1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
22 #include <linux/capability.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
45 #include "blockcheck.h"
55 #include "buffer_head_io.h"
60 struct ocfs2_xattr_def_value_root
{
61 struct ocfs2_xattr_value_root xv
;
62 struct ocfs2_extent_rec er
;
65 struct ocfs2_xattr_bucket
{
66 /* The inode these xattrs are associated with */
67 struct inode
*bu_inode
;
69 /* The actual buffers that make up the bucket */
70 struct buffer_head
*bu_bhs
[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
];
72 /* How many blocks make up one bucket for this filesystem */
76 struct ocfs2_xattr_set_ctxt
{
78 struct ocfs2_alloc_context
*meta_ac
;
79 struct ocfs2_alloc_context
*data_ac
;
80 struct ocfs2_cached_dealloc_ctxt dealloc
;
83 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
84 #define OCFS2_XATTR_INLINE_SIZE 80
85 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
86 - sizeof(struct ocfs2_xattr_header) \
88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
89 - sizeof(struct ocfs2_xattr_block) \
90 - sizeof(struct ocfs2_xattr_header) \
93 static struct ocfs2_xattr_def_value_root def_xv
= {
94 .xv
.xr_list
.l_count
= cpu_to_le16(1),
97 struct xattr_handler
*ocfs2_xattr_handlers
[] = {
98 &ocfs2_xattr_user_handler
,
99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
100 &ocfs2_xattr_acl_access_handler
,
101 &ocfs2_xattr_acl_default_handler
,
103 &ocfs2_xattr_trusted_handler
,
104 &ocfs2_xattr_security_handler
,
108 static struct xattr_handler
*ocfs2_xattr_handler_map
[OCFS2_XATTR_MAX
] = {
109 [OCFS2_XATTR_INDEX_USER
] = &ocfs2_xattr_user_handler
,
110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
111 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS
]
112 = &ocfs2_xattr_acl_access_handler
,
113 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
]
114 = &ocfs2_xattr_acl_default_handler
,
116 [OCFS2_XATTR_INDEX_TRUSTED
] = &ocfs2_xattr_trusted_handler
,
117 [OCFS2_XATTR_INDEX_SECURITY
] = &ocfs2_xattr_security_handler
,
120 struct ocfs2_xattr_info
{
127 struct ocfs2_xattr_search
{
128 struct buffer_head
*inode_bh
;
130 * xattr_bh point to the block buffer head which has extended attribute
131 * when extended attribute in inode, xattr_bh is equal to inode_bh.
133 struct buffer_head
*xattr_bh
;
134 struct ocfs2_xattr_header
*header
;
135 struct ocfs2_xattr_bucket
*bucket
;
138 struct ocfs2_xattr_entry
*here
;
142 static int ocfs2_xattr_bucket_get_name_value(struct inode
*inode
,
143 struct ocfs2_xattr_header
*xh
,
148 static int ocfs2_xattr_block_find(struct inode
*inode
,
151 struct ocfs2_xattr_search
*xs
);
152 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
153 struct buffer_head
*root_bh
,
156 struct ocfs2_xattr_search
*xs
);
158 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
159 struct ocfs2_xattr_tree_root
*xt
,
163 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
164 struct ocfs2_xattr_search
*xs
,
165 struct ocfs2_xattr_set_ctxt
*ctxt
);
167 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
168 struct ocfs2_xattr_info
*xi
,
169 struct ocfs2_xattr_search
*xs
,
170 struct ocfs2_xattr_set_ctxt
*ctxt
);
172 static int ocfs2_delete_xattr_index_block(struct inode
*inode
,
173 struct buffer_head
*xb_bh
);
174 static int ocfs2_mv_xattr_buckets(struct inode
*inode
, handle_t
*handle
,
175 u64 src_blk
, u64 last_blk
, u64 to_blk
,
176 unsigned int start_bucket
,
179 static inline u16
ocfs2_xattr_buckets_per_cluster(struct ocfs2_super
*osb
)
181 return (1 << osb
->s_clustersize_bits
) / OCFS2_XATTR_BUCKET_SIZE
;
184 static inline u16
ocfs2_blocks_per_xattr_bucket(struct super_block
*sb
)
186 return OCFS2_XATTR_BUCKET_SIZE
/ (1 << sb
->s_blocksize_bits
);
189 static inline u16
ocfs2_xattr_max_xe_in_bucket(struct super_block
*sb
)
191 u16 len
= sb
->s_blocksize
-
192 offsetof(struct ocfs2_xattr_header
, xh_entries
);
194 return len
/ sizeof(struct ocfs2_xattr_entry
);
197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
201 static struct ocfs2_xattr_bucket
*ocfs2_xattr_bucket_new(struct inode
*inode
)
203 struct ocfs2_xattr_bucket
*bucket
;
204 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
206 BUG_ON(blks
> OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
);
208 bucket
= kzalloc(sizeof(struct ocfs2_xattr_bucket
), GFP_NOFS
);
210 bucket
->bu_inode
= inode
;
211 bucket
->bu_blocks
= blks
;
217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket
*bucket
)
221 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
222 brelse(bucket
->bu_bhs
[i
]);
223 bucket
->bu_bhs
[i
] = NULL
;
227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket
*bucket
)
230 ocfs2_xattr_bucket_relse(bucket
);
231 bucket
->bu_inode
= NULL
;
237 * A bucket that has never been written to disk doesn't need to be
238 * read. We just need the buffer_heads. Don't call this for
239 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
247 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
248 bucket
->bu_bhs
[i
] = sb_getblk(bucket
->bu_inode
->i_sb
,
250 if (!bucket
->bu_bhs
[i
]) {
256 if (!ocfs2_buffer_uptodate(bucket
->bu_inode
,
258 ocfs2_set_new_buffer_uptodate(bucket
->bu_inode
,
263 ocfs2_xattr_bucket_relse(bucket
);
267 /* Read the xattr bucket at xb_blkno */
268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
273 rc
= ocfs2_read_blocks(bucket
->bu_inode
, xb_blkno
,
274 bucket
->bu_blocks
, bucket
->bu_bhs
, 0,
277 rc
= ocfs2_validate_meta_ecc_bhs(bucket
->bu_inode
->i_sb
,
280 &bucket_xh(bucket
)->xh_check
);
286 ocfs2_xattr_bucket_relse(bucket
);
290 static int ocfs2_xattr_bucket_journal_access(handle_t
*handle
,
291 struct ocfs2_xattr_bucket
*bucket
,
296 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
297 rc
= ocfs2_journal_access(handle
, bucket
->bu_inode
,
298 bucket
->bu_bhs
[i
], type
);
308 static void ocfs2_xattr_bucket_journal_dirty(handle_t
*handle
,
309 struct ocfs2_xattr_bucket
*bucket
)
313 ocfs2_compute_meta_ecc_bhs(bucket
->bu_inode
->i_sb
,
314 bucket
->bu_bhs
, bucket
->bu_blocks
,
315 &bucket_xh(bucket
)->xh_check
);
317 for (i
= 0; i
< bucket
->bu_blocks
; i
++)
318 ocfs2_journal_dirty(handle
, bucket
->bu_bhs
[i
]);
321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket
*dest
,
322 struct ocfs2_xattr_bucket
*src
)
325 int blocksize
= src
->bu_inode
->i_sb
->s_blocksize
;
327 BUG_ON(dest
->bu_blocks
!= src
->bu_blocks
);
328 BUG_ON(dest
->bu_inode
!= src
->bu_inode
);
330 for (i
= 0; i
< src
->bu_blocks
; i
++) {
331 memcpy(bucket_block(dest
, i
), bucket_block(src
, i
),
336 static int ocfs2_validate_xattr_block(struct super_block
*sb
,
337 struct buffer_head
*bh
)
340 struct ocfs2_xattr_block
*xb
=
341 (struct ocfs2_xattr_block
*)bh
->b_data
;
343 mlog(0, "Validating xattr block %llu\n",
344 (unsigned long long)bh
->b_blocknr
);
346 BUG_ON(!buffer_uptodate(bh
));
349 * If the ecc fails, we return the error but otherwise
350 * leave the filesystem running. We know any error is
351 * local to this block.
353 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &xb
->xb_check
);
358 * Errors after here are fatal
361 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
363 "Extended attribute block #%llu has bad "
365 (unsigned long long)bh
->b_blocknr
, 7,
370 if (le64_to_cpu(xb
->xb_blkno
) != bh
->b_blocknr
) {
372 "Extended attribute block #%llu has an "
373 "invalid xb_blkno of %llu",
374 (unsigned long long)bh
->b_blocknr
,
375 (unsigned long long)le64_to_cpu(xb
->xb_blkno
));
379 if (le32_to_cpu(xb
->xb_fs_generation
) != OCFS2_SB(sb
)->fs_generation
) {
381 "Extended attribute block #%llu has an invalid "
382 "xb_fs_generation of #%u",
383 (unsigned long long)bh
->b_blocknr
,
384 le32_to_cpu(xb
->xb_fs_generation
));
391 static int ocfs2_read_xattr_block(struct inode
*inode
, u64 xb_blkno
,
392 struct buffer_head
**bh
)
395 struct buffer_head
*tmp
= *bh
;
397 rc
= ocfs2_read_block(inode
, xb_blkno
, &tmp
,
398 ocfs2_validate_xattr_block
);
400 /* If ocfs2_read_block() got us a new bh, pass it up. */
407 static inline const char *ocfs2_xattr_prefix(int name_index
)
409 struct xattr_handler
*handler
= NULL
;
411 if (name_index
> 0 && name_index
< OCFS2_XATTR_MAX
)
412 handler
= ocfs2_xattr_handler_map
[name_index
];
414 return handler
? handler
->prefix
: NULL
;
417 static u32
ocfs2_xattr_name_hash(struct inode
*inode
,
421 /* Get hash value of uuid from super block */
422 u32 hash
= OCFS2_SB(inode
->i_sb
)->uuid_hash
;
425 /* hash extended attribute name */
426 for (i
= 0; i
< name_len
; i
++) {
427 hash
= (hash
<< OCFS2_HASH_SHIFT
) ^
428 (hash
>> (8*sizeof(hash
) - OCFS2_HASH_SHIFT
)) ^
436 * ocfs2_xattr_hash_entry()
438 * Compute the hash of an extended attribute.
440 static void ocfs2_xattr_hash_entry(struct inode
*inode
,
441 struct ocfs2_xattr_header
*header
,
442 struct ocfs2_xattr_entry
*entry
)
445 char *name
= (char *)header
+ le16_to_cpu(entry
->xe_name_offset
);
447 hash
= ocfs2_xattr_name_hash(inode
, name
, entry
->xe_name_len
);
448 entry
->xe_name_hash
= cpu_to_le32(hash
);
453 static int ocfs2_xattr_entry_real_size(int name_len
, size_t value_len
)
457 if (value_len
<= OCFS2_XATTR_INLINE_SIZE
)
458 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(value_len
);
460 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
461 size
+= sizeof(struct ocfs2_xattr_entry
);
466 int ocfs2_calc_security_init(struct inode
*dir
,
467 struct ocfs2_security_xattr_info
*si
,
470 struct ocfs2_alloc_context
**xattr_ac
)
473 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
474 int s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
478 * The max space of security xattr taken inline is
479 * 256(name) + 80(value) + 16(entry) = 352 bytes,
480 * So reserve one metadata block for it is ok.
482 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
483 s_size
> OCFS2_XATTR_FREE_IN_IBODY
) {
484 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
489 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
492 /* reserve clusters for xattr value which will be set in B tree*/
493 if (si
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
494 int new_clusters
= ocfs2_clusters_for_bytes(dir
->i_sb
,
497 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
499 *want_clusters
+= new_clusters
;
504 int ocfs2_calc_xattr_init(struct inode
*dir
,
505 struct buffer_head
*dir_bh
,
507 struct ocfs2_security_xattr_info
*si
,
510 struct ocfs2_alloc_context
**xattr_ac
)
513 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
514 int s_size
= 0, a_size
= 0, acl_len
= 0, new_clusters
;
517 s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
520 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
) {
521 acl_len
= ocfs2_xattr_get_nolock(dir
, dir_bh
,
522 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
,
525 a_size
= ocfs2_xattr_entry_real_size(0, acl_len
);
528 } else if (acl_len
!= 0 && acl_len
!= -ENODATA
) {
534 if (!(s_size
+ a_size
))
538 * The max space of security xattr taken inline is
539 * 256(name) + 80(value) + 16(entry) = 352 bytes,
540 * The max space of acl xattr taken inline is
541 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
542 * when blocksize = 512, may reserve one more cluser for
543 * xattr bucket, otherwise reserve one metadata block
546 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
547 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_IBODY
) {
548 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
553 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
556 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
&&
557 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_BLOCK(dir
)) {
559 *xattr_credits
+= ocfs2_blocks_per_xattr_bucket(dir
->i_sb
);
563 * reserve credits and clusters for xattrs which has large value
564 * and have to be set outside
566 if (si
->enable
&& si
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
567 new_clusters
= ocfs2_clusters_for_bytes(dir
->i_sb
,
569 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
571 *want_clusters
+= new_clusters
;
573 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
&&
574 acl_len
> OCFS2_XATTR_INLINE_SIZE
) {
575 /* for directory, it has DEFAULT and ACCESS two types of acls */
576 new_clusters
= (S_ISDIR(mode
) ? 2 : 1) *
577 ocfs2_clusters_for_bytes(dir
->i_sb
, acl_len
);
578 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
580 *want_clusters
+= new_clusters
;
586 static int ocfs2_xattr_extend_allocation(struct inode
*inode
,
588 struct ocfs2_xattr_value_buf
*vb
,
589 struct ocfs2_xattr_set_ctxt
*ctxt
)
592 handle_t
*handle
= ctxt
->handle
;
593 enum ocfs2_alloc_restarted why
;
594 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
595 u32 prev_clusters
, logical_start
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
596 struct ocfs2_extent_tree et
;
598 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add
);
600 ocfs2_init_xattr_value_extent_tree(&et
, inode
, vb
);
602 status
= vb
->vb_access(handle
, inode
, vb
->vb_bh
,
603 OCFS2_JOURNAL_ACCESS_WRITE
);
609 prev_clusters
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
610 status
= ocfs2_add_clusters_in_btree(osb
,
625 status
= ocfs2_journal_dirty(handle
, vb
->vb_bh
);
631 clusters_to_add
-= le32_to_cpu(vb
->vb_xv
->xr_clusters
) - prev_clusters
;
634 * We should have already allocated enough space before the transaction,
635 * so no need to restart.
