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,
278 (unsigned long)btrfs_header_fsid(cow),
281 WARN_ON(btrfs_header_generation(buf) > trans->transid);
282 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
283 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
285 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
290 btrfs_mark_buffer_dirty(cow);
299 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
300 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
302 MOD_LOG_ROOT_REPLACE,
305 struct tree_mod_move {
310 struct tree_mod_root {
315 struct tree_mod_elem {
317 u64 index; /* shifted logical */
321 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
324 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
327 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
328 struct btrfs_disk_key key;
331 /* this is used for op == MOD_LOG_MOVE_KEYS */
332 struct tree_mod_move move;
334 /* this is used for op == MOD_LOG_ROOT_REPLACE */
335 struct tree_mod_root old_root;
338 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
340 read_lock(&fs_info->tree_mod_log_lock);
343 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
345 read_unlock(&fs_info->tree_mod_log_lock);
348 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
350 write_lock(&fs_info->tree_mod_log_lock);
353 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
355 write_unlock(&fs_info->tree_mod_log_lock);
359 * Increment the upper half of tree_mod_seq, set lower half zero.
361 * Must be called with fs_info->tree_mod_seq_lock held.
363 static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info)
365 u64 seq = atomic64_read(&fs_info->tree_mod_seq);
366 seq &= 0xffffffff00000000ull;
368 atomic64_set(&fs_info->tree_mod_seq, seq);
373 * Increment the lower half of tree_mod_seq.
375 * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers
376 * are generated should not technically require a spin lock here. (Rationale:
377 * incrementing the minor while incrementing the major seq number is between its
378 * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it
379 * just returns a unique sequence number as usual.) We have decided to leave
380 * that requirement in here and rethink it once we notice it really imposes a
381 * problem on some workload.
383 static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info)
385 return atomic64_inc_return(&fs_info->tree_mod_seq);
389 * return the last minor in the previous major tree_mod_seq number
391 u64 btrfs_tree_mod_seq_prev(u64 seq)
393 return (seq & 0xffffffff00000000ull) - 1ull;
397 * This adds a new blocker to the tree mod log's blocker list if the @elem
398 * passed does not already have a sequence number set. So when a caller expects
399 * to record tree modifications, it should ensure to set elem->seq to zero
400 * before calling btrfs_get_tree_mod_seq.
401 * Returns a fresh, unused tree log modification sequence number, even if no new
404 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
405 struct seq_list *elem)
409 tree_mod_log_write_lock(fs_info);
410 spin_lock(&fs_info->tree_mod_seq_lock);
412 elem->seq = btrfs_inc_tree_mod_seq_major(fs_info);
413 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
415 seq = btrfs_inc_tree_mod_seq_minor(fs_info);
416 spin_unlock(&fs_info->tree_mod_seq_lock);
417 tree_mod_log_write_unlock(fs_info);
422 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
423 struct seq_list *elem)
425 struct rb_root *tm_root;
426 struct rb_node *node;
427 struct rb_node *next;
428 struct seq_list *cur_elem;
429 struct tree_mod_elem *tm;
430 u64 min_seq = (u64)-1;
431 u64 seq_putting = elem->seq;
436 spin_lock(&fs_info->tree_mod_seq_lock);
437 list_del(&elem->list);
440 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
441 if (cur_elem->seq < min_seq) {
442 if (seq_putting > cur_elem->seq) {
444 * blocker with lower sequence number exists, we
445 * cannot remove anything from the log
447 spin_unlock(&fs_info->tree_mod_seq_lock);
450 min_seq = cur_elem->seq;
453 spin_unlock(&fs_info->tree_mod_seq_lock);
456 * anything that's lower than the lowest existing (read: blocked)
457 * sequence number can be removed from the tree.
459 tree_mod_log_write_lock(fs_info);
460 tm_root = &fs_info->tree_mod_log;
461 for (node = rb_first(tm_root); node; node = next) {
462 next = rb_next(node);
463 tm = container_of(node, struct tree_mod_elem, node);
464 if (tm->seq > min_seq)
466 rb_erase(node, tm_root);
469 tree_mod_log_write_unlock(fs_info);
473 * key order of the log:
476 * the index is the shifted logical of the *new* root node for root replace
477 * operations, or the shifted logical of the affected block for all other
481 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
483 struct rb_root *tm_root;
484 struct rb_node **new;
485 struct rb_node *parent = NULL;
486 struct tree_mod_elem *cur;
488 BUG_ON(!tm || !tm->seq);
490 tm_root = &fs_info->tree_mod_log;
491 new = &tm_root->rb_node;
493 cur = container_of(*new, struct tree_mod_elem, node);
495 if (cur->index < tm->index)
496 new = &((*new)->rb_left);
497 else if (cur->index > tm->index)
498 new = &((*new)->rb_right);
499 else if (cur->seq < tm->seq)
500 new = &((*new)->rb_left);
501 else if (cur->seq > tm->seq)
502 new = &((*new)->rb_right);
509 rb_link_node(&tm->node, parent, new);
510 rb_insert_color(&tm->node, tm_root);
515 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
516 * returns zero with the tree_mod_log_lock acquired. The caller must hold
517 * this until all tree mod log insertions are recorded in the rb tree and then
518 * call tree_mod_log_write_unlock() to release.
520 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
521 struct extent_buffer *eb) {
523 if (list_empty(&(fs_info)->tree_mod_seq_list))
525 if (eb && btrfs_header_level(eb) == 0)
528 tree_mod_log_write_lock(fs_info);
529 if (list_empty(&fs_info->tree_mod_seq_list)) {
531 * someone emptied the list while we were waiting for the lock.
532 * we must not add to the list when no blocker exists.
534 tree_mod_log_write_unlock(fs_info);
542 * This allocates memory and gets a tree modification sequence number.
544 * Returns <0 on error.
545 * Returns >0 (the added sequence number) on success.
547 static inline int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags,
548 struct tree_mod_elem **tm_ret)
550 struct tree_mod_elem *tm;
553 * once we switch from spin locks to something different, we should
554 * honor the flags parameter here.
556 tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
560 spin_lock(&fs_info->tree_mod_seq_lock);
561 tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info);
562 spin_unlock(&fs_info->tree_mod_seq_lock);
568 __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
569 struct extent_buffer *eb, int slot,
570 enum mod_log_op op, gfp_t flags)
573 struct tree_mod_elem *tm;
575 ret = tree_mod_alloc(fs_info, flags, &tm);
579 tm->index = eb->start >> PAGE_CACHE_SHIFT;
580 if (op != MOD_LOG_KEY_ADD) {
581 btrfs_node_key(eb, &tm->key, slot);
582 tm->blockptr = btrfs_node_blockptr(eb, slot);
586 tm->generation = btrfs_node_ptr_generation(eb, slot);
588 return __tree_mod_log_insert(fs_info, tm);
592 tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
593 struct extent_buffer *eb, int slot,
594 enum mod_log_op op, gfp_t flags)
598 if (tree_mod_dont_log(fs_info, eb))
601 ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
603 tree_mod_log_write_unlock(fs_info);
608 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
609 int slot, enum mod_log_op op)
611 return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS);
615 tree_mod_log_insert_key_locked(struct btrfs_fs_info *fs_info,
616 struct extent_buffer *eb, int slot,
619 return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
623 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
624 struct extent_buffer *eb, int dst_slot, int src_slot,
625 int nr_items, gfp_t flags)
627 struct tree_mod_elem *tm;
631 if (tree_mod_dont_log(fs_info, eb))
635 * When we override something during the move, we log these removals.
636 * This can only happen when we move towards the beginning of the
637 * buffer, i.e. dst_slot < src_slot.
639 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
640 ret = tree_mod_log_insert_key_locked(fs_info, eb, i + dst_slot,
641 MOD_LOG_KEY_REMOVE_WHILE_MOVING);
645 ret = tree_mod_alloc(fs_info, flags, &tm);
649 tm->index = eb->start >> PAGE_CACHE_SHIFT;
651 tm->move.dst_slot = dst_slot;
652 tm->move.nr_items = nr_items;
653 tm->op = MOD_LOG_MOVE_KEYS;
655 ret = __tree_mod_log_insert(fs_info, tm);
657 tree_mod_log_write_unlock(fs_info);
662 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
668 if (btrfs_header_level(eb) == 0)
671 nritems = btrfs_header_nritems(eb);
672 for (i = nritems - 1; i >= 0; i--) {
673 ret = tree_mod_log_insert_key_locked(fs_info, eb, i,
674 MOD_LOG_KEY_REMOVE_WHILE_FREEING);
680 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
681 struct extent_buffer *old_root,
682 struct extent_buffer *new_root, gfp_t flags,
685 struct tree_mod_elem *tm;
688 if (tree_mod_dont_log(fs_info, NULL))
692 __tree_mod_log_free_eb(fs_info, old_root);
694 ret = tree_mod_alloc(fs_info, flags, &tm);
698 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
699 tm->old_root.logical = old_root->start;
700 tm->old_root.level = btrfs_header_level(old_root);
701 tm->generation = btrfs_header_generation(old_root);
702 tm->op = MOD_LOG_ROOT_REPLACE;
704 ret = __tree_mod_log_insert(fs_info, tm);
706 tree_mod_log_write_unlock(fs_info);
710 static struct tree_mod_elem *
711 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
714 struct rb_root *tm_root;
715 struct rb_node *node;
716 struct tree_mod_elem *cur = NULL;
717 struct tree_mod_elem *found = NULL;
718 u64 index = start >> PAGE_CACHE_SHIFT;
720 tree_mod_log_read_lock(fs_info);
721 tm_root = &fs_info->tree_mod_log;
722 node = tm_root->rb_node;
724 cur = container_of(node, struct tree_mod_elem, node);
725 if (cur->index < index) {
726 node = node->rb_left;
727 } else if (cur->index > index) {
728 node = node->rb_right;
729 } else if (cur->seq < min_seq) {
730 node = node->rb_left;
731 } else if (!smallest) {
732 /* we want the node with the highest seq */
734 BUG_ON(found->seq > cur->seq);
736 node = node->rb_left;
737 } else if (cur->seq > min_seq) {
738 /* we want the node with the smallest seq */
740 BUG_ON(found->seq < cur->seq);
742 node = node->rb_right;
748 tree_mod_log_read_unlock(fs_info);
754 * this returns the element from the log with the smallest time sequence
755 * value that's in the log (the oldest log item). any element with a time
756 * sequence lower than min_seq will be ignored.
758 static struct tree_mod_elem *
759 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
762 return __tree_mod_log_search(fs_info, start, min_seq, 1);
766 * this returns the element from the log with the largest time sequence
767 * value that's in the log (the most recent log item). any element with
768 * a time sequence lower than min_seq will be ignored.
770 static struct tree_mod_elem *
771 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
773 return __tree_mod_log_search(fs_info, start, min_seq, 0);
777 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
778 struct extent_buffer *src, unsigned long dst_offset,
779 unsigned long src_offset, int nr_items)
784 if (tree_mod_dont_log(fs_info, NULL))
787 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) {
788 tree_mod_log_write_unlock(fs_info);
792 for (i = 0; i < nr_items; i++) {
793 ret = tree_mod_log_insert_key_locked(fs_info, src,
797 ret = tree_mod_log_insert_key_locked(fs_info, dst,
803 tree_mod_log_write_unlock(fs_info);
807 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
808 int dst_offset, int src_offset, int nr_items)
811 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
817 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
818 struct extent_buffer *eb, int slot, int atomic)
822 ret = tree_mod_log_insert_key_mask(fs_info, eb, slot,
824 atomic ? GFP_ATOMIC : GFP_NOFS);
829 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
831 if (tree_mod_dont_log(fs_info, eb))
834 __tree_mod_log_free_eb(fs_info, eb);
836 tree_mod_log_write_unlock(fs_info);
840 tree_mod_log_set_root_pointer(struct btrfs_root *root,
841 struct extent_buffer *new_root_node,
845 ret = tree_mod_log_insert_root(root->fs_info, root->node,
846 new_root_node, GFP_NOFS, log_removal);
851 * check if the tree block can be shared by multiple trees
853 int btrfs_block_can_be_shared(struct btrfs_root *root,
854 struct extent_buffer *buf)
857 * Tree blocks not in refernece counted trees and tree roots
858 * are never shared. If a block was allocated after the last
859 * snapshot and the block was not allocated by tree relocation,
860 * we know the block is not shared.
862 if (root->ref_cows &&
863 buf != root->node && buf != root->commit_root &&
864 (btrfs_header_generation(buf) <=
865 btrfs_root_last_snapshot(&root->root_item) ||
866 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
868 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
869 if (root->ref_cows &&
870 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
876 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
877 struct btrfs_root *root,
878 struct extent_buffer *buf,
879 struct extent_buffer *cow,
889 * Backrefs update rules:
891 * Always use full backrefs for extent pointers in tree block
892 * allocated by tree relocation.
894 * If a shared tree block is no longer referenced by its owner
895 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
896 * use full backrefs for extent pointers in tree block.
898 * If a tree block is been relocating
899 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
900 * use full backrefs for extent pointers in tree block.
901 * The reason for this is some operations (such as drop tree)
902 * are only allowed for blocks use full backrefs.
905 if (btrfs_block_can_be_shared(root, buf)) {
906 ret = btrfs_lookup_extent_info(trans, root, buf->start,
907 btrfs_header_level(buf), 1,
913 btrfs_std_error(root->fs_info, ret);
918 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
919 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
920 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
925 owner = btrfs_header_owner(buf);
926 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
927 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
930 if ((owner == root->root_key.objectid ||
931 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
932 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
933 ret = btrfs_inc_ref(trans, root, buf, 1, 1);
934 BUG_ON(ret); /* -ENOMEM */
936 if (root->root_key.objectid ==
937 BTRFS_TREE_RELOC_OBJECTID) {
938 ret = btrfs_dec_ref(trans, root, buf, 0, 1);
939 BUG_ON(ret); /* -ENOMEM */
940 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
941 BUG_ON(ret); /* -ENOMEM */
943 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
946 if (root->root_key.objectid ==
947 BTRFS_TREE_RELOC_OBJECTID)
948 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
950 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
951 BUG_ON(ret); /* -ENOMEM */
953 if (new_flags != 0) {
954 int level = btrfs_header_level(buf);
956 ret = btrfs_set_disk_extent_flags(trans, root,
959 new_flags, level, 0);
964 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
965 if (root->root_key.objectid ==
966 BTRFS_TREE_RELOC_OBJECTID)
967 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
969 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
970 BUG_ON(ret); /* -ENOMEM */
971 ret = btrfs_dec_ref(trans, root, buf, 1, 1);
972 BUG_ON(ret); /* -ENOMEM */
974 clean_tree_block(trans, root, buf);
981 * does the dirty work in cow of a single block. The parent block (if
982 * supplied) is updated to point to the new cow copy. The new buffer is marked
983 * dirty and returned locked. If you modify the block it needs to be marked
986 * search_start -- an allocation hint for the new block
988 * empty_size -- a hint that you plan on doing more cow. This is the size in
989 * bytes the allocator should try to find free next to the block it returns.
