Loading...
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
42STATIC int
43xfs_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 */
68static int
69xfs_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
78 ASSERT(context != NULL);
79 dp = context->dp;
80 ASSERT(dp != NULL);
81 ASSERT(dp->i_afp != NULL);
82 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
83 ASSERT(sf != NULL);
84 if (!sf->hdr.count)
85 return 0;
86 cursor = context->cursor;
87 ASSERT(cursor != NULL);
88
89 trace_xfs_attr_list_sf(context);
90
91 /*
92 * If the buffer is large enough and the cursor is at the start,
93 * do not bother with sorting since we will return everything in
94 * one buffer and another call using the cursor won't need to be
95 * made.
96 * Note the generous fudge factor of 16 overhead bytes per entry.
97 * If bufsize is zero then put_listent must be a search function
98 * and can just scan through what we have.
99 */
100 if (context->bufsize == 0 ||
101 (XFS_ISRESET_CURSOR(cursor) &&
102 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
103 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
104 context->put_listent(context,
105 sfe->flags,
106 sfe->nameval,
107 (int)sfe->namelen,
108 (int)sfe->valuelen);
109 /*
110 * Either search callback finished early or
111 * didn't fit it all in the buffer after all.
112 */
113 if (context->seen_enough)
114 break;
115 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
116 }
117 trace_xfs_attr_list_sf_all(context);
118 return 0;
119 }
120
121 /* do no more for a search callback */
122 if (context->bufsize == 0)
123 return 0;
124
125 /*
126 * It didn't all fit, so we have to sort everything on hashval.
127 */
128 sbsize = sf->hdr.count * sizeof(*sbuf);
129 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
130
131 /*
132 * Scan the attribute list for the rest of the entries, storing
133 * the relevant info from only those that match into a buffer.
134 */
135 nsbuf = 0;
136 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
137 if (unlikely(
138 ((char *)sfe < (char *)sf) ||
139 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
140 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
141 XFS_ERRLEVEL_LOW,
142 context->dp->i_mount, sfe);
143 kmem_free(sbuf);
144 return -EFSCORRUPTED;
145 }
146
147 sbp->entno = i;
148 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
149 sbp->name = sfe->nameval;
150 sbp->namelen = sfe->namelen;
151 /* These are bytes, and both on-disk, don't endian-flip */
152 sbp->valuelen = sfe->valuelen;
153 sbp->flags = sfe->flags;
154 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
155 sbp++;
156 nsbuf++;
157 }
158
159 /*
160 * Sort the entries on hash then entno.
161 */
162 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
163
164 /*
165 * Re-find our place IN THE SORTED LIST.
166 */
167 count = 0;
168 cursor->initted = 1;
169 cursor->blkno = 0;
170 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
171 if (sbp->hash == cursor->hashval) {
172 if (cursor->offset == count) {
173 break;
174 }
175 count++;
176 } else if (sbp->hash > cursor->hashval) {
177 break;
178 }
179 }
180 if (i == nsbuf) {
181 kmem_free(sbuf);
182 return 0;
183 }
184
185 /*
186 * Loop putting entries into the user buffer.
187 */
188 for ( ; i < nsbuf; i++, sbp++) {
189 if (cursor->hashval != sbp->hash) {
190 cursor->hashval = sbp->hash;
191 cursor->offset = 0;
192 }
193 context->put_listent(context,
194 sbp->flags,
195 sbp->name,
196 sbp->namelen,
197 sbp->valuelen);
198 if (context->seen_enough)
199 break;
200 cursor->offset++;
201 }
202
203 kmem_free(sbuf);
204 return 0;
205}
206
207/*
208 * We didn't find the block & hash mentioned in the cursor state, so
209 * walk down the attr btree looking for the hash.
