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);
271 ocfs2_xattr_bucket_relse(bucket
);
275 static int ocfs2_xattr_bucket_journal_access(handle_t
*handle
,
276 struct ocfs2_xattr_bucket
*bucket
,
281 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
282 rc
= ocfs2_journal_access(handle
, bucket
->bu_inode
,
283 bucket
->bu_bhs
[i
], type
);
293 static void ocfs2_xattr_bucket_journal_dirty(handle_t
*handle
,
294 struct ocfs2_xattr_bucket
*bucket
)
298 for (i
= 0; i
< bucket
->bu_blocks
; i
++)
299 ocfs2_journal_dirty(handle
, bucket
->bu_bhs
[i
]);
302 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket
*dest
,
303 struct ocfs2_xattr_bucket
*src
)
306 int blocksize
= src
->bu_inode
->i_sb
->s_blocksize
;
308 BUG_ON(dest
->bu_blocks
!= src
->bu_blocks
);
309 BUG_ON(dest
->bu_inode
!= src
->bu_inode
);
311 for (i
= 0; i
< src
->bu_blocks
; i
++) {
312 memcpy(bucket_block(dest
, i
), bucket_block(src
, i
),
317 static inline const char *ocfs2_xattr_prefix(int name_index
)
319 struct xattr_handler
*handler
= NULL
;
321 if (name_index
> 0 && name_index
< OCFS2_XATTR_MAX
)
322 handler
= ocfs2_xattr_handler_map
[name_index
];
324 return handler
? handler
->prefix
: NULL
;
327 static u32
ocfs2_xattr_name_hash(struct inode
*inode
,
331 /* Get hash value of uuid from super block */
332 u32 hash
= OCFS2_SB(inode
->i_sb
)->uuid_hash
;
335 /* hash extended attribute name */
336 for (i
= 0; i
< name_len
; i
++) {
337 hash
= (hash
<< OCFS2_HASH_SHIFT
) ^
338 (hash
>> (8*sizeof(hash
) - OCFS2_HASH_SHIFT
)) ^
346 * ocfs2_xattr_hash_entry()
348 * Compute the hash of an extended attribute.
350 static void ocfs2_xattr_hash_entry(struct inode
*inode
,
351 struct ocfs2_xattr_header
*header
,
352 struct ocfs2_xattr_entry
*entry
)
355 char *name
= (char *)header
+ le16_to_cpu(entry
->xe_name_offset
);
357 hash
= ocfs2_xattr_name_hash(inode
, name
, entry
->xe_name_len
);
358 entry
->xe_name_hash
= cpu_to_le32(hash
);
363 static int ocfs2_xattr_entry_real_size(int name_len
, size_t value_len
)
367 if (value_len
<= OCFS2_XATTR_INLINE_SIZE
)
368 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(value_len
);
370 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
371 size
+= sizeof(struct ocfs2_xattr_entry
);
376 int ocfs2_calc_security_init(struct inode
*dir
,
377 struct ocfs2_security_xattr_info
*si
,
380 struct ocfs2_alloc_context
**xattr_ac
)
383 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
384 int s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
388 * The max space of security xattr taken inline is
389 * 256(name) + 80(value) + 16(entry) = 352 bytes,
390 * So reserve one metadata block for it is ok.
392 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
393 s_size
> OCFS2_XATTR_FREE_IN_IBODY
) {
394 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
399 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
402 /* reserve clusters for xattr value which will be set in B tree*/
403 if (si
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
404 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
,
409 int ocfs2_calc_xattr_init(struct inode
*dir
,
410 struct buffer_head
*dir_bh
,
412 struct ocfs2_security_xattr_info
*si
,
415 struct ocfs2_alloc_context
**xattr_ac
)
418 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
424 s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
427 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
) {
428 acl_len
= ocfs2_xattr_get_nolock(dir
, dir_bh
,
429 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
,
432 a_size
= ocfs2_xattr_entry_real_size(0, acl_len
);
435 } else if (acl_len
!= 0 && acl_len
!= -ENODATA
) {
441 if (!(s_size
+ a_size
))
445 * The max space of security xattr taken inline is
446 * 256(name) + 80(value) + 16(entry) = 352 bytes,
447 * The max space of acl xattr taken inline is
448 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
449 * when blocksize = 512, may reserve one more cluser for
450 * xattr bucket, otherwise reserve one metadata block
453 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
454 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_IBODY
) {
455 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
460 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
463 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
&&
464 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_BLOCK(dir
)) {
466 *xattr_credits
+= ocfs2_blocks_per_xattr_bucket(dir
->i_sb
);
469 /* reserve clusters for xattr value which will be set in B tree*/
470 if (si
->enable
&& si
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
471 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
,
473 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
&&
474 acl_len
> OCFS2_XATTR_INLINE_SIZE
) {
475 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
, acl_len
);
477 *want_clusters
+= ocfs2_clusters_for_bytes(dir
->i_sb
,
484 static int ocfs2_xattr_extend_allocation(struct inode
*inode
,
486 struct buffer_head
*xattr_bh
,
487 struct ocfs2_xattr_value_root
*xv
,
488 struct ocfs2_xattr_set_ctxt
*ctxt
)
491 handle_t
*handle
= ctxt
->handle
;
492 enum ocfs2_alloc_restarted why
;
493 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
494 u32 prev_clusters
, logical_start
= le32_to_cpu(xv
->xr_clusters
);
495 struct ocfs2_extent_tree et
;
497 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add
);
499 ocfs2_init_xattr_value_extent_tree(&et
, inode
, xattr_bh
, xv
);
501 status
= ocfs2_journal_access(handle
, inode
, xattr_bh
,
502 OCFS2_JOURNAL_ACCESS_WRITE
);
508 prev_clusters
= le32_to_cpu(xv
->xr_clusters
);
509 status
= ocfs2_add_clusters_in_btree(osb
,
524 status
= ocfs2_journal_dirty(handle
, xattr_bh
);
530 clusters_to_add
-= le32_to_cpu(xv
->xr_clusters
) - prev_clusters
;
533 * We should have already allocated enough space before the transaction,
534 * so no need to restart.
536 BUG_ON(why
!= RESTART_NONE
|| clusters_to_add
);
543 static int __ocfs2_remove_xattr_range(struct inode
*inode
,
544 struct buffer_head
*root_bh
,
545 struct ocfs2_xattr_value_root
*xv
,
546 u32 cpos
, u32 phys_cpos
, u32 len
,
547 struct ocfs2_xattr_set_ctxt
*ctxt
)
550 u64 phys_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
551 handle_t
*handle
= ctxt
->handle
;
552 struct ocfs2_extent_tree et
;
554 ocfs2_init_xattr_value_extent_tree(&et
, inode
, root_bh
, xv
);
556 ret
= ocfs2_journal_access(handle
, inode
, root_bh
,
557 OCFS2_JOURNAL_ACCESS_WRITE
);
563 ret
= ocfs2_remove_extent(inode
, &et
, cpos
, len
, handle
, ctxt
->meta_ac
,
570 le32_add_cpu(&xv
->xr_clusters
, -len
);
572 ret
= ocfs2_journal_dirty(handle
, root_bh
);
578 ret
= ocfs2_cache_cluster_dealloc(&ctxt
->dealloc
, phys_blkno
, len
);
586 static int ocfs2_xattr_shrink_size(struct inode
*inode
,
589 struct buffer_head
*root_bh
,
590 struct ocfs2_xattr_value_root
*xv
,
591 struct ocfs2_xattr_set_ctxt
*ctxt
)
594 u32 trunc_len
, cpos
, phys_cpos
, alloc_size
;
597 if (old_clusters
<= new_clusters
)
601 trunc_len
= old_clusters
- new_clusters
;
603 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &phys_cpos
,
604 &alloc_size
, &xv
->xr_list
);
610 if (alloc_size
> trunc_len
)
611 alloc_size
= trunc_len
;
613 ret
= __ocfs2_remove_xattr_range(inode
, root_bh
, xv
, cpos
,
614 phys_cpos
, alloc_size
,
621 block
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
622 ocfs2_remove_xattr_clusters_from_cache(inode
, block
,
625 trunc_len
-= alloc_size
;
632 static int ocfs2_xattr_value_truncate(struct inode
*inode
,
633 struct buffer_head
*root_bh
,
634 struct ocfs2_xattr_value_root
*xv
,
636 struct ocfs2_xattr_set_ctxt
*ctxt
)
639 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
640 u32 old_clusters
= le32_to_cpu(xv
->xr_clusters
);
642 if (new_clusters
== old_clusters
)
645 if (new_clusters
> old_clusters
)
646 ret
= ocfs2_xattr_extend_allocation(inode
,
647 new_clusters
- old_clusters
,
650 ret
= ocfs2_xattr_shrink_size(inode
,
651 old_clusters
, new_clusters
,
657 static int ocfs2_xattr_list_entry(char *buffer
, size_t size
,
658 size_t *result
, const char *prefix
,
659 const char *name
, int name_len
)
661 char *p
= buffer
+ *result
;
662 int prefix_len
= strlen(prefix
);
663 int total_len
= prefix_len
+ name_len
+ 1;
665 *result
+= total_len
;
667 /* we are just looking for how big our buffer needs to be */
674 memcpy(p
, prefix
, prefix_len
);
675 memcpy(p
+ prefix_len
, name
, name_len
);
676 p
[prefix_len
+ name_len
] = '\0';
681 static int ocfs2_xattr_list_entries(struct inode
*inode
,
682 struct ocfs2_xattr_header
*header
,
683 char *buffer
, size_t buffer_size
)
687 const char *prefix
, *name
;
689 for (i
= 0 ; i
< le16_to_cpu(header
->xh_count
); i
++) {
690 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
691 type
= ocfs2_xattr_get_type(entry
);
692 prefix
= ocfs2_xattr_prefix(type
);
695 name
= (const char *)header
+
696 le16_to_cpu(entry
->xe_name_offset
);
698 ret
= ocfs2_xattr_list_entry(buffer
, buffer_size
,
699 &result
, prefix
, name
,
709 static int ocfs2_xattr_ibody_list(struct inode
*inode
,
710 struct ocfs2_dinode
*di
,
714 struct ocfs2_xattr_header
*header
= NULL
;
715 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
718 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
721 header
= (struct ocfs2_xattr_header
*)
722 ((void *)di
+ inode
->i_sb
->s_blocksize
-
723 le16_to_cpu(di
->i_xattr_inline_size
));
725 ret
= ocfs2_xattr_list_entries(inode
, header
, buffer
, buffer_size
);
730 static int ocfs2_xattr_block_list(struct inode
*inode
,
731 struct ocfs2_dinode
*di
,
735 struct buffer_head
*blk_bh
= NULL
;
736 struct ocfs2_xattr_block
*xb
;
739 if (!di
->i_xattr_loc
)
742 ret
= ocfs2_read_block(inode
, le64_to_cpu(di
->i_xattr_loc
), &blk_bh
);
748 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
749 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
754 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
755 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
756 ret
= ocfs2_xattr_list_entries(inode
, header
,
757 buffer
, buffer_size
);
759 struct ocfs2_xattr_tree_root
*xt
= &xb
->xb_attrs
.xb_root
;
760 ret
= ocfs2_xattr_tree_list_index_block(inode
, xt
,
761 buffer
, buffer_size
);
769 ssize_t
ocfs2_listxattr(struct dentry
*dentry
,
773 int ret
= 0, i_ret
= 0, b_ret
= 0;
774 struct buffer_head
*di_bh
= NULL
;
775 struct ocfs2_dinode
*di
= NULL
;
776 struct ocfs2_inode_info
*oi
= OCFS2_I(dentry
->d_inode
);
778 if (!ocfs2_supports_xattr(OCFS2_SB(dentry
->d_sb
)))
781 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
784 ret
= ocfs2_inode_lock(dentry
->d_inode
, &di_bh
, 0);
790 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
792 down_read(&oi
->ip_xattr_sem
);
793 i_ret
= ocfs2_xattr_ibody_list(dentry
->d_inode
, di
, buffer
, size
);
801 b_ret
= ocfs2_xattr_block_list(dentry
->d_inode
, di
,
806 up_read(&oi
->ip_xattr_sem
);
807 ocfs2_inode_unlock(dentry
->d_inode
, 0);
811 return i_ret
+ b_ret
;
814 static int ocfs2_xattr_find_entry(int name_index
,
816 struct ocfs2_xattr_search
*xs
)
818 struct ocfs2_xattr_entry
*entry
;
825 name_len
= strlen(name
);
827 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
828 cmp
= name_index
- ocfs2_xattr_get_type(entry
);
830 cmp
= name_len
- entry
->xe_name_len
;
832 cmp
= memcmp(name
, (xs
->base
+
833 le16_to_cpu(entry
->xe_name_offset
)),
841 return cmp
? -ENODATA
: 0;
844 static int ocfs2_xattr_get_value_outside(struct inode
*inode
,
845 struct ocfs2_xattr_value_root
*xv
,
849 u32 cpos
, p_cluster
, num_clusters
, bpc
, clusters
;
852 size_t cplen
, blocksize
;
853 struct buffer_head
*bh
= NULL
;
854 struct ocfs2_extent_list
*el
;
857 clusters
= le32_to_cpu(xv
->xr_clusters
);
858 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
859 blocksize
= inode
->i_sb
->s_blocksize
;
862 while (cpos
< clusters
) {
863 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
870 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
871 /* Copy ocfs2_xattr_value */
872 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
873 ret
= ocfs2_read_block(inode
, blkno
, &bh
);
879 cplen
= len
>= blocksize
? blocksize
: len
;
880 memcpy(buffer
, bh
->b_data
, cplen
);
889 cpos
+= num_clusters
;
895 static int ocfs2_xattr_ibody_get(struct inode
*inode
,
900 struct ocfs2_xattr_search
*xs
)
902 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
903 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
904 struct ocfs2_xattr_value_root
*xv
;
908 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
911 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
912 xs
->header
= (struct ocfs2_xattr_header
*)
913 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
914 xs
->base
= (void *)xs
->header
;
915 xs
->here
= xs
->header
->xh_entries
;
917 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
920 size
= le64_to_cpu(xs
->here
->xe_value_size
);
922 if (size
> buffer_size
)
924 if (ocfs2_xattr_is_local(xs
->here
)) {
925 memcpy(buffer
, (void *)xs
->base
+
926 le16_to_cpu(xs
->here
->xe_name_offset
) +
927 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
), size
);
929 xv
= (struct ocfs2_xattr_value_root
*)
930 (xs
->base
+ le16_to_cpu(
931 xs
->here
->xe_name_offset
) +
932 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
));
933 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
945 static int ocfs2_xattr_block_get(struct inode
*inode
,
950 struct ocfs2_xattr_search
*xs
)
952 struct ocfs2_xattr_block
*xb
;
953 struct ocfs2_xattr_value_root
*xv
;
955 int ret
= -ENODATA
, name_offset
, name_len
, block_off
, i
;
957 xs
->bucket
= ocfs2_xattr_bucket_new(inode
);
964 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, xs
);
975 xb
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
976 size
= le64_to_cpu(xs
->here
->xe_value_size
);
979 if (size
> buffer_size
)
982 name_offset
= le16_to_cpu(xs
->here
->xe_name_offset
);
983 name_len
= OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
);
984 i
= xs
->here
- xs
->header
->xh_entries
;
986 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
987 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
988 bucket_xh(xs
->bucket
),
992 xs
->base
= bucket_block(xs
->bucket
, block_off
);
994 if (ocfs2_xattr_is_local(xs
->here
)) {
995 memcpy(buffer
, (void *)xs
->base
+
996 name_offset
+ name_len
, size
);
998 xv
= (struct ocfs2_xattr_value_root
*)
999 (xs
->base
+ name_offset
+ name_len
);
1000 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1010 ocfs2_xattr_bucket_free(xs
->bucket
);
1012 brelse(xs
->xattr_bh
);
1013 xs
->xattr_bh
= NULL
;
1017 int ocfs2_xattr_get_nolock(struct inode
*inode
,
1018 struct buffer_head
*di_bh
,
1025 struct ocfs2_dinode
*di
= NULL
;
1026 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1027 struct ocfs2_xattr_search xis
= {
1028 .not_found
= -ENODATA
,
1030 struct ocfs2_xattr_search xbs
= {
1031 .not_found
= -ENODATA
,
1034 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1037 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1040 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
1041 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1043 down_read(&oi
->ip_xattr_sem
);
1044 ret
= ocfs2_xattr_ibody_get(inode
, name_index
, name
, buffer
,
1046 if (ret
== -ENODATA
&& di
->i_xattr_loc
)
1047 ret
= ocfs2_xattr_block_get(inode
, name_index
, name
, buffer
,
1049 up_read(&oi
->ip_xattr_sem
);
1054 /* ocfs2_xattr_get()
1056 * Copy an extended attribute into the buffer provided.
