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 void 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])
59 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
60 if (p->locks[i] == BTRFS_READ_LOCK)
61 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
62 else if (p->locks[i] == BTRFS_WRITE_LOCK)
63 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
68 * reset all the locked nodes in the patch to spinning locks.
70 * held is used to keep lockdep happy, when lockdep is enabled
71 * we set held to a blocking lock before we go around and
72 * retake all the spinlocks in the path. You can safely use NULL
75 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
76 struct extent_buffer *held, int held_rw)
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 /* lockdep really cares that we take all of these spinlocks
82 * in the right order. If any of the locks in the path are not
83 * currently blocking, it is going to complain. So, make really
84 * really sure by forcing the path to blocking before we clear
88 btrfs_set_lock_blocking_rw(held, held_rw);
89 if (held_rw == BTRFS_WRITE_LOCK)
90 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
91 else if (held_rw == BTRFS_READ_LOCK)
92 held_rw = BTRFS_READ_LOCK_BLOCKING;
94 btrfs_set_path_blocking(p);
97 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
98 if (p->nodes[i] && p->locks[i]) {
99 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
100 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
101 p->locks[i] = BTRFS_WRITE_LOCK;
102 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
103 p->locks[i] = BTRFS_READ_LOCK;
107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
109 btrfs_clear_lock_blocking_rw(held, held_rw);
113 /* this also releases the path */
114 void btrfs_free_path(struct btrfs_path *p)
118 btrfs_release_path(p);
119 kmem_cache_free(btrfs_path_cachep, p);
123 * path release drops references on the extent buffers in the path
124 * and it drops any locks held by this path
126 * It is safe to call this on paths that no locks or extent buffers held.
128 noinline void btrfs_release_path(struct btrfs_path *p)
132 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
137 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
140 free_extent_buffer(p->nodes[i]);
146 * safely gets a reference on the root node of a tree. A lock
147 * is not taken, so a concurrent writer may put a different node
148 * at the root of the tree. See btrfs_lock_root_node for the
151 * The extent buffer returned by this has a reference taken, so
152 * it won't disappear. It may stop being the root of the tree
153 * at any time because there are no locks held.
155 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
157 struct extent_buffer *eb;
161 eb = rcu_dereference(root->node);
164 * RCU really hurts here, we could free up the root node because
165 * it was cow'ed but we may not get the new root node yet so do
166 * the inc_not_zero dance and if it doesn't work then
167 * synchronize_rcu and try again.
169 if (atomic_inc_not_zero(&eb->refs)) {
179 /* loop around taking references on and locking the root node of the
180 * tree until you end up with a lock on the root. A locked buffer
181 * is returned, with a reference held.
183 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
185 struct extent_buffer *eb;
188 eb = btrfs_root_node(root);
190 if (eb == root->node)
192 btrfs_tree_unlock(eb);
193 free_extent_buffer(eb);
198 /* loop around taking references on and locking the root node of the
199 * tree until you end up with a lock on the root. A locked buffer
200 * is returned, with a reference held.
202 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
204 struct extent_buffer *eb;
207 eb = btrfs_root_node(root);
208 btrfs_tree_read_lock(eb);
209 if (eb == root->node)
211 btrfs_tree_read_unlock(eb);
212 free_extent_buffer(eb);
217 /* cowonly root (everything not a reference counted cow subvolume), just get
218 * put onto a simple dirty list. transaction.c walks this to make sure they
219 * get properly updated on disk.
221 static void add_root_to_dirty_list(struct btrfs_root *root)
223 spin_lock(&root->fs_info->trans_lock);
224 if (root->track_dirty && list_empty(&root->dirty_list)) {
225 list_add(&root->dirty_list,
226 &root->fs_info->dirty_cowonly_roots);
228 spin_unlock(&root->fs_info->trans_lock);
232 * used by snapshot creation to make a copy of a root for a tree with
233 * a given objectid. The buffer with the new root node is returned in
234 * cow_ret, and this func returns zero on success or a negative error code.
236 int btrfs_copy_root(struct btrfs_trans_handle *trans,
237 struct btrfs_root *root,
238 struct extent_buffer *buf,
239 struct extent_buffer **cow_ret, u64 new_root_objectid)
241 struct extent_buffer *cow;
244 struct btrfs_disk_key disk_key;
246 WARN_ON(root->ref_cows && trans->transid !=
247 root->fs_info->running_transaction->transid);
248 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
250 level = btrfs_header_level(buf);
252 btrfs_item_key(buf, &disk_key, 0);
254 btrfs_node_key(buf, &disk_key, 0);
256 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
257 new_root_objectid, &disk_key, level,
262 copy_extent_buffer(cow, buf, 0, 0, cow->len);
263 btrfs_set_header_bytenr(cow, cow->start);
264 btrfs_set_header_generation(cow, trans->transid);
265 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
266 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
267 BTRFS_HEADER_FLAG_RELOC);
268 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
269 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
271 btrfs_set_header_owner(cow, new_root_objectid);
273 write_extent_buffer(cow, root->fs_info->fsid,
274 (unsigned long)btrfs_header_fsid(cow),
277 WARN_ON(btrfs_header_generation(buf) > trans->transid);
278 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
279 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
281 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
286 btrfs_mark_buffer_dirty(cow);
292 * check if the tree block can be shared by multiple trees
294 int btrfs_block_can_be_shared(struct btrfs_root *root,
295 struct extent_buffer *buf)
298 * Tree blocks not in refernece counted trees and tree roots
299 * are never shared. If a block was allocated after the last
300 * snapshot and the block was not allocated by tree relocation,
301 * we know the block is not shared.
303 if (root->ref_cows &&
304 buf != root->node && buf != root->commit_root &&
305 (btrfs_header_generation(buf) <=
306 btrfs_root_last_snapshot(&root->root_item) ||
307 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
309 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
310 if (root->ref_cows &&
311 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
317 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
318 struct btrfs_root *root,
319 struct extent_buffer *buf,
320 struct extent_buffer *cow,
330 * Backrefs update rules:
332 * Always use full backrefs for extent pointers in tree block
333 * allocated by tree relocation.
335 * If a shared tree block is no longer referenced by its owner
336 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
337 * use full backrefs for extent pointers in tree block.
339 * If a tree block is been relocating
340 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
341 * use full backrefs for extent pointers in tree block.
342 * The reason for this is some operations (such as drop tree)
343 * are only allowed for blocks use full backrefs.
346 if (btrfs_block_can_be_shared(root, buf)) {
347 ret = btrfs_lookup_extent_info(trans, root, buf->start,
348 buf->len, &refs, &flags);
353 btrfs_std_error(root->fs_info, ret);
358 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
359 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
360 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
365 owner = btrfs_header_owner(buf);
366 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
367 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
370 if ((owner == root->root_key.objectid ||
371 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
372 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
373 ret = btrfs_inc_ref(trans, root, buf, 1, 1);
374 BUG_ON(ret); /* -ENOMEM */
376 if (root->root_key.objectid ==
377 BTRFS_TREE_RELOC_OBJECTID) {
378 ret = btrfs_dec_ref(trans, root, buf, 0, 1);
379 BUG_ON(ret); /* -ENOMEM */
380 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
381 BUG_ON(ret); /* -ENOMEM */
383 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
386 if (root->root_key.objectid ==
387 BTRFS_TREE_RELOC_OBJECTID)
388 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
390 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
391 BUG_ON(ret); /* -ENOMEM */
393 if (new_flags != 0) {
394 ret = btrfs_set_disk_extent_flags(trans, root,
402 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
403 if (root->root_key.objectid ==
404 BTRFS_TREE_RELOC_OBJECTID)
405 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
407 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
408 BUG_ON(ret); /* -ENOMEM */
409 ret = btrfs_dec_ref(trans, root, buf, 1, 1);
410 BUG_ON(ret); /* -ENOMEM */
412 clean_tree_block(trans, root, buf);
419 * does the dirty work in cow of a single block. The parent block (if
420 * supplied) is updated to point to the new cow copy. The new buffer is marked
421 * dirty and returned locked. If you modify the block it needs to be marked
424 * search_start -- an allocation hint for the new block
426 * empty_size -- a hint that you plan on doing more cow. This is the size in
427 * bytes the allocator should try to find free next to the block it returns.
428 * This is just a hint and may be ignored by the allocator.
