1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2017-2023 Oracle.  All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
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 "xfs_attr_sf.h"
19#include "scrub/scrub.h"
20#include "scrub/common.h"
21#include "scrub/dabtree.h"
22#include "scrub/attr.h"
23
24/* Free the buffers linked from the xattr buffer. */
25static void
26xchk_xattr_buf_cleanup(
27	void			*priv)
28{
29	struct xchk_xattr_buf	*ab = priv;
30
31	kvfree(ab->freemap);
32	ab->freemap = NULL;
33	kvfree(ab->usedmap);
34	ab->usedmap = NULL;
35	kvfree(ab->value);
36	ab->value = NULL;
37	ab->value_sz = 0;
38}
39
40/*
41 * Allocate the free space bitmap if we're trying harder; there are leaf blocks
42 * in the attr fork; or we can't tell if there are leaf blocks.
43 */
44static inline bool
45xchk_xattr_want_freemap(
46	struct xfs_scrub	*sc)
47{
48	struct xfs_ifork	*ifp;
49
50	if (sc->flags & XCHK_TRY_HARDER)
51		return true;
52
53	if (!sc->ip)
54		return true;
55
56	ifp = xfs_ifork_ptr(sc->ip, XFS_ATTR_FORK);
57	if (!ifp)
58		return false;
59
60	return xfs_ifork_has_extents(ifp);
61}
62
63/*
64 * Allocate enough memory to hold an attr value and attr block bitmaps,
65 * reallocating the buffer if necessary.  Buffer contents are not preserved
66 * across a reallocation.
67 */
68static int
69xchk_setup_xattr_buf(
70	struct xfs_scrub	*sc,
71	size_t			value_size)
72{
73	size_t			bmp_sz;
74	struct xchk_xattr_buf	*ab = sc->buf;
75	void			*new_val;
76
77	bmp_sz = sizeof(long) * BITS_TO_LONGS(sc->mp->m_attr_geo->blksize);
78
79	if (ab)
80		goto resize_value;
81
82	ab = kvzalloc(sizeof(struct xchk_xattr_buf), XCHK_GFP_FLAGS);
83	if (!ab)
84		return -ENOMEM;
85	sc->buf = ab;
86	sc->buf_cleanup = xchk_xattr_buf_cleanup;
87
88	ab->usedmap = kvmalloc(bmp_sz, XCHK_GFP_FLAGS);
89	if (!ab->usedmap)
90		return -ENOMEM;
91
92	if (xchk_xattr_want_freemap(sc)) {
93		ab->freemap = kvmalloc(bmp_sz, XCHK_GFP_FLAGS);
94		if (!ab->freemap)
95			return -ENOMEM;
96	}
97
98resize_value:
99	if (ab->value_sz >= value_size)
100		return 0;
101
102	if (ab->value) {
103		kvfree(ab->value);
104		ab->value = NULL;
105		ab->value_sz = 0;
106	}
107
108	new_val = kvmalloc(value_size, XCHK_GFP_FLAGS);
109	if (!new_val)
110		return -ENOMEM;
111
112	ab->value = new_val;
113	ab->value_sz = value_size;
114	return 0;
115}
116
117/* Set us up to scrub an inode's extended attributes. */
118int
119xchk_setup_xattr(
120	struct xfs_scrub	*sc)
121{
122	int			error;
123
124	/*
125	 * We failed to get memory while checking attrs, so this time try to
126	 * get all the memory we're ever going to need.  Allocate the buffer
127	 * without the inode lock held, which means we can sleep.
128	 */
129	if (sc->flags & XCHK_TRY_HARDER) {
130		error = xchk_setup_xattr_buf(sc, XATTR_SIZE_MAX);
131		if (error)
132			return error;
133	}
134
135	return xchk_setup_inode_contents(sc, 0);
136}
137
138/* Extended Attributes */
139
140struct xchk_xattr {
141	struct xfs_attr_list_context	context;
142	struct xfs_scrub		*sc;
143};
144
145/*
146 * Check that an extended attribute key can be looked up by hash.
147 *
148 * We use the XFS attribute list iterator (i.e. xfs_attr_list_ilocked)
149 * to call this function for every attribute key in an inode.  Once
150 * we're here, we load the attribute value to see if any errors happen,
151 * or if we get more or less data than we expected.
