xref: /openbmc/linux/fs/xfs/xfs_attr_list.c (revision 5e012745) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_da_format.h"
15 #include "xfs_inode.h"
16 #include "xfs_trans.h"
17 #include "xfs_bmap.h"
18 #include "xfs_attr.h"
19 #include "xfs_attr_sf.h"
20 #include "xfs_attr_leaf.h"
21 #include "xfs_error.h"
22 #include "xfs_trace.h"
23 #include "xfs_dir2.h"
24 
25 STATIC int
26 xfs_attr_shortform_compare(const void *a, const void *b)
27 {
28 	xfs_attr_sf_sort_t *sa, *sb;
29 
30 	sa = (xfs_attr_sf_sort_t *)a;
31 	sb = (xfs_attr_sf_sort_t *)b;
32 	if (sa->hash < sb->hash) {
33 		return -1;
34 	} else if (sa->hash > sb->hash) {
35 		return 1;
36 	} else {
37 		return sa->entno - sb->entno;
38 	}
39 }
40 
41 #define XFS_ISRESET_CURSOR(cursor) \
42 	(!((cursor)->initted) && !((cursor)->hashval) && \
43 	 !((cursor)->blkno) && !((cursor)->offset))
44 /*
45  * Copy out entries of shortform attribute lists for attr_list().
46  * Shortform attribute lists are not stored in hashval sorted order.
47  * If the output buffer is not large enough to hold them all, then we
48  * we have to calculate each entries' hashvalue and sort them before
49  * we can begin returning them to the user.
50  */
51 static int
52 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
53 {
54 	attrlist_cursor_kern_t *cursor;
55 	xfs_attr_sf_sort_t *sbuf, *sbp;
56 	xfs_attr_shortform_t *sf;
57 	xfs_attr_sf_entry_t *sfe;
58 	xfs_inode_t *dp;
59 	int sbsize, nsbuf, count, i;
60 
61 	ASSERT(context != NULL);
62 	dp = context->dp;
63 	ASSERT(dp != NULL);
64 	ASSERT(dp->i_afp != NULL);
65 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
66 	ASSERT(sf != NULL);
67 	if (!sf->hdr.count)
68 		return 0;
69 	cursor = context->cursor;
70 	ASSERT(cursor != NULL);
71 
72 	trace_xfs_attr_list_sf(context);
73 
74 	/*
75 	 * If the buffer is large enough and the cursor is at the start,
76 	 * do not bother with sorting since we will return everything in
77 	 * one buffer and another call using the cursor won't need to be
78 	 * made.
79 	 * Note the generous fudge factor of 16 overhead bytes per entry.
80 	 * If bufsize is zero then put_listent must be a search function
81 	 * and can just scan through what we have.
82 	 */
83 	if (context->bufsize == 0 ||
84 	    (XFS_ISRESET_CURSOR(cursor) &&
85 	     (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
86 		for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
87 			context->put_listent(context,
88 					     sfe->flags,
89 					     sfe->nameval,
90 					     (int)sfe->namelen,
91 					     (int)sfe->valuelen);
92 			/*
93 			 * Either search callback finished early or
94 			 * didn't fit it all in the buffer after all.
95 			 */
96 			if (context->seen_enough)
97 				break;
98 			sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
99 		}
100 		trace_xfs_attr_list_sf_all(context);
101 		return 0;
102 	}
103 
104 	/* do no more for a search callback */
105 	if (context->bufsize == 0)
106 		return 0;
107 
108 	/*
109 	 * It didn't all fit, so we have to sort everything on hashval.
110 	 */
111 	sbsize = sf->hdr.count * sizeof(*sbuf);
112 	sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
113 
114 	/*
115 	 * Scan the attribute list for the rest of the entries, storing
116 	 * the relevant info from only those that match into a buffer.
117 	 */
118 	nsbuf = 0;
119 	for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
120 		if (unlikely(
121 		    ((char *)sfe < (char *)sf) ||
122 		    ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
123 			XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
124 					     XFS_ERRLEVEL_LOW,
125 					     context->dp->i_mount, sfe,
126 					     sizeof(*sfe));
127 			kmem_free(sbuf);
128 			return -EFSCORRUPTED;
129 		}
130 
131 		sbp->entno = i;
132 		sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
133 		sbp->name = sfe->nameval;
134 		sbp->namelen = sfe->namelen;
135 		/* These are bytes, and both on-disk, don't endian-flip */
136 		sbp->valuelen = sfe->valuelen;
137 		sbp->flags = sfe->flags;
138 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
139 		sbp++;
140 		nsbuf++;
141 	}
142 
143 	/*
144 	 * Sort the entries on hash then entno.
