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.
380 * Note: must be called with write lock for fs_info::tree_mod_log_lock.
383 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
385 struct rb_root *tm_root;
386 struct rb_node **new;
387 struct rb_node *parent = NULL;
388 struct tree_mod_elem *cur;
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, fs_info,
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 struct extent_buffer *old;
1347 struct tree_mod_root *old_root = NULL;
1348 u64 old_generation = 0;
1352 eb_root = btrfs_read_lock_root_node(root);
1353 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1357 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1358 old_root = &tm->old_root;
1359 old_generation = tm->generation;
1360 logical = old_root->logical;
1361 level = old_root->level;
1363 logical = eb_root->start;
1364 level = btrfs_header_level(eb_root);
1367 tm = tree_mod_log_search(fs_info, logical, time_seq);
1368 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1369 btrfs_tree_read_unlock(eb_root);
1370 free_extent_buffer(eb_root);
1371 old = read_tree_block(fs_info, logical, 0, level, NULL);
1372 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1374 free_extent_buffer(old);
1376 "failed to read tree block %llu from get_old_root",
1379 eb = btrfs_clone_extent_buffer(old);
1380 free_extent_buffer(old);
1382 } else if (old_root) {
1383 btrfs_tree_read_unlock(eb_root);
1384 free_extent_buffer(eb_root);
1385 eb = alloc_dummy_extent_buffer(fs_info, logical);
1387 btrfs_set_lock_blocking_read(eb_root);
1388 eb = btrfs_clone_extent_buffer(eb_root);
1389 btrfs_tree_read_unlock_blocking(eb_root);
1390 free_extent_buffer(eb_root);
1395 btrfs_tree_read_lock(eb);
1397 btrfs_set_header_bytenr(eb, eb->start);
1398 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1399 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1400 btrfs_set_header_level(eb, old_root->level);
1401 btrfs_set_header_generation(eb, old_generation);
1404 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1406 WARN_ON(btrfs_header_level(eb) != 0);
1407 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1412 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1414 struct tree_mod_elem *tm;
1416 struct extent_buffer *eb_root = btrfs_root_node(root);
1418 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1419 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1420 level = tm->old_root.level;
1422 level = btrfs_header_level(eb_root);
1424 free_extent_buffer(eb_root);
1429 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1430 struct btrfs_root *root,
1431 struct extent_buffer *buf)
1433 if (btrfs_is_testing(root->fs_info))
1436 /* Ensure we can see the FORCE_COW bit */
1437 smp_mb__before_atomic();
1440 * We do not need to cow a block if
1441 * 1) this block is not created or changed in this transaction;
1442 * 2) this block does not belong to TREE_RELOC tree;
1443 * 3) the root is not forced COW.
1445 * What is forced COW:
1446 * when we create snapshot during committing the transaction,
1447 * after we've finished copying src root, we must COW the shared
1448 * block to ensure the metadata consistency.
1450 if (btrfs_header_generation(buf) == trans->transid &&
1451 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1452 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1453 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1454 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1460 * cows a single block, see __btrfs_cow_block for the real work.
1461 * This version of it has extra checks so that a block isn't COWed more than
1462 * once per transaction, as long as it hasn't been written yet
1464 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1465 struct btrfs_root *root, struct extent_buffer *buf,
1466 struct extent_buffer *parent, int parent_slot,
1467 struct extent_buffer **cow_ret)
1469 struct btrfs_fs_info *fs_info = root->fs_info;
1473 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1475 "COW'ing blocks on a fs root that's being dropped");
1477 if (trans->transaction != fs_info->running_transaction)
1478 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1480 fs_info->running_transaction->transid);
1482 if (trans->transid != fs_info->generation)
1483 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1484 trans->transid, fs_info->generation);
1486 if (!should_cow_block(trans, root, buf)) {
1487 trans->dirty = true;
1492 search_start = buf->start & ~((u64)SZ_1G - 1);
1495 btrfs_set_lock_blocking_write(parent);
1496 btrfs_set_lock_blocking_write(buf);
1499 * Before CoWing this block for later modification, check if it's
1500 * the subtree root and do the delayed subtree trace if needed.
1502 * Also We don't care about the error, as it's handled internally.
1504 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1505 ret = __btrfs_cow_block(trans, root, buf, parent,
1506 parent_slot, cow_ret, search_start, 0);
1508 trace_btrfs_cow_block(root, buf, *cow_ret);
1514 * helper function for defrag to decide if two blocks pointed to by a
1515 * node are actually close by
1517 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1519 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1521 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1527 * compare two keys in a memcmp fashion
1529 static int comp_keys(const struct btrfs_disk_key *disk,
1530 const struct btrfs_key *k2)
1532 struct btrfs_key k1;
1534 btrfs_disk_key_to_cpu(&k1, disk);
1536 return btrfs_comp_cpu_keys(&k1, k2);
1540 * same as comp_keys only with two btrfs_key's
1542 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1544 if (k1->objectid > k2->objectid)
1546 if (k1->objectid < k2->objectid)
1548 if (k1->type > k2->type)
1550 if (k1->type < k2->type)
1552 if (k1->offset > k2->offset)
1554 if (k1->offset < k2->offset)
1560 * this is used by the defrag code to go through all the
1561 * leaves pointed to by a node and reallocate them so that
1562 * disk order is close to key order
1564 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1565 struct btrfs_root *root, struct extent_buffer *parent,
1566 int start_slot, u64 *last_ret,
1567 struct btrfs_key *progress)
1569 struct btrfs_fs_info *fs_info = root->fs_info;
1570 struct extent_buffer *cur;
1573 u64 search_start = *last_ret;
1583 int progress_passed = 0;
1584 struct btrfs_disk_key disk_key;
1586 parent_level = btrfs_header_level(parent);
1588 WARN_ON(trans->transaction != fs_info->running_transaction);
1589 WARN_ON(trans->transid != fs_info->generation);
1591 parent_nritems = btrfs_header_nritems(parent);
1592 blocksize = fs_info->nodesize;
1593 end_slot = parent_nritems - 1;
1595 if (parent_nritems <= 1)
1598 btrfs_set_lock_blocking_write(parent);
1600 for (i = start_slot; i <= end_slot; i++) {
1601 struct btrfs_key first_key;
1604 btrfs_node_key(parent, &disk_key, i);
1605 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1608 progress_passed = 1;
1609 blocknr = btrfs_node_blockptr(parent, i);
1610 gen = btrfs_node_ptr_generation(parent, i);
1611 btrfs_node_key_to_cpu(parent, &first_key, i);
1612 if (last_block == 0)
1613 last_block = blocknr;
1616 other = btrfs_node_blockptr(parent, i - 1);
1617 close = close_blocks(blocknr, other, blocksize);
1619 if (!close && i < end_slot) {
1620 other = btrfs_node_blockptr(parent, i + 1);
1621 close = close_blocks(blocknr, other, blocksize);
1624 last_block = blocknr;
1628 cur = find_extent_buffer(fs_info, blocknr);
1630 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1633 if (!cur || !uptodate) {
1635 cur = read_tree_block(fs_info, blocknr, gen,
1639 return PTR_ERR(cur);
1640 } else if (!extent_buffer_uptodate(cur)) {
1641 free_extent_buffer(cur);
1644 } else if (!uptodate) {
1645 err = btrfs_read_buffer(cur, gen,
1646 parent_level - 1,&first_key);
1648 free_extent_buffer(cur);
1653 if (search_start == 0)
1654 search_start = last_block;
1656 btrfs_tree_lock(cur);
1657 btrfs_set_lock_blocking_write(cur);
1658 err = __btrfs_cow_block(trans, root, cur, parent, i,
1661 (end_slot - i) * blocksize));
1663 btrfs_tree_unlock(cur);
1664 free_extent_buffer(cur);
1667 search_start = cur->start;
1668 last_block = cur->start;
1669 *last_ret = search_start;
1670 btrfs_tree_unlock(cur);
1671 free_extent_buffer(cur);
1677 * search for key in the extent_buffer. The items start at offset p,
1678 * and they are item_size apart. There are 'max' items in p.
1680 * the slot in the array is returned via slot, and it points to
1681 * the place where you would insert key if it is not found in
1684 * slot may point to max if the key is bigger than all of the keys
1686 static noinline int generic_bin_search(struct extent_buffer *eb,
1687 unsigned long p, int item_size,
1688 const struct btrfs_key *key,
1695 struct btrfs_disk_key *tmp = NULL;
1696 struct btrfs_disk_key unaligned;
1697 unsigned long offset;
1699 unsigned long map_start = 0;
1700 unsigned long map_len = 0;
1704 btrfs_err(eb->fs_info,
1705 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1706 __func__, low, high, eb->start,
1707 btrfs_header_owner(eb), btrfs_header_level(eb));
1711 while (low < high) {
1712 mid = (low + high) / 2;
1713 offset = p + mid * item_size;
1715 if (!kaddr || offset < map_start ||
1716 (offset + sizeof(struct btrfs_disk_key)) >
1717 map_start + map_len) {
1719 err = map_private_extent_buffer(eb, offset,
1720 sizeof(struct btrfs_disk_key),
1721 &kaddr, &map_start, &map_len);
1724 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1726 } else if (err == 1) {
1727 read_extent_buffer(eb, &unaligned,
1728 offset, sizeof(unaligned));
1735 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1738 ret = comp_keys(tmp, key);
1754 * simple bin_search frontend that does the right thing for
1757 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1758 int level, int *slot)
1761 return generic_bin_search(eb,
1762 offsetof(struct btrfs_leaf, items),
1763 sizeof(struct btrfs_item),
1764 key, btrfs_header_nritems(eb),
1767 return generic_bin_search(eb,
1768 offsetof(struct btrfs_node, ptrs),
1769 sizeof(struct btrfs_key_ptr),
1770 key, btrfs_header_nritems(eb),
1774 static void root_add_used(struct btrfs_root *root, u32 size)
1776 spin_lock(&root->accounting_lock);
1777 btrfs_set_root_used(&root->root_item,
1778 btrfs_root_used(&root->root_item) + size);
1779 spin_unlock(&root->accounting_lock);
1782 static void root_sub_used(struct btrfs_root *root, u32 size)
1784 spin_lock(&root->accounting_lock);
1785 btrfs_set_root_used(&root->root_item,
1786 btrfs_root_used(&root->root_item) - size);
1787 spin_unlock(&root->accounting_lock);
1790 /* given a node and slot number, this reads the blocks it points to. The
1791 * extent buffer is returned with a reference taken (but unlocked).
1793 static noinline struct extent_buffer *read_node_slot(
1794 struct extent_buffer *parent, int slot)
1796 int level = btrfs_header_level(parent);
1797 struct extent_buffer *eb;
1798 struct btrfs_key first_key;
1800 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1801 return ERR_PTR(-ENOENT);
1805 btrfs_node_key_to_cpu(parent, &first_key, slot);
1806 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1807 btrfs_node_ptr_generation(parent, slot),
1808 level - 1, &first_key);
1809 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1810 free_extent_buffer(eb);
1818 * node level balancing, used to make sure nodes are in proper order for
1819 * item deletion. We balance from the top down, so we have to make sure
1820 * that a deletion won't leave an node completely empty later on.
