1 // SPDX-License-Identifier: GPL-2.0-only
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * Copyright (C) 2009 Oracle. All rights reserved.
10 #include <linux/sort.h>
11 #include <cluster/masklog.h>
19 #include "buffer_head_io.h"
20 #include "blockcheck.h"
21 #include "refcounttree.h"
24 #include "extent_map.h"
28 #include "ocfs2_trace.h"
31 #include <linux/bio.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/swap.h>
37 #include <linux/security.h>
38 #include <linux/fsnotify.h>
39 #include <linux/quotaops.h>
40 #include <linux/namei.h>
41 #include <linux/mount.h>
42 #include <linux/posix_acl.h>
44 struct ocfs2_cow_context {
48 struct ocfs2_extent_tree data_et;
49 struct ocfs2_refcount_tree *ref_tree;
50 struct buffer_head *ref_root_bh;
51 struct ocfs2_alloc_context *meta_ac;
52 struct ocfs2_alloc_context *data_ac;
53 struct ocfs2_cached_dealloc_ctxt dealloc;
55 struct ocfs2_post_refcount *post_refcount;
57 int (*get_clusters)(struct ocfs2_cow_context *context,
58 u32 v_cluster, u32 *p_cluster,
60 unsigned int *extent_flags);
61 int (*cow_duplicate_clusters)(handle_t *handle,
63 u32 cpos, u32 old_cluster,
64 u32 new_cluster, u32 new_len);
67 static inline struct ocfs2_refcount_tree *
68 cache_info_to_refcount(struct ocfs2_caching_info *ci)
70 return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
73 static int ocfs2_validate_refcount_block(struct super_block *sb,
74 struct buffer_head *bh)
77 struct ocfs2_refcount_block *rb =
78 (struct ocfs2_refcount_block *)bh->b_data;
80 trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
82 BUG_ON(!buffer_uptodate(bh));
85 * If the ecc fails, we return the error but otherwise
86 * leave the filesystem running. We know any error is
87 * local to this block.
89 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
91 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
92 (unsigned long long)bh->b_blocknr);
97 if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
99 "Refcount block #%llu has bad signature %.*s\n",
100 (unsigned long long)bh->b_blocknr, 7,
105 if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
107 "Refcount block #%llu has an invalid rf_blkno of %llu\n",
108 (unsigned long long)bh->b_blocknr,
109 (unsigned long long)le64_to_cpu(rb->rf_blkno));
113 if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
115 "Refcount block #%llu has an invalid rf_fs_generation of #%u\n",
116 (unsigned long long)bh->b_blocknr,
117 le32_to_cpu(rb->rf_fs_generation));
124 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
126 struct buffer_head **bh)
129 struct buffer_head *tmp = *bh;
131 rc = ocfs2_read_block(ci, rb_blkno, &tmp,
132 ocfs2_validate_refcount_block);
134 /* If ocfs2_read_block() got us a new bh, pass it up. */
141 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
143 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
148 static struct super_block *
149 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
151 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
156 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
157 __acquires(&rf->rf_lock)
159 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
161 spin_lock(&rf->rf_lock);
164 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
165 __releases(&rf->rf_lock)
167 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
169 spin_unlock(&rf->rf_lock);
172 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
174 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
176 mutex_lock(&rf->rf_io_mutex);
179 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
181 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
183 mutex_unlock(&rf->rf_io_mutex);
186 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
187 .co_owner = ocfs2_refcount_cache_owner,
188 .co_get_super = ocfs2_refcount_cache_get_super,
189 .co_cache_lock = ocfs2_refcount_cache_lock,
190 .co_cache_unlock = ocfs2_refcount_cache_unlock,
191 .co_io_lock = ocfs2_refcount_cache_io_lock,
192 .co_io_unlock = ocfs2_refcount_cache_io_unlock,
195 static struct ocfs2_refcount_tree *
196 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
198 struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
199 struct ocfs2_refcount_tree *tree = NULL;
202 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
204 if (blkno < tree->rf_blkno)
206 else if (blkno > tree->rf_blkno)
215 /* osb_lock is already locked. */
216 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
217 struct ocfs2_refcount_tree *new)
219 u64 rf_blkno = new->rf_blkno;
220 struct rb_node *parent = NULL;
221 struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
222 struct ocfs2_refcount_tree *tmp;
227 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
230 if (rf_blkno < tmp->rf_blkno)
232 else if (rf_blkno > tmp->rf_blkno)
235 /* This should never happen! */
236 mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
237 (unsigned long long)rf_blkno);
242 rb_link_node(&new->rf_node, parent, p);
243 rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
246 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
248 ocfs2_metadata_cache_exit(&tree->rf_ci);
249 ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
250 ocfs2_lock_res_free(&tree->rf_lockres);
255 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
256 struct ocfs2_refcount_tree *tree)
258 rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
259 if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
260 osb->osb_ref_tree_lru = NULL;
263 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
264 struct ocfs2_refcount_tree *tree)
266 spin_lock(&osb->osb_lock);
267 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
268 spin_unlock(&osb->osb_lock);
271 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
273 struct ocfs2_refcount_tree *tree =
274 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
276 ocfs2_free_refcount_tree(tree);
280 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
282 kref_get(&tree->rf_getcnt);
286 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
288 kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
291 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
292 struct super_block *sb)
294 ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
295 mutex_init(&new->rf_io_mutex);
297 spin_lock_init(&new->rf_lock);
300 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
301 struct ocfs2_refcount_tree *new,
302 u64 rf_blkno, u32 generation)
304 init_rwsem(&new->rf_sem);
305 ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
306 rf_blkno, generation);
309 static struct ocfs2_refcount_tree*
310 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
312 struct ocfs2_refcount_tree *new;
314 new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
318 new->rf_blkno = rf_blkno;
319 kref_init(&new->rf_getcnt);
320 ocfs2_init_refcount_tree_ci(new, osb->sb);
325 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
326 struct ocfs2_refcount_tree **ret_tree)
329 struct ocfs2_refcount_tree *tree, *new = NULL;
330 struct buffer_head *ref_root_bh = NULL;
331 struct ocfs2_refcount_block *ref_rb;
333 spin_lock(&osb->osb_lock);
334 if (osb->osb_ref_tree_lru &&
335 osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
336 tree = osb->osb_ref_tree_lru;
338 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
342 spin_unlock(&osb->osb_lock);
344 new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
351 * We need the generation to create the refcount tree lock and since
352 * it isn't changed during the tree modification, we are safe here to
353 * read without protection.
354 * We also have to purge the cache after we create the lock since the
355 * refcount block may have the stale data. It can only be trusted when
356 * we hold the refcount lock.
358 ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
361 ocfs2_metadata_cache_exit(&new->rf_ci);
366 ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
367 new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
368 ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
370 ocfs2_metadata_cache_purge(&new->rf_ci);
372 spin_lock(&osb->osb_lock);
373 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
377 ocfs2_insert_refcount_tree(osb, new);
385 osb->osb_ref_tree_lru = tree;
387 spin_unlock(&osb->osb_lock);
390 ocfs2_free_refcount_tree(new);
396 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
399 struct buffer_head *di_bh = NULL;
400 struct ocfs2_dinode *di;
402 ret = ocfs2_read_inode_block(inode, &di_bh);
408 BUG_ON(!ocfs2_is_refcount_inode(inode));
410 di = (struct ocfs2_dinode *)di_bh->b_data;
411 *ref_blkno = le64_to_cpu(di->i_refcount_loc);
417 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
418 struct ocfs2_refcount_tree *tree, int rw)
422 ret = ocfs2_refcount_lock(tree, rw);
429 down_write(&tree->rf_sem);
431 down_read(&tree->rf_sem);
438 * Lock the refcount tree pointed by ref_blkno and return the tree.
439 * In most case, we lock the tree and read the refcount block.
440 * So read it here if the caller really needs it.
442 * If the tree has been re-created by other node, it will free the
443 * old one and re-create it.
445 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
446 u64 ref_blkno, int rw,
447 struct ocfs2_refcount_tree **ret_tree,
448 struct buffer_head **ref_bh)
450 int ret, delete_tree = 0;
451 struct ocfs2_refcount_tree *tree = NULL;
452 struct buffer_head *ref_root_bh = NULL;
453 struct ocfs2_refcount_block *rb;
456 ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
462 ocfs2_refcount_tree_get(tree);
464 ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
467 ocfs2_refcount_tree_put(tree);
471 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
475 ocfs2_unlock_refcount_tree(osb, tree, rw);
479 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
481 * If the refcount block has been freed and re-created, we may need
482 * to recreate the refcount tree also.
484 * Here we just remove the tree from the rb-tree, and the last
485 * kref holder will unlock and delete this refcount_tree.
486 * Then we goto "again" and ocfs2_get_refcount_tree will create
487 * the new refcount tree for us.
489 if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
490 if (!tree->rf_removed) {
491 ocfs2_erase_refcount_tree_from_list(osb, tree);
492 tree->rf_removed = 1;
496 ocfs2_unlock_refcount_tree(osb, tree, rw);
498 * We get an extra reference when we create the refcount
499 * tree, so another put will destroy it.
502 ocfs2_refcount_tree_put(tree);
510 *ref_bh = ref_root_bh;
518 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
519 struct ocfs2_refcount_tree *tree, int rw)
522 up_write(&tree->rf_sem);
524 up_read(&tree->rf_sem);
526 ocfs2_refcount_unlock(tree, rw);
527 ocfs2_refcount_tree_put(tree);
530 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
532 struct rb_node *node;
533 struct ocfs2_refcount_tree *tree;
534 struct rb_root *root = &osb->osb_rf_lock_tree;
536 while ((node = rb_last(root)) != NULL) {
537 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
539 trace_ocfs2_purge_refcount_trees(
540 (unsigned long long) tree->rf_blkno);
542 rb_erase(&tree->rf_node, root);
543 ocfs2_free_refcount_tree(tree);
548 * Create a refcount tree for an inode.
549 * We take for granted that the inode is already locked.
551 static int ocfs2_create_refcount_tree(struct inode *inode,
552 struct buffer_head *di_bh)
555 handle_t *handle = NULL;
556 struct ocfs2_alloc_context *meta_ac = NULL;
557 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
558 struct ocfs2_inode_info *oi = OCFS2_I(inode);
559 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
560 struct buffer_head *new_bh = NULL;
561 struct ocfs2_refcount_block *rb;
562 struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
563 u16 suballoc_bit_start;
565 u64 suballoc_loc, first_blkno;
567 BUG_ON(ocfs2_is_refcount_inode(inode));
569 trace_ocfs2_create_refcount_tree(
570 (unsigned long long)oi->ip_blkno);
572 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
578 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
579 if (IS_ERR(handle)) {
580 ret = PTR_ERR(handle);
585 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
586 OCFS2_JOURNAL_ACCESS_WRITE);
592 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
593 &suballoc_bit_start, &num_got,
600 new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
607 new_bh = sb_getblk(inode->i_sb, first_blkno);
613 ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
615 ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
616 OCFS2_JOURNAL_ACCESS_CREATE);
622 /* Initialize ocfs2_refcount_block. */
623 rb = (struct ocfs2_refcount_block *)new_bh->b_data;
624 memset(rb, 0, inode->i_sb->s_blocksize);
625 strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
626 rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
627 rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
628 rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
629 rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
630 rb->rf_blkno = cpu_to_le64(first_blkno);
631 rb->rf_count = cpu_to_le32(1);
632 rb->rf_records.rl_count =
633 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
634 spin_lock(&osb->osb_lock);
635 rb->rf_generation = osb->s_next_generation++;
636 spin_unlock(&osb->osb_lock);
638 ocfs2_journal_dirty(handle, new_bh);
640 spin_lock(&oi->ip_lock);
641 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
642 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
643 di->i_refcount_loc = cpu_to_le64(first_blkno);
644 spin_unlock(&oi->ip_lock);
646 trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
648 ocfs2_journal_dirty(handle, di_bh);
651 * We have to init the tree lock here since it will use
652 * the generation number to create it.
