2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_path *path, int level);
30 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
31 *root, struct btrfs_key *ins_key,
32 struct btrfs_path *path, int data_size, int extend);
33 static int push_node_left(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root, struct extent_buffer *dst,
35 struct extent_buffer *src, int empty);
36 static int balance_node_right(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct extent_buffer *dst_buf,
39 struct extent_buffer *src_buf);
40 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
42 static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
43 struct extent_buffer *eb);
44 static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
46 struct btrfs_path *btrfs_alloc_path(void)
48 struct btrfs_path *path;
49 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
54 * set all locked nodes in the path to blocking locks. This should
55 * be done before scheduling
57 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
60 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
61 if (!p->nodes[i] || !p->locks[i])
63 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
64 if (p->locks[i] == BTRFS_READ_LOCK)
65 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
66 else if (p->locks[i] == BTRFS_WRITE_LOCK)
67 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
72 * reset all the locked nodes in the patch to spinning locks.
74 * held is used to keep lockdep happy, when lockdep is enabled
75 * we set held to a blocking lock before we go around and
76 * retake all the spinlocks in the path. You can safely use NULL
79 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
80 struct extent_buffer *held, int held_rw)
84 #ifdef CONFIG_DEBUG_LOCK_ALLOC
85 /* lockdep really cares that we take all of these spinlocks
86 * in the right order. If any of the locks in the path are not
87 * currently blocking, it is going to complain. So, make really
88 * really sure by forcing the path to blocking before we clear
92 btrfs_set_lock_blocking_rw(held, held_rw);
93 if (held_rw == BTRFS_WRITE_LOCK)
94 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
95 else if (held_rw == BTRFS_READ_LOCK)
96 held_rw = BTRFS_READ_LOCK_BLOCKING;
98 btrfs_set_path_blocking(p);
101 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
102 if (p->nodes[i] && p->locks[i]) {
103 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
104 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
105 p->locks[i] = BTRFS_WRITE_LOCK;
106 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
107 p->locks[i] = BTRFS_READ_LOCK;
111 #ifdef CONFIG_DEBUG_LOCK_ALLOC
113 btrfs_clear_lock_blocking_rw(held, held_rw);
117 /* this also releases the path */
118 void btrfs_free_path(struct btrfs_path *p)
122 btrfs_release_path(p);
123 kmem_cache_free(btrfs_path_cachep, p);
127 * path release drops references on the extent buffers in the path
128 * and it drops any locks held by this path
130 * It is safe to call this on paths that no locks or extent buffers held.
132 noinline void btrfs_release_path(struct btrfs_path *p)
136 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
141 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
144 free_extent_buffer(p->nodes[i]);
150 * safely gets a reference on the root node of a tree. A lock
151 * is not taken, so a concurrent writer may put a different node
152 * at the root of the tree. See btrfs_lock_root_node for the
155 * The extent buffer returned by this has a reference taken, so
156 * it won't disappear. It may stop being the root of the tree
157 * at any time because there are no locks held.
159 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
161 struct extent_buffer *eb;
165 eb = rcu_dereference(root->node);
168 * RCU really hurts here, we could free up the root node because
169 * it was cow'ed but we may not get the new root node yet so do
170 * the inc_not_zero dance and if it doesn't work then
171 * synchronize_rcu and try again.
173 if (atomic_inc_not_zero(&eb->refs)) {
183 /* loop around taking references on and locking the root node of the
184 * tree until you end up with a lock on the root. A locked buffer
185 * is returned, with a reference held.
187 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
189 struct extent_buffer *eb;
192 eb = btrfs_root_node(root);
194 if (eb == root->node)
196 btrfs_tree_unlock(eb);
197 free_extent_buffer(eb);
202 /* loop around taking references on and locking the root node of the
203 * tree until you end up with a lock on the root. A locked buffer
204 * is returned, with a reference held.
206 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
208 struct extent_buffer *eb;
211 eb = btrfs_root_node(root);
212 btrfs_tree_read_lock(eb);
213 if (eb == root->node)
215 btrfs_tree_read_unlock(eb);
216 free_extent_buffer(eb);
221 /* cowonly root (everything not a reference counted cow subvolume), just get
222 * put onto a simple dirty list. transaction.c walks this to make sure they
223 * get properly updated on disk.
225 static void add_root_to_dirty_list(struct btrfs_root *root)
227 spin_lock(&root->fs_info->trans_lock);
228 if (root->track_dirty && list_empty(&root->dirty_list)) {
229 list_add(&root->dirty_list,
230 &root->fs_info->dirty_cowonly_roots);
232 spin_unlock(&root->fs_info->trans_lock);
236 * used by snapshot creation to make a copy of a root for a tree with
237 * a given objectid. The buffer with the new root node is returned in
238 * cow_ret, and this func returns zero on success or a negative error code.
240 int btrfs_copy_root(struct btrfs_trans_handle *trans,
241 struct btrfs_root *root,
242 struct extent_buffer *buf,
243 struct extent_buffer **cow_ret, u64 new_root_objectid)
245 struct extent_buffer *cow;
248 struct btrfs_disk_key disk_key;
250 WARN_ON(root->ref_cows && trans->transid !=
251 root->fs_info->running_transaction->transid);
252 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
254 level = btrfs_header_level(buf);
256 btrfs_item_key(buf, &disk_key, 0);
258 btrfs_node_key(buf, &disk_key, 0);
260 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
261 new_root_objectid, &disk_key, level,
266 copy_extent_buffer(cow, buf, 0, 0, cow->len);
267 btrfs_set_header_bytenr(cow, cow->start);
268 btrfs_set_header_generation(cow, trans->transid);
269 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
270 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
271 BTRFS_HEADER_FLAG_RELOC);
272 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
273 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
275 btrfs_set_header_owner(cow, new_root_objectid);
277 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(cow),
280 WARN_ON(btrfs_header_generation(buf) > trans->transid);
281 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
282 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
284 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
289 btrfs_mark_buffer_dirty(cow);
298 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
299 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
301 MOD_LOG_ROOT_REPLACE,
304 struct tree_mod_move {
309 struct tree_mod_root {
314 struct tree_mod_elem {
316 u64 index; /* shifted logical */
320 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
323 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
326 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
327 struct btrfs_disk_key key;
330 /* this is used for op == MOD_LOG_MOVE_KEYS */
331 struct tree_mod_move move;
333 /* this is used for op == MOD_LOG_ROOT_REPLACE */
334 struct tree_mod_root old_root;
337 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
339 read_lock(&fs_info->tree_mod_log_lock);
342 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
344 read_unlock(&fs_info->tree_mod_log_lock);
347 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
349 write_lock(&fs_info->tree_mod_log_lock);
352 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
354 write_unlock(&fs_info->tree_mod_log_lock);
358 * Increment the upper half of tree_mod_seq, set lower half zero.
360 * Must be called with fs_info->tree_mod_seq_lock held.
362 static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info)
364 u64 seq = atomic64_read(&fs_info->tree_mod_seq);
365 seq &= 0xffffffff00000000ull;
367 atomic64_set(&fs_info->tree_mod_seq, seq);
372 * Increment the lower half of tree_mod_seq.
374 * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers
375 * are generated should not technically require a spin lock here. (Rationale:
376 * incrementing the minor while incrementing the major seq number is between its
377 * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it
378 * just returns a unique sequence number as usual.) We have decided to leave
379 * that requirement in here and rethink it once we notice it really imposes a
380 * problem on some workload.
382 static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info)
384 return atomic64_inc_return(&fs_info->tree_mod_seq);
388 * return the last minor in the previous major tree_mod_seq number
390 u64 btrfs_tree_mod_seq_prev(u64 seq)
392 return (seq & 0xffffffff00000000ull) - 1ull;
396 * This adds a new blocker to the tree mod log's blocker list if the @elem
397 * passed does not already have a sequence number set. So when a caller expects
398 * to record tree modifications, it should ensure to set elem->seq to zero
399 * before calling btrfs_get_tree_mod_seq.
400 * Returns a fresh, unused tree log modification sequence number, even if no new
403 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
404 struct seq_list *elem)
408 tree_mod_log_write_lock(fs_info);
409 spin_lock(&fs_info->tree_mod_seq_lock);
411 elem->seq = btrfs_inc_tree_mod_seq_major(fs_info);
412 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
414 seq = btrfs_inc_tree_mod_seq_minor(fs_info);
415 spin_unlock(&fs_info->tree_mod_seq_lock);
416 tree_mod_log_write_unlock(fs_info);
421 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
422 struct seq_list *elem)
424 struct rb_root *tm_root;
425 struct rb_node *node;
426 struct rb_node *next;
427 struct seq_list *cur_elem;
428 struct tree_mod_elem *tm;
429 u64 min_seq = (u64)-1;
430 u64 seq_putting = elem->seq;
435 spin_lock(&fs_info->tree_mod_seq_lock);
436 list_del(&elem->list);
439 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
440 if (cur_elem->seq < min_seq) {
441 if (seq_putting > cur_elem->seq) {
443 * blocker with lower sequence number exists, we
444 * cannot remove anything from the log
446 spin_unlock(&fs_info->tree_mod_seq_lock);
449 min_seq = cur_elem->seq;
452 spin_unlock(&fs_info->tree_mod_seq_lock);
455 * anything that's lower than the lowest existing (read: blocked)
456 * sequence number can be removed from the tree.
458 tree_mod_log_write_lock(fs_info);
459 tm_root = &fs_info->tree_mod_log;
460 for (node = rb_first(tm_root); node; node = next) {
461 next = rb_next(node);
462 tm = container_of(node, struct tree_mod_elem, node);
463 if (tm->seq > min_seq)
465 rb_erase(node, tm_root);
468 tree_mod_log_write_unlock(fs_info);
472 * key order of the log:
475 * the index is the shifted logical of the *new* root node for root replace
476 * operations, or the shifted logical of the affected block for all other
480 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
482 struct rb_root *tm_root;
483 struct rb_node **new;
484 struct rb_node *parent = NULL;
485 struct tree_mod_elem *cur;
490 tree_mod_log_write_lock(fs_info);
491 if (list_empty(&fs_info->tree_mod_seq_list)) {
492 tree_mod_log_write_unlock(fs_info);
494 * Ok we no longer care about logging modifications, free up tm
495 * and return 0. Any callers shouldn't be using tm after
496 * calling tree_mod_log_insert, but if they do we can just
497 * change this to return a special error code to let the callers
498 * do their own thing.
504 spin_lock(&fs_info->tree_mod_seq_lock);
505 tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info);
506 spin_unlock(&fs_info->tree_mod_seq_lock);
508 tm_root = &fs_info->tree_mod_log;
509 new = &tm_root->rb_node;
511 cur = container_of(*new, struct tree_mod_elem, node);
513 if (cur->index < tm->index)
514 new = &((*new)->rb_left);
515 else if (cur->index > tm->index)
516 new = &((*new)->rb_right);
517 else if (cur->seq < tm->seq)
518 new = &((*new)->rb_left);
519 else if (cur->seq > tm->seq)
520 new = &((*new)->rb_right);
528 rb_link_node(&tm->node, parent, new);
529 rb_insert_color(&tm->node, tm_root);
531 tree_mod_log_write_unlock(fs_info);
536 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
537 * returns zero with the tree_mod_log_lock acquired. The caller must hold
538 * this until all tree mod log insertions are recorded in the rb tree and then
539 * call tree_mod_log_write_unlock() to release.
541 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
542 struct extent_buffer *eb) {
544 if (list_empty(&(fs_info)->tree_mod_seq_list))
546 if (eb && btrfs_header_level(eb) == 0)
552 __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
553 struct extent_buffer *eb, int slot,
554 enum mod_log_op op, gfp_t flags)
556 struct tree_mod_elem *tm;
558 tm = kzalloc(sizeof(*tm), flags);
562 tm->index = eb->start >> PAGE_CACHE_SHIFT;
563 if (op != MOD_LOG_KEY_ADD) {
564 btrfs_node_key(eb, &tm->key, slot);
565 tm->blockptr = btrfs_node_blockptr(eb, slot);
569 tm->generation = btrfs_node_ptr_generation(eb, slot);
571 return __tree_mod_log_insert(fs_info, tm);
575 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
576 struct extent_buffer *eb, int slot,
577 enum mod_log_op op, gfp_t flags)
579 if (tree_mod_dont_log(fs_info, eb))
582 return __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
586 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
587 struct extent_buffer *eb, int dst_slot, int src_slot,
588 int nr_items, gfp_t flags)
590 struct tree_mod_elem *tm;
594 if (tree_mod_dont_log(fs_info, eb))
598 * When we override something during the move, we log these removals.
599 * This can only happen when we move towards the beginning of the
600 * buffer, i.e. dst_slot < src_slot.
602 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
603 ret = __tree_mod_log_insert_key(fs_info, eb, i + dst_slot,
604 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
608 tm = kzalloc(sizeof(*tm), flags);
612 tm->index = eb->start >> PAGE_CACHE_SHIFT;
614 tm->move.dst_slot = dst_slot;
615 tm->move.nr_items = nr_items;
616 tm->op = MOD_LOG_MOVE_KEYS;
618 return __tree_mod_log_insert(fs_info, tm);
622 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
628 if (btrfs_header_level(eb) == 0)
631 nritems = btrfs_header_nritems(eb);
632 for (i = nritems - 1; i >= 0; i--) {
633 ret = __tree_mod_log_insert_key(fs_info, eb, i,
634 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
640 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
641 struct extent_buffer *old_root,
642 struct extent_buffer *new_root, gfp_t flags,
645 struct tree_mod_elem *tm;
647 if (tree_mod_dont_log(fs_info, NULL))
651 __tree_mod_log_free_eb(fs_info, old_root);
653 tm = kzalloc(sizeof(*tm), flags);
657 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
658 tm->old_root.logical = old_root->start;
659 tm->old_root.level = btrfs_header_level(old_root);
660 tm->generation = btrfs_header_generation(old_root);
661 tm->op = MOD_LOG_ROOT_REPLACE;
663 return __tree_mod_log_insert(fs_info, tm);
666 static struct tree_mod_elem *
667 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
670 struct rb_root *tm_root;
671 struct rb_node *node;
672 struct tree_mod_elem *cur = NULL;
673 struct tree_mod_elem *found = NULL;
674 u64 index = start >> PAGE_CACHE_SHIFT;
676 tree_mod_log_read_lock(fs_info);
677 tm_root = &fs_info->tree_mod_log;
678 node = tm_root->rb_node;
680 cur = container_of(node, struct tree_mod_elem, node);
681 if (cur->index < index) {
682 node = node->rb_left;
683 } else if (cur->index > index) {
684 node = node->rb_right;
685 } else if (cur->seq < min_seq) {
686 node = node->rb_left;
687 } else if (!smallest) {
688 /* we want the node with the highest seq */
690 BUG_ON(found->seq > cur->seq);
692 node = node->rb_left;
693 } else if (cur->seq > min_seq) {
694 /* we want the node with the smallest seq */
696 BUG_ON(found->seq < cur->seq);
698 node = node->rb_right;
704 tree_mod_log_read_unlock(fs_info);
710 * this returns the element from the log with the smallest time sequence
711 * value that's in the log (the oldest log item). any element with a time
712 * sequence lower than min_seq will be ignored.
714 static struct tree_mod_elem *
715 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
718 return __tree_mod_log_search(fs_info, start, min_seq, 1);
722 * this returns the element from the log with the largest time sequence
723 * value that's in the log (the most recent log item). any element with
724 * a time sequence lower than min_seq will be ignored.
