1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007,2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/rbtree.h>
12 #include "transaction.h"
13 #include "print-tree.h"
16 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
17 *root, struct btrfs_path *path, int level);
18 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
19 const struct btrfs_key *ins_key, struct btrfs_path *path,
20 int data_size, int extend);
21 static int push_node_left(struct btrfs_trans_handle *trans,
22 struct btrfs_fs_info *fs_info,
23 struct extent_buffer *dst,
24 struct extent_buffer *src, int empty);
25 static int balance_node_right(struct btrfs_trans_handle *trans,
26 struct btrfs_fs_info *fs_info,
27 struct extent_buffer *dst_buf,
28 struct extent_buffer *src_buf);
29 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
32 struct btrfs_path *btrfs_alloc_path(void)
34 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
38 * set all locked nodes in the path to blocking locks. This should
39 * be done before scheduling
41 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
44 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
45 if (!p->nodes[i] || !p->locks[i])
47 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
48 if (p->locks[i] == BTRFS_READ_LOCK)
49 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
50 else if (p->locks[i] == BTRFS_WRITE_LOCK)
51 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
55 /* this also releases the path */
56 void btrfs_free_path(struct btrfs_path *p)
60 btrfs_release_path(p);
61 kmem_cache_free(btrfs_path_cachep, p);
65 * path release drops references on the extent buffers in the path
66 * and it drops any locks held by this path
68 * It is safe to call this on paths that no locks or extent buffers held.
70 noinline void btrfs_release_path(struct btrfs_path *p)
74 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
79 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
82 free_extent_buffer(p->nodes[i]);
88 * safely gets a reference on the root node of a tree. A lock
89 * is not taken, so a concurrent writer may put a different node
90 * at the root of the tree. See btrfs_lock_root_node for the
93 * The extent buffer returned by this has a reference taken, so
94 * it won't disappear. It may stop being the root of the tree
95 * at any time because there are no locks held.
97 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
99 struct extent_buffer *eb;
103 eb = rcu_dereference(root->node);
106 * RCU really hurts here, we could free up the root node because
107 * it was COWed but we may not get the new root node yet so do
108 * the inc_not_zero dance and if it doesn't work then
109 * synchronize_rcu and try again.
111 if (atomic_inc_not_zero(&eb->refs)) {
121 /* loop around taking references on and locking the root node of the
122 * tree until you end up with a lock on the root. A locked buffer
123 * is returned, with a reference held.
125 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
127 struct extent_buffer *eb;
130 eb = btrfs_root_node(root);
132 if (eb == root->node)
134 btrfs_tree_unlock(eb);
135 free_extent_buffer(eb);
140 /* loop around taking references on and locking the root node of the
141 * tree until you end up with a lock on the root. A locked buffer
142 * is returned, with a reference held.
144 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
146 struct extent_buffer *eb;
149 eb = btrfs_root_node(root);
150 btrfs_tree_read_lock(eb);
151 if (eb == root->node)
153 btrfs_tree_read_unlock(eb);
154 free_extent_buffer(eb);
159 /* cowonly root (everything not a reference counted cow subvolume), just get
160 * put onto a simple dirty list. transaction.c walks this to make sure they
161 * get properly updated on disk.
163 static void add_root_to_dirty_list(struct btrfs_root *root)
165 struct btrfs_fs_info *fs_info = root->fs_info;
167 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
168 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
171 spin_lock(&fs_info->trans_lock);
172 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
173 /* Want the extent tree to be the last on the list */
174 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
175 list_move_tail(&root->dirty_list,
176 &fs_info->dirty_cowonly_roots);
178 list_move(&root->dirty_list,
179 &fs_info->dirty_cowonly_roots);
181 spin_unlock(&fs_info->trans_lock);
185 * used by snapshot creation to make a copy of a root for a tree with
186 * a given objectid. The buffer with the new root node is returned in
187 * cow_ret, and this func returns zero on success or a negative error code.
189 int btrfs_copy_root(struct btrfs_trans_handle *trans,
190 struct btrfs_root *root,
191 struct extent_buffer *buf,
192 struct extent_buffer **cow_ret, u64 new_root_objectid)
194 struct btrfs_fs_info *fs_info = root->fs_info;
195 struct extent_buffer *cow;
198 struct btrfs_disk_key disk_key;
200 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
201 trans->transid != fs_info->running_transaction->transid);
202 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
203 trans->transid != root->last_trans);
205 level = btrfs_header_level(buf);
207 btrfs_item_key(buf, &disk_key, 0);
209 btrfs_node_key(buf, &disk_key, 0);
211 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
212 &disk_key, level, buf->start, 0);
216 copy_extent_buffer_full(cow, buf);
217 btrfs_set_header_bytenr(cow, cow->start);
218 btrfs_set_header_generation(cow, trans->transid);
219 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
220 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
221 BTRFS_HEADER_FLAG_RELOC);
222 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
223 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
225 btrfs_set_header_owner(cow, new_root_objectid);
227 write_extent_buffer_fsid(cow, fs_info->fsid);
229 WARN_ON(btrfs_header_generation(buf) > trans->transid);
230 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
231 ret = btrfs_inc_ref(trans, root, cow, 1);
233 ret = btrfs_inc_ref(trans, root, cow, 0);
238 btrfs_mark_buffer_dirty(cow);
247 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
248 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
250 MOD_LOG_ROOT_REPLACE,
253 struct tree_mod_root {
258 struct tree_mod_elem {
264 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
267 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
270 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
271 struct btrfs_disk_key key;
274 /* this is used for op == MOD_LOG_MOVE_KEYS */
280 /* this is used for op == MOD_LOG_ROOT_REPLACE */
281 struct tree_mod_root old_root;
285 * Pull a new tree mod seq number for our operation.
287 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
289 return atomic64_inc_return(&fs_info->tree_mod_seq);
293 * This adds a new blocker to the tree mod log's blocker list if the @elem
294 * passed does not already have a sequence number set. So when a caller expects
295 * to record tree modifications, it should ensure to set elem->seq to zero
296 * before calling btrfs_get_tree_mod_seq.
297 * Returns a fresh, unused tree log modification sequence number, even if no new
300 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
301 struct seq_list *elem)
303 write_lock(&fs_info->tree_mod_log_lock);
304 spin_lock(&fs_info->tree_mod_seq_lock);
306 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
307 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
309 spin_unlock(&fs_info->tree_mod_seq_lock);
310 write_unlock(&fs_info->tree_mod_log_lock);
315 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
316 struct seq_list *elem)
318 struct rb_root *tm_root;
319 struct rb_node *node;
320 struct rb_node *next;
321 struct seq_list *cur_elem;
322 struct tree_mod_elem *tm;
323 u64 min_seq = (u64)-1;
324 u64 seq_putting = elem->seq;
329 spin_lock(&fs_info->tree_mod_seq_lock);
330 list_del(&elem->list);
333 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
334 if (cur_elem->seq < min_seq) {
335 if (seq_putting > cur_elem->seq) {
337 * blocker with lower sequence number exists, we
338 * cannot remove anything from the log
340 spin_unlock(&fs_info->tree_mod_seq_lock);
343 min_seq = cur_elem->seq;
346 spin_unlock(&fs_info->tree_mod_seq_lock);
349 * anything that's lower than the lowest existing (read: blocked)
350 * sequence number can be removed from the tree.
352 write_lock(&fs_info->tree_mod_log_lock);
353 tm_root = &fs_info->tree_mod_log;
354 for (node = rb_first(tm_root); node; node = next) {
355 next = rb_next(node);
356 tm = rb_entry(node, struct tree_mod_elem, node);
357 if (tm->seq > min_seq)
359 rb_erase(node, tm_root);
362 write_unlock(&fs_info->tree_mod_log_lock);
366 * key order of the log:
367 * node/leaf start address -> sequence
369 * The 'start address' is the logical address of the *new* root node
370 * for root replace operations, or the logical address of the affected
371 * block for all other operations.
373 * Note: must be called with write lock for fs_info::tree_mod_log_lock.
376 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
378 struct rb_root *tm_root;
379 struct rb_node **new;
380 struct rb_node *parent = NULL;
381 struct tree_mod_elem *cur;
383 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
385 tm_root = &fs_info->tree_mod_log;
386 new = &tm_root->rb_node;
388 cur = rb_entry(*new, struct tree_mod_elem, node);
390 if (cur->logical < tm->logical)
391 new = &((*new)->rb_left);
392 else if (cur->logical > tm->logical)
393 new = &((*new)->rb_right);
394 else if (cur->seq < tm->seq)
395 new = &((*new)->rb_left);
396 else if (cur->seq > tm->seq)
397 new = &((*new)->rb_right);
402 rb_link_node(&tm->node, parent, new);
403 rb_insert_color(&tm->node, tm_root);
408 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
409 * returns zero with the tree_mod_log_lock acquired. The caller must hold
410 * this until all tree mod log insertions are recorded in the rb tree and then
411 * write unlock fs_info::tree_mod_log_lock.
413 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
414 struct extent_buffer *eb) {
416 if (list_empty(&(fs_info)->tree_mod_seq_list))
418 if (eb && btrfs_header_level(eb) == 0)
421 write_lock(&fs_info->tree_mod_log_lock);
422 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
423 write_unlock(&fs_info->tree_mod_log_lock);
430 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
431 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
432 struct extent_buffer *eb)
435 if (list_empty(&(fs_info)->tree_mod_seq_list))
437 if (eb && btrfs_header_level(eb) == 0)
443 static struct tree_mod_elem *
444 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
445 enum mod_log_op op, gfp_t flags)
447 struct tree_mod_elem *tm;
449 tm = kzalloc(sizeof(*tm), flags);
453 tm->logical = eb->start;
454 if (op != MOD_LOG_KEY_ADD) {
455 btrfs_node_key(eb, &tm->key, slot);
456 tm->blockptr = btrfs_node_blockptr(eb, slot);
460 tm->generation = btrfs_node_ptr_generation(eb, slot);
461 RB_CLEAR_NODE(&tm->node);
466 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
467 enum mod_log_op op, gfp_t flags)
469 struct tree_mod_elem *tm;
472 if (!tree_mod_need_log(eb->fs_info, eb))
475 tm = alloc_tree_mod_elem(eb, slot, op, flags);
479 if (tree_mod_dont_log(eb->fs_info, eb)) {
484 ret = __tree_mod_log_insert(eb->fs_info, tm);
485 write_unlock(&eb->fs_info->tree_mod_log_lock);
492 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
493 int dst_slot, int src_slot, int nr_items)
495 struct tree_mod_elem *tm = NULL;
496 struct tree_mod_elem **tm_list = NULL;
501 if (!tree_mod_need_log(eb->fs_info, eb))
504 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
508 tm = kzalloc(sizeof(*tm), GFP_NOFS);
514 tm->logical = eb->start;
516 tm->move.dst_slot = dst_slot;
517 tm->move.nr_items = nr_items;
518 tm->op = MOD_LOG_MOVE_KEYS;
520 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
521 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
522 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
529 if (tree_mod_dont_log(eb->fs_info, eb))
534 * When we override something during the move, we log these removals.
535 * This can only happen when we move towards the beginning of the
536 * buffer, i.e. dst_slot < src_slot.
538 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
539 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
544 ret = __tree_mod_log_insert(eb->fs_info, tm);
547 write_unlock(&eb->fs_info->tree_mod_log_lock);
552 for (i = 0; i < nr_items; i++) {
553 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
554 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
558 write_unlock(&eb->fs_info->tree_mod_log_lock);
566 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
567 struct tree_mod_elem **tm_list,
573 for (i = nritems - 1; i >= 0; i--) {
574 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
576 for (j = nritems - 1; j > i; j--)
577 rb_erase(&tm_list[j]->node,
578 &fs_info->tree_mod_log);
586 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
587 struct extent_buffer *new_root, int log_removal)
589 struct btrfs_fs_info *fs_info = old_root->fs_info;
590 struct tree_mod_elem *tm = NULL;
591 struct tree_mod_elem **tm_list = NULL;
596 if (!tree_mod_need_log(fs_info, NULL))
599 if (log_removal && btrfs_header_level(old_root) > 0) {
600 nritems = btrfs_header_nritems(old_root);
601 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
607 for (i = 0; i < nritems; i++) {
608 tm_list[i] = alloc_tree_mod_elem(old_root, i,
609 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
617 tm = kzalloc(sizeof(*tm), GFP_NOFS);
623 tm->logical = new_root->start;
624 tm->old_root.logical = old_root->start;
625 tm->old_root.level = btrfs_header_level(old_root);
626 tm->generation = btrfs_header_generation(old_root);
627 tm->op = MOD_LOG_ROOT_REPLACE;
629 if (tree_mod_dont_log(fs_info, NULL))
633 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
635 ret = __tree_mod_log_insert(fs_info, tm);
637 write_unlock(&fs_info->tree_mod_log_lock);
646 for (i = 0; i < nritems; i++)
655 static struct tree_mod_elem *
656 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
659 struct rb_root *tm_root;
660 struct rb_node *node;
661 struct tree_mod_elem *cur = NULL;
662 struct tree_mod_elem *found = NULL;
664 read_lock(&fs_info->tree_mod_log_lock);
665 tm_root = &fs_info->tree_mod_log;
666 node = tm_root->rb_node;
668 cur = rb_entry(node, struct tree_mod_elem, node);
669 if (cur->logical < start) {
670 node = node->rb_left;
671 } else if (cur->logical > start) {
672 node = node->rb_right;
673 } else if (cur->seq < min_seq) {
674 node = node->rb_left;
675 } else if (!smallest) {
676 /* we want the node with the highest seq */
678 BUG_ON(found->seq > cur->seq);
680 node = node->rb_left;
681 } else if (cur->seq > min_seq) {
682 /* we want the node with the smallest seq */
684 BUG_ON(found->seq < cur->seq);
686 node = node->rb_right;
692 read_unlock(&fs_info->tree_mod_log_lock);
698 * this returns the element from the log with the smallest time sequence
699 * value that's in the log (the oldest log item). any element with a time
700 * sequence lower than min_seq will be ignored.
702 static struct tree_mod_elem *
703 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
706 return __tree_mod_log_search(fs_info, start, min_seq, 1);
710 * this returns the element from the log with the largest time sequence
711 * value that's in the log (the most recent log item). any element with
712 * a time sequence lower than min_seq will be ignored.
714 static struct tree_mod_elem *
715 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
717 return __tree_mod_log_search(fs_info, start, min_seq, 0);
721 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
722 struct extent_buffer *src, unsigned long dst_offset,
723 unsigned long src_offset, int nr_items)
726 struct tree_mod_elem **tm_list = NULL;
727 struct tree_mod_elem **tm_list_add, **tm_list_rem;
731 if (!tree_mod_need_log(fs_info, NULL))
734 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
737 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
742 tm_list_add = tm_list;
743 tm_list_rem = tm_list + nr_items;
744 for (i = 0; i < nr_items; i++) {
745 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
746 MOD_LOG_KEY_REMOVE, GFP_NOFS);
747 if (!tm_list_rem[i]) {
752 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
753 MOD_LOG_KEY_ADD, GFP_NOFS);
754 if (!tm_list_add[i]) {
760 if (tree_mod_dont_log(fs_info, NULL))
764 for (i = 0; i < nr_items; i++) {
765 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
768 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
773 write_unlock(&fs_info->tree_mod_log_lock);
779 for (i = 0; i < nr_items * 2; i++) {
780 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
781 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
785 write_unlock(&fs_info->tree_mod_log_lock);
791 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
793 struct tree_mod_elem **tm_list = NULL;
798 if (btrfs_header_level(eb) == 0)
801 if (!tree_mod_need_log(eb->fs_info, NULL))
804 nritems = btrfs_header_nritems(eb);
805 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
809 for (i = 0; i < nritems; i++) {
810 tm_list[i] = alloc_tree_mod_elem(eb, i,
811 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
818 if (tree_mod_dont_log(eb->fs_info, eb))
821 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
822 write_unlock(&eb->fs_info->tree_mod_log_lock);
830 for (i = 0; i < nritems; i++)
838 * check if the tree block can be shared by multiple trees
840 int btrfs_block_can_be_shared(struct btrfs_root *root,
841 struct extent_buffer *buf)
844 * Tree blocks not in reference counted trees and tree roots
845 * are never shared. If a block was allocated after the last
846 * snapshot and the block was not allocated by tree relocation,
847 * we know the block is not shared.