990 * This is just a hint and may be ignored by the allocator.
992 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
993 struct btrfs_root *root,
994 struct extent_buffer *buf,
995 struct extent_buffer *parent, int parent_slot,
996 struct extent_buffer **cow_ret,
997 u64 search_start, u64 empty_size)
999 struct btrfs_disk_key disk_key;
1000 struct extent_buffer *cow;
1003 int unlock_orig = 0;
1006 if (*cow_ret == buf)
1009 btrfs_assert_tree_locked(buf);
1011 WARN_ON(root->ref_cows && trans->transid !=
1012 root->fs_info->running_transaction->transid);
1013 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
1015 level = btrfs_header_level(buf);
1018 btrfs_item_key(buf, &disk_key, 0);
1020 btrfs_node_key(buf, &disk_key, 0);
1022 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1024 parent_start = parent->start;
1030 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
1031 root->root_key.objectid, &disk_key,
1032 level, search_start, empty_size);
1034 return PTR_ERR(cow);
1036 /* cow is set to blocking by btrfs_init_new_buffer */
1038 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1039 btrfs_set_header_bytenr(cow, cow->start);
1040 btrfs_set_header_generation(cow, trans->transid);
1041 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1042 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1043 BTRFS_HEADER_FLAG_RELOC);
1044 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1045 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1047 btrfs_set_header_owner(cow, root->root_key.objectid);
1049 write_extent_buffer(cow, root->fs_info->fsid,
1050 (unsigned long)btrfs_header_fsid(cow),
1053 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1055 btrfs_abort_transaction(trans, root, ret);
1060 btrfs_reloc_cow_block(trans, root, buf, cow);
1062 if (buf == root->node) {
1063 WARN_ON(parent && parent != buf);
1064 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1065 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1066 parent_start = buf->start;
1070 extent_buffer_get(cow);
1071 tree_mod_log_set_root_pointer(root, cow, 1);
1072 rcu_assign_pointer(root->node, cow);
1074 btrfs_free_tree_block(trans, root, buf, parent_start,
1076 free_extent_buffer(buf);
1077 add_root_to_dirty_list(root);
1079 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1080 parent_start = parent->start;
1084 WARN_ON(trans->transid != btrfs_header_generation(parent));
1085 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1086 MOD_LOG_KEY_REPLACE);
1087 btrfs_set_node_blockptr(parent, parent_slot,
1089 btrfs_set_node_ptr_generation(parent, parent_slot,
1091 btrfs_mark_buffer_dirty(parent);
1093 tree_mod_log_free_eb(root->fs_info, buf);
1094 btrfs_free_tree_block(trans, root, buf, parent_start,
1098 btrfs_tree_unlock(buf);
1099 free_extent_buffer_stale(buf);
1100 btrfs_mark_buffer_dirty(cow);
1106 * returns the logical address of the oldest predecessor of the given root.
1107 * entries older than time_seq are ignored.
1109 static struct tree_mod_elem *
1110 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1111 struct extent_buffer *eb_root, u64 time_seq)
1113 struct tree_mod_elem *tm;
1114 struct tree_mod_elem *found = NULL;
1115 u64 root_logical = eb_root->start;
1122 * the very last operation that's logged for a root is the replacement
1123 * operation (if it is replaced at all). this has the index of the *new*
1124 * root, making it the very first operation that's logged for this root.
1127 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1132 * if there are no tree operation for the oldest root, we simply
1133 * return it. this should only happen if that (old) root is at
1140 * if there's an operation that's not a root replacement, we
1141 * found the oldest version of our root. normally, we'll find a
1142 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1144 if (tm->op != MOD_LOG_ROOT_REPLACE)
1148 root_logical = tm->old_root.logical;
1152 /* if there's no old root to return, return what we found instead */
1160 * tm is a pointer to the first operation to rewind within eb. then, all
1161 * previous operations will be rewinded (until we reach something older than
1165 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1166 u64 time_seq, struct tree_mod_elem *first_tm)
1169 struct rb_node *next;
1170 struct tree_mod_elem *tm = first_tm;
1171 unsigned long o_dst;
1172 unsigned long o_src;
1173 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1175 n = btrfs_header_nritems(eb);
1176 tree_mod_log_read_lock(fs_info);
1177 while (tm && tm->seq >= time_seq) {
1179 * all the operations are recorded with the operator used for
1180 * the modification. as we're going backwards, we do the
1181 * opposite of each operation here.
1184 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1185 BUG_ON(tm->slot < n);
1187 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1188 case MOD_LOG_KEY_REMOVE:
1189 btrfs_set_node_key(eb, &tm->key, tm->slot);
1190 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1191 btrfs_set_node_ptr_generation(eb, tm->slot,
1195 case MOD_LOG_KEY_REPLACE:
1196 BUG_ON(tm->slot >= n);
1197 btrfs_set_node_key(eb, &tm->key, tm->slot);
1198 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1199 btrfs_set_node_ptr_generation(eb, tm->slot,
1202 case MOD_LOG_KEY_ADD:
1203 /* if a move operation is needed it's in the log */
1206 case MOD_LOG_MOVE_KEYS:
1207 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1208 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1209 memmove_extent_buffer(eb, o_dst, o_src,
1210 tm->move.nr_items * p_size);
1212 case MOD_LOG_ROOT_REPLACE:
1214 * this operation is special. for roots, this must be
1215 * handled explicitly before rewinding.
1216 * for non-roots, this operation may exist if the node
1217 * was a root: root A -> child B; then A gets empty and
1218 * B is promoted to the new root. in the mod log, we'll
1219 * have a root-replace operation for B, a tree block
1220 * that is no root. we simply ignore that operation.
1224 next = rb_next(&tm->node);
1227 tm = container_of(next, struct tree_mod_elem, node);
1228 if (tm->index != first_tm->index)
1231 tree_mod_log_read_unlock(fs_info);
1232 btrfs_set_header_nritems(eb, n);
1236 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1237 * is returned. If rewind operations happen, a fresh buffer is returned. The
1238 * returned buffer is always read-locked. If the returned buffer is not the
1239 * input buffer, the lock on the input buffer is released and the input buffer
1240 * is freed (its refcount is decremented).
1242 static struct extent_buffer *
1243 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1246 struct extent_buffer *eb_rewin;
1247 struct tree_mod_elem *tm;
1252 if (btrfs_header_level(eb) == 0)
1255 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1259 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1260 BUG_ON(tm->slot != 0);
1261 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1262 fs_info->tree_root->nodesize);
1264 btrfs_set_header_bytenr(eb_rewin, eb->start);
1265 btrfs_set_header_backref_rev(eb_rewin,
1266 btrfs_header_backref_rev(eb));
1267 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1268 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1270 eb_rewin = btrfs_clone_extent_buffer(eb);
1274 btrfs_tree_read_unlock(eb);
1275 free_extent_buffer(eb);
1277 extent_buffer_get(eb_rewin);
1278 btrfs_tree_read_lock(eb_rewin);
1279 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1280 WARN_ON(btrfs_header_nritems(eb_rewin) >
1281 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1287 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1288 * value. If there are no changes, the current root->root_node is returned. If
1289 * anything changed in between, there's a fresh buffer allocated on which the
1290 * rewind operations are done. In any case, the returned buffer is read locked.
1291 * Returns NULL on error (with no locks held).
1293 static inline struct extent_buffer *
1294 get_old_root(struct btrfs_root *root, u64 time_seq)
1296 struct tree_mod_elem *tm;
1297 struct extent_buffer *eb = NULL;
1298 struct extent_buffer *eb_root;
1299 struct extent_buffer *old;
1300 struct tree_mod_root *old_root = NULL;
1301 u64 old_generation = 0;
1305 eb_root = btrfs_read_lock_root_node(root);
1306 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1310 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1311 old_root = &tm->old_root;
1312 old_generation = tm->generation;
1313 logical = old_root->logical;
1315 logical = eb_root->start;
1318 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1319 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1320 btrfs_tree_read_unlock(eb_root);
1321 free_extent_buffer(eb_root);
1322 blocksize = btrfs_level_size(root, old_root->level);
1323 old = read_tree_block(root, logical, blocksize, 0);
1324 if (!old || !extent_buffer_uptodate(old)) {
1325 free_extent_buffer(old);
1326 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1330 eb = btrfs_clone_extent_buffer(old);
1331 free_extent_buffer(old);
1333 } else if (old_root) {
1334 btrfs_tree_read_unlock(eb_root);
1335 free_extent_buffer(eb_root);
1336 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1338 eb = btrfs_clone_extent_buffer(eb_root);
1339 btrfs_tree_read_unlock(eb_root);
1340 free_extent_buffer(eb_root);
1345 extent_buffer_get(eb);
1346 btrfs_tree_read_lock(eb);
1348 btrfs_set_header_bytenr(eb, eb->start);
1349 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1350 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1351 btrfs_set_header_level(eb, old_root->level);
1352 btrfs_set_header_generation(eb, old_generation);
1355 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1357 WARN_ON(btrfs_header_level(eb) != 0);
1358 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1363 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1365 struct tree_mod_elem *tm;
1367 struct extent_buffer *eb_root = btrfs_root_node(root);
1369 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1370 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1371 level = tm->old_root.level;
1373 level = btrfs_header_level(eb_root);
1375 free_extent_buffer(eb_root);
1380 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1381 struct btrfs_root *root,
1382 struct extent_buffer *buf)
1384 /* ensure we can see the force_cow */
1388 * We do not need to cow a block if
1389 * 1) this block is not created or changed in this transaction;
1390 * 2) this block does not belong to TREE_RELOC tree;
1391 * 3) the root is not forced COW.
1393 * What is forced COW:
1394 * when we create snapshot during commiting the transaction,
1395 * after we've finished coping src root, we must COW the shared
1396 * block to ensure the metadata consistency.
1398 if (btrfs_header_generation(buf) == trans->transid &&
1399 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1400 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1401 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1408 * cows a single block, see __btrfs_cow_block for the real work.
1409 * This version of it has extra checks so that a block isn't cow'd more than
1410 * once per transaction, as long as it hasn't been written yet
1412 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1413 struct btrfs_root *root, struct extent_buffer *buf,
1414 struct extent_buffer *parent, int parent_slot,
1415 struct extent_buffer **cow_ret)
1420 if (trans->transaction != root->fs_info->running_transaction)
1421 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1422 (unsigned long long)trans->transid,
1423 (unsigned long long)
1424 root->fs_info->running_transaction->transid);
1426 if (trans->transid != root->fs_info->generation)
1427 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1428 (unsigned long long)trans->transid,
1429 (unsigned long long)root->fs_info->generation);
1431 if (!should_cow_block(trans, root, buf)) {
1436 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1439 btrfs_set_lock_blocking(parent);
1440 btrfs_set_lock_blocking(buf);
1442 ret = __btrfs_cow_block(trans, root, buf, parent,
1443 parent_slot, cow_ret, search_start, 0);
1445 trace_btrfs_cow_block(root, buf, *cow_ret);
1451 * helper function for defrag to decide if two blocks pointed to by a
1452 * node are actually close by
1454 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1456 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1458 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1464 * compare two keys in a memcmp fashion
1466 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1468 struct btrfs_key k1;
1470 btrfs_disk_key_to_cpu(&k1, disk);
1472 return btrfs_comp_cpu_keys(&k1, k2);
1476 * same as comp_keys only with two btrfs_key's
1478 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1480 if (k1->objectid > k2->objectid)
1482 if (k1->objectid < k2->objectid)
1484 if (k1->type > k2->type)
1486 if (k1->type < k2->type)
1488 if (k1->offset > k2->offset)
1490 if (k1->offset < k2->offset)
1496 * this is used by the defrag code to go through all the
1497 * leaves pointed to by a node and reallocate them so that
1498 * disk order is close to key order
1500 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1501 struct btrfs_root *root, struct extent_buffer *parent,
1502 int start_slot, u64 *last_ret,
1503 struct btrfs_key *progress)
1505 struct extent_buffer *cur;
1508 u64 search_start = *last_ret;
1518 int progress_passed = 0;
1519 struct btrfs_disk_key disk_key;
1521 parent_level = btrfs_header_level(parent);
1523 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1524 WARN_ON(trans->transid != root->fs_info->generation);
1526 parent_nritems = btrfs_header_nritems(parent);
1527 blocksize = btrfs_level_size(root, parent_level - 1);
1528 end_slot = parent_nritems;
1530 if (parent_nritems == 1)
1533 btrfs_set_lock_blocking(parent);
1535 for (i = start_slot; i < end_slot; i++) {
1538 btrfs_node_key(parent, &disk_key, i);
1539 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1542 progress_passed = 1;
1543 blocknr = btrfs_node_blockptr(parent, i);
1544 gen = btrfs_node_ptr_generation(parent, i);
1545 if (last_block == 0)
1546 last_block = blocknr;
1549 other = btrfs_node_blockptr(parent, i - 1);
1550 close = close_blocks(blocknr, other, blocksize);
1552 if (!close && i < end_slot - 2) {
1553 other = btrfs_node_blockptr(parent, i + 1);
1554 close = close_blocks(blocknr, other, blocksize);
1557 last_block = blocknr;
1561 cur = btrfs_find_tree_block(root, blocknr, blocksize);
1563 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1566 if (!cur || !uptodate) {
1568 cur = read_tree_block(root, blocknr,
1570 if (!cur || !extent_buffer_uptodate(cur)) {
1571 free_extent_buffer(cur);
1574 } else if (!uptodate) {
1575 err = btrfs_read_buffer(cur, gen);
1577 free_extent_buffer(cur);
1582 if (search_start == 0)
1583 search_start = last_block;
1585 btrfs_tree_lock(cur);
1586 btrfs_set_lock_blocking(cur);
1587 err = __btrfs_cow_block(trans, root, cur, parent, i,
1590 (end_slot - i) * blocksize));
1592 btrfs_tree_unlock(cur);
1593 free_extent_buffer(cur);
1596 search_start = cur->start;
1597 last_block = cur->start;
1598 *last_ret = search_start;
1599 btrfs_tree_unlock(cur);
1600 free_extent_buffer(cur);
1606 * The leaf data grows from end-to-front in the node.
1607 * this returns the address of the start of the last item,
1608 * which is the stop of the leaf data stack
1610 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1611 struct extent_buffer *leaf)
1613 u32 nr = btrfs_header_nritems(leaf);
1615 return BTRFS_LEAF_DATA_SIZE(root);
1616 return btrfs_item_offset_nr(leaf, nr - 1);
1621 * search for key in the extent_buffer. The items start at offset p,
1622 * and they are item_size apart. There are 'max' items in p.
