Loading...
1/*
2 * linux/fs/ext3/xattr.c
3 *
4 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
5 *
6 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7 * Ext3 code with a lot of help from Eric Jarman <ejarman@acm.org>.
8 * Extended attributes for symlinks and special files added per
9 * suggestion of Luka Renko <luka.renko@hermes.si>.
10 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
11 * Red Hat Inc.
12 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
13 * and Andreas Gruenbacher <agruen@suse.de>.
14 */
15
16/*
17 * Extended attributes are stored directly in inodes (on file systems with
18 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
19 * field contains the block number if an inode uses an additional block. All
20 * attributes must fit in the inode and one additional block. Blocks that
21 * contain the identical set of attributes may be shared among several inodes.
22 * Identical blocks are detected by keeping a cache of blocks that have
23 * recently been accessed.
24 *
25 * The attributes in inodes and on blocks have a different header; the entries
26 * are stored in the same format:
27 *
28 * +------------------+
29 * | header |
30 * | entry 1 | |
31 * | entry 2 | | growing downwards
32 * | entry 3 | v
33 * | four null bytes |
34 * | . . . |
35 * | value 1 | ^
36 * | value 3 | | growing upwards
37 * | value 2 | |
38 * +------------------+
39 *
40 * The header is followed by multiple entry descriptors. In disk blocks, the
41 * entry descriptors are kept sorted. In inodes, they are unsorted. The
42 * attribute values are aligned to the end of the block in no specific order.
43 *
44 * Locking strategy
45 * ----------------
46 * EXT3_I(inode)->i_file_acl is protected by EXT3_I(inode)->xattr_sem.
47 * EA blocks are only changed if they are exclusive to an inode, so
48 * holding xattr_sem also means that nothing but the EA block's reference
49 * count can change. Multiple writers to the same block are synchronized
50 * by the buffer lock.
51 */
52
53#include "ext3.h"
54#include <linux/mbcache.h>
55#include <linux/quotaops.h>
56#include "xattr.h"
57#include "acl.h"
58
59#define BHDR(bh) ((struct ext3_xattr_header *)((bh)->b_data))
60#define ENTRY(ptr) ((struct ext3_xattr_entry *)(ptr))
61#define BFIRST(bh) ENTRY(BHDR(bh)+1)
62#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
63
64#define IHDR(inode, raw_inode) \
65 ((struct ext3_xattr_ibody_header *) \
66 ((void *)raw_inode + \
67 EXT3_GOOD_OLD_INODE_SIZE + \
68 EXT3_I(inode)->i_extra_isize))
69#define IFIRST(hdr) ((struct ext3_xattr_entry *)((hdr)+1))
70
71#ifdef EXT3_XATTR_DEBUG
72# define ea_idebug(inode, f...) do { \
73 printk(KERN_DEBUG "inode %s:%lu: ", \
74 inode->i_sb->s_id, inode->i_ino); \
75 printk(f); \
76 printk("\n"); \
77 } while (0)
78# define ea_bdebug(bh, f...) do { \
79 char b[BDEVNAME_SIZE]; \
80 printk(KERN_DEBUG "block %s:%lu: ", \
81 bdevname(bh->b_bdev, b), \
82 (unsigned long) bh->b_blocknr); \
83 printk(f); \
84 printk("\n"); \
85 } while (0)
86#else
87# define ea_idebug(f...)
88# define ea_bdebug(f...)
89#endif
90
91static void ext3_xattr_cache_insert(struct buffer_head *);
92static struct buffer_head *ext3_xattr_cache_find(struct inode *,
93 struct ext3_xattr_header *,
94 struct mb_cache_entry **);
95static void ext3_xattr_rehash(struct ext3_xattr_header *,
96 struct ext3_xattr_entry *);
97static int ext3_xattr_list(struct dentry *dentry, char *buffer,
98 size_t buffer_size);
99
100static struct mb_cache *ext3_xattr_cache;
101
102static const struct xattr_handler *ext3_xattr_handler_map[] = {
103 [EXT3_XATTR_INDEX_USER] = &ext3_xattr_user_handler,
104#ifdef CONFIG_EXT3_FS_POSIX_ACL
105 [EXT3_XATTR_INDEX_POSIX_ACL_ACCESS] = &ext3_xattr_acl_access_handler,
106 [EXT3_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext3_xattr_acl_default_handler,
107#endif
108 [EXT3_XATTR_INDEX_TRUSTED] = &ext3_xattr_trusted_handler,
109#ifdef CONFIG_EXT3_FS_SECURITY
110 [EXT3_XATTR_INDEX_SECURITY] = &ext3_xattr_security_handler,
111#endif
112};
113
114const struct xattr_handler *ext3_xattr_handlers[] = {
115 &ext3_xattr_user_handler,
116 &ext3_xattr_trusted_handler,
117#ifdef CONFIG_EXT3_FS_POSIX_ACL
118 &ext3_xattr_acl_access_handler,
119 &ext3_xattr_acl_default_handler,
120#endif
121#ifdef CONFIG_EXT3_FS_SECURITY
122 &ext3_xattr_security_handler,
123#endif
124 NULL
125};
126
127static inline const struct xattr_handler *
128ext3_xattr_handler(int name_index)
129{
130 const struct xattr_handler *handler = NULL;
131
132 if (name_index > 0 && name_index < ARRAY_SIZE(ext3_xattr_handler_map))
133 handler = ext3_xattr_handler_map[name_index];
134 return handler;
135}
136
137/*
138 * Inode operation listxattr()
139 *
140 * dentry->d_inode->i_mutex: don't care
141 */
142ssize_t
143ext3_listxattr(struct dentry *dentry, char *buffer, size_t size)
144{
145 return ext3_xattr_list(dentry, buffer, size);
146}
147
148static int
149ext3_xattr_check_names(struct ext3_xattr_entry *entry, void *end)
150{
151 while (!IS_LAST_ENTRY(entry)) {
152 struct ext3_xattr_entry *next = EXT3_XATTR_NEXT(entry);
153 if ((void *)next >= end)
154 return -EIO;
155 entry = next;
156 }
157 return 0;
158}
159
160static inline int
161ext3_xattr_check_block(struct buffer_head *bh)
162{
163 int error;
164
165 if (BHDR(bh)->h_magic != cpu_to_le32(EXT3_XATTR_MAGIC) ||
166 BHDR(bh)->h_blocks != cpu_to_le32(1))
167 return -EIO;
168 error = ext3_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size);
169 return error;
170}
171
172static inline int
173ext3_xattr_check_entry(struct ext3_xattr_entry *entry, size_t size)
174{
175 size_t value_size = le32_to_cpu(entry->e_value_size);
176
177 if (entry->e_value_block != 0 || value_size > size ||
178 le16_to_cpu(entry->e_value_offs) + value_size > size)
179 return -EIO;
180 return 0;
181}
182
183static int
184ext3_xattr_find_entry(struct ext3_xattr_entry **pentry, int name_index,
185 const char *name, size_t size, int sorted)
186{
187 struct ext3_xattr_entry *entry;
188 size_t name_len;
189 int cmp = 1;
190
191 if (name == NULL)
192 return -EINVAL;
193 name_len = strlen(name);
194 entry = *pentry;
195 for (; !IS_LAST_ENTRY(entry); entry = EXT3_XATTR_NEXT(entry)) {
196 cmp = name_index - entry->e_name_index;
197 if (!cmp)
198 cmp = name_len - entry->e_name_len;
199 if (!cmp)
200 cmp = memcmp(name, entry->e_name, name_len);
201 if (cmp <= 0 && (sorted || cmp == 0))
202 break;
203 }
204 *pentry = entry;
205 if (!cmp && ext3_xattr_check_entry(entry, size))
206 return -EIO;
207 return cmp ? -ENODATA : 0;
208}
209
210static int
211ext3_xattr_block_get(struct inode *inode, int name_index, const char *name,
212 void *buffer, size_t buffer_size)
213{
214 struct buffer_head *bh = NULL;
215 struct ext3_xattr_entry *entry;
216 size_t size;
217 int error;
218
219 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
220 name_index, name, buffer, (long)buffer_size);
221
222 error = -ENODATA;
223 if (!EXT3_I(inode)->i_file_acl)
224 goto cleanup;
225 ea_idebug(inode, "reading block %u", EXT3_I(inode)->i_file_acl);
226 bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
227 if (!bh)
228 goto cleanup;
229 ea_bdebug(bh, "b_count=%d, refcount=%d",
230 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
231 if (ext3_xattr_check_block(bh)) {
232bad_block: ext3_error(inode->i_sb, __func__,
233 "inode %lu: bad block "E3FSBLK, inode->i_ino,
234 EXT3_I(inode)->i_file_acl);
235 error = -EIO;
236 goto cleanup;
237 }
238 ext3_xattr_cache_insert(bh);
239 entry = BFIRST(bh);
240 error = ext3_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
241 if (error == -EIO)
242 goto bad_block;
243 if (error)
244 goto cleanup;
245 size = le32_to_cpu(entry->e_value_size);
246 if (buffer) {
247 error = -ERANGE;
248 if (size > buffer_size)
249 goto cleanup;
250 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
251 size);
252 }
253 error = size;
254
255cleanup:
256 brelse(bh);
257 return error;
258}
259
260static int
261ext3_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
262 void *buffer, size_t buffer_size)
263{
264 struct ext3_xattr_ibody_header *header;
265 struct ext3_xattr_entry *entry;
266 struct ext3_inode *raw_inode;
267 struct ext3_iloc iloc;
268 size_t size;
269 void *end;
270 int error;
271
272 if (!ext3_test_inode_state(inode, EXT3_STATE_XATTR))
273 return -ENODATA;
274 error = ext3_get_inode_loc(inode, &iloc);
275 if (error)
276 return error;
277 raw_inode = ext3_raw_inode(&iloc);
278 header = IHDR(inode, raw_inode);
279 entry = IFIRST(header);
280 end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
281 error = ext3_xattr_check_names(entry, end);
282 if (error)
283 goto cleanup;
284 error = ext3_xattr_find_entry(&entry, name_index, name,
285 end - (void *)entry, 0);
286 if (error)
287 goto cleanup;
288 size = le32_to_cpu(entry->e_value_size);
289 if (buffer) {
290 error = -ERANGE;
291 if (size > buffer_size)
292 goto cleanup;
293 memcpy(buffer, (void *)IFIRST(header) +
294 le16_to_cpu(entry->e_value_offs), size);
295 }
296 error = size;
297
298cleanup:
299 brelse(iloc.bh);
300 return error;
301}
302
303/*
304 * ext3_xattr_get()
305 *
306 * Copy an extended attribute into the buffer
307 * provided, or compute the buffer size required.
308 * Buffer is NULL to compute the size of the buffer required.
309 *
310 * Returns a negative error number on failure, or the number of bytes
311 * used / required on success.