849 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
850 buf != root->node && buf != root->commit_root &&
851 (btrfs_header_generation(buf) <=
852 btrfs_root_last_snapshot(&root->root_item) ||
853 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
859 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
860 struct btrfs_root *root,
861 struct extent_buffer *buf,
862 struct extent_buffer *cow,
865 struct btrfs_fs_info *fs_info = root->fs_info;
873 * Backrefs update rules:
875 * Always use full backrefs for extent pointers in tree block
876 * allocated by tree relocation.
878 * If a shared tree block is no longer referenced by its owner
879 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
880 * use full backrefs for extent pointers in tree block.
882 * If a tree block is been relocating
883 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
884 * use full backrefs for extent pointers in tree block.
885 * The reason for this is some operations (such as drop tree)
886 * are only allowed for blocks use full backrefs.
889 if (btrfs_block_can_be_shared(root, buf)) {
890 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
891 btrfs_header_level(buf), 1,
897 btrfs_handle_fs_error(fs_info, ret, NULL);
902 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
903 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
904 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
909 owner = btrfs_header_owner(buf);
910 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
911 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
914 if ((owner == root->root_key.objectid ||
915 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
916 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
917 ret = btrfs_inc_ref(trans, root, buf, 1);
921 if (root->root_key.objectid ==
922 BTRFS_TREE_RELOC_OBJECTID) {
923 ret = btrfs_dec_ref(trans, root, buf, 0);
926 ret = btrfs_inc_ref(trans, root, cow, 1);
930 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
933 if (root->root_key.objectid ==
934 BTRFS_TREE_RELOC_OBJECTID)
935 ret = btrfs_inc_ref(trans, root, cow, 1);
937 ret = btrfs_inc_ref(trans, root, cow, 0);
941 if (new_flags != 0) {
942 int level = btrfs_header_level(buf);
944 ret = btrfs_set_disk_extent_flags(trans, fs_info,
947 new_flags, level, 0);
952 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
953 if (root->root_key.objectid ==
954 BTRFS_TREE_RELOC_OBJECTID)
955 ret = btrfs_inc_ref(trans, root, cow, 1);
957 ret = btrfs_inc_ref(trans, root, cow, 0);
960 ret = btrfs_dec_ref(trans, root, buf, 1);
964 clean_tree_block(fs_info, buf);
971 * does the dirty work in cow of a single block. The parent block (if
972 * supplied) is updated to point to the new cow copy. The new buffer is marked
973 * dirty and returned locked. If you modify the block it needs to be marked
976 * search_start -- an allocation hint for the new block
978 * empty_size -- a hint that you plan on doing more cow. This is the size in
979 * bytes the allocator should try to find free next to the block it returns.
980 * This is just a hint and may be ignored by the allocator.
982 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
983 struct btrfs_root *root,
984 struct extent_buffer *buf,
985 struct extent_buffer *parent, int parent_slot,
986 struct extent_buffer **cow_ret,
987 u64 search_start, u64 empty_size)
989 struct btrfs_fs_info *fs_info = root->fs_info;
990 struct btrfs_disk_key disk_key;
991 struct extent_buffer *cow;
995 u64 parent_start = 0;
1000 btrfs_assert_tree_locked(buf);
1002 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1003 trans->transid != fs_info->running_transaction->transid);
1004 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1005 trans->transid != root->last_trans);
1007 level = btrfs_header_level(buf);
1010 btrfs_item_key(buf, &disk_key, 0);
1012 btrfs_node_key(buf, &disk_key, 0);
1014 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1015 parent_start = parent->start;
1018 * If we are COWing a node/leaf from the extent, chunk or device trees,
1019 * make sure that we do not finish block group creation of pending block
1020 * groups. We do this to avoid a deadlock.
1021 * COWing can result in allocation of a new chunk, and flushing pending
1022 * block groups (btrfs_create_pending_block_groups()) can be triggered
1023 * when finishing allocation of a new chunk. Creation of a pending block
1024 * group modifies the extent, chunk and device trees, therefore we could
1025 * deadlock with ourselves since we are holding a lock on an extent
1026 * buffer that btrfs_create_pending_block_groups() may try to COW later.
1028 if (root == fs_info->extent_root ||
1029 root == fs_info->chunk_root ||
1030 root == fs_info->dev_root)
1031 trans->can_flush_pending_bgs = false;
1033 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1034 root->root_key.objectid, &disk_key, level,
1035 search_start, empty_size);
1036 trans->can_flush_pending_bgs = true;
1038 return PTR_ERR(cow);
1040 /* cow is set to blocking by btrfs_init_new_buffer */
1042 copy_extent_buffer_full(cow, buf);
1043 btrfs_set_header_bytenr(cow, cow->start);
1044 btrfs_set_header_generation(cow, trans->transid);
1045 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1046 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1047 BTRFS_HEADER_FLAG_RELOC);
1048 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1049 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1051 btrfs_set_header_owner(cow, root->root_key.objectid);
1053 write_extent_buffer_fsid(cow, fs_info->fsid);
1055 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1057 btrfs_abort_transaction(trans, ret);
1061 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1062 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1064 btrfs_abort_transaction(trans, ret);
1069 if (buf == root->node) {
1070 WARN_ON(parent && parent != buf);
1071 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1072 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1073 parent_start = buf->start;
1075 extent_buffer_get(cow);
1076 ret = tree_mod_log_insert_root(root->node, cow, 1);
1078 rcu_assign_pointer(root->node, cow);
1080 btrfs_free_tree_block(trans, root, buf, parent_start,
1082 free_extent_buffer(buf);
1083 add_root_to_dirty_list(root);
1085 WARN_ON(trans->transid != btrfs_header_generation(parent));
1086 tree_mod_log_insert_key(parent, parent_slot,
1087 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1088 btrfs_set_node_blockptr(parent, parent_slot,
1090 btrfs_set_node_ptr_generation(parent, parent_slot,
1092 btrfs_mark_buffer_dirty(parent);
1094 ret = tree_mod_log_free_eb(buf);
1096 btrfs_abort_transaction(trans, ret);
1100 btrfs_free_tree_block(trans, root, buf, parent_start,
1104 btrfs_tree_unlock(buf);
1105 free_extent_buffer_stale(buf);
1106 btrfs_mark_buffer_dirty(cow);
1112 * returns the logical address of the oldest predecessor of the given root.
1113 * entries older than time_seq are ignored.
1115 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1116 struct extent_buffer *eb_root, u64 time_seq)
1118 struct tree_mod_elem *tm;
1119 struct tree_mod_elem *found = NULL;
1120 u64 root_logical = eb_root->start;
1127 * the very last operation that's logged for a root is the
1128 * replacement operation (if it is replaced at all). this has
1129 * the logical address of the *new* root, making it the very
1130 * first operation that's logged for this root.
1133 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1138 * if there are no tree operation for the oldest root, we simply
1139 * return it. this should only happen if that (old) root is at
1146 * if there's an operation that's not a root replacement, we
1147 * found the oldest version of our root. normally, we'll find a
1148 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1150 if (tm->op != MOD_LOG_ROOT_REPLACE)
1154 root_logical = tm->old_root.logical;
1158 /* if there's no old root to return, return what we found instead */
1166 * tm is a pointer to the first operation to rewind within eb. then, all
1167 * previous operations will be rewound (until we reach something older than
1171 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1172 u64 time_seq, struct tree_mod_elem *first_tm)
1175 struct rb_node *next;
1176 struct tree_mod_elem *tm = first_tm;
1177 unsigned long o_dst;
1178 unsigned long o_src;
1179 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1181 n = btrfs_header_nritems(eb);
1182 read_lock(&fs_info->tree_mod_log_lock);
1183 while (tm && tm->seq >= time_seq) {
1185 * all the operations are recorded with the operator used for
1186 * the modification. as we're going backwards, we do the
1187 * opposite of each operation here.
1190 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1191 BUG_ON(tm->slot < n);
1193 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1194 case MOD_LOG_KEY_REMOVE:
1195 btrfs_set_node_key(eb, &tm->key, tm->slot);
1196 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1197 btrfs_set_node_ptr_generation(eb, tm->slot,
1201 case MOD_LOG_KEY_REPLACE:
1202 BUG_ON(tm->slot >= n);
1203 btrfs_set_node_key(eb, &tm->key, tm->slot);
1204 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1205 btrfs_set_node_ptr_generation(eb, tm->slot,
1208 case MOD_LOG_KEY_ADD:
1209 /* if a move operation is needed it's in the log */
1212 case MOD_LOG_MOVE_KEYS:
1213 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1214 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1215 memmove_extent_buffer(eb, o_dst, o_src,
1216 tm->move.nr_items * p_size);
1218 case MOD_LOG_ROOT_REPLACE:
1220 * this operation is special. for roots, this must be
1221 * handled explicitly before rewinding.
1222 * for non-roots, this operation may exist if the node
1223 * was a root: root A -> child B; then A gets empty and
1224 * B is promoted to the new root. in the mod log, we'll
1225 * have a root-replace operation for B, a tree block
1226 * that is no root. we simply ignore that operation.
1230 next = rb_next(&tm->node);
1233 tm = rb_entry(next, struct tree_mod_elem, node);
1234 if (tm->logical != first_tm->logical)
1237 read_unlock(&fs_info->tree_mod_log_lock);
1238 btrfs_set_header_nritems(eb, n);
1242 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1243 * is returned. If rewind operations happen, a fresh buffer is returned. The
1244 * returned buffer is always read-locked. If the returned buffer is not the
1245 * input buffer, the lock on the input buffer is released and the input buffer
1246 * is freed (its refcount is decremented).
1248 static struct extent_buffer *
1249 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1250 struct extent_buffer *eb, u64 time_seq)
1252 struct extent_buffer *eb_rewin;
1253 struct tree_mod_elem *tm;
1258 if (btrfs_header_level(eb) == 0)
1261 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1265 btrfs_set_path_blocking(path);
1266 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1268 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1269 BUG_ON(tm->slot != 0);
1270 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1272 btrfs_tree_read_unlock_blocking(eb);
1273 free_extent_buffer(eb);
1276 btrfs_set_header_bytenr(eb_rewin, eb->start);
1277 btrfs_set_header_backref_rev(eb_rewin,
1278 btrfs_header_backref_rev(eb));
1279 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1280 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1282 eb_rewin = btrfs_clone_extent_buffer(eb);
1284 btrfs_tree_read_unlock_blocking(eb);
1285 free_extent_buffer(eb);
1290 btrfs_tree_read_unlock_blocking(eb);
1291 free_extent_buffer(eb);
1293 extent_buffer_get(eb_rewin);
1294 btrfs_tree_read_lock(eb_rewin);
1295 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1296 WARN_ON(btrfs_header_nritems(eb_rewin) >
1297 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1303 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1304 * value. If there are no changes, the current root->root_node is returned. If
1305 * anything changed in between, there's a fresh buffer allocated on which the
1306 * rewind operations are done. In any case, the returned buffer is read locked.
1307 * Returns NULL on error (with no locks held).
1309 static inline struct extent_buffer *
1310 get_old_root(struct btrfs_root *root, u64 time_seq)
1312 struct btrfs_fs_info *fs_info = root->fs_info;
1313 struct tree_mod_elem *tm;
1314 struct extent_buffer *eb = NULL;
1315 struct extent_buffer *eb_root;
1316 struct extent_buffer *old;
1317 struct tree_mod_root *old_root = NULL;
1318 u64 old_generation = 0;
1322 eb_root = btrfs_read_lock_root_node(root);
1323 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1327 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1328 old_root = &tm->old_root;
1329 old_generation = tm->generation;
1330 logical = old_root->logical;
1331 level = old_root->level;
1333 logical = eb_root->start;
1334 level = btrfs_header_level(eb_root);
1337 tm = tree_mod_log_search(fs_info, logical, time_seq);
1338 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1339 btrfs_tree_read_unlock(eb_root);
1340 free_extent_buffer(eb_root);
1341 old = read_tree_block(fs_info, logical, 0, level, NULL);
1342 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1344 free_extent_buffer(old);
1346 "failed to read tree block %llu from get_old_root",
1349 eb = btrfs_clone_extent_buffer(old);
1350 free_extent_buffer(old);
1352 } else if (old_root) {
1353 btrfs_tree_read_unlock(eb_root);
1354 free_extent_buffer(eb_root);
1355 eb = alloc_dummy_extent_buffer(fs_info, logical);
1357 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1358 eb = btrfs_clone_extent_buffer(eb_root);
1359 btrfs_tree_read_unlock_blocking(eb_root);
1360 free_extent_buffer(eb_root);
1365 extent_buffer_get(eb);
1366 btrfs_tree_read_lock(eb);
1368 btrfs_set_header_bytenr(eb, eb->start);
1369 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1370 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1371 btrfs_set_header_level(eb, old_root->level);
1372 btrfs_set_header_generation(eb, old_generation);
1375 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1377 WARN_ON(btrfs_header_level(eb) != 0);
1378 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1383 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1385 struct tree_mod_elem *tm;
1387 struct extent_buffer *eb_root = btrfs_root_node(root);
1389 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1390 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1391 level = tm->old_root.level;
1393 level = btrfs_header_level(eb_root);
1395 free_extent_buffer(eb_root);
1400 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1401 struct btrfs_root *root,
1402 struct extent_buffer *buf)
1404 if (btrfs_is_testing(root->fs_info))
1407 /* Ensure we can see the FORCE_COW bit */
1408 smp_mb__before_atomic();
1411 * We do not need to cow a block if
1412 * 1) this block is not created or changed in this transaction;
1413 * 2) this block does not belong to TREE_RELOC tree;
1414 * 3) the root is not forced COW.
1416 * What is forced COW:
1417 * when we create snapshot during committing the transaction,
1418 * after we've finished coping src root, we must COW the shared
1419 * block to ensure the metadata consistency.
1421 if (btrfs_header_generation(buf) == trans->transid &&
1422 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1423 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1424 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1425 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1431 * cows a single block, see __btrfs_cow_block for the real work.
1432 * This version of it has extra checks so that a block isn't COWed more than
1433 * once per transaction, as long as it hasn't been written yet
1435 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1436 struct btrfs_root *root, struct extent_buffer *buf,
1437 struct extent_buffer *parent, int parent_slot,
1438 struct extent_buffer **cow_ret)
1440 struct btrfs_fs_info *fs_info = root->fs_info;
1444 if (trans->transaction != fs_info->running_transaction)
1445 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1447 fs_info->running_transaction->transid);
1449 if (trans->transid != fs_info->generation)
1450 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1451 trans->transid, fs_info->generation);
1453 if (!should_cow_block(trans, root, buf)) {
1454 trans->dirty = true;
1459 search_start = buf->start & ~((u64)SZ_1G - 1);
1462 btrfs_set_lock_blocking(parent);
1463 btrfs_set_lock_blocking(buf);
1465 ret = __btrfs_cow_block(trans, root, buf, parent,
1466 parent_slot, cow_ret, search_start, 0);
1468 trace_btrfs_cow_block(root, buf, *cow_ret);
1474 * helper function for defrag to decide if two blocks pointed to by a
1475 * node are actually close by
1477 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1479 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1481 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1487 * compare two keys in a memcmp fashion
1489 static int comp_keys(const struct btrfs_disk_key *disk,
1490 const struct btrfs_key *k2)
1492 struct btrfs_key k1;
1494 btrfs_disk_key_to_cpu(&k1, disk);
1496 return btrfs_comp_cpu_keys(&k1, k2);
1500 * same as comp_keys only with two btrfs_key's
1502 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1504 if (k1->objectid > k2->objectid)
1506 if (k1->objectid < k2->objectid)
1508 if (k1->type > k2->type)
1510 if (k1->type < k2->type)
1512 if (k1->offset > k2->offset)
1514 if (k1->offset < k2->offset)
1520 * this is used by the defrag code to go through all the
1521 * leaves pointed to by a node and reallocate them so that
1522 * disk order is close to key order
1524 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1525 struct btrfs_root *root, struct extent_buffer *parent,
1526 int start_slot, u64 *last_ret,
1527 struct btrfs_key *progress)
1529 struct btrfs_fs_info *fs_info = root->fs_info;
1530 struct extent_buffer *cur;
1533 u64 search_start = *last_ret;
1543 int progress_passed = 0;
1544 struct btrfs_disk_key disk_key;
1546 parent_level = btrfs_header_level(parent);
1548 WARN_ON(trans->transaction != fs_info->running_transaction);
1549 WARN_ON(trans->transid != fs_info->generation);
1551 parent_nritems = btrfs_header_nritems(parent);
1552 blocksize = fs_info->nodesize;
1553 end_slot = parent_nritems - 1;
1555 if (parent_nritems <= 1)
1558 btrfs_set_lock_blocking(parent);
1560 for (i = start_slot; i <= end_slot; i++) {
1561 struct btrfs_key first_key;
1564 btrfs_node_key(parent, &disk_key, i);
1565 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1568 progress_passed = 1;
1569 blocknr = btrfs_node_blockptr(parent, i);
1570 gen = btrfs_node_ptr_generation(parent, i);
1571 btrfs_node_key_to_cpu(parent, &first_key, i);
1572 if (last_block == 0)
1573 last_block = blocknr;
1576 other = btrfs_node_blockptr(parent, i - 1);
1577 close = close_blocks(blocknr, other, blocksize);
1579 if (!close && i < end_slot) {
1580 other = btrfs_node_blockptr(parent, i + 1);
1581 close = close_blocks(blocknr, other, blocksize);
1584 last_block = blocknr;
1588 cur = find_extent_buffer(fs_info, blocknr);
1590 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1593 if (!cur || !uptodate) {
1595 cur = read_tree_block(fs_info, blocknr, gen,
1599 return PTR_ERR(cur);
1600 } else if (!extent_buffer_uptodate(cur)) {
1601 free_extent_buffer(cur);
1604 } else if (!uptodate) {
1605 err = btrfs_read_buffer(cur, gen,
1606 parent_level - 1,&first_key);
1608 free_extent_buffer(cur);
1613 if (search_start == 0)
1614 search_start = last_block;
1616 btrfs_tree_lock(cur);
1617 btrfs_set_lock_blocking(cur);
1618 err = __btrfs_cow_block(trans, root, cur, parent, i,
1621 (end_slot - i) * blocksize));
1623 btrfs_tree_unlock(cur);
1624 free_extent_buffer(cur);
1627 search_start = cur->start;
1628 last_block = cur->start;
1629 *last_ret = search_start;
1630 btrfs_tree_unlock(cur);
1631 free_extent_buffer(cur);
1637 * search for key in the extent_buffer. The items start at offset p,
1638 * and they are item_size apart. There are 'max' items in p.
