1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
22 #include <linux/capability.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
45 #include "blockcheck.h"
55 #include "buffer_head_io.h"
58 #include "refcounttree.h"
61 struct ocfs2_xattr_def_value_root {
62 struct ocfs2_xattr_value_root xv;
63 struct ocfs2_extent_rec er;
66 struct ocfs2_xattr_bucket {
67 /* The inode these xattrs are associated with */
68 struct inode *bu_inode;
70 /* The actual buffers that make up the bucket */
71 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
73 /* How many blocks make up one bucket for this filesystem */
77 struct ocfs2_xattr_set_ctxt {
79 struct ocfs2_alloc_context *meta_ac;
80 struct ocfs2_alloc_context *data_ac;
81 struct ocfs2_cached_dealloc_ctxt dealloc;
84 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
85 #define OCFS2_XATTR_INLINE_SIZE 80
86 #define OCFS2_XATTR_HEADER_GAP 4
87 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
88 - sizeof(struct ocfs2_xattr_header) \
89 - OCFS2_XATTR_HEADER_GAP)
90 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
91 - sizeof(struct ocfs2_xattr_block) \
92 - sizeof(struct ocfs2_xattr_header) \
93 - OCFS2_XATTR_HEADER_GAP)
95 static struct ocfs2_xattr_def_value_root def_xv = {
96 .xv.xr_list.l_count = cpu_to_le16(1),
99 struct xattr_handler *ocfs2_xattr_handlers[] = {
100 &ocfs2_xattr_user_handler,
101 &ocfs2_xattr_acl_access_handler,
102 &ocfs2_xattr_acl_default_handler,
103 &ocfs2_xattr_trusted_handler,
104 &ocfs2_xattr_security_handler,
108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
109 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
110 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
111 = &ocfs2_xattr_acl_access_handler,
112 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
113 = &ocfs2_xattr_acl_default_handler,
114 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
115 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
118 struct ocfs2_xattr_info {
122 const void *xi_value;
126 struct ocfs2_xattr_search {
127 struct buffer_head *inode_bh;
129 * xattr_bh point to the block buffer head which has extended attribute
130 * when extended attribute in inode, xattr_bh is equal to inode_bh.
132 struct buffer_head *xattr_bh;
133 struct ocfs2_xattr_header *header;
134 struct ocfs2_xattr_bucket *bucket;
137 struct ocfs2_xattr_entry *here;
141 /* Operations on struct ocfs2_xa_entry */
143 struct ocfs2_xa_loc_operations {
147 int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
149 void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
152 * Return a pointer to the appropriate buffer in loc->xl_storage
153 * at the given offset from loc->xl_header.
155 void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
157 /* Can we reuse the existing entry for the new value? */
158 int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
159 struct ocfs2_xattr_info *xi);
161 /* How much space is needed for the new value? */
162 int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
163 struct ocfs2_xattr_info *xi);
166 * Return the offset of the first name+value pair. This is
167 * the start of our downward-filling free space.
169 int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
172 * Remove the name+value at this location. Do whatever is
173 * appropriate with the remaining name+value pairs.
175 void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
177 /* Fill xl_entry with a new entry */
178 void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
180 /* Add name+value storage to an entry */
181 void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
184 * Initialize the value buf's access and bh fields for this entry.
185 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
187 void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
188 struct ocfs2_xattr_value_buf *vb);
192 * Describes an xattr entry location. This is a memory structure
193 * tracking the on-disk structure.
195 struct ocfs2_xa_loc {
196 /* This xattr belongs to this inode */
197 struct inode *xl_inode;
199 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
200 struct ocfs2_xattr_header *xl_header;
202 /* Bytes from xl_header to the end of the storage */
206 * The ocfs2_xattr_entry this location describes. If this is
207 * NULL, this location describes the on-disk structure where it
210 struct ocfs2_xattr_entry *xl_entry;
213 * Internal housekeeping
216 /* Buffer(s) containing this entry */
219 /* Operations on the storage backing this location */
220 const struct ocfs2_xa_loc_operations *xl_ops;
224 * Convenience functions to calculate how much space is needed for a
225 * given name+value pair
227 static int namevalue_size(int name_len, uint64_t value_len)
229 if (value_len > OCFS2_XATTR_INLINE_SIZE)
230 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
232 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
235 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
237 return namevalue_size(xi->xi_name_len, xi->xi_value_len);
240 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
242 u64 value_len = le64_to_cpu(xe->xe_value_size);
244 BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
245 ocfs2_xattr_is_local(xe));
246 return namevalue_size(xe->xe_name_len, value_len);
250 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
251 struct ocfs2_xattr_header *xh,
256 static int ocfs2_xattr_block_find(struct inode *inode,
259 struct ocfs2_xattr_search *xs);
260 static int ocfs2_xattr_index_block_find(struct inode *inode,
261 struct buffer_head *root_bh,
264 struct ocfs2_xattr_search *xs);
266 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
267 struct buffer_head *blk_bh,
271 static int ocfs2_xattr_create_index_block(struct inode *inode,
272 struct ocfs2_xattr_search *xs,
273 struct ocfs2_xattr_set_ctxt *ctxt);
275 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
276 struct ocfs2_xattr_info *xi,
277 struct ocfs2_xattr_search *xs,
278 struct ocfs2_xattr_set_ctxt *ctxt);
280 typedef int (xattr_tree_rec_func)(struct inode *inode,
281 struct buffer_head *root_bh,
282 u64 blkno, u32 cpos, u32 len, void *para);
283 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
284 struct buffer_head *root_bh,
285 xattr_tree_rec_func *rec_func,
287 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
288 struct ocfs2_xattr_bucket *bucket,
290 static int ocfs2_rm_xattr_cluster(struct inode *inode,
291 struct buffer_head *root_bh,
297 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
298 u64 src_blk, u64 last_blk, u64 to_blk,
299 unsigned int start_bucket,
301 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
302 struct ocfs2_dinode *di,
303 struct ocfs2_xattr_info *xi,
304 struct ocfs2_xattr_search *xis,
305 struct ocfs2_xattr_search *xbs,
306 struct ocfs2_refcount_tree **ref_tree,
309 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
310 struct ocfs2_xattr_bucket *bucket,
312 struct ocfs2_xattr_value_root **xv,
313 struct buffer_head **bh);
315 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
317 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
320 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
322 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
325 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
326 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
327 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
329 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
331 struct ocfs2_xattr_bucket *bucket;
332 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
334 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
336 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
338 bucket->bu_inode = inode;
339 bucket->bu_blocks = blks;
345 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
349 for (i = 0; i < bucket->bu_blocks; i++) {
350 brelse(bucket->bu_bhs[i]);
351 bucket->bu_bhs[i] = NULL;
355 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
358 ocfs2_xattr_bucket_relse(bucket);
359 bucket->bu_inode = NULL;
365 * A bucket that has never been written to disk doesn't need to be
366 * read. We just need the buffer_heads. Don't call this for
367 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
370 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
375 for (i = 0; i < bucket->bu_blocks; i++) {
376 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
378 if (!bucket->bu_bhs[i]) {
384 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
386 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
391 ocfs2_xattr_bucket_relse(bucket);
395 /* Read the xattr bucket at xb_blkno */
396 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
401 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
402 bucket->bu_blocks, bucket->bu_bhs, 0,
405 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
406 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
409 &bucket_xh(bucket)->xh_check);
410 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
416 ocfs2_xattr_bucket_relse(bucket);
420 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
421 struct ocfs2_xattr_bucket *bucket,
426 for (i = 0; i < bucket->bu_blocks; i++) {
427 rc = ocfs2_journal_access(handle,
428 INODE_CACHE(bucket->bu_inode),
429 bucket->bu_bhs[i], type);
439 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
440 struct ocfs2_xattr_bucket *bucket)
444 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
445 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
446 bucket->bu_bhs, bucket->bu_blocks,
447 &bucket_xh(bucket)->xh_check);
448 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
450 for (i = 0; i < bucket->bu_blocks; i++)
451 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
454 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
455 struct ocfs2_xattr_bucket *src)
458 int blocksize = src->bu_inode->i_sb->s_blocksize;
460 BUG_ON(dest->bu_blocks != src->bu_blocks);
461 BUG_ON(dest->bu_inode != src->bu_inode);
463 for (i = 0; i < src->bu_blocks; i++) {
464 memcpy(bucket_block(dest, i), bucket_block(src, i),
469 static int ocfs2_validate_xattr_block(struct super_block *sb,
470 struct buffer_head *bh)
473 struct ocfs2_xattr_block *xb =
474 (struct ocfs2_xattr_block *)bh->b_data;
476 mlog(0, "Validating xattr block %llu\n",
477 (unsigned long long)bh->b_blocknr);
479 BUG_ON(!buffer_uptodate(bh));
482 * If the ecc fails, we return the error but otherwise
483 * leave the filesystem running. We know any error is
484 * local to this block.
486 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
491 * Errors after here are fatal
494 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
496 "Extended attribute block #%llu has bad "
498 (unsigned long long)bh->b_blocknr, 7,
503 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
505 "Extended attribute block #%llu has an "
506 "invalid xb_blkno of %llu",
507 (unsigned long long)bh->b_blocknr,
508 (unsigned long long)le64_to_cpu(xb->xb_blkno));
512 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
514 "Extended attribute block #%llu has an invalid "
515 "xb_fs_generation of #%u",
516 (unsigned long long)bh->b_blocknr,
517 le32_to_cpu(xb->xb_fs_generation));
524 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
525 struct buffer_head **bh)
528 struct buffer_head *tmp = *bh;
530 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
531 ocfs2_validate_xattr_block);
533 /* If ocfs2_read_block() got us a new bh, pass it up. */
540 static inline const char *ocfs2_xattr_prefix(int name_index)
542 struct xattr_handler *handler = NULL;
544 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
545 handler = ocfs2_xattr_handler_map[name_index];
547 return handler ? handler->prefix : NULL;
550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
554 /* Get hash value of uuid from super block */
555 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
558 /* hash extended attribute name */
559 for (i = 0; i < name_len; i++) {
560 hash = (hash << OCFS2_HASH_SHIFT) ^
561 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
570 return namevalue_size(name_len, value_len) +
571 sizeof(struct ocfs2_xattr_entry);
574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
576 return namevalue_size_xi(xi) +
577 sizeof(struct ocfs2_xattr_entry);
580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
582 return namevalue_size_xe(xe) +
583 sizeof(struct ocfs2_xattr_entry);
586 int ocfs2_calc_security_init(struct inode *dir,
587 struct ocfs2_security_xattr_info *si,
590 struct ocfs2_alloc_context **xattr_ac)
593 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
594 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
598 * The max space of security xattr taken inline is
599 * 256(name) + 80(value) + 16(entry) = 352 bytes,
600 * So reserve one metadata block for it is ok.
602 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
603 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
604 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
609 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
612 /* reserve clusters for xattr value which will be set in B tree*/
613 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
614 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
617 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
619 *want_clusters += new_clusters;
624 int ocfs2_calc_xattr_init(struct inode *dir,
625 struct buffer_head *dir_bh,
627 struct ocfs2_security_xattr_info *si,
633 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
634 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
637 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
640 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
641 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
642 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
645 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
648 } else if (acl_len != 0 && acl_len != -ENODATA) {
654 if (!(s_size + a_size))
658 * The max space of security xattr taken inline is
659 * 256(name) + 80(value) + 16(entry) = 352 bytes,
660 * The max space of acl xattr taken inline is
661 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
662 * when blocksize = 512, may reserve one more cluser for
663 * xattr bucket, otherwise reserve one metadata block
665 * If this is a new directory with inline data,
666 * we choose to reserve the entire inline area for
667 * directory contents and force an external xattr block.
669 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
670 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
671 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
672 *want_meta = *want_meta + 1;
673 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
676 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
677 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
679 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
683 * reserve credits and clusters for xattrs which has large value
684 * and have to be set outside
686 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
687 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
689 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
691 *want_clusters += new_clusters;
693 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
694 acl_len > OCFS2_XATTR_INLINE_SIZE) {
695 /* for directory, it has DEFAULT and ACCESS two types of acls */
696 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
697 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
698 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
700 *want_clusters += new_clusters;
706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
708 struct ocfs2_xattr_value_buf *vb,
709 struct ocfs2_xattr_set_ctxt *ctxt)
712 handle_t *handle = ctxt->handle;
713 enum ocfs2_alloc_restarted why;
714 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
715 struct ocfs2_extent_tree et;
717 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
719 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
721 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
722 OCFS2_JOURNAL_ACCESS_WRITE);
728 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
729 status = ocfs2_add_clusters_in_btree(handle,
742 status = ocfs2_journal_dirty(handle, vb->vb_bh);
748 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
751 * We should have already allocated enough space before the transaction,
752 * so no need to restart.
754 BUG_ON(why != RESTART_NONE || clusters_to_add);
761 static int __ocfs2_remove_xattr_range(struct inode *inode,
762 struct ocfs2_xattr_value_buf *vb,
763 u32 cpos, u32 phys_cpos, u32 len,
764 unsigned int ext_flags,
765 struct ocfs2_xattr_set_ctxt *ctxt)
768 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
769 handle_t *handle = ctxt->handle;
770 struct ocfs2_extent_tree et;
772 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
774 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
775 OCFS2_JOURNAL_ACCESS_WRITE);
781 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
788 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
790 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
796 if (ext_flags & OCFS2_EXT_REFCOUNTED)
797 ret = ocfs2_decrease_refcount(inode, handle,
798 ocfs2_blocks_to_clusters(inode->i_sb,
800 len, ctxt->meta_ac, &ctxt->dealloc, 1);
802 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
811 static int ocfs2_xattr_shrink_size(struct inode *inode,
814 struct ocfs2_xattr_value_buf *vb,
815 struct ocfs2_xattr_set_ctxt *ctxt)
818 unsigned int ext_flags;
819 u32 trunc_len, cpos, phys_cpos, alloc_size;
822 if (old_clusters <= new_clusters)
826 trunc_len = old_clusters - new_clusters;
828 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
830 &vb->vb_xv->xr_list, &ext_flags);
836 if (alloc_size > trunc_len)
837 alloc_size = trunc_len;
839 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
840 phys_cpos, alloc_size,
847 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
848 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
851 trunc_len -= alloc_size;
858 static int ocfs2_xattr_value_truncate(struct inode *inode,
859 struct ocfs2_xattr_value_buf *vb,
861 struct ocfs2_xattr_set_ctxt *ctxt)
864 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
865 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
867 if (new_clusters == old_clusters)
870 if (new_clusters > old_clusters)
871 ret = ocfs2_xattr_extend_allocation(inode,
872 new_clusters - old_clusters,
875 ret = ocfs2_xattr_shrink_size(inode,
876 old_clusters, new_clusters,
882 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
883 size_t *result, const char *prefix,
884 const char *name, int name_len)
886 char *p = buffer + *result;
887 int prefix_len = strlen(prefix);
888 int total_len = prefix_len + name_len + 1;
890 *result += total_len;
892 /* we are just looking for how big our buffer needs to be */
899 memcpy(p, prefix, prefix_len);
900 memcpy(p + prefix_len, name, name_len);
901 p[prefix_len + name_len] = '\0';
906 static int ocfs2_xattr_list_entries(struct inode *inode,
907 struct ocfs2_xattr_header *header,
908 char *buffer, size_t buffer_size)
912 const char *prefix, *name;
914 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
915 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
916 type = ocfs2_xattr_get_type(entry);
917 prefix = ocfs2_xattr_prefix(type);
920 name = (const char *)header +
921 le16_to_cpu(entry->xe_name_offset);
923 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
924 &result, prefix, name,
934 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
935 struct ocfs2_dinode *di)
937 struct ocfs2_xattr_header *xh;
940 xh = (struct ocfs2_xattr_header *)
941 ((void *)di + inode->i_sb->s_blocksize -
942 le16_to_cpu(di->i_xattr_inline_size));
944 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
945 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
951 static int ocfs2_xattr_ibody_list(struct inode *inode,
952 struct ocfs2_dinode *di,
956 struct ocfs2_xattr_header *header = NULL;
957 struct ocfs2_inode_info *oi = OCFS2_I(inode);
960 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
963 header = (struct ocfs2_xattr_header *)
964 ((void *)di + inode->i_sb->s_blocksize -
965 le16_to_cpu(di->i_xattr_inline_size));
967 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
972 static int ocfs2_xattr_block_list(struct inode *inode,
973 struct ocfs2_dinode *di,
977 struct buffer_head *blk_bh = NULL;
978 struct ocfs2_xattr_block *xb;
981 if (!di->i_xattr_loc)
984 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
991 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
992 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
993 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
994 ret = ocfs2_xattr_list_entries(inode, header,
995 buffer, buffer_size);
997 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
998 buffer, buffer_size);
1005 ssize_t ocfs2_listxattr(struct dentry *dentry,
1009 int ret = 0, i_ret = 0, b_ret = 0;
1010 struct buffer_head *di_bh = NULL;
1011 struct ocfs2_dinode *di = NULL;
1012 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1014 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1017 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1020 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1026 di = (struct ocfs2_dinode *)di_bh->b_data;
1028 down_read(&oi->ip_xattr_sem);
1029 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1037 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1042 up_read(&oi->ip_xattr_sem);
1043 ocfs2_inode_unlock(dentry->d_inode, 0);
1047 return i_ret + b_ret;
1050 static int ocfs2_xattr_find_entry(int name_index,
1052 struct ocfs2_xattr_search *xs)
1054 struct ocfs2_xattr_entry *entry;
1061 name_len = strlen(name);
1063 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1064 cmp = name_index - ocfs2_xattr_get_type(entry);
1066 cmp = name_len - entry->xe_name_len;
1068 cmp = memcmp(name, (xs->base +
1069 le16_to_cpu(entry->xe_name_offset)),
1077 return cmp ? -ENODATA : 0;
1080 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1081 struct ocfs2_xattr_value_root *xv,
1085 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1088 size_t cplen, blocksize;
1089 struct buffer_head *bh = NULL;
1090 struct ocfs2_extent_list *el;
1093 clusters = le32_to_cpu(xv->xr_clusters);
1094 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1095 blocksize = inode->i_sb->s_blocksize;
1098 while (cpos < clusters) {
1099 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1100 &num_clusters, el, NULL);
1106 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1107 /* Copy ocfs2_xattr_value */
1108 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1109 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1116 cplen = len >= blocksize ? blocksize : len;
1117 memcpy(buffer, bh->b_data, cplen);
1126 cpos += num_clusters;
1132 static int ocfs2_xattr_ibody_get(struct inode *inode,
1137 struct ocfs2_xattr_search *xs)
1139 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1140 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1141 struct ocfs2_xattr_value_root *xv;
1145 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1148 xs->end = (void *)di + inode->i_sb->s_blocksize;
1149 xs->header = (struct ocfs2_xattr_header *)
1150 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1151 xs->base = (void *)xs->header;
1152 xs->here = xs->header->xh_entries;
1154 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1157 size = le64_to_cpu(xs->here->xe_value_size);
1159 if (size > buffer_size)
1161 if (ocfs2_xattr_is_local(xs->here)) {
1162 memcpy(buffer, (void *)xs->base +
1163 le16_to_cpu(xs->here->xe_name_offset) +
1164 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1166 xv = (struct ocfs2_xattr_value_root *)
1167 (xs->base + le16_to_cpu(
1168 xs->here->xe_name_offset) +
1169 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1170 ret = ocfs2_xattr_get_value_outside(inode, xv,
1182 static int ocfs2_xattr_block_get(struct inode *inode,
1187 struct ocfs2_xattr_search *xs)
1189 struct ocfs2_xattr_block *xb;
1190 struct ocfs2_xattr_value_root *xv;
1192 int ret = -ENODATA, name_offset, name_len, i;
1193 int uninitialized_var(block_off);
1195 xs->bucket = ocfs2_xattr_bucket_new(inode);
1202 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1208 if (xs->not_found) {
1213 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1214 size = le64_to_cpu(xs->here->xe_value_size);
1217 if (size > buffer_size)
1220 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1221 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1222 i = xs->here - xs->header->xh_entries;
1224 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1225 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1226 bucket_xh(xs->bucket),
1230 xs->base = bucket_block(xs->bucket, block_off);
1232 if (ocfs2_xattr_is_local(xs->here)) {
1233 memcpy(buffer, (void *)xs->base +
1234 name_offset + name_len, size);
1236 xv = (struct ocfs2_xattr_value_root *)
1237 (xs->base + name_offset + name_len);
1238 ret = ocfs2_xattr_get_value_outside(inode, xv,
1248 ocfs2_xattr_bucket_free(xs->bucket);
1250 brelse(xs->xattr_bh);
1251 xs->xattr_bh = NULL;
1255 int ocfs2_xattr_get_nolock(struct inode *inode,
1256 struct buffer_head *di_bh,
1263 struct ocfs2_dinode *di = NULL;
1264 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1265 struct ocfs2_xattr_search xis = {
1266 .not_found = -ENODATA,
1268 struct ocfs2_xattr_search xbs = {
1269 .not_found = -ENODATA,
1272 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1275 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1278 xis.inode_bh = xbs.inode_bh = di_bh;
1279 di = (struct ocfs2_dinode *)di_bh->b_data;
1281 down_read(&oi->ip_xattr_sem);
1282 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1284 if (ret == -ENODATA && di->i_xattr_loc)
1285 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1287 up_read(&oi->ip_xattr_sem);
1292 /* ocfs2_xattr_get()
1294 * Copy an extended attribute into the buffer provided.
