1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2009 Oracle. All rights reserved.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public
10 * License version 2 as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
18 #include <linux/sort.h>
19 #define MLOG_MASK_PREFIX ML_REFCOUNT
20 #include <cluster/masklog.h>
28 #include "buffer_head_io.h"
29 #include "blockcheck.h"
30 #include "refcounttree.h"
33 #include "extent_map.h"
38 #include <linux/bio.h>
39 #include <linux/blkdev.h>
40 #include <linux/gfp.h>
41 #include <linux/slab.h>
42 #include <linux/writeback.h>
43 #include <linux/pagevec.h>
44 #include <linux/swap.h>
45 #include <linux/security.h>
46 #include <linux/fsnotify.h>
47 #include <linux/quotaops.h>
48 #include <linux/namei.h>
49 #include <linux/mount.h>
51 struct ocfs2_cow_context
{
55 struct ocfs2_extent_tree data_et
;
56 struct ocfs2_refcount_tree
*ref_tree
;
57 struct buffer_head
*ref_root_bh
;
58 struct ocfs2_alloc_context
*meta_ac
;
59 struct ocfs2_alloc_context
*data_ac
;
60 struct ocfs2_cached_dealloc_ctxt dealloc
;
62 struct ocfs2_post_refcount
*post_refcount
;
64 int (*get_clusters
)(struct ocfs2_cow_context
*context
,
65 u32 v_cluster
, u32
*p_cluster
,
67 unsigned int *extent_flags
);
68 int (*cow_duplicate_clusters
)(handle_t
*handle
,
69 struct ocfs2_cow_context
*context
,
70 u32 cpos
, u32 old_cluster
,
71 u32 new_cluster
, u32 new_len
);
74 static inline struct ocfs2_refcount_tree
*
75 cache_info_to_refcount(struct ocfs2_caching_info
*ci
)
77 return container_of(ci
, struct ocfs2_refcount_tree
, rf_ci
);
80 static int ocfs2_validate_refcount_block(struct super_block
*sb
,
81 struct buffer_head
*bh
)
84 struct ocfs2_refcount_block
*rb
=
85 (struct ocfs2_refcount_block
*)bh
->b_data
;
87 mlog(0, "Validating refcount block %llu\n",
88 (unsigned long long)bh
->b_blocknr
);
90 BUG_ON(!buffer_uptodate(bh
));
93 * If the ecc fails, we return the error but otherwise
94 * leave the filesystem running. We know any error is
95 * local to this block.
97 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &rb
->rf_check
);
99 mlog(ML_ERROR
, "Checksum failed for refcount block %llu\n",
100 (unsigned long long)bh
->b_blocknr
);
105 if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb
)) {
107 "Refcount block #%llu has bad signature %.*s",
108 (unsigned long long)bh
->b_blocknr
, 7,
113 if (le64_to_cpu(rb
->rf_blkno
) != bh
->b_blocknr
) {
115 "Refcount block #%llu has an invalid rf_blkno "
117 (unsigned long long)bh
->b_blocknr
,
118 (unsigned long long)le64_to_cpu(rb
->rf_blkno
));
122 if (le32_to_cpu(rb
->rf_fs_generation
) != OCFS2_SB(sb
)->fs_generation
) {
124 "Refcount block #%llu has an invalid "
125 "rf_fs_generation of #%u",
126 (unsigned long long)bh
->b_blocknr
,
127 le32_to_cpu(rb
->rf_fs_generation
));
134 static int ocfs2_read_refcount_block(struct ocfs2_caching_info
*ci
,
136 struct buffer_head
**bh
)
139 struct buffer_head
*tmp
= *bh
;
141 rc
= ocfs2_read_block(ci
, rb_blkno
, &tmp
,
142 ocfs2_validate_refcount_block
);
144 /* If ocfs2_read_block() got us a new bh, pass it up. */
151 static u64
ocfs2_refcount_cache_owner(struct ocfs2_caching_info
*ci
)
153 struct ocfs2_refcount_tree
*rf
= cache_info_to_refcount(ci
);
158 static struct super_block
*
159 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info
*ci
)
161 struct ocfs2_refcount_tree
*rf
= cache_info_to_refcount(ci
);
166 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info
*ci
)
168 struct ocfs2_refcount_tree
*rf
= cache_info_to_refcount(ci
);
170 spin_lock(&rf
->rf_lock
);
173 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info
*ci
)
175 struct ocfs2_refcount_tree
*rf
= cache_info_to_refcount(ci
);
177 spin_unlock(&rf
->rf_lock
);
180 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info
*ci
)
182 struct ocfs2_refcount_tree
*rf
= cache_info_to_refcount(ci
);
184 mutex_lock(&rf
->rf_io_mutex
);
187 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info
*ci
)
189 struct ocfs2_refcount_tree
*rf
= cache_info_to_refcount(ci
);
191 mutex_unlock(&rf
->rf_io_mutex
);
194 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops
= {
195 .co_owner
= ocfs2_refcount_cache_owner
,
196 .co_get_super
= ocfs2_refcount_cache_get_super
,
197 .co_cache_lock
= ocfs2_refcount_cache_lock
,
198 .co_cache_unlock
= ocfs2_refcount_cache_unlock
,
199 .co_io_lock
= ocfs2_refcount_cache_io_lock
,
200 .co_io_unlock
= ocfs2_refcount_cache_io_unlock
,
203 static struct ocfs2_refcount_tree
*
204 ocfs2_find_refcount_tree(struct ocfs2_super
*osb
, u64 blkno
)
206 struct rb_node
*n
= osb
->osb_rf_lock_tree
.rb_node
;
207 struct ocfs2_refcount_tree
*tree
= NULL
;
210 tree
= rb_entry(n
, struct ocfs2_refcount_tree
, rf_node
);
212 if (blkno
< tree
->rf_blkno
)
214 else if (blkno
> tree
->rf_blkno
)
223 /* osb_lock is already locked. */
224 static void ocfs2_insert_refcount_tree(struct ocfs2_super
*osb
,
225 struct ocfs2_refcount_tree
*new)
227 u64 rf_blkno
= new->rf_blkno
;
228 struct rb_node
*parent
= NULL
;
229 struct rb_node
**p
= &osb
->osb_rf_lock_tree
.rb_node
;
230 struct ocfs2_refcount_tree
*tmp
;
235 tmp
= rb_entry(parent
, struct ocfs2_refcount_tree
,
238 if (rf_blkno
< tmp
->rf_blkno
)
240 else if (rf_blkno
> tmp
->rf_blkno
)
243 /* This should never happen! */
244 mlog(ML_ERROR
, "Duplicate refcount block %llu found!\n",
245 (unsigned long long)rf_blkno
);
250 rb_link_node(&new->rf_node
, parent
, p
);
251 rb_insert_color(&new->rf_node
, &osb
->osb_rf_lock_tree
);
254 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree
*tree
)
256 ocfs2_metadata_cache_exit(&tree
->rf_ci
);
257 ocfs2_simple_drop_lockres(OCFS2_SB(tree
->rf_sb
), &tree
->rf_lockres
);
258 ocfs2_lock_res_free(&tree
->rf_lockres
);
263 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super
*osb
,
264 struct ocfs2_refcount_tree
*tree
)
266 rb_erase(&tree
->rf_node
, &osb
->osb_rf_lock_tree
);
267 if (osb
->osb_ref_tree_lru
&& osb
->osb_ref_tree_lru
== tree
)
268 osb
->osb_ref_tree_lru
= NULL
;
271 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super
*osb
,
272 struct ocfs2_refcount_tree
*tree
)
274 spin_lock(&osb
->osb_lock
);
275 ocfs2_erase_refcount_tree_from_list_no_lock(osb
, tree
);
276 spin_unlock(&osb
->osb_lock
);
279 void ocfs2_kref_remove_refcount_tree(struct kref
*kref
)
281 struct ocfs2_refcount_tree
*tree
=
282 container_of(kref
, struct ocfs2_refcount_tree
, rf_getcnt
);
284 ocfs2_free_refcount_tree(tree
);
288 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree
*tree
)
290 kref_get(&tree
->rf_getcnt
);
294 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree
*tree
)
296 kref_put(&tree
->rf_getcnt
, ocfs2_kref_remove_refcount_tree
);
299 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree
*new,
300 struct super_block
*sb
)
302 ocfs2_metadata_cache_init(&new->rf_ci
, &ocfs2_refcount_caching_ops
);
303 mutex_init(&new->rf_io_mutex
);
305 spin_lock_init(&new->rf_lock
);
308 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super
*osb
,
309 struct ocfs2_refcount_tree
*new,
310 u64 rf_blkno
, u32 generation
)
312 init_rwsem(&new->rf_sem
);
313 ocfs2_refcount_lock_res_init(&new->rf_lockres
, osb
,
314 rf_blkno
, generation
);
317 static struct ocfs2_refcount_tree
*
318 ocfs2_allocate_refcount_tree(struct ocfs2_super
*osb
, u64 rf_blkno
)
320 struct ocfs2_refcount_tree
*new;
322 new = kzalloc(sizeof(struct ocfs2_refcount_tree
), GFP_NOFS
);
326 new->rf_blkno
= rf_blkno
;
327 kref_init(&new->rf_getcnt
);
328 ocfs2_init_refcount_tree_ci(new, osb
->sb
);
333 static int ocfs2_get_refcount_tree(struct ocfs2_super
*osb
, u64 rf_blkno
,
334 struct ocfs2_refcount_tree
**ret_tree
)
337 struct ocfs2_refcount_tree
*tree
, *new = NULL
;
338 struct buffer_head
*ref_root_bh
= NULL
;
339 struct ocfs2_refcount_block
*ref_rb
;
341 spin_lock(&osb
->osb_lock
);
342 if (osb
->osb_ref_tree_lru
&&
343 osb
->osb_ref_tree_lru
->rf_blkno
== rf_blkno
)
344 tree
= osb
->osb_ref_tree_lru
;
346 tree
= ocfs2_find_refcount_tree(osb
, rf_blkno
);
350 spin_unlock(&osb
->osb_lock
);
352 new = ocfs2_allocate_refcount_tree(osb
, rf_blkno
);
359 * We need the generation to create the refcount tree lock and since
360 * it isn't changed during the tree modification, we are safe here to
361 * read without protection.
362 * We also have to purge the cache after we create the lock since the
363 * refcount block may have the stale data. It can only be trusted when
364 * we hold the refcount lock.
366 ret
= ocfs2_read_refcount_block(&new->rf_ci
, rf_blkno
, &ref_root_bh
);
369 ocfs2_metadata_cache_exit(&new->rf_ci
);
374 ref_rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
375 new->rf_generation
= le32_to_cpu(ref_rb
->rf_generation
);
376 ocfs2_init_refcount_tree_lock(osb
, new, rf_blkno
,
378 ocfs2_metadata_cache_purge(&new->rf_ci
);
380 spin_lock(&osb
->osb_lock
);
381 tree
= ocfs2_find_refcount_tree(osb
, rf_blkno
);
385 ocfs2_insert_refcount_tree(osb
, new);
393 osb
->osb_ref_tree_lru
= tree
;
395 spin_unlock(&osb
->osb_lock
);
398 ocfs2_free_refcount_tree(new);
404 static int ocfs2_get_refcount_block(struct inode
*inode
, u64
*ref_blkno
)
407 struct buffer_head
*di_bh
= NULL
;
408 struct ocfs2_dinode
*di
;
410 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
416 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
));
418 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
419 *ref_blkno
= le64_to_cpu(di
->i_refcount_loc
);
425 static int __ocfs2_lock_refcount_tree(struct ocfs2_super
*osb
,
426 struct ocfs2_refcount_tree
*tree
, int rw
)
430 ret
= ocfs2_refcount_lock(tree
, rw
);
437 down_write(&tree
->rf_sem
);
439 down_read(&tree
->rf_sem
);
446 * Lock the refcount tree pointed by ref_blkno and return the tree.
447 * In most case, we lock the tree and read the refcount block.
448 * So read it here if the caller really needs it.
450 * If the tree has been re-created by other node, it will free the
451 * old one and re-create it.
453 int ocfs2_lock_refcount_tree(struct ocfs2_super
*osb
,
454 u64 ref_blkno
, int rw
,
455 struct ocfs2_refcount_tree
**ret_tree
,
456 struct buffer_head
**ref_bh
)
458 int ret
, delete_tree
= 0;
459 struct ocfs2_refcount_tree
*tree
= NULL
;
460 struct buffer_head
*ref_root_bh
= NULL
;
461 struct ocfs2_refcount_block
*rb
;
464 ret
= ocfs2_get_refcount_tree(osb
, ref_blkno
, &tree
);
470 ocfs2_refcount_tree_get(tree
);
472 ret
= __ocfs2_lock_refcount_tree(osb
, tree
, rw
);
475 ocfs2_refcount_tree_put(tree
);
479 ret
= ocfs2_read_refcount_block(&tree
->rf_ci
, tree
->rf_blkno
,
483 ocfs2_unlock_refcount_tree(osb
, tree
, rw
);
484 ocfs2_refcount_tree_put(tree
);
488 rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
490 * If the refcount block has been freed and re-created, we may need
491 * to recreate the refcount tree also.
493 * Here we just remove the tree from the rb-tree, and the last
494 * kref holder will unlock and delete this refcount_tree.
495 * Then we goto "again" and ocfs2_get_refcount_tree will create
496 * the new refcount tree for us.
498 if (tree
->rf_generation
!= le32_to_cpu(rb
->rf_generation
)) {
499 if (!tree
->rf_removed
) {
500 ocfs2_erase_refcount_tree_from_list(osb
, tree
);
501 tree
->rf_removed
= 1;
505 ocfs2_unlock_refcount_tree(osb
, tree
, rw
);
507 * We get an extra reference when we create the refcount
508 * tree, so another put will destroy it.
511 ocfs2_refcount_tree_put(tree
);
519 *ref_bh
= ref_root_bh
;
527 int ocfs2_lock_refcount_tree_by_inode(struct inode
*inode
, int rw
,
528 struct ocfs2_refcount_tree
**ret_tree
,
529 struct buffer_head
**ref_bh
)
534 ret
= ocfs2_get_refcount_block(inode
, &ref_blkno
);
540 return ocfs2_lock_refcount_tree(OCFS2_SB(inode
->i_sb
), ref_blkno
,
541 rw
, ret_tree
, ref_bh
);
544 void ocfs2_unlock_refcount_tree(struct ocfs2_super
*osb
,
545 struct ocfs2_refcount_tree
*tree
, int rw
)
548 up_write(&tree
->rf_sem
);
550 up_read(&tree
->rf_sem
);
552 ocfs2_refcount_unlock(tree
, rw
);
553 ocfs2_refcount_tree_put(tree
);
556 void ocfs2_purge_refcount_trees(struct ocfs2_super
*osb
)
558 struct rb_node
*node
;
559 struct ocfs2_refcount_tree
*tree
;
560 struct rb_root
*root
= &osb
->osb_rf_lock_tree
;
562 while ((node
= rb_last(root
)) != NULL
) {
563 tree
= rb_entry(node
, struct ocfs2_refcount_tree
, rf_node
);
565 mlog(0, "Purge tree %llu\n",
566 (unsigned long long) tree
->rf_blkno
);
568 rb_erase(&tree
->rf_node
, root
);
569 ocfs2_free_refcount_tree(tree
);
574 * Create a refcount tree for an inode.
575 * We take for granted that the inode is already locked.
