1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2009 Oracle. All rights reserved.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public
10 * License version 2 as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
18 #include <linux/sort.h>
19 #include <cluster/masklog.h>
27 #include "buffer_head_io.h"
28 #include "blockcheck.h"
29 #include "refcounttree.h"
32 #include "extent_map.h"
36 #include "ocfs2_trace.h"
38 #include <linux/bio.h>
39 #include <linux/blkdev.h>
40 #include <linux/slab.h>
41 #include <linux/writeback.h>
42 #include <linux/pagevec.h>
43 #include <linux/swap.h>
44 #include <linux/security.h>
45 #include <linux/fsnotify.h>
46 #include <linux/quotaops.h>
47 #include <linux/namei.h>
48 #include <linux/mount.h>
49 #include <linux/posix_acl.h>
51 struct ocfs2_cow_context {
55 struct ocfs2_extent_tree data_et;
56 struct ocfs2_refcount_tree *ref_tree;
57 struct buffer_head *ref_root_bh;
58 struct ocfs2_alloc_context *meta_ac;
59 struct ocfs2_alloc_context *data_ac;
60 struct ocfs2_cached_dealloc_ctxt dealloc;
62 struct ocfs2_post_refcount *post_refcount;
64 int (*get_clusters)(struct ocfs2_cow_context *context,
65 u32 v_cluster, u32 *p_cluster,
67 unsigned int *extent_flags);
68 int (*cow_duplicate_clusters)(handle_t *handle,
70 u32 cpos, u32 old_cluster,
71 u32 new_cluster, u32 new_len);
74 static inline struct ocfs2_refcount_tree *
75 cache_info_to_refcount(struct ocfs2_caching_info *ci)
77 return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
80 static int ocfs2_validate_refcount_block(struct super_block *sb,
81 struct buffer_head *bh)
84 struct ocfs2_refcount_block *rb =
85 (struct ocfs2_refcount_block *)bh->b_data;
87 trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
89 BUG_ON(!buffer_uptodate(bh));
92 * If the ecc fails, we return the error but otherwise
93 * leave the filesystem running. We know any error is
94 * local to this block.
96 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
98 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
99 (unsigned long long)bh->b_blocknr);
104 if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
106 "Refcount block #%llu has bad signature %.*s\n",
107 (unsigned long long)bh->b_blocknr, 7,
112 if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
114 "Refcount block #%llu has an invalid rf_blkno of %llu\n",
115 (unsigned long long)bh->b_blocknr,
116 (unsigned long long)le64_to_cpu(rb->rf_blkno));
120 if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
122 "Refcount block #%llu has an invalid rf_fs_generation of #%u\n",
123 (unsigned long long)bh->b_blocknr,
124 le32_to_cpu(rb->rf_fs_generation));
131 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
133 struct buffer_head **bh)
136 struct buffer_head *tmp = *bh;
138 rc = ocfs2_read_block(ci, rb_blkno, &tmp,
139 ocfs2_validate_refcount_block);
141 /* If ocfs2_read_block() got us a new bh, pass it up. */
148 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
150 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
155 static struct super_block *
156 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
158 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
163 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
165 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
167 spin_lock(&rf->rf_lock);
170 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
172 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
174 spin_unlock(&rf->rf_lock);
177 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
179 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
181 mutex_lock(&rf->rf_io_mutex);
184 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
186 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
188 mutex_unlock(&rf->rf_io_mutex);
191 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
192 .co_owner = ocfs2_refcount_cache_owner,
193 .co_get_super = ocfs2_refcount_cache_get_super,
194 .co_cache_lock = ocfs2_refcount_cache_lock,
195 .co_cache_unlock = ocfs2_refcount_cache_unlock,
196 .co_io_lock = ocfs2_refcount_cache_io_lock,
197 .co_io_unlock = ocfs2_refcount_cache_io_unlock,
200 static struct ocfs2_refcount_tree *
201 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
203 struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
204 struct ocfs2_refcount_tree *tree = NULL;
207 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
209 if (blkno < tree->rf_blkno)
211 else if (blkno > tree->rf_blkno)
220 /* osb_lock is already locked. */
221 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
222 struct ocfs2_refcount_tree *new)
224 u64 rf_blkno = new->rf_blkno;
225 struct rb_node *parent = NULL;
226 struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
227 struct ocfs2_refcount_tree *tmp;
232 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
235 if (rf_blkno < tmp->rf_blkno)
237 else if (rf_blkno > tmp->rf_blkno)
240 /* This should never happen! */
241 mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
242 (unsigned long long)rf_blkno);
247 rb_link_node(&new->rf_node, parent, p);
248 rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
251 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
253 ocfs2_metadata_cache_exit(&tree->rf_ci);
254 ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
255 ocfs2_lock_res_free(&tree->rf_lockres);
260 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
261 struct ocfs2_refcount_tree *tree)
263 rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
264 if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
265 osb->osb_ref_tree_lru = NULL;
268 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
269 struct ocfs2_refcount_tree *tree)
271 spin_lock(&osb->osb_lock);
272 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
273 spin_unlock(&osb->osb_lock);
276 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
278 struct ocfs2_refcount_tree *tree =
279 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
281 ocfs2_free_refcount_tree(tree);
285 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
287 kref_get(&tree->rf_getcnt);
291 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
293 kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
296 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
297 struct super_block *sb)
299 ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
300 mutex_init(&new->rf_io_mutex);
302 spin_lock_init(&new->rf_lock);
305 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
306 struct ocfs2_refcount_tree *new,
307 u64 rf_blkno, u32 generation)
309 init_rwsem(&new->rf_sem);
310 ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
311 rf_blkno, generation);
314 static struct ocfs2_refcount_tree*
315 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
317 struct ocfs2_refcount_tree *new;
319 new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
323 new->rf_blkno = rf_blkno;
324 kref_init(&new->rf_getcnt);
325 ocfs2_init_refcount_tree_ci(new, osb->sb);
330 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
331 struct ocfs2_refcount_tree **ret_tree)
334 struct ocfs2_refcount_tree *tree, *new = NULL;
335 struct buffer_head *ref_root_bh = NULL;
336 struct ocfs2_refcount_block *ref_rb;
338 spin_lock(&osb->osb_lock);
339 if (osb->osb_ref_tree_lru &&
340 osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
341 tree = osb->osb_ref_tree_lru;
343 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
347 spin_unlock(&osb->osb_lock);
349 new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
356 * We need the generation to create the refcount tree lock and since
357 * it isn't changed during the tree modification, we are safe here to
358 * read without protection.
359 * We also have to purge the cache after we create the lock since the
360 * refcount block may have the stale data. It can only be trusted when
361 * we hold the refcount lock.
363 ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
366 ocfs2_metadata_cache_exit(&new->rf_ci);
371 ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
372 new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
373 ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
375 ocfs2_metadata_cache_purge(&new->rf_ci);
377 spin_lock(&osb->osb_lock);
378 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
382 ocfs2_insert_refcount_tree(osb, new);
390 osb->osb_ref_tree_lru = tree;
392 spin_unlock(&osb->osb_lock);
395 ocfs2_free_refcount_tree(new);
401 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
404 struct buffer_head *di_bh = NULL;
405 struct ocfs2_dinode *di;
407 ret = ocfs2_read_inode_block(inode, &di_bh);
413 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
415 di = (struct ocfs2_dinode *)di_bh->b_data;
416 *ref_blkno = le64_to_cpu(di->i_refcount_loc);
422 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
423 struct ocfs2_refcount_tree *tree, int rw)
427 ret = ocfs2_refcount_lock(tree, rw);
434 down_write(&tree->rf_sem);
436 down_read(&tree->rf_sem);
443 * Lock the refcount tree pointed by ref_blkno and return the tree.
444 * In most case, we lock the tree and read the refcount block.
445 * So read it here if the caller really needs it.
447 * If the tree has been re-created by other node, it will free the
448 * old one and re-create it.
450 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
451 u64 ref_blkno, int rw,
452 struct ocfs2_refcount_tree **ret_tree,
453 struct buffer_head **ref_bh)
455 int ret, delete_tree = 0;
456 struct ocfs2_refcount_tree *tree = NULL;
457 struct buffer_head *ref_root_bh = NULL;
458 struct ocfs2_refcount_block *rb;
461 ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
467 ocfs2_refcount_tree_get(tree);
469 ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
472 ocfs2_refcount_tree_put(tree);
476 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
480 ocfs2_unlock_refcount_tree(osb, tree, rw);
484 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
486 * If the refcount block has been freed and re-created, we may need
487 * to recreate the refcount tree also.
489 * Here we just remove the tree from the rb-tree, and the last
490 * kref holder will unlock and delete this refcount_tree.
491 * Then we goto "again" and ocfs2_get_refcount_tree will create
492 * the new refcount tree for us.
494 if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
495 if (!tree->rf_removed) {
496 ocfs2_erase_refcount_tree_from_list(osb, tree);
497 tree->rf_removed = 1;
501 ocfs2_unlock_refcount_tree(osb, tree, rw);
503 * We get an extra reference when we create the refcount
504 * tree, so another put will destroy it.
507 ocfs2_refcount_tree_put(tree);
515 *ref_bh = ref_root_bh;
523 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
524 struct ocfs2_refcount_tree *tree, int rw)
527 up_write(&tree->rf_sem);
529 up_read(&tree->rf_sem);
531 ocfs2_refcount_unlock(tree, rw);
532 ocfs2_refcount_tree_put(tree);
535 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
537 struct rb_node *node;
538 struct ocfs2_refcount_tree *tree;
539 struct rb_root *root = &osb->osb_rf_lock_tree;
541 while ((node = rb_last(root)) != NULL) {
542 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
544 trace_ocfs2_purge_refcount_trees(
545 (unsigned long long) tree->rf_blkno);
547 rb_erase(&tree->rf_node, root);
548 ocfs2_free_refcount_tree(tree);
553 * Create a refcount tree for an inode.
554 * We take for granted that the inode is already locked.