654 new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
655 ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
656 new_tree->rf_generation);
658 spin_lock(&osb->osb_lock);
659 tree = ocfs2_find_refcount_tree(osb, first_blkno);
662 * We've just created a new refcount tree in this block. If
663 * we found a refcount tree on the ocfs2_super, it must be
664 * one we just deleted. We free the old tree before
665 * inserting the new tree.
667 BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
669 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
670 ocfs2_insert_refcount_tree(osb, new_tree);
671 spin_unlock(&osb->osb_lock);
674 ocfs2_refcount_tree_put(tree);
677 ocfs2_commit_trans(osb, handle);
681 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
687 ocfs2_free_alloc_context(meta_ac);
692 static int ocfs2_set_refcount_tree(struct inode *inode,
693 struct buffer_head *di_bh,
697 handle_t *handle = NULL;
698 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
699 struct ocfs2_inode_info *oi = OCFS2_I(inode);
700 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
701 struct buffer_head *ref_root_bh = NULL;
702 struct ocfs2_refcount_block *rb;
703 struct ocfs2_refcount_tree *ref_tree;
705 BUG_ON(ocfs2_is_refcount_inode(inode));
707 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
708 &ref_tree, &ref_root_bh);
714 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
715 if (IS_ERR(handle)) {
716 ret = PTR_ERR(handle);
721 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
722 OCFS2_JOURNAL_ACCESS_WRITE);
728 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
729 OCFS2_JOURNAL_ACCESS_WRITE);
735 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
736 le32_add_cpu(&rb->rf_count, 1);
738 ocfs2_journal_dirty(handle, ref_root_bh);
740 spin_lock(&oi->ip_lock);
741 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
742 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
743 di->i_refcount_loc = cpu_to_le64(refcount_loc);
744 spin_unlock(&oi->ip_lock);
745 ocfs2_journal_dirty(handle, di_bh);
748 ocfs2_commit_trans(osb, handle);
750 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
756 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
758 int ret, delete_tree = 0;
759 handle_t *handle = NULL;
760 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
761 struct ocfs2_inode_info *oi = OCFS2_I(inode);
762 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
763 struct ocfs2_refcount_block *rb;
764 struct inode *alloc_inode = NULL;
765 struct buffer_head *alloc_bh = NULL;
766 struct buffer_head *blk_bh = NULL;
767 struct ocfs2_refcount_tree *ref_tree;
768 int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
769 u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
772 if (!ocfs2_is_refcount_inode(inode))
776 ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
782 rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
785 * If we are the last user, we need to free the block.
786 * So lock the allocator ahead.
788 if (le32_to_cpu(rb->rf_count) == 1) {
789 blk = le64_to_cpu(rb->rf_blkno);
790 bit = le16_to_cpu(rb->rf_suballoc_bit);
791 if (rb->rf_suballoc_loc)
792 bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
794 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
796 alloc_inode = ocfs2_get_system_file_inode(osb,
797 EXTENT_ALLOC_SYSTEM_INODE,
798 le16_to_cpu(rb->rf_suballoc_slot));
804 inode_lock(alloc_inode);
806 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
812 credits += OCFS2_SUBALLOC_FREE;
815 handle = ocfs2_start_trans(osb, credits);
816 if (IS_ERR(handle)) {
817 ret = PTR_ERR(handle);
822 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
823 OCFS2_JOURNAL_ACCESS_WRITE);
829 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
830 OCFS2_JOURNAL_ACCESS_WRITE);
836 spin_lock(&oi->ip_lock);
837 oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
838 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
839 di->i_refcount_loc = 0;
840 spin_unlock(&oi->ip_lock);
841 ocfs2_journal_dirty(handle, di_bh);
843 le32_add_cpu(&rb->rf_count , -1);
844 ocfs2_journal_dirty(handle, blk_bh);
848 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
849 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
850 alloc_bh, bit, bg_blkno, 1);
856 ocfs2_commit_trans(osb, handle);
859 ocfs2_inode_unlock(alloc_inode, 1);
864 inode_unlock(alloc_inode);
868 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
870 ocfs2_refcount_tree_put(ref_tree);
876 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
877 struct buffer_head *ref_leaf_bh,
878 u64 cpos, unsigned int len,
879 struct ocfs2_refcount_rec *ret_rec,
883 struct ocfs2_refcount_block *rb =
884 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
885 struct ocfs2_refcount_rec *rec = NULL;
887 for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
888 rec = &rb->rf_records.rl_recs[i];
890 if (le64_to_cpu(rec->r_cpos) +
891 le32_to_cpu(rec->r_clusters) <= cpos)
893 else if (le64_to_cpu(rec->r_cpos) > cpos)
896 /* ok, cpos fail in this rec. Just return. */
903 /* We meet with a hole here, so fake the rec. */
904 ret_rec->r_cpos = cpu_to_le64(cpos);
905 ret_rec->r_refcount = 0;
906 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
907 le64_to_cpu(rec->r_cpos) < cpos + len)
908 ret_rec->r_clusters =
909 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
911 ret_rec->r_clusters = cpu_to_le32(len);
919 * Try to remove refcount tree. The mechanism is:
920 * 1) Check whether i_clusters == 0, if no, exit.
921 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
922 * 3) Check whether we have inline xattr stored outside, if yes, exit.
923 * 4) Remove the tree.
925 int ocfs2_try_remove_refcount_tree(struct inode *inode,
926 struct buffer_head *di_bh)
929 struct ocfs2_inode_info *oi = OCFS2_I(inode);
930 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
932 down_write(&oi->ip_xattr_sem);
933 down_write(&oi->ip_alloc_sem);
938 if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
941 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
942 ocfs2_has_inline_xattr_value_outside(inode, di))
945 ret = ocfs2_remove_refcount_tree(inode, di_bh);
949 up_write(&oi->ip_alloc_sem);
950 up_write(&oi->ip_xattr_sem);
955 * Find the end range for a leaf refcount block indicated by
956 * el->l_recs[index].e_blkno.
958 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
959 struct buffer_head *ref_root_bh,
960 struct ocfs2_extent_block *eb,
961 struct ocfs2_extent_list *el,
962 int index, u32 *cpos_end)
964 int ret, i, subtree_root;
967 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
968 struct ocfs2_path *left_path = NULL, *right_path = NULL;
969 struct ocfs2_extent_tree et;
970 struct ocfs2_extent_list *tmp_el;
972 if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
974 * We have a extent rec after index, so just use the e_cpos
975 * of the next extent rec.
977 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
981 if (!eb || !eb->h_next_leaf_blk) {
983 * We are the last extent rec, so any high cpos should
984 * be stored in this leaf refcount block.
986 *cpos_end = UINT_MAX;
991 * If the extent block isn't the last one, we have to find
992 * the subtree root between this extent block and the next
993 * leaf extent block and get the corresponding e_cpos from
994 * the subroot. Otherwise we may corrupt the b-tree.
996 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
998 left_path = ocfs2_new_path_from_et(&et);
1005 cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1006 ret = ocfs2_find_path(ci, left_path, cpos);
1012 right_path = ocfs2_new_path_from_path(left_path);
1019 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1025 ret = ocfs2_find_path(ci, right_path, cpos);
1031 subtree_root = ocfs2_find_subtree_root(&et, left_path,
1034 tmp_el = left_path->p_node[subtree_root].el;
1035 blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1036 for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1037 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1038 *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1043 BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1046 ocfs2_free_path(left_path);
1047 ocfs2_free_path(right_path);
1052 * Given a cpos and len, try to find the refcount record which contains cpos.
1053 * 1. If cpos can be found in one refcount record, return the record.
1054 * 2. If cpos can't be found, return a fake record which start from cpos
1055 * and end at a small value between cpos+len and start of the next record.
1056 * This fake record has r_refcount = 0.
1058 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1059 struct buffer_head *ref_root_bh,
1060 u64 cpos, unsigned int len,
1061 struct ocfs2_refcount_rec *ret_rec,
1063 struct buffer_head **ret_bh)
1065 int ret = 0, i, found;
1066 u32 low_cpos, cpos_end;
1067 struct ocfs2_extent_list *el;
1068 struct ocfs2_extent_rec *rec = NULL;
1069 struct ocfs2_extent_block *eb = NULL;
1070 struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1071 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1072 struct ocfs2_refcount_block *rb =
1073 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1075 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1076 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1078 *ret_bh = ref_root_bh;
1079 get_bh(ref_root_bh);
1084 low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1086 if (el->l_tree_depth) {
1087 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1093 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1096 if (el->l_tree_depth) {
1097 ret = ocfs2_error(sb,
1098 "refcount tree %llu has non zero tree depth in leaf btree tree block %llu\n",
1099 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1100 (unsigned long long)eb_bh->b_blocknr);
1106 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1107 rec = &el->l_recs[i];
1109 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1116 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1117 eb, el, i, &cpos_end);
1123 if (cpos_end < low_cpos + len)
1124 len = cpos_end - low_cpos;
1127 ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1134 ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1136 *ret_bh = ref_leaf_bh;
1142 enum ocfs2_ref_rec_contig {
1143 REF_CONTIG_NONE = 0,
1146 REF_CONTIG_LEFTRIGHT,
1149 static enum ocfs2_ref_rec_contig
1150 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1153 if ((rb->rf_records.rl_recs[index].r_refcount ==
1154 rb->rf_records.rl_recs[index + 1].r_refcount) &&
1155 (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1156 le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1157 le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1158 return REF_CONTIG_RIGHT;
1160 return REF_CONTIG_NONE;
1163 static enum ocfs2_ref_rec_contig
1164 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1167 enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1169 if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1170 ret = ocfs2_refcount_rec_adjacent(rb, index);
1173 enum ocfs2_ref_rec_contig tmp;
1175 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1177 if (tmp == REF_CONTIG_RIGHT) {
1178 if (ret == REF_CONTIG_RIGHT)
1179 ret = REF_CONTIG_LEFTRIGHT;
1181 ret = REF_CONTIG_LEFT;
1188 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1191 BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1192 rb->rf_records.rl_recs[index+1].r_refcount);
1194 le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1195 le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1197 if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1198 memmove(&rb->rf_records.rl_recs[index + 1],
1199 &rb->rf_records.rl_recs[index + 2],
1200 sizeof(struct ocfs2_refcount_rec) *
1201 (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1203 memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1204 0, sizeof(struct ocfs2_refcount_rec));
1205 le16_add_cpu(&rb->rf_records.rl_used, -1);
1209 * Merge the refcount rec if we are contiguous with the adjacent recs.
1211 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1214 enum ocfs2_ref_rec_contig contig =
1215 ocfs2_refcount_rec_contig(rb, index);
1217 if (contig == REF_CONTIG_NONE)
1220 if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1225 ocfs2_rotate_refcount_rec_left(rb, index);
1227 if (contig == REF_CONTIG_LEFTRIGHT)
1228 ocfs2_rotate_refcount_rec_left(rb, index);
1232 * Change the refcount indexed by "index" in ref_bh.
1233 * If refcount reaches 0, remove it.