210 */
211STATIC int
212xfs_attr_node_list_lookup(
213 struct xfs_attr_list_context *context,
214 struct attrlist_cursor_kern *cursor,
215 struct xfs_buf **pbp)
216{
217 struct xfs_da3_icnode_hdr nodehdr;
218 struct xfs_da_intnode *node;
219 struct xfs_da_node_entry *btree;
220 struct xfs_inode *dp = context->dp;
221 struct xfs_mount *mp = dp->i_mount;
222 struct xfs_trans *tp = context->tp;
223 struct xfs_buf *bp;
224 int i;
225 int error = 0;
226 unsigned int expected_level = 0;
227 uint16_t magic;
228
229 ASSERT(*pbp == NULL);
230 cursor->blkno = 0;
231 for (;;) {
232 error = xfs_da3_node_read(tp, dp, cursor->blkno, -1, &bp,
233 XFS_ATTR_FORK);
234 if (error)
235 return error;
236 node = bp->b_addr;
237 magic = be16_to_cpu(node->hdr.info.magic);
238 if (magic == XFS_ATTR_LEAF_MAGIC ||
239 magic == XFS_ATTR3_LEAF_MAGIC)
240 break;
241 if (magic != XFS_DA_NODE_MAGIC &&
242 magic != XFS_DA3_NODE_MAGIC) {
243 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
244 node);
245 goto out_corruptbuf;
246 }
247
248 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
249
250 /* Tree taller than we can handle; bail out! */
251 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
252 goto out_corruptbuf;
253
254 /* Check the level from the root node. */
255 if (cursor->blkno == 0)
256 expected_level = nodehdr.level - 1;
257 else if (expected_level != nodehdr.level)
258 goto out_corruptbuf;
259 else
260 expected_level--;
261
262 btree = dp->d_ops->node_tree_p(node);
263 for (i = 0; i < nodehdr.count; btree++, i++) {
264 if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
265 cursor->blkno = be32_to_cpu(btree->before);
266 trace_xfs_attr_list_node_descend(context,
267 btree);
268 break;
269 }
270 }
271 xfs_trans_brelse(tp, bp);
272
273 if (i == nodehdr.count)
274 return 0;
275
276 /* We can't point back to the root. */
277 if (cursor->blkno == 0)
278 return -EFSCORRUPTED;
279 }
280
281 if (expected_level != 0)
282 goto out_corruptbuf;
283
284 *pbp = bp;
285 return 0;
286
287out_corruptbuf:
288 xfs_trans_brelse(tp, bp);
289 return -EFSCORRUPTED;
290}
291
292STATIC int
293xfs_attr_node_list(
294 struct xfs_attr_list_context *context)
295{
296 struct xfs_attr3_icleaf_hdr leafhdr;
297 struct attrlist_cursor_kern *cursor;
298 struct xfs_attr_leafblock *leaf;
299 struct xfs_da_intnode *node;
300 struct xfs_buf *bp;
301 struct xfs_inode *dp = context->dp;
302 struct xfs_mount *mp = dp->i_mount;
303 int error;
304
305 trace_xfs_attr_node_list(context);
306
307 cursor = context->cursor;
308 cursor->initted = 1;
309
310 /*
311 * Do all sorts of validation on the passed-in cursor structure.
312 * If anything is amiss, ignore the cursor and look up the hashval
313 * starting from the btree root.
314 */
315 bp = NULL;
316 if (cursor->blkno > 0) {
317 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
318 &bp, XFS_ATTR_FORK);
319 if ((error != 0) && (error != -EFSCORRUPTED))
320 return error;
321 if (bp) {
322 struct xfs_attr_leaf_entry *entries;
323
324 node = bp->b_addr;
325 switch (be16_to_cpu(node->hdr.info.magic)) {
326 case XFS_DA_NODE_MAGIC:
327 case XFS_DA3_NODE_MAGIC:
328 trace_xfs_attr_list_wrong_blk(context);
329 xfs_trans_brelse(context->tp, bp);
330 bp = NULL;
331 break;
332 case XFS_ATTR_LEAF_MAGIC:
333 case XFS_ATTR3_LEAF_MAGIC:
334 leaf = bp->b_addr;
335 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
336 &leafhdr, leaf);
337 entries = xfs_attr3_leaf_entryp(leaf);
338 if (cursor->hashval > be32_to_cpu(
339 entries[leafhdr.count - 1].hashval)) {
340 trace_xfs_attr_list_wrong_blk(context);
341 xfs_trans_brelse(context->tp, bp);
342 bp = NULL;
343 } else if (cursor->hashval <= be32_to_cpu(
344 entries[0].hashval)) {
345 trace_xfs_attr_list_wrong_blk(context);
346 xfs_trans_brelse(context->tp, bp);
347 bp = NULL;
348 }
349 break;
350 default:
351 trace_xfs_attr_list_wrong_blk(context);
352 xfs_trans_brelse(context->tp, bp);
353 bp = NULL;
354 }
355 }
356 }
357
358 /*
359 * We did not find what we expected given the cursor's contents,
360 * so we start from the top and work down based on the hash value.
