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);
1009 btrfs_reloc_cow_block(trans, root, buf, cow);
1011 if (buf == root->node) {
1012 WARN_ON(parent && parent != buf);
1013 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1014 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1015 parent_start = buf->start;
1019 extent_buffer_get(cow);
1020 tree_mod_log_set_root_pointer(root, cow, 1);
1021 rcu_assign_pointer(root->node, cow);
1023 btrfs_free_tree_block(trans, root, buf, parent_start,
1025 free_extent_buffer(buf);
1026 add_root_to_dirty_list(root);
1028 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1029 parent_start = parent->start;
1033 WARN_ON(trans->transid != btrfs_header_generation(parent));
1034 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1035 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1036 btrfs_set_node_blockptr(parent, parent_slot,
1038 btrfs_set_node_ptr_generation(parent, parent_slot,
1040 btrfs_mark_buffer_dirty(parent);
1042 tree_mod_log_free_eb(root->fs_info, buf);
1043 btrfs_free_tree_block(trans, root, buf, parent_start,
1047 btrfs_tree_unlock(buf);
1048 free_extent_buffer_stale(buf);
1049 btrfs_mark_buffer_dirty(cow);
1055 * returns the logical address of the oldest predecessor of the given root.
1056 * entries older than time_seq are ignored.
1058 static struct tree_mod_elem *
1059 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1060 struct extent_buffer *eb_root, u64 time_seq)
1062 struct tree_mod_elem *tm;
1063 struct tree_mod_elem *found = NULL;
1064 u64 root_logical = eb_root->start;
1071 * the very last operation that's logged for a root is the replacement
1072 * operation (if it is replaced at all). this has the index of the *new*
1073 * root, making it the very first operation that's logged for this root.
1076 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1081 * if there are no tree operation for the oldest root, we simply
1082 * return it. this should only happen if that (old) root is at
1089 * if there's an operation that's not a root replacement, we
1090 * found the oldest version of our root. normally, we'll find a
1091 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1093 if (tm->op != MOD_LOG_ROOT_REPLACE)
1097 root_logical = tm->old_root.logical;
1101 /* if there's no old root to return, return what we found instead */
1109 * tm is a pointer to the first operation to rewind within eb. then, all
1110 * previous operations will be rewinded (until we reach something older than
1114 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1115 u64 time_seq, struct tree_mod_elem *first_tm)
1118 struct rb_node *next;
1119 struct tree_mod_elem *tm = first_tm;
1120 unsigned long o_dst;
1121 unsigned long o_src;
1122 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1124 n = btrfs_header_nritems(eb);
1125 tree_mod_log_read_lock(fs_info);
1126 while (tm && tm->seq >= time_seq) {
1128 * all the operations are recorded with the operator used for
1129 * the modification. as we're going backwards, we do the
1130 * opposite of each operation here.
1133 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1134 BUG_ON(tm->slot < n);
1136 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1137 case MOD_LOG_KEY_REMOVE:
1138 btrfs_set_node_key(eb, &tm->key, tm->slot);
1139 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1140 btrfs_set_node_ptr_generation(eb, tm->slot,
1144 case MOD_LOG_KEY_REPLACE:
1145 BUG_ON(tm->slot >= n);
1146 btrfs_set_node_key(eb, &tm->key, tm->slot);
1147 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1148 btrfs_set_node_ptr_generation(eb, tm->slot,
1151 case MOD_LOG_KEY_ADD:
1152 /* if a move operation is needed it's in the log */
1155 case MOD_LOG_MOVE_KEYS:
1156 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1157 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1158 memmove_extent_buffer(eb, o_dst, o_src,
1159 tm->move.nr_items * p_size);
1161 case MOD_LOG_ROOT_REPLACE:
1163 * this operation is special. for roots, this must be
1164 * handled explicitly before rewinding.
1165 * for non-roots, this operation may exist if the node
1166 * was a root: root A -> child B; then A gets empty and
1167 * B is promoted to the new root. in the mod log, we'll
1168 * have a root-replace operation for B, a tree block
1169 * that is no root. we simply ignore that operation.
1173 next = rb_next(&tm->node);
1176 tm = container_of(next, struct tree_mod_elem, node);
1177 if (tm->index != first_tm->index)
1180 tree_mod_log_read_unlock(fs_info);
1181 btrfs_set_header_nritems(eb, n);
1185 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1186 * is returned. If rewind operations happen, a fresh buffer is returned. The
1187 * returned buffer is always read-locked. If the returned buffer is not the
1188 * input buffer, the lock on the input buffer is released and the input buffer
1189 * is freed (its refcount is decremented).
1191 static struct extent_buffer *
1192 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1193 struct extent_buffer *eb, u64 time_seq)
1195 struct extent_buffer *eb_rewin;
1196 struct tree_mod_elem *tm;
1201 if (btrfs_header_level(eb) == 0)
1204 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1208 btrfs_set_path_blocking(path);
1209 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1211 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1212 BUG_ON(tm->slot != 0);
1213 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1214 fs_info->tree_root->nodesize);
1216 btrfs_tree_read_unlock_blocking(eb);
1217 free_extent_buffer(eb);
1220 btrfs_set_header_bytenr(eb_rewin, eb->start);
1221 btrfs_set_header_backref_rev(eb_rewin,
1222 btrfs_header_backref_rev(eb));
1223 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1224 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1226 eb_rewin = btrfs_clone_extent_buffer(eb);
1228 btrfs_tree_read_unlock_blocking(eb);
1229 free_extent_buffer(eb);
1234 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1235 btrfs_tree_read_unlock_blocking(eb);
1236 free_extent_buffer(eb);
1238 extent_buffer_get(eb_rewin);
1239 btrfs_tree_read_lock(eb_rewin);
1240 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1241 WARN_ON(btrfs_header_nritems(eb_rewin) >
1242 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1248 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1249 * value. If there are no changes, the current root->root_node is returned. If
1250 * anything changed in between, there's a fresh buffer allocated on which the
1251 * rewind operations are done. In any case, the returned buffer is read locked.
1252 * Returns NULL on error (with no locks held).
1254 static inline struct extent_buffer *
1255 get_old_root(struct btrfs_root *root, u64 time_seq)
1257 struct tree_mod_elem *tm;
1258 struct extent_buffer *eb = NULL;
1259 struct extent_buffer *eb_root;
1260 struct extent_buffer *old;
1261 struct tree_mod_root *old_root = NULL;
1262 u64 old_generation = 0;
1266 eb_root = btrfs_read_lock_root_node(root);
1267 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1271 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1272 old_root = &tm->old_root;
1273 old_generation = tm->generation;
1274 logical = old_root->logical;
1276 logical = eb_root->start;
1279 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1280 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1281 btrfs_tree_read_unlock(eb_root);
1282 free_extent_buffer(eb_root);
1283 blocksize = btrfs_level_size(root, old_root->level);
1284 old = read_tree_block(root, logical, blocksize, 0);
1285 if (!old || !extent_buffer_uptodate(old)) {
1286 free_extent_buffer(old);
1287 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1291 eb = btrfs_clone_extent_buffer(old);
1292 free_extent_buffer(old);
1294 } else if (old_root) {
1295 btrfs_tree_read_unlock(eb_root);
1296 free_extent_buffer(eb_root);
1297 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1299 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1300 eb = btrfs_clone_extent_buffer(eb_root);
1301 btrfs_tree_read_unlock_blocking(eb_root);
1302 free_extent_buffer(eb_root);
1307 extent_buffer_get(eb);
1308 btrfs_tree_read_lock(eb);
1310 btrfs_set_header_bytenr(eb, eb->start);
1311 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1312 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1313 btrfs_set_header_level(eb, old_root->level);
1314 btrfs_set_header_generation(eb, old_generation);
1317 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1319 WARN_ON(btrfs_header_level(eb) != 0);
1320 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1325 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1327 struct tree_mod_elem *tm;
1329 struct extent_buffer *eb_root = btrfs_root_node(root);
1331 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1332 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1333 level = tm->old_root.level;
1335 level = btrfs_header_level(eb_root);
1337 free_extent_buffer(eb_root);
1342 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1343 struct btrfs_root *root,
1344 struct extent_buffer *buf)
1346 /* ensure we can see the force_cow */
1350 * We do not need to cow a block if
1351 * 1) this block is not created or changed in this transaction;
1352 * 2) this block does not belong to TREE_RELOC tree;
1353 * 3) the root is not forced COW.
1355 * What is forced COW:
1356 * when we create snapshot during commiting the transaction,
1357 * after we've finished coping src root, we must COW the shared
1358 * block to ensure the metadata consistency.
1360 if (btrfs_header_generation(buf) == trans->transid &&
1361 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1362 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1363 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1370 * cows a single block, see __btrfs_cow_block for the real work.
1371 * This version of it has extra checks so that a block isn't cow'd more than
1372 * once per transaction, as long as it hasn't been written yet
1374 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1375 struct btrfs_root *root, struct extent_buffer *buf,
1376 struct extent_buffer *parent, int parent_slot,
1377 struct extent_buffer **cow_ret)
1382 if (trans->transaction != root->fs_info->running_transaction)
1383 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1385 root->fs_info->running_transaction->transid);
1387 if (trans->transid != root->fs_info->generation)
1388 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1389 trans->transid, root->fs_info->generation);
1391 if (!should_cow_block(trans, root, buf)) {
1396 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1399 btrfs_set_lock_blocking(parent);
1400 btrfs_set_lock_blocking(buf);
1402 ret = __btrfs_cow_block(trans, root, buf, parent,
1403 parent_slot, cow_ret, search_start, 0);
1405 trace_btrfs_cow_block(root, buf, *cow_ret);
1411 * helper function for defrag to decide if two blocks pointed to by a
1412 * node are actually close by
1414 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1416 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1418 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1424 * compare two keys in a memcmp fashion
1426 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1428 struct btrfs_key k1;
1430 btrfs_disk_key_to_cpu(&k1, disk);
1432 return btrfs_comp_cpu_keys(&k1, k2);
1436 * same as comp_keys only with two btrfs_key's
1438 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1440 if (k1->objectid > k2->objectid)
1442 if (k1->objectid < k2->objectid)
1444 if (k1->type > k2->type)
1446 if (k1->type < k2->type)
1448 if (k1->offset > k2->offset)
1450 if (k1->offset < k2->offset)
1456 * this is used by the defrag code to go through all the
1457 * leaves pointed to by a node and reallocate them so that
1458 * disk order is close to key order
1460 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1461 struct btrfs_root *root, struct extent_buffer *parent,
1462 int start_slot, u64 *last_ret,
1463 struct btrfs_key *progress)
1465 struct extent_buffer *cur;
1468 u64 search_start = *last_ret;
1478 int progress_passed = 0;
1479 struct btrfs_disk_key disk_key;
1481 parent_level = btrfs_header_level(parent);
1483 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1484 WARN_ON(trans->transid != root->fs_info->generation);
1486 parent_nritems = btrfs_header_nritems(parent);
1487 blocksize = btrfs_level_size(root, parent_level - 1);
1488 end_slot = parent_nritems;
1490 if (parent_nritems == 1)
1493 btrfs_set_lock_blocking(parent);
1495 for (i = start_slot; i < end_slot; i++) {
1498 btrfs_node_key(parent, &disk_key, i);
1499 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1502 progress_passed = 1;
1503 blocknr = btrfs_node_blockptr(parent, i);
1504 gen = btrfs_node_ptr_generation(parent, i);
1505 if (last_block == 0)
1506 last_block = blocknr;
1509 other = btrfs_node_blockptr(parent, i - 1);
1510 close = close_blocks(blocknr, other, blocksize);
1512 if (!close && i < end_slot - 2) {
1513 other = btrfs_node_blockptr(parent, i + 1);
1514 close = close_blocks(blocknr, other, blocksize);
1517 last_block = blocknr;
1521 cur = btrfs_find_tree_block(root, blocknr, blocksize);
1523 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1526 if (!cur || !uptodate) {
1528 cur = read_tree_block(root, blocknr,
1530 if (!cur || !extent_buffer_uptodate(cur)) {
1531 free_extent_buffer(cur);
1534 } else if (!uptodate) {
1535 err = btrfs_read_buffer(cur, gen);
1537 free_extent_buffer(cur);
1542 if (search_start == 0)
1543 search_start = last_block;
1545 btrfs_tree_lock(cur);
1546 btrfs_set_lock_blocking(cur);
1547 err = __btrfs_cow_block(trans, root, cur, parent, i,
1550 (end_slot - i) * blocksize));
1552 btrfs_tree_unlock(cur);
1553 free_extent_buffer(cur);
1556 search_start = cur->start;
1557 last_block = cur->start;
1558 *last_ret = search_start;
1559 btrfs_tree_unlock(cur);
1560 free_extent_buffer(cur);
1566 * The leaf data grows from end-to-front in the node.
1567 * this returns the address of the start of the last item,
1568 * which is the stop of the leaf data stack
1570 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1571 struct extent_buffer *leaf)
1573 u32 nr = btrfs_header_nritems(leaf);
1575 return BTRFS_LEAF_DATA_SIZE(root);
1576 return btrfs_item_offset_nr(leaf, nr - 1);
1581 * search for key in the extent_buffer. The items start at offset p,
1582 * and they are item_size apart. There are 'max' items in p.