1640 * the slot in the array is returned via slot, and it points to
1641 * the place where you would insert key if it is not found in
1644 * slot may point to max if the key is bigger than all of the keys
1646 static noinline int generic_bin_search(struct extent_buffer *eb,
1647 unsigned long p, int item_size,
1648 const struct btrfs_key *key,
1655 struct btrfs_disk_key *tmp = NULL;
1656 struct btrfs_disk_key unaligned;
1657 unsigned long offset;
1659 unsigned long map_start = 0;
1660 unsigned long map_len = 0;
1664 btrfs_err(eb->fs_info,
1665 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1666 __func__, low, high, eb->start,
1667 btrfs_header_owner(eb), btrfs_header_level(eb));
1671 while (low < high) {
1672 mid = (low + high) / 2;
1673 offset = p + mid * item_size;
1675 if (!kaddr || offset < map_start ||
1676 (offset + sizeof(struct btrfs_disk_key)) >
1677 map_start + map_len) {
1679 err = map_private_extent_buffer(eb, offset,
1680 sizeof(struct btrfs_disk_key),
1681 &kaddr, &map_start, &map_len);
1684 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1686 } else if (err == 1) {
1687 read_extent_buffer(eb, &unaligned,
1688 offset, sizeof(unaligned));
1695 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1698 ret = comp_keys(tmp, key);
1714 * simple bin_search frontend that does the right thing for
1717 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1718 int level, int *slot)
1721 return generic_bin_search(eb,
1722 offsetof(struct btrfs_leaf, items),
1723 sizeof(struct btrfs_item),
1724 key, btrfs_header_nritems(eb),
1727 return generic_bin_search(eb,
1728 offsetof(struct btrfs_node, ptrs),
1729 sizeof(struct btrfs_key_ptr),
1730 key, btrfs_header_nritems(eb),
1734 static void root_add_used(struct btrfs_root *root, u32 size)
1736 spin_lock(&root->accounting_lock);
1737 btrfs_set_root_used(&root->root_item,
1738 btrfs_root_used(&root->root_item) + size);
1739 spin_unlock(&root->accounting_lock);
1742 static void root_sub_used(struct btrfs_root *root, u32 size)
1744 spin_lock(&root->accounting_lock);
1745 btrfs_set_root_used(&root->root_item,
1746 btrfs_root_used(&root->root_item) - size);
1747 spin_unlock(&root->accounting_lock);
1750 /* given a node and slot number, this reads the blocks it points to. The
1751 * extent buffer is returned with a reference taken (but unlocked).
1753 static noinline struct extent_buffer *
1754 read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
1757 int level = btrfs_header_level(parent);
1758 struct extent_buffer *eb;
1759 struct btrfs_key first_key;
1761 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1762 return ERR_PTR(-ENOENT);
1766 btrfs_node_key_to_cpu(parent, &first_key, slot);
1767 eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
1768 btrfs_node_ptr_generation(parent, slot),
1769 level - 1, &first_key);
1770 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1771 free_extent_buffer(eb);
1779 * node level balancing, used to make sure nodes are in proper order for
1780 * item deletion. We balance from the top down, so we have to make sure
1781 * that a deletion won't leave an node completely empty later on.
1783 static noinline int balance_level(struct btrfs_trans_handle *trans,
1784 struct btrfs_root *root,
1785 struct btrfs_path *path, int level)
1787 struct btrfs_fs_info *fs_info = root->fs_info;
1788 struct extent_buffer *right = NULL;
1789 struct extent_buffer *mid;
1790 struct extent_buffer *left = NULL;
1791 struct extent_buffer *parent = NULL;
1795 int orig_slot = path->slots[level];
1800 mid = path->nodes[level];
1802 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1803 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1804 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1806 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1808 if (level < BTRFS_MAX_LEVEL - 1) {
1809 parent = path->nodes[level + 1];
1810 pslot = path->slots[level + 1];
1814 * deal with the case where there is only one pointer in the root
1815 * by promoting the node below to a root
1818 struct extent_buffer *child;
1820 if (btrfs_header_nritems(mid) != 1)
1823 /* promote the child to a root */
1824 child = read_node_slot(fs_info, mid, 0);
1825 if (IS_ERR(child)) {
1826 ret = PTR_ERR(child);
1827 btrfs_handle_fs_error(fs_info, ret, NULL);
1831 btrfs_tree_lock(child);
1832 btrfs_set_lock_blocking(child);
1833 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1835 btrfs_tree_unlock(child);
1836 free_extent_buffer(child);
1840 ret = tree_mod_log_insert_root(root->node, child, 1);
1842 rcu_assign_pointer(root->node, child);
1844 add_root_to_dirty_list(root);
1845 btrfs_tree_unlock(child);
1847 path->locks[level] = 0;
1848 path->nodes[level] = NULL;
1849 clean_tree_block(fs_info, mid);
1850 btrfs_tree_unlock(mid);
1851 /* once for the path */
1852 free_extent_buffer(mid);
1854 root_sub_used(root, mid->len);
1855 btrfs_free_tree_block(trans, root, mid, 0, 1);
1856 /* once for the root ptr */
1857 free_extent_buffer_stale(mid);
1860 if (btrfs_header_nritems(mid) >
1861 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1864 left = read_node_slot(fs_info, parent, pslot - 1);
1869 btrfs_tree_lock(left);
1870 btrfs_set_lock_blocking(left);
1871 wret = btrfs_cow_block(trans, root, left,
1872 parent, pslot - 1, &left);
1879 right = read_node_slot(fs_info, parent, pslot + 1);
1884 btrfs_tree_lock(right);
1885 btrfs_set_lock_blocking(right);
1886 wret = btrfs_cow_block(trans, root, right,
1887 parent, pslot + 1, &right);
1894 /* first, try to make some room in the middle buffer */
1896 orig_slot += btrfs_header_nritems(left);
1897 wret = push_node_left(trans, fs_info, left, mid, 1);
1903 * then try to empty the right most buffer into the middle
1906 wret = push_node_left(trans, fs_info, mid, right, 1);
1907 if (wret < 0 && wret != -ENOSPC)
1909 if (btrfs_header_nritems(right) == 0) {
1910 clean_tree_block(fs_info, right);
1911 btrfs_tree_unlock(right);
1912 del_ptr(root, path, level + 1, pslot + 1);
1913 root_sub_used(root, right->len);
1914 btrfs_free_tree_block(trans, root, right, 0, 1);
1915 free_extent_buffer_stale(right);
1918 struct btrfs_disk_key right_key;
1919 btrfs_node_key(right, &right_key, 0);
1920 ret = tree_mod_log_insert_key(parent, pslot + 1,
1921 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1923 btrfs_set_node_key(parent, &right_key, pslot + 1);
1924 btrfs_mark_buffer_dirty(parent);
1927 if (btrfs_header_nritems(mid) == 1) {
1929 * we're not allowed to leave a node with one item in the
1930 * tree during a delete. A deletion from lower in the tree
1931 * could try to delete the only pointer in this node.
1932 * So, pull some keys from the left.
1933 * There has to be a left pointer at this point because
1934 * otherwise we would have pulled some pointers from the
1939 btrfs_handle_fs_error(fs_info, ret, NULL);
1942 wret = balance_node_right(trans, fs_info, mid, left);
1948 wret = push_node_left(trans, fs_info, left, mid, 1);
1954 if (btrfs_header_nritems(mid) == 0) {
1955 clean_tree_block(fs_info, mid);
1956 btrfs_tree_unlock(mid);
1957 del_ptr(root, path, level + 1, pslot);
1958 root_sub_used(root, mid->len);
1959 btrfs_free_tree_block(trans, root, mid, 0, 1);
1960 free_extent_buffer_stale(mid);
1963 /* update the parent key to reflect our changes */
1964 struct btrfs_disk_key mid_key;
1965 btrfs_node_key(mid, &mid_key, 0);
1966 ret = tree_mod_log_insert_key(parent, pslot,
1967 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1969 btrfs_set_node_key(parent, &mid_key, pslot);
1970 btrfs_mark_buffer_dirty(parent);
1973 /* update the path */
1975 if (btrfs_header_nritems(left) > orig_slot) {
1976 extent_buffer_get(left);
1977 /* left was locked after cow */
1978 path->nodes[level] = left;
1979 path->slots[level + 1] -= 1;
1980 path->slots[level] = orig_slot;
1982 btrfs_tree_unlock(mid);
1983 free_extent_buffer(mid);
1986 orig_slot -= btrfs_header_nritems(left);
1987 path->slots[level] = orig_slot;
1990 /* double check we haven't messed things up */
1992 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1996 btrfs_tree_unlock(right);
1997 free_extent_buffer(right);
2000 if (path->nodes[level] != left)
2001 btrfs_tree_unlock(left);
2002 free_extent_buffer(left);
2007 /* Node balancing for insertion. Here we only split or push nodes around
2008 * when they are completely full. This is also done top down, so we
2009 * have to be pessimistic.
2011 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2012 struct btrfs_root *root,
2013 struct btrfs_path *path, int level)
2015 struct btrfs_fs_info *fs_info = root->fs_info;
2016 struct extent_buffer *right = NULL;
2017 struct extent_buffer *mid;
2018 struct extent_buffer *left = NULL;
2019 struct extent_buffer *parent = NULL;
2023 int orig_slot = path->slots[level];
2028 mid = path->nodes[level];
2029 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2031 if (level < BTRFS_MAX_LEVEL - 1) {
2032 parent = path->nodes[level + 1];
2033 pslot = path->slots[level + 1];
2039 left = read_node_slot(fs_info, parent, pslot - 1);
2043 /* first, try to make some room in the middle buffer */
2047 btrfs_tree_lock(left);
2048 btrfs_set_lock_blocking(left);
2050 left_nr = btrfs_header_nritems(left);
2051 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2054 ret = btrfs_cow_block(trans, root, left, parent,
2059 wret = push_node_left(trans, fs_info,
2066 struct btrfs_disk_key disk_key;
2067 orig_slot += left_nr;
2068 btrfs_node_key(mid, &disk_key, 0);
2069 ret = tree_mod_log_insert_key(parent, pslot,
2070 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2072 btrfs_set_node_key(parent, &disk_key, pslot);
2073 btrfs_mark_buffer_dirty(parent);
2074 if (btrfs_header_nritems(left) > orig_slot) {
2075 path->nodes[level] = left;
2076 path->slots[level + 1] -= 1;
2077 path->slots[level] = orig_slot;
2078 btrfs_tree_unlock(mid);
2079 free_extent_buffer(mid);
2082 btrfs_header_nritems(left);
2083 path->slots[level] = orig_slot;
2084 btrfs_tree_unlock(left);
2085 free_extent_buffer(left);
2089 btrfs_tree_unlock(left);
2090 free_extent_buffer(left);
2092 right = read_node_slot(fs_info, parent, pslot + 1);
2097 * then try to empty the right most buffer into the middle
2102 btrfs_tree_lock(right);
2103 btrfs_set_lock_blocking(right);
2105 right_nr = btrfs_header_nritems(right);
2106 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2109 ret = btrfs_cow_block(trans, root, right,
2115 wret = balance_node_right(trans, fs_info,
2122 struct btrfs_disk_key disk_key;
2124 btrfs_node_key(right, &disk_key, 0);
2125 ret = tree_mod_log_insert_key(parent, pslot + 1,
2126 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2128 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2129 btrfs_mark_buffer_dirty(parent);
2131 if (btrfs_header_nritems(mid) <= orig_slot) {
2132 path->nodes[level] = right;
2133 path->slots[level + 1] += 1;
2134 path->slots[level] = orig_slot -
2135 btrfs_header_nritems(mid);
2136 btrfs_tree_unlock(mid);
2137 free_extent_buffer(mid);
2139 btrfs_tree_unlock(right);
2140 free_extent_buffer(right);
2144 btrfs_tree_unlock(right);
2145 free_extent_buffer(right);
2151 * readahead one full node of leaves, finding things that are close
2152 * to the block in 'slot', and triggering ra on them.
2154 static void reada_for_search(struct btrfs_fs_info *fs_info,
2155 struct btrfs_path *path,
2156 int level, int slot, u64 objectid)
2158 struct extent_buffer *node;
2159 struct btrfs_disk_key disk_key;
2164 struct extent_buffer *eb;
2172 if (!path->nodes[level])
2175 node = path->nodes[level];
2177 search = btrfs_node_blockptr(node, slot);
2178 blocksize = fs_info->nodesize;
2179 eb = find_extent_buffer(fs_info, search);
2181 free_extent_buffer(eb);
2187 nritems = btrfs_header_nritems(node);
2191 if (path->reada == READA_BACK) {
2195 } else if (path->reada == READA_FORWARD) {
2200 if (path->reada == READA_BACK && objectid) {
2201 btrfs_node_key(node, &disk_key, nr);
2202 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2205 search = btrfs_node_blockptr(node, nr);
2206 if ((search <= target && target - search <= 65536) ||
2207 (search > target && search - target <= 65536)) {
2208 readahead_tree_block(fs_info, search);
2212 if ((nread > 65536 || nscan > 32))
2217 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2218 struct btrfs_path *path, int level)
2222 struct extent_buffer *parent;
2223 struct extent_buffer *eb;
2228 parent = path->nodes[level + 1];
2232 nritems = btrfs_header_nritems(parent);
2233 slot = path->slots[level + 1];
2236 block1 = btrfs_node_blockptr(parent, slot - 1);
2237 gen = btrfs_node_ptr_generation(parent, slot - 1);
2238 eb = find_extent_buffer(fs_info, block1);
2240 * if we get -eagain from btrfs_buffer_uptodate, we
2241 * don't want to return eagain here. That will loop
2244 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2246 free_extent_buffer(eb);
2248 if (slot + 1 < nritems) {
2249 block2 = btrfs_node_blockptr(parent, slot + 1);
2250 gen = btrfs_node_ptr_generation(parent, slot + 1);
2251 eb = find_extent_buffer(fs_info, block2);
2252 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2254 free_extent_buffer(eb);
2258 readahead_tree_block(fs_info, block1);
2260 readahead_tree_block(fs_info, block2);
2265 * when we walk down the tree, it is usually safe to unlock the higher layers
2266 * in the tree. The exceptions are when our path goes through slot 0, because
2267 * operations on the tree might require changing key pointers higher up in the
2270 * callers might also have set path->keep_locks, which tells this code to keep
2271 * the lock if the path points to the last slot in the block. This is part of
2272 * walking through the tree, and selecting the next slot in the higher block.
