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"
18 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
19 *root, struct btrfs_path *path, int level);
20 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
21 const struct btrfs_key *ins_key, struct btrfs_path *path,
22 int data_size, int extend);
23 static int push_node_left(struct btrfs_trans_handle *trans,
24 struct extent_buffer *dst,
25 struct extent_buffer *src, int empty);
26 static int balance_node_right(struct btrfs_trans_handle *trans,
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])
48 * If we currently have a spinning reader or writer lock this
49 * will bump the count of blocking holders and drop the
52 if (p->locks[i] == BTRFS_READ_LOCK) {
53 btrfs_set_lock_blocking_read(p->nodes[i]);
54 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
55 } else if (p->locks[i] == BTRFS_WRITE_LOCK) {
56 btrfs_set_lock_blocking_write(p->nodes[i]);
57 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
62 /* this also releases the path */
63 void btrfs_free_path(struct btrfs_path *p)
67 btrfs_release_path(p);
68 kmem_cache_free(btrfs_path_cachep, p);
72 * path release drops references on the extent buffers in the path
73 * and it drops any locks held by this path
75 * It is safe to call this on paths that no locks or extent buffers held.
77 noinline void btrfs_release_path(struct btrfs_path *p)
81 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
86 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
89 free_extent_buffer(p->nodes[i]);
95 * safely gets a reference on the root node of a tree. A lock
96 * is not taken, so a concurrent writer may put a different node
97 * at the root of the tree. See btrfs_lock_root_node for the
100 * The extent buffer returned by this has a reference taken, so
101 * it won't disappear. It may stop being the root of the tree
102 * at any time because there are no locks held.
104 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
106 struct extent_buffer *eb;
110 eb = rcu_dereference(root->node);
113 * RCU really hurts here, we could free up the root node because
114 * it was COWed but we may not get the new root node yet so do
115 * the inc_not_zero dance and if it doesn't work then
116 * synchronize_rcu and try again.
118 if (atomic_inc_not_zero(&eb->refs)) {
128 /* loop around taking references on and locking the root node of the
129 * tree until you end up with a lock on the root. A locked buffer
130 * is returned, with a reference held.
132 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
134 struct extent_buffer *eb;
137 eb = btrfs_root_node(root);
139 if (eb == root->node)
141 btrfs_tree_unlock(eb);
142 free_extent_buffer(eb);
147 /* loop around taking references on and locking the root node of the
148 * tree until you end up with a lock on the root. A locked buffer
149 * is returned, with a reference held.
151 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
153 struct extent_buffer *eb;
156 eb = btrfs_root_node(root);
157 btrfs_tree_read_lock(eb);
158 if (eb == root->node)
160 btrfs_tree_read_unlock(eb);
161 free_extent_buffer(eb);
166 /* cowonly root (everything not a reference counted cow subvolume), just get
167 * put onto a simple dirty list. transaction.c walks this to make sure they
168 * get properly updated on disk.
170 static void add_root_to_dirty_list(struct btrfs_root *root)
172 struct btrfs_fs_info *fs_info = root->fs_info;
174 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
175 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
178 spin_lock(&fs_info->trans_lock);
179 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
180 /* Want the extent tree to be the last on the list */
181 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
182 list_move_tail(&root->dirty_list,
183 &fs_info->dirty_cowonly_roots);
185 list_move(&root->dirty_list,
186 &fs_info->dirty_cowonly_roots);
188 spin_unlock(&fs_info->trans_lock);
192 * used by snapshot creation to make a copy of a root for a tree with
193 * a given objectid. The buffer with the new root node is returned in
194 * cow_ret, and this func returns zero on success or a negative error code.
196 int btrfs_copy_root(struct btrfs_trans_handle *trans,
197 struct btrfs_root *root,
198 struct extent_buffer *buf,
199 struct extent_buffer **cow_ret, u64 new_root_objectid)
201 struct btrfs_fs_info *fs_info = root->fs_info;
202 struct extent_buffer *cow;
205 struct btrfs_disk_key disk_key;
207 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
208 trans->transid != fs_info->running_transaction->transid);
209 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
210 trans->transid != root->last_trans);
212 level = btrfs_header_level(buf);
214 btrfs_item_key(buf, &disk_key, 0);
216 btrfs_node_key(buf, &disk_key, 0);
218 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
219 &disk_key, level, buf->start, 0);
223 copy_extent_buffer_full(cow, buf);
224 btrfs_set_header_bytenr(cow, cow->start);
225 btrfs_set_header_generation(cow, trans->transid);
226 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
227 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
228 BTRFS_HEADER_FLAG_RELOC);
229 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
230 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
232 btrfs_set_header_owner(cow, new_root_objectid);
234 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
236 WARN_ON(btrfs_header_generation(buf) > trans->transid);
237 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
238 ret = btrfs_inc_ref(trans, root, cow, 1);
240 ret = btrfs_inc_ref(trans, root, cow, 0);
245 btrfs_mark_buffer_dirty(cow);
254 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
255 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
257 MOD_LOG_ROOT_REPLACE,
260 struct tree_mod_root {
265 struct tree_mod_elem {
271 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
274 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
277 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
278 struct btrfs_disk_key key;
281 /* this is used for op == MOD_LOG_MOVE_KEYS */
287 /* this is used for op == MOD_LOG_ROOT_REPLACE */
288 struct tree_mod_root old_root;
292 * Pull a new tree mod seq number for our operation.
294 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
296 return atomic64_inc_return(&fs_info->tree_mod_seq);
300 * This adds a new blocker to the tree mod log's blocker list if the @elem
301 * passed does not already have a sequence number set. So when a caller expects
302 * to record tree modifications, it should ensure to set elem->seq to zero
303 * before calling btrfs_get_tree_mod_seq.
304 * Returns a fresh, unused tree log modification sequence number, even if no new
307 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
308 struct seq_list *elem)
310 write_lock(&fs_info->tree_mod_log_lock);
311 spin_lock(&fs_info->tree_mod_seq_lock);
313 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
314 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
316 spin_unlock(&fs_info->tree_mod_seq_lock);
317 write_unlock(&fs_info->tree_mod_log_lock);
322 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
323 struct seq_list *elem)
325 struct rb_root *tm_root;
326 struct rb_node *node;
327 struct rb_node *next;
328 struct seq_list *cur_elem;
329 struct tree_mod_elem *tm;
330 u64 min_seq = (u64)-1;
331 u64 seq_putting = elem->seq;
336 spin_lock(&fs_info->tree_mod_seq_lock);
337 list_del(&elem->list);
340 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
341 if (cur_elem->seq < min_seq) {
342 if (seq_putting > cur_elem->seq) {
344 * blocker with lower sequence number exists, we
345 * cannot remove anything from the log
347 spin_unlock(&fs_info->tree_mod_seq_lock);
350 min_seq = cur_elem->seq;
353 spin_unlock(&fs_info->tree_mod_seq_lock);
356 * anything that's lower than the lowest existing (read: blocked)
357 * sequence number can be removed from the tree.
359 write_lock(&fs_info->tree_mod_log_lock);
360 tm_root = &fs_info->tree_mod_log;
361 for (node = rb_first(tm_root); node; node = next) {
362 next = rb_next(node);
363 tm = rb_entry(node, struct tree_mod_elem, node);
364 if (tm->seq > min_seq)
366 rb_erase(node, tm_root);
369 write_unlock(&fs_info->tree_mod_log_lock);
373 * key order of the log:
374 * node/leaf start address -> sequence
376 * The 'start address' is the logical address of the *new* root node
377 * for root replace operations, or the logical address of the affected
378 * block for all other operations.
381 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
383 struct rb_root *tm_root;
384 struct rb_node **new;
385 struct rb_node *parent = NULL;
386 struct tree_mod_elem *cur;
388 lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
390 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
392 tm_root = &fs_info->tree_mod_log;
393 new = &tm_root->rb_node;
395 cur = rb_entry(*new, struct tree_mod_elem, node);
397 if (cur->logical < tm->logical)
398 new = &((*new)->rb_left);
399 else if (cur->logical > tm->logical)
400 new = &((*new)->rb_right);
401 else if (cur->seq < tm->seq)
402 new = &((*new)->rb_left);
403 else if (cur->seq > tm->seq)
404 new = &((*new)->rb_right);
409 rb_link_node(&tm->node, parent, new);
410 rb_insert_color(&tm->node, tm_root);
415 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
416 * returns zero with the tree_mod_log_lock acquired. The caller must hold
417 * this until all tree mod log insertions are recorded in the rb tree and then
418 * write unlock fs_info::tree_mod_log_lock.
420 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
421 struct extent_buffer *eb) {
423 if (list_empty(&(fs_info)->tree_mod_seq_list))
425 if (eb && btrfs_header_level(eb) == 0)
428 write_lock(&fs_info->tree_mod_log_lock);
429 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
430 write_unlock(&fs_info->tree_mod_log_lock);
437 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
438 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
439 struct extent_buffer *eb)
442 if (list_empty(&(fs_info)->tree_mod_seq_list))
444 if (eb && btrfs_header_level(eb) == 0)
450 static struct tree_mod_elem *
451 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
452 enum mod_log_op op, gfp_t flags)
454 struct tree_mod_elem *tm;
456 tm = kzalloc(sizeof(*tm), flags);
460 tm->logical = eb->start;
461 if (op != MOD_LOG_KEY_ADD) {
462 btrfs_node_key(eb, &tm->key, slot);
463 tm->blockptr = btrfs_node_blockptr(eb, slot);
467 tm->generation = btrfs_node_ptr_generation(eb, slot);
468 RB_CLEAR_NODE(&tm->node);
473 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
474 enum mod_log_op op, gfp_t flags)
476 struct tree_mod_elem *tm;
479 if (!tree_mod_need_log(eb->fs_info, eb))
482 tm = alloc_tree_mod_elem(eb, slot, op, flags);
486 if (tree_mod_dont_log(eb->fs_info, eb)) {
491 ret = __tree_mod_log_insert(eb->fs_info, tm);
492 write_unlock(&eb->fs_info->tree_mod_log_lock);
499 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
500 int dst_slot, int src_slot, int nr_items)
502 struct tree_mod_elem *tm = NULL;
503 struct tree_mod_elem **tm_list = NULL;
508 if (!tree_mod_need_log(eb->fs_info, eb))
511 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
515 tm = kzalloc(sizeof(*tm), GFP_NOFS);
521 tm->logical = eb->start;
523 tm->move.dst_slot = dst_slot;
524 tm->move.nr_items = nr_items;
525 tm->op = MOD_LOG_MOVE_KEYS;
527 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
528 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
529 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
536 if (tree_mod_dont_log(eb->fs_info, eb))
541 * When we override something during the move, we log these removals.
542 * This can only happen when we move towards the beginning of the
543 * buffer, i.e. dst_slot < src_slot.
545 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
546 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
551 ret = __tree_mod_log_insert(eb->fs_info, tm);
554 write_unlock(&eb->fs_info->tree_mod_log_lock);
559 for (i = 0; i < nr_items; i++) {
560 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
561 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
565 write_unlock(&eb->fs_info->tree_mod_log_lock);
573 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
574 struct tree_mod_elem **tm_list,
580 for (i = nritems - 1; i >= 0; i--) {
581 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
583 for (j = nritems - 1; j > i; j--)
584 rb_erase(&tm_list[j]->node,
585 &fs_info->tree_mod_log);
593 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
594 struct extent_buffer *new_root, int log_removal)
596 struct btrfs_fs_info *fs_info = old_root->fs_info;
597 struct tree_mod_elem *tm = NULL;
598 struct tree_mod_elem **tm_list = NULL;
603 if (!tree_mod_need_log(fs_info, NULL))
606 if (log_removal && btrfs_header_level(old_root) > 0) {
607 nritems = btrfs_header_nritems(old_root);
608 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
614 for (i = 0; i < nritems; i++) {
615 tm_list[i] = alloc_tree_mod_elem(old_root, i,
616 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
624 tm = kzalloc(sizeof(*tm), GFP_NOFS);
630 tm->logical = new_root->start;
631 tm->old_root.logical = old_root->start;
632 tm->old_root.level = btrfs_header_level(old_root);
633 tm->generation = btrfs_header_generation(old_root);
634 tm->op = MOD_LOG_ROOT_REPLACE;
636 if (tree_mod_dont_log(fs_info, NULL))
640 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
642 ret = __tree_mod_log_insert(fs_info, tm);
644 write_unlock(&fs_info->tree_mod_log_lock);
653 for (i = 0; i < nritems; i++)
662 static struct tree_mod_elem *
663 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
666 struct rb_root *tm_root;
667 struct rb_node *node;
668 struct tree_mod_elem *cur = NULL;
669 struct tree_mod_elem *found = NULL;
671 read_lock(&fs_info->tree_mod_log_lock);
672 tm_root = &fs_info->tree_mod_log;
673 node = tm_root->rb_node;
675 cur = rb_entry(node, struct tree_mod_elem, node);
676 if (cur->logical < start) {
677 node = node->rb_left;
678 } else if (cur->logical > start) {
679 node = node->rb_right;
680 } else if (cur->seq < min_seq) {
681 node = node->rb_left;
682 } else if (!smallest) {
683 /* we want the node with the highest seq */
685 BUG_ON(found->seq > cur->seq);
687 node = node->rb_left;
688 } else if (cur->seq > min_seq) {
689 /* we want the node with the smallest seq */
691 BUG_ON(found->seq < cur->seq);
693 node = node->rb_right;
699 read_unlock(&fs_info->tree_mod_log_lock);
705 * this returns the element from the log with the smallest time sequence
706 * value that's in the log (the oldest log item). any element with a time
707 * sequence lower than min_seq will be ignored.
709 static struct tree_mod_elem *
710 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
713 return __tree_mod_log_search(fs_info, start, min_seq, 1);
717 * this returns the element from the log with the largest time sequence
718 * value that's in the log (the most recent log item). any element with
719 * a time sequence lower than min_seq will be ignored.
721 static struct tree_mod_elem *
722 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
724 return __tree_mod_log_search(fs_info, start, min_seq, 0);
727 static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst,
728 struct extent_buffer *src, unsigned long dst_offset,
729 unsigned long src_offset, int nr_items)
731 struct btrfs_fs_info *fs_info = dst->fs_info;
733 struct tree_mod_elem **tm_list = NULL;
734 struct tree_mod_elem **tm_list_add, **tm_list_rem;
738 if (!tree_mod_need_log(fs_info, NULL))
741 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
744 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
749 tm_list_add = tm_list;
750 tm_list_rem = tm_list + nr_items;
751 for (i = 0; i < nr_items; i++) {
752 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
753 MOD_LOG_KEY_REMOVE, GFP_NOFS);
754 if (!tm_list_rem[i]) {
759 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
760 MOD_LOG_KEY_ADD, GFP_NOFS);
761 if (!tm_list_add[i]) {
767 if (tree_mod_dont_log(fs_info, NULL))
771 for (i = 0; i < nr_items; i++) {
772 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
775 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
780 write_unlock(&fs_info->tree_mod_log_lock);
786 for (i = 0; i < nr_items * 2; i++) {
787 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
788 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
792 write_unlock(&fs_info->tree_mod_log_lock);
798 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
800 struct tree_mod_elem **tm_list = NULL;
805 if (btrfs_header_level(eb) == 0)
808 if (!tree_mod_need_log(eb->fs_info, NULL))
811 nritems = btrfs_header_nritems(eb);
812 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
816 for (i = 0; i < nritems; i++) {
817 tm_list[i] = alloc_tree_mod_elem(eb, i,
818 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
825 if (tree_mod_dont_log(eb->fs_info, eb))
828 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
829 write_unlock(&eb->fs_info->tree_mod_log_lock);
837 for (i = 0; i < nritems; i++)
845 * check if the tree block can be shared by multiple trees
847 int btrfs_block_can_be_shared(struct btrfs_root *root,
848 struct extent_buffer *buf)
851 * Tree blocks not in reference counted trees and tree roots
852 * are never shared. If a block was allocated after the last
853 * snapshot and the block was not allocated by tree relocation,
854 * we know the block is not shared.
