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>
54 #include "buffer_head_io.h"
59 struct ocfs2_xattr_def_value_root
{
60 struct ocfs2_xattr_value_root xv
;
61 struct ocfs2_extent_rec er
;
64 struct ocfs2_xattr_bucket
{
65 /* The inode these xattrs are associated with */
66 struct inode
*bu_inode
;
68 /* The actual buffers that make up the bucket */
69 struct buffer_head
*bu_bhs
[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
];
71 /* How many blocks make up one bucket for this filesystem */
75 struct ocfs2_xattr_set_ctxt
{
77 struct ocfs2_alloc_context
*meta_ac
;
78 struct ocfs2_alloc_context
*data_ac
;
79 struct ocfs2_cached_dealloc_ctxt dealloc
;
82 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
83 #define OCFS2_XATTR_INLINE_SIZE 80
84 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
85 - sizeof(struct ocfs2_xattr_header) \
87 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
88 - sizeof(struct ocfs2_xattr_block) \
89 - sizeof(struct ocfs2_xattr_header) \
92 static struct ocfs2_xattr_def_value_root def_xv
= {
93 .xv
.xr_list
.l_count
= cpu_to_le16(1),
96 struct xattr_handler
*ocfs2_xattr_handlers
[] = {
97 &ocfs2_xattr_user_handler
,
98 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
99 &ocfs2_xattr_acl_access_handler
,
100 &ocfs2_xattr_acl_default_handler
,
102 &ocfs2_xattr_trusted_handler
,
103 &ocfs2_xattr_security_handler
,
107 static struct xattr_handler
*ocfs2_xattr_handler_map
[OCFS2_XATTR_MAX
] = {
108 [OCFS2_XATTR_INDEX_USER
] = &ocfs2_xattr_user_handler
,
109 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
110 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS
]
111 = &ocfs2_xattr_acl_access_handler
,
112 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
]
113 = &ocfs2_xattr_acl_default_handler
,
115 [OCFS2_XATTR_INDEX_TRUSTED
] = &ocfs2_xattr_trusted_handler
,
116 [OCFS2_XATTR_INDEX_SECURITY
] = &ocfs2_xattr_security_handler
,
119 struct ocfs2_xattr_info
{
126 struct ocfs2_xattr_search
{
127 struct buffer_head
*inode_bh
;
129 * xattr_bh point to the block buffer head which has extended attribute
130 * when extended attribute in inode, xattr_bh is equal to inode_bh.
132 struct buffer_head
*xattr_bh
;
133 struct ocfs2_xattr_header
*header
;
134 struct ocfs2_xattr_bucket
*bucket
;
137 struct ocfs2_xattr_entry
*here
;
141 static int ocfs2_xattr_bucket_get_name_value(struct inode
*inode
,
142 struct ocfs2_xattr_header
*xh
,
147 static int ocfs2_xattr_block_find(struct inode
*inode
,
150 struct ocfs2_xattr_search
*xs
);
151 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
152 struct buffer_head
*root_bh
,
155 struct ocfs2_xattr_search
*xs
);
157 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
158 struct ocfs2_xattr_tree_root
*xt
,
162 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
163 struct ocfs2_xattr_search
*xs
,
164 struct ocfs2_xattr_set_ctxt
*ctxt
);
166 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
167 struct ocfs2_xattr_info
*xi
,
168 struct ocfs2_xattr_search
*xs
,
169 struct ocfs2_xattr_set_ctxt
*ctxt
);
171 static int ocfs2_delete_xattr_index_block(struct inode
*inode
,
172 struct buffer_head
*xb_bh
);
174 static inline u16
ocfs2_xattr_buckets_per_cluster(struct ocfs2_super
*osb
)
176 return (1 << osb
->s_clustersize_bits
) / OCFS2_XATTR_BUCKET_SIZE
;
179 static inline u16
ocfs2_blocks_per_xattr_bucket(struct super_block
*sb
)
181 return OCFS2_XATTR_BUCKET_SIZE
/ (1 << sb
->s_blocksize_bits
);
184 static inline u16
ocfs2_xattr_max_xe_in_bucket(struct super_block
*sb
)
186 u16 len
= sb
->s_blocksize
-
187 offsetof(struct ocfs2_xattr_header
, xh_entries
);
189 return len
/ sizeof(struct ocfs2_xattr_entry
);
192 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
193 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
194 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
196 static struct ocfs2_xattr_bucket
*ocfs2_xattr_bucket_new(struct inode
*inode
)
198 struct ocfs2_xattr_bucket
*bucket
;
199 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
201 BUG_ON(blks
> OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
);
203 bucket
= kzalloc(sizeof(struct ocfs2_xattr_bucket
), GFP_NOFS
);
205 bucket
->bu_inode
= inode
;
206 bucket
->bu_blocks
= blks
;
212 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket
*bucket
)
216 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
217 brelse(bucket
->bu_bhs
[i
]);
218 bucket
->bu_bhs
[i
] = NULL
;
222 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket
*bucket
)
225 ocfs2_xattr_bucket_relse(bucket
);
226 bucket
->bu_inode
= NULL
;
232 * A bucket that has never been written to disk doesn't need to be
233 * read. We just need the buffer_heads. Don't call this for
234 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
237 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
242 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
243 bucket
->bu_bhs
[i
] = sb_getblk(bucket
->bu_inode
->i_sb
,
245 if (!bucket
->bu_bhs
[i
]) {
251 if (!ocfs2_buffer_uptodate(bucket
->bu_inode
,
253 ocfs2_set_new_buffer_uptodate(bucket
->bu_inode
,
258 ocfs2_xattr_bucket_relse(bucket
);
262 /* Read the xattr bucket at xb_blkno */
263 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
268 rc
= ocfs2_read_blocks(bucket
->bu_inode
, xb_blkno
,
269 bucket
->bu_blocks
, bucket
->bu_bhs
, 0,
272 ocfs2_xattr_bucket_relse(bucket
);
276 static int ocfs2_xattr_bucket_journal_access(handle_t
*handle
,
277 struct ocfs2_xattr_bucket
*bucket
,
282 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
283 rc
= ocfs2_journal_access(handle
, bucket
->bu_inode
,
284 bucket
->bu_bhs
[i
], type
);
294 static void ocfs2_xattr_bucket_journal_dirty(handle_t
*handle
,
295 struct ocfs2_xattr_bucket
*bucket
)
299 for (i
= 0; i
< bucket
->bu_blocks
; i
++)
300 ocfs2_journal_dirty(handle
, bucket
->bu_bhs
[i
]);
303 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket
*dest
,
304 struct ocfs2_xattr_bucket
*src
)
307 int blocksize
= src
->bu_inode
->i_sb
->s_blocksize
;
309 BUG_ON(dest
->bu_blocks
!= src
->bu_blocks
);
310 BUG_ON(dest
->bu_inode
!= src
->bu_inode
);
312 for (i
= 0; i
< src
->bu_blocks
; i
++) {
313 memcpy(bucket_block(dest
, i
), bucket_block(src
, i
),
318 static int ocfs2_validate_xattr_block(struct super_block
*sb
,
319 struct buffer_head
*bh
)
321 struct ocfs2_xattr_block
*xb
=
322 (struct ocfs2_xattr_block
*)bh
->b_data
;
324 mlog(0, "Validating xattr block %llu\n",
325 (unsigned long long)bh
->b_blocknr
);
327 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
329 "Extended attribute block #%llu has bad "
331 (unsigned long long)bh
->b_blocknr
, 7,
336 if (le64_to_cpu(xb
->xb_blkno
) != bh
->b_blocknr
) {
338 "Extended attribute block #%llu has an "
339 "invalid xb_blkno of %llu",
340 (unsigned long long)bh
->b_blocknr
,
341 (unsigned long long)le64_to_cpu(xb
->xb_blkno
));
345 if (le32_to_cpu(xb
->xb_fs_generation
) != OCFS2_SB(sb
)->fs_generation
) {
347 "Extended attribute block #%llu has an invalid "
348 "xb_fs_generation of #%u",
349 (unsigned long long)bh
->b_blocknr
,
350 le32_to_cpu(xb
->xb_fs_generation
));
357 static int ocfs2_read_xattr_block(struct inode
*inode
, u64 xb_blkno
,
358 struct buffer_head
**bh
)
361 struct buffer_head
*tmp
= *bh
;
363 rc
= ocfs2_read_block(inode
, xb_blkno
, &tmp
,
364 ocfs2_validate_xattr_block
);
366 /* If ocfs2_read_block() got us a new bh, pass it up. */
373 static inline const char *ocfs2_xattr_prefix(int name_index
)
375 struct xattr_handler
*handler
= NULL
;
377 if (name_index
> 0 && name_index
< OCFS2_XATTR_MAX
)
378 handler
= ocfs2_xattr_handler_map
[name_index
];
380 return handler
? handler
->prefix
: NULL
;
383 static u32
ocfs2_xattr_name_hash(struct inode
*inode
,
387 /* Get hash value of uuid from super block */
388 u32 hash
= OCFS2_SB(inode
->i_sb
)->uuid_hash
;
391 /* hash extended attribute name */
392 for (i
= 0; i
< name_len
; i
++) {
393 hash
= (hash
<< OCFS2_HASH_SHIFT
) ^
394 (hash
>> (8*sizeof(hash
) - OCFS2_HASH_SHIFT
)) ^
402 * ocfs2_xattr_hash_entry()
404 * Compute the hash of an extended attribute.
406 static void ocfs2_xattr_hash_entry(struct inode
*inode
,
407 struct ocfs2_xattr_header
*header
,
408 struct ocfs2_xattr_entry
*entry
)
411 char *name
= (char *)header
+ le16_to_cpu(entry
->xe_name_offset
);
413 hash
= ocfs2_xattr_name_hash(inode
, name
, entry
->xe_name_len
);
414 entry
->xe_name_hash
= cpu_to_le32(hash
);
419 static int ocfs2_xattr_entry_real_size(int name_len
, size_t value_len
)
423 if (value_len
<= OCFS2_XATTR_INLINE_SIZE
)
424 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(value_len
);
426 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
427 size
+= sizeof(struct ocfs2_xattr_entry
);
432 int ocfs2_calc_security_init(struct inode
*dir
,
433 struct ocfs2_security_xattr_info
*si
,
436 struct ocfs2_alloc_context
**xattr_ac
)
439 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
440 int s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
444 * The max space of security xattr taken inline is
445 * 256(name) + 80(value) + 16(entry) = 352 bytes,
446 * So reserve one metadata block for it is ok.
448 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
449 s_size
> OCFS2_XATTR_FREE_IN_IBODY
) {
450 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
455 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
458 /* reserve clusters for xattr value which will be set in B tree*/
459 if (si
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
460 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
,
465 int ocfs2_calc_xattr_init(struct inode
*dir
,
466 struct buffer_head
*dir_bh
,
468 struct ocfs2_security_xattr_info
*si
,
471 struct ocfs2_alloc_context
**xattr_ac
)
474 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
480 s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
483 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
) {
484 acl_len
= ocfs2_xattr_get_nolock(dir
, dir_bh
,
485 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
,
488 a_size
= ocfs2_xattr_entry_real_size(0, acl_len
);
491 } else if (acl_len
!= 0 && acl_len
!= -ENODATA
) {
497 if (!(s_size
+ a_size
))
501 * The max space of security xattr taken inline is
502 * 256(name) + 80(value) + 16(entry) = 352 bytes,
503 * The max space of acl xattr taken inline is
504 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
505 * when blocksize = 512, may reserve one more cluser for
506 * xattr bucket, otherwise reserve one metadata block
509 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
510 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_IBODY
) {
511 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
516 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
519 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
&&
520 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_BLOCK(dir
)) {
522 *xattr_credits
+= ocfs2_blocks_per_xattr_bucket(dir
->i_sb
);
525 /* reserve clusters for xattr value which will be set in B tree*/
526 if (si
->enable
&& si
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
527 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
,
529 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
&&
530 acl_len
> OCFS2_XATTR_INLINE_SIZE
) {
531 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
, acl_len
);
533 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
,
540 static int ocfs2_xattr_extend_allocation(struct inode
*inode
,
542 struct buffer_head
*xattr_bh
,
543 struct ocfs2_xattr_value_root
*xv
,
544 struct ocfs2_xattr_set_ctxt
*ctxt
)
547 handle_t
*handle
= ctxt
->handle
;
548 enum ocfs2_alloc_restarted why
;
549 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
550 u32 prev_clusters
, logical_start
= le32_to_cpu(xv
->xr_clusters
);
551 struct ocfs2_extent_tree et
;
553 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add
);
555 ocfs2_init_xattr_value_extent_tree(&et
, inode
, xattr_bh
, xv
);
557 status
= ocfs2_journal_access(handle
, inode
, xattr_bh
,
558 OCFS2_JOURNAL_ACCESS_WRITE
);
564 prev_clusters
= le32_to_cpu(xv
->xr_clusters
);
565 status
= ocfs2_add_clusters_in_btree(osb
,
580 status
= ocfs2_journal_dirty(handle
, xattr_bh
);
586 clusters_to_add
-= le32_to_cpu(xv
->xr_clusters
) - prev_clusters
;
589 * We should have already allocated enough space before the transaction,
590 * so no need to restart.
