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 #include <cluster/masklog.h>
27 #include "buffer_head_io.h"
28 #include "blockcheck.h"
29 #include "refcounttree.h"
32 #include "extent_map.h"
36 #include "ocfs2_trace.h"
38 #include <linux/bio.h>
39 #include <linux/blkdev.h>
40 #include <linux/slab.h>
41 #include <linux/writeback.h>
42 #include <linux/pagevec.h>
43 #include <linux/swap.h>
44 #include <linux/security.h>
45 #include <linux/fsnotify.h>
46 #include <linux/quotaops.h>
47 #include <linux/namei.h>
48 #include <linux/mount.h>
49 #include <linux/posix_acl.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,
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 trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
89 BUG_ON(!buffer_uptodate(bh));
92 * If the ecc fails, we return the error but otherwise
93 * leave the filesystem running. We know any error is
94 * local to this block.
96 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
98 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
99 (unsigned long long)bh->b_blocknr);
104 if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
106 "Refcount block #%llu has bad signature %.*s",
107 (unsigned long long)bh->b_blocknr, 7,
112 if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
114 "Refcount block #%llu has an invalid rf_blkno "
116 (unsigned long long)bh->b_blocknr,
117 (unsigned long long)le64_to_cpu(rb->rf_blkno));
121 if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
123 "Refcount block #%llu has an invalid "
124 "rf_fs_generation of #%u",
125 (unsigned long long)bh->b_blocknr,
126 le32_to_cpu(rb->rf_fs_generation));
133 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
135 struct buffer_head **bh)
138 struct buffer_head *tmp = *bh;
140 rc = ocfs2_read_block(ci, rb_blkno, &tmp,
141 ocfs2_validate_refcount_block);
143 /* If ocfs2_read_block() got us a new bh, pass it up. */
150 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
152 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
157 static struct super_block *
158 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
160 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
165 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
167 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
169 spin_lock(&rf->rf_lock);
172 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
174 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
176 spin_unlock(&rf->rf_lock);
179 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
181 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
183 mutex_lock(&rf->rf_io_mutex);
186 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
188 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
190 mutex_unlock(&rf->rf_io_mutex);
193 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
194 .co_owner = ocfs2_refcount_cache_owner,
195 .co_get_super = ocfs2_refcount_cache_get_super,
196 .co_cache_lock = ocfs2_refcount_cache_lock,
197 .co_cache_unlock = ocfs2_refcount_cache_unlock,
198 .co_io_lock = ocfs2_refcount_cache_io_lock,
199 .co_io_unlock = ocfs2_refcount_cache_io_unlock,
202 static struct ocfs2_refcount_tree *
203 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
205 struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
206 struct ocfs2_refcount_tree *tree = NULL;
209 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
211 if (blkno < tree->rf_blkno)
213 else if (blkno > tree->rf_blkno)
222 /* osb_lock is already locked. */
223 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
224 struct ocfs2_refcount_tree *new)
226 u64 rf_blkno = new->rf_blkno;
227 struct rb_node *parent = NULL;
228 struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
229 struct ocfs2_refcount_tree *tmp;
234 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
237 if (rf_blkno < tmp->rf_blkno)
239 else if (rf_blkno > tmp->rf_blkno)
242 /* This should never happen! */
243 mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
244 (unsigned long long)rf_blkno);
249 rb_link_node(&new->rf_node, parent, p);
250 rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
253 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
255 ocfs2_metadata_cache_exit(&tree->rf_ci);
256 ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
257 ocfs2_lock_res_free(&tree->rf_lockres);
262 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
263 struct ocfs2_refcount_tree *tree)
265 rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
266 if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
267 osb->osb_ref_tree_lru = NULL;
270 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
271 struct ocfs2_refcount_tree *tree)
273 spin_lock(&osb->osb_lock);
274 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
275 spin_unlock(&osb->osb_lock);
278 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
280 struct ocfs2_refcount_tree *tree =
281 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
283 ocfs2_free_refcount_tree(tree);
287 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
289 kref_get(&tree->rf_getcnt);
293 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
295 kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
298 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
299 struct super_block *sb)
301 ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
302 mutex_init(&new->rf_io_mutex);
304 spin_lock_init(&new->rf_lock);
307 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
308 struct ocfs2_refcount_tree *new,
309 u64 rf_blkno, u32 generation)
311 init_rwsem(&new->rf_sem);
312 ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
313 rf_blkno, generation);
316 static struct ocfs2_refcount_tree*
317 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
319 struct ocfs2_refcount_tree *new;
321 new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
325 new->rf_blkno = rf_blkno;
326 kref_init(&new->rf_getcnt);
327 ocfs2_init_refcount_tree_ci(new, osb->sb);
332 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
333 struct ocfs2_refcount_tree **ret_tree)
336 struct ocfs2_refcount_tree *tree, *new = NULL;
337 struct buffer_head *ref_root_bh = NULL;
338 struct ocfs2_refcount_block *ref_rb;
340 spin_lock(&osb->osb_lock);
341 if (osb->osb_ref_tree_lru &&
342 osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
343 tree = osb->osb_ref_tree_lru;
345 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
349 spin_unlock(&osb->osb_lock);
351 new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
358 * We need the generation to create the refcount tree lock and since
359 * it isn't changed during the tree modification, we are safe here to
360 * read without protection.
361 * We also have to purge the cache after we create the lock since the
362 * refcount block may have the stale data. It can only be trusted when
363 * we hold the refcount lock.
365 ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
368 ocfs2_metadata_cache_exit(&new->rf_ci);
373 ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
374 new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
375 ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
377 ocfs2_metadata_cache_purge(&new->rf_ci);
379 spin_lock(&osb->osb_lock);
380 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
384 ocfs2_insert_refcount_tree(osb, new);
392 osb->osb_ref_tree_lru = tree;
394 spin_unlock(&osb->osb_lock);
397 ocfs2_free_refcount_tree(new);
403 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
406 struct buffer_head *di_bh = NULL;
407 struct ocfs2_dinode *di;
409 ret = ocfs2_read_inode_block(inode, &di_bh);
415 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
417 di = (struct ocfs2_dinode *)di_bh->b_data;
418 *ref_blkno = le64_to_cpu(di->i_refcount_loc);
424 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
425 struct ocfs2_refcount_tree *tree, int rw)
429 ret = ocfs2_refcount_lock(tree, rw);
436 down_write(&tree->rf_sem);
438 down_read(&tree->rf_sem);
445 * Lock the refcount tree pointed by ref_blkno and return the tree.
446 * In most case, we lock the tree and read the refcount block.
447 * So read it here if the caller really needs it.
449 * If the tree has been re-created by other node, it will free the
450 * old one and re-create it.
452 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
453 u64 ref_blkno, int rw,
454 struct ocfs2_refcount_tree **ret_tree,
455 struct buffer_head **ref_bh)
457 int ret, delete_tree = 0;
458 struct ocfs2_refcount_tree *tree = NULL;
459 struct buffer_head *ref_root_bh = NULL;
460 struct ocfs2_refcount_block *rb;
463 ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
469 ocfs2_refcount_tree_get(tree);
471 ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
474 ocfs2_refcount_tree_put(tree);
478 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
482 ocfs2_unlock_refcount_tree(osb, tree, rw);
483 ocfs2_refcount_tree_put(tree);
487 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
489 * If the refcount block has been freed and re-created, we may need
490 * to recreate the refcount tree also.
492 * Here we just remove the tree from the rb-tree, and the last
493 * kref holder will unlock and delete this refcount_tree.
494 * Then we goto "again" and ocfs2_get_refcount_tree will create
495 * the new refcount tree for us.
497 if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
498 if (!tree->rf_removed) {
499 ocfs2_erase_refcount_tree_from_list(osb, tree);
500 tree->rf_removed = 1;
504 ocfs2_unlock_refcount_tree(osb, tree, rw);
506 * We get an extra reference when we create the refcount
507 * tree, so another put will destroy it.
510 ocfs2_refcount_tree_put(tree);
518 *ref_bh = ref_root_bh;
526 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
527 struct ocfs2_refcount_tree *tree, int rw)
530 up_write(&tree->rf_sem);
532 up_read(&tree->rf_sem);
534 ocfs2_refcount_unlock(tree, rw);
535 ocfs2_refcount_tree_put(tree);
538 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
540 struct rb_node *node;
541 struct ocfs2_refcount_tree *tree;
542 struct rb_root *root = &osb->osb_rf_lock_tree;
544 while ((node = rb_last(root)) != NULL) {
545 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
547 trace_ocfs2_purge_refcount_trees(
548 (unsigned long long) tree->rf_blkno);
550 rb_erase(&tree->rf_node, root);
551 ocfs2_free_refcount_tree(tree);
556 * Create a refcount tree for an inode.
557 * We take for granted that the inode is already locked.
559 static int ocfs2_create_refcount_tree(struct inode *inode,
560 struct buffer_head *di_bh)
563 handle_t *handle = NULL;
564 struct ocfs2_alloc_context *meta_ac = NULL;
565 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
566 struct ocfs2_inode_info *oi = OCFS2_I(inode);
567 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
568 struct buffer_head *new_bh = NULL;
569 struct ocfs2_refcount_block *rb;
570 struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
571 u16 suballoc_bit_start;
573 u64 suballoc_loc, first_blkno;
575 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
577 trace_ocfs2_create_refcount_tree(
578 (unsigned long long)OCFS2_I(inode)->ip_blkno);
580 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
586 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
587 if (IS_ERR(handle)) {
588 ret = PTR_ERR(handle);
593 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
594 OCFS2_JOURNAL_ACCESS_WRITE);
600 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
601 &suballoc_bit_start, &num_got,
608 new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
615 new_bh = sb_getblk(inode->i_sb, first_blkno);
621 ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
623 ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
624 OCFS2_JOURNAL_ACCESS_CREATE);
630 /* Initialize ocfs2_refcount_block. */
631 rb = (struct ocfs2_refcount_block *)new_bh->b_data;
632 memset(rb, 0, inode->i_sb->s_blocksize);
633 strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
634 rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
635 rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
636 rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
637 rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
638 rb->rf_blkno = cpu_to_le64(first_blkno);
639 rb->rf_count = cpu_to_le32(1);
640 rb->rf_records.rl_count =
641 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
642 spin_lock(&osb->osb_lock);
643 rb->rf_generation = osb->s_next_generation++;
644 spin_unlock(&osb->osb_lock);
646 ocfs2_journal_dirty(handle, new_bh);
648 spin_lock(&oi->ip_lock);
649 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
650 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
651 di->i_refcount_loc = cpu_to_le64(first_blkno);
652 spin_unlock(&oi->ip_lock);
654 trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
656 ocfs2_journal_dirty(handle, di_bh);
659 * We have to init the tree lock here since it will use
660 * the generation number to create it.