312 */
313int
314ext3_xattr_get(struct inode *inode, int name_index, const char *name,
315 void *buffer, size_t buffer_size)
316{
317 int error;
318
319 down_read(&EXT3_I(inode)->xattr_sem);
320 error = ext3_xattr_ibody_get(inode, name_index, name, buffer,
321 buffer_size);
322 if (error == -ENODATA)
323 error = ext3_xattr_block_get(inode, name_index, name, buffer,
324 buffer_size);
325 up_read(&EXT3_I(inode)->xattr_sem);
326 return error;
327}
328
329static int
330ext3_xattr_list_entries(struct dentry *dentry, struct ext3_xattr_entry *entry,
331 char *buffer, size_t buffer_size)
332{
333 size_t rest = buffer_size;
334
335 for (; !IS_LAST_ENTRY(entry); entry = EXT3_XATTR_NEXT(entry)) {
336 const struct xattr_handler *handler =
337 ext3_xattr_handler(entry->e_name_index);
338
339 if (handler) {
340 size_t size = handler->list(dentry, buffer, rest,
341 entry->e_name,
342 entry->e_name_len,
343 handler->flags);
344 if (buffer) {
345 if (size > rest)
346 return -ERANGE;
347 buffer += size;
348 }
349 rest -= size;
350 }
351 }
352 return buffer_size - rest;
353}
354
355static int
356ext3_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
357{
358 struct inode *inode = dentry->d_inode;
359 struct buffer_head *bh = NULL;
360 int error;
361
362 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
363 buffer, (long)buffer_size);
364
365 error = 0;
366 if (!EXT3_I(inode)->i_file_acl)
367 goto cleanup;
368 ea_idebug(inode, "reading block %u", EXT3_I(inode)->i_file_acl);
369 bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
370 error = -EIO;
371 if (!bh)
372 goto cleanup;
373 ea_bdebug(bh, "b_count=%d, refcount=%d",
374 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
375 if (ext3_xattr_check_block(bh)) {
376 ext3_error(inode->i_sb, __func__,
377 "inode %lu: bad block "E3FSBLK, inode->i_ino,
378 EXT3_I(inode)->i_file_acl);
379 error = -EIO;
380 goto cleanup;
381 }
382 ext3_xattr_cache_insert(bh);
383 error = ext3_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
384
385cleanup:
386 brelse(bh);
387
388 return error;
389}
390
391static int
392ext3_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
393{
394 struct inode *inode = dentry->d_inode;
395 struct ext3_xattr_ibody_header *header;
396 struct ext3_inode *raw_inode;
397 struct ext3_iloc iloc;
398 void *end;
399 int error;
400
401 if (!ext3_test_inode_state(inode, EXT3_STATE_XATTR))
402 return 0;
403 error = ext3_get_inode_loc(inode, &iloc);
404 if (error)
405 return error;
406 raw_inode = ext3_raw_inode(&iloc);
407 header = IHDR(inode, raw_inode);
408 end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
409 error = ext3_xattr_check_names(IFIRST(header), end);
410 if (error)
411 goto cleanup;
412 error = ext3_xattr_list_entries(dentry, IFIRST(header),
413 buffer, buffer_size);
414
415cleanup:
416 brelse(iloc.bh);
417 return error;
418}
419
420/*
421 * ext3_xattr_list()
422 *
423 * Copy a list of attribute names into the buffer
424 * provided, or compute the buffer size required.
425 * Buffer is NULL to compute the size of the buffer required.
426 *
427 * Returns a negative error number on failure, or the number of bytes
428 * used / required on success.
429 */
430static int
431ext3_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
432{
433 int i_error, b_error;
434
435 down_read(&EXT3_I(dentry->d_inode)->xattr_sem);
436 i_error = ext3_xattr_ibody_list(dentry, buffer, buffer_size);
437 if (i_error < 0) {
438 b_error = 0;
439 } else {
440 if (buffer) {
441 buffer += i_error;
442 buffer_size -= i_error;
443 }
444 b_error = ext3_xattr_block_list(dentry, buffer, buffer_size);
445 if (b_error < 0)
446 i_error = 0;
447 }
448 up_read(&EXT3_I(dentry->d_inode)->xattr_sem);
449 return i_error + b_error;
450}
451
452/*
453 * If the EXT3_FEATURE_COMPAT_EXT_ATTR feature of this file system is
454 * not set, set it.
455 */
456static void ext3_xattr_update_super_block(handle_t *handle,
457 struct super_block *sb)
458{
459 if (EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_EXT_ATTR))
460 return;
461
462 if (ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh) == 0) {
463 EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_EXT_ATTR);
464 ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
465 }
466}
467
468/*
469 * Release the xattr block BH: If the reference count is > 1, decrement
470 * it; otherwise free the block.
471 */
472static void
473ext3_xattr_release_block(handle_t *handle, struct inode *inode,
474 struct buffer_head *bh)
475{
476 struct mb_cache_entry *ce = NULL;
477 int error = 0;
478
479 ce = mb_cache_entry_get(ext3_xattr_cache, bh->b_bdev, bh->b_blocknr);
480 error = ext3_journal_get_write_access(handle, bh);
481 if (error)
482 goto out;
483
484 lock_buffer(bh);
485
486 if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
487 ea_bdebug(bh, "refcount now=0; freeing");
488 if (ce)
489 mb_cache_entry_free(ce);
490 ext3_free_blocks(handle, inode, bh->b_blocknr, 1);
491 get_bh(bh);
492 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
493 } else {
494 le32_add_cpu(&BHDR(bh)->h_refcount, -1);
495 error = ext3_journal_dirty_metadata(handle, bh);
496 if (IS_SYNC(inode))
497 handle->h_sync = 1;
498 dquot_free_block(inode, 1);
499 ea_bdebug(bh, "refcount now=%d; releasing",
500 le32_to_cpu(BHDR(bh)->h_refcount));
501 if (ce)
502 mb_cache_entry_release(ce);
503 }
504 unlock_buffer(bh);
505out:
506 ext3_std_error(inode->i_sb, error);
507 return;
508}
509
510struct ext3_xattr_info {
511 int name_index;
512 const char *name;
513 const void *value;
514 size_t value_len;
515};
516
517struct ext3_xattr_search {
518 struct ext3_xattr_entry *first;
519 void *base;
520 void *end;
521 struct ext3_xattr_entry *here;
522 int not_found;
523};
524
525static int
526ext3_xattr_set_entry(struct ext3_xattr_info *i, struct ext3_xattr_search *s)
527{
528 struct ext3_xattr_entry *last;
529 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
530
531 /* Compute min_offs and last. */
532 last = s->first;
533 for (; !IS_LAST_ENTRY(last); last = EXT3_XATTR_NEXT(last)) {
534 if (!last->e_value_block && last->e_value_size) {
535 size_t offs = le16_to_cpu(last->e_value_offs);
536 if (offs < min_offs)
537 min_offs = offs;
538 }
539 }
540 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
541 if (!s->not_found) {
542 if (!s->here->e_value_block && s->here->e_value_size) {
543 size_t size = le32_to_cpu(s->here->e_value_size);
544 free += EXT3_XATTR_SIZE(size);
545 }
546 free += EXT3_XATTR_LEN(name_len);
547 }
548 if (i->value) {
549 if (free < EXT3_XATTR_SIZE(i->value_len) ||
550 free < EXT3_XATTR_LEN(name_len) +
551 EXT3_XATTR_SIZE(i->value_len))
552 return -ENOSPC;
553 }
554
555 if (i->value && s->not_found) {
556 /* Insert the new name. */
557 size_t size = EXT3_XATTR_LEN(name_len);
558 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
559 memmove((void *)s->here + size, s->here, rest);
560 memset(s->here, 0, size);
561 s->here->e_name_index = i->name_index;
562 s->here->e_name_len = name_len;
563 memcpy(s->here->e_name, i->name, name_len);
564 } else {
565 if (!s->here->e_value_block && s->here->e_value_size) {
566 void *first_val = s->base + min_offs;
567 size_t offs = le16_to_cpu(s->here->e_value_offs);
568 void *val = s->base + offs;
569 size_t size = EXT3_XATTR_SIZE(
570 le32_to_cpu(s->here->e_value_size));
571
572 if (i->value && size == EXT3_XATTR_SIZE(i->value_len)) {
573 /* The old and the new value have the same
574 size. Just replace. */
575 s->here->e_value_size =
576 cpu_to_le32(i->value_len);
577 memset(val + size - EXT3_XATTR_PAD, 0,
578 EXT3_XATTR_PAD); /* Clear pad bytes. */
579 memcpy(val, i->value, i->value_len);
580 return 0;
581 }
582
583 /* Remove the old value. */
584 memmove(first_val + size, first_val, val - first_val);
585 memset(first_val, 0, size);
586 s->here->e_value_size = 0;
587 s->here->e_value_offs = 0;
588 min_offs += size;
589
590 /* Adjust all value offsets. */
591 last = s->first;
592 while (!IS_LAST_ENTRY(last)) {
593 size_t o = le16_to_cpu(last->e_value_offs);
594 if (!last->e_value_block &&
595 last->e_value_size && o < offs)
596 last->e_value_offs =
597 cpu_to_le16(o + size);
598 last = EXT3_XATTR_NEXT(last);
599 }
600 }
601 if (!i->value) {
602 /* Remove the old name. */
603 size_t size = EXT3_XATTR_LEN(name_len);
604 last = ENTRY((void *)last - size);
605 memmove(s->here, (void *)s->here + size,
606 (void *)last - (void *)s->here + sizeof(__u32));
607 memset(last, 0, size);
608 }
609 }
610
611 if (i->value) {
612 /* Insert the new value. */
613 s->here->e_value_size = cpu_to_le32(i->value_len);
614 if (i->value_len) {
615 size_t size = EXT3_XATTR_SIZE(i->value_len);
616 void *val = s->base + min_offs - size;
617 s->here->e_value_offs = cpu_to_le16(min_offs - size);
618 memset(val + size - EXT3_XATTR_PAD, 0,
619 EXT3_XATTR_PAD); /* Clear the pad bytes. */
620 memcpy(val, i->value, i->value_len);
621 }
622 }
623 return 0;
624}
625
626struct ext3_xattr_block_find {
627 struct ext3_xattr_search s;
628 struct buffer_head *bh;
629};
630
631static int
632ext3_xattr_block_find(struct inode *inode, struct ext3_xattr_info *i,
633 struct ext3_xattr_block_find *bs)
634{
635 struct super_block *sb = inode->i_sb;
636 int error;
637
638 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
639 i->name_index, i->name, i->value, (long)i->value_len);
640
641 if (EXT3_I(inode)->i_file_acl) {
642 /* The inode already has an extended attribute block. */
643 bs->bh = sb_bread(sb, EXT3_I(inode)->i_file_acl);
644 error = -EIO;
645 if (!bs->bh)
646 goto cleanup;
647 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
648 atomic_read(&(bs->bh->b_count)),
649 le32_to_cpu(BHDR(bs->bh)->h_refcount));
650 if (ext3_xattr_check_block(bs->bh)) {
651 ext3_error(sb, __func__,
652 "inode %lu: bad block "E3FSBLK, inode->i_ino,
653 EXT3_I(inode)->i_file_acl);
654 error = -EIO;
655 goto cleanup;
656 }
657 /* Find the named attribute. */
658 bs->s.base = BHDR(bs->bh);
659 bs->s.first = BFIRST(bs->bh);
660 bs->s.end = bs->bh->b_data + bs->bh->b_size;
661 bs->s.here = bs->s.first;
662 error = ext3_xattr_find_entry(&bs->s.here, i->name_index,
663 i->name, bs->bh->b_size, 1);
664 if (error && error != -ENODATA)
665 goto cleanup;
666 bs->s.not_found = error;
667 }
668 error = 0;
669
670cleanup:
671 return error;