1057 * Buffer is NULL to compute the size of buffer required.
1059 static int ocfs2_xattr_get(struct inode
*inode
,
1066 struct buffer_head
*di_bh
= NULL
;
1068 ret
= ocfs2_inode_lock(inode
, &di_bh
, 0);
1073 ret
= ocfs2_xattr_get_nolock(inode
, di_bh
, name_index
,
1074 name
, buffer
, buffer_size
);
1076 ocfs2_inode_unlock(inode
, 0);
1083 static int __ocfs2_xattr_set_value_outside(struct inode
*inode
,
1085 struct ocfs2_xattr_value_root
*xv
,
1089 int ret
= 0, i
, cp_len
, credits
;
1090 u16 blocksize
= inode
->i_sb
->s_blocksize
;
1091 u32 p_cluster
, num_clusters
;
1092 u32 cpos
= 0, bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1093 u32 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, value_len
);
1095 struct buffer_head
*bh
= NULL
;
1097 BUG_ON(clusters
> le32_to_cpu(xv
->xr_clusters
));
1100 * In __ocfs2_xattr_set_value_outside has already been dirtied,
1101 * so we don't need to worry about whether ocfs2_extend_trans
1102 * will create a new transactio for us or not.
1104 credits
= clusters
* bpc
;
1105 ret
= ocfs2_extend_trans(handle
, credits
);
1111 while (cpos
< clusters
) {
1112 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1113 &num_clusters
, &xv
->xr_list
);
1119 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1121 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1122 ret
= ocfs2_read_block(inode
, blkno
, &bh
);
1128 ret
= ocfs2_journal_access(handle
,
1131 OCFS2_JOURNAL_ACCESS_WRITE
);
1137 cp_len
= value_len
> blocksize
? blocksize
: value_len
;
1138 memcpy(bh
->b_data
, value
, cp_len
);
1139 value_len
-= cp_len
;
1141 if (cp_len
< blocksize
)
1142 memset(bh
->b_data
+ cp_len
, 0,
1143 blocksize
- cp_len
);
1145 ret
= ocfs2_journal_dirty(handle
, bh
);
1154 * XXX: do we need to empty all the following
1155 * blocks in this cluster?
1160 cpos
+= num_clusters
;
1168 static int ocfs2_xattr_cleanup(struct inode
*inode
,
1170 struct ocfs2_xattr_info
*xi
,
1171 struct ocfs2_xattr_search
*xs
,
1175 size_t name_len
= strlen(xi
->name
);
1176 void *val
= xs
->base
+ offs
;
1177 size_t size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1179 ret
= ocfs2_journal_access(handle
, inode
, xs
->xattr_bh
,
1180 OCFS2_JOURNAL_ACCESS_WRITE
);
1185 /* Decrease xattr count */
1186 le16_add_cpu(&xs
->header
->xh_count
, -1);
1187 /* Remove the xattr entry and tree root which has already be set*/
1188 memset((void *)xs
->here
, 0, sizeof(struct ocfs2_xattr_entry
));
1189 memset(val
, 0, size
);
1191 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1198 static int ocfs2_xattr_update_entry(struct inode
*inode
,
1200 struct ocfs2_xattr_info
*xi
,
1201 struct ocfs2_xattr_search
*xs
,
1206 ret
= ocfs2_journal_access(handle
, inode
, xs
->xattr_bh
,
1207 OCFS2_JOURNAL_ACCESS_WRITE
);
1213 xs
->here
->xe_name_offset
= cpu_to_le16(offs
);
1214 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1215 if (xi
->value_len
<= OCFS2_XATTR_INLINE_SIZE
)
1216 ocfs2_xattr_set_local(xs
->here
, 1);
1218 ocfs2_xattr_set_local(xs
->here
, 0);
1219 ocfs2_xattr_hash_entry(inode
, xs
->header
, xs
->here
);
1221 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1229 * ocfs2_xattr_set_value_outside()
1231 * Set large size value in B tree.
1233 static int ocfs2_xattr_set_value_outside(struct inode
*inode
,
1234 struct ocfs2_xattr_info
*xi
,
1235 struct ocfs2_xattr_search
*xs
,
1236 struct ocfs2_xattr_set_ctxt
*ctxt
,
1239 size_t name_len
= strlen(xi
->name
);
1240 void *val
= xs
->base
+ offs
;
1241 struct ocfs2_xattr_value_root
*xv
= NULL
;
1242 size_t size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1245 memset(val
, 0, size
);
1246 memcpy(val
, xi
->name
, name_len
);
1247 xv
= (struct ocfs2_xattr_value_root
*)
1248 (val
+ OCFS2_XATTR_SIZE(name_len
));
1249 xv
->xr_clusters
= 0;
1250 xv
->xr_last_eb_blk
= 0;
1251 xv
->xr_list
.l_tree_depth
= 0;
1252 xv
->xr_list
.l_count
= cpu_to_le16(1);
1253 xv
->xr_list
.l_next_free_rec
= 0;
1255 ret
= ocfs2_xattr_value_truncate(inode
, xs
->xattr_bh
, xv
,
1256 xi
->value_len
, ctxt
);
1261 ret
= ocfs2_xattr_update_entry(inode
, ctxt
->handle
, xi
, xs
, offs
);
1266 ret
= __ocfs2_xattr_set_value_outside(inode
, ctxt
->handle
, xv
,
1267 xi
->value
, xi
->value_len
);
1275 * ocfs2_xattr_set_entry_local()
1277 * Set, replace or remove extended attribute in local.
1279 static void ocfs2_xattr_set_entry_local(struct inode
*inode
,
1280 struct ocfs2_xattr_info
*xi
,
1281 struct ocfs2_xattr_search
*xs
,
1282 struct ocfs2_xattr_entry
*last
,
1285 size_t name_len
= strlen(xi
->name
);
1288 if (xi
->value
&& xs
->not_found
) {
1289 /* Insert the new xattr entry. */
1290 le16_add_cpu(&xs
->header
->xh_count
, 1);
1291 ocfs2_xattr_set_type(last
, xi
->name_index
);
1292 ocfs2_xattr_set_local(last
, 1);
1293 last
->xe_name_len
= name_len
;
1299 first_val
= xs
->base
+ min_offs
;
1300 offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1301 val
= xs
->base
+ offs
;
1303 if (le64_to_cpu(xs
->here
->xe_value_size
) >
1304 OCFS2_XATTR_INLINE_SIZE
)
1305 size
= OCFS2_XATTR_SIZE(name_len
) +
1306 OCFS2_XATTR_ROOT_SIZE
;
1308 size
= OCFS2_XATTR_SIZE(name_len
) +
1309 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1311 if (xi
->value
&& size
== OCFS2_XATTR_SIZE(name_len
) +
1312 OCFS2_XATTR_SIZE(xi
->value_len
)) {
1313 /* The old and the new value have the
1314 same size. Just replace the value. */
1315 ocfs2_xattr_set_local(xs
->here
, 1);
1316 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1317 /* Clear value bytes. */
1318 memset(val
+ OCFS2_XATTR_SIZE(name_len
),
1320 OCFS2_XATTR_SIZE(xi
->value_len
));
1321 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
1326 /* Remove the old name+value. */
1327 memmove(first_val
+ size
, first_val
, val
- first_val
);
1328 memset(first_val
, 0, size
);
1329 xs
->here
->xe_name_hash
= 0;
1330 xs
->here
->xe_name_offset
= 0;
1331 ocfs2_xattr_set_local(xs
->here
, 1);
1332 xs
->here
->xe_value_size
= 0;
1336 /* Adjust all value offsets. */
1337 last
= xs
->header
->xh_entries
;
1338 for (i
= 0 ; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1339 size_t o
= le16_to_cpu(last
->xe_name_offset
);
1342 last
->xe_name_offset
= cpu_to_le16(o
+ size
);
1347 /* Remove the old entry. */
1349 memmove(xs
->here
, xs
->here
+ 1,
1350 (void *)last
- (void *)xs
->here
);
1351 memset(last
, 0, sizeof(struct ocfs2_xattr_entry
));
1352 le16_add_cpu(&xs
->header
->xh_count
, -1);
1356 /* Insert the new name+value. */
1357 size_t size
= OCFS2_XATTR_SIZE(name_len
) +
1358 OCFS2_XATTR_SIZE(xi
->value_len
);
1359 void *val
= xs
->base
+ min_offs
- size
;
1361 xs
->here
->xe_name_offset
= cpu_to_le16(min_offs
- size
);
1362 memset(val
, 0, size
);
1363 memcpy(val
, xi
->name
, name_len
);
1364 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
1367 xs
->here
->xe_value_size
= cpu_to_le64(xi
->value_len
);
1368 ocfs2_xattr_set_local(xs
->here
, 1);
1369 ocfs2_xattr_hash_entry(inode
, xs
->header
, xs
->here
);
1376 * ocfs2_xattr_set_entry()
1378 * Set extended attribute entry into inode or block.
1380 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1381 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1382 * then set value in B tree with set_value_outside().
1384 static int ocfs2_xattr_set_entry(struct inode
*inode
,
1385 struct ocfs2_xattr_info
*xi
,
1386 struct ocfs2_xattr_search
*xs
,
1387 struct ocfs2_xattr_set_ctxt
*ctxt
,
1390 struct ocfs2_xattr_entry
*last
;
1391 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1392 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1393 size_t min_offs
= xs
->end
- xs
->base
, name_len
= strlen(xi
->name
);
1395 handle_t
*handle
= ctxt
->handle
;
1397 struct ocfs2_xattr_info xi_l
= {
1398 .name_index
= xi
->name_index
,
1401 .value_len
= xi
->value_len
,
1404 /* Compute min_offs, last and free space. */
1405 last
= xs
->header
->xh_entries
;
1407 for (i
= 0 ; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1408 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
1409 if (offs
< min_offs
)
1414 free
= min_offs
- ((void *)last
- xs
->base
) - sizeof(__u32
);
1418 if (!xs
->not_found
) {
1420 if (ocfs2_xattr_is_local(xs
->here
))
1421 size
= OCFS2_XATTR_SIZE(name_len
) +
1422 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1424 size
= OCFS2_XATTR_SIZE(name_len
) +
1425 OCFS2_XATTR_ROOT_SIZE
;
1426 free
+= (size
+ sizeof(struct ocfs2_xattr_entry
));
1428 /* Check free space in inode or block */
1429 if (xi
->value
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1430 if (free
< sizeof(struct ocfs2_xattr_entry
) +
1431 OCFS2_XATTR_SIZE(name_len
) +
1432 OCFS2_XATTR_ROOT_SIZE
) {
1436 size_l
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
1437 xi_l
.value
= (void *)&def_xv
;
1438 xi_l
.value_len
= OCFS2_XATTR_ROOT_SIZE
;
1439 } else if (xi
->value
) {
1440 if (free
< sizeof(struct ocfs2_xattr_entry
) +
1441 OCFS2_XATTR_SIZE(name_len
) +
1442 OCFS2_XATTR_SIZE(xi
->value_len
)) {
1448 if (!xs
->not_found
) {
1449 /* For existing extended attribute */
1450 size_t size
= OCFS2_XATTR_SIZE(name_len
) +
1451 OCFS2_XATTR_SIZE(le64_to_cpu(xs
->here
->xe_value_size
));
1452 size_t offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1453 void *val
= xs
->base
+ offs
;
1455 if (ocfs2_xattr_is_local(xs
->here
) && size
== size_l
) {
1456 /* Replace existing local xattr with tree root */
1457 ret
= ocfs2_xattr_set_value_outside(inode
, xi
, xs
,
1462 } else if (!ocfs2_xattr_is_local(xs
->here
)) {
1463 /* For existing xattr which has value outside */
1464 struct ocfs2_xattr_value_root
*xv
= NULL
;
1465 xv
= (struct ocfs2_xattr_value_root
*)(val
+
1466 OCFS2_XATTR_SIZE(name_len
));
1468 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1470 * If new value need set outside also,
1471 * first truncate old value to new value,
1472 * then set new value with set_value_outside().
1474 ret
= ocfs2_xattr_value_truncate(inode
,
1484 ret
= ocfs2_xattr_update_entry(inode
,
1494 ret
= __ocfs2_xattr_set_value_outside(inode
,
1504 * If new value need set in local,
1505 * just trucate old value to zero.
1507 ret
= ocfs2_xattr_value_truncate(inode
,
1518 ret
= ocfs2_journal_access(handle
, inode
, xs
->inode_bh
,
1519 OCFS2_JOURNAL_ACCESS_WRITE
);
1525 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1526 ret
= ocfs2_journal_access(handle
, inode
, xs
->xattr_bh
,
1527 OCFS2_JOURNAL_ACCESS_WRITE
);
1535 * Set value in local, include set tree root in local.
1536 * This is the first step for value size >INLINE_SIZE.
1538 ocfs2_xattr_set_entry_local(inode
, &xi_l
, xs
, last
, min_offs
);
1540 if (!(flag
& OCFS2_INLINE_XATTR_FL
)) {
1541 ret
= ocfs2_journal_dirty(handle
, xs
->xattr_bh
);
1548 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) &&
1549 (flag
& OCFS2_INLINE_XATTR_FL
)) {
1550 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1551 unsigned int xattrsize
= osb
->s_xattr_inline_size
;
1554 * Adjust extent record count or inline data size
1555 * to reserve space for extended attribute.
1557 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1558 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
1559 le16_add_cpu(&idata
->id_count
, -xattrsize
);
1560 } else if (!(ocfs2_inode_is_fast_symlink(inode
))) {
1561 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
1562 le16_add_cpu(&el
->l_count
, -(xattrsize
/
1563 sizeof(struct ocfs2_extent_rec
)));
1565 di
->i_xattr_inline_size
= cpu_to_le16(xattrsize
);
1567 /* Update xattr flag */
1568 spin_lock(&oi
->ip_lock
);
1569 oi
->ip_dyn_features
|= flag
;
1570 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
1571 spin_unlock(&oi
->ip_lock
);
1572 /* Update inode ctime */
1573 inode
->i_ctime
= CURRENT_TIME
;
1574 di
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1575 di
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1577 ret
= ocfs2_journal_dirty(handle
, xs
->inode_bh
);
1581 if (!ret
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
1583 * Set value outside in B tree.
1584 * This is the second step for value size > INLINE_SIZE.
1586 size_t offs
= le16_to_cpu(xs
->here
->xe_name_offset
);
1587 ret
= ocfs2_xattr_set_value_outside(inode
, xi
, xs
, ctxt
, offs
);
1593 * If set value outside failed, we have to clean
1594 * the junk tree root we have already set in local.