637 BUG_ON(why
!= RESTART_NONE
|| clusters_to_add
);
644 static int __ocfs2_remove_xattr_range(struct inode
*inode
,
645 struct ocfs2_xattr_value_buf
*vb
,
646 u32 cpos
, u32 phys_cpos
, u32 len
,
647 struct ocfs2_xattr_set_ctxt
*ctxt
)
650 u64 phys_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
651 handle_t
*handle
= ctxt
->handle
;
652 struct ocfs2_extent_tree et
;
654 ocfs2_init_xattr_value_extent_tree(&et
, inode
, vb
);
656 ret
= vb
->vb_access(handle
, inode
, vb
->vb_bh
,
657 OCFS2_JOURNAL_ACCESS_WRITE
);
663 ret
= ocfs2_remove_extent(inode
, &et
, cpos
, len
, handle
, ctxt
->meta_ac
,
670 le32_add_cpu(&vb
->vb_xv
->xr_clusters
, -len
);
672 ret
= ocfs2_journal_dirty(handle
, vb
->vb_bh
);
678 ret
= ocfs2_cache_cluster_dealloc(&ctxt
->dealloc
, phys_blkno
, len
);
686 static int ocfs2_xattr_shrink_size(struct inode
*inode
,
689 struct ocfs2_xattr_value_buf
*vb
,
690 struct ocfs2_xattr_set_ctxt
*ctxt
)
693 u32 trunc_len
, cpos
, phys_cpos
, alloc_size
;
696 if (old_clusters
<= new_clusters
)
700 trunc_len
= old_clusters
- new_clusters
;
702 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &phys_cpos
,
704 &vb
->vb_xv
->xr_list
);
710 if (alloc_size
> trunc_len
)
711 alloc_size
= trunc_len
;
713 ret
= __ocfs2_remove_xattr_range(inode
, vb
, cpos
,
714 phys_cpos
, alloc_size
,
721 block
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
722 ocfs2_remove_xattr_clusters_from_cache(inode
, block
,
725 trunc_len
-= alloc_size
;
732 static int ocfs2_xattr_value_truncate(struct inode
*inode
,
733 struct ocfs2_xattr_value_buf
*vb
,
735 struct ocfs2_xattr_set_ctxt
*ctxt
)
738 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
739 u32 old_clusters
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
741 if (new_clusters
== old_clusters
)
744 if (new_clusters
> old_clusters
)
745 ret
= ocfs2_xattr_extend_allocation(inode
,
746 new_clusters
- old_clusters
,
749 ret
= ocfs2_xattr_shrink_size(inode
,
750 old_clusters
, new_clusters
,
756 static int ocfs2_xattr_list_entry(char *buffer
, size_t size
,
757 size_t *result
, const char *prefix
,
758 const char *name
, int name_len
)
760 char *p
= buffer
+ *result
;
761 int prefix_len
= strlen(prefix
);
762 int total_len
= prefix_len
+ name_len
+ 1;
764 *result
+= total_len
;
766 /* we are just looking for how big our buffer needs to be */
773 memcpy(p
, prefix
, prefix_len
);
774 memcpy(p
+ prefix_len
, name
, name_len
);
775 p
[prefix_len
+ name_len
] = '\0';
780 static int ocfs2_xattr_list_entries(struct inode
*inode
,
781 struct ocfs2_xattr_header
*header
,
782 char *buffer
, size_t buffer_size
)
786 const char *prefix
, *name
;
788 for (i
= 0 ; i
< le16_to_cpu(header
->xh_count
); i
++) {
789 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
790 type
= ocfs2_xattr_get_type(entry
);
791 prefix
= ocfs2_xattr_prefix(type
);
794 name
= (const char *)header
+
795 le16_to_cpu(entry
->xe_name_offset
);
797 ret
= ocfs2_xattr_list_entry(buffer
, buffer_size
,
798 &result
, prefix
, name
,
808 static int ocfs2_xattr_ibody_list(struct inode
*inode
,
809 struct ocfs2_dinode
*di
,
813 struct ocfs2_xattr_header
*header
= NULL
;
814 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
817 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
820 header
= (struct ocfs2_xattr_header
*)
821 ((void *)di
+ inode
->i_sb
->s_blocksize
-
822 le16_to_cpu(di
->i_xattr_inline_size
));
824 ret
= ocfs2_xattr_list_entries(inode
, header
, buffer
, buffer_size
);
829 static int ocfs2_xattr_block_list(struct inode
*inode
,
830 struct ocfs2_dinode
*di
,
834 struct buffer_head
*blk_bh
= NULL
;
835 struct ocfs2_xattr_block
*xb
;
838 if (!di
->i_xattr_loc
)
841 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
848 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
849 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
850 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
851 ret
= ocfs2_xattr_list_entries(inode
, header
,
852 buffer
, buffer_size
);
854 struct ocfs2_xattr_tree_root
*xt
= &xb
->xb_attrs
.xb_root
;
855 ret
= ocfs2_xattr_tree_list_index_block(inode
, xt
,
856 buffer
, buffer_size
);
864 ssize_t
ocfs2_listxattr(struct dentry
*dentry
,
868 int ret
= 0, i_ret
= 0, b_ret
= 0;
869 struct buffer_head
*di_bh
= NULL
;
870 struct ocfs2_dinode
*di
= NULL
;
871 struct ocfs2_inode_info
*oi
= OCFS2_I(dentry
->d_inode
);
873 if (!ocfs2_supports_xattr(OCFS2_SB(dentry
->d_sb
)))
876 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
879 ret
= ocfs2_inode_lock(dentry
->d_inode
, &di_bh
, 0);
885 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
887 down_read(&oi
->ip_xattr_sem
);
888 i_ret
= ocfs2_xattr_ibody_list(dentry
->d_inode
, di
, buffer
, size
);
896 b_ret
= ocfs2_xattr_block_list(dentry
->d_inode
, di
,
901 up_read(&oi
->ip_xattr_sem
);
902 ocfs2_inode_unlock(dentry
->d_inode
, 0);
906 return i_ret
+ b_ret
;
909 static int ocfs2_xattr_find_entry(int name_index
,
911 struct ocfs2_xattr_search
*xs
)
913 struct ocfs2_xattr_entry
*entry
;
920 name_len
= strlen(name
);
922 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
923 cmp
= name_index
- ocfs2_xattr_get_type(entry
);
925 cmp
= name_len
- entry
->xe_name_len
;
927 cmp
= memcmp(name
, (xs
->base
+
928 le16_to_cpu(entry
->xe_name_offset
)),
936 return cmp
? -ENODATA
: 0;
939 static int ocfs2_xattr_get_value_outside(struct inode
*inode
,
940 struct ocfs2_xattr_value_root
*xv
,
944 u32 cpos
, p_cluster
, num_clusters
, bpc
, clusters
;
947 size_t cplen
, blocksize
;
948 struct buffer_head
*bh
= NULL
;
949 struct ocfs2_extent_list
*el
;
952 clusters
= le32_to_cpu(xv
->xr_clusters
);
953 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
954 blocksize
= inode
->i_sb
->s_blocksize
;
957 while (cpos
< clusters
) {
958 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
965 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
966 /* Copy ocfs2_xattr_value */
967 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
968 ret
= ocfs2_read_block(inode
, blkno
, &bh
, NULL
);
974 cplen
= len
>= blocksize
? blocksize
: len
;
975 memcpy(buffer
, bh
->b_data
, cplen
);
984 cpos
+= num_clusters
;
990 static int ocfs2_xattr_ibody_get(struct inode
*inode
,
995 struct ocfs2_xattr_search
*xs
)
997 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
998 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
999 struct ocfs2_xattr_value_root
*xv
;
1003 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
1006 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
1007 xs
->header
= (struct ocfs2_xattr_header
*)
1008 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
1009 xs
->base
= (void *)xs
->header
;
1010 xs
->here
= xs
->header
->xh_entries
;
1012 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1015 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1017 if (size
> buffer_size
)
1019 if (ocfs2_xattr_is_local(xs
->here
)) {
1020 memcpy(buffer
, (void *)xs
->base
+
1021 le16_to_cpu(xs
->here
->xe_name_offset
) +
1022 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
), size
);
1024 xv
= (struct ocfs2_xattr_value_root
*)
1025 (xs
->base
+ le16_to_cpu(
1026 xs
->here
->xe_name_offset
) +
1027 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
));
1028 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1040 static int ocfs2_xattr_block_get(struct inode
*inode
,
1045 struct ocfs2_xattr_search
*xs
)
1047 struct ocfs2_xattr_block
*xb
;
1048 struct ocfs2_xattr_value_root
*xv
;
1050 int ret
= -ENODATA
, name_offset
, name_len
, block_off
, i
;
1052 xs
->bucket
= ocfs2_xattr_bucket_new(inode
);
1059 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, xs
);
1065 if (xs
->not_found
) {
1070 xb
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
1071 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1074 if (size
> buffer_size
)
1077 name_offset
= le16_to_cpu(xs
->here
->xe_name_offset
);
1078 name_len
= OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
);
1079 i
= xs
->here
- xs
->header
->xh_entries
;
1081 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
1082 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
1083 bucket_xh(xs
->bucket
),
1087 xs
->base
= bucket_block(xs
->bucket
, block_off
);
1089 if (ocfs2_xattr_is_local(xs
->here
)) {
1090 memcpy(buffer
, (void *)xs
->base
+
1091 name_offset
+ name_len
, size
);
1093 xv
= (struct ocfs2_xattr_value_root
*)
1094 (xs
->base
+ name_offset
+ name_len
);
1095 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1105 ocfs2_xattr_bucket_free(xs
->bucket
);
1107 brelse(xs
->xattr_bh
);
1108 xs
->xattr_bh
= NULL
;
1112 int ocfs2_xattr_get_nolock(struct inode
*inode
,
1113 struct buffer_head
*di_bh
,
1120 struct ocfs2_dinode
*di
= NULL
;
1121 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1122 struct ocfs2_xattr_search xis
= {
1123 .not_found
= -ENODATA
,
1125 struct ocfs2_xattr_search xbs
= {
1126 .not_found
= -ENODATA
,
1129 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1132 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1135 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
1136 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1138 down_read(&oi
->ip_xattr_sem
);
1139 ret
= ocfs2_xattr_ibody_get(inode
, name_index
, name
, buffer
,
1141 if (ret
== -ENODATA
&& di
->i_xattr_loc
)
1142 ret
= ocfs2_xattr_block_get(inode
, name_index
, name
, buffer
,
1144 up_read(&oi
->ip_xattr_sem
);
1149 /* ocfs2_xattr_get()
1151 * Copy an extended attribute into the buffer provided.
1152 * Buffer is NULL to compute the size of buffer required.
1154 static int ocfs2_xattr_get(struct inode
*inode
,
1161 struct buffer_head
*di_bh
= NULL
;
1163 ret
= ocfs2_inode_lock(inode
, &di_bh
, 0);
1168 ret
= ocfs2_xattr_get_nolock(inode
, di_bh
, name_index
,
1169 name
, buffer
, buffer_size
);
1171 ocfs2_inode_unlock(inode
, 0);
1178 static int __ocfs2_xattr_set_value_outside(struct inode
*inode
,
1180 struct ocfs2_xattr_value_root
*xv
,
1184 int ret
= 0, i
, cp_len
;
1185 u16 blocksize
= inode
->i_sb
->s_blocksize
;
1186 u32 p_cluster
, num_clusters
;
1187 u32 cpos
= 0, bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1188 u32 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, value_len
);
1190 struct buffer_head
*bh
= NULL
;
1192 BUG_ON(clusters
> le32_to_cpu(xv
->xr_clusters
));
1194 while (cpos
< clusters
) {
1195 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1196 &num_clusters
, &xv
->xr_list
);
1202 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1204 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1205 ret
= ocfs2_read_block(inode
, blkno
, &bh
, NULL
);
1211 ret
= ocfs2_journal_access(handle
,
1214 OCFS2_JOURNAL_ACCESS_WRITE
);
1220 cp_len
= value_len
> blocksize
? blocksize
: value_len
;
1221 memcpy(bh
->b_data
, value
, cp_len
);
1222 value_len
-= cp_len
;
1224 if (cp_len
< blocksize
)
1225 memset(bh
->b_data
+ cp_len
, 0,
1226 blocksize
- cp_len
);
1228 ret
= ocfs2_journal_dirty(handle
, bh
);
1237 * XXX: do we need to empty all the following
1238 * blocks in this cluster?
1243 cpos
+= num_clusters
;
1251 static int ocfs2_xattr_cleanup(struct inode
*inode
,
1253 struct ocfs2_xattr_info
*xi
,
1254 struct ocfs2_xattr_search
*xs
,
1255 struct ocfs2_xattr_value_buf
*vb
,
1259 size_t name_len
= strlen(xi
->name
);
1260 void *val
= xs
->base
+ offs
;
1261 size_t size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1263 ret
= vb
->vb_access(handle
, inode
, vb
->vb_bh
,
1264 OCFS2_JOURNAL_ACCESS_WRITE
);
1269 /* Decrease xattr count */
1270 le16_add_cpu(&xs
->header
->xh_count
, -1);
1271 /* Remove the xattr entry and tree root which has already be set*/
1272 memset((void *)xs
->here
, 0, sizeof(struct ocfs2_xattr_entry
));
1273 memset(val
, 0, size
);
1275 ret
= ocfs2_journal_dirty(handle
, vb
->vb_bh
);
1282 static int ocfs2_xattr_update_entry(struct inode
*inode
,
1284 struct ocfs2_xattr_info
*xi
,
1285 struct ocfs2_xattr_search
*xs
,
1286 struct ocfs2_xattr_value_buf
*vb
,
1291 ret
= vb
->vb_access(handle
, inode
, vb
->vb_bh
,
1292 OCFS2_JOURNAL_ACCESS_WRITE
);
1298 xs
->here
->xe_name_offset
= cpu_to_le16(offs
);
1299 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1300 if (xi
->value_len
<= OCFS2_XATTR_INLINE_SIZE
)
1301 ocfs2_xattr_set_local(xs
->here
, 1);
1303 ocfs2_xattr_set_local(xs
->here
, 0);
1304 ocfs2_xattr_hash_entry(inode
, xs
->header
, xs
->here
);
1306 ret
= ocfs2_journal_dirty(handle
, vb
->vb_bh
);
1314 * ocfs2_xattr_set_value_outside()
1316 * Set large size value in B tree.
1318 static int ocfs2_xattr_set_value_outside(struct inode
*inode
,
1319 struct ocfs2_xattr_info
*xi
,
1320 struct ocfs2_xattr_search
*xs
,
1321 struct ocfs2_xattr_set_ctxt
*ctxt
,
1322 struct ocfs2_xattr_value_buf
*vb
,
1325 size_t name_len
= strlen(xi
->name
);
1326 void *val
= xs
->base
+ offs
;
1327 struct ocfs2_xattr_value_root
*xv
= NULL
;
1328 size_t size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1331 memset(val
, 0, size
);
1332 memcpy(val
, xi
->name
, name_len
);
1333 xv
= (struct ocfs2_xattr_value_root
*)
1334 (val
+ OCFS2_XATTR_SIZE(name_len
));
1335 xv
->xr_clusters
= 0;
1336 xv
->xr_last_eb_blk
= 0;
1337 xv
->xr_list
.l_tree_depth
= 0;
1338 xv
->xr_list
.l_count
= cpu_to_le16(1);
1339 xv
->xr_list
.l_next_free_rec
= 0;
1342 ret
= ocfs2_xattr_value_truncate(inode
, vb
, xi
->value_len
, ctxt
);
1347 ret
= ocfs2_xattr_update_entry(inode
, ctxt
->handle
, xi
, xs
, vb
, offs
);
1352 ret
= __ocfs2_xattr_set_value_outside(inode
, ctxt
->handle
, vb
->vb_xv
,
1353 xi
->value
, xi
->value_len
);
1361 * ocfs2_xattr_set_entry_local()
1363 * Set, replace or remove extended attribute in local.
1365 static void ocfs2_xattr_set_entry_local(struct inode
*inode
,
1366 struct ocfs2_xattr_info
*xi
,
1367 struct ocfs2_xattr_search
*xs
,
1368 struct ocfs2_xattr_entry
*last
,
1371 size_t name_len
= strlen(xi
->name
);
1374 if (xi
->value
&& xs
->not_found
) {
1375 /* Insert the new xattr entry. */
1376 le16_add_cpu(&xs
->header
->xh_count
, 1);
1377 ocfs2_xattr_set_type(last
, xi
->name_index
);
1378 ocfs2_xattr_set_local(last
, 1);
1379 last
->xe_name_len
= name_len
;
1385 first_val
= xs
->base
+ min_offs
;
1386 offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1387 val
= xs
->base
+ offs
;
1389 if (le64_to_cpu(xs
->here
->xe_value_size
) >
1390 OCFS2_XATTR_INLINE_SIZE
)
1391 size
= OCFS2_XATTR_SIZE(name_len
) +
1392 OCFS2_XATTR_ROOT_SIZE
;
1394 size
= OCFS2_XATTR_SIZE(name_len
) +
1395 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1397 if (xi
->value
&& size
== OCFS2_XATTR_SIZE(name_len
) +
1398 OCFS2_XATTR_SIZE(xi
->value_len
)) {
1399 /* The old and the new value have the
1400 same size. Just replace the value. */
1401 ocfs2_xattr_set_local(xs
->here
, 1);
1402 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1403 /* Clear value bytes. */
1404 memset(val
+ OCFS2_XATTR_SIZE(name_len
),
1406 OCFS2_XATTR_SIZE(xi
->value_len
));
1407 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
1412 /* Remove the old name+value. */
1413 memmove(first_val
+ size
, first_val
, val
- first_val
);
1414 memset(first_val
, 0, size
);
1415 xs
->here
->xe_name_hash
= 0;
1416 xs
->here
->xe_name_offset
= 0;
1417 ocfs2_xattr_set_local(xs
->here
, 1);
1418 xs
->here
->xe_value_size
= 0;
1422 /* Adjust all value offsets. */
1423 last
= xs
->header
->xh_entries
;
1424 for (i
= 0 ; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1425 size_t o
= le16_to_cpu(last
->xe_name_offset
);
1428 last
->xe_name_offset
= cpu_to_le16(o
+ size
);
1433 /* Remove the old entry. */
1435 memmove(xs
->here
, xs
->here
+ 1,
1436 (void *)last
- (void *)xs
->here
);
1437 memset(last
, 0, sizeof(struct ocfs2_xattr_entry
));
1438 le16_add_cpu(&xs
->header
->xh_count
, -1);
1442 /* Insert the new name+value. */
1443 size_t size
= OCFS2_XATTR_SIZE(name_len
) +
1444 OCFS2_XATTR_SIZE(xi
->value_len
);
1445 void *val
= xs
->base
+ min_offs
- size
;
1447 xs
->here
->xe_name_offset
= cpu_to_le16(min_offs
- size
);
1448 memset(val
, 0, size
);
1449 memcpy(val
, xi
->name
, name_len
);
1450 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
1453 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1454 ocfs2_xattr_set_local(xs
->here
, 1);
1455 ocfs2_xattr_hash_entry(inode
, xs
->header
, xs
->here
);
1462 * ocfs2_xattr_set_entry()
1464 * Set extended attribute entry into inode or block.
1466 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1467 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1468 * then set value in B tree with set_value_outside().
