2 * linux/fs/ext2/xattr.c
4 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
6 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7 * Extended attributes for symlinks and special files added per
8 * suggestion of Luka Renko <luka.renko@hermes.si>.
9 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
15 * Extended attributes are stored on disk blocks allocated outside of
16 * any inode. The i_file_acl field is then made to point to this allocated
17 * block. If all extended attributes of an inode are identical, these
18 * inodes may share the same extended attribute block. Such situations
19 * are automatically detected by keeping a cache of recent attribute block
20 * numbers and hashes over the block's contents in memory.
23 * Extended attribute block layout:
25 * +------------------+
28 * | entry 2 | | growing downwards
33 * | value 3 | | growing upwards
35 * +------------------+
37 * The block header is followed by multiple entry descriptors. These entry
38 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
39 * byte boundaries. The entry descriptors are sorted by attribute name,
40 * so that two extended attribute blocks can be compared efficiently.
42 * Attribute values are aligned to the end of the block, stored in
43 * no specific order. They are also padded to EXT2_XATTR_PAD byte
44 * boundaries. No additional gaps are left between them.
48 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
49 * EA blocks are only changed if they are exclusive to an inode, so
50 * holding xattr_sem also means that nothing but the EA block's reference
51 * count will change. Multiple writers to an EA block are synchronized
52 * by the bh lock. No more than a single bh lock is held at any time
56 #include <linux/buffer_head.h>
57 #include <linux/init.h>
58 #include <linux/slab.h>
59 #include <linux/mbcache.h>
60 #include <linux/quotaops.h>
61 #include <linux/rwsem.h>
62 #include <linux/security.h>
67 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
68 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
69 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
70 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
72 #ifdef EXT2_XATTR_DEBUG
73 # define ea_idebug(inode, f...) do { \
74 printk(KERN_DEBUG "inode %s:%ld: ", \
75 inode->i_sb->s_id, inode->i_ino); \
79 # define ea_bdebug(bh, f...) do { \
80 printk(KERN_DEBUG "block %pg:%lu: ", \
81 bh->b_bdev, (unsigned long) bh->b_blocknr); \
86 # define ea_idebug(f...)
87 # define ea_bdebug(f...)
90 static int ext2_xattr_set2(struct inode
*, struct buffer_head
*,
91 struct ext2_xattr_header
*);
93 static int ext2_xattr_cache_insert(struct mb_cache
*, struct buffer_head
*);
94 static struct buffer_head
*ext2_xattr_cache_find(struct inode
*,
95 struct ext2_xattr_header
*);
96 static void ext2_xattr_rehash(struct ext2_xattr_header
*,
97 struct ext2_xattr_entry
*);
99 static const struct xattr_handler
*ext2_xattr_handler_map
[] = {
100 [EXT2_XATTR_INDEX_USER
] = &ext2_xattr_user_handler
,
101 #ifdef CONFIG_EXT2_FS_POSIX_ACL
102 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS
] = &posix_acl_access_xattr_handler
,
103 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT
] = &posix_acl_default_xattr_handler
,
105 [EXT2_XATTR_INDEX_TRUSTED
] = &ext2_xattr_trusted_handler
,
106 #ifdef CONFIG_EXT2_FS_SECURITY
107 [EXT2_XATTR_INDEX_SECURITY
] = &ext2_xattr_security_handler
,
111 const struct xattr_handler
*ext2_xattr_handlers
[] = {
112 &ext2_xattr_user_handler
,
113 &ext2_xattr_trusted_handler
,
114 #ifdef CONFIG_EXT2_FS_POSIX_ACL
115 &posix_acl_access_xattr_handler
,
116 &posix_acl_default_xattr_handler
,
118 #ifdef CONFIG_EXT2_FS_SECURITY
119 &ext2_xattr_security_handler
,
124 static inline const struct xattr_handler
*
125 ext2_xattr_handler(int name_index
)
127 const struct xattr_handler
*handler
= NULL
;
129 if (name_index
> 0 && name_index
< ARRAY_SIZE(ext2_xattr_handler_map
))
130 handler
= ext2_xattr_handler_map
[name_index
];
137 * Copy an extended attribute into the buffer
138 * provided, or compute the buffer size required.
139 * Buffer is NULL to compute the size of the buffer required.
141 * Returns a negative error number on failure, or the number of bytes
142 * used / required on success.