592 BUG_ON(why
!= RESTART_NONE
|| clusters_to_add
);
599 static int __ocfs2_remove_xattr_range(struct inode
*inode
,
600 struct buffer_head
*root_bh
,
601 struct ocfs2_xattr_value_root
*xv
,
602 u32 cpos
, u32 phys_cpos
, u32 len
,
603 struct ocfs2_xattr_set_ctxt
*ctxt
)
606 u64 phys_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
607 handle_t
*handle
= ctxt
->handle
;
608 struct ocfs2_extent_tree et
;
610 ocfs2_init_xattr_value_extent_tree(&et
, inode
, root_bh
, xv
);
612 ret
= ocfs2_journal_access(handle
, inode
, root_bh
,
613 OCFS2_JOURNAL_ACCESS_WRITE
);
619 ret
= ocfs2_remove_extent(inode
, &et
, cpos
, len
, handle
, ctxt
->meta_ac
,
626 le32_add_cpu(&xv
->xr_clusters
, -len
);
628 ret
= ocfs2_journal_dirty(handle
, root_bh
);
634 ret
= ocfs2_cache_cluster_dealloc(&ctxt
->dealloc
, phys_blkno
, len
);
642 static int ocfs2_xattr_shrink_size(struct inode
*inode
,
645 struct buffer_head
*root_bh
,
646 struct ocfs2_xattr_value_root
*xv
,
647 struct ocfs2_xattr_set_ctxt
*ctxt
)
650 u32 trunc_len
, cpos
, phys_cpos
, alloc_size
;
653 if (old_clusters
<= new_clusters
)
657 trunc_len
= old_clusters
- new_clusters
;
659 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &phys_cpos
,
660 &alloc_size
, &xv
->xr_list
);
666 if (alloc_size
> trunc_len
)
667 alloc_size
= trunc_len
;
669 ret
= __ocfs2_remove_xattr_range(inode
, root_bh
, xv
, cpos
,
670 phys_cpos
, alloc_size
,
677 block
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
678 ocfs2_remove_xattr_clusters_from_cache(inode
, block
,
681 trunc_len
-= alloc_size
;
688 static int ocfs2_xattr_value_truncate(struct inode
*inode
,
689 struct buffer_head
*root_bh
,
690 struct ocfs2_xattr_value_root
*xv
,
692 struct ocfs2_xattr_set_ctxt
*ctxt
)
695 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
696 u32 old_clusters
= le32_to_cpu(xv
->xr_clusters
);
698 if (new_clusters
== old_clusters
)
701 if (new_clusters
> old_clusters
)
702 ret
= ocfs2_xattr_extend_allocation(inode
,
703 new_clusters
- old_clusters
,
706 ret
= ocfs2_xattr_shrink_size(inode
,
707 old_clusters
, new_clusters
,
713 static int ocfs2_xattr_list_entry(char *buffer
, size_t size
,
714 size_t *result
, const char *prefix
,
715 const char *name
, int name_len
)
717 char *p
= buffer
+ *result
;
718 int prefix_len
= strlen(prefix
);
719 int total_len
= prefix_len
+ name_len
+ 1;
721 *result
+= total_len
;
723 /* we are just looking for how big our buffer needs to be */
730 memcpy(p
, prefix
, prefix_len
);
731 memcpy(p
+ prefix_len
, name
, name_len
);
732 p
[prefix_len
+ name_len
] = '\0';
737 static int ocfs2_xattr_list_entries(struct inode
*inode
,
738 struct ocfs2_xattr_header
*header
,
739 char *buffer
, size_t buffer_size
)
743 const char *prefix
, *name
;
745 for (i
= 0 ; i
< le16_to_cpu(header
->xh_count
); i
++) {
746 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
747 type
= ocfs2_xattr_get_type(entry
);
748 prefix
= ocfs2_xattr_prefix(type
);
751 name
= (const char *)header
+
752 le16_to_cpu(entry
->xe_name_offset
);
754 ret
= ocfs2_xattr_list_entry(buffer
, buffer_size
,
755 &result
, prefix
, name
,
765 static int ocfs2_xattr_ibody_list(struct inode
*inode
,
766 struct ocfs2_dinode
*di
,
770 struct ocfs2_xattr_header
*header
= NULL
;
771 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
774 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
777 header
= (struct ocfs2_xattr_header
*)
778 ((void *)di
+ inode
->i_sb
->s_blocksize
-
779 le16_to_cpu(di
->i_xattr_inline_size
));
781 ret
= ocfs2_xattr_list_entries(inode
, header
, buffer
, buffer_size
);
786 static int ocfs2_xattr_block_list(struct inode
*inode
,
787 struct ocfs2_dinode
*di
,
791 struct buffer_head
*blk_bh
= NULL
;
792 struct ocfs2_xattr_block
*xb
;
795 if (!di
->i_xattr_loc
)
798 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
805 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
806 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
807 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
808 ret
= ocfs2_xattr_list_entries(inode
, header
,
809 buffer
, buffer_size
);
811 struct ocfs2_xattr_tree_root
*xt
= &xb
->xb_attrs
.xb_root
;
812 ret
= ocfs2_xattr_tree_list_index_block(inode
, xt
,
813 buffer
, buffer_size
);
821 ssize_t
ocfs2_listxattr(struct dentry
*dentry
,
825 int ret
= 0, i_ret
= 0, b_ret
= 0;
826 struct buffer_head
*di_bh
= NULL
;
827 struct ocfs2_dinode
*di
= NULL
;
828 struct ocfs2_inode_info
*oi
= OCFS2_I(dentry
->d_inode
);
830 if (!ocfs2_supports_xattr(OCFS2_SB(dentry
->d_sb
)))
833 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
836 ret
= ocfs2_inode_lock(dentry
->d_inode
, &di_bh
, 0);
842 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
844 down_read(&oi
->ip_xattr_sem
);
845 i_ret
= ocfs2_xattr_ibody_list(dentry
->d_inode
, di
, buffer
, size
);
853 b_ret
= ocfs2_xattr_block_list(dentry
->d_inode
, di
,
858 up_read(&oi
->ip_xattr_sem
);
859 ocfs2_inode_unlock(dentry
->d_inode
, 0);
863 return i_ret
+ b_ret
;
866 static int ocfs2_xattr_find_entry(int name_index
,
868 struct ocfs2_xattr_search
*xs
)
870 struct ocfs2_xattr_entry
*entry
;
877 name_len
= strlen(name
);
879 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
880 cmp
= name_index
- ocfs2_xattr_get_type(entry
);
882 cmp
= name_len
- entry
->xe_name_len
;
884 cmp
= memcmp(name
, (xs
->base
+
885 le16_to_cpu(entry
->xe_name_offset
)),
893 return cmp
? -ENODATA
: 0;
896 static int ocfs2_xattr_get_value_outside(struct inode
*inode
,
897 struct ocfs2_xattr_value_root
*xv
,
901 u32 cpos
, p_cluster
, num_clusters
, bpc
, clusters
;
904 size_t cplen
, blocksize
;
905 struct buffer_head
*bh
= NULL
;
906 struct ocfs2_extent_list
*el
;
909 clusters
= le32_to_cpu(xv
->xr_clusters
);
910 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
911 blocksize
= inode
->i_sb
->s_blocksize
;
914 while (cpos
< clusters
) {
915 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
922 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
923 /* Copy ocfs2_xattr_value */
924 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
925 ret
= ocfs2_read_block(inode
, blkno
, &bh
, NULL
);
931 cplen
= len
>= blocksize
? blocksize
: len
;
932 memcpy(buffer
, bh
->b_data
, cplen
);
941 cpos
+= num_clusters
;
947 static int ocfs2_xattr_ibody_get(struct inode
*inode
,
952 struct ocfs2_xattr_search
*xs
)
954 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
955 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
956 struct ocfs2_xattr_value_root
*xv
;
960 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
963 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
964 xs
->header
= (struct ocfs2_xattr_header
*)
965 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
966 xs
->base
= (void *)xs
->header
;
967 xs
->here
= xs
->header
->xh_entries
;
969 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
972 size
= le64_to_cpu(xs
->here
->xe_value_size
);
974 if (size
> buffer_size
)
976 if (ocfs2_xattr_is_local(xs
->here
)) {
977 memcpy(buffer
, (void *)xs
->base
+
978 le16_to_cpu(xs
->here
->xe_name_offset
) +
979 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
), size
);
981 xv
= (struct ocfs2_xattr_value_root
*)
982 (xs
->base
+ le16_to_cpu(
983 xs
->here
->xe_name_offset
) +
984 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
));
985 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
997 static int ocfs2_xattr_block_get(struct inode
*inode
,
1002 struct ocfs2_xattr_search
*xs
)
1004 struct ocfs2_xattr_block
*xb
;
1005 struct ocfs2_xattr_value_root
*xv
;
1007 int ret
= -ENODATA
, name_offset
, name_len
, block_off
, i
;
1009 xs
->bucket
= ocfs2_xattr_bucket_new(inode
);
1016 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, xs
);
1022 if (xs
->not_found
) {
1027 xb
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
1028 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1031 if (size
> buffer_size
)
1034 name_offset
= le16_to_cpu(xs
->here
->xe_name_offset
);
1035 name_len
= OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
);
1036 i
= xs
->here
- xs
->header
->xh_entries
;
1038 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
1039 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
1040 bucket_xh(xs
->bucket
),
1044 xs
->base
= bucket_block(xs
->bucket
, block_off
);
1046 if (ocfs2_xattr_is_local(xs
->here
)) {
1047 memcpy(buffer
, (void *)xs
->base
+
1048 name_offset
+ name_len
, size
);
1050 xv
= (struct ocfs2_xattr_value_root
*)
1051 (xs
->base
+ name_offset
+ name_len
);
1052 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1062 ocfs2_xattr_bucket_free(xs
->bucket
);
1064 brelse(xs
->xattr_bh
);
1065 xs
->xattr_bh
= NULL
;
1069 int ocfs2_xattr_get_nolock(struct inode
*inode
,
1070 struct buffer_head
*di_bh
,
1077 struct ocfs2_dinode
*di
= NULL
;
1078 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1079 struct ocfs2_xattr_search xis
= {
1080 .not_found
= -ENODATA
,
1082 struct ocfs2_xattr_search xbs
= {
1083 .not_found
= -ENODATA
,
1086 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1089 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1092 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
1093 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1095 down_read(&oi
->ip_xattr_sem
);
1096 ret
= ocfs2_xattr_ibody_get(inode
, name_index
, name
, buffer
,
1098 if (ret
== -ENODATA
&& di
->i_xattr_loc
)
1099 ret
= ocfs2_xattr_block_get(inode
, name_index
, name
, buffer
,
1101 up_read(&oi
->ip_xattr_sem
);
1106 /* ocfs2_xattr_get()
1108 * Copy an extended attribute into the buffer provided.
1109 * Buffer is NULL to compute the size of buffer required.
1111 static int ocfs2_xattr_get(struct inode
*inode
,
1118 struct buffer_head
*di_bh
= NULL
;
1120 ret
= ocfs2_inode_lock(inode
, &di_bh
, 0);
1125 ret
= ocfs2_xattr_get_nolock(inode
, di_bh
, name_index
,
1126 name
, buffer
, buffer_size
);
1128 ocfs2_inode_unlock(inode
, 0);
1135 static int __ocfs2_xattr_set_value_outside(struct inode
*inode
,
1137 struct ocfs2_xattr_value_root
*xv
,
1141 int ret
= 0, i
, cp_len
, credits
;
1142 u16 blocksize
= inode
->i_sb
->s_blocksize
;
1143 u32 p_cluster
, num_clusters
;
1144 u32 cpos
= 0, bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1145 u32 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, value_len
);
1147 struct buffer_head
*bh
= NULL
;
1149 BUG_ON(clusters
> le32_to_cpu(xv
->xr_clusters
));
1152 * In __ocfs2_xattr_set_value_outside has already been dirtied,
1153 * so we don't need to worry about whether ocfs2_extend_trans
1154 * will create a new transactio for us or not.
1156 credits
= clusters
* bpc
;
1157 ret
= ocfs2_extend_trans(handle
, credits
);
1163 while (cpos
< clusters
) {
1164 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1165 &num_clusters
, &xv
->xr_list
);
1171 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1173 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1174 ret
= ocfs2_read_block(inode
, blkno
, &bh
, NULL
);
1180 ret
= ocfs2_journal_access(handle
,
1183 OCFS2_JOURNAL_ACCESS_WRITE
);
1189 cp_len
= value_len
> blocksize
? blocksize
: value_len
;
1190 memcpy(bh
->b_data
, value
, cp_len
);
1191 value_len
-= cp_len
;
1193 if (cp_len
< blocksize
)
1194 memset(bh
->b_data
+ cp_len
, 0,
1195 blocksize
- cp_len
);
1197 ret
= ocfs2_journal_dirty(handle
, bh
);
1206 * XXX: do we need to empty all the following
1207 * blocks in this cluster?
1212 cpos
+= num_clusters
;
1220 static int ocfs2_xattr_cleanup(struct inode
*inode
,
1222 struct ocfs2_xattr_info
*xi
,
1223 struct ocfs2_xattr_search
*xs
,
1227 size_t name_len
= strlen(xi
->name
);
1228 void *val
= xs
->base
+ offs
;
1229 size_t size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1231 ret
= ocfs2_journal_access(handle
, inode
, xs
->xattr_bh
,
1232 OCFS2_JOURNAL_ACCESS_WRITE
);
1237 /* Decrease xattr count */
1238 le16_add_cpu(&xs
->header
->xh_count
, -1);
1239 /* Remove the xattr entry and tree root which has already be set*/
1240 memset((void *)xs
->here
, 0, sizeof(struct ocfs2_xattr_entry
));
1241 memset(val
, 0, size
);
1243 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1250 static int ocfs2_xattr_update_entry(struct inode
*inode
,
1252 struct ocfs2_xattr_info
*xi
,
1253 struct ocfs2_xattr_search
*xs
,
1258 ret
= ocfs2_journal_access(handle
, inode
, xs
->xattr_bh
,
1259 OCFS2_JOURNAL_ACCESS_WRITE
);
1265 xs
->here
->xe_name_offset
= cpu_to_le16(offs
);
1266 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1267 if (xi
->value_len
<= OCFS2_XATTR_INLINE_SIZE
)
1268 ocfs2_xattr_set_local(xs
->here
, 1);
1270 ocfs2_xattr_set_local(xs
->here
, 0);
1271 ocfs2_xattr_hash_entry(inode
, xs
->header
, xs
->here
);
1273 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1281 * ocfs2_xattr_set_value_outside()
1283 * Set large size value in B tree.
1285 static int ocfs2_xattr_set_value_outside(struct inode
*inode
,
1286 struct ocfs2_xattr_info
*xi
,
1287 struct ocfs2_xattr_search
*xs
,
1288 struct ocfs2_xattr_set_ctxt
*ctxt
,
1291 size_t name_len
= strlen(xi
->name
);
1292 void *val
= xs
->base
+ offs
;
1293 struct ocfs2_xattr_value_root
*xv
= NULL
;
1294 size_t size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1297 memset(val
, 0, size
);
1298 memcpy(val
, xi
->name
, name_len
);
1299 xv
= (struct ocfs2_xattr_value_root
*)
1300 (val
+ OCFS2_XATTR_SIZE(name_len
));
1301 xv
->xr_clusters
= 0;
1302 xv
->xr_last_eb_blk
= 0;
1303 xv
->xr_list
.l_tree_depth
= 0;
1304 xv
->xr_list
.l_count
= cpu_to_le16(1);
1305 xv
->xr_list
.l_next_free_rec
= 0;
1307 ret
= ocfs2_xattr_value_truncate(inode
, xs
->xattr_bh
, xv
,
1308 xi
->value_len
, ctxt
);
1313 ret
= ocfs2_xattr_update_entry(inode
, ctxt
->handle
, xi
, xs
, offs
);
1318 ret
= __ocfs2_xattr_set_value_outside(inode
, ctxt
->handle
, xv
,
1319 xi
->value
, xi
->value_len
);
1327 * ocfs2_xattr_set_entry_local()
1329 * Set, replace or remove extended attribute in local.
1331 static void ocfs2_xattr_set_entry_local(struct inode
*inode
,
1332 struct ocfs2_xattr_info
*xi
,
1333 struct ocfs2_xattr_search
*xs
,
1334 struct ocfs2_xattr_entry
*last
,
1337 size_t name_len
= strlen(xi
->name
);
1340 if (xi
->value
&& xs
->not_found
) {
1341 /* Insert the new xattr entry. */
1342 le16_add_cpu(&xs
->header
->xh_count
, 1);
1343 ocfs2_xattr_set_type(last
, xi
->name_index
);
1344 ocfs2_xattr_set_local(last
, 1);
1345 last
->xe_name_len
= name_len
;
1351 first_val
= xs
->base
+ min_offs
;
1352 offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1353 val
= xs
->base
+ offs
;
1355 if (le64_to_cpu(xs
->here
->xe_value_size
) >
1356 OCFS2_XATTR_INLINE_SIZE
)
1357 size
= OCFS2_XATTR_SIZE(name_len
) +
1358 OCFS2_XATTR_ROOT_SIZE
;
1360 size
= OCFS2_XATTR_SIZE(name_len
) +
1361 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1363 if (xi
->value
&& size
== OCFS2_XATTR_SIZE(name_len
) +
1364 OCFS2_XATTR_SIZE(xi
->value_len
)) {
1365 /* The old and the new value have the
1366 same size. Just replace the value. */
1367 ocfs2_xattr_set_local(xs
->here
, 1);
1368 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1369 /* Clear value bytes. */
1370 memset(val
+ OCFS2_XATTR_SIZE(name_len
),
1372 OCFS2_XATTR_SIZE(xi
->value_len
));
1373 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
1378 /* Remove the old name+value. */
1379 memmove(first_val
+ size
, first_val
, val
- first_val
);
1380 memset(first_val
, 0, size
);
1381 xs
->here
->xe_name_hash
= 0;
1382 xs
->here
->xe_name_offset
= 0;
1383 ocfs2_xattr_set_local(xs
->here
, 1);
1384 xs
->here
->xe_value_size
= 0;
1388 /* Adjust all value offsets. */
1389 last
= xs
->header
->xh_entries
;
1390 for (i
= 0 ; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1391 size_t o
= le16_to_cpu(last
->xe_name_offset
);
1394 last
->xe_name_offset
= cpu_to_le16(o
+ size
);
1399 /* Remove the old entry. */
1401 memmove(xs
->here
, xs
->here
+ 1,
1402 (void *)last
- (void *)xs
->here
);
1403 memset(last
, 0, sizeof(struct ocfs2_xattr_entry
));
1404 le16_add_cpu(&xs
->header
->xh_count
, -1);
1408 /* Insert the new name+value. */
1409 size_t size
= OCFS2_XATTR_SIZE(name_len
) +
1410 OCFS2_XATTR_SIZE(xi
->value_len
);
1411 void *val
= xs
->base
+ min_offs
- size
;
1413 xs
->here
->xe_name_offset
= cpu_to_le16(min_offs
- size
);
1414 memset(val
, 0, size
);
1415 memcpy(val
, xi
->name
, name_len
);
1416 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
1419 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1420 ocfs2_xattr_set_local(xs
->here
, 1);
1421 ocfs2_xattr_hash_entry(inode
, xs
->header
, xs
->here
);
1428 * ocfs2_xattr_set_entry()
1430 * Set extended attribute entry into inode or block.
1432 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1433 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1434 * then set value in B tree with set_value_outside().