1295 * Buffer is NULL to compute the size of buffer required.
1297 static int ocfs2_xattr_get(struct inode *inode,
1304 struct buffer_head *di_bh = NULL;
1306 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1311 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1312 name, buffer, buffer_size);
1314 ocfs2_inode_unlock(inode, 0);
1321 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1323 struct ocfs2_xattr_value_buf *vb,
1327 int ret = 0, i, cp_len;
1328 u16 blocksize = inode->i_sb->s_blocksize;
1329 u32 p_cluster, num_clusters;
1330 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1331 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1333 struct buffer_head *bh = NULL;
1334 unsigned int ext_flags;
1335 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1337 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1339 while (cpos < clusters) {
1340 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1341 &num_clusters, &xv->xr_list,
1348 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1350 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1352 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1353 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1360 ret = ocfs2_journal_access(handle,
1363 OCFS2_JOURNAL_ACCESS_WRITE);
1369 cp_len = value_len > blocksize ? blocksize : value_len;
1370 memcpy(bh->b_data, value, cp_len);
1371 value_len -= cp_len;
1373 if (cp_len < blocksize)
1374 memset(bh->b_data + cp_len, 0,
1375 blocksize - cp_len);
1377 ret = ocfs2_journal_dirty(handle, bh);
1386 * XXX: do we need to empty all the following
1387 * blocks in this cluster?
1392 cpos += num_clusters;
1400 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1408 free_space = free_start -
1409 sizeof(struct ocfs2_xattr_header) -
1410 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1411 OCFS2_XATTR_HEADER_GAP;
1414 if (free_space < needed_space)
1420 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1423 return loc->xl_ops->xlo_journal_access(handle, loc, type);
1426 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1428 loc->xl_ops->xlo_journal_dirty(handle, loc);
1431 /* Give a pointer into the storage for the given offset */
1432 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1434 BUG_ON(offset >= loc->xl_size);
1435 return loc->xl_ops->xlo_offset_pointer(loc, offset);
1439 * Wipe the name+value pair and allow the storage to reclaim it. This
1440 * must be followed by either removal of the entry or a call to
1441 * ocfs2_xa_add_namevalue().
1443 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1445 loc->xl_ops->xlo_wipe_namevalue(loc);
1449 * Find lowest offset to a name+value pair. This is the start of our
1450 * downward-growing free space.
1452 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1454 return loc->xl_ops->xlo_get_free_start(loc);
1457 /* Can we reuse loc->xl_entry for xi? */
1458 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1459 struct ocfs2_xattr_info *xi)
1461 return loc->xl_ops->xlo_can_reuse(loc, xi);
1464 /* How much free space is needed to set the new value */
1465 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1466 struct ocfs2_xattr_info *xi)
1468 return loc->xl_ops->xlo_check_space(loc, xi);
1471 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1473 loc->xl_ops->xlo_add_entry(loc, name_hash);
1474 loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1476 * We can't leave the new entry's xe_name_offset at zero or
1477 * add_namevalue() will go nuts. We set it to the size of our
1478 * storage so that it can never be less than any other entry.
1480 loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1483 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1484 struct ocfs2_xattr_info *xi)
1486 int size = namevalue_size_xi(xi);
1490 loc->xl_ops->xlo_add_namevalue(loc, size);
1491 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1492 loc->xl_entry->xe_name_len = xi->xi_name_len;
1493 ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1494 ocfs2_xattr_set_local(loc->xl_entry,
1495 xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1497 nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1498 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1499 memset(nameval_buf, 0, size);
1500 memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1503 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1504 struct ocfs2_xattr_value_buf *vb)
1506 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1507 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1509 /* Value bufs are for value trees */
1510 BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1511 BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1512 (name_size + OCFS2_XATTR_ROOT_SIZE));
1514 loc->xl_ops->xlo_fill_value_buf(loc, vb);
1516 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1521 static int ocfs2_xa_block_journal_access(handle_t *handle,
1522 struct ocfs2_xa_loc *loc, int type)
1524 struct buffer_head *bh = loc->xl_storage;
1525 ocfs2_journal_access_func access;
1527 if (loc->xl_size == (bh->b_size -
1528 offsetof(struct ocfs2_xattr_block,
1529 xb_attrs.xb_header)))
1530 access = ocfs2_journal_access_xb;
1532 access = ocfs2_journal_access_di;
1533 return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1536 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1537 struct ocfs2_xa_loc *loc)
1539 struct buffer_head *bh = loc->xl_storage;
1541 ocfs2_journal_dirty(handle, bh);
1544 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1547 return (char *)loc->xl_header + offset;
1550 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1551 struct ocfs2_xattr_info *xi)
1554 * Block storage is strict. If the sizes aren't exact, we will
1555 * remove the old one and reinsert the new.
1557 return namevalue_size_xe(loc->xl_entry) ==
1558 namevalue_size_xi(xi);
1561 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1563 struct ocfs2_xattr_header *xh = loc->xl_header;
1564 int i, count = le16_to_cpu(xh->xh_count);
1565 int offset, free_start = loc->xl_size;
1567 for (i = 0; i < count; i++) {
1568 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1569 if (offset < free_start)
1570 free_start = offset;
1576 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1577 struct ocfs2_xattr_info *xi)
1579 int count = le16_to_cpu(loc->xl_header->xh_count);
1580 int free_start = ocfs2_xa_get_free_start(loc);
1581 int needed_space = ocfs2_xi_entry_usage(xi);
1584 * Block storage will reclaim the original entry before inserting
1585 * the new value, so we only need the difference. If the new
1586 * entry is smaller than the old one, we don't need anything.
1588 if (loc->xl_entry) {
1589 /* Don't need space if we're reusing! */
1590 if (ocfs2_xa_can_reuse_entry(loc, xi))
1593 needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1595 if (needed_space < 0)
1597 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1601 * Block storage for xattrs keeps the name+value pairs compacted. When
1602 * we remove one, we have to shift any that preceded it towards the end.
1604 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1607 int namevalue_offset, first_namevalue_offset, namevalue_size;
1608 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1609 struct ocfs2_xattr_header *xh = loc->xl_header;
1610 int count = le16_to_cpu(xh->xh_count);
1612 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1613 namevalue_size = namevalue_size_xe(entry);
1614 first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1616 /* Shift the name+value pairs */
1617 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1618 (char *)xh + first_namevalue_offset,
1619 namevalue_offset - first_namevalue_offset);
1620 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1622 /* Now tell xh->xh_entries about it */
1623 for (i = 0; i < count; i++) {
1624 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1625 if (offset < namevalue_offset)
1626 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1631 * Note that we don't update xh_free_start or xh_name_value_len
1632 * because they're not used in block-stored xattrs.
1636 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1638 int count = le16_to_cpu(loc->xl_header->xh_count);
1639 loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1640 le16_add_cpu(&loc->xl_header->xh_count, 1);
1641 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1644 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1646 int free_start = ocfs2_xa_get_free_start(loc);
1648 loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1651 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1652 struct ocfs2_xattr_value_buf *vb)
1654 struct buffer_head *bh = loc->xl_storage;
1656 if (loc->xl_size == (bh->b_size -
1657 offsetof(struct ocfs2_xattr_block,
1658 xb_attrs.xb_header)))
1659 vb->vb_access = ocfs2_journal_access_xb;
1661 vb->vb_access = ocfs2_journal_access_di;
1666 * Operations for xattrs stored in blocks. This includes inline inode
1667 * storage and unindexed ocfs2_xattr_blocks.
1669 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1670 .xlo_journal_access = ocfs2_xa_block_journal_access,
1671 .xlo_journal_dirty = ocfs2_xa_block_journal_dirty,
1672 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1673 .xlo_check_space = ocfs2_xa_block_check_space,
1674 .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1675 .xlo_get_free_start = ocfs2_xa_block_get_free_start,
1676 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1677 .xlo_add_entry = ocfs2_xa_block_add_entry,
1678 .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
1679 .xlo_fill_value_buf = ocfs2_xa_block_fill_value_buf,
1682 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1683 struct ocfs2_xa_loc *loc, int type)
1685 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1687 return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1690 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1691 struct ocfs2_xa_loc *loc)
1693 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1695 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1698 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1701 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1702 int block, block_offset;
1704 /* The header is at the front of the bucket */
1705 block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1706 block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1708 return bucket_block(bucket, block) + block_offset;
1711 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1712 struct ocfs2_xattr_info *xi)
1714 return namevalue_size_xe(loc->xl_entry) >=
1715 namevalue_size_xi(xi);
1718 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1720 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1721 return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1724 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1725 int free_start, int size)
1728 * We need to make sure that the name+value pair fits within
1731 if (((free_start - size) >> sb->s_blocksize_bits) !=
1732 ((free_start - 1) >> sb->s_blocksize_bits))
1733 free_start -= free_start % sb->s_blocksize;
1738 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1739 struct ocfs2_xattr_info *xi)
1742 int count = le16_to_cpu(loc->xl_header->xh_count);
1743 int free_start = ocfs2_xa_get_free_start(loc);
1744 int needed_space = ocfs2_xi_entry_usage(xi);
1745 int size = namevalue_size_xi(xi);
1746 struct super_block *sb = loc->xl_inode->i_sb;
1749 * Bucket storage does not reclaim name+value pairs it cannot
1750 * reuse. They live as holes until the bucket fills, and then
1751 * the bucket is defragmented. However, the bucket can reclaim
1752 * the ocfs2_xattr_entry.
1754 if (loc->xl_entry) {
1755 /* Don't need space if we're reusing! */
1756 if (ocfs2_xa_can_reuse_entry(loc, xi))
1759 needed_space -= sizeof(struct ocfs2_xattr_entry);
1761 BUG_ON(needed_space < 0);
1763 if (free_start < size) {
1768 * First we check if it would fit in the first place.
1769 * Below, we align the free start to a block. This may
1770 * slide us below the minimum gap. By checking unaligned
1771 * first, we avoid that error.
1773 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1777 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1780 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1783 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1785 le16_add_cpu(&loc->xl_header->xh_name_value_len,
1786 -namevalue_size_xe(loc->xl_entry));
1789 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1791 struct ocfs2_xattr_header *xh = loc->xl_header;
1792 int count = le16_to_cpu(xh->xh_count);
1793 int low = 0, high = count - 1, tmp;
1794 struct ocfs2_xattr_entry *tmp_xe;
1797 * We keep buckets sorted by name_hash, so we need to find
1800 while (low <= high && count) {
1801 tmp = (low + high) / 2;
1802 tmp_xe = &xh->xh_entries[tmp];
1804 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1806 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1815 memmove(&xh->xh_entries[low + 1],
1816 &xh->xh_entries[low],
1817 ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1819 le16_add_cpu(&xh->xh_count, 1);
1820 loc->xl_entry = &xh->xh_entries[low];
1821 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1824 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1826 int free_start = ocfs2_xa_get_free_start(loc);
1827 struct ocfs2_xattr_header *xh = loc->xl_header;
1828 struct super_block *sb = loc->xl_inode->i_sb;
1831 free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1832 nameval_offset = free_start - size;
1833 loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1834 xh->xh_free_start = cpu_to_le16(nameval_offset);
1835 le16_add_cpu(&xh->xh_name_value_len, size);
1839 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1840 struct ocfs2_xattr_value_buf *vb)
1842 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1843 struct super_block *sb = loc->xl_inode->i_sb;
1844 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1845 int size = namevalue_size_xe(loc->xl_entry);
1846 int block_offset = nameval_offset >> sb->s_blocksize_bits;
1848 /* Values are not allowed to straddle block boundaries */
1849 BUG_ON(block_offset !=
1850 ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1851 /* We expect the bucket to be filled in */
1852 BUG_ON(!bucket->bu_bhs[block_offset]);
1854 vb->vb_access = ocfs2_journal_access;
1855 vb->vb_bh = bucket->bu_bhs[block_offset];
1858 /* Operations for xattrs stored in buckets. */
1859 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1860 .xlo_journal_access = ocfs2_xa_bucket_journal_access,
1861 .xlo_journal_dirty = ocfs2_xa_bucket_journal_dirty,
1862 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1863 .xlo_check_space = ocfs2_xa_bucket_check_space,
1864 .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1865 .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
1866 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1867 .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1868 .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
1869 .xlo_fill_value_buf = ocfs2_xa_bucket_fill_value_buf,
1872 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1873 struct ocfs2_xattr_set_ctxt *ctxt)
1875 int trunc_rc, access_rc;
1876 struct ocfs2_xattr_value_buf vb;
1878 ocfs2_xa_fill_value_buf(loc, &vb);
1879 trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1883 * The caller of ocfs2_xa_value_truncate() has already called
1884 * ocfs2_xa_journal_access on the loc. However, The truncate code
1885 * calls ocfs2_extend_trans(). This may commit the previous
1886 * transaction and open a new one. If this is a bucket, truncate
1887 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1888 * the caller is expecting to dirty the entire bucket. So we must
1889 * reset the journal work. We do this even if truncate has failed,
1890 * as it could have failed after committing the extend.
1892 access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1893 OCFS2_JOURNAL_ACCESS_WRITE);
1895 /* Errors in truncate take precedence */
1896 return trunc_rc ? trunc_rc : access_rc;
1899 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1902 struct ocfs2_xattr_header *xh = loc->xl_header;
1903 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1905 ocfs2_xa_wipe_namevalue(loc);
1906 loc->xl_entry = NULL;
1908 le16_add_cpu(&xh->xh_count, -1);
1909 count = le16_to_cpu(xh->xh_count);
1912 * Only zero out the entry if there are more remaining. This is
1913 * important for an empty bucket, as it keeps track of the
1914 * bucket's hash value. It doesn't hurt empty block storage.