1470 static int ocfs2_xattr_set_entry(struct inode
*inode
,
1471 struct ocfs2_xattr_info
*xi
,
1472 struct ocfs2_xattr_search
*xs
,
1473 struct ocfs2_xattr_set_ctxt
*ctxt
,
1476 struct ocfs2_xattr_entry
*last
;
1477 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1478 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1479 size_t min_offs
= xs
->end
- xs
->base
, name_len
= strlen(xi
->name
);
1481 handle_t
*handle
= ctxt
->handle
;
1483 struct ocfs2_xattr_info xi_l
= {
1484 .name_index
= xi
->name_index
,
1487 .value_len
= xi
->value_len
,
1489 struct ocfs2_xattr_value_buf vb
= {
1490 .vb_bh
= xs
->xattr_bh
,
1491 .vb_access
= ocfs2_journal_access_di
,
1494 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1495 BUG_ON(xs
->xattr_bh
== xs
->inode_bh
);
1496 vb
.vb_access
= ocfs2_journal_access_xb
;
1498 BUG_ON(xs
->xattr_bh
!= xs
->inode_bh
);
1500 /* Compute min_offs, last and free space. */
1501 last
= xs
->header
->xh_entries
;
1503 for (i
= 0 ; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1504 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
1505 if (offs
< min_offs
)
1510 free
= min_offs
- ((void *)last
- xs
->base
) - sizeof(__u32
);
1514 if (!xs
->not_found
) {
1516 if (ocfs2_xattr_is_local(xs
->here
))
1517 size
= OCFS2_XATTR_SIZE(name_len
) +
1518 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1520 size
= OCFS2_XATTR_SIZE(name_len
) +
1521 OCFS2_XATTR_ROOT_SIZE
;
1522 free
+= (size
+ sizeof(struct ocfs2_xattr_entry
));
1524 /* Check free space in inode or block */
1525 if (xi
->value
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1526 if (free
< sizeof(struct ocfs2_xattr_entry
) +
1527 OCFS2_XATTR_SIZE(name_len
) +
1528 OCFS2_XATTR_ROOT_SIZE
) {
1532 size_l
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1533 xi_l
.value
= (void *)&def_xv
;
1534 xi_l
.value_len
= OCFS2_XATTR_ROOT_SIZE
;
1535 } else if (xi
->value
) {
1536 if (free
< sizeof(struct ocfs2_xattr_entry
) +
1537 OCFS2_XATTR_SIZE(name_len
) +
1538 OCFS2_XATTR_SIZE(xi
->value_len
)) {
1544 if (!xs
->not_found
) {
1545 /* For existing extended attribute */
1546 size_t size
= OCFS2_XATTR_SIZE(name_len
) +
1547 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1548 size_t offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1549 void *val
= xs
->base
+ offs
;
1551 if (ocfs2_xattr_is_local(xs
->here
) && size
== size_l
) {
1552 /* Replace existing local xattr with tree root */
1553 ret
= ocfs2_xattr_set_value_outside(inode
, xi
, xs
,
1558 } else if (!ocfs2_xattr_is_local(xs
->here
)) {
1559 /* For existing xattr which has value outside */
1560 vb
.vb_xv
= (struct ocfs2_xattr_value_root
*)
1561 (val
+ OCFS2_XATTR_SIZE(name_len
));
1563 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1565 * If new value need set outside also,
1566 * first truncate old value to new value,
1567 * then set new value with set_value_outside().
1569 ret
= ocfs2_xattr_value_truncate(inode
,
1578 ret
= ocfs2_xattr_update_entry(inode
,
1589 ret
= __ocfs2_xattr_set_value_outside(inode
,
1599 * If new value need set in local,
1600 * just trucate old value to zero.
1602 ret
= ocfs2_xattr_value_truncate(inode
,
1612 ret
= ocfs2_journal_access_di(handle
, inode
, xs
->inode_bh
,
1613 OCFS2_JOURNAL_ACCESS_WRITE
);
1619 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1620 ret
= vb
.vb_access(handle
, inode
, vb
.vb_bh
,
1621 OCFS2_JOURNAL_ACCESS_WRITE
);
1629 * Set value in local, include set tree root in local.
1630 * This is the first step for value size >INLINE_SIZE.
1632 ocfs2_xattr_set_entry_local(inode
, &xi_l
, xs
, last
, min_offs
);
1634 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1635 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1642 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) &&
1643 (flag
& OCFS2_INLINE_XATTR_FL
)) {
1644 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1645 unsigned int xattrsize
= osb
->s_xattr_inline_size
;
1648 * Adjust extent record count or inline data size
1649 * to reserve space for extended attribute.
1651 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1652 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
1653 le16_add_cpu(&idata
->id_count
, -xattrsize
);
1654 } else if (!(ocfs2_inode_is_fast_symlink(inode
))) {
1655 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
1656 le16_add_cpu(&el
->l_count
, -(xattrsize
/
1657 sizeof(struct ocfs2_extent_rec
)));
1659 di
->i_xattr_inline_size
= cpu_to_le16(xattrsize
);
1661 /* Update xattr flag */
1662 spin_lock(&oi
->ip_lock
);
1663 oi
->ip_dyn_features
|= flag
;
1664 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
1665 spin_unlock(&oi
->ip_lock
);
1667 ret
= ocfs2_journal_dirty(handle
, xs
->inode_bh
);
1671 if (!ret
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1673 * Set value outside in B tree.
1674 * This is the second step for value size > INLINE_SIZE.
1676 size_t offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1677 ret
= ocfs2_xattr_set_value_outside(inode
, xi
, xs
, ctxt
,
1684 * If set value outside failed, we have to clean
1685 * the junk tree root we have already set in local.
1687 ret2
= ocfs2_xattr_cleanup(inode
, ctxt
->handle
,
1697 static int ocfs2_remove_value_outside(struct inode
*inode
,
1698 struct ocfs2_xattr_value_buf
*vb
,
1699 struct ocfs2_xattr_header
*header
)
1702 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1703 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
1705 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
1707 ctxt
.handle
= ocfs2_start_trans(osb
,
1708 ocfs2_remove_extent_credits(osb
->sb
));
1709 if (IS_ERR(ctxt
.handle
)) {
1710 ret
= PTR_ERR(ctxt
.handle
);
1715 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
1716 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
1718 if (!ocfs2_xattr_is_local(entry
)) {
1721 val
= (void *)header
+
1722 le16_to_cpu(entry
->xe_name_offset
);
1723 vb
->vb_xv
= (struct ocfs2_xattr_value_root
*)
1724 (val
+ OCFS2_XATTR_SIZE(entry
->xe_name_len
));
1725 ret
= ocfs2_xattr_value_truncate(inode
, vb
, 0, &ctxt
);
1733 ocfs2_commit_trans(osb
, ctxt
.handle
);
1734 ocfs2_schedule_truncate_log_flush(osb
, 1);
1735 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
1740 static int ocfs2_xattr_ibody_remove(struct inode
*inode
,
1741 struct buffer_head
*di_bh
)
1744 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1745 struct ocfs2_xattr_header
*header
;
1747 struct ocfs2_xattr_value_buf vb
= {
1749 .vb_access
= ocfs2_journal_access_di
,
1752 header
= (struct ocfs2_xattr_header
*)
1753 ((void *)di
+ inode
->i_sb
->s_blocksize
-
1754 le16_to_cpu(di
->i_xattr_inline_size
));
1756 ret
= ocfs2_remove_value_outside(inode
, &vb
, header
);
1761 static int ocfs2_xattr_block_remove(struct inode
*inode
,
1762 struct buffer_head
*blk_bh
)
1764 struct ocfs2_xattr_block
*xb
;
1766 struct ocfs2_xattr_value_buf vb
= {
1768 .vb_access
= ocfs2_journal_access_xb
,
1771 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1772 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
1773 struct ocfs2_xattr_header
*header
= &(xb
->xb_attrs
.xb_header
);
1774 ret
= ocfs2_remove_value_outside(inode
, &vb
, header
);
1776 ret
= ocfs2_delete_xattr_index_block(inode
, blk_bh
);
1781 static int ocfs2_xattr_free_block(struct inode
*inode
,
1784 struct inode
*xb_alloc_inode
;
1785 struct buffer_head
*xb_alloc_bh
= NULL
;
1786 struct buffer_head
*blk_bh
= NULL
;
1787 struct ocfs2_xattr_block
*xb
;
1788 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1794 ret
= ocfs2_read_xattr_block(inode
, block
, &blk_bh
);
1800 ret
= ocfs2_xattr_block_remove(inode
, blk_bh
);
1806 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1807 blk
= le64_to_cpu(xb
->xb_blkno
);
1808 bit
= le16_to_cpu(xb
->xb_suballoc_bit
);
1809 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
1811 xb_alloc_inode
= ocfs2_get_system_file_inode(osb
,
1812 EXTENT_ALLOC_SYSTEM_INODE
,
1813 le16_to_cpu(xb
->xb_suballoc_slot
));
1814 if (!xb_alloc_inode
) {
1819 mutex_lock(&xb_alloc_inode
->i_mutex
);
1821 ret
= ocfs2_inode_lock(xb_alloc_inode
, &xb_alloc_bh
, 1);
1827 handle
= ocfs2_start_trans(osb
, OCFS2_SUBALLOC_FREE
);
1828 if (IS_ERR(handle
)) {
1829 ret
= PTR_ERR(handle
);
1834 ret
= ocfs2_free_suballoc_bits(handle
, xb_alloc_inode
, xb_alloc_bh
,
1839 ocfs2_commit_trans(osb
, handle
);
1841 ocfs2_inode_unlock(xb_alloc_inode
, 1);
1842 brelse(xb_alloc_bh
);
1844 mutex_unlock(&xb_alloc_inode
->i_mutex
);
1845 iput(xb_alloc_inode
);
1852 * ocfs2_xattr_remove()
1854 * Free extended attribute resources associated with this inode.
1856 int ocfs2_xattr_remove(struct inode
*inode
, struct buffer_head
*di_bh
)
1858 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1859 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1863 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1866 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1869 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
1870 ret
= ocfs2_xattr_ibody_remove(inode
, di_bh
);
1877 if (di
->i_xattr_loc
) {
1878 ret
= ocfs2_xattr_free_block(inode
,
1879 le64_to_cpu(di
->i_xattr_loc
));
1886 handle
= ocfs2_start_trans((OCFS2_SB(inode
->i_sb
)),
1887 OCFS2_INODE_UPDATE_CREDITS
);
1888 if (IS_ERR(handle
)) {
1889 ret
= PTR_ERR(handle
);
1893 ret
= ocfs2_journal_access_di(handle
, inode
, di_bh
,
1894 OCFS2_JOURNAL_ACCESS_WRITE
);
1900 di
->i_xattr_loc
= 0;
1902 spin_lock(&oi
->ip_lock
);
1903 oi
->ip_dyn_features
&= ~(OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
);
1904 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
1905 spin_unlock(&oi
->ip_lock
);
1907 ret
= ocfs2_journal_dirty(handle
, di_bh
);
1911 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
1916 static int ocfs2_xattr_has_space_inline(struct inode
*inode
,
1917 struct ocfs2_dinode
*di
)
1919 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1920 unsigned int xattrsize
= OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
;
1923 if (xattrsize
< OCFS2_MIN_XATTR_INLINE_SIZE
)
1926 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1927 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
1928 free
= le16_to_cpu(idata
->id_count
) - le64_to_cpu(di
->i_size
);
1929 } else if (ocfs2_inode_is_fast_symlink(inode
)) {
1930 free
= ocfs2_fast_symlink_chars(inode
->i_sb
) -
1931 le64_to_cpu(di
->i_size
);
1933 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
1934 free
= (le16_to_cpu(el
->l_count
) -
1935 le16_to_cpu(el
->l_next_free_rec
)) *
1936 sizeof(struct ocfs2_extent_rec
);
1938 if (free
>= xattrsize
)
1945 * ocfs2_xattr_ibody_find()
1947 * Find extended attribute in inode block and
1948 * fill search info into struct ocfs2_xattr_search.
1950 static int ocfs2_xattr_ibody_find(struct inode
*inode
,
1953 struct ocfs2_xattr_search
*xs
)
1955 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1956 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1960 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
1963 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
1964 down_read(&oi
->ip_alloc_sem
);
1965 has_space
= ocfs2_xattr_has_space_inline(inode
, di
);
1966 up_read(&oi
->ip_alloc_sem
);
1971 xs
->xattr_bh
= xs
->inode_bh
;
1972 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
1973 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)
1974 xs
->header
= (struct ocfs2_xattr_header
*)
1975 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
1977 xs
->header
= (struct ocfs2_xattr_header
*)
1978 (xs
->end
- OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
);
1979 xs
->base
= (void *)xs
->header
;
1980 xs
->here
= xs
->header
->xh_entries
;
1982 /* Find the named attribute. */
1983 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
1984 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1985 if (ret
&& ret
!= -ENODATA
)
1987 xs
->not_found
= ret
;
1994 * ocfs2_xattr_ibody_set()
1996 * Set, replace or remove an extended attribute into inode block.
1999 static int ocfs2_xattr_ibody_set(struct inode
*inode
,
2000 struct ocfs2_xattr_info
*xi
,
2001 struct ocfs2_xattr_search
*xs
,
2002 struct ocfs2_xattr_set_ctxt
*ctxt
)
2004 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2005 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2008 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
2011 down_write(&oi
->ip_alloc_sem
);
2012 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
2013 if (!ocfs2_xattr_has_space_inline(inode
, di
)) {
2019 ret
= ocfs2_xattr_set_entry(inode
, xi
, xs
, ctxt
,
2020 (OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
));
2022 up_write(&oi
->ip_alloc_sem
);
2028 * ocfs2_xattr_block_find()
2030 * Find extended attribute in external block and
2031 * fill search info into struct ocfs2_xattr_search.
2033 static int ocfs2_xattr_block_find(struct inode
*inode
,
2036 struct ocfs2_xattr_search
*xs
)
2038 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2039 struct buffer_head
*blk_bh
= NULL
;
2040 struct ocfs2_xattr_block
*xb
;
2043 if (!di
->i_xattr_loc
)
2046 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
2053 xs
->xattr_bh
= blk_bh
;
2054 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2056 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2057 xs
->header
= &xb
->xb_attrs
.xb_header
;
2058 xs
->base
= (void *)xs
->header
;
2059 xs
->end
= (void *)(blk_bh
->b_data
) + blk_bh
->b_size
;
2060 xs
->here
= xs
->header
->xh_entries
;
2062 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
2064 ret
= ocfs2_xattr_index_block_find(inode
, blk_bh
,
2068 if (ret
&& ret
!= -ENODATA
) {
2069 xs
->xattr_bh
= NULL
;
2072 xs
->not_found
= ret
;
2081 * ocfs2_xattr_block_set()
2083 * Set, replace or remove an extended attribute into external block.
2086 static int ocfs2_xattr_block_set(struct inode
*inode
,
2087 struct ocfs2_xattr_info
*xi
,
2088 struct ocfs2_xattr_search
*xs
,
2089 struct ocfs2_xattr_set_ctxt
*ctxt
)
2091 struct buffer_head
*new_bh
= NULL
;
2092 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2093 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2094 handle_t
*handle
= ctxt
->handle
;
2095 struct ocfs2_xattr_block
*xblk
= NULL
;
2096 u16 suballoc_bit_start
;
2101 if (!xs
->xattr_bh
) {
2102 ret
= ocfs2_journal_access_di(handle
, inode
, xs
->inode_bh
,
2103 OCFS2_JOURNAL_ACCESS_CREATE
);
2109 ret
= ocfs2_claim_metadata(osb
, handle
, ctxt
->meta_ac
, 1,
2110 &suballoc_bit_start
, &num_got
,
2117 new_bh
= sb_getblk(inode
->i_sb
, first_blkno
);
2118 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
2120 ret
= ocfs2_journal_access_xb(handle
, inode
, new_bh
,
2121 OCFS2_JOURNAL_ACCESS_CREATE
);
2127 /* Initialize ocfs2_xattr_block */
2128 xs
->xattr_bh
= new_bh
;
2129 xblk
= (struct ocfs2_xattr_block
*)new_bh
->b_data
;
2130 memset(xblk
, 0, inode
->i_sb
->s_blocksize
);
2131 strcpy((void *)xblk
, OCFS2_XATTR_BLOCK_SIGNATURE
);
2132 xblk
->xb_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
2133 xblk
->xb_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
2134 xblk
->xb_fs_generation
= cpu_to_le32(osb
->fs_generation
);
2135 xblk
->xb_blkno
= cpu_to_le64(first_blkno
);
2137 xs
->header
= &xblk
->xb_attrs
.xb_header
;
2138 xs
->base
= (void *)xs
->header
;
2139 xs
->end
= (void *)xblk
+ inode
->i_sb
->s_blocksize
;
2140 xs
->here
= xs
->header
->xh_entries
;
2142 ret
= ocfs2_journal_dirty(handle
, new_bh
);
2147 di
->i_xattr_loc
= cpu_to_le64(first_blkno
);
2148 ocfs2_journal_dirty(handle
, xs
->inode_bh
);
2150 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2152 if (!(le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2153 /* Set extended attribute into external block */
2154 ret
= ocfs2_xattr_set_entry(inode
, xi
, xs
, ctxt
,
2155 OCFS2_HAS_XATTR_FL
);
2156 if (!ret
|| ret
!= -ENOSPC
)
2159 ret
= ocfs2_xattr_create_index_block(inode
, xs
, ctxt
);
2164 ret
= ocfs2_xattr_set_entry_index_block(inode
, xi
, xs
, ctxt
);
2171 /* Check whether the new xattr can be inserted into the inode. */
2172 static int ocfs2_xattr_can_be_in_inode(struct inode
*inode
,
2173 struct ocfs2_xattr_info
*xi
,
2174 struct ocfs2_xattr_search
*xs
)
2177 struct ocfs2_xattr_entry
*last
;
2179 size_t min_offs
= xs
->end
- xs
->base
;
2184 last
= xs
->header
->xh_entries
;
2186 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
2187 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
2188 if (offs
< min_offs
)
2193 free
= min_offs
- ((void *)last
- xs
->base
) - sizeof(__u32
);
2197 BUG_ON(!xs
->not_found
);
2199 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
2200 value_size
= OCFS2_XATTR_ROOT_SIZE
;
2202 value_size
= OCFS2_XATTR_SIZE(xi
->value_len
);
2204 if (free
>= sizeof(struct ocfs2_xattr_entry
) +
2205 OCFS2_XATTR_SIZE(strlen(xi
->name
)) + value_size
)
2211 static int ocfs2_calc_xattr_set_need(struct inode
*inode
,
2212 struct ocfs2_dinode
*di
,
2213 struct ocfs2_xattr_info
*xi
,
2214 struct ocfs2_xattr_search
*xis
,
2215 struct ocfs2_xattr_search
*xbs
,
2220 int ret
= 0, old_in_xb
= 0;
2221 int clusters_add
= 0, meta_add
= 0, credits
= 0;
2222 struct buffer_head
*bh
= NULL
;
2223 struct ocfs2_xattr_block
*xb
= NULL
;
2224 struct ocfs2_xattr_entry
*xe
= NULL
;
2225 struct ocfs2_xattr_value_root
*xv
= NULL
;
2227 int name_offset
, name_len
= 0;
2228 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
2233 * Calculate the clusters we need to write.
