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;
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(osb, handle, meta_ac, 1,
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_bit = cpu_to_le16(suballoc_bit_start);
631 rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
632 rb->rf_blkno = cpu_to_le64(first_blkno);
633 rb->rf_count = cpu_to_le32(1);
634 rb->rf_records.rl_count =
635 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
636 spin_lock(&osb->osb_lock);
637 rb->rf_generation = osb->s_next_generation++;
638 spin_unlock(&osb->osb_lock);
640 ocfs2_journal_dirty(handle, new_bh);
642 spin_lock(&oi->ip_lock);
643 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
644 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
645 di->i_refcount_loc = cpu_to_le64(first_blkno);
646 spin_unlock(&oi->ip_lock);
648 mlog(0, "created tree for inode %lu, refblock %llu\n",
649 inode->i_ino, (unsigned long long)first_blkno);
651 ocfs2_journal_dirty(handle, di_bh);
654 * We have to init the tree lock here since it will use
655 * the generation number to create it.
657 new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
658 ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
659 new_tree->rf_generation);
661 spin_lock(&osb->osb_lock);
662 tree = ocfs2_find_refcount_tree(osb, first_blkno);
665 * We've just created a new refcount tree in this block. If
666 * we found a refcount tree on the ocfs2_super, it must be
667 * one we just deleted. We free the old tree before
668 * inserting the new tree.
670 BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
672 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
673 ocfs2_insert_refcount_tree(osb, new_tree);
674 spin_unlock(&osb->osb_lock);
677 ocfs2_refcount_tree_put(tree);
680 ocfs2_commit_trans(osb, handle);
684 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
690 ocfs2_free_alloc_context(meta_ac);
695 static int ocfs2_set_refcount_tree(struct inode *inode,
696 struct buffer_head *di_bh,
700 handle_t *handle = NULL;
701 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
702 struct ocfs2_inode_info *oi = OCFS2_I(inode);
703 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
704 struct buffer_head *ref_root_bh = NULL;
705 struct ocfs2_refcount_block *rb;
706 struct ocfs2_refcount_tree *ref_tree;
708 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
710 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
711 &ref_tree, &ref_root_bh);
717 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
718 if (IS_ERR(handle)) {
719 ret = PTR_ERR(handle);
724 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
725 OCFS2_JOURNAL_ACCESS_WRITE);
731 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
732 OCFS2_JOURNAL_ACCESS_WRITE);
738 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
739 le32_add_cpu(&rb->rf_count, 1);
741 ocfs2_journal_dirty(handle, ref_root_bh);
743 spin_lock(&oi->ip_lock);
744 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
745 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
746 di->i_refcount_loc = cpu_to_le64(refcount_loc);
747 spin_unlock(&oi->ip_lock);
748 ocfs2_journal_dirty(handle, di_bh);
751 ocfs2_commit_trans(osb, handle);
753 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
759 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
761 int ret, delete_tree = 0;
762 handle_t *handle = NULL;
763 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
764 struct ocfs2_inode_info *oi = OCFS2_I(inode);
765 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
766 struct ocfs2_refcount_block *rb;
767 struct inode *alloc_inode = NULL;
768 struct buffer_head *alloc_bh = NULL;
769 struct buffer_head *blk_bh = NULL;
770 struct ocfs2_refcount_tree *ref_tree;
771 int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
772 u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
775 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
779 ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
785 rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
788 * If we are the last user, we need to free the block.
789 * So lock the allocator ahead.
791 if (le32_to_cpu(rb->rf_count) == 1) {
792 blk = le64_to_cpu(rb->rf_blkno);
793 bit = le16_to_cpu(rb->rf_suballoc_bit);
794 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
796 alloc_inode = ocfs2_get_system_file_inode(osb,
797 EXTENT_ALLOC_SYSTEM_INODE,
798 le16_to_cpu(rb->rf_suballoc_slot));
804 mutex_lock(&alloc_inode->i_mutex);
806 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
812 credits += OCFS2_SUBALLOC_FREE;
815 handle = ocfs2_start_trans(osb, credits);
816 if (IS_ERR(handle)) {
817 ret = PTR_ERR(handle);
822 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
823 OCFS2_JOURNAL_ACCESS_WRITE);
829 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
830 OCFS2_JOURNAL_ACCESS_WRITE);
836 spin_lock(&oi->ip_lock);
837 oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
838 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
839 di->i_refcount_loc = 0;
840 spin_unlock(&oi->ip_lock);
841 ocfs2_journal_dirty(handle, di_bh);
843 le32_add_cpu(&rb->rf_count , -1);
844 ocfs2_journal_dirty(handle, blk_bh);
848 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
849 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
850 alloc_bh, bit, bg_blkno, 1);
856 ocfs2_commit_trans(osb, handle);
859 ocfs2_inode_unlock(alloc_inode, 1);
864 mutex_unlock(&alloc_inode->i_mutex);
868 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
870 ocfs2_refcount_tree_put(ref_tree);
876 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
877 struct buffer_head *ref_leaf_bh,
878 u64 cpos, unsigned int len,
879 struct ocfs2_refcount_rec *ret_rec,
883 struct ocfs2_refcount_block *rb =
884 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
885 struct ocfs2_refcount_rec *rec = NULL;
887 for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
888 rec = &rb->rf_records.rl_recs[i];
890 if (le64_to_cpu(rec->r_cpos) +
891 le32_to_cpu(rec->r_clusters) <= cpos)
893 else if (le64_to_cpu(rec->r_cpos) > cpos)
896 /* ok, cpos fail in this rec. Just return. */
903 /* We meet with a hole here, so fake the rec. */
904 ret_rec->r_cpos = cpu_to_le64(cpos);
905 ret_rec->r_refcount = 0;
906 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
907 le64_to_cpu(rec->r_cpos) < cpos + len)
908 ret_rec->r_clusters =
909 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
911 ret_rec->r_clusters = cpu_to_le32(len);
919 * Try to remove refcount tree. The mechanism is:
920 * 1) Check whether i_clusters == 0, if no, exit.
921 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
922 * 3) Check whether we have inline xattr stored outside, if yes, exit.
923 * 4) Remove the tree.
925 int ocfs2_try_remove_refcount_tree(struct inode *inode,
926 struct buffer_head *di_bh)
929 struct ocfs2_inode_info *oi = OCFS2_I(inode);
930 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
932 down_write(&oi->ip_xattr_sem);
933 down_write(&oi->ip_alloc_sem);
938 if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
941 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
942 ocfs2_has_inline_xattr_value_outside(inode, di))
945 ret = ocfs2_remove_refcount_tree(inode, di_bh);
949 up_write(&oi->ip_alloc_sem);
950 up_write(&oi->ip_xattr_sem);
955 * Find the end range for a leaf refcount block indicated by
956 * el->l_recs[index].e_blkno.
958 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
959 struct buffer_head *ref_root_bh,
960 struct ocfs2_extent_block *eb,
961 struct ocfs2_extent_list *el,
962 int index, u32 *cpos_end)
964 int ret, i, subtree_root;
967 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
968 struct ocfs2_path *left_path = NULL, *right_path = NULL;
969 struct ocfs2_extent_tree et;
970 struct ocfs2_extent_list *tmp_el;
972 if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
974 * We have a extent rec after index, so just use the e_cpos
975 * of the next extent rec.
977 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
981 if (!eb || (eb && !eb->h_next_leaf_blk)) {
983 * We are the last extent rec, so any high cpos should
984 * be stored in this leaf refcount block.
986 *cpos_end = UINT_MAX;
991 * If the extent block isn't the last one, we have to find
992 * the subtree root between this extent block and the next
993 * leaf extent block and get the corresponding e_cpos from
994 * the subroot. Otherwise we may corrupt the b-tree.
996 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
998 left_path = ocfs2_new_path_from_et(&et);
1005 cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1006 ret = ocfs2_find_path(ci, left_path, cpos);
1012 right_path = ocfs2_new_path_from_path(left_path);
1019 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1025 ret = ocfs2_find_path(ci, right_path, cpos);
1031 subtree_root = ocfs2_find_subtree_root(&et, left_path,
1034 tmp_el = left_path->p_node[subtree_root].el;
1035 blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1036 for (i = 0; i < le32_to_cpu(tmp_el->l_next_free_rec); i++) {
1037 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1038 *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1043 BUG_ON(i == le32_to_cpu(tmp_el->l_next_free_rec));
1046 ocfs2_free_path(left_path);
1047 ocfs2_free_path(right_path);
1052 * Given a cpos and len, try to find the refcount record which contains cpos.
1053 * 1. If cpos can be found in one refcount record, return the record.
1054 * 2. If cpos can't be found, return a fake record which start from cpos
1055 * and end at a small value between cpos+len and start of the next record.
1056 * This fake record has r_refcount = 0.