672}
673
674static int
675ext3_xattr_block_set(handle_t *handle, struct inode *inode,
676 struct ext3_xattr_info *i,
677 struct ext3_xattr_block_find *bs)
678{
679 struct super_block *sb = inode->i_sb;
680 struct buffer_head *new_bh = NULL;
681 struct ext3_xattr_search *s = &bs->s;
682 struct mb_cache_entry *ce = NULL;
683 int error = 0;
684
685#define header(x) ((struct ext3_xattr_header *)(x))
686
687 if (i->value && i->value_len > sb->s_blocksize)
688 return -ENOSPC;
689 if (s->base) {
690 ce = mb_cache_entry_get(ext3_xattr_cache, bs->bh->b_bdev,
691 bs->bh->b_blocknr);
692 error = ext3_journal_get_write_access(handle, bs->bh);
693 if (error)
694 goto cleanup;
695 lock_buffer(bs->bh);
696
697 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
698 if (ce) {
699 mb_cache_entry_free(ce);
700 ce = NULL;
701 }
702 ea_bdebug(bs->bh, "modifying in-place");
703 error = ext3_xattr_set_entry(i, s);
704 if (!error) {
705 if (!IS_LAST_ENTRY(s->first))
706 ext3_xattr_rehash(header(s->base),
707 s->here);
708 ext3_xattr_cache_insert(bs->bh);
709 }
710 unlock_buffer(bs->bh);
711 if (error == -EIO)
712 goto bad_block;
713 if (!error)
714 error = ext3_journal_dirty_metadata(handle,
715 bs->bh);
716 if (error)
717 goto cleanup;
718 goto inserted;
719 } else {
720 int offset = (char *)s->here - bs->bh->b_data;
721
722 unlock_buffer(bs->bh);
723 journal_release_buffer(handle, bs->bh);
724
725 if (ce) {
726 mb_cache_entry_release(ce);
727 ce = NULL;
728 }
729 ea_bdebug(bs->bh, "cloning");
730 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
731 error = -ENOMEM;
732 if (s->base == NULL)
733 goto cleanup;
734 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
735 s->first = ENTRY(header(s->base)+1);
736 header(s->base)->h_refcount = cpu_to_le32(1);
737 s->here = ENTRY(s->base + offset);
738 s->end = s->base + bs->bh->b_size;
739 }
740 } else {
741 /* Allocate a buffer where we construct the new block. */
742 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
743 /* assert(header == s->base) */
744 error = -ENOMEM;
745 if (s->base == NULL)
746 goto cleanup;
747 header(s->base)->h_magic = cpu_to_le32(EXT3_XATTR_MAGIC);
748 header(s->base)->h_blocks = cpu_to_le32(1);
749 header(s->base)->h_refcount = cpu_to_le32(1);
750 s->first = ENTRY(header(s->base)+1);
751 s->here = ENTRY(header(s->base)+1);
752 s->end = s->base + sb->s_blocksize;
753 }
754
755 error = ext3_xattr_set_entry(i, s);
756 if (error == -EIO)
757 goto bad_block;
758 if (error)
759 goto cleanup;
760 if (!IS_LAST_ENTRY(s->first))
761 ext3_xattr_rehash(header(s->base), s->here);
762
763inserted:
764 if (!IS_LAST_ENTRY(s->first)) {
765 new_bh = ext3_xattr_cache_find(inode, header(s->base), &ce);
766 if (new_bh) {
767 /* We found an identical block in the cache. */
768 if (new_bh == bs->bh)
769 ea_bdebug(new_bh, "keeping");
770 else {
771 /* The old block is released after updating
772 the inode. */
773 error = dquot_alloc_block(inode, 1);
774 if (error)
775 goto cleanup;
776 error = ext3_journal_get_write_access(handle,
777 new_bh);
778 if (error)
779 goto cleanup_dquot;
780 lock_buffer(new_bh);
781 le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
782 ea_bdebug(new_bh, "reusing; refcount now=%d",
783 le32_to_cpu(BHDR(new_bh)->h_refcount));
784 unlock_buffer(new_bh);
785 error = ext3_journal_dirty_metadata(handle,
786 new_bh);
787 if (error)
788 goto cleanup_dquot;
789 }
790 mb_cache_entry_release(ce);
791 ce = NULL;
792 } else if (bs->bh && s->base == bs->bh->b_data) {
793 /* We were modifying this block in-place. */
794 ea_bdebug(bs->bh, "keeping this block");
795 new_bh = bs->bh;
796 get_bh(new_bh);
797 } else {
798 /* We need to allocate a new block */
799 ext3_fsblk_t goal = ext3_group_first_block_no(sb,
800 EXT3_I(inode)->i_block_group);
801 ext3_fsblk_t block;
802
803 /*
804 * Protect us agaist concurrent allocations to the
805 * same inode from ext3_..._writepage(). Reservation
806 * code does not expect racing allocations.
807 */
808 mutex_lock(&EXT3_I(inode)->truncate_mutex);
809 block = ext3_new_block(handle, inode, goal, &error);
810 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
811 if (error)
812 goto cleanup;
813 ea_idebug(inode, "creating block %d", block);
814
815 new_bh = sb_getblk(sb, block);
816 if (!new_bh) {
817getblk_failed:
818 ext3_free_blocks(handle, inode, block, 1);
819 error = -EIO;
820 goto cleanup;
821 }
822 lock_buffer(new_bh);
823 error = ext3_journal_get_create_access(handle, new_bh);
824 if (error) {
825 unlock_buffer(new_bh);
826 goto getblk_failed;
827 }
828 memcpy(new_bh->b_data, s->base, new_bh->b_size);
829 set_buffer_uptodate(new_bh);
830 unlock_buffer(new_bh);
831 ext3_xattr_cache_insert(new_bh);
832 error = ext3_journal_dirty_metadata(handle, new_bh);
833 if (error)
834 goto cleanup;
835 }
836 }
837
838 /* Update the inode. */
839 EXT3_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
840
841 /* Drop the previous xattr block. */
842 if (bs->bh && bs->bh != new_bh)
843 ext3_xattr_release_block(handle, inode, bs->bh);
844 error = 0;
845
846cleanup:
847 if (ce)
848 mb_cache_entry_release(ce);
849 brelse(new_bh);
850 if (!(bs->bh && s->base == bs->bh->b_data))
851 kfree(s->base);
852
853 return error;
854
855cleanup_dquot:
856 dquot_free_block(inode, 1);
857 goto cleanup;
858
859bad_block:
860 ext3_error(inode->i_sb, __func__,
861 "inode %lu: bad block "E3FSBLK, inode->i_ino,
862 EXT3_I(inode)->i_file_acl);
863 goto cleanup;
864
865#undef header
866}
867
868struct ext3_xattr_ibody_find {
869 struct ext3_xattr_search s;
870 struct ext3_iloc iloc;
871};
872
873static int
874ext3_xattr_ibody_find(struct inode *inode, struct ext3_xattr_info *i,
875 struct ext3_xattr_ibody_find *is)
876{
877 struct ext3_xattr_ibody_header *header;
878 struct ext3_inode *raw_inode;
879 int error;
880
881 if (EXT3_I(inode)->i_extra_isize == 0)
882 return 0;
883 raw_inode = ext3_raw_inode(&is->iloc);
884 header = IHDR(inode, raw_inode);
885 is->s.base = is->s.first = IFIRST(header);
886 is->s.here = is->s.first;
887 is->s.end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
888 if (ext3_test_inode_state(inode, EXT3_STATE_XATTR)) {
889 error = ext3_xattr_check_names(IFIRST(header), is->s.end);
890 if (error)
891 return error;
892 /* Find the named attribute. */
893 error = ext3_xattr_find_entry(&is->s.here, i->name_index,
894 i->name, is->s.end -
895 (void *)is->s.base, 0);
896 if (error && error != -ENODATA)
897 return error;
898 is->s.not_found = error;
899 }
900 return 0;
901}
902
903static int
904ext3_xattr_ibody_set(handle_t *handle, struct inode *inode,
905 struct ext3_xattr_info *i,
906 struct ext3_xattr_ibody_find *is)
907{
908 struct ext3_xattr_ibody_header *header;
909 struct ext3_xattr_search *s = &is->s;
910 int error;
911
912 if (EXT3_I(inode)->i_extra_isize == 0)
913 return -ENOSPC;
914 error = ext3_xattr_set_entry(i, s);
915 if (error)
916 return error;
917 header = IHDR(inode, ext3_raw_inode(&is->iloc));
918 if (!IS_LAST_ENTRY(s->first)) {
919 header->h_magic = cpu_to_le32(EXT3_XATTR_MAGIC);
920 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
921 } else {
922 header->h_magic = cpu_to_le32(0);
923 ext3_clear_inode_state(inode, EXT3_STATE_XATTR);
924 }
925 return 0;
926}
927
928/*
929 * ext3_xattr_set_handle()
930 *
931 * Create, replace or remove an extended attribute for this inode. Value
932 * is NULL to remove an existing extended attribute, and non-NULL to
933 * either replace an existing extended attribute, or create a new extended
934 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
935 * specify that an extended attribute must exist and must not exist
936 * previous to the call, respectively.
937 *
938 * Returns 0, or a negative error number on failure.
939 */
940int
941ext3_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
942 const char *name, const void *value, size_t value_len,
943 int flags)
944{
945 struct ext3_xattr_info i = {
946 .name_index = name_index,
947 .name = name,
948 .value = value,
949 .value_len = value_len,
950
951 };
952 struct ext3_xattr_ibody_find is = {
953 .s = { .not_found = -ENODATA, },
954 };
955 struct ext3_xattr_block_find bs = {
956 .s = { .not_found = -ENODATA, },
957 };
958 int error;
959
960 if (!name)
961 return -EINVAL;
962 if (strlen(name) > 255)
963 return -ERANGE;
964 down_write(&EXT3_I(inode)->xattr_sem);
965 error = ext3_get_inode_loc(inode, &is.iloc);
966 if (error)
967 goto cleanup;
968
969 error = ext3_journal_get_write_access(handle, is.iloc.bh);
970 if (error)
971 goto cleanup;
972
973 if (ext3_test_inode_state(inode, EXT3_STATE_NEW)) {
974 struct ext3_inode *raw_inode = ext3_raw_inode(&is.iloc);
975 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
976 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
977 }
978
979 error = ext3_xattr_ibody_find(inode, &i, &is);
980 if (error)
981 goto cleanup;
982 if (is.s.not_found)
983 error = ext3_xattr_block_find(inode, &i, &bs);
984 if (error)
985 goto cleanup;
986 if (is.s.not_found && bs.s.not_found) {
987 error = -ENODATA;
988 if (flags & XATTR_REPLACE)
989 goto cleanup;
990 error = 0;
991 if (!value)
992 goto cleanup;
993 } else {
994 error = -EEXIST;
995 if (flags & XATTR_CREATE)
996 goto cleanup;
997 }
998 if (!value) {
999 if (!is.s.not_found)
1000 error = ext3_xattr_ibody_set(handle, inode, &i, &is);
1001 else if (!bs.s.not_found)
1002 error = ext3_xattr_block_set(handle, inode, &i, &bs);
1003 } else {
1004 error = ext3_xattr_ibody_set(handle, inode, &i, &is);
1005 if (!error && !bs.s.not_found) {
1006 i.value = NULL;
1007 error = ext3_xattr_block_set(handle, inode, &i, &bs);
1008 } else if (error == -ENOSPC) {
1009 if (EXT3_I(inode)->i_file_acl && !bs.s.base) {
1010 error = ext3_xattr_block_find(inode, &i, &bs);
1011 if (error)
1012 goto cleanup;
1013 }
1014 error = ext3_xattr_block_set(handle, inode, &i, &bs);
1015 if (error)
1016 goto cleanup;
1017 if (!is.s.not_found) {
1018 i.value = NULL;
1019 error = ext3_xattr_ibody_set(handle, inode, &i,
1020 &is);
1021 }
1022 }
1023 }
1024 if (!error) {
1025 ext3_xattr_update_super_block(handle, inode->i_sb);
1026 inode->i_ctime = CURRENT_TIME_SEC;
1027 error = ext3_mark_iloc_dirty(handle, inode, &is.iloc);
1028 /*
1029 * The bh is consumed by ext3_mark_iloc_dirty, even with
1030 * error != 0.