145 	 */
146 	xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
147 
148 	/*
149 	 * Re-find our place IN THE SORTED LIST.
150 	 */
151 	count = 0;
152 	cursor->initted = 1;
153 	cursor->blkno = 0;
154 	for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
155 		if (sbp->hash == cursor->hashval) {
156 			if (cursor->offset == count) {
157 				break;
158 			}
159 			count++;
160 		} else if (sbp->hash > cursor->hashval) {
161 			break;
162 		}
163 	}
164 	if (i == nsbuf) {
165 		kmem_free(sbuf);
166 		return 0;
167 	}
168 
169 	/*
170 	 * Loop putting entries into the user buffer.
171 	 */
172 	for ( ; i < nsbuf; i++, sbp++) {
173 		if (cursor->hashval != sbp->hash) {
174 			cursor->hashval = sbp->hash;
175 			cursor->offset = 0;
176 		}
177 		context->put_listent(context,
178 				     sbp->flags,
179 				     sbp->name,
180 				     sbp->namelen,
181 				     sbp->valuelen);
182 		if (context->seen_enough)
183 			break;
184 		cursor->offset++;
185 	}
186 
187 	kmem_free(sbuf);
188 	return 0;
189 }
190 
191 /*
192  * We didn't find the block & hash mentioned in the cursor state, so
193  * walk down the attr btree looking for the hash.
194  */
195 STATIC int
196 xfs_attr_node_list_lookup(
197 	struct xfs_attr_list_context	*context,
198 	struct attrlist_cursor_kern	*cursor,
199 	struct xfs_buf			**pbp)
200 {
201 	struct xfs_da3_icnode_hdr	nodehdr;
202 	struct xfs_da_intnode		*node;
203 	struct xfs_da_node_entry	*btree;
204 	struct xfs_inode		*dp = context->dp;
205 	struct xfs_mount		*mp = dp->i_mount;
206 	struct xfs_trans		*tp = context->tp;
207 	struct xfs_buf			*bp;
208 	int				i;
209 	int				error = 0;
210 	unsigned int			expected_level = 0;
211 	uint16_t			magic;
212 
213 	ASSERT(*pbp == NULL);
214 	cursor->blkno = 0;
215 	for (;;) {
216 		error = xfs_da3_node_read(tp, dp, cursor->blkno, -1, &bp,
217 				XFS_ATTR_FORK);
218 		if (error)
219 			return error;
220 		node = bp->b_addr;
221 		magic = be16_to_cpu(node->hdr.info.magic);
222 		if (magic == XFS_ATTR_LEAF_MAGIC ||
223 		    magic == XFS_ATTR3_LEAF_MAGIC)
224 			break;
225 		if (magic != XFS_DA_NODE_MAGIC &&
226 		    magic != XFS_DA3_NODE_MAGIC) {
227 			XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
228 					node, sizeof(*node));
229 			goto out_corruptbuf;
230 		}
231 
232 		dp->d_ops->node_hdr_from_disk(&nodehdr, node);
233 
234 		/* Tree taller than we can handle; bail out! */
235 		if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
236 			goto out_corruptbuf;
237 
238 		/* Check the level from the root node. */
239 		if (cursor->blkno == 0)
240 			expected_level = nodehdr.level - 1;
241 		else if (expected_level != nodehdr.level)
242 			goto out_corruptbuf;
243 		else
244 			expected_level--;
245 
246 		btree = dp->d_ops->node_tree_p(node);
247 		for (i = 0; i < nodehdr.count; btree++, i++) {
248 			if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
249 				cursor->blkno = be32_to_cpu(btree->before);
250 				trace_xfs_attr_list_node_descend(context,
251 						btree);
252 				break;
253 			}
254 		}
255 		xfs_trans_brelse(tp, bp);
256 
257 		if (i == nodehdr.count)
258 			return 0;
259 
260 		/* We can't point back to the root. */
261 		if (cursor->blkno == 0)
262 			return -EFSCORRUPTED;
263 	}
264 
265 	if (expected_level != 0)
266 		goto out_corruptbuf;
267 
268 	*pbp = bp;
269 	return 0;
270 
271 out_corruptbuf:
272 	xfs_trans_brelse(tp, bp);
273 	return -EFSCORRUPTED;
274 }
275 
276 STATIC int
277 xfs_attr_node_list(
278 	struct xfs_attr_list_context	*context)
279 {
280 	struct xfs_attr3_icleaf_hdr	leafhdr;
281 	struct attrlist_cursor_kern	*cursor;
282 	struct xfs_attr_leafblock	*leaf;
283 	struct xfs_da_intnode		*node;
284 	struct xfs_buf			*bp;
285 	struct xfs_inode		*dp = context->dp;
286 	struct xfs_mount		*mp = dp->i_mount;
287 	int				error;
288 
289 	trace_xfs_attr_node_list(context);
290 
291 	cursor = context->cursor;
292 	cursor->initted = 1;
293 
294 	/*
295 	 * Do all sorts of validation on the passed-in cursor structure.