1235 static int ocfs2_change_refcount_rec(handle_t *handle,
1236 struct ocfs2_caching_info *ci,
1237 struct buffer_head *ref_leaf_bh,
1238 int index, int merge, int change)
1241 struct ocfs2_refcount_block *rb =
1242 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1243 struct ocfs2_refcount_list *rl = &rb->rf_records;
1244 struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1246 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1247 OCFS2_JOURNAL_ACCESS_WRITE);
1253 trace_ocfs2_change_refcount_rec(
1254 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1255 index, le32_to_cpu(rec->r_refcount), change);
1256 le32_add_cpu(&rec->r_refcount, change);
1258 if (!rec->r_refcount) {
1259 if (index != le16_to_cpu(rl->rl_used) - 1) {
1260 memmove(rec, rec + 1,
1261 (le16_to_cpu(rl->rl_used) - index - 1) *
1262 sizeof(struct ocfs2_refcount_rec));
1263 memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1264 0, sizeof(struct ocfs2_refcount_rec));
1267 le16_add_cpu(&rl->rl_used, -1);
1269 ocfs2_refcount_rec_merge(rb, index);
1271 ocfs2_journal_dirty(handle, ref_leaf_bh);
1276 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1277 struct ocfs2_caching_info *ci,
1278 struct buffer_head *ref_root_bh,
1279 struct buffer_head **ref_leaf_bh,
1280 struct ocfs2_alloc_context *meta_ac)
1283 u16 suballoc_bit_start;
1285 u64 suballoc_loc, blkno;
1286 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1287 struct buffer_head *new_bh = NULL;
1288 struct ocfs2_refcount_block *new_rb;
1289 struct ocfs2_refcount_block *root_rb =
1290 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1292 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1293 OCFS2_JOURNAL_ACCESS_WRITE);
1299 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1300 &suballoc_bit_start, &num_got,
1307 new_bh = sb_getblk(sb, blkno);
1308 if (new_bh == NULL) {
1313 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1315 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1316 OCFS2_JOURNAL_ACCESS_CREATE);
1323 * Initialize ocfs2_refcount_block.
1324 * It should contain the same information as the old root.
1325 * so just memcpy it and change the corresponding field.
1327 memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1329 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1330 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1331 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1332 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1333 new_rb->rf_blkno = cpu_to_le64(blkno);
1334 new_rb->rf_cpos = cpu_to_le32(0);
1335 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1336 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1337 ocfs2_journal_dirty(handle, new_bh);
1339 /* Now change the root. */
1340 memset(&root_rb->rf_list, 0, sb->s_blocksize -
1341 offsetof(struct ocfs2_refcount_block, rf_list));
1342 root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1343 root_rb->rf_clusters = cpu_to_le32(1);
1344 root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1345 root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1346 root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1347 root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1349 ocfs2_journal_dirty(handle, ref_root_bh);
1351 trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1352 le16_to_cpu(new_rb->rf_records.rl_used));
1354 *ref_leaf_bh = new_bh;
1361 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1362 struct ocfs2_refcount_rec *next)
1364 if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1365 ocfs2_get_ref_rec_low_cpos(next))
1371 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1373 const struct ocfs2_refcount_rec *l = a, *r = b;
1374 u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1375 u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1377 if (l_cpos > r_cpos)
1379 if (l_cpos < r_cpos)
1384 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1386 const struct ocfs2_refcount_rec *l = a, *r = b;
1387 u64 l_cpos = le64_to_cpu(l->r_cpos);
1388 u64 r_cpos = le64_to_cpu(r->r_cpos);
1390 if (l_cpos > r_cpos)
1392 if (l_cpos < r_cpos)
1397 static void swap_refcount_rec(void *a, void *b, int size)
1399 struct ocfs2_refcount_rec *l = a, *r = b;
1405 * The refcount cpos are ordered by their 64bit cpos,
1406 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1407 * So we need to make sure that this pos isn't intersected with others.
1409 * Note: The refcount block is already sorted by their low 32 bit cpos,
1410 * So just try the middle pos first, and we will exit when we find
1411 * the good position.
1413 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1414 u32 *split_pos, int *split_index)
1416 int num_used = le16_to_cpu(rl->rl_used);
1417 int delta, middle = num_used / 2;
1419 for (delta = 0; delta < middle; delta++) {
1420 /* Let's check delta earlier than middle */
1421 if (ocfs2_refcount_rec_no_intersect(
1422 &rl->rl_recs[middle - delta - 1],
1423 &rl->rl_recs[middle - delta])) {
1424 *split_index = middle - delta;
1428 /* For even counts, don't walk off the end */
1429 if ((middle + delta + 1) == num_used)
1432 /* Now try delta past middle */
1433 if (ocfs2_refcount_rec_no_intersect(
1434 &rl->rl_recs[middle + delta],
1435 &rl->rl_recs[middle + delta + 1])) {
1436 *split_index = middle + delta + 1;
1441 if (delta >= middle)
1444 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1448 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1449 struct buffer_head *new_bh,
1452 int split_index = 0, num_moved, ret;
1454 struct ocfs2_refcount_block *rb =
1455 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1456 struct ocfs2_refcount_list *rl = &rb->rf_records;
1457 struct ocfs2_refcount_block *new_rb =
1458 (struct ocfs2_refcount_block *)new_bh->b_data;
1459 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1461 trace_ocfs2_divide_leaf_refcount_block(
1462 (unsigned long long)ref_leaf_bh->b_blocknr,
1463 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1466 * XXX: Improvement later.
1467 * If we know all the high 32 bit cpos is the same, no need to sort.
1469 * In order to make the whole process safe, we do:
1470 * 1. sort the entries by their low 32 bit cpos first so that we can
1471 * find the split cpos easily.
1472 * 2. call ocfs2_insert_extent to insert the new refcount block.
1473 * 3. move the refcount rec to the new block.
1474 * 4. sort the entries by their 64 bit cpos.
1475 * 5. dirty the new_rb and rb.
1477 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1478 sizeof(struct ocfs2_refcount_rec),
1479 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1481 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1487 new_rb->rf_cpos = cpu_to_le32(cpos);
1489 /* move refcount records starting from split_index to the new block. */
1490 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1491 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1492 num_moved * sizeof(struct ocfs2_refcount_rec));
1494 /*ok, remove the entries we just moved over to the other block. */
1495 memset(&rl->rl_recs[split_index], 0,
1496 num_moved * sizeof(struct ocfs2_refcount_rec));
1498 /* change old and new rl_used accordingly. */
1499 le16_add_cpu(&rl->rl_used, -num_moved);
1500 new_rl->rl_used = cpu_to_le16(num_moved);
1502 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1503 sizeof(struct ocfs2_refcount_rec),
1504 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1506 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1507 sizeof(struct ocfs2_refcount_rec),
1508 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1514 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1515 struct ocfs2_caching_info *ci,
1516 struct buffer_head *ref_root_bh,
1517 struct buffer_head *ref_leaf_bh,
1518 struct ocfs2_alloc_context *meta_ac)
1521 u16 suballoc_bit_start;
1522 u32 num_got, new_cpos;
1523 u64 suballoc_loc, blkno;
1524 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1525 struct ocfs2_refcount_block *root_rb =
1526 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1527 struct buffer_head *new_bh = NULL;
1528 struct ocfs2_refcount_block *new_rb;
1529 struct ocfs2_extent_tree ref_et;
1531 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1533 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1534 OCFS2_JOURNAL_ACCESS_WRITE);
1540 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1541 OCFS2_JOURNAL_ACCESS_WRITE);
1547 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1548 &suballoc_bit_start, &num_got,
1555 new_bh = sb_getblk(sb, blkno);
1556 if (new_bh == NULL) {
1561 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1563 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1564 OCFS2_JOURNAL_ACCESS_CREATE);
1570 /* Initialize ocfs2_refcount_block. */
1571 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1572 memset(new_rb, 0, sb->s_blocksize);
1573 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1574 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1575 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1576 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1577 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1578 new_rb->rf_blkno = cpu_to_le64(blkno);
1579 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1580 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1581 new_rb->rf_records.rl_count =
1582 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1583 new_rb->rf_generation = root_rb->rf_generation;
1585 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1591 ocfs2_journal_dirty(handle, ref_leaf_bh);
1592 ocfs2_journal_dirty(handle, new_bh);
1594 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1596 trace_ocfs2_new_leaf_refcount_block(
1597 (unsigned long long)new_bh->b_blocknr, new_cpos);
1599 /* Insert the new leaf block with the specific offset cpos. */
1600 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1610 static int ocfs2_expand_refcount_tree(handle_t *handle,
1611 struct ocfs2_caching_info *ci,
1612 struct buffer_head *ref_root_bh,
1613 struct buffer_head *ref_leaf_bh,
1614 struct ocfs2_alloc_context *meta_ac)
1617 struct buffer_head *expand_bh = NULL;
1619 if (ref_root_bh == ref_leaf_bh) {
1621 * the old root bh hasn't been expanded to a b-tree,
1622 * so expand it first.
1624 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1625 &expand_bh, meta_ac);
1631 expand_bh = ref_leaf_bh;
1636 /* Now add a new refcount block into the tree.*/
1637 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1638 expand_bh, meta_ac);
1647 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1649 * Only called when we have inserted a new refcount rec at index 0
1650 * which means ocfs2_extent_rec.e_cpos may need some change.
1652 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1653 struct ocfs2_caching_info *ci,
1654 struct buffer_head *ref_root_bh,
1655 struct buffer_head *ref_leaf_bh,
1656 struct ocfs2_refcount_rec *rec)
1659 u32 new_cpos, old_cpos;
1660 struct ocfs2_path *path = NULL;
1661 struct ocfs2_extent_tree et;
1662 struct ocfs2_refcount_block *rb =
1663 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1664 struct ocfs2_extent_list *el;
1666 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1669 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1670 old_cpos = le32_to_cpu(rb->rf_cpos);
1671 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1672 if (old_cpos <= new_cpos)
1675 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1677 path = ocfs2_new_path_from_et(&et);
1684 ret = ocfs2_find_path(ci, path, old_cpos);
1691 * 2 more credits, one for the leaf refcount block, one for
1692 * the extent block contains the extent rec.
1694 ret = ocfs2_extend_trans(handle, 2);
1700 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1701 OCFS2_JOURNAL_ACCESS_WRITE);
1707 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1708 OCFS2_JOURNAL_ACCESS_WRITE);
1714 /* change the leaf extent block first. */
1715 el = path_leaf_el(path);
1717 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1718 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1721 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1723 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1725 /* change the r_cpos in the leaf block. */
1726 rb->rf_cpos = cpu_to_le32(new_cpos);
1728 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1729 ocfs2_journal_dirty(handle, ref_leaf_bh);
1732 ocfs2_free_path(path);
1736 static int ocfs2_insert_refcount_rec(handle_t *handle,
1737 struct ocfs2_caching_info *ci,
1738 struct buffer_head *ref_root_bh,
1739 struct buffer_head *ref_leaf_bh,
1740 struct ocfs2_refcount_rec *rec,
1741 int index, int merge,
1742 struct ocfs2_alloc_context *meta_ac)
1745 struct ocfs2_refcount_block *rb =
1746 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1747 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1748 struct buffer_head *new_bh = NULL;
1750 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1752 if (rf_list->rl_used == rf_list->rl_count) {
1753 u64 cpos = le64_to_cpu(rec->r_cpos);
1754 u32 len = le32_to_cpu(rec->r_clusters);
1756 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1757 ref_leaf_bh, meta_ac);
1763 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1764 cpos, len, NULL, &index,
1771 ref_leaf_bh = new_bh;
1772 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1773 rf_list = &rb->rf_records;
1776 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1777 OCFS2_JOURNAL_ACCESS_WRITE);
1783 if (index < le16_to_cpu(rf_list->rl_used))
1784 memmove(&rf_list->rl_recs[index + 1],
1785 &rf_list->rl_recs[index],
1786 (le16_to_cpu(rf_list->rl_used) - index) *
1787 sizeof(struct ocfs2_refcount_rec));
1789 trace_ocfs2_insert_refcount_rec(
1790 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1791 (unsigned long long)le64_to_cpu(rec->r_cpos),
1792 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1794 rf_list->rl_recs[index] = *rec;
1796 le16_add_cpu(&rf_list->rl_used, 1);
1799 ocfs2_refcount_rec_merge(rb, index);
1801 ocfs2_journal_dirty(handle, ref_leaf_bh);
1804 ret = ocfs2_adjust_refcount_rec(handle, ci,
1816 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1817 * This is much simple than our b-tree code.