1436 static int ocfs2_xattr_set_entry(struct inode
*inode
,
1437 struct ocfs2_xattr_info
*xi
,
1438 struct ocfs2_xattr_search
*xs
,
1439 struct ocfs2_xattr_set_ctxt
*ctxt
,
1442 struct ocfs2_xattr_entry
*last
;
1443 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1444 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1445 size_t min_offs
= xs
->end
- xs
->base
, name_len
= strlen(xi
->name
);
1447 handle_t
*handle
= ctxt
->handle
;
1449 struct ocfs2_xattr_info xi_l
= {
1450 .name_index
= xi
->name_index
,
1453 .value_len
= xi
->value_len
,
1456 /* Compute min_offs, last and free space. */
1457 last
= xs
->header
->xh_entries
;
1459 for (i
= 0 ; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1460 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
1461 if (offs
< min_offs
)
1466 free
= min_offs
- ((void *)last
- xs
->base
) - sizeof(__u32
);
1470 if (!xs
->not_found
) {
1472 if (ocfs2_xattr_is_local(xs
->here
))
1473 size
= OCFS2_XATTR_SIZE(name_len
) +
1474 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1476 size
= OCFS2_XATTR_SIZE(name_len
) +
1477 OCFS2_XATTR_ROOT_SIZE
;
1478 free
+= (size
+ sizeof(struct ocfs2_xattr_entry
));
1480 /* Check free space in inode or block */
1481 if (xi
->value
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1482 if (free
< sizeof(struct ocfs2_xattr_entry
) +
1483 OCFS2_XATTR_SIZE(name_len
) +
1484 OCFS2_XATTR_ROOT_SIZE
) {
1488 size_l
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1489 xi_l
.value
= (void *)&def_xv
;
1490 xi_l
.value_len
= OCFS2_XATTR_ROOT_SIZE
;
1491 } else if (xi
->value
) {
1492 if (free
< sizeof(struct ocfs2_xattr_entry
) +
1493 OCFS2_XATTR_SIZE(name_len
) +
1494 OCFS2_XATTR_SIZE(xi
->value_len
)) {
1500 if (!xs
->not_found
) {
1501 /* For existing extended attribute */
1502 size_t size
= OCFS2_XATTR_SIZE(name_len
) +
1503 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1504 size_t offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1505 void *val
= xs
->base
+ offs
;
1507 if (ocfs2_xattr_is_local(xs
->here
) && size
== size_l
) {
1508 /* Replace existing local xattr with tree root */
1509 ret
= ocfs2_xattr_set_value_outside(inode
, xi
, xs
,
1514 } else if (!ocfs2_xattr_is_local(xs
->here
)) {
1515 /* For existing xattr which has value outside */
1516 struct ocfs2_xattr_value_root
*xv
= NULL
;
1517 xv
= (struct ocfs2_xattr_value_root
*)(val
+
1518 OCFS2_XATTR_SIZE(name_len
));
1520 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1522 * If new value need set outside also,
1523 * first truncate old value to new value,
1524 * then set new value with set_value_outside().
1526 ret
= ocfs2_xattr_value_truncate(inode
,
1536 ret
= ocfs2_xattr_update_entry(inode
,
1546 ret
= __ocfs2_xattr_set_value_outside(inode
,
1556 * If new value need set in local,
1557 * just trucate old value to zero.
1559 ret
= ocfs2_xattr_value_truncate(inode
,
1570 ret
= ocfs2_journal_access(handle
, inode
, xs
->inode_bh
,
1571 OCFS2_JOURNAL_ACCESS_WRITE
);
1577 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1578 ret
= ocfs2_journal_access(handle
, inode
, xs
->xattr_bh
,
1579 OCFS2_JOURNAL_ACCESS_WRITE
);
1587 * Set value in local, include set tree root in local.
1588 * This is the first step for value size >INLINE_SIZE.
1590 ocfs2_xattr_set_entry_local(inode
, &xi_l
, xs
, last
, min_offs
);
1592 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1593 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1600 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) &&
1601 (flag
& OCFS2_INLINE_XATTR_FL
)) {
1602 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1603 unsigned int xattrsize
= osb
->s_xattr_inline_size
;
1606 * Adjust extent record count or inline data size
1607 * to reserve space for extended attribute.
1609 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1610 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
1611 le16_add_cpu(&idata
->id_count
, -xattrsize
);
1612 } else if (!(ocfs2_inode_is_fast_symlink(inode
))) {
1613 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
1614 le16_add_cpu(&el
->l_count
, -(xattrsize
/
1615 sizeof(struct ocfs2_extent_rec
)));
1617 di
->i_xattr_inline_size
= cpu_to_le16(xattrsize
);
1619 /* Update xattr flag */
1620 spin_lock(&oi
->ip_lock
);
1621 oi
->ip_dyn_features
|= flag
;
1622 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
1623 spin_unlock(&oi
->ip_lock
);
1624 /* Update inode ctime */
1625 inode
->i_ctime
= CURRENT_TIME
;
1626 di
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1627 di
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1629 ret
= ocfs2_journal_dirty(handle
, xs
->inode_bh
);
1633 if (!ret
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1635 * Set value outside in B tree.
1636 * This is the second step for value size > INLINE_SIZE.
1638 size_t offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1639 ret
= ocfs2_xattr_set_value_outside(inode
, xi
, xs
, ctxt
, offs
);
1645 * If set value outside failed, we have to clean
1646 * the junk tree root we have already set in local.
1648 ret2
= ocfs2_xattr_cleanup(inode
, ctxt
->handle
,
1658 static int ocfs2_remove_value_outside(struct inode
*inode
,
1659 struct buffer_head
*bh
,
1660 struct ocfs2_xattr_header
*header
)
1663 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1664 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
1666 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
1668 ctxt
.handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
1669 if (IS_ERR(ctxt
.handle
)) {
1670 ret
= PTR_ERR(ctxt
.handle
);
1675 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
1676 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
1678 if (!ocfs2_xattr_is_local(entry
)) {
1679 struct ocfs2_xattr_value_root
*xv
;
1682 val
= (void *)header
+
1683 le16_to_cpu(entry
->xe_name_offset
);
1684 xv
= (struct ocfs2_xattr_value_root
*)
1685 (val
+ OCFS2_XATTR_SIZE(entry
->xe_name_len
));
1686 ret
= ocfs2_xattr_value_truncate(inode
, bh
, xv
,
1695 ocfs2_commit_trans(osb
, ctxt
.handle
);
1696 ocfs2_schedule_truncate_log_flush(osb
, 1);
1697 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
1702 static int ocfs2_xattr_ibody_remove(struct inode
*inode
,
1703 struct buffer_head
*di_bh
)
1706 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1707 struct ocfs2_xattr_header
*header
;
1710 header
= (struct ocfs2_xattr_header
*)
1711 ((void *)di
+ inode
->i_sb
->s_blocksize
-
1712 le16_to_cpu(di
->i_xattr_inline_size
));
1714 ret
= ocfs2_remove_value_outside(inode
, di_bh
, header
);
1719 static int ocfs2_xattr_block_remove(struct inode
*inode
,
1720 struct buffer_head
*blk_bh
)
1722 struct ocfs2_xattr_block
*xb
;
1725 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1726 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
1727 struct ocfs2_xattr_header
*header
= &(xb
->xb_attrs
.xb_header
);
1728 ret
= ocfs2_remove_value_outside(inode
, blk_bh
, header
);
1730 ret
= ocfs2_delete_xattr_index_block(inode
, blk_bh
);
1735 static int ocfs2_xattr_free_block(struct inode
*inode
,
1738 struct inode
*xb_alloc_inode
;
1739 struct buffer_head
*xb_alloc_bh
= NULL
;
1740 struct buffer_head
*blk_bh
= NULL
;
1741 struct ocfs2_xattr_block
*xb
;
1742 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1748 ret
= ocfs2_read_xattr_block(inode
, block
, &blk_bh
);
1754 ret
= ocfs2_xattr_block_remove(inode
, blk_bh
);
1760 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1761 blk
= le64_to_cpu(xb
->xb_blkno
);
1762 bit
= le16_to_cpu(xb
->xb_suballoc_bit
);
1763 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
1765 xb_alloc_inode
= ocfs2_get_system_file_inode(osb
,
1766 EXTENT_ALLOC_SYSTEM_INODE
,
1767 le16_to_cpu(xb
->xb_suballoc_slot
));
1768 if (!xb_alloc_inode
) {
1773 mutex_lock(&xb_alloc_inode
->i_mutex
);
1775 ret
= ocfs2_inode_lock(xb_alloc_inode
, &xb_alloc_bh
, 1);
1781 handle
= ocfs2_start_trans(osb
, OCFS2_SUBALLOC_FREE
);
1782 if (IS_ERR(handle
)) {
1783 ret
= PTR_ERR(handle
);
1788 ret
= ocfs2_free_suballoc_bits(handle
, xb_alloc_inode
, xb_alloc_bh
,
1793 ocfs2_commit_trans(osb
, handle
);
1795 ocfs2_inode_unlock(xb_alloc_inode
, 1);
1796 brelse(xb_alloc_bh
);
1798 mutex_unlock(&xb_alloc_inode
->i_mutex
);
1799 iput(xb_alloc_inode
);
1806 * ocfs2_xattr_remove()
1808 * Free extended attribute resources associated with this inode.
1810 int ocfs2_xattr_remove(struct inode
*inode
, struct buffer_head
*di_bh
)
1812 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1813 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1817 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1820 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1823 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
1824 ret
= ocfs2_xattr_ibody_remove(inode
, di_bh
);
1831 if (di
->i_xattr_loc
) {
1832 ret
= ocfs2_xattr_free_block(inode
,
1833 le64_to_cpu(di
->i_xattr_loc
));
1840 handle
= ocfs2_start_trans((OCFS2_SB(inode
->i_sb
)),
1841 OCFS2_INODE_UPDATE_CREDITS
);
1842 if (IS_ERR(handle
)) {
1843 ret
= PTR_ERR(handle
);
1847 ret
= ocfs2_journal_access(handle
, inode
, di_bh
,
1848 OCFS2_JOURNAL_ACCESS_WRITE
);
1854 di
->i_xattr_loc
= 0;
1856 spin_lock(&oi
->ip_lock
);
1857 oi
->ip_dyn_features
&= ~(OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
);
1858 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
1859 spin_unlock(&oi
->ip_lock
);
1861 ret
= ocfs2_journal_dirty(handle
, di_bh
);
1865 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
1870 static int ocfs2_xattr_has_space_inline(struct inode
*inode
,
1871 struct ocfs2_dinode
*di
)
1873 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1874 unsigned int xattrsize
= OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
;
1877 if (xattrsize
< OCFS2_MIN_XATTR_INLINE_SIZE
)
1880 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1881 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
1882 free
= le16_to_cpu(idata
->id_count
) - le64_to_cpu(di
->i_size
);
1883 } else if (ocfs2_inode_is_fast_symlink(inode
)) {
1884 free
= ocfs2_fast_symlink_chars(inode
->i_sb
) -
1885 le64_to_cpu(di
->i_size
);
1887 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
1888 free
= (le16_to_cpu(el
->l_count
) -
1889 le16_to_cpu(el
->l_next_free_rec
)) *
1890 sizeof(struct ocfs2_extent_rec
);
1892 if (free
>= xattrsize
)
1899 * ocfs2_xattr_ibody_find()
1901 * Find extended attribute in inode block and
1902 * fill search info into struct ocfs2_xattr_search.
1904 static int ocfs2_xattr_ibody_find(struct inode
*inode
,
1907 struct ocfs2_xattr_search
*xs
)
1909 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1910 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1914 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
1917 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
1918 down_read(&oi
->ip_alloc_sem
);
1919 has_space
= ocfs2_xattr_has_space_inline(inode
, di
);
1920 up_read(&oi
->ip_alloc_sem
);
1925 xs
->xattr_bh
= xs
->inode_bh
;
1926 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
1927 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)
1928 xs
->header
= (struct ocfs2_xattr_header
*)
1929 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
1931 xs
->header
= (struct ocfs2_xattr_header
*)
1932 (xs
->end
- OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
);
1933 xs
->base
= (void *)xs
->header
;
1934 xs
->here
= xs
->header
->xh_entries
;
1936 /* Find the named attribute. */
1937 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
1938 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1939 if (ret
&& ret
!= -ENODATA
)
1941 xs
->not_found
= ret
;
1948 * ocfs2_xattr_ibody_set()
1950 * Set, replace or remove an extended attribute into inode block.
1953 static int ocfs2_xattr_ibody_set(struct inode
*inode
,
1954 struct ocfs2_xattr_info
*xi
,
1955 struct ocfs2_xattr_search
*xs
,
1956 struct ocfs2_xattr_set_ctxt
*ctxt
)
1958 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1959 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1962 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
1965 down_write(&oi
->ip_alloc_sem
);
1966 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
1967 if (!ocfs2_xattr_has_space_inline(inode
, di
)) {
1973 ret
= ocfs2_xattr_set_entry(inode
, xi
, xs
, ctxt
,
1974 (OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
));
1976 up_write(&oi
->ip_alloc_sem
);
1982 * ocfs2_xattr_block_find()
1984 * Find extended attribute in external block and
1985 * fill search info into struct ocfs2_xattr_search.
1987 static int ocfs2_xattr_block_find(struct inode
*inode
,
1990 struct ocfs2_xattr_search
*xs
)
1992 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1993 struct buffer_head
*blk_bh
= NULL
;
1994 struct ocfs2_xattr_block
*xb
;
1997 if (!di
->i_xattr_loc
)
2000 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
2007 xs
->xattr_bh
= blk_bh
;
2008 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2010 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2011 xs
->header
= &xb
->xb_attrs
.xb_header
;
2012 xs
->base
= (void *)xs
->header
;
2013 xs
->end
= (void *)(blk_bh
->b_data
) + blk_bh
->b_size
;
2014 xs
->here
= xs
->header
->xh_entries
;
2016 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
2018 ret
= ocfs2_xattr_index_block_find(inode
, blk_bh
,
2022 if (ret
&& ret
!= -ENODATA
) {
2023 xs
->xattr_bh
= NULL
;
2026 xs
->not_found
= ret
;
2035 * ocfs2_xattr_block_set()
2037 * Set, replace or remove an extended attribute into external block.
2040 static int ocfs2_xattr_block_set(struct inode
*inode
,
2041 struct ocfs2_xattr_info
*xi
,
2042 struct ocfs2_xattr_search
*xs
,
2043 struct ocfs2_xattr_set_ctxt
*ctxt
)
2045 struct buffer_head
*new_bh
= NULL
;
2046 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2047 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2048 handle_t
*handle
= ctxt
->handle
;
2049 struct ocfs2_xattr_block
*xblk
= NULL
;
2050 u16 suballoc_bit_start
;
2055 if (!xs
->xattr_bh
) {
2056 ret
= ocfs2_journal_access(handle
, inode
, xs
->inode_bh
,
2057 OCFS2_JOURNAL_ACCESS_CREATE
);
2063 ret
= ocfs2_claim_metadata(osb
, handle
, ctxt
->meta_ac
, 1,
2064 &suballoc_bit_start
, &num_got
,
2071 new_bh
= sb_getblk(inode
->i_sb
, first_blkno
);
2072 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
2074 ret
= ocfs2_journal_access(handle
, inode
, new_bh
,
2075 OCFS2_JOURNAL_ACCESS_CREATE
);
2081 /* Initialize ocfs2_xattr_block */
2082 xs
->xattr_bh
= new_bh
;
2083 xblk
= (struct ocfs2_xattr_block
*)new_bh
->b_data
;
2084 memset(xblk
, 0, inode
->i_sb
->s_blocksize
);
2085 strcpy((void *)xblk
, OCFS2_XATTR_BLOCK_SIGNATURE
);
2086 xblk
->xb_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
2087 xblk
->xb_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
2088 xblk
->xb_fs_generation
= cpu_to_le32(osb
->fs_generation
);
2089 xblk
->xb_blkno
= cpu_to_le64(first_blkno
);
2091 xs
->header
= &xblk
->xb_attrs
.xb_header
;
2092 xs
->base
= (void *)xs
->header
;
2093 xs
->end
= (void *)xblk
+ inode
->i_sb
->s_blocksize
;
2094 xs
->here
= xs
->header
->xh_entries
;
2096 ret
= ocfs2_journal_dirty(handle
, new_bh
);
2101 di
->i_xattr_loc
= cpu_to_le64(first_blkno
);
2102 ocfs2_journal_dirty(handle
, xs
->inode_bh
);
2104 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2106 if (!(le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2107 /* Set extended attribute into external block */
2108 ret
= ocfs2_xattr_set_entry(inode
, xi
, xs
, ctxt
,
2109 OCFS2_HAS_XATTR_FL
);
2110 if (!ret
|| ret
!= -ENOSPC
)
2113 ret
= ocfs2_xattr_create_index_block(inode
, xs
, ctxt
);
2118 ret
= ocfs2_xattr_set_entry_index_block(inode
, xi
, xs
, ctxt
);
2125 /* Check whether the new xattr can be inserted into the inode. */
2126 static int ocfs2_xattr_can_be_in_inode(struct inode
*inode
,
2127 struct ocfs2_xattr_info
*xi
,
2128 struct ocfs2_xattr_search
*xs
)
2131 struct ocfs2_xattr_entry
*last
;
2133 size_t min_offs
= xs
->end
- xs
->base
;
2138 last
= xs
->header
->xh_entries
;
2140 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
2141 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
2142 if (offs
< min_offs
)
2147 free
= min_offs
- ((void *)last
- xs
->base
) - sizeof(__u32
);
2151 BUG_ON(!xs
->not_found
);
2153 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
2154 value_size
= OCFS2_XATTR_ROOT_SIZE
;
2156 value_size
= OCFS2_XATTR_SIZE(xi
->value_len
);
2158 if (free
>= sizeof(struct ocfs2_xattr_entry
) +
2159 OCFS2_XATTR_SIZE(strlen(xi
->name
)) + value_size
)
2165 static int ocfs2_calc_xattr_set_need(struct inode
*inode
,
2166 struct ocfs2_dinode
*di
,
2167 struct ocfs2_xattr_info
*xi
,
2168 struct ocfs2_xattr_search
*xis
,
2169 struct ocfs2_xattr_search
*xbs
,
2174 int ret
= 0, old_in_xb
= 0;
2175 int clusters_add
= 0, meta_add
= 0, credits
= 0;
2176 struct buffer_head
*bh
= NULL
;
2177 struct ocfs2_xattr_block
*xb
= NULL
;
2178 struct ocfs2_xattr_entry
*xe
= NULL
;
2179 struct ocfs2_xattr_value_root
*xv
= NULL
;
2181 int name_offset
, name_len
= 0;
2182 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
2186 if (xis
->not_found
&& xbs
->not_found
) {
2187 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2189 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2190 clusters_add
+= new_clusters
;
2191 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2199 if (!xis
->not_found
) {
2201 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
2202 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
2204 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
2206 int i
, block_off
= 0;
2207 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
2209 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
2210 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
2211 i
= xbs
->here
- xbs
->header
->xh_entries
;
2214 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
2215 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
2216 bucket_xh(xbs
->bucket
),
2219 base
= bucket_block(xbs
->bucket
, block_off
);
2220 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2223 credits
+= OCFS2_XATTR_BLOCK_UPDATE_CREDITS
;
2228 * delete a xattr doesn't need metadata and cluster allocation.
