2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
23 #include "transaction.h"
24 #include "print-tree.h"
27 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
28 *root, struct btrfs_path *path, int level);
29 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
30 *root, struct btrfs_key *ins_key,
31 struct btrfs_path *path, int data_size, int extend);
32 static int push_node_left(struct btrfs_trans_handle *trans,
33 struct btrfs_root *root, struct extent_buffer *dst,
34 struct extent_buffer *src, int empty);
35 static int balance_node_right(struct btrfs_trans_handle *trans,
36 struct btrfs_root *root,
37 struct extent_buffer *dst_buf,
38 struct extent_buffer *src_buf);
39 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
40 struct btrfs_path *path, int level, int slot);
42 struct btrfs_path *btrfs_alloc_path(void)
44 struct btrfs_path *path;
45 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
50 * set all locked nodes in the path to blocking locks. This should
51 * be done before scheduling
53 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
56 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
57 if (p->nodes[i] && p->locks[i])
58 btrfs_set_lock_blocking(p->nodes[i]);
63 * reset all the locked nodes in the patch to spinning locks.
65 * held is used to keep lockdep happy, when lockdep is enabled
66 * we set held to a blocking lock before we go around and
67 * retake all the spinlocks in the path. You can safely use NULL
70 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
71 struct extent_buffer *held)
75 #ifdef CONFIG_DEBUG_LOCK_ALLOC
76 /* lockdep really cares that we take all of these spinlocks
77 * in the right order. If any of the locks in the path are not
78 * currently blocking, it is going to complain. So, make really
79 * really sure by forcing the path to blocking before we clear
83 btrfs_set_lock_blocking(held);
84 btrfs_set_path_blocking(p);
87 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
88 if (p->nodes[i] && p->locks[i])
89 btrfs_clear_lock_blocking(p->nodes[i]);
92 #ifdef CONFIG_DEBUG_LOCK_ALLOC
94 btrfs_clear_lock_blocking(held);
98 /* this also releases the path */
99 void btrfs_free_path(struct btrfs_path *p)
103 btrfs_release_path(p);
104 kmem_cache_free(btrfs_path_cachep, p);
108 * path release drops references on the extent buffers in the path
109 * and it drops any locks held by this path
111 * It is safe to call this on paths that no locks or extent buffers held.
113 noinline void btrfs_release_path(struct btrfs_path *p)
117 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
122 btrfs_tree_unlock(p->nodes[i]);
125 free_extent_buffer(p->nodes[i]);
131 * safely gets a reference on the root node of a tree. A lock
132 * is not taken, so a concurrent writer may put a different node
133 * at the root of the tree. See btrfs_lock_root_node for the
136 * The extent buffer returned by this has a reference taken, so
137 * it won't disappear. It may stop being the root of the tree
138 * at any time because there are no locks held.
140 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
142 struct extent_buffer *eb;
145 eb = rcu_dereference(root->node);
146 extent_buffer_get(eb);
151 /* loop around taking references on and locking the root node of the
152 * tree until you end up with a lock on the root. A locked buffer
153 * is returned, with a reference held.
155 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
157 struct extent_buffer *eb;
160 eb = btrfs_root_node(root);
162 if (eb == root->node)
164 btrfs_tree_unlock(eb);
165 free_extent_buffer(eb);
170 /* cowonly root (everything not a reference counted cow subvolume), just get
171 * put onto a simple dirty list. transaction.c walks this to make sure they
172 * get properly updated on disk.
174 static void add_root_to_dirty_list(struct btrfs_root *root)
176 if (root->track_dirty && list_empty(&root->dirty_list)) {
177 list_add(&root->dirty_list,
178 &root->fs_info->dirty_cowonly_roots);
183 * used by snapshot creation to make a copy of a root for a tree with
184 * a given objectid. The buffer with the new root node is returned in
185 * cow_ret, and this func returns zero on success or a negative error code.
187 int btrfs_copy_root(struct btrfs_trans_handle *trans,
188 struct btrfs_root *root,
189 struct extent_buffer *buf,
190 struct extent_buffer **cow_ret, u64 new_root_objectid)
192 struct extent_buffer *cow;
195 struct btrfs_disk_key disk_key;
197 WARN_ON(root->ref_cows && trans->transid !=
198 root->fs_info->running_transaction->transid);
199 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
201 level = btrfs_header_level(buf);
203 btrfs_item_key(buf, &disk_key, 0);
205 btrfs_node_key(buf, &disk_key, 0);
207 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
208 new_root_objectid, &disk_key, level,
213 copy_extent_buffer(cow, buf, 0, 0, cow->len);
214 btrfs_set_header_bytenr(cow, cow->start);
215 btrfs_set_header_generation(cow, trans->transid);
216 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
217 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
218 BTRFS_HEADER_FLAG_RELOC);
219 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
220 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
222 btrfs_set_header_owner(cow, new_root_objectid);
224 write_extent_buffer(cow, root->fs_info->fsid,
225 (unsigned long)btrfs_header_fsid(cow),
228 WARN_ON(btrfs_header_generation(buf) > trans->transid);
229 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
230 ret = btrfs_inc_ref(trans, root, cow, 1);
232 ret = btrfs_inc_ref(trans, root, cow, 0);
237 btrfs_mark_buffer_dirty(cow);
243 * check if the tree block can be shared by multiple trees
245 int btrfs_block_can_be_shared(struct btrfs_root *root,
246 struct extent_buffer *buf)
249 * Tree blocks not in refernece counted trees and tree roots
250 * are never shared. If a block was allocated after the last
251 * snapshot and the block was not allocated by tree relocation,
252 * we know the block is not shared.
254 if (root->ref_cows &&
255 buf != root->node && buf != root->commit_root &&
256 (btrfs_header_generation(buf) <=
257 btrfs_root_last_snapshot(&root->root_item) ||
258 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
261 if (root->ref_cows &&
262 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
268 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
269 struct btrfs_root *root,
270 struct extent_buffer *buf,
271 struct extent_buffer *cow,
281 * Backrefs update rules:
283 * Always use full backrefs for extent pointers in tree block
284 * allocated by tree relocation.
286 * If a shared tree block is no longer referenced by its owner
287 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
288 * use full backrefs for extent pointers in tree block.
290 * If a tree block is been relocating
291 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
292 * use full backrefs for extent pointers in tree block.
293 * The reason for this is some operations (such as drop tree)
294 * are only allowed for blocks use full backrefs.
297 if (btrfs_block_can_be_shared(root, buf)) {
298 ret = btrfs_lookup_extent_info(trans, root, buf->start,
299 buf->len, &refs, &flags);
304 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
305 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
306 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
311 owner = btrfs_header_owner(buf);
312 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
313 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
316 if ((owner == root->root_key.objectid ||
317 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
318 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
319 ret = btrfs_inc_ref(trans, root, buf, 1);
322 if (root->root_key.objectid ==
323 BTRFS_TREE_RELOC_OBJECTID) {
324 ret = btrfs_dec_ref(trans, root, buf, 0);
326 ret = btrfs_inc_ref(trans, root, cow, 1);
329 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
332 if (root->root_key.objectid ==
333 BTRFS_TREE_RELOC_OBJECTID)
334 ret = btrfs_inc_ref(trans, root, cow, 1);
336 ret = btrfs_inc_ref(trans, root, cow, 0);
339 if (new_flags != 0) {
340 ret = btrfs_set_disk_extent_flags(trans, root,
347 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
348 if (root->root_key.objectid ==
349 BTRFS_TREE_RELOC_OBJECTID)
350 ret = btrfs_inc_ref(trans, root, cow, 1);
352 ret = btrfs_inc_ref(trans, root, cow, 0);
354 ret = btrfs_dec_ref(trans, root, buf, 1);
357 clean_tree_block(trans, root, buf);
364 * does the dirty work in cow of a single block. The parent block (if
365 * supplied) is updated to point to the new cow copy. The new buffer is marked
366 * dirty and returned locked. If you modify the block it needs to be marked
369 * search_start -- an allocation hint for the new block
371 * empty_size -- a hint that you plan on doing more cow. This is the size in
372 * bytes the allocator should try to find free next to the block it returns.
373 * This is just a hint and may be ignored by the allocator.
375 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
376 struct btrfs_root *root,
377 struct extent_buffer *buf,
378 struct extent_buffer *parent, int parent_slot,
379 struct extent_buffer **cow_ret,
380 u64 search_start, u64 empty_size)
382 struct btrfs_disk_key disk_key;
383 struct extent_buffer *cow;
392 btrfs_assert_tree_locked(buf);
394 WARN_ON(root->ref_cows && trans->transid !=
395 root->fs_info->running_transaction->transid);
396 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
398 level = btrfs_header_level(buf);
401 btrfs_item_key(buf, &disk_key, 0);
403 btrfs_node_key(buf, &disk_key, 0);
405 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
407 parent_start = parent->start;
413 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
414 root->root_key.objectid, &disk_key,
415 level, search_start, empty_size);
419 /* cow is set to blocking by btrfs_init_new_buffer */
421 copy_extent_buffer(cow, buf, 0, 0, cow->len);
422 btrfs_set_header_bytenr(cow, cow->start);
423 btrfs_set_header_generation(cow, trans->transid);
424 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
425 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
426 BTRFS_HEADER_FLAG_RELOC);
427 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
428 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
430 btrfs_set_header_owner(cow, root->root_key.objectid);
432 write_extent_buffer(cow, root->fs_info->fsid,
433 (unsigned long)btrfs_header_fsid(cow),
436 update_ref_for_cow(trans, root, buf, cow, &last_ref);
439 btrfs_reloc_cow_block(trans, root, buf, cow);
441 if (buf == root->node) {
442 WARN_ON(parent && parent != buf);
443 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
444 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
445 parent_start = buf->start;
449 extent_buffer_get(cow);
450 rcu_assign_pointer(root->node, cow);
452 btrfs_free_tree_block(trans, root, buf, parent_start,
454 free_extent_buffer(buf);
455 add_root_to_dirty_list(root);
457 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
458 parent_start = parent->start;
462 WARN_ON(trans->transid != btrfs_header_generation(parent));
463 btrfs_set_node_blockptr(parent, parent_slot,
465 btrfs_set_node_ptr_generation(parent, parent_slot,
467 btrfs_mark_buffer_dirty(parent);
468 btrfs_free_tree_block(trans, root, buf, parent_start,
472 btrfs_tree_unlock(buf);
473 free_extent_buffer(buf);
474 btrfs_mark_buffer_dirty(cow);
479 static inline int should_cow_block(struct btrfs_trans_handle *trans,
480 struct btrfs_root *root,
481 struct extent_buffer *buf)
483 if (btrfs_header_generation(buf) == trans->transid &&
484 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
485 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
486 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
492 * cows a single block, see __btrfs_cow_block for the real work.