1584 * the slot in the array is returned via slot, and it points to
1585 * the place where you would insert key if it is not found in
1588 * slot may point to max if the key is bigger than all of the keys
1590 static noinline int generic_bin_search(struct extent_buffer *eb,
1592 int item_size, struct btrfs_key *key,
1599 struct btrfs_disk_key *tmp = NULL;
1600 struct btrfs_disk_key unaligned;
1601 unsigned long offset;
1603 unsigned long map_start = 0;
1604 unsigned long map_len = 0;
1607 while (low < high) {
1608 mid = (low + high) / 2;
1609 offset = p + mid * item_size;
1611 if (!kaddr || offset < map_start ||
1612 (offset + sizeof(struct btrfs_disk_key)) >
1613 map_start + map_len) {
1615 err = map_private_extent_buffer(eb, offset,
1616 sizeof(struct btrfs_disk_key),
1617 &kaddr, &map_start, &map_len);
1620 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1623 read_extent_buffer(eb, &unaligned,
1624 offset, sizeof(unaligned));
1629 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1632 ret = comp_keys(tmp, key);
1648 * simple bin_search frontend that does the right thing for
1651 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1652 int level, int *slot)
1655 return generic_bin_search(eb,
1656 offsetof(struct btrfs_leaf, items),
1657 sizeof(struct btrfs_item),
1658 key, btrfs_header_nritems(eb),
1661 return generic_bin_search(eb,
1662 offsetof(struct btrfs_node, ptrs),
1663 sizeof(struct btrfs_key_ptr),
1664 key, btrfs_header_nritems(eb),
1668 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1669 int level, int *slot)
1671 return bin_search(eb, key, level, slot);
1674 static void root_add_used(struct btrfs_root *root, u32 size)
1676 spin_lock(&root->accounting_lock);
1677 btrfs_set_root_used(&root->root_item,
1678 btrfs_root_used(&root->root_item) + size);
1679 spin_unlock(&root->accounting_lock);
1682 static void root_sub_used(struct btrfs_root *root, u32 size)
1684 spin_lock(&root->accounting_lock);
1685 btrfs_set_root_used(&root->root_item,
1686 btrfs_root_used(&root->root_item) - size);
1687 spin_unlock(&root->accounting_lock);
1690 /* given a node and slot number, this reads the blocks it points to. The
1691 * extent buffer is returned with a reference taken (but unlocked).
1692 * NULL is returned on error.
1694 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1695 struct extent_buffer *parent, int slot)
1697 int level = btrfs_header_level(parent);
1698 struct extent_buffer *eb;
1702 if (slot >= btrfs_header_nritems(parent))
1707 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1708 btrfs_level_size(root, level - 1),
1709 btrfs_node_ptr_generation(parent, slot));
1710 if (eb && !extent_buffer_uptodate(eb)) {
1711 free_extent_buffer(eb);
1719 * node level balancing, used to make sure nodes are in proper order for
1720 * item deletion. We balance from the top down, so we have to make sure
1721 * that a deletion won't leave an node completely empty later on.
1723 static noinline int balance_level(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path, int level)
1727 struct extent_buffer *right = NULL;
1728 struct extent_buffer *mid;
1729 struct extent_buffer *left = NULL;
1730 struct extent_buffer *parent = NULL;
1734 int orig_slot = path->slots[level];
1740 mid = path->nodes[level];
1742 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1743 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1744 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1746 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1748 if (level < BTRFS_MAX_LEVEL - 1) {
1749 parent = path->nodes[level + 1];
1750 pslot = path->slots[level + 1];
1754 * deal with the case where there is only one pointer in the root
1755 * by promoting the node below to a root
1758 struct extent_buffer *child;
1760 if (btrfs_header_nritems(mid) != 1)
1763 /* promote the child to a root */
1764 child = read_node_slot(root, mid, 0);
1767 btrfs_std_error(root->fs_info, ret);
1771 btrfs_tree_lock(child);
1772 btrfs_set_lock_blocking(child);
1773 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1775 btrfs_tree_unlock(child);
1776 free_extent_buffer(child);
1780 tree_mod_log_set_root_pointer(root, child, 1);
1781 rcu_assign_pointer(root->node, child);
1783 add_root_to_dirty_list(root);
1784 btrfs_tree_unlock(child);
1786 path->locks[level] = 0;
1787 path->nodes[level] = NULL;
1788 clean_tree_block(trans, root, mid);
1789 btrfs_tree_unlock(mid);
1790 /* once for the path */
1791 free_extent_buffer(mid);
1793 root_sub_used(root, mid->len);
1794 btrfs_free_tree_block(trans, root, mid, 0, 1);
1795 /* once for the root ptr */
1796 free_extent_buffer_stale(mid);
1799 if (btrfs_header_nritems(mid) >
1800 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1803 left = read_node_slot(root, parent, pslot - 1);
1805 btrfs_tree_lock(left);
1806 btrfs_set_lock_blocking(left);
1807 wret = btrfs_cow_block(trans, root, left,
1808 parent, pslot - 1, &left);
1814 right = read_node_slot(root, parent, pslot + 1);
1816 btrfs_tree_lock(right);
1817 btrfs_set_lock_blocking(right);
1818 wret = btrfs_cow_block(trans, root, right,
1819 parent, pslot + 1, &right);
1826 /* first, try to make some room in the middle buffer */
1828 orig_slot += btrfs_header_nritems(left);
1829 wret = push_node_left(trans, root, left, mid, 1);
1835 * then try to empty the right most buffer into the middle
1838 wret = push_node_left(trans, root, mid, right, 1);
1839 if (wret < 0 && wret != -ENOSPC)
1841 if (btrfs_header_nritems(right) == 0) {
1842 clean_tree_block(trans, root, right);
1843 btrfs_tree_unlock(right);
1844 del_ptr(root, path, level + 1, pslot + 1);
1845 root_sub_used(root, right->len);
1846 btrfs_free_tree_block(trans, root, right, 0, 1);
1847 free_extent_buffer_stale(right);
1850 struct btrfs_disk_key right_key;
1851 btrfs_node_key(right, &right_key, 0);
1852 tree_mod_log_set_node_key(root->fs_info, parent,
1854 btrfs_set_node_key(parent, &right_key, pslot + 1);
1855 btrfs_mark_buffer_dirty(parent);
1858 if (btrfs_header_nritems(mid) == 1) {
1860 * we're not allowed to leave a node with one item in the
1861 * tree during a delete. A deletion from lower in the tree
1862 * could try to delete the only pointer in this node.
1863 * So, pull some keys from the left.
1864 * There has to be a left pointer at this point because
1865 * otherwise we would have pulled some pointers from the
1870 btrfs_std_error(root->fs_info, ret);
1873 wret = balance_node_right(trans, root, mid, left);
1879 wret = push_node_left(trans, root, left, mid, 1);
1885 if (btrfs_header_nritems(mid) == 0) {
1886 clean_tree_block(trans, root, mid);
1887 btrfs_tree_unlock(mid);
1888 del_ptr(root, path, level + 1, pslot);
1889 root_sub_used(root, mid->len);
1890 btrfs_free_tree_block(trans, root, mid, 0, 1);
1891 free_extent_buffer_stale(mid);
1894 /* update the parent key to reflect our changes */
1895 struct btrfs_disk_key mid_key;
1896 btrfs_node_key(mid, &mid_key, 0);
1897 tree_mod_log_set_node_key(root->fs_info, parent,
1899 btrfs_set_node_key(parent, &mid_key, pslot);
1900 btrfs_mark_buffer_dirty(parent);
1903 /* update the path */
1905 if (btrfs_header_nritems(left) > orig_slot) {
1906 extent_buffer_get(left);
1907 /* left was locked after cow */
1908 path->nodes[level] = left;
1909 path->slots[level + 1] -= 1;
1910 path->slots[level] = orig_slot;
1912 btrfs_tree_unlock(mid);
1913 free_extent_buffer(mid);
1916 orig_slot -= btrfs_header_nritems(left);
1917 path->slots[level] = orig_slot;
1920 /* double check we haven't messed things up */
1922 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1926 btrfs_tree_unlock(right);
1927 free_extent_buffer(right);
1930 if (path->nodes[level] != left)
1931 btrfs_tree_unlock(left);
1932 free_extent_buffer(left);
1937 /* Node balancing for insertion. Here we only split or push nodes around
1938 * when they are completely full. This is also done top down, so we
1939 * have to be pessimistic.
1941 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 struct btrfs_path *path, int level)
1945 struct extent_buffer *right = NULL;
1946 struct extent_buffer *mid;
1947 struct extent_buffer *left = NULL;
1948 struct extent_buffer *parent = NULL;
1952 int orig_slot = path->slots[level];
1957 mid = path->nodes[level];
1958 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1960 if (level < BTRFS_MAX_LEVEL - 1) {
1961 parent = path->nodes[level + 1];
1962 pslot = path->slots[level + 1];
1968 left = read_node_slot(root, parent, pslot - 1);
1970 /* first, try to make some room in the middle buffer */
1974 btrfs_tree_lock(left);
1975 btrfs_set_lock_blocking(left);
1977 left_nr = btrfs_header_nritems(left);
1978 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1981 ret = btrfs_cow_block(trans, root, left, parent,
1986 wret = push_node_left(trans, root,
1993 struct btrfs_disk_key disk_key;
1994 orig_slot += left_nr;
1995 btrfs_node_key(mid, &disk_key, 0);
1996 tree_mod_log_set_node_key(root->fs_info, parent,
1998 btrfs_set_node_key(parent, &disk_key, pslot);
1999 btrfs_mark_buffer_dirty(parent);
2000 if (btrfs_header_nritems(left) > orig_slot) {
2001 path->nodes[level] = left;
2002 path->slots[level + 1] -= 1;
2003 path->slots[level] = orig_slot;
2004 btrfs_tree_unlock(mid);
2005 free_extent_buffer(mid);
2008 btrfs_header_nritems(left);
2009 path->slots[level] = orig_slot;
2010 btrfs_tree_unlock(left);
2011 free_extent_buffer(left);
2015 btrfs_tree_unlock(left);
2016 free_extent_buffer(left);
2018 right = read_node_slot(root, parent, pslot + 1);
2021 * then try to empty the right most buffer into the middle
2026 btrfs_tree_lock(right);
2027 btrfs_set_lock_blocking(right);
2029 right_nr = btrfs_header_nritems(right);
2030 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2033 ret = btrfs_cow_block(trans, root, right,
2039 wret = balance_node_right(trans, root,
2046 struct btrfs_disk_key disk_key;
2048 btrfs_node_key(right, &disk_key, 0);
2049 tree_mod_log_set_node_key(root->fs_info, parent,
2051 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2052 btrfs_mark_buffer_dirty(parent);
2054 if (btrfs_header_nritems(mid) <= orig_slot) {
2055 path->nodes[level] = right;
2056 path->slots[level + 1] += 1;
2057 path->slots[level] = orig_slot -
2058 btrfs_header_nritems(mid);
2059 btrfs_tree_unlock(mid);
2060 free_extent_buffer(mid);
2062 btrfs_tree_unlock(right);
2063 free_extent_buffer(right);
2067 btrfs_tree_unlock(right);
2068 free_extent_buffer(right);
2074 * readahead one full node of leaves, finding things that are close
2075 * to the block in 'slot', and triggering ra on them.
2077 static void reada_for_search(struct btrfs_root *root,
2078 struct btrfs_path *path,
2079 int level, int slot, u64 objectid)
2081 struct extent_buffer *node;
2082 struct btrfs_disk_key disk_key;
2088 int direction = path->reada;
2089 struct extent_buffer *eb;
2097 if (!path->nodes[level])
2100 node = path->nodes[level];
2102 search = btrfs_node_blockptr(node, slot);
2103 blocksize = btrfs_level_size(root, level - 1);
2104 eb = btrfs_find_tree_block(root, search, blocksize);
2106 free_extent_buffer(eb);
2112 nritems = btrfs_header_nritems(node);
2116 if (direction < 0) {
2120 } else if (direction > 0) {
2125 if (path->reada < 0 && objectid) {
2126 btrfs_node_key(node, &disk_key, nr);
2127 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2130 search = btrfs_node_blockptr(node, nr);
2131 if ((search <= target && target - search <= 65536) ||
2132 (search > target && search - target <= 65536)) {
2133 gen = btrfs_node_ptr_generation(node, nr);
2134 readahead_tree_block(root, search, blocksize, gen);
2138 if ((nread > 65536 || nscan > 32))
2143 static noinline void reada_for_balance(struct btrfs_root *root,
2144 struct btrfs_path *path, int level)
2148 struct extent_buffer *parent;
2149 struct extent_buffer *eb;
2155 parent = path->nodes[level + 1];
2159 nritems = btrfs_header_nritems(parent);
2160 slot = path->slots[level + 1];
2161 blocksize = btrfs_level_size(root, level);
2164 block1 = btrfs_node_blockptr(parent, slot - 1);
2165 gen = btrfs_node_ptr_generation(parent, slot - 1);
2166 eb = btrfs_find_tree_block(root, block1, blocksize);
2168 * if we get -eagain from btrfs_buffer_uptodate, we
2169 * don't want to return eagain here. That will loop
2172 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2174 free_extent_buffer(eb);
2176 if (slot + 1 < nritems) {
2177 block2 = btrfs_node_blockptr(parent, slot + 1);
2178 gen = btrfs_node_ptr_generation(parent, slot + 1);
2179 eb = btrfs_find_tree_block(root, block2, blocksize);
2180 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2182 free_extent_buffer(eb);
2186 readahead_tree_block(root, block1, blocksize, 0);
2188 readahead_tree_block(root, block2, blocksize, 0);
2193 * when we walk down the tree, it is usually safe to unlock the higher layers
2194 * in the tree. The exceptions are when our path goes through slot 0, because
2195 * operations on the tree might require changing key pointers higher up in the
2198 * callers might also have set path->keep_locks, which tells this code to keep
2199 * the lock if the path points to the last slot in the block. This is part of
2200 * walking through the tree, and selecting the next slot in the higher block.