2274 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2275 * if lowest_unlock is 1, level 0 won't be unlocked
2277 static noinline void unlock_up(struct btrfs_path *path, int level,
2278 int lowest_unlock, int min_write_lock_level,
2279 int *write_lock_level)
2282 int skip_level = level;
2284 struct extent_buffer *t;
2286 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2287 if (!path->nodes[i])
2289 if (!path->locks[i])
2291 if (!no_skips && path->slots[i] == 0) {
2295 if (!no_skips && path->keep_locks) {
2298 nritems = btrfs_header_nritems(t);
2299 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2304 if (skip_level < i && i >= lowest_unlock)
2308 if (i >= lowest_unlock && i > skip_level) {
2309 btrfs_tree_unlock_rw(t, path->locks[i]);
2311 if (write_lock_level &&
2312 i > min_write_lock_level &&
2313 i <= *write_lock_level) {
2314 *write_lock_level = i - 1;
2321 * This releases any locks held in the path starting at level and
2322 * going all the way up to the root.
2324 * btrfs_search_slot will keep the lock held on higher nodes in a few
2325 * corner cases, such as COW of the block at slot zero in the node. This
2326 * ignores those rules, and it should only be called when there are no
2327 * more updates to be done higher up in the tree.
2329 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2333 if (path->keep_locks)
2336 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2337 if (!path->nodes[i])
2339 if (!path->locks[i])
2341 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2347 * helper function for btrfs_search_slot. The goal is to find a block
2348 * in cache without setting the path to blocking. If we find the block
2349 * we return zero and the path is unchanged.
2351 * If we can't find the block, we set the path blocking and do some
2352 * reada. -EAGAIN is returned and the search must be repeated.
2355 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2356 struct extent_buffer **eb_ret, int level, int slot,
2357 const struct btrfs_key *key)
2359 struct btrfs_fs_info *fs_info = root->fs_info;
2362 struct extent_buffer *b = *eb_ret;
2363 struct extent_buffer *tmp;
2364 struct btrfs_key first_key;
2368 blocknr = btrfs_node_blockptr(b, slot);
2369 gen = btrfs_node_ptr_generation(b, slot);
2370 parent_level = btrfs_header_level(b);
2371 btrfs_node_key_to_cpu(b, &first_key, slot);
2373 tmp = find_extent_buffer(fs_info, blocknr);
2375 /* first we do an atomic uptodate check */
2376 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2381 /* the pages were up to date, but we failed
2382 * the generation number check. Do a full
2383 * read for the generation number that is correct.
2384 * We must do this without dropping locks so
2385 * we can trust our generation number
2387 btrfs_set_path_blocking(p);
2389 /* now we're allowed to do a blocking uptodate check */
2390 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2395 free_extent_buffer(tmp);
2396 btrfs_release_path(p);
2401 * reduce lock contention at high levels
2402 * of the btree by dropping locks before
2403 * we read. Don't release the lock on the current
2404 * level because we need to walk this node to figure
2405 * out which blocks to read.
2407 btrfs_unlock_up_safe(p, level + 1);
2408 btrfs_set_path_blocking(p);
2410 if (p->reada != READA_NONE)
2411 reada_for_search(fs_info, p, level, slot, key->objectid);
2414 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2418 * If the read above didn't mark this buffer up to date,
2419 * it will never end up being up to date. Set ret to EIO now
2420 * and give up so that our caller doesn't loop forever
2423 if (!extent_buffer_uptodate(tmp))
2425 free_extent_buffer(tmp);
2430 btrfs_release_path(p);
2435 * helper function for btrfs_search_slot. This does all of the checks
2436 * for node-level blocks and does any balancing required based on
2439 * If no extra work was required, zero is returned. If we had to
2440 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2444 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2445 struct btrfs_root *root, struct btrfs_path *p,
2446 struct extent_buffer *b, int level, int ins_len,
2447 int *write_lock_level)
2449 struct btrfs_fs_info *fs_info = root->fs_info;
2452 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2453 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2456 if (*write_lock_level < level + 1) {
2457 *write_lock_level = level + 1;
2458 btrfs_release_path(p);
2462 btrfs_set_path_blocking(p);
2463 reada_for_balance(fs_info, p, level);
2464 sret = split_node(trans, root, p, level);
2471 b = p->nodes[level];
2472 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2473 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2476 if (*write_lock_level < level + 1) {
2477 *write_lock_level = level + 1;
2478 btrfs_release_path(p);
2482 btrfs_set_path_blocking(p);
2483 reada_for_balance(fs_info, p, level);
2484 sret = balance_level(trans, root, p, level);
2490 b = p->nodes[level];
2492 btrfs_release_path(p);
2495 BUG_ON(btrfs_header_nritems(b) == 1);
2505 static void key_search_validate(struct extent_buffer *b,
2506 const struct btrfs_key *key,
2509 #ifdef CONFIG_BTRFS_ASSERT
2510 struct btrfs_disk_key disk_key;
2512 btrfs_cpu_key_to_disk(&disk_key, key);
2515 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2516 offsetof(struct btrfs_leaf, items[0].key),
2519 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2520 offsetof(struct btrfs_node, ptrs[0].key),
2525 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2526 int level, int *prev_cmp, int *slot)
2528 if (*prev_cmp != 0) {
2529 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2533 key_search_validate(b, key, level);
2539 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2540 u64 iobjectid, u64 ioff, u8 key_type,
2541 struct btrfs_key *found_key)
2544 struct btrfs_key key;
2545 struct extent_buffer *eb;
2550 key.type = key_type;
2551 key.objectid = iobjectid;
2554 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2558 eb = path->nodes[0];
2559 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2560 ret = btrfs_next_leaf(fs_root, path);
2563 eb = path->nodes[0];
2566 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2567 if (found_key->type != key.type ||
2568 found_key->objectid != key.objectid)
2574 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2575 struct btrfs_path *p,
2576 int write_lock_level)
2578 struct btrfs_fs_info *fs_info = root->fs_info;
2579 struct extent_buffer *b;
2583 /* We try very hard to do read locks on the root */
2584 root_lock = BTRFS_READ_LOCK;
2586 if (p->search_commit_root) {
2587 /* The commit roots are read only so we always do read locks */
2588 if (p->need_commit_sem)
2589 down_read(&fs_info->commit_root_sem);
2590 b = root->commit_root;
2591 extent_buffer_get(b);
2592 level = btrfs_header_level(b);
2593 if (p->need_commit_sem)
2594 up_read(&fs_info->commit_root_sem);
2596 * Ensure that all callers have set skip_locking when
2597 * p->search_commit_root = 1.
2599 ASSERT(p->skip_locking == 1);
2604 if (p->skip_locking) {
2605 b = btrfs_root_node(root);
2606 level = btrfs_header_level(b);
2611 * If the level is set to maximum, we can skip trying to get the read
2614 if (write_lock_level < BTRFS_MAX_LEVEL) {
2616 * We don't know the level of the root node until we actually
2617 * have it read locked
2619 b = btrfs_read_lock_root_node(root);
2620 level = btrfs_header_level(b);
2621 if (level > write_lock_level)
2624 /* Whoops, must trade for write lock */
2625 btrfs_tree_read_unlock(b);
2626 free_extent_buffer(b);
2629 b = btrfs_lock_root_node(root);
2630 root_lock = BTRFS_WRITE_LOCK;
2632 /* The level might have changed, check again */
2633 level = btrfs_header_level(b);
2636 p->nodes[level] = b;
2637 if (!p->skip_locking)
2638 p->locks[level] = root_lock;
2640 * Callers are responsible for dropping b's references.
2647 * btrfs_search_slot - look for a key in a tree and perform necessary
2648 * modifications to preserve tree invariants.
2650 * @trans: Handle of transaction, used when modifying the tree
2651 * @p: Holds all btree nodes along the search path
2652 * @root: The root node of the tree
2653 * @key: The key we are looking for
2654 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2655 * deletions it's -1. 0 for plain searches
2656 * @cow: boolean should CoW operations be performed. Must always be 1
2657 * when modifying the tree.
2659 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2660 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2662 * If @key is found, 0 is returned and you can find the item in the leaf level
2663 * of the path (level 0)
2665 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2666 * points to the slot where it should be inserted
2668 * If an error is encountered while searching the tree a negative error number
2671 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2672 const struct btrfs_key *key, struct btrfs_path *p,
2673 int ins_len, int cow)
2675 struct btrfs_fs_info *fs_info = root->fs_info;
2676 struct extent_buffer *b;
2681 int lowest_unlock = 1;
2682 /* everything at write_lock_level or lower must be write locked */
2683 int write_lock_level = 0;
2684 u8 lowest_level = 0;
2685 int min_write_lock_level;
2688 lowest_level = p->lowest_level;
2689 WARN_ON(lowest_level && ins_len > 0);
2690 WARN_ON(p->nodes[0] != NULL);
2691 BUG_ON(!cow && ins_len);
2696 /* when we are removing items, we might have to go up to level
2697 * two as we update tree pointers Make sure we keep write
2698 * for those levels as well
2700 write_lock_level = 2;
2701 } else if (ins_len > 0) {
2703 * for inserting items, make sure we have a write lock on
2704 * level 1 so we can update keys
2706 write_lock_level = 1;
2710 write_lock_level = -1;
2712 if (cow && (p->keep_locks || p->lowest_level))
2713 write_lock_level = BTRFS_MAX_LEVEL;
2715 min_write_lock_level = write_lock_level;
2719 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2722 level = btrfs_header_level(b);
2725 * setup the path here so we can release it under lock
2726 * contention with the cow code
2729 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2732 * if we don't really need to cow this block
2733 * then we don't want to set the path blocking,
2734 * so we test it here
2736 if (!should_cow_block(trans, root, b)) {
2737 trans->dirty = true;
2742 * must have write locks on this node and the
2745 if (level > write_lock_level ||
2746 (level + 1 > write_lock_level &&
2747 level + 1 < BTRFS_MAX_LEVEL &&
2748 p->nodes[level + 1])) {
2749 write_lock_level = level + 1;
2750 btrfs_release_path(p);
2754 btrfs_set_path_blocking(p);
2756 err = btrfs_cow_block(trans, root, b, NULL, 0,
2759 err = btrfs_cow_block(trans, root, b,
2760 p->nodes[level + 1],
2761 p->slots[level + 1], &b);
2768 p->nodes[level] = b;
2770 * Leave path with blocking locks to avoid massive
2771 * lock context switch, this is made on purpose.
2775 * we have a lock on b and as long as we aren't changing
2776 * the tree, there is no way to for the items in b to change.
2777 * It is safe to drop the lock on our parent before we
2778 * go through the expensive btree search on b.
2780 * If we're inserting or deleting (ins_len != 0), then we might
2781 * be changing slot zero, which may require changing the parent.
2782 * So, we can't drop the lock until after we know which slot
2783 * we're operating on.
2785 if (!ins_len && !p->keep_locks) {
2788 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2789 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2794 ret = key_search(b, key, level, &prev_cmp, &slot);
2800 if (ret && slot > 0) {
2804 p->slots[level] = slot;
2805 err = setup_nodes_for_search(trans, root, p, b, level,
2806 ins_len, &write_lock_level);
2813 b = p->nodes[level];
2814 slot = p->slots[level];
2817 * slot 0 is special, if we change the key
2818 * we have to update the parent pointer
2819 * which means we must have a write lock
2822 if (slot == 0 && ins_len &&
2823 write_lock_level < level + 1) {
2824 write_lock_level = level + 1;
2825 btrfs_release_path(p);
2829 unlock_up(p, level, lowest_unlock,
2830 min_write_lock_level, &write_lock_level);
2832 if (level == lowest_level) {
2838 err = read_block_for_search(root, p, &b, level,
2847 if (!p->skip_locking) {
2848 level = btrfs_header_level(b);
2849 if (level <= write_lock_level) {
2850 err = btrfs_try_tree_write_lock(b);
2852 btrfs_set_path_blocking(p);
2855 p->locks[level] = BTRFS_WRITE_LOCK;
2857 err = btrfs_tree_read_lock_atomic(b);
2859 btrfs_set_path_blocking(p);
2860 btrfs_tree_read_lock(b);
2862 p->locks[level] = BTRFS_READ_LOCK;
2864 p->nodes[level] = b;
2867 p->slots[level] = slot;
2869 btrfs_leaf_free_space(fs_info, b) < ins_len) {
2870 if (write_lock_level < 1) {
2871 write_lock_level = 1;
2872 btrfs_release_path(p);
2876 btrfs_set_path_blocking(p);
2877 err = split_leaf(trans, root, key,
2878 p, ins_len, ret == 0);
2886 if (!p->search_for_split)
2887 unlock_up(p, level, lowest_unlock,
2888 min_write_lock_level, NULL);
2895 * we don't really know what they plan on doing with the path
2896 * from here on, so for now just mark it as blocking
2898 if (!p->leave_spinning)
2899 btrfs_set_path_blocking(p);
2900 if (ret < 0 && !p->skip_release_on_error)
2901 btrfs_release_path(p);
2906 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2907 * current state of the tree together with the operations recorded in the tree
2908 * modification log to search for the key in a previous version of this tree, as
2909 * denoted by the time_seq parameter.
2911 * Naturally, there is no support for insert, delete or cow operations.
2913 * The resulting path and return value will be set up as if we called
2914 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2916 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2917 struct btrfs_path *p, u64 time_seq)
2919 struct btrfs_fs_info *fs_info = root->fs_info;
2920 struct extent_buffer *b;
2925 int lowest_unlock = 1;
2926 u8 lowest_level = 0;
2929 lowest_level = p->lowest_level;
2930 WARN_ON(p->nodes[0] != NULL);
2932 if (p->search_commit_root) {
2934 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2938 b = get_old_root(root, time_seq);
2943 level = btrfs_header_level(b);
2944 p->locks[level] = BTRFS_READ_LOCK;
2947 level = btrfs_header_level(b);
2948 p->nodes[level] = b;
2951 * we have a lock on b and as long as we aren't changing
2952 * the tree, there is no way to for the items in b to change.
2953 * It is safe to drop the lock on our parent before we
2954 * go through the expensive btree search on b.
2956 btrfs_unlock_up_safe(p, level + 1);
2959 * Since we can unwind ebs we want to do a real search every
2963 ret = key_search(b, key, level, &prev_cmp, &slot);
2967 if (ret && slot > 0) {
2971 p->slots[level] = slot;
2972 unlock_up(p, level, lowest_unlock, 0, NULL);
2974 if (level == lowest_level) {
2980 err = read_block_for_search(root, p, &b, level,
2989 level = btrfs_header_level(b);
2990 err = btrfs_tree_read_lock_atomic(b);
2992 btrfs_set_path_blocking(p);
2993 btrfs_tree_read_lock(b);
2995 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3000 p->locks[level] = BTRFS_READ_LOCK;
3001 p->nodes[level] = b;
3003 p->slots[level] = slot;
3004 unlock_up(p, level, lowest_unlock, 0, NULL);
3010 if (!p->leave_spinning)
3011 btrfs_set_path_blocking(p);
3013 btrfs_release_path(p);
3019 * helper to use instead of search slot if no exact match is needed but
3020 * instead the next or previous item should be returned.
3021 * When find_higher is true, the next higher item is returned, the next lower
3023 * When return_any and find_higher are both true, and no higher item is found,
3024 * return the next lower instead.
3025 * When return_any is true and find_higher is false, and no lower item is found,
3026 * return the next higher instead.
3027 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3030 int btrfs_search_slot_for_read(struct btrfs_root *root,
3031 const struct btrfs_key *key,
3032 struct btrfs_path *p, int find_higher,
3036 struct extent_buffer *leaf;
3039 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3043 * a return value of 1 means the path is at the position where the
3044 * item should be inserted. Normally this is the next bigger item,
3045 * but in case the previous item is the last in a leaf, path points
3046 * to the first free slot in the previous leaf, i.e. at an invalid
3052 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3053 ret = btrfs_next_leaf(root, p);
3059 * no higher item found, return the next
3064 btrfs_release_path(p);
3068 if (p->slots[0] == 0) {
3069 ret = btrfs_prev_leaf(root, p);
3074 if (p->slots[0] == btrfs_header_nritems(leaf))
3081 * no lower item found, return the next
3086 btrfs_release_path(p);
3096 * adjust the pointers going up the tree, starting at level
3097 * making sure the right key of each node is points to 'key'.