1822 static noinline int balance_level(struct btrfs_trans_handle *trans,
1823 struct btrfs_root *root,
1824 struct btrfs_path *path, int level)
1826 struct btrfs_fs_info *fs_info = root->fs_info;
1827 struct extent_buffer *right = NULL;
1828 struct extent_buffer *mid;
1829 struct extent_buffer *left = NULL;
1830 struct extent_buffer *parent = NULL;
1834 int orig_slot = path->slots[level];
1839 mid = path->nodes[level];
1841 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1842 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1843 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1845 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1847 if (level < BTRFS_MAX_LEVEL - 1) {
1848 parent = path->nodes[level + 1];
1849 pslot = path->slots[level + 1];
1853 * deal with the case where there is only one pointer in the root
1854 * by promoting the node below to a root
1857 struct extent_buffer *child;
1859 if (btrfs_header_nritems(mid) != 1)
1862 /* promote the child to a root */
1863 child = read_node_slot(mid, 0);
1864 if (IS_ERR(child)) {
1865 ret = PTR_ERR(child);
1866 btrfs_handle_fs_error(fs_info, ret, NULL);
1870 btrfs_tree_lock(child);
1871 btrfs_set_lock_blocking_write(child);
1872 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1874 btrfs_tree_unlock(child);
1875 free_extent_buffer(child);
1879 ret = tree_mod_log_insert_root(root->node, child, 1);
1881 rcu_assign_pointer(root->node, child);
1883 add_root_to_dirty_list(root);
1884 btrfs_tree_unlock(child);
1886 path->locks[level] = 0;
1887 path->nodes[level] = NULL;
1888 btrfs_clean_tree_block(mid);
1889 btrfs_tree_unlock(mid);
1890 /* once for the path */
1891 free_extent_buffer(mid);
1893 root_sub_used(root, mid->len);
1894 btrfs_free_tree_block(trans, root, mid, 0, 1);
1895 /* once for the root ptr */
1896 free_extent_buffer_stale(mid);
1899 if (btrfs_header_nritems(mid) >
1900 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1903 left = read_node_slot(parent, pslot - 1);
1908 btrfs_tree_lock(left);
1909 btrfs_set_lock_blocking_write(left);
1910 wret = btrfs_cow_block(trans, root, left,
1911 parent, pslot - 1, &left);
1918 right = read_node_slot(parent, pslot + 1);
1923 btrfs_tree_lock(right);
1924 btrfs_set_lock_blocking_write(right);
1925 wret = btrfs_cow_block(trans, root, right,
1926 parent, pslot + 1, &right);
1933 /* first, try to make some room in the middle buffer */
1935 orig_slot += btrfs_header_nritems(left);
1936 wret = push_node_left(trans, left, mid, 1);
1942 * then try to empty the right most buffer into the middle
1945 wret = push_node_left(trans, mid, right, 1);
1946 if (wret < 0 && wret != -ENOSPC)
1948 if (btrfs_header_nritems(right) == 0) {
1949 btrfs_clean_tree_block(right);
1950 btrfs_tree_unlock(right);
1951 del_ptr(root, path, level + 1, pslot + 1);
1952 root_sub_used(root, right->len);
1953 btrfs_free_tree_block(trans, root, right, 0, 1);
1954 free_extent_buffer_stale(right);
1957 struct btrfs_disk_key right_key;
1958 btrfs_node_key(right, &right_key, 0);
1959 ret = tree_mod_log_insert_key(parent, pslot + 1,
1960 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1962 btrfs_set_node_key(parent, &right_key, pslot + 1);
1963 btrfs_mark_buffer_dirty(parent);
1966 if (btrfs_header_nritems(mid) == 1) {
1968 * we're not allowed to leave a node with one item in the
1969 * tree during a delete. A deletion from lower in the tree
1970 * could try to delete the only pointer in this node.
1971 * So, pull some keys from the left.
1972 * There has to be a left pointer at this point because
1973 * otherwise we would have pulled some pointers from the
1978 btrfs_handle_fs_error(fs_info, ret, NULL);
1981 wret = balance_node_right(trans, mid, left);
1987 wret = push_node_left(trans, left, mid, 1);
1993 if (btrfs_header_nritems(mid) == 0) {
1994 btrfs_clean_tree_block(mid);
1995 btrfs_tree_unlock(mid);
1996 del_ptr(root, path, level + 1, pslot);
1997 root_sub_used(root, mid->len);
1998 btrfs_free_tree_block(trans, root, mid, 0, 1);
1999 free_extent_buffer_stale(mid);
2002 /* update the parent key to reflect our changes */
2003 struct btrfs_disk_key mid_key;
2004 btrfs_node_key(mid, &mid_key, 0);
2005 ret = tree_mod_log_insert_key(parent, pslot,
2006 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2008 btrfs_set_node_key(parent, &mid_key, pslot);
2009 btrfs_mark_buffer_dirty(parent);
2012 /* update the path */
2014 if (btrfs_header_nritems(left) > orig_slot) {
2015 extent_buffer_get(left);
2016 /* left was locked after cow */
2017 path->nodes[level] = left;
2018 path->slots[level + 1] -= 1;
2019 path->slots[level] = orig_slot;
2021 btrfs_tree_unlock(mid);
2022 free_extent_buffer(mid);
2025 orig_slot -= btrfs_header_nritems(left);
2026 path->slots[level] = orig_slot;
2029 /* double check we haven't messed things up */
2031 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2035 btrfs_tree_unlock(right);
2036 free_extent_buffer(right);
2039 if (path->nodes[level] != left)
2040 btrfs_tree_unlock(left);
2041 free_extent_buffer(left);
2046 /* Node balancing for insertion. Here we only split or push nodes around
2047 * when they are completely full. This is also done top down, so we
2048 * have to be pessimistic.
2050 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2051 struct btrfs_root *root,
2052 struct btrfs_path *path, int level)
2054 struct btrfs_fs_info *fs_info = root->fs_info;
2055 struct extent_buffer *right = NULL;
2056 struct extent_buffer *mid;
2057 struct extent_buffer *left = NULL;
2058 struct extent_buffer *parent = NULL;
2062 int orig_slot = path->slots[level];
2067 mid = path->nodes[level];
2068 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2070 if (level < BTRFS_MAX_LEVEL - 1) {
2071 parent = path->nodes[level + 1];
2072 pslot = path->slots[level + 1];
2078 left = read_node_slot(parent, pslot - 1);
2082 /* first, try to make some room in the middle buffer */
2086 btrfs_tree_lock(left);
2087 btrfs_set_lock_blocking_write(left);
2089 left_nr = btrfs_header_nritems(left);
2090 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2093 ret = btrfs_cow_block(trans, root, left, parent,
2098 wret = push_node_left(trans, left, mid, 0);
2104 struct btrfs_disk_key disk_key;
2105 orig_slot += left_nr;
2106 btrfs_node_key(mid, &disk_key, 0);
2107 ret = tree_mod_log_insert_key(parent, pslot,
2108 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2110 btrfs_set_node_key(parent, &disk_key, pslot);
2111 btrfs_mark_buffer_dirty(parent);
2112 if (btrfs_header_nritems(left) > orig_slot) {
2113 path->nodes[level] = left;
2114 path->slots[level + 1] -= 1;
2115 path->slots[level] = orig_slot;
2116 btrfs_tree_unlock(mid);
2117 free_extent_buffer(mid);
2120 btrfs_header_nritems(left);
2121 path->slots[level] = orig_slot;
2122 btrfs_tree_unlock(left);
2123 free_extent_buffer(left);
2127 btrfs_tree_unlock(left);
2128 free_extent_buffer(left);
2130 right = read_node_slot(parent, pslot + 1);
2135 * then try to empty the right most buffer into the middle
2140 btrfs_tree_lock(right);
2141 btrfs_set_lock_blocking_write(right);
2143 right_nr = btrfs_header_nritems(right);
2144 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2147 ret = btrfs_cow_block(trans, root, right,
2153 wret = balance_node_right(trans, right, mid);
2159 struct btrfs_disk_key disk_key;
2161 btrfs_node_key(right, &disk_key, 0);
2162 ret = tree_mod_log_insert_key(parent, pslot + 1,
2163 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2165 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2166 btrfs_mark_buffer_dirty(parent);
2168 if (btrfs_header_nritems(mid) <= orig_slot) {
2169 path->nodes[level] = right;
2170 path->slots[level + 1] += 1;
2171 path->slots[level] = orig_slot -
2172 btrfs_header_nritems(mid);
2173 btrfs_tree_unlock(mid);
2174 free_extent_buffer(mid);
2176 btrfs_tree_unlock(right);
2177 free_extent_buffer(right);
2181 btrfs_tree_unlock(right);
2182 free_extent_buffer(right);
2188 * readahead one full node of leaves, finding things that are close
2189 * to the block in 'slot', and triggering ra on them.
2191 static void reada_for_search(struct btrfs_fs_info *fs_info,
2192 struct btrfs_path *path,
2193 int level, int slot, u64 objectid)
2195 struct extent_buffer *node;
2196 struct btrfs_disk_key disk_key;
2201 struct extent_buffer *eb;
2209 if (!path->nodes[level])
2212 node = path->nodes[level];
2214 search = btrfs_node_blockptr(node, slot);
2215 blocksize = fs_info->nodesize;
2216 eb = find_extent_buffer(fs_info, search);
2218 free_extent_buffer(eb);
2224 nritems = btrfs_header_nritems(node);
2228 if (path->reada == READA_BACK) {
2232 } else if (path->reada == READA_FORWARD) {
2237 if (path->reada == READA_BACK && objectid) {
2238 btrfs_node_key(node, &disk_key, nr);
2239 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2242 search = btrfs_node_blockptr(node, nr);
2243 if ((search <= target && target - search <= 65536) ||
2244 (search > target && search - target <= 65536)) {
2245 readahead_tree_block(fs_info, search);
2249 if ((nread > 65536 || nscan > 32))
2254 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2255 struct btrfs_path *path, int level)
2259 struct extent_buffer *parent;
2260 struct extent_buffer *eb;
2265 parent = path->nodes[level + 1];
2269 nritems = btrfs_header_nritems(parent);
2270 slot = path->slots[level + 1];
2273 block1 = btrfs_node_blockptr(parent, slot - 1);
2274 gen = btrfs_node_ptr_generation(parent, slot - 1);
2275 eb = find_extent_buffer(fs_info, block1);
2277 * if we get -eagain from btrfs_buffer_uptodate, we
2278 * don't want to return eagain here. That will loop
2281 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2283 free_extent_buffer(eb);
2285 if (slot + 1 < nritems) {
2286 block2 = btrfs_node_blockptr(parent, slot + 1);
2287 gen = btrfs_node_ptr_generation(parent, slot + 1);
2288 eb = find_extent_buffer(fs_info, block2);
2289 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2291 free_extent_buffer(eb);
2295 readahead_tree_block(fs_info, block1);
2297 readahead_tree_block(fs_info, block2);
2302 * when we walk down the tree, it is usually safe to unlock the higher layers
2303 * in the tree. The exceptions are when our path goes through slot 0, because
2304 * operations on the tree might require changing key pointers higher up in the
2307 * callers might also have set path->keep_locks, which tells this code to keep
2308 * the lock if the path points to the last slot in the block. This is part of
2309 * walking through the tree, and selecting the next slot in the higher block.
2311 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2312 * if lowest_unlock is 1, level 0 won't be unlocked
2314 static noinline void unlock_up(struct btrfs_path *path, int level,
2315 int lowest_unlock, int min_write_lock_level,
2316 int *write_lock_level)
2319 int skip_level = level;
2321 struct extent_buffer *t;
2323 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2324 if (!path->nodes[i])
2326 if (!path->locks[i])
2328 if (!no_skips && path->slots[i] == 0) {
2332 if (!no_skips && path->keep_locks) {
2335 nritems = btrfs_header_nritems(t);
2336 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2341 if (skip_level < i && i >= lowest_unlock)
2345 if (i >= lowest_unlock && i > skip_level) {
2346 btrfs_tree_unlock_rw(t, path->locks[i]);
2348 if (write_lock_level &&
2349 i > min_write_lock_level &&
2350 i <= *write_lock_level) {
2351 *write_lock_level = i - 1;
2358 * This releases any locks held in the path starting at level and
2359 * going all the way up to the root.
2361 * btrfs_search_slot will keep the lock held on higher nodes in a few
2362 * corner cases, such as COW of the block at slot zero in the node. This
2363 * ignores those rules, and it should only be called when there are no
2364 * more updates to be done higher up in the tree.
2366 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2370 if (path->keep_locks)
2373 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2374 if (!path->nodes[i])
2376 if (!path->locks[i])
2378 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2384 * helper function for btrfs_search_slot. The goal is to find a block
2385 * in cache without setting the path to blocking. If we find the block
2386 * we return zero and the path is unchanged.
2388 * If we can't find the block, we set the path blocking and do some
2389 * reada. -EAGAIN is returned and the search must be repeated.
2392 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2393 struct extent_buffer **eb_ret, int level, int slot,
2394 const struct btrfs_key *key)
2396 struct btrfs_fs_info *fs_info = root->fs_info;
2399 struct extent_buffer *b = *eb_ret;
2400 struct extent_buffer *tmp;
2401 struct btrfs_key first_key;
2405 blocknr = btrfs_node_blockptr(b, slot);
2406 gen = btrfs_node_ptr_generation(b, slot);
2407 parent_level = btrfs_header_level(b);
2408 btrfs_node_key_to_cpu(b, &first_key, slot);
2410 tmp = find_extent_buffer(fs_info, blocknr);
2412 /* first we do an atomic uptodate check */
2413 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2415 * Do extra check for first_key, eb can be stale due to
2416 * being cached, read from scrub, or have multiple
2417 * parents (shared tree blocks).
2419 if (btrfs_verify_level_key(tmp,
2420 parent_level - 1, &first_key, gen)) {
2421 free_extent_buffer(tmp);
2428 /* the pages were up to date, but we failed
2429 * the generation number check. Do a full
2430 * read for the generation number that is correct.
2431 * We must do this without dropping locks so
2432 * we can trust our generation number
2434 btrfs_set_path_blocking(p);
2436 /* now we're allowed to do a blocking uptodate check */
2437 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2442 free_extent_buffer(tmp);
2443 btrfs_release_path(p);
2448 * reduce lock contention at high levels
2449 * of the btree by dropping locks before
2450 * we read. Don't release the lock on the current
2451 * level because we need to walk this node to figure
2452 * out which blocks to read.