145 ext2_xattr_get(struct inode
*inode
, int name_index
, const char *name
,
146 void *buffer
, size_t buffer_size
)
148 struct buffer_head
*bh
= NULL
;
149 struct ext2_xattr_entry
*entry
;
150 size_t name_len
, size
;
153 struct mb_cache
*ext2_mb_cache
= EXT2_SB(inode
->i_sb
)->s_mb_cache
;
155 ea_idebug(inode
, "name=%d.%s, buffer=%p, buffer_size=%ld",
156 name_index
, name
, buffer
, (long)buffer_size
);
160 name_len
= strlen(name
);
164 down_read(&EXT2_I(inode
)->xattr_sem
);
166 if (!EXT2_I(inode
)->i_file_acl
)
168 ea_idebug(inode
, "reading block %d", EXT2_I(inode
)->i_file_acl
);
169 bh
= sb_bread(inode
->i_sb
, EXT2_I(inode
)->i_file_acl
);
173 ea_bdebug(bh
, "b_count=%d, refcount=%d",
174 atomic_read(&(bh
->b_count
)), le32_to_cpu(HDR(bh
)->h_refcount
));
175 end
= bh
->b_data
+ bh
->b_size
;
176 if (HDR(bh
)->h_magic
!= cpu_to_le32(EXT2_XATTR_MAGIC
) ||
177 HDR(bh
)->h_blocks
!= cpu_to_le32(1)) {
178 bad_block
: ext2_error(inode
->i_sb
, "ext2_xattr_get",
179 "inode %ld: bad block %d", inode
->i_ino
,
180 EXT2_I(inode
)->i_file_acl
);
185 /* find named attribute */
186 entry
= FIRST_ENTRY(bh
);
187 while (!IS_LAST_ENTRY(entry
)) {
188 struct ext2_xattr_entry
*next
=
189 EXT2_XATTR_NEXT(entry
);
190 if ((char *)next
>= end
)
192 if (name_index
== entry
->e_name_index
&&
193 name_len
== entry
->e_name_len
&&
194 memcmp(name
, entry
->e_name
, name_len
) == 0)
198 if (ext2_xattr_cache_insert(ext2_mb_cache
, bh
))
199 ea_idebug(inode
, "cache insert failed");
203 /* check the buffer size */
204 if (entry
->e_value_block
!= 0)
206 size
= le32_to_cpu(entry
->e_value_size
);
207 if (size
> inode
->i_sb
->s_blocksize
||
208 le16_to_cpu(entry
->e_value_offs
) + size
> inode
->i_sb
->s_blocksize
)
211 if (ext2_xattr_cache_insert(ext2_mb_cache
, bh
))
212 ea_idebug(inode
, "cache insert failed");
215 if (size
> buffer_size
)
217 /* return value of attribute */
218 memcpy(buffer
, bh
->b_data
+ le16_to_cpu(entry
->e_value_offs
),
225 up_read(&EXT2_I(inode
)->xattr_sem
);
233 * Copy a list of attribute names into the buffer
234 * provided, or compute the buffer size required.
235 * Buffer is NULL to compute the size of the buffer required.
237 * Returns a negative error number on failure, or the number of bytes
238 * used / required on success.