3098 * This is used after shifting pointers to the left, so it stops
3099 * fixing up pointers when a given leaf/node is not in slot 0 of the
3103 static void fixup_low_keys(struct btrfs_path *path,
3104 struct btrfs_disk_key *key, int level)
3107 struct extent_buffer *t;
3110 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3111 int tslot = path->slots[i];
3113 if (!path->nodes[i])
3116 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3119 btrfs_set_node_key(t, key, tslot);
3120 btrfs_mark_buffer_dirty(path->nodes[i]);
3129 * This function isn't completely safe. It's the caller's responsibility
3130 * that the new key won't break the order
3132 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3133 struct btrfs_path *path,
3134 const struct btrfs_key *new_key)
3136 struct btrfs_disk_key disk_key;
3137 struct extent_buffer *eb;
3140 eb = path->nodes[0];
3141 slot = path->slots[0];
3143 btrfs_item_key(eb, &disk_key, slot - 1);
3144 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3146 if (slot < btrfs_header_nritems(eb) - 1) {
3147 btrfs_item_key(eb, &disk_key, slot + 1);
3148 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3151 btrfs_cpu_key_to_disk(&disk_key, new_key);
3152 btrfs_set_item_key(eb, &disk_key, slot);
3153 btrfs_mark_buffer_dirty(eb);
3155 fixup_low_keys(path, &disk_key, 1);
3159 * try to push data from one node into the next node left in the
3162 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3163 * error, and > 0 if there was no room in the left hand block.
3165 static int push_node_left(struct btrfs_trans_handle *trans,
3166 struct btrfs_fs_info *fs_info,
3167 struct extent_buffer *dst,
3168 struct extent_buffer *src, int empty)
3175 src_nritems = btrfs_header_nritems(src);
3176 dst_nritems = btrfs_header_nritems(dst);
3177 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3178 WARN_ON(btrfs_header_generation(src) != trans->transid);
3179 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3181 if (!empty && src_nritems <= 8)
3184 if (push_items <= 0)
3188 push_items = min(src_nritems, push_items);
3189 if (push_items < src_nritems) {
3190 /* leave at least 8 pointers in the node if
3191 * we aren't going to empty it
3193 if (src_nritems - push_items < 8) {
3194 if (push_items <= 8)
3200 push_items = min(src_nritems - 8, push_items);
3202 ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
3205 btrfs_abort_transaction(trans, ret);
3208 copy_extent_buffer(dst, src,
3209 btrfs_node_key_ptr_offset(dst_nritems),
3210 btrfs_node_key_ptr_offset(0),
3211 push_items * sizeof(struct btrfs_key_ptr));
3213 if (push_items < src_nritems) {
3215 * Don't call tree_mod_log_insert_move here, key removal was
3216 * already fully logged by tree_mod_log_eb_copy above.
3218 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3219 btrfs_node_key_ptr_offset(push_items),
3220 (src_nritems - push_items) *
3221 sizeof(struct btrfs_key_ptr));
3223 btrfs_set_header_nritems(src, src_nritems - push_items);
3224 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3225 btrfs_mark_buffer_dirty(src);
3226 btrfs_mark_buffer_dirty(dst);
3232 * try to push data from one node into the next node right in the
3235 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3236 * error, and > 0 if there was no room in the right hand block.
3238 * this will only push up to 1/2 the contents of the left node over
3240 static int balance_node_right(struct btrfs_trans_handle *trans,
3241 struct btrfs_fs_info *fs_info,
3242 struct extent_buffer *dst,
3243 struct extent_buffer *src)
3251 WARN_ON(btrfs_header_generation(src) != trans->transid);
3252 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3254 src_nritems = btrfs_header_nritems(src);
3255 dst_nritems = btrfs_header_nritems(dst);
3256 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3257 if (push_items <= 0)
3260 if (src_nritems < 4)
3263 max_push = src_nritems / 2 + 1;
3264 /* don't try to empty the node */
3265 if (max_push >= src_nritems)
3268 if (max_push < push_items)
3269 push_items = max_push;
3271 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3273 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3274 btrfs_node_key_ptr_offset(0),
3276 sizeof(struct btrfs_key_ptr));
3278 ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
3279 src_nritems - push_items, push_items);
3281 btrfs_abort_transaction(trans, ret);
3284 copy_extent_buffer(dst, src,
3285 btrfs_node_key_ptr_offset(0),
3286 btrfs_node_key_ptr_offset(src_nritems - push_items),
3287 push_items * sizeof(struct btrfs_key_ptr));
3289 btrfs_set_header_nritems(src, src_nritems - push_items);
3290 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3292 btrfs_mark_buffer_dirty(src);
3293 btrfs_mark_buffer_dirty(dst);
3299 * helper function to insert a new root level in the tree.
3300 * A new node is allocated, and a single item is inserted to
3301 * point to the existing root
3303 * returns zero on success or < 0 on failure.
3305 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3306 struct btrfs_root *root,
3307 struct btrfs_path *path, int level)
3309 struct btrfs_fs_info *fs_info = root->fs_info;
3311 struct extent_buffer *lower;
3312 struct extent_buffer *c;
3313 struct extent_buffer *old;
3314 struct btrfs_disk_key lower_key;
3317 BUG_ON(path->nodes[level]);
3318 BUG_ON(path->nodes[level-1] != root->node);
3320 lower = path->nodes[level-1];
3322 btrfs_item_key(lower, &lower_key, 0);
3324 btrfs_node_key(lower, &lower_key, 0);
3326 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3327 &lower_key, level, root->node->start, 0);
3331 root_add_used(root, fs_info->nodesize);
3333 btrfs_set_header_nritems(c, 1);
3334 btrfs_set_node_key(c, &lower_key, 0);
3335 btrfs_set_node_blockptr(c, 0, lower->start);
3336 lower_gen = btrfs_header_generation(lower);
3337 WARN_ON(lower_gen != trans->transid);
3339 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3341 btrfs_mark_buffer_dirty(c);
3344 ret = tree_mod_log_insert_root(root->node, c, 0);
3346 rcu_assign_pointer(root->node, c);
3348 /* the super has an extra ref to root->node */
3349 free_extent_buffer(old);
3351 add_root_to_dirty_list(root);
3352 extent_buffer_get(c);
3353 path->nodes[level] = c;
3354 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3355 path->slots[level] = 0;
3360 * worker function to insert a single pointer in a node.
3361 * the node should have enough room for the pointer already
3363 * slot and level indicate where you want the key to go, and
3364 * blocknr is the block the key points to.
3366 static void insert_ptr(struct btrfs_trans_handle *trans,
3367 struct btrfs_fs_info *fs_info, struct btrfs_path *path,
3368 struct btrfs_disk_key *key, u64 bytenr,
3369 int slot, int level)
3371 struct extent_buffer *lower;
3375 BUG_ON(!path->nodes[level]);
3376 btrfs_assert_tree_locked(path->nodes[level]);
3377 lower = path->nodes[level];
3378 nritems = btrfs_header_nritems(lower);
3379 BUG_ON(slot > nritems);
3380 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info));
3381 if (slot != nritems) {
3383 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3387 memmove_extent_buffer(lower,
3388 btrfs_node_key_ptr_offset(slot + 1),
3389 btrfs_node_key_ptr_offset(slot),
3390 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3393 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3397 btrfs_set_node_key(lower, key, slot);
3398 btrfs_set_node_blockptr(lower, slot, bytenr);
3399 WARN_ON(trans->transid == 0);
3400 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3401 btrfs_set_header_nritems(lower, nritems + 1);
3402 btrfs_mark_buffer_dirty(lower);
3406 * split the node at the specified level in path in two.
3407 * The path is corrected to point to the appropriate node after the split
3409 * Before splitting this tries to make some room in the node by pushing
3410 * left and right, if either one works, it returns right away.
3412 * returns 0 on success and < 0 on failure
3414 static noinline int split_node(struct btrfs_trans_handle *trans,
3415 struct btrfs_root *root,
3416 struct btrfs_path *path, int level)
3418 struct btrfs_fs_info *fs_info = root->fs_info;
3419 struct extent_buffer *c;
3420 struct extent_buffer *split;
3421 struct btrfs_disk_key disk_key;
3426 c = path->nodes[level];
3427 WARN_ON(btrfs_header_generation(c) != trans->transid);
3428 if (c == root->node) {
3430 * trying to split the root, lets make a new one
3432 * tree mod log: We don't log_removal old root in
3433 * insert_new_root, because that root buffer will be kept as a
3434 * normal node. We are going to log removal of half of the
3435 * elements below with tree_mod_log_eb_copy. We're holding a
3436 * tree lock on the buffer, which is why we cannot race with
3437 * other tree_mod_log users.
3439 ret = insert_new_root(trans, root, path, level + 1);
3443 ret = push_nodes_for_insert(trans, root, path, level);
3444 c = path->nodes[level];
3445 if (!ret && btrfs_header_nritems(c) <
3446 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3452 c_nritems = btrfs_header_nritems(c);
3453 mid = (c_nritems + 1) / 2;
3454 btrfs_node_key(c, &disk_key, mid);
3456 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3457 &disk_key, level, c->start, 0);
3459 return PTR_ERR(split);
3461 root_add_used(root, fs_info->nodesize);
3462 ASSERT(btrfs_header_level(c) == level);
3464 ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
3466 btrfs_abort_transaction(trans, ret);
3469 copy_extent_buffer(split, c,
3470 btrfs_node_key_ptr_offset(0),
3471 btrfs_node_key_ptr_offset(mid),
3472 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3473 btrfs_set_header_nritems(split, c_nritems - mid);
3474 btrfs_set_header_nritems(c, mid);
3477 btrfs_mark_buffer_dirty(c);
3478 btrfs_mark_buffer_dirty(split);
3480 insert_ptr(trans, fs_info, path, &disk_key, split->start,
3481 path->slots[level + 1] + 1, level + 1);
3483 if (path->slots[level] >= mid) {
3484 path->slots[level] -= mid;
3485 btrfs_tree_unlock(c);
3486 free_extent_buffer(c);
3487 path->nodes[level] = split;
3488 path->slots[level + 1] += 1;
3490 btrfs_tree_unlock(split);
3491 free_extent_buffer(split);
3497 * how many bytes are required to store the items in a leaf. start
3498 * and nr indicate which items in the leaf to check. This totals up the
3499 * space used both by the item structs and the item data
3501 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3503 struct btrfs_item *start_item;
3504 struct btrfs_item *end_item;
3505 struct btrfs_map_token token;
3507 int nritems = btrfs_header_nritems(l);
3508 int end = min(nritems, start + nr) - 1;
3512 btrfs_init_map_token(&token);
3513 start_item = btrfs_item_nr(start);
3514 end_item = btrfs_item_nr(end);
3515 data_len = btrfs_token_item_offset(l, start_item, &token) +
3516 btrfs_token_item_size(l, start_item, &token);
3517 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3518 data_len += sizeof(struct btrfs_item) * nr;
3519 WARN_ON(data_len < 0);
3524 * The space between the end of the leaf items and
3525 * the start of the leaf data. IOW, how much room
3526 * the leaf has left for both items and data
3528 noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
3529 struct extent_buffer *leaf)
3531 int nritems = btrfs_header_nritems(leaf);
3534 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3537 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3539 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3540 leaf_space_used(leaf, 0, nritems), nritems);
3546 * min slot controls the lowest index we're willing to push to the
3547 * right. We'll push up to and including min_slot, but no lower
3549 static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info,
3550 struct btrfs_path *path,
3551 int data_size, int empty,
3552 struct extent_buffer *right,
3553 int free_space, u32 left_nritems,
3556 struct extent_buffer *left = path->nodes[0];
3557 struct extent_buffer *upper = path->nodes[1];
3558 struct btrfs_map_token token;
3559 struct btrfs_disk_key disk_key;
3564 struct btrfs_item *item;
3570 btrfs_init_map_token(&token);
3575 nr = max_t(u32, 1, min_slot);
3577 if (path->slots[0] >= left_nritems)
3578 push_space += data_size;
3580 slot = path->slots[1];
3581 i = left_nritems - 1;
3583 item = btrfs_item_nr(i);
3585 if (!empty && push_items > 0) {
3586 if (path->slots[0] > i)
3588 if (path->slots[0] == i) {
3589 int space = btrfs_leaf_free_space(fs_info, left);
3590 if (space + push_space * 2 > free_space)
3595 if (path->slots[0] == i)
3596 push_space += data_size;
3598 this_item_size = btrfs_item_size(left, item);
3599 if (this_item_size + sizeof(*item) + push_space > free_space)
3603 push_space += this_item_size + sizeof(*item);
3609 if (push_items == 0)
3612 WARN_ON(!empty && push_items == left_nritems);
3614 /* push left to right */
3615 right_nritems = btrfs_header_nritems(right);
3617 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3618 push_space -= leaf_data_end(fs_info, left);
3620 /* make room in the right data area */
3621 data_end = leaf_data_end(fs_info, right);
3622 memmove_extent_buffer(right,
3623 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3624 BTRFS_LEAF_DATA_OFFSET + data_end,
3625 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3627 /* copy from the left data area */
3628 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3629 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3630 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left),
3633 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3634 btrfs_item_nr_offset(0),
3635 right_nritems * sizeof(struct btrfs_item));
3637 /* copy the items from left to right */
3638 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3639 btrfs_item_nr_offset(left_nritems - push_items),
3640 push_items * sizeof(struct btrfs_item));
3642 /* update the item pointers */
3643 right_nritems += push_items;
3644 btrfs_set_header_nritems(right, right_nritems);
3645 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3646 for (i = 0; i < right_nritems; i++) {
3647 item = btrfs_item_nr(i);
3648 push_space -= btrfs_token_item_size(right, item, &token);
3649 btrfs_set_token_item_offset(right, item, push_space, &token);
3652 left_nritems -= push_items;
3653 btrfs_set_header_nritems(left, left_nritems);
3656 btrfs_mark_buffer_dirty(left);
3658 clean_tree_block(fs_info, left);
3660 btrfs_mark_buffer_dirty(right);
3662 btrfs_item_key(right, &disk_key, 0);
3663 btrfs_set_node_key(upper, &disk_key, slot + 1);
3664 btrfs_mark_buffer_dirty(upper);
3666 /* then fixup the leaf pointer in the path */
3667 if (path->slots[0] >= left_nritems) {
3668 path->slots[0] -= left_nritems;
3669 if (btrfs_header_nritems(path->nodes[0]) == 0)
3670 clean_tree_block(fs_info, path->nodes[0]);
3671 btrfs_tree_unlock(path->nodes[0]);
3672 free_extent_buffer(path->nodes[0]);
3673 path->nodes[0] = right;
3674 path->slots[1] += 1;
3676 btrfs_tree_unlock(right);
3677 free_extent_buffer(right);
3682 btrfs_tree_unlock(right);
3683 free_extent_buffer(right);
3688 * push some data in the path leaf to the right, trying to free up at
3689 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3691 * returns 1 if the push failed because the other node didn't have enough
3692 * room, 0 if everything worked out and < 0 if there were major errors.
3694 * this will push starting from min_slot to the end of the leaf. It won't
3695 * push any slot lower than min_slot
3697 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3698 *root, struct btrfs_path *path,
3699 int min_data_size, int data_size,
3700 int empty, u32 min_slot)
3702 struct btrfs_fs_info *fs_info = root->fs_info;
3703 struct extent_buffer *left = path->nodes[0];
3704 struct extent_buffer *right;
3705 struct extent_buffer *upper;
3711 if (!path->nodes[1])
3714 slot = path->slots[1];
3715 upper = path->nodes[1];
3716 if (slot >= btrfs_header_nritems(upper) - 1)
3719 btrfs_assert_tree_locked(path->nodes[1]);
3721 right = read_node_slot(fs_info, upper, slot + 1);
3723 * slot + 1 is not valid or we fail to read the right node,
3724 * no big deal, just return.
