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, tmp;
1417 * The refcount cpos are ordered by their 64bit cpos,
1418 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1419 * So we need to make sure that this pos isn't intersected with others.
1421 * Note: The refcount block is already sorted by their low 32 bit cpos,
1422 * So just try the middle pos first, and we will exit when we find
1423 * the good position.
1425 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1426 u32 *split_pos, int *split_index)
1428 int num_used = le16_to_cpu(rl->rl_used);
1429 int delta, middle = num_used / 2;
1431 for (delta = 0; delta < middle; delta++) {
1432 /* Let's check delta earlier than middle */
1433 if (ocfs2_refcount_rec_no_intersect(
1434 &rl->rl_recs[middle - delta - 1],
1435 &rl->rl_recs[middle - delta])) {
1436 *split_index = middle - delta;
1440 /* For even counts, don't walk off the end */
1441 if ((middle + delta + 1) == num_used)
1444 /* Now try delta past middle */
1445 if (ocfs2_refcount_rec_no_intersect(
1446 &rl->rl_recs[middle + delta],
1447 &rl->rl_recs[middle + delta + 1])) {
1448 *split_index = middle + delta + 1;
1453 if (delta >= middle)
1456 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1460 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1461 struct buffer_head *new_bh,
1464 int split_index = 0, num_moved, ret;
1466 struct ocfs2_refcount_block *rb =
1467 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1468 struct ocfs2_refcount_list *rl = &rb->rf_records;
1469 struct ocfs2_refcount_block *new_rb =
1470 (struct ocfs2_refcount_block *)new_bh->b_data;
1471 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1473 trace_ocfs2_divide_leaf_refcount_block(
1474 (unsigned long long)ref_leaf_bh->b_blocknr,
1475 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1478 * XXX: Improvement later.
1479 * If we know all the high 32 bit cpos is the same, no need to sort.
1481 * In order to make the whole process safe, we do:
1482 * 1. sort the entries by their low 32 bit cpos first so that we can
1483 * find the split cpos easily.
1484 * 2. call ocfs2_insert_extent to insert the new refcount block.
1485 * 3. move the refcount rec to the new block.
1486 * 4. sort the entries by their 64 bit cpos.
1487 * 5. dirty the new_rb and rb.
1489 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1490 sizeof(struct ocfs2_refcount_rec),
1491 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1493 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1499 new_rb->rf_cpos = cpu_to_le32(cpos);
1501 /* move refcount records starting from split_index to the new block. */
1502 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1503 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1504 num_moved * sizeof(struct ocfs2_refcount_rec));
1506 /*ok, remove the entries we just moved over to the other block. */
1507 memset(&rl->rl_recs[split_index], 0,
1508 num_moved * sizeof(struct ocfs2_refcount_rec));
1510 /* change old and new rl_used accordingly. */
1511 le16_add_cpu(&rl->rl_used, -num_moved);
1512 new_rl->rl_used = cpu_to_le16(num_moved);
1514 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1515 sizeof(struct ocfs2_refcount_rec),
1516 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1518 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1519 sizeof(struct ocfs2_refcount_rec),
1520 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1526 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1527 struct ocfs2_caching_info *ci,
1528 struct buffer_head *ref_root_bh,
1529 struct buffer_head *ref_leaf_bh,
1530 struct ocfs2_alloc_context *meta_ac)
1533 u16 suballoc_bit_start;
1534 u32 num_got, new_cpos;
1535 u64 suballoc_loc, blkno;
1536 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1537 struct ocfs2_refcount_block *root_rb =
1538 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1539 struct buffer_head *new_bh = NULL;
1540 struct ocfs2_refcount_block *new_rb;
1541 struct ocfs2_extent_tree ref_et;
1543 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1545 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1546 OCFS2_JOURNAL_ACCESS_WRITE);
1552 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1553 OCFS2_JOURNAL_ACCESS_WRITE);
1559 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1560 &suballoc_bit_start, &num_got,
1567 new_bh = sb_getblk(sb, blkno);
1568 if (new_bh == NULL) {
1573 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1575 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1576 OCFS2_JOURNAL_ACCESS_CREATE);
1582 /* Initialize ocfs2_refcount_block. */
1583 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1584 memset(new_rb, 0, sb->s_blocksize);
1585 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1586 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1587 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1588 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1589 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1590 new_rb->rf_blkno = cpu_to_le64(blkno);
1591 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1592 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1593 new_rb->rf_records.rl_count =
1594 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1595 new_rb->rf_generation = root_rb->rf_generation;
1597 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1603 ocfs2_journal_dirty(handle, ref_leaf_bh);
1604 ocfs2_journal_dirty(handle, new_bh);
1606 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1608 trace_ocfs2_new_leaf_refcount_block(
1609 (unsigned long long)new_bh->b_blocknr, new_cpos);
1611 /* Insert the new leaf block with the specific offset cpos. */
1612 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1622 static int ocfs2_expand_refcount_tree(handle_t *handle,
1623 struct ocfs2_caching_info *ci,
1624 struct buffer_head *ref_root_bh,
1625 struct buffer_head *ref_leaf_bh,
1626 struct ocfs2_alloc_context *meta_ac)
1629 struct buffer_head *expand_bh = NULL;
1631 if (ref_root_bh == ref_leaf_bh) {
1633 * the old root bh hasn't been expanded to a b-tree,
1634 * so expand it first.
1636 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1637 &expand_bh, meta_ac);
1643 expand_bh = ref_leaf_bh;
1648 /* Now add a new refcount block into the tree.*/
1649 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1650 expand_bh, meta_ac);
1659 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1661 * Only called when we have inserted a new refcount rec at index 0
1662 * which means ocfs2_extent_rec.e_cpos may need some change.
1664 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1665 struct ocfs2_caching_info *ci,
1666 struct buffer_head *ref_root_bh,
1667 struct buffer_head *ref_leaf_bh,
1668 struct ocfs2_refcount_rec *rec)
1671 u32 new_cpos, old_cpos;
1672 struct ocfs2_path *path = NULL;
1673 struct ocfs2_extent_tree et;
1674 struct ocfs2_refcount_block *rb =
1675 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1676 struct ocfs2_extent_list *el;
1678 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1681 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1682 old_cpos = le32_to_cpu(rb->rf_cpos);
1683 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1684 if (old_cpos <= new_cpos)
1687 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1689 path = ocfs2_new_path_from_et(&et);
1696 ret = ocfs2_find_path(ci, path, old_cpos);
1703 * 2 more credits, one for the leaf refcount block, one for
1704 * the extent block contains the extent rec.
