xref: /openbmc/linux/fs/xfs/xfs_attr_list.c (revision 15b7cc78)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_inode.h"
29 #include "xfs_trans.h"
30 #include "xfs_inode_item.h"
31 #include "xfs_bmap.h"
32 #include "xfs_attr.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_attr_remote.h"
35 #include "xfs_attr_leaf.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_cksum.h"
40 #include "xfs_dir2.h"
41 
42 STATIC int
43 xfs_attr_shortform_compare(const void *a, const void *b)
44 {
45 	xfs_attr_sf_sort_t *sa, *sb;
46 
47 	sa = (xfs_attr_sf_sort_t *)a;
48 	sb = (xfs_attr_sf_sort_t *)b;
49 	if (sa->hash < sb->hash) {
50 		return -1;
51 	} else if (sa->hash > sb->hash) {
52 		return 1;
53 	} else {
54 		return sa->entno - sb->entno;
55 	}
56 }
57 
58 #define XFS_ISRESET_CURSOR(cursor) \
59 	(!((cursor)->initted) && !((cursor)->hashval) && \
60 	 !((cursor)->blkno) && !((cursor)->offset))
61 /*
62  * Copy out entries of shortform attribute lists for attr_list().
63  * Shortform attribute lists are not stored in hashval sorted order.
64  * If the output buffer is not large enough to hold them all, then we
65  * we have to calculate each entries' hashvalue and sort them before
66  * we can begin returning them to the user.
67  */
68 int
69 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
70 {
71 	attrlist_cursor_kern_t *cursor;
72 	xfs_attr_sf_sort_t *sbuf, *sbp;
73 	xfs_attr_shortform_t *sf;
74 	xfs_attr_sf_entry_t *sfe;
75 	xfs_inode_t *dp;
76 	int sbsize, nsbuf, count, i;
77 	int error;
78 
79 	ASSERT(context != NULL);
80 	dp = context->dp;
81 	ASSERT(dp != NULL);
82 	ASSERT(dp->i_afp != NULL);
83 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
84 	ASSERT(sf != NULL);
85 	if (!sf->hdr.count)
86 		return 0;
87 	cursor = context->cursor;
88 	ASSERT(cursor != NULL);
89 
90 	trace_xfs_attr_list_sf(context);
91 
92 	/*
93 	 * If the buffer is large enough and the cursor is at the start,
94 	 * do not bother with sorting since we will return everything in
95 	 * one buffer and another call using the cursor won't need to be
96 	 * made.
97 	 * Note the generous fudge factor of 16 overhead bytes per entry.
98 	 * If bufsize is zero then put_listent must be a search function
99 	 * and can just scan through what we have.
100 	 */
101 	if (context->bufsize == 0 ||
102 	    (XFS_ISRESET_CURSOR(cursor) &&
103              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
104 		for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
105 			error = context->put_listent(context,
106 					   sfe->flags,
107 					   sfe->nameval,
108 					   (int)sfe->namelen,
109 					   (int)sfe->valuelen);
110 			if (error)
111 				return error;
112 			/*
113 			 * Either search callback finished early or
114 			 * didn't fit it all in the buffer after all.
115 			 */
116 			if (context->seen_enough)
117 				break;
118 			sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
119 		}
120 		trace_xfs_attr_list_sf_all(context);
121 		return 0;
122 	}
123 
124 	/* do no more for a search callback */
125 	if (context->bufsize == 0)
126 		return 0;
127 
128 	/*
129 	 * It didn't all fit, so we have to sort everything on hashval.
130 	 */
131 	sbsize = sf->hdr.count * sizeof(*sbuf);
132 	sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
133 
134 	/*
135 	 * Scan the attribute list for the rest of the entries, storing
136 	 * the relevant info from only those that match into a buffer.
