xref: /openbmc/linux/fs/xfs/xfs_attr_list.c (revision d623f60d) 
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_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_da_format.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_inode.h"
17 #include "xfs_trans.h"
18 #include "xfs_inode_item.h"
19 #include "xfs_bmap.h"
20 #include "xfs_attr.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr_remote.h"
23 #include "xfs_attr_leaf.h"
24 #include "xfs_error.h"
25 #include "xfs_trace.h"
26 #include "xfs_buf_item.h"
27 #include "xfs_cksum.h"
28 #include "xfs_dir2.h"
29 
30 STATIC int
31 xfs_attr_shortform_compare(const void *a, const void *b)
32 {
33 	xfs_attr_sf_sort_t *sa, *sb;
34 
35 	sa = (xfs_attr_sf_sort_t *)a;
36 	sb = (xfs_attr_sf_sort_t *)b;
37 	if (sa->hash < sb->hash) {
38 		return -1;
39 	} else if (sa->hash > sb->hash) {
40 		return 1;
41 	} else {
42 		return sa->entno - sb->entno;
43 	}
44 }
45 
46 #define XFS_ISRESET_CURSOR(cursor) \
47 	(!((cursor)->initted) && !((cursor)->hashval) && \
48 	 !((cursor)->blkno) && !((cursor)->offset))
49 /*
50  * Copy out entries of shortform attribute lists for attr_list().
51  * Shortform attribute lists are not stored in hashval sorted order.
52  * If the output buffer is not large enough to hold them all, then we
53  * we have to calculate each entries' hashvalue and sort them before
54  * we can begin returning them to the user.
55  */
56 static int
57 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
58 {
59 	attrlist_cursor_kern_t *cursor;
60 	xfs_attr_sf_sort_t *sbuf, *sbp;
61 	xfs_attr_shortform_t *sf;
62 	xfs_attr_sf_entry_t *sfe;
63 	xfs_inode_t *dp;
64 	int sbsize, nsbuf, count, i;
65 
66 	ASSERT(context != NULL);
67 	dp = context->dp;
68 	ASSERT(dp != NULL);
69 	ASSERT(dp->i_afp != NULL);
70 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
71 	ASSERT(sf != NULL);
72 	if (!sf->hdr.count)
73 		return 0;
74 	cursor = context->cursor;
75 	ASSERT(cursor != NULL);
76 
77 	trace_xfs_attr_list_sf(context);
78 
79 	/*
80 	 * If the buffer is large enough and the cursor is at the start,
81 	 * do not bother with sorting since we will return everything in
82 	 * one buffer and another call using the cursor won't need to be
83 	 * made.
84 	 * Note the generous fudge factor of 16 overhead bytes per entry.
85 	 * If bufsize is zero then put_listent must be a search function
86 	 * and can just scan through what we have.
87 	 */
88 	if (context->bufsize == 0 ||
89 	    (XFS_ISRESET_CURSOR(cursor) &&
90              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
91 		for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
92 			context->put_listent(context,
93 					     sfe->flags,
94 					     sfe->nameval,
95 					     (int)sfe->namelen,
96 					     (int)sfe->valuelen);
97 			/*
98 			 * Either search callback finished early or
99 			 * didn't fit it all in the buffer after all.
100 			 */
101 			if (context->seen_enough)
102 				break;
103 			sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
104 		}
105 		trace_xfs_attr_list_sf_all(context);
106 		return 0;
107 	}
108 
109 	/* do no more for a search callback */
110 	if (context->bufsize == 0)
111 		return 0;
112 
113 	/*
114 	 * It didn't all fit, so we have to sort everything on hashval.
115 	 */
116 	sbsize = sf->hdr.count * sizeof(*sbuf);
117 	sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
118 
119 	/*
120 	 * Scan the attribute list for the rest of the entries, storing
121 	 * the relevant info from only those that match into a buffer.
122 	 */
123 	nsbuf = 0;
124 	for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
125 		if (unlikely(
126 		    ((char *)sfe < (char *)sf) ||
127 		    ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
128 			XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
129 					     XFS_ERRLEVEL_LOW,
130 					     context->dp->i_mount, sfe,
131 					     sizeof(*sfe));
132 			kmem_free(sbuf);
133 			return -EFSCORRUPTED;
134 		}
135 
136 		sbp->entno = i;
137 		sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
138 		sbp->name = sfe->nameval;
139 		sbp->namelen = sfe->namelen;
140 		/* These are bytes, and both on-disk, don't endian-flip */
141 		sbp->valuelen = sfe->valuelen;
142 		sbp->flags = sfe->flags;
143 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
144 		sbp++;
145 		nsbuf++;
146 	}
147 
148 	/*
149 	 * Sort the entries on hash then entno.