1818 * split_rec is the new refcount rec we want to insert.
1819 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1820 * increase refcount or decrease a refcount to non-zero).
1821 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1822 * rec( in case we decrease a refcount to zero).
1824 static int ocfs2_split_refcount_rec(handle_t *handle,
1825 struct ocfs2_caching_info *ci,
1826 struct buffer_head *ref_root_bh,
1827 struct buffer_head *ref_leaf_bh,
1828 struct ocfs2_refcount_rec *split_rec,
1829 int index, int merge,
1830 struct ocfs2_alloc_context *meta_ac,
1831 struct ocfs2_cached_dealloc_ctxt *dealloc)
1835 struct ocfs2_refcount_block *rb =
1836 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1837 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1838 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1839 struct ocfs2_refcount_rec *tail_rec = NULL;
1840 struct buffer_head *new_bh = NULL;
1842 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1844 trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1845 le32_to_cpu(orig_rec->r_clusters),
1846 le32_to_cpu(orig_rec->r_refcount),
1847 le64_to_cpu(split_rec->r_cpos),
1848 le32_to_cpu(split_rec->r_clusters),
1849 le32_to_cpu(split_rec->r_refcount));
1852 * If we just need to split the header or tail clusters,
1853 * no more recs are needed, just split is OK.
1854 * Otherwise we at least need one new recs.
1856 if (!split_rec->r_refcount &&
1857 (split_rec->r_cpos == orig_rec->r_cpos ||
1858 le64_to_cpu(split_rec->r_cpos) +
1859 le32_to_cpu(split_rec->r_clusters) ==
1860 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1866 * We need one more rec if we split in the middle and the new rec have
1867 * some refcount in it.
1869 if (split_rec->r_refcount &&
1870 (split_rec->r_cpos != orig_rec->r_cpos &&
1871 le64_to_cpu(split_rec->r_cpos) +
1872 le32_to_cpu(split_rec->r_clusters) !=
1873 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1876 /* If the leaf block don't have enough record, expand it. */
1877 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1878 le16_to_cpu(rf_list->rl_count)) {
1879 struct ocfs2_refcount_rec tmp_rec;
1880 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1881 len = le32_to_cpu(orig_rec->r_clusters);
1882 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1883 ref_leaf_bh, meta_ac);
1890 * We have to re-get it since now cpos may be moved to
1891 * another leaf block.
1893 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1894 cpos, len, &tmp_rec, &index,
1901 ref_leaf_bh = new_bh;
1902 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1903 rf_list = &rb->rf_records;
1904 orig_rec = &rf_list->rl_recs[index];
1907 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1908 OCFS2_JOURNAL_ACCESS_WRITE);
1915 * We have calculated out how many new records we need and store
1916 * in recs_need, so spare enough space first by moving the records
1917 * after "index" to the end.
1919 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1920 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1921 &rf_list->rl_recs[index + 1],
1922 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1923 sizeof(struct ocfs2_refcount_rec));
1925 len = (le64_to_cpu(orig_rec->r_cpos) +
1926 le32_to_cpu(orig_rec->r_clusters)) -
1927 (le64_to_cpu(split_rec->r_cpos) +
1928 le32_to_cpu(split_rec->r_clusters));
1931 * If we have "len", the we will split in the tail and move it
1932 * to the end of the space we have just spared.
1935 tail_rec = &rf_list->rl_recs[index + recs_need];
1937 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1938 le64_add_cpu(&tail_rec->r_cpos,
1939 le32_to_cpu(tail_rec->r_clusters) - len);
1940 tail_rec->r_clusters = cpu_to_le32(len);
1944 * If the split pos isn't the same as the original one, we need to
1945 * split in the head.
1947 * Note: We have the chance that split_rec.r_refcount = 0,
1948 * recs_need = 0 and len > 0, which means we just cut the head from
1949 * the orig_rec and in that case we have done some modification in
1950 * orig_rec above, so the check for r_cpos is faked.
1952 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1953 len = le64_to_cpu(split_rec->r_cpos) -
1954 le64_to_cpu(orig_rec->r_cpos);
1955 orig_rec->r_clusters = cpu_to_le32(len);
1959 le16_add_cpu(&rf_list->rl_used, recs_need);
1961 if (split_rec->r_refcount) {
1962 rf_list->rl_recs[index] = *split_rec;
1963 trace_ocfs2_split_refcount_rec_insert(
1964 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1965 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1966 le32_to_cpu(split_rec->r_clusters),
1967 le32_to_cpu(split_rec->r_refcount));
1970 ocfs2_refcount_rec_merge(rb, index);
1973 ocfs2_journal_dirty(handle, ref_leaf_bh);
1980 static int __ocfs2_increase_refcount(handle_t *handle,
1981 struct ocfs2_caching_info *ci,
1982 struct buffer_head *ref_root_bh,
1983 u64 cpos, u32 len, int merge,
1984 struct ocfs2_alloc_context *meta_ac,
1985 struct ocfs2_cached_dealloc_ctxt *dealloc)
1988 struct buffer_head *ref_leaf_bh = NULL;
1989 struct ocfs2_refcount_rec rec;
1990 unsigned int set_len = 0;
1992 trace_ocfs2_increase_refcount_begin(
1993 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1994 (unsigned long long)cpos, len);
1997 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1998 cpos, len, &rec, &index,
2005 set_len = le32_to_cpu(rec.r_clusters);
2008 * Here we may meet with 3 situations:
2010 * 1. If we find an already existing record, and the length
2011 * is the same, cool, we just need to increase the r_refcount
2013 * 2. If we find a hole, just insert it with r_refcount = 1.
2014 * 3. If we are in the middle of one extent record, split
2017 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2019 trace_ocfs2_increase_refcount_change(
2020 (unsigned long long)cpos, set_len,
2021 le32_to_cpu(rec.r_refcount));
2022 ret = ocfs2_change_refcount_rec(handle, ci,
2029 } else if (!rec.r_refcount) {
2030 rec.r_refcount = cpu_to_le32(1);
2032 trace_ocfs2_increase_refcount_insert(
2033 (unsigned long long)le64_to_cpu(rec.r_cpos),
2035 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2044 set_len = min((u64)(cpos + len),
2045 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2046 rec.r_cpos = cpu_to_le64(cpos);
2047 rec.r_clusters = cpu_to_le32(set_len);
2048 le32_add_cpu(&rec.r_refcount, 1);
2050 trace_ocfs2_increase_refcount_split(
2051 (unsigned long long)le64_to_cpu(rec.r_cpos),
2052 set_len, le32_to_cpu(rec.r_refcount));
2053 ret = ocfs2_split_refcount_rec(handle, ci,
2054 ref_root_bh, ref_leaf_bh,
2065 brelse(ref_leaf_bh);
2070 brelse(ref_leaf_bh);
2074 static int ocfs2_remove_refcount_extent(handle_t *handle,
2075 struct ocfs2_caching_info *ci,
2076 struct buffer_head *ref_root_bh,
2077 struct buffer_head *ref_leaf_bh,
2078 struct ocfs2_alloc_context *meta_ac,
2079 struct ocfs2_cached_dealloc_ctxt *dealloc)
2082 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2083 struct ocfs2_refcount_block *rb =
2084 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2085 struct ocfs2_extent_tree et;
2087 BUG_ON(rb->rf_records.rl_used);
2089 trace_ocfs2_remove_refcount_extent(
2090 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2091 (unsigned long long)ref_leaf_bh->b_blocknr,
2092 le32_to_cpu(rb->rf_cpos));
2094 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2095 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2096 1, meta_ac, dealloc);
2102 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2105 * add the freed block to the dealloc so that it will be freed
2106 * when we run dealloc.
2108 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2109 le16_to_cpu(rb->rf_suballoc_slot),
2110 le64_to_cpu(rb->rf_suballoc_loc),
2111 le64_to_cpu(rb->rf_blkno),
2112 le16_to_cpu(rb->rf_suballoc_bit));
2118 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2119 OCFS2_JOURNAL_ACCESS_WRITE);
2125 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2127 le32_add_cpu(&rb->rf_clusters, -1);
2130 * check whether we need to restore the root refcount block if
2131 * there is no leaf extent block at atll.
2133 if (!rb->rf_list.l_next_free_rec) {
2134 BUG_ON(rb->rf_clusters);
2136 trace_ocfs2_restore_refcount_block(
2137 (unsigned long long)ref_root_bh->b_blocknr);
2142 memset(&rb->rf_records, 0, sb->s_blocksize -
2143 offsetof(struct ocfs2_refcount_block, rf_records));
2144 rb->rf_records.rl_count =
2145 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2148 ocfs2_journal_dirty(handle, ref_root_bh);
2154 int ocfs2_increase_refcount(handle_t *handle,
2155 struct ocfs2_caching_info *ci,
2156 struct buffer_head *ref_root_bh,
2158 struct ocfs2_alloc_context *meta_ac,
2159 struct ocfs2_cached_dealloc_ctxt *dealloc)
2161 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2166 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2167 struct ocfs2_caching_info *ci,
2168 struct buffer_head *ref_root_bh,
2169 struct buffer_head *ref_leaf_bh,
2170 int index, u64 cpos, unsigned int len,
2171 struct ocfs2_alloc_context *meta_ac,
2172 struct ocfs2_cached_dealloc_ctxt *dealloc)
2175 struct ocfs2_refcount_block *rb =
2176 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2177 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2179 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2181 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2183 trace_ocfs2_decrease_refcount_rec(
2184 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2185 (unsigned long long)cpos, len);
2187 if (cpos == le64_to_cpu(rec->r_cpos) &&
2188 len == le32_to_cpu(rec->r_clusters))
2189 ret = ocfs2_change_refcount_rec(handle, ci,
2190 ref_leaf_bh, index, 1, -1);
2192 struct ocfs2_refcount_rec split = *rec;
2193 split.r_cpos = cpu_to_le64(cpos);
2194 split.r_clusters = cpu_to_le32(len);
2196 le32_add_cpu(&split.r_refcount, -1);
2198 ret = ocfs2_split_refcount_rec(handle, ci,
2199 ref_root_bh, ref_leaf_bh,
2209 /* Remove the leaf refcount block if it contains no refcount record. */
2210 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2211 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2212 ref_leaf_bh, meta_ac,
2222 static int __ocfs2_decrease_refcount(handle_t *handle,
2223 struct ocfs2_caching_info *ci,
2224 struct buffer_head *ref_root_bh,
2226 struct ocfs2_alloc_context *meta_ac,
2227 struct ocfs2_cached_dealloc_ctxt *dealloc,
2230 int ret = 0, index = 0;
2231 struct ocfs2_refcount_rec rec;
2232 unsigned int r_count = 0, r_len;
2233 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2234 struct buffer_head *ref_leaf_bh = NULL;
2236 trace_ocfs2_decrease_refcount(
2237 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2238 (unsigned long long)cpos, len, delete);
2241 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2242 cpos, len, &rec, &index,
2249 r_count = le32_to_cpu(rec.r_refcount);
2250 BUG_ON(r_count == 0);
2252 BUG_ON(r_count > 1);
2254 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2255 le32_to_cpu(rec.r_clusters)) - cpos;
2257 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2266 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2267 ret = ocfs2_cache_cluster_dealloc(dealloc,
2268 ocfs2_clusters_to_blocks(sb, cpos),
2278 brelse(ref_leaf_bh);
2283 brelse(ref_leaf_bh);
2287 /* Caller must hold refcount tree lock. */
2288 int ocfs2_decrease_refcount(struct inode *inode,
2289 handle_t *handle, u32 cpos, u32 len,
2290 struct ocfs2_alloc_context *meta_ac,
2291 struct ocfs2_cached_dealloc_ctxt *dealloc,
2296 struct buffer_head *ref_root_bh = NULL;
2297 struct ocfs2_refcount_tree *tree;
2299 BUG_ON(!ocfs2_is_refcount_inode(inode));
2301 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2307 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2313 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2320 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2321 cpos, len, meta_ac, dealloc, delete);
2325 brelse(ref_root_bh);
2330 * Mark the already-existing extent at cpos as refcounted for len clusters.
