1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2009 Oracle. All rights reserved.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public
10 * License version 2 as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
18 #include <linux/sort.h>
19 #define MLOG_MASK_PREFIX ML_REFCOUNT
20 #include <cluster/masklog.h>
28 #include "buffer_head_io.h"
29 #include "blockcheck.h"
30 #include "refcounttree.h"
33 #include "extent_map.h"
38 #include <linux/bio.h>
39 #include <linux/blkdev.h>
40 #include <linux/gfp.h>
41 #include <linux/slab.h>
42 #include <linux/writeback.h>
43 #include <linux/pagevec.h>
44 #include <linux/swap.h>
45 #include <linux/security.h>
46 #include <linux/fsnotify.h>
47 #include <linux/quotaops.h>
48 #include <linux/namei.h>
49 #include <linux/mount.h>
51 struct ocfs2_cow_context {
55 struct ocfs2_extent_tree data_et;
56 struct ocfs2_refcount_tree *ref_tree;
57 struct buffer_head *ref_root_bh;
58 struct ocfs2_alloc_context *meta_ac;
59 struct ocfs2_alloc_context *data_ac;
60 struct ocfs2_cached_dealloc_ctxt dealloc;
62 struct ocfs2_post_refcount *post_refcount;
64 int (*get_clusters)(struct ocfs2_cow_context *context,
65 u32 v_cluster, u32 *p_cluster,
67 unsigned int *extent_flags);
68 int (*cow_duplicate_clusters)(handle_t *handle,
69 struct ocfs2_cow_context *context,
70 u32 cpos, u32 old_cluster,
71 u32 new_cluster, u32 new_len);
74 static inline struct ocfs2_refcount_tree *
75 cache_info_to_refcount(struct ocfs2_caching_info *ci)
77 return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
80 static int ocfs2_validate_refcount_block(struct super_block *sb,
81 struct buffer_head *bh)
84 struct ocfs2_refcount_block *rb =
85 (struct ocfs2_refcount_block *)bh->b_data;
87 mlog(0, "Validating refcount block %llu\n",
88 (unsigned long long)bh->b_blocknr);
90 BUG_ON(!buffer_uptodate(bh));
93 * If the ecc fails, we return the error but otherwise
94 * leave the filesystem running. We know any error is
95 * local to this block.
97 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
99 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
100 (unsigned long long)bh->b_blocknr);
105 if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
107 "Refcount block #%llu has bad signature %.*s",
108 (unsigned long long)bh->b_blocknr, 7,
113 if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
115 "Refcount block #%llu has an invalid rf_blkno "
117 (unsigned long long)bh->b_blocknr,
118 (unsigned long long)le64_to_cpu(rb->rf_blkno));
122 if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
124 "Refcount block #%llu has an invalid "
125 "rf_fs_generation of #%u",
126 (unsigned long long)bh->b_blocknr,
127 le32_to_cpu(rb->rf_fs_generation));
134 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
136 struct buffer_head **bh)
139 struct buffer_head *tmp = *bh;
141 rc = ocfs2_read_block(ci, rb_blkno, &tmp,
142 ocfs2_validate_refcount_block);
144 /* If ocfs2_read_block() got us a new bh, pass it up. */
151 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
153 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
158 static struct super_block *
159 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
161 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
166 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
168 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
170 spin_lock(&rf->rf_lock);
173 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
175 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
177 spin_unlock(&rf->rf_lock);
180 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
182 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
184 mutex_lock(&rf->rf_io_mutex);
187 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
189 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
191 mutex_unlock(&rf->rf_io_mutex);
194 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
195 .co_owner = ocfs2_refcount_cache_owner,
196 .co_get_super = ocfs2_refcount_cache_get_super,
197 .co_cache_lock = ocfs2_refcount_cache_lock,
198 .co_cache_unlock = ocfs2_refcount_cache_unlock,
199 .co_io_lock = ocfs2_refcount_cache_io_lock,
200 .co_io_unlock = ocfs2_refcount_cache_io_unlock,
203 static struct ocfs2_refcount_tree *
204 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
206 struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
207 struct ocfs2_refcount_tree *tree = NULL;
210 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
212 if (blkno < tree->rf_blkno)
214 else if (blkno > tree->rf_blkno)
223 /* osb_lock is already locked. */
224 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
225 struct ocfs2_refcount_tree *new)
227 u64 rf_blkno = new->rf_blkno;
228 struct rb_node *parent = NULL;
229 struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
230 struct ocfs2_refcount_tree *tmp;
235 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
238 if (rf_blkno < tmp->rf_blkno)
240 else if (rf_blkno > tmp->rf_blkno)
243 /* This should never happen! */
244 mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
245 (unsigned long long)rf_blkno);
250 rb_link_node(&new->rf_node, parent, p);
251 rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
254 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
256 ocfs2_metadata_cache_exit(&tree->rf_ci);
257 ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
258 ocfs2_lock_res_free(&tree->rf_lockres);
263 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
264 struct ocfs2_refcount_tree *tree)
266 rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
267 if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
268 osb->osb_ref_tree_lru = NULL;
271 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
272 struct ocfs2_refcount_tree *tree)
274 spin_lock(&osb->osb_lock);
275 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
276 spin_unlock(&osb->osb_lock);
279 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
281 struct ocfs2_refcount_tree *tree =
282 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
284 ocfs2_free_refcount_tree(tree);
288 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
290 kref_get(&tree->rf_getcnt);
294 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
296 kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
299 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
300 struct super_block *sb)
302 ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
303 mutex_init(&new->rf_io_mutex);
305 spin_lock_init(&new->rf_lock);
308 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
309 struct ocfs2_refcount_tree *new,
310 u64 rf_blkno, u32 generation)
312 init_rwsem(&new->rf_sem);
313 ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
314 rf_blkno, generation);
317 static struct ocfs2_refcount_tree*
318 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
320 struct ocfs2_refcount_tree *new;
322 new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
326 new->rf_blkno = rf_blkno;
327 kref_init(&new->rf_getcnt);
328 ocfs2_init_refcount_tree_ci(new, osb->sb);
333 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
334 struct ocfs2_refcount_tree **ret_tree)
337 struct ocfs2_refcount_tree *tree, *new = NULL;
338 struct buffer_head *ref_root_bh = NULL;
339 struct ocfs2_refcount_block *ref_rb;
341 spin_lock(&osb->osb_lock);
342 if (osb->osb_ref_tree_lru &&
343 osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
344 tree = osb->osb_ref_tree_lru;
346 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
350 spin_unlock(&osb->osb_lock);
352 new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
359 * We need the generation to create the refcount tree lock and since
360 * it isn't changed during the tree modification, we are safe here to
361 * read without protection.
362 * We also have to purge the cache after we create the lock since the
363 * refcount block may have the stale data. It can only be trusted when
364 * we hold the refcount lock.
366 ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
369 ocfs2_metadata_cache_exit(&new->rf_ci);
374 ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
375 new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
376 ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
378 ocfs2_metadata_cache_purge(&new->rf_ci);
380 spin_lock(&osb->osb_lock);
381 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
385 ocfs2_insert_refcount_tree(osb, new);
393 osb->osb_ref_tree_lru = tree;
395 spin_unlock(&osb->osb_lock);
398 ocfs2_free_refcount_tree(new);
404 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
407 struct buffer_head *di_bh = NULL;
408 struct ocfs2_dinode *di;
410 ret = ocfs2_read_inode_block(inode, &di_bh);
416 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
418 di = (struct ocfs2_dinode *)di_bh->b_data;
419 *ref_blkno = le64_to_cpu(di->i_refcount_loc);
425 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
426 struct ocfs2_refcount_tree *tree, int rw)
430 ret = ocfs2_refcount_lock(tree, rw);
437 down_write(&tree->rf_sem);
439 down_read(&tree->rf_sem);
446 * Lock the refcount tree pointed by ref_blkno and return the tree.
447 * In most case, we lock the tree and read the refcount block.
448 * So read it here if the caller really needs it.
450 * If the tree has been re-created by other node, it will free the
451 * old one and re-create it.
453 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
454 u64 ref_blkno, int rw,
455 struct ocfs2_refcount_tree **ret_tree,
456 struct buffer_head **ref_bh)
458 int ret, delete_tree = 0;
459 struct ocfs2_refcount_tree *tree = NULL;
460 struct buffer_head *ref_root_bh = NULL;
461 struct ocfs2_refcount_block *rb;
464 ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
470 ocfs2_refcount_tree_get(tree);
472 ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
475 ocfs2_refcount_tree_put(tree);
479 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
483 ocfs2_unlock_refcount_tree(osb, tree, rw);
484 ocfs2_refcount_tree_put(tree);
488 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
490 * If the refcount block has been freed and re-created, we may need
491 * to recreate the refcount tree also.
493 * Here we just remove the tree from the rb-tree, and the last
494 * kref holder will unlock and delete this refcount_tree.
495 * Then we goto "again" and ocfs2_get_refcount_tree will create
496 * the new refcount tree for us.
498 if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
499 if (!tree->rf_removed) {
500 ocfs2_erase_refcount_tree_from_list(osb, tree);
501 tree->rf_removed = 1;
505 ocfs2_unlock_refcount_tree(osb, tree, rw);
507 * We get an extra reference when we create the refcount
508 * tree, so another put will destroy it.
511 ocfs2_refcount_tree_put(tree);
519 *ref_bh = ref_root_bh;
527 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
528 struct ocfs2_refcount_tree *tree, int rw)
531 up_write(&tree->rf_sem);
533 up_read(&tree->rf_sem);
535 ocfs2_refcount_unlock(tree, rw);
536 ocfs2_refcount_tree_put(tree);
539 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
541 struct rb_node *node;
542 struct ocfs2_refcount_tree *tree;
543 struct rb_root *root = &osb->osb_rf_lock_tree;
545 while ((node = rb_last(root)) != NULL) {
546 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
548 mlog(0, "Purge tree %llu\n",
549 (unsigned long long) tree->rf_blkno);
551 rb_erase(&tree->rf_node, root);
552 ocfs2_free_refcount_tree(tree);
557 * Create a refcount tree for an inode.
558 * We take for granted that the inode is already locked.