726 static struct tree_mod_elem *
727 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
729 return __tree_mod_log_search(fs_info, start, min_seq, 0);
733 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
734 struct extent_buffer *src, unsigned long dst_offset,
735 unsigned long src_offset, int nr_items)
740 if (tree_mod_dont_log(fs_info, NULL))
743 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
746 for (i = 0; i < nr_items; i++) {
747 ret = __tree_mod_log_insert_key(fs_info, src,
749 MOD_LOG_KEY_REMOVE, GFP_NOFS);
751 ret = __tree_mod_log_insert_key(fs_info, dst,
760 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
761 int dst_offset, int src_offset, int nr_items)
764 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
770 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
771 struct extent_buffer *eb, int slot, int atomic)
775 ret = __tree_mod_log_insert_key(fs_info, eb, slot,
777 atomic ? GFP_ATOMIC : GFP_NOFS);
782 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
784 if (tree_mod_dont_log(fs_info, eb))
786 __tree_mod_log_free_eb(fs_info, eb);
790 tree_mod_log_set_root_pointer(struct btrfs_root *root,
791 struct extent_buffer *new_root_node,
795 ret = tree_mod_log_insert_root(root->fs_info, root->node,
796 new_root_node, GFP_NOFS, log_removal);
801 * check if the tree block can be shared by multiple trees
803 int btrfs_block_can_be_shared(struct btrfs_root *root,
804 struct extent_buffer *buf)
807 * Tree blocks not in refernece counted trees and tree roots
808 * are never shared. If a block was allocated after the last
809 * snapshot and the block was not allocated by tree relocation,
810 * we know the block is not shared.
812 if (root->ref_cows &&
813 buf != root->node && buf != root->commit_root &&
814 (btrfs_header_generation(buf) <=
815 btrfs_root_last_snapshot(&root->root_item) ||
816 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
818 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
819 if (root->ref_cows &&
820 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
826 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct extent_buffer *buf,
829 struct extent_buffer *cow,
839 * Backrefs update rules:
841 * Always use full backrefs for extent pointers in tree block
842 * allocated by tree relocation.
844 * If a shared tree block is no longer referenced by its owner
845 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
846 * use full backrefs for extent pointers in tree block.
848 * If a tree block is been relocating
849 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
850 * use full backrefs for extent pointers in tree block.
851 * The reason for this is some operations (such as drop tree)
852 * are only allowed for blocks use full backrefs.
855 if (btrfs_block_can_be_shared(root, buf)) {
856 ret = btrfs_lookup_extent_info(trans, root, buf->start,
857 btrfs_header_level(buf), 1,
863 btrfs_std_error(root->fs_info, ret);
868 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
869 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
870 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
875 owner = btrfs_header_owner(buf);
876 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
877 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
880 if ((owner == root->root_key.objectid ||
881 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
882 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
883 ret = btrfs_inc_ref(trans, root, buf, 1, 1);
884 BUG_ON(ret); /* -ENOMEM */
886 if (root->root_key.objectid ==
887 BTRFS_TREE_RELOC_OBJECTID) {
888 ret = btrfs_dec_ref(trans, root, buf, 0, 1);
889 BUG_ON(ret); /* -ENOMEM */
890 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
891 BUG_ON(ret); /* -ENOMEM */
893 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
896 if (root->root_key.objectid ==
897 BTRFS_TREE_RELOC_OBJECTID)
898 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
900 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
901 BUG_ON(ret); /* -ENOMEM */
903 if (new_flags != 0) {
904 int level = btrfs_header_level(buf);
906 ret = btrfs_set_disk_extent_flags(trans, root,
909 new_flags, level, 0);
914 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
915 if (root->root_key.objectid ==
916 BTRFS_TREE_RELOC_OBJECTID)
917 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
919 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
920 BUG_ON(ret); /* -ENOMEM */
921 ret = btrfs_dec_ref(trans, root, buf, 1, 1);
922 BUG_ON(ret); /* -ENOMEM */
924 clean_tree_block(trans, root, buf);
931 * does the dirty work in cow of a single block. The parent block (if
932 * supplied) is updated to point to the new cow copy. The new buffer is marked
933 * dirty and returned locked. If you modify the block it needs to be marked
936 * search_start -- an allocation hint for the new block
938 * empty_size -- a hint that you plan on doing more cow. This is the size in
939 * bytes the allocator should try to find free next to the block it returns.
940 * This is just a hint and may be ignored by the allocator.
942 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
943 struct btrfs_root *root,
944 struct extent_buffer *buf,
945 struct extent_buffer *parent, int parent_slot,
946 struct extent_buffer **cow_ret,
947 u64 search_start, u64 empty_size)
949 struct btrfs_disk_key disk_key;
950 struct extent_buffer *cow;
959 btrfs_assert_tree_locked(buf);
961 WARN_ON(root->ref_cows && trans->transid !=
962 root->fs_info->running_transaction->transid);
963 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
965 level = btrfs_header_level(buf);
968 btrfs_item_key(buf, &disk_key, 0);
970 btrfs_node_key(buf, &disk_key, 0);
972 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
974 parent_start = parent->start;
980 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
981 root->root_key.objectid, &disk_key,
982 level, search_start, empty_size);
986 /* cow is set to blocking by btrfs_init_new_buffer */
988 copy_extent_buffer(cow, buf, 0, 0, cow->len);
989 btrfs_set_header_bytenr(cow, cow->start);
990 btrfs_set_header_generation(cow, trans->transid);
991 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
992 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
993 BTRFS_HEADER_FLAG_RELOC);
994 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
995 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
997 btrfs_set_header_owner(cow, root->root_key.objectid);
999 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(cow),
1002 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1004 btrfs_abort_transaction(trans, root, ret);
1008 if (root->ref_cows) {
1009 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1014 if (buf == root->node) {
1015 WARN_ON(parent && parent != buf);
1016 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1017 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1018 parent_start = buf->start;
1022 extent_buffer_get(cow);
1023 tree_mod_log_set_root_pointer(root, cow, 1);
1024 rcu_assign_pointer(root->node, cow);
1026 btrfs_free_tree_block(trans, root, buf, parent_start,
1028 free_extent_buffer(buf);
1029 add_root_to_dirty_list(root);
1031 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1032 parent_start = parent->start;
1036 WARN_ON(trans->transid != btrfs_header_generation(parent));
1037 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1038 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1039 btrfs_set_node_blockptr(parent, parent_slot,
1041 btrfs_set_node_ptr_generation(parent, parent_slot,
1043 btrfs_mark_buffer_dirty(parent);
1045 tree_mod_log_free_eb(root->fs_info, buf);
1046 btrfs_free_tree_block(trans, root, buf, parent_start,
1050 btrfs_tree_unlock(buf);
1051 free_extent_buffer_stale(buf);
1052 btrfs_mark_buffer_dirty(cow);
1058 * returns the logical address of the oldest predecessor of the given root.
1059 * entries older than time_seq are ignored.
1061 static struct tree_mod_elem *
1062 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1063 struct extent_buffer *eb_root, u64 time_seq)
1065 struct tree_mod_elem *tm;
1066 struct tree_mod_elem *found = NULL;
1067 u64 root_logical = eb_root->start;
1074 * the very last operation that's logged for a root is the replacement
1075 * operation (if it is replaced at all). this has the index of the *new*
1076 * root, making it the very first operation that's logged for this root.
1079 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1084 * if there are no tree operation for the oldest root, we simply
1085 * return it. this should only happen if that (old) root is at
1092 * if there's an operation that's not a root replacement, we
1093 * found the oldest version of our root. normally, we'll find a
1094 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1096 if (tm->op != MOD_LOG_ROOT_REPLACE)
1100 root_logical = tm->old_root.logical;
1104 /* if there's no old root to return, return what we found instead */
1112 * tm is a pointer to the first operation to rewind within eb. then, all
1113 * previous operations will be rewinded (until we reach something older than
1117 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1118 u64 time_seq, struct tree_mod_elem *first_tm)
1121 struct rb_node *next;
1122 struct tree_mod_elem *tm = first_tm;
1123 unsigned long o_dst;
1124 unsigned long o_src;
1125 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1127 n = btrfs_header_nritems(eb);
1128 tree_mod_log_read_lock(fs_info);
1129 while (tm && tm->seq >= time_seq) {
1131 * all the operations are recorded with the operator used for
1132 * the modification. as we're going backwards, we do the
1133 * opposite of each operation here.
1136 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1137 BUG_ON(tm->slot < n);
1139 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1140 case MOD_LOG_KEY_REMOVE:
1141 btrfs_set_node_key(eb, &tm->key, tm->slot);
1142 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1143 btrfs_set_node_ptr_generation(eb, tm->slot,
1147 case MOD_LOG_KEY_REPLACE:
1148 BUG_ON(tm->slot >= n);
1149 btrfs_set_node_key(eb, &tm->key, tm->slot);
1150 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1151 btrfs_set_node_ptr_generation(eb, tm->slot,
1154 case MOD_LOG_KEY_ADD:
1155 /* if a move operation is needed it's in the log */
1158 case MOD_LOG_MOVE_KEYS:
1159 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1160 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1161 memmove_extent_buffer(eb, o_dst, o_src,
1162 tm->move.nr_items * p_size);
1164 case MOD_LOG_ROOT_REPLACE:
1166 * this operation is special. for roots, this must be
1167 * handled explicitly before rewinding.
1168 * for non-roots, this operation may exist if the node
1169 * was a root: root A -> child B; then A gets empty and
1170 * B is promoted to the new root. in the mod log, we'll
1171 * have a root-replace operation for B, a tree block
1172 * that is no root. we simply ignore that operation.
1176 next = rb_next(&tm->node);
1179 tm = container_of(next, struct tree_mod_elem, node);
1180 if (tm->index != first_tm->index)
1183 tree_mod_log_read_unlock(fs_info);
1184 btrfs_set_header_nritems(eb, n);
1188 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1189 * is returned. If rewind operations happen, a fresh buffer is returned. The
1190 * returned buffer is always read-locked. If the returned buffer is not the
1191 * input buffer, the lock on the input buffer is released and the input buffer
1192 * is freed (its refcount is decremented).
1194 static struct extent_buffer *
1195 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1196 struct extent_buffer *eb, u64 time_seq)
1198 struct extent_buffer *eb_rewin;
1199 struct tree_mod_elem *tm;
1204 if (btrfs_header_level(eb) == 0)
1207 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1211 btrfs_set_path_blocking(path);
1212 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1214 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1215 BUG_ON(tm->slot != 0);
1216 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1217 fs_info->tree_root->nodesize);
1219 btrfs_tree_read_unlock_blocking(eb);
1220 free_extent_buffer(eb);
1223 btrfs_set_header_bytenr(eb_rewin, eb->start);
1224 btrfs_set_header_backref_rev(eb_rewin,
1225 btrfs_header_backref_rev(eb));
1226 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1227 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1229 eb_rewin = btrfs_clone_extent_buffer(eb);
1231 btrfs_tree_read_unlock_blocking(eb);
1232 free_extent_buffer(eb);
1237 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1238 btrfs_tree_read_unlock_blocking(eb);
1239 free_extent_buffer(eb);
1241 extent_buffer_get(eb_rewin);
1242 btrfs_tree_read_lock(eb_rewin);
1243 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1244 WARN_ON(btrfs_header_nritems(eb_rewin) >
1245 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1251 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1252 * value. If there are no changes, the current root->root_node is returned. If
1253 * anything changed in between, there's a fresh buffer allocated on which the
1254 * rewind operations are done. In any case, the returned buffer is read locked.
1255 * Returns NULL on error (with no locks held).
1257 static inline struct extent_buffer *
1258 get_old_root(struct btrfs_root *root, u64 time_seq)
1260 struct tree_mod_elem *tm;
1261 struct extent_buffer *eb = NULL;
1262 struct extent_buffer *eb_root;
1263 struct extent_buffer *old;
1264 struct tree_mod_root *old_root = NULL;
1265 u64 old_generation = 0;
1269 eb_root = btrfs_read_lock_root_node(root);
1270 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1274 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1275 old_root = &tm->old_root;
1276 old_generation = tm->generation;
1277 logical = old_root->logical;
1279 logical = eb_root->start;
1282 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1283 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1284 btrfs_tree_read_unlock(eb_root);
1285 free_extent_buffer(eb_root);
1286 blocksize = btrfs_level_size(root, old_root->level);
1287 old = read_tree_block(root, logical, blocksize, 0);
1288 if (!old || !extent_buffer_uptodate(old)) {
1289 free_extent_buffer(old);
1290 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1294 eb = btrfs_clone_extent_buffer(old);
1295 free_extent_buffer(old);
1297 } else if (old_root) {
1298 btrfs_tree_read_unlock(eb_root);
1299 free_extent_buffer(eb_root);
1300 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1302 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1303 eb = btrfs_clone_extent_buffer(eb_root);
1304 btrfs_tree_read_unlock_blocking(eb_root);
1305 free_extent_buffer(eb_root);
1310 extent_buffer_get(eb);
1311 btrfs_tree_read_lock(eb);
1313 btrfs_set_header_bytenr(eb, eb->start);
1314 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1315 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1316 btrfs_set_header_level(eb, old_root->level);
1317 btrfs_set_header_generation(eb, old_generation);
1320 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1322 WARN_ON(btrfs_header_level(eb) != 0);
1323 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1328 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1330 struct tree_mod_elem *tm;
1332 struct extent_buffer *eb_root = btrfs_root_node(root);
1334 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1335 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1336 level = tm->old_root.level;
1338 level = btrfs_header_level(eb_root);
1340 free_extent_buffer(eb_root);
1345 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1346 struct btrfs_root *root,
1347 struct extent_buffer *buf)
1349 /* ensure we can see the force_cow */
1353 * We do not need to cow a block if
1354 * 1) this block is not created or changed in this transaction;
1355 * 2) this block does not belong to TREE_RELOC tree;
1356 * 3) the root is not forced COW.
1358 * What is forced COW:
1359 * when we create snapshot during commiting the transaction,
1360 * after we've finished coping src root, we must COW the shared
1361 * block to ensure the metadata consistency.
1363 if (btrfs_header_generation(buf) == trans->transid &&
1364 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1365 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1366 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1373 * cows a single block, see __btrfs_cow_block for the real work.
1374 * This version of it has extra checks so that a block isn't cow'd more than
1375 * once per transaction, as long as it hasn't been written yet
1377 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1378 struct btrfs_root *root, struct extent_buffer *buf,
1379 struct extent_buffer *parent, int parent_slot,
1380 struct extent_buffer **cow_ret)
1385 if (trans->transaction != root->fs_info->running_transaction)
1386 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1388 root->fs_info->running_transaction->transid);
1390 if (trans->transid != root->fs_info->generation)
1391 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1392 trans->transid, root->fs_info->generation);
1394 if (!should_cow_block(trans, root, buf)) {
1399 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1402 btrfs_set_lock_blocking(parent);
1403 btrfs_set_lock_blocking(buf);
1405 ret = __btrfs_cow_block(trans, root, buf, parent,
1406 parent_slot, cow_ret, search_start, 0);
1408 trace_btrfs_cow_block(root, buf, *cow_ret);
1414 * helper function for defrag to decide if two blocks pointed to by a
1415 * node are actually close by
1417 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1419 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1421 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1427 * compare two keys in a memcmp fashion
1429 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1431 struct btrfs_key k1;
1433 btrfs_disk_key_to_cpu(&k1, disk);
1435 return btrfs_comp_cpu_keys(&k1, k2);
1439 * same as comp_keys only with two btrfs_key's
1441 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1443 if (k1->objectid > k2->objectid)
1445 if (k1->objectid < k2->objectid)
1447 if (k1->type > k2->type)
1449 if (k1->type < k2->type)
1451 if (k1->offset > k2->offset)
1453 if (k1->offset < k2->offset)
1459 * this is used by the defrag code to go through all the
1460 * leaves pointed to by a node and reallocate them so that
1461 * disk order is close to key order
1463 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1464 struct btrfs_root *root, struct extent_buffer *parent,
1465 int start_slot, u64 *last_ret,
1466 struct btrfs_key *progress)
1468 struct extent_buffer *cur;
1471 u64 search_start = *last_ret;
1481 int progress_passed = 0;
1482 struct btrfs_disk_key disk_key;
1484 parent_level = btrfs_header_level(parent);
1486 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1487 WARN_ON(trans->transid != root->fs_info->generation);
1489 parent_nritems = btrfs_header_nritems(parent);
1490 blocksize = btrfs_level_size(root, parent_level - 1);
1491 end_slot = parent_nritems;
1493 if (parent_nritems == 1)
1496 btrfs_set_lock_blocking(parent);
1498 for (i = start_slot; i < end_slot; i++) {
1501 btrfs_node_key(parent, &disk_key, i);
1502 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1505 progress_passed = 1;
1506 blocknr = btrfs_node_blockptr(parent, i);
1507 gen = btrfs_node_ptr_generation(parent, i);
1508 if (last_block == 0)
1509 last_block = blocknr;
1512 other = btrfs_node_blockptr(parent, i - 1);
1513 close = close_blocks(blocknr, other, blocksize);
1515 if (!close && i < end_slot - 2) {
1516 other = btrfs_node_blockptr(parent, i + 1);
1517 close = close_blocks(blocknr, other, blocksize);
1520 last_block = blocknr;
1524 cur = btrfs_find_tree_block(root, blocknr, blocksize);
1526 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1529 if (!cur || !uptodate) {
1531 cur = read_tree_block(root, blocknr,
1533 if (!cur || !extent_buffer_uptodate(cur)) {
1534 free_extent_buffer(cur);
1537 } else if (!uptodate) {
1538 err = btrfs_read_buffer(cur, gen);
1540 free_extent_buffer(cur);
1545 if (search_start == 0)
1546 search_start = last_block;
1548 btrfs_tree_lock(cur);
1549 btrfs_set_lock_blocking(cur);
1550 err = __btrfs_cow_block(trans, root, cur, parent, i,
1553 (end_slot - i) * blocksize));
1555 btrfs_tree_unlock(cur);
1556 free_extent_buffer(cur);
1559 search_start = cur->start;
1560 last_block = cur->start;
1561 *last_ret = search_start;
1562 btrfs_tree_unlock(cur);
1563 free_extent_buffer(cur);
1569 * The leaf data grows from end-to-front in the node.
