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#include "xfs_health.h"
26
27STATIC int
28xfs_attr_shortform_compare(const void *a, const void *b)
29{
30	xfs_attr_sf_sort_t *sa, *sb;
31
32	sa = (xfs_attr_sf_sort_t *)a;
33	sb = (xfs_attr_sf_sort_t *)b;
34	if (sa->hash < sb->hash) {
35		return -1;
36	} else if (sa->hash > sb->hash) {
37		return 1;
38	} else {
39		return sa->entno - sb->entno;
40	}
41}
42
43#define XFS_ISRESET_CURSOR(cursor) \
44	(!((cursor)->initted) && !((cursor)->hashval) && \
45	 !((cursor)->blkno) && !((cursor)->offset))
46/*
47 * Copy out entries of shortform attribute lists for attr_list().
48 * Shortform attribute lists are not stored in hashval sorted order.
49 * If the output buffer is not large enough to hold them all, then
50 * we have to calculate each entries' hashvalue and sort them before
51 * we can begin returning them to the user.
52 */
53static int
54xfs_attr_shortform_list(
55	struct xfs_attr_list_context	*context)
56{
57	struct xfs_attrlist_cursor_kern	*cursor = &context->cursor;
58	struct xfs_inode		*dp = context->dp;
59	struct xfs_attr_sf_sort		*sbuf, *sbp;
60	struct xfs_attr_sf_hdr		*sf = dp->i_af.if_data;
61	struct xfs_attr_sf_entry	*sfe;
62	int				sbsize, nsbuf, count, i;
63	int				error = 0;
64
65	ASSERT(sf != NULL);
66	if (!sf->count)
67		return 0;
68
69	trace_xfs_attr_list_sf(context);
70
71	/*
72	 * If the buffer is large enough and the cursor is at the start,
73	 * do not bother with sorting since we will return everything in
74	 * one buffer and another call using the cursor won't need to be
75	 * made.
76	 * Note the generous fudge factor of 16 overhead bytes per entry.
77	 * If bufsize is zero then put_listent must be a search function
78	 * and can just scan through what we have.
79	 */
80	if (context->bufsize == 0 ||
81	    (XFS_ISRESET_CURSOR(cursor) &&
82	     (dp->i_af.if_bytes + sf->count * 16) < context->bufsize)) {
83		for (i = 0, sfe = xfs_attr_sf_firstentry(sf); i < sf->count; i++) {
84			if (XFS_IS_CORRUPT(context->dp->i_mount,
85					   !xfs_attr_namecheck(sfe->nameval,
86							       sfe->namelen))) {
87				xfs_dirattr_mark_sick(context->dp, XFS_ATTR_FORK);
88				return -EFSCORRUPTED;
89			}
90			context->put_listent(context,
91					     sfe->flags,
92					     sfe->nameval,
93					     (int)sfe->namelen,
94					     (int)sfe->valuelen);
95			/*
96			 * Either search callback finished early or
97			 * didn't fit it all in the buffer after all.
98			 */
99			if (context->seen_enough)
100				break;
101			sfe = xfs_attr_sf_nextentry(sfe);
102		}
103		trace_xfs_attr_list_sf_all(context);
104		return 0;
105	}
106
107	/* do no more for a search callback */
108	if (context->bufsize == 0)
109		return 0;
110
111	/*
112	 * It didn't all fit, so we have to sort everything on hashval.
113	 */
114	sbsize = sf->count * sizeof(*sbuf);
115	sbp = sbuf = kmalloc(sbsize, GFP_KERNEL | __GFP_NOFAIL);
116
117	/*
118	 * Scan the attribute list for the rest of the entries, storing
119	 * the relevant info from only those that match into a buffer.
120	 */
121	nsbuf = 0;
122	for (i = 0, sfe = xfs_attr_sf_firstentry(sf); i < sf->count; i++) {
123		if (unlikely(
124		    ((char *)sfe < (char *)sf) ||
125		    ((char *)sfe >= ((char *)sf + dp->i_af.if_bytes)))) {
126			XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
127					     XFS_ERRLEVEL_LOW,
128					     context->dp->i_mount, sfe,
129					     sizeof(*sfe));
130			kfree(sbuf);
131			xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
132			return -EFSCORRUPTED;
133		}
134
135		sbp->entno = i;
136		sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
137		sbp->name = sfe->nameval;
138		sbp->namelen = sfe->namelen;
139		/* These are bytes, and both on-disk, don't endian-flip */
140		sbp->valuelen = sfe->valuelen;
141		sbp->flags = sfe->flags;
142		sfe = xfs_attr_sf_nextentry(sfe);
143		sbp++;
144		nsbuf++;
145	}
146
147	/*
148	 * Sort the entries on hash then entno.
