1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
8 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
9 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 #include <linux/capability.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/highmem.h>
17 #include <linux/pagemap.h>
18 #include <linux/uio.h>
19 #include <linux/sched.h>
20 #include <linux/splice.h>
21 #include <linux/mount.h>
22 #include <linux/writeback.h>
23 #include <linux/falloc.h>
24 #include <linux/sort.h>
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/string.h>
28 #include <linux/security.h>
30 #include <cluster/masklog.h>
34 #include "blockcheck.h"
44 #include "buffer_head_io.h"
47 #include "refcounttree.h"
49 #include "ocfs2_trace.h"
51 struct ocfs2_xattr_def_value_root
{
52 struct ocfs2_xattr_value_root xv
;
53 struct ocfs2_extent_rec er
;
56 struct ocfs2_xattr_bucket
{
57 /* The inode these xattrs are associated with */
58 struct inode
*bu_inode
;
60 /* The actual buffers that make up the bucket */
61 struct buffer_head
*bu_bhs
[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
];
63 /* How many blocks make up one bucket for this filesystem */
67 struct ocfs2_xattr_set_ctxt
{
69 struct ocfs2_alloc_context
*meta_ac
;
70 struct ocfs2_alloc_context
*data_ac
;
71 struct ocfs2_cached_dealloc_ctxt dealloc
;
75 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
76 #define OCFS2_XATTR_INLINE_SIZE 80
77 #define OCFS2_XATTR_HEADER_GAP 4
78 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
79 - sizeof(struct ocfs2_xattr_header) \
80 - OCFS2_XATTR_HEADER_GAP)
81 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
82 - sizeof(struct ocfs2_xattr_block) \
83 - sizeof(struct ocfs2_xattr_header) \
84 - OCFS2_XATTR_HEADER_GAP)
86 static struct ocfs2_xattr_def_value_root def_xv
= {
87 .xv
.xr_list
.l_count
= cpu_to_le16(1),
90 const struct xattr_handler
*ocfs2_xattr_handlers
[] = {
91 &ocfs2_xattr_user_handler
,
92 &ocfs2_xattr_trusted_handler
,
93 &ocfs2_xattr_security_handler
,
97 static const struct xattr_handler
*ocfs2_xattr_handler_map
[OCFS2_XATTR_MAX
] = {
98 [OCFS2_XATTR_INDEX_USER
] = &ocfs2_xattr_user_handler
,
99 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS
] = &nop_posix_acl_access
,
100 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
] = &nop_posix_acl_default
,
101 [OCFS2_XATTR_INDEX_TRUSTED
] = &ocfs2_xattr_trusted_handler
,
102 [OCFS2_XATTR_INDEX_SECURITY
] = &ocfs2_xattr_security_handler
,
105 struct ocfs2_xattr_info
{
109 const void *xi_value
;
113 struct ocfs2_xattr_search
{
114 struct buffer_head
*inode_bh
;
116 * xattr_bh point to the block buffer head which has extended attribute
117 * when extended attribute in inode, xattr_bh is equal to inode_bh.
119 struct buffer_head
*xattr_bh
;
120 struct ocfs2_xattr_header
*header
;
121 struct ocfs2_xattr_bucket
*bucket
;
124 struct ocfs2_xattr_entry
*here
;
128 /* Operations on struct ocfs2_xa_entry */
130 struct ocfs2_xa_loc_operations
{
134 int (*xlo_journal_access
)(handle_t
*handle
, struct ocfs2_xa_loc
*loc
,
136 void (*xlo_journal_dirty
)(handle_t
*handle
, struct ocfs2_xa_loc
*loc
);
139 * Return a pointer to the appropriate buffer in loc->xl_storage
140 * at the given offset from loc->xl_header.
142 void *(*xlo_offset_pointer
)(struct ocfs2_xa_loc
*loc
, int offset
);
144 /* Can we reuse the existing entry for the new value? */
145 int (*xlo_can_reuse
)(struct ocfs2_xa_loc
*loc
,
146 struct ocfs2_xattr_info
*xi
);
148 /* How much space is needed for the new value? */
149 int (*xlo_check_space
)(struct ocfs2_xa_loc
*loc
,
150 struct ocfs2_xattr_info
*xi
);
153 * Return the offset of the first name+value pair. This is
154 * the start of our downward-filling free space.
156 int (*xlo_get_free_start
)(struct ocfs2_xa_loc
*loc
);
159 * Remove the name+value at this location. Do whatever is
160 * appropriate with the remaining name+value pairs.
162 void (*xlo_wipe_namevalue
)(struct ocfs2_xa_loc
*loc
);
164 /* Fill xl_entry with a new entry */
165 void (*xlo_add_entry
)(struct ocfs2_xa_loc
*loc
, u32 name_hash
);
167 /* Add name+value storage to an entry */
168 void (*xlo_add_namevalue
)(struct ocfs2_xa_loc
*loc
, int size
);
171 * Initialize the value buf's access and bh fields for this entry.
172 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
174 void (*xlo_fill_value_buf
)(struct ocfs2_xa_loc
*loc
,
175 struct ocfs2_xattr_value_buf
*vb
);
179 * Describes an xattr entry location. This is a memory structure
180 * tracking the on-disk structure.
182 struct ocfs2_xa_loc
{
183 /* This xattr belongs to this inode */
184 struct inode
*xl_inode
;
186 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
187 struct ocfs2_xattr_header
*xl_header
;
189 /* Bytes from xl_header to the end of the storage */
193 * The ocfs2_xattr_entry this location describes. If this is
194 * NULL, this location describes the on-disk structure where it
197 struct ocfs2_xattr_entry
*xl_entry
;
200 * Internal housekeeping
203 /* Buffer(s) containing this entry */
206 /* Operations on the storage backing this location */
207 const struct ocfs2_xa_loc_operations
*xl_ops
;
211 * Convenience functions to calculate how much space is needed for a
212 * given name+value pair
214 static int namevalue_size(int name_len
, uint64_t value_len
)
216 if (value_len
> OCFS2_XATTR_INLINE_SIZE
)
217 return OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_ROOT_SIZE
;
219 return OCFS2_XATTR_SIZE(name_len
) + OCFS2_XATTR_SIZE(value_len
);
222 static int namevalue_size_xi(struct ocfs2_xattr_info
*xi
)
224 return namevalue_size(xi
->xi_name_len
, xi
->xi_value_len
);
227 static int namevalue_size_xe(struct ocfs2_xattr_entry
*xe
)
229 u64 value_len
= le64_to_cpu(xe
->xe_value_size
);
231 BUG_ON((value_len
> OCFS2_XATTR_INLINE_SIZE
) &&
232 ocfs2_xattr_is_local(xe
));
233 return namevalue_size(xe
->xe_name_len
, value_len
);
237 static int ocfs2_xattr_bucket_get_name_value(struct super_block
*sb
,
238 struct ocfs2_xattr_header
*xh
,
243 static int ocfs2_xattr_block_find(struct inode
*inode
,
246 struct ocfs2_xattr_search
*xs
);
247 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
248 struct buffer_head
*root_bh
,
251 struct ocfs2_xattr_search
*xs
);
253 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
254 struct buffer_head
*blk_bh
,
258 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
259 struct ocfs2_xattr_search
*xs
,
260 struct ocfs2_xattr_set_ctxt
*ctxt
);
262 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
263 struct ocfs2_xattr_info
*xi
,
264 struct ocfs2_xattr_search
*xs
,
265 struct ocfs2_xattr_set_ctxt
*ctxt
);
267 typedef int (xattr_tree_rec_func
)(struct inode
*inode
,
268 struct buffer_head
*root_bh
,
269 u64 blkno
, u32 cpos
, u32 len
, void *para
);
270 static int ocfs2_iterate_xattr_index_block(struct inode
*inode
,
271 struct buffer_head
*root_bh
,
272 xattr_tree_rec_func
*rec_func
,
274 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
275 struct ocfs2_xattr_bucket
*bucket
,
277 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
278 struct buffer_head
*root_bh
,
284 static int ocfs2_mv_xattr_buckets(struct inode
*inode
, handle_t
*handle
,
285 u64 src_blk
, u64 last_blk
, u64 to_blk
,
286 unsigned int start_bucket
,
288 static int ocfs2_prepare_refcount_xattr(struct inode
*inode
,
289 struct ocfs2_dinode
*di
,
290 struct ocfs2_xattr_info
*xi
,
291 struct ocfs2_xattr_search
*xis
,
292 struct ocfs2_xattr_search
*xbs
,
293 struct ocfs2_refcount_tree
**ref_tree
,
296 static int ocfs2_get_xattr_tree_value_root(struct super_block
*sb
,
297 struct ocfs2_xattr_bucket
*bucket
,
299 struct ocfs2_xattr_value_root
**xv
,
300 struct buffer_head
**bh
);
302 static inline u16
ocfs2_xattr_buckets_per_cluster(struct ocfs2_super
*osb
)
304 return (1 << osb
->s_clustersize_bits
) / OCFS2_XATTR_BUCKET_SIZE
;
307 static inline u16
ocfs2_blocks_per_xattr_bucket(struct super_block
*sb
)
309 return OCFS2_XATTR_BUCKET_SIZE
/ (1 << sb
->s_blocksize_bits
);
312 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
313 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
314 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
316 static struct ocfs2_xattr_bucket
*ocfs2_xattr_bucket_new(struct inode
*inode
)
318 struct ocfs2_xattr_bucket
*bucket
;
319 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
321 BUG_ON(blks
> OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET
);
323 bucket
= kzalloc(sizeof(struct ocfs2_xattr_bucket
), GFP_NOFS
);
325 bucket
->bu_inode
= inode
;
326 bucket
->bu_blocks
= blks
;
332 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket
*bucket
)
336 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
337 brelse(bucket
->bu_bhs
[i
]);
338 bucket
->bu_bhs
[i
] = NULL
;
342 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket
*bucket
)
345 ocfs2_xattr_bucket_relse(bucket
);
346 bucket
->bu_inode
= NULL
;
352 * A bucket that has never been written to disk doesn't need to be
353 * read. We just need the buffer_heads. Don't call this for
354 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
357 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
358 u64 xb_blkno
, int new)
362 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
363 bucket
->bu_bhs
[i
] = sb_getblk(bucket
->bu_inode
->i_sb
,
365 if (!bucket
->bu_bhs
[i
]) {
371 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket
->bu_inode
),
372 bucket
->bu_bhs
[i
])) {
374 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket
->bu_inode
),
377 set_buffer_uptodate(bucket
->bu_bhs
[i
]);
378 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket
->bu_inode
),
385 ocfs2_xattr_bucket_relse(bucket
);
389 /* Read the xattr bucket at xb_blkno */
390 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket
*bucket
,
395 rc
= ocfs2_read_blocks(INODE_CACHE(bucket
->bu_inode
), xb_blkno
,
396 bucket
->bu_blocks
, bucket
->bu_bhs
, 0,
399 spin_lock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
400 rc
= ocfs2_validate_meta_ecc_bhs(bucket
->bu_inode
->i_sb
,
403 &bucket_xh(bucket
)->xh_check
);
404 spin_unlock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
410 ocfs2_xattr_bucket_relse(bucket
);
414 static int ocfs2_xattr_bucket_journal_access(handle_t
*handle
,
415 struct ocfs2_xattr_bucket
*bucket
,
420 for (i
= 0; i
< bucket
->bu_blocks
; i
++) {
421 rc
= ocfs2_journal_access(handle
,
422 INODE_CACHE(bucket
->bu_inode
),
423 bucket
->bu_bhs
[i
], type
);
433 static void ocfs2_xattr_bucket_journal_dirty(handle_t
*handle
,
434 struct ocfs2_xattr_bucket
*bucket
)
438 spin_lock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
439 ocfs2_compute_meta_ecc_bhs(bucket
->bu_inode
->i_sb
,
440 bucket
->bu_bhs
, bucket
->bu_blocks
,
441 &bucket_xh(bucket
)->xh_check
);
442 spin_unlock(&OCFS2_SB(bucket
->bu_inode
->i_sb
)->osb_xattr_lock
);
444 for (i
= 0; i
< bucket
->bu_blocks
; i
++)
445 ocfs2_journal_dirty(handle
, bucket
->bu_bhs
[i
]);
448 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket
*dest
,
449 struct ocfs2_xattr_bucket
*src
)
452 int blocksize
= src
->bu_inode
->i_sb
->s_blocksize
;
454 BUG_ON(dest
->bu_blocks
!= src
->bu_blocks
);
455 BUG_ON(dest
->bu_inode
!= src
->bu_inode
);
457 for (i
= 0; i
< src
->bu_blocks
; i
++) {
458 memcpy(bucket_block(dest
, i
), bucket_block(src
, i
),
463 static int ocfs2_validate_xattr_block(struct super_block
*sb
,
464 struct buffer_head
*bh
)
467 struct ocfs2_xattr_block
*xb
=
468 (struct ocfs2_xattr_block
*)bh
->b_data
;
470 trace_ocfs2_validate_xattr_block((unsigned long long)bh
->b_blocknr
);
472 BUG_ON(!buffer_uptodate(bh
));
475 * If the ecc fails, we return the error but otherwise
476 * leave the filesystem running. We know any error is
477 * local to this block.
479 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &xb
->xb_check
);
484 * Errors after here are fatal
487 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb
)) {
488 return ocfs2_error(sb
,
489 "Extended attribute block #%llu has bad signature %.*s\n",
490 (unsigned long long)bh
->b_blocknr
, 7,
494 if (le64_to_cpu(xb
->xb_blkno
) != bh
->b_blocknr
) {
495 return ocfs2_error(sb
,
496 "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
497 (unsigned long long)bh
->b_blocknr
,
498 (unsigned long long)le64_to_cpu(xb
->xb_blkno
));
501 if (le32_to_cpu(xb
->xb_fs_generation
) != OCFS2_SB(sb
)->fs_generation
) {
502 return ocfs2_error(sb
,
503 "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
504 (unsigned long long)bh
->b_blocknr
,
505 le32_to_cpu(xb
->xb_fs_generation
));
511 static int ocfs2_read_xattr_block(struct inode
*inode
, u64 xb_blkno
,
512 struct buffer_head
**bh
)
515 struct buffer_head
*tmp
= *bh
;
517 rc
= ocfs2_read_block(INODE_CACHE(inode
), xb_blkno
, &tmp
,
518 ocfs2_validate_xattr_block
);
520 /* If ocfs2_read_block() got us a new bh, pass it up. */
527 static inline const char *ocfs2_xattr_prefix(int name_index
)
529 const struct xattr_handler
*handler
= NULL
;
531 if (name_index
> 0 && name_index
< OCFS2_XATTR_MAX
)
532 handler
= ocfs2_xattr_handler_map
[name_index
];
533 return handler
? xattr_prefix(handler
) : NULL
;
536 static u32
ocfs2_xattr_name_hash(struct inode
*inode
,
540 /* Get hash value of uuid from super block */
541 u32 hash
= OCFS2_SB(inode
->i_sb
)->uuid_hash
;
544 /* hash extended attribute name */
545 for (i
= 0; i
< name_len
; i
++) {
546 hash
= (hash
<< OCFS2_HASH_SHIFT
) ^
547 (hash
>> (8*sizeof(hash
) - OCFS2_HASH_SHIFT
)) ^
554 static int ocfs2_xattr_entry_real_size(int name_len
, size_t value_len
)
556 return namevalue_size(name_len
, value_len
) +
557 sizeof(struct ocfs2_xattr_entry
);
560 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info
*xi
)
562 return namevalue_size_xi(xi
) +
563 sizeof(struct ocfs2_xattr_entry
);
566 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry
*xe
)
568 return namevalue_size_xe(xe
) +
569 sizeof(struct ocfs2_xattr_entry
);
572 int ocfs2_calc_security_init(struct inode
*dir
,
573 struct ocfs2_security_xattr_info
*si
,
576 struct ocfs2_alloc_context
**xattr_ac
)
579 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
580 int s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
584 * The max space of security xattr taken inline is
585 * 256(name) + 80(value) + 16(entry) = 352 bytes,
586 * So reserve one metadata block for it is ok.
588 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
589 s_size
> OCFS2_XATTR_FREE_IN_IBODY
) {
590 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, xattr_ac
);
595 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
598 /* reserve clusters for xattr value which will be set in B tree*/
599 if (si
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
600 int new_clusters
= ocfs2_clusters_for_bytes(dir
->i_sb
,
603 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
605 *want_clusters
+= new_clusters
;
610 int ocfs2_calc_xattr_init(struct inode
*dir
,
611 struct buffer_head
*dir_bh
,
613 struct ocfs2_security_xattr_info
*si
,
619 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
620 int s_size
= 0, a_size
= 0, acl_len
= 0, new_clusters
;
623 s_size
= ocfs2_xattr_entry_real_size(strlen(si
->name
),
626 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
) {
627 down_read(&OCFS2_I(dir
)->ip_xattr_sem
);
628 acl_len
= ocfs2_xattr_get_nolock(dir
, dir_bh
,
629 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
,
631 up_read(&OCFS2_I(dir
)->ip_xattr_sem
);
633 a_size
= ocfs2_xattr_entry_real_size(0, acl_len
);
636 } else if (acl_len
!= 0 && acl_len
!= -ENODATA
) {
643 if (!(s_size
+ a_size
))
647 * The max space of security xattr taken inline is
648 * 256(name) + 80(value) + 16(entry) = 352 bytes,
649 * The max space of acl xattr taken inline is
650 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
651 * when blocksize = 512, may reserve one more cluser for
652 * xattr bucket, otherwise reserve one metadata block
654 * If this is a new directory with inline data,
655 * we choose to reserve the entire inline area for
656 * directory contents and force an external xattr block.
658 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
||
659 (S_ISDIR(mode
) && ocfs2_supports_inline_data(osb
)) ||
660 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_IBODY
) {
661 *want_meta
= *want_meta
+ 1;
662 *xattr_credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
665 if (dir
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
&&
666 (s_size
+ a_size
) > OCFS2_XATTR_FREE_IN_BLOCK(dir
)) {
668 *xattr_credits
+= ocfs2_blocks_per_xattr_bucket(dir
->i_sb
);
672 * reserve credits and clusters for xattrs which has large value
673 * and have to be set outside
675 if (si
->enable
&& si
->value_len
> OCFS2_XATTR_INLINE_SIZE
) {
676 new_clusters
= ocfs2_clusters_for_bytes(dir
->i_sb
,
678 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
680 *want_clusters
+= new_clusters
;
682 if (osb
->s_mount_opt
& OCFS2_MOUNT_POSIX_ACL
&&
683 acl_len
> OCFS2_XATTR_INLINE_SIZE
) {
684 /* for directory, it has DEFAULT and ACCESS two types of acls */
685 new_clusters
= (S_ISDIR(mode
) ? 2 : 1) *
686 ocfs2_clusters_for_bytes(dir
->i_sb
, acl_len
);
687 *xattr_credits
+= ocfs2_clusters_to_blocks(dir
->i_sb
,
689 *want_clusters
+= new_clusters
;
695 static int ocfs2_xattr_extend_allocation(struct inode
*inode
,
697 struct ocfs2_xattr_value_buf
*vb
,
698 struct ocfs2_xattr_set_ctxt
*ctxt
)
700 int status
= 0, credits
;
701 handle_t
*handle
= ctxt
->handle
;
702 enum ocfs2_alloc_restarted why
;
703 u32 prev_clusters
, logical_start
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
704 struct ocfs2_extent_tree et
;
706 ocfs2_init_xattr_value_extent_tree(&et
, INODE_CACHE(inode
), vb
);
708 while (clusters_to_add
) {
709 trace_ocfs2_xattr_extend_allocation(clusters_to_add
);
711 status
= vb
->vb_access(handle
, INODE_CACHE(inode
), vb
->vb_bh
,
712 OCFS2_JOURNAL_ACCESS_WRITE
);
718 prev_clusters
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
719 status
= ocfs2_add_clusters_in_btree(handle
,
727 if ((status
< 0) && (status
!= -EAGAIN
)) {
728 if (status
!= -ENOSPC
)
733 ocfs2_journal_dirty(handle
, vb
->vb_bh
);
735 clusters_to_add
-= le32_to_cpu(vb
->vb_xv
->xr_clusters
) -
738 if (why
!= RESTART_NONE
&& clusters_to_add
) {
740 * We can only fail in case the alloc file doesn't give
741 * up enough clusters.
743 BUG_ON(why
== RESTART_META
);
745 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
746 &vb
->vb_xv
->xr_list
);
747 status
= ocfs2_extend_trans(handle
, credits
);
759 static int __ocfs2_remove_xattr_range(struct inode
*inode
,
760 struct ocfs2_xattr_value_buf
*vb
,
761 u32 cpos
, u32 phys_cpos
, u32 len
,
762 unsigned int ext_flags
,
763 struct ocfs2_xattr_set_ctxt
*ctxt
)
766 u64 phys_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
767 handle_t
*handle
= ctxt
->handle
;
768 struct ocfs2_extent_tree et
;
770 ocfs2_init_xattr_value_extent_tree(&et
, INODE_CACHE(inode
), vb
);
772 ret
= vb
->vb_access(handle
, INODE_CACHE(inode
), vb
->vb_bh
,
773 OCFS2_JOURNAL_ACCESS_WRITE
);
779 ret
= ocfs2_remove_extent(handle
, &et
, cpos
, len
, ctxt
->meta_ac
,
786 le32_add_cpu(&vb
->vb_xv
->xr_clusters
, -len
);
787 ocfs2_journal_dirty(handle
, vb
->vb_bh
);
789 if (ext_flags
& OCFS2_EXT_REFCOUNTED
)
790 ret
= ocfs2_decrease_refcount(inode
, handle
,
791 ocfs2_blocks_to_clusters(inode
->i_sb
,
793 len
, ctxt
->meta_ac
, &ctxt
->dealloc
, 1);
795 ret
= ocfs2_cache_cluster_dealloc(&ctxt
->dealloc
,
804 static int ocfs2_xattr_shrink_size(struct inode
*inode
,
807 struct ocfs2_xattr_value_buf
*vb
,
808 struct ocfs2_xattr_set_ctxt
*ctxt
)
811 unsigned int ext_flags
;
812 u32 trunc_len
, cpos
, phys_cpos
, alloc_size
;
815 if (old_clusters
<= new_clusters
)
819 trunc_len
= old_clusters
- new_clusters
;
821 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &phys_cpos
,
823 &vb
->vb_xv
->xr_list
, &ext_flags
);
829 if (alloc_size
> trunc_len
)
830 alloc_size
= trunc_len
;
832 ret
= __ocfs2_remove_xattr_range(inode
, vb
, cpos
,
833 phys_cpos
, alloc_size
,
840 block
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
841 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode
),
844 trunc_len
-= alloc_size
;
851 static int ocfs2_xattr_value_truncate(struct inode
*inode
,
852 struct ocfs2_xattr_value_buf
*vb
,
854 struct ocfs2_xattr_set_ctxt
*ctxt
)
857 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
858 u32 old_clusters
= le32_to_cpu(vb
->vb_xv
->xr_clusters
);
860 if (new_clusters
== old_clusters
)
863 if (new_clusters
> old_clusters
)
864 ret
= ocfs2_xattr_extend_allocation(inode
,
865 new_clusters
- old_clusters
,
868 ret
= ocfs2_xattr_shrink_size(inode
,
869 old_clusters
, new_clusters
,
875 static int ocfs2_xattr_list_entry(struct super_block
*sb
,
876 char *buffer
, size_t size
,
877 size_t *result
, int type
,
878 const char *name
, int name_len
)
880 char *p
= buffer
+ *result
;
886 case OCFS2_XATTR_INDEX_USER
:
887 if (OCFS2_SB(sb
)->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
891 case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS
:
892 case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
:
893 if (!(sb
->s_flags
& SB_POSIXACL
))
897 case OCFS2_XATTR_INDEX_TRUSTED
:
898 if (!capable(CAP_SYS_ADMIN
))
903 prefix
= ocfs2_xattr_prefix(type
);
906 prefix_len
= strlen(prefix
);
907 total_len
= prefix_len
+ name_len
+ 1;
908 *result
+= total_len
;
910 /* we are just looking for how big our buffer needs to be */
917 memcpy(p
, prefix
, prefix_len
);
918 memcpy(p
+ prefix_len
, name
, name_len
);
919 p
[prefix_len
+ name_len
] = '\0';
924 static int ocfs2_xattr_list_entries(struct inode
*inode
,
925 struct ocfs2_xattr_header
*header
,
926 char *buffer
, size_t buffer_size
)
932 for (i
= 0 ; i
< le16_to_cpu(header
->xh_count
); i
++) {
933 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
934 type
= ocfs2_xattr_get_type(entry
);
935 name
= (const char *)header
+
936 le16_to_cpu(entry
->xe_name_offset
);
938 ret
= ocfs2_xattr_list_entry(inode
->i_sb
,
949 int ocfs2_has_inline_xattr_value_outside(struct inode
*inode
,
950 struct ocfs2_dinode
*di
)
952 struct ocfs2_xattr_header
*xh
;
955 xh
= (struct ocfs2_xattr_header
*)
956 ((void *)di
+ inode
->i_sb
->s_blocksize
-
957 le16_to_cpu(di
->i_xattr_inline_size
));
959 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++)
960 if (!ocfs2_xattr_is_local(&xh
->xh_entries
[i
]))
966 static int ocfs2_xattr_ibody_list(struct inode
*inode
,
967 struct ocfs2_dinode
*di
,
971 struct ocfs2_xattr_header
*header
= NULL
;
972 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
975 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
978 header
= (struct ocfs2_xattr_header
*)
979 ((void *)di
+ inode
->i_sb
->s_blocksize
-
980 le16_to_cpu(di
->i_xattr_inline_size
));
982 ret
= ocfs2_xattr_list_entries(inode
, header
, buffer
, buffer_size
);
987 static int ocfs2_xattr_block_list(struct inode
*inode
,
988 struct ocfs2_dinode
*di
,
992 struct buffer_head
*blk_bh
= NULL
;
993 struct ocfs2_xattr_block
*xb
;
996 if (!di
->i_xattr_loc
)
999 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
1006 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
1007 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
1008 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
1009 ret
= ocfs2_xattr_list_entries(inode
, header
,
1010 buffer
, buffer_size
);
1012 ret
= ocfs2_xattr_tree_list_index_block(inode
, blk_bh
,
1013 buffer
, buffer_size
);
1020 ssize_t
ocfs2_listxattr(struct dentry
*dentry
,
1024 int ret
= 0, i_ret
= 0, b_ret
= 0;
1025 struct buffer_head
*di_bh
= NULL
;
1026 struct ocfs2_dinode
*di
= NULL
;
1027 struct ocfs2_inode_info
*oi
= OCFS2_I(d_inode(dentry
));
1029 if (!ocfs2_supports_xattr(OCFS2_SB(dentry
->d_sb
)))
1032 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1035 ret
= ocfs2_inode_lock(d_inode(dentry
), &di_bh
, 0);
1041 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1043 down_read(&oi
->ip_xattr_sem
);
1044 i_ret
= ocfs2_xattr_ibody_list(d_inode(dentry
), di
, buffer
, size
);
1052 b_ret
= ocfs2_xattr_block_list(d_inode(dentry
), di
,
1057 up_read(&oi
->ip_xattr_sem
);
1058 ocfs2_inode_unlock(d_inode(dentry
), 0);
1062 return i_ret
+ b_ret
;
1065 static int ocfs2_xattr_find_entry(int name_index
,
1067 struct ocfs2_xattr_search
*xs
)
1069 struct ocfs2_xattr_entry
*entry
;
1076 name_len
= strlen(name
);
1078 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
1079 cmp
= name_index
- ocfs2_xattr_get_type(entry
);
1081 cmp
= name_len
- entry
->xe_name_len
;
1083 cmp
= memcmp(name
, (xs
->base
+
1084 le16_to_cpu(entry
->xe_name_offset
)),
1092 return cmp
? -ENODATA
: 0;
1095 static int ocfs2_xattr_get_value_outside(struct inode
*inode
,
1096 struct ocfs2_xattr_value_root
*xv
,
1100 u32 cpos
, p_cluster
, num_clusters
, bpc
, clusters
;
1103 size_t cplen
, blocksize
;
1104 struct buffer_head
*bh
= NULL
;
1105 struct ocfs2_extent_list
*el
;
1108 clusters
= le32_to_cpu(xv
->xr_clusters
);
1109 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1110 blocksize
= inode
->i_sb
->s_blocksize
;
1113 while (cpos
< clusters
) {
1114 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1115 &num_clusters
, el
, NULL
);
1121 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1122 /* Copy ocfs2_xattr_value */
1123 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1124 ret
= ocfs2_read_block(INODE_CACHE(inode
), blkno
,
1131 cplen
= len
>= blocksize
? blocksize
: len
;
1132 memcpy(buffer
, bh
->b_data
, cplen
);
1141 cpos
+= num_clusters
;
1147 static int ocfs2_xattr_ibody_get(struct inode
*inode
,
1152 struct ocfs2_xattr_search
*xs
)
1154 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1155 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
1156 struct ocfs2_xattr_value_root
*xv
;
1160 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
))
1163 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
1164 xs
->header
= (struct ocfs2_xattr_header
*)
1165 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
1166 xs
->base
= (void *)xs
->header
;
1167 xs
->here
= xs
->header
->xh_entries
;
1169 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
1172 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1174 if (size
> buffer_size
)
1176 if (ocfs2_xattr_is_local(xs
->here
)) {
1177 memcpy(buffer
, (void *)xs
->base
+
1178 le16_to_cpu(xs
->here
->xe_name_offset
) +
1179 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
), size
);
1181 xv
= (struct ocfs2_xattr_value_root
*)
1182 (xs
->base
+ le16_to_cpu(
1183 xs
->here
->xe_name_offset
) +
1184 OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
));
1185 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1197 static int ocfs2_xattr_block_get(struct inode
*inode
,
1202 struct ocfs2_xattr_search
*xs
)
1204 struct ocfs2_xattr_block
*xb
;
1205 struct ocfs2_xattr_value_root
*xv
;
1207 int ret
= -ENODATA
, name_offset
, name_len
, i
;
1210 xs
->bucket
= ocfs2_xattr_bucket_new(inode
);
1217 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, xs
);
1223 if (xs
->not_found
) {
1228 xb
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
1229 size
= le64_to_cpu(xs
->here
->xe_value_size
);
1232 if (size
> buffer_size
)
1235 name_offset
= le16_to_cpu(xs
->here
->xe_name_offset
);
1236 name_len
= OCFS2_XATTR_SIZE(xs
->here
->xe_name_len
);
1237 i
= xs
->here
- xs
->header
->xh_entries
;
1239 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
1240 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
1241 bucket_xh(xs
->bucket
),
1249 xs
->base
= bucket_block(xs
->bucket
, block_off
);
1251 if (ocfs2_xattr_is_local(xs
->here
)) {
1252 memcpy(buffer
, (void *)xs
->base
+
1253 name_offset
+ name_len
, size
);
1255 xv
= (struct ocfs2_xattr_value_root
*)
1256 (xs
->base
+ name_offset
+ name_len
);
1257 ret
= ocfs2_xattr_get_value_outside(inode
, xv
,
1267 ocfs2_xattr_bucket_free(xs
->bucket
);
1269 brelse(xs
->xattr_bh
);
1270 xs
->xattr_bh
= NULL
;
1274 int ocfs2_xattr_get_nolock(struct inode
*inode
,
1275 struct buffer_head
*di_bh
,
1282 struct ocfs2_dinode
*di
= NULL
;
1283 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1284 struct ocfs2_xattr_search xis
= {
1285 .not_found
= -ENODATA
,
1287 struct ocfs2_xattr_search xbs
= {
1288 .not_found
= -ENODATA
,
1291 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
1294 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
1297 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
1298 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1300 ret
= ocfs2_xattr_ibody_get(inode
, name_index
, name
, buffer
,
1302 if (ret
== -ENODATA
&& di
->i_xattr_loc
)
1303 ret
= ocfs2_xattr_block_get(inode
, name_index
, name
, buffer
,
1309 /* ocfs2_xattr_get()
1311 * Copy an extended attribute into the buffer provided.