1570 * this returns the address of the start of the last item,
1571 * which is the stop of the leaf data stack
1573 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1574 struct extent_buffer *leaf)
1576 u32 nr = btrfs_header_nritems(leaf);
1578 return BTRFS_LEAF_DATA_SIZE(root);
1579 return btrfs_item_offset_nr(leaf, nr - 1);
1584 * search for key in the extent_buffer. The items start at offset p,
1585 * and they are item_size apart. There are 'max' items in p.
1587 * the slot in the array is returned via slot, and it points to
1588 * the place where you would insert key if it is not found in
1591 * slot may point to max if the key is bigger than all of the keys
1593 static noinline int generic_bin_search(struct extent_buffer *eb,
1595 int item_size, struct btrfs_key *key,
1602 struct btrfs_disk_key *tmp = NULL;
1603 struct btrfs_disk_key unaligned;
1604 unsigned long offset;
1606 unsigned long map_start = 0;
1607 unsigned long map_len = 0;
1610 while (low < high) {
1611 mid = (low + high) / 2;
1612 offset = p + mid * item_size;
1614 if (!kaddr || offset < map_start ||
1615 (offset + sizeof(struct btrfs_disk_key)) >
1616 map_start + map_len) {
1618 err = map_private_extent_buffer(eb, offset,
1619 sizeof(struct btrfs_disk_key),
1620 &kaddr, &map_start, &map_len);
1623 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1626 read_extent_buffer(eb, &unaligned,
1627 offset, sizeof(unaligned));
1632 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1635 ret = comp_keys(tmp, key);
1651 * simple bin_search frontend that does the right thing for
1654 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1655 int level, int *slot)
1658 return generic_bin_search(eb,
1659 offsetof(struct btrfs_leaf, items),
1660 sizeof(struct btrfs_item),
1661 key, btrfs_header_nritems(eb),
1664 return generic_bin_search(eb,
1665 offsetof(struct btrfs_node, ptrs),
1666 sizeof(struct btrfs_key_ptr),
1667 key, btrfs_header_nritems(eb),
1671 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1672 int level, int *slot)
1674 return bin_search(eb, key, level, slot);
1677 static void root_add_used(struct btrfs_root *root, u32 size)
1679 spin_lock(&root->accounting_lock);
1680 btrfs_set_root_used(&root->root_item,
1681 btrfs_root_used(&root->root_item) + size);
1682 spin_unlock(&root->accounting_lock);
1685 static void root_sub_used(struct btrfs_root *root, u32 size)
1687 spin_lock(&root->accounting_lock);
1688 btrfs_set_root_used(&root->root_item,
1689 btrfs_root_used(&root->root_item) - size);
1690 spin_unlock(&root->accounting_lock);
1693 /* given a node and slot number, this reads the blocks it points to. The
1694 * extent buffer is returned with a reference taken (but unlocked).
1695 * NULL is returned on error.
1697 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1698 struct extent_buffer *parent, int slot)
1700 int level = btrfs_header_level(parent);
1701 struct extent_buffer *eb;
1705 if (slot >= btrfs_header_nritems(parent))
1710 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1711 btrfs_level_size(root, level - 1),
1712 btrfs_node_ptr_generation(parent, slot));
1713 if (eb && !extent_buffer_uptodate(eb)) {
1714 free_extent_buffer(eb);
1722 * node level balancing, used to make sure nodes are in proper order for
1723 * item deletion. We balance from the top down, so we have to make sure
1724 * that a deletion won't leave an node completely empty later on.
1726 static noinline int balance_level(struct btrfs_trans_handle *trans,
1727 struct btrfs_root *root,
1728 struct btrfs_path *path, int level)
1730 struct extent_buffer *right = NULL;
1731 struct extent_buffer *mid;
1732 struct extent_buffer *left = NULL;
1733 struct extent_buffer *parent = NULL;
1737 int orig_slot = path->slots[level];
1743 mid = path->nodes[level];
1745 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1746 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1747 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1749 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1751 if (level < BTRFS_MAX_LEVEL - 1) {
1752 parent = path->nodes[level + 1];
1753 pslot = path->slots[level + 1];
1757 * deal with the case where there is only one pointer in the root
1758 * by promoting the node below to a root
1761 struct extent_buffer *child;
1763 if (btrfs_header_nritems(mid) != 1)
1766 /* promote the child to a root */
1767 child = read_node_slot(root, mid, 0);
1770 btrfs_std_error(root->fs_info, ret);
1774 btrfs_tree_lock(child);
1775 btrfs_set_lock_blocking(child);
1776 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1778 btrfs_tree_unlock(child);
1779 free_extent_buffer(child);
1783 tree_mod_log_set_root_pointer(root, child, 1);
1784 rcu_assign_pointer(root->node, child);
1786 add_root_to_dirty_list(root);
1787 btrfs_tree_unlock(child);
1789 path->locks[level] = 0;
1790 path->nodes[level] = NULL;
1791 clean_tree_block(trans, root, mid);
1792 btrfs_tree_unlock(mid);
1793 /* once for the path */
1794 free_extent_buffer(mid);
1796 root_sub_used(root, mid->len);
1797 btrfs_free_tree_block(trans, root, mid, 0, 1);
1798 /* once for the root ptr */
1799 free_extent_buffer_stale(mid);
1802 if (btrfs_header_nritems(mid) >
1803 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1806 left = read_node_slot(root, parent, pslot - 1);
1808 btrfs_tree_lock(left);
1809 btrfs_set_lock_blocking(left);
1810 wret = btrfs_cow_block(trans, root, left,
1811 parent, pslot - 1, &left);
1817 right = read_node_slot(root, parent, pslot + 1);
1819 btrfs_tree_lock(right);
1820 btrfs_set_lock_blocking(right);
1821 wret = btrfs_cow_block(trans, root, right,
1822 parent, pslot + 1, &right);
1829 /* first, try to make some room in the middle buffer */
1831 orig_slot += btrfs_header_nritems(left);
1832 wret = push_node_left(trans, root, left, mid, 1);
1838 * then try to empty the right most buffer into the middle
1841 wret = push_node_left(trans, root, mid, right, 1);
1842 if (wret < 0 && wret != -ENOSPC)
1844 if (btrfs_header_nritems(right) == 0) {
1845 clean_tree_block(trans, root, right);
1846 btrfs_tree_unlock(right);
1847 del_ptr(root, path, level + 1, pslot + 1);
1848 root_sub_used(root, right->len);
1849 btrfs_free_tree_block(trans, root, right, 0, 1);
1850 free_extent_buffer_stale(right);
1853 struct btrfs_disk_key right_key;
1854 btrfs_node_key(right, &right_key, 0);
1855 tree_mod_log_set_node_key(root->fs_info, parent,
1857 btrfs_set_node_key(parent, &right_key, pslot + 1);
1858 btrfs_mark_buffer_dirty(parent);
1861 if (btrfs_header_nritems(mid) == 1) {
1863 * we're not allowed to leave a node with one item in the
1864 * tree during a delete. A deletion from lower in the tree
1865 * could try to delete the only pointer in this node.
1866 * So, pull some keys from the left.
1867 * There has to be a left pointer at this point because
1868 * otherwise we would have pulled some pointers from the
1873 btrfs_std_error(root->fs_info, ret);
1876 wret = balance_node_right(trans, root, mid, left);
1882 wret = push_node_left(trans, root, left, mid, 1);
1888 if (btrfs_header_nritems(mid) == 0) {
1889 clean_tree_block(trans, root, mid);
1890 btrfs_tree_unlock(mid);
1891 del_ptr(root, path, level + 1, pslot);
1892 root_sub_used(root, mid->len);
1893 btrfs_free_tree_block(trans, root, mid, 0, 1);
1894 free_extent_buffer_stale(mid);
1897 /* update the parent key to reflect our changes */
1898 struct btrfs_disk_key mid_key;
1899 btrfs_node_key(mid, &mid_key, 0);
1900 tree_mod_log_set_node_key(root->fs_info, parent,
1902 btrfs_set_node_key(parent, &mid_key, pslot);
1903 btrfs_mark_buffer_dirty(parent);
1906 /* update the path */
1908 if (btrfs_header_nritems(left) > orig_slot) {
1909 extent_buffer_get(left);
1910 /* left was locked after cow */
1911 path->nodes[level] = left;
1912 path->slots[level + 1] -= 1;
1913 path->slots[level] = orig_slot;
1915 btrfs_tree_unlock(mid);
1916 free_extent_buffer(mid);
1919 orig_slot -= btrfs_header_nritems(left);
1920 path->slots[level] = orig_slot;
1923 /* double check we haven't messed things up */
1925 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1929 btrfs_tree_unlock(right);
1930 free_extent_buffer(right);
1933 if (path->nodes[level] != left)
1934 btrfs_tree_unlock(left);
1935 free_extent_buffer(left);
1940 /* Node balancing for insertion. Here we only split or push nodes around
1941 * when they are completely full. This is also done top down, so we
1942 * have to be pessimistic.
1944 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1945 struct btrfs_root *root,
1946 struct btrfs_path *path, int level)
1948 struct extent_buffer *right = NULL;
1949 struct extent_buffer *mid;
1950 struct extent_buffer *left = NULL;
1951 struct extent_buffer *parent = NULL;
1955 int orig_slot = path->slots[level];
1960 mid = path->nodes[level];
1961 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1963 if (level < BTRFS_MAX_LEVEL - 1) {
1964 parent = path->nodes[level + 1];
1965 pslot = path->slots[level + 1];
1971 left = read_node_slot(root, parent, pslot - 1);
1973 /* first, try to make some room in the middle buffer */
1977 btrfs_tree_lock(left);
1978 btrfs_set_lock_blocking(left);
1980 left_nr = btrfs_header_nritems(left);
1981 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1984 ret = btrfs_cow_block(trans, root, left, parent,
1989 wret = push_node_left(trans, root,
1996 struct btrfs_disk_key disk_key;
1997 orig_slot += left_nr;
1998 btrfs_node_key(mid, &disk_key, 0);
1999 tree_mod_log_set_node_key(root->fs_info, parent,
2001 btrfs_set_node_key(parent, &disk_key, pslot);
2002 btrfs_mark_buffer_dirty(parent);
2003 if (btrfs_header_nritems(left) > orig_slot) {
2004 path->nodes[level] = left;
2005 path->slots[level + 1] -= 1;
2006 path->slots[level] = orig_slot;
2007 btrfs_tree_unlock(mid);
2008 free_extent_buffer(mid);
2011 btrfs_header_nritems(left);
2012 path->slots[level] = orig_slot;
2013 btrfs_tree_unlock(left);
2014 free_extent_buffer(left);
2018 btrfs_tree_unlock(left);
2019 free_extent_buffer(left);
2021 right = read_node_slot(root, parent, pslot + 1);
2024 * then try to empty the right most buffer into the middle
2029 btrfs_tree_lock(right);
2030 btrfs_set_lock_blocking(right);
2032 right_nr = btrfs_header_nritems(right);
2033 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2036 ret = btrfs_cow_block(trans, root, right,
2042 wret = balance_node_right(trans, root,
2049 struct btrfs_disk_key disk_key;
2051 btrfs_node_key(right, &disk_key, 0);
2052 tree_mod_log_set_node_key(root->fs_info, parent,
2054 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2055 btrfs_mark_buffer_dirty(parent);
2057 if (btrfs_header_nritems(mid) <= orig_slot) {
2058 path->nodes[level] = right;
2059 path->slots[level + 1] += 1;
2060 path->slots[level] = orig_slot -
2061 btrfs_header_nritems(mid);
2062 btrfs_tree_unlock(mid);
2063 free_extent_buffer(mid);
2065 btrfs_tree_unlock(right);
2066 free_extent_buffer(right);
2070 btrfs_tree_unlock(right);
2071 free_extent_buffer(right);
2077 * readahead one full node of leaves, finding things that are close
2078 * to the block in 'slot', and triggering ra on them.
2080 static void reada_for_search(struct btrfs_root *root,
2081 struct btrfs_path *path,
2082 int level, int slot, u64 objectid)
2084 struct extent_buffer *node;
2085 struct btrfs_disk_key disk_key;
2091 int direction = path->reada;
2092 struct extent_buffer *eb;
2100 if (!path->nodes[level])
2103 node = path->nodes[level];
2105 search = btrfs_node_blockptr(node, slot);
2106 blocksize = btrfs_level_size(root, level - 1);
2107 eb = btrfs_find_tree_block(root, search, blocksize);
2109 free_extent_buffer(eb);
2115 nritems = btrfs_header_nritems(node);
2119 if (direction < 0) {
2123 } else if (direction > 0) {
2128 if (path->reada < 0 && objectid) {
2129 btrfs_node_key(node, &disk_key, nr);
2130 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2133 search = btrfs_node_blockptr(node, nr);
2134 if ((search <= target && target - search <= 65536) ||
2135 (search > target && search - target <= 65536)) {
2136 gen = btrfs_node_ptr_generation(node, nr);
2137 readahead_tree_block(root, search, blocksize, gen);
2141 if ((nread > 65536 || nscan > 32))
2146 static noinline void reada_for_balance(struct btrfs_root *root,
2147 struct btrfs_path *path, int level)
2151 struct extent_buffer *parent;
2152 struct extent_buffer *eb;
2158 parent = path->nodes[level + 1];
2162 nritems = btrfs_header_nritems(parent);
2163 slot = path->slots[level + 1];
2164 blocksize = btrfs_level_size(root, level);
2167 block1 = btrfs_node_blockptr(parent, slot - 1);
2168 gen = btrfs_node_ptr_generation(parent, slot - 1);
2169 eb = btrfs_find_tree_block(root, block1, blocksize);
2171 * if we get -eagain from btrfs_buffer_uptodate, we
2172 * don't want to return eagain here. That will loop
2175 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2177 free_extent_buffer(eb);
2179 if (slot + 1 < nritems) {
2180 block2 = btrfs_node_blockptr(parent, slot + 1);
2181 gen = btrfs_node_ptr_generation(parent, slot + 1);
2182 eb = btrfs_find_tree_block(root, block2, blocksize);
2183 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2185 free_extent_buffer(eb);
2189 readahead_tree_block(root, block1, blocksize, 0);
2191 readahead_tree_block(root, block2, blocksize, 0);
2196 * when we walk down the tree, it is usually safe to unlock the higher layers
2197 * in the tree. The exceptions are when our path goes through slot 0, because
2198 * operations on the tree might require changing key pointers higher up in the
2201 * callers might also have set path->keep_locks, which tells this code to keep
2202 * the lock if the path points to the last slot in the block. This is part of
2203 * walking through the tree, and selecting the next slot in the higher block.