577 static int ocfs2_create_refcount_tree(struct inode
*inode
,
578 struct buffer_head
*di_bh
)
581 handle_t
*handle
= NULL
;
582 struct ocfs2_alloc_context
*meta_ac
= NULL
;
583 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
584 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
585 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
586 struct buffer_head
*new_bh
= NULL
;
587 struct ocfs2_refcount_block
*rb
;
588 struct ocfs2_refcount_tree
*new_tree
= NULL
, *tree
= NULL
;
589 u16 suballoc_bit_start
;
593 BUG_ON(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
);
595 mlog(0, "create tree for inode %lu\n", inode
->i_ino
);
597 ret
= ocfs2_reserve_new_metadata_blocks(osb
, 1, &meta_ac
);
603 handle
= ocfs2_start_trans(osb
, OCFS2_REFCOUNT_TREE_CREATE_CREDITS
);
604 if (IS_ERR(handle
)) {
605 ret
= PTR_ERR(handle
);
610 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
611 OCFS2_JOURNAL_ACCESS_WRITE
);
617 ret
= ocfs2_claim_metadata(osb
, handle
, meta_ac
, 1,
618 &suballoc_bit_start
, &num_got
,
625 new_tree
= ocfs2_allocate_refcount_tree(osb
, first_blkno
);
632 new_bh
= sb_getblk(inode
->i_sb
, first_blkno
);
633 ocfs2_set_new_buffer_uptodate(&new_tree
->rf_ci
, new_bh
);
635 ret
= ocfs2_journal_access_rb(handle
, &new_tree
->rf_ci
, new_bh
,
636 OCFS2_JOURNAL_ACCESS_CREATE
);
642 /* Initialize ocfs2_refcount_block. */
643 rb
= (struct ocfs2_refcount_block
*)new_bh
->b_data
;
644 memset(rb
, 0, inode
->i_sb
->s_blocksize
);
645 strcpy((void *)rb
, OCFS2_REFCOUNT_BLOCK_SIGNATURE
);
646 rb
->rf_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
647 rb
->rf_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
648 rb
->rf_fs_generation
= cpu_to_le32(osb
->fs_generation
);
649 rb
->rf_blkno
= cpu_to_le64(first_blkno
);
650 rb
->rf_count
= cpu_to_le32(1);
651 rb
->rf_records
.rl_count
=
652 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb
->sb
));
653 spin_lock(&osb
->osb_lock
);
654 rb
->rf_generation
= osb
->s_next_generation
++;
655 spin_unlock(&osb
->osb_lock
);
657 ocfs2_journal_dirty(handle
, new_bh
);
659 spin_lock(&oi
->ip_lock
);
660 oi
->ip_dyn_features
|= OCFS2_HAS_REFCOUNT_FL
;
661 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
662 di
->i_refcount_loc
= cpu_to_le64(first_blkno
);
663 spin_unlock(&oi
->ip_lock
);
665 mlog(0, "created tree for inode %lu, refblock %llu\n",
666 inode
->i_ino
, (unsigned long long)first_blkno
);
668 ocfs2_journal_dirty(handle
, di_bh
);
671 * We have to init the tree lock here since it will use
672 * the generation number to create it.
674 new_tree
->rf_generation
= le32_to_cpu(rb
->rf_generation
);
675 ocfs2_init_refcount_tree_lock(osb
, new_tree
, first_blkno
,
676 new_tree
->rf_generation
);
678 spin_lock(&osb
->osb_lock
);
679 tree
= ocfs2_find_refcount_tree(osb
, first_blkno
);
682 * We've just created a new refcount tree in this block. If
683 * we found a refcount tree on the ocfs2_super, it must be
684 * one we just deleted. We free the old tree before
685 * inserting the new tree.
687 BUG_ON(tree
&& tree
->rf_generation
== new_tree
->rf_generation
);
689 ocfs2_erase_refcount_tree_from_list_no_lock(osb
, tree
);
690 ocfs2_insert_refcount_tree(osb
, new_tree
);
691 spin_unlock(&osb
->osb_lock
);
694 ocfs2_refcount_tree_put(tree
);
697 ocfs2_commit_trans(osb
, handle
);
701 ocfs2_metadata_cache_exit(&new_tree
->rf_ci
);
707 ocfs2_free_alloc_context(meta_ac
);
712 static int ocfs2_set_refcount_tree(struct inode
*inode
,
713 struct buffer_head
*di_bh
,
717 handle_t
*handle
= NULL
;
718 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
719 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
720 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
721 struct buffer_head
*ref_root_bh
= NULL
;
722 struct ocfs2_refcount_block
*rb
;
723 struct ocfs2_refcount_tree
*ref_tree
;
725 BUG_ON(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
);
727 ret
= ocfs2_lock_refcount_tree(osb
, refcount_loc
, 1,
728 &ref_tree
, &ref_root_bh
);
734 handle
= ocfs2_start_trans(osb
, OCFS2_REFCOUNT_TREE_SET_CREDITS
);
735 if (IS_ERR(handle
)) {
736 ret
= PTR_ERR(handle
);
741 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
742 OCFS2_JOURNAL_ACCESS_WRITE
);
748 ret
= ocfs2_journal_access_rb(handle
, &ref_tree
->rf_ci
, ref_root_bh
,
749 OCFS2_JOURNAL_ACCESS_WRITE
);
755 rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
756 le32_add_cpu(&rb
->rf_count
, 1);
758 ocfs2_journal_dirty(handle
, ref_root_bh
);
760 spin_lock(&oi
->ip_lock
);
761 oi
->ip_dyn_features
|= OCFS2_HAS_REFCOUNT_FL
;
762 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
763 di
->i_refcount_loc
= cpu_to_le64(refcount_loc
);
764 spin_unlock(&oi
->ip_lock
);
765 ocfs2_journal_dirty(handle
, di_bh
);
768 ocfs2_commit_trans(osb
, handle
);
770 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
776 int ocfs2_remove_refcount_tree(struct inode
*inode
, struct buffer_head
*di_bh
)
778 int ret
, delete_tree
= 0;
779 handle_t
*handle
= NULL
;
780 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
781 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
782 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
783 struct ocfs2_refcount_block
*rb
;
784 struct inode
*alloc_inode
= NULL
;
785 struct buffer_head
*alloc_bh
= NULL
;
786 struct buffer_head
*blk_bh
= NULL
;
787 struct ocfs2_refcount_tree
*ref_tree
;
788 int credits
= OCFS2_REFCOUNT_TREE_REMOVE_CREDITS
;
789 u64 blk
= 0, bg_blkno
= 0, ref_blkno
= le64_to_cpu(di
->i_refcount_loc
);
792 if (!(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
))
796 ret
= ocfs2_lock_refcount_tree(osb
, ref_blkno
, 1, &ref_tree
, &blk_bh
);
802 rb
= (struct ocfs2_refcount_block
*)blk_bh
->b_data
;
805 * If we are the last user, we need to free the block.
806 * So lock the allocator ahead.
808 if (le32_to_cpu(rb
->rf_count
) == 1) {
809 blk
= le64_to_cpu(rb
->rf_blkno
);
810 bit
= le16_to_cpu(rb
->rf_suballoc_bit
);
811 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
813 alloc_inode
= ocfs2_get_system_file_inode(osb
,
814 EXTENT_ALLOC_SYSTEM_INODE
,
815 le16_to_cpu(rb
->rf_suballoc_slot
));
821 mutex_lock(&alloc_inode
->i_mutex
);
823 ret
= ocfs2_inode_lock(alloc_inode
, &alloc_bh
, 1);
829 credits
+= OCFS2_SUBALLOC_FREE
;
832 handle
= ocfs2_start_trans(osb
, credits
);
833 if (IS_ERR(handle
)) {
834 ret
= PTR_ERR(handle
);
839 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
840 OCFS2_JOURNAL_ACCESS_WRITE
);
846 ret
= ocfs2_journal_access_rb(handle
, &ref_tree
->rf_ci
, blk_bh
,
847 OCFS2_JOURNAL_ACCESS_WRITE
);
853 spin_lock(&oi
->ip_lock
);
854 oi
->ip_dyn_features
&= ~OCFS2_HAS_REFCOUNT_FL
;
855 di
->i_dyn_features
= cpu_to_le16(oi
->ip_dyn_features
);
856 di
->i_refcount_loc
= 0;
857 spin_unlock(&oi
->ip_lock
);
858 ocfs2_journal_dirty(handle
, di_bh
);
860 le32_add_cpu(&rb
->rf_count
, -1);
861 ocfs2_journal_dirty(handle
, blk_bh
);
865 ocfs2_erase_refcount_tree_from_list(osb
, ref_tree
);
866 ret
= ocfs2_free_suballoc_bits(handle
, alloc_inode
,
867 alloc_bh
, bit
, bg_blkno
, 1);
873 ocfs2_commit_trans(osb
, handle
);
876 ocfs2_inode_unlock(alloc_inode
, 1);
881 mutex_unlock(&alloc_inode
->i_mutex
);
885 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
887 ocfs2_refcount_tree_put(ref_tree
);
893 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info
*ci
,
894 struct buffer_head
*ref_leaf_bh
,
895 u64 cpos
, unsigned int len
,
896 struct ocfs2_refcount_rec
*ret_rec
,
900 struct ocfs2_refcount_block
*rb
=
901 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
902 struct ocfs2_refcount_rec
*rec
= NULL
;
904 for (; i
< le16_to_cpu(rb
->rf_records
.rl_used
); i
++) {
905 rec
= &rb
->rf_records
.rl_recs
[i
];
907 if (le64_to_cpu(rec
->r_cpos
) +
908 le32_to_cpu(rec
->r_clusters
) <= cpos
)
910 else if (le64_to_cpu(rec
->r_cpos
) > cpos
)
913 /* ok, cpos fail in this rec. Just return. */
920 /* We meet with a hole here, so fake the rec. */
921 ret_rec
->r_cpos
= cpu_to_le64(cpos
);
922 ret_rec
->r_refcount
= 0;
923 if (i
< le16_to_cpu(rb
->rf_records
.rl_used
) &&
924 le64_to_cpu(rec
->r_cpos
) < cpos
+ len
)
925 ret_rec
->r_clusters
=
926 cpu_to_le32(le64_to_cpu(rec
->r_cpos
) - cpos
);
928 ret_rec
->r_clusters
= cpu_to_le32(len
);
936 * Try to remove refcount tree. The mechanism is:
937 * 1) Check whether i_clusters == 0, if no, exit.
938 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
939 * 3) Check whether we have inline xattr stored outside, if yes, exit.
940 * 4) Remove the tree.
942 int ocfs2_try_remove_refcount_tree(struct inode
*inode
,
943 struct buffer_head
*di_bh
)
946 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
947 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
949 down_write(&oi
->ip_xattr_sem
);
950 down_write(&oi
->ip_alloc_sem
);
955 if ((oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
) && di
->i_xattr_loc
)
958 if (oi
->ip_dyn_features
& OCFS2_INLINE_XATTR_FL
&&
959 ocfs2_has_inline_xattr_value_outside(inode
, di
))
962 ret
= ocfs2_remove_refcount_tree(inode
, di_bh
);
966 up_write(&oi
->ip_alloc_sem
);
967 up_write(&oi
->ip_xattr_sem
);
972 * Find the end range for a leaf refcount block indicated by
973 * el->l_recs[index].e_blkno.
975 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info
*ci
,
976 struct buffer_head
*ref_root_bh
,
977 struct ocfs2_extent_block
*eb
,
978 struct ocfs2_extent_list
*el
,
979 int index
, u32
*cpos_end
)
981 int ret
, i
, subtree_root
;
984 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
985 struct ocfs2_path
*left_path
= NULL
, *right_path
= NULL
;
986 struct ocfs2_extent_tree et
;
987 struct ocfs2_extent_list
*tmp_el
;
989 if (index
< le16_to_cpu(el
->l_next_free_rec
) - 1) {
991 * We have a extent rec after index, so just use the e_cpos
992 * of the next extent rec.
994 *cpos_end
= le32_to_cpu(el
->l_recs
[index
+1].e_cpos
);
998 if (!eb
|| (eb
&& !eb
->h_next_leaf_blk
)) {
1000 * We are the last extent rec, so any high cpos should
1001 * be stored in this leaf refcount block.
1003 *cpos_end
= UINT_MAX
;
1008 * If the extent block isn't the last one, we have to find
1009 * the subtree root between this extent block and the next
1010 * leaf extent block and get the corresponding e_cpos from
1011 * the subroot. Otherwise we may corrupt the b-tree.
1013 ocfs2_init_refcount_extent_tree(&et
, ci
, ref_root_bh
);
1015 left_path
= ocfs2_new_path_from_et(&et
);
1022 cpos
= le32_to_cpu(eb
->h_list
.l_recs
[index
].e_cpos
);
1023 ret
= ocfs2_find_path(ci
, left_path
, cpos
);
1029 right_path
= ocfs2_new_path_from_path(left_path
);
1036 ret
= ocfs2_find_cpos_for_right_leaf(sb
, left_path
, &cpos
);
1042 ret
= ocfs2_find_path(ci
, right_path
, cpos
);
1048 subtree_root
= ocfs2_find_subtree_root(&et
, left_path
,
1051 tmp_el
= left_path
->p_node
[subtree_root
].el
;
1052 blkno
= left_path
->p_node
[subtree_root
+1].bh
->b_blocknr
;
1053 for (i
= 0; i
< le32_to_cpu(tmp_el
->l_next_free_rec
); i
++) {
1054 if (le64_to_cpu(tmp_el
->l_recs
[i
].e_blkno
) == blkno
) {
1055 *cpos_end
= le32_to_cpu(tmp_el
->l_recs
[i
+1].e_cpos
);
1060 BUG_ON(i
== le32_to_cpu(tmp_el
->l_next_free_rec
));
1063 ocfs2_free_path(left_path
);
1064 ocfs2_free_path(right_path
);
1069 * Given a cpos and len, try to find the refcount record which contains cpos.
1070 * 1. If cpos can be found in one refcount record, return the record.
1071 * 2. If cpos can't be found, return a fake record which start from cpos
1072 * and end at a small value between cpos+len and start of the next record.
1073 * This fake record has r_refcount = 0.
1075 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info
*ci
,
1076 struct buffer_head
*ref_root_bh
,
1077 u64 cpos
, unsigned int len
,
1078 struct ocfs2_refcount_rec
*ret_rec
,
1080 struct buffer_head
**ret_bh
)
1082 int ret
= 0, i
, found
;
1083 u32 low_cpos
, uninitialized_var(cpos_end
);
1084 struct ocfs2_extent_list
*el
;
1085 struct ocfs2_extent_rec
*rec
= NULL
;
1086 struct ocfs2_extent_block
*eb
= NULL
;
1087 struct buffer_head
*eb_bh
= NULL
, *ref_leaf_bh
= NULL
;
1088 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
1089 struct ocfs2_refcount_block
*rb
=
1090 (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
1092 if (!(le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
)) {
1093 ocfs2_find_refcount_rec_in_rl(ci
, ref_root_bh
, cpos
, len
,
1095 *ret_bh
= ref_root_bh
;
1096 get_bh(ref_root_bh
);
1101 low_cpos
= cpos
& OCFS2_32BIT_POS_MASK
;
1103 if (el
->l_tree_depth
) {
1104 ret
= ocfs2_find_leaf(ci
, el
, low_cpos
, &eb_bh
);
1110 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
1113 if (el
->l_tree_depth
) {
1115 "refcount tree %llu has non zero tree "
1116 "depth in leaf btree tree block %llu\n",
1117 (unsigned long long)ocfs2_metadata_cache_owner(ci
),
1118 (unsigned long long)eb_bh
->b_blocknr
);
1125 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
1126 rec
= &el
->l_recs
[i
];
1128 if (le32_to_cpu(rec
->e_cpos
) <= low_cpos
) {
1135 ret
= ocfs2_get_refcount_cpos_end(ci
, ref_root_bh
,
1136 eb
, el
, i
, &cpos_end
);
1142 if (cpos_end
< low_cpos
+ len
)
1143 len
= cpos_end
- low_cpos
;
1146 ret
= ocfs2_read_refcount_block(ci
, le64_to_cpu(rec
->e_blkno
),
1153 ocfs2_find_refcount_rec_in_rl(ci
, ref_leaf_bh
, cpos
, len
,
1155 *ret_bh
= ref_leaf_bh
;
1161 enum ocfs2_ref_rec_contig
{
1162 REF_CONTIG_NONE
= 0,
1165 REF_CONTIG_LEFTRIGHT
,
1168 static enum ocfs2_ref_rec_contig
1169 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block
*rb
,
1172 if ((rb
->rf_records
.rl_recs
[index
].r_refcount
==
1173 rb
->rf_records
.rl_recs
[index
+ 1].r_refcount
) &&
1174 (le64_to_cpu(rb
->rf_records
.rl_recs
[index
].r_cpos
) +
1175 le32_to_cpu(rb
->rf_records
.rl_recs
[index
].r_clusters
) ==
1176 le64_to_cpu(rb
->rf_records
.rl_recs
[index
+ 1].r_cpos
)))
1177 return REF_CONTIG_RIGHT
;
1179 return REF_CONTIG_NONE
;
1182 static enum ocfs2_ref_rec_contig
1183 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block
*rb
,
1186 enum ocfs2_ref_rec_contig ret
= REF_CONTIG_NONE
;
1188 if (index
< le16_to_cpu(rb
->rf_records
.rl_used
) - 1)
1189 ret
= ocfs2_refcount_rec_adjacent(rb
, index
);
1192 enum ocfs2_ref_rec_contig tmp
;
1194 tmp
= ocfs2_refcount_rec_adjacent(rb
, index
- 1);
1196 if (tmp
== REF_CONTIG_RIGHT
) {
1197 if (ret
== REF_CONTIG_RIGHT
)
1198 ret
= REF_CONTIG_LEFTRIGHT
;
1200 ret
= REF_CONTIG_LEFT
;
1207 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block
*rb
,
1210 BUG_ON(rb
->rf_records
.rl_recs
[index
].r_refcount
!=
1211 rb
->rf_records
.rl_recs
[index
+1].r_refcount
);
1213 le32_add_cpu(&rb
->rf_records
.rl_recs
[index
].r_clusters
,
1214 le32_to_cpu(rb
->rf_records
.rl_recs
[index
+1].r_clusters
));
1216 if (index
< le16_to_cpu(rb
->rf_records
.rl_used
) - 2)
1217 memmove(&rb
->rf_records
.rl_recs
[index
+ 1],
1218 &rb
->rf_records
.rl_recs
[index
+ 2],
1219 sizeof(struct ocfs2_refcount_rec
) *
1220 (le16_to_cpu(rb
->rf_records
.rl_used
) - index
- 2));
1222 memset(&rb
->rf_records
.rl_recs
[le16_to_cpu(rb
->rf_records
.rl_used
) - 1],
1223 0, sizeof(struct ocfs2_refcount_rec
));
1224 le16_add_cpu(&rb
->rf_records
.rl_used
, -1);
1228 * Merge the refcount rec if we are contiguous with the adjacent recs.