560 static int ocfs2_create_refcount_tree(struct inode *inode,
561 struct buffer_head *di_bh)
564 handle_t *handle = NULL;
565 struct ocfs2_alloc_context *meta_ac = NULL;
566 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
567 struct ocfs2_inode_info *oi = OCFS2_I(inode);
568 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
569 struct buffer_head *new_bh = NULL;
570 struct ocfs2_refcount_block *rb;
571 struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
572 u16 suballoc_bit_start;
574 u64 suballoc_loc, first_blkno;
576 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
578 mlog(0, "create tree for inode %lu\n", inode->i_ino);
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);
616 ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
618 ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
619 OCFS2_JOURNAL_ACCESS_CREATE);
625 /* Initialize ocfs2_refcount_block. */
626 rb = (struct ocfs2_refcount_block *)new_bh->b_data;
627 memset(rb, 0, inode->i_sb->s_blocksize);
628 strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
629 rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
630 rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
631 rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
632 rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
633 rb->rf_blkno = cpu_to_le64(first_blkno);
634 rb->rf_count = cpu_to_le32(1);
635 rb->rf_records.rl_count =
636 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
637 spin_lock(&osb->osb_lock);
638 rb->rf_generation = osb->s_next_generation++;
639 spin_unlock(&osb->osb_lock);
641 ocfs2_journal_dirty(handle, new_bh);
643 spin_lock(&oi->ip_lock);
644 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
645 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
646 di->i_refcount_loc = cpu_to_le64(first_blkno);
647 spin_unlock(&oi->ip_lock);
649 mlog(0, "created tree for inode %lu, refblock %llu\n",
650 inode->i_ino, (unsigned long long)first_blkno);
652 ocfs2_journal_dirty(handle, di_bh);
655 * We have to init the tree lock here since it will use
656 * the generation number to create it.
658 new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
659 ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
660 new_tree->rf_generation);
662 spin_lock(&osb->osb_lock);
663 tree = ocfs2_find_refcount_tree(osb, first_blkno);
666 * We've just created a new refcount tree in this block. If
667 * we found a refcount tree on the ocfs2_super, it must be
668 * one we just deleted. We free the old tree before
669 * inserting the new tree.
671 BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
673 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
674 ocfs2_insert_refcount_tree(osb, new_tree);
675 spin_unlock(&osb->osb_lock);
678 ocfs2_refcount_tree_put(tree);
681 ocfs2_commit_trans(osb, handle);
685 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
691 ocfs2_free_alloc_context(meta_ac);
696 static int ocfs2_set_refcount_tree(struct inode *inode,
697 struct buffer_head *di_bh,
701 handle_t *handle = NULL;
702 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
703 struct ocfs2_inode_info *oi = OCFS2_I(inode);
704 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
705 struct buffer_head *ref_root_bh = NULL;
706 struct ocfs2_refcount_block *rb;
707 struct ocfs2_refcount_tree *ref_tree;
709 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
711 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
712 &ref_tree, &ref_root_bh);
718 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
719 if (IS_ERR(handle)) {
720 ret = PTR_ERR(handle);
725 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
726 OCFS2_JOURNAL_ACCESS_WRITE);
732 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
733 OCFS2_JOURNAL_ACCESS_WRITE);
739 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
740 le32_add_cpu(&rb->rf_count, 1);
742 ocfs2_journal_dirty(handle, ref_root_bh);
744 spin_lock(&oi->ip_lock);
745 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
746 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
747 di->i_refcount_loc = cpu_to_le64(refcount_loc);
748 spin_unlock(&oi->ip_lock);
749 ocfs2_journal_dirty(handle, di_bh);
752 ocfs2_commit_trans(osb, handle);
754 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
760 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
762 int ret, delete_tree = 0;
763 handle_t *handle = NULL;
764 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
765 struct ocfs2_inode_info *oi = OCFS2_I(inode);
766 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
767 struct ocfs2_refcount_block *rb;
768 struct inode *alloc_inode = NULL;
769 struct buffer_head *alloc_bh = NULL;
770 struct buffer_head *blk_bh = NULL;
771 struct ocfs2_refcount_tree *ref_tree;
772 int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
773 u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
776 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
780 ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
786 rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
789 * If we are the last user, we need to free the block.
790 * So lock the allocator ahead.
792 if (le32_to_cpu(rb->rf_count) == 1) {
793 blk = le64_to_cpu(rb->rf_blkno);
794 bit = le16_to_cpu(rb->rf_suballoc_bit);
795 if (rb->rf_suballoc_loc)
796 bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
798 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
800 alloc_inode = ocfs2_get_system_file_inode(osb,
801 EXTENT_ALLOC_SYSTEM_INODE,
802 le16_to_cpu(rb->rf_suballoc_slot));
808 mutex_lock(&alloc_inode->i_mutex);
810 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
816 credits += OCFS2_SUBALLOC_FREE;
819 handle = ocfs2_start_trans(osb, credits);
820 if (IS_ERR(handle)) {
821 ret = PTR_ERR(handle);
826 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
827 OCFS2_JOURNAL_ACCESS_WRITE);
833 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
834 OCFS2_JOURNAL_ACCESS_WRITE);
840 spin_lock(&oi->ip_lock);
841 oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
842 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
843 di->i_refcount_loc = 0;
844 spin_unlock(&oi->ip_lock);
845 ocfs2_journal_dirty(handle, di_bh);
847 le32_add_cpu(&rb->rf_count , -1);
848 ocfs2_journal_dirty(handle, blk_bh);
852 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
853 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
854 alloc_bh, bit, bg_blkno, 1);
860 ocfs2_commit_trans(osb, handle);
863 ocfs2_inode_unlock(alloc_inode, 1);
868 mutex_unlock(&alloc_inode->i_mutex);
872 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
874 ocfs2_refcount_tree_put(ref_tree);
880 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
881 struct buffer_head *ref_leaf_bh,
882 u64 cpos, unsigned int len,
883 struct ocfs2_refcount_rec *ret_rec,
887 struct ocfs2_refcount_block *rb =
888 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
889 struct ocfs2_refcount_rec *rec = NULL;
891 for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
892 rec = &rb->rf_records.rl_recs[i];
894 if (le64_to_cpu(rec->r_cpos) +
895 le32_to_cpu(rec->r_clusters) <= cpos)
897 else if (le64_to_cpu(rec->r_cpos) > cpos)
900 /* ok, cpos fail in this rec. Just return. */
907 /* We meet with a hole here, so fake the rec. */
908 ret_rec->r_cpos = cpu_to_le64(cpos);
909 ret_rec->r_refcount = 0;
910 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
911 le64_to_cpu(rec->r_cpos) < cpos + len)
912 ret_rec->r_clusters =
913 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
915 ret_rec->r_clusters = cpu_to_le32(len);
923 * Try to remove refcount tree. The mechanism is:
924 * 1) Check whether i_clusters == 0, if no, exit.
925 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
926 * 3) Check whether we have inline xattr stored outside, if yes, exit.
927 * 4) Remove the tree.
929 int ocfs2_try_remove_refcount_tree(struct inode *inode,
930 struct buffer_head *di_bh)
933 struct ocfs2_inode_info *oi = OCFS2_I(inode);
934 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
936 down_write(&oi->ip_xattr_sem);
937 down_write(&oi->ip_alloc_sem);
942 if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
945 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
946 ocfs2_has_inline_xattr_value_outside(inode, di))
949 ret = ocfs2_remove_refcount_tree(inode, di_bh);
953 up_write(&oi->ip_alloc_sem);
954 up_write(&oi->ip_xattr_sem);
959 * Find the end range for a leaf refcount block indicated by
960 * el->l_recs[index].e_blkno.
962 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
963 struct buffer_head *ref_root_bh,
964 struct ocfs2_extent_block *eb,
965 struct ocfs2_extent_list *el,
966 int index, u32 *cpos_end)
968 int ret, i, subtree_root;
971 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
972 struct ocfs2_path *left_path = NULL, *right_path = NULL;
973 struct ocfs2_extent_tree et;
974 struct ocfs2_extent_list *tmp_el;
976 if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
978 * We have a extent rec after index, so just use the e_cpos
979 * of the next extent rec.
981 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
985 if (!eb || (eb && !eb->h_next_leaf_blk)) {
987 * We are the last extent rec, so any high cpos should
988 * be stored in this leaf refcount block.
990 *cpos_end = UINT_MAX;
995 * If the extent block isn't the last one, we have to find
996 * the subtree root between this extent block and the next
997 * leaf extent block and get the corresponding e_cpos from
998 * the subroot. Otherwise we may corrupt the b-tree.
1000 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1002 left_path = ocfs2_new_path_from_et(&et);
1009 cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1010 ret = ocfs2_find_path(ci, left_path, cpos);
1016 right_path = ocfs2_new_path_from_path(left_path);
1023 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1029 ret = ocfs2_find_path(ci, right_path, cpos);
1035 subtree_root = ocfs2_find_subtree_root(&et, left_path,
1038 tmp_el = left_path->p_node[subtree_root].el;
1039 blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1040 for (i = 0; i < le32_to_cpu(tmp_el->l_next_free_rec); i++) {
1041 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1042 *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1047 BUG_ON(i == le32_to_cpu(tmp_el->l_next_free_rec));
1050 ocfs2_free_path(left_path);
1051 ocfs2_free_path(right_path);
1056 * Given a cpos and len, try to find the refcount record which contains cpos.
1057 * 1. If cpos can be found in one refcount record, return the record.
1058 * 2. If cpos can't be found, return a fake record which start from cpos
1059 * and end at a small value between cpos+len and start of the next record.
1060 * This fake record has r_refcount = 0.
1062 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1063 struct buffer_head *ref_root_bh,
1064 u64 cpos, unsigned int len,
1065 struct ocfs2_refcount_rec *ret_rec,
1067 struct buffer_head **ret_bh)
1069 int ret = 0, i, found;
1070 u32 low_cpos, uninitialized_var(cpos_end);
1071 struct ocfs2_extent_list *el;
1072 struct ocfs2_extent_rec *rec = NULL;
1073 struct ocfs2_extent_block *eb = NULL;
1074 struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1075 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1076 struct ocfs2_refcount_block *rb =
1077 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1079 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1080 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1082 *ret_bh = ref_root_bh;
1083 get_bh(ref_root_bh);
1088 low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1090 if (el->l_tree_depth) {
1091 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1097 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1100 if (el->l_tree_depth) {
1102 "refcount tree %llu has non zero tree "
1103 "depth in leaf btree tree block %llu\n",
1104 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1105 (unsigned long long)eb_bh->b_blocknr);
1112 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1113 rec = &el->l_recs[i];
1115 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1122 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1123 eb, el, i, &cpos_end);
1129 if (cpos_end < low_cpos + len)
1130 len = cpos_end - low_cpos;
1133 ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1140 ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1142 *ret_bh = ref_leaf_bh;
1148 enum ocfs2_ref_rec_contig {
1149 REF_CONTIG_NONE = 0,
1152 REF_CONTIG_LEFTRIGHT,
1155 static enum ocfs2_ref_rec_contig
1156 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1159 if ((rb->rf_records.rl_recs[index].r_refcount ==
1160 rb->rf_records.rl_recs[index + 1].r_refcount) &&
1161 (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1162 le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1163 le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1164 return REF_CONTIG_RIGHT;
1166 return REF_CONTIG_NONE;
1169 static enum ocfs2_ref_rec_contig
1170 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1173 enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1175 if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1176 ret = ocfs2_refcount_rec_adjacent(rb, index);
1179 enum ocfs2_ref_rec_contig tmp;
1181 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1183 if (tmp == REF_CONTIG_RIGHT) {
1184 if (ret == REF_CONTIG_RIGHT)
1185 ret = REF_CONTIG_LEFTRIGHT;
1187 ret = REF_CONTIG_LEFT;
1194 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1197 BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1198 rb->rf_records.rl_recs[index+1].r_refcount);
1200 le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1201 le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1203 if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1204 memmove(&rb->rf_records.rl_recs[index + 1],
1205 &rb->rf_records.rl_recs[index + 2],
1206 sizeof(struct ocfs2_refcount_rec) *
1207 (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1209 memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1210 0, sizeof(struct ocfs2_refcount_rec));
1211 le16_add_cpu(&rb->rf_records.rl_used, -1);
1215 * Merge the refcount rec if we are contiguous with the adjacent recs.