856 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
857 buf != root->node && buf != root->commit_root &&
858 (btrfs_header_generation(buf) <=
859 btrfs_root_last_snapshot(&root->root_item) ||
860 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
866 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
867 struct btrfs_root *root,
868 struct extent_buffer *buf,
869 struct extent_buffer *cow,
872 struct btrfs_fs_info *fs_info = root->fs_info;
880 * Backrefs update rules:
882 * Always use full backrefs for extent pointers in tree block
883 * allocated by tree relocation.
885 * If a shared tree block is no longer referenced by its owner
886 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
887 * use full backrefs for extent pointers in tree block.
889 * If a tree block is been relocating
890 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
891 * use full backrefs for extent pointers in tree block.
892 * The reason for this is some operations (such as drop tree)
893 * are only allowed for blocks use full backrefs.
896 if (btrfs_block_can_be_shared(root, buf)) {
897 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
898 btrfs_header_level(buf), 1,
904 btrfs_handle_fs_error(fs_info, ret, NULL);
909 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
910 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
911 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
916 owner = btrfs_header_owner(buf);
917 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
918 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
921 if ((owner == root->root_key.objectid ||
922 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
923 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
924 ret = btrfs_inc_ref(trans, root, buf, 1);
928 if (root->root_key.objectid ==
929 BTRFS_TREE_RELOC_OBJECTID) {
930 ret = btrfs_dec_ref(trans, root, buf, 0);
933 ret = btrfs_inc_ref(trans, root, cow, 1);
937 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
940 if (root->root_key.objectid ==
941 BTRFS_TREE_RELOC_OBJECTID)
942 ret = btrfs_inc_ref(trans, root, cow, 1);
944 ret = btrfs_inc_ref(trans, root, cow, 0);
948 if (new_flags != 0) {
949 int level = btrfs_header_level(buf);
951 ret = btrfs_set_disk_extent_flags(trans,
954 new_flags, level, 0);
959 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
960 if (root->root_key.objectid ==
961 BTRFS_TREE_RELOC_OBJECTID)
962 ret = btrfs_inc_ref(trans, root, cow, 1);
964 ret = btrfs_inc_ref(trans, root, cow, 0);
967 ret = btrfs_dec_ref(trans, root, buf, 1);
971 btrfs_clean_tree_block(buf);
977 static struct extent_buffer *alloc_tree_block_no_bg_flush(
978 struct btrfs_trans_handle *trans,
979 struct btrfs_root *root,
981 const struct btrfs_disk_key *disk_key,
986 struct btrfs_fs_info *fs_info = root->fs_info;
987 struct extent_buffer *ret;
990 * If we are COWing a node/leaf from the extent, chunk, device or free
991 * space trees, make sure that we do not finish block group creation of
992 * pending block groups. We do this to avoid a deadlock.
993 * COWing can result in allocation of a new chunk, and flushing pending
994 * block groups (btrfs_create_pending_block_groups()) can be triggered
995 * when finishing allocation of a new chunk. Creation of a pending block
996 * group modifies the extent, chunk, device and free space trees,
997 * therefore we could deadlock with ourselves since we are holding a
998 * lock on an extent buffer that btrfs_create_pending_block_groups() may
1000 * For similar reasons, we also need to delay flushing pending block
1001 * groups when splitting a leaf or node, from one of those trees, since
1002 * we are holding a write lock on it and its parent or when inserting a
1003 * new root node for one of those trees.
1005 if (root == fs_info->extent_root ||
1006 root == fs_info->chunk_root ||
1007 root == fs_info->dev_root ||
1008 root == fs_info->free_space_root)
1009 trans->can_flush_pending_bgs = false;
1011 ret = btrfs_alloc_tree_block(trans, root, parent_start,
1012 root->root_key.objectid, disk_key, level,
1014 trans->can_flush_pending_bgs = true;
1020 * does the dirty work in cow of a single block. The parent block (if
1021 * supplied) is updated to point to the new cow copy. The new buffer is marked
1022 * dirty and returned locked. If you modify the block it needs to be marked
1025 * search_start -- an allocation hint for the new block
1027 * empty_size -- a hint that you plan on doing more cow. This is the size in
1028 * bytes the allocator should try to find free next to the block it returns.
1029 * This is just a hint and may be ignored by the allocator.
1031 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1032 struct btrfs_root *root,
1033 struct extent_buffer *buf,
1034 struct extent_buffer *parent, int parent_slot,
1035 struct extent_buffer **cow_ret,
1036 u64 search_start, u64 empty_size)
1038 struct btrfs_fs_info *fs_info = root->fs_info;
1039 struct btrfs_disk_key disk_key;
1040 struct extent_buffer *cow;
1043 int unlock_orig = 0;
1044 u64 parent_start = 0;
1046 if (*cow_ret == buf)
1049 btrfs_assert_tree_locked(buf);
1051 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1052 trans->transid != fs_info->running_transaction->transid);
1053 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1054 trans->transid != root->last_trans);
1056 level = btrfs_header_level(buf);
1059 btrfs_item_key(buf, &disk_key, 0);
1061 btrfs_node_key(buf, &disk_key, 0);
1063 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1064 parent_start = parent->start;
1066 cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1067 level, search_start, empty_size);
1069 return PTR_ERR(cow);
1071 /* cow is set to blocking by btrfs_init_new_buffer */
1073 copy_extent_buffer_full(cow, buf);
1074 btrfs_set_header_bytenr(cow, cow->start);
1075 btrfs_set_header_generation(cow, trans->transid);
1076 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1077 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1078 BTRFS_HEADER_FLAG_RELOC);
1079 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1080 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1082 btrfs_set_header_owner(cow, root->root_key.objectid);
1084 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1086 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1088 btrfs_abort_transaction(trans, ret);
1092 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1093 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1095 btrfs_abort_transaction(trans, ret);
1100 if (buf == root->node) {
1101 WARN_ON(parent && parent != buf);
1102 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1103 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1104 parent_start = buf->start;
1106 extent_buffer_get(cow);
1107 ret = tree_mod_log_insert_root(root->node, cow, 1);
1109 rcu_assign_pointer(root->node, cow);
1111 btrfs_free_tree_block(trans, root, buf, parent_start,
1113 free_extent_buffer(buf);
1114 add_root_to_dirty_list(root);
1116 WARN_ON(trans->transid != btrfs_header_generation(parent));
1117 tree_mod_log_insert_key(parent, parent_slot,
1118 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1119 btrfs_set_node_blockptr(parent, parent_slot,
1121 btrfs_set_node_ptr_generation(parent, parent_slot,
1123 btrfs_mark_buffer_dirty(parent);
1125 ret = tree_mod_log_free_eb(buf);
1127 btrfs_abort_transaction(trans, ret);
1131 btrfs_free_tree_block(trans, root, buf, parent_start,
1135 btrfs_tree_unlock(buf);
1136 free_extent_buffer_stale(buf);
1137 btrfs_mark_buffer_dirty(cow);
1143 * returns the logical address of the oldest predecessor of the given root.
1144 * entries older than time_seq are ignored.
1146 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1147 struct extent_buffer *eb_root, u64 time_seq)
1149 struct tree_mod_elem *tm;
1150 struct tree_mod_elem *found = NULL;
1151 u64 root_logical = eb_root->start;
1158 * the very last operation that's logged for a root is the
1159 * replacement operation (if it is replaced at all). this has
1160 * the logical address of the *new* root, making it the very
1161 * first operation that's logged for this root.
1164 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1169 * if there are no tree operation for the oldest root, we simply
1170 * return it. this should only happen if that (old) root is at
1177 * if there's an operation that's not a root replacement, we
1178 * found the oldest version of our root. normally, we'll find a
1179 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1181 if (tm->op != MOD_LOG_ROOT_REPLACE)
1185 root_logical = tm->old_root.logical;
1189 /* if there's no old root to return, return what we found instead */
1197 * tm is a pointer to the first operation to rewind within eb. then, all
1198 * previous operations will be rewound (until we reach something older than
1202 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1203 u64 time_seq, struct tree_mod_elem *first_tm)
1206 struct rb_node *next;
1207 struct tree_mod_elem *tm = first_tm;
1208 unsigned long o_dst;
1209 unsigned long o_src;
1210 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1212 n = btrfs_header_nritems(eb);
1213 read_lock(&fs_info->tree_mod_log_lock);
1214 while (tm && tm->seq >= time_seq) {
1216 * all the operations are recorded with the operator used for
1217 * the modification. as we're going backwards, we do the
1218 * opposite of each operation here.
1221 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1222 BUG_ON(tm->slot < n);
1224 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1225 case MOD_LOG_KEY_REMOVE:
1226 btrfs_set_node_key(eb, &tm->key, tm->slot);
1227 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1228 btrfs_set_node_ptr_generation(eb, tm->slot,
1232 case MOD_LOG_KEY_REPLACE:
1233 BUG_ON(tm->slot >= n);
1234 btrfs_set_node_key(eb, &tm->key, tm->slot);
1235 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1236 btrfs_set_node_ptr_generation(eb, tm->slot,
1239 case MOD_LOG_KEY_ADD:
1240 /* if a move operation is needed it's in the log */
1243 case MOD_LOG_MOVE_KEYS:
1244 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1245 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1246 memmove_extent_buffer(eb, o_dst, o_src,
1247 tm->move.nr_items * p_size);
1249 case MOD_LOG_ROOT_REPLACE:
1251 * this operation is special. for roots, this must be
1252 * handled explicitly before rewinding.
1253 * for non-roots, this operation may exist if the node
1254 * was a root: root A -> child B; then A gets empty and
1255 * B is promoted to the new root. in the mod log, we'll
1256 * have a root-replace operation for B, a tree block
1257 * that is no root. we simply ignore that operation.
1261 next = rb_next(&tm->node);
1264 tm = rb_entry(next, struct tree_mod_elem, node);
1265 if (tm->logical != first_tm->logical)
1268 read_unlock(&fs_info->tree_mod_log_lock);
1269 btrfs_set_header_nritems(eb, n);
1273 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1274 * is returned. If rewind operations happen, a fresh buffer is returned. The
1275 * returned buffer is always read-locked. If the returned buffer is not the
1276 * input buffer, the lock on the input buffer is released and the input buffer
1277 * is freed (its refcount is decremented).
1279 static struct extent_buffer *
1280 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1281 struct extent_buffer *eb, u64 time_seq)
1283 struct extent_buffer *eb_rewin;
1284 struct tree_mod_elem *tm;
1289 if (btrfs_header_level(eb) == 0)
1292 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1296 btrfs_set_path_blocking(path);
1297 btrfs_set_lock_blocking_read(eb);
1299 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1300 BUG_ON(tm->slot != 0);
1301 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1303 btrfs_tree_read_unlock_blocking(eb);
1304 free_extent_buffer(eb);
1307 btrfs_set_header_bytenr(eb_rewin, eb->start);
1308 btrfs_set_header_backref_rev(eb_rewin,
1309 btrfs_header_backref_rev(eb));
1310 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1311 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1313 eb_rewin = btrfs_clone_extent_buffer(eb);
1315 btrfs_tree_read_unlock_blocking(eb);
1316 free_extent_buffer(eb);
1321 btrfs_tree_read_unlock_blocking(eb);
1322 free_extent_buffer(eb);
1324 btrfs_tree_read_lock(eb_rewin);
1325 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1326 WARN_ON(btrfs_header_nritems(eb_rewin) >
1327 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1333 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1334 * value. If there are no changes, the current root->root_node is returned. If
1335 * anything changed in between, there's a fresh buffer allocated on which the
1336 * rewind operations are done. In any case, the returned buffer is read locked.
1337 * Returns NULL on error (with no locks held).
1339 static inline struct extent_buffer *
1340 get_old_root(struct btrfs_root *root, u64 time_seq)
1342 struct btrfs_fs_info *fs_info = root->fs_info;
1343 struct tree_mod_elem *tm;
1344 struct extent_buffer *eb = NULL;
1345 struct extent_buffer *eb_root;
1346 u64 eb_root_owner = 0;
1347 struct extent_buffer *old;
1348 struct tree_mod_root *old_root = NULL;
1349 u64 old_generation = 0;
1353 eb_root = btrfs_read_lock_root_node(root);
1354 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1358 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1359 old_root = &tm->old_root;
1360 old_generation = tm->generation;
1361 logical = old_root->logical;
1362 level = old_root->level;
1364 logical = eb_root->start;
1365 level = btrfs_header_level(eb_root);
1368 tm = tree_mod_log_search(fs_info, logical, time_seq);
1369 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1370 btrfs_tree_read_unlock(eb_root);
1371 free_extent_buffer(eb_root);
1372 old = read_tree_block(fs_info, logical, 0, level, NULL);
1373 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1375 free_extent_buffer(old);
1377 "failed to read tree block %llu from get_old_root",
1380 eb = btrfs_clone_extent_buffer(old);
1381 free_extent_buffer(old);
1383 } else if (old_root) {
1384 eb_root_owner = btrfs_header_owner(eb_root);
1385 btrfs_tree_read_unlock(eb_root);
1386 free_extent_buffer(eb_root);
1387 eb = alloc_dummy_extent_buffer(fs_info, logical);
1389 btrfs_set_lock_blocking_read(eb_root);
1390 eb = btrfs_clone_extent_buffer(eb_root);
1391 btrfs_tree_read_unlock_blocking(eb_root);
1392 free_extent_buffer(eb_root);
1397 btrfs_tree_read_lock(eb);
1399 btrfs_set_header_bytenr(eb, eb->start);
1400 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1401 btrfs_set_header_owner(eb, eb_root_owner);
1402 btrfs_set_header_level(eb, old_root->level);
1403 btrfs_set_header_generation(eb, old_generation);
1406 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1408 WARN_ON(btrfs_header_level(eb) != 0);
1409 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1414 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1416 struct tree_mod_elem *tm;
1418 struct extent_buffer *eb_root = btrfs_root_node(root);
1420 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1421 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1422 level = tm->old_root.level;
1424 level = btrfs_header_level(eb_root);
1426 free_extent_buffer(eb_root);
1431 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1432 struct btrfs_root *root,
1433 struct extent_buffer *buf)
1435 if (btrfs_is_testing(root->fs_info))
1438 /* Ensure we can see the FORCE_COW bit */
1439 smp_mb__before_atomic();
1442 * We do not need to cow a block if
1443 * 1) this block is not created or changed in this transaction;
1444 * 2) this block does not belong to TREE_RELOC tree;
1445 * 3) the root is not forced COW.
1447 * What is forced COW:
1448 * when we create snapshot during committing the transaction,
1449 * after we've finished copying src root, we must COW the shared
1450 * block to ensure the metadata consistency.
1452 if (btrfs_header_generation(buf) == trans->transid &&
1453 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1454 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1455 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1456 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1462 * cows a single block, see __btrfs_cow_block for the real work.