1031 */
1032 is.iloc.bh = NULL;
1033 if (IS_SYNC(inode))
1034 handle->h_sync = 1;
1035 }
1036
1037cleanup:
1038 brelse(is.iloc.bh);
1039 brelse(bs.bh);
1040 up_write(&EXT3_I(inode)->xattr_sem);
1041 return error;
1042}
1043
1044/*
1045 * ext3_xattr_set()
1046 *
1047 * Like ext3_xattr_set_handle, but start from an inode. This extended
1048 * attribute modification is a filesystem transaction by itself.
1049 *
1050 * Returns 0, or a negative error number on failure.
1051 */
1052int
1053ext3_xattr_set(struct inode *inode, int name_index, const char *name,
1054 const void *value, size_t value_len, int flags)
1055{
1056 handle_t *handle;
1057 int error, retries = 0;
1058
1059retry:
1060 handle = ext3_journal_start(inode, EXT3_DATA_TRANS_BLOCKS(inode->i_sb));
1061 if (IS_ERR(handle)) {
1062 error = PTR_ERR(handle);
1063 } else {
1064 int error2;
1065
1066 error = ext3_xattr_set_handle(handle, inode, name_index, name,
1067 value, value_len, flags);
1068 error2 = ext3_journal_stop(handle);
1069 if (error == -ENOSPC &&
1070 ext3_should_retry_alloc(inode->i_sb, &retries))
1071 goto retry;
1072 if (error == 0)
1073 error = error2;
1074 }
1075
1076 return error;
1077}
1078
1079/*
1080 * ext3_xattr_delete_inode()
1081 *
1082 * Free extended attribute resources associated with this inode. This
1083 * is called immediately before an inode is freed. We have exclusive
1084 * access to the inode.
1085 */
1086void
1087ext3_xattr_delete_inode(handle_t *handle, struct inode *inode)
1088{
1089 struct buffer_head *bh = NULL;
1090
1091 if (!EXT3_I(inode)->i_file_acl)
1092 goto cleanup;
1093 bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
1094 if (!bh) {
1095 ext3_error(inode->i_sb, __func__,
1096 "inode %lu: block "E3FSBLK" read error", inode->i_ino,
1097 EXT3_I(inode)->i_file_acl);
1098 goto cleanup;
1099 }
1100 if (BHDR(bh)->h_magic != cpu_to_le32(EXT3_XATTR_MAGIC) ||
1101 BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1102 ext3_error(inode->i_sb, __func__,
1103 "inode %lu: bad block "E3FSBLK, inode->i_ino,
1104 EXT3_I(inode)->i_file_acl);
1105 goto cleanup;
1106 }
1107 ext3_xattr_release_block(handle, inode, bh);
1108 EXT3_I(inode)->i_file_acl = 0;
1109
1110cleanup:
1111 brelse(bh);
1112}
1113
1114/*
1115 * ext3_xattr_put_super()
1116 *
1117 * This is called when a file system is unmounted.
1118 */
1119void
1120ext3_xattr_put_super(struct super_block *sb)
1121{
1122 mb_cache_shrink(sb->s_bdev);
1123}
1124
1125/*
1126 * ext3_xattr_cache_insert()
1127 *
1128 * Create a new entry in the extended attribute cache, and insert
1129 * it unless such an entry is already in the cache.
1130 *
1131 * Returns 0, or a negative error number on failure.
1132 */
1133static void
1134ext3_xattr_cache_insert(struct buffer_head *bh)
1135{
1136 __u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1137 struct mb_cache_entry *ce;
1138 int error;
1139
1140 ce = mb_cache_entry_alloc(ext3_xattr_cache, GFP_NOFS);
1141 if (!ce) {
1142 ea_bdebug(bh, "out of memory");
1143 return;
1144 }
1145 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
1146 if (error) {
1147 mb_cache_entry_free(ce);
1148 if (error == -EBUSY) {
1149 ea_bdebug(bh, "already in cache");
1150 error = 0;
1151 }
1152 } else {
1153 ea_bdebug(bh, "inserting [%x]", (int)hash);
1154 mb_cache_entry_release(ce);
1155 }
1156}
1157
1158/*
1159 * ext3_xattr_cmp()
1160 *
1161 * Compare two extended attribute blocks for equality.
1162 *
1163 * Returns 0 if the blocks are equal, 1 if they differ, and
1164 * a negative error number on errors.
1165 */
1166static int
1167ext3_xattr_cmp(struct ext3_xattr_header *header1,
1168 struct ext3_xattr_header *header2)
1169{
1170 struct ext3_xattr_entry *entry1, *entry2;
1171
1172 entry1 = ENTRY(header1+1);
1173 entry2 = ENTRY(header2+1);
1174 while (!IS_LAST_ENTRY(entry1)) {
1175 if (IS_LAST_ENTRY(entry2))
1176 return 1;
1177 if (entry1->e_hash != entry2->e_hash ||
1178 entry1->e_name_index != entry2->e_name_index ||
1179 entry1->e_name_len != entry2->e_name_len ||
1180 entry1->e_value_size != entry2->e_value_size ||
1181 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1182 return 1;
1183 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1184 return -EIO;
1185 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1186 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1187 le32_to_cpu(entry1->e_value_size)))
1188 return 1;
1189
1190 entry1 = EXT3_XATTR_NEXT(entry1);
1191 entry2 = EXT3_XATTR_NEXT(entry2);
1192 }
1193 if (!IS_LAST_ENTRY(entry2))
1194 return 1;
1195 return 0;
1196}
1197
1198/*
1199 * ext3_xattr_cache_find()
1200 *
1201 * Find an identical extended attribute block.
1202 *
1203 * Returns a pointer to the block found, or NULL if such a block was
1204 * not found or an error occurred.
1205 */
1206static struct buffer_head *
1207ext3_xattr_cache_find(struct inode *inode, struct ext3_xattr_header *header,
1208 struct mb_cache_entry **pce)
1209{
1210 __u32 hash = le32_to_cpu(header->h_hash);
1211 struct mb_cache_entry *ce;
1212
1213 if (!header->h_hash)
1214 return NULL; /* never share */
1215 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1216again:
1217 ce = mb_cache_entry_find_first(ext3_xattr_cache, inode->i_sb->s_bdev,
1218 hash);
1219 while (ce) {
1220 struct buffer_head *bh;
1221
1222 if (IS_ERR(ce)) {
1223 if (PTR_ERR(ce) == -EAGAIN)
1224 goto again;
1225 break;
1226 }
1227 bh = sb_bread(inode->i_sb, ce->e_block);
1228 if (!bh) {
1229 ext3_error(inode->i_sb, __func__,
1230 "inode %lu: block %lu read error",
1231 inode->i_ino, (unsigned long) ce->e_block);
1232 } else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1233 EXT3_XATTR_REFCOUNT_MAX) {
1234 ea_idebug(inode, "block %lu refcount %d>=%d",
1235 (unsigned long) ce->e_block,
1236 le32_to_cpu(BHDR(bh)->h_refcount),
1237 EXT3_XATTR_REFCOUNT_MAX);
1238 } else if (ext3_xattr_cmp(header, BHDR(bh)) == 0) {
1239 *pce = ce;
1240 return bh;
1241 }
1242 brelse(bh);
1243 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
1244 }
1245 return NULL;
1246}
1247
1248#define NAME_HASH_SHIFT 5
1249#define VALUE_HASH_SHIFT 16
1250
1251/*
1252 * ext3_xattr_hash_entry()
1253 *
1254 * Compute the hash of an extended attribute.
1255 */
1256static inline void ext3_xattr_hash_entry(struct ext3_xattr_header *header,
1257 struct ext3_xattr_entry *entry)
1258{
1259 __u32 hash = 0;
1260 char *name = entry->e_name;
1261 int n;
1262
1263 for (n=0; n < entry->e_name_len; n++) {
1264 hash = (hash << NAME_HASH_SHIFT) ^
1265 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1266 *name++;
1267 }
1268
1269 if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1270 __le32 *value = (__le32 *)((char *)header +
1271 le16_to_cpu(entry->e_value_offs));
1272 for (n = (le32_to_cpu(entry->e_value_size) +
1273 EXT3_XATTR_ROUND) >> EXT3_XATTR_PAD_BITS; n; n--) {
1274 hash = (hash << VALUE_HASH_SHIFT) ^
1275 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1276 le32_to_cpu(*value++);
1277 }
1278 }
1279 entry->e_hash = cpu_to_le32(hash);
1280}
1281
1282#undef NAME_HASH_SHIFT
1283#undef VALUE_HASH_SHIFT
1284
1285#define BLOCK_HASH_SHIFT 16
1286
1287/*
1288 * ext3_xattr_rehash()
1289 *
1290 * Re-compute the extended attribute hash value after an entry has changed.
1291 */
1292static void ext3_xattr_rehash(struct ext3_xattr_header *header,
1293 struct ext3_xattr_entry *entry)
1294{
1295 struct ext3_xattr_entry *here;
1296 __u32 hash = 0;
1297
1298 ext3_xattr_hash_entry(header, entry);
1299 here = ENTRY(header+1);
1300 while (!IS_LAST_ENTRY(here)) {
1301 if (!here->e_hash) {
1302 /* Block is not shared if an entry's hash value == 0 */
1303 hash = 0;
1304 break;
1305 }
1306 hash = (hash << BLOCK_HASH_SHIFT) ^
1307 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1308 le32_to_cpu(here->e_hash);
1309 here = EXT3_XATTR_NEXT(here);
1310 }
1311 header->h_hash = cpu_to_le32(hash);
1312}
1313
1314#undef BLOCK_HASH_SHIFT
1315
1316int __init
1317init_ext3_xattr(void)
1318{
1319 ext3_xattr_cache = mb_cache_create("ext3_xattr", 6);
1320 if (!ext3_xattr_cache)
1321 return -ENOMEM;
1322 return 0;
1323}
1324
1325void
1326exit_ext3_xattr(void)
1327{
1328 if (ext3_xattr_cache)
1329 mb_cache_destroy(ext3_xattr_cache);
1330 ext3_xattr_cache = NULL;
1331}
1/*
2 * linux/fs/ext3/xattr.c
3 *
4 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
5 *
6 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7 * Ext3 code with a lot of help from Eric Jarman <ejarman@acm.org>.