1706 ret = ocfs2_extend_trans(handle, 2);
1712 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1713 OCFS2_JOURNAL_ACCESS_WRITE);
1719 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1720 OCFS2_JOURNAL_ACCESS_WRITE);
1726 /* change the leaf extent block first. */
1727 el = path_leaf_el(path);
1729 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1730 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1733 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1735 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1737 /* change the r_cpos in the leaf block. */
1738 rb->rf_cpos = cpu_to_le32(new_cpos);
1740 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1741 ocfs2_journal_dirty(handle, ref_leaf_bh);
1744 ocfs2_free_path(path);
1748 static int ocfs2_insert_refcount_rec(handle_t *handle,
1749 struct ocfs2_caching_info *ci,
1750 struct buffer_head *ref_root_bh,
1751 struct buffer_head *ref_leaf_bh,
1752 struct ocfs2_refcount_rec *rec,
1753 int index, int merge,
1754 struct ocfs2_alloc_context *meta_ac)
1757 struct ocfs2_refcount_block *rb =
1758 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1759 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1760 struct buffer_head *new_bh = NULL;
1762 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1764 if (rf_list->rl_used == rf_list->rl_count) {
1765 u64 cpos = le64_to_cpu(rec->r_cpos);
1766 u32 len = le32_to_cpu(rec->r_clusters);
1768 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1769 ref_leaf_bh, meta_ac);
1775 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1776 cpos, len, NULL, &index,
1783 ref_leaf_bh = new_bh;
1784 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1785 rf_list = &rb->rf_records;
1788 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1789 OCFS2_JOURNAL_ACCESS_WRITE);
1795 if (index < le16_to_cpu(rf_list->rl_used))
1796 memmove(&rf_list->rl_recs[index + 1],
1797 &rf_list->rl_recs[index],
1798 (le16_to_cpu(rf_list->rl_used) - index) *
1799 sizeof(struct ocfs2_refcount_rec));
1801 trace_ocfs2_insert_refcount_rec(
1802 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1803 (unsigned long long)le64_to_cpu(rec->r_cpos),
1804 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1806 rf_list->rl_recs[index] = *rec;
1808 le16_add_cpu(&rf_list->rl_used, 1);
1811 ocfs2_refcount_rec_merge(rb, index);
1813 ocfs2_journal_dirty(handle, ref_leaf_bh);
1816 ret = ocfs2_adjust_refcount_rec(handle, ci,
1828 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1829 * This is much simple than our b-tree code.
1830 * split_rec is the new refcount rec we want to insert.
1831 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1832 * increase refcount or decrease a refcount to non-zero).
1833 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1834 * rec( in case we decrease a refcount to zero).
1836 static int ocfs2_split_refcount_rec(handle_t *handle,
1837 struct ocfs2_caching_info *ci,
1838 struct buffer_head *ref_root_bh,
1839 struct buffer_head *ref_leaf_bh,
1840 struct ocfs2_refcount_rec *split_rec,
1841 int index, int merge,
1842 struct ocfs2_alloc_context *meta_ac,
1843 struct ocfs2_cached_dealloc_ctxt *dealloc)
1847 struct ocfs2_refcount_block *rb =
1848 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1849 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1850 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1851 struct ocfs2_refcount_rec *tail_rec = NULL;
1852 struct buffer_head *new_bh = NULL;
1854 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1856 trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1857 le32_to_cpu(orig_rec->r_clusters),
1858 le32_to_cpu(orig_rec->r_refcount),
1859 le64_to_cpu(split_rec->r_cpos),
1860 le32_to_cpu(split_rec->r_clusters),
1861 le32_to_cpu(split_rec->r_refcount));
1864 * If we just need to split the header or tail clusters,
1865 * no more recs are needed, just split is OK.
1866 * Otherwise we at least need one new recs.
1868 if (!split_rec->r_refcount &&
1869 (split_rec->r_cpos == orig_rec->r_cpos ||
1870 le64_to_cpu(split_rec->r_cpos) +
1871 le32_to_cpu(split_rec->r_clusters) ==
1872 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1878 * We need one more rec if we split in the middle and the new rec have
1879 * some refcount in it.
1881 if (split_rec->r_refcount &&
1882 (split_rec->r_cpos != orig_rec->r_cpos &&
1883 le64_to_cpu(split_rec->r_cpos) +
1884 le32_to_cpu(split_rec->r_clusters) !=
1885 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1888 /* If the leaf block don't have enough record, expand it. */
1889 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1890 le16_to_cpu(rf_list->rl_count)) {
1891 struct ocfs2_refcount_rec tmp_rec;
1892 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1893 len = le32_to_cpu(orig_rec->r_clusters);
1894 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1895 ref_leaf_bh, meta_ac);
1902 * We have to re-get it since now cpos may be moved to
1903 * another leaf block.
1905 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1906 cpos, len, &tmp_rec, &index,
1913 ref_leaf_bh = new_bh;
1914 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1915 rf_list = &rb->rf_records;
1916 orig_rec = &rf_list->rl_recs[index];
1919 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1920 OCFS2_JOURNAL_ACCESS_WRITE);
1927 * We have calculated out how many new records we need and store
1928 * in recs_need, so spare enough space first by moving the records
1929 * after "index" to the end.
1931 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1932 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1933 &rf_list->rl_recs[index + 1],
1934 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1935 sizeof(struct ocfs2_refcount_rec));
1937 len = (le64_to_cpu(orig_rec->r_cpos) +
1938 le32_to_cpu(orig_rec->r_clusters)) -
1939 (le64_to_cpu(split_rec->r_cpos) +
1940 le32_to_cpu(split_rec->r_clusters));
1943 * If we have "len", the we will split in the tail and move it
1944 * to the end of the space we have just spared.
1947 tail_rec = &rf_list->rl_recs[index + recs_need];
1949 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1950 le64_add_cpu(&tail_rec->r_cpos,
1951 le32_to_cpu(tail_rec->r_clusters) - len);
1952 tail_rec->r_clusters = cpu_to_le32(len);
1956 * If the split pos isn't the same as the original one, we need to
1957 * split in the head.
1959 * Note: We have the chance that split_rec.r_refcount = 0,
1960 * recs_need = 0 and len > 0, which means we just cut the head from
1961 * the orig_rec and in that case we have done some modification in
1962 * orig_rec above, so the check for r_cpos is faked.
1964 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1965 len = le64_to_cpu(split_rec->r_cpos) -
1966 le64_to_cpu(orig_rec->r_cpos);
1967 orig_rec->r_clusters = cpu_to_le32(len);
1971 le16_add_cpu(&rf_list->rl_used, recs_need);
1973 if (split_rec->r_refcount) {
1974 rf_list->rl_recs[index] = *split_rec;
1975 trace_ocfs2_split_refcount_rec_insert(
1976 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1977 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1978 le32_to_cpu(split_rec->r_clusters),
1979 le32_to_cpu(split_rec->r_refcount));
1982 ocfs2_refcount_rec_merge(rb, index);
1985 ocfs2_journal_dirty(handle, ref_leaf_bh);
1992 static int __ocfs2_increase_refcount(handle_t *handle,
1993 struct ocfs2_caching_info *ci,
1994 struct buffer_head *ref_root_bh,
1995 u64 cpos, u32 len, int merge,
1996 struct ocfs2_alloc_context *meta_ac,
1997 struct ocfs2_cached_dealloc_ctxt *dealloc)
2000 struct buffer_head *ref_leaf_bh = NULL;
2001 struct ocfs2_refcount_rec rec;
2002 unsigned int set_len = 0;
2004 trace_ocfs2_increase_refcount_begin(
2005 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2006 (unsigned long long)cpos, len);
2009 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2010 cpos, len, &rec, &index,
2017 set_len = le32_to_cpu(rec.r_clusters);
2020 * Here we may meet with 3 situations:
2022 * 1. If we find an already existing record, and the length
2023 * is the same, cool, we just need to increase the r_refcount
2025 * 2. If we find a hole, just insert it with r_refcount = 1.