150 	 */
151 	xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
152 
153 	/*
154 	 * Re-find our place IN THE SORTED LIST.
155 	 */
156 	count = 0;
157 	cursor->initted = 1;
158 	cursor->blkno = 0;
159 	for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
160 		if (sbp->hash == cursor->hashval) {
161 			if (cursor->offset == count) {
162 				break;
163 			}
164 			count++;
165 		} else if (sbp->hash > cursor->hashval) {
166 			break;
167 		}
168 	}
169 	if (i == nsbuf) {
170 		kmem_free(sbuf);
171 		return 0;
172 	}
173 
174 	/*
175 	 * Loop putting entries into the user buffer.
176 	 */
177 	for ( ; i < nsbuf; i++, sbp++) {
178 		if (cursor->hashval != sbp->hash) {
179 			cursor->hashval = sbp->hash;
180 			cursor->offset = 0;
181 		}
182 		context->put_listent(context,
183 				     sbp->flags,
184 				     sbp->name,
185 				     sbp->namelen,
186 				     sbp->valuelen);
187 		if (context->seen_enough)
188 			break;
189 		cursor->offset++;
190 	}
191 
192 	kmem_free(sbuf);
193 	return 0;
194 }
195 
196 /*
197  * We didn't find the block & hash mentioned in the cursor state, so
198  * walk down the attr btree looking for the hash.
199  */
200 STATIC int
201 xfs_attr_node_list_lookup(
202 	struct xfs_attr_list_context	*context,
203 	struct attrlist_cursor_kern	*cursor,
204 	struct xfs_buf			**pbp)
205 {
206 	struct xfs_da3_icnode_hdr	nodehdr;
207 	struct xfs_da_intnode		*node;
208 	struct xfs_da_node_entry	*btree;
209 	struct xfs_inode		*dp = context->dp;
210 	struct xfs_mount		*mp = dp->i_mount;
211 	struct xfs_trans		*tp = context->tp;
212 	struct xfs_buf			*bp;
213 	int				i;
214 	int				error = 0;
215 	unsigned int			expected_level = 0;
216 	uint16_t			magic;
217 
218 	ASSERT(*pbp == NULL);
219 	cursor->blkno = 0;
220 	for (;;) {
221 		error = xfs_da3_node_read(tp, dp, cursor->blkno, -1, &bp,
222 				XFS_ATTR_FORK);
223 		if (error)
224 			return error;
225 		node = bp->b_addr;
226 		magic = be16_to_cpu(node->hdr.info.magic);
227 		if (magic == XFS_ATTR_LEAF_MAGIC ||
228 		    magic == XFS_ATTR3_LEAF_MAGIC)
229 			break;
230 		if (magic != XFS_DA_NODE_MAGIC &&
231 		    magic != XFS_DA3_NODE_MAGIC) {
232 			XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
233 					node, sizeof(*node));
234 			goto out_corruptbuf;
235 		}
236 
237 		dp->d_ops->node_hdr_from_disk(&nodehdr, node);
238 
239 		/* Tree taller than we can handle; bail out! */
240 		if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
241 			goto out_corruptbuf;
242 
243 		/* Check the level from the root node. */
244 		if (cursor->blkno == 0)
245 			expected_level = nodehdr.level - 1;
246 		else if (expected_level != nodehdr.level)
247 			goto out_corruptbuf;
248 		else
249 			expected_level--;
250 
251 		btree = dp->d_ops->node_tree_p(node);
252 		for (i = 0; i < nodehdr.count; btree++, i++) {
253 			if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
254 				cursor->blkno = be32_to_cpu(btree->before);
255 				trace_xfs_attr_list_node_descend(context,
256 						btree);
257 				break;
258 			}
259 		}
260 		xfs_trans_brelse(tp, bp);
261 
262 		if (i == nodehdr.count)
263 			return 0;
264 
265 		/* We can't point back to the root. */
266 		if (cursor->blkno == 0)
267 			return -EFSCORRUPTED;
268 	}
269 
270 	if (expected_level != 0)
271 		goto out_corruptbuf;
272 
273 	*pbp = bp;
274 	return 0;
275 
276 out_corruptbuf:
277 	xfs_trans_brelse(tp, bp);
278 	return -EFSCORRUPTED;
279 }
280 
281 STATIC int
282 xfs_attr_node_list(
283 	struct xfs_attr_list_context	*context)
284 {
285 	struct xfs_attr3_icleaf_hdr	leafhdr;
286 	struct attrlist_cursor_kern	*cursor;
287 	struct xfs_attr_leafblock	*leaf;
288 	struct xfs_da_intnode		*node;
289 	struct xfs_buf			*bp;
290 	struct xfs_inode		*dp = context->dp;
291 	struct xfs_mount		*mp = dp->i_mount;
292 	int				error;
293 
294 	trace_xfs_attr_node_list(context);
295 
296 	cursor = context->cursor;
297 	cursor->initted = 1;
298 
299 	/*
300 	 * Do all sorts of validation on the passed-in cursor structure.