1917 index = ((char *)entry - (char *)&xh->xh_entries) /
1918 sizeof(struct ocfs2_xattr_entry);
1919 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1920 (count - index) * sizeof(struct ocfs2_xattr_entry));
1921 memset(&xh->xh_entries[count], 0,
1922 sizeof(struct ocfs2_xattr_entry));
1926 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1927 struct ocfs2_xattr_set_ctxt *ctxt)
1931 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
1932 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
1939 ocfs2_xa_remove_entry(loc);
1945 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
1947 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1950 nameval_buf = ocfs2_xa_offset_pointer(loc,
1951 le16_to_cpu(loc->xl_entry->xe_name_offset));
1952 memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
1956 * Take an existing entry and make it ready for the new value. This
1957 * won't allocate space, but it may free space. It should be ready for
1958 * ocfs2_xa_prepare_entry() to finish the work.
1960 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
1961 struct ocfs2_xattr_info *xi,
1962 struct ocfs2_xattr_set_ctxt *ctxt)
1965 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
1967 int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
1968 int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
1970 BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
1973 nameval_buf = ocfs2_xa_offset_pointer(loc,
1974 le16_to_cpu(loc->xl_entry->xe_name_offset));
1976 memset(nameval_buf + name_size, 0,
1977 namevalue_size_xe(loc->xl_entry) - name_size);
1979 ocfs2_xa_install_value_root(loc);
1982 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
1987 memset(nameval_buf + name_size, 0,
1988 namevalue_size_xe(loc->xl_entry) -
1990 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
1992 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2001 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2002 ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2009 * Prepares loc->xl_entry to receive the new xattr. This includes
2010 * properly setting up the name+value pair region. If loc->xl_entry
2011 * already exists, it will take care of modifying it appropriately.
2013 * Note that this modifies the data. You did journal_access already,
2016 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2017 struct ocfs2_xattr_info *xi,
2019 struct ocfs2_xattr_set_ctxt *ctxt)
2023 rc = ocfs2_xa_check_space(loc, xi);
2027 if (loc->xl_entry) {
2028 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2029 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2035 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2036 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2042 ocfs2_xa_wipe_namevalue(loc);
2044 ocfs2_xa_add_entry(loc, name_hash);
2047 * If we get here, we have a blank entry. Fill it. We grow our
2048 * name+value pair back from the end.
2050 ocfs2_xa_add_namevalue(loc, xi);
2051 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2052 ocfs2_xa_install_value_root(loc);
2055 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2056 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2066 * Store the value portion of the name+value pair. This will skip
2067 * values that are stored externally. Their tree roots were set up
2068 * by ocfs2_xa_prepare_entry().
2070 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2071 struct ocfs2_xattr_info *xi,
2072 struct ocfs2_xattr_set_ctxt *ctxt)
2075 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2076 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2078 struct ocfs2_xattr_value_buf vb;
2080 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2081 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2082 ocfs2_xa_fill_value_buf(loc, &vb);
2083 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2088 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2093 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2094 struct ocfs2_xattr_info *xi,
2095 struct ocfs2_xattr_set_ctxt *ctxt)
2098 u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2101 ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2102 OCFS2_JOURNAL_ACCESS_WRITE);
2108 /* Don't worry, we are never called with !xi_value and !xl_entry */
2109 if (!xi->xi_value) {
2110 ret = ocfs2_xa_remove(loc, ctxt);
2114 ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2121 ret = ocfs2_xa_store_value(loc, xi, ctxt);
2127 ocfs2_xa_journal_dirty(ctxt->handle, loc);
2133 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2134 struct inode *inode,
2135 struct buffer_head *bh,
2136 struct ocfs2_xattr_entry *entry)
2138 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2140 loc->xl_inode = inode;
2141 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2142 loc->xl_storage = bh;
2143 loc->xl_entry = entry;
2145 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2146 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2149 loc->xl_size = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2152 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2156 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2157 struct inode *inode,
2158 struct buffer_head *bh,
2159 struct ocfs2_xattr_entry *entry)
2161 struct ocfs2_xattr_block *xb =
2162 (struct ocfs2_xattr_block *)bh->b_data;
2164 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2166 loc->xl_inode = inode;
2167 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2168 loc->xl_storage = bh;
2169 loc->xl_header = &(xb->xb_attrs.xb_header);
2170 loc->xl_entry = entry;
2171 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2172 xb_attrs.xb_header);
2175 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2176 struct ocfs2_xattr_bucket *bucket,
2177 struct ocfs2_xattr_entry *entry)
2179 loc->xl_inode = bucket->bu_inode;
2180 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2181 loc->xl_storage = bucket;
2182 loc->xl_header = bucket_xh(bucket);
2183 loc->xl_entry = entry;
2184 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2189 * ocfs2_xattr_set_entry()
2191 * Set extended attribute entry into inode or block.
2193 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
2194 * We first insert tree root(ocfs2_xattr_value_root) like a normal value,
2195 * then set value in B tree with set_value_outside().
2197 static int ocfs2_xattr_set_entry(struct inode *inode,
2198 struct ocfs2_xattr_info *xi,
2199 struct ocfs2_xattr_search *xs,
2200 struct ocfs2_xattr_set_ctxt *ctxt,
2203 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2204 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2205 handle_t *handle = ctxt->handle;
2207 struct ocfs2_xa_loc loc;
2209 if (!(flag & OCFS2_INLINE_XATTR_FL))
2210 BUG_ON(xs->xattr_bh == xs->inode_bh);
2212 BUG_ON(xs->xattr_bh != xs->inode_bh);
2214 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
2215 OCFS2_JOURNAL_ACCESS_WRITE);
2221 if (xs->xattr_bh == xs->inode_bh)
2222 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2223 xs->not_found ? NULL : xs->here);
2225 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2226 xs->not_found ? NULL : xs->here);
2228 ret = ocfs2_xa_set(&loc, xi, ctxt);
2234 xs->here = loc.xl_entry;
2236 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
2237 (flag & OCFS2_INLINE_XATTR_FL)) {
2238 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2239 unsigned int xattrsize = osb->s_xattr_inline_size;
2242 * Adjust extent record count or inline data size
2243 * to reserve space for extended attribute.
2245 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2246 struct ocfs2_inline_data *idata = &di->id2.i_data;
2247 le16_add_cpu(&idata->id_count, -xattrsize);
2248 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2249 struct ocfs2_extent_list *el = &di->id2.i_list;
2250 le16_add_cpu(&el->l_count, -(xattrsize /
2251 sizeof(struct ocfs2_extent_rec)));
2253 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2255 /* Update xattr flag */
2256 spin_lock(&oi->ip_lock);
2257 oi->ip_dyn_features |= flag;
2258 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2259 spin_unlock(&oi->ip_lock);
2261 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
2270 * In xattr remove, if it is stored outside and refcounted, we may have
2271 * the chance to split the refcount tree. So need the allocators.
2273 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2274 struct ocfs2_xattr_value_root *xv,
2275 struct ocfs2_caching_info *ref_ci,
2276 struct buffer_head *ref_root_bh,
2277 struct ocfs2_alloc_context **meta_ac,
2280 int ret, meta_add = 0;
2281 u32 p_cluster, num_clusters;
2282 unsigned int ext_flags;
2285 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2294 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2297 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2299 &meta_add, ref_credits);
2305 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2314 static int ocfs2_remove_value_outside(struct inode*inode,
2315 struct ocfs2_xattr_value_buf *vb,
2316 struct ocfs2_xattr_header *header,
2317 struct ocfs2_caching_info *ref_ci,
2318 struct buffer_head *ref_root_bh)
2320 int ret = 0, i, ref_credits;
2321 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2322 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2325 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2327 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2328 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2330 if (ocfs2_xattr_is_local(entry))
2333 val = (void *)header +
2334 le16_to_cpu(entry->xe_name_offset);
2335 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2336 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2338 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2339 ref_ci, ref_root_bh,
2343 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2344 ocfs2_remove_extent_credits(osb->sb));
2345 if (IS_ERR(ctxt.handle)) {
2346 ret = PTR_ERR(ctxt.handle);
2351 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2357 ocfs2_commit_trans(osb, ctxt.handle);
2359 ocfs2_free_alloc_context(ctxt.meta_ac);
2360 ctxt.meta_ac = NULL;
2365 ocfs2_free_alloc_context(ctxt.meta_ac);
2366 ocfs2_schedule_truncate_log_flush(osb, 1);
2367 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2371 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2372 struct buffer_head *di_bh,
2373 struct ocfs2_caching_info *ref_ci,
2374 struct buffer_head *ref_root_bh)
2377 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2378 struct ocfs2_xattr_header *header;
2380 struct ocfs2_xattr_value_buf vb = {
2382 .vb_access = ocfs2_journal_access_di,
2385 header = (struct ocfs2_xattr_header *)
2386 ((void *)di + inode->i_sb->s_blocksize -
2387 le16_to_cpu(di->i_xattr_inline_size));
2389 ret = ocfs2_remove_value_outside(inode, &vb, header,
2390 ref_ci, ref_root_bh);
2395 struct ocfs2_rm_xattr_bucket_para {
2396 struct ocfs2_caching_info *ref_ci;
2397 struct buffer_head *ref_root_bh;
2400 static int ocfs2_xattr_block_remove(struct inode *inode,
2401 struct buffer_head *blk_bh,
2402 struct ocfs2_caching_info *ref_ci,
2403 struct buffer_head *ref_root_bh)
2405 struct ocfs2_xattr_block *xb;
2407 struct ocfs2_xattr_value_buf vb = {
2409 .vb_access = ocfs2_journal_access_xb,
2411 struct ocfs2_rm_xattr_bucket_para args = {
2413 .ref_root_bh = ref_root_bh,
2416 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2417 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2418 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2419 ret = ocfs2_remove_value_outside(inode, &vb, header,
2420 ref_ci, ref_root_bh);
2422 ret = ocfs2_iterate_xattr_index_block(inode,
2424 ocfs2_rm_xattr_cluster,
2430 static int ocfs2_xattr_free_block(struct inode *inode,
2432 struct ocfs2_caching_info *ref_ci,
2433 struct buffer_head *ref_root_bh)
2435 struct inode *xb_alloc_inode;
2436 struct buffer_head *xb_alloc_bh = NULL;
2437 struct buffer_head *blk_bh = NULL;
2438 struct ocfs2_xattr_block *xb;
2439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2445 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2451 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2457 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2458 blk = le64_to_cpu(xb->xb_blkno);
2459 bit = le16_to_cpu(xb->xb_suballoc_bit);
2460 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2462 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2463 EXTENT_ALLOC_SYSTEM_INODE,
2464 le16_to_cpu(xb->xb_suballoc_slot));
2465 if (!xb_alloc_inode) {
2470 mutex_lock(&xb_alloc_inode->i_mutex);
2472 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2478 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2479 if (IS_ERR(handle)) {
2480 ret = PTR_ERR(handle);
2485 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2490 ocfs2_commit_trans(osb, handle);
2492 ocfs2_inode_unlock(xb_alloc_inode, 1);
2493 brelse(xb_alloc_bh);
2495 mutex_unlock(&xb_alloc_inode->i_mutex);
2496 iput(xb_alloc_inode);
2503 * ocfs2_xattr_remove()
2505 * Free extended attribute resources associated with this inode.
2507 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2509 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2510 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2511 struct ocfs2_refcount_tree *ref_tree = NULL;
2512 struct buffer_head *ref_root_bh = NULL;
2513 struct ocfs2_caching_info *ref_ci = NULL;
2517 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2520 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2523 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2524 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2525 le64_to_cpu(di->i_refcount_loc),
2526 1, &ref_tree, &ref_root_bh);
2531 ref_ci = &ref_tree->rf_ci;
2535 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2536 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2537 ref_ci, ref_root_bh);
2544 if (di->i_xattr_loc) {
2545 ret = ocfs2_xattr_free_block(inode,
2546 le64_to_cpu(di->i_xattr_loc),
2547 ref_ci, ref_root_bh);
2554 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2555 OCFS2_INODE_UPDATE_CREDITS);
2556 if (IS_ERR(handle)) {
2557 ret = PTR_ERR(handle);
2561 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2562 OCFS2_JOURNAL_ACCESS_WRITE);
2568 di->i_xattr_loc = 0;
2570 spin_lock(&oi->ip_lock);
2571 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2572 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2573 spin_unlock(&oi->ip_lock);
2575 ret = ocfs2_journal_dirty(handle, di_bh);
2579 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2582 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2583 brelse(ref_root_bh);
2587 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2588 struct ocfs2_dinode *di)
2590 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2591 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2594 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2597 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2598 struct ocfs2_inline_data *idata = &di->id2.i_data;
2599 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2600 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2601 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2602 le64_to_cpu(di->i_size);
2604 struct ocfs2_extent_list *el = &di->id2.i_list;
2605 free = (le16_to_cpu(el->l_count) -
2606 le16_to_cpu(el->l_next_free_rec)) *
2607 sizeof(struct ocfs2_extent_rec);
2609 if (free >= xattrsize)
2616 * ocfs2_xattr_ibody_find()
2618 * Find extended attribute in inode block and
2619 * fill search info into struct ocfs2_xattr_search.
2621 static int ocfs2_xattr_ibody_find(struct inode *inode,
2624 struct ocfs2_xattr_search *xs)
2626 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2627 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2631 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2634 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2635 down_read(&oi->ip_alloc_sem);
2636 has_space = ocfs2_xattr_has_space_inline(inode, di);
2637 up_read(&oi->ip_alloc_sem);
2642 xs->xattr_bh = xs->inode_bh;
2643 xs->end = (void *)di + inode->i_sb->s_blocksize;
2644 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2645 xs->header = (struct ocfs2_xattr_header *)
2646 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2648 xs->header = (struct ocfs2_xattr_header *)
2649 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2650 xs->base = (void *)xs->header;
2651 xs->here = xs->header->xh_entries;
2653 /* Find the named attribute. */
2654 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2655 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2656 if (ret && ret != -ENODATA)
2658 xs->not_found = ret;
2665 * ocfs2_xattr_ibody_set()
2667 * Set, replace or remove an extended attribute into inode block.
2670 static int ocfs2_xattr_ibody_set(struct inode *inode,
2671 struct ocfs2_xattr_info *xi,
2672 struct ocfs2_xattr_search *xs,
2673 struct ocfs2_xattr_set_ctxt *ctxt)
2675 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2676 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2679 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2682 down_write(&oi->ip_alloc_sem);
2683 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2684 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2690 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2691 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2693 up_write(&oi->ip_alloc_sem);
2699 * ocfs2_xattr_block_find()
2701 * Find extended attribute in external block and
2702 * fill search info into struct ocfs2_xattr_search.
2704 static int ocfs2_xattr_block_find(struct inode *inode,
2707 struct ocfs2_xattr_search *xs)
2709 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2710 struct buffer_head *blk_bh = NULL;
2711 struct ocfs2_xattr_block *xb;
2714 if (!di->i_xattr_loc)
2717 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2724 xs->xattr_bh = blk_bh;
2725 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2727 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2728 xs->header = &xb->xb_attrs.xb_header;
2729 xs->base = (void *)xs->header;
2730 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2731 xs->here = xs->header->xh_entries;
2733 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2735 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2739 if (ret && ret != -ENODATA) {
2740 xs->xattr_bh = NULL;
2743 xs->not_found = ret;
2751 static int ocfs2_create_xattr_block(handle_t *handle,
2752 struct inode *inode,
2753 struct buffer_head *inode_bh,
2754 struct ocfs2_alloc_context *meta_ac,
2755 struct buffer_head **ret_bh,
2759 u16 suballoc_bit_start;
2762 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2763 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2764 struct buffer_head *new_bh = NULL;
2765 struct ocfs2_xattr_block *xblk;
2767 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2768 OCFS2_JOURNAL_ACCESS_CREATE);
2774 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2775 &suballoc_bit_start, &num_got,
2782 new_bh = sb_getblk(inode->i_sb, first_blkno);
2783 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2785 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2787 OCFS2_JOURNAL_ACCESS_CREATE);
2793 /* Initialize ocfs2_xattr_block */
2794 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2795 memset(xblk, 0, inode->i_sb->s_blocksize);
2796 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2797 xblk->xb_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2798 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2799 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2800 xblk->xb_blkno = cpu_to_le64(first_blkno);
2803 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2804 xr->xt_clusters = cpu_to_le32(1);
2805 xr->xt_last_eb_blk = 0;
2806 xr->xt_list.l_tree_depth = 0;
2807 xr->xt_list.l_count = cpu_to_le16(
2808 ocfs2_xattr_recs_per_xb(inode->i_sb));
2809 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2810 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2813 ret = ocfs2_journal_dirty(handle, new_bh);
2818 di->i_xattr_loc = cpu_to_le64(first_blkno);
2819 ocfs2_journal_dirty(handle, inode_bh);
2830 * ocfs2_xattr_block_set()
2832 * Set, replace or remove an extended attribute into external block.