2026 * 3. If we are in the middle of one extent record, split
2029 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2031 trace_ocfs2_increase_refcount_change(
2032 (unsigned long long)cpos, set_len,
2033 le32_to_cpu(rec.r_refcount));
2034 ret = ocfs2_change_refcount_rec(handle, ci,
2041 } else if (!rec.r_refcount) {
2042 rec.r_refcount = cpu_to_le32(1);
2044 trace_ocfs2_increase_refcount_insert(
2045 (unsigned long long)le64_to_cpu(rec.r_cpos),
2047 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2056 set_len = min((u64)(cpos + len),
2057 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2058 rec.r_cpos = cpu_to_le64(cpos);
2059 rec.r_clusters = cpu_to_le32(set_len);
2060 le32_add_cpu(&rec.r_refcount, 1);
2062 trace_ocfs2_increase_refcount_split(
2063 (unsigned long long)le64_to_cpu(rec.r_cpos),
2064 set_len, le32_to_cpu(rec.r_refcount));
2065 ret = ocfs2_split_refcount_rec(handle, ci,
2066 ref_root_bh, ref_leaf_bh,
2077 brelse(ref_leaf_bh);
2082 brelse(ref_leaf_bh);
2086 static int ocfs2_remove_refcount_extent(handle_t *handle,
2087 struct ocfs2_caching_info *ci,
2088 struct buffer_head *ref_root_bh,
2089 struct buffer_head *ref_leaf_bh,
2090 struct ocfs2_alloc_context *meta_ac,
2091 struct ocfs2_cached_dealloc_ctxt *dealloc)
2094 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2095 struct ocfs2_refcount_block *rb =
2096 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2097 struct ocfs2_extent_tree et;
2099 BUG_ON(rb->rf_records.rl_used);
2101 trace_ocfs2_remove_refcount_extent(
2102 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2103 (unsigned long long)ref_leaf_bh->b_blocknr,
2104 le32_to_cpu(rb->rf_cpos));
2106 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2107 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2108 1, meta_ac, dealloc);
2114 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2117 * add the freed block to the dealloc so that it will be freed
2118 * when we run dealloc.
2120 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2121 le16_to_cpu(rb->rf_suballoc_slot),
2122 le64_to_cpu(rb->rf_suballoc_loc),
2123 le64_to_cpu(rb->rf_blkno),
2124 le16_to_cpu(rb->rf_suballoc_bit));
2130 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2131 OCFS2_JOURNAL_ACCESS_WRITE);
2137 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2139 le32_add_cpu(&rb->rf_clusters, -1);
2142 * check whether we need to restore the root refcount block if
2143 * there is no leaf extent block at atll.
2145 if (!rb->rf_list.l_next_free_rec) {
2146 BUG_ON(rb->rf_clusters);
2148 trace_ocfs2_restore_refcount_block(
2149 (unsigned long long)ref_root_bh->b_blocknr);
2154 memset(&rb->rf_records, 0, sb->s_blocksize -
2155 offsetof(struct ocfs2_refcount_block, rf_records));
2156 rb->rf_records.rl_count =
2157 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2160 ocfs2_journal_dirty(handle, ref_root_bh);
2166 int ocfs2_increase_refcount(handle_t *handle,
2167 struct ocfs2_caching_info *ci,
2168 struct buffer_head *ref_root_bh,
2170 struct ocfs2_alloc_context *meta_ac,
2171 struct ocfs2_cached_dealloc_ctxt *dealloc)
2173 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2178 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2179 struct ocfs2_caching_info *ci,
2180 struct buffer_head *ref_root_bh,
2181 struct buffer_head *ref_leaf_bh,
2182 int index, u64 cpos, unsigned int len,
2183 struct ocfs2_alloc_context *meta_ac,
2184 struct ocfs2_cached_dealloc_ctxt *dealloc)
2187 struct ocfs2_refcount_block *rb =
2188 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2189 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2191 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2193 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2195 trace_ocfs2_decrease_refcount_rec(
2196 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2197 (unsigned long long)cpos, len);
2199 if (cpos == le64_to_cpu(rec->r_cpos) &&
2200 len == le32_to_cpu(rec->r_clusters))
2201 ret = ocfs2_change_refcount_rec(handle, ci,
2202 ref_leaf_bh, index, 1, -1);
2204 struct ocfs2_refcount_rec split = *rec;
2205 split.r_cpos = cpu_to_le64(cpos);
2206 split.r_clusters = cpu_to_le32(len);
2208 le32_add_cpu(&split.r_refcount, -1);
2210 ret = ocfs2_split_refcount_rec(handle, ci,
2211 ref_root_bh, ref_leaf_bh,
2221 /* Remove the leaf refcount block if it contains no refcount record. */
2222 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2223 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2224 ref_leaf_bh, meta_ac,
2234 static int __ocfs2_decrease_refcount(handle_t *handle,
2235 struct ocfs2_caching_info *ci,
2236 struct buffer_head *ref_root_bh,
2238 struct ocfs2_alloc_context *meta_ac,
2239 struct ocfs2_cached_dealloc_ctxt *dealloc,
2242 int ret = 0, index = 0;
2243 struct ocfs2_refcount_rec rec;
2244 unsigned int r_count = 0, r_len;
2245 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2246 struct buffer_head *ref_leaf_bh = NULL;
2248 trace_ocfs2_decrease_refcount(
2249 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2250 (unsigned long long)cpos, len, delete);
2253 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2254 cpos, len, &rec, &index,
2261 r_count = le32_to_cpu(rec.r_refcount);
2262 BUG_ON(r_count == 0);
2264 BUG_ON(r_count > 1);
2266 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2267 le32_to_cpu(rec.r_clusters)) - cpos;
2269 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2278 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2279 ret = ocfs2_cache_cluster_dealloc(dealloc,
2280 ocfs2_clusters_to_blocks(sb, cpos),
2290 brelse(ref_leaf_bh);
2295 brelse(ref_leaf_bh);
2299 /* Caller must hold refcount tree lock. */
2300 int ocfs2_decrease_refcount(struct inode *inode,
2301 handle_t *handle, u32 cpos, u32 len,
2302 struct ocfs2_alloc_context *meta_ac,
2303 struct ocfs2_cached_dealloc_ctxt *dealloc,
2308 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2309 struct buffer_head *ref_root_bh = NULL;
2310 struct ocfs2_refcount_tree *tree;
2312 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2314 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2320 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2326 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2333 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2334 cpos, len, meta_ac, dealloc, delete);
2338 brelse(ref_root_bh);
2343 * Mark the already-existing extent at cpos as refcounted for len clusters.
2344 * This adds the refcount extent flag.
2346 * If the existing extent is larger than the request, initiate a
2347 * split. An attempt will be made at merging with adjacent extents.
2349 * The caller is responsible for passing down meta_ac if we'll need it.
2351 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2352 struct ocfs2_extent_tree *et,
2353 handle_t *handle, u32 cpos,
2355 struct ocfs2_alloc_context *meta_ac,
2356 struct ocfs2_cached_dealloc_ctxt *dealloc)
2360 trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2363 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2364 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2365 "tree, but the feature bit is not set in the "
2366 "super block.", inode->i_ino);
2371 ret = ocfs2_change_extent_flag(handle, et, cpos,
2372 len, phys, meta_ac, dealloc,
2373 OCFS2_EXT_REFCOUNTED, 0);
2382 * Given some contiguous physical clusters, calculate what we need
2383 * for modifying their refcount.
2385 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2386 struct ocfs2_caching_info *ci,
2387 struct buffer_head *ref_root_bh,
2393 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2394 u64 cpos = start_cpos;
2395 struct ocfs2_refcount_block *rb;
2396 struct ocfs2_refcount_rec rec;
2397 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2401 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2402 cpos, clusters, &rec,
2403 &index, &ref_leaf_bh);
2409 if (ref_leaf_bh != prev_bh) {
2411 * Now we encounter a new leaf block, so calculate
2412 * whether we need to extend the old leaf.