241 ext2_xattr_list(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
243 struct inode
*inode
= d_inode(dentry
);
244 struct buffer_head
*bh
= NULL
;
245 struct ext2_xattr_entry
*entry
;
247 size_t rest
= buffer_size
;
249 struct mb_cache
*ext2_mb_cache
= EXT2_SB(inode
->i_sb
)->s_mb_cache
;
251 ea_idebug(inode
, "buffer=%p, buffer_size=%ld",
252 buffer
, (long)buffer_size
);
254 down_read(&EXT2_I(inode
)->xattr_sem
);
256 if (!EXT2_I(inode
)->i_file_acl
)
258 ea_idebug(inode
, "reading block %d", EXT2_I(inode
)->i_file_acl
);
259 bh
= sb_bread(inode
->i_sb
, EXT2_I(inode
)->i_file_acl
);
263 ea_bdebug(bh
, "b_count=%d, refcount=%d",
264 atomic_read(&(bh
->b_count
)), le32_to_cpu(HDR(bh
)->h_refcount
));
265 end
= bh
->b_data
+ bh
->b_size
;
266 if (HDR(bh
)->h_magic
!= cpu_to_le32(EXT2_XATTR_MAGIC
) ||
267 HDR(bh
)->h_blocks
!= cpu_to_le32(1)) {
268 bad_block
: ext2_error(inode
->i_sb
, "ext2_xattr_list",
269 "inode %ld: bad block %d", inode
->i_ino
,
270 EXT2_I(inode
)->i_file_acl
);
275 /* check the on-disk data structure */
276 entry
= FIRST_ENTRY(bh
);
277 while (!IS_LAST_ENTRY(entry
)) {
278 struct ext2_xattr_entry
*next
= EXT2_XATTR_NEXT(entry
);
280 if ((char *)next
>= end
)
284 if (ext2_xattr_cache_insert(ext2_mb_cache
, bh
))
285 ea_idebug(inode
, "cache insert failed");
287 /* list the attribute names */
288 for (entry
= FIRST_ENTRY(bh
); !IS_LAST_ENTRY(entry
);
289 entry
= EXT2_XATTR_NEXT(entry
)) {
290 const struct xattr_handler
*handler
=
291 ext2_xattr_handler(entry
->e_name_index
);
293 if (handler
&& (!handler
->list
|| handler
->list(dentry
))) {
294 const char *prefix
= handler
->prefix
?: handler
->name
;
295 size_t prefix_len
= strlen(prefix
);
296 size_t size
= prefix_len
+ entry
->e_name_len
+ 1;
303 memcpy(buffer
, prefix
, prefix_len
);
304 buffer
+= prefix_len
;
305 memcpy(buffer
, entry
->e_name
, entry
->e_name_len
);
306 buffer
+= entry
->e_name_len
;
312 error
= buffer_size
- rest
; /* total size */
316 up_read(&EXT2_I(inode
)->xattr_sem
);
322 * Inode operation listxattr()
324 * d_inode(dentry)->i_mutex: don't care
327 ext2_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
329 return ext2_xattr_list(dentry
, buffer
, size
);
333 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
336 static void ext2_xattr_update_super_block(struct super_block
*sb
)
338 if (EXT2_HAS_COMPAT_FEATURE(sb
, EXT2_FEATURE_COMPAT_EXT_ATTR
))
341 spin_lock(&EXT2_SB(sb
)->s_lock
);
342 EXT2_SET_COMPAT_FEATURE(sb
, EXT2_FEATURE_COMPAT_EXT_ATTR
);
343 spin_unlock(&EXT2_SB(sb
)->s_lock
);
344 mark_buffer_dirty(EXT2_SB(sb
)->s_sbh
);
350 * Create, replace or remove an extended attribute for this inode. Value
351 * is NULL to remove an existing extended attribute, and non-NULL to
352 * either replace an existing extended attribute, or create a new extended
353 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
354 * specify that an extended attribute must exist and must not exist
355 * previous to the call, respectively.
357 * Returns 0, or a negative error number on failure.
360 ext2_xattr_set(struct inode
*inode
, int name_index
, const char *name
,
361 const void *value
, size_t value_len
, int flags
)
363 struct super_block
*sb
= inode
->i_sb
;
364 struct buffer_head
*bh
= NULL
;
365 struct ext2_xattr_header
*header
= NULL
;
366 struct ext2_xattr_entry
*here
, *last
;
367 size_t name_len
, free
, min_offs
= sb
->s_blocksize
;
368 int not_found
= 1, error
;
372 * header -- Points either into bh, or to a temporarily
374 * here -- The named entry found, or the place for inserting, within
375 * the block pointed to by header.
376 * last -- Points right after the last named entry within the block
377 * pointed to by header.
378 * min_offs -- The offset of the first value (values are aligned
379 * towards the end of the block).
380 * end -- Points right after the block pointed to by header.