1312 * Buffer is NULL to compute the size of buffer required.
1314 static int ocfs2_xattr_get(struct inode
*inode
,
1321 struct buffer_head
*di_bh
= NULL
;
1322 struct ocfs2_lock_holder oh
;
1324 had_lock
= ocfs2_inode_lock_tracker(inode
, &di_bh
, 0, &oh
);
1326 mlog_errno(had_lock
);
1329 down_read(&OCFS2_I(inode
)->ip_xattr_sem
);
1330 ret
= ocfs2_xattr_get_nolock(inode
, di_bh
, name_index
,
1331 name
, buffer
, buffer_size
);
1332 up_read(&OCFS2_I(inode
)->ip_xattr_sem
);
1334 ocfs2_inode_unlock_tracker(inode
, 0, &oh
, had_lock
);
1341 static int __ocfs2_xattr_set_value_outside(struct inode
*inode
,
1343 struct ocfs2_xattr_value_buf
*vb
,
1347 int ret
= 0, i
, cp_len
;
1348 u16 blocksize
= inode
->i_sb
->s_blocksize
;
1349 u32 p_cluster
, num_clusters
;
1350 u32 cpos
= 0, bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
1351 u32 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, value_len
);
1353 struct buffer_head
*bh
= NULL
;
1354 unsigned int ext_flags
;
1355 struct ocfs2_xattr_value_root
*xv
= vb
->vb_xv
;
1357 BUG_ON(clusters
> le32_to_cpu(xv
->xr_clusters
));
1359 while (cpos
< clusters
) {
1360 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
1361 &num_clusters
, &xv
->xr_list
,
1368 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1370 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
1372 for (i
= 0; i
< num_clusters
* bpc
; i
++, blkno
++) {
1373 ret
= ocfs2_read_block(INODE_CACHE(inode
), blkno
,
1380 ret
= ocfs2_journal_access(handle
,
1383 OCFS2_JOURNAL_ACCESS_WRITE
);
1389 cp_len
= value_len
> blocksize
? blocksize
: value_len
;
1390 memcpy(bh
->b_data
, value
, cp_len
);
1391 value_len
-= cp_len
;
1393 if (cp_len
< blocksize
)
1394 memset(bh
->b_data
+ cp_len
, 0,
1395 blocksize
- cp_len
);
1397 ocfs2_journal_dirty(handle
, bh
);
1402 * XXX: do we need to empty all the following
1403 * blocks in this cluster?
1408 cpos
+= num_clusters
;
1416 static int ocfs2_xa_check_space_helper(int needed_space
, int free_start
,
1424 free_space
= free_start
-
1425 sizeof(struct ocfs2_xattr_header
) -
1426 (num_entries
* sizeof(struct ocfs2_xattr_entry
)) -
1427 OCFS2_XATTR_HEADER_GAP
;
1430 if (free_space
< needed_space
)
1436 static int ocfs2_xa_journal_access(handle_t
*handle
, struct ocfs2_xa_loc
*loc
,
1439 return loc
->xl_ops
->xlo_journal_access(handle
, loc
, type
);
1442 static void ocfs2_xa_journal_dirty(handle_t
*handle
, struct ocfs2_xa_loc
*loc
)
1444 loc
->xl_ops
->xlo_journal_dirty(handle
, loc
);
1447 /* Give a pointer into the storage for the given offset */
1448 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc
*loc
, int offset
)
1450 BUG_ON(offset
>= loc
->xl_size
);
1451 return loc
->xl_ops
->xlo_offset_pointer(loc
, offset
);
1455 * Wipe the name+value pair and allow the storage to reclaim it. This
1456 * must be followed by either removal of the entry or a call to
1457 * ocfs2_xa_add_namevalue().
1459 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc
*loc
)
1461 loc
->xl_ops
->xlo_wipe_namevalue(loc
);
1465 * Find lowest offset to a name+value pair. This is the start of our
1466 * downward-growing free space.
1468 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc
*loc
)
1470 return loc
->xl_ops
->xlo_get_free_start(loc
);
1473 /* Can we reuse loc->xl_entry for xi? */
1474 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc
*loc
,
1475 struct ocfs2_xattr_info
*xi
)
1477 return loc
->xl_ops
->xlo_can_reuse(loc
, xi
);
1480 /* How much free space is needed to set the new value */
1481 static int ocfs2_xa_check_space(struct ocfs2_xa_loc
*loc
,
1482 struct ocfs2_xattr_info
*xi
)
1484 return loc
->xl_ops
->xlo_check_space(loc
, xi
);
1487 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc
*loc
, u32 name_hash
)
1489 loc
->xl_ops
->xlo_add_entry(loc
, name_hash
);
1490 loc
->xl_entry
->xe_name_hash
= cpu_to_le32(name_hash
);
1492 * We can't leave the new entry's xe_name_offset at zero or
1493 * add_namevalue() will go nuts. We set it to the size of our
1494 * storage so that it can never be less than any other entry.
1496 loc
->xl_entry
->xe_name_offset
= cpu_to_le16(loc
->xl_size
);
1499 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc
*loc
,
1500 struct ocfs2_xattr_info
*xi
)
1502 int size
= namevalue_size_xi(xi
);
1506 loc
->xl_ops
->xlo_add_namevalue(loc
, size
);
1507 loc
->xl_entry
->xe_value_size
= cpu_to_le64(xi
->xi_value_len
);
1508 loc
->xl_entry
->xe_name_len
= xi
->xi_name_len
;
1509 ocfs2_xattr_set_type(loc
->xl_entry
, xi
->xi_name_index
);
1510 ocfs2_xattr_set_local(loc
->xl_entry
,
1511 xi
->xi_value_len
<= OCFS2_XATTR_INLINE_SIZE
);
1513 nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
1514 nameval_buf
= ocfs2_xa_offset_pointer(loc
, nameval_offset
);
1515 memset(nameval_buf
, 0, size
);
1516 memcpy(nameval_buf
, xi
->xi_name
, xi
->xi_name_len
);
1519 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc
*loc
,
1520 struct ocfs2_xattr_value_buf
*vb
)
1522 int nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
1523 int name_size
= OCFS2_XATTR_SIZE(loc
->xl_entry
->xe_name_len
);
1525 /* Value bufs are for value trees */
1526 BUG_ON(ocfs2_xattr_is_local(loc
->xl_entry
));
1527 BUG_ON(namevalue_size_xe(loc
->xl_entry
) !=
1528 (name_size
+ OCFS2_XATTR_ROOT_SIZE
));
1530 loc
->xl_ops
->xlo_fill_value_buf(loc
, vb
);
1532 (struct ocfs2_xattr_value_root
*)ocfs2_xa_offset_pointer(loc
,
1537 static int ocfs2_xa_block_journal_access(handle_t
*handle
,
1538 struct ocfs2_xa_loc
*loc
, int type
)
1540 struct buffer_head
*bh
= loc
->xl_storage
;
1541 ocfs2_journal_access_func access
;
1543 if (loc
->xl_size
== (bh
->b_size
-
1544 offsetof(struct ocfs2_xattr_block
,
1545 xb_attrs
.xb_header
)))
1546 access
= ocfs2_journal_access_xb
;
1548 access
= ocfs2_journal_access_di
;
1549 return access(handle
, INODE_CACHE(loc
->xl_inode
), bh
, type
);
1552 static void ocfs2_xa_block_journal_dirty(handle_t
*handle
,
1553 struct ocfs2_xa_loc
*loc
)
1555 struct buffer_head
*bh
= loc
->xl_storage
;
1557 ocfs2_journal_dirty(handle
, bh
);
1560 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc
*loc
,
1563 return (char *)loc
->xl_header
+ offset
;
1566 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc
*loc
,
1567 struct ocfs2_xattr_info
*xi
)
1570 * Block storage is strict. If the sizes aren't exact, we will
1571 * remove the old one and reinsert the new.
1573 return namevalue_size_xe(loc
->xl_entry
) ==
1574 namevalue_size_xi(xi
);
1577 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc
*loc
)
1579 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1580 int i
, count
= le16_to_cpu(xh
->xh_count
);
1581 int offset
, free_start
= loc
->xl_size
;
1583 for (i
= 0; i
< count
; i
++) {
1584 offset
= le16_to_cpu(xh
->xh_entries
[i
].xe_name_offset
);
1585 if (offset
< free_start
)
1586 free_start
= offset
;
1592 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc
*loc
,
1593 struct ocfs2_xattr_info
*xi
)
1595 int count
= le16_to_cpu(loc
->xl_header
->xh_count
);
1596 int free_start
= ocfs2_xa_get_free_start(loc
);
1597 int needed_space
= ocfs2_xi_entry_usage(xi
);
1600 * Block storage will reclaim the original entry before inserting
1601 * the new value, so we only need the difference. If the new
1602 * entry is smaller than the old one, we don't need anything.
1604 if (loc
->xl_entry
) {
1605 /* Don't need space if we're reusing! */
1606 if (ocfs2_xa_can_reuse_entry(loc
, xi
))
1609 needed_space
-= ocfs2_xe_entry_usage(loc
->xl_entry
);
1611 if (needed_space
< 0)
1613 return ocfs2_xa_check_space_helper(needed_space
, free_start
, count
);
1617 * Block storage for xattrs keeps the name+value pairs compacted. When
1618 * we remove one, we have to shift any that preceded it towards the end.
1620 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc
*loc
)
1623 int namevalue_offset
, first_namevalue_offset
, namevalue_size
;
1624 struct ocfs2_xattr_entry
*entry
= loc
->xl_entry
;
1625 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1626 int count
= le16_to_cpu(xh
->xh_count
);
1628 namevalue_offset
= le16_to_cpu(entry
->xe_name_offset
);
1629 namevalue_size
= namevalue_size_xe(entry
);
1630 first_namevalue_offset
= ocfs2_xa_get_free_start(loc
);
1632 /* Shift the name+value pairs */
1633 memmove((char *)xh
+ first_namevalue_offset
+ namevalue_size
,
1634 (char *)xh
+ first_namevalue_offset
,
1635 namevalue_offset
- first_namevalue_offset
);
1636 memset((char *)xh
+ first_namevalue_offset
, 0, namevalue_size
);
1638 /* Now tell xh->xh_entries about it */
1639 for (i
= 0; i
< count
; i
++) {
1640 offset
= le16_to_cpu(xh
->xh_entries
[i
].xe_name_offset
);
1641 if (offset
<= namevalue_offset
)
1642 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
,
1647 * Note that we don't update xh_free_start or xh_name_value_len
1648 * because they're not used in block-stored xattrs.
1652 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc
*loc
, u32 name_hash
)
1654 int count
= le16_to_cpu(loc
->xl_header
->xh_count
);
1655 loc
->xl_entry
= &(loc
->xl_header
->xh_entries
[count
]);
1656 le16_add_cpu(&loc
->xl_header
->xh_count
, 1);
1657 memset(loc
->xl_entry
, 0, sizeof(struct ocfs2_xattr_entry
));
1660 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc
*loc
, int size
)
1662 int free_start
= ocfs2_xa_get_free_start(loc
);
1664 loc
->xl_entry
->xe_name_offset
= cpu_to_le16(free_start
- size
);
1667 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc
*loc
,
1668 struct ocfs2_xattr_value_buf
*vb
)
1670 struct buffer_head
*bh
= loc
->xl_storage
;
1672 if (loc
->xl_size
== (bh
->b_size
-
1673 offsetof(struct ocfs2_xattr_block
,
1674 xb_attrs
.xb_header
)))
1675 vb
->vb_access
= ocfs2_journal_access_xb
;
1677 vb
->vb_access
= ocfs2_journal_access_di
;
1682 * Operations for xattrs stored in blocks. This includes inline inode
1683 * storage and unindexed ocfs2_xattr_blocks.
1685 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops
= {
1686 .xlo_journal_access
= ocfs2_xa_block_journal_access
,
1687 .xlo_journal_dirty
= ocfs2_xa_block_journal_dirty
,
1688 .xlo_offset_pointer
= ocfs2_xa_block_offset_pointer
,
1689 .xlo_check_space
= ocfs2_xa_block_check_space
,
1690 .xlo_can_reuse
= ocfs2_xa_block_can_reuse
,
1691 .xlo_get_free_start
= ocfs2_xa_block_get_free_start
,
1692 .xlo_wipe_namevalue
= ocfs2_xa_block_wipe_namevalue
,
1693 .xlo_add_entry
= ocfs2_xa_block_add_entry
,
1694 .xlo_add_namevalue
= ocfs2_xa_block_add_namevalue
,
1695 .xlo_fill_value_buf
= ocfs2_xa_block_fill_value_buf
,
1698 static int ocfs2_xa_bucket_journal_access(handle_t
*handle
,
1699 struct ocfs2_xa_loc
*loc
, int type
)
1701 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1703 return ocfs2_xattr_bucket_journal_access(handle
, bucket
, type
);
1706 static void ocfs2_xa_bucket_journal_dirty(handle_t
*handle
,
1707 struct ocfs2_xa_loc
*loc
)
1709 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1711 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
1714 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc
*loc
,
1717 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1718 int block
, block_offset
;
1720 /* The header is at the front of the bucket */
1721 block
= offset
>> loc
->xl_inode
->i_sb
->s_blocksize_bits
;
1722 block_offset
= offset
% loc
->xl_inode
->i_sb
->s_blocksize
;
1724 return bucket_block(bucket
, block
) + block_offset
;
1727 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc
*loc
,
1728 struct ocfs2_xattr_info
*xi
)
1730 return namevalue_size_xe(loc
->xl_entry
) >=
1731 namevalue_size_xi(xi
);
1734 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc
*loc
)
1736 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1737 return le16_to_cpu(bucket_xh(bucket
)->xh_free_start
);
1740 static int ocfs2_bucket_align_free_start(struct super_block
*sb
,
1741 int free_start
, int size
)
1744 * We need to make sure that the name+value pair fits within
1747 if (((free_start
- size
) >> sb
->s_blocksize_bits
) !=
1748 ((free_start
- 1) >> sb
->s_blocksize_bits
))
1749 free_start
-= free_start
% sb
->s_blocksize
;
1754 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc
*loc
,
1755 struct ocfs2_xattr_info
*xi
)
1758 int count
= le16_to_cpu(loc
->xl_header
->xh_count
);
1759 int free_start
= ocfs2_xa_get_free_start(loc
);
1760 int needed_space
= ocfs2_xi_entry_usage(xi
);
1761 int size
= namevalue_size_xi(xi
);
1762 struct super_block
*sb
= loc
->xl_inode
->i_sb
;
1765 * Bucket storage does not reclaim name+value pairs it cannot
1766 * reuse. They live as holes until the bucket fills, and then
1767 * the bucket is defragmented. However, the bucket can reclaim
1768 * the ocfs2_xattr_entry.
1770 if (loc
->xl_entry
) {
1771 /* Don't need space if we're reusing! */
1772 if (ocfs2_xa_can_reuse_entry(loc
, xi
))
1775 needed_space
-= sizeof(struct ocfs2_xattr_entry
);
1777 BUG_ON(needed_space
< 0);
1779 if (free_start
< size
) {
1784 * First we check if it would fit in the first place.
1785 * Below, we align the free start to a block. This may
1786 * slide us below the minimum gap. By checking unaligned
1787 * first, we avoid that error.
1789 rc
= ocfs2_xa_check_space_helper(needed_space
, free_start
,
1793 free_start
= ocfs2_bucket_align_free_start(sb
, free_start
,
1796 return ocfs2_xa_check_space_helper(needed_space
, free_start
, count
);
1799 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc
*loc
)
1801 le16_add_cpu(&loc
->xl_header
->xh_name_value_len
,
1802 -namevalue_size_xe(loc
->xl_entry
));
1805 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc
*loc
, u32 name_hash
)
1807 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1808 int count
= le16_to_cpu(xh
->xh_count
);
1809 int low
= 0, high
= count
- 1, tmp
;
1810 struct ocfs2_xattr_entry
*tmp_xe
;
1813 * We keep buckets sorted by name_hash, so we need to find
1816 while (low
<= high
&& count
) {
1817 tmp
= (low
+ high
) / 2;
1818 tmp_xe
= &xh
->xh_entries
[tmp
];
1820 if (name_hash
> le32_to_cpu(tmp_xe
->xe_name_hash
))
1822 else if (name_hash
< le32_to_cpu(tmp_xe
->xe_name_hash
))
1831 memmove(&xh
->xh_entries
[low
+ 1],
1832 &xh
->xh_entries
[low
],
1833 ((count
- low
) * sizeof(struct ocfs2_xattr_entry
)));
1835 le16_add_cpu(&xh
->xh_count
, 1);
1836 loc
->xl_entry
= &xh
->xh_entries
[low
];
1837 memset(loc
->xl_entry
, 0, sizeof(struct ocfs2_xattr_entry
));
1840 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc
*loc
, int size
)
1842 int free_start
= ocfs2_xa_get_free_start(loc
);
1843 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1844 struct super_block
*sb
= loc
->xl_inode
->i_sb
;
1847 free_start
= ocfs2_bucket_align_free_start(sb
, free_start
, size
);
1848 nameval_offset
= free_start
- size
;
1849 loc
->xl_entry
->xe_name_offset
= cpu_to_le16(nameval_offset
);
1850 xh
->xh_free_start
= cpu_to_le16(nameval_offset
);
1851 le16_add_cpu(&xh
->xh_name_value_len
, size
);
1855 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc
*loc
,
1856 struct ocfs2_xattr_value_buf
*vb
)
1858 struct ocfs2_xattr_bucket
*bucket
= loc
->xl_storage
;
1859 struct super_block
*sb
= loc
->xl_inode
->i_sb
;
1860 int nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
1861 int size
= namevalue_size_xe(loc
->xl_entry
);
1862 int block_offset
= nameval_offset
>> sb
->s_blocksize_bits
;
1864 /* Values are not allowed to straddle block boundaries */
1865 BUG_ON(block_offset
!=
1866 ((nameval_offset
+ size
- 1) >> sb
->s_blocksize_bits
));
1867 /* We expect the bucket to be filled in */
1868 BUG_ON(!bucket
->bu_bhs
[block_offset
]);
1870 vb
->vb_access
= ocfs2_journal_access
;
1871 vb
->vb_bh
= bucket
->bu_bhs
[block_offset
];
1874 /* Operations for xattrs stored in buckets. */
1875 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops
= {
1876 .xlo_journal_access
= ocfs2_xa_bucket_journal_access
,
1877 .xlo_journal_dirty
= ocfs2_xa_bucket_journal_dirty
,
1878 .xlo_offset_pointer
= ocfs2_xa_bucket_offset_pointer
,
1879 .xlo_check_space
= ocfs2_xa_bucket_check_space
,
1880 .xlo_can_reuse
= ocfs2_xa_bucket_can_reuse
,
1881 .xlo_get_free_start
= ocfs2_xa_bucket_get_free_start
,
1882 .xlo_wipe_namevalue
= ocfs2_xa_bucket_wipe_namevalue
,
1883 .xlo_add_entry
= ocfs2_xa_bucket_add_entry
,
1884 .xlo_add_namevalue
= ocfs2_xa_bucket_add_namevalue
,
1885 .xlo_fill_value_buf
= ocfs2_xa_bucket_fill_value_buf
,
1888 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc
*loc
)
1890 struct ocfs2_xattr_value_buf vb
;
1892 if (ocfs2_xattr_is_local(loc
->xl_entry
))
1895 ocfs2_xa_fill_value_buf(loc
, &vb
);
1896 return le32_to_cpu(vb
.vb_xv
->xr_clusters
);
1899 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc
*loc
, u64 bytes
,
1900 struct ocfs2_xattr_set_ctxt
*ctxt
)
1902 int trunc_rc
, access_rc
;
1903 struct ocfs2_xattr_value_buf vb
;
1905 ocfs2_xa_fill_value_buf(loc
, &vb
);
1906 trunc_rc
= ocfs2_xattr_value_truncate(loc
->xl_inode
, &vb
, bytes
,
1910 * The caller of ocfs2_xa_value_truncate() has already called
1911 * ocfs2_xa_journal_access on the loc. However, The truncate code
1912 * calls ocfs2_extend_trans(). This may commit the previous
1913 * transaction and open a new one. If this is a bucket, truncate
1914 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1915 * the caller is expecting to dirty the entire bucket. So we must
1916 * reset the journal work. We do this even if truncate has failed,
1917 * as it could have failed after committing the extend.
