1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
22 #include <linux/capability.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
45 #include "blockcheck.h"
55 #include "buffer_head_io.h"
58 #include "refcounttree.h"
61 struct ocfs2_xattr_def_value_root
{
62 struct ocfs2_xattr_value_root xv
;
63 struct ocfs2_extent_rec er
;
66 struct ocfs2_xattr_bucket
{
67 /* The inode these xattrs are associated with */
68 struct inode
*bu_inode
;
70 /* The actual buffers that make up the bucket */
71 struct buffer_head
*bu_bhs
[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
];
73 /* How many blocks make up one bucket for this filesystem */
77 struct ocfs2_xattr_set_ctxt
{
79 struct ocfs2_alloc_context
*meta_ac
;
80 struct ocfs2_alloc_context
*data_ac
;
81 struct ocfs2_cached_dealloc_ctxt dealloc
;
84 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
85 #define OCFS2_XATTR_INLINE_SIZE 80
86 #define OCFS2_XATTR_HEADER_GAP 4
87 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
88 - sizeof(struct ocfs2_xattr_header) \
89 - OCFS2_XATTR_HEADER_GAP)
90 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
91 - sizeof(struct ocfs2_xattr_block) \
92 - sizeof(struct ocfs2_xattr_header) \
93 - OCFS2_XATTR_HEADER_GAP)
95 static struct ocfs2_xattr_def_value_root def_xv
= {
96 .xv
.xr_list
.l_count
= cpu_to_le16(1),
99 struct xattr_handler
*ocfs2_xattr_handlers
[] = {
100 &ocfs2_xattr_user_handler
,
101 &ocfs2_xattr_acl_access_handler
,
102 &ocfs2_xattr_acl_default_handler
,
103 &ocfs2_xattr_trusted_handler
,
104 &ocfs2_xattr_security_handler
,
108 static struct xattr_handler
*ocfs2_xattr_handler_map
[OCFS2_XATTR_MAX
] = {
109 [OCFS2_XATTR_INDEX_USER
] = &ocfs2_xattr_user_handler
,
110 [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
,
114 [OCFS2_XATTR_INDEX_TRUSTED
] = &ocfs2_xattr_trusted_handler
,
115 [OCFS2_XATTR_INDEX_SECURITY
] = &ocfs2_xattr_security_handler
,
118 struct ocfs2_xattr_info
{
122 const void *xi_value
;
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 /* Operations on struct ocfs2_xa_entry */
143 struct ocfs2_xa_loc_operations
{
147 int (*xlo_journal_access
)(handle_t
*handle
, struct ocfs2_xa_loc
*loc
,
149 void (*xlo_journal_dirty
)(handle_t
*handle
, struct ocfs2_xa_loc
*loc
);
152 * Return a pointer to the appropriate buffer in loc->xl_storage
153 * at the given offset from loc->xl_header.
155 void *(*xlo_offset_pointer
)(struct ocfs2_xa_loc
*loc
, int offset
);
157 /* Can we reuse the existing entry for the new value? */
158 int (*xlo_can_reuse
)(struct ocfs2_xa_loc
*loc
,
159 struct ocfs2_xattr_info
*xi
);
161 /* How much space is needed for the new value? */
162 int (*xlo_check_space
)(struct ocfs2_xa_loc
*loc
,
163 struct ocfs2_xattr_info
*xi
);
166 * Return the offset of the first name+value pair. This is
167 * the start of our downward-filling free space.
169 int (*xlo_get_free_start
)(struct ocfs2_xa_loc
*loc
);
172 * Remove the name+value at this location. Do whatever is
173 * appropriate with the remaining name+value pairs.
175 void (*xlo_wipe_namevalue
)(struct ocfs2_xa_loc
*loc
);
177 /* Fill xl_entry with a new entry */
178 void (*xlo_add_entry
)(struct ocfs2_xa_loc
*loc
, u32 name_hash
);
180 /* Add name+value storage to an entry */
181 void (*xlo_add_namevalue
)(struct ocfs2_xa_loc
*loc
, int size
);
184 * Initialize the value buf's access and bh fields for this entry.
185 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
187 void (*xlo_fill_value_buf
)(struct ocfs2_xa_loc
*loc
,
188 struct ocfs2_xattr_value_buf
*vb
);
192 * Describes an xattr entry location. This is a memory structure
193 * tracking the on-disk structure.
195 struct ocfs2_xa_loc
{
196 /* This xattr belongs to this inode */
197 struct inode
*xl_inode
;
199 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
200 struct ocfs2_xattr_header
*xl_header
;
202 /* Bytes from xl_header to the end of the storage */
206 * The ocfs2_xattr_entry this location describes. If this is
207 * NULL, this location describes the on-disk structure where it
210 struct ocfs2_xattr_entry
*xl_entry
;
213 * Internal housekeeping
216 /* Buffer(s) containing this entry */
219 /* Operations on the storage backing this location */
220 const struct ocfs2_xa_loc_operations
*xl_ops
;
224 * Convenience functions to calculate how much space is needed for a
225 * given name+value pair
227 static int namevalue_size(int name_len
, uint64_t value_len
)
229 if (value_len
> OCFS2_XATTR_INLINE_SIZE
)
230 return OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
232 return OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(value_len
);
235 static int namevalue_size_xi(struct ocfs2_xattr_info
*xi
)
237 return namevalue_size(xi
->xi_name_len
, xi
->xi_value_len
);
240 static int namevalue_size_xe(struct ocfs2_xattr_entry
*xe
)
242 u64 value_len
= le64_to_cpu(xe
->xe_value_size
);
244 BUG_ON((value_len
> OCFS2_XATTR_INLINE_SIZE
) &&
245 ocfs2_xattr_is_local(xe
));
246 return namevalue_size(xe
->xe_name_len
, value_len
);
250 static int ocfs2_xattr_bucket_get_name_value(struct super_block
*sb
,
251 struct ocfs2_xattr_header
*xh
,
256 static int ocfs2_xattr_block_find(struct inode
*inode
,
259 struct ocfs2_xattr_search
*xs
);
260 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
261 struct buffer_head
*root_bh
,
264 struct ocfs2_xattr_search
*xs
);
266 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
267 struct buffer_head
*blk_bh
,
271 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
272 struct ocfs2_xattr_search
*xs
,
273 struct ocfs2_xattr_set_ctxt
*ctxt
);
275 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
276 struct ocfs2_xattr_info
*xi
,
277 struct ocfs2_xattr_search
*xs
,
278 struct ocfs2_xattr_set_ctxt
*ctxt
);
280 typedef int (xattr_tree_rec_func
)(struct inode
*inode
,
281 struct buffer_head
*root_bh
,
282 u64 blkno
, u32 cpos
, u32 len
, void *para
);
283 static int ocfs2_iterate_xattr_index_block(struct inode
*inode
,
284 struct buffer_head
*root_bh
,
285 xattr_tree_rec_func
*rec_func
,
287 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
288 struct ocfs2_xattr_bucket
*bucket
,
290 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
291 struct buffer_head
*root_bh
,
297 static int ocfs2_mv_xattr_buckets(struct inode
*inode
, handle_t
*handle
,
298 u64 src_blk
, u64 last_blk
, u64 to_blk
,
299 unsigned int start_bucket
,
301 static int ocfs2_prepare_refcount_xattr(struct inode
*inode
,
302 struct ocfs2_dinode
*di
,
303 struct ocfs2_xattr_info
*xi
,
304 struct ocfs2_xattr_search
*xis
,
305 struct ocfs2_xattr_search
*xbs
,
306 struct ocfs2_refcount_tree
**ref_tree
,
309 static int ocfs2_get_xattr_tree_value_root(struct super_block
*sb
,
310 struct ocfs2_xattr_bucket
*bucket
,
312 struct ocfs2_xattr_value_root
**xv
,
313 struct buffer_head
**bh
);
315 static inline u16
ocfs2_xattr_buckets_per_cluster(struct ocfs2_super
*osb
)
317 return (1 << osb
->s_clustersize_bits
) / OCFS2_XATTR_BUCKET_SIZE
;
320 static inline u16
ocfs2_blocks_per_xattr_bucket(struct super_block
*sb
)
322 return OCFS2_XATTR_BUCKET_SIZE
/ (1 << sb
->s_blocksize_bits
);
325 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
326 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
327 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
329 static struct ocfs2_xattr_bucket
*ocfs2_xattr_bucket_new(struct inode
*inode
)
331 struct ocfs2_xattr_bucket
*bucket
;
332 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
334 BUG_ON(blks
> OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
);
336 bucket
= kzalloc(sizeof(struct ocfs2_xattr_bucket
), GFP_NOFS
);
338 bucket
->bu_inode
= inode
;
339 bucket
->bu_blocks
= blks
;
345 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket
*bucket
)
349 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
350 brelse(bucket
->bu_bhs
[i
]);
351 bucket
->bu_bhs
[i
] = NULL
;
355 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket
*bucket
)
358 ocfs2_xattr_bucket_relse(bucket
);
359 bucket
->bu_inode
= NULL
;
365 * A bucket that has never been written to disk doesn't need to be
366 * read. We just need the buffer_heads. Don't call this for
367 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
370 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
375 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
376 bucket
->bu_bhs
[i
] = sb_getblk(bucket
->bu_inode
->i_sb
,
378 if (!bucket
->bu_bhs
[i
]) {
384 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket
->bu_inode
),
386 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket
->bu_inode
),
391 ocfs2_xattr_bucket_relse(bucket
);
395 /* Read the xattr bucket at xb_blkno */
396 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
401 rc
= ocfs2_read_blocks(INODE_CACHE(bucket
->bu_inode
), xb_blkno
,
402 bucket
->bu_blocks
, bucket
->bu_bhs
, 0,
405 spin_lock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
406 rc
= ocfs2_validate_meta_ecc_bhs(bucket
->bu_inode
->i_sb
,
409 &bucket_xh(bucket
)->xh_check
);
410 spin_unlock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
416 ocfs2_xattr_bucket_relse(bucket
);
420 static int ocfs2_xattr_bucket_journal_access(handle_t
*handle
,
421 struct ocfs2_xattr_bucket
*bucket
,
426 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
427 rc
= ocfs2_journal_access(handle
,
428 INODE_CACHE(bucket
->bu_inode
),
429 bucket
->bu_bhs
[i
], type
);
439 static void ocfs2_xattr_bucket_journal_dirty(handle_t
*handle
,
440 struct ocfs2_xattr_bucket
*bucket
)
444 spin_lock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
445 ocfs2_compute_meta_ecc_bhs(bucket
->bu_inode
->i_sb
,
446 bucket
->bu_bhs
, bucket
->bu_blocks
,
447 &bucket_xh(bucket
)->xh_check
);
448 spin_unlock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
450 for (i
= 0; i
< bucket
->bu_blocks
; i
++)
451 ocfs2_journal_dirty(handle
, bucket
->bu_bhs
[i
]);
454 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket
*dest
,
455 struct ocfs2_xattr_bucket
*src
)
458 int blocksize
= src
->bu_inode
->i_sb
->s_blocksize
;
460 BUG_ON(dest
->bu_blocks
!= src
->bu_blocks
);
461 BUG_ON(dest
->bu_inode
!= src
->bu_inode
);
463 for (i
= 0; i
< src
->bu_blocks
; i
++) {
464 memcpy(bucket_block(dest
, i
), bucket_block(src
, i
),
469 static int ocfs2_validate_xattr_block(struct super_block
*sb
,
470 struct buffer_head
*bh
)
473 struct ocfs2_xattr_block
*xb
=
474 (struct ocfs2_xattr_block
*)bh
->b_data
;
476 mlog(0, "Validating xattr block %llu\n",
477 (unsigned long long)bh
->b_blocknr
);
479 BUG_ON(!buffer_uptodate(bh
));
482 * If the ecc fails, we return the error but otherwise
483 * leave the filesystem running. We know any error is
484 * local to this block.
486 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &xb
->xb_check
);
491 * Errors after here are fatal
494 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
496 "Extended attribute block #%llu has bad "
498 (unsigned long long)bh
->b_blocknr
, 7,
503 if (le64_to_cpu(xb
->xb_blkno
) != bh
->b_blocknr
) {
505 "Extended attribute block #%llu has an "
506 "invalid xb_blkno of %llu",
507 (unsigned long long)bh
->b_blocknr
,
508 (unsigned long long)le64_to_cpu(xb
->xb_blkno
));
512 if (le32_to_cpu(xb
->xb_fs_generation
) != OCFS2_SB(sb
)->fs_generation
) {
514 "Extended attribute block #%llu has an invalid "
515 "xb_fs_generation of #%u",
516 (unsigned long long)bh
->b_blocknr
,
517 le32_to_cpu(xb
->xb_fs_generation
));
524 static int ocfs2_read_xattr_block(struct inode
*inode
, u64 xb_blkno
,
525 struct buffer_head
**bh
)
528 struct buffer_head
*tmp
= *bh
;
530 rc
= ocfs2_read_block(INODE_CACHE(inode
), xb_blkno
, &tmp
,
531 ocfs2_validate_xattr_block
);
533 /* If ocfs2_read_block() got us a new bh, pass it up. */
540 static inline const char *ocfs2_xattr_prefix(int name_index
)
542 struct xattr_handler
*handler
= NULL
;
544 if (name_index
> 0 && name_index
< OCFS2_XATTR_MAX
)
545 handler
= ocfs2_xattr_handler_map
[name_index
];
547 return handler
? handler
->prefix
: NULL
;
550 static u32
ocfs2_xattr_name_hash(struct inode
*inode
,
554 /* Get hash value of uuid from super block */
555 u32 hash
= OCFS2_SB(inode
->i_sb
)->uuid_hash
;
558 /* hash extended attribute name */
559 for (i
= 0; i
< name_len
; i
++) {
560 hash
= (hash
<< OCFS2_HASH_SHIFT
) ^
561 (hash
>> (8*sizeof(hash
) - OCFS2_HASH_SHIFT
)) ^
568 static int ocfs2_xattr_entry_real_size(int name_len
, size_t value_len
)
570 return namevalue_size(name_len
, value_len
) +
571 sizeof(struct ocfs2_xattr_entry
);
574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info
*xi
)
576 return namevalue_size_xi(xi
) +
577 sizeof(struct ocfs2_xattr_entry
);
580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry
*xe
)
582 return namevalue_size_xe(xe
) +
583 sizeof(struct ocfs2_xattr_entry
);
586 int ocfs2_calc_security_init(struct inode
*dir
,
587 struct ocfs2_security_xattr_info
*si
,
590 struct ocfs2_alloc_context
**xattr_ac
)
593 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
594 int s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
598 * The max space of security xattr taken inline is
599 * 256(name) + 80(value) + 16(entry) = 352 bytes,
600 * So reserve one metadata block for it is ok.
602 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
603 s_size
> OCFS2_XATTR_FREE_IN_IBODY
) {
604 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
609 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
612 /* reserve clusters for xattr value which will be set in B tree*/
613 if (si
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
614 int new_clusters
= ocfs2_clusters_for_bytes(dir
->i_sb
,
617 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
619 *want_clusters
+= new_clusters
;
624 int ocfs2_calc_xattr_init(struct inode
*dir
,
625 struct buffer_head
*dir_bh
,
627 struct ocfs2_security_xattr_info
*si
,
633 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
634 int s_size
= 0, a_size
= 0, acl_len
= 0, new_clusters
;
637 s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
640 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
) {
641 acl_len
= ocfs2_xattr_get_nolock(dir
, dir_bh
,
642 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
,
645 a_size
= ocfs2_xattr_entry_real_size(0, acl_len
);
648 } else if (acl_len
!= 0 && acl_len
!= -ENODATA
) {
654 if (!(s_size
+ a_size
))
658 * The max space of security xattr taken inline is
659 * 256(name) + 80(value) + 16(entry) = 352 bytes,
660 * The max space of acl xattr taken inline is
661 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
662 * when blocksize = 512, may reserve one more cluser for
663 * xattr bucket, otherwise reserve one metadata block
665 * If this is a new directory with inline data,
666 * we choose to reserve the entire inline area for
667 * directory contents and force an external xattr block.
669 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
670 (S_ISDIR(mode
) && ocfs2_supports_inline_data(osb
)) ||
671 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_IBODY
) {
672 *want_meta
= *want_meta
+ 1;
673 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
676 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
&&
677 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_BLOCK(dir
)) {
679 *xattr_credits
+= ocfs2_blocks_per_xattr_bucket(dir
->i_sb
);
683 * reserve credits and clusters for xattrs which has large value
684 * and have to be set outside
686 if (si
->enable
&& si
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
687 new_clusters
= ocfs2_clusters_for_bytes(dir
->i_sb
,
689 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
691 *want_clusters
+= new_clusters
;
693 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
&&
694 acl_len
> OCFS2_XATTR_INLINE_SIZE
) {
695 /* for directory, it has DEFAULT and ACCESS two types of acls */
696 new_clusters
= (S_ISDIR(mode
) ? 2 : 1) *
697 ocfs2_clusters_for_bytes(dir
->i_sb
, acl_len
);
698 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
700 *want_clusters
+= new_clusters
;
706 static int ocfs2_xattr_extend_allocation(struct inode
*inode
,
708 struct ocfs2_xattr_value_buf
*vb
,
709 struct ocfs2_xattr_set_ctxt
*ctxt
)
712 handle_t
*handle
= ctxt
->handle
;
713 enum ocfs2_alloc_restarted why
;
714 u32 prev_clusters
, logical_start
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
715 struct ocfs2_extent_tree et
;
717 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add
);
719 ocfs2_init_xattr_value_extent_tree(&et
, INODE_CACHE(inode
), vb
);
721 status
= vb
->vb_access(handle
, INODE_CACHE(inode
), vb
->vb_bh
,
722 OCFS2_JOURNAL_ACCESS_WRITE
);
728 prev_clusters
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
729 status
= ocfs2_add_clusters_in_btree(handle
,
742 ocfs2_journal_dirty(handle
, vb
->vb_bh
);
744 clusters_to_add
-= le32_to_cpu(vb
->vb_xv
->xr_clusters
) - prev_clusters
;
747 * We should have already allocated enough space before the transaction,
748 * so no need to restart.
750 BUG_ON(why
!= RESTART_NONE
|| clusters_to_add
);
757 static int __ocfs2_remove_xattr_range(struct inode
*inode
,
758 struct ocfs2_xattr_value_buf
*vb
,
759 u32 cpos
, u32 phys_cpos
, u32 len
,
760 unsigned int ext_flags
,
761 struct ocfs2_xattr_set_ctxt
*ctxt
)
764 u64 phys_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
765 handle_t
*handle
= ctxt
->handle
;
766 struct ocfs2_extent_tree et
;
768 ocfs2_init_xattr_value_extent_tree(&et
, INODE_CACHE(inode
), vb
);
770 ret
= vb
->vb_access(handle
, INODE_CACHE(inode
), vb
->vb_bh
,
771 OCFS2_JOURNAL_ACCESS_WRITE
);
777 ret
= ocfs2_remove_extent(handle
, &et
, cpos
, len
, ctxt
->meta_ac
,
784 le32_add_cpu(&vb
->vb_xv
->xr_clusters
, -len
);
785 ocfs2_journal_dirty(handle
, vb
->vb_bh
);
787 if (ext_flags
& OCFS2_EXT_REFCOUNTED
)
788 ret
= ocfs2_decrease_refcount(inode
, handle
,
789 ocfs2_blocks_to_clusters(inode
->i_sb
,
791 len
, ctxt
->meta_ac
, &ctxt
->dealloc
, 1);
793 ret
= ocfs2_cache_cluster_dealloc(&ctxt
->dealloc
,
802 static int ocfs2_xattr_shrink_size(struct inode
*inode
,
805 struct ocfs2_xattr_value_buf
*vb
,
806 struct ocfs2_xattr_set_ctxt
*ctxt
)
809 unsigned int ext_flags
;
810 u32 trunc_len
, cpos
, phys_cpos
, alloc_size
;
813 if (old_clusters
<= new_clusters
)
817 trunc_len
= old_clusters
- new_clusters
;
819 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &phys_cpos
,
821 &vb
->vb_xv
->xr_list
, &ext_flags
);
827 if (alloc_size
> trunc_len
)
828 alloc_size
= trunc_len
;
830 ret
= __ocfs2_remove_xattr_range(inode
, vb
, cpos
,
831 phys_cpos
, alloc_size
,
838 block
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
839 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode
),
842 trunc_len
-= alloc_size
;
849 static int ocfs2_xattr_value_truncate(struct inode
*inode
,
850 struct ocfs2_xattr_value_buf
*vb
,
852 struct ocfs2_xattr_set_ctxt
*ctxt
)
855 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
856 u32 old_clusters
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
858 if (new_clusters
== old_clusters
)
861 if (new_clusters
> old_clusters
)
862 ret
= ocfs2_xattr_extend_allocation(inode
,
863 new_clusters
- old_clusters
,
866 ret
= ocfs2_xattr_shrink_size(inode
,
867 old_clusters
, new_clusters
,
873 static int ocfs2_xattr_list_entry(char *buffer
, size_t size
,
874 size_t *result
, const char *prefix
,
875 const char *name
, int name_len
)
877 char *p
= buffer
+ *result
;
878 int prefix_len
= strlen(prefix
);
879 int total_len
= prefix_len
+ name_len
+ 1;
881 *result
+= total_len
;
883 /* we are just looking for how big our buffer needs to be */
890 memcpy(p
, prefix
, prefix_len
);
891 memcpy(p
+ prefix_len
, name
, name_len
);
892 p
[prefix_len
+ name_len
] = '\0';
897 static int ocfs2_xattr_list_entries(struct inode
*inode
,
898 struct ocfs2_xattr_header
*header
,
899 char *buffer
, size_t buffer_size
)
903 const char *prefix
, *name
;
905 for (i
= 0 ; i
< le16_to_cpu(header
->xh_count
); i
++) {
906 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
907 type
= ocfs2_xattr_get_type(entry
);
908 prefix
= ocfs2_xattr_prefix(type
);
911 name
= (const char *)header
+
912 le16_to_cpu(entry
->xe_name_offset
);
914 ret
= ocfs2_xattr_list_entry(buffer
, buffer_size
,
915 &result
, prefix
, name
,
925 int ocfs2_has_inline_xattr_value_outside(struct inode
*inode
,
926 struct ocfs2_dinode
*di
)
928 struct ocfs2_xattr_header
*xh
;
931 xh
= (struct ocfs2_xattr_header
*)
932 ((void *)di
+ inode
->i_sb
->s_blocksize
-
933 le16_to_cpu(di
->i_xattr_inline_size
));
935 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++)
936 if (!ocfs2_xattr_is_local(&xh
->xh_entries
[i
]))
942 static int ocfs2_xattr_ibody_list(struct inode
*inode
,
943 struct ocfs2_dinode
*di
,
947 struct ocfs2_xattr_header
*header
= NULL
;
948 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
951 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
954 header
= (struct ocfs2_xattr_header
*)
955 ((void *)di
+ inode
->i_sb
->s_blocksize
-
956 le16_to_cpu(di
->i_xattr_inline_size
));
958 ret
= ocfs2_xattr_list_entries(inode
, header
, buffer
, buffer_size
);
963 static int ocfs2_xattr_block_list(struct inode
*inode
,
964 struct ocfs2_dinode
*di
,
968 struct buffer_head
*blk_bh
= NULL
;
969 struct ocfs2_xattr_block
*xb
;
972 if (!di
->i_xattr_loc
)
975 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
982 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
983 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
984 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
985 ret
= ocfs2_xattr_list_entries(inode
, header
,
986 buffer
, buffer_size
);
988 ret
= ocfs2_xattr_tree_list_index_block(inode
, blk_bh
,
989 buffer
, buffer_size
);
996 ssize_t
ocfs2_listxattr(struct dentry
*dentry
,
1000 int ret
= 0, i_ret
= 0, b_ret
= 0;
1001 struct buffer_head
*di_bh
= NULL
;
1002 struct ocfs2_dinode
*di
= NULL
;
1003 struct ocfs2_inode_info
*oi
= OCFS2_I(dentry
->d_inode
);
1005 if (!ocfs2_supports_xattr(OCFS2_SB(dentry
->d_sb
)))
1008 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1011 ret
= ocfs2_inode_lock(dentry
->d_inode
, &di_bh
, 0);
1017 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1019 down_read(&oi
->ip_xattr_sem
);
1020 i_ret
= ocfs2_xattr_ibody_list(dentry
->d_inode
, di
, buffer
, size
);
1028 b_ret
= ocfs2_xattr_block_list(dentry
->d_inode
, di
,
1033 up_read(&oi
->ip_xattr_sem
);
1034 ocfs2_inode_unlock(dentry
->d_inode
, 0);
1038 return i_ret
+ b_ret
;
1041 static int ocfs2_xattr_find_entry(int name_index
,
1043 struct ocfs2_xattr_search
*xs
)
1045 struct ocfs2_xattr_entry
*entry
;
1052 name_len
= strlen(name
);
1054 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1055 cmp
= name_index
- ocfs2_xattr_get_type(entry
);
1057 cmp
= name_len
- entry
->xe_name_len
;
1059 cmp
= memcmp(name
, (xs
->base
+
1060 le16_to_cpu(entry
->xe_name_offset
)),
1068 return cmp
? -ENODATA
: 0;
1071 static int ocfs2_xattr_get_value_outside(struct inode
*inode
,
1072 struct ocfs2_xattr_value_root
*xv
,
1076 u32 cpos
, p_cluster
, num_clusters
, bpc
, clusters
;
1079 size_t cplen
, blocksize
;
1080 struct buffer_head
*bh
= NULL
;
1081 struct ocfs2_extent_list
*el
;
1084 clusters
= le32_to_cpu(xv
->xr_clusters
);
1085 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1086 blocksize
= inode
->i_sb
->s_blocksize
;
1089 while (cpos
< clusters
) {
1090 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1091 &num_clusters
, el
, NULL
);
1097 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1098 /* Copy ocfs2_xattr_value */
1099 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1100 ret
= ocfs2_read_block(INODE_CACHE(inode
), blkno
,
1107 cplen
= len
>= blocksize
? blocksize
: len
;
1108 memcpy(buffer
, bh
->b_data
, cplen
);
1117 cpos
+= num_clusters
;
1123 static int ocfs2_xattr_ibody_get(struct inode
*inode
,
1128 struct ocfs2_xattr_search
*xs
)
1130 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1131 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1132 struct ocfs2_xattr_value_root
*xv
;
1136 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
1139 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
1140 xs
->header
= (struct ocfs2_xattr_header
*)
1141 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
1142 xs
->base
= (void *)xs
->header
;
1143 xs
->here
= xs
->header
->xh_entries
;
1145 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1148 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1150 if (size
> buffer_size
)
1152 if (ocfs2_xattr_is_local(xs
->here
)) {
1153 memcpy(buffer
, (void *)xs
->base
+
1154 le16_to_cpu(xs
->here
->xe_name_offset
) +
1155 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
), size
);
1157 xv
= (struct ocfs2_xattr_value_root
*)
1158 (xs
->base
+ le16_to_cpu(
1159 xs
->here
->xe_name_offset
) +
1160 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
));
1161 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1173 static int ocfs2_xattr_block_get(struct inode
*inode
,
1178 struct ocfs2_xattr_search
*xs
)
1180 struct ocfs2_xattr_block
*xb
;
1181 struct ocfs2_xattr_value_root
*xv
;
1183 int ret
= -ENODATA
, name_offset
, name_len
, i
;
1184 int uninitialized_var(block_off
);
1186 xs
->bucket
= ocfs2_xattr_bucket_new(inode
);
1193 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, xs
);
1199 if (xs
->not_found
) {
1204 xb
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
1205 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1208 if (size
> buffer_size
)
1211 name_offset
= le16_to_cpu(xs
->here
->xe_name_offset
);
1212 name_len
= OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
);
1213 i
= xs
->here
- xs
->header
->xh_entries
;
1215 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
1216 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
1217 bucket_xh(xs
->bucket
),
1221 xs
->base
= bucket_block(xs
->bucket
, block_off
);
1223 if (ocfs2_xattr_is_local(xs
->here
)) {
1224 memcpy(buffer
, (void *)xs
->base
+
1225 name_offset
+ name_len
, size
);
1227 xv
= (struct ocfs2_xattr_value_root
*)
1228 (xs
->base
+ name_offset
+ name_len
);
1229 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1239 ocfs2_xattr_bucket_free(xs
->bucket
);
1241 brelse(xs
->xattr_bh
);
1242 xs
->xattr_bh
= NULL
;
1246 int ocfs2_xattr_get_nolock(struct inode
*inode
,
1247 struct buffer_head
*di_bh
,
1254 struct ocfs2_dinode
*di
= NULL
;
1255 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1256 struct ocfs2_xattr_search xis
= {
1257 .not_found
= -ENODATA
,
1259 struct ocfs2_xattr_search xbs
= {
1260 .not_found
= -ENODATA
,
1263 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1266 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1269 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
1270 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1272 down_read(&oi
->ip_xattr_sem
);
1273 ret
= ocfs2_xattr_ibody_get(inode
, name_index
, name
, buffer
,
1275 if (ret
== -ENODATA
&& di
->i_xattr_loc
)
1276 ret
= ocfs2_xattr_block_get(inode
, name_index
, name
, buffer
,
1278 up_read(&oi
->ip_xattr_sem
);
1283 /* ocfs2_xattr_get()
1285 * Copy an extended attribute into the buffer provided.