149	 */
150	xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
151
152	/*
153	 * Re-find our place IN THE SORTED LIST.
154	 */
155	count = 0;
156	cursor->initted = 1;
157	cursor->blkno = 0;
158	for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
159		if (sbp->hash == cursor->hashval) {
160			if (cursor->offset == count) {
161				break;
162			}
163			count++;
164		} else if (sbp->hash > cursor->hashval) {
165			break;
166		}
167	}
168	if (i == nsbuf)
169		goto out;
170
171	/*
172	 * Loop putting entries into the user buffer.
173	 */
174	for ( ; i < nsbuf; i++, sbp++) {
175		if (cursor->hashval != sbp->hash) {
176			cursor->hashval = sbp->hash;
177			cursor->offset = 0;
178		}
179		if (XFS_IS_CORRUPT(context->dp->i_mount,
180				   !xfs_attr_namecheck(sbp->name,
181						       sbp->namelen))) {
182			xfs_dirattr_mark_sick(context->dp, XFS_ATTR_FORK);
183			error = -EFSCORRUPTED;
184			goto out;
185		}
186		context->put_listent(context,
187				     sbp->flags,
188				     sbp->name,
189				     sbp->namelen,
190				     sbp->valuelen);
191		if (context->seen_enough)
192			break;
193		cursor->offset++;
194	}
195out:
196	kfree(sbuf);
197	return error;
198}
199
200/*
201 * We didn't find the block & hash mentioned in the cursor state, so
202 * walk down the attr btree looking for the hash.
203 */
204STATIC int
205xfs_attr_node_list_lookup(
206	struct xfs_attr_list_context	*context,
207	struct xfs_attrlist_cursor_kern	*cursor,
208	struct xfs_buf			**pbp)
209{
210	struct xfs_da3_icnode_hdr	nodehdr;
211	struct xfs_da_intnode		*node;
212	struct xfs_da_node_entry	*btree;
213	struct xfs_inode		*dp = context->dp;
214	struct xfs_mount		*mp = dp->i_mount;
215	struct xfs_trans		*tp = context->tp;
216	struct xfs_buf			*bp;
217	int				i;
218	int				error = 0;
219	unsigned int			expected_level = 0;
220	uint16_t			magic;
221
222	ASSERT(*pbp == NULL);
223	cursor->blkno = 0;
224	for (;;) {
225		error = xfs_da3_node_read(tp, dp, cursor->blkno, &bp,
226				XFS_ATTR_FORK);
227		if (error)
228			return error;
229		node = bp->b_addr;
230		magic = be16_to_cpu(node->hdr.info.magic);
231		if (magic == XFS_ATTR_LEAF_MAGIC ||
232		    magic == XFS_ATTR3_LEAF_MAGIC)
233			break;
234		if (magic != XFS_DA_NODE_MAGIC &&
235		    magic != XFS_DA3_NODE_MAGIC) {
236			XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
237					node, sizeof(*node));
238			goto out_corruptbuf;
239		}
240
241		xfs_da3_node_hdr_from_disk(mp, &nodehdr, node);
242
243		/* Tree taller than we can handle; bail out! */
244		if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
245			goto out_corruptbuf;
246
247		/* Check the level from the root node. */
248		if (cursor->blkno == 0)
249			expected_level = nodehdr.level - 1;
250		else if (expected_level != nodehdr.level)
251			goto out_corruptbuf;
252		else
253			expected_level--;
254
255		btree = nodehdr.btree;
256		for (i = 0; i < nodehdr.count; btree++, i++) {
257			if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
258				cursor->blkno = be32_to_cpu(btree->before);
259				trace_xfs_attr_list_node_descend(context,
260						btree);
261				break;
262			}
263		}
264		xfs_trans_brelse(tp, bp);
265
266		if (i == nodehdr.count)
267			return 0;
268
269		/* We can't point back to the root. */
270		if (XFS_IS_CORRUPT(mp, cursor->blkno == 0)) {
271			xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
272			return -EFSCORRUPTED;
273		}
274	}
275
276	if (expected_level != 0)
277		goto out_corruptbuf;
278
279	*pbp = bp;
280	return 0;
281
282out_corruptbuf:
283	xfs_buf_mark_corrupt(bp);
284	xfs_trans_brelse(tp, bp);
285	xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
286	return -EFSCORRUPTED;
287}
288
289STATIC int
290xfs_attr_node_list(
291	struct xfs_attr_list_context	*context)
292{
293	struct xfs_attrlist_cursor_kern	*cursor = &context->cursor;
294	struct xfs_attr3_icleaf_hdr	leafhdr;
295	struct xfs_attr_leafblock	*leaf;
296	struct xfs_da_intnode		*node;
297	struct xfs_buf			*bp;
298	struct xfs_inode		*dp = context->dp;
299	struct xfs_mount		*mp = dp->i_mount;
300	int				error = 0;
301
302	trace_xfs_attr_node_list(context);
303
304	cursor->initted = 1;
305
306	/*
307	 * Do all sorts of validation on the passed-in cursor structure.