430 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
431 struct btrfs_root *root,
432 struct extent_buffer *buf,
433 struct extent_buffer *parent, int parent_slot,
434 struct extent_buffer **cow_ret,
435 u64 search_start, u64 empty_size)
437 struct btrfs_disk_key disk_key;
438 struct extent_buffer *cow;
447 btrfs_assert_tree_locked(buf);
449 WARN_ON(root->ref_cows && trans->transid !=
450 root->fs_info->running_transaction->transid);
451 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
453 level = btrfs_header_level(buf);
456 btrfs_item_key(buf, &disk_key, 0);
458 btrfs_node_key(buf, &disk_key, 0);
460 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
462 parent_start = parent->start;
468 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
469 root->root_key.objectid, &disk_key,
470 level, search_start, empty_size, 1);
474 /* cow is set to blocking by btrfs_init_new_buffer */
476 copy_extent_buffer(cow, buf, 0, 0, cow->len);
477 btrfs_set_header_bytenr(cow, cow->start);
478 btrfs_set_header_generation(cow, trans->transid);
479 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
480 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
481 BTRFS_HEADER_FLAG_RELOC);
482 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
483 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
485 btrfs_set_header_owner(cow, root->root_key.objectid);
487 write_extent_buffer(cow, root->fs_info->fsid,
488 (unsigned long)btrfs_header_fsid(cow),
491 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
493 btrfs_abort_transaction(trans, root, ret);
498 btrfs_reloc_cow_block(trans, root, buf, cow);
500 if (buf == root->node) {
501 WARN_ON(parent && parent != buf);
502 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
503 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
504 parent_start = buf->start;
508 extent_buffer_get(cow);
509 rcu_assign_pointer(root->node, cow);
511 btrfs_free_tree_block(trans, root, buf, parent_start,
513 free_extent_buffer(buf);
514 add_root_to_dirty_list(root);
516 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
517 parent_start = parent->start;
521 WARN_ON(trans->transid != btrfs_header_generation(parent));
522 btrfs_set_node_blockptr(parent, parent_slot,
524 btrfs_set_node_ptr_generation(parent, parent_slot,
526 btrfs_mark_buffer_dirty(parent);
527 btrfs_free_tree_block(trans, root, buf, parent_start,
531 btrfs_tree_unlock(buf);
532 free_extent_buffer_stale(buf);
533 btrfs_mark_buffer_dirty(cow);
538 static inline int should_cow_block(struct btrfs_trans_handle *trans,
539 struct btrfs_root *root,
540 struct extent_buffer *buf)
542 /* ensure we can see the force_cow */
546 * We do not need to cow a block if
547 * 1) this block is not created or changed in this transaction;
548 * 2) this block does not belong to TREE_RELOC tree;
549 * 3) the root is not forced COW.
551 * What is forced COW:
552 * when we create snapshot during commiting the transaction,
553 * after we've finished coping src root, we must COW the shared
554 * block to ensure the metadata consistency.
556 if (btrfs_header_generation(buf) == trans->transid &&
557 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
558 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
559 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
566 * cows a single block, see __btrfs_cow_block for the real work.
567 * This version of it has extra checks so that a block isn't cow'd more than
568 * once per transaction, as long as it hasn't been written yet
570 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
571 struct btrfs_root *root, struct extent_buffer *buf,
572 struct extent_buffer *parent, int parent_slot,
573 struct extent_buffer **cow_ret)
578 if (trans->transaction != root->fs_info->running_transaction) {
579 printk(KERN_CRIT "trans %llu running %llu\n",
580 (unsigned long long)trans->transid,
582 root->fs_info->running_transaction->transid);
585 if (trans->transid != root->fs_info->generation) {
586 printk(KERN_CRIT "trans %llu running %llu\n",
587 (unsigned long long)trans->transid,
588 (unsigned long long)root->fs_info->generation);
592 if (!should_cow_block(trans, root, buf)) {
597 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
600 btrfs_set_lock_blocking(parent);
601 btrfs_set_lock_blocking(buf);
603 ret = __btrfs_cow_block(trans, root, buf, parent,
604 parent_slot, cow_ret, search_start, 0);
606 trace_btrfs_cow_block(root, buf, *cow_ret);
612 * helper function for defrag to decide if two blocks pointed to by a
613 * node are actually close by
615 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
617 if (blocknr < other && other - (blocknr + blocksize) < 32768)
619 if (blocknr > other && blocknr - (other + blocksize) < 32768)
625 * compare two keys in a memcmp fashion
627 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
631 btrfs_disk_key_to_cpu(&k1, disk);
633 return btrfs_comp_cpu_keys(&k1, k2);
637 * same as comp_keys only with two btrfs_key's
639 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
641 if (k1->objectid > k2->objectid)
643 if (k1->objectid < k2->objectid)
645 if (k1->type > k2->type)
647 if (k1->type < k2->type)
649 if (k1->offset > k2->offset)
651 if (k1->offset < k2->offset)
657 * this is used by the defrag code to go through all the
658 * leaves pointed to by a node and reallocate them so that
659 * disk order is close to key order
661 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
662 struct btrfs_root *root, struct extent_buffer *parent,
663 int start_slot, int cache_only, u64 *last_ret,
664 struct btrfs_key *progress)
666 struct extent_buffer *cur;
669 u64 search_start = *last_ret;
679 int progress_passed = 0;
680 struct btrfs_disk_key disk_key;
682 parent_level = btrfs_header_level(parent);
683 if (cache_only && parent_level != 1)
686 if (trans->transaction != root->fs_info->running_transaction)
688 if (trans->transid != root->fs_info->generation)
691 parent_nritems = btrfs_header_nritems(parent);
692 blocksize = btrfs_level_size(root, parent_level - 1);
693 end_slot = parent_nritems;
695 if (parent_nritems == 1)
698 btrfs_set_lock_blocking(parent);
700 for (i = start_slot; i < end_slot; i++) {
703 btrfs_node_key(parent, &disk_key, i);
704 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
708 blocknr = btrfs_node_blockptr(parent, i);
709 gen = btrfs_node_ptr_generation(parent, i);
711 last_block = blocknr;
714 other = btrfs_node_blockptr(parent, i - 1);
715 close = close_blocks(blocknr, other, blocksize);
717 if (!close && i < end_slot - 2) {
718 other = btrfs_node_blockptr(parent, i + 1);
719 close = close_blocks(blocknr, other, blocksize);
722 last_block = blocknr;
726 cur = btrfs_find_tree_block(root, blocknr, blocksize);
728 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
731 if (!cur || !uptodate) {
733 free_extent_buffer(cur);
737 cur = read_tree_block(root, blocknr,
741 } else if (!uptodate) {
742 btrfs_read_buffer(cur, gen);
745 if (search_start == 0)
746 search_start = last_block;
748 btrfs_tree_lock(cur);
749 btrfs_set_lock_blocking(cur);
750 err = __btrfs_cow_block(trans, root, cur, parent, i,
753 (end_slot - i) * blocksize));
755 btrfs_tree_unlock(cur);
756 free_extent_buffer(cur);
759 search_start = cur->start;
760 last_block = cur->start;
761 *last_ret = search_start;
762 btrfs_tree_unlock(cur);
763 free_extent_buffer(cur);
769 * The leaf data grows from end-to-front in the node.
770 * this returns the address of the start of the last item,
771 * which is the stop of the leaf data stack
773 static inline unsigned int leaf_data_end(struct btrfs_root *root,
774 struct extent_buffer *leaf)
776 u32 nr = btrfs_header_nritems(leaf);
778 return BTRFS_LEAF_DATA_SIZE(root);
779 return btrfs_item_offset_nr(leaf, nr - 1);
784 * search for key in the extent_buffer. The items start at offset p,
785 * and they are item_size apart. There are 'max' items in p.
787 * the slot in the array is returned via slot, and it points to
788 * the place where you would insert key if it is not found in
791 * slot may point to max if the key is bigger than all of the keys
793 static noinline int generic_bin_search(struct extent_buffer *eb,
795 int item_size, struct btrfs_key *key,
802 struct btrfs_disk_key *tmp = NULL;
803 struct btrfs_disk_key unaligned;
804 unsigned long offset;
806 unsigned long map_start = 0;
807 unsigned long map_len = 0;
811 mid = (low + high) / 2;
812 offset = p + mid * item_size;
814 if (!kaddr || offset < map_start ||
815 (offset + sizeof(struct btrfs_disk_key)) >
816 map_start + map_len) {
818 err = map_private_extent_buffer(eb, offset,
819 sizeof(struct btrfs_disk_key),
820 &kaddr, &map_start, &map_len);
823 tmp = (struct btrfs_disk_key *)(kaddr + offset -
826 read_extent_buffer(eb, &unaligned,
827 offset, sizeof(unaligned));
832 tmp = (struct btrfs_disk_key *)(kaddr + offset -
835 ret = comp_keys(tmp, key);
851 * simple bin_search frontend that does the right thing for
854 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
855 int level, int *slot)
858 return generic_bin_search(eb,
859 offsetof(struct btrfs_leaf, items),
860 sizeof(struct btrfs_item),
861 key, btrfs_header_nritems(eb),
864 return generic_bin_search(eb,
865 offsetof(struct btrfs_node, ptrs),
866 sizeof(struct btrfs_key_ptr),
867 key, btrfs_header_nritems(eb),
873 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
874 int level, int *slot)
876 return bin_search(eb, key, level, slot);
879 static void root_add_used(struct btrfs_root *root, u32 size)
881 spin_lock(&root->accounting_lock);
882 btrfs_set_root_used(&root->root_item,
883 btrfs_root_used(&root->root_item) + size);
884 spin_unlock(&root->accounting_lock);
887 static void root_sub_used(struct btrfs_root *root, u32 size)
889 spin_lock(&root->accounting_lock);
890 btrfs_set_root_used(&root->root_item,
891 btrfs_root_used(&root->root_item) - size);
892 spin_unlock(&root->accounting_lock);
895 /* given a node and slot number, this reads the blocks it points to. The
896 * extent buffer is returned with a reference taken (but unlocked).
