2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
22 #include "transaction.h"
23 #include "print-tree.h"
26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_key *ins_key,
30 struct btrfs_path *path, int data_size, int extend);
31 static int push_node_left(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root, struct extent_buffer *dst,
33 struct extent_buffer *src, int empty);
34 static int balance_node_right(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 struct extent_buffer *dst_buf,
37 struct extent_buffer *src_buf);
38 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 struct btrfs_path *path, int level, int slot);
41 struct btrfs_path *btrfs_alloc_path(void)
43 struct btrfs_path *path;
44 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
51 * set all locked nodes in the path to blocking locks. This should
52 * be done before scheduling
54 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
57 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
58 if (p->nodes[i] && p->locks[i])
59 btrfs_set_lock_blocking(p->nodes[i]);
64 * reset all the locked nodes in the patch to spinning locks.
66 * held is used to keep lockdep happy, when lockdep is enabled
67 * we set held to a blocking lock before we go around and
68 * retake all the spinlocks in the path. You can safely use NULL
71 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
72 struct extent_buffer *held)
76 #ifdef CONFIG_DEBUG_LOCK_ALLOC
77 /* lockdep really cares that we take all of these spinlocks
78 * in the right order. If any of the locks in the path are not
79 * currently blocking, it is going to complain. So, make really
80 * really sure by forcing the path to blocking before we clear
84 btrfs_set_lock_blocking(held);
85 btrfs_set_path_blocking(p);
88 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
89 if (p->nodes[i] && p->locks[i])
90 btrfs_clear_lock_blocking(p->nodes[i]);
93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
95 btrfs_clear_lock_blocking(held);
99 /* this also releases the path */
100 void btrfs_free_path(struct btrfs_path *p)
102 btrfs_release_path(NULL, p);
103 kmem_cache_free(btrfs_path_cachep, p);
107 * path release drops references on the extent buffers in the path
108 * and it drops any locks held by this path
110 * It is safe to call this on paths that no locks or extent buffers held.
112 noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
116 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
121 btrfs_tree_unlock(p->nodes[i]);
124 free_extent_buffer(p->nodes[i]);
130 * safely gets a reference on the root node of a tree. A lock
131 * is not taken, so a concurrent writer may put a different node
132 * at the root of the tree. See btrfs_lock_root_node for the
135 * The extent buffer returned by this has a reference taken, so
136 * it won't disappear. It may stop being the root of the tree
137 * at any time because there are no locks held.
139 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
141 struct extent_buffer *eb;
142 spin_lock(&root->node_lock);
144 extent_buffer_get(eb);
145 spin_unlock(&root->node_lock);
149 /* loop around taking references on and locking the root node of the
150 * tree until you end up with a lock on the root. A locked buffer
151 * is returned, with a reference held.
153 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
155 struct extent_buffer *eb;
158 eb = btrfs_root_node(root);
161 spin_lock(&root->node_lock);
162 if (eb == root->node) {
163 spin_unlock(&root->node_lock);
166 spin_unlock(&root->node_lock);
168 btrfs_tree_unlock(eb);
169 free_extent_buffer(eb);
174 /* cowonly root (everything not a reference counted cow subvolume), just get
175 * put onto a simple dirty list. transaction.c walks this to make sure they
176 * get properly updated on disk.
178 static void add_root_to_dirty_list(struct btrfs_root *root)
180 if (root->track_dirty && list_empty(&root->dirty_list)) {
181 list_add(&root->dirty_list,
182 &root->fs_info->dirty_cowonly_roots);
187 * used by snapshot creation to make a copy of a root for a tree with
188 * a given objectid. The buffer with the new root node is returned in
189 * cow_ret, and this func returns zero on success or a negative error code.
191 int btrfs_copy_root(struct btrfs_trans_handle *trans,
192 struct btrfs_root *root,
193 struct extent_buffer *buf,
194 struct extent_buffer **cow_ret, u64 new_root_objectid)
196 struct extent_buffer *cow;
200 struct btrfs_root *new_root;
202 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
206 memcpy(new_root, root, sizeof(*new_root));
207 new_root->root_key.objectid = new_root_objectid;
209 WARN_ON(root->ref_cows && trans->transid !=
210 root->fs_info->running_transaction->transid);
211 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
213 level = btrfs_header_level(buf);
214 nritems = btrfs_header_nritems(buf);
216 cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
217 new_root_objectid, trans->transid,
218 level, buf->start, 0);
224 copy_extent_buffer(cow, buf, 0, 0, cow->len);
225 btrfs_set_header_bytenr(cow, cow->start);
226 btrfs_set_header_generation(cow, trans->transid);
227 btrfs_set_header_owner(cow, new_root_objectid);
228 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
230 write_extent_buffer(cow, root->fs_info->fsid,
231 (unsigned long)btrfs_header_fsid(cow),
234 WARN_ON(btrfs_header_generation(buf) > trans->transid);
235 ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
241 btrfs_mark_buffer_dirty(cow);
247 * does the dirty work in cow of a single block. The parent block (if
248 * supplied) is updated to point to the new cow copy. The new buffer is marked
249 * dirty and returned locked. If you modify the block it needs to be marked
252 * search_start -- an allocation hint for the new block
254 * empty_size -- a hint that you plan on doing more cow. This is the size in
255 * bytes the allocator should try to find free next to the block it returns.
256 * This is just a hint and may be ignored by the allocator.
258 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
259 struct btrfs_root *root,
260 struct extent_buffer *buf,
261 struct extent_buffer *parent, int parent_slot,
262 struct extent_buffer **cow_ret,
263 u64 search_start, u64 empty_size)
266 struct extent_buffer *cow;
275 btrfs_assert_tree_locked(buf);
278 parent_start = parent->start;
282 WARN_ON(root->ref_cows && trans->transid !=
283 root->fs_info->running_transaction->transid);
284 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
286 level = btrfs_header_level(buf);
287 nritems = btrfs_header_nritems(buf);
289 cow = btrfs_alloc_free_block(trans, root, buf->len,
290 parent_start, root->root_key.objectid,
291 trans->transid, level,
292 search_start, empty_size);
296 /* cow is set to blocking by btrfs_init_new_buffer */
298 copy_extent_buffer(cow, buf, 0, 0, cow->len);
299 btrfs_set_header_bytenr(cow, cow->start);
300 btrfs_set_header_generation(cow, trans->transid);
301 btrfs_set_header_owner(cow, root->root_key.objectid);
302 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
304 write_extent_buffer(cow, root->fs_info->fsid,
305 (unsigned long)btrfs_header_fsid(cow),
308 WARN_ON(btrfs_header_generation(buf) > trans->transid);
309 if (btrfs_header_generation(buf) != trans->transid) {
311 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
315 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
317 } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
319 * There are only two places that can drop reference to
320 * tree blocks owned by living reloc trees, one is here,
321 * the other place is btrfs_drop_subtree. In both places,
322 * we check reference count while tree block is locked.
323 * Furthermore, if reference count is one, it won't get
324 * increased by someone else.
327 ret = btrfs_lookup_extent_ref(trans, root, buf->start,
331 ret = btrfs_update_ref(trans, root, buf, cow,
333 clean_tree_block(trans, root, buf);
335 ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
339 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
342 clean_tree_block(trans, root, buf);
345 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
346 ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
350 if (buf == root->node) {
351 WARN_ON(parent && parent != buf);
353 spin_lock(&root->node_lock);
355 extent_buffer_get(cow);
356 spin_unlock(&root->node_lock);
358 if (buf != root->commit_root) {
359 btrfs_free_extent(trans, root, buf->start,
360 buf->len, buf->start,
361 root->root_key.objectid,
362 btrfs_header_generation(buf),
365 free_extent_buffer(buf);
366 add_root_to_dirty_list(root);
368 btrfs_set_node_blockptr(parent, parent_slot,
370 WARN_ON(trans->transid == 0);
371 btrfs_set_node_ptr_generation(parent, parent_slot,
373 btrfs_mark_buffer_dirty(parent);
374 WARN_ON(btrfs_header_generation(parent) != trans->transid);
375 btrfs_free_extent(trans, root, buf->start, buf->len,
376 parent_start, btrfs_header_owner(parent),
377 btrfs_header_generation(parent), level, 1);
380 btrfs_tree_unlock(buf);
381 free_extent_buffer(buf);
382 btrfs_mark_buffer_dirty(cow);
388 * cows a single block, see __btrfs_cow_block for the real work.
389 * This version of it has extra checks so that a block isn't cow'd more than
390 * once per transaction, as long as it hasn't been written yet
392 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
393 struct btrfs_root *root, struct extent_buffer *buf,
394 struct extent_buffer *parent, int parent_slot,
395 struct extent_buffer **cow_ret)
400 if (trans->transaction != root->fs_info->running_transaction) {
401 printk(KERN_CRIT "trans %llu running %llu\n",
402 (unsigned long long)trans->transid,
404 root->fs_info->running_transaction->transid);
407 if (trans->transid != root->fs_info->generation) {
408 printk(KERN_CRIT "trans %llu running %llu\n",
409 (unsigned long long)trans->transid,
410 (unsigned long long)root->fs_info->generation);
414 if (btrfs_header_generation(buf) == trans->transid &&
415 btrfs_header_owner(buf) == root->root_key.objectid &&
416 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
421 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
424 btrfs_set_lock_blocking(parent);
425 btrfs_set_lock_blocking(buf);
427 ret = __btrfs_cow_block(trans, root, buf, parent,
428 parent_slot, cow_ret, search_start, 0);
433 * helper function for defrag to decide if two blocks pointed to by a
434 * node are actually close by
436 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
438 if (blocknr < other && other - (blocknr + blocksize) < 32768)
440 if (blocknr > other && blocknr - (other + blocksize) < 32768)
446 * compare two keys in a memcmp fashion
448 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
452 btrfs_disk_key_to_cpu(&k1, disk);
454 if (k1.objectid > k2->objectid)
456 if (k1.objectid < k2->objectid)
458 if (k1.type > k2->type)
460 if (k1.type < k2->type)
462 if (k1.offset > k2->offset)
464 if (k1.offset < k2->offset)
470 * same as comp_keys only with two btrfs_key's
472 static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
474 if (k1->objectid > k2->objectid)
476 if (k1->objectid < k2->objectid)
478 if (k1->type > k2->type)
480 if (k1->type < k2->type)
482 if (k1->offset > k2->offset)
484 if (k1->offset < k2->offset)
490 * this is used by the defrag code to go through all the
491 * leaves pointed to by a node and reallocate them so that
492 * disk order is close to key order
494 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
495 struct btrfs_root *root, struct extent_buffer *parent,
496 int start_slot, int cache_only, u64 *last_ret,
497 struct btrfs_key *progress)
499 struct extent_buffer *cur;
502 u64 search_start = *last_ret;
512 int progress_passed = 0;
513 struct btrfs_disk_key disk_key;
515 parent_level = btrfs_header_level(parent);
516 if (cache_only && parent_level != 1)
519 if (trans->transaction != root->fs_info->running_transaction)
521 if (trans->transid != root->fs_info->generation)
524 parent_nritems = btrfs_header_nritems(parent);
525 blocksize = btrfs_level_size(root, parent_level - 1);
526 end_slot = parent_nritems;
528 if (parent_nritems == 1)
531 btrfs_set_lock_blocking(parent);
533 for (i = start_slot; i < end_slot; i++) {
536 if (!parent->map_token) {
537 map_extent_buffer(parent,
538 btrfs_node_key_ptr_offset(i),
539 sizeof(struct btrfs_key_ptr),
540 &parent->map_token, &parent->kaddr,
541 &parent->map_start, &parent->map_len,
544 btrfs_node_key(parent, &disk_key, i);
545 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
549 blocknr = btrfs_node_blockptr(parent, i);
550 gen = btrfs_node_ptr_generation(parent, i);
552 last_block = blocknr;
555 other = btrfs_node_blockptr(parent, i - 1);
556 close = close_blocks(blocknr, other, blocksize);
558 if (!close && i < end_slot - 2) {
559 other = btrfs_node_blockptr(parent, i + 1);
560 close = close_blocks(blocknr, other, blocksize);
563 last_block = blocknr;
566 if (parent->map_token) {
567 unmap_extent_buffer(parent, parent->map_token,
569 parent->map_token = NULL;
572 cur = btrfs_find_tree_block(root, blocknr, blocksize);
574 uptodate = btrfs_buffer_uptodate(cur, gen);
577 if (!cur || !uptodate) {
579 free_extent_buffer(cur);
583 cur = read_tree_block(root, blocknr,
585 } else if (!uptodate) {
586 btrfs_read_buffer(cur, gen);
589 if (search_start == 0)
590 search_start = last_block;
592 btrfs_tree_lock(cur);
593 btrfs_set_lock_blocking(cur);
594 err = __btrfs_cow_block(trans, root, cur, parent, i,
597 (end_slot - i) * blocksize));
599 btrfs_tree_unlock(cur);
600 free_extent_buffer(cur);
603 search_start = cur->start;
604 last_block = cur->start;
605 *last_ret = search_start;
606 btrfs_tree_unlock(cur);
607 free_extent_buffer(cur);
609 if (parent->map_token) {
610 unmap_extent_buffer(parent, parent->map_token,
612 parent->map_token = NULL;
618 * The leaf data grows from end-to-front in the node.