296 	 * If anything is amiss, ignore the cursor and look up the hashval
297 	 * starting from the btree root.
298 	 */
299 	bp = NULL;
300 	if (cursor->blkno > 0) {
301 		error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
302 					      &bp, XFS_ATTR_FORK);
303 		if ((error != 0) && (error != -EFSCORRUPTED))
304 			return error;
305 		if (bp) {
306 			struct xfs_attr_leaf_entry *entries;
307 
308 			node = bp->b_addr;
309 			switch (be16_to_cpu(node->hdr.info.magic)) {
310 			case XFS_DA_NODE_MAGIC:
311 			case XFS_DA3_NODE_MAGIC:
312 				trace_xfs_attr_list_wrong_blk(context);
313 				xfs_trans_brelse(context->tp, bp);
314 				bp = NULL;
315 				break;
316 			case XFS_ATTR_LEAF_MAGIC:
317 			case XFS_ATTR3_LEAF_MAGIC:
318 				leaf = bp->b_addr;
319 				xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
320 							     &leafhdr, leaf);
321 				entries = xfs_attr3_leaf_entryp(leaf);
322 				if (cursor->hashval > be32_to_cpu(
323 						entries[leafhdr.count - 1].hashval)) {
324 					trace_xfs_attr_list_wrong_blk(context);
325 					xfs_trans_brelse(context->tp, bp);
326 					bp = NULL;
327 				} else if (cursor->hashval <= be32_to_cpu(
328 						entries[0].hashval)) {
329 					trace_xfs_attr_list_wrong_blk(context);
330 					xfs_trans_brelse(context->tp, bp);
331 					bp = NULL;
332 				}
333 				break;
334 			default:
335 				trace_xfs_attr_list_wrong_blk(context);
336 				xfs_trans_brelse(context->tp, bp);
337 				bp = NULL;
338 			}
339 		}
340 	}
341 
342 	/*
343 	 * We did not find what we expected given the cursor's contents,
344 	 * so we start from the top and work down based on the hash value.
345 	 * Note that start of node block is same as start of leaf block.
346 	 */
347 	if (bp == NULL) {
348 		error = xfs_attr_node_list_lookup(context, cursor, &bp);
349 		if (error || !bp)
350 			return error;
351 	}
352 	ASSERT(bp != NULL);
353 
354 	/*
355 	 * Roll upward through the blocks, processing each leaf block in
356 	 * order.  As long as there is space in the result buffer, keep
357 	 * adding the information.
358 	 */
359 	for (;;) {
360 		leaf = bp->b_addr;
361 		xfs_attr3_leaf_list_int(bp, context);
362 		xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
363 		if (context->seen_enough || leafhdr.forw == 0)
364 			break;
365 		cursor->blkno = leafhdr.forw;
366 		xfs_trans_brelse(context->tp, bp);
367 		error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, -1, &bp);
368 		if (error)
369 			return error;
370 	}
371 	xfs_trans_brelse(context->tp, bp);
372 	return 0;
373 }
374 
375 /*
376  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
377  */
378 void
379 xfs_attr3_leaf_list_int(
380 	struct xfs_buf			*bp,
381 	struct xfs_attr_list_context	*context)
382 {
383 	struct attrlist_cursor_kern	*cursor;
384 	struct xfs_attr_leafblock	*leaf;
385 	struct xfs_attr3_icleaf_hdr	ichdr;
386 	struct xfs_attr_leaf_entry	*entries;
387 	struct xfs_attr_leaf_entry	*entry;
388 	int				i;
389 	struct xfs_mount		*mp = context->dp->i_mount;
390 
391 	trace_xfs_attr_list_leaf(context);
392 
393 	leaf = bp->b_addr;
394 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
395 	entries = xfs_attr3_leaf_entryp(leaf);
396 
397 	cursor = context->cursor;
398 	cursor->initted = 1;
399 
400 	/*
401 	 * Re-find our place in the leaf block if this is a new syscall.
