xref: /linux/fs/xfs/xfs_attr_list.c (revision be239684b18e1cdcafcf8c7face4a2f562c745ad)
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_da_btree.h"
19 #include "xfs_attr.h"
20 #include "xfs_attr_sf.h"
21 #include "xfs_attr_leaf.h"
22 #include "xfs_error.h"
23 #include "xfs_trace.h"
24 #include "xfs_dir2.h"
25 
26 STATIC int
27 xfs_attr_shortform_compare(const void *a, const void *b)
28 {
29 	xfs_attr_sf_sort_t *sa, *sb;
30 
31 	sa = (xfs_attr_sf_sort_t *)a;
32 	sb = (xfs_attr_sf_sort_t *)b;
33 	if (sa->hash < sb->hash) {
34 		return -1;
35 	} else if (sa->hash > sb->hash) {
36 		return 1;
37 	} else {
38 		return sa->entno - sb->entno;
39 	}
40 }
41 
42 #define XFS_ISRESET_CURSOR(cursor) \
43 	(!((cursor)->initted) && !((cursor)->hashval) && \
44 	 !((cursor)->blkno) && !((cursor)->offset))
45 /*
46  * Copy out entries of shortform attribute lists for attr_list().
47  * Shortform attribute lists are not stored in hashval sorted order.
48  * If the output buffer is not large enough to hold them all, then
49  * we have to calculate each entries' hashvalue and sort them before
50  * we can begin returning them to the user.
51  */
52 static int
53 xfs_attr_shortform_list(
54 	struct xfs_attr_list_context	*context)
55 {
56 	struct xfs_attrlist_cursor_kern	*cursor = &context->cursor;
57 	struct xfs_inode		*dp = context->dp;
58 	struct xfs_attr_sf_sort		*sbuf, *sbp;
59 	struct xfs_attr_sf_hdr		*sf = dp->i_af.if_data;
60 	struct xfs_attr_sf_entry	*sfe;
61 	int				sbsize, nsbuf, count, i;
62 	int				error = 0;
63 
64 	ASSERT(sf != NULL);
65 	if (!sf->count)
66 		return 0;
67 
68 	trace_xfs_attr_list_sf(context);
69 
70 	/*
71 	 * If the buffer is large enough and the cursor is at the start,
72 	 * do not bother with sorting since we will return everything in
73 	 * one buffer and another call using the cursor won't need to be
74 	 * made.
75 	 * Note the generous fudge factor of 16 overhead bytes per entry.
76 	 * If bufsize is zero then put_listent must be a search function
77 	 * and can just scan through what we have.
78 	 */
79 	if (context->bufsize == 0 ||
80 	    (XFS_ISRESET_CURSOR(cursor) &&
81 	     (dp->i_af.if_bytes + sf->count * 16) < context->bufsize)) {
82 		for (i = 0, sfe = xfs_attr_sf_firstentry(sf); i < sf->count; i++) {
83 			if (XFS_IS_CORRUPT(context->dp->i_mount,
84 					   !xfs_attr_namecheck(sfe->nameval,
85 							       sfe->namelen)))
86 				return -EFSCORRUPTED;
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->count * sizeof(*sbuf);
112 	sbp = sbuf = kmem_alloc(sbsize, 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 = xfs_attr_sf_firstentry(sf); i < sf->count; i++) {
120 		if (unlikely(
121 		    ((char *)sfe < (char *)sf) ||
122 		    ((char *)sfe >= ((char *)sf + dp->i_af.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 		goto out;
166 
167 	/*
168 	 * Loop putting entries into the user buffer.
169 	 */
170 	for ( ; i < nsbuf; i++, sbp++) {
171 		if (cursor->hashval != sbp->hash) {
172 			cursor->hashval = sbp->hash;
173 			cursor->offset = 0;
174 		}
175 		if (XFS_IS_CORRUPT(context->dp->i_mount,
176 				   !xfs_attr_namecheck(sbp->name,
177 						       sbp->namelen))) {
178 			error = -EFSCORRUPTED;
179 			goto out;
180 		}
181 		context->put_listent(context,
182 				     sbp->flags,
183 				     sbp->name,
184 				     sbp->namelen,
185 				     sbp->valuelen);
186 		if (context->seen_enough)
187 			break;
188 		cursor->offset++;
189 	}
190 out:
191 	kmem_free(sbuf);
192 	return error;
193 }
194 
195 /*
196  * We didn't find the block & hash mentioned in the cursor state, so
197  * walk down the attr btree looking for the hash.