2234 * No matter whether we replace an old one or add a new one,
2235 * we need this for writing.
2237 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
2238 credits
+= new_clusters
*
2239 ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
2241 if (xis
->not_found
&& xbs
->not_found
) {
2242 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2244 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2245 clusters_add
+= new_clusters
;
2246 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2254 if (!xis
->not_found
) {
2256 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
2257 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
2259 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
2261 int i
, block_off
= 0;
2262 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
2264 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
2265 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
2266 i
= xbs
->here
- xbs
->header
->xh_entries
;
2269 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
2270 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
2271 bucket_xh(xbs
->bucket
),
2274 base
= bucket_block(xbs
->bucket
, block_off
);
2275 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2278 credits
+= OCFS2_XATTR_BLOCK_UPDATE_CREDITS
;
2283 * delete a xattr doesn't need metadata and cluster allocation.
2284 * so just calculate the credits and return.
2286 * The credits for removing the value tree will be extended
2287 * by ocfs2_remove_extent itself.
2290 if (!ocfs2_xattr_is_local(xe
))
2291 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
);
2296 /* do cluster allocation guess first. */
2297 value_size
= le64_to_cpu(xe
->xe_value_size
);
2301 * In xattr set, we always try to set the xe in inode first,
2302 * so if it can be inserted into inode successfully, the old
2303 * one will be removed from the xattr block, and this xattr
2304 * will be inserted into inode as a new xattr in inode.
2306 if (ocfs2_xattr_can_be_in_inode(inode
, xi
, xis
)) {
2307 clusters_add
+= new_clusters
;
2308 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
) +
2309 OCFS2_INODE_UPDATE_CREDITS
;
2310 if (!ocfs2_xattr_is_local(xe
))
2311 credits
+= ocfs2_calc_extend_credits(
2319 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2320 /* the new values will be stored outside. */
2321 u32 old_clusters
= 0;
2323 if (!ocfs2_xattr_is_local(xe
)) {
2324 old_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
2326 xv
= (struct ocfs2_xattr_value_root
*)
2327 (base
+ name_offset
+ name_len
);
2328 value_size
= OCFS2_XATTR_ROOT_SIZE
;
2332 if (old_clusters
>= new_clusters
) {
2333 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
);
2336 meta_add
+= ocfs2_extend_meta_needed(&xv
->xr_list
);
2337 clusters_add
+= new_clusters
- old_clusters
;
2338 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2342 if (value_size
>= OCFS2_XATTR_ROOT_SIZE
)
2347 * Now the new value will be stored inside. So if the new
2348 * value is smaller than the size of value root or the old
2349 * value, we don't need any allocation, otherwise we have
2350 * to guess metadata allocation.
2352 if ((ocfs2_xattr_is_local(xe
) && value_size
>= xi
->value_len
) ||
2353 (!ocfs2_xattr_is_local(xe
) &&
2354 OCFS2_XATTR_ROOT_SIZE
>= xi
->value_len
))
2359 /* calculate metadata allocation. */
2360 if (di
->i_xattr_loc
) {
2361 if (!xbs
->xattr_bh
) {
2362 ret
= ocfs2_read_xattr_block(inode
,
2363 le64_to_cpu(di
->i_xattr_loc
),
2370 xb
= (struct ocfs2_xattr_block
*)bh
->b_data
;
2372 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
2375 * If there is already an xattr tree, good, we can calculate
2376 * like other b-trees. Otherwise we may have the chance of
2377 * create a tree, the credit calculation is borrowed from
2378 * ocfs2_calc_extend_credits with root_el = NULL. And the
2379 * new tree will be cluster based, so no meta is needed.
2381 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
2382 struct ocfs2_extent_list
*el
=
2383 &xb
->xb_attrs
.xb_root
.xt_list
;
2384 meta_add
+= ocfs2_extend_meta_needed(el
);
2385 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2388 credits
+= OCFS2_SUBALLOC_ALLOC
+ 1;
2391 * This cluster will be used either for new bucket or for
2393 * If the cluster size is the same as the bucket size, one
2394 * more is needed since we may need to extend the bucket
2398 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2399 if (OCFS2_XATTR_BUCKET_SIZE
==
2400 OCFS2_SB(inode
->i_sb
)->s_clustersize
) {
2401 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2406 credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
2410 *clusters_need
= clusters_add
;
2412 *meta_need
= meta_add
;
2414 *credits_need
= credits
;
2419 static int ocfs2_init_xattr_set_ctxt(struct inode
*inode
,
2420 struct ocfs2_dinode
*di
,
2421 struct ocfs2_xattr_info
*xi
,
2422 struct ocfs2_xattr_search
*xis
,
2423 struct ocfs2_xattr_search
*xbs
,
2424 struct ocfs2_xattr_set_ctxt
*ctxt
,
2427 int clusters_add
, meta_add
, ret
;
2428 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2430 memset(ctxt
, 0, sizeof(struct ocfs2_xattr_set_ctxt
));
2432 ocfs2_init_dealloc_ctxt(&ctxt
->dealloc
);
2434 ret
= ocfs2_calc_xattr_set_need(inode
, di
, xi
, xis
, xbs
,
2435 &clusters_add
, &meta_add
, credits
);
2441 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2442 "credits = %d\n", xi
->name
, meta_add
, clusters_add
, *credits
);
2445 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
,
2454 ret
= ocfs2_reserve_clusters(osb
, clusters_add
, &ctxt
->data_ac
);
2460 if (ctxt
->meta_ac
) {
2461 ocfs2_free_alloc_context(ctxt
->meta_ac
);
2462 ctxt
->meta_ac
= NULL
;
2466 * We cannot have an error and a non null ctxt->data_ac.
2473 static int __ocfs2_xattr_set_handle(struct inode
*inode
,
2474 struct ocfs2_dinode
*di
,
2475 struct ocfs2_xattr_info
*xi
,
2476 struct ocfs2_xattr_search
*xis
,
2477 struct ocfs2_xattr_search
*xbs
,
2478 struct ocfs2_xattr_set_ctxt
*ctxt
)
2480 int ret
= 0, credits
, old_found
;
2483 /* Remove existing extended attribute */
2484 if (!xis
->not_found
)
2485 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
2486 else if (!xbs
->not_found
)
2487 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2489 /* We always try to set extended attribute into inode first*/
2490 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
2491 if (!ret
&& !xbs
->not_found
) {
2493 * If succeed and that extended attribute existing in
2494 * external block, then we will remove it.
2499 old_found
= xis
->not_found
;
2500 xis
->not_found
= -ENODATA
;
2501 ret
= ocfs2_calc_xattr_set_need(inode
,
2509 xis
->not_found
= old_found
;
2515 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2516 ctxt
->handle
->h_buffer_credits
);
2521 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2522 } else if (ret
== -ENOSPC
) {
2523 if (di
->i_xattr_loc
&& !xbs
->xattr_bh
) {
2524 ret
= ocfs2_xattr_block_find(inode
,
2530 old_found
= xis
->not_found
;
2531 xis
->not_found
= -ENODATA
;
2532 ret
= ocfs2_calc_xattr_set_need(inode
,
2540 xis
->not_found
= old_found
;
2546 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2547 ctxt
->handle
->h_buffer_credits
);
2554 * If no space in inode, we will set extended attribute
2555 * into external block.
2557 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2560 if (!xis
->not_found
) {
2562 * If succeed and that extended attribute
2563 * existing in inode, we will remove it.
2567 xbs
->not_found
= -ENODATA
;
2568 ret
= ocfs2_calc_xattr_set_need(inode
,
2581 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2582 ctxt
->handle
->h_buffer_credits
);
2587 ret
= ocfs2_xattr_ibody_set(inode
, xi
,
2594 /* Update inode ctime. */
2595 ret
= ocfs2_journal_access(ctxt
->handle
, inode
, xis
->inode_bh
,
2596 OCFS2_JOURNAL_ACCESS_WRITE
);
2602 inode
->i_ctime
= CURRENT_TIME
;
2603 di
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
2604 di
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
2605 ocfs2_journal_dirty(ctxt
->handle
, xis
->inode_bh
);
2612 * This function only called duing creating inode
2613 * for init security/acl xattrs of the new inode.
2614 * The xattrs could be put into ibody or extent block,
2615 * xattr bucket would not be use in this case.
2616 * transanction credits also be reserved in here.
2618 int ocfs2_xattr_set_handle(handle_t
*handle
,
2619 struct inode
*inode
,
2620 struct buffer_head
*di_bh
,
2626 struct ocfs2_alloc_context
*meta_ac
,
2627 struct ocfs2_alloc_context
*data_ac
)
2629 struct ocfs2_dinode
*di
;
2632 struct ocfs2_xattr_info xi
= {
2633 .name_index
= name_index
,
2636 .value_len
= value_len
,
2639 struct ocfs2_xattr_search xis
= {
2640 .not_found
= -ENODATA
,
2643 struct ocfs2_xattr_search xbs
= {
2644 .not_found
= -ENODATA
,
2647 struct ocfs2_xattr_set_ctxt ctxt
= {
2653 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2656 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
2657 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2659 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2661 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
2664 if (xis
.not_found
) {
2665 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
2670 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
2673 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2674 brelse(xbs
.xattr_bh
);
2682 * Set, replace or remove an extended attribute for this inode.
2683 * value is NULL to remove an existing extended attribute, else either
2684 * create or replace an extended attribute.
2686 int ocfs2_xattr_set(struct inode
*inode
,
2693 struct buffer_head
*di_bh
= NULL
;
2694 struct ocfs2_dinode
*di
;
2696 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2697 struct inode
*tl_inode
= osb
->osb_tl_inode
;
2698 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
2700 struct ocfs2_xattr_info xi
= {
2701 .name_index
= name_index
,
2704 .value_len
= value_len
,
2707 struct ocfs2_xattr_search xis
= {
2708 .not_found
= -ENODATA
,
2711 struct ocfs2_xattr_search xbs
= {
2712 .not_found
= -ENODATA
,
2715 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2719 * Only xbs will be used on indexed trees. xis doesn't need a
2722 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
2724 mlog_errno(-ENOMEM
);
2728 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2731 goto cleanup_nolock
;
2733 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
2734 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2736 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2738 * Scan inode and external block to find the same name
2739 * extended attribute and collect search infomation.
2741 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
2744 if (xis
.not_found
) {
2745 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
2750 if (xis
.not_found
&& xbs
.not_found
) {
2752 if (flags
& XATTR_REPLACE
)
2759 if (flags
& XATTR_CREATE
)
2764 mutex_lock(&tl_inode
->i_mutex
);
2766 if (ocfs2_truncate_log_needs_flush(osb
)) {
2767 ret
= __ocfs2_flush_truncate_log(osb
);
2769 mutex_unlock(&tl_inode
->i_mutex
);
2774 mutex_unlock(&tl_inode
->i_mutex
);
2776 ret
= ocfs2_init_xattr_set_ctxt(inode
, di
, &xi
, &xis
,
2777 &xbs
, &ctxt
, &credits
);
2783 /* we need to update inode's ctime field, so add credit for it. */
2784 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
2785 ctxt
.handle
= ocfs2_start_trans(osb
, credits
);
2786 if (IS_ERR(ctxt
.handle
)) {
2787 ret
= PTR_ERR(ctxt
.handle
);
2792 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
2794 ocfs2_commit_trans(osb
, ctxt
.handle
);
2797 ocfs2_free_alloc_context(ctxt
.data_ac
);
2799 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2800 if (ocfs2_dealloc_has_cluster(&ctxt
.dealloc
))
2801 ocfs2_schedule_truncate_log_flush(osb
, 1);
2802 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
2804 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2805 ocfs2_inode_unlock(inode
, 1);
2808 brelse(xbs
.xattr_bh
);
2809 ocfs2_xattr_bucket_free(xbs
.bucket
);
2815 * Find the xattr extent rec which may contains name_hash.
2816 * e_cpos will be the first name hash of the xattr rec.
2817 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2819 static int ocfs2_xattr_get_rec(struct inode
*inode
,
2824 struct ocfs2_extent_list
*el
)
2827 struct buffer_head
*eb_bh
= NULL
;
2828 struct ocfs2_extent_block
*eb
;
2829 struct ocfs2_extent_rec
*rec
= NULL
;
2832 if (el
->l_tree_depth
) {
2833 ret
= ocfs2_find_leaf(inode
, el
, name_hash
, &eb_bh
);
2839 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
2842 if (el
->l_tree_depth
) {
2843 ocfs2_error(inode
->i_sb
,
2844 "Inode %lu has non zero tree depth in "
2845 "xattr tree block %llu\n", inode
->i_ino
,
2846 (unsigned long long)eb_bh
->b_blocknr
);
2852 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
2853 rec
= &el
->l_recs
[i
];
2855 if (le32_to_cpu(rec
->e_cpos
) <= name_hash
) {
2856 e_blkno
= le64_to_cpu(rec
->e_blkno
);
2862 ocfs2_error(inode
->i_sb
, "Inode %lu has bad extent "
2863 "record (%u, %u, 0) in xattr", inode
->i_ino
,
2864 le32_to_cpu(rec
->e_cpos
),
2865 ocfs2_rec_clusters(el
, rec
));
2870 *p_blkno
= le64_to_cpu(rec
->e_blkno
);
2871 *num_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
2873 *e_cpos
= le32_to_cpu(rec
->e_cpos
);
2879 typedef int (xattr_bucket_func
)(struct inode
*inode
,
2880 struct ocfs2_xattr_bucket
*bucket
,
2883 static int ocfs2_find_xe_in_bucket(struct inode
*inode
,
2884 struct ocfs2_xattr_bucket
*bucket
,
2891 int i
, ret
= 0, cmp
= 1, block_off
, new_offset
;
2892 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
2893 size_t name_len
= strlen(name
);
2894 struct ocfs2_xattr_entry
*xe
= NULL
;
2898 * We don't use binary search in the bucket because there
2899 * may be multiple entries with the same name hash.
2901 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
2902 xe
= &xh
->xh_entries
[i
];
2904 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
))
2906 else if (name_hash
< le32_to_cpu(xe
->xe_name_hash
))
2909 cmp
= name_index
- ocfs2_xattr_get_type(xe
);
2911 cmp
= name_len
- xe
->xe_name_len
;
2915 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
2926 xe_name
= bucket_block(bucket
, block_off
) + new_offset
;
2927 if (!memcmp(name
, xe_name
, name_len
)) {
2939 * Find the specified xattr entry in a series of buckets.
2940 * This series start from p_blkno and last for num_clusters.
2941 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2942 * the num of the valid buckets.
2944 * Return the buffer_head this xattr should reside in. And if the xattr's
2945 * hash is in the gap of 2 buckets, return the lower bucket.
2947 static int ocfs2_xattr_bucket_find(struct inode
*inode
,
2954 struct ocfs2_xattr_search
*xs
)
2957 struct ocfs2_xattr_header
*xh
= NULL
;
2958 struct ocfs2_xattr_entry
*xe
= NULL
;
2960 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2961 int low_bucket
= 0, bucket
, high_bucket
;
2962 struct ocfs2_xattr_bucket
*search
;
2964 u64 blkno
, lower_blkno
= 0;
2966 search
= ocfs2_xattr_bucket_new(inode
);
2973 ret
= ocfs2_read_xattr_bucket(search
, p_blkno
);
2979 xh
= bucket_xh(search
);
2980 high_bucket
= le16_to_cpu(xh
->xh_num_buckets
) - 1;
2981 while (low_bucket
<= high_bucket
) {
2982 ocfs2_xattr_bucket_relse(search
);
2984 bucket
= (low_bucket
+ high_bucket
) / 2;
2985 blkno
= p_blkno
+ bucket
* blk_per_bucket
;
2986 ret
= ocfs2_read_xattr_bucket(search
, blkno
);
2992 xh
= bucket_xh(search
);
2993 xe
= &xh
->xh_entries
[0];
2994 if (name_hash
< le32_to_cpu(xe
->xe_name_hash
)) {
2995 high_bucket
= bucket
- 1;
3000 * Check whether the hash of the last entry in our
3001 * bucket is larger than the search one. for an empty
3002 * bucket, the last one is also the first one.