2205 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2206 * if lowest_unlock is 1, level 0 won't be unlocked
2208 static noinline void unlock_up(struct btrfs_path *path, int level,
2209 int lowest_unlock, int min_write_lock_level,
2210 int *write_lock_level)
2213 int skip_level = level;
2215 struct extent_buffer *t;
2217 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2218 if (!path->nodes[i])
2220 if (!path->locks[i])
2222 if (!no_skips && path->slots[i] == 0) {
2226 if (!no_skips && path->keep_locks) {
2229 nritems = btrfs_header_nritems(t);
2230 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2235 if (skip_level < i && i >= lowest_unlock)
2239 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2240 btrfs_tree_unlock_rw(t, path->locks[i]);
2242 if (write_lock_level &&
2243 i > min_write_lock_level &&
2244 i <= *write_lock_level) {
2245 *write_lock_level = i - 1;
2252 * This releases any locks held in the path starting at level and
2253 * going all the way up to the root.
2255 * btrfs_search_slot will keep the lock held on higher nodes in a few
2256 * corner cases, such as COW of the block at slot zero in the node. This
2257 * ignores those rules, and it should only be called when there are no
2258 * more updates to be done higher up in the tree.
2260 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2264 if (path->keep_locks)
2267 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2268 if (!path->nodes[i])
2270 if (!path->locks[i])
2272 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2278 * helper function for btrfs_search_slot. The goal is to find a block
2279 * in cache without setting the path to blocking. If we find the block
2280 * we return zero and the path is unchanged.
2282 * If we can't find the block, we set the path blocking and do some
2283 * reada. -EAGAIN is returned and the search must be repeated.
2286 read_block_for_search(struct btrfs_trans_handle *trans,
2287 struct btrfs_root *root, struct btrfs_path *p,
2288 struct extent_buffer **eb_ret, int level, int slot,
2289 struct btrfs_key *key, u64 time_seq)
2294 struct extent_buffer *b = *eb_ret;
2295 struct extent_buffer *tmp;
2298 blocknr = btrfs_node_blockptr(b, slot);
2299 gen = btrfs_node_ptr_generation(b, slot);
2300 blocksize = btrfs_level_size(root, level - 1);
2302 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2304 /* first we do an atomic uptodate check */
2305 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2310 /* the pages were up to date, but we failed
2311 * the generation number check. Do a full
2312 * read for the generation number that is correct.
2313 * We must do this without dropping locks so
2314 * we can trust our generation number
2316 btrfs_set_path_blocking(p);
2318 /* now we're allowed to do a blocking uptodate check */
2319 ret = btrfs_read_buffer(tmp, gen);
2324 free_extent_buffer(tmp);
2325 btrfs_release_path(p);
2330 * reduce lock contention at high levels
2331 * of the btree by dropping locks before
2332 * we read. Don't release the lock on the current
2333 * level because we need to walk this node to figure
2334 * out which blocks to read.
2336 btrfs_unlock_up_safe(p, level + 1);
2337 btrfs_set_path_blocking(p);
2339 free_extent_buffer(tmp);
2341 reada_for_search(root, p, level, slot, key->objectid);
2343 btrfs_release_path(p);
2346 tmp = read_tree_block(root, blocknr, blocksize, 0);
2349 * If the read above didn't mark this buffer up to date,
2350 * it will never end up being up to date. Set ret to EIO now
2351 * and give up so that our caller doesn't loop forever
2354 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2356 free_extent_buffer(tmp);
2362 * helper function for btrfs_search_slot. This does all of the checks
2363 * for node-level blocks and does any balancing required based on
2366 * If no extra work was required, zero is returned. If we had to
2367 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2371 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2372 struct btrfs_root *root, struct btrfs_path *p,
2373 struct extent_buffer *b, int level, int ins_len,
2374 int *write_lock_level)
2377 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2378 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2381 if (*write_lock_level < level + 1) {
2382 *write_lock_level = level + 1;
2383 btrfs_release_path(p);
2387 btrfs_set_path_blocking(p);
2388 reada_for_balance(root, p, level);
2389 sret = split_node(trans, root, p, level);
2390 btrfs_clear_path_blocking(p, NULL, 0);
2397 b = p->nodes[level];
2398 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2399 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2402 if (*write_lock_level < level + 1) {
2403 *write_lock_level = level + 1;
2404 btrfs_release_path(p);
2408 btrfs_set_path_blocking(p);
2409 reada_for_balance(root, p, level);
2410 sret = balance_level(trans, root, p, level);
2411 btrfs_clear_path_blocking(p, NULL, 0);
2417 b = p->nodes[level];
2419 btrfs_release_path(p);
2422 BUG_ON(btrfs_header_nritems(b) == 1);
2432 static void key_search_validate(struct extent_buffer *b,
2433 struct btrfs_key *key,
2436 #ifdef CONFIG_BTRFS_ASSERT
2437 struct btrfs_disk_key disk_key;
2439 btrfs_cpu_key_to_disk(&disk_key, key);
2442 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2443 offsetof(struct btrfs_leaf, items[0].key),
2446 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2447 offsetof(struct btrfs_node, ptrs[0].key),
2452 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2453 int level, int *prev_cmp, int *slot)
2455 if (*prev_cmp != 0) {
2456 *prev_cmp = bin_search(b, key, level, slot);
2460 key_search_validate(b, key, level);
2467 * look for key in the tree. path is filled in with nodes along the way
2468 * if key is found, we return zero and you can find the item in the leaf
2469 * level of the path (level 0)
2471 * If the key isn't found, the path points to the slot where it should
2472 * be inserted, and 1 is returned. If there are other errors during the
2473 * search a negative error number is returned.
2475 * if ins_len > 0, nodes and leaves will be split as we walk down the
2476 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2479 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2480 *root, struct btrfs_key *key, struct btrfs_path *p, int
2483 struct extent_buffer *b;
2488 int lowest_unlock = 1;
2490 /* everything at write_lock_level or lower must be write locked */
2491 int write_lock_level = 0;
2492 u8 lowest_level = 0;
2493 int min_write_lock_level;
2496 lowest_level = p->lowest_level;
2497 WARN_ON(lowest_level && ins_len > 0);
2498 WARN_ON(p->nodes[0] != NULL);
2503 /* when we are removing items, we might have to go up to level
2504 * two as we update tree pointers Make sure we keep write
2505 * for those levels as well
2507 write_lock_level = 2;
2508 } else if (ins_len > 0) {
2510 * for inserting items, make sure we have a write lock on
2511 * level 1 so we can update keys
2513 write_lock_level = 1;
2517 write_lock_level = -1;
2519 if (cow && (p->keep_locks || p->lowest_level))
2520 write_lock_level = BTRFS_MAX_LEVEL;
2522 min_write_lock_level = write_lock_level;
2527 * we try very hard to do read locks on the root
2529 root_lock = BTRFS_READ_LOCK;
2531 if (p->search_commit_root) {
2533 * the commit roots are read only
2534 * so we always do read locks
2536 b = root->commit_root;
2537 extent_buffer_get(b);
2538 level = btrfs_header_level(b);
2539 if (!p->skip_locking)
2540 btrfs_tree_read_lock(b);
2542 if (p->skip_locking) {
2543 b = btrfs_root_node(root);
2544 level = btrfs_header_level(b);
2546 /* we don't know the level of the root node
2547 * until we actually have it read locked
2549 b = btrfs_read_lock_root_node(root);
2550 level = btrfs_header_level(b);
2551 if (level <= write_lock_level) {
2552 /* whoops, must trade for write lock */
2553 btrfs_tree_read_unlock(b);
2554 free_extent_buffer(b);
2555 b = btrfs_lock_root_node(root);
2556 root_lock = BTRFS_WRITE_LOCK;
2558 /* the level might have changed, check again */
2559 level = btrfs_header_level(b);
2563 p->nodes[level] = b;
2564 if (!p->skip_locking)
2565 p->locks[level] = root_lock;
2568 level = btrfs_header_level(b);
2571 * setup the path here so we can release it under lock
2572 * contention with the cow code
2576 * if we don't really need to cow this block
2577 * then we don't want to set the path blocking,
2578 * so we test it here
2580 if (!should_cow_block(trans, root, b))
2583 btrfs_set_path_blocking(p);
2586 * must have write locks on this node and the
2589 if (level > write_lock_level ||
2590 (level + 1 > write_lock_level &&
2591 level + 1 < BTRFS_MAX_LEVEL &&
2592 p->nodes[level + 1])) {
2593 write_lock_level = level + 1;
2594 btrfs_release_path(p);
2598 err = btrfs_cow_block(trans, root, b,
2599 p->nodes[level + 1],
2600 p->slots[level + 1], &b);
2607 BUG_ON(!cow && ins_len);
2609 p->nodes[level] = b;
2610 btrfs_clear_path_blocking(p, NULL, 0);
2613 * we have a lock on b and as long as we aren't changing
2614 * the tree, there is no way to for the items in b to change.
2615 * It is safe to drop the lock on our parent before we
2616 * go through the expensive btree search on b.
2618 * If cow is true, then we might be changing slot zero,
2619 * which may require changing the parent. So, we can't
2620 * drop the lock until after we know which slot we're
2624 btrfs_unlock_up_safe(p, level + 1);
2626 ret = key_search(b, key, level, &prev_cmp, &slot);
2630 if (ret && slot > 0) {
2634 p->slots[level] = slot;
2635 err = setup_nodes_for_search(trans, root, p, b, level,
2636 ins_len, &write_lock_level);
2643 b = p->nodes[level];
2644 slot = p->slots[level];
2647 * slot 0 is special, if we change the key
2648 * we have to update the parent pointer
2649 * which means we must have a write lock
2652 if (slot == 0 && cow &&
2653 write_lock_level < level + 1) {
2654 write_lock_level = level + 1;
2655 btrfs_release_path(p);
2659 unlock_up(p, level, lowest_unlock,
2660 min_write_lock_level, &write_lock_level);
2662 if (level == lowest_level) {
2668 err = read_block_for_search(trans, root, p,
2669 &b, level, slot, key, 0);
2677 if (!p->skip_locking) {
2678 level = btrfs_header_level(b);
2679 if (level <= write_lock_level) {
2680 err = btrfs_try_tree_write_lock(b);
2682 btrfs_set_path_blocking(p);
2684 btrfs_clear_path_blocking(p, b,
2687 p->locks[level] = BTRFS_WRITE_LOCK;
2689 err = btrfs_try_tree_read_lock(b);
2691 btrfs_set_path_blocking(p);
2692 btrfs_tree_read_lock(b);
2693 btrfs_clear_path_blocking(p, b,
2696 p->locks[level] = BTRFS_READ_LOCK;
2698 p->nodes[level] = b;
2701 p->slots[level] = slot;
2703 btrfs_leaf_free_space(root, b) < ins_len) {
2704 if (write_lock_level < 1) {
2705 write_lock_level = 1;
2706 btrfs_release_path(p);
2710 btrfs_set_path_blocking(p);
2711 err = split_leaf(trans, root, key,
2712 p, ins_len, ret == 0);
2713 btrfs_clear_path_blocking(p, NULL, 0);
2721 if (!p->search_for_split)
2722 unlock_up(p, level, lowest_unlock,
2723 min_write_lock_level, &write_lock_level);
2730 * we don't really know what they plan on doing with the path
2731 * from here on, so for now just mark it as blocking
2733 if (!p->leave_spinning)
2734 btrfs_set_path_blocking(p);
2736 btrfs_release_path(p);
2741 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2742 * current state of the tree together with the operations recorded in the tree
2743 * modification log to search for the key in a previous version of this tree, as
2744 * denoted by the time_seq parameter.
2746 * Naturally, there is no support for insert, delete or cow operations.
2748 * The resulting path and return value will be set up as if we called
2749 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2751 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2752 struct btrfs_path *p, u64 time_seq)
2754 struct extent_buffer *b;
2759 int lowest_unlock = 1;
2760 u8 lowest_level = 0;
2763 lowest_level = p->lowest_level;
2764 WARN_ON(p->nodes[0] != NULL);
2766 if (p->search_commit_root) {
2768 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2772 b = get_old_root(root, time_seq);
2773 level = btrfs_header_level(b);
2774 p->locks[level] = BTRFS_READ_LOCK;
2777 level = btrfs_header_level(b);
2778 p->nodes[level] = b;
2779 btrfs_clear_path_blocking(p, NULL, 0);
2782 * we have a lock on b and as long as we aren't changing
2783 * the tree, there is no way to for the items in b to change.
2784 * It is safe to drop the lock on our parent before we
2785 * go through the expensive btree search on b.
2787 btrfs_unlock_up_safe(p, level + 1);
2790 * Since we can unwind eb's we want to do a real search every
2794 ret = key_search(b, key, level, &prev_cmp, &slot);
2798 if (ret && slot > 0) {
2802 p->slots[level] = slot;
2803 unlock_up(p, level, lowest_unlock, 0, NULL);
2805 if (level == lowest_level) {
2811 err = read_block_for_search(NULL, root, p, &b, level,
2812 slot, key, time_seq);
2820 level = btrfs_header_level(b);
2821 err = btrfs_try_tree_read_lock(b);
2823 btrfs_set_path_blocking(p);
2824 btrfs_tree_read_lock(b);
2825 btrfs_clear_path_blocking(p, b,
2828 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
2833 p->locks[level] = BTRFS_READ_LOCK;
2834 p->nodes[level] = b;
2836 p->slots[level] = slot;
2837 unlock_up(p, level, lowest_unlock, 0, NULL);
2843 if (!p->leave_spinning)
2844 btrfs_set_path_blocking(p);
2846 btrfs_release_path(p);
2852 * helper to use instead of search slot if no exact match is needed but
2853 * instead the next or previous item should be returned.
2854 * When find_higher is true, the next higher item is returned, the next lower
2856 * When return_any and find_higher are both true, and no higher item is found,
2857 * return the next lower instead.
2858 * When return_any is true and find_higher is false, and no lower item is found,
2859 * return the next higher instead.
2860 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2863 int btrfs_search_slot_for_read(struct btrfs_root *root,
2864 struct btrfs_key *key, struct btrfs_path *p,
2865 int find_higher, int return_any)
2868 struct extent_buffer *leaf;
2871 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2875 * a return value of 1 means the path is at the position where the
2876 * item should be inserted. Normally this is the next bigger item,
2877 * but in case the previous item is the last in a leaf, path points
2878 * to the first free slot in the previous leaf, i.e. at an invalid
2884 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2885 ret = btrfs_next_leaf(root, p);
2891 * no higher item found, return the next
2896 btrfs_release_path(p);
2900 if (p->slots[0] == 0) {
2901 ret = btrfs_prev_leaf(root, p);
2905 p->slots[0] = btrfs_header_nritems(leaf) - 1;
2911 * no lower item found, return the next
2916 btrfs_release_path(p);
2926 * adjust the pointers going up the tree, starting at level
2927 * making sure the right key of each node is points to 'key'.
2928 * This is used after shifting pointers to the left, so it stops
2929 * fixing up pointers when a given leaf/node is not in slot 0 of the
2933 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
2934 struct btrfs_disk_key *key, int level)
2937 struct extent_buffer *t;
2939 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2940 int tslot = path->slots[i];
2941 if (!path->nodes[i])
2944 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
2945 btrfs_set_node_key(t, key, tslot);
2946 btrfs_mark_buffer_dirty(path->nodes[i]);
2955 * This function isn't completely safe. It's the caller's responsibility
2956 * that the new key won't break the order
2958 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
2959 struct btrfs_key *new_key)
2961 struct btrfs_disk_key disk_key;
2962 struct extent_buffer *eb;
2965 eb = path->nodes[0];
2966 slot = path->slots[0];
2968 btrfs_item_key(eb, &disk_key, slot - 1);
2969 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
2971 if (slot < btrfs_header_nritems(eb) - 1) {
2972 btrfs_item_key(eb, &disk_key, slot + 1);
2973 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
2976 btrfs_cpu_key_to_disk(&disk_key, new_key);
2977 btrfs_set_item_key(eb, &disk_key, slot);
2978 btrfs_mark_buffer_dirty(eb);
2980 fixup_low_keys(root, path, &disk_key, 1);
2984 * try to push data from one node into the next node left in the
2987 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2988 * error, and > 0 if there was no room in the left hand block.