1058 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1059 struct buffer_head *ref_root_bh,
1060 u64 cpos, unsigned int len,
1061 struct ocfs2_refcount_rec *ret_rec,
1063 struct buffer_head **ret_bh)
1065 int ret = 0, i, found;
1066 u32 low_cpos, uninitialized_var(cpos_end);
1067 struct ocfs2_extent_list *el;
1068 struct ocfs2_extent_rec *rec = NULL;
1069 struct ocfs2_extent_block *eb = NULL;
1070 struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1071 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1072 struct ocfs2_refcount_block *rb =
1073 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1075 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1076 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1078 *ret_bh = ref_root_bh;
1079 get_bh(ref_root_bh);
1084 low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1086 if (el->l_tree_depth) {
1087 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1093 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1096 if (el->l_tree_depth) {
1098 "refcount tree %llu has non zero tree "
1099 "depth in leaf btree tree block %llu\n",
1100 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1101 (unsigned long long)eb_bh->b_blocknr);
1108 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1109 rec = &el->l_recs[i];
1111 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1118 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1119 eb, el, i, &cpos_end);
1125 if (cpos_end < low_cpos + len)
1126 len = cpos_end - low_cpos;
1129 ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1136 ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1138 *ret_bh = ref_leaf_bh;
1144 enum ocfs2_ref_rec_contig {
1145 REF_CONTIG_NONE = 0,
1148 REF_CONTIG_LEFTRIGHT,
1151 static enum ocfs2_ref_rec_contig
1152 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1155 if ((rb->rf_records.rl_recs[index].r_refcount ==
1156 rb->rf_records.rl_recs[index + 1].r_refcount) &&
1157 (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1158 le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1159 le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1160 return REF_CONTIG_RIGHT;
1162 return REF_CONTIG_NONE;
1165 static enum ocfs2_ref_rec_contig
1166 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1169 enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1171 if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1172 ret = ocfs2_refcount_rec_adjacent(rb, index);
1175 enum ocfs2_ref_rec_contig tmp;
1177 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1179 if (tmp == REF_CONTIG_RIGHT) {
1180 if (ret == REF_CONTIG_RIGHT)
1181 ret = REF_CONTIG_LEFTRIGHT;
1183 ret = REF_CONTIG_LEFT;
1190 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1193 BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1194 rb->rf_records.rl_recs[index+1].r_refcount);
1196 le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1197 le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1199 if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1200 memmove(&rb->rf_records.rl_recs[index + 1],
1201 &rb->rf_records.rl_recs[index + 2],
1202 sizeof(struct ocfs2_refcount_rec) *
1203 (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1205 memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1206 0, sizeof(struct ocfs2_refcount_rec));
1207 le16_add_cpu(&rb->rf_records.rl_used, -1);
1211 * Merge the refcount rec if we are contiguous with the adjacent recs.
1213 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1216 enum ocfs2_ref_rec_contig contig =
1217 ocfs2_refcount_rec_contig(rb, index);
1219 if (contig == REF_CONTIG_NONE)
1222 if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1227 ocfs2_rotate_refcount_rec_left(rb, index);
1229 if (contig == REF_CONTIG_LEFTRIGHT)
1230 ocfs2_rotate_refcount_rec_left(rb, index);
1234 * Change the refcount indexed by "index" in ref_bh.
1235 * If refcount reaches 0, remove it.
1237 static int ocfs2_change_refcount_rec(handle_t *handle,
1238 struct ocfs2_caching_info *ci,
1239 struct buffer_head *ref_leaf_bh,
1240 int index, int merge, int change)
1243 struct ocfs2_refcount_block *rb =
1244 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1245 struct ocfs2_refcount_list *rl = &rb->rf_records;
1246 struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1248 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1249 OCFS2_JOURNAL_ACCESS_WRITE);
1255 mlog(0, "change index %d, old count %u, change %d\n", index,
1256 le32_to_cpu(rec->r_refcount), change);
1257 le32_add_cpu(&rec->r_refcount, change);
1259 if (!rec->r_refcount) {
1260 if (index != le16_to_cpu(rl->rl_used) - 1) {
1261 memmove(rec, rec + 1,
1262 (le16_to_cpu(rl->rl_used) - index - 1) *
1263 sizeof(struct ocfs2_refcount_rec));
1264 memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1265 0, sizeof(struct ocfs2_refcount_rec));
1268 le16_add_cpu(&rl->rl_used, -1);
1270 ocfs2_refcount_rec_merge(rb, index);
1272 ocfs2_journal_dirty(handle, ref_leaf_bh);
1277 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1278 struct ocfs2_caching_info *ci,
1279 struct buffer_head *ref_root_bh,
1280 struct buffer_head **ref_leaf_bh,
1281 struct ocfs2_alloc_context *meta_ac)
1284 u16 suballoc_bit_start;
1287 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1288 struct buffer_head *new_bh = NULL;
1289 struct ocfs2_refcount_block *new_rb;
1290 struct ocfs2_refcount_block *root_rb =
1291 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1293 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1294 OCFS2_JOURNAL_ACCESS_WRITE);
1300 ret = ocfs2_claim_metadata(OCFS2_SB(sb), handle, meta_ac, 1,
1301 &suballoc_bit_start, &num_got,
1308 new_bh = sb_getblk(sb, blkno);
1309 if (new_bh == NULL) {
1314 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1316 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1317 OCFS2_JOURNAL_ACCESS_CREATE);
1324 * Initialize ocfs2_refcount_block.
1325 * It should contain the same information as the old root.
1326 * so just memcpy it and change the corresponding field.
1328 memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1330 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1331 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1332 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1333 new_rb->rf_blkno = cpu_to_le64(blkno);
1334 new_rb->rf_cpos = cpu_to_le32(0);
1335 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1336 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1337 ocfs2_journal_dirty(handle, new_bh);
1339 /* Now change the root. */
1340 memset(&root_rb->rf_list, 0, sb->s_blocksize -
1341 offsetof(struct ocfs2_refcount_block, rf_list));
1342 root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1343 root_rb->rf_clusters = cpu_to_le32(1);
1344 root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1345 root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1346 root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1347 root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1349 ocfs2_journal_dirty(handle, ref_root_bh);
1351 mlog(0, "new leaf block %llu, used %u\n", (unsigned long long)blkno,
1352 le16_to_cpu(new_rb->rf_records.rl_used));
1354 *ref_leaf_bh = new_bh;
1361 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1362 struct ocfs2_refcount_rec *next)
1364 if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1365 ocfs2_get_ref_rec_low_cpos(next))
1371 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1373 const struct ocfs2_refcount_rec *l = a, *r = b;
1374 u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1375 u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1377 if (l_cpos > r_cpos)
1379 if (l_cpos < r_cpos)
1384 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1386 const struct ocfs2_refcount_rec *l = a, *r = b;
1387 u64 l_cpos = le64_to_cpu(l->r_cpos);
1388 u64 r_cpos = le64_to_cpu(r->r_cpos);
1390 if (l_cpos > r_cpos)
1392 if (l_cpos < r_cpos)
1397 static void swap_refcount_rec(void *a, void *b, int size)
1399 struct ocfs2_refcount_rec *l = a, *r = b, tmp;
1401 tmp = *(struct ocfs2_refcount_rec *)l;
1402 *(struct ocfs2_refcount_rec *)l =
1403 *(struct ocfs2_refcount_rec *)r;
1404 *(struct ocfs2_refcount_rec *)r = tmp;
1408 * The refcount cpos are ordered by their 64bit cpos,
1409 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1410 * So we need to make sure that this pos isn't intersected with others.
1412 * Note: The refcount block is already sorted by their low 32 bit cpos,
1413 * So just try the middle pos first, and we will exit when we find
1414 * the good position.
1416 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1417 u32 *split_pos, int *split_index)
1419 int num_used = le16_to_cpu(rl->rl_used);
1420 int delta, middle = num_used / 2;
1422 for (delta = 0; delta < middle; delta++) {
1423 /* Let's check delta earlier than middle */
1424 if (ocfs2_refcount_rec_no_intersect(
1425 &rl->rl_recs[middle - delta - 1],
1426 &rl->rl_recs[middle - delta])) {
1427 *split_index = middle - delta;
1431 /* For even counts, don't walk off the end */
1432 if ((middle + delta + 1) == num_used)
1435 /* Now try delta past middle */
1436 if (ocfs2_refcount_rec_no_intersect(
1437 &rl->rl_recs[middle + delta],
1438 &rl->rl_recs[middle + delta + 1])) {
1439 *split_index = middle + delta + 1;
1444 if (delta >= middle)
1447 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1451 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1452 struct buffer_head *new_bh,
1455 int split_index = 0, num_moved, ret;
1457 struct ocfs2_refcount_block *rb =
1458 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1459 struct ocfs2_refcount_list *rl = &rb->rf_records;
1460 struct ocfs2_refcount_block *new_rb =
1461 (struct ocfs2_refcount_block *)new_bh->b_data;
1462 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1464 mlog(0, "split old leaf refcount block %llu, count = %u, used = %u\n",
1465 (unsigned long long)ref_leaf_bh->b_blocknr,
1466 le32_to_cpu(rl->rl_count), le32_to_cpu(rl->rl_used));
1469 * XXX: Improvement later.
1470 * If we know all the high 32 bit cpos is the same, no need to sort.
1472 * In order to make the whole process safe, we do:
1473 * 1. sort the entries by their low 32 bit cpos first so that we can
1474 * find the split cpos easily.
1475 * 2. call ocfs2_insert_extent to insert the new refcount block.
1476 * 3. move the refcount rec to the new block.
1477 * 4. sort the entries by their 64 bit cpos.
1478 * 5. dirty the new_rb and rb.