2229 * so just calculate the credits and return.
2231 * The credits for removing the value tree will be extended
2232 * by ocfs2_remove_extent itself.
2235 if (!ocfs2_xattr_is_local(xe
))
2236 credits
+= OCFS2_REMOVE_EXTENT_CREDITS
;
2241 /* do cluster allocation guess first. */
2242 value_size
= le64_to_cpu(xe
->xe_value_size
);
2246 * In xattr set, we always try to set the xe in inode first,
2247 * so if it can be inserted into inode successfully, the old
2248 * one will be removed from the xattr block, and this xattr
2249 * will be inserted into inode as a new xattr in inode.
2251 if (ocfs2_xattr_can_be_in_inode(inode
, xi
, xis
)) {
2252 clusters_add
+= new_clusters
;
2253 credits
+= OCFS2_REMOVE_EXTENT_CREDITS
+
2254 OCFS2_INODE_UPDATE_CREDITS
;
2255 if (!ocfs2_xattr_is_local(xe
))
2256 credits
+= ocfs2_calc_extend_credits(
2264 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2265 /* the new values will be stored outside. */
2266 u32 old_clusters
= 0;
2268 if (!ocfs2_xattr_is_local(xe
)) {
2269 old_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
2271 xv
= (struct ocfs2_xattr_value_root
*)
2272 (base
+ name_offset
+ name_len
);
2273 value_size
= OCFS2_XATTR_ROOT_SIZE
;
2277 if (old_clusters
>= new_clusters
) {
2278 credits
+= OCFS2_REMOVE_EXTENT_CREDITS
;
2281 meta_add
+= ocfs2_extend_meta_needed(&xv
->xr_list
);
2282 clusters_add
+= new_clusters
- old_clusters
;
2283 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2287 if (value_size
>= OCFS2_XATTR_ROOT_SIZE
)
2292 * Now the new value will be stored inside. So if the new
2293 * value is smaller than the size of value root or the old
2294 * value, we don't need any allocation, otherwise we have
2295 * to guess metadata allocation.
2297 if ((ocfs2_xattr_is_local(xe
) && value_size
>= xi
->value_len
) ||
2298 (!ocfs2_xattr_is_local(xe
) &&
2299 OCFS2_XATTR_ROOT_SIZE
>= xi
->value_len
))
2304 /* calculate metadata allocation. */
2305 if (di
->i_xattr_loc
) {
2306 if (!xbs
->xattr_bh
) {
2307 ret
= ocfs2_read_xattr_block(inode
,
2308 le64_to_cpu(di
->i_xattr_loc
),
2315 xb
= (struct ocfs2_xattr_block
*)bh
->b_data
;
2317 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
2319 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
2320 struct ocfs2_extent_list
*el
=
2321 &xb
->xb_attrs
.xb_root
.xt_list
;
2322 meta_add
+= ocfs2_extend_meta_needed(el
);
2323 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2328 * This cluster will be used either for new bucket or for
2330 * If the cluster size is the same as the bucket size, one
2331 * more is needed since we may need to extend the bucket
2335 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2336 if (OCFS2_XATTR_BUCKET_SIZE
==
2337 OCFS2_SB(inode
->i_sb
)->s_clustersize
) {
2338 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2343 credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
2347 *clusters_need
= clusters_add
;
2349 *meta_need
= meta_add
;
2351 *credits_need
= credits
;
2356 static int ocfs2_init_xattr_set_ctxt(struct inode
*inode
,
2357 struct ocfs2_dinode
*di
,
2358 struct ocfs2_xattr_info
*xi
,
2359 struct ocfs2_xattr_search
*xis
,
2360 struct ocfs2_xattr_search
*xbs
,
2361 struct ocfs2_xattr_set_ctxt
*ctxt
,
2364 int clusters_add
, meta_add
, ret
;
2365 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2367 memset(ctxt
, 0, sizeof(struct ocfs2_xattr_set_ctxt
));
2369 ocfs2_init_dealloc_ctxt(&ctxt
->dealloc
);
2371 ret
= ocfs2_calc_xattr_set_need(inode
, di
, xi
, xis
, xbs
,
2372 &clusters_add
, &meta_add
, credits
);
2378 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2379 "credits = %d\n", xi
->name
, meta_add
, clusters_add
, *credits
);
2382 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
,
2391 ret
= ocfs2_reserve_clusters(osb
, clusters_add
, &ctxt
->data_ac
);
2397 if (ctxt
->meta_ac
) {
2398 ocfs2_free_alloc_context(ctxt
->meta_ac
);
2399 ctxt
->meta_ac
= NULL
;
2403 * We cannot have an error and a non null ctxt->data_ac.
2410 static int __ocfs2_xattr_set_handle(struct inode
*inode
,
2411 struct ocfs2_dinode
*di
,
2412 struct ocfs2_xattr_info
*xi
,
2413 struct ocfs2_xattr_search
*xis
,
2414 struct ocfs2_xattr_search
*xbs
,
2415 struct ocfs2_xattr_set_ctxt
*ctxt
)
2417 int ret
= 0, credits
, old_found
;
2420 /* Remove existing extended attribute */
2421 if (!xis
->not_found
)
2422 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
2423 else if (!xbs
->not_found
)
2424 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2426 /* We always try to set extended attribute into inode first*/
2427 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
2428 if (!ret
&& !xbs
->not_found
) {
2430 * If succeed and that extended attribute existing in
2431 * external block, then we will remove it.
2436 old_found
= xis
->not_found
;
2437 xis
->not_found
= -ENODATA
;
2438 ret
= ocfs2_calc_xattr_set_need(inode
,
2446 xis
->not_found
= old_found
;
2452 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2453 ctxt
->handle
->h_buffer_credits
);
2458 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2459 } else if (ret
== -ENOSPC
) {
2460 if (di
->i_xattr_loc
&& !xbs
->xattr_bh
) {
2461 ret
= ocfs2_xattr_block_find(inode
,
2467 old_found
= xis
->not_found
;
2468 xis
->not_found
= -ENODATA
;
2469 ret
= ocfs2_calc_xattr_set_need(inode
,
2477 xis
->not_found
= old_found
;
2483 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2484 ctxt
->handle
->h_buffer_credits
);
2491 * If no space in inode, we will set extended attribute
2492 * into external block.
2494 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2497 if (!xis
->not_found
) {
2499 * If succeed and that extended attribute
2500 * existing in inode, we will remove it.
2504 xbs
->not_found
= -ENODATA
;
2505 ret
= ocfs2_calc_xattr_set_need(inode
,
2518 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2519 ctxt
->handle
->h_buffer_credits
);
2524 ret
= ocfs2_xattr_ibody_set(inode
, xi
,
2535 * This function only called duing creating inode
2536 * for init security/acl xattrs of the new inode.
2537 * The xattrs could be put into ibody or extent block,
2538 * xattr bucket would not be use in this case.
2539 * transanction credits also be reserved in here.
2541 int ocfs2_xattr_set_handle(handle_t
*handle
,
2542 struct inode
*inode
,
2543 struct buffer_head
*di_bh
,
2549 struct ocfs2_alloc_context
*meta_ac
,
2550 struct ocfs2_alloc_context
*data_ac
)
2552 struct ocfs2_dinode
*di
;
2555 struct ocfs2_xattr_info xi
= {
2556 .name_index
= name_index
,
2559 .value_len
= value_len
,
2562 struct ocfs2_xattr_search xis
= {
2563 .not_found
= -ENODATA
,
2566 struct ocfs2_xattr_search xbs
= {
2567 .not_found
= -ENODATA
,
2570 struct ocfs2_xattr_set_ctxt ctxt
= {
2576 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2579 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
2580 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2582 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2584 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
2587 if (xis
.not_found
) {
2588 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
2593 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
2596 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2597 brelse(xbs
.xattr_bh
);
2605 * Set, replace or remove an extended attribute for this inode.
2606 * value is NULL to remove an existing extended attribute, else either
2607 * create or replace an extended attribute.
2609 int ocfs2_xattr_set(struct inode
*inode
,
2616 struct buffer_head
*di_bh
= NULL
;
2617 struct ocfs2_dinode
*di
;
2619 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2620 struct inode
*tl_inode
= osb
->osb_tl_inode
;
2621 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
2623 struct ocfs2_xattr_info xi
= {
2624 .name_index
= name_index
,
2627 .value_len
= value_len
,
2630 struct ocfs2_xattr_search xis
= {
2631 .not_found
= -ENODATA
,
2634 struct ocfs2_xattr_search xbs
= {
2635 .not_found
= -ENODATA
,
2638 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2642 * Only xbs will be used on indexed trees. xis doesn't need a
2645 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
2647 mlog_errno(-ENOMEM
);
2651 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2654 goto cleanup_nolock
;
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 * Scan inode and external block to find the same name
2662 * extended attribute and collect search infomation.
2664 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
2667 if (xis
.not_found
) {
2668 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
2673 if (xis
.not_found
&& xbs
.not_found
) {
2675 if (flags
& XATTR_REPLACE
)
2682 if (flags
& XATTR_CREATE
)
2687 mutex_lock(&tl_inode
->i_mutex
);
2689 if (ocfs2_truncate_log_needs_flush(osb
)) {
2690 ret
= __ocfs2_flush_truncate_log(osb
);
2692 mutex_unlock(&tl_inode
->i_mutex
);
2697 mutex_unlock(&tl_inode
->i_mutex
);
2699 ret
= ocfs2_init_xattr_set_ctxt(inode
, di
, &xi
, &xis
,
2700 &xbs
, &ctxt
, &credits
);
2706 ctxt
.handle
= ocfs2_start_trans(osb
, credits
);
2707 if (IS_ERR(ctxt
.handle
)) {
2708 ret
= PTR_ERR(ctxt
.handle
);
2713 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
2715 ocfs2_commit_trans(osb
, ctxt
.handle
);
2718 ocfs2_free_alloc_context(ctxt
.data_ac
);
2720 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2721 if (ocfs2_dealloc_has_cluster(&ctxt
.dealloc
))
2722 ocfs2_schedule_truncate_log_flush(osb
, 1);
2723 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
2725 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2726 ocfs2_inode_unlock(inode
, 1);
2729 brelse(xbs
.xattr_bh
);
2730 ocfs2_xattr_bucket_free(xbs
.bucket
);
2736 * Find the xattr extent rec which may contains name_hash.
2737 * e_cpos will be the first name hash of the xattr rec.
2738 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2740 static int ocfs2_xattr_get_rec(struct inode
*inode
,
2745 struct ocfs2_extent_list
*el
)
2748 struct buffer_head
*eb_bh
= NULL
;
2749 struct ocfs2_extent_block
*eb
;
2750 struct ocfs2_extent_rec
*rec
= NULL
;
2753 if (el
->l_tree_depth
) {
2754 ret
= ocfs2_find_leaf(inode
, el
, name_hash
, &eb_bh
);
2760 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
2763 if (el
->l_tree_depth
) {
2764 ocfs2_error(inode
->i_sb
,
2765 "Inode %lu has non zero tree depth in "
2766 "xattr tree block %llu\n", inode
->i_ino
,
2767 (unsigned long long)eb_bh
->b_blocknr
);
2773 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
2774 rec
= &el
->l_recs
[i
];
2776 if (le32_to_cpu(rec
->e_cpos
) <= name_hash
) {
2777 e_blkno
= le64_to_cpu(rec
->e_blkno
);
2783 ocfs2_error(inode
->i_sb
, "Inode %lu has bad extent "
2784 "record (%u, %u, 0) in xattr", inode
->i_ino
,
2785 le32_to_cpu(rec
->e_cpos
),
2786 ocfs2_rec_clusters(el
, rec
));
2791 *p_blkno
= le64_to_cpu(rec
->e_blkno
);
2792 *num_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
2794 *e_cpos
= le32_to_cpu(rec
->e_cpos
);
2800 typedef int (xattr_bucket_func
)(struct inode
*inode
,
2801 struct ocfs2_xattr_bucket
*bucket
,
2804 static int ocfs2_find_xe_in_bucket(struct inode
*inode
,
2805 struct ocfs2_xattr_bucket
*bucket
,
2812 int i
, ret
= 0, cmp
= 1, block_off
, new_offset
;
2813 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
2814 size_t name_len
= strlen(name
);
2815 struct ocfs2_xattr_entry
*xe
= NULL
;
2819 * We don't use binary search in the bucket because there
2820 * may be multiple entries with the same name hash.
2822 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
2823 xe
= &xh
->xh_entries
[i
];
2825 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
))
2827 else if (name_hash
< le32_to_cpu(xe
->xe_name_hash
))
2830 cmp
= name_index
- ocfs2_xattr_get_type(xe
);
2832 cmp
= name_len
- xe
->xe_name_len
;
2836 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
2847 xe_name
= bucket_block(bucket
, block_off
) + new_offset
;
2848 if (!memcmp(name
, xe_name
, name_len
)) {
2860 * Find the specified xattr entry in a series of buckets.
2861 * This series start from p_blkno and last for num_clusters.
2862 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2863 * the num of the valid buckets.
2865 * Return the buffer_head this xattr should reside in. And if the xattr's
2866 * hash is in the gap of 2 buckets, return the lower bucket.
2868 static int ocfs2_xattr_bucket_find(struct inode
*inode
,
2875 struct ocfs2_xattr_search
*xs
)
2878 struct ocfs2_xattr_header
*xh
= NULL
;
2879 struct ocfs2_xattr_entry
*xe
= NULL
;
2881 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2882 int low_bucket
= 0, bucket
, high_bucket
;
2883 struct ocfs2_xattr_bucket
*search
;
2885 u64 blkno
, lower_blkno
= 0;
2887 search
= ocfs2_xattr_bucket_new(inode
);
2894 ret
= ocfs2_read_xattr_bucket(search
, p_blkno
);
2900 xh
= bucket_xh(search
);
2901 high_bucket
= le16_to_cpu(xh
->xh_num_buckets
) - 1;
2902 while (low_bucket
<= high_bucket
) {
2903 ocfs2_xattr_bucket_relse(search
);
2905 bucket
= (low_bucket
+ high_bucket
) / 2;
2906 blkno
= p_blkno
+ bucket
* blk_per_bucket
;
2907 ret
= ocfs2_read_xattr_bucket(search
, blkno
);
2913 xh
= bucket_xh(search
);
2914 xe
= &xh
->xh_entries
[0];
2915 if (name_hash
< le32_to_cpu(xe
->xe_name_hash
)) {
2916 high_bucket
= bucket
- 1;
2921 * Check whether the hash of the last entry in our
2922 * bucket is larger than the search one. for an empty
2923 * bucket, the last one is also the first one.