2990 static int push_node_left(struct btrfs_trans_handle *trans,
2991 struct btrfs_root *root, struct extent_buffer *dst,
2992 struct extent_buffer *src, int empty)
2999 src_nritems = btrfs_header_nritems(src);
3000 dst_nritems = btrfs_header_nritems(dst);
3001 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3002 WARN_ON(btrfs_header_generation(src) != trans->transid);
3003 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3005 if (!empty && src_nritems <= 8)
3008 if (push_items <= 0)
3012 push_items = min(src_nritems, push_items);
3013 if (push_items < src_nritems) {
3014 /* leave at least 8 pointers in the node if
3015 * we aren't going to empty it
3017 if (src_nritems - push_items < 8) {
3018 if (push_items <= 8)
3024 push_items = min(src_nritems - 8, push_items);
3026 tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3028 copy_extent_buffer(dst, src,
3029 btrfs_node_key_ptr_offset(dst_nritems),
3030 btrfs_node_key_ptr_offset(0),
3031 push_items * sizeof(struct btrfs_key_ptr));
3033 if (push_items < src_nritems) {
3035 * don't call tree_mod_log_eb_move here, key removal was already
3036 * fully logged by tree_mod_log_eb_copy above.
3038 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3039 btrfs_node_key_ptr_offset(push_items),
3040 (src_nritems - push_items) *
3041 sizeof(struct btrfs_key_ptr));
3043 btrfs_set_header_nritems(src, src_nritems - push_items);
3044 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3045 btrfs_mark_buffer_dirty(src);
3046 btrfs_mark_buffer_dirty(dst);
3052 * try to push data from one node into the next node right in the
3055 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3056 * error, and > 0 if there was no room in the right hand block.
3058 * this will only push up to 1/2 the contents of the left node over
3060 static int balance_node_right(struct btrfs_trans_handle *trans,
3061 struct btrfs_root *root,
3062 struct extent_buffer *dst,
3063 struct extent_buffer *src)
3071 WARN_ON(btrfs_header_generation(src) != trans->transid);
3072 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3074 src_nritems = btrfs_header_nritems(src);
3075 dst_nritems = btrfs_header_nritems(dst);
3076 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3077 if (push_items <= 0)
3080 if (src_nritems < 4)
3083 max_push = src_nritems / 2 + 1;
3084 /* don't try to empty the node */
3085 if (max_push >= src_nritems)
3088 if (max_push < push_items)
3089 push_items = max_push;
3091 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3092 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3093 btrfs_node_key_ptr_offset(0),
3095 sizeof(struct btrfs_key_ptr));
3097 tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3098 src_nritems - push_items, push_items);
3099 copy_extent_buffer(dst, src,
3100 btrfs_node_key_ptr_offset(0),
3101 btrfs_node_key_ptr_offset(src_nritems - push_items),
3102 push_items * sizeof(struct btrfs_key_ptr));
3104 btrfs_set_header_nritems(src, src_nritems - push_items);
3105 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3107 btrfs_mark_buffer_dirty(src);
3108 btrfs_mark_buffer_dirty(dst);
3114 * helper function to insert a new root level in the tree.
3115 * A new node is allocated, and a single item is inserted to
3116 * point to the existing root
3118 * returns zero on success or < 0 on failure.
3120 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3121 struct btrfs_root *root,
3122 struct btrfs_path *path, int level)
3125 struct extent_buffer *lower;
3126 struct extent_buffer *c;
3127 struct extent_buffer *old;
3128 struct btrfs_disk_key lower_key;
3130 BUG_ON(path->nodes[level]);
3131 BUG_ON(path->nodes[level-1] != root->node);
3133 lower = path->nodes[level-1];
3135 btrfs_item_key(lower, &lower_key, 0);
3137 btrfs_node_key(lower, &lower_key, 0);
3139 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3140 root->root_key.objectid, &lower_key,
3141 level, root->node->start, 0);
3145 root_add_used(root, root->nodesize);
3147 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3148 btrfs_set_header_nritems(c, 1);
3149 btrfs_set_header_level(c, level);
3150 btrfs_set_header_bytenr(c, c->start);
3151 btrfs_set_header_generation(c, trans->transid);
3152 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3153 btrfs_set_header_owner(c, root->root_key.objectid);
3155 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(c),
3158 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3159 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3161 btrfs_set_node_key(c, &lower_key, 0);
3162 btrfs_set_node_blockptr(c, 0, lower->start);
3163 lower_gen = btrfs_header_generation(lower);
3164 WARN_ON(lower_gen != trans->transid);
3166 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3168 btrfs_mark_buffer_dirty(c);
3171 tree_mod_log_set_root_pointer(root, c, 0);
3172 rcu_assign_pointer(root->node, c);
3174 /* the super has an extra ref to root->node */
3175 free_extent_buffer(old);
3177 add_root_to_dirty_list(root);
3178 extent_buffer_get(c);
3179 path->nodes[level] = c;
3180 path->locks[level] = BTRFS_WRITE_LOCK;
3181 path->slots[level] = 0;
3186 * worker function to insert a single pointer in a node.
3187 * the node should have enough room for the pointer already
3189 * slot and level indicate where you want the key to go, and
3190 * blocknr is the block the key points to.
3192 static void insert_ptr(struct btrfs_trans_handle *trans,
3193 struct btrfs_root *root, struct btrfs_path *path,
3194 struct btrfs_disk_key *key, u64 bytenr,
3195 int slot, int level)
3197 struct extent_buffer *lower;
3201 BUG_ON(!path->nodes[level]);
3202 btrfs_assert_tree_locked(path->nodes[level]);
3203 lower = path->nodes[level];
3204 nritems = btrfs_header_nritems(lower);
3205 BUG_ON(slot > nritems);
3206 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3207 if (slot != nritems) {
3209 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3210 slot, nritems - slot);
3211 memmove_extent_buffer(lower,
3212 btrfs_node_key_ptr_offset(slot + 1),
3213 btrfs_node_key_ptr_offset(slot),
3214 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3217 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3218 MOD_LOG_KEY_ADD, GFP_NOFS);
3221 btrfs_set_node_key(lower, key, slot);
3222 btrfs_set_node_blockptr(lower, slot, bytenr);
3223 WARN_ON(trans->transid == 0);
3224 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3225 btrfs_set_header_nritems(lower, nritems + 1);
3226 btrfs_mark_buffer_dirty(lower);
3230 * split the node at the specified level in path in two.
3231 * The path is corrected to point to the appropriate node after the split
3233 * Before splitting this tries to make some room in the node by pushing
3234 * left and right, if either one works, it returns right away.
3236 * returns 0 on success and < 0 on failure
3238 static noinline int split_node(struct btrfs_trans_handle *trans,
3239 struct btrfs_root *root,
3240 struct btrfs_path *path, int level)
3242 struct extent_buffer *c;
3243 struct extent_buffer *split;
3244 struct btrfs_disk_key disk_key;
3249 c = path->nodes[level];
3250 WARN_ON(btrfs_header_generation(c) != trans->transid);
3251 if (c == root->node) {
3253 * trying to split the root, lets make a new one
3255 * tree mod log: We don't log_removal old root in
3256 * insert_new_root, because that root buffer will be kept as a
3257 * normal node. We are going to log removal of half of the
3258 * elements below with tree_mod_log_eb_copy. We're holding a
3259 * tree lock on the buffer, which is why we cannot race with
3260 * other tree_mod_log users.
3262 ret = insert_new_root(trans, root, path, level + 1);
3266 ret = push_nodes_for_insert(trans, root, path, level);
3267 c = path->nodes[level];
3268 if (!ret && btrfs_header_nritems(c) <
3269 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3275 c_nritems = btrfs_header_nritems(c);
3276 mid = (c_nritems + 1) / 2;
3277 btrfs_node_key(c, &disk_key, mid);
3279 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3280 root->root_key.objectid,
3281 &disk_key, level, c->start, 0);
3283 return PTR_ERR(split);
3285 root_add_used(root, root->nodesize);
3287 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3288 btrfs_set_header_level(split, btrfs_header_level(c));
3289 btrfs_set_header_bytenr(split, split->start);
3290 btrfs_set_header_generation(split, trans->transid);
3291 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3292 btrfs_set_header_owner(split, root->root_key.objectid);
3293 write_extent_buffer(split, root->fs_info->fsid,
3294 btrfs_header_fsid(split), BTRFS_FSID_SIZE);
3295 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3296 btrfs_header_chunk_tree_uuid(split),
3299 tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid);
3300 copy_extent_buffer(split, c,
3301 btrfs_node_key_ptr_offset(0),
3302 btrfs_node_key_ptr_offset(mid),
3303 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3304 btrfs_set_header_nritems(split, c_nritems - mid);
3305 btrfs_set_header_nritems(c, mid);
3308 btrfs_mark_buffer_dirty(c);
3309 btrfs_mark_buffer_dirty(split);
3311 insert_ptr(trans, root, path, &disk_key, split->start,
3312 path->slots[level + 1] + 1, level + 1);
3314 if (path->slots[level] >= mid) {
3315 path->slots[level] -= mid;
3316 btrfs_tree_unlock(c);
3317 free_extent_buffer(c);
3318 path->nodes[level] = split;
3319 path->slots[level + 1] += 1;
3321 btrfs_tree_unlock(split);
3322 free_extent_buffer(split);
3328 * how many bytes are required to store the items in a leaf. start
3329 * and nr indicate which items in the leaf to check. This totals up the
3330 * space used both by the item structs and the item data
3332 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3334 struct btrfs_item *start_item;
3335 struct btrfs_item *end_item;
3336 struct btrfs_map_token token;
3338 int nritems = btrfs_header_nritems(l);
3339 int end = min(nritems, start + nr) - 1;
3343 btrfs_init_map_token(&token);
3344 start_item = btrfs_item_nr(start);
3345 end_item = btrfs_item_nr(end);
3346 data_len = btrfs_token_item_offset(l, start_item, &token) +
3347 btrfs_token_item_size(l, start_item, &token);
3348 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3349 data_len += sizeof(struct btrfs_item) * nr;
3350 WARN_ON(data_len < 0);
3355 * The space between the end of the leaf items and
3356 * the start of the leaf data. IOW, how much room
3357 * the leaf has left for both items and data
3359 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3360 struct extent_buffer *leaf)
3362 int nritems = btrfs_header_nritems(leaf);
3364 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3366 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
3367 "used %d nritems %d\n",
3368 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3369 leaf_space_used(leaf, 0, nritems), nritems);
3375 * min slot controls the lowest index we're willing to push to the
3376 * right. We'll push up to and including min_slot, but no lower
3378 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3379 struct btrfs_root *root,
3380 struct btrfs_path *path,
3381 int data_size, int empty,
3382 struct extent_buffer *right,
3383 int free_space, u32 left_nritems,
3386 struct extent_buffer *left = path->nodes[0];
3387 struct extent_buffer *upper = path->nodes[1];
3388 struct btrfs_map_token token;
3389 struct btrfs_disk_key disk_key;
3394 struct btrfs_item *item;
3400 btrfs_init_map_token(&token);
3405 nr = max_t(u32, 1, min_slot);
3407 if (path->slots[0] >= left_nritems)
3408 push_space += data_size;
3410 slot = path->slots[1];
3411 i = left_nritems - 1;
3413 item = btrfs_item_nr(i);
3415 if (!empty && push_items > 0) {
3416 if (path->slots[0] > i)
3418 if (path->slots[0] == i) {
3419 int space = btrfs_leaf_free_space(root, left);
3420 if (space + push_space * 2 > free_space)
3425 if (path->slots[0] == i)
3426 push_space += data_size;
3428 this_item_size = btrfs_item_size(left, item);
3429 if (this_item_size + sizeof(*item) + push_space > free_space)
3433 push_space += this_item_size + sizeof(*item);
3439 if (push_items == 0)
3442 WARN_ON(!empty && push_items == left_nritems);
3444 /* push left to right */
3445 right_nritems = btrfs_header_nritems(right);
3447 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3448 push_space -= leaf_data_end(root, left);
3450 /* make room in the right data area */
3451 data_end = leaf_data_end(root, right);
3452 memmove_extent_buffer(right,
3453 btrfs_leaf_data(right) + data_end - push_space,
3454 btrfs_leaf_data(right) + data_end,
3455 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3457 /* copy from the left data area */
3458 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3459 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3460 btrfs_leaf_data(left) + leaf_data_end(root, left),
3463 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3464 btrfs_item_nr_offset(0),
3465 right_nritems * sizeof(struct btrfs_item));
3467 /* copy the items from left to right */
3468 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3469 btrfs_item_nr_offset(left_nritems - push_items),
3470 push_items * sizeof(struct btrfs_item));
3472 /* update the item pointers */
3473 right_nritems += push_items;
3474 btrfs_set_header_nritems(right, right_nritems);
3475 push_space = BTRFS_LEAF_DATA_SIZE(root);
3476 for (i = 0; i < right_nritems; i++) {
3477 item = btrfs_item_nr(i);
3478 push_space -= btrfs_token_item_size(right, item, &token);
3479 btrfs_set_token_item_offset(right, item, push_space, &token);
3482 left_nritems -= push_items;
3483 btrfs_set_header_nritems(left, left_nritems);
3486 btrfs_mark_buffer_dirty(left);
3488 clean_tree_block(trans, root, left);
3490 btrfs_mark_buffer_dirty(right);
3492 btrfs_item_key(right, &disk_key, 0);
3493 btrfs_set_node_key(upper, &disk_key, slot + 1);
3494 btrfs_mark_buffer_dirty(upper);
3496 /* then fixup the leaf pointer in the path */
3497 if (path->slots[0] >= left_nritems) {
3498 path->slots[0] -= left_nritems;
3499 if (btrfs_header_nritems(path->nodes[0]) == 0)
3500 clean_tree_block(trans, root, path->nodes[0]);
3501 btrfs_tree_unlock(path->nodes[0]);
3502 free_extent_buffer(path->nodes[0]);
3503 path->nodes[0] = right;
3504 path->slots[1] += 1;
3506 btrfs_tree_unlock(right);
3507 free_extent_buffer(right);
3512 btrfs_tree_unlock(right);
3513 free_extent_buffer(right);
3518 * push some data in the path leaf to the right, trying to free up at
3519 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3521 * returns 1 if the push failed because the other node didn't have enough
3522 * room, 0 if everything worked out and < 0 if there were major errors.