2202 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2203 * if lowest_unlock is 1, level 0 won't be unlocked
2205 static noinline void unlock_up(struct btrfs_path *path, int level,
2206 int lowest_unlock, int min_write_lock_level,
2207 int *write_lock_level)
2210 int skip_level = level;
2212 struct extent_buffer *t;
2214 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2215 if (!path->nodes[i])
2217 if (!path->locks[i])
2219 if (!no_skips && path->slots[i] == 0) {
2223 if (!no_skips && path->keep_locks) {
2226 nritems = btrfs_header_nritems(t);
2227 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2232 if (skip_level < i && i >= lowest_unlock)
2236 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2237 btrfs_tree_unlock_rw(t, path->locks[i]);
2239 if (write_lock_level &&
2240 i > min_write_lock_level &&
2241 i <= *write_lock_level) {
2242 *write_lock_level = i - 1;
2249 * This releases any locks held in the path starting at level and
2250 * going all the way up to the root.
2252 * btrfs_search_slot will keep the lock held on higher nodes in a few
2253 * corner cases, such as COW of the block at slot zero in the node. This
2254 * ignores those rules, and it should only be called when there are no
2255 * more updates to be done higher up in the tree.
2257 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2261 if (path->keep_locks)
2264 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2265 if (!path->nodes[i])
2267 if (!path->locks[i])
2269 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2275 * helper function for btrfs_search_slot. The goal is to find a block
2276 * in cache without setting the path to blocking. If we find the block
2277 * we return zero and the path is unchanged.
2279 * If we can't find the block, we set the path blocking and do some
2280 * reada. -EAGAIN is returned and the search must be repeated.
2283 read_block_for_search(struct btrfs_trans_handle *trans,
2284 struct btrfs_root *root, struct btrfs_path *p,
2285 struct extent_buffer **eb_ret, int level, int slot,
2286 struct btrfs_key *key, u64 time_seq)
2291 struct extent_buffer *b = *eb_ret;
2292 struct extent_buffer *tmp;
2295 blocknr = btrfs_node_blockptr(b, slot);
2296 gen = btrfs_node_ptr_generation(b, slot);
2297 blocksize = btrfs_level_size(root, level - 1);
2299 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2301 /* first we do an atomic uptodate check */
2302 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2307 /* the pages were up to date, but we failed
2308 * the generation number check. Do a full
2309 * read for the generation number that is correct.
2310 * We must do this without dropping locks so
2311 * we can trust our generation number
2313 btrfs_set_path_blocking(p);
2315 /* now we're allowed to do a blocking uptodate check */
2316 ret = btrfs_read_buffer(tmp, gen);
2321 free_extent_buffer(tmp);
2322 btrfs_release_path(p);
2327 * reduce lock contention at high levels
2328 * of the btree by dropping locks before
2329 * we read. Don't release the lock on the current
2330 * level because we need to walk this node to figure
2331 * out which blocks to read.
2333 btrfs_unlock_up_safe(p, level + 1);
2334 btrfs_set_path_blocking(p);
2336 free_extent_buffer(tmp);
2338 reada_for_search(root, p, level, slot, key->objectid);
2340 btrfs_release_path(p);
2343 tmp = read_tree_block(root, blocknr, blocksize, 0);
2346 * If the read above didn't mark this buffer up to date,
2347 * it will never end up being up to date. Set ret to EIO now
2348 * and give up so that our caller doesn't loop forever
2351 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2353 free_extent_buffer(tmp);
2359 * helper function for btrfs_search_slot. This does all of the checks
2360 * for node-level blocks and does any balancing required based on
2363 * If no extra work was required, zero is returned. If we had to
2364 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2368 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2369 struct btrfs_root *root, struct btrfs_path *p,
2370 struct extent_buffer *b, int level, int ins_len,
2371 int *write_lock_level)
2374 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2375 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2378 if (*write_lock_level < level + 1) {
2379 *write_lock_level = level + 1;
2380 btrfs_release_path(p);
2384 btrfs_set_path_blocking(p);
2385 reada_for_balance(root, p, level);
2386 sret = split_node(trans, root, p, level);
2387 btrfs_clear_path_blocking(p, NULL, 0);
2394 b = p->nodes[level];
2395 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2396 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2399 if (*write_lock_level < level + 1) {
2400 *write_lock_level = level + 1;
2401 btrfs_release_path(p);
2405 btrfs_set_path_blocking(p);
2406 reada_for_balance(root, p, level);
2407 sret = balance_level(trans, root, p, level);
2408 btrfs_clear_path_blocking(p, NULL, 0);
2414 b = p->nodes[level];
2416 btrfs_release_path(p);
2419 BUG_ON(btrfs_header_nritems(b) == 1);
2429 static void key_search_validate(struct extent_buffer *b,
2430 struct btrfs_key *key,
2433 #ifdef CONFIG_BTRFS_ASSERT
2434 struct btrfs_disk_key disk_key;
2436 btrfs_cpu_key_to_disk(&disk_key, key);
2439 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2440 offsetof(struct btrfs_leaf, items[0].key),
2443 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2444 offsetof(struct btrfs_node, ptrs[0].key),
2449 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2450 int level, int *prev_cmp, int *slot)
2452 if (*prev_cmp != 0) {
2453 *prev_cmp = bin_search(b, key, level, slot);
2457 key_search_validate(b, key, level);
2464 * look for key in the tree. path is filled in with nodes along the way
2465 * if key is found, we return zero and you can find the item in the leaf
2466 * level of the path (level 0)
2468 * If the key isn't found, the path points to the slot where it should
2469 * be inserted, and 1 is returned. If there are other errors during the
2470 * search a negative error number is returned.
2472 * if ins_len > 0, nodes and leaves will be split as we walk down the
2473 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2476 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2477 *root, struct btrfs_key *key, struct btrfs_path *p, int
2480 struct extent_buffer *b;
2485 int lowest_unlock = 1;
2487 /* everything at write_lock_level or lower must be write locked */
2488 int write_lock_level = 0;
2489 u8 lowest_level = 0;
2490 int min_write_lock_level;
2493 lowest_level = p->lowest_level;
2494 WARN_ON(lowest_level && ins_len > 0);
2495 WARN_ON(p->nodes[0] != NULL);
2500 /* when we are removing items, we might have to go up to level
2501 * two as we update tree pointers Make sure we keep write
2502 * for those levels as well
2504 write_lock_level = 2;
2505 } else if (ins_len > 0) {
2507 * for inserting items, make sure we have a write lock on
2508 * level 1 so we can update keys
2510 write_lock_level = 1;
2514 write_lock_level = -1;
2516 if (cow && (p->keep_locks || p->lowest_level))
2517 write_lock_level = BTRFS_MAX_LEVEL;
2519 min_write_lock_level = write_lock_level;
2524 * we try very hard to do read locks on the root
2526 root_lock = BTRFS_READ_LOCK;
2528 if (p->search_commit_root) {
2530 * the commit roots are read only
2531 * so we always do read locks
2533 b = root->commit_root;
2534 extent_buffer_get(b);
2535 level = btrfs_header_level(b);
2536 if (!p->skip_locking)
2537 btrfs_tree_read_lock(b);
2539 if (p->skip_locking) {
2540 b = btrfs_root_node(root);
2541 level = btrfs_header_level(b);
2543 /* we don't know the level of the root node
2544 * until we actually have it read locked
2546 b = btrfs_read_lock_root_node(root);
2547 level = btrfs_header_level(b);
2548 if (level <= write_lock_level) {
2549 /* whoops, must trade for write lock */
2550 btrfs_tree_read_unlock(b);
2551 free_extent_buffer(b);
2552 b = btrfs_lock_root_node(root);
2553 root_lock = BTRFS_WRITE_LOCK;
2555 /* the level might have changed, check again */
2556 level = btrfs_header_level(b);
2560 p->nodes[level] = b;
2561 if (!p->skip_locking)
2562 p->locks[level] = root_lock;
2565 level = btrfs_header_level(b);
2568 * setup the path here so we can release it under lock
2569 * contention with the cow code
2573 * if we don't really need to cow this block
2574 * then we don't want to set the path blocking,
2575 * so we test it here
2577 if (!should_cow_block(trans, root, b))
2580 btrfs_set_path_blocking(p);
2583 * must have write locks on this node and the
2586 if (level > write_lock_level ||
2587 (level + 1 > write_lock_level &&
2588 level + 1 < BTRFS_MAX_LEVEL &&
2589 p->nodes[level + 1])) {
2590 write_lock_level = level + 1;
2591 btrfs_release_path(p);
2595 err = btrfs_cow_block(trans, root, b,
2596 p->nodes[level + 1],
2597 p->slots[level + 1], &b);
2604 BUG_ON(!cow && ins_len);
2606 p->nodes[level] = b;
2607 btrfs_clear_path_blocking(p, NULL, 0);
2610 * we have a lock on b and as long as we aren't changing
2611 * the tree, there is no way to for the items in b to change.
2612 * It is safe to drop the lock on our parent before we
2613 * go through the expensive btree search on b.
2615 * If cow is true, then we might be changing slot zero,
2616 * which may require changing the parent. So, we can't
2617 * drop the lock until after we know which slot we're
2621 btrfs_unlock_up_safe(p, level + 1);
2623 ret = key_search(b, key, level, &prev_cmp, &slot);
2627 if (ret && slot > 0) {
2631 p->slots[level] = slot;
2632 err = setup_nodes_for_search(trans, root, p, b, level,
2633 ins_len, &write_lock_level);
2640 b = p->nodes[level];
2641 slot = p->slots[level];
2644 * slot 0 is special, if we change the key
2645 * we have to update the parent pointer
2646 * which means we must have a write lock
2649 if (slot == 0 && cow &&
2650 write_lock_level < level + 1) {
2651 write_lock_level = level + 1;
2652 btrfs_release_path(p);
2656 unlock_up(p, level, lowest_unlock,
2657 min_write_lock_level, &write_lock_level);
2659 if (level == lowest_level) {
2665 err = read_block_for_search(trans, root, p,
2666 &b, level, slot, key, 0);
2674 if (!p->skip_locking) {
2675 level = btrfs_header_level(b);
2676 if (level <= write_lock_level) {
2677 err = btrfs_try_tree_write_lock(b);
2679 btrfs_set_path_blocking(p);
2681 btrfs_clear_path_blocking(p, b,
2684 p->locks[level] = BTRFS_WRITE_LOCK;
2686 err = btrfs_try_tree_read_lock(b);
2688 btrfs_set_path_blocking(p);
2689 btrfs_tree_read_lock(b);
2690 btrfs_clear_path_blocking(p, b,
2693 p->locks[level] = BTRFS_READ_LOCK;
2695 p->nodes[level] = b;
2698 p->slots[level] = slot;
2700 btrfs_leaf_free_space(root, b) < ins_len) {
2701 if (write_lock_level < 1) {
2702 write_lock_level = 1;
2703 btrfs_release_path(p);
2707 btrfs_set_path_blocking(p);
2708 err = split_leaf(trans, root, key,
2709 p, ins_len, ret == 0);
2710 btrfs_clear_path_blocking(p, NULL, 0);
2718 if (!p->search_for_split)
2719 unlock_up(p, level, lowest_unlock,
2720 min_write_lock_level, &write_lock_level);
2727 * we don't really know what they plan on doing with the path
2728 * from here on, so for now just mark it as blocking
2730 if (!p->leave_spinning)
2731 btrfs_set_path_blocking(p);
2733 btrfs_release_path(p);
2738 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2739 * current state of the tree together with the operations recorded in the tree
2740 * modification log to search for the key in a previous version of this tree, as
2741 * denoted by the time_seq parameter.
2743 * Naturally, there is no support for insert, delete or cow operations.
2745 * The resulting path and return value will be set up as if we called
2746 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2748 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2749 struct btrfs_path *p, u64 time_seq)
2751 struct extent_buffer *b;
2756 int lowest_unlock = 1;
2757 u8 lowest_level = 0;
2760 lowest_level = p->lowest_level;
2761 WARN_ON(p->nodes[0] != NULL);
2763 if (p->search_commit_root) {
2765 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2770 b = get_old_root(root, time_seq);
2771 level = btrfs_header_level(b);
2772 p->locks[level] = BTRFS_READ_LOCK;
2775 level = btrfs_header_level(b);
2776 p->nodes[level] = b;
2777 btrfs_clear_path_blocking(p, NULL, 0);
2780 * we have a lock on b and as long as we aren't changing
2781 * the tree, there is no way to for the items in b to change.
2782 * It is safe to drop the lock on our parent before we
2783 * go through the expensive btree search on b.
2785 btrfs_unlock_up_safe(p, level + 1);
2787 ret = key_search(b, key, level, &prev_cmp, &slot);
2791 if (ret && slot > 0) {
2795 p->slots[level] = slot;
2796 unlock_up(p, level, lowest_unlock, 0, NULL);
2798 if (level == lowest_level) {
2804 err = read_block_for_search(NULL, root, p, &b, level,
2805 slot, key, time_seq);
2813 level = btrfs_header_level(b);
2814 err = btrfs_try_tree_read_lock(b);
2816 btrfs_set_path_blocking(p);
2817 btrfs_tree_read_lock(b);
2818 btrfs_clear_path_blocking(p, b,
2821 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
2826 p->locks[level] = BTRFS_READ_LOCK;
2827 p->nodes[level] = b;
2829 p->slots[level] = slot;
2830 unlock_up(p, level, lowest_unlock, 0, NULL);
2836 if (!p->leave_spinning)
2837 btrfs_set_path_blocking(p);
2839 btrfs_release_path(p);
2845 * helper to use instead of search slot if no exact match is needed but
2846 * instead the next or previous item should be returned.