556 static int ocfs2_create_refcount_tree(struct inode *inode,
557 struct buffer_head *di_bh)
560 handle_t *handle = NULL;
561 struct ocfs2_alloc_context *meta_ac = NULL;
562 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
563 struct ocfs2_inode_info *oi = OCFS2_I(inode);
564 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
565 struct buffer_head *new_bh = NULL;
566 struct ocfs2_refcount_block *rb;
567 struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
568 u16 suballoc_bit_start;
570 u64 suballoc_loc, first_blkno;
572 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
574 trace_ocfs2_create_refcount_tree(
575 (unsigned long long)OCFS2_I(inode)->ip_blkno);
577 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
583 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
584 if (IS_ERR(handle)) {
585 ret = PTR_ERR(handle);
590 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
591 OCFS2_JOURNAL_ACCESS_WRITE);
597 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
598 &suballoc_bit_start, &num_got,
605 new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
612 new_bh = sb_getblk(inode->i_sb, first_blkno);
618 ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
620 ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
621 OCFS2_JOURNAL_ACCESS_CREATE);
627 /* Initialize ocfs2_refcount_block. */
628 rb = (struct ocfs2_refcount_block *)new_bh->b_data;
629 memset(rb, 0, inode->i_sb->s_blocksize);
630 strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
631 rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
632 rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
633 rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
634 rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
635 rb->rf_blkno = cpu_to_le64(first_blkno);
636 rb->rf_count = cpu_to_le32(1);
637 rb->rf_records.rl_count =
638 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
639 spin_lock(&osb->osb_lock);
640 rb->rf_generation = osb->s_next_generation++;
641 spin_unlock(&osb->osb_lock);
643 ocfs2_journal_dirty(handle, new_bh);
645 spin_lock(&oi->ip_lock);
646 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
647 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
648 di->i_refcount_loc = cpu_to_le64(first_blkno);
649 spin_unlock(&oi->ip_lock);
651 trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
653 ocfs2_journal_dirty(handle, di_bh);
656 * We have to init the tree lock here since it will use
657 * the generation number to create it.
659 new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
660 ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
661 new_tree->rf_generation);
663 spin_lock(&osb->osb_lock);
664 tree = ocfs2_find_refcount_tree(osb, first_blkno);
667 * We've just created a new refcount tree in this block. If
668 * we found a refcount tree on the ocfs2_super, it must be
669 * one we just deleted. We free the old tree before
670 * inserting the new tree.
672 BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
674 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
675 ocfs2_insert_refcount_tree(osb, new_tree);
676 spin_unlock(&osb->osb_lock);
679 ocfs2_refcount_tree_put(tree);
682 ocfs2_commit_trans(osb, handle);
686 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
692 ocfs2_free_alloc_context(meta_ac);
697 static int ocfs2_set_refcount_tree(struct inode *inode,
698 struct buffer_head *di_bh,
702 handle_t *handle = NULL;
703 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
704 struct ocfs2_inode_info *oi = OCFS2_I(inode);
705 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
706 struct buffer_head *ref_root_bh = NULL;
707 struct ocfs2_refcount_block *rb;
708 struct ocfs2_refcount_tree *ref_tree;
710 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
712 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
713 &ref_tree, &ref_root_bh);
719 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
720 if (IS_ERR(handle)) {
721 ret = PTR_ERR(handle);
726 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
727 OCFS2_JOURNAL_ACCESS_WRITE);
733 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
734 OCFS2_JOURNAL_ACCESS_WRITE);
740 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
741 le32_add_cpu(&rb->rf_count, 1);
743 ocfs2_journal_dirty(handle, ref_root_bh);
745 spin_lock(&oi->ip_lock);
746 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
747 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
748 di->i_refcount_loc = cpu_to_le64(refcount_loc);
749 spin_unlock(&oi->ip_lock);
750 ocfs2_journal_dirty(handle, di_bh);
753 ocfs2_commit_trans(osb, handle);
755 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
761 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
763 int ret, delete_tree = 0;
764 handle_t *handle = NULL;
765 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
766 struct ocfs2_inode_info *oi = OCFS2_I(inode);
767 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
768 struct ocfs2_refcount_block *rb;
769 struct inode *alloc_inode = NULL;
770 struct buffer_head *alloc_bh = NULL;
771 struct buffer_head *blk_bh = NULL;
772 struct ocfs2_refcount_tree *ref_tree;
773 int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
774 u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
777 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
781 ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
787 rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
790 * If we are the last user, we need to free the block.
791 * So lock the allocator ahead.
793 if (le32_to_cpu(rb->rf_count) == 1) {
794 blk = le64_to_cpu(rb->rf_blkno);
795 bit = le16_to_cpu(rb->rf_suballoc_bit);
796 if (rb->rf_suballoc_loc)
797 bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
799 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
801 alloc_inode = ocfs2_get_system_file_inode(osb,
802 EXTENT_ALLOC_SYSTEM_INODE,
803 le16_to_cpu(rb->rf_suballoc_slot));
809 inode_lock(alloc_inode);
811 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
817 credits += OCFS2_SUBALLOC_FREE;
820 handle = ocfs2_start_trans(osb, credits);
821 if (IS_ERR(handle)) {
822 ret = PTR_ERR(handle);
827 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
828 OCFS2_JOURNAL_ACCESS_WRITE);
834 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
835 OCFS2_JOURNAL_ACCESS_WRITE);
841 spin_lock(&oi->ip_lock);
842 oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
843 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
844 di->i_refcount_loc = 0;
845 spin_unlock(&oi->ip_lock);
846 ocfs2_journal_dirty(handle, di_bh);
848 le32_add_cpu(&rb->rf_count , -1);
849 ocfs2_journal_dirty(handle, blk_bh);
853 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
854 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
855 alloc_bh, bit, bg_blkno, 1);
861 ocfs2_commit_trans(osb, handle);
864 ocfs2_inode_unlock(alloc_inode, 1);
869 inode_unlock(alloc_inode);
873 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
875 ocfs2_refcount_tree_put(ref_tree);
881 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
882 struct buffer_head *ref_leaf_bh,
883 u64 cpos, unsigned int len,
884 struct ocfs2_refcount_rec *ret_rec,
888 struct ocfs2_refcount_block *rb =
889 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
890 struct ocfs2_refcount_rec *rec = NULL;
892 for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
893 rec = &rb->rf_records.rl_recs[i];
895 if (le64_to_cpu(rec->r_cpos) +
896 le32_to_cpu(rec->r_clusters) <= cpos)
898 else if (le64_to_cpu(rec->r_cpos) > cpos)
901 /* ok, cpos fail in this rec. Just return. */
908 /* We meet with a hole here, so fake the rec. */
909 ret_rec->r_cpos = cpu_to_le64(cpos);
910 ret_rec->r_refcount = 0;
911 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
912 le64_to_cpu(rec->r_cpos) < cpos + len)
913 ret_rec->r_clusters =
914 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
916 ret_rec->r_clusters = cpu_to_le32(len);
924 * Try to remove refcount tree. The mechanism is:
925 * 1) Check whether i_clusters == 0, if no, exit.
926 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
927 * 3) Check whether we have inline xattr stored outside, if yes, exit.
928 * 4) Remove the tree.
930 int ocfs2_try_remove_refcount_tree(struct inode *inode,
931 struct buffer_head *di_bh)
934 struct ocfs2_inode_info *oi = OCFS2_I(inode);
935 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
937 down_write(&oi->ip_xattr_sem);
938 down_write(&oi->ip_alloc_sem);
943 if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
946 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
947 ocfs2_has_inline_xattr_value_outside(inode, di))
950 ret = ocfs2_remove_refcount_tree(inode, di_bh);
954 up_write(&oi->ip_alloc_sem);
955 up_write(&oi->ip_xattr_sem);
960 * Find the end range for a leaf refcount block indicated by
961 * el->l_recs[index].e_blkno.
963 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
964 struct buffer_head *ref_root_bh,
965 struct ocfs2_extent_block *eb,
966 struct ocfs2_extent_list *el,
967 int index, u32 *cpos_end)
969 int ret, i, subtree_root;
972 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
973 struct ocfs2_path *left_path = NULL, *right_path = NULL;
974 struct ocfs2_extent_tree et;
975 struct ocfs2_extent_list *tmp_el;
977 if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
979 * We have a extent rec after index, so just use the e_cpos
980 * of the next extent rec.
982 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
986 if (!eb || (eb && !eb->h_next_leaf_blk)) {
988 * We are the last extent rec, so any high cpos should
989 * be stored in this leaf refcount block.
991 *cpos_end = UINT_MAX;
996 * If the extent block isn't the last one, we have to find
997 * the subtree root between this extent block and the next
998 * leaf extent block and get the corresponding e_cpos from
999 * the subroot. Otherwise we may corrupt the b-tree.
1001 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1003 left_path = ocfs2_new_path_from_et(&et);
1010 cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1011 ret = ocfs2_find_path(ci, left_path, cpos);
1017 right_path = ocfs2_new_path_from_path(left_path);
1024 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1030 ret = ocfs2_find_path(ci, right_path, cpos);
1036 subtree_root = ocfs2_find_subtree_root(&et, left_path,
1039 tmp_el = left_path->p_node[subtree_root].el;
1040 blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1041 for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1042 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1043 *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1048 BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1051 ocfs2_free_path(left_path);
1052 ocfs2_free_path(right_path);
1057 * Given a cpos and len, try to find the refcount record which contains cpos.
1058 * 1. If cpos can be found in one refcount record, return the record.
1059 * 2. If cpos can't be found, return a fake record which start from cpos
1060 * and end at a small value between cpos+len and start of the next record.
1061 * This fake record has r_refcount = 0.
1063 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1064 struct buffer_head *ref_root_bh,
1065 u64 cpos, unsigned int len,
1066 struct ocfs2_refcount_rec *ret_rec,
1068 struct buffer_head **ret_bh)
1070 int ret = 0, i, found;
1071 u32 low_cpos, uninitialized_var(cpos_end);
1072 struct ocfs2_extent_list *el;
1073 struct ocfs2_extent_rec *rec = NULL;
1074 struct ocfs2_extent_block *eb = NULL;
1075 struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1076 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1077 struct ocfs2_refcount_block *rb =
1078 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1080 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1081 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1083 *ret_bh = ref_root_bh;
1084 get_bh(ref_root_bh);
1089 low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1091 if (el->l_tree_depth) {
1092 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1098 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1101 if (el->l_tree_depth) {
1102 ret = ocfs2_error(sb,
1103 "refcount tree %llu has non zero tree depth in leaf btree tree block %llu\n",
1104 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1105 (unsigned long long)eb_bh->b_blocknr);
1111 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1112 rec = &el->l_recs[i];
1114 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1121 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1122 eb, el, i, &cpos_end);
1128 if (cpos_end < low_cpos + len)
1129 len = cpos_end - low_cpos;
1132 ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1139 ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1141 *ret_bh = ref_leaf_bh;
1147 enum ocfs2_ref_rec_contig {
1148 REF_CONTIG_NONE = 0,
1151 REF_CONTIG_LEFTRIGHT,
1154 static enum ocfs2_ref_rec_contig
1155 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1158 if ((rb->rf_records.rl_recs[index].r_refcount ==
1159 rb->rf_records.rl_recs[index + 1].r_refcount) &&
1160 (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1161 le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1162 le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1163 return REF_CONTIG_RIGHT;
1165 return REF_CONTIG_NONE;
1168 static enum ocfs2_ref_rec_contig
1169 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1172 enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1174 if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1175 ret = ocfs2_refcount_rec_adjacent(rb, index);
1178 enum ocfs2_ref_rec_contig tmp;
1180 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1182 if (tmp == REF_CONTIG_RIGHT) {
1183 if (ret == REF_CONTIG_RIGHT)
1184 ret = REF_CONTIG_LEFTRIGHT;
1186 ret = REF_CONTIG_LEFT;
1193 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1196 BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1197 rb->rf_records.rl_recs[index+1].r_refcount);
1199 le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1200 le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1202 if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1203 memmove(&rb->rf_records.rl_recs[index + 1],
1204 &rb->rf_records.rl_recs[index + 2],
1205 sizeof(struct ocfs2_refcount_rec) *
1206 (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1208 memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1209 0, sizeof(struct ocfs2_refcount_rec));
1210 le16_add_cpu(&rb->rf_records.rl_used, -1);
1214 * Merge the refcount rec if we are contiguous with the adjacent recs.