619 * this returns the address of the start of the last item,
620 * which is the stop of the leaf data stack
622 static inline unsigned int leaf_data_end(struct btrfs_root *root,
623 struct extent_buffer *leaf)
625 u32 nr = btrfs_header_nritems(leaf);
627 return BTRFS_LEAF_DATA_SIZE(root);
628 return btrfs_item_offset_nr(leaf, nr - 1);
632 * extra debugging checks to make sure all the items in a key are
633 * well formed and in the proper order
635 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
638 struct extent_buffer *parent = NULL;
639 struct extent_buffer *node = path->nodes[level];
640 struct btrfs_disk_key parent_key;
641 struct btrfs_disk_key node_key;
644 struct btrfs_key cpukey;
645 u32 nritems = btrfs_header_nritems(node);
647 if (path->nodes[level + 1])
648 parent = path->nodes[level + 1];
650 slot = path->slots[level];
651 BUG_ON(nritems == 0);
653 parent_slot = path->slots[level + 1];
654 btrfs_node_key(parent, &parent_key, parent_slot);
655 btrfs_node_key(node, &node_key, 0);
656 BUG_ON(memcmp(&parent_key, &node_key,
657 sizeof(struct btrfs_disk_key)));
658 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
659 btrfs_header_bytenr(node));
661 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
663 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
664 btrfs_node_key(node, &node_key, slot);
665 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
667 if (slot < nritems - 1) {
668 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
669 btrfs_node_key(node, &node_key, slot);
670 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
676 * extra checking to make sure all the items in a leaf are
677 * well formed and in the proper order
679 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
682 struct extent_buffer *leaf = path->nodes[level];
683 struct extent_buffer *parent = NULL;
685 struct btrfs_key cpukey;
686 struct btrfs_disk_key parent_key;
687 struct btrfs_disk_key leaf_key;
688 int slot = path->slots[0];
690 u32 nritems = btrfs_header_nritems(leaf);
692 if (path->nodes[level + 1])
693 parent = path->nodes[level + 1];
699 parent_slot = path->slots[level + 1];
700 btrfs_node_key(parent, &parent_key, parent_slot);
701 btrfs_item_key(leaf, &leaf_key, 0);
703 BUG_ON(memcmp(&parent_key, &leaf_key,
704 sizeof(struct btrfs_disk_key)));
705 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
706 btrfs_header_bytenr(leaf));
708 if (slot != 0 && slot < nritems - 1) {
709 btrfs_item_key(leaf, &leaf_key, slot);
710 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
711 if (comp_keys(&leaf_key, &cpukey) <= 0) {
712 btrfs_print_leaf(root, leaf);
713 printk(KERN_CRIT "slot %d offset bad key\n", slot);
716 if (btrfs_item_offset_nr(leaf, slot - 1) !=
717 btrfs_item_end_nr(leaf, slot)) {
718 btrfs_print_leaf(root, leaf);
719 printk(KERN_CRIT "slot %d offset bad\n", slot);
723 if (slot < nritems - 1) {
724 btrfs_item_key(leaf, &leaf_key, slot);
725 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
726 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
727 if (btrfs_item_offset_nr(leaf, slot) !=
728 btrfs_item_end_nr(leaf, slot + 1)) {
729 btrfs_print_leaf(root, leaf);
730 printk(KERN_CRIT "slot %d offset bad\n", slot);
734 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
735 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
739 static noinline int check_block(struct btrfs_root *root,
740 struct btrfs_path *path, int level)
744 return check_leaf(root, path, level);
745 return check_node(root, path, level);
749 * search for key in the extent_buffer. The items start at offset p,
750 * and they are item_size apart. There are 'max' items in p.
752 * the slot in the array is returned via slot, and it points to
753 * the place where you would insert key if it is not found in
756 * slot may point to max if the key is bigger than all of the keys
758 static noinline int generic_bin_search(struct extent_buffer *eb,
760 int item_size, struct btrfs_key *key,
767 struct btrfs_disk_key *tmp = NULL;
768 struct btrfs_disk_key unaligned;
769 unsigned long offset;
770 char *map_token = NULL;
772 unsigned long map_start = 0;
773 unsigned long map_len = 0;
777 mid = (low + high) / 2;
778 offset = p + mid * item_size;
780 if (!map_token || offset < map_start ||
781 (offset + sizeof(struct btrfs_disk_key)) >
782 map_start + map_len) {
784 unmap_extent_buffer(eb, map_token, KM_USER0);
788 err = map_private_extent_buffer(eb, offset,
789 sizeof(struct btrfs_disk_key),
791 &map_start, &map_len, KM_USER0);
794 tmp = (struct btrfs_disk_key *)(kaddr + offset -
797 read_extent_buffer(eb, &unaligned,
798 offset, sizeof(unaligned));
803 tmp = (struct btrfs_disk_key *)(kaddr + offset -
806 ret = comp_keys(tmp, key);
815 unmap_extent_buffer(eb, map_token, KM_USER0);
821 unmap_extent_buffer(eb, map_token, KM_USER0);
826 * simple bin_search frontend that does the right thing for
829 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
830 int level, int *slot)
833 return generic_bin_search(eb,
834 offsetof(struct btrfs_leaf, items),
835 sizeof(struct btrfs_item),
836 key, btrfs_header_nritems(eb),
839 return generic_bin_search(eb,
840 offsetof(struct btrfs_node, ptrs),
841 sizeof(struct btrfs_key_ptr),
842 key, btrfs_header_nritems(eb),
848 /* given a node and slot number, this reads the blocks it points to. The
849 * extent buffer is returned with a reference taken (but unlocked).
850 * NULL is returned on error.
852 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
853 struct extent_buffer *parent, int slot)
855 int level = btrfs_header_level(parent);
858 if (slot >= btrfs_header_nritems(parent))
863 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
864 btrfs_level_size(root, level - 1),
865 btrfs_node_ptr_generation(parent, slot));
869 * node level balancing, used to make sure nodes are in proper order for
870 * item deletion. We balance from the top down, so we have to make sure
871 * that a deletion won't leave an node completely empty later on.
873 static noinline int balance_level(struct btrfs_trans_handle *trans,
874 struct btrfs_root *root,
875 struct btrfs_path *path, int level)
877 struct extent_buffer *right = NULL;
878 struct extent_buffer *mid;
879 struct extent_buffer *left = NULL;
880 struct extent_buffer *parent = NULL;
884 int orig_slot = path->slots[level];
885 int err_on_enospc = 0;
891 mid = path->nodes[level];
893 WARN_ON(!path->locks[level]);
894 WARN_ON(btrfs_header_generation(mid) != trans->transid);
896 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
898 if (level < BTRFS_MAX_LEVEL - 1)
899 parent = path->nodes[level + 1];
900 pslot = path->slots[level + 1];
903 * deal with the case where there is only one pointer in the root
904 * by promoting the node below to a root
907 struct extent_buffer *child;
909 if (btrfs_header_nritems(mid) != 1)
912 /* promote the child to a root */
913 child = read_node_slot(root, mid, 0);
915 btrfs_tree_lock(child);
916 btrfs_set_lock_blocking(child);
917 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
920 spin_lock(&root->node_lock);
922 spin_unlock(&root->node_lock);
924 ret = btrfs_update_extent_ref(trans, root, child->start,
926 mid->start, child->start,
927 root->root_key.objectid,
928 trans->transid, level - 1);
931 add_root_to_dirty_list(root);
932 btrfs_tree_unlock(child);
934 path->locks[level] = 0;
935 path->nodes[level] = NULL;
936 clean_tree_block(trans, root, mid);
937 btrfs_tree_unlock(mid);
938 /* once for the path */
939 free_extent_buffer(mid);
940 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
941 mid->start, root->root_key.objectid,
942 btrfs_header_generation(mid),
944 /* once for the root ptr */
945 free_extent_buffer(mid);
948 if (btrfs_header_nritems(mid) >
949 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
952 if (btrfs_header_nritems(mid) < 2)
955 left = read_node_slot(root, parent, pslot - 1);
957 btrfs_tree_lock(left);
958 btrfs_set_lock_blocking(left);
959 wret = btrfs_cow_block(trans, root, left,
960 parent, pslot - 1, &left);
966 right = read_node_slot(root, parent, pslot + 1);
968 btrfs_tree_lock(right);
969 btrfs_set_lock_blocking(right);
970 wret = btrfs_cow_block(trans, root, right,
971 parent, pslot + 1, &right);
978 /* first, try to make some room in the middle buffer */
980 orig_slot += btrfs_header_nritems(left);
981 wret = push_node_left(trans, root, left, mid, 1);
984 if (btrfs_header_nritems(mid) < 2)
989 * then try to empty the right most buffer into the middle
992 wret = push_node_left(trans, root, mid, right, 1);
993 if (wret < 0 && wret != -ENOSPC)
995 if (btrfs_header_nritems(right) == 0) {
996 u64 bytenr = right->start;
997 u64 generation = btrfs_header_generation(parent);
998 u32 blocksize = right->len;
1000 clean_tree_block(trans, root, right);
1001 btrfs_tree_unlock(right);
1002 free_extent_buffer(right);
1004 wret = del_ptr(trans, root, path, level + 1, pslot +
1008 wret = btrfs_free_extent(trans, root, bytenr,
1009 blocksize, parent->start,
1010 btrfs_header_owner(parent),
1011 generation, level, 1);
1015 struct btrfs_disk_key right_key;
1016 btrfs_node_key(right, &right_key, 0);
1017 btrfs_set_node_key(parent, &right_key, pslot + 1);
1018 btrfs_mark_buffer_dirty(parent);
1021 if (btrfs_header_nritems(mid) == 1) {
1023 * we're not allowed to leave a node with one item in the
1024 * tree during a delete. A deletion from lower in the tree
1025 * could try to delete the only pointer in this node.
1026 * So, pull some keys from the left.
1027 * There has to be a left pointer at this point because
1028 * otherwise we would have pulled some pointers from the
1032 wret = balance_node_right(trans, root, mid, left);
1038 wret = push_node_left(trans, root, left, mid, 1);
1044 if (btrfs_header_nritems(mid) == 0) {
1045 /* we've managed to empty the middle node, drop it */
1046 u64 root_gen = btrfs_header_generation(parent);
1047 u64 bytenr = mid->start;
1048 u32 blocksize = mid->len;
1050 clean_tree_block(trans, root, mid);
1051 btrfs_tree_unlock(mid);
1052 free_extent_buffer(mid);
1054 wret = del_ptr(trans, root, path, level + 1, pslot);
1057 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1059 btrfs_header_owner(parent),
1060 root_gen, level, 1);
1064 /* update the parent key to reflect our changes */
1065 struct btrfs_disk_key mid_key;
1066 btrfs_node_key(mid, &mid_key, 0);
1067 btrfs_set_node_key(parent, &mid_key, pslot);
1068 btrfs_mark_buffer_dirty(parent);
1071 /* update the path */
1073 if (btrfs_header_nritems(left) > orig_slot) {
1074 extent_buffer_get(left);
1075 /* left was locked after cow */
1076 path->nodes[level] = left;
1077 path->slots[level + 1] -= 1;
1078 path->slots[level] = orig_slot;
1080 btrfs_tree_unlock(mid);
1081 free_extent_buffer(mid);
1084 orig_slot -= btrfs_header_nritems(left);
1085 path->slots[level] = orig_slot;
1088 /* double check we haven't messed things up */
1089 check_block(root, path, level);
1091 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1095 btrfs_tree_unlock(right);
1096 free_extent_buffer(right);
1099 if (path->nodes[level] != left)
1100 btrfs_tree_unlock(left);
1101 free_extent_buffer(left);
1106 /* Node balancing for insertion. Here we only split or push nodes around
1107 * when they are completely full. This is also done top down, so we
1108 * have to be pessimistic.