2847 * When find_higher is true, the next higher item is returned, the next lower
2849 * When return_any and find_higher are both true, and no higher item is found,
2850 * return the next lower instead.
2851 * When return_any is true and find_higher is false, and no lower item is found,
2852 * return the next higher instead.
2853 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2856 int btrfs_search_slot_for_read(struct btrfs_root *root,
2857 struct btrfs_key *key, struct btrfs_path *p,
2858 int find_higher, int return_any)
2861 struct extent_buffer *leaf;
2864 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2868 * a return value of 1 means the path is at the position where the
2869 * item should be inserted. Normally this is the next bigger item,
2870 * but in case the previous item is the last in a leaf, path points
2871 * to the first free slot in the previous leaf, i.e. at an invalid
2877 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2878 ret = btrfs_next_leaf(root, p);
2884 * no higher item found, return the next
2889 btrfs_release_path(p);
2893 if (p->slots[0] == 0) {
2894 ret = btrfs_prev_leaf(root, p);
2898 p->slots[0] = btrfs_header_nritems(leaf) - 1;
2904 * no lower item found, return the next
2909 btrfs_release_path(p);
2919 * adjust the pointers going up the tree, starting at level
2920 * making sure the right key of each node is points to 'key'.
2921 * This is used after shifting pointers to the left, so it stops
2922 * fixing up pointers when a given leaf/node is not in slot 0 of the
2926 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
2927 struct btrfs_disk_key *key, int level)
2930 struct extent_buffer *t;
2932 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2933 int tslot = path->slots[i];
2934 if (!path->nodes[i])
2937 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
2938 btrfs_set_node_key(t, key, tslot);
2939 btrfs_mark_buffer_dirty(path->nodes[i]);
2948 * This function isn't completely safe. It's the caller's responsibility
2949 * that the new key won't break the order
2951 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
2952 struct btrfs_key *new_key)
2954 struct btrfs_disk_key disk_key;
2955 struct extent_buffer *eb;
2958 eb = path->nodes[0];
2959 slot = path->slots[0];
2961 btrfs_item_key(eb, &disk_key, slot - 1);
2962 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
2964 if (slot < btrfs_header_nritems(eb) - 1) {
2965 btrfs_item_key(eb, &disk_key, slot + 1);
2966 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
2969 btrfs_cpu_key_to_disk(&disk_key, new_key);
2970 btrfs_set_item_key(eb, &disk_key, slot);
2971 btrfs_mark_buffer_dirty(eb);
2973 fixup_low_keys(root, path, &disk_key, 1);
2977 * try to push data from one node into the next node left in the
2980 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2981 * error, and > 0 if there was no room in the left hand block.
2983 static int push_node_left(struct btrfs_trans_handle *trans,
2984 struct btrfs_root *root, struct extent_buffer *dst,
2985 struct extent_buffer *src, int empty)
2992 src_nritems = btrfs_header_nritems(src);
2993 dst_nritems = btrfs_header_nritems(dst);
2994 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2995 WARN_ON(btrfs_header_generation(src) != trans->transid);
2996 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2998 if (!empty && src_nritems <= 8)
3001 if (push_items <= 0)
3005 push_items = min(src_nritems, push_items);
3006 if (push_items < src_nritems) {
3007 /* leave at least 8 pointers in the node if
3008 * we aren't going to empty it
3010 if (src_nritems - push_items < 8) {
3011 if (push_items <= 8)
3017 push_items = min(src_nritems - 8, push_items);
3019 tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3021 copy_extent_buffer(dst, src,
3022 btrfs_node_key_ptr_offset(dst_nritems),
3023 btrfs_node_key_ptr_offset(0),
3024 push_items * sizeof(struct btrfs_key_ptr));
3026 if (push_items < src_nritems) {
3028 * don't call tree_mod_log_eb_move here, key removal was already
3029 * fully logged by tree_mod_log_eb_copy above.
3031 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3032 btrfs_node_key_ptr_offset(push_items),
3033 (src_nritems - push_items) *
3034 sizeof(struct btrfs_key_ptr));
3036 btrfs_set_header_nritems(src, src_nritems - push_items);
3037 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3038 btrfs_mark_buffer_dirty(src);
3039 btrfs_mark_buffer_dirty(dst);
3045 * try to push data from one node into the next node right in the
3048 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3049 * error, and > 0 if there was no room in the right hand block.
3051 * this will only push up to 1/2 the contents of the left node over
3053 static int balance_node_right(struct btrfs_trans_handle *trans,
3054 struct btrfs_root *root,
3055 struct extent_buffer *dst,
3056 struct extent_buffer *src)
3064 WARN_ON(btrfs_header_generation(src) != trans->transid);
3065 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3067 src_nritems = btrfs_header_nritems(src);
3068 dst_nritems = btrfs_header_nritems(dst);
3069 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3070 if (push_items <= 0)
3073 if (src_nritems < 4)
3076 max_push = src_nritems / 2 + 1;
3077 /* don't try to empty the node */
3078 if (max_push >= src_nritems)
3081 if (max_push < push_items)
3082 push_items = max_push;
3084 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3085 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3086 btrfs_node_key_ptr_offset(0),
3088 sizeof(struct btrfs_key_ptr));
3090 tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3091 src_nritems - push_items, push_items);
3092 copy_extent_buffer(dst, src,
3093 btrfs_node_key_ptr_offset(0),
3094 btrfs_node_key_ptr_offset(src_nritems - push_items),
3095 push_items * sizeof(struct btrfs_key_ptr));
3097 btrfs_set_header_nritems(src, src_nritems - push_items);
3098 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3100 btrfs_mark_buffer_dirty(src);
3101 btrfs_mark_buffer_dirty(dst);
3107 * helper function to insert a new root level in the tree.
3108 * A new node is allocated, and a single item is inserted to
3109 * point to the existing root
3111 * returns zero on success or < 0 on failure.
3113 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3114 struct btrfs_root *root,
3115 struct btrfs_path *path, int level)
3118 struct extent_buffer *lower;
3119 struct extent_buffer *c;
3120 struct extent_buffer *old;
3121 struct btrfs_disk_key lower_key;
3123 BUG_ON(path->nodes[level]);
3124 BUG_ON(path->nodes[level-1] != root->node);
3126 lower = path->nodes[level-1];
3128 btrfs_item_key(lower, &lower_key, 0);
3130 btrfs_node_key(lower, &lower_key, 0);
3132 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3133 root->root_key.objectid, &lower_key,
3134 level, root->node->start, 0);
3138 root_add_used(root, root->nodesize);
3140 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3141 btrfs_set_header_nritems(c, 1);
3142 btrfs_set_header_level(c, level);
3143 btrfs_set_header_bytenr(c, c->start);
3144 btrfs_set_header_generation(c, trans->transid);
3145 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3146 btrfs_set_header_owner(c, root->root_key.objectid);
3148 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(c),
3151 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3152 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3154 btrfs_set_node_key(c, &lower_key, 0);
3155 btrfs_set_node_blockptr(c, 0, lower->start);
3156 lower_gen = btrfs_header_generation(lower);
3157 WARN_ON(lower_gen != trans->transid);
3159 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3161 btrfs_mark_buffer_dirty(c);
3164 tree_mod_log_set_root_pointer(root, c, 0);
3165 rcu_assign_pointer(root->node, c);
3167 /* the super has an extra ref to root->node */
3168 free_extent_buffer(old);
3170 add_root_to_dirty_list(root);
3171 extent_buffer_get(c);
3172 path->nodes[level] = c;
3173 path->locks[level] = BTRFS_WRITE_LOCK;
3174 path->slots[level] = 0;
3179 * worker function to insert a single pointer in a node.
3180 * the node should have enough room for the pointer already
3182 * slot and level indicate where you want the key to go, and
3183 * blocknr is the block the key points to.
3185 static void insert_ptr(struct btrfs_trans_handle *trans,
3186 struct btrfs_root *root, struct btrfs_path *path,
3187 struct btrfs_disk_key *key, u64 bytenr,
3188 int slot, int level)
3190 struct extent_buffer *lower;
3194 BUG_ON(!path->nodes[level]);
3195 btrfs_assert_tree_locked(path->nodes[level]);
3196 lower = path->nodes[level];
3197 nritems = btrfs_header_nritems(lower);
3198 BUG_ON(slot > nritems);
3199 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3200 if (slot != nritems) {
3202 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3203 slot, nritems - slot);
3204 memmove_extent_buffer(lower,
3205 btrfs_node_key_ptr_offset(slot + 1),
3206 btrfs_node_key_ptr_offset(slot),
3207 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3210 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3211 MOD_LOG_KEY_ADD, GFP_NOFS);
3214 btrfs_set_node_key(lower, key, slot);
3215 btrfs_set_node_blockptr(lower, slot, bytenr);
3216 WARN_ON(trans->transid == 0);
3217 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3218 btrfs_set_header_nritems(lower, nritems + 1);
3219 btrfs_mark_buffer_dirty(lower);
3223 * split the node at the specified level in path in two.
3224 * The path is corrected to point to the appropriate node after the split
3226 * Before splitting this tries to make some room in the node by pushing
3227 * left and right, if either one works, it returns right away.
3229 * returns 0 on success and < 0 on failure
3231 static noinline int split_node(struct btrfs_trans_handle *trans,
3232 struct btrfs_root *root,
3233 struct btrfs_path *path, int level)
3235 struct extent_buffer *c;
3236 struct extent_buffer *split;
3237 struct btrfs_disk_key disk_key;
3242 c = path->nodes[level];
3243 WARN_ON(btrfs_header_generation(c) != trans->transid);
3244 if (c == root->node) {
3246 * trying to split the root, lets make a new one
3248 * tree mod log: We don't log_removal old root in
3249 * insert_new_root, because that root buffer will be kept as a
3250 * normal node. We are going to log removal of half of the
3251 * elements below with tree_mod_log_eb_copy. We're holding a
3252 * tree lock on the buffer, which is why we cannot race with
3253 * other tree_mod_log users.