2331 * This adds the refcount extent flag.
2333 * If the existing extent is larger than the request, initiate a
2334 * split. An attempt will be made at merging with adjacent extents.
2336 * The caller is responsible for passing down meta_ac if we'll need it.
2338 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2339 struct ocfs2_extent_tree *et,
2340 handle_t *handle, u32 cpos,
2342 struct ocfs2_alloc_context *meta_ac,
2343 struct ocfs2_cached_dealloc_ctxt *dealloc)
2347 trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2350 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2351 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",
2356 ret = ocfs2_change_extent_flag(handle, et, cpos,
2357 len, phys, meta_ac, dealloc,
2358 OCFS2_EXT_REFCOUNTED, 0);
2367 * Given some contiguous physical clusters, calculate what we need
2368 * for modifying their refcount.
2370 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2371 struct ocfs2_caching_info *ci,
2372 struct buffer_head *ref_root_bh,
2378 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2379 u64 cpos = start_cpos;
2380 struct ocfs2_refcount_block *rb;
2381 struct ocfs2_refcount_rec rec;
2382 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2386 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2387 cpos, clusters, &rec,
2388 &index, &ref_leaf_bh);
2394 if (ref_leaf_bh != prev_bh) {
2396 * Now we encounter a new leaf block, so calculate
2397 * whether we need to extend the old leaf.
2400 rb = (struct ocfs2_refcount_block *)
2403 if (le16_to_cpu(rb->rf_records.rl_used) +
2405 le16_to_cpu(rb->rf_records.rl_count))
2412 prev_bh = ref_leaf_bh;
2416 trace_ocfs2_calc_refcount_meta_credits_iterate(
2417 recs_add, (unsigned long long)cpos, clusters,
2418 (unsigned long long)le64_to_cpu(rec.r_cpos),
2419 le32_to_cpu(rec.r_clusters),
2420 le32_to_cpu(rec.r_refcount), index);
2422 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2423 le32_to_cpu(rec.r_clusters)) - cpos;
2425 * We record all the records which will be inserted to the
2426 * same refcount block, so that we can tell exactly whether
2427 * we need a new refcount block or not.
2429 * If we will insert a new one, this is easy and only happens
2430 * during adding refcounted flag to the extent, so we don't
2431 * have a chance of spliting. We just need one record.
2433 * If the refcount rec already exists, that would be a little
2434 * complicated. we may have to:
2435 * 1) split at the beginning if the start pos isn't aligned.
2436 * we need 1 more record in this case.
2437 * 2) split int the end if the end pos isn't aligned.
2438 * we need 1 more record in this case.
2439 * 3) split in the middle because of file system fragmentation.
2440 * we need 2 more records in this case(we can't detect this
2441 * beforehand, so always think of the worst case).
2443 if (rec.r_refcount) {
2445 /* Check whether we need a split at the beginning. */
2446 if (cpos == start_cpos &&
2447 cpos != le64_to_cpu(rec.r_cpos))
2450 /* Check whether we need a split in the end. */
2451 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2452 le32_to_cpu(rec.r_clusters))
2457 brelse(ref_leaf_bh);
2464 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2466 if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2467 le16_to_cpu(rb->rf_records.rl_count))
2476 *meta_add += ref_blocks;
2477 *credits += ref_blocks;
2480 * So we may need ref_blocks to insert into the tree.
2481 * That also means we need to change the b-tree and add that number
2482 * of records since we never merge them.
2483 * We need one more block for expansion since the new created leaf
2484 * block is also full and needs split.
2486 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2487 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2488 struct ocfs2_extent_tree et;
2490 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2491 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2492 *credits += ocfs2_calc_extend_credits(sb,
2495 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2501 trace_ocfs2_calc_refcount_meta_credits(
2502 (unsigned long long)start_cpos, clusters,
2503 *meta_add, *credits);
2504 brelse(ref_leaf_bh);
2510 * For refcount tree, we will decrease some contiguous clusters
2511 * refcount count, so just go through it to see how many blocks
2512 * we gonna touch and whether we need to create new blocks.
2514 * Normally the refcount blocks store these refcount should be
2515 * contiguous also, so that we can get the number easily.
2516 * We will at most add split 2 refcount records and 2 more
2517 * refcount blocks, so just check it in a rough way.
2519 * Caller must hold refcount tree lock.
2521 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2529 struct buffer_head *ref_root_bh = NULL;
2530 struct ocfs2_refcount_tree *tree;
2531 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2533 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2534 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",
2539 BUG_ON(!ocfs2_is_refcount_inode(inode));
2541 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2542 refcount_loc, &tree);
2548 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2555 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2558 start_cpos, clusters,
2559 ref_blocks, credits);
2565 trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2568 brelse(ref_root_bh);
2572 #define MAX_CONTIG_BYTES 1048576
2574 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2576 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2579 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2581 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2585 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2586 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2587 * while still being less than or equal to it.
2589 * The goal is to break the extent at a multiple of contig_clusters.
2591 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2595 BUG_ON(start > cpos);
2597 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2601 * Given a cluster count of len, pad it out so that it is a multiple
2602 * of contig_clusters.
2604 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2607 unsigned int padded =
2608 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2609 ocfs2_cow_contig_mask(sb);
2619 * Calculate out the start and number of virtual clusters we need to to CoW.
2621 * cpos is vitual start cluster position we want to do CoW in a
2622 * file and write_len is the cluster length.
2623 * max_cpos is the place where we want to stop CoW intentionally.
2625 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2626 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2627 * get good I/O from the resulting extent tree.
2629 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2630 struct ocfs2_extent_list *el,
2638 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2639 struct buffer_head *eb_bh = NULL;
2640 struct ocfs2_extent_block *eb = NULL;
2641 struct ocfs2_extent_rec *rec;
2642 unsigned int want_clusters, rec_end = 0;
2643 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2646 BUG_ON(cpos + write_len > max_cpos);
2648 if (tree_height > 0) {
2649 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2655 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2658 if (el->l_tree_depth) {
2659 ret = ocfs2_error(inode->i_sb,
2660 "Inode %lu has non zero tree depth in leaf block %llu\n",
2662 (unsigned long long)eb_bh->b_blocknr);
2668 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2669 rec = &el->l_recs[i];
2671 if (ocfs2_is_empty_extent(rec)) {
2672 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2673 "index %d\n", inode->i_ino, i);
2677 if (le32_to_cpu(rec->e_cpos) +
2678 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2681 if (*cow_len == 0) {
2683 * We should find a refcounted record in the
2686 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2687 *cow_start = le32_to_cpu(rec->e_cpos);
2691 * If we encounter a hole, a non-refcounted record or
2692 * pass the max_cpos, stop the search.
2694 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2695 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2696 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2699 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2700 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2701 if (rec_end > max_cpos) {
2703 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2707 * How many clusters do we actually need from
2708 * this extent? First we see how many we actually
2709 * need to complete the write. If that's smaller
2710 * than contig_clusters, we try for contig_clusters.
2713 want_clusters = write_len;
2715 want_clusters = (cpos + write_len) -
2716 (*cow_start + *cow_len);
2717 if (want_clusters < contig_clusters)
2718 want_clusters = contig_clusters;
2721 * If the write does not cover the whole extent, we
2722 * need to calculate how we're going to split the extent.
2723 * We try to do it on contig_clusters boundaries.
2725 * Any extent smaller than contig_clusters will be
2726 * CoWed in its entirety.
2728 if (leaf_clusters <= contig_clusters)
2729 *cow_len += leaf_clusters;
2730 else if (*cow_len || (*cow_start == cpos)) {
2732 * This extent needs to be CoW'd from its
2733 * beginning, so all we have to do is compute
2734 * how many clusters to grab. We align
2735 * want_clusters to the edge of contig_clusters
2736 * to get better I/O.
2738 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2741 if (leaf_clusters < want_clusters)
2742 *cow_len += leaf_clusters;
2744 *cow_len += want_clusters;
2745 } else if ((*cow_start + contig_clusters) >=
2746 (cpos + write_len)) {
2748 * Breaking off contig_clusters at the front
2749 * of the extent will cover our write. That's
2752 *cow_len = contig_clusters;
2753 } else if ((rec_end - cpos) <= contig_clusters) {
2755 * Breaking off contig_clusters at the tail of
2756 * this extent will cover cpos.
2758 *cow_start = rec_end - contig_clusters;
2759 *cow_len = contig_clusters;
2760 } else if ((rec_end - cpos) <= want_clusters) {
2762 * While we can't fit the entire write in this
2763 * extent, we know that the write goes from cpos
2764 * to the end of the extent. Break that off.
2765 * We try to break it at some multiple of
2766 * contig_clusters from the front of the extent.
2767 * Failing that (ie, cpos is within
2768 * contig_clusters of the front), we'll CoW the
2771 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2773 *cow_len = rec_end - *cow_start;
2776 * Ok, the entire write lives in the middle of
2777 * this extent. Let's try to slice the extent up
2778 * nicely. Optimally, our CoW region starts at
2779 * m*contig_clusters from the beginning of the
2780 * extent and goes for n*contig_clusters,
2781 * covering the entire write.
2783 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2786 want_clusters = (cpos + write_len) - *cow_start;
2787 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2789 if (*cow_start + want_clusters <= rec_end)
2790 *cow_len = want_clusters;
2792 *cow_len = rec_end - *cow_start;
2795 /* Have we covered our entire write yet? */
2796 if ((*cow_start + *cow_len) >= (cpos + write_len))
2800 * If we reach the end of the extent block and don't get enough
2801 * clusters, continue with the next extent block if possible.
2803 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2804 eb && eb->h_next_leaf_blk) {
2808 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2809 le64_to_cpu(eb->h_next_leaf_blk),
2816 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2828 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2829 * num_clusters in data_tree "et" and change the refcount for the old
2830 * clusters(starting form p_cluster) in the refcount tree.