2926 xe
= &xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1];
2928 last_hash
= le32_to_cpu(xe
->xe_name_hash
);
2930 /* record lower_blkno which may be the insert place. */
2931 lower_blkno
= blkno
;
2933 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
)) {
2934 low_bucket
= bucket
+ 1;
2938 /* the searched xattr should reside in this bucket if exists. */
2939 ret
= ocfs2_find_xe_in_bucket(inode
, search
,
2940 name_index
, name
, name_hash
,
2950 * Record the bucket we have found.
2951 * When the xattr's hash value is in the gap of 2 buckets, we will
2952 * always set it to the previous bucket.
2955 lower_blkno
= p_blkno
;
2957 /* This should be in cache - we just read it during the search */
2958 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, lower_blkno
);
2964 xs
->header
= bucket_xh(xs
->bucket
);
2965 xs
->base
= bucket_block(xs
->bucket
, 0);
2966 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
2969 xs
->here
= &xs
->header
->xh_entries
[index
];
2970 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name
,
2971 (unsigned long long)bucket_blkno(xs
->bucket
), index
);
2976 ocfs2_xattr_bucket_free(search
);
2980 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
2981 struct buffer_head
*root_bh
,
2984 struct ocfs2_xattr_search
*xs
)
2987 struct ocfs2_xattr_block
*xb
=
2988 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
2989 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
2990 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
2992 u32 first_hash
, num_clusters
= 0;
2993 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
2995 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
2998 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2999 name
, name_hash
, name_index
);
3001 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &first_hash
,
3008 BUG_ON(p_blkno
== 0 || num_clusters
== 0 || first_hash
> name_hash
);
3010 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3011 "in the rec is %u\n", num_clusters
, (unsigned long long)p_blkno
,
3014 ret
= ocfs2_xattr_bucket_find(inode
, name_index
, name
, name_hash
,
3015 p_blkno
, first_hash
, num_clusters
, xs
);
3021 static int ocfs2_iterate_xattr_buckets(struct inode
*inode
,
3024 xattr_bucket_func
*func
,
3028 u32 bpc
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
));
3029 u32 num_buckets
= clusters
* bpc
;
3030 struct ocfs2_xattr_bucket
*bucket
;
3032 bucket
= ocfs2_xattr_bucket_new(inode
);
3034 mlog_errno(-ENOMEM
);
3038 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3039 clusters
, (unsigned long long)blkno
);
3041 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bucket
->bu_blocks
) {
3042 ret
= ocfs2_read_xattr_bucket(bucket
, blkno
);
3049 * The real bucket num in this series of blocks is stored
3050 * in the 1st bucket.
3053 num_buckets
= le16_to_cpu(bucket_xh(bucket
)->xh_num_buckets
);
3055 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3056 (unsigned long long)blkno
,
3057 le32_to_cpu(bucket_xh(bucket
)->xh_entries
[0].xe_name_hash
));
3059 ret
= func(inode
, bucket
, para
);
3062 /* Fall through to bucket_relse() */
3065 ocfs2_xattr_bucket_relse(bucket
);
3070 ocfs2_xattr_bucket_free(bucket
);
3074 struct ocfs2_xattr_tree_list
{
3080 static int ocfs2_xattr_bucket_get_name_value(struct inode
*inode
,
3081 struct ocfs2_xattr_header
*xh
,
3088 if (index
< 0 || index
>= le16_to_cpu(xh
->xh_count
))
3091 name_offset
= le16_to_cpu(xh
->xh_entries
[index
].xe_name_offset
);
3093 *block_off
= name_offset
>> inode
->i_sb
->s_blocksize_bits
;
3094 *new_offset
= name_offset
% inode
->i_sb
->s_blocksize
;
3099 static int ocfs2_list_xattr_bucket(struct inode
*inode
,
3100 struct ocfs2_xattr_bucket
*bucket
,
3104 struct ocfs2_xattr_tree_list
*xl
= (struct ocfs2_xattr_tree_list
*)para
;
3105 int i
, block_off
, new_offset
;
3106 const char *prefix
, *name
;
3108 for (i
= 0 ; i
< le16_to_cpu(bucket_xh(bucket
)->xh_count
); i
++) {
3109 struct ocfs2_xattr_entry
*entry
= &bucket_xh(bucket
)->xh_entries
[i
];
3110 type
= ocfs2_xattr_get_type(entry
);
3111 prefix
= ocfs2_xattr_prefix(type
);
3114 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
3122 name
= (const char *)bucket_block(bucket
, block_off
) +
3124 ret
= ocfs2_xattr_list_entry(xl
->buffer
,
3128 entry
->xe_name_len
);
3137 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
3138 struct ocfs2_xattr_tree_root
*xt
,
3142 struct ocfs2_extent_list
*el
= &xt
->xt_list
;
3144 u32 name_hash
= UINT_MAX
, e_cpos
= 0, num_clusters
= 0;
3146 struct ocfs2_xattr_tree_list xl
= {
3148 .buffer_size
= buffer_size
,
3152 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
3155 while (name_hash
> 0) {
3156 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
3157 &e_cpos
, &num_clusters
, el
);
3163 ret
= ocfs2_iterate_xattr_buckets(inode
, p_blkno
, num_clusters
,
3164 ocfs2_list_xattr_bucket
,
3174 name_hash
= e_cpos
- 1;
3182 static int cmp_xe(const void *a
, const void *b
)
3184 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
3185 u32 l_hash
= le32_to_cpu(l
->xe_name_hash
);
3186 u32 r_hash
= le32_to_cpu(r
->xe_name_hash
);
3188 if (l_hash
> r_hash
)
3190 if (l_hash
< r_hash
)
3195 static void swap_xe(void *a
, void *b
, int size
)
3197 struct ocfs2_xattr_entry
*l
= a
, *r
= b
, tmp
;
3200 memcpy(l
, r
, sizeof(struct ocfs2_xattr_entry
));
3201 memcpy(r
, &tmp
, sizeof(struct ocfs2_xattr_entry
));
3205 * When the ocfs2_xattr_block is filled up, new bucket will be created
3206 * and all the xattr entries will be moved to the new bucket.
3207 * The header goes at the start of the bucket, and the names+values are
3208 * filled from the end. This is why *target starts as the last buffer.
3209 * Note: we need to sort the entries since they are not saved in order
3210 * in the ocfs2_xattr_block.
3212 static void ocfs2_cp_xattr_block_to_bucket(struct inode
*inode
,
3213 struct buffer_head
*xb_bh
,
3214 struct ocfs2_xattr_bucket
*bucket
)
3216 int i
, blocksize
= inode
->i_sb
->s_blocksize
;
3217 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3218 u16 offset
, size
, off_change
;
3219 struct ocfs2_xattr_entry
*xe
;
3220 struct ocfs2_xattr_block
*xb
=
3221 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
3222 struct ocfs2_xattr_header
*xb_xh
= &xb
->xb_attrs
.xb_header
;
3223 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
3224 u16 count
= le16_to_cpu(xb_xh
->xh_count
);
3225 char *src
= xb_bh
->b_data
;
3226 char *target
= bucket_block(bucket
, blks
- 1);
3228 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3229 (unsigned long long)xb_bh
->b_blocknr
,
3230 (unsigned long long)bucket_blkno(bucket
));
3232 for (i
= 0; i
< blks
; i
++)
3233 memset(bucket_block(bucket
, i
), 0, blocksize
);
3236 * Since the xe_name_offset is based on ocfs2_xattr_header,
3237 * there is a offset change corresponding to the change of
3238 * ocfs2_xattr_header's position.
3240 off_change
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
3241 xe
= &xb_xh
->xh_entries
[count
- 1];
3242 offset
= le16_to_cpu(xe
->xe_name_offset
) + off_change
;
3243 size
= blocksize
- offset
;
3245 /* copy all the names and values. */
3246 memcpy(target
+ offset
, src
+ offset
, size
);
3248 /* Init new header now. */
3249 xh
->xh_count
= xb_xh
->xh_count
;
3250 xh
->xh_num_buckets
= cpu_to_le16(1);
3251 xh
->xh_name_value_len
= cpu_to_le16(size
);
3252 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
- size
);
3254 /* copy all the entries. */
3255 target
= bucket_block(bucket
, 0);
3256 offset
= offsetof(struct ocfs2_xattr_header
, xh_entries
);
3257 size
= count
* sizeof(struct ocfs2_xattr_entry
);
3258 memcpy(target
+ offset
, (char *)xb_xh
+ offset
, size
);
3260 /* Change the xe offset for all the xe because of the move. */
3261 off_change
= OCFS2_XATTR_BUCKET_SIZE
- blocksize
+
3262 offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
3263 for (i
= 0; i
< count
; i
++)
3264 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
, off_change
);
3266 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3267 offset
, size
, off_change
);
3269 sort(target
+ offset
, count
, sizeof(struct ocfs2_xattr_entry
),
3274 * After we move xattr from block to index btree, we have to
3275 * update ocfs2_xattr_search to the new xe and base.
3277 * When the entry is in xattr block, xattr_bh indicates the storage place.
3278 * While if the entry is in index b-tree, "bucket" indicates the
3279 * real place of the xattr.
3281 static void ocfs2_xattr_update_xattr_search(struct inode
*inode
,
3282 struct ocfs2_xattr_search
*xs
,
3283 struct buffer_head
*old_bh
)
3285 char *buf
= old_bh
->b_data
;
3286 struct ocfs2_xattr_block
*old_xb
= (struct ocfs2_xattr_block
*)buf
;
3287 struct ocfs2_xattr_header
*old_xh
= &old_xb
->xb_attrs
.xb_header
;
3290 xs
->header
= bucket_xh(xs
->bucket
);
3291 xs
->base
= bucket_block(xs
->bucket
, 0);
3292 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
3297 i
= xs
->here
- old_xh
->xh_entries
;
3298 xs
->here
= &xs
->header
->xh_entries
[i
];
3301 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
3302 struct ocfs2_xattr_search
*xs
,
3303 struct ocfs2_xattr_set_ctxt
*ctxt
)
3308 handle_t
*handle
= ctxt
->handle
;
3309 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3310 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
3311 struct buffer_head
*xb_bh
= xs
->xattr_bh
;
3312 struct ocfs2_xattr_block
*xb
=
3313 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
3314 struct ocfs2_xattr_tree_root
*xr
;
3315 u16 xb_flags
= le16_to_cpu(xb
->xb_flags
);
3317 mlog(0, "create xattr index block for %llu\n",
3318 (unsigned long long)xb_bh
->b_blocknr
);
3320 BUG_ON(xb_flags
& OCFS2_XATTR_INDEXED
);
3321 BUG_ON(!xs
->bucket
);
3325 * We can use this lock for now, and maybe move to a dedicated mutex
3326 * if performance becomes a problem later.
3328 down_write(&oi
->ip_alloc_sem
);
3330 ret
= ocfs2_journal_access(handle
, inode
, xb_bh
,
3331 OCFS2_JOURNAL_ACCESS_WRITE
);
3337 ret
= __ocfs2_claim_clusters(osb
, handle
, ctxt
->data_ac
,
3338 1, 1, &bit_off
, &len
);
3345 * The bucket may spread in many blocks, and
3346 * we will only touch the 1st block and the last block
3347 * in the whole bucket(one for entry and one for data).
3349 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, bit_off
);
3351 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3352 (unsigned long long)blkno
);
3354 ret
= ocfs2_init_xattr_bucket(xs
->bucket
, blkno
);
3360 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
3361 OCFS2_JOURNAL_ACCESS_CREATE
);
3367 ocfs2_cp_xattr_block_to_bucket(inode
, xb_bh
, xs
->bucket
);
3368 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
3370 ocfs2_xattr_update_xattr_search(inode
, xs
, xb_bh
);
3372 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3373 memset(&xb
->xb_attrs
, 0, inode
->i_sb
->s_blocksize
-
3374 offsetof(struct ocfs2_xattr_block
, xb_attrs
));
3376 xr
= &xb
->xb_attrs
.xb_root
;
3377 xr
->xt_clusters
= cpu_to_le32(1);
3378 xr
->xt_last_eb_blk
= 0;
3379 xr
->xt_list
.l_tree_depth
= 0;
3380 xr
->xt_list
.l_count
= cpu_to_le16(ocfs2_xattr_recs_per_xb(inode
->i_sb
));
3381 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
3383 xr
->xt_list
.l_recs
[0].e_cpos
= 0;
3384 xr
->xt_list
.l_recs
[0].e_blkno
= cpu_to_le64(blkno
);
3385 xr
->xt_list
.l_recs
[0].e_leaf_clusters
= cpu_to_le16(1);
3387 xb
->xb_flags
= cpu_to_le16(xb_flags
| OCFS2_XATTR_INDEXED
);
3389 ocfs2_journal_dirty(handle
, xb_bh
);
3392 up_write(&oi
->ip_alloc_sem
);
3397 static int cmp_xe_offset(const void *a
, const void *b
)
3399 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
3400 u32 l_name_offset
= le16_to_cpu(l
->xe_name_offset
);
3401 u32 r_name_offset
= le16_to_cpu(r
->xe_name_offset
);
3403 if (l_name_offset
< r_name_offset
)
3405 if (l_name_offset
> r_name_offset
)
3411 * defrag a xattr bucket if we find that the bucket has some
3412 * holes beteen name/value pairs.
3413 * We will move all the name/value pairs to the end of the bucket
3414 * so that we can spare some space for insertion.
3416 static int ocfs2_defrag_xattr_bucket(struct inode
*inode
,
3418 struct ocfs2_xattr_bucket
*bucket
)
3421 size_t end
, offset
, len
, value_len
;
3422 struct ocfs2_xattr_header
*xh
;
3423 char *entries
, *buf
, *bucket_buf
= NULL
;
3424 u64 blkno
= bucket_blkno(bucket
);
3426 size_t blocksize
= inode
->i_sb
->s_blocksize
;
3427 struct ocfs2_xattr_entry
*xe
;
3430 * In order to make the operation more efficient and generic,
3431 * we copy all the blocks into a contiguous memory and do the
3432 * defragment there, so if anything is error, we will not touch
3435 bucket_buf
= kmalloc(OCFS2_XATTR_BUCKET_SIZE
, GFP_NOFS
);
3442 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
3443 memcpy(buf
, bucket_block(bucket
, i
), blocksize
);
3445 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
3446 OCFS2_JOURNAL_ACCESS_WRITE
);
3452 xh
= (struct ocfs2_xattr_header
*)bucket_buf
;
3453 entries
= (char *)xh
->xh_entries
;
3454 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
3456 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3457 "xh_free_start = %u, xh_name_value_len = %u.\n",
3458 (unsigned long long)blkno
, le16_to_cpu(xh
->xh_count
),
3459 xh_free_start
, le16_to_cpu(xh
->xh_name_value_len
));
3462 * sort all the entries by their offset.
3463 * the largest will be the first, so that we can
3464 * move them to the end one by one.
3466 sort(entries
, le16_to_cpu(xh
->xh_count
),
3467 sizeof(struct ocfs2_xattr_entry
),
3468 cmp_xe_offset
, swap_xe
);
3470 /* Move all name/values to the end of the bucket. */
3471 xe
= xh
->xh_entries
;
3472 end
= OCFS2_XATTR_BUCKET_SIZE
;
3473 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, xe
++) {
3474 offset
= le16_to_cpu(xe
->xe_name_offset
);
3475 if (ocfs2_xattr_is_local(xe
))
3476 value_len
= OCFS2_XATTR_SIZE(
3477 le64_to_cpu(xe
->xe_value_size
));
3479 value_len
= OCFS2_XATTR_ROOT_SIZE
;
3480 len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
) + value_len
;
3483 * We must make sure that the name/value pair
3484 * exist in the same block. So adjust end to
3485 * the previous block end if needed.
3487 if (((end
- len
) / blocksize
!=
3488 (end
- 1) / blocksize
))
3489 end
= end
- end
% blocksize
;
3491 if (end
> offset
+ len
) {
3492 memmove(bucket_buf
+ end
- len
,
3493 bucket_buf
+ offset
, len
);
3494 xe
->xe_name_offset
= cpu_to_le16(end
- len
);
3497 mlog_bug_on_msg(end
< offset
+ len
, "Defrag check failed for "
3498 "bucket %llu\n", (unsigned long long)blkno
);
3503 mlog_bug_on_msg(xh_free_start
> end
, "Defrag check failed for "
3504 "bucket %llu\n", (unsigned long long)blkno
);
3506 if (xh_free_start
== end
)
3509 memset(bucket_buf
+ xh_free_start
, 0, end
- xh_free_start
);
3510 xh
->xh_free_start
= cpu_to_le16(end
);
3512 /* sort the entries by their name_hash. */
3513 sort(entries
, le16_to_cpu(xh
->xh_count
),
3514 sizeof(struct ocfs2_xattr_entry
),
3518 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
3519 memcpy(bucket_block(bucket
, i
), buf
, blocksize
);
3520 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
3528 * Move half nums of the xattr bucket in the previous cluster to this new
3529 * cluster. We only touch the last cluster of the previous extend record.