3255 ret = insert_new_root(trans, root, path, level + 1);
3259 ret = push_nodes_for_insert(trans, root, path, level);
3260 c = path->nodes[level];
3261 if (!ret && btrfs_header_nritems(c) <
3262 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3268 c_nritems = btrfs_header_nritems(c);
3269 mid = (c_nritems + 1) / 2;
3270 btrfs_node_key(c, &disk_key, mid);
3272 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3273 root->root_key.objectid,
3274 &disk_key, level, c->start, 0);
3276 return PTR_ERR(split);
3278 root_add_used(root, root->nodesize);
3280 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3281 btrfs_set_header_level(split, btrfs_header_level(c));
3282 btrfs_set_header_bytenr(split, split->start);
3283 btrfs_set_header_generation(split, trans->transid);
3284 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3285 btrfs_set_header_owner(split, root->root_key.objectid);
3286 write_extent_buffer(split, root->fs_info->fsid,
3287 btrfs_header_fsid(split), BTRFS_FSID_SIZE);
3288 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3289 btrfs_header_chunk_tree_uuid(split),
3292 tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid);
3293 copy_extent_buffer(split, c,
3294 btrfs_node_key_ptr_offset(0),
3295 btrfs_node_key_ptr_offset(mid),
3296 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3297 btrfs_set_header_nritems(split, c_nritems - mid);
3298 btrfs_set_header_nritems(c, mid);
3301 btrfs_mark_buffer_dirty(c);
3302 btrfs_mark_buffer_dirty(split);
3304 insert_ptr(trans, root, path, &disk_key, split->start,
3305 path->slots[level + 1] + 1, level + 1);
3307 if (path->slots[level] >= mid) {
3308 path->slots[level] -= mid;
3309 btrfs_tree_unlock(c);
3310 free_extent_buffer(c);
3311 path->nodes[level] = split;
3312 path->slots[level + 1] += 1;
3314 btrfs_tree_unlock(split);
3315 free_extent_buffer(split);
3321 * how many bytes are required to store the items in a leaf. start
3322 * and nr indicate which items in the leaf to check. This totals up the
3323 * space used both by the item structs and the item data
3325 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3327 struct btrfs_item *start_item;
3328 struct btrfs_item *end_item;
3329 struct btrfs_map_token token;
3331 int nritems = btrfs_header_nritems(l);
3332 int end = min(nritems, start + nr) - 1;
3336 btrfs_init_map_token(&token);
3337 start_item = btrfs_item_nr(l, start);
3338 end_item = btrfs_item_nr(l, end);
3339 data_len = btrfs_token_item_offset(l, start_item, &token) +
3340 btrfs_token_item_size(l, start_item, &token);
3341 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3342 data_len += sizeof(struct btrfs_item) * nr;
3343 WARN_ON(data_len < 0);
3348 * The space between the end of the leaf items and
3349 * the start of the leaf data. IOW, how much room
3350 * the leaf has left for both items and data
3352 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3353 struct extent_buffer *leaf)
3355 int nritems = btrfs_header_nritems(leaf);
3357 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3359 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
3360 "used %d nritems %d\n",
3361 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3362 leaf_space_used(leaf, 0, nritems), nritems);
3368 * min slot controls the lowest index we're willing to push to the
3369 * right. We'll push up to and including min_slot, but no lower
3371 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3372 struct btrfs_root *root,
3373 struct btrfs_path *path,
3374 int data_size, int empty,
3375 struct extent_buffer *right,
3376 int free_space, u32 left_nritems,
3379 struct extent_buffer *left = path->nodes[0];
3380 struct extent_buffer *upper = path->nodes[1];
3381 struct btrfs_map_token token;
3382 struct btrfs_disk_key disk_key;
3387 struct btrfs_item *item;
3393 btrfs_init_map_token(&token);
3398 nr = max_t(u32, 1, min_slot);
3400 if (path->slots[0] >= left_nritems)
3401 push_space += data_size;
3403 slot = path->slots[1];
3404 i = left_nritems - 1;
3406 item = btrfs_item_nr(left, i);
3408 if (!empty && push_items > 0) {
3409 if (path->slots[0] > i)
3411 if (path->slots[0] == i) {
3412 int space = btrfs_leaf_free_space(root, left);
3413 if (space + push_space * 2 > free_space)
3418 if (path->slots[0] == i)
3419 push_space += data_size;
3421 this_item_size = btrfs_item_size(left, item);
3422 if (this_item_size + sizeof(*item) + push_space > free_space)
3426 push_space += this_item_size + sizeof(*item);
3432 if (push_items == 0)
3435 WARN_ON(!empty && push_items == left_nritems);
3437 /* push left to right */
3438 right_nritems = btrfs_header_nritems(right);
3440 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3441 push_space -= leaf_data_end(root, left);
3443 /* make room in the right data area */
3444 data_end = leaf_data_end(root, right);
3445 memmove_extent_buffer(right,
3446 btrfs_leaf_data(right) + data_end - push_space,
3447 btrfs_leaf_data(right) + data_end,
3448 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3450 /* copy from the left data area */
3451 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3452 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3453 btrfs_leaf_data(left) + leaf_data_end(root, left),
3456 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3457 btrfs_item_nr_offset(0),
3458 right_nritems * sizeof(struct btrfs_item));
3460 /* copy the items from left to right */
3461 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3462 btrfs_item_nr_offset(left_nritems - push_items),
3463 push_items * sizeof(struct btrfs_item));
3465 /* update the item pointers */
3466 right_nritems += push_items;
3467 btrfs_set_header_nritems(right, right_nritems);
3468 push_space = BTRFS_LEAF_DATA_SIZE(root);
3469 for (i = 0; i < right_nritems; i++) {
3470 item = btrfs_item_nr(right, i);
3471 push_space -= btrfs_token_item_size(right, item, &token);
3472 btrfs_set_token_item_offset(right, item, push_space, &token);
3475 left_nritems -= push_items;
3476 btrfs_set_header_nritems(left, left_nritems);
3479 btrfs_mark_buffer_dirty(left);
3481 clean_tree_block(trans, root, left);
3483 btrfs_mark_buffer_dirty(right);
3485 btrfs_item_key(right, &disk_key, 0);
3486 btrfs_set_node_key(upper, &disk_key, slot + 1);
3487 btrfs_mark_buffer_dirty(upper);
3489 /* then fixup the leaf pointer in the path */
3490 if (path->slots[0] >= left_nritems) {
3491 path->slots[0] -= left_nritems;
3492 if (btrfs_header_nritems(path->nodes[0]) == 0)
3493 clean_tree_block(trans, root, path->nodes[0]);
3494 btrfs_tree_unlock(path->nodes[0]);
3495 free_extent_buffer(path->nodes[0]);
3496 path->nodes[0] = right;
3497 path->slots[1] += 1;
3499 btrfs_tree_unlock(right);
3500 free_extent_buffer(right);
3505 btrfs_tree_unlock(right);
3506 free_extent_buffer(right);
3511 * push some data in the path leaf to the right, trying to free up at
3512 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3514 * returns 1 if the push failed because the other node didn't have enough
3515 * room, 0 if everything worked out and < 0 if there were major errors.
3517 * this will push starting from min_slot to the end of the leaf. It won't
3518 * push any slot lower than min_slot
3520 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3521 *root, struct btrfs_path *path,
3522 int min_data_size, int data_size,
3523 int empty, u32 min_slot)
3525 struct extent_buffer *left = path->nodes[0];
3526 struct extent_buffer *right;
3527 struct extent_buffer *upper;
3533 if (!path->nodes[1])
3536 slot = path->slots[1];
3537 upper = path->nodes[1];
3538 if (slot >= btrfs_header_nritems(upper) - 1)
3541 btrfs_assert_tree_locked(path->nodes[1]);
3543 right = read_node_slot(root, upper, slot + 1);
3547 btrfs_tree_lock(right);
3548 btrfs_set_lock_blocking(right);
3550 free_space = btrfs_leaf_free_space(root, right);
3551 if (free_space < data_size)
3554 /* cow and double check */
3555 ret = btrfs_cow_block(trans, root, right, upper,
3560 free_space = btrfs_leaf_free_space(root, right);
3561 if (free_space < data_size)
3564 left_nritems = btrfs_header_nritems(left);
3565 if (left_nritems == 0)
3568 return __push_leaf_right(trans, root, path, min_data_size, empty,
3569 right, free_space, left_nritems, min_slot);
3571 btrfs_tree_unlock(right);
3572 free_extent_buffer(right);
3577 * push some data in the path leaf to the left, trying to free up at
3578 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3580 * max_slot can put a limit on how far into the leaf we'll push items. The
3581 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3584 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3585 struct btrfs_root *root,
3586 struct btrfs_path *path, int data_size,
3587 int empty, struct extent_buffer *left,
3588 int free_space, u32 right_nritems,
3591 struct btrfs_disk_key disk_key;
3592 struct extent_buffer *right = path->nodes[0];
3596 struct btrfs_item *item;
3597 u32 old_left_nritems;
3601 u32 old_left_item_size;
3602 struct btrfs_map_token token;
3604 btrfs_init_map_token(&token);
3607 nr = min(right_nritems, max_slot);
3609 nr = min(right_nritems - 1, max_slot);
3611 for (i = 0; i < nr; i++) {
3612 item = btrfs_item_nr(right, i);
3614 if (!empty && push_items > 0) {
3615 if (path->slots[0] < i)
3617 if (path->slots[0] == i) {
3618 int space = btrfs_leaf_free_space(root, right);
3619 if (space + push_space * 2 > free_space)
3624 if (path->slots[0] == i)
3625 push_space += data_size;
3627 this_item_size = btrfs_item_size(right, item);
3628 if (this_item_size + sizeof(*item) + push_space > free_space)
3632 push_space += this_item_size + sizeof(*item);
3635 if (push_items == 0) {
3639 if (!empty && push_items == btrfs_header_nritems(right))
3642 /* push data from right to left */
3643 copy_extent_buffer(left, right,
3644 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3645 btrfs_item_nr_offset(0),
3646 push_items * sizeof(struct btrfs_item));
3648 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3649 btrfs_item_offset_nr(right, push_items - 1);
3651 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3652 leaf_data_end(root, left) - push_space,
3653 btrfs_leaf_data(right) +
3654 btrfs_item_offset_nr(right, push_items - 1),
3656 old_left_nritems = btrfs_header_nritems(left);
3657 BUG_ON(old_left_nritems <= 0);
3659 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3660 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3663 item = btrfs_item_nr(left, i);
3665 ioff = btrfs_token_item_offset(left, item, &token);
3666 btrfs_set_token_item_offset(left, item,
3667 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3670 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3672 /* fixup right node */
3673 if (push_items > right_nritems)
3674 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3677 if (push_items < right_nritems) {
3678 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3679 leaf_data_end(root, right);
3680 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3681 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3682 btrfs_leaf_data(right) +
3683 leaf_data_end(root, right), push_space);
3685 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3686 btrfs_item_nr_offset(push_items),
3687 (btrfs_header_nritems(right) - push_items) *
3688 sizeof(struct btrfs_item));
3690 right_nritems -= push_items;
3691 btrfs_set_header_nritems(right, right_nritems);
3692 push_space = BTRFS_LEAF_DATA_SIZE(root);
3693 for (i = 0; i < right_nritems; i++) {
3694 item = btrfs_item_nr(right, i);
3696 push_space = push_space - btrfs_token_item_size(right,
3698 btrfs_set_token_item_offset(right, item, push_space, &token);
3701 btrfs_mark_buffer_dirty(left);
3703 btrfs_mark_buffer_dirty(right);
3705 clean_tree_block(trans, root, right);
3707 btrfs_item_key(right, &disk_key, 0);
3708 fixup_low_keys(root, path, &disk_key, 1);
3710 /* then fixup the leaf pointer in the path */
3711 if (path->slots[0] < push_items) {
3712 path->slots[0] += old_left_nritems;
3713 btrfs_tree_unlock(path->nodes[0]);
3714 free_extent_buffer(path->nodes[0]);
3715 path->nodes[0] = left;
3716 path->slots[1] -= 1;
3718 btrfs_tree_unlock(left);
3719 free_extent_buffer(left);
3720 path->slots[0] -= push_items;
3722 BUG_ON(path->slots[0] < 0);
3725 btrfs_tree_unlock(left);
3726 free_extent_buffer(left);
3731 * push some data in the path leaf to the left, trying to free up at
3732 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3734 * max_slot can put a limit on how far into the leaf we'll push items. The
3735 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3738 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3739 *root, struct btrfs_path *path, int min_data_size,
3740 int data_size, int empty, u32 max_slot)
3742 struct extent_buffer *right = path->nodes[0];
3743 struct extent_buffer *left;
3749 slot = path->slots[1];
3752 if (!path->nodes[1])
3755 right_nritems = btrfs_header_nritems(right);
3756 if (right_nritems == 0)
3759 btrfs_assert_tree_locked(path->nodes[1]);
3761 left = read_node_slot(root, path->nodes[1], slot - 1);
3765 btrfs_tree_lock(left);
3766 btrfs_set_lock_blocking(left);
3768 free_space = btrfs_leaf_free_space(root, left);
3769 if (free_space < data_size) {
3774 /* cow and double check */
3775 ret = btrfs_cow_block(trans, root, left,
3776 path->nodes[1], slot - 1, &left);
3778 /* we hit -ENOSPC, but it isn't fatal here */
3784 free_space = btrfs_leaf_free_space(root, left);
3785 if (free_space < data_size) {
3790 return __push_leaf_left(trans, root, path, min_data_size,
3791 empty, left, free_space, right_nritems,
3794 btrfs_tree_unlock(left);
3795 free_extent_buffer(left);
3800 * split the path's leaf in two, making sure there is at least data_size
3801 * available for the resulting leaf level of the path.
3803 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3804 struct btrfs_root *root,
3805 struct btrfs_path *path,
3806 struct extent_buffer *l,
3807 struct extent_buffer *right,
3808 int slot, int mid, int nritems)
3813 struct btrfs_disk_key disk_key;
3814 struct btrfs_map_token token;
3816 btrfs_init_map_token(&token);
3818 nritems = nritems - mid;
3819 btrfs_set_header_nritems(right, nritems);
3820 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
3822 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3823 btrfs_item_nr_offset(mid),
3824 nritems * sizeof(struct btrfs_item));
3826 copy_extent_buffer(right, l,
3827 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
3828 data_copy_size, btrfs_leaf_data(l) +
3829 leaf_data_end(root, l), data_copy_size);
3831 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
3832 btrfs_item_end_nr(l, mid);
3834 for (i = 0; i < nritems; i++) {
3835 struct btrfs_item *item = btrfs_item_nr(right, i);
3838 ioff = btrfs_token_item_offset(right, item, &token);
3839 btrfs_set_token_item_offset(right, item,
3840 ioff + rt_data_off, &token);
3843 btrfs_set_header_nritems(l, mid);
3844 btrfs_item_key(right, &disk_key, 0);
3845 insert_ptr(trans, root, path, &disk_key, right->start,
3846 path->slots[1] + 1, 1);
3848 btrfs_mark_buffer_dirty(right);
3849 btrfs_mark_buffer_dirty(l);
3850 BUG_ON(path->slots[0] != slot);
3853 btrfs_tree_unlock(path->nodes[0]);
3854 free_extent_buffer(path->nodes[0]);
3855 path->nodes[0] = right;
3856 path->slots[0] -= mid;
3857 path->slots[1] += 1;
3859 btrfs_tree_unlock(right);
3860 free_extent_buffer(right);
3863 BUG_ON(path->slots[0] < 0);
3867 * double splits happen when we need to insert a big item in the middle
3868 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3869 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3872 * We avoid this by trying to push the items on either side of our target
3873 * into the adjacent leaves. If all goes well we can avoid the double split
3876 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3877 struct btrfs_root *root,
3878 struct btrfs_path *path,
3886 slot = path->slots[0];
3889 * try to push all the items after our slot into the
3892 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
3899 nritems = btrfs_header_nritems(path->nodes[0]);
3901 * our goal is to get our slot at the start or end of a leaf. If
3902 * we've done so we're done
3904 if (path->slots[0] == 0 || path->slots[0] == nritems)
3907 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3910 /* try to push all the items before our slot into the next leaf */
3911 slot = path->slots[0];
3912 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
3925 * split the path's leaf in two, making sure there is at least data_size
3926 * available for the resulting leaf level of the path.
3928 * returns 0 if all went well and < 0 on failure.