402 	 */
403 	if (context->resynch) {
404 		entry = &entries[0];
405 		for (i = 0; i < ichdr.count; entry++, i++) {
406 			if (be32_to_cpu(entry->hashval) == cursor->hashval) {
407 				if (cursor->offset == context->dupcnt) {
408 					context->dupcnt = 0;
409 					break;
410 				}
411 				context->dupcnt++;
412 			} else if (be32_to_cpu(entry->hashval) >
413 					cursor->hashval) {
414 				context->dupcnt = 0;
415 				break;
416 			}
417 		}
418 		if (i == ichdr.count) {
419 			trace_xfs_attr_list_notfound(context);
420 			return;
421 		}
422 	} else {
423 		entry = &entries[0];
424 		i = 0;
425 	}
426 	context->resynch = 0;
427 
428 	/*
429 	 * We have found our place, start copying out the new attributes.
430 	 */
431 	for (; i < ichdr.count; entry++, i++) {
432 		char *name;
433 		int namelen, valuelen;
434 
435 		if (be32_to_cpu(entry->hashval) != cursor->hashval) {
436 			cursor->hashval = be32_to_cpu(entry->hashval);
437 			cursor->offset = 0;
438 		}
439 
440 		if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
441 		    !(context->flags & ATTR_INCOMPLETE))
442 			continue;		/* skip incomplete entries */
443 
444 		if (entry->flags & XFS_ATTR_LOCAL) {
445 			xfs_attr_leaf_name_local_t *name_loc;
446 
447 			name_loc = xfs_attr3_leaf_name_local(leaf, i);
448 			name = name_loc->nameval;
449 			namelen = name_loc->namelen;
450 			valuelen = be16_to_cpu(name_loc->valuelen);
451 		} else {
452 			xfs_attr_leaf_name_remote_t *name_rmt;
453 
454 			name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
455 			name = name_rmt->name;
456 			namelen = name_rmt->namelen;
457 			valuelen = be32_to_cpu(name_rmt->valuelen);
458 		}
459 
460 		context->put_listent(context, entry->flags,
461 					      name, namelen, valuelen);
462 		if (context->seen_enough)
463 			break;
464 		cursor->offset++;
465 	}
466 	trace_xfs_attr_list_leaf_end(context);
467 	return;
468 }
469 
470 /*
471  * Copy out attribute entries for attr_list(), for leaf attribute lists.
472  */
473 STATIC int
474 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
475 {
476 	int error;
477 	struct xfs_buf *bp;
478 
479 	trace_xfs_attr_leaf_list(context);
480 
481 	context->cursor->blkno = 0;
482 	error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
483 	if (error)
484 		return error;
485 
486 	xfs_attr3_leaf_list_int(bp, context);
487 	xfs_trans_brelse(context->tp, bp);
488 	return 0;
489 }
490 
491 int
492 xfs_attr_list_int_ilocked(
493 	struct xfs_attr_list_context	*context)
494 {
495 	struct xfs_inode		*dp = context->dp;
496 
497 	ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
498 
499 	/*
500 	 * Decide on what work routines to call based on the inode size.
501 	 */
502 	if (!xfs_inode_hasattr(dp))
503 		return 0;
504 	else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
505 		return xfs_attr_shortform_list(context);
506 	else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
507 		return xfs_attr_leaf_list(context);
508 	return xfs_attr_node_list(context);
509 }
510 
511 int
512 xfs_attr_list_int(
513 	xfs_attr_list_context_t *context)
514 {
515 	int error;
516 	xfs_inode_t *dp = context->dp;
517 	uint		lock_mode;
518 
519 	XFS_STATS_INC(dp->i_mount, xs_attr_list);
520 
521 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
522 		return -EIO;
523 
524 	lock_mode = xfs_ilock_attr_map_shared(dp);
525 	error = xfs_attr_list_int_ilocked(context);
526 	xfs_iunlock(dp, lock_mode);
527 	return error;
528 }
529 
530 #define	ATTR_ENTBASESIZE		/* minimum bytes used by an attr */ \
531 	(((struct attrlist_ent *) 0)->a_name - (char *) 0)
532 #define	ATTR_ENTSIZE(namelen)		/* actual bytes used by an attr */ \
533 	((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(uint32_t)-1) \
534 	 & ~(sizeof(uint32_t)-1))
535 
536 /*
537  * Format an attribute and copy it out to the user's buffer.