1230 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block
*rb
,
1233 enum ocfs2_ref_rec_contig contig
=
1234 ocfs2_refcount_rec_contig(rb
, index
);
1236 if (contig
== REF_CONTIG_NONE
)
1239 if (contig
== REF_CONTIG_LEFT
|| contig
== REF_CONTIG_LEFTRIGHT
) {
1244 ocfs2_rotate_refcount_rec_left(rb
, index
);
1246 if (contig
== REF_CONTIG_LEFTRIGHT
)
1247 ocfs2_rotate_refcount_rec_left(rb
, index
);
1251 * Change the refcount indexed by "index" in ref_bh.
1252 * If refcount reaches 0, remove it.
1254 static int ocfs2_change_refcount_rec(handle_t
*handle
,
1255 struct ocfs2_caching_info
*ci
,
1256 struct buffer_head
*ref_leaf_bh
,
1257 int index
, int merge
, int change
)
1260 struct ocfs2_refcount_block
*rb
=
1261 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1262 struct ocfs2_refcount_list
*rl
= &rb
->rf_records
;
1263 struct ocfs2_refcount_rec
*rec
= &rl
->rl_recs
[index
];
1265 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_leaf_bh
,
1266 OCFS2_JOURNAL_ACCESS_WRITE
);
1272 mlog(0, "change index %d, old count %u, change %d\n", index
,
1273 le32_to_cpu(rec
->r_refcount
), change
);
1274 le32_add_cpu(&rec
->r_refcount
, change
);
1276 if (!rec
->r_refcount
) {
1277 if (index
!= le16_to_cpu(rl
->rl_used
) - 1) {
1278 memmove(rec
, rec
+ 1,
1279 (le16_to_cpu(rl
->rl_used
) - index
- 1) *
1280 sizeof(struct ocfs2_refcount_rec
));
1281 memset(&rl
->rl_recs
[le16_to_cpu(rl
->rl_used
) - 1],
1282 0, sizeof(struct ocfs2_refcount_rec
));
1285 le16_add_cpu(&rl
->rl_used
, -1);
1287 ocfs2_refcount_rec_merge(rb
, index
);
1289 ret
= ocfs2_journal_dirty(handle
, ref_leaf_bh
);
1296 static int ocfs2_expand_inline_ref_root(handle_t
*handle
,
1297 struct ocfs2_caching_info
*ci
,
1298 struct buffer_head
*ref_root_bh
,
1299 struct buffer_head
**ref_leaf_bh
,
1300 struct ocfs2_alloc_context
*meta_ac
)
1303 u16 suballoc_bit_start
;
1306 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
1307 struct buffer_head
*new_bh
= NULL
;
1308 struct ocfs2_refcount_block
*new_rb
;
1309 struct ocfs2_refcount_block
*root_rb
=
1310 (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
1312 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_root_bh
,
1313 OCFS2_JOURNAL_ACCESS_WRITE
);
1319 ret
= ocfs2_claim_metadata(OCFS2_SB(sb
), handle
, meta_ac
, 1,
1320 &suballoc_bit_start
, &num_got
,
1327 new_bh
= sb_getblk(sb
, blkno
);
1328 if (new_bh
== NULL
) {
1333 ocfs2_set_new_buffer_uptodate(ci
, new_bh
);
1335 ret
= ocfs2_journal_access_rb(handle
, ci
, new_bh
,
1336 OCFS2_JOURNAL_ACCESS_CREATE
);
1343 * Initialize ocfs2_refcount_block.
1344 * It should contain the same information as the old root.
1345 * so just memcpy it and change the corresponding field.
1347 memcpy(new_bh
->b_data
, ref_root_bh
->b_data
, sb
->s_blocksize
);
1349 new_rb
= (struct ocfs2_refcount_block
*)new_bh
->b_data
;
1350 new_rb
->rf_suballoc_slot
= cpu_to_le16(OCFS2_SB(sb
)->slot_num
);
1351 new_rb
->rf_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
1352 new_rb
->rf_blkno
= cpu_to_le64(blkno
);
1353 new_rb
->rf_cpos
= cpu_to_le32(0);
1354 new_rb
->rf_parent
= cpu_to_le64(ref_root_bh
->b_blocknr
);
1355 new_rb
->rf_flags
= cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL
);
1356 ocfs2_journal_dirty(handle
, new_bh
);
1358 /* Now change the root. */
1359 memset(&root_rb
->rf_list
, 0, sb
->s_blocksize
-
1360 offsetof(struct ocfs2_refcount_block
, rf_list
));
1361 root_rb
->rf_list
.l_count
= cpu_to_le16(ocfs2_extent_recs_per_rb(sb
));
1362 root_rb
->rf_clusters
= cpu_to_le32(1);
1363 root_rb
->rf_list
.l_next_free_rec
= cpu_to_le16(1);
1364 root_rb
->rf_list
.l_recs
[0].e_blkno
= cpu_to_le64(blkno
);
1365 root_rb
->rf_list
.l_recs
[0].e_leaf_clusters
= cpu_to_le16(1);
1366 root_rb
->rf_flags
= cpu_to_le32(OCFS2_REFCOUNT_TREE_FL
);
1368 ocfs2_journal_dirty(handle
, ref_root_bh
);
1370 mlog(0, "new leaf block %llu, used %u\n", (unsigned long long)blkno
,
1371 le16_to_cpu(new_rb
->rf_records
.rl_used
));
1373 *ref_leaf_bh
= new_bh
;
1380 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec
*prev
,
1381 struct ocfs2_refcount_rec
*next
)
1383 if (ocfs2_get_ref_rec_low_cpos(prev
) + le32_to_cpu(prev
->r_clusters
) <=
1384 ocfs2_get_ref_rec_low_cpos(next
))
1390 static int cmp_refcount_rec_by_low_cpos(const void *a
, const void *b
)
1392 const struct ocfs2_refcount_rec
*l
= a
, *r
= b
;
1393 u32 l_cpos
= ocfs2_get_ref_rec_low_cpos(l
);
1394 u32 r_cpos
= ocfs2_get_ref_rec_low_cpos(r
);
1396 if (l_cpos
> r_cpos
)
1398 if (l_cpos
< r_cpos
)
1403 static int cmp_refcount_rec_by_cpos(const void *a
, const void *b
)
1405 const struct ocfs2_refcount_rec
*l
= a
, *r
= b
;
1406 u64 l_cpos
= le64_to_cpu(l
->r_cpos
);
1407 u64 r_cpos
= le64_to_cpu(r
->r_cpos
);
1409 if (l_cpos
> r_cpos
)
1411 if (l_cpos
< r_cpos
)
1416 static void swap_refcount_rec(void *a
, void *b
, int size
)
1418 struct ocfs2_refcount_rec
*l
= a
, *r
= b
, tmp
;
1420 tmp
= *(struct ocfs2_refcount_rec
*)l
;
1421 *(struct ocfs2_refcount_rec
*)l
=
1422 *(struct ocfs2_refcount_rec
*)r
;
1423 *(struct ocfs2_refcount_rec
*)r
= tmp
;
1427 * The refcount cpos are ordered by their 64bit cpos,
1428 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1429 * So we need to make sure that this pos isn't intersected with others.
1431 * Note: The refcount block is already sorted by their low 32 bit cpos,
1432 * So just try the middle pos first, and we will exit when we find
1433 * the good position.
1435 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list
*rl
,
1436 u32
*split_pos
, int *split_index
)
1438 int num_used
= le16_to_cpu(rl
->rl_used
);
1439 int delta
, middle
= num_used
/ 2;
1441 for (delta
= 0; delta
< middle
; delta
++) {
1442 /* Let's check delta earlier than middle */
1443 if (ocfs2_refcount_rec_no_intersect(
1444 &rl
->rl_recs
[middle
- delta
- 1],
1445 &rl
->rl_recs
[middle
- delta
])) {
1446 *split_index
= middle
- delta
;
1450 /* For even counts, don't walk off the end */
1451 if ((middle
+ delta
+ 1) == num_used
)
1454 /* Now try delta past middle */
1455 if (ocfs2_refcount_rec_no_intersect(
1456 &rl
->rl_recs
[middle
+ delta
],
1457 &rl
->rl_recs
[middle
+ delta
+ 1])) {
1458 *split_index
= middle
+ delta
+ 1;
1463 if (delta
>= middle
)
1466 *split_pos
= ocfs2_get_ref_rec_low_cpos(&rl
->rl_recs
[*split_index
]);
1470 static int ocfs2_divide_leaf_refcount_block(struct buffer_head
*ref_leaf_bh
,
1471 struct buffer_head
*new_bh
,
1474 int split_index
= 0, num_moved
, ret
;
1476 struct ocfs2_refcount_block
*rb
=
1477 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1478 struct ocfs2_refcount_list
*rl
= &rb
->rf_records
;
1479 struct ocfs2_refcount_block
*new_rb
=
1480 (struct ocfs2_refcount_block
*)new_bh
->b_data
;
1481 struct ocfs2_refcount_list
*new_rl
= &new_rb
->rf_records
;
1483 mlog(0, "split old leaf refcount block %llu, count = %u, used = %u\n",
1484 (unsigned long long)ref_leaf_bh
->b_blocknr
,
1485 le32_to_cpu(rl
->rl_count
), le32_to_cpu(rl
->rl_used
));
1488 * XXX: Improvement later.
1489 * If we know all the high 32 bit cpos is the same, no need to sort.
1491 * In order to make the whole process safe, we do:
1492 * 1. sort the entries by their low 32 bit cpos first so that we can
1493 * find the split cpos easily.
1494 * 2. call ocfs2_insert_extent to insert the new refcount block.
1495 * 3. move the refcount rec to the new block.
1496 * 4. sort the entries by their 64 bit cpos.
1497 * 5. dirty the new_rb and rb.
1499 sort(&rl
->rl_recs
, le16_to_cpu(rl
->rl_used
),
1500 sizeof(struct ocfs2_refcount_rec
),
1501 cmp_refcount_rec_by_low_cpos
, swap_refcount_rec
);
1503 ret
= ocfs2_find_refcount_split_pos(rl
, &cpos
, &split_index
);
1509 new_rb
->rf_cpos
= cpu_to_le32(cpos
);
1511 /* move refcount records starting from split_index to the new block. */
1512 num_moved
= le16_to_cpu(rl
->rl_used
) - split_index
;
1513 memcpy(new_rl
->rl_recs
, &rl
->rl_recs
[split_index
],
1514 num_moved
* sizeof(struct ocfs2_refcount_rec
));
1516 /*ok, remove the entries we just moved over to the other block. */
1517 memset(&rl
->rl_recs
[split_index
], 0,
1518 num_moved
* sizeof(struct ocfs2_refcount_rec
));
1520 /* change old and new rl_used accordingly. */
1521 le16_add_cpu(&rl
->rl_used
, -num_moved
);
1522 new_rl
->rl_used
= cpu_to_le32(num_moved
);
1524 sort(&rl
->rl_recs
, le16_to_cpu(rl
->rl_used
),
1525 sizeof(struct ocfs2_refcount_rec
),
1526 cmp_refcount_rec_by_cpos
, swap_refcount_rec
);
1528 sort(&new_rl
->rl_recs
, le16_to_cpu(new_rl
->rl_used
),
1529 sizeof(struct ocfs2_refcount_rec
),
1530 cmp_refcount_rec_by_cpos
, swap_refcount_rec
);
1536 static int ocfs2_new_leaf_refcount_block(handle_t
*handle
,
1537 struct ocfs2_caching_info
*ci
,
1538 struct buffer_head
*ref_root_bh
,
1539 struct buffer_head
*ref_leaf_bh
,
1540 struct ocfs2_alloc_context
*meta_ac
)
1543 u16 suballoc_bit_start
;
1544 u32 num_got
, new_cpos
;
1546 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
1547 struct ocfs2_refcount_block
*root_rb
=
1548 (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
1549 struct buffer_head
*new_bh
= NULL
;
1550 struct ocfs2_refcount_block
*new_rb
;
1551 struct ocfs2_extent_tree ref_et
;
1553 BUG_ON(!(le32_to_cpu(root_rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
));
1555 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_root_bh
,
1556 OCFS2_JOURNAL_ACCESS_WRITE
);
1562 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_leaf_bh
,
1563 OCFS2_JOURNAL_ACCESS_WRITE
);
1569 ret
= ocfs2_claim_metadata(OCFS2_SB(sb
), handle
, meta_ac
, 1,
1570 &suballoc_bit_start
, &num_got
,
1577 new_bh
= sb_getblk(sb
, blkno
);
1578 if (new_bh
== NULL
) {
1583 ocfs2_set_new_buffer_uptodate(ci
, new_bh
);
1585 ret
= ocfs2_journal_access_rb(handle
, ci
, new_bh
,
1586 OCFS2_JOURNAL_ACCESS_CREATE
);
1592 /* Initialize ocfs2_refcount_block. */
1593 new_rb
= (struct ocfs2_refcount_block
*)new_bh
->b_data
;
1594 memset(new_rb
, 0, sb
->s_blocksize
);
1595 strcpy((void *)new_rb
, OCFS2_REFCOUNT_BLOCK_SIGNATURE
);
1596 new_rb
->rf_suballoc_slot
= cpu_to_le16(OCFS2_SB(sb
)->slot_num
);
1597 new_rb
->rf_suballoc_bit
= cpu_to_le16(suballoc_bit_start
);
1598 new_rb
->rf_fs_generation
= cpu_to_le32(OCFS2_SB(sb
)->fs_generation
);
1599 new_rb
->rf_blkno
= cpu_to_le64(blkno
);
1600 new_rb
->rf_parent
= cpu_to_le64(ref_root_bh
->b_blocknr
);
1601 new_rb
->rf_flags
= cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL
);
1602 new_rb
->rf_records
.rl_count
=
1603 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb
));
1604 new_rb
->rf_generation
= root_rb
->rf_generation
;
1606 ret
= ocfs2_divide_leaf_refcount_block(ref_leaf_bh
, new_bh
, &new_cpos
);
1612 ocfs2_journal_dirty(handle
, ref_leaf_bh
);
1613 ocfs2_journal_dirty(handle
, new_bh
);
1615 ocfs2_init_refcount_extent_tree(&ref_et
, ci
, ref_root_bh
);
1617 mlog(0, "insert new leaf block %llu at %u\n",
1618 (unsigned long long)new_bh
->b_blocknr
, new_cpos
);
1620 /* Insert the new leaf block with the specific offset cpos. */
1621 ret
= ocfs2_insert_extent(handle
, &ref_et
, new_cpos
, new_bh
->b_blocknr
,
1631 static int ocfs2_expand_refcount_tree(handle_t
*handle
,
1632 struct ocfs2_caching_info
*ci
,
1633 struct buffer_head
*ref_root_bh
,
1634 struct buffer_head
*ref_leaf_bh
,
1635 struct ocfs2_alloc_context
*meta_ac
)
1638 struct buffer_head
*expand_bh
= NULL
;
1640 if (ref_root_bh
== ref_leaf_bh
) {
1642 * the old root bh hasn't been expanded to a b-tree,
1643 * so expand it first.