493 * This version of it has extra checks so that a block isn't cow'd more than
494 * once per transaction, as long as it hasn't been written yet
496 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
497 struct btrfs_root *root, struct extent_buffer *buf,
498 struct extent_buffer *parent, int parent_slot,
499 struct extent_buffer **cow_ret)
504 if (trans->transaction != root->fs_info->running_transaction) {
505 printk(KERN_CRIT "trans %llu running %llu\n",
506 (unsigned long long)trans->transid,
508 root->fs_info->running_transaction->transid);
511 if (trans->transid != root->fs_info->generation) {
512 printk(KERN_CRIT "trans %llu running %llu\n",
513 (unsigned long long)trans->transid,
514 (unsigned long long)root->fs_info->generation);
518 if (!should_cow_block(trans, root, buf)) {
523 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
526 btrfs_set_lock_blocking(parent);
527 btrfs_set_lock_blocking(buf);
529 ret = __btrfs_cow_block(trans, root, buf, parent,
530 parent_slot, cow_ret, search_start, 0);
532 trace_btrfs_cow_block(root, buf, *cow_ret);
538 * helper function for defrag to decide if two blocks pointed to by a
539 * node are actually close by
541 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
543 if (blocknr < other && other - (blocknr + blocksize) < 32768)
545 if (blocknr > other && blocknr - (other + blocksize) < 32768)
551 * compare two keys in a memcmp fashion
553 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
557 btrfs_disk_key_to_cpu(&k1, disk);
559 return btrfs_comp_cpu_keys(&k1, k2);
563 * same as comp_keys only with two btrfs_key's
565 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
567 if (k1->objectid > k2->objectid)
569 if (k1->objectid < k2->objectid)
571 if (k1->type > k2->type)
573 if (k1->type < k2->type)
575 if (k1->offset > k2->offset)
577 if (k1->offset < k2->offset)
583 * this is used by the defrag code to go through all the
584 * leaves pointed to by a node and reallocate them so that
585 * disk order is close to key order
587 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
588 struct btrfs_root *root, struct extent_buffer *parent,
589 int start_slot, int cache_only, u64 *last_ret,
590 struct btrfs_key *progress)
592 struct extent_buffer *cur;
595 u64 search_start = *last_ret;
605 int progress_passed = 0;
606 struct btrfs_disk_key disk_key;
608 parent_level = btrfs_header_level(parent);
609 if (cache_only && parent_level != 1)
612 if (trans->transaction != root->fs_info->running_transaction)
614 if (trans->transid != root->fs_info->generation)
617 parent_nritems = btrfs_header_nritems(parent);
618 blocksize = btrfs_level_size(root, parent_level - 1);
619 end_slot = parent_nritems;
621 if (parent_nritems == 1)
624 btrfs_set_lock_blocking(parent);
626 for (i = start_slot; i < end_slot; i++) {
629 if (!parent->map_token) {
630 map_extent_buffer(parent,
631 btrfs_node_key_ptr_offset(i),
632 sizeof(struct btrfs_key_ptr),
633 &parent->map_token, &parent->kaddr,
634 &parent->map_start, &parent->map_len,
637 btrfs_node_key(parent, &disk_key, i);
638 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
642 blocknr = btrfs_node_blockptr(parent, i);
643 gen = btrfs_node_ptr_generation(parent, i);
645 last_block = blocknr;
648 other = btrfs_node_blockptr(parent, i - 1);
649 close = close_blocks(blocknr, other, blocksize);
651 if (!close && i < end_slot - 2) {
652 other = btrfs_node_blockptr(parent, i + 1);
653 close = close_blocks(blocknr, other, blocksize);
656 last_block = blocknr;
659 if (parent->map_token) {
660 unmap_extent_buffer(parent, parent->map_token,
662 parent->map_token = NULL;
665 cur = btrfs_find_tree_block(root, blocknr, blocksize);
667 uptodate = btrfs_buffer_uptodate(cur, gen);
670 if (!cur || !uptodate) {
672 free_extent_buffer(cur);
676 cur = read_tree_block(root, blocknr,
680 } else if (!uptodate) {
681 btrfs_read_buffer(cur, gen);
684 if (search_start == 0)
685 search_start = last_block;
687 btrfs_tree_lock(cur);
688 btrfs_set_lock_blocking(cur);
689 err = __btrfs_cow_block(trans, root, cur, parent, i,
692 (end_slot - i) * blocksize));
694 btrfs_tree_unlock(cur);
695 free_extent_buffer(cur);
698 search_start = cur->start;
699 last_block = cur->start;
700 *last_ret = search_start;
701 btrfs_tree_unlock(cur);
702 free_extent_buffer(cur);
704 if (parent->map_token) {
705 unmap_extent_buffer(parent, parent->map_token,
707 parent->map_token = NULL;
713 * The leaf data grows from end-to-front in the node.
714 * this returns the address of the start of the last item,
715 * which is the stop of the leaf data stack
717 static inline unsigned int leaf_data_end(struct btrfs_root *root,
718 struct extent_buffer *leaf)
720 u32 nr = btrfs_header_nritems(leaf);
722 return BTRFS_LEAF_DATA_SIZE(root);
723 return btrfs_item_offset_nr(leaf, nr - 1);
728 * search for key in the extent_buffer. The items start at offset p,
729 * and they are item_size apart. There are 'max' items in p.
731 * the slot in the array is returned via slot, and it points to
732 * the place where you would insert key if it is not found in
735 * slot may point to max if the key is bigger than all of the keys
737 static noinline int generic_bin_search(struct extent_buffer *eb,
739 int item_size, struct btrfs_key *key,
746 struct btrfs_disk_key *tmp = NULL;
747 struct btrfs_disk_key unaligned;
748 unsigned long offset;
749 char *map_token = NULL;
751 unsigned long map_start = 0;
752 unsigned long map_len = 0;
756 mid = (low + high) / 2;
757 offset = p + mid * item_size;
759 if (!map_token || offset < map_start ||
760 (offset + sizeof(struct btrfs_disk_key)) >
761 map_start + map_len) {
763 unmap_extent_buffer(eb, map_token, KM_USER0);
767 err = map_private_extent_buffer(eb, offset,
768 sizeof(struct btrfs_disk_key),
770 &map_start, &map_len, KM_USER0);
773 tmp = (struct btrfs_disk_key *)(kaddr + offset -
776 read_extent_buffer(eb, &unaligned,
777 offset, sizeof(unaligned));
782 tmp = (struct btrfs_disk_key *)(kaddr + offset -
785 ret = comp_keys(tmp, key);
794 unmap_extent_buffer(eb, map_token, KM_USER0);
800 unmap_extent_buffer(eb, map_token, KM_USER0);
805 * simple bin_search frontend that does the right thing for
808 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
809 int level, int *slot)
812 return generic_bin_search(eb,
813 offsetof(struct btrfs_leaf, items),
814 sizeof(struct btrfs_item),
815 key, btrfs_header_nritems(eb),
818 return generic_bin_search(eb,
819 offsetof(struct btrfs_node, ptrs),
820 sizeof(struct btrfs_key_ptr),
821 key, btrfs_header_nritems(eb),
827 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
828 int level, int *slot)
830 return bin_search(eb, key, level, slot);
833 static void root_add_used(struct btrfs_root *root, u32 size)
835 spin_lock(&root->accounting_lock);
836 btrfs_set_root_used(&root->root_item,
837 btrfs_root_used(&root->root_item) + size);
838 spin_unlock(&root->accounting_lock);
841 static void root_sub_used(struct btrfs_root *root, u32 size)
843 spin_lock(&root->accounting_lock);
844 btrfs_set_root_used(&root->root_item,
845 btrfs_root_used(&root->root_item) - size);
846 spin_unlock(&root->accounting_lock);
849 /* given a node and slot number, this reads the blocks it points to. The
850 * extent buffer is returned with a reference taken (but unlocked).
851 * NULL is returned on error.
853 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
854 struct extent_buffer *parent, int slot)
856 int level = btrfs_header_level(parent);
859 if (slot >= btrfs_header_nritems(parent))
864 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
865 btrfs_level_size(root, level - 1),
866 btrfs_node_ptr_generation(parent, slot));
870 * node level balancing, used to make sure nodes are in proper order for
871 * item deletion. We balance from the top down, so we have to make sure
872 * that a deletion won't leave an node completely empty later on.
874 static noinline int balance_level(struct btrfs_trans_handle *trans,
875 struct btrfs_root *root,
876 struct btrfs_path *path, int level)
878 struct extent_buffer *right = NULL;
879 struct extent_buffer *mid;
880 struct extent_buffer *left = NULL;
881 struct extent_buffer *parent = NULL;
885 int orig_slot = path->slots[level];
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);
919 btrfs_tree_unlock(child);
920 free_extent_buffer(child);
924 rcu_assign_pointer(root->node, child);
926 add_root_to_dirty_list(root);
927 btrfs_tree_unlock(child);
929 path->locks[level] = 0;
930 path->nodes[level] = NULL;
931 clean_tree_block(trans, root, mid);
932 btrfs_tree_unlock(mid);
933 /* once for the path */
934 free_extent_buffer(mid);
936 root_sub_used(root, mid->len);
937 btrfs_free_tree_block(trans, root, mid, 0, 1);
938 /* once for the root ptr */
939 free_extent_buffer(mid);
942 if (btrfs_header_nritems(mid) >
943 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
946 btrfs_header_nritems(mid);
948 left = read_node_slot(root, parent, pslot - 1);
950 btrfs_tree_lock(left);
951 btrfs_set_lock_blocking(left);
952 wret = btrfs_cow_block(trans, root, left,
953 parent, pslot - 1, &left);
959 right = read_node_slot(root, parent, pslot + 1);
961 btrfs_tree_lock(right);
962 btrfs_set_lock_blocking(right);
963 wret = btrfs_cow_block(trans, root, right,
964 parent, pslot + 1, &right);
971 /* first, try to make some room in the middle buffer */
973 orig_slot += btrfs_header_nritems(left);
974 wret = push_node_left(trans, root, left, mid, 1);
977 btrfs_header_nritems(mid);
981 * then try to empty the right most buffer into the middle
984 wret = push_node_left(trans, root, mid, right, 1);
985 if (wret < 0 && wret != -ENOSPC)
987 if (btrfs_header_nritems(right) == 0) {
988 clean_tree_block(trans, root, right);
989 btrfs_tree_unlock(right);
990 wret = del_ptr(trans, root, path, level + 1, pslot +
994 root_sub_used(root, right->len);
995 btrfs_free_tree_block(trans, root, right, 0, 1);
996 free_extent_buffer(right);
999 struct btrfs_disk_key right_key;
1000 btrfs_node_key(right, &right_key, 0);
1001 btrfs_set_node_key(parent, &right_key, pslot + 1);
1002 btrfs_mark_buffer_dirty(parent);
1005 if (btrfs_header_nritems(mid) == 1) {
1007 * we're not allowed to leave a node with one item in the
1008 * tree during a delete. A deletion from lower in the tree
1009 * could try to delete the only pointer in this node.
1010 * So, pull some keys from the left.
1011 * There has to be a left pointer at this point because
1012 * otherwise we would have pulled some pointers from the
1016 wret = balance_node_right(trans, root, mid, left);
1022 wret = push_node_left(trans, root, left, mid, 1);
1028 if (btrfs_header_nritems(mid) == 0) {
1029 clean_tree_block(trans, root, mid);
1030 btrfs_tree_unlock(mid);
1031 wret = del_ptr(trans, root, path, level + 1, pslot);
1034 root_sub_used(root, mid->len);
1035 btrfs_free_tree_block(trans, root, mid, 0, 1);
1036 free_extent_buffer(mid);
1039 /* update the parent key to reflect our changes */
1040 struct btrfs_disk_key mid_key;
1041 btrfs_node_key(mid, &mid_key, 0);
1042 btrfs_set_node_key(parent, &mid_key, pslot);
1043 btrfs_mark_buffer_dirty(parent);
1046 /* update the path */
1048 if (btrfs_header_nritems(left) > orig_slot) {
1049 extent_buffer_get(left);
1050 /* left was locked after cow */
1051 path->nodes[level] = left;
1052 path->slots[level + 1] -= 1;
1053 path->slots[level] = orig_slot;
1055 btrfs_tree_unlock(mid);
1056 free_extent_buffer(mid);
1059 orig_slot -= btrfs_header_nritems(left);
1060 path->slots[level] = orig_slot;
1063 /* double check we haven't messed things up */
1065 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1069 btrfs_tree_unlock(right);
1070 free_extent_buffer(right);
1073 if (path->nodes[level] != left)
1074 btrfs_tree_unlock(left);
1075 free_extent_buffer(left);
1080 /* Node balancing for insertion. Here we only split or push nodes around
1081 * when they are completely full. This is also done top down, so we
1082 * have to be pessimistic.