198  */
199 STATIC int
200 xfs_attr_node_list_lookup(
201 	struct xfs_attr_list_context	*context,
202 	struct xfs_attrlist_cursor_kern	*cursor,
203 	struct xfs_buf			**pbp)
204 {
205 	struct xfs_da3_icnode_hdr	nodehdr;
206 	struct xfs_da_intnode		*node;
207 	struct xfs_da_node_entry	*btree;
208 	struct xfs_inode		*dp = context->dp;
209 	struct xfs_mount		*mp = dp->i_mount;
210 	struct xfs_trans		*tp = context->tp;
211 	struct xfs_buf			*bp;
212 	int				i;
213 	int				error = 0;
214 	unsigned int			expected_level = 0;
215 	uint16_t			magic;
216 
217 	ASSERT(*pbp == NULL);
218 	cursor->blkno = 0;
219 	for (;;) {
220 		error = xfs_da3_node_read(tp, dp, cursor->blkno, &bp,
221 				XFS_ATTR_FORK);
222 		if (error)
223 			return error;
224 		node = bp->b_addr;
225 		magic = be16_to_cpu(node->hdr.info.magic);
226 		if (magic == XFS_ATTR_LEAF_MAGIC ||
227 		    magic == XFS_ATTR3_LEAF_MAGIC)
228 			break;
229 		if (magic != XFS_DA_NODE_MAGIC &&
230 		    magic != XFS_DA3_NODE_MAGIC) {
231 			XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
232 					node, sizeof(*node));
233 			goto out_corruptbuf;
234 		}
235 
236 		xfs_da3_node_hdr_from_disk(mp, &nodehdr, node);
237 
238 		/* Tree taller than we can handle; bail out! */
239 		if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
240 			goto out_corruptbuf;
241 
242 		/* Check the level from the root node. */
243 		if (cursor->blkno == 0)
244 			expected_level = nodehdr.level - 1;
245 		else if (expected_level != nodehdr.level)
246 			goto out_corruptbuf;
247 		else
248 			expected_level--;
249 
250 		btree = nodehdr.btree;
251 		for (i = 0; i < nodehdr.count; btree++, i++) {
252 			if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
253 				cursor->blkno = be32_to_cpu(btree->before);
254 				trace_xfs_attr_list_node_descend(context,
255 						btree);
256 				break;
257 			}
258 		}
259 		xfs_trans_brelse(tp, bp);
260 
261 		if (i == nodehdr.count)
262 			return 0;
263 
264 		/* We can't point back to the root. */
265 		if (XFS_IS_CORRUPT(mp, cursor->blkno == 0))
266 			return -EFSCORRUPTED;
267 	}
268 
269 	if (expected_level != 0)
270 		goto out_corruptbuf;
271 
272 	*pbp = bp;
273 	return 0;
274 
275 out_corruptbuf:
276 	xfs_buf_mark_corrupt(bp);
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_attrlist_cursor_kern	*cursor = &context->cursor;
286 	struct xfs_attr3_icleaf_hdr	leafhdr;
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 = 0;
293 
294 	trace_xfs_attr_node_list(context);
295 
296 	cursor->initted = 1;
297 
298 	/*
299 	 * Do all sorts of validation on the passed-in cursor structure.
300 	 * If anything is amiss, ignore the cursor and look up the hashval
301 	 * starting from the btree root.