1919 access_rc
= ocfs2_xa_journal_access(ctxt
->handle
, loc
,
1920 OCFS2_JOURNAL_ACCESS_WRITE
);
1922 /* Errors in truncate take precedence */
1923 return trunc_rc
? trunc_rc
: access_rc
;
1926 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc
*loc
)
1929 struct ocfs2_xattr_header
*xh
= loc
->xl_header
;
1930 struct ocfs2_xattr_entry
*entry
= loc
->xl_entry
;
1932 ocfs2_xa_wipe_namevalue(loc
);
1933 loc
->xl_entry
= NULL
;
1935 le16_add_cpu(&xh
->xh_count
, -1);
1936 count
= le16_to_cpu(xh
->xh_count
);
1939 * Only zero out the entry if there are more remaining. This is
1940 * important for an empty bucket, as it keeps track of the
1941 * bucket's hash value. It doesn't hurt empty block storage.
1944 index
= ((char *)entry
- (char *)&xh
->xh_entries
) /
1945 sizeof(struct ocfs2_xattr_entry
);
1946 memmove(&xh
->xh_entries
[index
], &xh
->xh_entries
[index
+ 1],
1947 (count
- index
) * sizeof(struct ocfs2_xattr_entry
));
1948 memset(&xh
->xh_entries
[count
], 0,
1949 sizeof(struct ocfs2_xattr_entry
));
1954 * If we have a problem adjusting the size of an external value during
1955 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1956 * in an intermediate state. For example, the value may be partially
1959 * If the value tree hasn't changed, the extend/truncate went nowhere.
1960 * We have nothing to do. The caller can treat it as a straight error.
1962 * If the value tree got partially truncated, we now have a corrupted
1963 * extended attribute. We're going to wipe its entry and leak the
1964 * clusters. Better to leak some storage than leave a corrupt entry.
1966 * If the value tree grew, it obviously didn't grow enough for the
1967 * new entry. We're not going to try and reclaim those clusters either.
1968 * If there was already an external value there (orig_clusters != 0),
1969 * the new clusters are attached safely and we can just leave the old
1970 * value in place. If there was no external value there, we remove
1973 * This way, the xattr block we store in the journal will be consistent.
1974 * If the size change broke because of the journal, no changes will hit
1977 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc
*loc
,
1979 unsigned int orig_clusters
)
1981 unsigned int new_clusters
= ocfs2_xa_value_clusters(loc
);
1982 char *nameval_buf
= ocfs2_xa_offset_pointer(loc
,
1983 le16_to_cpu(loc
->xl_entry
->xe_name_offset
));
1985 if (new_clusters
< orig_clusters
) {
1987 "Partial truncate while %s xattr %.*s. Leaking "
1988 "%u clusters and removing the entry\n",
1989 what
, loc
->xl_entry
->xe_name_len
, nameval_buf
,
1990 orig_clusters
- new_clusters
);
1991 ocfs2_xa_remove_entry(loc
);
1992 } else if (!orig_clusters
) {
1994 "Unable to allocate an external value for xattr "
1995 "%.*s safely. Leaking %u clusters and removing the "
1997 loc
->xl_entry
->xe_name_len
, nameval_buf
,
1998 new_clusters
- orig_clusters
);
1999 ocfs2_xa_remove_entry(loc
);
2000 } else if (new_clusters
> orig_clusters
)
2002 "Unable to grow xattr %.*s safely. %u new clusters "
2003 "have been added, but the value will not be "
2005 loc
->xl_entry
->xe_name_len
, nameval_buf
,
2006 new_clusters
- orig_clusters
);
2009 static int ocfs2_xa_remove(struct ocfs2_xa_loc
*loc
,
2010 struct ocfs2_xattr_set_ctxt
*ctxt
)
2013 unsigned int orig_clusters
;
2015 if (!ocfs2_xattr_is_local(loc
->xl_entry
)) {
2016 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2017 rc
= ocfs2_xa_value_truncate(loc
, 0, ctxt
);
2021 * Since this is remove, we can return 0 if
2022 * ocfs2_xa_cleanup_value_truncate() is going to
2023 * wipe the entry anyway. So we check the
2024 * cluster count as well.
2026 if (orig_clusters
!= ocfs2_xa_value_clusters(loc
))
2028 ocfs2_xa_cleanup_value_truncate(loc
, "removing",
2035 ocfs2_xa_remove_entry(loc
);
2041 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc
*loc
)
2043 int name_size
= OCFS2_XATTR_SIZE(loc
->xl_entry
->xe_name_len
);
2046 nameval_buf
= ocfs2_xa_offset_pointer(loc
,
2047 le16_to_cpu(loc
->xl_entry
->xe_name_offset
));
2048 memcpy(nameval_buf
+ name_size
, &def_xv
, OCFS2_XATTR_ROOT_SIZE
);
2052 * Take an existing entry and make it ready for the new value. This
2053 * won't allocate space, but it may free space. It should be ready for
2054 * ocfs2_xa_prepare_entry() to finish the work.
2056 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc
*loc
,
2057 struct ocfs2_xattr_info
*xi
,
2058 struct ocfs2_xattr_set_ctxt
*ctxt
)
2061 int name_size
= OCFS2_XATTR_SIZE(xi
->xi_name_len
);
2062 unsigned int orig_clusters
;
2064 int xe_local
= ocfs2_xattr_is_local(loc
->xl_entry
);
2065 int xi_local
= xi
->xi_value_len
<= OCFS2_XATTR_INLINE_SIZE
;
2067 BUG_ON(OCFS2_XATTR_SIZE(loc
->xl_entry
->xe_name_len
) !=
2070 nameval_buf
= ocfs2_xa_offset_pointer(loc
,
2071 le16_to_cpu(loc
->xl_entry
->xe_name_offset
));
2073 memset(nameval_buf
+ name_size
, 0,
2074 namevalue_size_xe(loc
->xl_entry
) - name_size
);
2076 ocfs2_xa_install_value_root(loc
);
2078 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2080 rc
= ocfs2_xa_value_truncate(loc
, 0, ctxt
);
2084 memset(nameval_buf
+ name_size
, 0,
2085 namevalue_size_xe(loc
->xl_entry
) -
2087 } else if (le64_to_cpu(loc
->xl_entry
->xe_value_size
) >
2089 rc
= ocfs2_xa_value_truncate(loc
, xi
->xi_value_len
,
2096 ocfs2_xa_cleanup_value_truncate(loc
, "reusing",
2102 loc
->xl_entry
->xe_value_size
= cpu_to_le64(xi
->xi_value_len
);
2103 ocfs2_xattr_set_local(loc
->xl_entry
, xi_local
);
2110 * Prepares loc->xl_entry to receive the new xattr. This includes
2111 * properly setting up the name+value pair region. If loc->xl_entry
2112 * already exists, it will take care of modifying it appropriately.
2114 * Note that this modifies the data. You did journal_access already,
2117 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc
*loc
,
2118 struct ocfs2_xattr_info
*xi
,
2120 struct ocfs2_xattr_set_ctxt
*ctxt
)
2123 unsigned int orig_clusters
;
2124 __le64 orig_value_size
= 0;
2126 rc
= ocfs2_xa_check_space(loc
, xi
);
2130 if (loc
->xl_entry
) {
2131 if (ocfs2_xa_can_reuse_entry(loc
, xi
)) {
2132 orig_value_size
= loc
->xl_entry
->xe_value_size
;
2133 rc
= ocfs2_xa_reuse_entry(loc
, xi
, ctxt
);
2139 if (!ocfs2_xattr_is_local(loc
->xl_entry
)) {
2140 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2141 rc
= ocfs2_xa_value_truncate(loc
, 0, ctxt
);
2144 ocfs2_xa_cleanup_value_truncate(loc
,
2150 ocfs2_xa_wipe_namevalue(loc
);
2152 ocfs2_xa_add_entry(loc
, name_hash
);
2155 * If we get here, we have a blank entry. Fill it. We grow our
2156 * name+value pair back from the end.
2158 ocfs2_xa_add_namevalue(loc
, xi
);
2159 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
)
2160 ocfs2_xa_install_value_root(loc
);
2163 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2164 orig_clusters
= ocfs2_xa_value_clusters(loc
);
2165 rc
= ocfs2_xa_value_truncate(loc
, xi
->xi_value_len
, ctxt
);
2167 ctxt
->set_abort
= 1;
2168 ocfs2_xa_cleanup_value_truncate(loc
, "growing",
2171 * If we were growing an existing value,
2172 * ocfs2_xa_cleanup_value_truncate() won't remove
2173 * the entry. We need to restore the original value
2176 if (loc
->xl_entry
) {
2177 BUG_ON(!orig_value_size
);
2178 loc
->xl_entry
->xe_value_size
= orig_value_size
;
2189 * Store the value portion of the name+value pair. This will skip
2190 * values that are stored externally. Their tree roots were set up
2191 * by ocfs2_xa_prepare_entry().
2193 static int ocfs2_xa_store_value(struct ocfs2_xa_loc
*loc
,
2194 struct ocfs2_xattr_info
*xi
,
2195 struct ocfs2_xattr_set_ctxt
*ctxt
)
2198 int nameval_offset
= le16_to_cpu(loc
->xl_entry
->xe_name_offset
);
2199 int name_size
= OCFS2_XATTR_SIZE(xi
->xi_name_len
);
2201 struct ocfs2_xattr_value_buf vb
;
2203 nameval_buf
= ocfs2_xa_offset_pointer(loc
, nameval_offset
);
2204 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
2205 ocfs2_xa_fill_value_buf(loc
, &vb
);
2206 rc
= __ocfs2_xattr_set_value_outside(loc
->xl_inode
,
2211 memcpy(nameval_buf
+ name_size
, xi
->xi_value
, xi
->xi_value_len
);
2216 static int ocfs2_xa_set(struct ocfs2_xa_loc
*loc
,
2217 struct ocfs2_xattr_info
*xi
,
2218 struct ocfs2_xattr_set_ctxt
*ctxt
)
2221 u32 name_hash
= ocfs2_xattr_name_hash(loc
->xl_inode
, xi
->xi_name
,
2224 ret
= ocfs2_xa_journal_access(ctxt
->handle
, loc
,
2225 OCFS2_JOURNAL_ACCESS_WRITE
);
2232 * From here on out, everything is going to modify the buffer a
2233 * little. Errors are going to leave the xattr header in a
2234 * sane state. Thus, even with errors we dirty the sucker.
2237 /* Don't worry, we are never called with !xi_value and !xl_entry */
2238 if (!xi
->xi_value
) {
2239 ret
= ocfs2_xa_remove(loc
, ctxt
);
2243 ret
= ocfs2_xa_prepare_entry(loc
, xi
, name_hash
, ctxt
);
2250 ret
= ocfs2_xa_store_value(loc
, xi
, ctxt
);
2255 ocfs2_xa_journal_dirty(ctxt
->handle
, loc
);
2261 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc
*loc
,
2262 struct inode
*inode
,
2263 struct buffer_head
*bh
,
2264 struct ocfs2_xattr_entry
*entry
)
2266 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
2268 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
));
2270 loc
->xl_inode
= inode
;
2271 loc
->xl_ops
= &ocfs2_xa_block_loc_ops
;
2272 loc
->xl_storage
= bh
;
2273 loc
->xl_entry
= entry
;
2274 loc
->xl_size
= le16_to_cpu(di
->i_xattr_inline_size
);
2276 (struct ocfs2_xattr_header
*)(bh
->b_data
+ bh
->b_size
-
2280 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc
*loc
,
2281 struct inode
*inode
,
2282 struct buffer_head
*bh
,
2283 struct ocfs2_xattr_entry
*entry
)
2285 struct ocfs2_xattr_block
*xb
=
2286 (struct ocfs2_xattr_block
*)bh
->b_data
;
2288 BUG_ON(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
);
2290 loc
->xl_inode
= inode
;
2291 loc
->xl_ops
= &ocfs2_xa_block_loc_ops
;
2292 loc
->xl_storage
= bh
;
2293 loc
->xl_header
= &(xb
->xb_attrs
.xb_header
);
2294 loc
->xl_entry
= entry
;
2295 loc
->xl_size
= bh
->b_size
- offsetof(struct ocfs2_xattr_block
,
2296 xb_attrs
.xb_header
);
2299 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc
*loc
,
2300 struct ocfs2_xattr_bucket
*bucket
,
2301 struct ocfs2_xattr_entry
*entry
)
2303 loc
->xl_inode
= bucket
->bu_inode
;
2304 loc
->xl_ops
= &ocfs2_xa_bucket_loc_ops
;
2305 loc
->xl_storage
= bucket
;
2306 loc
->xl_header
= bucket_xh(bucket
);
2307 loc
->xl_entry
= entry
;
2308 loc
->xl_size
= OCFS2_XATTR_BUCKET_SIZE
;
2312 * In xattr remove, if it is stored outside and refcounted, we may have
2313 * the chance to split the refcount tree. So need the allocators.
2315 static int ocfs2_lock_xattr_remove_allocators(struct inode
*inode
,
2316 struct ocfs2_xattr_value_root
*xv
,
2317 struct ocfs2_caching_info
*ref_ci
,
2318 struct buffer_head
*ref_root_bh
,
2319 struct ocfs2_alloc_context
**meta_ac
,
2322 int ret
, meta_add
= 0;
2323 u32 p_cluster
, num_clusters
;
2324 unsigned int ext_flags
;
2327 ret
= ocfs2_xattr_get_clusters(inode
, 0, &p_cluster
,
2336 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
2339 ret
= ocfs2_refcounted_xattr_delete_need(inode
, ref_ci
,
2341 &meta_add
, ref_credits
);
2347 ret
= ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode
->i_sb
),
2356 static int ocfs2_remove_value_outside(struct inode
*inode
,
2357 struct ocfs2_xattr_value_buf
*vb
,
2358 struct ocfs2_xattr_header
*header
,
2359 struct ocfs2_caching_info
*ref_ci
,
2360 struct buffer_head
*ref_root_bh
)
2362 int ret
= 0, i
, ref_credits
;
2363 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2364 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, };
2367 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
2369 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
2370 struct ocfs2_xattr_entry
*entry
= &header
->xh_entries
[i
];
2372 if (ocfs2_xattr_is_local(entry
))
2375 val
= (void *)header
+
2376 le16_to_cpu(entry
->xe_name_offset
);
2377 vb
->vb_xv
= (struct ocfs2_xattr_value_root
*)
2378 (val
+ OCFS2_XATTR_SIZE(entry
->xe_name_len
));
2380 ret
= ocfs2_lock_xattr_remove_allocators(inode
, vb
->vb_xv
,
2381 ref_ci
, ref_root_bh
,
2385 ctxt
.handle
= ocfs2_start_trans(osb
, ref_credits
+
2386 ocfs2_remove_extent_credits(osb
->sb
));
2387 if (IS_ERR(ctxt
.handle
)) {
2388 ret
= PTR_ERR(ctxt
.handle
);
2393 ret
= ocfs2_xattr_value_truncate(inode
, vb
, 0, &ctxt
);
2395 ocfs2_commit_trans(osb
, ctxt
.handle
);
2397 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2398 ctxt
.meta_ac
= NULL
;
2409 ocfs2_free_alloc_context(ctxt
.meta_ac
);
2410 ocfs2_schedule_truncate_log_flush(osb
, 1);
2411 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
2415 static int ocfs2_xattr_ibody_remove(struct inode
*inode
,
2416 struct buffer_head
*di_bh
,
2417 struct ocfs2_caching_info
*ref_ci
,
2418 struct buffer_head
*ref_root_bh
)
2421 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2422 struct ocfs2_xattr_header
*header
;
2424 struct ocfs2_xattr_value_buf vb
= {
2426 .vb_access
= ocfs2_journal_access_di
,
2429 header
= (struct ocfs2_xattr_header
*)
2430 ((void *)di
+ inode
->i_sb
->s_blocksize
-
2431 le16_to_cpu(di
->i_xattr_inline_size
));
2433 ret
= ocfs2_remove_value_outside(inode
, &vb
, header
,
2434 ref_ci
, ref_root_bh
);
2439 struct ocfs2_rm_xattr_bucket_para
{
2440 struct ocfs2_caching_info
*ref_ci
;
2441 struct buffer_head
*ref_root_bh
;
2444 static int ocfs2_xattr_block_remove(struct inode
*inode
,
2445 struct buffer_head
*blk_bh
,
2446 struct ocfs2_caching_info
*ref_ci
,
2447 struct buffer_head
*ref_root_bh
)
2449 struct ocfs2_xattr_block
*xb
;
2451 struct ocfs2_xattr_value_buf vb
= {
2453 .vb_access
= ocfs2_journal_access_xb
,
2455 struct ocfs2_rm_xattr_bucket_para args
= {
2457 .ref_root_bh
= ref_root_bh
,
2460 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2461 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2462 struct ocfs2_xattr_header
*header
= &(xb
->xb_attrs
.xb_header
);
2463 ret
= ocfs2_remove_value_outside(inode
, &vb
, header
,
2464 ref_ci
, ref_root_bh
);
2466 ret
= ocfs2_iterate_xattr_index_block(inode
,
2468 ocfs2_rm_xattr_cluster
,
2474 static int ocfs2_xattr_free_block(struct inode
*inode
,
2476 struct ocfs2_caching_info
*ref_ci
,
2477 struct buffer_head
*ref_root_bh
)
2479 struct inode
*xb_alloc_inode
;
2480 struct buffer_head
*xb_alloc_bh
= NULL
;
2481 struct buffer_head
*blk_bh
= NULL
;
2482 struct ocfs2_xattr_block
*xb
;
2483 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2489 ret
= ocfs2_read_xattr_block(inode
, block
, &blk_bh
);
2495 ret
= ocfs2_xattr_block_remove(inode
, blk_bh
, ref_ci
, ref_root_bh
);
2501 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2502 blk
= le64_to_cpu(xb
->xb_blkno
);
2503 bit
= le16_to_cpu(xb
->xb_suballoc_bit
);
2504 if (xb
->xb_suballoc_loc
)
2505 bg_blkno
= le64_to_cpu(xb
->xb_suballoc_loc
);
2507 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
2509 xb_alloc_inode
= ocfs2_get_system_file_inode(osb
,
2510 EXTENT_ALLOC_SYSTEM_INODE
,
2511 le16_to_cpu(xb
->xb_suballoc_slot
));
2512 if (!xb_alloc_inode
) {
2517 inode_lock(xb_alloc_inode
);
2519 ret
= ocfs2_inode_lock(xb_alloc_inode
, &xb_alloc_bh
, 1);
2525 handle
= ocfs2_start_trans(osb
, OCFS2_SUBALLOC_FREE
);
2526 if (IS_ERR(handle
)) {
2527 ret
= PTR_ERR(handle
);
2532 ret
= ocfs2_free_suballoc_bits(handle
, xb_alloc_inode
, xb_alloc_bh
,
2537 ocfs2_commit_trans(osb
, handle
);
2539 ocfs2_inode_unlock(xb_alloc_inode
, 1);
2540 brelse(xb_alloc_bh
);
2542 inode_unlock(xb_alloc_inode
);
2543 iput(xb_alloc_inode
);
2550 * ocfs2_xattr_remove()
2552 * Free extended attribute resources associated with this inode.
2554 int ocfs2_xattr_remove(struct inode
*inode
, struct buffer_head
*di_bh
)
2556 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2557 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2558 struct ocfs2_refcount_tree
*ref_tree
= NULL
;
2559 struct buffer_head
*ref_root_bh
= NULL
;
2560 struct ocfs2_caching_info
*ref_ci
= NULL
;
2564 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
2567 if (!(oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
))
2570 if (ocfs2_is_refcount_inode(inode
)) {
2571 ret
= ocfs2_lock_refcount_tree(OCFS2_SB(inode
->i_sb
),
2572 le64_to_cpu(di
->i_refcount_loc
),
2573 1, &ref_tree
, &ref_root_bh
);
2578 ref_ci
= &ref_tree
->rf_ci
;
2582 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
2583 ret
= ocfs2_xattr_ibody_remove(inode
, di_bh
,
2584 ref_ci
, ref_root_bh
);
2591 if (di
->i_xattr_loc
) {
2592 ret
= ocfs2_xattr_free_block(inode
,
2593 le64_to_cpu(di
->i_xattr_loc
),
2594 ref_ci
, ref_root_bh
);
2601 handle
= ocfs2_start_trans((OCFS2_SB(inode
->i_sb
)),
2602 OCFS2_INODE_UPDATE_CREDITS
);
2603 if (IS_ERR(handle
)) {
2604 ret
= PTR_ERR(handle
);
2608 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
2609 OCFS2_JOURNAL_ACCESS_WRITE
);
2615 di
->i_xattr_loc
= 0;
2617 spin_lock(&oi
->ip_lock
);
2618 oi
->ip_dyn_features
&= ~(OCFS2_INLINE_XATTR_FL
| OCFS2_HAS_XATTR_FL
);
2619 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
2620 spin_unlock(&oi
->ip_lock
);
2621 ocfs2_update_inode_fsync_trans(handle
, inode
, 0);
2623 ocfs2_journal_dirty(handle
, di_bh
);
2625 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
2628 ocfs2_unlock_refcount_tree(OCFS2_SB(inode
->i_sb
), ref_tree
, 1);
2629 brelse(ref_root_bh
);
2633 static int ocfs2_xattr_has_space_inline(struct inode
*inode
,
2634 struct ocfs2_dinode
*di
)
2636 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2637 unsigned int xattrsize
= OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
;
2640 if (xattrsize
< OCFS2_MIN_XATTR_INLINE_SIZE
)
2643 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2644 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
2645 free
= le16_to_cpu(idata
->id_count
) - le64_to_cpu(di
->i_size
);
2646 } else if (ocfs2_inode_is_fast_symlink(inode
)) {
2647 free
= ocfs2_fast_symlink_chars(inode
->i_sb
) -
2648 le64_to_cpu(di
->i_size
);
2650 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
2651 free
= (le16_to_cpu(el
->l_count
) -
2652 le16_to_cpu(el
->l_next_free_rec
)) *
2653 sizeof(struct ocfs2_extent_rec
);
2655 if (free
>= xattrsize
)
2662 * ocfs2_xattr_ibody_find()
2664 * Find extended attribute in inode block and
2665 * fill search info into struct ocfs2_xattr_search.
2667 static int ocfs2_xattr_ibody_find(struct inode
*inode
,
2670 struct ocfs2_xattr_search
*xs
)
2672 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2673 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2677 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
2680 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
2681 down_read(&oi
->ip_alloc_sem
);
2682 has_space
= ocfs2_xattr_has_space_inline(inode
, di
);
2683 up_read(&oi
->ip_alloc_sem
);
2688 xs
->xattr_bh
= xs
->inode_bh
;
2689 xs
->end
= (void *)di
+ inode
->i_sb
->s_blocksize
;
2690 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)
2691 xs
->header
= (struct ocfs2_xattr_header
*)
2692 (xs
->end
- le16_to_cpu(di
->i_xattr_inline_size
));
2694 xs
->header
= (struct ocfs2_xattr_header
*)
2695 (xs
->end
- OCFS2_SB(inode
->i_sb
)->s_xattr_inline_size
);
2696 xs
->base
= (void *)xs
->header
;
2697 xs
->here
= xs
->header
->xh_entries
;
2699 /* Find the named attribute. */
2700 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
2701 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
2702 if (ret
&& ret
!= -ENODATA
)
2704 xs
->not_found
= ret
;
2710 static int ocfs2_xattr_ibody_init(struct inode
*inode
,
2711 struct buffer_head
*di_bh
,
2712 struct ocfs2_xattr_set_ctxt
*ctxt
)
2715 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2716 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2717 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2718 unsigned int xattrsize
= osb
->s_xattr_inline_size
;
2720 if (!ocfs2_xattr_has_space_inline(inode
, di
)) {
2725 ret
= ocfs2_journal_access_di(ctxt
->handle
, INODE_CACHE(inode
), di_bh
,
2726 OCFS2_JOURNAL_ACCESS_WRITE
);
2733 * Adjust extent record count or inline data size
2734 * to reserve space for extended attribute.
2736 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2737 struct ocfs2_inline_data
*idata
= &di
->id2
.i_data
;
2738 le16_add_cpu(&idata
->id_count
, -xattrsize
);
2739 } else if (!(ocfs2_inode_is_fast_symlink(inode
))) {
2740 struct ocfs2_extent_list
*el
= &di
->id2
.i_list
;
2741 le16_add_cpu(&el
->l_count
, -(xattrsize
/
2742 sizeof(struct ocfs2_extent_rec
)));
2744 di
->i_xattr_inline_size
= cpu_to_le16(xattrsize
);
2746 spin_lock(&oi
->ip_lock
);
2747 oi
->ip_dyn_features
|= OCFS2_INLINE_XATTR_FL
|OCFS2_HAS_XATTR_FL
;
2748 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
2749 spin_unlock(&oi
->ip_lock
);
2751 ocfs2_journal_dirty(ctxt
->handle
, di_bh
);
2758 * ocfs2_xattr_ibody_set()
2760 * Set, replace or remove an extended attribute into inode block.
2763 static int ocfs2_xattr_ibody_set(struct inode
*inode
,
2764 struct ocfs2_xattr_info
*xi
,
2765 struct ocfs2_xattr_search
*xs
,
2766 struct ocfs2_xattr_set_ctxt
*ctxt
)
2769 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2770 struct ocfs2_xa_loc loc
;
2772 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
)
2775 down_write(&oi
->ip_alloc_sem
);
2776 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
)) {
2777 ret
= ocfs2_xattr_ibody_init(inode
, xs
->inode_bh
, ctxt
);
2785 ocfs2_init_dinode_xa_loc(&loc
, inode
, xs
->inode_bh
,
2786 xs
->not_found
? NULL
: xs
->here
);
2787 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
2793 xs
->here
= loc
.xl_entry
;
2796 up_write(&oi
->ip_alloc_sem
);
2802 * ocfs2_xattr_block_find()
2804 * Find extended attribute in external block and
2805 * fill search info into struct ocfs2_xattr_search.