3531 * first_bh is the first buffer_head of a series of bucket in the same
3532 * extent rec and header_bh is the header of one bucket in this cluster.
3533 * They will be updated if we move the data header_bh contains to the new
3534 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3536 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode
*inode
,
3538 struct buffer_head
**first_bh
,
3539 struct buffer_head
**header_bh
,
3545 int i
, ret
, credits
;
3546 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3547 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
3548 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
3549 int blocksize
= inode
->i_sb
->s_blocksize
;
3550 struct buffer_head
*old_bh
, *new_bh
, *prev_bh
, *new_first_bh
= NULL
;
3551 struct ocfs2_xattr_header
*new_xh
;
3552 struct ocfs2_xattr_header
*xh
=
3553 (struct ocfs2_xattr_header
*)((*first_bh
)->b_data
);
3555 BUG_ON(le16_to_cpu(xh
->xh_num_buckets
) < num_buckets
);
3556 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
== osb
->s_clustersize
);
3558 prev_bh
= *first_bh
;
3560 xh
= (struct ocfs2_xattr_header
*)prev_bh
->b_data
;
3562 prev_blkno
+= (num_clusters
- 1) * bpc
+ bpc
/ 2;
3564 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3565 (unsigned long long)prev_blkno
, (unsigned long long)new_blkno
);
3568 * We need to update the 1st half of the new cluster and
3569 * 1 more for the update of the 1st bucket of the previous
3572 credits
= bpc
/ 2 + 1 + handle
->h_buffer_credits
;
3573 ret
= ocfs2_extend_trans(handle
, credits
);
3579 ret
= ocfs2_journal_access(handle
, inode
, prev_bh
,
3580 OCFS2_JOURNAL_ACCESS_WRITE
);
3586 for (i
= 0; i
< bpc
/ 2; i
++, prev_blkno
++, new_blkno
++) {
3587 old_bh
= new_bh
= NULL
;
3588 new_bh
= sb_getblk(inode
->i_sb
, new_blkno
);
3595 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
3597 ret
= ocfs2_journal_access(handle
, inode
, new_bh
,
3598 OCFS2_JOURNAL_ACCESS_CREATE
);
3605 ret
= ocfs2_read_block(inode
, prev_blkno
, &old_bh
, NULL
);
3612 memcpy(new_bh
->b_data
, old_bh
->b_data
, blocksize
);
3615 new_xh
= (struct ocfs2_xattr_header
*)new_bh
->b_data
;
3616 new_xh
->xh_num_buckets
= cpu_to_le16(num_buckets
/ 2);
3619 *first_hash
= le32_to_cpu(
3620 new_xh
->xh_entries
[0].xe_name_hash
);
3621 new_first_bh
= new_bh
;
3622 get_bh(new_first_bh
);
3625 ocfs2_journal_dirty(handle
, new_bh
);
3627 if (*header_bh
== old_bh
) {
3629 *header_bh
= new_bh
;
3633 *first_bh
= new_first_bh
;
3640 le16_add_cpu(&xh
->xh_num_buckets
, -(num_buckets
/ 2));
3642 ocfs2_journal_dirty(handle
, prev_bh
);
3645 brelse(new_first_bh
);
3650 * Find the suitable pos when we divide a bucket into 2.
3651 * We have to make sure the xattrs with the same hash value exist
3652 * in the same bucket.
3654 * If this ocfs2_xattr_header covers more than one hash value, find a
3655 * place where the hash value changes. Try to find the most even split.
3656 * The most common case is that all entries have different hash values,
3657 * and the first check we make will find a place to split.
3659 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header
*xh
)
3661 struct ocfs2_xattr_entry
*entries
= xh
->xh_entries
;
3662 int count
= le16_to_cpu(xh
->xh_count
);
3663 int delta
, middle
= count
/ 2;
3666 * We start at the middle. Each step gets farther away in both
3667 * directions. We therefore hit the change in hash value
3668 * nearest to the middle. Note that this loop does not execute for
3671 for (delta
= 0; delta
< middle
; delta
++) {
3672 /* Let's check delta earlier than middle */
3673 if (cmp_xe(&entries
[middle
- delta
- 1],
3674 &entries
[middle
- delta
]))
3675 return middle
- delta
;
3677 /* For even counts, don't walk off the end */
3678 if ((middle
+ delta
+ 1) == count
)
3681 /* Now try delta past middle */
3682 if (cmp_xe(&entries
[middle
+ delta
],
3683 &entries
[middle
+ delta
+ 1]))
3684 return middle
+ delta
+ 1;
3687 /* Every entry had the same hash */
3692 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3693 * first_hash will record the 1st hash of the new bucket.
3695 * Normally half of the xattrs will be moved. But we have to make
3696 * sure that the xattrs with the same hash value are stored in the
3697 * same bucket. If all the xattrs in this bucket have the same hash
3698 * value, the new bucket will be initialized as an empty one and the
3699 * first_hash will be initialized as (hash_value+1).
3701 static int ocfs2_divide_xattr_bucket(struct inode
*inode
,
3706 int new_bucket_head
)
3709 int count
, start
, len
, name_value_len
= 0, xe_len
, name_offset
= 0;
3710 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
3711 struct ocfs2_xattr_header
*xh
;
3712 struct ocfs2_xattr_entry
*xe
;
3713 int blocksize
= inode
->i_sb
->s_blocksize
;
3715 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3716 (unsigned long long)blk
, (unsigned long long)new_blk
);
3718 s_bucket
= ocfs2_xattr_bucket_new(inode
);
3719 t_bucket
= ocfs2_xattr_bucket_new(inode
);
3720 if (!s_bucket
|| !t_bucket
) {
3726 ret
= ocfs2_read_xattr_bucket(s_bucket
, blk
);
3732 ret
= ocfs2_xattr_bucket_journal_access(handle
, s_bucket
,
3733 OCFS2_JOURNAL_ACCESS_WRITE
);
3740 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3741 * there's no need to read it.
3743 ret
= ocfs2_init_xattr_bucket(t_bucket
, new_blk
);
3749 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
3751 OCFS2_JOURNAL_ACCESS_CREATE
:
3752 OCFS2_JOURNAL_ACCESS_WRITE
);
3758 xh
= bucket_xh(s_bucket
);
3759 count
= le16_to_cpu(xh
->xh_count
);
3760 start
= ocfs2_xattr_find_divide_pos(xh
);
3762 if (start
== count
) {
3763 xe
= &xh
->xh_entries
[start
-1];
3766 * initialized a new empty bucket here.
3767 * The hash value is set as one larger than
3768 * that of the last entry in the previous bucket.
3770 for (i
= 0; i
< t_bucket
->bu_blocks
; i
++)
3771 memset(bucket_block(t_bucket
, i
), 0, blocksize
);
3773 xh
= bucket_xh(t_bucket
);
3774 xh
->xh_free_start
= cpu_to_le16(blocksize
);
3775 xh
->xh_entries
[0].xe_name_hash
= xe
->xe_name_hash
;
3776 le32_add_cpu(&xh
->xh_entries
[0].xe_name_hash
, 1);
3778 goto set_num_buckets
;
3781 /* copy the whole bucket to the new first. */
3782 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
3784 /* update the new bucket. */
3785 xh
= bucket_xh(t_bucket
);
3788 * Calculate the total name/value len and xh_free_start for
3789 * the old bucket first.
3791 name_offset
= OCFS2_XATTR_BUCKET_SIZE
;
3793 for (i
= 0; i
< start
; i
++) {
3794 xe
= &xh
->xh_entries
[i
];
3795 xe_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3796 if (ocfs2_xattr_is_local(xe
))
3798 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
3800 xe_len
+= OCFS2_XATTR_ROOT_SIZE
;
3801 name_value_len
+= xe_len
;
3802 if (le16_to_cpu(xe
->xe_name_offset
) < name_offset
)
3803 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3807 * Now begin the modification to the new bucket.
3809 * In the new bucket, We just move the xattr entry to the beginning
3810 * and don't touch the name/value. So there will be some holes in the
3811 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3814 xe
= &xh
->xh_entries
[start
];
3815 len
= sizeof(struct ocfs2_xattr_entry
) * (count
- start
);
3816 mlog(0, "mv xattr entry len %d from %d to %d\n", len
,
3817 (int)((char *)xe
- (char *)xh
),
3818 (int)((char *)xh
->xh_entries
- (char *)xh
));
3819 memmove((char *)xh
->xh_entries
, (char *)xe
, len
);
3820 xe
= &xh
->xh_entries
[count
- start
];
3821 len
= sizeof(struct ocfs2_xattr_entry
) * start
;
3822 memset((char *)xe
, 0, len
);
3824 le16_add_cpu(&xh
->xh_count
, -start
);
3825 le16_add_cpu(&xh
->xh_name_value_len
, -name_value_len
);
3827 /* Calculate xh_free_start for the new bucket. */
3828 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
3829 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
3830 xe
= &xh
->xh_entries
[i
];
3831 xe_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3832 if (ocfs2_xattr_is_local(xe
))
3834 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
3836 xe_len
+= OCFS2_XATTR_ROOT_SIZE
;
3837 if (le16_to_cpu(xe
->xe_name_offset
) <
3838 le16_to_cpu(xh
->xh_free_start
))
3839 xh
->xh_free_start
= xe
->xe_name_offset
;
3843 /* set xh->xh_num_buckets for the new xh. */
3844 if (new_bucket_head
)
3845 xh
->xh_num_buckets
= cpu_to_le16(1);
3847 xh
->xh_num_buckets
= 0;
3849 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
3851 /* store the first_hash of the new bucket. */
3853 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
3856 * Now only update the 1st block of the old bucket. If we
3857 * just added a new empty bucket, there is no need to modify
3863 xh
= bucket_xh(s_bucket
);
3864 memset(&xh
->xh_entries
[start
], 0,
3865 sizeof(struct ocfs2_xattr_entry
) * (count
- start
));
3866 xh
->xh_count
= cpu_to_le16(start
);
3867 xh
->xh_free_start
= cpu_to_le16(name_offset
);
3868 xh
->xh_name_value_len
= cpu_to_le16(name_value_len
);
3870 ocfs2_xattr_bucket_journal_dirty(handle
, s_bucket
);
3873 ocfs2_xattr_bucket_free(s_bucket
);
3874 ocfs2_xattr_bucket_free(t_bucket
);
3880 * Copy xattr from one bucket to another bucket.
3882 * The caller must make sure that the journal transaction
3883 * has enough space for journaling.
3885 static int ocfs2_cp_xattr_bucket(struct inode
*inode
,
3892 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
3894 BUG_ON(s_blkno
== t_blkno
);
3896 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3897 (unsigned long long)s_blkno
, (unsigned long long)t_blkno
,
3900 s_bucket
= ocfs2_xattr_bucket_new(inode
);
3901 t_bucket
= ocfs2_xattr_bucket_new(inode
);
3902 if (!s_bucket
|| !t_bucket
) {
3908 ret
= ocfs2_read_xattr_bucket(s_bucket
, s_blkno
);
3913 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3914 * there's no need to read it.
3916 ret
= ocfs2_init_xattr_bucket(t_bucket
, t_blkno
);
3920 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
3922 OCFS2_JOURNAL_ACCESS_CREATE
:
3923 OCFS2_JOURNAL_ACCESS_WRITE
);
3927 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
3928 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
3931 ocfs2_xattr_bucket_free(t_bucket
);
3932 ocfs2_xattr_bucket_free(s_bucket
);
3938 * Copy one xattr cluster from src_blk to to_blk.
3939 * The to_blk will become the first bucket header of the cluster, so its
3940 * xh_num_buckets will be initialized as the bucket num in the cluster.
3942 static int ocfs2_cp_xattr_cluster(struct inode
*inode
,
3944 struct buffer_head
*first_bh
,
3949 int i
, ret
, credits
;
3950 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3951 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
3952 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
3953 struct buffer_head
*bh
= NULL
;
3954 struct ocfs2_xattr_header
*xh
;
3955 u64 to_blk_start
= to_blk
;
3957 mlog(0, "cp xattrs from cluster %llu to %llu\n",
3958 (unsigned long long)src_blk
, (unsigned long long)to_blk
);
3961 * We need to update the new cluster and 1 more for the update of
3962 * the 1st bucket of the previous extent rec.
3964 credits
= bpc
+ 1 + handle
->h_buffer_credits
;
3965 ret
= ocfs2_extend_trans(handle
, credits
);
3971 ret
= ocfs2_journal_access(handle
, inode
, first_bh
,
3972 OCFS2_JOURNAL_ACCESS_WRITE
);
3978 for (i
= 0; i
< num_buckets
; i
++) {
3979 ret
= ocfs2_cp_xattr_bucket(inode
, handle
,
3980 src_blk
, to_blk
, 1);
3986 src_blk
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3987 to_blk
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3990 /* update the old bucket header. */
3991 xh
= (struct ocfs2_xattr_header
*)first_bh
->b_data
;
3992 le16_add_cpu(&xh
->xh_num_buckets
, -num_buckets
);
3994 ocfs2_journal_dirty(handle
, first_bh
);
3996 /* update the new bucket header. */
3997 ret
= ocfs2_read_block(inode
, to_blk_start
, &bh
, NULL
);
4003 ret
= ocfs2_journal_access(handle
, inode
, bh
,
4004 OCFS2_JOURNAL_ACCESS_WRITE
);
4010 xh
= (struct ocfs2_xattr_header
*)bh
->b_data
;
4011 xh
->xh_num_buckets
= cpu_to_le16(num_buckets
);
4013 ocfs2_journal_dirty(handle
, bh
);
4016 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
4023 * Move some xattrs in this cluster to the new cluster.
4024 * This function should only be called when bucket size == cluster size.
4025 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4027 static int ocfs2_divide_xattr_cluster(struct inode
*inode
,
4033 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4034 int ret
, credits
= 2 * blk_per_bucket
+ handle
->h_buffer_credits
;
4036 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
< OCFS2_SB(inode
->i_sb
)->s_clustersize
);
4038 ret
= ocfs2_extend_trans(handle
, credits
);
4044 /* Move half of the xattr in start_blk to the next bucket. */
4045 return ocfs2_divide_xattr_bucket(inode
, handle
, prev_blk
,
4046 new_blk
, first_hash
, 1);
4050 * Move some xattrs from the old cluster to the new one since they are not
4051 * contiguous in ocfs2 xattr tree.
4053 * new_blk starts a new separate cluster, and we will move some xattrs from
4054 * prev_blk to it. v_start will be set as the first name hash value in this
4055 * new cluster so that it can be used as e_cpos during tree insertion and
4056 * don't collide with our original b-tree operations. first_bh and header_bh
4057 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4058 * to extend the insert bucket.
4060 * The problem is how much xattr should we move to the new one and when should
4061 * we update first_bh and header_bh?
4062 * 1. If cluster size > bucket size, that means the previous cluster has more
4063 * than 1 bucket, so just move half nums of bucket into the new cluster and
4064 * update the first_bh and header_bh if the insert bucket has been moved
4065 * to the new cluster.
4066 * 2. If cluster_size == bucket_size:
4067 * a) If the previous extent rec has more than one cluster and the insert
4068 * place isn't in the last cluster, copy the entire last cluster to the
4069 * new one. This time, we don't need to upate the first_bh and header_bh
4070 * since they will not be moved into the new cluster.
4071 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4072 * the new one. And we set the extend flag to zero if the insert place is
4073 * moved into the new allocated cluster since no extend is needed.
4075 static int ocfs2_adjust_xattr_cross_cluster(struct inode
*inode
,
4077 struct buffer_head
**first_bh
,
4078 struct buffer_head
**header_bh
,
4086 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
4088 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4089 (unsigned long long)prev_blk
, prev_clusters
,
4090 (unsigned long long)new_blk
);
4092 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
)) > 1)
4093 ret
= ocfs2_mv_xattr_bucket_cross_cluster(inode
,
4102 u64 last_blk
= prev_blk
+ bpc
* (prev_clusters
- 1);
4104 if (prev_clusters
> 1 && (*header_bh
)->b_blocknr
!= last_blk
)
4105 ret
= ocfs2_cp_xattr_cluster(inode
, handle
, *first_bh
,
4109 ret
= ocfs2_divide_xattr_cluster(inode
, handle
,
4113 if ((*header_bh
)->b_blocknr
== last_blk
&& extend
)
4122 * Add a new cluster for xattr storage.