2415 rb = (struct ocfs2_refcount_block *)
2418 if (le16_to_cpu(rb->rf_records.rl_used) +
2420 le16_to_cpu(rb->rf_records.rl_count))
2427 prev_bh = ref_leaf_bh;
2431 trace_ocfs2_calc_refcount_meta_credits_iterate(
2432 recs_add, (unsigned long long)cpos, clusters,
2433 (unsigned long long)le64_to_cpu(rec.r_cpos),
2434 le32_to_cpu(rec.r_clusters),
2435 le32_to_cpu(rec.r_refcount), index);
2437 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2438 le32_to_cpu(rec.r_clusters)) - cpos;
2440 * We record all the records which will be inserted to the
2441 * same refcount block, so that we can tell exactly whether
2442 * we need a new refcount block or not.
2444 * If we will insert a new one, this is easy and only happens
2445 * during adding refcounted flag to the extent, so we don't
2446 * have a chance of spliting. We just need one record.
2448 * If the refcount rec already exists, that would be a little
2449 * complicated. we may have to:
2450 * 1) split at the beginning if the start pos isn't aligned.
2451 * we need 1 more record in this case.
2452 * 2) split int the end if the end pos isn't aligned.
2453 * we need 1 more record in this case.
2454 * 3) split in the middle because of file system fragmentation.
2455 * we need 2 more records in this case(we can't detect this
2456 * beforehand, so always think of the worst case).
2458 if (rec.r_refcount) {
2460 /* Check whether we need a split at the beginning. */
2461 if (cpos == start_cpos &&
2462 cpos != le64_to_cpu(rec.r_cpos))
2465 /* Check whether we need a split in the end. */
2466 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2467 le32_to_cpu(rec.r_clusters))
2472 brelse(ref_leaf_bh);
2479 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2481 if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2482 le16_to_cpu(rb->rf_records.rl_count))
2491 *meta_add += ref_blocks;
2492 *credits += ref_blocks;
2495 * So we may need ref_blocks to insert into the tree.
2496 * That also means we need to change the b-tree and add that number
2497 * of records since we never merge them.
2498 * We need one more block for expansion since the new created leaf
2499 * block is also full and needs split.
2501 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2502 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2503 struct ocfs2_extent_tree et;
2505 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2506 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2507 *credits += ocfs2_calc_extend_credits(sb,
2510 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2516 trace_ocfs2_calc_refcount_meta_credits(
2517 (unsigned long long)start_cpos, clusters,
2518 *meta_add, *credits);
2519 brelse(ref_leaf_bh);
2525 * For refcount tree, we will decrease some contiguous clusters
2526 * refcount count, so just go through it to see how many blocks
2527 * we gonna touch and whether we need to create new blocks.
2529 * Normally the refcount blocks store these refcount should be
2530 * contiguous also, so that we can get the number easily.
2531 * We will at most add split 2 refcount records and 2 more
2532 * refcount blocks, so just check it in a rough way.
2534 * Caller must hold refcount tree lock.
2536 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2544 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2545 struct buffer_head *ref_root_bh = NULL;
2546 struct ocfs2_refcount_tree *tree;
2547 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2549 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2550 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2551 "tree, but the feature bit is not set in the "
2552 "super block.", inode->i_ino);
2557 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2559 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2560 refcount_loc, &tree);
2566 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2573 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2576 start_cpos, clusters,
2577 ref_blocks, credits);
2583 trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2586 brelse(ref_root_bh);
2590 #define MAX_CONTIG_BYTES 1048576
2592 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2594 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2597 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2599 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2603 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2604 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2605 * while still being less than or equal to it.
2607 * The goal is to break the extent at a multiple of contig_clusters.
2609 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2613 BUG_ON(start > cpos);
2615 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2619 * Given a cluster count of len, pad it out so that it is a multiple
2620 * of contig_clusters.
2622 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2625 unsigned int padded =
2626 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2627 ocfs2_cow_contig_mask(sb);
2637 * Calculate out the start and number of virtual clusters we need to to CoW.
2639 * cpos is vitual start cluster position we want to do CoW in a
2640 * file and write_len is the cluster length.
2641 * max_cpos is the place where we want to stop CoW intentionally.
2643 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2644 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2645 * get good I/O from the resulting extent tree.
2647 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2648 struct ocfs2_extent_list *el,
2656 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2657 struct buffer_head *eb_bh = NULL;
2658 struct ocfs2_extent_block *eb = NULL;
2659 struct ocfs2_extent_rec *rec;
2660 unsigned int want_clusters, rec_end = 0;
2661 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2664 BUG_ON(cpos + write_len > max_cpos);
2666 if (tree_height > 0) {
2667 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2673 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2676 if (el->l_tree_depth) {
2677 ocfs2_error(inode->i_sb,
2678 "Inode %lu has non zero tree depth in "
2679 "leaf block %llu\n", inode->i_ino,
2680 (unsigned long long)eb_bh->b_blocknr);
2687 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2688 rec = &el->l_recs[i];
2690 if (ocfs2_is_empty_extent(rec)) {
2691 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2692 "index %d\n", inode->i_ino, i);
2696 if (le32_to_cpu(rec->e_cpos) +
2697 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2700 if (*cow_len == 0) {
2702 * We should find a refcounted record in the
2705 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2706 *cow_start = le32_to_cpu(rec->e_cpos);
2710 * If we encounter a hole, a non-refcounted record or
2711 * pass the max_cpos, stop the search.
2713 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2714 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2715 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2718 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2719 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2720 if (rec_end > max_cpos) {
2722 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2726 * How many clusters do we actually need from
2727 * this extent? First we see how many we actually
2728 * need to complete the write. If that's smaller
2729 * than contig_clusters, we try for contig_clusters.
2732 want_clusters = write_len;
2734 want_clusters = (cpos + write_len) -
2735 (*cow_start + *cow_len);
2736 if (want_clusters < contig_clusters)
2737 want_clusters = contig_clusters;
2740 * If the write does not cover the whole extent, we
2741 * need to calculate how we're going to split the extent.
2742 * We try to do it on contig_clusters boundaries.
2744 * Any extent smaller than contig_clusters will be
2745 * CoWed in its entirety.
2747 if (leaf_clusters <= contig_clusters)
2748 *cow_len += leaf_clusters;
2749 else if (*cow_len || (*cow_start == cpos)) {
2751 * This extent needs to be CoW'd from its
2752 * beginning, so all we have to do is compute
2753 * how many clusters to grab. We align
2754 * want_clusters to the edge of contig_clusters
2755 * to get better I/O.
2757 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2760 if (leaf_clusters < want_clusters)
2761 *cow_len += leaf_clusters;
2763 *cow_len += want_clusters;
2764 } else if ((*cow_start + contig_clusters) >=
2765 (cpos + write_len)) {
2767 * Breaking off contig_clusters at the front
2768 * of the extent will cover our write. That's
2771 *cow_len = contig_clusters;
2772 } else if ((rec_end - cpos) <= contig_clusters) {
2774 * Breaking off contig_clusters at the tail of
2775 * this extent will cover cpos.
2777 *cow_start = rec_end - contig_clusters;
2778 *cow_len = contig_clusters;
2779 } else if ((rec_end - cpos) <= want_clusters) {
2781 * While we can't fit the entire write in this
2782 * extent, we know that the write goes from cpos
2783 * to the end of the extent. Break that off.