137 	 */
138 	nsbuf = 0;
139 	for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
140 		if (unlikely(
141 		    ((char *)sfe < (char *)sf) ||
142 		    ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
143 			XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
144 					     XFS_ERRLEVEL_LOW,
145 					     context->dp->i_mount, sfe);
146 			kmem_free(sbuf);
147 			return -EFSCORRUPTED;
148 		}
149 
150 		sbp->entno = i;
151 		sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
152 		sbp->name = sfe->nameval;
153 		sbp->namelen = sfe->namelen;
154 		/* These are bytes, and both on-disk, don't endian-flip */
155 		sbp->valuelen = sfe->valuelen;
156 		sbp->flags = sfe->flags;
157 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
158 		sbp++;
159 		nsbuf++;
160 	}
161 
162 	/*
163 	 * Sort the entries on hash then entno.
164 	 */
165 	xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
166 
167 	/*
168 	 * Re-find our place IN THE SORTED LIST.
169 	 */
170 	count = 0;
171 	cursor->initted = 1;
172 	cursor->blkno = 0;
173 	for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
174 		if (sbp->hash == cursor->hashval) {
175 			if (cursor->offset == count) {
176 				break;
177 			}
178 			count++;
179 		} else if (sbp->hash > cursor->hashval) {
180 			break;
181 		}
182 	}
183 	if (i == nsbuf) {
184 		kmem_free(sbuf);
185 		return 0;
186 	}
187 
188 	/*
189 	 * Loop putting entries into the user buffer.
190 	 */
191 	for ( ; i < nsbuf; i++, sbp++) {
192 		if (cursor->hashval != sbp->hash) {
193 			cursor->hashval = sbp->hash;
194 			cursor->offset = 0;
195 		}
196 		error = context->put_listent(context,
197 					sbp->flags,
198 					sbp->name,
199 					sbp->namelen,
200 					sbp->valuelen);
201 		if (error) {
202 			kmem_free(sbuf);
203 			return error;
204 		}
205 		if (context->seen_enough)
206 			break;
207 		cursor->offset++;
208 	}
209 
210 	kmem_free(sbuf);
211 	return 0;
212 }
213 
214 STATIC int
215 xfs_attr_node_list(xfs_attr_list_context_t *context)
216 {
217 	attrlist_cursor_kern_t *cursor;
218 	xfs_attr_leafblock_t *leaf;
219 	xfs_da_intnode_t *node;
220 	struct xfs_attr3_icleaf_hdr leafhdr;
221 	struct xfs_da3_icnode_hdr nodehdr;
222 	struct xfs_da_node_entry *btree;
223 	int error, i;
224 	struct xfs_buf *bp;
225 	struct xfs_inode	*dp = context->dp;
226 	struct xfs_mount	*mp = dp->i_mount;
227 
228 	trace_xfs_attr_node_list(context);
229 
230 	cursor = context->cursor;
231 	cursor->initted = 1;
232 
233 	/*
234 	 * Do all sorts of validation on the passed-in cursor structure.
235 	 * If anything is amiss, ignore the cursor and look up the hashval
236 	 * starting from the btree root.
237 	 */
238 	bp = NULL;
239 	if (cursor->blkno > 0) {
240 		error = xfs_da3_node_read(NULL, dp, cursor->blkno, -1,
241 					      &bp, XFS_ATTR_FORK);
242 		if ((error != 0) && (error != -EFSCORRUPTED))
243 			return error;
244 		if (bp) {
245 			struct xfs_attr_leaf_entry *entries;
246 
247 			node = bp->b_addr;
248 			switch (be16_to_cpu(node->hdr.info.magic)) {
249 			case XFS_DA_NODE_MAGIC:
250 			case XFS_DA3_NODE_MAGIC:
251 				trace_xfs_attr_list_wrong_blk(context);
252 				xfs_trans_brelse(NULL, bp);
253 				bp = NULL;
254 				break;
255 			case XFS_ATTR_LEAF_MAGIC:
256 			case XFS_ATTR3_LEAF_MAGIC:
257 				leaf = bp->b_addr;
258 				xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
259 							     &leafhdr, leaf);
260 				entries = xfs_attr3_leaf_entryp(leaf);
261 				if (cursor->hashval > be32_to_cpu(
262 						entries[leafhdr.count - 1].hashval)) {
263 					trace_xfs_attr_list_wrong_blk(context);
264 					xfs_trans_brelse(NULL, bp);
265 					bp = NULL;
266 				} else if (cursor->hashval <= be32_to_cpu(
267 						entries[0].hashval)) {
268 					trace_xfs_attr_list_wrong_blk(context);
269 					xfs_trans_brelse(NULL, bp);
270 					bp = NULL;
271 				}
272 				break;
273 			default:
274 				trace_xfs_attr_list_wrong_blk(context);
275 				xfs_trans_brelse(NULL, bp);
276 				bp = NULL;
277 			}
278 		}
279 	}
280 
281 	/*
282 	 * We did not find what we expected given the cursor's contents,
283 	 * so we start from the top and work down based on the hash value.