361 * Note that start of node block is same as start of leaf block.
362 */
363 if (bp == NULL) {
364 error = xfs_attr_node_list_lookup(context, cursor, &bp);
365 if (error || !bp)
366 return error;
367 }
368 ASSERT(bp != NULL);
369
370 /*
371 * Roll upward through the blocks, processing each leaf block in
372 * order. As long as there is space in the result buffer, keep
373 * adding the information.
374 */
375 for (;;) {
376 leaf = bp->b_addr;
377 xfs_attr3_leaf_list_int(bp, context);
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, -1, &bp);
384 if (error)
385 return error;
386 }
387 xfs_trans_brelse(context->tp, bp);
388 return 0;
389}
390
391/*
392 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
393 */
394void
395xfs_attr3_leaf_list_int(
396 struct xfs_buf *bp,
397 struct xfs_attr_list_context *context)
398{
399 struct attrlist_cursor_kern *cursor;
400 struct xfs_attr_leafblock *leaf;
401 struct xfs_attr3_icleaf_hdr ichdr;
402 struct xfs_attr_leaf_entry *entries;
403 struct xfs_attr_leaf_entry *entry;
404 int i;
405 struct xfs_mount *mp = context->dp->i_mount;
406
407 trace_xfs_attr_list_leaf(context);
408
409 leaf = bp->b_addr;
410 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
411 entries = xfs_attr3_leaf_entryp(leaf);
412
413 cursor = context->cursor;
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;
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->flags & ATTR_INCOMPLETE))
458 continue; /* skip incomplete entries */
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 context->put_listent(context, entry->flags,
477 name, namelen, valuelen);
478 if (context->seen_enough)
479 break;
480 cursor->offset++;
481 }
482 trace_xfs_attr_list_leaf_end(context);
483 return;
484}
485
486/*
487 * Copy out attribute entries for attr_list(), for leaf attribute lists.
488 */
489STATIC int
490xfs_attr_leaf_list(xfs_attr_list_context_t *context)
491{
492 int error;
493 struct xfs_buf *bp;
494
495 trace_xfs_attr_leaf_list(context);
496
497 context->cursor->blkno = 0;
498 error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
499 if (error)
500 return error;
501
502 xfs_attr3_leaf_list_int(bp, context);
503 xfs_trans_brelse(context->tp, bp);
504 return 0;
505}
506
507int
508xfs_attr_list_int_ilocked(
509 struct xfs_attr_list_context *context)
510{
511 struct xfs_inode *dp = context->dp;
512
513 ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
514
515 /*
516 * Decide on what work routines to call based on the inode size.
517 */
518 if (!xfs_inode_hasattr(dp))
519 return 0;
520 else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
521 return xfs_attr_shortform_list(context);
522 else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
523 return xfs_attr_leaf_list(context);
524 return xfs_attr_node_list(context);
525}
526
527int
528xfs_attr_list_int(
529 xfs_attr_list_context_t *context)
530{
531 int error;
532 xfs_inode_t *dp = context->dp;
533 uint lock_mode;
534
535 XFS_STATS_INC(dp->i_mount, xs_attr_list);
536
537 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
538 return -EIO;
539
540 lock_mode = xfs_ilock_attr_map_shared(dp);
541 error = xfs_attr_list_int_ilocked(context);
542 xfs_iunlock(dp, lock_mode);
543 return error;
544}
545
546#define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
547 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
548#define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
549 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(uint32_t)-1) \
550 & ~(sizeof(uint32_t)-1))
551
552/*
553 * Format an attribute and copy it out to the user's buffer.