2833 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2834 * more new leaf records.
2835 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2836 * just give data_ac = NULL.
2838 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2839 u32 p_cluster, u32 num_clusters,
2840 struct ocfs2_extent_tree *et,
2841 struct ocfs2_caching_info *ref_ci,
2842 struct buffer_head *ref_root_bh,
2843 struct ocfs2_alloc_context **meta_ac,
2844 struct ocfs2_alloc_context **data_ac,
2847 int ret = 0, meta_add = 0;
2848 int num_free_extents = ocfs2_num_free_extents(et);
2850 if (num_free_extents < 0) {
2851 ret = num_free_extents;
2856 if (num_free_extents < num_clusters + 2)
2858 ocfs2_extend_meta_needed(et->et_root_el);
2860 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2862 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2863 p_cluster, num_clusters,
2864 &meta_add, credits);
2870 trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2871 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2879 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2888 ocfs2_free_alloc_context(*meta_ac);
2896 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2898 BUG_ON(buffer_dirty(bh));
2900 clear_buffer_mapped(bh);
2905 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2906 struct inode *inode,
2907 u32 cpos, u32 old_cluster,
2908 u32 new_cluster, u32 new_len)
2910 int ret = 0, partial;
2911 struct super_block *sb = inode->i_sb;
2912 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2915 unsigned int from, to;
2916 loff_t offset, end, map_end;
2917 struct address_space *mapping = inode->i_mapping;
2919 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2920 new_cluster, new_len);
2922 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2923 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2925 * We only duplicate pages until we reach the page contains i_size - 1.
2926 * So trim 'end' to i_size.
2928 if (end > i_size_read(inode))
2929 end = i_size_read(inode);
2931 while (offset < end) {
2932 page_index = offset >> PAGE_SHIFT;
2933 map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
2937 /* from, to is the offset within the page. */
2938 from = offset & (PAGE_SIZE - 1);
2940 if (map_end & (PAGE_SIZE - 1))
2941 to = map_end & (PAGE_SIZE - 1);
2944 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2952 * In case PAGE_SIZE <= CLUSTER_SIZE, we do not expect a dirty
2953 * page, so write it back.
2955 if (PAGE_SIZE <= OCFS2_SB(sb)->s_clustersize) {
2956 if (PageDirty(page)) {
2958 * write_on_page will unlock the page on return
2960 ret = write_one_page(page);
2965 if (!PageUptodate(page)) {
2966 ret = block_read_full_page(page, ocfs2_get_block);
2974 if (page_has_buffers(page)) {
2975 ret = walk_page_buffers(handle, page_buffers(page),
2977 ocfs2_clear_cow_buffer);
2984 ocfs2_map_and_dirty_page(inode,
2986 page, 0, &new_block);
2987 mark_page_accessed(page);
3000 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3001 struct inode *inode,
3002 u32 cpos, u32 old_cluster,
3003 u32 new_cluster, u32 new_len)
3006 struct super_block *sb = inode->i_sb;
3007 struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3008 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3009 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3010 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3011 struct ocfs2_super *osb = OCFS2_SB(sb);
3012 struct buffer_head *old_bh = NULL;
3013 struct buffer_head *new_bh = NULL;
3015 trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3016 new_cluster, new_len);
3018 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3019 new_bh = sb_getblk(osb->sb, new_block);
3020 if (new_bh == NULL) {
3026 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3028 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3034 ret = ocfs2_journal_access(handle, ci, new_bh,
3035 OCFS2_JOURNAL_ACCESS_CREATE);
3041 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3042 ocfs2_journal_dirty(handle, new_bh);
3055 static int ocfs2_clear_ext_refcount(handle_t *handle,
3056 struct ocfs2_extent_tree *et,
3057 u32 cpos, u32 p_cluster, u32 len,
3058 unsigned int ext_flags,
3059 struct ocfs2_alloc_context *meta_ac,
3060 struct ocfs2_cached_dealloc_ctxt *dealloc)
3063 struct ocfs2_extent_rec replace_rec;
3064 struct ocfs2_path *path = NULL;
3065 struct ocfs2_extent_list *el;
3066 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3067 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3069 trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3070 cpos, len, p_cluster, ext_flags);
3072 memset(&replace_rec, 0, sizeof(replace_rec));
3073 replace_rec.e_cpos = cpu_to_le32(cpos);
3074 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3075 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3077 replace_rec.e_flags = ext_flags;
3078 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3080 path = ocfs2_new_path_from_et(et);
3087 ret = ocfs2_find_path(et->et_ci, path, cpos);
3093 el = path_leaf_el(path);
3095 index = ocfs2_search_extent_list(el, cpos);
3097 ret = ocfs2_error(sb,
3098 "Inode %llu has an extent at cpos %u which can no longer be found\n",
3099 (unsigned long long)ino, cpos);
3103 ret = ocfs2_split_extent(handle, et, path, index,
3104 &replace_rec, meta_ac, dealloc);
3109 ocfs2_free_path(path);
3113 static int ocfs2_replace_clusters(handle_t *handle,
3114 struct ocfs2_cow_context *context,
3117 unsigned int ext_flags)
3120 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3121 u64 ino = ocfs2_metadata_cache_owner(ci);
3123 trace_ocfs2_replace_clusters((unsigned long long)ino,
3124 cpos, old, new, len, ext_flags);
3126 /*If the old clusters is unwritten, no need to duplicate. */
3127 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3128 ret = context->cow_duplicate_clusters(handle, context->inode,
3129 cpos, old, new, len);
3136 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3137 cpos, new, len, ext_flags,
3138 context->meta_ac, &context->dealloc);
3145 int ocfs2_cow_sync_writeback(struct super_block *sb,
3146 struct inode *inode,
3147 u32 cpos, u32 num_clusters)
3150 loff_t offset, end, map_end;
3154 if (ocfs2_should_order_data(inode))
3157 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3158 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3160 ret = filemap_fdatawrite_range(inode->i_mapping,
3167 while (offset < end) {
3168 page_index = offset >> PAGE_SHIFT;
3169 map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
3173 page = find_or_create_page(inode->i_mapping,
3174 page_index, GFP_NOFS);
3177 wait_on_page_writeback(page);
3178 if (PageError(page)) {
3182 mark_page_accessed(page);
3195 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3196 u32 v_cluster, u32 *p_cluster,
3198 unsigned int *extent_flags)
3200 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3201 num_clusters, extent_flags);
3204 static int ocfs2_make_clusters_writable(struct super_block *sb,
3205 struct ocfs2_cow_context *context,
3206 u32 cpos, u32 p_cluster,
3207 u32 num_clusters, unsigned int e_flags)
3209 int ret, delete, index, credits = 0;
3210 u32 new_bit, new_len, orig_num_clusters;
3211 unsigned int set_len;
3212 struct ocfs2_super *osb = OCFS2_SB(sb);
3214 struct buffer_head *ref_leaf_bh = NULL;
3215 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3216 struct ocfs2_refcount_rec rec;
3218 trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3219 num_clusters, e_flags);
3221 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3224 context->ref_root_bh,
3226 &context->data_ac, &credits);
3232 if (context->post_refcount)
3233 credits += context->post_refcount->credits;
3235 credits += context->extra_credits;
3236 handle = ocfs2_start_trans(osb, credits);
3237 if (IS_ERR(handle)) {
3238 ret = PTR_ERR(handle);
3243 orig_num_clusters = num_clusters;
3245 while (num_clusters) {
3246 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3247 p_cluster, num_clusters,
3248 &rec, &index, &ref_leaf_bh);
3254 BUG_ON(!rec.r_refcount);
3255 set_len = min((u64)p_cluster + num_clusters,
3256 le64_to_cpu(rec.r_cpos) +
3257 le32_to_cpu(rec.r_clusters)) - p_cluster;
3260 * There are many different situation here.
3261 * 1. If refcount == 1, remove the flag and don't COW.
3262 * 2. If refcount > 1, allocate clusters.
3263 * Here we may not allocate r_len once at a time, so continue
3264 * until we reach num_clusters.
3266 if (le32_to_cpu(rec.r_refcount) == 1) {
3268 ret = ocfs2_clear_ext_refcount(handle,
3281 ret = __ocfs2_claim_clusters(handle,
3284 &new_bit, &new_len);
3290 ret = ocfs2_replace_clusters(handle, context,
3291 cpos, p_cluster, new_bit,
3300 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3301 context->ref_root_bh,
3304 &context->dealloc, delete);
3311 p_cluster += set_len;
3312 num_clusters -= set_len;
3313 brelse(ref_leaf_bh);
3317 /* handle any post_cow action. */
3318 if (context->post_refcount && context->post_refcount->func) {
3319 ret = context->post_refcount->func(context->inode, handle,
3320 context->post_refcount->para);
3328 * Here we should write the new page out first if we are
3329 * in write-back mode.
3331 if (context->get_clusters == ocfs2_di_get_clusters) {
3332 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3339 ocfs2_commit_trans(osb, handle);
3342 if (context->data_ac) {
3343 ocfs2_free_alloc_context(context->data_ac);
3344 context->data_ac = NULL;
3346 if (context->meta_ac) {
3347 ocfs2_free_alloc_context(context->meta_ac);
3348 context->meta_ac = NULL;
3350 brelse(ref_leaf_bh);
3355 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3358 struct inode *inode = context->inode;
3359 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3360 u32 p_cluster, num_clusters;
3361 unsigned int ext_flags;
3362 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3364 if (!ocfs2_refcount_tree(osb)) {
3365 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",
3369 ocfs2_init_dealloc_ctxt(&context->dealloc);
3372 ret = context->get_clusters(context, cow_start, &p_cluster,
3373 &num_clusters, &ext_flags);
3379 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3381 if (cow_len < num_clusters)
3382 num_clusters = cow_len;
3384 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3385 cow_start, p_cluster,
3386 num_clusters, ext_flags);
3392 cow_len -= num_clusters;
3393 cow_start += num_clusters;
3396 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3397 ocfs2_schedule_truncate_log_flush(osb, 1);
3398 ocfs2_run_deallocs(osb, &context->dealloc);
3405 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3406 * past max_cpos. This will stop when it runs into a hole or an
3407 * unrefcounted extent.
3409 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3410 struct buffer_head *di_bh,
3411 u32 cpos, u32 write_len, u32 max_cpos)
3414 u32 cow_start = 0, cow_len = 0;
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(!ocfs2_is_refcount_inode(inode));
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_cow_context *context = NULL;
3632 u32 cow_start, cow_len;
3634 BUG_ON(!ocfs2_is_refcount_inode(inode));
3636 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3637 cpos, write_len, UINT_MAX,
3638 &cow_start, &cow_len);
3644 BUG_ON(cow_len == 0);
3646 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3653 context->inode = inode;
3654 context->cow_start = cow_start;
3655 context->cow_len = cow_len;
3656 context->ref_tree = ref_tree;
3657 context->ref_root_bh = ref_root_bh;
3658 context->cow_object = xv;
3660 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3661 /* We need the extra credits for duplicate_clusters by jbd. */
3662 context->extra_credits =
3663 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3664 context->get_clusters = ocfs2_xattr_value_get_clusters;
3665 context->post_refcount = post;
3667 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3668 INODE_CACHE(inode), vb);
3670 ret = ocfs2_replace_cow(context);
3680 * Insert a new extent into refcount tree and mark a extent rec
3681 * as refcounted in the dinode tree.