1463 * This version of it has extra checks so that a block isn't COWed more than
1464 * once per transaction, as long as it hasn't been written yet
1466 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1467 struct btrfs_root *root, struct extent_buffer *buf,
1468 struct extent_buffer *parent, int parent_slot,
1469 struct extent_buffer **cow_ret)
1471 struct btrfs_fs_info *fs_info = root->fs_info;
1475 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1477 "COW'ing blocks on a fs root that's being dropped");
1479 if (trans->transaction != fs_info->running_transaction)
1480 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1482 fs_info->running_transaction->transid);
1484 if (trans->transid != fs_info->generation)
1485 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1486 trans->transid, fs_info->generation);
1488 if (!should_cow_block(trans, root, buf)) {
1489 trans->dirty = true;
1494 search_start = buf->start & ~((u64)SZ_1G - 1);
1497 btrfs_set_lock_blocking_write(parent);
1498 btrfs_set_lock_blocking_write(buf);
1501 * Before CoWing this block for later modification, check if it's
1502 * the subtree root and do the delayed subtree trace if needed.
1504 * Also We don't care about the error, as it's handled internally.
1506 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1507 ret = __btrfs_cow_block(trans, root, buf, parent,
1508 parent_slot, cow_ret, search_start, 0);
1510 trace_btrfs_cow_block(root, buf, *cow_ret);
1516 * helper function for defrag to decide if two blocks pointed to by a
1517 * node are actually close by
1519 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1521 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1523 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1529 * compare two keys in a memcmp fashion
1531 static int comp_keys(const struct btrfs_disk_key *disk,
1532 const struct btrfs_key *k2)
1534 struct btrfs_key k1;
1536 btrfs_disk_key_to_cpu(&k1, disk);
1538 return btrfs_comp_cpu_keys(&k1, k2);
1542 * same as comp_keys only with two btrfs_key's
1544 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1546 if (k1->objectid > k2->objectid)
1548 if (k1->objectid < k2->objectid)
1550 if (k1->type > k2->type)
1552 if (k1->type < k2->type)
1554 if (k1->offset > k2->offset)
1556 if (k1->offset < k2->offset)
1562 * this is used by the defrag code to go through all the
1563 * leaves pointed to by a node and reallocate them so that
1564 * disk order is close to key order
1566 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1567 struct btrfs_root *root, struct extent_buffer *parent,
1568 int start_slot, u64 *last_ret,
1569 struct btrfs_key *progress)
1571 struct btrfs_fs_info *fs_info = root->fs_info;
1572 struct extent_buffer *cur;
1575 u64 search_start = *last_ret;
1585 int progress_passed = 0;
1586 struct btrfs_disk_key disk_key;
1588 parent_level = btrfs_header_level(parent);
1590 WARN_ON(trans->transaction != fs_info->running_transaction);
1591 WARN_ON(trans->transid != fs_info->generation);
1593 parent_nritems = btrfs_header_nritems(parent);
1594 blocksize = fs_info->nodesize;
1595 end_slot = parent_nritems - 1;
1597 if (parent_nritems <= 1)
1600 btrfs_set_lock_blocking_write(parent);
1602 for (i = start_slot; i <= end_slot; i++) {
1603 struct btrfs_key first_key;
1606 btrfs_node_key(parent, &disk_key, i);
1607 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1610 progress_passed = 1;
1611 blocknr = btrfs_node_blockptr(parent, i);
1612 gen = btrfs_node_ptr_generation(parent, i);
1613 btrfs_node_key_to_cpu(parent, &first_key, i);
1614 if (last_block == 0)
1615 last_block = blocknr;
1618 other = btrfs_node_blockptr(parent, i - 1);
1619 close = close_blocks(blocknr, other, blocksize);
1621 if (!close && i < end_slot) {
1622 other = btrfs_node_blockptr(parent, i + 1);
1623 close = close_blocks(blocknr, other, blocksize);
1626 last_block = blocknr;
1630 cur = find_extent_buffer(fs_info, blocknr);
1632 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1635 if (!cur || !uptodate) {
1637 cur = read_tree_block(fs_info, blocknr, gen,
1641 return PTR_ERR(cur);
1642 } else if (!extent_buffer_uptodate(cur)) {
1643 free_extent_buffer(cur);
1646 } else if (!uptodate) {
1647 err = btrfs_read_buffer(cur, gen,
1648 parent_level - 1,&first_key);
1650 free_extent_buffer(cur);
1655 if (search_start == 0)
1656 search_start = last_block;
1658 btrfs_tree_lock(cur);
1659 btrfs_set_lock_blocking_write(cur);
1660 err = __btrfs_cow_block(trans, root, cur, parent, i,
1663 (end_slot - i) * blocksize));
1665 btrfs_tree_unlock(cur);
1666 free_extent_buffer(cur);
1669 search_start = cur->start;
1670 last_block = cur->start;
1671 *last_ret = search_start;
1672 btrfs_tree_unlock(cur);
1673 free_extent_buffer(cur);
1679 * search for key in the extent_buffer. The items start at offset p,
1680 * and they are item_size apart. There are 'max' items in p.
1682 * the slot in the array is returned via slot, and it points to
1683 * the place where you would insert key if it is not found in
1686 * slot may point to max if the key is bigger than all of the keys
1688 static noinline int generic_bin_search(struct extent_buffer *eb,
1689 unsigned long p, int item_size,
1690 const struct btrfs_key *key,
1697 struct btrfs_disk_key *tmp = NULL;
1698 struct btrfs_disk_key unaligned;
1699 unsigned long offset;
1701 unsigned long map_start = 0;
1702 unsigned long map_len = 0;
1706 btrfs_err(eb->fs_info,
1707 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1708 __func__, low, high, eb->start,
1709 btrfs_header_owner(eb), btrfs_header_level(eb));
1713 while (low < high) {
1714 mid = (low + high) / 2;
1715 offset = p + mid * item_size;
1717 if (!kaddr || offset < map_start ||
1718 (offset + sizeof(struct btrfs_disk_key)) >
1719 map_start + map_len) {
1721 err = map_private_extent_buffer(eb, offset,
1722 sizeof(struct btrfs_disk_key),
1723 &kaddr, &map_start, &map_len);
1726 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1728 } else if (err == 1) {
1729 read_extent_buffer(eb, &unaligned,
1730 offset, sizeof(unaligned));
1737 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1740 ret = comp_keys(tmp, key);
1756 * simple bin_search frontend that does the right thing for
1759 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1760 int level, int *slot)
1763 return generic_bin_search(eb,
1764 offsetof(struct btrfs_leaf, items),
1765 sizeof(struct btrfs_item),
1766 key, btrfs_header_nritems(eb),
1769 return generic_bin_search(eb,
1770 offsetof(struct btrfs_node, ptrs),
1771 sizeof(struct btrfs_key_ptr),
1772 key, btrfs_header_nritems(eb),
1776 static void root_add_used(struct btrfs_root *root, u32 size)
1778 spin_lock(&root->accounting_lock);
1779 btrfs_set_root_used(&root->root_item,
1780 btrfs_root_used(&root->root_item) + size);
1781 spin_unlock(&root->accounting_lock);
1784 static void root_sub_used(struct btrfs_root *root, u32 size)
1786 spin_lock(&root->accounting_lock);
1787 btrfs_set_root_used(&root->root_item,
1788 btrfs_root_used(&root->root_item) - size);
1789 spin_unlock(&root->accounting_lock);
1792 /* given a node and slot number, this reads the blocks it points to. The
1793 * extent buffer is returned with a reference taken (but unlocked).
1795 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1798 int level = btrfs_header_level(parent);
1799 struct extent_buffer *eb;
1800 struct btrfs_key first_key;
1802 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1803 return ERR_PTR(-ENOENT);
1807 btrfs_node_key_to_cpu(parent, &first_key, slot);
1808 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1809 btrfs_node_ptr_generation(parent, slot),
1810 level - 1, &first_key);
1811 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1812 free_extent_buffer(eb);
1820 * node level balancing, used to make sure nodes are in proper order for
1821 * item deletion. We balance from the top down, so we have to make sure
1822 * that a deletion won't leave an node completely empty later on.
1824 static noinline int balance_level(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path, int level)
1828 struct btrfs_fs_info *fs_info = root->fs_info;
1829 struct extent_buffer *right = NULL;
1830 struct extent_buffer *mid;
1831 struct extent_buffer *left = NULL;
1832 struct extent_buffer *parent = NULL;
1836 int orig_slot = path->slots[level];
1841 mid = path->nodes[level];
1843 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1844 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1845 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1847 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1849 if (level < BTRFS_MAX_LEVEL - 1) {
1850 parent = path->nodes[level + 1];
1851 pslot = path->slots[level + 1];
1855 * deal with the case where there is only one pointer in the root
1856 * by promoting the node below to a root
1859 struct extent_buffer *child;
1861 if (btrfs_header_nritems(mid) != 1)
1864 /* promote the child to a root */
1865 child = btrfs_read_node_slot(mid, 0);
1866 if (IS_ERR(child)) {
1867 ret = PTR_ERR(child);
1868 btrfs_handle_fs_error(fs_info, ret, NULL);
1872 btrfs_tree_lock(child);
1873 btrfs_set_lock_blocking_write(child);
1874 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1876 btrfs_tree_unlock(child);
1877 free_extent_buffer(child);
1881 ret = tree_mod_log_insert_root(root->node, child, 1);
1883 rcu_assign_pointer(root->node, child);
1885 add_root_to_dirty_list(root);
1886 btrfs_tree_unlock(child);
1888 path->locks[level] = 0;
1889 path->nodes[level] = NULL;
1890 btrfs_clean_tree_block(mid);
1891 btrfs_tree_unlock(mid);
1892 /* once for the path */
1893 free_extent_buffer(mid);
1895 root_sub_used(root, mid->len);
1896 btrfs_free_tree_block(trans, root, mid, 0, 1);
1897 /* once for the root ptr */
1898 free_extent_buffer_stale(mid);
1901 if (btrfs_header_nritems(mid) >
1902 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1905 left = btrfs_read_node_slot(parent, pslot - 1);
1910 btrfs_tree_lock(left);
1911 btrfs_set_lock_blocking_write(left);
1912 wret = btrfs_cow_block(trans, root, left,
1913 parent, pslot - 1, &left);
1920 right = btrfs_read_node_slot(parent, pslot + 1);
1925 btrfs_tree_lock(right);
1926 btrfs_set_lock_blocking_write(right);
1927 wret = btrfs_cow_block(trans, root, right,
1928 parent, pslot + 1, &right);
1935 /* first, try to make some room in the middle buffer */
1937 orig_slot += btrfs_header_nritems(left);
1938 wret = push_node_left(trans, left, mid, 1);
1944 * then try to empty the right most buffer into the middle
1947 wret = push_node_left(trans, mid, right, 1);
1948 if (wret < 0 && wret != -ENOSPC)
1950 if (btrfs_header_nritems(right) == 0) {
1951 btrfs_clean_tree_block(right);
1952 btrfs_tree_unlock(right);
1953 del_ptr(root, path, level + 1, pslot + 1);
1954 root_sub_used(root, right->len);
1955 btrfs_free_tree_block(trans, root, right, 0, 1);
1956 free_extent_buffer_stale(right);
1959 struct btrfs_disk_key right_key;
1960 btrfs_node_key(right, &right_key, 0);
1961 ret = tree_mod_log_insert_key(parent, pslot + 1,
1962 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1964 btrfs_set_node_key(parent, &right_key, pslot + 1);
1965 btrfs_mark_buffer_dirty(parent);
1968 if (btrfs_header_nritems(mid) == 1) {
1970 * we're not allowed to leave a node with one item in the
1971 * tree during a delete. A deletion from lower in the tree
1972 * could try to delete the only pointer in this node.
1973 * So, pull some keys from the left.
1974 * There has to be a left pointer at this point because
1975 * otherwise we would have pulled some pointers from the
1980 btrfs_handle_fs_error(fs_info, ret, NULL);
1983 wret = balance_node_right(trans, mid, left);
1989 wret = push_node_left(trans, left, mid, 1);
1995 if (btrfs_header_nritems(mid) == 0) {
1996 btrfs_clean_tree_block(mid);
1997 btrfs_tree_unlock(mid);
1998 del_ptr(root, path, level + 1, pslot);
1999 root_sub_used(root, mid->len);
2000 btrfs_free_tree_block(trans, root, mid, 0, 1);
2001 free_extent_buffer_stale(mid);
2004 /* update the parent key to reflect our changes */
2005 struct btrfs_disk_key mid_key;
2006 btrfs_node_key(mid, &mid_key, 0);
2007 ret = tree_mod_log_insert_key(parent, pslot,
2008 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2010 btrfs_set_node_key(parent, &mid_key, pslot);
2011 btrfs_mark_buffer_dirty(parent);
2014 /* update the path */
2016 if (btrfs_header_nritems(left) > orig_slot) {
2017 extent_buffer_get(left);
2018 /* left was locked after cow */
2019 path->nodes[level] = left;
2020 path->slots[level + 1] -= 1;
2021 path->slots[level] = orig_slot;
2023 btrfs_tree_unlock(mid);
2024 free_extent_buffer(mid);
2027 orig_slot -= btrfs_header_nritems(left);
2028 path->slots[level] = orig_slot;
2031 /* double check we haven't messed things up */
2033 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2037 btrfs_tree_unlock(right);
2038 free_extent_buffer(right);
2041 if (path->nodes[level] != left)
2042 btrfs_tree_unlock(left);
2043 free_extent_buffer(left);
2048 /* Node balancing for insertion. Here we only split or push nodes around
2049 * when they are completely full. This is also done top down, so we
2050 * have to be pessimistic.
2052 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2053 struct btrfs_root *root,
2054 struct btrfs_path *path, int level)
2056 struct btrfs_fs_info *fs_info = root->fs_info;
2057 struct extent_buffer *right = NULL;
2058 struct extent_buffer *mid;
2059 struct extent_buffer *left = NULL;
2060 struct extent_buffer *parent = NULL;
2064 int orig_slot = path->slots[level];
2069 mid = path->nodes[level];
2070 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2072 if (level < BTRFS_MAX_LEVEL - 1) {
2073 parent = path->nodes[level + 1];
2074 pslot = path->slots[level + 1];
2080 left = btrfs_read_node_slot(parent, pslot - 1);
2084 /* first, try to make some room in the middle buffer */
2088 btrfs_tree_lock(left);
2089 btrfs_set_lock_blocking_write(left);
2091 left_nr = btrfs_header_nritems(left);
2092 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2095 ret = btrfs_cow_block(trans, root, left, parent,
2100 wret = push_node_left(trans, left, mid, 0);
2106 struct btrfs_disk_key disk_key;
2107 orig_slot += left_nr;
2108 btrfs_node_key(mid, &disk_key, 0);
2109 ret = tree_mod_log_insert_key(parent, pslot,
2110 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2112 btrfs_set_node_key(parent, &disk_key, pslot);
2113 btrfs_mark_buffer_dirty(parent);
2114 if (btrfs_header_nritems(left) > orig_slot) {
2115 path->nodes[level] = left;
2116 path->slots[level + 1] -= 1;
2117 path->slots[level] = orig_slot;
2118 btrfs_tree_unlock(mid);
2119 free_extent_buffer(mid);
2122 btrfs_header_nritems(left);
2123 path->slots[level] = orig_slot;
2124 btrfs_tree_unlock(left);
2125 free_extent_buffer(left);
2129 btrfs_tree_unlock(left);
2130 free_extent_buffer(left);
2132 right = btrfs_read_node_slot(parent, pslot + 1);
2137 * then try to empty the right most buffer into the middle
2142 btrfs_tree_lock(right);
2143 btrfs_set_lock_blocking_write(right);
2145 right_nr = btrfs_header_nritems(right);
2146 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2149 ret = btrfs_cow_block(trans, root, right,
2155 wret = balance_node_right(trans, right, mid);
2161 struct btrfs_disk_key disk_key;
2163 btrfs_node_key(right, &disk_key, 0);
2164 ret = tree_mod_log_insert_key(parent, pslot + 1,
2165 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2167 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2168 btrfs_mark_buffer_dirty(parent);
2170 if (btrfs_header_nritems(mid) <= orig_slot) {
2171 path->nodes[level] = right;
2172 path->slots[level + 1] += 1;
2173 path->slots[level] = orig_slot -
2174 btrfs_header_nritems(mid);
2175 btrfs_tree_unlock(mid);
2176 free_extent_buffer(mid);
2178 btrfs_tree_unlock(right);
2179 free_extent_buffer(right);
2183 btrfs_tree_unlock(right);
2184 free_extent_buffer(right);
2190 * readahead one full node of leaves, finding things that are close
2191 * to the block in 'slot', and triggering ra on them.