554 * Take care to check values and protect against them changing later,
555 * we may be reading them directly out of a user buffer.
556 */
557STATIC void
558xfs_attr_put_listent(
559 xfs_attr_list_context_t *context,
560 int flags,
561 unsigned char *name,
562 int namelen,
563 int valuelen)
564{
565 struct attrlist *alist = (struct attrlist *)context->alist;
566 attrlist_ent_t *aep;
567 int arraytop;
568
569 ASSERT(!(context->flags & ATTR_KERNOVAL));
570 ASSERT(context->count >= 0);
571 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
572 ASSERT(context->firstu >= sizeof(*alist));
573 ASSERT(context->firstu <= context->bufsize);
574
575 /*
576 * Only list entries in the right namespace.
577 */
578 if (((context->flags & ATTR_SECURE) == 0) !=
579 ((flags & XFS_ATTR_SECURE) == 0))
580 return;
581 if (((context->flags & ATTR_ROOT) == 0) !=
582 ((flags & XFS_ATTR_ROOT) == 0))
583 return;
584
585 arraytop = sizeof(*alist) +
586 context->count * sizeof(alist->al_offset[0]);
587 context->firstu -= ATTR_ENTSIZE(namelen);
588 if (context->firstu < arraytop) {
589 trace_xfs_attr_list_full(context);
590 alist->al_more = 1;
591 context->seen_enough = 1;
592 return;
593 }
594
595 aep = (attrlist_ent_t *)&context->alist[context->firstu];
596 aep->a_valuelen = valuelen;
597 memcpy(aep->a_name, name, namelen);
598 aep->a_name[namelen] = 0;
599 alist->al_offset[context->count++] = context->firstu;
600 alist->al_count = context->count;
601 trace_xfs_attr_list_add(context);
602 return;
603}
604
605/*
606 * Generate a list of extended attribute names and optionally
607 * also value lengths. Positive return value follows the XFS
608 * convention of being an error, zero or negative return code
609 * is the length of the buffer returned (negated), indicating
610 * success.
611 */
612int
613xfs_attr_list(
614 xfs_inode_t *dp,
615 char *buffer,
616 int bufsize,
617 int flags,
618 attrlist_cursor_kern_t *cursor)
619{
620 xfs_attr_list_context_t context;
621 struct attrlist *alist;
622 int error;
623
624 /*
625 * Validate the cursor.
626 */
627 if (cursor->pad1 || cursor->pad2)
628 return -EINVAL;
629 if ((cursor->initted == 0) &&
630 (cursor->hashval || cursor->blkno || cursor->offset))
631 return -EINVAL;
632
633 /* Only internal consumers can retrieve incomplete attrs. */
634 if (flags & ATTR_INCOMPLETE)
635 return -EINVAL;
636
637 /*
638 * Check for a properly aligned buffer.
639 */
640 if (((long)buffer) & (sizeof(int)-1))
641 return -EFAULT;
642 if (flags & ATTR_KERNOVAL)
643 bufsize = 0;
644
645 /*
646 * Initialize the output buffer.
647 */
648 memset(&context, 0, sizeof(context));
649 context.dp = dp;
650 context.cursor = cursor;
651 context.resynch = 1;
652 context.flags = flags;
653 context.alist = buffer;
654 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
655 context.firstu = context.bufsize;
656 context.put_listent = xfs_attr_put_listent;
657
658 alist = (struct attrlist *)context.alist;
659 alist->al_count = 0;
660 alist->al_more = 0;
661 alist->al_offset[0] = context.bufsize;
662
663 error = xfs_attr_list_int(&context);
664 ASSERT(error <= 0);
665 return error;
666}
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->flags,
86 sfe->nameval,
87 sfe->namelen))) {
88 xfs_dirattr_mark_sick(context->dp, XFS_ATTR_FORK);
89 return -EFSCORRUPTED;
90 }
91 context->put_listent(context,
92 sfe->flags,
93 sfe->nameval,
94 (int)sfe->namelen,
95 &sfe->nameval[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->count * sizeof(*sbuf);
117 sbp = sbuf = kmalloc(sbsize,
118 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
119
120 /*
121 * Scan the attribute list for the rest of the entries, storing
122 * the relevant info from only those that match into a buffer.