1084 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1085 struct btrfs_root *root,
1086 struct btrfs_path *path, int level)
1088 struct extent_buffer *right = NULL;
1089 struct extent_buffer *mid;
1090 struct extent_buffer *left = NULL;
1091 struct extent_buffer *parent = NULL;
1095 int orig_slot = path->slots[level];
1100 mid = path->nodes[level];
1101 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1103 if (level < BTRFS_MAX_LEVEL - 1)
1104 parent = path->nodes[level + 1];
1105 pslot = path->slots[level + 1];
1110 left = read_node_slot(root, parent, pslot - 1);
1112 /* first, try to make some room in the middle buffer */
1116 btrfs_tree_lock(left);
1117 btrfs_set_lock_blocking(left);
1119 left_nr = btrfs_header_nritems(left);
1120 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1123 ret = btrfs_cow_block(trans, root, left, parent,
1128 wret = push_node_left(trans, root,
1135 struct btrfs_disk_key disk_key;
1136 orig_slot += left_nr;
1137 btrfs_node_key(mid, &disk_key, 0);
1138 btrfs_set_node_key(parent, &disk_key, pslot);
1139 btrfs_mark_buffer_dirty(parent);
1140 if (btrfs_header_nritems(left) > orig_slot) {
1141 path->nodes[level] = left;
1142 path->slots[level + 1] -= 1;
1143 path->slots[level] = orig_slot;
1144 btrfs_tree_unlock(mid);
1145 free_extent_buffer(mid);
1148 btrfs_header_nritems(left);
1149 path->slots[level] = orig_slot;
1150 btrfs_tree_unlock(left);
1151 free_extent_buffer(left);
1155 btrfs_tree_unlock(left);
1156 free_extent_buffer(left);
1158 right = read_node_slot(root, parent, pslot + 1);
1161 * then try to empty the right most buffer into the middle
1166 btrfs_tree_lock(right);
1167 btrfs_set_lock_blocking(right);
1169 right_nr = btrfs_header_nritems(right);
1170 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1173 ret = btrfs_cow_block(trans, root, right,
1179 wret = balance_node_right(trans, root,
1186 struct btrfs_disk_key disk_key;
1188 btrfs_node_key(right, &disk_key, 0);
1189 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1190 btrfs_mark_buffer_dirty(parent);
1192 if (btrfs_header_nritems(mid) <= orig_slot) {
1193 path->nodes[level] = right;
1194 path->slots[level + 1] += 1;
1195 path->slots[level] = orig_slot -
1196 btrfs_header_nritems(mid);
1197 btrfs_tree_unlock(mid);
1198 free_extent_buffer(mid);
1200 btrfs_tree_unlock(right);
1201 free_extent_buffer(right);
1205 btrfs_tree_unlock(right);
1206 free_extent_buffer(right);
1212 * readahead one full node of leaves, finding things that are close
1213 * to the block in 'slot', and triggering ra on them.
1215 static void reada_for_search(struct btrfs_root *root,
1216 struct btrfs_path *path,
1217 int level, int slot, u64 objectid)
1219 struct extent_buffer *node;
1220 struct btrfs_disk_key disk_key;
1226 int direction = path->reada;
1227 struct extent_buffer *eb;
1235 if (!path->nodes[level])
1238 node = path->nodes[level];
1240 search = btrfs_node_blockptr(node, slot);
1241 blocksize = btrfs_level_size(root, level - 1);
1242 eb = btrfs_find_tree_block(root, search, blocksize);
1244 free_extent_buffer(eb);
1250 nritems = btrfs_header_nritems(node);
1253 if (!node->map_token) {
1254 unsigned long offset = btrfs_node_key_ptr_offset(nr);
1255 map_private_extent_buffer(node, offset,
1256 sizeof(struct btrfs_key_ptr),
1260 &node->map_len, KM_USER1);
1262 if (direction < 0) {
1266 } else if (direction > 0) {
1271 if (path->reada < 0 && objectid) {
1272 btrfs_node_key(node, &disk_key, nr);
1273 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1276 search = btrfs_node_blockptr(node, nr);
1277 if ((search <= target && target - search <= 65536) ||
1278 (search > target && search - target <= 65536)) {
1279 gen = btrfs_node_ptr_generation(node, nr);
1280 if (node->map_token) {
1281 unmap_extent_buffer(node, node->map_token,
1283 node->map_token = NULL;
1285 readahead_tree_block(root, search, blocksize, gen);
1289 if ((nread > 65536 || nscan > 32))
1292 if (node->map_token) {
1293 unmap_extent_buffer(node, node->map_token, KM_USER1);
1294 node->map_token = NULL;
1299 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1302 static noinline int reada_for_balance(struct btrfs_root *root,
1303 struct btrfs_path *path, int level)
1307 struct extent_buffer *parent;
1308 struct extent_buffer *eb;
1315 parent = path->nodes[level + 1];
1319 nritems = btrfs_header_nritems(parent);
1320 slot = path->slots[level + 1];
1321 blocksize = btrfs_level_size(root, level);
1324 block1 = btrfs_node_blockptr(parent, slot - 1);
1325 gen = btrfs_node_ptr_generation(parent, slot - 1);
1326 eb = btrfs_find_tree_block(root, block1, blocksize);
1327 if (eb && btrfs_buffer_uptodate(eb, gen))
1329 free_extent_buffer(eb);
1331 if (slot + 1 < nritems) {
1332 block2 = btrfs_node_blockptr(parent, slot + 1);
1333 gen = btrfs_node_ptr_generation(parent, slot + 1);
1334 eb = btrfs_find_tree_block(root, block2, blocksize);
1335 if (eb && btrfs_buffer_uptodate(eb, gen))
1337 free_extent_buffer(eb);
1339 if (block1 || block2) {
1342 /* release the whole path */
1343 btrfs_release_path(path);
1345 /* read the blocks */
1347 readahead_tree_block(root, block1, blocksize, 0);
1349 readahead_tree_block(root, block2, blocksize, 0);
1352 eb = read_tree_block(root, block1, blocksize, 0);
1353 free_extent_buffer(eb);
1356 eb = read_tree_block(root, block2, blocksize, 0);
1357 free_extent_buffer(eb);
1365 * when we walk down the tree, it is usually safe to unlock the higher layers
1366 * in the tree. The exceptions are when our path goes through slot 0, because
1367 * operations on the tree might require changing key pointers higher up in the
1370 * callers might also have set path->keep_locks, which tells this code to keep
1371 * the lock if the path points to the last slot in the block. This is part of
1372 * walking through the tree, and selecting the next slot in the higher block.
1374 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1375 * if lowest_unlock is 1, level 0 won't be unlocked
1377 static noinline void unlock_up(struct btrfs_path *path, int level,
1381 int skip_level = level;
1383 struct extent_buffer *t;
1385 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1386 if (!path->nodes[i])
1388 if (!path->locks[i])
1390 if (!no_skips && path->slots[i] == 0) {
1394 if (!no_skips && path->keep_locks) {
1397 nritems = btrfs_header_nritems(t);
1398 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1403 if (skip_level < i && i >= lowest_unlock)
1407 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1408 btrfs_tree_unlock(t);
1415 * This releases any locks held in the path starting at level and
1416 * going all the way up to the root.
1418 * btrfs_search_slot will keep the lock held on higher nodes in a few
1419 * corner cases, such as COW of the block at slot zero in the node. This
1420 * ignores those rules, and it should only be called when there are no
1421 * more updates to be done higher up in the tree.
1423 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1427 if (path->keep_locks)
1430 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1431 if (!path->nodes[i])
1433 if (!path->locks[i])
1435 btrfs_tree_unlock(path->nodes[i]);
1441 * helper function for btrfs_search_slot. The goal is to find a block
1442 * in cache without setting the path to blocking. If we find the block
1443 * we return zero and the path is unchanged.
1445 * If we can't find the block, we set the path blocking and do some
1446 * reada. -EAGAIN is returned and the search must be repeated.
1449 read_block_for_search(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root, struct btrfs_path *p,
1451 struct extent_buffer **eb_ret, int level, int slot,
1452 struct btrfs_key *key)
1457 struct extent_buffer *b = *eb_ret;
1458 struct extent_buffer *tmp;
1461 blocknr = btrfs_node_blockptr(b, slot);
1462 gen = btrfs_node_ptr_generation(b, slot);
1463 blocksize = btrfs_level_size(root, level - 1);
1465 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1467 if (btrfs_buffer_uptodate(tmp, 0)) {
1468 if (btrfs_buffer_uptodate(tmp, gen)) {
1470 * we found an up to date block without
1477 /* the pages were up to date, but we failed
1478 * the generation number check. Do a full
1479 * read for the generation number that is correct.
1480 * We must do this without dropping locks so
1481 * we can trust our generation number
1483 free_extent_buffer(tmp);
1484 tmp = read_tree_block(root, blocknr, blocksize, gen);
1485 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1489 free_extent_buffer(tmp);
1490 btrfs_release_path(p);
1496 * reduce lock contention at high levels
1497 * of the btree by dropping locks before
1498 * we read. Don't release the lock on the current
1499 * level because we need to walk this node to figure
1500 * out which blocks to read.
1502 btrfs_unlock_up_safe(p, level + 1);
1503 btrfs_set_path_blocking(p);
1505 free_extent_buffer(tmp);
1507 reada_for_search(root, p, level, slot, key->objectid);
1509 btrfs_release_path(p);
1512 tmp = read_tree_block(root, blocknr, blocksize, 0);
1515 * If the read above didn't mark this buffer up to date,
1516 * it will never end up being up to date. Set ret to EIO now
1517 * and give up so that our caller doesn't loop forever
1520 if (!btrfs_buffer_uptodate(tmp, 0))
1522 free_extent_buffer(tmp);
1528 * helper function for btrfs_search_slot. This does all of the checks
1529 * for node-level blocks and does any balancing required based on
1532 * If no extra work was required, zero is returned. If we had to
1533 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1537 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1538 struct btrfs_root *root, struct btrfs_path *p,
1539 struct extent_buffer *b, int level, int ins_len)
1542 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1543 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1546 sret = reada_for_balance(root, p, level);
1550 btrfs_set_path_blocking(p);
1551 sret = split_node(trans, root, p, level);
1552 btrfs_clear_path_blocking(p, NULL);
1559 b = p->nodes[level];
1560 } else if (ins_len < 0 && btrfs_header_nritems(b) <
1561 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1564 sret = reada_for_balance(root, p, level);
1568 btrfs_set_path_blocking(p);
1569 sret = balance_level(trans, root, p, level);
1570 btrfs_clear_path_blocking(p, NULL);
1576 b = p->nodes[level];
1578 btrfs_release_path(p);
1581 BUG_ON(btrfs_header_nritems(b) == 1);
1592 * look for key in the tree. path is filled in with nodes along the way
1593 * if key is found, we return zero and you can find the item in the leaf
1594 * level of the path (level 0)
1596 * If the key isn't found, the path points to the slot where it should
1597 * be inserted, and 1 is returned. If there are other errors during the
1598 * search a negative error number is returned.
1600 * if ins_len > 0, nodes and leaves will be split as we walk down the
1601 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1604 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1605 *root, struct btrfs_key *key, struct btrfs_path *p, int
1608 struct extent_buffer *b;
1613 int lowest_unlock = 1;
1614 u8 lowest_level = 0;
1616 lowest_level = p->lowest_level;
1617 WARN_ON(lowest_level && ins_len > 0);
1618 WARN_ON(p->nodes[0] != NULL);
1624 if (p->search_commit_root) {
1625 b = root->commit_root;
1626 extent_buffer_get(b);
1627 if (!p->skip_locking)
1630 if (p->skip_locking)
1631 b = btrfs_root_node(root);
1633 b = btrfs_lock_root_node(root);
1637 level = btrfs_header_level(b);
1640 * setup the path here so we can release it under lock
1641 * contention with the cow code
1643 p->nodes[level] = b;
1644 if (!p->skip_locking)
1645 p->locks[level] = 1;
1649 * if we don't really need to cow this block
1650 * then we don't want to set the path blocking,
1651 * so we test it here
1653 if (!should_cow_block(trans, root, b))
1656 btrfs_set_path_blocking(p);
1658 err = btrfs_cow_block(trans, root, b,
1659 p->nodes[level + 1],
1660 p->slots[level + 1], &b);
1667 BUG_ON(!cow && ins_len);
1669 p->nodes[level] = b;
1670 if (!p->skip_locking)
1671 p->locks[level] = 1;
1673 btrfs_clear_path_blocking(p, NULL);
1676 * we have a lock on b and as long as we aren't changing
1677 * the tree, there is no way to for the items in b to change.
1678 * It is safe to drop the lock on our parent before we
1679 * go through the expensive btree search on b.
1681 * If cow is true, then we might be changing slot zero,
1682 * which may require changing the parent. So, we can't
1683 * drop the lock until after we know which slot we're
1687 btrfs_unlock_up_safe(p, level + 1);
1689 ret = bin_search(b, key, level, &slot);
1693 if (ret && slot > 0) {
1697 p->slots[level] = slot;
1698 err = setup_nodes_for_search(trans, root, p, b, level,
1706 b = p->nodes[level];
1707 slot = p->slots[level];
1709 unlock_up(p, level, lowest_unlock);
1711 if (level == lowest_level) {
1717 err = read_block_for_search(trans, root, p,
1718 &b, level, slot, key);
1726 if (!p->skip_locking) {
1727 btrfs_clear_path_blocking(p, NULL);
1728 err = btrfs_try_spin_lock(b);
1731 btrfs_set_path_blocking(p);
1733 btrfs_clear_path_blocking(p, b);
1737 p->slots[level] = slot;
1739 btrfs_leaf_free_space(root, b) < ins_len) {
1740 btrfs_set_path_blocking(p);
1741 err = split_leaf(trans, root, key,
1742 p, ins_len, ret == 0);
1743 btrfs_clear_path_blocking(p, NULL);
1751 if (!p->search_for_split)
1752 unlock_up(p, level, lowest_unlock);
1759 * we don't really know what they plan on doing with the path
1760 * from here on, so for now just mark it as blocking
1762 if (!p->leave_spinning)
1763 btrfs_set_path_blocking(p);
1765 btrfs_release_path(p);
1770 * adjust the pointers going up the tree, starting at level
1771 * making sure the right key of each node is points to 'key'.