897 * NULL is returned on error.
899 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
900 struct extent_buffer *parent, int slot)
902 int level = btrfs_header_level(parent);
905 if (slot >= btrfs_header_nritems(parent))
910 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
911 btrfs_level_size(root, level - 1),
912 btrfs_node_ptr_generation(parent, slot));
916 * node level balancing, used to make sure nodes are in proper order for
917 * item deletion. We balance from the top down, so we have to make sure
918 * that a deletion won't leave an node completely empty later on.
920 static noinline int balance_level(struct btrfs_trans_handle *trans,
921 struct btrfs_root *root,
922 struct btrfs_path *path, int level)
924 struct extent_buffer *right = NULL;
925 struct extent_buffer *mid;
926 struct extent_buffer *left = NULL;
927 struct extent_buffer *parent = NULL;
931 int orig_slot = path->slots[level];
937 mid = path->nodes[level];
939 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
940 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
941 WARN_ON(btrfs_header_generation(mid) != trans->transid);
943 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
945 if (level < BTRFS_MAX_LEVEL - 1) {
946 parent = path->nodes[level + 1];
947 pslot = path->slots[level + 1];
951 * deal with the case where there is only one pointer in the root
952 * by promoting the node below to a root
955 struct extent_buffer *child;
957 if (btrfs_header_nritems(mid) != 1)
960 /* promote the child to a root */
961 child = read_node_slot(root, mid, 0);
964 btrfs_std_error(root->fs_info, ret);
968 btrfs_tree_lock(child);
969 btrfs_set_lock_blocking(child);
970 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
972 btrfs_tree_unlock(child);
973 free_extent_buffer(child);
977 rcu_assign_pointer(root->node, child);
979 add_root_to_dirty_list(root);
980 btrfs_tree_unlock(child);
982 path->locks[level] = 0;
983 path->nodes[level] = NULL;
984 clean_tree_block(trans, root, mid);
985 btrfs_tree_unlock(mid);
986 /* once for the path */
987 free_extent_buffer(mid);
989 root_sub_used(root, mid->len);
990 btrfs_free_tree_block(trans, root, mid, 0, 1, 0);
991 /* once for the root ptr */
992 free_extent_buffer_stale(mid);
995 if (btrfs_header_nritems(mid) >
996 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
999 btrfs_header_nritems(mid);
1001 left = read_node_slot(root, parent, pslot - 1);
1003 btrfs_tree_lock(left);
1004 btrfs_set_lock_blocking(left);
1005 wret = btrfs_cow_block(trans, root, left,
1006 parent, pslot - 1, &left);
1012 right = read_node_slot(root, parent, pslot + 1);
1014 btrfs_tree_lock(right);
1015 btrfs_set_lock_blocking(right);
1016 wret = btrfs_cow_block(trans, root, right,
1017 parent, pslot + 1, &right);
1024 /* first, try to make some room in the middle buffer */
1026 orig_slot += btrfs_header_nritems(left);
1027 wret = push_node_left(trans, root, left, mid, 1);
1030 btrfs_header_nritems(mid);
1034 * then try to empty the right most buffer into the middle
1037 wret = push_node_left(trans, root, mid, right, 1);
1038 if (wret < 0 && wret != -ENOSPC)
1040 if (btrfs_header_nritems(right) == 0) {
1041 clean_tree_block(trans, root, right);
1042 btrfs_tree_unlock(right);
1043 del_ptr(trans, root, path, level + 1, pslot + 1);
1044 root_sub_used(root, right->len);
1045 btrfs_free_tree_block(trans, root, right, 0, 1, 0);
1046 free_extent_buffer_stale(right);
1049 struct btrfs_disk_key right_key;
1050 btrfs_node_key(right, &right_key, 0);
1051 btrfs_set_node_key(parent, &right_key, pslot + 1);
1052 btrfs_mark_buffer_dirty(parent);
1055 if (btrfs_header_nritems(mid) == 1) {
1057 * we're not allowed to leave a node with one item in the
1058 * tree during a delete. A deletion from lower in the tree
1059 * could try to delete the only pointer in this node.
1060 * So, pull some keys from the left.
1061 * There has to be a left pointer at this point because
1062 * otherwise we would have pulled some pointers from the
1067 btrfs_std_error(root->fs_info, ret);
1070 wret = balance_node_right(trans, root, mid, left);
1076 wret = push_node_left(trans, root, left, mid, 1);
1082 if (btrfs_header_nritems(mid) == 0) {
1083 clean_tree_block(trans, root, mid);
1084 btrfs_tree_unlock(mid);
1085 del_ptr(trans, root, path, level + 1, pslot);
1086 root_sub_used(root, mid->len);
1087 btrfs_free_tree_block(trans, root, mid, 0, 1, 0);
1088 free_extent_buffer_stale(mid);
1091 /* update the parent key to reflect our changes */
1092 struct btrfs_disk_key mid_key;
1093 btrfs_node_key(mid, &mid_key, 0);
1094 btrfs_set_node_key(parent, &mid_key, pslot);
1095 btrfs_mark_buffer_dirty(parent);
1098 /* update the path */
1100 if (btrfs_header_nritems(left) > orig_slot) {
1101 extent_buffer_get(left);
1102 /* left was locked after cow */
1103 path->nodes[level] = left;
1104 path->slots[level + 1] -= 1;
1105 path->slots[level] = orig_slot;
1107 btrfs_tree_unlock(mid);
1108 free_extent_buffer(mid);
1111 orig_slot -= btrfs_header_nritems(left);
1112 path->slots[level] = orig_slot;
1115 /* double check we haven't messed things up */
1117 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1121 btrfs_tree_unlock(right);
1122 free_extent_buffer(right);
1125 if (path->nodes[level] != left)
1126 btrfs_tree_unlock(left);
1127 free_extent_buffer(left);
1132 /* Node balancing for insertion. Here we only split or push nodes around
1133 * when they are completely full. This is also done top down, so we
1134 * have to be pessimistic.
1136 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1137 struct btrfs_root *root,
1138 struct btrfs_path *path, int level)
1140 struct extent_buffer *right = NULL;
1141 struct extent_buffer *mid;
1142 struct extent_buffer *left = NULL;
1143 struct extent_buffer *parent = NULL;
1147 int orig_slot = path->slots[level];
1152 mid = path->nodes[level];
1153 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1155 if (level < BTRFS_MAX_LEVEL - 1) {
1156 parent = path->nodes[level + 1];
1157 pslot = path->slots[level + 1];
1163 left = read_node_slot(root, parent, pslot - 1);
1165 /* first, try to make some room in the middle buffer */
1169 btrfs_tree_lock(left);
1170 btrfs_set_lock_blocking(left);
1172 left_nr = btrfs_header_nritems(left);
1173 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1176 ret = btrfs_cow_block(trans, root, left, parent,
1181 wret = push_node_left(trans, root,
1188 struct btrfs_disk_key disk_key;
1189 orig_slot += left_nr;
1190 btrfs_node_key(mid, &disk_key, 0);
1191 btrfs_set_node_key(parent, &disk_key, pslot);
1192 btrfs_mark_buffer_dirty(parent);
1193 if (btrfs_header_nritems(left) > orig_slot) {
1194 path->nodes[level] = left;
1195 path->slots[level + 1] -= 1;
1196 path->slots[level] = orig_slot;
1197 btrfs_tree_unlock(mid);
1198 free_extent_buffer(mid);
1201 btrfs_header_nritems(left);
1202 path->slots[level] = orig_slot;
1203 btrfs_tree_unlock(left);
1204 free_extent_buffer(left);
1208 btrfs_tree_unlock(left);
1209 free_extent_buffer(left);
1211 right = read_node_slot(root, parent, pslot + 1);
1214 * then try to empty the right most buffer into the middle
1219 btrfs_tree_lock(right);
1220 btrfs_set_lock_blocking(right);
1222 right_nr = btrfs_header_nritems(right);
1223 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1226 ret = btrfs_cow_block(trans, root, right,
1232 wret = balance_node_right(trans, root,
1239 struct btrfs_disk_key disk_key;
1241 btrfs_node_key(right, &disk_key, 0);
1242 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1243 btrfs_mark_buffer_dirty(parent);
1245 if (btrfs_header_nritems(mid) <= orig_slot) {
1246 path->nodes[level] = right;
1247 path->slots[level + 1] += 1;
1248 path->slots[level] = orig_slot -
1249 btrfs_header_nritems(mid);
1250 btrfs_tree_unlock(mid);
1251 free_extent_buffer(mid);
1253 btrfs_tree_unlock(right);
1254 free_extent_buffer(right);
1258 btrfs_tree_unlock(right);
1259 free_extent_buffer(right);
1265 * readahead one full node of leaves, finding things that are close
1266 * to the block in 'slot', and triggering ra on them.