1480 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1481 sizeof(struct ocfs2_refcount_rec),
1482 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1484 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1490 new_rb->rf_cpos = cpu_to_le32(cpos);
1492 /* move refcount records starting from split_index to the new block. */
1493 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1494 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1495 num_moved * sizeof(struct ocfs2_refcount_rec));
1497 /*ok, remove the entries we just moved over to the other block. */
1498 memset(&rl->rl_recs[split_index], 0,
1499 num_moved * sizeof(struct ocfs2_refcount_rec));
1501 /* change old and new rl_used accordingly. */
1502 le16_add_cpu(&rl->rl_used, -num_moved);
1503 new_rl->rl_used = cpu_to_le16(num_moved);
1505 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1506 sizeof(struct ocfs2_refcount_rec),
1507 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1509 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1510 sizeof(struct ocfs2_refcount_rec),
1511 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1517 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1518 struct ocfs2_caching_info *ci,
1519 struct buffer_head *ref_root_bh,
1520 struct buffer_head *ref_leaf_bh,
1521 struct ocfs2_alloc_context *meta_ac)
1524 u16 suballoc_bit_start;
1525 u32 num_got, new_cpos;
1527 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1528 struct ocfs2_refcount_block *root_rb =
1529 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1530 struct buffer_head *new_bh = NULL;
1531 struct ocfs2_refcount_block *new_rb;
1532 struct ocfs2_extent_tree ref_et;
1534 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1536 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1537 OCFS2_JOURNAL_ACCESS_WRITE);
1543 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1544 OCFS2_JOURNAL_ACCESS_WRITE);
1550 ret = ocfs2_claim_metadata(OCFS2_SB(sb), handle, meta_ac, 1,
1551 &suballoc_bit_start, &num_got,
1558 new_bh = sb_getblk(sb, blkno);
1559 if (new_bh == NULL) {
1564 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1566 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1567 OCFS2_JOURNAL_ACCESS_CREATE);
1573 /* Initialize ocfs2_refcount_block. */
1574 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1575 memset(new_rb, 0, sb->s_blocksize);
1576 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1577 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1578 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1579 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1580 new_rb->rf_blkno = cpu_to_le64(blkno);
1581 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1582 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1583 new_rb->rf_records.rl_count =
1584 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1585 new_rb->rf_generation = root_rb->rf_generation;
1587 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1593 ocfs2_journal_dirty(handle, ref_leaf_bh);
1594 ocfs2_journal_dirty(handle, new_bh);
1596 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1598 mlog(0, "insert new leaf block %llu at %u\n",
1599 (unsigned long long)new_bh->b_blocknr, new_cpos);
1601 /* Insert the new leaf block with the specific offset cpos. */
1602 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1612 static int ocfs2_expand_refcount_tree(handle_t *handle,
1613 struct ocfs2_caching_info *ci,
1614 struct buffer_head *ref_root_bh,
1615 struct buffer_head *ref_leaf_bh,
1616 struct ocfs2_alloc_context *meta_ac)
1619 struct buffer_head *expand_bh = NULL;
1621 if (ref_root_bh == ref_leaf_bh) {
1623 * the old root bh hasn't been expanded to a b-tree,
1624 * so expand it first.
1626 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1627 &expand_bh, meta_ac);
1633 expand_bh = ref_leaf_bh;
1638 /* Now add a new refcount block into the tree.*/
1639 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1640 expand_bh, meta_ac);
1649 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1651 * Only called when we have inserted a new refcount rec at index 0
1652 * which means ocfs2_extent_rec.e_cpos may need some change.
1654 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1655 struct ocfs2_caching_info *ci,
1656 struct buffer_head *ref_root_bh,
1657 struct buffer_head *ref_leaf_bh,
1658 struct ocfs2_refcount_rec *rec)
1661 u32 new_cpos, old_cpos;
1662 struct ocfs2_path *path = NULL;
1663 struct ocfs2_extent_tree et;
1664 struct ocfs2_refcount_block *rb =
1665 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1666 struct ocfs2_extent_list *el;
1668 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1671 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1672 old_cpos = le32_to_cpu(rb->rf_cpos);
1673 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1674 if (old_cpos <= new_cpos)
1677 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1679 path = ocfs2_new_path_from_et(&et);
1686 ret = ocfs2_find_path(ci, path, old_cpos);
1693 * 2 more credits, one for the leaf refcount block, one for
1694 * the extent block contains the extent rec.
1696 ret = ocfs2_extend_trans(handle, handle->h_buffer_credits + 2);
1702 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1703 OCFS2_JOURNAL_ACCESS_WRITE);
1709 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1710 OCFS2_JOURNAL_ACCESS_WRITE);
1716 /* change the leaf extent block first. */
1717 el = path_leaf_el(path);
1719 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1720 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1723 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1725 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1727 /* change the r_cpos in the leaf block. */
1728 rb->rf_cpos = cpu_to_le32(new_cpos);
1730 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1731 ocfs2_journal_dirty(handle, ref_leaf_bh);
1734 ocfs2_free_path(path);
1738 static int ocfs2_insert_refcount_rec(handle_t *handle,
1739 struct ocfs2_caching_info *ci,
1740 struct buffer_head *ref_root_bh,
1741 struct buffer_head *ref_leaf_bh,
1742 struct ocfs2_refcount_rec *rec,
1743 int index, int merge,
1744 struct ocfs2_alloc_context *meta_ac)
1747 struct ocfs2_refcount_block *rb =
1748 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1749 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1750 struct buffer_head *new_bh = NULL;
1752 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1754 if (rf_list->rl_used == rf_list->rl_count) {
1755 u64 cpos = le64_to_cpu(rec->r_cpos);
1756 u32 len = le32_to_cpu(rec->r_clusters);
1758 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1759 ref_leaf_bh, meta_ac);
1765 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1766 cpos, len, NULL, &index,
1773 ref_leaf_bh = new_bh;
1774 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1775 rf_list = &rb->rf_records;
1778 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1779 OCFS2_JOURNAL_ACCESS_WRITE);
1785 if (index < le16_to_cpu(rf_list->rl_used))
1786 memmove(&rf_list->rl_recs[index + 1],
1787 &rf_list->rl_recs[index],
1788 (le16_to_cpu(rf_list->rl_used) - index) *
1789 sizeof(struct ocfs2_refcount_rec));
1791 mlog(0, "insert refcount record start %llu, len %u, count %u "
1792 "to leaf block %llu at index %d\n",
1793 (unsigned long long)le64_to_cpu(rec->r_cpos),
1794 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount),
1795 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1797 rf_list->rl_recs[index] = *rec;
1799 le16_add_cpu(&rf_list->rl_used, 1);
1802 ocfs2_refcount_rec_merge(rb, index);
1804 ocfs2_journal_dirty(handle, ref_leaf_bh);
1807 ret = ocfs2_adjust_refcount_rec(handle, ci,
1819 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1820 * This is much simple than our b-tree code.
1821 * split_rec is the new refcount rec we want to insert.
1822 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1823 * increase refcount or decrease a refcount to non-zero).
1824 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1825 * rec( in case we decrease a refcount to zero).
1827 static int ocfs2_split_refcount_rec(handle_t *handle,
1828 struct ocfs2_caching_info *ci,
1829 struct buffer_head *ref_root_bh,
1830 struct buffer_head *ref_leaf_bh,
1831 struct ocfs2_refcount_rec *split_rec,
1832 int index, int merge,
1833 struct ocfs2_alloc_context *meta_ac,
1834 struct ocfs2_cached_dealloc_ctxt *dealloc)
1838 struct ocfs2_refcount_block *rb =
1839 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1840 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1841 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1842 struct ocfs2_refcount_rec *tail_rec = NULL;
1843 struct buffer_head *new_bh = NULL;
1845 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1847 mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
1848 le64_to_cpu(orig_rec->r_cpos), le32_to_cpu(orig_rec->r_clusters),
1849 le64_to_cpu(split_rec->r_cpos),
1850 le32_to_cpu(split_rec->r_clusters));
1853 * If we just need to split the header or tail clusters,
1854 * no more recs are needed, just split is OK.
1855 * Otherwise we at least need one new recs.
1857 if (!split_rec->r_refcount &&
1858 (split_rec->r_cpos == orig_rec->r_cpos ||
1859 le64_to_cpu(split_rec->r_cpos) +
1860 le32_to_cpu(split_rec->r_clusters) ==
1861 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1867 * We need one more rec if we split in the middle and the new rec have
1868 * some refcount in it.
1870 if (split_rec->r_refcount &&
1871 (split_rec->r_cpos != orig_rec->r_cpos &&
1872 le64_to_cpu(split_rec->r_cpos) +
1873 le32_to_cpu(split_rec->r_clusters) !=
1874 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1877 /* If the leaf block don't have enough record, expand it. */
1878 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1879 le16_to_cpu(rf_list->rl_count)) {
1880 struct ocfs2_refcount_rec tmp_rec;
1881 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1882 len = le32_to_cpu(orig_rec->r_clusters);
1883 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1884 ref_leaf_bh, meta_ac);
1891 * We have to re-get it since now cpos may be moved to
1892 * another leaf block.
1894 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1895 cpos, len, &tmp_rec, &index,
1902 ref_leaf_bh = new_bh;
1903 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1904 rf_list = &rb->rf_records;
1905 orig_rec = &rf_list->rl_recs[index];
1908 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1909 OCFS2_JOURNAL_ACCESS_WRITE);
1916 * We have calculated out how many new records we need and store
1917 * in recs_need, so spare enough space first by moving the records
1918 * after "index" to the end.
1920 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1921 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1922 &rf_list->rl_recs[index + 1],
1923 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1924 sizeof(struct ocfs2_refcount_rec));
1926 len = (le64_to_cpu(orig_rec->r_cpos) +
1927 le32_to_cpu(orig_rec->r_clusters)) -
1928 (le64_to_cpu(split_rec->r_cpos) +
1929 le32_to_cpu(split_rec->r_clusters));
1932 * If we have "len", the we will split in the tail and move it
1933 * to the end of the space we have just spared.