662 new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
663 ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
664 new_tree->rf_generation);
666 spin_lock(&osb->osb_lock);
667 tree = ocfs2_find_refcount_tree(osb, first_blkno);
670 * We've just created a new refcount tree in this block. If
671 * we found a refcount tree on the ocfs2_super, it must be
672 * one we just deleted. We free the old tree before
673 * inserting the new tree.
675 BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
677 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
678 ocfs2_insert_refcount_tree(osb, new_tree);
679 spin_unlock(&osb->osb_lock);
682 ocfs2_refcount_tree_put(tree);
685 ocfs2_commit_trans(osb, handle);
689 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
695 ocfs2_free_alloc_context(meta_ac);
700 static int ocfs2_set_refcount_tree(struct inode *inode,
701 struct buffer_head *di_bh,
705 handle_t *handle = NULL;
706 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
707 struct ocfs2_inode_info *oi = OCFS2_I(inode);
708 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
709 struct buffer_head *ref_root_bh = NULL;
710 struct ocfs2_refcount_block *rb;
711 struct ocfs2_refcount_tree *ref_tree;
713 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
715 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
716 &ref_tree, &ref_root_bh);
722 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
723 if (IS_ERR(handle)) {
724 ret = PTR_ERR(handle);
729 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
730 OCFS2_JOURNAL_ACCESS_WRITE);
736 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
737 OCFS2_JOURNAL_ACCESS_WRITE);
743 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
744 le32_add_cpu(&rb->rf_count, 1);
746 ocfs2_journal_dirty(handle, ref_root_bh);
748 spin_lock(&oi->ip_lock);
749 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
750 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
751 di->i_refcount_loc = cpu_to_le64(refcount_loc);
752 spin_unlock(&oi->ip_lock);
753 ocfs2_journal_dirty(handle, di_bh);
756 ocfs2_commit_trans(osb, handle);
758 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
764 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
766 int ret, delete_tree = 0;
767 handle_t *handle = NULL;
768 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
769 struct ocfs2_inode_info *oi = OCFS2_I(inode);
770 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
771 struct ocfs2_refcount_block *rb;
772 struct inode *alloc_inode = NULL;
773 struct buffer_head *alloc_bh = NULL;
774 struct buffer_head *blk_bh = NULL;
775 struct ocfs2_refcount_tree *ref_tree;
776 int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
777 u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
780 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
784 ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
790 rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
793 * If we are the last user, we need to free the block.
794 * So lock the allocator ahead.
796 if (le32_to_cpu(rb->rf_count) == 1) {
797 blk = le64_to_cpu(rb->rf_blkno);
798 bit = le16_to_cpu(rb->rf_suballoc_bit);
799 if (rb->rf_suballoc_loc)
800 bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
802 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
804 alloc_inode = ocfs2_get_system_file_inode(osb,
805 EXTENT_ALLOC_SYSTEM_INODE,
806 le16_to_cpu(rb->rf_suballoc_slot));
812 mutex_lock(&alloc_inode->i_mutex);
814 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
820 credits += OCFS2_SUBALLOC_FREE;
823 handle = ocfs2_start_trans(osb, credits);
824 if (IS_ERR(handle)) {
825 ret = PTR_ERR(handle);
830 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
831 OCFS2_JOURNAL_ACCESS_WRITE);
837 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
838 OCFS2_JOURNAL_ACCESS_WRITE);
844 spin_lock(&oi->ip_lock);
845 oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
846 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
847 di->i_refcount_loc = 0;
848 spin_unlock(&oi->ip_lock);
849 ocfs2_journal_dirty(handle, di_bh);
851 le32_add_cpu(&rb->rf_count , -1);
852 ocfs2_journal_dirty(handle, blk_bh);
856 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
857 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
858 alloc_bh, bit, bg_blkno, 1);
864 ocfs2_commit_trans(osb, handle);
867 ocfs2_inode_unlock(alloc_inode, 1);
872 mutex_unlock(&alloc_inode->i_mutex);
876 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
878 ocfs2_refcount_tree_put(ref_tree);
884 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
885 struct buffer_head *ref_leaf_bh,
886 u64 cpos, unsigned int len,
887 struct ocfs2_refcount_rec *ret_rec,
891 struct ocfs2_refcount_block *rb =
892 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
893 struct ocfs2_refcount_rec *rec = NULL;
895 for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
896 rec = &rb->rf_records.rl_recs[i];
898 if (le64_to_cpu(rec->r_cpos) +
899 le32_to_cpu(rec->r_clusters) <= cpos)
901 else if (le64_to_cpu(rec->r_cpos) > cpos)
904 /* ok, cpos fail in this rec. Just return. */
911 /* We meet with a hole here, so fake the rec. */
912 ret_rec->r_cpos = cpu_to_le64(cpos);
913 ret_rec->r_refcount = 0;
914 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
915 le64_to_cpu(rec->r_cpos) < cpos + len)
916 ret_rec->r_clusters =
917 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
919 ret_rec->r_clusters = cpu_to_le32(len);
927 * Try to remove refcount tree. The mechanism is:
928 * 1) Check whether i_clusters == 0, if no, exit.
929 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
930 * 3) Check whether we have inline xattr stored outside, if yes, exit.
931 * 4) Remove the tree.
933 int ocfs2_try_remove_refcount_tree(struct inode *inode,
934 struct buffer_head *di_bh)
937 struct ocfs2_inode_info *oi = OCFS2_I(inode);
938 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
940 down_write(&oi->ip_xattr_sem);
941 down_write(&oi->ip_alloc_sem);
946 if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
949 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
950 ocfs2_has_inline_xattr_value_outside(inode, di))
953 ret = ocfs2_remove_refcount_tree(inode, di_bh);
957 up_write(&oi->ip_alloc_sem);
958 up_write(&oi->ip_xattr_sem);
963 * Find the end range for a leaf refcount block indicated by
964 * el->l_recs[index].e_blkno.
966 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
967 struct buffer_head *ref_root_bh,
968 struct ocfs2_extent_block *eb,
969 struct ocfs2_extent_list *el,
970 int index, u32 *cpos_end)
972 int ret, i, subtree_root;
975 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
976 struct ocfs2_path *left_path = NULL, *right_path = NULL;
977 struct ocfs2_extent_tree et;
978 struct ocfs2_extent_list *tmp_el;
980 if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
982 * We have a extent rec after index, so just use the e_cpos
983 * of the next extent rec.
985 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
989 if (!eb || (eb && !eb->h_next_leaf_blk)) {
991 * We are the last extent rec, so any high cpos should
992 * be stored in this leaf refcount block.
994 *cpos_end = UINT_MAX;
999 * If the extent block isn't the last one, we have to find
1000 * the subtree root between this extent block and the next
1001 * leaf extent block and get the corresponding e_cpos from
1002 * the subroot. Otherwise we may corrupt the b-tree.
1004 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1006 left_path = ocfs2_new_path_from_et(&et);
1013 cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1014 ret = ocfs2_find_path(ci, left_path, cpos);
1020 right_path = ocfs2_new_path_from_path(left_path);
1027 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1033 ret = ocfs2_find_path(ci, right_path, cpos);
1039 subtree_root = ocfs2_find_subtree_root(&et, left_path,
1042 tmp_el = left_path->p_node[subtree_root].el;
1043 blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1044 for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1045 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1046 *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1051 BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1054 ocfs2_free_path(left_path);
1055 ocfs2_free_path(right_path);
1060 * Given a cpos and len, try to find the refcount record which contains cpos.
1061 * 1. If cpos can be found in one refcount record, return the record.
1062 * 2. If cpos can't be found, return a fake record which start from cpos
1063 * and end at a small value between cpos+len and start of the next record.
1064 * This fake record has r_refcount = 0.
1066 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1067 struct buffer_head *ref_root_bh,
1068 u64 cpos, unsigned int len,
1069 struct ocfs2_refcount_rec *ret_rec,
1071 struct buffer_head **ret_bh)
1073 int ret = 0, i, found;
1074 u32 low_cpos, uninitialized_var(cpos_end);
1075 struct ocfs2_extent_list *el;
1076 struct ocfs2_extent_rec *rec = NULL;
1077 struct ocfs2_extent_block *eb = NULL;
1078 struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1079 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1080 struct ocfs2_refcount_block *rb =
1081 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1083 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1084 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1086 *ret_bh = ref_root_bh;
1087 get_bh(ref_root_bh);
1092 low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1094 if (el->l_tree_depth) {
1095 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1101 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1104 if (el->l_tree_depth) {
1106 "refcount tree %llu has non zero tree "
1107 "depth in leaf btree tree block %llu\n",
1108 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1109 (unsigned long long)eb_bh->b_blocknr);
1116 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1117 rec = &el->l_recs[i];
1119 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1126 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1127 eb, el, i, &cpos_end);
1133 if (cpos_end < low_cpos + len)
1134 len = cpos_end - low_cpos;
1137 ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1144 ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1146 *ret_bh = ref_leaf_bh;
1152 enum ocfs2_ref_rec_contig {
1153 REF_CONTIG_NONE = 0,
1156 REF_CONTIG_LEFTRIGHT,
1159 static enum ocfs2_ref_rec_contig
1160 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1163 if ((rb->rf_records.rl_recs[index].r_refcount ==
1164 rb->rf_records.rl_recs[index + 1].r_refcount) &&
1165 (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1166 le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1167 le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1168 return REF_CONTIG_RIGHT;
1170 return REF_CONTIG_NONE;
1173 static enum ocfs2_ref_rec_contig
1174 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1177 enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1179 if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1180 ret = ocfs2_refcount_rec_adjacent(rb, index);
1183 enum ocfs2_ref_rec_contig tmp;
1185 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1187 if (tmp == REF_CONTIG_RIGHT) {
1188 if (ret == REF_CONTIG_RIGHT)
1189 ret = REF_CONTIG_LEFTRIGHT;
1191 ret = REF_CONTIG_LEFT;
1198 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1201 BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1202 rb->rf_records.rl_recs[index+1].r_refcount);
1204 le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1205 le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1207 if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1208 memmove(&rb->rf_records.rl_recs[index + 1],
1209 &rb->rf_records.rl_recs[index + 2],
1210 sizeof(struct ocfs2_refcount_rec) *
1211 (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1213 memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1214 0, sizeof(struct ocfs2_refcount_rec));
1215 le16_add_cpu(&rb->rf_records.rl_used, -1);
1219 * Merge the refcount rec if we are contiguous with the adjacent recs.