1216 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1219 enum ocfs2_ref_rec_contig contig =
1220 ocfs2_refcount_rec_contig(rb, index);
1222 if (contig == REF_CONTIG_NONE)
1225 if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1230 ocfs2_rotate_refcount_rec_left(rb, index);
1232 if (contig == REF_CONTIG_LEFTRIGHT)
1233 ocfs2_rotate_refcount_rec_left(rb, index);
1237 * Change the refcount indexed by "index" in ref_bh.
1238 * If refcount reaches 0, remove it.
1240 static int ocfs2_change_refcount_rec(handle_t *handle,
1241 struct ocfs2_caching_info *ci,
1242 struct buffer_head *ref_leaf_bh,
1243 int index, int merge, int change)
1246 struct ocfs2_refcount_block *rb =
1247 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1248 struct ocfs2_refcount_list *rl = &rb->rf_records;
1249 struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1251 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1252 OCFS2_JOURNAL_ACCESS_WRITE);
1258 trace_ocfs2_change_refcount_rec(
1259 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1260 index, le32_to_cpu(rec->r_refcount), change);
1261 le32_add_cpu(&rec->r_refcount, change);
1263 if (!rec->r_refcount) {
1264 if (index != le16_to_cpu(rl->rl_used) - 1) {
1265 memmove(rec, rec + 1,
1266 (le16_to_cpu(rl->rl_used) - index - 1) *
1267 sizeof(struct ocfs2_refcount_rec));
1268 memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1269 0, sizeof(struct ocfs2_refcount_rec));
1272 le16_add_cpu(&rl->rl_used, -1);
1274 ocfs2_refcount_rec_merge(rb, index);
1276 ocfs2_journal_dirty(handle, ref_leaf_bh);
1281 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1282 struct ocfs2_caching_info *ci,
1283 struct buffer_head *ref_root_bh,
1284 struct buffer_head **ref_leaf_bh,
1285 struct ocfs2_alloc_context *meta_ac)
1288 u16 suballoc_bit_start;
1290 u64 suballoc_loc, blkno;
1291 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1292 struct buffer_head *new_bh = NULL;
1293 struct ocfs2_refcount_block *new_rb;
1294 struct ocfs2_refcount_block *root_rb =
1295 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1297 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1298 OCFS2_JOURNAL_ACCESS_WRITE);
1304 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1305 &suballoc_bit_start, &num_got,
1312 new_bh = sb_getblk(sb, blkno);
1313 if (new_bh == NULL) {
1318 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1320 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1321 OCFS2_JOURNAL_ACCESS_CREATE);
1328 * Initialize ocfs2_refcount_block.
1329 * It should contain the same information as the old root.
1330 * so just memcpy it and change the corresponding field.
1332 memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1334 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1335 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1336 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1337 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1338 new_rb->rf_blkno = cpu_to_le64(blkno);
1339 new_rb->rf_cpos = cpu_to_le32(0);
1340 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1341 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1342 ocfs2_journal_dirty(handle, new_bh);
1344 /* Now change the root. */
1345 memset(&root_rb->rf_list, 0, sb->s_blocksize -
1346 offsetof(struct ocfs2_refcount_block, rf_list));
1347 root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1348 root_rb->rf_clusters = cpu_to_le32(1);
1349 root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1350 root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1351 root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1352 root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1354 ocfs2_journal_dirty(handle, ref_root_bh);
1356 trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1357 le16_to_cpu(new_rb->rf_records.rl_used));
1359 *ref_leaf_bh = new_bh;
1366 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1367 struct ocfs2_refcount_rec *next)
1369 if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1370 ocfs2_get_ref_rec_low_cpos(next))
1376 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1378 const struct ocfs2_refcount_rec *l = a, *r = b;
1379 u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1380 u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1382 if (l_cpos > r_cpos)
1384 if (l_cpos < r_cpos)
1389 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1391 const struct ocfs2_refcount_rec *l = a, *r = b;
1392 u64 l_cpos = le64_to_cpu(l->r_cpos);
1393 u64 r_cpos = le64_to_cpu(r->r_cpos);
1395 if (l_cpos > r_cpos)
1397 if (l_cpos < r_cpos)
1402 static void swap_refcount_rec(void *a, void *b, int size)
1404 struct ocfs2_refcount_rec *l = a, *r = b;
1410 * The refcount cpos are ordered by their 64bit cpos,
1411 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1412 * So we need to make sure that this pos isn't intersected with others.
1414 * Note: The refcount block is already sorted by their low 32 bit cpos,
1415 * So just try the middle pos first, and we will exit when we find
1416 * the good position.
1418 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1419 u32 *split_pos, int *split_index)
1421 int num_used = le16_to_cpu(rl->rl_used);
1422 int delta, middle = num_used / 2;
1424 for (delta = 0; delta < middle; delta++) {
1425 /* Let's check delta earlier than middle */
1426 if (ocfs2_refcount_rec_no_intersect(
1427 &rl->rl_recs[middle - delta - 1],
1428 &rl->rl_recs[middle - delta])) {
1429 *split_index = middle - delta;
1433 /* For even counts, don't walk off the end */
1434 if ((middle + delta + 1) == num_used)
1437 /* Now try delta past middle */
1438 if (ocfs2_refcount_rec_no_intersect(
1439 &rl->rl_recs[middle + delta],
1440 &rl->rl_recs[middle + delta + 1])) {
1441 *split_index = middle + delta + 1;
1446 if (delta >= middle)
1449 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1453 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1454 struct buffer_head *new_bh,
1457 int split_index = 0, num_moved, ret;
1459 struct ocfs2_refcount_block *rb =
1460 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1461 struct ocfs2_refcount_list *rl = &rb->rf_records;
1462 struct ocfs2_refcount_block *new_rb =
1463 (struct ocfs2_refcount_block *)new_bh->b_data;
1464 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1466 trace_ocfs2_divide_leaf_refcount_block(
1467 (unsigned long long)ref_leaf_bh->b_blocknr,
1468 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1471 * XXX: Improvement later.
1472 * If we know all the high 32 bit cpos is the same, no need to sort.
1474 * In order to make the whole process safe, we do:
1475 * 1. sort the entries by their low 32 bit cpos first so that we can
1476 * find the split cpos easily.
1477 * 2. call ocfs2_insert_extent to insert the new refcount block.
1478 * 3. move the refcount rec to the new block.
1479 * 4. sort the entries by their 64 bit cpos.
1480 * 5. dirty the new_rb and rb.
1482 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1483 sizeof(struct ocfs2_refcount_rec),
1484 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1486 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1492 new_rb->rf_cpos = cpu_to_le32(cpos);
1494 /* move refcount records starting from split_index to the new block. */
1495 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1496 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1497 num_moved * sizeof(struct ocfs2_refcount_rec));
1499 /*ok, remove the entries we just moved over to the other block. */
1500 memset(&rl->rl_recs[split_index], 0,
1501 num_moved * sizeof(struct ocfs2_refcount_rec));
1503 /* change old and new rl_used accordingly. */
1504 le16_add_cpu(&rl->rl_used, -num_moved);
1505 new_rl->rl_used = cpu_to_le16(num_moved);
1507 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1508 sizeof(struct ocfs2_refcount_rec),
1509 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1511 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1512 sizeof(struct ocfs2_refcount_rec),
1513 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1519 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1520 struct ocfs2_caching_info *ci,
1521 struct buffer_head *ref_root_bh,
1522 struct buffer_head *ref_leaf_bh,
1523 struct ocfs2_alloc_context *meta_ac)
1526 u16 suballoc_bit_start;
1527 u32 num_got, new_cpos;
1528 u64 suballoc_loc, blkno;
1529 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1530 struct ocfs2_refcount_block *root_rb =
1531 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1532 struct buffer_head *new_bh = NULL;
1533 struct ocfs2_refcount_block *new_rb;
1534 struct ocfs2_extent_tree ref_et;
1536 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1538 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1539 OCFS2_JOURNAL_ACCESS_WRITE);
1545 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1546 OCFS2_JOURNAL_ACCESS_WRITE);
1552 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1553 &suballoc_bit_start, &num_got,
1560 new_bh = sb_getblk(sb, blkno);
1561 if (new_bh == NULL) {
1566 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1568 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1569 OCFS2_JOURNAL_ACCESS_CREATE);
1575 /* Initialize ocfs2_refcount_block. */
1576 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1577 memset(new_rb, 0, sb->s_blocksize);
1578 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1579 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1580 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1581 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1582 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1583 new_rb->rf_blkno = cpu_to_le64(blkno);
1584 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1585 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1586 new_rb->rf_records.rl_count =
1587 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1588 new_rb->rf_generation = root_rb->rf_generation;
1590 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1596 ocfs2_journal_dirty(handle, ref_leaf_bh);
1597 ocfs2_journal_dirty(handle, new_bh);
1599 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1601 trace_ocfs2_new_leaf_refcount_block(
1602 (unsigned long long)new_bh->b_blocknr, new_cpos);
1604 /* Insert the new leaf block with the specific offset cpos. */
1605 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1615 static int ocfs2_expand_refcount_tree(handle_t *handle,
1616 struct ocfs2_caching_info *ci,
1617 struct buffer_head *ref_root_bh,
1618 struct buffer_head *ref_leaf_bh,
1619 struct ocfs2_alloc_context *meta_ac)
1622 struct buffer_head *expand_bh = NULL;
1624 if (ref_root_bh == ref_leaf_bh) {
1626 * the old root bh hasn't been expanded to a b-tree,
1627 * so expand it first.