1772 * This is used after shifting pointers to the left, so it stops
1773 * fixing up pointers when a given leaf/node is not in slot 0 of the
1776 * If this fails to write a tree block, it returns -1, but continues
1777 * fixing up the blocks in ram so the tree is consistent.
1779 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1780 struct btrfs_root *root, struct btrfs_path *path,
1781 struct btrfs_disk_key *key, int level)
1785 struct extent_buffer *t;
1787 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1788 int tslot = path->slots[i];
1789 if (!path->nodes[i])
1792 btrfs_set_node_key(t, key, tslot);
1793 btrfs_mark_buffer_dirty(path->nodes[i]);
1803 * This function isn't completely safe. It's the caller's responsibility
1804 * that the new key won't break the order
1806 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1807 struct btrfs_root *root, struct btrfs_path *path,
1808 struct btrfs_key *new_key)
1810 struct btrfs_disk_key disk_key;
1811 struct extent_buffer *eb;
1814 eb = path->nodes[0];
1815 slot = path->slots[0];
1817 btrfs_item_key(eb, &disk_key, slot - 1);
1818 if (comp_keys(&disk_key, new_key) >= 0)
1821 if (slot < btrfs_header_nritems(eb) - 1) {
1822 btrfs_item_key(eb, &disk_key, slot + 1);
1823 if (comp_keys(&disk_key, new_key) <= 0)
1827 btrfs_cpu_key_to_disk(&disk_key, new_key);
1828 btrfs_set_item_key(eb, &disk_key, slot);
1829 btrfs_mark_buffer_dirty(eb);
1831 fixup_low_keys(trans, root, path, &disk_key, 1);
1836 * try to push data from one node into the next node left in the
1839 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1840 * error, and > 0 if there was no room in the left hand block.
1842 static int push_node_left(struct btrfs_trans_handle *trans,
1843 struct btrfs_root *root, struct extent_buffer *dst,
1844 struct extent_buffer *src, int empty)
1851 src_nritems = btrfs_header_nritems(src);
1852 dst_nritems = btrfs_header_nritems(dst);
1853 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1854 WARN_ON(btrfs_header_generation(src) != trans->transid);
1855 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1857 if (!empty && src_nritems <= 8)
1860 if (push_items <= 0)
1864 push_items = min(src_nritems, push_items);
1865 if (push_items < src_nritems) {
1866 /* leave at least 8 pointers in the node if
1867 * we aren't going to empty it
1869 if (src_nritems - push_items < 8) {
1870 if (push_items <= 8)
1876 push_items = min(src_nritems - 8, push_items);
1878 copy_extent_buffer(dst, src,
1879 btrfs_node_key_ptr_offset(dst_nritems),
1880 btrfs_node_key_ptr_offset(0),
1881 push_items * sizeof(struct btrfs_key_ptr));
1883 if (push_items < src_nritems) {
1884 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1885 btrfs_node_key_ptr_offset(push_items),
1886 (src_nritems - push_items) *
1887 sizeof(struct btrfs_key_ptr));
1889 btrfs_set_header_nritems(src, src_nritems - push_items);
1890 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1891 btrfs_mark_buffer_dirty(src);
1892 btrfs_mark_buffer_dirty(dst);
1898 * try to push data from one node into the next node right in the
1901 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1902 * error, and > 0 if there was no room in the right hand block.
1904 * this will only push up to 1/2 the contents of the left node over
1906 static int balance_node_right(struct btrfs_trans_handle *trans,
1907 struct btrfs_root *root,
1908 struct extent_buffer *dst,
1909 struct extent_buffer *src)
1917 WARN_ON(btrfs_header_generation(src) != trans->transid);
1918 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1920 src_nritems = btrfs_header_nritems(src);
1921 dst_nritems = btrfs_header_nritems(dst);
1922 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1923 if (push_items <= 0)
1926 if (src_nritems < 4)
1929 max_push = src_nritems / 2 + 1;
1930 /* don't try to empty the node */
1931 if (max_push >= src_nritems)
1934 if (max_push < push_items)
1935 push_items = max_push;
1937 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1938 btrfs_node_key_ptr_offset(0),
1940 sizeof(struct btrfs_key_ptr));
1942 copy_extent_buffer(dst, src,
1943 btrfs_node_key_ptr_offset(0),
1944 btrfs_node_key_ptr_offset(src_nritems - push_items),
1945 push_items * sizeof(struct btrfs_key_ptr));
1947 btrfs_set_header_nritems(src, src_nritems - push_items);
1948 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1950 btrfs_mark_buffer_dirty(src);
1951 btrfs_mark_buffer_dirty(dst);
1957 * helper function to insert a new root level in the tree.
1958 * A new node is allocated, and a single item is inserted to
1959 * point to the existing root
1961 * returns zero on success or < 0 on failure.
1963 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
1964 struct btrfs_root *root,
1965 struct btrfs_path *path, int level)
1968 struct extent_buffer *lower;
1969 struct extent_buffer *c;
1970 struct extent_buffer *old;
1971 struct btrfs_disk_key lower_key;
1973 BUG_ON(path->nodes[level]);
1974 BUG_ON(path->nodes[level-1] != root->node);
1976 lower = path->nodes[level-1];
1978 btrfs_item_key(lower, &lower_key, 0);
1980 btrfs_node_key(lower, &lower_key, 0);
1982 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1983 root->root_key.objectid, &lower_key,
1984 level, root->node->start, 0);
1988 root_add_used(root, root->nodesize);
1990 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
1991 btrfs_set_header_nritems(c, 1);
1992 btrfs_set_header_level(c, level);
1993 btrfs_set_header_bytenr(c, c->start);
1994 btrfs_set_header_generation(c, trans->transid);
1995 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
1996 btrfs_set_header_owner(c, root->root_key.objectid);
1998 write_extent_buffer(c, root->fs_info->fsid,
1999 (unsigned long)btrfs_header_fsid(c),
2002 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2003 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2006 btrfs_set_node_key(c, &lower_key, 0);
2007 btrfs_set_node_blockptr(c, 0, lower->start);
2008 lower_gen = btrfs_header_generation(lower);
2009 WARN_ON(lower_gen != trans->transid);
2011 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2013 btrfs_mark_buffer_dirty(c);
2016 rcu_assign_pointer(root->node, c);
2018 /* the super has an extra ref to root->node */
2019 free_extent_buffer(old);
2021 add_root_to_dirty_list(root);
2022 extent_buffer_get(c);
2023 path->nodes[level] = c;
2024 path->locks[level] = 1;
2025 path->slots[level] = 0;
2030 * worker function to insert a single pointer in a node.
2031 * the node should have enough room for the pointer already
2033 * slot and level indicate where you want the key to go, and
2034 * blocknr is the block the key points to.
2036 * returns zero on success and < 0 on any error
2038 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2039 *root, struct btrfs_path *path, struct btrfs_disk_key
2040 *key, u64 bytenr, int slot, int level)
2042 struct extent_buffer *lower;
2045 BUG_ON(!path->nodes[level]);
2046 btrfs_assert_tree_locked(path->nodes[level]);
2047 lower = path->nodes[level];
2048 nritems = btrfs_header_nritems(lower);
2049 BUG_ON(slot > nritems);
2050 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2052 if (slot != nritems) {
2053 memmove_extent_buffer(lower,
2054 btrfs_node_key_ptr_offset(slot + 1),
2055 btrfs_node_key_ptr_offset(slot),
2056 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2058 btrfs_set_node_key(lower, key, slot);
2059 btrfs_set_node_blockptr(lower, slot, bytenr);
2060 WARN_ON(trans->transid == 0);
2061 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2062 btrfs_set_header_nritems(lower, nritems + 1);
2063 btrfs_mark_buffer_dirty(lower);
2068 * split the node at the specified level in path in two.
2069 * The path is corrected to point to the appropriate node after the split
2071 * Before splitting this tries to make some room in the node by pushing
2072 * left and right, if either one works, it returns right away.
2074 * returns 0 on success and < 0 on failure
2076 static noinline int split_node(struct btrfs_trans_handle *trans,
2077 struct btrfs_root *root,
2078 struct btrfs_path *path, int level)
2080 struct extent_buffer *c;
2081 struct extent_buffer *split;
2082 struct btrfs_disk_key disk_key;
2088 c = path->nodes[level];
2089 WARN_ON(btrfs_header_generation(c) != trans->transid);
2090 if (c == root->node) {
2091 /* trying to split the root, lets make a new one */
2092 ret = insert_new_root(trans, root, path, level + 1);
2096 ret = push_nodes_for_insert(trans, root, path, level);
2097 c = path->nodes[level];
2098 if (!ret && btrfs_header_nritems(c) <
2099 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2105 c_nritems = btrfs_header_nritems(c);
2106 mid = (c_nritems + 1) / 2;
2107 btrfs_node_key(c, &disk_key, mid);
2109 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2110 root->root_key.objectid,
2111 &disk_key, level, c->start, 0);
2113 return PTR_ERR(split);
2115 root_add_used(root, root->nodesize);
2117 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2118 btrfs_set_header_level(split, btrfs_header_level(c));
2119 btrfs_set_header_bytenr(split, split->start);
2120 btrfs_set_header_generation(split, trans->transid);
2121 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2122 btrfs_set_header_owner(split, root->root_key.objectid);
2123 write_extent_buffer(split, root->fs_info->fsid,
2124 (unsigned long)btrfs_header_fsid(split),
2126 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2127 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2131 copy_extent_buffer(split, c,
2132 btrfs_node_key_ptr_offset(0),
2133 btrfs_node_key_ptr_offset(mid),
2134 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2135 btrfs_set_header_nritems(split, c_nritems - mid);
2136 btrfs_set_header_nritems(c, mid);
2139 btrfs_mark_buffer_dirty(c);
2140 btrfs_mark_buffer_dirty(split);
2142 wret = insert_ptr(trans, root, path, &disk_key, split->start,
2143 path->slots[level + 1] + 1,
2148 if (path->slots[level] >= mid) {
2149 path->slots[level] -= mid;
2150 btrfs_tree_unlock(c);
2151 free_extent_buffer(c);
2152 path->nodes[level] = split;
2153 path->slots[level + 1] += 1;
2155 btrfs_tree_unlock(split);
2156 free_extent_buffer(split);
2162 * how many bytes are required to store the items in a leaf. start
2163 * and nr indicate which items in the leaf to check. This totals up the
2164 * space used both by the item structs and the item data
2166 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2169 int nritems = btrfs_header_nritems(l);
2170 int end = min(nritems, start + nr) - 1;
2174 data_len = btrfs_item_end_nr(l, start);
2175 data_len = data_len - btrfs_item_offset_nr(l, end);
2176 data_len += sizeof(struct btrfs_item) * nr;
2177 WARN_ON(data_len < 0);
2182 * The space between the end of the leaf items and
2183 * the start of the leaf data. IOW, how much room
2184 * the leaf has left for both items and data
2186 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2187 struct extent_buffer *leaf)
2189 int nritems = btrfs_header_nritems(leaf);
2191 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2193 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2194 "used %d nritems %d\n",
2195 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2196 leaf_space_used(leaf, 0, nritems), nritems);
2202 * min slot controls the lowest index we're willing to push to the
2203 * right. We'll push up to and including min_slot, but no lower
2205 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2206 struct btrfs_root *root,
2207 struct btrfs_path *path,
2208 int data_size, int empty,
2209 struct extent_buffer *right,