3524 * this will push starting from min_slot to the end of the leaf. It won't
3525 * push any slot lower than min_slot
3527 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3528 *root, struct btrfs_path *path,
3529 int min_data_size, int data_size,
3530 int empty, u32 min_slot)
3532 struct extent_buffer *left = path->nodes[0];
3533 struct extent_buffer *right;
3534 struct extent_buffer *upper;
3540 if (!path->nodes[1])
3543 slot = path->slots[1];
3544 upper = path->nodes[1];
3545 if (slot >= btrfs_header_nritems(upper) - 1)
3548 btrfs_assert_tree_locked(path->nodes[1]);
3550 right = read_node_slot(root, upper, slot + 1);
3554 btrfs_tree_lock(right);
3555 btrfs_set_lock_blocking(right);
3557 free_space = btrfs_leaf_free_space(root, right);
3558 if (free_space < data_size)
3561 /* cow and double check */
3562 ret = btrfs_cow_block(trans, root, right, upper,
3567 free_space = btrfs_leaf_free_space(root, right);
3568 if (free_space < data_size)
3571 left_nritems = btrfs_header_nritems(left);
3572 if (left_nritems == 0)
3575 return __push_leaf_right(trans, root, path, min_data_size, empty,
3576 right, free_space, left_nritems, min_slot);
3578 btrfs_tree_unlock(right);
3579 free_extent_buffer(right);
3584 * push some data in the path leaf to the left, trying to free up at
3585 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3587 * max_slot can put a limit on how far into the leaf we'll push items. The
3588 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3591 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3592 struct btrfs_root *root,
3593 struct btrfs_path *path, int data_size,
3594 int empty, struct extent_buffer *left,
3595 int free_space, u32 right_nritems,
3598 struct btrfs_disk_key disk_key;
3599 struct extent_buffer *right = path->nodes[0];
3603 struct btrfs_item *item;
3604 u32 old_left_nritems;
3608 u32 old_left_item_size;
3609 struct btrfs_map_token token;
3611 btrfs_init_map_token(&token);
3614 nr = min(right_nritems, max_slot);
3616 nr = min(right_nritems - 1, max_slot);
3618 for (i = 0; i < nr; i++) {
3619 item = btrfs_item_nr(i);
3621 if (!empty && push_items > 0) {
3622 if (path->slots[0] < i)
3624 if (path->slots[0] == i) {
3625 int space = btrfs_leaf_free_space(root, right);
3626 if (space + push_space * 2 > free_space)
3631 if (path->slots[0] == i)
3632 push_space += data_size;
3634 this_item_size = btrfs_item_size(right, item);
3635 if (this_item_size + sizeof(*item) + push_space > free_space)
3639 push_space += this_item_size + sizeof(*item);
3642 if (push_items == 0) {
3646 if (!empty && push_items == btrfs_header_nritems(right))
3649 /* push data from right to left */
3650 copy_extent_buffer(left, right,
3651 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3652 btrfs_item_nr_offset(0),
3653 push_items * sizeof(struct btrfs_item));
3655 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3656 btrfs_item_offset_nr(right, push_items - 1);
3658 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3659 leaf_data_end(root, left) - push_space,
3660 btrfs_leaf_data(right) +
3661 btrfs_item_offset_nr(right, push_items - 1),
3663 old_left_nritems = btrfs_header_nritems(left);
3664 BUG_ON(old_left_nritems <= 0);
3666 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3667 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3670 item = btrfs_item_nr(i);
3672 ioff = btrfs_token_item_offset(left, item, &token);
3673 btrfs_set_token_item_offset(left, item,
3674 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3677 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3679 /* fixup right node */
3680 if (push_items > right_nritems)
3681 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3684 if (push_items < right_nritems) {
3685 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3686 leaf_data_end(root, right);
3687 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3688 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3689 btrfs_leaf_data(right) +
3690 leaf_data_end(root, right), push_space);
3692 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3693 btrfs_item_nr_offset(push_items),
3694 (btrfs_header_nritems(right) - push_items) *
3695 sizeof(struct btrfs_item));
3697 right_nritems -= push_items;
3698 btrfs_set_header_nritems(right, right_nritems);
3699 push_space = BTRFS_LEAF_DATA_SIZE(root);
3700 for (i = 0; i < right_nritems; i++) {
3701 item = btrfs_item_nr(i);
3703 push_space = push_space - btrfs_token_item_size(right,
3705 btrfs_set_token_item_offset(right, item, push_space, &token);
3708 btrfs_mark_buffer_dirty(left);
3710 btrfs_mark_buffer_dirty(right);
3712 clean_tree_block(trans, root, right);
3714 btrfs_item_key(right, &disk_key, 0);
3715 fixup_low_keys(root, path, &disk_key, 1);
3717 /* then fixup the leaf pointer in the path */
3718 if (path->slots[0] < push_items) {
3719 path->slots[0] += old_left_nritems;
3720 btrfs_tree_unlock(path->nodes[0]);
3721 free_extent_buffer(path->nodes[0]);
3722 path->nodes[0] = left;
3723 path->slots[1] -= 1;
3725 btrfs_tree_unlock(left);
3726 free_extent_buffer(left);
3727 path->slots[0] -= push_items;
3729 BUG_ON(path->slots[0] < 0);
3732 btrfs_tree_unlock(left);
3733 free_extent_buffer(left);
3738 * push some data in the path leaf to the left, trying to free up at
3739 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3741 * max_slot can put a limit on how far into the leaf we'll push items. The
3742 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3745 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3746 *root, struct btrfs_path *path, int min_data_size,
3747 int data_size, int empty, u32 max_slot)
3749 struct extent_buffer *right = path->nodes[0];
3750 struct extent_buffer *left;
3756 slot = path->slots[1];
3759 if (!path->nodes[1])
3762 right_nritems = btrfs_header_nritems(right);
3763 if (right_nritems == 0)
3766 btrfs_assert_tree_locked(path->nodes[1]);
3768 left = read_node_slot(root, path->nodes[1], slot - 1);
3772 btrfs_tree_lock(left);
3773 btrfs_set_lock_blocking(left);
3775 free_space = btrfs_leaf_free_space(root, left);
3776 if (free_space < data_size) {
3781 /* cow and double check */
3782 ret = btrfs_cow_block(trans, root, left,
3783 path->nodes[1], slot - 1, &left);
3785 /* we hit -ENOSPC, but it isn't fatal here */
3791 free_space = btrfs_leaf_free_space(root, left);
3792 if (free_space < data_size) {
3797 return __push_leaf_left(trans, root, path, min_data_size,
3798 empty, left, free_space, right_nritems,
3801 btrfs_tree_unlock(left);
3802 free_extent_buffer(left);
3807 * split the path's leaf in two, making sure there is at least data_size
3808 * available for the resulting leaf level of the path.
3810 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3811 struct btrfs_root *root,
3812 struct btrfs_path *path,
3813 struct extent_buffer *l,
3814 struct extent_buffer *right,
3815 int slot, int mid, int nritems)
3820 struct btrfs_disk_key disk_key;
3821 struct btrfs_map_token token;
3823 btrfs_init_map_token(&token);
3825 nritems = nritems - mid;
3826 btrfs_set_header_nritems(right, nritems);
3827 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
3829 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3830 btrfs_item_nr_offset(mid),
3831 nritems * sizeof(struct btrfs_item));
3833 copy_extent_buffer(right, l,
3834 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
3835 data_copy_size, btrfs_leaf_data(l) +
3836 leaf_data_end(root, l), data_copy_size);
3838 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
3839 btrfs_item_end_nr(l, mid);
3841 for (i = 0; i < nritems; i++) {
3842 struct btrfs_item *item = btrfs_item_nr(i);
3845 ioff = btrfs_token_item_offset(right, item, &token);
3846 btrfs_set_token_item_offset(right, item,
3847 ioff + rt_data_off, &token);
3850 btrfs_set_header_nritems(l, mid);
3851 btrfs_item_key(right, &disk_key, 0);
3852 insert_ptr(trans, root, path, &disk_key, right->start,
3853 path->slots[1] + 1, 1);
3855 btrfs_mark_buffer_dirty(right);
3856 btrfs_mark_buffer_dirty(l);
3857 BUG_ON(path->slots[0] != slot);
3860 btrfs_tree_unlock(path->nodes[0]);
3861 free_extent_buffer(path->nodes[0]);
3862 path->nodes[0] = right;
3863 path->slots[0] -= mid;
3864 path->slots[1] += 1;
3866 btrfs_tree_unlock(right);
3867 free_extent_buffer(right);
3870 BUG_ON(path->slots[0] < 0);
3874 * double splits happen when we need to insert a big item in the middle
3875 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3876 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3879 * We avoid this by trying to push the items on either side of our target
3880 * into the adjacent leaves. If all goes well we can avoid the double split
3883 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3884 struct btrfs_root *root,
3885 struct btrfs_path *path,
3893 slot = path->slots[0];
3896 * try to push all the items after our slot into the
3899 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
3906 nritems = btrfs_header_nritems(path->nodes[0]);
3908 * our goal is to get our slot at the start or end of a leaf. If
3909 * we've done so we're done
3911 if (path->slots[0] == 0 || path->slots[0] == nritems)
3914 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3917 /* try to push all the items before our slot into the next leaf */
3918 slot = path->slots[0];
3919 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
3932 * split the path's leaf in two, making sure there is at least data_size
3933 * available for the resulting leaf level of the path.
3935 * returns 0 if all went well and < 0 on failure.
3937 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3938 struct btrfs_root *root,
3939 struct btrfs_key *ins_key,
3940 struct btrfs_path *path, int data_size,
3943 struct btrfs_disk_key disk_key;
3944 struct extent_buffer *l;
3948 struct extent_buffer *right;
3952 int num_doubles = 0;
3953 int tried_avoid_double = 0;
3956 slot = path->slots[0];
3957 if (extend && data_size + btrfs_item_size_nr(l, slot) +
3958 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
3961 /* first try to make some room by pushing left and right */
3962 if (data_size && path->nodes[1]) {
3963 wret = push_leaf_right(trans, root, path, data_size,
3968 wret = push_leaf_left(trans, root, path, data_size,
3969 data_size, 0, (u32)-1);
3975 /* did the pushes work? */
3976 if (btrfs_leaf_free_space(root, l) >= data_size)
3980 if (!path->nodes[1]) {
3981 ret = insert_new_root(trans, root, path, 1);
3988 slot = path->slots[0];
3989 nritems = btrfs_header_nritems(l);
3990 mid = (nritems + 1) / 2;
3994 leaf_space_used(l, mid, nritems - mid) + data_size >
3995 BTRFS_LEAF_DATA_SIZE(root)) {
3996 if (slot >= nritems) {
4000 if (mid != nritems &&
4001 leaf_space_used(l, mid, nritems - mid) +
4002 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4003 if (data_size && !tried_avoid_double)
4004 goto push_for_double;
4010 if (leaf_space_used(l, 0, mid) + data_size >
4011 BTRFS_LEAF_DATA_SIZE(root)) {
4012 if (!extend && data_size && slot == 0) {
4014 } else if ((extend || !data_size) && slot == 0) {
4018 if (mid != nritems &&
4019 leaf_space_used(l, mid, nritems - mid) +
4020 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4021 if (data_size && !tried_avoid_double)
4022 goto push_for_double;
4030 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4032 btrfs_item_key(l, &disk_key, mid);
4034 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
4035 root->root_key.objectid,
4036 &disk_key, 0, l->start, 0);
4038 return PTR_ERR(right);
4040 root_add_used(root, root->leafsize);
4042 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4043 btrfs_set_header_bytenr(right, right->start);
4044 btrfs_set_header_generation(right, trans->transid);
4045 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4046 btrfs_set_header_owner(right, root->root_key.objectid);
4047 btrfs_set_header_level(right, 0);
4048 write_extent_buffer(right, root->fs_info->fsid,
4049 btrfs_header_fsid(right), BTRFS_FSID_SIZE);
4051 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4052 btrfs_header_chunk_tree_uuid(right),
4057 btrfs_set_header_nritems(right, 0);
4058 insert_ptr(trans, root, path, &disk_key, right->start,
4059 path->slots[1] + 1, 1);
4060 btrfs_tree_unlock(path->nodes[0]);
4061 free_extent_buffer(path->nodes[0]);
4062 path->nodes[0] = right;
4064 path->slots[1] += 1;
4066 btrfs_set_header_nritems(right, 0);
4067 insert_ptr(trans, root, path, &disk_key, right->start,
4069 btrfs_tree_unlock(path->nodes[0]);
4070 free_extent_buffer(path->nodes[0]);
4071 path->nodes[0] = right;
4073 if (path->slots[1] == 0)
4074 fixup_low_keys(root, path, &disk_key, 1);
4076 btrfs_mark_buffer_dirty(right);
4080 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4083 BUG_ON(num_doubles != 0);
4091 push_for_double_split(trans, root, path, data_size);
4092 tried_avoid_double = 1;
4093 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4098 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4099 struct btrfs_root *root,
4100 struct btrfs_path *path, int ins_len)
4102 struct btrfs_key key;
4103 struct extent_buffer *leaf;
4104 struct btrfs_file_extent_item *fi;
4109 leaf = path->nodes[0];
4110 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4112 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4113 key.type != BTRFS_EXTENT_CSUM_KEY);
4115 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4118 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4119 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4120 fi = btrfs_item_ptr(leaf, path->slots[0],
4121 struct btrfs_file_extent_item);
4122 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4124 btrfs_release_path(path);
4126 path->keep_locks = 1;
4127 path->search_for_split = 1;
4128 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4129 path->search_for_split = 0;
4134 leaf = path->nodes[0];
4135 /* if our item isn't there or got smaller, return now */
4136 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4139 /* the leaf has changed, it now has room. return now */
4140 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4143 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4144 fi = btrfs_item_ptr(leaf, path->slots[0],
4145 struct btrfs_file_extent_item);
4146 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4150 btrfs_set_path_blocking(path);
4151 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4155 path->keep_locks = 0;
4156 btrfs_unlock_up_safe(path, 1);
4159 path->keep_locks = 0;
4163 static noinline int split_item(struct btrfs_trans_handle *trans,
4164 struct btrfs_root *root,
4165 struct btrfs_path *path,
4166 struct btrfs_key *new_key,
4167 unsigned long split_offset)
4169 struct extent_buffer *leaf;
4170 struct btrfs_item *item;
4171 struct btrfs_item *new_item;
4177 struct btrfs_disk_key disk_key;
4179 leaf = path->nodes[0];
4180 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4182 btrfs_set_path_blocking(path);
4184 item = btrfs_item_nr(path->slots[0]);
4185 orig_offset = btrfs_item_offset(leaf, item);
4186 item_size = btrfs_item_size(leaf, item);
4188 buf = kmalloc(item_size, GFP_NOFS);
4192 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4193 path->slots[0]), item_size);
4195 slot = path->slots[0] + 1;
4196 nritems = btrfs_header_nritems(leaf);
4197 if (slot != nritems) {
4198 /* shift the items */
4199 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4200 btrfs_item_nr_offset(slot),
4201 (nritems - slot) * sizeof(struct btrfs_item));
4204 btrfs_cpu_key_to_disk(&disk_key, new_key);
4205 btrfs_set_item_key(leaf, &disk_key, slot);
4207 new_item = btrfs_item_nr(slot);
4209 btrfs_set_item_offset(leaf, new_item, orig_offset);
4210 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4212 btrfs_set_item_offset(leaf, item,
4213 orig_offset + item_size - split_offset);
4214 btrfs_set_item_size(leaf, item, split_offset);
4216 btrfs_set_header_nritems(leaf, nritems + 1);
4218 /* write the data for the start of the original item */
4219 write_extent_buffer(leaf, buf,
4220 btrfs_item_ptr_offset(leaf, path->slots[0]),
4223 /* write the data for the new item */
4224 write_extent_buffer(leaf, buf + split_offset,
4225 btrfs_item_ptr_offset(leaf, slot),
4226 item_size - split_offset);
4227 btrfs_mark_buffer_dirty(leaf);
4229 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4235 * This function splits a single item into two items,
4236 * giving 'new_key' to the new item and splitting the
4237 * old one at split_offset (from the start of the item).
4239 * The path may be released by this operation. After
4240 * the split, the path is pointing to the old item. The
4241 * new item is going to be in the same node as the old one.
4243 * Note, the item being split must be smaller enough to live alone on
4244 * a tree block with room for one extra struct btrfs_item
4246 * This allows us to split the item in place, keeping a lock on the
4247 * leaf the entire time.
4249 int btrfs_split_item(struct btrfs_trans_handle *trans,
4250 struct btrfs_root *root,
4251 struct btrfs_path *path,
4252 struct btrfs_key *new_key,
4253 unsigned long split_offset)
4256 ret = setup_leaf_for_split(trans, root, path,
4257 sizeof(struct btrfs_item));
4261 ret = split_item(trans, root, path, new_key, split_offset);
4266 * This function duplicate a item, giving 'new_key' to the new item.
4267 * It guarantees both items live in the same tree leaf and the new item
4268 * is contiguous with the original item.
4270 * This allows us to split file extent in place, keeping a lock on the
4271 * leaf the entire time.