1624 * the slot in the array is returned via slot, and it points to
1625 * the place where you would insert key if it is not found in
1628 * slot may point to max if the key is bigger than all of the keys
1630 static noinline int generic_bin_search(struct extent_buffer *eb,
1632 int item_size, struct btrfs_key *key,
1639 struct btrfs_disk_key *tmp = NULL;
1640 struct btrfs_disk_key unaligned;
1641 unsigned long offset;
1643 unsigned long map_start = 0;
1644 unsigned long map_len = 0;
1647 while (low < high) {
1648 mid = (low + high) / 2;
1649 offset = p + mid * item_size;
1651 if (!kaddr || offset < map_start ||
1652 (offset + sizeof(struct btrfs_disk_key)) >
1653 map_start + map_len) {
1655 err = map_private_extent_buffer(eb, offset,
1656 sizeof(struct btrfs_disk_key),
1657 &kaddr, &map_start, &map_len);
1660 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1663 read_extent_buffer(eb, &unaligned,
1664 offset, sizeof(unaligned));
1669 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1672 ret = comp_keys(tmp, key);
1688 * simple bin_search frontend that does the right thing for
1691 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1692 int level, int *slot)
1695 return generic_bin_search(eb,
1696 offsetof(struct btrfs_leaf, items),
1697 sizeof(struct btrfs_item),
1698 key, btrfs_header_nritems(eb),
1701 return generic_bin_search(eb,
1702 offsetof(struct btrfs_node, ptrs),
1703 sizeof(struct btrfs_key_ptr),
1704 key, btrfs_header_nritems(eb),
1708 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1709 int level, int *slot)
1711 return bin_search(eb, key, level, slot);
1714 static void root_add_used(struct btrfs_root *root, u32 size)
1716 spin_lock(&root->accounting_lock);
1717 btrfs_set_root_used(&root->root_item,
1718 btrfs_root_used(&root->root_item) + size);
1719 spin_unlock(&root->accounting_lock);
1722 static void root_sub_used(struct btrfs_root *root, u32 size)
1724 spin_lock(&root->accounting_lock);
1725 btrfs_set_root_used(&root->root_item,
1726 btrfs_root_used(&root->root_item) - size);
1727 spin_unlock(&root->accounting_lock);
1730 /* given a node and slot number, this reads the blocks it points to. The
1731 * extent buffer is returned with a reference taken (but unlocked).
1732 * NULL is returned on error.
1734 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1735 struct extent_buffer *parent, int slot)
1737 int level = btrfs_header_level(parent);
1738 struct extent_buffer *eb;
1742 if (slot >= btrfs_header_nritems(parent))
1747 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1748 btrfs_level_size(root, level - 1),
1749 btrfs_node_ptr_generation(parent, slot));
1750 if (eb && !extent_buffer_uptodate(eb)) {
1751 free_extent_buffer(eb);
1759 * node level balancing, used to make sure nodes are in proper order for
1760 * item deletion. We balance from the top down, so we have to make sure
1761 * that a deletion won't leave an node completely empty later on.
1763 static noinline int balance_level(struct btrfs_trans_handle *trans,
1764 struct btrfs_root *root,
1765 struct btrfs_path *path, int level)
1767 struct extent_buffer *right = NULL;
1768 struct extent_buffer *mid;
1769 struct extent_buffer *left = NULL;
1770 struct extent_buffer *parent = NULL;
1774 int orig_slot = path->slots[level];
1780 mid = path->nodes[level];
1782 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1783 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1784 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1786 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1788 if (level < BTRFS_MAX_LEVEL - 1) {
1789 parent = path->nodes[level + 1];
1790 pslot = path->slots[level + 1];
1794 * deal with the case where there is only one pointer in the root
1795 * by promoting the node below to a root
1798 struct extent_buffer *child;
1800 if (btrfs_header_nritems(mid) != 1)
1803 /* promote the child to a root */
1804 child = read_node_slot(root, mid, 0);
1807 btrfs_std_error(root->fs_info, ret);
1811 btrfs_tree_lock(child);
1812 btrfs_set_lock_blocking(child);
1813 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1815 btrfs_tree_unlock(child);
1816 free_extent_buffer(child);
1820 tree_mod_log_set_root_pointer(root, child, 1);
1821 rcu_assign_pointer(root->node, child);
1823 add_root_to_dirty_list(root);
1824 btrfs_tree_unlock(child);
1826 path->locks[level] = 0;
1827 path->nodes[level] = NULL;
1828 clean_tree_block(trans, root, mid);
1829 btrfs_tree_unlock(mid);
1830 /* once for the path */
1831 free_extent_buffer(mid);
1833 root_sub_used(root, mid->len);
1834 btrfs_free_tree_block(trans, root, mid, 0, 1);
1835 /* once for the root ptr */
1836 free_extent_buffer_stale(mid);
1839 if (btrfs_header_nritems(mid) >
1840 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1843 left = read_node_slot(root, parent, pslot - 1);
1845 btrfs_tree_lock(left);
1846 btrfs_set_lock_blocking(left);
1847 wret = btrfs_cow_block(trans, root, left,
1848 parent, pslot - 1, &left);
1854 right = read_node_slot(root, parent, pslot + 1);
1856 btrfs_tree_lock(right);
1857 btrfs_set_lock_blocking(right);
1858 wret = btrfs_cow_block(trans, root, right,
1859 parent, pslot + 1, &right);
1866 /* first, try to make some room in the middle buffer */
1868 orig_slot += btrfs_header_nritems(left);
1869 wret = push_node_left(trans, root, left, mid, 1);
1875 * then try to empty the right most buffer into the middle
1878 wret = push_node_left(trans, root, mid, right, 1);
1879 if (wret < 0 && wret != -ENOSPC)
1881 if (btrfs_header_nritems(right) == 0) {
1882 clean_tree_block(trans, root, right);
1883 btrfs_tree_unlock(right);
1884 del_ptr(root, path, level + 1, pslot + 1);
1885 root_sub_used(root, right->len);
1886 btrfs_free_tree_block(trans, root, right, 0, 1);
1887 free_extent_buffer_stale(right);
1890 struct btrfs_disk_key right_key;
1891 btrfs_node_key(right, &right_key, 0);
1892 tree_mod_log_set_node_key(root->fs_info, parent,
1894 btrfs_set_node_key(parent, &right_key, pslot + 1);
1895 btrfs_mark_buffer_dirty(parent);
1898 if (btrfs_header_nritems(mid) == 1) {
1900 * we're not allowed to leave a node with one item in the
1901 * tree during a delete. A deletion from lower in the tree
1902 * could try to delete the only pointer in this node.
1903 * So, pull some keys from the left.
1904 * There has to be a left pointer at this point because
1905 * otherwise we would have pulled some pointers from the
1910 btrfs_std_error(root->fs_info, ret);
1913 wret = balance_node_right(trans, root, mid, left);
1919 wret = push_node_left(trans, root, left, mid, 1);
1925 if (btrfs_header_nritems(mid) == 0) {
1926 clean_tree_block(trans, root, mid);
1927 btrfs_tree_unlock(mid);
1928 del_ptr(root, path, level + 1, pslot);
1929 root_sub_used(root, mid->len);
1930 btrfs_free_tree_block(trans, root, mid, 0, 1);
1931 free_extent_buffer_stale(mid);
1934 /* update the parent key to reflect our changes */
1935 struct btrfs_disk_key mid_key;
1936 btrfs_node_key(mid, &mid_key, 0);
1937 tree_mod_log_set_node_key(root->fs_info, parent,
1939 btrfs_set_node_key(parent, &mid_key, pslot);
1940 btrfs_mark_buffer_dirty(parent);
1943 /* update the path */
1945 if (btrfs_header_nritems(left) > orig_slot) {
1946 extent_buffer_get(left);
1947 /* left was locked after cow */
1948 path->nodes[level] = left;
1949 path->slots[level + 1] -= 1;
1950 path->slots[level] = orig_slot;
1952 btrfs_tree_unlock(mid);
1953 free_extent_buffer(mid);
1956 orig_slot -= btrfs_header_nritems(left);
1957 path->slots[level] = orig_slot;
1960 /* double check we haven't messed things up */
1962 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1966 btrfs_tree_unlock(right);
1967 free_extent_buffer(right);
1970 if (path->nodes[level] != left)
1971 btrfs_tree_unlock(left);
1972 free_extent_buffer(left);
1977 /* Node balancing for insertion. Here we only split or push nodes around
1978 * when they are completely full. This is also done top down, so we
1979 * have to be pessimistic.
1981 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1982 struct btrfs_root *root,
1983 struct btrfs_path *path, int level)
1985 struct extent_buffer *right = NULL;
1986 struct extent_buffer *mid;
1987 struct extent_buffer *left = NULL;
1988 struct extent_buffer *parent = NULL;
1992 int orig_slot = path->slots[level];
1997 mid = path->nodes[level];
1998 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2000 if (level < BTRFS_MAX_LEVEL - 1) {
2001 parent = path->nodes[level + 1];
2002 pslot = path->slots[level + 1];
2008 left = read_node_slot(root, parent, pslot - 1);
2010 /* first, try to make some room in the middle buffer */
2014 btrfs_tree_lock(left);
2015 btrfs_set_lock_blocking(left);
2017 left_nr = btrfs_header_nritems(left);
2018 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2021 ret = btrfs_cow_block(trans, root, left, parent,
2026 wret = push_node_left(trans, root,
2033 struct btrfs_disk_key disk_key;
2034 orig_slot += left_nr;
2035 btrfs_node_key(mid, &disk_key, 0);
2036 tree_mod_log_set_node_key(root->fs_info, parent,
2038 btrfs_set_node_key(parent, &disk_key, pslot);
2039 btrfs_mark_buffer_dirty(parent);
2040 if (btrfs_header_nritems(left) > orig_slot) {
2041 path->nodes[level] = left;
2042 path->slots[level + 1] -= 1;
2043 path->slots[level] = orig_slot;
2044 btrfs_tree_unlock(mid);
2045 free_extent_buffer(mid);
2048 btrfs_header_nritems(left);
2049 path->slots[level] = orig_slot;
2050 btrfs_tree_unlock(left);
2051 free_extent_buffer(left);
2055 btrfs_tree_unlock(left);
2056 free_extent_buffer(left);
2058 right = read_node_slot(root, parent, pslot + 1);
2061 * then try to empty the right most buffer into the middle
2066 btrfs_tree_lock(right);
2067 btrfs_set_lock_blocking(right);
2069 right_nr = btrfs_header_nritems(right);
2070 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2073 ret = btrfs_cow_block(trans, root, right,
2079 wret = balance_node_right(trans, root,
2086 struct btrfs_disk_key disk_key;
2088 btrfs_node_key(right, &disk_key, 0);
2089 tree_mod_log_set_node_key(root->fs_info, parent,
2091 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2092 btrfs_mark_buffer_dirty(parent);
2094 if (btrfs_header_nritems(mid) <= orig_slot) {
2095 path->nodes[level] = right;
2096 path->slots[level + 1] += 1;
2097 path->slots[level] = orig_slot -
2098 btrfs_header_nritems(mid);
2099 btrfs_tree_unlock(mid);
2100 free_extent_buffer(mid);
2102 btrfs_tree_unlock(right);
2103 free_extent_buffer(right);
2107 btrfs_tree_unlock(right);
2108 free_extent_buffer(right);
2114 * readahead one full node of leaves, finding things that are close
2115 * to the block in 'slot', and triggering ra on them.
2117 static void reada_for_search(struct btrfs_root *root,
2118 struct btrfs_path *path,
2119 int level, int slot, u64 objectid)
2121 struct extent_buffer *node;
2122 struct btrfs_disk_key disk_key;
2128 int direction = path->reada;
2129 struct extent_buffer *eb;
2137 if (!path->nodes[level])
2140 node = path->nodes[level];
2142 search = btrfs_node_blockptr(node, slot);
2143 blocksize = btrfs_level_size(root, level - 1);
2144 eb = btrfs_find_tree_block(root, search, blocksize);
2146 free_extent_buffer(eb);
2152 nritems = btrfs_header_nritems(node);
2156 if (direction < 0) {
2160 } else if (direction > 0) {
2165 if (path->reada < 0 && objectid) {
2166 btrfs_node_key(node, &disk_key, nr);
2167 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2170 search = btrfs_node_blockptr(node, nr);
2171 if ((search <= target && target - search <= 65536) ||
2172 (search > target && search - target <= 65536)) {
2173 gen = btrfs_node_ptr_generation(node, nr);
2174 readahead_tree_block(root, search, blocksize, gen);
2178 if ((nread > 65536 || nscan > 32))
2183 static noinline void reada_for_balance(struct btrfs_root *root,
2184 struct btrfs_path *path, int level)
2188 struct extent_buffer *parent;
2189 struct extent_buffer *eb;
2195 parent = path->nodes[level + 1];
2199 nritems = btrfs_header_nritems(parent);
2200 slot = path->slots[level + 1];
2201 blocksize = btrfs_level_size(root, level);
2204 block1 = btrfs_node_blockptr(parent, slot - 1);
2205 gen = btrfs_node_ptr_generation(parent, slot - 1);
2206 eb = btrfs_find_tree_block(root, block1, blocksize);
2208 * if we get -eagain from btrfs_buffer_uptodate, we
2209 * don't want to return eagain here. That will loop
2212 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2214 free_extent_buffer(eb);
2216 if (slot + 1 < nritems) {
2217 block2 = btrfs_node_blockptr(parent, slot + 1);
2218 gen = btrfs_node_ptr_generation(parent, slot + 1);
2219 eb = btrfs_find_tree_block(root, block2, blocksize);
2220 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2222 free_extent_buffer(eb);
2226 readahead_tree_block(root, block1, blocksize, 0);
2228 readahead_tree_block(root, block2, blocksize, 0);
2233 * when we walk down the tree, it is usually safe to unlock the higher layers
2234 * in the tree. The exceptions are when our path goes through slot 0, because
2235 * operations on the tree might require changing key pointers higher up in the
2238 * callers might also have set path->keep_locks, which tells this code to keep
2239 * the lock if the path points to the last slot in the block. This is part of
2240 * walking through the tree, and selecting the next slot in the higher block.
2242 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2243 * if lowest_unlock is 1, level 0 won't be unlocked
2245 static noinline void unlock_up(struct btrfs_path *path, int level,
2246 int lowest_unlock, int min_write_lock_level,
2247 int *write_lock_level)
2250 int skip_level = level;
2252 struct extent_buffer *t;
2254 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2255 if (!path->nodes[i])
2257 if (!path->locks[i])
2259 if (!no_skips && path->slots[i] == 0) {
2263 if (!no_skips && path->keep_locks) {
2266 nritems = btrfs_header_nritems(t);
2267 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2272 if (skip_level < i && i >= lowest_unlock)
2276 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2277 btrfs_tree_unlock_rw(t, path->locks[i]);
2279 if (write_lock_level &&
2280 i > min_write_lock_level &&
2281 i <= *write_lock_level) {
2282 *write_lock_level = i - 1;
2289 * This releases any locks held in the path starting at level and
2290 * going all the way up to the root.
2292 * btrfs_search_slot will keep the lock held on higher nodes in a few
2293 * corner cases, such as COW of the block at slot zero in the node. This
2294 * ignores those rules, and it should only be called when there are no
2295 * more updates to be done higher up in the tree.
2297 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2301 if (path->keep_locks)
2304 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2305 if (!path->nodes[i])
2307 if (!path->locks[i])
2309 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2315 * helper function for btrfs_search_slot. The goal is to find a block
2316 * in cache without setting the path to blocking. If we find the block
2317 * we return zero and the path is unchanged.