3729 btrfs_tree_lock(right);
3730 btrfs_set_lock_blocking(right);
3732 free_space = btrfs_leaf_free_space(fs_info, right);
3733 if (free_space < data_size)
3736 /* cow and double check */
3737 ret = btrfs_cow_block(trans, root, right, upper,
3742 free_space = btrfs_leaf_free_space(fs_info, right);
3743 if (free_space < data_size)
3746 left_nritems = btrfs_header_nritems(left);
3747 if (left_nritems == 0)
3750 if (path->slots[0] == left_nritems && !empty) {
3751 /* Key greater than all keys in the leaf, right neighbor has
3752 * enough room for it and we're not emptying our leaf to delete
3753 * it, therefore use right neighbor to insert the new item and
3754 * no need to touch/dirty our left leaft. */
3755 btrfs_tree_unlock(left);
3756 free_extent_buffer(left);
3757 path->nodes[0] = right;
3763 return __push_leaf_right(fs_info, path, min_data_size, empty,
3764 right, free_space, left_nritems, min_slot);
3766 btrfs_tree_unlock(right);
3767 free_extent_buffer(right);
3772 * push some data in the path leaf to the left, trying to free up at
3773 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3775 * max_slot can put a limit on how far into the leaf we'll push items. The
3776 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3779 static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info,
3780 struct btrfs_path *path, int data_size,
3781 int empty, struct extent_buffer *left,
3782 int free_space, u32 right_nritems,
3785 struct btrfs_disk_key disk_key;
3786 struct extent_buffer *right = path->nodes[0];
3790 struct btrfs_item *item;
3791 u32 old_left_nritems;
3795 u32 old_left_item_size;
3796 struct btrfs_map_token token;
3798 btrfs_init_map_token(&token);
3801 nr = min(right_nritems, max_slot);
3803 nr = min(right_nritems - 1, max_slot);
3805 for (i = 0; i < nr; i++) {
3806 item = btrfs_item_nr(i);
3808 if (!empty && push_items > 0) {
3809 if (path->slots[0] < i)
3811 if (path->slots[0] == i) {
3812 int space = btrfs_leaf_free_space(fs_info, right);
3813 if (space + push_space * 2 > free_space)
3818 if (path->slots[0] == i)
3819 push_space += data_size;
3821 this_item_size = btrfs_item_size(right, item);
3822 if (this_item_size + sizeof(*item) + push_space > free_space)
3826 push_space += this_item_size + sizeof(*item);
3829 if (push_items == 0) {
3833 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3835 /* push data from right to left */
3836 copy_extent_buffer(left, right,
3837 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3838 btrfs_item_nr_offset(0),
3839 push_items * sizeof(struct btrfs_item));
3841 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3842 btrfs_item_offset_nr(right, push_items - 1);
3844 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3845 leaf_data_end(fs_info, left) - push_space,
3846 BTRFS_LEAF_DATA_OFFSET +
3847 btrfs_item_offset_nr(right, push_items - 1),
3849 old_left_nritems = btrfs_header_nritems(left);
3850 BUG_ON(old_left_nritems <= 0);
3852 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3853 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3856 item = btrfs_item_nr(i);
3858 ioff = btrfs_token_item_offset(left, item, &token);
3859 btrfs_set_token_item_offset(left, item,
3860 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3863 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3865 /* fixup right node */
3866 if (push_items > right_nritems)
3867 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3870 if (push_items < right_nritems) {
3871 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3872 leaf_data_end(fs_info, right);
3873 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3874 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3875 BTRFS_LEAF_DATA_OFFSET +
3876 leaf_data_end(fs_info, right), push_space);
3878 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3879 btrfs_item_nr_offset(push_items),
3880 (btrfs_header_nritems(right) - push_items) *
3881 sizeof(struct btrfs_item));
3883 right_nritems -= push_items;
3884 btrfs_set_header_nritems(right, right_nritems);
3885 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3886 for (i = 0; i < right_nritems; i++) {
3887 item = btrfs_item_nr(i);
3889 push_space = push_space - btrfs_token_item_size(right,
3891 btrfs_set_token_item_offset(right, item, push_space, &token);
3894 btrfs_mark_buffer_dirty(left);
3896 btrfs_mark_buffer_dirty(right);
3898 clean_tree_block(fs_info, right);
3900 btrfs_item_key(right, &disk_key, 0);
3901 fixup_low_keys(path, &disk_key, 1);
3903 /* then fixup the leaf pointer in the path */
3904 if (path->slots[0] < push_items) {
3905 path->slots[0] += old_left_nritems;
3906 btrfs_tree_unlock(path->nodes[0]);
3907 free_extent_buffer(path->nodes[0]);
3908 path->nodes[0] = left;
3909 path->slots[1] -= 1;
3911 btrfs_tree_unlock(left);
3912 free_extent_buffer(left);
3913 path->slots[0] -= push_items;
3915 BUG_ON(path->slots[0] < 0);
3918 btrfs_tree_unlock(left);
3919 free_extent_buffer(left);
3924 * push some data in the path leaf to the left, trying to free up at
3925 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3927 * max_slot can put a limit on how far into the leaf we'll push items. The
3928 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3931 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3932 *root, struct btrfs_path *path, int min_data_size,
3933 int data_size, int empty, u32 max_slot)
3935 struct btrfs_fs_info *fs_info = root->fs_info;
3936 struct extent_buffer *right = path->nodes[0];
3937 struct extent_buffer *left;
3943 slot = path->slots[1];
3946 if (!path->nodes[1])
3949 right_nritems = btrfs_header_nritems(right);
3950 if (right_nritems == 0)
3953 btrfs_assert_tree_locked(path->nodes[1]);
3955 left = read_node_slot(fs_info, path->nodes[1], slot - 1);
3957 * slot - 1 is not valid or we fail to read the left node,
3958 * no big deal, just return.
3963 btrfs_tree_lock(left);
3964 btrfs_set_lock_blocking(left);
3966 free_space = btrfs_leaf_free_space(fs_info, left);
3967 if (free_space < data_size) {
3972 /* cow and double check */
3973 ret = btrfs_cow_block(trans, root, left,
3974 path->nodes[1], slot - 1, &left);
3976 /* we hit -ENOSPC, but it isn't fatal here */
3982 free_space = btrfs_leaf_free_space(fs_info, left);
3983 if (free_space < data_size) {
3988 return __push_leaf_left(fs_info, path, min_data_size,
3989 empty, left, free_space, right_nritems,
3992 btrfs_tree_unlock(left);
3993 free_extent_buffer(left);
3998 * split the path's leaf in two, making sure there is at least data_size
3999 * available for the resulting leaf level of the path.
4001 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4002 struct btrfs_fs_info *fs_info,
4003 struct btrfs_path *path,
4004 struct extent_buffer *l,
4005 struct extent_buffer *right,
4006 int slot, int mid, int nritems)
4011 struct btrfs_disk_key disk_key;
4012 struct btrfs_map_token token;
4014 btrfs_init_map_token(&token);
4016 nritems = nritems - mid;
4017 btrfs_set_header_nritems(right, nritems);
4018 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
4020 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4021 btrfs_item_nr_offset(mid),
4022 nritems * sizeof(struct btrfs_item));
4024 copy_extent_buffer(right, l,
4025 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4026 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4027 leaf_data_end(fs_info, l), data_copy_size);
4029 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4031 for (i = 0; i < nritems; i++) {
4032 struct btrfs_item *item = btrfs_item_nr(i);
4035 ioff = btrfs_token_item_offset(right, item, &token);
4036 btrfs_set_token_item_offset(right, item,
4037 ioff + rt_data_off, &token);
4040 btrfs_set_header_nritems(l, mid);
4041 btrfs_item_key(right, &disk_key, 0);
4042 insert_ptr(trans, fs_info, path, &disk_key, right->start,
4043 path->slots[1] + 1, 1);
4045 btrfs_mark_buffer_dirty(right);
4046 btrfs_mark_buffer_dirty(l);
4047 BUG_ON(path->slots[0] != slot);
4050 btrfs_tree_unlock(path->nodes[0]);
4051 free_extent_buffer(path->nodes[0]);
4052 path->nodes[0] = right;
4053 path->slots[0] -= mid;
4054 path->slots[1] += 1;
4056 btrfs_tree_unlock(right);
4057 free_extent_buffer(right);
4060 BUG_ON(path->slots[0] < 0);
4064 * double splits happen when we need to insert a big item in the middle
4065 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4066 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4069 * We avoid this by trying to push the items on either side of our target
4070 * into the adjacent leaves. If all goes well we can avoid the double split
4073 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4074 struct btrfs_root *root,
4075 struct btrfs_path *path,
4078 struct btrfs_fs_info *fs_info = root->fs_info;
4083 int space_needed = data_size;
4085 slot = path->slots[0];
4086 if (slot < btrfs_header_nritems(path->nodes[0]))
4087 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4090 * try to push all the items after our slot into the
4093 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4100 nritems = btrfs_header_nritems(path->nodes[0]);
4102 * our goal is to get our slot at the start or end of a leaf. If
4103 * we've done so we're done
4105 if (path->slots[0] == 0 || path->slots[0] == nritems)
4108 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4111 /* try to push all the items before our slot into the next leaf */
4112 slot = path->slots[0];
4113 space_needed = data_size;
4115 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4116 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4129 * split the path's leaf in two, making sure there is at least data_size
4130 * available for the resulting leaf level of the path.
4132 * returns 0 if all went well and < 0 on failure.
4134 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4135 struct btrfs_root *root,
4136 const struct btrfs_key *ins_key,
4137 struct btrfs_path *path, int data_size,
4140 struct btrfs_disk_key disk_key;
4141 struct extent_buffer *l;
4145 struct extent_buffer *right;
4146 struct btrfs_fs_info *fs_info = root->fs_info;
4150 int num_doubles = 0;
4151 int tried_avoid_double = 0;
4154 slot = path->slots[0];
4155 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4156 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4159 /* first try to make some room by pushing left and right */
4160 if (data_size && path->nodes[1]) {
4161 int space_needed = data_size;
4163 if (slot < btrfs_header_nritems(l))
4164 space_needed -= btrfs_leaf_free_space(fs_info, l);
4166 wret = push_leaf_right(trans, root, path, space_needed,
4167 space_needed, 0, 0);
4171 space_needed = data_size;
4173 space_needed -= btrfs_leaf_free_space(fs_info,
4175 wret = push_leaf_left(trans, root, path, space_needed,
4176 space_needed, 0, (u32)-1);
4182 /* did the pushes work? */
4183 if (btrfs_leaf_free_space(fs_info, l) >= data_size)
4187 if (!path->nodes[1]) {
4188 ret = insert_new_root(trans, root, path, 1);
4195 slot = path->slots[0];
4196 nritems = btrfs_header_nritems(l);
4197 mid = (nritems + 1) / 2;
4201 leaf_space_used(l, mid, nritems - mid) + data_size >
4202 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4203 if (slot >= nritems) {
4207 if (mid != nritems &&
4208 leaf_space_used(l, mid, nritems - mid) +
4209 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4210 if (data_size && !tried_avoid_double)
4211 goto push_for_double;
4217 if (leaf_space_used(l, 0, mid) + data_size >
4218 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4219 if (!extend && data_size && slot == 0) {
4221 } else if ((extend || !data_size) && slot == 0) {
4225 if (mid != nritems &&
4226 leaf_space_used(l, mid, nritems - mid) +
4227 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4228 if (data_size && !tried_avoid_double)
4229 goto push_for_double;
4237 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4239 btrfs_item_key(l, &disk_key, mid);
4241 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4242 &disk_key, 0, l->start, 0);
4244 return PTR_ERR(right);
4246 root_add_used(root, fs_info->nodesize);
4250 btrfs_set_header_nritems(right, 0);
4251 insert_ptr(trans, fs_info, path, &disk_key,
4252 right->start, path->slots[1] + 1, 1);
4253 btrfs_tree_unlock(path->nodes[0]);
4254 free_extent_buffer(path->nodes[0]);
4255 path->nodes[0] = right;
4257 path->slots[1] += 1;
4259 btrfs_set_header_nritems(right, 0);
4260 insert_ptr(trans, fs_info, path, &disk_key,
4261 right->start, path->slots[1], 1);
4262 btrfs_tree_unlock(path->nodes[0]);
4263 free_extent_buffer(path->nodes[0]);
4264 path->nodes[0] = right;
4266 if (path->slots[1] == 0)
4267 fixup_low_keys(path, &disk_key, 1);
4270 * We create a new leaf 'right' for the required ins_len and
4271 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4272 * the content of ins_len to 'right'.
4277 copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
4280 BUG_ON(num_doubles != 0);
4288 push_for_double_split(trans, root, path, data_size);
4289 tried_avoid_double = 1;
4290 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4295 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4296 struct btrfs_root *root,
4297 struct btrfs_path *path, int ins_len)
4299 struct btrfs_fs_info *fs_info = root->fs_info;
4300 struct btrfs_key key;
4301 struct extent_buffer *leaf;
4302 struct btrfs_file_extent_item *fi;
4307 leaf = path->nodes[0];
4308 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4310 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4311 key.type != BTRFS_EXTENT_CSUM_KEY);
4313 if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
4316 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4317 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4318 fi = btrfs_item_ptr(leaf, path->slots[0],
4319 struct btrfs_file_extent_item);
4320 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4322 btrfs_release_path(path);
4324 path->keep_locks = 1;
4325 path->search_for_split = 1;
4326 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4327 path->search_for_split = 0;
4334 leaf = path->nodes[0];
4335 /* if our item isn't there, return now */
4336 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4339 /* the leaf has changed, it now has room. return now */
4340 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
4343 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4344 fi = btrfs_item_ptr(leaf, path->slots[0],
4345 struct btrfs_file_extent_item);
4346 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4350 btrfs_set_path_blocking(path);
4351 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4355 path->keep_locks = 0;
4356 btrfs_unlock_up_safe(path, 1);
4359 path->keep_locks = 0;
4363 static noinline int split_item(struct btrfs_fs_info *fs_info,
4364 struct btrfs_path *path,
4365 const struct btrfs_key *new_key,
4366 unsigned long split_offset)
4368 struct extent_buffer *leaf;
4369 struct btrfs_item *item;
4370 struct btrfs_item *new_item;
4376 struct btrfs_disk_key disk_key;
4378 leaf = path->nodes[0];
4379 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
4381 btrfs_set_path_blocking(path);
4383 item = btrfs_item_nr(path->slots[0]);
4384 orig_offset = btrfs_item_offset(leaf, item);
4385 item_size = btrfs_item_size(leaf, item);
4387 buf = kmalloc(item_size, GFP_NOFS);
4391 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4392 path->slots[0]), item_size);
4394 slot = path->slots[0] + 1;
4395 nritems = btrfs_header_nritems(leaf);
4396 if (slot != nritems) {
4397 /* shift the items */
4398 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4399 btrfs_item_nr_offset(slot),
4400 (nritems - slot) * sizeof(struct btrfs_item));
4403 btrfs_cpu_key_to_disk(&disk_key, new_key);
4404 btrfs_set_item_key(leaf, &disk_key, slot);
4406 new_item = btrfs_item_nr(slot);
4408 btrfs_set_item_offset(leaf, new_item, orig_offset);
4409 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4411 btrfs_set_item_offset(leaf, item,
4412 orig_offset + item_size - split_offset);
4413 btrfs_set_item_size(leaf, item, split_offset);
4415 btrfs_set_header_nritems(leaf, nritems + 1);
4417 /* write the data for the start of the original item */
4418 write_extent_buffer(leaf, buf,
4419 btrfs_item_ptr_offset(leaf, path->slots[0]),
4422 /* write the data for the new item */
4423 write_extent_buffer(leaf, buf + split_offset,
4424 btrfs_item_ptr_offset(leaf, slot),
4425 item_size - split_offset);
4426 btrfs_mark_buffer_dirty(leaf);
4428 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
4434 * This function splits a single item into two items,
4435 * giving 'new_key' to the new item and splitting the
4436 * old one at split_offset (from the start of the item).
4438 * The path may be released by this operation. After
4439 * the split, the path is pointing to the old item. The
4440 * new item is going to be in the same node as the old one.
4442 * Note, the item being split must be smaller enough to live alone on
4443 * a tree block with room for one extra struct btrfs_item
4445 * This allows us to split the item in place, keeping a lock on the
4446 * leaf the entire time.
4448 int btrfs_split_item(struct btrfs_trans_handle *trans,
4449 struct btrfs_root *root,
4450 struct btrfs_path *path,
4451 const struct btrfs_key *new_key,
4452 unsigned long split_offset)
4455 ret = setup_leaf_for_split(trans, root, path,
4456 sizeof(struct btrfs_item));
4460 ret = split_item(root->fs_info, path, new_key, split_offset);
4465 * This function duplicate a item, giving 'new_key' to the new item.