1596 ret2
= ocfs2_xattr_cleanup(inode
, ctxt
->handle
,
1606 static int ocfs2_remove_value_outside(struct inode
*inode
,
1607 struct buffer_head
*bh
,
1608 struct ocfs2_xattr_header
*header
)
1611 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1612 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
1614 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
1616 ctxt
.handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
1617 if (IS_ERR(ctxt
.handle
)) {
1618 ret
= PTR_ERR(ctxt
.handle
);
1623 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
1624 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
1626 if (!ocfs2_xattr_is_local(entry
)) {
1627 struct ocfs2_xattr_value_root
*xv
;
1630 val
= (void *)header
+
1631 le16_to_cpu(entry
->xe_name_offset
);
1632 xv
= (struct ocfs2_xattr_value_root
*)
1633 (val
+ OCFS2_XATTR_SIZE(entry
->xe_name_len
));
1634 ret
= ocfs2_xattr_value_truncate(inode
, bh
, xv
,
1643 ocfs2_commit_trans(osb
, ctxt
.handle
);
1644 ocfs2_schedule_truncate_log_flush(osb
, 1);
1645 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
1650 static int ocfs2_xattr_ibody_remove(struct inode
*inode
,
1651 struct buffer_head
*di_bh
)
1654 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1655 struct ocfs2_xattr_header
*header
;
1658 header
= (struct ocfs2_xattr_header
*)
1659 ((void *)di
+ inode
->i_sb
->s_blocksize
-
1660 le16_to_cpu(di
->i_xattr_inline_size
));
1662 ret
= ocfs2_remove_value_outside(inode
, di_bh
, header
);
1667 static int ocfs2_xattr_block_remove(struct inode
*inode
,
1668 struct buffer_head
*blk_bh
)
1670 struct ocfs2_xattr_block
*xb
;
1673 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1674 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
1675 struct ocfs2_xattr_header
*header
= &(xb
->xb_attrs
.xb_header
);
1676 ret
= ocfs2_remove_value_outside(inode
, blk_bh
, header
);
1678 ret
= ocfs2_delete_xattr_index_block(inode
, blk_bh
);
1683 static int ocfs2_xattr_free_block(struct inode
*inode
,
1686 struct inode
*xb_alloc_inode
;
1687 struct buffer_head
*xb_alloc_bh
= NULL
;
1688 struct buffer_head
*blk_bh
= NULL
;
1689 struct ocfs2_xattr_block
*xb
;
1690 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1696 ret
= ocfs2_read_block(inode
, block
, &blk_bh
);
1702 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1703 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
1708 ret
= ocfs2_xattr_block_remove(inode
, blk_bh
);
1714 blk
= le64_to_cpu(xb
->xb_blkno
);
1715 bit
= le16_to_cpu(xb
->xb_suballoc_bit
);
1716 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
1718 xb_alloc_inode
= ocfs2_get_system_file_inode(osb
,
1719 EXTENT_ALLOC_SYSTEM_INODE
,
1720 le16_to_cpu(xb
->xb_suballoc_slot
));
1721 if (!xb_alloc_inode
) {
1726 mutex_lock(&xb_alloc_inode
->i_mutex
);
1728 ret
= ocfs2_inode_lock(xb_alloc_inode
, &xb_alloc_bh
, 1);
1734 handle
= ocfs2_start_trans(osb
, OCFS2_SUBALLOC_FREE
);
1735 if (IS_ERR(handle
)) {
1736 ret
= PTR_ERR(handle
);
1741 ret
= ocfs2_free_suballoc_bits(handle
, xb_alloc_inode
, xb_alloc_bh
,
1746 ocfs2_commit_trans(osb
, handle
);
1748 ocfs2_inode_unlock(xb_alloc_inode
, 1);
1749 brelse(xb_alloc_bh
);
1751 mutex_unlock(&xb_alloc_inode
->i_mutex
);
1752 iput(xb_alloc_inode
);
1759 * ocfs2_xattr_remove()
1761 * Free extended attribute resources associated with this inode.
1763 int ocfs2_xattr_remove(struct inode
*inode
, struct buffer_head
*di_bh
)
1765 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1766 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1770 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1773 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1776 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
1777 ret
= ocfs2_xattr_ibody_remove(inode
, di_bh
);
1784 if (di
->i_xattr_loc
) {
1785 ret
= ocfs2_xattr_free_block(inode
,
1786 le64_to_cpu(di
->i_xattr_loc
));
1793 handle
= ocfs2_start_trans((OCFS2_SB(inode
->i_sb
)),
1794 OCFS2_INODE_UPDATE_CREDITS
);
1795 if (IS_ERR(handle
)) {
1796 ret
= PTR_ERR(handle
);
1800 ret
= ocfs2_journal_access(handle
, inode
, di_bh
,
1801 OCFS2_JOURNAL_ACCESS_WRITE
);
1807 di
->i_xattr_loc
= 0;
1809 spin_lock(&oi
->ip_lock
);
1810 oi
->ip_dyn_features
&= ~(OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
);
1811 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
1812 spin_unlock(&oi
->ip_lock
);
1814 ret
= ocfs2_journal_dirty(handle
, di_bh
);
1818 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
1823 static int ocfs2_xattr_has_space_inline(struct inode
*inode
,
1824 struct ocfs2_dinode
*di
)
1826 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1827 unsigned int xattrsize
= OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
;
1830 if (xattrsize
< OCFS2_MIN_XATTR_INLINE_SIZE
)
1833 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1834 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
1835 free
= le16_to_cpu(idata
->id_count
) - le64_to_cpu(di
->i_size
);
1836 } else if (ocfs2_inode_is_fast_symlink(inode
)) {
1837 free
= ocfs2_fast_symlink_chars(inode
->i_sb
) -
1838 le64_to_cpu(di
->i_size
);
1840 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
1841 free
= (le16_to_cpu(el
->l_count
) -
1842 le16_to_cpu(el
->l_next_free_rec
)) *
1843 sizeof(struct ocfs2_extent_rec
);
1845 if (free
>= xattrsize
)
1852 * ocfs2_xattr_ibody_find()
1854 * Find extended attribute in inode block and
1855 * fill search info into struct ocfs2_xattr_search.
1857 static int ocfs2_xattr_ibody_find(struct inode
*inode
,
1860 struct ocfs2_xattr_search
*xs
)
1862 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1863 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1867 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
1870 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
1871 down_read(&oi
->ip_alloc_sem
);
1872 has_space
= ocfs2_xattr_has_space_inline(inode
, di
);
1873 up_read(&oi
->ip_alloc_sem
);
1878 xs
->xattr_bh
= xs
->inode_bh
;
1879 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
1880 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)
1881 xs
->header
= (struct ocfs2_xattr_header
*)
1882 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
1884 xs
->header
= (struct ocfs2_xattr_header
*)
1885 (xs
->end
- OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
);
1886 xs
->base
= (void *)xs
->header
;
1887 xs
->here
= xs
->header
->xh_entries
;
1889 /* Find the named attribute. */
1890 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
1891 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1892 if (ret
&& ret
!= -ENODATA
)
1894 xs
->not_found
= ret
;
1901 * ocfs2_xattr_ibody_set()
1903 * Set, replace or remove an extended attribute into inode block.
1906 static int ocfs2_xattr_ibody_set(struct inode
*inode
,
1907 struct ocfs2_xattr_info
*xi
,
1908 struct ocfs2_xattr_search
*xs
,
1909 struct ocfs2_xattr_set_ctxt
*ctxt
)
1911 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1912 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1915 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
1918 down_write(&oi
->ip_alloc_sem
);
1919 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
1920 if (!ocfs2_xattr_has_space_inline(inode
, di
)) {
1926 ret
= ocfs2_xattr_set_entry(inode
, xi
, xs
, ctxt
,
1927 (OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
));
1929 up_write(&oi
->ip_alloc_sem
);
1935 * ocfs2_xattr_block_find()
1937 * Find extended attribute in external block and
1938 * fill search info into struct ocfs2_xattr_search.
1940 static int ocfs2_xattr_block_find(struct inode
*inode
,
1943 struct ocfs2_xattr_search
*xs
)
1945 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1946 struct buffer_head
*blk_bh
= NULL
;
1947 struct ocfs2_xattr_block
*xb
;
1950 if (!di
->i_xattr_loc
)
1953 ret
= ocfs2_read_block(inode
, le64_to_cpu(di
->i_xattr_loc
), &blk_bh
);
1959 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1960 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
1965 xs
->xattr_bh
= blk_bh
;
1967 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
1968 xs
->header
= &xb
->xb_attrs
.xb_header
;
1969 xs
->base
= (void *)xs
->header
;
1970 xs
->end
= (void *)(blk_bh
->b_data
) + blk_bh
->b_size
;
1971 xs
->here
= xs
->header
->xh_entries
;
1973 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1975 ret
= ocfs2_xattr_index_block_find(inode
, blk_bh
,
1979 if (ret
&& ret
!= -ENODATA
) {
1980 xs
->xattr_bh
= NULL
;
1983 xs
->not_found
= ret
;
1992 * ocfs2_xattr_block_set()
1994 * Set, replace or remove an extended attribute into external block.
1997 static int ocfs2_xattr_block_set(struct inode
*inode
,
1998 struct ocfs2_xattr_info
*xi
,
1999 struct ocfs2_xattr_search
*xs
,
2000 struct ocfs2_xattr_set_ctxt
*ctxt
)
2002 struct buffer_head
*new_bh
= NULL
;
2003 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2004 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2005 handle_t
*handle
= ctxt
->handle
;
2006 struct ocfs2_xattr_block
*xblk
= NULL
;
2007 u16 suballoc_bit_start
;
2012 if (!xs
->xattr_bh
) {
2013 ret
= ocfs2_journal_access(handle
, inode
, xs
->inode_bh
,
2014 OCFS2_JOURNAL_ACCESS_CREATE
);
2020 ret
= ocfs2_claim_metadata(osb
, handle
, ctxt
->meta_ac
, 1,
2021 &suballoc_bit_start
, &num_got
,
2028 new_bh
= sb_getblk(inode
->i_sb
, first_blkno
);
2029 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
2031 ret
= ocfs2_journal_access(handle
, inode
, new_bh
,
2032 OCFS2_JOURNAL_ACCESS_CREATE
);
2038 /* Initialize ocfs2_xattr_block */
2039 xs
->xattr_bh
= new_bh
;
2040 xblk
= (struct ocfs2_xattr_block
*)new_bh
->b_data
;
2041 memset(xblk
, 0, inode
->i_sb
->s_blocksize
);
2042 strcpy((void *)xblk
, OCFS2_XATTR_BLOCK_SIGNATURE
);
2043 xblk
->xb_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
2044 xblk
->xb_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
2045 xblk
->xb_fs_generation
= cpu_to_le32(osb
->fs_generation
);
2046 xblk
->xb_blkno
= cpu_to_le64(first_blkno
);
2048 xs
->header
= &xblk
->xb_attrs
.xb_header
;
2049 xs
->base
= (void *)xs
->header
;
2050 xs
->end
= (void *)xblk
+ inode
->i_sb
->s_blocksize
;
2051 xs
->here
= xs
->header
->xh_entries
;
2053 ret
= ocfs2_journal_dirty(handle
, new_bh
);
2058 di
->i_xattr_loc
= cpu_to_le64(first_blkno
);
2059 ocfs2_journal_dirty(handle
, xs
->inode_bh
);
2061 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2063 if (!(le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2064 /* Set extended attribute into external block */
2065 ret
= ocfs2_xattr_set_entry(inode
, xi
, xs
, ctxt
,
2066 OCFS2_HAS_XATTR_FL
);
2067 if (!ret
|| ret
!= -ENOSPC
)
2070 ret
= ocfs2_xattr_create_index_block(inode
, xs
, ctxt
);
2075 ret
= ocfs2_xattr_set_entry_index_block(inode
, xi
, xs
, ctxt
);
2082 /* Check whether the new xattr can be inserted into the inode. */
2083 static int ocfs2_xattr_can_be_in_inode(struct inode
*inode
,
2084 struct ocfs2_xattr_info
*xi
,
2085 struct ocfs2_xattr_search
*xs
)
2088 struct ocfs2_xattr_entry
*last
;
2090 size_t min_offs
= xs
->end
- xs
->base
;
2095 last
= xs
->header
->xh_entries
;
2097 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
2098 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
2099 if (offs
< min_offs
)
2104 free
= min_offs
- ((void *)last
- xs
->base
) - sizeof(__u32
);
2108 BUG_ON(!xs
->not_found
);
2110 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
2111 value_size
= OCFS2_XATTR_ROOT_SIZE
;
2113 value_size
= OCFS2_XATTR_SIZE(xi
->value_len
);
2115 if (free
>= sizeof(struct ocfs2_xattr_entry
) +
2116 OCFS2_XATTR_SIZE(strlen(xi
->name
)) + value_size
)
2122 static int ocfs2_calc_xattr_set_need(struct inode
*inode
,
2123 struct ocfs2_dinode
*di
,
2124 struct ocfs2_xattr_info
*xi
,
2125 struct ocfs2_xattr_search
*xis
,
2126 struct ocfs2_xattr_search
*xbs
,
2131 int ret
= 0, old_in_xb
= 0;
2132 int clusters_add
= 0, meta_add
= 0, credits
= 0;
2133 struct buffer_head
*bh
= NULL
;
2134 struct ocfs2_xattr_block
*xb
= NULL
;
2135 struct ocfs2_xattr_entry
*xe
= NULL
;
2136 struct ocfs2_xattr_value_root
*xv
= NULL
;
2138 int name_offset
, name_len
= 0;
2139 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
2143 if (xis
->not_found
&& xbs
->not_found
) {
2144 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2146 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2147 clusters_add
+= new_clusters
;
2148 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2156 if (!xis
->not_found
) {
2158 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
2159 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
2161 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
2164 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
2166 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
2167 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
2168 i
= xbs
->here
- xbs
->header
->xh_entries
;
2171 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
2172 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
2173 bucket_xh(xbs
->bucket
),
2176 base
= bucket_block(xbs
->bucket
, block_off
);
2177 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2180 credits
+= OCFS2_XATTR_BLOCK_UPDATE_CREDITS
;
2185 * delete a xattr doesn't need metadata and cluster allocation.
2186 * so just calculate the credits and return.
2188 * The credits for removing the value tree will be extended
2189 * by ocfs2_remove_extent itself.
2192 if (!ocfs2_xattr_is_local(xe
))
2193 credits
+= OCFS2_REMOVE_EXTENT_CREDITS
;
2198 /* do cluster allocation guess first. */
2199 value_size
= le64_to_cpu(xe
->xe_value_size
);
2203 * In xattr set, we always try to set the xe in inode first,
2204 * so if it can be inserted into inode successfully, the old
2205 * one will be removed from the xattr block, and this xattr
2206 * will be inserted into inode as a new xattr in inode.
2208 if (ocfs2_xattr_can_be_in_inode(inode
, xi
, xis
)) {
2209 clusters_add
+= new_clusters
;
2210 credits
+= OCFS2_REMOVE_EXTENT_CREDITS
+
2211 OCFS2_INODE_UPDATE_CREDITS
;
2212 if (!ocfs2_xattr_is_local(xe
))
2213 credits
+= ocfs2_calc_extend_credits(
2221 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2222 /* the new values will be stored outside. */
2223 u32 old_clusters
= 0;
2225 if (!ocfs2_xattr_is_local(xe
)) {
2226 old_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
2228 xv
= (struct ocfs2_xattr_value_root
*)
2229 (base
+ name_offset
+ name_len
);
2233 if (old_clusters
>= new_clusters
) {
2234 credits
+= OCFS2_REMOVE_EXTENT_CREDITS
;
2237 meta_add
+= ocfs2_extend_meta_needed(&xv
->xr_list
);
2238 clusters_add
+= new_clusters
- old_clusters
;
2239 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2247 * Now the new value will be stored inside. So if the new
2248 * value is smaller than the size of value root or the old
2249 * value, we don't need any allocation, otherwise we have
2250 * to guess metadata allocation.