2319 * If we can't find the block, we set the path blocking and do some
2320 * reada. -EAGAIN is returned and the search must be repeated.
2323 read_block_for_search(struct btrfs_trans_handle *trans,
2324 struct btrfs_root *root, struct btrfs_path *p,
2325 struct extent_buffer **eb_ret, int level, int slot,
2326 struct btrfs_key *key, u64 time_seq)
2331 struct extent_buffer *b = *eb_ret;
2332 struct extent_buffer *tmp;
2335 blocknr = btrfs_node_blockptr(b, slot);
2336 gen = btrfs_node_ptr_generation(b, slot);
2337 blocksize = btrfs_level_size(root, level - 1);
2339 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2341 /* first we do an atomic uptodate check */
2342 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2347 /* the pages were up to date, but we failed
2348 * the generation number check. Do a full
2349 * read for the generation number that is correct.
2350 * We must do this without dropping locks so
2351 * we can trust our generation number
2353 btrfs_set_path_blocking(p);
2355 /* now we're allowed to do a blocking uptodate check */
2356 ret = btrfs_read_buffer(tmp, gen);
2361 free_extent_buffer(tmp);
2362 btrfs_release_path(p);
2367 * reduce lock contention at high levels
2368 * of the btree by dropping locks before
2369 * we read. Don't release the lock on the current
2370 * level because we need to walk this node to figure
2371 * out which blocks to read.
2373 btrfs_unlock_up_safe(p, level + 1);
2374 btrfs_set_path_blocking(p);
2376 free_extent_buffer(tmp);
2378 reada_for_search(root, p, level, slot, key->objectid);
2380 btrfs_release_path(p);
2383 tmp = read_tree_block(root, blocknr, blocksize, 0);
2386 * If the read above didn't mark this buffer up to date,
2387 * it will never end up being up to date. Set ret to EIO now
2388 * and give up so that our caller doesn't loop forever
2391 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2393 free_extent_buffer(tmp);
2399 * helper function for btrfs_search_slot. This does all of the checks
2400 * for node-level blocks and does any balancing required based on
2403 * If no extra work was required, zero is returned. If we had to
2404 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2408 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2409 struct btrfs_root *root, struct btrfs_path *p,
2410 struct extent_buffer *b, int level, int ins_len,
2411 int *write_lock_level)
2414 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2415 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2418 if (*write_lock_level < level + 1) {
2419 *write_lock_level = level + 1;
2420 btrfs_release_path(p);
2424 btrfs_set_path_blocking(p);
2425 reada_for_balance(root, p, level);
2426 sret = split_node(trans, root, p, level);
2427 btrfs_clear_path_blocking(p, NULL, 0);
2434 b = p->nodes[level];
2435 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2436 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2439 if (*write_lock_level < level + 1) {
2440 *write_lock_level = level + 1;
2441 btrfs_release_path(p);
2445 btrfs_set_path_blocking(p);
2446 reada_for_balance(root, p, level);
2447 sret = balance_level(trans, root, p, level);
2448 btrfs_clear_path_blocking(p, NULL, 0);
2454 b = p->nodes[level];
2456 btrfs_release_path(p);
2459 BUG_ON(btrfs_header_nritems(b) == 1);
2470 * look for key in the tree. path is filled in with nodes along the way
2471 * if key is found, we return zero and you can find the item in the leaf
2472 * level of the path (level 0)
2474 * If the key isn't found, the path points to the slot where it should
2475 * be inserted, and 1 is returned. If there are other errors during the
2476 * search a negative error number is returned.
2478 * if ins_len > 0, nodes and leaves will be split as we walk down the
2479 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2482 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2483 *root, struct btrfs_key *key, struct btrfs_path *p, int
2486 struct extent_buffer *b;
2491 int lowest_unlock = 1;
2493 /* everything at write_lock_level or lower must be write locked */
2494 int write_lock_level = 0;
2495 u8 lowest_level = 0;
2496 int min_write_lock_level;
2498 lowest_level = p->lowest_level;
2499 WARN_ON(lowest_level && ins_len > 0);
2500 WARN_ON(p->nodes[0] != NULL);
2505 /* when we are removing items, we might have to go up to level
2506 * two as we update tree pointers Make sure we keep write
2507 * for those levels as well
2509 write_lock_level = 2;
2510 } else if (ins_len > 0) {
2512 * for inserting items, make sure we have a write lock on
2513 * level 1 so we can update keys
2515 write_lock_level = 1;
2519 write_lock_level = -1;
2521 if (cow && (p->keep_locks || p->lowest_level))
2522 write_lock_level = BTRFS_MAX_LEVEL;
2524 min_write_lock_level = write_lock_level;
2528 * we try very hard to do read locks on the root
2530 root_lock = BTRFS_READ_LOCK;
2532 if (p->search_commit_root) {
2534 * the commit roots are read only
2535 * so we always do read locks
2537 b = root->commit_root;
2538 extent_buffer_get(b);
2539 level = btrfs_header_level(b);
2540 if (!p->skip_locking)
2541 btrfs_tree_read_lock(b);
2543 if (p->skip_locking) {
2544 b = btrfs_root_node(root);
2545 level = btrfs_header_level(b);
2547 /* we don't know the level of the root node
2548 * until we actually have it read locked
2550 b = btrfs_read_lock_root_node(root);
2551 level = btrfs_header_level(b);
2552 if (level <= write_lock_level) {
2553 /* whoops, must trade for write lock */
2554 btrfs_tree_read_unlock(b);
2555 free_extent_buffer(b);
2556 b = btrfs_lock_root_node(root);
2557 root_lock = BTRFS_WRITE_LOCK;
2559 /* the level might have changed, check again */
2560 level = btrfs_header_level(b);
2564 p->nodes[level] = b;
2565 if (!p->skip_locking)
2566 p->locks[level] = root_lock;
2569 level = btrfs_header_level(b);
2572 * setup the path here so we can release it under lock
2573 * contention with the cow code
2577 * if we don't really need to cow this block
2578 * then we don't want to set the path blocking,
2579 * so we test it here
2581 if (!should_cow_block(trans, root, b))
2584 btrfs_set_path_blocking(p);
2587 * must have write locks on this node and the
2590 if (level > write_lock_level ||
2591 (level + 1 > write_lock_level &&
2592 level + 1 < BTRFS_MAX_LEVEL &&
2593 p->nodes[level + 1])) {
2594 write_lock_level = level + 1;
2595 btrfs_release_path(p);
2599 err = btrfs_cow_block(trans, root, b,
2600 p->nodes[level + 1],
2601 p->slots[level + 1], &b);
2608 BUG_ON(!cow && ins_len);
2610 p->nodes[level] = b;
2611 btrfs_clear_path_blocking(p, NULL, 0);
2614 * we have a lock on b and as long as we aren't changing
2615 * the tree, there is no way to for the items in b to change.
2616 * It is safe to drop the lock on our parent before we
2617 * go through the expensive btree search on b.
2619 * If cow is true, then we might be changing slot zero,
2620 * which may require changing the parent. So, we can't
2621 * drop the lock until after we know which slot we're
2625 btrfs_unlock_up_safe(p, level + 1);
2627 ret = bin_search(b, key, level, &slot);
2631 if (ret && slot > 0) {
2635 p->slots[level] = slot;
2636 err = setup_nodes_for_search(trans, root, p, b, level,
2637 ins_len, &write_lock_level);
2644 b = p->nodes[level];
2645 slot = p->slots[level];
2648 * slot 0 is special, if we change the key
2649 * we have to update the parent pointer
2650 * which means we must have a write lock
2653 if (slot == 0 && cow &&
2654 write_lock_level < level + 1) {
2655 write_lock_level = level + 1;
2656 btrfs_release_path(p);
2660 unlock_up(p, level, lowest_unlock,
2661 min_write_lock_level, &write_lock_level);
2663 if (level == lowest_level) {
2669 err = read_block_for_search(trans, root, p,
2670 &b, level, slot, key, 0);
2678 if (!p->skip_locking) {
2679 level = btrfs_header_level(b);
2680 if (level <= write_lock_level) {
2681 err = btrfs_try_tree_write_lock(b);
2683 btrfs_set_path_blocking(p);
2685 btrfs_clear_path_blocking(p, b,
2688 p->locks[level] = BTRFS_WRITE_LOCK;
2690 err = btrfs_try_tree_read_lock(b);
2692 btrfs_set_path_blocking(p);
2693 btrfs_tree_read_lock(b);
2694 btrfs_clear_path_blocking(p, b,
2697 p->locks[level] = BTRFS_READ_LOCK;
2699 p->nodes[level] = b;
2702 p->slots[level] = slot;
2704 btrfs_leaf_free_space(root, b) < ins_len) {
2705 if (write_lock_level < 1) {
2706 write_lock_level = 1;
2707 btrfs_release_path(p);
2711 btrfs_set_path_blocking(p);
2712 err = split_leaf(trans, root, key,
2713 p, ins_len, ret == 0);
2714 btrfs_clear_path_blocking(p, NULL, 0);
2722 if (!p->search_for_split)
2723 unlock_up(p, level, lowest_unlock,
2724 min_write_lock_level, &write_lock_level);
2731 * we don't really know what they plan on doing with the path
2732 * from here on, so for now just mark it as blocking
2734 if (!p->leave_spinning)
2735 btrfs_set_path_blocking(p);
2737 btrfs_release_path(p);
2742 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2743 * current state of the tree together with the operations recorded in the tree
2744 * modification log to search for the key in a previous version of this tree, as
2745 * denoted by the time_seq parameter.
2747 * Naturally, there is no support for insert, delete or cow operations.
2749 * The resulting path and return value will be set up as if we called
2750 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2752 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2753 struct btrfs_path *p, u64 time_seq)
2755 struct extent_buffer *b;
2760 int lowest_unlock = 1;
2761 u8 lowest_level = 0;
2763 lowest_level = p->lowest_level;
2764 WARN_ON(p->nodes[0] != NULL);
2766 if (p->search_commit_root) {
2768 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2772 b = get_old_root(root, time_seq);
2773 level = btrfs_header_level(b);
2774 p->locks[level] = BTRFS_READ_LOCK;
2777 level = btrfs_header_level(b);
2778 p->nodes[level] = b;
2779 btrfs_clear_path_blocking(p, NULL, 0);
2782 * we have a lock on b and as long as we aren't changing
2783 * the tree, there is no way to for the items in b to change.
2784 * It is safe to drop the lock on our parent before we
2785 * go through the expensive btree search on b.
2787 btrfs_unlock_up_safe(p, level + 1);
2789 ret = bin_search(b, key, level, &slot);
2793 if (ret && slot > 0) {
2797 p->slots[level] = slot;
2798 unlock_up(p, level, lowest_unlock, 0, NULL);
2800 if (level == lowest_level) {
2806 err = read_block_for_search(NULL, root, p, &b, level,
2807 slot, key, time_seq);
2815 level = btrfs_header_level(b);
2816 err = btrfs_try_tree_read_lock(b);
2818 btrfs_set_path_blocking(p);
2819 btrfs_tree_read_lock(b);
2820 btrfs_clear_path_blocking(p, b,
2823 b = tree_mod_log_rewind(root->fs_info, b, time_seq);
2824 p->locks[level] = BTRFS_READ_LOCK;
2825 p->nodes[level] = b;
2827 p->slots[level] = slot;
2828 unlock_up(p, level, lowest_unlock, 0, NULL);
2834 if (!p->leave_spinning)
2835 btrfs_set_path_blocking(p);
2837 btrfs_release_path(p);
2843 * helper to use instead of search slot if no exact match is needed but
2844 * instead the next or previous item should be returned.
2845 * When find_higher is true, the next higher item is returned, the next lower
2847 * When return_any and find_higher are both true, and no higher item is found,
2848 * return the next lower instead.
2849 * When return_any is true and find_higher is false, and no lower item is found,
2850 * return the next higher instead.
2851 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2854 int btrfs_search_slot_for_read(struct btrfs_root *root,
2855 struct btrfs_key *key, struct btrfs_path *p,
2856 int find_higher, int return_any)
2859 struct extent_buffer *leaf;
2862 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2866 * a return value of 1 means the path is at the position where the
2867 * item should be inserted. Normally this is the next bigger item,
2868 * but in case the previous item is the last in a leaf, path points
2869 * to the first free slot in the previous leaf, i.e. at an invalid
2875 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2876 ret = btrfs_next_leaf(root, p);
2882 * no higher item found, return the next
2887 btrfs_release_path(p);
2891 if (p->slots[0] == 0) {
2892 ret = btrfs_prev_leaf(root, p);
2896 p->slots[0] = btrfs_header_nritems(leaf) - 1;
2902 * no lower item found, return the next
2907 btrfs_release_path(p);
2917 * adjust the pointers going up the tree, starting at level
2918 * making sure the right key of each node is points to 'key'.
2919 * This is used after shifting pointers to the left, so it stops
2920 * fixing up pointers when a given leaf/node is not in slot 0 of the
2924 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
2925 struct btrfs_disk_key *key, int level)
2928 struct extent_buffer *t;
2930 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2931 int tslot = path->slots[i];
2932 if (!path->nodes[i])
2935 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
2936 btrfs_set_node_key(t, key, tslot);
2937 btrfs_mark_buffer_dirty(path->nodes[i]);
2946 * This function isn't completely safe. It's the caller's responsibility
2947 * that the new key won't break the order
2949 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
2950 struct btrfs_key *new_key)
2952 struct btrfs_disk_key disk_key;
2953 struct extent_buffer *eb;
2956 eb = path->nodes[0];
2957 slot = path->slots[0];
2959 btrfs_item_key(eb, &disk_key, slot - 1);
2960 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
2962 if (slot < btrfs_header_nritems(eb) - 1) {
2963 btrfs_item_key(eb, &disk_key, slot + 1);
2964 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
2967 btrfs_cpu_key_to_disk(&disk_key, new_key);
2968 btrfs_set_item_key(eb, &disk_key, slot);
2969 btrfs_mark_buffer_dirty(eb);
2971 fixup_low_keys(root, path, &disk_key, 1);
2975 * try to push data from one node into the next node left in the
2978 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2979 * error, and > 0 if there was no room in the left hand block.
2981 static int push_node_left(struct btrfs_trans_handle *trans,
2982 struct btrfs_root *root, struct extent_buffer *dst,
2983 struct extent_buffer *src, int empty)
2990 src_nritems = btrfs_header_nritems(src);
2991 dst_nritems = btrfs_header_nritems(dst);
2992 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2993 WARN_ON(btrfs_header_generation(src) != trans->transid);
2994 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2996 if (!empty && src_nritems <= 8)
2999 if (push_items <= 0)
3003 push_items = min(src_nritems, push_items);
3004 if (push_items < src_nritems) {
3005 /* leave at least 8 pointers in the node if
3006 * we aren't going to empty it
3008 if (src_nritems - push_items < 8) {
3009 if (push_items <= 8)
3015 push_items = min(src_nritems - 8, push_items);
3017 tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3019 copy_extent_buffer(dst, src,
3020 btrfs_node_key_ptr_offset(dst_nritems),
3021 btrfs_node_key_ptr_offset(0),
3022 push_items * sizeof(struct btrfs_key_ptr));
3024 if (push_items < src_nritems) {
3026 * don't call tree_mod_log_eb_move here, key removal was already
3027 * fully logged by tree_mod_log_eb_copy above.