1217 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1220 enum ocfs2_ref_rec_contig contig =
1221 ocfs2_refcount_rec_contig(rb, index);
1223 if (contig == REF_CONTIG_NONE)
1226 if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1231 ocfs2_rotate_refcount_rec_left(rb, index);
1233 if (contig == REF_CONTIG_LEFTRIGHT)
1234 ocfs2_rotate_refcount_rec_left(rb, index);
1238 * Change the refcount indexed by "index" in ref_bh.
1239 * If refcount reaches 0, remove it.
1241 static int ocfs2_change_refcount_rec(handle_t *handle,
1242 struct ocfs2_caching_info *ci,
1243 struct buffer_head *ref_leaf_bh,
1244 int index, int merge, int change)
1247 struct ocfs2_refcount_block *rb =
1248 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1249 struct ocfs2_refcount_list *rl = &rb->rf_records;
1250 struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1252 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1253 OCFS2_JOURNAL_ACCESS_WRITE);
1259 mlog(0, "change index %d, old count %u, change %d\n", index,
1260 le32_to_cpu(rec->r_refcount), change);
1261 le32_add_cpu(&rec->r_refcount, change);
1263 if (!rec->r_refcount) {
1264 if (index != le16_to_cpu(rl->rl_used) - 1) {
1265 memmove(rec, rec + 1,
1266 (le16_to_cpu(rl->rl_used) - index - 1) *
1267 sizeof(struct ocfs2_refcount_rec));
1268 memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1269 0, sizeof(struct ocfs2_refcount_rec));
1272 le16_add_cpu(&rl->rl_used, -1);
1274 ocfs2_refcount_rec_merge(rb, index);
1276 ocfs2_journal_dirty(handle, ref_leaf_bh);
1281 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1282 struct ocfs2_caching_info *ci,
1283 struct buffer_head *ref_root_bh,
1284 struct buffer_head **ref_leaf_bh,
1285 struct ocfs2_alloc_context *meta_ac)
1288 u16 suballoc_bit_start;
1290 u64 suballoc_loc, blkno;
1291 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1292 struct buffer_head *new_bh = NULL;
1293 struct ocfs2_refcount_block *new_rb;
1294 struct ocfs2_refcount_block *root_rb =
1295 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1297 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1298 OCFS2_JOURNAL_ACCESS_WRITE);
1304 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1305 &suballoc_bit_start, &num_got,
1312 new_bh = sb_getblk(sb, blkno);
1313 if (new_bh == NULL) {
1318 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1320 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1321 OCFS2_JOURNAL_ACCESS_CREATE);
1328 * Initialize ocfs2_refcount_block.
1329 * It should contain the same information as the old root.
1330 * so just memcpy it and change the corresponding field.
1332 memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1334 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1335 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1336 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1337 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1338 new_rb->rf_blkno = cpu_to_le64(blkno);
1339 new_rb->rf_cpos = cpu_to_le32(0);
1340 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1341 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1342 ocfs2_journal_dirty(handle, new_bh);
1344 /* Now change the root. */
1345 memset(&root_rb->rf_list, 0, sb->s_blocksize -
1346 offsetof(struct ocfs2_refcount_block, rf_list));
1347 root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1348 root_rb->rf_clusters = cpu_to_le32(1);
1349 root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1350 root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1351 root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1352 root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1354 ocfs2_journal_dirty(handle, ref_root_bh);
1356 mlog(0, "new leaf block %llu, used %u\n", (unsigned long long)blkno,
1357 le16_to_cpu(new_rb->rf_records.rl_used));
1359 *ref_leaf_bh = new_bh;
1366 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1367 struct ocfs2_refcount_rec *next)
1369 if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1370 ocfs2_get_ref_rec_low_cpos(next))
1376 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1378 const struct ocfs2_refcount_rec *l = a, *r = b;
1379 u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1380 u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1382 if (l_cpos > r_cpos)
1384 if (l_cpos < r_cpos)
1389 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1391 const struct ocfs2_refcount_rec *l = a, *r = b;
1392 u64 l_cpos = le64_to_cpu(l->r_cpos);
1393 u64 r_cpos = le64_to_cpu(r->r_cpos);
1395 if (l_cpos > r_cpos)
1397 if (l_cpos < r_cpos)
1402 static void swap_refcount_rec(void *a, void *b, int size)
1404 struct ocfs2_refcount_rec *l = a, *r = b, tmp;
1406 tmp = *(struct ocfs2_refcount_rec *)l;
1407 *(struct ocfs2_refcount_rec *)l =
1408 *(struct ocfs2_refcount_rec *)r;
1409 *(struct ocfs2_refcount_rec *)r = tmp;
1413 * The refcount cpos are ordered by their 64bit cpos,
1414 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1415 * So we need to make sure that this pos isn't intersected with others.
1417 * Note: The refcount block is already sorted by their low 32 bit cpos,
1418 * So just try the middle pos first, and we will exit when we find
1419 * the good position.
1421 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1422 u32 *split_pos, int *split_index)
1424 int num_used = le16_to_cpu(rl->rl_used);
1425 int delta, middle = num_used / 2;
1427 for (delta = 0; delta < middle; delta++) {
1428 /* Let's check delta earlier than middle */
1429 if (ocfs2_refcount_rec_no_intersect(
1430 &rl->rl_recs[middle - delta - 1],
1431 &rl->rl_recs[middle - delta])) {
1432 *split_index = middle - delta;
1436 /* For even counts, don't walk off the end */
1437 if ((middle + delta + 1) == num_used)
1440 /* Now try delta past middle */
1441 if (ocfs2_refcount_rec_no_intersect(
1442 &rl->rl_recs[middle + delta],
1443 &rl->rl_recs[middle + delta + 1])) {
1444 *split_index = middle + delta + 1;
1449 if (delta >= middle)
1452 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1456 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1457 struct buffer_head *new_bh,
1460 int split_index = 0, num_moved, ret;
1462 struct ocfs2_refcount_block *rb =
1463 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1464 struct ocfs2_refcount_list *rl = &rb->rf_records;
1465 struct ocfs2_refcount_block *new_rb =
1466 (struct ocfs2_refcount_block *)new_bh->b_data;
1467 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1469 mlog(0, "split old leaf refcount block %llu, count = %u, used = %u\n",
1470 (unsigned long long)ref_leaf_bh->b_blocknr,
1471 le32_to_cpu(rl->rl_count), le32_to_cpu(rl->rl_used));
1474 * XXX: Improvement later.
1475 * If we know all the high 32 bit cpos is the same, no need to sort.
1477 * In order to make the whole process safe, we do:
1478 * 1. sort the entries by their low 32 bit cpos first so that we can
1479 * find the split cpos easily.
1480 * 2. call ocfs2_insert_extent to insert the new refcount block.
1481 * 3. move the refcount rec to the new block.
1482 * 4. sort the entries by their 64 bit cpos.
1483 * 5. dirty the new_rb and rb.
1485 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1486 sizeof(struct ocfs2_refcount_rec),
1487 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1489 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1495 new_rb->rf_cpos = cpu_to_le32(cpos);
1497 /* move refcount records starting from split_index to the new block. */
1498 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1499 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1500 num_moved * sizeof(struct ocfs2_refcount_rec));
1502 /*ok, remove the entries we just moved over to the other block. */
1503 memset(&rl->rl_recs[split_index], 0,
1504 num_moved * sizeof(struct ocfs2_refcount_rec));
1506 /* change old and new rl_used accordingly. */
1507 le16_add_cpu(&rl->rl_used, -num_moved);
1508 new_rl->rl_used = cpu_to_le16(num_moved);
1510 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1511 sizeof(struct ocfs2_refcount_rec),
1512 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1514 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1515 sizeof(struct ocfs2_refcount_rec),
1516 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1522 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1523 struct ocfs2_caching_info *ci,
1524 struct buffer_head *ref_root_bh,
1525 struct buffer_head *ref_leaf_bh,
1526 struct ocfs2_alloc_context *meta_ac)
1529 u16 suballoc_bit_start;
1530 u32 num_got, new_cpos;
1531 u64 suballoc_loc, blkno;
1532 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1533 struct ocfs2_refcount_block *root_rb =
1534 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1535 struct buffer_head *new_bh = NULL;
1536 struct ocfs2_refcount_block *new_rb;
1537 struct ocfs2_extent_tree ref_et;
1539 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1541 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1542 OCFS2_JOURNAL_ACCESS_WRITE);
1548 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1549 OCFS2_JOURNAL_ACCESS_WRITE);
1555 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1556 &suballoc_bit_start, &num_got,
1563 new_bh = sb_getblk(sb, blkno);
1564 if (new_bh == NULL) {
1569 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1571 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1572 OCFS2_JOURNAL_ACCESS_CREATE);
1578 /* Initialize ocfs2_refcount_block. */
1579 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1580 memset(new_rb, 0, sb->s_blocksize);
1581 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1582 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1583 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1584 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1585 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1586 new_rb->rf_blkno = cpu_to_le64(blkno);
1587 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1588 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1589 new_rb->rf_records.rl_count =
1590 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1591 new_rb->rf_generation = root_rb->rf_generation;
1593 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1599 ocfs2_journal_dirty(handle, ref_leaf_bh);
1600 ocfs2_journal_dirty(handle, new_bh);
1602 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1604 mlog(0, "insert new leaf block %llu at %u\n",
1605 (unsigned long long)new_bh->b_blocknr, new_cpos);
1607 /* Insert the new leaf block with the specific offset cpos. */
1608 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1618 static int ocfs2_expand_refcount_tree(handle_t *handle,
1619 struct ocfs2_caching_info *ci,
1620 struct buffer_head *ref_root_bh,
1621 struct buffer_head *ref_leaf_bh,
1622 struct ocfs2_alloc_context *meta_ac)
1625 struct buffer_head *expand_bh = NULL;
1627 if (ref_root_bh == ref_leaf_bh) {
1629 * the old root bh hasn't been expanded to a b-tree,
1630 * so expand it first.