152 */
153static void
154xchk_xattr_listent(
155	struct xfs_attr_list_context	*context,
156	int				flags,
157	unsigned char			*name,
158	int				namelen,
159	int				valuelen)
160{
161	struct xfs_da_args		args = {
162		.op_flags		= XFS_DA_OP_NOTIME,
163		.attr_filter		= flags & XFS_ATTR_NSP_ONDISK_MASK,
164		.geo			= context->dp->i_mount->m_attr_geo,
165		.whichfork		= XFS_ATTR_FORK,
166		.dp			= context->dp,
167		.name			= name,
168		.namelen		= namelen,
169		.hashval		= xfs_da_hashname(name, namelen),
170		.trans			= context->tp,
171		.valuelen		= valuelen,
172	};
173	struct xchk_xattr_buf		*ab;
174	struct xchk_xattr		*sx;
175	int				error = 0;
176
177	sx = container_of(context, struct xchk_xattr, context);
178	ab = sx->sc->buf;
179
180	if (xchk_should_terminate(sx->sc, &error)) {
181		context->seen_enough = error;
182		return;
183	}
184
185	if (flags & XFS_ATTR_INCOMPLETE) {
186		/* Incomplete attr key, just mark the inode for preening. */
187		xchk_ino_set_preen(sx->sc, context->dp->i_ino);
188		return;
189	}
190
191	/* Only one namespace bit allowed. */
192	if (hweight32(flags & XFS_ATTR_NSP_ONDISK_MASK) > 1) {
193		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK, args.blkno);
194		goto fail_xref;
195	}
196
197	/* Does this name make sense? */
198	if (!xfs_attr_namecheck(name, namelen)) {
199		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK, args.blkno);
200		goto fail_xref;
201	}
202
203	/*
204	 * Local xattr values are stored in the attr leaf block, so we don't
205	 * need to retrieve the value from a remote block to detect corruption
206	 * problems.
207	 */
208	if (flags & XFS_ATTR_LOCAL)
209		goto fail_xref;
210
211	/*
212	 * Try to allocate enough memory to extrat the attr value.  If that
213	 * doesn't work, we overload the seen_enough variable to convey
214	 * the error message back to the main scrub function.
215	 */
216	error = xchk_setup_xattr_buf(sx->sc, valuelen);
217	if (error == -ENOMEM)
218		error = -EDEADLOCK;
219	if (error) {
220		context->seen_enough = error;
221		return;
222	}
223
224	args.value = ab->value;
225
226	error = xfs_attr_get_ilocked(&args);
227	/* ENODATA means the hash lookup failed and the attr is bad */
228	if (error == -ENODATA)
229		error = -EFSCORRUPTED;
230	if (!xchk_fblock_process_error(sx->sc, XFS_ATTR_FORK, args.blkno,
231			&error))
232		goto fail_xref;
233	if (args.valuelen != valuelen)
234		xchk_fblock_set_corrupt(sx->sc, XFS_ATTR_FORK,
235					     args.blkno);
236fail_xref:
237	if (sx->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
238		context->seen_enough = 1;
239	return;
240}
241
242/*
243 * Mark a range [start, start+len) in this map.  Returns true if the
244 * region was free, and false if there's a conflict or a problem.
245 *
246 * Within a char, the lowest bit of the char represents the byte with
247 * the smallest address
248 */
249STATIC bool
250xchk_xattr_set_map(
251	struct xfs_scrub	*sc,
252	unsigned long		*map,
253	unsigned int		start,
254	unsigned int		len)
255{
256	unsigned int		mapsize = sc->mp->m_attr_geo->blksize;
257	bool			ret = true;
258
259	if (start >= mapsize)
260		return false;
261	if (start + len > mapsize) {
262		len = mapsize - start;
263		ret = false;
264	}
265
266	if (find_next_bit(map, mapsize, start) < start + len)
267		ret = false;
268	bitmap_set(map, start, len);
269
270	return ret;
271}
272
273/*
274 * Check the leaf freemap from the usage bitmap.  Returns false if the
275 * attr freemap has problems or points to used space.
276 */
277STATIC bool
278xchk_xattr_check_freemap(
279	struct xfs_scrub		*sc,
280	struct xfs_attr3_icleaf_hdr	*leafhdr)
281{
282	struct xchk_xattr_buf		*ab = sc->buf;
283	unsigned int			mapsize = sc->mp->m_attr_geo->blksize;
284	int				i;
285
286	/* Construct bitmap of freemap contents. */
287	bitmap_zero(ab->freemap, mapsize);
288	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
289		if (!xchk_xattr_set_map(sc, ab->freemap,
290				leafhdr->freemap[i].base,
291				leafhdr->freemap[i].size))
292			return false;
293	}
294
295	/* Look for bits that are set in freemap and are marked in use. */
296	return !bitmap_intersects(ab->freemap, ab->usedmap, mapsize);
297}
298
299/*
300 * Check this leaf entry's relations to everything else.