2210 int free_space, u32 left_nritems,
2213 struct extent_buffer *left = path->nodes[0];
2214 struct extent_buffer *upper = path->nodes[1];
2215 struct btrfs_disk_key disk_key;
2220 struct btrfs_item *item;
2229 nr = max_t(u32, 1, min_slot);
2231 if (path->slots[0] >= left_nritems)
2232 push_space += data_size;
2234 slot = path->slots[1];
2235 i = left_nritems - 1;
2237 item = btrfs_item_nr(left, i);
2239 if (!empty && push_items > 0) {
2240 if (path->slots[0] > i)
2242 if (path->slots[0] == i) {
2243 int space = btrfs_leaf_free_space(root, left);
2244 if (space + push_space * 2 > free_space)
2249 if (path->slots[0] == i)
2250 push_space += data_size;
2252 if (!left->map_token) {
2253 map_extent_buffer(left, (unsigned long)item,
2254 sizeof(struct btrfs_item),
2255 &left->map_token, &left->kaddr,
2256 &left->map_start, &left->map_len,
2260 this_item_size = btrfs_item_size(left, item);
2261 if (this_item_size + sizeof(*item) + push_space > free_space)
2265 push_space += this_item_size + sizeof(*item);
2270 if (left->map_token) {
2271 unmap_extent_buffer(left, left->map_token, KM_USER1);
2272 left->map_token = NULL;
2275 if (push_items == 0)
2278 if (!empty && push_items == left_nritems)
2281 /* push left to right */
2282 right_nritems = btrfs_header_nritems(right);
2284 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2285 push_space -= leaf_data_end(root, left);
2287 /* make room in the right data area */
2288 data_end = leaf_data_end(root, right);
2289 memmove_extent_buffer(right,
2290 btrfs_leaf_data(right) + data_end - push_space,
2291 btrfs_leaf_data(right) + data_end,
2292 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2294 /* copy from the left data area */
2295 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2296 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2297 btrfs_leaf_data(left) + leaf_data_end(root, left),
2300 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2301 btrfs_item_nr_offset(0),
2302 right_nritems * sizeof(struct btrfs_item));
2304 /* copy the items from left to right */
2305 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2306 btrfs_item_nr_offset(left_nritems - push_items),
2307 push_items * sizeof(struct btrfs_item));
2309 /* update the item pointers */
2310 right_nritems += push_items;
2311 btrfs_set_header_nritems(right, right_nritems);
2312 push_space = BTRFS_LEAF_DATA_SIZE(root);
2313 for (i = 0; i < right_nritems; i++) {
2314 item = btrfs_item_nr(right, i);
2315 if (!right->map_token) {
2316 map_extent_buffer(right, (unsigned long)item,
2317 sizeof(struct btrfs_item),
2318 &right->map_token, &right->kaddr,
2319 &right->map_start, &right->map_len,
2322 push_space -= btrfs_item_size(right, item);
2323 btrfs_set_item_offset(right, item, push_space);
2326 if (right->map_token) {
2327 unmap_extent_buffer(right, right->map_token, KM_USER1);
2328 right->map_token = NULL;
2330 left_nritems -= push_items;
2331 btrfs_set_header_nritems(left, left_nritems);
2334 btrfs_mark_buffer_dirty(left);
2336 clean_tree_block(trans, root, left);
2338 btrfs_mark_buffer_dirty(right);
2340 btrfs_item_key(right, &disk_key, 0);
2341 btrfs_set_node_key(upper, &disk_key, slot + 1);
2342 btrfs_mark_buffer_dirty(upper);
2344 /* then fixup the leaf pointer in the path */
2345 if (path->slots[0] >= left_nritems) {
2346 path->slots[0] -= left_nritems;
2347 if (btrfs_header_nritems(path->nodes[0]) == 0)
2348 clean_tree_block(trans, root, path->nodes[0]);
2349 btrfs_tree_unlock(path->nodes[0]);
2350 free_extent_buffer(path->nodes[0]);
2351 path->nodes[0] = right;
2352 path->slots[1] += 1;
2354 btrfs_tree_unlock(right);
2355 free_extent_buffer(right);
2360 btrfs_tree_unlock(right);
2361 free_extent_buffer(right);
2366 * push some data in the path leaf to the right, trying to free up at
2367 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2369 * returns 1 if the push failed because the other node didn't have enough
2370 * room, 0 if everything worked out and < 0 if there were major errors.
2372 * this will push starting from min_slot to the end of the leaf. It won't
2373 * push any slot lower than min_slot
2375 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2376 *root, struct btrfs_path *path,
2377 int min_data_size, int data_size,
2378 int empty, u32 min_slot)
2380 struct extent_buffer *left = path->nodes[0];
2381 struct extent_buffer *right;
2382 struct extent_buffer *upper;
2388 if (!path->nodes[1])
2391 slot = path->slots[1];
2392 upper = path->nodes[1];
2393 if (slot >= btrfs_header_nritems(upper) - 1)
2396 btrfs_assert_tree_locked(path->nodes[1]);
2398 right = read_node_slot(root, upper, slot + 1);
2402 btrfs_tree_lock(right);
2403 btrfs_set_lock_blocking(right);
2405 free_space = btrfs_leaf_free_space(root, right);
2406 if (free_space < data_size)
2409 /* cow and double check */
2410 ret = btrfs_cow_block(trans, root, right, upper,
2415 free_space = btrfs_leaf_free_space(root, right);
2416 if (free_space < data_size)
2419 left_nritems = btrfs_header_nritems(left);
2420 if (left_nritems == 0)
2423 return __push_leaf_right(trans, root, path, min_data_size, empty,
2424 right, free_space, left_nritems, min_slot);
2426 btrfs_tree_unlock(right);
2427 free_extent_buffer(right);
2432 * push some data in the path leaf to the left, trying to free up at
2433 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2435 * max_slot can put a limit on how far into the leaf we'll push items. The
2436 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
2439 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2440 struct btrfs_root *root,
2441 struct btrfs_path *path, int data_size,
2442 int empty, struct extent_buffer *left,
2443 int free_space, u32 right_nritems,
2446 struct btrfs_disk_key disk_key;
2447 struct extent_buffer *right = path->nodes[0];
2451 struct btrfs_item *item;
2452 u32 old_left_nritems;
2457 u32 old_left_item_size;
2460 nr = min(right_nritems, max_slot);
2462 nr = min(right_nritems - 1, max_slot);
2464 for (i = 0; i < nr; i++) {
2465 item = btrfs_item_nr(right, i);
2466 if (!right->map_token) {
2467 map_extent_buffer(right, (unsigned long)item,
2468 sizeof(struct btrfs_item),
2469 &right->map_token, &right->kaddr,
2470 &right->map_start, &right->map_len,
2474 if (!empty && push_items > 0) {
2475 if (path->slots[0] < i)
2477 if (path->slots[0] == i) {
2478 int space = btrfs_leaf_free_space(root, right);
2479 if (space + push_space * 2 > free_space)
2484 if (path->slots[0] == i)
2485 push_space += data_size;
2487 this_item_size = btrfs_item_size(right, item);
2488 if (this_item_size + sizeof(*item) + push_space > free_space)
2492 push_space += this_item_size + sizeof(*item);
2495 if (right->map_token) {
2496 unmap_extent_buffer(right, right->map_token, KM_USER1);
2497 right->map_token = NULL;
2500 if (push_items == 0) {
2504 if (!empty && push_items == btrfs_header_nritems(right))
2507 /* push data from right to left */
2508 copy_extent_buffer(left, right,
2509 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2510 btrfs_item_nr_offset(0),
2511 push_items * sizeof(struct btrfs_item));
2513 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2514 btrfs_item_offset_nr(right, push_items - 1);
2516 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2517 leaf_data_end(root, left) - push_space,
2518 btrfs_leaf_data(right) +
2519 btrfs_item_offset_nr(right, push_items - 1),
2521 old_left_nritems = btrfs_header_nritems(left);
2522 BUG_ON(old_left_nritems <= 0);
2524 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2525 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2528 item = btrfs_item_nr(left, i);
2529 if (!left->map_token) {
2530 map_extent_buffer(left, (unsigned long)item,
2531 sizeof(struct btrfs_item),
2532 &left->map_token, &left->kaddr,
2533 &left->map_start, &left->map_len,
2537 ioff = btrfs_item_offset(left, item);
2538 btrfs_set_item_offset(left, item,
2539 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2541 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2542 if (left->map_token) {
2543 unmap_extent_buffer(left, left->map_token, KM_USER1);
2544 left->map_token = NULL;
2547 /* fixup right node */
2548 if (push_items > right_nritems) {
2549 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2554 if (push_items < right_nritems) {
2555 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2556 leaf_data_end(root, right);
2557 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2558 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2559 btrfs_leaf_data(right) +
2560 leaf_data_end(root, right), push_space);
2562 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2563 btrfs_item_nr_offset(push_items),
2564 (btrfs_header_nritems(right) - push_items) *
2565 sizeof(struct btrfs_item));
2567 right_nritems -= push_items;
2568 btrfs_set_header_nritems(right, right_nritems);
2569 push_space = BTRFS_LEAF_DATA_SIZE(root);
2570 for (i = 0; i < right_nritems; i++) {
2571 item = btrfs_item_nr(right, i);
2573 if (!right->map_token) {
2574 map_extent_buffer(right, (unsigned long)item,
2575 sizeof(struct btrfs_item),
2576 &right->map_token, &right->kaddr,
2577 &right->map_start, &right->map_len,
2581 push_space = push_space - btrfs_item_size(right, item);
2582 btrfs_set_item_offset(right, item, push_space);
2584 if (right->map_token) {
2585 unmap_extent_buffer(right, right->map_token, KM_USER1);
2586 right->map_token = NULL;
2589 btrfs_mark_buffer_dirty(left);
2591 btrfs_mark_buffer_dirty(right);
2593 clean_tree_block(trans, root, right);
2595 btrfs_item_key(right, &disk_key, 0);
2596 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2600 /* then fixup the leaf pointer in the path */
2601 if (path->slots[0] < push_items) {
2602 path->slots[0] += old_left_nritems;
2603 btrfs_tree_unlock(path->nodes[0]);
2604 free_extent_buffer(path->nodes[0]);
2605 path->nodes[0] = left;
2606 path->slots[1] -= 1;
2608 btrfs_tree_unlock(left);
2609 free_extent_buffer(left);
2610 path->slots[0] -= push_items;
2612 BUG_ON(path->slots[0] < 0);
2615 btrfs_tree_unlock(left);
2616 free_extent_buffer(left);
2621 * push some data in the path leaf to the left, trying to free up at
2622 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2624 * max_slot can put a limit on how far into the leaf we'll push items. The
2625 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
2628 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2629 *root, struct btrfs_path *path, int min_data_size,
2630 int data_size, int empty, u32 max_slot)
2632 struct extent_buffer *right = path->nodes[0];
2633 struct extent_buffer *left;
2639 slot = path->slots[1];
2642 if (!path->nodes[1])
2645 right_nritems = btrfs_header_nritems(right);
2646 if (right_nritems == 0)
2649 btrfs_assert_tree_locked(path->nodes[1]);
2651 left = read_node_slot(root, path->nodes[1], slot - 1);
2655 btrfs_tree_lock(left);
2656 btrfs_set_lock_blocking(left);
2658 free_space = btrfs_leaf_free_space(root, left);
2659 if (free_space < data_size) {
2664 /* cow and double check */
2665 ret = btrfs_cow_block(trans, root, left,
2666 path->nodes[1], slot - 1, &left);
2668 /* we hit -ENOSPC, but it isn't fatal here */
2673 free_space = btrfs_leaf_free_space(root, left);
2674 if (free_space < data_size) {
2679 return __push_leaf_left(trans, root, path, min_data_size,
2680 empty, left, free_space, right_nritems,
2683 btrfs_tree_unlock(left);
2684 free_extent_buffer(left);
2689 * split the path's leaf in two, making sure there is at least data_size
2690 * available for the resulting leaf level of the path.
2692 * returns 0 if all went well and < 0 on failure.