301 	 * If anything is amiss, ignore the cursor and look up the hashval
302 	 * starting from the btree root.
303 	 */
304 	bp = NULL;
305 	if (cursor->blkno > 0) {
306 		error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
307 					      &bp, XFS_ATTR_FORK);
308 		if ((error != 0) && (error != -EFSCORRUPTED))
309 			return error;
310 		if (bp) {
311 			struct xfs_attr_leaf_entry *entries;
312 
313 			node = bp->b_addr;
314 			switch (be16_to_cpu(node->hdr.info.magic)) {
315 			case XFS_DA_NODE_MAGIC:
316 			case XFS_DA3_NODE_MAGIC:
317 				trace_xfs_attr_list_wrong_blk(context);
318 				xfs_trans_brelse(context->tp, bp);
319 				bp = NULL;
320 				break;
321 			case XFS_ATTR_LEAF_MAGIC:
322 			case XFS_ATTR3_LEAF_MAGIC:
323 				leaf = bp->b_addr;
324 				xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
325 							     &leafhdr, leaf);
326 				entries = xfs_attr3_leaf_entryp(leaf);
327 				if (cursor->hashval > be32_to_cpu(
328 						entries[leafhdr.count - 1].hashval)) {
329 					trace_xfs_attr_list_wrong_blk(context);
330 					xfs_trans_brelse(context->tp, bp);
331 					bp = NULL;
332 				} else if (cursor->hashval <= be32_to_cpu(
333 						entries[0].hashval)) {
334 					trace_xfs_attr_list_wrong_blk(context);
335 					xfs_trans_brelse(context->tp, bp);
336 					bp = NULL;
337 				}
338 				break;
339 			default:
340 				trace_xfs_attr_list_wrong_blk(context);
341 				xfs_trans_brelse(context->tp, bp);
342 				bp = NULL;
343 			}
344 		}
345 	}
346 
347 	/*
348 	 * We did not find what we expected given the cursor's contents,
349 	 * so we start from the top and work down based on the hash value.
350 	 * Note that start of node block is same as start of leaf block.
351 	 */
352 	if (bp == NULL) {
353 		error = xfs_attr_node_list_lookup(context, cursor, &bp);
354 		if (error || !bp)
355 			return error;
356 	}
357 	ASSERT(bp != NULL);
358 
359 	/*
360 	 * Roll upward through the blocks, processing each leaf block in
361 	 * order.  As long as there is space in the result buffer, keep
362 	 * adding the information.
363 	 */
364 	for (;;) {
365 		leaf = bp->b_addr;
366 		xfs_attr3_leaf_list_int(bp, context);
367 		xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
368 		if (context->seen_enough || leafhdr.forw == 0)
369 			break;
370 		cursor->blkno = leafhdr.forw;
371 		xfs_trans_brelse(context->tp, bp);
372 		error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, -1, &bp);
373 		if (error)
374 			return error;
375 	}
376 	xfs_trans_brelse(context->tp, bp);
377 	return 0;
378 }
379 
380 /*
381  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
382  */
383 void
384 xfs_attr3_leaf_list_int(
385 	struct xfs_buf			*bp,
386 	struct xfs_attr_list_context	*context)
387 {
388 	struct attrlist_cursor_kern	*cursor;
389 	struct xfs_attr_leafblock	*leaf;
390 	struct xfs_attr3_icleaf_hdr	ichdr;
391 	struct xfs_attr_leaf_entry	*entries;
392 	struct xfs_attr_leaf_entry	*entry;
393 	int				i;
394 	struct xfs_mount		*mp = context->dp->i_mount;
395 
396 	trace_xfs_attr_list_leaf(context);
397 
398 	leaf = bp->b_addr;
399 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
400 	entries = xfs_attr3_leaf_entryp(leaf);
401 
402 	cursor = context->cursor;
403 	cursor->initted = 1;
404 
405 	/*
406 	 * Re-find our place in the leaf block if this is a new syscall.