1286 * Buffer is NULL to compute the size of buffer required.
1288 static int ocfs2_xattr_get(struct inode
*inode
,
1295 struct buffer_head
*di_bh
= NULL
;
1297 ret
= ocfs2_inode_lock(inode
, &di_bh
, 0);
1302 ret
= ocfs2_xattr_get_nolock(inode
, di_bh
, name_index
,
1303 name
, buffer
, buffer_size
);
1305 ocfs2_inode_unlock(inode
, 0);
1312 static int __ocfs2_xattr_set_value_outside(struct inode
*inode
,
1314 struct ocfs2_xattr_value_buf
*vb
,
1318 int ret
= 0, i
, cp_len
;
1319 u16 blocksize
= inode
->i_sb
->s_blocksize
;
1320 u32 p_cluster
, num_clusters
;
1321 u32 cpos
= 0, bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1322 u32 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, value_len
);
1324 struct buffer_head
*bh
= NULL
;
1325 unsigned int ext_flags
;
1326 struct ocfs2_xattr_value_root
*xv
= vb
->vb_xv
;
1328 BUG_ON(clusters
> le32_to_cpu(xv
->xr_clusters
));
1330 while (cpos
< clusters
) {
1331 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1332 &num_clusters
, &xv
->xr_list
,
1339 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1341 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1343 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1344 ret
= ocfs2_read_block(INODE_CACHE(inode
), blkno
,
1351 ret
= ocfs2_journal_access(handle
,
1354 OCFS2_JOURNAL_ACCESS_WRITE
);
1360 cp_len
= value_len
> blocksize
? blocksize
: value_len
;
1361 memcpy(bh
->b_data
, value
, cp_len
);
1362 value_len
-= cp_len
;
1364 if (cp_len
< blocksize
)
1365 memset(bh
->b_data
+ cp_len
, 0,
1366 blocksize
- cp_len
);
1368 ocfs2_journal_dirty(handle
, bh
);
1373 * XXX: do we need to empty all the following
1374 * blocks in this cluster?
1379 cpos
+= num_clusters
;
1387 static int ocfs2_xa_check_space_helper(int needed_space
, int free_start
,
1395 free_space
= free_start
-
1396 sizeof(struct ocfs2_xattr_header
) -
1397 (num_entries
* sizeof(struct ocfs2_xattr_entry
)) -
1398 OCFS2_XATTR_HEADER_GAP
;
1401 if (free_space
< needed_space
)
1407 static int ocfs2_xa_journal_access(handle_t
*handle
, struct ocfs2_xa_loc
*loc
,
1410 return loc
->xl_ops
->xlo_journal_access(handle
, loc
, type
);
1413 static void ocfs2_xa_journal_dirty(handle_t
*handle
, struct ocfs2_xa_loc
*loc
)
1415 loc
->xl_ops
->xlo_journal_dirty(handle
, loc
);
1418 /* Give a pointer into the storage for the given offset */
1419 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc
*loc
, int offset
)
1421 BUG_ON(offset
>= loc
->xl_size
);
1422 return loc
->xl_ops
->xlo_offset_pointer(loc
, offset
);
1426 * Wipe the name+value pair and allow the storage to reclaim it. This
1427 * must be followed by either removal of the entry or a call to
1428 * ocfs2_xa_add_namevalue().
1430 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc
*loc
)
1432 loc
->xl_ops
->xlo_wipe_namevalue(loc
);
1436 * Find lowest offset to a name+value pair. This is the start of our
1437 * downward-growing free space.
1439 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc
*loc
)
1441 return loc
->xl_ops
->xlo_get_free_start(loc
);
1444 /* Can we reuse loc->xl_entry for xi? */
1445 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc
*loc
,
1446 struct ocfs2_xattr_info
*xi
)
1448 return loc
->xl_ops
->xlo_can_reuse(loc
, xi
);
1451 /* How much free space is needed to set the new value */
1452 static int ocfs2_xa_check_space(struct ocfs2_xa_loc
*loc
,
1453 struct ocfs2_xattr_info
*xi
)
1455 return loc
->xl_ops
->xlo_check_space(loc
, xi
);
1458 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc
*loc
, u32 name_hash
)
1460 loc
->xl_ops
->xlo_add_entry(loc
, name_hash
);
1461 loc
->xl_entry
->xe_name_hash
= cpu_to_le32(name_hash
);
1463 * We can't leave the new entry's xe_name_offset at zero or
1464 * add_namevalue() will go nuts. We set it to the size of our
1465 * storage so that it can never be less than any other entry.
1467 loc
->xl_entry
->xe_name_offset
= cpu_to_le16(loc
->xl_size
);
1470 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc
*loc
,
1471 struct ocfs2_xattr_info
*xi
)
1473 int size
= namevalue_size_xi(xi
);
1477 loc
->xl_ops
->xlo_add_namevalue(loc
, size
);
1478 loc
->xl_entry
->xe_value_size
= cpu_to_le64(xi
->xi_value_len
);
1479 loc
->xl_entry
->xe_name_len
= xi
->xi_name_len
;
1480 ocfs2_xattr_set_type(loc
->xl_entry
, xi
->xi_name_index
);
1481 ocfs2_xattr_set_local(loc
->xl_entry
,
1482 xi
->xi_value_len
<= OCFS2_XATTR_INLINE_SIZE
);
1484 nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
1485 nameval_buf
= ocfs2_xa_offset_pointer(loc
, nameval_offset
);
1486 memset(nameval_buf
, 0, size
);
1487 memcpy(nameval_buf
, xi
->xi_name
, xi
->xi_name_len
);
1490 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc
*loc
,
1491 struct ocfs2_xattr_value_buf
*vb
)
1493 int nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
1494 int name_size
= OCFS2_XATTR_SIZE(loc
->xl_entry
->xe_name_len
);
1496 /* Value bufs are for value trees */
1497 BUG_ON(ocfs2_xattr_is_local(loc
->xl_entry
));
1498 BUG_ON(namevalue_size_xe(loc
->xl_entry
) !=
1499 (name_size
+ OCFS2_XATTR_ROOT_SIZE
));
1501 loc
->xl_ops
->xlo_fill_value_buf(loc
, vb
);
1503 (struct ocfs2_xattr_value_root
*)ocfs2_xa_offset_pointer(loc
,
1508 static int ocfs2_xa_block_journal_access(handle_t
*handle
,
1509 struct ocfs2_xa_loc
*loc
, int type
)
1511 struct buffer_head
*bh
= loc
->xl_storage
;
1512 ocfs2_journal_access_func access
;
1514 if (loc
->xl_size
== (bh
->b_size
-
1515 offsetof(struct ocfs2_xattr_block
,
1516 xb_attrs
.xb_header
)))
1517 access
= ocfs2_journal_access_xb
;
1519 access
= ocfs2_journal_access_di
;
1520 return access(handle
, INODE_CACHE(loc
->xl_inode
), bh
, type
);
1523 static void ocfs2_xa_block_journal_dirty(handle_t
*handle
,
1524 struct ocfs2_xa_loc
*loc
)
1526 struct buffer_head
*bh
= loc
->xl_storage
;
1528 ocfs2_journal_dirty(handle
, bh
);
1531 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc
*loc
,
1534 return (char *)loc
->xl_header
+ offset
;
1537 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc
*loc
,
1538 struct ocfs2_xattr_info
*xi
)
1541 * Block storage is strict. If the sizes aren't exact, we will
1542 * remove the old one and reinsert the new.
1544 return namevalue_size_xe(loc
->xl_entry
) ==
1545 namevalue_size_xi(xi
);
1548 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc
*loc
)
1550 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1551 int i
, count
= le16_to_cpu(xh
->xh_count
);
1552 int offset
, free_start
= loc
->xl_size
;
1554 for (i
= 0; i
< count
; i
++) {
1555 offset
= le16_to_cpu(xh
->xh_entries
[i
].xe_name_offset
);
1556 if (offset
< free_start
)
1557 free_start
= offset
;
1563 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc
*loc
,
1564 struct ocfs2_xattr_info
*xi
)
1566 int count
= le16_to_cpu(loc
->xl_header
->xh_count
);
1567 int free_start
= ocfs2_xa_get_free_start(loc
);
1568 int needed_space
= ocfs2_xi_entry_usage(xi
);
1571 * Block storage will reclaim the original entry before inserting
1572 * the new value, so we only need the difference. If the new
1573 * entry is smaller than the old one, we don't need anything.
1575 if (loc
->xl_entry
) {
1576 /* Don't need space if we're reusing! */
1577 if (ocfs2_xa_can_reuse_entry(loc
, xi
))
1580 needed_space
-= ocfs2_xe_entry_usage(loc
->xl_entry
);
1582 if (needed_space
< 0)
1584 return ocfs2_xa_check_space_helper(needed_space
, free_start
, count
);
1588 * Block storage for xattrs keeps the name+value pairs compacted. When
1589 * we remove one, we have to shift any that preceded it towards the end.
1591 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc
*loc
)
1594 int namevalue_offset
, first_namevalue_offset
, namevalue_size
;
1595 struct ocfs2_xattr_entry
*entry
= loc
->xl_entry
;
1596 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1597 int count
= le16_to_cpu(xh
->xh_count
);
1599 namevalue_offset
= le16_to_cpu(entry
->xe_name_offset
);
1600 namevalue_size
= namevalue_size_xe(entry
);
1601 first_namevalue_offset
= ocfs2_xa_get_free_start(loc
);
1603 /* Shift the name+value pairs */
1604 memmove((char *)xh
+ first_namevalue_offset
+ namevalue_size
,
1605 (char *)xh
+ first_namevalue_offset
,
1606 namevalue_offset
- first_namevalue_offset
);
1607 memset((char *)xh
+ first_namevalue_offset
, 0, namevalue_size
);
1609 /* Now tell xh->xh_entries about it */
1610 for (i
= 0; i
< count
; i
++) {
1611 offset
= le16_to_cpu(xh
->xh_entries
[i
].xe_name_offset
);
1612 if (offset
<= namevalue_offset
)
1613 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
,
1618 * Note that we don't update xh_free_start or xh_name_value_len
1619 * because they're not used in block-stored xattrs.
1623 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc
*loc
, u32 name_hash
)
1625 int count
= le16_to_cpu(loc
->xl_header
->xh_count
);
1626 loc
->xl_entry
= &(loc
->xl_header
->xh_entries
[count
]);
1627 le16_add_cpu(&loc
->xl_header
->xh_count
, 1);
1628 memset(loc
->xl_entry
, 0, sizeof(struct ocfs2_xattr_entry
));
1631 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc
*loc
, int size
)
1633 int free_start
= ocfs2_xa_get_free_start(loc
);
1635 loc
->xl_entry
->xe_name_offset
= cpu_to_le16(free_start
- size
);
1638 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc
*loc
,
1639 struct ocfs2_xattr_value_buf
*vb
)
1641 struct buffer_head
*bh
= loc
->xl_storage
;
1643 if (loc
->xl_size
== (bh
->b_size
-
1644 offsetof(struct ocfs2_xattr_block
,
1645 xb_attrs
.xb_header
)))
1646 vb
->vb_access
= ocfs2_journal_access_xb
;
1648 vb
->vb_access
= ocfs2_journal_access_di
;
1653 * Operations for xattrs stored in blocks. This includes inline inode
1654 * storage and unindexed ocfs2_xattr_blocks.
1656 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops
= {
1657 .xlo_journal_access
= ocfs2_xa_block_journal_access
,
1658 .xlo_journal_dirty
= ocfs2_xa_block_journal_dirty
,
1659 .xlo_offset_pointer
= ocfs2_xa_block_offset_pointer
,
1660 .xlo_check_space
= ocfs2_xa_block_check_space
,
1661 .xlo_can_reuse
= ocfs2_xa_block_can_reuse
,
1662 .xlo_get_free_start
= ocfs2_xa_block_get_free_start
,
1663 .xlo_wipe_namevalue
= ocfs2_xa_block_wipe_namevalue
,
1664 .xlo_add_entry
= ocfs2_xa_block_add_entry
,
1665 .xlo_add_namevalue
= ocfs2_xa_block_add_namevalue
,
1666 .xlo_fill_value_buf
= ocfs2_xa_block_fill_value_buf
,
1669 static int ocfs2_xa_bucket_journal_access(handle_t
*handle
,
1670 struct ocfs2_xa_loc
*loc
, int type
)
1672 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1674 return ocfs2_xattr_bucket_journal_access(handle
, bucket
, type
);
1677 static void ocfs2_xa_bucket_journal_dirty(handle_t
*handle
,
1678 struct ocfs2_xa_loc
*loc
)
1680 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1682 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
1685 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc
*loc
,
1688 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1689 int block
, block_offset
;
1691 /* The header is at the front of the bucket */
1692 block
= offset
>> loc
->xl_inode
->i_sb
->s_blocksize_bits
;
1693 block_offset
= offset
% loc
->xl_inode
->i_sb
->s_blocksize
;
1695 return bucket_block(bucket
, block
) + block_offset
;
1698 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc
*loc
,
1699 struct ocfs2_xattr_info
*xi
)
1701 return namevalue_size_xe(loc
->xl_entry
) >=
1702 namevalue_size_xi(xi
);
1705 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc
*loc
)
1707 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1708 return le16_to_cpu(bucket_xh(bucket
)->xh_free_start
);
1711 static int ocfs2_bucket_align_free_start(struct super_block
*sb
,
1712 int free_start
, int size
)
1715 * We need to make sure that the name+value pair fits within
1718 if (((free_start
- size
) >> sb
->s_blocksize_bits
) !=
1719 ((free_start
- 1) >> sb
->s_blocksize_bits
))
1720 free_start
-= free_start
% sb
->s_blocksize
;
1725 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc
*loc
,
1726 struct ocfs2_xattr_info
*xi
)
1729 int count
= le16_to_cpu(loc
->xl_header
->xh_count
);
1730 int free_start
= ocfs2_xa_get_free_start(loc
);
1731 int needed_space
= ocfs2_xi_entry_usage(xi
);
1732 int size
= namevalue_size_xi(xi
);
1733 struct super_block
*sb
= loc
->xl_inode
->i_sb
;
1736 * Bucket storage does not reclaim name+value pairs it cannot
1737 * reuse. They live as holes until the bucket fills, and then
1738 * the bucket is defragmented. However, the bucket can reclaim
1739 * the ocfs2_xattr_entry.
1741 if (loc
->xl_entry
) {
1742 /* Don't need space if we're reusing! */
1743 if (ocfs2_xa_can_reuse_entry(loc
, xi
))
1746 needed_space
-= sizeof(struct ocfs2_xattr_entry
);
1748 BUG_ON(needed_space
< 0);
1750 if (free_start
< size
) {
1755 * First we check if it would fit in the first place.
1756 * Below, we align the free start to a block. This may
1757 * slide us below the minimum gap. By checking unaligned
1758 * first, we avoid that error.
1760 rc
= ocfs2_xa_check_space_helper(needed_space
, free_start
,
1764 free_start
= ocfs2_bucket_align_free_start(sb
, free_start
,
1767 return ocfs2_xa_check_space_helper(needed_space
, free_start
, count
);
1770 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc
*loc
)
1772 le16_add_cpu(&loc
->xl_header
->xh_name_value_len
,
1773 -namevalue_size_xe(loc
->xl_entry
));
1776 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc
*loc
, u32 name_hash
)
1778 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1779 int count
= le16_to_cpu(xh
->xh_count
);
1780 int low
= 0, high
= count
- 1, tmp
;
1781 struct ocfs2_xattr_entry
*tmp_xe
;
1784 * We keep buckets sorted by name_hash, so we need to find
1787 while (low
<= high
&& count
) {
1788 tmp
= (low
+ high
) / 2;
1789 tmp_xe
= &xh
->xh_entries
[tmp
];
1791 if (name_hash
> le32_to_cpu(tmp_xe
->xe_name_hash
))
1793 else if (name_hash
< le32_to_cpu(tmp_xe
->xe_name_hash
))
1802 memmove(&xh
->xh_entries
[low
+ 1],
1803 &xh
->xh_entries
[low
],
1804 ((count
- low
) * sizeof(struct ocfs2_xattr_entry
)));
1806 le16_add_cpu(&xh
->xh_count
, 1);
1807 loc
->xl_entry
= &xh
->xh_entries
[low
];
1808 memset(loc
->xl_entry
, 0, sizeof(struct ocfs2_xattr_entry
));
1811 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc
*loc
, int size
)
1813 int free_start
= ocfs2_xa_get_free_start(loc
);
1814 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1815 struct super_block
*sb
= loc
->xl_inode
->i_sb
;
1818 free_start
= ocfs2_bucket_align_free_start(sb
, free_start
, size
);
1819 nameval_offset
= free_start
- size
;
1820 loc
->xl_entry
->xe_name_offset
= cpu_to_le16(nameval_offset
);
1821 xh
->xh_free_start
= cpu_to_le16(nameval_offset
);
1822 le16_add_cpu(&xh
->xh_name_value_len
, size
);
1826 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc
*loc
,
1827 struct ocfs2_xattr_value_buf
*vb
)
1829 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1830 struct super_block
*sb
= loc
->xl_inode
->i_sb
;
1831 int nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
1832 int size
= namevalue_size_xe(loc
->xl_entry
);
1833 int block_offset
= nameval_offset
>> sb
->s_blocksize_bits
;
1835 /* Values are not allowed to straddle block boundaries */
1836 BUG_ON(block_offset
!=
1837 ((nameval_offset
+ size
- 1) >> sb
->s_blocksize_bits
));
1838 /* We expect the bucket to be filled in */
1839 BUG_ON(!bucket
->bu_bhs
[block_offset
]);
1841 vb
->vb_access
= ocfs2_journal_access
;
1842 vb
->vb_bh
= bucket
->bu_bhs
[block_offset
];
1845 /* Operations for xattrs stored in buckets. */
1846 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops
= {
1847 .xlo_journal_access
= ocfs2_xa_bucket_journal_access
,
1848 .xlo_journal_dirty
= ocfs2_xa_bucket_journal_dirty
,
1849 .xlo_offset_pointer
= ocfs2_xa_bucket_offset_pointer
,
1850 .xlo_check_space
= ocfs2_xa_bucket_check_space
,
1851 .xlo_can_reuse
= ocfs2_xa_bucket_can_reuse
,
1852 .xlo_get_free_start
= ocfs2_xa_bucket_get_free_start
,
1853 .xlo_wipe_namevalue
= ocfs2_xa_bucket_wipe_namevalue
,
1854 .xlo_add_entry
= ocfs2_xa_bucket_add_entry
,
1855 .xlo_add_namevalue
= ocfs2_xa_bucket_add_namevalue
,
1856 .xlo_fill_value_buf
= ocfs2_xa_bucket_fill_value_buf
,
1859 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc
*loc
)
1861 struct ocfs2_xattr_value_buf vb
;
1863 if (ocfs2_xattr_is_local(loc
->xl_entry
))
1866 ocfs2_xa_fill_value_buf(loc
, &vb
);
1867 return le32_to_cpu(vb
.vb_xv
->xr_clusters
);
1870 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc
*loc
, u64 bytes
,
1871 struct ocfs2_xattr_set_ctxt
*ctxt
)
1873 int trunc_rc
, access_rc
;
1874 struct ocfs2_xattr_value_buf vb
;
1876 ocfs2_xa_fill_value_buf(loc
, &vb
);
1877 trunc_rc
= ocfs2_xattr_value_truncate(loc
->xl_inode
, &vb
, bytes
,
1881 * The caller of ocfs2_xa_value_truncate() has already called
1882 * ocfs2_xa_journal_access on the loc. However, The truncate code
1883 * calls ocfs2_extend_trans(). This may commit the previous
1884 * transaction and open a new one. If this is a bucket, truncate
1885 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1886 * the caller is expecting to dirty the entire bucket. So we must
1887 * reset the journal work. We do this even if truncate has failed,
1888 * as it could have failed after committing the extend.
1890 access_rc
= ocfs2_xa_journal_access(ctxt
->handle
, loc
,
1891 OCFS2_JOURNAL_ACCESS_WRITE
);
1893 /* Errors in truncate take precedence */
1894 return trunc_rc
? trunc_rc
: access_rc
;
1897 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc
*loc
)
1900 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1901 struct ocfs2_xattr_entry
*entry
= loc
->xl_entry
;
1903 ocfs2_xa_wipe_namevalue(loc
);
1904 loc
->xl_entry
= NULL
;
1906 le16_add_cpu(&xh
->xh_count
, -1);
1907 count
= le16_to_cpu(xh
->xh_count
);
1910 * Only zero out the entry if there are more remaining. This is
1911 * important for an empty bucket, as it keeps track of the
1912 * bucket's hash value. It doesn't hurt empty block storage.
