keep all
514 	 * three locks held because that's the only way to ensure we're
515 	 * the only thread poking into the da btree.  We traverse the da
516 	 * btree while holding a leaf buffer locked for the xattr name
517 	 * iteration, which doesn't really follow the usual buffer
518 	 * locking order.
519 	 */
520 	error = xfs_attr_list_int_ilocked(&sx.context);
521 	if (!xchk_fblock_process_error(sc, XFS_ATTR_FORK, 0, &error))
522 		goto out;
523 
524 	/* Did our listent function try to return any errors? */
525 	if (sx.context.seen_enough < 0)
526 		error = sx.context.seen_enough;
527 out:
528 	return error;
529 }
530