4273 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4274 struct btrfs_root *root,
4275 struct btrfs_path *path,
4276 struct btrfs_key *new_key)
4278 struct extent_buffer *leaf;
4282 leaf = path->nodes[0];
4283 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4284 ret = setup_leaf_for_split(trans, root, path,
4285 item_size + sizeof(struct btrfs_item));
4290 setup_items_for_insert(root, path, new_key, &item_size,
4291 item_size, item_size +
4292 sizeof(struct btrfs_item), 1);
4293 leaf = path->nodes[0];
4294 memcpy_extent_buffer(leaf,
4295 btrfs_item_ptr_offset(leaf, path->slots[0]),
4296 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4302 * make the item pointed to by the path smaller. new_size indicates
4303 * how small to make it, and from_end tells us if we just chop bytes
4304 * off the end of the item or if we shift the item to chop bytes off
4307 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4308 u32 new_size, int from_end)
4311 struct extent_buffer *leaf;
4312 struct btrfs_item *item;
4314 unsigned int data_end;
4315 unsigned int old_data_start;
4316 unsigned int old_size;
4317 unsigned int size_diff;
4319 struct btrfs_map_token token;
4321 btrfs_init_map_token(&token);
4323 leaf = path->nodes[0];
4324 slot = path->slots[0];
4326 old_size = btrfs_item_size_nr(leaf, slot);
4327 if (old_size == new_size)
4330 nritems = btrfs_header_nritems(leaf);
4331 data_end = leaf_data_end(root, leaf);
4333 old_data_start = btrfs_item_offset_nr(leaf, slot);
4335 size_diff = old_size - new_size;
4338 BUG_ON(slot >= nritems);
4341 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4343 /* first correct the data pointers */
4344 for (i = slot; i < nritems; i++) {
4346 item = btrfs_item_nr(i);
4348 ioff = btrfs_token_item_offset(leaf, item, &token);
4349 btrfs_set_token_item_offset(leaf, item,
4350 ioff + size_diff, &token);
4353 /* shift the data */
4355 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4356 data_end + size_diff, btrfs_leaf_data(leaf) +
4357 data_end, old_data_start + new_size - data_end);
4359 struct btrfs_disk_key disk_key;
4362 btrfs_item_key(leaf, &disk_key, slot);
4364 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4366 struct btrfs_file_extent_item *fi;
4368 fi = btrfs_item_ptr(leaf, slot,
4369 struct btrfs_file_extent_item);
4370 fi = (struct btrfs_file_extent_item *)(
4371 (unsigned long)fi - size_diff);
4373 if (btrfs_file_extent_type(leaf, fi) ==
4374 BTRFS_FILE_EXTENT_INLINE) {
4375 ptr = btrfs_item_ptr_offset(leaf, slot);
4376 memmove_extent_buffer(leaf, ptr,
4378 offsetof(struct btrfs_file_extent_item,
4383 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4384 data_end + size_diff, btrfs_leaf_data(leaf) +
4385 data_end, old_data_start - data_end);
4387 offset = btrfs_disk_key_offset(&disk_key);
4388 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4389 btrfs_set_item_key(leaf, &disk_key, slot);
4391 fixup_low_keys(root, path, &disk_key, 1);
4394 item = btrfs_item_nr(slot);
4395 btrfs_set_item_size(leaf, item, new_size);
4396 btrfs_mark_buffer_dirty(leaf);
4398 if (btrfs_leaf_free_space(root, leaf) < 0) {
4399 btrfs_print_leaf(root, leaf);
4405 * make the item pointed to by the path bigger, data_size is the added size.
4407 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4411 struct extent_buffer *leaf;
4412 struct btrfs_item *item;
4414 unsigned int data_end;
4415 unsigned int old_data;
4416 unsigned int old_size;
4418 struct btrfs_map_token token;
4420 btrfs_init_map_token(&token);
4422 leaf = path->nodes[0];
4424 nritems = btrfs_header_nritems(leaf);
4425 data_end = leaf_data_end(root, leaf);
4427 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4428 btrfs_print_leaf(root, leaf);
4431 slot = path->slots[0];
4432 old_data = btrfs_item_end_nr(leaf, slot);
4435 if (slot >= nritems) {
4436 btrfs_print_leaf(root, leaf);
4437 printk(KERN_CRIT "slot %d too large, nritems %d\n",
4443 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4445 /* first correct the data pointers */
4446 for (i = slot; i < nritems; i++) {
4448 item = btrfs_item_nr(i);
4450 ioff = btrfs_token_item_offset(leaf, item, &token);
4451 btrfs_set_token_item_offset(leaf, item,
4452 ioff - data_size, &token);
4455 /* shift the data */
4456 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4457 data_end - data_size, btrfs_leaf_data(leaf) +
4458 data_end, old_data - data_end);
4460 data_end = old_data;
4461 old_size = btrfs_item_size_nr(leaf, slot);
4462 item = btrfs_item_nr(slot);
4463 btrfs_set_item_size(leaf, item, old_size + data_size);
4464 btrfs_mark_buffer_dirty(leaf);
4466 if (btrfs_leaf_free_space(root, leaf) < 0) {
4467 btrfs_print_leaf(root, leaf);
4473 * this is a helper for btrfs_insert_empty_items, the main goal here is
4474 * to save stack depth by doing the bulk of the work in a function
4475 * that doesn't call btrfs_search_slot
4477 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4478 struct btrfs_key *cpu_key, u32 *data_size,
4479 u32 total_data, u32 total_size, int nr)
4481 struct btrfs_item *item;
4484 unsigned int data_end;
4485 struct btrfs_disk_key disk_key;
4486 struct extent_buffer *leaf;
4488 struct btrfs_map_token token;
4490 btrfs_init_map_token(&token);
4492 leaf = path->nodes[0];
4493 slot = path->slots[0];
4495 nritems = btrfs_header_nritems(leaf);
4496 data_end = leaf_data_end(root, leaf);
4498 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4499 btrfs_print_leaf(root, leaf);
4500 printk(KERN_CRIT "not enough freespace need %u have %d\n",
4501 total_size, btrfs_leaf_free_space(root, leaf));
4505 if (slot != nritems) {
4506 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4508 if (old_data < data_end) {
4509 btrfs_print_leaf(root, leaf);
4510 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
4511 slot, old_data, data_end);
4515 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4517 /* first correct the data pointers */
4518 for (i = slot; i < nritems; i++) {
4521 item = btrfs_item_nr( i);
4522 ioff = btrfs_token_item_offset(leaf, item, &token);
4523 btrfs_set_token_item_offset(leaf, item,
4524 ioff - total_data, &token);
4526 /* shift the items */
4527 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4528 btrfs_item_nr_offset(slot),
4529 (nritems - slot) * sizeof(struct btrfs_item));
4531 /* shift the data */
4532 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4533 data_end - total_data, btrfs_leaf_data(leaf) +
4534 data_end, old_data - data_end);
4535 data_end = old_data;
4538 /* setup the item for the new data */
4539 for (i = 0; i < nr; i++) {
4540 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4541 btrfs_set_item_key(leaf, &disk_key, slot + i);
4542 item = btrfs_item_nr(slot + i);
4543 btrfs_set_token_item_offset(leaf, item,
4544 data_end - data_size[i], &token);
4545 data_end -= data_size[i];
4546 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4549 btrfs_set_header_nritems(leaf, nritems + nr);
4552 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4553 fixup_low_keys(root, path, &disk_key, 1);
4555 btrfs_unlock_up_safe(path, 1);
4556 btrfs_mark_buffer_dirty(leaf);
4558 if (btrfs_leaf_free_space(root, leaf) < 0) {
4559 btrfs_print_leaf(root, leaf);
4565 * Given a key and some data, insert items into the tree.
4566 * This does all the path init required, making room in the tree if needed.
4568 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4569 struct btrfs_root *root,
4570 struct btrfs_path *path,
4571 struct btrfs_key *cpu_key, u32 *data_size,
4580 for (i = 0; i < nr; i++)
4581 total_data += data_size[i];
4583 total_size = total_data + (nr * sizeof(struct btrfs_item));
4584 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4590 slot = path->slots[0];
4593 setup_items_for_insert(root, path, cpu_key, data_size,
4594 total_data, total_size, nr);
4599 * Given a key and some data, insert an item into the tree.
4600 * This does all the path init required, making room in the tree if needed.
4602 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4603 *root, struct btrfs_key *cpu_key, void *data, u32
4607 struct btrfs_path *path;
4608 struct extent_buffer *leaf;
4611 path = btrfs_alloc_path();
4614 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4616 leaf = path->nodes[0];
4617 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4618 write_extent_buffer(leaf, data, ptr, data_size);
4619 btrfs_mark_buffer_dirty(leaf);
4621 btrfs_free_path(path);
4626 * delete the pointer from a given node.
4628 * the tree should have been previously balanced so the deletion does not
4631 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4632 int level, int slot)
4634 struct extent_buffer *parent = path->nodes[level];
4638 nritems = btrfs_header_nritems(parent);
4639 if (slot != nritems - 1) {
4641 tree_mod_log_eb_move(root->fs_info, parent, slot,
4642 slot + 1, nritems - slot - 1);
4643 memmove_extent_buffer(parent,
4644 btrfs_node_key_ptr_offset(slot),
4645 btrfs_node_key_ptr_offset(slot + 1),
4646 sizeof(struct btrfs_key_ptr) *
4647 (nritems - slot - 1));
4649 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4650 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4655 btrfs_set_header_nritems(parent, nritems);
4656 if (nritems == 0 && parent == root->node) {
4657 BUG_ON(btrfs_header_level(root->node) != 1);
4658 /* just turn the root into a leaf and break */
4659 btrfs_set_header_level(root->node, 0);
4660 } else if (slot == 0) {
4661 struct btrfs_disk_key disk_key;
4663 btrfs_node_key(parent, &disk_key, 0);
4664 fixup_low_keys(root, path, &disk_key, level + 1);
4666 btrfs_mark_buffer_dirty(parent);
4670 * a helper function to delete the leaf pointed to by path->slots[1] and
4673 * This deletes the pointer in path->nodes[1] and frees the leaf
4674 * block extent. zero is returned if it all worked out, < 0 otherwise.
4676 * The path must have already been setup for deleting the leaf, including
4677 * all the proper balancing. path->nodes[1] must be locked.
4679 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4680 struct btrfs_root *root,
4681 struct btrfs_path *path,
4682 struct extent_buffer *leaf)
4684 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4685 del_ptr(root, path, 1, path->slots[1]);
4688 * btrfs_free_extent is expensive, we want to make sure we
4689 * aren't holding any locks when we call it
4691 btrfs_unlock_up_safe(path, 0);
4693 root_sub_used(root, leaf->len);
4695 extent_buffer_get(leaf);
4696 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4697 free_extent_buffer_stale(leaf);
4700 * delete the item at the leaf level in path. If that empties
4701 * the leaf, remove it from the tree
4703 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4704 struct btrfs_path *path, int slot, int nr)
4706 struct extent_buffer *leaf;
4707 struct btrfs_item *item;
4714 struct btrfs_map_token token;
4716 btrfs_init_map_token(&token);
4718 leaf = path->nodes[0];
4719 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4721 for (i = 0; i < nr; i++)
4722 dsize += btrfs_item_size_nr(leaf, slot + i);
4724 nritems = btrfs_header_nritems(leaf);
4726 if (slot + nr != nritems) {
4727 int data_end = leaf_data_end(root, leaf);
4729 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4731 btrfs_leaf_data(leaf) + data_end,
4732 last_off - data_end);
4734 for (i = slot + nr; i < nritems; i++) {
4737 item = btrfs_item_nr(i);
4738 ioff = btrfs_token_item_offset(leaf, item, &token);
4739 btrfs_set_token_item_offset(leaf, item,
4740 ioff + dsize, &token);
4743 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4744 btrfs_item_nr_offset(slot + nr),
4745 sizeof(struct btrfs_item) *
4746 (nritems - slot - nr));
4748 btrfs_set_header_nritems(leaf, nritems - nr);
4751 /* delete the leaf if we've emptied it */
4753 if (leaf == root->node) {
4754 btrfs_set_header_level(leaf, 0);
4756 btrfs_set_path_blocking(path);
4757 clean_tree_block(trans, root, leaf);
4758 btrfs_del_leaf(trans, root, path, leaf);
4761 int used = leaf_space_used(leaf, 0, nritems);
4763 struct btrfs_disk_key disk_key;
4765 btrfs_item_key(leaf, &disk_key, 0);
4766 fixup_low_keys(root, path, &disk_key, 1);
4769 /* delete the leaf if it is mostly empty */
4770 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
4771 /* push_leaf_left fixes the path.
4772 * make sure the path still points to our leaf
4773 * for possible call to del_ptr below
4775 slot = path->slots[1];
4776 extent_buffer_get(leaf);
4778 btrfs_set_path_blocking(path);
4779 wret = push_leaf_left(trans, root, path, 1, 1,
4781 if (wret < 0 && wret != -ENOSPC)
4784 if (path->nodes[0] == leaf &&
4785 btrfs_header_nritems(leaf)) {
4786 wret = push_leaf_right(trans, root, path, 1,
4788 if (wret < 0 && wret != -ENOSPC)
4792 if (btrfs_header_nritems(leaf) == 0) {
4793 path->slots[1] = slot;
4794 btrfs_del_leaf(trans, root, path, leaf);
4795 free_extent_buffer(leaf);
4798 /* if we're still in the path, make sure
4799 * we're dirty. Otherwise, one of the
4800 * push_leaf functions must have already
4801 * dirtied this buffer
4803 if (path->nodes[0] == leaf)
4804 btrfs_mark_buffer_dirty(leaf);
4805 free_extent_buffer(leaf);
4808 btrfs_mark_buffer_dirty(leaf);
4815 * search the tree again to find a leaf with lesser keys
4816 * returns 0 if it found something or 1 if there are no lesser leaves.
4817 * returns < 0 on io errors.
4819 * This may release the path, and so you may lose any locks held at the
4822 static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4824 struct btrfs_key key;
4825 struct btrfs_disk_key found_key;
4828 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4832 else if (key.type > 0)
4834 else if (key.objectid > 0)
4839 btrfs_release_path(path);
4840 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4843 btrfs_item_key(path->nodes[0], &found_key, 0);
4844 ret = comp_keys(&found_key, &key);
4851 * A helper function to walk down the tree starting at min_key, and looking
4852 * for nodes or leaves that are have a minimum transaction id.
4853 * This is used by the btree defrag code, and tree logging
4855 * This does not cow, but it does stuff the starting key it finds back
4856 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4857 * key and get a writable path.
4859 * This does lock as it descends, and path->keep_locks should be set
4860 * to 1 by the caller.
4862 * This honors path->lowest_level to prevent descent past a given level
4865 * min_trans indicates the oldest transaction that you are interested
4866 * in walking through. Any nodes or leaves older than min_trans are
4867 * skipped over (without reading them).
4869 * returns zero if something useful was found, < 0 on error and 1 if there
4870 * was nothing in the tree that matched the search criteria.
4872 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4873 struct btrfs_key *max_key,
4874 struct btrfs_path *path,
4877 struct extent_buffer *cur;
4878 struct btrfs_key found_key;
4885 WARN_ON(!path->keep_locks);
4887 cur = btrfs_read_lock_root_node(root);
4888 level = btrfs_header_level(cur);
4889 WARN_ON(path->nodes[level]);
4890 path->nodes[level] = cur;
4891 path->locks[level] = BTRFS_READ_LOCK;
4893 if (btrfs_header_generation(cur) < min_trans) {
4898 nritems = btrfs_header_nritems(cur);
4899 level = btrfs_header_level(cur);
4900 sret = bin_search(cur, min_key, level, &slot);
4902 /* at the lowest level, we're done, setup the path and exit */
4903 if (level == path->lowest_level) {
4904 if (slot >= nritems)
4907 path->slots[level] = slot;
4908 btrfs_item_key_to_cpu(cur, &found_key, slot);
4911 if (sret && slot > 0)
4914 * check this node pointer against the min_trans parameters.
4915 * If it is too old, old, skip to the next one.