2193 static void reada_for_search(struct btrfs_fs_info *fs_info,
2194 struct btrfs_path *path,
2195 int level, int slot, u64 objectid)
2197 struct extent_buffer *node;
2198 struct btrfs_disk_key disk_key;
2203 struct extent_buffer *eb;
2211 if (!path->nodes[level])
2214 node = path->nodes[level];
2216 search = btrfs_node_blockptr(node, slot);
2217 blocksize = fs_info->nodesize;
2218 eb = find_extent_buffer(fs_info, search);
2220 free_extent_buffer(eb);
2226 nritems = btrfs_header_nritems(node);
2230 if (path->reada == READA_BACK) {
2234 } else if (path->reada == READA_FORWARD) {
2239 if (path->reada == READA_BACK && objectid) {
2240 btrfs_node_key(node, &disk_key, nr);
2241 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2244 search = btrfs_node_blockptr(node, nr);
2245 if ((search <= target && target - search <= 65536) ||
2246 (search > target && search - target <= 65536)) {
2247 readahead_tree_block(fs_info, search);
2251 if ((nread > 65536 || nscan > 32))
2256 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2257 struct btrfs_path *path, int level)
2261 struct extent_buffer *parent;
2262 struct extent_buffer *eb;
2267 parent = path->nodes[level + 1];
2271 nritems = btrfs_header_nritems(parent);
2272 slot = path->slots[level + 1];
2275 block1 = btrfs_node_blockptr(parent, slot - 1);
2276 gen = btrfs_node_ptr_generation(parent, slot - 1);
2277 eb = find_extent_buffer(fs_info, block1);
2279 * if we get -eagain from btrfs_buffer_uptodate, we
2280 * don't want to return eagain here. That will loop
2283 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2285 free_extent_buffer(eb);
2287 if (slot + 1 < nritems) {
2288 block2 = btrfs_node_blockptr(parent, slot + 1);
2289 gen = btrfs_node_ptr_generation(parent, slot + 1);
2290 eb = find_extent_buffer(fs_info, block2);
2291 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2293 free_extent_buffer(eb);
2297 readahead_tree_block(fs_info, block1);
2299 readahead_tree_block(fs_info, block2);
2304 * when we walk down the tree, it is usually safe to unlock the higher layers
2305 * in the tree. The exceptions are when our path goes through slot 0, because
2306 * operations on the tree might require changing key pointers higher up in the
2309 * callers might also have set path->keep_locks, which tells this code to keep
2310 * the lock if the path points to the last slot in the block. This is part of
2311 * walking through the tree, and selecting the next slot in the higher block.
2313 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2314 * if lowest_unlock is 1, level 0 won't be unlocked
2316 static noinline void unlock_up(struct btrfs_path *path, int level,
2317 int lowest_unlock, int min_write_lock_level,
2318 int *write_lock_level)
2321 int skip_level = level;
2323 struct extent_buffer *t;
2325 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2326 if (!path->nodes[i])
2328 if (!path->locks[i])
2330 if (!no_skips && path->slots[i] == 0) {
2334 if (!no_skips && path->keep_locks) {
2337 nritems = btrfs_header_nritems(t);
2338 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2343 if (skip_level < i && i >= lowest_unlock)
2347 if (i >= lowest_unlock && i > skip_level) {
2348 btrfs_tree_unlock_rw(t, path->locks[i]);
2350 if (write_lock_level &&
2351 i > min_write_lock_level &&
2352 i <= *write_lock_level) {
2353 *write_lock_level = i - 1;
2360 * This releases any locks held in the path starting at level and
2361 * going all the way up to the root.
2363 * btrfs_search_slot will keep the lock held on higher nodes in a few
2364 * corner cases, such as COW of the block at slot zero in the node. This
2365 * ignores those rules, and it should only be called when there are no
2366 * more updates to be done higher up in the tree.
2368 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2372 if (path->keep_locks)
2375 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2376 if (!path->nodes[i])
2378 if (!path->locks[i])
2380 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2386 * helper function for btrfs_search_slot. The goal is to find a block
2387 * in cache without setting the path to blocking. If we find the block
2388 * we return zero and the path is unchanged.
2390 * If we can't find the block, we set the path blocking and do some
2391 * reada. -EAGAIN is returned and the search must be repeated.
2394 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2395 struct extent_buffer **eb_ret, int level, int slot,
2396 const struct btrfs_key *key)
2398 struct btrfs_fs_info *fs_info = root->fs_info;
2401 struct extent_buffer *b = *eb_ret;
2402 struct extent_buffer *tmp;
2403 struct btrfs_key first_key;
2407 blocknr = btrfs_node_blockptr(b, slot);
2408 gen = btrfs_node_ptr_generation(b, slot);
2409 parent_level = btrfs_header_level(b);
2410 btrfs_node_key_to_cpu(b, &first_key, slot);
2412 tmp = find_extent_buffer(fs_info, blocknr);
2414 /* first we do an atomic uptodate check */
2415 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2417 * Do extra check for first_key, eb can be stale due to
2418 * being cached, read from scrub, or have multiple
2419 * parents (shared tree blocks).
2421 if (btrfs_verify_level_key(tmp,
2422 parent_level - 1, &first_key, gen)) {
2423 free_extent_buffer(tmp);
2430 /* the pages were up to date, but we failed
2431 * the generation number check. Do a full
2432 * read for the generation number that is correct.
2433 * We must do this without dropping locks so
2434 * we can trust our generation number
2436 btrfs_set_path_blocking(p);
2438 /* now we're allowed to do a blocking uptodate check */
2439 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2444 free_extent_buffer(tmp);
2445 btrfs_release_path(p);
2450 * reduce lock contention at high levels
2451 * of the btree by dropping locks before
2452 * we read. Don't release the lock on the current
2453 * level because we need to walk this node to figure
2454 * out which blocks to read.
2456 btrfs_unlock_up_safe(p, level + 1);
2457 btrfs_set_path_blocking(p);
2459 if (p->reada != READA_NONE)
2460 reada_for_search(fs_info, p, level, slot, key->objectid);
2463 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2467 * If the read above didn't mark this buffer up to date,
2468 * it will never end up being up to date. Set ret to EIO now
2469 * and give up so that our caller doesn't loop forever
2472 if (!extent_buffer_uptodate(tmp))
2474 free_extent_buffer(tmp);
2479 btrfs_release_path(p);
2484 * helper function for btrfs_search_slot. This does all of the checks
2485 * for node-level blocks and does any balancing required based on
2488 * If no extra work was required, zero is returned. If we had to
2489 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2493 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2494 struct btrfs_root *root, struct btrfs_path *p,
2495 struct extent_buffer *b, int level, int ins_len,
2496 int *write_lock_level)
2498 struct btrfs_fs_info *fs_info = root->fs_info;
2501 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2502 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2505 if (*write_lock_level < level + 1) {
2506 *write_lock_level = level + 1;
2507 btrfs_release_path(p);
2511 btrfs_set_path_blocking(p);
2512 reada_for_balance(fs_info, p, level);
2513 sret = split_node(trans, root, p, level);
2520 b = p->nodes[level];
2521 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2522 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2525 if (*write_lock_level < level + 1) {
2526 *write_lock_level = level + 1;
2527 btrfs_release_path(p);
2531 btrfs_set_path_blocking(p);
2532 reada_for_balance(fs_info, p, level);
2533 sret = balance_level(trans, root, p, level);
2539 b = p->nodes[level];
2541 btrfs_release_path(p);
2544 BUG_ON(btrfs_header_nritems(b) == 1);
2554 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2555 int level, int *prev_cmp, int *slot)
2557 if (*prev_cmp != 0) {
2558 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2567 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2568 u64 iobjectid, u64 ioff, u8 key_type,
2569 struct btrfs_key *found_key)
2572 struct btrfs_key key;
2573 struct extent_buffer *eb;
2578 key.type = key_type;
2579 key.objectid = iobjectid;
2582 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2586 eb = path->nodes[0];
2587 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2588 ret = btrfs_next_leaf(fs_root, path);
2591 eb = path->nodes[0];
2594 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2595 if (found_key->type != key.type ||
2596 found_key->objectid != key.objectid)
2602 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2603 struct btrfs_path *p,
2604 int write_lock_level)
2606 struct btrfs_fs_info *fs_info = root->fs_info;
2607 struct extent_buffer *b;
2611 /* We try very hard to do read locks on the root */
2612 root_lock = BTRFS_READ_LOCK;
2614 if (p->search_commit_root) {
2616 * The commit roots are read only so we always do read locks,
2617 * and we always must hold the commit_root_sem when doing
2618 * searches on them, the only exception is send where we don't
2619 * want to block transaction commits for a long time, so
2620 * we need to clone the commit root in order to avoid races
2621 * with transaction commits that create a snapshot of one of
2622 * the roots used by a send operation.
2624 if (p->need_commit_sem) {
2625 down_read(&fs_info->commit_root_sem);
2626 b = btrfs_clone_extent_buffer(root->commit_root);
2627 up_read(&fs_info->commit_root_sem);
2629 return ERR_PTR(-ENOMEM);
2632 b = root->commit_root;
2633 extent_buffer_get(b);
2635 level = btrfs_header_level(b);
2637 * Ensure that all callers have set skip_locking when
2638 * p->search_commit_root = 1.
2640 ASSERT(p->skip_locking == 1);
2645 if (p->skip_locking) {
2646 b = btrfs_root_node(root);
2647 level = btrfs_header_level(b);
2652 * If the level is set to maximum, we can skip trying to get the read
2655 if (write_lock_level < BTRFS_MAX_LEVEL) {
2657 * We don't know the level of the root node until we actually
2658 * have it read locked
2660 b = btrfs_read_lock_root_node(root);
2661 level = btrfs_header_level(b);
2662 if (level > write_lock_level)
2665 /* Whoops, must trade for write lock */
2666 btrfs_tree_read_unlock(b);
2667 free_extent_buffer(b);
2670 b = btrfs_lock_root_node(root);
2671 root_lock = BTRFS_WRITE_LOCK;
2673 /* The level might have changed, check again */
2674 level = btrfs_header_level(b);
2677 p->nodes[level] = b;
2678 if (!p->skip_locking)
2679 p->locks[level] = root_lock;
2681 * Callers are responsible for dropping b's references.
2688 * btrfs_search_slot - look for a key in a tree and perform necessary
2689 * modifications to preserve tree invariants.
2691 * @trans: Handle of transaction, used when modifying the tree
2692 * @p: Holds all btree nodes along the search path
2693 * @root: The root node of the tree
2694 * @key: The key we are looking for
2695 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2696 * deletions it's -1. 0 for plain searches
2697 * @cow: boolean should CoW operations be performed. Must always be 1
2698 * when modifying the tree.
2700 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2701 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2703 * If @key is found, 0 is returned and you can find the item in the leaf level
2704 * of the path (level 0)
2706 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2707 * points to the slot where it should be inserted
2709 * If an error is encountered while searching the tree a negative error number
2712 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2713 const struct btrfs_key *key, struct btrfs_path *p,
2714 int ins_len, int cow)
2716 struct extent_buffer *b;
2721 int lowest_unlock = 1;
2722 /* everything at write_lock_level or lower must be write locked */
2723 int write_lock_level = 0;
2724 u8 lowest_level = 0;
2725 int min_write_lock_level;
2728 lowest_level = p->lowest_level;
2729 WARN_ON(lowest_level && ins_len > 0);
2730 WARN_ON(p->nodes[0] != NULL);
2731 BUG_ON(!cow && ins_len);
2736 /* when we are removing items, we might have to go up to level
2737 * two as we update tree pointers Make sure we keep write
2738 * for those levels as well
2740 write_lock_level = 2;
2741 } else if (ins_len > 0) {
2743 * for inserting items, make sure we have a write lock on
2744 * level 1 so we can update keys
2746 write_lock_level = 1;
2750 write_lock_level = -1;
2752 if (cow && (p->keep_locks || p->lowest_level))
2753 write_lock_level = BTRFS_MAX_LEVEL;
2755 min_write_lock_level = write_lock_level;
2759 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2766 level = btrfs_header_level(b);
2769 * setup the path here so we can release it under lock
2770 * contention with the cow code
2773 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2776 * if we don't really need to cow this block
2777 * then we don't want to set the path blocking,
2778 * so we test it here
2780 if (!should_cow_block(trans, root, b)) {
2781 trans->dirty = true;
2786 * must have write locks on this node and the
2789 if (level > write_lock_level ||
2790 (level + 1 > write_lock_level &&
2791 level + 1 < BTRFS_MAX_LEVEL &&
2792 p->nodes[level + 1])) {
2793 write_lock_level = level + 1;
2794 btrfs_release_path(p);
2798 btrfs_set_path_blocking(p);
2800 err = btrfs_cow_block(trans, root, b, NULL, 0,
2803 err = btrfs_cow_block(trans, root, b,
2804 p->nodes[level + 1],
2805 p->slots[level + 1], &b);
2812 p->nodes[level] = b;
2814 * Leave path with blocking locks to avoid massive
2815 * lock context switch, this is made on purpose.
2819 * we have a lock on b and as long as we aren't changing
2820 * the tree, there is no way to for the items in b to change.
2821 * It is safe to drop the lock on our parent before we
2822 * go through the expensive btree search on b.
2824 * If we're inserting or deleting (ins_len != 0), then we might
2825 * be changing slot zero, which may require changing the parent.
2826 * So, we can't drop the lock until after we know which slot
2827 * we're operating on.