8 * Extended attributes for symlinks and special files added per
9 * suggestion of Luka Renko <luka.renko@hermes.si>.
10 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
11 * Red Hat Inc.
12 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
13 * and Andreas Gruenbacher <agruen@suse.de>.
14 */
15
16/*
17 * Extended attributes are stored directly in inodes (on file systems with
18 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
19 * field contains the block number if an inode uses an additional block. All
20 * attributes must fit in the inode and one additional block. Blocks that
21 * contain the identical set of attributes may be shared among several inodes.
22 * Identical blocks are detected by keeping a cache of blocks that have
23 * recently been accessed.
24 *
25 * The attributes in inodes and on blocks have a different header; the entries
26 * are stored in the same format:
27 *
28 * +------------------+
29 * | header |
30 * | entry 1 | |
31 * | entry 2 | | growing downwards
32 * | entry 3 | v
33 * | four null bytes |
34 * | . . . |
35 * | value 1 | ^
36 * | value 3 | | growing upwards
37 * | value 2 | |
38 * +------------------+
39 *
40 * The header is followed by multiple entry descriptors. In disk blocks, the
41 * entry descriptors are kept sorted. In inodes, they are unsorted. The
42 * attribute values are aligned to the end of the block in no specific order.
43 *
44 * Locking strategy
45 * ----------------
46 * EXT3_I(inode)->i_file_acl is protected by EXT3_I(inode)->xattr_sem.
47 * EA blocks are only changed if they are exclusive to an inode, so
48 * holding xattr_sem also means that nothing but the EA block's reference
49 * count can change. Multiple writers to the same block are synchronized
50 * by the buffer lock.
51 */
52
53#include <linux/init.h>
54#include <linux/fs.h>
55#include <linux/slab.h>
56#include <linux/ext3_jbd.h>
57#include <linux/ext3_fs.h>
58#include <linux/mbcache.h>
59#include <linux/quotaops.h>
60#include <linux/rwsem.h>
61#include "xattr.h"
62#include "acl.h"
63
64#define BHDR(bh) ((struct ext3_xattr_header *)((bh)->b_data))
65#define ENTRY(ptr) ((struct ext3_xattr_entry *)(ptr))
66#define BFIRST(bh) ENTRY(BHDR(bh)+1)
67#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
68
69#define IHDR(inode, raw_inode) \
70 ((struct ext3_xattr_ibody_header *) \
71 ((void *)raw_inode + \
72 EXT3_GOOD_OLD_INODE_SIZE + \
73 EXT3_I(inode)->i_extra_isize))
74#define IFIRST(hdr) ((struct ext3_xattr_entry *)((hdr)+1))
75
76#ifdef EXT3_XATTR_DEBUG
77# define ea_idebug(inode, f...) do { \
78 printk(KERN_DEBUG "inode %s:%lu: ", \
79 inode->i_sb->s_id, inode->i_ino); \
80 printk(f); \
81 printk("\n"); \
82 } while (0)
83# define ea_bdebug(bh, f...) do { \
84 char b[BDEVNAME_SIZE]; \
85 printk(KERN_DEBUG "block %s:%lu: ", \
86 bdevname(bh->b_bdev, b), \
87 (unsigned long) bh->b_blocknr); \
88 printk(f); \
89 printk("\n"); \
90 } while (0)
91#else
92# define ea_idebug(f...)
93# define ea_bdebug(f...)
94#endif
95
96static void ext3_xattr_cache_insert(struct buffer_head *);
97static struct buffer_head *ext3_xattr_cache_find(struct inode *,
98 struct ext3_xattr_header *,
99 struct mb_cache_entry **);
100static void ext3_xattr_rehash(struct ext3_xattr_header *,
101 struct ext3_xattr_entry *);
102static int ext3_xattr_list(struct dentry *dentry, char *buffer,
103 size_t buffer_size);
104
105static struct mb_cache *ext3_xattr_cache;
106
107static const struct xattr_handler *ext3_xattr_handler_map[] = {
108 [EXT3_XATTR_INDEX_USER] = &ext3_xattr_user_handler,
109#ifdef CONFIG_EXT3_FS_POSIX_ACL
110 [EXT3_XATTR_INDEX_POSIX_ACL_ACCESS] = &ext3_xattr_acl_access_handler,
111 [EXT3_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext3_xattr_acl_default_handler,
112#endif
113 [EXT3_XATTR_INDEX_TRUSTED] = &ext3_xattr_trusted_handler,
114#ifdef CONFIG_EXT3_FS_SECURITY
115 [EXT3_XATTR_INDEX_SECURITY] = &ext3_xattr_security_handler,
116#endif
117};
118
119const struct xattr_handler *ext3_xattr_handlers[] = {
120 &ext3_xattr_user_handler,
121 &ext3_xattr_trusted_handler,
122#ifdef CONFIG_EXT3_FS_POSIX_ACL
123 &ext3_xattr_acl_access_handler,
124 &ext3_xattr_acl_default_handler,
125#endif
126#ifdef CONFIG_EXT3_FS_SECURITY
127 &ext3_xattr_security_handler,
128#endif
129 NULL
130};
131
132static inline const struct xattr_handler *
133ext3_xattr_handler(int name_index)
134{
135 const struct xattr_handler *handler = NULL;
136
137 if (name_index > 0 && name_index < ARRAY_SIZE(ext3_xattr_handler_map))
138 handler = ext3_xattr_handler_map[name_index];
139 return handler;
140}
141
142/*
143 * Inode operation listxattr()
144 *
145 * dentry->d_inode->i_mutex: don't care
146 */
147ssize_t
148ext3_listxattr(struct dentry *dentry, char *buffer, size_t size)
149{
150 return ext3_xattr_list(dentry, buffer, size);
151}
152
153static int
154ext3_xattr_check_names(struct ext3_xattr_entry *entry, void *end)
155{
156 while (!IS_LAST_ENTRY(entry)) {
157 struct ext3_xattr_entry *next = EXT3_XATTR_NEXT(entry);
158 if ((void *)next >= end)
159 return -EIO;
160 entry = next;
161 }
162 return 0;
163}
164
165static inline int
166ext3_xattr_check_block(struct buffer_head *bh)
167{
168 int error;
169
170 if (BHDR(bh)->h_magic != cpu_to_le32(EXT3_XATTR_MAGIC) ||
171 BHDR(bh)->h_blocks != cpu_to_le32(1))
172 return -EIO;
173 error = ext3_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size);
174 return error;
175}
176
177static inline int
178ext3_xattr_check_entry(struct ext3_xattr_entry *entry, size_t size)
179{
180 size_t value_size = le32_to_cpu(entry->e_value_size);
181
182 if (entry->e_value_block != 0 || value_size > size ||
183 le16_to_cpu(entry->e_value_offs) + value_size > size)
184 return -EIO;
185 return 0;
186}
187
188static int
189ext3_xattr_find_entry(struct ext3_xattr_entry **pentry, int name_index,
190 const char *name, size_t size, int sorted)
191{
192 struct ext3_xattr_entry *entry;
193 size_t name_len;
194 int cmp = 1;
195
196 if (name == NULL)
197 return -EINVAL;
198 name_len = strlen(name);
199 entry = *pentry;
200 for (; !IS_LAST_ENTRY(entry); entry = EXT3_XATTR_NEXT(entry)) {
201 cmp = name_index - entry->e_name_index;
202 if (!cmp)
203 cmp = name_len - entry->e_name_len;
204 if (!cmp)
205 cmp = memcmp(name, entry->e_name, name_len);
206 if (cmp <= 0 && (sorted || cmp == 0))
207 break;
208 }
209 *pentry = entry;
210 if (!cmp && ext3_xattr_check_entry(entry, size))
211 return -EIO;
212 return cmp ? -ENODATA : 0;
213}
214
215static int
216ext3_xattr_block_get(struct inode *inode, int name_index, const char *name,
217 void *buffer, size_t buffer_size)
218{
219 struct buffer_head *bh = NULL;
220 struct ext3_xattr_entry *entry;
221 size_t size;
222 int error;
223
224 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
225 name_index, name, buffer, (long)buffer_size);
226
227 error = -ENODATA;
228 if (!EXT3_I(inode)->i_file_acl)
229 goto cleanup;
230 ea_idebug(inode, "reading block %u", EXT3_I(inode)->i_file_acl);
231 bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
232 if (!bh)
233 goto cleanup;
234 ea_bdebug(bh, "b_count=%d, refcount=%d",
235 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
236 if (ext3_xattr_check_block(bh)) {
237bad_block: ext3_error(inode->i_sb, __func__,
238 "inode %lu: bad block "E3FSBLK, inode->i_ino,
239 EXT3_I(inode)->i_file_acl);
240 error = -EIO;
241 goto cleanup;
242 }
243 ext3_xattr_cache_insert(bh);
244 entry = BFIRST(bh);
245 error = ext3_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
246 if (error == -EIO)
247 goto bad_block;
248 if (error)
249 goto cleanup;
250 size = le32_to_cpu(entry->e_value_size);
251 if (buffer) {
252 error = -ERANGE;
253 if (size > buffer_size)
254 goto cleanup;
255 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
256 size);
257 }
258 error = size;
259
260cleanup:
261 brelse(bh);
262 return error;
263}
264
265static int
266ext3_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
267 void *buffer, size_t buffer_size)
268{
269 struct ext3_xattr_ibody_header *header;
270 struct ext3_xattr_entry *entry;
271 struct ext3_inode *raw_inode;
272 struct ext3_iloc iloc;
273 size_t size;
274 void *end;
275 int error;
276
277 if (!ext3_test_inode_state(inode, EXT3_STATE_XATTR))
278 return -ENODATA;
279 error = ext3_get_inode_loc(inode, &iloc);
280 if (error)
281 return error;
282 raw_inode = ext3_raw_inode(&iloc);
283 header = IHDR(inode, raw_inode);
284 entry = IFIRST(header);
285 end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
286 error = ext3_xattr_check_names(entry, end);
287 if (error)
288 goto cleanup;
289 error = ext3_xattr_find_entry(&entry, name_index, name,
290 end - (void *)entry, 0);
291 if (error)
292 goto cleanup;
293 size = le32_to_cpu(entry->e_value_size);
294 if (buffer) {
295 error = -ERANGE;
296 if (size > buffer_size)
297 goto cleanup;
298 memcpy(buffer, (void *)IFIRST(header) +
299 le16_to_cpu(entry->e_value_offs), size);
300 }
301 error = size;
302
303cleanup:
304 brelse(iloc.bh);
305 return error;
306}
307
308/*
309 * ext3_xattr_get()
310 *
311 * Copy an extended attribute into the buffer
312 * provided, or compute the buffer size required.
313 * Buffer is NULL to compute the size of the buffer required.
314 *
315 * Returns a negative error number on failure, or the number of bytes
316 * used / required on success.