123 */
124 nsbuf = 0;
125 for (i = 0, sfe = xfs_attr_sf_firstentry(sf); i < sf->count; i++) {
126 if (unlikely(
127 ((char *)sfe < (char *)sf) ||
128 ((char *)sfe >= ((char *)sf + dp->i_af.if_bytes)) ||
129 !xfs_attr_check_namespace(sfe->flags))) {
130 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
131 XFS_ERRLEVEL_LOW,
132 context->dp->i_mount, sfe,
133 sizeof(*sfe));
134 kfree(sbuf);
135 xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
136 return -EFSCORRUPTED;
137 }
138
139 sbp->entno = i;
140 sbp->name = sfe->nameval;
141 sbp->namelen = sfe->namelen;
142 /* These are bytes, and both on-disk, don't endian-flip */
143 sbp->value = &sfe->nameval[sfe->namelen];
144 sbp->valuelen = sfe->valuelen;
145 sbp->flags = sfe->flags;
146 sbp->hash = xfs_attr_hashval(dp->i_mount, sfe->flags,
147 sfe->nameval, sfe->namelen,
148 sfe->nameval + sfe->namelen,
149 sfe->valuelen);
150 sfe = xfs_attr_sf_nextentry(sfe);
151 sbp++;
152 nsbuf++;
153 }
154
155 /*
156 * Sort the entries on hash then entno.
157 */
158 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
159
160 /*
161 * Re-find our place IN THE SORTED LIST.
162 */
163 count = 0;
164 cursor->initted = 1;
165 cursor->blkno = 0;
166 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
167 if (sbp->hash == cursor->hashval) {
168 if (cursor->offset == count) {
169 break;
170 }
171 count++;
172 } else if (sbp->hash > cursor->hashval) {
173 break;
174 }
175 }
176 if (i == nsbuf)
177 goto out;
178
179 /*
180 * Loop putting entries into the user buffer.
181 */
182 for ( ; i < nsbuf; i++, sbp++) {
183 if (cursor->hashval != sbp->hash) {
184 cursor->hashval = sbp->hash;
185 cursor->offset = 0;
186 }
187 if (XFS_IS_CORRUPT(context->dp->i_mount,
188 !xfs_attr_namecheck(sbp->flags, sbp->name,
189 sbp->namelen))) {
190 xfs_dirattr_mark_sick(context->dp, XFS_ATTR_FORK);
191 error = -EFSCORRUPTED;
192 goto out;
193 }
194 context->put_listent(context,
195 sbp->flags,
196 sbp->name,
197 sbp->namelen,
198 sbp->value,
199 sbp->valuelen);
200 if (context->seen_enough)
201 break;
202 cursor->offset++;
203 }
204out:
205 kfree(sbuf);
206 return error;
207}
208
209/*
210 * We didn't find the block & hash mentioned in the cursor state, so
211 * walk down the attr btree looking for the hash.