2784 * We try to break it at some multiple of
2785 * contig_clusters from the front of the extent.
2786 * Failing that (ie, cpos is within
2787 * contig_clusters of the front), we'll CoW the
2790 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2792 *cow_len = rec_end - *cow_start;
2795 * Ok, the entire write lives in the middle of
2796 * this extent. Let's try to slice the extent up
2797 * nicely. Optimally, our CoW region starts at
2798 * m*contig_clusters from the beginning of the
2799 * extent and goes for n*contig_clusters,
2800 * covering the entire write.
2802 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2805 want_clusters = (cpos + write_len) - *cow_start;
2806 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2808 if (*cow_start + want_clusters <= rec_end)
2809 *cow_len = want_clusters;
2811 *cow_len = rec_end - *cow_start;
2814 /* Have we covered our entire write yet? */
2815 if ((*cow_start + *cow_len) >= (cpos + write_len))
2819 * If we reach the end of the extent block and don't get enough
2820 * clusters, continue with the next extent block if possible.
2822 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2823 eb && eb->h_next_leaf_blk) {
2827 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2828 le64_to_cpu(eb->h_next_leaf_blk),
2835 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2847 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2848 * num_clusters in data_tree "et" and change the refcount for the old
2849 * clusters(starting form p_cluster) in the refcount tree.
2852 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2853 * more new leaf records.
2854 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2855 * just give data_ac = NULL.
2857 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2858 u32 p_cluster, u32 num_clusters,
2859 struct ocfs2_extent_tree *et,
2860 struct ocfs2_caching_info *ref_ci,
2861 struct buffer_head *ref_root_bh,
2862 struct ocfs2_alloc_context **meta_ac,
2863 struct ocfs2_alloc_context **data_ac,
2866 int ret = 0, meta_add = 0;
2867 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2869 if (num_free_extents < 0) {
2870 ret = num_free_extents;
2875 if (num_free_extents < num_clusters + 2)
2877 ocfs2_extend_meta_needed(et->et_root_el);
2879 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2881 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2882 p_cluster, num_clusters,
2883 &meta_add, credits);
2889 trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2890 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2898 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2907 ocfs2_free_alloc_context(*meta_ac);
2915 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2917 BUG_ON(buffer_dirty(bh));
2919 clear_buffer_mapped(bh);
2924 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2925 struct inode *inode,
2926 u32 cpos, u32 old_cluster,
2927 u32 new_cluster, u32 new_len)
2929 int ret = 0, partial;
2930 struct super_block *sb = inode->i_sb;
2931 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2934 unsigned int from, to, readahead_pages;
2935 loff_t offset, end, map_end;
2936 struct address_space *mapping = inode->i_mapping;
2938 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2939 new_cluster, new_len);
2942 (ocfs2_cow_contig_clusters(sb) <<
2943 OCFS2_SB(sb)->s_clustersize_bits) >> PAGE_CACHE_SHIFT;
2944 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2945 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2947 * We only duplicate pages until we reach the page contains i_size - 1.
2948 * So trim 'end' to i_size.
2950 if (end > i_size_read(inode))
2951 end = i_size_read(inode);
2953 while (offset < end) {
2954 page_index = offset >> PAGE_CACHE_SHIFT;
2955 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2959 /* from, to is the offset within the page. */
2960 from = offset & (PAGE_CACHE_SIZE - 1);
2961 to = PAGE_CACHE_SIZE;
2962 if (map_end & (PAGE_CACHE_SIZE - 1))
2963 to = map_end & (PAGE_CACHE_SIZE - 1);
2965 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2973 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2974 * can't be dirtied before we CoW it out.
2976 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2977 BUG_ON(PageDirty(page));
2979 if (!PageUptodate(page)) {
2980 ret = block_read_full_page(page, ocfs2_get_block);
2988 if (page_has_buffers(page)) {
2989 ret = walk_page_buffers(handle, page_buffers(page),
2991 ocfs2_clear_cow_buffer);
2998 ocfs2_map_and_dirty_page(inode,
3000 page, 0, &new_block);
3001 mark_page_accessed(page);
3004 page_cache_release(page);
3014 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3015 struct inode *inode,
3016 u32 cpos, u32 old_cluster,
3017 u32 new_cluster, u32 new_len)
3020 struct super_block *sb = inode->i_sb;
3021 struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3022 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3023 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3024 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3025 struct ocfs2_super *osb = OCFS2_SB(sb);
3026 struct buffer_head *old_bh = NULL;
3027 struct buffer_head *new_bh = NULL;
3029 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3030 new_cluster, new_len);
3032 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3033 new_bh = sb_getblk(osb->sb, new_block);
3034 if (new_bh == NULL) {
3040 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3042 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3048 ret = ocfs2_journal_access(handle, ci, new_bh,
3049 OCFS2_JOURNAL_ACCESS_CREATE);
3055 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3056 ocfs2_journal_dirty(handle, new_bh);
3069 static int ocfs2_clear_ext_refcount(handle_t *handle,
3070 struct ocfs2_extent_tree *et,
3071 u32 cpos, u32 p_cluster, u32 len,
3072 unsigned int ext_flags,
3073 struct ocfs2_alloc_context *meta_ac,
3074 struct ocfs2_cached_dealloc_ctxt *dealloc)
3077 struct ocfs2_extent_rec replace_rec;
3078 struct ocfs2_path *path = NULL;
3079 struct ocfs2_extent_list *el;
3080 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3081 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3083 trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3084 cpos, len, p_cluster, ext_flags);
3086 memset(&replace_rec, 0, sizeof(replace_rec));
3087 replace_rec.e_cpos = cpu_to_le32(cpos);
3088 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3089 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3091 replace_rec.e_flags = ext_flags;
3092 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3094 path = ocfs2_new_path_from_et(et);
3101 ret = ocfs2_find_path(et->et_ci, path, cpos);
3107 el = path_leaf_el(path);
3109 index = ocfs2_search_extent_list(el, cpos);
3112 "Inode %llu has an extent at cpos %u which can no "
3113 "longer be found.\n",
3114 (unsigned long long)ino, cpos);
3119 ret = ocfs2_split_extent(handle, et, path, index,
3120 &replace_rec, meta_ac, dealloc);
3125 ocfs2_free_path(path);
3129 static int ocfs2_replace_clusters(handle_t *handle,
3130 struct ocfs2_cow_context *context,
3133 unsigned int ext_flags)
3136 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3137 u64 ino = ocfs2_metadata_cache_owner(ci);
3139 trace_ocfs2_replace_clusters((unsigned long long)ino,
3140 cpos, old, new, len, ext_flags);
3142 /*If the old clusters is unwritten, no need to duplicate. */
3143 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3144 ret = context->cow_duplicate_clusters(handle, context->inode,
3145 cpos, old, new, len);
3152 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3153 cpos, new, len, ext_flags,
3154 context->meta_ac, &context->dealloc);
3161 int ocfs2_cow_sync_writeback(struct super_block *sb,
3162 struct inode *inode,
3163 u32 cpos, u32 num_clusters)
3166 loff_t offset, end, map_end;
3170 if (ocfs2_should_order_data(inode))
3173 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3174 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3176 ret = filemap_fdatawrite_range(inode->i_mapping,
3183 while (offset < end) {
3184 page_index = offset >> PAGE_CACHE_SHIFT;
3185 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3189 page = find_or_create_page(inode->i_mapping,
3190 page_index, GFP_NOFS);
3193 wait_on_page_writeback(page);
3194 if (PageError(page)) {
3198 mark_page_accessed(page);
3201 page_cache_release(page);
3211 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3212 u32 v_cluster, u32 *p_cluster,
3214 unsigned int *extent_flags)
3216 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3217 num_clusters, extent_flags);
3220 static int ocfs2_make_clusters_writable(struct super_block *sb,
3221 struct ocfs2_cow_context *context,
3222 u32 cpos, u32 p_cluster,
3223 u32 num_clusters, unsigned int e_flags)
3225 int ret, delete, index, credits = 0;
3226 u32 new_bit, new_len, orig_num_clusters;
3227 unsigned int set_len;
3228 struct ocfs2_super *osb = OCFS2_SB(sb);
3230 struct buffer_head *ref_leaf_bh = NULL;
3231 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3232 struct ocfs2_refcount_rec rec;
3234 trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3235 num_clusters, e_flags);
3237 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3240 context->ref_root_bh,
3242 &context->data_ac, &credits);
3248 if (context->post_refcount)
3249 credits += context->post_refcount->credits;
3251 credits += context->extra_credits;
3252 handle = ocfs2_start_trans(osb, credits);
3253 if (IS_ERR(handle)) {
3254 ret = PTR_ERR(handle);
3259 orig_num_clusters = num_clusters;
3261 while (num_clusters) {
3262 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3263 p_cluster, num_clusters,
3264 &rec, &index, &ref_leaf_bh);
3270 BUG_ON(!rec.r_refcount);
3271 set_len = min((u64)p_cluster + num_clusters,
3272 le64_to_cpu(rec.r_cpos) +
3273 le32_to_cpu(rec.r_clusters)) - p_cluster;
3276 * There are many different situation here.