3683 int ocfs2_add_refcount_flag(struct inode *inode,
3684 struct ocfs2_extent_tree *data_et,
3685 struct ocfs2_caching_info *ref_ci,
3686 struct buffer_head *ref_root_bh,
3687 u32 cpos, u32 p_cluster, u32 num_clusters,
3688 struct ocfs2_cached_dealloc_ctxt *dealloc,
3689 struct ocfs2_post_refcount *post)
3693 int credits = 1, ref_blocks = 0;
3694 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3695 struct ocfs2_alloc_context *meta_ac = NULL;
3697 /* We need to be able to handle at least an extent tree split. */
3698 ref_blocks = ocfs2_extend_meta_needed(data_et->et_root_el);
3700 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3701 ref_ci, ref_root_bh,
3702 p_cluster, num_clusters,
3703 &ref_blocks, &credits);
3709 trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3712 ret = ocfs2_reserve_new_metadata_blocks(osb,
3713 ref_blocks, &meta_ac);
3721 credits += post->credits;
3723 handle = ocfs2_start_trans(osb, credits);
3724 if (IS_ERR(handle)) {
3725 ret = PTR_ERR(handle);
3730 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3731 cpos, num_clusters, p_cluster,
3738 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3739 p_cluster, num_clusters, 0,
3746 if (post && post->func) {
3747 ret = post->func(inode, handle, post->para);
3753 ocfs2_commit_trans(osb, handle);
3756 ocfs2_free_alloc_context(meta_ac);
3760 static int ocfs2_change_ctime(struct inode *inode,
3761 struct buffer_head *di_bh)
3765 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3767 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3768 OCFS2_INODE_UPDATE_CREDITS);
3769 if (IS_ERR(handle)) {
3770 ret = PTR_ERR(handle);
3775 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3776 OCFS2_JOURNAL_ACCESS_WRITE);
3782 inode->i_ctime = current_time(inode);
3783 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3784 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3786 ocfs2_journal_dirty(handle, di_bh);
3789 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3794 static int ocfs2_attach_refcount_tree(struct inode *inode,
3795 struct buffer_head *di_bh)
3797 int ret, data_changed = 0;
3798 struct buffer_head *ref_root_bh = NULL;
3799 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3800 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3801 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3802 struct ocfs2_refcount_tree *ref_tree;
3803 unsigned int ext_flags;
3805 u32 cpos, num_clusters, clusters, p_cluster;
3806 struct ocfs2_cached_dealloc_ctxt dealloc;
3807 struct ocfs2_extent_tree di_et;
3809 ocfs2_init_dealloc_ctxt(&dealloc);
3811 if (!ocfs2_is_refcount_inode(inode)) {
3812 ret = ocfs2_create_refcount_tree(inode, di_bh);
3819 BUG_ON(!di->i_refcount_loc);
3820 ret = ocfs2_lock_refcount_tree(osb,
3821 le64_to_cpu(di->i_refcount_loc), 1,
3822 &ref_tree, &ref_root_bh);
3828 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3831 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3833 size = i_size_read(inode);
3834 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3837 while (cpos < clusters) {
3838 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3839 &num_clusters, &ext_flags);
3844 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3845 ret = ocfs2_add_refcount_flag(inode, &di_et,
3848 p_cluster, num_clusters,
3857 cpos += num_clusters;
3861 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3862 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3873 ret = ocfs2_change_ctime(inode, di_bh);
3879 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3880 brelse(ref_root_bh);
3882 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3883 ocfs2_schedule_truncate_log_flush(osb, 1);
3884 ocfs2_run_deallocs(osb, &dealloc);
3888 * Empty the extent map so that we may get the right extent
3889 * record from the disk.
3891 ocfs2_extent_map_trunc(inode, 0);
3896 static int ocfs2_add_refcounted_extent(struct inode *inode,
3897 struct ocfs2_extent_tree *et,
3898 struct ocfs2_caching_info *ref_ci,
3899 struct buffer_head *ref_root_bh,
3900 u32 cpos, u32 p_cluster, u32 num_clusters,
3901 unsigned int ext_flags,
3902 struct ocfs2_cached_dealloc_ctxt *dealloc)
3907 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3908 struct ocfs2_alloc_context *meta_ac = NULL;
3910 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3911 p_cluster, num_clusters,
3913 ref_root_bh, &meta_ac,
3920 handle = ocfs2_start_trans(osb, credits);
3921 if (IS_ERR(handle)) {
3922 ret = PTR_ERR(handle);
3927 ret = ocfs2_insert_extent(handle, et, cpos,
3928 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3929 num_clusters, ext_flags, meta_ac);
3935 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3936 p_cluster, num_clusters,
3943 ret = dquot_alloc_space_nodirty(inode,
3944 ocfs2_clusters_to_bytes(osb->sb, num_clusters));
3949 ocfs2_commit_trans(osb, handle);
3952 ocfs2_free_alloc_context(meta_ac);
3956 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3957 struct buffer_head *s_bh,
3958 struct inode *t_inode,
3959 struct buffer_head *t_bh)
3963 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3964 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3965 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3967 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3969 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3970 if (IS_ERR(handle)) {
3971 ret = PTR_ERR(handle);
3976 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3977 OCFS2_JOURNAL_ACCESS_WRITE);
3983 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3984 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3985 le16_to_cpu(s_di->id2.i_data.id_count));
3986 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3987 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3988 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3989 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3991 ocfs2_journal_dirty(handle, t_bh);
3994 ocfs2_commit_trans(osb, handle);
3999 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
4000 struct inode *t_inode,
4001 struct buffer_head *t_bh,
4002 struct ocfs2_caching_info *ref_ci,
4003 struct buffer_head *ref_root_bh,
4004 struct ocfs2_cached_dealloc_ctxt *dealloc)
4007 u32 p_cluster, num_clusters, clusters, cpos;
4009 unsigned int ext_flags;
4010 struct ocfs2_extent_tree et;
4012 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4014 size = i_size_read(s_inode);
4015 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4018 while (cpos < clusters) {
4019 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4020 &num_clusters, &ext_flags);
4026 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4027 ref_ci, ref_root_bh,
4038 cpos += num_clusters;
4046 * change the new file's attributes to the src.
4048 * reflink creates a snapshot of a file, that means the attributes
4049 * must be identical except for three exceptions - nlink, ino, and ctime.
4051 static int ocfs2_complete_reflink(struct inode *s_inode,
4052 struct buffer_head *s_bh,
4053 struct inode *t_inode,
4054 struct buffer_head *t_bh,
4059 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4060 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4061 loff_t size = i_size_read(s_inode);
4063 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4064 OCFS2_INODE_UPDATE_CREDITS);
4065 if (IS_ERR(handle)) {
4066 ret = PTR_ERR(handle);
4071 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4072 OCFS2_JOURNAL_ACCESS_WRITE);
4078 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4079 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4080 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4081 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4082 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4083 i_size_write(t_inode, size);
4084 t_inode->i_blocks = s_inode->i_blocks;
4086 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4087 di->i_clusters = s_di->i_clusters;
4088 di->i_size = s_di->i_size;
4089 di->i_dyn_features = s_di->i_dyn_features;
4090 di->i_attr = s_di->i_attr;
4093 t_inode->i_uid = s_inode->i_uid;
4094 t_inode->i_gid = s_inode->i_gid;
4095 t_inode->i_mode = s_inode->i_mode;
4096 di->i_uid = s_di->i_uid;
4097 di->i_gid = s_di->i_gid;
4098 di->i_mode = s_di->i_mode;
4102 * we want mtime to appear identical to the source and
4105 t_inode->i_ctime = current_time(t_inode);
4107 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4108 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4110 t_inode->i_mtime = s_inode->i_mtime;
4111 di->i_mtime = s_di->i_mtime;
4112 di->i_mtime_nsec = s_di->i_mtime_nsec;
4115 ocfs2_journal_dirty(handle, t_bh);
4118 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4122 static int ocfs2_create_reflink_node(struct inode *s_inode,
4123 struct buffer_head *s_bh,
4124 struct inode *t_inode,
4125 struct buffer_head *t_bh,
4129 struct buffer_head *ref_root_bh = NULL;
4130 struct ocfs2_cached_dealloc_ctxt dealloc;
4131 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4132 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4133 struct ocfs2_refcount_tree *ref_tree;
4135 ocfs2_init_dealloc_ctxt(&dealloc);
4137 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4138 le64_to_cpu(di->i_refcount_loc));
4144 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4145 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4152 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4153 1, &ref_tree, &ref_root_bh);
4159 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4160 &ref_tree->rf_ci, ref_root_bh,
4164 goto out_unlock_refcount;
4167 out_unlock_refcount:
4168 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4169 brelse(ref_root_bh);
4171 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4172 ocfs2_schedule_truncate_log_flush(osb, 1);
4173 ocfs2_run_deallocs(osb, &dealloc);
4179 static int __ocfs2_reflink(struct dentry *old_dentry,
4180 struct buffer_head *old_bh,
4181 struct inode *new_inode,
4185 struct inode *inode = d_inode(old_dentry);
4186 struct buffer_head *new_bh = NULL;
4188 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4194 ret = filemap_fdatawrite(inode->i_mapping);
4200 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4206 inode_lock_nested(new_inode, I_MUTEX_CHILD);
4207 ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4208 OI_LS_REFLINK_TARGET);
4214 ret = ocfs2_create_reflink_node(inode, old_bh,
4215 new_inode, new_bh, preserve);
4221 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4222 ret = ocfs2_reflink_xattrs(inode, old_bh,
4231 ret = ocfs2_complete_reflink(inode, old_bh,
4232 new_inode, new_bh, preserve);
4237 ocfs2_inode_unlock(new_inode, 1);
4240 inode_unlock(new_inode);
4243 ret = filemap_fdatawait(inode->i_mapping);
4250 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4251 struct dentry *new_dentry, bool preserve)
4253 int error, had_lock;
4254 struct inode *inode = d_inode(old_dentry);
4255 struct buffer_head *old_bh = NULL;
4256 struct inode *new_orphan_inode = NULL;
4257 struct ocfs2_lock_holder oh;
4259 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4263 error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4270 error = ocfs2_rw_lock(inode, 1);
4276 error = ocfs2_inode_lock(inode, &old_bh, 1);
4279 ocfs2_rw_unlock(inode, 1);
4283 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4284 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4285 error = __ocfs2_reflink(old_dentry, old_bh,
4286 new_orphan_inode, preserve);
4287 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4288 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4290 ocfs2_inode_unlock(inode, 1);
4291 ocfs2_rw_unlock(inode, 1);
4299 had_lock = ocfs2_inode_lock_tracker(new_orphan_inode, NULL, 1,
4307 /* If the security isn't preserved, we need to re-initialize them. */
4309 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4310 &new_dentry->d_name);
4315 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4320 ocfs2_inode_unlock_tracker(new_orphan_inode, 1, &oh, had_lock);
4323 if (new_orphan_inode) {
4325 * We need to open_unlock the inode no matter whether we
4326 * succeed or not, so that other nodes can delete it later.
4328 ocfs2_open_unlock(new_orphan_inode);
4330 iput(new_orphan_inode);
4337 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4338 * sys_reflink(). This will go away when vfs_reflink() exists in
4342 /* copied from may_create in VFS. */
4343 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4345 if (d_really_is_positive(child))
4347 if (IS_DEADDIR(dir))
4349 return inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
4353 * ocfs2_vfs_reflink - Create a reference-counted link
4355 * @old_dentry: source dentry + inode
4356 * @dir: directory to create the target
4357 * @new_dentry: target dentry
4358 * @preserve: if true, preserve all file attributes
4360 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4361 struct dentry *new_dentry, bool preserve)
4363 struct inode *inode = d_inode(old_dentry);
4369 error = ocfs2_may_create(dir, new_dentry);
4373 if (dir->i_sb != inode->i_sb)
4377 * A reflink to an append-only or immutable file cannot be created.