3005 xe
= &xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1];
3007 last_hash
= le32_to_cpu(xe
->xe_name_hash
);
3009 /* record lower_blkno which may be the insert place. */
3010 lower_blkno
= blkno
;
3012 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
)) {
3013 low_bucket
= bucket
+ 1;
3017 /* the searched xattr should reside in this bucket if exists. */
3018 ret
= ocfs2_find_xe_in_bucket(inode
, search
,
3019 name_index
, name
, name_hash
,
3029 * Record the bucket we have found.
3030 * When the xattr's hash value is in the gap of 2 buckets, we will
3031 * always set it to the previous bucket.
3034 lower_blkno
= p_blkno
;
3036 /* This should be in cache - we just read it during the search */
3037 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, lower_blkno
);
3043 xs
->header
= bucket_xh(xs
->bucket
);
3044 xs
->base
= bucket_block(xs
->bucket
, 0);
3045 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
3048 xs
->here
= &xs
->header
->xh_entries
[index
];
3049 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name
,
3050 (unsigned long long)bucket_blkno(xs
->bucket
), index
);
3055 ocfs2_xattr_bucket_free(search
);
3059 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
3060 struct buffer_head
*root_bh
,
3063 struct ocfs2_xattr_search
*xs
)
3066 struct ocfs2_xattr_block
*xb
=
3067 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
3068 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
3069 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
3071 u32 first_hash
, num_clusters
= 0;
3072 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
3074 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
3077 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3078 name
, name_hash
, name_index
);
3080 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &first_hash
,
3087 BUG_ON(p_blkno
== 0 || num_clusters
== 0 || first_hash
> name_hash
);
3089 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3090 "in the rec is %u\n", num_clusters
, (unsigned long long)p_blkno
,
3093 ret
= ocfs2_xattr_bucket_find(inode
, name_index
, name
, name_hash
,
3094 p_blkno
, first_hash
, num_clusters
, xs
);
3100 static int ocfs2_iterate_xattr_buckets(struct inode
*inode
,
3103 xattr_bucket_func
*func
,
3107 u32 bpc
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
));
3108 u32 num_buckets
= clusters
* bpc
;
3109 struct ocfs2_xattr_bucket
*bucket
;
3111 bucket
= ocfs2_xattr_bucket_new(inode
);
3113 mlog_errno(-ENOMEM
);
3117 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3118 clusters
, (unsigned long long)blkno
);
3120 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bucket
->bu_blocks
) {
3121 ret
= ocfs2_read_xattr_bucket(bucket
, blkno
);
3128 * The real bucket num in this series of blocks is stored
3129 * in the 1st bucket.
3132 num_buckets
= le16_to_cpu(bucket_xh(bucket
)->xh_num_buckets
);
3134 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3135 (unsigned long long)blkno
,
3136 le32_to_cpu(bucket_xh(bucket
)->xh_entries
[0].xe_name_hash
));
3138 ret
= func(inode
, bucket
, para
);
3141 /* Fall through to bucket_relse() */
3144 ocfs2_xattr_bucket_relse(bucket
);
3149 ocfs2_xattr_bucket_free(bucket
);
3153 struct ocfs2_xattr_tree_list
{
3159 static int ocfs2_xattr_bucket_get_name_value(struct inode
*inode
,
3160 struct ocfs2_xattr_header
*xh
,
3167 if (index
< 0 || index
>= le16_to_cpu(xh
->xh_count
))
3170 name_offset
= le16_to_cpu(xh
->xh_entries
[index
].xe_name_offset
);
3172 *block_off
= name_offset
>> inode
->i_sb
->s_blocksize_bits
;
3173 *new_offset
= name_offset
% inode
->i_sb
->s_blocksize
;
3178 static int ocfs2_list_xattr_bucket(struct inode
*inode
,
3179 struct ocfs2_xattr_bucket
*bucket
,
3183 struct ocfs2_xattr_tree_list
*xl
= (struct ocfs2_xattr_tree_list
*)para
;
3184 int i
, block_off
, new_offset
;
3185 const char *prefix
, *name
;
3187 for (i
= 0 ; i
< le16_to_cpu(bucket_xh(bucket
)->xh_count
); i
++) {
3188 struct ocfs2_xattr_entry
*entry
= &bucket_xh(bucket
)->xh_entries
[i
];
3189 type
= ocfs2_xattr_get_type(entry
);
3190 prefix
= ocfs2_xattr_prefix(type
);
3193 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
3201 name
= (const char *)bucket_block(bucket
, block_off
) +
3203 ret
= ocfs2_xattr_list_entry(xl
->buffer
,
3207 entry
->xe_name_len
);
3216 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
3217 struct ocfs2_xattr_tree_root
*xt
,
3221 struct ocfs2_extent_list
*el
= &xt
->xt_list
;
3223 u32 name_hash
= UINT_MAX
, e_cpos
= 0, num_clusters
= 0;
3225 struct ocfs2_xattr_tree_list xl
= {
3227 .buffer_size
= buffer_size
,
3231 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
3234 while (name_hash
> 0) {
3235 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
3236 &e_cpos
, &num_clusters
, el
);
3242 ret
= ocfs2_iterate_xattr_buckets(inode
, p_blkno
, num_clusters
,
3243 ocfs2_list_xattr_bucket
,
3253 name_hash
= e_cpos
- 1;
3261 static int cmp_xe(const void *a
, const void *b
)
3263 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
3264 u32 l_hash
= le32_to_cpu(l
->xe_name_hash
);
3265 u32 r_hash
= le32_to_cpu(r
->xe_name_hash
);
3267 if (l_hash
> r_hash
)
3269 if (l_hash
< r_hash
)
3274 static void swap_xe(void *a
, void *b
, int size
)
3276 struct ocfs2_xattr_entry
*l
= a
, *r
= b
, tmp
;
3279 memcpy(l
, r
, sizeof(struct ocfs2_xattr_entry
));
3280 memcpy(r
, &tmp
, sizeof(struct ocfs2_xattr_entry
));
3284 * When the ocfs2_xattr_block is filled up, new bucket will be created
3285 * and all the xattr entries will be moved to the new bucket.
3286 * The header goes at the start of the bucket, and the names+values are
3287 * filled from the end. This is why *target starts as the last buffer.
3288 * Note: we need to sort the entries since they are not saved in order
3289 * in the ocfs2_xattr_block.
3291 static void ocfs2_cp_xattr_block_to_bucket(struct inode
*inode
,
3292 struct buffer_head
*xb_bh
,
3293 struct ocfs2_xattr_bucket
*bucket
)
3295 int i
, blocksize
= inode
->i_sb
->s_blocksize
;
3296 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3297 u16 offset
, size
, off_change
;
3298 struct ocfs2_xattr_entry
*xe
;
3299 struct ocfs2_xattr_block
*xb
=
3300 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
3301 struct ocfs2_xattr_header
*xb_xh
= &xb
->xb_attrs
.xb_header
;
3302 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
3303 u16 count
= le16_to_cpu(xb_xh
->xh_count
);
3304 char *src
= xb_bh
->b_data
;
3305 char *target
= bucket_block(bucket
, blks
- 1);
3307 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3308 (unsigned long long)xb_bh
->b_blocknr
,
3309 (unsigned long long)bucket_blkno(bucket
));
3311 for (i
= 0; i
< blks
; i
++)
3312 memset(bucket_block(bucket
, i
), 0, blocksize
);
3315 * Since the xe_name_offset is based on ocfs2_xattr_header,
3316 * there is a offset change corresponding to the change of
3317 * ocfs2_xattr_header's position.
3319 off_change
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
3320 xe
= &xb_xh
->xh_entries
[count
- 1];
3321 offset
= le16_to_cpu(xe
->xe_name_offset
) + off_change
;
3322 size
= blocksize
- offset
;
3324 /* copy all the names and values. */
3325 memcpy(target
+ offset
, src
+ offset
, size
);
3327 /* Init new header now. */
3328 xh
->xh_count
= xb_xh
->xh_count
;
3329 xh
->xh_num_buckets
= cpu_to_le16(1);
3330 xh
->xh_name_value_len
= cpu_to_le16(size
);
3331 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
- size
);
3333 /* copy all the entries. */
3334 target
= bucket_block(bucket
, 0);
3335 offset
= offsetof(struct ocfs2_xattr_header
, xh_entries
);
3336 size
= count
* sizeof(struct ocfs2_xattr_entry
);
3337 memcpy(target
+ offset
, (char *)xb_xh
+ offset
, size
);
3339 /* Change the xe offset for all the xe because of the move. */
3340 off_change
= OCFS2_XATTR_BUCKET_SIZE
- blocksize
+
3341 offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
3342 for (i
= 0; i
< count
; i
++)
3343 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
, off_change
);
3345 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3346 offset
, size
, off_change
);
3348 sort(target
+ offset
, count
, sizeof(struct ocfs2_xattr_entry
),
3353 * After we move xattr from block to index btree, we have to
3354 * update ocfs2_xattr_search to the new xe and base.
3356 * When the entry is in xattr block, xattr_bh indicates the storage place.
3357 * While if the entry is in index b-tree, "bucket" indicates the
3358 * real place of the xattr.
3360 static void ocfs2_xattr_update_xattr_search(struct inode
*inode
,
3361 struct ocfs2_xattr_search
*xs
,
3362 struct buffer_head
*old_bh
)
3364 char *buf
= old_bh
->b_data
;
3365 struct ocfs2_xattr_block
*old_xb
= (struct ocfs2_xattr_block
*)buf
;
3366 struct ocfs2_xattr_header
*old_xh
= &old_xb
->xb_attrs
.xb_header
;
3369 xs
->header
= bucket_xh(xs
->bucket
);
3370 xs
->base
= bucket_block(xs
->bucket
, 0);
3371 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
3376 i
= xs
->here
- old_xh
->xh_entries
;
3377 xs
->here
= &xs
->header
->xh_entries
[i
];
3380 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
3381 struct ocfs2_xattr_search
*xs
,
3382 struct ocfs2_xattr_set_ctxt
*ctxt
)
3387 handle_t
*handle
= ctxt
->handle
;
3388 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3389 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
3390 struct buffer_head
*xb_bh
= xs
->xattr_bh
;
3391 struct ocfs2_xattr_block
*xb
=
3392 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
3393 struct ocfs2_xattr_tree_root
*xr
;
3394 u16 xb_flags
= le16_to_cpu(xb
->xb_flags
);
3396 mlog(0, "create xattr index block for %llu\n",
3397 (unsigned long long)xb_bh
->b_blocknr
);
3399 BUG_ON(xb_flags
& OCFS2_XATTR_INDEXED
);
3400 BUG_ON(!xs
->bucket
);
3404 * We can use this lock for now, and maybe move to a dedicated mutex
3405 * if performance becomes a problem later.
3407 down_write(&oi
->ip_alloc_sem
);
3409 ret
= ocfs2_journal_access_xb(handle
, inode
, xb_bh
,
3410 OCFS2_JOURNAL_ACCESS_WRITE
);
3416 ret
= __ocfs2_claim_clusters(osb
, handle
, ctxt
->data_ac
,
3417 1, 1, &bit_off
, &len
);
3424 * The bucket may spread in many blocks, and
3425 * we will only touch the 1st block and the last block
3426 * in the whole bucket(one for entry and one for data).
3428 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, bit_off
);
3430 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3431 (unsigned long long)blkno
);
3433 ret
= ocfs2_init_xattr_bucket(xs
->bucket
, blkno
);
3439 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
3440 OCFS2_JOURNAL_ACCESS_CREATE
);
3446 ocfs2_cp_xattr_block_to_bucket(inode
, xb_bh
, xs
->bucket
);
3447 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
3449 ocfs2_xattr_update_xattr_search(inode
, xs
, xb_bh
);
3451 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3452 memset(&xb
->xb_attrs
, 0, inode
->i_sb
->s_blocksize
-
3453 offsetof(struct ocfs2_xattr_block
, xb_attrs
));
3455 xr
= &xb
->xb_attrs
.xb_root
;
3456 xr
->xt_clusters
= cpu_to_le32(1);
3457 xr
->xt_last_eb_blk
= 0;
3458 xr
->xt_list
.l_tree_depth
= 0;
3459 xr
->xt_list
.l_count
= cpu_to_le16(ocfs2_xattr_recs_per_xb(inode
->i_sb
));
3460 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
3462 xr
->xt_list
.l_recs
[0].e_cpos
= 0;
3463 xr
->xt_list
.l_recs
[0].e_blkno
= cpu_to_le64(blkno
);
3464 xr
->xt_list
.l_recs
[0].e_leaf_clusters
= cpu_to_le16(1);
3466 xb
->xb_flags
= cpu_to_le16(xb_flags
| OCFS2_XATTR_INDEXED
);
3468 ocfs2_journal_dirty(handle
, xb_bh
);
3471 up_write(&oi
->ip_alloc_sem
);
3476 static int cmp_xe_offset(const void *a
, const void *b
)
3478 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
3479 u32 l_name_offset
= le16_to_cpu(l
->xe_name_offset
);
3480 u32 r_name_offset
= le16_to_cpu(r
->xe_name_offset
);
3482 if (l_name_offset
< r_name_offset
)
3484 if (l_name_offset
> r_name_offset
)
3490 * defrag a xattr bucket if we find that the bucket has some
3491 * holes beteen name/value pairs.
3492 * We will move all the name/value pairs to the end of the bucket
3493 * so that we can spare some space for insertion.
3495 static int ocfs2_defrag_xattr_bucket(struct inode
*inode
,
3497 struct ocfs2_xattr_bucket
*bucket
)
3500 size_t end
, offset
, len
, value_len
;
3501 struct ocfs2_xattr_header
*xh
;
3502 char *entries
, *buf
, *bucket_buf
= NULL
;
3503 u64 blkno
= bucket_blkno(bucket
);
3505 size_t blocksize
= inode
->i_sb
->s_blocksize
;
3506 struct ocfs2_xattr_entry
*xe
;
3509 * In order to make the operation more efficient and generic,
3510 * we copy all the blocks into a contiguous memory and do the
3511 * defragment there, so if anything is error, we will not touch
3514 bucket_buf
= kmalloc(OCFS2_XATTR_BUCKET_SIZE
, GFP_NOFS
);
3521 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
3522 memcpy(buf
, bucket_block(bucket
, i
), blocksize
);
3524 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
3525 OCFS2_JOURNAL_ACCESS_WRITE
);
3531 xh
= (struct ocfs2_xattr_header
*)bucket_buf
;
3532 entries
= (char *)xh
->xh_entries
;
3533 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
3535 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3536 "xh_free_start = %u, xh_name_value_len = %u.\n",
3537 (unsigned long long)blkno
, le16_to_cpu(xh
->xh_count
),
3538 xh_free_start
, le16_to_cpu(xh
->xh_name_value_len
));
3541 * sort all the entries by their offset.
3542 * the largest will be the first, so that we can
3543 * move them to the end one by one.
3545 sort(entries
, le16_to_cpu(xh
->xh_count
),
3546 sizeof(struct ocfs2_xattr_entry
),
3547 cmp_xe_offset
, swap_xe
);
3549 /* Move all name/values to the end of the bucket. */
3550 xe
= xh
->xh_entries
;
3551 end
= OCFS2_XATTR_BUCKET_SIZE
;
3552 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, xe
++) {
3553 offset
= le16_to_cpu(xe
->xe_name_offset
);
3554 if (ocfs2_xattr_is_local(xe
))
3555 value_len
= OCFS2_XATTR_SIZE(
3556 le64_to_cpu(xe
->xe_value_size
));
3558 value_len
= OCFS2_XATTR_ROOT_SIZE
;
3559 len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
) + value_len
;
3562 * We must make sure that the name/value pair
3563 * exist in the same block. So adjust end to
3564 * the previous block end if needed.
3566 if (((end
- len
) / blocksize
!=
3567 (end
- 1) / blocksize
))
3568 end
= end
- end
% blocksize
;
3570 if (end
> offset
+ len
) {
3571 memmove(bucket_buf
+ end
- len
,
3572 bucket_buf
+ offset
, len
);
3573 xe
->xe_name_offset
= cpu_to_le16(end
- len
);
3576 mlog_bug_on_msg(end
< offset
+ len
, "Defrag check failed for "
3577 "bucket %llu\n", (unsigned long long)blkno
);
3582 mlog_bug_on_msg(xh_free_start
> end
, "Defrag check failed for "
3583 "bucket %llu\n", (unsigned long long)blkno
);
3585 if (xh_free_start
== end
)
3588 memset(bucket_buf
+ xh_free_start
, 0, end
- xh_free_start
);
3589 xh
->xh_free_start
= cpu_to_le16(end
);
3591 /* sort the entries by their name_hash. */
3592 sort(entries
, le16_to_cpu(xh
->xh_count
),
3593 sizeof(struct ocfs2_xattr_entry
),
3597 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
3598 memcpy(bucket_block(bucket
, i
), buf
, blocksize
);
3599 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
3607 * prev_blkno points to the start of an existing extent. new_blkno
3608 * points to a newly allocated extent. Because we know each of our
3609 * clusters contains more than bucket, we can easily split one cluster
3610 * at a bucket boundary. So we take the last cluster of the existing
3611 * extent and split it down the middle. We move the last half of the
3612 * buckets in the last cluster of the existing extent over to the new
3615 * first_bh is the buffer at prev_blkno so we can update the existing
3616 * extent's bucket count. header_bh is the bucket were we were hoping
3617 * to insert our xattr. If the bucket move places the target in the new
3618 * extent, we'll update first_bh and header_bh after modifying the old
3621 * first_hash will be set as the 1st xe's name_hash in the new extent.