2694 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2695 struct btrfs_root *root,
2696 struct btrfs_path *path,
2697 struct extent_buffer *l,
2698 struct extent_buffer *right,
2699 int slot, int mid, int nritems)
2706 struct btrfs_disk_key disk_key;
2708 nritems = nritems - mid;
2709 btrfs_set_header_nritems(right, nritems);
2710 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2712 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2713 btrfs_item_nr_offset(mid),
2714 nritems * sizeof(struct btrfs_item));
2716 copy_extent_buffer(right, l,
2717 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2718 data_copy_size, btrfs_leaf_data(l) +
2719 leaf_data_end(root, l), data_copy_size);
2721 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2722 btrfs_item_end_nr(l, mid);
2724 for (i = 0; i < nritems; i++) {
2725 struct btrfs_item *item = btrfs_item_nr(right, i);
2728 if (!right->map_token) {
2729 map_extent_buffer(right, (unsigned long)item,
2730 sizeof(struct btrfs_item),
2731 &right->map_token, &right->kaddr,
2732 &right->map_start, &right->map_len,
2736 ioff = btrfs_item_offset(right, item);
2737 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2740 if (right->map_token) {
2741 unmap_extent_buffer(right, right->map_token, KM_USER1);
2742 right->map_token = NULL;
2745 btrfs_set_header_nritems(l, mid);
2747 btrfs_item_key(right, &disk_key, 0);
2748 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2749 path->slots[1] + 1, 1);
2753 btrfs_mark_buffer_dirty(right);
2754 btrfs_mark_buffer_dirty(l);
2755 BUG_ON(path->slots[0] != slot);
2758 btrfs_tree_unlock(path->nodes[0]);
2759 free_extent_buffer(path->nodes[0]);
2760 path->nodes[0] = right;
2761 path->slots[0] -= mid;
2762 path->slots[1] += 1;
2764 btrfs_tree_unlock(right);
2765 free_extent_buffer(right);
2768 BUG_ON(path->slots[0] < 0);
2774 * double splits happen when we need to insert a big item in the middle
2775 * of a leaf. A double split can leave us with 3 mostly empty leaves:
2776 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
2779 * We avoid this by trying to push the items on either side of our target
2780 * into the adjacent leaves. If all goes well we can avoid the double split
2783 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root,
2785 struct btrfs_path *path,
2793 slot = path->slots[0];
2796 * try to push all the items after our slot into the
2799 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
2806 nritems = btrfs_header_nritems(path->nodes[0]);
2808 * our goal is to get our slot at the start or end of a leaf. If
2809 * we've done so we're done
2811 if (path->slots[0] == 0 || path->slots[0] == nritems)
2814 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
2817 /* try to push all the items before our slot into the next leaf */
2818 slot = path->slots[0];
2819 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
2832 * split the path's leaf in two, making sure there is at least data_size
2833 * available for the resulting leaf level of the path.
2835 * returns 0 if all went well and < 0 on failure.
2837 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2838 struct btrfs_root *root,
2839 struct btrfs_key *ins_key,
2840 struct btrfs_path *path, int data_size,
2843 struct btrfs_disk_key disk_key;
2844 struct extent_buffer *l;
2848 struct extent_buffer *right;
2852 int num_doubles = 0;
2853 int tried_avoid_double = 0;
2856 slot = path->slots[0];
2857 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2858 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2861 /* first try to make some room by pushing left and right */
2863 wret = push_leaf_right(trans, root, path, data_size,
2868 wret = push_leaf_left(trans, root, path, data_size,
2869 data_size, 0, (u32)-1);
2875 /* did the pushes work? */
2876 if (btrfs_leaf_free_space(root, l) >= data_size)
2880 if (!path->nodes[1]) {
2881 ret = insert_new_root(trans, root, path, 1);
2888 slot = path->slots[0];
2889 nritems = btrfs_header_nritems(l);
2890 mid = (nritems + 1) / 2;
2894 leaf_space_used(l, mid, nritems - mid) + data_size >
2895 BTRFS_LEAF_DATA_SIZE(root)) {
2896 if (slot >= nritems) {
2900 if (mid != nritems &&
2901 leaf_space_used(l, mid, nritems - mid) +
2902 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2903 if (data_size && !tried_avoid_double)
2904 goto push_for_double;
2910 if (leaf_space_used(l, 0, mid) + data_size >
2911 BTRFS_LEAF_DATA_SIZE(root)) {
2912 if (!extend && data_size && slot == 0) {
2914 } else if ((extend || !data_size) && slot == 0) {
2918 if (mid != nritems &&
2919 leaf_space_used(l, mid, nritems - mid) +
2920 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2921 if (data_size && !tried_avoid_double)
2922 goto push_for_double;
2930 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2932 btrfs_item_key(l, &disk_key, mid);
2934 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2935 root->root_key.objectid,
2936 &disk_key, 0, l->start, 0);
2938 return PTR_ERR(right);
2940 root_add_used(root, root->leafsize);
2942 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2943 btrfs_set_header_bytenr(right, right->start);
2944 btrfs_set_header_generation(right, trans->transid);
2945 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2946 btrfs_set_header_owner(right, root->root_key.objectid);
2947 btrfs_set_header_level(right, 0);
2948 write_extent_buffer(right, root->fs_info->fsid,
2949 (unsigned long)btrfs_header_fsid(right),
2952 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2953 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2958 btrfs_set_header_nritems(right, 0);
2959 wret = insert_ptr(trans, root, path,
2960 &disk_key, right->start,
2961 path->slots[1] + 1, 1);
2965 btrfs_tree_unlock(path->nodes[0]);
2966 free_extent_buffer(path->nodes[0]);
2967 path->nodes[0] = right;
2969 path->slots[1] += 1;
2971 btrfs_set_header_nritems(right, 0);
2972 wret = insert_ptr(trans, root, path,
2978 btrfs_tree_unlock(path->nodes[0]);
2979 free_extent_buffer(path->nodes[0]);
2980 path->nodes[0] = right;
2982 if (path->slots[1] == 0) {
2983 wret = fixup_low_keys(trans, root,
2984 path, &disk_key, 1);
2989 btrfs_mark_buffer_dirty(right);
2993 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2997 BUG_ON(num_doubles != 0);
3005 push_for_double_split(trans, root, path, data_size);
3006 tried_avoid_double = 1;
3007 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3012 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3013 struct btrfs_root *root,
3014 struct btrfs_path *path, int ins_len)
3016 struct btrfs_key key;
3017 struct extent_buffer *leaf;
3018 struct btrfs_file_extent_item *fi;
3023 leaf = path->nodes[0];
3024 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3026 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3027 key.type != BTRFS_EXTENT_CSUM_KEY);
3029 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3032 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3033 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3034 fi = btrfs_item_ptr(leaf, path->slots[0],
3035 struct btrfs_file_extent_item);
3036 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3038 btrfs_release_path(path);
3040 path->keep_locks = 1;
3041 path->search_for_split = 1;
3042 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3043 path->search_for_split = 0;
3048 leaf = path->nodes[0];
3049 /* if our item isn't there or got smaller, return now */
3050 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3053 /* the leaf has changed, it now has room. return now */
3054 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3057 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3058 fi = btrfs_item_ptr(leaf, path->slots[0],
3059 struct btrfs_file_extent_item);
3060 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3064 btrfs_set_path_blocking(path);
3065 ret = split_leaf(trans, root, &key, path, ins_len, 1);
3069 path->keep_locks = 0;
3070 btrfs_unlock_up_safe(path, 1);
3073 path->keep_locks = 0;
3077 static noinline int split_item(struct btrfs_trans_handle *trans,
3078 struct btrfs_root *root,
3079 struct btrfs_path *path,
3080 struct btrfs_key *new_key,
3081 unsigned long split_offset)
3083 struct extent_buffer *leaf;
3084 struct btrfs_item *item;
3085 struct btrfs_item *new_item;
3091 struct btrfs_disk_key disk_key;
3093 leaf = path->nodes[0];
3094 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3096 btrfs_set_path_blocking(path);
3098 item = btrfs_item_nr(leaf, path->slots[0]);
3099 orig_offset = btrfs_item_offset(leaf, item);
3100 item_size = btrfs_item_size(leaf, item);
3102 buf = kmalloc(item_size, GFP_NOFS);
3106 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3107 path->slots[0]), item_size);
3109 slot = path->slots[0] + 1;
3110 nritems = btrfs_header_nritems(leaf);
3111 if (slot != nritems) {
3112 /* shift the items */
3113 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3114 btrfs_item_nr_offset(slot),
3115 (nritems - slot) * sizeof(struct btrfs_item));
3118 btrfs_cpu_key_to_disk(&disk_key, new_key);
3119 btrfs_set_item_key(leaf, &disk_key, slot);
3121 new_item = btrfs_item_nr(leaf, slot);
3123 btrfs_set_item_offset(leaf, new_item, orig_offset);
3124 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3126 btrfs_set_item_offset(leaf, item,
3127 orig_offset + item_size - split_offset);
3128 btrfs_set_item_size(leaf, item, split_offset);
3130 btrfs_set_header_nritems(leaf, nritems + 1);
3132 /* write the data for the start of the original item */
3133 write_extent_buffer(leaf, buf,
3134 btrfs_item_ptr_offset(leaf, path->slots[0]),
3137 /* write the data for the new item */
3138 write_extent_buffer(leaf, buf + split_offset,
3139 btrfs_item_ptr_offset(leaf, slot),
3140 item_size - split_offset);
3141 btrfs_mark_buffer_dirty(leaf);
3143 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
3149 * This function splits a single item into two items,
3150 * giving 'new_key' to the new item and splitting the
3151 * old one at split_offset (from the start of the item).
3153 * The path may be released by this operation. After
3154 * the split, the path is pointing to the old item. The
3155 * new item is going to be in the same node as the old one.
3157 * Note, the item being split must be smaller enough to live alone on
3158 * a tree block with room for one extra struct btrfs_item
3160 * This allows us to split the item in place, keeping a lock on the
3161 * leaf the entire time.
3163 int btrfs_split_item(struct btrfs_trans_handle *trans,
3164 struct btrfs_root *root,
3165 struct btrfs_path *path,
3166 struct btrfs_key *new_key,
3167 unsigned long split_offset)
3170 ret = setup_leaf_for_split(trans, root, path,
3171 sizeof(struct btrfs_item));
3175 ret = split_item(trans, root, path, new_key, split_offset);
3180 * This function duplicate a item, giving 'new_key' to the new item.
3181 * It guarantees both items live in the same tree leaf and the new item
3182 * is contiguous with the original item.
3184 * This allows us to split file extent in place, keeping a lock on the
3185 * leaf the entire time.
3187 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3188 struct btrfs_root *root,
3189 struct btrfs_path *path,
3190 struct btrfs_key *new_key)
3192 struct extent_buffer *leaf;
3196 leaf = path->nodes[0];
3197 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3198 ret = setup_leaf_for_split(trans, root, path,
3199 item_size + sizeof(struct btrfs_item));
3204 ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3205 item_size, item_size +
3206 sizeof(struct btrfs_item), 1);
3209 leaf = path->nodes[0];
3210 memcpy_extent_buffer(leaf,
3211 btrfs_item_ptr_offset(leaf, path->slots[0]),
3212 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3218 * make the item pointed to by the path smaller. new_size indicates
3219 * how small to make it, and from_end tells us if we just chop bytes
3220 * off the end of the item or if we shift the item to chop bytes off
3223 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3224 struct btrfs_root *root,
3225 struct btrfs_path *path,
3226 u32 new_size, int from_end)
3229 struct extent_buffer *leaf;
3230 struct btrfs_item *item;
3232 unsigned int data_end;
3233 unsigned int old_data_start;
3234 unsigned int old_size;
3235 unsigned int size_diff;
3238 leaf = path->nodes[0];
3239 slot = path->slots[0];
3241 old_size = btrfs_item_size_nr(leaf, slot);
3242 if (old_size == new_size)
3245 nritems = btrfs_header_nritems(leaf);
3246 data_end = leaf_data_end(root, leaf);
3248 old_data_start = btrfs_item_offset_nr(leaf, slot);
3250 size_diff = old_size - new_size;
3253 BUG_ON(slot >= nritems);
3256 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3258 /* first correct the data pointers */
3259 for (i = slot; i < nritems; i++) {
3261 item = btrfs_item_nr(leaf, i);
3263 if (!leaf->map_token) {
3264 map_extent_buffer(leaf, (unsigned long)item,
3265 sizeof(struct btrfs_item),
3266 &leaf->map_token, &leaf->kaddr,
3267 &leaf->map_start, &leaf->map_len,
3271 ioff = btrfs_item_offset(leaf, item);
3272 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3275 if (leaf->map_token) {
3276 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3277 leaf->map_token = NULL;
3280 /* shift the data */
3282 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3283 data_end + size_diff, btrfs_leaf_data(leaf) +
3284 data_end, old_data_start + new_size - data_end);
3286 struct btrfs_disk_key disk_key;
3289 btrfs_item_key(leaf, &disk_key, slot);
3291 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3293 struct btrfs_file_extent_item *fi;
3295 fi = btrfs_item_ptr(leaf, slot,
3296 struct btrfs_file_extent_item);
3297 fi = (struct btrfs_file_extent_item *)(
3298 (unsigned long)fi - size_diff);
3300 if (btrfs_file_extent_type(leaf, fi) ==
3301 BTRFS_FILE_EXTENT_INLINE) {
3302 ptr = btrfs_item_ptr_offset(leaf, slot);
3303 memmove_extent_buffer(leaf, ptr,
3305 offsetof(struct btrfs_file_extent_item,
3310 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3311 data_end + size_diff, btrfs_leaf_data(leaf) +
3312 data_end, old_data_start - data_end);
3314 offset = btrfs_disk_key_offset(&disk_key);
3315 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3316 btrfs_set_item_key(leaf, &disk_key, slot);
3318 fixup_low_keys(trans, root, path, &disk_key, 1);
3321 item = btrfs_item_nr(leaf, slot);
3322 btrfs_set_item_size(leaf, item, new_size);
3323 btrfs_mark_buffer_dirty(leaf);
3325 if (btrfs_leaf_free_space(root, leaf) < 0) {
3326 btrfs_print_leaf(root, leaf);
3333 * make the item pointed to by the path bigger, data_size is the new size.