1268 static void reada_for_search(struct btrfs_root *root,
1269 struct btrfs_path *path,
1270 int level, int slot, u64 objectid)
1272 struct extent_buffer *node;
1273 struct btrfs_disk_key disk_key;
1279 int direction = path->reada;
1280 struct extent_buffer *eb;
1288 if (!path->nodes[level])
1291 node = path->nodes[level];
1293 search = btrfs_node_blockptr(node, slot);
1294 blocksize = btrfs_level_size(root, level - 1);
1295 eb = btrfs_find_tree_block(root, search, blocksize);
1297 free_extent_buffer(eb);
1303 nritems = btrfs_header_nritems(node);
1307 if (direction < 0) {
1311 } else if (direction > 0) {
1316 if (path->reada < 0 && objectid) {
1317 btrfs_node_key(node, &disk_key, nr);
1318 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1321 search = btrfs_node_blockptr(node, nr);
1322 if ((search <= target && target - search <= 65536) ||
1323 (search > target && search - target <= 65536)) {
1324 gen = btrfs_node_ptr_generation(node, nr);
1325 readahead_tree_block(root, search, blocksize, gen);
1329 if ((nread > 65536 || nscan > 32))
1335 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1338 static noinline int reada_for_balance(struct btrfs_root *root,
1339 struct btrfs_path *path, int level)
1343 struct extent_buffer *parent;
1344 struct extent_buffer *eb;
1351 parent = path->nodes[level + 1];
1355 nritems = btrfs_header_nritems(parent);
1356 slot = path->slots[level + 1];
1357 blocksize = btrfs_level_size(root, level);
1360 block1 = btrfs_node_blockptr(parent, slot - 1);
1361 gen = btrfs_node_ptr_generation(parent, slot - 1);
1362 eb = btrfs_find_tree_block(root, block1, blocksize);
1364 * if we get -eagain from btrfs_buffer_uptodate, we
1365 * don't want to return eagain here. That will loop
1368 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
1370 free_extent_buffer(eb);
1372 if (slot + 1 < nritems) {
1373 block2 = btrfs_node_blockptr(parent, slot + 1);
1374 gen = btrfs_node_ptr_generation(parent, slot + 1);
1375 eb = btrfs_find_tree_block(root, block2, blocksize);
1376 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
1378 free_extent_buffer(eb);
1380 if (block1 || block2) {
1383 /* release the whole path */
1384 btrfs_release_path(path);
1386 /* read the blocks */
1388 readahead_tree_block(root, block1, blocksize, 0);
1390 readahead_tree_block(root, block2, blocksize, 0);
1393 eb = read_tree_block(root, block1, blocksize, 0);
1394 free_extent_buffer(eb);
1397 eb = read_tree_block(root, block2, blocksize, 0);
1398 free_extent_buffer(eb);
1406 * when we walk down the tree, it is usually safe to unlock the higher layers
1407 * in the tree. The exceptions are when our path goes through slot 0, because
1408 * operations on the tree might require changing key pointers higher up in the
1411 * callers might also have set path->keep_locks, which tells this code to keep
1412 * the lock if the path points to the last slot in the block. This is part of
1413 * walking through the tree, and selecting the next slot in the higher block.
1415 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1416 * if lowest_unlock is 1, level 0 won't be unlocked
1418 static noinline void unlock_up(struct btrfs_path *path, int level,
1419 int lowest_unlock, int min_write_lock_level,
1420 int *write_lock_level)
1423 int skip_level = level;
1425 struct extent_buffer *t;
1427 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1428 if (!path->nodes[i])
1430 if (!path->locks[i])
1432 if (!no_skips && path->slots[i] == 0) {
1436 if (!no_skips && path->keep_locks) {
1439 nritems = btrfs_header_nritems(t);
1440 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1445 if (skip_level < i && i >= lowest_unlock)
1449 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1450 btrfs_tree_unlock_rw(t, path->locks[i]);
1452 if (write_lock_level &&
1453 i > min_write_lock_level &&
1454 i <= *write_lock_level) {
1455 *write_lock_level = i - 1;
1462 * This releases any locks held in the path starting at level and
1463 * going all the way up to the root.
1465 * btrfs_search_slot will keep the lock held on higher nodes in a few
1466 * corner cases, such as COW of the block at slot zero in the node. This
1467 * ignores those rules, and it should only be called when there are no
1468 * more updates to be done higher up in the tree.
1470 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1474 if (path->keep_locks)
1477 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1478 if (!path->nodes[i])
1480 if (!path->locks[i])
1482 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
1488 * helper function for btrfs_search_slot. The goal is to find a block
1489 * in cache without setting the path to blocking. If we find the block
1490 * we return zero and the path is unchanged.
1492 * If we can't find the block, we set the path blocking and do some
1493 * reada. -EAGAIN is returned and the search must be repeated.
1496 read_block_for_search(struct btrfs_trans_handle *trans,
1497 struct btrfs_root *root, struct btrfs_path *p,
1498 struct extent_buffer **eb_ret, int level, int slot,
1499 struct btrfs_key *key)
1504 struct extent_buffer *b = *eb_ret;
1505 struct extent_buffer *tmp;
1508 blocknr = btrfs_node_blockptr(b, slot);
1509 gen = btrfs_node_ptr_generation(b, slot);
1510 blocksize = btrfs_level_size(root, level - 1);
1512 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1514 /* first we do an atomic uptodate check */
1515 if (btrfs_buffer_uptodate(tmp, 0, 1) > 0) {
1516 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
1518 * we found an up to date block without
1525 /* the pages were up to date, but we failed
1526 * the generation number check. Do a full
1527 * read for the generation number that is correct.
1528 * We must do this without dropping locks so
1529 * we can trust our generation number
1531 free_extent_buffer(tmp);
1532 btrfs_set_path_blocking(p);
1534 /* now we're allowed to do a blocking uptodate check */
1535 tmp = read_tree_block(root, blocknr, blocksize, gen);
1536 if (tmp && btrfs_buffer_uptodate(tmp, gen, 0) > 0) {
1540 free_extent_buffer(tmp);
1541 btrfs_release_path(p);
1547 * reduce lock contention at high levels
1548 * of the btree by dropping locks before
1549 * we read. Don't release the lock on the current
1550 * level because we need to walk this node to figure
1551 * out which blocks to read.
1553 btrfs_unlock_up_safe(p, level + 1);
1554 btrfs_set_path_blocking(p);
1556 free_extent_buffer(tmp);
1558 reada_for_search(root, p, level, slot, key->objectid);
1560 btrfs_release_path(p);
1563 tmp = read_tree_block(root, blocknr, blocksize, 0);
1566 * If the read above didn't mark this buffer up to date,
1567 * it will never end up being up to date. Set ret to EIO now
1568 * and give up so that our caller doesn't loop forever
1571 if (!btrfs_buffer_uptodate(tmp, 0, 0))
1573 free_extent_buffer(tmp);
1579 * helper function for btrfs_search_slot. This does all of the checks
1580 * for node-level blocks and does any balancing required based on
1583 * If no extra work was required, zero is returned. If we had to
1584 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1588 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1589 struct btrfs_root *root, struct btrfs_path *p,
1590 struct extent_buffer *b, int level, int ins_len,
1591 int *write_lock_level)
1594 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1595 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1598 if (*write_lock_level < level + 1) {
1599 *write_lock_level = level + 1;
1600 btrfs_release_path(p);
1604 sret = reada_for_balance(root, p, level);
1608 btrfs_set_path_blocking(p);
1609 sret = split_node(trans, root, p, level);
1610 btrfs_clear_path_blocking(p, NULL, 0);
1617 b = p->nodes[level];
1618 } else if (ins_len < 0 && btrfs_header_nritems(b) <
1619 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1622 if (*write_lock_level < level + 1) {
1623 *write_lock_level = level + 1;
1624 btrfs_release_path(p);
1628 sret = reada_for_balance(root, p, level);
1632 btrfs_set_path_blocking(p);
1633 sret = balance_level(trans, root, p, level);
1634 btrfs_clear_path_blocking(p, NULL, 0);
1640 b = p->nodes[level];
1642 btrfs_release_path(p);
1645 BUG_ON(btrfs_header_nritems(b) == 1);
1656 * look for key in the tree. path is filled in with nodes along the way
1657 * if key is found, we return zero and you can find the item in the leaf
1658 * level of the path (level 0)
1660 * If the key isn't found, the path points to the slot where it should
1661 * be inserted, and 1 is returned. If there are other errors during the
1662 * search a negative error number is returned.