3930 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3931 struct btrfs_root *root,
3932 struct btrfs_key *ins_key,
3933 struct btrfs_path *path, int data_size,
3936 struct btrfs_disk_key disk_key;
3937 struct extent_buffer *l;
3941 struct extent_buffer *right;
3945 int num_doubles = 0;
3946 int tried_avoid_double = 0;
3949 slot = path->slots[0];
3950 if (extend && data_size + btrfs_item_size_nr(l, slot) +
3951 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
3954 /* first try to make some room by pushing left and right */
3955 if (data_size && path->nodes[1]) {
3956 wret = push_leaf_right(trans, root, path, data_size,
3961 wret = push_leaf_left(trans, root, path, data_size,
3962 data_size, 0, (u32)-1);
3968 /* did the pushes work? */
3969 if (btrfs_leaf_free_space(root, l) >= data_size)
3973 if (!path->nodes[1]) {
3974 ret = insert_new_root(trans, root, path, 1);
3981 slot = path->slots[0];
3982 nritems = btrfs_header_nritems(l);
3983 mid = (nritems + 1) / 2;
3987 leaf_space_used(l, mid, nritems - mid) + data_size >
3988 BTRFS_LEAF_DATA_SIZE(root)) {
3989 if (slot >= nritems) {
3993 if (mid != nritems &&
3994 leaf_space_used(l, mid, nritems - mid) +
3995 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
3996 if (data_size && !tried_avoid_double)
3997 goto push_for_double;
4003 if (leaf_space_used(l, 0, mid) + data_size >
4004 BTRFS_LEAF_DATA_SIZE(root)) {
4005 if (!extend && data_size && slot == 0) {
4007 } else if ((extend || !data_size) && slot == 0) {
4011 if (mid != nritems &&
4012 leaf_space_used(l, mid, nritems - mid) +
4013 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4014 if (data_size && !tried_avoid_double)
4015 goto push_for_double;
4023 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4025 btrfs_item_key(l, &disk_key, mid);
4027 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
4028 root->root_key.objectid,
4029 &disk_key, 0, l->start, 0);
4031 return PTR_ERR(right);
4033 root_add_used(root, root->leafsize);
4035 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4036 btrfs_set_header_bytenr(right, right->start);
4037 btrfs_set_header_generation(right, trans->transid);
4038 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4039 btrfs_set_header_owner(right, root->root_key.objectid);
4040 btrfs_set_header_level(right, 0);
4041 write_extent_buffer(right, root->fs_info->fsid,
4042 btrfs_header_fsid(right), BTRFS_FSID_SIZE);
4044 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4045 btrfs_header_chunk_tree_uuid(right),
4050 btrfs_set_header_nritems(right, 0);
4051 insert_ptr(trans, root, path, &disk_key, right->start,
4052 path->slots[1] + 1, 1);
4053 btrfs_tree_unlock(path->nodes[0]);
4054 free_extent_buffer(path->nodes[0]);
4055 path->nodes[0] = right;
4057 path->slots[1] += 1;
4059 btrfs_set_header_nritems(right, 0);
4060 insert_ptr(trans, root, path, &disk_key, right->start,
4062 btrfs_tree_unlock(path->nodes[0]);
4063 free_extent_buffer(path->nodes[0]);
4064 path->nodes[0] = right;
4066 if (path->slots[1] == 0)
4067 fixup_low_keys(root, path, &disk_key, 1);
4069 btrfs_mark_buffer_dirty(right);
4073 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4076 BUG_ON(num_doubles != 0);
4084 push_for_double_split(trans, root, path, data_size);
4085 tried_avoid_double = 1;
4086 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4091 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4092 struct btrfs_root *root,
4093 struct btrfs_path *path, int ins_len)
4095 struct btrfs_key key;
4096 struct extent_buffer *leaf;
4097 struct btrfs_file_extent_item *fi;
4102 leaf = path->nodes[0];
4103 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4105 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4106 key.type != BTRFS_EXTENT_CSUM_KEY);
4108 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4111 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4112 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4113 fi = btrfs_item_ptr(leaf, path->slots[0],
4114 struct btrfs_file_extent_item);
4115 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4117 btrfs_release_path(path);
4119 path->keep_locks = 1;
4120 path->search_for_split = 1;
4121 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4122 path->search_for_split = 0;
4127 leaf = path->nodes[0];
4128 /* if our item isn't there or got smaller, return now */
4129 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4132 /* the leaf has changed, it now has room. return now */
4133 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4136 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4137 fi = btrfs_item_ptr(leaf, path->slots[0],
4138 struct btrfs_file_extent_item);
4139 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4143 btrfs_set_path_blocking(path);
4144 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4148 path->keep_locks = 0;
4149 btrfs_unlock_up_safe(path, 1);
4152 path->keep_locks = 0;
4156 static noinline int split_item(struct btrfs_trans_handle *trans,
4157 struct btrfs_root *root,
4158 struct btrfs_path *path,
4159 struct btrfs_key *new_key,
4160 unsigned long split_offset)
4162 struct extent_buffer *leaf;
4163 struct btrfs_item *item;
4164 struct btrfs_item *new_item;
4170 struct btrfs_disk_key disk_key;
4172 leaf = path->nodes[0];
4173 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4175 btrfs_set_path_blocking(path);
4177 item = btrfs_item_nr(leaf, path->slots[0]);
4178 orig_offset = btrfs_item_offset(leaf, item);
4179 item_size = btrfs_item_size(leaf, item);
4181 buf = kmalloc(item_size, GFP_NOFS);
4185 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4186 path->slots[0]), item_size);
4188 slot = path->slots[0] + 1;
4189 nritems = btrfs_header_nritems(leaf);
4190 if (slot != nritems) {
4191 /* shift the items */
4192 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4193 btrfs_item_nr_offset(slot),
4194 (nritems - slot) * sizeof(struct btrfs_item));
4197 btrfs_cpu_key_to_disk(&disk_key, new_key);
4198 btrfs_set_item_key(leaf, &disk_key, slot);
4200 new_item = btrfs_item_nr(leaf, slot);
4202 btrfs_set_item_offset(leaf, new_item, orig_offset);
4203 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4205 btrfs_set_item_offset(leaf, item,
4206 orig_offset + item_size - split_offset);
4207 btrfs_set_item_size(leaf, item, split_offset);
4209 btrfs_set_header_nritems(leaf, nritems + 1);
4211 /* write the data for the start of the original item */
4212 write_extent_buffer(leaf, buf,
4213 btrfs_item_ptr_offset(leaf, path->slots[0]),
4216 /* write the data for the new item */
4217 write_extent_buffer(leaf, buf + split_offset,
4218 btrfs_item_ptr_offset(leaf, slot),
4219 item_size - split_offset);
4220 btrfs_mark_buffer_dirty(leaf);
4222 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4228 * This function splits a single item into two items,
4229 * giving 'new_key' to the new item and splitting the
4230 * old one at split_offset (from the start of the item).
4232 * The path may be released by this operation. After
4233 * the split, the path is pointing to the old item. The
4234 * new item is going to be in the same node as the old one.
4236 * Note, the item being split must be smaller enough to live alone on
4237 * a tree block with room for one extra struct btrfs_item
4239 * This allows us to split the item in place, keeping a lock on the
4240 * leaf the entire time.
4242 int btrfs_split_item(struct btrfs_trans_handle *trans,
4243 struct btrfs_root *root,
4244 struct btrfs_path *path,
4245 struct btrfs_key *new_key,
4246 unsigned long split_offset)
4249 ret = setup_leaf_for_split(trans, root, path,
4250 sizeof(struct btrfs_item));
4254 ret = split_item(trans, root, path, new_key, split_offset);
4259 * This function duplicate a item, giving 'new_key' to the new item.
4260 * It guarantees both items live in the same tree leaf and the new item
4261 * is contiguous with the original item.
4263 * This allows us to split file extent in place, keeping a lock on the
4264 * leaf the entire time.
4266 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4267 struct btrfs_root *root,
4268 struct btrfs_path *path,
4269 struct btrfs_key *new_key)
4271 struct extent_buffer *leaf;
4275 leaf = path->nodes[0];
4276 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4277 ret = setup_leaf_for_split(trans, root, path,
4278 item_size + sizeof(struct btrfs_item));
4283 setup_items_for_insert(root, path, new_key, &item_size,
4284 item_size, item_size +
4285 sizeof(struct btrfs_item), 1);
4286 leaf = path->nodes[0];
4287 memcpy_extent_buffer(leaf,
4288 btrfs_item_ptr_offset(leaf, path->slots[0]),
4289 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4295 * make the item pointed to by the path smaller. new_size indicates
4296 * how small to make it, and from_end tells us if we just chop bytes
4297 * off the end of the item or if we shift the item to chop bytes off
4300 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4301 u32 new_size, int from_end)
4304 struct extent_buffer *leaf;
4305 struct btrfs_item *item;
4307 unsigned int data_end;
4308 unsigned int old_data_start;
4309 unsigned int old_size;
4310 unsigned int size_diff;
4312 struct btrfs_map_token token;
4314 btrfs_init_map_token(&token);
4316 leaf = path->nodes[0];
4317 slot = path->slots[0];
4319 old_size = btrfs_item_size_nr(leaf, slot);
4320 if (old_size == new_size)
4323 nritems = btrfs_header_nritems(leaf);
4324 data_end = leaf_data_end(root, leaf);
4326 old_data_start = btrfs_item_offset_nr(leaf, slot);
4328 size_diff = old_size - new_size;
4331 BUG_ON(slot >= nritems);
4334 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4336 /* first correct the data pointers */
4337 for (i = slot; i < nritems; i++) {
4339 item = btrfs_item_nr(leaf, i);
4341 ioff = btrfs_token_item_offset(leaf, item, &token);
4342 btrfs_set_token_item_offset(leaf, item,
4343 ioff + size_diff, &token);
4346 /* shift the data */
4348 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4349 data_end + size_diff, btrfs_leaf_data(leaf) +
4350 data_end, old_data_start + new_size - data_end);
4352 struct btrfs_disk_key disk_key;
4355 btrfs_item_key(leaf, &disk_key, slot);
4357 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4359 struct btrfs_file_extent_item *fi;
4361 fi = btrfs_item_ptr(leaf, slot,
4362 struct btrfs_file_extent_item);
4363 fi = (struct btrfs_file_extent_item *)(
4364 (unsigned long)fi - size_diff);
4366 if (btrfs_file_extent_type(leaf, fi) ==
4367 BTRFS_FILE_EXTENT_INLINE) {
4368 ptr = btrfs_item_ptr_offset(leaf, slot);
4369 memmove_extent_buffer(leaf, ptr,
4371 offsetof(struct btrfs_file_extent_item,
4376 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4377 data_end + size_diff, btrfs_leaf_data(leaf) +
4378 data_end, old_data_start - data_end);
4380 offset = btrfs_disk_key_offset(&disk_key);
4381 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4382 btrfs_set_item_key(leaf, &disk_key, slot);
4384 fixup_low_keys(root, path, &disk_key, 1);
4387 item = btrfs_item_nr(leaf, slot);
4388 btrfs_set_item_size(leaf, item, new_size);
4389 btrfs_mark_buffer_dirty(leaf);
4391 if (btrfs_leaf_free_space(root, leaf) < 0) {
4392 btrfs_print_leaf(root, leaf);
4398 * make the item pointed to by the path bigger, data_size is the added size.
4400 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4404 struct extent_buffer *leaf;
4405 struct btrfs_item *item;
4407 unsigned int data_end;
4408 unsigned int old_data;
4409 unsigned int old_size;
4411 struct btrfs_map_token token;
4413 btrfs_init_map_token(&token);
4415 leaf = path->nodes[0];
4417 nritems = btrfs_header_nritems(leaf);
4418 data_end = leaf_data_end(root, leaf);
4420 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4421 btrfs_print_leaf(root, leaf);
4424 slot = path->slots[0];
4425 old_data = btrfs_item_end_nr(leaf, slot);
4428 if (slot >= nritems) {
4429 btrfs_print_leaf(root, leaf);
4430 printk(KERN_CRIT "slot %d too large, nritems %d\n",
4436 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4438 /* first correct the data pointers */
4439 for (i = slot; i < nritems; i++) {
4441 item = btrfs_item_nr(leaf, i);
4443 ioff = btrfs_token_item_offset(leaf, item, &token);
4444 btrfs_set_token_item_offset(leaf, item,
4445 ioff - data_size, &token);
4448 /* shift the data */
4449 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4450 data_end - data_size, btrfs_leaf_data(leaf) +
4451 data_end, old_data - data_end);
4453 data_end = old_data;
4454 old_size = btrfs_item_size_nr(leaf, slot);
4455 item = btrfs_item_nr(leaf, slot);
4456 btrfs_set_item_size(leaf, item, old_size + data_size);
4457 btrfs_mark_buffer_dirty(leaf);
4459 if (btrfs_leaf_free_space(root, leaf) < 0) {
4460 btrfs_print_leaf(root, leaf);
4466 * this is a helper for btrfs_insert_empty_items, the main goal here is
4467 * to save stack depth by doing the bulk of the work in a function
4468 * that doesn't call btrfs_search_slot
4470 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4471 struct btrfs_key *cpu_key, u32 *data_size,
4472 u32 total_data, u32 total_size, int nr)
4474 struct btrfs_item *item;
4477 unsigned int data_end;
4478 struct btrfs_disk_key disk_key;
4479 struct extent_buffer *leaf;
4481 struct btrfs_map_token token;
4483 btrfs_init_map_token(&token);
4485 leaf = path->nodes[0];
4486 slot = path->slots[0];
4488 nritems = btrfs_header_nritems(leaf);
4489 data_end = leaf_data_end(root, leaf);
4491 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4492 btrfs_print_leaf(root, leaf);
4493 printk(KERN_CRIT "not enough freespace need %u have %d\n",
4494 total_size, btrfs_leaf_free_space(root, leaf));
4498 if (slot != nritems) {
4499 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4501 if (old_data < data_end) {
4502 btrfs_print_leaf(root, leaf);
4503 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
4504 slot, old_data, data_end);
4508 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4510 /* first correct the data pointers */
4511 for (i = slot; i < nritems; i++) {
4514 item = btrfs_item_nr(leaf, i);
4515 ioff = btrfs_token_item_offset(leaf, item, &token);
4516 btrfs_set_token_item_offset(leaf, item,
4517 ioff - total_data, &token);
4519 /* shift the items */
4520 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4521 btrfs_item_nr_offset(slot),
4522 (nritems - slot) * sizeof(struct btrfs_item));
4524 /* shift the data */
4525 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4526 data_end - total_data, btrfs_leaf_data(leaf) +
4527 data_end, old_data - data_end);
4528 data_end = old_data;
4531 /* setup the item for the new data */
4532 for (i = 0; i < nr; i++) {
4533 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4534 btrfs_set_item_key(leaf, &disk_key, slot + i);
4535 item = btrfs_item_nr(leaf, slot + i);
4536 btrfs_set_token_item_offset(leaf, item,
4537 data_end - data_size[i], &token);
4538 data_end -= data_size[i];
4539 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4542 btrfs_set_header_nritems(leaf, nritems + nr);
4545 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4546 fixup_low_keys(root, path, &disk_key, 1);
4548 btrfs_unlock_up_safe(path, 1);
4549 btrfs_mark_buffer_dirty(leaf);
4551 if (btrfs_leaf_free_space(root, leaf) < 0) {
4552 btrfs_print_leaf(root, leaf);
4558 * Given a key and some data, insert items into the tree.