1936 tail_rec = &rf_list->rl_recs[index + recs_need];
1938 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1939 le64_add_cpu(&tail_rec->r_cpos,
1940 le32_to_cpu(tail_rec->r_clusters) - len);
1941 tail_rec->r_clusters = cpu_to_le32(len);
1945 * If the split pos isn't the same as the original one, we need to
1946 * split in the head.
1948 * Note: We have the chance that split_rec.r_refcount = 0,
1949 * recs_need = 0 and len > 0, which means we just cut the head from
1950 * the orig_rec and in that case we have done some modification in
1951 * orig_rec above, so the check for r_cpos is faked.
1953 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1954 len = le64_to_cpu(split_rec->r_cpos) -
1955 le64_to_cpu(orig_rec->r_cpos);
1956 orig_rec->r_clusters = cpu_to_le32(len);
1960 le16_add_cpu(&rf_list->rl_used, recs_need);
1962 if (split_rec->r_refcount) {
1963 rf_list->rl_recs[index] = *split_rec;
1964 mlog(0, "insert refcount record start %llu, len %u, count %u "
1965 "to leaf block %llu at index %d\n",
1966 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1967 le32_to_cpu(split_rec->r_clusters),
1968 le32_to_cpu(split_rec->r_refcount),
1969 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1972 ocfs2_refcount_rec_merge(rb, index);
1975 ocfs2_journal_dirty(handle, ref_leaf_bh);
1982 static int __ocfs2_increase_refcount(handle_t *handle,
1983 struct ocfs2_caching_info *ci,
1984 struct buffer_head *ref_root_bh,
1985 u64 cpos, u32 len, int merge,
1986 struct ocfs2_alloc_context *meta_ac,
1987 struct ocfs2_cached_dealloc_ctxt *dealloc)
1990 struct buffer_head *ref_leaf_bh = NULL;
1991 struct ocfs2_refcount_rec rec;
1992 unsigned int set_len = 0;
1994 mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
1995 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1996 (unsigned long long)cpos, len);
1999 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2000 cpos, len, &rec, &index,
2007 set_len = le32_to_cpu(rec.r_clusters);
2010 * Here we may meet with 3 situations:
2012 * 1. If we find an already existing record, and the length
2013 * is the same, cool, we just need to increase the r_refcount
2015 * 2. If we find a hole, just insert it with r_refcount = 1.
2016 * 3. If we are in the middle of one extent record, split
2019 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2021 mlog(0, "increase refcount rec, start %llu, len %u, "
2022 "count %u\n", (unsigned long long)cpos, set_len,
2023 le32_to_cpu(rec.r_refcount));
2024 ret = ocfs2_change_refcount_rec(handle, ci,
2031 } else if (!rec.r_refcount) {
2032 rec.r_refcount = cpu_to_le32(1);
2034 mlog(0, "insert refcount rec, start %llu, len %u\n",
2035 (unsigned long long)le64_to_cpu(rec.r_cpos),
2037 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2046 set_len = min((u64)(cpos + len),
2047 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2048 rec.r_cpos = cpu_to_le64(cpos);
2049 rec.r_clusters = cpu_to_le32(set_len);
2050 le32_add_cpu(&rec.r_refcount, 1);
2052 mlog(0, "split refcount rec, start %llu, "
2053 "len %u, count %u\n",
2054 (unsigned long long)le64_to_cpu(rec.r_cpos),
2055 set_len, le32_to_cpu(rec.r_refcount));
2056 ret = ocfs2_split_refcount_rec(handle, ci,
2057 ref_root_bh, ref_leaf_bh,
2068 brelse(ref_leaf_bh);
2073 brelse(ref_leaf_bh);
2077 static int ocfs2_remove_refcount_extent(handle_t *handle,
2078 struct ocfs2_caching_info *ci,
2079 struct buffer_head *ref_root_bh,
2080 struct buffer_head *ref_leaf_bh,
2081 struct ocfs2_alloc_context *meta_ac,
2082 struct ocfs2_cached_dealloc_ctxt *dealloc)
2085 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2086 struct ocfs2_refcount_block *rb =
2087 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2088 struct ocfs2_extent_tree et;
2090 BUG_ON(rb->rf_records.rl_used);
2092 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2093 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2094 1, meta_ac, dealloc);
2100 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2103 * add the freed block to the dealloc so that it will be freed
2104 * when we run dealloc.
2106 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2107 le16_to_cpu(rb->rf_suballoc_slot),
2108 le64_to_cpu(rb->rf_blkno),
2109 le16_to_cpu(rb->rf_suballoc_bit));
2115 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2116 OCFS2_JOURNAL_ACCESS_WRITE);
2122 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2124 le32_add_cpu(&rb->rf_clusters, -1);
2127 * check whether we need to restore the root refcount block if
2128 * there is no leaf extent block at atll.
2130 if (!rb->rf_list.l_next_free_rec) {
2131 BUG_ON(rb->rf_clusters);
2133 mlog(0, "reset refcount tree root %llu to be a record block.\n",
2134 (unsigned long long)ref_root_bh->b_blocknr);
2139 memset(&rb->rf_records, 0, sb->s_blocksize -
2140 offsetof(struct ocfs2_refcount_block, rf_records));
2141 rb->rf_records.rl_count =
2142 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2145 ocfs2_journal_dirty(handle, ref_root_bh);
2151 int ocfs2_increase_refcount(handle_t *handle,
2152 struct ocfs2_caching_info *ci,
2153 struct buffer_head *ref_root_bh,
2155 struct ocfs2_alloc_context *meta_ac,
2156 struct ocfs2_cached_dealloc_ctxt *dealloc)
2158 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2163 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2164 struct ocfs2_caching_info *ci,
2165 struct buffer_head *ref_root_bh,
2166 struct buffer_head *ref_leaf_bh,
2167 int index, u64 cpos, unsigned int len,
2168 struct ocfs2_alloc_context *meta_ac,
2169 struct ocfs2_cached_dealloc_ctxt *dealloc)
2172 struct ocfs2_refcount_block *rb =
2173 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2174 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2176 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2178 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2180 if (cpos == le64_to_cpu(rec->r_cpos) &&
2181 len == le32_to_cpu(rec->r_clusters))
2182 ret = ocfs2_change_refcount_rec(handle, ci,
2183 ref_leaf_bh, index, 1, -1);
2185 struct ocfs2_refcount_rec split = *rec;
2186 split.r_cpos = cpu_to_le64(cpos);
2187 split.r_clusters = cpu_to_le32(len);
2189 le32_add_cpu(&split.r_refcount, -1);
2191 mlog(0, "split refcount rec, start %llu, "
2192 "len %u, count %u, original start %llu, len %u\n",
2193 (unsigned long long)le64_to_cpu(split.r_cpos),
2194 len, le32_to_cpu(split.r_refcount),
2195 (unsigned long long)le64_to_cpu(rec->r_cpos),
2196 le32_to_cpu(rec->r_clusters));
2197 ret = ocfs2_split_refcount_rec(handle, ci,
2198 ref_root_bh, ref_leaf_bh,
2208 /* Remove the leaf refcount block if it contains no refcount record. */
2209 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2210 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2211 ref_leaf_bh, meta_ac,
2221 static int __ocfs2_decrease_refcount(handle_t *handle,
2222 struct ocfs2_caching_info *ci,
2223 struct buffer_head *ref_root_bh,
2225 struct ocfs2_alloc_context *meta_ac,
2226 struct ocfs2_cached_dealloc_ctxt *dealloc,
2229 int ret = 0, index = 0;
2230 struct ocfs2_refcount_rec rec;
2231 unsigned int r_count = 0, r_len;
2232 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2233 struct buffer_head *ref_leaf_bh = NULL;
2235 mlog(0, "Tree owner %llu, decrease refcount start %llu, "
2236 "len %u, delete %u\n",
2237 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2238 (unsigned long long)cpos, len, delete);
2241 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2242 cpos, len, &rec, &index,
2249 r_count = le32_to_cpu(rec.r_refcount);
2250 BUG_ON(r_count == 0);
2252 BUG_ON(r_count > 1);
2254 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2255 le32_to_cpu(rec.r_clusters)) - cpos;
2257 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2266 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2267 ret = ocfs2_cache_cluster_dealloc(dealloc,
2268 ocfs2_clusters_to_blocks(sb, cpos),
2278 brelse(ref_leaf_bh);
2283 brelse(ref_leaf_bh);
2287 /* Caller must hold refcount tree lock. */
2288 int ocfs2_decrease_refcount(struct inode *inode,
2289 handle_t *handle, u32 cpos, u32 len,
2290 struct ocfs2_alloc_context *meta_ac,
2291 struct ocfs2_cached_dealloc_ctxt *dealloc,
2296 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2297 struct buffer_head *ref_root_bh = NULL;
2298 struct ocfs2_refcount_tree *tree;
2300 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2302 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2308 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2314 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2321 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2322 cpos, len, meta_ac, dealloc, delete);
2326 brelse(ref_root_bh);
2331 * Mark the already-existing extent at cpos as refcounted for len clusters.
2332 * This adds the refcount extent flag.
2334 * If the existing extent is larger than the request, initiate a
2335 * split. An attempt will be made at merging with adjacent extents.
2337 * The caller is responsible for passing down meta_ac if we'll need it.