1664 * if ins_len > 0, nodes and leaves will be split as we walk down the
1665 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1668 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1669 *root, struct btrfs_key *key, struct btrfs_path *p, int
1672 struct extent_buffer *b;
1677 int lowest_unlock = 1;
1679 /* everything at write_lock_level or lower must be write locked */
1680 int write_lock_level = 0;
1681 u8 lowest_level = 0;
1682 int min_write_lock_level;
1684 lowest_level = p->lowest_level;
1685 WARN_ON(lowest_level && ins_len > 0);
1686 WARN_ON(p->nodes[0] != NULL);
1691 /* when we are removing items, we might have to go up to level
1692 * two as we update tree pointers Make sure we keep write
1693 * for those levels as well
1695 write_lock_level = 2;
1696 } else if (ins_len > 0) {
1698 * for inserting items, make sure we have a write lock on
1699 * level 1 so we can update keys
1701 write_lock_level = 1;
1705 write_lock_level = -1;
1707 if (cow && (p->keep_locks || p->lowest_level))
1708 write_lock_level = BTRFS_MAX_LEVEL;
1710 min_write_lock_level = write_lock_level;
1714 * we try very hard to do read locks on the root
1716 root_lock = BTRFS_READ_LOCK;
1718 if (p->search_commit_root) {
1720 * the commit roots are read only
1721 * so we always do read locks
1723 b = root->commit_root;
1724 extent_buffer_get(b);
1725 level = btrfs_header_level(b);
1726 if (!p->skip_locking)
1727 btrfs_tree_read_lock(b);
1729 if (p->skip_locking) {
1730 b = btrfs_root_node(root);
1731 level = btrfs_header_level(b);
1733 /* we don't know the level of the root node
1734 * until we actually have it read locked
1736 b = btrfs_read_lock_root_node(root);
1737 level = btrfs_header_level(b);
1738 if (level <= write_lock_level) {
1739 /* whoops, must trade for write lock */
1740 btrfs_tree_read_unlock(b);
1741 free_extent_buffer(b);
1742 b = btrfs_lock_root_node(root);
1743 root_lock = BTRFS_WRITE_LOCK;
1745 /* the level might have changed, check again */
1746 level = btrfs_header_level(b);
1750 p->nodes[level] = b;
1751 if (!p->skip_locking)
1752 p->locks[level] = root_lock;
1755 level = btrfs_header_level(b);
1758 * setup the path here so we can release it under lock
1759 * contention with the cow code
1763 * if we don't really need to cow this block
1764 * then we don't want to set the path blocking,
1765 * so we test it here
1767 if (!should_cow_block(trans, root, b))
1770 btrfs_set_path_blocking(p);
1773 * must have write locks on this node and the
1776 if (level + 1 > write_lock_level) {
1777 write_lock_level = level + 1;
1778 btrfs_release_path(p);
1782 err = btrfs_cow_block(trans, root, b,
1783 p->nodes[level + 1],
1784 p->slots[level + 1], &b);
1791 BUG_ON(!cow && ins_len);
1793 p->nodes[level] = b;
1794 btrfs_clear_path_blocking(p, NULL, 0);
1797 * we have a lock on b and as long as we aren't changing
1798 * the tree, there is no way to for the items in b to change.
1799 * It is safe to drop the lock on our parent before we
1800 * go through the expensive btree search on b.
1802 * If cow is true, then we might be changing slot zero,
1803 * which may require changing the parent. So, we can't
1804 * drop the lock until after we know which slot we're
1808 btrfs_unlock_up_safe(p, level + 1);
1810 ret = bin_search(b, key, level, &slot);
1814 if (ret && slot > 0) {
1818 p->slots[level] = slot;
1819 err = setup_nodes_for_search(trans, root, p, b, level,
1820 ins_len, &write_lock_level);
1827 b = p->nodes[level];
1828 slot = p->slots[level];
1831 * slot 0 is special, if we change the key
1832 * we have to update the parent pointer
1833 * which means we must have a write lock
1836 if (slot == 0 && cow &&
1837 write_lock_level < level + 1) {
1838 write_lock_level = level + 1;
1839 btrfs_release_path(p);
1843 unlock_up(p, level, lowest_unlock,
1844 min_write_lock_level, &write_lock_level);
1846 if (level == lowest_level) {
1852 err = read_block_for_search(trans, root, p,
1853 &b, level, slot, key);
1861 if (!p->skip_locking) {
1862 level = btrfs_header_level(b);
1863 if (level <= write_lock_level) {
1864 err = btrfs_try_tree_write_lock(b);
1866 btrfs_set_path_blocking(p);
1868 btrfs_clear_path_blocking(p, b,
1871 p->locks[level] = BTRFS_WRITE_LOCK;
1873 err = btrfs_try_tree_read_lock(b);
1875 btrfs_set_path_blocking(p);
1876 btrfs_tree_read_lock(b);
1877 btrfs_clear_path_blocking(p, b,
1880 p->locks[level] = BTRFS_READ_LOCK;
1882 p->nodes[level] = b;
1885 p->slots[level] = slot;
1887 btrfs_leaf_free_space(root, b) < ins_len) {
1888 if (write_lock_level < 1) {
1889 write_lock_level = 1;
1890 btrfs_release_path(p);
1894 btrfs_set_path_blocking(p);
1895 err = split_leaf(trans, root, key,
1896 p, ins_len, ret == 0);
1897 btrfs_clear_path_blocking(p, NULL, 0);
1905 if (!p->search_for_split)
1906 unlock_up(p, level, lowest_unlock,
1907 min_write_lock_level, &write_lock_level);
1914 * we don't really know what they plan on doing with the path
1915 * from here on, so for now just mark it as blocking
1917 if (!p->leave_spinning)
1918 btrfs_set_path_blocking(p);
1920 btrfs_release_path(p);
1925 * adjust the pointers going up the tree, starting at level
1926 * making sure the right key of each node is points to 'key'.
1927 * This is used after shifting pointers to the left, so it stops
1928 * fixing up pointers when a given leaf/node is not in slot 0 of the
1932 static void fixup_low_keys(struct btrfs_trans_handle *trans,
1933 struct btrfs_root *root, struct btrfs_path *path,
1934 struct btrfs_disk_key *key, int level)
1937 struct extent_buffer *t;
1939 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1940 int tslot = path->slots[i];
1941 if (!path->nodes[i])
1944 btrfs_set_node_key(t, key, tslot);
1945 btrfs_mark_buffer_dirty(path->nodes[i]);
1954 * This function isn't completely safe. It's the caller's responsibility
1955 * that the new key won't break the order
1957 void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root, struct btrfs_path *path,
1959 struct btrfs_key *new_key)
1961 struct btrfs_disk_key disk_key;
1962 struct extent_buffer *eb;
1965 eb = path->nodes[0];
1966 slot = path->slots[0];
1968 btrfs_item_key(eb, &disk_key, slot - 1);
1969 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
1971 if (slot < btrfs_header_nritems(eb) - 1) {
1972 btrfs_item_key(eb, &disk_key, slot + 1);
1973 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
1976 btrfs_cpu_key_to_disk(&disk_key, new_key);
1977 btrfs_set_item_key(eb, &disk_key, slot);
1978 btrfs_mark_buffer_dirty(eb);
1980 fixup_low_keys(trans, root, path, &disk_key, 1);
1984 * try to push data from one node into the next node left in the
1987 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1988 * error, and > 0 if there was no room in the left hand block.