3029 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3030 btrfs_node_key_ptr_offset(push_items),
3031 (src_nritems - push_items) *
3032 sizeof(struct btrfs_key_ptr));
3034 btrfs_set_header_nritems(src, src_nritems - push_items);
3035 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3036 btrfs_mark_buffer_dirty(src);
3037 btrfs_mark_buffer_dirty(dst);
3043 * try to push data from one node into the next node right in the
3046 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3047 * error, and > 0 if there was no room in the right hand block.
3049 * this will only push up to 1/2 the contents of the left node over
3051 static int balance_node_right(struct btrfs_trans_handle *trans,
3052 struct btrfs_root *root,
3053 struct extent_buffer *dst,
3054 struct extent_buffer *src)
3062 WARN_ON(btrfs_header_generation(src) != trans->transid);
3063 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3065 src_nritems = btrfs_header_nritems(src);
3066 dst_nritems = btrfs_header_nritems(dst);
3067 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3068 if (push_items <= 0)
3071 if (src_nritems < 4)
3074 max_push = src_nritems / 2 + 1;
3075 /* don't try to empty the node */
3076 if (max_push >= src_nritems)
3079 if (max_push < push_items)
3080 push_items = max_push;
3082 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3083 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3084 btrfs_node_key_ptr_offset(0),
3086 sizeof(struct btrfs_key_ptr));
3088 tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3089 src_nritems - push_items, push_items);
3090 copy_extent_buffer(dst, src,
3091 btrfs_node_key_ptr_offset(0),
3092 btrfs_node_key_ptr_offset(src_nritems - push_items),
3093 push_items * sizeof(struct btrfs_key_ptr));
3095 btrfs_set_header_nritems(src, src_nritems - push_items);
3096 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3098 btrfs_mark_buffer_dirty(src);
3099 btrfs_mark_buffer_dirty(dst);
3105 * helper function to insert a new root level in the tree.
3106 * A new node is allocated, and a single item is inserted to
3107 * point to the existing root
3109 * returns zero on success or < 0 on failure.
3111 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3112 struct btrfs_root *root,
3113 struct btrfs_path *path, int level)
3116 struct extent_buffer *lower;
3117 struct extent_buffer *c;
3118 struct extent_buffer *old;
3119 struct btrfs_disk_key lower_key;
3121 BUG_ON(path->nodes[level]);
3122 BUG_ON(path->nodes[level-1] != root->node);
3124 lower = path->nodes[level-1];
3126 btrfs_item_key(lower, &lower_key, 0);
3128 btrfs_node_key(lower, &lower_key, 0);
3130 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3131 root->root_key.objectid, &lower_key,
3132 level, root->node->start, 0);
3136 root_add_used(root, root->nodesize);
3138 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3139 btrfs_set_header_nritems(c, 1);
3140 btrfs_set_header_level(c, level);
3141 btrfs_set_header_bytenr(c, c->start);
3142 btrfs_set_header_generation(c, trans->transid);
3143 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3144 btrfs_set_header_owner(c, root->root_key.objectid);
3146 write_extent_buffer(c, root->fs_info->fsid,
3147 (unsigned long)btrfs_header_fsid(c),
3150 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3151 (unsigned long)btrfs_header_chunk_tree_uuid(c),
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,
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 (unsigned long)btrfs_header_fsid(split),
3289 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3290 (unsigned long)btrfs_header_chunk_tree_uuid(split),
3293 tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid);
3294 copy_extent_buffer(split, c,
3295 btrfs_node_key_ptr_offset(0),
3296 btrfs_node_key_ptr_offset(mid),
3297 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3298 btrfs_set_header_nritems(split, c_nritems - mid);
3299 btrfs_set_header_nritems(c, mid);
3302 btrfs_mark_buffer_dirty(c);
3303 btrfs_mark_buffer_dirty(split);
3305 insert_ptr(trans, root, path, &disk_key, split->start,
3306 path->slots[level + 1] + 1, level + 1);
3308 if (path->slots[level] >= mid) {
3309 path->slots[level] -= mid;
3310 btrfs_tree_unlock(c);
3311 free_extent_buffer(c);
3312 path->nodes[level] = split;
3313 path->slots[level + 1] += 1;
3315 btrfs_tree_unlock(split);
3316 free_extent_buffer(split);
3322 * how many bytes are required to store the items in a leaf. start
3323 * and nr indicate which items in the leaf to check. This totals up the
3324 * space used both by the item structs and the item data
3326 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3328 struct btrfs_item *start_item;
3329 struct btrfs_item *end_item;
3330 struct btrfs_map_token token;
3332 int nritems = btrfs_header_nritems(l);
3333 int end = min(nritems, start + nr) - 1;
3337 btrfs_init_map_token(&token);
3338 start_item = btrfs_item_nr(l, start);
3339 end_item = btrfs_item_nr(l, end);
3340 data_len = btrfs_token_item_offset(l, start_item, &token) +
3341 btrfs_token_item_size(l, start_item, &token);
3342 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3343 data_len += sizeof(struct btrfs_item) * nr;
3344 WARN_ON(data_len < 0);
3349 * The space between the end of the leaf items and
3350 * the start of the leaf data. IOW, how much room
3351 * the leaf has left for both items and data
3353 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3354 struct extent_buffer *leaf)
3356 int nritems = btrfs_header_nritems(leaf);
3358 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3360 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
3361 "used %d nritems %d\n",
3362 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3363 leaf_space_used(leaf, 0, nritems), nritems);
3369 * min slot controls the lowest index we're willing to push to the
3370 * right. We'll push up to and including min_slot, but no lower
3372 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3373 struct btrfs_root *root,
3374 struct btrfs_path *path,
3375 int data_size, int empty,
3376 struct extent_buffer *right,
3377 int free_space, u32 left_nritems,
3380 struct extent_buffer *left = path->nodes[0];
3381 struct extent_buffer *upper = path->nodes[1];
3382 struct btrfs_map_token token;
3383 struct btrfs_disk_key disk_key;
3388 struct btrfs_item *item;
3394 btrfs_init_map_token(&token);
3399 nr = max_t(u32, 1, min_slot);
3401 if (path->slots[0] >= left_nritems)
3402 push_space += data_size;
3404 slot = path->slots[1];
3405 i = left_nritems - 1;
3407 item = btrfs_item_nr(left, i);
3409 if (!empty && push_items > 0) {
3410 if (path->slots[0] > i)
3412 if (path->slots[0] == i) {
3413 int space = btrfs_leaf_free_space(root, left);
3414 if (space + push_space * 2 > free_space)
3419 if (path->slots[0] == i)
3420 push_space += data_size;
3422 this_item_size = btrfs_item_size(left, item);
3423 if (this_item_size + sizeof(*item) + push_space > free_space)
3427 push_space += this_item_size + sizeof(*item);
3433 if (push_items == 0)
3436 WARN_ON(!empty && push_items == left_nritems);
3438 /* push left to right */
3439 right_nritems = btrfs_header_nritems(right);
3441 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3442 push_space -= leaf_data_end(root, left);
3444 /* make room in the right data area */
3445 data_end = leaf_data_end(root, right);
3446 memmove_extent_buffer(right,
3447 btrfs_leaf_data(right) + data_end - push_space,
3448 btrfs_leaf_data(right) + data_end,
3449 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3451 /* copy from the left data area */
3452 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3453 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3454 btrfs_leaf_data(left) + leaf_data_end(root, left),
3457 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3458 btrfs_item_nr_offset(0),
3459 right_nritems * sizeof(struct btrfs_item));
3461 /* copy the items from left to right */
3462 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3463 btrfs_item_nr_offset(left_nritems - push_items),
3464 push_items * sizeof(struct btrfs_item));
3466 /* update the item pointers */
3467 right_nritems += push_items;
3468 btrfs_set_header_nritems(right, right_nritems);
3469 push_space = BTRFS_LEAF_DATA_SIZE(root);
3470 for (i = 0; i < right_nritems; i++) {
3471 item = btrfs_item_nr(right, i);
3472 push_space -= btrfs_token_item_size(right, item, &token);
3473 btrfs_set_token_item_offset(right, item, push_space, &token);
3476 left_nritems -= push_items;
3477 btrfs_set_header_nritems(left, left_nritems);
3480 btrfs_mark_buffer_dirty(left);
3482 clean_tree_block(trans, root, left);
3484 btrfs_mark_buffer_dirty(right);
3486 btrfs_item_key(right, &disk_key, 0);
3487 btrfs_set_node_key(upper, &disk_key, slot + 1);
3488 btrfs_mark_buffer_dirty(upper);
3490 /* then fixup the leaf pointer in the path */
3491 if (path->slots[0] >= left_nritems) {
3492 path->slots[0] -= left_nritems;
3493 if (btrfs_header_nritems(path->nodes[0]) == 0)
3494 clean_tree_block(trans, root, path->nodes[0]);
3495 btrfs_tree_unlock(path->nodes[0]);
3496 free_extent_buffer(path->nodes[0]);
3497 path->nodes[0] = right;
3498 path->slots[1] += 1;
3500 btrfs_tree_unlock(right);
3501 free_extent_buffer(right);
3506 btrfs_tree_unlock(right);
3507 free_extent_buffer(right);
3512 * push some data in the path leaf to the right, trying to free up at
3513 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3515 * returns 1 if the push failed because the other node didn't have enough
3516 * room, 0 if everything worked out and < 0 if there were major errors.
3518 * this will push starting from min_slot to the end of the leaf. It won't
3519 * push any slot lower than min_slot
3521 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3522 *root, struct btrfs_path *path,
3523 int min_data_size, int data_size,
3524 int empty, u32 min_slot)
3526 struct extent_buffer *left = path->nodes[0];
3527 struct extent_buffer *right;
3528 struct extent_buffer *upper;
3534 if (!path->nodes[1])
3537 slot = path->slots[1];
3538 upper = path->nodes[1];
3539 if (slot >= btrfs_header_nritems(upper) - 1)
3542 btrfs_assert_tree_locked(path->nodes[1]);
3544 right = read_node_slot(root, upper, slot + 1);
3548 btrfs_tree_lock(right);
3549 btrfs_set_lock_blocking(right);
3551 free_space = btrfs_leaf_free_space(root, right);
3552 if (free_space < data_size)
3555 /* cow and double check */
3556 ret = btrfs_cow_block(trans, root, right, upper,
3561 free_space = btrfs_leaf_free_space(root, right);
3562 if (free_space < data_size)
3565 left_nritems = btrfs_header_nritems(left);
3566 if (left_nritems == 0)
3569 return __push_leaf_right(trans, root, path, min_data_size, empty,
3570 right, free_space, left_nritems, min_slot);
3572 btrfs_tree_unlock(right);
3573 free_extent_buffer(right);
3578 * push some data in the path leaf to the left, trying to free up at
3579 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3581 * max_slot can put a limit on how far into the leaf we'll push items. The
3582 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3585 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3586 struct btrfs_root *root,
3587 struct btrfs_path *path, int data_size,
3588 int empty, struct extent_buffer *left,
3589 int free_space, u32 right_nritems,
3592 struct btrfs_disk_key disk_key;
3593 struct extent_buffer *right = path->nodes[0];
3597 struct btrfs_item *item;
3598 u32 old_left_nritems;
3602 u32 old_left_item_size;
3603 struct btrfs_map_token token;
3605 btrfs_init_map_token(&token);
3608 nr = min(right_nritems, max_slot);
3610 nr = min(right_nritems - 1, max_slot);
3612 for (i = 0; i < nr; i++) {
3613 item = btrfs_item_nr(right, i);
3615 if (!empty && push_items > 0) {
3616 if (path->slots[0] < i)
3618 if (path->slots[0] == i) {
3619 int space = btrfs_leaf_free_space(root, right);
3620 if (space + push_space * 2 > free_space)
3625 if (path->slots[0] == i)
3626 push_space += data_size;
3628 this_item_size = btrfs_item_size(right, item);
3629 if (this_item_size + sizeof(*item) + push_space > free_space)
3633 push_space += this_item_size + sizeof(*item);
3636 if (push_items == 0) {
3640 if (!empty && push_items == btrfs_header_nritems(right))
3643 /* push data from right to left */
3644 copy_extent_buffer(left, right,
3645 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3646 btrfs_item_nr_offset(0),
3647 push_items * sizeof(struct btrfs_item));
3649 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3650 btrfs_item_offset_nr(right, push_items - 1);
3652 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3653 leaf_data_end(root, left) - push_space,
3654 btrfs_leaf_data(right) +
3655 btrfs_item_offset_nr(right, push_items - 1),
3657 old_left_nritems = btrfs_header_nritems(left);
3658 BUG_ON(old_left_nritems <= 0);
3660 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3661 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3664 item = btrfs_item_nr(left, i);
3666 ioff = btrfs_token_item_offset(left, item, &token);
3667 btrfs_set_token_item_offset(left, item,
3668 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3671 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3673 /* fixup right node */
3674 if (push_items > right_nritems)
3675 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3678 if (push_items < right_nritems) {
3679 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3680 leaf_data_end(root, right);
3681 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3682 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3683 btrfs_leaf_data(right) +
3684 leaf_data_end(root, right), push_space);
3686 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3687 btrfs_item_nr_offset(push_items),
3688 (btrfs_header_nritems(right) - push_items) *
3689 sizeof(struct btrfs_item));
3691 right_nritems -= push_items;
3692 btrfs_set_header_nritems(right, right_nritems);
3693 push_space = BTRFS_LEAF_DATA_SIZE(root);
3694 for (i = 0; i < right_nritems; i++) {
3695 item = btrfs_item_nr(right, i);
3697 push_space = push_space - btrfs_token_item_size(right,
3699 btrfs_set_token_item_offset(right, item, push_space, &token);
3702 btrfs_mark_buffer_dirty(left);
3704 btrfs_mark_buffer_dirty(right);
3706 clean_tree_block(trans, root, right);
3708 btrfs_item_key(right, &disk_key, 0);
3709 fixup_low_keys(root, path, &disk_key, 1);
3711 /* then fixup the leaf pointer in the path */
3712 if (path->slots[0] < push_items) {
3713 path->slots[0] += old_left_nritems;
3714 btrfs_tree_unlock(path->nodes[0]);
3715 free_extent_buffer(path->nodes[0]);
3716 path->nodes[0] = left;
3717 path->slots[1] -= 1;
3719 btrfs_tree_unlock(left);
3720 free_extent_buffer(left);
3721 path->slots[0] -= push_items;
3723 BUG_ON(path->slots[0] < 0);
3726 btrfs_tree_unlock(left);
3727 free_extent_buffer(left);
3732 * push some data in the path leaf to the left, trying to free up at
3733 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3735 * max_slot can put a limit on how far into the leaf we'll push items. The
3736 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3739 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3740 *root, struct btrfs_path *path, int min_data_size,
3741 int data_size, int empty, u32 max_slot)
3743 struct extent_buffer *right = path->nodes[0];
3744 struct extent_buffer *left;
3750 slot = path->slots[1];
3753 if (!path->nodes[1])
3756 right_nritems = btrfs_header_nritems(right);
3757 if (right_nritems == 0)
3760 btrfs_assert_tree_locked(path->nodes[1]);
3762 left = read_node_slot(root, path->nodes[1], slot - 1);
3766 btrfs_tree_lock(left);
3767 btrfs_set_lock_blocking(left);
3769 free_space = btrfs_leaf_free_space(root, left);
3770 if (free_space < data_size) {
3775 /* cow and double check */
3776 ret = btrfs_cow_block(trans, root, left,
3777 path->nodes[1], slot - 1, &left);
3779 /* we hit -ENOSPC, but it isn't fatal here */
3785 free_space = btrfs_leaf_free_space(root, left);
3786 if (free_space < data_size) {
3791 return __push_leaf_left(trans, root, path, min_data_size,
3792 empty, left, free_space, right_nritems,
3795 btrfs_tree_unlock(left);
3796 free_extent_buffer(left);
3801 * split the path's leaf in two, making sure there is at least data_size
3802 * available for the resulting leaf level of the path.