1629 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1630 &expand_bh, meta_ac);
1636 expand_bh = ref_leaf_bh;
1641 /* Now add a new refcount block into the tree.*/
1642 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1643 expand_bh, meta_ac);
1652 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1654 * Only called when we have inserted a new refcount rec at index 0
1655 * which means ocfs2_extent_rec.e_cpos may need some change.
1657 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1658 struct ocfs2_caching_info *ci,
1659 struct buffer_head *ref_root_bh,
1660 struct buffer_head *ref_leaf_bh,
1661 struct ocfs2_refcount_rec *rec)
1664 u32 new_cpos, old_cpos;
1665 struct ocfs2_path *path = NULL;
1666 struct ocfs2_extent_tree et;
1667 struct ocfs2_refcount_block *rb =
1668 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1669 struct ocfs2_extent_list *el;
1671 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1674 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1675 old_cpos = le32_to_cpu(rb->rf_cpos);
1676 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1677 if (old_cpos <= new_cpos)
1680 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1682 path = ocfs2_new_path_from_et(&et);
1689 ret = ocfs2_find_path(ci, path, old_cpos);
1696 * 2 more credits, one for the leaf refcount block, one for
1697 * the extent block contains the extent rec.
1699 ret = ocfs2_extend_trans(handle, 2);
1705 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1706 OCFS2_JOURNAL_ACCESS_WRITE);
1712 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1713 OCFS2_JOURNAL_ACCESS_WRITE);
1719 /* change the leaf extent block first. */
1720 el = path_leaf_el(path);
1722 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1723 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1726 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1728 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1730 /* change the r_cpos in the leaf block. */
1731 rb->rf_cpos = cpu_to_le32(new_cpos);
1733 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1734 ocfs2_journal_dirty(handle, ref_leaf_bh);
1737 ocfs2_free_path(path);
1741 static int ocfs2_insert_refcount_rec(handle_t *handle,
1742 struct ocfs2_caching_info *ci,
1743 struct buffer_head *ref_root_bh,
1744 struct buffer_head *ref_leaf_bh,
1745 struct ocfs2_refcount_rec *rec,
1746 int index, int merge,
1747 struct ocfs2_alloc_context *meta_ac)
1750 struct ocfs2_refcount_block *rb =
1751 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1752 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1753 struct buffer_head *new_bh = NULL;
1755 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1757 if (rf_list->rl_used == rf_list->rl_count) {
1758 u64 cpos = le64_to_cpu(rec->r_cpos);
1759 u32 len = le32_to_cpu(rec->r_clusters);
1761 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1762 ref_leaf_bh, meta_ac);
1768 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1769 cpos, len, NULL, &index,
1776 ref_leaf_bh = new_bh;
1777 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1778 rf_list = &rb->rf_records;
1781 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1782 OCFS2_JOURNAL_ACCESS_WRITE);
1788 if (index < le16_to_cpu(rf_list->rl_used))
1789 memmove(&rf_list->rl_recs[index + 1],
1790 &rf_list->rl_recs[index],
1791 (le16_to_cpu(rf_list->rl_used) - index) *
1792 sizeof(struct ocfs2_refcount_rec));
1794 trace_ocfs2_insert_refcount_rec(
1795 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1796 (unsigned long long)le64_to_cpu(rec->r_cpos),
1797 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1799 rf_list->rl_recs[index] = *rec;
1801 le16_add_cpu(&rf_list->rl_used, 1);
1804 ocfs2_refcount_rec_merge(rb, index);
1806 ocfs2_journal_dirty(handle, ref_leaf_bh);
1809 ret = ocfs2_adjust_refcount_rec(handle, ci,
1821 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1822 * This is much simple than our b-tree code.
1823 * split_rec is the new refcount rec we want to insert.
1824 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1825 * increase refcount or decrease a refcount to non-zero).
1826 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1827 * rec( in case we decrease a refcount to zero).
1829 static int ocfs2_split_refcount_rec(handle_t *handle,
1830 struct ocfs2_caching_info *ci,
1831 struct buffer_head *ref_root_bh,
1832 struct buffer_head *ref_leaf_bh,
1833 struct ocfs2_refcount_rec *split_rec,
1834 int index, int merge,
1835 struct ocfs2_alloc_context *meta_ac,
1836 struct ocfs2_cached_dealloc_ctxt *dealloc)
1840 struct ocfs2_refcount_block *rb =
1841 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1842 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1843 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1844 struct ocfs2_refcount_rec *tail_rec = NULL;
1845 struct buffer_head *new_bh = NULL;
1847 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1849 trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1850 le32_to_cpu(orig_rec->r_clusters),
1851 le32_to_cpu(orig_rec->r_refcount),
1852 le64_to_cpu(split_rec->r_cpos),
1853 le32_to_cpu(split_rec->r_clusters),
1854 le32_to_cpu(split_rec->r_refcount));
1857 * If we just need to split the header or tail clusters,
1858 * no more recs are needed, just split is OK.
1859 * Otherwise we at least need one new recs.
1861 if (!split_rec->r_refcount &&
1862 (split_rec->r_cpos == orig_rec->r_cpos ||
1863 le64_to_cpu(split_rec->r_cpos) +
1864 le32_to_cpu(split_rec->r_clusters) ==
1865 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1871 * We need one more rec if we split in the middle and the new rec have
1872 * some refcount in it.
1874 if (split_rec->r_refcount &&
1875 (split_rec->r_cpos != orig_rec->r_cpos &&
1876 le64_to_cpu(split_rec->r_cpos) +
1877 le32_to_cpu(split_rec->r_clusters) !=
1878 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1881 /* If the leaf block don't have enough record, expand it. */
1882 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1883 le16_to_cpu(rf_list->rl_count)) {
1884 struct ocfs2_refcount_rec tmp_rec;
1885 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1886 len = le32_to_cpu(orig_rec->r_clusters);
1887 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1888 ref_leaf_bh, meta_ac);
1895 * We have to re-get it since now cpos may be moved to
1896 * another leaf block.
1898 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1899 cpos, len, &tmp_rec, &index,
1906 ref_leaf_bh = new_bh;
1907 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1908 rf_list = &rb->rf_records;
1909 orig_rec = &rf_list->rl_recs[index];
1912 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1913 OCFS2_JOURNAL_ACCESS_WRITE);
1920 * We have calculated out how many new records we need and store
1921 * in recs_need, so spare enough space first by moving the records
1922 * after "index" to the end.
1924 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1925 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1926 &rf_list->rl_recs[index + 1],
1927 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1928 sizeof(struct ocfs2_refcount_rec));
1930 len = (le64_to_cpu(orig_rec->r_cpos) +
1931 le32_to_cpu(orig_rec->r_clusters)) -
1932 (le64_to_cpu(split_rec->r_cpos) +
1933 le32_to_cpu(split_rec->r_clusters));
1936 * If we have "len", the we will split in the tail and move it
1937 * to the end of the space we have just spared.
1940 tail_rec = &rf_list->rl_recs[index + recs_need];
1942 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1943 le64_add_cpu(&tail_rec->r_cpos,
1944 le32_to_cpu(tail_rec->r_clusters) - len);
1945 tail_rec->r_clusters = cpu_to_le32(len);
1949 * If the split pos isn't the same as the original one, we need to
1950 * split in the head.
1952 * Note: We have the chance that split_rec.r_refcount = 0,
1953 * recs_need = 0 and len > 0, which means we just cut the head from
1954 * the orig_rec and in that case we have done some modification in
1955 * orig_rec above, so the check for r_cpos is faked.
1957 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1958 len = le64_to_cpu(split_rec->r_cpos) -
1959 le64_to_cpu(orig_rec->r_cpos);
1960 orig_rec->r_clusters = cpu_to_le32(len);
1964 le16_add_cpu(&rf_list->rl_used, recs_need);
1966 if (split_rec->r_refcount) {
1967 rf_list->rl_recs[index] = *split_rec;
1968 trace_ocfs2_split_refcount_rec_insert(
1969 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1970 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1971 le32_to_cpu(split_rec->r_clusters),
1972 le32_to_cpu(split_rec->r_refcount));
1975 ocfs2_refcount_rec_merge(rb, index);
1978 ocfs2_journal_dirty(handle, ref_leaf_bh);
1985 static int __ocfs2_increase_refcount(handle_t *handle,
1986 struct ocfs2_caching_info *ci,
1987 struct buffer_head *ref_root_bh,
1988 u64 cpos, u32 len, int merge,
1989 struct ocfs2_alloc_context *meta_ac,
1990 struct ocfs2_cached_dealloc_ctxt *dealloc)
1993 struct buffer_head *ref_leaf_bh = NULL;
1994 struct ocfs2_refcount_rec rec;
1995 unsigned int set_len = 0;
1997 trace_ocfs2_increase_refcount_begin(
1998 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1999 (unsigned long long)cpos, len);
2002 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2003 cpos, len, &rec, &index,
2010 set_len = le32_to_cpu(rec.r_clusters);
2013 * Here we may meet with 3 situations:
2015 * 1. If we find an already existing record, and the length
2016 * is the same, cool, we just need to increase the r_refcount
2018 * 2. If we find a hole, just insert it with r_refcount = 1.
2019 * 3. If we are in the middle of one extent record, split
2022 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2024 trace_ocfs2_increase_refcount_change(
2025 (unsigned long long)cpos, set_len,
2026 le32_to_cpu(rec.r_refcount));
2027 ret = ocfs2_change_refcount_rec(handle, ci,
2034 } else if (!rec.r_refcount) {
2035 rec.r_refcount = cpu_to_le32(1);
2037 trace_ocfs2_increase_refcount_insert(
2038 (unsigned long long)le64_to_cpu(rec.r_cpos),
2040 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2049 set_len = min((u64)(cpos + len),
2050 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2051 rec.r_cpos = cpu_to_le64(cpos);
2052 rec.r_clusters = cpu_to_le32(set_len);
2053 le32_add_cpu(&rec.r_refcount, 1);
2055 trace_ocfs2_increase_refcount_split(
2056 (unsigned long long)le64_to_cpu(rec.r_cpos),
2057 set_len, le32_to_cpu(rec.r_refcount));
2058 ret = ocfs2_split_refcount_rec(handle, ci,
2059 ref_root_bh, ref_leaf_bh,
2070 brelse(ref_leaf_bh);
2075 brelse(ref_leaf_bh);
2079 static int ocfs2_remove_refcount_extent(handle_t *handle,
2080 struct ocfs2_caching_info *ci,
2081 struct buffer_head *ref_root_bh,
2082 struct buffer_head *ref_leaf_bh,
2083 struct ocfs2_alloc_context *meta_ac,
2084 struct ocfs2_cached_dealloc_ctxt *dealloc)
2087 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2088 struct ocfs2_refcount_block *rb =
2089 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2090 struct ocfs2_extent_tree et;
2092 BUG_ON(rb->rf_records.rl_used);
2094 trace_ocfs2_remove_refcount_extent(
2095 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2096 (unsigned long long)ref_leaf_bh->b_blocknr,
2097 le32_to_cpu(rb->rf_cpos));
2099 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2100 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2101 1, meta_ac, dealloc);
2107 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2110 * add the freed block to the dealloc so that it will be freed
2111 * when we run dealloc.