212 */
213STATIC int
214xfs_attr_node_list_lookup(
215 struct xfs_attr_list_context *context,
216 struct xfs_attrlist_cursor_kern *cursor,
217 struct xfs_buf **pbp)
218{
219 struct xfs_da3_icnode_hdr nodehdr;
220 struct xfs_da_intnode *node;
221 struct xfs_da_node_entry *btree;
222 struct xfs_inode *dp = context->dp;
223 struct xfs_mount *mp = dp->i_mount;
224 struct xfs_trans *tp = context->tp;
225 struct xfs_buf *bp;
226 xfs_failaddr_t fa;
227 int i;
228 int error = 0;
229 unsigned int expected_level = 0;
230 uint16_t magic;
231
232 ASSERT(*pbp == NULL);
233 cursor->blkno = 0;
234 for (;;) {
235 error = xfs_da3_node_read(tp, dp, cursor->blkno, &bp,
236 XFS_ATTR_FORK);
237 if (error)
238 return error;
239 node = bp->b_addr;
240 magic = be16_to_cpu(node->hdr.info.magic);
241 if (magic == XFS_ATTR_LEAF_MAGIC ||
242 magic == XFS_ATTR3_LEAF_MAGIC)
243 break;
244 if (magic != XFS_DA_NODE_MAGIC &&
245 magic != XFS_DA3_NODE_MAGIC) {
246 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
247 node, sizeof(*node));
248 goto out_corruptbuf;
249 }
250
251 fa = xfs_da3_node_header_check(bp, dp->i_ino);
252 if (fa)
253 goto out_corruptbuf;
254
255 xfs_da3_node_hdr_from_disk(mp, &nodehdr, node);
256
257 /* Tree taller than we can handle; bail out! */
258 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
259 goto out_corruptbuf;
260
261 /* Check the level from the root node. */
262 if (cursor->blkno == 0)
263 expected_level = nodehdr.level - 1;
264 else if (expected_level != nodehdr.level)
265 goto out_corruptbuf;
266 else
267 expected_level--;
268
269 btree = nodehdr.btree;
270 for (i = 0; i < nodehdr.count; btree++, i++) {
271 if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
272 cursor->blkno = be32_to_cpu(btree->before);
273 trace_xfs_attr_list_node_descend(context,
274 btree);
275 break;
276 }
277 }
278 xfs_trans_brelse(tp, bp);
279
280 if (i == nodehdr.count)
281 return 0;
282
283 /* We can't point back to the root. */
284 if (XFS_IS_CORRUPT(mp, cursor->blkno == 0)) {
285 xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
286 return -EFSCORRUPTED;
287 }
288 }
289
290 fa = xfs_attr3_leaf_header_check(bp, dp->i_ino);
291 if (fa) {
292 __xfs_buf_mark_corrupt(bp, fa);
293 goto out_releasebuf;
294 }
295
296 if (expected_level != 0)
297 goto out_corruptbuf;
298
299 *pbp = bp;
300 return 0;
301
302out_corruptbuf:
303 xfs_buf_mark_corrupt(bp);
304out_releasebuf:
305 xfs_trans_brelse(tp, bp);
306 xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
307 return -EFSCORRUPTED;
308}
309
310STATIC int
311xfs_attr_node_list(
312 struct xfs_attr_list_context *context)
313{
314 struct xfs_attrlist_cursor_kern *cursor = &context->cursor;
315 struct xfs_attr3_icleaf_hdr leafhdr;
316 struct xfs_attr_leafblock *leaf;
317 struct xfs_da_intnode *node;
318 struct xfs_buf *bp;
319 struct xfs_inode *dp = context->dp;
320 struct xfs_mount *mp = dp->i_mount;
321 xfs_failaddr_t fa;
322 int error = 0;
323
324 trace_xfs_attr_node_list(context);
325
326 cursor->initted = 1;
327
328 /*
329 * Do all sorts of validation on the passed-in cursor structure.
330 * If anything is amiss, ignore the cursor and look up the hashval
331 * starting from the btree root.