1645 ret
= ocfs2_expand_inline_ref_root(handle
, ci
, ref_root_bh
,
1646 &expand_bh
, meta_ac
);
1652 expand_bh
= ref_leaf_bh
;
1657 /* Now add a new refcount block into the tree.*/
1658 ret
= ocfs2_new_leaf_refcount_block(handle
, ci
, ref_root_bh
,
1659 expand_bh
, meta_ac
);
1668 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1670 * Only called when we have inserted a new refcount rec at index 0
1671 * which means ocfs2_extent_rec.e_cpos may need some change.
1673 static int ocfs2_adjust_refcount_rec(handle_t
*handle
,
1674 struct ocfs2_caching_info
*ci
,
1675 struct buffer_head
*ref_root_bh
,
1676 struct buffer_head
*ref_leaf_bh
,
1677 struct ocfs2_refcount_rec
*rec
)
1680 u32 new_cpos
, old_cpos
;
1681 struct ocfs2_path
*path
= NULL
;
1682 struct ocfs2_extent_tree et
;
1683 struct ocfs2_refcount_block
*rb
=
1684 (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
1685 struct ocfs2_extent_list
*el
;
1687 if (!(le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
))
1690 rb
= (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1691 old_cpos
= le32_to_cpu(rb
->rf_cpos
);
1692 new_cpos
= le64_to_cpu(rec
->r_cpos
) & OCFS2_32BIT_POS_MASK
;
1693 if (old_cpos
<= new_cpos
)
1696 ocfs2_init_refcount_extent_tree(&et
, ci
, ref_root_bh
);
1698 path
= ocfs2_new_path_from_et(&et
);
1705 ret
= ocfs2_find_path(ci
, path
, old_cpos
);
1712 * 2 more credits, one for the leaf refcount block, one for
1713 * the extent block contains the extent rec.
1715 ret
= ocfs2_extend_trans(handle
, handle
->h_buffer_credits
+ 2);
1721 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_leaf_bh
,
1722 OCFS2_JOURNAL_ACCESS_WRITE
);
1728 ret
= ocfs2_journal_access_eb(handle
, ci
, path_leaf_bh(path
),
1729 OCFS2_JOURNAL_ACCESS_WRITE
);
1735 /* change the leaf extent block first. */
1736 el
= path_leaf_el(path
);
1738 for (i
= 0; i
< le16_to_cpu(el
->l_next_free_rec
); i
++)
1739 if (le32_to_cpu(el
->l_recs
[i
].e_cpos
) == old_cpos
)
1742 BUG_ON(i
== le16_to_cpu(el
->l_next_free_rec
));
1744 el
->l_recs
[i
].e_cpos
= cpu_to_le32(new_cpos
);
1746 /* change the r_cpos in the leaf block. */
1747 rb
->rf_cpos
= cpu_to_le32(new_cpos
);
1749 ocfs2_journal_dirty(handle
, path_leaf_bh(path
));
1750 ocfs2_journal_dirty(handle
, ref_leaf_bh
);
1753 ocfs2_free_path(path
);
1757 static int ocfs2_insert_refcount_rec(handle_t
*handle
,
1758 struct ocfs2_caching_info
*ci
,
1759 struct buffer_head
*ref_root_bh
,
1760 struct buffer_head
*ref_leaf_bh
,
1761 struct ocfs2_refcount_rec
*rec
,
1762 int index
, int merge
,
1763 struct ocfs2_alloc_context
*meta_ac
)
1766 struct ocfs2_refcount_block
*rb
=
1767 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1768 struct ocfs2_refcount_list
*rf_list
= &rb
->rf_records
;
1769 struct buffer_head
*new_bh
= NULL
;
1771 BUG_ON(le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
);
1773 if (rf_list
->rl_used
== rf_list
->rl_count
) {
1774 u64 cpos
= le64_to_cpu(rec
->r_cpos
);
1775 u32 len
= le32_to_cpu(rec
->r_clusters
);
1777 ret
= ocfs2_expand_refcount_tree(handle
, ci
, ref_root_bh
,
1778 ref_leaf_bh
, meta_ac
);
1784 ret
= ocfs2_get_refcount_rec(ci
, ref_root_bh
,
1785 cpos
, len
, NULL
, &index
,
1792 ref_leaf_bh
= new_bh
;
1793 rb
= (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1794 rf_list
= &rb
->rf_records
;
1797 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_leaf_bh
,
1798 OCFS2_JOURNAL_ACCESS_WRITE
);
1804 if (index
< le16_to_cpu(rf_list
->rl_used
))
1805 memmove(&rf_list
->rl_recs
[index
+ 1],
1806 &rf_list
->rl_recs
[index
],
1807 (le16_to_cpu(rf_list
->rl_used
) - index
) *
1808 sizeof(struct ocfs2_refcount_rec
));
1810 mlog(0, "insert refcount record start %llu, len %u, count %u "
1811 "to leaf block %llu at index %d\n",
1812 (unsigned long long)le64_to_cpu(rec
->r_cpos
),
1813 le32_to_cpu(rec
->r_clusters
), le32_to_cpu(rec
->r_refcount
),
1814 (unsigned long long)ref_leaf_bh
->b_blocknr
, index
);
1816 rf_list
->rl_recs
[index
] = *rec
;
1818 le16_add_cpu(&rf_list
->rl_used
, 1);
1821 ocfs2_refcount_rec_merge(rb
, index
);
1823 ret
= ocfs2_journal_dirty(handle
, ref_leaf_bh
);
1830 ret
= ocfs2_adjust_refcount_rec(handle
, ci
,
1842 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1843 * This is much simple than our b-tree code.
1844 * split_rec is the new refcount rec we want to insert.
1845 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1846 * increase refcount or decrease a refcount to non-zero).
1847 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1848 * rec( in case we decrease a refcount to zero).
1850 static int ocfs2_split_refcount_rec(handle_t
*handle
,
1851 struct ocfs2_caching_info
*ci
,
1852 struct buffer_head
*ref_root_bh
,
1853 struct buffer_head
*ref_leaf_bh
,
1854 struct ocfs2_refcount_rec
*split_rec
,
1855 int index
, int merge
,
1856 struct ocfs2_alloc_context
*meta_ac
,
1857 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
1861 struct ocfs2_refcount_block
*rb
=
1862 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1863 struct ocfs2_refcount_list
*rf_list
= &rb
->rf_records
;
1864 struct ocfs2_refcount_rec
*orig_rec
= &rf_list
->rl_recs
[index
];
1865 struct ocfs2_refcount_rec
*tail_rec
= NULL
;
1866 struct buffer_head
*new_bh
= NULL
;
1868 BUG_ON(le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
);
1870 mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
1871 le64_to_cpu(orig_rec
->r_cpos
), le32_to_cpu(orig_rec
->r_clusters
),
1872 le64_to_cpu(split_rec
->r_cpos
),
1873 le32_to_cpu(split_rec
->r_clusters
));
1876 * If we just need to split the header or tail clusters,
1877 * no more recs are needed, just split is OK.
1878 * Otherwise we at least need one new recs.
1880 if (!split_rec
->r_refcount
&&
1881 (split_rec
->r_cpos
== orig_rec
->r_cpos
||
1882 le64_to_cpu(split_rec
->r_cpos
) +
1883 le32_to_cpu(split_rec
->r_clusters
) ==
1884 le64_to_cpu(orig_rec
->r_cpos
) + le32_to_cpu(orig_rec
->r_clusters
)))
1890 * We need one more rec if we split in the middle and the new rec have
1891 * some refcount in it.
1893 if (split_rec
->r_refcount
&&
1894 (split_rec
->r_cpos
!= orig_rec
->r_cpos
&&
1895 le64_to_cpu(split_rec
->r_cpos
) +
1896 le32_to_cpu(split_rec
->r_clusters
) !=
1897 le64_to_cpu(orig_rec
->r_cpos
) + le32_to_cpu(orig_rec
->r_clusters
)))
1900 /* If the leaf block don't have enough record, expand it. */
1901 if (le16_to_cpu(rf_list
->rl_used
) + recs_need
> rf_list
->rl_count
) {
1902 struct ocfs2_refcount_rec tmp_rec
;
1903 u64 cpos
= le64_to_cpu(orig_rec
->r_cpos
);
1904 len
= le32_to_cpu(orig_rec
->r_clusters
);
1905 ret
= ocfs2_expand_refcount_tree(handle
, ci
, ref_root_bh
,
1906 ref_leaf_bh
, meta_ac
);
1913 * We have to re-get it since now cpos may be moved to
1914 * another leaf block.
1916 ret
= ocfs2_get_refcount_rec(ci
, ref_root_bh
,
1917 cpos
, len
, &tmp_rec
, &index
,
1924 ref_leaf_bh
= new_bh
;
1925 rb
= (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
1926 rf_list
= &rb
->rf_records
;
1927 orig_rec
= &rf_list
->rl_recs
[index
];
1930 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_leaf_bh
,
1931 OCFS2_JOURNAL_ACCESS_WRITE
);
1938 * We have calculated out how many new records we need and store
1939 * in recs_need, so spare enough space first by moving the records
1940 * after "index" to the end.
1942 if (index
!= le16_to_cpu(rf_list
->rl_used
) - 1)
1943 memmove(&rf_list
->rl_recs
[index
+ 1 + recs_need
],
1944 &rf_list
->rl_recs
[index
+ 1],
1945 (le16_to_cpu(rf_list
->rl_used
) - index
- 1) *
1946 sizeof(struct ocfs2_refcount_rec
));
1948 len
= (le64_to_cpu(orig_rec
->r_cpos
) +
1949 le32_to_cpu(orig_rec
->r_clusters
)) -
1950 (le64_to_cpu(split_rec
->r_cpos
) +
1951 le32_to_cpu(split_rec
->r_clusters
));
1954 * If we have "len", the we will split in the tail and move it
1955 * to the end of the space we have just spared.
1958 tail_rec
= &rf_list
->rl_recs
[index
+ recs_need
];
1960 memcpy(tail_rec
, orig_rec
, sizeof(struct ocfs2_refcount_rec
));
1961 le64_add_cpu(&tail_rec
->r_cpos
,
1962 le32_to_cpu(tail_rec
->r_clusters
) - len
);
1963 tail_rec
->r_clusters
= le32_to_cpu(len
);
1967 * If the split pos isn't the same as the original one, we need to
1968 * split in the head.
1970 * Note: We have the chance that split_rec.r_refcount = 0,
1971 * recs_need = 0 and len > 0, which means we just cut the head from
1972 * the orig_rec and in that case we have done some modification in
1973 * orig_rec above, so the check for r_cpos is faked.
1975 if (split_rec
->r_cpos
!= orig_rec
->r_cpos
&& tail_rec
!= orig_rec
) {
1976 len
= le64_to_cpu(split_rec
->r_cpos
) -
1977 le64_to_cpu(orig_rec
->r_cpos
);
1978 orig_rec
->r_clusters
= cpu_to_le32(len
);
1982 le16_add_cpu(&rf_list
->rl_used
, recs_need
);
1984 if (split_rec
->r_refcount
) {
1985 rf_list
->rl_recs
[index
] = *split_rec
;
1986 mlog(0, "insert refcount record start %llu, len %u, count %u "
1987 "to leaf block %llu at index %d\n",
1988 (unsigned long long)le64_to_cpu(split_rec
->r_cpos
),
1989 le32_to_cpu(split_rec
->r_clusters
),
1990 le32_to_cpu(split_rec
->r_refcount
),
1991 (unsigned long long)ref_leaf_bh
->b_blocknr
, index
);
1994 ocfs2_refcount_rec_merge(rb
, index
);
1997 ret
= ocfs2_journal_dirty(handle
, ref_leaf_bh
);
2006 static int __ocfs2_increase_refcount(handle_t
*handle
,
2007 struct ocfs2_caching_info
*ci
,
2008 struct buffer_head
*ref_root_bh
,
2009 u64 cpos
, u32 len
, int merge
,
2010 struct ocfs2_alloc_context
*meta_ac
,
2011 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
2014 struct buffer_head
*ref_leaf_bh
= NULL
;
2015 struct ocfs2_refcount_rec rec
;
2016 unsigned int set_len
= 0;
2018 mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
2019 (unsigned long long)ocfs2_metadata_cache_owner(ci
),
2020 (unsigned long long)cpos
, len
);
2023 ret
= ocfs2_get_refcount_rec(ci
, ref_root_bh
,
2024 cpos
, len
, &rec
, &index
,
2031 set_len
= le32_to_cpu(rec
.r_clusters
);
2034 * Here we may meet with 3 situations:
2036 * 1. If we find an already existing record, and the length
2037 * is the same, cool, we just need to increase the r_refcount
2039 * 2. If we find a hole, just insert it with r_refcount = 1.
2040 * 3. If we are in the middle of one extent record, split
2043 if (rec
.r_refcount
&& le64_to_cpu(rec
.r_cpos
) == cpos
&&
2045 mlog(0, "increase refcount rec, start %llu, len %u, "
2046 "count %u\n", (unsigned long long)cpos
, set_len
,
2047 le32_to_cpu(rec
.r_refcount
));
2048 ret
= ocfs2_change_refcount_rec(handle
, ci
,
2055 } else if (!rec
.r_refcount
) {
2056 rec
.r_refcount
= cpu_to_le32(1);
2058 mlog(0, "insert refcount rec, start %llu, len %u\n",
2059 (unsigned long long)le64_to_cpu(rec
.r_cpos
),
2061 ret
= ocfs2_insert_refcount_rec(handle
, ci
, ref_root_bh
,
2070 set_len
= min((u64
)(cpos
+ len
),
2071 le64_to_cpu(rec
.r_cpos
) + set_len
) - cpos
;
2072 rec
.r_cpos
= cpu_to_le64(cpos
);
2073 rec
.r_clusters
= cpu_to_le32(set_len
);
2074 le32_add_cpu(&rec
.r_refcount
, 1);
2076 mlog(0, "split refcount rec, start %llu, "
2077 "len %u, count %u\n",
2078 (unsigned long long)le64_to_cpu(rec
.r_cpos
),
2079 set_len
, le32_to_cpu(rec
.r_refcount
));
2080 ret
= ocfs2_split_refcount_rec(handle
, ci
,
2081 ref_root_bh
, ref_leaf_bh
,
2092 brelse(ref_leaf_bh
);
2097 brelse(ref_leaf_bh
);
2101 static int ocfs2_remove_refcount_extent(handle_t
*handle
,
2102 struct ocfs2_caching_info
*ci
,
2103 struct buffer_head
*ref_root_bh
,
2104 struct buffer_head
*ref_leaf_bh
,
2105 struct ocfs2_alloc_context
*meta_ac
,
2106 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
2109 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
2110 struct ocfs2_refcount_block
*rb
=
2111 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
2112 struct ocfs2_extent_tree et
;
2114 BUG_ON(rb
->rf_records
.rl_used
);
2116 ocfs2_init_refcount_extent_tree(&et
, ci
, ref_root_bh
);
2117 ret
= ocfs2_remove_extent(handle
, &et
, le32_to_cpu(rb
->rf_cpos
),
2118 1, meta_ac
, dealloc
);
2124 ocfs2_remove_from_cache(ci
, ref_leaf_bh
);
2127 * add the freed block to the dealloc so that it will be freed
2128 * when we run dealloc.