1915 index
= ((char *)entry
- (char *)&xh
->xh_entries
) /
1916 sizeof(struct ocfs2_xattr_entry
);
1917 memmove(&xh
->xh_entries
[index
], &xh
->xh_entries
[index
+ 1],
1918 (count
- index
) * sizeof(struct ocfs2_xattr_entry
));
1919 memset(&xh
->xh_entries
[count
], 0,
1920 sizeof(struct ocfs2_xattr_entry
));
1925 * If we have a problem adjusting the size of an external value during
1926 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1927 * in an intermediate state. For example, the value may be partially
1930 * If the value tree hasn't changed, the extend/truncate went nowhere.
1931 * We have nothing to do. The caller can treat it as a straight error.
1933 * If the value tree got partially truncated, we now have a corrupted
1934 * extended attribute. We're going to wipe its entry and leak the
1935 * clusters. Better to leak some storage than leave a corrupt entry.
1937 * If the value tree grew, it obviously didn't grow enough for the
1938 * new entry. We're not going to try and reclaim those clusters either.
1939 * If there was already an external value there (orig_clusters != 0),
1940 * the new clusters are attached safely and we can just leave the old
1941 * value in place. If there was no external value there, we remove
1944 * This way, the xattr block we store in the journal will be consistent.
1945 * If the size change broke because of the journal, no changes will hit
1948 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc
*loc
,
1950 unsigned int orig_clusters
)
1952 unsigned int new_clusters
= ocfs2_xa_value_clusters(loc
);
1953 char *nameval_buf
= ocfs2_xa_offset_pointer(loc
,
1954 le16_to_cpu(loc
->xl_entry
->xe_name_offset
));
1956 if (new_clusters
< orig_clusters
) {
1958 "Partial truncate while %s xattr %.*s. Leaking "
1959 "%u clusters and removing the entry\n",
1960 what
, loc
->xl_entry
->xe_name_len
, nameval_buf
,
1961 orig_clusters
- new_clusters
);
1962 ocfs2_xa_remove_entry(loc
);
1963 } else if (!orig_clusters
) {
1965 "Unable to allocate an external value for xattr "
1966 "%.*s safely. Leaking %u clusters and removing the "
1968 loc
->xl_entry
->xe_name_len
, nameval_buf
,
1969 new_clusters
- orig_clusters
);
1970 ocfs2_xa_remove_entry(loc
);
1971 } else if (new_clusters
> orig_clusters
)
1973 "Unable to grow xattr %.*s safely. %u new clusters "
1974 "have been added, but the value will not be "
1976 loc
->xl_entry
->xe_name_len
, nameval_buf
,
1977 new_clusters
- orig_clusters
);
1980 static int ocfs2_xa_remove(struct ocfs2_xa_loc
*loc
,
1981 struct ocfs2_xattr_set_ctxt
*ctxt
)
1984 unsigned int orig_clusters
;
1986 if (!ocfs2_xattr_is_local(loc
->xl_entry
)) {
1987 orig_clusters
= ocfs2_xa_value_clusters(loc
);
1988 rc
= ocfs2_xa_value_truncate(loc
, 0, ctxt
);
1992 * Since this is remove, we can return 0 if
1993 * ocfs2_xa_cleanup_value_truncate() is going to
1994 * wipe the entry anyway. So we check the
1995 * cluster count as well.
1997 if (orig_clusters
!= ocfs2_xa_value_clusters(loc
))
1999 ocfs2_xa_cleanup_value_truncate(loc
, "removing",
2006 ocfs2_xa_remove_entry(loc
);
2012 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc
*loc
)
2014 int name_size
= OCFS2_XATTR_SIZE(loc
->xl_entry
->xe_name_len
);
2017 nameval_buf
= ocfs2_xa_offset_pointer(loc
,
2018 le16_to_cpu(loc
->xl_entry
->xe_name_offset
));
2019 memcpy(nameval_buf
+ name_size
, &def_xv
, OCFS2_XATTR_ROOT_SIZE
);
2023 * Take an existing entry and make it ready for the new value. This
2024 * won't allocate space, but it may free space. It should be ready for
2025 * ocfs2_xa_prepare_entry() to finish the work.
2027 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc
*loc
,
2028 struct ocfs2_xattr_info
*xi
,
2029 struct ocfs2_xattr_set_ctxt
*ctxt
)
2032 int name_size
= OCFS2_XATTR_SIZE(xi
->xi_name_len
);
2033 unsigned int orig_clusters
;
2035 int xe_local
= ocfs2_xattr_is_local(loc
->xl_entry
);
2036 int xi_local
= xi
->xi_value_len
<= OCFS2_XATTR_INLINE_SIZE
;
2038 BUG_ON(OCFS2_XATTR_SIZE(loc
->xl_entry
->xe_name_len
) !=
2041 nameval_buf
= ocfs2_xa_offset_pointer(loc
,
2042 le16_to_cpu(loc
->xl_entry
->xe_name_offset
));
2044 memset(nameval_buf
+ name_size
, 0,
2045 namevalue_size_xe(loc
->xl_entry
) - name_size
);
2047 ocfs2_xa_install_value_root(loc
);
2049 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2051 rc
= ocfs2_xa_value_truncate(loc
, 0, ctxt
);
2055 memset(nameval_buf
+ name_size
, 0,
2056 namevalue_size_xe(loc
->xl_entry
) -
2058 } else if (le64_to_cpu(loc
->xl_entry
->xe_value_size
) >
2060 rc
= ocfs2_xa_value_truncate(loc
, xi
->xi_value_len
,
2067 ocfs2_xa_cleanup_value_truncate(loc
, "reusing",
2073 loc
->xl_entry
->xe_value_size
= cpu_to_le64(xi
->xi_value_len
);
2074 ocfs2_xattr_set_local(loc
->xl_entry
, xi_local
);
2081 * Prepares loc->xl_entry to receive the new xattr. This includes
2082 * properly setting up the name+value pair region. If loc->xl_entry
2083 * already exists, it will take care of modifying it appropriately.
2085 * Note that this modifies the data. You did journal_access already,
2088 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc
*loc
,
2089 struct ocfs2_xattr_info
*xi
,
2091 struct ocfs2_xattr_set_ctxt
*ctxt
)
2094 unsigned int orig_clusters
;
2095 __le64 orig_value_size
= 0;
2097 rc
= ocfs2_xa_check_space(loc
, xi
);
2101 if (loc
->xl_entry
) {
2102 if (ocfs2_xa_can_reuse_entry(loc
, xi
)) {
2103 orig_value_size
= loc
->xl_entry
->xe_value_size
;
2104 rc
= ocfs2_xa_reuse_entry(loc
, xi
, ctxt
);
2110 if (!ocfs2_xattr_is_local(loc
->xl_entry
)) {
2111 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2112 rc
= ocfs2_xa_value_truncate(loc
, 0, ctxt
);
2115 ocfs2_xa_cleanup_value_truncate(loc
,
2121 ocfs2_xa_wipe_namevalue(loc
);
2123 ocfs2_xa_add_entry(loc
, name_hash
);
2126 * If we get here, we have a blank entry. Fill it. We grow our
2127 * name+value pair back from the end.
2129 ocfs2_xa_add_namevalue(loc
, xi
);
2130 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
)
2131 ocfs2_xa_install_value_root(loc
);
2134 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2135 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2136 rc
= ocfs2_xa_value_truncate(loc
, xi
->xi_value_len
, ctxt
);
2139 * If we tried to grow an existing external value,
2140 * ocfs2_xa_cleanuP-value_truncate() is going to
2141 * let it stand. We have to restore its original
2144 loc
->xl_entry
->xe_value_size
= orig_value_size
;
2145 ocfs2_xa_cleanup_value_truncate(loc
, "growing",
2156 * Store the value portion of the name+value pair. This will skip
2157 * values that are stored externally. Their tree roots were set up
2158 * by ocfs2_xa_prepare_entry().
2160 static int ocfs2_xa_store_value(struct ocfs2_xa_loc
*loc
,
2161 struct ocfs2_xattr_info
*xi
,
2162 struct ocfs2_xattr_set_ctxt
*ctxt
)
2165 int nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
2166 int name_size
= OCFS2_XATTR_SIZE(xi
->xi_name_len
);
2168 struct ocfs2_xattr_value_buf vb
;
2170 nameval_buf
= ocfs2_xa_offset_pointer(loc
, nameval_offset
);
2171 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2172 ocfs2_xa_fill_value_buf(loc
, &vb
);
2173 rc
= __ocfs2_xattr_set_value_outside(loc
->xl_inode
,
2178 memcpy(nameval_buf
+ name_size
, xi
->xi_value
, xi
->xi_value_len
);
2183 static int ocfs2_xa_set(struct ocfs2_xa_loc
*loc
,
2184 struct ocfs2_xattr_info
*xi
,
2185 struct ocfs2_xattr_set_ctxt
*ctxt
)
2188 u32 name_hash
= ocfs2_xattr_name_hash(loc
->xl_inode
, xi
->xi_name
,
2191 ret
= ocfs2_xa_journal_access(ctxt
->handle
, loc
,
2192 OCFS2_JOURNAL_ACCESS_WRITE
);
2199 * From here on out, everything is going to modify the buffer a
2200 * little. Errors are going to leave the xattr header in a
2201 * sane state. Thus, even with errors we dirty the sucker.
2204 /* Don't worry, we are never called with !xi_value and !xl_entry */
2205 if (!xi
->xi_value
) {
2206 ret
= ocfs2_xa_remove(loc
, ctxt
);
2210 ret
= ocfs2_xa_prepare_entry(loc
, xi
, name_hash
, ctxt
);
2217 ret
= ocfs2_xa_store_value(loc
, xi
, ctxt
);
2222 ocfs2_xa_journal_dirty(ctxt
->handle
, loc
);
2228 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc
*loc
,
2229 struct inode
*inode
,
2230 struct buffer_head
*bh
,
2231 struct ocfs2_xattr_entry
*entry
)
2233 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
2235 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
));
2237 loc
->xl_inode
= inode
;
2238 loc
->xl_ops
= &ocfs2_xa_block_loc_ops
;
2239 loc
->xl_storage
= bh
;
2240 loc
->xl_entry
= entry
;
2241 loc
->xl_size
= le16_to_cpu(di
->i_xattr_inline_size
);
2243 (struct ocfs2_xattr_header
*)(bh
->b_data
+ bh
->b_size
-
2247 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc
*loc
,
2248 struct inode
*inode
,
2249 struct buffer_head
*bh
,
2250 struct ocfs2_xattr_entry
*entry
)
2252 struct ocfs2_xattr_block
*xb
=
2253 (struct ocfs2_xattr_block
*)bh
->b_data
;
2255 BUG_ON(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
);
2257 loc
->xl_inode
= inode
;
2258 loc
->xl_ops
= &ocfs2_xa_block_loc_ops
;
2259 loc
->xl_storage
= bh
;
2260 loc
->xl_header
= &(xb
->xb_attrs
.xb_header
);
2261 loc
->xl_entry
= entry
;
2262 loc
->xl_size
= bh
->b_size
- offsetof(struct ocfs2_xattr_block
,
2263 xb_attrs
.xb_header
);
2266 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc
*loc
,
2267 struct ocfs2_xattr_bucket
*bucket
,
2268 struct ocfs2_xattr_entry
*entry
)
2270 loc
->xl_inode
= bucket
->bu_inode
;
2271 loc
->xl_ops
= &ocfs2_xa_bucket_loc_ops
;
2272 loc
->xl_storage
= bucket
;
2273 loc
->xl_header
= bucket_xh(bucket
);
2274 loc
->xl_entry
= entry
;
2275 loc
->xl_size
= OCFS2_XATTR_BUCKET_SIZE
;
2279 * In xattr remove, if it is stored outside and refcounted, we may have
2280 * the chance to split the refcount tree. So need the allocators.
2282 static int ocfs2_lock_xattr_remove_allocators(struct inode
*inode
,
2283 struct ocfs2_xattr_value_root
*xv
,
2284 struct ocfs2_caching_info
*ref_ci
,
2285 struct buffer_head
*ref_root_bh
,
2286 struct ocfs2_alloc_context
**meta_ac
,
2289 int ret
, meta_add
= 0;
2290 u32 p_cluster
, num_clusters
;
2291 unsigned int ext_flags
;
2294 ret
= ocfs2_xattr_get_clusters(inode
, 0, &p_cluster
,
2303 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
2306 ret
= ocfs2_refcounted_xattr_delete_need(inode
, ref_ci
,
2308 &meta_add
, ref_credits
);
2314 ret
= ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode
->i_sb
),
2323 static int ocfs2_remove_value_outside(struct inode
*inode
,
2324 struct ocfs2_xattr_value_buf
*vb
,
2325 struct ocfs2_xattr_header
*header
,
2326 struct ocfs2_caching_info
*ref_ci
,
2327 struct buffer_head
*ref_root_bh
)
2329 int ret
= 0, i
, ref_credits
;
2330 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2331 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
2334 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
2336 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
2337 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
2339 if (ocfs2_xattr_is_local(entry
))
2342 val
= (void *)header
+
2343 le16_to_cpu(entry
->xe_name_offset
);
2344 vb
->vb_xv
= (struct ocfs2_xattr_value_root
*)
2345 (val
+ OCFS2_XATTR_SIZE(entry
->xe_name_len
));
2347 ret
= ocfs2_lock_xattr_remove_allocators(inode
, vb
->vb_xv
,
2348 ref_ci
, ref_root_bh
,
2352 ctxt
.handle
= ocfs2_start_trans(osb
, ref_credits
+
2353 ocfs2_remove_extent_credits(osb
->sb
));
2354 if (IS_ERR(ctxt
.handle
)) {
2355 ret
= PTR_ERR(ctxt
.handle
);
2360 ret
= ocfs2_xattr_value_truncate(inode
, vb
, 0, &ctxt
);
2366 ocfs2_commit_trans(osb
, ctxt
.handle
);
2368 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2369 ctxt
.meta_ac
= NULL
;
2374 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2375 ocfs2_schedule_truncate_log_flush(osb
, 1);
2376 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
2380 static int ocfs2_xattr_ibody_remove(struct inode
*inode
,
2381 struct buffer_head
*di_bh
,
2382 struct ocfs2_caching_info
*ref_ci
,
2383 struct buffer_head
*ref_root_bh
)
2386 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2387 struct ocfs2_xattr_header
*header
;
2389 struct ocfs2_xattr_value_buf vb
= {
2391 .vb_access
= ocfs2_journal_access_di
,
2394 header
= (struct ocfs2_xattr_header
*)
2395 ((void *)di
+ inode
->i_sb
->s_blocksize
-
2396 le16_to_cpu(di
->i_xattr_inline_size
));
2398 ret
= ocfs2_remove_value_outside(inode
, &vb
, header
,
2399 ref_ci
, ref_root_bh
);
2404 struct ocfs2_rm_xattr_bucket_para
{
2405 struct ocfs2_caching_info
*ref_ci
;
2406 struct buffer_head
*ref_root_bh
;
2409 static int ocfs2_xattr_block_remove(struct inode
*inode
,
2410 struct buffer_head
*blk_bh
,
2411 struct ocfs2_caching_info
*ref_ci
,
2412 struct buffer_head
*ref_root_bh
)
2414 struct ocfs2_xattr_block
*xb
;
2416 struct ocfs2_xattr_value_buf vb
= {
2418 .vb_access
= ocfs2_journal_access_xb
,
2420 struct ocfs2_rm_xattr_bucket_para args
= {
2422 .ref_root_bh
= ref_root_bh
,
2425 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2426 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2427 struct ocfs2_xattr_header
*header
= &(xb
->xb_attrs
.xb_header
);
2428 ret
= ocfs2_remove_value_outside(inode
, &vb
, header
,
2429 ref_ci
, ref_root_bh
);
2431 ret
= ocfs2_iterate_xattr_index_block(inode
,
2433 ocfs2_rm_xattr_cluster
,
2439 static int ocfs2_xattr_free_block(struct inode
*inode
,
2441 struct ocfs2_caching_info
*ref_ci
,
2442 struct buffer_head
*ref_root_bh
)
2444 struct inode
*xb_alloc_inode
;
2445 struct buffer_head
*xb_alloc_bh
= NULL
;
2446 struct buffer_head
*blk_bh
= NULL
;
2447 struct ocfs2_xattr_block
*xb
;
2448 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2454 ret
= ocfs2_read_xattr_block(inode
, block
, &blk_bh
);
2460 ret
= ocfs2_xattr_block_remove(inode
, blk_bh
, ref_ci
, ref_root_bh
);
2466 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2467 blk
= le64_to_cpu(xb
->xb_blkno
);
2468 bit
= le16_to_cpu(xb
->xb_suballoc_bit
);
2469 if (xb
->xb_suballoc_loc
)
2470 bg_blkno
= le64_to_cpu(xb
->xb_suballoc_loc
);
2472 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
2474 xb_alloc_inode
= ocfs2_get_system_file_inode(osb
,
2475 EXTENT_ALLOC_SYSTEM_INODE
,
2476 le16_to_cpu(xb
->xb_suballoc_slot
));
2477 if (!xb_alloc_inode
) {
2482 mutex_lock(&xb_alloc_inode
->i_mutex
);
2484 ret
= ocfs2_inode_lock(xb_alloc_inode
, &xb_alloc_bh
, 1);
2490 handle
= ocfs2_start_trans(osb
, OCFS2_SUBALLOC_FREE
);
2491 if (IS_ERR(handle
)) {
2492 ret
= PTR_ERR(handle
);
2497 ret
= ocfs2_free_suballoc_bits(handle
, xb_alloc_inode
, xb_alloc_bh
,
2502 ocfs2_commit_trans(osb
, handle
);
2504 ocfs2_inode_unlock(xb_alloc_inode
, 1);
2505 brelse(xb_alloc_bh
);
2507 mutex_unlock(&xb_alloc_inode
->i_mutex
);
2508 iput(xb_alloc_inode
);
2515 * ocfs2_xattr_remove()
2517 * Free extended attribute resources associated with this inode.
2519 int ocfs2_xattr_remove(struct inode
*inode
, struct buffer_head
*di_bh
)
2521 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2522 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2523 struct ocfs2_refcount_tree
*ref_tree
= NULL
;
2524 struct buffer_head
*ref_root_bh
= NULL
;
2525 struct ocfs2_caching_info
*ref_ci
= NULL
;
2529 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2532 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
2535 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
) {
2536 ret
= ocfs2_lock_refcount_tree(OCFS2_SB(inode
->i_sb
),
2537 le64_to_cpu(di
->i_refcount_loc
),
2538 1, &ref_tree
, &ref_root_bh
);
2543 ref_ci
= &ref_tree
->rf_ci
;
2547 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
2548 ret
= ocfs2_xattr_ibody_remove(inode
, di_bh
,
2549 ref_ci
, ref_root_bh
);
2556 if (di
->i_xattr_loc
) {
2557 ret
= ocfs2_xattr_free_block(inode
,
2558 le64_to_cpu(di
->i_xattr_loc
),
2559 ref_ci
, ref_root_bh
);
2566 handle
= ocfs2_start_trans((OCFS2_SB(inode
->i_sb
)),
2567 OCFS2_INODE_UPDATE_CREDITS
);
2568 if (IS_ERR(handle
)) {
2569 ret
= PTR_ERR(handle
);
2573 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
2574 OCFS2_JOURNAL_ACCESS_WRITE
);
2580 di
->i_xattr_loc
= 0;
2582 spin_lock(&oi
->ip_lock
);
2583 oi
->ip_dyn_features
&= ~(OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
);
2584 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
2585 spin_unlock(&oi
->ip_lock
);
2587 ocfs2_journal_dirty(handle
, di_bh
);
2589 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
2592 ocfs2_unlock_refcount_tree(OCFS2_SB(inode
->i_sb
), ref_tree
, 1);
2593 brelse(ref_root_bh
);
2597 static int ocfs2_xattr_has_space_inline(struct inode
*inode
,
2598 struct ocfs2_dinode
*di
)
2600 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2601 unsigned int xattrsize
= OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
;
2604 if (xattrsize
< OCFS2_MIN_XATTR_INLINE_SIZE
)
2607 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2608 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
2609 free
= le16_to_cpu(idata
->id_count
) - le64_to_cpu(di
->i_size
);
2610 } else if (ocfs2_inode_is_fast_symlink(inode
)) {
2611 free
= ocfs2_fast_symlink_chars(inode
->i_sb
) -
2612 le64_to_cpu(di
->i_size
);
2614 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
2615 free
= (le16_to_cpu(el
->l_count
) -
2616 le16_to_cpu(el
->l_next_free_rec
)) *
2617 sizeof(struct ocfs2_extent_rec
);
2619 if (free
>= xattrsize
)
2626 * ocfs2_xattr_ibody_find()
2628 * Find extended attribute in inode block and
2629 * fill search info into struct ocfs2_xattr_search.
2631 static int ocfs2_xattr_ibody_find(struct inode
*inode
,
2634 struct ocfs2_xattr_search
*xs
)
2636 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2637 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2641 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
2644 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
2645 down_read(&oi
->ip_alloc_sem
);
2646 has_space
= ocfs2_xattr_has_space_inline(inode
, di
);
2647 up_read(&oi
->ip_alloc_sem
);
2652 xs
->xattr_bh
= xs
->inode_bh
;
2653 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
2654 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)
2655 xs
->header
= (struct ocfs2_xattr_header
*)
2656 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
2658 xs
->header
= (struct ocfs2_xattr_header
*)
2659 (xs
->end
- OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
);
2660 xs
->base
= (void *)xs
->header
;
2661 xs
->here
= xs
->header
->xh_entries
;
2663 /* Find the named attribute. */
2664 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
2665 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
2666 if (ret
&& ret
!= -ENODATA
)
2668 xs
->not_found
= ret
;
2674 static int ocfs2_xattr_ibody_init(struct inode
*inode
,
2675 struct buffer_head
*di_bh
,
2676 struct ocfs2_xattr_set_ctxt
*ctxt
)
2679 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2680 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2681 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2682 unsigned int xattrsize
= osb
->s_xattr_inline_size
;
2684 if (!ocfs2_xattr_has_space_inline(inode
, di
)) {
2689 ret
= ocfs2_journal_access_di(ctxt
->handle
, INODE_CACHE(inode
), di_bh
,
2690 OCFS2_JOURNAL_ACCESS_WRITE
);
2697 * Adjust extent record count or inline data size
2698 * to reserve space for extended attribute.
2700 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2701 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
2702 le16_add_cpu(&idata
->id_count
, -xattrsize
);
2703 } else if (!(ocfs2_inode_is_fast_symlink(inode
))) {
2704 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
2705 le16_add_cpu(&el
->l_count
, -(xattrsize
/
2706 sizeof(struct ocfs2_extent_rec
)));
2708 di
->i_xattr_inline_size
= cpu_to_le16(xattrsize
);
2710 spin_lock(&oi
->ip_lock
);
2711 oi
->ip_dyn_features
|= OCFS2_INLINE_XATTR_FL
|OCFS2_HAS_XATTR_FL
;
2712 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
2713 spin_unlock(&oi
->ip_lock
);
2715 ocfs2_journal_dirty(ctxt
->handle
, di_bh
);
2722 * ocfs2_xattr_ibody_set()
2724 * Set, replace or remove an extended attribute into inode block.
2727 static int ocfs2_xattr_ibody_set(struct inode
*inode
,
2728 struct ocfs2_xattr_info
*xi
,
2729 struct ocfs2_xattr_search
*xs
,
2730 struct ocfs2_xattr_set_ctxt
*ctxt
)
2733 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2734 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2735 struct ocfs2_xa_loc loc
;
2737 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
2740 down_write(&oi
->ip_alloc_sem
);
2741 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
2742 if (!ocfs2_xattr_has_space_inline(inode
, di
)) {
2748 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
2749 ret
= ocfs2_xattr_ibody_init(inode
, xs
->inode_bh
, ctxt
);
2757 ocfs2_init_dinode_xa_loc(&loc
, inode
, xs
->inode_bh
,
2758 xs
->not_found
? NULL
: xs
->here
);
2759 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
2765 xs
->here
= loc
.xl_entry
;
2768 up_write(&oi
->ip_alloc_sem
);
2774 * ocfs2_xattr_block_find()
2776 * Find extended attribute in external block and
2777 * fill search info into struct ocfs2_xattr_search.