2454 btrfs_unlock_up_safe(p, level + 1);
2455 btrfs_set_path_blocking(p);
2457 if (p->reada != READA_NONE)
2458 reada_for_search(fs_info, p, level, slot, key->objectid);
2461 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2465 * If the read above didn't mark this buffer up to date,
2466 * it will never end up being up to date. Set ret to EIO now
2467 * and give up so that our caller doesn't loop forever
2470 if (!extent_buffer_uptodate(tmp))
2472 free_extent_buffer(tmp);
2477 btrfs_release_path(p);
2482 * helper function for btrfs_search_slot. This does all of the checks
2483 * for node-level blocks and does any balancing required based on
2486 * If no extra work was required, zero is returned. If we had to
2487 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2491 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2492 struct btrfs_root *root, struct btrfs_path *p,
2493 struct extent_buffer *b, int level, int ins_len,
2494 int *write_lock_level)
2496 struct btrfs_fs_info *fs_info = root->fs_info;
2499 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2500 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2503 if (*write_lock_level < level + 1) {
2504 *write_lock_level = level + 1;
2505 btrfs_release_path(p);
2509 btrfs_set_path_blocking(p);
2510 reada_for_balance(fs_info, p, level);
2511 sret = split_node(trans, root, p, level);
2518 b = p->nodes[level];
2519 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2520 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2523 if (*write_lock_level < level + 1) {
2524 *write_lock_level = level + 1;
2525 btrfs_release_path(p);
2529 btrfs_set_path_blocking(p);
2530 reada_for_balance(fs_info, p, level);
2531 sret = balance_level(trans, root, p, level);
2537 b = p->nodes[level];
2539 btrfs_release_path(p);
2542 BUG_ON(btrfs_header_nritems(b) == 1);
2552 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2553 int level, int *prev_cmp, int *slot)
2555 if (*prev_cmp != 0) {
2556 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2565 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2566 u64 iobjectid, u64 ioff, u8 key_type,
2567 struct btrfs_key *found_key)
2570 struct btrfs_key key;
2571 struct extent_buffer *eb;
2576 key.type = key_type;
2577 key.objectid = iobjectid;
2580 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2584 eb = path->nodes[0];
2585 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2586 ret = btrfs_next_leaf(fs_root, path);
2589 eb = path->nodes[0];
2592 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2593 if (found_key->type != key.type ||
2594 found_key->objectid != key.objectid)
2600 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2601 struct btrfs_path *p,
2602 int write_lock_level)
2604 struct btrfs_fs_info *fs_info = root->fs_info;
2605 struct extent_buffer *b;
2609 /* We try very hard to do read locks on the root */
2610 root_lock = BTRFS_READ_LOCK;
2612 if (p->search_commit_root) {
2614 * The commit roots are read only so we always do read locks,
2615 * and we always must hold the commit_root_sem when doing
2616 * searches on them, the only exception is send where we don't
2617 * want to block transaction commits for a long time, so
2618 * we need to clone the commit root in order to avoid races
2619 * with transaction commits that create a snapshot of one of
2620 * the roots used by a send operation.
2622 if (p->need_commit_sem) {
2623 down_read(&fs_info->commit_root_sem);
2624 b = btrfs_clone_extent_buffer(root->commit_root);
2625 up_read(&fs_info->commit_root_sem);
2627 return ERR_PTR(-ENOMEM);
2630 b = root->commit_root;
2631 extent_buffer_get(b);
2633 level = btrfs_header_level(b);
2635 * Ensure that all callers have set skip_locking when
2636 * p->search_commit_root = 1.
2638 ASSERT(p->skip_locking == 1);
2643 if (p->skip_locking) {
2644 b = btrfs_root_node(root);
2645 level = btrfs_header_level(b);
2650 * If the level is set to maximum, we can skip trying to get the read
2653 if (write_lock_level < BTRFS_MAX_LEVEL) {
2655 * We don't know the level of the root node until we actually
2656 * have it read locked
2658 b = btrfs_read_lock_root_node(root);
2659 level = btrfs_header_level(b);
2660 if (level > write_lock_level)
2663 /* Whoops, must trade for write lock */
2664 btrfs_tree_read_unlock(b);
2665 free_extent_buffer(b);
2668 b = btrfs_lock_root_node(root);
2669 root_lock = BTRFS_WRITE_LOCK;
2671 /* The level might have changed, check again */
2672 level = btrfs_header_level(b);
2675 p->nodes[level] = b;
2676 if (!p->skip_locking)
2677 p->locks[level] = root_lock;
2679 * Callers are responsible for dropping b's references.
2686 * btrfs_search_slot - look for a key in a tree and perform necessary
2687 * modifications to preserve tree invariants.
2689 * @trans: Handle of transaction, used when modifying the tree
2690 * @p: Holds all btree nodes along the search path
2691 * @root: The root node of the tree
2692 * @key: The key we are looking for
2693 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2694 * deletions it's -1. 0 for plain searches
2695 * @cow: boolean should CoW operations be performed. Must always be 1
2696 * when modifying the tree.
2698 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2699 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2701 * If @key is found, 0 is returned and you can find the item in the leaf level
2702 * of the path (level 0)
2704 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2705 * points to the slot where it should be inserted
2707 * If an error is encountered while searching the tree a negative error number
2710 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2711 const struct btrfs_key *key, struct btrfs_path *p,
2712 int ins_len, int cow)
2714 struct extent_buffer *b;
2719 int lowest_unlock = 1;
2720 /* everything at write_lock_level or lower must be write locked */
2721 int write_lock_level = 0;
2722 u8 lowest_level = 0;
2723 int min_write_lock_level;
2726 lowest_level = p->lowest_level;
2727 WARN_ON(lowest_level && ins_len > 0);
2728 WARN_ON(p->nodes[0] != NULL);
2729 BUG_ON(!cow && ins_len);
2734 /* when we are removing items, we might have to go up to level
2735 * two as we update tree pointers Make sure we keep write
2736 * for those levels as well
2738 write_lock_level = 2;
2739 } else if (ins_len > 0) {
2741 * for inserting items, make sure we have a write lock on
2742 * level 1 so we can update keys
2744 write_lock_level = 1;
2748 write_lock_level = -1;
2750 if (cow && (p->keep_locks || p->lowest_level))
2751 write_lock_level = BTRFS_MAX_LEVEL;
2753 min_write_lock_level = write_lock_level;
2757 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2764 level = btrfs_header_level(b);
2767 * setup the path here so we can release it under lock
2768 * contention with the cow code
2771 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2774 * if we don't really need to cow this block
2775 * then we don't want to set the path blocking,
2776 * so we test it here
2778 if (!should_cow_block(trans, root, b)) {
2779 trans->dirty = true;
2784 * must have write locks on this node and the
2787 if (level > write_lock_level ||
2788 (level + 1 > write_lock_level &&
2789 level + 1 < BTRFS_MAX_LEVEL &&
2790 p->nodes[level + 1])) {
2791 write_lock_level = level + 1;
2792 btrfs_release_path(p);
2796 btrfs_set_path_blocking(p);
2798 err = btrfs_cow_block(trans, root, b, NULL, 0,
2801 err = btrfs_cow_block(trans, root, b,
2802 p->nodes[level + 1],
2803 p->slots[level + 1], &b);
2810 p->nodes[level] = b;
2812 * Leave path with blocking locks to avoid massive
2813 * lock context switch, this is made on purpose.
2817 * we have a lock on b and as long as we aren't changing
2818 * the tree, there is no way to for the items in b to change.
2819 * It is safe to drop the lock on our parent before we
2820 * go through the expensive btree search on b.
2822 * If we're inserting or deleting (ins_len != 0), then we might
2823 * be changing slot zero, which may require changing the parent.
2824 * So, we can't drop the lock until after we know which slot
2825 * we're operating on.
2827 if (!ins_len && !p->keep_locks) {
2830 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2831 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2836 ret = key_search(b, key, level, &prev_cmp, &slot);
2842 if (ret && slot > 0) {
2846 p->slots[level] = slot;
2847 err = setup_nodes_for_search(trans, root, p, b, level,
2848 ins_len, &write_lock_level);
2855 b = p->nodes[level];
2856 slot = p->slots[level];
2859 * slot 0 is special, if we change the key
2860 * we have to update the parent pointer
2861 * which means we must have a write lock
2864 if (slot == 0 && ins_len &&
2865 write_lock_level < level + 1) {
2866 write_lock_level = level + 1;
2867 btrfs_release_path(p);
2871 unlock_up(p, level, lowest_unlock,
2872 min_write_lock_level, &write_lock_level);
2874 if (level == lowest_level) {
2880 err = read_block_for_search(root, p, &b, level,
2889 if (!p->skip_locking) {
2890 level = btrfs_header_level(b);
2891 if (level <= write_lock_level) {
2892 err = btrfs_try_tree_write_lock(b);
2894 btrfs_set_path_blocking(p);
2897 p->locks[level] = BTRFS_WRITE_LOCK;
2899 err = btrfs_tree_read_lock_atomic(b);
2901 btrfs_set_path_blocking(p);
2902 btrfs_tree_read_lock(b);
2904 p->locks[level] = BTRFS_READ_LOCK;
2906 p->nodes[level] = b;
2909 p->slots[level] = slot;
2911 btrfs_leaf_free_space(b) < ins_len) {
2912 if (write_lock_level < 1) {
2913 write_lock_level = 1;
2914 btrfs_release_path(p);
2918 btrfs_set_path_blocking(p);
2919 err = split_leaf(trans, root, key,
2920 p, ins_len, ret == 0);
2928 if (!p->search_for_split)
2929 unlock_up(p, level, lowest_unlock,
2930 min_write_lock_level, NULL);
2937 * we don't really know what they plan on doing with the path
2938 * from here on, so for now just mark it as blocking
2940 if (!p->leave_spinning)
2941 btrfs_set_path_blocking(p);
2942 if (ret < 0 && !p->skip_release_on_error)
2943 btrfs_release_path(p);
2948 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2949 * current state of the tree together with the operations recorded in the tree
2950 * modification log to search for the key in a previous version of this tree, as
2951 * denoted by the time_seq parameter.
2953 * Naturally, there is no support for insert, delete or cow operations.
2955 * The resulting path and return value will be set up as if we called
2956 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2958 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2959 struct btrfs_path *p, u64 time_seq)
2961 struct btrfs_fs_info *fs_info = root->fs_info;
2962 struct extent_buffer *b;
2967 int lowest_unlock = 1;
2968 u8 lowest_level = 0;
2971 lowest_level = p->lowest_level;
2972 WARN_ON(p->nodes[0] != NULL);
2974 if (p->search_commit_root) {
2976 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2980 b = get_old_root(root, time_seq);
2985 level = btrfs_header_level(b);
2986 p->locks[level] = BTRFS_READ_LOCK;
2989 level = btrfs_header_level(b);
2990 p->nodes[level] = b;
2993 * we have a lock on b and as long as we aren't changing
2994 * the tree, there is no way to for the items in b to change.
2995 * It is safe to drop the lock on our parent before we
2996 * go through the expensive btree search on b.
2998 btrfs_unlock_up_safe(p, level + 1);
3001 * Since we can unwind ebs we want to do a real search every
3005 ret = key_search(b, key, level, &prev_cmp, &slot);
3011 if (ret && slot > 0) {
3015 p->slots[level] = slot;
3016 unlock_up(p, level, lowest_unlock, 0, NULL);
3018 if (level == lowest_level) {
3024 err = read_block_for_search(root, p, &b, level,
3033 level = btrfs_header_level(b);
3034 err = btrfs_tree_read_lock_atomic(b);
3036 btrfs_set_path_blocking(p);
3037 btrfs_tree_read_lock(b);
3039 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3044 p->locks[level] = BTRFS_READ_LOCK;
3045 p->nodes[level] = b;
3047 p->slots[level] = slot;
3048 unlock_up(p, level, lowest_unlock, 0, NULL);
3054 if (!p->leave_spinning)
3055 btrfs_set_path_blocking(p);
3057 btrfs_release_path(p);
3063 * helper to use instead of search slot if no exact match is needed but
3064 * instead the next or previous item should be returned.
3065 * When find_higher is true, the next higher item is returned, the next lower
3067 * When return_any and find_higher are both true, and no higher item is found,
3068 * return the next lower instead.
3069 * When return_any is true and find_higher is false, and no lower item is found,
3070 * return the next higher instead.
3071 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3074 int btrfs_search_slot_for_read(struct btrfs_root *root,
3075 const struct btrfs_key *key,
3076 struct btrfs_path *p, int find_higher,
3080 struct extent_buffer *leaf;
3083 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3087 * a return value of 1 means the path is at the position where the
3088 * item should be inserted. Normally this is the next bigger item,
3089 * but in case the previous item is the last in a leaf, path points
3090 * to the first free slot in the previous leaf, i.e. at an invalid
3096 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3097 ret = btrfs_next_leaf(root, p);
3103 * no higher item found, return the next
3108 btrfs_release_path(p);
3112 if (p->slots[0] == 0) {
3113 ret = btrfs_prev_leaf(root, p);
3118 if (p->slots[0] == btrfs_header_nritems(leaf))
3125 * no lower item found, return the next
3130 btrfs_release_path(p);
3140 * adjust the pointers going up the tree, starting at level
3141 * making sure the right key of each node is points to 'key'.