2113 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2114 le16_to_cpu(rb->rf_suballoc_slot),
2115 le64_to_cpu(rb->rf_suballoc_loc),
2116 le64_to_cpu(rb->rf_blkno),
2117 le16_to_cpu(rb->rf_suballoc_bit));
2123 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2124 OCFS2_JOURNAL_ACCESS_WRITE);
2130 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2132 le32_add_cpu(&rb->rf_clusters, -1);
2135 * check whether we need to restore the root refcount block if
2136 * there is no leaf extent block at atll.
2138 if (!rb->rf_list.l_next_free_rec) {
2139 BUG_ON(rb->rf_clusters);
2141 trace_ocfs2_restore_refcount_block(
2142 (unsigned long long)ref_root_bh->b_blocknr);
2147 memset(&rb->rf_records, 0, sb->s_blocksize -
2148 offsetof(struct ocfs2_refcount_block, rf_records));
2149 rb->rf_records.rl_count =
2150 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2153 ocfs2_journal_dirty(handle, ref_root_bh);
2159 int ocfs2_increase_refcount(handle_t *handle,
2160 struct ocfs2_caching_info *ci,
2161 struct buffer_head *ref_root_bh,
2163 struct ocfs2_alloc_context *meta_ac,
2164 struct ocfs2_cached_dealloc_ctxt *dealloc)
2166 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2171 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2172 struct ocfs2_caching_info *ci,
2173 struct buffer_head *ref_root_bh,
2174 struct buffer_head *ref_leaf_bh,
2175 int index, u64 cpos, unsigned int len,
2176 struct ocfs2_alloc_context *meta_ac,
2177 struct ocfs2_cached_dealloc_ctxt *dealloc)
2180 struct ocfs2_refcount_block *rb =
2181 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2182 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2184 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2186 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2188 trace_ocfs2_decrease_refcount_rec(
2189 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2190 (unsigned long long)cpos, len);
2192 if (cpos == le64_to_cpu(rec->r_cpos) &&
2193 len == le32_to_cpu(rec->r_clusters))
2194 ret = ocfs2_change_refcount_rec(handle, ci,
2195 ref_leaf_bh, index, 1, -1);
2197 struct ocfs2_refcount_rec split = *rec;
2198 split.r_cpos = cpu_to_le64(cpos);
2199 split.r_clusters = cpu_to_le32(len);
2201 le32_add_cpu(&split.r_refcount, -1);
2203 ret = ocfs2_split_refcount_rec(handle, ci,
2204 ref_root_bh, ref_leaf_bh,
2214 /* Remove the leaf refcount block if it contains no refcount record. */
2215 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2216 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2217 ref_leaf_bh, meta_ac,
2227 static int __ocfs2_decrease_refcount(handle_t *handle,
2228 struct ocfs2_caching_info *ci,
2229 struct buffer_head *ref_root_bh,
2231 struct ocfs2_alloc_context *meta_ac,
2232 struct ocfs2_cached_dealloc_ctxt *dealloc,
2235 int ret = 0, index = 0;
2236 struct ocfs2_refcount_rec rec;
2237 unsigned int r_count = 0, r_len;
2238 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2239 struct buffer_head *ref_leaf_bh = NULL;
2241 trace_ocfs2_decrease_refcount(
2242 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2243 (unsigned long long)cpos, len, delete);
2246 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2247 cpos, len, &rec, &index,
2254 r_count = le32_to_cpu(rec.r_refcount);
2255 BUG_ON(r_count == 0);
2257 BUG_ON(r_count > 1);
2259 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2260 le32_to_cpu(rec.r_clusters)) - cpos;
2262 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2271 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2272 ret = ocfs2_cache_cluster_dealloc(dealloc,
2273 ocfs2_clusters_to_blocks(sb, cpos),
2283 brelse(ref_leaf_bh);
2288 brelse(ref_leaf_bh);
2292 /* Caller must hold refcount tree lock. */
2293 int ocfs2_decrease_refcount(struct inode *inode,
2294 handle_t *handle, u32 cpos, u32 len,
2295 struct ocfs2_alloc_context *meta_ac,
2296 struct ocfs2_cached_dealloc_ctxt *dealloc,
2301 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2302 struct buffer_head *ref_root_bh = NULL;
2303 struct ocfs2_refcount_tree *tree;
2305 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2307 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2313 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2319 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2326 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2327 cpos, len, meta_ac, dealloc, delete);
2331 brelse(ref_root_bh);
2336 * Mark the already-existing extent at cpos as refcounted for len clusters.
2337 * This adds the refcount extent flag.
2339 * If the existing extent is larger than the request, initiate a
2340 * split. An attempt will be made at merging with adjacent extents.
2342 * The caller is responsible for passing down meta_ac if we'll need it.
2344 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2345 struct ocfs2_extent_tree *et,
2346 handle_t *handle, u32 cpos,
2348 struct ocfs2_alloc_context *meta_ac,
2349 struct ocfs2_cached_dealloc_ctxt *dealloc)
2353 trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2356 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2357 ret = ocfs2_error(inode->i_sb, "Inode %lu want to use refcount tree, but the feature bit is not set in the super block\n",
2362 ret = ocfs2_change_extent_flag(handle, et, cpos,
2363 len, phys, meta_ac, dealloc,
2364 OCFS2_EXT_REFCOUNTED, 0);
2373 * Given some contiguous physical clusters, calculate what we need
2374 * for modifying their refcount.
2376 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2377 struct ocfs2_caching_info *ci,
2378 struct buffer_head *ref_root_bh,
2384 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2385 u64 cpos = start_cpos;
2386 struct ocfs2_refcount_block *rb;
2387 struct ocfs2_refcount_rec rec;
2388 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2392 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2393 cpos, clusters, &rec,
2394 &index, &ref_leaf_bh);
2400 if (ref_leaf_bh != prev_bh) {
2402 * Now we encounter a new leaf block, so calculate
2403 * whether we need to extend the old leaf.
2406 rb = (struct ocfs2_refcount_block *)
2409 if (le16_to_cpu(rb->rf_records.rl_used) +
2411 le16_to_cpu(rb->rf_records.rl_count))
2418 prev_bh = ref_leaf_bh;
2422 trace_ocfs2_calc_refcount_meta_credits_iterate(
2423 recs_add, (unsigned long long)cpos, clusters,
2424 (unsigned long long)le64_to_cpu(rec.r_cpos),
2425 le32_to_cpu(rec.r_clusters),
2426 le32_to_cpu(rec.r_refcount), index);
2428 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2429 le32_to_cpu(rec.r_clusters)) - cpos;
2431 * We record all the records which will be inserted to the
2432 * same refcount block, so that we can tell exactly whether
2433 * we need a new refcount block or not.
2435 * If we will insert a new one, this is easy and only happens
2436 * during adding refcounted flag to the extent, so we don't
2437 * have a chance of spliting. We just need one record.
2439 * If the refcount rec already exists, that would be a little
2440 * complicated. we may have to:
2441 * 1) split at the beginning if the start pos isn't aligned.
2442 * we need 1 more record in this case.
2443 * 2) split int the end if the end pos isn't aligned.
2444 * we need 1 more record in this case.
2445 * 3) split in the middle because of file system fragmentation.
2446 * we need 2 more records in this case(we can't detect this
2447 * beforehand, so always think of the worst case).
2449 if (rec.r_refcount) {
2451 /* Check whether we need a split at the beginning. */
2452 if (cpos == start_cpos &&
2453 cpos != le64_to_cpu(rec.r_cpos))
2456 /* Check whether we need a split in the end. */
2457 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2458 le32_to_cpu(rec.r_clusters))
2463 brelse(ref_leaf_bh);
2470 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2472 if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2473 le16_to_cpu(rb->rf_records.rl_count))
2482 *meta_add += ref_blocks;
2483 *credits += ref_blocks;
2486 * So we may need ref_blocks to insert into the tree.
2487 * That also means we need to change the b-tree and add that number
2488 * of records since we never merge them.
2489 * We need one more block for expansion since the new created leaf
2490 * block is also full and needs split.
2492 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2493 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2494 struct ocfs2_extent_tree et;
2496 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2497 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2498 *credits += ocfs2_calc_extend_credits(sb,
2501 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2507 trace_ocfs2_calc_refcount_meta_credits(
2508 (unsigned long long)start_cpos, clusters,
2509 *meta_add, *credits);
2510 brelse(ref_leaf_bh);
2516 * For refcount tree, we will decrease some contiguous clusters
2517 * refcount count, so just go through it to see how many blocks
2518 * we gonna touch and whether we need to create new blocks.
2520 * Normally the refcount blocks store these refcount should be
2521 * contiguous also, so that we can get the number easily.
2522 * We will at most add split 2 refcount records and 2 more
2523 * refcount blocks, so just check it in a rough way.
2525 * Caller must hold refcount tree lock.
2527 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2535 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2536 struct buffer_head *ref_root_bh = NULL;
2537 struct ocfs2_refcount_tree *tree;
2538 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2540 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2541 ret = ocfs2_error(inode->i_sb, "Inode %lu want to use refcount tree, but the feature bit is not set in the super block\n",
2546 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2548 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2549 refcount_loc, &tree);
2555 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2562 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2565 start_cpos, clusters,
2566 ref_blocks, credits);
2572 trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2575 brelse(ref_root_bh);
2579 #define MAX_CONTIG_BYTES 1048576
2581 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2583 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2586 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2588 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2592 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2593 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2594 * while still being less than or equal to it.
2596 * The goal is to break the extent at a multiple of contig_clusters.
2598 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2602 BUG_ON(start > cpos);
2604 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2608 * Given a cluster count of len, pad it out so that it is a multiple
2609 * of contig_clusters.
2611 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2614 unsigned int padded =
2615 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2616 ocfs2_cow_contig_mask(sb);
2626 * Calculate out the start and number of virtual clusters we need to to CoW.
2628 * cpos is vitual start cluster position we want to do CoW in a
2629 * file and write_len is the cluster length.
2630 * max_cpos is the place where we want to stop CoW intentionally.
2632 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2633 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2634 * get good I/O from the resulting extent tree.