2779 static int ocfs2_xattr_block_find(struct inode
*inode
,
2782 struct ocfs2_xattr_search
*xs
)
2784 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2785 struct buffer_head
*blk_bh
= NULL
;
2786 struct ocfs2_xattr_block
*xb
;
2789 if (!di
->i_xattr_loc
)
2792 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
2799 xs
->xattr_bh
= blk_bh
;
2800 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2802 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2803 xs
->header
= &xb
->xb_attrs
.xb_header
;
2804 xs
->base
= (void *)xs
->header
;
2805 xs
->end
= (void *)(blk_bh
->b_data
) + blk_bh
->b_size
;
2806 xs
->here
= xs
->header
->xh_entries
;
2808 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
2810 ret
= ocfs2_xattr_index_block_find(inode
, blk_bh
,
2814 if (ret
&& ret
!= -ENODATA
) {
2815 xs
->xattr_bh
= NULL
;
2818 xs
->not_found
= ret
;
2826 static int ocfs2_create_xattr_block(struct inode
*inode
,
2827 struct buffer_head
*inode_bh
,
2828 struct ocfs2_xattr_set_ctxt
*ctxt
,
2830 struct buffer_head
**ret_bh
)
2833 u16 suballoc_bit_start
;
2835 u64 suballoc_loc
, first_blkno
;
2836 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)inode_bh
->b_data
;
2837 struct buffer_head
*new_bh
= NULL
;
2838 struct ocfs2_xattr_block
*xblk
;
2840 ret
= ocfs2_journal_access_di(ctxt
->handle
, INODE_CACHE(inode
),
2841 inode_bh
, OCFS2_JOURNAL_ACCESS_CREATE
);
2847 ret
= ocfs2_claim_metadata(ctxt
->handle
, ctxt
->meta_ac
, 1,
2848 &suballoc_loc
, &suballoc_bit_start
,
2849 &num_got
, &first_blkno
);
2855 new_bh
= sb_getblk(inode
->i_sb
, first_blkno
);
2856 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), new_bh
);
2858 ret
= ocfs2_journal_access_xb(ctxt
->handle
, INODE_CACHE(inode
),
2860 OCFS2_JOURNAL_ACCESS_CREATE
);
2866 /* Initialize ocfs2_xattr_block */
2867 xblk
= (struct ocfs2_xattr_block
*)new_bh
->b_data
;
2868 memset(xblk
, 0, inode
->i_sb
->s_blocksize
);
2869 strcpy((void *)xblk
, OCFS2_XATTR_BLOCK_SIGNATURE
);
2870 xblk
->xb_suballoc_slot
= cpu_to_le16(ctxt
->meta_ac
->ac_alloc_slot
);
2871 xblk
->xb_suballoc_loc
= cpu_to_le64(suballoc_loc
);
2872 xblk
->xb_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
2873 xblk
->xb_fs_generation
=
2874 cpu_to_le32(OCFS2_SB(inode
->i_sb
)->fs_generation
);
2875 xblk
->xb_blkno
= cpu_to_le64(first_blkno
);
2877 struct ocfs2_xattr_tree_root
*xr
= &xblk
->xb_attrs
.xb_root
;
2878 xr
->xt_clusters
= cpu_to_le32(1);
2879 xr
->xt_last_eb_blk
= 0;
2880 xr
->xt_list
.l_tree_depth
= 0;
2881 xr
->xt_list
.l_count
= cpu_to_le16(
2882 ocfs2_xattr_recs_per_xb(inode
->i_sb
));
2883 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
2884 xblk
->xb_flags
= cpu_to_le16(OCFS2_XATTR_INDEXED
);
2886 ocfs2_journal_dirty(ctxt
->handle
, new_bh
);
2888 /* Add it to the inode */
2889 di
->i_xattr_loc
= cpu_to_le64(first_blkno
);
2891 spin_lock(&OCFS2_I(inode
)->ip_lock
);
2892 OCFS2_I(inode
)->ip_dyn_features
|= OCFS2_HAS_XATTR_FL
;
2893 di
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
2894 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
2896 ocfs2_journal_dirty(ctxt
->handle
, inode_bh
);
2907 * ocfs2_xattr_block_set()
2909 * Set, replace or remove an extended attribute into external block.
2912 static int ocfs2_xattr_block_set(struct inode
*inode
,
2913 struct ocfs2_xattr_info
*xi
,
2914 struct ocfs2_xattr_search
*xs
,
2915 struct ocfs2_xattr_set_ctxt
*ctxt
)
2917 struct buffer_head
*new_bh
= NULL
;
2918 struct ocfs2_xattr_block
*xblk
= NULL
;
2920 struct ocfs2_xa_loc loc
;
2922 if (!xs
->xattr_bh
) {
2923 ret
= ocfs2_create_xattr_block(inode
, xs
->inode_bh
, ctxt
,
2930 xs
->xattr_bh
= new_bh
;
2931 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2932 xs
->header
= &xblk
->xb_attrs
.xb_header
;
2933 xs
->base
= (void *)xs
->header
;
2934 xs
->end
= (void *)xblk
+ inode
->i_sb
->s_blocksize
;
2935 xs
->here
= xs
->header
->xh_entries
;
2937 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2939 if (!(le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2940 ocfs2_init_xattr_block_xa_loc(&loc
, inode
, xs
->xattr_bh
,
2941 xs
->not_found
? NULL
: xs
->here
);
2943 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
2945 xs
->here
= loc
.xl_entry
;
2946 else if (ret
!= -ENOSPC
)
2949 ret
= ocfs2_xattr_create_index_block(inode
, xs
, ctxt
);
2955 if (le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)
2956 ret
= ocfs2_xattr_set_entry_index_block(inode
, xi
, xs
, ctxt
);
2962 /* Check whether the new xattr can be inserted into the inode. */
2963 static int ocfs2_xattr_can_be_in_inode(struct inode
*inode
,
2964 struct ocfs2_xattr_info
*xi
,
2965 struct ocfs2_xattr_search
*xs
)
2967 struct ocfs2_xattr_entry
*last
;
2969 size_t min_offs
= xs
->end
- xs
->base
;
2974 last
= xs
->header
->xh_entries
;
2976 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
2977 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
2978 if (offs
< min_offs
)
2983 free
= min_offs
- ((void *)last
- xs
->base
) - OCFS2_XATTR_HEADER_GAP
;
2987 BUG_ON(!xs
->not_found
);
2989 if (free
>= (sizeof(struct ocfs2_xattr_entry
) + namevalue_size_xi(xi
)))
2995 static int ocfs2_calc_xattr_set_need(struct inode
*inode
,
2996 struct ocfs2_dinode
*di
,
2997 struct ocfs2_xattr_info
*xi
,
2998 struct ocfs2_xattr_search
*xis
,
2999 struct ocfs2_xattr_search
*xbs
,
3004 int ret
= 0, old_in_xb
= 0;
3005 int clusters_add
= 0, meta_add
= 0, credits
= 0;
3006 struct buffer_head
*bh
= NULL
;
3007 struct ocfs2_xattr_block
*xb
= NULL
;
3008 struct ocfs2_xattr_entry
*xe
= NULL
;
3009 struct ocfs2_xattr_value_root
*xv
= NULL
;
3011 int name_offset
, name_len
= 0;
3012 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
3017 * Calculate the clusters we need to write.
3018 * No matter whether we replace an old one or add a new one,
3019 * we need this for writing.
3021 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
)
3022 credits
+= new_clusters
*
3023 ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
3025 if (xis
->not_found
&& xbs
->not_found
) {
3026 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3028 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
3029 clusters_add
+= new_clusters
;
3030 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3038 if (!xis
->not_found
) {
3040 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3041 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3043 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
3045 int i
, block_off
= 0;
3046 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
3048 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3049 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3050 i
= xbs
->here
- xbs
->header
->xh_entries
;
3053 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
3054 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
3055 bucket_xh(xbs
->bucket
),
3058 base
= bucket_block(xbs
->bucket
, block_off
);
3059 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3062 credits
+= OCFS2_XATTR_BLOCK_UPDATE_CREDITS
;
3067 * delete a xattr doesn't need metadata and cluster allocation.
3068 * so just calculate the credits and return.
3070 * The credits for removing the value tree will be extended
3071 * by ocfs2_remove_extent itself.
3073 if (!xi
->xi_value
) {
3074 if (!ocfs2_xattr_is_local(xe
))
3075 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
);
3080 /* do cluster allocation guess first. */
3081 value_size
= le64_to_cpu(xe
->xe_value_size
);
3085 * In xattr set, we always try to set the xe in inode first,
3086 * so if it can be inserted into inode successfully, the old
3087 * one will be removed from the xattr block, and this xattr
3088 * will be inserted into inode as a new xattr in inode.
3090 if (ocfs2_xattr_can_be_in_inode(inode
, xi
, xis
)) {
3091 clusters_add
+= new_clusters
;
3092 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
) +
3093 OCFS2_INODE_UPDATE_CREDITS
;
3094 if (!ocfs2_xattr_is_local(xe
))
3095 credits
+= ocfs2_calc_extend_credits(
3103 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
3104 /* the new values will be stored outside. */
3105 u32 old_clusters
= 0;
3107 if (!ocfs2_xattr_is_local(xe
)) {
3108 old_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
3110 xv
= (struct ocfs2_xattr_value_root
*)
3111 (base
+ name_offset
+ name_len
);
3112 value_size
= OCFS2_XATTR_ROOT_SIZE
;
3116 if (old_clusters
>= new_clusters
) {
3117 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
);
3120 meta_add
+= ocfs2_extend_meta_needed(&xv
->xr_list
);
3121 clusters_add
+= new_clusters
- old_clusters
;
3122 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3126 if (value_size
>= OCFS2_XATTR_ROOT_SIZE
)
3131 * Now the new value will be stored inside. So if the new
3132 * value is smaller than the size of value root or the old
3133 * value, we don't need any allocation, otherwise we have
3134 * to guess metadata allocation.
3136 if ((ocfs2_xattr_is_local(xe
) &&
3137 (value_size
>= xi
->xi_value_len
)) ||
3138 (!ocfs2_xattr_is_local(xe
) &&
3139 OCFS2_XATTR_ROOT_SIZE
>= xi
->xi_value_len
))
3144 /* calculate metadata allocation. */
3145 if (di
->i_xattr_loc
) {
3146 if (!xbs
->xattr_bh
) {
3147 ret
= ocfs2_read_xattr_block(inode
,
3148 le64_to_cpu(di
->i_xattr_loc
),
3155 xb
= (struct ocfs2_xattr_block
*)bh
->b_data
;
3157 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
3160 * If there is already an xattr tree, good, we can calculate
3161 * like other b-trees. Otherwise we may have the chance of
3162 * create a tree, the credit calculation is borrowed from
3163 * ocfs2_calc_extend_credits with root_el = NULL. And the
3164 * new tree will be cluster based, so no meta is needed.
3166 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
3167 struct ocfs2_extent_list
*el
=
3168 &xb
->xb_attrs
.xb_root
.xt_list
;
3169 meta_add
+= ocfs2_extend_meta_needed(el
);
3170 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3173 credits
+= OCFS2_SUBALLOC_ALLOC
+ 1;
3176 * This cluster will be used either for new bucket or for
3178 * If the cluster size is the same as the bucket size, one
3179 * more is needed since we may need to extend the bucket
3183 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3184 if (OCFS2_XATTR_BUCKET_SIZE
==
3185 OCFS2_SB(inode
->i_sb
)->s_clustersize
) {
3186 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3191 credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
3195 *clusters_need
= clusters_add
;
3197 *meta_need
= meta_add
;
3199 *credits_need
= credits
;
3204 static int ocfs2_init_xattr_set_ctxt(struct inode
*inode
,
3205 struct ocfs2_dinode
*di
,
3206 struct ocfs2_xattr_info
*xi
,
3207 struct ocfs2_xattr_search
*xis
,
3208 struct ocfs2_xattr_search
*xbs
,
3209 struct ocfs2_xattr_set_ctxt
*ctxt
,
3213 int clusters_add
, meta_add
, ret
;
3214 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3216 memset(ctxt
, 0, sizeof(struct ocfs2_xattr_set_ctxt
));
3218 ocfs2_init_dealloc_ctxt(&ctxt
->dealloc
);
3220 ret
= ocfs2_calc_xattr_set_need(inode
, di
, xi
, xis
, xbs
,
3221 &clusters_add
, &meta_add
, credits
);
3227 meta_add
+= extra_meta
;
3228 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
3229 "credits = %d\n", xi
->xi_name
, meta_add
, clusters_add
, *credits
);
3232 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
,
3241 ret
= ocfs2_reserve_clusters(osb
, clusters_add
, &ctxt
->data_ac
);
3247 if (ctxt
->meta_ac
) {
3248 ocfs2_free_alloc_context(ctxt
->meta_ac
);
3249 ctxt
->meta_ac
= NULL
;
3253 * We cannot have an error and a non null ctxt->data_ac.
3260 static int __ocfs2_xattr_set_handle(struct inode
*inode
,
3261 struct ocfs2_dinode
*di
,
3262 struct ocfs2_xattr_info
*xi
,
3263 struct ocfs2_xattr_search
*xis
,
3264 struct ocfs2_xattr_search
*xbs
,
3265 struct ocfs2_xattr_set_ctxt
*ctxt
)
3267 int ret
= 0, credits
, old_found
;
3269 if (!xi
->xi_value
) {
3270 /* Remove existing extended attribute */
3271 if (!xis
->not_found
)
3272 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
3273 else if (!xbs
->not_found
)
3274 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
3276 /* We always try to set extended attribute into inode first*/
3277 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
3278 if (!ret
&& !xbs
->not_found
) {
3280 * If succeed and that extended attribute existing in
3281 * external block, then we will remove it.
3283 xi
->xi_value
= NULL
;
3284 xi
->xi_value_len
= 0;
3286 old_found
= xis
->not_found
;
3287 xis
->not_found
= -ENODATA
;
3288 ret
= ocfs2_calc_xattr_set_need(inode
,
3296 xis
->not_found
= old_found
;
3302 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
);
3307 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
3308 } else if (ret
== -ENOSPC
) {
3309 if (di
->i_xattr_loc
&& !xbs
->xattr_bh
) {
3310 ret
= ocfs2_xattr_block_find(inode
,
3316 old_found
= xis
->not_found
;
3317 xis
->not_found
= -ENODATA
;
3318 ret
= ocfs2_calc_xattr_set_need(inode
,
3326 xis
->not_found
= old_found
;
3332 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
);
3339 * If no space in inode, we will set extended attribute
3340 * into external block.
3342 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
3345 if (!xis
->not_found
) {
3347 * If succeed and that extended attribute
3348 * existing in inode, we will remove it.
3350 xi
->xi_value
= NULL
;
3351 xi
->xi_value_len
= 0;
3352 xbs
->not_found
= -ENODATA
;
3353 ret
= ocfs2_calc_xattr_set_need(inode
,
3366 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
);
3371 ret
= ocfs2_xattr_ibody_set(inode
, xi
,
3378 /* Update inode ctime. */
3379 ret
= ocfs2_journal_access_di(ctxt
->handle
, INODE_CACHE(inode
),
3381 OCFS2_JOURNAL_ACCESS_WRITE
);
3387 inode
->i_ctime
= CURRENT_TIME
;
3388 di
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
3389 di
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
3390 ocfs2_journal_dirty(ctxt
->handle
, xis
->inode_bh
);
3397 * This function only called duing creating inode
3398 * for init security/acl xattrs of the new inode.
3399 * All transanction credits have been reserved in mknod.
3401 int ocfs2_xattr_set_handle(handle_t
*handle
,
3402 struct inode
*inode
,
3403 struct buffer_head
*di_bh
,
3409 struct ocfs2_alloc_context
*meta_ac
,
3410 struct ocfs2_alloc_context
*data_ac
)
3412 struct ocfs2_dinode
*di
;
3415 struct ocfs2_xattr_info xi
= {
3416 .xi_name_index
= name_index
,
3418 .xi_name_len
= strlen(name
),
3420 .xi_value_len
= value_len
,
3423 struct ocfs2_xattr_search xis
= {
3424 .not_found
= -ENODATA
,
3427 struct ocfs2_xattr_search xbs
= {
3428 .not_found
= -ENODATA
,
3431 struct ocfs2_xattr_set_ctxt ctxt
= {
3437 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
3441 * In extreme situation, may need xattr bucket when
3442 * block size is too small. And we have already reserved
3443 * the credits for bucket in mknod.
3445 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
) {
3446 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
3448 mlog_errno(-ENOMEM
);
3453 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
3454 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3456 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3458 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
3461 if (xis
.not_found
) {
3462 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
3467 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
3470 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3471 brelse(xbs
.xattr_bh
);
3472 ocfs2_xattr_bucket_free(xbs
.bucket
);
3480 * Set, replace or remove an extended attribute for this inode.
3481 * value is NULL to remove an existing extended attribute, else either
3482 * create or replace an extended attribute.
3484 int ocfs2_xattr_set(struct inode
*inode
,
3491 struct buffer_head
*di_bh
= NULL
;
3492 struct ocfs2_dinode
*di
;
3493 int ret
, credits
, ref_meta
= 0, ref_credits
= 0;
3494 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3495 struct inode
*tl_inode
= osb
->osb_tl_inode
;
3496 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
3497 struct ocfs2_refcount_tree
*ref_tree
= NULL
;
3499 struct ocfs2_xattr_info xi
= {
3500 .xi_name_index
= name_index
,
3502 .xi_name_len
= strlen(name
),
3504 .xi_value_len
= value_len
,
3507 struct ocfs2_xattr_search xis
= {
3508 .not_found
= -ENODATA
,
3511 struct ocfs2_xattr_search xbs
= {
3512 .not_found
= -ENODATA
,
3515 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
3519 * Only xbs will be used on indexed trees. xis doesn't need a
3522 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
3524 mlog_errno(-ENOMEM
);
3528 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
3531 goto cleanup_nolock
;
3533 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
3534 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3536 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3538 * Scan inode and external block to find the same name
3539 * extended attribute and collect search infomation.
3541 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
3544 if (xis
.not_found
) {
3545 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
3550 if (xis
.not_found
&& xbs
.not_found
) {
3552 if (flags
& XATTR_REPLACE
)
3559 if (flags
& XATTR_CREATE
)
3563 /* Check whether the value is refcounted and do some prepartion. */
3564 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
&&
3565 (!xis
.not_found
|| !xbs
.not_found
)) {
3566 ret
= ocfs2_prepare_refcount_xattr(inode
, di
, &xi
,
3567 &xis
, &xbs
, &ref_tree
,
3568 &ref_meta
, &ref_credits
);
3575 mutex_lock(&tl_inode
->i_mutex
);
3577 if (ocfs2_truncate_log_needs_flush(osb
)) {
3578 ret
= __ocfs2_flush_truncate_log(osb
);
3580 mutex_unlock(&tl_inode
->i_mutex
);
3585 mutex_unlock(&tl_inode
->i_mutex
);
3587 ret
= ocfs2_init_xattr_set_ctxt(inode
, di
, &xi
, &xis
,
3588 &xbs
, &ctxt
, ref_meta
, &credits
);
3594 /* we need to update inode's ctime field, so add credit for it. */
3595 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
3596 ctxt
.handle
= ocfs2_start_trans(osb
, credits
+ ref_credits
);
3597 if (IS_ERR(ctxt
.handle
)) {
3598 ret
= PTR_ERR(ctxt
.handle
);
3603 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
3605 ocfs2_commit_trans(osb
, ctxt
.handle
);
3608 ocfs2_free_alloc_context(ctxt
.data_ac
);
3610 ocfs2_free_alloc_context(ctxt
.meta_ac
);
3611 if (ocfs2_dealloc_has_cluster(&ctxt
.dealloc
))
3612 ocfs2_schedule_truncate_log_flush(osb
, 1);
3613 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
3617 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
3618 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3619 if (!value
&& !ret
) {
3620 ret
= ocfs2_try_remove_refcount_tree(inode
, di_bh
);
3624 ocfs2_inode_unlock(inode
, 1);
3627 brelse(xbs
.xattr_bh
);
3628 ocfs2_xattr_bucket_free(xbs
.bucket
);
3634 * Find the xattr extent rec which may contains name_hash.
3635 * e_cpos will be the first name hash of the xattr rec.
3636 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3638 static int ocfs2_xattr_get_rec(struct inode
*inode
,
3643 struct ocfs2_extent_list
*el
)
3646 struct buffer_head
*eb_bh
= NULL
;
3647 struct ocfs2_extent_block
*eb
;
3648 struct ocfs2_extent_rec
*rec
= NULL
;
3651 if (el
->l_tree_depth
) {
3652 ret
= ocfs2_find_leaf(INODE_CACHE(inode
), el
, name_hash
,
3659 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
3662 if (el
->l_tree_depth
) {
3663 ocfs2_error(inode
->i_sb
,
3664 "Inode %lu has non zero tree depth in "
3665 "xattr tree block %llu\n", inode
->i_ino
,
3666 (unsigned long long)eb_bh
->b_blocknr
);
3672 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
3673 rec
= &el
->l_recs
[i
];
3675 if (le32_to_cpu(rec
->e_cpos
) <= name_hash
) {
3676 e_blkno
= le64_to_cpu(rec
->e_blkno
);
3682 ocfs2_error(inode
->i_sb
, "Inode %lu has bad extent "
3683 "record (%u, %u, 0) in xattr", inode
->i_ino
,
3684 le32_to_cpu(rec
->e_cpos
),
3685 ocfs2_rec_clusters(el
, rec
));
3690 *p_blkno
= le64_to_cpu(rec
->e_blkno
);
3691 *num_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
3693 *e_cpos
= le32_to_cpu(rec
->e_cpos
);
3699 typedef int (xattr_bucket_func
)(struct inode
*inode
,
3700 struct ocfs2_xattr_bucket
*bucket
,
3703 static int ocfs2_find_xe_in_bucket(struct inode
*inode
,
3704 struct ocfs2_xattr_bucket
*bucket
,
3711 int i
, ret
= 0, cmp
= 1, block_off
, new_offset
;
3712 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
3713 size_t name_len
= strlen(name
);
3714 struct ocfs2_xattr_entry
*xe
= NULL
;
3718 * We don't use binary search in the bucket because there
3719 * may be multiple entries with the same name hash.
3721 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
3722 xe
= &xh
->xh_entries
[i
];
3724 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
))
3726 else if (name_hash
< le32_to_cpu(xe
->xe_name_hash
))
3729 cmp
= name_index
- ocfs2_xattr_get_type(xe
);
3731 cmp
= name_len
- xe
->xe_name_len
;
3735 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
3746 xe_name
= bucket_block(bucket
, block_off
) + new_offset
;
3747 if (!memcmp(name
, xe_name
, name_len
)) {
3759 * Find the specified xattr entry in a series of buckets.
3760 * This series start from p_blkno and last for num_clusters.
3761 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3762 * the num of the valid buckets.
3764 * Return the buffer_head this xattr should reside in. And if the xattr's
3765 * hash is in the gap of 2 buckets, return the lower bucket.
3767 static int ocfs2_xattr_bucket_find(struct inode
*inode
,
3774 struct ocfs2_xattr_search
*xs
)
3777 struct ocfs2_xattr_header
*xh
= NULL
;
3778 struct ocfs2_xattr_entry
*xe
= NULL
;
3780 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3781 int low_bucket
= 0, bucket
, high_bucket
;
3782 struct ocfs2_xattr_bucket
*search
;
3784 u64 blkno
, lower_blkno
= 0;
3786 search
= ocfs2_xattr_bucket_new(inode
);
3793 ret
= ocfs2_read_xattr_bucket(search
, p_blkno
);
3799 xh
= bucket_xh(search
);
3800 high_bucket
= le16_to_cpu(xh
->xh_num_buckets
) - 1;
3801 while (low_bucket
<= high_bucket
) {
3802 ocfs2_xattr_bucket_relse(search
);
3804 bucket
= (low_bucket
+ high_bucket
) / 2;
3805 blkno
= p_blkno
+ bucket
* blk_per_bucket
;
3806 ret
= ocfs2_read_xattr_bucket(search
, blkno
);
3812 xh
= bucket_xh(search
);
3813 xe
= &xh
->xh_entries
[0];
3814 if (name_hash
< le32_to_cpu(xe
->xe_name_hash
)) {
3815 high_bucket
= bucket
- 1;
3820 * Check whether the hash of the last entry in our
3821 * bucket is larger than the search one. for an empty
3822 * bucket, the last one is also the first one.
3825 xe
= &xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1];
3827 last_hash
= le32_to_cpu(xe
->xe_name_hash
);
3829 /* record lower_blkno which may be the insert place. */
3830 lower_blkno
= blkno
;
3832 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
)) {
3833 low_bucket
= bucket
+ 1;
3837 /* the searched xattr should reside in this bucket if exists. */
3838 ret
= ocfs2_find_xe_in_bucket(inode
, search
,
3839 name_index
, name
, name_hash
,
3849 * Record the bucket we have found.
3850 * When the xattr's hash value is in the gap of 2 buckets, we will
3851 * always set it to the previous bucket.