2130 ret
= ocfs2_cache_block_dealloc(dealloc
, EXTENT_ALLOC_SYSTEM_INODE
,
2131 le16_to_cpu(rb
->rf_suballoc_slot
),
2132 le64_to_cpu(rb
->rf_blkno
),
2133 le16_to_cpu(rb
->rf_suballoc_bit
));
2139 ret
= ocfs2_journal_access_rb(handle
, ci
, ref_root_bh
,
2140 OCFS2_JOURNAL_ACCESS_WRITE
);
2146 rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
2148 le32_add_cpu(&rb
->rf_clusters
, -1);
2151 * check whether we need to restore the root refcount block if
2152 * there is no leaf extent block at atll.
2154 if (!rb
->rf_list
.l_next_free_rec
) {
2155 BUG_ON(rb
->rf_clusters
);
2157 mlog(0, "reset refcount tree root %llu to be a record block.\n",
2158 (unsigned long long)ref_root_bh
->b_blocknr
);
2163 memset(&rb
->rf_records
, 0, sb
->s_blocksize
-
2164 offsetof(struct ocfs2_refcount_block
, rf_records
));
2165 rb
->rf_records
.rl_count
=
2166 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb
));
2169 ocfs2_journal_dirty(handle
, ref_root_bh
);
2175 int ocfs2_increase_refcount(handle_t
*handle
,
2176 struct ocfs2_caching_info
*ci
,
2177 struct buffer_head
*ref_root_bh
,
2179 struct ocfs2_alloc_context
*meta_ac
,
2180 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
2182 return __ocfs2_increase_refcount(handle
, ci
, ref_root_bh
,
2187 static int ocfs2_decrease_refcount_rec(handle_t
*handle
,
2188 struct ocfs2_caching_info
*ci
,
2189 struct buffer_head
*ref_root_bh
,
2190 struct buffer_head
*ref_leaf_bh
,
2191 int index
, u64 cpos
, unsigned int len
,
2192 struct ocfs2_alloc_context
*meta_ac
,
2193 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
2196 struct ocfs2_refcount_block
*rb
=
2197 (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
2198 struct ocfs2_refcount_rec
*rec
= &rb
->rf_records
.rl_recs
[index
];
2200 BUG_ON(cpos
< le64_to_cpu(rec
->r_cpos
));
2202 le64_to_cpu(rec
->r_cpos
) + le32_to_cpu(rec
->r_clusters
));
2204 if (cpos
== le64_to_cpu(rec
->r_cpos
) &&
2205 len
== le32_to_cpu(rec
->r_clusters
))
2206 ret
= ocfs2_change_refcount_rec(handle
, ci
,
2207 ref_leaf_bh
, index
, 1, -1);
2209 struct ocfs2_refcount_rec split
= *rec
;
2210 split
.r_cpos
= cpu_to_le64(cpos
);
2211 split
.r_clusters
= cpu_to_le32(len
);
2213 le32_add_cpu(&split
.r_refcount
, -1);
2215 mlog(0, "split refcount rec, start %llu, "
2216 "len %u, count %u, original start %llu, len %u\n",
2217 (unsigned long long)le64_to_cpu(split
.r_cpos
),
2218 len
, le32_to_cpu(split
.r_refcount
),
2219 (unsigned long long)le64_to_cpu(rec
->r_cpos
),
2220 le32_to_cpu(rec
->r_clusters
));
2221 ret
= ocfs2_split_refcount_rec(handle
, ci
,
2222 ref_root_bh
, ref_leaf_bh
,
2232 /* Remove the leaf refcount block if it contains no refcount record. */
2233 if (!rb
->rf_records
.rl_used
&& ref_leaf_bh
!= ref_root_bh
) {
2234 ret
= ocfs2_remove_refcount_extent(handle
, ci
, ref_root_bh
,
2235 ref_leaf_bh
, meta_ac
,
2245 static int __ocfs2_decrease_refcount(handle_t
*handle
,
2246 struct ocfs2_caching_info
*ci
,
2247 struct buffer_head
*ref_root_bh
,
2249 struct ocfs2_alloc_context
*meta_ac
,
2250 struct ocfs2_cached_dealloc_ctxt
*dealloc
,
2253 int ret
= 0, index
= 0;
2254 struct ocfs2_refcount_rec rec
;
2255 unsigned int r_count
= 0, r_len
;
2256 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
2257 struct buffer_head
*ref_leaf_bh
= NULL
;
2259 mlog(0, "Tree owner %llu, decrease refcount start %llu, "
2260 "len %u, delete %u\n",
2261 (unsigned long long)ocfs2_metadata_cache_owner(ci
),
2262 (unsigned long long)cpos
, len
, delete);
2265 ret
= ocfs2_get_refcount_rec(ci
, ref_root_bh
,
2266 cpos
, len
, &rec
, &index
,
2273 r_count
= le32_to_cpu(rec
.r_refcount
);
2274 BUG_ON(r_count
== 0);
2276 BUG_ON(r_count
> 1);
2278 r_len
= min((u64
)(cpos
+ len
), le64_to_cpu(rec
.r_cpos
) +
2279 le32_to_cpu(rec
.r_clusters
)) - cpos
;
2281 ret
= ocfs2_decrease_refcount_rec(handle
, ci
, ref_root_bh
,
2290 if (le32_to_cpu(rec
.r_refcount
) == 1 && delete) {
2291 ret
= ocfs2_cache_cluster_dealloc(dealloc
,
2292 ocfs2_clusters_to_blocks(sb
, cpos
),
2302 brelse(ref_leaf_bh
);
2307 brelse(ref_leaf_bh
);
2311 /* Caller must hold refcount tree lock. */
2312 int ocfs2_decrease_refcount(struct inode
*inode
,
2313 handle_t
*handle
, u32 cpos
, u32 len
,
2314 struct ocfs2_alloc_context
*meta_ac
,
2315 struct ocfs2_cached_dealloc_ctxt
*dealloc
,
2320 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2321 struct buffer_head
*ref_root_bh
= NULL
;
2322 struct ocfs2_refcount_tree
*tree
;
2324 BUG_ON(!(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
));
2326 ret
= ocfs2_get_refcount_block(inode
, &ref_blkno
);
2332 ret
= ocfs2_get_refcount_tree(OCFS2_SB(inode
->i_sb
), ref_blkno
, &tree
);
2338 ret
= ocfs2_read_refcount_block(&tree
->rf_ci
, tree
->rf_blkno
,
2345 ret
= __ocfs2_decrease_refcount(handle
, &tree
->rf_ci
, ref_root_bh
,
2346 cpos
, len
, meta_ac
, dealloc
, delete);
2350 brelse(ref_root_bh
);
2355 * Mark the already-existing extent at cpos as refcounted for len clusters.
2356 * This adds the refcount extent flag.
2358 * If the existing extent is larger than the request, initiate a
2359 * split. An attempt will be made at merging with adjacent extents.
2361 * The caller is responsible for passing down meta_ac if we'll need it.
2363 static int ocfs2_mark_extent_refcounted(struct inode
*inode
,
2364 struct ocfs2_extent_tree
*et
,
2365 handle_t
*handle
, u32 cpos
,
2367 struct ocfs2_alloc_context
*meta_ac
,
2368 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
2372 mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
2373 inode
->i_ino
, cpos
, len
, phys
);
2375 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
))) {
2376 ocfs2_error(inode
->i_sb
, "Inode %lu want to use refcount "
2377 "tree, but the feature bit is not set in the "
2378 "super block.", inode
->i_ino
);
2383 ret
= ocfs2_change_extent_flag(handle
, et
, cpos
,
2384 len
, phys
, meta_ac
, dealloc
,
2385 OCFS2_EXT_REFCOUNTED
, 0);
2394 * Given some contiguous physical clusters, calculate what we need
2395 * for modifying their refcount.
2397 static int ocfs2_calc_refcount_meta_credits(struct super_block
*sb
,
2398 struct ocfs2_caching_info
*ci
,
2399 struct buffer_head
*ref_root_bh
,
2405 int ret
= 0, index
, ref_blocks
= 0, recs_add
= 0;
2406 u64 cpos
= start_cpos
;
2407 struct ocfs2_refcount_block
*rb
;
2408 struct ocfs2_refcount_rec rec
;
2409 struct buffer_head
*ref_leaf_bh
= NULL
, *prev_bh
= NULL
;
2412 mlog(0, "start_cpos %llu, clusters %u\n",
2413 (unsigned long long)start_cpos
, clusters
);
2415 ret
= ocfs2_get_refcount_rec(ci
, ref_root_bh
,
2416 cpos
, clusters
, &rec
,
2417 &index
, &ref_leaf_bh
);
2423 if (ref_leaf_bh
!= prev_bh
) {
2425 * Now we encounter a new leaf block, so calculate
2426 * whether we need to extend the old leaf.
2429 rb
= (struct ocfs2_refcount_block
*)
2432 if (le64_to_cpu(rb
->rf_records
.rl_used
) +
2434 le16_to_cpu(rb
->rf_records
.rl_count
))
2441 prev_bh
= ref_leaf_bh
;
2445 rb
= (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
2447 mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
2448 "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
2449 recs_add
, (unsigned long long)cpos
, clusters
,
2450 (unsigned long long)le64_to_cpu(rec
.r_cpos
),
2451 le32_to_cpu(rec
.r_clusters
),
2452 le32_to_cpu(rec
.r_refcount
), index
);
2454 len
= min((u64
)cpos
+ clusters
, le64_to_cpu(rec
.r_cpos
) +
2455 le32_to_cpu(rec
.r_clusters
)) - cpos
;
2457 * We record all the records which will be inserted to the
2458 * same refcount block, so that we can tell exactly whether
2459 * we need a new refcount block or not.
2461 * If we will insert a new one, this is easy and only happens
2462 * during adding refcounted flag to the extent, so we don't
2463 * have a chance of spliting. We just need one record.
2465 * If the refcount rec already exists, that would be a little
2466 * complicated. we may have to:
2467 * 1) split at the beginning if the start pos isn't aligned.
2468 * we need 1 more record in this case.
2469 * 2) split int the end if the end pos isn't aligned.
2470 * we need 1 more record in this case.
2471 * 3) split in the middle because of file system fragmentation.
2472 * we need 2 more records in this case(we can't detect this
2473 * beforehand, so always think of the worst case).
2475 if (rec
.r_refcount
) {
2477 /* Check whether we need a split at the beginning. */
2478 if (cpos
== start_cpos
&&
2479 cpos
!= le64_to_cpu(rec
.r_cpos
))
2482 /* Check whether we need a split in the end. */
2483 if (cpos
+ clusters
< le64_to_cpu(rec
.r_cpos
) +
2484 le32_to_cpu(rec
.r_clusters
))
2489 brelse(ref_leaf_bh
);
2496 rb
= (struct ocfs2_refcount_block
*)prev_bh
->b_data
;
2498 if (le64_to_cpu(rb
->rf_records
.rl_used
) + recs_add
>
2499 le16_to_cpu(rb
->rf_records
.rl_count
))
2508 mlog(0, "we need ref_blocks %d\n", ref_blocks
);
2509 *meta_add
+= ref_blocks
;
2510 *credits
+= ref_blocks
;
2513 * So we may need ref_blocks to insert into the tree.
2514 * That also means we need to change the b-tree and add that number
2515 * of records since we never merge them.
2516 * We need one more block for expansion since the new created leaf
2517 * block is also full and needs split.
2519 rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
2520 if (le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
) {
2521 struct ocfs2_extent_tree et
;
2523 ocfs2_init_refcount_extent_tree(&et
, ci
, ref_root_bh
);
2524 *meta_add
+= ocfs2_extend_meta_needed(et
.et_root_el
);
2525 *credits
+= ocfs2_calc_extend_credits(sb
,
2529 *credits
+= OCFS2_EXPAND_REFCOUNT_TREE_CREDITS
;
2534 brelse(ref_leaf_bh
);
2540 * For refcount tree, we will decrease some contiguous clusters
2541 * refcount count, so just go through it to see how many blocks
2542 * we gonna touch and whether we need to create new blocks.
2544 * Normally the refcount blocks store these refcount should be
2545 * continguous also, so that we can get the number easily.
2546 * As for meta_ac, we will at most add split 2 refcount record and
2547 * 2 more refcount block, so just check it in a rough way.
2549 * Caller must hold refcount tree lock.
2551 int ocfs2_prepare_refcount_change_for_del(struct inode
*inode
,
2552 struct buffer_head
*di_bh
,
2556 struct ocfs2_alloc_context
**meta_ac
)
2558 int ret
, ref_blocks
= 0;
2559 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2560 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
2561 struct buffer_head
*ref_root_bh
= NULL
;
2562 struct ocfs2_refcount_tree
*tree
;
2563 u64 start_cpos
= ocfs2_blocks_to_clusters(inode
->i_sb
, phys_blkno
);
2565 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
))) {
2566 ocfs2_error(inode
->i_sb
, "Inode %lu want to use refcount "
2567 "tree, but the feature bit is not set in the "
2568 "super block.", inode
->i_ino
);
2573 BUG_ON(!(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
));
2575 ret
= ocfs2_get_refcount_tree(OCFS2_SB(inode
->i_sb
),
2576 le64_to_cpu(di
->i_refcount_loc
), &tree
);
2582 ret
= ocfs2_read_refcount_block(&tree
->rf_ci
,
2583 le64_to_cpu(di
->i_refcount_loc
),
2590 ret
= ocfs2_calc_refcount_meta_credits(inode
->i_sb
,
2593 start_cpos
, clusters
,
2594 &ref_blocks
, credits
);
2600 mlog(0, "reserve new metadata %d, credits = %d\n",
2601 ref_blocks
, *credits
);
2604 ret
= ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode
->i_sb
),
2605 ref_blocks
, meta_ac
);
2611 brelse(ref_root_bh
);
2615 #define MAX_CONTIG_BYTES 1048576
2617 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block
*sb
)
2619 return ocfs2_clusters_for_bytes(sb
, MAX_CONTIG_BYTES
);
2622 static inline unsigned int ocfs2_cow_contig_mask(struct super_block
*sb
)
2624 return ~(ocfs2_cow_contig_clusters(sb
) - 1);
2628 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2629 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2630 * while still being less than or equal to it.
2632 * The goal is to break the extent at a multiple of contig_clusters.
2634 static inline unsigned int ocfs2_cow_align_start(struct super_block
*sb
,
2638 BUG_ON(start
> cpos
);
2640 return start
+ ((cpos
- start
) & ocfs2_cow_contig_mask(sb
));
2644 * Given a cluster count of len, pad it out so that it is a multiple
2645 * of contig_clusters.
2647 static inline unsigned int ocfs2_cow_align_length(struct super_block
*sb
,
2650 unsigned int padded
=
2651 (len
+ (ocfs2_cow_contig_clusters(sb
) - 1)) &
2652 ocfs2_cow_contig_mask(sb
);
2662 * Calculate out the start and number of virtual clusters we need to to CoW.
2664 * cpos is vitual start cluster position we want to do CoW in a
2665 * file and write_len is the cluster length.
2666 * max_cpos is the place where we want to stop CoW intentionally.
2668 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2669 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2670 * get good I/O from the resulting extent tree.