1110 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1111 struct btrfs_root *root,
1112 struct btrfs_path *path, int level)
1114 struct extent_buffer *right = NULL;
1115 struct extent_buffer *mid;
1116 struct extent_buffer *left = NULL;
1117 struct extent_buffer *parent = NULL;
1121 int orig_slot = path->slots[level];
1127 mid = path->nodes[level];
1128 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1129 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1131 if (level < BTRFS_MAX_LEVEL - 1)
1132 parent = path->nodes[level + 1];
1133 pslot = path->slots[level + 1];
1138 left = read_node_slot(root, parent, pslot - 1);
1140 /* first, try to make some room in the middle buffer */
1144 btrfs_tree_lock(left);
1145 btrfs_set_lock_blocking(left);
1147 left_nr = btrfs_header_nritems(left);
1148 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1151 ret = btrfs_cow_block(trans, root, left, parent,
1156 wret = push_node_left(trans, root,
1163 struct btrfs_disk_key disk_key;
1164 orig_slot += left_nr;
1165 btrfs_node_key(mid, &disk_key, 0);
1166 btrfs_set_node_key(parent, &disk_key, pslot);
1167 btrfs_mark_buffer_dirty(parent);
1168 if (btrfs_header_nritems(left) > orig_slot) {
1169 path->nodes[level] = left;
1170 path->slots[level + 1] -= 1;
1171 path->slots[level] = orig_slot;
1172 btrfs_tree_unlock(mid);
1173 free_extent_buffer(mid);
1176 btrfs_header_nritems(left);
1177 path->slots[level] = orig_slot;
1178 btrfs_tree_unlock(left);
1179 free_extent_buffer(left);
1183 btrfs_tree_unlock(left);
1184 free_extent_buffer(left);
1186 right = read_node_slot(root, parent, pslot + 1);
1189 * then try to empty the right most buffer into the middle
1194 btrfs_tree_lock(right);
1195 btrfs_set_lock_blocking(right);
1197 right_nr = btrfs_header_nritems(right);
1198 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1201 ret = btrfs_cow_block(trans, root, right,
1207 wret = balance_node_right(trans, root,
1214 struct btrfs_disk_key disk_key;
1216 btrfs_node_key(right, &disk_key, 0);
1217 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1218 btrfs_mark_buffer_dirty(parent);
1220 if (btrfs_header_nritems(mid) <= orig_slot) {
1221 path->nodes[level] = right;
1222 path->slots[level + 1] += 1;
1223 path->slots[level] = orig_slot -
1224 btrfs_header_nritems(mid);
1225 btrfs_tree_unlock(mid);
1226 free_extent_buffer(mid);
1228 btrfs_tree_unlock(right);
1229 free_extent_buffer(right);
1233 btrfs_tree_unlock(right);
1234 free_extent_buffer(right);
1240 * readahead one full node of leaves, finding things that are close
1241 * to the block in 'slot', and triggering ra on them.
1243 static noinline void reada_for_search(struct btrfs_root *root,
1244 struct btrfs_path *path,
1245 int level, int slot, u64 objectid)
1247 struct extent_buffer *node;
1248 struct btrfs_disk_key disk_key;
1253 int direction = path->reada;
1254 struct extent_buffer *eb;
1262 if (!path->nodes[level])
1265 node = path->nodes[level];
1267 search = btrfs_node_blockptr(node, slot);
1268 blocksize = btrfs_level_size(root, level - 1);
1269 eb = btrfs_find_tree_block(root, search, blocksize);
1271 free_extent_buffer(eb);
1277 nritems = btrfs_header_nritems(node);
1280 if (direction < 0) {
1284 } else if (direction > 0) {
1289 if (path->reada < 0 && objectid) {
1290 btrfs_node_key(node, &disk_key, nr);
1291 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1294 search = btrfs_node_blockptr(node, nr);
1295 if ((search <= target && target - search <= 65536) ||
1296 (search > target && search - target <= 65536)) {
1297 readahead_tree_block(root, search, blocksize,
1298 btrfs_node_ptr_generation(node, nr));
1302 if ((nread > 65536 || nscan > 32))
1308 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1311 static noinline int reada_for_balance(struct btrfs_root *root,
1312 struct btrfs_path *path, int level)
1316 struct extent_buffer *parent;
1317 struct extent_buffer *eb;
1324 parent = path->nodes[level - 1];
1328 nritems = btrfs_header_nritems(parent);
1329 slot = path->slots[level];
1330 blocksize = btrfs_level_size(root, level);
1333 block1 = btrfs_node_blockptr(parent, slot - 1);
1334 gen = btrfs_node_ptr_generation(parent, slot - 1);
1335 eb = btrfs_find_tree_block(root, block1, blocksize);
1336 if (eb && btrfs_buffer_uptodate(eb, gen))
1338 free_extent_buffer(eb);
1340 if (slot < nritems) {
1341 block2 = btrfs_node_blockptr(parent, slot + 1);
1342 gen = btrfs_node_ptr_generation(parent, slot + 1);
1343 eb = btrfs_find_tree_block(root, block2, blocksize);
1344 if (eb && btrfs_buffer_uptodate(eb, gen))
1346 free_extent_buffer(eb);
1348 if (block1 || block2) {
1350 btrfs_release_path(root, path);
1352 readahead_tree_block(root, block1, blocksize, 0);
1354 readahead_tree_block(root, block2, blocksize, 0);
1357 eb = read_tree_block(root, block1, blocksize, 0);
1358 free_extent_buffer(eb);
1361 eb = read_tree_block(root, block2, blocksize, 0);
1362 free_extent_buffer(eb);
1370 * when we walk down the tree, it is usually safe to unlock the higher layers
1371 * in the tree. The exceptions are when our path goes through slot 0, because
1372 * operations on the tree might require changing key pointers higher up in the
1375 * callers might also have set path->keep_locks, which tells this code to keep
1376 * the lock if the path points to the last slot in the block. This is part of
1377 * walking through the tree, and selecting the next slot in the higher block.
1379 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1380 * if lowest_unlock is 1, level 0 won't be unlocked
1382 static noinline void unlock_up(struct btrfs_path *path, int level,
1386 int skip_level = level;
1388 struct extent_buffer *t;
1390 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1391 if (!path->nodes[i])
1393 if (!path->locks[i])
1395 if (!no_skips && path->slots[i] == 0) {
1399 if (!no_skips && path->keep_locks) {
1402 nritems = btrfs_header_nritems(t);
1403 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1408 if (skip_level < i && i >= lowest_unlock)
1412 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1413 btrfs_tree_unlock(t);
1420 * This releases any locks held in the path starting at level and
1421 * going all the way up to the root.
1423 * btrfs_search_slot will keep the lock held on higher nodes in a few
1424 * corner cases, such as COW of the block at slot zero in the node. This
1425 * ignores those rules, and it should only be called when there are no
1426 * more updates to be done higher up in the tree.
1428 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1432 if (path->keep_locks || path->lowest_level)
1435 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1436 if (!path->nodes[i])
1438 if (!path->locks[i])
1440 btrfs_tree_unlock(path->nodes[i]);
1446 * look for key in the tree. path is filled in with nodes along the way
1447 * if key is found, we return zero and you can find the item in the leaf
1448 * level of the path (level 0)
1450 * If the key isn't found, the path points to the slot where it should
1451 * be inserted, and 1 is returned. If there are other errors during the
1452 * search a negative error number is returned.
1454 * if ins_len > 0, nodes and leaves will be split as we walk down the
1455 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1458 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1459 *root, struct btrfs_key *key, struct btrfs_path *p, int
1462 struct extent_buffer *b;
1463 struct extent_buffer *tmp;
1467 int should_reada = p->reada;
1468 int lowest_unlock = 1;
1470 u8 lowest_level = 0;
1474 lowest_level = p->lowest_level;
1475 WARN_ON(lowest_level && ins_len > 0);
1476 WARN_ON(p->nodes[0] != NULL);
1482 if (p->skip_locking)
1483 b = btrfs_root_node(root);
1485 b = btrfs_lock_root_node(root);
1488 level = btrfs_header_level(b);
1491 * setup the path here so we can release it under lock
1492 * contention with the cow code
1494 p->nodes[level] = b;
1495 if (!p->skip_locking)
1496 p->locks[level] = 1;
1501 /* is a cow on this block not required */
1502 if (btrfs_header_generation(b) == trans->transid &&
1503 btrfs_header_owner(b) == root->root_key.objectid &&
1504 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1507 btrfs_set_path_blocking(p);
1509 wret = btrfs_cow_block(trans, root, b,
1510 p->nodes[level + 1],
1511 p->slots[level + 1], &b);
1513 free_extent_buffer(b);
1519 BUG_ON(!cow && ins_len);
1520 if (level != btrfs_header_level(b))
1522 level = btrfs_header_level(b);
1524 p->nodes[level] = b;
1525 if (!p->skip_locking)
1526 p->locks[level] = 1;
1528 btrfs_clear_path_blocking(p, NULL);
1531 * we have a lock on b and as long as we aren't changing
1532 * the tree, there is no way to for the items in b to change.
1533 * It is safe to drop the lock on our parent before we
1534 * go through the expensive btree search on b.
1536 * If cow is true, then we might be changing slot zero,
1537 * which may require changing the parent. So, we can't
1538 * drop the lock until after we know which slot we're
1542 btrfs_unlock_up_safe(p, level + 1);
1544 ret = check_block(root, p, level);
1550 ret = bin_search(b, key, level, &slot);
1553 if (ret && slot > 0)
1555 p->slots[level] = slot;
1556 if ((p->search_for_split || ins_len > 0) &&
1557 btrfs_header_nritems(b) >=
1558 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1561 sret = reada_for_balance(root, p, level);
1565 btrfs_set_path_blocking(p);
1566 sret = split_node(trans, root, p, level);
1567 btrfs_clear_path_blocking(p, NULL);
1574 b = p->nodes[level];
1575 slot = p->slots[level];
1576 } else if (ins_len < 0 &&
1577 btrfs_header_nritems(b) <
1578 BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
1581 sret = reada_for_balance(root, p, level);
1585 btrfs_set_path_blocking(p);
1586 sret = balance_level(trans, root, p, level);
1587 btrfs_clear_path_blocking(p, NULL);
1593 b = p->nodes[level];
1595 btrfs_release_path(NULL, p);
1598 slot = p->slots[level];
1599 BUG_ON(btrfs_header_nritems(b) == 1);
1601 unlock_up(p, level, lowest_unlock);
1603 /* this is only true while dropping a snapshot */
1604 if (level == lowest_level) {
1609 blocknr = btrfs_node_blockptr(b, slot);
1610 gen = btrfs_node_ptr_generation(b, slot);
1611 blocksize = btrfs_level_size(root, level - 1);
1613 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1614 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1618 * reduce lock contention at high levels
1619 * of the btree by dropping locks before
1623 btrfs_release_path(NULL, p);
1625 free_extent_buffer(tmp);
1627 reada_for_search(root, p,
1631 tmp = read_tree_block(root, blocknr,
1634 free_extent_buffer(tmp);
1637 btrfs_set_path_blocking(p);
1639 free_extent_buffer(tmp);
1641 reada_for_search(root, p,
1644 b = read_node_slot(root, b, slot);
1647 if (!p->skip_locking) {
1650 btrfs_clear_path_blocking(p, NULL);
1651 lret = btrfs_try_spin_lock(b);
1654 btrfs_set_path_blocking(p);
1656 btrfs_clear_path_blocking(p, b);
1660 p->slots[level] = slot;
1662 btrfs_leaf_free_space(root, b) < ins_len) {
1665 btrfs_set_path_blocking(p);
1666 sret = split_leaf(trans, root, key,
1667 p, ins_len, ret == 0);
1668 btrfs_clear_path_blocking(p, NULL);
1676 if (!p->search_for_split)
1677 unlock_up(p, level, lowest_unlock);
1684 * we don't really know what they plan on doing with the path
1685 * from here on, so for now just mark it as blocking
1687 if (!p->leave_spinning)
1688 btrfs_set_path_blocking(p);
1692 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_key *node_keys,
1695 u64 *nodes, int lowest_level)
1697 struct extent_buffer *eb;
1698 struct extent_buffer *parent;
1699 struct btrfs_key key;
1708 eb = btrfs_lock_root_node(root);
1709 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
1712 btrfs_set_lock_blocking(eb);
1716 level = btrfs_header_level(parent);
1717 if (level == 0 || level <= lowest_level)
1720 ret = bin_search(parent, &node_keys[lowest_level], level,
1722 if (ret && slot > 0)
1725 bytenr = btrfs_node_blockptr(parent, slot);
1726 if (nodes[level - 1] == bytenr)
1729 blocksize = btrfs_level_size(root, level - 1);
1730 generation = btrfs_node_ptr_generation(parent, slot);
1731 btrfs_node_key_to_cpu(eb, &key, slot);
1732 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1734 if (generation == trans->transid) {
1735 eb = read_tree_block(root, bytenr, blocksize,
1737 btrfs_tree_lock(eb);
1738 btrfs_set_lock_blocking(eb);
1742 * if node keys match and node pointer hasn't been modified
1743 * in the running transaction, we can merge the path. for
1744 * blocks owened by reloc trees, the node pointer check is
1745 * skipped, this is because these blocks are fully controlled
1746 * by the space balance code, no one else can modify them.