4466 * It guarantees both items live in the same tree leaf and the new item
4467 * is contiguous with the original item.
4469 * This allows us to split file extent in place, keeping a lock on the
4470 * leaf the entire time.
4472 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4473 struct btrfs_root *root,
4474 struct btrfs_path *path,
4475 const struct btrfs_key *new_key)
4477 struct extent_buffer *leaf;
4481 leaf = path->nodes[0];
4482 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4483 ret = setup_leaf_for_split(trans, root, path,
4484 item_size + sizeof(struct btrfs_item));
4489 setup_items_for_insert(root, path, new_key, &item_size,
4490 item_size, item_size +
4491 sizeof(struct btrfs_item), 1);
4492 leaf = path->nodes[0];
4493 memcpy_extent_buffer(leaf,
4494 btrfs_item_ptr_offset(leaf, path->slots[0]),
4495 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4501 * make the item pointed to by the path smaller. new_size indicates
4502 * how small to make it, and from_end tells us if we just chop bytes
4503 * off the end of the item or if we shift the item to chop bytes off
4506 void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
4507 struct btrfs_path *path, u32 new_size, int from_end)
4510 struct extent_buffer *leaf;
4511 struct btrfs_item *item;
4513 unsigned int data_end;
4514 unsigned int old_data_start;
4515 unsigned int old_size;
4516 unsigned int size_diff;
4518 struct btrfs_map_token token;
4520 btrfs_init_map_token(&token);
4522 leaf = path->nodes[0];
4523 slot = path->slots[0];
4525 old_size = btrfs_item_size_nr(leaf, slot);
4526 if (old_size == new_size)
4529 nritems = btrfs_header_nritems(leaf);
4530 data_end = leaf_data_end(fs_info, leaf);
4532 old_data_start = btrfs_item_offset_nr(leaf, slot);
4534 size_diff = old_size - new_size;
4537 BUG_ON(slot >= nritems);
4540 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4542 /* first correct the data pointers */
4543 for (i = slot; i < nritems; i++) {
4545 item = btrfs_item_nr(i);
4547 ioff = btrfs_token_item_offset(leaf, item, &token);
4548 btrfs_set_token_item_offset(leaf, item,
4549 ioff + size_diff, &token);
4552 /* shift the data */
4554 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4555 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4556 data_end, old_data_start + new_size - data_end);
4558 struct btrfs_disk_key disk_key;
4561 btrfs_item_key(leaf, &disk_key, slot);
4563 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4565 struct btrfs_file_extent_item *fi;
4567 fi = btrfs_item_ptr(leaf, slot,
4568 struct btrfs_file_extent_item);
4569 fi = (struct btrfs_file_extent_item *)(
4570 (unsigned long)fi - size_diff);
4572 if (btrfs_file_extent_type(leaf, fi) ==
4573 BTRFS_FILE_EXTENT_INLINE) {
4574 ptr = btrfs_item_ptr_offset(leaf, slot);
4575 memmove_extent_buffer(leaf, ptr,
4577 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4581 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4582 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4583 data_end, old_data_start - data_end);
4585 offset = btrfs_disk_key_offset(&disk_key);
4586 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4587 btrfs_set_item_key(leaf, &disk_key, slot);
4589 fixup_low_keys(path, &disk_key, 1);
4592 item = btrfs_item_nr(slot);
4593 btrfs_set_item_size(leaf, item, new_size);
4594 btrfs_mark_buffer_dirty(leaf);
4596 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4597 btrfs_print_leaf(leaf);
4603 * make the item pointed to by the path bigger, data_size is the added size.
4605 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4609 struct extent_buffer *leaf;
4610 struct btrfs_item *item;
4612 unsigned int data_end;
4613 unsigned int old_data;
4614 unsigned int old_size;
4616 struct btrfs_map_token token;
4618 btrfs_init_map_token(&token);
4620 leaf = path->nodes[0];
4622 nritems = btrfs_header_nritems(leaf);
4623 data_end = leaf_data_end(fs_info, leaf);
4625 if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
4626 btrfs_print_leaf(leaf);
4629 slot = path->slots[0];
4630 old_data = btrfs_item_end_nr(leaf, slot);
4633 if (slot >= nritems) {
4634 btrfs_print_leaf(leaf);
4635 btrfs_crit(fs_info, "slot %d too large, nritems %d",
4641 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4643 /* first correct the data pointers */
4644 for (i = slot; i < nritems; i++) {
4646 item = btrfs_item_nr(i);
4648 ioff = btrfs_token_item_offset(leaf, item, &token);
4649 btrfs_set_token_item_offset(leaf, item,
4650 ioff - data_size, &token);
4653 /* shift the data */
4654 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4655 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4656 data_end, old_data - data_end);
4658 data_end = old_data;
4659 old_size = btrfs_item_size_nr(leaf, slot);
4660 item = btrfs_item_nr(slot);
4661 btrfs_set_item_size(leaf, item, old_size + data_size);
4662 btrfs_mark_buffer_dirty(leaf);
4664 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4665 btrfs_print_leaf(leaf);
4671 * this is a helper for btrfs_insert_empty_items, the main goal here is
4672 * to save stack depth by doing the bulk of the work in a function
4673 * that doesn't call btrfs_search_slot
4675 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4676 const struct btrfs_key *cpu_key, u32 *data_size,
4677 u32 total_data, u32 total_size, int nr)
4679 struct btrfs_fs_info *fs_info = root->fs_info;
4680 struct btrfs_item *item;
4683 unsigned int data_end;
4684 struct btrfs_disk_key disk_key;
4685 struct extent_buffer *leaf;
4687 struct btrfs_map_token token;
4689 if (path->slots[0] == 0) {
4690 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4691 fixup_low_keys(path, &disk_key, 1);
4693 btrfs_unlock_up_safe(path, 1);
4695 btrfs_init_map_token(&token);
4697 leaf = path->nodes[0];
4698 slot = path->slots[0];
4700 nritems = btrfs_header_nritems(leaf);
4701 data_end = leaf_data_end(fs_info, leaf);
4703 if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
4704 btrfs_print_leaf(leaf);
4705 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4706 total_size, btrfs_leaf_free_space(fs_info, leaf));
4710 if (slot != nritems) {
4711 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4713 if (old_data < data_end) {
4714 btrfs_print_leaf(leaf);
4715 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4716 slot, old_data, data_end);
4720 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4722 /* first correct the data pointers */
4723 for (i = slot; i < nritems; i++) {
4726 item = btrfs_item_nr(i);
4727 ioff = btrfs_token_item_offset(leaf, item, &token);
4728 btrfs_set_token_item_offset(leaf, item,
4729 ioff - total_data, &token);
4731 /* shift the items */
4732 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4733 btrfs_item_nr_offset(slot),
4734 (nritems - slot) * sizeof(struct btrfs_item));
4736 /* shift the data */
4737 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4738 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4739 data_end, old_data - data_end);
4740 data_end = old_data;
4743 /* setup the item for the new data */
4744 for (i = 0; i < nr; i++) {
4745 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4746 btrfs_set_item_key(leaf, &disk_key, slot + i);
4747 item = btrfs_item_nr(slot + i);
4748 btrfs_set_token_item_offset(leaf, item,
4749 data_end - data_size[i], &token);
4750 data_end -= data_size[i];
4751 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4754 btrfs_set_header_nritems(leaf, nritems + nr);
4755 btrfs_mark_buffer_dirty(leaf);
4757 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4758 btrfs_print_leaf(leaf);
4764 * Given a key and some data, insert items into the tree.
4765 * This does all the path init required, making room in the tree if needed.
4767 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4768 struct btrfs_root *root,
4769 struct btrfs_path *path,
4770 const struct btrfs_key *cpu_key, u32 *data_size,
4779 for (i = 0; i < nr; i++)
4780 total_data += data_size[i];
4782 total_size = total_data + (nr * sizeof(struct btrfs_item));
4783 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4789 slot = path->slots[0];
4792 setup_items_for_insert(root, path, cpu_key, data_size,
4793 total_data, total_size, nr);
4798 * Given a key and some data, insert an item into the tree.
4799 * This does all the path init required, making room in the tree if needed.
4801 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4802 const struct btrfs_key *cpu_key, void *data,
4806 struct btrfs_path *path;
4807 struct extent_buffer *leaf;
4810 path = btrfs_alloc_path();
4813 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4815 leaf = path->nodes[0];
4816 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4817 write_extent_buffer(leaf, data, ptr, data_size);
4818 btrfs_mark_buffer_dirty(leaf);
4820 btrfs_free_path(path);
4825 * delete the pointer from a given node.
4827 * the tree should have been previously balanced so the deletion does not
4830 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4831 int level, int slot)
4833 struct extent_buffer *parent = path->nodes[level];
4837 nritems = btrfs_header_nritems(parent);
4838 if (slot != nritems - 1) {
4840 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4841 nritems - slot - 1);
4844 memmove_extent_buffer(parent,
4845 btrfs_node_key_ptr_offset(slot),
4846 btrfs_node_key_ptr_offset(slot + 1),
4847 sizeof(struct btrfs_key_ptr) *
4848 (nritems - slot - 1));
4850 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4856 btrfs_set_header_nritems(parent, nritems);
4857 if (nritems == 0 && parent == root->node) {
4858 BUG_ON(btrfs_header_level(root->node) != 1);
4859 /* just turn the root into a leaf and break */
4860 btrfs_set_header_level(root->node, 0);
4861 } else if (slot == 0) {
4862 struct btrfs_disk_key disk_key;
4864 btrfs_node_key(parent, &disk_key, 0);
4865 fixup_low_keys(path, &disk_key, level + 1);
4867 btrfs_mark_buffer_dirty(parent);
4871 * a helper function to delete the leaf pointed to by path->slots[1] and
4874 * This deletes the pointer in path->nodes[1] and frees the leaf
4875 * block extent. zero is returned if it all worked out, < 0 otherwise.
4877 * The path must have already been setup for deleting the leaf, including
4878 * all the proper balancing. path->nodes[1] must be locked.
4880 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4881 struct btrfs_root *root,
4882 struct btrfs_path *path,
4883 struct extent_buffer *leaf)
4885 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4886 del_ptr(root, path, 1, path->slots[1]);
4889 * btrfs_free_extent is expensive, we want to make sure we
4890 * aren't holding any locks when we call it
4892 btrfs_unlock_up_safe(path, 0);
4894 root_sub_used(root, leaf->len);
4896 extent_buffer_get(leaf);
4897 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4898 free_extent_buffer_stale(leaf);
4901 * delete the item at the leaf level in path. If that empties
4902 * the leaf, remove it from the tree
4904 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4905 struct btrfs_path *path, int slot, int nr)
4907 struct btrfs_fs_info *fs_info = root->fs_info;
4908 struct extent_buffer *leaf;
4909 struct btrfs_item *item;
4916 struct btrfs_map_token token;
4918 btrfs_init_map_token(&token);
4920 leaf = path->nodes[0];
4921 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4923 for (i = 0; i < nr; i++)
4924 dsize += btrfs_item_size_nr(leaf, slot + i);
4926 nritems = btrfs_header_nritems(leaf);
4928 if (slot + nr != nritems) {
4929 int data_end = leaf_data_end(fs_info, leaf);
4931 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4933 BTRFS_LEAF_DATA_OFFSET + data_end,
4934 last_off - data_end);
4936 for (i = slot + nr; i < nritems; i++) {
4939 item = btrfs_item_nr(i);
4940 ioff = btrfs_token_item_offset(leaf, item, &token);
4941 btrfs_set_token_item_offset(leaf, item,
4942 ioff + dsize, &token);
4945 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4946 btrfs_item_nr_offset(slot + nr),
4947 sizeof(struct btrfs_item) *
4948 (nritems - slot - nr));
4950 btrfs_set_header_nritems(leaf, nritems - nr);
4953 /* delete the leaf if we've emptied it */
4955 if (leaf == root->node) {
4956 btrfs_set_header_level(leaf, 0);
4958 btrfs_set_path_blocking(path);
4959 clean_tree_block(fs_info, leaf);
4960 btrfs_del_leaf(trans, root, path, leaf);
4963 int used = leaf_space_used(leaf, 0, nritems);
4965 struct btrfs_disk_key disk_key;
4967 btrfs_item_key(leaf, &disk_key, 0);
4968 fixup_low_keys(path, &disk_key, 1);
4971 /* delete the leaf if it is mostly empty */
4972 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4973 /* push_leaf_left fixes the path.
4974 * make sure the path still points to our leaf
4975 * for possible call to del_ptr below
4977 slot = path->slots[1];
4978 extent_buffer_get(leaf);
4980 btrfs_set_path_blocking(path);
4981 wret = push_leaf_left(trans, root, path, 1, 1,
4983 if (wret < 0 && wret != -ENOSPC)
4986 if (path->nodes[0] == leaf &&
4987 btrfs_header_nritems(leaf)) {
4988 wret = push_leaf_right(trans, root, path, 1,
4990 if (wret < 0 && wret != -ENOSPC)
4994 if (btrfs_header_nritems(leaf) == 0) {
4995 path->slots[1] = slot;
4996 btrfs_del_leaf(trans, root, path, leaf);
4997 free_extent_buffer(leaf);
5000 /* if we're still in the path, make sure
5001 * we're dirty. Otherwise, one of the
5002 * push_leaf functions must have already
5003 * dirtied this buffer
5005 if (path->nodes[0] == leaf)
5006 btrfs_mark_buffer_dirty(leaf);
5007 free_extent_buffer(leaf);
5010 btrfs_mark_buffer_dirty(leaf);
5017 * search the tree again to find a leaf with lesser keys
5018 * returns 0 if it found something or 1 if there are no lesser leaves.
5019 * returns < 0 on io errors.
5021 * This may release the path, and so you may lose any locks held at the
5024 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5026 struct btrfs_key key;
5027 struct btrfs_disk_key found_key;
5030 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5032 if (key.offset > 0) {
5034 } else if (key.type > 0) {
5036 key.offset = (u64)-1;
5037 } else if (key.objectid > 0) {
5040 key.offset = (u64)-1;
5045 btrfs_release_path(path);
5046 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5049 btrfs_item_key(path->nodes[0], &found_key, 0);
5050 ret = comp_keys(&found_key, &key);
5052 * We might have had an item with the previous key in the tree right
5053 * before we released our path. And after we released our path, that
5054 * item might have been pushed to the first slot (0) of the leaf we
5055 * were holding due to a tree balance. Alternatively, an item with the
5056 * previous key can exist as the only element of a leaf (big fat item).
5057 * Therefore account for these 2 cases, so that our callers (like
5058 * btrfs_previous_item) don't miss an existing item with a key matching
5059 * the previous key we computed above.
5067 * A helper function to walk down the tree starting at min_key, and looking
5068 * for nodes or leaves that are have a minimum transaction id.
5069 * This is used by the btree defrag code, and tree logging
5071 * This does not cow, but it does stuff the starting key it finds back
5072 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5073 * key and get a writable path.
5075 * This honors path->lowest_level to prevent descent past a given level
5078 * min_trans indicates the oldest transaction that you are interested
5079 * in walking through. Any nodes or leaves older than min_trans are
5080 * skipped over (without reading them).
5082 * returns zero if something useful was found, < 0 on error and 1 if there
5083 * was nothing in the tree that matched the search criteria.
5085 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5086 struct btrfs_path *path,
5089 struct btrfs_fs_info *fs_info = root->fs_info;
5090 struct extent_buffer *cur;
5091 struct btrfs_key found_key;
5097 int keep_locks = path->keep_locks;
5099 path->keep_locks = 1;
5101 cur = btrfs_read_lock_root_node(root);
5102 level = btrfs_header_level(cur);
5103 WARN_ON(path->nodes[level]);
5104 path->nodes[level] = cur;
5105 path->locks[level] = BTRFS_READ_LOCK;
5107 if (btrfs_header_generation(cur) < min_trans) {
5112 nritems = btrfs_header_nritems(cur);
5113 level = btrfs_header_level(cur);
5114 sret = btrfs_bin_search(cur, min_key, level, &slot);
5116 /* at the lowest level, we're done, setup the path and exit */
5117 if (level == path->lowest_level) {
5118 if (slot >= nritems)
5121 path->slots[level] = slot;
5122 btrfs_item_key_to_cpu(cur, &found_key, slot);
5125 if (sret && slot > 0)
5128 * check this node pointer against the min_trans parameters.
5129 * If it is too old, old, skip to the next one.