2807 static int ocfs2_xattr_block_find(struct inode
*inode
,
2810 struct ocfs2_xattr_search
*xs
)
2812 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)xs
->inode_bh
->b_data
;
2813 struct buffer_head
*blk_bh
= NULL
;
2814 struct ocfs2_xattr_block
*xb
;
2817 if (!di
->i_xattr_loc
)
2820 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
2827 xs
->xattr_bh
= blk_bh
;
2828 xb
= (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
2830 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2831 xs
->header
= &xb
->xb_attrs
.xb_header
;
2832 xs
->base
= (void *)xs
->header
;
2833 xs
->end
= (void *)(blk_bh
->b_data
) + blk_bh
->b_size
;
2834 xs
->here
= xs
->header
->xh_entries
;
2836 ret
= ocfs2_xattr_find_entry(name_index
, name
, xs
);
2838 ret
= ocfs2_xattr_index_block_find(inode
, blk_bh
,
2842 if (ret
&& ret
!= -ENODATA
) {
2843 xs
->xattr_bh
= NULL
;
2846 xs
->not_found
= ret
;
2854 static int ocfs2_create_xattr_block(struct inode
*inode
,
2855 struct buffer_head
*inode_bh
,
2856 struct ocfs2_xattr_set_ctxt
*ctxt
,
2858 struct buffer_head
**ret_bh
)
2861 u16 suballoc_bit_start
;
2863 u64 suballoc_loc
, first_blkno
;
2864 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)inode_bh
->b_data
;
2865 struct buffer_head
*new_bh
= NULL
;
2866 struct ocfs2_xattr_block
*xblk
;
2868 ret
= ocfs2_journal_access_di(ctxt
->handle
, INODE_CACHE(inode
),
2869 inode_bh
, OCFS2_JOURNAL_ACCESS_CREATE
);
2875 ret
= ocfs2_claim_metadata(ctxt
->handle
, ctxt
->meta_ac
, 1,
2876 &suballoc_loc
, &suballoc_bit_start
,
2877 &num_got
, &first_blkno
);
2883 new_bh
= sb_getblk(inode
->i_sb
, first_blkno
);
2890 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), new_bh
);
2892 ret
= ocfs2_journal_access_xb(ctxt
->handle
, INODE_CACHE(inode
),
2894 OCFS2_JOURNAL_ACCESS_CREATE
);
2900 /* Initialize ocfs2_xattr_block */
2901 xblk
= (struct ocfs2_xattr_block
*)new_bh
->b_data
;
2902 memset(xblk
, 0, inode
->i_sb
->s_blocksize
);
2903 strcpy((void *)xblk
, OCFS2_XATTR_BLOCK_SIGNATURE
);
2904 xblk
->xb_suballoc_slot
= cpu_to_le16(ctxt
->meta_ac
->ac_alloc_slot
);
2905 xblk
->xb_suballoc_loc
= cpu_to_le64(suballoc_loc
);
2906 xblk
->xb_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
2907 xblk
->xb_fs_generation
=
2908 cpu_to_le32(OCFS2_SB(inode
->i_sb
)->fs_generation
);
2909 xblk
->xb_blkno
= cpu_to_le64(first_blkno
);
2911 struct ocfs2_xattr_tree_root
*xr
= &xblk
->xb_attrs
.xb_root
;
2912 xr
->xt_clusters
= cpu_to_le32(1);
2913 xr
->xt_last_eb_blk
= 0;
2914 xr
->xt_list
.l_tree_depth
= 0;
2915 xr
->xt_list
.l_count
= cpu_to_le16(
2916 ocfs2_xattr_recs_per_xb(inode
->i_sb
));
2917 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
2918 xblk
->xb_flags
= cpu_to_le16(OCFS2_XATTR_INDEXED
);
2920 ocfs2_journal_dirty(ctxt
->handle
, new_bh
);
2922 /* Add it to the inode */
2923 di
->i_xattr_loc
= cpu_to_le64(first_blkno
);
2925 spin_lock(&OCFS2_I(inode
)->ip_lock
);
2926 OCFS2_I(inode
)->ip_dyn_features
|= OCFS2_HAS_XATTR_FL
;
2927 di
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
2928 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
2930 ocfs2_journal_dirty(ctxt
->handle
, inode_bh
);
2941 * ocfs2_xattr_block_set()
2943 * Set, replace or remove an extended attribute into external block.
2946 static int ocfs2_xattr_block_set(struct inode
*inode
,
2947 struct ocfs2_xattr_info
*xi
,
2948 struct ocfs2_xattr_search
*xs
,
2949 struct ocfs2_xattr_set_ctxt
*ctxt
)
2951 struct buffer_head
*new_bh
= NULL
;
2952 struct ocfs2_xattr_block
*xblk
= NULL
;
2954 struct ocfs2_xa_loc loc
;
2956 if (!xs
->xattr_bh
) {
2957 ret
= ocfs2_create_xattr_block(inode
, xs
->inode_bh
, ctxt
,
2964 xs
->xattr_bh
= new_bh
;
2965 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2966 xs
->header
= &xblk
->xb_attrs
.xb_header
;
2967 xs
->base
= (void *)xs
->header
;
2968 xs
->end
= (void *)xblk
+ inode
->i_sb
->s_blocksize
;
2969 xs
->here
= xs
->header
->xh_entries
;
2971 xblk
= (struct ocfs2_xattr_block
*)xs
->xattr_bh
->b_data
;
2973 if (!(le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
2974 ocfs2_init_xattr_block_xa_loc(&loc
, inode
, xs
->xattr_bh
,
2975 xs
->not_found
? NULL
: xs
->here
);
2977 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
2979 xs
->here
= loc
.xl_entry
;
2980 else if ((ret
!= -ENOSPC
) || ctxt
->set_abort
)
2983 ret
= ocfs2_xattr_create_index_block(inode
, xs
, ctxt
);
2989 if (le16_to_cpu(xblk
->xb_flags
) & OCFS2_XATTR_INDEXED
)
2990 ret
= ocfs2_xattr_set_entry_index_block(inode
, xi
, xs
, ctxt
);
2996 /* Check whether the new xattr can be inserted into the inode. */
2997 static int ocfs2_xattr_can_be_in_inode(struct inode
*inode
,
2998 struct ocfs2_xattr_info
*xi
,
2999 struct ocfs2_xattr_search
*xs
)
3001 struct ocfs2_xattr_entry
*last
;
3003 size_t min_offs
= xs
->end
- xs
->base
;
3008 last
= xs
->header
->xh_entries
;
3010 for (i
= 0; i
< le16_to_cpu(xs
->header
->xh_count
); i
++) {
3011 size_t offs
= le16_to_cpu(last
->xe_name_offset
);
3012 if (offs
< min_offs
)
3017 free
= min_offs
- ((void *)last
- xs
->base
) - OCFS2_XATTR_HEADER_GAP
;
3021 BUG_ON(!xs
->not_found
);
3023 if (free
>= (sizeof(struct ocfs2_xattr_entry
) + namevalue_size_xi(xi
)))
3029 static int ocfs2_calc_xattr_set_need(struct inode
*inode
,
3030 struct ocfs2_dinode
*di
,
3031 struct ocfs2_xattr_info
*xi
,
3032 struct ocfs2_xattr_search
*xis
,
3033 struct ocfs2_xattr_search
*xbs
,
3038 int ret
= 0, old_in_xb
= 0;
3039 int clusters_add
= 0, meta_add
= 0, credits
= 0;
3040 struct buffer_head
*bh
= NULL
;
3041 struct ocfs2_xattr_block
*xb
= NULL
;
3042 struct ocfs2_xattr_entry
*xe
= NULL
;
3043 struct ocfs2_xattr_value_root
*xv
= NULL
;
3045 int name_offset
, name_len
= 0;
3046 u32 new_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
3051 * Calculate the clusters we need to write.
3052 * No matter whether we replace an old one or add a new one,
3053 * we need this for writing.
3055 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
)
3056 credits
+= new_clusters
*
3057 ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
3059 if (xis
->not_found
&& xbs
->not_found
) {
3060 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3062 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
3063 clusters_add
+= new_clusters
;
3064 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3065 &def_xv
.xv
.xr_list
);
3071 if (!xis
->not_found
) {
3073 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3074 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3076 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
3078 int i
, block_off
= 0;
3079 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
3081 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
3082 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
3083 i
= xbs
->here
- xbs
->header
->xh_entries
;
3086 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
3087 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
3088 bucket_xh(xbs
->bucket
),
3091 base
= bucket_block(xbs
->bucket
, block_off
);
3092 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3095 credits
+= OCFS2_XATTR_BLOCK_UPDATE_CREDITS
;
3100 * delete a xattr doesn't need metadata and cluster allocation.
3101 * so just calculate the credits and return.
3103 * The credits for removing the value tree will be extended
3104 * by ocfs2_remove_extent itself.
3106 if (!xi
->xi_value
) {
3107 if (!ocfs2_xattr_is_local(xe
))
3108 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
);
3113 /* do cluster allocation guess first. */
3114 value_size
= le64_to_cpu(xe
->xe_value_size
);
3118 * In xattr set, we always try to set the xe in inode first,
3119 * so if it can be inserted into inode successfully, the old
3120 * one will be removed from the xattr block, and this xattr
3121 * will be inserted into inode as a new xattr in inode.
3123 if (ocfs2_xattr_can_be_in_inode(inode
, xi
, xis
)) {
3124 clusters_add
+= new_clusters
;
3125 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
) +
3126 OCFS2_INODE_UPDATE_CREDITS
;
3127 if (!ocfs2_xattr_is_local(xe
))
3128 credits
+= ocfs2_calc_extend_credits(
3130 &def_xv
.xv
.xr_list
);
3135 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
3136 /* the new values will be stored outside. */
3137 u32 old_clusters
= 0;
3139 if (!ocfs2_xattr_is_local(xe
)) {
3140 old_clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
,
3142 xv
= (struct ocfs2_xattr_value_root
*)
3143 (base
+ name_offset
+ name_len
);
3144 value_size
= OCFS2_XATTR_ROOT_SIZE
;
3148 if (old_clusters
>= new_clusters
) {
3149 credits
+= ocfs2_remove_extent_credits(inode
->i_sb
);
3152 meta_add
+= ocfs2_extend_meta_needed(&xv
->xr_list
);
3153 clusters_add
+= new_clusters
- old_clusters
;
3154 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3156 if (value_size
>= OCFS2_XATTR_ROOT_SIZE
)
3161 * Now the new value will be stored inside. So if the new
3162 * value is smaller than the size of value root or the old
3163 * value, we don't need any allocation, otherwise we have
3164 * to guess metadata allocation.
3166 if ((ocfs2_xattr_is_local(xe
) &&
3167 (value_size
>= xi
->xi_value_len
)) ||
3168 (!ocfs2_xattr_is_local(xe
) &&
3169 OCFS2_XATTR_ROOT_SIZE
>= xi
->xi_value_len
))
3174 /* calculate metadata allocation. */
3175 if (di
->i_xattr_loc
) {
3176 if (!xbs
->xattr_bh
) {
3177 ret
= ocfs2_read_xattr_block(inode
,
3178 le64_to_cpu(di
->i_xattr_loc
),
3185 xb
= (struct ocfs2_xattr_block
*)bh
->b_data
;
3187 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
3190 * If there is already an xattr tree, good, we can calculate
3191 * like other b-trees. Otherwise we may have the chance of
3192 * create a tree, the credit calculation is borrowed from
3193 * ocfs2_calc_extend_credits with root_el = NULL. And the
3194 * new tree will be cluster based, so no meta is needed.
3196 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
3197 struct ocfs2_extent_list
*el
=
3198 &xb
->xb_attrs
.xb_root
.xt_list
;
3199 meta_add
+= ocfs2_extend_meta_needed(el
);
3200 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3203 credits
+= OCFS2_SUBALLOC_ALLOC
+ 1;
3206 * This cluster will be used either for new bucket or for
3208 * If the cluster size is the same as the bucket size, one
3209 * more is needed since we may need to extend the bucket
3213 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3214 if (OCFS2_XATTR_BUCKET_SIZE
==
3215 OCFS2_SB(inode
->i_sb
)->s_clustersize
) {
3216 credits
+= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3220 credits
+= OCFS2_XATTR_BLOCK_CREATE_CREDITS
;
3221 if (xi
->xi_value_len
> OCFS2_XATTR_INLINE_SIZE
) {
3222 struct ocfs2_extent_list
*el
= &def_xv
.xv
.xr_list
;
3223 meta_add
+= ocfs2_extend_meta_needed(el
);
3224 credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3232 *clusters_need
= clusters_add
;
3234 *meta_need
= meta_add
;
3236 *credits_need
= credits
;
3241 static int ocfs2_init_xattr_set_ctxt(struct inode
*inode
,
3242 struct ocfs2_dinode
*di
,
3243 struct ocfs2_xattr_info
*xi
,
3244 struct ocfs2_xattr_search
*xis
,
3245 struct ocfs2_xattr_search
*xbs
,
3246 struct ocfs2_xattr_set_ctxt
*ctxt
,
3250 int clusters_add
, meta_add
, ret
;
3251 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3253 memset(ctxt
, 0, sizeof(struct ocfs2_xattr_set_ctxt
));
3255 ocfs2_init_dealloc_ctxt(&ctxt
->dealloc
);
3257 ret
= ocfs2_calc_xattr_set_need(inode
, di
, xi
, xis
, xbs
,
3258 &clusters_add
, &meta_add
, credits
);
3264 meta_add
+= extra_meta
;
3265 trace_ocfs2_init_xattr_set_ctxt(xi
->xi_name
, meta_add
,
3266 clusters_add
, *credits
);
3269 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
,
3278 ret
= ocfs2_reserve_clusters(osb
, clusters_add
, &ctxt
->data_ac
);
3284 if (ctxt
->meta_ac
) {
3285 ocfs2_free_alloc_context(ctxt
->meta_ac
);
3286 ctxt
->meta_ac
= NULL
;
3290 * We cannot have an error and a non null ctxt->data_ac.
3297 static int __ocfs2_xattr_set_handle(struct inode
*inode
,
3298 struct ocfs2_dinode
*di
,
3299 struct ocfs2_xattr_info
*xi
,
3300 struct ocfs2_xattr_search
*xis
,
3301 struct ocfs2_xattr_search
*xbs
,
3302 struct ocfs2_xattr_set_ctxt
*ctxt
)
3304 int ret
= 0, credits
, old_found
;
3306 if (!xi
->xi_value
) {
3307 /* Remove existing extended attribute */
3308 if (!xis
->not_found
)
3309 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
3310 else if (!xbs
->not_found
)
3311 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
3313 /* We always try to set extended attribute into inode first*/
3314 ret
= ocfs2_xattr_ibody_set(inode
, xi
, xis
, ctxt
);
3315 if (!ret
&& !xbs
->not_found
) {
3317 * If succeed and that extended attribute existing in
3318 * external block, then we will remove it.
3320 xi
->xi_value
= NULL
;
3321 xi
->xi_value_len
= 0;
3323 old_found
= xis
->not_found
;
3324 xis
->not_found
= -ENODATA
;
3325 ret
= ocfs2_calc_xattr_set_need(inode
,
3333 xis
->not_found
= old_found
;
3339 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
);
3344 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
3345 } else if ((ret
== -ENOSPC
) && !ctxt
->set_abort
) {
3346 if (di
->i_xattr_loc
&& !xbs
->xattr_bh
) {
3347 ret
= ocfs2_xattr_block_find(inode
,
3353 old_found
= xis
->not_found
;
3354 xis
->not_found
= -ENODATA
;
3355 ret
= ocfs2_calc_xattr_set_need(inode
,
3363 xis
->not_found
= old_found
;
3369 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
);
3376 * If no space in inode, we will set extended attribute
3377 * into external block.
3379 ret
= ocfs2_xattr_block_set(inode
, xi
, xbs
, ctxt
);
3382 if (!xis
->not_found
) {
3384 * If succeed and that extended attribute
3385 * existing in inode, we will remove it.
3387 xi
->xi_value
= NULL
;
3388 xi
->xi_value_len
= 0;
3389 xbs
->not_found
= -ENODATA
;
3390 ret
= ocfs2_calc_xattr_set_need(inode
,
3403 ret
= ocfs2_extend_trans(ctxt
->handle
, credits
);
3408 ret
= ocfs2_xattr_ibody_set(inode
, xi
,
3415 /* Update inode ctime. */
3416 ret
= ocfs2_journal_access_di(ctxt
->handle
, INODE_CACHE(inode
),
3418 OCFS2_JOURNAL_ACCESS_WRITE
);
3424 inode
->i_ctime
= current_time(inode
);
3425 di
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
3426 di
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
3427 ocfs2_journal_dirty(ctxt
->handle
, xis
->inode_bh
);
3434 * This function only called duing creating inode
3435 * for init security/acl xattrs of the new inode.
3436 * All transanction credits have been reserved in mknod.
3438 int ocfs2_xattr_set_handle(handle_t
*handle
,
3439 struct inode
*inode
,
3440 struct buffer_head
*di_bh
,
3446 struct ocfs2_alloc_context
*meta_ac
,
3447 struct ocfs2_alloc_context
*data_ac
)
3449 struct ocfs2_dinode
*di
;
3452 struct ocfs2_xattr_info xi
= {
3453 .xi_name_index
= name_index
,
3455 .xi_name_len
= strlen(name
),
3457 .xi_value_len
= value_len
,
3460 struct ocfs2_xattr_search xis
= {
3461 .not_found
= -ENODATA
,
3464 struct ocfs2_xattr_search xbs
= {
3465 .not_found
= -ENODATA
,
3468 struct ocfs2_xattr_set_ctxt ctxt
= {
3474 if (!ocfs2_supports_xattr(OCFS2_SB(inode
->i_sb
)))
3478 * In extreme situation, may need xattr bucket when
3479 * block size is too small. And we have already reserved
3480 * the credits for bucket in mknod.
3482 if (inode
->i_sb
->s_blocksize
== OCFS2_MIN_BLOCKSIZE
) {
3483 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
3485 mlog_errno(-ENOMEM
);
3490 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
3491 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3493 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3495 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
3498 if (xis
.not_found
) {
3499 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
3504 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
3507 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3508 brelse(xbs
.xattr_bh
);
3509 ocfs2_xattr_bucket_free(xbs
.bucket
);
3517 * Set, replace or remove an extended attribute for this inode.
3518 * value is NULL to remove an existing extended attribute, else either
3519 * create or replace an extended attribute.
3521 int ocfs2_xattr_set(struct inode
*inode
,
3528 struct buffer_head
*di_bh
= NULL
;
3529 struct ocfs2_dinode
*di
;
3530 int ret
, credits
, had_lock
, ref_meta
= 0, ref_credits
= 0;
3531 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3532 struct inode
*tl_inode
= osb
->osb_tl_inode
;
3533 struct ocfs2_xattr_set_ctxt ctxt
= { NULL
, NULL
, NULL
, };
3534 struct ocfs2_refcount_tree
*ref_tree
= NULL
;
3535 struct ocfs2_lock_holder oh
;
3537 struct ocfs2_xattr_info xi
= {
3538 .xi_name_index
= name_index
,
3540 .xi_name_len
= strlen(name
),
3542 .xi_value_len
= value_len
,
3545 struct ocfs2_xattr_search xis
= {
3546 .not_found
= -ENODATA
,
3549 struct ocfs2_xattr_search xbs
= {
3550 .not_found
= -ENODATA
,
3553 if (!ocfs2_supports_xattr(osb
))
3557 * Only xbs will be used on indexed trees. xis doesn't need a
3560 xbs
.bucket
= ocfs2_xattr_bucket_new(inode
);
3562 mlog_errno(-ENOMEM
);
3566 had_lock
= ocfs2_inode_lock_tracker(inode
, &di_bh
, 1, &oh
);
3570 goto cleanup_nolock
;
3572 xis
.inode_bh
= xbs
.inode_bh
= di_bh
;
3573 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3575 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3577 * Scan inode and external block to find the same name
3578 * extended attribute and collect search information.
3580 ret
= ocfs2_xattr_ibody_find(inode
, name_index
, name
, &xis
);
3583 if (xis
.not_found
) {
3584 ret
= ocfs2_xattr_block_find(inode
, name_index
, name
, &xbs
);
3589 if (xis
.not_found
&& xbs
.not_found
) {
3591 if (flags
& XATTR_REPLACE
)
3598 if (flags
& XATTR_CREATE
)
3602 /* Check whether the value is refcounted and do some preparation. */
3603 if (ocfs2_is_refcount_inode(inode
) &&
3604 (!xis
.not_found
|| !xbs
.not_found
)) {
3605 ret
= ocfs2_prepare_refcount_xattr(inode
, di
, &xi
,
3606 &xis
, &xbs
, &ref_tree
,
3607 &ref_meta
, &ref_credits
);
3614 inode_lock(tl_inode
);
3616 if (ocfs2_truncate_log_needs_flush(osb
)) {
3617 ret
= __ocfs2_flush_truncate_log(osb
);
3619 inode_unlock(tl_inode
);
3624 inode_unlock(tl_inode
);
3626 ret
= ocfs2_init_xattr_set_ctxt(inode
, di
, &xi
, &xis
,
3627 &xbs
, &ctxt
, ref_meta
, &credits
);
3633 /* we need to update inode's ctime field, so add credit for it. */
3634 credits
+= OCFS2_INODE_UPDATE_CREDITS
;
3635 ctxt
.handle
= ocfs2_start_trans(osb
, credits
+ ref_credits
);
3636 if (IS_ERR(ctxt
.handle
)) {
3637 ret
= PTR_ERR(ctxt
.handle
);
3642 ret
= __ocfs2_xattr_set_handle(inode
, di
, &xi
, &xis
, &xbs
, &ctxt
);
3643 ocfs2_update_inode_fsync_trans(ctxt
.handle
, inode
, 0);
3645 ocfs2_commit_trans(osb
, ctxt
.handle
);
3649 ocfs2_free_alloc_context(ctxt
.data_ac
);
3651 ocfs2_free_alloc_context(ctxt
.meta_ac
);
3652 if (ocfs2_dealloc_has_cluster(&ctxt
.dealloc
))
3653 ocfs2_schedule_truncate_log_flush(osb
, 1);
3654 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
3658 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
3659 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
3660 if (!value
&& !ret
) {
3661 ret
= ocfs2_try_remove_refcount_tree(inode
, di_bh
);
3665 ocfs2_inode_unlock_tracker(inode
, 1, &oh
, had_lock
);
3668 brelse(xbs
.xattr_bh
);
3669 ocfs2_xattr_bucket_free(xbs
.bucket
);
3675 * Find the xattr extent rec which may contains name_hash.
3676 * e_cpos will be the first name hash of the xattr rec.
3677 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3679 static int ocfs2_xattr_get_rec(struct inode
*inode
,
3684 struct ocfs2_extent_list
*el
)
3687 struct buffer_head
*eb_bh
= NULL
;
3688 struct ocfs2_extent_block
*eb
;
3689 struct ocfs2_extent_rec
*rec
= NULL
;
3692 if (el
->l_tree_depth
) {
3693 ret
= ocfs2_find_leaf(INODE_CACHE(inode
), el
, name_hash
,
3700 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
3703 if (el
->l_tree_depth
) {
3704 ret
= ocfs2_error(inode
->i_sb
,
3705 "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3707 (unsigned long long)eb_bh
->b_blocknr
);
3712 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
3713 rec
= &el
->l_recs
[i
];
3715 if (le32_to_cpu(rec
->e_cpos
) <= name_hash
) {
3716 e_blkno
= le64_to_cpu(rec
->e_blkno
);
3722 ret
= ocfs2_error(inode
->i_sb
, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3724 le32_to_cpu(rec
->e_cpos
),
3725 ocfs2_rec_clusters(el
, rec
));
3729 *p_blkno
= le64_to_cpu(rec
->e_blkno
);
3730 *num_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
3732 *e_cpos
= le32_to_cpu(rec
->e_cpos
);
3738 typedef int (xattr_bucket_func
)(struct inode
*inode
,
3739 struct ocfs2_xattr_bucket
*bucket
,
3742 static int ocfs2_find_xe_in_bucket(struct inode
*inode
,
3743 struct ocfs2_xattr_bucket
*bucket
,
3750 int i
, ret
= 0, cmp
= 1, block_off
, new_offset
;
3751 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
3752 size_t name_len
= strlen(name
);
3753 struct ocfs2_xattr_entry
*xe
= NULL
;
3757 * We don't use binary search in the bucket because there
3758 * may be multiple entries with the same name hash.
3760 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
3761 xe
= &xh
->xh_entries
[i
];
3763 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
))
3765 else if (name_hash
< le32_to_cpu(xe
->xe_name_hash
))
3768 cmp
= name_index
- ocfs2_xattr_get_type(xe
);
3770 cmp
= name_len
- xe
->xe_name_len
;
3774 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
3785 xe_name
= bucket_block(bucket
, block_off
) + new_offset
;
3786 if (!memcmp(name
, xe_name
, name_len
)) {
3798 * Find the specified xattr entry in a series of buckets.
3799 * This series start from p_blkno and last for num_clusters.
3800 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3801 * the num of the valid buckets.
3803 * Return the buffer_head this xattr should reside in. And if the xattr's
3804 * hash is in the gap of 2 buckets, return the lower bucket.
3806 static int ocfs2_xattr_bucket_find(struct inode
*inode
,
3813 struct ocfs2_xattr_search
*xs
)
3816 struct ocfs2_xattr_header
*xh
= NULL
;
3817 struct ocfs2_xattr_entry
*xe
= NULL
;
3819 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
3820 int low_bucket
= 0, bucket
, high_bucket
;
3821 struct ocfs2_xattr_bucket
*search
;
3822 u64 blkno
, lower_blkno
= 0;
3824 search
= ocfs2_xattr_bucket_new(inode
);
3831 ret
= ocfs2_read_xattr_bucket(search
, p_blkno
);
3837 xh
= bucket_xh(search
);
3838 high_bucket
= le16_to_cpu(xh
->xh_num_buckets
) - 1;
3839 while (low_bucket
<= high_bucket
) {
3840 ocfs2_xattr_bucket_relse(search
);
3842 bucket
= (low_bucket
+ high_bucket
) / 2;
3843 blkno
= p_blkno
+ bucket
* blk_per_bucket
;
3844 ret
= ocfs2_read_xattr_bucket(search
, blkno
);
3850 xh
= bucket_xh(search
);
3851 xe
= &xh
->xh_entries
[0];
3852 if (name_hash
< le32_to_cpu(xe
->xe_name_hash
)) {
3853 high_bucket
= bucket
- 1;
3858 * Check whether the hash of the last entry in our
3859 * bucket is larger than the search one. for an empty
3860 * bucket, the last one is also the first one.