317 */
318int
319ext3_xattr_get(struct inode *inode, int name_index, const char *name,
320 void *buffer, size_t buffer_size)
321{
322 int error;
323
324 down_read(&EXT3_I(inode)->xattr_sem);
325 error = ext3_xattr_ibody_get(inode, name_index, name, buffer,
326 buffer_size);
327 if (error == -ENODATA)
328 error = ext3_xattr_block_get(inode, name_index, name, buffer,
329 buffer_size);
330 up_read(&EXT3_I(inode)->xattr_sem);
331 return error;
332}
333
334static int
335ext3_xattr_list_entries(struct dentry *dentry, struct ext3_xattr_entry *entry,
336 char *buffer, size_t buffer_size)
337{
338 size_t rest = buffer_size;
339
340 for (; !IS_LAST_ENTRY(entry); entry = EXT3_XATTR_NEXT(entry)) {
341 const struct xattr_handler *handler =
342 ext3_xattr_handler(entry->e_name_index);
343
344 if (handler) {
345 size_t size = handler->list(dentry, buffer, rest,
346 entry->e_name,
347 entry->e_name_len,
348 handler->flags);
349 if (buffer) {
350 if (size > rest)
351 return -ERANGE;
352 buffer += size;
353 }
354 rest -= size;
355 }
356 }
357 return buffer_size - rest;
358}
359
360static int
361ext3_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
362{
363 struct inode *inode = dentry->d_inode;
364 struct buffer_head *bh = NULL;
365 int error;
366
367 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
368 buffer, (long)buffer_size);
369
370 error = 0;
371 if (!EXT3_I(inode)->i_file_acl)
372 goto cleanup;
373 ea_idebug(inode, "reading block %u", EXT3_I(inode)->i_file_acl);
374 bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
375 error = -EIO;
376 if (!bh)
377 goto cleanup;
378 ea_bdebug(bh, "b_count=%d, refcount=%d",
379 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
380 if (ext3_xattr_check_block(bh)) {
381 ext3_error(inode->i_sb, __func__,
382 "inode %lu: bad block "E3FSBLK, inode->i_ino,
383 EXT3_I(inode)->i_file_acl);
384 error = -EIO;
385 goto cleanup;
386 }
387 ext3_xattr_cache_insert(bh);
388 error = ext3_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
389
390cleanup:
391 brelse(bh);
392
393 return error;
394}
395
396static int
397ext3_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
398{
399 struct inode *inode = dentry->d_inode;
400 struct ext3_xattr_ibody_header *header;
401 struct ext3_inode *raw_inode;
402 struct ext3_iloc iloc;
403 void *end;
404 int error;
405
406 if (!ext3_test_inode_state(inode, EXT3_STATE_XATTR))
407 return 0;
408 error = ext3_get_inode_loc(inode, &iloc);
409 if (error)
410 return error;
411 raw_inode = ext3_raw_inode(&iloc);
412 header = IHDR(inode, raw_inode);
413 end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
414 error = ext3_xattr_check_names(IFIRST(header), end);
415 if (error)
416 goto cleanup;
417 error = ext3_xattr_list_entries(dentry, IFIRST(header),
418 buffer, buffer_size);
419
420cleanup:
421 brelse(iloc.bh);
422 return error;
423}
424
425/*
426 * ext3_xattr_list()
427 *
428 * Copy a list of attribute names into the buffer
429 * provided, or compute the buffer size required.
430 * Buffer is NULL to compute the size of the buffer required.
431 *
432 * Returns a negative error number on failure, or the number of bytes
433 * used / required on success.
434 */
435static int
436ext3_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
437{
438 int i_error, b_error;
439
440 down_read(&EXT3_I(dentry->d_inode)->xattr_sem);
441 i_error = ext3_xattr_ibody_list(dentry, buffer, buffer_size);
442 if (i_error < 0) {
443 b_error = 0;
444 } else {
445 if (buffer) {
446 buffer += i_error;
447 buffer_size -= i_error;
448 }
449 b_error = ext3_xattr_block_list(dentry, buffer, buffer_size);
450 if (b_error < 0)
451 i_error = 0;
452 }
453 up_read(&EXT3_I(dentry->d_inode)->xattr_sem);
454 return i_error + b_error;
455}
456
457/*
458 * If the EXT3_FEATURE_COMPAT_EXT_ATTR feature of this file system is
459 * not set, set it.
460 */
461static void ext3_xattr_update_super_block(handle_t *handle,
462 struct super_block *sb)
463{
464 if (EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_EXT_ATTR))
465 return;
466
467 if (ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh) == 0) {
468 EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_EXT_ATTR);
469 ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
470 }
471}
472
473/*
474 * Release the xattr block BH: If the reference count is > 1, decrement
475 * it; otherwise free the block.
476 */
477static void
478ext3_xattr_release_block(handle_t *handle, struct inode *inode,
479 struct buffer_head *bh)
480{
481 struct mb_cache_entry *ce = NULL;
482 int error = 0;
483
484 ce = mb_cache_entry_get(ext3_xattr_cache, bh->b_bdev, bh->b_blocknr);
485 error = ext3_journal_get_write_access(handle, bh);
486 if (error)
487 goto out;
488
489 lock_buffer(bh);
490
491 if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
492 ea_bdebug(bh, "refcount now=0; freeing");
493 if (ce)
494 mb_cache_entry_free(ce);
495 ext3_free_blocks(handle, inode, bh->b_blocknr, 1);
496 get_bh(bh);
497 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
498 } else {
499 le32_add_cpu(&BHDR(bh)->h_refcount, -1);
500 error = ext3_journal_dirty_metadata(handle, bh);
501 if (IS_SYNC(inode))
502 handle->h_sync = 1;
503 dquot_free_block(inode, 1);
504 ea_bdebug(bh, "refcount now=%d; releasing",
505 le32_to_cpu(BHDR(bh)->h_refcount));
506 if (ce)
507 mb_cache_entry_release(ce);
508 }
509 unlock_buffer(bh);
510out:
511 ext3_std_error(inode->i_sb, error);
512 return;
513}
514
515struct ext3_xattr_info {
516 int name_index;
517 const char *name;
518 const void *value;
519 size_t value_len;
520};
521
522struct ext3_xattr_search {
523 struct ext3_xattr_entry *first;
524 void *base;
525 void *end;
526 struct ext3_xattr_entry *here;
527 int not_found;
528};
529
530static int
531ext3_xattr_set_entry(struct ext3_xattr_info *i, struct ext3_xattr_search *s)
532{
533 struct ext3_xattr_entry *last;
534 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
535
536 /* Compute min_offs and last. */
537 last = s->first;
538 for (; !IS_LAST_ENTRY(last); last = EXT3_XATTR_NEXT(last)) {
539 if (!last->e_value_block && last->e_value_size) {
540 size_t offs = le16_to_cpu(last->e_value_offs);
541 if (offs < min_offs)
542 min_offs = offs;
543 }
544 }
545 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
546 if (!s->not_found) {
547 if (!s->here->e_value_block && s->here->e_value_size) {
548 size_t size = le32_to_cpu(s->here->e_value_size);
549 free += EXT3_XATTR_SIZE(size);
550 }
551 free += EXT3_XATTR_LEN(name_len);
552 }
553 if (i->value) {
554 if (free < EXT3_XATTR_SIZE(i->value_len) ||
555 free < EXT3_XATTR_LEN(name_len) +
556 EXT3_XATTR_SIZE(i->value_len))
557 return -ENOSPC;
558 }
559
560 if (i->value && s->not_found) {
561 /* Insert the new name. */
562 size_t size = EXT3_XATTR_LEN(name_len);
563 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
564 memmove((void *)s->here + size, s->here, rest);
565 memset(s->here, 0, size);
566 s->here->e_name_index = i->name_index;
567 s->here->e_name_len = name_len;
568 memcpy(s->here->e_name, i->name, name_len);
569 } else {
570 if (!s->here->e_value_block && s->here->e_value_size) {
571 void *first_val = s->base + min_offs;
572 size_t offs = le16_to_cpu(s->here->e_value_offs);
573 void *val = s->base + offs;
574 size_t size = EXT3_XATTR_SIZE(
575 le32_to_cpu(s->here->e_value_size));
576
577 if (i->value && size == EXT3_XATTR_SIZE(i->value_len)) {
578 /* The old and the new value have the same
579 size. Just replace. */
580 s->here->e_value_size =
581 cpu_to_le32(i->value_len);
582 memset(val + size - EXT3_XATTR_PAD, 0,
583 EXT3_XATTR_PAD); /* Clear pad bytes. */
584 memcpy(val, i->value, i->value_len);
585 return 0;
586 }
587
588 /* Remove the old value. */
589 memmove(first_val + size, first_val, val - first_val);
590 memset(first_val, 0, size);
591 s->here->e_value_size = 0;
592 s->here->e_value_offs = 0;
593 min_offs += size;
594
595 /* Adjust all value offsets. */
596 last = s->first;
597 while (!IS_LAST_ENTRY(last)) {
598 size_t o = le16_to_cpu(last->e_value_offs);
599 if (!last->e_value_block &&
600 last->e_value_size && o < offs)
601 last->e_value_offs =
602 cpu_to_le16(o + size);
603 last = EXT3_XATTR_NEXT(last);
604 }
605 }
606 if (!i->value) {
607 /* Remove the old name. */
608 size_t size = EXT3_XATTR_LEN(name_len);
609 last = ENTRY((void *)last - size);
610 memmove(s->here, (void *)s->here + size,
611 (void *)last - (void *)s->here + sizeof(__u32));
612 memset(last, 0, size);
613 }
614 }
615
616 if (i->value) {
617 /* Insert the new value. */
618 s->here->e_value_size = cpu_to_le32(i->value_len);
619 if (i->value_len) {
620 size_t size = EXT3_XATTR_SIZE(i->value_len);
621 void *val = s->base + min_offs - size;
622 s->here->e_value_offs = cpu_to_le16(min_offs - size);
623 memset(val + size - EXT3_XATTR_PAD, 0,
624 EXT3_XATTR_PAD); /* Clear the pad bytes. */
625 memcpy(val, i->value, i->value_len);
626 }
627 }
628 return 0;
629}
630
631struct ext3_xattr_block_find {
632 struct ext3_xattr_search s;
633 struct buffer_head *bh;
634};
635
636static int
637ext3_xattr_block_find(struct inode *inode, struct ext3_xattr_info *i,
638 struct ext3_xattr_block_find *bs)
639{
640 struct super_block *sb = inode->i_sb;
641 int error;
642
643 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
644 i->name_index, i->name, i->value, (long)i->value_len);
645
646 if (EXT3_I(inode)->i_file_acl) {
647 /* The inode already has an extended attribute block. */
648 bs->bh = sb_bread(sb, EXT3_I(inode)->i_file_acl);
649 error = -EIO;
650 if (!bs->bh)
651 goto cleanup;
652 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
653 atomic_read(&(bs->bh->b_count)),
654 le32_to_cpu(BHDR(bs->bh)->h_refcount));
655 if (ext3_xattr_check_block(bs->bh)) {
656 ext3_error(sb, __func__,
657 "inode %lu: bad block "E3FSBLK, inode->i_ino,
658 EXT3_I(inode)->i_file_acl);
659 error = -EIO;
660 goto cleanup;
661 }
662 /* Find the named attribute. */
663 bs->s.base = BHDR(bs->bh);
664 bs->s.first = BFIRST(bs->bh);
665 bs->s.end = bs->bh->b_data + bs->bh->b_size;
666 bs->s.here = bs->s.first;
667 error = ext3_xattr_find_entry(&bs->s.here, i->name_index,
668 i->name, bs->bh->b_size, 1);
669 if (error && error != -ENODATA)
670 goto cleanup;
671 bs->s.not_found = error;
672 }
673 error = 0;
674
675cleanup:
676 return error;