538  * Take care to check values and protect against them changing later,
539  * we may be reading them directly out of a user buffer.
540  */
541 STATIC void
542 xfs_attr_put_listent(
543 	xfs_attr_list_context_t *context,
544 	int		flags,
545 	unsigned char	*name,
546 	int		namelen,
547 	int		valuelen)
548 {
549 	struct attrlist *alist = (struct attrlist *)context->alist;
550 	attrlist_ent_t *aep;
551 	int arraytop;
552 
553 	ASSERT(!context->seen_enough);
554 	ASSERT(!(context->flags & ATTR_KERNOVAL));
555 	ASSERT(context->count >= 0);
556 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
557 	ASSERT(context->firstu >= sizeof(*alist));
558 	ASSERT(context->firstu <= context->bufsize);
559 
560 	/*
561 	 * Only list entries in the right namespace.
562 	 */
563 	if (((context->flags & ATTR_SECURE) == 0) !=
564 	    ((flags & XFS_ATTR_SECURE) == 0))
565 		return;
566 	if (((context->flags & ATTR_ROOT) == 0) !=
567 	    ((flags & XFS_ATTR_ROOT) == 0))
568 		return;
569 
570 	arraytop = sizeof(*alist) +
571 			context->count * sizeof(alist->al_offset[0]);
572 	context->firstu -= ATTR_ENTSIZE(namelen);
573 	if (context->firstu < arraytop) {
574 		trace_xfs_attr_list_full(context);
575 		alist->al_more = 1;
576 		context->seen_enough = 1;
577 		return;
578 	}
579 
580 	aep = (attrlist_ent_t *)&context->alist[context->firstu];
581 	aep->a_valuelen = valuelen;
582 	memcpy(aep->a_name, name, namelen);
583 	aep->a_name[namelen] = 0;
584 	alist->al_offset[context->count++] = context->firstu;
585 	alist->al_count = context->count;
586 	trace_xfs_attr_list_add(context);
587 	return;
588 }
589 
590 /*
591  * Generate a list of extended attribute names and optionally
592  * also value lengths.  Positive return value follows the XFS
593  * convention of being an error, zero or negative return code
594  * is the length of the buffer returned (negated), indicating
595  * success.
596  */
597 int
598 xfs_attr_list(
599 	xfs_inode_t	*dp,
600 	char		*buffer,
601 	int		bufsize,
602 	int		flags,
603 	attrlist_cursor_kern_t *cursor)
604 {
605 	xfs_attr_list_context_t context;
606 	struct attrlist *alist;
607 	int error;
608 
609 	/*
610 	 * Validate the cursor.
611 	 */
612 	if (cursor->pad1 || cursor->pad2)
613 		return -EINVAL;
614 	if ((cursor->initted == 0) &&
615 	    (cursor->hashval || cursor->blkno || cursor->offset))
616 		return -EINVAL;
617 
618 	/* Only internal consumers can retrieve incomplete attrs. */
619 	if (flags & ATTR_INCOMPLETE)
620 		return -EINVAL;
621 
622 	/*
623 	 * Check for a properly aligned buffer.
624 	 */
625 	if (((long)buffer) & (sizeof(int)-1))
626 		return -EFAULT;
627 	if (flags & ATTR_KERNOVAL)
628 		bufsize = 0;
629 
630 	/*
631 	 * Initialize the output buffer.
632 	 */
633 	memset(&context, 0, sizeof(context));
634 	context.dp = dp;
635 	context.cursor = cursor;
636 	context.resynch = 1;
637 	context.flags = flags;
638 	context.alist = buffer;
639 	context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
640 	context.firstu = context.bufsize;
641 	context.put_listent = xfs_attr_put_listent;
642 
643 	alist = (struct attrlist *)context.alist;
644 	alist->al_count = 0;
645 	alist->al_more = 0;
646 	alist->al_offset[0] = context.bufsize;
647 
648 	error = xfs_attr_list_int(&context);
649 	ASSERT(error <= 0);
650 	return error;
651 }
652