4124 * If the new cluster is contiguous with the previous one, it will be
4125 * appended to the same extent record, and num_clusters will be updated.
4126 * If not, we will insert a new extent for it and move some xattrs in
4127 * the last cluster into the new allocated one.
4128 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4129 * lose the benefits of hashing because we'll have to search large leaves.
4130 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4133 * first_bh is the first block of the previous extent rec and header_bh
4134 * indicates the bucket we will insert the new xattrs. They will be updated
4135 * when the header_bh is moved into the new cluster.
4137 static int ocfs2_add_new_xattr_cluster(struct inode
*inode
,
4138 struct buffer_head
*root_bh
,
4139 struct buffer_head
**first_bh
,
4140 struct buffer_head
**header_bh
,
4145 struct ocfs2_xattr_set_ctxt
*ctxt
)
4148 u16 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
4149 u32 prev_clusters
= *num_clusters
;
4150 u32 clusters_to_add
= 1, bit_off
, num_bits
, v_start
= 0;
4152 handle_t
*handle
= ctxt
->handle
;
4153 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4154 struct ocfs2_extent_tree et
;
4156 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4157 "previous xattr blkno = %llu\n",
4158 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
4159 prev_cpos
, (unsigned long long)prev_blkno
);
4161 ocfs2_init_xattr_tree_extent_tree(&et
, inode
, root_bh
);
4163 ret
= ocfs2_journal_access(handle
, inode
, root_bh
,
4164 OCFS2_JOURNAL_ACCESS_WRITE
);
4170 ret
= __ocfs2_claim_clusters(osb
, handle
, ctxt
->data_ac
, 1,
4171 clusters_to_add
, &bit_off
, &num_bits
);
4178 BUG_ON(num_bits
> clusters_to_add
);
4180 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
4181 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4182 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
4184 if (prev_blkno
+ prev_clusters
* bpc
== block
&&
4185 (prev_clusters
+ num_bits
) << osb
->s_clustersize_bits
<=
4186 OCFS2_MAX_XATTR_TREE_LEAF_SIZE
) {
4188 * If this cluster is contiguous with the old one and
4189 * adding this new cluster, we don't surpass the limit of
4190 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4191 * initialized and used like other buckets in the previous
4193 * So add it as a contiguous one. The caller will handle
4196 v_start
= prev_cpos
+ prev_clusters
;
4197 *num_clusters
= prev_clusters
+ num_bits
;
4198 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4201 ret
= ocfs2_adjust_xattr_cross_cluster(inode
,
4216 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4217 num_bits
, (unsigned long long)block
, v_start
);
4218 ret
= ocfs2_insert_extent(osb
, handle
, inode
, &et
, v_start
, block
,
4219 num_bits
, 0, ctxt
->meta_ac
);
4225 ret
= ocfs2_journal_dirty(handle
, root_bh
);
4234 * Extend a new xattr bucket and move xattrs to the end one by one until
4235 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
4237 static int ocfs2_extend_xattr_bucket(struct inode
*inode
,
4239 struct buffer_head
*first_bh
,
4240 struct buffer_head
*start_bh
,
4244 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4245 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4246 u64 start_blk
= start_bh
->b_blocknr
, end_blk
;
4247 u32 num_buckets
= num_clusters
* ocfs2_xattr_buckets_per_cluster(osb
);
4248 struct ocfs2_xattr_header
*first_xh
=
4249 (struct ocfs2_xattr_header
*)first_bh
->b_data
;
4250 u16 bucket
= le16_to_cpu(first_xh
->xh_num_buckets
);
4252 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4253 "from %llu, len = %u\n", (unsigned long long)start_blk
,
4254 (unsigned long long)first_bh
->b_blocknr
, num_clusters
);
4256 BUG_ON(bucket
>= num_buckets
);
4258 end_blk
= first_bh
->b_blocknr
+ (bucket
- 1) * blk_per_bucket
;
4261 * We will touch all the buckets after the start_bh(include it).
4262 * Then we add one more bucket.
4264 credits
= end_blk
- start_blk
+ 3 * blk_per_bucket
+ 1 +
4265 handle
->h_buffer_credits
;
4266 ret
= ocfs2_extend_trans(handle
, credits
);
4272 ret
= ocfs2_journal_access(handle
, inode
, first_bh
,
4273 OCFS2_JOURNAL_ACCESS_WRITE
);
4279 while (end_blk
!= start_blk
) {
4280 ret
= ocfs2_cp_xattr_bucket(inode
, handle
, end_blk
,
4281 end_blk
+ blk_per_bucket
, 0);
4284 end_blk
-= blk_per_bucket
;
4287 /* Move half of the xattr in start_blk to the next bucket. */
4288 ret
= ocfs2_divide_xattr_bucket(inode
, handle
, start_blk
,
4289 start_blk
+ blk_per_bucket
, NULL
, 0);
4291 le16_add_cpu(&first_xh
->xh_num_buckets
, 1);
4292 ocfs2_journal_dirty(handle
, first_bh
);
4299 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
4300 * xb_bh is the ocfs2_xattr_block.
4301 * We will move all the buckets starting from header_bh to the next place. As
4302 * for this one, half num of its xattrs will be moved to the next one.
4304 * We will allocate a new cluster if current cluster is full and adjust
4305 * header_bh and first_bh if the insert place is moved to the new cluster.
4307 static int ocfs2_add_new_xattr_bucket(struct inode
*inode
,
4308 struct buffer_head
*xb_bh
,
4309 struct buffer_head
*header_bh
,
4310 struct ocfs2_xattr_set_ctxt
*ctxt
)
4312 struct ocfs2_xattr_header
*first_xh
= NULL
;
4313 struct buffer_head
*first_bh
= NULL
;
4314 struct ocfs2_xattr_block
*xb
=
4315 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4316 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
4317 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
4318 struct ocfs2_xattr_header
*xh
=
4319 (struct ocfs2_xattr_header
*)header_bh
->b_data
;
4320 u32 name_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
4321 struct super_block
*sb
= inode
->i_sb
;
4322 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
4323 int ret
, num_buckets
, extend
= 1;
4325 u32 e_cpos
, num_clusters
;
4327 mlog(0, "Add new xattr bucket starting form %llu\n",
4328 (unsigned long long)header_bh
->b_blocknr
);
4331 * Add refrence for header_bh here because it may be
4332 * changed in ocfs2_add_new_xattr_cluster and we need
4333 * to free it in the end.
4337 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &e_cpos
,
4344 ret
= ocfs2_read_block(inode
, p_blkno
, &first_bh
, NULL
);
4350 num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
) * num_clusters
;
4351 first_xh
= (struct ocfs2_xattr_header
*)first_bh
->b_data
;
4353 if (num_buckets
== le16_to_cpu(first_xh
->xh_num_buckets
)) {
4354 ret
= ocfs2_add_new_xattr_cluster(inode
,
4370 ret
= ocfs2_extend_xattr_bucket(inode
,
4383 static inline char *ocfs2_xattr_bucket_get_val(struct inode
*inode
,
4384 struct ocfs2_xattr_bucket
*bucket
,
4387 int block_off
= offs
>> inode
->i_sb
->s_blocksize_bits
;
4389 offs
= offs
% inode
->i_sb
->s_blocksize
;
4390 return bucket_block(bucket
, block_off
) + offs
;
4394 * Handle the normal xattr set, including replace, delete and new.
4396 * Note: "local" indicates the real data's locality. So we can't
4397 * just its bucket locality by its length.
4399 static void ocfs2_xattr_set_entry_normal(struct inode
*inode
,
4400 struct ocfs2_xattr_info
*xi
,
4401 struct ocfs2_xattr_search
*xs
,
4405 struct ocfs2_xattr_entry
*last
, *xe
;
4406 int name_len
= strlen(xi
->name
);
4407 struct ocfs2_xattr_header
*xh
= xs
->header
;
4408 u16 count
= le16_to_cpu(xh
->xh_count
), start
;
4409 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4411 size_t offs
, size
, new_size
;
4413 last
= &xh
->xh_entries
[count
];
4414 if (!xs
->not_found
) {
4416 offs
= le16_to_cpu(xe
->xe_name_offset
);
4417 if (ocfs2_xattr_is_local(xe
))
4418 size
= OCFS2_XATTR_SIZE(name_len
) +
4419 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
4421 size
= OCFS2_XATTR_SIZE(name_len
) +
4422 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE
);
4425 * If the new value will be stored outside, xi->value has been
4426 * initalized as an empty ocfs2_xattr_value_root, and the same
4427 * goes with xi->value_len, so we can set new_size safely here.
4428 * See ocfs2_xattr_set_in_bucket.
4430 new_size
= OCFS2_XATTR_SIZE(name_len
) +
4431 OCFS2_XATTR_SIZE(xi
->value_len
);
4433 le16_add_cpu(&xh
->xh_name_value_len
, -size
);
4435 if (new_size
> size
)
4436 goto set_new_name_value
;
4438 /* Now replace the old value with new one. */
4440 xe
->xe_value_size
= cpu_to_le64(xi
->value_len
);
4442 xe
->xe_value_size
= 0;
4444 val
= ocfs2_xattr_bucket_get_val(inode
,
4446 memset(val
+ OCFS2_XATTR_SIZE(name_len
), 0,
4447 size
- OCFS2_XATTR_SIZE(name_len
));
4448 if (OCFS2_XATTR_SIZE(xi
->value_len
) > 0)
4449 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
4450 xi
->value
, xi
->value_len
);
4452 le16_add_cpu(&xh
->xh_name_value_len
, new_size
);
4453 ocfs2_xattr_set_local(xe
, local
);
4457 * Remove the old entry if there is more than one.
4458 * We don't remove the last entry so that we can
4459 * use it to indicate the hash value of the empty
4463 le16_add_cpu(&xh
->xh_count
, -1);
4466 (void *)last
- (void *)xe
);
4468 sizeof(struct ocfs2_xattr_entry
));
4471 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
4476 /* find a new entry for insert. */
4477 int low
= 0, high
= count
- 1, tmp
;
4478 struct ocfs2_xattr_entry
*tmp_xe
;
4480 while (low
<= high
&& count
) {
4481 tmp
= (low
+ high
) / 2;
4482 tmp_xe
= &xh
->xh_entries
[tmp
];
4484 if (name_hash
> le32_to_cpu(tmp_xe
->xe_name_hash
))
4486 else if (name_hash
<
4487 le32_to_cpu(tmp_xe
->xe_name_hash
))
4495 xe
= &xh
->xh_entries
[low
];
4497 memmove(xe
+ 1, xe
, (void *)last
- (void *)xe
);
4499 le16_add_cpu(&xh
->xh_count
, 1);
4500 memset(xe
, 0, sizeof(struct ocfs2_xattr_entry
));
4501 xe
->xe_name_hash
= cpu_to_le32(name_hash
);
4502 xe
->xe_name_len
= name_len
;
4503 ocfs2_xattr_set_type(xe
, xi
->name_index
);
4507 /* Insert the new name+value. */
4508 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(xi
->value_len
);
4511 * We must make sure that the name/value pair
4512 * exists in the same block.
4514 offs
= le16_to_cpu(xh
->xh_free_start
);
4515 start
= offs
- size
;
4517 if (start
>> inode
->i_sb
->s_blocksize_bits
!=
4518 (offs
- 1) >> inode
->i_sb
->s_blocksize_bits
) {
4519 offs
= offs
- offs
% blocksize
;
4520 xh
->xh_free_start
= cpu_to_le16(offs
);
4523 val
= ocfs2_xattr_bucket_get_val(inode
, xs
->bucket
, offs
- size
);
4524 xe
->xe_name_offset
= cpu_to_le16(offs
- size
);
4526 memset(val
, 0, size
);
4527 memcpy(val
, xi
->name
, name_len
);
4528 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
), xi
->value
, xi
->value_len
);
4530 xe
->xe_value_size
= cpu_to_le64(xi
->value_len
);
4531 ocfs2_xattr_set_local(xe
, local
);
4533 le16_add_cpu(&xh
->xh_free_start
, -size
);
4534 le16_add_cpu(&xh
->xh_name_value_len
, size
);
4540 * Set the xattr entry in the specified bucket.
4541 * The bucket is indicated by xs->bucket and it should have the enough
4542 * space for the xattr insertion.
4544 static int ocfs2_xattr_set_entry_in_bucket(struct inode
*inode
,
4546 struct ocfs2_xattr_info
*xi
,
4547 struct ocfs2_xattr_search
*xs
,
4554 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4555 (unsigned long)xi
->value_len
, xi
->name_index
,
4556 (unsigned long long)bucket_blkno(xs
->bucket
));
4558 if (!xs
->bucket
->bu_bhs
[1]) {
4559 blkno
= bucket_blkno(xs
->bucket
);
4560 ocfs2_xattr_bucket_relse(xs
->bucket
);
4561 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, blkno
);
4568 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4569 OCFS2_JOURNAL_ACCESS_WRITE
);
4575 ocfs2_xattr_set_entry_normal(inode
, xi
, xs
, name_hash
, local
);
4576 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4582 static int ocfs2_xattr_value_update_size(struct inode
*inode
,
4584 struct buffer_head
*xe_bh
,
4585 struct ocfs2_xattr_entry
*xe
,
4590 ret
= ocfs2_journal_access(handle
, inode
, xe_bh
,
4591 OCFS2_JOURNAL_ACCESS_WRITE
);
4597 xe
->xe_value_size
= cpu_to_le64(new_size
);
4599 ret
= ocfs2_journal_dirty(handle
, xe_bh
);
4608 * Truncate the specified xe_off entry in xattr bucket.
4609 * bucket is indicated by header_bh and len is the new length.
4610 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4612 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4614 static int ocfs2_xattr_bucket_value_truncate(struct inode
*inode
,
4615 struct buffer_head
*header_bh
,
4618 struct ocfs2_xattr_set_ctxt
*ctxt
)
4622 struct buffer_head
*value_bh
= NULL
;
4623 struct ocfs2_xattr_value_root
*xv
;
4624 struct ocfs2_xattr_entry
*xe
;
4625 struct ocfs2_xattr_header
*xh
=
4626 (struct ocfs2_xattr_header
*)header_bh
->b_data
;
4627 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4629 xe
= &xh
->xh_entries
[xe_off
];
4631 BUG_ON(!xe
|| ocfs2_xattr_is_local(xe
));
4633 offset
= le16_to_cpu(xe
->xe_name_offset
) +
4634 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
4636 value_blk
= offset
/ blocksize
;
4638 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4639 BUG_ON(value_blk
!= (offset
+ OCFS2_XATTR_ROOT_SIZE
- 1) / blocksize
);
4640 value_blk
+= header_bh
->b_blocknr
;
4642 ret
= ocfs2_read_block(inode
, value_blk
, &value_bh
, NULL
);
4648 xv
= (struct ocfs2_xattr_value_root
*)
4649 (value_bh
->b_data
+ offset
% blocksize
);
4651 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4652 xe_off
, (unsigned long long)header_bh
->b_blocknr
, len
);
4653 ret
= ocfs2_xattr_value_truncate(inode
, value_bh
, xv
, len
, ctxt
);
4659 ret
= ocfs2_xattr_value_update_size(inode
, ctxt
->handle
,
4660 header_bh
, xe
, len
);
4671 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode
*inode
,
4672 struct ocfs2_xattr_search
*xs
,
4674 struct ocfs2_xattr_set_ctxt
*ctxt
)
4677 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4678 struct ocfs2_xattr_header
*xh
= (struct ocfs2_xattr_header
*)xs
->base
;
4680 BUG_ON(!xs
->bucket
->bu_bhs
[0] || !xe
|| ocfs2_xattr_is_local(xe
));
4682 offset
= xe
- xh
->xh_entries
;
4683 ret
= ocfs2_xattr_bucket_value_truncate(inode
, xs
->bucket
->bu_bhs
[0],
4691 static int ocfs2_xattr_bucket_set_value_outside(struct inode
*inode
,
4693 struct ocfs2_xattr_search
*xs
,
4698 struct ocfs2_xattr_value_root
*xv
;
4699 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4701 BUG_ON(!xs
->base
|| !xe
|| ocfs2_xattr_is_local(xe
));
4703 offset
= le16_to_cpu(xe
->xe_name_offset
) +
4704 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
4706 xv
= (struct ocfs2_xattr_value_root
*)(xs
->base
+ offset
);
4708 return __ocfs2_xattr_set_value_outside(inode
, handle
,
4709 xv
, val
, value_len
);
4712 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
4713 struct buffer_head
*root_bh
,
4719 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4720 struct inode
*tl_inode
= osb
->osb_tl_inode
;
4722 struct ocfs2_xattr_block
*xb
=
4723 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
4724 struct ocfs2_alloc_context
*meta_ac
= NULL
;
4725 struct ocfs2_cached_dealloc_ctxt dealloc
;
4726 struct ocfs2_extent_tree et
;
4728 ocfs2_init_xattr_tree_extent_tree(&et
, inode
, root_bh
);
4730 ocfs2_init_dealloc_ctxt(&dealloc
);
4732 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4733 cpos
, len
, (unsigned long long)blkno
);
4735 ocfs2_remove_xattr_clusters_from_cache(inode
, blkno
, len
);
4737 ret
= ocfs2_lock_allocators(inode
, &et
, 0, 1, NULL
, &meta_ac
);
4743 mutex_lock(&tl_inode
->i_mutex
);
4745 if (ocfs2_truncate_log_needs_flush(osb
)) {
4746 ret
= __ocfs2_flush_truncate_log(osb
);
4753 handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
4754 if (IS_ERR(handle
)) {
4760 ret
= ocfs2_journal_access(handle
, inode
, root_bh
,
4761 OCFS2_JOURNAL_ACCESS_WRITE
);
4767 ret
= ocfs2_remove_extent(inode
, &et
, cpos
, len
, handle
, meta_ac
,
4774 le32_add_cpu(&xb
->xb_attrs
.xb_root
.xt_clusters
, -len
);
4776 ret
= ocfs2_journal_dirty(handle
, root_bh
);
4782 ret
= ocfs2_truncate_log_append(osb
, handle
, blkno
, len
);
4787 ocfs2_commit_trans(osb
, handle
);
4789 ocfs2_schedule_truncate_log_flush(osb
, 1);
4791 mutex_unlock(&tl_inode
->i_mutex
);
4794 ocfs2_free_alloc_context(meta_ac
);
4796 ocfs2_run_deallocs(osb
, &dealloc
);
4801 static void ocfs2_xattr_bucket_remove_xs(struct inode
*inode
,
4803 struct ocfs2_xattr_search
*xs
)
4805 struct ocfs2_xattr_header
*xh
= bucket_xh(xs
->bucket
);
4806 struct ocfs2_xattr_entry
*last
= &xh
->xh_entries
[
4807 le16_to_cpu(xh
->xh_count
) - 1];
4810 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4811 OCFS2_JOURNAL_ACCESS_WRITE
);
4817 /* Remove the old entry. */
4818 memmove(xs
->here
, xs
->here
+ 1,
4819 (void *)last
- (void *)xs
->here
);
4820 memset(last
, 0, sizeof(struct ocfs2_xattr_entry
));
4821 le16_add_cpu(&xh
->xh_count
, -1);
4823 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4827 * Set the xattr name/value in the bucket specified in xs.