1632 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1633 &expand_bh, meta_ac);
1639 expand_bh = ref_leaf_bh;
1644 /* Now add a new refcount block into the tree.*/
1645 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1646 expand_bh, meta_ac);
1655 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1657 * Only called when we have inserted a new refcount rec at index 0
1658 * which means ocfs2_extent_rec.e_cpos may need some change.
1660 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1661 struct ocfs2_caching_info *ci,
1662 struct buffer_head *ref_root_bh,
1663 struct buffer_head *ref_leaf_bh,
1664 struct ocfs2_refcount_rec *rec)
1667 u32 new_cpos, old_cpos;
1668 struct ocfs2_path *path = NULL;
1669 struct ocfs2_extent_tree et;
1670 struct ocfs2_refcount_block *rb =
1671 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1672 struct ocfs2_extent_list *el;
1674 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1677 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1678 old_cpos = le32_to_cpu(rb->rf_cpos);
1679 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1680 if (old_cpos <= new_cpos)
1683 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1685 path = ocfs2_new_path_from_et(&et);
1692 ret = ocfs2_find_path(ci, path, old_cpos);
1699 * 2 more credits, one for the leaf refcount block, one for
1700 * the extent block contains the extent rec.
1702 ret = ocfs2_extend_trans(handle, 2);
1708 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1709 OCFS2_JOURNAL_ACCESS_WRITE);
1715 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1716 OCFS2_JOURNAL_ACCESS_WRITE);
1722 /* change the leaf extent block first. */
1723 el = path_leaf_el(path);
1725 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1726 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1729 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1731 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1733 /* change the r_cpos in the leaf block. */
1734 rb->rf_cpos = cpu_to_le32(new_cpos);
1736 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1737 ocfs2_journal_dirty(handle, ref_leaf_bh);
1740 ocfs2_free_path(path);
1744 static int ocfs2_insert_refcount_rec(handle_t *handle,
1745 struct ocfs2_caching_info *ci,
1746 struct buffer_head *ref_root_bh,
1747 struct buffer_head *ref_leaf_bh,
1748 struct ocfs2_refcount_rec *rec,
1749 int index, int merge,
1750 struct ocfs2_alloc_context *meta_ac)
1753 struct ocfs2_refcount_block *rb =
1754 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1755 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1756 struct buffer_head *new_bh = NULL;
1758 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1760 if (rf_list->rl_used == rf_list->rl_count) {
1761 u64 cpos = le64_to_cpu(rec->r_cpos);
1762 u32 len = le32_to_cpu(rec->r_clusters);
1764 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1765 ref_leaf_bh, meta_ac);
1771 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1772 cpos, len, NULL, &index,
1779 ref_leaf_bh = new_bh;
1780 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1781 rf_list = &rb->rf_records;
1784 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1785 OCFS2_JOURNAL_ACCESS_WRITE);
1791 if (index < le16_to_cpu(rf_list->rl_used))
1792 memmove(&rf_list->rl_recs[index + 1],
1793 &rf_list->rl_recs[index],
1794 (le16_to_cpu(rf_list->rl_used) - index) *
1795 sizeof(struct ocfs2_refcount_rec));
1797 mlog(0, "insert refcount record start %llu, len %u, count %u "
1798 "to leaf block %llu at index %d\n",
1799 (unsigned long long)le64_to_cpu(rec->r_cpos),
1800 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount),
1801 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1803 rf_list->rl_recs[index] = *rec;
1805 le16_add_cpu(&rf_list->rl_used, 1);
1808 ocfs2_refcount_rec_merge(rb, index);
1810 ocfs2_journal_dirty(handle, ref_leaf_bh);
1813 ret = ocfs2_adjust_refcount_rec(handle, ci,
1825 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1826 * This is much simple than our b-tree code.
1827 * split_rec is the new refcount rec we want to insert.
1828 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1829 * increase refcount or decrease a refcount to non-zero).
1830 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1831 * rec( in case we decrease a refcount to zero).
1833 static int ocfs2_split_refcount_rec(handle_t *handle,
1834 struct ocfs2_caching_info *ci,
1835 struct buffer_head *ref_root_bh,
1836 struct buffer_head *ref_leaf_bh,
1837 struct ocfs2_refcount_rec *split_rec,
1838 int index, int merge,
1839 struct ocfs2_alloc_context *meta_ac,
1840 struct ocfs2_cached_dealloc_ctxt *dealloc)
1844 struct ocfs2_refcount_block *rb =
1845 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1846 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1847 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1848 struct ocfs2_refcount_rec *tail_rec = NULL;
1849 struct buffer_head *new_bh = NULL;
1851 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1853 mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
1854 le64_to_cpu(orig_rec->r_cpos), le32_to_cpu(orig_rec->r_clusters),
1855 le64_to_cpu(split_rec->r_cpos),
1856 le32_to_cpu(split_rec->r_clusters));
1859 * If we just need to split the header or tail clusters,
1860 * no more recs are needed, just split is OK.
1861 * Otherwise we at least need one new recs.
1863 if (!split_rec->r_refcount &&
1864 (split_rec->r_cpos == orig_rec->r_cpos ||
1865 le64_to_cpu(split_rec->r_cpos) +
1866 le32_to_cpu(split_rec->r_clusters) ==
1867 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1873 * We need one more rec if we split in the middle and the new rec have
1874 * some refcount in it.
1876 if (split_rec->r_refcount &&
1877 (split_rec->r_cpos != orig_rec->r_cpos &&
1878 le64_to_cpu(split_rec->r_cpos) +
1879 le32_to_cpu(split_rec->r_clusters) !=
1880 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1883 /* If the leaf block don't have enough record, expand it. */
1884 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1885 le16_to_cpu(rf_list->rl_count)) {
1886 struct ocfs2_refcount_rec tmp_rec;
1887 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1888 len = le32_to_cpu(orig_rec->r_clusters);
1889 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1890 ref_leaf_bh, meta_ac);
1897 * We have to re-get it since now cpos may be moved to
1898 * another leaf block.
1900 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1901 cpos, len, &tmp_rec, &index,
1908 ref_leaf_bh = new_bh;
1909 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1910 rf_list = &rb->rf_records;
1911 orig_rec = &rf_list->rl_recs[index];
1914 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1915 OCFS2_JOURNAL_ACCESS_WRITE);
1922 * We have calculated out how many new records we need and store
1923 * in recs_need, so spare enough space first by moving the records
1924 * after "index" to the end.
1926 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1927 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1928 &rf_list->rl_recs[index + 1],
1929 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1930 sizeof(struct ocfs2_refcount_rec));
1932 len = (le64_to_cpu(orig_rec->r_cpos) +
1933 le32_to_cpu(orig_rec->r_clusters)) -
1934 (le64_to_cpu(split_rec->r_cpos) +
1935 le32_to_cpu(split_rec->r_clusters));
1938 * If we have "len", the we will split in the tail and move it
1939 * to the end of the space we have just spared.
1942 tail_rec = &rf_list->rl_recs[index + recs_need];
1944 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1945 le64_add_cpu(&tail_rec->r_cpos,
1946 le32_to_cpu(tail_rec->r_clusters) - len);
1947 tail_rec->r_clusters = cpu_to_le32(len);
1951 * If the split pos isn't the same as the original one, we need to
1952 * split in the head.
1954 * Note: We have the chance that split_rec.r_refcount = 0,
1955 * recs_need = 0 and len > 0, which means we just cut the head from
1956 * the orig_rec and in that case we have done some modification in
1957 * orig_rec above, so the check for r_cpos is faked.
1959 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1960 len = le64_to_cpu(split_rec->r_cpos) -
1961 le64_to_cpu(orig_rec->r_cpos);
1962 orig_rec->r_clusters = cpu_to_le32(len);
1966 le16_add_cpu(&rf_list->rl_used, recs_need);
1968 if (split_rec->r_refcount) {
1969 rf_list->rl_recs[index] = *split_rec;
1970 mlog(0, "insert refcount record start %llu, len %u, count %u "
1971 "to leaf block %llu at index %d\n",
1972 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1973 le32_to_cpu(split_rec->r_clusters),
1974 le32_to_cpu(split_rec->r_refcount),
1975 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1978 ocfs2_refcount_rec_merge(rb, index);
1981 ocfs2_journal_dirty(handle, ref_leaf_bh);
1988 static int __ocfs2_increase_refcount(handle_t *handle,
1989 struct ocfs2_caching_info *ci,
1990 struct buffer_head *ref_root_bh,
1991 u64 cpos, u32 len, int merge,
1992 struct ocfs2_alloc_context *meta_ac,
1993 struct ocfs2_cached_dealloc_ctxt *dealloc)
1996 struct buffer_head *ref_leaf_bh = NULL;
1997 struct ocfs2_refcount_rec rec;
1998 unsigned int set_len = 0;
2000 mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
2001 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2002 (unsigned long long)cpos, len);
2005 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2006 cpos, len, &rec, &index,
2013 set_len = le32_to_cpu(rec.r_clusters);
2016 * Here we may meet with 3 situations:
2018 * 1. If we find an already existing record, and the length
2019 * is the same, cool, we just need to increase the r_refcount
2021 * 2. If we find a hole, just insert it with r_refcount = 1.
2022 * 3. If we are in the middle of one extent record, split
2025 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2027 mlog(0, "increase refcount rec, start %llu, len %u, "
2028 "count %u\n", (unsigned long long)cpos, set_len,
2029 le32_to_cpu(rec.r_refcount));
2030 ret = ocfs2_change_refcount_rec(handle, ci,
2037 } else if (!rec.r_refcount) {
2038 rec.r_refcount = cpu_to_le32(1);
2040 mlog(0, "insert refcount rec, start %llu, len %u\n",
2041 (unsigned long long)le64_to_cpu(rec.r_cpos),
2043 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2052 set_len = min((u64)(cpos + len),
2053 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2054 rec.r_cpos = cpu_to_le64(cpos);
2055 rec.r_clusters = cpu_to_le32(set_len);
2056 le32_add_cpu(&rec.r_refcount, 1);
2058 mlog(0, "split refcount rec, start %llu, "
2059 "len %u, count %u\n",
2060 (unsigned long long)le64_to_cpu(rec.r_cpos),
2061 set_len, le32_to_cpu(rec.r_refcount));
2062 ret = ocfs2_split_refcount_rec(handle, ci,
2063 ref_root_bh, ref_leaf_bh,
2074 brelse(ref_leaf_bh);
2079 brelse(ref_leaf_bh);
2083 static int ocfs2_remove_refcount_extent(handle_t *handle,
2084 struct ocfs2_caching_info *ci,
2085 struct buffer_head *ref_root_bh,
2086 struct buffer_head *ref_leaf_bh,
2087 struct ocfs2_alloc_context *meta_ac,
2088 struct ocfs2_cached_dealloc_ctxt *dealloc)
2091 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2092 struct ocfs2_refcount_block *rb =
2093 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2094 struct ocfs2_extent_tree et;
2096 BUG_ON(rb->rf_records.rl_used);
2098 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2099 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2100 1, meta_ac, dealloc);
2106 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2109 * add the freed block to the dealloc so that it will be freed
2110 * when we run dealloc.