2672 static int ocfs2_refcount_cal_cow_clusters(struct inode
*inode
,
2673 struct ocfs2_extent_list
*el
,
2681 int tree_height
= le16_to_cpu(el
->l_tree_depth
), i
;
2682 struct buffer_head
*eb_bh
= NULL
;
2683 struct ocfs2_extent_block
*eb
= NULL
;
2684 struct ocfs2_extent_rec
*rec
;
2685 unsigned int want_clusters
, rec_end
= 0;
2686 int contig_clusters
= ocfs2_cow_contig_clusters(inode
->i_sb
);
2689 BUG_ON(cpos
+ write_len
> max_cpos
);
2691 if (tree_height
> 0) {
2692 ret
= ocfs2_find_leaf(INODE_CACHE(inode
), el
, cpos
, &eb_bh
);
2698 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
2701 if (el
->l_tree_depth
) {
2702 ocfs2_error(inode
->i_sb
,
2703 "Inode %lu has non zero tree depth in "
2704 "leaf block %llu\n", inode
->i_ino
,
2705 (unsigned long long)eb_bh
->b_blocknr
);
2712 for (i
= 0; i
< le16_to_cpu(el
->l_next_free_rec
); i
++) {
2713 rec
= &el
->l_recs
[i
];
2715 if (ocfs2_is_empty_extent(rec
)) {
2716 mlog_bug_on_msg(i
!= 0, "Inode %lu has empty record in "
2717 "index %d\n", inode
->i_ino
, i
);
2721 if (le32_to_cpu(rec
->e_cpos
) +
2722 le16_to_cpu(rec
->e_leaf_clusters
) <= cpos
)
2725 if (*cow_len
== 0) {
2727 * We should find a refcounted record in the
2730 BUG_ON(!(rec
->e_flags
& OCFS2_EXT_REFCOUNTED
));
2731 *cow_start
= le32_to_cpu(rec
->e_cpos
);
2735 * If we encounter a hole, a non-refcounted record or
2736 * pass the max_cpos, stop the search.
2738 if ((!(rec
->e_flags
& OCFS2_EXT_REFCOUNTED
)) ||
2739 (*cow_len
&& rec_end
!= le32_to_cpu(rec
->e_cpos
)) ||
2740 (max_cpos
<= le32_to_cpu(rec
->e_cpos
)))
2743 leaf_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
2744 rec_end
= le32_to_cpu(rec
->e_cpos
) + leaf_clusters
;
2745 if (rec_end
> max_cpos
) {
2747 leaf_clusters
= rec_end
- le32_to_cpu(rec
->e_cpos
);
2751 * How many clusters do we actually need from
2752 * this extent? First we see how many we actually
2753 * need to complete the write. If that's smaller
2754 * than contig_clusters, we try for contig_clusters.
2757 want_clusters
= write_len
;
2759 want_clusters
= (cpos
+ write_len
) -
2760 (*cow_start
+ *cow_len
);
2761 if (want_clusters
< contig_clusters
)
2762 want_clusters
= contig_clusters
;
2765 * If the write does not cover the whole extent, we
2766 * need to calculate how we're going to split the extent.
2767 * We try to do it on contig_clusters boundaries.
2769 * Any extent smaller than contig_clusters will be
2770 * CoWed in its entirety.
2772 if (leaf_clusters
<= contig_clusters
)
2773 *cow_len
+= leaf_clusters
;
2774 else if (*cow_len
|| (*cow_start
== cpos
)) {
2776 * This extent needs to be CoW'd from its
2777 * beginning, so all we have to do is compute
2778 * how many clusters to grab. We align
2779 * want_clusters to the edge of contig_clusters
2780 * to get better I/O.
2782 want_clusters
= ocfs2_cow_align_length(inode
->i_sb
,
2785 if (leaf_clusters
< want_clusters
)
2786 *cow_len
+= leaf_clusters
;
2788 *cow_len
+= want_clusters
;
2789 } else if ((*cow_start
+ contig_clusters
) >=
2790 (cpos
+ write_len
)) {
2792 * Breaking off contig_clusters at the front
2793 * of the extent will cover our write. That's
2796 *cow_len
= contig_clusters
;
2797 } else if ((rec_end
- cpos
) <= contig_clusters
) {
2799 * Breaking off contig_clusters at the tail of
2800 * this extent will cover cpos.
2802 *cow_start
= rec_end
- contig_clusters
;
2803 *cow_len
= contig_clusters
;
2804 } else if ((rec_end
- cpos
) <= want_clusters
) {
2806 * While we can't fit the entire write in this
2807 * extent, we know that the write goes from cpos
2808 * to the end of the extent. Break that off.
2809 * We try to break it at some multiple of
2810 * contig_clusters from the front of the extent.
2811 * Failing that (ie, cpos is within
2812 * contig_clusters of the front), we'll CoW the
2815 *cow_start
= ocfs2_cow_align_start(inode
->i_sb
,
2817 *cow_len
= rec_end
- *cow_start
;
2820 * Ok, the entire write lives in the middle of
2821 * this extent. Let's try to slice the extent up
2822 * nicely. Optimally, our CoW region starts at
2823 * m*contig_clusters from the beginning of the
2824 * extent and goes for n*contig_clusters,
2825 * covering the entire write.
2827 *cow_start
= ocfs2_cow_align_start(inode
->i_sb
,
2830 want_clusters
= (cpos
+ write_len
) - *cow_start
;
2831 want_clusters
= ocfs2_cow_align_length(inode
->i_sb
,
2833 if (*cow_start
+ want_clusters
<= rec_end
)
2834 *cow_len
= want_clusters
;
2836 *cow_len
= rec_end
- *cow_start
;
2839 /* Have we covered our entire write yet? */
2840 if ((*cow_start
+ *cow_len
) >= (cpos
+ write_len
))
2844 * If we reach the end of the extent block and don't get enough
2845 * clusters, continue with the next extent block if possible.
2847 if (i
+ 1 == le16_to_cpu(el
->l_next_free_rec
) &&
2848 eb
&& eb
->h_next_leaf_blk
) {
2852 ret
= ocfs2_read_extent_block(INODE_CACHE(inode
),
2853 le64_to_cpu(eb
->h_next_leaf_blk
),
2860 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
2872 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2873 * num_clusters in data_tree "et" and change the refcount for the old
2874 * clusters(starting form p_cluster) in the refcount tree.
2877 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2878 * more new leaf records.
2879 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2880 * just give data_ac = NULL.
2882 static int ocfs2_lock_refcount_allocators(struct super_block
*sb
,
2883 u32 p_cluster
, u32 num_clusters
,
2884 struct ocfs2_extent_tree
*et
,
2885 struct ocfs2_caching_info
*ref_ci
,
2886 struct buffer_head
*ref_root_bh
,
2887 struct ocfs2_alloc_context
**meta_ac
,
2888 struct ocfs2_alloc_context
**data_ac
,
2891 int ret
= 0, meta_add
= 0;
2892 int num_free_extents
= ocfs2_num_free_extents(OCFS2_SB(sb
), et
);
2894 if (num_free_extents
< 0) {
2895 ret
= num_free_extents
;
2900 if (num_free_extents
< num_clusters
+ 2)
2902 ocfs2_extend_meta_needed(et
->et_root_el
);
2904 *credits
+= ocfs2_calc_extend_credits(sb
, et
->et_root_el
,
2907 ret
= ocfs2_calc_refcount_meta_credits(sb
, ref_ci
, ref_root_bh
,
2908 p_cluster
, num_clusters
,
2909 &meta_add
, credits
);
2915 mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
2916 meta_add
, num_clusters
, *credits
);
2917 ret
= ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb
), meta_add
,
2925 ret
= ocfs2_reserve_clusters(OCFS2_SB(sb
), num_clusters
,
2934 ocfs2_free_alloc_context(*meta_ac
);
2942 static int ocfs2_clear_cow_buffer(handle_t
*handle
, struct buffer_head
*bh
)
2944 BUG_ON(buffer_dirty(bh
));
2946 clear_buffer_mapped(bh
);
2951 static int ocfs2_duplicate_clusters_by_page(handle_t
*handle
,
2952 struct ocfs2_cow_context
*context
,
2953 u32 cpos
, u32 old_cluster
,
2954 u32 new_cluster
, u32 new_len
)
2956 int ret
= 0, partial
;
2957 struct ocfs2_caching_info
*ci
= context
->data_et
.et_ci
;
2958 struct super_block
*sb
= ocfs2_metadata_cache_get_super(ci
);
2959 u64 new_block
= ocfs2_clusters_to_blocks(sb
, new_cluster
);
2962 unsigned int from
, to
;
2963 loff_t offset
, end
, map_end
;
2964 struct address_space
*mapping
= context
->inode
->i_mapping
;
2966 mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster
,
2967 new_cluster
, new_len
, cpos
);
2969 offset
= ((loff_t
)cpos
) << OCFS2_SB(sb
)->s_clustersize_bits
;
2970 end
= offset
+ (new_len
<< OCFS2_SB(sb
)->s_clustersize_bits
);
2972 while (offset
< end
) {
2973 page_index
= offset
>> PAGE_CACHE_SHIFT
;
2974 map_end
= (page_index
+ 1) << PAGE_CACHE_SHIFT
;
2978 /* from, to is the offset within the page. */
2979 from
= offset
& (PAGE_CACHE_SIZE
- 1);
2980 to
= PAGE_CACHE_SIZE
;
2981 if (map_end
& (PAGE_CACHE_SIZE
- 1))
2982 to
= map_end
& (PAGE_CACHE_SIZE
- 1);
2984 page
= grab_cache_page(mapping
, page_index
);
2986 /* This page can't be dirtied before we CoW it out. */
2987 BUG_ON(PageDirty(page
));
2989 if (!PageUptodate(page
)) {
2990 ret
= block_read_full_page(page
, ocfs2_get_block
);
2998 if (page_has_buffers(page
)) {
2999 ret
= walk_page_buffers(handle
, page_buffers(page
),
3001 ocfs2_clear_cow_buffer
);
3008 ocfs2_map_and_dirty_page(context
->inode
,
3010 page
, 0, &new_block
);
3011 mark_page_accessed(page
);
3014 page_cache_release(page
);
3024 static int ocfs2_duplicate_clusters_by_jbd(handle_t
*handle
,
3025 struct ocfs2_cow_context
*context
,
3026 u32 cpos
, u32 old_cluster
,
3027 u32 new_cluster
, u32 new_len
)
3030 struct super_block
*sb
= context
->inode
->i_sb
;
3031 struct ocfs2_caching_info
*ci
= context
->data_et
.et_ci
;
3032 int i
, blocks
= ocfs2_clusters_to_blocks(sb
, new_len
);
3033 u64 old_block
= ocfs2_clusters_to_blocks(sb
, old_cluster
);
3034 u64 new_block
= ocfs2_clusters_to_blocks(sb
, new_cluster
);
3035 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
3036 struct buffer_head
*old_bh
= NULL
;
3037 struct buffer_head
*new_bh
= NULL
;
3039 mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster
,
3040 new_cluster
, new_len
);
3042 for (i
= 0; i
< blocks
; i
++, old_block
++, new_block
++) {
3043 new_bh
= sb_getblk(osb
->sb
, new_block
);
3044 if (new_bh
== NULL
) {
3050 ocfs2_set_new_buffer_uptodate(ci
, new_bh
);
3052 ret
= ocfs2_read_block(ci
, old_block
, &old_bh
, NULL
);
3058 ret
= ocfs2_journal_access(handle
, ci
, new_bh
,
3059 OCFS2_JOURNAL_ACCESS_CREATE
);
3065 memcpy(new_bh
->b_data
, old_bh
->b_data
, sb
->s_blocksize
);
3066 ret
= ocfs2_journal_dirty(handle
, new_bh
);
3083 static int ocfs2_clear_ext_refcount(handle_t
*handle
,
3084 struct ocfs2_extent_tree
*et
,
3085 u32 cpos
, u32 p_cluster
, u32 len
,
3086 unsigned int ext_flags
,
3087 struct ocfs2_alloc_context
*meta_ac
,
3088 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
3091 struct ocfs2_extent_rec replace_rec
;
3092 struct ocfs2_path
*path
= NULL
;
3093 struct ocfs2_extent_list
*el
;
3094 struct super_block
*sb
= ocfs2_metadata_cache_get_super(et
->et_ci
);
3095 u64 ino
= ocfs2_metadata_cache_owner(et
->et_ci
);
3097 mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
3098 (unsigned long long)ino
, cpos
, len
, p_cluster
, ext_flags
);
3100 memset(&replace_rec
, 0, sizeof(replace_rec
));
3101 replace_rec
.e_cpos
= cpu_to_le32(cpos
);
3102 replace_rec
.e_leaf_clusters
= cpu_to_le16(len
);
3103 replace_rec
.e_blkno
= cpu_to_le64(ocfs2_clusters_to_blocks(sb
,
3105 replace_rec
.e_flags
= ext_flags
;
3106 replace_rec
.e_flags
&= ~OCFS2_EXT_REFCOUNTED
;
3108 path
= ocfs2_new_path_from_et(et
);
3115 ret
= ocfs2_find_path(et
->et_ci
, path
, cpos
);
3121 el
= path_leaf_el(path
);
3123 index
= ocfs2_search_extent_list(el
, cpos
);
3124 if (index
== -1 || index
>= le16_to_cpu(el
->l_next_free_rec
)) {
3126 "Inode %llu has an extent at cpos %u which can no "
3127 "longer be found.\n",
3128 (unsigned long long)ino
, cpos
);
3133 ret
= ocfs2_split_extent(handle
, et
, path
, index
,
3134 &replace_rec
, meta_ac
, dealloc
);
3139 ocfs2_free_path(path
);
3143 static int ocfs2_replace_clusters(handle_t
*handle
,
3144 struct ocfs2_cow_context
*context
,
3147 unsigned int ext_flags
)
3150 struct ocfs2_caching_info
*ci
= context
->data_et
.et_ci
;
3151 u64 ino
= ocfs2_metadata_cache_owner(ci
);
3153 mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
3154 (unsigned long long)ino
, cpos
, old
, new, len
, ext_flags
);
3156 /*If the old clusters is unwritten, no need to duplicate. */
3157 if (!(ext_flags
& OCFS2_EXT_UNWRITTEN
)) {
3158 ret
= context
->cow_duplicate_clusters(handle
, context
, cpos
,
3166 ret
= ocfs2_clear_ext_refcount(handle
, &context
->data_et
,
3167 cpos
, new, len
, ext_flags
,
3168 context
->meta_ac
, &context
->dealloc
);
3175 static int ocfs2_cow_sync_writeback(struct super_block
*sb
,
3176 struct ocfs2_cow_context
*context
,
3177 u32 cpos
, u32 num_clusters
)
3180 loff_t offset
, end
, map_end
;
3184 if (ocfs2_should_order_data(context
->inode
))
3187 offset
= ((loff_t
)cpos
) << OCFS2_SB(sb
)->s_clustersize_bits
;
3188 end
= offset
+ (num_clusters
<< OCFS2_SB(sb
)->s_clustersize_bits
);
3190 ret
= filemap_fdatawrite_range(context
->inode
->i_mapping
,
3197 while (offset
< end
) {
3198 page_index
= offset
>> PAGE_CACHE_SHIFT
;
3199 map_end
= (page_index
+ 1) << PAGE_CACHE_SHIFT
;
3203 page
= grab_cache_page(context
->inode
->i_mapping
, page_index
);
3206 wait_on_page_writeback(page
);
3207 if (PageError(page
)) {
3211 mark_page_accessed(page
);
3214 page_cache_release(page
);
3224 static int ocfs2_di_get_clusters(struct ocfs2_cow_context
*context
,
3225 u32 v_cluster
, u32
*p_cluster
,
3227 unsigned int *extent_flags
)
3229 return ocfs2_get_clusters(context
->inode
, v_cluster
, p_cluster
,
3230 num_clusters
, extent_flags
);
3233 static int ocfs2_make_clusters_writable(struct super_block
*sb
,
3234 struct ocfs2_cow_context
*context
,
3235 u32 cpos
, u32 p_cluster
,
3236 u32 num_clusters
, unsigned int e_flags
)
3238 int ret
, delete, index
, credits
= 0;
3239 u32 new_bit
, new_len
;
3240 unsigned int set_len
;
3241 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
3243 struct buffer_head
*ref_leaf_bh
= NULL
;
3244 struct ocfs2_caching_info
*ref_ci
= &context
->ref_tree
->rf_ci
;
3245 struct ocfs2_refcount_rec rec
;
3247 mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
3248 cpos
, p_cluster
, num_clusters
, e_flags
);
3250 ret
= ocfs2_lock_refcount_allocators(sb
, p_cluster
, num_clusters
,
3253 context
->ref_root_bh
,
3255 &context
->data_ac
, &credits
);
3261 if (context
->post_refcount
)
3262 credits
+= context
->post_refcount
->credits
;
3264 credits
+= context
->extra_credits
;
3265 handle
= ocfs2_start_trans(osb
, credits
);
3266 if (IS_ERR(handle
)) {
3267 ret
= PTR_ERR(handle
);
3272 while (num_clusters
) {
3273 ret
= ocfs2_get_refcount_rec(ref_ci
, context
->ref_root_bh
,
3274 p_cluster
, num_clusters
,
3275 &rec
, &index
, &ref_leaf_bh
);
3281 BUG_ON(!rec
.r_refcount
);
3282 set_len
= min((u64
)p_cluster
+ num_clusters
,
3283 le64_to_cpu(rec
.r_cpos
) +
3284 le32_to_cpu(rec
.r_clusters
)) - p_cluster
;
3287 * There are many different situation here.