4829 * As the new value in xi may be stored in the bucket or in an outside cluster,
4830 * we divide the whole process into 3 steps:
4831 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4832 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4833 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4834 * 4. If the clusters for the new outside value can't be allocated, we need
4835 * to free the xattr we allocated in set.
4837 static int ocfs2_xattr_set_in_bucket(struct inode
*inode
,
4838 struct ocfs2_xattr_info
*xi
,
4839 struct ocfs2_xattr_search
*xs
,
4840 struct ocfs2_xattr_set_ctxt
*ctxt
)
4844 char *val
= (char *)xi
->value
;
4845 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4846 u32 name_hash
= ocfs2_xattr_name_hash(inode
, xi
->name
,
4849 if (!xs
->not_found
&& !ocfs2_xattr_is_local(xe
)) {
4851 * We need to truncate the xattr storage first.
4853 * If both the old and new value are stored to
4854 * outside block, we only need to truncate
4855 * the storage and then set the value outside.
4857 * If the new value should be stored within block,
4858 * we should free all the outside block first and
4859 * the modification to the xattr block will be done
4860 * by following steps.
4862 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
4863 value_len
= xi
->value_len
;
4867 ret
= ocfs2_xattr_bucket_value_truncate_xs(inode
, xs
,
4874 goto set_value_outside
;
4877 value_len
= xi
->value_len
;
4878 /* So we have to handle the inside block change now. */
4879 if (value_len
> OCFS2_XATTR_INLINE_SIZE
) {
4881 * If the new value will be stored outside of block,
4882 * initalize a new empty value root and insert it first.
4885 xi
->value
= &def_xv
;
4886 xi
->value_len
= OCFS2_XATTR_ROOT_SIZE
;
4889 ret
= ocfs2_xattr_set_entry_in_bucket(inode
, ctxt
->handle
, xi
, xs
,
4896 if (value_len
<= OCFS2_XATTR_INLINE_SIZE
)
4899 /* allocate the space now for the outside block storage. */
4900 ret
= ocfs2_xattr_bucket_value_truncate_xs(inode
, xs
,
4905 if (xs
->not_found
) {
4907 * We can't allocate enough clusters for outside
4908 * storage and we have allocated xattr already,
4909 * so need to remove it.
4911 ocfs2_xattr_bucket_remove_xs(inode
, ctxt
->handle
, xs
);
4917 ret
= ocfs2_xattr_bucket_set_value_outside(inode
, ctxt
->handle
,
4918 xs
, val
, value_len
);
4924 * check whether the xattr bucket is filled up with the same hash value.
4925 * If we want to insert the xattr with the same hash, return -ENOSPC.
4926 * If we want to insert a xattr with different hash value, go ahead
4927 * and ocfs2_divide_xattr_bucket will handle this.
4929 static int ocfs2_check_xattr_bucket_collision(struct inode
*inode
,
4930 struct ocfs2_xattr_bucket
*bucket
,
4933 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
4934 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
4936 if (name_hash
!= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
))
4939 if (xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1].xe_name_hash
==
4940 xh
->xh_entries
[0].xe_name_hash
) {
4941 mlog(ML_ERROR
, "Too much hash collision in xattr bucket %llu, "
4943 (unsigned long long)bucket_blkno(bucket
),
4944 le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
));
4951 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
4952 struct ocfs2_xattr_info
*xi
,
4953 struct ocfs2_xattr_search
*xs
,
4954 struct ocfs2_xattr_set_ctxt
*ctxt
)
4956 struct ocfs2_xattr_header
*xh
;
4957 struct ocfs2_xattr_entry
*xe
;
4958 u16 count
, header_size
, xh_free_start
;
4959 int free
, max_free
, need
, old
;
4960 size_t value_size
= 0, name_len
= strlen(xi
->name
);
4961 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4962 int ret
, allocation
= 0;
4964 mlog_entry("Set xattr %s in xattr index block\n", xi
->name
);
4968 count
= le16_to_cpu(xh
->xh_count
);
4969 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
4970 header_size
= sizeof(struct ocfs2_xattr_header
) +
4971 count
* sizeof(struct ocfs2_xattr_entry
);
4972 max_free
= OCFS2_XATTR_BUCKET_SIZE
-
4973 le16_to_cpu(xh
->xh_name_value_len
) - header_size
;
4975 mlog_bug_on_msg(header_size
> blocksize
, "bucket %llu has header size "
4976 "of %u which exceed block size\n",
4977 (unsigned long long)bucket_blkno(xs
->bucket
),
4980 if (xi
->value
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
4981 value_size
= OCFS2_XATTR_ROOT_SIZE
;
4983 value_size
= OCFS2_XATTR_SIZE(xi
->value_len
);
4986 need
= sizeof(struct ocfs2_xattr_entry
) +
4987 OCFS2_XATTR_SIZE(name_len
) + value_size
;
4989 need
= value_size
+ OCFS2_XATTR_SIZE(name_len
);
4992 * We only replace the old value if the new length is smaller
4993 * than the old one. Otherwise we will allocate new space in the
4994 * bucket to store it.
4997 if (ocfs2_xattr_is_local(xe
))
4998 old
= OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
5000 old
= OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE
);
5002 if (old
>= value_size
)
5006 free
= xh_free_start
- header_size
;
5008 * We need to make sure the new name/value pair
5009 * can exist in the same block.
5011 if (xh_free_start
% blocksize
< need
)
5012 free
-= xh_free_start
% blocksize
;
5014 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5015 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5016 " %u\n", xs
->not_found
,
5017 (unsigned long long)bucket_blkno(xs
->bucket
),
5018 free
, need
, max_free
, le16_to_cpu(xh
->xh_free_start
),
5019 le16_to_cpu(xh
->xh_name_value_len
));
5023 count
== ocfs2_xattr_max_xe_in_bucket(inode
->i_sb
))) {
5024 if (need
<= max_free
&&
5025 count
< ocfs2_xattr_max_xe_in_bucket(inode
->i_sb
)) {
5027 * We can create the space by defragment. Since only the
5028 * name/value will be moved, the xe shouldn't be changed
5031 ret
= ocfs2_defrag_xattr_bucket(inode
, ctxt
->handle
,
5038 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
5039 free
= xh_free_start
- header_size
;
5040 if (xh_free_start
% blocksize
< need
)
5041 free
-= xh_free_start
% blocksize
;
5046 mlog(0, "Can't get enough space for xattr insert by "
5047 "defragment. Need %u bytes, but we have %d, so "
5048 "allocate new bucket for it.\n", need
, free
);
5052 * We have to add new buckets or clusters and one
5053 * allocation should leave us enough space for insert.
5058 * We do not allow for overlapping ranges between buckets. And
5059 * the maximum number of collisions we will allow for then is
5060 * one bucket's worth, so check it here whether we need to
5061 * add a new bucket for the insert.
5063 ret
= ocfs2_check_xattr_bucket_collision(inode
,
5071 ret
= ocfs2_add_new_xattr_bucket(inode
,
5073 xs
->bucket
->bu_bhs
[0],
5080 ocfs2_xattr_bucket_relse(xs
->bucket
);
5082 ret
= ocfs2_xattr_index_block_find(inode
, xs
->xattr_bh
,
5085 if (ret
&& ret
!= -ENODATA
)
5087 xs
->not_found
= ret
;
5093 ret
= ocfs2_xattr_set_in_bucket(inode
, xi
, xs
, ctxt
);
5099 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
5100 struct ocfs2_xattr_bucket
*bucket
,
5104 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5106 struct ocfs2_xattr_entry
*xe
;
5107 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5108 struct ocfs2_xattr_set_ctxt ctxt
= {NULL
, NULL
,};
5110 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
5112 ctxt
.handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
5113 if (IS_ERR(ctxt
.handle
)) {
5114 ret
= PTR_ERR(ctxt
.handle
);
5119 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
5120 xe
= &xh
->xh_entries
[i
];
5121 if (ocfs2_xattr_is_local(xe
))
5124 ret
= ocfs2_xattr_bucket_value_truncate(inode
,
5133 ret
= ocfs2_commit_trans(osb
, ctxt
.handle
);
5134 ocfs2_schedule_truncate_log_flush(osb
, 1);
5135 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
5140 static int ocfs2_delete_xattr_index_block(struct inode
*inode
,
5141 struct buffer_head
*xb_bh
)
5143 struct ocfs2_xattr_block
*xb
=
5144 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
5145 struct ocfs2_extent_list
*el
= &xb
->xb_attrs
.xb_root
.xt_list
;
5147 u32 name_hash
= UINT_MAX
, e_cpos
, num_clusters
;
5150 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
5153 while (name_hash
> 0) {
5154 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
5155 &e_cpos
, &num_clusters
, el
);
5161 ret
= ocfs2_iterate_xattr_buckets(inode
, p_blkno
, num_clusters
,
5162 ocfs2_delete_xattr_in_bucket
,
5169 ret
= ocfs2_rm_xattr_cluster(inode
, xb_bh
,
5170 p_blkno
, e_cpos
, num_clusters
);
5179 name_hash
= e_cpos
- 1;
5187 * 'security' attributes support
5189 static size_t ocfs2_xattr_security_list(struct inode
*inode
, char *list
,
5190 size_t list_size
, const char *name
,
5193 const size_t prefix_len
= XATTR_SECURITY_PREFIX_LEN
;
5194 const size_t total_len
= prefix_len
+ name_len
+ 1;
5196 if (list
&& total_len
<= list_size
) {
5197 memcpy(list
, XATTR_SECURITY_PREFIX
, prefix_len
);
5198 memcpy(list
+ prefix_len
, name
, name_len
);
5199 list
[prefix_len
+ name_len
] = '\0';
5204 static int ocfs2_xattr_security_get(struct inode
*inode
, const char *name
,
5205 void *buffer
, size_t size
)
5207 if (strcmp(name
, "") == 0)
5209 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_SECURITY
, name
,
5213 static int ocfs2_xattr_security_set(struct inode
*inode
, const char *name
,
5214 const void *value
, size_t size
, int flags
)
5216 if (strcmp(name
, "") == 0)
5219 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_SECURITY
, name
, value
,
5223 int ocfs2_init_security_get(struct inode
*inode
,
5225 struct ocfs2_security_xattr_info
*si
)
5227 return security_inode_init_security(inode
, dir
, &si
->name
, &si
->value
,
5231 int ocfs2_init_security_set(handle_t
*handle
,
5232 struct inode
*inode
,
5233 struct buffer_head
*di_bh
,
5234 struct ocfs2_security_xattr_info
*si
,
5235 struct ocfs2_alloc_context
*xattr_ac
,
5236 struct ocfs2_alloc_context
*data_ac
)
5238 return ocfs2_xattr_set_handle(handle
, inode
, di_bh
,
5239 OCFS2_XATTR_INDEX_SECURITY
,
5240 si
->name
, si
->value
, si
->value_len
, 0,
5244 struct xattr_handler ocfs2_xattr_security_handler
= {
5245 .prefix
= XATTR_SECURITY_PREFIX
,
5246 .list
= ocfs2_xattr_security_list
,
5247 .get
= ocfs2_xattr_security_get
,
5248 .set
= ocfs2_xattr_security_set
,
5252 * 'trusted' attributes support
5254 static size_t ocfs2_xattr_trusted_list(struct inode
*inode
, char *list
,
5255 size_t list_size
, const char *name
,
5258 const size_t prefix_len
= XATTR_TRUSTED_PREFIX_LEN
;
5259 const size_t total_len
= prefix_len
+ name_len
+ 1;
5261 if (list
&& total_len
<= list_size
) {
5262 memcpy(list
, XATTR_TRUSTED_PREFIX
, prefix_len
);
5263 memcpy(list
+ prefix_len
, name
, name_len
);
5264 list
[prefix_len
+ name_len
] = '\0';
5269 static int ocfs2_xattr_trusted_get(struct inode
*inode
, const char *name
,
5270 void *buffer
, size_t size
)
5272 if (strcmp(name
, "") == 0)
5274 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_TRUSTED
, name
,
5278 static int ocfs2_xattr_trusted_set(struct inode
*inode
, const char *name
,
5279 const void *value
, size_t size
, int flags
)
5281 if (strcmp(name
, "") == 0)
5284 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_TRUSTED
, name
, value
,
5288 struct xattr_handler ocfs2_xattr_trusted_handler
= {
5289 .prefix
= XATTR_TRUSTED_PREFIX
,
5290 .list
= ocfs2_xattr_trusted_list
,
5291 .get
= ocfs2_xattr_trusted_get
,
5292 .set
= ocfs2_xattr_trusted_set
,
5296 * 'user' attributes support
5298 static size_t ocfs2_xattr_user_list(struct inode
*inode
, char *list
,
5299 size_t list_size
, const char *name
,
5302 const size_t prefix_len
= XATTR_USER_PREFIX_LEN
;
5303 const size_t total_len
= prefix_len
+ name_len
+ 1;
5304 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5306 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5309 if (list
&& total_len
<= list_size
) {
5310 memcpy(list
, XATTR_USER_PREFIX
, prefix_len
);
5311 memcpy(list
+ prefix_len
, name
, name_len
);
5312 list
[prefix_len
+ name_len
] = '\0';
5317 static int ocfs2_xattr_user_get(struct inode
*inode
, const char *name
,
5318 void *buffer
, size_t size
)
5320 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5322 if (strcmp(name
, "") == 0)
5324 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5326 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_USER
, name
,
5330 static int ocfs2_xattr_user_set(struct inode
*inode
, const char *name
,
5331 const void *value
, size_t size
, int flags
)
5333 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5335 if (strcmp(name
, "") == 0)
5337 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5340 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_USER
, name
, value
,
5344 struct xattr_handler ocfs2_xattr_user_handler
= {
5345 .prefix
= XATTR_USER_PREFIX
,
5346 .list
= ocfs2_xattr_user_list
,
5347 .get
= ocfs2_xattr_user_get
,
5348 .set
= ocfs2_xattr_user_set
,