2835 static int ocfs2_xattr_block_set(struct inode *inode,
2836 struct ocfs2_xattr_info *xi,
2837 struct ocfs2_xattr_search *xs,
2838 struct ocfs2_xattr_set_ctxt *ctxt)
2840 struct buffer_head *new_bh = NULL;
2841 handle_t *handle = ctxt->handle;
2842 struct ocfs2_xattr_block *xblk = NULL;
2845 if (!xs->xattr_bh) {
2846 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
2847 ctxt->meta_ac, &new_bh, 0);
2853 xs->xattr_bh = new_bh;
2854 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2855 xs->header = &xblk->xb_attrs.xb_header;
2856 xs->base = (void *)xs->header;
2857 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2858 xs->here = xs->header->xh_entries;
2860 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2862 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2863 /* Set extended attribute into external block */
2864 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2865 OCFS2_HAS_XATTR_FL);
2866 if (!ret || ret != -ENOSPC)
2869 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2874 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2881 /* Check whether the new xattr can be inserted into the inode. */
2882 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2883 struct ocfs2_xattr_info *xi,
2884 struct ocfs2_xattr_search *xs)
2886 struct ocfs2_xattr_entry *last;
2888 size_t min_offs = xs->end - xs->base;
2893 last = xs->header->xh_entries;
2895 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2896 size_t offs = le16_to_cpu(last->xe_name_offset);
2897 if (offs < min_offs)
2902 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2906 BUG_ON(!xs->not_found);
2908 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
2914 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2915 struct ocfs2_dinode *di,
2916 struct ocfs2_xattr_info *xi,
2917 struct ocfs2_xattr_search *xis,
2918 struct ocfs2_xattr_search *xbs,
2923 int ret = 0, old_in_xb = 0;
2924 int clusters_add = 0, meta_add = 0, credits = 0;
2925 struct buffer_head *bh = NULL;
2926 struct ocfs2_xattr_block *xb = NULL;
2927 struct ocfs2_xattr_entry *xe = NULL;
2928 struct ocfs2_xattr_value_root *xv = NULL;
2930 int name_offset, name_len = 0;
2931 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2936 * Calculate the clusters we need to write.
2937 * No matter whether we replace an old one or add a new one,
2938 * we need this for writing.
2940 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2941 credits += new_clusters *
2942 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2944 if (xis->not_found && xbs->not_found) {
2945 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2947 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2948 clusters_add += new_clusters;
2949 credits += ocfs2_calc_extend_credits(inode->i_sb,
2957 if (!xis->not_found) {
2959 name_offset = le16_to_cpu(xe->xe_name_offset);
2960 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2962 credits += OCFS2_INODE_UPDATE_CREDITS;
2964 int i, block_off = 0;
2965 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2967 name_offset = le16_to_cpu(xe->xe_name_offset);
2968 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2969 i = xbs->here - xbs->header->xh_entries;
2972 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2973 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2974 bucket_xh(xbs->bucket),
2977 base = bucket_block(xbs->bucket, block_off);
2978 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2981 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2986 * delete a xattr doesn't need metadata and cluster allocation.
2987 * so just calculate the credits and return.
2989 * The credits for removing the value tree will be extended
2990 * by ocfs2_remove_extent itself.
2992 if (!xi->xi_value) {
2993 if (!ocfs2_xattr_is_local(xe))
2994 credits += ocfs2_remove_extent_credits(inode->i_sb);
2999 /* do cluster allocation guess first. */
3000 value_size = le64_to_cpu(xe->xe_value_size);
3004 * In xattr set, we always try to set the xe in inode first,
3005 * so if it can be inserted into inode successfully, the old
3006 * one will be removed from the xattr block, and this xattr
3007 * will be inserted into inode as a new xattr in inode.
3009 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3010 clusters_add += new_clusters;
3011 credits += ocfs2_remove_extent_credits(inode->i_sb) +
3012 OCFS2_INODE_UPDATE_CREDITS;
3013 if (!ocfs2_xattr_is_local(xe))
3014 credits += ocfs2_calc_extend_credits(
3022 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3023 /* the new values will be stored outside. */
3024 u32 old_clusters = 0;
3026 if (!ocfs2_xattr_is_local(xe)) {
3027 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3029 xv = (struct ocfs2_xattr_value_root *)
3030 (base + name_offset + name_len);
3031 value_size = OCFS2_XATTR_ROOT_SIZE;
3035 if (old_clusters >= new_clusters) {
3036 credits += ocfs2_remove_extent_credits(inode->i_sb);
3039 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3040 clusters_add += new_clusters - old_clusters;
3041 credits += ocfs2_calc_extend_credits(inode->i_sb,
3045 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3050 * Now the new value will be stored inside. So if the new
3051 * value is smaller than the size of value root or the old
3052 * value, we don't need any allocation, otherwise we have
3053 * to guess metadata allocation.
3055 if ((ocfs2_xattr_is_local(xe) &&
3056 (value_size >= xi->xi_value_len)) ||
3057 (!ocfs2_xattr_is_local(xe) &&
3058 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3063 /* calculate metadata allocation. */
3064 if (di->i_xattr_loc) {
3065 if (!xbs->xattr_bh) {
3066 ret = ocfs2_read_xattr_block(inode,
3067 le64_to_cpu(di->i_xattr_loc),
3074 xb = (struct ocfs2_xattr_block *)bh->b_data;
3076 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3079 * If there is already an xattr tree, good, we can calculate
3080 * like other b-trees. Otherwise we may have the chance of
3081 * create a tree, the credit calculation is borrowed from
3082 * ocfs2_calc_extend_credits with root_el = NULL. And the
3083 * new tree will be cluster based, so no meta is needed.
3085 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3086 struct ocfs2_extent_list *el =
3087 &xb->xb_attrs.xb_root.xt_list;
3088 meta_add += ocfs2_extend_meta_needed(el);
3089 credits += ocfs2_calc_extend_credits(inode->i_sb,
3092 credits += OCFS2_SUBALLOC_ALLOC + 1;
3095 * This cluster will be used either for new bucket or for
3097 * If the cluster size is the same as the bucket size, one
3098 * more is needed since we may need to extend the bucket
3102 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3103 if (OCFS2_XATTR_BUCKET_SIZE ==
3104 OCFS2_SB(inode->i_sb)->s_clustersize) {
3105 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3110 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3114 *clusters_need = clusters_add;
3116 *meta_need = meta_add;
3118 *credits_need = credits;
3123 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3124 struct ocfs2_dinode *di,
3125 struct ocfs2_xattr_info *xi,
3126 struct ocfs2_xattr_search *xis,
3127 struct ocfs2_xattr_search *xbs,
3128 struct ocfs2_xattr_set_ctxt *ctxt,
3132 int clusters_add, meta_add, ret;
3133 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3135 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3137 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3139 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3140 &clusters_add, &meta_add, credits);
3146 meta_add += extra_meta;
3147 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
3148 "credits = %d\n", xi->xi_name, meta_add, clusters_add, *credits);
3151 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3160 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3166 if (ctxt->meta_ac) {
3167 ocfs2_free_alloc_context(ctxt->meta_ac);
3168 ctxt->meta_ac = NULL;
3172 * We cannot have an error and a non null ctxt->data_ac.
3179 static int __ocfs2_xattr_set_handle(struct inode *inode,
3180 struct ocfs2_dinode *di,
3181 struct ocfs2_xattr_info *xi,
3182 struct ocfs2_xattr_search *xis,
3183 struct ocfs2_xattr_search *xbs,
3184 struct ocfs2_xattr_set_ctxt *ctxt)
3186 int ret = 0, credits, old_found;
3188 if (!xi->xi_value) {
3189 /* Remove existing extended attribute */
3190 if (!xis->not_found)
3191 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3192 else if (!xbs->not_found)
3193 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3195 /* We always try to set extended attribute into inode first*/
3196 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3197 if (!ret && !xbs->not_found) {
3199 * If succeed and that extended attribute existing in
3200 * external block, then we will remove it.
3202 xi->xi_value = NULL;
3203 xi->xi_value_len = 0;
3205 old_found = xis->not_found;
3206 xis->not_found = -ENODATA;
3207 ret = ocfs2_calc_xattr_set_need(inode,
3215 xis->not_found = old_found;
3221 ret = ocfs2_extend_trans(ctxt->handle, credits +
3222 ctxt->handle->h_buffer_credits);
3227 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3228 } else if (ret == -ENOSPC) {
3229 if (di->i_xattr_loc && !xbs->xattr_bh) {
3230 ret = ocfs2_xattr_block_find(inode,
3236 old_found = xis->not_found;
3237 xis->not_found = -ENODATA;
3238 ret = ocfs2_calc_xattr_set_need(inode,
3246 xis->not_found = old_found;
3252 ret = ocfs2_extend_trans(ctxt->handle, credits +
3253 ctxt->handle->h_buffer_credits);
3260 * If no space in inode, we will set extended attribute
3261 * into external block.
3263 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3266 if (!xis->not_found) {
3268 * If succeed and that extended attribute
3269 * existing in inode, we will remove it.
3271 xi->xi_value = NULL;
3272 xi->xi_value_len = 0;
3273 xbs->not_found = -ENODATA;
3274 ret = ocfs2_calc_xattr_set_need(inode,
3287 ret = ocfs2_extend_trans(ctxt->handle, credits +
3288 ctxt->handle->h_buffer_credits);
3293 ret = ocfs2_xattr_ibody_set(inode, xi,
3300 /* Update inode ctime. */
3301 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3303 OCFS2_JOURNAL_ACCESS_WRITE);
3309 inode->i_ctime = CURRENT_TIME;
3310 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3311 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3312 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3319 * This function only called duing creating inode
3320 * for init security/acl xattrs of the new inode.
3321 * All transanction credits have been reserved in mknod.
3323 int ocfs2_xattr_set_handle(handle_t *handle,
3324 struct inode *inode,
3325 struct buffer_head *di_bh,
3331 struct ocfs2_alloc_context *meta_ac,
3332 struct ocfs2_alloc_context *data_ac)
3334 struct ocfs2_dinode *di;
3337 struct ocfs2_xattr_info xi = {
3338 .xi_name_index = name_index,
3340 .xi_name_len = strlen(name),
3342 .xi_value_len = value_len,
3345 struct ocfs2_xattr_search xis = {
3346 .not_found = -ENODATA,
3349 struct ocfs2_xattr_search xbs = {
3350 .not_found = -ENODATA,
3353 struct ocfs2_xattr_set_ctxt ctxt = {
3359 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3363 * In extreme situation, may need xattr bucket when
3364 * block size is too small. And we have already reserved
3365 * the credits for bucket in mknod.
3367 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3368 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3370 mlog_errno(-ENOMEM);
3375 xis.inode_bh = xbs.inode_bh = di_bh;
3376 di = (struct ocfs2_dinode *)di_bh->b_data;
3378 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3380 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3383 if (xis.not_found) {
3384 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3389 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3392 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3393 brelse(xbs.xattr_bh);
3394 ocfs2_xattr_bucket_free(xbs.bucket);
3402 * Set, replace or remove an extended attribute for this inode.
3403 * value is NULL to remove an existing extended attribute, else either
3404 * create or replace an extended attribute.
3406 int ocfs2_xattr_set(struct inode *inode,
3413 struct buffer_head *di_bh = NULL;
3414 struct ocfs2_dinode *di;
3415 int ret, credits, ref_meta = 0, ref_credits = 0;
3416 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3417 struct inode *tl_inode = osb->osb_tl_inode;
3418 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3419 struct ocfs2_refcount_tree *ref_tree = NULL;
3421 struct ocfs2_xattr_info xi = {
3422 .xi_name_index = name_index,
3424 .xi_name_len = strlen(name),
3426 .xi_value_len = value_len,
3429 struct ocfs2_xattr_search xis = {
3430 .not_found = -ENODATA,
3433 struct ocfs2_xattr_search xbs = {
3434 .not_found = -ENODATA,
3437 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3441 * Only xbs will be used on indexed trees. xis doesn't need a
3444 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3446 mlog_errno(-ENOMEM);
3450 ret = ocfs2_inode_lock(inode, &di_bh, 1);
3453 goto cleanup_nolock;
3455 xis.inode_bh = xbs.inode_bh = di_bh;
3456 di = (struct ocfs2_dinode *)di_bh->b_data;
3458 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3460 * Scan inode and external block to find the same name
3461 * extended attribute and collect search infomation.
3463 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3466 if (xis.not_found) {
3467 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3472 if (xis.not_found && xbs.not_found) {
3474 if (flags & XATTR_REPLACE)
3481 if (flags & XATTR_CREATE)
3485 /* Check whether the value is refcounted and do some prepartion. */
3486 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3487 (!xis.not_found || !xbs.not_found)) {
3488 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3489 &xis, &xbs, &ref_tree,
3490 &ref_meta, &ref_credits);
3497 mutex_lock(&tl_inode->i_mutex);
3499 if (ocfs2_truncate_log_needs_flush(osb)) {
3500 ret = __ocfs2_flush_truncate_log(osb);
3502 mutex_unlock(&tl_inode->i_mutex);
3507 mutex_unlock(&tl_inode->i_mutex);
3509 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3510 &xbs, &ctxt, ref_meta, &credits);
3516 /* we need to update inode's ctime field, so add credit for it. */
3517 credits += OCFS2_INODE_UPDATE_CREDITS;
3518 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3519 if (IS_ERR(ctxt.handle)) {
3520 ret = PTR_ERR(ctxt.handle);
3525 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3527 ocfs2_commit_trans(osb, ctxt.handle);
3530 ocfs2_free_alloc_context(ctxt.data_ac);
3532 ocfs2_free_alloc_context(ctxt.meta_ac);
3533 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3534 ocfs2_schedule_truncate_log_flush(osb, 1);
3535 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3539 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3540 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3541 if (!value && !ret) {
3542 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3546 ocfs2_inode_unlock(inode, 1);
3549 brelse(xbs.xattr_bh);
3550 ocfs2_xattr_bucket_free(xbs.bucket);
3556 * Find the xattr extent rec which may contains name_hash.
3557 * e_cpos will be the first name hash of the xattr rec.
3558 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3560 static int ocfs2_xattr_get_rec(struct inode *inode,
3565 struct ocfs2_extent_list *el)
3568 struct buffer_head *eb_bh = NULL;
3569 struct ocfs2_extent_block *eb;
3570 struct ocfs2_extent_rec *rec = NULL;
3573 if (el->l_tree_depth) {
3574 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3581 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3584 if (el->l_tree_depth) {
3585 ocfs2_error(inode->i_sb,
3586 "Inode %lu has non zero tree depth in "
3587 "xattr tree block %llu\n", inode->i_ino,
3588 (unsigned long long)eb_bh->b_blocknr);
3594 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3595 rec = &el->l_recs[i];
3597 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3598 e_blkno = le64_to_cpu(rec->e_blkno);
3604 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3605 "record (%u, %u, 0) in xattr", inode->i_ino,
3606 le32_to_cpu(rec->e_cpos),
3607 ocfs2_rec_clusters(el, rec));
3612 *p_blkno = le64_to_cpu(rec->e_blkno);
3613 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3615 *e_cpos = le32_to_cpu(rec->e_cpos);
3621 typedef int (xattr_bucket_func)(struct inode *inode,
3622 struct ocfs2_xattr_bucket *bucket,
3625 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3626 struct ocfs2_xattr_bucket *bucket,
3633 int i, ret = 0, cmp = 1, block_off, new_offset;
3634 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3635 size_t name_len = strlen(name);
3636 struct ocfs2_xattr_entry *xe = NULL;
3640 * We don't use binary search in the bucket because there
3641 * may be multiple entries with the same name hash.
3643 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3644 xe = &xh->xh_entries[i];
3646 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3648 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3651 cmp = name_index - ocfs2_xattr_get_type(xe);
3653 cmp = name_len - xe->xe_name_len;
3657 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3668 xe_name = bucket_block(bucket, block_off) + new_offset;
3669 if (!memcmp(name, xe_name, name_len)) {
3681 * Find the specified xattr entry in a series of buckets.
3682 * This series start from p_blkno and last for num_clusters.
3683 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3684 * the num of the valid buckets.
3686 * Return the buffer_head this xattr should reside in. And if the xattr's
3687 * hash is in the gap of 2 buckets, return the lower bucket.
3689 static int ocfs2_xattr_bucket_find(struct inode *inode,
3696 struct ocfs2_xattr_search *xs)
3699 struct ocfs2_xattr_header *xh = NULL;
3700 struct ocfs2_xattr_entry *xe = NULL;
3702 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3703 int low_bucket = 0, bucket, high_bucket;
3704 struct ocfs2_xattr_bucket *search;
3706 u64 blkno, lower_blkno = 0;
3708 search = ocfs2_xattr_bucket_new(inode);
3715 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3721 xh = bucket_xh(search);
3722 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3723 while (low_bucket <= high_bucket) {
3724 ocfs2_xattr_bucket_relse(search);
3726 bucket = (low_bucket + high_bucket) / 2;
3727 blkno = p_blkno + bucket * blk_per_bucket;
3728 ret = ocfs2_read_xattr_bucket(search, blkno);
3734 xh = bucket_xh(search);
3735 xe = &xh->xh_entries[0];
3736 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3737 high_bucket = bucket - 1;
3742 * Check whether the hash of the last entry in our
3743 * bucket is larger than the search one. for an empty
3744 * bucket, the last one is also the first one.
3747 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3749 last_hash = le32_to_cpu(xe->xe_name_hash);
3751 /* record lower_blkno which may be the insert place. */
3752 lower_blkno = blkno;
3754 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3755 low_bucket = bucket + 1;
3759 /* the searched xattr should reside in this bucket if exists. */
3760 ret = ocfs2_find_xe_in_bucket(inode, search,
3761 name_index, name, name_hash,
3771 * Record the bucket we have found.
3772 * When the xattr's hash value is in the gap of 2 buckets, we will
3773 * always set it to the previous bucket.