1221 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1224 enum ocfs2_ref_rec_contig contig =
1225 ocfs2_refcount_rec_contig(rb, index);
1227 if (contig == REF_CONTIG_NONE)
1230 if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1235 ocfs2_rotate_refcount_rec_left(rb, index);
1237 if (contig == REF_CONTIG_LEFTRIGHT)
1238 ocfs2_rotate_refcount_rec_left(rb, index);
1242 * Change the refcount indexed by "index" in ref_bh.
1243 * If refcount reaches 0, remove it.
1245 static int ocfs2_change_refcount_rec(handle_t *handle,
1246 struct ocfs2_caching_info *ci,
1247 struct buffer_head *ref_leaf_bh,
1248 int index, int merge, int change)
1251 struct ocfs2_refcount_block *rb =
1252 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1253 struct ocfs2_refcount_list *rl = &rb->rf_records;
1254 struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1256 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1257 OCFS2_JOURNAL_ACCESS_WRITE);
1263 trace_ocfs2_change_refcount_rec(
1264 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1265 index, le32_to_cpu(rec->r_refcount), change);
1266 le32_add_cpu(&rec->r_refcount, change);
1268 if (!rec->r_refcount) {
1269 if (index != le16_to_cpu(rl->rl_used) - 1) {
1270 memmove(rec, rec + 1,
1271 (le16_to_cpu(rl->rl_used) - index - 1) *
1272 sizeof(struct ocfs2_refcount_rec));
1273 memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1274 0, sizeof(struct ocfs2_refcount_rec));
1277 le16_add_cpu(&rl->rl_used, -1);
1279 ocfs2_refcount_rec_merge(rb, index);
1281 ocfs2_journal_dirty(handle, ref_leaf_bh);
1286 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1287 struct ocfs2_caching_info *ci,
1288 struct buffer_head *ref_root_bh,
1289 struct buffer_head **ref_leaf_bh,
1290 struct ocfs2_alloc_context *meta_ac)
1293 u16 suballoc_bit_start;
1295 u64 suballoc_loc, blkno;
1296 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1297 struct buffer_head *new_bh = NULL;
1298 struct ocfs2_refcount_block *new_rb;
1299 struct ocfs2_refcount_block *root_rb =
1300 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1302 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1303 OCFS2_JOURNAL_ACCESS_WRITE);
1309 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1310 &suballoc_bit_start, &num_got,
1317 new_bh = sb_getblk(sb, blkno);
1318 if (new_bh == NULL) {
1323 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1325 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1326 OCFS2_JOURNAL_ACCESS_CREATE);
1333 * Initialize ocfs2_refcount_block.
1334 * It should contain the same information as the old root.
1335 * so just memcpy it and change the corresponding field.
1337 memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1339 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1340 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1341 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1342 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1343 new_rb->rf_blkno = cpu_to_le64(blkno);
1344 new_rb->rf_cpos = cpu_to_le32(0);
1345 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1346 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1347 ocfs2_journal_dirty(handle, new_bh);
1349 /* Now change the root. */
1350 memset(&root_rb->rf_list, 0, sb->s_blocksize -
1351 offsetof(struct ocfs2_refcount_block, rf_list));
1352 root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1353 root_rb->rf_clusters = cpu_to_le32(1);
1354 root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1355 root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1356 root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1357 root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1359 ocfs2_journal_dirty(handle, ref_root_bh);
1361 trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1362 le16_to_cpu(new_rb->rf_records.rl_used));
1364 *ref_leaf_bh = new_bh;
1371 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1372 struct ocfs2_refcount_rec *next)
1374 if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1375 ocfs2_get_ref_rec_low_cpos(next))
1381 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1383 const struct ocfs2_refcount_rec *l = a, *r = b;
1384 u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1385 u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1387 if (l_cpos > r_cpos)
1389 if (l_cpos < r_cpos)
1394 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1396 const struct ocfs2_refcount_rec *l = a, *r = b;
1397 u64 l_cpos = le64_to_cpu(l->r_cpos);
1398 u64 r_cpos = le64_to_cpu(r->r_cpos);
1400 if (l_cpos > r_cpos)
1402 if (l_cpos < r_cpos)
1407 static void swap_refcount_rec(void *a, void *b, int size)
1409 struct ocfs2_refcount_rec *l = a, *r = b;
1415 * The refcount cpos are ordered by their 64bit cpos,
1416 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1417 * So we need to make sure that this pos isn't intersected with others.
1419 * Note: The refcount block is already sorted by their low 32 bit cpos,
1420 * So just try the middle pos first, and we will exit when we find
1421 * the good position.
1423 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1424 u32 *split_pos, int *split_index)
1426 int num_used = le16_to_cpu(rl->rl_used);
1427 int delta, middle = num_used / 2;
1429 for (delta = 0; delta < middle; delta++) {
1430 /* Let's check delta earlier than middle */
1431 if (ocfs2_refcount_rec_no_intersect(
1432 &rl->rl_recs[middle - delta - 1],
1433 &rl->rl_recs[middle - delta])) {
1434 *split_index = middle - delta;
1438 /* For even counts, don't walk off the end */
1439 if ((middle + delta + 1) == num_used)
1442 /* Now try delta past middle */
1443 if (ocfs2_refcount_rec_no_intersect(
1444 &rl->rl_recs[middle + delta],
1445 &rl->rl_recs[middle + delta + 1])) {
1446 *split_index = middle + delta + 1;
1451 if (delta >= middle)
1454 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1458 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1459 struct buffer_head *new_bh,
1462 int split_index = 0, num_moved, ret;
1464 struct ocfs2_refcount_block *rb =
1465 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1466 struct ocfs2_refcount_list *rl = &rb->rf_records;
1467 struct ocfs2_refcount_block *new_rb =
1468 (struct ocfs2_refcount_block *)new_bh->b_data;
1469 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1471 trace_ocfs2_divide_leaf_refcount_block(
1472 (unsigned long long)ref_leaf_bh->b_blocknr,
1473 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1476 * XXX: Improvement later.
1477 * If we know all the high 32 bit cpos is the same, no need to sort.
1479 * In order to make the whole process safe, we do:
1480 * 1. sort the entries by their low 32 bit cpos first so that we can
1481 * find the split cpos easily.
1482 * 2. call ocfs2_insert_extent to insert the new refcount block.
1483 * 3. move the refcount rec to the new block.
1484 * 4. sort the entries by their 64 bit cpos.
1485 * 5. dirty the new_rb and rb.
1487 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1488 sizeof(struct ocfs2_refcount_rec),
1489 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1491 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1497 new_rb->rf_cpos = cpu_to_le32(cpos);
1499 /* move refcount records starting from split_index to the new block. */
1500 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1501 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1502 num_moved * sizeof(struct ocfs2_refcount_rec));
1504 /*ok, remove the entries we just moved over to the other block. */
1505 memset(&rl->rl_recs[split_index], 0,
1506 num_moved * sizeof(struct ocfs2_refcount_rec));
1508 /* change old and new rl_used accordingly. */
1509 le16_add_cpu(&rl->rl_used, -num_moved);
1510 new_rl->rl_used = cpu_to_le16(num_moved);
1512 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1513 sizeof(struct ocfs2_refcount_rec),
1514 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1516 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1517 sizeof(struct ocfs2_refcount_rec),
1518 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1524 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1525 struct ocfs2_caching_info *ci,
1526 struct buffer_head *ref_root_bh,
1527 struct buffer_head *ref_leaf_bh,
1528 struct ocfs2_alloc_context *meta_ac)
1531 u16 suballoc_bit_start;
1532 u32 num_got, new_cpos;
1533 u64 suballoc_loc, blkno;
1534 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1535 struct ocfs2_refcount_block *root_rb =
1536 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1537 struct buffer_head *new_bh = NULL;
1538 struct ocfs2_refcount_block *new_rb;
1539 struct ocfs2_extent_tree ref_et;
1541 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1543 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1544 OCFS2_JOURNAL_ACCESS_WRITE);
1550 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1551 OCFS2_JOURNAL_ACCESS_WRITE);
1557 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1558 &suballoc_bit_start, &num_got,
1565 new_bh = sb_getblk(sb, blkno);
1566 if (new_bh == NULL) {
1571 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1573 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1574 OCFS2_JOURNAL_ACCESS_CREATE);
1580 /* Initialize ocfs2_refcount_block. */
1581 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1582 memset(new_rb, 0, sb->s_blocksize);
1583 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1584 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1585 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1586 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1587 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1588 new_rb->rf_blkno = cpu_to_le64(blkno);
1589 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1590 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1591 new_rb->rf_records.rl_count =
1592 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1593 new_rb->rf_generation = root_rb->rf_generation;
1595 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1601 ocfs2_journal_dirty(handle, ref_leaf_bh);
1602 ocfs2_journal_dirty(handle, new_bh);
1604 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1606 trace_ocfs2_new_leaf_refcount_block(
1607 (unsigned long long)new_bh->b_blocknr, new_cpos);
1609 /* Insert the new leaf block with the specific offset cpos. */
1610 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1620 static int ocfs2_expand_refcount_tree(handle_t *handle,
1621 struct ocfs2_caching_info *ci,
1622 struct buffer_head *ref_root_bh,
1623 struct buffer_head *ref_leaf_bh,
1624 struct ocfs2_alloc_context *meta_ac)
1627 struct buffer_head *expand_bh = NULL;
1629 if (ref_root_bh == ref_leaf_bh) {
1631 * the old root bh hasn't been expanded to a b-tree,
1632 * so expand it first.
1634 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1635 &expand_bh, meta_ac);
1641 expand_bh = ref_leaf_bh;
1646 /* Now add a new refcount block into the tree.*/
1647 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1648 expand_bh, meta_ac);
1657 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1659 * Only called when we have inserted a new refcount rec at index 0
1660 * which means ocfs2_extent_rec.e_cpos may need some change.
1662 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1663 struct ocfs2_caching_info *ci,
1664 struct buffer_head *ref_root_bh,
1665 struct buffer_head *ref_leaf_bh,
1666 struct ocfs2_refcount_rec *rec)
1669 u32 new_cpos, old_cpos;
1670 struct ocfs2_path *path = NULL;
1671 struct ocfs2_extent_tree et;
1672 struct ocfs2_refcount_block *rb =
1673 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1674 struct ocfs2_extent_list *el;
1676 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1679 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1680 old_cpos = le32_to_cpu(rb->rf_cpos);
1681 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1682 if (old_cpos <= new_cpos)
1685 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1687 path = ocfs2_new_path_from_et(&et);
1694 ret = ocfs2_find_path(ci, path, old_cpos);
1701 * 2 more credits, one for the leaf refcount block, one for
1702 * the extent block contains the extent rec.