2636 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2637 struct ocfs2_extent_list *el,
2645 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2646 struct buffer_head *eb_bh = NULL;
2647 struct ocfs2_extent_block *eb = NULL;
2648 struct ocfs2_extent_rec *rec;
2649 unsigned int want_clusters, rec_end = 0;
2650 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2653 BUG_ON(cpos + write_len > max_cpos);
2655 if (tree_height > 0) {
2656 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2662 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2665 if (el->l_tree_depth) {
2666 ret = ocfs2_error(inode->i_sb,
2667 "Inode %lu has non zero tree depth in leaf block %llu\n",
2669 (unsigned long long)eb_bh->b_blocknr);
2675 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2676 rec = &el->l_recs[i];
2678 if (ocfs2_is_empty_extent(rec)) {
2679 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2680 "index %d\n", inode->i_ino, i);
2684 if (le32_to_cpu(rec->e_cpos) +
2685 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2688 if (*cow_len == 0) {
2690 * We should find a refcounted record in the
2693 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2694 *cow_start = le32_to_cpu(rec->e_cpos);
2698 * If we encounter a hole, a non-refcounted record or
2699 * pass the max_cpos, stop the search.
2701 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2702 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2703 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2706 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2707 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2708 if (rec_end > max_cpos) {
2710 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2714 * How many clusters do we actually need from
2715 * this extent? First we see how many we actually
2716 * need to complete the write. If that's smaller
2717 * than contig_clusters, we try for contig_clusters.
2720 want_clusters = write_len;
2722 want_clusters = (cpos + write_len) -
2723 (*cow_start + *cow_len);
2724 if (want_clusters < contig_clusters)
2725 want_clusters = contig_clusters;
2728 * If the write does not cover the whole extent, we
2729 * need to calculate how we're going to split the extent.
2730 * We try to do it on contig_clusters boundaries.
2732 * Any extent smaller than contig_clusters will be
2733 * CoWed in its entirety.
2735 if (leaf_clusters <= contig_clusters)
2736 *cow_len += leaf_clusters;
2737 else if (*cow_len || (*cow_start == cpos)) {
2739 * This extent needs to be CoW'd from its
2740 * beginning, so all we have to do is compute
2741 * how many clusters to grab. We align
2742 * want_clusters to the edge of contig_clusters
2743 * to get better I/O.
2745 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2748 if (leaf_clusters < want_clusters)
2749 *cow_len += leaf_clusters;
2751 *cow_len += want_clusters;
2752 } else if ((*cow_start + contig_clusters) >=
2753 (cpos + write_len)) {
2755 * Breaking off contig_clusters at the front
2756 * of the extent will cover our write. That's
2759 *cow_len = contig_clusters;
2760 } else if ((rec_end - cpos) <= contig_clusters) {
2762 * Breaking off contig_clusters at the tail of
2763 * this extent will cover cpos.
2765 *cow_start = rec_end - contig_clusters;
2766 *cow_len = contig_clusters;
2767 } else if ((rec_end - cpos) <= want_clusters) {
2769 * While we can't fit the entire write in this
2770 * extent, we know that the write goes from cpos
2771 * to the end of the extent. Break that off.
2772 * We try to break it at some multiple of
2773 * contig_clusters from the front of the extent.
2774 * Failing that (ie, cpos is within
2775 * contig_clusters of the front), we'll CoW the
2778 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2780 *cow_len = rec_end - *cow_start;
2783 * Ok, the entire write lives in the middle of
2784 * this extent. Let's try to slice the extent up
2785 * nicely. Optimally, our CoW region starts at
2786 * m*contig_clusters from the beginning of the
2787 * extent and goes for n*contig_clusters,
2788 * covering the entire write.
2790 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2793 want_clusters = (cpos + write_len) - *cow_start;
2794 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2796 if (*cow_start + want_clusters <= rec_end)
2797 *cow_len = want_clusters;
2799 *cow_len = rec_end - *cow_start;
2802 /* Have we covered our entire write yet? */
2803 if ((*cow_start + *cow_len) >= (cpos + write_len))
2807 * If we reach the end of the extent block and don't get enough
2808 * clusters, continue with the next extent block if possible.
2810 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2811 eb && eb->h_next_leaf_blk) {
2815 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2816 le64_to_cpu(eb->h_next_leaf_blk),
2823 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2835 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2836 * num_clusters in data_tree "et" and change the refcount for the old
2837 * clusters(starting form p_cluster) in the refcount tree.
2840 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2841 * more new leaf records.
2842 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2843 * just give data_ac = NULL.
2845 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2846 u32 p_cluster, u32 num_clusters,
2847 struct ocfs2_extent_tree *et,
2848 struct ocfs2_caching_info *ref_ci,
2849 struct buffer_head *ref_root_bh,
2850 struct ocfs2_alloc_context **meta_ac,
2851 struct ocfs2_alloc_context **data_ac,
2854 int ret = 0, meta_add = 0;
2855 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2857 if (num_free_extents < 0) {
2858 ret = num_free_extents;
2863 if (num_free_extents < num_clusters + 2)
2865 ocfs2_extend_meta_needed(et->et_root_el);
2867 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2869 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2870 p_cluster, num_clusters,
2871 &meta_add, credits);
2877 trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2878 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2886 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2895 ocfs2_free_alloc_context(*meta_ac);
2903 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2905 BUG_ON(buffer_dirty(bh));
2907 clear_buffer_mapped(bh);
2912 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2913 struct inode *inode,
2914 u32 cpos, u32 old_cluster,
2915 u32 new_cluster, u32 new_len)
2917 int ret = 0, partial;
2918 struct super_block *sb = inode->i_sb;
2919 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2922 unsigned int from, to;
2923 loff_t offset, end, map_end;
2924 struct address_space *mapping = inode->i_mapping;
2926 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2927 new_cluster, new_len);
2929 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2930 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2932 * We only duplicate pages until we reach the page contains i_size - 1.
2933 * So trim 'end' to i_size.
2935 if (end > i_size_read(inode))
2936 end = i_size_read(inode);
2938 while (offset < end) {
2939 page_index = offset >> PAGE_SHIFT;
2940 map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
2944 /* from, to is the offset within the page. */
2945 from = offset & (PAGE_SIZE - 1);
2947 if (map_end & (PAGE_SIZE - 1))
2948 to = map_end & (PAGE_SIZE - 1);
2950 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2958 * In case PAGE_SIZE <= CLUSTER_SIZE, This page
2959 * can't be dirtied before we CoW it out.
2961 if (PAGE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2962 BUG_ON(PageDirty(page));
2964 if (!PageUptodate(page)) {
2965 ret = block_read_full_page(page, ocfs2_get_block);
2973 if (page_has_buffers(page)) {
2974 ret = walk_page_buffers(handle, page_buffers(page),
2976 ocfs2_clear_cow_buffer);
2983 ocfs2_map_and_dirty_page(inode,
2985 page, 0, &new_block);
2986 mark_page_accessed(page);
2999 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3000 struct inode *inode,
3001 u32 cpos, u32 old_cluster,
3002 u32 new_cluster, u32 new_len)
3005 struct super_block *sb = inode->i_sb;
3006 struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3007 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3008 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3009 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3010 struct ocfs2_super *osb = OCFS2_SB(sb);
3011 struct buffer_head *old_bh = NULL;
3012 struct buffer_head *new_bh = NULL;
3014 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3015 new_cluster, new_len);
3017 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3018 new_bh = sb_getblk(osb->sb, new_block);
3019 if (new_bh == NULL) {
3025 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3027 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3033 ret = ocfs2_journal_access(handle, ci, new_bh,
3034 OCFS2_JOURNAL_ACCESS_CREATE);
3040 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3041 ocfs2_journal_dirty(handle, new_bh);
3054 static int ocfs2_clear_ext_refcount(handle_t *handle,
3055 struct ocfs2_extent_tree *et,
3056 u32 cpos, u32 p_cluster, u32 len,
3057 unsigned int ext_flags,
3058 struct ocfs2_alloc_context *meta_ac,
3059 struct ocfs2_cached_dealloc_ctxt *dealloc)
3062 struct ocfs2_extent_rec replace_rec;
3063 struct ocfs2_path *path = NULL;
3064 struct ocfs2_extent_list *el;
3065 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3066 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3068 trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3069 cpos, len, p_cluster, ext_flags);
3071 memset(&replace_rec, 0, sizeof(replace_rec));
3072 replace_rec.e_cpos = cpu_to_le32(cpos);
3073 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3074 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3076 replace_rec.e_flags = ext_flags;
3077 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3079 path = ocfs2_new_path_from_et(et);
3086 ret = ocfs2_find_path(et->et_ci, path, cpos);
3092 el = path_leaf_el(path);
3094 index = ocfs2_search_extent_list(el, cpos);
3096 ret = ocfs2_error(sb,
3097 "Inode %llu has an extent at cpos %u which can no longer be found\n",
3098 (unsigned long long)ino, cpos);
3102 ret = ocfs2_split_extent(handle, et, path, index,
3103 &replace_rec, meta_ac, dealloc);
3108 ocfs2_free_path(path);
3112 static int ocfs2_replace_clusters(handle_t *handle,
3113 struct ocfs2_cow_context *context,
3116 unsigned int ext_flags)
3119 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3120 u64 ino = ocfs2_metadata_cache_owner(ci);
3122 trace_ocfs2_replace_clusters((unsigned long long)ino,
3123 cpos, old, new, len, ext_flags);
3125 /*If the old clusters is unwritten, no need to duplicate. */
3126 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3127 ret = context->cow_duplicate_clusters(handle, context->inode,
3128 cpos, old, new, len);
3135 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3136 cpos, new, len, ext_flags,
3137 context->meta_ac, &context->dealloc);
3144 int ocfs2_cow_sync_writeback(struct super_block *sb,
3145 struct inode *inode,
3146 u32 cpos, u32 num_clusters)
3149 loff_t offset, end, map_end;
3153 if (ocfs2_should_order_data(inode))
3156 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3157 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3159 ret = filemap_fdatawrite_range(inode->i_mapping,
3166 while (offset < end) {
3167 page_index = offset >> PAGE_SHIFT;
3168 map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
3172 page = find_or_create_page(inode->i_mapping,
3173 page_index, GFP_NOFS);
3176 wait_on_page_writeback(page);
3177 if (PageError(page)) {
3181 mark_page_accessed(page);
3194 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3195 u32 v_cluster, u32 *p_cluster,
3197 unsigned int *extent_flags)
3199 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3200 num_clusters, extent_flags);
3203 static int ocfs2_make_clusters_writable(struct super_block *sb,
3204 struct ocfs2_cow_context *context,
3205 u32 cpos, u32 p_cluster,
3206 u32 num_clusters, unsigned int e_flags)
3208 int ret, delete, index, credits = 0;
3209 u32 new_bit, new_len, orig_num_clusters;
3210 unsigned int set_len;
3211 struct ocfs2_super *osb = OCFS2_SB(sb);
3213 struct buffer_head *ref_leaf_bh = NULL;
3214 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3215 struct ocfs2_refcount_rec rec;
3217 trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3218 num_clusters, e_flags);
3220 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3223 context->ref_root_bh,
3225 &context->data_ac, &credits);
3231 if (context->post_refcount)
3232 credits += context->post_refcount->credits;
3234 credits += context->extra_credits;
3235 handle = ocfs2_start_trans(osb, credits);
3236 if (IS_ERR(handle)) {
3237 ret = PTR_ERR(handle);
3242 orig_num_clusters = num_clusters;
3244 while (num_clusters) {
3245 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3246 p_cluster, num_clusters,
3247 &rec, &index, &ref_leaf_bh);
3253 BUG_ON(!rec.r_refcount);
3254 set_len = min((u64)p_cluster + num_clusters,
3255 le64_to_cpu(rec.r_cpos) +
3256 le32_to_cpu(rec.r_clusters)) - p_cluster;
3259 * There are many different situation here.