3288 * 1. If refcount == 1, remove the flag and don't COW.
3289 * 2. If refcount > 1, allocate clusters.
3290 * Here we may not allocate r_len once at a time, so continue
3291 * until we reach num_clusters.
3293 if (le32_to_cpu(rec
.r_refcount
) == 1) {
3295 ret
= ocfs2_clear_ext_refcount(handle
,
3308 ret
= __ocfs2_claim_clusters(osb
, handle
,
3311 &new_bit
, &new_len
);
3317 ret
= ocfs2_replace_clusters(handle
, context
,
3318 cpos
, p_cluster
, new_bit
,
3327 ret
= __ocfs2_decrease_refcount(handle
, ref_ci
,
3328 context
->ref_root_bh
,
3331 &context
->dealloc
, delete);
3338 p_cluster
+= set_len
;
3339 num_clusters
-= set_len
;
3340 brelse(ref_leaf_bh
);
3344 /* handle any post_cow action. */
3345 if (context
->post_refcount
&& context
->post_refcount
->func
) {
3346 ret
= context
->post_refcount
->func(context
->inode
, handle
,
3347 context
->post_refcount
->para
);
3355 * Here we should write the new page out first if we are
3356 * in write-back mode.
3358 if (context
->get_clusters
== ocfs2_di_get_clusters
) {
3359 ret
= ocfs2_cow_sync_writeback(sb
, context
, cpos
, num_clusters
);
3365 ocfs2_commit_trans(osb
, handle
);
3368 if (context
->data_ac
) {
3369 ocfs2_free_alloc_context(context
->data_ac
);
3370 context
->data_ac
= NULL
;
3372 if (context
->meta_ac
) {
3373 ocfs2_free_alloc_context(context
->meta_ac
);
3374 context
->meta_ac
= NULL
;
3376 brelse(ref_leaf_bh
);
3381 static int ocfs2_replace_cow(struct ocfs2_cow_context
*context
)
3384 struct inode
*inode
= context
->inode
;
3385 u32 cow_start
= context
->cow_start
, cow_len
= context
->cow_len
;
3386 u32 p_cluster
, num_clusters
;
3387 unsigned int ext_flags
;
3388 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3390 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
))) {
3391 ocfs2_error(inode
->i_sb
, "Inode %lu want to use refcount "
3392 "tree, but the feature bit is not set in the "
3393 "super block.", inode
->i_ino
);
3397 ocfs2_init_dealloc_ctxt(&context
->dealloc
);
3400 ret
= context
->get_clusters(context
, cow_start
, &p_cluster
,
3401 &num_clusters
, &ext_flags
);
3407 BUG_ON(!(ext_flags
& OCFS2_EXT_REFCOUNTED
));
3409 if (cow_len
< num_clusters
)
3410 num_clusters
= cow_len
;
3412 ret
= ocfs2_make_clusters_writable(inode
->i_sb
, context
,
3413 cow_start
, p_cluster
,
3414 num_clusters
, ext_flags
);
3420 cow_len
-= num_clusters
;
3421 cow_start
+= num_clusters
;
3424 if (ocfs2_dealloc_has_cluster(&context
->dealloc
)) {
3425 ocfs2_schedule_truncate_log_flush(osb
, 1);
3426 ocfs2_run_deallocs(osb
, &context
->dealloc
);
3433 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3434 * past max_cpos. This will stop when it runs into a hole or an
3435 * unrefcounted extent.
3437 static int ocfs2_refcount_cow_hunk(struct inode
*inode
,
3438 struct buffer_head
*di_bh
,
3439 u32 cpos
, u32 write_len
, u32 max_cpos
)
3442 u32 cow_start
= 0, cow_len
= 0;
3443 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
3444 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3445 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3446 struct buffer_head
*ref_root_bh
= NULL
;
3447 struct ocfs2_refcount_tree
*ref_tree
;
3448 struct ocfs2_cow_context
*context
= NULL
;
3450 BUG_ON(!(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
));
3452 ret
= ocfs2_refcount_cal_cow_clusters(inode
, &di
->id2
.i_list
,
3453 cpos
, write_len
, max_cpos
,
3454 &cow_start
, &cow_len
);
3460 mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
3461 "cow_len %u\n", inode
->i_ino
,
3462 cpos
, write_len
, cow_start
, cow_len
);
3464 BUG_ON(cow_len
== 0);
3466 context
= kzalloc(sizeof(struct ocfs2_cow_context
), GFP_NOFS
);
3473 ret
= ocfs2_lock_refcount_tree(osb
, le64_to_cpu(di
->i_refcount_loc
),
3474 1, &ref_tree
, &ref_root_bh
);
3480 context
->inode
= inode
;
3481 context
->cow_start
= cow_start
;
3482 context
->cow_len
= cow_len
;
3483 context
->ref_tree
= ref_tree
;
3484 context
->ref_root_bh
= ref_root_bh
;
3485 context
->cow_duplicate_clusters
= ocfs2_duplicate_clusters_by_page
;
3486 context
->get_clusters
= ocfs2_di_get_clusters
;
3488 ocfs2_init_dinode_extent_tree(&context
->data_et
,
3489 INODE_CACHE(inode
), di_bh
);
3491 ret
= ocfs2_replace_cow(context
);
3496 * truncate the extent map here since no matter whether we meet with
3497 * any error during the action, we shouldn't trust cached extent map
3500 ocfs2_extent_map_trunc(inode
, cow_start
);
3502 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
3503 brelse(ref_root_bh
);
3510 * CoW any and all clusters between cpos and cpos+write_len.
3511 * Don't CoW past max_cpos. If this returns successfully, all
3512 * clusters between cpos and cpos+write_len are safe to modify.
3514 int ocfs2_refcount_cow(struct inode
*inode
,
3515 struct buffer_head
*di_bh
,
3516 u32 cpos
, u32 write_len
, u32 max_cpos
)
3519 u32 p_cluster
, num_clusters
;
3520 unsigned int ext_flags
;
3523 ret
= ocfs2_get_clusters(inode
, cpos
, &p_cluster
,
3524 &num_clusters
, &ext_flags
);
3530 if (write_len
< num_clusters
)
3531 num_clusters
= write_len
;
3533 if (ext_flags
& OCFS2_EXT_REFCOUNTED
) {
3534 ret
= ocfs2_refcount_cow_hunk(inode
, di_bh
, cpos
,
3535 num_clusters
, max_cpos
);
3542 write_len
-= num_clusters
;
3543 cpos
+= num_clusters
;
3549 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context
*context
,
3550 u32 v_cluster
, u32
*p_cluster
,
3552 unsigned int *extent_flags
)
3554 struct inode
*inode
= context
->inode
;
3555 struct ocfs2_xattr_value_root
*xv
= context
->cow_object
;
3557 return ocfs2_xattr_get_clusters(inode
, v_cluster
, p_cluster
,
3558 num_clusters
, &xv
->xr_list
,
3563 * Given a xattr value root, calculate the most meta/credits we need for
3564 * refcount tree change if we truncate it to 0.
3566 int ocfs2_refcounted_xattr_delete_need(struct inode
*inode
,
3567 struct ocfs2_caching_info
*ref_ci
,
3568 struct buffer_head
*ref_root_bh
,
3569 struct ocfs2_xattr_value_root
*xv
,
3570 int *meta_add
, int *credits
)
3572 int ret
= 0, index
, ref_blocks
= 0;
3573 u32 p_cluster
, num_clusters
;
3574 u32 cpos
= 0, clusters
= le32_to_cpu(xv
->xr_clusters
);
3575 struct ocfs2_refcount_block
*rb
;
3576 struct ocfs2_refcount_rec rec
;
3577 struct buffer_head
*ref_leaf_bh
= NULL
;
3579 while (cpos
< clusters
) {
3580 ret
= ocfs2_xattr_get_clusters(inode
, cpos
, &p_cluster
,
3581 &num_clusters
, &xv
->xr_list
,
3588 cpos
+= num_clusters
;
3590 while (num_clusters
) {
3591 ret
= ocfs2_get_refcount_rec(ref_ci
, ref_root_bh
,
3592 p_cluster
, num_clusters
,
3600 BUG_ON(!rec
.r_refcount
);
3602 rb
= (struct ocfs2_refcount_block
*)ref_leaf_bh
->b_data
;
3605 * We really don't know whether the other clusters is in
3606 * this refcount block or not, so just take the worst
3607 * case that all the clusters are in this block and each
3608 * one will split a refcount rec, so totally we need
3609 * clusters * 2 new refcount rec.
3611 if (le64_to_cpu(rb
->rf_records
.rl_used
) + clusters
* 2 >
3612 le16_to_cpu(rb
->rf_records
.rl_count
))
3616 brelse(ref_leaf_bh
);
3619 if (num_clusters
<= le32_to_cpu(rec
.r_clusters
))
3622 num_clusters
-= le32_to_cpu(rec
.r_clusters
);
3623 p_cluster
+= num_clusters
;
3627 *meta_add
+= ref_blocks
;
3631 rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
3632 if (le32_to_cpu(rb
->rf_flags
) & OCFS2_REFCOUNT_TREE_FL
)
3633 *credits
+= OCFS2_EXPAND_REFCOUNT_TREE_CREDITS
;
3635 struct ocfs2_extent_tree et
;
3637 ocfs2_init_refcount_extent_tree(&et
, ref_ci
, ref_root_bh
);
3638 *credits
+= ocfs2_calc_extend_credits(inode
->i_sb
,
3644 brelse(ref_leaf_bh
);
3651 int ocfs2_refcount_cow_xattr(struct inode
*inode
,
3652 struct ocfs2_dinode
*di
,
3653 struct ocfs2_xattr_value_buf
*vb
,
3654 struct ocfs2_refcount_tree
*ref_tree
,
3655 struct buffer_head
*ref_root_bh
,
3656 u32 cpos
, u32 write_len
,
3657 struct ocfs2_post_refcount
*post
)
3660 struct ocfs2_xattr_value_root
*xv
= vb
->vb_xv
;
3661 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
3662 struct ocfs2_cow_context
*context
= NULL
;
3663 u32 cow_start
, cow_len
;
3665 BUG_ON(!(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
));
3667 ret
= ocfs2_refcount_cal_cow_clusters(inode
, &xv
->xr_list
,
3668 cpos
, write_len
, UINT_MAX
,
3669 &cow_start
, &cow_len
);
3675 BUG_ON(cow_len
== 0);
3677 context
= kzalloc(sizeof(struct ocfs2_cow_context
), GFP_NOFS
);
3684 context
->inode
= inode
;
3685 context
->cow_start
= cow_start
;
3686 context
->cow_len
= cow_len
;
3687 context
->ref_tree
= ref_tree
;
3688 context
->ref_root_bh
= ref_root_bh
;;
3689 context
->cow_object
= xv
;
3691 context
->cow_duplicate_clusters
= ocfs2_duplicate_clusters_by_jbd
;
3692 /* We need the extra credits for duplicate_clusters by jbd. */
3693 context
->extra_credits
=
3694 ocfs2_clusters_to_blocks(inode
->i_sb
, 1) * cow_len
;
3695 context
->get_clusters
= ocfs2_xattr_value_get_clusters
;
3696 context
->post_refcount
= post
;
3698 ocfs2_init_xattr_value_extent_tree(&context
->data_et
,
3699 INODE_CACHE(inode
), vb
);
3701 ret
= ocfs2_replace_cow(context
);
3711 * Insert a new extent into refcount tree and mark a extent rec
3712 * as refcounted in the dinode tree.
3714 int ocfs2_add_refcount_flag(struct inode
*inode
,
3715 struct ocfs2_extent_tree
*data_et
,
3716 struct ocfs2_caching_info
*ref_ci
,
3717 struct buffer_head
*ref_root_bh
,
3718 u32 cpos
, u32 p_cluster
, u32 num_clusters
,
3719 struct ocfs2_cached_dealloc_ctxt
*dealloc
,
3720 struct ocfs2_post_refcount
*post
)
3724 int credits
= 1, ref_blocks
= 0;
3725 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3726 struct ocfs2_alloc_context
*meta_ac
= NULL
;
3728 ret
= ocfs2_calc_refcount_meta_credits(inode
->i_sb
,
3729 ref_ci
, ref_root_bh
,
3730 p_cluster
, num_clusters
,
3731 &ref_blocks
, &credits
);
3737 mlog(0, "reserve new metadata %d, credits = %d\n",
3738 ref_blocks
, credits
);
3741 ret
= ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode
->i_sb
),
3742 ref_blocks
, &meta_ac
);
3750 credits
+= post
->credits
;
3752 handle
= ocfs2_start_trans(osb
, credits
);
3753 if (IS_ERR(handle
)) {
3754 ret
= PTR_ERR(handle
);
3759 ret
= ocfs2_mark_extent_refcounted(inode
, data_et
, handle
,
3760 cpos
, num_clusters
, p_cluster
,
3767 ret
= __ocfs2_increase_refcount(handle
, ref_ci
, ref_root_bh
,
3768 p_cluster
, num_clusters
, 0,
3775 if (post
&& post
->func
) {
3776 ret
= post
->func(inode
, handle
, post
->para
);
3782 ocfs2_commit_trans(osb
, handle
);
3785 ocfs2_free_alloc_context(meta_ac
);
3789 static int ocfs2_change_ctime(struct inode
*inode
,
3790 struct buffer_head
*di_bh
)
3794 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3796 handle
= ocfs2_start_trans(OCFS2_SB(inode
->i_sb
),
3797 OCFS2_INODE_UPDATE_CREDITS
);
3798 if (IS_ERR(handle
)) {
3799 ret
= PTR_ERR(handle
);
3804 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
3805 OCFS2_JOURNAL_ACCESS_WRITE
);
3811 inode
->i_ctime
= CURRENT_TIME
;
3812 di
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
3813 di
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
3815 ocfs2_journal_dirty(handle
, di_bh
);
3818 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
3823 static int ocfs2_attach_refcount_tree(struct inode
*inode
,
3824 struct buffer_head
*di_bh
)
3826 int ret
, data_changed
= 0;
3827 struct buffer_head
*ref_root_bh
= NULL
;
3828 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
3829 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
3830 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3831 struct ocfs2_refcount_tree
*ref_tree
;
3832 unsigned int ext_flags
;
3834 u32 cpos
, num_clusters
, clusters
, p_cluster
;
3835 struct ocfs2_cached_dealloc_ctxt dealloc
;
3836 struct ocfs2_extent_tree di_et
;
3838 ocfs2_init_dealloc_ctxt(&dealloc
);
3840 if (!(oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
)) {
3841 ret
= ocfs2_create_refcount_tree(inode
, di_bh
);
3848 BUG_ON(!di
->i_refcount_loc
);
3849 ret
= ocfs2_lock_refcount_tree(osb
,
3850 le64_to_cpu(di
->i_refcount_loc
), 1,
3851 &ref_tree
, &ref_root_bh
);
3857 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
3860 ocfs2_init_dinode_extent_tree(&di_et
, INODE_CACHE(inode
), di_bh
);
3862 size
= i_size_read(inode
);
3863 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, size
);
3866 while (cpos
< clusters
) {
3867 ret
= ocfs2_get_clusters(inode
, cpos
, &p_cluster
,
3868 &num_clusters
, &ext_flags
);
3870 if (p_cluster
&& !(ext_flags
& OCFS2_EXT_REFCOUNTED
)) {
3871 ret
= ocfs2_add_refcount_flag(inode
, &di_et
,
3874 p_cluster
, num_clusters
,
3883 cpos
+= num_clusters
;
3887 if (oi
->ip_dyn_features
& OCFS2_HAS_XATTR_FL
) {
3888 ret
= ocfs2_xattr_attach_refcount_tree(inode
, di_bh
,
3899 ret
= ocfs2_change_ctime(inode
, di_bh
);
3905 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
3906 brelse(ref_root_bh
);
3908 if (!ret
&& ocfs2_dealloc_has_cluster(&dealloc
)) {
3909 ocfs2_schedule_truncate_log_flush(osb
, 1);
3910 ocfs2_run_deallocs(osb
, &dealloc
);
3914 * Empty the extent map so that we may get the right extent
3915 * record from the disk.