3863 xe
= &xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1];
3865 /* record lower_blkno which may be the insert place. */
3866 lower_blkno
= blkno
;
3868 if (name_hash
> le32_to_cpu(xe
->xe_name_hash
)) {
3869 low_bucket
= bucket
+ 1;
3873 /* the searched xattr should reside in this bucket if exists. */
3874 ret
= ocfs2_find_xe_in_bucket(inode
, search
,
3875 name_index
, name
, name_hash
,
3885 * Record the bucket we have found.
3886 * When the xattr's hash value is in the gap of 2 buckets, we will
3887 * always set it to the previous bucket.
3890 lower_blkno
= p_blkno
;
3892 /* This should be in cache - we just read it during the search */
3893 ret
= ocfs2_read_xattr_bucket(xs
->bucket
, lower_blkno
);
3899 xs
->header
= bucket_xh(xs
->bucket
);
3900 xs
->base
= bucket_block(xs
->bucket
, 0);
3901 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
3904 xs
->here
= &xs
->header
->xh_entries
[index
];
3905 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode
)->ip_blkno
,
3906 name
, name_index
, name_hash
,
3907 (unsigned long long)bucket_blkno(xs
->bucket
),
3913 ocfs2_xattr_bucket_free(search
);
3917 static int ocfs2_xattr_index_block_find(struct inode
*inode
,
3918 struct buffer_head
*root_bh
,
3921 struct ocfs2_xattr_search
*xs
)
3924 struct ocfs2_xattr_block
*xb
=
3925 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
3926 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
3927 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
3929 u32 first_hash
, num_clusters
= 0;
3930 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
3932 if (le16_to_cpu(el
->l_next_free_rec
) == 0)
3935 trace_ocfs2_xattr_index_block_find(OCFS2_I(inode
)->ip_blkno
,
3936 name
, name_index
, name_hash
,
3937 (unsigned long long)root_bh
->b_blocknr
,
3940 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &first_hash
,
3947 BUG_ON(p_blkno
== 0 || num_clusters
== 0 || first_hash
> name_hash
);
3949 trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode
)->ip_blkno
,
3950 name
, name_index
, first_hash
,
3951 (unsigned long long)p_blkno
,
3954 ret
= ocfs2_xattr_bucket_find(inode
, name_index
, name
, name_hash
,
3955 p_blkno
, first_hash
, num_clusters
, xs
);
3961 static int ocfs2_iterate_xattr_buckets(struct inode
*inode
,
3964 xattr_bucket_func
*func
,
3968 u32 bpc
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
));
3969 u32 num_buckets
= clusters
* bpc
;
3970 struct ocfs2_xattr_bucket
*bucket
;
3972 bucket
= ocfs2_xattr_bucket_new(inode
);
3974 mlog_errno(-ENOMEM
);
3978 trace_ocfs2_iterate_xattr_buckets(
3979 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
3980 (unsigned long long)blkno
, clusters
);
3982 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bucket
->bu_blocks
) {
3983 ret
= ocfs2_read_xattr_bucket(bucket
, blkno
);
3990 * The real bucket num in this series of blocks is stored
3991 * in the 1st bucket.
3994 num_buckets
= le16_to_cpu(bucket_xh(bucket
)->xh_num_buckets
);
3996 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno
,
3997 le32_to_cpu(bucket_xh(bucket
)->xh_entries
[0].xe_name_hash
));
3999 ret
= func(inode
, bucket
, para
);
4000 if (ret
&& ret
!= -ERANGE
)
4002 /* Fall through to bucket_relse() */
4005 ocfs2_xattr_bucket_relse(bucket
);
4010 ocfs2_xattr_bucket_free(bucket
);
4014 struct ocfs2_xattr_tree_list
{
4020 static int ocfs2_xattr_bucket_get_name_value(struct super_block
*sb
,
4021 struct ocfs2_xattr_header
*xh
,
4028 if (index
< 0 || index
>= le16_to_cpu(xh
->xh_count
))
4031 name_offset
= le16_to_cpu(xh
->xh_entries
[index
].xe_name_offset
);
4033 *block_off
= name_offset
>> sb
->s_blocksize_bits
;
4034 *new_offset
= name_offset
% sb
->s_blocksize
;
4039 static int ocfs2_list_xattr_bucket(struct inode
*inode
,
4040 struct ocfs2_xattr_bucket
*bucket
,
4044 struct ocfs2_xattr_tree_list
*xl
= (struct ocfs2_xattr_tree_list
*)para
;
4045 int i
, block_off
, new_offset
;
4048 for (i
= 0 ; i
< le16_to_cpu(bucket_xh(bucket
)->xh_count
); i
++) {
4049 struct ocfs2_xattr_entry
*entry
= &bucket_xh(bucket
)->xh_entries
[i
];
4050 type
= ocfs2_xattr_get_type(entry
);
4052 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
4060 name
= (const char *)bucket_block(bucket
, block_off
) +
4062 ret
= ocfs2_xattr_list_entry(inode
->i_sb
,
4067 entry
->xe_name_len
);
4075 static int ocfs2_iterate_xattr_index_block(struct inode
*inode
,
4076 struct buffer_head
*blk_bh
,
4077 xattr_tree_rec_func
*rec_func
,
4080 struct ocfs2_xattr_block
*xb
=
4081 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
4082 struct ocfs2_extent_list
*el
= &xb
->xb_attrs
.xb_root
.xt_list
;
4084 u32 name_hash
= UINT_MAX
, e_cpos
= 0, num_clusters
= 0;
4087 if (!el
->l_next_free_rec
|| !rec_func
)
4090 while (name_hash
> 0) {
4091 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
,
4092 &e_cpos
, &num_clusters
, el
);
4098 ret
= rec_func(inode
, blk_bh
, p_blkno
, e_cpos
,
4099 num_clusters
, para
);
4109 name_hash
= e_cpos
- 1;
4116 static int ocfs2_list_xattr_tree_rec(struct inode
*inode
,
4117 struct buffer_head
*root_bh
,
4118 u64 blkno
, u32 cpos
, u32 len
, void *para
)
4120 return ocfs2_iterate_xattr_buckets(inode
, blkno
, len
,
4121 ocfs2_list_xattr_bucket
, para
);
4124 static int ocfs2_xattr_tree_list_index_block(struct inode
*inode
,
4125 struct buffer_head
*blk_bh
,
4130 struct ocfs2_xattr_tree_list xl
= {
4132 .buffer_size
= buffer_size
,
4136 ret
= ocfs2_iterate_xattr_index_block(inode
, blk_bh
,
4137 ocfs2_list_xattr_tree_rec
, &xl
);
4148 static int cmp_xe(const void *a
, const void *b
)
4150 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
4151 u32 l_hash
= le32_to_cpu(l
->xe_name_hash
);
4152 u32 r_hash
= le32_to_cpu(r
->xe_name_hash
);
4154 if (l_hash
> r_hash
)
4156 if (l_hash
< r_hash
)
4161 static void swap_xe(void *a
, void *b
, int size
)
4163 struct ocfs2_xattr_entry
*l
= a
, *r
= b
, tmp
;
4166 memcpy(l
, r
, sizeof(struct ocfs2_xattr_entry
));
4167 memcpy(r
, &tmp
, sizeof(struct ocfs2_xattr_entry
));
4171 * When the ocfs2_xattr_block is filled up, new bucket will be created
4172 * and all the xattr entries will be moved to the new bucket.
4173 * The header goes at the start of the bucket, and the names+values are
4174 * filled from the end. This is why *target starts as the last buffer.
4175 * Note: we need to sort the entries since they are not saved in order
4176 * in the ocfs2_xattr_block.
4178 static void ocfs2_cp_xattr_block_to_bucket(struct inode
*inode
,
4179 struct buffer_head
*xb_bh
,
4180 struct ocfs2_xattr_bucket
*bucket
)
4182 int i
, blocksize
= inode
->i_sb
->s_blocksize
;
4183 int blks
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4184 u16 offset
, size
, off_change
;
4185 struct ocfs2_xattr_entry
*xe
;
4186 struct ocfs2_xattr_block
*xb
=
4187 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4188 struct ocfs2_xattr_header
*xb_xh
= &xb
->xb_attrs
.xb_header
;
4189 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
4190 u16 count
= le16_to_cpu(xb_xh
->xh_count
);
4191 char *src
= xb_bh
->b_data
;
4192 char *target
= bucket_block(bucket
, blks
- 1);
4194 trace_ocfs2_cp_xattr_block_to_bucket_begin(
4195 (unsigned long long)xb_bh
->b_blocknr
,
4196 (unsigned long long)bucket_blkno(bucket
));
4198 for (i
= 0; i
< blks
; i
++)
4199 memset(bucket_block(bucket
, i
), 0, blocksize
);
4202 * Since the xe_name_offset is based on ocfs2_xattr_header,
4203 * there is a offset change corresponding to the change of
4204 * ocfs2_xattr_header's position.
4206 off_change
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
4207 xe
= &xb_xh
->xh_entries
[count
- 1];
4208 offset
= le16_to_cpu(xe
->xe_name_offset
) + off_change
;
4209 size
= blocksize
- offset
;
4211 /* copy all the names and values. */
4212 memcpy(target
+ offset
, src
+ offset
, size
);
4214 /* Init new header now. */
4215 xh
->xh_count
= xb_xh
->xh_count
;
4216 xh
->xh_num_buckets
= cpu_to_le16(1);
4217 xh
->xh_name_value_len
= cpu_to_le16(size
);
4218 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
- size
);
4220 /* copy all the entries. */
4221 target
= bucket_block(bucket
, 0);
4222 offset
= offsetof(struct ocfs2_xattr_header
, xh_entries
);
4223 size
= count
* sizeof(struct ocfs2_xattr_entry
);
4224 memcpy(target
+ offset
, (char *)xb_xh
+ offset
, size
);
4226 /* Change the xe offset for all the xe because of the move. */
4227 off_change
= OCFS2_XATTR_BUCKET_SIZE
- blocksize
+
4228 offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
4229 for (i
= 0; i
< count
; i
++)
4230 le16_add_cpu(&xh
->xh_entries
[i
].xe_name_offset
, off_change
);
4232 trace_ocfs2_cp_xattr_block_to_bucket_end(offset
, size
, off_change
);
4234 sort(target
+ offset
, count
, sizeof(struct ocfs2_xattr_entry
),
4239 * After we move xattr from block to index btree, we have to
4240 * update ocfs2_xattr_search to the new xe and base.
4242 * When the entry is in xattr block, xattr_bh indicates the storage place.
4243 * While if the entry is in index b-tree, "bucket" indicates the
4244 * real place of the xattr.
4246 static void ocfs2_xattr_update_xattr_search(struct inode
*inode
,
4247 struct ocfs2_xattr_search
*xs
,
4248 struct buffer_head
*old_bh
)
4250 char *buf
= old_bh
->b_data
;
4251 struct ocfs2_xattr_block
*old_xb
= (struct ocfs2_xattr_block
*)buf
;
4252 struct ocfs2_xattr_header
*old_xh
= &old_xb
->xb_attrs
.xb_header
;
4255 xs
->header
= bucket_xh(xs
->bucket
);
4256 xs
->base
= bucket_block(xs
->bucket
, 0);
4257 xs
->end
= xs
->base
+ inode
->i_sb
->s_blocksize
;
4262 i
= xs
->here
- old_xh
->xh_entries
;
4263 xs
->here
= &xs
->header
->xh_entries
[i
];
4266 static int ocfs2_xattr_create_index_block(struct inode
*inode
,
4267 struct ocfs2_xattr_search
*xs
,
4268 struct ocfs2_xattr_set_ctxt
*ctxt
)
4273 handle_t
*handle
= ctxt
->handle
;
4274 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
4275 struct buffer_head
*xb_bh
= xs
->xattr_bh
;
4276 struct ocfs2_xattr_block
*xb
=
4277 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
4278 struct ocfs2_xattr_tree_root
*xr
;
4279 u16 xb_flags
= le16_to_cpu(xb
->xb_flags
);
4281 trace_ocfs2_xattr_create_index_block_begin(
4282 (unsigned long long)xb_bh
->b_blocknr
);
4284 BUG_ON(xb_flags
& OCFS2_XATTR_INDEXED
);
4285 BUG_ON(!xs
->bucket
);
4289 * We can use this lock for now, and maybe move to a dedicated mutex
4290 * if performance becomes a problem later.
4292 down_write(&oi
->ip_alloc_sem
);
4294 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(inode
), xb_bh
,
4295 OCFS2_JOURNAL_ACCESS_WRITE
);
4301 ret
= __ocfs2_claim_clusters(handle
, ctxt
->data_ac
,
4302 1, 1, &bit_off
, &len
);
4309 * The bucket may spread in many blocks, and
4310 * we will only touch the 1st block and the last block
4311 * in the whole bucket(one for entry and one for data).
4313 blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, bit_off
);
4315 trace_ocfs2_xattr_create_index_block((unsigned long long)blkno
);
4317 ret
= ocfs2_init_xattr_bucket(xs
->bucket
, blkno
, 1);
4323 ret
= ocfs2_xattr_bucket_journal_access(handle
, xs
->bucket
,
4324 OCFS2_JOURNAL_ACCESS_CREATE
);
4330 ocfs2_cp_xattr_block_to_bucket(inode
, xb_bh
, xs
->bucket
);
4331 ocfs2_xattr_bucket_journal_dirty(handle
, xs
->bucket
);
4333 ocfs2_xattr_update_xattr_search(inode
, xs
, xb_bh
);
4335 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4336 memset(&xb
->xb_attrs
, 0, inode
->i_sb
->s_blocksize
-
4337 offsetof(struct ocfs2_xattr_block
, xb_attrs
));
4339 xr
= &xb
->xb_attrs
.xb_root
;
4340 xr
->xt_clusters
= cpu_to_le32(1);
4341 xr
->xt_last_eb_blk
= 0;
4342 xr
->xt_list
.l_tree_depth
= 0;
4343 xr
->xt_list
.l_count
= cpu_to_le16(ocfs2_xattr_recs_per_xb(inode
->i_sb
));
4344 xr
->xt_list
.l_next_free_rec
= cpu_to_le16(1);
4346 xr
->xt_list
.l_recs
[0].e_cpos
= 0;
4347 xr
->xt_list
.l_recs
[0].e_blkno
= cpu_to_le64(blkno
);
4348 xr
->xt_list
.l_recs
[0].e_leaf_clusters
= cpu_to_le16(1);
4350 xb
->xb_flags
= cpu_to_le16(xb_flags
| OCFS2_XATTR_INDEXED
);
4352 ocfs2_journal_dirty(handle
, xb_bh
);
4355 up_write(&oi
->ip_alloc_sem
);
4360 static int cmp_xe_offset(const void *a
, const void *b
)
4362 const struct ocfs2_xattr_entry
*l
= a
, *r
= b
;
4363 u32 l_name_offset
= le16_to_cpu(l
->xe_name_offset
);
4364 u32 r_name_offset
= le16_to_cpu(r
->xe_name_offset
);
4366 if (l_name_offset
< r_name_offset
)
4368 if (l_name_offset
> r_name_offset
)
4374 * defrag a xattr bucket if we find that the bucket has some
4375 * holes beteen name/value pairs.
4376 * We will move all the name/value pairs to the end of the bucket
4377 * so that we can spare some space for insertion.
4379 static int ocfs2_defrag_xattr_bucket(struct inode
*inode
,
4381 struct ocfs2_xattr_bucket
*bucket
)
4384 size_t end
, offset
, len
;
4385 struct ocfs2_xattr_header
*xh
;
4386 char *entries
, *buf
, *bucket_buf
= NULL
;
4387 u64 blkno
= bucket_blkno(bucket
);
4389 size_t blocksize
= inode
->i_sb
->s_blocksize
;
4390 struct ocfs2_xattr_entry
*xe
;
4393 * In order to make the operation more efficient and generic,
4394 * we copy all the blocks into a contiguous memory and do the
4395 * defragment there, so if anything is error, we will not touch
4398 bucket_buf
= kmalloc(OCFS2_XATTR_BUCKET_SIZE
, GFP_NOFS
);
4405 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
4406 memcpy(buf
, bucket_block(bucket
, i
), blocksize
);
4408 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
4409 OCFS2_JOURNAL_ACCESS_WRITE
);
4415 xh
= (struct ocfs2_xattr_header
*)bucket_buf
;
4416 entries
= (char *)xh
->xh_entries
;
4417 xh_free_start
= le16_to_cpu(xh
->xh_free_start
);
4419 trace_ocfs2_defrag_xattr_bucket(
4420 (unsigned long long)blkno
, le16_to_cpu(xh
->xh_count
),
4421 xh_free_start
, le16_to_cpu(xh
->xh_name_value_len
));
4424 * sort all the entries by their offset.
4425 * the largest will be the first, so that we can
4426 * move them to the end one by one.
4428 sort(entries
, le16_to_cpu(xh
->xh_count
),
4429 sizeof(struct ocfs2_xattr_entry
),
4430 cmp_xe_offset
, swap_xe
);
4432 /* Move all name/values to the end of the bucket. */
4433 xe
= xh
->xh_entries
;
4434 end
= OCFS2_XATTR_BUCKET_SIZE
;
4435 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, xe
++) {
4436 offset
= le16_to_cpu(xe
->xe_name_offset
);
4437 len
= namevalue_size_xe(xe
);
4440 * We must make sure that the name/value pair
4441 * exist in the same block. So adjust end to
4442 * the previous block end if needed.
4444 if (((end
- len
) / blocksize
!=
4445 (end
- 1) / blocksize
))
4446 end
= end
- end
% blocksize
;
4448 if (end
> offset
+ len
) {
4449 memmove(bucket_buf
+ end
- len
,
4450 bucket_buf
+ offset
, len
);
4451 xe
->xe_name_offset
= cpu_to_le16(end
- len
);
4454 mlog_bug_on_msg(end
< offset
+ len
, "Defrag check failed for "
4455 "bucket %llu\n", (unsigned long long)blkno
);
4460 mlog_bug_on_msg(xh_free_start
> end
, "Defrag check failed for "
4461 "bucket %llu\n", (unsigned long long)blkno
);
4463 if (xh_free_start
== end
)
4466 memset(bucket_buf
+ xh_free_start
, 0, end
- xh_free_start
);
4467 xh
->xh_free_start
= cpu_to_le16(end
);
4469 /* sort the entries by their name_hash. */
4470 sort(entries
, le16_to_cpu(xh
->xh_count
),
4471 sizeof(struct ocfs2_xattr_entry
),
4475 for (i
= 0; i
< bucket
->bu_blocks
; i
++, buf
+= blocksize
)
4476 memcpy(bucket_block(bucket
, i
), buf
, blocksize
);
4477 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
4485 * prev_blkno points to the start of an existing extent. new_blkno
4486 * points to a newly allocated extent. Because we know each of our
4487 * clusters contains more than bucket, we can easily split one cluster
4488 * at a bucket boundary. So we take the last cluster of the existing
4489 * extent and split it down the middle. We move the last half of the
4490 * buckets in the last cluster of the existing extent over to the new
4493 * first_bh is the buffer at prev_blkno so we can update the existing
4494 * extent's bucket count. header_bh is the bucket were we were hoping
4495 * to insert our xattr. If the bucket move places the target in the new
4496 * extent, we'll update first_bh and header_bh after modifying the old
4499 * first_hash will be set as the 1st xe's name_hash in the new extent.
4501 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode
*inode
,
4503 struct ocfs2_xattr_bucket
*first
,
4504 struct ocfs2_xattr_bucket
*target
,
4510 struct super_block
*sb
= inode
->i_sb
;
4511 int blks_per_bucket
= ocfs2_blocks_per_xattr_bucket(sb
);
4512 int num_buckets
= ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb
));
4513 int to_move
= num_buckets
/ 2;
4515 u64 last_cluster_blkno
= bucket_blkno(first
) +
4516 ((num_clusters
- 1) * ocfs2_clusters_to_blocks(sb
, 1));
4518 BUG_ON(le16_to_cpu(bucket_xh(first
)->xh_num_buckets
) < num_buckets
);
4519 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
== OCFS2_SB(sb
)->s_clustersize
);
4521 trace_ocfs2_mv_xattr_bucket_cross_cluster(
4522 (unsigned long long)last_cluster_blkno
,
4523 (unsigned long long)new_blkno
);
4525 ret
= ocfs2_mv_xattr_buckets(inode
, handle
, bucket_blkno(first
),
4526 last_cluster_blkno
, new_blkno
,
4527 to_move
, first_hash
);
4533 /* This is the first bucket that got moved */
4534 src_blkno
= last_cluster_blkno
+ (to_move
* blks_per_bucket
);
4537 * If the target bucket was part of the moved buckets, we need to
4538 * update first and target.
4540 if (bucket_blkno(target
) >= src_blkno
) {
4541 /* Find the block for the new target bucket */
4542 src_blkno
= new_blkno
+
4543 (bucket_blkno(target
) - src_blkno
);
4545 ocfs2_xattr_bucket_relse(first
);
4546 ocfs2_xattr_bucket_relse(target
);
4549 * These shouldn't fail - the buffers are in the
4550 * journal from ocfs2_cp_xattr_bucket().
4552 ret
= ocfs2_read_xattr_bucket(first
, new_blkno
);
4557 ret
= ocfs2_read_xattr_bucket(target
, src_blkno
);
4568 * Find the suitable pos when we divide a bucket into 2.
4569 * We have to make sure the xattrs with the same hash value exist
4570 * in the same bucket.
4572 * If this ocfs2_xattr_header covers more than one hash value, find a
4573 * place where the hash value changes. Try to find the most even split.
4574 * The most common case is that all entries have different hash values,
4575 * and the first check we make will find a place to split.
4577 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header
*xh
)
4579 struct ocfs2_xattr_entry
*entries
= xh
->xh_entries
;
4580 int count
= le16_to_cpu(xh
->xh_count
);
4581 int delta
, middle
= count
/ 2;
4584 * We start at the middle. Each step gets farther away in both
4585 * directions. We therefore hit the change in hash value
4586 * nearest to the middle. Note that this loop does not execute for
4589 for (delta
= 0; delta
< middle
; delta
++) {
4590 /* Let's check delta earlier than middle */
4591 if (cmp_xe(&entries
[middle
- delta
- 1],
4592 &entries
[middle
- delta
]))
4593 return middle
- delta
;
4595 /* For even counts, don't walk off the end */
4596 if ((middle
+ delta
+ 1) == count
)
4599 /* Now try delta past middle */
4600 if (cmp_xe(&entries
[middle
+ delta
],
4601 &entries
[middle
+ delta
+ 1]))
4602 return middle
+ delta
+ 1;
4605 /* Every entry had the same hash */
4610 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4611 * first_hash will record the 1st hash of the new bucket.
4613 * Normally half of the xattrs will be moved. But we have to make
4614 * sure that the xattrs with the same hash value are stored in the
4615 * same bucket. If all the xattrs in this bucket have the same hash
4616 * value, the new bucket will be initialized as an empty one and the
4617 * first_hash will be initialized as (hash_value+1).
4619 static int ocfs2_divide_xattr_bucket(struct inode
*inode
,
4624 int new_bucket_head
)
4627 int count
, start
, len
, name_value_len
= 0, name_offset
= 0;
4628 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
4629 struct ocfs2_xattr_header
*xh
;
4630 struct ocfs2_xattr_entry
*xe
;
4631 int blocksize
= inode
->i_sb
->s_blocksize
;
4633 trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk
,
4634 (unsigned long long)new_blk
);
4636 s_bucket
= ocfs2_xattr_bucket_new(inode
);
4637 t_bucket
= ocfs2_xattr_bucket_new(inode
);
4638 if (!s_bucket
|| !t_bucket
) {
4644 ret
= ocfs2_read_xattr_bucket(s_bucket
, blk
);
4650 ret
= ocfs2_xattr_bucket_journal_access(handle
, s_bucket
,
4651 OCFS2_JOURNAL_ACCESS_WRITE
);
4658 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4659 * there's no need to read it.
4661 ret
= ocfs2_init_xattr_bucket(t_bucket
, new_blk
, new_bucket_head
);
4668 * Hey, if we're overwriting t_bucket, what difference does
4669 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4670 * same part of ocfs2_cp_xattr_bucket().
4672 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
4674 OCFS2_JOURNAL_ACCESS_CREATE
:
4675 OCFS2_JOURNAL_ACCESS_WRITE
);
4681 xh
= bucket_xh(s_bucket
);
4682 count
= le16_to_cpu(xh
->xh_count
);
4683 start
= ocfs2_xattr_find_divide_pos(xh
);
4685 if (start
== count
) {
4686 xe
= &xh
->xh_entries
[start
-1];
4689 * initialized a new empty bucket here.
4690 * The hash value is set as one larger than
4691 * that of the last entry in the previous bucket.
4693 for (i
= 0; i
< t_bucket
->bu_blocks
; i
++)
4694 memset(bucket_block(t_bucket
, i
), 0, blocksize
);
4696 xh
= bucket_xh(t_bucket
);
4697 xh
->xh_free_start
= cpu_to_le16(blocksize
);
4698 xh
->xh_entries
[0].xe_name_hash
= xe
->xe_name_hash
;
4699 le32_add_cpu(&xh
->xh_entries
[0].xe_name_hash
, 1);
4701 goto set_num_buckets
;
4704 /* copy the whole bucket to the new first. */
4705 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
4707 /* update the new bucket. */
4708 xh
= bucket_xh(t_bucket
);
4711 * Calculate the total name/value len and xh_free_start for
4712 * the old bucket first.
4714 name_offset
= OCFS2_XATTR_BUCKET_SIZE
;
4716 for (i
= 0; i
< start
; i
++) {
4717 xe
= &xh
->xh_entries
[i
];
4718 name_value_len
+= namevalue_size_xe(xe
);
4719 if (le16_to_cpu(xe
->xe_name_offset
) < name_offset
)
4720 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
4724 * Now begin the modification to the new bucket.