284 	 * Note that start of node block is same as start of leaf block.
285 	 */
286 	if (bp == NULL) {
287 		cursor->blkno = 0;
288 		for (;;) {
289 			__uint16_t magic;
290 
291 			error = xfs_da3_node_read(NULL, dp,
292 						      cursor->blkno, -1, &bp,
293 						      XFS_ATTR_FORK);
294 			if (error)
295 				return error;
296 			node = bp->b_addr;
297 			magic = be16_to_cpu(node->hdr.info.magic);
298 			if (magic == XFS_ATTR_LEAF_MAGIC ||
299 			    magic == XFS_ATTR3_LEAF_MAGIC)
300 				break;
301 			if (magic != XFS_DA_NODE_MAGIC &&
302 			    magic != XFS_DA3_NODE_MAGIC) {
303 				XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
304 						     XFS_ERRLEVEL_LOW,
305 						     context->dp->i_mount,
306 						     node);
307 				xfs_trans_brelse(NULL, bp);
308 				return -EFSCORRUPTED;
309 			}
310 
311 			dp->d_ops->node_hdr_from_disk(&nodehdr, node);
312 			btree = dp->d_ops->node_tree_p(node);
313 			for (i = 0; i < nodehdr.count; btree++, i++) {
314 				if (cursor->hashval
315 						<= be32_to_cpu(btree->hashval)) {
316 					cursor->blkno = be32_to_cpu(btree->before);
317 					trace_xfs_attr_list_node_descend(context,
318 									 btree);
319 					break;
320 				}
321 			}
322 			if (i == nodehdr.count) {
323 				xfs_trans_brelse(NULL, bp);
324 				return 0;
325 			}
326 			xfs_trans_brelse(NULL, bp);
327 		}
328 	}
329 	ASSERT(bp != NULL);
330 
331 	/*
332 	 * Roll upward through the blocks, processing each leaf block in
333 	 * order.  As long as there is space in the result buffer, keep
334 	 * adding the information.
335 	 */
336 	for (;;) {
337 		leaf = bp->b_addr;
338 		error = xfs_attr3_leaf_list_int(bp, context);
339 		if (error) {
340 			xfs_trans_brelse(NULL, bp);
341 			return error;
342 		}
343 		xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
344 		if (context->seen_enough || leafhdr.forw == 0)
345 			break;
346 		cursor->blkno = leafhdr.forw;
347 		xfs_trans_brelse(NULL, bp);
348 		error = xfs_attr3_leaf_read(NULL, dp, cursor->blkno, -1, &bp);
349 		if (error)
350 			return error;
351 	}
352 	xfs_trans_brelse(NULL, bp);
353 	return 0;
354 }
355 
356 /*
357  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
358  */
359 int
360 xfs_attr3_leaf_list_int(
361 	struct xfs_buf			*bp,
362 	struct xfs_attr_list_context	*context)
363 {
364 	struct attrlist_cursor_kern	*cursor;
365 	struct xfs_attr_leafblock	*leaf;
366 	struct xfs_attr3_icleaf_hdr	ichdr;
367 	struct xfs_attr_leaf_entry	*entries;
368 	struct xfs_attr_leaf_entry	*entry;
369 	int				retval;
370 	int				i;
371 	struct xfs_mount		*mp = context->dp->i_mount;
372 
373 	trace_xfs_attr_list_leaf(context);
374 
375 	leaf = bp->b_addr;
376 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
377 	entries = xfs_attr3_leaf_entryp(leaf);
378 
379 	cursor = context->cursor;
380 	cursor->initted = 1;
381 
382 	/*
383 	 * Re-find our place in the leaf block if this is a new syscall.