2829 if (!ins_len && !p->keep_locks) {
2832 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2833 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2838 ret = key_search(b, key, level, &prev_cmp, &slot);
2844 if (ret && slot > 0) {
2848 p->slots[level] = slot;
2849 err = setup_nodes_for_search(trans, root, p, b, level,
2850 ins_len, &write_lock_level);
2857 b = p->nodes[level];
2858 slot = p->slots[level];
2861 * slot 0 is special, if we change the key
2862 * we have to update the parent pointer
2863 * which means we must have a write lock
2866 if (slot == 0 && ins_len &&
2867 write_lock_level < level + 1) {
2868 write_lock_level = level + 1;
2869 btrfs_release_path(p);
2873 unlock_up(p, level, lowest_unlock,
2874 min_write_lock_level, &write_lock_level);
2876 if (level == lowest_level) {
2882 err = read_block_for_search(root, p, &b, level,
2891 if (!p->skip_locking) {
2892 level = btrfs_header_level(b);
2893 if (level <= write_lock_level) {
2894 err = btrfs_try_tree_write_lock(b);
2896 btrfs_set_path_blocking(p);
2899 p->locks[level] = BTRFS_WRITE_LOCK;
2901 err = btrfs_tree_read_lock_atomic(b);
2903 btrfs_set_path_blocking(p);
2904 btrfs_tree_read_lock(b);
2906 p->locks[level] = BTRFS_READ_LOCK;
2908 p->nodes[level] = b;
2911 p->slots[level] = slot;
2913 btrfs_leaf_free_space(b) < ins_len) {
2914 if (write_lock_level < 1) {
2915 write_lock_level = 1;
2916 btrfs_release_path(p);
2920 btrfs_set_path_blocking(p);
2921 err = split_leaf(trans, root, key,
2922 p, ins_len, ret == 0);
2930 if (!p->search_for_split)
2931 unlock_up(p, level, lowest_unlock,
2932 min_write_lock_level, NULL);
2939 * we don't really know what they plan on doing with the path
2940 * from here on, so for now just mark it as blocking
2942 if (!p->leave_spinning)
2943 btrfs_set_path_blocking(p);
2944 if (ret < 0 && !p->skip_release_on_error)
2945 btrfs_release_path(p);
2950 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2951 * current state of the tree together with the operations recorded in the tree
2952 * modification log to search for the key in a previous version of this tree, as
2953 * denoted by the time_seq parameter.
2955 * Naturally, there is no support for insert, delete or cow operations.
2957 * The resulting path and return value will be set up as if we called
2958 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2960 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2961 struct btrfs_path *p, u64 time_seq)
2963 struct btrfs_fs_info *fs_info = root->fs_info;
2964 struct extent_buffer *b;
2969 int lowest_unlock = 1;
2970 u8 lowest_level = 0;
2973 lowest_level = p->lowest_level;
2974 WARN_ON(p->nodes[0] != NULL);
2976 if (p->search_commit_root) {
2978 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2982 b = get_old_root(root, time_seq);
2987 level = btrfs_header_level(b);
2988 p->locks[level] = BTRFS_READ_LOCK;
2991 level = btrfs_header_level(b);
2992 p->nodes[level] = b;
2995 * we have a lock on b and as long as we aren't changing
2996 * the tree, there is no way to for the items in b to change.
2997 * It is safe to drop the lock on our parent before we
2998 * go through the expensive btree search on b.
3000 btrfs_unlock_up_safe(p, level + 1);
3003 * Since we can unwind ebs we want to do a real search every
3007 ret = key_search(b, key, level, &prev_cmp, &slot);
3013 if (ret && slot > 0) {
3017 p->slots[level] = slot;
3018 unlock_up(p, level, lowest_unlock, 0, NULL);
3020 if (level == lowest_level) {
3026 err = read_block_for_search(root, p, &b, level,
3035 level = btrfs_header_level(b);
3036 err = btrfs_tree_read_lock_atomic(b);
3038 btrfs_set_path_blocking(p);
3039 btrfs_tree_read_lock(b);
3041 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3046 p->locks[level] = BTRFS_READ_LOCK;
3047 p->nodes[level] = b;
3049 p->slots[level] = slot;
3050 unlock_up(p, level, lowest_unlock, 0, NULL);
3056 if (!p->leave_spinning)
3057 btrfs_set_path_blocking(p);
3059 btrfs_release_path(p);
3065 * helper to use instead of search slot if no exact match is needed but
3066 * instead the next or previous item should be returned.
3067 * When find_higher is true, the next higher item is returned, the next lower
3069 * When return_any and find_higher are both true, and no higher item is found,
3070 * return the next lower instead.
3071 * When return_any is true and find_higher is false, and no lower item is found,
3072 * return the next higher instead.
3073 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3076 int btrfs_search_slot_for_read(struct btrfs_root *root,
3077 const struct btrfs_key *key,
3078 struct btrfs_path *p, int find_higher,
3082 struct extent_buffer *leaf;
3085 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3089 * a return value of 1 means the path is at the position where the
3090 * item should be inserted. Normally this is the next bigger item,
3091 * but in case the previous item is the last in a leaf, path points
3092 * to the first free slot in the previous leaf, i.e. at an invalid
3098 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3099 ret = btrfs_next_leaf(root, p);
3105 * no higher item found, return the next
3110 btrfs_release_path(p);
3114 if (p->slots[0] == 0) {
3115 ret = btrfs_prev_leaf(root, p);
3120 if (p->slots[0] == btrfs_header_nritems(leaf))
3127 * no lower item found, return the next
3132 btrfs_release_path(p);
3142 * adjust the pointers going up the tree, starting at level
3143 * making sure the right key of each node is points to 'key'.
3144 * This is used after shifting pointers to the left, so it stops
3145 * fixing up pointers when a given leaf/node is not in slot 0 of the
3149 static void fixup_low_keys(struct btrfs_path *path,
3150 struct btrfs_disk_key *key, int level)
3153 struct extent_buffer *t;
3156 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3157 int tslot = path->slots[i];
3159 if (!path->nodes[i])
3162 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3165 btrfs_set_node_key(t, key, tslot);
3166 btrfs_mark_buffer_dirty(path->nodes[i]);
3175 * This function isn't completely safe. It's the caller's responsibility
3176 * that the new key won't break the order
3178 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3179 struct btrfs_path *path,
3180 const struct btrfs_key *new_key)
3182 struct btrfs_disk_key disk_key;
3183 struct extent_buffer *eb;
3186 eb = path->nodes[0];
3187 slot = path->slots[0];
3189 btrfs_item_key(eb, &disk_key, slot - 1);
3190 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
3192 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3193 slot, btrfs_disk_key_objectid(&disk_key),
3194 btrfs_disk_key_type(&disk_key),
3195 btrfs_disk_key_offset(&disk_key),
3196 new_key->objectid, new_key->type,
3198 btrfs_print_leaf(eb);
3202 if (slot < btrfs_header_nritems(eb) - 1) {
3203 btrfs_item_key(eb, &disk_key, slot + 1);
3204 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
3206 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3207 slot, btrfs_disk_key_objectid(&disk_key),
3208 btrfs_disk_key_type(&disk_key),
3209 btrfs_disk_key_offset(&disk_key),
3210 new_key->objectid, new_key->type,
3212 btrfs_print_leaf(eb);
3217 btrfs_cpu_key_to_disk(&disk_key, new_key);
3218 btrfs_set_item_key(eb, &disk_key, slot);
3219 btrfs_mark_buffer_dirty(eb);
3221 fixup_low_keys(path, &disk_key, 1);
3225 * try to push data from one node into the next node left in the
3228 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3229 * error, and > 0 if there was no room in the left hand block.
3231 static int push_node_left(struct btrfs_trans_handle *trans,
3232 struct extent_buffer *dst,
3233 struct extent_buffer *src, int empty)
3235 struct btrfs_fs_info *fs_info = trans->fs_info;
3241 src_nritems = btrfs_header_nritems(src);
3242 dst_nritems = btrfs_header_nritems(dst);
3243 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3244 WARN_ON(btrfs_header_generation(src) != trans->transid);
3245 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3247 if (!empty && src_nritems <= 8)
3250 if (push_items <= 0)
3254 push_items = min(src_nritems, push_items);
3255 if (push_items < src_nritems) {
3256 /* leave at least 8 pointers in the node if
3257 * we aren't going to empty it
3259 if (src_nritems - push_items < 8) {
3260 if (push_items <= 8)
3266 push_items = min(src_nritems - 8, push_items);
3268 ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3270 btrfs_abort_transaction(trans, ret);
3273 copy_extent_buffer(dst, src,
3274 btrfs_node_key_ptr_offset(dst_nritems),
3275 btrfs_node_key_ptr_offset(0),
3276 push_items * sizeof(struct btrfs_key_ptr));
3278 if (push_items < src_nritems) {
3280 * Don't call tree_mod_log_insert_move here, key removal was
3281 * already fully logged by tree_mod_log_eb_copy above.
3283 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3284 btrfs_node_key_ptr_offset(push_items),
3285 (src_nritems - push_items) *
3286 sizeof(struct btrfs_key_ptr));
3288 btrfs_set_header_nritems(src, src_nritems - push_items);
3289 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3290 btrfs_mark_buffer_dirty(src);
3291 btrfs_mark_buffer_dirty(dst);
3297 * try to push data from one node into the next node right in the
3300 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3301 * error, and > 0 if there was no room in the right hand block.
3303 * this will only push up to 1/2 the contents of the left node over
3305 static int balance_node_right(struct btrfs_trans_handle *trans,
3306 struct extent_buffer *dst,
3307 struct extent_buffer *src)
3309 struct btrfs_fs_info *fs_info = trans->fs_info;
3316 WARN_ON(btrfs_header_generation(src) != trans->transid);
3317 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3319 src_nritems = btrfs_header_nritems(src);
3320 dst_nritems = btrfs_header_nritems(dst);
3321 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3322 if (push_items <= 0)
3325 if (src_nritems < 4)
3328 max_push = src_nritems / 2 + 1;
3329 /* don't try to empty the node */
3330 if (max_push >= src_nritems)
3333 if (max_push < push_items)
3334 push_items = max_push;
3336 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3338 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3339 btrfs_node_key_ptr_offset(0),
3341 sizeof(struct btrfs_key_ptr));
3343 ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3346 btrfs_abort_transaction(trans, ret);
3349 copy_extent_buffer(dst, src,
3350 btrfs_node_key_ptr_offset(0),
3351 btrfs_node_key_ptr_offset(src_nritems - push_items),
3352 push_items * sizeof(struct btrfs_key_ptr));
3354 btrfs_set_header_nritems(src, src_nritems - push_items);
3355 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3357 btrfs_mark_buffer_dirty(src);
3358 btrfs_mark_buffer_dirty(dst);
3364 * helper function to insert a new root level in the tree.
3365 * A new node is allocated, and a single item is inserted to
3366 * point to the existing root
3368 * returns zero on success or < 0 on failure.
3370 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *root,
3372 struct btrfs_path *path, int level)
3374 struct btrfs_fs_info *fs_info = root->fs_info;
3376 struct extent_buffer *lower;
3377 struct extent_buffer *c;
3378 struct extent_buffer *old;
3379 struct btrfs_disk_key lower_key;
3382 BUG_ON(path->nodes[level]);
3383 BUG_ON(path->nodes[level-1] != root->node);
3385 lower = path->nodes[level-1];
3387 btrfs_item_key(lower, &lower_key, 0);
3389 btrfs_node_key(lower, &lower_key, 0);
3391 c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3392 root->node->start, 0);
3396 root_add_used(root, fs_info->nodesize);
3398 btrfs_set_header_nritems(c, 1);
3399 btrfs_set_node_key(c, &lower_key, 0);
3400 btrfs_set_node_blockptr(c, 0, lower->start);
3401 lower_gen = btrfs_header_generation(lower);
3402 WARN_ON(lower_gen != trans->transid);
3404 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3406 btrfs_mark_buffer_dirty(c);
3409 ret = tree_mod_log_insert_root(root->node, c, 0);
3411 rcu_assign_pointer(root->node, c);
3413 /* the super has an extra ref to root->node */
3414 free_extent_buffer(old);
3416 add_root_to_dirty_list(root);
3417 extent_buffer_get(c);
3418 path->nodes[level] = c;
3419 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3420 path->slots[level] = 0;
3425 * worker function to insert a single pointer in a node.
3426 * the node should have enough room for the pointer already
3428 * slot and level indicate where you want the key to go, and
3429 * blocknr is the block the key points to.
3431 static void insert_ptr(struct btrfs_trans_handle *trans,
3432 struct btrfs_path *path,
3433 struct btrfs_disk_key *key, u64 bytenr,
3434 int slot, int level)
3436 struct extent_buffer *lower;
3440 BUG_ON(!path->nodes[level]);
3441 btrfs_assert_tree_locked(path->nodes[level]);
3442 lower = path->nodes[level];
3443 nritems = btrfs_header_nritems(lower);
3444 BUG_ON(slot > nritems);
3445 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3446 if (slot != nritems) {
3448 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3452 memmove_extent_buffer(lower,
3453 btrfs_node_key_ptr_offset(slot + 1),
3454 btrfs_node_key_ptr_offset(slot),
3455 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3458 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3462 btrfs_set_node_key(lower, key, slot);
3463 btrfs_set_node_blockptr(lower, slot, bytenr);
3464 WARN_ON(trans->transid == 0);
3465 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3466 btrfs_set_header_nritems(lower, nritems + 1);
3467 btrfs_mark_buffer_dirty(lower);
3471 * split the node at the specified level in path in two.
3472 * The path is corrected to point to the appropriate node after the split
3474 * Before splitting this tries to make some room in the node by pushing
3475 * left and right, if either one works, it returns right away.
3477 * returns 0 on success and < 0 on failure
3479 static noinline int split_node(struct btrfs_trans_handle *trans,
3480 struct btrfs_root *root,
3481 struct btrfs_path *path, int level)
3483 struct btrfs_fs_info *fs_info = root->fs_info;
3484 struct extent_buffer *c;
3485 struct extent_buffer *split;
3486 struct btrfs_disk_key disk_key;
3491 c = path->nodes[level];
3492 WARN_ON(btrfs_header_generation(c) != trans->transid);
3493 if (c == root->node) {
3495 * trying to split the root, lets make a new one
3497 * tree mod log: We don't log_removal old root in
3498 * insert_new_root, because that root buffer will be kept as a
3499 * normal node. We are going to log removal of half of the
3500 * elements below with tree_mod_log_eb_copy. We're holding a
3501 * tree lock on the buffer, which is why we cannot race with
3502 * other tree_mod_log users.