1748 if (!nodes[level - 1] || !key_match ||
1749 (generation == trans->transid &&
1750 btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1751 if (level == 1 || level == lowest_level + 1) {
1752 if (generation == trans->transid) {
1753 btrfs_tree_unlock(eb);
1754 free_extent_buffer(eb);
1759 if (generation != trans->transid) {
1760 eb = read_tree_block(root, bytenr, blocksize,
1762 btrfs_tree_lock(eb);
1763 btrfs_set_lock_blocking(eb);
1766 ret = btrfs_cow_block(trans, root, eb, parent, slot,
1770 if (root->root_key.objectid ==
1771 BTRFS_TREE_RELOC_OBJECTID) {
1772 if (!nodes[level - 1]) {
1773 nodes[level - 1] = eb->start;
1774 memcpy(&node_keys[level - 1], &key,
1775 sizeof(node_keys[0]));
1781 btrfs_tree_unlock(parent);
1782 free_extent_buffer(parent);
1787 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1788 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1789 btrfs_mark_buffer_dirty(parent);
1791 ret = btrfs_inc_extent_ref(trans, root,
1793 blocksize, parent->start,
1794 btrfs_header_owner(parent),
1795 btrfs_header_generation(parent),
1800 * If the block was created in the running transaction,
1801 * it's possible this is the last reference to it, so we
1802 * should drop the subtree.
1804 if (generation == trans->transid) {
1805 ret = btrfs_drop_subtree(trans, root, eb, parent);
1807 btrfs_tree_unlock(eb);
1808 free_extent_buffer(eb);
1810 ret = btrfs_free_extent(trans, root, bytenr,
1811 blocksize, parent->start,
1812 btrfs_header_owner(parent),
1813 btrfs_header_generation(parent),
1819 btrfs_tree_unlock(parent);
1820 free_extent_buffer(parent);
1825 * adjust the pointers going up the tree, starting at level
1826 * making sure the right key of each node is points to 'key'.
1827 * This is used after shifting pointers to the left, so it stops
1828 * fixing up pointers when a given leaf/node is not in slot 0 of the
1831 * If this fails to write a tree block, it returns -1, but continues
1832 * fixing up the blocks in ram so the tree is consistent.
1834 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1835 struct btrfs_root *root, struct btrfs_path *path,
1836 struct btrfs_disk_key *key, int level)
1840 struct extent_buffer *t;
1842 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1843 int tslot = path->slots[i];
1844 if (!path->nodes[i])
1847 btrfs_set_node_key(t, key, tslot);
1848 btrfs_mark_buffer_dirty(path->nodes[i]);
1858 * This function isn't completely safe. It's the caller's responsibility
1859 * that the new key won't break the order
1861 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1862 struct btrfs_root *root, struct btrfs_path *path,
1863 struct btrfs_key *new_key)
1865 struct btrfs_disk_key disk_key;
1866 struct extent_buffer *eb;
1869 eb = path->nodes[0];
1870 slot = path->slots[0];
1872 btrfs_item_key(eb, &disk_key, slot - 1);
1873 if (comp_keys(&disk_key, new_key) >= 0)
1876 if (slot < btrfs_header_nritems(eb) - 1) {
1877 btrfs_item_key(eb, &disk_key, slot + 1);
1878 if (comp_keys(&disk_key, new_key) <= 0)
1882 btrfs_cpu_key_to_disk(&disk_key, new_key);
1883 btrfs_set_item_key(eb, &disk_key, slot);
1884 btrfs_mark_buffer_dirty(eb);
1886 fixup_low_keys(trans, root, path, &disk_key, 1);
1891 * try to push data from one node into the next node left in the
1894 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1895 * error, and > 0 if there was no room in the left hand block.
1897 static int push_node_left(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root, struct extent_buffer *dst,
1899 struct extent_buffer *src, int empty)
1906 src_nritems = btrfs_header_nritems(src);
1907 dst_nritems = btrfs_header_nritems(dst);
1908 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1909 WARN_ON(btrfs_header_generation(src) != trans->transid);
1910 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1912 if (!empty && src_nritems <= 8)
1915 if (push_items <= 0)
1919 push_items = min(src_nritems, push_items);
1920 if (push_items < src_nritems) {
1921 /* leave at least 8 pointers in the node if
1922 * we aren't going to empty it
1924 if (src_nritems - push_items < 8) {
1925 if (push_items <= 8)
1931 push_items = min(src_nritems - 8, push_items);
1933 copy_extent_buffer(dst, src,
1934 btrfs_node_key_ptr_offset(dst_nritems),
1935 btrfs_node_key_ptr_offset(0),
1936 push_items * sizeof(struct btrfs_key_ptr));
1938 if (push_items < src_nritems) {
1939 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1940 btrfs_node_key_ptr_offset(push_items),
1941 (src_nritems - push_items) *
1942 sizeof(struct btrfs_key_ptr));
1944 btrfs_set_header_nritems(src, src_nritems - push_items);
1945 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1946 btrfs_mark_buffer_dirty(src);
1947 btrfs_mark_buffer_dirty(dst);
1949 ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1956 * try to push data from one node into the next node right in the
1959 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1960 * error, and > 0 if there was no room in the right hand block.
1962 * this will only push up to 1/2 the contents of the left node over
1964 static int balance_node_right(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root,
1966 struct extent_buffer *dst,
1967 struct extent_buffer *src)
1975 WARN_ON(btrfs_header_generation(src) != trans->transid);
1976 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1978 src_nritems = btrfs_header_nritems(src);
1979 dst_nritems = btrfs_header_nritems(dst);
1980 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1981 if (push_items <= 0)
1984 if (src_nritems < 4)
1987 max_push = src_nritems / 2 + 1;
1988 /* don't try to empty the node */
1989 if (max_push >= src_nritems)
1992 if (max_push < push_items)
1993 push_items = max_push;
1995 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1996 btrfs_node_key_ptr_offset(0),
1998 sizeof(struct btrfs_key_ptr));
2000 copy_extent_buffer(dst, src,
2001 btrfs_node_key_ptr_offset(0),
2002 btrfs_node_key_ptr_offset(src_nritems - push_items),
2003 push_items * sizeof(struct btrfs_key_ptr));
2005 btrfs_set_header_nritems(src, src_nritems - push_items);
2006 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2008 btrfs_mark_buffer_dirty(src);
2009 btrfs_mark_buffer_dirty(dst);
2011 ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
2018 * helper function to insert a new root level in the tree.
2019 * A new node is allocated, and a single item is inserted to
2020 * point to the existing root
2022 * returns zero on success or < 0 on failure.
2024 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2025 struct btrfs_root *root,
2026 struct btrfs_path *path, int level)
2029 struct extent_buffer *lower;
2030 struct extent_buffer *c;
2031 struct extent_buffer *old;
2032 struct btrfs_disk_key lower_key;
2035 BUG_ON(path->nodes[level]);
2036 BUG_ON(path->nodes[level-1] != root->node);
2038 lower = path->nodes[level-1];
2040 btrfs_item_key(lower, &lower_key, 0);
2042 btrfs_node_key(lower, &lower_key, 0);
2044 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2045 root->root_key.objectid, trans->transid,
2046 level, root->node->start, 0);
2050 memset_extent_buffer(c, 0, 0, root->nodesize);
2051 btrfs_set_header_nritems(c, 1);
2052 btrfs_set_header_level(c, level);
2053 btrfs_set_header_bytenr(c, c->start);
2054 btrfs_set_header_generation(c, trans->transid);
2055 btrfs_set_header_owner(c, root->root_key.objectid);
2057 write_extent_buffer(c, root->fs_info->fsid,
2058 (unsigned long)btrfs_header_fsid(c),
2061 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2062 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2065 btrfs_set_node_key(c, &lower_key, 0);
2066 btrfs_set_node_blockptr(c, 0, lower->start);
2067 lower_gen = btrfs_header_generation(lower);
2068 WARN_ON(lower_gen != trans->transid);
2070 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2072 btrfs_mark_buffer_dirty(c);
2074 spin_lock(&root->node_lock);
2077 spin_unlock(&root->node_lock);
2079 ret = btrfs_update_extent_ref(trans, root, lower->start,
2080 lower->len, lower->start, c->start,
2081 root->root_key.objectid,
2082 trans->transid, level - 1);
2085 /* the super has an extra ref to root->node */
2086 free_extent_buffer(old);
2088 add_root_to_dirty_list(root);
2089 extent_buffer_get(c);
2090 path->nodes[level] = c;
2091 path->locks[level] = 1;
2092 path->slots[level] = 0;
2097 * worker function to insert a single pointer in a node.
2098 * the node should have enough room for the pointer already
2100 * slot and level indicate where you want the key to go, and
2101 * blocknr is the block the key points to.
2103 * returns zero on success and < 0 on any error
2105 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2106 *root, struct btrfs_path *path, struct btrfs_disk_key
2107 *key, u64 bytenr, int slot, int level)
2109 struct extent_buffer *lower;
2112 BUG_ON(!path->nodes[level]);
2113 lower = path->nodes[level];
2114 nritems = btrfs_header_nritems(lower);
2117 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2119 if (slot != nritems) {
2120 memmove_extent_buffer(lower,
2121 btrfs_node_key_ptr_offset(slot + 1),
2122 btrfs_node_key_ptr_offset(slot),
2123 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2125 btrfs_set_node_key(lower, key, slot);
2126 btrfs_set_node_blockptr(lower, slot, bytenr);
2127 WARN_ON(trans->transid == 0);
2128 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2129 btrfs_set_header_nritems(lower, nritems + 1);
2130 btrfs_mark_buffer_dirty(lower);
2135 * split the node at the specified level in path in two.
2136 * The path is corrected to point to the appropriate node after the split
2138 * Before splitting this tries to make some room in the node by pushing
2139 * left and right, if either one works, it returns right away.