3277 * 1. If refcount == 1, remove the flag and don't COW.
3278 * 2. If refcount > 1, allocate clusters.
3279 * Here we may not allocate r_len once at a time, so continue
3280 * until we reach num_clusters.
3282 if (le32_to_cpu(rec.r_refcount) == 1) {
3284 ret = ocfs2_clear_ext_refcount(handle,
3297 ret = __ocfs2_claim_clusters(handle,
3300 &new_bit, &new_len);
3306 ret = ocfs2_replace_clusters(handle, context,
3307 cpos, p_cluster, new_bit,
3316 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3317 context->ref_root_bh,
3320 &context->dealloc, delete);
3327 p_cluster += set_len;
3328 num_clusters -= set_len;
3329 brelse(ref_leaf_bh);
3333 /* handle any post_cow action. */
3334 if (context->post_refcount && context->post_refcount->func) {
3335 ret = context->post_refcount->func(context->inode, handle,
3336 context->post_refcount->para);
3344 * Here we should write the new page out first if we are
3345 * in write-back mode.
3347 if (context->get_clusters == ocfs2_di_get_clusters) {
3348 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3355 ocfs2_commit_trans(osb, handle);
3358 if (context->data_ac) {
3359 ocfs2_free_alloc_context(context->data_ac);
3360 context->data_ac = NULL;
3362 if (context->meta_ac) {
3363 ocfs2_free_alloc_context(context->meta_ac);
3364 context->meta_ac = NULL;
3366 brelse(ref_leaf_bh);
3371 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3374 struct inode *inode = context->inode;
3375 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3376 u32 p_cluster, num_clusters;
3377 unsigned int ext_flags;
3378 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3380 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3381 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3382 "tree, but the feature bit is not set in the "
3383 "super block.", inode->i_ino);
3387 ocfs2_init_dealloc_ctxt(&context->dealloc);
3390 ret = context->get_clusters(context, cow_start, &p_cluster,
3391 &num_clusters, &ext_flags);
3397 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3399 if (cow_len < num_clusters)
3400 num_clusters = cow_len;
3402 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3403 cow_start, p_cluster,
3404 num_clusters, ext_flags);
3410 cow_len -= num_clusters;
3411 cow_start += num_clusters;
3414 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3415 ocfs2_schedule_truncate_log_flush(osb, 1);
3416 ocfs2_run_deallocs(osb, &context->dealloc);
3423 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3424 * past max_cpos. This will stop when it runs into a hole or an
3425 * unrefcounted extent.
3427 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3428 struct buffer_head *di_bh,
3429 u32 cpos, u32 write_len, u32 max_cpos)
3432 u32 cow_start = 0, cow_len = 0;
3433 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3434 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3435 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3436 struct buffer_head *ref_root_bh = NULL;
3437 struct ocfs2_refcount_tree *ref_tree;
3438 struct ocfs2_cow_context *context = NULL;
3440 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3442 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3443 cpos, write_len, max_cpos,
3444 &cow_start, &cow_len);
3450 trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3451 cpos, write_len, max_cpos,
3452 cow_start, cow_len);
3454 BUG_ON(cow_len == 0);
3456 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3463 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3464 1, &ref_tree, &ref_root_bh);
3470 context->inode = inode;
3471 context->cow_start = cow_start;
3472 context->cow_len = cow_len;
3473 context->ref_tree = ref_tree;
3474 context->ref_root_bh = ref_root_bh;
3475 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3476 context->get_clusters = ocfs2_di_get_clusters;
3478 ocfs2_init_dinode_extent_tree(&context->data_et,
3479 INODE_CACHE(inode), di_bh);
3481 ret = ocfs2_replace_cow(context);
3486 * truncate the extent map here since no matter whether we meet with
3487 * any error during the action, we shouldn't trust cached extent map
3490 ocfs2_extent_map_trunc(inode, cow_start);
3492 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3493 brelse(ref_root_bh);
3500 * CoW any and all clusters between cpos and cpos+write_len.
3501 * Don't CoW past max_cpos. If this returns successfully, all
3502 * clusters between cpos and cpos+write_len are safe to modify.
3504 int ocfs2_refcount_cow(struct inode *inode,
3505 struct buffer_head *di_bh,
3506 u32 cpos, u32 write_len, u32 max_cpos)
3509 u32 p_cluster, num_clusters;
3510 unsigned int ext_flags;
3513 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3514 &num_clusters, &ext_flags);
3520 if (write_len < num_clusters)
3521 num_clusters = write_len;
3523 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3524 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3525 num_clusters, max_cpos);
3532 write_len -= num_clusters;
3533 cpos += num_clusters;
3539 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3540 u32 v_cluster, u32 *p_cluster,
3542 unsigned int *extent_flags)
3544 struct inode *inode = context->inode;
3545 struct ocfs2_xattr_value_root *xv = context->cow_object;
3547 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3548 num_clusters, &xv->xr_list,
3553 * Given a xattr value root, calculate the most meta/credits we need for
3554 * refcount tree change if we truncate it to 0.
3556 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3557 struct ocfs2_caching_info *ref_ci,
3558 struct buffer_head *ref_root_bh,
3559 struct ocfs2_xattr_value_root *xv,
3560 int *meta_add, int *credits)
3562 int ret = 0, index, ref_blocks = 0;
3563 u32 p_cluster, num_clusters;
3564 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3565 struct ocfs2_refcount_block *rb;
3566 struct ocfs2_refcount_rec rec;
3567 struct buffer_head *ref_leaf_bh = NULL;
3569 while (cpos < clusters) {
3570 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3571 &num_clusters, &xv->xr_list,
3578 cpos += num_clusters;
3580 while (num_clusters) {
3581 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3582 p_cluster, num_clusters,
3590 BUG_ON(!rec.r_refcount);
3592 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3595 * We really don't know whether the other clusters is in
3596 * this refcount block or not, so just take the worst
3597 * case that all the clusters are in this block and each
3598 * one will split a refcount rec, so totally we need
3599 * clusters * 2 new refcount rec.