383 ea_idebug(inode
, "name=%d.%s, value=%p, value_len=%ld",
384 name_index
, name
, value
, (long)value_len
);
390 name_len
= strlen(name
);
391 if (name_len
> 255 || value_len
> sb
->s_blocksize
)
393 down_write(&EXT2_I(inode
)->xattr_sem
);
394 if (EXT2_I(inode
)->i_file_acl
) {
395 /* The inode already has an extended attribute block. */
396 bh
= sb_bread(sb
, EXT2_I(inode
)->i_file_acl
);
400 ea_bdebug(bh
, "b_count=%d, refcount=%d",
401 atomic_read(&(bh
->b_count
)),
402 le32_to_cpu(HDR(bh
)->h_refcount
));
404 end
= bh
->b_data
+ bh
->b_size
;
405 if (header
->h_magic
!= cpu_to_le32(EXT2_XATTR_MAGIC
) ||
406 header
->h_blocks
!= cpu_to_le32(1)) {
407 bad_block
: ext2_error(sb
, "ext2_xattr_set",
408 "inode %ld: bad block %d", inode
->i_ino
,
409 EXT2_I(inode
)->i_file_acl
);
413 /* Find the named attribute. */
414 here
= FIRST_ENTRY(bh
);
415 while (!IS_LAST_ENTRY(here
)) {
416 struct ext2_xattr_entry
*next
= EXT2_XATTR_NEXT(here
);
417 if ((char *)next
>= end
)
419 if (!here
->e_value_block
&& here
->e_value_size
) {
420 size_t offs
= le16_to_cpu(here
->e_value_offs
);
424 not_found
= name_index
- here
->e_name_index
;
426 not_found
= name_len
- here
->e_name_len
;
428 not_found
= memcmp(name
, here
->e_name
,name_len
);
434 /* We still need to compute min_offs and last. */
435 while (!IS_LAST_ENTRY(last
)) {
436 struct ext2_xattr_entry
*next
= EXT2_XATTR_NEXT(last
);
437 if ((char *)next
>= end
)
439 if (!last
->e_value_block
&& last
->e_value_size
) {
440 size_t offs
= le16_to_cpu(last
->e_value_offs
);
447 /* Check whether we have enough space left. */
448 free
= min_offs
- ((char*)last
- (char*)header
) - sizeof(__u32
);
450 /* We will use a new extended attribute block. */
451 free
= sb
->s_blocksize
-
452 sizeof(struct ext2_xattr_header
) - sizeof(__u32
);
453 here
= last
= NULL
; /* avoid gcc uninitialized warning. */
457 /* Request to remove a nonexistent attribute? */
459 if (flags
& XATTR_REPLACE
)
465 /* Request to create an existing attribute? */
467 if (flags
& XATTR_CREATE
)
469 if (!here
->e_value_block
&& here
->e_value_size
) {
470 size_t size
= le32_to_cpu(here
->e_value_size
);
472 if (le16_to_cpu(here
->e_value_offs
) + size
>
473 sb
->s_blocksize
|| size
> sb
->s_blocksize
)
475 free
+= EXT2_XATTR_SIZE(size
);
477 free
+= EXT2_XATTR_LEN(name_len
);
480 if (free
< EXT2_XATTR_LEN(name_len
) + EXT2_XATTR_SIZE(value_len
))
483 /* Here we know that we can set the new attribute. */
486 /* assert(header == HDR(bh)); */
488 if (header
->h_refcount
== cpu_to_le32(1)) {
489 __u32 hash
= le32_to_cpu(header
->h_hash
);
491 ea_bdebug(bh
, "modifying in-place");
493 * This must happen under buffer lock for
494 * ext2_xattr_set2() to reliably detect modified block
496 mb_cache_entry_delete_block(EXT2_SB(sb
)->s_mb_cache
,
497 hash
, bh
->b_blocknr
);
499 /* keep the buffer locked while modifying it. */
504 ea_bdebug(bh
, "cloning");
505 header
= kmalloc(bh
->b_size
, GFP_KERNEL
);
509 memcpy(header
, HDR(bh
), bh
->b_size
);
510 header
->h_refcount