2141 * returns 0 on success and < 0 on failure
2143 static noinline int split_node(struct btrfs_trans_handle *trans,
2144 struct btrfs_root *root,
2145 struct btrfs_path *path, int level)
2147 struct extent_buffer *c;
2148 struct extent_buffer *split;
2149 struct btrfs_disk_key disk_key;
2155 c = path->nodes[level];
2156 WARN_ON(btrfs_header_generation(c) != trans->transid);
2157 if (c == root->node) {
2158 /* trying to split the root, lets make a new one */
2159 ret = insert_new_root(trans, root, path, level + 1);
2163 ret = push_nodes_for_insert(trans, root, path, level);
2164 c = path->nodes[level];
2165 if (!ret && btrfs_header_nritems(c) <
2166 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2172 c_nritems = btrfs_header_nritems(c);
2174 split = btrfs_alloc_free_block(trans, root, root->nodesize,
2175 path->nodes[level + 1]->start,
2176 root->root_key.objectid,
2177 trans->transid, level, c->start, 0);
2179 return PTR_ERR(split);
2181 btrfs_set_header_flags(split, btrfs_header_flags(c));
2182 btrfs_set_header_level(split, btrfs_header_level(c));
2183 btrfs_set_header_bytenr(split, split->start);
2184 btrfs_set_header_generation(split, trans->transid);
2185 btrfs_set_header_owner(split, root->root_key.objectid);
2186 btrfs_set_header_flags(split, 0);
2187 write_extent_buffer(split, root->fs_info->fsid,
2188 (unsigned long)btrfs_header_fsid(split),
2190 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2191 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2194 mid = (c_nritems + 1) / 2;
2196 copy_extent_buffer(split, c,
2197 btrfs_node_key_ptr_offset(0),
2198 btrfs_node_key_ptr_offset(mid),
2199 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2200 btrfs_set_header_nritems(split, c_nritems - mid);
2201 btrfs_set_header_nritems(c, mid);
2204 btrfs_mark_buffer_dirty(c);
2205 btrfs_mark_buffer_dirty(split);
2207 btrfs_node_key(split, &disk_key, 0);
2208 wret = insert_ptr(trans, root, path, &disk_key, split->start,
2209 path->slots[level + 1] + 1,
2214 ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2217 if (path->slots[level] >= mid) {
2218 path->slots[level] -= mid;
2219 btrfs_tree_unlock(c);
2220 free_extent_buffer(c);
2221 path->nodes[level] = split;
2222 path->slots[level + 1] += 1;
2224 btrfs_tree_unlock(split);
2225 free_extent_buffer(split);
2231 * how many bytes are required to store the items in a leaf. start
2232 * and nr indicate which items in the leaf to check. This totals up the
2233 * space used both by the item structs and the item data
2235 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2238 int nritems = btrfs_header_nritems(l);
2239 int end = min(nritems, start + nr) - 1;
2243 data_len = btrfs_item_end_nr(l, start);
2244 data_len = data_len - btrfs_item_offset_nr(l, end);
2245 data_len += sizeof(struct btrfs_item) * nr;
2246 WARN_ON(data_len < 0);
2251 * The space between the end of the leaf items and
2252 * the start of the leaf data. IOW, how much room
2253 * the leaf has left for both items and data
2255 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2256 struct extent_buffer *leaf)
2258 int nritems = btrfs_header_nritems(leaf);
2260 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2262 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2263 "used %d nritems %d\n",
2264 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2265 leaf_space_used(leaf, 0, nritems), nritems);
2270 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2271 struct btrfs_root *root,
2272 struct btrfs_path *path,
2273 int data_size, int empty,
2274 struct extent_buffer *right,
2275 int free_space, u32 left_nritems)
2277 struct extent_buffer *left = path->nodes[0];
2278 struct extent_buffer *upper = path->nodes[1];
2279 struct btrfs_disk_key disk_key;
2284 struct btrfs_item *item;
2296 if (path->slots[0] >= left_nritems)
2297 push_space += data_size;
2299 slot = path->slots[1];
2300 i = left_nritems - 1;
2302 item = btrfs_item_nr(left, i);
2304 if (!empty && push_items > 0) {
2305 if (path->slots[0] > i)
2307 if (path->slots[0] == i) {
2308 int space = btrfs_leaf_free_space(root, left);
2309 if (space + push_space * 2 > free_space)
2314 if (path->slots[0] == i)
2315 push_space += data_size;
2317 if (!left->map_token) {
2318 map_extent_buffer(left, (unsigned long)item,
2319 sizeof(struct btrfs_item),
2320 &left->map_token, &left->kaddr,
2321 &left->map_start, &left->map_len,
2325 this_item_size = btrfs_item_size(left, item);
2326 if (this_item_size + sizeof(*item) + push_space > free_space)
2330 push_space += this_item_size + sizeof(*item);
2335 if (left->map_token) {
2336 unmap_extent_buffer(left, left->map_token, KM_USER1);
2337 left->map_token = NULL;
2340 if (push_items == 0)
2343 if (!empty && push_items == left_nritems)
2346 /* push left to right */
2347 right_nritems = btrfs_header_nritems(right);
2349 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2350 push_space -= leaf_data_end(root, left);
2352 /* make room in the right data area */
2353 data_end = leaf_data_end(root, right);
2354 memmove_extent_buffer(right,
2355 btrfs_leaf_data(right) + data_end - push_space,
2356 btrfs_leaf_data(right) + data_end,
2357 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2359 /* copy from the left data area */
2360 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2361 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2362 btrfs_leaf_data(left) + leaf_data_end(root, left),
2365 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2366 btrfs_item_nr_offset(0),
2367 right_nritems * sizeof(struct btrfs_item));
2369 /* copy the items from left to right */
2370 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2371 btrfs_item_nr_offset(left_nritems - push_items),
2372 push_items * sizeof(struct btrfs_item));
2374 /* update the item pointers */
2375 right_nritems += push_items;
2376 btrfs_set_header_nritems(right, right_nritems);
2377 push_space = BTRFS_LEAF_DATA_SIZE(root);
2378 for (i = 0; i < right_nritems; i++) {
2379 item = btrfs_item_nr(right, i);
2380 if (!right->map_token) {
2381 map_extent_buffer(right, (unsigned long)item,
2382 sizeof(struct btrfs_item),
2383 &right->map_token, &right->kaddr,
2384 &right->map_start, &right->map_len,
2387 push_space -= btrfs_item_size(right, item);
2388 btrfs_set_item_offset(right, item, push_space);
2391 if (right->map_token) {
2392 unmap_extent_buffer(right, right->map_token, KM_USER1);
2393 right->map_token = NULL;
2395 left_nritems -= push_items;
2396 btrfs_set_header_nritems(left, left_nritems);
2399 btrfs_mark_buffer_dirty(left);
2400 btrfs_mark_buffer_dirty(right);
2402 ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2405 btrfs_item_key(right, &disk_key, 0);
2406 btrfs_set_node_key(upper, &disk_key, slot + 1);
2407 btrfs_mark_buffer_dirty(upper);
2409 /* then fixup the leaf pointer in the path */
2410 if (path->slots[0] >= left_nritems) {
2411 path->slots[0] -= left_nritems;
2412 if (btrfs_header_nritems(path->nodes[0]) == 0)
2413 clean_tree_block(trans, root, path->nodes[0]);
2414 btrfs_tree_unlock(path->nodes[0]);
2415 free_extent_buffer(path->nodes[0]);
2416 path->nodes[0] = right;
2417 path->slots[1] += 1;
2419 btrfs_tree_unlock(right);
2420 free_extent_buffer(right);
2425 btrfs_tree_unlock(right);
2426 free_extent_buffer(right);
2431 * push some data in the path leaf to the right, trying to free up at
2432 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2434 * returns 1 if the push failed because the other node didn't have enough
2435 * room, 0 if everything worked out and < 0 if there were major errors.
2437 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2438 *root, struct btrfs_path *path, int data_size,
2441 struct extent_buffer *left = path->nodes[0];
2442 struct extent_buffer *right;
2443 struct extent_buffer *upper;
2449 if (!path->nodes[1])
2452 slot = path->slots[1];
2453 upper = path->nodes[1];
2454 if (slot >= btrfs_header_nritems(upper) - 1)
2457 btrfs_assert_tree_locked(path->nodes[1]);
2459 right = read_node_slot(root, upper, slot + 1);
2460 btrfs_tree_lock(right);
2461 btrfs_set_lock_blocking(right);
2463 free_space = btrfs_leaf_free_space(root, right);
2464 if (free_space < data_size)
2467 /* cow and double check */
2468 ret = btrfs_cow_block(trans, root, right, upper,
2473 free_space = btrfs_leaf_free_space(root, right);
2474 if (free_space < data_size)
2477 left_nritems = btrfs_header_nritems(left);
2478 if (left_nritems == 0)
2481 return __push_leaf_right(trans, root, path, data_size, empty,
2482 right, free_space, left_nritems);
2484 btrfs_tree_unlock(right);
2485 free_extent_buffer(right);
2490 * push some data in the path leaf to the left, trying to free up at
2491 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2493 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2494 struct btrfs_root *root,
2495 struct btrfs_path *path, int data_size,
2496 int empty, struct extent_buffer *left,
2497 int free_space, int right_nritems)
2499 struct btrfs_disk_key disk_key;
2500 struct extent_buffer *right = path->nodes[0];
2505 struct btrfs_item *item;
2506 u32 old_left_nritems;
2511 u32 old_left_item_size;
2513 slot = path->slots[1];
2518 nr = right_nritems - 1;
2520 for (i = 0; i < nr; i++) {
2521 item = btrfs_item_nr(right, i);
2522 if (!right->map_token) {
2523 map_extent_buffer(right, (unsigned long)item,
2524 sizeof(struct btrfs_item),
2525 &right->map_token, &right->kaddr,
2526 &right->map_start, &right->map_len,
2530 if (!empty && push_items > 0) {
2531 if (path->slots[0] < i)
2533 if (path->slots[0] == i) {
2534 int space = btrfs_leaf_free_space(root, right);
2535 if (space + push_space * 2 > free_space)
2540 if (path->slots[0] == i)
2541 push_space += data_size;
2543 this_item_size = btrfs_item_size(right, item);
2544 if (this_item_size + sizeof(*item) + push_space > free_space)
2548 push_space += this_item_size + sizeof(*item);
2551 if (right->map_token) {
2552 unmap_extent_buffer(right, right->map_token, KM_USER1);
2553 right->map_token = NULL;
2556 if (push_items == 0) {
2560 if (!empty && push_items == btrfs_header_nritems(right))
2563 /* push data from right to left */
2564 copy_extent_buffer(left, right,
2565 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2566 btrfs_item_nr_offset(0),
2567 push_items * sizeof(struct btrfs_item));
2569 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2570 btrfs_item_offset_nr(right, push_items - 1);
2572 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2573 leaf_data_end(root, left) - push_space,
2574 btrfs_leaf_data(right) +
2575 btrfs_item_offset_nr(right, push_items - 1),
2577 old_left_nritems = btrfs_header_nritems(left);
2578 BUG_ON(old_left_nritems <= 0);
2580 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2581 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2584 item = btrfs_item_nr(left, i);
2585 if (!left->map_token) {
2586 map_extent_buffer(left, (unsigned long)item,
2587 sizeof(struct btrfs_item),
2588 &left->map_token, &left->kaddr,
2589 &left->map_start, &left->map_len,
2593 ioff = btrfs_item_offset(left, item);
2594 btrfs_set_item_offset(left, item,
2595 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2597 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2598 if (left->map_token) {
2599 unmap_extent_buffer(left, left->map_token, KM_USER1);
2600 left->map_token = NULL;
2603 /* fixup right node */
2604 if (push_items > right_nritems) {
2605 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2610 if (push_items < right_nritems) {
2611 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2612 leaf_data_end(root, right);
2613 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2614 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2615 btrfs_leaf_data(right) +
2616 leaf_data_end(root, right), push_space);
2618 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2619 btrfs_item_nr_offset(push_items),
2620 (btrfs_header_nritems(right) - push_items) *
2621 sizeof(struct btrfs_item));
2623 right_nritems -= push_items;
2624 btrfs_set_header_nritems(right, right_nritems);
2625 push_space = BTRFS_LEAF_DATA_SIZE(root);
2626 for (i = 0; i < right_nritems; i++) {
2627 item = btrfs_item_nr(right, i);
2629 if (!right->map_token) {
2630 map_extent_buffer(right, (unsigned long)item,
2631 sizeof(struct btrfs_item),
2632 &right->map_token, &right->kaddr,
2633 &right->map_start, &right->map_len,
2637 push_space = push_space - btrfs_item_size(right, item);
2638 btrfs_set_item_offset(right, item, push_space);
2640 if (right->map_token) {
2641 unmap_extent_buffer(right, right->map_token, KM_USER1);
2642 right->map_token = NULL;
2645 btrfs_mark_buffer_dirty(left);
2647 btrfs_mark_buffer_dirty(right);
2649 ret = btrfs_update_ref(trans, root, right, left,
2650 old_left_nritems, push_items);
2653 btrfs_item_key(right, &disk_key, 0);
2654 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2658 /* then fixup the leaf pointer in the path */
2659 if (path->slots[0] < push_items) {
2660 path->slots[0] += old_left_nritems;
2661 if (btrfs_header_nritems(path->nodes[0]) == 0)
2662 clean_tree_block(trans, root, path->nodes[0]);
2663 btrfs_tree_unlock(path->nodes[0]);
2664 free_extent_buffer(path->nodes[0]);
2665 path->nodes[0] = left;
2666 path->slots[1] -= 1;
2668 btrfs_tree_unlock(left);
2669 free_extent_buffer(left);
2670 path->slots[0] -= push_items;
2672 BUG_ON(path->slots[0] < 0);
2675 btrfs_tree_unlock(left);
2676 free_extent_buffer(left);
2681 * push some data in the path leaf to the left, trying to free up at
2682 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2684 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2685 *root, struct btrfs_path *path, int data_size,
2688 struct extent_buffer *right = path->nodes[0];
2689 struct extent_buffer *left;
2695 slot = path->slots[1];
2698 if (!path->nodes[1])
2701 right_nritems = btrfs_header_nritems(right);
2702 if (right_nritems == 0)
2705 btrfs_assert_tree_locked(path->nodes[1]);
2707 left = read_node_slot(root, path->nodes[1], slot - 1);
2708 btrfs_tree_lock(left);
2709 btrfs_set_lock_blocking(left);
2711 free_space = btrfs_leaf_free_space(root, left);
2712 if (free_space < data_size) {
2717 /* cow and double check */
2718 ret = btrfs_cow_block(trans, root, left,
2719 path->nodes[1], slot - 1, &left);
2721 /* we hit -ENOSPC, but it isn't fatal here */
2726 free_space = btrfs_leaf_free_space(root, left);
2727 if (free_space < data_size) {
2732 return __push_leaf_left(trans, root, path, data_size,
2733 empty, left, free_space, right_nritems);
2735 btrfs_tree_unlock(left);
2736 free_extent_buffer(left);
2741 * split the path's leaf in two, making sure there is at least data_size
2742 * available for the resulting leaf level of the path.