1990 static int push_node_left(struct btrfs_trans_handle *trans,
1991 struct btrfs_root *root, struct extent_buffer *dst,
1992 struct extent_buffer *src, int empty)
1999 src_nritems = btrfs_header_nritems(src);
2000 dst_nritems = btrfs_header_nritems(dst);
2001 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2002 WARN_ON(btrfs_header_generation(src) != trans->transid);
2003 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2005 if (!empty && src_nritems <= 8)
2008 if (push_items <= 0)
2012 push_items = min(src_nritems, push_items);
2013 if (push_items < src_nritems) {
2014 /* leave at least 8 pointers in the node if
2015 * we aren't going to empty it
2017 if (src_nritems - push_items < 8) {
2018 if (push_items <= 8)
2024 push_items = min(src_nritems - 8, push_items);
2026 copy_extent_buffer(dst, src,
2027 btrfs_node_key_ptr_offset(dst_nritems),
2028 btrfs_node_key_ptr_offset(0),
2029 push_items * sizeof(struct btrfs_key_ptr));
2031 if (push_items < src_nritems) {
2032 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
2033 btrfs_node_key_ptr_offset(push_items),
2034 (src_nritems - push_items) *
2035 sizeof(struct btrfs_key_ptr));
2037 btrfs_set_header_nritems(src, src_nritems - push_items);
2038 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2039 btrfs_mark_buffer_dirty(src);
2040 btrfs_mark_buffer_dirty(dst);
2046 * try to push data from one node into the next node right in the
2049 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2050 * error, and > 0 if there was no room in the right hand block.
2052 * this will only push up to 1/2 the contents of the left node over
2054 static int balance_node_right(struct btrfs_trans_handle *trans,
2055 struct btrfs_root *root,
2056 struct extent_buffer *dst,
2057 struct extent_buffer *src)
2065 WARN_ON(btrfs_header_generation(src) != trans->transid);
2066 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2068 src_nritems = btrfs_header_nritems(src);
2069 dst_nritems = btrfs_header_nritems(dst);
2070 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2071 if (push_items <= 0)
2074 if (src_nritems < 4)
2077 max_push = src_nritems / 2 + 1;
2078 /* don't try to empty the node */
2079 if (max_push >= src_nritems)
2082 if (max_push < push_items)
2083 push_items = max_push;
2085 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2086 btrfs_node_key_ptr_offset(0),
2088 sizeof(struct btrfs_key_ptr));
2090 copy_extent_buffer(dst, src,
2091 btrfs_node_key_ptr_offset(0),
2092 btrfs_node_key_ptr_offset(src_nritems - push_items),
2093 push_items * sizeof(struct btrfs_key_ptr));
2095 btrfs_set_header_nritems(src, src_nritems - push_items);
2096 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2098 btrfs_mark_buffer_dirty(src);
2099 btrfs_mark_buffer_dirty(dst);
2105 * helper function to insert a new root level in the tree.
2106 * A new node is allocated, and a single item is inserted to
2107 * point to the existing root
2109 * returns zero on success or < 0 on failure.
2111 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2112 struct btrfs_root *root,
2113 struct btrfs_path *path, int level)
2116 struct extent_buffer *lower;
2117 struct extent_buffer *c;
2118 struct extent_buffer *old;
2119 struct btrfs_disk_key lower_key;
2121 BUG_ON(path->nodes[level]);
2122 BUG_ON(path->nodes[level-1] != root->node);
2124 lower = path->nodes[level-1];
2126 btrfs_item_key(lower, &lower_key, 0);
2128 btrfs_node_key(lower, &lower_key, 0);
2130 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2131 root->root_key.objectid, &lower_key,
2132 level, root->node->start, 0, 0);
2136 root_add_used(root, root->nodesize);
2138 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2139 btrfs_set_header_nritems(c, 1);
2140 btrfs_set_header_level(c, level);
2141 btrfs_set_header_bytenr(c, c->start);
2142 btrfs_set_header_generation(c, trans->transid);
2143 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2144 btrfs_set_header_owner(c, root->root_key.objectid);
2146 write_extent_buffer(c, root->fs_info->fsid,
2147 (unsigned long)btrfs_header_fsid(c),
2150 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2151 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2154 btrfs_set_node_key(c, &lower_key, 0);
2155 btrfs_set_node_blockptr(c, 0, lower->start);
2156 lower_gen = btrfs_header_generation(lower);
2157 WARN_ON(lower_gen != trans->transid);
2159 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2161 btrfs_mark_buffer_dirty(c);
2164 rcu_assign_pointer(root->node, c);
2166 /* the super has an extra ref to root->node */
2167 free_extent_buffer(old);
2169 add_root_to_dirty_list(root);
2170 extent_buffer_get(c);
2171 path->nodes[level] = c;
2172 path->locks[level] = BTRFS_WRITE_LOCK;
2173 path->slots[level] = 0;
2178 * worker function to insert a single pointer in a node.
2179 * the node should have enough room for the pointer already
2181 * slot and level indicate where you want the key to go, and
2182 * blocknr is the block the key points to.
2184 static void insert_ptr(struct btrfs_trans_handle *trans,
2185 struct btrfs_root *root, struct btrfs_path *path,
2186 struct btrfs_disk_key *key, u64 bytenr,
2187 int slot, int level)
2189 struct extent_buffer *lower;
2192 BUG_ON(!path->nodes[level]);
2193 btrfs_assert_tree_locked(path->nodes[level]);
2194 lower = path->nodes[level];
2195 nritems = btrfs_header_nritems(lower);
2196 BUG_ON(slot > nritems);
2197 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
2198 if (slot != nritems) {
2199 memmove_extent_buffer(lower,
2200 btrfs_node_key_ptr_offset(slot + 1),
2201 btrfs_node_key_ptr_offset(slot),
2202 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2204 btrfs_set_node_key(lower, key, slot);
2205 btrfs_set_node_blockptr(lower, slot, bytenr);
2206 WARN_ON(trans->transid == 0);
2207 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2208 btrfs_set_header_nritems(lower, nritems + 1);
2209 btrfs_mark_buffer_dirty(lower);
2213 * split the node at the specified level in path in two.
2214 * The path is corrected to point to the appropriate node after the split
2216 * Before splitting this tries to make some room in the node by pushing
2217 * left and right, if either one works, it returns right away.