= cpu_to_le32(1);
512 offset
= (char *)here
- bh
->b_data
;
513 here
= ENTRY((char *)header
+ offset
);
514 offset
= (char *)last
- bh
->b_data
;
515 last
= ENTRY((char *)header
+ offset
);
518 /* Allocate a buffer where we construct the new block. */
519 header
= kzalloc(sb
->s_blocksize
, GFP_KERNEL
);
523 end
= (char *)header
+ sb
->s_blocksize
;
524 header
->h_magic
= cpu_to_le32(EXT2_XATTR_MAGIC
);
525 header
->h_blocks
= header
->h_refcount
= cpu_to_le32(1);
526 last
= here
= ENTRY(header
+1);
529 /* Iff we are modifying the block in-place, bh is locked here. */
532 /* Insert the new name. */
533 size_t size
= EXT2_XATTR_LEN(name_len
);
534 size_t rest
= (char *)last
- (char *)here
;
535 memmove((char *)here
+ size
, here
, rest
);
536 memset(here
, 0, size
);
537 here
->e_name_index
= name_index
;
538 here
->e_name_len
= name_len
;
539 memcpy(here
->e_name
, name
, name_len
);
541 if (!here
->e_value_block
&& here
->e_value_size
) {
542 char *first_val
= (char *)header
+ min_offs
;
543 size_t offs
= le16_to_cpu(here
->e_value_offs
);
544 char *val
= (char *)header
+ offs
;
545 size_t size
= EXT2_XATTR_SIZE(
546 le32_to_cpu(here
->e_value_size
));
548 if (size
== EXT2_XATTR_SIZE(value_len
)) {
549 /* The old and the new value have the same
550 size. Just replace. */
551 here
->e_value_size
= cpu_to_le32(value_len
);
552 memset(val
+ size
- EXT2_XATTR_PAD
, 0,
553 EXT2_XATTR_PAD
); /* Clear pad bytes. */
554 memcpy(val
, value
, value_len
);
558 /* Remove the old value. */
559 memmove(first_val
+ size
, first_val
, val
- first_val
);
560 memset(first_val
, 0, size
);
561 here
->e_value_offs
= 0;
564 /* Adjust all value offsets. */
565 last
= ENTRY(header
+1);
566 while (!IS_LAST_ENTRY(last
)) {
567 size_t o
= le16_to_cpu(last
->e_value_offs
);
568 if (!last
->e_value_block
&& o
< offs
)
570 cpu_to_le16(o
+ size
);
571 last
= EXT2_XATTR_NEXT(last
);
575 /* Remove the old name. */
576 size_t size
= EXT2_XATTR_LEN(name_len
);
577 last
= ENTRY((char *)last
- size
);
578 memmove(here
, (char*)here
+ size
,
579 (char*)last
- (char*)here
);
580 memset(last
, 0, size
);
585 /* Insert the new value. */
586 here
->e_value_size
= cpu_to_le32(value_len
);
588 size_t size
= EXT2_XATTR_SIZE(value_len
);
589 char *val
= (char *)header
+ min_offs
- size
;
591 cpu_to_le16((char *)val
- (char *)header
);
592 memset(val
+ size
- EXT2_XATTR_PAD
, 0,
593 EXT2_XATTR_PAD
); /* Clear the pad bytes. */
594 memcpy(val
, value
, value_len
);
599 if (IS_LAST_ENTRY(ENTRY(header
+1))) {
600 /* This block is now empty. */
601 if (bh
&& header
== HDR(bh
))
602 unlock_buffer(bh
); /* we were modifying in-place. */
603 error
= ext2_xattr_set2(inode
, bh
, NULL
);
605 ext2_xattr_rehash(header
, here
);
606 if (bh
&& header
== HDR(bh
))
607 unlock_buffer(bh
); /* we were modifying in-place. */
608 error
= ext2_xattr_set2(inode
, bh
, header
);
613 if (!(bh
&& header
== HDR(bh
)))
615 up_write(&EXT2_I(inode
)->xattr_sem
);
621 * Second half of ext2_xattr_set(): Update the file system.