3804 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3805 struct btrfs_root *root,
3806 struct btrfs_path *path,
3807 struct extent_buffer *l,
3808 struct extent_buffer *right,
3809 int slot, int mid, int nritems)
3814 struct btrfs_disk_key disk_key;
3815 struct btrfs_map_token token;
3817 btrfs_init_map_token(&token);
3819 nritems = nritems - mid;
3820 btrfs_set_header_nritems(right, nritems);
3821 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
3823 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3824 btrfs_item_nr_offset(mid),
3825 nritems * sizeof(struct btrfs_item));
3827 copy_extent_buffer(right, l,
3828 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
3829 data_copy_size, btrfs_leaf_data(l) +
3830 leaf_data_end(root, l), data_copy_size);
3832 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
3833 btrfs_item_end_nr(l, mid);
3835 for (i = 0; i < nritems; i++) {
3836 struct btrfs_item *item = btrfs_item_nr(right, i);
3839 ioff = btrfs_token_item_offset(right, item, &token);
3840 btrfs_set_token_item_offset(right, item,
3841 ioff + rt_data_off, &token);
3844 btrfs_set_header_nritems(l, mid);
3845 btrfs_item_key(right, &disk_key, 0);
3846 insert_ptr(trans, root, path, &disk_key, right->start,
3847 path->slots[1] + 1, 1);
3849 btrfs_mark_buffer_dirty(right);
3850 btrfs_mark_buffer_dirty(l);
3851 BUG_ON(path->slots[0] != slot);
3854 btrfs_tree_unlock(path->nodes[0]);
3855 free_extent_buffer(path->nodes[0]);
3856 path->nodes[0] = right;
3857 path->slots[0] -= mid;
3858 path->slots[1] += 1;
3860 btrfs_tree_unlock(right);
3861 free_extent_buffer(right);
3864 BUG_ON(path->slots[0] < 0);
3868 * double splits happen when we need to insert a big item in the middle
3869 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3870 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3873 * We avoid this by trying to push the items on either side of our target
3874 * into the adjacent leaves. If all goes well we can avoid the double split
3877 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3878 struct btrfs_root *root,
3879 struct btrfs_path *path,
3887 slot = path->slots[0];
3890 * try to push all the items after our slot into the
3893 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
3900 nritems = btrfs_header_nritems(path->nodes[0]);
3902 * our goal is to get our slot at the start or end of a leaf. If
3903 * we've done so we're done
3905 if (path->slots[0] == 0 || path->slots[0] == nritems)
3908 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3911 /* try to push all the items before our slot into the next leaf */
3912 slot = path->slots[0];
3913 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
3926 * split the path's leaf in two, making sure there is at least data_size
3927 * available for the resulting leaf level of the path.
3929 * returns 0 if all went well and < 0 on failure.
3931 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3932 struct btrfs_root *root,
3933 struct btrfs_key *ins_key,
3934 struct btrfs_path *path, int data_size,
3937 struct btrfs_disk_key disk_key;
3938 struct extent_buffer *l;
3942 struct extent_buffer *right;
3946 int num_doubles = 0;
3947 int tried_avoid_double = 0;
3950 slot = path->slots[0];
3951 if (extend && data_size + btrfs_item_size_nr(l, slot) +
3952 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
3955 /* first try to make some room by pushing left and right */
3956 if (data_size && path->nodes[1]) {
3957 wret = push_leaf_right(trans, root, path, data_size,
3962 wret = push_leaf_left(trans, root, path, data_size,
3963 data_size, 0, (u32)-1);
3969 /* did the pushes work? */
3970 if (btrfs_leaf_free_space(root, l) >= data_size)
3974 if (!path->nodes[1]) {
3975 ret = insert_new_root(trans, root, path, 1);
3982 slot = path->slots[0];
3983 nritems = btrfs_header_nritems(l);
3984 mid = (nritems + 1) / 2;
3988 leaf_space_used(l, mid, nritems - mid) + data_size >
3989 BTRFS_LEAF_DATA_SIZE(root)) {
3990 if (slot >= nritems) {
3994 if (mid != nritems &&
3995 leaf_space_used(l, mid, nritems - mid) +
3996 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
3997 if (data_size && !tried_avoid_double)
3998 goto push_for_double;
4004 if (leaf_space_used(l, 0, mid) + data_size >
4005 BTRFS_LEAF_DATA_SIZE(root)) {
4006 if (!extend && data_size && slot == 0) {
4008 } else if ((extend || !data_size) && slot == 0) {
4012 if (mid != nritems &&
4013 leaf_space_used(l, mid, nritems - mid) +
4014 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4015 if (data_size && !tried_avoid_double)
4016 goto push_for_double;
4024 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4026 btrfs_item_key(l, &disk_key, mid);
4028 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
4029 root->root_key.objectid,
4030 &disk_key, 0, l->start, 0);
4032 return PTR_ERR(right);
4034 root_add_used(root, root->leafsize);
4036 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4037 btrfs_set_header_bytenr(right, right->start);
4038 btrfs_set_header_generation(right, trans->transid);
4039 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4040 btrfs_set_header_owner(right, root->root_key.objectid);
4041 btrfs_set_header_level(right, 0);
4042 write_extent_buffer(right, root->fs_info->fsid,
4043 (unsigned long)btrfs_header_fsid(right),
4046 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4047 (unsigned long)btrfs_header_chunk_tree_uuid(right),
4052 btrfs_set_header_nritems(right, 0);
4053 insert_ptr(trans, root, path, &disk_key, right->start,
4054 path->slots[1] + 1, 1);
4055 btrfs_tree_unlock(path->nodes[0]);
4056 free_extent_buffer(path->nodes[0]);
4057 path->nodes[0] = right;
4059 path->slots[1] += 1;
4061 btrfs_set_header_nritems(right, 0);
4062 insert_ptr(trans, root, path, &disk_key, right->start,
4064 btrfs_tree_unlock(path->nodes[0]);
4065 free_extent_buffer(path->nodes[0]);
4066 path->nodes[0] = right;
4068 if (path->slots[1] == 0)
4069 fixup_low_keys(root, path, &disk_key, 1);
4071 btrfs_mark_buffer_dirty(right);
4075 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4078 BUG_ON(num_doubles != 0);
4086 push_for_double_split(trans, root, path, data_size);
4087 tried_avoid_double = 1;
4088 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4093 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4094 struct btrfs_root *root,
4095 struct btrfs_path *path, int ins_len)
4097 struct btrfs_key key;
4098 struct extent_buffer *leaf;
4099 struct btrfs_file_extent_item *fi;
4104 leaf = path->nodes[0];
4105 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4107 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4108 key.type != BTRFS_EXTENT_CSUM_KEY);
4110 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4113 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4114 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4115 fi = btrfs_item_ptr(leaf, path->slots[0],
4116 struct btrfs_file_extent_item);
4117 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4119 btrfs_release_path(path);
4121 path->keep_locks = 1;
4122 path->search_for_split = 1;
4123 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4124 path->search_for_split = 0;
4129 leaf = path->nodes[0];
4130 /* if our item isn't there or got smaller, return now */
4131 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4134 /* the leaf has changed, it now has room. return now */
4135 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4138 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4139 fi = btrfs_item_ptr(leaf, path->slots[0],
4140 struct btrfs_file_extent_item);
4141 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4145 btrfs_set_path_blocking(path);
4146 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4150 path->keep_locks = 0;
4151 btrfs_unlock_up_safe(path, 1);
4154 path->keep_locks = 0;
4158 static noinline int split_item(struct btrfs_trans_handle *trans,
4159 struct btrfs_root *root,
4160 struct btrfs_path *path,
4161 struct btrfs_key *new_key,
4162 unsigned long split_offset)
4164 struct extent_buffer *leaf;
4165 struct btrfs_item *item;
4166 struct btrfs_item *new_item;
4172 struct btrfs_disk_key disk_key;
4174 leaf = path->nodes[0];
4175 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4177 btrfs_set_path_blocking(path);
4179 item = btrfs_item_nr(leaf, path->slots[0]);
4180 orig_offset = btrfs_item_offset(leaf, item);
4181 item_size = btrfs_item_size(leaf, item);
4183 buf = kmalloc(item_size, GFP_NOFS);
4187 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4188 path->slots[0]), item_size);
4190 slot = path->slots[0] + 1;
4191 nritems = btrfs_header_nritems(leaf);
4192 if (slot != nritems) {
4193 /* shift the items */
4194 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4195 btrfs_item_nr_offset(slot),
4196 (nritems - slot) * sizeof(struct btrfs_item));
4199 btrfs_cpu_key_to_disk(&disk_key, new_key);
4200 btrfs_set_item_key(leaf, &disk_key, slot);
4202 new_item = btrfs_item_nr(leaf, slot);
4204 btrfs_set_item_offset(leaf, new_item, orig_offset);
4205 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4207 btrfs_set_item_offset(leaf, item,
4208 orig_offset + item_size - split_offset);
4209 btrfs_set_item_size(leaf, item, split_offset);
4211 btrfs_set_header_nritems(leaf, nritems + 1);
4213 /* write the data for the start of the original item */
4214 write_extent_buffer(leaf, buf,
4215 btrfs_item_ptr_offset(leaf, path->slots[0]),
4218 /* write the data for the new item */
4219 write_extent_buffer(leaf, buf + split_offset,
4220 btrfs_item_ptr_offset(leaf, slot),
4221 item_size - split_offset);
4222 btrfs_mark_buffer_dirty(leaf);
4224 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4230 * This function splits a single item into two items,
4231 * giving 'new_key' to the new item and splitting the
4232 * old one at split_offset (from the start of the item).
4234 * The path may be released by this operation. After
4235 * the split, the path is pointing to the old item. The
4236 * new item is going to be in the same node as the old one.
4238 * Note, the item being split must be smaller enough to live alone on
4239 * a tree block with room for one extra struct btrfs_item
4241 * This allows us to split the item in place, keeping a lock on the
4242 * leaf the entire time.
4244 int btrfs_split_item(struct btrfs_trans_handle *trans,
4245 struct btrfs_root *root,
4246 struct btrfs_path *path,
4247 struct btrfs_key *new_key,
4248 unsigned long split_offset)
4251 ret = setup_leaf_for_split(trans, root, path,
4252 sizeof(struct btrfs_item));
4256 ret = split_item(trans, root, path, new_key, split_offset);
4261 * This function duplicate a item, giving 'new_key' to the new item.
4262 * It guarantees both items live in the same tree leaf and the new item
4263 * is contiguous with the original item.
4265 * This allows us to split file extent in place, keeping a lock on the
4266 * leaf the entire time.
4268 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4269 struct btrfs_root *root,
4270 struct btrfs_path *path,
4271 struct btrfs_key *new_key)
4273 struct extent_buffer *leaf;
4277 leaf = path->nodes[0];
4278 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4279 ret = setup_leaf_for_split(trans, root, path,
4280 item_size + sizeof(struct btrfs_item));
4285 setup_items_for_insert(root, path, new_key, &item_size,
4286 item_size, item_size +
4287 sizeof(struct btrfs_item), 1);
4288 leaf = path->nodes[0];
4289 memcpy_extent_buffer(leaf,
4290 btrfs_item_ptr_offset(leaf, path->slots[0]),
4291 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4297 * make the item pointed to by the path smaller. new_size indicates
4298 * how small to make it, and from_end tells us if we just chop bytes
4299 * off the end of the item or if we shift the item to chop bytes off
4302 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4303 u32 new_size, int from_end)
4306 struct extent_buffer *leaf;
4307 struct btrfs_item *item;
4309 unsigned int data_end;
4310 unsigned int old_data_start;
4311 unsigned int old_size;
4312 unsigned int size_diff;
4314 struct btrfs_map_token token;
4316 btrfs_init_map_token(&token);
4318 leaf = path->nodes[0];
4319 slot = path->slots[0];
4321 old_size = btrfs_item_size_nr(leaf, slot);
4322 if (old_size == new_size)
4325 nritems = btrfs_header_nritems(leaf);
4326 data_end = leaf_data_end(root, leaf);
4328 old_data_start = btrfs_item_offset_nr(leaf, slot);
4330 size_diff = old_size - new_size;
4333 BUG_ON(slot >= nritems);
4336 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4338 /* first correct the data pointers */
4339 for (i = slot; i < nritems; i++) {
4341 item = btrfs_item_nr(leaf, i);
4343 ioff = btrfs_token_item_offset(leaf, item, &token);
4344 btrfs_set_token_item_offset(leaf, item,
4345 ioff + size_diff, &token);
4348 /* shift the data */
4350 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4351 data_end + size_diff, btrfs_leaf_data(leaf) +
4352 data_end, old_data_start + new_size - data_end);
4354 struct btrfs_disk_key disk_key;
4357 btrfs_item_key(leaf, &disk_key, slot);
4359 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4361 struct btrfs_file_extent_item *fi;
4363 fi = btrfs_item_ptr(leaf, slot,
4364 struct btrfs_file_extent_item);
4365 fi = (struct btrfs_file_extent_item *)(
4366 (unsigned long)fi - size_diff);
4368 if (btrfs_file_extent_type(leaf, fi) ==
4369 BTRFS_FILE_EXTENT_INLINE) {
4370 ptr = btrfs_item_ptr_offset(leaf, slot);
4371 memmove_extent_buffer(leaf, ptr,
4373 offsetof(struct btrfs_file_extent_item,
4378 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4379 data_end + size_diff, btrfs_leaf_data(leaf) +
4380 data_end, old_data_start - data_end);
4382 offset = btrfs_disk_key_offset(&disk_key);
4383 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4384 btrfs_set_item_key(leaf, &disk_key, slot);
4386 fixup_low_keys(root, path, &disk_key, 1);
4389 item = btrfs_item_nr(leaf, slot);
4390 btrfs_set_item_size(leaf, item, new_size);
4391 btrfs_mark_buffer_dirty(leaf);
4393 if (btrfs_leaf_free_space(root, leaf) < 0) {
4394 btrfs_print_leaf(root, leaf);
4400 * make the item pointed to by the path bigger, data_size is the added size.