3335 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3336 struct btrfs_root *root, struct btrfs_path *path,
3340 struct extent_buffer *leaf;
3341 struct btrfs_item *item;
3343 unsigned int data_end;
3344 unsigned int old_data;
3345 unsigned int old_size;
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) < data_size) {
3354 btrfs_print_leaf(root, leaf);
3357 slot = path->slots[0];
3358 old_data = btrfs_item_end_nr(leaf, slot);
3361 if (slot >= nritems) {
3362 btrfs_print_leaf(root, leaf);
3363 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3369 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3371 /* first correct the data pointers */
3372 for (i = slot; i < nritems; i++) {
3374 item = btrfs_item_nr(leaf, i);
3376 if (!leaf->map_token) {
3377 map_extent_buffer(leaf, (unsigned long)item,
3378 sizeof(struct btrfs_item),
3379 &leaf->map_token, &leaf->kaddr,
3380 &leaf->map_start, &leaf->map_len,
3383 ioff = btrfs_item_offset(leaf, item);
3384 btrfs_set_item_offset(leaf, item, ioff - data_size);
3387 if (leaf->map_token) {
3388 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3389 leaf->map_token = NULL;
3392 /* shift the data */
3393 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3394 data_end - data_size, btrfs_leaf_data(leaf) +
3395 data_end, old_data - data_end);
3397 data_end = old_data;
3398 old_size = btrfs_item_size_nr(leaf, slot);
3399 item = btrfs_item_nr(leaf, slot);
3400 btrfs_set_item_size(leaf, item, old_size + data_size);
3401 btrfs_mark_buffer_dirty(leaf);
3403 if (btrfs_leaf_free_space(root, leaf) < 0) {
3404 btrfs_print_leaf(root, leaf);
3411 * Given a key and some data, insert items into the tree.
3412 * This does all the path init required, making room in the tree if needed.
3413 * Returns the number of keys that were inserted.
3415 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3416 struct btrfs_root *root,
3417 struct btrfs_path *path,
3418 struct btrfs_key *cpu_key, u32 *data_size,
3421 struct extent_buffer *leaf;
3422 struct btrfs_item *item;
3429 unsigned int data_end;
3430 struct btrfs_disk_key disk_key;
3431 struct btrfs_key found_key;
3433 for (i = 0; i < nr; i++) {
3434 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3435 BTRFS_LEAF_DATA_SIZE(root)) {
3439 total_data += data_size[i];
3440 total_size += data_size[i] + sizeof(struct btrfs_item);
3444 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3450 leaf = path->nodes[0];
3452 nritems = btrfs_header_nritems(leaf);
3453 data_end = leaf_data_end(root, leaf);
3455 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3456 for (i = nr; i >= 0; i--) {
3457 total_data -= data_size[i];
3458 total_size -= data_size[i] + sizeof(struct btrfs_item);
3459 if (total_size < btrfs_leaf_free_space(root, leaf))
3465 slot = path->slots[0];
3468 if (slot != nritems) {
3469 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3471 item = btrfs_item_nr(leaf, slot);
3472 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3474 /* figure out how many keys we can insert in here */
3475 total_data = data_size[0];
3476 for (i = 1; i < nr; i++) {
3477 if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3479 total_data += data_size[i];
3483 if (old_data < data_end) {
3484 btrfs_print_leaf(root, leaf);
3485 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3486 slot, old_data, data_end);
3490 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3492 /* first correct the data pointers */
3493 WARN_ON(leaf->map_token);
3494 for (i = slot; i < nritems; i++) {
3497 item = btrfs_item_nr(leaf, i);
3498 if (!leaf->map_token) {
3499 map_extent_buffer(leaf, (unsigned long)item,
3500 sizeof(struct btrfs_item),
3501 &leaf->map_token, &leaf->kaddr,
3502 &leaf->map_start, &leaf->map_len,
3506 ioff = btrfs_item_offset(leaf, item);
3507 btrfs_set_item_offset(leaf, item, ioff - total_data);
3509 if (leaf->map_token) {
3510 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3511 leaf->map_token = NULL;
3514 /* shift the items */
3515 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3516 btrfs_item_nr_offset(slot),
3517 (nritems - slot) * sizeof(struct btrfs_item));
3519 /* shift the data */
3520 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3521 data_end - total_data, btrfs_leaf_data(leaf) +
3522 data_end, old_data - data_end);
3523 data_end = old_data;
3526 * this sucks but it has to be done, if we are inserting at
3527 * the end of the leaf only insert 1 of the items, since we
3528 * have no way of knowing whats on the next leaf and we'd have
3529 * to drop our current locks to figure it out
3534 /* setup the item for the new data */
3535 for (i = 0; i < nr; i++) {
3536 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3537 btrfs_set_item_key(leaf, &disk_key, slot + i);
3538 item = btrfs_item_nr(leaf, slot + i);
3539 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3540 data_end -= data_size[i];
3541 btrfs_set_item_size(leaf, item, data_size[i]);
3543 btrfs_set_header_nritems(leaf, nritems + nr);
3544 btrfs_mark_buffer_dirty(leaf);
3548 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3549 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3552 if (btrfs_leaf_free_space(root, leaf) < 0) {
3553 btrfs_print_leaf(root, leaf);
3563 * this is a helper for btrfs_insert_empty_items, the main goal here is
3564 * to save stack depth by doing the bulk of the work in a function
3565 * that doesn't call btrfs_search_slot
3567 int setup_items_for_insert(struct btrfs_trans_handle *trans,
3568 struct btrfs_root *root, struct btrfs_path *path,
3569 struct btrfs_key *cpu_key, u32 *data_size,
3570 u32 total_data, u32 total_size, int nr)
3572 struct btrfs_item *item;
3575 unsigned int data_end;
3576 struct btrfs_disk_key disk_key;
3578 struct extent_buffer *leaf;
3581 leaf = path->nodes[0];
3582 slot = path->slots[0];
3584 nritems = btrfs_header_nritems(leaf);
3585 data_end = leaf_data_end(root, leaf);
3587 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3588 btrfs_print_leaf(root, leaf);
3589 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3590 total_size, btrfs_leaf_free_space(root, leaf));
3594 if (slot != nritems) {
3595 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3597 if (old_data < data_end) {
3598 btrfs_print_leaf(root, leaf);
3599 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3600 slot, old_data, data_end);
3604 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3606 /* first correct the data pointers */
3607 WARN_ON(leaf->map_token);
3608 for (i = slot; i < nritems; i++) {
3611 item = btrfs_item_nr(leaf, i);
3612 if (!leaf->map_token) {
3613 map_extent_buffer(leaf, (unsigned long)item,
3614 sizeof(struct btrfs_item),
3615 &leaf->map_token, &leaf->kaddr,
3616 &leaf->map_start, &leaf->map_len,
3620 ioff = btrfs_item_offset(leaf, item);
3621 btrfs_set_item_offset(leaf, item, ioff - total_data);
3623 if (leaf->map_token) {
3624 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3625 leaf->map_token = NULL;
3628 /* shift the items */
3629 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3630 btrfs_item_nr_offset(slot),
3631 (nritems - slot) * sizeof(struct btrfs_item));
3633 /* shift the data */
3634 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3635 data_end - total_data, btrfs_leaf_data(leaf) +
3636 data_end, old_data - data_end);
3637 data_end = old_data;
3640 /* setup the item for the new data */
3641 for (i = 0; i < nr; i++) {
3642 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3643 btrfs_set_item_key(leaf, &disk_key, slot + i);
3644 item = btrfs_item_nr(leaf, slot + i);
3645 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3646 data_end -= data_size[i];
3647 btrfs_set_item_size(leaf, item, data_size[i]);
3650 btrfs_set_header_nritems(leaf, nritems + nr);
3654 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3655 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3657 btrfs_unlock_up_safe(path, 1);
3658 btrfs_mark_buffer_dirty(leaf);
3660 if (btrfs_leaf_free_space(root, leaf) < 0) {
3661 btrfs_print_leaf(root, leaf);
3668 * Given a key and some data, insert items into the tree.
3669 * This does all the path init required, making room in the tree if needed.
3671 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3672 struct btrfs_root *root,
3673 struct btrfs_path *path,
3674 struct btrfs_key *cpu_key, u32 *data_size,
3683 for (i = 0; i < nr; i++)
3684 total_data += data_size[i];
3686 total_size = total_data + (nr * sizeof(struct btrfs_item));
3687 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3693 slot = path->slots[0];
3696 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3697 total_data, total_size, nr);
3704 * Given a key and some data, insert an item into the tree.
3705 * This does all the path init required, making room in the tree if needed.
3707 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3708 *root, struct btrfs_key *cpu_key, void *data, u32
3712 struct btrfs_path *path;
3713 struct extent_buffer *leaf;
3716 path = btrfs_alloc_path();
3719 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3721 leaf = path->nodes[0];
3722 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3723 write_extent_buffer(leaf, data, ptr, data_size);
3724 btrfs_mark_buffer_dirty(leaf);
3726 btrfs_free_path(path);
3731 * delete the pointer from a given node.
3733 * the tree should have been previously balanced so the deletion does not
3736 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3737 struct btrfs_path *path, int level, int slot)
3739 struct extent_buffer *parent = path->nodes[level];
3744 nritems = btrfs_header_nritems(parent);
3745 if (slot != nritems - 1) {
3746 memmove_extent_buffer(parent,
3747 btrfs_node_key_ptr_offset(slot),
3748 btrfs_node_key_ptr_offset(slot + 1),
3749 sizeof(struct btrfs_key_ptr) *
3750 (nritems - slot - 1));
3753 btrfs_set_header_nritems(parent, nritems);
3754 if (nritems == 0 && parent == root->node) {
3755 BUG_ON(btrfs_header_level(root->node) != 1);
3756 /* just turn the root into a leaf and break */
3757 btrfs_set_header_level(root->node, 0);
3758 } else if (slot == 0) {
3759 struct btrfs_disk_key disk_key;
3761 btrfs_node_key(parent, &disk_key, 0);
3762 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3766 btrfs_mark_buffer_dirty(parent);
3771 * a helper function to delete the leaf pointed to by path->slots[1] and
3774 * This deletes the pointer in path->nodes[1] and frees the leaf
3775 * block extent. zero is returned if it all worked out, < 0 otherwise.
3777 * The path must have already been setup for deleting the leaf, including
3778 * all the proper balancing. path->nodes[1] must be locked.