4726 * In the new bucket, We just move the xattr entry to the beginning
4727 * and don't touch the name/value. So there will be some holes in the
4728 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4731 xe
= &xh
->xh_entries
[start
];
4732 len
= sizeof(struct ocfs2_xattr_entry
) * (count
- start
);
4733 trace_ocfs2_divide_xattr_bucket_move(len
,
4734 (int)((char *)xe
- (char *)xh
),
4735 (int)((char *)xh
->xh_entries
- (char *)xh
));
4736 memmove((char *)xh
->xh_entries
, (char *)xe
, len
);
4737 xe
= &xh
->xh_entries
[count
- start
];
4738 len
= sizeof(struct ocfs2_xattr_entry
) * start
;
4739 memset((char *)xe
, 0, len
);
4741 le16_add_cpu(&xh
->xh_count
, -start
);
4742 le16_add_cpu(&xh
->xh_name_value_len
, -name_value_len
);
4744 /* Calculate xh_free_start for the new bucket. */
4745 xh
->xh_free_start
= cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE
);
4746 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
4747 xe
= &xh
->xh_entries
[i
];
4748 if (le16_to_cpu(xe
->xe_name_offset
) <
4749 le16_to_cpu(xh
->xh_free_start
))
4750 xh
->xh_free_start
= xe
->xe_name_offset
;
4754 /* set xh->xh_num_buckets for the new xh. */
4755 if (new_bucket_head
)
4756 xh
->xh_num_buckets
= cpu_to_le16(1);
4758 xh
->xh_num_buckets
= 0;
4760 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
4762 /* store the first_hash of the new bucket. */
4764 *first_hash
= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
);
4767 * Now only update the 1st block of the old bucket. If we
4768 * just added a new empty bucket, there is no need to modify
4774 xh
= bucket_xh(s_bucket
);
4775 memset(&xh
->xh_entries
[start
], 0,
4776 sizeof(struct ocfs2_xattr_entry
) * (count
- start
));
4777 xh
->xh_count
= cpu_to_le16(start
);
4778 xh
->xh_free_start
= cpu_to_le16(name_offset
);
4779 xh
->xh_name_value_len
= cpu_to_le16(name_value_len
);
4781 ocfs2_xattr_bucket_journal_dirty(handle
, s_bucket
);
4784 ocfs2_xattr_bucket_free(s_bucket
);
4785 ocfs2_xattr_bucket_free(t_bucket
);
4791 * Copy xattr from one bucket to another bucket.
4793 * The caller must make sure that the journal transaction
4794 * has enough space for journaling.
4796 static int ocfs2_cp_xattr_bucket(struct inode
*inode
,
4803 struct ocfs2_xattr_bucket
*s_bucket
= NULL
, *t_bucket
= NULL
;
4805 BUG_ON(s_blkno
== t_blkno
);
4807 trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno
,
4808 (unsigned long long)t_blkno
,
4811 s_bucket
= ocfs2_xattr_bucket_new(inode
);
4812 t_bucket
= ocfs2_xattr_bucket_new(inode
);
4813 if (!s_bucket
|| !t_bucket
) {
4819 ret
= ocfs2_read_xattr_bucket(s_bucket
, s_blkno
);
4824 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4825 * there's no need to read it.
4827 ret
= ocfs2_init_xattr_bucket(t_bucket
, t_blkno
, t_is_new
);
4832 * Hey, if we're overwriting t_bucket, what difference does
4833 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4834 * cluster to fill, we came here from
4835 * ocfs2_mv_xattr_buckets(), and it is really new -
4836 * ACCESS_CREATE is required. But we also might have moved data
4837 * out of t_bucket before extending back into it.
4838 * ocfs2_add_new_xattr_bucket() can do this - its call to
4839 * ocfs2_add_new_xattr_cluster() may have created a new extent
4840 * and copied out the end of the old extent. Then it re-extends
4841 * the old extent back to create space for new xattrs. That's
4842 * how we get here, and the bucket isn't really new.
4844 ret
= ocfs2_xattr_bucket_journal_access(handle
, t_bucket
,
4846 OCFS2_JOURNAL_ACCESS_CREATE
:
4847 OCFS2_JOURNAL_ACCESS_WRITE
);
4851 ocfs2_xattr_bucket_copy_data(t_bucket
, s_bucket
);
4852 ocfs2_xattr_bucket_journal_dirty(handle
, t_bucket
);
4855 ocfs2_xattr_bucket_free(t_bucket
);
4856 ocfs2_xattr_bucket_free(s_bucket
);
4862 * src_blk points to the start of an existing extent. last_blk points to
4863 * last cluster in that extent. to_blk points to a newly allocated
4864 * extent. We copy the buckets from the cluster at last_blk to the new
4865 * extent. If start_bucket is non-zero, we skip that many buckets before
4866 * we start copying. The new extent's xh_num_buckets gets set to the
4867 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4868 * by the same amount.
4870 static int ocfs2_mv_xattr_buckets(struct inode
*inode
, handle_t
*handle
,
4871 u64 src_blk
, u64 last_blk
, u64 to_blk
,
4872 unsigned int start_bucket
,
4875 int i
, ret
, credits
;
4876 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
4877 int blks_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4878 int num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
);
4879 struct ocfs2_xattr_bucket
*old_first
, *new_first
;
4881 trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk
,
4882 (unsigned long long)to_blk
);
4884 BUG_ON(start_bucket
>= num_buckets
);
4886 num_buckets
-= start_bucket
;
4887 last_blk
+= (start_bucket
* blks_per_bucket
);
4890 /* The first bucket of the original extent */
4891 old_first
= ocfs2_xattr_bucket_new(inode
);
4892 /* The first bucket of the new extent */
4893 new_first
= ocfs2_xattr_bucket_new(inode
);
4894 if (!old_first
|| !new_first
) {
4900 ret
= ocfs2_read_xattr_bucket(old_first
, src_blk
);
4907 * We need to update the first bucket of the old extent and all
4908 * the buckets going to the new extent.
4910 credits
= ((num_buckets
+ 1) * blks_per_bucket
);
4911 ret
= ocfs2_extend_trans(handle
, credits
);
4917 ret
= ocfs2_xattr_bucket_journal_access(handle
, old_first
,
4918 OCFS2_JOURNAL_ACCESS_WRITE
);
4924 for (i
= 0; i
< num_buckets
; i
++) {
4925 ret
= ocfs2_cp_xattr_bucket(inode
, handle
,
4926 last_blk
+ (i
* blks_per_bucket
),
4927 to_blk
+ (i
* blks_per_bucket
),
4936 * Get the new bucket ready before we dirty anything
4937 * (This actually shouldn't fail, because we already dirtied
4938 * it once in ocfs2_cp_xattr_bucket()).
4940 ret
= ocfs2_read_xattr_bucket(new_first
, to_blk
);
4945 ret
= ocfs2_xattr_bucket_journal_access(handle
, new_first
,
4946 OCFS2_JOURNAL_ACCESS_WRITE
);
4952 /* Now update the headers */
4953 le16_add_cpu(&bucket_xh(old_first
)->xh_num_buckets
, -num_buckets
);
4954 ocfs2_xattr_bucket_journal_dirty(handle
, old_first
);
4956 bucket_xh(new_first
)->xh_num_buckets
= cpu_to_le16(num_buckets
);
4957 ocfs2_xattr_bucket_journal_dirty(handle
, new_first
);
4960 *first_hash
= le32_to_cpu(bucket_xh(new_first
)->xh_entries
[0].xe_name_hash
);
4963 ocfs2_xattr_bucket_free(new_first
);
4964 ocfs2_xattr_bucket_free(old_first
);
4969 * Move some xattrs in this cluster to the new cluster.
4970 * This function should only be called when bucket size == cluster size.
4971 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4973 static int ocfs2_divide_xattr_cluster(struct inode
*inode
,
4979 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
4980 int ret
, credits
= 2 * blk_per_bucket
;
4982 BUG_ON(OCFS2_XATTR_BUCKET_SIZE
< OCFS2_SB(inode
->i_sb
)->s_clustersize
);
4984 ret
= ocfs2_extend_trans(handle
, credits
);
4990 /* Move half of the xattr in start_blk to the next bucket. */
4991 return ocfs2_divide_xattr_bucket(inode
, handle
, prev_blk
,
4992 new_blk
, first_hash
, 1);
4996 * Move some xattrs from the old cluster to the new one since they are not
4997 * contiguous in ocfs2 xattr tree.
4999 * new_blk starts a new separate cluster, and we will move some xattrs from
5000 * prev_blk to it. v_start will be set as the first name hash value in this
5001 * new cluster so that it can be used as e_cpos during tree insertion and
5002 * don't collide with our original b-tree operations. first_bh and header_bh
5003 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5004 * to extend the insert bucket.
5006 * The problem is how much xattr should we move to the new one and when should
5007 * we update first_bh and header_bh?
5008 * 1. If cluster size > bucket size, that means the previous cluster has more
5009 * than 1 bucket, so just move half nums of bucket into the new cluster and
5010 * update the first_bh and header_bh if the insert bucket has been moved
5011 * to the new cluster.
5012 * 2. If cluster_size == bucket_size:
5013 * a) If the previous extent rec has more than one cluster and the insert
5014 * place isn't in the last cluster, copy the entire last cluster to the
5015 * new one. This time, we don't need to upate the first_bh and header_bh
5016 * since they will not be moved into the new cluster.
5017 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
5018 * the new one. And we set the extend flag to zero if the insert place is
5019 * moved into the new allocated cluster since no extend is needed.
5021 static int ocfs2_adjust_xattr_cross_cluster(struct inode
*inode
,
5023 struct ocfs2_xattr_bucket
*first
,
5024 struct ocfs2_xattr_bucket
*target
,
5032 trace_ocfs2_adjust_xattr_cross_cluster(
5033 (unsigned long long)bucket_blkno(first
),
5034 (unsigned long long)new_blk
, prev_clusters
);
5036 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
)) > 1) {
5037 ret
= ocfs2_mv_xattr_bucket_cross_cluster(inode
,
5046 /* The start of the last cluster in the first extent */
5047 u64 last_blk
= bucket_blkno(first
) +
5048 ((prev_clusters
- 1) *
5049 ocfs2_clusters_to_blocks(inode
->i_sb
, 1));
5051 if (prev_clusters
> 1 && bucket_blkno(target
) != last_blk
) {
5052 ret
= ocfs2_mv_xattr_buckets(inode
, handle
,
5053 bucket_blkno(first
),
5054 last_blk
, new_blk
, 0,
5059 ret
= ocfs2_divide_xattr_cluster(inode
, handle
,
5065 if ((bucket_blkno(target
) == last_blk
) && extend
)
5074 * Add a new cluster for xattr storage.
5076 * If the new cluster is contiguous with the previous one, it will be
5077 * appended to the same extent record, and num_clusters will be updated.
5078 * If not, we will insert a new extent for it and move some xattrs in
5079 * the last cluster into the new allocated one.
5080 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5081 * lose the benefits of hashing because we'll have to search large leaves.
5082 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5085 * first_bh is the first block of the previous extent rec and header_bh
5086 * indicates the bucket we will insert the new xattrs. They will be updated
5087 * when the header_bh is moved into the new cluster.
5089 static int ocfs2_add_new_xattr_cluster(struct inode
*inode
,
5090 struct buffer_head
*root_bh
,
5091 struct ocfs2_xattr_bucket
*first
,
5092 struct ocfs2_xattr_bucket
*target
,
5096 struct ocfs2_xattr_set_ctxt
*ctxt
)
5099 u16 bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
5100 u32 prev_clusters
= *num_clusters
;
5101 u32 clusters_to_add
= 1, bit_off
, num_bits
, v_start
= 0;
5103 handle_t
*handle
= ctxt
->handle
;
5104 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5105 struct ocfs2_extent_tree et
;
5107 trace_ocfs2_add_new_xattr_cluster_begin(
5108 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
5109 (unsigned long long)bucket_blkno(first
),
5110 prev_cpos
, prev_clusters
);
5112 ocfs2_init_xattr_tree_extent_tree(&et
, INODE_CACHE(inode
), root_bh
);
5114 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(inode
), root_bh
,
5115 OCFS2_JOURNAL_ACCESS_WRITE
);
5121 ret
= __ocfs2_claim_clusters(handle
, ctxt
->data_ac
, 1,
5122 clusters_to_add
, &bit_off
, &num_bits
);
5129 BUG_ON(num_bits
> clusters_to_add
);
5131 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
5132 trace_ocfs2_add_new_xattr_cluster((unsigned long long)block
, num_bits
);
5134 if (bucket_blkno(first
) + (prev_clusters
* bpc
) == block
&&
5135 (prev_clusters
+ num_bits
) << osb
->s_clustersize_bits
<=
5136 OCFS2_MAX_XATTR_TREE_LEAF_SIZE
) {
5138 * If this cluster is contiguous with the old one and
5139 * adding this new cluster, we don't surpass the limit of
5140 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5141 * initialized and used like other buckets in the previous
5143 * So add it as a contiguous one. The caller will handle
5146 v_start
= prev_cpos
+ prev_clusters
;
5147 *num_clusters
= prev_clusters
+ num_bits
;
5149 ret
= ocfs2_adjust_xattr_cross_cluster(inode
,
5163 trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block
,
5165 ret
= ocfs2_insert_extent(handle
, &et
, v_start
, block
,
5166 num_bits
, 0, ctxt
->meta_ac
);
5172 ocfs2_journal_dirty(handle
, root_bh
);
5179 * We are given an extent. 'first' is the bucket at the very front of
5180 * the extent. The extent has space for an additional bucket past
5181 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5182 * of the target bucket. We wish to shift every bucket past the target
5183 * down one, filling in that additional space. When we get back to the
5184 * target, we split the target between itself and the now-empty bucket
5185 * at target+1 (aka, target_blkno + blks_per_bucket).
5187 static int ocfs2_extend_xattr_bucket(struct inode
*inode
,
5189 struct ocfs2_xattr_bucket
*first
,
5194 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5195 u16 blk_per_bucket
= ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
5197 u16 new_bucket
= le16_to_cpu(bucket_xh(first
)->xh_num_buckets
);
5199 trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk
,
5200 (unsigned long long)bucket_blkno(first
),
5201 num_clusters
, new_bucket
);
5203 /* The extent must have room for an additional bucket */
5204 BUG_ON(new_bucket
>=
5205 (num_clusters
* ocfs2_xattr_buckets_per_cluster(osb
)));
5207 /* end_blk points to the last existing bucket */
5208 end_blk
= bucket_blkno(first
) + ((new_bucket
- 1) * blk_per_bucket
);
5211 * end_blk is the start of the last existing bucket.
5212 * Thus, (end_blk - target_blk) covers the target bucket and
5213 * every bucket after it up to, but not including, the last
5214 * existing bucket. Then we add the last existing bucket, the
5215 * new bucket, and the first bucket (3 * blk_per_bucket).
5217 credits
= (end_blk
- target_blk
) + (3 * blk_per_bucket
);
5218 ret
= ocfs2_extend_trans(handle
, credits
);
5224 ret
= ocfs2_xattr_bucket_journal_access(handle
, first
,
5225 OCFS2_JOURNAL_ACCESS_WRITE
);
5231 while (end_blk
!= target_blk
) {
5232 ret
= ocfs2_cp_xattr_bucket(inode
, handle
, end_blk
,
5233 end_blk
+ blk_per_bucket
, 0);
5236 end_blk
-= blk_per_bucket
;
5239 /* Move half of the xattr in target_blkno to the next bucket. */
5240 ret
= ocfs2_divide_xattr_bucket(inode
, handle
, target_blk
,
5241 target_blk
+ blk_per_bucket
, NULL
, 0);
5243 le16_add_cpu(&bucket_xh(first
)->xh_num_buckets
, 1);
5244 ocfs2_xattr_bucket_journal_dirty(handle
, first
);
5251 * Add new xattr bucket in an extent record and adjust the buckets
5252 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5253 * bucket we want to insert into.
5255 * In the easy case, we will move all the buckets after target down by
5256 * one. Half of target's xattrs will be moved to the next bucket.
5258 * If current cluster is full, we'll allocate a new one. This may not
5259 * be contiguous. The underlying calls will make sure that there is
5260 * space for the insert, shifting buckets around if necessary.
5261 * 'target' may be moved by those calls.
5263 static int ocfs2_add_new_xattr_bucket(struct inode
*inode
,
5264 struct buffer_head
*xb_bh
,
5265 struct ocfs2_xattr_bucket
*target
,
5266 struct ocfs2_xattr_set_ctxt
*ctxt
)
5268 struct ocfs2_xattr_block
*xb
=
5269 (struct ocfs2_xattr_block
*)xb_bh
->b_data
;
5270 struct ocfs2_xattr_tree_root
*xb_root
= &xb
->xb_attrs
.xb_root
;
5271 struct ocfs2_extent_list
*el
= &xb_root
->xt_list
;
5273 le32_to_cpu(bucket_xh(target
)->xh_entries
[0].xe_name_hash
);
5274 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5275 int ret
, num_buckets
, extend
= 1;
5277 u32 e_cpos
, num_clusters
;
5278 /* The bucket at the front of the extent */
5279 struct ocfs2_xattr_bucket
*first
;
5281 trace_ocfs2_add_new_xattr_bucket(
5282 (unsigned long long)bucket_blkno(target
));
5284 /* The first bucket of the original extent */
5285 first
= ocfs2_xattr_bucket_new(inode
);
5292 ret
= ocfs2_xattr_get_rec(inode
, name_hash
, &p_blkno
, &e_cpos
,
5299 ret
= ocfs2_read_xattr_bucket(first
, p_blkno
);
5305 num_buckets
= ocfs2_xattr_buckets_per_cluster(osb
) * num_clusters
;
5306 if (num_buckets
== le16_to_cpu(bucket_xh(first
)->xh_num_buckets
)) {
5308 * This can move first+target if the target bucket moves
5309 * to the new extent.
5311 ret
= ocfs2_add_new_xattr_cluster(inode
,
5326 ret
= ocfs2_extend_xattr_bucket(inode
,
5329 bucket_blkno(target
),
5336 ocfs2_xattr_bucket_free(first
);
5342 * Truncate the specified xe_off entry in xattr bucket.
5343 * bucket is indicated by header_bh and len is the new length.
5344 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5346 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5348 static int ocfs2_xattr_bucket_value_truncate(struct inode
*inode
,
5349 struct ocfs2_xattr_bucket
*bucket
,
5352 struct ocfs2_xattr_set_ctxt
*ctxt
)
5356 struct ocfs2_xattr_entry
*xe
;
5357 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5358 size_t blocksize
= inode
->i_sb
->s_blocksize
;
5359 struct ocfs2_xattr_value_buf vb
= {
5360 .vb_access
= ocfs2_journal_access
,
5363 xe
= &xh
->xh_entries
[xe_off
];
5365 BUG_ON(!xe
|| ocfs2_xattr_is_local(xe
));
5367 offset
= le16_to_cpu(xe
->xe_name_offset
) +
5368 OCFS2_XATTR_SIZE(xe
->xe_name_len
);
5370 value_blk
= offset
/ blocksize
;
5372 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5373 BUG_ON(value_blk
!= (offset
+ OCFS2_XATTR_ROOT_SIZE
- 1) / blocksize
);
5375 vb
.vb_bh
= bucket
->bu_bhs
[value_blk
];
5378 vb
.vb_xv
= (struct ocfs2_xattr_value_root
*)
5379 (vb
.vb_bh
->b_data
+ offset
% blocksize
);
5382 * From here on out we have to dirty the bucket. The generic
5383 * value calls only modify one of the bucket's bhs, but we need
5384 * to send the bucket at once. So if they error, they *could* have
5385 * modified something. We have to assume they did, and dirty
5386 * the whole bucket. This leaves us in a consistent state.
5388 trace_ocfs2_xattr_bucket_value_truncate(
5389 (unsigned long long)bucket_blkno(bucket
), xe_off
, len
);
5390 ret
= ocfs2_xattr_value_truncate(inode
, &vb
, len
, ctxt
);
5396 ret
= ocfs2_xattr_bucket_journal_access(ctxt
->handle
, bucket
,
5397 OCFS2_JOURNAL_ACCESS_WRITE
);
5403 xe
->xe_value_size
= cpu_to_le64(len
);
5405 ocfs2_xattr_bucket_journal_dirty(ctxt
->handle
, bucket
);
5411 static int ocfs2_rm_xattr_cluster(struct inode
*inode
,
5412 struct buffer_head
*root_bh
,
5419 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5420 struct inode
*tl_inode
= osb
->osb_tl_inode
;
5422 struct ocfs2_xattr_block
*xb
=
5423 (struct ocfs2_xattr_block
*)root_bh
->b_data
;
5424 struct ocfs2_alloc_context
*meta_ac
= NULL
;
5425 struct ocfs2_cached_dealloc_ctxt dealloc
;
5426 struct ocfs2_extent_tree et
;
5428 ret
= ocfs2_iterate_xattr_buckets(inode
, blkno
, len
,
5429 ocfs2_delete_xattr_in_bucket
, para
);
5435 ocfs2_init_xattr_tree_extent_tree(&et
, INODE_CACHE(inode
), root_bh
);
5437 ocfs2_init_dealloc_ctxt(&dealloc
);
5439 trace_ocfs2_rm_xattr_cluster(
5440 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
5441 (unsigned long long)blkno
, cpos
, len
);
5443 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode
), blkno
,
5446 ret
= ocfs2_lock_allocators(inode
, &et
, 0, 1, NULL
, &meta_ac
);
5452 inode_lock(tl_inode
);
5454 if (ocfs2_truncate_log_needs_flush(osb
)) {
5455 ret
= __ocfs2_flush_truncate_log(osb
);
5462 handle
= ocfs2_start_trans(osb
, ocfs2_remove_extent_credits(osb
->sb
));
5463 if (IS_ERR(handle
)) {
5469 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(inode
), root_bh
,
5470 OCFS2_JOURNAL_ACCESS_WRITE
);
5476 ret
= ocfs2_remove_extent(handle
, &et
, cpos
, len
, meta_ac
,
5483 le32_add_cpu(&xb
->xb_attrs
.xb_root
.xt_clusters
, -len
);
5484 ocfs2_journal_dirty(handle
, root_bh
);
5486 ret
= ocfs2_truncate_log_append(osb
, handle
, blkno
, len
);
5489 ocfs2_update_inode_fsync_trans(handle
, inode
, 0);
5492 ocfs2_commit_trans(osb
, handle
);
5494 ocfs2_schedule_truncate_log_flush(osb
, 1);
5496 inode_unlock(tl_inode
);
5499 ocfs2_free_alloc_context(meta_ac
);
5501 ocfs2_run_deallocs(osb
, &dealloc
);
5507 * check whether the xattr bucket is filled up with the same hash value.
5508 * If we want to insert the xattr with the same hash, return -ENOSPC.
5509 * If we want to insert a xattr with different hash value, go ahead
5510 * and ocfs2_divide_xattr_bucket will handle this.
5512 static int ocfs2_check_xattr_bucket_collision(struct inode
*inode
,
5513 struct ocfs2_xattr_bucket
*bucket
,
5516 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5517 u32 name_hash
= ocfs2_xattr_name_hash(inode
, name
, strlen(name
));
5519 if (name_hash
!= le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
))
5522 if (xh
->xh_entries
[le16_to_cpu(xh
->xh_count
) - 1].xe_name_hash
==
5523 xh
->xh_entries
[0].xe_name_hash
) {
5524 mlog(ML_ERROR
, "Too much hash collision in xattr bucket %llu, "
5526 (unsigned long long)bucket_blkno(bucket
),
5527 le32_to_cpu(xh
->xh_entries
[0].xe_name_hash
));
5535 * Try to set the entry in the current bucket. If we fail, the caller
5536 * will handle getting us another bucket.
5538 static int ocfs2_xattr_set_entry_bucket(struct inode
*inode
,
5539 struct ocfs2_xattr_info
*xi
,
5540 struct ocfs2_xattr_search
*xs
,
5541 struct ocfs2_xattr_set_ctxt
*ctxt
)
5544 struct ocfs2_xa_loc loc
;
5546 trace_ocfs2_xattr_set_entry_bucket(xi
->xi_name
);
5548 ocfs2_init_xattr_bucket_xa_loc(&loc
, xs
->bucket
,
5549 xs
->not_found
? NULL
: xs
->here
);
5550 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
5552 xs
->here
= loc
.xl_entry
;
5555 if (ret
!= -ENOSPC
) {
5560 /* Ok, we need space. Let's try defragmenting the bucket. */
5561 ret
= ocfs2_defrag_xattr_bucket(inode
, ctxt
->handle
,
5568 ret
= ocfs2_xa_set(&loc
, xi
, ctxt
);
5570 xs
->here
= loc
.xl_entry
;
5581 static int ocfs2_xattr_set_entry_index_block(struct inode
*inode
,
5582 struct ocfs2_xattr_info
*xi
,
5583 struct ocfs2_xattr_search
*xs
,
5584 struct ocfs2_xattr_set_ctxt
*ctxt
)
5588 trace_ocfs2_xattr_set_entry_index_block(xi
->xi_name
);
5590 ret
= ocfs2_xattr_set_entry_bucket(inode
, xi
, xs
, ctxt
);
5593 if (ret
!= -ENOSPC
) {
5598 /* Ack, need more space. Let's try to get another bucket! */
5601 * We do not allow for overlapping ranges between buckets. And
5602 * the maximum number of collisions we will allow for then is
5603 * one bucket's worth, so check it here whether we need to
5604 * add a new bucket for the insert.