1704 ret = ocfs2_extend_trans(handle, 2);
1710 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1711 OCFS2_JOURNAL_ACCESS_WRITE);
1717 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1718 OCFS2_JOURNAL_ACCESS_WRITE);
1724 /* change the leaf extent block first. */
1725 el = path_leaf_el(path);
1727 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1728 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1731 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1733 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1735 /* change the r_cpos in the leaf block. */
1736 rb->rf_cpos = cpu_to_le32(new_cpos);
1738 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1739 ocfs2_journal_dirty(handle, ref_leaf_bh);
1742 ocfs2_free_path(path);
1746 static int ocfs2_insert_refcount_rec(handle_t *handle,
1747 struct ocfs2_caching_info *ci,
1748 struct buffer_head *ref_root_bh,
1749 struct buffer_head *ref_leaf_bh,
1750 struct ocfs2_refcount_rec *rec,
1751 int index, int merge,
1752 struct ocfs2_alloc_context *meta_ac)
1755 struct ocfs2_refcount_block *rb =
1756 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1757 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1758 struct buffer_head *new_bh = NULL;
1760 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1762 if (rf_list->rl_used == rf_list->rl_count) {
1763 u64 cpos = le64_to_cpu(rec->r_cpos);
1764 u32 len = le32_to_cpu(rec->r_clusters);
1766 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1767 ref_leaf_bh, meta_ac);
1773 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1774 cpos, len, NULL, &index,
1781 ref_leaf_bh = new_bh;
1782 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1783 rf_list = &rb->rf_records;
1786 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1787 OCFS2_JOURNAL_ACCESS_WRITE);
1793 if (index < le16_to_cpu(rf_list->rl_used))
1794 memmove(&rf_list->rl_recs[index + 1],
1795 &rf_list->rl_recs[index],
1796 (le16_to_cpu(rf_list->rl_used) - index) *
1797 sizeof(struct ocfs2_refcount_rec));
1799 trace_ocfs2_insert_refcount_rec(
1800 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1801 (unsigned long long)le64_to_cpu(rec->r_cpos),
1802 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1804 rf_list->rl_recs[index] = *rec;
1806 le16_add_cpu(&rf_list->rl_used, 1);
1809 ocfs2_refcount_rec_merge(rb, index);
1811 ocfs2_journal_dirty(handle, ref_leaf_bh);
1814 ret = ocfs2_adjust_refcount_rec(handle, ci,
1826 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1827 * This is much simple than our b-tree code.
1828 * split_rec is the new refcount rec we want to insert.
1829 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1830 * increase refcount or decrease a refcount to non-zero).
1831 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1832 * rec( in case we decrease a refcount to zero).
1834 static int ocfs2_split_refcount_rec(handle_t *handle,
1835 struct ocfs2_caching_info *ci,
1836 struct buffer_head *ref_root_bh,
1837 struct buffer_head *ref_leaf_bh,
1838 struct ocfs2_refcount_rec *split_rec,
1839 int index, int merge,
1840 struct ocfs2_alloc_context *meta_ac,
1841 struct ocfs2_cached_dealloc_ctxt *dealloc)
1845 struct ocfs2_refcount_block *rb =
1846 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1847 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1848 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1849 struct ocfs2_refcount_rec *tail_rec = NULL;
1850 struct buffer_head *new_bh = NULL;
1852 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1854 trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1855 le32_to_cpu(orig_rec->r_clusters),
1856 le32_to_cpu(orig_rec->r_refcount),
1857 le64_to_cpu(split_rec->r_cpos),
1858 le32_to_cpu(split_rec->r_clusters),
1859 le32_to_cpu(split_rec->r_refcount));
1862 * If we just need to split the header or tail clusters,
1863 * no more recs are needed, just split is OK.
1864 * Otherwise we at least need one new recs.
1866 if (!split_rec->r_refcount &&
1867 (split_rec->r_cpos == orig_rec->r_cpos ||
1868 le64_to_cpu(split_rec->r_cpos) +
1869 le32_to_cpu(split_rec->r_clusters) ==
1870 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1876 * We need one more rec if we split in the middle and the new rec have
1877 * some refcount in it.
1879 if (split_rec->r_refcount &&
1880 (split_rec->r_cpos != orig_rec->r_cpos &&
1881 le64_to_cpu(split_rec->r_cpos) +
1882 le32_to_cpu(split_rec->r_clusters) !=
1883 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1886 /* If the leaf block don't have enough record, expand it. */
1887 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1888 le16_to_cpu(rf_list->rl_count)) {
1889 struct ocfs2_refcount_rec tmp_rec;
1890 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1891 len = le32_to_cpu(orig_rec->r_clusters);
1892 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1893 ref_leaf_bh, meta_ac);
1900 * We have to re-get it since now cpos may be moved to
1901 * another leaf block.
1903 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1904 cpos, len, &tmp_rec, &index,
1911 ref_leaf_bh = new_bh;
1912 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1913 rf_list = &rb->rf_records;
1914 orig_rec = &rf_list->rl_recs[index];
1917 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1918 OCFS2_JOURNAL_ACCESS_WRITE);
1925 * We have calculated out how many new records we need and store
1926 * in recs_need, so spare enough space first by moving the records
1927 * after "index" to the end.
1929 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1930 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1931 &rf_list->rl_recs[index + 1],
1932 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1933 sizeof(struct ocfs2_refcount_rec));
1935 len = (le64_to_cpu(orig_rec->r_cpos) +
1936 le32_to_cpu(orig_rec->r_clusters)) -
1937 (le64_to_cpu(split_rec->r_cpos) +
1938 le32_to_cpu(split_rec->r_clusters));
1941 * If we have "len", the we will split in the tail and move it
1942 * to the end of the space we have just spared.
1945 tail_rec = &rf_list->rl_recs[index + recs_need];
1947 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1948 le64_add_cpu(&tail_rec->r_cpos,
1949 le32_to_cpu(tail_rec->r_clusters) - len);
1950 tail_rec->r_clusters = cpu_to_le32(len);
1954 * If the split pos isn't the same as the original one, we need to
1955 * split in the head.
1957 * Note: We have the chance that split_rec.r_refcount = 0,
1958 * recs_need = 0 and len > 0, which means we just cut the head from
1959 * the orig_rec and in that case we have done some modification in
1960 * orig_rec above, so the check for r_cpos is faked.
1962 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1963 len = le64_to_cpu(split_rec->r_cpos) -
1964 le64_to_cpu(orig_rec->r_cpos);
1965 orig_rec->r_clusters = cpu_to_le32(len);
1969 le16_add_cpu(&rf_list->rl_used, recs_need);
1971 if (split_rec->r_refcount) {
1972 rf_list->rl_recs[index] = *split_rec;
1973 trace_ocfs2_split_refcount_rec_insert(
1974 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1975 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1976 le32_to_cpu(split_rec->r_clusters),
1977 le32_to_cpu(split_rec->r_refcount));
1980 ocfs2_refcount_rec_merge(rb, index);
1983 ocfs2_journal_dirty(handle, ref_leaf_bh);
1990 static int __ocfs2_increase_refcount(handle_t *handle,
1991 struct ocfs2_caching_info *ci,
1992 struct buffer_head *ref_root_bh,
1993 u64 cpos, u32 len, int merge,
1994 struct ocfs2_alloc_context *meta_ac,
1995 struct ocfs2_cached_dealloc_ctxt *dealloc)
1998 struct buffer_head *ref_leaf_bh = NULL;
1999 struct ocfs2_refcount_rec rec;
2000 unsigned int set_len = 0;
2002 trace_ocfs2_increase_refcount_begin(
2003 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2004 (unsigned long long)cpos, len);
2007 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2008 cpos, len, &rec, &index,
2015 set_len = le32_to_cpu(rec.r_clusters);
2018 * Here we may meet with 3 situations:
2020 * 1. If we find an already existing record, and the length
2021 * is the same, cool, we just need to increase the r_refcount
2023 * 2. If we find a hole, just insert it with r_refcount = 1.
2024 * 3. If we are in the middle of one extent record, split
2027 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2029 trace_ocfs2_increase_refcount_change(
2030 (unsigned long long)cpos, set_len,
2031 le32_to_cpu(rec.r_refcount));
2032 ret = ocfs2_change_refcount_rec(handle, ci,
2039 } else if (!rec.r_refcount) {
2040 rec.r_refcount = cpu_to_le32(1);
2042 trace_ocfs2_increase_refcount_insert(
2043 (unsigned long long)le64_to_cpu(rec.r_cpos),
2045 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2054 set_len = min((u64)(cpos + len),
2055 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2056 rec.r_cpos = cpu_to_le64(cpos);
2057 rec.r_clusters = cpu_to_le32(set_len);
2058 le32_add_cpu(&rec.r_refcount, 1);
2060 trace_ocfs2_increase_refcount_split(
2061 (unsigned long long)le64_to_cpu(rec.r_cpos),
2062 set_len, le32_to_cpu(rec.r_refcount));
2063 ret = ocfs2_split_refcount_rec(handle, ci,
2064 ref_root_bh, ref_leaf_bh,
2075 brelse(ref_leaf_bh);
2080 brelse(ref_leaf_bh);
2084 static int ocfs2_remove_refcount_extent(handle_t *handle,
2085 struct ocfs2_caching_info *ci,
2086 struct buffer_head *ref_root_bh,
2087 struct buffer_head *ref_leaf_bh,
2088 struct ocfs2_alloc_context *meta_ac,
2089 struct ocfs2_cached_dealloc_ctxt *dealloc)
2092 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2093 struct ocfs2_refcount_block *rb =
2094 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2095 struct ocfs2_extent_tree et;
2097 BUG_ON(rb->rf_records.rl_used);
2099 trace_ocfs2_remove_refcount_extent(
2100 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2101 (unsigned long long)ref_leaf_bh->b_blocknr,
2102 le32_to_cpu(rb->rf_cpos));
2104 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2105 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2106 1, meta_ac, dealloc);
2112 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2115 * add the freed block to the dealloc so that it will be freed
2116 * when we run dealloc.
2118 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2119 le16_to_cpu(rb->rf_suballoc_slot),
2120 le64_to_cpu(rb->rf_suballoc_loc),
2121 le64_to_cpu(rb->rf_blkno),
2122 le16_to_cpu(rb->rf_suballoc_bit));
2128 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2129 OCFS2_JOURNAL_ACCESS_WRITE);
2135 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2137 le32_add_cpu(&rb->rf_clusters, -1);
2140 * check whether we need to restore the root refcount block if
2141 * there is no leaf extent block at atll.