3623 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode
*inode
,
3625 struct ocfs2_xattr_bucket
*first
,
3626 struct ocfs2_xattr_bucket
*target
,
3632 struct super_block
*sb
= inode
->i_sb
;
3633 int blks_per_bucket
= ocfs2_blocks_per_xattr_bucket(sb
);
3634 int num_buckets
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb
));
3635 int to_move
= num_buckets
/ 2;
3637 u64 last_cluster_blkno
= bucket_blkno(first
) +
3638 ((num_clusters
- 1) * ocfs2_clusters_to_blocks(sb
, 1));
3640 BUG_ON(le16_to_cpu(bucket_xh(first
)->xh_num_buckets
) < num_buckets
);
3641 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
== OCFS2_SB(sb
)->s_clustersize
);
3643 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3644 (unsigned long long)last_cluster_blkno
, (unsigned long long)new_blkno
);
3646 ret
= ocfs2_mv_xattr_buckets(inode
, handle
, bucket_blkno(first
),
3647 last_cluster_blkno
, new_blkno
,
3648 to_move
, first_hash
);
3654 /* This is the first bucket that got moved */
3655 src_blkno
= last_cluster_blkno
+ (to_move
* blks_per_bucket
);
3658 * If the target bucket was part of the moved buckets, we need to
3659 * update first and target.
3661 if (bucket_blkno(target
) >= src_blkno
) {
3662 /* Find the block for the new target bucket */
3663 src_blkno
= new_blkno
+
3664 (bucket_blkno(target
) - src_blkno
);
3666 ocfs2_xattr_bucket_relse(first
);
3667 ocfs2_xattr_bucket_relse(target
);
3670 * These shouldn't fail - the buffers are in the
3671 * journal from ocfs2_cp_xattr_bucket().
3673 ret
= ocfs2_read_xattr_bucket(first
, new_blkno
);
3678 ret
= ocfs2_read_xattr_bucket(target
, src_blkno
);
3689 * Find the suitable pos when we divide a bucket into 2.
3690 * We have to make sure the xattrs with the same hash value exist
3691 * in the same bucket.
3693 * If this ocfs2_xattr_header covers more than one hash value, find a
3694 * place where the hash value changes. Try to find the most even split.
3695 * The most common case is that all entries have different hash values,
3696 * and the first check we make will find a place to split.
3698 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header
*xh
)
3700 struct ocfs2_xattr_entry
*entries
= xh
->xh_entries
;
3701 int count
= le16_to_cpu(xh
->xh_count
);
3702 int delta
, middle
= count
/ 2;
3705 * We start at the middle. Each step gets farther away in both
3706 * directions. We therefore hit the change in hash value
3707 * nearest to the middle. Note that this loop does not execute for
3710 for (delta
= 0; delta
< middle
; delta
++) {
3711 /* Let's check delta earlier than middle */
3712 if (cmp_xe(&entries
[middle
- delta
- 1],
3713 &entries
[middle
- delta
]))
3714 return middle
- delta
;
3716 /* For even counts, don't walk off the end */
3717 if ((middle
+ delta
+ 1) == count
)
3720 /* Now try delta past middle */
3721 if (cmp_xe(&entries
[middle
+ delta
],
3722 &entries
[middle
+ delta
+ 1]))
3723 return middle
+ delta
+ 1;
3726 /* Every entry had the same hash */
3731 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3732 * first_hash will record the 1st hash of the new bucket.
3734 * Normally half of the xattrs will be moved. But we have to make
3735 * sure that the xattrs with the same hash value are stored in the
3736 * same bucket. If all the xattrs in this bucket have the same hash
3737 * value, the new bucket will be initialized as an empty one and the
3738 * first_hash will be initialized as (hash_value+1).
3740 static int ocfs2_divide_xattr_bucket(struct inode
*inode
,
3745 int new_bucket_head
)
3748 int count
, start
, len
, name_value_len
= 0, xe_len
, name_offset
= 0;
3749 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
3750 struct ocfs2_xattr_header
*xh
;
3751 struct ocfs2_xattr_entry
*xe
;
3752 int blocksize
= inode
->i_sb
->s_blocksize
;
3754 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3755 (unsigned long long)blk
, (unsigned long long)new_blk
);
3757 s_bucket
= ocfs2_xattr_bucket_new(inode
);
3758 t_bucket
= ocfs2_xattr_bucket_new(inode
);
3759 if (!s_bucket
|| !t_bucket
) {
3765 ret
= ocfs2_read_xattr_bucket(s_bucket
, blk
);
3771 ret
= ocfs2_xattr_bucket_journal_access(handle
, s_bucket
,
3772 OCFS2_JOURNAL_ACCESS_WRITE
);
3779 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3780 * there's no need to read it.
3782 ret
= ocfs2_init_xattr_bucket(t_bucket
, new_blk
);
3789 * Hey, if we're overwriting t_bucket, what difference does
3790 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3791 * same part of ocfs2_cp_xattr_bucket().
3793 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
3795 OCFS2_JOURNAL_ACCESS_CREATE
:
3796 OCFS2_JOURNAL_ACCESS_WRITE
);
3802 xh
= bucket_xh(s_bucket
);
3803 count
= le16_to_cpu(xh
->xh_count
);
3804 start
= ocfs2_xattr_find_divide_pos(xh
);
3806 if (start
== count
) {
3807 xe
= &xh
->xh_entries
[start
-1];
3810 * initialized a new empty bucket here.
3811 * The hash value is set as one larger than
3812 * that of the last entry in the previous bucket.
3814 for (i
= 0; i
< t_bucket
->bu_blocks
; i
++)
3815 memset(bucket_block(t_bucket
, i
), 0, blocksize
);
3817 xh
= bucket_xh(t_bucket
);
3818 xh
->xh_free_start
= cpu_to_le16(blocksize
);
3819 xh
->xh_entries
[0].xe_name_hash
= xe
->xe_name_hash
;
3820 le32_add_cpu(&xh
->xh_entries
[0].xe_name_hash
, 1);
3822 goto set_num_buckets
;
3825 /* copy the whole bucket to the new first. */
3826 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
3828 /* update the new bucket. */
3829 xh
= bucket_xh(t_bucket
);
3832 * Calculate the total name/value len and xh_free_start for
3833 * the old bucket first.
3835 name_offset
= OCFS2_XATTR_BUCKET_SIZE
;
3837 for (i
= 0; i
< start
; i
++) {
3838 xe
= &xh
->xh_entries
[i
];
3839 xe_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3840 if (ocfs2_xattr_is_local(xe
))
3842 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
3844 xe_len
+= OCFS2_XATTR_ROOT_SIZE
;
3845 name_value_len
+= xe_len
;
3846 if (le16_to_cpu(xe
->xe_name_offset
) < name_offset
)
3847 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3851 * Now begin the modification to the new bucket.
3853 * In the new bucket, We just move the xattr entry to the beginning
3854 * and don't touch the name/value. So there will be some holes in the
3855 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3858 xe
= &xh
->xh_entries
[start
];
3859 len
= sizeof(struct ocfs2_xattr_entry
) * (count
- start
);
3860 mlog(0, "mv xattr entry len %d from %d to %d\n", len
,
3861 (int)((char *)xe
- (char *)xh
),
3862 (int)((char *)xh
->xh_entries
- (char *)xh
));
3863 memmove((char *)xh
->xh_entries
, (char *)xe
, len
);
3864 xe
= &xh
->xh_entries
[count
- start
];
3865 len
= sizeof(struct ocfs2_xattr_entry
) * start
;
3866 memset((char *)xe
, 0, len
);
3868 le16_add_cpu(&xh
->xh_count
, -start
);
3869 le16_add_cpu(&xh
->xh_name_value_len
, -name_value_len
);
3871 /* Calculate xh_free_start for the new bucket. */
3872 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
3873 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
3874 xe
= &xh
->xh_entries
[i
];
3875 xe_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3876 if (ocfs2_xattr_is_local(xe
))
3878 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
3880 xe_len
+= OCFS2_XATTR_ROOT_SIZE
;
3881 if (le16_to_cpu(xe
->xe_name_offset
) <
3882 le16_to_cpu(xh
->xh_free_start
))
3883 xh
->xh_free_start
= xe
->xe_name_offset
;
3887 /* set xh->xh_num_buckets for the new xh. */
3888 if (new_bucket_head
)
3889 xh
->xh_num_buckets
= cpu_to_le16(1);
3891 xh
->xh_num_buckets
= 0;
3893 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
3895 /* store the first_hash of the new bucket. */
3897 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
3900 * Now only update the 1st block of the old bucket. If we
3901 * just added a new empty bucket, there is no need to modify
3907 xh
= bucket_xh(s_bucket
);
3908 memset(&xh
->xh_entries
[start
], 0,
3909 sizeof(struct ocfs2_xattr_entry
) * (count
- start
));
3910 xh
->xh_count
= cpu_to_le16(start
);
3911 xh
->xh_free_start
= cpu_to_le16(name_offset
);
3912 xh
->xh_name_value_len
= cpu_to_le16(name_value_len
);
3914 ocfs2_xattr_bucket_journal_dirty(handle
, s_bucket
);
3917 ocfs2_xattr_bucket_free(s_bucket
);
3918 ocfs2_xattr_bucket_free(t_bucket
);
3924 * Copy xattr from one bucket to another bucket.
3926 * The caller must make sure that the journal transaction
3927 * has enough space for journaling.
3929 static int ocfs2_cp_xattr_bucket(struct inode
*inode
,
3936 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
3938 BUG_ON(s_blkno
== t_blkno
);
3940 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3941 (unsigned long long)s_blkno
, (unsigned long long)t_blkno
,
3944 s_bucket
= ocfs2_xattr_bucket_new(inode
);
3945 t_bucket
= ocfs2_xattr_bucket_new(inode
);
3946 if (!s_bucket
|| !t_bucket
) {
3952 ret
= ocfs2_read_xattr_bucket(s_bucket
, s_blkno
);
3957 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3958 * there's no need to read it.
3960 ret
= ocfs2_init_xattr_bucket(t_bucket
, t_blkno
);
3965 * Hey, if we're overwriting t_bucket, what difference does
3966 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
3967 * cluster to fill, we came here from
3968 * ocfs2_mv_xattr_buckets(), and it is really new -
3969 * ACCESS_CREATE is required. But we also might have moved data
3970 * out of t_bucket before extending back into it.
3971 * ocfs2_add_new_xattr_bucket() can do this - its call to
3972 * ocfs2_add_new_xattr_cluster() may have created a new extent
3973 * and copied out the end of the old extent. Then it re-extends
3974 * the old extent back to create space for new xattrs. That's
3975 * how we get here, and the bucket isn't really new.
3977 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
3979 OCFS2_JOURNAL_ACCESS_CREATE
:
3980 OCFS2_JOURNAL_ACCESS_WRITE
);
3984 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
3985 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
3988 ocfs2_xattr_bucket_free(t_bucket
);
3989 ocfs2_xattr_bucket_free(s_bucket
);
3995 * src_blk points to the start of an existing extent. last_blk points to
3996 * last cluster in that extent. to_blk points to a newly allocated
3997 * extent. We copy the buckets from the cluster at last_blk to the new
3998 * extent. If start_bucket is non-zero, we skip that many buckets before
3999 * we start copying. The new extent's xh_num_buckets gets set to the
4000 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4001 * by the same amount.
4003 static int ocfs2_mv_xattr_buckets(struct inode
*inode
, handle_t
*handle
,
4004 u64 src_blk
, u64 last_blk
, u64 to_blk
,
4005 unsigned int start_bucket
,
4008 int i
, ret
, credits
;
4009 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4010 int blks_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4011 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
4012 struct ocfs2_xattr_bucket
*old_first
, *new_first
;
4014 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4015 (unsigned long long)last_blk
, (unsigned long long)to_blk
);
4017 BUG_ON(start_bucket
>= num_buckets
);
4019 num_buckets
-= start_bucket
;
4020 last_blk
+= (start_bucket
* blks_per_bucket
);
4023 /* The first bucket of the original extent */
4024 old_first
= ocfs2_xattr_bucket_new(inode
);
4025 /* The first bucket of the new extent */
4026 new_first
= ocfs2_xattr_bucket_new(inode
);
4027 if (!old_first
|| !new_first
) {
4033 ret
= ocfs2_read_xattr_bucket(old_first
, src_blk
);
4040 * We need to update the first bucket of the old extent and all
4041 * the buckets going to the new extent.
4043 credits
= ((num_buckets
+ 1) * blks_per_bucket
) +
4044 handle
->h_buffer_credits
;
4045 ret
= ocfs2_extend_trans(handle
, credits
);
4051 ret
= ocfs2_xattr_bucket_journal_access(handle
, old_first
,
4052 OCFS2_JOURNAL_ACCESS_WRITE
);
4058 for (i
= 0; i
< num_buckets
; i
++) {
4059 ret
= ocfs2_cp_xattr_bucket(inode
, handle
,
4060 last_blk
+ (i
* blks_per_bucket
),
4061 to_blk
+ (i
* blks_per_bucket
),
4070 * Get the new bucket ready before we dirty anything
4071 * (This actually shouldn't fail, because we already dirtied
4072 * it once in ocfs2_cp_xattr_bucket()).
4074 ret
= ocfs2_read_xattr_bucket(new_first
, to_blk
);
4079 ret
= ocfs2_xattr_bucket_journal_access(handle
, new_first
,
4080 OCFS2_JOURNAL_ACCESS_WRITE
);
4086 /* Now update the headers */
4087 le16_add_cpu(&bucket_xh(old_first
)->xh_num_buckets
, -num_buckets
);
4088 ocfs2_xattr_bucket_journal_dirty(handle
, old_first
);
4090 bucket_xh(new_first
)->xh_num_buckets
= cpu_to_le16(num_buckets
);
4091 ocfs2_xattr_bucket_journal_dirty(handle
, new_first
);
4094 *first_hash
= le32_to_cpu(bucket_xh(new_first
)->xh_entries
[0].xe_name_hash
);
4097 ocfs2_xattr_bucket_free(new_first
);
4098 ocfs2_xattr_bucket_free(old_first
);
4103 * Move some xattrs in this cluster to the new cluster.
4104 * This function should only be called when bucket size == cluster size.
4105 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4107 static int ocfs2_divide_xattr_cluster(struct inode
*inode
,
4113 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4114 int ret
, credits
= 2 * blk_per_bucket
+ handle
->h_buffer_credits
;
4116 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
< OCFS2_SB(inode
->i_sb
)->s_clustersize
);
4118 ret
= ocfs2_extend_trans(handle
, credits
);
4124 /* Move half of the xattr in start_blk to the next bucket. */
4125 return ocfs2_divide_xattr_bucket(inode
, handle
, prev_blk
,
4126 new_blk
, first_hash
, 1);
4130 * Move some xattrs from the old cluster to the new one since they are not
4131 * contiguous in ocfs2 xattr tree.
4133 * new_blk starts a new separate cluster, and we will move some xattrs from
4134 * prev_blk to it. v_start will be set as the first name hash value in this
4135 * new cluster so that it can be used as e_cpos during tree insertion and
4136 * don't collide with our original b-tree operations. first_bh and header_bh
4137 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4138 * to extend the insert bucket.
4140 * The problem is how much xattr should we move to the new one and when should
4141 * we update first_bh and header_bh?
4142 * 1. If cluster size > bucket size, that means the previous cluster has more
4143 * than 1 bucket, so just move half nums of bucket into the new cluster and
4144 * update the first_bh and header_bh if the insert bucket has been moved
4145 * to the new cluster.
4146 * 2. If cluster_size == bucket_size:
4147 * a) If the previous extent rec has more than one cluster and the insert
4148 * place isn't in the last cluster, copy the entire last cluster to the
4149 * new one. This time, we don't need to upate the first_bh and header_bh
4150 * since they will not be moved into the new cluster.
4151 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4152 * the new one. And we set the extend flag to zero if the insert place is
4153 * moved into the new allocated cluster since no extend is needed.