2219 * returns 0 on success and < 0 on failure
2221 static noinline int split_node(struct btrfs_trans_handle *trans,
2222 struct btrfs_root *root,
2223 struct btrfs_path *path, int level)
2225 struct extent_buffer *c;
2226 struct extent_buffer *split;
2227 struct btrfs_disk_key disk_key;
2232 c = path->nodes[level];
2233 WARN_ON(btrfs_header_generation(c) != trans->transid);
2234 if (c == root->node) {
2235 /* trying to split the root, lets make a new one */
2236 ret = insert_new_root(trans, root, path, level + 1);
2240 ret = push_nodes_for_insert(trans, root, path, level);
2241 c = path->nodes[level];
2242 if (!ret && btrfs_header_nritems(c) <
2243 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2249 c_nritems = btrfs_header_nritems(c);
2250 mid = (c_nritems + 1) / 2;
2251 btrfs_node_key(c, &disk_key, mid);
2253 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2254 root->root_key.objectid,
2255 &disk_key, level, c->start, 0, 0);
2257 return PTR_ERR(split);
2259 root_add_used(root, root->nodesize);
2261 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2262 btrfs_set_header_level(split, btrfs_header_level(c));
2263 btrfs_set_header_bytenr(split, split->start);
2264 btrfs_set_header_generation(split, trans->transid);
2265 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2266 btrfs_set_header_owner(split, root->root_key.objectid);
2267 write_extent_buffer(split, root->fs_info->fsid,
2268 (unsigned long)btrfs_header_fsid(split),
2270 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2271 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2275 copy_extent_buffer(split, c,
2276 btrfs_node_key_ptr_offset(0),
2277 btrfs_node_key_ptr_offset(mid),
2278 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2279 btrfs_set_header_nritems(split, c_nritems - mid);
2280 btrfs_set_header_nritems(c, mid);
2283 btrfs_mark_buffer_dirty(c);
2284 btrfs_mark_buffer_dirty(split);
2286 insert_ptr(trans, root, path, &disk_key, split->start,
2287 path->slots[level + 1] + 1, level + 1);
2289 if (path->slots[level] >= mid) {
2290 path->slots[level] -= mid;
2291 btrfs_tree_unlock(c);
2292 free_extent_buffer(c);
2293 path->nodes[level] = split;
2294 path->slots[level + 1] += 1;
2296 btrfs_tree_unlock(split);
2297 free_extent_buffer(split);
2303 * how many bytes are required to store the items in a leaf. start
2304 * and nr indicate which items in the leaf to check. This totals up the
2305 * space used both by the item structs and the item data
2307 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2310 int nritems = btrfs_header_nritems(l);
2311 int end = min(nritems, start + nr) - 1;
2315 data_len = btrfs_item_end_nr(l, start);
2316 data_len = data_len - btrfs_item_offset_nr(l, end);
2317 data_len += sizeof(struct btrfs_item) * nr;
2318 WARN_ON(data_len < 0);
2323 * The space between the end of the leaf items and
2324 * the start of the leaf data. IOW, how much room
2325 * the leaf has left for both items and data
2327 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2328 struct extent_buffer *leaf)
2330 int nritems = btrfs_header_nritems(leaf);
2332 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2334 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2335 "used %d nritems %d\n",
2336 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2337 leaf_space_used(leaf, 0, nritems), nritems);
2343 * min slot controls the lowest index we're willing to push to the
2344 * right. We'll push up to and including min_slot, but no lower
2346 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2347 struct btrfs_root *root,
2348 struct btrfs_path *path,
2349 int data_size, int empty,
2350 struct extent_buffer *right,
2351 int free_space, u32 left_nritems,
2354 struct extent_buffer *left = path->nodes[0];
2355 struct extent_buffer *upper = path->nodes[1];
2356 struct btrfs_map_token token;
2357 struct btrfs_disk_key disk_key;
2362 struct btrfs_item *item;
2368 btrfs_init_map_token(&token);
2373 nr = max_t(u32, 1, min_slot);
2375 if (path->slots[0] >= left_nritems)
2376 push_space += data_size;
2378 slot = path->slots[1];
2379 i = left_nritems - 1;
2381 item = btrfs_item_nr(left, i);
2383 if (!empty && push_items > 0) {
2384 if (path->slots[0] > i)
2386 if (path->slots[0] == i) {
2387 int space = btrfs_leaf_free_space(root, left);
2388 if (space + push_space * 2 > free_space)
2393 if (path->slots[0] == i)
2394 push_space += data_size;
2396 this_item_size = btrfs_item_size(left, item);
2397 if (this_item_size + sizeof(*item) + push_space > free_space)
2401 push_space += this_item_size + sizeof(*item);
2407 if (push_items == 0)
2410 if (!empty && push_items == left_nritems)
2413 /* push left to right */
2414 right_nritems = btrfs_header_nritems(right);
2416 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2417 push_space -= leaf_data_end(root, left);
2419 /* make room in the right data area */
2420 data_end = leaf_data_end(root, right);
2421 memmove_extent_buffer(right,
2422 btrfs_leaf_data(right) + data_end - push_space,
2423 btrfs_leaf_data(right) + data_end,
2424 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2426 /* copy from the left data area */
2427 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2428 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2429 btrfs_leaf_data(left) + leaf_data_end(root, left),
2432 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2433 btrfs_item_nr_offset(0),
2434 right_nritems * sizeof(struct btrfs_item));
2436 /* copy the items from left to right */
2437 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2438 btrfs_item_nr_offset(left_nritems - push_items),
2439 push_items * sizeof(struct btrfs_item));
2441 /* update the item pointers */
2442 right_nritems += push_items;
2443 btrfs_set_header_nritems(right, right_nritems);
2444 push_space = BTRFS_LEAF_DATA_SIZE(root);
2445 for (i = 0; i < right_nritems; i++) {
2446 item = btrfs_item_nr(right, i);
2447 push_space -= btrfs_token_item_size(right, item, &token);
2448 btrfs_set_token_item_offset(right, item, push_space, &token);
2451 left_nritems -= push_items;
2452 btrfs_set_header_nritems(left, left_nritems);
2455 btrfs_mark_buffer_dirty(left);
2457 clean_tree_block(trans, root, left);
2459 btrfs_mark_buffer_dirty(right);
2461 btrfs_item_key(right, &disk_key, 0);
2462 btrfs_set_node_key(upper, &disk_key, slot + 1);
2463 btrfs_mark_buffer_dirty(upper);
2465 /* then fixup the leaf pointer in the path */
2466 if (path->slots[0] >= left_nritems) {
2467 path->slots[0] -= left_nritems;
2468 if (btrfs_header_nritems(path->nodes[0]) == 0)
2469 clean_tree_block(trans, root, path->nodes[0]);
2470 btrfs_tree_unlock(path->nodes[0]);
2471 free_extent_buffer(path->nodes[0]);
2472 path->nodes[0] = right;
2473 path->slots[1] += 1;
2475 btrfs_tree_unlock(right);
2476 free_extent_buffer(right);
2481 btrfs_tree_unlock(right);
2482 free_extent_buffer(right);
2487 * push some data in the path leaf to the right, trying to free up at
2488 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2490 * returns 1 if the push failed because the other node didn't have enough
2491 * room, 0 if everything worked out and < 0 if there were major errors.
2493 * this will push starting from min_slot to the end of the leaf. It won't
2494 * push any slot lower than min_slot
2496 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2497 *root, struct btrfs_path *path,
2498 int min_data_size, int data_size,
2499 int empty, u32 min_slot)
2501 struct extent_buffer *left = path->nodes[0];
2502 struct extent_buffer *right;
2503 struct extent_buffer *upper;
2509 if (!path->nodes[1])
2512 slot = path->slots[1];
2513 upper = path->nodes[1];
2514 if (slot >= btrfs_header_nritems(upper) - 1)
2517 btrfs_assert_tree_locked(path->nodes[1]);
2519 right = read_node_slot(root, upper, slot + 1);
2523 btrfs_tree_lock(right);
2524 btrfs_set_lock_blocking(right);
2526 free_space = btrfs_leaf_free_space(root, right);
2527 if (free_space < data_size)
2530 /* cow and double check */
2531 ret = btrfs_cow_block(trans, root, right, upper,
2536 free_space = btrfs_leaf_free_space(root, right);
2537 if (free_space < data_size)
2540 left_nritems = btrfs_header_nritems(left);
2541 if (left_nritems == 0)
2544 return __push_leaf_right(trans, root, path, min_data_size, empty,
2545 right, free_space, left_nritems, min_slot);
2547 btrfs_tree_unlock(right);
2548 free_extent_buffer(right);
2553 * push some data in the path leaf to the left, trying to free up at
2554 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2556 * max_slot can put a limit on how far into the leaf we'll push items. The
2557 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
2560 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2561 struct btrfs_root *root,
2562 struct btrfs_path *path, int data_size,
2563 int empty, struct extent_buffer *left,
2564 int free_space, u32 right_nritems,
2567 struct btrfs_disk_key disk_key;
2568 struct extent_buffer *right = path->nodes[0];
2572 struct btrfs_item *item;
2573 u32 old_left_nritems;
2577 u32 old_left_item_size;
2578 struct btrfs_map_token token;
2580 btrfs_init_map_token(&token);
2583 nr = min(right_nritems, max_slot);
2585 nr = min(right_nritems - 1, max_slot);
2587 for (i = 0; i < nr; i++) {
2588 item = btrfs_item_nr(right, i);
2590 if (!empty && push_items > 0) {
2591 if (path->slots[0] < i)
2593 if (path->slots[0] == i) {
2594 int space = btrfs_leaf_free_space(root, right);
2595 if (space + push_space * 2 > free_space)
2600 if (path->slots[0] == i)
2601 push_space += data_size;
2603 this_item_size = btrfs_item_size(right, item);
2604 if (this_item_size + sizeof(*item) + push_space > free_space)
2608 push_space += this_item_size + sizeof(*item);
2611 if (push_items == 0) {
2615 if (!empty && push_items == btrfs_header_nritems(right))