3142 * This is used after shifting pointers to the left, so it stops
3143 * fixing up pointers when a given leaf/node is not in slot 0 of the
3147 static void fixup_low_keys(struct btrfs_path *path,
3148 struct btrfs_disk_key *key, int level)
3151 struct extent_buffer *t;
3154 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3155 int tslot = path->slots[i];
3157 if (!path->nodes[i])
3160 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3163 btrfs_set_node_key(t, key, tslot);
3164 btrfs_mark_buffer_dirty(path->nodes[i]);
3173 * This function isn't completely safe. It's the caller's responsibility
3174 * that the new key won't break the order
3176 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3177 struct btrfs_path *path,
3178 const struct btrfs_key *new_key)
3180 struct btrfs_disk_key disk_key;
3181 struct extent_buffer *eb;
3184 eb = path->nodes[0];
3185 slot = path->slots[0];
3187 btrfs_item_key(eb, &disk_key, slot - 1);
3188 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3190 if (slot < btrfs_header_nritems(eb) - 1) {
3191 btrfs_item_key(eb, &disk_key, slot + 1);
3192 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3195 btrfs_cpu_key_to_disk(&disk_key, new_key);
3196 btrfs_set_item_key(eb, &disk_key, slot);
3197 btrfs_mark_buffer_dirty(eb);
3199 fixup_low_keys(path, &disk_key, 1);
3203 * try to push data from one node into the next node left in the
3206 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3207 * error, and > 0 if there was no room in the left hand block.
3209 static int push_node_left(struct btrfs_trans_handle *trans,
3210 struct extent_buffer *dst,
3211 struct extent_buffer *src, int empty)
3213 struct btrfs_fs_info *fs_info = trans->fs_info;
3219 src_nritems = btrfs_header_nritems(src);
3220 dst_nritems = btrfs_header_nritems(dst);
3221 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3222 WARN_ON(btrfs_header_generation(src) != trans->transid);
3223 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3225 if (!empty && src_nritems <= 8)
3228 if (push_items <= 0)
3232 push_items = min(src_nritems, push_items);
3233 if (push_items < src_nritems) {
3234 /* leave at least 8 pointers in the node if
3235 * we aren't going to empty it
3237 if (src_nritems - push_items < 8) {
3238 if (push_items <= 8)
3244 push_items = min(src_nritems - 8, push_items);
3246 ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3248 btrfs_abort_transaction(trans, ret);
3251 copy_extent_buffer(dst, src,
3252 btrfs_node_key_ptr_offset(dst_nritems),
3253 btrfs_node_key_ptr_offset(0),
3254 push_items * sizeof(struct btrfs_key_ptr));
3256 if (push_items < src_nritems) {
3258 * Don't call tree_mod_log_insert_move here, key removal was
3259 * already fully logged by tree_mod_log_eb_copy above.
3261 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3262 btrfs_node_key_ptr_offset(push_items),
3263 (src_nritems - push_items) *
3264 sizeof(struct btrfs_key_ptr));
3266 btrfs_set_header_nritems(src, src_nritems - push_items);
3267 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3268 btrfs_mark_buffer_dirty(src);
3269 btrfs_mark_buffer_dirty(dst);
3275 * try to push data from one node into the next node right in the
3278 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3279 * error, and > 0 if there was no room in the right hand block.
3281 * this will only push up to 1/2 the contents of the left node over
3283 static int balance_node_right(struct btrfs_trans_handle *trans,
3284 struct extent_buffer *dst,
3285 struct extent_buffer *src)
3287 struct btrfs_fs_info *fs_info = trans->fs_info;
3294 WARN_ON(btrfs_header_generation(src) != trans->transid);
3295 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3297 src_nritems = btrfs_header_nritems(src);
3298 dst_nritems = btrfs_header_nritems(dst);
3299 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3300 if (push_items <= 0)
3303 if (src_nritems < 4)
3306 max_push = src_nritems / 2 + 1;
3307 /* don't try to empty the node */
3308 if (max_push >= src_nritems)
3311 if (max_push < push_items)
3312 push_items = max_push;
3314 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3316 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3317 btrfs_node_key_ptr_offset(0),
3319 sizeof(struct btrfs_key_ptr));
3321 ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3324 btrfs_abort_transaction(trans, ret);
3327 copy_extent_buffer(dst, src,
3328 btrfs_node_key_ptr_offset(0),
3329 btrfs_node_key_ptr_offset(src_nritems - push_items),
3330 push_items * sizeof(struct btrfs_key_ptr));
3332 btrfs_set_header_nritems(src, src_nritems - push_items);
3333 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3335 btrfs_mark_buffer_dirty(src);
3336 btrfs_mark_buffer_dirty(dst);
3342 * helper function to insert a new root level in the tree.
3343 * A new node is allocated, and a single item is inserted to
3344 * point to the existing root
3346 * returns zero on success or < 0 on failure.
3348 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3349 struct btrfs_root *root,
3350 struct btrfs_path *path, int level)
3352 struct btrfs_fs_info *fs_info = root->fs_info;
3354 struct extent_buffer *lower;
3355 struct extent_buffer *c;
3356 struct extent_buffer *old;
3357 struct btrfs_disk_key lower_key;
3360 BUG_ON(path->nodes[level]);
3361 BUG_ON(path->nodes[level-1] != root->node);
3363 lower = path->nodes[level-1];
3365 btrfs_item_key(lower, &lower_key, 0);
3367 btrfs_node_key(lower, &lower_key, 0);
3369 c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3370 root->node->start, 0);
3374 root_add_used(root, fs_info->nodesize);
3376 btrfs_set_header_nritems(c, 1);
3377 btrfs_set_node_key(c, &lower_key, 0);
3378 btrfs_set_node_blockptr(c, 0, lower->start);
3379 lower_gen = btrfs_header_generation(lower);
3380 WARN_ON(lower_gen != trans->transid);
3382 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3384 btrfs_mark_buffer_dirty(c);
3387 ret = tree_mod_log_insert_root(root->node, c, 0);
3389 rcu_assign_pointer(root->node, c);
3391 /* the super has an extra ref to root->node */
3392 free_extent_buffer(old);
3394 add_root_to_dirty_list(root);
3395 extent_buffer_get(c);
3396 path->nodes[level] = c;
3397 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3398 path->slots[level] = 0;
3403 * worker function to insert a single pointer in a node.
3404 * the node should have enough room for the pointer already
3406 * slot and level indicate where you want the key to go, and
3407 * blocknr is the block the key points to.
3409 static void insert_ptr(struct btrfs_trans_handle *trans,
3410 struct btrfs_path *path,
3411 struct btrfs_disk_key *key, u64 bytenr,
3412 int slot, int level)
3414 struct extent_buffer *lower;
3418 BUG_ON(!path->nodes[level]);
3419 btrfs_assert_tree_locked(path->nodes[level]);
3420 lower = path->nodes[level];
3421 nritems = btrfs_header_nritems(lower);
3422 BUG_ON(slot > nritems);
3423 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3424 if (slot != nritems) {
3426 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3430 memmove_extent_buffer(lower,
3431 btrfs_node_key_ptr_offset(slot + 1),
3432 btrfs_node_key_ptr_offset(slot),
3433 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3436 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3440 btrfs_set_node_key(lower, key, slot);
3441 btrfs_set_node_blockptr(lower, slot, bytenr);
3442 WARN_ON(trans->transid == 0);
3443 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3444 btrfs_set_header_nritems(lower, nritems + 1);
3445 btrfs_mark_buffer_dirty(lower);
3449 * split the node at the specified level in path in two.
3450 * The path is corrected to point to the appropriate node after the split
3452 * Before splitting this tries to make some room in the node by pushing
3453 * left and right, if either one works, it returns right away.
3455 * returns 0 on success and < 0 on failure
3457 static noinline int split_node(struct btrfs_trans_handle *trans,
3458 struct btrfs_root *root,
3459 struct btrfs_path *path, int level)
3461 struct btrfs_fs_info *fs_info = root->fs_info;
3462 struct extent_buffer *c;
3463 struct extent_buffer *split;
3464 struct btrfs_disk_key disk_key;
3469 c = path->nodes[level];
3470 WARN_ON(btrfs_header_generation(c) != trans->transid);
3471 if (c == root->node) {
3473 * trying to split the root, lets make a new one
3475 * tree mod log: We don't log_removal old root in
3476 * insert_new_root, because that root buffer will be kept as a
3477 * normal node. We are going to log removal of half of the
3478 * elements below with tree_mod_log_eb_copy. We're holding a
3479 * tree lock on the buffer, which is why we cannot race with
3480 * other tree_mod_log users.
3482 ret = insert_new_root(trans, root, path, level + 1);
3486 ret = push_nodes_for_insert(trans, root, path, level);
3487 c = path->nodes[level];
3488 if (!ret && btrfs_header_nritems(c) <
3489 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3495 c_nritems = btrfs_header_nritems(c);
3496 mid = (c_nritems + 1) / 2;
3497 btrfs_node_key(c, &disk_key, mid);
3499 split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3502 return PTR_ERR(split);
3504 root_add_used(root, fs_info->nodesize);
3505 ASSERT(btrfs_header_level(c) == level);
3507 ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3509 btrfs_abort_transaction(trans, ret);
3512 copy_extent_buffer(split, c,
3513 btrfs_node_key_ptr_offset(0),
3514 btrfs_node_key_ptr_offset(mid),
3515 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3516 btrfs_set_header_nritems(split, c_nritems - mid);
3517 btrfs_set_header_nritems(c, mid);
3520 btrfs_mark_buffer_dirty(c);
3521 btrfs_mark_buffer_dirty(split);
3523 insert_ptr(trans, path, &disk_key, split->start,
3524 path->slots[level + 1] + 1, level + 1);
3526 if (path->slots[level] >= mid) {
3527 path->slots[level] -= mid;
3528 btrfs_tree_unlock(c);
3529 free_extent_buffer(c);
3530 path->nodes[level] = split;
3531 path->slots[level + 1] += 1;
3533 btrfs_tree_unlock(split);
3534 free_extent_buffer(split);
3540 * how many bytes are required to store the items in a leaf. start
3541 * and nr indicate which items in the leaf to check. This totals up the
3542 * space used both by the item structs and the item data
3544 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3546 struct btrfs_item *start_item;
3547 struct btrfs_item *end_item;
3548 struct btrfs_map_token token;
3550 int nritems = btrfs_header_nritems(l);
3551 int end = min(nritems, start + nr) - 1;
3555 btrfs_init_map_token(&token);
3556 start_item = btrfs_item_nr(start);
3557 end_item = btrfs_item_nr(end);
3558 data_len = btrfs_token_item_offset(l, start_item, &token) +
3559 btrfs_token_item_size(l, start_item, &token);
3560 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3561 data_len += sizeof(struct btrfs_item) * nr;
3562 WARN_ON(data_len < 0);
3567 * The space between the end of the leaf items and
3568 * the start of the leaf data. IOW, how much room
3569 * the leaf has left for both items and data
3571 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3573 struct btrfs_fs_info *fs_info = leaf->fs_info;
3574 int nritems = btrfs_header_nritems(leaf);
3577 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3580 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3582 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3583 leaf_space_used(leaf, 0, nritems), nritems);
3589 * min slot controls the lowest index we're willing to push to the
3590 * right. We'll push up to and including min_slot, but no lower
3592 static noinline int __push_leaf_right(struct btrfs_path *path,
3593 int data_size, int empty,
3594 struct extent_buffer *right,
3595 int free_space, u32 left_nritems,
3598 struct btrfs_fs_info *fs_info = right->fs_info;
3599 struct extent_buffer *left = path->nodes[0];
3600 struct extent_buffer *upper = path->nodes[1];
3601 struct btrfs_map_token token;
3602 struct btrfs_disk_key disk_key;
3607 struct btrfs_item *item;
3613 btrfs_init_map_token(&token);
3618 nr = max_t(u32, 1, min_slot);
3620 if (path->slots[0] >= left_nritems)
3621 push_space += data_size;
3623 slot = path->slots[1];
3624 i = left_nritems - 1;
3626 item = btrfs_item_nr(i);
3628 if (!empty && push_items > 0) {
3629 if (path->slots[0] > i)
3631 if (path->slots[0] == i) {
3632 int space = btrfs_leaf_free_space(left);
3634 if (space + push_space * 2 > free_space)
3639 if (path->slots[0] == i)
3640 push_space += data_size;
3642 this_item_size = btrfs_item_size(left, item);
3643 if (this_item_size + sizeof(*item) + push_space > free_space)
3647 push_space += this_item_size + sizeof(*item);
3653 if (push_items == 0)
3656 WARN_ON(!empty && push_items == left_nritems);
3658 /* push left to right */
3659 right_nritems = btrfs_header_nritems(right);
3661 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3662 push_space -= leaf_data_end(left);
3664 /* make room in the right data area */
3665 data_end = leaf_data_end(right);
3666 memmove_extent_buffer(right,
3667 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3668 BTRFS_LEAF_DATA_OFFSET + data_end,
3669 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3671 /* copy from the left data area */
3672 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3673 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3674 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3677 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3678 btrfs_item_nr_offset(0),
3679 right_nritems * sizeof(struct btrfs_item));
3681 /* copy the items from left to right */
3682 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3683 btrfs_item_nr_offset(left_nritems - push_items),
3684 push_items * sizeof(struct btrfs_item));
3686 /* update the item pointers */
3687 right_nritems += push_items;
3688 btrfs_set_header_nritems(right, right_nritems);
3689 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3690 for (i = 0; i < right_nritems; i++) {
3691 item = btrfs_item_nr(i);
3692 push_space -= btrfs_token_item_size(right, item, &token);
3693 btrfs_set_token_item_offset(right, item, push_space, &token);
3696 left_nritems -= push_items;
3697 btrfs_set_header_nritems(left, left_nritems);
3700 btrfs_mark_buffer_dirty(left);
3702 btrfs_clean_tree_block(left);
3704 btrfs_mark_buffer_dirty(right);
3706 btrfs_item_key(right, &disk_key, 0);
3707 btrfs_set_node_key(upper, &disk_key, slot + 1);
3708 btrfs_mark_buffer_dirty(upper);
3710 /* then fixup the leaf pointer in the path */
3711 if (path->slots[0] >= left_nritems) {
3712 path->slots[0] -= left_nritems;
3713 if (btrfs_header_nritems(path->nodes[0]) == 0)
3714 btrfs_clean_tree_block(path->nodes[0]);
3715 btrfs_tree_unlock(path->nodes[0]);
3716 free_extent_buffer(path->nodes[0]);
3717 path->nodes[0] = right;
3718 path->slots[1] += 1;
3720 btrfs_tree_unlock(right);
3721 free_extent_buffer(right);
3726 btrfs_tree_unlock(right);
3727 free_extent_buffer(right);
3732 * push some data in the path leaf to the right, trying to free up at
3733 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3735 * returns 1 if the push failed because the other node didn't have enough
3736 * room, 0 if everything worked out and < 0 if there were major errors.