2143 if (!rb->rf_list.l_next_free_rec) {
2144 BUG_ON(rb->rf_clusters);
2146 trace_ocfs2_restore_refcount_block(
2147 (unsigned long long)ref_root_bh->b_blocknr);
2152 memset(&rb->rf_records, 0, sb->s_blocksize -
2153 offsetof(struct ocfs2_refcount_block, rf_records));
2154 rb->rf_records.rl_count =
2155 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2158 ocfs2_journal_dirty(handle, ref_root_bh);
2164 int ocfs2_increase_refcount(handle_t *handle,
2165 struct ocfs2_caching_info *ci,
2166 struct buffer_head *ref_root_bh,
2168 struct ocfs2_alloc_context *meta_ac,
2169 struct ocfs2_cached_dealloc_ctxt *dealloc)
2171 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2176 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2177 struct ocfs2_caching_info *ci,
2178 struct buffer_head *ref_root_bh,
2179 struct buffer_head *ref_leaf_bh,
2180 int index, u64 cpos, unsigned int len,
2181 struct ocfs2_alloc_context *meta_ac,
2182 struct ocfs2_cached_dealloc_ctxt *dealloc)
2185 struct ocfs2_refcount_block *rb =
2186 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2187 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2189 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2191 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2193 trace_ocfs2_decrease_refcount_rec(
2194 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2195 (unsigned long long)cpos, len);
2197 if (cpos == le64_to_cpu(rec->r_cpos) &&
2198 len == le32_to_cpu(rec->r_clusters))
2199 ret = ocfs2_change_refcount_rec(handle, ci,
2200 ref_leaf_bh, index, 1, -1);
2202 struct ocfs2_refcount_rec split = *rec;
2203 split.r_cpos = cpu_to_le64(cpos);
2204 split.r_clusters = cpu_to_le32(len);
2206 le32_add_cpu(&split.r_refcount, -1);
2208 ret = ocfs2_split_refcount_rec(handle, ci,
2209 ref_root_bh, ref_leaf_bh,
2219 /* Remove the leaf refcount block if it contains no refcount record. */
2220 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2221 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2222 ref_leaf_bh, meta_ac,
2232 static int __ocfs2_decrease_refcount(handle_t *handle,
2233 struct ocfs2_caching_info *ci,
2234 struct buffer_head *ref_root_bh,
2236 struct ocfs2_alloc_context *meta_ac,
2237 struct ocfs2_cached_dealloc_ctxt *dealloc,
2240 int ret = 0, index = 0;
2241 struct ocfs2_refcount_rec rec;
2242 unsigned int r_count = 0, r_len;
2243 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2244 struct buffer_head *ref_leaf_bh = NULL;
2246 trace_ocfs2_decrease_refcount(
2247 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2248 (unsigned long long)cpos, len, delete);
2251 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2252 cpos, len, &rec, &index,
2259 r_count = le32_to_cpu(rec.r_refcount);
2260 BUG_ON(r_count == 0);
2262 BUG_ON(r_count > 1);
2264 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2265 le32_to_cpu(rec.r_clusters)) - cpos;
2267 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2276 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2277 ret = ocfs2_cache_cluster_dealloc(dealloc,
2278 ocfs2_clusters_to_blocks(sb, cpos),
2288 brelse(ref_leaf_bh);
2293 brelse(ref_leaf_bh);
2297 /* Caller must hold refcount tree lock. */
2298 int ocfs2_decrease_refcount(struct inode *inode,
2299 handle_t *handle, u32 cpos, u32 len,
2300 struct ocfs2_alloc_context *meta_ac,
2301 struct ocfs2_cached_dealloc_ctxt *dealloc,
2306 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2307 struct buffer_head *ref_root_bh = NULL;
2308 struct ocfs2_refcount_tree *tree;
2310 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2312 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2318 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2324 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2331 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2332 cpos, len, meta_ac, dealloc, delete);
2336 brelse(ref_root_bh);
2341 * Mark the already-existing extent at cpos as refcounted for len clusters.
2342 * This adds the refcount extent flag.
2344 * If the existing extent is larger than the request, initiate a
2345 * split. An attempt will be made at merging with adjacent extents.
2347 * The caller is responsible for passing down meta_ac if we'll need it.
2349 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2350 struct ocfs2_extent_tree *et,
2351 handle_t *handle, u32 cpos,
2353 struct ocfs2_alloc_context *meta_ac,
2354 struct ocfs2_cached_dealloc_ctxt *dealloc)
2358 trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2361 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2362 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2363 "tree, but the feature bit is not set in the "
2364 "super block.", inode->i_ino);
2369 ret = ocfs2_change_extent_flag(handle, et, cpos,
2370 len, phys, meta_ac, dealloc,
2371 OCFS2_EXT_REFCOUNTED, 0);
2380 * Given some contiguous physical clusters, calculate what we need
2381 * for modifying their refcount.
2383 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2384 struct ocfs2_caching_info *ci,
2385 struct buffer_head *ref_root_bh,
2391 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2392 u64 cpos = start_cpos;
2393 struct ocfs2_refcount_block *rb;
2394 struct ocfs2_refcount_rec rec;
2395 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2399 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2400 cpos, clusters, &rec,
2401 &index, &ref_leaf_bh);
2407 if (ref_leaf_bh != prev_bh) {
2409 * Now we encounter a new leaf block, so calculate
2410 * whether we need to extend the old leaf.
2413 rb = (struct ocfs2_refcount_block *)
2416 if (le16_to_cpu(rb->rf_records.rl_used) +
2418 le16_to_cpu(rb->rf_records.rl_count))
2425 prev_bh = ref_leaf_bh;
2429 trace_ocfs2_calc_refcount_meta_credits_iterate(
2430 recs_add, (unsigned long long)cpos, clusters,
2431 (unsigned long long)le64_to_cpu(rec.r_cpos),
2432 le32_to_cpu(rec.r_clusters),
2433 le32_to_cpu(rec.r_refcount), index);
2435 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2436 le32_to_cpu(rec.r_clusters)) - cpos;
2438 * We record all the records which will be inserted to the
2439 * same refcount block, so that we can tell exactly whether
2440 * we need a new refcount block or not.
2442 * If we will insert a new one, this is easy and only happens
2443 * during adding refcounted flag to the extent, so we don't
2444 * have a chance of spliting. We just need one record.
2446 * If the refcount rec already exists, that would be a little
2447 * complicated. we may have to:
2448 * 1) split at the beginning if the start pos isn't aligned.
2449 * we need 1 more record in this case.
2450 * 2) split int the end if the end pos isn't aligned.
2451 * we need 1 more record in this case.
2452 * 3) split in the middle because of file system fragmentation.
2453 * we need 2 more records in this case(we can't detect this
2454 * beforehand, so always think of the worst case).
2456 if (rec.r_refcount) {
2458 /* Check whether we need a split at the beginning. */
2459 if (cpos == start_cpos &&
2460 cpos != le64_to_cpu(rec.r_cpos))
2463 /* Check whether we need a split in the end. */
2464 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2465 le32_to_cpu(rec.r_clusters))
2470 brelse(ref_leaf_bh);
2477 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2479 if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2480 le16_to_cpu(rb->rf_records.rl_count))
2489 *meta_add += ref_blocks;
2490 *credits += ref_blocks;
2493 * So we may need ref_blocks to insert into the tree.
2494 * That also means we need to change the b-tree and add that number
2495 * of records since we never merge them.
2496 * We need one more block for expansion since the new created leaf
2497 * block is also full and needs split.
2499 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2500 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2501 struct ocfs2_extent_tree et;
2503 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2504 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2505 *credits += ocfs2_calc_extend_credits(sb,
2508 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2514 trace_ocfs2_calc_refcount_meta_credits(
2515 (unsigned long long)start_cpos, clusters,
2516 *meta_add, *credits);
2517 brelse(ref_leaf_bh);
2523 * For refcount tree, we will decrease some contiguous clusters
2524 * refcount count, so just go through it to see how many blocks
2525 * we gonna touch and whether we need to create new blocks.
2527 * Normally the refcount blocks store these refcount should be
2528 * contiguous also, so that we can get the number easily.
2529 * We will at most add split 2 refcount records and 2 more
2530 * refcount blocks, so just check it in a rough way.
2532 * Caller must hold refcount tree lock.
2534 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2542 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2543 struct buffer_head *ref_root_bh = NULL;
2544 struct ocfs2_refcount_tree *tree;
2545 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2547 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2548 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2549 "tree, but the feature bit is not set in the "
2550 "super block.", inode->i_ino);
2555 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2557 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2558 refcount_loc, &tree);
2564 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2571 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2574 start_cpos, clusters,
2575 ref_blocks, credits);
2581 trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2584 brelse(ref_root_bh);
2588 #define MAX_CONTIG_BYTES 1048576
2590 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2592 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2595 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2597 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2601 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2602 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2603 * while still being less than or equal to it.
2605 * The goal is to break the extent at a multiple of contig_clusters.
2607 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2611 BUG_ON(start > cpos);
2613 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2617 * Given a cluster count of len, pad it out so that it is a multiple
2618 * of contig_clusters.
2620 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2623 unsigned int padded =
2624 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2625 ocfs2_cow_contig_mask(sb);
2635 * Calculate out the start and number of virtual clusters we need to to CoW.
2637 * cpos is vitual start cluster position we want to do CoW in a
2638 * file and write_len is the cluster length.
2639 * max_cpos is the place where we want to stop CoW intentionally.
2641 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2642 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2643 * get good I/O from the resulting extent tree.
2645 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2646 struct ocfs2_extent_list *el,
2654 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2655 struct buffer_head *eb_bh = NULL;
2656 struct ocfs2_extent_block *eb = NULL;
2657 struct ocfs2_extent_rec *rec;
2658 unsigned int want_clusters, rec_end = 0;
2659 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2662 BUG_ON(cpos + write_len > max_cpos);
2664 if (tree_height > 0) {
2665 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2671 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2674 if (el->l_tree_depth) {
2675 ocfs2_error(inode->i_sb,
2676 "Inode %lu has non zero tree depth in "
2677 "leaf block %llu\n", inode->i_ino,
2678 (unsigned long long)eb_bh->b_blocknr);
2685 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2686 rec = &el->l_recs[i];
2688 if (ocfs2_is_empty_extent(rec)) {
2689 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2690 "index %d\n", inode->i_ino, i);
2694 if (le32_to_cpu(rec->e_cpos) +
2695 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2698 if (*cow_len == 0) {
2700 * We should find a refcounted record in the
2703 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2704 *cow_start = le32_to_cpu(rec->e_cpos);
2708 * If we encounter a hole, a non-refcounted record or
2709 * pass the max_cpos, stop the search.