2339 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2340 struct ocfs2_extent_tree *et,
2341 handle_t *handle, u32 cpos,
2343 struct ocfs2_alloc_context *meta_ac,
2344 struct ocfs2_cached_dealloc_ctxt *dealloc)
2348 mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
2349 inode->i_ino, cpos, len, phys);
2351 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2352 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2353 "tree, but the feature bit is not set in the "
2354 "super block.", inode->i_ino);
2359 ret = ocfs2_change_extent_flag(handle, et, cpos,
2360 len, phys, meta_ac, dealloc,
2361 OCFS2_EXT_REFCOUNTED, 0);
2370 * Given some contiguous physical clusters, calculate what we need
2371 * for modifying their refcount.
2373 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2374 struct ocfs2_caching_info *ci,
2375 struct buffer_head *ref_root_bh,
2381 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2382 u64 cpos = start_cpos;
2383 struct ocfs2_refcount_block *rb;
2384 struct ocfs2_refcount_rec rec;
2385 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2388 mlog(0, "start_cpos %llu, clusters %u\n",
2389 (unsigned long long)start_cpos, clusters);
2391 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2392 cpos, clusters, &rec,
2393 &index, &ref_leaf_bh);
2399 if (ref_leaf_bh != prev_bh) {
2401 * Now we encounter a new leaf block, so calculate
2402 * whether we need to extend the old leaf.
2405 rb = (struct ocfs2_refcount_block *)
2408 if (le64_to_cpu(rb->rf_records.rl_used) +
2410 le16_to_cpu(rb->rf_records.rl_count))
2417 prev_bh = ref_leaf_bh;
2421 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2423 mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
2424 "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
2425 recs_add, (unsigned long long)cpos, clusters,
2426 (unsigned long long)le64_to_cpu(rec.r_cpos),
2427 le32_to_cpu(rec.r_clusters),
2428 le32_to_cpu(rec.r_refcount), index);
2430 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2431 le32_to_cpu(rec.r_clusters)) - cpos;
2433 * If the refcount rec already exist, cool. We just need
2434 * to check whether there is a split. Otherwise we just need
2435 * to increase the refcount.
2436 * If we will insert one, increases recs_add.
2438 * We record all the records which will be inserted to the
2439 * same refcount block, so that we can tell exactly whether
2440 * we need a new refcount block or not.
2442 if (rec.r_refcount) {
2443 /* Check whether we need a split at the beginning. */
2444 if (cpos == start_cpos &&
2445 cpos != le64_to_cpu(rec.r_cpos))
2448 /* Check whether we need a split in the end. */
2449 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2450 le32_to_cpu(rec.r_clusters))
2455 brelse(ref_leaf_bh);
2462 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2464 if (le64_to_cpu(rb->rf_records.rl_used) + recs_add >
2465 le16_to_cpu(rb->rf_records.rl_count))
2474 mlog(0, "we need ref_blocks %d\n", ref_blocks);
2475 *meta_add += ref_blocks;
2476 *credits += ref_blocks;
2479 * So we may need ref_blocks to insert into the tree.
2480 * That also means we need to change the b-tree and add that number
2481 * of records since we never merge them.
2482 * We need one more block for expansion since the new created leaf
2483 * block is also full and needs split.
2485 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2486 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2487 struct ocfs2_extent_tree et;
2489 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2490 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2491 *credits += ocfs2_calc_extend_credits(sb,
2495 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2500 brelse(ref_leaf_bh);
2506 * For refcount tree, we will decrease some contiguous clusters
2507 * refcount count, so just go through it to see how many blocks
2508 * we gonna touch and whether we need to create new blocks.
2510 * Normally the refcount blocks store these refcount should be
2511 * contiguous also, so that we can get the number easily.
2512 * As for meta_ac, we will at most add split 2 refcount record and
2513 * 2 more refcount block, so just check it in a rough way.
2515 * Caller must hold refcount tree lock.
2517 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2518 struct buffer_head *di_bh,
2522 struct ocfs2_alloc_context **meta_ac)
2524 int ret, ref_blocks = 0;
2525 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2526 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2527 struct buffer_head *ref_root_bh = NULL;
2528 struct ocfs2_refcount_tree *tree;
2529 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2531 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2532 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2533 "tree, but the feature bit is not set in the "
2534 "super block.", inode->i_ino);
2539 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2541 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2542 le64_to_cpu(di->i_refcount_loc), &tree);
2548 ret = ocfs2_read_refcount_block(&tree->rf_ci,
2549 le64_to_cpu(di->i_refcount_loc),
2556 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2559 start_cpos, clusters,
2560 &ref_blocks, credits);
2566 mlog(0, "reserve new metadata %d, credits = %d\n",
2567 ref_blocks, *credits);
2570 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2571 ref_blocks, meta_ac);
2577 brelse(ref_root_bh);
2581 #define MAX_CONTIG_BYTES 1048576
2583 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2585 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2588 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2590 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2594 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2595 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2596 * while still being less than or equal to it.
2598 * The goal is to break the extent at a multiple of contig_clusters.
2600 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2604 BUG_ON(start > cpos);
2606 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2610 * Given a cluster count of len, pad it out so that it is a multiple
2611 * of contig_clusters.
2613 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2616 unsigned int padded =
2617 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2618 ocfs2_cow_contig_mask(sb);
2628 * Calculate out the start and number of virtual clusters we need to to CoW.
2630 * cpos is vitual start cluster position we want to do CoW in a
2631 * file and write_len is the cluster length.
2632 * max_cpos is the place where we want to stop CoW intentionally.
2634 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2635 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2636 * get good I/O from the resulting extent tree.
2638 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2639 struct ocfs2_extent_list *el,
2647 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2648 struct buffer_head *eb_bh = NULL;
2649 struct ocfs2_extent_block *eb = NULL;
2650 struct ocfs2_extent_rec *rec;
2651 unsigned int want_clusters, rec_end = 0;
2652 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2655 BUG_ON(cpos + write_len > max_cpos);
2657 if (tree_height > 0) {
2658 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2664 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2667 if (el->l_tree_depth) {
2668 ocfs2_error(inode->i_sb,
2669 "Inode %lu has non zero tree depth in "
2670 "leaf block %llu\n", inode->i_ino,
2671 (unsigned long long)eb_bh->b_blocknr);
2678 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2679 rec = &el->l_recs[i];
2681 if (ocfs2_is_empty_extent(rec)) {
2682 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2683 "index %d\n", inode->i_ino, i);
2687 if (le32_to_cpu(rec->e_cpos) +
2688 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2691 if (*cow_len == 0) {
2693 * We should find a refcounted record in the
2696 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2697 *cow_start = le32_to_cpu(rec->e_cpos);
2701 * If we encounter a hole, a non-refcounted record or
2702 * pass the max_cpos, stop the search.
2704 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2705 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2706 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2709 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2710 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2711 if (rec_end > max_cpos) {
2713 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2717 * How many clusters do we actually need from
2718 * this extent? First we see how many we actually
2719 * need to complete the write. If that's smaller
2720 * than contig_clusters, we try for contig_clusters.
2723 want_clusters = write_len;
2725 want_clusters = (cpos + write_len) -
2726 (*cow_start + *cow_len);
2727 if (want_clusters < contig_clusters)
2728 want_clusters = contig_clusters;
2731 * If the write does not cover the whole extent, we
2732 * need to calculate how we're going to split the extent.
2733 * We try to do it on contig_clusters boundaries.
2735 * Any extent smaller than contig_clusters will be
2736 * CoWed in its entirety.
2738 if (leaf_clusters <= contig_clusters)
2739 *cow_len += leaf_clusters;
2740 else if (*cow_len || (*cow_start == cpos)) {
2742 * This extent needs to be CoW'd from its
2743 * beginning, so all we have to do is compute
2744 * how many clusters to grab. We align
2745 * want_clusters to the edge of contig_clusters
2746 * to get better I/O.
2748 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2751 if (leaf_clusters < want_clusters)
2752 *cow_len += leaf_clusters;
2754 *cow_len += want_clusters;
2755 } else if ((*cow_start + contig_clusters) >=
2756 (cpos + write_len)) {
2758 * Breaking off contig_clusters at the front
2759 * of the extent will cover our write. That's
2762 *cow_len = contig_clusters;
2763 } else if ((rec_end - cpos) <= contig_clusters) {
2765 * Breaking off contig_clusters at the tail of
2766 * this extent will cover cpos.
2768 *cow_start = rec_end - contig_clusters;
2769 *cow_len = contig_clusters;
2770 } else if ((rec_end - cpos) <= want_clusters) {
2772 * While we can't fit the entire write in this
2773 * extent, we know that the write goes from cpos
2774 * to the end of the extent. Break that off.
2775 * We try to break it at some multiple of
2776 * contig_clusters from the front of the extent.
2777 * Failing that (ie, cpos is within
2778 * contig_clusters of the front), we'll CoW the
2781 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2783 *cow_len = rec_end - *cow_start;
2786 * Ok, the entire write lives in the middle of
2787 * this extent. Let's try to slice the extent up
2788 * nicely. Optimally, our CoW region starts at
2789 * m*contig_clusters from the beginning of the
2790 * extent and goes for n*contig_clusters,
2791 * covering the entire write.
2793 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2796 want_clusters = (cpos + write_len) - *cow_start;
2797 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2799 if (*cow_start + want_clusters <= rec_end)
2800 *cow_len = want_clusters;
2802 *cow_len = rec_end - *cow_start;
2805 /* Have we covered our entire write yet? */
2806 if ((*cow_start + *cow_len) >= (cpos + write_len))
2810 * If we reach the end of the extent block and don't get enough
2811 * clusters, continue with the next extent block if possible.