301 * Returns the number of bytes used for the name/value data.
302 */
303STATIC void
304xchk_xattr_entry(
305	struct xchk_da_btree		*ds,
306	int				level,
307	char				*buf_end,
308	struct xfs_attr_leafblock	*leaf,
309	struct xfs_attr3_icleaf_hdr	*leafhdr,
310	struct xfs_attr_leaf_entry	*ent,
311	int				idx,
312	unsigned int			*usedbytes,
313	__u32				*last_hashval)
314{
315	struct xfs_mount		*mp = ds->state->mp;
316	struct xchk_xattr_buf		*ab = ds->sc->buf;
317	char				*name_end;
318	struct xfs_attr_leaf_name_local	*lentry;
319	struct xfs_attr_leaf_name_remote *rentry;
320	unsigned int			nameidx;
321	unsigned int			namesize;
322
323	if (ent->pad2 != 0)
324		xchk_da_set_corrupt(ds, level);
325
326	/* Hash values in order? */
327	if (be32_to_cpu(ent->hashval) < *last_hashval)
328		xchk_da_set_corrupt(ds, level);
329	*last_hashval = be32_to_cpu(ent->hashval);
330
331	nameidx = be16_to_cpu(ent->nameidx);
332	if (nameidx < leafhdr->firstused ||
333	    nameidx >= mp->m_attr_geo->blksize) {
334		xchk_da_set_corrupt(ds, level);
335		return;
336	}
337
338	/* Check the name information. */
339	if (ent->flags & XFS_ATTR_LOCAL) {
340		lentry = xfs_attr3_leaf_name_local(leaf, idx);
341		namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
342				be16_to_cpu(lentry->valuelen));
343		name_end = (char *)lentry + namesize;
344		if (lentry->namelen == 0)
345			xchk_da_set_corrupt(ds, level);
346	} else {
347		rentry = xfs_attr3_leaf_name_remote(leaf, idx);
348		namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
349		name_end = (char *)rentry + namesize;
350		if (rentry->namelen == 0 || rentry->valueblk == 0)
351			xchk_da_set_corrupt(ds, level);
352	}
353	if (name_end > buf_end)
354		xchk_da_set_corrupt(ds, level);
355
356	if (!xchk_xattr_set_map(ds->sc, ab->usedmap, nameidx, namesize))
357		xchk_da_set_corrupt(ds, level);
358	if (!(ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
359		*usedbytes += namesize;
360}
361
362/* Scrub an attribute leaf. */
363STATIC int
364xchk_xattr_block(
365	struct xchk_da_btree		*ds,
366	int				level)
367{
368	struct xfs_attr3_icleaf_hdr	leafhdr;
369	struct xfs_mount		*mp = ds->state->mp;
370	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
371	struct xfs_buf			*bp = blk->bp;
372	xfs_dablk_t			*last_checked = ds->private;
373	struct xfs_attr_leafblock	*leaf = bp->b_addr;
374	struct xfs_attr_leaf_entry	*ent;
375	struct xfs_attr_leaf_entry	*entries;
376	struct xchk_xattr_buf		*ab = ds->sc->buf;
377	char				*buf_end;
378	size_t				off;
379	__u32				last_hashval = 0;
380	unsigned int			usedbytes = 0;
381	unsigned int			hdrsize;
382	int				i;
383
384	if (*last_checked == blk->blkno)
385		return 0;
386
387	*last_checked = blk->blkno;
388	bitmap_zero(ab->usedmap, mp->m_attr_geo->blksize);
389
390	/* Check all the padding. */
391	if (xfs_has_crc(ds->sc->mp)) {
392		struct xfs_attr3_leafblock	*leaf3 = bp->b_addr;
393
394		if (leaf3->hdr.pad1 != 0 || leaf3->hdr.pad2 != 0 ||
395		    leaf3->hdr.info.hdr.pad != 0)
396			xchk_da_set_corrupt(ds, level);
397	} else {
398		if (leaf->hdr.pad1 != 0 || leaf->hdr.info.pad != 0)
399			xchk_da_set_corrupt(ds, level);
400	}
401
402	/* Check the leaf header */
403	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
404	hdrsize = xfs_attr3_leaf_hdr_size(leaf);
405
406	if (leafhdr.usedbytes > mp->m_attr_geo->blksize)
407		xchk_da_set_corrupt(ds, level);
408	if (leafhdr.firstused > mp->m_attr_geo->blksize)
409		xchk_da_set_corrupt(ds, level);
410	if (leafhdr.firstused < hdrsize)
411		xchk_da_set_corrupt(ds, level);