3260 * 1. If refcount == 1, remove the flag and don't COW.
3261 * 2. If refcount > 1, allocate clusters.
3262 * Here we may not allocate r_len once at a time, so continue
3263 * until we reach num_clusters.
3265 if (le32_to_cpu(rec.r_refcount) == 1) {
3267 ret = ocfs2_clear_ext_refcount(handle,
3280 ret = __ocfs2_claim_clusters(handle,
3283 &new_bit, &new_len);
3289 ret = ocfs2_replace_clusters(handle, context,
3290 cpos, p_cluster, new_bit,
3299 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3300 context->ref_root_bh,
3303 &context->dealloc, delete);
3310 p_cluster += set_len;
3311 num_clusters -= set_len;
3312 brelse(ref_leaf_bh);
3316 /* handle any post_cow action. */
3317 if (context->post_refcount && context->post_refcount->func) {
3318 ret = context->post_refcount->func(context->inode, handle,
3319 context->post_refcount->para);
3327 * Here we should write the new page out first if we are
3328 * in write-back mode.
3330 if (context->get_clusters == ocfs2_di_get_clusters) {
3331 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3338 ocfs2_commit_trans(osb, handle);
3341 if (context->data_ac) {
3342 ocfs2_free_alloc_context(context->data_ac);
3343 context->data_ac = NULL;
3345 if (context->meta_ac) {
3346 ocfs2_free_alloc_context(context->meta_ac);
3347 context->meta_ac = NULL;
3349 brelse(ref_leaf_bh);
3354 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3357 struct inode *inode = context->inode;
3358 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3359 u32 p_cluster, num_clusters;
3360 unsigned int ext_flags;
3361 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3363 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3364 return ocfs2_error(inode->i_sb, "Inode %lu want to use refcount tree, but the feature bit is not set in the super block\n",
3368 ocfs2_init_dealloc_ctxt(&context->dealloc);
3371 ret = context->get_clusters(context, cow_start, &p_cluster,
3372 &num_clusters, &ext_flags);
3378 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3380 if (cow_len < num_clusters)
3381 num_clusters = cow_len;
3383 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3384 cow_start, p_cluster,
3385 num_clusters, ext_flags);
3391 cow_len -= num_clusters;
3392 cow_start += num_clusters;
3395 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3396 ocfs2_schedule_truncate_log_flush(osb, 1);
3397 ocfs2_run_deallocs(osb, &context->dealloc);
3404 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3405 * past max_cpos. This will stop when it runs into a hole or an
3406 * unrefcounted extent.
3408 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3409 struct buffer_head *di_bh,
3410 u32 cpos, u32 write_len, u32 max_cpos)
3413 u32 cow_start = 0, cow_len = 0;
3414 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3415 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3416 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3417 struct buffer_head *ref_root_bh = NULL;
3418 struct ocfs2_refcount_tree *ref_tree;
3419 struct ocfs2_cow_context *context = NULL;
3421 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3423 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3424 cpos, write_len, max_cpos,
3425 &cow_start, &cow_len);
3431 trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3432 cpos, write_len, max_cpos,
3433 cow_start, cow_len);
3435 BUG_ON(cow_len == 0);
3437 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3444 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3445 1, &ref_tree, &ref_root_bh);
3451 context->inode = inode;
3452 context->cow_start = cow_start;
3453 context->cow_len = cow_len;
3454 context->ref_tree = ref_tree;
3455 context->ref_root_bh = ref_root_bh;
3456 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3457 context->get_clusters = ocfs2_di_get_clusters;
3459 ocfs2_init_dinode_extent_tree(&context->data_et,
3460 INODE_CACHE(inode), di_bh);
3462 ret = ocfs2_replace_cow(context);
3467 * truncate the extent map here since no matter whether we meet with
3468 * any error during the action, we shouldn't trust cached extent map
3471 ocfs2_extent_map_trunc(inode, cow_start);
3473 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3474 brelse(ref_root_bh);
3481 * CoW any and all clusters between cpos and cpos+write_len.
3482 * Don't CoW past max_cpos. If this returns successfully, all
3483 * clusters between cpos and cpos+write_len are safe to modify.
3485 int ocfs2_refcount_cow(struct inode *inode,
3486 struct buffer_head *di_bh,
3487 u32 cpos, u32 write_len, u32 max_cpos)
3490 u32 p_cluster, num_clusters;
3491 unsigned int ext_flags;
3494 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3495 &num_clusters, &ext_flags);
3501 if (write_len < num_clusters)
3502 num_clusters = write_len;
3504 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3505 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3506 num_clusters, max_cpos);
3513 write_len -= num_clusters;
3514 cpos += num_clusters;
3520 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3521 u32 v_cluster, u32 *p_cluster,
3523 unsigned int *extent_flags)
3525 struct inode *inode = context->inode;
3526 struct ocfs2_xattr_value_root *xv = context->cow_object;
3528 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3529 num_clusters, &xv->xr_list,
3534 * Given a xattr value root, calculate the most meta/credits we need for
3535 * refcount tree change if we truncate it to 0.
3537 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3538 struct ocfs2_caching_info *ref_ci,
3539 struct buffer_head *ref_root_bh,
3540 struct ocfs2_xattr_value_root *xv,
3541 int *meta_add, int *credits)
3543 int ret = 0, index, ref_blocks = 0;
3544 u32 p_cluster, num_clusters;
3545 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3546 struct ocfs2_refcount_block *rb;
3547 struct ocfs2_refcount_rec rec;
3548 struct buffer_head *ref_leaf_bh = NULL;
3550 while (cpos < clusters) {
3551 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3552 &num_clusters, &xv->xr_list,
3559 cpos += num_clusters;
3561 while (num_clusters) {
3562 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3563 p_cluster, num_clusters,
3571 BUG_ON(!rec.r_refcount);
3573 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3576 * We really don't know whether the other clusters is in
3577 * this refcount block or not, so just take the worst
3578 * case that all the clusters are in this block and each
3579 * one will split a refcount rec, so totally we need
3580 * clusters * 2 new refcount rec.
3582 if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3583 le16_to_cpu(rb->rf_records.rl_count))
3587 brelse(ref_leaf_bh);
3590 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3593 num_clusters -= le32_to_cpu(rec.r_clusters);
3594 p_cluster += num_clusters;
3598 *meta_add += ref_blocks;
3602 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3603 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3604 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3606 struct ocfs2_extent_tree et;
3608 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3609 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3614 brelse(ref_leaf_bh);
3621 int ocfs2_refcount_cow_xattr(struct inode *inode,
3622 struct ocfs2_dinode *di,
3623 struct ocfs2_xattr_value_buf *vb,
3624 struct ocfs2_refcount_tree *ref_tree,
3625 struct buffer_head *ref_root_bh,
3626 u32 cpos, u32 write_len,
3627 struct ocfs2_post_refcount *post)
3630 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3631 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3632 struct ocfs2_cow_context *context = NULL;
3633 u32 cow_start, cow_len;
3635 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3637 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3638 cpos, write_len, UINT_MAX,
3639 &cow_start, &cow_len);
3645 BUG_ON(cow_len == 0);
3647 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3654 context->inode = inode;
3655 context->cow_start = cow_start;
3656 context->cow_len = cow_len;
3657 context->ref_tree = ref_tree;
3658 context->ref_root_bh = ref_root_bh;
3659 context->cow_object = xv;
3661 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3662 /* We need the extra credits for duplicate_clusters by jbd. */
3663 context->extra_credits =
3664 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3665 context->get_clusters = ocfs2_xattr_value_get_clusters;
3666 context->post_refcount = post;
3668 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3669 INODE_CACHE(inode), vb);
3671 ret = ocfs2_replace_cow(context);
3681 * Insert a new extent into refcount tree and mark a extent rec
3682 * as refcounted in the dinode tree.
3684 int ocfs2_add_refcount_flag(struct inode *inode,
3685 struct ocfs2_extent_tree *data_et,
3686 struct ocfs2_caching_info *ref_ci,
3687 struct buffer_head *ref_root_bh,
3688 u32 cpos, u32 p_cluster, u32 num_clusters,
3689 struct ocfs2_cached_dealloc_ctxt *dealloc,
3690 struct ocfs2_post_refcount *post)
3694 int credits = 1, ref_blocks = 0;
3695 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3696 struct ocfs2_alloc_context *meta_ac = NULL;
3698 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3699 ref_ci, ref_root_bh,
3700 p_cluster, num_clusters,
3701 &ref_blocks, &credits);
3707 trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3710 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3711 ref_blocks, &meta_ac);
3719 credits += post->credits;
3721 handle = ocfs2_start_trans(osb, credits);
3722 if (IS_ERR(handle)) {
3723 ret = PTR_ERR(handle);
3728 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3729 cpos, num_clusters, p_cluster,
3736 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3737 p_cluster, num_clusters, 0,
3744 if (post && post->func) {
3745 ret = post->func(inode, handle, post->para);
3751 ocfs2_commit_trans(osb, handle);
3754 ocfs2_free_alloc_context(meta_ac);
3758 static int ocfs2_change_ctime(struct inode *inode,
3759 struct buffer_head *di_bh)
3763 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3765 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3766 OCFS2_INODE_UPDATE_CREDITS);
3767 if (IS_ERR(handle)) {
3768 ret = PTR_ERR(handle);
3773 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3774 OCFS2_JOURNAL_ACCESS_WRITE);
3780 inode->i_ctime = current_time(inode);
3781 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3782 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3784 ocfs2_journal_dirty(handle, di_bh);
3787 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3792 static int ocfs2_attach_refcount_tree(struct inode *inode,
3793 struct buffer_head *di_bh)
3795 int ret, data_changed = 0;
3796 struct buffer_head *ref_root_bh = NULL;
3797 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3798 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3799 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3800 struct ocfs2_refcount_tree *ref_tree;
3801 unsigned int ext_flags;
3803 u32 cpos, num_clusters, clusters, p_cluster;
3804 struct ocfs2_cached_dealloc_ctxt dealloc;
3805 struct ocfs2_extent_tree di_et;
3807 ocfs2_init_dealloc_ctxt(&dealloc);
3809 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3810 ret = ocfs2_create_refcount_tree(inode, di_bh);
3817 BUG_ON(!di->i_refcount_loc);
3818 ret = ocfs2_lock_refcount_tree(osb,
3819 le64_to_cpu(di->i_refcount_loc), 1,
3820 &ref_tree, &ref_root_bh);
3826 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3829 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3831 size = i_size_read(inode);
3832 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3835 while (cpos < clusters) {
3836 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3837 &num_clusters, &ext_flags);
3842 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3843 ret = ocfs2_add_refcount_flag(inode, &di_et,
3846 p_cluster, num_clusters,
3855 cpos += num_clusters;
3859 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3860 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3871 ret = ocfs2_change_ctime(inode, di_bh);
3877 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3878 brelse(ref_root_bh);
3880 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3881 ocfs2_schedule_truncate_log_flush(osb, 1);
3882 ocfs2_run_deallocs(osb, &dealloc);
3886 * Empty the extent map so that we may get the right extent
3887 * record from the disk.