2744 * returns 0 if all went well and < 0 on failure.
2746 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2747 struct btrfs_root *root,
2748 struct btrfs_path *path,
2749 struct extent_buffer *l,
2750 struct extent_buffer *right,
2751 int slot, int mid, int nritems)
2758 struct btrfs_disk_key disk_key;
2760 nritems = nritems - mid;
2761 btrfs_set_header_nritems(right, nritems);
2762 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2764 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2765 btrfs_item_nr_offset(mid),
2766 nritems * sizeof(struct btrfs_item));
2768 copy_extent_buffer(right, l,
2769 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2770 data_copy_size, btrfs_leaf_data(l) +
2771 leaf_data_end(root, l), data_copy_size);
2773 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2774 btrfs_item_end_nr(l, mid);
2776 for (i = 0; i < nritems; i++) {
2777 struct btrfs_item *item = btrfs_item_nr(right, i);
2780 if (!right->map_token) {
2781 map_extent_buffer(right, (unsigned long)item,
2782 sizeof(struct btrfs_item),
2783 &right->map_token, &right->kaddr,
2784 &right->map_start, &right->map_len,
2788 ioff = btrfs_item_offset(right, item);
2789 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2792 if (right->map_token) {
2793 unmap_extent_buffer(right, right->map_token, KM_USER1);
2794 right->map_token = NULL;
2797 btrfs_set_header_nritems(l, mid);
2799 btrfs_item_key(right, &disk_key, 0);
2800 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2801 path->slots[1] + 1, 1);
2805 btrfs_mark_buffer_dirty(right);
2806 btrfs_mark_buffer_dirty(l);
2807 BUG_ON(path->slots[0] != slot);
2809 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2813 btrfs_tree_unlock(path->nodes[0]);
2814 free_extent_buffer(path->nodes[0]);
2815 path->nodes[0] = right;
2816 path->slots[0] -= mid;
2817 path->slots[1] += 1;
2819 btrfs_tree_unlock(right);
2820 free_extent_buffer(right);
2823 BUG_ON(path->slots[0] < 0);
2829 * split the path's leaf in two, making sure there is at least data_size
2830 * available for the resulting leaf level of the path.
2832 * returns 0 if all went well and < 0 on failure.
2834 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2835 struct btrfs_root *root,
2836 struct btrfs_key *ins_key,
2837 struct btrfs_path *path, int data_size,
2840 struct extent_buffer *l;
2844 struct extent_buffer *right;
2848 int num_doubles = 0;
2850 /* first try to make some room by pushing left and right */
2851 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2852 wret = push_leaf_right(trans, root, path, data_size, 0);
2856 wret = push_leaf_left(trans, root, path, data_size, 0);
2862 /* did the pushes work? */
2863 if (btrfs_leaf_free_space(root, l) >= data_size)
2867 if (!path->nodes[1]) {
2868 ret = insert_new_root(trans, root, path, 1);
2875 slot = path->slots[0];
2876 nritems = btrfs_header_nritems(l);
2877 mid = (nritems + 1) / 2;
2879 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2880 path->nodes[1]->start,
2881 root->root_key.objectid,
2882 trans->transid, 0, l->start, 0);
2883 if (IS_ERR(right)) {
2885 return PTR_ERR(right);
2888 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2889 btrfs_set_header_bytenr(right, right->start);
2890 btrfs_set_header_generation(right, trans->transid);
2891 btrfs_set_header_owner(right, root->root_key.objectid);
2892 btrfs_set_header_level(right, 0);
2893 write_extent_buffer(right, root->fs_info->fsid,
2894 (unsigned long)btrfs_header_fsid(right),
2897 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2898 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2903 leaf_space_used(l, mid, nritems - mid) + data_size >
2904 BTRFS_LEAF_DATA_SIZE(root)) {
2905 if (slot >= nritems) {
2906 struct btrfs_disk_key disk_key;
2908 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2909 btrfs_set_header_nritems(right, 0);
2910 wret = insert_ptr(trans, root, path,
2911 &disk_key, right->start,
2912 path->slots[1] + 1, 1);
2916 btrfs_tree_unlock(path->nodes[0]);
2917 free_extent_buffer(path->nodes[0]);
2918 path->nodes[0] = right;
2920 path->slots[1] += 1;
2921 btrfs_mark_buffer_dirty(right);
2925 if (mid != nritems &&
2926 leaf_space_used(l, mid, nritems - mid) +
2927 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2932 if (leaf_space_used(l, 0, mid) + data_size >
2933 BTRFS_LEAF_DATA_SIZE(root)) {
2934 if (!extend && data_size && slot == 0) {
2935 struct btrfs_disk_key disk_key;
2937 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2938 btrfs_set_header_nritems(right, 0);
2939 wret = insert_ptr(trans, root, path,
2945 btrfs_tree_unlock(path->nodes[0]);
2946 free_extent_buffer(path->nodes[0]);
2947 path->nodes[0] = right;
2949 if (path->slots[1] == 0) {
2950 wret = fixup_low_keys(trans, root,
2951 path, &disk_key, 1);
2955 btrfs_mark_buffer_dirty(right);
2957 } else if ((extend || !data_size) && slot == 0) {
2961 if (mid != nritems &&
2962 leaf_space_used(l, mid, nritems - mid) +
2963 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2970 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2974 BUG_ON(num_doubles != 0);
2983 * This function splits a single item into two items,
2984 * giving 'new_key' to the new item and splitting the
2985 * old one at split_offset (from the start of the item).
2987 * The path may be released by this operation. After
2988 * the split, the path is pointing to the old item. The
2989 * new item is going to be in the same node as the old one.
2991 * Note, the item being split must be smaller enough to live alone on
2992 * a tree block with room for one extra struct btrfs_item
2994 * This allows us to split the item in place, keeping a lock on the
2995 * leaf the entire time.
2997 int btrfs_split_item(struct btrfs_trans_handle *trans,
2998 struct btrfs_root *root,
2999 struct btrfs_path *path,
3000 struct btrfs_key *new_key,
3001 unsigned long split_offset)
3004 struct extent_buffer *leaf;
3005 struct btrfs_key orig_key;
3006 struct btrfs_item *item;
3007 struct btrfs_item *new_item;
3012 struct btrfs_disk_key disk_key;
3015 leaf = path->nodes[0];
3016 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
3017 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
3020 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3021 btrfs_release_path(root, path);
3023 path->search_for_split = 1;
3024 path->keep_locks = 1;
3026 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
3027 path->search_for_split = 0;
3029 /* if our item isn't there or got smaller, return now */
3030 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
3032 path->keep_locks = 0;
3036 btrfs_set_path_blocking(path);
3037 ret = split_leaf(trans, root, &orig_key, path,
3038 sizeof(struct btrfs_item), 1);
3039 path->keep_locks = 0;
3042 btrfs_unlock_up_safe(path, 1);
3043 leaf = path->nodes[0];
3044 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3048 * make sure any changes to the path from split_leaf leave it
3049 * in a blocking state
3051 btrfs_set_path_blocking(path);
3053 item = btrfs_item_nr(leaf, path->slots[0]);
3054 orig_offset = btrfs_item_offset(leaf, item);
3055 item_size = btrfs_item_size(leaf, item);
3057 buf = kmalloc(item_size, GFP_NOFS);
3058 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3059 path->slots[0]), item_size);
3060 slot = path->slots[0] + 1;
3061 leaf = path->nodes[0];
3063 nritems = btrfs_header_nritems(leaf);
3065 if (slot != nritems) {
3066 /* shift the items */
3067 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3068 btrfs_item_nr_offset(slot),
3069 (nritems - slot) * sizeof(struct btrfs_item));
3073 btrfs_cpu_key_to_disk(&disk_key, new_key);
3074 btrfs_set_item_key(leaf, &disk_key, slot);
3076 new_item = btrfs_item_nr(leaf, slot);
3078 btrfs_set_item_offset(leaf, new_item, orig_offset);
3079 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3081 btrfs_set_item_offset(leaf, item,
3082 orig_offset + item_size - split_offset);
3083 btrfs_set_item_size(leaf, item, split_offset);
3085 btrfs_set_header_nritems(leaf, nritems + 1);
3087 /* write the data for the start of the original item */
3088 write_extent_buffer(leaf, buf,
3089 btrfs_item_ptr_offset(leaf, path->slots[0]),
3092 /* write the data for the new item */
3093 write_extent_buffer(leaf, buf + split_offset,
3094 btrfs_item_ptr_offset(leaf, slot),
3095 item_size - split_offset);
3096 btrfs_mark_buffer_dirty(leaf);
3099 if (btrfs_leaf_free_space(root, leaf) < 0) {
3100 btrfs_print_leaf(root, leaf);
3108 * make the item pointed to by the path smaller. new_size indicates
3109 * how small to make it, and from_end tells us if we just chop bytes
3110 * off the end of the item or if we shift the item to chop bytes off
3113 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3114 struct btrfs_root *root,
3115 struct btrfs_path *path,
3116 u32 new_size, int from_end)
3121 struct extent_buffer *leaf;
3122 struct btrfs_item *item;
3124 unsigned int data_end;
3125 unsigned int old_data_start;
3126 unsigned int old_size;
3127 unsigned int size_diff;
3130 slot_orig = path->slots[0];
3131 leaf = path->nodes[0];
3132 slot = path->slots[0];
3134 old_size = btrfs_item_size_nr(leaf, slot);
3135 if (old_size == new_size)
3138 nritems = btrfs_header_nritems(leaf);
3139 data_end = leaf_data_end(root, leaf);
3141 old_data_start = btrfs_item_offset_nr(leaf, slot);
3143 size_diff = old_size - new_size;
3146 BUG_ON(slot >= nritems);
3149 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3151 /* first correct the data pointers */
3152 for (i = slot; i < nritems; i++) {
3154 item = btrfs_item_nr(leaf, i);
3156 if (!leaf->map_token) {
3157 map_extent_buffer(leaf, (unsigned long)item,
3158 sizeof(struct btrfs_item),
3159 &leaf->map_token, &leaf->kaddr,
3160 &leaf->map_start, &leaf->map_len,
3164 ioff = btrfs_item_offset(leaf, item);
3165 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3168 if (leaf->map_token) {
3169 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3170 leaf->map_token = NULL;
3173 /* shift the data */
3175 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3176 data_end + size_diff, btrfs_leaf_data(leaf) +
3177 data_end, old_data_start + new_size - data_end);
3179 struct btrfs_disk_key disk_key;
3182 btrfs_item_key(leaf, &disk_key, slot);
3184 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3186 struct btrfs_file_extent_item *fi;
3188 fi = btrfs_item_ptr(leaf, slot,
3189 struct btrfs_file_extent_item);
3190 fi = (struct btrfs_file_extent_item *)(
3191 (unsigned long)fi - size_diff);
3193 if (btrfs_file_extent_type(leaf, fi) ==
3194 BTRFS_FILE_EXTENT_INLINE) {
3195 ptr = btrfs_item_ptr_offset(leaf, slot);
3196 memmove_extent_buffer(leaf, ptr,
3198 offsetof(struct btrfs_file_extent_item,
3203 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3204 data_end + size_diff, btrfs_leaf_data(leaf) +
3205 data_end, old_data_start - data_end);
3207 offset = btrfs_disk_key_offset(&disk_key);
3208 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3209 btrfs_set_item_key(leaf, &disk_key, slot);
3211 fixup_low_keys(trans, root, path, &disk_key, 1);
3214 item = btrfs_item_nr(leaf, slot);
3215 btrfs_set_item_size(leaf, item, new_size);
3216 btrfs_mark_buffer_dirty(leaf);
3219 if (btrfs_leaf_free_space(root, leaf) < 0) {
3220 btrfs_print_leaf(root, leaf);
3227 * make the item pointed to by the path bigger, data_size is the new size.