302 	 */
303 	bp = NULL;
304 	if (cursor->blkno > 0) {
305 		error = xfs_da3_node_read(context->tp, dp, cursor->blkno, &bp,
306 				XFS_ATTR_FORK);
307 		if ((error != 0) && (error != -EFSCORRUPTED))
308 			return error;
309 		if (bp) {
310 			struct xfs_attr_leaf_entry *entries;
311 
312 			node = bp->b_addr;
313 			switch (be16_to_cpu(node->hdr.info.magic)) {
314 			case XFS_DA_NODE_MAGIC:
315 			case XFS_DA3_NODE_MAGIC:
316 				trace_xfs_attr_list_wrong_blk(context);
317 				xfs_trans_brelse(context->tp, bp);
318 				bp = NULL;
319 				break;
320 			case XFS_ATTR_LEAF_MAGIC:
321 			case XFS_ATTR3_LEAF_MAGIC:
322 				leaf = bp->b_addr;
323 				xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
324 							     &leafhdr, leaf);
325 				entries = xfs_attr3_leaf_entryp(leaf);
326 				if (cursor->hashval > be32_to_cpu(
327 						entries[leafhdr.count - 1].hashval)) {
328 					trace_xfs_attr_list_wrong_blk(context);
329 					xfs_trans_brelse(context->tp, bp);
330 					bp = NULL;
331 				} else if (cursor->hashval <= be32_to_cpu(
332 						entries[0].hashval)) {
333 					trace_xfs_attr_list_wrong_blk(context);
334 					xfs_trans_brelse(context->tp, bp);
335 					bp = NULL;
336 				}
337 				break;
338 			default:
339 				trace_xfs_attr_list_wrong_blk(context);
340 				xfs_trans_brelse(context->tp, bp);
341 				bp = NULL;
342 			}
343 		}
344 	}
345 
346 	/*
347 	 * We did not find what we expected given the cursor's contents,
348 	 * so we start from the top and work down based on the hash value.
349 	 * Note that start of node block is same as start of leaf block.
350 	 */
351 	if (bp == NULL) {
352 		error = xfs_attr_node_list_lookup(context, cursor, &bp);
353 		if (error || !bp)
354 			return error;
355 	}
356 	ASSERT(bp != NULL);
357 
358 	/*
359 	 * Roll upward through the blocks, processing each leaf block in
360 	 * order.  As long as there is space in the result buffer, keep
361 	 * adding the information.
362 	 */
363 	for (;;) {
364 		leaf = bp->b_addr;
365 		error = xfs_attr3_leaf_list_int(bp, context);
366 		if (error)
367 			break;
368 		xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
369 		if (context->seen_enough || leafhdr.forw == 0)
370 			break;
371 		cursor->blkno = leafhdr.forw;
372 		xfs_trans_brelse(context->tp, bp);
373 		error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno,
374 					    &bp);
375 		if (error)
376 			return error;
377 	}
378 	xfs_trans_brelse(context->tp, bp);
379 	return error;
380 }
381 
382 /*
383  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
384  */
385 int
386 xfs_attr3_leaf_list_int(
387 	struct xfs_buf			*bp,
388 	struct xfs_attr_list_context	*context)
389 {
390 	struct xfs_attrlist_cursor_kern	*cursor = &context->cursor;
391 	struct xfs_attr_leafblock	*leaf;
392 	struct xfs_attr3_icleaf_hdr	ichdr;
393 	struct xfs_attr_leaf_entry	*entries;
394 	struct xfs_attr_leaf_entry	*entry;
395 	int				i;
396 	struct xfs_mount		*mp = context->dp->i_mount;
397 
398 	trace_xfs_attr_list_leaf(context);
399 
400 	leaf = bp->b_addr;
401 	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
402 	entries = xfs_attr3_leaf_entryp(leaf);
403 
404 	cursor->initted = 1;
405 
406 	/*
407 	 * Re-find our place in the leaf block if this is a new syscall.