3917 ocfs2_extent_map_trunc(inode
, 0);
3922 static int ocfs2_add_refcounted_extent(struct inode
*inode
,
3923 struct ocfs2_extent_tree
*et
,
3924 struct ocfs2_caching_info
*ref_ci
,
3925 struct buffer_head
*ref_root_bh
,
3926 u32 cpos
, u32 p_cluster
, u32 num_clusters
,
3927 unsigned int ext_flags
,
3928 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
3933 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
3934 struct ocfs2_alloc_context
*meta_ac
= NULL
;
3936 ret
= ocfs2_lock_refcount_allocators(inode
->i_sb
,
3937 p_cluster
, num_clusters
,
3939 ref_root_bh
, &meta_ac
,
3946 handle
= ocfs2_start_trans(osb
, credits
);
3947 if (IS_ERR(handle
)) {
3948 ret
= PTR_ERR(handle
);
3953 ret
= ocfs2_insert_extent(handle
, et
, cpos
,
3954 cpu_to_le64(ocfs2_clusters_to_blocks(inode
->i_sb
,
3956 num_clusters
, ext_flags
, meta_ac
);
3962 ret
= ocfs2_increase_refcount(handle
, ref_ci
, ref_root_bh
,
3963 p_cluster
, num_clusters
,
3969 ocfs2_commit_trans(osb
, handle
);
3972 ocfs2_free_alloc_context(meta_ac
);
3976 static int ocfs2_duplicate_inline_data(struct inode
*s_inode
,
3977 struct buffer_head
*s_bh
,
3978 struct inode
*t_inode
,
3979 struct buffer_head
*t_bh
)
3983 struct ocfs2_super
*osb
= OCFS2_SB(s_inode
->i_sb
);
3984 struct ocfs2_dinode
*s_di
= (struct ocfs2_dinode
*)s_bh
->b_data
;
3985 struct ocfs2_dinode
*t_di
= (struct ocfs2_dinode
*)t_bh
->b_data
;
3987 BUG_ON(!(OCFS2_I(s_inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
3989 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
3990 if (IS_ERR(handle
)) {
3991 ret
= PTR_ERR(handle
);
3996 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(t_inode
), t_bh
,
3997 OCFS2_JOURNAL_ACCESS_WRITE
);
4003 t_di
->id2
.i_data
.id_count
= s_di
->id2
.i_data
.id_count
;
4004 memcpy(t_di
->id2
.i_data
.id_data
, s_di
->id2
.i_data
.id_data
,
4005 le16_to_cpu(s_di
->id2
.i_data
.id_count
));
4006 spin_lock(&OCFS2_I(t_inode
)->ip_lock
);
4007 OCFS2_I(t_inode
)->ip_dyn_features
|= OCFS2_INLINE_DATA_FL
;
4008 t_di
->i_dyn_features
= cpu_to_le16(OCFS2_I(t_inode
)->ip_dyn_features
);
4009 spin_unlock(&OCFS2_I(t_inode
)->ip_lock
);
4011 ocfs2_journal_dirty(handle
, t_bh
);
4014 ocfs2_commit_trans(osb
, handle
);
4019 static int ocfs2_duplicate_extent_list(struct inode
*s_inode
,
4020 struct inode
*t_inode
,
4021 struct buffer_head
*t_bh
,
4022 struct ocfs2_caching_info
*ref_ci
,
4023 struct buffer_head
*ref_root_bh
,
4024 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
4027 u32 p_cluster
, num_clusters
, clusters
, cpos
;
4029 unsigned int ext_flags
;
4030 struct ocfs2_extent_tree et
;
4032 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(t_inode
), t_bh
);
4034 size
= i_size_read(s_inode
);
4035 clusters
= ocfs2_clusters_for_bytes(s_inode
->i_sb
, size
);
4038 while (cpos
< clusters
) {
4039 ret
= ocfs2_get_clusters(s_inode
, cpos
, &p_cluster
,
4040 &num_clusters
, &ext_flags
);
4043 ret
= ocfs2_add_refcounted_extent(t_inode
, &et
,
4044 ref_ci
, ref_root_bh
,
4055 cpos
+= num_clusters
;
4063 * change the new file's attributes to the src.
4065 * reflink creates a snapshot of a file, that means the attributes
4066 * must be identical except for three exceptions - nlink, ino, and ctime.
4068 static int ocfs2_complete_reflink(struct inode
*s_inode
,
4069 struct buffer_head
*s_bh
,
4070 struct inode
*t_inode
,
4071 struct buffer_head
*t_bh
,
4076 struct ocfs2_dinode
*s_di
= (struct ocfs2_dinode
*)s_bh
->b_data
;
4077 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)t_bh
->b_data
;
4078 loff_t size
= i_size_read(s_inode
);
4080 handle
= ocfs2_start_trans(OCFS2_SB(t_inode
->i_sb
),
4081 OCFS2_INODE_UPDATE_CREDITS
);
4082 if (IS_ERR(handle
)) {
4083 ret
= PTR_ERR(handle
);
4088 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(t_inode
), t_bh
,
4089 OCFS2_JOURNAL_ACCESS_WRITE
);
4095 spin_lock(&OCFS2_I(t_inode
)->ip_lock
);
4096 OCFS2_I(t_inode
)->ip_clusters
= OCFS2_I(s_inode
)->ip_clusters
;
4097 OCFS2_I(t_inode
)->ip_attr
= OCFS2_I(s_inode
)->ip_attr
;
4098 OCFS2_I(t_inode
)->ip_dyn_features
= OCFS2_I(s_inode
)->ip_dyn_features
;
4099 spin_unlock(&OCFS2_I(t_inode
)->ip_lock
);
4100 i_size_write(t_inode
, size
);
4102 di
->i_xattr_inline_size
= s_di
->i_xattr_inline_size
;
4103 di
->i_clusters
= s_di
->i_clusters
;
4104 di
->i_size
= s_di
->i_size
;
4105 di
->i_dyn_features
= s_di
->i_dyn_features
;
4106 di
->i_attr
= s_di
->i_attr
;
4109 t_inode
->i_uid
= s_inode
->i_uid
;
4110 t_inode
->i_gid
= s_inode
->i_gid
;
4111 t_inode
->i_mode
= s_inode
->i_mode
;
4112 di
->i_uid
= s_di
->i_uid
;
4113 di
->i_gid
= s_di
->i_gid
;
4114 di
->i_mode
= s_di
->i_mode
;
4118 * we want mtime to appear identical to the source and
4121 t_inode
->i_ctime
= CURRENT_TIME
;
4123 di
->i_ctime
= cpu_to_le64(t_inode
->i_ctime
.tv_sec
);
4124 di
->i_ctime_nsec
= cpu_to_le32(t_inode
->i_ctime
.tv_nsec
);
4126 t_inode
->i_mtime
= s_inode
->i_mtime
;
4127 di
->i_mtime
= s_di
->i_mtime
;
4128 di
->i_mtime_nsec
= s_di
->i_mtime_nsec
;
4131 ocfs2_journal_dirty(handle
, t_bh
);
4134 ocfs2_commit_trans(OCFS2_SB(t_inode
->i_sb
), handle
);
4138 static int ocfs2_create_reflink_node(struct inode
*s_inode
,
4139 struct buffer_head
*s_bh
,
4140 struct inode
*t_inode
,
4141 struct buffer_head
*t_bh
,
4145 struct buffer_head
*ref_root_bh
= NULL
;
4146 struct ocfs2_cached_dealloc_ctxt dealloc
;
4147 struct ocfs2_super
*osb
= OCFS2_SB(s_inode
->i_sb
);
4148 struct ocfs2_refcount_block
*rb
;
4149 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)s_bh
->b_data
;
4150 struct ocfs2_refcount_tree
*ref_tree
;
4152 ocfs2_init_dealloc_ctxt(&dealloc
);
4154 ret
= ocfs2_set_refcount_tree(t_inode
, t_bh
,
4155 le64_to_cpu(di
->i_refcount_loc
));
4161 if (OCFS2_I(s_inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
4162 ret
= ocfs2_duplicate_inline_data(s_inode
, s_bh
,
4169 ret
= ocfs2_lock_refcount_tree(osb
, le64_to_cpu(di
->i_refcount_loc
),
4170 1, &ref_tree
, &ref_root_bh
);
4175 rb
= (struct ocfs2_refcount_block
*)ref_root_bh
->b_data
;
4177 ret
= ocfs2_duplicate_extent_list(s_inode
, t_inode
, t_bh
,
4178 &ref_tree
->rf_ci
, ref_root_bh
,
4182 goto out_unlock_refcount
;
4185 out_unlock_refcount
:
4186 ocfs2_unlock_refcount_tree(osb
, ref_tree
, 1);
4187 brelse(ref_root_bh
);
4189 if (ocfs2_dealloc_has_cluster(&dealloc
)) {
4190 ocfs2_schedule_truncate_log_flush(osb
, 1);
4191 ocfs2_run_deallocs(osb
, &dealloc
);
4197 static int __ocfs2_reflink(struct dentry
*old_dentry
,
4198 struct buffer_head
*old_bh
,
4199 struct inode
*new_inode
,
4203 struct inode
*inode
= old_dentry
->d_inode
;
4204 struct buffer_head
*new_bh
= NULL
;
4206 ret
= filemap_fdatawrite(inode
->i_mapping
);
4212 ret
= ocfs2_attach_refcount_tree(inode
, old_bh
);
4218 mutex_lock(&new_inode
->i_mutex
);
4219 ret
= ocfs2_inode_lock(new_inode
, &new_bh
, 1);
4225 ret
= ocfs2_create_reflink_node(inode
, old_bh
,
4226 new_inode
, new_bh
, preserve
);
4232 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_XATTR_FL
) {
4233 ret
= ocfs2_reflink_xattrs(inode
, old_bh
,
4242 ret
= ocfs2_complete_reflink(inode
, old_bh
,
4243 new_inode
, new_bh
, preserve
);
4248 ocfs2_inode_unlock(new_inode
, 1);
4251 mutex_unlock(&new_inode
->i_mutex
);
4254 ret
= filemap_fdatawait(inode
->i_mapping
);
4261 static int ocfs2_reflink(struct dentry
*old_dentry
, struct inode
*dir
,
4262 struct dentry
*new_dentry
, bool preserve
)
4265 struct inode
*inode
= old_dentry
->d_inode
;
4266 struct buffer_head
*old_bh
= NULL
;
4267 struct inode
*new_orphan_inode
= NULL
;
4269 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
)))
4272 error
= ocfs2_create_inode_in_orphan(dir
, inode
->i_mode
,
4279 error
= ocfs2_inode_lock(inode
, &old_bh
, 1);
4285 down_write(&OCFS2_I(inode
)->ip_xattr_sem
);
4286 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
4287 error
= __ocfs2_reflink(old_dentry
, old_bh
,
4288 new_orphan_inode
, preserve
);
4289 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
4290 up_write(&OCFS2_I(inode
)->ip_xattr_sem
);
4292 ocfs2_inode_unlock(inode
, 1);
4300 /* If the security isn't preserved, we need to re-initialize them. */
4302 error
= ocfs2_init_security_and_acl(dir
, new_orphan_inode
);
4308 error
= ocfs2_mv_orphaned_inode_to_new(dir
, new_orphan_inode
,
4314 if (new_orphan_inode
) {
4316 * We need to open_unlock the inode no matter whether we
4317 * succeed or not, so that other nodes can delete it later.
4319 ocfs2_open_unlock(new_orphan_inode
);
4321 iput(new_orphan_inode
);
4328 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4329 * sys_reflink(). This will go away when vfs_reflink() exists in
4333 /* copied from may_create in VFS. */
4334 static inline int ocfs2_may_create(struct inode
*dir
, struct dentry
*child
)
4338 if (IS_DEADDIR(dir
))
4340 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
4343 /* copied from user_path_parent. */
4344 static int ocfs2_user_path_parent(const char __user
*path
,
4345 struct nameidata
*nd
, char **name
)
4347 char *s
= getname(path
);
4353 error
= path_lookup(s
, LOOKUP_PARENT
, nd
);
4363 * ocfs2_vfs_reflink - Create a reference-counted link
4365 * @old_dentry: source dentry + inode
4366 * @dir: directory to create the target
4367 * @new_dentry: target dentry
4368 * @preserve: if true, preserve all file attributes
4370 int ocfs2_vfs_reflink(struct dentry
*old_dentry
, struct inode
*dir
,
4371 struct dentry
*new_dentry
, bool preserve
)
4373 struct inode
*inode
= old_dentry
->d_inode
;
4379 error
= ocfs2_may_create(dir
, new_dentry
);
4383 if (dir
->i_sb
!= inode
->i_sb
)
4387 * A reflink to an append-only or immutable file cannot be created.
4389 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4392 /* Only regular files can be reflinked. */
4393 if (!S_ISREG(inode
->i_mode
))
4397 * If the caller wants to preserve ownership, they require the
4401 if ((current_fsuid() != inode
->i_uid
) && !capable(CAP_CHOWN
))
4403 if (!in_group_p(inode
->i_gid
) && !capable(CAP_CHOWN
))
4408 * If the caller is modifying any aspect of the attributes, they
4409 * are not creating a snapshot. They need read permission on the
4413 error
= inode_permission(inode
, MAY_READ
);
4418 mutex_lock(&inode
->i_mutex
);
4420 error
= ocfs2_reflink(old_dentry
, dir
, new_dentry
, preserve
);
4421 mutex_unlock(&inode
->i_mutex
);
4423 fsnotify_create(dir
, new_dentry
);
4427 * Most codes are copied from sys_linkat.
4429 int ocfs2_reflink_ioctl(struct inode
*inode
,
4430 const char __user
*oldname
,
4431 const char __user
*newname
,
4434 struct dentry
*new_dentry
;
4435 struct nameidata nd
;
4436 struct path old_path
;
4440 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
)))
4443 error
= user_path_at(AT_FDCWD
, oldname
, 0, &old_path
);
4449 error
= ocfs2_user_path_parent(newname
, &nd
, &to
);
4456 if (old_path
.mnt
!= nd
.path
.mnt
)
4458 new_dentry
= lookup_create(&nd
, 0);
4459 error
= PTR_ERR(new_dentry
);
4460 if (IS_ERR(new_dentry
)) {
4465 error
= mnt_want_write(nd
.path
.mnt
);
4471 error
= ocfs2_vfs_reflink(old_path
.dentry
,
4472 nd
.path
.dentry
->d_inode
,
4473 new_dentry
, preserve
);
4474 mnt_drop_write(nd
.path
.mnt
);
4478 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
4483 path_put(&old_path
);