3229 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3230 struct btrfs_root *root, struct btrfs_path *path,
3236 struct extent_buffer *leaf;
3237 struct btrfs_item *item;
3239 unsigned int data_end;
3240 unsigned int old_data;
3241 unsigned int old_size;
3244 slot_orig = path->slots[0];
3245 leaf = path->nodes[0];
3247 nritems = btrfs_header_nritems(leaf);
3248 data_end = leaf_data_end(root, leaf);
3250 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3251 btrfs_print_leaf(root, leaf);
3254 slot = path->slots[0];
3255 old_data = btrfs_item_end_nr(leaf, slot);
3258 if (slot >= nritems) {
3259 btrfs_print_leaf(root, leaf);
3260 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3266 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3268 /* first correct the data pointers */
3269 for (i = slot; i < nritems; i++) {
3271 item = btrfs_item_nr(leaf, i);
3273 if (!leaf->map_token) {
3274 map_extent_buffer(leaf, (unsigned long)item,
3275 sizeof(struct btrfs_item),
3276 &leaf->map_token, &leaf->kaddr,
3277 &leaf->map_start, &leaf->map_len,
3280 ioff = btrfs_item_offset(leaf, item);
3281 btrfs_set_item_offset(leaf, item, ioff - data_size);
3284 if (leaf->map_token) {
3285 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3286 leaf->map_token = NULL;
3289 /* shift the data */
3290 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3291 data_end - data_size, btrfs_leaf_data(leaf) +
3292 data_end, old_data - data_end);
3294 data_end = old_data;
3295 old_size = btrfs_item_size_nr(leaf, slot);
3296 item = btrfs_item_nr(leaf, slot);
3297 btrfs_set_item_size(leaf, item, old_size + data_size);
3298 btrfs_mark_buffer_dirty(leaf);
3301 if (btrfs_leaf_free_space(root, leaf) < 0) {
3302 btrfs_print_leaf(root, leaf);
3309 * Given a key and some data, insert items into the tree.
3310 * This does all the path init required, making room in the tree if needed.
3311 * Returns the number of keys that were inserted.
3313 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3314 struct btrfs_root *root,
3315 struct btrfs_path *path,
3316 struct btrfs_key *cpu_key, u32 *data_size,
3319 struct extent_buffer *leaf;
3320 struct btrfs_item *item;
3327 unsigned int data_end;
3328 struct btrfs_disk_key disk_key;
3329 struct btrfs_key found_key;
3331 for (i = 0; i < nr; i++) {
3332 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3333 BTRFS_LEAF_DATA_SIZE(root)) {
3337 total_data += data_size[i];
3338 total_size += data_size[i] + sizeof(struct btrfs_item);
3342 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3348 leaf = path->nodes[0];
3350 nritems = btrfs_header_nritems(leaf);
3351 data_end = leaf_data_end(root, leaf);
3353 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3354 for (i = nr; i >= 0; i--) {
3355 total_data -= data_size[i];
3356 total_size -= data_size[i] + sizeof(struct btrfs_item);
3357 if (total_size < btrfs_leaf_free_space(root, leaf))
3363 slot = path->slots[0];
3366 if (slot != nritems) {
3367 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3369 item = btrfs_item_nr(leaf, slot);
3370 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3372 /* figure out how many keys we can insert in here */
3373 total_data = data_size[0];
3374 for (i = 1; i < nr; i++) {
3375 if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3377 total_data += data_size[i];
3381 if (old_data < data_end) {
3382 btrfs_print_leaf(root, leaf);
3383 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3384 slot, old_data, data_end);
3388 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3390 /* first correct the data pointers */
3391 WARN_ON(leaf->map_token);
3392 for (i = slot; i < nritems; i++) {
3395 item = btrfs_item_nr(leaf, i);
3396 if (!leaf->map_token) {
3397 map_extent_buffer(leaf, (unsigned long)item,
3398 sizeof(struct btrfs_item),
3399 &leaf->map_token, &leaf->kaddr,
3400 &leaf->map_start, &leaf->map_len,
3404 ioff = btrfs_item_offset(leaf, item);
3405 btrfs_set_item_offset(leaf, item, ioff - total_data);
3407 if (leaf->map_token) {
3408 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3409 leaf->map_token = NULL;
3412 /* shift the items */
3413 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3414 btrfs_item_nr_offset(slot),
3415 (nritems - slot) * sizeof(struct btrfs_item));
3417 /* shift the data */
3418 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3419 data_end - total_data, btrfs_leaf_data(leaf) +
3420 data_end, old_data - data_end);
3421 data_end = old_data;
3424 * this sucks but it has to be done, if we are inserting at
3425 * the end of the leaf only insert 1 of the items, since we
3426 * have no way of knowing whats on the next leaf and we'd have
3427 * to drop our current locks to figure it out
3432 /* setup the item for the new data */
3433 for (i = 0; i < nr; i++) {
3434 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3435 btrfs_set_item_key(leaf, &disk_key, slot + i);
3436 item = btrfs_item_nr(leaf, slot + i);
3437 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3438 data_end -= data_size[i];
3439 btrfs_set_item_size(leaf, item, data_size[i]);
3441 btrfs_set_header_nritems(leaf, nritems + nr);
3442 btrfs_mark_buffer_dirty(leaf);
3446 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3447 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3450 if (btrfs_leaf_free_space(root, leaf) < 0) {
3451 btrfs_print_leaf(root, leaf);
3461 * this is a helper for btrfs_insert_empty_items, the main goal here is
3462 * to save stack depth by doing the bulk of the work in a function
3463 * that doesn't call btrfs_search_slot
3465 static noinline_for_stack int
3466 setup_items_for_insert(struct btrfs_trans_handle *trans,
3467 struct btrfs_root *root, struct btrfs_path *path,
3468 struct btrfs_key *cpu_key, u32 *data_size,
3469 u32 total_data, u32 total_size, int nr)
3471 struct btrfs_item *item;
3474 unsigned int data_end;
3475 struct btrfs_disk_key disk_key;
3477 struct extent_buffer *leaf;
3480 leaf = path->nodes[0];
3481 slot = path->slots[0];
3483 nritems = btrfs_header_nritems(leaf);
3484 data_end = leaf_data_end(root, leaf);
3486 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3487 btrfs_print_leaf(root, leaf);
3488 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3489 total_size, btrfs_leaf_free_space(root, leaf));
3493 if (slot != nritems) {
3494 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3496 if (old_data < data_end) {
3497 btrfs_print_leaf(root, leaf);
3498 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3499 slot, old_data, data_end);
3503 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3505 /* first correct the data pointers */
3506 WARN_ON(leaf->map_token);
3507 for (i = slot; i < nritems; i++) {
3510 item = btrfs_item_nr(leaf, i);
3511 if (!leaf->map_token) {
3512 map_extent_buffer(leaf, (unsigned long)item,
3513 sizeof(struct btrfs_item),
3514 &leaf->map_token, &leaf->kaddr,
3515 &leaf->map_start, &leaf->map_len,
3519 ioff = btrfs_item_offset(leaf, item);
3520 btrfs_set_item_offset(leaf, item, ioff - total_data);
3522 if (leaf->map_token) {
3523 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3524 leaf->map_token = NULL;
3527 /* shift the items */
3528 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3529 btrfs_item_nr_offset(slot),
3530 (nritems - slot) * sizeof(struct btrfs_item));
3532 /* shift the data */
3533 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3534 data_end - total_data, btrfs_leaf_data(leaf) +
3535 data_end, old_data - data_end);
3536 data_end = old_data;
3539 /* setup the item for the new data */
3540 for (i = 0; i < nr; i++) {
3541 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3542 btrfs_set_item_key(leaf, &disk_key, slot + i);
3543 item = btrfs_item_nr(leaf, slot + i);
3544 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3545 data_end -= data_size[i];
3546 btrfs_set_item_size(leaf, item, data_size[i]);
3549 btrfs_set_header_nritems(leaf, nritems + nr);
3553 struct btrfs_disk_key disk_key;
3554 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3555 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3557 btrfs_unlock_up_safe(path, 1);
3558 btrfs_mark_buffer_dirty(leaf);
3560 if (btrfs_leaf_free_space(root, leaf) < 0) {
3561 btrfs_print_leaf(root, leaf);
3568 * Given a key and some data, insert items into the tree.
3569 * This does all the path init required, making room in the tree if needed.
3571 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3572 struct btrfs_root *root,
3573 struct btrfs_path *path,
3574 struct btrfs_key *cpu_key, u32 *data_size,
3577 struct extent_buffer *leaf;
3584 for (i = 0; i < nr; i++)
3585 total_data += data_size[i];
3587 total_size = total_data + (nr * sizeof(struct btrfs_item));
3588 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3594 leaf = path->nodes[0];
3595 slot = path->slots[0];
3598 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3599 total_data, total_size, nr);
3606 * Given a key and some data, insert an item into the tree.
3607 * This does all the path init required, making room in the tree if needed.
3609 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3610 *root, struct btrfs_key *cpu_key, void *data, u32
3614 struct btrfs_path *path;
3615 struct extent_buffer *leaf;
3618 path = btrfs_alloc_path();
3620 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3622 leaf = path->nodes[0];
3623 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3624 write_extent_buffer(leaf, data, ptr, data_size);
3625 btrfs_mark_buffer_dirty(leaf);
3627 btrfs_free_path(path);
3632 * delete the pointer from a given node.
3634 * the tree should have been previously balanced so the deletion does not
3637 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3638 struct btrfs_path *path, int level, int slot)
3640 struct extent_buffer *parent = path->nodes[level];
3645 nritems = btrfs_header_nritems(parent);
3646 if (slot != nritems - 1) {
3647 memmove_extent_buffer(parent,
3648 btrfs_node_key_ptr_offset(slot),
3649 btrfs_node_key_ptr_offset(slot + 1),
3650 sizeof(struct btrfs_key_ptr) *
3651 (nritems - slot - 1));
3654 btrfs_set_header_nritems(parent, nritems);
3655 if (nritems == 0 && parent == root->node) {
3656 BUG_ON(btrfs_header_level(root->node) != 1);
3657 /* just turn the root into a leaf and break */
3658 btrfs_set_header_level(root->node, 0);
3659 } else if (slot == 0) {
3660 struct btrfs_disk_key disk_key;
3662 btrfs_node_key(parent, &disk_key, 0);
3663 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3667 btrfs_mark_buffer_dirty(parent);
3672 * a helper function to delete the leaf pointed to by path->slots[1] and
3673 * path->nodes[1]. bytenr is the node block pointer, but since the callers
3674 * already know it, it is faster to have them pass it down than to
3675 * read it out of the node again.
3677 * This deletes the pointer in path->nodes[1] and frees the leaf
3678 * block extent. zero is returned if it all worked out, < 0 otherwise.
3680 * The path must have already been setup for deleting the leaf, including
3681 * all the proper balancing. path->nodes[1] must be locked.