2252 if ((ocfs2_xattr_is_local(xe
) && value_size
>= xi
->value_len
) ||
2253 (!ocfs2_xattr_is_local(xe
) &&
2254 OCFS2_XATTR_ROOT_SIZE
>= xi
->value_len
))
2259 /* calculate metadata allocation. */
2260 if (di
->i_xattr_loc
) {
2261 if (!xbs
->xattr_bh
) {
2262 ret
= ocfs2_read_block(inode
,
2263 le64_to_cpu(di
->i_xattr_loc
),
2270 xb
= (struct ocfs2_xattr_block
*)bh
->b_data
;
2272 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
2274 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
2275 struct ocfs2_extent_list
*el
=
2276 &xb
->xb_attrs
.xb_root
.xt_list
;
2277 meta_add
+= ocfs2_extend_meta_needed(el
);
2278 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
2283 * This cluster will be used either for new bucket or for
2285 * If the cluster size is the same as the bucket size, one
2286 * more is needed since we may need to extend the bucket
2290 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2291 if (OCFS2_XATTR_BUCKET_SIZE
==
2292 OCFS2_SB(inode
->i_sb
)->s_clustersize
) {
2293 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2298 credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
2302 *clusters_need
= clusters_add
;
2304 *meta_need
= meta_add
;
2306 *credits_need
= credits
;
2311 static int ocfs2_init_xattr_set_ctxt(struct inode
*inode
,
2312 struct ocfs2_dinode
*di
,
2313 struct ocfs2_xattr_info
*xi
,
2314 struct ocfs2_xattr_search
*xis
,
2315 struct ocfs2_xattr_search
*xbs
,
2316 struct ocfs2_xattr_set_ctxt
*ctxt
,
2319 int clusters_add
, meta_add
, ret
;
2320 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2322 memset(ctxt
, 0, sizeof(struct ocfs2_xattr_set_ctxt
));
2324 ocfs2_init_dealloc_ctxt(&ctxt
->dealloc
);
2326 ret
= ocfs2_calc_xattr_set_need(inode
, di
, xi
, xis
, xbs
,
2327 &clusters_add
, &meta_add
, credits
);
2333 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2334 "credits = %d\n", xi
->name
, meta_add
, clusters_add
, *credits
);
2337 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
,
2346 ret
= ocfs2_reserve_clusters(osb
, clusters_add
, &ctxt
->data_ac
);
2352 if (ctxt
->meta_ac
) {
2353 ocfs2_free_alloc_context(ctxt
->meta_ac
);
2354 ctxt
->meta_ac
= NULL
;
2358 * We cannot have an error and a non null ctxt->data_ac.
2365 static int __ocfs2_xattr_set_handle(struct inode
*inode
,
2366 struct ocfs2_dinode
*di
,
2367 struct ocfs2_xattr_info
*xi
,
2368 struct ocfs2_xattr_search
*xis
,
2369 struct ocfs2_xattr_search
*xbs
,
2370 struct ocfs2_xattr_set_ctxt
*ctxt
)
2372 int ret
= 0, credits
;
2375 /* Remove existing extended attribute */
2376 if (!xis
->not_found
)
2377 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
2378 else if (!xbs
->not_found
)
2379 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2381 /* We always try to set extended attribute into inode first*/
2382 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
2383 if (!ret
&& !xbs
->not_found
) {
2385 * If succeed and that extended attribute existing in
2386 * external block, then we will remove it.
2391 xis
->not_found
= -ENODATA
;
2392 ret
= ocfs2_calc_xattr_set_need(inode
,
2405 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2406 ctxt
->handle
->h_buffer_credits
);
2411 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2412 } else if (ret
== -ENOSPC
) {
2413 if (di
->i_xattr_loc
&& !xbs
->xattr_bh
) {
2414 ret
= ocfs2_xattr_block_find(inode
,
2420 xis
->not_found
= -ENODATA
;
2421 ret
= ocfs2_calc_xattr_set_need(inode
,
2434 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2435 ctxt
->handle
->h_buffer_credits
);
2442 * If no space in inode, we will set extended attribute
2443 * into external block.
2445 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
2448 if (!xis
->not_found
) {
2450 * If succeed and that extended attribute
2451 * existing in inode, we will remove it.
2455 xbs
->not_found
= -ENODATA
;
2456 ret
= ocfs2_calc_xattr_set_need(inode
,
2469 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
+
2470 ctxt
->handle
->h_buffer_credits
);
2475 ret
= ocfs2_xattr_ibody_set(inode
, xi
,
2486 * This function only called duing creating inode
2487 * for init security/acl xattrs of the new inode.
2488 * The xattrs could be put into ibody or extent block,
2489 * xattr bucket would not be use in this case.
2490 * transanction credits also be reserved in here.
2492 int ocfs2_xattr_set_handle(handle_t
*handle
,
2493 struct inode
*inode
,
2494 struct buffer_head
*di_bh
,
2500 struct ocfs2_alloc_context
*meta_ac
,
2501 struct ocfs2_alloc_context
*data_ac
)
2503 struct ocfs2_dinode
*di
;
2506 struct ocfs2_xattr_info xi
= {
2507 .name_index
= name_index
,
2510 .value_len
= value_len
,
2513 struct ocfs2_xattr_search xis
= {
2514 .not_found
= -ENODATA
,
2517 struct ocfs2_xattr_search xbs
= {
2518 .not_found
= -ENODATA
,
2521 struct ocfs2_xattr_set_ctxt ctxt
= {
2527 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2530 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
2531 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2533 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2535 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
2538 if (xis
.not_found
) {
2539 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
2544 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
2547 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2548 brelse(xbs
.xattr_bh
);
2556 * Set, replace or remove an extended attribute for this inode.
2557 * value is NULL to remove an existing extended attribute, else either
2558 * create or replace an extended attribute.
2560 int ocfs2_xattr_set(struct inode
*inode
,
2567 struct buffer_head
*di_bh
= NULL
;
2568 struct ocfs2_dinode
*di
;
2570 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2571 struct inode
*tl_inode
= osb
->osb_tl_inode
;
2572 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
2574 struct ocfs2_xattr_info xi
= {
2575 .name_index
= name_index
,
2578 .value_len
= value_len
,
2581 struct ocfs2_xattr_search xis
= {
2582 .not_found
= -ENODATA
,
2585 struct ocfs2_xattr_search xbs
= {
2586 .not_found
= -ENODATA
,
2589 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2593 * Only xbs will be used on indexed trees. xis doesn't need a
2596 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
2598 mlog_errno(-ENOMEM
);
2602 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2605 goto cleanup_nolock
;
2607 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
2608 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2610 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2612 * Scan inode and external block to find the same name
2613 * extended attribute and collect search infomation.
2615 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
2618 if (xis
.not_found
) {
2619 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
2624 if (xis
.not_found
&& xbs
.not_found
) {
2626 if (flags
& XATTR_REPLACE
)
2633 if (flags
& XATTR_CREATE
)
2638 mutex_lock(&tl_inode
->i_mutex
);
2640 if (ocfs2_truncate_log_needs_flush(osb
)) {
2641 ret
= __ocfs2_flush_truncate_log(osb
);
2643 mutex_unlock(&tl_inode
->i_mutex
);
2648 mutex_unlock(&tl_inode
->i_mutex
);
2650 ret
= ocfs2_init_xattr_set_ctxt(inode
, di
, &xi
, &xis
,
2651 &xbs
, &ctxt
, &credits
);
2657 ctxt
.handle
= ocfs2_start_trans(osb
, credits
);
2658 if (IS_ERR(ctxt
.handle
)) {
2659 ret
= PTR_ERR(ctxt
.handle
);
2664 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
2666 ocfs2_commit_trans(osb
, ctxt
.handle
);
2669 ocfs2_free_alloc_context(ctxt
.data_ac
);
2671 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2672 if (ocfs2_dealloc_has_cluster(&ctxt
.dealloc
))
2673 ocfs2_schedule_truncate_log_flush(osb
, 1);
2674 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
2676 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
2677 ocfs2_inode_unlock(inode
, 1);
2680 brelse(xbs
.xattr_bh
);
2681 ocfs2_xattr_bucket_free(xbs
.bucket
);
2687 * Find the xattr extent rec which may contains name_hash.
2688 * e_cpos will be the first name hash of the xattr rec.
2689 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2691 static int ocfs2_xattr_get_rec(struct inode
*inode
,
2696 struct ocfs2_extent_list
*el
)
2699 struct buffer_head
*eb_bh
= NULL
;
2700 struct ocfs2_extent_block
*eb
;
2701 struct ocfs2_extent_rec
*rec
= NULL
;
2704 if (el
->l_tree_depth
) {
2705 ret
= ocfs2_find_leaf(inode
, el
, name_hash
, &eb_bh
);
2711 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
2714 if (el
->l_tree_depth
) {
2715 ocfs2_error(inode
->i_sb
,
2716 "Inode %lu has non zero tree depth in "
2717 "xattr tree block %llu\n", inode
->i_ino
,
2718 (unsigned long long)eb_bh
->b_blocknr
);
2724 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
2725 rec
= &el
->l_recs
[i
];
2727 if (le32_to_cpu(rec
->e_cpos
) <= name_hash
) {
2728 e_blkno
= le64_to_cpu(rec
->e_blkno
);
2734 ocfs2_error(inode
->i_sb
, "Inode %lu has bad extent "
2735 "record (%u, %u, 0) in xattr", inode
->i_ino
,
2736 le32_to_cpu(rec
->e_cpos
),
2737 ocfs2_rec_clusters(el
, rec
));
2742 *p_blkno
= le64_to_cpu(rec
->e_blkno
);
2743 *num_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
2745 *e_cpos
= le32_to_cpu(rec
->e_cpos
);
2751 typedef int (xattr_bucket_func
)(struct inode
*inode
,
2752 struct ocfs2_xattr_bucket
*bucket
,
2755 static int ocfs2_find_xe_in_bucket(struct inode
*inode
,
2756 struct ocfs2_xattr_bucket
*bucket
,
2763 int i
, ret
= 0, cmp
= 1, block_off
, new_offset
;
2764 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
2765 size_t name_len
= strlen(name
);
2766 struct ocfs2_xattr_entry
*xe
= NULL
;
2770 * We don't use binary search in the bucket because there
2771 * may be multiple entries with the same name hash.
2773 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
2774 xe
= &xh
->xh_entries
[i
];
2776 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
))
2778 else if (name_hash
< le32_to_cpu(xe
->xe_name_hash
))
2781 cmp
= name_index
- ocfs2_xattr_get_type(xe
);
2783 cmp
= name_len
- xe
->xe_name_len
;
2787 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
2797 xe_name
= bucket_block(bucket
, block_off
) + new_offset
;
2798 if (!memcmp(name
, xe_name
, name_len
)) {
2810 * Find the specified xattr entry in a series of buckets.
2811 * This series start from p_blkno and last for num_clusters.
2812 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2813 * the num of the valid buckets.
2815 * Return the buffer_head this xattr should reside in. And if the xattr's
2816 * hash is in the gap of 2 buckets, return the lower bucket.
2818 static int ocfs2_xattr_bucket_find(struct inode
*inode
,
2825 struct ocfs2_xattr_search
*xs
)
2828 struct ocfs2_xattr_header
*xh
= NULL
;
2829 struct ocfs2_xattr_entry
*xe
= NULL
;
2831 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
2832 int low_bucket
= 0, bucket
, high_bucket
;
2833 struct ocfs2_xattr_bucket
*search
;
2835 u64 blkno
, lower_blkno
= 0;
2837 search
= ocfs2_xattr_bucket_new(inode
);
2844 ret
= ocfs2_read_xattr_bucket(search
, p_blkno
);
2850 xh
= bucket_xh(search
);
2851 high_bucket
= le16_to_cpu(xh
->xh_num_buckets
) - 1;
2852 while (low_bucket
<= high_bucket
) {
2853 ocfs2_xattr_bucket_relse(search
);
2855 bucket
= (low_bucket
+ high_bucket
) / 2;
2856 blkno
= p_blkno
+ bucket
* blk_per_bucket
;
2857 ret
= ocfs2_read_xattr_bucket(search
, blkno
);
2863 xh
= bucket_xh(search
);
2864 xe
= &xh
->xh_entries
[0];
2865 if (name_hash
< le32_to_cpu(xe
->xe_name_hash
)) {
2866 high_bucket
= bucket
- 1;
2871 * Check whether the hash of the last entry in our
2872 * bucket is larger than the search one. for an empty
2873 * bucket, the last one is also the first one.
2876 xe
= &xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1];
2878 last_hash
= le32_to_cpu(xe
->xe_name_hash
);
2880 /* record lower_blkno which may be the insert place. */
2881 lower_blkno
= blkno
;
2883 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
)) {
2884 low_bucket
= bucket
+ 1;
2888 /* the searched xattr should reside in this bucket if exists. */
2889 ret
= ocfs2_find_xe_in_bucket(inode
, search
,
2890 name_index
, name
, name_hash
,
2900 * Record the bucket we have found.
2901 * When the xattr's hash value is in the gap of 2 buckets, we will
2902 * always set it to the previous bucket.
2905 lower_blkno
= p_blkno
;
2907 /* This should be in cache - we just read it during the search */
2908 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, lower_blkno
);
2914 xs
->header
= bucket_xh(xs
->bucket
);
2915 xs
->base
= bucket_block(xs
->bucket
, 0);
2916 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
2919 xs
->here
= &xs
->header
->xh_entries
[index
];
2920 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name
,
2921 (unsigned long long)bucket_blkno(xs
->bucket
), index
);
2926 ocfs2_xattr_bucket_free(search
);
2930 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
2931 struct buffer_head
*root_bh
,
2934 struct ocfs2_xattr_search
*xs
)
2937 struct ocfs2_xattr_block
*xb
=
2938 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
2939 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
2940 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
2942 u32 first_hash
, num_clusters
= 0;
2943 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
2945 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
2948 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2949 name
, name_hash
, name_index
);
2951 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &first_hash
,
2958 BUG_ON(p_blkno
== 0 || num_clusters
== 0 || first_hash
> name_hash
);
2960 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2961 "in the rec is %u\n", num_clusters
, (unsigned long long)p_blkno
,
2964 ret
= ocfs2_xattr_bucket_find(inode
, name_index
, name
, name_hash
,
2965 p_blkno
, first_hash
, num_clusters
, xs
);
2971 static int ocfs2_iterate_xattr_buckets(struct inode
*inode
,
2974 xattr_bucket_func
*func
,
2978 u32 bpc
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
));
2979 u32 num_buckets
= clusters
* bpc
;
2980 struct ocfs2_xattr_bucket
*bucket
;
2982 bucket
= ocfs2_xattr_bucket_new(inode
);
2984 mlog_errno(-ENOMEM
);
2988 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2989 clusters
, (unsigned long long)blkno
);
2991 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bucket
->bu_blocks
) {
2992 ret
= ocfs2_read_xattr_bucket(bucket
, blkno
);
2999 * The real bucket num in this series of blocks is stored
3000 * in the 1st bucket.