3504 ret = insert_new_root(trans, root, path, level + 1);
3508 ret = push_nodes_for_insert(trans, root, path, level);
3509 c = path->nodes[level];
3510 if (!ret && btrfs_header_nritems(c) <
3511 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3517 c_nritems = btrfs_header_nritems(c);
3518 mid = (c_nritems + 1) / 2;
3519 btrfs_node_key(c, &disk_key, mid);
3521 split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3524 return PTR_ERR(split);
3526 root_add_used(root, fs_info->nodesize);
3527 ASSERT(btrfs_header_level(c) == level);
3529 ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3531 btrfs_abort_transaction(trans, ret);
3534 copy_extent_buffer(split, c,
3535 btrfs_node_key_ptr_offset(0),
3536 btrfs_node_key_ptr_offset(mid),
3537 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3538 btrfs_set_header_nritems(split, c_nritems - mid);
3539 btrfs_set_header_nritems(c, mid);
3542 btrfs_mark_buffer_dirty(c);
3543 btrfs_mark_buffer_dirty(split);
3545 insert_ptr(trans, path, &disk_key, split->start,
3546 path->slots[level + 1] + 1, level + 1);
3548 if (path->slots[level] >= mid) {
3549 path->slots[level] -= mid;
3550 btrfs_tree_unlock(c);
3551 free_extent_buffer(c);
3552 path->nodes[level] = split;
3553 path->slots[level + 1] += 1;
3555 btrfs_tree_unlock(split);
3556 free_extent_buffer(split);
3562 * how many bytes are required to store the items in a leaf. start
3563 * and nr indicate which items in the leaf to check. This totals up the
3564 * space used both by the item structs and the item data
3566 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3568 struct btrfs_item *start_item;
3569 struct btrfs_item *end_item;
3570 struct btrfs_map_token token;
3572 int nritems = btrfs_header_nritems(l);
3573 int end = min(nritems, start + nr) - 1;
3577 btrfs_init_map_token(&token, l);
3578 start_item = btrfs_item_nr(start);
3579 end_item = btrfs_item_nr(end);
3580 data_len = btrfs_token_item_offset(l, start_item, &token) +
3581 btrfs_token_item_size(l, start_item, &token);
3582 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3583 data_len += sizeof(struct btrfs_item) * nr;
3584 WARN_ON(data_len < 0);
3589 * The space between the end of the leaf items and
3590 * the start of the leaf data. IOW, how much room
3591 * the leaf has left for both items and data
3593 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3595 struct btrfs_fs_info *fs_info = leaf->fs_info;
3596 int nritems = btrfs_header_nritems(leaf);
3599 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3602 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3604 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3605 leaf_space_used(leaf, 0, nritems), nritems);
3611 * min slot controls the lowest index we're willing to push to the
3612 * right. We'll push up to and including min_slot, but no lower
3614 static noinline int __push_leaf_right(struct btrfs_path *path,
3615 int data_size, int empty,
3616 struct extent_buffer *right,
3617 int free_space, u32 left_nritems,
3620 struct btrfs_fs_info *fs_info = right->fs_info;
3621 struct extent_buffer *left = path->nodes[0];
3622 struct extent_buffer *upper = path->nodes[1];
3623 struct btrfs_map_token token;
3624 struct btrfs_disk_key disk_key;
3629 struct btrfs_item *item;
3638 nr = max_t(u32, 1, min_slot);
3640 if (path->slots[0] >= left_nritems)
3641 push_space += data_size;
3643 slot = path->slots[1];
3644 i = left_nritems - 1;
3646 item = btrfs_item_nr(i);
3648 if (!empty && push_items > 0) {
3649 if (path->slots[0] > i)
3651 if (path->slots[0] == i) {
3652 int space = btrfs_leaf_free_space(left);
3654 if (space + push_space * 2 > free_space)
3659 if (path->slots[0] == i)
3660 push_space += data_size;
3662 this_item_size = btrfs_item_size(left, item);
3663 if (this_item_size + sizeof(*item) + push_space > free_space)
3667 push_space += this_item_size + sizeof(*item);
3673 if (push_items == 0)
3676 WARN_ON(!empty && push_items == left_nritems);
3678 /* push left to right */
3679 right_nritems = btrfs_header_nritems(right);
3681 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3682 push_space -= leaf_data_end(left);
3684 /* make room in the right data area */
3685 data_end = leaf_data_end(right);
3686 memmove_extent_buffer(right,
3687 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3688 BTRFS_LEAF_DATA_OFFSET + data_end,
3689 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3691 /* copy from the left data area */
3692 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3693 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3694 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3697 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3698 btrfs_item_nr_offset(0),
3699 right_nritems * sizeof(struct btrfs_item));
3701 /* copy the items from left to right */
3702 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3703 btrfs_item_nr_offset(left_nritems - push_items),
3704 push_items * sizeof(struct btrfs_item));
3706 /* update the item pointers */
3707 btrfs_init_map_token(&token, right);
3708 right_nritems += push_items;
3709 btrfs_set_header_nritems(right, right_nritems);
3710 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3711 for (i = 0; i < right_nritems; i++) {
3712 item = btrfs_item_nr(i);
3713 push_space -= btrfs_token_item_size(right, item, &token);
3714 btrfs_set_token_item_offset(right, item, push_space, &token);
3717 left_nritems -= push_items;
3718 btrfs_set_header_nritems(left, left_nritems);
3721 btrfs_mark_buffer_dirty(left);
3723 btrfs_clean_tree_block(left);
3725 btrfs_mark_buffer_dirty(right);
3727 btrfs_item_key(right, &disk_key, 0);
3728 btrfs_set_node_key(upper, &disk_key, slot + 1);
3729 btrfs_mark_buffer_dirty(upper);
3731 /* then fixup the leaf pointer in the path */
3732 if (path->slots[0] >= left_nritems) {
3733 path->slots[0] -= left_nritems;
3734 if (btrfs_header_nritems(path->nodes[0]) == 0)
3735 btrfs_clean_tree_block(path->nodes[0]);
3736 btrfs_tree_unlock(path->nodes[0]);
3737 free_extent_buffer(path->nodes[0]);
3738 path->nodes[0] = right;
3739 path->slots[1] += 1;
3741 btrfs_tree_unlock(right);
3742 free_extent_buffer(right);
3747 btrfs_tree_unlock(right);
3748 free_extent_buffer(right);
3753 * push some data in the path leaf to the right, trying to free up at
3754 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3756 * returns 1 if the push failed because the other node didn't have enough
3757 * room, 0 if everything worked out and < 0 if there were major errors.
3759 * this will push starting from min_slot to the end of the leaf. It won't
3760 * push any slot lower than min_slot
3762 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3763 *root, struct btrfs_path *path,
3764 int min_data_size, int data_size,
3765 int empty, u32 min_slot)
3767 struct extent_buffer *left = path->nodes[0];
3768 struct extent_buffer *right;
3769 struct extent_buffer *upper;
3775 if (!path->nodes[1])
3778 slot = path->slots[1];
3779 upper = path->nodes[1];
3780 if (slot >= btrfs_header_nritems(upper) - 1)
3783 btrfs_assert_tree_locked(path->nodes[1]);
3785 right = btrfs_read_node_slot(upper, slot + 1);
3787 * slot + 1 is not valid or we fail to read the right node,
3788 * no big deal, just return.
3793 btrfs_tree_lock(right);
3794 btrfs_set_lock_blocking_write(right);
3796 free_space = btrfs_leaf_free_space(right);
3797 if (free_space < data_size)
3800 /* cow and double check */
3801 ret = btrfs_cow_block(trans, root, right, upper,
3806 free_space = btrfs_leaf_free_space(right);
3807 if (free_space < data_size)
3810 left_nritems = btrfs_header_nritems(left);
3811 if (left_nritems == 0)
3814 if (path->slots[0] == left_nritems && !empty) {
3815 /* Key greater than all keys in the leaf, right neighbor has
3816 * enough room for it and we're not emptying our leaf to delete
3817 * it, therefore use right neighbor to insert the new item and
3818 * no need to touch/dirty our left leaf. */
3819 btrfs_tree_unlock(left);
3820 free_extent_buffer(left);
3821 path->nodes[0] = right;
3827 return __push_leaf_right(path, min_data_size, empty,
3828 right, free_space, left_nritems, min_slot);
3830 btrfs_tree_unlock(right);
3831 free_extent_buffer(right);
3836 * push some data in the path leaf to the left, trying to free up at
3837 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3839 * max_slot can put a limit on how far into the leaf we'll push items. The
3840 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3843 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3844 int empty, struct extent_buffer *left,
3845 int free_space, u32 right_nritems,
3848 struct btrfs_fs_info *fs_info = left->fs_info;
3849 struct btrfs_disk_key disk_key;
3850 struct extent_buffer *right = path->nodes[0];
3854 struct btrfs_item *item;
3855 u32 old_left_nritems;
3859 u32 old_left_item_size;
3860 struct btrfs_map_token token;
3863 nr = min(right_nritems, max_slot);
3865 nr = min(right_nritems - 1, max_slot);
3867 for (i = 0; i < nr; i++) {
3868 item = btrfs_item_nr(i);
3870 if (!empty && push_items > 0) {
3871 if (path->slots[0] < i)
3873 if (path->slots[0] == i) {
3874 int space = btrfs_leaf_free_space(right);
3876 if (space + push_space * 2 > free_space)
3881 if (path->slots[0] == i)
3882 push_space += data_size;
3884 this_item_size = btrfs_item_size(right, item);
3885 if (this_item_size + sizeof(*item) + push_space > free_space)
3889 push_space += this_item_size + sizeof(*item);
3892 if (push_items == 0) {
3896 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3898 /* push data from right to left */
3899 copy_extent_buffer(left, right,
3900 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3901 btrfs_item_nr_offset(0),
3902 push_items * sizeof(struct btrfs_item));
3904 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3905 btrfs_item_offset_nr(right, push_items - 1);
3907 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3908 leaf_data_end(left) - push_space,
3909 BTRFS_LEAF_DATA_OFFSET +
3910 btrfs_item_offset_nr(right, push_items - 1),
3912 old_left_nritems = btrfs_header_nritems(left);
3913 BUG_ON(old_left_nritems <= 0);
3915 btrfs_init_map_token(&token, left);
3916 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3917 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3920 item = btrfs_item_nr(i);
3922 ioff = btrfs_token_item_offset(left, item, &token);
3923 btrfs_set_token_item_offset(left, item,
3924 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3927 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3929 /* fixup right node */
3930 if (push_items > right_nritems)
3931 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3934 if (push_items < right_nritems) {
3935 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3936 leaf_data_end(right);
3937 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3938 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3939 BTRFS_LEAF_DATA_OFFSET +
3940 leaf_data_end(right), push_space);
3942 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3943 btrfs_item_nr_offset(push_items),
3944 (btrfs_header_nritems(right) - push_items) *
3945 sizeof(struct btrfs_item));
3948 btrfs_init_map_token(&token, right);
3949 right_nritems -= push_items;
3950 btrfs_set_header_nritems(right, right_nritems);
3951 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3952 for (i = 0; i < right_nritems; i++) {
3953 item = btrfs_item_nr(i);
3955 push_space = push_space - btrfs_token_item_size(right,
3957 btrfs_set_token_item_offset(right, item, push_space, &token);
3960 btrfs_mark_buffer_dirty(left);
3962 btrfs_mark_buffer_dirty(right);
3964 btrfs_clean_tree_block(right);
3966 btrfs_item_key(right, &disk_key, 0);
3967 fixup_low_keys(path, &disk_key, 1);
3969 /* then fixup the leaf pointer in the path */
3970 if (path->slots[0] < push_items) {
3971 path->slots[0] += old_left_nritems;
3972 btrfs_tree_unlock(path->nodes[0]);
3973 free_extent_buffer(path->nodes[0]);
3974 path->nodes[0] = left;
3975 path->slots[1] -= 1;
3977 btrfs_tree_unlock(left);
3978 free_extent_buffer(left);
3979 path->slots[0] -= push_items;
3981 BUG_ON(path->slots[0] < 0);
3984 btrfs_tree_unlock(left);
3985 free_extent_buffer(left);
3990 * push some data in the path leaf to the left, trying to free up at
3991 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3993 * max_slot can put a limit on how far into the leaf we'll push items. The
3994 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3997 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3998 *root, struct btrfs_path *path, int min_data_size,
3999 int data_size, int empty, u32 max_slot)
4001 struct extent_buffer *right = path->nodes[0];
4002 struct extent_buffer *left;
4008 slot = path->slots[1];
4011 if (!path->nodes[1])
4014 right_nritems = btrfs_header_nritems(right);
4015 if (right_nritems == 0)
4018 btrfs_assert_tree_locked(path->nodes[1]);
4020 left = btrfs_read_node_slot(path->nodes[1], slot - 1);
4022 * slot - 1 is not valid or we fail to read the left node,
4023 * no big deal, just return.
4028 btrfs_tree_lock(left);
4029 btrfs_set_lock_blocking_write(left);
4031 free_space = btrfs_leaf_free_space(left);
4032 if (free_space < data_size) {
4037 /* cow and double check */
4038 ret = btrfs_cow_block(trans, root, left,
4039 path->nodes[1], slot - 1, &left);
4041 /* we hit -ENOSPC, but it isn't fatal here */
4047 free_space = btrfs_leaf_free_space(left);
4048 if (free_space < data_size) {
4053 return __push_leaf_left(path, min_data_size,
4054 empty, left, free_space, right_nritems,
4057 btrfs_tree_unlock(left);
4058 free_extent_buffer(left);
4063 * split the path's leaf in two, making sure there is at least data_size
4064 * available for the resulting leaf level of the path.
4066 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4067 struct btrfs_path *path,
4068 struct extent_buffer *l,
4069 struct extent_buffer *right,
4070 int slot, int mid, int nritems)
4072 struct btrfs_fs_info *fs_info = trans->fs_info;
4076 struct btrfs_disk_key disk_key;
4077 struct btrfs_map_token token;
4079 nritems = nritems - mid;
4080 btrfs_set_header_nritems(right, nritems);
4081 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4083 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4084 btrfs_item_nr_offset(mid),
4085 nritems * sizeof(struct btrfs_item));
4087 copy_extent_buffer(right, l,
4088 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4089 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4090 leaf_data_end(l), data_copy_size);
4092 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4094 btrfs_init_map_token(&token, right);
4095 for (i = 0; i < nritems; i++) {
4096 struct btrfs_item *item = btrfs_item_nr(i);
4099 ioff = btrfs_token_item_offset(right, item, &token);
4100 btrfs_set_token_item_offset(right, item,
4101 ioff + rt_data_off, &token);
4104 btrfs_set_header_nritems(l, mid);
4105 btrfs_item_key(right, &disk_key, 0);
4106 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4108 btrfs_mark_buffer_dirty(right);
4109 btrfs_mark_buffer_dirty(l);
4110 BUG_ON(path->slots[0] != slot);
4113 btrfs_tree_unlock(path->nodes[0]);
4114 free_extent_buffer(path->nodes[0]);
4115 path->nodes[0] = right;
4116 path->slots[0] -= mid;
4117 path->slots[1] += 1;
4119 btrfs_tree_unlock(right);
4120 free_extent_buffer(right);
4123 BUG_ON(path->slots[0] < 0);
4127 * double splits happen when we need to insert a big item in the middle
4128 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4129 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4132 * We avoid this by trying to push the items on either side of our target
4133 * into the adjacent leaves. If all goes well we can avoid the double split
4136 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4137 struct btrfs_root *root,
4138 struct btrfs_path *path,
4145 int space_needed = data_size;
4147 slot = path->slots[0];
4148 if (slot < btrfs_header_nritems(path->nodes[0]))
4149 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4152 * try to push all the items after our slot into the
4155 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4162 nritems = btrfs_header_nritems(path->nodes[0]);
4164 * our goal is to get our slot at the start or end of a leaf. If
4165 * we've done so we're done
4167 if (path->slots[0] == 0 || path->slots[0] == nritems)
4170 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4173 /* try to push all the items before our slot into the next leaf */
4174 slot = path->slots[0];
4175 space_needed = data_size;
4177 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4178 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4191 * split the path's leaf in two, making sure there is at least data_size
4192 * available for the resulting leaf level of the path.
4194 * returns 0 if all went well and < 0 on failure.