4155 static int ocfs2_adjust_xattr_cross_cluster(struct inode
*inode
,
4157 struct ocfs2_xattr_bucket
*first
,
4158 struct ocfs2_xattr_bucket
*target
,
4166 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4167 (unsigned long long)bucket_blkno(first
), prev_clusters
,
4168 (unsigned long long)new_blk
);
4170 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
)) > 1) {
4171 ret
= ocfs2_mv_xattr_bucket_cross_cluster(inode
,
4180 /* The start of the last cluster in the first extent */
4181 u64 last_blk
= bucket_blkno(first
) +
4182 ((prev_clusters
- 1) *
4183 ocfs2_clusters_to_blocks(inode
->i_sb
, 1));
4185 if (prev_clusters
> 1 && bucket_blkno(target
) != last_blk
) {
4186 ret
= ocfs2_mv_xattr_buckets(inode
, handle
,
4187 bucket_blkno(first
),
4188 last_blk
, new_blk
, 0,
4193 ret
= ocfs2_divide_xattr_cluster(inode
, handle
,
4199 if ((bucket_blkno(target
) == last_blk
) && extend
)
4208 * Add a new cluster for xattr storage.
4210 * If the new cluster is contiguous with the previous one, it will be
4211 * appended to the same extent record, and num_clusters will be updated.
4212 * If not, we will insert a new extent for it and move some xattrs in
4213 * the last cluster into the new allocated one.
4214 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4215 * lose the benefits of hashing because we'll have to search large leaves.
4216 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4219 * first_bh is the first block of the previous extent rec and header_bh
4220 * indicates the bucket we will insert the new xattrs. They will be updated
4221 * when the header_bh is moved into the new cluster.
4223 static int ocfs2_add_new_xattr_cluster(struct inode
*inode
,
4224 struct buffer_head
*root_bh
,
4225 struct ocfs2_xattr_bucket
*first
,
4226 struct ocfs2_xattr_bucket
*target
,
4230 struct ocfs2_xattr_set_ctxt
*ctxt
)
4233 u16 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
4234 u32 prev_clusters
= *num_clusters
;
4235 u32 clusters_to_add
= 1, bit_off
, num_bits
, v_start
= 0;
4237 handle_t
*handle
= ctxt
->handle
;
4238 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4239 struct ocfs2_extent_tree et
;
4241 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4242 "previous xattr blkno = %llu\n",
4243 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
4244 prev_cpos
, (unsigned long long)bucket_blkno(first
));
4246 ocfs2_init_xattr_tree_extent_tree(&et
, inode
, root_bh
);
4248 ret
= ocfs2_journal_access_xb(handle
, inode
, root_bh
,
4249 OCFS2_JOURNAL_ACCESS_WRITE
);
4255 ret
= __ocfs2_claim_clusters(osb
, handle
, ctxt
->data_ac
, 1,
4256 clusters_to_add
, &bit_off
, &num_bits
);
4263 BUG_ON(num_bits
> clusters_to_add
);
4265 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
4266 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4267 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
4269 if (bucket_blkno(first
) + (prev_clusters
* bpc
) == block
&&
4270 (prev_clusters
+ num_bits
) << osb
->s_clustersize_bits
<=
4271 OCFS2_MAX_XATTR_TREE_LEAF_SIZE
) {
4273 * If this cluster is contiguous with the old one and
4274 * adding this new cluster, we don't surpass the limit of
4275 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4276 * initialized and used like other buckets in the previous
4278 * So add it as a contiguous one. The caller will handle
4281 v_start
= prev_cpos
+ prev_clusters
;
4282 *num_clusters
= prev_clusters
+ num_bits
;
4283 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4286 ret
= ocfs2_adjust_xattr_cross_cluster(inode
,
4300 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4301 num_bits
, (unsigned long long)block
, v_start
);
4302 ret
= ocfs2_insert_extent(osb
, handle
, inode
, &et
, v_start
, block
,
4303 num_bits
, 0, ctxt
->meta_ac
);
4309 ret
= ocfs2_journal_dirty(handle
, root_bh
);
4318 * We are given an extent. 'first' is the bucket at the very front of
4319 * the extent. The extent has space for an additional bucket past
4320 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4321 * of the target bucket. We wish to shift every bucket past the target
4322 * down one, filling in that additional space. When we get back to the
4323 * target, we split the target between itself and the now-empty bucket
4324 * at target+1 (aka, target_blkno + blks_per_bucket).
4326 static int ocfs2_extend_xattr_bucket(struct inode
*inode
,
4328 struct ocfs2_xattr_bucket
*first
,
4333 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4334 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4336 u16 new_bucket
= le16_to_cpu(bucket_xh(first
)->xh_num_buckets
);
4338 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4339 "from %llu, len = %u\n", (unsigned long long)target_blk
,
4340 (unsigned long long)bucket_blkno(first
), num_clusters
);
4342 /* The extent must have room for an additional bucket */
4343 BUG_ON(new_bucket
>=
4344 (num_clusters
* ocfs2_xattr_buckets_per_cluster(osb
)));
4346 /* end_blk points to the last existing bucket */
4347 end_blk
= bucket_blkno(first
) + ((new_bucket
- 1) * blk_per_bucket
);
4350 * end_blk is the start of the last existing bucket.
4351 * Thus, (end_blk - target_blk) covers the target bucket and
4352 * every bucket after it up to, but not including, the last
4353 * existing bucket. Then we add the last existing bucket, the
4354 * new bucket, and the first bucket (3 * blk_per_bucket).
4356 credits
= (end_blk
- target_blk
) + (3 * blk_per_bucket
) +
4357 handle
->h_buffer_credits
;
4358 ret
= ocfs2_extend_trans(handle
, credits
);
4364 ret
= ocfs2_xattr_bucket_journal_access(handle
, first
,
4365 OCFS2_JOURNAL_ACCESS_WRITE
);
4371 while (end_blk
!= target_blk
) {
4372 ret
= ocfs2_cp_xattr_bucket(inode
, handle
, end_blk
,
4373 end_blk
+ blk_per_bucket
, 0);
4376 end_blk
-= blk_per_bucket
;
4379 /* Move half of the xattr in target_blkno to the next bucket. */
4380 ret
= ocfs2_divide_xattr_bucket(inode
, handle
, target_blk
,
4381 target_blk
+ blk_per_bucket
, NULL
, 0);
4383 le16_add_cpu(&bucket_xh(first
)->xh_num_buckets
, 1);
4384 ocfs2_xattr_bucket_journal_dirty(handle
, first
);
4391 * Add new xattr bucket in an extent record and adjust the buckets
4392 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4393 * bucket we want to insert into.
4395 * In the easy case, we will move all the buckets after target down by
4396 * one. Half of target's xattrs will be moved to the next bucket.
4398 * If current cluster is full, we'll allocate a new one. This may not
4399 * be contiguous. The underlying calls will make sure that there is
4400 * space for the insert, shifting buckets around if necessary.
4401 * 'target' may be moved by those calls.
4403 static int ocfs2_add_new_xattr_bucket(struct inode
*inode
,
4404 struct buffer_head
*xb_bh
,
4405 struct ocfs2_xattr_bucket
*target
,
4406 struct ocfs2_xattr_set_ctxt
*ctxt
)
4408 struct ocfs2_xattr_block
*xb
=
4409 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4410 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
4411 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
4413 le32_to_cpu(bucket_xh(target
)->xh_entries
[0].xe_name_hash
);
4414 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4415 int ret
, num_buckets
, extend
= 1;
4417 u32 e_cpos
, num_clusters
;
4418 /* The bucket at the front of the extent */
4419 struct ocfs2_xattr_bucket
*first
;
4421 mlog(0, "Add new xattr bucket starting from %llu\n",
4422 (unsigned long long)bucket_blkno(target
));
4424 /* The first bucket of the original extent */
4425 first
= ocfs2_xattr_bucket_new(inode
);
4432 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &e_cpos
,
4439 ret
= ocfs2_read_xattr_bucket(first
, p_blkno
);
4445 num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
) * num_clusters
;
4446 if (num_buckets
== le16_to_cpu(bucket_xh(first
)->xh_num_buckets
)) {
4448 * This can move first+target if the target bucket moves
4449 * to the new extent.
4451 ret
= ocfs2_add_new_xattr_cluster(inode
,
4466 ret
= ocfs2_extend_xattr_bucket(inode
,
4469 bucket_blkno(target
),
4476 ocfs2_xattr_bucket_free(first
);
4481 static inline char *ocfs2_xattr_bucket_get_val(struct inode
*inode
,
4482 struct ocfs2_xattr_bucket
*bucket
,
4485 int block_off
= offs
>> inode
->i_sb
->s_blocksize_bits
;
4487 offs
= offs
% inode
->i_sb
->s_blocksize
;
4488 return bucket_block(bucket
, block_off
) + offs
;
4492 * Handle the normal xattr set, including replace, delete and new.
4494 * Note: "local" indicates the real data's locality. So we can't
4495 * just its bucket locality by its length.
4497 static void ocfs2_xattr_set_entry_normal(struct inode
*inode
,
4498 struct ocfs2_xattr_info
*xi
,
4499 struct ocfs2_xattr_search
*xs
,
4503 struct ocfs2_xattr_entry
*last
, *xe
;
4504 int name_len
= strlen(xi
->name
);
4505 struct ocfs2_xattr_header
*xh
= xs
->header
;
4506 u16 count
= le16_to_cpu(xh
->xh_count
), start
;
4507 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4509 size_t offs
, size
, new_size
;
4511 last
= &xh
->xh_entries
[count
];
4512 if (!xs
->not_found
) {
4514 offs
= le16_to_cpu(xe
->xe_name_offset
);
4515 if (ocfs2_xattr_is_local(xe
))
4516 size
= OCFS2_XATTR_SIZE(name_len
) +
4517 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
4519 size
= OCFS2_XATTR_SIZE(name_len
) +
4520 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE
);
4523 * If the new value will be stored outside, xi->value has been
4524 * initalized as an empty ocfs2_xattr_value_root, and the same
4525 * goes with xi->value_len, so we can set new_size safely here.
4526 * See ocfs2_xattr_set_in_bucket.
4528 new_size
= OCFS2_XATTR_SIZE(name_len
) +
4529 OCFS2_XATTR_SIZE(xi
->value_len
);
4531 le16_add_cpu(&xh
->xh_name_value_len
, -size
);
4533 if (new_size
> size
)
4534 goto set_new_name_value
;
4536 /* Now replace the old value with new one. */
4538 xe
->xe_value_size
= cpu_to_le64(xi
->value_len
);
4540 xe
->xe_value_size
= 0;
4542 val
= ocfs2_xattr_bucket_get_val(inode
,
4544 memset(val
+ OCFS2_XATTR_SIZE(name_len
), 0,
4545 size
- OCFS2_XATTR_SIZE(name_len
));
4546 if (OCFS2_XATTR_SIZE(xi
->value_len
) > 0)
4547 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
4548 xi
->value
, xi
->value_len
);
4550 le16_add_cpu(&xh
->xh_name_value_len
, new_size
);
4551 ocfs2_xattr_set_local(xe
, local
);
4555 * Remove the old entry if there is more than one.
4556 * We don't remove the last entry so that we can
4557 * use it to indicate the hash value of the empty
4561 le16_add_cpu(&xh
->xh_count
, -1);
4564 (void *)last
- (void *)xe
);
4566 sizeof(struct ocfs2_xattr_entry
));
4569 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
4574 /* find a new entry for insert. */
4575 int low
= 0, high
= count
- 1, tmp
;
4576 struct ocfs2_xattr_entry
*tmp_xe
;
4578 while (low
<= high
&& count
) {
4579 tmp
= (low
+ high
) / 2;
4580 tmp_xe
= &xh
->xh_entries
[tmp
];
4582 if (name_hash
> le32_to_cpu(tmp_xe
->xe_name_hash
))
4584 else if (name_hash
<
4585 le32_to_cpu(tmp_xe
->xe_name_hash
))
4593 xe
= &xh
->xh_entries
[low
];
4595 memmove(xe
+ 1, xe
, (void *)last
- (void *)xe
);
4597 le16_add_cpu(&xh
->xh_count
, 1);
4598 memset(xe
, 0, sizeof(struct ocfs2_xattr_entry
));
4599 xe
->xe_name_hash
= cpu_to_le32(name_hash
);
4600 xe
->xe_name_len
= name_len
;
4601 ocfs2_xattr_set_type(xe
, xi
->name_index
);
4605 /* Insert the new name+value. */
4606 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(xi
->value_len
);
4609 * We must make sure that the name/value pair
4610 * exists in the same block.
4612 offs
= le16_to_cpu(xh
->xh_free_start
);
4613 start
= offs
- size
;
4615 if (start
>> inode
->i_sb
->s_blocksize_bits
!=
4616 (offs
- 1) >> inode
->i_sb
->s_blocksize_bits
) {
4617 offs
= offs
- offs
% blocksize
;
4618 xh
->xh_free_start
= cpu_to_le16(offs
);
4621 val
= ocfs2_xattr_bucket_get_val(inode
, xs
->bucket
, offs
- size
);
4622 xe
->xe_name_offset
= cpu_to_le16(offs
- size
);
4624 memset(val
, 0, size
);
4625 memcpy(val
, xi
->name
, name_len
);
4626 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
), xi
->value
, xi
->value_len
);
4628 xe
->xe_value_size
= cpu_to_le64(xi
->value_len
);
4629 ocfs2_xattr_set_local(xe
, local
);
4631 le16_add_cpu(&xh
->xh_free_start
, -size
);
4632 le16_add_cpu(&xh
->xh_name_value_len
, size
);
4638 * Set the xattr entry in the specified bucket.
4639 * The bucket is indicated by xs->bucket and it should have the enough
4640 * space for the xattr insertion.
4642 static int ocfs2_xattr_set_entry_in_bucket(struct inode
*inode
,
4644 struct ocfs2_xattr_info
*xi
,
4645 struct ocfs2_xattr_search
*xs
,
4652 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4653 (unsigned long)xi
->value_len
, xi
->name_index
,
4654 (unsigned long long)bucket_blkno(xs
->bucket
));
4656 if (!xs
->bucket
->bu_bhs
[1]) {
4657 blkno
= bucket_blkno(xs
->bucket
);
4658 ocfs2_xattr_bucket_relse(xs
->bucket
);
4659 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, blkno
);
4666 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4667 OCFS2_JOURNAL_ACCESS_WRITE
);
4673 ocfs2_xattr_set_entry_normal(inode
, xi
, xs
, name_hash
, local
);
4674 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4681 * Truncate the specified xe_off entry in xattr bucket.
4682 * bucket is indicated by header_bh and len is the new length.
4683 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4685 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4687 static int ocfs2_xattr_bucket_value_truncate(struct inode
*inode
,
4688 struct ocfs2_xattr_bucket
*bucket
,
4691 struct ocfs2_xattr_set_ctxt
*ctxt
)
4695 struct ocfs2_xattr_entry
*xe
;
4696 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
4697 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4698 struct ocfs2_xattr_value_buf vb
= {
4699 .vb_access
= ocfs2_journal_access
,
4702 xe
= &xh
->xh_entries
[xe_off
];
4704 BUG_ON(!xe
|| ocfs2_xattr_is_local(xe
));
4706 offset
= le16_to_cpu(xe
->xe_name_offset
) +
4707 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
4709 value_blk
= offset
/ blocksize
;
4711 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4712 BUG_ON(value_blk
!= (offset
+ OCFS2_XATTR_ROOT_SIZE
- 1) / blocksize
);
4714 vb
.vb_bh
= bucket
->bu_bhs
[value_blk
];
4717 vb
.vb_xv
= (struct ocfs2_xattr_value_root
*)
4718 (vb
.vb_bh
->b_data
+ offset
% blocksize
);
4720 ret
= ocfs2_xattr_bucket_journal_access(ctxt
->handle
, bucket
,
4721 OCFS2_JOURNAL_ACCESS_WRITE
);
4728 * From here on out we have to dirty the bucket. The generic
4729 * value calls only modify one of the bucket's bhs, but we need
4730 * to send the bucket at once. So if they error, they *could* have
4731 * modified something. We have to assume they did, and dirty
4732 * the whole bucket. This leaves us in a consistent state.