3003 num_buckets
= le16_to_cpu(bucket_xh(bucket
)->xh_num_buckets
);
3005 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3006 (unsigned long long)blkno
,
3007 le32_to_cpu(bucket_xh(bucket
)->xh_entries
[0].xe_name_hash
));
3009 ret
= func(inode
, bucket
, para
);
3012 /* Fall through to bucket_relse() */
3015 ocfs2_xattr_bucket_relse(bucket
);
3020 ocfs2_xattr_bucket_free(bucket
);
3024 struct ocfs2_xattr_tree_list
{
3030 static int ocfs2_xattr_bucket_get_name_value(struct inode
*inode
,
3031 struct ocfs2_xattr_header
*xh
,
3038 if (index
< 0 || index
>= le16_to_cpu(xh
->xh_count
))
3041 name_offset
= le16_to_cpu(xh
->xh_entries
[index
].xe_name_offset
);
3043 *block_off
= name_offset
>> inode
->i_sb
->s_blocksize_bits
;
3044 *new_offset
= name_offset
% inode
->i_sb
->s_blocksize
;
3049 static int ocfs2_list_xattr_bucket(struct inode
*inode
,
3050 struct ocfs2_xattr_bucket
*bucket
,
3054 struct ocfs2_xattr_tree_list
*xl
= (struct ocfs2_xattr_tree_list
*)para
;
3055 int i
, block_off
, new_offset
;
3056 const char *prefix
, *name
;
3058 for (i
= 0 ; i
< le16_to_cpu(bucket_xh(bucket
)->xh_count
); i
++) {
3059 struct ocfs2_xattr_entry
*entry
= &bucket_xh(bucket
)->xh_entries
[i
];
3060 type
= ocfs2_xattr_get_type(entry
);
3061 prefix
= ocfs2_xattr_prefix(type
);
3064 ret
= ocfs2_xattr_bucket_get_name_value(inode
,
3072 name
= (const char *)bucket_block(bucket
, block_off
) +
3074 ret
= ocfs2_xattr_list_entry(xl
->buffer
,
3078 entry
->xe_name_len
);
3087 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
3088 struct ocfs2_xattr_tree_root
*xt
,
3092 struct ocfs2_extent_list
*el
= &xt
->xt_list
;
3094 u32 name_hash
= UINT_MAX
, e_cpos
= 0, num_clusters
= 0;
3096 struct ocfs2_xattr_tree_list xl
= {
3098 .buffer_size
= buffer_size
,
3102 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
3105 while (name_hash
> 0) {
3106 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
3107 &e_cpos
, &num_clusters
, el
);
3113 ret
= ocfs2_iterate_xattr_buckets(inode
, p_blkno
, num_clusters
,
3114 ocfs2_list_xattr_bucket
,
3124 name_hash
= e_cpos
- 1;
3132 static int cmp_xe(const void *a
, const void *b
)
3134 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
3135 u32 l_hash
= le32_to_cpu(l
->xe_name_hash
);
3136 u32 r_hash
= le32_to_cpu(r
->xe_name_hash
);
3138 if (l_hash
> r_hash
)
3140 if (l_hash
< r_hash
)
3145 static void swap_xe(void *a
, void *b
, int size
)
3147 struct ocfs2_xattr_entry
*l
= a
, *r
= b
, tmp
;
3150 memcpy(l
, r
, sizeof(struct ocfs2_xattr_entry
));
3151 memcpy(r
, &tmp
, sizeof(struct ocfs2_xattr_entry
));
3155 * When the ocfs2_xattr_block is filled up, new bucket will be created
3156 * and all the xattr entries will be moved to the new bucket.
3157 * The header goes at the start of the bucket, and the names+values are
3158 * filled from the end. This is why *target starts as the last buffer.
3159 * Note: we need to sort the entries since they are not saved in order
3160 * in the ocfs2_xattr_block.
3162 static void ocfs2_cp_xattr_block_to_bucket(struct inode
*inode
,
3163 struct buffer_head
*xb_bh
,
3164 struct ocfs2_xattr_bucket
*bucket
)
3166 int i
, blocksize
= inode
->i_sb
->s_blocksize
;
3167 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3168 u16 offset
, size
, off_change
;
3169 struct ocfs2_xattr_entry
*xe
;
3170 struct ocfs2_xattr_block
*xb
=
3171 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
3172 struct ocfs2_xattr_header
*xb_xh
= &xb
->xb_attrs
.xb_header
;
3173 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
3174 u16 count
= le16_to_cpu(xb_xh
->xh_count
);
3175 char *src
= xb_bh
->b_data
;
3176 char *target
= bucket_block(bucket
, blks
- 1);
3178 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3179 (unsigned long long)xb_bh
->b_blocknr
,
3180 (unsigned long long)bucket_blkno(bucket
));
3182 for (i
= 0; i
< blks
; i
++)
3183 memset(bucket_block(bucket
, i
), 0, blocksize
);
3186 * Since the xe_name_offset is based on ocfs2_xattr_header,
3187 * there is a offset change corresponding to the change of
3188 * ocfs2_xattr_header's position.
3190 off_change
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
3191 xe
= &xb_xh
->xh_entries
[count
- 1];
3192 offset
= le16_to_cpu(xe
->xe_name_offset
) + off_change
;
3193 size
= blocksize
- offset
;
3195 /* copy all the names and values. */
3196 memcpy(target
+ offset
, src
+ offset
, size
);
3198 /* Init new header now. */
3199 xh
->xh_count
= xb_xh
->xh_count
;
3200 xh
->xh_num_buckets
= cpu_to_le16(1);
3201 xh
->xh_name_value_len
= cpu_to_le16(size
);
3202 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
- size
);
3204 /* copy all the entries. */
3205 target
= bucket_block(bucket
, 0);
3206 offset
= offsetof(struct ocfs2_xattr_header
, xh_entries
);
3207 size
= count
* sizeof(struct ocfs2_xattr_entry
);
3208 memcpy(target
+ offset
, (char *)xb_xh
+ offset
, size
);
3210 /* Change the xe offset for all the xe because of the move. */
3211 off_change
= OCFS2_XATTR_BUCKET_SIZE
- blocksize
+
3212 offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
3213 for (i
= 0; i
< count
; i
++)
3214 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
, off_change
);
3216 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3217 offset
, size
, off_change
);
3219 sort(target
+ offset
, count
, sizeof(struct ocfs2_xattr_entry
),
3224 * After we move xattr from block to index btree, we have to
3225 * update ocfs2_xattr_search to the new xe and base.
3227 * When the entry is in xattr block, xattr_bh indicates the storage place.
3228 * While if the entry is in index b-tree, "bucket" indicates the
3229 * real place of the xattr.
3231 static void ocfs2_xattr_update_xattr_search(struct inode
*inode
,
3232 struct ocfs2_xattr_search
*xs
,
3233 struct buffer_head
*old_bh
)
3235 char *buf
= old_bh
->b_data
;
3236 struct ocfs2_xattr_block
*old_xb
= (struct ocfs2_xattr_block
*)buf
;
3237 struct ocfs2_xattr_header
*old_xh
= &old_xb
->xb_attrs
.xb_header
;
3240 xs
->header
= bucket_xh(xs
->bucket
);
3241 xs
->base
= bucket_block(xs
->bucket
, 0);
3242 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
3247 i
= xs
->here
- old_xh
->xh_entries
;
3248 xs
->here
= &xs
->header
->xh_entries
[i
];
3251 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
3252 struct ocfs2_xattr_search
*xs
,
3253 struct ocfs2_xattr_set_ctxt
*ctxt
)
3258 handle_t
*handle
= ctxt
->handle
;
3259 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3260 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
3261 struct buffer_head
*xb_bh
= xs
->xattr_bh
;
3262 struct ocfs2_xattr_block
*xb
=
3263 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
3264 struct ocfs2_xattr_tree_root
*xr
;
3265 u16 xb_flags
= le16_to_cpu(xb
->xb_flags
);
3267 mlog(0, "create xattr index block for %llu\n",
3268 (unsigned long long)xb_bh
->b_blocknr
);
3270 BUG_ON(xb_flags
& OCFS2_XATTR_INDEXED
);
3271 BUG_ON(!xs
->bucket
);
3275 * We can use this lock for now, and maybe move to a dedicated mutex
3276 * if performance becomes a problem later.
3278 down_write(&oi
->ip_alloc_sem
);
3280 ret
= ocfs2_journal_access(handle
, inode
, xb_bh
,
3281 OCFS2_JOURNAL_ACCESS_WRITE
);
3287 ret
= __ocfs2_claim_clusters(osb
, handle
, ctxt
->data_ac
,
3288 1, 1, &bit_off
, &len
);
3295 * The bucket may spread in many blocks, and
3296 * we will only touch the 1st block and the last block
3297 * in the whole bucket(one for entry and one for data).
3299 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, bit_off
);
3301 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3302 (unsigned long long)blkno
);
3304 ret
= ocfs2_init_xattr_bucket(xs
->bucket
, blkno
);
3310 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
3311 OCFS2_JOURNAL_ACCESS_CREATE
);
3317 ocfs2_cp_xattr_block_to_bucket(inode
, xb_bh
, xs
->bucket
);
3318 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
3320 ocfs2_xattr_update_xattr_search(inode
, xs
, xb_bh
);
3322 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3323 memset(&xb
->xb_attrs
, 0, inode
->i_sb
->s_blocksize
-
3324 offsetof(struct ocfs2_xattr_block
, xb_attrs
));
3326 xr
= &xb
->xb_attrs
.xb_root
;
3327 xr
->xt_clusters
= cpu_to_le32(1);
3328 xr
->xt_last_eb_blk
= 0;
3329 xr
->xt_list
.l_tree_depth
= 0;
3330 xr
->xt_list
.l_count
= cpu_to_le16(ocfs2_xattr_recs_per_xb(inode
->i_sb
));
3331 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
3333 xr
->xt_list
.l_recs
[0].e_cpos
= 0;
3334 xr
->xt_list
.l_recs
[0].e_blkno
= cpu_to_le64(blkno
);
3335 xr
->xt_list
.l_recs
[0].e_leaf_clusters
= cpu_to_le16(1);
3337 xb
->xb_flags
= cpu_to_le16(xb_flags
| OCFS2_XATTR_INDEXED
);
3339 ocfs2_journal_dirty(handle
, xb_bh
);
3342 up_write(&oi
->ip_alloc_sem
);
3347 static int cmp_xe_offset(const void *a
, const void *b
)
3349 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
3350 u32 l_name_offset
= le16_to_cpu(l
->xe_name_offset
);
3351 u32 r_name_offset
= le16_to_cpu(r
->xe_name_offset
);
3353 if (l_name_offset
< r_name_offset
)
3355 if (l_name_offset
> r_name_offset
)
3361 * defrag a xattr bucket if we find that the bucket has some
3362 * holes beteen name/value pairs.
3363 * We will move all the name/value pairs to the end of the bucket
3364 * so that we can spare some space for insertion.
3366 static int ocfs2_defrag_xattr_bucket(struct inode
*inode
,
3368 struct ocfs2_xattr_bucket
*bucket
)
3371 size_t end
, offset
, len
, value_len
;
3372 struct ocfs2_xattr_header
*xh
;
3373 char *entries
, *buf
, *bucket_buf
= NULL
;
3374 u64 blkno
= bucket_blkno(bucket
);
3376 size_t blocksize
= inode
->i_sb
->s_blocksize
;
3377 struct ocfs2_xattr_entry
*xe
;
3380 * In order to make the operation more efficient and generic,
3381 * we copy all the blocks into a contiguous memory and do the
3382 * defragment there, so if anything is error, we will not touch
3385 bucket_buf
= kmalloc(OCFS2_XATTR_BUCKET_SIZE
, GFP_NOFS
);
3392 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
3393 memcpy(buf
, bucket_block(bucket
, i
), blocksize
);
3395 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
3396 OCFS2_JOURNAL_ACCESS_WRITE
);
3402 xh
= (struct ocfs2_xattr_header
*)bucket_buf
;
3403 entries
= (char *)xh
->xh_entries
;
3404 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
3406 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3407 "xh_free_start = %u, xh_name_value_len = %u.\n",
3408 (unsigned long long)blkno
, le16_to_cpu(xh
->xh_count
),
3409 xh_free_start
, le16_to_cpu(xh
->xh_name_value_len
));
3412 * sort all the entries by their offset.
3413 * the largest will be the first, so that we can
3414 * move them to the end one by one.
3416 sort(entries
, le16_to_cpu(xh
->xh_count
),
3417 sizeof(struct ocfs2_xattr_entry
),
3418 cmp_xe_offset
, swap_xe
);
3420 /* Move all name/values to the end of the bucket. */
3421 xe
= xh
->xh_entries
;
3422 end
= OCFS2_XATTR_BUCKET_SIZE
;
3423 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, xe
++) {
3424 offset
= le16_to_cpu(xe
->xe_name_offset
);
3425 if (ocfs2_xattr_is_local(xe
))
3426 value_len
= OCFS2_XATTR_SIZE(
3427 le64_to_cpu(xe
->xe_value_size
));
3429 value_len
= OCFS2_XATTR_ROOT_SIZE
;
3430 len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
) + value_len
;
3433 * We must make sure that the name/value pair
3434 * exist in the same block. So adjust end to
3435 * the previous block end if needed.
3437 if (((end
- len
) / blocksize
!=
3438 (end
- 1) / blocksize
))
3439 end
= end
- end
% blocksize
;
3441 if (end
> offset
+ len
) {
3442 memmove(bucket_buf
+ end
- len
,
3443 bucket_buf
+ offset
, len
);
3444 xe
->xe_name_offset
= cpu_to_le16(end
- len
);
3447 mlog_bug_on_msg(end
< offset
+ len
, "Defrag check failed for "
3448 "bucket %llu\n", (unsigned long long)blkno
);
3453 mlog_bug_on_msg(xh_free_start
> end
, "Defrag check failed for "
3454 "bucket %llu\n", (unsigned long long)blkno
);
3456 if (xh_free_start
== end
)
3459 memset(bucket_buf
+ xh_free_start
, 0, end
- xh_free_start
);
3460 xh
->xh_free_start
= cpu_to_le16(end
);
3462 /* sort the entries by their name_hash. */
3463 sort(entries
, le16_to_cpu(xh
->xh_count
),
3464 sizeof(struct ocfs2_xattr_entry
),
3468 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
3469 memcpy(bucket_block(bucket
, i
), buf
, blocksize
);
3470 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
3478 * Move half nums of the xattr bucket in the previous cluster to this new
3479 * cluster. We only touch the last cluster of the previous extend record.
3481 * first_bh is the first buffer_head of a series of bucket in the same
3482 * extent rec and header_bh is the header of one bucket in this cluster.
3483 * They will be updated if we move the data header_bh contains to the new
3484 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3486 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode
*inode
,
3488 struct buffer_head
**first_bh
,
3489 struct buffer_head
**header_bh
,
3495 int i
, ret
, credits
;
3496 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3497 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
3498 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
3499 int blocksize
= inode
->i_sb
->s_blocksize
;
3500 struct buffer_head
*old_bh
, *new_bh
, *prev_bh
, *new_first_bh
= NULL
;
3501 struct ocfs2_xattr_header
*new_xh
;
3502 struct ocfs2_xattr_header
*xh
=
3503 (struct ocfs2_xattr_header
*)((*first_bh
)->b_data
);
3505 BUG_ON(le16_to_cpu(xh
->xh_num_buckets
) < num_buckets
);
3506 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
== osb
->s_clustersize
);
3508 prev_bh
= *first_bh
;
3510 xh
= (struct ocfs2_xattr_header
*)prev_bh
->b_data
;
3512 prev_blkno
+= (num_clusters
- 1) * bpc
+ bpc
/ 2;
3514 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3515 (unsigned long long)prev_blkno
, (unsigned long long)new_blkno
);
3518 * We need to update the 1st half of the new cluster and
3519 * 1 more for the update of the 1st bucket of the previous
3522 credits
= bpc
/ 2 + 1 + handle
->h_buffer_credits
;
3523 ret
= ocfs2_extend_trans(handle
, credits
);
3529 ret
= ocfs2_journal_access(handle
, inode
, prev_bh
,
3530 OCFS2_JOURNAL_ACCESS_WRITE
);
3536 for (i
= 0; i
< bpc
/ 2; i
++, prev_blkno
++, new_blkno
++) {
3537 old_bh
= new_bh
= NULL
;
3538 new_bh
= sb_getblk(inode
->i_sb
, new_blkno
);
3545 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
3547 ret
= ocfs2_journal_access(handle
, inode
, new_bh
,
3548 OCFS2_JOURNAL_ACCESS_CREATE
);
3555 ret
= ocfs2_read_block(inode
, prev_blkno
, &old_bh
);
3562 memcpy(new_bh
->b_data
, old_bh
->b_data
, blocksize
);
3565 new_xh
= (struct ocfs2_xattr_header
*)new_bh
->b_data
;
3566 new_xh
->xh_num_buckets
= cpu_to_le16(num_buckets
/ 2);
3569 *first_hash
= le32_to_cpu(
3570 new_xh
->xh_entries
[0].xe_name_hash
);
3571 new_first_bh
= new_bh
;
3572 get_bh(new_first_bh
);
3575 ocfs2_journal_dirty(handle
, new_bh
);
3577 if (*header_bh
== old_bh
) {
3579 *header_bh
= new_bh
;
3583 *first_bh
= new_first_bh
;
3590 le16_add_cpu(&xh
->xh_num_buckets
, -(num_buckets
/ 2));
3592 ocfs2_journal_dirty(handle
, prev_bh
);
3595 brelse(new_first_bh
);
3600 * Find the suitable pos when we divide a bucket into 2.