5131 while (slot < nritems) {
5134 gen = btrfs_node_ptr_generation(cur, slot);
5135 if (gen < min_trans) {
5143 * we didn't find a candidate key in this node, walk forward
5144 * and find another one
5146 if (slot >= nritems) {
5147 path->slots[level] = slot;
5148 btrfs_set_path_blocking(path);
5149 sret = btrfs_find_next_key(root, path, min_key, level,
5152 btrfs_release_path(path);
5158 /* save our key for returning back */
5159 btrfs_node_key_to_cpu(cur, &found_key, slot);
5160 path->slots[level] = slot;
5161 if (level == path->lowest_level) {
5165 btrfs_set_path_blocking(path);
5166 cur = read_node_slot(fs_info, cur, slot);
5172 btrfs_tree_read_lock(cur);
5174 path->locks[level - 1] = BTRFS_READ_LOCK;
5175 path->nodes[level - 1] = cur;
5176 unlock_up(path, level, 1, 0, NULL);
5179 path->keep_locks = keep_locks;
5181 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5182 btrfs_set_path_blocking(path);
5183 memcpy(min_key, &found_key, sizeof(found_key));
5188 static int tree_move_down(struct btrfs_fs_info *fs_info,
5189 struct btrfs_path *path,
5192 struct extent_buffer *eb;
5194 BUG_ON(*level == 0);
5195 eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
5199 path->nodes[*level - 1] = eb;
5200 path->slots[*level - 1] = 0;
5205 static int tree_move_next_or_upnext(struct btrfs_path *path,
5206 int *level, int root_level)
5210 nritems = btrfs_header_nritems(path->nodes[*level]);
5212 path->slots[*level]++;
5214 while (path->slots[*level] >= nritems) {
5215 if (*level == root_level)
5219 path->slots[*level] = 0;
5220 free_extent_buffer(path->nodes[*level]);
5221 path->nodes[*level] = NULL;
5223 path->slots[*level]++;
5225 nritems = btrfs_header_nritems(path->nodes[*level]);
5232 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5235 static int tree_advance(struct btrfs_fs_info *fs_info,
5236 struct btrfs_path *path,
5237 int *level, int root_level,
5239 struct btrfs_key *key)
5243 if (*level == 0 || !allow_down) {
5244 ret = tree_move_next_or_upnext(path, level, root_level);
5246 ret = tree_move_down(fs_info, path, level);
5250 btrfs_item_key_to_cpu(path->nodes[*level], key,
5251 path->slots[*level]);
5253 btrfs_node_key_to_cpu(path->nodes[*level], key,
5254 path->slots[*level]);
5259 static int tree_compare_item(struct btrfs_path *left_path,
5260 struct btrfs_path *right_path,
5265 unsigned long off1, off2;
5267 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5268 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5272 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5273 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5274 right_path->slots[0]);
5276 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5278 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5285 #define ADVANCE_ONLY_NEXT -1
5288 * This function compares two trees and calls the provided callback for
5289 * every changed/new/deleted item it finds.
5290 * If shared tree blocks are encountered, whole subtrees are skipped, making
5291 * the compare pretty fast on snapshotted subvolumes.
5293 * This currently works on commit roots only. As commit roots are read only,
5294 * we don't do any locking. The commit roots are protected with transactions.
5295 * Transactions are ended and rejoined when a commit is tried in between.
5297 * This function checks for modifications done to the trees while comparing.
5298 * If it detects a change, it aborts immediately.
5300 int btrfs_compare_trees(struct btrfs_root *left_root,
5301 struct btrfs_root *right_root,
5302 btrfs_changed_cb_t changed_cb, void *ctx)
5304 struct btrfs_fs_info *fs_info = left_root->fs_info;
5307 struct btrfs_path *left_path = NULL;
5308 struct btrfs_path *right_path = NULL;
5309 struct btrfs_key left_key;
5310 struct btrfs_key right_key;
5311 char *tmp_buf = NULL;
5312 int left_root_level;
5313 int right_root_level;
5316 int left_end_reached;
5317 int right_end_reached;
5325 left_path = btrfs_alloc_path();
5330 right_path = btrfs_alloc_path();
5336 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
5342 left_path->search_commit_root = 1;
5343 left_path->skip_locking = 1;
5344 right_path->search_commit_root = 1;
5345 right_path->skip_locking = 1;
5348 * Strategy: Go to the first items of both trees. Then do
5350 * If both trees are at level 0
5351 * Compare keys of current items
5352 * If left < right treat left item as new, advance left tree
5354 * If left > right treat right item as deleted, advance right tree
5356 * If left == right do deep compare of items, treat as changed if
5357 * needed, advance both trees and repeat
5358 * If both trees are at the same level but not at level 0
5359 * Compare keys of current nodes/leafs
5360 * If left < right advance left tree and repeat
5361 * If left > right advance right tree and repeat
5362 * If left == right compare blockptrs of the next nodes/leafs
5363 * If they match advance both trees but stay at the same level
5365 * If they don't match advance both trees while allowing to go
5367 * If tree levels are different
5368 * Advance the tree that needs it and repeat
5370 * Advancing a tree means:
5371 * If we are at level 0, try to go to the next slot. If that's not
5372 * possible, go one level up and repeat. Stop when we found a level
5373 * where we could go to the next slot. We may at this point be on a
5376 * If we are not at level 0 and not on shared tree blocks, go one
5379 * If we are not at level 0 and on shared tree blocks, go one slot to
5380 * the right if possible or go up and right.
5383 down_read(&fs_info->commit_root_sem);
5384 left_level = btrfs_header_level(left_root->commit_root);
5385 left_root_level = left_level;
5386 left_path->nodes[left_level] =
5387 btrfs_clone_extent_buffer(left_root->commit_root);
5388 if (!left_path->nodes[left_level]) {
5389 up_read(&fs_info->commit_root_sem);
5393 extent_buffer_get(left_path->nodes[left_level]);
5395 right_level = btrfs_header_level(right_root->commit_root);
5396 right_root_level = right_level;
5397 right_path->nodes[right_level] =
5398 btrfs_clone_extent_buffer(right_root->commit_root);
5399 if (!right_path->nodes[right_level]) {
5400 up_read(&fs_info->commit_root_sem);
5404 extent_buffer_get(right_path->nodes[right_level]);
5405 up_read(&fs_info->commit_root_sem);
5407 if (left_level == 0)
5408 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5409 &left_key, left_path->slots[left_level]);
5411 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5412 &left_key, left_path->slots[left_level]);
5413 if (right_level == 0)
5414 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5415 &right_key, right_path->slots[right_level]);
5417 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5418 &right_key, right_path->slots[right_level]);
5420 left_end_reached = right_end_reached = 0;
5421 advance_left = advance_right = 0;
5424 if (advance_left && !left_end_reached) {
5425 ret = tree_advance(fs_info, left_path, &left_level,
5427 advance_left != ADVANCE_ONLY_NEXT,
5430 left_end_reached = ADVANCE;
5435 if (advance_right && !right_end_reached) {
5436 ret = tree_advance(fs_info, right_path, &right_level,
5438 advance_right != ADVANCE_ONLY_NEXT,
5441 right_end_reached = ADVANCE;
5447 if (left_end_reached && right_end_reached) {
5450 } else if (left_end_reached) {
5451 if (right_level == 0) {
5452 ret = changed_cb(left_path, right_path,
5454 BTRFS_COMPARE_TREE_DELETED,
5459 advance_right = ADVANCE;
5461 } else if (right_end_reached) {
5462 if (left_level == 0) {
5463 ret = changed_cb(left_path, right_path,
5465 BTRFS_COMPARE_TREE_NEW,
5470 advance_left = ADVANCE;
5474 if (left_level == 0 && right_level == 0) {
5475 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5477 ret = changed_cb(left_path, right_path,
5479 BTRFS_COMPARE_TREE_NEW,
5483 advance_left = ADVANCE;
5484 } else if (cmp > 0) {
5485 ret = changed_cb(left_path, right_path,
5487 BTRFS_COMPARE_TREE_DELETED,
5491 advance_right = ADVANCE;
5493 enum btrfs_compare_tree_result result;
5495 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5496 ret = tree_compare_item(left_path, right_path,
5499 result = BTRFS_COMPARE_TREE_CHANGED;
5501 result = BTRFS_COMPARE_TREE_SAME;
5502 ret = changed_cb(left_path, right_path,
5503 &left_key, result, ctx);
5506 advance_left = ADVANCE;
5507 advance_right = ADVANCE;
5509 } else if (left_level == right_level) {
5510 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5512 advance_left = ADVANCE;
5513 } else if (cmp > 0) {
5514 advance_right = ADVANCE;
5516 left_blockptr = btrfs_node_blockptr(
5517 left_path->nodes[left_level],
5518 left_path->slots[left_level]);
5519 right_blockptr = btrfs_node_blockptr(
5520 right_path->nodes[right_level],
5521 right_path->slots[right_level]);
5522 left_gen = btrfs_node_ptr_generation(
5523 left_path->nodes[left_level],
5524 left_path->slots[left_level]);
5525 right_gen = btrfs_node_ptr_generation(
5526 right_path->nodes[right_level],
5527 right_path->slots[right_level]);
5528 if (left_blockptr == right_blockptr &&
5529 left_gen == right_gen) {
5531 * As we're on a shared block, don't
5532 * allow to go deeper.
5534 advance_left = ADVANCE_ONLY_NEXT;
5535 advance_right = ADVANCE_ONLY_NEXT;
5537 advance_left = ADVANCE;
5538 advance_right = ADVANCE;
5541 } else if (left_level < right_level) {
5542 advance_right = ADVANCE;
5544 advance_left = ADVANCE;
5549 btrfs_free_path(left_path);
5550 btrfs_free_path(right_path);
5556 * this is similar to btrfs_next_leaf, but does not try to preserve
5557 * and fixup the path. It looks for and returns the next key in the
5558 * tree based on the current path and the min_trans parameters.
5560 * 0 is returned if another key is found, < 0 if there are any errors
5561 * and 1 is returned if there are no higher keys in the tree
5563 * path->keep_locks should be set to 1 on the search made before
5564 * calling this function.
5566 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5567 struct btrfs_key *key, int level, u64 min_trans)
5570 struct extent_buffer *c;
5572 WARN_ON(!path->keep_locks);
5573 while (level < BTRFS_MAX_LEVEL) {
5574 if (!path->nodes[level])
5577 slot = path->slots[level] + 1;
5578 c = path->nodes[level];
5580 if (slot >= btrfs_header_nritems(c)) {
5583 struct btrfs_key cur_key;
5584 if (level + 1 >= BTRFS_MAX_LEVEL ||
5585 !path->nodes[level + 1])
5588 if (path->locks[level + 1]) {
5593 slot = btrfs_header_nritems(c) - 1;
5595 btrfs_item_key_to_cpu(c, &cur_key, slot);
5597 btrfs_node_key_to_cpu(c, &cur_key, slot);
5599 orig_lowest = path->lowest_level;
5600 btrfs_release_path(path);
5601 path->lowest_level = level;
5602 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5604 path->lowest_level = orig_lowest;
5608 c = path->nodes[level];
5609 slot = path->slots[level];
5616 btrfs_item_key_to_cpu(c, key, slot);
5618 u64 gen = btrfs_node_ptr_generation(c, slot);
5620 if (gen < min_trans) {
5624 btrfs_node_key_to_cpu(c, key, slot);
5632 * search the tree again to find a leaf with greater keys
5633 * returns 0 if it found something or 1 if there are no greater leaves.
5634 * returns < 0 on io errors.
5636 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5638 return btrfs_next_old_leaf(root, path, 0);
5641 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5646 struct extent_buffer *c;
5647 struct extent_buffer *next;
5648 struct btrfs_key key;
5651 int old_spinning = path->leave_spinning;
5652 int next_rw_lock = 0;
5654 nritems = btrfs_header_nritems(path->nodes[0]);
5658 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5663 btrfs_release_path(path);
5665 path->keep_locks = 1;
5666 path->leave_spinning = 1;
5669 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5671 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5672 path->keep_locks = 0;
5677 nritems = btrfs_header_nritems(path->nodes[0]);
5679 * by releasing the path above we dropped all our locks. A balance
5680 * could have added more items next to the key that used to be
5681 * at the very end of the block. So, check again here and
5682 * advance the path if there are now more items available.
5684 if (nritems > 0 && path->slots[0] < nritems - 1) {
5691 * So the above check misses one case:
5692 * - after releasing the path above, someone has removed the item that
5693 * used to be at the very end of the block, and balance between leafs
5694 * gets another one with bigger key.offset to replace it.
5696 * This one should be returned as well, or we can get leaf corruption
5697 * later(esp. in __btrfs_drop_extents()).
5699 * And a bit more explanation about this check,
5700 * with ret > 0, the key isn't found, the path points to the slot
5701 * where it should be inserted, so the path->slots[0] item must be the
5704 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5709 while (level < BTRFS_MAX_LEVEL) {
5710 if (!path->nodes[level]) {
5715 slot = path->slots[level] + 1;
5716 c = path->nodes[level];
5717 if (slot >= btrfs_header_nritems(c)) {
5719 if (level == BTRFS_MAX_LEVEL) {
5727 btrfs_tree_unlock_rw(next, next_rw_lock);
5728 free_extent_buffer(next);
5732 next_rw_lock = path->locks[level];
5733 ret = read_block_for_search(root, path, &next, level,
5739 btrfs_release_path(path);
5743 if (!path->skip_locking) {
5744 ret = btrfs_try_tree_read_lock(next);
5745 if (!ret && time_seq) {
5747 * If we don't get the lock, we may be racing
5748 * with push_leaf_left, holding that lock while
5749 * itself waiting for the leaf we've currently
5750 * locked. To solve this situation, we give up
5751 * on our lock and cycle.
5753 free_extent_buffer(next);
5754 btrfs_release_path(path);
5759 btrfs_set_path_blocking(path);
5760 btrfs_tree_read_lock(next);
5762 next_rw_lock = BTRFS_READ_LOCK;
5766 path->slots[level] = slot;
5769 c = path->nodes[level];
5770 if (path->locks[level])
5771 btrfs_tree_unlock_rw(c, path->locks[level]);
5773 free_extent_buffer(c);
5774 path->nodes[level] = next;
5775 path->slots[level] = 0;
5776 if (!path->skip_locking)
5777 path->locks[level] = next_rw_lock;
5781 ret = read_block_for_search(root, path, &next, level,
5787 btrfs_release_path(path);
5791 if (!path->skip_locking) {
5792 ret = btrfs_try_tree_read_lock(next);
5794 btrfs_set_path_blocking(path);
5795 btrfs_tree_read_lock(next);
5797 next_rw_lock = BTRFS_READ_LOCK;
5802 unlock_up(path, 0, 1, 0, NULL);
5803 path->leave_spinning = old_spinning;
5805 btrfs_set_path_blocking(path);
5811 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5812 * searching until it gets past min_objectid or finds an item of 'type'
5814 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5816 int btrfs_previous_item(struct btrfs_root *root,
5817 struct btrfs_path *path, u64 min_objectid,
5820 struct btrfs_key found_key;
5821 struct extent_buffer *leaf;
5826 if (path->slots[0] == 0) {
5827 btrfs_set_path_blocking(path);
5828 ret = btrfs_prev_leaf(root, path);
5834 leaf = path->nodes[0];
5835 nritems = btrfs_header_nritems(leaf);
5838 if (path->slots[0] == nritems)
5841 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5842 if (found_key.objectid < min_objectid)
5844 if (found_key.type == type)
5846 if (found_key.objectid == min_objectid &&
5847 found_key.type < type)
5854 * search in extent tree to find a previous Metadata/Data extent item with
5857 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5859 int btrfs_previous_extent_item(struct btrfs_root *root,
5860 struct btrfs_path *path, u64 min_objectid)
5862 struct btrfs_key found_key;
5863 struct extent_buffer *leaf;
5868 if (path->slots[0] == 0) {
5869 btrfs_set_path_blocking(path);
5870 ret = btrfs_prev_leaf(root, path);
5876 leaf = path->nodes[0];
5877 nritems = btrfs_header_nritems(leaf);
5880 if (path->slots[0] == nritems)
5883 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5884 if (found_key.objectid < min_objectid)
5886 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5887 found_key.type == BTRFS_METADATA_ITEM_KEY)
5889 if (found_key.objectid == min_objectid &&
5890 found_key.type < BTRFS_EXTENT_ITEM_KEY)