3683 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3684 struct btrfs_root *root,
3685 struct btrfs_path *path, u64 bytenr)
3688 u64 root_gen = btrfs_header_generation(path->nodes[1]);
3689 u64 parent_start = path->nodes[1]->start;
3690 u64 parent_owner = btrfs_header_owner(path->nodes[1]);
3692 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3697 * btrfs_free_extent is expensive, we want to make sure we
3698 * aren't holding any locks when we call it
3700 btrfs_unlock_up_safe(path, 0);
3702 ret = btrfs_free_extent(trans, root, bytenr,
3703 btrfs_level_size(root, 0),
3704 parent_start, parent_owner,
3709 * delete the item at the leaf level in path. If that empties
3710 * the leaf, remove it from the tree
3712 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3713 struct btrfs_path *path, int slot, int nr)
3715 struct extent_buffer *leaf;
3716 struct btrfs_item *item;
3724 leaf = path->nodes[0];
3725 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3727 for (i = 0; i < nr; i++)
3728 dsize += btrfs_item_size_nr(leaf, slot + i);
3730 nritems = btrfs_header_nritems(leaf);
3732 if (slot + nr != nritems) {
3733 int data_end = leaf_data_end(root, leaf);
3735 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3737 btrfs_leaf_data(leaf) + data_end,
3738 last_off - data_end);
3740 for (i = slot + nr; i < nritems; i++) {
3743 item = btrfs_item_nr(leaf, i);
3744 if (!leaf->map_token) {
3745 map_extent_buffer(leaf, (unsigned long)item,
3746 sizeof(struct btrfs_item),
3747 &leaf->map_token, &leaf->kaddr,
3748 &leaf->map_start, &leaf->map_len,
3751 ioff = btrfs_item_offset(leaf, item);
3752 btrfs_set_item_offset(leaf, item, ioff + dsize);
3755 if (leaf->map_token) {
3756 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3757 leaf->map_token = NULL;
3760 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3761 btrfs_item_nr_offset(slot + nr),
3762 sizeof(struct btrfs_item) *
3763 (nritems - slot - nr));
3765 btrfs_set_header_nritems(leaf, nritems - nr);
3768 /* delete the leaf if we've emptied it */
3770 if (leaf == root->node) {
3771 btrfs_set_header_level(leaf, 0);
3773 ret = btrfs_del_leaf(trans, root, path, leaf->start);
3777 int used = leaf_space_used(leaf, 0, nritems);
3779 struct btrfs_disk_key disk_key;
3781 btrfs_item_key(leaf, &disk_key, 0);
3782 wret = fixup_low_keys(trans, root, path,
3788 /* delete the leaf if it is mostly empty */
3789 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3790 /* push_leaf_left fixes the path.
3791 * make sure the path still points to our leaf
3792 * for possible call to del_ptr below
3794 slot = path->slots[1];
3795 extent_buffer_get(leaf);
3797 btrfs_set_path_blocking(path);
3798 wret = push_leaf_left(trans, root, path, 1, 1);
3799 if (wret < 0 && wret != -ENOSPC)
3802 if (path->nodes[0] == leaf &&
3803 btrfs_header_nritems(leaf)) {
3804 wret = push_leaf_right(trans, root, path, 1, 1);
3805 if (wret < 0 && wret != -ENOSPC)
3809 if (btrfs_header_nritems(leaf) == 0) {
3810 path->slots[1] = slot;
3811 ret = btrfs_del_leaf(trans, root, path,
3814 free_extent_buffer(leaf);
3816 /* if we're still in the path, make sure
3817 * we're dirty. Otherwise, one of the
3818 * push_leaf functions must have already
3819 * dirtied this buffer
3821 if (path->nodes[0] == leaf)
3822 btrfs_mark_buffer_dirty(leaf);
3823 free_extent_buffer(leaf);
3826 btrfs_mark_buffer_dirty(leaf);
3833 * search the tree again to find a leaf with lesser keys
3834 * returns 0 if it found something or 1 if there are no lesser leaves.
3835 * returns < 0 on io errors.
3837 * This may release the path, and so you may lose any locks held at the
3840 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3842 struct btrfs_key key;
3843 struct btrfs_disk_key found_key;
3846 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3850 else if (key.type > 0)
3852 else if (key.objectid > 0)
3857 btrfs_release_path(root, path);
3858 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3861 btrfs_item_key(path->nodes[0], &found_key, 0);
3862 ret = comp_keys(&found_key, &key);
3869 * A helper function to walk down the tree starting at min_key, and looking
3870 * for nodes or leaves that are either in cache or have a minimum
3871 * transaction id. This is used by the btree defrag code, and tree logging
3873 * This does not cow, but it does stuff the starting key it finds back
3874 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3875 * key and get a writable path.
3877 * This does lock as it descends, and path->keep_locks should be set
3878 * to 1 by the caller.
3880 * This honors path->lowest_level to prevent descent past a given level
3883 * min_trans indicates the oldest transaction that you are interested
3884 * in walking through. Any nodes or leaves older than min_trans are
3885 * skipped over (without reading them).
3887 * returns zero if something useful was found, < 0 on error and 1 if there
3888 * was nothing in the tree that matched the search criteria.
3890 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3891 struct btrfs_key *max_key,
3892 struct btrfs_path *path, int cache_only,
3895 struct extent_buffer *cur;
3896 struct btrfs_key found_key;
3903 WARN_ON(!path->keep_locks);
3905 cur = btrfs_lock_root_node(root);
3906 level = btrfs_header_level(cur);
3907 WARN_ON(path->nodes[level]);
3908 path->nodes[level] = cur;
3909 path->locks[level] = 1;
3911 if (btrfs_header_generation(cur) < min_trans) {
3916 nritems = btrfs_header_nritems(cur);
3917 level = btrfs_header_level(cur);
3918 sret = bin_search(cur, min_key, level, &slot);
3920 /* at the lowest level, we're done, setup the path and exit */
3921 if (level == path->lowest_level) {
3922 if (slot >= nritems)
3925 path->slots[level] = slot;
3926 btrfs_item_key_to_cpu(cur, &found_key, slot);
3929 if (sret && slot > 0)
3932 * check this node pointer against the cache_only and
3933 * min_trans parameters. If it isn't in cache or is too
3934 * old, skip to the next one.
3936 while (slot < nritems) {
3939 struct extent_buffer *tmp;
3940 struct btrfs_disk_key disk_key;
3942 blockptr = btrfs_node_blockptr(cur, slot);
3943 gen = btrfs_node_ptr_generation(cur, slot);
3944 if (gen < min_trans) {
3952 btrfs_node_key(cur, &disk_key, slot);
3953 if (comp_keys(&disk_key, max_key) >= 0) {
3959 tmp = btrfs_find_tree_block(root, blockptr,
3960 btrfs_level_size(root, level - 1));
3962 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3963 free_extent_buffer(tmp);
3967 free_extent_buffer(tmp);
3972 * we didn't find a candidate key in this node, walk forward
3973 * and find another one
3975 if (slot >= nritems) {
3976 path->slots[level] = slot;
3977 btrfs_set_path_blocking(path);
3978 sret = btrfs_find_next_key(root, path, min_key, level,
3979 cache_only, min_trans);
3981 btrfs_release_path(root, path);
3987 /* save our key for returning back */
3988 btrfs_node_key_to_cpu(cur, &found_key, slot);
3989 path->slots[level] = slot;
3990 if (level == path->lowest_level) {
3992 unlock_up(path, level, 1);
3995 btrfs_set_path_blocking(path);
3996 cur = read_node_slot(root, cur, slot);
3998 btrfs_tree_lock(cur);
4000 path->locks[level - 1] = 1;
4001 path->nodes[level - 1] = cur;
4002 unlock_up(path, level, 1);
4003 btrfs_clear_path_blocking(path, NULL);
4007 memcpy(min_key, &found_key, sizeof(found_key));
4008 btrfs_set_path_blocking(path);
4013 * this is similar to btrfs_next_leaf, but does not try to preserve
4014 * and fixup the path. It looks for and returns the next key in the
4015 * tree based on the current path and the cache_only and min_trans
4018 * 0 is returned if another key is found, < 0 if there are any errors
4019 * and 1 is returned if there are no higher keys in the tree
4021 * path->keep_locks should be set to 1 on the search made before
4022 * calling this function.
4024 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4025 struct btrfs_key *key, int lowest_level,
4026 int cache_only, u64 min_trans)
4028 int level = lowest_level;
4030 struct extent_buffer *c;
4032 WARN_ON(!path->keep_locks);
4033 while (level < BTRFS_MAX_LEVEL) {
4034 if (!path->nodes[level])
4037 slot = path->slots[level] + 1;
4038 c = path->nodes[level];
4040 if (slot >= btrfs_header_nritems(c)) {
4042 if (level == BTRFS_MAX_LEVEL)
4047 btrfs_item_key_to_cpu(c, key, slot);
4049 u64 blockptr = btrfs_node_blockptr(c, slot);
4050 u64 gen = btrfs_node_ptr_generation(c, slot);
4053 struct extent_buffer *cur;
4054 cur = btrfs_find_tree_block(root, blockptr,
4055 btrfs_level_size(root, level - 1));
4056 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4059 free_extent_buffer(cur);
4062 free_extent_buffer(cur);
4064 if (gen < min_trans) {
4068 btrfs_node_key_to_cpu(c, key, slot);
4076 * search the tree again to find a leaf with greater keys
4077 * returns 0 if it found something or 1 if there are no greater leaves.
4078 * returns < 0 on io errors.
4080 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4084 struct extent_buffer *c;
4085 struct extent_buffer *next = NULL;
4086 struct btrfs_key key;
4090 nritems = btrfs_header_nritems(path->nodes[0]);
4094 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4096 btrfs_release_path(root, path);
4097 path->keep_locks = 1;
4098 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4099 path->keep_locks = 0;
4104 btrfs_set_path_blocking(path);
4105 nritems = btrfs_header_nritems(path->nodes[0]);
4107 * by releasing the path above we dropped all our locks. A balance
4108 * could have added more items next to the key that used to be
4109 * at the very end of the block. So, check again here and
4110 * advance the path if there are now more items available.
4112 if (nritems > 0 && path->slots[0] < nritems - 1) {
4117 while (level < BTRFS_MAX_LEVEL) {
4118 if (!path->nodes[level])
4121 slot = path->slots[level] + 1;
4122 c = path->nodes[level];
4123 if (slot >= btrfs_header_nritems(c)) {
4125 if (level == BTRFS_MAX_LEVEL)
4131 btrfs_tree_unlock(next);
4132 free_extent_buffer(next);
4135 /* the path was set to blocking above */
4136 if (level == 1 && (path->locks[1] || path->skip_locking) &&
4138 reada_for_search(root, path, level, slot, 0);
4140 next = read_node_slot(root, c, slot);
4141 if (!path->skip_locking) {
4142 btrfs_assert_tree_locked(c);
4143 btrfs_tree_lock(next);
4144 btrfs_set_lock_blocking(next);
4148 path->slots[level] = slot;
4151 c = path->nodes[level];
4152 if (path->locks[level])
4153 btrfs_tree_unlock(c);
4154 free_extent_buffer(c);
4155 path->nodes[level] = next;
4156 path->slots[level] = 0;
4157 if (!path->skip_locking)
4158 path->locks[level] = 1;
4162 btrfs_set_path_blocking(path);
4163 if (level == 1 && path->locks[1] && path->reada)
4164 reada_for_search(root, path, level, slot, 0);
4165 next = read_node_slot(root, next, 0);
4166 if (!path->skip_locking) {
4167 btrfs_assert_tree_locked(path->nodes[level]);
4168 btrfs_tree_lock(next);
4169 btrfs_set_lock_blocking(next);
4173 unlock_up(path, 0, 1);
4178 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4179 * searching until it gets past min_objectid or finds an item of 'type'
4181 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4183 int btrfs_previous_item(struct btrfs_root *root,
4184 struct btrfs_path *path, u64 min_objectid,
4187 struct btrfs_key found_key;
4188 struct extent_buffer *leaf;
4193 if (path->slots[0] == 0) {
4194 btrfs_set_path_blocking(path);
4195 ret = btrfs_prev_leaf(root, path);
4201 leaf = path->nodes[0];
4202 nritems = btrfs_header_nritems(leaf);
4205 if (path->slots[0] == nritems)
4208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4209 if (found_key.type == type)
4211 if (found_key.objectid < min_objectid)
4213 if (found_key.objectid == min_objectid &&
4214 found_key.type < type)