4402 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4406 struct extent_buffer *leaf;
4407 struct btrfs_item *item;
4409 unsigned int data_end;
4410 unsigned int old_data;
4411 unsigned int old_size;
4413 struct btrfs_map_token token;
4415 btrfs_init_map_token(&token);
4417 leaf = path->nodes[0];
4419 nritems = btrfs_header_nritems(leaf);
4420 data_end = leaf_data_end(root, leaf);
4422 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4423 btrfs_print_leaf(root, leaf);
4426 slot = path->slots[0];
4427 old_data = btrfs_item_end_nr(leaf, slot);
4430 if (slot >= nritems) {
4431 btrfs_print_leaf(root, leaf);
4432 printk(KERN_CRIT "slot %d too large, nritems %d\n",
4438 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4440 /* first correct the data pointers */
4441 for (i = slot; i < nritems; i++) {
4443 item = btrfs_item_nr(leaf, i);
4445 ioff = btrfs_token_item_offset(leaf, item, &token);
4446 btrfs_set_token_item_offset(leaf, item,
4447 ioff - data_size, &token);
4450 /* shift the data */
4451 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4452 data_end - data_size, btrfs_leaf_data(leaf) +
4453 data_end, old_data - data_end);
4455 data_end = old_data;
4456 old_size = btrfs_item_size_nr(leaf, slot);
4457 item = btrfs_item_nr(leaf, slot);
4458 btrfs_set_item_size(leaf, item, old_size + data_size);
4459 btrfs_mark_buffer_dirty(leaf);
4461 if (btrfs_leaf_free_space(root, leaf) < 0) {
4462 btrfs_print_leaf(root, leaf);
4468 * this is a helper for btrfs_insert_empty_items, the main goal here is
4469 * to save stack depth by doing the bulk of the work in a function
4470 * that doesn't call btrfs_search_slot
4472 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4473 struct btrfs_key *cpu_key, u32 *data_size,
4474 u32 total_data, u32 total_size, int nr)
4476 struct btrfs_item *item;
4479 unsigned int data_end;
4480 struct btrfs_disk_key disk_key;
4481 struct extent_buffer *leaf;
4483 struct btrfs_map_token token;
4485 btrfs_init_map_token(&token);
4487 leaf = path->nodes[0];
4488 slot = path->slots[0];
4490 nritems = btrfs_header_nritems(leaf);
4491 data_end = leaf_data_end(root, leaf);
4493 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4494 btrfs_print_leaf(root, leaf);
4495 printk(KERN_CRIT "not enough freespace need %u have %d\n",
4496 total_size, btrfs_leaf_free_space(root, leaf));
4500 if (slot != nritems) {
4501 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4503 if (old_data < data_end) {
4504 btrfs_print_leaf(root, leaf);
4505 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
4506 slot, old_data, data_end);
4510 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4512 /* first correct the data pointers */
4513 for (i = slot; i < nritems; i++) {
4516 item = btrfs_item_nr(leaf, i);
4517 ioff = btrfs_token_item_offset(leaf, item, &token);
4518 btrfs_set_token_item_offset(leaf, item,
4519 ioff - total_data, &token);
4521 /* shift the items */
4522 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4523 btrfs_item_nr_offset(slot),
4524 (nritems - slot) * sizeof(struct btrfs_item));
4526 /* shift the data */
4527 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4528 data_end - total_data, btrfs_leaf_data(leaf) +
4529 data_end, old_data - data_end);
4530 data_end = old_data;
4533 /* setup the item for the new data */
4534 for (i = 0; i < nr; i++) {
4535 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4536 btrfs_set_item_key(leaf, &disk_key, slot + i);
4537 item = btrfs_item_nr(leaf, slot + i);
4538 btrfs_set_token_item_offset(leaf, item,
4539 data_end - data_size[i], &token);
4540 data_end -= data_size[i];
4541 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4544 btrfs_set_header_nritems(leaf, nritems + nr);
4547 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4548 fixup_low_keys(root, path, &disk_key, 1);
4550 btrfs_unlock_up_safe(path, 1);
4551 btrfs_mark_buffer_dirty(leaf);
4553 if (btrfs_leaf_free_space(root, leaf) < 0) {
4554 btrfs_print_leaf(root, leaf);
4560 * Given a key and some data, insert items into the tree.
4561 * This does all the path init required, making room in the tree if needed.
4563 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4564 struct btrfs_root *root,
4565 struct btrfs_path *path,
4566 struct btrfs_key *cpu_key, u32 *data_size,
4575 for (i = 0; i < nr; i++)
4576 total_data += data_size[i];
4578 total_size = total_data + (nr * sizeof(struct btrfs_item));
4579 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4585 slot = path->slots[0];
4588 setup_items_for_insert(root, path, cpu_key, data_size,
4589 total_data, total_size, nr);
4594 * Given a key and some data, insert an item into the tree.
4595 * This does all the path init required, making room in the tree if needed.
4597 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4598 *root, struct btrfs_key *cpu_key, void *data, u32
4602 struct btrfs_path *path;
4603 struct extent_buffer *leaf;
4606 path = btrfs_alloc_path();
4609 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4611 leaf = path->nodes[0];
4612 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4613 write_extent_buffer(leaf, data, ptr, data_size);
4614 btrfs_mark_buffer_dirty(leaf);
4616 btrfs_free_path(path);
4621 * delete the pointer from a given node.
4623 * the tree should have been previously balanced so the deletion does not
4626 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4627 int level, int slot)
4629 struct extent_buffer *parent = path->nodes[level];
4633 nritems = btrfs_header_nritems(parent);
4634 if (slot != nritems - 1) {
4636 tree_mod_log_eb_move(root->fs_info, parent, slot,
4637 slot + 1, nritems - slot - 1);
4638 memmove_extent_buffer(parent,
4639 btrfs_node_key_ptr_offset(slot),
4640 btrfs_node_key_ptr_offset(slot + 1),
4641 sizeof(struct btrfs_key_ptr) *
4642 (nritems - slot - 1));
4644 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4645 MOD_LOG_KEY_REMOVE);
4650 btrfs_set_header_nritems(parent, nritems);
4651 if (nritems == 0 && parent == root->node) {
4652 BUG_ON(btrfs_header_level(root->node) != 1);
4653 /* just turn the root into a leaf and break */
4654 btrfs_set_header_level(root->node, 0);
4655 } else if (slot == 0) {
4656 struct btrfs_disk_key disk_key;
4658 btrfs_node_key(parent, &disk_key, 0);
4659 fixup_low_keys(root, path, &disk_key, level + 1);
4661 btrfs_mark_buffer_dirty(parent);
4665 * a helper function to delete the leaf pointed to by path->slots[1] and
4668 * This deletes the pointer in path->nodes[1] and frees the leaf
4669 * block extent. zero is returned if it all worked out, < 0 otherwise.
4671 * The path must have already been setup for deleting the leaf, including
4672 * all the proper balancing. path->nodes[1] must be locked.
4674 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4675 struct btrfs_root *root,
4676 struct btrfs_path *path,
4677 struct extent_buffer *leaf)
4679 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4680 del_ptr(root, path, 1, path->slots[1]);
4683 * btrfs_free_extent is expensive, we want to make sure we
4684 * aren't holding any locks when we call it
4686 btrfs_unlock_up_safe(path, 0);
4688 root_sub_used(root, leaf->len);
4690 extent_buffer_get(leaf);
4691 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4692 free_extent_buffer_stale(leaf);
4695 * delete the item at the leaf level in path. If that empties
4696 * the leaf, remove it from the tree
4698 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4699 struct btrfs_path *path, int slot, int nr)
4701 struct extent_buffer *leaf;
4702 struct btrfs_item *item;
4709 struct btrfs_map_token token;
4711 btrfs_init_map_token(&token);
4713 leaf = path->nodes[0];
4714 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4716 for (i = 0; i < nr; i++)
4717 dsize += btrfs_item_size_nr(leaf, slot + i);
4719 nritems = btrfs_header_nritems(leaf);
4721 if (slot + nr != nritems) {
4722 int data_end = leaf_data_end(root, leaf);
4724 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4726 btrfs_leaf_data(leaf) + data_end,
4727 last_off - data_end);
4729 for (i = slot + nr; i < nritems; i++) {
4732 item = btrfs_item_nr(leaf, i);
4733 ioff = btrfs_token_item_offset(leaf, item, &token);
4734 btrfs_set_token_item_offset(leaf, item,
4735 ioff + dsize, &token);
4738 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4739 btrfs_item_nr_offset(slot + nr),
4740 sizeof(struct btrfs_item) *
4741 (nritems - slot - nr));
4743 btrfs_set_header_nritems(leaf, nritems - nr);
4746 /* delete the leaf if we've emptied it */
4748 if (leaf == root->node) {
4749 btrfs_set_header_level(leaf, 0);
4751 btrfs_set_path_blocking(path);
4752 clean_tree_block(trans, root, leaf);
4753 btrfs_del_leaf(trans, root, path, leaf);
4756 int used = leaf_space_used(leaf, 0, nritems);
4758 struct btrfs_disk_key disk_key;
4760 btrfs_item_key(leaf, &disk_key, 0);
4761 fixup_low_keys(root, path, &disk_key, 1);
4764 /* delete the leaf if it is mostly empty */
4765 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
4766 /* push_leaf_left fixes the path.
4767 * make sure the path still points to our leaf
4768 * for possible call to del_ptr below
4770 slot = path->slots[1];
4771 extent_buffer_get(leaf);
4773 btrfs_set_path_blocking(path);
4774 wret = push_leaf_left(trans, root, path, 1, 1,
4776 if (wret < 0 && wret != -ENOSPC)
4779 if (path->nodes[0] == leaf &&
4780 btrfs_header_nritems(leaf)) {
4781 wret = push_leaf_right(trans, root, path, 1,
4783 if (wret < 0 && wret != -ENOSPC)
4787 if (btrfs_header_nritems(leaf) == 0) {
4788 path->slots[1] = slot;
4789 btrfs_del_leaf(trans, root, path, leaf);
4790 free_extent_buffer(leaf);
4793 /* if we're still in the path, make sure
4794 * we're dirty. Otherwise, one of the
4795 * push_leaf functions must have already
4796 * dirtied this buffer
4798 if (path->nodes[0] == leaf)
4799 btrfs_mark_buffer_dirty(leaf);
4800 free_extent_buffer(leaf);
4803 btrfs_mark_buffer_dirty(leaf);
4810 * search the tree again to find a leaf with lesser keys
4811 * returns 0 if it found something or 1 if there are no lesser leaves.
4812 * returns < 0 on io errors.
4814 * This may release the path, and so you may lose any locks held at the
4817 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4819 struct btrfs_key key;
4820 struct btrfs_disk_key found_key;
4823 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4827 else if (key.type > 0)
4829 else if (key.objectid > 0)
4834 btrfs_release_path(path);
4835 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4838 btrfs_item_key(path->nodes[0], &found_key, 0);
4839 ret = comp_keys(&found_key, &key);
4846 * A helper function to walk down the tree starting at min_key, and looking
4847 * for nodes or leaves that are have a minimum transaction id.
4848 * This is used by the btree defrag code, and tree logging
4850 * This does not cow, but it does stuff the starting key it finds back
4851 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4852 * key and get a writable path.
4854 * This does lock as it descends, and path->keep_locks should be set
4855 * to 1 by the caller.
4857 * This honors path->lowest_level to prevent descent past a given level
4860 * min_trans indicates the oldest transaction that you are interested
4861 * in walking through. Any nodes or leaves older than min_trans are
4862 * skipped over (without reading them).
4864 * returns zero if something useful was found, < 0 on error and 1 if there
4865 * was nothing in the tree that matched the search criteria.
4867 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4868 struct btrfs_key *max_key,
4869 struct btrfs_path *path,
4872 struct extent_buffer *cur;
4873 struct btrfs_key found_key;
4880 WARN_ON(!path->keep_locks);
4882 cur = btrfs_read_lock_root_node(root);
4883 level = btrfs_header_level(cur);
4884 WARN_ON(path->nodes[level]);
4885 path->nodes[level] = cur;
4886 path->locks[level] = BTRFS_READ_LOCK;
4888 if (btrfs_header_generation(cur) < min_trans) {
4893 nritems = btrfs_header_nritems(cur);
4894 level = btrfs_header_level(cur);
4895 sret = bin_search(cur, min_key, level, &slot);
4897 /* at the lowest level, we're done, setup the path and exit */
4898 if (level == path->lowest_level) {
4899 if (slot >= nritems)
4902 path->slots[level] = slot;
4903 btrfs_item_key_to_cpu(cur, &found_key, slot);
4906 if (sret && slot > 0)
4909 * check this node pointer against the min_trans parameters.
4910 * If it is too old, old, skip to the next one.
4912 while (slot < nritems) {
4916 blockptr = btrfs_node_blockptr(cur, slot);
4917 gen = btrfs_node_ptr_generation(cur, slot);
4918 if (gen < min_trans) {
4926 * we didn't find a candidate key in this node, walk forward
4927 * and find another one
4929 if (slot >= nritems) {
4930 path->slots[level] = slot;
4931 btrfs_set_path_blocking(path);
4932 sret = btrfs_find_next_key(root, path, min_key, level,
4935 btrfs_release_path(path);
4941 /* save our key for returning back */
4942 btrfs_node_key_to_cpu(cur, &found_key, slot);
4943 path->slots[level] = slot;
4944 if (level == path->lowest_level) {
4946 unlock_up(path, level, 1, 0, NULL);
4949 btrfs_set_path_blocking(path);
4950 cur = read_node_slot(root, cur, slot);
4951 BUG_ON(!cur); /* -ENOMEM */
4953 btrfs_tree_read_lock(cur);
4955 path->locks[level - 1] = BTRFS_READ_LOCK;
4956 path->nodes[level - 1] = cur;
4957 unlock_up(path, level, 1, 0, NULL);
4958 btrfs_clear_path_blocking(path, NULL, 0);
4962 memcpy(min_key, &found_key, sizeof(found_key));
4963 btrfs_set_path_blocking(path);
4967 static void tree_move_down(struct btrfs_root *root,
4968 struct btrfs_path *path,
4969 int *level, int root_level)
4971 BUG_ON(*level == 0);
4972 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
4973 path->slots[*level]);
4974 path->slots[*level - 1] = 0;
4978 static int tree_move_next_or_upnext(struct btrfs_root *root,
4979 struct btrfs_path *path,
4980 int *level, int root_level)
4984 nritems = btrfs_header_nritems(path->nodes[*level]);
4986 path->slots[*level]++;
4988 while (path->slots[*level] >= nritems) {
4989 if (*level == root_level)
4993 path->slots[*level] = 0;
4994 free_extent_buffer(path->nodes[*level]);
4995 path->nodes[*level] = NULL;
4997 path->slots[*level]++;
4999 nritems = btrfs_header_nritems(path->nodes[*level]);
5006 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5009 static int tree_advance(struct btrfs_root *root,
5010 struct btrfs_path *path,
5011 int *level, int root_level,
5013 struct btrfs_key *key)
5017 if (*level == 0 || !allow_down) {
5018 ret = tree_move_next_or_upnext(root, path, level, root_level);
5020 tree_move_down(root, path, level, root_level);
5025 btrfs_item_key_to_cpu(path->nodes[*level], key,
5026 path->slots[*level]);
5028 btrfs_node_key_to_cpu(path->nodes[*level], key,
5029 path->slots[*level]);
5034 static int tree_compare_item(struct btrfs_root *left_root,
5035 struct btrfs_path *left_path,
5036 struct btrfs_path *right_path,
5041 unsigned long off1, off2;
5043 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5044 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5048 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5049 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5050 right_path->slots[0]);
5052 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5054 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5061 #define ADVANCE_ONLY_NEXT -1
5064 * This function compares two trees and calls the provided callback for
5065 * every changed/new/deleted item it finds.