3601 * We have to make sure the xattrs with the same hash value exist
3602 * in the same bucket.
3604 * If this ocfs2_xattr_header covers more than one hash value, find a
3605 * place where the hash value changes. Try to find the most even split.
3606 * The most common case is that all entries have different hash values,
3607 * and the first check we make will find a place to split.
3609 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header
*xh
)
3611 struct ocfs2_xattr_entry
*entries
= xh
->xh_entries
;
3612 int count
= le16_to_cpu(xh
->xh_count
);
3613 int delta
, middle
= count
/ 2;
3616 * We start at the middle. Each step gets farther away in both
3617 * directions. We therefore hit the change in hash value
3618 * nearest to the middle. Note that this loop does not execute for
3621 for (delta
= 0; delta
< middle
; delta
++) {
3622 /* Let's check delta earlier than middle */
3623 if (cmp_xe(&entries
[middle
- delta
- 1],
3624 &entries
[middle
- delta
]))
3625 return middle
- delta
;
3627 /* For even counts, don't walk off the end */
3628 if ((middle
+ delta
+ 1) == count
)
3631 /* Now try delta past middle */
3632 if (cmp_xe(&entries
[middle
+ delta
],
3633 &entries
[middle
+ delta
+ 1]))
3634 return middle
+ delta
+ 1;
3637 /* Every entry had the same hash */
3642 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3643 * first_hash will record the 1st hash of the new bucket.
3645 * Normally half of the xattrs will be moved. But we have to make
3646 * sure that the xattrs with the same hash value are stored in the
3647 * same bucket. If all the xattrs in this bucket have the same hash
3648 * value, the new bucket will be initialized as an empty one and the
3649 * first_hash will be initialized as (hash_value+1).
3651 static int ocfs2_divide_xattr_bucket(struct inode
*inode
,
3656 int new_bucket_head
)
3659 int count
, start
, len
, name_value_len
= 0, xe_len
, name_offset
= 0;
3660 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
3661 struct ocfs2_xattr_header
*xh
;
3662 struct ocfs2_xattr_entry
*xe
;
3663 int blocksize
= inode
->i_sb
->s_blocksize
;
3665 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3666 (unsigned long long)blk
, (unsigned long long)new_blk
);
3668 s_bucket
= ocfs2_xattr_bucket_new(inode
);
3669 t_bucket
= ocfs2_xattr_bucket_new(inode
);
3670 if (!s_bucket
|| !t_bucket
) {
3676 ret
= ocfs2_read_xattr_bucket(s_bucket
, blk
);
3682 ret
= ocfs2_xattr_bucket_journal_access(handle
, s_bucket
,
3683 OCFS2_JOURNAL_ACCESS_WRITE
);
3690 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3691 * there's no need to read it.
3693 ret
= ocfs2_init_xattr_bucket(t_bucket
, new_blk
);
3699 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
3701 OCFS2_JOURNAL_ACCESS_CREATE
:
3702 OCFS2_JOURNAL_ACCESS_WRITE
);
3708 xh
= bucket_xh(s_bucket
);
3709 count
= le16_to_cpu(xh
->xh_count
);
3710 start
= ocfs2_xattr_find_divide_pos(xh
);
3712 if (start
== count
) {
3713 xe
= &xh
->xh_entries
[start
-1];
3716 * initialized a new empty bucket here.
3717 * The hash value is set as one larger than
3718 * that of the last entry in the previous bucket.
3720 for (i
= 0; i
< t_bucket
->bu_blocks
; i
++)
3721 memset(bucket_block(t_bucket
, i
), 0, blocksize
);
3723 xh
= bucket_xh(t_bucket
);
3724 xh
->xh_free_start
= cpu_to_le16(blocksize
);
3725 xh
->xh_entries
[0].xe_name_hash
= xe
->xe_name_hash
;
3726 le32_add_cpu(&xh
->xh_entries
[0].xe_name_hash
, 1);
3728 goto set_num_buckets
;
3731 /* copy the whole bucket to the new first. */
3732 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
3734 /* update the new bucket. */
3735 xh
= bucket_xh(t_bucket
);
3738 * Calculate the total name/value len and xh_free_start for
3739 * the old bucket first.
3741 name_offset
= OCFS2_XATTR_BUCKET_SIZE
;
3743 for (i
= 0; i
< start
; i
++) {
3744 xe
= &xh
->xh_entries
[i
];
3745 xe_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3746 if (ocfs2_xattr_is_local(xe
))
3748 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
3750 xe_len
+= OCFS2_XATTR_ROOT_SIZE
;
3751 name_value_len
+= xe_len
;
3752 if (le16_to_cpu(xe
->xe_name_offset
) < name_offset
)
3753 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3757 * Now begin the modification to the new bucket.
3759 * In the new bucket, We just move the xattr entry to the beginning
3760 * and don't touch the name/value. So there will be some holes in the
3761 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3764 xe
= &xh
->xh_entries
[start
];
3765 len
= sizeof(struct ocfs2_xattr_entry
) * (count
- start
);
3766 mlog(0, "mv xattr entry len %d from %d to %d\n", len
,
3767 (int)((char *)xe
- (char *)xh
),
3768 (int)((char *)xh
->xh_entries
- (char *)xh
));
3769 memmove((char *)xh
->xh_entries
, (char *)xe
, len
);
3770 xe
= &xh
->xh_entries
[count
- start
];
3771 len
= sizeof(struct ocfs2_xattr_entry
) * start
;
3772 memset((char *)xe
, 0, len
);
3774 le16_add_cpu(&xh
->xh_count
, -start
);
3775 le16_add_cpu(&xh
->xh_name_value_len
, -name_value_len
);
3777 /* Calculate xh_free_start for the new bucket. */
3778 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
3779 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
3780 xe
= &xh
->xh_entries
[i
];
3781 xe_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3782 if (ocfs2_xattr_is_local(xe
))
3784 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
3786 xe_len
+= OCFS2_XATTR_ROOT_SIZE
;
3787 if (le16_to_cpu(xe
->xe_name_offset
) <
3788 le16_to_cpu(xh
->xh_free_start
))
3789 xh
->xh_free_start
= xe
->xe_name_offset
;
3793 /* set xh->xh_num_buckets for the new xh. */
3794 if (new_bucket_head
)
3795 xh
->xh_num_buckets
= cpu_to_le16(1);
3797 xh
->xh_num_buckets
= 0;
3799 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
3801 /* store the first_hash of the new bucket. */
3803 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
3806 * Now only update the 1st block of the old bucket. If we
3807 * just added a new empty bucket, there is no need to modify
3813 xh
= bucket_xh(s_bucket
);
3814 memset(&xh
->xh_entries
[start
], 0,
3815 sizeof(struct ocfs2_xattr_entry
) * (count
- start
));
3816 xh
->xh_count
= cpu_to_le16(start
);
3817 xh
->xh_free_start
= cpu_to_le16(name_offset
);
3818 xh
->xh_name_value_len
= cpu_to_le16(name_value_len
);
3820 ocfs2_xattr_bucket_journal_dirty(handle
, s_bucket
);
3823 ocfs2_xattr_bucket_free(s_bucket
);
3824 ocfs2_xattr_bucket_free(t_bucket
);
3830 * Copy xattr from one bucket to another bucket.
3832 * The caller must make sure that the journal transaction
3833 * has enough space for journaling.
3835 static int ocfs2_cp_xattr_bucket(struct inode
*inode
,
3842 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
3844 BUG_ON(s_blkno
== t_blkno
);
3846 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3847 (unsigned long long)s_blkno
, (unsigned long long)t_blkno
,
3850 s_bucket
= ocfs2_xattr_bucket_new(inode
);
3851 t_bucket
= ocfs2_xattr_bucket_new(inode
);
3852 if (!s_bucket
|| !t_bucket
) {
3858 ret
= ocfs2_read_xattr_bucket(s_bucket
, s_blkno
);
3863 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3864 * there's no need to read it.
3866 ret
= ocfs2_init_xattr_bucket(t_bucket
, t_blkno
);
3870 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
3872 OCFS2_JOURNAL_ACCESS_CREATE
:
3873 OCFS2_JOURNAL_ACCESS_WRITE
);
3877 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
3878 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
3881 ocfs2_xattr_bucket_free(t_bucket
);
3882 ocfs2_xattr_bucket_free(s_bucket
);
3888 * Copy one xattr cluster from src_blk to to_blk.
3889 * The to_blk will become the first bucket header of the cluster, so its
3890 * xh_num_buckets will be initialized as the bucket num in the cluster.
3892 static int ocfs2_cp_xattr_cluster(struct inode
*inode
,
3894 struct buffer_head
*first_bh
,
3899 int i
, ret
, credits
;
3900 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3901 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
3902 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
3903 struct buffer_head
*bh
= NULL
;
3904 struct ocfs2_xattr_header
*xh
;
3905 u64 to_blk_start
= to_blk
;
3907 mlog(0, "cp xattrs from cluster %llu to %llu\n",
3908 (unsigned long long)src_blk
, (unsigned long long)to_blk
);
3911 * We need to update the new cluster and 1 more for the update of
3912 * the 1st bucket of the previous extent rec.
3914 credits
= bpc
+ 1 + handle
->h_buffer_credits
;
3915 ret
= ocfs2_extend_trans(handle
, credits
);
3921 ret
= ocfs2_journal_access(handle
, inode
, first_bh
,
3922 OCFS2_JOURNAL_ACCESS_WRITE
);
3928 for (i
= 0; i
< num_buckets
; i
++) {
3929 ret
= ocfs2_cp_xattr_bucket(inode
, handle
,
3930 src_blk
, to_blk
, 1);
3936 src_blk
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3937 to_blk
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3940 /* update the old bucket header. */
3941 xh
= (struct ocfs2_xattr_header
*)first_bh
->b_data
;
3942 le16_add_cpu(&xh
->xh_num_buckets
, -num_buckets
);
3944 ocfs2_journal_dirty(handle
, first_bh
);
3946 /* update the new bucket header. */
3947 ret
= ocfs2_read_block(inode
, to_blk_start
, &bh
);
3953 ret
= ocfs2_journal_access(handle
, inode
, bh
,
3954 OCFS2_JOURNAL_ACCESS_WRITE
);
3960 xh
= (struct ocfs2_xattr_header
*)bh
->b_data
;
3961 xh
->xh_num_buckets
= cpu_to_le16(num_buckets
);
3963 ocfs2_journal_dirty(handle
, bh
);
3966 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
3973 * Move some xattrs in this cluster to the new cluster.
3974 * This function should only be called when bucket size == cluster size.
3975 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3977 static int ocfs2_divide_xattr_cluster(struct inode
*inode
,
3983 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3984 int ret
, credits
= 2 * blk_per_bucket
+ handle
->h_buffer_credits
;
3986 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
< OCFS2_SB(inode
->i_sb
)->s_clustersize
);
3988 ret
= ocfs2_extend_trans(handle
, credits
);
3994 /* Move half of the xattr in start_blk to the next bucket. */
3995 return ocfs2_divide_xattr_bucket(inode
, handle
, prev_blk
,
3996 new_blk
, first_hash
, 1);
4000 * Move some xattrs from the old cluster to the new one since they are not
4001 * contiguous in ocfs2 xattr tree.
4003 * new_blk starts a new separate cluster, and we will move some xattrs from
4004 * prev_blk to it. v_start will be set as the first name hash value in this
4005 * new cluster so that it can be used as e_cpos during tree insertion and
4006 * don't collide with our original b-tree operations. first_bh and header_bh
4007 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4008 * to extend the insert bucket.
4010 * The problem is how much xattr should we move to the new one and when should
4011 * we update first_bh and header_bh?
4012 * 1. If cluster size > bucket size, that means the previous cluster has more
4013 * than 1 bucket, so just move half nums of bucket into the new cluster and
4014 * update the first_bh and header_bh if the insert bucket has been moved
4015 * to the new cluster.
4016 * 2. If cluster_size == bucket_size:
4017 * a) If the previous extent rec has more than one cluster and the insert
4018 * place isn't in the last cluster, copy the entire last cluster to the
4019 * new one. This time, we don't need to upate the first_bh and header_bh
4020 * since they will not be moved into the new cluster.
4021 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4022 * the new one. And we set the extend flag to zero if the insert place is
4023 * moved into the new allocated cluster since no extend is needed.
4025 static int ocfs2_adjust_xattr_cross_cluster(struct inode
*inode
,
4027 struct buffer_head
**first_bh
,
4028 struct buffer_head
**header_bh
,
4036 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
4038 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4039 (unsigned long long)prev_blk
, prev_clusters
,
4040 (unsigned long long)new_blk
);
4042 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
)) > 1)
4043 ret
= ocfs2_mv_xattr_bucket_cross_cluster(inode
,
4052 u64 last_blk
= prev_blk
+ bpc
* (prev_clusters
- 1);
4054 if (prev_clusters
> 1 && (*header_bh
)->b_blocknr
!= last_blk
)
4055 ret
= ocfs2_cp_xattr_cluster(inode
, handle
, *first_bh
,
4059 ret
= ocfs2_divide_xattr_cluster(inode
, handle
,
4063 if ((*header_bh
)->b_blocknr
== last_blk
&& extend
)
4072 * Add a new cluster for xattr storage.
4074 * If the new cluster is contiguous with the previous one, it will be
4075 * appended to the same extent record, and num_clusters will be updated.
4076 * If not, we will insert a new extent for it and move some xattrs in
4077 * the last cluster into the new allocated one.
4078 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4079 * lose the benefits of hashing because we'll have to search large leaves.
4080 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4083 * first_bh is the first block of the previous extent rec and header_bh
4084 * indicates the bucket we will insert the new xattrs. They will be updated
4085 * when the header_bh is moved into the new cluster.