3738 * this will push starting from min_slot to the end of the leaf. It won't
3739 * push any slot lower than min_slot
3741 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3742 *root, struct btrfs_path *path,
3743 int min_data_size, int data_size,
3744 int empty, u32 min_slot)
3746 struct extent_buffer *left = path->nodes[0];
3747 struct extent_buffer *right;
3748 struct extent_buffer *upper;
3754 if (!path->nodes[1])
3757 slot = path->slots[1];
3758 upper = path->nodes[1];
3759 if (slot >= btrfs_header_nritems(upper) - 1)
3762 btrfs_assert_tree_locked(path->nodes[1]);
3764 right = read_node_slot(upper, slot + 1);
3766 * slot + 1 is not valid or we fail to read the right node,
3767 * no big deal, just return.
3772 btrfs_tree_lock(right);
3773 btrfs_set_lock_blocking_write(right);
3775 free_space = btrfs_leaf_free_space(right);
3776 if (free_space < data_size)
3779 /* cow and double check */
3780 ret = btrfs_cow_block(trans, root, right, upper,
3785 free_space = btrfs_leaf_free_space(right);
3786 if (free_space < data_size)
3789 left_nritems = btrfs_header_nritems(left);
3790 if (left_nritems == 0)
3793 if (path->slots[0] == left_nritems && !empty) {
3794 /* Key greater than all keys in the leaf, right neighbor has
3795 * enough room for it and we're not emptying our leaf to delete
3796 * it, therefore use right neighbor to insert the new item and
3797 * no need to touch/dirty our left leaf. */
3798 btrfs_tree_unlock(left);
3799 free_extent_buffer(left);
3800 path->nodes[0] = right;
3806 return __push_leaf_right(path, min_data_size, empty,
3807 right, free_space, left_nritems, min_slot);
3809 btrfs_tree_unlock(right);
3810 free_extent_buffer(right);
3815 * push some data in the path leaf to the left, trying to free up at
3816 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3818 * max_slot can put a limit on how far into the leaf we'll push items. The
3819 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3822 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3823 int empty, struct extent_buffer *left,
3824 int free_space, u32 right_nritems,
3827 struct btrfs_fs_info *fs_info = left->fs_info;
3828 struct btrfs_disk_key disk_key;
3829 struct extent_buffer *right = path->nodes[0];
3833 struct btrfs_item *item;
3834 u32 old_left_nritems;
3838 u32 old_left_item_size;
3839 struct btrfs_map_token token;
3841 btrfs_init_map_token(&token);
3844 nr = min(right_nritems, max_slot);
3846 nr = min(right_nritems - 1, max_slot);
3848 for (i = 0; i < nr; i++) {
3849 item = btrfs_item_nr(i);
3851 if (!empty && push_items > 0) {
3852 if (path->slots[0] < i)
3854 if (path->slots[0] == i) {
3855 int space = btrfs_leaf_free_space(right);
3857 if (space + push_space * 2 > free_space)
3862 if (path->slots[0] == i)
3863 push_space += data_size;
3865 this_item_size = btrfs_item_size(right, item);
3866 if (this_item_size + sizeof(*item) + push_space > free_space)
3870 push_space += this_item_size + sizeof(*item);
3873 if (push_items == 0) {
3877 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3879 /* push data from right to left */
3880 copy_extent_buffer(left, right,
3881 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3882 btrfs_item_nr_offset(0),
3883 push_items * sizeof(struct btrfs_item));
3885 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3886 btrfs_item_offset_nr(right, push_items - 1);
3888 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3889 leaf_data_end(left) - push_space,
3890 BTRFS_LEAF_DATA_OFFSET +
3891 btrfs_item_offset_nr(right, push_items - 1),
3893 old_left_nritems = btrfs_header_nritems(left);
3894 BUG_ON(old_left_nritems <= 0);
3896 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3897 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3900 item = btrfs_item_nr(i);
3902 ioff = btrfs_token_item_offset(left, item, &token);
3903 btrfs_set_token_item_offset(left, item,
3904 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3907 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3909 /* fixup right node */
3910 if (push_items > right_nritems)
3911 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3914 if (push_items < right_nritems) {
3915 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3916 leaf_data_end(right);
3917 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3918 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3919 BTRFS_LEAF_DATA_OFFSET +
3920 leaf_data_end(right), push_space);
3922 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3923 btrfs_item_nr_offset(push_items),
3924 (btrfs_header_nritems(right) - push_items) *
3925 sizeof(struct btrfs_item));
3927 right_nritems -= push_items;
3928 btrfs_set_header_nritems(right, right_nritems);
3929 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3930 for (i = 0; i < right_nritems; i++) {
3931 item = btrfs_item_nr(i);
3933 push_space = push_space - btrfs_token_item_size(right,
3935 btrfs_set_token_item_offset(right, item, push_space, &token);
3938 btrfs_mark_buffer_dirty(left);
3940 btrfs_mark_buffer_dirty(right);
3942 btrfs_clean_tree_block(right);
3944 btrfs_item_key(right, &disk_key, 0);
3945 fixup_low_keys(path, &disk_key, 1);
3947 /* then fixup the leaf pointer in the path */
3948 if (path->slots[0] < push_items) {
3949 path->slots[0] += old_left_nritems;
3950 btrfs_tree_unlock(path->nodes[0]);
3951 free_extent_buffer(path->nodes[0]);
3952 path->nodes[0] = left;
3953 path->slots[1] -= 1;
3955 btrfs_tree_unlock(left);
3956 free_extent_buffer(left);
3957 path->slots[0] -= push_items;
3959 BUG_ON(path->slots[0] < 0);
3962 btrfs_tree_unlock(left);
3963 free_extent_buffer(left);
3968 * push some data in the path leaf to the left, trying to free up at
3969 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3971 * max_slot can put a limit on how far into the leaf we'll push items. The
3972 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3975 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3976 *root, struct btrfs_path *path, int min_data_size,
3977 int data_size, int empty, u32 max_slot)
3979 struct extent_buffer *right = path->nodes[0];
3980 struct extent_buffer *left;
3986 slot = path->slots[1];
3989 if (!path->nodes[1])
3992 right_nritems = btrfs_header_nritems(right);
3993 if (right_nritems == 0)
3996 btrfs_assert_tree_locked(path->nodes[1]);
3998 left = read_node_slot(path->nodes[1], slot - 1);
4000 * slot - 1 is not valid or we fail to read the left node,
4001 * no big deal, just return.
4006 btrfs_tree_lock(left);
4007 btrfs_set_lock_blocking_write(left);
4009 free_space = btrfs_leaf_free_space(left);
4010 if (free_space < data_size) {
4015 /* cow and double check */
4016 ret = btrfs_cow_block(trans, root, left,
4017 path->nodes[1], slot - 1, &left);
4019 /* we hit -ENOSPC, but it isn't fatal here */
4025 free_space = btrfs_leaf_free_space(left);
4026 if (free_space < data_size) {
4031 return __push_leaf_left(path, min_data_size,
4032 empty, left, free_space, right_nritems,
4035 btrfs_tree_unlock(left);
4036 free_extent_buffer(left);
4041 * split the path's leaf in two, making sure there is at least data_size
4042 * available for the resulting leaf level of the path.
4044 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4045 struct btrfs_path *path,
4046 struct extent_buffer *l,
4047 struct extent_buffer *right,
4048 int slot, int mid, int nritems)
4050 struct btrfs_fs_info *fs_info = trans->fs_info;
4054 struct btrfs_disk_key disk_key;
4055 struct btrfs_map_token token;
4057 btrfs_init_map_token(&token);
4059 nritems = nritems - mid;
4060 btrfs_set_header_nritems(right, nritems);
4061 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4063 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4064 btrfs_item_nr_offset(mid),
4065 nritems * sizeof(struct btrfs_item));
4067 copy_extent_buffer(right, l,
4068 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4069 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4070 leaf_data_end(l), data_copy_size);
4072 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4074 for (i = 0; i < nritems; i++) {
4075 struct btrfs_item *item = btrfs_item_nr(i);
4078 ioff = btrfs_token_item_offset(right, item, &token);
4079 btrfs_set_token_item_offset(right, item,
4080 ioff + rt_data_off, &token);
4083 btrfs_set_header_nritems(l, mid);
4084 btrfs_item_key(right, &disk_key, 0);
4085 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4087 btrfs_mark_buffer_dirty(right);
4088 btrfs_mark_buffer_dirty(l);
4089 BUG_ON(path->slots[0] != slot);
4092 btrfs_tree_unlock(path->nodes[0]);
4093 free_extent_buffer(path->nodes[0]);
4094 path->nodes[0] = right;
4095 path->slots[0] -= mid;
4096 path->slots[1] += 1;
4098 btrfs_tree_unlock(right);
4099 free_extent_buffer(right);
4102 BUG_ON(path->slots[0] < 0);
4106 * double splits happen when we need to insert a big item in the middle
4107 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4108 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4111 * We avoid this by trying to push the items on either side of our target
4112 * into the adjacent leaves. If all goes well we can avoid the double split
4115 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4116 struct btrfs_root *root,
4117 struct btrfs_path *path,
4124 int space_needed = data_size;
4126 slot = path->slots[0];
4127 if (slot < btrfs_header_nritems(path->nodes[0]))
4128 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4131 * try to push all the items after our slot into the
4134 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4141 nritems = btrfs_header_nritems(path->nodes[0]);
4143 * our goal is to get our slot at the start or end of a leaf. If
4144 * we've done so we're done
4146 if (path->slots[0] == 0 || path->slots[0] == nritems)
4149 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4152 /* try to push all the items before our slot into the next leaf */
4153 slot = path->slots[0];
4154 space_needed = data_size;
4156 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4157 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4170 * split the path's leaf in two, making sure there is at least data_size
4171 * available for the resulting leaf level of the path.
4173 * returns 0 if all went well and < 0 on failure.