3854 lower_blkno
= p_blkno
;
3856 /* This should be in cache - we just read it during the search */
3857 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, lower_blkno
);
3863 xs
->header
= bucket_xh(xs
->bucket
);
3864 xs
->base
= bucket_block(xs
->bucket
, 0);
3865 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
3868 xs
->here
= &xs
->header
->xh_entries
[index
];
3869 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name
,
3870 (unsigned long long)bucket_blkno(xs
->bucket
), index
);
3875 ocfs2_xattr_bucket_free(search
);
3879 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
3880 struct buffer_head
*root_bh
,
3883 struct ocfs2_xattr_search
*xs
)
3886 struct ocfs2_xattr_block
*xb
=
3887 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
3888 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
3889 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
3891 u32 first_hash
, num_clusters
= 0;
3892 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
3894 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
3897 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3898 name
, name_hash
, name_index
);
3900 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &first_hash
,
3907 BUG_ON(p_blkno
== 0 || num_clusters
== 0 || first_hash
> name_hash
);
3909 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3910 "in the rec is %u\n", num_clusters
, (unsigned long long)p_blkno
,
3913 ret
= ocfs2_xattr_bucket_find(inode
, name_index
, name
, name_hash
,
3914 p_blkno
, first_hash
, num_clusters
, xs
);
3920 static int ocfs2_iterate_xattr_buckets(struct inode
*inode
,
3923 xattr_bucket_func
*func
,
3927 u32 bpc
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
));
3928 u32 num_buckets
= clusters
* bpc
;
3929 struct ocfs2_xattr_bucket
*bucket
;
3931 bucket
= ocfs2_xattr_bucket_new(inode
);
3933 mlog_errno(-ENOMEM
);
3937 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3938 clusters
, (unsigned long long)blkno
);
3940 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bucket
->bu_blocks
) {
3941 ret
= ocfs2_read_xattr_bucket(bucket
, blkno
);
3948 * The real bucket num in this series of blocks is stored
3949 * in the 1st bucket.
3952 num_buckets
= le16_to_cpu(bucket_xh(bucket
)->xh_num_buckets
);
3954 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3955 (unsigned long long)blkno
,
3956 le32_to_cpu(bucket_xh(bucket
)->xh_entries
[0].xe_name_hash
));
3958 ret
= func(inode
, bucket
, para
);
3959 if (ret
&& ret
!= -ERANGE
)
3961 /* Fall through to bucket_relse() */
3964 ocfs2_xattr_bucket_relse(bucket
);
3969 ocfs2_xattr_bucket_free(bucket
);
3973 struct ocfs2_xattr_tree_list
{
3979 static int ocfs2_xattr_bucket_get_name_value(struct super_block
*sb
,
3980 struct ocfs2_xattr_header
*xh
,
3987 if (index
< 0 || index
>= le16_to_cpu(xh
->xh_count
))
3990 name_offset
= le16_to_cpu(xh
->xh_entries
[index
].xe_name_offset
);
3992 *block_off
= name_offset
>> sb
->s_blocksize_bits
;
3993 *new_offset
= name_offset
% sb
->s_blocksize
;
3998 static int ocfs2_list_xattr_bucket(struct inode
*inode
,
3999 struct ocfs2_xattr_bucket
*bucket
,
4003 struct ocfs2_xattr_tree_list
*xl
= (struct ocfs2_xattr_tree_list
*)para
;
4004 int i
, block_off
, new_offset
;
4005 const char *prefix
, *name
;
4007 for (i
= 0 ; i
< le16_to_cpu(bucket_xh(bucket
)->xh_count
); i
++) {
4008 struct ocfs2_xattr_entry
*entry
= &bucket_xh(bucket
)->xh_entries
[i
];
4009 type
= ocfs2_xattr_get_type(entry
);
4010 prefix
= ocfs2_xattr_prefix(type
);
4013 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
4021 name
= (const char *)bucket_block(bucket
, block_off
) +
4023 ret
= ocfs2_xattr_list_entry(xl
->buffer
,
4027 entry
->xe_name_len
);
4036 static int ocfs2_iterate_xattr_index_block(struct inode
*inode
,
4037 struct buffer_head
*blk_bh
,
4038 xattr_tree_rec_func
*rec_func
,
4041 struct ocfs2_xattr_block
*xb
=
4042 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
4043 struct ocfs2_extent_list
*el
= &xb
->xb_attrs
.xb_root
.xt_list
;
4045 u32 name_hash
= UINT_MAX
, e_cpos
= 0, num_clusters
= 0;
4048 if (!el
->l_next_free_rec
|| !rec_func
)
4051 while (name_hash
> 0) {
4052 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
4053 &e_cpos
, &num_clusters
, el
);
4059 ret
= rec_func(inode
, blk_bh
, p_blkno
, e_cpos
,
4060 num_clusters
, para
);
4070 name_hash
= e_cpos
- 1;
4077 static int ocfs2_list_xattr_tree_rec(struct inode
*inode
,
4078 struct buffer_head
*root_bh
,
4079 u64 blkno
, u32 cpos
, u32 len
, void *para
)
4081 return ocfs2_iterate_xattr_buckets(inode
, blkno
, len
,
4082 ocfs2_list_xattr_bucket
, para
);
4085 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
4086 struct buffer_head
*blk_bh
,
4091 struct ocfs2_xattr_tree_list xl
= {
4093 .buffer_size
= buffer_size
,
4097 ret
= ocfs2_iterate_xattr_index_block(inode
, blk_bh
,
4098 ocfs2_list_xattr_tree_rec
, &xl
);
4109 static int cmp_xe(const void *a
, const void *b
)
4111 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
4112 u32 l_hash
= le32_to_cpu(l
->xe_name_hash
);
4113 u32 r_hash
= le32_to_cpu(r
->xe_name_hash
);
4115 if (l_hash
> r_hash
)
4117 if (l_hash
< r_hash
)
4122 static void swap_xe(void *a
, void *b
, int size
)
4124 struct ocfs2_xattr_entry
*l
= a
, *r
= b
, tmp
;
4127 memcpy(l
, r
, sizeof(struct ocfs2_xattr_entry
));
4128 memcpy(r
, &tmp
, sizeof(struct ocfs2_xattr_entry
));
4132 * When the ocfs2_xattr_block is filled up, new bucket will be created
4133 * and all the xattr entries will be moved to the new bucket.
4134 * The header goes at the start of the bucket, and the names+values are
4135 * filled from the end. This is why *target starts as the last buffer.
4136 * Note: we need to sort the entries since they are not saved in order
4137 * in the ocfs2_xattr_block.
4139 static void ocfs2_cp_xattr_block_to_bucket(struct inode
*inode
,
4140 struct buffer_head
*xb_bh
,
4141 struct ocfs2_xattr_bucket
*bucket
)
4143 int i
, blocksize
= inode
->i_sb
->s_blocksize
;
4144 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4145 u16 offset
, size
, off_change
;
4146 struct ocfs2_xattr_entry
*xe
;
4147 struct ocfs2_xattr_block
*xb
=
4148 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4149 struct ocfs2_xattr_header
*xb_xh
= &xb
->xb_attrs
.xb_header
;
4150 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
4151 u16 count
= le16_to_cpu(xb_xh
->xh_count
);
4152 char *src
= xb_bh
->b_data
;
4153 char *target
= bucket_block(bucket
, blks
- 1);
4155 mlog(0, "cp xattr from block %llu to bucket %llu\n",
4156 (unsigned long long)xb_bh
->b_blocknr
,
4157 (unsigned long long)bucket_blkno(bucket
));
4159 for (i
= 0; i
< blks
; i
++)
4160 memset(bucket_block(bucket
, i
), 0, blocksize
);
4163 * Since the xe_name_offset is based on ocfs2_xattr_header,
4164 * there is a offset change corresponding to the change of
4165 * ocfs2_xattr_header's position.
4167 off_change
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
4168 xe
= &xb_xh
->xh_entries
[count
- 1];
4169 offset
= le16_to_cpu(xe
->xe_name_offset
) + off_change
;
4170 size
= blocksize
- offset
;
4172 /* copy all the names and values. */
4173 memcpy(target
+ offset
, src
+ offset
, size
);
4175 /* Init new header now. */
4176 xh
->xh_count
= xb_xh
->xh_count
;
4177 xh
->xh_num_buckets
= cpu_to_le16(1);
4178 xh
->xh_name_value_len
= cpu_to_le16(size
);
4179 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
- size
);
4181 /* copy all the entries. */
4182 target
= bucket_block(bucket
, 0);
4183 offset
= offsetof(struct ocfs2_xattr_header
, xh_entries
);
4184 size
= count
* sizeof(struct ocfs2_xattr_entry
);
4185 memcpy(target
+ offset
, (char *)xb_xh
+ offset
, size
);
4187 /* Change the xe offset for all the xe because of the move. */
4188 off_change
= OCFS2_XATTR_BUCKET_SIZE
- blocksize
+
4189 offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
4190 for (i
= 0; i
< count
; i
++)
4191 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
, off_change
);
4193 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
4194 offset
, size
, off_change
);
4196 sort(target
+ offset
, count
, sizeof(struct ocfs2_xattr_entry
),
4201 * After we move xattr from block to index btree, we have to
4202 * update ocfs2_xattr_search to the new xe and base.
4204 * When the entry is in xattr block, xattr_bh indicates the storage place.
4205 * While if the entry is in index b-tree, "bucket" indicates the
4206 * real place of the xattr.
4208 static void ocfs2_xattr_update_xattr_search(struct inode
*inode
,
4209 struct ocfs2_xattr_search
*xs
,
4210 struct buffer_head
*old_bh
)
4212 char *buf
= old_bh
->b_data
;
4213 struct ocfs2_xattr_block
*old_xb
= (struct ocfs2_xattr_block
*)buf
;
4214 struct ocfs2_xattr_header
*old_xh
= &old_xb
->xb_attrs
.xb_header
;
4217 xs
->header
= bucket_xh(xs
->bucket
);
4218 xs
->base
= bucket_block(xs
->bucket
, 0);
4219 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
4224 i
= xs
->here
- old_xh
->xh_entries
;
4225 xs
->here
= &xs
->header
->xh_entries
[i
];
4228 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
4229 struct ocfs2_xattr_search
*xs
,
4230 struct ocfs2_xattr_set_ctxt
*ctxt
)
4235 handle_t
*handle
= ctxt
->handle
;
4236 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
4237 struct buffer_head
*xb_bh
= xs
->xattr_bh
;
4238 struct ocfs2_xattr_block
*xb
=
4239 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4240 struct ocfs2_xattr_tree_root
*xr
;
4241 u16 xb_flags
= le16_to_cpu(xb
->xb_flags
);
4243 mlog(0, "create xattr index block for %llu\n",
4244 (unsigned long long)xb_bh
->b_blocknr
);
4246 BUG_ON(xb_flags
& OCFS2_XATTR_INDEXED
);
4247 BUG_ON(!xs
->bucket
);
4251 * We can use this lock for now, and maybe move to a dedicated mutex
4252 * if performance becomes a problem later.
4254 down_write(&oi
->ip_alloc_sem
);
4256 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(inode
), xb_bh
,
4257 OCFS2_JOURNAL_ACCESS_WRITE
);
4263 ret
= __ocfs2_claim_clusters(handle
, ctxt
->data_ac
,
4264 1, 1, &bit_off
, &len
);
4271 * The bucket may spread in many blocks, and
4272 * we will only touch the 1st block and the last block
4273 * in the whole bucket(one for entry and one for data).
4275 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, bit_off
);
4277 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
4278 (unsigned long long)blkno
);
4280 ret
= ocfs2_init_xattr_bucket(xs
->bucket
, blkno
);
4286 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4287 OCFS2_JOURNAL_ACCESS_CREATE
);
4293 ocfs2_cp_xattr_block_to_bucket(inode
, xb_bh
, xs
->bucket
);
4294 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4296 ocfs2_xattr_update_xattr_search(inode
, xs
, xb_bh
);
4298 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4299 memset(&xb
->xb_attrs
, 0, inode
->i_sb
->s_blocksize
-
4300 offsetof(struct ocfs2_xattr_block
, xb_attrs
));
4302 xr
= &xb
->xb_attrs
.xb_root
;
4303 xr
->xt_clusters
= cpu_to_le32(1);
4304 xr
->xt_last_eb_blk
= 0;
4305 xr
->xt_list
.l_tree_depth
= 0;
4306 xr
->xt_list
.l_count
= cpu_to_le16(ocfs2_xattr_recs_per_xb(inode
->i_sb
));
4307 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
4309 xr
->xt_list
.l_recs
[0].e_cpos
= 0;
4310 xr
->xt_list
.l_recs
[0].e_blkno
= cpu_to_le64(blkno
);
4311 xr
->xt_list
.l_recs
[0].e_leaf_clusters
= cpu_to_le16(1);
4313 xb
->xb_flags
= cpu_to_le16(xb_flags
| OCFS2_XATTR_INDEXED
);
4315 ocfs2_journal_dirty(handle
, xb_bh
);
4318 up_write(&oi
->ip_alloc_sem
);
4323 static int cmp_xe_offset(const void *a
, const void *b
)
4325 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
4326 u32 l_name_offset
= le16_to_cpu(l
->xe_name_offset
);
4327 u32 r_name_offset
= le16_to_cpu(r
->xe_name_offset
);
4329 if (l_name_offset
< r_name_offset
)
4331 if (l_name_offset
> r_name_offset
)
4337 * defrag a xattr bucket if we find that the bucket has some
4338 * holes beteen name/value pairs.
4339 * We will move all the name/value pairs to the end of the bucket
4340 * so that we can spare some space for insertion.
4342 static int ocfs2_defrag_xattr_bucket(struct inode
*inode
,
4344 struct ocfs2_xattr_bucket
*bucket
)
4347 size_t end
, offset
, len
;
4348 struct ocfs2_xattr_header
*xh
;
4349 char *entries
, *buf
, *bucket_buf
= NULL
;
4350 u64 blkno
= bucket_blkno(bucket
);
4352 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4353 struct ocfs2_xattr_entry
*xe
;
4356 * In order to make the operation more efficient and generic,
4357 * we copy all the blocks into a contiguous memory and do the
4358 * defragment there, so if anything is error, we will not touch
4361 bucket_buf
= kmalloc(OCFS2_XATTR_BUCKET_SIZE
, GFP_NOFS
);
4368 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
4369 memcpy(buf
, bucket_block(bucket
, i
), blocksize
);
4371 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
4372 OCFS2_JOURNAL_ACCESS_WRITE
);
4378 xh
= (struct ocfs2_xattr_header
*)bucket_buf
;
4379 entries
= (char *)xh
->xh_entries
;
4380 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
4382 mlog(0, "adjust xattr bucket in %llu, count = %u, "
4383 "xh_free_start = %u, xh_name_value_len = %u.\n",
4384 (unsigned long long)blkno
, le16_to_cpu(xh
->xh_count
),
4385 xh_free_start
, le16_to_cpu(xh
->xh_name_value_len
));
4388 * sort all the entries by their offset.
4389 * the largest will be the first, so that we can
4390 * move them to the end one by one.
4392 sort(entries
, le16_to_cpu(xh
->xh_count
),
4393 sizeof(struct ocfs2_xattr_entry
),
4394 cmp_xe_offset
, swap_xe
);
4396 /* Move all name/values to the end of the bucket. */
4397 xe
= xh
->xh_entries
;
4398 end
= OCFS2_XATTR_BUCKET_SIZE
;
4399 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, xe
++) {
4400 offset
= le16_to_cpu(xe
->xe_name_offset
);
4401 len
= namevalue_size_xe(xe
);
4404 * We must make sure that the name/value pair
4405 * exist in the same block. So adjust end to
4406 * the previous block end if needed.
4408 if (((end
- len
) / blocksize
!=
4409 (end
- 1) / blocksize
))
4410 end
= end
- end
% blocksize
;
4412 if (end
> offset
+ len
) {
4413 memmove(bucket_buf
+ end
- len
,
4414 bucket_buf
+ offset
, len
);
4415 xe
->xe_name_offset
= cpu_to_le16(end
- len
);
4418 mlog_bug_on_msg(end
< offset
+ len
, "Defrag check failed for "
4419 "bucket %llu\n", (unsigned long long)blkno
);
4424 mlog_bug_on_msg(xh_free_start
> end
, "Defrag check failed for "
4425 "bucket %llu\n", (unsigned long long)blkno
);
4427 if (xh_free_start
== end
)
4430 memset(bucket_buf
+ xh_free_start
, 0, end
- xh_free_start
);
4431 xh
->xh_free_start
= cpu_to_le16(end
);
4433 /* sort the entries by their name_hash. */
4434 sort(entries
, le16_to_cpu(xh
->xh_count
),
4435 sizeof(struct ocfs2_xattr_entry
),
4439 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
4440 memcpy(bucket_block(bucket
, i
), buf
, blocksize
);
4441 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
4449 * prev_blkno points to the start of an existing extent. new_blkno
4450 * points to a newly allocated extent. Because we know each of our
4451 * clusters contains more than bucket, we can easily split one cluster
4452 * at a bucket boundary. So we take the last cluster of the existing
4453 * extent and split it down the middle. We move the last half of the
4454 * buckets in the last cluster of the existing extent over to the new
4457 * first_bh is the buffer at prev_blkno so we can update the existing
4458 * extent's bucket count. header_bh is the bucket were we were hoping
4459 * to insert our xattr. If the bucket move places the target in the new
4460 * extent, we'll update first_bh and header_bh after modifying the old
4463 * first_hash will be set as the 1st xe's name_hash in the new extent.
4465 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode
*inode
,
4467 struct ocfs2_xattr_bucket
*first
,
4468 struct ocfs2_xattr_bucket
*target
,
4474 struct super_block
*sb
= inode
->i_sb
;
4475 int blks_per_bucket
= ocfs2_blocks_per_xattr_bucket(sb
);
4476 int num_buckets
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb
));
4477 int to_move
= num_buckets
/ 2;
4479 u64 last_cluster_blkno
= bucket_blkno(first
) +
4480 ((num_clusters
- 1) * ocfs2_clusters_to_blocks(sb
, 1));
4482 BUG_ON(le16_to_cpu(bucket_xh(first
)->xh_num_buckets
) < num_buckets
);
4483 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
== OCFS2_SB(sb
)->s_clustersize
);
4485 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4486 (unsigned long long)last_cluster_blkno
, (unsigned long long)new_blkno
);
4488 ret
= ocfs2_mv_xattr_buckets(inode
, handle
, bucket_blkno(first
),
4489 last_cluster_blkno
, new_blkno
,
4490 to_move
, first_hash
);
4496 /* This is the first bucket that got moved */
4497 src_blkno
= last_cluster_blkno
+ (to_move
* blks_per_bucket
);
4500 * If the target bucket was part of the moved buckets, we need to
4501 * update first and target.
4503 if (bucket_blkno(target
) >= src_blkno
) {
4504 /* Find the block for the new target bucket */
4505 src_blkno
= new_blkno
+
4506 (bucket_blkno(target
) - src_blkno
);
4508 ocfs2_xattr_bucket_relse(first
);
4509 ocfs2_xattr_bucket_relse(target
);
4512 * These shouldn't fail - the buffers are in the
4513 * journal from ocfs2_cp_xattr_bucket().
4515 ret
= ocfs2_read_xattr_bucket(first
, new_blkno
);
4520 ret
= ocfs2_read_xattr_bucket(target
, src_blkno
);
4531 * Find the suitable pos when we divide a bucket into 2.
4532 * We have to make sure the xattrs with the same hash value exist
4533 * in the same bucket.
4535 * If this ocfs2_xattr_header covers more than one hash value, find a
4536 * place where the hash value changes. Try to find the most even split.
4537 * The most common case is that all entries have different hash values,
4538 * and the first check we make will find a place to split.
4540 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header
*xh
)
4542 struct ocfs2_xattr_entry
*entries
= xh
->xh_entries
;
4543 int count
= le16_to_cpu(xh
->xh_count
);
4544 int delta
, middle
= count
/ 2;
4547 * We start at the middle. Each step gets farther away in both
4548 * directions. We therefore hit the change in hash value
4549 * nearest to the middle. Note that this loop does not execute for
4552 for (delta
= 0; delta
< middle
; delta
++) {
4553 /* Let's check delta earlier than middle */
4554 if (cmp_xe(&entries
[middle
- delta
- 1],
4555 &entries
[middle
- delta
]))
4556 return middle
- delta
;
4558 /* For even counts, don't walk off the end */
4559 if ((middle
+ delta
+ 1) == count
)
4562 /* Now try delta past middle */
4563 if (cmp_xe(&entries
[middle
+ delta
],
4564 &entries
[middle
+ delta
+ 1]))
4565 return middle
+ delta
+ 1;
4568 /* Every entry had the same hash */
4573 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4574 * first_hash will record the 1st hash of the new bucket.
4576 * Normally half of the xattrs will be moved. But we have to make
4577 * sure that the xattrs with the same hash value are stored in the
4578 * same bucket. If all the xattrs in this bucket have the same hash
4579 * value, the new bucket will be initialized as an empty one and the
4580 * first_hash will be initialized as (hash_value+1).
4582 static int ocfs2_divide_xattr_bucket(struct inode
*inode
,
4587 int new_bucket_head
)
4590 int count
, start
, len
, name_value_len
= 0, name_offset
= 0;
4591 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
4592 struct ocfs2_xattr_header
*xh
;
4593 struct ocfs2_xattr_entry
*xe
;
4594 int blocksize
= inode
->i_sb
->s_blocksize
;
4596 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4597 (unsigned long long)blk
, (unsigned long long)new_blk
);
4599 s_bucket
= ocfs2_xattr_bucket_new(inode
);
4600 t_bucket
= ocfs2_xattr_bucket_new(inode
);
4601 if (!s_bucket
|| !t_bucket
) {
4607 ret
= ocfs2_read_xattr_bucket(s_bucket
, blk
);
4613 ret
= ocfs2_xattr_bucket_journal_access(handle
, s_bucket
,
4614 OCFS2_JOURNAL_ACCESS_WRITE
);
4621 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4622 * there's no need to read it.
4624 ret
= ocfs2_init_xattr_bucket(t_bucket
, new_blk
);
4631 * Hey, if we're overwriting t_bucket, what difference does
4632 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4633 * same part of ocfs2_cp_xattr_bucket().
4635 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
4637 OCFS2_JOURNAL_ACCESS_CREATE
:
4638 OCFS2_JOURNAL_ACCESS_WRITE
);
4644 xh
= bucket_xh(s_bucket
);
4645 count
= le16_to_cpu(xh
->xh_count
);
4646 start
= ocfs2_xattr_find_divide_pos(xh
);
4648 if (start
== count
) {
4649 xe
= &xh
->xh_entries
[start
-1];
4652 * initialized a new empty bucket here.
4653 * The hash value is set as one larger than
4654 * that of the last entry in the previous bucket.
4656 for (i
= 0; i
< t_bucket
->bu_blocks
; i
++)
4657 memset(bucket_block(t_bucket
, i
), 0, blocksize
);
4659 xh
= bucket_xh(t_bucket
);
4660 xh
->xh_free_start
= cpu_to_le16(blocksize
);
4661 xh
->xh_entries
[0].xe_name_hash
= xe
->xe_name_hash
;
4662 le32_add_cpu(&xh
->xh_entries
[0].xe_name_hash
, 1);
4664 goto set_num_buckets
;
4667 /* copy the whole bucket to the new first. */
4668 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
4670 /* update the new bucket. */
4671 xh
= bucket_xh(t_bucket
);
4674 * Calculate the total name/value len and xh_free_start for
4675 * the old bucket first.
4677 name_offset
= OCFS2_XATTR_BUCKET_SIZE
;
4679 for (i
= 0; i
< start
; i
++) {
4680 xe
= &xh
->xh_entries
[i
];
4681 name_value_len
+= namevalue_size_xe(xe
);
4682 if (le16_to_cpu(xe
->xe_name_offset
) < name_offset
)
4683 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
4687 * Now begin the modification to the new bucket.
4689 * In the new bucket, We just move the xattr entry to the beginning
4690 * and don't touch the name/value. So there will be some holes in the
4691 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4694 xe
= &xh
->xh_entries
[start
];
4695 len
= sizeof(struct ocfs2_xattr_entry
) * (count
- start
);
4696 mlog(0, "mv xattr entry len %d from %d to %d\n", len
,
4697 (int)((char *)xe
- (char *)xh
),
4698 (int)((char *)xh
->xh_entries
- (char *)xh
));
4699 memmove((char *)xh
->xh_entries
, (char *)xe
, len
);
4700 xe
= &xh
->xh_entries
[count
- start
];
4701 len
= sizeof(struct ocfs2_xattr_entry
) * start
;
4702 memset((char *)xe
, 0, len
);
4704 le16_add_cpu(&xh
->xh_count
, -start
);
4705 le16_add_cpu(&xh
->xh_name_value_len
, -name_value_len
);
4707 /* Calculate xh_free_start for the new bucket. */
4708 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
4709 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
4710 xe
= &xh
->xh_entries
[i
];
4711 if (le16_to_cpu(xe
->xe_name_offset
) <
4712 le16_to_cpu(xh
->xh_free_start
))
4713 xh
->xh_free_start
= xe
->xe_name_offset
;
4717 /* set xh->xh_num_buckets for the new xh. */
4718 if (new_bucket_head
)
4719 xh
->xh_num_buckets
= cpu_to_le16(1);
4721 xh
->xh_num_buckets
= 0;
4723 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
4725 /* store the first_hash of the new bucket. */
4727 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
4730 * Now only update the 1st block of the old bucket. If we
4731 * just added a new empty bucket, there is no need to modify
4737 xh
= bucket_xh(s_bucket
);
4738 memset(&xh
->xh_entries
[start
], 0,
4739 sizeof(struct ocfs2_xattr_entry
) * (count
- start
));
4740 xh
->xh_count
= cpu_to_le16(start
);
4741 xh
->xh_free_start
= cpu_to_le16(name_offset
);
4742 xh
->xh_name_value_len
= cpu_to_le16(name_value_len
);
4744 ocfs2_xattr_bucket_journal_dirty(handle
, s_bucket
);
4747 ocfs2_xattr_bucket_free(s_bucket
);
4748 ocfs2_xattr_bucket_free(t_bucket
);
4754 * Copy xattr from one bucket to another bucket.