332 */
333 bp = NULL;
334 if (cursor->blkno > 0) {
335 struct xfs_attr_leaf_entry *entries;
336
337 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, &bp,
338 XFS_ATTR_FORK);
339 if (xfs_metadata_is_sick(error))
340 xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
341 if (error != 0 && error != -EFSCORRUPTED)
342 return error;
343 if (!bp)
344 goto need_lookup;
345
346 node = bp->b_addr;
347 switch (be16_to_cpu(node->hdr.info.magic)) {
348 case XFS_DA_NODE_MAGIC:
349 case XFS_DA3_NODE_MAGIC:
350 trace_xfs_attr_list_wrong_blk(context);
351 fa = xfs_da3_node_header_check(bp, dp->i_ino);
352 if (fa) {
353 __xfs_buf_mark_corrupt(bp, fa);
354 xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
355 }
356 xfs_trans_brelse(context->tp, bp);
357 bp = NULL;
358 break;
359 case XFS_ATTR_LEAF_MAGIC:
360 case XFS_ATTR3_LEAF_MAGIC:
361 leaf = bp->b_addr;
362 fa = xfs_attr3_leaf_header_check(bp, dp->i_ino);
363 if (fa) {
364 __xfs_buf_mark_corrupt(bp, fa);
365 xfs_trans_brelse(context->tp, bp);
366 xfs_dirattr_mark_sick(dp, XFS_ATTR_FORK);
367 bp = NULL;
368 break;
369 }
370 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
371 &leafhdr, leaf);
372 entries = xfs_attr3_leaf_entryp(leaf);
373 if (cursor->hashval > be32_to_cpu(
374 entries[leafhdr.count - 1].hashval)) {
375 trace_xfs_attr_list_wrong_blk(context);
376 xfs_trans_brelse(context->tp, bp);
377 bp = NULL;
378 } else if (cursor->hashval <= be32_to_cpu(
379 entries[0].hashval)) {
380 trace_xfs_attr_list_wrong_blk(context);
381 xfs_trans_brelse(context->tp, bp);
382 bp = NULL;
383 }
384 break;
385 default:
386 trace_xfs_attr_list_wrong_blk(context);
387 xfs_trans_brelse(context->tp, bp);
388 bp = NULL;
389 }
390 }
391
392 /*
393 * We did not find what we expected given the cursor's contents,
394 * so we start from the top and work down based on the hash value.
395 * Note that start of node block is same as start of leaf block.
396 */
397 if (bp == NULL) {
398need_lookup:
399 error = xfs_attr_node_list_lookup(context, cursor, &bp);
400 if (error || !bp)
401 return error;
402 }
403 ASSERT(bp != NULL);
404
405 /*
406 * Roll upward through the blocks, processing each leaf block in
407 * order. As long as there is space in the result buffer, keep
408 * adding the information.
409 */
410 for (;;) {
411 leaf = bp->b_addr;
412 error = xfs_attr3_leaf_list_int(bp, context);
413 if (error)
414 break;
415 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
416 if (context->seen_enough || leafhdr.forw == 0)
417 break;
418 cursor->blkno = leafhdr.forw;
419 xfs_trans_brelse(context->tp, bp);
420 error = xfs_attr3_leaf_read(context->tp, dp, dp->i_ino,
421 cursor->blkno, &bp);
422 if (error)
423 return error;
424 }
425 xfs_trans_brelse(context->tp, bp);
426 return error;
427}
428
429/*
430 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
431 */
432int
433xfs_attr3_leaf_list_int(
434 struct xfs_buf *bp,
435 struct xfs_attr_list_context *context)
436{
437 struct xfs_attrlist_cursor_kern *cursor = &context->cursor;
438 struct xfs_attr_leafblock *leaf;
439 struct xfs_attr3_icleaf_hdr ichdr;
440 struct xfs_attr_leaf_entry *entries;
441 struct xfs_attr_leaf_entry *entry;
442 int i;
443 struct xfs_mount *mp = context->dp->i_mount;
444
445 trace_xfs_attr_list_leaf(context);
446
447 leaf = bp->b_addr;
448 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
449 entries = xfs_attr3_leaf_entryp(leaf);
450
451 cursor->initted = 1;
452
453 /*
454 * Re-find our place in the leaf block if this is a new syscall.