3780 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3781 struct btrfs_root *root,
3782 struct btrfs_path *path,
3783 struct extent_buffer *leaf)
3787 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3788 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3793 * btrfs_free_extent is expensive, we want to make sure we
3794 * aren't holding any locks when we call it
3796 btrfs_unlock_up_safe(path, 0);
3798 root_sub_used(root, leaf->len);
3800 btrfs_free_tree_block(trans, root, leaf, 0, 1);
3804 * delete the item at the leaf level in path. If that empties
3805 * the leaf, remove it from the tree
3807 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3808 struct btrfs_path *path, int slot, int nr)
3810 struct extent_buffer *leaf;
3811 struct btrfs_item *item;
3819 leaf = path->nodes[0];
3820 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3822 for (i = 0; i < nr; i++)
3823 dsize += btrfs_item_size_nr(leaf, slot + i);
3825 nritems = btrfs_header_nritems(leaf);
3827 if (slot + nr != nritems) {
3828 int data_end = leaf_data_end(root, leaf);
3830 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3832 btrfs_leaf_data(leaf) + data_end,
3833 last_off - data_end);
3835 for (i = slot + nr; i < nritems; i++) {
3838 item = btrfs_item_nr(leaf, i);
3839 if (!leaf->map_token) {
3840 map_extent_buffer(leaf, (unsigned long)item,
3841 sizeof(struct btrfs_item),
3842 &leaf->map_token, &leaf->kaddr,
3843 &leaf->map_start, &leaf->map_len,
3846 ioff = btrfs_item_offset(leaf, item);
3847 btrfs_set_item_offset(leaf, item, ioff + dsize);
3850 if (leaf->map_token) {
3851 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3852 leaf->map_token = NULL;
3855 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3856 btrfs_item_nr_offset(slot + nr),
3857 sizeof(struct btrfs_item) *
3858 (nritems - slot - nr));
3860 btrfs_set_header_nritems(leaf, nritems - nr);
3863 /* delete the leaf if we've emptied it */
3865 if (leaf == root->node) {
3866 btrfs_set_header_level(leaf, 0);
3868 btrfs_set_path_blocking(path);
3869 clean_tree_block(trans, root, leaf);
3870 ret = btrfs_del_leaf(trans, root, path, leaf);
3874 int used = leaf_space_used(leaf, 0, nritems);
3876 struct btrfs_disk_key disk_key;
3878 btrfs_item_key(leaf, &disk_key, 0);
3879 wret = fixup_low_keys(trans, root, path,
3885 /* delete the leaf if it is mostly empty */
3886 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3887 /* push_leaf_left fixes the path.
3888 * make sure the path still points to our leaf
3889 * for possible call to del_ptr below
3891 slot = path->slots[1];
3892 extent_buffer_get(leaf);
3894 btrfs_set_path_blocking(path);
3895 wret = push_leaf_left(trans, root, path, 1, 1,
3897 if (wret < 0 && wret != -ENOSPC)
3900 if (path->nodes[0] == leaf &&
3901 btrfs_header_nritems(leaf)) {
3902 wret = push_leaf_right(trans, root, path, 1,
3904 if (wret < 0 && wret != -ENOSPC)
3908 if (btrfs_header_nritems(leaf) == 0) {
3909 path->slots[1] = slot;
3910 ret = btrfs_del_leaf(trans, root, path, leaf);
3912 free_extent_buffer(leaf);
3914 /* if we're still in the path, make sure
3915 * we're dirty. Otherwise, one of the
3916 * push_leaf functions must have already
3917 * dirtied this buffer
3919 if (path->nodes[0] == leaf)
3920 btrfs_mark_buffer_dirty(leaf);
3921 free_extent_buffer(leaf);
3924 btrfs_mark_buffer_dirty(leaf);
3931 * search the tree again to find a leaf with lesser keys
3932 * returns 0 if it found something or 1 if there are no lesser leaves.
3933 * returns < 0 on io errors.
3935 * This may release the path, and so you may lose any locks held at the
3938 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3940 struct btrfs_key key;
3941 struct btrfs_disk_key found_key;
3944 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3948 else if (key.type > 0)
3950 else if (key.objectid > 0)
3955 btrfs_release_path(path);
3956 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3959 btrfs_item_key(path->nodes[0], &found_key, 0);
3960 ret = comp_keys(&found_key, &key);
3967 * A helper function to walk down the tree starting at min_key, and looking
3968 * for nodes or leaves that are either in cache or have a minimum
3969 * transaction id. This is used by the btree defrag code, and tree logging
3971 * This does not cow, but it does stuff the starting key it finds back
3972 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3973 * key and get a writable path.
3975 * This does lock as it descends, and path->keep_locks should be set
3976 * to 1 by the caller.
3978 * This honors path->lowest_level to prevent descent past a given level
3981 * min_trans indicates the oldest transaction that you are interested
3982 * in walking through. Any nodes or leaves older than min_trans are
3983 * skipped over (without reading them).
3985 * returns zero if something useful was found, < 0 on error and 1 if there
3986 * was nothing in the tree that matched the search criteria.
3988 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3989 struct btrfs_key *max_key,
3990 struct btrfs_path *path, int cache_only,
3993 struct extent_buffer *cur;
3994 struct btrfs_key found_key;
4001 WARN_ON(!path->keep_locks);
4003 cur = btrfs_lock_root_node(root);
4004 level = btrfs_header_level(cur);
4005 WARN_ON(path->nodes[level]);
4006 path->nodes[level] = cur;
4007 path->locks[level] = 1;
4009 if (btrfs_header_generation(cur) < min_trans) {
4014 nritems = btrfs_header_nritems(cur);
4015 level = btrfs_header_level(cur);
4016 sret = bin_search(cur, min_key, level, &slot);
4018 /* at the lowest level, we're done, setup the path and exit */
4019 if (level == path->lowest_level) {
4020 if (slot >= nritems)
4023 path->slots[level] = slot;
4024 btrfs_item_key_to_cpu(cur, &found_key, slot);
4027 if (sret && slot > 0)
4030 * check this node pointer against the cache_only and
4031 * min_trans parameters. If it isn't in cache or is too
4032 * old, skip to the next one.
4034 while (slot < nritems) {
4037 struct extent_buffer *tmp;
4038 struct btrfs_disk_key disk_key;
4040 blockptr = btrfs_node_blockptr(cur, slot);
4041 gen = btrfs_node_ptr_generation(cur, slot);
4042 if (gen < min_trans) {
4050 btrfs_node_key(cur, &disk_key, slot);
4051 if (comp_keys(&disk_key, max_key) >= 0) {
4057 tmp = btrfs_find_tree_block(root, blockptr,
4058 btrfs_level_size(root, level - 1));
4060 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4061 free_extent_buffer(tmp);
4065 free_extent_buffer(tmp);
4070 * we didn't find a candidate key in this node, walk forward
4071 * and find another one
4073 if (slot >= nritems) {
4074 path->slots[level] = slot;
4075 btrfs_set_path_blocking(path);
4076 sret = btrfs_find_next_key(root, path, min_key, level,
4077 cache_only, min_trans);
4079 btrfs_release_path(path);
4085 /* save our key for returning back */
4086 btrfs_node_key_to_cpu(cur, &found_key, slot);
4087 path->slots[level] = slot;
4088 if (level == path->lowest_level) {
4090 unlock_up(path, level, 1);
4093 btrfs_set_path_blocking(path);
4094 cur = read_node_slot(root, cur, slot);
4097 btrfs_tree_lock(cur);
4099 path->locks[level - 1] = 1;
4100 path->nodes[level - 1] = cur;
4101 unlock_up(path, level, 1);
4102 btrfs_clear_path_blocking(path, NULL);
4106 memcpy(min_key, &found_key, sizeof(found_key));
4107 btrfs_set_path_blocking(path);
4112 * this is similar to btrfs_next_leaf, but does not try to preserve
4113 * and fixup the path. It looks for and returns the next key in the
4114 * tree based on the current path and the cache_only and min_trans
4117 * 0 is returned if another key is found, < 0 if there are any errors
4118 * and 1 is returned if there are no higher keys in the tree
4120 * path->keep_locks should be set to 1 on the search made before
4121 * calling this function.
4123 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4124 struct btrfs_key *key, int level,
4125 int cache_only, u64 min_trans)
4128 struct extent_buffer *c;
4130 WARN_ON(!path->keep_locks);
4131 while (level < BTRFS_MAX_LEVEL) {
4132 if (!path->nodes[level])
4135 slot = path->slots[level] + 1;
4136 c = path->nodes[level];
4138 if (slot >= btrfs_header_nritems(c)) {
4141 struct btrfs_key cur_key;
4142 if (level + 1 >= BTRFS_MAX_LEVEL ||
4143 !path->nodes[level + 1])
4146 if (path->locks[level + 1]) {
4151 slot = btrfs_header_nritems(c) - 1;
4153 btrfs_item_key_to_cpu(c, &cur_key, slot);
4155 btrfs_node_key_to_cpu(c, &cur_key, slot);
4157 orig_lowest = path->lowest_level;
4158 btrfs_release_path(path);
4159 path->lowest_level = level;
4160 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4162 path->lowest_level = orig_lowest;
4166 c = path->nodes[level];
4167 slot = path->slots[level];
4174 btrfs_item_key_to_cpu(c, key, slot);
4176 u64 blockptr = btrfs_node_blockptr(c, slot);
4177 u64 gen = btrfs_node_ptr_generation(c, slot);
4180 struct extent_buffer *cur;
4181 cur = btrfs_find_tree_block(root, blockptr,
4182 btrfs_level_size(root, level - 1));
4183 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4186 free_extent_buffer(cur);
4189 free_extent_buffer(cur);
4191 if (gen < min_trans) {
4195 btrfs_node_key_to_cpu(c, key, slot);
4203 * search the tree again to find a leaf with greater keys
4204 * returns 0 if it found something or 1 if there are no greater leaves.
4205 * returns < 0 on io errors.
4207 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4211 struct extent_buffer *c;
4212 struct extent_buffer *next;
4213 struct btrfs_key key;
4216 int old_spinning = path->leave_spinning;
4217 int force_blocking = 0;
4219 nritems = btrfs_header_nritems(path->nodes[0]);
4224 * we take the blocks in an order that upsets lockdep. Using
4225 * blocking mode is the only way around it.
4227 #ifdef CONFIG_DEBUG_LOCK_ALLOC
4231 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4235 btrfs_release_path(path);
4237 path->keep_locks = 1;
4239 if (!force_blocking)
4240 path->leave_spinning = 1;
4242 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4243 path->keep_locks = 0;
4248 nritems = btrfs_header_nritems(path->nodes[0]);
4250 * by releasing the path above we dropped all our locks. A balance
4251 * could have added more items next to the key that used to be
4252 * at the very end of the block. So, check again here and
4253 * advance the path if there are now more items available.
4255 if (nritems > 0 && path->slots[0] < nritems - 1) {
4262 while (level < BTRFS_MAX_LEVEL) {
4263 if (!path->nodes[level]) {
4268 slot = path->slots[level] + 1;
4269 c = path->nodes[level];
4270 if (slot >= btrfs_header_nritems(c)) {
4272 if (level == BTRFS_MAX_LEVEL) {
4280 btrfs_tree_unlock(next);
4281 free_extent_buffer(next);
4285 ret = read_block_for_search(NULL, root, path, &next, level,
4291 btrfs_release_path(path);
4295 if (!path->skip_locking) {
4296 ret = btrfs_try_spin_lock(next);
4298 btrfs_set_path_blocking(path);
4299 btrfs_tree_lock(next);
4300 if (!force_blocking)
4301 btrfs_clear_path_blocking(path, next);
4304 btrfs_set_lock_blocking(next);
4308 path->slots[level] = slot;
4311 c = path->nodes[level];
4312 if (path->locks[level])
4313 btrfs_tree_unlock(c);
4315 free_extent_buffer(c);
4316 path->nodes[level] = next;
4317 path->slots[level] = 0;
4318 if (!path->skip_locking)
4319 path->locks[level] = 1;
4324 ret = read_block_for_search(NULL, root, path, &next, level,
4330 btrfs_release_path(path);
4334 if (!path->skip_locking) {
4335 btrfs_assert_tree_locked(path->nodes[level]);
4336 ret = btrfs_try_spin_lock(next);
4338 btrfs_set_path_blocking(path);
4339 btrfs_tree_lock(next);
4340 if (!force_blocking)
4341 btrfs_clear_path_blocking(path, next);
4344 btrfs_set_lock_blocking(next);
4349 unlock_up(path, 0, 1);
4350 path->leave_spinning = old_spinning;
4352 btrfs_set_path_blocking(path);
4358 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4359 * searching until it gets past min_objectid or finds an item of 'type'
4361 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4363 int btrfs_previous_item(struct btrfs_root *root,
4364 struct btrfs_path *path, u64 min_objectid,
4367 struct btrfs_key found_key;
4368 struct extent_buffer *leaf;
4373 if (path->slots[0] == 0) {
4374 btrfs_set_path_blocking(path);
4375 ret = btrfs_prev_leaf(root, path);
4381 leaf = path->nodes[0];
4382 nritems = btrfs_header_nritems(leaf);
4385 if (path->slots[0] == nritems)
4388 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4389 if (found_key.objectid < min_objectid)
4391 if (found_key.type == type)
4393 if (found_key.objectid == min_objectid &&
4394 found_key.type < type)