4175 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4176 struct btrfs_root *root,
4177 const struct btrfs_key *ins_key,
4178 struct btrfs_path *path, int data_size,
4181 struct btrfs_disk_key disk_key;
4182 struct extent_buffer *l;
4186 struct extent_buffer *right;
4187 struct btrfs_fs_info *fs_info = root->fs_info;
4191 int num_doubles = 0;
4192 int tried_avoid_double = 0;
4195 slot = path->slots[0];
4196 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4197 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4200 /* first try to make some room by pushing left and right */
4201 if (data_size && path->nodes[1]) {
4202 int space_needed = data_size;
4204 if (slot < btrfs_header_nritems(l))
4205 space_needed -= btrfs_leaf_free_space(l);
4207 wret = push_leaf_right(trans, root, path, space_needed,
4208 space_needed, 0, 0);
4212 space_needed = data_size;
4214 space_needed -= btrfs_leaf_free_space(l);
4215 wret = push_leaf_left(trans, root, path, space_needed,
4216 space_needed, 0, (u32)-1);
4222 /* did the pushes work? */
4223 if (btrfs_leaf_free_space(l) >= data_size)
4227 if (!path->nodes[1]) {
4228 ret = insert_new_root(trans, root, path, 1);
4235 slot = path->slots[0];
4236 nritems = btrfs_header_nritems(l);
4237 mid = (nritems + 1) / 2;
4241 leaf_space_used(l, mid, nritems - mid) + data_size >
4242 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4243 if (slot >= nritems) {
4247 if (mid != nritems &&
4248 leaf_space_used(l, mid, nritems - mid) +
4249 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4250 if (data_size && !tried_avoid_double)
4251 goto push_for_double;
4257 if (leaf_space_used(l, 0, mid) + data_size >
4258 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4259 if (!extend && data_size && slot == 0) {
4261 } else if ((extend || !data_size) && slot == 0) {
4265 if (mid != nritems &&
4266 leaf_space_used(l, mid, nritems - mid) +
4267 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4268 if (data_size && !tried_avoid_double)
4269 goto push_for_double;
4277 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4279 btrfs_item_key(l, &disk_key, mid);
4281 right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4284 return PTR_ERR(right);
4286 root_add_used(root, fs_info->nodesize);
4290 btrfs_set_header_nritems(right, 0);
4291 insert_ptr(trans, path, &disk_key,
4292 right->start, path->slots[1] + 1, 1);
4293 btrfs_tree_unlock(path->nodes[0]);
4294 free_extent_buffer(path->nodes[0]);
4295 path->nodes[0] = right;
4297 path->slots[1] += 1;
4299 btrfs_set_header_nritems(right, 0);
4300 insert_ptr(trans, path, &disk_key,
4301 right->start, path->slots[1], 1);
4302 btrfs_tree_unlock(path->nodes[0]);
4303 free_extent_buffer(path->nodes[0]);
4304 path->nodes[0] = right;
4306 if (path->slots[1] == 0)
4307 fixup_low_keys(path, &disk_key, 1);
4310 * We create a new leaf 'right' for the required ins_len and
4311 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4312 * the content of ins_len to 'right'.
4317 copy_for_split(trans, path, l, right, slot, mid, nritems);
4320 BUG_ON(num_doubles != 0);
4328 push_for_double_split(trans, root, path, data_size);
4329 tried_avoid_double = 1;
4330 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4335 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4336 struct btrfs_root *root,
4337 struct btrfs_path *path, int ins_len)
4339 struct btrfs_key key;
4340 struct extent_buffer *leaf;
4341 struct btrfs_file_extent_item *fi;
4346 leaf = path->nodes[0];
4347 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4349 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4350 key.type != BTRFS_EXTENT_CSUM_KEY);
4352 if (btrfs_leaf_free_space(leaf) >= ins_len)
4355 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4356 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4357 fi = btrfs_item_ptr(leaf, path->slots[0],
4358 struct btrfs_file_extent_item);
4359 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4361 btrfs_release_path(path);
4363 path->keep_locks = 1;
4364 path->search_for_split = 1;
4365 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4366 path->search_for_split = 0;
4373 leaf = path->nodes[0];
4374 /* if our item isn't there, return now */
4375 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4378 /* the leaf has changed, it now has room. return now */
4379 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4382 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4383 fi = btrfs_item_ptr(leaf, path->slots[0],
4384 struct btrfs_file_extent_item);
4385 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4389 btrfs_set_path_blocking(path);
4390 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4394 path->keep_locks = 0;
4395 btrfs_unlock_up_safe(path, 1);
4398 path->keep_locks = 0;
4402 static noinline int split_item(struct btrfs_path *path,
4403 const struct btrfs_key *new_key,
4404 unsigned long split_offset)
4406 struct extent_buffer *leaf;
4407 struct btrfs_item *item;
4408 struct btrfs_item *new_item;
4414 struct btrfs_disk_key disk_key;
4416 leaf = path->nodes[0];
4417 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4419 btrfs_set_path_blocking(path);
4421 item = btrfs_item_nr(path->slots[0]);
4422 orig_offset = btrfs_item_offset(leaf, item);
4423 item_size = btrfs_item_size(leaf, item);
4425 buf = kmalloc(item_size, GFP_NOFS);
4429 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4430 path->slots[0]), item_size);
4432 slot = path->slots[0] + 1;
4433 nritems = btrfs_header_nritems(leaf);
4434 if (slot != nritems) {
4435 /* shift the items */
4436 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4437 btrfs_item_nr_offset(slot),
4438 (nritems - slot) * sizeof(struct btrfs_item));
4441 btrfs_cpu_key_to_disk(&disk_key, new_key);
4442 btrfs_set_item_key(leaf, &disk_key, slot);
4444 new_item = btrfs_item_nr(slot);
4446 btrfs_set_item_offset(leaf, new_item, orig_offset);
4447 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4449 btrfs_set_item_offset(leaf, item,
4450 orig_offset + item_size - split_offset);
4451 btrfs_set_item_size(leaf, item, split_offset);
4453 btrfs_set_header_nritems(leaf, nritems + 1);
4455 /* write the data for the start of the original item */
4456 write_extent_buffer(leaf, buf,
4457 btrfs_item_ptr_offset(leaf, path->slots[0]),
4460 /* write the data for the new item */
4461 write_extent_buffer(leaf, buf + split_offset,
4462 btrfs_item_ptr_offset(leaf, slot),
4463 item_size - split_offset);
4464 btrfs_mark_buffer_dirty(leaf);
4466 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4472 * This function splits a single item into two items,
4473 * giving 'new_key' to the new item and splitting the
4474 * old one at split_offset (from the start of the item).
4476 * The path may be released by this operation. After
4477 * the split, the path is pointing to the old item. The
4478 * new item is going to be in the same node as the old one.
4480 * Note, the item being split must be smaller enough to live alone on
4481 * a tree block with room for one extra struct btrfs_item
4483 * This allows us to split the item in place, keeping a lock on the
4484 * leaf the entire time.
4486 int btrfs_split_item(struct btrfs_trans_handle *trans,
4487 struct btrfs_root *root,
4488 struct btrfs_path *path,
4489 const struct btrfs_key *new_key,
4490 unsigned long split_offset)
4493 ret = setup_leaf_for_split(trans, root, path,
4494 sizeof(struct btrfs_item));
4498 ret = split_item(path, new_key, split_offset);
4503 * This function duplicate a item, giving 'new_key' to the new item.
4504 * It guarantees both items live in the same tree leaf and the new item
4505 * is contiguous with the original item.
4507 * This allows us to split file extent in place, keeping a lock on the
4508 * leaf the entire time.
4510 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4511 struct btrfs_root *root,
4512 struct btrfs_path *path,
4513 const struct btrfs_key *new_key)
4515 struct extent_buffer *leaf;
4519 leaf = path->nodes[0];
4520 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4521 ret = setup_leaf_for_split(trans, root, path,
4522 item_size + sizeof(struct btrfs_item));
4527 setup_items_for_insert(root, path, new_key, &item_size,
4528 item_size, item_size +
4529 sizeof(struct btrfs_item), 1);
4530 leaf = path->nodes[0];
4531 memcpy_extent_buffer(leaf,
4532 btrfs_item_ptr_offset(leaf, path->slots[0]),
4533 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4539 * make the item pointed to by the path smaller. new_size indicates
4540 * how small to make it, and from_end tells us if we just chop bytes
4541 * off the end of the item or if we shift the item to chop bytes off
4544 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4547 struct extent_buffer *leaf;
4548 struct btrfs_item *item;
4550 unsigned int data_end;
4551 unsigned int old_data_start;
4552 unsigned int old_size;
4553 unsigned int size_diff;
4555 struct btrfs_map_token token;
4557 btrfs_init_map_token(&token);
4559 leaf = path->nodes[0];
4560 slot = path->slots[0];
4562 old_size = btrfs_item_size_nr(leaf, slot);
4563 if (old_size == new_size)
4566 nritems = btrfs_header_nritems(leaf);
4567 data_end = leaf_data_end(leaf);
4569 old_data_start = btrfs_item_offset_nr(leaf, slot);
4571 size_diff = old_size - new_size;
4574 BUG_ON(slot >= nritems);
4577 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4579 /* first correct the data pointers */
4580 for (i = slot; i < nritems; i++) {
4582 item = btrfs_item_nr(i);
4584 ioff = btrfs_token_item_offset(leaf, item, &token);
4585 btrfs_set_token_item_offset(leaf, item,
4586 ioff + size_diff, &token);
4589 /* shift the data */
4591 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4592 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4593 data_end, old_data_start + new_size - data_end);
4595 struct btrfs_disk_key disk_key;
4598 btrfs_item_key(leaf, &disk_key, slot);
4600 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4602 struct btrfs_file_extent_item *fi;
4604 fi = btrfs_item_ptr(leaf, slot,
4605 struct btrfs_file_extent_item);
4606 fi = (struct btrfs_file_extent_item *)(
4607 (unsigned long)fi - size_diff);
4609 if (btrfs_file_extent_type(leaf, fi) ==
4610 BTRFS_FILE_EXTENT_INLINE) {
4611 ptr = btrfs_item_ptr_offset(leaf, slot);
4612 memmove_extent_buffer(leaf, ptr,
4614 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4618 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4619 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4620 data_end, old_data_start - data_end);
4622 offset = btrfs_disk_key_offset(&disk_key);
4623 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4624 btrfs_set_item_key(leaf, &disk_key, slot);
4626 fixup_low_keys(path, &disk_key, 1);
4629 item = btrfs_item_nr(slot);
4630 btrfs_set_item_size(leaf, item, new_size);
4631 btrfs_mark_buffer_dirty(leaf);
4633 if (btrfs_leaf_free_space(leaf) < 0) {
4634 btrfs_print_leaf(leaf);
4640 * make the item pointed to by the path bigger, data_size is the added size.
4642 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4646 struct extent_buffer *leaf;
4647 struct btrfs_item *item;
4649 unsigned int data_end;
4650 unsigned int old_data;
4651 unsigned int old_size;
4653 struct btrfs_map_token token;
4655 btrfs_init_map_token(&token);
4657 leaf = path->nodes[0];
4659 nritems = btrfs_header_nritems(leaf);
4660 data_end = leaf_data_end(leaf);
4662 if (btrfs_leaf_free_space(leaf) < data_size) {
4663 btrfs_print_leaf(leaf);
4666 slot = path->slots[0];
4667 old_data = btrfs_item_end_nr(leaf, slot);
4670 if (slot >= nritems) {
4671 btrfs_print_leaf(leaf);
4672 btrfs_crit(fs_info, "slot %d too large, nritems %d",
4678 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4680 /* first correct the data pointers */
4681 for (i = slot; i < nritems; i++) {
4683 item = btrfs_item_nr(i);
4685 ioff = btrfs_token_item_offset(leaf, item, &token);
4686 btrfs_set_token_item_offset(leaf, item,
4687 ioff - data_size, &token);
4690 /* shift the data */
4691 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4692 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4693 data_end, old_data - data_end);
4695 data_end = old_data;
4696 old_size = btrfs_item_size_nr(leaf, slot);
4697 item = btrfs_item_nr(slot);
4698 btrfs_set_item_size(leaf, item, old_size + data_size);
4699 btrfs_mark_buffer_dirty(leaf);
4701 if (btrfs_leaf_free_space(leaf) < 0) {
4702 btrfs_print_leaf(leaf);
4708 * this is a helper for btrfs_insert_empty_items, the main goal here is
4709 * to save stack depth by doing the bulk of the work in a function
4710 * that doesn't call btrfs_search_slot
4712 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4713 const struct btrfs_key *cpu_key, u32 *data_size,
4714 u32 total_data, u32 total_size, int nr)
4716 struct btrfs_fs_info *fs_info = root->fs_info;
4717 struct btrfs_item *item;
4720 unsigned int data_end;
4721 struct btrfs_disk_key disk_key;
4722 struct extent_buffer *leaf;
4724 struct btrfs_map_token token;
4726 if (path->slots[0] == 0) {
4727 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4728 fixup_low_keys(path, &disk_key, 1);
4730 btrfs_unlock_up_safe(path, 1);
4732 btrfs_init_map_token(&token);
4734 leaf = path->nodes[0];
4735 slot = path->slots[0];
4737 nritems = btrfs_header_nritems(leaf);
4738 data_end = leaf_data_end(leaf);
4740 if (btrfs_leaf_free_space(leaf) < total_size) {
4741 btrfs_print_leaf(leaf);
4742 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4743 total_size, btrfs_leaf_free_space(leaf));
4747 if (slot != nritems) {
4748 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4750 if (old_data < data_end) {
4751 btrfs_print_leaf(leaf);
4752 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4753 slot, old_data, data_end);
4757 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4759 /* first correct the data pointers */
4760 for (i = slot; i < nritems; i++) {
4763 item = btrfs_item_nr(i);
4764 ioff = btrfs_token_item_offset(leaf, item, &token);
4765 btrfs_set_token_item_offset(leaf, item,
4766 ioff - total_data, &token);
4768 /* shift the items */
4769 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4770 btrfs_item_nr_offset(slot),
4771 (nritems - slot) * sizeof(struct btrfs_item));
4773 /* shift the data */
4774 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4775 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4776 data_end, old_data - data_end);
4777 data_end = old_data;
4780 /* setup the item for the new data */
4781 for (i = 0; i < nr; i++) {
4782 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4783 btrfs_set_item_key(leaf, &disk_key, slot + i);
4784 item = btrfs_item_nr(slot + i);
4785 btrfs_set_token_item_offset(leaf, item,
4786 data_end - data_size[i], &token);
4787 data_end -= data_size[i];
4788 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4791 btrfs_set_header_nritems(leaf, nritems + nr);
4792 btrfs_mark_buffer_dirty(leaf);
4794 if (btrfs_leaf_free_space(leaf) < 0) {
4795 btrfs_print_leaf(leaf);
4801 * Given a key and some data, insert items into the tree.