5606 ret
= ocfs2_check_xattr_bucket_collision(inode
,
5614 ret
= ocfs2_add_new_xattr_bucket(inode
,
5624 * ocfs2_add_new_xattr_bucket() will have updated
5625 * xs->bucket if it moved, but it will not have updated
5626 * any of the other search fields. Thus, we drop it and
5627 * re-search. Everything should be cached, so it'll be
5630 ocfs2_xattr_bucket_relse(xs
->bucket
);
5631 ret
= ocfs2_xattr_index_block_find(inode
, xs
->xattr_bh
,
5634 if (ret
&& ret
!= -ENODATA
)
5636 xs
->not_found
= ret
;
5638 /* Ok, we have a new bucket, let's try again */
5639 ret
= ocfs2_xattr_set_entry_bucket(inode
, xi
, xs
, ctxt
);
5640 if (ret
&& (ret
!= -ENOSPC
))
5647 static int ocfs2_delete_xattr_in_bucket(struct inode
*inode
,
5648 struct ocfs2_xattr_bucket
*bucket
,
5651 int ret
= 0, ref_credits
;
5652 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
5654 struct ocfs2_xattr_entry
*xe
;
5655 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5656 struct ocfs2_xattr_set_ctxt ctxt
= {NULL
, NULL
,};
5657 int credits
= ocfs2_remove_extent_credits(osb
->sb
) +
5658 ocfs2_blocks_per_xattr_bucket(inode
->i_sb
);
5659 struct ocfs2_xattr_value_root
*xv
;
5660 struct ocfs2_rm_xattr_bucket_para
*args
=
5661 (struct ocfs2_rm_xattr_bucket_para
*)para
;
5663 ocfs2_init_dealloc_ctxt(&ctxt
.dealloc
);
5665 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
5666 xe
= &xh
->xh_entries
[i
];
5667 if (ocfs2_xattr_is_local(xe
))
5670 ret
= ocfs2_get_xattr_tree_value_root(inode
->i_sb
, bucket
,
5677 ret
= ocfs2_lock_xattr_remove_allocators(inode
, xv
,
5683 ctxt
.handle
= ocfs2_start_trans(osb
, credits
+ ref_credits
);
5684 if (IS_ERR(ctxt
.handle
)) {
5685 ret
= PTR_ERR(ctxt
.handle
);
5690 ret
= ocfs2_xattr_bucket_value_truncate(inode
, bucket
,
5693 ocfs2_commit_trans(osb
, ctxt
.handle
);
5695 ocfs2_free_alloc_context(ctxt
.meta_ac
);
5696 ctxt
.meta_ac
= NULL
;
5705 ocfs2_free_alloc_context(ctxt
.meta_ac
);
5706 ocfs2_schedule_truncate_log_flush(osb
, 1);
5707 ocfs2_run_deallocs(osb
, &ctxt
.dealloc
);
5712 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5713 * or change the extent record flag), we need to recalculate
5714 * the metaecc for the whole bucket. So it is done here.
5717 * We have to give the extra credits for the caller.
5719 static int ocfs2_xattr_bucket_post_refcount(struct inode
*inode
,
5724 struct ocfs2_xattr_bucket
*bucket
=
5725 (struct ocfs2_xattr_bucket
*)para
;
5727 ret
= ocfs2_xattr_bucket_journal_access(handle
, bucket
,
5728 OCFS2_JOURNAL_ACCESS_WRITE
);
5734 ocfs2_xattr_bucket_journal_dirty(handle
, bucket
);
5740 * Special action we need if the xattr value is refcounted.
5742 * 1. If the xattr is refcounted, lock the tree.
5743 * 2. CoW the xattr if we are setting the new value and the value
5744 * will be stored outside.
5745 * 3. In other case, decrease_refcount will work for us, so just
5746 * lock the refcount tree, calculate the meta and credits is OK.
5748 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5749 * currently CoW is a completed transaction, while this function
5750 * will also lock the allocators and let us deadlock. So we will
5751 * CoW the whole xattr value.
5753 static int ocfs2_prepare_refcount_xattr(struct inode
*inode
,
5754 struct ocfs2_dinode
*di
,
5755 struct ocfs2_xattr_info
*xi
,
5756 struct ocfs2_xattr_search
*xis
,
5757 struct ocfs2_xattr_search
*xbs
,
5758 struct ocfs2_refcount_tree
**ref_tree
,
5763 struct ocfs2_xattr_block
*xb
;
5764 struct ocfs2_xattr_entry
*xe
;
5766 u32 p_cluster
, num_clusters
;
5767 unsigned int ext_flags
;
5768 int name_offset
, name_len
;
5769 struct ocfs2_xattr_value_buf vb
;
5770 struct ocfs2_xattr_bucket
*bucket
= NULL
;
5771 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
5772 struct ocfs2_post_refcount refcount
;
5773 struct ocfs2_post_refcount
*p
= NULL
;
5774 struct buffer_head
*ref_root_bh
= NULL
;
5776 if (!xis
->not_found
) {
5778 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
5779 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
5781 vb
.vb_bh
= xis
->inode_bh
;
5782 vb
.vb_access
= ocfs2_journal_access_di
;
5784 int i
, block_off
= 0;
5785 xb
= (struct ocfs2_xattr_block
*)xbs
->xattr_bh
->b_data
;
5787 name_offset
= le16_to_cpu(xe
->xe_name_offset
);
5788 name_len
= OCFS2_XATTR_SIZE(xe
->xe_name_len
);
5789 i
= xbs
->here
- xbs
->header
->xh_entries
;
5791 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
) {
5792 ret
= ocfs2_xattr_bucket_get_name_value(inode
->i_sb
,
5793 bucket_xh(xbs
->bucket
),
5800 base
= bucket_block(xbs
->bucket
, block_off
);
5801 vb
.vb_bh
= xbs
->bucket
->bu_bhs
[block_off
];
5802 vb
.vb_access
= ocfs2_journal_access
;
5804 if (ocfs2_meta_ecc(osb
)) {
5805 /*create parameters for ocfs2_post_refcount. */
5806 bucket
= xbs
->bucket
;
5807 refcount
.credits
= bucket
->bu_blocks
;
5808 refcount
.para
= bucket
;
5810 ocfs2_xattr_bucket_post_refcount
;
5815 vb
.vb_bh
= xbs
->xattr_bh
;
5816 vb
.vb_access
= ocfs2_journal_access_xb
;
5820 if (ocfs2_xattr_is_local(xe
))
5823 vb
.vb_xv
= (struct ocfs2_xattr_value_root
*)
5824 (base
+ name_offset
+ name_len
);
5826 ret
= ocfs2_xattr_get_clusters(inode
, 0, &p_cluster
,
5827 &num_clusters
, &vb
.vb_xv
->xr_list
,
5835 * We just need to check the 1st extent record, since we always
5836 * CoW the whole xattr. So there shouldn't be a xattr with
5837 * some REFCOUNT extent recs after the 1st one.
5839 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
5842 ret
= ocfs2_lock_refcount_tree(osb
, le64_to_cpu(di
->i_refcount_loc
),
5843 1, ref_tree
, &ref_root_bh
);
5850 * If we are deleting the xattr or the new size will be stored inside,
5851 * cool, leave it there, the xattr truncate process will remove them
5852 * for us(it still needs the refcount tree lock and the meta, credits).
5853 * And the worse case is that every cluster truncate will split the
5854 * refcount tree, and make the original extent become 3. So we will need
5855 * 2 * cluster more extent recs at most.
5857 if (!xi
->xi_value
|| xi
->xi_value_len
<= OCFS2_XATTR_INLINE_SIZE
) {
5859 ret
= ocfs2_refcounted_xattr_delete_need(inode
,
5860 &(*ref_tree
)->rf_ci
,
5861 ref_root_bh
, vb
.vb_xv
,
5868 ret
= ocfs2_refcount_cow_xattr(inode
, di
, &vb
,
5869 *ref_tree
, ref_root_bh
, 0,
5870 le32_to_cpu(vb
.vb_xv
->xr_clusters
), p
);
5875 brelse(ref_root_bh
);
5880 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5881 * The physical clusters will be added to refcount tree.
5883 static int ocfs2_xattr_value_attach_refcount(struct inode
*inode
,
5884 struct ocfs2_xattr_value_root
*xv
,
5885 struct ocfs2_extent_tree
*value_et
,
5886 struct ocfs2_caching_info
*ref_ci
,
5887 struct buffer_head
*ref_root_bh
,
5888 struct ocfs2_cached_dealloc_ctxt
*dealloc
,
5889 struct ocfs2_post_refcount
*refcount
)
5892 u32 clusters
= le32_to_cpu(xv
->xr_clusters
);
5893 u32 cpos
, p_cluster
, num_clusters
;
5894 struct ocfs2_extent_list
*el
= &xv
->xr_list
;
5895 unsigned int ext_flags
;
5898 while (cpos
< clusters
) {
5899 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
5900 &num_clusters
, el
, &ext_flags
);
5906 cpos
+= num_clusters
;
5907 if ((ext_flags
& OCFS2_EXT_REFCOUNTED
))
5912 ret
= ocfs2_add_refcount_flag(inode
, value_et
,
5913 ref_ci
, ref_root_bh
,
5914 cpos
- num_clusters
,
5915 p_cluster
, num_clusters
,
5927 * Given a normal ocfs2_xattr_header, refcount all the entries which
5928 * have value stored outside.
5929 * Used for xattrs stored in inode and ocfs2_xattr_block.
5931 static int ocfs2_xattr_attach_refcount_normal(struct inode
*inode
,
5932 struct ocfs2_xattr_value_buf
*vb
,
5933 struct ocfs2_xattr_header
*header
,
5934 struct ocfs2_caching_info
*ref_ci
,
5935 struct buffer_head
*ref_root_bh
,
5936 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
5939 struct ocfs2_xattr_entry
*xe
;
5940 struct ocfs2_xattr_value_root
*xv
;
5941 struct ocfs2_extent_tree et
;
5944 for (i
= 0; i
< le16_to_cpu(header
->xh_count
); i
++) {
5945 xe
= &header
->xh_entries
[i
];
5947 if (ocfs2_xattr_is_local(xe
))
5950 xv
= (struct ocfs2_xattr_value_root
*)((void *)header
+
5951 le16_to_cpu(xe
->xe_name_offset
) +
5952 OCFS2_XATTR_SIZE(xe
->xe_name_len
));
5955 ocfs2_init_xattr_value_extent_tree(&et
, INODE_CACHE(inode
), vb
);
5957 ret
= ocfs2_xattr_value_attach_refcount(inode
, xv
, &et
,
5958 ref_ci
, ref_root_bh
,
5969 static int ocfs2_xattr_inline_attach_refcount(struct inode
*inode
,
5970 struct buffer_head
*fe_bh
,
5971 struct ocfs2_caching_info
*ref_ci
,
5972 struct buffer_head
*ref_root_bh
,
5973 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
5975 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)fe_bh
->b_data
;
5976 struct ocfs2_xattr_header
*header
= (struct ocfs2_xattr_header
*)
5977 (fe_bh
->b_data
+ inode
->i_sb
->s_blocksize
-
5978 le16_to_cpu(di
->i_xattr_inline_size
));
5979 struct ocfs2_xattr_value_buf vb
= {
5981 .vb_access
= ocfs2_journal_access_di
,
5984 return ocfs2_xattr_attach_refcount_normal(inode
, &vb
, header
,
5985 ref_ci
, ref_root_bh
, dealloc
);
5988 struct ocfs2_xattr_tree_value_refcount_para
{
5989 struct ocfs2_caching_info
*ref_ci
;
5990 struct buffer_head
*ref_root_bh
;
5991 struct ocfs2_cached_dealloc_ctxt
*dealloc
;
5994 static int ocfs2_get_xattr_tree_value_root(struct super_block
*sb
,
5995 struct ocfs2_xattr_bucket
*bucket
,
5997 struct ocfs2_xattr_value_root
**xv
,
5998 struct buffer_head
**bh
)
6000 int ret
, block_off
, name_offset
;
6001 struct ocfs2_xattr_header
*xh
= bucket_xh(bucket
);
6002 struct ocfs2_xattr_entry
*xe
= &xh
->xh_entries
[offset
];
6005 ret
= ocfs2_xattr_bucket_get_name_value(sb
,
6015 base
= bucket_block(bucket
, block_off
);
6017 *xv
= (struct ocfs2_xattr_value_root
*)(base
+ name_offset
+
6018 OCFS2_XATTR_SIZE(xe
->xe_name_len
));
6021 *bh
= bucket
->bu_bhs
[block_off
];
6027 * For a given xattr bucket, refcount all the entries which
6028 * have value stored outside.
6030 static int ocfs2_xattr_bucket_value_refcount(struct inode
*inode
,
6031 struct ocfs2_xattr_bucket
*bucket
,
6035 struct ocfs2_extent_tree et
;
6036 struct ocfs2_xattr_tree_value_refcount_para
*ref
=
6037 (struct ocfs2_xattr_tree_value_refcount_para
*)para
;
6038 struct ocfs2_xattr_header
*xh
=
6039 (struct ocfs2_xattr_header
*)bucket
->bu_bhs
[0]->b_data
;
6040 struct ocfs2_xattr_entry
*xe
;
6041 struct ocfs2_xattr_value_buf vb
= {
6042 .vb_access
= ocfs2_journal_access
,
6044 struct ocfs2_post_refcount refcount
= {
6045 .credits
= bucket
->bu_blocks
,
6047 .func
= ocfs2_xattr_bucket_post_refcount
,
6049 struct ocfs2_post_refcount
*p
= NULL
;
6051 /* We only need post_refcount if we support metaecc. */
6052 if (ocfs2_meta_ecc(OCFS2_SB(inode
->i_sb
)))
6055 trace_ocfs2_xattr_bucket_value_refcount(
6056 (unsigned long long)bucket_blkno(bucket
),
6057 le16_to_cpu(xh
->xh_count
));
6058 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
6059 xe
= &xh
->xh_entries
[i
];
6061 if (ocfs2_xattr_is_local(xe
))
6064 ret
= ocfs2_get_xattr_tree_value_root(inode
->i_sb
, bucket
, i
,
6065 &vb
.vb_xv
, &vb
.vb_bh
);
6071 ocfs2_init_xattr_value_extent_tree(&et
,
6072 INODE_CACHE(inode
), &vb
);
6074 ret
= ocfs2_xattr_value_attach_refcount(inode
, vb
.vb_xv
,
6088 static int ocfs2_refcount_xattr_tree_rec(struct inode
*inode
,
6089 struct buffer_head
*root_bh
,
6090 u64 blkno
, u32 cpos
, u32 len
, void *para
)
6092 return ocfs2_iterate_xattr_buckets(inode
, blkno
, len
,
6093 ocfs2_xattr_bucket_value_refcount
,
6097 static int ocfs2_xattr_block_attach_refcount(struct inode
*inode
,
6098 struct buffer_head
*blk_bh
,
6099 struct ocfs2_caching_info
*ref_ci
,
6100 struct buffer_head
*ref_root_bh
,
6101 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
6104 struct ocfs2_xattr_block
*xb
=
6105 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
6107 if (!(le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)) {
6108 struct ocfs2_xattr_header
*header
= &xb
->xb_attrs
.xb_header
;
6109 struct ocfs2_xattr_value_buf vb
= {
6111 .vb_access
= ocfs2_journal_access_xb
,
6114 ret
= ocfs2_xattr_attach_refcount_normal(inode
, &vb
, header
,
6115 ref_ci
, ref_root_bh
,
6118 struct ocfs2_xattr_tree_value_refcount_para para
= {
6120 .ref_root_bh
= ref_root_bh
,
6124 ret
= ocfs2_iterate_xattr_index_block(inode
, blk_bh
,
6125 ocfs2_refcount_xattr_tree_rec
,
6132 int ocfs2_xattr_attach_refcount_tree(struct inode
*inode
,
6133 struct buffer_head
*fe_bh
,
6134 struct ocfs2_caching_info
*ref_ci
,
6135 struct buffer_head
*ref_root_bh
,
6136 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
6139 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
6140 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)fe_bh
->b_data
;
6141 struct buffer_head
*blk_bh
= NULL
;
6143 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
6144 ret
= ocfs2_xattr_inline_attach_refcount(inode
, fe_bh
,
6145 ref_ci
, ref_root_bh
,
6153 if (!di
->i_xattr_loc
)
6156 ret
= ocfs2_read_xattr_block(inode
, le64_to_cpu(di
->i_xattr_loc
),
6163 ret
= ocfs2_xattr_block_attach_refcount(inode
, blk_bh
, ref_ci
,
6164 ref_root_bh
, dealloc
);
6174 typedef int (should_xattr_reflinked
)(struct ocfs2_xattr_entry
*xe
);
6176 * Store the information we need in xattr reflink.
6177 * old_bh and new_bh are inode bh for the old and new inode.
6179 struct ocfs2_xattr_reflink
{
6180 struct inode
*old_inode
;
6181 struct inode
*new_inode
;
6182 struct buffer_head
*old_bh
;
6183 struct buffer_head
*new_bh
;
6184 struct ocfs2_caching_info
*ref_ci
;
6185 struct buffer_head
*ref_root_bh
;
6186 struct ocfs2_cached_dealloc_ctxt
*dealloc
;
6187 should_xattr_reflinked
*xattr_reflinked
;
6191 * Given a xattr header and xe offset,
6192 * return the proper xv and the corresponding bh.
6193 * xattr in inode, block and xattr tree have different implementaions.
6195 typedef int (get_xattr_value_root
)(struct super_block
*sb
,
6196 struct buffer_head
*bh
,
6197 struct ocfs2_xattr_header
*xh
,
6199 struct ocfs2_xattr_value_root
**xv
,
6200 struct buffer_head
**ret_bh
,
6204 * Calculate all the xattr value root metadata stored in this xattr header and
6205 * credits we need if we create them from the scratch.
6206 * We use get_xattr_value_root so that all types of xattr container can use it.
6208 static int ocfs2_value_metas_in_xattr_header(struct super_block
*sb
,
6209 struct buffer_head
*bh
,
6210 struct ocfs2_xattr_header
*xh
,
6211 int *metas
, int *credits
,
6213 get_xattr_value_root
*func
,
6217 struct ocfs2_xattr_value_root
*xv
;
6218 struct ocfs2_xattr_entry
*xe
;
6220 for (i
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++) {
6221 xe
= &xh
->xh_entries
[i
];
6222 if (ocfs2_xattr_is_local(xe
))
6225 ret
= func(sb
, bh
, xh
, i
, &xv
, NULL
, para
);
6231 *metas
+= le16_to_cpu(xv
->xr_list
.l_tree_depth
) *
6232 le16_to_cpu(xv
->xr_list
.l_next_free_rec
);
6234 *credits
+= ocfs2_calc_extend_credits(sb
,
6235 &def_xv
.xv
.xr_list
);
6238 * If the value is a tree with depth > 1, We don't go deep
6239 * to the extent block, so just calculate a maximum record num.
6241 if (!xv
->xr_list
.l_tree_depth
)
6242 *num_recs
+= le16_to_cpu(xv
->xr_list
.l_next_free_rec
);
6244 *num_recs
+= ocfs2_clusters_for_bytes(sb
,
6251 /* Used by xattr inode and block to return the right xv and buffer_head. */
6252 static int ocfs2_get_xattr_value_root(struct super_block
*sb
,
6253 struct buffer_head
*bh
,
6254 struct ocfs2_xattr_header
*xh
,
6256 struct ocfs2_xattr_value_root
**xv
,
6257 struct buffer_head
**ret_bh
,
6260 struct ocfs2_xattr_entry
*xe
= &xh
->xh_entries
[offset
];
6262 *xv
= (struct ocfs2_xattr_value_root
*)((void *)xh
+
6263 le16_to_cpu(xe
->xe_name_offset
) +
6264 OCFS2_XATTR_SIZE(xe
->xe_name_len
));
6273 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6274 * It is only used for inline xattr and xattr block.
6276 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super
*osb
,
6277 struct ocfs2_xattr_header
*xh
,
6278 struct buffer_head
*ref_root_bh
,
6280 struct ocfs2_alloc_context
**meta_ac
)
6282 int ret
, meta_add
= 0, num_recs
= 0;
6283 struct ocfs2_refcount_block
*rb
=
6284 (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
6288 ret
= ocfs2_value_metas_in_xattr_header(osb
->sb
, NULL
, xh
,
6289 &meta_add
, credits
, &num_recs
,
6290 ocfs2_get_xattr_value_root
,
6298 * We need to add/modify num_recs in refcount tree, so just calculate
6299 * an approximate number we need for refcount tree change.
6300 * Sometimes we need to split the tree, and after split, half recs
6301 * will be moved to the new block, and a new block can only provide
6302 * half number of recs. So we multiple new blocks by 2.
6304 num_recs
= num_recs
/ ocfs2_refcount_recs_per_rb(osb
->sb
) * 2;
6305 meta_add
+= num_recs
;
6306 *credits
+= num_recs
+ num_recs
* OCFS2_EXPAND_REFCOUNT_TREE_CREDITS
;
6307 if (le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
)
6308 *credits
+= le16_to_cpu(rb
->rf_list
.l_tree_depth
) *
6309 le16_to_cpu(rb
->rf_list
.l_next_free_rec
) + 1;
6313 ret
= ocfs2_reserve_new_metadata_blocks(osb
, meta_add
, meta_ac
);
6322 * Given a xattr header, reflink all the xattrs in this container.
6323 * It can be used for inode, block and bucket.
6326 * Before we call this function, the caller has memcpy the xattr in
6327 * old_xh to the new_xh.
6329 * If args.xattr_reflinked is set, call it to decide whether the xe should
6330 * be reflinked or not. If not, remove it from the new xattr header.
6332 static int ocfs2_reflink_xattr_header(handle_t
*handle
,
6333 struct ocfs2_xattr_reflink
*args
,
6334 struct buffer_head
*old_bh
,
6335 struct ocfs2_xattr_header
*xh
,
6336 struct buffer_head
*new_bh
,
6337 struct ocfs2_xattr_header
*new_xh
,
6338 struct ocfs2_xattr_value_buf
*vb
,
6339 struct ocfs2_alloc_context
*meta_ac
,
6340 get_xattr_value_root
*func
,
6344 struct super_block
*sb
= args
->old_inode
->i_sb
;
6345 struct buffer_head
*value_bh
;
6346 struct ocfs2_xattr_entry
*xe
, *last
;
6347 struct ocfs2_xattr_value_root
*xv
, *new_xv
;
6348 struct ocfs2_extent_tree data_et
;
6349 u32 clusters
, cpos
, p_cluster
, num_clusters
;
6350 unsigned int ext_flags
= 0;
6352 trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh
->b_blocknr
,
6353 le16_to_cpu(xh
->xh_count
));
6355 last
= &new_xh
->xh_entries
[le16_to_cpu(new_xh
->xh_count
)];
6356 for (i
= 0, j
= 0; i
< le16_to_cpu(xh
->xh_count
); i
++, j
++) {
6357 xe
= &xh
->xh_entries
[i
];
6359 if (args
->xattr_reflinked
&& !args
->xattr_reflinked(xe
)) {
6360 xe
= &new_xh
->xh_entries
[j
];
6362 le16_add_cpu(&new_xh
->xh_count
, -1);
6363 if (new_xh
->xh_count
) {
6365 (void *)last
- (void *)xe
);
6367 sizeof(struct ocfs2_xattr_entry
));
6371 * We don't want j to increase in the next round since
6372 * it is already moved ahead.
6378 if (ocfs2_xattr_is_local(xe
))
6381 ret
= func(sb
, old_bh
, xh
, i
, &xv
, NULL
, para
);
6387 ret
= func(sb
, new_bh
, new_xh
, j
, &new_xv
, &value_bh
, para
);
6394 * For the xattr which has l_tree_depth = 0, all the extent
6395 * recs have already be copied to the new xh with the
6396 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6397 * increase the refount count int the refcount tree.
6399 * For the xattr which has l_tree_depth > 0, we need
6400 * to initialize it to the empty default value root,
6401 * and then insert the extents one by one.
6403 if (xv
->xr_list
.l_tree_depth
) {
6404 memcpy(new_xv
, &def_xv
, OCFS2_XATTR_ROOT_SIZE
);
6406 vb
->vb_bh
= value_bh
;
6407 ocfs2_init_xattr_value_extent_tree(&data_et
,
6408 INODE_CACHE(args
->new_inode
), vb
);
6411 clusters
= le32_to_cpu(xv
->xr_clusters
);
6413 while (cpos
< clusters
) {
6414 ret
= ocfs2_xattr_get_clusters(args
->old_inode
,
6427 if (xv
->xr_list
.l_tree_depth
) {
6428 ret
= ocfs2_insert_extent(handle
,
6430 ocfs2_clusters_to_blocks(
6431 args
->old_inode
->i_sb
,
6433 num_clusters
, ext_flags
,
6441 ret
= ocfs2_increase_refcount(handle
, args
->ref_ci
,
6443 p_cluster
, num_clusters
,
6444 meta_ac
, args
->dealloc
);
6450 cpos
+= num_clusters
;
6458 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink
*args
)
6460 int ret
= 0, credits
= 0;
6462 struct ocfs2_super
*osb
= OCFS2_SB(args
->old_inode
->i_sb
);
6463 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)args
->old_bh
->b_data
;
6464 int inline_size
= le16_to_cpu(di
->i_xattr_inline_size
);
6465 int header_off
= osb
->sb
->s_blocksize
- inline_size
;
6466 struct ocfs2_xattr_header
*xh
= (struct ocfs2_xattr_header
*)
6467 (args
->old_bh
->b_data
+ header_off
);
6468 struct ocfs2_xattr_header
*new_xh
= (struct ocfs2_xattr_header
*)
6469 (args
->new_bh
->b_data
+ header_off
);
6470 struct ocfs2_alloc_context
*meta_ac
= NULL
;
6471 struct ocfs2_inode_info
*new_oi
;
6472 struct ocfs2_dinode
*new_di
;
6473 struct ocfs2_xattr_value_buf vb
= {
6474 .vb_bh
= args
->new_bh
,
6475 .vb_access
= ocfs2_journal_access_di
,
6478 ret
= ocfs2_reflink_lock_xattr_allocators(osb
, xh
, args
->ref_root_bh
,
6479 &credits
, &meta_ac
);
6485 handle
= ocfs2_start_trans(osb
, credits
);
6486 if (IS_ERR(handle
)) {
6487 ret
= PTR_ERR(handle
);
6492 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(args
->new_inode
),
6493 args
->new_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
6499 memcpy(args
->new_bh
->b_data
+ header_off
,
6500 args
->old_bh
->b_data
+ header_off
, inline_size
);
6502 new_di
= (struct ocfs2_dinode
*)args
->new_bh
->b_data
;
6503 new_di
->i_xattr_inline_size
= cpu_to_le16(inline_size
);
6505 ret
= ocfs2_reflink_xattr_header(handle
, args
, args
->old_bh
, xh
,
6506 args
->new_bh
, new_xh
, &vb
, meta_ac
,
6507 ocfs2_get_xattr_value_root
, NULL
);
6513 new_oi
= OCFS2_I(args
->new_inode
);
6515 * Adjust extent record count to reserve space for extended attribute.