412	if (!xchk_xattr_set_map(ds->sc, ab->usedmap, 0, hdrsize))
413		xchk_da_set_corrupt(ds, level);
414
415	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
416		goto out;
417
418	entries = xfs_attr3_leaf_entryp(leaf);
419	if ((char *)&entries[leafhdr.count] > (char *)leaf + leafhdr.firstused)
420		xchk_da_set_corrupt(ds, level);
421
422	buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
423	for (i = 0, ent = entries; i < leafhdr.count; ent++, i++) {
424		/* Mark the leaf entry itself. */
425		off = (char *)ent - (char *)leaf;
426		if (!xchk_xattr_set_map(ds->sc, ab->usedmap, off,
427				sizeof(xfs_attr_leaf_entry_t))) {
428			xchk_da_set_corrupt(ds, level);
429			goto out;
430		}
431
432		/* Check the entry and nameval. */
433		xchk_xattr_entry(ds, level, buf_end, leaf, &leafhdr,
434				ent, i, &usedbytes, &last_hashval);
435
436		if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
437			goto out;
438	}
439
440	if (!xchk_xattr_check_freemap(ds->sc, &leafhdr))
441		xchk_da_set_corrupt(ds, level);
442
443	if (leafhdr.usedbytes != usedbytes)
444		xchk_da_set_corrupt(ds, level);
445
446out:
447	return 0;
448}
449
450/* Scrub a attribute btree record. */
451STATIC int
452xchk_xattr_rec(
453	struct xchk_da_btree		*ds,
454	int				level)
455{
456	struct xfs_mount		*mp = ds->state->mp;
457	struct xfs_da_state_blk		*blk = &ds->state->path.blk[level];
458	struct xfs_attr_leaf_name_local	*lentry;
459	struct xfs_attr_leaf_name_remote	*rentry;
460	struct xfs_buf			*bp;
461	struct xfs_attr_leaf_entry	*ent;
462	xfs_dahash_t			calc_hash;
463	xfs_dahash_t			hash;
464	int				nameidx;
465	int				hdrsize;
466	unsigned int			badflags;
467	int				error;
468
469	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
470
471	ent = xfs_attr3_leaf_entryp(blk->bp->b_addr) + blk->index;
472
473	/* Check the whole block, if necessary. */
474	error = xchk_xattr_block(ds, level);
475	if (error)
476		goto out;
477	if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
478		goto out;
479
480	/* Check the hash of the entry. */
481	error = xchk_da_btree_hash(ds, level, &ent->hashval);
482	if (error)
483		goto out;
484
485	/* Find the attr entry's location. */
486	bp = blk->bp;
487	hdrsize = xfs_attr3_leaf_hdr_size(bp->b_addr);
488	nameidx = be16_to_cpu(ent->nameidx);
489	if (nameidx < hdrsize || nameidx >= mp->m_attr_geo->blksize) {
490		xchk_da_set_corrupt(ds, level);
491		goto out;
492	}
493
494	/* Retrieve the entry and check it. */
495	hash = be32_to_cpu(ent->hashval);
496	badflags = ~(XFS_ATTR_LOCAL | XFS_ATTR_ROOT | XFS_ATTR_SECURE |
497			XFS_ATTR_INCOMPLETE);
498	if ((ent->flags & badflags) != 0)
499		xchk_da_set_corrupt(ds, level);
500	if (ent->flags & XFS_ATTR_LOCAL) {
501		lentry = (struct xfs_attr_leaf_name_local *)
502				(((char *)bp->b_addr) + nameidx);
503		if (lentry->namelen <= 0) {
504			xchk_da_set_corrupt(ds, level);
505			goto out;
506		}
507		calc_hash = xfs_da_hashname(lentry->nameval, lentry->namelen);
508	} else {
509		rentry = (struct xfs_attr_leaf_name_remote *)
510				(((char *)bp->b_addr) + nameidx);
511		if (rentry->namelen <= 0) {
512			xchk_da_set_corrupt(ds, level);
513			goto out;
514		}
515		calc_hash = xfs_da_hashname(rentry->name, rentry->namelen);
516	}
517	if (calc_hash != hash)
518		xchk_da_set_corrupt(ds, level);
519
520out:
521	return error;
522}
523
524/* Check space usage of shortform attrs. */
525STATIC int
526xchk_xattr_check_sf(
527	struct xfs_scrub		*sc)
528{
529	struct xchk_xattr_buf		*ab = sc->buf;
530	struct xfs_ifork		*ifp = &sc->ip->i_af;