2711 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2712 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2713 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2716 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2717 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2718 if (rec_end > max_cpos) {
2720 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2724 * How many clusters do we actually need from
2725 * this extent? First we see how many we actually
2726 * need to complete the write. If that's smaller
2727 * than contig_clusters, we try for contig_clusters.
2730 want_clusters = write_len;
2732 want_clusters = (cpos + write_len) -
2733 (*cow_start + *cow_len);
2734 if (want_clusters < contig_clusters)
2735 want_clusters = contig_clusters;
2738 * If the write does not cover the whole extent, we
2739 * need to calculate how we're going to split the extent.
2740 * We try to do it on contig_clusters boundaries.
2742 * Any extent smaller than contig_clusters will be
2743 * CoWed in its entirety.
2745 if (leaf_clusters <= contig_clusters)
2746 *cow_len += leaf_clusters;
2747 else if (*cow_len || (*cow_start == cpos)) {
2749 * This extent needs to be CoW'd from its
2750 * beginning, so all we have to do is compute
2751 * how many clusters to grab. We align
2752 * want_clusters to the edge of contig_clusters
2753 * to get better I/O.
2755 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2758 if (leaf_clusters < want_clusters)
2759 *cow_len += leaf_clusters;
2761 *cow_len += want_clusters;
2762 } else if ((*cow_start + contig_clusters) >=
2763 (cpos + write_len)) {
2765 * Breaking off contig_clusters at the front
2766 * of the extent will cover our write. That's
2769 *cow_len = contig_clusters;
2770 } else if ((rec_end - cpos) <= contig_clusters) {
2772 * Breaking off contig_clusters at the tail of
2773 * this extent will cover cpos.
2775 *cow_start = rec_end - contig_clusters;
2776 *cow_len = contig_clusters;
2777 } else if ((rec_end - cpos) <= want_clusters) {
2779 * While we can't fit the entire write in this
2780 * extent, we know that the write goes from cpos
2781 * to the end of the extent. Break that off.
2782 * We try to break it at some multiple of
2783 * contig_clusters from the front of the extent.
2784 * Failing that (ie, cpos is within
2785 * contig_clusters of the front), we'll CoW the
2788 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2790 *cow_len = rec_end - *cow_start;
2793 * Ok, the entire write lives in the middle of
2794 * this extent. Let's try to slice the extent up
2795 * nicely. Optimally, our CoW region starts at
2796 * m*contig_clusters from the beginning of the
2797 * extent and goes for n*contig_clusters,
2798 * covering the entire write.
2800 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2803 want_clusters = (cpos + write_len) - *cow_start;
2804 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2806 if (*cow_start + want_clusters <= rec_end)
2807 *cow_len = want_clusters;
2809 *cow_len = rec_end - *cow_start;
2812 /* Have we covered our entire write yet? */
2813 if ((*cow_start + *cow_len) >= (cpos + write_len))
2817 * If we reach the end of the extent block and don't get enough
2818 * clusters, continue with the next extent block if possible.
2820 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2821 eb && eb->h_next_leaf_blk) {
2825 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2826 le64_to_cpu(eb->h_next_leaf_blk),
2833 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2845 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2846 * num_clusters in data_tree "et" and change the refcount for the old
2847 * clusters(starting form p_cluster) in the refcount tree.
2850 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2851 * more new leaf records.
2852 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2853 * just give data_ac = NULL.
2855 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2856 u32 p_cluster, u32 num_clusters,
2857 struct ocfs2_extent_tree *et,
2858 struct ocfs2_caching_info *ref_ci,
2859 struct buffer_head *ref_root_bh,
2860 struct ocfs2_alloc_context **meta_ac,
2861 struct ocfs2_alloc_context **data_ac,
2864 int ret = 0, meta_add = 0;
2865 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2867 if (num_free_extents < 0) {
2868 ret = num_free_extents;
2873 if (num_free_extents < num_clusters + 2)
2875 ocfs2_extend_meta_needed(et->et_root_el);
2877 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2879 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2880 p_cluster, num_clusters,
2881 &meta_add, credits);
2887 trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2888 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2896 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2905 ocfs2_free_alloc_context(*meta_ac);
2913 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2915 BUG_ON(buffer_dirty(bh));
2917 clear_buffer_mapped(bh);
2922 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2923 struct inode *inode,
2924 u32 cpos, u32 old_cluster,
2925 u32 new_cluster, u32 new_len)
2927 int ret = 0, partial;
2928 struct super_block *sb = inode->i_sb;
2929 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2932 unsigned int from, to, readahead_pages;
2933 loff_t offset, end, map_end;
2934 struct address_space *mapping = inode->i_mapping;
2936 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2937 new_cluster, new_len);
2940 (ocfs2_cow_contig_clusters(sb) <<
2941 OCFS2_SB(sb)->s_clustersize_bits) >> PAGE_CACHE_SHIFT;
2942 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2943 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2945 * We only duplicate pages until we reach the page contains i_size - 1.
2946 * So trim 'end' to i_size.
2948 if (end > i_size_read(inode))
2949 end = i_size_read(inode);
2951 while (offset < end) {
2952 page_index = offset >> PAGE_CACHE_SHIFT;
2953 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2957 /* from, to is the offset within the page. */
2958 from = offset & (PAGE_CACHE_SIZE - 1);
2959 to = PAGE_CACHE_SIZE;
2960 if (map_end & (PAGE_CACHE_SIZE - 1))
2961 to = map_end & (PAGE_CACHE_SIZE - 1);
2963 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2971 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2972 * can't be dirtied before we CoW it out.
2974 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2975 BUG_ON(PageDirty(page));
2977 if (!PageUptodate(page)) {
2978 ret = block_read_full_page(page, ocfs2_get_block);
2986 if (page_has_buffers(page)) {
2987 ret = walk_page_buffers(handle, page_buffers(page),
2989 ocfs2_clear_cow_buffer);
2996 ocfs2_map_and_dirty_page(inode,
2998 page, 0, &new_block);
2999 mark_page_accessed(page);
3002 page_cache_release(page);
3012 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3013 struct inode *inode,
3014 u32 cpos, u32 old_cluster,
3015 u32 new_cluster, u32 new_len)
3018 struct super_block *sb = inode->i_sb;
3019 struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3020 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3021 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3022 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3023 struct ocfs2_super *osb = OCFS2_SB(sb);
3024 struct buffer_head *old_bh = NULL;
3025 struct buffer_head *new_bh = NULL;
3027 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3028 new_cluster, new_len);
3030 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3031 new_bh = sb_getblk(osb->sb, new_block);
3032 if (new_bh == NULL) {
3038 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3040 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3046 ret = ocfs2_journal_access(handle, ci, new_bh,
3047 OCFS2_JOURNAL_ACCESS_CREATE);
3053 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3054 ocfs2_journal_dirty(handle, new_bh);
3067 static int ocfs2_clear_ext_refcount(handle_t *handle,
3068 struct ocfs2_extent_tree *et,
3069 u32 cpos, u32 p_cluster, u32 len,
3070 unsigned int ext_flags,
3071 struct ocfs2_alloc_context *meta_ac,
3072 struct ocfs2_cached_dealloc_ctxt *dealloc)
3075 struct ocfs2_extent_rec replace_rec;
3076 struct ocfs2_path *path = NULL;
3077 struct ocfs2_extent_list *el;
3078 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3079 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3081 trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3082 cpos, len, p_cluster, ext_flags);
3084 memset(&replace_rec, 0, sizeof(replace_rec));
3085 replace_rec.e_cpos = cpu_to_le32(cpos);
3086 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3087 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3089 replace_rec.e_flags = ext_flags;
3090 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3092 path = ocfs2_new_path_from_et(et);
3099 ret = ocfs2_find_path(et->et_ci, path, cpos);
3105 el = path_leaf_el(path);
3107 index = ocfs2_search_extent_list(el, cpos);
3110 "Inode %llu has an extent at cpos %u which can no "
3111 "longer be found.\n",
3112 (unsigned long long)ino, cpos);
3117 ret = ocfs2_split_extent(handle, et, path, index,
3118 &replace_rec, meta_ac, dealloc);
3123 ocfs2_free_path(path);
3127 static int ocfs2_replace_clusters(handle_t *handle,
3128 struct ocfs2_cow_context *context,
3131 unsigned int ext_flags)
3134 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3135 u64 ino = ocfs2_metadata_cache_owner(ci);
3137 trace_ocfs2_replace_clusters((unsigned long long)ino,
3138 cpos, old, new, len, ext_flags);
3140 /*If the old clusters is unwritten, no need to duplicate. */
3141 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3142 ret = context->cow_duplicate_clusters(handle, context->inode,
3143 cpos, old, new, len);
3150 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3151 cpos, new, len, ext_flags,
3152 context->meta_ac, &context->dealloc);
3159 int ocfs2_cow_sync_writeback(struct super_block *sb,
3160 struct inode *inode,
3161 u32 cpos, u32 num_clusters)
3164 loff_t offset, end, map_end;
3168 if (ocfs2_should_order_data(inode))
3171 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3172 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3174 ret = filemap_fdatawrite_range(inode->i_mapping,
3181 while (offset < end) {
3182 page_index = offset >> PAGE_CACHE_SHIFT;
3183 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3187 page = find_or_create_page(inode->i_mapping,
3188 page_index, GFP_NOFS);
3191 wait_on_page_writeback(page);
3192 if (PageError(page)) {
3196 mark_page_accessed(page);
3199 page_cache_release(page);
3209 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3210 u32 v_cluster, u32 *p_cluster,
3212 unsigned int *extent_flags)
3214 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3215 num_clusters, extent_flags);
3218 static int ocfs2_make_clusters_writable(struct super_block *sb,
3219 struct ocfs2_cow_context *context,
3220 u32 cpos, u32 p_cluster,
3221 u32 num_clusters, unsigned int e_flags)
3223 int ret, delete, index, credits = 0;
3224 u32 new_bit, new_len, orig_num_clusters;
3225 unsigned int set_len;
3226 struct ocfs2_super *osb = OCFS2_SB(sb);
3228 struct buffer_head *ref_leaf_bh = NULL;
3229 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3230 struct ocfs2_refcount_rec rec;
3232 trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3233 num_clusters, e_flags);
3235 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3238 context->ref_root_bh,
3240 &context->data_ac, &credits);
3246 if (context->post_refcount)
3247 credits += context->post_refcount->credits;
3249 credits += context->extra_credits;
3250 handle = ocfs2_start_trans(osb, credits);
3251 if (IS_ERR(handle)) {
3252 ret = PTR_ERR(handle);
3257 orig_num_clusters = num_clusters;
3259 while (num_clusters) {
3260 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3261 p_cluster, num_clusters,
3262 &rec, &index, &ref_leaf_bh);
3268 BUG_ON(!rec.r_refcount);
3269 set_len = min((u64)p_cluster + num_clusters,
3270 le64_to_cpu(rec.r_cpos) +
3271 le32_to_cpu(rec.r_clusters)) - p_cluster;
3274 * There are many different situation here.