2813 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2814 eb && eb->h_next_leaf_blk) {
2818 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2819 le64_to_cpu(eb->h_next_leaf_blk),
2826 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2838 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2839 * num_clusters in data_tree "et" and change the refcount for the old
2840 * clusters(starting form p_cluster) in the refcount tree.
2843 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2844 * more new leaf records.
2845 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2846 * just give data_ac = NULL.
2848 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2849 u32 p_cluster, u32 num_clusters,
2850 struct ocfs2_extent_tree *et,
2851 struct ocfs2_caching_info *ref_ci,
2852 struct buffer_head *ref_root_bh,
2853 struct ocfs2_alloc_context **meta_ac,
2854 struct ocfs2_alloc_context **data_ac,
2857 int ret = 0, meta_add = 0;
2858 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2860 if (num_free_extents < 0) {
2861 ret = num_free_extents;
2866 if (num_free_extents < num_clusters + 2)
2868 ocfs2_extend_meta_needed(et->et_root_el);
2870 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
2873 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2874 p_cluster, num_clusters,
2875 &meta_add, credits);
2881 mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
2882 meta_add, num_clusters, *credits);
2883 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2891 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2900 ocfs2_free_alloc_context(*meta_ac);
2908 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2910 BUG_ON(buffer_dirty(bh));
2912 clear_buffer_mapped(bh);
2917 static int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2918 struct ocfs2_cow_context *context,
2919 u32 cpos, u32 old_cluster,
2920 u32 new_cluster, u32 new_len)
2922 int ret = 0, partial;
2923 struct ocfs2_caching_info *ci = context->data_et.et_ci;
2924 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2925 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2928 unsigned int from, to;
2929 loff_t offset, end, map_end;
2930 struct address_space *mapping = context->inode->i_mapping;
2932 mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster,
2933 new_cluster, new_len, cpos);
2935 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2936 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2938 while (offset < end) {
2939 page_index = offset >> PAGE_CACHE_SHIFT;
2940 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2944 /* from, to is the offset within the page. */
2945 from = offset & (PAGE_CACHE_SIZE - 1);
2946 to = PAGE_CACHE_SIZE;
2947 if (map_end & (PAGE_CACHE_SIZE - 1))
2948 to = map_end & (PAGE_CACHE_SIZE - 1);
2950 page = grab_cache_page(mapping, page_index);
2953 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2954 * can't be dirtied before we CoW it out.
2956 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2957 BUG_ON(PageDirty(page));
2959 if (!PageUptodate(page)) {
2960 ret = block_read_full_page(page, ocfs2_get_block);
2968 if (page_has_buffers(page)) {
2969 ret = walk_page_buffers(handle, page_buffers(page),
2971 ocfs2_clear_cow_buffer);
2978 ocfs2_map_and_dirty_page(context->inode,
2980 page, 0, &new_block);
2981 mark_page_accessed(page);
2984 page_cache_release(page);
2994 static int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
2995 struct ocfs2_cow_context *context,
2996 u32 cpos, u32 old_cluster,
2997 u32 new_cluster, u32 new_len)
3000 struct super_block *sb = context->inode->i_sb;
3001 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3002 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3003 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3004 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3005 struct ocfs2_super *osb = OCFS2_SB(sb);
3006 struct buffer_head *old_bh = NULL;
3007 struct buffer_head *new_bh = NULL;
3009 mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster,
3010 new_cluster, new_len);
3012 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3013 new_bh = sb_getblk(osb->sb, new_block);
3014 if (new_bh == NULL) {
3020 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3022 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3028 ret = ocfs2_journal_access(handle, ci, new_bh,
3029 OCFS2_JOURNAL_ACCESS_CREATE);
3035 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3036 ocfs2_journal_dirty(handle, new_bh);
3049 static int ocfs2_clear_ext_refcount(handle_t *handle,
3050 struct ocfs2_extent_tree *et,
3051 u32 cpos, u32 p_cluster, u32 len,
3052 unsigned int ext_flags,
3053 struct ocfs2_alloc_context *meta_ac,
3054 struct ocfs2_cached_dealloc_ctxt *dealloc)
3057 struct ocfs2_extent_rec replace_rec;
3058 struct ocfs2_path *path = NULL;
3059 struct ocfs2_extent_list *el;
3060 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3061 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3063 mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
3064 (unsigned long long)ino, cpos, len, p_cluster, ext_flags);
3066 memset(&replace_rec, 0, sizeof(replace_rec));
3067 replace_rec.e_cpos = cpu_to_le32(cpos);
3068 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3069 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3071 replace_rec.e_flags = ext_flags;
3072 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3074 path = ocfs2_new_path_from_et(et);
3081 ret = ocfs2_find_path(et->et_ci, path, cpos);
3087 el = path_leaf_el(path);
3089 index = ocfs2_search_extent_list(el, cpos);
3090 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
3092 "Inode %llu has an extent at cpos %u which can no "
3093 "longer be found.\n",
3094 (unsigned long long)ino, cpos);
3099 ret = ocfs2_split_extent(handle, et, path, index,
3100 &replace_rec, meta_ac, dealloc);
3105 ocfs2_free_path(path);
3109 static int ocfs2_replace_clusters(handle_t *handle,
3110 struct ocfs2_cow_context *context,
3113 unsigned int ext_flags)
3116 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3117 u64 ino = ocfs2_metadata_cache_owner(ci);
3119 mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
3120 (unsigned long long)ino, cpos, old, new, len, ext_flags);
3122 /*If the old clusters is unwritten, no need to duplicate. */
3123 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3124 ret = context->cow_duplicate_clusters(handle, context, cpos,
3132 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3133 cpos, new, len, ext_flags,
3134 context->meta_ac, &context->dealloc);
3141 static int ocfs2_cow_sync_writeback(struct super_block *sb,
3142 struct ocfs2_cow_context *context,
3143 u32 cpos, u32 num_clusters)
3146 loff_t offset, end, map_end;
3150 if (ocfs2_should_order_data(context->inode))
3153 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3154 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3156 ret = filemap_fdatawrite_range(context->inode->i_mapping,
3163 while (offset < end) {
3164 page_index = offset >> PAGE_CACHE_SHIFT;
3165 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3169 page = grab_cache_page(context->inode->i_mapping, page_index);
3172 wait_on_page_writeback(page);
3173 if (PageError(page)) {
3177 mark_page_accessed(page);
3180 page_cache_release(page);
3190 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3191 u32 v_cluster, u32 *p_cluster,
3193 unsigned int *extent_flags)
3195 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3196 num_clusters, extent_flags);
3199 static int ocfs2_make_clusters_writable(struct super_block *sb,
3200 struct ocfs2_cow_context *context,
3201 u32 cpos, u32 p_cluster,
3202 u32 num_clusters, unsigned int e_flags)
3204 int ret, delete, index, credits = 0;
3205 u32 new_bit, new_len;
3206 unsigned int set_len;
3207 struct ocfs2_super *osb = OCFS2_SB(sb);
3209 struct buffer_head *ref_leaf_bh = NULL;
3210 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3211 struct ocfs2_refcount_rec rec;
3213 mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
3214 cpos, p_cluster, num_clusters, e_flags);
3216 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3219 context->ref_root_bh,
3221 &context->data_ac, &credits);
3227 if (context->post_refcount)
3228 credits += context->post_refcount->credits;
3230 credits += context->extra_credits;
3231 handle = ocfs2_start_trans(osb, credits);
3232 if (IS_ERR(handle)) {
3233 ret = PTR_ERR(handle);
3238 while (num_clusters) {
3239 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3240 p_cluster, num_clusters,
3241 &rec, &index, &ref_leaf_bh);
3247 BUG_ON(!rec.r_refcount);
3248 set_len = min((u64)p_cluster + num_clusters,
3249 le64_to_cpu(rec.r_cpos) +
3250 le32_to_cpu(rec.r_clusters)) - p_cluster;
3253 * There are many different situation here.
3254 * 1. If refcount == 1, remove the flag and don't COW.
3255 * 2. If refcount > 1, allocate clusters.
3256 * Here we may not allocate r_len once at a time, so continue
3257 * until we reach num_clusters.
3259 if (le32_to_cpu(rec.r_refcount) == 1) {
3261 ret = ocfs2_clear_ext_refcount(handle,
3274 ret = __ocfs2_claim_clusters(osb, handle,
3277 &new_bit, &new_len);
3283 ret = ocfs2_replace_clusters(handle, context,
3284 cpos, p_cluster, new_bit,
3293 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3294 context->ref_root_bh,
3297 &context->dealloc, delete);
3304 p_cluster += set_len;
3305 num_clusters -= set_len;
3306 brelse(ref_leaf_bh);
3310 /* handle any post_cow action. */
3311 if (context->post_refcount && context->post_refcount->func) {
3312 ret = context->post_refcount->func(context->inode, handle,
3313 context->post_refcount->para);
3321 * Here we should write the new page out first if we are
3322 * in write-back mode.