407 	 */
408 	if (context->resynch) {
409 		entry = &entries[0];
410 		for (i = 0; i < ichdr.count; entry++, i++) {
411 			if (be32_to_cpu(entry->hashval) == cursor->hashval) {
412 				if (cursor->offset == context->dupcnt) {
413 					context->dupcnt = 0;
414 					break;
415 				}
416 				context->dupcnt++;
417 			} else if (be32_to_cpu(entry->hashval) >
418 					cursor->hashval) {
419 				context->dupcnt = 0;
420 				break;
421 			}
422 		}
423 		if (i == ichdr.count) {
424 			trace_xfs_attr_list_notfound(context);
425 			return;
426 		}
427 	} else {
428 		entry = &entries[0];
429 		i = 0;
430 	}
431 	context->resynch = 0;
432 
433 	/*
434 	 * We have found our place, start copying out the new attributes.
435 	 */
436 	for (; i < ichdr.count; entry++, i++) {
437 		char *name;
438 		int namelen, valuelen;
439 
440 		if (be32_to_cpu(entry->hashval) != cursor->hashval) {
441 			cursor->hashval = be32_to_cpu(entry->hashval);
442 			cursor->offset = 0;
443 		}
444 
445 		if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
446 		    !(context->flags & ATTR_INCOMPLETE))
447 			continue;		/* skip incomplete entries */
448 
449 		if (entry->flags & XFS_ATTR_LOCAL) {
450 			xfs_attr_leaf_name_local_t *name_loc;
451 
452 			name_loc = xfs_attr3_leaf_name_local(leaf, i);
453 			name = name_loc->nameval;
454 			namelen = name_loc->namelen;
455 			valuelen = be16_to_cpu(name_loc->valuelen);
456 		} else {
457 			xfs_attr_leaf_name_remote_t *name_rmt;
458 
459 			name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
460 			name = name_rmt->name;
461 			namelen = name_rmt->namelen;
462 			valuelen = be32_to_cpu(name_rmt->valuelen);
463 		}
464 
465 		context->put_listent(context, entry->flags,
466 					      name, namelen, valuelen);
467 		if (context->seen_enough)
468 			break;
469 		cursor->offset++;
470 	}
471 	trace_xfs_attr_list_leaf_end(context);
472 	return;
473 }
474 
475 /*
476  * Copy out attribute entries for attr_list(), for leaf attribute lists.
477  */
478 STATIC int
479 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
480 {
481 	int error;
482 	struct xfs_buf *bp;
483 
484 	trace_xfs_attr_leaf_list(context);
485 
486 	context->cursor->blkno = 0;
487 	error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
488 	if (error)
489 		return error;
490 
491 	xfs_attr3_leaf_list_int(bp, context);
492 	xfs_trans_brelse(context->tp, bp);
493 	return 0;
494 }
495 
496 int
497 xfs_attr_list_int_ilocked(
498 	struct xfs_attr_list_context	*context)
499 {
500 	struct xfs_inode		*dp = context->dp;
501 
502 	ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
503 
504 	/*
505 	 * Decide on what work routines to call based on the inode size.
506 	 */
507 	if (!xfs_inode_hasattr(dp))
508 		return 0;
509 	else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
510 		return xfs_attr_shortform_list(context);
511 	else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
512 		return xfs_attr_leaf_list(context);
513 	return xfs_attr_node_list(context);
514 }
515 
516 int
517 xfs_attr_list_int(
518 	xfs_attr_list_context_t *context)
519 {
520 	int error;
521 	xfs_inode_t *dp = context->dp;
522 	uint		lock_mode;
523 
524 	XFS_STATS_INC(dp->i_mount, xs_attr_list);
525 
526 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
527 		return -EIO;
528 
529 	lock_mode = xfs_ilock_attr_map_shared(dp);
530 	error = xfs_attr_list_int_ilocked(context);
531 	xfs_iunlock(dp, lock_mode);
532 	return error;
533 }
534 
535 #define	ATTR_ENTBASESIZE		/* minimum bytes used by an attr */ \
536 	(((struct attrlist_ent *) 0)->a_name - (char *) 0)
537 #define	ATTR_ENTSIZE(namelen)		/* actual bytes used by an attr */ \
538 	((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(uint32_t)-1) \
539 	 & ~(sizeof(uint32_t)-1))
540 
541 /*
542  * Format an attribute and copy it out to the user's buffer.