4802 * This does all the path init required, making room in the tree if needed.
4804 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4805 struct btrfs_root *root,
4806 struct btrfs_path *path,
4807 const struct btrfs_key *cpu_key, u32 *data_size,
4816 for (i = 0; i < nr; i++)
4817 total_data += data_size[i];
4819 total_size = total_data + (nr * sizeof(struct btrfs_item));
4820 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4826 slot = path->slots[0];
4829 setup_items_for_insert(root, path, cpu_key, data_size,
4830 total_data, total_size, nr);
4835 * Given a key and some data, insert an item into the tree.
4836 * This does all the path init required, making room in the tree if needed.
4838 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4839 const struct btrfs_key *cpu_key, void *data,
4843 struct btrfs_path *path;
4844 struct extent_buffer *leaf;
4847 path = btrfs_alloc_path();
4850 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4852 leaf = path->nodes[0];
4853 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4854 write_extent_buffer(leaf, data, ptr, data_size);
4855 btrfs_mark_buffer_dirty(leaf);
4857 btrfs_free_path(path);
4862 * delete the pointer from a given node.
4864 * the tree should have been previously balanced so the deletion does not
4867 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4868 int level, int slot)
4870 struct extent_buffer *parent = path->nodes[level];
4874 nritems = btrfs_header_nritems(parent);
4875 if (slot != nritems - 1) {
4877 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4878 nritems - slot - 1);
4881 memmove_extent_buffer(parent,
4882 btrfs_node_key_ptr_offset(slot),
4883 btrfs_node_key_ptr_offset(slot + 1),
4884 sizeof(struct btrfs_key_ptr) *
4885 (nritems - slot - 1));
4887 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4893 btrfs_set_header_nritems(parent, nritems);
4894 if (nritems == 0 && parent == root->node) {
4895 BUG_ON(btrfs_header_level(root->node) != 1);
4896 /* just turn the root into a leaf and break */
4897 btrfs_set_header_level(root->node, 0);
4898 } else if (slot == 0) {
4899 struct btrfs_disk_key disk_key;
4901 btrfs_node_key(parent, &disk_key, 0);
4902 fixup_low_keys(path, &disk_key, level + 1);
4904 btrfs_mark_buffer_dirty(parent);
4908 * a helper function to delete the leaf pointed to by path->slots[1] and
4911 * This deletes the pointer in path->nodes[1] and frees the leaf
4912 * block extent. zero is returned if it all worked out, < 0 otherwise.
4914 * The path must have already been setup for deleting the leaf, including
4915 * all the proper balancing. path->nodes[1] must be locked.
4917 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4918 struct btrfs_root *root,
4919 struct btrfs_path *path,
4920 struct extent_buffer *leaf)
4922 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4923 del_ptr(root, path, 1, path->slots[1]);
4926 * btrfs_free_extent is expensive, we want to make sure we
4927 * aren't holding any locks when we call it
4929 btrfs_unlock_up_safe(path, 0);
4931 root_sub_used(root, leaf->len);
4933 extent_buffer_get(leaf);
4934 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4935 free_extent_buffer_stale(leaf);
4938 * delete the item at the leaf level in path. If that empties
4939 * the leaf, remove it from the tree
4941 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4942 struct btrfs_path *path, int slot, int nr)
4944 struct btrfs_fs_info *fs_info = root->fs_info;
4945 struct extent_buffer *leaf;
4946 struct btrfs_item *item;
4953 struct btrfs_map_token token;
4955 btrfs_init_map_token(&token);
4957 leaf = path->nodes[0];
4958 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4960 for (i = 0; i < nr; i++)
4961 dsize += btrfs_item_size_nr(leaf, slot + i);
4963 nritems = btrfs_header_nritems(leaf);
4965 if (slot + nr != nritems) {
4966 int data_end = leaf_data_end(leaf);
4968 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4970 BTRFS_LEAF_DATA_OFFSET + data_end,
4971 last_off - data_end);
4973 for (i = slot + nr; i < nritems; i++) {
4976 item = btrfs_item_nr(i);
4977 ioff = btrfs_token_item_offset(leaf, item, &token);
4978 btrfs_set_token_item_offset(leaf, item,
4979 ioff + dsize, &token);
4982 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4983 btrfs_item_nr_offset(slot + nr),
4984 sizeof(struct btrfs_item) *
4985 (nritems - slot - nr));
4987 btrfs_set_header_nritems(leaf, nritems - nr);
4990 /* delete the leaf if we've emptied it */
4992 if (leaf == root->node) {
4993 btrfs_set_header_level(leaf, 0);
4995 btrfs_set_path_blocking(path);
4996 btrfs_clean_tree_block(leaf);
4997 btrfs_del_leaf(trans, root, path, leaf);
5000 int used = leaf_space_used(leaf, 0, nritems);
5002 struct btrfs_disk_key disk_key;
5004 btrfs_item_key(leaf, &disk_key, 0);
5005 fixup_low_keys(path, &disk_key, 1);
5008 /* delete the leaf if it is mostly empty */
5009 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5010 /* push_leaf_left fixes the path.
5011 * make sure the path still points to our leaf
5012 * for possible call to del_ptr below
5014 slot = path->slots[1];
5015 extent_buffer_get(leaf);
5017 btrfs_set_path_blocking(path);
5018 wret = push_leaf_left(trans, root, path, 1, 1,
5020 if (wret < 0 && wret != -ENOSPC)
5023 if (path->nodes[0] == leaf &&
5024 btrfs_header_nritems(leaf)) {
5025 wret = push_leaf_right(trans, root, path, 1,
5027 if (wret < 0 && wret != -ENOSPC)
5031 if (btrfs_header_nritems(leaf) == 0) {
5032 path->slots[1] = slot;
5033 btrfs_del_leaf(trans, root, path, leaf);
5034 free_extent_buffer(leaf);
5037 /* if we're still in the path, make sure
5038 * we're dirty. Otherwise, one of the
5039 * push_leaf functions must have already
5040 * dirtied this buffer
5042 if (path->nodes[0] == leaf)
5043 btrfs_mark_buffer_dirty(leaf);
5044 free_extent_buffer(leaf);
5047 btrfs_mark_buffer_dirty(leaf);
5054 * search the tree again to find a leaf with lesser keys
5055 * returns 0 if it found something or 1 if there are no lesser leaves.
5056 * returns < 0 on io errors.
5058 * This may release the path, and so you may lose any locks held at the
5061 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5063 struct btrfs_key key;
5064 struct btrfs_disk_key found_key;
5067 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5069 if (key.offset > 0) {
5071 } else if (key.type > 0) {
5073 key.offset = (u64)-1;
5074 } else if (key.objectid > 0) {
5077 key.offset = (u64)-1;
5082 btrfs_release_path(path);
5083 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5086 btrfs_item_key(path->nodes[0], &found_key, 0);
5087 ret = comp_keys(&found_key, &key);
5089 * We might have had an item with the previous key in the tree right
5090 * before we released our path. And after we released our path, that
5091 * item might have been pushed to the first slot (0) of the leaf we
5092 * were holding due to a tree balance. Alternatively, an item with the
5093 * previous key can exist as the only element of a leaf (big fat item).
5094 * Therefore account for these 2 cases, so that our callers (like
5095 * btrfs_previous_item) don't miss an existing item with a key matching
5096 * the previous key we computed above.
5104 * A helper function to walk down the tree starting at min_key, and looking
5105 * for nodes or leaves that are have a minimum transaction id.
5106 * This is used by the btree defrag code, and tree logging
5108 * This does not cow, but it does stuff the starting key it finds back
5109 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5110 * key and get a writable path.
5112 * This honors path->lowest_level to prevent descent past a given level
5115 * min_trans indicates the oldest transaction that you are interested
5116 * in walking through. Any nodes or leaves older than min_trans are
5117 * skipped over (without reading them).
5119 * returns zero if something useful was found, < 0 on error and 1 if there
5120 * was nothing in the tree that matched the search criteria.
5122 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5123 struct btrfs_path *path,
5126 struct extent_buffer *cur;
5127 struct btrfs_key found_key;
5133 int keep_locks = path->keep_locks;
5135 path->keep_locks = 1;
5137 cur = btrfs_read_lock_root_node(root);
5138 level = btrfs_header_level(cur);
5139 WARN_ON(path->nodes[level]);
5140 path->nodes[level] = cur;
5141 path->locks[level] = BTRFS_READ_LOCK;
5143 if (btrfs_header_generation(cur) < min_trans) {
5148 nritems = btrfs_header_nritems(cur);
5149 level = btrfs_header_level(cur);
5150 sret = btrfs_bin_search(cur, min_key, level, &slot);
5156 /* at the lowest level, we're done, setup the path and exit */
5157 if (level == path->lowest_level) {
5158 if (slot >= nritems)
5161 path->slots[level] = slot;
5162 btrfs_item_key_to_cpu(cur, &found_key, slot);
5165 if (sret && slot > 0)
5168 * check this node pointer against the min_trans parameters.
5169 * If it is too old, old, skip to the next one.
5171 while (slot < nritems) {
5174 gen = btrfs_node_ptr_generation(cur, slot);
5175 if (gen < min_trans) {
5183 * we didn't find a candidate key in this node, walk forward
5184 * and find another one
5186 if (slot >= nritems) {
5187 path->slots[level] = slot;
5188 btrfs_set_path_blocking(path);
5189 sret = btrfs_find_next_key(root, path, min_key, level,
5192 btrfs_release_path(path);
5198 /* save our key for returning back */
5199 btrfs_node_key_to_cpu(cur, &found_key, slot);
5200 path->slots[level] = slot;
5201 if (level == path->lowest_level) {
5205 btrfs_set_path_blocking(path);
5206 cur = read_node_slot(cur, slot);
5212 btrfs_tree_read_lock(cur);
5214 path->locks[level - 1] = BTRFS_READ_LOCK;
5215 path->nodes[level - 1] = cur;
5216 unlock_up(path, level, 1, 0, NULL);
5219 path->keep_locks = keep_locks;
5221 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5222 btrfs_set_path_blocking(path);
5223 memcpy(min_key, &found_key, sizeof(found_key));
5228 static int tree_move_down(struct btrfs_fs_info *fs_info,
5229 struct btrfs_path *path,
5232 struct extent_buffer *eb;
5234 BUG_ON(*level == 0);
5235 eb = read_node_slot(path->nodes[*level], path->slots[*level]);
5239 path->nodes[*level - 1] = eb;
5240 path->slots[*level - 1] = 0;
5245 static int tree_move_next_or_upnext(struct btrfs_path *path,
5246 int *level, int root_level)
5250 nritems = btrfs_header_nritems(path->nodes[*level]);
5252 path->slots[*level]++;
5254 while (path->slots[*level] >= nritems) {
5255 if (*level == root_level)
5259 path->slots[*level] = 0;
5260 free_extent_buffer(path->nodes[*level]);
5261 path->nodes[*level] = NULL;
5263 path->slots[*level]++;
5265 nritems = btrfs_header_nritems(path->nodes[*level]);
5272 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5275 static int tree_advance(struct btrfs_fs_info *fs_info,
5276 struct btrfs_path *path,
5277 int *level, int root_level,
5279 struct btrfs_key *key)
5283 if (*level == 0 || !allow_down) {
5284 ret = tree_move_next_or_upnext(path, level, root_level);
5286 ret = tree_move_down(fs_info, path, level);
5290 btrfs_item_key_to_cpu(path->nodes[*level], key,
5291 path->slots[*level]);
5293 btrfs_node_key_to_cpu(path->nodes[*level], key,
5294 path->slots[*level]);
5299 static int tree_compare_item(struct btrfs_path *left_path,
5300 struct btrfs_path *right_path,
5305 unsigned long off1, off2;
5307 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5308 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5312 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5313 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5314 right_path->slots[0]);
5316 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5318 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5325 #define ADVANCE_ONLY_NEXT -1
5328 * This function compares two trees and calls the provided callback for
5329 * every changed/new/deleted item it finds.
5330 * If shared tree blocks are encountered, whole subtrees are skipped, making
5331 * the compare pretty fast on snapshotted subvolumes.
5333 * This currently works on commit roots only. As commit roots are read only,
5334 * we don't do any locking. The commit roots are protected with transactions.
5335 * Transactions are ended and rejoined when a commit is tried in between.
5337 * This function checks for modifications done to the trees while comparing.
5338 * If it detects a change, it aborts immediately.