677}
678
679static int
680ext3_xattr_block_set(handle_t *handle, struct inode *inode,
681 struct ext3_xattr_info *i,
682 struct ext3_xattr_block_find *bs)
683{
684 struct super_block *sb = inode->i_sb;
685 struct buffer_head *new_bh = NULL;
686 struct ext3_xattr_search *s = &bs->s;
687 struct mb_cache_entry *ce = NULL;
688 int error = 0;
689
690#define header(x) ((struct ext3_xattr_header *)(x))
691
692 if (i->value && i->value_len > sb->s_blocksize)
693 return -ENOSPC;
694 if (s->base) {
695 ce = mb_cache_entry_get(ext3_xattr_cache, bs->bh->b_bdev,
696 bs->bh->b_blocknr);
697 error = ext3_journal_get_write_access(handle, bs->bh);
698 if (error)
699 goto cleanup;
700 lock_buffer(bs->bh);
701
702 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
703 if (ce) {
704 mb_cache_entry_free(ce);
705 ce = NULL;
706 }
707 ea_bdebug(bs->bh, "modifying in-place");
708 error = ext3_xattr_set_entry(i, s);
709 if (!error) {
710 if (!IS_LAST_ENTRY(s->first))
711 ext3_xattr_rehash(header(s->base),
712 s->here);
713 ext3_xattr_cache_insert(bs->bh);
714 }
715 unlock_buffer(bs->bh);
716 if (error == -EIO)
717 goto bad_block;
718 if (!error)
719 error = ext3_journal_dirty_metadata(handle,
720 bs->bh);
721 if (error)
722 goto cleanup;
723 goto inserted;
724 } else {
725 int offset = (char *)s->here - bs->bh->b_data;
726
727 unlock_buffer(bs->bh);
728 journal_release_buffer(handle, bs->bh);
729
730 if (ce) {
731 mb_cache_entry_release(ce);
732 ce = NULL;
733 }
734 ea_bdebug(bs->bh, "cloning");
735 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
736 error = -ENOMEM;
737 if (s->base == NULL)
738 goto cleanup;
739 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
740 s->first = ENTRY(header(s->base)+1);
741 header(s->base)->h_refcount = cpu_to_le32(1);
742 s->here = ENTRY(s->base + offset);
743 s->end = s->base + bs->bh->b_size;
744 }
745 } else {
746 /* Allocate a buffer where we construct the new block. */
747 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
748 /* assert(header == s->base) */
749 error = -ENOMEM;
750 if (s->base == NULL)
751 goto cleanup;
752 header(s->base)->h_magic = cpu_to_le32(EXT3_XATTR_MAGIC);
753 header(s->base)->h_blocks = cpu_to_le32(1);
754 header(s->base)->h_refcount = cpu_to_le32(1);
755 s->first = ENTRY(header(s->base)+1);
756 s->here = ENTRY(header(s->base)+1);
757 s->end = s->base + sb->s_blocksize;
758 }
759
760 error = ext3_xattr_set_entry(i, s);
761 if (error == -EIO)
762 goto bad_block;
763 if (error)
764 goto cleanup;
765 if (!IS_LAST_ENTRY(s->first))
766 ext3_xattr_rehash(header(s->base), s->here);
767
768inserted:
769 if (!IS_LAST_ENTRY(s->first)) {
770 new_bh = ext3_xattr_cache_find(inode, header(s->base), &ce);
771 if (new_bh) {
772 /* We found an identical block in the cache. */
773 if (new_bh == bs->bh)
774 ea_bdebug(new_bh, "keeping");
775 else {
776 /* The old block is released after updating
777 the inode. */
778 error = dquot_alloc_block(inode, 1);
779 if (error)
780 goto cleanup;
781 error = ext3_journal_get_write_access(handle,
782 new_bh);
783 if (error)
784 goto cleanup_dquot;
785 lock_buffer(new_bh);
786 le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
787 ea_bdebug(new_bh, "reusing; refcount now=%d",
788 le32_to_cpu(BHDR(new_bh)->h_refcount));
789 unlock_buffer(new_bh);
790 error = ext3_journal_dirty_metadata(handle,
791 new_bh);
792 if (error)
793 goto cleanup_dquot;
794 }
795 mb_cache_entry_release(ce);
796 ce = NULL;
797 } else if (bs->bh && s->base == bs->bh->b_data) {
798 /* We were modifying this block in-place. */
799 ea_bdebug(bs->bh, "keeping this block");
800 new_bh = bs->bh;
801 get_bh(new_bh);
802 } else {
803 /* We need to allocate a new block */
804 ext3_fsblk_t goal = ext3_group_first_block_no(sb,
805 EXT3_I(inode)->i_block_group);
806 ext3_fsblk_t block;
807
808 /*
809 * Protect us agaist concurrent allocations to the
810 * same inode from ext3_..._writepage(). Reservation
811 * code does not expect racing allocations.
812 */
813 mutex_lock(&EXT3_I(inode)->truncate_mutex);
814 block = ext3_new_block(handle, inode, goal, &error);
815 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
816 if (error)
817 goto cleanup;
818 ea_idebug(inode, "creating block %d", block);
819
820 new_bh = sb_getblk(sb, block);
821 if (!new_bh) {
822getblk_failed:
823 ext3_free_blocks(handle, inode, block, 1);
824 error = -EIO;
825 goto cleanup;
826 }
827 lock_buffer(new_bh);
828 error = ext3_journal_get_create_access(handle, new_bh);
829 if (error) {
830 unlock_buffer(new_bh);
831 goto getblk_failed;
832 }
833 memcpy(new_bh->b_data, s->base, new_bh->b_size);
834 set_buffer_uptodate(new_bh);
835 unlock_buffer(new_bh);
836 ext3_xattr_cache_insert(new_bh);
837 error = ext3_journal_dirty_metadata(handle, new_bh);
838 if (error)
839 goto cleanup;
840 }
841 }
842
843 /* Update the inode. */
844 EXT3_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
845
846 /* Drop the previous xattr block. */
847 if (bs->bh && bs->bh != new_bh)
848 ext3_xattr_release_block(handle, inode, bs->bh);
849 error = 0;
850
851cleanup:
852 if (ce)
853 mb_cache_entry_release(ce);
854 brelse(new_bh);
855 if (!(bs->bh && s->base == bs->bh->b_data))
856 kfree(s->base);
857
858 return error;
859
860cleanup_dquot:
861 dquot_free_block(inode, 1);
862 goto cleanup;
863
864bad_block:
865 ext3_error(inode->i_sb, __func__,
866 "inode %lu: bad block "E3FSBLK, inode->i_ino,
867 EXT3_I(inode)->i_file_acl);
868 goto cleanup;
869
870#undef header
871}
872
873struct ext3_xattr_ibody_find {
874 struct ext3_xattr_search s;
875 struct ext3_iloc iloc;
876};
877
878static int
879ext3_xattr_ibody_find(struct inode *inode, struct ext3_xattr_info *i,
880 struct ext3_xattr_ibody_find *is)
881{
882 struct ext3_xattr_ibody_header *header;
883 struct ext3_inode *raw_inode;
884 int error;
885
886 if (EXT3_I(inode)->i_extra_isize == 0)
887 return 0;
888 raw_inode = ext3_raw_inode(&is->iloc);
889 header = IHDR(inode, raw_inode);
890 is->s.base = is->s.first = IFIRST(header);
891 is->s.here = is->s.first;
892 is->s.end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
893 if (ext3_test_inode_state(inode, EXT3_STATE_XATTR)) {
894 error = ext3_xattr_check_names(IFIRST(header), is->s.end);
895 if (error)
896 return error;
897 /* Find the named attribute. */
898 error = ext3_xattr_find_entry(&is->s.here, i->name_index,
899 i->name, is->s.end -
900 (void *)is->s.base, 0);
901 if (error && error != -ENODATA)
902 return error;
903 is->s.not_found = error;
904 }
905 return 0;
906}
907
908static int
909ext3_xattr_ibody_set(handle_t *handle, struct inode *inode,
910 struct ext3_xattr_info *i,
911 struct ext3_xattr_ibody_find *is)
912{
913 struct ext3_xattr_ibody_header *header;
914 struct ext3_xattr_search *s = &is->s;
915 int error;
916
917 if (EXT3_I(inode)->i_extra_isize == 0)
918 return -ENOSPC;
919 error = ext3_xattr_set_entry(i, s);
920 if (error)
921 return error;
922 header = IHDR(inode, ext3_raw_inode(&is->iloc));
923 if (!IS_LAST_ENTRY(s->first)) {
924 header->h_magic = cpu_to_le32(EXT3_XATTR_MAGIC);
925 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
926 } else {
927 header->h_magic = cpu_to_le32(0);
928 ext3_clear_inode_state(inode, EXT3_STATE_XATTR);
929 }
930 return 0;
931}
932
933/*
934 * ext3_xattr_set_handle()
935 *
936 * Create, replace or remove an extended attribute for this inode. Value
937 * is NULL to remove an existing extended attribute, and non-NULL to
938 * either replace an existing extended attribute, or create a new extended
939 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
940 * specify that an extended attribute must exist and must not exist
941 * previous to the call, respectively.
942 *
943 * Returns 0, or a negative error number on failure.
944 */
945int
946ext3_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
947 const char *name, const void *value, size_t value_len,
948 int flags)
949{
950 struct ext3_xattr_info i = {
951 .name_index = name_index,
952 .name = name,
953 .value = value,
954 .value_len = value_len,
955
956 };
957 struct ext3_xattr_ibody_find is = {
958 .s = { .not_found = -ENODATA, },
959 };
960 struct ext3_xattr_block_find bs = {
961 .s = { .not_found = -ENODATA, },
962 };
963 int error;
964
965 if (!name)
966 return -EINVAL;
967 if (strlen(name) > 255)
968 return -ERANGE;
969 down_write(&EXT3_I(inode)->xattr_sem);
970 error = ext3_get_inode_loc(inode, &is.iloc);
971 if (error)
972 goto cleanup;
973
974 error = ext3_journal_get_write_access(handle, is.iloc.bh);
975 if (error)
976 goto cleanup;
977
978 if (ext3_test_inode_state(inode, EXT3_STATE_NEW)) {
979 struct ext3_inode *raw_inode = ext3_raw_inode(&is.iloc);
980 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
981 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
982 }
983
984 error = ext3_xattr_ibody_find(inode, &i, &is);
985 if (error)
986 goto cleanup;
987 if (is.s.not_found)
988 error = ext3_xattr_block_find(inode, &i, &bs);
989 if (error)
990 goto cleanup;
991 if (is.s.not_found && bs.s.not_found) {
992 error = -ENODATA;
993 if (flags & XATTR_REPLACE)
994 goto cleanup;
995 error = 0;
996 if (!value)
997 goto cleanup;
998 } else {
999 error = -EEXIST;
1000 if (flags & XATTR_CREATE)
1001 goto cleanup;
1002 }
1003 if (!value) {
1004 if (!is.s.not_found)
1005 error = ext3_xattr_ibody_set(handle, inode, &i, &is);
1006 else if (!bs.s.not_found)
1007 error = ext3_xattr_block_set(handle, inode, &i, &bs);
1008 } else {
1009 error = ext3_xattr_ibody_set(handle, inode, &i, &is);
1010 if (!error && !bs.s.not_found) {
1011 i.value = NULL;
1012 error = ext3_xattr_block_set(handle, inode, &i, &bs);
1013 } else if (error == -ENOSPC) {
1014 if (EXT3_I(inode)->i_file_acl && !bs.s.base) {
1015 error = ext3_xattr_block_find(inode, &i, &bs);
1016 if (error)
1017 goto cleanup;
1018 }
1019 error = ext3_xattr_block_set(handle, inode, &i, &bs);
1020 if (error)
1021 goto cleanup;
1022 if (!is.s.not_found) {
1023 i.value = NULL;
1024 error = ext3_xattr_ibody_set(handle, inode, &i,
1025 &is);
1026 }
1027 }
1028 }
1029 if (!error) {
1030 ext3_xattr_update_super_block(handle, inode->i_sb);
1031 inode->i_ctime = CURRENT_TIME_SEC;
1032 error = ext3_mark_iloc_dirty(handle, inode, &is.iloc);
1033 /*
1034 * The bh is consumed by ext3_mark_iloc_dirty, even with
1035 * error != 0.