4756 * The caller must make sure that the journal transaction
4757 * has enough space for journaling.
4759 static int ocfs2_cp_xattr_bucket(struct inode
*inode
,
4766 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
4768 BUG_ON(s_blkno
== t_blkno
);
4770 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4771 (unsigned long long)s_blkno
, (unsigned long long)t_blkno
,
4774 s_bucket
= ocfs2_xattr_bucket_new(inode
);
4775 t_bucket
= ocfs2_xattr_bucket_new(inode
);
4776 if (!s_bucket
|| !t_bucket
) {
4782 ret
= ocfs2_read_xattr_bucket(s_bucket
, s_blkno
);
4787 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4788 * there's no need to read it.
4790 ret
= ocfs2_init_xattr_bucket(t_bucket
, t_blkno
);
4795 * Hey, if we're overwriting t_bucket, what difference does
4796 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4797 * cluster to fill, we came here from
4798 * ocfs2_mv_xattr_buckets(), and it is really new -
4799 * ACCESS_CREATE is required. But we also might have moved data
4800 * out of t_bucket before extending back into it.
4801 * ocfs2_add_new_xattr_bucket() can do this - its call to
4802 * ocfs2_add_new_xattr_cluster() may have created a new extent
4803 * and copied out the end of the old extent. Then it re-extends
4804 * the old extent back to create space for new xattrs. That's
4805 * how we get here, and the bucket isn't really new.
4807 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
4809 OCFS2_JOURNAL_ACCESS_CREATE
:
4810 OCFS2_JOURNAL_ACCESS_WRITE
);
4814 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
4815 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
4818 ocfs2_xattr_bucket_free(t_bucket
);
4819 ocfs2_xattr_bucket_free(s_bucket
);
4825 * src_blk points to the start of an existing extent. last_blk points to
4826 * last cluster in that extent. to_blk points to a newly allocated
4827 * extent. We copy the buckets from the cluster at last_blk to the new
4828 * extent. If start_bucket is non-zero, we skip that many buckets before
4829 * we start copying. The new extent's xh_num_buckets gets set to the
4830 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4831 * by the same amount.
4833 static int ocfs2_mv_xattr_buckets(struct inode
*inode
, handle_t
*handle
,
4834 u64 src_blk
, u64 last_blk
, u64 to_blk
,
4835 unsigned int start_bucket
,
4838 int i
, ret
, credits
;
4839 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4840 int blks_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4841 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
4842 struct ocfs2_xattr_bucket
*old_first
, *new_first
;
4844 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4845 (unsigned long long)last_blk
, (unsigned long long)to_blk
);
4847 BUG_ON(start_bucket
>= num_buckets
);
4849 num_buckets
-= start_bucket
;
4850 last_blk
+= (start_bucket
* blks_per_bucket
);
4853 /* The first bucket of the original extent */
4854 old_first
= ocfs2_xattr_bucket_new(inode
);
4855 /* The first bucket of the new extent */
4856 new_first
= ocfs2_xattr_bucket_new(inode
);
4857 if (!old_first
|| !new_first
) {
4863 ret
= ocfs2_read_xattr_bucket(old_first
, src_blk
);
4870 * We need to update the first bucket of the old extent and all
4871 * the buckets going to the new extent.
4873 credits
= ((num_buckets
+ 1) * blks_per_bucket
);
4874 ret
= ocfs2_extend_trans(handle
, credits
);
4880 ret
= ocfs2_xattr_bucket_journal_access(handle
, old_first
,
4881 OCFS2_JOURNAL_ACCESS_WRITE
);
4887 for (i
= 0; i
< num_buckets
; i
++) {
4888 ret
= ocfs2_cp_xattr_bucket(inode
, handle
,
4889 last_blk
+ (i
* blks_per_bucket
),
4890 to_blk
+ (i
* blks_per_bucket
),
4899 * Get the new bucket ready before we dirty anything
4900 * (This actually shouldn't fail, because we already dirtied
4901 * it once in ocfs2_cp_xattr_bucket()).
4903 ret
= ocfs2_read_xattr_bucket(new_first
, to_blk
);
4908 ret
= ocfs2_xattr_bucket_journal_access(handle
, new_first
,
4909 OCFS2_JOURNAL_ACCESS_WRITE
);
4915 /* Now update the headers */
4916 le16_add_cpu(&bucket_xh(old_first
)->xh_num_buckets
, -num_buckets
);
4917 ocfs2_xattr_bucket_journal_dirty(handle
, old_first
);
4919 bucket_xh(new_first
)->xh_num_buckets
= cpu_to_le16(num_buckets
);
4920 ocfs2_xattr_bucket_journal_dirty(handle
, new_first
);
4923 *first_hash
= le32_to_cpu(bucket_xh(new_first
)->xh_entries
[0].xe_name_hash
);
4926 ocfs2_xattr_bucket_free(new_first
);
4927 ocfs2_xattr_bucket_free(old_first
);
4932 * Move some xattrs in this cluster to the new cluster.
4933 * This function should only be called when bucket size == cluster size.
4934 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4936 static int ocfs2_divide_xattr_cluster(struct inode
*inode
,
4942 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4943 int ret
, credits
= 2 * blk_per_bucket
;
4945 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
< OCFS2_SB(inode
->i_sb
)->s_clustersize
);
4947 ret
= ocfs2_extend_trans(handle
, credits
);
4953 /* Move half of the xattr in start_blk to the next bucket. */
4954 return ocfs2_divide_xattr_bucket(inode
, handle
, prev_blk
,
4955 new_blk
, first_hash
, 1);
4959 * Move some xattrs from the old cluster to the new one since they are not
4960 * contiguous in ocfs2 xattr tree.
4962 * new_blk starts a new separate cluster, and we will move some xattrs from
4963 * prev_blk to it. v_start will be set as the first name hash value in this
4964 * new cluster so that it can be used as e_cpos during tree insertion and
4965 * don't collide with our original b-tree operations. first_bh and header_bh
4966 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4967 * to extend the insert bucket.
4969 * The problem is how much xattr should we move to the new one and when should
4970 * we update first_bh and header_bh?
4971 * 1. If cluster size > bucket size, that means the previous cluster has more
4972 * than 1 bucket, so just move half nums of bucket into the new cluster and
4973 * update the first_bh and header_bh if the insert bucket has been moved
4974 * to the new cluster.
4975 * 2. If cluster_size == bucket_size:
4976 * a) If the previous extent rec has more than one cluster and the insert
4977 * place isn't in the last cluster, copy the entire last cluster to the
4978 * new one. This time, we don't need to upate the first_bh and header_bh
4979 * since they will not be moved into the new cluster.
4980 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4981 * the new one. And we set the extend flag to zero if the insert place is
4982 * moved into the new allocated cluster since no extend is needed.
4984 static int ocfs2_adjust_xattr_cross_cluster(struct inode
*inode
,
4986 struct ocfs2_xattr_bucket
*first
,
4987 struct ocfs2_xattr_bucket
*target
,
4995 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4996 (unsigned long long)bucket_blkno(first
), prev_clusters
,
4997 (unsigned long long)new_blk
);
4999 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
)) > 1) {
5000 ret
= ocfs2_mv_xattr_bucket_cross_cluster(inode
,
5009 /* The start of the last cluster in the first extent */
5010 u64 last_blk
= bucket_blkno(first
) +
5011 ((prev_clusters
- 1) *
5012 ocfs2_clusters_to_blocks(inode
->i_sb
, 1));
5014 if (prev_clusters
> 1 && bucket_blkno(target
) != last_blk
) {
5015 ret
= ocfs2_mv_xattr_buckets(inode
, handle
,
5016 bucket_blkno(first
),
5017 last_blk
, new_blk
, 0,
5022 ret
= ocfs2_divide_xattr_cluster(inode
, handle
,
5028 if ((bucket_blkno(target
) == last_blk
) && extend
)
5037 * Add a new cluster for xattr storage.
5039 * If the new cluster is contiguous with the previous one, it will be
5040 * appended to the same extent record, and num_clusters will be updated.
5041 * If not, we will insert a new extent for it and move some xattrs in
5042 * the last cluster into the new allocated one.
5043 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5044 * lose the benefits of hashing because we'll have to search large leaves.
5045 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5048 * first_bh is the first block of the previous extent rec and header_bh
5049 * indicates the bucket we will insert the new xattrs. They will be updated
5050 * when the header_bh is moved into the new cluster.
5052 static int ocfs2_add_new_xattr_cluster(struct inode
*inode
,
5053 struct buffer_head
*root_bh
,
5054 struct ocfs2_xattr_bucket
*first
,
5055 struct ocfs2_xattr_bucket
*target
,
5059 struct ocfs2_xattr_set_ctxt
*ctxt
)
5062 u16 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
5063 u32 prev_clusters
= *num_clusters
;
5064 u32 clusters_to_add
= 1, bit_off
, num_bits
, v_start
= 0;
5066 handle_t
*handle
= ctxt
->handle
;
5067 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5068 struct ocfs2_extent_tree et
;
5070 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
5071 "previous xattr blkno = %llu\n",
5072 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
5073 prev_cpos
, (unsigned long long)bucket_blkno(first
));
5075 ocfs2_init_xattr_tree_extent_tree(&et
, INODE_CACHE(inode
), root_bh
);
5077 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(inode
), root_bh
,
5078 OCFS2_JOURNAL_ACCESS_WRITE
);
5084 ret
= __ocfs2_claim_clusters(handle
, ctxt
->data_ac
, 1,
5085 clusters_to_add
, &bit_off
, &num_bits
);
5092 BUG_ON(num_bits
> clusters_to_add
);
5094 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
5095 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
5096 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
5098 if (bucket_blkno(first
) + (prev_clusters
* bpc
) == block
&&
5099 (prev_clusters
+ num_bits
) << osb
->s_clustersize_bits
<=
5100 OCFS2_MAX_XATTR_TREE_LEAF_SIZE
) {
5102 * If this cluster is contiguous with the old one and
5103 * adding this new cluster, we don't surpass the limit of
5104 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5105 * initialized and used like other buckets in the previous
5107 * So add it as a contiguous one. The caller will handle
5110 v_start
= prev_cpos
+ prev_clusters
;
5111 *num_clusters
= prev_clusters
+ num_bits
;
5112 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
5115 ret
= ocfs2_adjust_xattr_cross_cluster(inode
,
5129 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
5130 num_bits
, (unsigned long long)block
, v_start
);
5131 ret
= ocfs2_insert_extent(handle
, &et
, v_start
, block
,
5132 num_bits
, 0, ctxt
->meta_ac
);
5138 ocfs2_journal_dirty(handle
, root_bh
);
5145 * We are given an extent. 'first' is the bucket at the very front of
5146 * the extent. The extent has space for an additional bucket past
5147 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5148 * of the target bucket. We wish to shift every bucket past the target
5149 * down one, filling in that additional space. When we get back to the
5150 * target, we split the target between itself and the now-empty bucket
5151 * at target+1 (aka, target_blkno + blks_per_bucket).
5153 static int ocfs2_extend_xattr_bucket(struct inode
*inode
,
5155 struct ocfs2_xattr_bucket
*first
,
5160 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5161 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
5163 u16 new_bucket
= le16_to_cpu(bucket_xh(first
)->xh_num_buckets
);
5165 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
5166 "from %llu, len = %u\n", (unsigned long long)target_blk
,
5167 (unsigned long long)bucket_blkno(first
), num_clusters
);
5169 /* The extent must have room for an additional bucket */
5170 BUG_ON(new_bucket
>=
5171 (num_clusters
* ocfs2_xattr_buckets_per_cluster(osb
)));
5173 /* end_blk points to the last existing bucket */
5174 end_blk
= bucket_blkno(first
) + ((new_bucket
- 1) * blk_per_bucket
);
5177 * end_blk is the start of the last existing bucket.
5178 * Thus, (end_blk - target_blk) covers the target bucket and
5179 * every bucket after it up to, but not including, the last
5180 * existing bucket. Then we add the last existing bucket, the
5181 * new bucket, and the first bucket (3 * blk_per_bucket).
5183 credits
= (end_blk
- target_blk
) + (3 * blk_per_bucket
);
5184 ret
= ocfs2_extend_trans(handle
, credits
);
5190 ret
= ocfs2_xattr_bucket_journal_access(handle
, first
,
5191 OCFS2_JOURNAL_ACCESS_WRITE
);
5197 while (end_blk
!= target_blk
) {
5198 ret
= ocfs2_cp_xattr_bucket(inode
, handle
, end_blk
,
5199 end_blk
+ blk_per_bucket
, 0);
5202 end_blk
-= blk_per_bucket
;
5205 /* Move half of the xattr in target_blkno to the next bucket. */
5206 ret
= ocfs2_divide_xattr_bucket(inode
, handle
, target_blk
,
5207 target_blk
+ blk_per_bucket
, NULL
, 0);
5209 le16_add_cpu(&bucket_xh(first
)->xh_num_buckets
, 1);
5210 ocfs2_xattr_bucket_journal_dirty(handle
, first
);
5217 * Add new xattr bucket in an extent record and adjust the buckets
5218 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5219 * bucket we want to insert into.
5221 * In the easy case, we will move all the buckets after target down by
5222 * one. Half of target's xattrs will be moved to the next bucket.
5224 * If current cluster is full, we'll allocate a new one. This may not
5225 * be contiguous. The underlying calls will make sure that there is
5226 * space for the insert, shifting buckets around if necessary.
5227 * 'target' may be moved by those calls.
5229 static int ocfs2_add_new_xattr_bucket(struct inode
*inode
,
5230 struct buffer_head
*xb_bh
,
5231 struct ocfs2_xattr_bucket
*target
,
5232 struct ocfs2_xattr_set_ctxt
*ctxt
)
5234 struct ocfs2_xattr_block
*xb
=
5235 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
5236 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
5237 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
5239 le32_to_cpu(bucket_xh(target
)->xh_entries
[0].xe_name_hash
);
5240 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5241 int ret
, num_buckets
, extend
= 1;
5243 u32 e_cpos
, num_clusters
;
5244 /* The bucket at the front of the extent */
5245 struct ocfs2_xattr_bucket
*first
;
5247 mlog(0, "Add new xattr bucket starting from %llu\n",
5248 (unsigned long long)bucket_blkno(target
));
5250 /* The first bucket of the original extent */
5251 first
= ocfs2_xattr_bucket_new(inode
);
5258 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &e_cpos
,
5265 ret
= ocfs2_read_xattr_bucket(first
, p_blkno
);
5271 num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
) * num_clusters
;
5272 if (num_buckets
== le16_to_cpu(bucket_xh(first
)->xh_num_buckets
)) {
5274 * This can move first+target if the target bucket moves
5275 * to the new extent.
5277 ret
= ocfs2_add_new_xattr_cluster(inode
,
5292 ret
= ocfs2_extend_xattr_bucket(inode
,
5295 bucket_blkno(target
),
5302 ocfs2_xattr_bucket_free(first
);
5307 static inline char *ocfs2_xattr_bucket_get_val(struct inode
*inode
,
5308 struct ocfs2_xattr_bucket
*bucket
,
5311 int block_off
= offs
>> inode
->i_sb
->s_blocksize_bits
;
5313 offs
= offs
% inode
->i_sb
->s_blocksize
;
5314 return bucket_block(bucket
, block_off
) + offs
;
5318 * Truncate the specified xe_off entry in xattr bucket.
5319 * bucket is indicated by header_bh and len is the new length.
5320 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5322 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5324 static int ocfs2_xattr_bucket_value_truncate(struct inode
*inode
,
5325 struct ocfs2_xattr_bucket
*bucket
,
5328 struct ocfs2_xattr_set_ctxt
*ctxt
)
5332 struct ocfs2_xattr_entry
*xe
;
5333 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5334 size_t blocksize
= inode
->i_sb
->s_blocksize
;
5335 struct ocfs2_xattr_value_buf vb
= {
5336 .vb_access
= ocfs2_journal_access
,
5339 xe
= &xh
->xh_entries
[xe_off
];
5341 BUG_ON(!xe
|| ocfs2_xattr_is_local(xe
));
5343 offset
= le16_to_cpu(xe
->xe_name_offset
) +
5344 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
5346 value_blk
= offset
/ blocksize
;
5348 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5349 BUG_ON(value_blk
!= (offset
+ OCFS2_XATTR_ROOT_SIZE
- 1) / blocksize
);
5351 vb
.vb_bh
= bucket
->bu_bhs
[value_blk
];
5354 vb
.vb_xv
= (struct ocfs2_xattr_value_root
*)
5355 (vb
.vb_bh
->b_data
+ offset
% blocksize
);
5358 * From here on out we have to dirty the bucket. The generic
5359 * value calls only modify one of the bucket's bhs, but we need
5360 * to send the bucket at once. So if they error, they *could* have
5361 * modified something. We have to assume they did, and dirty
5362 * the whole bucket. This leaves us in a consistent state.
5364 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5365 xe_off
, (unsigned long long)bucket_blkno(bucket
), len
);
5366 ret
= ocfs2_xattr_value_truncate(inode
, &vb
, len
, ctxt
);
5372 ret
= ocfs2_xattr_bucket_journal_access(ctxt
->handle
, bucket
,
5373 OCFS2_JOURNAL_ACCESS_WRITE
);
5379 xe
->xe_value_size
= cpu_to_le64(len
);
5381 ocfs2_xattr_bucket_journal_dirty(ctxt
->handle
, bucket
);
5387 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
5388 struct buffer_head
*root_bh
,
5395 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5396 struct inode
*tl_inode
= osb
->osb_tl_inode
;
5398 struct ocfs2_xattr_block
*xb
=
5399 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
5400 struct ocfs2_alloc_context
*meta_ac
= NULL
;
5401 struct ocfs2_cached_dealloc_ctxt dealloc
;
5402 struct ocfs2_extent_tree et
;
5404 ret
= ocfs2_iterate_xattr_buckets(inode
, blkno
, len
,
5405 ocfs2_delete_xattr_in_bucket
, para
);
5411 ocfs2_init_xattr_tree_extent_tree(&et
, INODE_CACHE(inode
), root_bh
);
5413 ocfs2_init_dealloc_ctxt(&dealloc
);
5415 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5416 cpos
, len
, (unsigned long long)blkno
);
5418 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode
), blkno
,
5421 ret
= ocfs2_lock_allocators(inode
, &et
, 0, 1, NULL
, &meta_ac
);
5427 mutex_lock(&tl_inode
->i_mutex
);
5429 if (ocfs2_truncate_log_needs_flush(osb
)) {
5430 ret
= __ocfs2_flush_truncate_log(osb
);
5437 handle
= ocfs2_start_trans(osb
, ocfs2_remove_extent_credits(osb
->sb
));
5438 if (IS_ERR(handle
)) {
5444 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(inode
), root_bh
,
5445 OCFS2_JOURNAL_ACCESS_WRITE
);
5451 ret
= ocfs2_remove_extent(handle
, &et
, cpos
, len
, meta_ac
,
5458 le32_add_cpu(&xb
->xb_attrs
.xb_root
.xt_clusters
, -len
);
5459 ocfs2_journal_dirty(handle
, root_bh
);
5461 ret
= ocfs2_truncate_log_append(osb
, handle
, blkno
, len
);
5466 ocfs2_commit_trans(osb
, handle
);
5468 ocfs2_schedule_truncate_log_flush(osb
, 1);
5470 mutex_unlock(&tl_inode
->i_mutex
);
5473 ocfs2_free_alloc_context(meta_ac
);
5475 ocfs2_run_deallocs(osb
, &dealloc
);
5481 * check whether the xattr bucket is filled up with the same hash value.
5482 * If we want to insert the xattr with the same hash, return -ENOSPC.
5483 * If we want to insert a xattr with different hash value, go ahead
5484 * and ocfs2_divide_xattr_bucket will handle this.
5486 static int ocfs2_check_xattr_bucket_collision(struct inode
*inode
,
5487 struct ocfs2_xattr_bucket
*bucket
,
5490 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5491 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
5493 if (name_hash
!= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
))
5496 if (xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1].xe_name_hash
==
5497 xh
->xh_entries
[0].xe_name_hash
) {
5498 mlog(ML_ERROR
, "Too much hash collision in xattr bucket %llu, "
5500 (unsigned long long)bucket_blkno(bucket
),
5501 le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
));
5509 * Try to set the entry in the current bucket. If we fail, the caller
5510 * will handle getting us another bucket.
5512 static int ocfs2_xattr_set_entry_bucket(struct inode
*inode
,
5513 struct ocfs2_xattr_info
*xi
,
5514 struct ocfs2_xattr_search
*xs
,
5515 struct ocfs2_xattr_set_ctxt
*ctxt
)
5518 struct ocfs2_xa_loc loc
;
5520 mlog_entry("Set xattr %s in xattr bucket\n", xi
->xi_name
);
5522 ocfs2_init_xattr_bucket_xa_loc(&loc
, xs
->bucket
,
5523 xs
->not_found
? NULL
: xs
->here
);
5524 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
5526 xs
->here
= loc
.xl_entry
;
5529 if (ret
!= -ENOSPC
) {
5534 /* Ok, we need space. Let's try defragmenting the bucket. */
5535 ret
= ocfs2_defrag_xattr_bucket(inode
, ctxt
->handle
,
5542 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
5544 xs
->here
= loc
.xl_entry
;
5556 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
5557 struct ocfs2_xattr_info
*xi
,
5558 struct ocfs2_xattr_search
*xs
,
5559 struct ocfs2_xattr_set_ctxt
*ctxt
)
5563 mlog_entry("Set xattr %s in xattr index block\n", xi
->xi_name
);
5565 ret
= ocfs2_xattr_set_entry_bucket(inode
, xi
, xs
, ctxt
);
5568 if (ret
!= -ENOSPC
) {
5573 /* Ack, need more space. Let's try to get another bucket! */
5576 * We do not allow for overlapping ranges between buckets. And
5577 * the maximum number of collisions we will allow for then is
5578 * one bucket's worth, so check it here whether we need to
5579 * add a new bucket for the insert.
5581 ret
= ocfs2_check_xattr_bucket_collision(inode
,
5589 ret
= ocfs2_add_new_xattr_bucket(inode
,
5599 * ocfs2_add_new_xattr_bucket() will have updated
5600 * xs->bucket if it moved, but it will not have updated
5601 * any of the other search fields. Thus, we drop it and
5602 * re-search. Everything should be cached, so it'll be
5605 ocfs2_xattr_bucket_relse(xs
->bucket
);
5606 ret
= ocfs2_xattr_index_block_find(inode
, xs
->xattr_bh
,
5609 if (ret
&& ret
!= -ENODATA
)
5611 xs
->not_found
= ret
;
5613 /* Ok, we have a new bucket, let's try again */
5614 ret
= ocfs2_xattr_set_entry_bucket(inode
, xi
, xs
, ctxt
);
5615 if (ret
&& (ret
!= -ENOSPC
))
5623 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
5624 struct ocfs2_xattr_bucket
*bucket
,
5627 int ret
= 0, ref_credits
;
5628 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5630 struct ocfs2_xattr_entry
*xe
;
5631 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5632 struct ocfs2_xattr_set_ctxt ctxt
= {NULL
, NULL
,};
5633 int credits
= ocfs2_remove_extent_credits(osb
->sb
) +
5634 ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
5635 struct ocfs2_xattr_value_root
*xv
;
5636 struct ocfs2_rm_xattr_bucket_para
*args
=
5637 (struct ocfs2_rm_xattr_bucket_para
*)para
;
5639 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
5641 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
5642 xe
= &xh
->xh_entries
[i
];
5643 if (ocfs2_xattr_is_local(xe
))
5646 ret
= ocfs2_get_xattr_tree_value_root(inode
->i_sb
, bucket
,
5649 ret
= ocfs2_lock_xattr_remove_allocators(inode
, xv
,
5655 ctxt
.handle
= ocfs2_start_trans(osb
, credits
+ ref_credits
);
5656 if (IS_ERR(ctxt
.handle
)) {
5657 ret
= PTR_ERR(ctxt
.handle
);
5662 ret
= ocfs2_xattr_bucket_value_truncate(inode
, bucket
,
5665 ocfs2_commit_trans(osb
, ctxt
.handle
);
5667 ocfs2_free_alloc_context(ctxt
.meta_ac
);
5668 ctxt
.meta_ac
= NULL
;
5677 ocfs2_free_alloc_context(ctxt
.meta_ac
);
5678 ocfs2_schedule_truncate_log_flush(osb
, 1);
5679 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
5684 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5685 * or change the extent record flag), we need to recalculate
5686 * the metaecc for the whole bucket. So it is done here.