455 */
456 if (context->resynch) {
457 entry = &entries[0];
458 for (i = 0; i < ichdr.count; entry++, i++) {
459 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
460 if (cursor->offset == context->dupcnt) {
461 context->dupcnt = 0;
462 break;
463 }
464 context->dupcnt++;
465 } else if (be32_to_cpu(entry->hashval) >
466 cursor->hashval) {
467 context->dupcnt = 0;
468 break;
469 }
470 }
471 if (i == ichdr.count) {
472 trace_xfs_attr_list_notfound(context);
473 return 0;
474 }
475 } else {
476 entry = &entries[0];
477 i = 0;
478 }
479 context->resynch = 0;
480
481 /*
482 * We have found our place, start copying out the new attributes.
483 */
484 for (; i < ichdr.count; entry++, i++) {
485 char *name;
486 void *value;
487 int namelen, valuelen;
488
489 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
490 cursor->hashval = be32_to_cpu(entry->hashval);
491 cursor->offset = 0;
492 }
493
494 if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
495 !context->allow_incomplete)
496 continue;
497
498 if (entry->flags & XFS_ATTR_LOCAL) {
499 xfs_attr_leaf_name_local_t *name_loc;
500
501 name_loc = xfs_attr3_leaf_name_local(leaf, i);
502 name = name_loc->nameval;
503 namelen = name_loc->namelen;
504 value = &name_loc->nameval[name_loc->namelen];
505 valuelen = be16_to_cpu(name_loc->valuelen);
506 } else {
507 xfs_attr_leaf_name_remote_t *name_rmt;
508
509 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
510 name = name_rmt->name;
511 namelen = name_rmt->namelen;
512 value = NULL;
513 valuelen = be32_to_cpu(name_rmt->valuelen);
514 }
515
516 if (XFS_IS_CORRUPT(context->dp->i_mount,
517 !xfs_attr_namecheck(entry->flags, name,
518 namelen))) {
519 xfs_dirattr_mark_sick(context->dp, XFS_ATTR_FORK);
520 return -EFSCORRUPTED;
521 }
522 context->put_listent(context, entry->flags,
523 name, namelen, value, valuelen);
524 if (context->seen_enough)
525 break;
526 cursor->offset++;
527 }
528 trace_xfs_attr_list_leaf_end(context);
529 return 0;
530}
531
532/*
533 * Copy out attribute entries for attr_list(), for leaf attribute lists.
534 */
535STATIC int
536xfs_attr_leaf_list(
537 struct xfs_attr_list_context *context)
538{
539 struct xfs_buf *bp;
540 int error;
541
542 trace_xfs_attr_leaf_list(context);
543
544 context->cursor.blkno = 0;
545 error = xfs_attr3_leaf_read(context->tp, context->dp,
546 context->dp->i_ino, 0, &bp);
547 if (error)
548 return error;
549
550 error = xfs_attr3_leaf_list_int(bp, context);
551 xfs_trans_brelse(context->tp, bp);
552 return error;
553}
554
555int
556xfs_attr_list_ilocked(
557 struct xfs_attr_list_context *context)
558{
559 struct xfs_inode *dp = context->dp;
560 int error;
561
562 xfs_assert_ilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL);
563
564 /*
565 * Decide on what work routines to call based on the inode size.
566 */
567 if (!xfs_inode_hasattr(dp))
568 return 0;
569 if (dp->i_af.if_format == XFS_DINODE_FMT_LOCAL)
570 return xfs_attr_shortform_list(context);
571
572 /* Prerequisite for xfs_attr_is_leaf */
573 error = xfs_iread_extents(NULL, dp, XFS_ATTR_FORK);
574 if (error)
575 return error;
576
577 if (xfs_attr_is_leaf(dp))
578 return xfs_attr_leaf_list(context);
579 return xfs_attr_node_list(context);
580}
581
582int
583xfs_attr_list(
584 struct xfs_attr_list_context *context)
585{
586 struct xfs_inode *dp = context->dp;
587 uint lock_mode;
588 int error;
589
590 XFS_STATS_INC(dp->i_mount, xs_attr_list);
591
592 if (xfs_is_shutdown(dp->i_mount))
593 return -EIO;
594
595 lock_mode = xfs_ilock_attr_map_shared(dp);
596 error = xfs_attr_list_ilocked(context);
597 xfs_iunlock(dp, lock_mode);
598 return error;
599}