2112 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2113 le16_to_cpu(rb->rf_suballoc_slot),
2114 le64_to_cpu(rb->rf_suballoc_loc),
2115 le64_to_cpu(rb->rf_blkno),
2116 le16_to_cpu(rb->rf_suballoc_bit));
2122 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2123 OCFS2_JOURNAL_ACCESS_WRITE);
2129 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2131 le32_add_cpu(&rb->rf_clusters, -1);
2134 * check whether we need to restore the root refcount block if
2135 * there is no leaf extent block at atll.
2137 if (!rb->rf_list.l_next_free_rec) {
2138 BUG_ON(rb->rf_clusters);
2140 mlog(0, "reset refcount tree root %llu to be a record block.\n",
2141 (unsigned long long)ref_root_bh->b_blocknr);
2146 memset(&rb->rf_records, 0, sb->s_blocksize -
2147 offsetof(struct ocfs2_refcount_block, rf_records));
2148 rb->rf_records.rl_count =
2149 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2152 ocfs2_journal_dirty(handle, ref_root_bh);
2158 int ocfs2_increase_refcount(handle_t *handle,
2159 struct ocfs2_caching_info *ci,
2160 struct buffer_head *ref_root_bh,
2162 struct ocfs2_alloc_context *meta_ac,
2163 struct ocfs2_cached_dealloc_ctxt *dealloc)
2165 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2170 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2171 struct ocfs2_caching_info *ci,
2172 struct buffer_head *ref_root_bh,
2173 struct buffer_head *ref_leaf_bh,
2174 int index, u64 cpos, unsigned int len,
2175 struct ocfs2_alloc_context *meta_ac,
2176 struct ocfs2_cached_dealloc_ctxt *dealloc)
2179 struct ocfs2_refcount_block *rb =
2180 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2181 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2183 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2185 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2187 if (cpos == le64_to_cpu(rec->r_cpos) &&
2188 len == le32_to_cpu(rec->r_clusters))
2189 ret = ocfs2_change_refcount_rec(handle, ci,
2190 ref_leaf_bh, index, 1, -1);
2192 struct ocfs2_refcount_rec split = *rec;
2193 split.r_cpos = cpu_to_le64(cpos);
2194 split.r_clusters = cpu_to_le32(len);
2196 le32_add_cpu(&split.r_refcount, -1);
2198 mlog(0, "split refcount rec, start %llu, "
2199 "len %u, count %u, original start %llu, len %u\n",
2200 (unsigned long long)le64_to_cpu(split.r_cpos),
2201 len, le32_to_cpu(split.r_refcount),
2202 (unsigned long long)le64_to_cpu(rec->r_cpos),
2203 le32_to_cpu(rec->r_clusters));
2204 ret = ocfs2_split_refcount_rec(handle, ci,
2205 ref_root_bh, ref_leaf_bh,
2215 /* Remove the leaf refcount block if it contains no refcount record. */
2216 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2217 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2218 ref_leaf_bh, meta_ac,
2228 static int __ocfs2_decrease_refcount(handle_t *handle,
2229 struct ocfs2_caching_info *ci,
2230 struct buffer_head *ref_root_bh,
2232 struct ocfs2_alloc_context *meta_ac,
2233 struct ocfs2_cached_dealloc_ctxt *dealloc,
2236 int ret = 0, index = 0;
2237 struct ocfs2_refcount_rec rec;
2238 unsigned int r_count = 0, r_len;
2239 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2240 struct buffer_head *ref_leaf_bh = NULL;
2242 mlog(0, "Tree owner %llu, decrease refcount start %llu, "
2243 "len %u, delete %u\n",
2244 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2245 (unsigned long long)cpos, len, delete);
2248 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2249 cpos, len, &rec, &index,
2256 r_count = le32_to_cpu(rec.r_refcount);
2257 BUG_ON(r_count == 0);
2259 BUG_ON(r_count > 1);
2261 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2262 le32_to_cpu(rec.r_clusters)) - cpos;
2264 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2273 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2274 ret = ocfs2_cache_cluster_dealloc(dealloc,
2275 ocfs2_clusters_to_blocks(sb, cpos),
2285 brelse(ref_leaf_bh);
2290 brelse(ref_leaf_bh);
2294 /* Caller must hold refcount tree lock. */
2295 int ocfs2_decrease_refcount(struct inode *inode,
2296 handle_t *handle, u32 cpos, u32 len,
2297 struct ocfs2_alloc_context *meta_ac,
2298 struct ocfs2_cached_dealloc_ctxt *dealloc,
2303 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2304 struct buffer_head *ref_root_bh = NULL;
2305 struct ocfs2_refcount_tree *tree;
2307 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2309 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2315 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2321 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2328 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2329 cpos, len, meta_ac, dealloc, delete);
2333 brelse(ref_root_bh);
2338 * Mark the already-existing extent at cpos as refcounted for len clusters.
2339 * This adds the refcount extent flag.
2341 * If the existing extent is larger than the request, initiate a
2342 * split. An attempt will be made at merging with adjacent extents.
2344 * The caller is responsible for passing down meta_ac if we'll need it.
2346 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2347 struct ocfs2_extent_tree *et,
2348 handle_t *handle, u32 cpos,
2350 struct ocfs2_alloc_context *meta_ac,
2351 struct ocfs2_cached_dealloc_ctxt *dealloc)
2355 mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
2356 inode->i_ino, cpos, len, phys);
2358 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2359 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2360 "tree, but the feature bit is not set in the "
2361 "super block.", inode->i_ino);
2366 ret = ocfs2_change_extent_flag(handle, et, cpos,
2367 len, phys, meta_ac, dealloc,
2368 OCFS2_EXT_REFCOUNTED, 0);
2377 * Given some contiguous physical clusters, calculate what we need
2378 * for modifying their refcount.
2380 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2381 struct ocfs2_caching_info *ci,
2382 struct buffer_head *ref_root_bh,
2388 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2389 u64 cpos = start_cpos;
2390 struct ocfs2_refcount_block *rb;
2391 struct ocfs2_refcount_rec rec;
2392 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2395 mlog(0, "start_cpos %llu, clusters %u\n",
2396 (unsigned long long)start_cpos, clusters);
2398 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2399 cpos, clusters, &rec,
2400 &index, &ref_leaf_bh);
2406 if (ref_leaf_bh != prev_bh) {
2408 * Now we encounter a new leaf block, so calculate
2409 * whether we need to extend the old leaf.
2412 rb = (struct ocfs2_refcount_block *)
2415 if (le64_to_cpu(rb->rf_records.rl_used) +
2417 le16_to_cpu(rb->rf_records.rl_count))
2424 prev_bh = ref_leaf_bh;
2428 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2430 mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
2431 "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
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 * If the refcount rec already exist, cool. We just need
2441 * to check whether there is a split. Otherwise we just need
2442 * to increase the refcount.
2443 * If we will insert one, increases recs_add.
2445 * We record all the records which will be inserted to the
2446 * same refcount block, so that we can tell exactly whether
2447 * we need a new refcount block or not.
2449 if (rec.r_refcount) {
2450 /* Check whether we need a split at the beginning. */
2451 if (cpos == start_cpos &&
2452 cpos != le64_to_cpu(rec.r_cpos))
2455 /* Check whether we need a split in the end. */
2456 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2457 le32_to_cpu(rec.r_clusters))
2462 brelse(ref_leaf_bh);
2469 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2471 if (le64_to_cpu(rb->rf_records.rl_used) + recs_add >
2472 le16_to_cpu(rb->rf_records.rl_count))
2481 mlog(0, "we need ref_blocks %d\n", ref_blocks);
2482 *meta_add += ref_blocks;
2483 *credits += ref_blocks;
2486 * So we may need ref_blocks to insert into the tree.
2487 * That also means we need to change the b-tree and add that number
2488 * of records since we never merge them.
2489 * We need one more block for expansion since the new created leaf
2490 * block is also full and needs split.
2492 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2493 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2494 struct ocfs2_extent_tree et;
2496 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2497 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2498 *credits += ocfs2_calc_extend_credits(sb,
2502 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2507 brelse(ref_leaf_bh);
2513 * For refcount tree, we will decrease some contiguous clusters
2514 * refcount count, so just go through it to see how many blocks
2515 * we gonna touch and whether we need to create new blocks.
2517 * Normally the refcount blocks store these refcount should be
2518 * contiguous also, so that we can get the number easily.
2519 * We will at most add split 2 refcount records and 2 more
2520 * refcount blocks, so just check it in a rough way.
2522 * Caller must hold refcount tree lock.
2524 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2532 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2533 struct buffer_head *ref_root_bh = NULL;
2534 struct ocfs2_refcount_tree *tree;
2535 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2537 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2538 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2539 "tree, but the feature bit is not set in the "
2540 "super block.", inode->i_ino);
2545 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2547 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2548 refcount_loc, &tree);
2554 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2561 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2564 start_cpos, clusters,
2565 ref_blocks, credits);
2571 mlog(0, "reserve new metadata %d blocks, credits = %d\n",
2572 *ref_blocks, *credits);
2575 brelse(ref_root_bh);
2579 #define MAX_CONTIG_BYTES 1048576
2581 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2583 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2586 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2588 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2592 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2593 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2594 * while still being less than or equal to it.
2596 * The goal is to break the extent at a multiple of contig_clusters.
2598 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2602 BUG_ON(start > cpos);
2604 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2608 * Given a cluster count of len, pad it out so that it is a multiple
2609 * of contig_clusters.
2611 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2614 unsigned int padded =
2615 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2616 ocfs2_cow_contig_mask(sb);
2626 * Calculate out the start and number of virtual clusters we need to to CoW.
2628 * cpos is vitual start cluster position we want to do CoW in a
2629 * file and write_len is the cluster length.
2630 * max_cpos is the place where we want to stop CoW intentionally.
2632 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2633 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2634 * get good I/O from the resulting extent tree.