4196 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4197 struct btrfs_root *root,
4198 const struct btrfs_key *ins_key,
4199 struct btrfs_path *path, int data_size,
4202 struct btrfs_disk_key disk_key;
4203 struct extent_buffer *l;
4207 struct extent_buffer *right;
4208 struct btrfs_fs_info *fs_info = root->fs_info;
4212 int num_doubles = 0;
4213 int tried_avoid_double = 0;
4216 slot = path->slots[0];
4217 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4218 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4221 /* first try to make some room by pushing left and right */
4222 if (data_size && path->nodes[1]) {
4223 int space_needed = data_size;
4225 if (slot < btrfs_header_nritems(l))
4226 space_needed -= btrfs_leaf_free_space(l);
4228 wret = push_leaf_right(trans, root, path, space_needed,
4229 space_needed, 0, 0);
4233 space_needed = data_size;
4235 space_needed -= btrfs_leaf_free_space(l);
4236 wret = push_leaf_left(trans, root, path, space_needed,
4237 space_needed, 0, (u32)-1);
4243 /* did the pushes work? */
4244 if (btrfs_leaf_free_space(l) >= data_size)
4248 if (!path->nodes[1]) {
4249 ret = insert_new_root(trans, root, path, 1);
4256 slot = path->slots[0];
4257 nritems = btrfs_header_nritems(l);
4258 mid = (nritems + 1) / 2;
4262 leaf_space_used(l, mid, nritems - mid) + data_size >
4263 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4264 if (slot >= nritems) {
4268 if (mid != nritems &&
4269 leaf_space_used(l, mid, nritems - mid) +
4270 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4271 if (data_size && !tried_avoid_double)
4272 goto push_for_double;
4278 if (leaf_space_used(l, 0, mid) + data_size >
4279 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4280 if (!extend && data_size && slot == 0) {
4282 } else if ((extend || !data_size) && slot == 0) {
4286 if (mid != nritems &&
4287 leaf_space_used(l, mid, nritems - mid) +
4288 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4289 if (data_size && !tried_avoid_double)
4290 goto push_for_double;
4298 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4300 btrfs_item_key(l, &disk_key, mid);
4302 right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4305 return PTR_ERR(right);
4307 root_add_used(root, fs_info->nodesize);
4311 btrfs_set_header_nritems(right, 0);
4312 insert_ptr(trans, path, &disk_key,
4313 right->start, path->slots[1] + 1, 1);
4314 btrfs_tree_unlock(path->nodes[0]);
4315 free_extent_buffer(path->nodes[0]);
4316 path->nodes[0] = right;
4318 path->slots[1] += 1;
4320 btrfs_set_header_nritems(right, 0);
4321 insert_ptr(trans, path, &disk_key,
4322 right->start, path->slots[1], 1);
4323 btrfs_tree_unlock(path->nodes[0]);
4324 free_extent_buffer(path->nodes[0]);
4325 path->nodes[0] = right;
4327 if (path->slots[1] == 0)
4328 fixup_low_keys(path, &disk_key, 1);
4331 * We create a new leaf 'right' for the required ins_len and
4332 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4333 * the content of ins_len to 'right'.
4338 copy_for_split(trans, path, l, right, slot, mid, nritems);
4341 BUG_ON(num_doubles != 0);
4349 push_for_double_split(trans, root, path, data_size);
4350 tried_avoid_double = 1;
4351 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4356 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4357 struct btrfs_root *root,
4358 struct btrfs_path *path, int ins_len)
4360 struct btrfs_key key;
4361 struct extent_buffer *leaf;
4362 struct btrfs_file_extent_item *fi;
4367 leaf = path->nodes[0];
4368 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4370 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4371 key.type != BTRFS_EXTENT_CSUM_KEY);
4373 if (btrfs_leaf_free_space(leaf) >= ins_len)
4376 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4377 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4378 fi = btrfs_item_ptr(leaf, path->slots[0],
4379 struct btrfs_file_extent_item);
4380 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4382 btrfs_release_path(path);
4384 path->keep_locks = 1;
4385 path->search_for_split = 1;
4386 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4387 path->search_for_split = 0;
4394 leaf = path->nodes[0];
4395 /* if our item isn't there, return now */
4396 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4399 /* the leaf has changed, it now has room. return now */
4400 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4403 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4404 fi = btrfs_item_ptr(leaf, path->slots[0],
4405 struct btrfs_file_extent_item);
4406 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4410 btrfs_set_path_blocking(path);
4411 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4415 path->keep_locks = 0;
4416 btrfs_unlock_up_safe(path, 1);
4419 path->keep_locks = 0;
4423 static noinline int split_item(struct btrfs_path *path,
4424 const struct btrfs_key *new_key,
4425 unsigned long split_offset)
4427 struct extent_buffer *leaf;
4428 struct btrfs_item *item;
4429 struct btrfs_item *new_item;
4435 struct btrfs_disk_key disk_key;
4437 leaf = path->nodes[0];
4438 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4440 btrfs_set_path_blocking(path);
4442 item = btrfs_item_nr(path->slots[0]);
4443 orig_offset = btrfs_item_offset(leaf, item);
4444 item_size = btrfs_item_size(leaf, item);
4446 buf = kmalloc(item_size, GFP_NOFS);
4450 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4451 path->slots[0]), item_size);
4453 slot = path->slots[0] + 1;
4454 nritems = btrfs_header_nritems(leaf);
4455 if (slot != nritems) {
4456 /* shift the items */
4457 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4458 btrfs_item_nr_offset(slot),
4459 (nritems - slot) * sizeof(struct btrfs_item));
4462 btrfs_cpu_key_to_disk(&disk_key, new_key);
4463 btrfs_set_item_key(leaf, &disk_key, slot);
4465 new_item = btrfs_item_nr(slot);
4467 btrfs_set_item_offset(leaf, new_item, orig_offset);
4468 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4470 btrfs_set_item_offset(leaf, item,
4471 orig_offset + item_size - split_offset);
4472 btrfs_set_item_size(leaf, item, split_offset);
4474 btrfs_set_header_nritems(leaf, nritems + 1);
4476 /* write the data for the start of the original item */
4477 write_extent_buffer(leaf, buf,
4478 btrfs_item_ptr_offset(leaf, path->slots[0]),
4481 /* write the data for the new item */
4482 write_extent_buffer(leaf, buf + split_offset,
4483 btrfs_item_ptr_offset(leaf, slot),
4484 item_size - split_offset);
4485 btrfs_mark_buffer_dirty(leaf);
4487 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4493 * This function splits a single item into two items,
4494 * giving 'new_key' to the new item and splitting the
4495 * old one at split_offset (from the start of the item).
4497 * The path may be released by this operation. After
4498 * the split, the path is pointing to the old item. The
4499 * new item is going to be in the same node as the old one.
4501 * Note, the item being split must be smaller enough to live alone on
4502 * a tree block with room for one extra struct btrfs_item
4504 * This allows us to split the item in place, keeping a lock on the
4505 * leaf the entire time.
4507 int btrfs_split_item(struct btrfs_trans_handle *trans,
4508 struct btrfs_root *root,
4509 struct btrfs_path *path,
4510 const struct btrfs_key *new_key,
4511 unsigned long split_offset)
4514 ret = setup_leaf_for_split(trans, root, path,
4515 sizeof(struct btrfs_item));
4519 ret = split_item(path, new_key, split_offset);
4524 * This function duplicate a item, giving 'new_key' to the new item.
4525 * It guarantees both items live in the same tree leaf and the new item
4526 * is contiguous with the original item.
4528 * This allows us to split file extent in place, keeping a lock on the
4529 * leaf the entire time.
4531 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4532 struct btrfs_root *root,
4533 struct btrfs_path *path,
4534 const struct btrfs_key *new_key)
4536 struct extent_buffer *leaf;
4540 leaf = path->nodes[0];
4541 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4542 ret = setup_leaf_for_split(trans, root, path,
4543 item_size + sizeof(struct btrfs_item));
4548 setup_items_for_insert(root, path, new_key, &item_size,
4549 item_size, item_size +
4550 sizeof(struct btrfs_item), 1);
4551 leaf = path->nodes[0];
4552 memcpy_extent_buffer(leaf,
4553 btrfs_item_ptr_offset(leaf, path->slots[0]),
4554 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4560 * make the item pointed to by the path smaller. new_size indicates
4561 * how small to make it, and from_end tells us if we just chop bytes
4562 * off the end of the item or if we shift the item to chop bytes off
4565 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4568 struct extent_buffer *leaf;
4569 struct btrfs_item *item;
4571 unsigned int data_end;
4572 unsigned int old_data_start;
4573 unsigned int old_size;
4574 unsigned int size_diff;
4576 struct btrfs_map_token token;
4578 leaf = path->nodes[0];
4579 slot = path->slots[0];
4581 old_size = btrfs_item_size_nr(leaf, slot);
4582 if (old_size == new_size)
4585 nritems = btrfs_header_nritems(leaf);
4586 data_end = leaf_data_end(leaf);
4588 old_data_start = btrfs_item_offset_nr(leaf, slot);
4590 size_diff = old_size - new_size;
4593 BUG_ON(slot >= nritems);
4596 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4598 /* first correct the data pointers */
4599 btrfs_init_map_token(&token, leaf);
4600 for (i = slot; i < nritems; i++) {
4602 item = btrfs_item_nr(i);
4604 ioff = btrfs_token_item_offset(leaf, item, &token);
4605 btrfs_set_token_item_offset(leaf, item,
4606 ioff + size_diff, &token);
4609 /* shift the data */
4611 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4612 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4613 data_end, old_data_start + new_size - data_end);
4615 struct btrfs_disk_key disk_key;
4618 btrfs_item_key(leaf, &disk_key, slot);
4620 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4622 struct btrfs_file_extent_item *fi;
4624 fi = btrfs_item_ptr(leaf, slot,
4625 struct btrfs_file_extent_item);
4626 fi = (struct btrfs_file_extent_item *)(
4627 (unsigned long)fi - size_diff);
4629 if (btrfs_file_extent_type(leaf, fi) ==
4630 BTRFS_FILE_EXTENT_INLINE) {
4631 ptr = btrfs_item_ptr_offset(leaf, slot);
4632 memmove_extent_buffer(leaf, ptr,
4634 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4638 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4639 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4640 data_end, old_data_start - data_end);
4642 offset = btrfs_disk_key_offset(&disk_key);
4643 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4644 btrfs_set_item_key(leaf, &disk_key, slot);
4646 fixup_low_keys(path, &disk_key, 1);
4649 item = btrfs_item_nr(slot);
4650 btrfs_set_item_size(leaf, item, new_size);
4651 btrfs_mark_buffer_dirty(leaf);
4653 if (btrfs_leaf_free_space(leaf) < 0) {
4654 btrfs_print_leaf(leaf);
4660 * make the item pointed to by the path bigger, data_size is the added size.
4662 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4665 struct extent_buffer *leaf;
4666 struct btrfs_item *item;
4668 unsigned int data_end;
4669 unsigned int old_data;
4670 unsigned int old_size;
4672 struct btrfs_map_token token;
4674 leaf = path->nodes[0];
4676 nritems = btrfs_header_nritems(leaf);
4677 data_end = leaf_data_end(leaf);
4679 if (btrfs_leaf_free_space(leaf) < data_size) {
4680 btrfs_print_leaf(leaf);
4683 slot = path->slots[0];
4684 old_data = btrfs_item_end_nr(leaf, slot);
4687 if (slot >= nritems) {
4688 btrfs_print_leaf(leaf);
4689 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4695 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4697 /* first correct the data pointers */
4698 btrfs_init_map_token(&token, leaf);
4699 for (i = slot; i < nritems; i++) {
4701 item = btrfs_item_nr(i);
4703 ioff = btrfs_token_item_offset(leaf, item, &token);
4704 btrfs_set_token_item_offset(leaf, item,
4705 ioff - data_size, &token);
4708 /* shift the data */
4709 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4710 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4711 data_end, old_data - data_end);
4713 data_end = old_data;
4714 old_size = btrfs_item_size_nr(leaf, slot);
4715 item = btrfs_item_nr(slot);
4716 btrfs_set_item_size(leaf, item, old_size + data_size);
4717 btrfs_mark_buffer_dirty(leaf);
4719 if (btrfs_leaf_free_space(leaf) < 0) {
4720 btrfs_print_leaf(leaf);
4726 * this is a helper for btrfs_insert_empty_items, the main goal here is
4727 * to save stack depth by doing the bulk of the work in a function
4728 * that doesn't call btrfs_search_slot
4730 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4731 const struct btrfs_key *cpu_key, u32 *data_size,
4732 u32 total_data, u32 total_size, int nr)
4734 struct btrfs_fs_info *fs_info = root->fs_info;
4735 struct btrfs_item *item;
4738 unsigned int data_end;
4739 struct btrfs_disk_key disk_key;
4740 struct extent_buffer *leaf;
4742 struct btrfs_map_token token;
4744 if (path->slots[0] == 0) {
4745 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4746 fixup_low_keys(path, &disk_key, 1);
4748 btrfs_unlock_up_safe(path, 1);
4750 leaf = path->nodes[0];
4751 slot = path->slots[0];
4753 nritems = btrfs_header_nritems(leaf);
4754 data_end = leaf_data_end(leaf);
4756 if (btrfs_leaf_free_space(leaf) < total_size) {
4757 btrfs_print_leaf(leaf);
4758 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4759 total_size, btrfs_leaf_free_space(leaf));
4763 btrfs_init_map_token(&token, leaf);
4764 if (slot != nritems) {
4765 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4767 if (old_data < data_end) {
4768 btrfs_print_leaf(leaf);
4769 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4770 slot, old_data, data_end);
4774 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4776 /* first correct the data pointers */
4777 for (i = slot; i < nritems; i++) {
4780 item = btrfs_item_nr(i);
4781 ioff = btrfs_token_item_offset(leaf, item, &token);
4782 btrfs_set_token_item_offset(leaf, item,
4783 ioff - total_data, &token);
4785 /* shift the items */
4786 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4787 btrfs_item_nr_offset(slot),
4788 (nritems - slot) * sizeof(struct btrfs_item));
4790 /* shift the data */
4791 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4792 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4793 data_end, old_data - data_end);
4794 data_end = old_data;
4797 /* setup the item for the new data */
4798 for (i = 0; i < nr; i++) {
4799 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4800 btrfs_set_item_key(leaf, &disk_key, slot + i);
4801 item = btrfs_item_nr(slot + i);
4802 btrfs_set_token_item_offset(leaf, item,
4803 data_end - data_size[i], &token);
4804 data_end -= data_size[i];
4805 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4808 btrfs_set_header_nritems(leaf, nritems + nr);
4809 btrfs_mark_buffer_dirty(leaf);
4811 if (btrfs_leaf_free_space(leaf) < 0) {
4812 btrfs_print_leaf(leaf);
4818 * Given a key and some data, insert items into the tree.
4819 * This does all the path init required, making room in the tree if needed.
4821 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4822 struct btrfs_root *root,
4823 struct btrfs_path *path,
4824 const struct btrfs_key *cpu_key, u32 *data_size,
4833 for (i = 0; i < nr; i++)
4834 total_data += data_size[i];
4836 total_size = total_data + (nr * sizeof(struct btrfs_item));
4837 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4843 slot = path->slots[0];
4846 setup_items_for_insert(root, path, cpu_key, data_size,
4847 total_data, total_size, nr);
4852 * Given a key and some data, insert an item into the tree.
4853 * This does all the path init required, making room in the tree if needed.
4855 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4856 const struct btrfs_key *cpu_key, void *data,
4860 struct btrfs_path *path;
4861 struct extent_buffer *leaf;
4864 path = btrfs_alloc_path();
4867 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4869 leaf = path->nodes[0];
4870 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4871 write_extent_buffer(leaf, data, ptr, data_size);
4872 btrfs_mark_buffer_dirty(leaf);
4874 btrfs_free_path(path);
4879 * delete the pointer from a given node.