308	 * If anything is amiss, ignore the cursor and look up the hashval
309	 * starting from the btree root.
310	 */
311	bp = NULL;
312	if (cursor->blkno > 0) {
313		error = xfs_da3_node_read(context->tp, dp, cursor->blkno, &bp,
314				XFS_ATTR_FORK);
315		if (xfs_metadata_is_sick(error))
316			xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
317		if ((error != 0) && (error != -EFSCORRUPTED))
318			return error;
319		if (bp) {
320			struct xfs_attr_leaf_entry *entries;
321
322			node = bp->b_addr;
323			switch (be16_to_cpu(node->hdr.info.magic)) {
324			case XFS_DA_NODE_MAGIC:
325			case XFS_DA3_NODE_MAGIC:
326				trace_xfs_attr_list_wrong_blk(context);
327				xfs_trans_brelse(context->tp, bp);
328				bp = NULL;
329				break;
330			case XFS_ATTR_LEAF_MAGIC:
331			case XFS_ATTR3_LEAF_MAGIC:
332				leaf = bp->b_addr;
333				xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
334							     &leafhdr, leaf);
335				entries = xfs_attr3_leaf_entryp(leaf);
336				if (cursor->hashval > be32_to_cpu(
337						entries[leafhdr.count - 1].hashval)) {
338					trace_xfs_attr_list_wrong_blk(context);
339					xfs_trans_brelse(context->tp, bp);
340					bp = NULL;
341				} else if (cursor->hashval <= be32_to_cpu(
342						entries[0].hashval)) {
343					trace_xfs_attr_list_wrong_blk(context);
344					xfs_trans_brelse(context->tp, bp);
345					bp = NULL;
346				}
347				break;
348			default:
349				trace_xfs_attr_list_wrong_blk(context);
350				xfs_trans_brelse(context->tp, bp);
351				bp = NULL;
352			}
353		}
354	}
355
356	/*
357	 * We did not find what we expected given the cursor's contents,
358	 * so we start from the top and work down based on the hash value.
359	 * Note that start of node block is same as start of leaf block.
360	 */
361	if (bp == NULL) {
362		error = xfs_attr_node_list_lookup(context, cursor, &bp);
363		if (error || !bp)
364			return error;
365	}
366	ASSERT(bp != NULL);
367
368	/*
369	 * Roll upward through the blocks, processing each leaf block in
370	 * order.  As long as there is space in the result buffer, keep
371	 * adding the information.
372	 */
373	for (;;) {
374		leaf = bp->b_addr;
375		error = xfs_attr3_leaf_list_int(bp, context);
376		if (error)
377			break;
378		xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
379		if (context->seen_enough || leafhdr.forw == 0)
380			break;
381		cursor->blkno = leafhdr.forw;
382		xfs_trans_brelse(context->tp, bp);
383		error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno,
384					    &bp);
385		if (error)
386			return error;
387	}
388	xfs_trans_brelse(context->tp, bp);
389	return error;
390}
391
392/*
393 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
394 */
395int
396xfs_attr3_leaf_list_int(
397	struct xfs_buf			*bp,
398	struct xfs_attr_list_context	*context)
399{
400	struct xfs_attrlist_cursor_kern	*cursor = &context->cursor;
401	struct xfs_attr_leafblock	*leaf;
402	struct xfs_attr3_icleaf_hdr	ichdr;
403	struct xfs_attr_leaf_entry	*entries;
404	struct xfs_attr_leaf_entry	*entry;
405	int				i;
406	struct xfs_mount		*mp = context->dp->i_mount;
407
408	trace_xfs_attr_list_leaf(context);
409
410	leaf = bp->b_addr;
411	xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
412	entries = xfs_attr3_leaf_entryp(leaf);
413
414	cursor->initted = 1;
415
416	/*
417	 * Re-find our place in the leaf block if this is a new syscall.