543  * Take care to check values and protect against them changing later,
544  * we may be reading them directly out of a user buffer.
545  */
546 STATIC void
547 xfs_attr_put_listent(
548 	xfs_attr_list_context_t *context,
549 	int		flags,
550 	unsigned char	*name,
551 	int		namelen,
552 	int		valuelen)
553 {
554 	struct attrlist *alist = (struct attrlist *)context->alist;
555 	attrlist_ent_t *aep;
556 	int arraytop;
557 
558 	ASSERT(!(context->flags & ATTR_KERNOVAL));
559 	ASSERT(context->count >= 0);
560 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
561 	ASSERT(context->firstu >= sizeof(*alist));
562 	ASSERT(context->firstu <= context->bufsize);
563 
564 	/*
565 	 * Only list entries in the right namespace.
566 	 */
567 	if (((context->flags & ATTR_SECURE) == 0) !=
568 	    ((flags & XFS_ATTR_SECURE) == 0))
569 		return;
570 	if (((context->flags & ATTR_ROOT) == 0) !=
571 	    ((flags & XFS_ATTR_ROOT) == 0))
572 		return;
573 
574 	arraytop = sizeof(*alist) +
575 			context->count * sizeof(alist->al_offset[0]);
576 	context->firstu -= ATTR_ENTSIZE(namelen);
577 	if (context->firstu < arraytop) {
578 		trace_xfs_attr_list_full(context);
579 		alist->al_more = 1;
580 		context->seen_enough = 1;
581 		return;
582 	}
583 
584 	aep = (attrlist_ent_t *)&context->alist[context->firstu];
585 	aep->a_valuelen = valuelen;
586 	memcpy(aep->a_name, name, namelen);
587 	aep->a_name[namelen] = 0;
588 	alist->al_offset[context->count++] = context->firstu;
589 	alist->al_count = context->count;
590 	trace_xfs_attr_list_add(context);
591 	return;
592 }
593 
594 /*
595  * Generate a list of extended attribute names and optionally
596  * also value lengths.  Positive return value follows the XFS
597  * convention of being an error, zero or negative return code
598  * is the length of the buffer returned (negated), indicating
599  * success.
600  */
601 int
602 xfs_attr_list(
603 	xfs_inode_t	*dp,
604 	char		*buffer,
605 	int		bufsize,
606 	int		flags,
607 	attrlist_cursor_kern_t *cursor)
608 {
609 	xfs_attr_list_context_t context;
610 	struct attrlist *alist;
611 	int error;
612 
613 	/*
614 	 * Validate the cursor.
615 	 */
616 	if (cursor->pad1 || cursor->pad2)
617 		return -EINVAL;
618 	if ((cursor->initted == 0) &&
619 	    (cursor->hashval || cursor->blkno || cursor->offset))
620 		return -EINVAL;
621 
622 	/* Only internal consumers can retrieve incomplete attrs. */
623 	if (flags & ATTR_INCOMPLETE)
624 		return -EINVAL;
625 
626 	/*
627 	 * Check for a properly aligned buffer.
628 	 */
629 	if (((long)buffer) & (sizeof(int)-1))
630 		return -EFAULT;
631 	if (flags & ATTR_KERNOVAL)
632 		bufsize = 0;
633 
634 	/*
635 	 * Initialize the output buffer.
636 	 */
637 	memset(&context, 0, sizeof(context));
638 	context.dp = dp;
639 	context.cursor = cursor;
640 	context.resynch = 1;
641 	context.flags = flags;
642 	context.alist = buffer;
643 	context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
644 	context.firstu = context.bufsize;
645 	context.put_listent = xfs_attr_put_listent;
646 
647 	alist = (struct attrlist *)context.alist;
648 	alist->al_count = 0;
649 	alist->al_more = 0;
650 	alist->al_offset[0] = context.bufsize;
651 
652 	error = xfs_attr_list_int(&context);
653 	ASSERT(error <= 0);
654 	return error;
655 }
656