384 	 */
385 	if (context->resynch) {
386 		entry = &entries[0];
387 		for (i = 0; i < ichdr.count; entry++, i++) {
388 			if (be32_to_cpu(entry->hashval) == cursor->hashval) {
389 				if (cursor->offset == context->dupcnt) {
390 					context->dupcnt = 0;
391 					break;
392 				}
393 				context->dupcnt++;
394 			} else if (be32_to_cpu(entry->hashval) >
395 					cursor->hashval) {
396 				context->dupcnt = 0;
397 				break;
398 			}
399 		}
400 		if (i == ichdr.count) {
401 			trace_xfs_attr_list_notfound(context);
402 			return 0;
403 		}
404 	} else {
405 		entry = &entries[0];
406 		i = 0;
407 	}
408 	context->resynch = 0;
409 
410 	/*
411 	 * We have found our place, start copying out the new attributes.
412 	 */
413 	retval = 0;
414 	for (; i < ichdr.count; entry++, i++) {
415 		char *name;
416 		int namelen, valuelen;
417 
418 		if (be32_to_cpu(entry->hashval) != cursor->hashval) {
419 			cursor->hashval = be32_to_cpu(entry->hashval);
420 			cursor->offset = 0;
421 		}
422 
423 		if (entry->flags & XFS_ATTR_INCOMPLETE)
424 			continue;		/* skip incomplete entries */
425 
426 		if (entry->flags & XFS_ATTR_LOCAL) {
427 			xfs_attr_leaf_name_local_t *name_loc;
428 
429 			name_loc = xfs_attr3_leaf_name_local(leaf, i);
430 			name = name_loc->nameval;
431 			namelen = name_loc->namelen;
432 			valuelen = be16_to_cpu(name_loc->valuelen);
433 		} else {
434 			xfs_attr_leaf_name_remote_t *name_rmt;
435 
436 			name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
437 			name = name_rmt->name;
438 			namelen = name_rmt->namelen;
439 			valuelen = be32_to_cpu(name_rmt->valuelen);
440 		}
441 
442 		retval = context->put_listent(context, entry->flags,
443 					      name, namelen, valuelen);
444 		if (retval)
445 			break;
446 		if (context->seen_enough)
447 			break;
448 		cursor->offset++;
449 	}
450 	trace_xfs_attr_list_leaf_end(context);
451 	return retval;
452 }
453 
454 /*
455  * Copy out attribute entries for attr_list(), for leaf attribute lists.
456  */
457 STATIC int
458 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
459 {
460 	int error;
461 	struct xfs_buf *bp;
462 
463 	trace_xfs_attr_leaf_list(context);
464 
465 	context->cursor->blkno = 0;
466 	error = xfs_attr3_leaf_read(NULL, context->dp, 0, -1, &bp);
467 	if (error)
468 		return error;
469 
470 	error = xfs_attr3_leaf_list_int(bp, context);
471 	xfs_trans_brelse(NULL, bp);
472 	return error;
473 }
474 
475 int
476 xfs_attr_list_int(
477 	xfs_attr_list_context_t *context)
478 {
479 	int error;
480 	xfs_inode_t *dp = context->dp;
481 	uint		lock_mode;
482 
483 	XFS_STATS_INC(dp->i_mount, xs_attr_list);
484 
485 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
486 		return -EIO;
487 
488 	/*
489 	 * Decide on what work routines to call based on the inode size.
490 	 */
491 	lock_mode = xfs_ilock_attr_map_shared(dp);
492 	if (!xfs_inode_hasattr(dp)) {
493 		error = 0;
494 	} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
495 		error = xfs_attr_shortform_list(context);
496 	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
497 		error = xfs_attr_leaf_list(context);
498 	} else {
499 		error = xfs_attr_node_list(context);
500 	}
501 	xfs_iunlock(dp, lock_mode);
502 	return error;
503 }
504 
505 #define	ATTR_ENTBASESIZE		/* minimum bytes used by an attr */ \
506 	(((struct attrlist_ent *) 0)->a_name - (char *) 0)
507 #define	ATTR_ENTSIZE(namelen)		/* actual bytes used by an attr */ \
508 	((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
509 	 & ~(sizeof(u_int32_t)-1))
510 
511 /*
512  * Format an attribute and copy it out to the user's buffer.