3601 if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3602 le16_to_cpu(rb->rf_records.rl_count))
3606 brelse(ref_leaf_bh);
3609 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3612 num_clusters -= le32_to_cpu(rec.r_clusters);
3613 p_cluster += num_clusters;
3617 *meta_add += ref_blocks;
3621 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3622 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3623 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3625 struct ocfs2_extent_tree et;
3627 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3628 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3633 brelse(ref_leaf_bh);
3640 int ocfs2_refcount_cow_xattr(struct inode *inode,
3641 struct ocfs2_dinode *di,
3642 struct ocfs2_xattr_value_buf *vb,
3643 struct ocfs2_refcount_tree *ref_tree,
3644 struct buffer_head *ref_root_bh,
3645 u32 cpos, u32 write_len,
3646 struct ocfs2_post_refcount *post)
3649 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3650 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3651 struct ocfs2_cow_context *context = NULL;
3652 u32 cow_start, cow_len;
3654 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3656 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3657 cpos, write_len, UINT_MAX,
3658 &cow_start, &cow_len);
3664 BUG_ON(cow_len == 0);
3666 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3673 context->inode = inode;
3674 context->cow_start = cow_start;
3675 context->cow_len = cow_len;
3676 context->ref_tree = ref_tree;
3677 context->ref_root_bh = ref_root_bh;
3678 context->cow_object = xv;
3680 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3681 /* We need the extra credits for duplicate_clusters by jbd. */
3682 context->extra_credits =
3683 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3684 context->get_clusters = ocfs2_xattr_value_get_clusters;
3685 context->post_refcount = post;
3687 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3688 INODE_CACHE(inode), vb);
3690 ret = ocfs2_replace_cow(context);
3700 * Insert a new extent into refcount tree and mark a extent rec
3701 * as refcounted in the dinode tree.
3703 int ocfs2_add_refcount_flag(struct inode *inode,
3704 struct ocfs2_extent_tree *data_et,
3705 struct ocfs2_caching_info *ref_ci,
3706 struct buffer_head *ref_root_bh,
3707 u32 cpos, u32 p_cluster, u32 num_clusters,
3708 struct ocfs2_cached_dealloc_ctxt *dealloc,
3709 struct ocfs2_post_refcount *post)
3713 int credits = 1, ref_blocks = 0;
3714 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3715 struct ocfs2_alloc_context *meta_ac = NULL;
3717 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3718 ref_ci, ref_root_bh,
3719 p_cluster, num_clusters,
3720 &ref_blocks, &credits);
3726 trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3729 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3730 ref_blocks, &meta_ac);
3738 credits += post->credits;
3740 handle = ocfs2_start_trans(osb, credits);
3741 if (IS_ERR(handle)) {
3742 ret = PTR_ERR(handle);
3747 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3748 cpos, num_clusters, p_cluster,
3755 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3756 p_cluster, num_clusters, 0,
3763 if (post && post->func) {
3764 ret = post->func(inode, handle, post->para);
3770 ocfs2_commit_trans(osb, handle);
3773 ocfs2_free_alloc_context(meta_ac);
3777 static int ocfs2_change_ctime(struct inode *inode,
3778 struct buffer_head *di_bh)
3782 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3784 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3785 OCFS2_INODE_UPDATE_CREDITS);
3786 if (IS_ERR(handle)) {
3787 ret = PTR_ERR(handle);
3792 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3793 OCFS2_JOURNAL_ACCESS_WRITE);
3799 inode->i_ctime = CURRENT_TIME;
3800 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3801 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3803 ocfs2_journal_dirty(handle, di_bh);
3806 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3811 static int ocfs2_attach_refcount_tree(struct inode *inode,
3812 struct buffer_head *di_bh)
3814 int ret, data_changed = 0;
3815 struct buffer_head *ref_root_bh = NULL;
3816 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3817 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3818 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3819 struct ocfs2_refcount_tree *ref_tree;
3820 unsigned int ext_flags;
3822 u32 cpos, num_clusters, clusters, p_cluster;
3823 struct ocfs2_cached_dealloc_ctxt dealloc;
3824 struct ocfs2_extent_tree di_et;
3826 ocfs2_init_dealloc_ctxt(&dealloc);
3828 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3829 ret = ocfs2_create_refcount_tree(inode, di_bh);
3836 BUG_ON(!di->i_refcount_loc);
3837 ret = ocfs2_lock_refcount_tree(osb,
3838 le64_to_cpu(di->i_refcount_loc), 1,
3839 &ref_tree, &ref_root_bh);
3845 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3848 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3850 size = i_size_read(inode);
3851 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3854 while (cpos < clusters) {
3855 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3856 &num_clusters, &ext_flags);
3861 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3862 ret = ocfs2_add_refcount_flag(inode, &di_et,
3865 p_cluster, num_clusters,
3874 cpos += num_clusters;
3878 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3879 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3890 ret = ocfs2_change_ctime(inode, di_bh);
3896 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3897 brelse(ref_root_bh);
3899 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3900 ocfs2_schedule_truncate_log_flush(osb, 1);
3901 ocfs2_run_deallocs(osb, &dealloc);
3905 * Empty the extent map so that we may get the right extent
3906 * record from the disk.
3908 ocfs2_extent_map_trunc(inode, 0);
3913 static int ocfs2_add_refcounted_extent(struct inode *inode,
3914 struct ocfs2_extent_tree *et,
3915 struct ocfs2_caching_info *ref_ci,
3916 struct buffer_head *ref_root_bh,
3917 u32 cpos, u32 p_cluster, u32 num_clusters,
3918 unsigned int ext_flags,
3919 struct ocfs2_cached_dealloc_ctxt *dealloc)
3924 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3925 struct ocfs2_alloc_context *meta_ac = NULL;
3927 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3928 p_cluster, num_clusters,
3930 ref_root_bh, &meta_ac,
3937 handle = ocfs2_start_trans(osb, credits);
3938 if (IS_ERR(handle)) {
3939 ret = PTR_ERR(handle);
3944 ret = ocfs2_insert_extent(handle, et, cpos,
3945 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3946 num_clusters, ext_flags, meta_ac);
3952 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3953 p_cluster, num_clusters,
3959 ocfs2_commit_trans(osb, handle);
3962 ocfs2_free_alloc_context(meta_ac);
3966 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3967 struct buffer_head *s_bh,
3968 struct inode *t_inode,
3969 struct buffer_head *t_bh)
3973 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3974 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3975 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3977 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3979 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3980 if (IS_ERR(handle)) {
3981 ret = PTR_ERR(handle);
3986 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3987 OCFS2_JOURNAL_ACCESS_WRITE);
3993 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3994 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3995 le16_to_cpu(s_di->id2.i_data.id_count));
3996 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3997 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3998 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3999 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4001 ocfs2_journal_dirty(handle, t_bh);
4004 ocfs2_commit_trans(osb, handle);
4009 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
4010 struct inode *t_inode,
4011 struct buffer_head *t_bh,
4012 struct ocfs2_caching_info *ref_ci,
4013 struct buffer_head *ref_root_bh,
4014 struct ocfs2_cached_dealloc_ctxt *dealloc)
4017 u32 p_cluster, num_clusters, clusters, cpos;
4019 unsigned int ext_flags;
4020 struct ocfs2_extent_tree et;
4022 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4024 size = i_size_read(s_inode);
4025 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4028 while (cpos < clusters) {
4029 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4030 &num_clusters, &ext_flags);
4036 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4037 ref_ci, ref_root_bh,
4048 cpos += num_clusters;
4056 * change the new file's attributes to the src.