5340 int btrfs_compare_trees(struct btrfs_root *left_root,
5341 struct btrfs_root *right_root,
5342 btrfs_changed_cb_t changed_cb, void *ctx)
5344 struct btrfs_fs_info *fs_info = left_root->fs_info;
5347 struct btrfs_path *left_path = NULL;
5348 struct btrfs_path *right_path = NULL;
5349 struct btrfs_key left_key;
5350 struct btrfs_key right_key;
5351 char *tmp_buf = NULL;
5352 int left_root_level;
5353 int right_root_level;
5356 int left_end_reached;
5357 int right_end_reached;
5365 left_path = btrfs_alloc_path();
5370 right_path = btrfs_alloc_path();
5376 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
5382 left_path->search_commit_root = 1;
5383 left_path->skip_locking = 1;
5384 right_path->search_commit_root = 1;
5385 right_path->skip_locking = 1;
5388 * Strategy: Go to the first items of both trees. Then do
5390 * If both trees are at level 0
5391 * Compare keys of current items
5392 * If left < right treat left item as new, advance left tree
5394 * If left > right treat right item as deleted, advance right tree
5396 * If left == right do deep compare of items, treat as changed if
5397 * needed, advance both trees and repeat
5398 * If both trees are at the same level but not at level 0
5399 * Compare keys of current nodes/leafs
5400 * If left < right advance left tree and repeat
5401 * If left > right advance right tree and repeat
5402 * If left == right compare blockptrs of the next nodes/leafs
5403 * If they match advance both trees but stay at the same level
5405 * If they don't match advance both trees while allowing to go
5407 * If tree levels are different
5408 * Advance the tree that needs it and repeat
5410 * Advancing a tree means:
5411 * If we are at level 0, try to go to the next slot. If that's not
5412 * possible, go one level up and repeat. Stop when we found a level
5413 * where we could go to the next slot. We may at this point be on a
5416 * If we are not at level 0 and not on shared tree blocks, go one
5419 * If we are not at level 0 and on shared tree blocks, go one slot to
5420 * the right if possible or go up and right.
5423 down_read(&fs_info->commit_root_sem);
5424 left_level = btrfs_header_level(left_root->commit_root);
5425 left_root_level = left_level;
5426 left_path->nodes[left_level] =
5427 btrfs_clone_extent_buffer(left_root->commit_root);
5428 if (!left_path->nodes[left_level]) {
5429 up_read(&fs_info->commit_root_sem);
5434 right_level = btrfs_header_level(right_root->commit_root);
5435 right_root_level = right_level;
5436 right_path->nodes[right_level] =
5437 btrfs_clone_extent_buffer(right_root->commit_root);
5438 if (!right_path->nodes[right_level]) {
5439 up_read(&fs_info->commit_root_sem);
5443 up_read(&fs_info->commit_root_sem);
5445 if (left_level == 0)
5446 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5447 &left_key, left_path->slots[left_level]);
5449 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5450 &left_key, left_path->slots[left_level]);
5451 if (right_level == 0)
5452 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5453 &right_key, right_path->slots[right_level]);
5455 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5456 &right_key, right_path->slots[right_level]);
5458 left_end_reached = right_end_reached = 0;
5459 advance_left = advance_right = 0;
5462 if (advance_left && !left_end_reached) {
5463 ret = tree_advance(fs_info, left_path, &left_level,
5465 advance_left != ADVANCE_ONLY_NEXT,
5468 left_end_reached = ADVANCE;
5473 if (advance_right && !right_end_reached) {
5474 ret = tree_advance(fs_info, right_path, &right_level,
5476 advance_right != ADVANCE_ONLY_NEXT,
5479 right_end_reached = ADVANCE;
5485 if (left_end_reached && right_end_reached) {
5488 } else if (left_end_reached) {
5489 if (right_level == 0) {
5490 ret = changed_cb(left_path, right_path,
5492 BTRFS_COMPARE_TREE_DELETED,
5497 advance_right = ADVANCE;
5499 } else if (right_end_reached) {
5500 if (left_level == 0) {
5501 ret = changed_cb(left_path, right_path,
5503 BTRFS_COMPARE_TREE_NEW,
5508 advance_left = ADVANCE;
5512 if (left_level == 0 && right_level == 0) {
5513 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5515 ret = changed_cb(left_path, right_path,
5517 BTRFS_COMPARE_TREE_NEW,
5521 advance_left = ADVANCE;
5522 } else if (cmp > 0) {
5523 ret = changed_cb(left_path, right_path,
5525 BTRFS_COMPARE_TREE_DELETED,
5529 advance_right = ADVANCE;
5531 enum btrfs_compare_tree_result result;
5533 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5534 ret = tree_compare_item(left_path, right_path,
5537 result = BTRFS_COMPARE_TREE_CHANGED;
5539 result = BTRFS_COMPARE_TREE_SAME;
5540 ret = changed_cb(left_path, right_path,
5541 &left_key, result, ctx);
5544 advance_left = ADVANCE;
5545 advance_right = ADVANCE;
5547 } else if (left_level == right_level) {
5548 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5550 advance_left = ADVANCE;
5551 } else if (cmp > 0) {
5552 advance_right = ADVANCE;
5554 left_blockptr = btrfs_node_blockptr(
5555 left_path->nodes[left_level],
5556 left_path->slots[left_level]);
5557 right_blockptr = btrfs_node_blockptr(
5558 right_path->nodes[right_level],
5559 right_path->slots[right_level]);
5560 left_gen = btrfs_node_ptr_generation(
5561 left_path->nodes[left_level],
5562 left_path->slots[left_level]);
5563 right_gen = btrfs_node_ptr_generation(
5564 right_path->nodes[right_level],
5565 right_path->slots[right_level]);
5566 if (left_blockptr == right_blockptr &&
5567 left_gen == right_gen) {
5569 * As we're on a shared block, don't
5570 * allow to go deeper.
5572 advance_left = ADVANCE_ONLY_NEXT;
5573 advance_right = ADVANCE_ONLY_NEXT;
5575 advance_left = ADVANCE;
5576 advance_right = ADVANCE;
5579 } else if (left_level < right_level) {
5580 advance_right = ADVANCE;
5582 advance_left = ADVANCE;
5587 btrfs_free_path(left_path);
5588 btrfs_free_path(right_path);
5594 * this is similar to btrfs_next_leaf, but does not try to preserve
5595 * and fixup the path. It looks for and returns the next key in the
5596 * tree based on the current path and the min_trans parameters.
5598 * 0 is returned if another key is found, < 0 if there are any errors
5599 * and 1 is returned if there are no higher keys in the tree
5601 * path->keep_locks should be set to 1 on the search made before
5602 * calling this function.
5604 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5605 struct btrfs_key *key, int level, u64 min_trans)
5608 struct extent_buffer *c;
5610 WARN_ON(!path->keep_locks);
5611 while (level < BTRFS_MAX_LEVEL) {
5612 if (!path->nodes[level])
5615 slot = path->slots[level] + 1;
5616 c = path->nodes[level];
5618 if (slot >= btrfs_header_nritems(c)) {
5621 struct btrfs_key cur_key;
5622 if (level + 1 >= BTRFS_MAX_LEVEL ||
5623 !path->nodes[level + 1])
5626 if (path->locks[level + 1]) {
5631 slot = btrfs_header_nritems(c) - 1;
5633 btrfs_item_key_to_cpu(c, &cur_key, slot);
5635 btrfs_node_key_to_cpu(c, &cur_key, slot);
5637 orig_lowest = path->lowest_level;
5638 btrfs_release_path(path);
5639 path->lowest_level = level;
5640 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5642 path->lowest_level = orig_lowest;
5646 c = path->nodes[level];
5647 slot = path->slots[level];
5654 btrfs_item_key_to_cpu(c, key, slot);
5656 u64 gen = btrfs_node_ptr_generation(c, slot);
5658 if (gen < min_trans) {
5662 btrfs_node_key_to_cpu(c, key, slot);
5670 * search the tree again to find a leaf with greater keys
5671 * returns 0 if it found something or 1 if there are no greater leaves.
5672 * returns < 0 on io errors.
5674 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5676 return btrfs_next_old_leaf(root, path, 0);
5679 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5684 struct extent_buffer *c;
5685 struct extent_buffer *next;
5686 struct btrfs_key key;
5689 int old_spinning = path->leave_spinning;
5690 int next_rw_lock = 0;
5692 nritems = btrfs_header_nritems(path->nodes[0]);
5696 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5701 btrfs_release_path(path);
5703 path->keep_locks = 1;
5704 path->leave_spinning = 1;
5707 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5710 path->keep_locks = 0;
5715 nritems = btrfs_header_nritems(path->nodes[0]);
5717 * by releasing the path above we dropped all our locks. A balance
5718 * could have added more items next to the key that used to be
5719 * at the very end of the block. So, check again here and
5720 * advance the path if there are now more items available.
5722 if (nritems > 0 && path->slots[0] < nritems - 1) {
5729 * So the above check misses one case:
5730 * - after releasing the path above, someone has removed the item that
5731 * used to be at the very end of the block, and balance between leafs
5732 * gets another one with bigger key.offset to replace it.
5734 * This one should be returned as well, or we can get leaf corruption
5735 * later(esp. in __btrfs_drop_extents()).
5737 * And a bit more explanation about this check,
5738 * with ret > 0, the key isn't found, the path points to the slot
5739 * where it should be inserted, so the path->slots[0] item must be the
5742 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5747 while (level < BTRFS_MAX_LEVEL) {
5748 if (!path->nodes[level]) {
5753 slot = path->slots[level] + 1;
5754 c = path->nodes[level];
5755 if (slot >= btrfs_header_nritems(c)) {
5757 if (level == BTRFS_MAX_LEVEL) {
5765 btrfs_tree_unlock_rw(next, next_rw_lock);
5766 free_extent_buffer(next);
5770 next_rw_lock = path->locks[level];
5771 ret = read_block_for_search(root, path, &next, level,
5777 btrfs_release_path(path);
5781 if (!path->skip_locking) {
5782 ret = btrfs_try_tree_read_lock(next);
5783 if (!ret && time_seq) {
5785 * If we don't get the lock, we may be racing
5786 * with push_leaf_left, holding that lock while
5787 * itself waiting for the leaf we've currently
5788 * locked. To solve this situation, we give up
5789 * on our lock and cycle.
5791 free_extent_buffer(next);
5792 btrfs_release_path(path);
5797 btrfs_set_path_blocking(path);
5798 btrfs_tree_read_lock(next);
5800 next_rw_lock = BTRFS_READ_LOCK;
5804 path->slots[level] = slot;
5807 c = path->nodes[level];
5808 if (path->locks[level])
5809 btrfs_tree_unlock_rw(c, path->locks[level]);
5811 free_extent_buffer(c);
5812 path->nodes[level] = next;
5813 path->slots[level] = 0;
5814 if (!path->skip_locking)
5815 path->locks[level] = next_rw_lock;
5819 ret = read_block_for_search(root, path, &next, level,
5825 btrfs_release_path(path);
5829 if (!path->skip_locking) {
5830 ret = btrfs_try_tree_read_lock(next);
5832 btrfs_set_path_blocking(path);
5833 btrfs_tree_read_lock(next);
5835 next_rw_lock = BTRFS_READ_LOCK;
5840 unlock_up(path, 0, 1, 0, NULL);
5841 path->leave_spinning = old_spinning;
5843 btrfs_set_path_blocking(path);
5849 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5850 * searching until it gets past min_objectid or finds an item of 'type'
5852 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5854 int btrfs_previous_item(struct btrfs_root *root,
5855 struct btrfs_path *path, u64 min_objectid,
5858 struct btrfs_key found_key;
5859 struct extent_buffer *leaf;
5864 if (path->slots[0] == 0) {
5865 btrfs_set_path_blocking(path);
5866 ret = btrfs_prev_leaf(root, path);
5872 leaf = path->nodes[0];
5873 nritems = btrfs_header_nritems(leaf);
5876 if (path->slots[0] == nritems)
5879 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5880 if (found_key.objectid < min_objectid)
5882 if (found_key.type == type)
5884 if (found_key.objectid == min_objectid &&
5885 found_key.type < type)
5892 * search in extent tree to find a previous Metadata/Data extent item with
5895 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5897 int btrfs_previous_extent_item(struct btrfs_root *root,
5898 struct btrfs_path *path, u64 min_objectid)
5900 struct btrfs_key found_key;
5901 struct extent_buffer *leaf;
5906 if (path->slots[0] == 0) {
5907 btrfs_set_path_blocking(path);
5908 ret = btrfs_prev_leaf(root, path);
5914 leaf = path->nodes[0];
5915 nritems = btrfs_header_nritems(leaf);
5918 if (path->slots[0] == nritems)
5921 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5922 if (found_key.objectid < min_objectid)
5924 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5925 found_key.type == BTRFS_METADATA_ITEM_KEY)
5927 if (found_key.objectid == min_objectid &&
5928 found_key.type < BTRFS_EXTENT_ITEM_KEY)