408 	 */
409 	if (context->resynch) {
410 		entry = &entries[0];
411 		for (i = 0; i < ichdr.count; entry++, i++) {
412 			if (be32_to_cpu(entry->hashval) == cursor->hashval) {
413 				if (cursor->offset == context->dupcnt) {
414 					context->dupcnt = 0;
415 					break;
416 				}
417 				context->dupcnt++;
418 			} else if (be32_to_cpu(entry->hashval) >
419 					cursor->hashval) {
420 				context->dupcnt = 0;
421 				break;
422 			}
423 		}
424 		if (i == ichdr.count) {
425 			trace_xfs_attr_list_notfound(context);
426 			return 0;
427 		}
428 	} else {
429 		entry = &entries[0];
430 		i = 0;
431 	}
432 	context->resynch = 0;
433 
434 	/*
435 	 * We have found our place, start copying out the new attributes.
436 	 */
437 	for (; i < ichdr.count; entry++, i++) {
438 		char *name;
439 		int namelen, valuelen;
440 
441 		if (be32_to_cpu(entry->hashval) != cursor->hashval) {
442 			cursor->hashval = be32_to_cpu(entry->hashval);
443 			cursor->offset = 0;
444 		}
445 
446 		if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
447 		    !context->allow_incomplete)
448 			continue;
449 
450 		if (entry->flags & XFS_ATTR_LOCAL) {
451 			xfs_attr_leaf_name_local_t *name_loc;
452 
453 			name_loc = xfs_attr3_leaf_name_local(leaf, i);
454 			name = name_loc->nameval;
455 			namelen = name_loc->namelen;
456 			valuelen = be16_to_cpu(name_loc->valuelen);
457 		} else {
458 			xfs_attr_leaf_name_remote_t *name_rmt;
459 
460 			name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
461 			name = name_rmt->name;
462 			namelen = name_rmt->namelen;
463 			valuelen = be32_to_cpu(name_rmt->valuelen);
464 		}
465 
466 		if (XFS_IS_CORRUPT(context->dp->i_mount,
467 				   !xfs_attr_namecheck(name, namelen)))
468 			return -EFSCORRUPTED;
469 		context->put_listent(context, entry->flags,
470 					      name, namelen, valuelen);
471 		if (context->seen_enough)
472 			break;
473 		cursor->offset++;
474 	}
475 	trace_xfs_attr_list_leaf_end(context);
476 	return 0;
477 }
478 
479 /*
480  * Copy out attribute entries for attr_list(), for leaf attribute lists.
481  */
482 STATIC int
483 xfs_attr_leaf_list(
484 	struct xfs_attr_list_context	*context)
485 {
486 	struct xfs_buf			*bp;
487 	int				error;
488 
489 	trace_xfs_attr_leaf_list(context);
490 
491 	context->cursor.blkno = 0;
492 	error = xfs_attr3_leaf_read(context->tp, context->dp, 0, &bp);
493 	if (error)
494 		return error;
495 
496 	error = xfs_attr3_leaf_list_int(bp, context);
497 	xfs_trans_brelse(context->tp, bp);
498 	return error;
499 }
500 
501 int
502 xfs_attr_list_ilocked(
503 	struct xfs_attr_list_context	*context)
504 {
505 	struct xfs_inode		*dp = context->dp;
506 
507 	ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
508 
509 	/*
510 	 * Decide on what work routines to call based on the inode size.
511 	 */
512 	if (!xfs_inode_hasattr(dp))
513 		return 0;
514 	if (dp->i_af.if_format == XFS_DINODE_FMT_LOCAL)
515 		return xfs_attr_shortform_list(context);
516 	if (xfs_attr_is_leaf(dp))
517 		return xfs_attr_leaf_list(context);
518 	return xfs_attr_node_list(context);
519 }
520 
521 int
522 xfs_attr_list(
523 	struct xfs_attr_list_context	*context)
524 {
525 	struct xfs_inode		*dp = context->dp;
526 	uint				lock_mode;
527 	int				error;
528 
529 	XFS_STATS_INC(dp->i_mount, xs_attr_list);
530 
531 	if (xfs_is_shutdown(dp->i_mount))
532 		return -EIO;
533 
534 	lock_mode = xfs_ilock_attr_map_shared(dp);
535 	error = xfs_attr_list_ilocked(context);
536 	xfs_iunlock(dp, lock_mode);
537 	return error;
538 }
539