624 ext2_xattr_set2(struct inode
*inode
, struct buffer_head
*old_bh
,
625 struct ext2_xattr_header
*header
)
627 struct super_block
*sb
= inode
->i_sb
;
628 struct buffer_head
*new_bh
= NULL
;
630 struct mb_cache
*ext2_mb_cache
= EXT2_SB(sb
)->s_mb_cache
;
633 new_bh
= ext2_xattr_cache_find(inode
, header
);
635 /* We found an identical block in the cache. */
636 if (new_bh
== old_bh
) {
637 ea_bdebug(new_bh
, "keeping this block");
639 /* The old block is released after updating
641 ea_bdebug(new_bh
, "reusing block");
643 error
= dquot_alloc_block(inode
, 1);
645 unlock_buffer(new_bh
);
648 le32_add_cpu(&HDR(new_bh
)->h_refcount
, 1);
649 ea_bdebug(new_bh
, "refcount now=%d",
650 le32_to_cpu(HDR(new_bh
)->h_refcount
));
652 unlock_buffer(new_bh
);
653 } else if (old_bh
&& header
== HDR(old_bh
)) {
654 /* Keep this block. No need to lock the block as we
655 don't need to change the reference count. */
658 ext2_xattr_cache_insert(ext2_mb_cache
, new_bh
);
660 /* We need to allocate a new block */
661 ext2_fsblk_t goal
= ext2_group_first_block_no(sb
,
662 EXT2_I(inode
)->i_block_group
);
663 int block
= ext2_new_block(inode
, goal
, &error
);
666 ea_idebug(inode
, "creating block %d", block
);
668 new_bh
= sb_getblk(sb
, block
);
669 if (unlikely(!new_bh
)) {
670 ext2_free_blocks(inode
, block
, 1);
671 mark_inode_dirty(inode
);
676 memcpy(new_bh
->b_data
, header
, new_bh
->b_size
);
677 set_buffer_uptodate(new_bh
);
678 unlock_buffer(new_bh
);
679 ext2_xattr_cache_insert(ext2_mb_cache
, new_bh
);
681 ext2_xattr_update_super_block(sb
);
683 mark_buffer_dirty(new_bh
);
684 if (IS_SYNC(inode
)) {
685 sync_dirty_buffer(new_bh
);
687 if (buffer_req(new_bh
) && !buffer_uptodate(new_bh
))
692 /* Update the inode. */
693 EXT2_I(inode
)->i_file_acl
= new_bh
? new_bh
->b_blocknr
: 0;
694 inode
->i_ctime
= CURRENT_TIME_SEC
;
695 if (IS_SYNC(inode
)) {
696 error
= sync_inode_metadata(inode
, 1);
697 /* In case sync failed due to ENOSPC the inode was actually
698 * written (only some dirty data were not) so we just proceed
699 * as if nothing happened and cleanup the unused block */
700 if (error
&& error
!= -ENOSPC
) {
701 if (new_bh
&& new_bh
!= old_bh
) {
702 dquot_free_block_nodirty(inode
, 1);
703 mark_inode_dirty(inode
);
708 mark_inode_dirty(inode
);
711 if (old_bh
&& old_bh
!= new_bh
) {
713 * If there was an old block and we are no longer using it,
714 * release the old block.
717 if (HDR(old_bh
)->h_refcount
== cpu_to_le32(1)) {
718 __u32 hash
= le32_to_cpu(HDR(old_bh
)->h_hash
);
721 * This must happen under buffer lock for
722 * ext2_xattr_set2() to reliably detect freed block
724 mb_cache_entry_delete_block(ext2_mb_cache
,
725 hash
, old_bh
->b_blocknr
);
726 /* Free the old block. */
727 ea_bdebug(old_bh
, "freeing");
728 ext2_free_blocks(inode
, old_bh
->b_blocknr
, 1);
729 mark_inode_dirty(inode
);
730 /* We let our caller release old_bh, so we
731 * need to duplicate the buffer before. */
735 /* Decrement the refcount only. */
736 le32_add_cpu(&HDR(old_bh
)->h_refcount
, -1);
737 dquot_free_block_nodirty(inode
, 1);
738 mark_inode_dirty(inode
);
739 mark_buffer_dirty(old_bh
);
740 ea_bdebug(old_bh
, "refcount now=%d",
741 le32_to_cpu(HDR(old_bh
)->h_refcount
));
743 unlock_buffer(old_bh
);
753 * ext2_xattr_delete_inode()
755 * Free extended attribute resources associated with this inode. This
756 * is called immediately before an inode is freed.