3776 lower_blkno = p_blkno;
3778 /* This should be in cache - we just read it during the search */
3779 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3785 xs->header = bucket_xh(xs->bucket);
3786 xs->base = bucket_block(xs->bucket, 0);
3787 xs->end = xs->base + inode->i_sb->s_blocksize;
3790 xs->here = &xs->header->xh_entries[index];
3791 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3792 (unsigned long long)bucket_blkno(xs->bucket), index);
3797 ocfs2_xattr_bucket_free(search);
3801 static int ocfs2_xattr_index_block_find(struct inode *inode,
3802 struct buffer_head *root_bh,
3805 struct ocfs2_xattr_search *xs)
3808 struct ocfs2_xattr_block *xb =
3809 (struct ocfs2_xattr_block *)root_bh->b_data;
3810 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3811 struct ocfs2_extent_list *el = &xb_root->xt_list;
3813 u32 first_hash, num_clusters = 0;
3814 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3816 if (le16_to_cpu(el->l_next_free_rec) == 0)
3819 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3820 name, name_hash, name_index);
3822 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3829 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3831 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3832 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3835 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3836 p_blkno, first_hash, num_clusters, xs);
3842 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3845 xattr_bucket_func *func,
3849 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3850 u32 num_buckets = clusters * bpc;
3851 struct ocfs2_xattr_bucket *bucket;
3853 bucket = ocfs2_xattr_bucket_new(inode);
3855 mlog_errno(-ENOMEM);
3859 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3860 clusters, (unsigned long long)blkno);
3862 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3863 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3870 * The real bucket num in this series of blocks is stored
3871 * in the 1st bucket.
3874 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3876 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3877 (unsigned long long)blkno,
3878 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3880 ret = func(inode, bucket, para);
3881 if (ret && ret != -ERANGE)
3883 /* Fall through to bucket_relse() */
3886 ocfs2_xattr_bucket_relse(bucket);
3891 ocfs2_xattr_bucket_free(bucket);
3895 struct ocfs2_xattr_tree_list {
3901 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3902 struct ocfs2_xattr_header *xh,
3909 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3912 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3914 *block_off = name_offset >> sb->s_blocksize_bits;
3915 *new_offset = name_offset % sb->s_blocksize;
3920 static int ocfs2_list_xattr_bucket(struct inode *inode,
3921 struct ocfs2_xattr_bucket *bucket,
3925 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3926 int i, block_off, new_offset;
3927 const char *prefix, *name;
3929 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3930 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3931 type = ocfs2_xattr_get_type(entry);
3932 prefix = ocfs2_xattr_prefix(type);
3935 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3943 name = (const char *)bucket_block(bucket, block_off) +
3945 ret = ocfs2_xattr_list_entry(xl->buffer,
3949 entry->xe_name_len);
3958 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
3959 struct buffer_head *blk_bh,
3960 xattr_tree_rec_func *rec_func,
3963 struct ocfs2_xattr_block *xb =
3964 (struct ocfs2_xattr_block *)blk_bh->b_data;
3965 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
3967 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3970 if (!el->l_next_free_rec || !rec_func)
3973 while (name_hash > 0) {
3974 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3975 &e_cpos, &num_clusters, el);
3981 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
3982 num_clusters, para);
3992 name_hash = e_cpos - 1;
3999 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4000 struct buffer_head *root_bh,
4001 u64 blkno, u32 cpos, u32 len, void *para)
4003 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4004 ocfs2_list_xattr_bucket, para);
4007 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4008 struct buffer_head *blk_bh,
4013 struct ocfs2_xattr_tree_list xl = {
4015 .buffer_size = buffer_size,
4019 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4020 ocfs2_list_xattr_tree_rec, &xl);
4031 static int cmp_xe(const void *a, const void *b)
4033 const struct ocfs2_xattr_entry *l = a, *r = b;
4034 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4035 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4037 if (l_hash > r_hash)
4039 if (l_hash < r_hash)
4044 static void swap_xe(void *a, void *b, int size)
4046 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4049 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4050 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4054 * When the ocfs2_xattr_block is filled up, new bucket will be created
4055 * and all the xattr entries will be moved to the new bucket.
4056 * The header goes at the start of the bucket, and the names+values are
4057 * filled from the end. This is why *target starts as the last buffer.
4058 * Note: we need to sort the entries since they are not saved in order
4059 * in the ocfs2_xattr_block.
4061 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4062 struct buffer_head *xb_bh,
4063 struct ocfs2_xattr_bucket *bucket)
4065 int i, blocksize = inode->i_sb->s_blocksize;
4066 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4067 u16 offset, size, off_change;
4068 struct ocfs2_xattr_entry *xe;
4069 struct ocfs2_xattr_block *xb =
4070 (struct ocfs2_xattr_block *)xb_bh->b_data;
4071 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4072 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4073 u16 count = le16_to_cpu(xb_xh->xh_count);
4074 char *src = xb_bh->b_data;
4075 char *target = bucket_block(bucket, blks - 1);
4077 mlog(0, "cp xattr from block %llu to bucket %llu\n",
4078 (unsigned long long)xb_bh->b_blocknr,
4079 (unsigned long long)bucket_blkno(bucket));
4081 for (i = 0; i < blks; i++)
4082 memset(bucket_block(bucket, i), 0, blocksize);
4085 * Since the xe_name_offset is based on ocfs2_xattr_header,
4086 * there is a offset change corresponding to the change of
4087 * ocfs2_xattr_header's position.
4089 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4090 xe = &xb_xh->xh_entries[count - 1];
4091 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4092 size = blocksize - offset;
4094 /* copy all the names and values. */
4095 memcpy(target + offset, src + offset, size);
4097 /* Init new header now. */
4098 xh->xh_count = xb_xh->xh_count;
4099 xh->xh_num_buckets = cpu_to_le16(1);
4100 xh->xh_name_value_len = cpu_to_le16(size);
4101 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4103 /* copy all the entries. */
4104 target = bucket_block(bucket, 0);
4105 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4106 size = count * sizeof(struct ocfs2_xattr_entry);
4107 memcpy(target + offset, (char *)xb_xh + offset, size);
4109 /* Change the xe offset for all the xe because of the move. */
4110 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4111 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4112 for (i = 0; i < count; i++)
4113 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4115 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
4116 offset, size, off_change);
4118 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4123 * After we move xattr from block to index btree, we have to
4124 * update ocfs2_xattr_search to the new xe and base.
4126 * When the entry is in xattr block, xattr_bh indicates the storage place.
4127 * While if the entry is in index b-tree, "bucket" indicates the
4128 * real place of the xattr.
4130 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4131 struct ocfs2_xattr_search *xs,
4132 struct buffer_head *old_bh)
4134 char *buf = old_bh->b_data;
4135 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4136 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4139 xs->header = bucket_xh(xs->bucket);
4140 xs->base = bucket_block(xs->bucket, 0);
4141 xs->end = xs->base + inode->i_sb->s_blocksize;
4146 i = xs->here - old_xh->xh_entries;
4147 xs->here = &xs->header->xh_entries[i];
4150 static int ocfs2_xattr_create_index_block(struct inode *inode,
4151 struct ocfs2_xattr_search *xs,
4152 struct ocfs2_xattr_set_ctxt *ctxt)
4157 handle_t *handle = ctxt->handle;
4158 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4159 struct ocfs2_inode_info *oi = OCFS2_I(inode);
4160 struct buffer_head *xb_bh = xs->xattr_bh;
4161 struct ocfs2_xattr_block *xb =
4162 (struct ocfs2_xattr_block *)xb_bh->b_data;
4163 struct ocfs2_xattr_tree_root *xr;
4164 u16 xb_flags = le16_to_cpu(xb->xb_flags);
4166 mlog(0, "create xattr index block for %llu\n",
4167 (unsigned long long)xb_bh->b_blocknr);
4169 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4170 BUG_ON(!xs->bucket);
4174 * We can use this lock for now, and maybe move to a dedicated mutex
4175 * if performance becomes a problem later.
4177 down_write(&oi->ip_alloc_sem);
4179 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4180 OCFS2_JOURNAL_ACCESS_WRITE);
4186 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
4187 1, 1, &bit_off, &len);
4194 * The bucket may spread in many blocks, and
4195 * we will only touch the 1st block and the last block
4196 * in the whole bucket(one for entry and one for data).
4198 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4200 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
4201 (unsigned long long)blkno);
4203 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4209 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4210 OCFS2_JOURNAL_ACCESS_CREATE);
4216 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4217 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4219 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4221 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4222 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4223 offsetof(struct ocfs2_xattr_block, xb_attrs));
4225 xr = &xb->xb_attrs.xb_root;
4226 xr->xt_clusters = cpu_to_le32(1);
4227 xr->xt_last_eb_blk = 0;
4228 xr->xt_list.l_tree_depth = 0;
4229 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4230 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4232 xr->xt_list.l_recs[0].e_cpos = 0;
4233 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4234 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4236 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4238 ocfs2_journal_dirty(handle, xb_bh);
4241 up_write(&oi->ip_alloc_sem);
4246 static int cmp_xe_offset(const void *a, const void *b)
4248 const struct ocfs2_xattr_entry *l = a, *r = b;
4249 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4250 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4252 if (l_name_offset < r_name_offset)
4254 if (l_name_offset > r_name_offset)
4260 * defrag a xattr bucket if we find that the bucket has some
4261 * holes beteen name/value pairs.
4262 * We will move all the name/value pairs to the end of the bucket
4263 * so that we can spare some space for insertion.
4265 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4267 struct ocfs2_xattr_bucket *bucket)
4270 size_t end, offset, len;
4271 struct ocfs2_xattr_header *xh;
4272 char *entries, *buf, *bucket_buf = NULL;
4273 u64 blkno = bucket_blkno(bucket);
4275 size_t blocksize = inode->i_sb->s_blocksize;
4276 struct ocfs2_xattr_entry *xe;
4279 * In order to make the operation more efficient and generic,
4280 * we copy all the blocks into a contiguous memory and do the
4281 * defragment there, so if anything is error, we will not touch
4284 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4291 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4292 memcpy(buf, bucket_block(bucket, i), blocksize);
4294 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4295 OCFS2_JOURNAL_ACCESS_WRITE);
4301 xh = (struct ocfs2_xattr_header *)bucket_buf;
4302 entries = (char *)xh->xh_entries;
4303 xh_free_start = le16_to_cpu(xh->xh_free_start);
4305 mlog(0, "adjust xattr bucket in %llu, count = %u, "
4306 "xh_free_start = %u, xh_name_value_len = %u.\n",
4307 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4308 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4311 * sort all the entries by their offset.
4312 * the largest will be the first, so that we can
4313 * move them to the end one by one.
4315 sort(entries, le16_to_cpu(xh->xh_count),
4316 sizeof(struct ocfs2_xattr_entry),
4317 cmp_xe_offset, swap_xe);
4319 /* Move all name/values to the end of the bucket. */
4320 xe = xh->xh_entries;
4321 end = OCFS2_XATTR_BUCKET_SIZE;
4322 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4323 offset = le16_to_cpu(xe->xe_name_offset);
4324 len = namevalue_size_xe(xe);
4327 * We must make sure that the name/value pair
4328 * exist in the same block. So adjust end to
4329 * the previous block end if needed.
4331 if (((end - len) / blocksize !=
4332 (end - 1) / blocksize))
4333 end = end - end % blocksize;
4335 if (end > offset + len) {
4336 memmove(bucket_buf + end - len,
4337 bucket_buf + offset, len);
4338 xe->xe_name_offset = cpu_to_le16(end - len);
4341 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4342 "bucket %llu\n", (unsigned long long)blkno);
4347 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4348 "bucket %llu\n", (unsigned long long)blkno);
4350 if (xh_free_start == end)
4353 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4354 xh->xh_free_start = cpu_to_le16(end);
4356 /* sort the entries by their name_hash. */
4357 sort(entries, le16_to_cpu(xh->xh_count),
4358 sizeof(struct ocfs2_xattr_entry),
4362 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4363 memcpy(bucket_block(bucket, i), buf, blocksize);
4364 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4372 * prev_blkno points to the start of an existing extent. new_blkno
4373 * points to a newly allocated extent. Because we know each of our
4374 * clusters contains more than bucket, we can easily split one cluster
4375 * at a bucket boundary. So we take the last cluster of the existing
4376 * extent and split it down the middle. We move the last half of the
4377 * buckets in the last cluster of the existing extent over to the new
4380 * first_bh is the buffer at prev_blkno so we can update the existing
4381 * extent's bucket count. header_bh is the bucket were we were hoping
4382 * to insert our xattr. If the bucket move places the target in the new
4383 * extent, we'll update first_bh and header_bh after modifying the old
4386 * first_hash will be set as the 1st xe's name_hash in the new extent.
4388 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4390 struct ocfs2_xattr_bucket *first,
4391 struct ocfs2_xattr_bucket *target,
4397 struct super_block *sb = inode->i_sb;
4398 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4399 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4400 int to_move = num_buckets / 2;
4402 u64 last_cluster_blkno = bucket_blkno(first) +
4403 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4405 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4406 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4408 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4409 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
4411 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4412 last_cluster_blkno, new_blkno,
4413 to_move, first_hash);
4419 /* This is the first bucket that got moved */
4420 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4423 * If the target bucket was part of the moved buckets, we need to
4424 * update first and target.
4426 if (bucket_blkno(target) >= src_blkno) {
4427 /* Find the block for the new target bucket */
4428 src_blkno = new_blkno +
4429 (bucket_blkno(target) - src_blkno);
4431 ocfs2_xattr_bucket_relse(first);
4432 ocfs2_xattr_bucket_relse(target);
4435 * These shouldn't fail - the buffers are in the
4436 * journal from ocfs2_cp_xattr_bucket().
4438 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4443 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4454 * Find the suitable pos when we divide a bucket into 2.
4455 * We have to make sure the xattrs with the same hash value exist
4456 * in the same bucket.
4458 * If this ocfs2_xattr_header covers more than one hash value, find a
4459 * place where the hash value changes. Try to find the most even split.
4460 * The most common case is that all entries have different hash values,
4461 * and the first check we make will find a place to split.
4463 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4465 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4466 int count = le16_to_cpu(xh->xh_count);
4467 int delta, middle = count / 2;
4470 * We start at the middle. Each step gets farther away in both
4471 * directions. We therefore hit the change in hash value
4472 * nearest to the middle. Note that this loop does not execute for
4475 for (delta = 0; delta < middle; delta++) {
4476 /* Let's check delta earlier than middle */
4477 if (cmp_xe(&entries[middle - delta - 1],
4478 &entries[middle - delta]))
4479 return middle - delta;
4481 /* For even counts, don't walk off the end */
4482 if ((middle + delta + 1) == count)
4485 /* Now try delta past middle */
4486 if (cmp_xe(&entries[middle + delta],
4487 &entries[middle + delta + 1]))
4488 return middle + delta + 1;
4491 /* Every entry had the same hash */
4496 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4497 * first_hash will record the 1st hash of the new bucket.
4499 * Normally half of the xattrs will be moved. But we have to make
4500 * sure that the xattrs with the same hash value are stored in the
4501 * same bucket. If all the xattrs in this bucket have the same hash
4502 * value, the new bucket will be initialized as an empty one and the
4503 * first_hash will be initialized as (hash_value+1).
4505 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4510 int new_bucket_head)
4513 int count, start, len, name_value_len = 0, name_offset = 0;
4514 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4515 struct ocfs2_xattr_header *xh;
4516 struct ocfs2_xattr_entry *xe;
4517 int blocksize = inode->i_sb->s_blocksize;
4519 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4520 (unsigned long long)blk, (unsigned long long)new_blk);
4522 s_bucket = ocfs2_xattr_bucket_new(inode);
4523 t_bucket = ocfs2_xattr_bucket_new(inode);
4524 if (!s_bucket || !t_bucket) {
4530 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4536 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4537 OCFS2_JOURNAL_ACCESS_WRITE);
4544 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4545 * there's no need to read it.
4547 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4554 * Hey, if we're overwriting t_bucket, what difference does
4555 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4556 * same part of ocfs2_cp_xattr_bucket().
4558 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4560 OCFS2_JOURNAL_ACCESS_CREATE :
4561 OCFS2_JOURNAL_ACCESS_WRITE);
4567 xh = bucket_xh(s_bucket);
4568 count = le16_to_cpu(xh->xh_count);
4569 start = ocfs2_xattr_find_divide_pos(xh);
4571 if (start == count) {
4572 xe = &xh->xh_entries[start-1];
4575 * initialized a new empty bucket here.
4576 * The hash value is set as one larger than
4577 * that of the last entry in the previous bucket.
4579 for (i = 0; i < t_bucket->bu_blocks; i++)
4580 memset(bucket_block(t_bucket, i), 0, blocksize);
4582 xh = bucket_xh(t_bucket);
4583 xh->xh_free_start = cpu_to_le16(blocksize);
4584 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4585 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4587 goto set_num_buckets;
4590 /* copy the whole bucket to the new first. */
4591 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4593 /* update the new bucket. */
4594 xh = bucket_xh(t_bucket);
4597 * Calculate the total name/value len and xh_free_start for
4598 * the old bucket first.
4600 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4602 for (i = 0; i < start; i++) {
4603 xe = &xh->xh_entries[i];
4604 name_value_len += namevalue_size_xe(xe);
4605 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4606 name_offset = le16_to_cpu(xe->xe_name_offset);
4610 * Now begin the modification to the new bucket.