4058 * reflink creates a snapshot of a file, that means the attributes
4059 * must be identical except for three exceptions - nlink, ino, and ctime.
4061 static int ocfs2_complete_reflink(struct inode *s_inode,
4062 struct buffer_head *s_bh,
4063 struct inode *t_inode,
4064 struct buffer_head *t_bh,
4069 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4070 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4071 loff_t size = i_size_read(s_inode);
4073 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4074 OCFS2_INODE_UPDATE_CREDITS);
4075 if (IS_ERR(handle)) {
4076 ret = PTR_ERR(handle);
4081 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4082 OCFS2_JOURNAL_ACCESS_WRITE);
4088 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4089 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4090 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4091 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4092 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4093 i_size_write(t_inode, size);
4094 t_inode->i_blocks = s_inode->i_blocks;
4096 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4097 di->i_clusters = s_di->i_clusters;
4098 di->i_size = s_di->i_size;
4099 di->i_dyn_features = s_di->i_dyn_features;
4100 di->i_attr = s_di->i_attr;
4103 t_inode->i_uid = s_inode->i_uid;
4104 t_inode->i_gid = s_inode->i_gid;
4105 t_inode->i_mode = s_inode->i_mode;
4106 di->i_uid = s_di->i_uid;
4107 di->i_gid = s_di->i_gid;
4108 di->i_mode = s_di->i_mode;
4112 * we want mtime to appear identical to the source and
4115 t_inode->i_ctime = CURRENT_TIME;
4117 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4118 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4120 t_inode->i_mtime = s_inode->i_mtime;
4121 di->i_mtime = s_di->i_mtime;
4122 di->i_mtime_nsec = s_di->i_mtime_nsec;
4125 ocfs2_journal_dirty(handle, t_bh);
4128 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4132 static int ocfs2_create_reflink_node(struct inode *s_inode,
4133 struct buffer_head *s_bh,
4134 struct inode *t_inode,
4135 struct buffer_head *t_bh,
4139 struct buffer_head *ref_root_bh = NULL;
4140 struct ocfs2_cached_dealloc_ctxt dealloc;
4141 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4142 struct ocfs2_refcount_block *rb;
4143 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4144 struct ocfs2_refcount_tree *ref_tree;
4146 ocfs2_init_dealloc_ctxt(&dealloc);
4148 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4149 le64_to_cpu(di->i_refcount_loc));
4155 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4156 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4163 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4164 1, &ref_tree, &ref_root_bh);
4169 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4171 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4172 &ref_tree->rf_ci, ref_root_bh,
4176 goto out_unlock_refcount;
4179 out_unlock_refcount:
4180 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4181 brelse(ref_root_bh);
4183 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4184 ocfs2_schedule_truncate_log_flush(osb, 1);
4185 ocfs2_run_deallocs(osb, &dealloc);
4191 static int __ocfs2_reflink(struct dentry *old_dentry,
4192 struct buffer_head *old_bh,
4193 struct inode *new_inode,
4197 struct inode *inode = old_dentry->d_inode;
4198 struct buffer_head *new_bh = NULL;
4200 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4206 ret = filemap_fdatawrite(inode->i_mapping);
4212 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4218 mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4219 ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4220 OI_LS_REFLINK_TARGET);
4226 ret = ocfs2_create_reflink_node(inode, old_bh,
4227 new_inode, new_bh, preserve);
4233 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4234 ret = ocfs2_reflink_xattrs(inode, old_bh,
4243 ret = ocfs2_complete_reflink(inode, old_bh,
4244 new_inode, new_bh, preserve);
4249 ocfs2_inode_unlock(new_inode, 1);
4252 mutex_unlock(&new_inode->i_mutex);
4255 ret = filemap_fdatawait(inode->i_mapping);
4262 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4263 struct dentry *new_dentry, bool preserve)
4266 struct inode *inode = old_dentry->d_inode;
4267 struct buffer_head *old_bh = NULL;
4268 struct inode *new_orphan_inode = NULL;
4269 struct posix_acl *default_acl, *acl;
4272 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4275 mode = inode->i_mode;
4276 error = posix_acl_create(dir, &mode, &default_acl, &acl);
4282 error = ocfs2_create_inode_in_orphan(dir, mode,
4289 error = ocfs2_rw_lock(inode, 1);
4295 error = ocfs2_inode_lock(inode, &old_bh, 1);
4298 ocfs2_rw_unlock(inode, 1);
4302 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4303 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4304 error = __ocfs2_reflink(old_dentry, old_bh,
4305 new_orphan_inode, preserve);
4306 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4307 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4309 ocfs2_inode_unlock(inode, 1);
4310 ocfs2_rw_unlock(inode, 1);
4318 /* If the security isn't preserved, we need to re-initialize them. */
4320 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4321 &new_dentry->d_name,
4328 posix_acl_release(default_acl);
4330 posix_acl_release(acl);
4332 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4338 if (new_orphan_inode) {
4340 * We need to open_unlock the inode no matter whether we
4341 * succeed or not, so that other nodes can delete it later.
4343 ocfs2_open_unlock(new_orphan_inode);
4345 iput(new_orphan_inode);
4352 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4353 * sys_reflink(). This will go away when vfs_reflink() exists in
4357 /* copied from may_create in VFS. */
4358 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4362 if (IS_DEADDIR(dir))
4364 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4368 * ocfs2_vfs_reflink - Create a reference-counted link
4370 * @old_dentry: source dentry + inode
4371 * @dir: directory to create the target
4372 * @new_dentry: target dentry
4373 * @preserve: if true, preserve all file attributes
4375 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4376 struct dentry *new_dentry, bool preserve)
4378 struct inode *inode = old_dentry->d_inode;
4384 error = ocfs2_may_create(dir, new_dentry);
4388 if (dir->i_sb != inode->i_sb)
4392 * A reflink to an append-only or immutable file cannot be created.
4394 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4397 /* Only regular files can be reflinked. */
4398 if (!S_ISREG(inode->i_mode))
4402 * If the caller wants to preserve ownership, they require the
4406 if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4408 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4413 * If the caller is modifying any aspect of the attributes, they
4414 * are not creating a snapshot. They need read permission on the
4418 error = inode_permission(inode, MAY_READ);
4423 mutex_lock(&inode->i_mutex);
4424 dquot_initialize(dir);
4425 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4426 mutex_unlock(&inode->i_mutex);
4428 fsnotify_create(dir, new_dentry);
4432 * Most codes are copied from sys_linkat.
4434 int ocfs2_reflink_ioctl(struct inode *inode,
4435 const char __user *oldname,
4436 const char __user *newname,
4439 struct dentry *new_dentry;
4440 struct path old_path, new_path;
4443 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4446 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4452 new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4453 error = PTR_ERR(new_dentry);
4454 if (IS_ERR(new_dentry)) {
4460 if (old_path.mnt != new_path.mnt) {
4465 error = ocfs2_vfs_reflink(old_path.dentry,
4466 new_path.dentry->d_inode,
4467 new_dentry, preserve);
4469 done_path_create(&new_path, new_dentry);
4471 path_put(&old_path);