759 ext2_xattr_delete_inode(struct inode
*inode
)
761 struct buffer_head
*bh
= NULL
;
763 down_write(&EXT2_I(inode
)->xattr_sem
);
764 if (!EXT2_I(inode
)->i_file_acl
)
766 bh
= sb_bread(inode
->i_sb
, EXT2_I(inode
)->i_file_acl
);
768 ext2_error(inode
->i_sb
, "ext2_xattr_delete_inode",
769 "inode %ld: block %d read error", inode
->i_ino
,
770 EXT2_I(inode
)->i_file_acl
);
773 ea_bdebug(bh
, "b_count=%d", atomic_read(&(bh
->b_count
)));
774 if (HDR(bh
)->h_magic
!= cpu_to_le32(EXT2_XATTR_MAGIC
) ||
775 HDR(bh
)->h_blocks
!= cpu_to_le32(1)) {
776 ext2_error(inode
->i_sb
, "ext2_xattr_delete_inode",
777 "inode %ld: bad block %d", inode
->i_ino
,
778 EXT2_I(inode
)->i_file_acl
);
782 if (HDR(bh
)->h_refcount
== cpu_to_le32(1)) {
783 __u32 hash
= le32_to_cpu(HDR(bh
)->h_hash
);
786 * This must happen under buffer lock for ext2_xattr_set2() to
787 * reliably detect freed block
789 mb_cache_entry_delete_block(EXT2_SB(inode
->i_sb
)->s_mb_cache
,
790 hash
, bh
->b_blocknr
);
791 ext2_free_blocks(inode
, EXT2_I(inode
)->i_file_acl
, 1);
796 le32_add_cpu(&HDR(bh
)->h_refcount
, -1);
797 ea_bdebug(bh
, "refcount now=%d",
798 le32_to_cpu(HDR(bh
)->h_refcount
));
800 mark_buffer_dirty(bh
);
802 sync_dirty_buffer(bh
);
803 dquot_free_block_nodirty(inode
, 1);
805 EXT2_I(inode
)->i_file_acl
= 0;
809 up_write(&EXT2_I(inode
)->xattr_sem
);
813 * ext2_xattr_cache_insert()
815 * Create a new entry in the extended attribute cache, and insert
816 * it unless such an entry is already in the cache.
818 * Returns 0, or a negative error number on failure.
821 ext2_xattr_cache_insert(struct mb_cache
*cache
, struct buffer_head
*bh
)
823 __u32 hash
= le32_to_cpu(HDR(bh
)->h_hash
);
826 error
= mb_cache_entry_create(cache
, GFP_NOFS
, hash
, bh
->b_blocknr
, 1);
828 if (error
== -EBUSY
) {
829 ea_bdebug(bh
, "already in cache (%d cache entries)",
830 atomic_read(&ext2_xattr_cache
->c_entry_count
));
834 ea_bdebug(bh
, "inserting [%x]", (int)hash
);
841 * Compare two extended attribute blocks for equality.
843 * Returns 0 if the blocks are equal, 1 if they differ, and
844 * a negative error number on errors.
847 ext2_xattr_cmp(struct ext2_xattr_header
*header1
,
848 struct ext2_xattr_header
*header2
)
850 struct ext2_xattr_entry
*entry1
, *entry2
;
852 entry1
= ENTRY(header1
+1);
853 entry2
= ENTRY(header2
+1);
854 while (!IS_LAST_ENTRY(entry1
)) {
855 if (IS_LAST_ENTRY(entry2
))
857 if (entry1
->e_hash
!= entry2
->e_hash
||
858 entry1
->e_name_index
!= entry2
->e_name_index
||
859 entry1
->e_name_len
!= entry2
->e_name_len
||
860 entry1
->e_value_size
!= entry2
->e_value_size
||
861 memcmp(entry1
->e_name
, entry2
->e_name
, entry1
->e_name_len
))
863 if (entry1
->e_value_block
!= 0 || entry2
->e_value_block
!= 0)
865 if (memcmp((char *)header1
+ le16_to_cpu(entry1
->e_value_offs
),
866 (char *)header2
+ le16_to_cpu(entry2
->e_value_offs
),
867 le32_to_cpu(entry1
->e_value_size
)))
870 entry1
= EXT2_XATTR_NEXT(entry1
);
871 entry2
= EXT2_XATTR_NEXT(entry2
);
873 if (!IS_LAST_ENTRY(entry2
))
879 * ext2_xattr_cache_find()
881 * Find an identical extended attribute block.
883 * Returns a locked buffer head to the block found, or NULL if such
884 * a block was not found or an error occurred.
886 static struct buffer_head
*
887 ext2_xattr_cache_find(struct inode
*inode
, struct ext2_xattr_header
*header
)
889 __u32 hash
= le32_to_cpu(header
->h_hash
);
890 struct mb_cache_entry
*ce
;
891 struct mb_cache
*ext2_mb_cache
= EXT2_SB(inode
->i_sb
)->s_mb_cache
;
894 return NULL
; /* never share */
895 ea_idebug(inode
, "looking for cached blocks [%x]", (int)hash
);
897 ce
= mb_cache_entry_find_first(ext2_mb_cache
, hash
);
899 struct buffer_head
*bh
;
901 bh
= sb_bread(inode
->i_sb
, ce
->e_block
);
903 ext2_error(inode
->i_sb
, "ext2_xattr_cache_find",
904 "inode %ld: block %ld read error",
905 inode
->i_ino
, (unsigned long) ce
->e_block
);
909 * We have to be careful about races with freeing or
910 * rehashing of xattr block. Once we hold buffer lock
911 * xattr block's state is stable so we can check
912 * whether the block got freed / rehashed or not.