1036 */
1037 is.iloc.bh = NULL;
1038 if (IS_SYNC(inode))
1039 handle->h_sync = 1;
1040 }
1041
1042cleanup:
1043 brelse(is.iloc.bh);
1044 brelse(bs.bh);
1045 up_write(&EXT3_I(inode)->xattr_sem);
1046 return error;
1047}
1048
1049/*
1050 * ext3_xattr_set()
1051 *
1052 * Like ext3_xattr_set_handle, but start from an inode. This extended
1053 * attribute modification is a filesystem transaction by itself.
1054 *
1055 * Returns 0, or a negative error number on failure.
1056 */
1057int
1058ext3_xattr_set(struct inode *inode, int name_index, const char *name,
1059 const void *value, size_t value_len, int flags)
1060{
1061 handle_t *handle;
1062 int error, retries = 0;
1063
1064retry:
1065 handle = ext3_journal_start(inode, EXT3_DATA_TRANS_BLOCKS(inode->i_sb));
1066 if (IS_ERR(handle)) {
1067 error = PTR_ERR(handle);
1068 } else {
1069 int error2;
1070
1071 error = ext3_xattr_set_handle(handle, inode, name_index, name,
1072 value, value_len, flags);
1073 error2 = ext3_journal_stop(handle);
1074 if (error == -ENOSPC &&
1075 ext3_should_retry_alloc(inode->i_sb, &retries))
1076 goto retry;
1077 if (error == 0)
1078 error = error2;
1079 }
1080
1081 return error;
1082}
1083
1084/*
1085 * ext3_xattr_delete_inode()
1086 *
1087 * Free extended attribute resources associated with this inode. This
1088 * is called immediately before an inode is freed. We have exclusive
1089 * access to the inode.
1090 */
1091void
1092ext3_xattr_delete_inode(handle_t *handle, struct inode *inode)
1093{
1094 struct buffer_head *bh = NULL;
1095
1096 if (!EXT3_I(inode)->i_file_acl)
1097 goto cleanup;
1098 bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
1099 if (!bh) {
1100 ext3_error(inode->i_sb, __func__,
1101 "inode %lu: block "E3FSBLK" read error", inode->i_ino,
1102 EXT3_I(inode)->i_file_acl);
1103 goto cleanup;
1104 }
1105 if (BHDR(bh)->h_magic != cpu_to_le32(EXT3_XATTR_MAGIC) ||
1106 BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1107 ext3_error(inode->i_sb, __func__,
1108 "inode %lu: bad block "E3FSBLK, inode->i_ino,
1109 EXT3_I(inode)->i_file_acl);
1110 goto cleanup;
1111 }
1112 ext3_xattr_release_block(handle, inode, bh);
1113 EXT3_I(inode)->i_file_acl = 0;
1114
1115cleanup:
1116 brelse(bh);
1117}
1118
1119/*
1120 * ext3_xattr_put_super()
1121 *
1122 * This is called when a file system is unmounted.
1123 */
1124void
1125ext3_xattr_put_super(struct super_block *sb)
1126{
1127 mb_cache_shrink(sb->s_bdev);
1128}
1129
1130/*
1131 * ext3_xattr_cache_insert()
1132 *
1133 * Create a new entry in the extended attribute cache, and insert
1134 * it unless such an entry is already in the cache.
1135 *
1136 * Returns 0, or a negative error number on failure.
1137 */
1138static void
1139ext3_xattr_cache_insert(struct buffer_head *bh)
1140{
1141 __u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1142 struct mb_cache_entry *ce;
1143 int error;
1144
1145 ce = mb_cache_entry_alloc(ext3_xattr_cache, GFP_NOFS);
1146 if (!ce) {
1147 ea_bdebug(bh, "out of memory");
1148 return;
1149 }
1150 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
1151 if (error) {
1152 mb_cache_entry_free(ce);
1153 if (error == -EBUSY) {
1154 ea_bdebug(bh, "already in cache");
1155 error = 0;
1156 }
1157 } else {
1158 ea_bdebug(bh, "inserting [%x]", (int)hash);
1159 mb_cache_entry_release(ce);
1160 }
1161}
1162
1163/*
1164 * ext3_xattr_cmp()
1165 *
1166 * Compare two extended attribute blocks for equality.
1167 *
1168 * Returns 0 if the blocks are equal, 1 if they differ, and
1169 * a negative error number on errors.
1170 */
1171static int
1172ext3_xattr_cmp(struct ext3_xattr_header *header1,
1173 struct ext3_xattr_header *header2)
1174{
1175 struct ext3_xattr_entry *entry1, *entry2;
1176
1177 entry1 = ENTRY(header1+1);
1178 entry2 = ENTRY(header2+1);
1179 while (!IS_LAST_ENTRY(entry1)) {
1180 if (IS_LAST_ENTRY(entry2))
1181 return 1;
1182 if (entry1->e_hash != entry2->e_hash ||
1183 entry1->e_name_index != entry2->e_name_index ||
1184 entry1->e_name_len != entry2->e_name_len ||
1185 entry1->e_value_size != entry2->e_value_size ||
1186 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1187 return 1;
1188 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1189 return -EIO;
1190 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1191 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1192 le32_to_cpu(entry1->e_value_size)))
1193 return 1;
1194
1195 entry1 = EXT3_XATTR_NEXT(entry1);
1196 entry2 = EXT3_XATTR_NEXT(entry2);
1197 }
1198 if (!IS_LAST_ENTRY(entry2))
1199 return 1;
1200 return 0;
1201}
1202
1203/*
1204 * ext3_xattr_cache_find()
1205 *
1206 * Find an identical extended attribute block.
1207 *
1208 * Returns a pointer to the block found, or NULL if such a block was
1209 * not found or an error occurred.
1210 */
1211static struct buffer_head *
1212ext3_xattr_cache_find(struct inode *inode, struct ext3_xattr_header *header,
1213 struct mb_cache_entry **pce)
1214{
1215 __u32 hash = le32_to_cpu(header->h_hash);
1216 struct mb_cache_entry *ce;
1217
1218 if (!header->h_hash)
1219 return NULL; /* never share */
1220 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1221again:
1222 ce = mb_cache_entry_find_first(ext3_xattr_cache, inode->i_sb->s_bdev,
1223 hash);
1224 while (ce) {
1225 struct buffer_head *bh;
1226
1227 if (IS_ERR(ce)) {
1228 if (PTR_ERR(ce) == -EAGAIN)
1229 goto again;
1230 break;
1231 }
1232 bh = sb_bread(inode->i_sb, ce->e_block);
1233 if (!bh) {
1234 ext3_error(inode->i_sb, __func__,
1235 "inode %lu: block %lu read error",
1236 inode->i_ino, (unsigned long) ce->e_block);
1237 } else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1238 EXT3_XATTR_REFCOUNT_MAX) {
1239 ea_idebug(inode, "block %lu refcount %d>=%d",
1240 (unsigned long) ce->e_block,
1241 le32_to_cpu(BHDR(bh)->h_refcount),
1242 EXT3_XATTR_REFCOUNT_MAX);
1243 } else if (ext3_xattr_cmp(header, BHDR(bh)) == 0) {
1244 *pce = ce;
1245 return bh;
1246 }
1247 brelse(bh);
1248 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
1249 }
1250 return NULL;
1251}
1252
1253#define NAME_HASH_SHIFT 5
1254#define VALUE_HASH_SHIFT 16
1255
1256/*
1257 * ext3_xattr_hash_entry()
1258 *
1259 * Compute the hash of an extended attribute.
1260 */
1261static inline void ext3_xattr_hash_entry(struct ext3_xattr_header *header,
1262 struct ext3_xattr_entry *entry)
1263{
1264 __u32 hash = 0;
1265 char *name = entry->e_name;
1266 int n;
1267
1268 for (n=0; n < entry->e_name_len; n++) {
1269 hash = (hash << NAME_HASH_SHIFT) ^
1270 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1271 *name++;
1272 }
1273
1274 if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1275 __le32 *value = (__le32 *)((char *)header +
1276 le16_to_cpu(entry->e_value_offs));
1277 for (n = (le32_to_cpu(entry->e_value_size) +
1278 EXT3_XATTR_ROUND) >> EXT3_XATTR_PAD_BITS; n; n--) {
1279 hash = (hash << VALUE_HASH_SHIFT) ^
1280 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1281 le32_to_cpu(*value++);
1282 }
1283 }
1284 entry->e_hash = cpu_to_le32(hash);
1285}
1286
1287#undef NAME_HASH_SHIFT
1288#undef VALUE_HASH_SHIFT
1289
1290#define BLOCK_HASH_SHIFT 16
1291
1292/*
1293 * ext3_xattr_rehash()
1294 *
1295 * Re-compute the extended attribute hash value after an entry has changed.
1296 */
1297static void ext3_xattr_rehash(struct ext3_xattr_header *header,
1298 struct ext3_xattr_entry *entry)
1299{
1300 struct ext3_xattr_entry *here;
1301 __u32 hash = 0;
1302
1303 ext3_xattr_hash_entry(header, entry);
1304 here = ENTRY(header+1);
1305 while (!IS_LAST_ENTRY(here)) {
1306 if (!here->e_hash) {
1307 /* Block is not shared if an entry's hash value == 0 */
1308 hash = 0;
1309 break;
1310 }
1311 hash = (hash << BLOCK_HASH_SHIFT) ^
1312 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1313 le32_to_cpu(here->e_hash);
1314 here = EXT3_XATTR_NEXT(here);
1315 }
1316 header->h_hash = cpu_to_le32(hash);
1317}
1318
1319#undef BLOCK_HASH_SHIFT
1320
1321int __init
1322init_ext3_xattr(void)
1323{
1324 ext3_xattr_cache = mb_cache_create("ext3_xattr", 6);
1325 if (!ext3_xattr_cache)
1326 return -ENOMEM;
1327 return 0;
1328}
1329
1330void
1331exit_ext3_xattr(void)
1332{
1333 if (ext3_xattr_cache)
1334 mb_cache_destroy(ext3_xattr_cache);
1335 ext3_xattr_cache = NULL;
1336}