3275 * 1. If refcount == 1, remove the flag and don't COW.
3276 * 2. If refcount > 1, allocate clusters.
3277 * Here we may not allocate r_len once at a time, so continue
3278 * until we reach num_clusters.
3280 if (le32_to_cpu(rec.r_refcount) == 1) {
3282 ret = ocfs2_clear_ext_refcount(handle,
3295 ret = __ocfs2_claim_clusters(handle,
3298 &new_bit, &new_len);
3304 ret = ocfs2_replace_clusters(handle, context,
3305 cpos, p_cluster, new_bit,
3314 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3315 context->ref_root_bh,
3318 &context->dealloc, delete);
3325 p_cluster += set_len;
3326 num_clusters -= set_len;
3327 brelse(ref_leaf_bh);
3331 /* handle any post_cow action. */
3332 if (context->post_refcount && context->post_refcount->func) {
3333 ret = context->post_refcount->func(context->inode, handle,
3334 context->post_refcount->para);
3342 * Here we should write the new page out first if we are
3343 * in write-back mode.
3345 if (context->get_clusters == ocfs2_di_get_clusters) {
3346 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3353 ocfs2_commit_trans(osb, handle);
3356 if (context->data_ac) {
3357 ocfs2_free_alloc_context(context->data_ac);
3358 context->data_ac = NULL;
3360 if (context->meta_ac) {
3361 ocfs2_free_alloc_context(context->meta_ac);
3362 context->meta_ac = NULL;
3364 brelse(ref_leaf_bh);
3369 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3372 struct inode *inode = context->inode;
3373 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3374 u32 p_cluster, num_clusters;
3375 unsigned int ext_flags;
3376 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3378 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3379 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3380 "tree, but the feature bit is not set in the "
3381 "super block.", inode->i_ino);
3385 ocfs2_init_dealloc_ctxt(&context->dealloc);
3388 ret = context->get_clusters(context, cow_start, &p_cluster,
3389 &num_clusters, &ext_flags);
3395 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3397 if (cow_len < num_clusters)
3398 num_clusters = cow_len;
3400 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3401 cow_start, p_cluster,
3402 num_clusters, ext_flags);
3408 cow_len -= num_clusters;
3409 cow_start += num_clusters;
3412 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3413 ocfs2_schedule_truncate_log_flush(osb, 1);
3414 ocfs2_run_deallocs(osb, &context->dealloc);
3421 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3422 * past max_cpos. This will stop when it runs into a hole or an
3423 * unrefcounted extent.
3425 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3426 struct buffer_head *di_bh,
3427 u32 cpos, u32 write_len, u32 max_cpos)
3430 u32 cow_start = 0, cow_len = 0;
3431 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3432 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3433 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3434 struct buffer_head *ref_root_bh = NULL;
3435 struct ocfs2_refcount_tree *ref_tree;
3436 struct ocfs2_cow_context *context = NULL;
3438 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3440 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3441 cpos, write_len, max_cpos,
3442 &cow_start, &cow_len);
3448 trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3449 cpos, write_len, max_cpos,
3450 cow_start, cow_len);
3452 BUG_ON(cow_len == 0);
3454 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3461 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3462 1, &ref_tree, &ref_root_bh);
3468 context->inode = inode;
3469 context->cow_start = cow_start;
3470 context->cow_len = cow_len;
3471 context->ref_tree = ref_tree;
3472 context->ref_root_bh = ref_root_bh;
3473 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3474 context->get_clusters = ocfs2_di_get_clusters;
3476 ocfs2_init_dinode_extent_tree(&context->data_et,
3477 INODE_CACHE(inode), di_bh);
3479 ret = ocfs2_replace_cow(context);
3484 * truncate the extent map here since no matter whether we meet with
3485 * any error during the action, we shouldn't trust cached extent map
3488 ocfs2_extent_map_trunc(inode, cow_start);
3490 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3491 brelse(ref_root_bh);
3498 * CoW any and all clusters between cpos and cpos+write_len.
3499 * Don't CoW past max_cpos. If this returns successfully, all
3500 * clusters between cpos and cpos+write_len are safe to modify.
3502 int ocfs2_refcount_cow(struct inode *inode,
3503 struct buffer_head *di_bh,
3504 u32 cpos, u32 write_len, u32 max_cpos)
3507 u32 p_cluster, num_clusters;
3508 unsigned int ext_flags;
3511 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3512 &num_clusters, &ext_flags);
3518 if (write_len < num_clusters)
3519 num_clusters = write_len;
3521 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3522 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3523 num_clusters, max_cpos);
3530 write_len -= num_clusters;
3531 cpos += num_clusters;
3537 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3538 u32 v_cluster, u32 *p_cluster,
3540 unsigned int *extent_flags)
3542 struct inode *inode = context->inode;
3543 struct ocfs2_xattr_value_root *xv = context->cow_object;
3545 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3546 num_clusters, &xv->xr_list,
3551 * Given a xattr value root, calculate the most meta/credits we need for
3552 * refcount tree change if we truncate it to 0.
3554 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3555 struct ocfs2_caching_info *ref_ci,
3556 struct buffer_head *ref_root_bh,
3557 struct ocfs2_xattr_value_root *xv,
3558 int *meta_add, int *credits)
3560 int ret = 0, index, ref_blocks = 0;
3561 u32 p_cluster, num_clusters;
3562 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3563 struct ocfs2_refcount_block *rb;
3564 struct ocfs2_refcount_rec rec;
3565 struct buffer_head *ref_leaf_bh = NULL;
3567 while (cpos < clusters) {
3568 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3569 &num_clusters, &xv->xr_list,
3576 cpos += num_clusters;
3578 while (num_clusters) {
3579 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3580 p_cluster, num_clusters,
3588 BUG_ON(!rec.r_refcount);
3590 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3593 * We really don't know whether the other clusters is in
3594 * this refcount block or not, so just take the worst
3595 * case that all the clusters are in this block and each
3596 * one will split a refcount rec, so totally we need
3597 * clusters * 2 new refcount rec.
3599 if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3600 le16_to_cpu(rb->rf_records.rl_count))
3604 brelse(ref_leaf_bh);
3607 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3610 num_clusters -= le32_to_cpu(rec.r_clusters);
3611 p_cluster += num_clusters;
3615 *meta_add += ref_blocks;
3619 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3620 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3621 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3623 struct ocfs2_extent_tree et;
3625 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3626 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3631 brelse(ref_leaf_bh);
3638 int ocfs2_refcount_cow_xattr(struct inode *inode,
3639 struct ocfs2_dinode *di,
3640 struct ocfs2_xattr_value_buf *vb,
3641 struct ocfs2_refcount_tree *ref_tree,
3642 struct buffer_head *ref_root_bh,
3643 u32 cpos, u32 write_len,
3644 struct ocfs2_post_refcount *post)
3647 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3648 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3649 struct ocfs2_cow_context *context = NULL;
3650 u32 cow_start, cow_len;
3652 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3654 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3655 cpos, write_len, UINT_MAX,
3656 &cow_start, &cow_len);
3662 BUG_ON(cow_len == 0);
3664 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3671 context->inode = inode;
3672 context->cow_start = cow_start;
3673 context->cow_len = cow_len;
3674 context->ref_tree = ref_tree;
3675 context->ref_root_bh = ref_root_bh;
3676 context->cow_object = xv;
3678 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3679 /* We need the extra credits for duplicate_clusters by jbd. */
3680 context->extra_credits =
3681 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3682 context->get_clusters = ocfs2_xattr_value_get_clusters;
3683 context->post_refcount = post;
3685 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3686 INODE_CACHE(inode), vb);
3688 ret = ocfs2_replace_cow(context);
3698 * Insert a new extent into refcount tree and mark a extent rec
3699 * as refcounted in the dinode tree.
3701 int ocfs2_add_refcount_flag(struct inode *inode,
3702 struct ocfs2_extent_tree *data_et,
3703 struct ocfs2_caching_info *ref_ci,
3704 struct buffer_head *ref_root_bh,
3705 u32 cpos, u32 p_cluster, u32 num_clusters,
3706 struct ocfs2_cached_dealloc_ctxt *dealloc,
3707 struct ocfs2_post_refcount *post)
3711 int credits = 1, ref_blocks = 0;
3712 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3713 struct ocfs2_alloc_context *meta_ac = NULL;
3715 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3716 ref_ci, ref_root_bh,
3717 p_cluster, num_clusters,
3718 &ref_blocks, &credits);
3724 trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3727 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3728 ref_blocks, &meta_ac);
3736 credits += post->credits;
3738 handle = ocfs2_start_trans(osb, credits);
3739 if (IS_ERR(handle)) {
3740 ret = PTR_ERR(handle);
3745 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3746 cpos, num_clusters, p_cluster,
3753 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3754 p_cluster, num_clusters, 0,
3761 if (post && post->func) {
3762 ret = post->func(inode, handle, post->para);
3768 ocfs2_commit_trans(osb, handle);
3771 ocfs2_free_alloc_context(meta_ac);
3775 static int ocfs2_change_ctime(struct inode *inode,
3776 struct buffer_head *di_bh)
3780 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3782 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3783 OCFS2_INODE_UPDATE_CREDITS);
3784 if (IS_ERR(handle)) {
3785 ret = PTR_ERR(handle);
3790 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3791 OCFS2_JOURNAL_ACCESS_WRITE);
3797 inode->i_ctime = CURRENT_TIME;
3798 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3799 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3801 ocfs2_journal_dirty(handle, di_bh);
3804 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3809 static int ocfs2_attach_refcount_tree(struct inode *inode,
3810 struct buffer_head *di_bh)
3812 int ret, data_changed = 0;
3813 struct buffer_head *ref_root_bh = NULL;
3814 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3815 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3816 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3817 struct ocfs2_refcount_tree *ref_tree;
3818 unsigned int ext_flags;
3820 u32 cpos, num_clusters, clusters, p_cluster;
3821 struct ocfs2_cached_dealloc_ctxt dealloc;
3822 struct ocfs2_extent_tree di_et;
3824 ocfs2_init_dealloc_ctxt(&dealloc);
3826 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3827 ret = ocfs2_create_refcount_tree(inode, di_bh);
3834 BUG_ON(!di->i_refcount_loc);
3835 ret = ocfs2_lock_refcount_tree(osb,
3836 le64_to_cpu(di->i_refcount_loc), 1,
3837 &ref_tree, &ref_root_bh);
3843 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3846 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3848 size = i_size_read(inode);
3849 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3852 while (cpos < clusters) {
3853 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3854 &num_clusters, &ext_flags);
3859 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3860 ret = ocfs2_add_refcount_flag(inode, &di_et,
3863 p_cluster, num_clusters,
3872 cpos += num_clusters;
3876 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3877 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3888 ret = ocfs2_change_ctime(inode, di_bh);
3894 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3895 brelse(ref_root_bh);
3897 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3898 ocfs2_schedule_truncate_log_flush(osb, 1);
3899 ocfs2_run_deallocs(osb, &dealloc);
3903 * Empty the extent map so that we may get the right extent
3904 * record from the disk.