2636 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2637 struct ocfs2_extent_list *el,
2645 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2646 struct buffer_head *eb_bh = NULL;
2647 struct ocfs2_extent_block *eb = NULL;
2648 struct ocfs2_extent_rec *rec;
2649 unsigned int want_clusters, rec_end = 0;
2650 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2653 BUG_ON(cpos + write_len > max_cpos);
2655 if (tree_height > 0) {
2656 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2662 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2665 if (el->l_tree_depth) {
2666 ocfs2_error(inode->i_sb,
2667 "Inode %lu has non zero tree depth in "
2668 "leaf block %llu\n", inode->i_ino,
2669 (unsigned long long)eb_bh->b_blocknr);
2676 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2677 rec = &el->l_recs[i];
2679 if (ocfs2_is_empty_extent(rec)) {
2680 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2681 "index %d\n", inode->i_ino, i);
2685 if (le32_to_cpu(rec->e_cpos) +
2686 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2689 if (*cow_len == 0) {
2691 * We should find a refcounted record in the
2694 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2695 *cow_start = le32_to_cpu(rec->e_cpos);
2699 * If we encounter a hole, a non-refcounted record or
2700 * pass the max_cpos, stop the search.
2702 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2703 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2704 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2707 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2708 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2709 if (rec_end > max_cpos) {
2711 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2715 * How many clusters do we actually need from
2716 * this extent? First we see how many we actually
2717 * need to complete the write. If that's smaller
2718 * than contig_clusters, we try for contig_clusters.
2721 want_clusters = write_len;
2723 want_clusters = (cpos + write_len) -
2724 (*cow_start + *cow_len);
2725 if (want_clusters < contig_clusters)
2726 want_clusters = contig_clusters;
2729 * If the write does not cover the whole extent, we
2730 * need to calculate how we're going to split the extent.
2731 * We try to do it on contig_clusters boundaries.
2733 * Any extent smaller than contig_clusters will be
2734 * CoWed in its entirety.
2736 if (leaf_clusters <= contig_clusters)
2737 *cow_len += leaf_clusters;
2738 else if (*cow_len || (*cow_start == cpos)) {
2740 * This extent needs to be CoW'd from its
2741 * beginning, so all we have to do is compute
2742 * how many clusters to grab. We align
2743 * want_clusters to the edge of contig_clusters
2744 * to get better I/O.
2746 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2749 if (leaf_clusters < want_clusters)
2750 *cow_len += leaf_clusters;
2752 *cow_len += want_clusters;
2753 } else if ((*cow_start + contig_clusters) >=
2754 (cpos + write_len)) {
2756 * Breaking off contig_clusters at the front
2757 * of the extent will cover our write. That's
2760 *cow_len = contig_clusters;
2761 } else if ((rec_end - cpos) <= contig_clusters) {
2763 * Breaking off contig_clusters at the tail of
2764 * this extent will cover cpos.
2766 *cow_start = rec_end - contig_clusters;
2767 *cow_len = contig_clusters;
2768 } else if ((rec_end - cpos) <= want_clusters) {
2770 * While we can't fit the entire write in this
2771 * extent, we know that the write goes from cpos
2772 * to the end of the extent. Break that off.
2773 * We try to break it at some multiple of
2774 * contig_clusters from the front of the extent.
2775 * Failing that (ie, cpos is within
2776 * contig_clusters of the front), we'll CoW the
2779 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2781 *cow_len = rec_end - *cow_start;
2784 * Ok, the entire write lives in the middle of
2785 * this extent. Let's try to slice the extent up
2786 * nicely. Optimally, our CoW region starts at
2787 * m*contig_clusters from the beginning of the
2788 * extent and goes for n*contig_clusters,
2789 * covering the entire write.
2791 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2794 want_clusters = (cpos + write_len) - *cow_start;
2795 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2797 if (*cow_start + want_clusters <= rec_end)
2798 *cow_len = want_clusters;
2800 *cow_len = rec_end - *cow_start;
2803 /* Have we covered our entire write yet? */
2804 if ((*cow_start + *cow_len) >= (cpos + write_len))
2808 * If we reach the end of the extent block and don't get enough
2809 * clusters, continue with the next extent block if possible.
2811 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2812 eb && eb->h_next_leaf_blk) {
2816 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2817 le64_to_cpu(eb->h_next_leaf_blk),
2824 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2836 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2837 * num_clusters in data_tree "et" and change the refcount for the old
2838 * clusters(starting form p_cluster) in the refcount tree.
2841 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2842 * more new leaf records.
2843 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2844 * just give data_ac = NULL.
2846 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2847 u32 p_cluster, u32 num_clusters,
2848 struct ocfs2_extent_tree *et,
2849 struct ocfs2_caching_info *ref_ci,
2850 struct buffer_head *ref_root_bh,
2851 struct ocfs2_alloc_context **meta_ac,
2852 struct ocfs2_alloc_context **data_ac,
2855 int ret = 0, meta_add = 0;
2856 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2858 if (num_free_extents < 0) {
2859 ret = num_free_extents;
2864 if (num_free_extents < num_clusters + 2)
2866 ocfs2_extend_meta_needed(et->et_root_el);
2868 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
2871 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2872 p_cluster, num_clusters,
2873 &meta_add, credits);
2879 mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
2880 meta_add, num_clusters, *credits);
2881 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2889 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2898 ocfs2_free_alloc_context(*meta_ac);
2906 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2908 BUG_ON(buffer_dirty(bh));
2910 clear_buffer_mapped(bh);
2915 static int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2916 struct ocfs2_cow_context *context,
2917 u32 cpos, u32 old_cluster,
2918 u32 new_cluster, u32 new_len)
2920 int ret = 0, partial;
2921 struct ocfs2_caching_info *ci = context->data_et.et_ci;
2922 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2923 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2926 unsigned int from, to;
2927 loff_t offset, end, map_end;
2928 struct address_space *mapping = context->inode->i_mapping;
2930 mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster,
2931 new_cluster, new_len, cpos);
2933 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2934 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2936 while (offset < end) {
2937 page_index = offset >> PAGE_CACHE_SHIFT;
2938 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2942 /* from, to is the offset within the page. */
2943 from = offset & (PAGE_CACHE_SIZE - 1);
2944 to = PAGE_CACHE_SIZE;
2945 if (map_end & (PAGE_CACHE_SIZE - 1))
2946 to = map_end & (PAGE_CACHE_SIZE - 1);
2948 page = grab_cache_page(mapping, page_index);
2951 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2952 * can't be dirtied before we CoW it out.
2954 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2955 BUG_ON(PageDirty(page));
2957 if (!PageUptodate(page)) {
2958 ret = block_read_full_page(page, ocfs2_get_block);
2966 if (page_has_buffers(page)) {
2967 ret = walk_page_buffers(handle, page_buffers(page),
2969 ocfs2_clear_cow_buffer);
2976 ocfs2_map_and_dirty_page(context->inode,
2978 page, 0, &new_block);
2979 mark_page_accessed(page);
2982 page_cache_release(page);
2992 static int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
2993 struct ocfs2_cow_context *context,
2994 u32 cpos, u32 old_cluster,
2995 u32 new_cluster, u32 new_len)
2998 struct super_block *sb = context->inode->i_sb;
2999 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3000 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3001 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3002 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3003 struct ocfs2_super *osb = OCFS2_SB(sb);
3004 struct buffer_head *old_bh = NULL;
3005 struct buffer_head *new_bh = NULL;
3007 mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster,
3008 new_cluster, new_len);
3010 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3011 new_bh = sb_getblk(osb->sb, new_block);
3012 if (new_bh == NULL) {
3018 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3020 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3026 ret = ocfs2_journal_access(handle, ci, new_bh,
3027 OCFS2_JOURNAL_ACCESS_CREATE);
3033 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3034 ocfs2_journal_dirty(handle, new_bh);
3047 static int ocfs2_clear_ext_refcount(handle_t *handle,
3048 struct ocfs2_extent_tree *et,
3049 u32 cpos, u32 p_cluster, u32 len,
3050 unsigned int ext_flags,
3051 struct ocfs2_alloc_context *meta_ac,
3052 struct ocfs2_cached_dealloc_ctxt *dealloc)
3055 struct ocfs2_extent_rec replace_rec;
3056 struct ocfs2_path *path = NULL;
3057 struct ocfs2_extent_list *el;
3058 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3059 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3061 mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
3062 (unsigned long long)ino, cpos, len, p_cluster, ext_flags);
3064 memset(&replace_rec, 0, sizeof(replace_rec));
3065 replace_rec.e_cpos = cpu_to_le32(cpos);
3066 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3067 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3069 replace_rec.e_flags = ext_flags;
3070 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3072 path = ocfs2_new_path_from_et(et);
3079 ret = ocfs2_find_path(et->et_ci, path, cpos);
3085 el = path_leaf_el(path);
3087 index = ocfs2_search_extent_list(el, cpos);
3088 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
3090 "Inode %llu has an extent at cpos %u which can no "
3091 "longer be found.\n",
3092 (unsigned long long)ino, cpos);
3097 ret = ocfs2_split_extent(handle, et, path, index,
3098 &replace_rec, meta_ac, dealloc);
3103 ocfs2_free_path(path);
3107 static int ocfs2_replace_clusters(handle_t *handle,
3108 struct ocfs2_cow_context *context,
3111 unsigned int ext_flags)
3114 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3115 u64 ino = ocfs2_metadata_cache_owner(ci);
3117 mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
3118 (unsigned long long)ino, cpos, old, new, len, ext_flags);
3120 /*If the old clusters is unwritten, no need to duplicate. */
3121 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3122 ret = context->cow_duplicate_clusters(handle, context, cpos,
3130 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3131 cpos, new, len, ext_flags,
3132 context->meta_ac, &context->dealloc);
3139 static int ocfs2_cow_sync_writeback(struct super_block *sb,
3140 struct ocfs2_cow_context *context,
3141 u32 cpos, u32 num_clusters)
3144 loff_t offset, end, map_end;
3148 if (ocfs2_should_order_data(context->inode))
3151 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3152 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3154 ret = filemap_fdatawrite_range(context->inode->i_mapping,
3161 while (offset < end) {
3162 page_index = offset >> PAGE_CACHE_SHIFT;
3163 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3167 page = grab_cache_page(context->inode->i_mapping, page_index);
3170 wait_on_page_writeback(page);
3171 if (PageError(page)) {
3175 mark_page_accessed(page);
3178 page_cache_release(page);
3188 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3189 u32 v_cluster, u32 *p_cluster,
3191 unsigned int *extent_flags)
3193 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3194 num_clusters, extent_flags);
3197 static int ocfs2_make_clusters_writable(struct super_block *sb,
3198 struct ocfs2_cow_context *context,
3199 u32 cpos, u32 p_cluster,
3200 u32 num_clusters, unsigned int e_flags)
3202 int ret, delete, index, credits = 0;
3203 u32 new_bit, new_len;
3204 unsigned int set_len;
3205 struct ocfs2_super *osb = OCFS2_SB(sb);
3207 struct buffer_head *ref_leaf_bh = NULL;
3208 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3209 struct ocfs2_refcount_rec rec;
3211 mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
3212 cpos, p_cluster, num_clusters, e_flags);
3214 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3217 context->ref_root_bh,
3219 &context->data_ac, &credits);
3225 if (context->post_refcount)
3226 credits += context->post_refcount->credits;
3228 credits += context->extra_credits;
3229 handle = ocfs2_start_trans(osb, credits);
3230 if (IS_ERR(handle)) {
3231 ret = PTR_ERR(handle);
3236 while (num_clusters) {
3237 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3238 p_cluster, num_clusters,
3239 &rec, &index, &ref_leaf_bh);
3245 BUG_ON(!rec.r_refcount);
3246 set_len = min((u64)p_cluster + num_clusters,
3247 le64_to_cpu(rec.r_cpos) +
3248 le32_to_cpu(rec.r_clusters)) - p_cluster;
3251 * There are many different situation here.