5689 * We have to give the extra credits for the caller.
5691 static int ocfs2_xattr_bucket_post_refcount(struct inode
*inode
,
5696 struct ocfs2_xattr_bucket
*bucket
=
5697 (struct ocfs2_xattr_bucket
*)para
;
5699 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
5700 OCFS2_JOURNAL_ACCESS_WRITE
);
5706 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
5712 * Special action we need if the xattr value is refcounted.
5714 * 1. If the xattr is refcounted, lock the tree.
5715 * 2. CoW the xattr if we are setting the new value and the value
5716 * will be stored outside.
5717 * 3. In other case, decrease_refcount will work for us, so just
5718 * lock the refcount tree, calculate the meta and credits is OK.
5720 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5721 * currently CoW is a completed transaction, while this function
5722 * will also lock the allocators and let us deadlock. So we will
5723 * CoW the whole xattr value.
5725 static int ocfs2_prepare_refcount_xattr(struct inode
*inode
,
5726 struct ocfs2_dinode
*di
,
5727 struct ocfs2_xattr_info
*xi
,
5728 struct ocfs2_xattr_search
*xis
,
5729 struct ocfs2_xattr_search
*xbs
,
5730 struct ocfs2_refcount_tree
**ref_tree
,
5735 struct ocfs2_xattr_block
*xb
;
5736 struct ocfs2_xattr_entry
*xe
;
5738 u32 p_cluster
, num_clusters
;
5739 unsigned int ext_flags
;
5740 int name_offset
, name_len
;
5741 struct ocfs2_xattr_value_buf vb
;
5742 struct ocfs2_xattr_bucket
*bucket
= NULL
;
5743 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5744 struct ocfs2_post_refcount refcount
;
5745 struct ocfs2_post_refcount
*p
= NULL
;
5746 struct buffer_head
*ref_root_bh
= NULL
;
5748 if (!xis
->not_found
) {
5750 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
5751 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
5753 vb
.vb_bh
= xis
->inode_bh
;
5754 vb
.vb_access
= ocfs2_journal_access_di
;
5756 int i
, block_off
= 0;
5757 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
5759 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
5760 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
5761 i
= xbs
->here
- xbs
->header
->xh_entries
;
5763 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
5764 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
5765 bucket_xh(xbs
->bucket
),
5772 base
= bucket_block(xbs
->bucket
, block_off
);
5773 vb
.vb_bh
= xbs
->bucket
->bu_bhs
[block_off
];
5774 vb
.vb_access
= ocfs2_journal_access
;
5776 if (ocfs2_meta_ecc(osb
)) {
5777 /*create parameters for ocfs2_post_refcount. */
5778 bucket
= xbs
->bucket
;
5779 refcount
.credits
= bucket
->bu_blocks
;
5780 refcount
.para
= bucket
;
5782 ocfs2_xattr_bucket_post_refcount
;
5787 vb
.vb_bh
= xbs
->xattr_bh
;
5788 vb
.vb_access
= ocfs2_journal_access_xb
;
5792 if (ocfs2_xattr_is_local(xe
))
5795 vb
.vb_xv
= (struct ocfs2_xattr_value_root
*)
5796 (base
+ name_offset
+ name_len
);
5798 ret
= ocfs2_xattr_get_clusters(inode
, 0, &p_cluster
,
5799 &num_clusters
, &vb
.vb_xv
->xr_list
,
5807 * We just need to check the 1st extent record, since we always
5808 * CoW the whole xattr. So there shouldn't be a xattr with
5809 * some REFCOUNT extent recs after the 1st one.
5811 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
5814 ret
= ocfs2_lock_refcount_tree(osb
, le64_to_cpu(di
->i_refcount_loc
),
5815 1, ref_tree
, &ref_root_bh
);
5822 * If we are deleting the xattr or the new size will be stored inside,
5823 * cool, leave it there, the xattr truncate process will remove them
5824 * for us(it still needs the refcount tree lock and the meta, credits).
5825 * And the worse case is that every cluster truncate will split the
5826 * refcount tree, and make the original extent become 3. So we will need
5827 * 2 * cluster more extent recs at most.
5829 if (!xi
->xi_value
|| xi
->xi_value_len
<= OCFS2_XATTR_INLINE_SIZE
) {
5831 ret
= ocfs2_refcounted_xattr_delete_need(inode
,
5832 &(*ref_tree
)->rf_ci
,
5833 ref_root_bh
, vb
.vb_xv
,
5840 ret
= ocfs2_refcount_cow_xattr(inode
, di
, &vb
,
5841 *ref_tree
, ref_root_bh
, 0,
5842 le32_to_cpu(vb
.vb_xv
->xr_clusters
), p
);
5847 brelse(ref_root_bh
);
5852 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5853 * The physical clusters will be added to refcount tree.
5855 static int ocfs2_xattr_value_attach_refcount(struct inode
*inode
,
5856 struct ocfs2_xattr_value_root
*xv
,
5857 struct ocfs2_extent_tree
*value_et
,
5858 struct ocfs2_caching_info
*ref_ci
,
5859 struct buffer_head
*ref_root_bh
,
5860 struct ocfs2_cached_dealloc_ctxt
*dealloc
,
5861 struct ocfs2_post_refcount
*refcount
)
5864 u32 clusters
= le32_to_cpu(xv
->xr_clusters
);
5865 u32 cpos
, p_cluster
, num_clusters
;
5866 struct ocfs2_extent_list
*el
= &xv
->xr_list
;
5867 unsigned int ext_flags
;
5870 while (cpos
< clusters
) {
5871 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
5872 &num_clusters
, el
, &ext_flags
);
5874 cpos
+= num_clusters
;
5875 if ((ext_flags
& OCFS2_EXT_REFCOUNTED
))
5880 ret
= ocfs2_add_refcount_flag(inode
, value_et
,
5881 ref_ci
, ref_root_bh
,
5882 cpos
- num_clusters
,
5883 p_cluster
, num_clusters
,
5895 * Given a normal ocfs2_xattr_header, refcount all the entries which
5896 * have value stored outside.
5897 * Used for xattrs stored in inode and ocfs2_xattr_block.
5899 static int ocfs2_xattr_attach_refcount_normal(struct inode
*inode
,
5900 struct ocfs2_xattr_value_buf
*vb
,
5901 struct ocfs2_xattr_header
*header
,
5902 struct ocfs2_caching_info
*ref_ci
,
5903 struct buffer_head
*ref_root_bh
,
5904 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
5907 struct ocfs2_xattr_entry
*xe
;
5908 struct ocfs2_xattr_value_root
*xv
;
5909 struct ocfs2_extent_tree et
;
5912 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
5913 xe
= &header
->xh_entries
[i
];
5915 if (ocfs2_xattr_is_local(xe
))
5918 xv
= (struct ocfs2_xattr_value_root
*)((void *)header
+
5919 le16_to_cpu(xe
->xe_name_offset
) +
5920 OCFS2_XATTR_SIZE(xe
->xe_name_len
));
5923 ocfs2_init_xattr_value_extent_tree(&et
, INODE_CACHE(inode
), vb
);
5925 ret
= ocfs2_xattr_value_attach_refcount(inode
, xv
, &et
,
5926 ref_ci
, ref_root_bh
,
5937 static int ocfs2_xattr_inline_attach_refcount(struct inode
*inode
,
5938 struct buffer_head
*fe_bh
,
5939 struct ocfs2_caching_info
*ref_ci
,
5940 struct buffer_head
*ref_root_bh
,
5941 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
5943 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)fe_bh
->b_data
;
5944 struct ocfs2_xattr_header
*header
= (struct ocfs2_xattr_header
*)
5945 (fe_bh
->b_data
+ inode
->i_sb
->s_blocksize
-
5946 le16_to_cpu(di
->i_xattr_inline_size
));
5947 struct ocfs2_xattr_value_buf vb
= {
5949 .vb_access
= ocfs2_journal_access_di
,
5952 return ocfs2_xattr_attach_refcount_normal(inode
, &vb
, header
,
5953 ref_ci
, ref_root_bh
, dealloc
);
5956 struct ocfs2_xattr_tree_value_refcount_para
{
5957 struct ocfs2_caching_info
*ref_ci
;
5958 struct buffer_head
*ref_root_bh
;
5959 struct ocfs2_cached_dealloc_ctxt
*dealloc
;
5962 static int ocfs2_get_xattr_tree_value_root(struct super_block
*sb
,
5963 struct ocfs2_xattr_bucket
*bucket
,
5965 struct ocfs2_xattr_value_root
**xv
,
5966 struct buffer_head
**bh
)
5968 int ret
, block_off
, name_offset
;
5969 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5970 struct ocfs2_xattr_entry
*xe
= &xh
->xh_entries
[offset
];
5973 ret
= ocfs2_xattr_bucket_get_name_value(sb
,
5983 base
= bucket_block(bucket
, block_off
);
5985 *xv
= (struct ocfs2_xattr_value_root
*)(base
+ name_offset
+
5986 OCFS2_XATTR_SIZE(xe
->xe_name_len
));
5989 *bh
= bucket
->bu_bhs
[block_off
];
5995 * For a given xattr bucket, refcount all the entries which
5996 * have value stored outside.
5998 static int ocfs2_xattr_bucket_value_refcount(struct inode
*inode
,
5999 struct ocfs2_xattr_bucket
*bucket
,
6003 struct ocfs2_extent_tree et
;
6004 struct ocfs2_xattr_tree_value_refcount_para
*ref
=
6005 (struct ocfs2_xattr_tree_value_refcount_para
*)para
;
6006 struct ocfs2_xattr_header
*xh
=
6007 (struct ocfs2_xattr_header
*)bucket
->bu_bhs
[0]->b_data
;
6008 struct ocfs2_xattr_entry
*xe
;
6009 struct ocfs2_xattr_value_buf vb
= {
6010 .vb_access
= ocfs2_journal_access
,
6012 struct ocfs2_post_refcount refcount
= {
6013 .credits
= bucket
->bu_blocks
,
6015 .func
= ocfs2_xattr_bucket_post_refcount
,
6017 struct ocfs2_post_refcount
*p
= NULL
;
6019 /* We only need post_refcount if we support metaecc. */
6020 if (ocfs2_meta_ecc(OCFS2_SB(inode
->i_sb
)))
6023 mlog(0, "refcount bucket %llu, count = %u\n",
6024 (unsigned long long)bucket_blkno(bucket
),
6025 le16_to_cpu(xh
->xh_count
));
6026 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
6027 xe
= &xh
->xh_entries
[i
];
6029 if (ocfs2_xattr_is_local(xe
))
6032 ret
= ocfs2_get_xattr_tree_value_root(inode
->i_sb
, bucket
, i
,
6033 &vb
.vb_xv
, &vb
.vb_bh
);
6039 ocfs2_init_xattr_value_extent_tree(&et
,
6040 INODE_CACHE(inode
), &vb
);
6042 ret
= ocfs2_xattr_value_attach_refcount(inode
, vb
.vb_xv
,
6056 static int ocfs2_refcount_xattr_tree_rec(struct inode
*inode
,
6057 struct buffer_head
*root_bh
,
6058 u64 blkno
, u32 cpos
, u32 len
, void *para
)
6060 return ocfs2_iterate_xattr_buckets(inode
, blkno
, len
,
6061 ocfs2_xattr_bucket_value_refcount
,
6065 static int ocfs2_xattr_block_attach_refcount(struct inode
*inode
,
6066 struct buffer_head
*blk_bh
,
6067 struct ocfs2_caching_info
*ref_ci
,
6068 struct buffer_head
*ref_root_bh
,
6069 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
6072 struct ocfs2_xattr_block
*xb
=
6073 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
6075 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
6076 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
6077 struct ocfs2_xattr_value_buf vb
= {
6079 .vb_access
= ocfs2_journal_access_xb
,
6082 ret
= ocfs2_xattr_attach_refcount_normal(inode
, &vb
, header
,
6083 ref_ci
, ref_root_bh
,
6086 struct ocfs2_xattr_tree_value_refcount_para para
= {
6088 .ref_root_bh
= ref_root_bh
,
6092 ret
= ocfs2_iterate_xattr_index_block(inode
, blk_bh
,
6093 ocfs2_refcount_xattr_tree_rec
,
6100 int ocfs2_xattr_attach_refcount_tree(struct inode
*inode
,
6101 struct buffer_head
*fe_bh
,
6102 struct ocfs2_caching_info
*ref_ci
,
6103 struct buffer_head
*ref_root_bh
,
6104 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
6107 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
6108 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)fe_bh
->b_data
;
6109 struct buffer_head
*blk_bh
= NULL
;
6111 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
6112 ret
= ocfs2_xattr_inline_attach_refcount(inode
, fe_bh
,
6113 ref_ci
, ref_root_bh
,
6121 if (!di
->i_xattr_loc
)
6124 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
6131 ret
= ocfs2_xattr_block_attach_refcount(inode
, blk_bh
, ref_ci
,
6132 ref_root_bh
, dealloc
);
6142 typedef int (should_xattr_reflinked
)(struct ocfs2_xattr_entry
*xe
);
6144 * Store the information we need in xattr reflink.
6145 * old_bh and new_bh are inode bh for the old and new inode.
6147 struct ocfs2_xattr_reflink
{
6148 struct inode
*old_inode
;
6149 struct inode
*new_inode
;
6150 struct buffer_head
*old_bh
;
6151 struct buffer_head
*new_bh
;
6152 struct ocfs2_caching_info
*ref_ci
;
6153 struct buffer_head
*ref_root_bh
;
6154 struct ocfs2_cached_dealloc_ctxt
*dealloc
;
6155 should_xattr_reflinked
*xattr_reflinked
;
6159 * Given a xattr header and xe offset,
6160 * return the proper xv and the corresponding bh.
6161 * xattr in inode, block and xattr tree have different implementaions.
6163 typedef int (get_xattr_value_root
)(struct super_block
*sb
,
6164 struct buffer_head
*bh
,
6165 struct ocfs2_xattr_header
*xh
,
6167 struct ocfs2_xattr_value_root
**xv
,
6168 struct buffer_head
**ret_bh
,
6172 * Calculate all the xattr value root metadata stored in this xattr header and
6173 * credits we need if we create them from the scratch.
6174 * We use get_xattr_value_root so that all types of xattr container can use it.
6176 static int ocfs2_value_metas_in_xattr_header(struct super_block
*sb
,
6177 struct buffer_head
*bh
,
6178 struct ocfs2_xattr_header
*xh
,
6179 int *metas
, int *credits
,
6181 get_xattr_value_root
*func
,
6185 struct ocfs2_xattr_value_root
*xv
;
6186 struct ocfs2_xattr_entry
*xe
;
6188 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
6189 xe
= &xh
->xh_entries
[i
];
6190 if (ocfs2_xattr_is_local(xe
))
6193 ret
= func(sb
, bh
, xh
, i
, &xv
, NULL
, para
);
6199 *metas
+= le16_to_cpu(xv
->xr_list
.l_tree_depth
) *
6200 le16_to_cpu(xv
->xr_list
.l_next_free_rec
);
6202 *credits
+= ocfs2_calc_extend_credits(sb
,
6204 le32_to_cpu(xv
->xr_clusters
));
6207 * If the value is a tree with depth > 1, We don't go deep
6208 * to the extent block, so just calculate a maximum record num.
6210 if (!xv
->xr_list
.l_tree_depth
)
6211 *num_recs
+= le16_to_cpu(xv
->xr_list
.l_next_free_rec
);
6213 *num_recs
+= ocfs2_clusters_for_bytes(sb
,
6220 /* Used by xattr inode and block to return the right xv and buffer_head. */
6221 static int ocfs2_get_xattr_value_root(struct super_block
*sb
,
6222 struct buffer_head
*bh
,
6223 struct ocfs2_xattr_header
*xh
,
6225 struct ocfs2_xattr_value_root
**xv
,
6226 struct buffer_head
**ret_bh
,
6229 struct ocfs2_xattr_entry
*xe
= &xh
->xh_entries
[offset
];
6231 *xv
= (struct ocfs2_xattr_value_root
*)((void *)xh
+
6232 le16_to_cpu(xe
->xe_name_offset
) +
6233 OCFS2_XATTR_SIZE(xe
->xe_name_len
));
6242 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6243 * It is only used for inline xattr and xattr block.
6245 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super
*osb
,
6246 struct ocfs2_xattr_header
*xh
,
6247 struct buffer_head
*ref_root_bh
,
6249 struct ocfs2_alloc_context
**meta_ac
)
6251 int ret
, meta_add
= 0, num_recs
= 0;
6252 struct ocfs2_refcount_block
*rb
=
6253 (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
6257 ret
= ocfs2_value_metas_in_xattr_header(osb
->sb
, NULL
, xh
,
6258 &meta_add
, credits
, &num_recs
,
6259 ocfs2_get_xattr_value_root
,
6267 * We need to add/modify num_recs in refcount tree, so just calculate
6268 * an approximate number we need for refcount tree change.
6269 * Sometimes we need to split the tree, and after split, half recs
6270 * will be moved to the new block, and a new block can only provide
6271 * half number of recs. So we multiple new blocks by 2.
6273 num_recs
= num_recs
/ ocfs2_refcount_recs_per_rb(osb
->sb
) * 2;
6274 meta_add
+= num_recs
;
6275 *credits
+= num_recs
+ num_recs
* OCFS2_EXPAND_REFCOUNT_TREE_CREDITS
;
6276 if (le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
)
6277 *credits
+= le16_to_cpu(rb
->rf_list
.l_tree_depth
) *
6278 le16_to_cpu(rb
->rf_list
.l_next_free_rec
) + 1;
6282 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
, meta_ac
);
6291 * Given a xattr header, reflink all the xattrs in this container.
6292 * It can be used for inode, block and bucket.
6295 * Before we call this function, the caller has memcpy the xattr in
6296 * old_xh to the new_xh.
6298 * If args.xattr_reflinked is set, call it to decide whether the xe should
6299 * be reflinked or not. If not, remove it from the new xattr header.
6301 static int ocfs2_reflink_xattr_header(handle_t
*handle
,
6302 struct ocfs2_xattr_reflink
*args
,
6303 struct buffer_head
*old_bh
,
6304 struct ocfs2_xattr_header
*xh
,
6305 struct buffer_head
*new_bh
,
6306 struct ocfs2_xattr_header
*new_xh
,
6307 struct ocfs2_xattr_value_buf
*vb
,
6308 struct ocfs2_alloc_context
*meta_ac
,
6309 get_xattr_value_root
*func
,
6313 struct super_block
*sb
= args
->old_inode
->i_sb
;
6314 struct buffer_head
*value_bh
;
6315 struct ocfs2_xattr_entry
*xe
, *last
;
6316 struct ocfs2_xattr_value_root
*xv
, *new_xv
;
6317 struct ocfs2_extent_tree data_et
;
6318 u32 clusters
, cpos
, p_cluster
, num_clusters
;
6319 unsigned int ext_flags
= 0;
6321 mlog(0, "reflink xattr in container %llu, count = %u\n",
6322 (unsigned long long)old_bh
->b_blocknr
, le16_to_cpu(xh
->xh_count
));
6324 last
= &new_xh
->xh_entries
[le16_to_cpu(new_xh
->xh_count
)];
6325 for (i
= 0, j
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, j
++) {
6326 xe
= &xh
->xh_entries
[i
];
6328 if (args
->xattr_reflinked
&& !args
->xattr_reflinked(xe
)) {
6329 xe
= &new_xh
->xh_entries
[j
];
6331 le16_add_cpu(&new_xh
->xh_count
, -1);
6332 if (new_xh
->xh_count
) {
6334 (void *)last
- (void *)xe
);
6336 sizeof(struct ocfs2_xattr_entry
));
6340 * We don't want j to increase in the next round since
6341 * it is already moved ahead.
6347 if (ocfs2_xattr_is_local(xe
))
6350 ret
= func(sb
, old_bh
, xh
, i
, &xv
, NULL
, para
);
6356 ret
= func(sb
, new_bh
, new_xh
, j
, &new_xv
, &value_bh
, para
);
6363 * For the xattr which has l_tree_depth = 0, all the extent
6364 * recs have already be copied to the new xh with the
6365 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6366 * increase the refount count int the refcount tree.
6368 * For the xattr which has l_tree_depth > 0, we need
6369 * to initialize it to the empty default value root,
6370 * and then insert the extents one by one.
6372 if (xv
->xr_list
.l_tree_depth
) {
6373 memcpy(new_xv
, &def_xv
, sizeof(def_xv
));
6375 vb
->vb_bh
= value_bh
;
6376 ocfs2_init_xattr_value_extent_tree(&data_et
,
6377 INODE_CACHE(args
->new_inode
), vb
);
6380 clusters
= le32_to_cpu(xv
->xr_clusters
);
6382 while (cpos
< clusters
) {
6383 ret
= ocfs2_xattr_get_clusters(args
->old_inode
,
6396 if (xv
->xr_list
.l_tree_depth
) {
6397 ret
= ocfs2_insert_extent(handle
,
6399 ocfs2_clusters_to_blocks(
6400 args
->old_inode
->i_sb
,
6402 num_clusters
, ext_flags
,
6410 ret
= ocfs2_increase_refcount(handle
, args
->ref_ci
,
6412 p_cluster
, num_clusters
,
6413 meta_ac
, args
->dealloc
);
6419 cpos
+= num_clusters
;
6427 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink
*args
)
6429 int ret
= 0, credits
= 0;
6431 struct ocfs2_super
*osb
= OCFS2_SB(args
->old_inode
->i_sb
);
6432 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)args
->old_bh
->b_data
;
6433 int inline_size
= le16_to_cpu(di
->i_xattr_inline_size
);
6434 int header_off
= osb
->sb
->s_blocksize
- inline_size
;
6435 struct ocfs2_xattr_header
*xh
= (struct ocfs2_xattr_header
*)
6436 (args
->old_bh
->b_data
+ header_off
);
6437 struct ocfs2_xattr_header
*new_xh
= (struct ocfs2_xattr_header
*)
6438 (args
->new_bh
->b_data
+ header_off
);
6439 struct ocfs2_alloc_context
*meta_ac
= NULL
;
6440 struct ocfs2_inode_info
*new_oi
;
6441 struct ocfs2_dinode
*new_di
;
6442 struct ocfs2_xattr_value_buf vb
= {
6443 .vb_bh
= args
->new_bh
,
6444 .vb_access
= ocfs2_journal_access_di
,
6447 ret
= ocfs2_reflink_lock_xattr_allocators(osb
, xh
, args
->ref_root_bh
,
6448 &credits
, &meta_ac
);
6454 handle
= ocfs2_start_trans(osb
, credits
);
6455 if (IS_ERR(handle
)) {
6456 ret
= PTR_ERR(handle
);
6461 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(args
->new_inode
),
6462 args
->new_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
6468 memcpy(args
->new_bh
->b_data
+ header_off
,
6469 args
->old_bh
->b_data
+ header_off
, inline_size
);
6471 new_di
= (struct ocfs2_dinode
*)args
->new_bh
->b_data
;
6472 new_di
->i_xattr_inline_size
= cpu_to_le16(inline_size
);
6474 ret
= ocfs2_reflink_xattr_header(handle
, args
, args
->old_bh
, xh
,
6475 args
->new_bh
, new_xh
, &vb
, meta_ac
,
6476 ocfs2_get_xattr_value_root
, NULL
);
6482 new_oi
= OCFS2_I(args
->new_inode
);
6483 spin_lock(&new_oi
->ip_lock
);
6484 new_oi
->ip_dyn_features
|= OCFS2_HAS_XATTR_FL
| OCFS2_INLINE_XATTR_FL
;
6485 new_di
->i_dyn_features
= cpu_to_le16(new_oi
->ip_dyn_features
);
6486 spin_unlock(&new_oi
->ip_lock
);
6488 ocfs2_journal_dirty(handle
, args
->new_bh
);
6491 ocfs2_commit_trans(osb
, handle
);
6495 ocfs2_free_alloc_context(meta_ac
);
6499 static int ocfs2_create_empty_xattr_block(struct inode
*inode
,
6500 struct buffer_head
*fe_bh
,
6501 struct buffer_head
**ret_bh
,
6505 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
6506 struct ocfs2_xattr_set_ctxt ctxt
;
6508 memset(&ctxt
, 0, sizeof(ctxt
));
6509 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, &ctxt
.meta_ac
);
6515 ctxt
.handle
= ocfs2_start_trans(osb
, OCFS2_XATTR_BLOCK_CREATE_CREDITS
);
6516 if (IS_ERR(ctxt
.handle
)) {
6517 ret
= PTR_ERR(ctxt
.handle
);
6522 mlog(0, "create new xattr block for inode %llu, index = %d\n",
6523 (unsigned long long)fe_bh
->b_blocknr
, indexed
);
6524 ret
= ocfs2_create_xattr_block(inode
, fe_bh
, &ctxt
, indexed
,
6529 ocfs2_commit_trans(osb
, ctxt
.handle
);
6531 ocfs2_free_alloc_context(ctxt
.meta_ac
);
6535 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink
*args
,
6536 struct buffer_head
*blk_bh
,
6537 struct buffer_head
*new_blk_bh
)
6539 int ret
= 0, credits
= 0;
6541 struct ocfs2_inode_info
*new_oi
= OCFS2_I(args
->new_inode
);
6542 struct ocfs2_dinode
*new_di
;
6543 struct ocfs2_super
*osb
= OCFS2_SB(args
->new_inode
->i_sb
);
6544 int header_off
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
6545 struct ocfs2_xattr_block
*xb
=
6546 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
6547 struct ocfs2_xattr_header
*xh
= &xb
->xb_attrs
.xb_header
;
6548 struct ocfs2_xattr_block
*new_xb
=
6549 (struct ocfs2_xattr_block
*)new_blk_bh
->b_data
;
6550 struct ocfs2_xattr_header
*new_xh
= &new_xb
->xb_attrs
.xb_header
;
6551 struct ocfs2_alloc_context
*meta_ac
;
6552 struct ocfs2_xattr_value_buf vb
= {
6553 .vb_bh
= new_blk_bh
,
6554 .vb_access
= ocfs2_journal_access_xb
,
6557 ret
= ocfs2_reflink_lock_xattr_allocators(osb
, xh
, args
->ref_root_bh
,
6558 &credits
, &meta_ac
);
6564 /* One more credits in case we need to add xattr flags in new inode. */
6565 handle
= ocfs2_start_trans(osb
, credits
+ 1);
6566 if (IS_ERR(handle
)) {
6567 ret
= PTR_ERR(handle
);
6572 if (!(new_oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
)) {
6573 ret
= ocfs2_journal_access_di(handle
,
6574 INODE_CACHE(args
->new_inode
),
6576 OCFS2_JOURNAL_ACCESS_WRITE
);
6583 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(args
->new_inode
),
6584 new_blk_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
6590 memcpy(new_blk_bh
->b_data
+ header_off
, blk_bh
->b_data
+ header_off
,
6591 osb
->sb
->s_blocksize
- header_off
);
6593 ret
= ocfs2_reflink_xattr_header(handle
, args
, blk_bh
, xh
,
6594 new_blk_bh
, new_xh
, &vb
, meta_ac
,
6595 ocfs2_get_xattr_value_root
, NULL
);
6601 ocfs2_journal_dirty(handle
, new_blk_bh
);
6603 if (!(new_oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
)) {
6604 new_di
= (struct ocfs2_dinode
*)args
->new_bh
->b_data
;
6605 spin_lock(&new_oi
->ip_lock
);
6606 new_oi
->ip_dyn_features
|= OCFS2_HAS_XATTR_FL
;
6607 new_di
->i_dyn_features
= cpu_to_le16(new_oi
->ip_dyn_features
);
6608 spin_unlock(&new_oi
->ip_lock
);
6610 ocfs2_journal_dirty(handle
, args
->new_bh
);
6614 ocfs2_commit_trans(osb
, handle
);
6617 ocfs2_free_alloc_context(meta_ac
);
6621 struct ocfs2_reflink_xattr_tree_args
{
6622 struct ocfs2_xattr_reflink
*reflink
;
6623 struct buffer_head
*old_blk_bh
;
6624 struct buffer_head
*new_blk_bh
;
6625 struct ocfs2_xattr_bucket
*old_bucket
;
6626 struct ocfs2_xattr_bucket
*new_bucket
;
6631 * We have to handle the case that both old bucket and new bucket
6632 * will call this function to get the right ret_bh.