3324 if (context->get_clusters == ocfs2_di_get_clusters) {
3325 ret = ocfs2_cow_sync_writeback(sb, context, cpos, num_clusters);
3331 ocfs2_commit_trans(osb, handle);
3334 if (context->data_ac) {
3335 ocfs2_free_alloc_context(context->data_ac);
3336 context->data_ac = NULL;
3338 if (context->meta_ac) {
3339 ocfs2_free_alloc_context(context->meta_ac);
3340 context->meta_ac = NULL;
3342 brelse(ref_leaf_bh);
3347 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3350 struct inode *inode = context->inode;
3351 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3352 u32 p_cluster, num_clusters;
3353 unsigned int ext_flags;
3354 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3356 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3357 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3358 "tree, but the feature bit is not set in the "
3359 "super block.", inode->i_ino);
3363 ocfs2_init_dealloc_ctxt(&context->dealloc);
3366 ret = context->get_clusters(context, cow_start, &p_cluster,
3367 &num_clusters, &ext_flags);
3373 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3375 if (cow_len < num_clusters)
3376 num_clusters = cow_len;
3378 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3379 cow_start, p_cluster,
3380 num_clusters, ext_flags);
3386 cow_len -= num_clusters;
3387 cow_start += num_clusters;
3390 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3391 ocfs2_schedule_truncate_log_flush(osb, 1);
3392 ocfs2_run_deallocs(osb, &context->dealloc);
3399 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3400 * past max_cpos. This will stop when it runs into a hole or an
3401 * unrefcounted extent.
3403 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3404 struct buffer_head *di_bh,
3405 u32 cpos, u32 write_len, u32 max_cpos)
3408 u32 cow_start = 0, cow_len = 0;
3409 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3410 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3411 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3412 struct buffer_head *ref_root_bh = NULL;
3413 struct ocfs2_refcount_tree *ref_tree;
3414 struct ocfs2_cow_context *context = NULL;
3416 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3418 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3419 cpos, write_len, max_cpos,
3420 &cow_start, &cow_len);
3426 mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
3427 "cow_len %u\n", inode->i_ino,
3428 cpos, write_len, cow_start, cow_len);
3430 BUG_ON(cow_len == 0);
3432 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3439 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3440 1, &ref_tree, &ref_root_bh);
3446 context->inode = inode;
3447 context->cow_start = cow_start;
3448 context->cow_len = cow_len;
3449 context->ref_tree = ref_tree;
3450 context->ref_root_bh = ref_root_bh;
3451 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3452 context->get_clusters = ocfs2_di_get_clusters;
3454 ocfs2_init_dinode_extent_tree(&context->data_et,
3455 INODE_CACHE(inode), di_bh);
3457 ret = ocfs2_replace_cow(context);
3462 * truncate the extent map here since no matter whether we meet with
3463 * any error during the action, we shouldn't trust cached extent map
3466 ocfs2_extent_map_trunc(inode, cow_start);
3468 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3469 brelse(ref_root_bh);
3476 * CoW any and all clusters between cpos and cpos+write_len.
3477 * Don't CoW past max_cpos. If this returns successfully, all
3478 * clusters between cpos and cpos+write_len are safe to modify.
3480 int ocfs2_refcount_cow(struct inode *inode,
3481 struct buffer_head *di_bh,
3482 u32 cpos, u32 write_len, u32 max_cpos)
3485 u32 p_cluster, num_clusters;
3486 unsigned int ext_flags;
3489 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3490 &num_clusters, &ext_flags);
3496 if (write_len < num_clusters)
3497 num_clusters = write_len;
3499 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3500 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3501 num_clusters, max_cpos);
3508 write_len -= num_clusters;
3509 cpos += num_clusters;
3515 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3516 u32 v_cluster, u32 *p_cluster,
3518 unsigned int *extent_flags)
3520 struct inode *inode = context->inode;
3521 struct ocfs2_xattr_value_root *xv = context->cow_object;
3523 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3524 num_clusters, &xv->xr_list,
3529 * Given a xattr value root, calculate the most meta/credits we need for
3530 * refcount tree change if we truncate it to 0.
3532 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3533 struct ocfs2_caching_info *ref_ci,
3534 struct buffer_head *ref_root_bh,
3535 struct ocfs2_xattr_value_root *xv,
3536 int *meta_add, int *credits)
3538 int ret = 0, index, ref_blocks = 0;
3539 u32 p_cluster, num_clusters;
3540 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3541 struct ocfs2_refcount_block *rb;
3542 struct ocfs2_refcount_rec rec;
3543 struct buffer_head *ref_leaf_bh = NULL;
3545 while (cpos < clusters) {
3546 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3547 &num_clusters, &xv->xr_list,
3554 cpos += num_clusters;
3556 while (num_clusters) {
3557 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3558 p_cluster, num_clusters,
3566 BUG_ON(!rec.r_refcount);
3568 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3571 * We really don't know whether the other clusters is in
3572 * this refcount block or not, so just take the worst
3573 * case that all the clusters are in this block and each
3574 * one will split a refcount rec, so totally we need
3575 * clusters * 2 new refcount rec.
3577 if (le64_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3578 le16_to_cpu(rb->rf_records.rl_count))
3582 brelse(ref_leaf_bh);
3585 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3588 num_clusters -= le32_to_cpu(rec.r_clusters);
3589 p_cluster += num_clusters;
3593 *meta_add += ref_blocks;
3597 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3598 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3599 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3601 struct ocfs2_extent_tree et;
3603 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3604 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3610 brelse(ref_leaf_bh);
3617 int ocfs2_refcount_cow_xattr(struct inode *inode,
3618 struct ocfs2_dinode *di,
3619 struct ocfs2_xattr_value_buf *vb,
3620 struct ocfs2_refcount_tree *ref_tree,
3621 struct buffer_head *ref_root_bh,
3622 u32 cpos, u32 write_len,
3623 struct ocfs2_post_refcount *post)
3626 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3627 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3628 struct ocfs2_cow_context *context = NULL;
3629 u32 cow_start, cow_len;
3631 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3633 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3634 cpos, write_len, UINT_MAX,
3635 &cow_start, &cow_len);
3641 BUG_ON(cow_len == 0);
3643 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3650 context->inode = inode;
3651 context->cow_start = cow_start;
3652 context->cow_len = cow_len;
3653 context->ref_tree = ref_tree;
3654 context->ref_root_bh = ref_root_bh;;
3655 context->cow_object = xv;
3657 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3658 /* We need the extra credits for duplicate_clusters by jbd. */
3659 context->extra_credits =
3660 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3661 context->get_clusters = ocfs2_xattr_value_get_clusters;
3662 context->post_refcount = post;
3664 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3665 INODE_CACHE(inode), vb);
3667 ret = ocfs2_replace_cow(context);
3677 * Insert a new extent into refcount tree and mark a extent rec
3678 * as refcounted in the dinode tree.
3680 int ocfs2_add_refcount_flag(struct inode *inode,
3681 struct ocfs2_extent_tree *data_et,
3682 struct ocfs2_caching_info *ref_ci,
3683 struct buffer_head *ref_root_bh,
3684 u32 cpos, u32 p_cluster, u32 num_clusters,
3685 struct ocfs2_cached_dealloc_ctxt *dealloc,
3686 struct ocfs2_post_refcount *post)
3690 int credits = 1, ref_blocks = 0;
3691 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3692 struct ocfs2_alloc_context *meta_ac = NULL;
3694 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3695 ref_ci, ref_root_bh,
3696 p_cluster, num_clusters,
3697 &ref_blocks, &credits);
3703 mlog(0, "reserve new metadata %d, credits = %d\n",
3704 ref_blocks, credits);
3707 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3708 ref_blocks, &meta_ac);
3716 credits += post->credits;
3718 handle = ocfs2_start_trans(osb, credits);
3719 if (IS_ERR(handle)) {
3720 ret = PTR_ERR(handle);
3725 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3726 cpos, num_clusters, p_cluster,
3733 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3734 p_cluster, num_clusters, 0,
3741 if (post && post->func) {
3742 ret = post->func(inode, handle, post->para);
3748 ocfs2_commit_trans(osb, handle);
3751 ocfs2_free_alloc_context(meta_ac);
3755 static int ocfs2_change_ctime(struct inode *inode,
3756 struct buffer_head *di_bh)
3760 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3762 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3763 OCFS2_INODE_UPDATE_CREDITS);
3764 if (IS_ERR(handle)) {
3765 ret = PTR_ERR(handle);
3770 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3771 OCFS2_JOURNAL_ACCESS_WRITE);
3777 inode->i_ctime = CURRENT_TIME;
3778 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3779 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3781 ocfs2_journal_dirty(handle, di_bh);
3784 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3789 static int ocfs2_attach_refcount_tree(struct inode *inode,
3790 struct buffer_head *di_bh)
3792 int ret, data_changed = 0;
3793 struct buffer_head *ref_root_bh = NULL;
3794 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3795 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3796 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3797 struct ocfs2_refcount_tree *ref_tree;
3798 unsigned int ext_flags;
3800 u32 cpos, num_clusters, clusters, p_cluster;
3801 struct ocfs2_cached_dealloc_ctxt dealloc;
3802 struct ocfs2_extent_tree di_et;
3804 ocfs2_init_dealloc_ctxt(&dealloc);
3806 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3807 ret = ocfs2_create_refcount_tree(inode, di_bh);
3814 BUG_ON(!di->i_refcount_loc);
3815 ret = ocfs2_lock_refcount_tree(osb,
3816 le64_to_cpu(di->i_refcount_loc), 1,
3817 &ref_tree, &ref_root_bh);
3823 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3826 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3828 size = i_size_read(inode);
3829 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3832 while (cpos < clusters) {
3833 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3834 &num_clusters, &ext_flags);
3836 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3837 ret = ocfs2_add_refcount_flag(inode, &di_et,
3840 p_cluster, num_clusters,
3849 cpos += num_clusters;
3853 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3854 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3865 ret = ocfs2_change_ctime(inode, di_bh);
3871 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3872 brelse(ref_root_bh);
3874 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3875 ocfs2_schedule_truncate_log_flush(osb, 1);
3876 ocfs2_run_deallocs(osb, &dealloc);
3880 * Empty the extent map so that we may get the right extent
3881 * record from the disk.