6516 * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6518 if (!(new_oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) &&
6519 !(ocfs2_inode_is_fast_symlink(args
->new_inode
))) {
6520 struct ocfs2_extent_list
*el
= &new_di
->id2
.i_list
;
6521 le16_add_cpu(&el
->l_count
, -(inline_size
/
6522 sizeof(struct ocfs2_extent_rec
)));
6524 spin_lock(&new_oi
->ip_lock
);
6525 new_oi
->ip_dyn_features
|= OCFS2_HAS_XATTR_FL
| OCFS2_INLINE_XATTR_FL
;
6526 new_di
->i_dyn_features
= cpu_to_le16(new_oi
->ip_dyn_features
);
6527 spin_unlock(&new_oi
->ip_lock
);
6529 ocfs2_journal_dirty(handle
, args
->new_bh
);
6532 ocfs2_commit_trans(osb
, handle
);
6536 ocfs2_free_alloc_context(meta_ac
);
6540 static int ocfs2_create_empty_xattr_block(struct inode
*inode
,
6541 struct buffer_head
*fe_bh
,
6542 struct buffer_head
**ret_bh
,
6546 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
6547 struct ocfs2_xattr_set_ctxt ctxt
;
6549 memset(&ctxt
, 0, sizeof(ctxt
));
6550 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, &ctxt
.meta_ac
);
6556 ctxt
.handle
= ocfs2_start_trans(osb
, OCFS2_XATTR_BLOCK_CREATE_CREDITS
);
6557 if (IS_ERR(ctxt
.handle
)) {
6558 ret
= PTR_ERR(ctxt
.handle
);
6563 trace_ocfs2_create_empty_xattr_block(
6564 (unsigned long long)fe_bh
->b_blocknr
, indexed
);
6565 ret
= ocfs2_create_xattr_block(inode
, fe_bh
, &ctxt
, indexed
,
6570 ocfs2_commit_trans(osb
, ctxt
.handle
);
6572 ocfs2_free_alloc_context(ctxt
.meta_ac
);
6576 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink
*args
,
6577 struct buffer_head
*blk_bh
,
6578 struct buffer_head
*new_blk_bh
)
6580 int ret
= 0, credits
= 0;
6582 struct ocfs2_inode_info
*new_oi
= OCFS2_I(args
->new_inode
);
6583 struct ocfs2_dinode
*new_di
;
6584 struct ocfs2_super
*osb
= OCFS2_SB(args
->new_inode
->i_sb
);
6585 int header_off
= offsetof(struct ocfs2_xattr_block
, xb_attrs
.xb_header
);
6586 struct ocfs2_xattr_block
*xb
=
6587 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
6588 struct ocfs2_xattr_header
*xh
= &xb
->xb_attrs
.xb_header
;
6589 struct ocfs2_xattr_block
*new_xb
=
6590 (struct ocfs2_xattr_block
*)new_blk_bh
->b_data
;
6591 struct ocfs2_xattr_header
*new_xh
= &new_xb
->xb_attrs
.xb_header
;
6592 struct ocfs2_alloc_context
*meta_ac
;
6593 struct ocfs2_xattr_value_buf vb
= {
6594 .vb_bh
= new_blk_bh
,
6595 .vb_access
= ocfs2_journal_access_xb
,
6598 ret
= ocfs2_reflink_lock_xattr_allocators(osb
, xh
, args
->ref_root_bh
,
6599 &credits
, &meta_ac
);
6605 /* One more credits in case we need to add xattr flags in new inode. */
6606 handle
= ocfs2_start_trans(osb
, credits
+ 1);
6607 if (IS_ERR(handle
)) {
6608 ret
= PTR_ERR(handle
);
6613 if (!(new_oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
)) {
6614 ret
= ocfs2_journal_access_di(handle
,
6615 INODE_CACHE(args
->new_inode
),
6617 OCFS2_JOURNAL_ACCESS_WRITE
);
6624 ret
= ocfs2_journal_access_xb(handle
, INODE_CACHE(args
->new_inode
),
6625 new_blk_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
6631 memcpy(new_blk_bh
->b_data
+ header_off
, blk_bh
->b_data
+ header_off
,
6632 osb
->sb
->s_blocksize
- header_off
);
6634 ret
= ocfs2_reflink_xattr_header(handle
, args
, blk_bh
, xh
,
6635 new_blk_bh
, new_xh
, &vb
, meta_ac
,
6636 ocfs2_get_xattr_value_root
, NULL
);
6642 ocfs2_journal_dirty(handle
, new_blk_bh
);
6644 if (!(new_oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
)) {
6645 new_di
= (struct ocfs2_dinode
*)args
->new_bh
->b_data
;
6646 spin_lock(&new_oi
->ip_lock
);
6647 new_oi
->ip_dyn_features
|= OCFS2_HAS_XATTR_FL
;
6648 new_di
->i_dyn_features
= cpu_to_le16(new_oi
->ip_dyn_features
);
6649 spin_unlock(&new_oi
->ip_lock
);
6651 ocfs2_journal_dirty(handle
, args
->new_bh
);
6655 ocfs2_commit_trans(osb
, handle
);
6658 ocfs2_free_alloc_context(meta_ac
);
6662 struct ocfs2_reflink_xattr_tree_args
{
6663 struct ocfs2_xattr_reflink
*reflink
;
6664 struct buffer_head
*old_blk_bh
;
6665 struct buffer_head
*new_blk_bh
;
6666 struct ocfs2_xattr_bucket
*old_bucket
;
6667 struct ocfs2_xattr_bucket
*new_bucket
;
6672 * We have to handle the case that both old bucket and new bucket
6673 * will call this function to get the right ret_bh.
6674 * So The caller must give us the right bh.
6676 static int ocfs2_get_reflink_xattr_value_root(struct super_block
*sb
,
6677 struct buffer_head
*bh
,
6678 struct ocfs2_xattr_header
*xh
,
6680 struct ocfs2_xattr_value_root
**xv
,
6681 struct buffer_head
**ret_bh
,
6684 struct ocfs2_reflink_xattr_tree_args
*args
=
6685 (struct ocfs2_reflink_xattr_tree_args
*)para
;
6686 struct ocfs2_xattr_bucket
*bucket
;
6688 if (bh
== args
->old_bucket
->bu_bhs
[0])
6689 bucket
= args
->old_bucket
;
6691 bucket
= args
->new_bucket
;
6693 return ocfs2_get_xattr_tree_value_root(sb
, bucket
, offset
,
6697 struct ocfs2_value_tree_metas
{
6703 static int ocfs2_value_tree_metas_in_bucket(struct super_block
*sb
,
6704 struct buffer_head
*bh
,
6705 struct ocfs2_xattr_header
*xh
,
6707 struct ocfs2_xattr_value_root
**xv
,
6708 struct buffer_head
**ret_bh
,
6711 struct ocfs2_xattr_bucket
*bucket
=
6712 (struct ocfs2_xattr_bucket
*)para
;
6714 return ocfs2_get_xattr_tree_value_root(sb
, bucket
, offset
,
6718 static int ocfs2_calc_value_tree_metas(struct inode
*inode
,
6719 struct ocfs2_xattr_bucket
*bucket
,
6722 struct ocfs2_value_tree_metas
*metas
=
6723 (struct ocfs2_value_tree_metas
*)para
;
6724 struct ocfs2_xattr_header
*xh
=
6725 (struct ocfs2_xattr_header
*)bucket
->bu_bhs
[0]->b_data
;
6727 /* Add the credits for this bucket first. */
6728 metas
->credits
+= bucket
->bu_blocks
;
6729 return ocfs2_value_metas_in_xattr_header(inode
->i_sb
, bucket
->bu_bhs
[0],
6730 xh
, &metas
->num_metas
,
6731 &metas
->credits
, &metas
->num_recs
,
6732 ocfs2_value_tree_metas_in_bucket
,
6737 * Given a xattr extent rec starting from blkno and having len clusters,
6738 * iterate all the buckets calculate how much metadata we need for reflinking
6739 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6741 static int ocfs2_lock_reflink_xattr_rec_allocators(
6742 struct ocfs2_reflink_xattr_tree_args
*args
,
6743 struct ocfs2_extent_tree
*xt_et
,
6744 u64 blkno
, u32 len
, int *credits
,
6745 struct ocfs2_alloc_context
**meta_ac
,
6746 struct ocfs2_alloc_context
**data_ac
)
6748 int ret
, num_free_extents
;
6749 struct ocfs2_value_tree_metas metas
;
6750 struct ocfs2_super
*osb
= OCFS2_SB(args
->reflink
->old_inode
->i_sb
);
6751 struct ocfs2_refcount_block
*rb
;
6753 memset(&metas
, 0, sizeof(metas
));
6755 ret
= ocfs2_iterate_xattr_buckets(args
->reflink
->old_inode
, blkno
, len
,
6756 ocfs2_calc_value_tree_metas
, &metas
);
6762 *credits
= metas
.credits
;
6765 * Calculate we need for refcount tree change.
6767 * We need to add/modify num_recs in refcount tree, so just calculate
6768 * an approximate number we need for refcount tree change.
6769 * Sometimes we need to split the tree, and after split, half recs
6770 * will be moved to the new block, and a new block can only provide
6771 * half number of recs. So we multiple new blocks by 2.
6772 * In the end, we have to add credits for modifying the already
6773 * existed refcount block.
6775 rb
= (struct ocfs2_refcount_block
*)args
->reflink
->ref_root_bh
->b_data
;
6777 (metas
.num_recs
+ ocfs2_refcount_recs_per_rb(osb
->sb
) - 1) /
6778 ocfs2_refcount_recs_per_rb(osb
->sb
) * 2;
6779 metas
.num_metas
+= metas
.num_recs
;
6780 *credits
+= metas
.num_recs
+
6781 metas
.num_recs
* OCFS2_EXPAND_REFCOUNT_TREE_CREDITS
;
6782 if (le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
)
6783 *credits
+= le16_to_cpu(rb
->rf_list
.l_tree_depth
) *
6784 le16_to_cpu(rb
->rf_list
.l_next_free_rec
) + 1;
6788 /* count in the xattr tree change. */
6789 num_free_extents
= ocfs2_num_free_extents(xt_et
);
6790 if (num_free_extents
< 0) {
6791 ret
= num_free_extents
;
6796 if (num_free_extents
< len
)
6797 metas
.num_metas
+= ocfs2_extend_meta_needed(xt_et
->et_root_el
);
6799 *credits
+= ocfs2_calc_extend_credits(osb
->sb
,
6802 if (metas
.num_metas
) {
6803 ret
= ocfs2_reserve_new_metadata_blocks(osb
, metas
.num_metas
,
6812 ret
= ocfs2_reserve_clusters(osb
, len
, data_ac
);
6819 ocfs2_free_alloc_context(*meta_ac
);
6827 static int ocfs2_reflink_xattr_bucket(handle_t
*handle
,
6828 u64 blkno
, u64 new_blkno
, u32 clusters
,
6829 u32
*cpos
, int num_buckets
,
6830 struct ocfs2_alloc_context
*meta_ac
,
6831 struct ocfs2_alloc_context
*data_ac
,
6832 struct ocfs2_reflink_xattr_tree_args
*args
)
6835 struct super_block
*sb
= args
->reflink
->old_inode
->i_sb
;
6836 int bpb
= args
->old_bucket
->bu_blocks
;
6837 struct ocfs2_xattr_value_buf vb
= {
6838 .vb_access
= ocfs2_journal_access
,
6841 for (i
= 0; i
< num_buckets
; i
++, blkno
+= bpb
, new_blkno
+= bpb
) {
6842 ret
= ocfs2_read_xattr_bucket(args
->old_bucket
, blkno
);
6848 ret
= ocfs2_init_xattr_bucket(args
->new_bucket
, new_blkno
, 1);
6854 ret
= ocfs2_xattr_bucket_journal_access(handle
,
6856 OCFS2_JOURNAL_ACCESS_CREATE
);
6862 for (j
= 0; j
< bpb
; j
++)
6863 memcpy(bucket_block(args
->new_bucket
, j
),
6864 bucket_block(args
->old_bucket
, j
),
6868 * Record the start cpos so that we can use it to initialize
6869 * our xattr tree we also set the xh_num_bucket for the new
6873 *cpos
= le32_to_cpu(bucket_xh(args
->new_bucket
)->
6874 xh_entries
[0].xe_name_hash
);
6875 bucket_xh(args
->new_bucket
)->xh_num_buckets
=
6876 cpu_to_le16(num_buckets
);
6879 ocfs2_xattr_bucket_journal_dirty(handle
, args
->new_bucket
);
6881 ret
= ocfs2_reflink_xattr_header(handle
, args
->reflink
,
6882 args
->old_bucket
->bu_bhs
[0],
6883 bucket_xh(args
->old_bucket
),
6884 args
->new_bucket
->bu_bhs
[0],
6885 bucket_xh(args
->new_bucket
),
6887 ocfs2_get_reflink_xattr_value_root
,
6895 * Re-access and dirty the bucket to calculate metaecc.
6896 * Because we may extend the transaction in reflink_xattr_header
6897 * which will let the already accessed block gone.
6899 ret
= ocfs2_xattr_bucket_journal_access(handle
,
6901 OCFS2_JOURNAL_ACCESS_WRITE
);
6907 ocfs2_xattr_bucket_journal_dirty(handle
, args
->new_bucket
);
6909 ocfs2_xattr_bucket_relse(args
->old_bucket
);
6910 ocfs2_xattr_bucket_relse(args
->new_bucket
);
6913 ocfs2_xattr_bucket_relse(args
->old_bucket
);
6914 ocfs2_xattr_bucket_relse(args
->new_bucket
);
6918 static int ocfs2_reflink_xattr_buckets(handle_t
*handle
,
6919 struct inode
*inode
,
6920 struct ocfs2_reflink_xattr_tree_args
*args
,
6921 struct ocfs2_extent_tree
*et
,
6922 struct ocfs2_alloc_context
*meta_ac
,
6923 struct ocfs2_alloc_context
*data_ac
,
6924 u64 blkno
, u32 cpos
, u32 len
)
6926 int ret
, first_inserted
= 0;
6927 u32 p_cluster
, num_clusters
, reflink_cpos
= 0;
6929 unsigned int num_buckets
, reflink_buckets
;
6931 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode
->i_sb
));
6933 ret
= ocfs2_read_xattr_bucket(args
->old_bucket
, blkno
);
6938 num_buckets
= le16_to_cpu(bucket_xh(args
->old_bucket
)->xh_num_buckets
);
6939 ocfs2_xattr_bucket_relse(args
->old_bucket
);
6941 while (len
&& num_buckets
) {
6942 ret
= ocfs2_claim_clusters(handle
, data_ac
,
6943 1, &p_cluster
, &num_clusters
);
6949 new_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
6950 reflink_buckets
= min(num_buckets
, bpc
* num_clusters
);
6952 ret
= ocfs2_reflink_xattr_bucket(handle
, blkno
,
6953 new_blkno
, num_clusters
,
6954 &reflink_cpos
, reflink_buckets
,
6955 meta_ac
, data_ac
, args
);
6962 * For the 1st allocated cluster, we make it use the same cpos
6963 * so that the xattr tree looks the same as the original one
6966 if (!first_inserted
) {
6967 reflink_cpos
= cpos
;
6970 ret
= ocfs2_insert_extent(handle
, et
, reflink_cpos
, new_blkno
,
6971 num_clusters
, 0, meta_ac
);
6975 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno
,
6976 num_clusters
, reflink_cpos
);
6978 len
-= num_clusters
;
6979 blkno
+= ocfs2_clusters_to_blocks(inode
->i_sb
, num_clusters
);
6980 num_buckets
-= reflink_buckets
;
6987 * Create the same xattr extent record in the new inode's xattr tree.
6989 static int ocfs2_reflink_xattr_rec(struct inode
*inode
,
6990 struct buffer_head
*root_bh
,
6996 int ret
, credits
= 0;
6998 struct ocfs2_reflink_xattr_tree_args
*args
=
6999 (struct ocfs2_reflink_xattr_tree_args
*)para
;
7000 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
7001 struct ocfs2_alloc_context
*meta_ac
= NULL
;
7002 struct ocfs2_alloc_context
*data_ac
= NULL
;
7003 struct ocfs2_extent_tree et
;
7005 trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno
, len
);
7007 ocfs2_init_xattr_tree_extent_tree(&et
,
7008 INODE_CACHE(args
->reflink
->new_inode
),
7011 ret
= ocfs2_lock_reflink_xattr_rec_allocators(args
, &et
, blkno
,
7013 &meta_ac
, &data_ac
);
7019 handle
= ocfs2_start_trans(osb
, credits
);
7020 if (IS_ERR(handle
)) {
7021 ret
= PTR_ERR(handle
);
7026 ret
= ocfs2_reflink_xattr_buckets(handle
, inode
, args
, &et
,
7032 ocfs2_commit_trans(osb
, handle
);
7036 ocfs2_free_alloc_context(meta_ac
);
7038 ocfs2_free_alloc_context(data_ac
);
7043 * Create reflinked xattr buckets.
7044 * We will add bucket one by one, and refcount all the xattrs in the bucket
7045 * if they are stored outside.
7047 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink
*args
,
7048 struct buffer_head
*blk_bh
,
7049 struct buffer_head
*new_blk_bh
)
7052 struct ocfs2_reflink_xattr_tree_args para
;
7054 memset(¶
, 0, sizeof(para
));
7055 para
.reflink
= args
;
7056 para
.old_blk_bh
= blk_bh
;
7057 para
.new_blk_bh
= new_blk_bh
;
7059 para
.old_bucket
= ocfs2_xattr_bucket_new(args
->old_inode
);
7060 if (!para
.old_bucket
) {
7061 mlog_errno(-ENOMEM
);
7065 para
.new_bucket
= ocfs2_xattr_bucket_new(args
->new_inode
);
7066 if (!para
.new_bucket
) {
7072 ret
= ocfs2_iterate_xattr_index_block(args
->old_inode
, blk_bh
,
7073 ocfs2_reflink_xattr_rec
,
7079 ocfs2_xattr_bucket_free(para
.old_bucket
);
7080 ocfs2_xattr_bucket_free(para
.new_bucket
);
7084 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink
*args
,
7085 struct buffer_head
*blk_bh
)
7087 int ret
, indexed
= 0;
7088 struct buffer_head
*new_blk_bh
= NULL
;
7089 struct ocfs2_xattr_block
*xb
=
7090 (struct ocfs2_xattr_block
*)blk_bh
->b_data
;
7093 if (le16_to_cpu(xb
->xb_flags
) & OCFS2_XATTR_INDEXED
)
7096 ret
= ocfs2_create_empty_xattr_block(args
->new_inode
, args
->new_bh
,
7097 &new_blk_bh
, indexed
);
7104 ret
= ocfs2_reflink_xattr_block(args
, blk_bh
, new_blk_bh
);
7106 ret
= ocfs2_reflink_xattr_tree(args
, blk_bh
, new_blk_bh
);
7115 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry
*xe
)
7117 int type
= ocfs2_xattr_get_type(xe
);
7119 return type
!= OCFS2_XATTR_INDEX_SECURITY
&&
7120 type
!= OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS
&&
7121 type
!= OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT
;
7124 int ocfs2_reflink_xattrs(struct inode
*old_inode
,
7125 struct buffer_head
*old_bh
,
7126 struct inode
*new_inode
,
7127 struct buffer_head
*new_bh
,
7128 bool preserve_security
)
7131 struct ocfs2_xattr_reflink args
;
7132 struct ocfs2_inode_info
*oi
= OCFS2_I(old_inode
);
7133 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)old_bh
->b_data
;
7134 struct buffer_head
*blk_bh
= NULL
;
7135 struct ocfs2_cached_dealloc_ctxt dealloc
;
7136 struct ocfs2_refcount_tree
*ref_tree
;
7137 struct buffer_head
*ref_root_bh
= NULL
;
7139 ret
= ocfs2_lock_refcount_tree(OCFS2_SB(old_inode
->i_sb
),
7140 le64_to_cpu(di
->i_refcount_loc
),
7141 1, &ref_tree
, &ref_root_bh
);
7147 ocfs2_init_dealloc_ctxt(&dealloc
);
7149 args
.old_inode
= old_inode
;
7150 args
.new_inode
= new_inode
;
7151 args
.old_bh
= old_bh
;
7152 args
.new_bh
= new_bh
;
7153 args
.ref_ci
= &ref_tree
->rf_ci
;
7154 args
.ref_root_bh
= ref_root_bh
;
7155 args
.dealloc
= &dealloc
;
7156 if (preserve_security
)
7157 args
.xattr_reflinked
= NULL
;
7159 args
.xattr_reflinked
= ocfs2_reflink_xattr_no_security
;
7161 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
) {
7162 ret
= ocfs2_reflink_xattr_inline(&args
);
7169 if (!di
->i_xattr_loc
)
7172 ret
= ocfs2_read_xattr_block(old_inode
, le64_to_cpu(di
->i_xattr_loc
),
7179 ret
= ocfs2_reflink_xattr_in_block(&args
, blk_bh
);
7186 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode
->i_sb
),
7188 brelse(ref_root_bh
);
7190 if (ocfs2_dealloc_has_cluster(&dealloc
)) {
7191 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode
->i_sb
), 1);
7192 ocfs2_run_deallocs(OCFS2_SB(old_inode
->i_sb
), &dealloc
);
7200 * Initialize security and acl for a already created inode.
7201 * Used for reflink a non-preserve-security file.
7203 * It uses common api like ocfs2_xattr_set, so the caller
7204 * must not hold any lock expect i_rwsem.
7206 int ocfs2_init_security_and_acl(struct inode
*dir
,
7207 struct inode
*inode
,
7208 const struct qstr
*qstr
)
7211 struct buffer_head
*dir_bh
= NULL
;
7213 ret
= ocfs2_init_security_get(inode
, dir
, qstr
, NULL
);
7219 ret
= ocfs2_inode_lock(dir
, &dir_bh
, 0);
7224 ret
= ocfs2_init_acl(NULL
, inode
, dir
, NULL
, dir_bh
, NULL
, NULL
);
7228 ocfs2_inode_unlock(dir
, 0);
7235 * 'security' attributes support
7237 static int ocfs2_xattr_security_get(const struct xattr_handler
*handler
,
7238 struct dentry
*unused
, struct inode
*inode
,
7239 const char *name
, void *buffer
, size_t size
)
7241 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_SECURITY
,
7242 name
, buffer
, size
);
7245 static int ocfs2_xattr_security_set(const struct xattr_handler
*handler
,
7246 struct mnt_idmap
*idmap
,
7247 struct dentry
*unused
, struct inode
*inode
,
7248 const char *name
, const void *value
,
7249 size_t size
, int flags
)
7251 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_SECURITY
,
7252 name
, value
, size
, flags
);
7255 static int ocfs2_initxattrs(struct inode
*inode
, const struct xattr
*xattr_array
,
7258 const struct xattr
*xattr
;
7261 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
7262 err
= ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_SECURITY
,
7263 xattr
->name
, xattr
->value
,
7264 xattr
->value_len
, XATTR_CREATE
);
7271 int ocfs2_init_security_get(struct inode
*inode
,
7273 const struct qstr
*qstr
,
7274 struct ocfs2_security_xattr_info
*si
)
7276 /* check whether ocfs2 support feature xattr */
7277 if (!ocfs2_supports_xattr(OCFS2_SB(dir
->i_sb
)))
7280 return security_old_inode_init_security(inode
, dir
, qstr
,
7281 &si
->name
, &si
->value
,
7284 return security_inode_init_security(inode
, dir
, qstr
,
7285 &ocfs2_initxattrs
, NULL
);
7288 int ocfs2_init_security_set(handle_t
*handle
,
7289 struct inode
*inode
,
7290 struct buffer_head
*di_bh
,
7291 struct ocfs2_security_xattr_info
*si
,
7292 struct ocfs2_alloc_context
*xattr_ac
,
7293 struct ocfs2_alloc_context
*data_ac
)
7295 return ocfs2_xattr_set_handle(handle
, inode
, di_bh
,
7296 OCFS2_XATTR_INDEX_SECURITY
,
7297 si
->name
, si
->value
, si
->value_len
, 0,
7301 const struct xattr_handler ocfs2_xattr_security_handler
= {
7302 .prefix
= XATTR_SECURITY_PREFIX
,
7303 .get
= ocfs2_xattr_security_get
,
7304 .set
= ocfs2_xattr_security_set
,
7308 * 'trusted' attributes support
7310 static int ocfs2_xattr_trusted_get(const struct xattr_handler
*handler
,
7311 struct dentry
*unused
, struct inode
*inode
,
7312 const char *name
, void *buffer
, size_t size
)
7314 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_TRUSTED
,
7315 name
, buffer
, size
);
7318 static int ocfs2_xattr_trusted_set(const struct xattr_handler
*handler
,
7319 struct mnt_idmap
*idmap
,
7320 struct dentry
*unused
, struct inode
*inode
,
7321 const char *name
, const void *value
,
7322 size_t size
, int flags
)
7324 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_TRUSTED
,
7325 name
, value
, size
, flags
);
7328 const struct xattr_handler ocfs2_xattr_trusted_handler
= {
7329 .prefix
= XATTR_TRUSTED_PREFIX
,
7330 .get
= ocfs2_xattr_trusted_get
,
7331 .set
= ocfs2_xattr_trusted_set
,
7335 * 'user' attributes support
7337 static int ocfs2_xattr_user_get(const struct xattr_handler
*handler
,
7338 struct dentry
*unused
, struct inode
*inode
,
7339 const char *name
, void *buffer
, size_t size
)
7341 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
7343 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
7345 return ocfs2_xattr_get(inode
, OCFS2_XATTR_INDEX_USER
, name
,
7349 static int ocfs2_xattr_user_set(const struct xattr_handler
*handler
,
7350 struct mnt_idmap
*idmap
,
7351 struct dentry
*unused
, struct inode
*inode
,
7352 const char *name
, const void *value
,
7353 size_t size
, int flags
)
7355 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
7357 if (osb
->s_mount_opt
& OCFS2_MOUNT_NOUSERXATTR
)
7360 return ocfs2_xattr_set(inode
, OCFS2_XATTR_INDEX_USER
,
7361 name
, value
, size
, flags
);
7364 const struct xattr_handler ocfs2_xattr_user_handler
= {
7365 .prefix
= XATTR_USER_PREFIX
,
7366 .get
= ocfs2_xattr_user_get
,
7367 .set
= ocfs2_xattr_user_set
,