4087 static int ocfs2_add_new_xattr_cluster(struct inode
*inode
,
4088 struct buffer_head
*root_bh
,
4089 struct buffer_head
**first_bh
,
4090 struct buffer_head
**header_bh
,
4095 struct ocfs2_xattr_set_ctxt
*ctxt
)
4098 u16 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
4099 u32 prev_clusters
= *num_clusters
;
4100 u32 clusters_to_add
= 1, bit_off
, num_bits
, v_start
= 0;
4102 handle_t
*handle
= ctxt
->handle
;
4103 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4104 struct ocfs2_extent_tree et
;
4106 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4107 "previous xattr blkno = %llu\n",
4108 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
4109 prev_cpos
, (unsigned long long)prev_blkno
);
4111 ocfs2_init_xattr_tree_extent_tree(&et
, inode
, root_bh
);
4113 ret
= ocfs2_journal_access(handle
, inode
, root_bh
,
4114 OCFS2_JOURNAL_ACCESS_WRITE
);
4120 ret
= __ocfs2_claim_clusters(osb
, handle
, ctxt
->data_ac
, 1,
4121 clusters_to_add
, &bit_off
, &num_bits
);
4128 BUG_ON(num_bits
> clusters_to_add
);
4130 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
4131 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4132 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
4134 if (prev_blkno
+ prev_clusters
* bpc
== block
&&
4135 (prev_clusters
+ num_bits
) << osb
->s_clustersize_bits
<=
4136 OCFS2_MAX_XATTR_TREE_LEAF_SIZE
) {
4138 * If this cluster is contiguous with the old one and
4139 * adding this new cluster, we don't surpass the limit of
4140 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4141 * initialized and used like other buckets in the previous
4143 * So add it as a contiguous one. The caller will handle
4146 v_start
= prev_cpos
+ prev_clusters
;
4147 *num_clusters
= prev_clusters
+ num_bits
;
4148 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4151 ret
= ocfs2_adjust_xattr_cross_cluster(inode
,
4166 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4167 num_bits
, (unsigned long long)block
, v_start
);
4168 ret
= ocfs2_insert_extent(osb
, handle
, inode
, &et
, v_start
, block
,
4169 num_bits
, 0, ctxt
->meta_ac
);
4175 ret
= ocfs2_journal_dirty(handle
, root_bh
);
4184 * Extend a new xattr bucket and move xattrs to the end one by one until
4185 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
4187 static int ocfs2_extend_xattr_bucket(struct inode
*inode
,
4189 struct buffer_head
*first_bh
,
4190 struct buffer_head
*start_bh
,
4194 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4195 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4196 u64 start_blk
= start_bh
->b_blocknr
, end_blk
;
4197 u32 num_buckets
= num_clusters
* ocfs2_xattr_buckets_per_cluster(osb
);
4198 struct ocfs2_xattr_header
*first_xh
=
4199 (struct ocfs2_xattr_header
*)first_bh
->b_data
;
4200 u16 bucket
= le16_to_cpu(first_xh
->xh_num_buckets
);
4202 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4203 "from %llu, len = %u\n", (unsigned long long)start_blk
,
4204 (unsigned long long)first_bh
->b_blocknr
, num_clusters
);
4206 BUG_ON(bucket
>= num_buckets
);
4208 end_blk
= first_bh
->b_blocknr
+ (bucket
- 1) * blk_per_bucket
;
4211 * We will touch all the buckets after the start_bh(include it).
4212 * Then we add one more bucket.
4214 credits
= end_blk
- start_blk
+ 3 * blk_per_bucket
+ 1 +
4215 handle
->h_buffer_credits
;
4216 ret
= ocfs2_extend_trans(handle
, credits
);
4222 ret
= ocfs2_journal_access(handle
, inode
, first_bh
,
4223 OCFS2_JOURNAL_ACCESS_WRITE
);
4229 while (end_blk
!= start_blk
) {
4230 ret
= ocfs2_cp_xattr_bucket(inode
, handle
, end_blk
,
4231 end_blk
+ blk_per_bucket
, 0);
4234 end_blk
-= blk_per_bucket
;
4237 /* Move half of the xattr in start_blk to the next bucket. */
4238 ret
= ocfs2_divide_xattr_bucket(inode
, handle
, start_blk
,
4239 start_blk
+ blk_per_bucket
, NULL
, 0);
4241 le16_add_cpu(&first_xh
->xh_num_buckets
, 1);
4242 ocfs2_journal_dirty(handle
, first_bh
);
4249 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
4250 * xb_bh is the ocfs2_xattr_block.
4251 * We will move all the buckets starting from header_bh to the next place. As
4252 * for this one, half num of its xattrs will be moved to the next one.
4254 * We will allocate a new cluster if current cluster is full and adjust
4255 * header_bh and first_bh if the insert place is moved to the new cluster.
4257 static int ocfs2_add_new_xattr_bucket(struct inode
*inode
,
4258 struct buffer_head
*xb_bh
,
4259 struct buffer_head
*header_bh
,
4260 struct ocfs2_xattr_set_ctxt
*ctxt
)
4262 struct ocfs2_xattr_header
*first_xh
= NULL
;
4263 struct buffer_head
*first_bh
= NULL
;
4264 struct ocfs2_xattr_block
*xb
=
4265 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4266 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
4267 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
4268 struct ocfs2_xattr_header
*xh
=
4269 (struct ocfs2_xattr_header
*)header_bh
->b_data
;
4270 u32 name_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
4271 struct super_block
*sb
= inode
->i_sb
;
4272 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
4273 int ret
, num_buckets
, extend
= 1;
4275 u32 e_cpos
, num_clusters
;
4277 mlog(0, "Add new xattr bucket starting form %llu\n",
4278 (unsigned long long)header_bh
->b_blocknr
);
4281 * Add refrence for header_bh here because it may be
4282 * changed in ocfs2_add_new_xattr_cluster and we need
4283 * to free it in the end.
4287 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &e_cpos
,
4294 ret
= ocfs2_read_block(inode
, p_blkno
, &first_bh
);
4300 num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
) * num_clusters
;
4301 first_xh
= (struct ocfs2_xattr_header
*)first_bh
->b_data
;
4303 if (num_buckets
== le16_to_cpu(first_xh
->xh_num_buckets
)) {
4304 ret
= ocfs2_add_new_xattr_cluster(inode
,
4320 ret
= ocfs2_extend_xattr_bucket(inode
,
4333 static inline char *ocfs2_xattr_bucket_get_val(struct inode
*inode
,
4334 struct ocfs2_xattr_bucket
*bucket
,
4337 int block_off
= offs
>> inode
->i_sb
->s_blocksize_bits
;
4339 offs
= offs
% inode
->i_sb
->s_blocksize
;
4340 return bucket_block(bucket
, block_off
) + offs
;
4344 * Handle the normal xattr set, including replace, delete and new.
4346 * Note: "local" indicates the real data's locality. So we can't
4347 * just its bucket locality by its length.
4349 static void ocfs2_xattr_set_entry_normal(struct inode
*inode
,
4350 struct ocfs2_xattr_info
*xi
,
4351 struct ocfs2_xattr_search
*xs
,
4355 struct ocfs2_xattr_entry
*last
, *xe
;
4356 int name_len
= strlen(xi
->name
);
4357 struct ocfs2_xattr_header
*xh
= xs
->header
;
4358 u16 count
= le16_to_cpu(xh
->xh_count
), start
;
4359 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4361 size_t offs
, size
, new_size
;
4363 last
= &xh
->xh_entries
[count
];
4364 if (!xs
->not_found
) {
4366 offs
= le16_to_cpu(xe
->xe_name_offset
);
4367 if (ocfs2_xattr_is_local(xe
))
4368 size
= OCFS2_XATTR_SIZE(name_len
) +
4369 OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
4371 size
= OCFS2_XATTR_SIZE(name_len
) +
4372 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE
);
4375 * If the new value will be stored outside, xi->value has been
4376 * initalized as an empty ocfs2_xattr_value_root, and the same
4377 * goes with xi->value_len, so we can set new_size safely here.
4378 * See ocfs2_xattr_set_in_bucket.
4380 new_size
= OCFS2_XATTR_SIZE(name_len
) +
4381 OCFS2_XATTR_SIZE(xi
->value_len
);
4383 le16_add_cpu(&xh
->xh_name_value_len
, -size
);
4385 if (new_size
> size
)
4386 goto set_new_name_value
;
4388 /* Now replace the old value with new one. */
4390 xe
->xe_value_size
= cpu_to_le64(xi
->value_len
);
4392 xe
->xe_value_size
= 0;
4394 val
= ocfs2_xattr_bucket_get_val(inode
,
4396 memset(val
+ OCFS2_XATTR_SIZE(name_len
), 0,
4397 size
- OCFS2_XATTR_SIZE(name_len
));
4398 if (OCFS2_XATTR_SIZE(xi
->value_len
) > 0)
4399 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
),
4400 xi
->value
, xi
->value_len
);
4402 le16_add_cpu(&xh
->xh_name_value_len
, new_size
);
4403 ocfs2_xattr_set_local(xe
, local
);
4407 * Remove the old entry if there is more than one.
4408 * We don't remove the last entry so that we can
4409 * use it to indicate the hash value of the empty
4413 le16_add_cpu(&xh
->xh_count
, -1);
4416 (void *)last
- (void *)xe
);
4418 sizeof(struct ocfs2_xattr_entry
));
4421 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
4426 /* find a new entry for insert. */
4427 int low
= 0, high
= count
- 1, tmp
;
4428 struct ocfs2_xattr_entry
*tmp_xe
;
4430 while (low
<= high
&& count
) {
4431 tmp
= (low
+ high
) / 2;
4432 tmp_xe
= &xh
->xh_entries
[tmp
];
4434 if (name_hash
> le32_to_cpu(tmp_xe
->xe_name_hash
))
4436 else if (name_hash
<
4437 le32_to_cpu(tmp_xe
->xe_name_hash
))
4445 xe
= &xh
->xh_entries
[low
];
4447 memmove(xe
+ 1, xe
, (void *)last
- (void *)xe
);
4449 le16_add_cpu(&xh
->xh_count
, 1);
4450 memset(xe
, 0, sizeof(struct ocfs2_xattr_entry
));
4451 xe
->xe_name_hash
= cpu_to_le32(name_hash
);
4452 xe
->xe_name_len
= name_len
;
4453 ocfs2_xattr_set_type(xe
, xi
->name_index
);
4457 /* Insert the new name+value. */
4458 size
= OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(xi
->value_len
);
4461 * We must make sure that the name/value pair
4462 * exists in the same block.
4464 offs
= le16_to_cpu(xh
->xh_free_start
);
4465 start
= offs
- size
;
4467 if (start
>> inode
->i_sb
->s_blocksize_bits
!=
4468 (offs
- 1) >> inode
->i_sb
->s_blocksize_bits
) {
4469 offs
= offs
- offs
% blocksize
;
4470 xh
->xh_free_start
= cpu_to_le16(offs
);
4473 val
= ocfs2_xattr_bucket_get_val(inode
, xs
->bucket
, offs
- size
);
4474 xe
->xe_name_offset
= cpu_to_le16(offs
- size
);
4476 memset(val
, 0, size
);
4477 memcpy(val
, xi
->name
, name_len
);
4478 memcpy(val
+ OCFS2_XATTR_SIZE(name_len
), xi
->value
, xi
->value_len
);
4480 xe
->xe_value_size
= cpu_to_le64(xi
->value_len
);
4481 ocfs2_xattr_set_local(xe
, local
);
4483 le16_add_cpu(&xh
->xh_free_start
, -size
);
4484 le16_add_cpu(&xh
->xh_name_value_len
, size
);
4490 * Set the xattr entry in the specified bucket.
4491 * The bucket is indicated by xs->bucket and it should have the enough
4492 * space for the xattr insertion.
4494 static int ocfs2_xattr_set_entry_in_bucket(struct inode
*inode
,
4496 struct ocfs2_xattr_info
*xi
,
4497 struct ocfs2_xattr_search
*xs
,
4504 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4505 (unsigned long)xi
->value_len
, xi
->name_index
,
4506 (unsigned long long)bucket_blkno(xs
->bucket
));
4508 if (!xs
->bucket
->bu_bhs
[1]) {
4509 blkno
= bucket_blkno(xs
->bucket
);
4510 ocfs2_xattr_bucket_relse(xs
->bucket
);
4511 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, blkno
);
4518 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4519 OCFS2_JOURNAL_ACCESS_WRITE
);
4525 ocfs2_xattr_set_entry_normal(inode
, xi
, xs
, name_hash
, local
);
4526 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4532 static int ocfs2_xattr_value_update_size(struct inode
*inode
,
4534 struct buffer_head
*xe_bh
,
4535 struct ocfs2_xattr_entry
*xe
,
4540 ret
= ocfs2_journal_access(handle
, inode
, xe_bh
,
4541 OCFS2_JOURNAL_ACCESS_WRITE
);
4547 xe
->xe_value_size
= cpu_to_le64(new_size
);
4549 ret
= ocfs2_journal_dirty(handle
, xe_bh
);
4558 * Truncate the specified xe_off entry in xattr bucket.
4559 * bucket is indicated by header_bh and len is the new length.
4560 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4562 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4564 static int ocfs2_xattr_bucket_value_truncate(struct inode
*inode
,
4565 struct buffer_head
*header_bh
,
4568 struct ocfs2_xattr_set_ctxt
*ctxt
)
4572 struct buffer_head
*value_bh
= NULL
;
4573 struct ocfs2_xattr_value_root
*xv
;
4574 struct ocfs2_xattr_entry
*xe
;
4575 struct ocfs2_xattr_header
*xh
=
4576 (struct ocfs2_xattr_header
*)header_bh
->b_data
;
4577 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4579 xe
= &xh
->xh_entries
[xe_off
];
4581 BUG_ON(!xe
|| ocfs2_xattr_is_local(xe
));
4583 offset
= le16_to_cpu(xe
->xe_name_offset
) +
4584 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
4586 value_blk
= offset
/ blocksize
;
4588 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4589 BUG_ON(value_blk
!= (offset
+ OCFS2_XATTR_ROOT_SIZE
- 1) / blocksize
);
4590 value_blk
+= header_bh
->b_blocknr
;
4592 ret
= ocfs2_read_block(inode
, value_blk
, &value_bh
);
4598 xv
= (struct ocfs2_xattr_value_root
*)
4599 (value_bh
->b_data
+ offset
% blocksize
);
4601 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4602 xe_off
, (unsigned long long)header_bh
->b_blocknr
, len
);
4603 ret
= ocfs2_xattr_value_truncate(inode
, value_bh
, xv
, len
, ctxt
);
4609 ret
= ocfs2_xattr_value_update_size(inode
, ctxt
->handle
,
4610 header_bh
, xe
, len
);
4621 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode
*inode
,
4622 struct ocfs2_xattr_search
*xs
,
4624 struct ocfs2_xattr_set_ctxt
*ctxt
)
4627 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4628 struct ocfs2_xattr_header
*xh
= (struct ocfs2_xattr_header
*)xs
->base
;
4630 BUG_ON(!xs
->bucket
->bu_bhs
[0] || !xe
|| ocfs2_xattr_is_local(xe
));
4632 offset
= xe
- xh
->xh_entries
;
4633 ret
= ocfs2_xattr_bucket_value_truncate(inode
, xs
->bucket
->bu_bhs
[0],
4641 static int ocfs2_xattr_bucket_set_value_outside(struct inode
*inode
,
4643 struct ocfs2_xattr_search
*xs
,
4648 struct ocfs2_xattr_value_root
*xv
;
4649 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4651 BUG_ON(!xs
->base
|| !xe
|| ocfs2_xattr_is_local(xe
));
4653 offset
= le16_to_cpu(xe
->xe_name_offset
) +
4654 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
4656 xv
= (struct ocfs2_xattr_value_root
*)(xs
->base
+ offset
);
4658 return __ocfs2_xattr_set_value_outside(inode
, handle
,
4659 xv
, val
, value_len
);
4662 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
4663 struct buffer_head
*root_bh
,
4669 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4670 struct inode
*tl_inode
= osb
->osb_tl_inode
;
4672 struct ocfs2_xattr_block
*xb
=
4673 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
4674 struct ocfs2_alloc_context
*meta_ac
= NULL
;
4675 struct ocfs2_cached_dealloc_ctxt dealloc
;
4676 struct ocfs2_extent_tree et
;
4678 ocfs2_init_xattr_tree_extent_tree(&et
, inode
, root_bh
);
4680 ocfs2_init_dealloc_ctxt(&dealloc
);
4682 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4683 cpos
, len
, (unsigned long long)blkno
);
4685 ocfs2_remove_xattr_clusters_from_cache(inode
, blkno
, len
);
4687 ret
= ocfs2_lock_allocators(inode
, &et
, 0, 1, NULL
, &meta_ac
);
4693 mutex_lock(&tl_inode
->i_mutex
);
4695 if (ocfs2_truncate_log_needs_flush(osb
)) {
4696 ret
= __ocfs2_flush_truncate_log(osb
);
4703 handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
4704 if (IS_ERR(handle
)) {
4710 ret
= ocfs2_journal_access(handle
, inode
, root_bh
,
4711 OCFS2_JOURNAL_ACCESS_WRITE
);
4717 ret
= ocfs2_remove_extent(inode
, &et
, cpos
, len
, handle
, meta_ac
,
4724 le32_add_cpu(&xb
->xb_attrs
.xb_root
.xt_clusters
, -len
);
4726 ret
= ocfs2_journal_dirty(handle
, root_bh
);
4732 ret
= ocfs2_truncate_log_append(osb
, handle
, blkno
, len
);
4737 ocfs2_commit_trans(osb
, handle
);
4739 ocfs2_schedule_truncate_log_flush(osb
, 1);
4741 mutex_unlock(&tl_inode
->i_mutex
);
4744 ocfs2_free_alloc_context(meta_ac
);
4746 ocfs2_run_deallocs(osb
, &dealloc
);
4751 static void ocfs2_xattr_bucket_remove_xs(struct inode
*inode
,
4753 struct ocfs2_xattr_search
*xs
)
4755 struct ocfs2_xattr_header
*xh
= bucket_xh(xs
->bucket
);
4756 struct ocfs2_xattr_entry
*last
= &xh
->xh_entries
[
4757 le16_to_cpu(xh
->xh_count
) - 1];
4760 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4761 OCFS2_JOURNAL_ACCESS_WRITE
);
4767 /* Remove the old entry. */
4768 memmove(xs
->here
, xs
->here
+ 1,
4769 (void *)last
- (void *)xs
->here
);
4770 memset(last
, 0, sizeof(struct ocfs2_xattr_entry
));
4771 le16_add_cpu(&xh
->xh_count
, -1);
4773 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4777 * Set the xattr name/value in the bucket specified in xs.