4612 * In the new bucket, We just move the xattr entry to the beginning
4613 * and don't touch the name/value. So there will be some holes in the
4614 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4617 xe = &xh->xh_entries[start];
4618 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4619 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
4620 (int)((char *)xe - (char *)xh),
4621 (int)((char *)xh->xh_entries - (char *)xh));
4622 memmove((char *)xh->xh_entries, (char *)xe, len);
4623 xe = &xh->xh_entries[count - start];
4624 len = sizeof(struct ocfs2_xattr_entry) * start;
4625 memset((char *)xe, 0, len);
4627 le16_add_cpu(&xh->xh_count, -start);
4628 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4630 /* Calculate xh_free_start for the new bucket. */
4631 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4632 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4633 xe = &xh->xh_entries[i];
4634 if (le16_to_cpu(xe->xe_name_offset) <
4635 le16_to_cpu(xh->xh_free_start))
4636 xh->xh_free_start = xe->xe_name_offset;
4640 /* set xh->xh_num_buckets for the new xh. */
4641 if (new_bucket_head)
4642 xh->xh_num_buckets = cpu_to_le16(1);
4644 xh->xh_num_buckets = 0;
4646 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4648 /* store the first_hash of the new bucket. */
4650 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4653 * Now only update the 1st block of the old bucket. If we
4654 * just added a new empty bucket, there is no need to modify
4660 xh = bucket_xh(s_bucket);
4661 memset(&xh->xh_entries[start], 0,
4662 sizeof(struct ocfs2_xattr_entry) * (count - start));
4663 xh->xh_count = cpu_to_le16(start);
4664 xh->xh_free_start = cpu_to_le16(name_offset);
4665 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4667 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4670 ocfs2_xattr_bucket_free(s_bucket);
4671 ocfs2_xattr_bucket_free(t_bucket);
4677 * Copy xattr from one bucket to another bucket.
4679 * The caller must make sure that the journal transaction
4680 * has enough space for journaling.
4682 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4689 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4691 BUG_ON(s_blkno == t_blkno);
4693 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4694 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4697 s_bucket = ocfs2_xattr_bucket_new(inode);
4698 t_bucket = ocfs2_xattr_bucket_new(inode);
4699 if (!s_bucket || !t_bucket) {
4705 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4710 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4711 * there's no need to read it.
4713 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4718 * Hey, if we're overwriting t_bucket, what difference does
4719 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4720 * cluster to fill, we came here from
4721 * ocfs2_mv_xattr_buckets(), and it is really new -
4722 * ACCESS_CREATE is required. But we also might have moved data
4723 * out of t_bucket before extending back into it.
4724 * ocfs2_add_new_xattr_bucket() can do this - its call to
4725 * ocfs2_add_new_xattr_cluster() may have created a new extent
4726 * and copied out the end of the old extent. Then it re-extends
4727 * the old extent back to create space for new xattrs. That's
4728 * how we get here, and the bucket isn't really new.
4730 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4732 OCFS2_JOURNAL_ACCESS_CREATE :
4733 OCFS2_JOURNAL_ACCESS_WRITE);
4737 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4738 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4741 ocfs2_xattr_bucket_free(t_bucket);
4742 ocfs2_xattr_bucket_free(s_bucket);
4748 * src_blk points to the start of an existing extent. last_blk points to
4749 * last cluster in that extent. to_blk points to a newly allocated
4750 * extent. We copy the buckets from the cluster at last_blk to the new
4751 * extent. If start_bucket is non-zero, we skip that many buckets before
4752 * we start copying. The new extent's xh_num_buckets gets set to the
4753 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4754 * by the same amount.
4756 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4757 u64 src_blk, u64 last_blk, u64 to_blk,
4758 unsigned int start_bucket,
4761 int i, ret, credits;
4762 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4763 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4764 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4765 struct ocfs2_xattr_bucket *old_first, *new_first;
4767 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4768 (unsigned long long)last_blk, (unsigned long long)to_blk);
4770 BUG_ON(start_bucket >= num_buckets);
4772 num_buckets -= start_bucket;
4773 last_blk += (start_bucket * blks_per_bucket);
4776 /* The first bucket of the original extent */
4777 old_first = ocfs2_xattr_bucket_new(inode);
4778 /* The first bucket of the new extent */
4779 new_first = ocfs2_xattr_bucket_new(inode);
4780 if (!old_first || !new_first) {
4786 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4793 * We need to update the first bucket of the old extent and all
4794 * the buckets going to the new extent.
4796 credits = ((num_buckets + 1) * blks_per_bucket) +
4797 handle->h_buffer_credits;
4798 ret = ocfs2_extend_trans(handle, credits);
4804 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4805 OCFS2_JOURNAL_ACCESS_WRITE);
4811 for (i = 0; i < num_buckets; i++) {
4812 ret = ocfs2_cp_xattr_bucket(inode, handle,
4813 last_blk + (i * blks_per_bucket),
4814 to_blk + (i * blks_per_bucket),
4823 * Get the new bucket ready before we dirty anything
4824 * (This actually shouldn't fail, because we already dirtied
4825 * it once in ocfs2_cp_xattr_bucket()).
4827 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4832 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4833 OCFS2_JOURNAL_ACCESS_WRITE);
4839 /* Now update the headers */
4840 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4841 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4843 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4844 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4847 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4850 ocfs2_xattr_bucket_free(new_first);
4851 ocfs2_xattr_bucket_free(old_first);
4856 * Move some xattrs in this cluster to the new cluster.
4857 * This function should only be called when bucket size == cluster size.
4858 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4860 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4866 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4867 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4869 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4871 ret = ocfs2_extend_trans(handle, credits);
4877 /* Move half of the xattr in start_blk to the next bucket. */
4878 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4879 new_blk, first_hash, 1);
4883 * Move some xattrs from the old cluster to the new one since they are not
4884 * contiguous in ocfs2 xattr tree.
4886 * new_blk starts a new separate cluster, and we will move some xattrs from
4887 * prev_blk to it. v_start will be set as the first name hash value in this
4888 * new cluster so that it can be used as e_cpos during tree insertion and
4889 * don't collide with our original b-tree operations. first_bh and header_bh
4890 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4891 * to extend the insert bucket.
4893 * The problem is how much xattr should we move to the new one and when should
4894 * we update first_bh and header_bh?
4895 * 1. If cluster size > bucket size, that means the previous cluster has more
4896 * than 1 bucket, so just move half nums of bucket into the new cluster and
4897 * update the first_bh and header_bh if the insert bucket has been moved
4898 * to the new cluster.
4899 * 2. If cluster_size == bucket_size:
4900 * a) If the previous extent rec has more than one cluster and the insert
4901 * place isn't in the last cluster, copy the entire last cluster to the
4902 * new one. This time, we don't need to upate the first_bh and header_bh
4903 * since they will not be moved into the new cluster.
4904 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4905 * the new one. And we set the extend flag to zero if the insert place is
4906 * moved into the new allocated cluster since no extend is needed.
4908 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4910 struct ocfs2_xattr_bucket *first,
4911 struct ocfs2_xattr_bucket *target,
4919 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4920 (unsigned long long)bucket_blkno(first), prev_clusters,
4921 (unsigned long long)new_blk);
4923 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4924 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4933 /* The start of the last cluster in the first extent */
4934 u64 last_blk = bucket_blkno(first) +
4935 ((prev_clusters - 1) *
4936 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4938 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4939 ret = ocfs2_mv_xattr_buckets(inode, handle,
4940 bucket_blkno(first),
4941 last_blk, new_blk, 0,
4946 ret = ocfs2_divide_xattr_cluster(inode, handle,
4952 if ((bucket_blkno(target) == last_blk) && extend)
4961 * Add a new cluster for xattr storage.
4963 * If the new cluster is contiguous with the previous one, it will be
4964 * appended to the same extent record, and num_clusters will be updated.
4965 * If not, we will insert a new extent for it and move some xattrs in
4966 * the last cluster into the new allocated one.
4967 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4968 * lose the benefits of hashing because we'll have to search large leaves.
4969 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4972 * first_bh is the first block of the previous extent rec and header_bh
4973 * indicates the bucket we will insert the new xattrs. They will be updated
4974 * when the header_bh is moved into the new cluster.
4976 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4977 struct buffer_head *root_bh,
4978 struct ocfs2_xattr_bucket *first,
4979 struct ocfs2_xattr_bucket *target,
4983 struct ocfs2_xattr_set_ctxt *ctxt)
4986 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4987 u32 prev_clusters = *num_clusters;
4988 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4990 handle_t *handle = ctxt->handle;
4991 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4992 struct ocfs2_extent_tree et;
4994 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4995 "previous xattr blkno = %llu\n",
4996 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4997 prev_cpos, (unsigned long long)bucket_blkno(first));
4999 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5001 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5002 OCFS2_JOURNAL_ACCESS_WRITE);
5008 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
5009 clusters_to_add, &bit_off, &num_bits);
5016 BUG_ON(num_bits > clusters_to_add);
5018 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5019 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
5020 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
5022 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5023 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5024 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5026 * If this cluster is contiguous with the old one and
5027 * adding this new cluster, we don't surpass the limit of
5028 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5029 * initialized and used like other buckets in the previous
5031 * So add it as a contiguous one. The caller will handle
5034 v_start = prev_cpos + prev_clusters;
5035 *num_clusters = prev_clusters + num_bits;
5036 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
5039 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5053 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
5054 num_bits, (unsigned long long)block, v_start);
5055 ret = ocfs2_insert_extent(handle, &et, v_start, block,
5056 num_bits, 0, ctxt->meta_ac);
5062 ret = ocfs2_journal_dirty(handle, root_bh);
5071 * We are given an extent. 'first' is the bucket at the very front of
5072 * the extent. The extent has space for an additional bucket past
5073 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5074 * of the target bucket. We wish to shift every bucket past the target
5075 * down one, filling in that additional space. When we get back to the
5076 * target, we split the target between itself and the now-empty bucket
5077 * at target+1 (aka, target_blkno + blks_per_bucket).
5079 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5081 struct ocfs2_xattr_bucket *first,
5086 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5087 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5089 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5091 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
5092 "from %llu, len = %u\n", (unsigned long long)target_blk,
5093 (unsigned long long)bucket_blkno(first), num_clusters);
5095 /* The extent must have room for an additional bucket */
5096 BUG_ON(new_bucket >=
5097 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5099 /* end_blk points to the last existing bucket */
5100 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5103 * end_blk is the start of the last existing bucket.
5104 * Thus, (end_blk - target_blk) covers the target bucket and
5105 * every bucket after it up to, but not including, the last
5106 * existing bucket. Then we add the last existing bucket, the
5107 * new bucket, and the first bucket (3 * blk_per_bucket).
5109 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
5110 handle->h_buffer_credits;
5111 ret = ocfs2_extend_trans(handle, credits);
5117 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5118 OCFS2_JOURNAL_ACCESS_WRITE);
5124 while (end_blk != target_blk) {
5125 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5126 end_blk + blk_per_bucket, 0);
5129 end_blk -= blk_per_bucket;
5132 /* Move half of the xattr in target_blkno to the next bucket. */
5133 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5134 target_blk + blk_per_bucket, NULL, 0);
5136 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5137 ocfs2_xattr_bucket_journal_dirty(handle, first);
5144 * Add new xattr bucket in an extent record and adjust the buckets
5145 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5146 * bucket we want to insert into.
5148 * In the easy case, we will move all the buckets after target down by
5149 * one. Half of target's xattrs will be moved to the next bucket.
5151 * If current cluster is full, we'll allocate a new one. This may not
5152 * be contiguous. The underlying calls will make sure that there is
5153 * space for the insert, shifting buckets around if necessary.
5154 * 'target' may be moved by those calls.
5156 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5157 struct buffer_head *xb_bh,
5158 struct ocfs2_xattr_bucket *target,
5159 struct ocfs2_xattr_set_ctxt *ctxt)
5161 struct ocfs2_xattr_block *xb =
5162 (struct ocfs2_xattr_block *)xb_bh->b_data;
5163 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5164 struct ocfs2_extent_list *el = &xb_root->xt_list;
5166 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5167 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5168 int ret, num_buckets, extend = 1;
5170 u32 e_cpos, num_clusters;
5171 /* The bucket at the front of the extent */
5172 struct ocfs2_xattr_bucket *first;
5174 mlog(0, "Add new xattr bucket starting from %llu\n",
5175 (unsigned long long)bucket_blkno(target));
5177 /* The first bucket of the original extent */
5178 first = ocfs2_xattr_bucket_new(inode);
5185 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5192 ret = ocfs2_read_xattr_bucket(first, p_blkno);
5198 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5199 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5201 * This can move first+target if the target bucket moves
5202 * to the new extent.
5204 ret = ocfs2_add_new_xattr_cluster(inode,
5219 ret = ocfs2_extend_xattr_bucket(inode,
5222 bucket_blkno(target),
5229 ocfs2_xattr_bucket_free(first);
5234 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5235 struct ocfs2_xattr_bucket *bucket,
5238 int block_off = offs >> inode->i_sb->s_blocksize_bits;
5240 offs = offs % inode->i_sb->s_blocksize;
5241 return bucket_block(bucket, block_off) + offs;
5245 * Truncate the specified xe_off entry in xattr bucket.
5246 * bucket is indicated by header_bh and len is the new length.
5247 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5249 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5251 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5252 struct ocfs2_xattr_bucket *bucket,
5255 struct ocfs2_xattr_set_ctxt *ctxt)
5259 struct ocfs2_xattr_entry *xe;
5260 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5261 size_t blocksize = inode->i_sb->s_blocksize;
5262 struct ocfs2_xattr_value_buf vb = {
5263 .vb_access = ocfs2_journal_access,
5266 xe = &xh->xh_entries[xe_off];
5268 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5270 offset = le16_to_cpu(xe->xe_name_offset) +
5271 OCFS2_XATTR_SIZE(xe->xe_name_len);
5273 value_blk = offset / blocksize;
5275 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5276 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5278 vb.vb_bh = bucket->bu_bhs[value_blk];
5281 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5282 (vb.vb_bh->b_data + offset % blocksize);
5285 * From here on out we have to dirty the bucket. The generic
5286 * value calls only modify one of the bucket's bhs, but we need
5287 * to send the bucket at once. So if they error, they *could* have
5288 * modified something. We have to assume they did, and dirty
5289 * the whole bucket. This leaves us in a consistent state.
5291 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5292 xe_off, (unsigned long long)bucket_blkno(bucket), len);
5293 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5299 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5300 OCFS2_JOURNAL_ACCESS_WRITE);
5306 xe->xe_value_size = cpu_to_le64(len);
5308 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5314 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5315 struct buffer_head *root_bh,
5322 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5323 struct inode *tl_inode = osb->osb_tl_inode;
5325 struct ocfs2_xattr_block *xb =
5326 (struct ocfs2_xattr_block *)root_bh->b_data;
5327 struct ocfs2_alloc_context *meta_ac = NULL;
5328 struct ocfs2_cached_dealloc_ctxt dealloc;
5329 struct ocfs2_extent_tree et;
5331 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5332 ocfs2_delete_xattr_in_bucket, para);
5338 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5340 ocfs2_init_dealloc_ctxt(&dealloc);
5342 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5343 cpos, len, (unsigned long long)blkno);
5345 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5348 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5354 mutex_lock(&tl_inode->i_mutex);
5356 if (ocfs2_truncate_log_needs_flush(osb)) {
5357 ret = __ocfs2_flush_truncate_log(osb);
5364 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5365 if (IS_ERR(handle)) {
5371 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5372 OCFS2_JOURNAL_ACCESS_WRITE);
5378 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5385 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5387 ret = ocfs2_journal_dirty(handle, root_bh);
5393 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5398 ocfs2_commit_trans(osb, handle);
5400 ocfs2_schedule_truncate_log_flush(osb, 1);
5402 mutex_unlock(&tl_inode->i_mutex);
5405 ocfs2_free_alloc_context(meta_ac);
5407 ocfs2_run_deallocs(osb, &dealloc);
5413 * check whether the xattr bucket is filled up with the same hash value.
5414 * If we want to insert the xattr with the same hash, return -ENOSPC.
5415 * If we want to insert a xattr with different hash value, go ahead
5416 * and ocfs2_divide_xattr_bucket will handle this.
5418 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5419 struct ocfs2_xattr_bucket *bucket,
5422 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5423 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5425 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5428 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5429 xh->xh_entries[0].xe_name_hash) {
5430 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5432 (unsigned long long)bucket_blkno(bucket),
5433 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5441 * Try to set the entry in the current bucket. If we fail, the caller
5442 * will handle getting us another bucket.
5444 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5445 struct ocfs2_xattr_info *xi,
5446 struct ocfs2_xattr_search *xs,
5447 struct ocfs2_xattr_set_ctxt *ctxt)
5450 struct ocfs2_xa_loc loc;
5452 mlog_entry("Set xattr %s in xattr bucket\n", xi->xi_name);
5454 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5455 xs->not_found ? NULL : xs->here);
5456 ret = ocfs2_xa_set(&loc, xi, ctxt);
5458 xs->here = loc.xl_entry;
5461 if (ret != -ENOSPC) {
5466 /* Ok, we need space. Let's try defragmenting the bucket. */
5467 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5474 ret = ocfs2_xa_set(&loc, xi, ctxt);
5476 xs->here = loc.xl_entry;
5488 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5489 struct ocfs2_xattr_info *xi,
5490 struct ocfs2_xattr_search *xs,
5491 struct ocfs2_xattr_set_ctxt *ctxt)
5495 mlog_entry("Set xattr %s in xattr index block\n", xi->xi_name);
5497 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5500 if (ret != -ENOSPC) {
5505 /* Ack, need more space. Let's try to get another bucket! */
5508 * We do not allow for overlapping ranges between buckets. And
5509 * the maximum number of collisions we will allow for then is
5510 * one bucket's worth, so check it here whether we need to
5511 * add a new bucket for the insert.