4559 * This does all the path init required, making room in the tree if needed.
4561 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4562 struct btrfs_root *root,
4563 struct btrfs_path *path,
4564 struct btrfs_key *cpu_key, u32 *data_size,
4573 for (i = 0; i < nr; i++)
4574 total_data += data_size[i];
4576 total_size = total_data + (nr * sizeof(struct btrfs_item));
4577 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4583 slot = path->slots[0];
4586 setup_items_for_insert(root, path, cpu_key, data_size,
4587 total_data, total_size, nr);
4592 * Given a key and some data, insert an item into the tree.
4593 * This does all the path init required, making room in the tree if needed.
4595 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4596 *root, struct btrfs_key *cpu_key, void *data, u32
4600 struct btrfs_path *path;
4601 struct extent_buffer *leaf;
4604 path = btrfs_alloc_path();
4607 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4609 leaf = path->nodes[0];
4610 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4611 write_extent_buffer(leaf, data, ptr, data_size);
4612 btrfs_mark_buffer_dirty(leaf);
4614 btrfs_free_path(path);
4619 * delete the pointer from a given node.
4621 * the tree should have been previously balanced so the deletion does not
4624 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4625 int level, int slot)
4627 struct extent_buffer *parent = path->nodes[level];
4631 nritems = btrfs_header_nritems(parent);
4632 if (slot != nritems - 1) {
4634 tree_mod_log_eb_move(root->fs_info, parent, slot,
4635 slot + 1, nritems - slot - 1);
4636 memmove_extent_buffer(parent,
4637 btrfs_node_key_ptr_offset(slot),
4638 btrfs_node_key_ptr_offset(slot + 1),
4639 sizeof(struct btrfs_key_ptr) *
4640 (nritems - slot - 1));
4642 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4643 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4648 btrfs_set_header_nritems(parent, nritems);
4649 if (nritems == 0 && parent == root->node) {
4650 BUG_ON(btrfs_header_level(root->node) != 1);
4651 /* just turn the root into a leaf and break */
4652 btrfs_set_header_level(root->node, 0);
4653 } else if (slot == 0) {
4654 struct btrfs_disk_key disk_key;
4656 btrfs_node_key(parent, &disk_key, 0);
4657 fixup_low_keys(root, path, &disk_key, level + 1);
4659 btrfs_mark_buffer_dirty(parent);
4663 * a helper function to delete the leaf pointed to by path->slots[1] and
4666 * This deletes the pointer in path->nodes[1] and frees the leaf
4667 * block extent. zero is returned if it all worked out, < 0 otherwise.
4669 * The path must have already been setup for deleting the leaf, including
4670 * all the proper balancing. path->nodes[1] must be locked.
4672 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4673 struct btrfs_root *root,
4674 struct btrfs_path *path,
4675 struct extent_buffer *leaf)
4677 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4678 del_ptr(root, path, 1, path->slots[1]);
4681 * btrfs_free_extent is expensive, we want to make sure we
4682 * aren't holding any locks when we call it
4684 btrfs_unlock_up_safe(path, 0);
4686 root_sub_used(root, leaf->len);
4688 extent_buffer_get(leaf);
4689 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4690 free_extent_buffer_stale(leaf);
4693 * delete the item at the leaf level in path. If that empties
4694 * the leaf, remove it from the tree
4696 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4697 struct btrfs_path *path, int slot, int nr)
4699 struct extent_buffer *leaf;
4700 struct btrfs_item *item;
4707 struct btrfs_map_token token;
4709 btrfs_init_map_token(&token);
4711 leaf = path->nodes[0];
4712 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4714 for (i = 0; i < nr; i++)
4715 dsize += btrfs_item_size_nr(leaf, slot + i);
4717 nritems = btrfs_header_nritems(leaf);
4719 if (slot + nr != nritems) {
4720 int data_end = leaf_data_end(root, leaf);
4722 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4724 btrfs_leaf_data(leaf) + data_end,
4725 last_off - data_end);
4727 for (i = slot + nr; i < nritems; i++) {
4730 item = btrfs_item_nr(leaf, i);
4731 ioff = btrfs_token_item_offset(leaf, item, &token);
4732 btrfs_set_token_item_offset(leaf, item,
4733 ioff + dsize, &token);
4736 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4737 btrfs_item_nr_offset(slot + nr),
4738 sizeof(struct btrfs_item) *
4739 (nritems - slot - nr));
4741 btrfs_set_header_nritems(leaf, nritems - nr);
4744 /* delete the leaf if we've emptied it */
4746 if (leaf == root->node) {
4747 btrfs_set_header_level(leaf, 0);
4749 btrfs_set_path_blocking(path);
4750 clean_tree_block(trans, root, leaf);
4751 btrfs_del_leaf(trans, root, path, leaf);
4754 int used = leaf_space_used(leaf, 0, nritems);
4756 struct btrfs_disk_key disk_key;
4758 btrfs_item_key(leaf, &disk_key, 0);
4759 fixup_low_keys(root, path, &disk_key, 1);
4762 /* delete the leaf if it is mostly empty */
4763 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
4764 /* push_leaf_left fixes the path.
4765 * make sure the path still points to our leaf
4766 * for possible call to del_ptr below
4768 slot = path->slots[1];
4769 extent_buffer_get(leaf);
4771 btrfs_set_path_blocking(path);
4772 wret = push_leaf_left(trans, root, path, 1, 1,
4774 if (wret < 0 && wret != -ENOSPC)
4777 if (path->nodes[0] == leaf &&
4778 btrfs_header_nritems(leaf)) {
4779 wret = push_leaf_right(trans, root, path, 1,
4781 if (wret < 0 && wret != -ENOSPC)
4785 if (btrfs_header_nritems(leaf) == 0) {
4786 path->slots[1] = slot;
4787 btrfs_del_leaf(trans, root, path, leaf);
4788 free_extent_buffer(leaf);
4791 /* if we're still in the path, make sure
4792 * we're dirty. Otherwise, one of the
4793 * push_leaf functions must have already
4794 * dirtied this buffer
4796 if (path->nodes[0] == leaf)
4797 btrfs_mark_buffer_dirty(leaf);
4798 free_extent_buffer(leaf);
4801 btrfs_mark_buffer_dirty(leaf);
4808 * search the tree again to find a leaf with lesser keys
4809 * returns 0 if it found something or 1 if there are no lesser leaves.
4810 * returns < 0 on io errors.
4812 * This may release the path, and so you may lose any locks held at the
4815 static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4817 struct btrfs_key key;
4818 struct btrfs_disk_key found_key;
4821 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4825 else if (key.type > 0)
4827 else if (key.objectid > 0)
4832 btrfs_release_path(path);
4833 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4836 btrfs_item_key(path->nodes[0], &found_key, 0);
4837 ret = comp_keys(&found_key, &key);
4844 * A helper function to walk down the tree starting at min_key, and looking
4845 * for nodes or leaves that are have a minimum transaction id.
4846 * This is used by the btree defrag code, and tree logging
4848 * This does not cow, but it does stuff the starting key it finds back
4849 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4850 * key and get a writable path.
4852 * This does lock as it descends, and path->keep_locks should be set
4853 * to 1 by the caller.
4855 * This honors path->lowest_level to prevent descent past a given level
4858 * min_trans indicates the oldest transaction that you are interested
4859 * in walking through. Any nodes or leaves older than min_trans are
4860 * skipped over (without reading them).
4862 * returns zero if something useful was found, < 0 on error and 1 if there
4863 * was nothing in the tree that matched the search criteria.
4865 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4866 struct btrfs_key *max_key,
4867 struct btrfs_path *path,
4870 struct extent_buffer *cur;
4871 struct btrfs_key found_key;
4878 WARN_ON(!path->keep_locks);
4880 cur = btrfs_read_lock_root_node(root);
4881 level = btrfs_header_level(cur);
4882 WARN_ON(path->nodes[level]);
4883 path->nodes[level] = cur;
4884 path->locks[level] = BTRFS_READ_LOCK;
4886 if (btrfs_header_generation(cur) < min_trans) {
4891 nritems = btrfs_header_nritems(cur);
4892 level = btrfs_header_level(cur);
4893 sret = bin_search(cur, min_key, level, &slot);
4895 /* at the lowest level, we're done, setup the path and exit */
4896 if (level == path->lowest_level) {
4897 if (slot >= nritems)
4900 path->slots[level] = slot;
4901 btrfs_item_key_to_cpu(cur, &found_key, slot);
4904 if (sret && slot > 0)
4907 * check this node pointer against the min_trans parameters.
4908 * If it is too old, old, skip to the next one.
4910 while (slot < nritems) {
4914 blockptr = btrfs_node_blockptr(cur, slot);
4915 gen = btrfs_node_ptr_generation(cur, slot);
4916 if (gen < min_trans) {
4924 * we didn't find a candidate key in this node, walk forward
4925 * and find another one
4927 if (slot >= nritems) {
4928 path->slots[level] = slot;
4929 btrfs_set_path_blocking(path);
4930 sret = btrfs_find_next_key(root, path, min_key, level,
4933 btrfs_release_path(path);
4939 /* save our key for returning back */
4940 btrfs_node_key_to_cpu(cur, &found_key, slot);
4941 path->slots[level] = slot;
4942 if (level == path->lowest_level) {
4944 unlock_up(path, level, 1, 0, NULL);
4947 btrfs_set_path_blocking(path);
4948 cur = read_node_slot(root, cur, slot);
4949 BUG_ON(!cur); /* -ENOMEM */
4951 btrfs_tree_read_lock(cur);
4953 path->locks[level - 1] = BTRFS_READ_LOCK;
4954 path->nodes[level - 1] = cur;
4955 unlock_up(path, level, 1, 0, NULL);
4956 btrfs_clear_path_blocking(path, NULL, 0);
4960 memcpy(min_key, &found_key, sizeof(found_key));
4961 btrfs_set_path_blocking(path);
4965 static void tree_move_down(struct btrfs_root *root,
4966 struct btrfs_path *path,
4967 int *level, int root_level)
4969 BUG_ON(*level == 0);
4970 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
4971 path->slots[*level]);
4972 path->slots[*level - 1] = 0;
4976 static int tree_move_next_or_upnext(struct btrfs_root *root,
4977 struct btrfs_path *path,
4978 int *level, int root_level)
4982 nritems = btrfs_header_nritems(path->nodes[*level]);
4984 path->slots[*level]++;
4986 while (path->slots[*level] >= nritems) {
4987 if (*level == root_level)
4991 path->slots[*level] = 0;
4992 free_extent_buffer(path->nodes[*level]);
4993 path->nodes[*level] = NULL;
4995 path->slots[*level]++;
4997 nritems = btrfs_header_nritems(path->nodes[*level]);
5004 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5007 static int tree_advance(struct btrfs_root *root,
5008 struct btrfs_path *path,
5009 int *level, int root_level,
5011 struct btrfs_key *key)
5015 if (*level == 0 || !allow_down) {
5016 ret = tree_move_next_or_upnext(root, path, level, root_level);
5018 tree_move_down(root, path, level, root_level);
5023 btrfs_item_key_to_cpu(path->nodes[*level], key,
5024 path->slots[*level]);
5026 btrfs_node_key_to_cpu(path->nodes[*level], key,
5027 path->slots[*level]);
5032 static int tree_compare_item(struct btrfs_root *left_root,
5033 struct btrfs_path *left_path,
5034 struct btrfs_path *right_path,
5039 unsigned long off1, off2;
5041 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5042 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5046 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5047 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5048 right_path->slots[0]);
5050 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5052 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5059 #define ADVANCE_ONLY_NEXT -1
5062 * This function compares two trees and calls the provided callback for
5063 * every changed/new/deleted item it finds.
5064 * If shared tree blocks are encountered, whole subtrees are skipped, making
5065 * the compare pretty fast on snapshotted subvolumes.
5067 * This currently works on commit roots only. As commit roots are read only,
5068 * we don't do any locking. The commit roots are protected with transactions.
5069 * Transactions are ended and rejoined when a commit is tried in between.
5071 * This function checks for modifications done to the trees while comparing.
5072 * If it detects a change, it aborts immediately.