513  * Take care to check values and protect against them changing later,
514  * we may be reading them directly out of a user buffer.
515  */
516 STATIC int
517 xfs_attr_put_listent(
518 	xfs_attr_list_context_t *context,
519 	int		flags,
520 	unsigned char	*name,
521 	int		namelen,
522 	int		valuelen)
523 {
524 	struct attrlist *alist = (struct attrlist *)context->alist;
525 	attrlist_ent_t *aep;
526 	int arraytop;
527 
528 	ASSERT(!(context->flags & ATTR_KERNOVAL));
529 	ASSERT(context->count >= 0);
530 	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
531 	ASSERT(context->firstu >= sizeof(*alist));
532 	ASSERT(context->firstu <= context->bufsize);
533 
534 	/*
535 	 * Only list entries in the right namespace.
536 	 */
537 	if (((context->flags & ATTR_SECURE) == 0) !=
538 	    ((flags & XFS_ATTR_SECURE) == 0))
539 		return 0;
540 	if (((context->flags & ATTR_ROOT) == 0) !=
541 	    ((flags & XFS_ATTR_ROOT) == 0))
542 		return 0;
543 
544 	arraytop = sizeof(*alist) +
545 			context->count * sizeof(alist->al_offset[0]);
546 	context->firstu -= ATTR_ENTSIZE(namelen);
547 	if (context->firstu < arraytop) {
548 		trace_xfs_attr_list_full(context);
549 		alist->al_more = 1;
550 		context->seen_enough = 1;
551 		return 0;
552 	}
553 
554 	aep = (attrlist_ent_t *)&context->alist[context->firstu];
555 	aep->a_valuelen = valuelen;
556 	memcpy(aep->a_name, name, namelen);
557 	aep->a_name[namelen] = 0;
558 	alist->al_offset[context->count++] = context->firstu;
559 	alist->al_count = context->count;
560 	trace_xfs_attr_list_add(context);
561 	return 0;
562 }
563 
564 /*
565  * Generate a list of extended attribute names and optionally
566  * also value lengths.  Positive return value follows the XFS
567  * convention of being an error, zero or negative return code
568  * is the length of the buffer returned (negated), indicating
569  * success.
570  */
571 int
572 xfs_attr_list(
573 	xfs_inode_t	*dp,
574 	char		*buffer,
575 	int		bufsize,
576 	int		flags,
577 	attrlist_cursor_kern_t *cursor)
578 {
579 	xfs_attr_list_context_t context;
580 	struct attrlist *alist;
581 	int error;
582 
583 	/*
584 	 * Validate the cursor.
585 	 */
586 	if (cursor->pad1 || cursor->pad2)
587 		return -EINVAL;
588 	if ((cursor->initted == 0) &&
589 	    (cursor->hashval || cursor->blkno || cursor->offset))
590 		return -EINVAL;
591 
592 	/*
593 	 * Check for a properly aligned buffer.
594 	 */
595 	if (((long)buffer) & (sizeof(int)-1))
596 		return -EFAULT;
597 	if (flags & ATTR_KERNOVAL)
598 		bufsize = 0;
599 
600 	/*
601 	 * Initialize the output buffer.
602 	 */
603 	memset(&context, 0, sizeof(context));
604 	context.dp = dp;
605 	context.cursor = cursor;
606 	context.resynch = 1;
607 	context.flags = flags;
608 	context.alist = buffer;
609 	context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
610 	context.firstu = context.bufsize;
611 	context.put_listent = xfs_attr_put_listent;
612 
613 	alist = (struct attrlist *)context.alist;
614 	alist->al_count = 0;
615 	alist->al_more = 0;
616 	alist->al_offset[0] = context.bufsize;
617 
618 	error = xfs_attr_list_int(&context);
619 	ASSERT(error <= 0);
620 	return error;
621 }
622