3906 ocfs2_extent_map_trunc(inode, 0);
3911 static int ocfs2_add_refcounted_extent(struct inode *inode,
3912 struct ocfs2_extent_tree *et,
3913 struct ocfs2_caching_info *ref_ci,
3914 struct buffer_head *ref_root_bh,
3915 u32 cpos, u32 p_cluster, u32 num_clusters,
3916 unsigned int ext_flags,
3917 struct ocfs2_cached_dealloc_ctxt *dealloc)
3922 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3923 struct ocfs2_alloc_context *meta_ac = NULL;
3925 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3926 p_cluster, num_clusters,
3928 ref_root_bh, &meta_ac,
3935 handle = ocfs2_start_trans(osb, credits);
3936 if (IS_ERR(handle)) {
3937 ret = PTR_ERR(handle);
3942 ret = ocfs2_insert_extent(handle, et, cpos,
3943 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3944 num_clusters, ext_flags, meta_ac);
3950 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3951 p_cluster, num_clusters,
3957 ocfs2_commit_trans(osb, handle);
3960 ocfs2_free_alloc_context(meta_ac);
3964 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3965 struct buffer_head *s_bh,
3966 struct inode *t_inode,
3967 struct buffer_head *t_bh)
3971 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3972 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3973 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3975 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3977 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3978 if (IS_ERR(handle)) {
3979 ret = PTR_ERR(handle);
3984 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3985 OCFS2_JOURNAL_ACCESS_WRITE);
3991 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3992 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3993 le16_to_cpu(s_di->id2.i_data.id_count));
3994 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3995 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3996 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3997 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3999 ocfs2_journal_dirty(handle, t_bh);
4002 ocfs2_commit_trans(osb, handle);
4007 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
4008 struct inode *t_inode,
4009 struct buffer_head *t_bh,
4010 struct ocfs2_caching_info *ref_ci,
4011 struct buffer_head *ref_root_bh,
4012 struct ocfs2_cached_dealloc_ctxt *dealloc)
4015 u32 p_cluster, num_clusters, clusters, cpos;
4017 unsigned int ext_flags;
4018 struct ocfs2_extent_tree et;
4020 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4022 size = i_size_read(s_inode);
4023 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4026 while (cpos < clusters) {
4027 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4028 &num_clusters, &ext_flags);
4034 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4035 ref_ci, ref_root_bh,
4046 cpos += num_clusters;
4054 * change the new file's attributes to the src.
4056 * reflink creates a snapshot of a file, that means the attributes
4057 * must be identical except for three exceptions - nlink, ino, and ctime.
4059 static int ocfs2_complete_reflink(struct inode *s_inode,
4060 struct buffer_head *s_bh,
4061 struct inode *t_inode,
4062 struct buffer_head *t_bh,
4067 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4068 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4069 loff_t size = i_size_read(s_inode);
4071 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4072 OCFS2_INODE_UPDATE_CREDITS);
4073 if (IS_ERR(handle)) {
4074 ret = PTR_ERR(handle);
4079 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4080 OCFS2_JOURNAL_ACCESS_WRITE);
4086 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4087 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4088 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4089 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4090 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4091 i_size_write(t_inode, size);
4092 t_inode->i_blocks = s_inode->i_blocks;
4094 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4095 di->i_clusters = s_di->i_clusters;
4096 di->i_size = s_di->i_size;
4097 di->i_dyn_features = s_di->i_dyn_features;
4098 di->i_attr = s_di->i_attr;
4101 t_inode->i_uid = s_inode->i_uid;
4102 t_inode->i_gid = s_inode->i_gid;
4103 t_inode->i_mode = s_inode->i_mode;
4104 di->i_uid = s_di->i_uid;
4105 di->i_gid = s_di->i_gid;
4106 di->i_mode = s_di->i_mode;
4110 * we want mtime to appear identical to the source and
4113 t_inode->i_ctime = CURRENT_TIME;
4115 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4116 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4118 t_inode->i_mtime = s_inode->i_mtime;
4119 di->i_mtime = s_di->i_mtime;
4120 di->i_mtime_nsec = s_di->i_mtime_nsec;
4123 ocfs2_journal_dirty(handle, t_bh);
4126 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4130 static int ocfs2_create_reflink_node(struct inode *s_inode,
4131 struct buffer_head *s_bh,
4132 struct inode *t_inode,
4133 struct buffer_head *t_bh,
4137 struct buffer_head *ref_root_bh = NULL;
4138 struct ocfs2_cached_dealloc_ctxt dealloc;
4139 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4140 struct ocfs2_refcount_block *rb;
4141 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4142 struct ocfs2_refcount_tree *ref_tree;
4144 ocfs2_init_dealloc_ctxt(&dealloc);
4146 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4147 le64_to_cpu(di->i_refcount_loc));
4153 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4154 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4161 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4162 1, &ref_tree, &ref_root_bh);
4167 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4169 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4170 &ref_tree->rf_ci, ref_root_bh,
4174 goto out_unlock_refcount;
4177 out_unlock_refcount:
4178 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4179 brelse(ref_root_bh);
4181 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4182 ocfs2_schedule_truncate_log_flush(osb, 1);
4183 ocfs2_run_deallocs(osb, &dealloc);
4189 static int __ocfs2_reflink(struct dentry *old_dentry,
4190 struct buffer_head *old_bh,
4191 struct inode *new_inode,
4195 struct inode *inode = d_inode(old_dentry);
4196 struct buffer_head *new_bh = NULL;
4198 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4204 ret = filemap_fdatawrite(inode->i_mapping);
4210 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4216 mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4217 ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4218 OI_LS_REFLINK_TARGET);
4224 ret = ocfs2_create_reflink_node(inode, old_bh,
4225 new_inode, new_bh, preserve);
4231 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4232 ret = ocfs2_reflink_xattrs(inode, old_bh,
4241 ret = ocfs2_complete_reflink(inode, old_bh,
4242 new_inode, new_bh, preserve);
4247 ocfs2_inode_unlock(new_inode, 1);
4250 mutex_unlock(&new_inode->i_mutex);
4253 ret = filemap_fdatawait(inode->i_mapping);
4260 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4261 struct dentry *new_dentry, bool preserve)
4264 struct inode *inode = d_inode(old_dentry);
4265 struct buffer_head *old_bh = NULL;
4266 struct inode *new_orphan_inode = NULL;
4267 struct posix_acl *default_acl, *acl;
4270 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4273 mode = inode->i_mode;
4274 error = posix_acl_create(dir, &mode, &default_acl, &acl);
4280 error = ocfs2_create_inode_in_orphan(dir, mode,
4287 error = ocfs2_rw_lock(inode, 1);
4293 error = ocfs2_inode_lock(inode, &old_bh, 1);
4296 ocfs2_rw_unlock(inode, 1);
4300 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4301 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4302 error = __ocfs2_reflink(old_dentry, old_bh,
4303 new_orphan_inode, preserve);
4304 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4305 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4307 ocfs2_inode_unlock(inode, 1);
4308 ocfs2_rw_unlock(inode, 1);
4316 /* If the security isn't preserved, we need to re-initialize them. */
4318 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4319 &new_dentry->d_name,
4326 posix_acl_release(default_acl);
4328 posix_acl_release(acl);
4330 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4336 if (new_orphan_inode) {
4338 * We need to open_unlock the inode no matter whether we
4339 * succeed or not, so that other nodes can delete it later.
4341 ocfs2_open_unlock(new_orphan_inode);
4343 iput(new_orphan_inode);
4350 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4351 * sys_reflink(). This will go away when vfs_reflink() exists in
4355 /* copied from may_create in VFS. */
4356 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4358 if (d_really_is_positive(child))
4360 if (IS_DEADDIR(dir))
4362 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4366 * ocfs2_vfs_reflink - Create a reference-counted link
4368 * @old_dentry: source dentry + inode
4369 * @dir: directory to create the target
4370 * @new_dentry: target dentry
4371 * @preserve: if true, preserve all file attributes
4373 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4374 struct dentry *new_dentry, bool preserve)
4376 struct inode *inode = d_inode(old_dentry);
4382 error = ocfs2_may_create(dir, new_dentry);
4386 if (dir->i_sb != inode->i_sb)
4390 * A reflink to an append-only or immutable file cannot be created.
4392 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4395 /* Only regular files can be reflinked. */
4396 if (!S_ISREG(inode->i_mode))
4400 * If the caller wants to preserve ownership, they require the
4404 if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4406 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4411 * If the caller is modifying any aspect of the attributes, they
4412 * are not creating a snapshot. They need read permission on the
4416 error = inode_permission(inode, MAY_READ);
4421 mutex_lock(&inode->i_mutex);
4422 dquot_initialize(dir);
4423 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4424 mutex_unlock(&inode->i_mutex);
4426 fsnotify_create(dir, new_dentry);
4430 * Most codes are copied from sys_linkat.
4432 int ocfs2_reflink_ioctl(struct inode *inode,
4433 const char __user *oldname,
4434 const char __user *newname,
4437 struct dentry *new_dentry;
4438 struct path old_path, new_path;
4441 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4444 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4450 new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4451 error = PTR_ERR(new_dentry);
4452 if (IS_ERR(new_dentry)) {
4458 if (old_path.mnt != new_path.mnt) {
4463 error = ocfs2_vfs_reflink(old_path.dentry,
4464 d_inode(new_path.dentry),
4465 new_dentry, preserve);
4467 done_path_create(&new_path, new_dentry);
4469 path_put(&old_path);