4881 * the tree should have been previously balanced so the deletion does not
4884 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4885 int level, int slot)
4887 struct extent_buffer *parent = path->nodes[level];
4891 nritems = btrfs_header_nritems(parent);
4892 if (slot != nritems - 1) {
4894 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4895 nritems - slot - 1);
4898 memmove_extent_buffer(parent,
4899 btrfs_node_key_ptr_offset(slot),
4900 btrfs_node_key_ptr_offset(slot + 1),
4901 sizeof(struct btrfs_key_ptr) *
4902 (nritems - slot - 1));
4904 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4910 btrfs_set_header_nritems(parent, nritems);
4911 if (nritems == 0 && parent == root->node) {
4912 BUG_ON(btrfs_header_level(root->node) != 1);
4913 /* just turn the root into a leaf and break */
4914 btrfs_set_header_level(root->node, 0);
4915 } else if (slot == 0) {
4916 struct btrfs_disk_key disk_key;
4918 btrfs_node_key(parent, &disk_key, 0);
4919 fixup_low_keys(path, &disk_key, level + 1);
4921 btrfs_mark_buffer_dirty(parent);
4925 * a helper function to delete the leaf pointed to by path->slots[1] and
4928 * This deletes the pointer in path->nodes[1] and frees the leaf
4929 * block extent. zero is returned if it all worked out, < 0 otherwise.
4931 * The path must have already been setup for deleting the leaf, including
4932 * all the proper balancing. path->nodes[1] must be locked.
4934 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4935 struct btrfs_root *root,
4936 struct btrfs_path *path,
4937 struct extent_buffer *leaf)
4939 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4940 del_ptr(root, path, 1, path->slots[1]);
4943 * btrfs_free_extent is expensive, we want to make sure we
4944 * aren't holding any locks when we call it
4946 btrfs_unlock_up_safe(path, 0);
4948 root_sub_used(root, leaf->len);
4950 extent_buffer_get(leaf);
4951 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4952 free_extent_buffer_stale(leaf);
4955 * delete the item at the leaf level in path. If that empties
4956 * the leaf, remove it from the tree
4958 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4959 struct btrfs_path *path, int slot, int nr)
4961 struct btrfs_fs_info *fs_info = root->fs_info;
4962 struct extent_buffer *leaf;
4963 struct btrfs_item *item;
4971 leaf = path->nodes[0];
4972 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4974 for (i = 0; i < nr; i++)
4975 dsize += btrfs_item_size_nr(leaf, slot + i);
4977 nritems = btrfs_header_nritems(leaf);
4979 if (slot + nr != nritems) {
4980 int data_end = leaf_data_end(leaf);
4981 struct btrfs_map_token token;
4983 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4985 BTRFS_LEAF_DATA_OFFSET + data_end,
4986 last_off - data_end);
4988 btrfs_init_map_token(&token, leaf);
4989 for (i = slot + nr; i < nritems; i++) {
4992 item = btrfs_item_nr(i);
4993 ioff = btrfs_token_item_offset(leaf, item, &token);
4994 btrfs_set_token_item_offset(leaf, item,
4995 ioff + dsize, &token);
4998 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4999 btrfs_item_nr_offset(slot + nr),
5000 sizeof(struct btrfs_item) *
5001 (nritems - slot - nr));
5003 btrfs_set_header_nritems(leaf, nritems - nr);
5006 /* delete the leaf if we've emptied it */
5008 if (leaf == root->node) {
5009 btrfs_set_header_level(leaf, 0);
5011 btrfs_set_path_blocking(path);
5012 btrfs_clean_tree_block(leaf);
5013 btrfs_del_leaf(trans, root, path, leaf);
5016 int used = leaf_space_used(leaf, 0, nritems);
5018 struct btrfs_disk_key disk_key;
5020 btrfs_item_key(leaf, &disk_key, 0);
5021 fixup_low_keys(path, &disk_key, 1);
5024 /* delete the leaf if it is mostly empty */
5025 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5026 /* push_leaf_left fixes the path.
5027 * make sure the path still points to our leaf
5028 * for possible call to del_ptr below
5030 slot = path->slots[1];
5031 extent_buffer_get(leaf);
5033 btrfs_set_path_blocking(path);
5034 wret = push_leaf_left(trans, root, path, 1, 1,
5036 if (wret < 0 && wret != -ENOSPC)
5039 if (path->nodes[0] == leaf &&
5040 btrfs_header_nritems(leaf)) {
5041 wret = push_leaf_right(trans, root, path, 1,
5043 if (wret < 0 && wret != -ENOSPC)
5047 if (btrfs_header_nritems(leaf) == 0) {
5048 path->slots[1] = slot;
5049 btrfs_del_leaf(trans, root, path, leaf);
5050 free_extent_buffer(leaf);
5053 /* if we're still in the path, make sure
5054 * we're dirty. Otherwise, one of the
5055 * push_leaf functions must have already
5056 * dirtied this buffer
5058 if (path->nodes[0] == leaf)
5059 btrfs_mark_buffer_dirty(leaf);
5060 free_extent_buffer(leaf);
5063 btrfs_mark_buffer_dirty(leaf);
5070 * search the tree again to find a leaf with lesser keys
5071 * returns 0 if it found something or 1 if there are no lesser leaves.
5072 * returns < 0 on io errors.
5074 * This may release the path, and so you may lose any locks held at the
5077 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5079 struct btrfs_key key;
5080 struct btrfs_disk_key found_key;
5083 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5085 if (key.offset > 0) {
5087 } else if (key.type > 0) {
5089 key.offset = (u64)-1;
5090 } else if (key.objectid > 0) {
5093 key.offset = (u64)-1;
5098 btrfs_release_path(path);
5099 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5102 btrfs_item_key(path->nodes[0], &found_key, 0);
5103 ret = comp_keys(&found_key, &key);
5105 * We might have had an item with the previous key in the tree right
5106 * before we released our path. And after we released our path, that
5107 * item might have been pushed to the first slot (0) of the leaf we
5108 * were holding due to a tree balance. Alternatively, an item with the
5109 * previous key can exist as the only element of a leaf (big fat item).
5110 * Therefore account for these 2 cases, so that our callers (like
5111 * btrfs_previous_item) don't miss an existing item with a key matching
5112 * the previous key we computed above.
5120 * A helper function to walk down the tree starting at min_key, and looking
5121 * for nodes or leaves that are have a minimum transaction id.
5122 * This is used by the btree defrag code, and tree logging
5124 * This does not cow, but it does stuff the starting key it finds back
5125 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5126 * key and get a writable path.
5128 * This honors path->lowest_level to prevent descent past a given level
5131 * min_trans indicates the oldest transaction that you are interested
5132 * in walking through. Any nodes or leaves older than min_trans are
5133 * skipped over (without reading them).
5135 * returns zero if something useful was found, < 0 on error and 1 if there
5136 * was nothing in the tree that matched the search criteria.
5138 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5139 struct btrfs_path *path,
5142 struct extent_buffer *cur;
5143 struct btrfs_key found_key;
5149 int keep_locks = path->keep_locks;
5151 path->keep_locks = 1;
5153 cur = btrfs_read_lock_root_node(root);
5154 level = btrfs_header_level(cur);
5155 WARN_ON(path->nodes[level]);
5156 path->nodes[level] = cur;
5157 path->locks[level] = BTRFS_READ_LOCK;
5159 if (btrfs_header_generation(cur) < min_trans) {
5164 nritems = btrfs_header_nritems(cur);
5165 level = btrfs_header_level(cur);
5166 sret = btrfs_bin_search(cur, min_key, level, &slot);
5172 /* at the lowest level, we're done, setup the path and exit */
5173 if (level == path->lowest_level) {
5174 if (slot >= nritems)
5177 path->slots[level] = slot;
5178 btrfs_item_key_to_cpu(cur, &found_key, slot);
5181 if (sret && slot > 0)
5184 * check this node pointer against the min_trans parameters.
5185 * If it is too old, old, skip to the next one.
5187 while (slot < nritems) {
5190 gen = btrfs_node_ptr_generation(cur, slot);
5191 if (gen < min_trans) {
5199 * we didn't find a candidate key in this node, walk forward
5200 * and find another one
5202 if (slot >= nritems) {
5203 path->slots[level] = slot;
5204 btrfs_set_path_blocking(path);
5205 sret = btrfs_find_next_key(root, path, min_key, level,
5208 btrfs_release_path(path);
5214 /* save our key for returning back */
5215 btrfs_node_key_to_cpu(cur, &found_key, slot);
5216 path->slots[level] = slot;
5217 if (level == path->lowest_level) {
5221 btrfs_set_path_blocking(path);
5222 cur = btrfs_read_node_slot(cur, slot);
5228 btrfs_tree_read_lock(cur);
5230 path->locks[level - 1] = BTRFS_READ_LOCK;
5231 path->nodes[level - 1] = cur;
5232 unlock_up(path, level, 1, 0, NULL);
5235 path->keep_locks = keep_locks;
5237 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5238 btrfs_set_path_blocking(path);
5239 memcpy(min_key, &found_key, sizeof(found_key));
5245 * this is similar to btrfs_next_leaf, but does not try to preserve
5246 * and fixup the path. It looks for and returns the next key in the
5247 * tree based on the current path and the min_trans parameters.
5249 * 0 is returned if another key is found, < 0 if there are any errors
5250 * and 1 is returned if there are no higher keys in the tree
5252 * path->keep_locks should be set to 1 on the search made before
5253 * calling this function.
5255 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5256 struct btrfs_key *key, int level, u64 min_trans)
5259 struct extent_buffer *c;
5261 WARN_ON(!path->keep_locks && !path->skip_locking);
5262 while (level < BTRFS_MAX_LEVEL) {
5263 if (!path->nodes[level])
5266 slot = path->slots[level] + 1;
5267 c = path->nodes[level];
5269 if (slot >= btrfs_header_nritems(c)) {
5272 struct btrfs_key cur_key;
5273 if (level + 1 >= BTRFS_MAX_LEVEL ||
5274 !path->nodes[level + 1])
5277 if (path->locks[level + 1] || path->skip_locking) {
5282 slot = btrfs_header_nritems(c) - 1;
5284 btrfs_item_key_to_cpu(c, &cur_key, slot);
5286 btrfs_node_key_to_cpu(c, &cur_key, slot);
5288 orig_lowest = path->lowest_level;
5289 btrfs_release_path(path);
5290 path->lowest_level = level;
5291 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5293 path->lowest_level = orig_lowest;
5297 c = path->nodes[level];
5298 slot = path->slots[level];
5305 btrfs_item_key_to_cpu(c, key, slot);
5307 u64 gen = btrfs_node_ptr_generation(c, slot);
5309 if (gen < min_trans) {
5313 btrfs_node_key_to_cpu(c, key, slot);
5321 * search the tree again to find a leaf with greater keys
5322 * returns 0 if it found something or 1 if there are no greater leaves.
5323 * returns < 0 on io errors.
5325 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5327 return btrfs_next_old_leaf(root, path, 0);
5330 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5335 struct extent_buffer *c;
5336 struct extent_buffer *next;
5337 struct btrfs_key key;
5340 int old_spinning = path->leave_spinning;
5341 int next_rw_lock = 0;
5343 nritems = btrfs_header_nritems(path->nodes[0]);
5347 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5352 btrfs_release_path(path);
5354 path->keep_locks = 1;
5355 path->leave_spinning = 1;
5358 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5360 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5361 path->keep_locks = 0;
5366 nritems = btrfs_header_nritems(path->nodes[0]);
5368 * by releasing the path above we dropped all our locks. A balance
5369 * could have added more items next to the key that used to be
5370 * at the very end of the block. So, check again here and
5371 * advance the path if there are now more items available.
5373 if (nritems > 0 && path->slots[0] < nritems - 1) {
5380 * So the above check misses one case:
5381 * - after releasing the path above, someone has removed the item that
5382 * used to be at the very end of the block, and balance between leafs
5383 * gets another one with bigger key.offset to replace it.
5385 * This one should be returned as well, or we can get leaf corruption
5386 * later(esp. in __btrfs_drop_extents()).
5388 * And a bit more explanation about this check,
5389 * with ret > 0, the key isn't found, the path points to the slot
5390 * where it should be inserted, so the path->slots[0] item must be the
5393 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5398 while (level < BTRFS_MAX_LEVEL) {
5399 if (!path->nodes[level]) {
5404 slot = path->slots[level] + 1;
5405 c = path->nodes[level];
5406 if (slot >= btrfs_header_nritems(c)) {
5408 if (level == BTRFS_MAX_LEVEL) {
5416 btrfs_tree_unlock_rw(next, next_rw_lock);
5417 free_extent_buffer(next);
5421 next_rw_lock = path->locks[level];
5422 ret = read_block_for_search(root, path, &next, level,
5428 btrfs_release_path(path);
5432 if (!path->skip_locking) {
5433 ret = btrfs_try_tree_read_lock(next);
5434 if (!ret && time_seq) {
5436 * If we don't get the lock, we may be racing
5437 * with push_leaf_left, holding that lock while
5438 * itself waiting for the leaf we've currently
5439 * locked. To solve this situation, we give up
5440 * on our lock and cycle.
5442 free_extent_buffer(next);
5443 btrfs_release_path(path);
5448 btrfs_set_path_blocking(path);
5449 btrfs_tree_read_lock(next);
5451 next_rw_lock = BTRFS_READ_LOCK;
5455 path->slots[level] = slot;
5458 c = path->nodes[level];
5459 if (path->locks[level])
5460 btrfs_tree_unlock_rw(c, path->locks[level]);
5462 free_extent_buffer(c);
5463 path->nodes[level] = next;
5464 path->slots[level] = 0;
5465 if (!path->skip_locking)
5466 path->locks[level] = next_rw_lock;
5470 ret = read_block_for_search(root, path, &next, level,
5476 btrfs_release_path(path);
5480 if (!path->skip_locking) {
5481 ret = btrfs_try_tree_read_lock(next);
5483 btrfs_set_path_blocking(path);
5484 btrfs_tree_read_lock(next);
5486 next_rw_lock = BTRFS_READ_LOCK;
5491 unlock_up(path, 0, 1, 0, NULL);
5492 path->leave_spinning = old_spinning;
5494 btrfs_set_path_blocking(path);
5500 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5501 * searching until it gets past min_objectid or finds an item of 'type'
5503 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5505 int btrfs_previous_item(struct btrfs_root *root,
5506 struct btrfs_path *path, u64 min_objectid,
5509 struct btrfs_key found_key;
5510 struct extent_buffer *leaf;
5515 if (path->slots[0] == 0) {
5516 btrfs_set_path_blocking(path);
5517 ret = btrfs_prev_leaf(root, path);
5523 leaf = path->nodes[0];
5524 nritems = btrfs_header_nritems(leaf);
5527 if (path->slots[0] == nritems)
5530 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5531 if (found_key.objectid < min_objectid)
5533 if (found_key.type == type)
5535 if (found_key.objectid == min_objectid &&
5536 found_key.type < type)
5543 * search in extent tree to find a previous Metadata/Data extent item with
5546 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5548 int btrfs_previous_extent_item(struct btrfs_root *root,
5549 struct btrfs_path *path, u64 min_objectid)
5551 struct btrfs_key found_key;
5552 struct extent_buffer *leaf;
5557 if (path->slots[0] == 0) {
5558 btrfs_set_path_blocking(path);
5559 ret = btrfs_prev_leaf(root, path);
5565 leaf = path->nodes[0];
5566 nritems = btrfs_header_nritems(leaf);
5569 if (path->slots[0] == nritems)
5572 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5573 if (found_key.objectid < min_objectid)
5575 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5576 found_key.type == BTRFS_METADATA_ITEM_KEY)
5578 if (found_key.objectid == min_objectid &&
5579 found_key.type < BTRFS_EXTENT_ITEM_KEY)