3252 * 1. If refcount == 1, remove the flag and don't COW.
3253 * 2. If refcount > 1, allocate clusters.
3254 * Here we may not allocate r_len once at a time, so continue
3255 * until we reach num_clusters.
3257 if (le32_to_cpu(rec.r_refcount) == 1) {
3259 ret = ocfs2_clear_ext_refcount(handle,
3272 ret = __ocfs2_claim_clusters(handle,
3275 &new_bit, &new_len);
3281 ret = ocfs2_replace_clusters(handle, context,
3282 cpos, p_cluster, new_bit,
3291 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3292 context->ref_root_bh,
3295 &context->dealloc, delete);
3302 p_cluster += set_len;
3303 num_clusters -= set_len;
3304 brelse(ref_leaf_bh);
3308 /* handle any post_cow action. */
3309 if (context->post_refcount && context->post_refcount->func) {
3310 ret = context->post_refcount->func(context->inode, handle,
3311 context->post_refcount->para);
3319 * Here we should write the new page out first if we are
3320 * in write-back mode.
3322 if (context->get_clusters == ocfs2_di_get_clusters) {
3323 ret = ocfs2_cow_sync_writeback(sb, context, cpos, num_clusters);
3329 ocfs2_commit_trans(osb, handle);
3332 if (context->data_ac) {
3333 ocfs2_free_alloc_context(context->data_ac);
3334 context->data_ac = NULL;
3336 if (context->meta_ac) {
3337 ocfs2_free_alloc_context(context->meta_ac);
3338 context->meta_ac = NULL;
3340 brelse(ref_leaf_bh);
3345 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3348 struct inode *inode = context->inode;
3349 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3350 u32 p_cluster, num_clusters;
3351 unsigned int ext_flags;
3352 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3354 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3355 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3356 "tree, but the feature bit is not set in the "
3357 "super block.", inode->i_ino);
3361 ocfs2_init_dealloc_ctxt(&context->dealloc);
3364 ret = context->get_clusters(context, cow_start, &p_cluster,
3365 &num_clusters, &ext_flags);
3371 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3373 if (cow_len < num_clusters)
3374 num_clusters = cow_len;
3376 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3377 cow_start, p_cluster,
3378 num_clusters, ext_flags);
3384 cow_len -= num_clusters;
3385 cow_start += num_clusters;
3388 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3389 ocfs2_schedule_truncate_log_flush(osb, 1);
3390 ocfs2_run_deallocs(osb, &context->dealloc);
3397 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3398 * past max_cpos. This will stop when it runs into a hole or an
3399 * unrefcounted extent.
3401 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3402 struct buffer_head *di_bh,
3403 u32 cpos, u32 write_len, u32 max_cpos)
3406 u32 cow_start = 0, cow_len = 0;
3407 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3408 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3409 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3410 struct buffer_head *ref_root_bh = NULL;
3411 struct ocfs2_refcount_tree *ref_tree;
3412 struct ocfs2_cow_context *context = NULL;
3414 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3416 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3417 cpos, write_len, max_cpos,
3418 &cow_start, &cow_len);
3424 mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
3425 "cow_len %u\n", inode->i_ino,
3426 cpos, write_len, cow_start, cow_len);
3428 BUG_ON(cow_len == 0);
3430 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3437 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3438 1, &ref_tree, &ref_root_bh);
3444 context->inode = inode;
3445 context->cow_start = cow_start;
3446 context->cow_len = cow_len;
3447 context->ref_tree = ref_tree;
3448 context->ref_root_bh = ref_root_bh;
3449 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3450 context->get_clusters = ocfs2_di_get_clusters;
3452 ocfs2_init_dinode_extent_tree(&context->data_et,
3453 INODE_CACHE(inode), di_bh);
3455 ret = ocfs2_replace_cow(context);
3460 * truncate the extent map here since no matter whether we meet with
3461 * any error during the action, we shouldn't trust cached extent map
3464 ocfs2_extent_map_trunc(inode, cow_start);
3466 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3467 brelse(ref_root_bh);
3474 * CoW any and all clusters between cpos and cpos+write_len.
3475 * Don't CoW past max_cpos. If this returns successfully, all
3476 * clusters between cpos and cpos+write_len are safe to modify.
3478 int ocfs2_refcount_cow(struct inode *inode,
3479 struct buffer_head *di_bh,
3480 u32 cpos, u32 write_len, u32 max_cpos)
3483 u32 p_cluster, num_clusters;
3484 unsigned int ext_flags;
3487 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3488 &num_clusters, &ext_flags);
3494 if (write_len < num_clusters)
3495 num_clusters = write_len;
3497 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3498 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3499 num_clusters, max_cpos);
3506 write_len -= num_clusters;
3507 cpos += num_clusters;
3513 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3514 u32 v_cluster, u32 *p_cluster,
3516 unsigned int *extent_flags)
3518 struct inode *inode = context->inode;
3519 struct ocfs2_xattr_value_root *xv = context->cow_object;
3521 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3522 num_clusters, &xv->xr_list,
3527 * Given a xattr value root, calculate the most meta/credits we need for
3528 * refcount tree change if we truncate it to 0.
3530 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3531 struct ocfs2_caching_info *ref_ci,
3532 struct buffer_head *ref_root_bh,
3533 struct ocfs2_xattr_value_root *xv,
3534 int *meta_add, int *credits)
3536 int ret = 0, index, ref_blocks = 0;
3537 u32 p_cluster, num_clusters;
3538 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3539 struct ocfs2_refcount_block *rb;
3540 struct ocfs2_refcount_rec rec;
3541 struct buffer_head *ref_leaf_bh = NULL;
3543 while (cpos < clusters) {
3544 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3545 &num_clusters, &xv->xr_list,
3552 cpos += num_clusters;
3554 while (num_clusters) {
3555 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3556 p_cluster, num_clusters,
3564 BUG_ON(!rec.r_refcount);
3566 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3569 * We really don't know whether the other clusters is in
3570 * this refcount block or not, so just take the worst
3571 * case that all the clusters are in this block and each
3572 * one will split a refcount rec, so totally we need
3573 * clusters * 2 new refcount rec.
3575 if (le64_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3576 le16_to_cpu(rb->rf_records.rl_count))
3580 brelse(ref_leaf_bh);
3583 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3586 num_clusters -= le32_to_cpu(rec.r_clusters);
3587 p_cluster += num_clusters;
3591 *meta_add += ref_blocks;
3595 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3596 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3597 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3599 struct ocfs2_extent_tree et;
3601 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3602 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3608 brelse(ref_leaf_bh);
3615 int ocfs2_refcount_cow_xattr(struct inode *inode,
3616 struct ocfs2_dinode *di,
3617 struct ocfs2_xattr_value_buf *vb,
3618 struct ocfs2_refcount_tree *ref_tree,
3619 struct buffer_head *ref_root_bh,
3620 u32 cpos, u32 write_len,
3621 struct ocfs2_post_refcount *post)
3624 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3625 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3626 struct ocfs2_cow_context *context = NULL;
3627 u32 cow_start, cow_len;
3629 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3631 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3632 cpos, write_len, UINT_MAX,
3633 &cow_start, &cow_len);
3639 BUG_ON(cow_len == 0);
3641 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3648 context->inode = inode;
3649 context->cow_start = cow_start;
3650 context->cow_len = cow_len;
3651 context->ref_tree = ref_tree;
3652 context->ref_root_bh = ref_root_bh;;
3653 context->cow_object = xv;
3655 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3656 /* We need the extra credits for duplicate_clusters by jbd. */
3657 context->extra_credits =
3658 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3659 context->get_clusters = ocfs2_xattr_value_get_clusters;
3660 context->post_refcount = post;
3662 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3663 INODE_CACHE(inode), vb);
3665 ret = ocfs2_replace_cow(context);
3675 * Insert a new extent into refcount tree and mark a extent rec
3676 * as refcounted in the dinode tree.
3678 int ocfs2_add_refcount_flag(struct inode *inode,
3679 struct ocfs2_extent_tree *data_et,
3680 struct ocfs2_caching_info *ref_ci,
3681 struct buffer_head *ref_root_bh,
3682 u32 cpos, u32 p_cluster, u32 num_clusters,
3683 struct ocfs2_cached_dealloc_ctxt *dealloc,
3684 struct ocfs2_post_refcount *post)
3688 int credits = 1, ref_blocks = 0;
3689 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3690 struct ocfs2_alloc_context *meta_ac = NULL;
3692 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3693 ref_ci, ref_root_bh,
3694 p_cluster, num_clusters,
3695 &ref_blocks, &credits);
3701 mlog(0, "reserve new metadata %d, credits = %d\n",
3702 ref_blocks, credits);
3705 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3706 ref_blocks, &meta_ac);
3714 credits += post->credits;
3716 handle = ocfs2_start_trans(osb, credits);
3717 if (IS_ERR(handle)) {
3718 ret = PTR_ERR(handle);
3723 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3724 cpos, num_clusters, p_cluster,
3731 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3732 p_cluster, num_clusters, 0,
3739 if (post && post->func) {
3740 ret = post->func(inode, handle, post->para);
3746 ocfs2_commit_trans(osb, handle);
3749 ocfs2_free_alloc_context(meta_ac);
3753 static int ocfs2_change_ctime(struct inode *inode,
3754 struct buffer_head *di_bh)
3758 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3760 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3761 OCFS2_INODE_UPDATE_CREDITS);
3762 if (IS_ERR(handle)) {
3763 ret = PTR_ERR(handle);
3768 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3769 OCFS2_JOURNAL_ACCESS_WRITE);
3775 inode->i_ctime = CURRENT_TIME;
3776 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3777 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3779 ocfs2_journal_dirty(handle, di_bh);
3782 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3787 static int ocfs2_attach_refcount_tree(struct inode *inode,
3788 struct buffer_head *di_bh)
3790 int ret, data_changed = 0;
3791 struct buffer_head *ref_root_bh = NULL;
3792 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3793 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3794 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3795 struct ocfs2_refcount_tree *ref_tree;
3796 unsigned int ext_flags;
3798 u32 cpos, num_clusters, clusters, p_cluster;
3799 struct ocfs2_cached_dealloc_ctxt dealloc;
3800 struct ocfs2_extent_tree di_et;
3802 ocfs2_init_dealloc_ctxt(&dealloc);
3804 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3805 ret = ocfs2_create_refcount_tree(inode, di_bh);
3812 BUG_ON(!di->i_refcount_loc);
3813 ret = ocfs2_lock_refcount_tree(osb,
3814 le64_to_cpu(di->i_refcount_loc), 1,
3815 &ref_tree, &ref_root_bh);
3821 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3824 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3826 size = i_size_read(inode);
3827 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3830 while (cpos < clusters) {
3831 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3832 &num_clusters, &ext_flags);
3834 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3835 ret = ocfs2_add_refcount_flag(inode, &di_et,
3838 p_cluster, num_clusters,
3847 cpos += num_clusters;
3851 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3852 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3863 ret = ocfs2_change_ctime(inode, di_bh);
3869 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3870 brelse(ref_root_bh);
3872 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3873 ocfs2_schedule_truncate_log_flush(osb, 1);
3874 ocfs2_run_deallocs(osb, &dealloc);
3878 * Empty the extent map so that we may get the right extent
3879 * record from the disk.