6633 * So The caller must give us the right bh.
6635 static int ocfs2_get_reflink_xattr_value_root(struct super_block
*sb
,
6636 struct buffer_head
*bh
,
6637 struct ocfs2_xattr_header
*xh
,
6639 struct ocfs2_xattr_value_root
**xv
,
6640 struct buffer_head
**ret_bh
,
6643 struct ocfs2_reflink_xattr_tree_args
*args
=
6644 (struct ocfs2_reflink_xattr_tree_args
*)para
;
6645 struct ocfs2_xattr_bucket
*bucket
;
6647 if (bh
== args
->old_bucket
->bu_bhs
[0])
6648 bucket
= args
->old_bucket
;
6650 bucket
= args
->new_bucket
;
6652 return ocfs2_get_xattr_tree_value_root(sb
, bucket
, offset
,
6656 struct ocfs2_value_tree_metas
{
6662 static int ocfs2_value_tree_metas_in_bucket(struct super_block
*sb
,
6663 struct buffer_head
*bh
,
6664 struct ocfs2_xattr_header
*xh
,
6666 struct ocfs2_xattr_value_root
**xv
,
6667 struct buffer_head
**ret_bh
,
6670 struct ocfs2_xattr_bucket
*bucket
=
6671 (struct ocfs2_xattr_bucket
*)para
;
6673 return ocfs2_get_xattr_tree_value_root(sb
, bucket
, offset
,
6677 static int ocfs2_calc_value_tree_metas(struct inode
*inode
,
6678 struct ocfs2_xattr_bucket
*bucket
,
6681 struct ocfs2_value_tree_metas
*metas
=
6682 (struct ocfs2_value_tree_metas
*)para
;
6683 struct ocfs2_xattr_header
*xh
=
6684 (struct ocfs2_xattr_header
*)bucket
->bu_bhs
[0]->b_data
;
6686 /* Add the credits for this bucket first. */
6687 metas
->credits
+= bucket
->bu_blocks
;
6688 return ocfs2_value_metas_in_xattr_header(inode
->i_sb
, bucket
->bu_bhs
[0],
6689 xh
, &metas
->num_metas
,
6690 &metas
->credits
, &metas
->num_recs
,
6691 ocfs2_value_tree_metas_in_bucket
,
6696 * Given a xattr extent rec starting from blkno and having len clusters,
6697 * iterate all the buckets calculate how much metadata we need for reflinking
6698 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6700 static int ocfs2_lock_reflink_xattr_rec_allocators(
6701 struct ocfs2_reflink_xattr_tree_args
*args
,
6702 struct ocfs2_extent_tree
*xt_et
,
6703 u64 blkno
, u32 len
, int *credits
,
6704 struct ocfs2_alloc_context
**meta_ac
,
6705 struct ocfs2_alloc_context
**data_ac
)
6707 int ret
, num_free_extents
;
6708 struct ocfs2_value_tree_metas metas
;
6709 struct ocfs2_super
*osb
= OCFS2_SB(args
->reflink
->old_inode
->i_sb
);
6710 struct ocfs2_refcount_block
*rb
;
6712 memset(&metas
, 0, sizeof(metas
));
6714 ret
= ocfs2_iterate_xattr_buckets(args
->reflink
->old_inode
, blkno
, len
,
6715 ocfs2_calc_value_tree_metas
, &metas
);
6721 *credits
= metas
.credits
;
6724 * Calculate we need for refcount tree change.
6726 * We need to add/modify num_recs in refcount tree, so just calculate
6727 * an approximate number we need for refcount tree change.
6728 * Sometimes we need to split the tree, and after split, half recs
6729 * will be moved to the new block, and a new block can only provide
6730 * half number of recs. So we multiple new blocks by 2.
6731 * In the end, we have to add credits for modifying the already
6732 * existed refcount block.
6734 rb
= (struct ocfs2_refcount_block
*)args
->reflink
->ref_root_bh
->b_data
;
6736 (metas
.num_recs
+ ocfs2_refcount_recs_per_rb(osb
->sb
) - 1) /
6737 ocfs2_refcount_recs_per_rb(osb
->sb
) * 2;
6738 metas
.num_metas
+= metas
.num_recs
;
6739 *credits
+= metas
.num_recs
+
6740 metas
.num_recs
* OCFS2_EXPAND_REFCOUNT_TREE_CREDITS
;
6741 if (le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
)
6742 *credits
+= le16_to_cpu(rb
->rf_list
.l_tree_depth
) *
6743 le16_to_cpu(rb
->rf_list
.l_next_free_rec
) + 1;
6747 /* count in the xattr tree change. */
6748 num_free_extents
= ocfs2_num_free_extents(osb
, xt_et
);
6749 if (num_free_extents
< 0) {
6750 ret
= num_free_extents
;
6755 if (num_free_extents
< len
)
6756 metas
.num_metas
+= ocfs2_extend_meta_needed(xt_et
->et_root_el
);
6758 *credits
+= ocfs2_calc_extend_credits(osb
->sb
,
6759 xt_et
->et_root_el
, len
);
6761 if (metas
.num_metas
) {
6762 ret
= ocfs2_reserve_new_metadata_blocks(osb
, metas
.num_metas
,
6771 ret
= ocfs2_reserve_clusters(osb
, len
, data_ac
);
6778 ocfs2_free_alloc_context(*meta_ac
);
6786 static int ocfs2_reflink_xattr_buckets(handle_t
*handle
,
6787 u64 blkno
, u64 new_blkno
, u32 clusters
,
6788 struct ocfs2_alloc_context
*meta_ac
,
6789 struct ocfs2_alloc_context
*data_ac
,
6790 struct ocfs2_reflink_xattr_tree_args
*args
)
6793 struct super_block
*sb
= args
->reflink
->old_inode
->i_sb
;
6794 u32 bpc
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb
));
6795 u32 num_buckets
= clusters
* bpc
;
6796 int bpb
= args
->old_bucket
->bu_blocks
;
6797 struct ocfs2_xattr_value_buf vb
= {
6798 .vb_access
= ocfs2_journal_access
,
6801 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bpb
, new_blkno
+= bpb
) {
6802 ret
= ocfs2_read_xattr_bucket(args
->old_bucket
, blkno
);
6808 ret
= ocfs2_init_xattr_bucket(args
->new_bucket
, new_blkno
);
6815 * The real bucket num in this series of blocks is stored
6816 * in the 1st bucket.
6819 num_buckets
= le16_to_cpu(
6820 bucket_xh(args
->old_bucket
)->xh_num_buckets
);
6822 ret
= ocfs2_xattr_bucket_journal_access(handle
,
6824 OCFS2_JOURNAL_ACCESS_CREATE
);
6830 for (j
= 0; j
< bpb
; j
++)
6831 memcpy(bucket_block(args
->new_bucket
, j
),
6832 bucket_block(args
->old_bucket
, j
),
6835 ocfs2_xattr_bucket_journal_dirty(handle
, args
->new_bucket
);
6837 ret
= ocfs2_reflink_xattr_header(handle
, args
->reflink
,
6838 args
->old_bucket
->bu_bhs
[0],
6839 bucket_xh(args
->old_bucket
),
6840 args
->new_bucket
->bu_bhs
[0],
6841 bucket_xh(args
->new_bucket
),
6843 ocfs2_get_reflink_xattr_value_root
,
6851 * Re-access and dirty the bucket to calculate metaecc.
6852 * Because we may extend the transaction in reflink_xattr_header
6853 * which will let the already accessed block gone.
6855 ret
= ocfs2_xattr_bucket_journal_access(handle
,
6857 OCFS2_JOURNAL_ACCESS_WRITE
);
6863 ocfs2_xattr_bucket_journal_dirty(handle
, args
->new_bucket
);
6864 ocfs2_xattr_bucket_relse(args
->old_bucket
);
6865 ocfs2_xattr_bucket_relse(args
->new_bucket
);
6868 ocfs2_xattr_bucket_relse(args
->old_bucket
);
6869 ocfs2_xattr_bucket_relse(args
->new_bucket
);
6873 * Create the same xattr extent record in the new inode's xattr tree.
6875 static int ocfs2_reflink_xattr_rec(struct inode
*inode
,
6876 struct buffer_head
*root_bh
,
6882 int ret
, credits
= 0;
6883 u32 p_cluster
, num_clusters
;
6886 struct ocfs2_reflink_xattr_tree_args
*args
=
6887 (struct ocfs2_reflink_xattr_tree_args
*)para
;
6888 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
6889 struct ocfs2_alloc_context
*meta_ac
= NULL
;
6890 struct ocfs2_alloc_context
*data_ac
= NULL
;
6891 struct ocfs2_extent_tree et
;
6893 ocfs2_init_xattr_tree_extent_tree(&et
,
6894 INODE_CACHE(args
->reflink
->new_inode
),
6897 ret
= ocfs2_lock_reflink_xattr_rec_allocators(args
, &et
, blkno
,
6899 &meta_ac
, &data_ac
);
6905 handle
= ocfs2_start_trans(osb
, credits
);
6906 if (IS_ERR(handle
)) {
6907 ret
= PTR_ERR(handle
);
6912 ret
= ocfs2_claim_clusters(handle
, data_ac
,
6913 len
, &p_cluster
, &num_clusters
);
6919 new_blkno
= ocfs2_clusters_to_blocks(osb
->sb
, p_cluster
);
6921 mlog(0, "reflink xattr buckets %llu to %llu, len %u\n",
6922 (unsigned long long)blkno
, (unsigned long long)new_blkno
, len
);
6923 ret
= ocfs2_reflink_xattr_buckets(handle
, blkno
, new_blkno
, len
,
6924 meta_ac
, data_ac
, args
);
6930 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
6931 (unsigned long long)new_blkno
, len
, cpos
);
6932 ret
= ocfs2_insert_extent(handle
, &et
, cpos
, new_blkno
,
6938 ocfs2_commit_trans(osb
, handle
);
6942 ocfs2_free_alloc_context(meta_ac
);
6944 ocfs2_free_alloc_context(data_ac
);
6949 * Create reflinked xattr buckets.
6950 * We will add bucket one by one, and refcount all the xattrs in the bucket
6951 * if they are stored outside.
6953 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink
*args
,
6954 struct buffer_head
*blk_bh
,
6955 struct buffer_head
*new_blk_bh
)
6958 struct ocfs2_reflink_xattr_tree_args para
;
6960 memset(¶
, 0, sizeof(para
));
6961 para
.reflink
= args
;
6962 para
.old_blk_bh
= blk_bh
;
6963 para
.new_blk_bh
= new_blk_bh
;
6965 para
.old_bucket
= ocfs2_xattr_bucket_new(args
->old_inode
);
6966 if (!para
.old_bucket
) {
6967 mlog_errno(-ENOMEM
);
6971 para
.new_bucket
= ocfs2_xattr_bucket_new(args
->new_inode
);
6972 if (!para
.new_bucket
) {
6978 ret
= ocfs2_iterate_xattr_index_block(args
->old_inode
, blk_bh
,
6979 ocfs2_reflink_xattr_rec
,
6985 ocfs2_xattr_bucket_free(para
.old_bucket
);
6986 ocfs2_xattr_bucket_free(para
.new_bucket
);
6990 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink
*args
,
6991 struct buffer_head
*blk_bh
)
6993 int ret
, indexed
= 0;
6994 struct buffer_head
*new_blk_bh
= NULL
;
6995 struct ocfs2_xattr_block
*xb
=
6996 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
6999 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)
7002 ret
= ocfs2_create_empty_xattr_block(args
->new_inode
, args
->new_bh
,
7003 &new_blk_bh
, indexed
);
7009 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
))
7010 ret
= ocfs2_reflink_xattr_block(args
, blk_bh
, new_blk_bh
);
7012 ret
= ocfs2_reflink_xattr_tree(args
, blk_bh
, new_blk_bh
);
7021 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry
*xe
)
7023 int type
= ocfs2_xattr_get_type(xe
);
7025 return type
!= OCFS2_XATTR_INDEX_SECURITY
&&
7026 type
!= OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS
&&
7027 type
!= OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
;
7030 int ocfs2_reflink_xattrs(struct inode
*old_inode
,
7031 struct buffer_head
*old_bh
,
7032 struct inode
*new_inode
,
7033 struct buffer_head
*new_bh
,
7034 bool preserve_security
)
7037 struct ocfs2_xattr_reflink args
;
7038 struct ocfs2_inode_info
*oi
= OCFS2_I(old_inode
);
7039 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)old_bh
->b_data
;
7040 struct buffer_head
*blk_bh
= NULL
;
7041 struct ocfs2_cached_dealloc_ctxt dealloc
;
7042 struct ocfs2_refcount_tree
*ref_tree
;
7043 struct buffer_head
*ref_root_bh
= NULL
;
7045 ret
= ocfs2_lock_refcount_tree(OCFS2_SB(old_inode
->i_sb
),
7046 le64_to_cpu(di
->i_refcount_loc
),
7047 1, &ref_tree
, &ref_root_bh
);
7053 ocfs2_init_dealloc_ctxt(&dealloc
);
7055 args
.old_inode
= old_inode
;
7056 args
.new_inode
= new_inode
;
7057 args
.old_bh
= old_bh
;
7058 args
.new_bh
= new_bh
;
7059 args
.ref_ci
= &ref_tree
->rf_ci
;
7060 args
.ref_root_bh
= ref_root_bh
;
7061 args
.dealloc
= &dealloc
;
7062 if (preserve_security
)
7063 args
.xattr_reflinked
= NULL
;
7065 args
.xattr_reflinked
= ocfs2_reflink_xattr_no_security
;
7067 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
7068 ret
= ocfs2_reflink_xattr_inline(&args
);
7075 if (!di
->i_xattr_loc
)
7078 ret
= ocfs2_read_xattr_block(old_inode
, le64_to_cpu(di
->i_xattr_loc
),
7085 ret
= ocfs2_reflink_xattr_in_block(&args
, blk_bh
);
7092 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode
->i_sb
),
7094 brelse(ref_root_bh
);
7096 if (ocfs2_dealloc_has_cluster(&dealloc
)) {
7097 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode
->i_sb
), 1);
7098 ocfs2_run_deallocs(OCFS2_SB(old_inode
->i_sb
), &dealloc
);
7106 * Initialize security and acl for a already created inode.
7107 * Used for reflink a non-preserve-security file.
7109 * It uses common api like ocfs2_xattr_set, so the caller
7110 * must not hold any lock expect i_mutex.
7112 int ocfs2_init_security_and_acl(struct inode
*dir
,
7113 struct inode
*inode
)
7116 struct buffer_head
*dir_bh
= NULL
;
7117 struct ocfs2_security_xattr_info si
= {
7121 ret
= ocfs2_init_security_get(inode
, dir
, &si
);
7123 ret
= ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_SECURITY
,
7124 si
.name
, si
.value
, si
.value_len
,
7130 } else if (ret
!= -EOPNOTSUPP
) {
7135 ret
= ocfs2_inode_lock(dir
, &dir_bh
, 0);
7141 ret
= ocfs2_init_acl(NULL
, inode
, dir
, NULL
, dir_bh
, NULL
, NULL
);
7145 ocfs2_inode_unlock(dir
, 0);
7151 * 'security' attributes support
7153 static size_t ocfs2_xattr_security_list(struct dentry
*dentry
, char *list
,
7154 size_t list_size
, const char *name
,
7155 size_t name_len
, int type
)
7157 const size_t prefix_len
= XATTR_SECURITY_PREFIX_LEN
;
7158 const size_t total_len
= prefix_len
+ name_len
+ 1;
7160 if (list
&& total_len
<= list_size
) {
7161 memcpy(list
, XATTR_SECURITY_PREFIX
, prefix_len
);
7162 memcpy(list
+ prefix_len
, name
, name_len
);
7163 list
[prefix_len
+ name_len
] = '\0';
7168 static int ocfs2_xattr_security_get(struct dentry
*dentry
, const char *name
,
7169 void *buffer
, size_t size
, int type
)
7171 if (strcmp(name
, "") == 0)
7173 return ocfs2_xattr_get(dentry
->d_inode
, OCFS2_XATTR_INDEX_SECURITY
,
7174 name
, buffer
, size
);
7177 static int ocfs2_xattr_security_set(struct dentry
*dentry
, const char *name
,
7178 const void *value
, size_t size
, int flags
, int type
)
7180 if (strcmp(name
, "") == 0)
7183 return ocfs2_xattr_set(dentry
->d_inode
, OCFS2_XATTR_INDEX_SECURITY
,
7184 name
, value
, size
, flags
);
7187 int ocfs2_init_security_get(struct inode
*inode
,
7189 struct ocfs2_security_xattr_info
*si
)
7191 /* check whether ocfs2 support feature xattr */
7192 if (!ocfs2_supports_xattr(OCFS2_SB(dir
->i_sb
)))
7194 return security_inode_init_security(inode
, dir
, &si
->name
, &si
->value
,
7198 int ocfs2_init_security_set(handle_t
*handle
,
7199 struct inode
*inode
,
7200 struct buffer_head
*di_bh
,
7201 struct ocfs2_security_xattr_info
*si
,
7202 struct ocfs2_alloc_context
*xattr_ac
,
7203 struct ocfs2_alloc_context
*data_ac
)
7205 return ocfs2_xattr_set_handle(handle
, inode
, di_bh
,
7206 OCFS2_XATTR_INDEX_SECURITY
,
7207 si
->name
, si
->value
, si
->value_len
, 0,
7211 struct xattr_handler ocfs2_xattr_security_handler
= {
7212 .prefix
= XATTR_SECURITY_PREFIX
,
7213 .list
= ocfs2_xattr_security_list
,
7214 .get
= ocfs2_xattr_security_get
,
7215 .set
= ocfs2_xattr_security_set
,
7219 * 'trusted' attributes support
7221 static size_t ocfs2_xattr_trusted_list(struct dentry
*dentry
, char *list
,
7222 size_t list_size
, const char *name
,
7223 size_t name_len
, int type
)
7225 const size_t prefix_len
= XATTR_TRUSTED_PREFIX_LEN
;
7226 const size_t total_len
= prefix_len
+ name_len
+ 1;
7228 if (list
&& total_len
<= list_size
) {
7229 memcpy(list
, XATTR_TRUSTED_PREFIX
, prefix_len
);
7230 memcpy(list
+ prefix_len
, name
, name_len
);
7231 list
[prefix_len
+ name_len
] = '\0';
7236 static int ocfs2_xattr_trusted_get(struct dentry
*dentry
, const char *name
,
7237 void *buffer
, size_t size
, int type
)
7239 if (strcmp(name
, "") == 0)
7241 return ocfs2_xattr_get(dentry
->d_inode
, OCFS2_XATTR_INDEX_TRUSTED
,
7242 name
, buffer
, size
);
7245 static int ocfs2_xattr_trusted_set(struct dentry
*dentry
, const char *name
,
7246 const void *value
, size_t size
, int flags
, int type
)
7248 if (strcmp(name
, "") == 0)
7251 return ocfs2_xattr_set(dentry
->d_inode
, OCFS2_XATTR_INDEX_TRUSTED
,
7252 name
, value
, size
, flags
);
7255 struct xattr_handler ocfs2_xattr_trusted_handler
= {
7256 .prefix
= XATTR_TRUSTED_PREFIX
,
7257 .list
= ocfs2_xattr_trusted_list
,
7258 .get
= ocfs2_xattr_trusted_get
,
7259 .set
= ocfs2_xattr_trusted_set
,
7263 * 'user' attributes support
7265 static size_t ocfs2_xattr_user_list(struct dentry
*dentry
, char *list
,
7266 size_t list_size
, const char *name
,
7267 size_t name_len
, int type
)
7269 const size_t prefix_len
= XATTR_USER_PREFIX_LEN
;
7270 const size_t total_len
= prefix_len
+ name_len
+ 1;
7271 struct ocfs2_super
*osb
= OCFS2_SB(dentry
->d_sb
);
7273 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
7276 if (list
&& total_len
<= list_size
) {
7277 memcpy(list
, XATTR_USER_PREFIX
, prefix_len
);
7278 memcpy(list
+ prefix_len
, name
, name_len
);
7279 list
[prefix_len
+ name_len
] = '\0';
7284 static int ocfs2_xattr_user_get(struct dentry
*dentry
, const char *name
,
7285 void *buffer
, size_t size
, int type
)
7287 struct ocfs2_super
*osb
= OCFS2_SB(dentry
->d_sb
);
7289 if (strcmp(name
, "") == 0)
7291 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
7293 return ocfs2_xattr_get(dentry
->d_inode
, OCFS2_XATTR_INDEX_USER
, name
,
7297 static int ocfs2_xattr_user_set(struct dentry
*dentry
, const char *name
,
7298 const void *value
, size_t size
, int flags
, int type
)
7300 struct ocfs2_super
*osb
= OCFS2_SB(dentry
->d_sb
);
7302 if (strcmp(name
, "") == 0)
7304 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
7307 return ocfs2_xattr_set(dentry
->d_inode
, OCFS2_XATTR_INDEX_USER
,
7308 name
, value
, size
, flags
);
7311 struct xattr_handler ocfs2_xattr_user_handler
= {
7312 .prefix
= XATTR_USER_PREFIX
,
7313 .list
= ocfs2_xattr_user_list
,
7314 .get
= ocfs2_xattr_user_get
,
7315 .set
= ocfs2_xattr_user_set
,