5513 ret = ocfs2_check_xattr_bucket_collision(inode,
5521 ret = ocfs2_add_new_xattr_bucket(inode,
5531 * ocfs2_add_new_xattr_bucket() will have updated
5532 * xs->bucket if it moved, but it will not have updated
5533 * any of the other search fields. Thus, we drop it and
5534 * re-search. Everything should be cached, so it'll be
5537 ocfs2_xattr_bucket_relse(xs->bucket);
5538 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5541 if (ret && ret != -ENODATA)
5543 xs->not_found = ret;
5545 /* Ok, we have a new bucket, let's try again */
5546 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5547 if (ret && (ret != -ENOSPC))
5555 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5556 struct ocfs2_xattr_bucket *bucket,
5559 int ret = 0, ref_credits;
5560 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5562 struct ocfs2_xattr_entry *xe;
5563 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5564 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5565 int credits = ocfs2_remove_extent_credits(osb->sb) +
5566 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5567 struct ocfs2_xattr_value_root *xv;
5568 struct ocfs2_rm_xattr_bucket_para *args =
5569 (struct ocfs2_rm_xattr_bucket_para *)para;
5571 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5573 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5574 xe = &xh->xh_entries[i];
5575 if (ocfs2_xattr_is_local(xe))
5578 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5581 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5587 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5588 if (IS_ERR(ctxt.handle)) {
5589 ret = PTR_ERR(ctxt.handle);
5594 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5597 ocfs2_commit_trans(osb, ctxt.handle);
5599 ocfs2_free_alloc_context(ctxt.meta_ac);
5600 ctxt.meta_ac = NULL;
5609 ocfs2_free_alloc_context(ctxt.meta_ac);
5610 ocfs2_schedule_truncate_log_flush(osb, 1);
5611 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5616 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5617 * or change the extent record flag), we need to recalculate
5618 * the metaecc for the whole bucket. So it is done here.
5621 * We have to give the extra credits for the caller.
5623 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5628 struct ocfs2_xattr_bucket *bucket =
5629 (struct ocfs2_xattr_bucket *)para;
5631 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5632 OCFS2_JOURNAL_ACCESS_WRITE);
5638 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5644 * Special action we need if the xattr value is refcounted.
5646 * 1. If the xattr is refcounted, lock the tree.
5647 * 2. CoW the xattr if we are setting the new value and the value
5648 * will be stored outside.
5649 * 3. In other case, decrease_refcount will work for us, so just
5650 * lock the refcount tree, calculate the meta and credits is OK.
5652 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5653 * currently CoW is a completed transaction, while this function
5654 * will also lock the allocators and let us deadlock. So we will
5655 * CoW the whole xattr value.
5657 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5658 struct ocfs2_dinode *di,
5659 struct ocfs2_xattr_info *xi,
5660 struct ocfs2_xattr_search *xis,
5661 struct ocfs2_xattr_search *xbs,
5662 struct ocfs2_refcount_tree **ref_tree,
5667 struct ocfs2_xattr_block *xb;
5668 struct ocfs2_xattr_entry *xe;
5670 u32 p_cluster, num_clusters;
5671 unsigned int ext_flags;
5672 int name_offset, name_len;
5673 struct ocfs2_xattr_value_buf vb;
5674 struct ocfs2_xattr_bucket *bucket = NULL;
5675 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5676 struct ocfs2_post_refcount refcount;
5677 struct ocfs2_post_refcount *p = NULL;
5678 struct buffer_head *ref_root_bh = NULL;
5680 if (!xis->not_found) {
5682 name_offset = le16_to_cpu(xe->xe_name_offset);
5683 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5685 vb.vb_bh = xis->inode_bh;
5686 vb.vb_access = ocfs2_journal_access_di;
5688 int i, block_off = 0;
5689 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5691 name_offset = le16_to_cpu(xe->xe_name_offset);
5692 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5693 i = xbs->here - xbs->header->xh_entries;
5695 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5696 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5697 bucket_xh(xbs->bucket),
5704 base = bucket_block(xbs->bucket, block_off);
5705 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5706 vb.vb_access = ocfs2_journal_access;
5708 if (ocfs2_meta_ecc(osb)) {
5709 /*create parameters for ocfs2_post_refcount. */
5710 bucket = xbs->bucket;
5711 refcount.credits = bucket->bu_blocks;
5712 refcount.para = bucket;
5714 ocfs2_xattr_bucket_post_refcount;
5719 vb.vb_bh = xbs->xattr_bh;
5720 vb.vb_access = ocfs2_journal_access_xb;
5724 if (ocfs2_xattr_is_local(xe))
5727 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5728 (base + name_offset + name_len);
5730 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5731 &num_clusters, &vb.vb_xv->xr_list,
5739 * We just need to check the 1st extent record, since we always
5740 * CoW the whole xattr. So there shouldn't be a xattr with
5741 * some REFCOUNT extent recs after the 1st one.
5743 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5746 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5747 1, ref_tree, &ref_root_bh);
5754 * If we are deleting the xattr or the new size will be stored inside,
5755 * cool, leave it there, the xattr truncate process will remove them
5756 * for us(it still needs the refcount tree lock and the meta, credits).
5757 * And the worse case is that every cluster truncate will split the
5758 * refcount tree, and make the original extent become 3. So we will need
5759 * 2 * cluster more extent recs at most.
5761 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5763 ret = ocfs2_refcounted_xattr_delete_need(inode,
5764 &(*ref_tree)->rf_ci,
5765 ref_root_bh, vb.vb_xv,
5772 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5773 *ref_tree, ref_root_bh, 0,
5774 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5779 brelse(ref_root_bh);
5784 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5785 * The physical clusters will be added to refcount tree.
5787 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5788 struct ocfs2_xattr_value_root *xv,
5789 struct ocfs2_extent_tree *value_et,
5790 struct ocfs2_caching_info *ref_ci,
5791 struct buffer_head *ref_root_bh,
5792 struct ocfs2_cached_dealloc_ctxt *dealloc,
5793 struct ocfs2_post_refcount *refcount)
5796 u32 clusters = le32_to_cpu(xv->xr_clusters);
5797 u32 cpos, p_cluster, num_clusters;
5798 struct ocfs2_extent_list *el = &xv->xr_list;
5799 unsigned int ext_flags;
5802 while (cpos < clusters) {
5803 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5804 &num_clusters, el, &ext_flags);
5806 cpos += num_clusters;
5807 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5812 ret = ocfs2_add_refcount_flag(inode, value_et,
5813 ref_ci, ref_root_bh,
5814 cpos - num_clusters,
5815 p_cluster, num_clusters,
5827 * Given a normal ocfs2_xattr_header, refcount all the entries which
5828 * have value stored outside.
5829 * Used for xattrs stored in inode and ocfs2_xattr_block.
5831 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5832 struct ocfs2_xattr_value_buf *vb,
5833 struct ocfs2_xattr_header *header,
5834 struct ocfs2_caching_info *ref_ci,
5835 struct buffer_head *ref_root_bh,
5836 struct ocfs2_cached_dealloc_ctxt *dealloc)
5839 struct ocfs2_xattr_entry *xe;
5840 struct ocfs2_xattr_value_root *xv;
5841 struct ocfs2_extent_tree et;
5844 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5845 xe = &header->xh_entries[i];
5847 if (ocfs2_xattr_is_local(xe))
5850 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5851 le16_to_cpu(xe->xe_name_offset) +
5852 OCFS2_XATTR_SIZE(xe->xe_name_len));
5855 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5857 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5858 ref_ci, ref_root_bh,
5869 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5870 struct buffer_head *fe_bh,
5871 struct ocfs2_caching_info *ref_ci,
5872 struct buffer_head *ref_root_bh,
5873 struct ocfs2_cached_dealloc_ctxt *dealloc)
5875 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5876 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5877 (fe_bh->b_data + inode->i_sb->s_blocksize -
5878 le16_to_cpu(di->i_xattr_inline_size));
5879 struct ocfs2_xattr_value_buf vb = {
5881 .vb_access = ocfs2_journal_access_di,
5884 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5885 ref_ci, ref_root_bh, dealloc);
5888 struct ocfs2_xattr_tree_value_refcount_para {
5889 struct ocfs2_caching_info *ref_ci;
5890 struct buffer_head *ref_root_bh;
5891 struct ocfs2_cached_dealloc_ctxt *dealloc;
5894 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5895 struct ocfs2_xattr_bucket *bucket,
5897 struct ocfs2_xattr_value_root **xv,
5898 struct buffer_head **bh)
5900 int ret, block_off, name_offset;
5901 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5902 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5905 ret = ocfs2_xattr_bucket_get_name_value(sb,
5915 base = bucket_block(bucket, block_off);
5917 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
5918 OCFS2_XATTR_SIZE(xe->xe_name_len));
5921 *bh = bucket->bu_bhs[block_off];
5927 * For a given xattr bucket, refcount all the entries which
5928 * have value stored outside.
5930 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
5931 struct ocfs2_xattr_bucket *bucket,
5935 struct ocfs2_extent_tree et;
5936 struct ocfs2_xattr_tree_value_refcount_para *ref =
5937 (struct ocfs2_xattr_tree_value_refcount_para *)para;
5938 struct ocfs2_xattr_header *xh =
5939 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
5940 struct ocfs2_xattr_entry *xe;
5941 struct ocfs2_xattr_value_buf vb = {
5942 .vb_access = ocfs2_journal_access,
5944 struct ocfs2_post_refcount refcount = {
5945 .credits = bucket->bu_blocks,
5947 .func = ocfs2_xattr_bucket_post_refcount,
5949 struct ocfs2_post_refcount *p = NULL;
5951 /* We only need post_refcount if we support metaecc. */
5952 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
5955 mlog(0, "refcount bucket %llu, count = %u\n",
5956 (unsigned long long)bucket_blkno(bucket),
5957 le16_to_cpu(xh->xh_count));
5958 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5959 xe = &xh->xh_entries[i];
5961 if (ocfs2_xattr_is_local(xe))
5964 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
5965 &vb.vb_xv, &vb.vb_bh);
5971 ocfs2_init_xattr_value_extent_tree(&et,
5972 INODE_CACHE(inode), &vb);
5974 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
5988 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
5989 struct buffer_head *root_bh,
5990 u64 blkno, u32 cpos, u32 len, void *para)
5992 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
5993 ocfs2_xattr_bucket_value_refcount,
5997 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
5998 struct buffer_head *blk_bh,
5999 struct ocfs2_caching_info *ref_ci,
6000 struct buffer_head *ref_root_bh,
6001 struct ocfs2_cached_dealloc_ctxt *dealloc)
6004 struct ocfs2_xattr_block *xb =
6005 (struct ocfs2_xattr_block *)blk_bh->b_data;
6007 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6008 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6009 struct ocfs2_xattr_value_buf vb = {
6011 .vb_access = ocfs2_journal_access_xb,
6014 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6015 ref_ci, ref_root_bh,
6018 struct ocfs2_xattr_tree_value_refcount_para para = {
6020 .ref_root_bh = ref_root_bh,
6024 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6025 ocfs2_refcount_xattr_tree_rec,
6032 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6033 struct buffer_head *fe_bh,
6034 struct ocfs2_caching_info *ref_ci,
6035 struct buffer_head *ref_root_bh,
6036 struct ocfs2_cached_dealloc_ctxt *dealloc)
6039 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6040 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6041 struct buffer_head *blk_bh = NULL;
6043 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6044 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6045 ref_ci, ref_root_bh,
6053 if (!di->i_xattr_loc)
6056 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6063 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6064 ref_root_bh, dealloc);
6074 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6076 * Store the information we need in xattr reflink.
6077 * old_bh and new_bh are inode bh for the old and new inode.
6079 struct ocfs2_xattr_reflink {
6080 struct inode *old_inode;
6081 struct inode *new_inode;
6082 struct buffer_head *old_bh;
6083 struct buffer_head *new_bh;
6084 struct ocfs2_caching_info *ref_ci;
6085 struct buffer_head *ref_root_bh;
6086 struct ocfs2_cached_dealloc_ctxt *dealloc;
6087 should_xattr_reflinked *xattr_reflinked;
6091 * Given a xattr header and xe offset,
6092 * return the proper xv and the corresponding bh.
6093 * xattr in inode, block and xattr tree have different implementaions.
6095 typedef int (get_xattr_value_root)(struct super_block *sb,
6096 struct buffer_head *bh,
6097 struct ocfs2_xattr_header *xh,
6099 struct ocfs2_xattr_value_root **xv,
6100 struct buffer_head **ret_bh,
6104 * Calculate all the xattr value root metadata stored in this xattr header and
6105 * credits we need if we create them from the scratch.
6106 * We use get_xattr_value_root so that all types of xattr container can use it.
6108 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6109 struct buffer_head *bh,
6110 struct ocfs2_xattr_header *xh,
6111 int *metas, int *credits,
6113 get_xattr_value_root *func,
6117 struct ocfs2_xattr_value_root *xv;
6118 struct ocfs2_xattr_entry *xe;
6120 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6121 xe = &xh->xh_entries[i];
6122 if (ocfs2_xattr_is_local(xe))
6125 ret = func(sb, bh, xh, i, &xv, NULL, para);
6131 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6132 le16_to_cpu(xv->xr_list.l_next_free_rec);
6134 *credits += ocfs2_calc_extend_credits(sb,
6136 le32_to_cpu(xv->xr_clusters));
6139 * If the value is a tree with depth > 1, We don't go deep
6140 * to the extent block, so just calculate a maximum record num.
6142 if (!xv->xr_list.l_tree_depth)
6143 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6145 *num_recs += ocfs2_clusters_for_bytes(sb,
6152 /* Used by xattr inode and block to return the right xv and buffer_head. */
6153 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6154 struct buffer_head *bh,
6155 struct ocfs2_xattr_header *xh,
6157 struct ocfs2_xattr_value_root **xv,
6158 struct buffer_head **ret_bh,
6161 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6163 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6164 le16_to_cpu(xe->xe_name_offset) +
6165 OCFS2_XATTR_SIZE(xe->xe_name_len));
6174 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6175 * It is only used for inline xattr and xattr block.
6177 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6178 struct ocfs2_xattr_header *xh,
6179 struct buffer_head *ref_root_bh,
6181 struct ocfs2_alloc_context **meta_ac)
6183 int ret, meta_add = 0, num_recs = 0;
6184 struct ocfs2_refcount_block *rb =
6185 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6189 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6190 &meta_add, credits, &num_recs,
6191 ocfs2_get_xattr_value_root,
6199 * We need to add/modify num_recs in refcount tree, so just calculate
6200 * an approximate number we need for refcount tree change.
6201 * Sometimes we need to split the tree, and after split, half recs
6202 * will be moved to the new block, and a new block can only provide
6203 * half number of recs. So we multiple new blocks by 2.
6205 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6206 meta_add += num_recs;
6207 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6208 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6209 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6210 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6214 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6223 * Given a xattr header, reflink all the xattrs in this container.
6224 * It can be used for inode, block and bucket.
6227 * Before we call this function, the caller has memcpy the xattr in
6228 * old_xh to the new_xh.
6230 * If args.xattr_reflinked is set, call it to decide whether the xe should
6231 * be reflinked or not. If not, remove it from the new xattr header.
6233 static int ocfs2_reflink_xattr_header(handle_t *handle,
6234 struct ocfs2_xattr_reflink *args,
6235 struct buffer_head *old_bh,
6236 struct ocfs2_xattr_header *xh,
6237 struct buffer_head *new_bh,
6238 struct ocfs2_xattr_header *new_xh,
6239 struct ocfs2_xattr_value_buf *vb,
6240 struct ocfs2_alloc_context *meta_ac,
6241 get_xattr_value_root *func,
6245 struct super_block *sb = args->old_inode->i_sb;
6246 struct buffer_head *value_bh;
6247 struct ocfs2_xattr_entry *xe, *last;
6248 struct ocfs2_xattr_value_root *xv, *new_xv;
6249 struct ocfs2_extent_tree data_et;
6250 u32 clusters, cpos, p_cluster, num_clusters;
6251 unsigned int ext_flags = 0;
6253 mlog(0, "reflink xattr in container %llu, count = %u\n",
6254 (unsigned long long)old_bh->b_blocknr, le16_to_cpu(xh->xh_count));
6256 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6257 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6258 xe = &xh->xh_entries[i];
6260 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6261 xe = &new_xh->xh_entries[j];
6263 le16_add_cpu(&new_xh->xh_count, -1);
6264 if (new_xh->xh_count) {
6266 (void *)last - (void *)xe);
6268 sizeof(struct ocfs2_xattr_entry));
6272 * We don't want j to increase in the next round since
6273 * it is already moved ahead.
6279 if (ocfs2_xattr_is_local(xe))
6282 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6288 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6295 * For the xattr which has l_tree_depth = 0, all the extent
6296 * recs have already be copied to the new xh with the
6297 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6298 * increase the refount count int the refcount tree.
6300 * For the xattr which has l_tree_depth > 0, we need
6301 * to initialize it to the empty default value root,
6302 * and then insert the extents one by one.