3881 ocfs2_extent_map_trunc(inode, 0);
3886 static int ocfs2_add_refcounted_extent(struct inode *inode,
3887 struct ocfs2_extent_tree *et,
3888 struct ocfs2_caching_info *ref_ci,
3889 struct buffer_head *ref_root_bh,
3890 u32 cpos, u32 p_cluster, u32 num_clusters,
3891 unsigned int ext_flags,
3892 struct ocfs2_cached_dealloc_ctxt *dealloc)
3897 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3898 struct ocfs2_alloc_context *meta_ac = NULL;
3900 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3901 p_cluster, num_clusters,
3903 ref_root_bh, &meta_ac,
3910 handle = ocfs2_start_trans(osb, credits);
3911 if (IS_ERR(handle)) {
3912 ret = PTR_ERR(handle);
3917 ret = ocfs2_insert_extent(handle, et, cpos,
3918 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3919 num_clusters, ext_flags, meta_ac);
3925 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3926 p_cluster, num_clusters,
3932 ocfs2_commit_trans(osb, handle);
3935 ocfs2_free_alloc_context(meta_ac);
3939 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3940 struct buffer_head *s_bh,
3941 struct inode *t_inode,
3942 struct buffer_head *t_bh)
3946 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3947 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3948 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3950 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3952 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3953 if (IS_ERR(handle)) {
3954 ret = PTR_ERR(handle);
3959 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3960 OCFS2_JOURNAL_ACCESS_WRITE);
3966 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3967 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3968 le16_to_cpu(s_di->id2.i_data.id_count));
3969 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3970 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3971 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3972 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3974 ocfs2_journal_dirty(handle, t_bh);
3977 ocfs2_commit_trans(osb, handle);
3982 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3983 struct inode *t_inode,
3984 struct buffer_head *t_bh,
3985 struct ocfs2_caching_info *ref_ci,
3986 struct buffer_head *ref_root_bh,
3987 struct ocfs2_cached_dealloc_ctxt *dealloc)
3990 u32 p_cluster, num_clusters, clusters, cpos;
3992 unsigned int ext_flags;
3993 struct ocfs2_extent_tree et;
3995 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
3997 size = i_size_read(s_inode);
3998 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4001 while (cpos < clusters) {
4002 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4003 &num_clusters, &ext_flags);
4006 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4007 ref_ci, ref_root_bh,
4018 cpos += num_clusters;
4026 * change the new file's attributes to the src.
4028 * reflink creates a snapshot of a file, that means the attributes
4029 * must be identical except for three exceptions - nlink, ino, and ctime.
4031 static int ocfs2_complete_reflink(struct inode *s_inode,
4032 struct buffer_head *s_bh,
4033 struct inode *t_inode,
4034 struct buffer_head *t_bh,
4039 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4040 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4041 loff_t size = i_size_read(s_inode);
4043 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4044 OCFS2_INODE_UPDATE_CREDITS);
4045 if (IS_ERR(handle)) {
4046 ret = PTR_ERR(handle);
4051 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4052 OCFS2_JOURNAL_ACCESS_WRITE);
4058 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4059 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4060 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4061 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4062 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4063 i_size_write(t_inode, size);
4064 t_inode->i_blocks = s_inode->i_blocks;
4066 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4067 di->i_clusters = s_di->i_clusters;
4068 di->i_size = s_di->i_size;
4069 di->i_dyn_features = s_di->i_dyn_features;
4070 di->i_attr = s_di->i_attr;
4073 di->i_uid = s_di->i_uid;
4074 di->i_gid = s_di->i_gid;
4075 di->i_mode = s_di->i_mode;
4079 * we want mtime to appear identical to the source and
4082 t_inode->i_ctime = CURRENT_TIME;
4084 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4085 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4087 t_inode->i_mtime = s_inode->i_mtime;
4088 di->i_mtime = s_di->i_mtime;
4089 di->i_mtime_nsec = s_di->i_mtime_nsec;
4092 ocfs2_journal_dirty(handle, t_bh);
4095 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4099 static int ocfs2_create_reflink_node(struct inode *s_inode,
4100 struct buffer_head *s_bh,
4101 struct inode *t_inode,
4102 struct buffer_head *t_bh,
4106 struct buffer_head *ref_root_bh = NULL;
4107 struct ocfs2_cached_dealloc_ctxt dealloc;
4108 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4109 struct ocfs2_refcount_block *rb;
4110 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4111 struct ocfs2_refcount_tree *ref_tree;
4113 ocfs2_init_dealloc_ctxt(&dealloc);
4115 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4116 le64_to_cpu(di->i_refcount_loc));
4122 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4123 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4130 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4131 1, &ref_tree, &ref_root_bh);
4136 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4138 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4139 &ref_tree->rf_ci, ref_root_bh,
4143 goto out_unlock_refcount;
4146 out_unlock_refcount:
4147 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4148 brelse(ref_root_bh);
4150 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4151 ocfs2_schedule_truncate_log_flush(osb, 1);
4152 ocfs2_run_deallocs(osb, &dealloc);
4158 static int __ocfs2_reflink(struct dentry *old_dentry,
4159 struct buffer_head *old_bh,
4160 struct inode *new_inode,
4164 struct inode *inode = old_dentry->d_inode;
4165 struct buffer_head *new_bh = NULL;
4167 ret = filemap_fdatawrite(inode->i_mapping);
4173 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4179 mutex_lock(&new_inode->i_mutex);
4180 ret = ocfs2_inode_lock(new_inode, &new_bh, 1);
4186 ret = ocfs2_create_reflink_node(inode, old_bh,
4187 new_inode, new_bh, preserve);
4193 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4194 ret = ocfs2_reflink_xattrs(inode, old_bh,
4203 ret = ocfs2_complete_reflink(inode, old_bh,
4204 new_inode, new_bh, preserve);
4209 ocfs2_inode_unlock(new_inode, 1);
4212 mutex_unlock(&new_inode->i_mutex);
4215 ret = filemap_fdatawait(inode->i_mapping);
4222 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4223 struct dentry *new_dentry, bool preserve)
4226 struct inode *inode = old_dentry->d_inode;
4227 struct buffer_head *old_bh = NULL;
4228 struct inode *new_orphan_inode = NULL;
4230 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4233 error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4240 error = ocfs2_inode_lock(inode, &old_bh, 1);
4246 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4247 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4248 error = __ocfs2_reflink(old_dentry, old_bh,
4249 new_orphan_inode, preserve);
4250 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4251 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4253 ocfs2_inode_unlock(inode, 1);
4261 /* If the security isn't preserved, we need to re-initialize them. */
4263 error = ocfs2_init_security_and_acl(dir, new_orphan_inode);
4269 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4275 if (new_orphan_inode) {
4277 * We need to open_unlock the inode no matter whether we
4278 * succeed or not, so that other nodes can delete it later.
4280 ocfs2_open_unlock(new_orphan_inode);
4282 iput(new_orphan_inode);
4289 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4290 * sys_reflink(). This will go away when vfs_reflink() exists in
4294 /* copied from may_create in VFS. */
4295 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4299 if (IS_DEADDIR(dir))
4301 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4304 /* copied from user_path_parent. */
4305 static int ocfs2_user_path_parent(const char __user *path,
4306 struct nameidata *nd, char **name)
4308 char *s = getname(path);
4314 error = path_lookup(s, LOOKUP_PARENT, nd);
4324 * ocfs2_vfs_reflink - Create a reference-counted link
4326 * @old_dentry: source dentry + inode
4327 * @dir: directory to create the target
4328 * @new_dentry: target dentry
4329 * @preserve: if true, preserve all file attributes
4331 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4332 struct dentry *new_dentry, bool preserve)
4334 struct inode *inode = old_dentry->d_inode;
4340 error = ocfs2_may_create(dir, new_dentry);
4344 if (dir->i_sb != inode->i_sb)
4348 * A reflink to an append-only or immutable file cannot be created.
4350 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4353 /* Only regular files can be reflinked. */
4354 if (!S_ISREG(inode->i_mode))
4358 * If the caller wants to preserve ownership, they require the
4362 if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
4364 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4369 * If the caller is modifying any aspect of the attributes, they
4370 * are not creating a snapshot. They need read permission on the
4374 error = inode_permission(inode, MAY_READ);
4379 mutex_lock(&inode->i_mutex);
4380 dquot_initialize(dir);
4381 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4382 mutex_unlock(&inode->i_mutex);
4384 fsnotify_create(dir, new_dentry);
4388 * Most codes are copied from sys_linkat.
4390 int ocfs2_reflink_ioctl(struct inode *inode,
4391 const char __user *oldname,
4392 const char __user *newname,
4395 struct dentry *new_dentry;
4396 struct nameidata nd;
4397 struct path old_path;
4401 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4404 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4410 error = ocfs2_user_path_parent(newname, &nd, &to);
4417 if (old_path.mnt != nd.path.mnt)
4419 new_dentry = lookup_create(&nd, 0);
4420 error = PTR_ERR(new_dentry);
4421 if (IS_ERR(new_dentry)) {
4426 error = mnt_want_write(nd.path.mnt);
4432 error = ocfs2_vfs_reflink(old_path.dentry,
4433 nd.path.dentry->d_inode,
4434 new_dentry, preserve);
4435 mnt_drop_write(nd.path.mnt);
4439 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
4444 path_put(&old_path);