5074 int btrfs_compare_trees(struct btrfs_root *left_root,
5075 struct btrfs_root *right_root,
5076 btrfs_changed_cb_t changed_cb, void *ctx)
5080 struct btrfs_trans_handle *trans = NULL;
5081 struct btrfs_path *left_path = NULL;
5082 struct btrfs_path *right_path = NULL;
5083 struct btrfs_key left_key;
5084 struct btrfs_key right_key;
5085 char *tmp_buf = NULL;
5086 int left_root_level;
5087 int right_root_level;
5090 int left_end_reached;
5091 int right_end_reached;
5096 u64 left_start_ctransid;
5097 u64 right_start_ctransid;
5100 left_path = btrfs_alloc_path();
5105 right_path = btrfs_alloc_path();
5111 tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5117 left_path->search_commit_root = 1;
5118 left_path->skip_locking = 1;
5119 right_path->search_commit_root = 1;
5120 right_path->skip_locking = 1;
5122 spin_lock(&left_root->root_item_lock);
5123 left_start_ctransid = btrfs_root_ctransid(&left_root->root_item);
5124 spin_unlock(&left_root->root_item_lock);
5126 spin_lock(&right_root->root_item_lock);
5127 right_start_ctransid = btrfs_root_ctransid(&right_root->root_item);
5128 spin_unlock(&right_root->root_item_lock);
5130 trans = btrfs_join_transaction(left_root);
5131 if (IS_ERR(trans)) {
5132 ret = PTR_ERR(trans);
5138 * Strategy: Go to the first items of both trees. Then do
5140 * If both trees are at level 0
5141 * Compare keys of current items
5142 * If left < right treat left item as new, advance left tree
5144 * If left > right treat right item as deleted, advance right tree
5146 * If left == right do deep compare of items, treat as changed if
5147 * needed, advance both trees and repeat
5148 * If both trees are at the same level but not at level 0
5149 * Compare keys of current nodes/leafs
5150 * If left < right advance left tree and repeat
5151 * If left > right advance right tree and repeat
5152 * If left == right compare blockptrs of the next nodes/leafs
5153 * If they match advance both trees but stay at the same level
5155 * If they don't match advance both trees while allowing to go
5157 * If tree levels are different
5158 * Advance the tree that needs it and repeat
5160 * Advancing a tree means:
5161 * If we are at level 0, try to go to the next slot. If that's not
5162 * possible, go one level up and repeat. Stop when we found a level
5163 * where we could go to the next slot. We may at this point be on a
5166 * If we are not at level 0 and not on shared tree blocks, go one
5169 * If we are not at level 0 and on shared tree blocks, go one slot to
5170 * the right if possible or go up and right.
5173 left_level = btrfs_header_level(left_root->commit_root);
5174 left_root_level = left_level;
5175 left_path->nodes[left_level] = left_root->commit_root;
5176 extent_buffer_get(left_path->nodes[left_level]);
5178 right_level = btrfs_header_level(right_root->commit_root);
5179 right_root_level = right_level;
5180 right_path->nodes[right_level] = right_root->commit_root;
5181 extent_buffer_get(right_path->nodes[right_level]);
5183 if (left_level == 0)
5184 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5185 &left_key, left_path->slots[left_level]);
5187 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5188 &left_key, left_path->slots[left_level]);
5189 if (right_level == 0)
5190 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5191 &right_key, right_path->slots[right_level]);
5193 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5194 &right_key, right_path->slots[right_level]);
5196 left_end_reached = right_end_reached = 0;
5197 advance_left = advance_right = 0;
5201 * We need to make sure the transaction does not get committed
5202 * while we do anything on commit roots. This means, we need to
5203 * join and leave transactions for every item that we process.
5205 if (trans && btrfs_should_end_transaction(trans, left_root)) {
5206 btrfs_release_path(left_path);
5207 btrfs_release_path(right_path);
5209 ret = btrfs_end_transaction(trans, left_root);
5214 /* now rejoin the transaction */
5216 trans = btrfs_join_transaction(left_root);
5217 if (IS_ERR(trans)) {
5218 ret = PTR_ERR(trans);
5223 spin_lock(&left_root->root_item_lock);
5224 ctransid = btrfs_root_ctransid(&left_root->root_item);
5225 spin_unlock(&left_root->root_item_lock);
5226 if (ctransid != left_start_ctransid)
5227 left_start_ctransid = 0;
5229 spin_lock(&right_root->root_item_lock);
5230 ctransid = btrfs_root_ctransid(&right_root->root_item);
5231 spin_unlock(&right_root->root_item_lock);
5232 if (ctransid != right_start_ctransid)
5233 right_start_ctransid = 0;
5235 if (!left_start_ctransid || !right_start_ctransid) {
5236 WARN(1, KERN_WARNING
5237 "btrfs: btrfs_compare_tree detected "
5238 "a change in one of the trees while "
5239 "iterating. This is probably a "
5246 * the commit root may have changed, so start again
5249 left_path->lowest_level = left_level;
5250 right_path->lowest_level = right_level;
5251 ret = btrfs_search_slot(NULL, left_root,
5252 &left_key, left_path, 0, 0);
5255 ret = btrfs_search_slot(NULL, right_root,
5256 &right_key, right_path, 0, 0);
5261 if (advance_left && !left_end_reached) {
5262 ret = tree_advance(left_root, left_path, &left_level,
5264 advance_left != ADVANCE_ONLY_NEXT,
5267 left_end_reached = ADVANCE;
5270 if (advance_right && !right_end_reached) {
5271 ret = tree_advance(right_root, right_path, &right_level,
5273 advance_right != ADVANCE_ONLY_NEXT,
5276 right_end_reached = ADVANCE;
5280 if (left_end_reached && right_end_reached) {
5283 } else if (left_end_reached) {
5284 if (right_level == 0) {
5285 ret = changed_cb(left_root, right_root,
5286 left_path, right_path,
5288 BTRFS_COMPARE_TREE_DELETED,
5293 advance_right = ADVANCE;
5295 } else if (right_end_reached) {
5296 if (left_level == 0) {
5297 ret = changed_cb(left_root, right_root,
5298 left_path, right_path,
5300 BTRFS_COMPARE_TREE_NEW,
5305 advance_left = ADVANCE;
5309 if (left_level == 0 && right_level == 0) {
5310 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5312 ret = changed_cb(left_root, right_root,
5313 left_path, right_path,
5315 BTRFS_COMPARE_TREE_NEW,
5319 advance_left = ADVANCE;
5320 } else if (cmp > 0) {
5321 ret = changed_cb(left_root, right_root,
5322 left_path, right_path,
5324 BTRFS_COMPARE_TREE_DELETED,
5328 advance_right = ADVANCE;
5330 enum btrfs_compare_tree_result cmp;
5332 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5333 ret = tree_compare_item(left_root, left_path,
5334 right_path, tmp_buf);
5336 cmp = BTRFS_COMPARE_TREE_CHANGED;
5338 cmp = BTRFS_COMPARE_TREE_SAME;
5339 ret = changed_cb(left_root, right_root,
5340 left_path, right_path,
5341 &left_key, cmp, ctx);
5344 advance_left = ADVANCE;
5345 advance_right = ADVANCE;
5347 } else if (left_level == right_level) {
5348 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5350 advance_left = ADVANCE;
5351 } else if (cmp > 0) {
5352 advance_right = ADVANCE;
5354 left_blockptr = btrfs_node_blockptr(
5355 left_path->nodes[left_level],
5356 left_path->slots[left_level]);
5357 right_blockptr = btrfs_node_blockptr(
5358 right_path->nodes[right_level],
5359 right_path->slots[right_level]);
5360 if (left_blockptr == right_blockptr) {
5362 * As we're on a shared block, don't
5363 * allow to go deeper.
5365 advance_left = ADVANCE_ONLY_NEXT;
5366 advance_right = ADVANCE_ONLY_NEXT;
5368 advance_left = ADVANCE;
5369 advance_right = ADVANCE;
5372 } else if (left_level < right_level) {
5373 advance_right = ADVANCE;
5375 advance_left = ADVANCE;
5380 btrfs_free_path(left_path);
5381 btrfs_free_path(right_path);
5386 ret = btrfs_end_transaction(trans, left_root);
5388 btrfs_end_transaction(trans, left_root);
5395 * this is similar to btrfs_next_leaf, but does not try to preserve
5396 * and fixup the path. It looks for and returns the next key in the
5397 * tree based on the current path and the min_trans parameters.
5399 * 0 is returned if another key is found, < 0 if there are any errors
5400 * and 1 is returned if there are no higher keys in the tree
5402 * path->keep_locks should be set to 1 on the search made before
5403 * calling this function.
5405 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5406 struct btrfs_key *key, int level, u64 min_trans)
5409 struct extent_buffer *c;
5411 WARN_ON(!path->keep_locks);
5412 while (level < BTRFS_MAX_LEVEL) {
5413 if (!path->nodes[level])
5416 slot = path->slots[level] + 1;
5417 c = path->nodes[level];
5419 if (slot >= btrfs_header_nritems(c)) {
5422 struct btrfs_key cur_key;
5423 if (level + 1 >= BTRFS_MAX_LEVEL ||
5424 !path->nodes[level + 1])
5427 if (path->locks[level + 1]) {
5432 slot = btrfs_header_nritems(c) - 1;
5434 btrfs_item_key_to_cpu(c, &cur_key, slot);
5436 btrfs_node_key_to_cpu(c, &cur_key, slot);
5438 orig_lowest = path->lowest_level;
5439 btrfs_release_path(path);
5440 path->lowest_level = level;
5441 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5443 path->lowest_level = orig_lowest;
5447 c = path->nodes[level];
5448 slot = path->slots[level];
5455 btrfs_item_key_to_cpu(c, key, slot);
5457 u64 gen = btrfs_node_ptr_generation(c, slot);
5459 if (gen < min_trans) {
5463 btrfs_node_key_to_cpu(c, key, slot);
5471 * search the tree again to find a leaf with greater keys
5472 * returns 0 if it found something or 1 if there are no greater leaves.
5473 * returns < 0 on io errors.
5475 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5477 return btrfs_next_old_leaf(root, path, 0);
5480 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5485 struct extent_buffer *c;
5486 struct extent_buffer *next;
5487 struct btrfs_key key;
5490 int old_spinning = path->leave_spinning;
5491 int next_rw_lock = 0;
5493 nritems = btrfs_header_nritems(path->nodes[0]);
5497 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5502 btrfs_release_path(path);
5504 path->keep_locks = 1;
5505 path->leave_spinning = 1;
5508 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5511 path->keep_locks = 0;
5516 nritems = btrfs_header_nritems(path->nodes[0]);
5518 * by releasing the path above we dropped all our locks. A balance
5519 * could have added more items next to the key that used to be
5520 * at the very end of the block. So, check again here and
5521 * advance the path if there are now more items available.
5523 if (nritems > 0 && path->slots[0] < nritems - 1) {
5530 while (level < BTRFS_MAX_LEVEL) {
5531 if (!path->nodes[level]) {
5536 slot = path->slots[level] + 1;
5537 c = path->nodes[level];
5538 if (slot >= btrfs_header_nritems(c)) {
5540 if (level == BTRFS_MAX_LEVEL) {
5548 btrfs_tree_unlock_rw(next, next_rw_lock);
5549 free_extent_buffer(next);
5553 next_rw_lock = path->locks[level];
5554 ret = read_block_for_search(NULL, root, path, &next, level,
5560 btrfs_release_path(path);
5564 if (!path->skip_locking) {
5565 ret = btrfs_try_tree_read_lock(next);
5566 if (!ret && time_seq) {
5568 * If we don't get the lock, we may be racing
5569 * with push_leaf_left, holding that lock while
5570 * itself waiting for the leaf we've currently
5571 * locked. To solve this situation, we give up
5572 * on our lock and cycle.
5574 free_extent_buffer(next);
5575 btrfs_release_path(path);
5580 btrfs_set_path_blocking(path);
5581 btrfs_tree_read_lock(next);
5582 btrfs_clear_path_blocking(path, next,
5585 next_rw_lock = BTRFS_READ_LOCK;
5589 path->slots[level] = slot;
5592 c = path->nodes[level];
5593 if (path->locks[level])
5594 btrfs_tree_unlock_rw(c, path->locks[level]);
5596 free_extent_buffer(c);
5597 path->nodes[level] = next;
5598 path->slots[level] = 0;
5599 if (!path->skip_locking)
5600 path->locks[level] = next_rw_lock;
5604 ret = read_block_for_search(NULL, root, path, &next, level,
5610 btrfs_release_path(path);
5614 if (!path->skip_locking) {
5615 ret = btrfs_try_tree_read_lock(next);
5617 btrfs_set_path_blocking(path);
5618 btrfs_tree_read_lock(next);
5619 btrfs_clear_path_blocking(path, next,
5622 next_rw_lock = BTRFS_READ_LOCK;
5627 unlock_up(path, 0, 1, 0, NULL);
5628 path->leave_spinning = old_spinning;
5630 btrfs_set_path_blocking(path);
5636 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5637 * searching until it gets past min_objectid or finds an item of 'type'
5639 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5641 int btrfs_previous_item(struct btrfs_root *root,
5642 struct btrfs_path *path, u64 min_objectid,
5645 struct btrfs_key found_key;
5646 struct extent_buffer *leaf;
5651 if (path->slots[0] == 0) {
5652 btrfs_set_path_blocking(path);
5653 ret = btrfs_prev_leaf(root, path);
5659 leaf = path->nodes[0];
5660 nritems = btrfs_header_nritems(leaf);
5663 if (path->slots[0] == nritems)
5666 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5667 if (found_key.objectid < min_objectid)
5669 if (found_key.type == type)
5671 if (found_key.objectid == min_objectid &&
5672 found_key.type < type)