531	struct xfs_attr_sf_hdr		*sf = ifp->if_data;
532	struct xfs_attr_sf_entry	*sfe = xfs_attr_sf_firstentry(sf);
533	struct xfs_attr_sf_entry	*next;
534	unsigned char			*end = ifp->if_data + ifp->if_bytes;
535	int				i;
536	int				error = 0;
537
538	bitmap_zero(ab->usedmap, ifp->if_bytes);
539	xchk_xattr_set_map(sc, ab->usedmap, 0, sizeof(*sf));
540
541	if ((unsigned char *)sfe > end) {
542		xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
543		return 0;
544	}
545
546	for (i = 0; i < sf->count; i++) {
547		unsigned char		*name = sfe->nameval;
548		unsigned char		*value = &sfe->nameval[sfe->namelen];
549
550		if (xchk_should_terminate(sc, &error))
551			return error;
552
553		next = xfs_attr_sf_nextentry(sfe);
554		if ((unsigned char *)next > end) {
555			xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
556			break;
557		}
558
559		if (!xchk_xattr_set_map(sc, ab->usedmap,
560				(char *)sfe - (char *)sf,
561				sizeof(struct xfs_attr_sf_entry))) {
562			xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
563			break;
564		}
565
566		if (!xchk_xattr_set_map(sc, ab->usedmap,
567				(char *)name - (char *)sf,
568				sfe->namelen)) {
569			xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
570			break;
571		}
572
573		if (!xchk_xattr_set_map(sc, ab->usedmap,
574				(char *)value - (char *)sf,
575				sfe->valuelen)) {
576			xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
577			break;
578		}
579
580		sfe = next;
581	}
582
583	return 0;
584}
585
586/* Scrub the extended attribute metadata. */
587int
588xchk_xattr(
589	struct xfs_scrub		*sc)
590{
591	struct xchk_xattr		sx = {
592		.sc			= sc,
593		.context		= {
594			.dp		= sc->ip,
595			.tp		= sc->tp,
596			.resynch	= 1,
597			.put_listent	= xchk_xattr_listent,
598			.allow_incomplete = true,
599		},
600	};
601	xfs_dablk_t			last_checked = -1U;
602	int				error = 0;
603
604	if (!xfs_inode_hasattr(sc->ip))
605		return -ENOENT;
606
607	/* Allocate memory for xattr checking. */
608	error = xchk_setup_xattr_buf(sc, 0);
609	if (error == -ENOMEM)
610		return -EDEADLOCK;
611	if (error)
612		return error;
613
614	/* Check the physical structure of the xattr. */
615	if (sc->ip->i_af.if_format == XFS_DINODE_FMT_LOCAL)
616		error = xchk_xattr_check_sf(sc);
617	else
618		error = xchk_da_btree(sc, XFS_ATTR_FORK, xchk_xattr_rec,
619				&last_checked);
620	if (error)
621		return error;
622
623	if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
624		return 0;
625
626	/*
627	 * Look up every xattr in this file by name and hash.
628	 *
629	 * Use the backend implementation of xfs_attr_list to call
630	 * xchk_xattr_listent on every attribute key in this inode.
631	 * In other words, we use the same iterator/callback mechanism
632	 * that listattr uses to scrub extended attributes, though in our
633	 * _listent function, we check the value of the attribute.
634	 *
635	 * The VFS only locks i_rwsem when modifying attrs, so keep all
636	 * three locks held because that's the only way to ensure we're
637	 * the only thread poking into the da btree.  We traverse the da
638	 * btree while holding a leaf buffer locked for the xattr name
639	 * iteration, which doesn't really follow the usual buffer
640	 * locking order.
641	 */
642	error = xfs_attr_list_ilocked(&sx.context);
643	if (!xchk_fblock_process_error(sc, XFS_ATTR_FORK, 0, &error))
644		return error;
645
646	/* Did our listent function try to return any errors? */
647	if (sx.context.seen_enough < 0)
648		return sx.context.seen_enough;
649
650	return 0;
651}
652