4779 * As the new value in xi may be stored in the bucket or in an outside cluster,
4780 * we divide the whole process into 3 steps:
4781 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4782 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4783 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4784 * 4. If the clusters for the new outside value can't be allocated, we need
4785 * to free the xattr we allocated in set.
4787 static int ocfs2_xattr_set_in_bucket(struct inode
*inode
,
4788 struct ocfs2_xattr_info
*xi
,
4789 struct ocfs2_xattr_search
*xs
,
4790 struct ocfs2_xattr_set_ctxt
*ctxt
)
4794 char *val
= (char *)xi
->value
;
4795 struct ocfs2_xattr_entry
*xe
= xs
->here
;
4796 u32 name_hash
= ocfs2_xattr_name_hash(inode
, xi
->name
,
4799 if (!xs
->not_found
&& !ocfs2_xattr_is_local(xe
)) {
4801 * We need to truncate the xattr storage first.
4803 * If both the old and new value are stored to
4804 * outside block, we only need to truncate
4805 * the storage and then set the value outside.
4807 * If the new value should be stored within block,
4808 * we should free all the outside block first and
4809 * the modification to the xattr block will be done
4810 * by following steps.
4812 if (xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
4813 value_len
= xi
->value_len
;
4817 ret
= ocfs2_xattr_bucket_value_truncate_xs(inode
, xs
,
4824 goto set_value_outside
;
4827 value_len
= xi
->value_len
;
4828 /* So we have to handle the inside block change now. */
4829 if (value_len
> OCFS2_XATTR_INLINE_SIZE
) {
4831 * If the new value will be stored outside of block,
4832 * initalize a new empty value root and insert it first.
4835 xi
->value
= &def_xv
;
4836 xi
->value_len
= OCFS2_XATTR_ROOT_SIZE
;
4839 ret
= ocfs2_xattr_set_entry_in_bucket(inode
, ctxt
->handle
, xi
, xs
,
4846 if (value_len
<= OCFS2_XATTR_INLINE_SIZE
)
4849 /* allocate the space now for the outside block storage. */
4850 ret
= ocfs2_xattr_bucket_value_truncate_xs(inode
, xs
,
4855 if (xs
->not_found
) {
4857 * We can't allocate enough clusters for outside
4858 * storage and we have allocated xattr already,
4859 * so need to remove it.
4861 ocfs2_xattr_bucket_remove_xs(inode
, ctxt
->handle
, xs
);
4867 ret
= ocfs2_xattr_bucket_set_value_outside(inode
, ctxt
->handle
,
4868 xs
, val
, value_len
);
4874 * check whether the xattr bucket is filled up with the same hash value.
4875 * If we want to insert the xattr with the same hash, return -ENOSPC.
4876 * If we want to insert a xattr with different hash value, go ahead
4877 * and ocfs2_divide_xattr_bucket will handle this.
4879 static int ocfs2_check_xattr_bucket_collision(struct inode
*inode
,
4880 struct ocfs2_xattr_bucket
*bucket
,
4883 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
4884 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
4886 if (name_hash
!= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
))
4889 if (xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1].xe_name_hash
==
4890 xh
->xh_entries
[0].xe_name_hash
) {
4891 mlog(ML_ERROR
, "Too much hash collision in xattr bucket %llu, "
4893 (unsigned long long)bucket_blkno(bucket
),
4894 le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
));
4901 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
4902 struct ocfs2_xattr_info
*xi
,
4903 struct ocfs2_xattr_search
*xs
,
4904 struct ocfs2_xattr_set_ctxt
*ctxt
)
4906 struct ocfs2_xattr_header
*xh
;
4907 struct ocfs2_xattr_entry
*xe
;
4908 u16 count
, header_size
, xh_free_start
;
4909 int free
, max_free
, need
, old
;
4910 size_t value_size
= 0, name_len
= strlen(xi
->name
);
4911 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4912 int ret
, allocation
= 0;
4914 mlog_entry("Set xattr %s in xattr index block\n", xi
->name
);
4918 count
= le16_to_cpu(xh
->xh_count
);
4919 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
4920 header_size
= sizeof(struct ocfs2_xattr_header
) +
4921 count
* sizeof(struct ocfs2_xattr_entry
);
4922 max_free
= OCFS2_XATTR_BUCKET_SIZE
-
4923 le16_to_cpu(xh
->xh_name_value_len
) - header_size
;
4925 mlog_bug_on_msg(header_size
> blocksize
, "bucket %llu has header size "
4926 "of %u which exceed block size\n",
4927 (unsigned long long)bucket_blkno(xs
->bucket
),
4930 if (xi
->value
&& xi
->value_len
> OCFS2_XATTR_INLINE_SIZE
)
4931 value_size
= OCFS2_XATTR_ROOT_SIZE
;
4933 value_size
= OCFS2_XATTR_SIZE(xi
->value_len
);
4936 need
= sizeof(struct ocfs2_xattr_entry
) +
4937 OCFS2_XATTR_SIZE(name_len
) + value_size
;
4939 need
= value_size
+ OCFS2_XATTR_SIZE(name_len
);
4942 * We only replace the old value if the new length is smaller
4943 * than the old one. Otherwise we will allocate new space in the
4944 * bucket to store it.
4947 if (ocfs2_xattr_is_local(xe
))
4948 old
= OCFS2_XATTR_SIZE(le64_to_cpu(xe
->xe_value_size
));
4950 old
= OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE
);
4952 if (old
>= value_size
)
4956 free
= xh_free_start
- header_size
;
4958 * We need to make sure the new name/value pair
4959 * can exist in the same block.
4961 if (xh_free_start
% blocksize
< need
)
4962 free
-= xh_free_start
% blocksize
;
4964 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4965 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4966 " %u\n", xs
->not_found
,
4967 (unsigned long long)bucket_blkno(xs
->bucket
),
4968 free
, need
, max_free
, le16_to_cpu(xh
->xh_free_start
),
4969 le16_to_cpu(xh
->xh_name_value_len
));
4973 count
== ocfs2_xattr_max_xe_in_bucket(inode
->i_sb
))) {
4974 if (need
<= max_free
&&
4975 count
< ocfs2_xattr_max_xe_in_bucket(inode
->i_sb
)) {
4977 * We can create the space by defragment. Since only the
4978 * name/value will be moved, the xe shouldn't be changed
4981 ret
= ocfs2_defrag_xattr_bucket(inode
, ctxt
->handle
,
4988 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
4989 free
= xh_free_start
- header_size
;
4990 if (xh_free_start
% blocksize
< need
)
4991 free
-= xh_free_start
% blocksize
;
4996 mlog(0, "Can't get enough space for xattr insert by "
4997 "defragment. Need %u bytes, but we have %d, so "
4998 "allocate new bucket for it.\n", need
, free
);
5002 * We have to add new buckets or clusters and one
5003 * allocation should leave us enough space for insert.
5008 * We do not allow for overlapping ranges between buckets. And
5009 * the maximum number of collisions we will allow for then is
5010 * one bucket's worth, so check it here whether we need to
5011 * add a new bucket for the insert.
5013 ret
= ocfs2_check_xattr_bucket_collision(inode
,
5021 ret
= ocfs2_add_new_xattr_bucket(inode
,
5023 xs
->bucket
->bu_bhs
[0],
5030 ocfs2_xattr_bucket_relse(xs
->bucket
);
5032 ret
= ocfs2_xattr_index_block_find(inode
, xs
->xattr_bh
,
5035 if (ret
&& ret
!= -ENODATA
)
5037 xs
->not_found
= ret
;
5043 ret
= ocfs2_xattr_set_in_bucket(inode
, xi
, xs
, ctxt
);
5049 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
5050 struct ocfs2_xattr_bucket
*bucket
,
5054 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5056 struct ocfs2_xattr_entry
*xe
;
5057 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5058 struct ocfs2_xattr_set_ctxt ctxt
= {NULL
, NULL
,};
5060 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
5062 ctxt
.handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
5063 if (IS_ERR(ctxt
.handle
)) {
5064 ret
= PTR_ERR(ctxt
.handle
);
5069 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
5070 xe
= &xh
->xh_entries
[i
];
5071 if (ocfs2_xattr_is_local(xe
))
5074 ret
= ocfs2_xattr_bucket_value_truncate(inode
,
5083 ret
= ocfs2_commit_trans(osb
, ctxt
.handle
);
5084 ocfs2_schedule_truncate_log_flush(osb
, 1);
5085 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
5090 static int ocfs2_delete_xattr_index_block(struct inode
*inode
,
5091 struct buffer_head
*xb_bh
)
5093 struct ocfs2_xattr_block
*xb
=
5094 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
5095 struct ocfs2_extent_list
*el
= &xb
->xb_attrs
.xb_root
.xt_list
;
5097 u32 name_hash
= UINT_MAX
, e_cpos
, num_clusters
;
5100 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
5103 while (name_hash
> 0) {
5104 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
5105 &e_cpos
, &num_clusters
, el
);
5111 ret
= ocfs2_iterate_xattr_buckets(inode
, p_blkno
, num_clusters
,
5112 ocfs2_delete_xattr_in_bucket
,
5119 ret
= ocfs2_rm_xattr_cluster(inode
, xb_bh
,
5120 p_blkno
, e_cpos
, num_clusters
);
5129 name_hash
= e_cpos
- 1;
5137 * 'security' attributes support
5139 static size_t ocfs2_xattr_security_list(struct inode
*inode
, char *list
,
5140 size_t list_size
, const char *name
,
5143 const size_t prefix_len
= XATTR_SECURITY_PREFIX_LEN
;
5144 const size_t total_len
= prefix_len
+ name_len
+ 1;
5146 if (list
&& total_len
<= list_size
) {
5147 memcpy(list
, XATTR_SECURITY_PREFIX
, prefix_len
);
5148 memcpy(list
+ prefix_len
, name
, name_len
);
5149 list
[prefix_len
+ name_len
] = '\0';
5154 static int ocfs2_xattr_security_get(struct inode
*inode
, const char *name
,
5155 void *buffer
, size_t size
)
5157 if (strcmp(name
, "") == 0)
5159 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_SECURITY
, name
,
5163 static int ocfs2_xattr_security_set(struct inode
*inode
, const char *name
,
5164 const void *value
, size_t size
, int flags
)
5166 if (strcmp(name
, "") == 0)
5169 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_SECURITY
, name
, value
,
5173 int ocfs2_init_security_get(struct inode
*inode
,
5175 struct ocfs2_security_xattr_info
*si
)
5177 return security_inode_init_security(inode
, dir
, &si
->name
, &si
->value
,
5181 int ocfs2_init_security_set(handle_t
*handle
,
5182 struct inode
*inode
,
5183 struct buffer_head
*di_bh
,
5184 struct ocfs2_security_xattr_info
*si
,
5185 struct ocfs2_alloc_context
*xattr_ac
,
5186 struct ocfs2_alloc_context
*data_ac
)
5188 return ocfs2_xattr_set_handle(handle
, inode
, di_bh
,
5189 OCFS2_XATTR_INDEX_SECURITY
,
5190 si
->name
, si
->value
, si
->value_len
, 0,
5194 struct xattr_handler ocfs2_xattr_security_handler
= {
5195 .prefix
= XATTR_SECURITY_PREFIX
,
5196 .list
= ocfs2_xattr_security_list
,
5197 .get
= ocfs2_xattr_security_get
,
5198 .set
= ocfs2_xattr_security_set
,
5202 * 'trusted' attributes support
5204 static size_t ocfs2_xattr_trusted_list(struct inode
*inode
, char *list
,
5205 size_t list_size
, const char *name
,
5208 const size_t prefix_len
= XATTR_TRUSTED_PREFIX_LEN
;
5209 const size_t total_len
= prefix_len
+ name_len
+ 1;
5211 if (list
&& total_len
<= list_size
) {
5212 memcpy(list
, XATTR_TRUSTED_PREFIX
, prefix_len
);
5213 memcpy(list
+ prefix_len
, name
, name_len
);
5214 list
[prefix_len
+ name_len
] = '\0';
5219 static int ocfs2_xattr_trusted_get(struct inode
*inode
, const char *name
,
5220 void *buffer
, size_t size
)
5222 if (strcmp(name
, "") == 0)
5224 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_TRUSTED
, name
,
5228 static int ocfs2_xattr_trusted_set(struct inode
*inode
, const char *name
,
5229 const void *value
, size_t size
, int flags
)
5231 if (strcmp(name
, "") == 0)
5234 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_TRUSTED
, name
, value
,
5238 struct xattr_handler ocfs2_xattr_trusted_handler
= {
5239 .prefix
= XATTR_TRUSTED_PREFIX
,
5240 .list
= ocfs2_xattr_trusted_list
,
5241 .get
= ocfs2_xattr_trusted_get
,
5242 .set
= ocfs2_xattr_trusted_set
,
5246 * 'user' attributes support
5248 static size_t ocfs2_xattr_user_list(struct inode
*inode
, char *list
,
5249 size_t list_size
, const char *name
,
5252 const size_t prefix_len
= XATTR_USER_PREFIX_LEN
;
5253 const size_t total_len
= prefix_len
+ name_len
+ 1;
5254 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5256 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5259 if (list
&& total_len
<= list_size
) {
5260 memcpy(list
, XATTR_USER_PREFIX
, prefix_len
);
5261 memcpy(list
+ prefix_len
, name
, name_len
);
5262 list
[prefix_len
+ name_len
] = '\0';
5267 static int ocfs2_xattr_user_get(struct inode
*inode
, const char *name
,
5268 void *buffer
, size_t size
)
5270 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5272 if (strcmp(name
, "") == 0)
5274 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5276 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_USER
, name
,
5280 static int ocfs2_xattr_user_set(struct inode
*inode
, const char *name
,
5281 const void *value
, size_t size
, int flags
)
5283 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5285 if (strcmp(name
, "") == 0)
5287 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
5290 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_USER
, name
, value
,
5294 struct xattr_handler ocfs2_xattr_user_handler
= {
5295 .prefix
= XATTR_USER_PREFIX
,
5296 .list
= ocfs2_xattr_user_list
,
5297 .get
= ocfs2_xattr_user_get
,
5298 .set
= ocfs2_xattr_user_set
,