5066 * If shared tree blocks are encountered, whole subtrees are skipped, making
5067 * the compare pretty fast on snapshotted subvolumes.
5069 * This currently works on commit roots only. As commit roots are read only,
5070 * we don't do any locking. The commit roots are protected with transactions.
5071 * Transactions are ended and rejoined when a commit is tried in between.
5073 * This function checks for modifications done to the trees while comparing.
5074 * If it detects a change, it aborts immediately.
5076 int btrfs_compare_trees(struct btrfs_root *left_root,
5077 struct btrfs_root *right_root,
5078 btrfs_changed_cb_t changed_cb, void *ctx)
5082 struct btrfs_trans_handle *trans = NULL;
5083 struct btrfs_path *left_path = NULL;
5084 struct btrfs_path *right_path = NULL;
5085 struct btrfs_key left_key;
5086 struct btrfs_key right_key;
5087 char *tmp_buf = NULL;
5088 int left_root_level;
5089 int right_root_level;
5092 int left_end_reached;
5093 int right_end_reached;
5098 u64 left_start_ctransid;
5099 u64 right_start_ctransid;
5102 left_path = btrfs_alloc_path();
5107 right_path = btrfs_alloc_path();
5113 tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5119 left_path->search_commit_root = 1;
5120 left_path->skip_locking = 1;
5121 right_path->search_commit_root = 1;
5122 right_path->skip_locking = 1;
5124 spin_lock(&left_root->root_item_lock);
5125 left_start_ctransid = btrfs_root_ctransid(&left_root->root_item);
5126 spin_unlock(&left_root->root_item_lock);
5128 spin_lock(&right_root->root_item_lock);
5129 right_start_ctransid = btrfs_root_ctransid(&right_root->root_item);
5130 spin_unlock(&right_root->root_item_lock);
5132 trans = btrfs_join_transaction(left_root);
5133 if (IS_ERR(trans)) {
5134 ret = PTR_ERR(trans);
5140 * Strategy: Go to the first items of both trees. Then do
5142 * If both trees are at level 0
5143 * Compare keys of current items
5144 * If left < right treat left item as new, advance left tree
5146 * If left > right treat right item as deleted, advance right tree
5148 * If left == right do deep compare of items, treat as changed if
5149 * needed, advance both trees and repeat
5150 * If both trees are at the same level but not at level 0
5151 * Compare keys of current nodes/leafs
5152 * If left < right advance left tree and repeat
5153 * If left > right advance right tree and repeat
5154 * If left == right compare blockptrs of the next nodes/leafs
5155 * If they match advance both trees but stay at the same level
5157 * If they don't match advance both trees while allowing to go
5159 * If tree levels are different
5160 * Advance the tree that needs it and repeat
5162 * Advancing a tree means:
5163 * If we are at level 0, try to go to the next slot. If that's not
5164 * possible, go one level up and repeat. Stop when we found a level
5165 * where we could go to the next slot. We may at this point be on a
5168 * If we are not at level 0 and not on shared tree blocks, go one
5171 * If we are not at level 0 and on shared tree blocks, go one slot to
5172 * the right if possible or go up and right.
5175 left_level = btrfs_header_level(left_root->commit_root);
5176 left_root_level = left_level;
5177 left_path->nodes[left_level] = left_root->commit_root;
5178 extent_buffer_get(left_path->nodes[left_level]);
5180 right_level = btrfs_header_level(right_root->commit_root);
5181 right_root_level = right_level;
5182 right_path->nodes[right_level] = right_root->commit_root;
5183 extent_buffer_get(right_path->nodes[right_level]);
5185 if (left_level == 0)
5186 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5187 &left_key, left_path->slots[left_level]);
5189 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5190 &left_key, left_path->slots[left_level]);
5191 if (right_level == 0)
5192 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5193 &right_key, right_path->slots[right_level]);
5195 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5196 &right_key, right_path->slots[right_level]);
5198 left_end_reached = right_end_reached = 0;
5199 advance_left = advance_right = 0;
5203 * We need to make sure the transaction does not get committed
5204 * while we do anything on commit roots. This means, we need to
5205 * join and leave transactions for every item that we process.
5207 if (trans && btrfs_should_end_transaction(trans, left_root)) {
5208 btrfs_release_path(left_path);
5209 btrfs_release_path(right_path);
5211 ret = btrfs_end_transaction(trans, left_root);
5216 /* now rejoin the transaction */
5218 trans = btrfs_join_transaction(left_root);
5219 if (IS_ERR(trans)) {
5220 ret = PTR_ERR(trans);
5225 spin_lock(&left_root->root_item_lock);
5226 ctransid = btrfs_root_ctransid(&left_root->root_item);
5227 spin_unlock(&left_root->root_item_lock);
5228 if (ctransid != left_start_ctransid)
5229 left_start_ctransid = 0;
5231 spin_lock(&right_root->root_item_lock);
5232 ctransid = btrfs_root_ctransid(&right_root->root_item);
5233 spin_unlock(&right_root->root_item_lock);
5234 if (ctransid != right_start_ctransid)
5235 right_start_ctransid = 0;
5237 if (!left_start_ctransid || !right_start_ctransid) {
5238 WARN(1, KERN_WARNING
5239 "btrfs: btrfs_compare_tree detected "
5240 "a change in one of the trees while "
5241 "iterating. This is probably a "
5248 * the commit root may have changed, so start again
5251 left_path->lowest_level = left_level;
5252 right_path->lowest_level = right_level;
5253 ret = btrfs_search_slot(NULL, left_root,
5254 &left_key, left_path, 0, 0);
5257 ret = btrfs_search_slot(NULL, right_root,
5258 &right_key, right_path, 0, 0);
5263 if (advance_left && !left_end_reached) {
5264 ret = tree_advance(left_root, left_path, &left_level,
5266 advance_left != ADVANCE_ONLY_NEXT,
5269 left_end_reached = ADVANCE;
5272 if (advance_right && !right_end_reached) {
5273 ret = tree_advance(right_root, right_path, &right_level,
5275 advance_right != ADVANCE_ONLY_NEXT,
5278 right_end_reached = ADVANCE;
5282 if (left_end_reached && right_end_reached) {
5285 } else if (left_end_reached) {
5286 if (right_level == 0) {
5287 ret = changed_cb(left_root, right_root,
5288 left_path, right_path,
5290 BTRFS_COMPARE_TREE_DELETED,
5295 advance_right = ADVANCE;
5297 } else if (right_end_reached) {
5298 if (left_level == 0) {
5299 ret = changed_cb(left_root, right_root,
5300 left_path, right_path,
5302 BTRFS_COMPARE_TREE_NEW,
5307 advance_left = ADVANCE;
5311 if (left_level == 0 && right_level == 0) {
5312 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5314 ret = changed_cb(left_root, right_root,
5315 left_path, right_path,
5317 BTRFS_COMPARE_TREE_NEW,
5321 advance_left = ADVANCE;
5322 } else if (cmp > 0) {
5323 ret = changed_cb(left_root, right_root,
5324 left_path, right_path,
5326 BTRFS_COMPARE_TREE_DELETED,
5330 advance_right = ADVANCE;
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 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5337 ret = changed_cb(left_root, right_root,
5338 left_path, right_path,
5340 BTRFS_COMPARE_TREE_CHANGED,
5345 advance_left = ADVANCE;
5346 advance_right = ADVANCE;
5348 } else if (left_level == right_level) {
5349 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5351 advance_left = ADVANCE;
5352 } else if (cmp > 0) {
5353 advance_right = ADVANCE;
5355 left_blockptr = btrfs_node_blockptr(
5356 left_path->nodes[left_level],
5357 left_path->slots[left_level]);
5358 right_blockptr = btrfs_node_blockptr(
5359 right_path->nodes[right_level],
5360 right_path->slots[right_level]);
5361 if (left_blockptr == right_blockptr) {
5363 * As we're on a shared block, don't
5364 * allow to go deeper.
5366 advance_left = ADVANCE_ONLY_NEXT;
5367 advance_right = ADVANCE_ONLY_NEXT;
5369 advance_left = ADVANCE;
5370 advance_right = ADVANCE;
5373 } else if (left_level < right_level) {
5374 advance_right = ADVANCE;
5376 advance_left = ADVANCE;
5381 btrfs_free_path(left_path);
5382 btrfs_free_path(right_path);
5387 ret = btrfs_end_transaction(trans, left_root);
5389 btrfs_end_transaction(trans, left_root);
5396 * this is similar to btrfs_next_leaf, but does not try to preserve
5397 * and fixup the path. It looks for and returns the next key in the
5398 * tree based on the current path and the min_trans parameters.
5400 * 0 is returned if another key is found, < 0 if there are any errors
5401 * and 1 is returned if there are no higher keys in the tree
5403 * path->keep_locks should be set to 1 on the search made before
5404 * calling this function.
5406 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5407 struct btrfs_key *key, int level, u64 min_trans)
5410 struct extent_buffer *c;
5412 WARN_ON(!path->keep_locks);
5413 while (level < BTRFS_MAX_LEVEL) {
5414 if (!path->nodes[level])
5417 slot = path->slots[level] + 1;
5418 c = path->nodes[level];
5420 if (slot >= btrfs_header_nritems(c)) {
5423 struct btrfs_key cur_key;
5424 if (level + 1 >= BTRFS_MAX_LEVEL ||
5425 !path->nodes[level + 1])
5428 if (path->locks[level + 1]) {
5433 slot = btrfs_header_nritems(c) - 1;
5435 btrfs_item_key_to_cpu(c, &cur_key, slot);
5437 btrfs_node_key_to_cpu(c, &cur_key, slot);
5439 orig_lowest = path->lowest_level;
5440 btrfs_release_path(path);
5441 path->lowest_level = level;
5442 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5444 path->lowest_level = orig_lowest;
5448 c = path->nodes[level];
5449 slot = path->slots[level];
5456 btrfs_item_key_to_cpu(c, key, slot);
5458 u64 gen = btrfs_node_ptr_generation(c, slot);
5460 if (gen < min_trans) {
5464 btrfs_node_key_to_cpu(c, key, slot);
5472 * search the tree again to find a leaf with greater keys
5473 * returns 0 if it found something or 1 if there are no greater leaves.
5474 * returns < 0 on io errors.
5476 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5478 return btrfs_next_old_leaf(root, path, 0);
5481 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5486 struct extent_buffer *c;
5487 struct extent_buffer *next;
5488 struct btrfs_key key;
5491 int old_spinning = path->leave_spinning;
5492 int next_rw_lock = 0;
5494 nritems = btrfs_header_nritems(path->nodes[0]);
5498 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5503 btrfs_release_path(path);
5505 path->keep_locks = 1;
5506 path->leave_spinning = 1;
5509 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5511 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5512 path->keep_locks = 0;
5517 nritems = btrfs_header_nritems(path->nodes[0]);
5519 * by releasing the path above we dropped all our locks. A balance
5520 * could have added more items next to the key that used to be
5521 * at the very end of the block. So, check again here and
5522 * advance the path if there are now more items available.
5524 if (nritems > 0 && path->slots[0] < nritems - 1) {
5531 while (level < BTRFS_MAX_LEVEL) {
5532 if (!path->nodes[level]) {
5537 slot = path->slots[level] + 1;
5538 c = path->nodes[level];
5539 if (slot >= btrfs_header_nritems(c)) {
5541 if (level == BTRFS_MAX_LEVEL) {
5549 btrfs_tree_unlock_rw(next, next_rw_lock);
5550 free_extent_buffer(next);
5554 next_rw_lock = path->locks[level];
5555 ret = read_block_for_search(NULL, root, path, &next, level,
5561 btrfs_release_path(path);
5565 if (!path->skip_locking) {
5566 ret = btrfs_try_tree_read_lock(next);
5567 if (!ret && time_seq) {
5569 * If we don't get the lock, we may be racing
5570 * with push_leaf_left, holding that lock while
5571 * itself waiting for the leaf we've currently
5572 * locked. To solve this situation, we give up
5573 * on our lock and cycle.
5575 free_extent_buffer(next);
5576 btrfs_release_path(path);
5581 btrfs_set_path_blocking(path);
5582 btrfs_tree_read_lock(next);
5583 btrfs_clear_path_blocking(path, next,
5586 next_rw_lock = BTRFS_READ_LOCK;
5590 path->slots[level] = slot;
5593 c = path->nodes[level];
5594 if (path->locks[level])
5595 btrfs_tree_unlock_rw(c, path->locks[level]);
5597 free_extent_buffer(c);
5598 path->nodes[level] = next;
5599 path->slots[level] = 0;
5600 if (!path->skip_locking)
5601 path->locks[level] = next_rw_lock;
5605 ret = read_block_for_search(NULL, root, path, &next, level,
5611 btrfs_release_path(path);
5615 if (!path->skip_locking) {
5616 ret = btrfs_try_tree_read_lock(next);
5618 btrfs_set_path_blocking(path);
5619 btrfs_tree_read_lock(next);
5620 btrfs_clear_path_blocking(path, next,
5623 next_rw_lock = BTRFS_READ_LOCK;
5628 unlock_up(path, 0, 1, 0, NULL);
5629 path->leave_spinning = old_spinning;
5631 btrfs_set_path_blocking(path);
5637 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5638 * searching until it gets past min_objectid or finds an item of 'type'
5640 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5642 int btrfs_previous_item(struct btrfs_root *root,
5643 struct btrfs_path *path, u64 min_objectid,
5646 struct btrfs_key found_key;
5647 struct extent_buffer *leaf;
5652 if (path->slots[0] == 0) {
5653 btrfs_set_path_blocking(path);
5654 ret = btrfs_prev_leaf(root, path);
5660 leaf = path->nodes[0];
5661 nritems = btrfs_header_nritems(leaf);
5664 if (path->slots[0] == nritems)
5667 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5668 if (found_key.objectid < min_objectid)
5670 if (found_key.type == type)
5672 if (found_key.objectid == min_objectid &&
5673 found_key.type < type)