913 * Since we unhash mbcache entry under buffer lock when
914 * freeing / rehashing xattr block, checking whether
915 * entry is still hashed is reliable.
917 if (hlist_bl_unhashed(&ce
->e_hash_list
)) {
918 mb_cache_entry_put(ext2_mb_cache
, ce
);
922 } else if (le32_to_cpu(HDR(bh
)->h_refcount
) >
923 EXT2_XATTR_REFCOUNT_MAX
) {
924 ea_idebug(inode
, "block %ld refcount %d>%d",
925 (unsigned long) ce
->e_block
,
926 le32_to_cpu(HDR(bh
)->h_refcount
),
927 EXT2_XATTR_REFCOUNT_MAX
);
928 } else if (!ext2_xattr_cmp(header
, HDR(bh
))) {
929 ea_bdebug(bh
, "b_count=%d",
930 atomic_read(&(bh
->b_count
)));
931 mb_cache_entry_touch(ext2_mb_cache
, ce
);
932 mb_cache_entry_put(ext2_mb_cache
, ce
);
938 ce
= mb_cache_entry_find_next(ext2_mb_cache
, ce
);
943 #define NAME_HASH_SHIFT 5
944 #define VALUE_HASH_SHIFT 16
947 * ext2_xattr_hash_entry()
949 * Compute the hash of an extended attribute.
951 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header
*header
,
952 struct ext2_xattr_entry
*entry
)
955 char *name
= entry
->e_name
;
958 for (n
=0; n
< entry
->e_name_len
; n
++) {
959 hash
= (hash
<< NAME_HASH_SHIFT
) ^
960 (hash
>> (8*sizeof(hash
) - NAME_HASH_SHIFT
)) ^
964 if (entry
->e_value_block
== 0 && entry
->e_value_size
!= 0) {
965 __le32
*value
= (__le32
*)((char *)header
+
966 le16_to_cpu(entry
->e_value_offs
));
967 for (n
= (le32_to_cpu(entry
->e_value_size
) +
968 EXT2_XATTR_ROUND
) >> EXT2_XATTR_PAD_BITS
; n
; n
--) {
969 hash
= (hash
<< VALUE_HASH_SHIFT
) ^
970 (hash
>> (8*sizeof(hash
) - VALUE_HASH_SHIFT
)) ^
971 le32_to_cpu(*value
++);
974 entry
->e_hash
= cpu_to_le32(hash
);
977 #undef NAME_HASH_SHIFT
978 #undef VALUE_HASH_SHIFT
980 #define BLOCK_HASH_SHIFT 16
983 * ext2_xattr_rehash()
985 * Re-compute the extended attribute hash value after an entry has changed.
987 static void ext2_xattr_rehash(struct ext2_xattr_header
*header
,
988 struct ext2_xattr_entry
*entry
)
990 struct ext2_xattr_entry
*here
;
993 ext2_xattr_hash_entry(header
, entry
);
994 here
= ENTRY(header
+1);
995 while (!IS_LAST_ENTRY(here
)) {
997 /* Block is not shared if an entry's hash value == 0 */
1001 hash
= (hash
<< BLOCK_HASH_SHIFT
) ^
1002 (hash
>> (8*sizeof(hash
) - BLOCK_HASH_SHIFT
)) ^
1003 le32_to_cpu(here
->e_hash
);
1004 here
= EXT2_XATTR_NEXT(here
);
1006 header
->h_hash
= cpu_to_le32(hash
);
1009 #undef BLOCK_HASH_SHIFT
1011 #define HASH_BUCKET_BITS 10
1013 struct mb_cache
*ext2_xattr_create_cache(void)
1015 return mb_cache_create(HASH_BUCKET_BITS
);
1018 void ext2_xattr_destroy_cache(struct mb_cache
*cache
)
1021 mb_cache_destroy(cache
);