4917 while (slot < nritems) {
4921 blockptr = btrfs_node_blockptr(cur, slot);
4922 gen = btrfs_node_ptr_generation(cur, slot);
4923 if (gen < min_trans) {
4931 * we didn't find a candidate key in this node, walk forward
4932 * and find another one
4934 if (slot >= nritems) {
4935 path->slots[level] = slot;
4936 btrfs_set_path_blocking(path);
4937 sret = btrfs_find_next_key(root, path, min_key, level,
4940 btrfs_release_path(path);
4946 /* save our key for returning back */
4947 btrfs_node_key_to_cpu(cur, &found_key, slot);
4948 path->slots[level] = slot;
4949 if (level == path->lowest_level) {
4951 unlock_up(path, level, 1, 0, NULL);
4954 btrfs_set_path_blocking(path);
4955 cur = read_node_slot(root, cur, slot);
4956 BUG_ON(!cur); /* -ENOMEM */
4958 btrfs_tree_read_lock(cur);
4960 path->locks[level - 1] = BTRFS_READ_LOCK;
4961 path->nodes[level - 1] = cur;
4962 unlock_up(path, level, 1, 0, NULL);
4963 btrfs_clear_path_blocking(path, NULL, 0);
4967 memcpy(min_key, &found_key, sizeof(found_key));
4968 btrfs_set_path_blocking(path);
4972 static void tree_move_down(struct btrfs_root *root,
4973 struct btrfs_path *path,
4974 int *level, int root_level)
4976 BUG_ON(*level == 0);
4977 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
4978 path->slots[*level]);
4979 path->slots[*level - 1] = 0;
4983 static int tree_move_next_or_upnext(struct btrfs_root *root,
4984 struct btrfs_path *path,
4985 int *level, int root_level)
4989 nritems = btrfs_header_nritems(path->nodes[*level]);
4991 path->slots[*level]++;
4993 while (path->slots[*level] >= nritems) {
4994 if (*level == root_level)
4998 path->slots[*level] = 0;
4999 free_extent_buffer(path->nodes[*level]);
5000 path->nodes[*level] = NULL;
5002 path->slots[*level]++;
5004 nritems = btrfs_header_nritems(path->nodes[*level]);
5011 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5014 static int tree_advance(struct btrfs_root *root,
5015 struct btrfs_path *path,
5016 int *level, int root_level,
5018 struct btrfs_key *key)
5022 if (*level == 0 || !allow_down) {
5023 ret = tree_move_next_or_upnext(root, path, level, root_level);
5025 tree_move_down(root, path, level, root_level);
5030 btrfs_item_key_to_cpu(path->nodes[*level], key,
5031 path->slots[*level]);
5033 btrfs_node_key_to_cpu(path->nodes[*level], key,
5034 path->slots[*level]);
5039 static int tree_compare_item(struct btrfs_root *left_root,
5040 struct btrfs_path *left_path,
5041 struct btrfs_path *right_path,
5046 unsigned long off1, off2;
5048 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5049 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5053 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5054 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5055 right_path->slots[0]);
5057 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5059 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5066 #define ADVANCE_ONLY_NEXT -1
5069 * This function compares two trees and calls the provided callback for
5070 * every changed/new/deleted item it finds.
5071 * If shared tree blocks are encountered, whole subtrees are skipped, making
5072 * the compare pretty fast on snapshotted subvolumes.
5074 * This currently works on commit roots only. As commit roots are read only,
5075 * we don't do any locking. The commit roots are protected with transactions.
5076 * Transactions are ended and rejoined when a commit is tried in between.
5078 * This function checks for modifications done to the trees while comparing.
5079 * If it detects a change, it aborts immediately.
5081 int btrfs_compare_trees(struct btrfs_root *left_root,
5082 struct btrfs_root *right_root,
5083 btrfs_changed_cb_t changed_cb, void *ctx)
5087 struct btrfs_trans_handle *trans = NULL;
5088 struct btrfs_path *left_path = NULL;
5089 struct btrfs_path *right_path = NULL;
5090 struct btrfs_key left_key;
5091 struct btrfs_key right_key;
5092 char *tmp_buf = NULL;
5093 int left_root_level;
5094 int right_root_level;
5097 int left_end_reached;
5098 int right_end_reached;
5103 u64 left_start_ctransid;
5104 u64 right_start_ctransid;
5107 left_path = btrfs_alloc_path();
5112 right_path = btrfs_alloc_path();
5118 tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5124 left_path->search_commit_root = 1;
5125 left_path->skip_locking = 1;
5126 right_path->search_commit_root = 1;
5127 right_path->skip_locking = 1;
5129 spin_lock(&left_root->root_item_lock);
5130 left_start_ctransid = btrfs_root_ctransid(&left_root->root_item);
5131 spin_unlock(&left_root->root_item_lock);
5133 spin_lock(&right_root->root_item_lock);
5134 right_start_ctransid = btrfs_root_ctransid(&right_root->root_item);
5135 spin_unlock(&right_root->root_item_lock);
5137 trans = btrfs_join_transaction(left_root);
5138 if (IS_ERR(trans)) {
5139 ret = PTR_ERR(trans);
5145 * Strategy: Go to the first items of both trees. Then do
5147 * If both trees are at level 0
5148 * Compare keys of current items
5149 * If left < right treat left item as new, advance left tree
5151 * If left > right treat right item as deleted, advance right tree
5153 * If left == right do deep compare of items, treat as changed if
5154 * needed, advance both trees and repeat
5155 * If both trees are at the same level but not at level 0
5156 * Compare keys of current nodes/leafs
5157 * If left < right advance left tree and repeat
5158 * If left > right advance right tree and repeat
5159 * If left == right compare blockptrs of the next nodes/leafs
5160 * If they match advance both trees but stay at the same level
5162 * If they don't match advance both trees while allowing to go
5164 * If tree levels are different
5165 * Advance the tree that needs it and repeat
5167 * Advancing a tree means:
5168 * If we are at level 0, try to go to the next slot. If that's not
5169 * possible, go one level up and repeat. Stop when we found a level
5170 * where we could go to the next slot. We may at this point be on a
5173 * If we are not at level 0 and not on shared tree blocks, go one
5176 * If we are not at level 0 and on shared tree blocks, go one slot to
5177 * the right if possible or go up and right.
5180 left_level = btrfs_header_level(left_root->commit_root);
5181 left_root_level = left_level;
5182 left_path->nodes[left_level] = left_root->commit_root;
5183 extent_buffer_get(left_path->nodes[left_level]);
5185 right_level = btrfs_header_level(right_root->commit_root);
5186 right_root_level = right_level;
5187 right_path->nodes[right_level] = right_root->commit_root;
5188 extent_buffer_get(right_path->nodes[right_level]);
5190 if (left_level == 0)
5191 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5192 &left_key, left_path->slots[left_level]);
5194 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5195 &left_key, left_path->slots[left_level]);
5196 if (right_level == 0)
5197 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5198 &right_key, right_path->slots[right_level]);
5200 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5201 &right_key, right_path->slots[right_level]);
5203 left_end_reached = right_end_reached = 0;
5204 advance_left = advance_right = 0;
5208 * We need to make sure the transaction does not get committed
5209 * while we do anything on commit roots. This means, we need to
5210 * join and leave transactions for every item that we process.
5212 if (trans && btrfs_should_end_transaction(trans, left_root)) {
5213 btrfs_release_path(left_path);
5214 btrfs_release_path(right_path);
5216 ret = btrfs_end_transaction(trans, left_root);
5221 /* now rejoin the transaction */
5223 trans = btrfs_join_transaction(left_root);
5224 if (IS_ERR(trans)) {
5225 ret = PTR_ERR(trans);
5230 spin_lock(&left_root->root_item_lock);
5231 ctransid = btrfs_root_ctransid(&left_root->root_item);
5232 spin_unlock(&left_root->root_item_lock);
5233 if (ctransid != left_start_ctransid)
5234 left_start_ctransid = 0;
5236 spin_lock(&right_root->root_item_lock);
5237 ctransid = btrfs_root_ctransid(&right_root->root_item);
5238 spin_unlock(&right_root->root_item_lock);
5239 if (ctransid != right_start_ctransid)
5240 right_start_ctransid = 0;
5242 if (!left_start_ctransid || !right_start_ctransid) {
5243 WARN(1, KERN_WARNING
5244 "btrfs: btrfs_compare_tree detected "
5245 "a change in one of the trees while "
5246 "iterating. This is probably a "
5253 * the commit root may have changed, so start again
5256 left_path->lowest_level = left_level;
5257 right_path->lowest_level = right_level;
5258 ret = btrfs_search_slot(NULL, left_root,
5259 &left_key, left_path, 0, 0);
5262 ret = btrfs_search_slot(NULL, right_root,
5263 &right_key, right_path, 0, 0);
5268 if (advance_left && !left_end_reached) {
5269 ret = tree_advance(left_root, left_path, &left_level,
5271 advance_left != ADVANCE_ONLY_NEXT,
5274 left_end_reached = ADVANCE;
5277 if (advance_right && !right_end_reached) {
5278 ret = tree_advance(right_root, right_path, &right_level,
5280 advance_right != ADVANCE_ONLY_NEXT,
5283 right_end_reached = ADVANCE;
5287 if (left_end_reached && right_end_reached) {
5290 } else if (left_end_reached) {
5291 if (right_level == 0) {
5292 ret = changed_cb(left_root, right_root,
5293 left_path, right_path,
5295 BTRFS_COMPARE_TREE_DELETED,
5300 advance_right = ADVANCE;
5302 } else if (right_end_reached) {
5303 if (left_level == 0) {
5304 ret = changed_cb(left_root, right_root,
5305 left_path, right_path,
5307 BTRFS_COMPARE_TREE_NEW,
5312 advance_left = ADVANCE;
5316 if (left_level == 0 && right_level == 0) {
5317 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5319 ret = changed_cb(left_root, right_root,
5320 left_path, right_path,
5322 BTRFS_COMPARE_TREE_NEW,
5326 advance_left = ADVANCE;
5327 } else if (cmp > 0) {
5328 ret = changed_cb(left_root, right_root,
5329 left_path, right_path,
5331 BTRFS_COMPARE_TREE_DELETED,
5335 advance_right = ADVANCE;
5337 enum btrfs_compare_tree_result cmp;
5339 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5340 ret = tree_compare_item(left_root, left_path,
5341 right_path, tmp_buf);
5343 cmp = BTRFS_COMPARE_TREE_CHANGED;
5345 cmp = BTRFS_COMPARE_TREE_SAME;
5346 ret = changed_cb(left_root, right_root,
5347 left_path, right_path,
5348 &left_key, cmp, ctx);
5351 advance_left = ADVANCE;
5352 advance_right = ADVANCE;
5354 } else if (left_level == right_level) {
5355 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5357 advance_left = ADVANCE;
5358 } else if (cmp > 0) {
5359 advance_right = ADVANCE;
5361 left_blockptr = btrfs_node_blockptr(
5362 left_path->nodes[left_level],
5363 left_path->slots[left_level]);
5364 right_blockptr = btrfs_node_blockptr(
5365 right_path->nodes[right_level],
5366 right_path->slots[right_level]);
5367 if (left_blockptr == right_blockptr) {
5369 * As we're on a shared block, don't
5370 * allow to go deeper.
5372 advance_left = ADVANCE_ONLY_NEXT;
5373 advance_right = ADVANCE_ONLY_NEXT;
5375 advance_left = ADVANCE;
5376 advance_right = ADVANCE;
5379 } else if (left_level < right_level) {
5380 advance_right = ADVANCE;
5382 advance_left = ADVANCE;
5387 btrfs_free_path(left_path);
5388 btrfs_free_path(right_path);
5393 ret = btrfs_end_transaction(trans, left_root);
5395 btrfs_end_transaction(trans, left_root);
5402 * this is similar to btrfs_next_leaf, but does not try to preserve
5403 * and fixup the path. It looks for and returns the next key in the
5404 * tree based on the current path and the min_trans parameters.
5406 * 0 is returned if another key is found, < 0 if there are any errors
5407 * and 1 is returned if there are no higher keys in the tree
5409 * path->keep_locks should be set to 1 on the search made before
5410 * calling this function.
5412 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5413 struct btrfs_key *key, int level, u64 min_trans)
5416 struct extent_buffer *c;
5418 WARN_ON(!path->keep_locks);
5419 while (level < BTRFS_MAX_LEVEL) {
5420 if (!path->nodes[level])
5423 slot = path->slots[level] + 1;
5424 c = path->nodes[level];
5426 if (slot >= btrfs_header_nritems(c)) {
5429 struct btrfs_key cur_key;
5430 if (level + 1 >= BTRFS_MAX_LEVEL ||
5431 !path->nodes[level + 1])
5434 if (path->locks[level + 1]) {
5439 slot = btrfs_header_nritems(c) - 1;
5441 btrfs_item_key_to_cpu(c, &cur_key, slot);
5443 btrfs_node_key_to_cpu(c, &cur_key, slot);
5445 orig_lowest = path->lowest_level;
5446 btrfs_release_path(path);
5447 path->lowest_level = level;
5448 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5450 path->lowest_level = orig_lowest;
5454 c = path->nodes[level];
5455 slot = path->slots[level];
5462 btrfs_item_key_to_cpu(c, key, slot);
5464 u64 gen = btrfs_node_ptr_generation(c, slot);
5466 if (gen < min_trans) {
5470 btrfs_node_key_to_cpu(c, key, slot);
5478 * search the tree again to find a leaf with greater keys
5479 * returns 0 if it found something or 1 if there are no greater leaves.
5480 * returns < 0 on io errors.
5482 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5484 return btrfs_next_old_leaf(root, path, 0);
5487 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5492 struct extent_buffer *c;
5493 struct extent_buffer *next;
5494 struct btrfs_key key;
5497 int old_spinning = path->leave_spinning;
5498 int next_rw_lock = 0;
5500 nritems = btrfs_header_nritems(path->nodes[0]);
5504 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5509 btrfs_release_path(path);
5511 path->keep_locks = 1;
5512 path->leave_spinning = 1;
5515 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5517 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5518 path->keep_locks = 0;
5523 nritems = btrfs_header_nritems(path->nodes[0]);
5525 * by releasing the path above we dropped all our locks. A balance
5526 * could have added more items next to the key that used to be
5527 * at the very end of the block. So, check again here and
5528 * advance the path if there are now more items available.
5530 if (nritems > 0 && path->slots[0] < nritems - 1) {
5537 while (level < BTRFS_MAX_LEVEL) {
5538 if (!path->nodes[level]) {
5543 slot = path->slots[level] + 1;
5544 c = path->nodes[level];
5545 if (slot >= btrfs_header_nritems(c)) {
5547 if (level == BTRFS_MAX_LEVEL) {
5555 btrfs_tree_unlock_rw(next, next_rw_lock);
5556 free_extent_buffer(next);
5560 next_rw_lock = path->locks[level];
5561 ret = read_block_for_search(NULL, root, path, &next, level,
5567 btrfs_release_path(path);
5571 if (!path->skip_locking) {
5572 ret = btrfs_try_tree_read_lock(next);
5573 if (!ret && time_seq) {
5575 * If we don't get the lock, we may be racing
5576 * with push_leaf_left, holding that lock while
5577 * itself waiting for the leaf we've currently
5578 * locked. To solve this situation, we give up
5579 * on our lock and cycle.
5581 free_extent_buffer(next);
5582 btrfs_release_path(path);
5587 btrfs_set_path_blocking(path);
5588 btrfs_tree_read_lock(next);
5589 btrfs_clear_path_blocking(path, next,
5592 next_rw_lock = BTRFS_READ_LOCK;
5596 path->slots[level] = slot;
5599 c = path->nodes[level];
5600 if (path->locks[level])
5601 btrfs_tree_unlock_rw(c, path->locks[level]);
5603 free_extent_buffer(c);
5604 path->nodes[level] = next;
5605 path->slots[level] = 0;
5606 if (!path->skip_locking)
5607 path->locks[level] = next_rw_lock;
5611 ret = read_block_for_search(NULL, root, path, &next, level,
5617 btrfs_release_path(path);
5621 if (!path->skip_locking) {
5622 ret = btrfs_try_tree_read_lock(next);
5624 btrfs_set_path_blocking(path);
5625 btrfs_tree_read_lock(next);
5626 btrfs_clear_path_blocking(path, next,
5629 next_rw_lock = BTRFS_READ_LOCK;
5634 unlock_up(path, 0, 1, 0, NULL);
5635 path->leave_spinning = old_spinning;
5637 btrfs_set_path_blocking(path);
5643 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5644 * searching until it gets past min_objectid or finds an item of 'type'
5646 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5648 int btrfs_previous_item(struct btrfs_root *root,
5649 struct btrfs_path *path, u64 min_objectid,
5652 struct btrfs_key found_key;
5653 struct extent_buffer *leaf;
5658 if (path->slots[0] == 0) {
5659 btrfs_set_path_blocking(path);
5660 ret = btrfs_prev_leaf(root, path);
5666 leaf = path->nodes[0];
5667 nritems = btrfs_header_nritems(leaf);
5670 if (path->slots[0] == nritems)
5673 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5674 if (found_key.objectid < min_objectid)
5676 if (found_key.type == type)
5678 if (found_key.objectid == min_objectid &&
5679 found_key.type < type)