4734 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4735 xe_off
, (unsigned long long)bucket_blkno(bucket
), len
);
4736 ret
= ocfs2_xattr_value_truncate(inode
, &vb
, len
, ctxt
);
4742 xe
->xe_value_size
= cpu_to_le64(len
);
4745 ocfs2_xattr_bucket_journal_dirty(ctxt
->handle
, bucket
);
4751 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode
*inode
,
4752 struct ocfs2_xattr_search
*xs
,
4754 struct ocfs2_xattr_set_ctxt
*ctxt
)
4757 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4758 struct ocfs2_xattr_header
*xh
= (struct ocfs2_xattr_header
*)xs
->base
;
4760 BUG_ON(!xs
->bucket
->bu_bhs
[0] || !xe
|| ocfs2_xattr_is_local(xe
));
4762 offset
= xe
- xh
->xh_entries
;
4763 ret
= ocfs2_xattr_bucket_value_truncate(inode
, xs
->bucket
,
4771 static int ocfs2_xattr_bucket_set_value_outside(struct inode
*inode
,
4773 struct ocfs2_xattr_search
*xs
,
4778 struct ocfs2_xattr_value_root
*xv
;
4779 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4781 BUG_ON(!xs
->base
|| !xe
|| ocfs2_xattr_is_local(xe
));
4783 offset
= le16_to_cpu(xe
->xe_name_offset
) +
4784 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
4786 xv
= (struct ocfs2_xattr_value_root
*)(xs
->base
+ offset
);
4788 return __ocfs2_xattr_set_value_outside(inode
, handle
,
4789 xv
, val
, value_len
);
4792 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
4793 struct buffer_head
*root_bh
,
4799 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4800 struct inode
*tl_inode
= osb
->osb_tl_inode
;
4802 struct ocfs2_xattr_block
*xb
=
4803 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
4804 struct ocfs2_alloc_context
*meta_ac
= NULL
;
4805 struct ocfs2_cached_dealloc_ctxt dealloc
;
4806 struct ocfs2_extent_tree et
;
4808 ocfs2_init_xattr_tree_extent_tree(&et
, inode
, root_bh
);
4810 ocfs2_init_dealloc_ctxt(&dealloc
);
4812 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4813 cpos
, len
, (unsigned long long)blkno
);
4815 ocfs2_remove_xattr_clusters_from_cache(inode
, blkno
, len
);
4817 ret
= ocfs2_lock_allocators(inode
, &et
, 0, 1, NULL
, &meta_ac
);
4823 mutex_lock(&tl_inode
->i_mutex
);
4825 if (ocfs2_truncate_log_needs_flush(osb
)) {
4826 ret
= __ocfs2_flush_truncate_log(osb
);
4833 handle
= ocfs2_start_trans(osb
, ocfs2_remove_extent_credits(osb
->sb
));
4834 if (IS_ERR(handle
)) {
4840 ret
= ocfs2_journal_access_xb(handle
, inode
, root_bh
,
4841 OCFS2_JOURNAL_ACCESS_WRITE
);
4847 ret
= ocfs2_remove_extent(inode
, &et
, cpos
, len
, handle
, meta_ac
,
4854 le32_add_cpu(&xb
->xb_attrs
.xb_root
.xt_clusters
, -len
);
4856 ret
= ocfs2_journal_dirty(handle
, root_bh
);
4862 ret
= ocfs2_truncate_log_append(osb
, handle
, blkno
, len
);
4867 ocfs2_commit_trans(osb
, handle
);
4869 ocfs2_schedule_truncate_log_flush(osb
, 1);
4871 mutex_unlock(&tl_inode
->i_mutex
);
4874 ocfs2_free_alloc_context(meta_ac
);
4876 ocfs2_run_deallocs(osb
, &dealloc
);
4881 static void ocfs2_xattr_bucket_remove_xs(struct inode
*inode
,
4883 struct ocfs2_xattr_search
*xs
)
4885 struct ocfs2_xattr_header
*xh
= bucket_xh(xs
->bucket
);
4886 struct ocfs2_xattr_entry
*last
= &xh
->xh_entries
[
4887 le16_to_cpu(xh
->xh_count
) - 1];
4890 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4891 OCFS2_JOURNAL_ACCESS_WRITE
);
4897 /* Remove the old entry. */
4898 memmove(xs
->here
, xs
->here
+ 1,
4899 (void *)last
- (void *)xs
->here
);
4900 memset(last
, 0, sizeof(struct ocfs2_xattr_entry
));
4901 le16_add_cpu(&xh
->xh_count
, -1);
4903 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4907 * Set the xattr name/value in the bucket specified in xs.
4909 * As the new value in xi may be stored in the bucket or in an outside cluster,
4910 * we divide the whole process into 3 steps:
4911 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4912 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4913 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4914 * 4. If the clusters for the new outside value can't be allocated, we need
4915 * to free the xattr we allocated in set.
4917 static int ocfs2_xattr_set_in_bucket(struct inode
*inode
,
4918 struct ocfs2_xattr_info
*xi
,
4919 struct ocfs2_xattr_search
*xs
,
4920 struct ocfs2_xattr_set_ctxt
*ctxt
)
4924 char *val
= (char *)xi
->value
;
4925 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4926 u32 name_hash
= ocfs2_xattr_name_hash(inode
, xi
->name
,
4929 if (!xs
->not_found
&& !ocfs2_xattr_is_local(xe
)) {
4931 * We need to truncate the xattr storage first.
4933 * If both the old and new value are stored to
4934 * outside block, we only need to truncate
4935 * the storage and then set the value outside.
4937 * If the new value should be stored within block,
4938 * we should free all the outside block first and
4939 * the modification to the xattr block will be done
4940 * by following steps.
4942 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
4943 value_len
= xi
->value_len
;
4947 ret
= ocfs2_xattr_bucket_value_truncate_xs(inode
, xs
,
4954 goto set_value_outside
;
4957 value_len
= xi
->value_len
;
4958 /* So we have to handle the inside block change now. */
4959 if (value_len
> OCFS2_XATTR_INLINE_SIZE
) {
4961 * If the new value will be stored outside of block,
4962 * initalize a new empty value root and insert it first.
4965 xi
->value
= &def_xv
;
4966 xi
->value_len
= OCFS2_XATTR_ROOT_SIZE
;
4969 ret
= ocfs2_xattr_set_entry_in_bucket(inode
, ctxt
->handle
, xi
, xs
,
4976 if (value_len
<= OCFS2_XATTR_INLINE_SIZE
)
4979 /* allocate the space now for the outside block storage. */
4980 ret
= ocfs2_xattr_bucket_value_truncate_xs(inode
, xs
,
4985 if (xs
->not_found
) {
4987 * We can't allocate enough clusters for outside
4988 * storage and we have allocated xattr already,
4989 * so need to remove it.
4991 ocfs2_xattr_bucket_remove_xs(inode
, ctxt
->handle
, xs
);
4997 ret
= ocfs2_xattr_bucket_set_value_outside(inode
, ctxt
->handle
,
4998 xs
, val
, value_len
);
5004 * check whether the xattr bucket is filled up with the same hash value.
5005 * If we want to insert the xattr with the same hash, return -ENOSPC.
5006 * If we want to insert a xattr with different hash value, go ahead
5007 * and ocfs2_divide_xattr_bucket will handle this.
5009 static int ocfs2_check_xattr_bucket_collision(struct inode
*inode
,
5010 struct ocfs2_xattr_bucket
*bucket
,
5013 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5014 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
5016 if (name_hash
!= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
))
5019 if (xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1].xe_name_hash
==
5020 xh
->xh_entries
[0].xe_name_hash
) {
5021 mlog(ML_ERROR
, "Too much hash collision in xattr bucket %llu, "
5023 (unsigned long long)bucket_blkno(bucket
),
5024 le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
));
5031 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
5032 struct ocfs2_xattr_info
*xi
,
5033 struct ocfs2_xattr_search
*xs
,
5034 struct ocfs2_xattr_set_ctxt
*ctxt
)
5036 struct ocfs2_xattr_header
*xh
;
5037 struct ocfs2_xattr_entry
*xe
;
5038 u16 count
, header_size
, xh_free_start
;
5039 int free
, max_free
, need
, old
;
5040 size_t value_size
= 0, name_len
= strlen(xi
->name
);
5041 size_t blocksize
= inode
->i_sb
->s_blocksize
;
5042 int ret
, allocation
= 0;
5044 mlog_entry("Set xattr %s in xattr index block\n", xi
->name
);
5048 count
= le16_to_cpu(xh
->xh_count
);
5049 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
5050 header_size
= sizeof(struct ocfs2_xattr_header
) +
5051 count
* sizeof(struct ocfs2_xattr_entry
);
5052 max_free
= OCFS2_XATTR_BUCKET_SIZE
-
5053 le16_to_cpu(xh
->xh_name_value_len
) - header_size
;
5055 mlog_bug_on_msg(header_size
> blocksize
, "bucket %llu has header size "
5056 "of %u which exceed block size\n",
5057 (unsigned long long)bucket_blkno(xs
->bucket
),
5060 if (xi
->value
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
5061 value_size
= OCFS2_XATTR_ROOT_SIZE
;
5063 value_size
= OCFS2_XATTR_SIZE(xi
->value_len
);
5066 need
= sizeof(struct ocfs2_xattr_entry
) +
5067 OCFS2_XATTR_SIZE(name_len
) + value_size
;
5069 need
= value_size
+ OCFS2_XATTR_SIZE(name_len
);
5072 * We only replace the old value if the new length is smaller
5073 * than the old one. Otherwise we will allocate new space in the
5074 * bucket to store it.
5077 if (ocfs2_xattr_is_local(xe
))
5078 old
= OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
5080 old
= OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE
);
5082 if (old
>= value_size
)
5086 free
= xh_free_start
- header_size
;
5088 * We need to make sure the new name/value pair
5089 * can exist in the same block.
5091 if (xh_free_start
% blocksize
< need
)
5092 free
-= xh_free_start
% blocksize
;
5094 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5095 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5096 " %u\n", xs
->not_found
,
5097 (unsigned long long)bucket_blkno(xs
->bucket
),
5098 free
, need
, max_free
, le16_to_cpu(xh
->xh_free_start
),
5099 le16_to_cpu(xh
->xh_name_value_len
));
5103 count
== ocfs2_xattr_max_xe_in_bucket(inode
->i_sb
))) {
5104 if (need
<= max_free
&&
5105 count
< ocfs2_xattr_max_xe_in_bucket(inode
->i_sb
)) {
5107 * We can create the space by defragment. Since only the
5108 * name/value will be moved, the xe shouldn't be changed
5111 ret
= ocfs2_defrag_xattr_bucket(inode
, ctxt
->handle
,
5118 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
5119 free
= xh_free_start
- header_size
;
5120 if (xh_free_start
% blocksize
< need
)
5121 free
-= xh_free_start
% blocksize
;
5126 mlog(0, "Can't get enough space for xattr insert by "
5127 "defragment. Need %u bytes, but we have %d, so "
5128 "allocate new bucket for it.\n", need
, free
);
5132 * We have to add new buckets or clusters and one
5133 * allocation should leave us enough space for insert.
5138 * We do not allow for overlapping ranges between buckets. And
5139 * the maximum number of collisions we will allow for then is
5140 * one bucket's worth, so check it here whether we need to
5141 * add a new bucket for the insert.
5143 ret
= ocfs2_check_xattr_bucket_collision(inode
,
5151 ret
= ocfs2_add_new_xattr_bucket(inode
,
5161 * ocfs2_add_new_xattr_bucket() will have updated
5162 * xs->bucket if it moved, but it will not have updated
5163 * any of the other search fields. Thus, we drop it and
5164 * re-search. Everything should be cached, so it'll be
5167 ocfs2_xattr_bucket_relse(xs
->bucket
);
5168 ret
= ocfs2_xattr_index_block_find(inode
, xs
->xattr_bh
,
5171 if (ret
&& ret
!= -ENODATA
)
5173 xs
->not_found
= ret
;
5179 ret
= ocfs2_xattr_set_in_bucket(inode
, xi
, xs
, ctxt
);
5185 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
5186 struct ocfs2_xattr_bucket
*bucket
,
5190 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5192 struct ocfs2_xattr_entry
*xe
;
5193 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5194 struct ocfs2_xattr_set_ctxt ctxt
= {NULL
, NULL
,};
5195 int credits
= ocfs2_remove_extent_credits(osb
->sb
) +
5196 ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
5199 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
5201 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
5202 xe
= &xh
->xh_entries
[i
];
5203 if (ocfs2_xattr_is_local(xe
))
5206 ctxt
.handle
= ocfs2_start_trans(osb
, credits
);
5207 if (IS_ERR(ctxt
.handle
)) {
5208 ret
= PTR_ERR(ctxt
.handle
);
5213 ret
= ocfs2_xattr_bucket_value_truncate(inode
, bucket
,
5216 ocfs2_commit_trans(osb
, ctxt
.handle
);
5223 ocfs2_schedule_truncate_log_flush(osb
, 1);
5224 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
5228 static int ocfs2_delete_xattr_index_block(struct inode
*inode
,
5229 struct buffer_head
*xb_bh
)
5231 struct ocfs2_xattr_block
*xb
=
5232 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
5233 struct ocfs2_extent_list
*el
= &xb
->xb_attrs
.xb_root
.xt_list
;
5235 u32 name_hash
= UINT_MAX
, e_cpos
, num_clusters
;
5238 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
5241 while (name_hash
> 0) {
5242 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
5243 &e_cpos
, &num_clusters
, el
);
5249 ret
= ocfs2_iterate_xattr_buckets(inode
, p_blkno
, num_clusters
,
5250 ocfs2_delete_xattr_in_bucket
,
5257 ret
= ocfs2_rm_xattr_cluster(inode
, xb_bh
,
5258 p_blkno
, e_cpos
, num_clusters
);
5267 name_hash
= e_cpos
- 1;
5275 * 'security' attributes support
5277 static size_t ocfs2_xattr_security_list(struct inode
*inode
, char *list
,
5278 size_t list_size
, const char *name
,
5281 const size_t prefix_len
= XATTR_SECURITY_PREFIX_LEN
;
5282 const size_t total_len
= prefix_len
+ name_len
+ 1;
5284 if (list
&& total_len
<= list_size
) {
5285 memcpy(list
, XATTR_SECURITY_PREFIX
, prefix_len
);
5286 memcpy(list
+ prefix_len
, name
, name_len
);
5287 list
[prefix_len
+ name_len
] = '\0';
5292 static int ocfs2_xattr_security_get(struct inode
*inode
, const char *name
,
5293 void *buffer
, size_t size
)
5295 if (strcmp(name
, "") == 0)
5297 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_SECURITY
, name
,
5301 static int ocfs2_xattr_security_set(struct inode
*inode
, const char *name
,
5302 const void *value
, size_t size
, int flags
)
5304 if (strcmp(name
, "") == 0)
5307 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_SECURITY
, name
, value
,
5311 int ocfs2_init_security_get(struct inode
*inode
,
5313 struct ocfs2_security_xattr_info
*si
)
5315 return security_inode_init_security(inode
, dir
, &si
->name
, &si
->value
,
5319 int ocfs2_init_security_set(handle_t
*handle
,
5320 struct inode
*inode
,
5321 struct buffer_head
*di_bh
,
5322 struct ocfs2_security_xattr_info
*si
,
5323 struct ocfs2_alloc_context
*xattr_ac
,
5324 struct ocfs2_alloc_context
*data_ac
)
5326 return ocfs2_xattr_set_handle(handle
, inode
, di_bh
,
5327 OCFS2_XATTR_INDEX_SECURITY
,
5328 si
->name
, si
->value
, si
->value_len
, 0,
5332 struct xattr_handler ocfs2_xattr_security_handler
= {
5333 .prefix
= XATTR_SECURITY_PREFIX
,
5334 .list
= ocfs2_xattr_security_list
,
5335 .get
= ocfs2_xattr_security_get
,
5336 .set
= ocfs2_xattr_security_set
,
5340 * 'trusted' attributes support
5342 static size_t ocfs2_xattr_trusted_list(struct inode
*inode
, char *list
,
5343 size_t list_size
, const char *name
,
5346 const size_t prefix_len
= XATTR_TRUSTED_PREFIX_LEN
;
5347 const size_t total_len
= prefix_len
+ name_len
+ 1;
5349 if (list
&& total_len
<= list_size
) {
5350 memcpy(list
, XATTR_TRUSTED_PREFIX
, prefix_len
);
5351 memcpy(list
+ prefix_len
, name
, name_len
);
5352 list
[prefix_len
+ name_len
] = '\0';
5357 static int ocfs2_xattr_trusted_get(struct inode
*inode
, const char *name
,
5358 void *buffer
, size_t size
)
5360 if (strcmp(name
, "") == 0)
5362 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_TRUSTED
, name
,
5366 static int ocfs2_xattr_trusted_set(struct inode
*inode
, const char *name
,
5367 const void *value
, size_t size
, int flags
)
5369 if (strcmp(name
, "") == 0)
5372 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_TRUSTED
, name
, value
,
5376 struct xattr_handler ocfs2_xattr_trusted_handler
= {
5377 .prefix
= XATTR_TRUSTED_PREFIX
,
5378 .list
= ocfs2_xattr_trusted_list
,
5379 .get
= ocfs2_xattr_trusted_get
,
5380 .set
= ocfs2_xattr_trusted_set
,
5384 * 'user' attributes support
5386 static size_t ocfs2_xattr_user_list(struct inode
*inode
, char *list
,
5387 size_t list_size
, const char *name
,
5390 const size_t prefix_len
= XATTR_USER_PREFIX_LEN
;
5391 const size_t total_len
= prefix_len
+ name_len
+ 1;
5392 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5394 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5397 if (list
&& total_len
<= list_size
) {
5398 memcpy(list
, XATTR_USER_PREFIX
, prefix_len
);
5399 memcpy(list
+ prefix_len
, name
, name_len
);
5400 list
[prefix_len
+ name_len
] = '\0';
5405 static int ocfs2_xattr_user_get(struct inode
*inode
, const char *name
,
5406 void *buffer
, size_t size
)
5408 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5410 if (strcmp(name
, "") == 0)
5412 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5414 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_USER
, name
,
5418 static int ocfs2_xattr_user_set(struct inode
*inode
, const char *name
,
5419 const void *value
, size_t size
, int flags
)
5421 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5423 if (strcmp(name
, "") == 0)
5425 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5428 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_USER
, name
, value
,
5432 struct xattr_handler ocfs2_xattr_user_handler
= {
5433 .prefix
= XATTR_USER_PREFIX
,
5434 .list
= ocfs2_xattr_user_list
,
5435 .get
= ocfs2_xattr_user_get
,
5436 .set
= ocfs2_xattr_user_set
,