6304 if (xv->xr_list.l_tree_depth) {
6305 memcpy(new_xv, &def_xv, sizeof(def_xv));
6307 vb->vb_bh = value_bh;
6308 ocfs2_init_xattr_value_extent_tree(&data_et,
6309 INODE_CACHE(args->new_inode), vb);
6312 clusters = le32_to_cpu(xv->xr_clusters);
6314 while (cpos < clusters) {
6315 ret = ocfs2_xattr_get_clusters(args->old_inode,
6328 if (xv->xr_list.l_tree_depth) {
6329 ret = ocfs2_insert_extent(handle,
6331 ocfs2_clusters_to_blocks(
6332 args->old_inode->i_sb,
6334 num_clusters, ext_flags,
6342 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6344 p_cluster, num_clusters,
6345 meta_ac, args->dealloc);
6351 cpos += num_clusters;
6359 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6361 int ret = 0, credits = 0;
6363 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6364 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6365 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6366 int header_off = osb->sb->s_blocksize - inline_size;
6367 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6368 (args->old_bh->b_data + header_off);
6369 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6370 (args->new_bh->b_data + header_off);
6371 struct ocfs2_alloc_context *meta_ac = NULL;
6372 struct ocfs2_inode_info *new_oi;
6373 struct ocfs2_dinode *new_di;
6374 struct ocfs2_xattr_value_buf vb = {
6375 .vb_bh = args->new_bh,
6376 .vb_access = ocfs2_journal_access_di,
6379 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6380 &credits, &meta_ac);
6386 handle = ocfs2_start_trans(osb, credits);
6387 if (IS_ERR(handle)) {
6388 ret = PTR_ERR(handle);
6393 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6394 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6400 memcpy(args->new_bh->b_data + header_off,
6401 args->old_bh->b_data + header_off, inline_size);
6403 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6404 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6406 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6407 args->new_bh, new_xh, &vb, meta_ac,
6408 ocfs2_get_xattr_value_root, NULL);
6414 new_oi = OCFS2_I(args->new_inode);
6415 spin_lock(&new_oi->ip_lock);
6416 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6417 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6418 spin_unlock(&new_oi->ip_lock);
6420 ocfs2_journal_dirty(handle, args->new_bh);
6423 ocfs2_commit_trans(osb, handle);
6427 ocfs2_free_alloc_context(meta_ac);
6431 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6432 struct buffer_head *fe_bh,
6433 struct buffer_head **ret_bh,
6438 struct ocfs2_alloc_context *meta_ac;
6439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6441 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
6447 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6448 if (IS_ERR(handle)) {
6449 ret = PTR_ERR(handle);
6454 mlog(0, "create new xattr block for inode %llu, index = %d\n",
6455 (unsigned long long)fe_bh->b_blocknr, indexed);
6456 ret = ocfs2_create_xattr_block(handle, inode, fe_bh,
6457 meta_ac, ret_bh, indexed);
6461 ocfs2_commit_trans(osb, handle);
6463 ocfs2_free_alloc_context(meta_ac);
6467 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6468 struct buffer_head *blk_bh,
6469 struct buffer_head *new_blk_bh)
6471 int ret = 0, credits = 0;
6473 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6474 struct ocfs2_dinode *new_di;
6475 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6476 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6477 struct ocfs2_xattr_block *xb =
6478 (struct ocfs2_xattr_block *)blk_bh->b_data;
6479 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6480 struct ocfs2_xattr_block *new_xb =
6481 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6482 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6483 struct ocfs2_alloc_context *meta_ac;
6484 struct ocfs2_xattr_value_buf vb = {
6485 .vb_bh = new_blk_bh,
6486 .vb_access = ocfs2_journal_access_xb,
6489 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6490 &credits, &meta_ac);
6496 /* One more credits in case we need to add xattr flags in new inode. */
6497 handle = ocfs2_start_trans(osb, credits + 1);
6498 if (IS_ERR(handle)) {
6499 ret = PTR_ERR(handle);
6504 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6505 ret = ocfs2_journal_access_di(handle,
6506 INODE_CACHE(args->new_inode),
6508 OCFS2_JOURNAL_ACCESS_WRITE);
6515 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6516 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6522 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6523 osb->sb->s_blocksize - header_off);
6525 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6526 new_blk_bh, new_xh, &vb, meta_ac,
6527 ocfs2_get_xattr_value_root, NULL);
6533 ocfs2_journal_dirty(handle, new_blk_bh);
6535 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6536 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6537 spin_lock(&new_oi->ip_lock);
6538 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6539 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6540 spin_unlock(&new_oi->ip_lock);
6542 ocfs2_journal_dirty(handle, args->new_bh);
6546 ocfs2_commit_trans(osb, handle);
6549 ocfs2_free_alloc_context(meta_ac);
6553 struct ocfs2_reflink_xattr_tree_args {
6554 struct ocfs2_xattr_reflink *reflink;
6555 struct buffer_head *old_blk_bh;
6556 struct buffer_head *new_blk_bh;
6557 struct ocfs2_xattr_bucket *old_bucket;
6558 struct ocfs2_xattr_bucket *new_bucket;
6563 * We have to handle the case that both old bucket and new bucket
6564 * will call this function to get the right ret_bh.
6565 * So The caller must give us the right bh.
6567 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6568 struct buffer_head *bh,
6569 struct ocfs2_xattr_header *xh,
6571 struct ocfs2_xattr_value_root **xv,
6572 struct buffer_head **ret_bh,
6575 struct ocfs2_reflink_xattr_tree_args *args =
6576 (struct ocfs2_reflink_xattr_tree_args *)para;
6577 struct ocfs2_xattr_bucket *bucket;
6579 if (bh == args->old_bucket->bu_bhs[0])
6580 bucket = args->old_bucket;
6582 bucket = args->new_bucket;
6584 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6588 struct ocfs2_value_tree_metas {
6594 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6595 struct buffer_head *bh,
6596 struct ocfs2_xattr_header *xh,
6598 struct ocfs2_xattr_value_root **xv,
6599 struct buffer_head **ret_bh,
6602 struct ocfs2_xattr_bucket *bucket =
6603 (struct ocfs2_xattr_bucket *)para;
6605 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6609 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6610 struct ocfs2_xattr_bucket *bucket,
6613 struct ocfs2_value_tree_metas *metas =
6614 (struct ocfs2_value_tree_metas *)para;
6615 struct ocfs2_xattr_header *xh =
6616 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6618 /* Add the credits for this bucket first. */
6619 metas->credits += bucket->bu_blocks;
6620 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6621 xh, &metas->num_metas,
6622 &metas->credits, &metas->num_recs,
6623 ocfs2_value_tree_metas_in_bucket,
6628 * Given a xattr extent rec starting from blkno and having len clusters,
6629 * iterate all the buckets calculate how much metadata we need for reflinking
6630 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6632 static int ocfs2_lock_reflink_xattr_rec_allocators(
6633 struct ocfs2_reflink_xattr_tree_args *args,
6634 struct ocfs2_extent_tree *xt_et,
6635 u64 blkno, u32 len, int *credits,
6636 struct ocfs2_alloc_context **meta_ac,
6637 struct ocfs2_alloc_context **data_ac)
6639 int ret, num_free_extents;
6640 struct ocfs2_value_tree_metas metas;
6641 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6642 struct ocfs2_refcount_block *rb;
6644 memset(&metas, 0, sizeof(metas));
6646 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6647 ocfs2_calc_value_tree_metas, &metas);
6653 *credits = metas.credits;
6656 * Calculate we need for refcount tree change.
6658 * We need to add/modify num_recs in refcount tree, so just calculate
6659 * an approximate number we need for refcount tree change.
6660 * Sometimes we need to split the tree, and after split, half recs
6661 * will be moved to the new block, and a new block can only provide
6662 * half number of recs. So we multiple new blocks by 2.
6663 * In the end, we have to add credits for modifying the already
6664 * existed refcount block.
6666 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6668 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6669 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6670 metas.num_metas += metas.num_recs;
6671 *credits += metas.num_recs +
6672 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6673 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6674 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6675 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6679 /* count in the xattr tree change. */
6680 num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6681 if (num_free_extents < 0) {
6682 ret = num_free_extents;
6687 if (num_free_extents < len)
6688 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6690 *credits += ocfs2_calc_extend_credits(osb->sb,
6691 xt_et->et_root_el, len);
6693 if (metas.num_metas) {
6694 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6703 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6710 ocfs2_free_alloc_context(*meta_ac);
6718 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6719 u64 blkno, u64 new_blkno, u32 clusters,
6720 struct ocfs2_alloc_context *meta_ac,
6721 struct ocfs2_alloc_context *data_ac,
6722 struct ocfs2_reflink_xattr_tree_args *args)
6725 struct super_block *sb = args->reflink->old_inode->i_sb;
6726 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
6727 u32 num_buckets = clusters * bpc;
6728 int bpb = args->old_bucket->bu_blocks;
6729 struct ocfs2_xattr_value_buf vb = {
6730 .vb_access = ocfs2_journal_access,
6733 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6734 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6740 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6747 * The real bucket num in this series of blocks is stored
6748 * in the 1st bucket.
6751 num_buckets = le16_to_cpu(
6752 bucket_xh(args->old_bucket)->xh_num_buckets);
6754 ret = ocfs2_xattr_bucket_journal_access(handle,
6756 OCFS2_JOURNAL_ACCESS_CREATE);
6762 for (j = 0; j < bpb; j++)
6763 memcpy(bucket_block(args->new_bucket, j),
6764 bucket_block(args->old_bucket, j),
6767 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6769 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6770 args->old_bucket->bu_bhs[0],
6771 bucket_xh(args->old_bucket),
6772 args->new_bucket->bu_bhs[0],
6773 bucket_xh(args->new_bucket),
6775 ocfs2_get_reflink_xattr_value_root,
6783 * Re-access and dirty the bucket to calculate metaecc.
6784 * Because we may extend the transaction in reflink_xattr_header
6785 * which will let the already accessed block gone.
6787 ret = ocfs2_xattr_bucket_journal_access(handle,
6789 OCFS2_JOURNAL_ACCESS_WRITE);
6795 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6796 ocfs2_xattr_bucket_relse(args->old_bucket);
6797 ocfs2_xattr_bucket_relse(args->new_bucket);
6800 ocfs2_xattr_bucket_relse(args->old_bucket);
6801 ocfs2_xattr_bucket_relse(args->new_bucket);
6805 * Create the same xattr extent record in the new inode's xattr tree.
6807 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6808 struct buffer_head *root_bh,
6814 int ret, credits = 0;
6815 u32 p_cluster, num_clusters;
6818 struct ocfs2_reflink_xattr_tree_args *args =
6819 (struct ocfs2_reflink_xattr_tree_args *)para;
6820 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6821 struct ocfs2_alloc_context *meta_ac = NULL;
6822 struct ocfs2_alloc_context *data_ac = NULL;
6823 struct ocfs2_extent_tree et;
6825 ocfs2_init_xattr_tree_extent_tree(&et,
6826 INODE_CACHE(args->reflink->new_inode),
6829 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6831 &meta_ac, &data_ac);
6837 handle = ocfs2_start_trans(osb, credits);
6838 if (IS_ERR(handle)) {
6839 ret = PTR_ERR(handle);
6844 ret = ocfs2_claim_clusters(osb, handle, data_ac,
6845 len, &p_cluster, &num_clusters);
6851 new_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cluster);
6853 mlog(0, "reflink xattr buckets %llu to %llu, len %u\n",
6854 (unsigned long long)blkno, (unsigned long long)new_blkno, len);
6855 ret = ocfs2_reflink_xattr_buckets(handle, blkno, new_blkno, len,
6856 meta_ac, data_ac, args);
6862 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
6863 (unsigned long long)new_blkno, len, cpos);
6864 ret = ocfs2_insert_extent(handle, &et, cpos, new_blkno,
6870 ocfs2_commit_trans(osb, handle);
6874 ocfs2_free_alloc_context(meta_ac);
6876 ocfs2_free_alloc_context(data_ac);
6881 * Create reflinked xattr buckets.
6882 * We will add bucket one by one, and refcount all the xattrs in the bucket
6883 * if they are stored outside.
6885 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
6886 struct buffer_head *blk_bh,
6887 struct buffer_head *new_blk_bh)
6890 struct ocfs2_reflink_xattr_tree_args para;
6892 memset(¶, 0, sizeof(para));
6893 para.reflink = args;
6894 para.old_blk_bh = blk_bh;
6895 para.new_blk_bh = new_blk_bh;
6897 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
6898 if (!para.old_bucket) {
6899 mlog_errno(-ENOMEM);
6903 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
6904 if (!para.new_bucket) {
6910 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
6911 ocfs2_reflink_xattr_rec,
6917 ocfs2_xattr_bucket_free(para.old_bucket);
6918 ocfs2_xattr_bucket_free(para.new_bucket);
6922 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
6923 struct buffer_head *blk_bh)
6925 int ret, indexed = 0;
6926 struct buffer_head *new_blk_bh = NULL;
6927 struct ocfs2_xattr_block *xb =
6928 (struct ocfs2_xattr_block *)blk_bh->b_data;
6931 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
6934 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
6935 &new_blk_bh, indexed);
6941 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED))
6942 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
6944 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
6953 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
6955 int type = ocfs2_xattr_get_type(xe);
6957 return type != OCFS2_XATTR_INDEX_SECURITY &&
6958 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
6959 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
6962 int ocfs2_reflink_xattrs(struct inode *old_inode,
6963 struct buffer_head *old_bh,
6964 struct inode *new_inode,
6965 struct buffer_head *new_bh,
6966 bool preserve_security)
6969 struct ocfs2_xattr_reflink args;
6970 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
6971 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
6972 struct buffer_head *blk_bh = NULL;
6973 struct ocfs2_cached_dealloc_ctxt dealloc;
6974 struct ocfs2_refcount_tree *ref_tree;
6975 struct buffer_head *ref_root_bh = NULL;
6977 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
6978 le64_to_cpu(di->i_refcount_loc),
6979 1, &ref_tree, &ref_root_bh);
6985 ocfs2_init_dealloc_ctxt(&dealloc);
6987 args.old_inode = old_inode;
6988 args.new_inode = new_inode;
6989 args.old_bh = old_bh;
6990 args.new_bh = new_bh;
6991 args.ref_ci = &ref_tree->rf_ci;
6992 args.ref_root_bh = ref_root_bh;
6993 args.dealloc = &dealloc;
6994 if (preserve_security)
6995 args.xattr_reflinked = NULL;
6997 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
6999 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7000 ret = ocfs2_reflink_xattr_inline(&args);
7007 if (!di->i_xattr_loc)
7010 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7017 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7024 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7026 brelse(ref_root_bh);
7028 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7029 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7030 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7038 * Initialize security and acl for a already created inode.
7039 * Used for reflink a non-preserve-security file.
7041 * It uses common api like ocfs2_xattr_set, so the caller
7042 * must not hold any lock expect i_mutex.
7044 int ocfs2_init_security_and_acl(struct inode *dir,
7045 struct inode *inode)
7048 struct buffer_head *dir_bh = NULL;
7049 struct ocfs2_security_xattr_info si = {
7053 ret = ocfs2_init_security_get(inode, dir, &si);
7055 ret = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7056 si.name, si.value, si.value_len,
7062 } else if (ret != -EOPNOTSUPP) {
7067 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7073 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7077 ocfs2_inode_unlock(dir, 0);
7083 * 'security' attributes support
7085 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7086 size_t list_size, const char *name,
7087 size_t name_len, int type)
7089 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7090 const size_t total_len = prefix_len + name_len + 1;
7092 if (list && total_len <= list_size) {
7093 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7094 memcpy(list + prefix_len, name, name_len);
7095 list[prefix_len + name_len] = '\0';
7100 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7101 void *buffer, size_t size, int type)
7103 if (strcmp(name, "") == 0)
7105 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7106 name, buffer, size);
7109 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7110 const void *value, size_t size, int flags, int type)
7112 if (strcmp(name, "") == 0)
7115 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7116 name, value, size, flags);
7119 int ocfs2_init_security_get(struct inode *inode,
7121 struct ocfs2_security_xattr_info *si)
7123 /* check whether ocfs2 support feature xattr */
7124 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7126 return security_inode_init_security(inode, dir, &si->name, &si->value,
7130 int ocfs2_init_security_set(handle_t *handle,
7131 struct inode *inode,
7132 struct buffer_head *di_bh,
7133 struct ocfs2_security_xattr_info *si,
7134 struct ocfs2_alloc_context *xattr_ac,
7135 struct ocfs2_alloc_context *data_ac)
7137 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7138 OCFS2_XATTR_INDEX_SECURITY,
7139 si->name, si->value, si->value_len, 0,
7143 struct xattr_handler ocfs2_xattr_security_handler = {
7144 .prefix = XATTR_SECURITY_PREFIX,
7145 .list = ocfs2_xattr_security_list,
7146 .get = ocfs2_xattr_security_get,
7147 .set = ocfs2_xattr_security_set,
7151 * 'trusted' attributes support
7153 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7154 size_t list_size, const char *name,
7155 size_t name_len, int type)
7157 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7158 const size_t total_len = prefix_len + name_len + 1;
7160 if (list && total_len <= list_size) {
7161 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7162 memcpy(list + prefix_len, name, name_len);
7163 list[prefix_len + name_len] = '\0';
7168 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7169 void *buffer, size_t size, int type)
7171 if (strcmp(name, "") == 0)
7173 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7174 name, buffer, size);
7177 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7178 const void *value, size_t size, int flags, int type)
7180 if (strcmp(name, "") == 0)
7183 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7184 name, value, size, flags);
7187 struct xattr_handler ocfs2_xattr_trusted_handler = {
7188 .prefix = XATTR_TRUSTED_PREFIX,
7189 .list = ocfs2_xattr_trusted_list,
7190 .get = ocfs2_xattr_trusted_get,
7191 .set = ocfs2_xattr_trusted_set,
7195 * 'user' attributes support
7197 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7198 size_t list_size, const char *name,
7199 size_t name_len, int type)
7201 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7202 const size_t total_len = prefix_len + name_len + 1;
7203 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7205 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7208 if (list && total_len <= list_size) {
7209 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7210 memcpy(list + prefix_len, name, name_len);
7211 list[prefix_len + name_len] = '\0';
7216 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7217 void *buffer, size_t size, int type)
7219 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7221 if (strcmp(name, "") == 0)
7223 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7225 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7229 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7230 const void *value, size_t size, int flags, int type)
7232 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7234 if (strcmp(name, "") == 0)
7236 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7239 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7240 name, value, size, flags);
7243 struct xattr_handler ocfs2_xattr_user_handler = {
7244 .prefix = XATTR_USER_PREFIX,
7245 .list = ocfs2_xattr_user_list,
7246 .get = ocfs2_xattr_user_get,
7247 .set = ocfs2_xattr_user_set,