3883 ocfs2_extent_map_trunc(inode, 0);
3888 static int ocfs2_add_refcounted_extent(struct inode *inode,
3889 struct ocfs2_extent_tree *et,
3890 struct ocfs2_caching_info *ref_ci,
3891 struct buffer_head *ref_root_bh,
3892 u32 cpos, u32 p_cluster, u32 num_clusters,
3893 unsigned int ext_flags,
3894 struct ocfs2_cached_dealloc_ctxt *dealloc)
3899 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3900 struct ocfs2_alloc_context *meta_ac = NULL;
3902 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3903 p_cluster, num_clusters,
3905 ref_root_bh, &meta_ac,
3912 handle = ocfs2_start_trans(osb, credits);
3913 if (IS_ERR(handle)) {
3914 ret = PTR_ERR(handle);
3919 ret = ocfs2_insert_extent(handle, et, cpos,
3920 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3921 num_clusters, ext_flags, meta_ac);
3927 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3928 p_cluster, num_clusters,
3934 ocfs2_commit_trans(osb, handle);
3937 ocfs2_free_alloc_context(meta_ac);
3941 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3942 struct buffer_head *s_bh,
3943 struct inode *t_inode,
3944 struct buffer_head *t_bh)
3948 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3949 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3950 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3952 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3954 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3955 if (IS_ERR(handle)) {
3956 ret = PTR_ERR(handle);
3961 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3962 OCFS2_JOURNAL_ACCESS_WRITE);
3968 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3969 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3970 le16_to_cpu(s_di->id2.i_data.id_count));
3971 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3972 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3973 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3974 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3976 ocfs2_journal_dirty(handle, t_bh);
3979 ocfs2_commit_trans(osb, handle);
3984 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3985 struct inode *t_inode,
3986 struct buffer_head *t_bh,
3987 struct ocfs2_caching_info *ref_ci,
3988 struct buffer_head *ref_root_bh,
3989 struct ocfs2_cached_dealloc_ctxt *dealloc)
3992 u32 p_cluster, num_clusters, clusters, cpos;
3994 unsigned int ext_flags;
3995 struct ocfs2_extent_tree et;
3997 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
3999 size = i_size_read(s_inode);
4000 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4003 while (cpos < clusters) {
4004 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4005 &num_clusters, &ext_flags);
4008 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4009 ref_ci, ref_root_bh,
4020 cpos += num_clusters;
4028 * change the new file's attributes to the src.
4030 * reflink creates a snapshot of a file, that means the attributes
4031 * must be identical except for three exceptions - nlink, ino, and ctime.
4033 static int ocfs2_complete_reflink(struct inode *s_inode,
4034 struct buffer_head *s_bh,
4035 struct inode *t_inode,
4036 struct buffer_head *t_bh,
4041 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4042 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4043 loff_t size = i_size_read(s_inode);
4045 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4046 OCFS2_INODE_UPDATE_CREDITS);
4047 if (IS_ERR(handle)) {
4048 ret = PTR_ERR(handle);
4053 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4054 OCFS2_JOURNAL_ACCESS_WRITE);
4060 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4061 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4062 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4063 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4064 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4065 i_size_write(t_inode, size);
4066 t_inode->i_blocks = s_inode->i_blocks;
4068 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4069 di->i_clusters = s_di->i_clusters;
4070 di->i_size = s_di->i_size;
4071 di->i_dyn_features = s_di->i_dyn_features;
4072 di->i_attr = s_di->i_attr;
4075 di->i_uid = s_di->i_uid;
4076 di->i_gid = s_di->i_gid;
4077 di->i_mode = s_di->i_mode;
4081 * we want mtime to appear identical to the source and
4084 t_inode->i_ctime = CURRENT_TIME;
4086 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4087 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4089 t_inode->i_mtime = s_inode->i_mtime;
4090 di->i_mtime = s_di->i_mtime;
4091 di->i_mtime_nsec = s_di->i_mtime_nsec;
4094 ocfs2_journal_dirty(handle, t_bh);
4097 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4101 static int ocfs2_create_reflink_node(struct inode *s_inode,
4102 struct buffer_head *s_bh,
4103 struct inode *t_inode,
4104 struct buffer_head *t_bh,
4108 struct buffer_head *ref_root_bh = NULL;
4109 struct ocfs2_cached_dealloc_ctxt dealloc;
4110 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4111 struct ocfs2_refcount_block *rb;
4112 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4113 struct ocfs2_refcount_tree *ref_tree;
4115 ocfs2_init_dealloc_ctxt(&dealloc);
4117 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4118 le64_to_cpu(di->i_refcount_loc));
4124 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4125 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4132 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4133 1, &ref_tree, &ref_root_bh);
4138 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4140 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4141 &ref_tree->rf_ci, ref_root_bh,
4145 goto out_unlock_refcount;
4148 out_unlock_refcount:
4149 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4150 brelse(ref_root_bh);
4152 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4153 ocfs2_schedule_truncate_log_flush(osb, 1);
4154 ocfs2_run_deallocs(osb, &dealloc);
4160 static int __ocfs2_reflink(struct dentry *old_dentry,
4161 struct buffer_head *old_bh,
4162 struct inode *new_inode,
4166 struct inode *inode = old_dentry->d_inode;
4167 struct buffer_head *new_bh = NULL;
4169 ret = filemap_fdatawrite(inode->i_mapping);
4175 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4181 mutex_lock(&new_inode->i_mutex);
4182 ret = ocfs2_inode_lock(new_inode, &new_bh, 1);
4188 ret = ocfs2_create_reflink_node(inode, old_bh,
4189 new_inode, new_bh, preserve);
4195 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4196 ret = ocfs2_reflink_xattrs(inode, old_bh,
4205 ret = ocfs2_complete_reflink(inode, old_bh,
4206 new_inode, new_bh, preserve);
4211 ocfs2_inode_unlock(new_inode, 1);
4214 mutex_unlock(&new_inode->i_mutex);
4217 ret = filemap_fdatawait(inode->i_mapping);
4224 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4225 struct dentry *new_dentry, bool preserve)
4228 struct inode *inode = old_dentry->d_inode;
4229 struct buffer_head *old_bh = NULL;
4230 struct inode *new_orphan_inode = NULL;
4232 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4235 error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4242 error = ocfs2_inode_lock(inode, &old_bh, 1);
4248 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4249 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4250 error = __ocfs2_reflink(old_dentry, old_bh,
4251 new_orphan_inode, preserve);
4252 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4253 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4255 ocfs2_inode_unlock(inode, 1);
4263 /* If the security isn't preserved, we need to re-initialize them. */
4265 error = ocfs2_init_security_and_acl(dir, new_orphan_inode);
4271 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4277 if (new_orphan_inode) {
4279 * We need to open_unlock the inode no matter whether we
4280 * succeed or not, so that other nodes can delete it later.
4282 ocfs2_open_unlock(new_orphan_inode);
4284 iput(new_orphan_inode);
4291 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4292 * sys_reflink(). This will go away when vfs_reflink() exists in
4296 /* copied from may_create in VFS. */
4297 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4301 if (IS_DEADDIR(dir))
4303 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4306 /* copied from user_path_parent. */
4307 static int ocfs2_user_path_parent(const char __user *path,
4308 struct nameidata *nd, char **name)
4310 char *s = getname(path);
4316 error = path_lookup(s, LOOKUP_PARENT, nd);
4326 * ocfs2_vfs_reflink - Create a reference-counted link
4328 * @old_dentry: source dentry + inode
4329 * @dir: directory to create the target
4330 * @new_dentry: target dentry
4331 * @preserve: if true, preserve all file attributes
4333 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4334 struct dentry *new_dentry, bool preserve)
4336 struct inode *inode = old_dentry->d_inode;
4342 error = ocfs2_may_create(dir, new_dentry);
4346 if (dir->i_sb != inode->i_sb)
4350 * A reflink to an append-only or immutable file cannot be created.
4352 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4355 /* Only regular files can be reflinked. */
4356 if (!S_ISREG(inode->i_mode))
4360 * If the caller wants to preserve ownership, they require the
4364 if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
4366 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4371 * If the caller is modifying any aspect of the attributes, they
4372 * are not creating a snapshot. They need read permission on the
4376 error = inode_permission(inode, MAY_READ);
4381 mutex_lock(&inode->i_mutex);
4382 dquot_initialize(dir);
4383 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4384 mutex_unlock(&inode->i_mutex);
4386 fsnotify_create(dir, new_dentry);
4390 * Most codes are copied from sys_linkat.
4392 int ocfs2_reflink_ioctl(struct inode *inode,
4393 const char __user *oldname,
4394 const char __user *newname,
4397 struct dentry *new_dentry;
4398 struct nameidata nd;
4399 struct path old_path;
4403 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4406 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4412 error = ocfs2_user_path_parent(newname, &nd, &to);
4419 if (old_path.mnt != nd.path.mnt)
4421 new_dentry = lookup_create(&nd, 0);
4422 error = PTR_ERR(new_dentry);
4423 if (IS_ERR(new_dentry)) {
4428 error = mnt_want_write(nd.path.mnt);
4434 error = ocfs2_vfs_reflink(old_path.dentry,
4435 nd.path.dentry->d_inode,
4436 new_dentry, preserve);
4437 mnt_drop_write(nd.path.mnt);
4441 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
4446 path_put(&old_path);