4379 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4382 /* Only regular files can be reflinked. */
4383 if (!S_ISREG(inode->i_mode))
4387 * If the caller wants to preserve ownership, they require the
4391 if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4393 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4398 * If the caller is modifying any aspect of the attributes, they
4399 * are not creating a snapshot. They need read permission on the
4403 error = inode_permission(&init_user_ns, inode, MAY_READ);
4409 error = dquot_initialize(dir);
4411 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4412 inode_unlock(inode);
4414 fsnotify_create(dir, new_dentry);
4418 * Most codes are copied from sys_linkat.
4420 int ocfs2_reflink_ioctl(struct inode *inode,
4421 const char __user *oldname,
4422 const char __user *newname,
4425 struct dentry *new_dentry;
4426 struct path old_path, new_path;
4429 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4432 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4438 new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4439 error = PTR_ERR(new_dentry);
4440 if (IS_ERR(new_dentry)) {
4446 if (old_path.mnt != new_path.mnt) {
4451 error = ocfs2_vfs_reflink(old_path.dentry,
4452 d_inode(new_path.dentry),
4453 new_dentry, preserve);
4455 done_path_create(&new_path, new_dentry);
4457 path_put(&old_path);
4462 /* Update destination inode size, if necessary. */
4463 int ocfs2_reflink_update_dest(struct inode *dest,
4464 struct buffer_head *d_bh,
4470 dest->i_blocks = ocfs2_inode_sector_count(dest);
4472 if (newlen <= i_size_read(dest))
4475 handle = ocfs2_start_trans(OCFS2_SB(dest->i_sb),
4476 OCFS2_INODE_UPDATE_CREDITS);
4477 if (IS_ERR(handle)) {
4478 ret = PTR_ERR(handle);
4483 /* Extend i_size if needed. */
4484 spin_lock(&OCFS2_I(dest)->ip_lock);
4485 if (newlen > i_size_read(dest))
4486 i_size_write(dest, newlen);
4487 spin_unlock(&OCFS2_I(dest)->ip_lock);
4488 dest->i_ctime = dest->i_mtime = current_time(dest);
4490 ret = ocfs2_mark_inode_dirty(handle, dest, d_bh);
4497 ocfs2_commit_trans(OCFS2_SB(dest->i_sb), handle);
4501 /* Remap the range pos_in:len in s_inode to pos_out:len in t_inode. */
4502 static loff_t ocfs2_reflink_remap_extent(struct inode *s_inode,
4503 struct buffer_head *s_bh,
4505 struct inode *t_inode,
4506 struct buffer_head *t_bh,
4509 struct ocfs2_cached_dealloc_ctxt *dealloc)
4511 struct ocfs2_extent_tree s_et;
4512 struct ocfs2_extent_tree t_et;
4513 struct ocfs2_dinode *dis;
4514 struct buffer_head *ref_root_bh = NULL;
4515 struct ocfs2_refcount_tree *ref_tree;
4516 struct ocfs2_super *osb;
4517 loff_t remapped_bytes = 0;
4518 loff_t pstart, plen;
4519 u32 p_cluster, num_clusters, slast, spos, tpos, remapped_clus = 0;
4520 unsigned int ext_flags;
4523 osb = OCFS2_SB(s_inode->i_sb);
4524 dis = (struct ocfs2_dinode *)s_bh->b_data;
4525 ocfs2_init_dinode_extent_tree(&s_et, INODE_CACHE(s_inode), s_bh);
4526 ocfs2_init_dinode_extent_tree(&t_et, INODE_CACHE(t_inode), t_bh);
4528 spos = ocfs2_bytes_to_clusters(s_inode->i_sb, pos_in);
4529 tpos = ocfs2_bytes_to_clusters(t_inode->i_sb, pos_out);
4530 slast = ocfs2_clusters_for_bytes(s_inode->i_sb, pos_in + len);
4532 while (spos < slast) {
4533 if (fatal_signal_pending(current)) {
4538 /* Look up the extent. */
4539 ret = ocfs2_get_clusters(s_inode, spos, &p_cluster,
4540 &num_clusters, &ext_flags);
4546 num_clusters = min_t(u32, num_clusters, slast - spos);
4548 /* Punch out the dest range. */
4549 pstart = ocfs2_clusters_to_bytes(t_inode->i_sb, tpos);
4550 plen = ocfs2_clusters_to_bytes(t_inode->i_sb, num_clusters);
4551 ret = ocfs2_remove_inode_range(t_inode, t_bh, pstart, plen);
4560 /* Lock the refcount btree... */
4561 ret = ocfs2_lock_refcount_tree(osb,
4562 le64_to_cpu(dis->i_refcount_loc),
4563 1, &ref_tree, &ref_root_bh);
4569 /* Mark s_inode's extent as refcounted. */
4570 if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) {
4571 ret = ocfs2_add_refcount_flag(s_inode, &s_et,
4574 p_cluster, num_clusters,
4578 goto out_unlock_refcount;
4582 /* Map in the new extent. */
4583 ext_flags |= OCFS2_EXT_REFCOUNTED;
4584 ret = ocfs2_add_refcounted_extent(t_inode, &t_et,
4593 goto out_unlock_refcount;
4596 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4597 brelse(ref_root_bh);
4599 spos += num_clusters;
4600 tpos += num_clusters;
4601 remapped_clus += num_clusters;
4605 out_unlock_refcount:
4606 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4607 brelse(ref_root_bh);
4609 remapped_bytes = ocfs2_clusters_to_bytes(t_inode->i_sb, remapped_clus);
4610 remapped_bytes = min_t(loff_t, len, remapped_bytes);
4612 return remapped_bytes > 0 ? remapped_bytes : ret;
4615 /* Set up refcount tree and remap s_inode to t_inode. */
4616 loff_t ocfs2_reflink_remap_blocks(struct inode *s_inode,
4617 struct buffer_head *s_bh,
4619 struct inode *t_inode,
4620 struct buffer_head *t_bh,
4624 struct ocfs2_cached_dealloc_ctxt dealloc;
4625 struct ocfs2_super *osb;
4626 struct ocfs2_dinode *dis;
4627 struct ocfs2_dinode *dit;
4630 osb = OCFS2_SB(s_inode->i_sb);
4631 dis = (struct ocfs2_dinode *)s_bh->b_data;
4632 dit = (struct ocfs2_dinode *)t_bh->b_data;
4633 ocfs2_init_dealloc_ctxt(&dealloc);
4636 * If we're reflinking the entire file and the source is inline
4637 * data, just copy the contents.
4639 if (pos_in == pos_out && pos_in == 0 && len == i_size_read(s_inode) &&
4640 i_size_read(t_inode) <= len &&
4641 (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) {
4642 ret = ocfs2_duplicate_inline_data(s_inode, s_bh, t_inode, t_bh);
4649 * If both inodes belong to two different refcount groups then
4650 * forget it because we don't know how (or want) to go merging
4654 if (ocfs2_is_refcount_inode(s_inode) &&
4655 ocfs2_is_refcount_inode(t_inode) &&
4656 le64_to_cpu(dis->i_refcount_loc) !=
4657 le64_to_cpu(dit->i_refcount_loc))
4660 /* Neither inode has a refcount tree. Add one to s_inode. */
4661 if (!ocfs2_is_refcount_inode(s_inode) &&
4662 !ocfs2_is_refcount_inode(t_inode)) {
4663 ret = ocfs2_create_refcount_tree(s_inode, s_bh);
4670 /* Ensure that both inodes end up with the same refcount tree. */
4671 if (!ocfs2_is_refcount_inode(s_inode)) {
4672 ret = ocfs2_set_refcount_tree(s_inode, s_bh,
4673 le64_to_cpu(dit->i_refcount_loc));
4679 if (!ocfs2_is_refcount_inode(t_inode)) {
4680 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4681 le64_to_cpu(dis->i_refcount_loc));
4688 /* Turn off inline data in the dest file. */
4689 if (OCFS2_I(t_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4690 ret = ocfs2_convert_inline_data_to_extents(t_inode, t_bh);
4697 /* Actually remap extents now. */
4698 ret = ocfs2_reflink_remap_extent(s_inode, s_bh, pos_in, t_inode, t_bh,
4699 pos_out, len, &dealloc);
4706 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4707 ocfs2_schedule_truncate_log_flush(osb, 1);
4708 ocfs2_run_deallocs(osb, &dealloc);
4714 /* Lock an inode and grab a bh pointing to the inode. */
4715 int ocfs2_reflink_inodes_lock(struct inode *s_inode,
4716 struct buffer_head **bh_s,
4717 struct inode *t_inode,
4718 struct buffer_head **bh_t)
4720 struct inode *inode1 = s_inode;
4721 struct inode *inode2 = t_inode;
4722 struct ocfs2_inode_info *oi1;
4723 struct ocfs2_inode_info *oi2;
4724 struct buffer_head *bh1 = NULL;
4725 struct buffer_head *bh2 = NULL;
4726 bool same_inode = (s_inode == t_inode);
4727 bool need_swap = (inode1->i_ino > inode2->i_ino);
4730 /* First grab the VFS and rw locks. */
4731 lock_two_nondirectories(s_inode, t_inode);
4733 swap(inode1, inode2);
4735 status = ocfs2_rw_lock(inode1, 1);
4741 status = ocfs2_rw_lock(inode2, 1);
4748 /* Now go for the cluster locks */
4749 oi1 = OCFS2_I(inode1);
4750 oi2 = OCFS2_I(inode2);
4752 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
4753 (unsigned long long)oi2->ip_blkno);
4755 /* We always want to lock the one with the lower lockid first. */
4756 if (oi1->ip_blkno > oi2->ip_blkno)
4757 mlog_errno(-ENOLCK);
4760 status = ocfs2_inode_lock_nested(inode1, &bh1, 1,
4761 OI_LS_REFLINK_TARGET);
4763 if (status != -ENOENT)
4770 status = ocfs2_inode_lock_nested(inode2, &bh2, 1,
4771 OI_LS_REFLINK_TARGET);
4773 if (status != -ENOENT)
4782 * If we swapped inode order above, we have to swap the buffer heads
4783 * before passing them back to the caller.
4790 trace_ocfs2_double_lock_end(
4791 (unsigned long long)oi1->ip_blkno,
4792 (unsigned long long)oi2->ip_blkno);
4797 ocfs2_inode_unlock(inode1, 1);
4800 ocfs2_rw_unlock(inode2, 1);
4802 ocfs2_rw_unlock(inode1, 1);
4804 unlock_two_nondirectories(s_inode, t_inode);
4808 /* Unlock both inodes and release buffers. */
4809 void ocfs2_reflink_inodes_unlock(struct inode *s_inode,
4810 struct buffer_head *s_bh,
4811 struct inode *t_inode,
4812 struct buffer_head *t_bh)
4814 ocfs2_inode_unlock(s_inode, 1);
4815 ocfs2_rw_unlock(s_inode, 1);
4817 if (s_inode != t_inode) {
4818 ocfs2_inode_unlock(t_inode, 1);
4819 ocfs2_rw_unlock(t_inode, 1);
4822 unlock_two_nondirectories(s_inode, t_inode);