418	 */
419	if (context->resynch) {
420		entry = &entries[0];
421		for (i = 0; i < ichdr.count; entry++, i++) {
422			if (be32_to_cpu(entry->hashval) == cursor->hashval) {
423				if (cursor->offset == context->dupcnt) {
424					context->dupcnt = 0;
425					break;
426				}
427				context->dupcnt++;
428			} else if (be32_to_cpu(entry->hashval) >
429					cursor->hashval) {
430				context->dupcnt = 0;
431				break;
432			}
433		}
434		if (i == ichdr.count) {
435			trace_xfs_attr_list_notfound(context);
436			return 0;
437		}
438	} else {
439		entry = &entries[0];
440		i = 0;
441	}
442	context->resynch = 0;
443
444	/*
445	 * We have found our place, start copying out the new attributes.
446	 */
447	for (; i < ichdr.count; entry++, i++) {
448		char *name;
449		int namelen, valuelen;
450
451		if (be32_to_cpu(entry->hashval) != cursor->hashval) {
452			cursor->hashval = be32_to_cpu(entry->hashval);
453			cursor->offset = 0;
454		}
455
456		if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
457		    !context->allow_incomplete)
458			continue;
459
460		if (entry->flags & XFS_ATTR_LOCAL) {
461			xfs_attr_leaf_name_local_t *name_loc;
462
463			name_loc = xfs_attr3_leaf_name_local(leaf, i);
464			name = name_loc->nameval;
465			namelen = name_loc->namelen;
466			valuelen = be16_to_cpu(name_loc->valuelen);
467		} else {
468			xfs_attr_leaf_name_remote_t *name_rmt;
469
470			name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
471			name = name_rmt->name;
472			namelen = name_rmt->namelen;
473			valuelen = be32_to_cpu(name_rmt->valuelen);
474		}
475
476		if (XFS_IS_CORRUPT(context->dp->i_mount,
477				   !xfs_attr_namecheck(name, namelen))) {
478			xfs_dirattr_mark_sick(context->dp, XFS_ATTR_FORK);
479			return -EFSCORRUPTED;
480		}
481		context->put_listent(context, entry->flags,
482					      name, namelen, valuelen);
483		if (context->seen_enough)
484			break;
485		cursor->offset++;
486	}
487	trace_xfs_attr_list_leaf_end(context);
488	return 0;
489}
490
491/*
492 * Copy out attribute entries for attr_list(), for leaf attribute lists.
493 */
494STATIC int
495xfs_attr_leaf_list(
496	struct xfs_attr_list_context	*context)
497{
498	struct xfs_buf			*bp;
499	int				error;
500
501	trace_xfs_attr_leaf_list(context);
502
503	context->cursor.blkno = 0;
504	error = xfs_attr3_leaf_read(context->tp, context->dp, 0, &bp);
505	if (error)
506		return error;
507
508	error = xfs_attr3_leaf_list_int(bp, context);
509	xfs_trans_brelse(context->tp, bp);
510	return error;
511}
512
513int
514xfs_attr_list_ilocked(
515	struct xfs_attr_list_context	*context)
516{
517	struct xfs_inode		*dp = context->dp;
518
519	xfs_assert_ilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL);
520
521	/*
522	 * Decide on what work routines to call based on the inode size.
523	 */
524	if (!xfs_inode_hasattr(dp))
525		return 0;
526	if (dp->i_af.if_format == XFS_DINODE_FMT_LOCAL)
527		return xfs_attr_shortform_list(context);
528	if (xfs_attr_is_leaf(dp))
529		return xfs_attr_leaf_list(context);
530	return xfs_attr_node_list(context);
531}
532
533int
534xfs_attr_list(
535	struct xfs_attr_list_context	*context)
536{
537	struct xfs_inode		*dp = context->dp;
538	uint				lock_mode;
539	int				error;
540
541	XFS_STATS_INC(dp->i_mount, xs_attr_list);
542
543	if (xfs_is_shutdown(dp->i_mount))
544		return -EIO;
545
546	lock_mode = xfs_ilock_attr_map_shared(dp);
547	error = xfs_attr_list_ilocked(context);
548	xfs_iunlock(dp, lock_mode);
549	return error;
550}
551