3889 ocfs2_extent_map_trunc(inode, 0);
3894 static int ocfs2_add_refcounted_extent(struct inode *inode,
3895 struct ocfs2_extent_tree *et,
3896 struct ocfs2_caching_info *ref_ci,
3897 struct buffer_head *ref_root_bh,
3898 u32 cpos, u32 p_cluster, u32 num_clusters,
3899 unsigned int ext_flags,
3900 struct ocfs2_cached_dealloc_ctxt *dealloc)
3905 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3906 struct ocfs2_alloc_context *meta_ac = NULL;
3908 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3909 p_cluster, num_clusters,
3911 ref_root_bh, &meta_ac,
3918 handle = ocfs2_start_trans(osb, credits);
3919 if (IS_ERR(handle)) {
3920 ret = PTR_ERR(handle);
3925 ret = ocfs2_insert_extent(handle, et, cpos,
3926 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3927 num_clusters, ext_flags, meta_ac);
3933 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3934 p_cluster, num_clusters,
3940 ocfs2_commit_trans(osb, handle);
3943 ocfs2_free_alloc_context(meta_ac);
3947 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3948 struct buffer_head *s_bh,
3949 struct inode *t_inode,
3950 struct buffer_head *t_bh)
3954 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3955 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3956 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3958 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3960 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3961 if (IS_ERR(handle)) {
3962 ret = PTR_ERR(handle);
3967 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3968 OCFS2_JOURNAL_ACCESS_WRITE);
3974 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3975 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3976 le16_to_cpu(s_di->id2.i_data.id_count));
3977 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3978 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3979 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3980 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3982 ocfs2_journal_dirty(handle, t_bh);
3985 ocfs2_commit_trans(osb, handle);
3990 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3991 struct inode *t_inode,
3992 struct buffer_head *t_bh,
3993 struct ocfs2_caching_info *ref_ci,
3994 struct buffer_head *ref_root_bh,
3995 struct ocfs2_cached_dealloc_ctxt *dealloc)
3998 u32 p_cluster, num_clusters, clusters, cpos;
4000 unsigned int ext_flags;
4001 struct ocfs2_extent_tree et;
4003 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4005 size = i_size_read(s_inode);
4006 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4009 while (cpos < clusters) {
4010 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4011 &num_clusters, &ext_flags);
4017 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4018 ref_ci, ref_root_bh,
4029 cpos += num_clusters;
4037 * change the new file's attributes to the src.
4039 * reflink creates a snapshot of a file, that means the attributes
4040 * must be identical except for three exceptions - nlink, ino, and ctime.
4042 static int ocfs2_complete_reflink(struct inode *s_inode,
4043 struct buffer_head *s_bh,
4044 struct inode *t_inode,
4045 struct buffer_head *t_bh,
4050 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4051 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4052 loff_t size = i_size_read(s_inode);
4054 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4055 OCFS2_INODE_UPDATE_CREDITS);
4056 if (IS_ERR(handle)) {
4057 ret = PTR_ERR(handle);
4062 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4063 OCFS2_JOURNAL_ACCESS_WRITE);
4069 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4070 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4071 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4072 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4073 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4074 i_size_write(t_inode, size);
4075 t_inode->i_blocks = s_inode->i_blocks;
4077 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4078 di->i_clusters = s_di->i_clusters;
4079 di->i_size = s_di->i_size;
4080 di->i_dyn_features = s_di->i_dyn_features;
4081 di->i_attr = s_di->i_attr;
4084 t_inode->i_uid = s_inode->i_uid;
4085 t_inode->i_gid = s_inode->i_gid;
4086 t_inode->i_mode = s_inode->i_mode;
4087 di->i_uid = s_di->i_uid;
4088 di->i_gid = s_di->i_gid;
4089 di->i_mode = s_di->i_mode;
4093 * we want mtime to appear identical to the source and
4096 t_inode->i_ctime = current_time(t_inode);
4098 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4099 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4101 t_inode->i_mtime = s_inode->i_mtime;
4102 di->i_mtime = s_di->i_mtime;
4103 di->i_mtime_nsec = s_di->i_mtime_nsec;
4106 ocfs2_journal_dirty(handle, t_bh);
4109 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4113 static int ocfs2_create_reflink_node(struct inode *s_inode,
4114 struct buffer_head *s_bh,
4115 struct inode *t_inode,
4116 struct buffer_head *t_bh,
4120 struct buffer_head *ref_root_bh = NULL;
4121 struct ocfs2_cached_dealloc_ctxt dealloc;
4122 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4123 struct ocfs2_refcount_block *rb;
4124 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4125 struct ocfs2_refcount_tree *ref_tree;
4127 ocfs2_init_dealloc_ctxt(&dealloc);
4129 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4130 le64_to_cpu(di->i_refcount_loc));
4136 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4137 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4144 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4145 1, &ref_tree, &ref_root_bh);
4150 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4152 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4153 &ref_tree->rf_ci, ref_root_bh,
4157 goto out_unlock_refcount;
4160 out_unlock_refcount:
4161 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4162 brelse(ref_root_bh);
4164 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4165 ocfs2_schedule_truncate_log_flush(osb, 1);
4166 ocfs2_run_deallocs(osb, &dealloc);
4172 static int __ocfs2_reflink(struct dentry *old_dentry,
4173 struct buffer_head *old_bh,
4174 struct inode *new_inode,
4178 struct inode *inode = d_inode(old_dentry);
4179 struct buffer_head *new_bh = NULL;
4181 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4187 ret = filemap_fdatawrite(inode->i_mapping);
4193 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4199 inode_lock_nested(new_inode, I_MUTEX_CHILD);
4200 ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4201 OI_LS_REFLINK_TARGET);
4207 ret = ocfs2_create_reflink_node(inode, old_bh,
4208 new_inode, new_bh, preserve);
4214 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4215 ret = ocfs2_reflink_xattrs(inode, old_bh,
4224 ret = ocfs2_complete_reflink(inode, old_bh,
4225 new_inode, new_bh, preserve);
4230 ocfs2_inode_unlock(new_inode, 1);
4233 inode_unlock(new_inode);
4236 ret = filemap_fdatawait(inode->i_mapping);
4243 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4244 struct dentry *new_dentry, bool preserve)
4247 struct inode *inode = d_inode(old_dentry);
4248 struct buffer_head *old_bh = NULL;
4249 struct inode *new_orphan_inode = NULL;
4251 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4255 error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4262 error = ocfs2_rw_lock(inode, 1);
4268 error = ocfs2_inode_lock(inode, &old_bh, 1);
4271 ocfs2_rw_unlock(inode, 1);
4275 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4276 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4277 error = __ocfs2_reflink(old_dentry, old_bh,
4278 new_orphan_inode, preserve);
4279 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4280 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4282 ocfs2_inode_unlock(inode, 1);
4283 ocfs2_rw_unlock(inode, 1);
4291 /* If the security isn't preserved, we need to re-initialize them. */
4293 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4294 &new_dentry->d_name);
4300 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4306 if (new_orphan_inode) {
4308 * We need to open_unlock the inode no matter whether we
4309 * succeed or not, so that other nodes can delete it later.
4311 ocfs2_open_unlock(new_orphan_inode);
4313 iput(new_orphan_inode);
4320 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4321 * sys_reflink(). This will go away when vfs_reflink() exists in
4325 /* copied from may_create in VFS. */
4326 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4328 if (d_really_is_positive(child))
4330 if (IS_DEADDIR(dir))
4332 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4336 * ocfs2_vfs_reflink - Create a reference-counted link
4338 * @old_dentry: source dentry + inode
4339 * @dir: directory to create the target
4340 * @new_dentry: target dentry
4341 * @preserve: if true, preserve all file attributes
4343 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4344 struct dentry *new_dentry, bool preserve)
4346 struct inode *inode = d_inode(old_dentry);
4352 error = ocfs2_may_create(dir, new_dentry);
4356 if (dir->i_sb != inode->i_sb)
4360 * A reflink to an append-only or immutable file cannot be created.
4362 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4365 /* Only regular files can be reflinked. */
4366 if (!S_ISREG(inode->i_mode))
4370 * If the caller wants to preserve ownership, they require the
4374 if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4376 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4381 * If the caller is modifying any aspect of the attributes, they
4382 * are not creating a snapshot. They need read permission on the
4386 error = inode_permission(inode, MAY_READ);
4392 error = dquot_initialize(dir);
4394 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4395 inode_unlock(inode);
4397 fsnotify_create(dir, new_dentry);
4401 * Most codes are copied from sys_linkat.
4403 int ocfs2_reflink_ioctl(struct inode *inode,
4404 const char __user *oldname,
4405 const char __user *newname,
4408 struct dentry *new_dentry;
4409 struct path old_path, new_path;
4412 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4415 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4421 new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4422 error = PTR_ERR(new_dentry);
4423 if (IS_ERR(new_dentry)) {
4429 if (old_path.mnt != new_path.mnt) {
4434 error = ocfs2_vfs_reflink(old_path.dentry,
4435 d_inode(new_path.dentry),
4436 new_dentry, preserve);
4438 done_path_create(&new_path, new_dentry);
4440 path_put(&old_path);