2618 /* push data from right to left */
2619 copy_extent_buffer(left, right,
2620 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2621 btrfs_item_nr_offset(0),
2622 push_items * sizeof(struct btrfs_item));
2624 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2625 btrfs_item_offset_nr(right, push_items - 1);
2627 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2628 leaf_data_end(root, left) - push_space,
2629 btrfs_leaf_data(right) +
2630 btrfs_item_offset_nr(right, push_items - 1),
2632 old_left_nritems = btrfs_header_nritems(left);
2633 BUG_ON(old_left_nritems <= 0);
2635 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2636 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2639 item = btrfs_item_nr(left, i);
2641 ioff = btrfs_token_item_offset(left, item, &token);
2642 btrfs_set_token_item_offset(left, item,
2643 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
2646 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2648 /* fixup right node */
2649 if (push_items > right_nritems) {
2650 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2655 if (push_items < right_nritems) {
2656 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2657 leaf_data_end(root, right);
2658 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2659 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2660 btrfs_leaf_data(right) +
2661 leaf_data_end(root, right), push_space);
2663 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2664 btrfs_item_nr_offset(push_items),
2665 (btrfs_header_nritems(right) - push_items) *
2666 sizeof(struct btrfs_item));
2668 right_nritems -= push_items;
2669 btrfs_set_header_nritems(right, right_nritems);
2670 push_space = BTRFS_LEAF_DATA_SIZE(root);
2671 for (i = 0; i < right_nritems; i++) {
2672 item = btrfs_item_nr(right, i);
2674 push_space = push_space - btrfs_token_item_size(right,
2676 btrfs_set_token_item_offset(right, item, push_space, &token);
2679 btrfs_mark_buffer_dirty(left);
2681 btrfs_mark_buffer_dirty(right);
2683 clean_tree_block(trans, root, right);
2685 btrfs_item_key(right, &disk_key, 0);
2686 fixup_low_keys(trans, root, path, &disk_key, 1);
2688 /* then fixup the leaf pointer in the path */
2689 if (path->slots[0] < push_items) {
2690 path->slots[0] += old_left_nritems;
2691 btrfs_tree_unlock(path->nodes[0]);
2692 free_extent_buffer(path->nodes[0]);
2693 path->nodes[0] = left;
2694 path->slots[1] -= 1;
2696 btrfs_tree_unlock(left);
2697 free_extent_buffer(left);
2698 path->slots[0] -= push_items;
2700 BUG_ON(path->slots[0] < 0);
2703 btrfs_tree_unlock(left);
2704 free_extent_buffer(left);
2709 * push some data in the path leaf to the left, trying to free up at
2710 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2712 * max_slot can put a limit on how far into the leaf we'll push items. The
2713 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
2716 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2717 *root, struct btrfs_path *path, int min_data_size,
2718 int data_size, int empty, u32 max_slot)
2720 struct extent_buffer *right = path->nodes[0];
2721 struct extent_buffer *left;
2727 slot = path->slots[1];
2730 if (!path->nodes[1])
2733 right_nritems = btrfs_header_nritems(right);
2734 if (right_nritems == 0)
2737 btrfs_assert_tree_locked(path->nodes[1]);
2739 left = read_node_slot(root, path->nodes[1], slot - 1);
2743 btrfs_tree_lock(left);
2744 btrfs_set_lock_blocking(left);
2746 free_space = btrfs_leaf_free_space(root, left);
2747 if (free_space < data_size) {
2752 /* cow and double check */
2753 ret = btrfs_cow_block(trans, root, left,
2754 path->nodes[1], slot - 1, &left);
2756 /* we hit -ENOSPC, but it isn't fatal here */
2762 free_space = btrfs_leaf_free_space(root, left);
2763 if (free_space < data_size) {
2768 return __push_leaf_left(trans, root, path, min_data_size,
2769 empty, left, free_space, right_nritems,
2772 btrfs_tree_unlock(left);
2773 free_extent_buffer(left);
2778 * split the path's leaf in two, making sure there is at least data_size
2779 * available for the resulting leaf level of the path.
2781 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
2782 struct btrfs_root *root,
2783 struct btrfs_path *path,
2784 struct extent_buffer *l,
2785 struct extent_buffer *right,
2786 int slot, int mid, int nritems)
2791 struct btrfs_disk_key disk_key;
2792 struct btrfs_map_token token;
2794 btrfs_init_map_token(&token);
2796 nritems = nritems - mid;
2797 btrfs_set_header_nritems(right, nritems);
2798 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2800 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2801 btrfs_item_nr_offset(mid),
2802 nritems * sizeof(struct btrfs_item));
2804 copy_extent_buffer(right, l,
2805 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2806 data_copy_size, btrfs_leaf_data(l) +
2807 leaf_data_end(root, l), data_copy_size);
2809 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2810 btrfs_item_end_nr(l, mid);
2812 for (i = 0; i < nritems; i++) {
2813 struct btrfs_item *item = btrfs_item_nr(right, i);
2816 ioff = btrfs_token_item_offset(right, item, &token);
2817 btrfs_set_token_item_offset(right, item,
2818 ioff + rt_data_off, &token);
2821 btrfs_set_header_nritems(l, mid);
2822 btrfs_item_key(right, &disk_key, 0);
2823 insert_ptr(trans, root, path, &disk_key, right->start,
2824 path->slots[1] + 1, 1);
2826 btrfs_mark_buffer_dirty(right);
2827 btrfs_mark_buffer_dirty(l);
2828 BUG_ON(path->slots[0] != slot);
2831 btrfs_tree_unlock(path->nodes[0]);
2832 free_extent_buffer(path->nodes[0]);
2833 path->nodes[0] = right;
2834 path->slots[0] -= mid;
2835 path->slots[1] += 1;
2837 btrfs_tree_unlock(right);
2838 free_extent_buffer(right);
2841 BUG_ON(path->slots[0] < 0);
2845 * double splits happen when we need to insert a big item in the middle
2846 * of a leaf. A double split can leave us with 3 mostly empty leaves:
2847 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
2850 * We avoid this by trying to push the items on either side of our target
2851 * into the adjacent leaves. If all goes well we can avoid the double split
2854 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
2855 struct btrfs_root *root,
2856 struct btrfs_path *path,
2864 slot = path->slots[0];
2867 * try to push all the items after our slot into the
2870 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
2877 nritems = btrfs_header_nritems(path->nodes[0]);
2879 * our goal is to get our slot at the start or end of a leaf. If
2880 * we've done so we're done
2882 if (path->slots[0] == 0 || path->slots[0] == nritems)
2885 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
2888 /* try to push all the items before our slot into the next leaf */
2889 slot = path->slots[0];
2890 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
2903 * split the path's leaf in two, making sure there is at least data_size
2904 * available for the resulting leaf level of the path.
2906 * returns 0 if all went well and < 0 on failure.
2908 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2909 struct btrfs_root *root,
2910 struct btrfs_key *ins_key,
2911 struct btrfs_path *path, int data_size,
2914 struct btrfs_disk_key disk_key;
2915 struct extent_buffer *l;
2919 struct extent_buffer *right;
2923 int num_doubles = 0;
2924 int tried_avoid_double = 0;
2927 slot = path->slots[0];
2928 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2929 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2932 /* first try to make some room by pushing left and right */
2934 wret = push_leaf_right(trans, root, path, data_size,
2939 wret = push_leaf_left(trans, root, path, data_size,
2940 data_size, 0, (u32)-1);
2946 /* did the pushes work? */
2947 if (btrfs_leaf_free_space(root, l) >= data_size)
2951 if (!path->nodes[1]) {
2952 ret = insert_new_root(trans, root, path, 1);
2959 slot = path->slots[0];
2960 nritems = btrfs_header_nritems(l);
2961 mid = (nritems + 1) / 2;
2965 leaf_space_used(l, mid, nritems - mid) + data_size >
2966 BTRFS_LEAF_DATA_SIZE(root)) {
2967 if (slot >= nritems) {
2971 if (mid != nritems &&
2972 leaf_space_used(l, mid, nritems - mid) +
2973 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2974 if (data_size && !tried_avoid_double)
2975 goto push_for_double;
2981 if (leaf_space_used(l, 0, mid) + data_size >
2982 BTRFS_LEAF_DATA_SIZE(root)) {
2983 if (!extend && data_size && slot == 0) {
2985 } else if ((extend || !data_size) && slot == 0) {
2989 if (mid != nritems &&
2990 leaf_space_used(l, mid, nritems - mid) +
2991 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2992 if (data_size && !tried_avoid_double)
2993 goto push_for_double;
3001 btrfs_cpu_key_to_disk(&disk_key, ins_key);
3003 btrfs_item_key(l, &disk_key, mid);
3005 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
3006 root->root_key.objectid,
3007 &disk_key, 0, l->start, 0, 0);
3009 return PTR_ERR(right);
3011 root_add_used(root, root->leafsize);
3013 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
3014 btrfs_set_header_bytenr(right, right->start);
3015 btrfs_set_header_generation(right, trans->transid);
3016 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
3017 btrfs_set_header_owner(right, root->root_key.objectid);
3018 btrfs_set_header_level(right, 0);
3019 write_extent_buffer(right, root->fs_info->fsid,
3020 (unsigned long)btrfs_header_fsid(right),
3023 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
3024 (unsigned long)btrfs_header_chunk_tree_uuid(right),
3029 btrfs_set_header_nritems(right, 0);
3030 insert_ptr(trans, root, path, &disk_key, right->start,
3031 path->slots[1] + 1, 1);
3032 btrfs_tree_unlock(path->nodes[0]);
3033 free_extent_buffer(path->nodes[0]);
3034 path->nodes[0] = right;
3036 path->slots[1] += 1;
3038 btrfs_set_header_nritems(right, 0);
3039 insert_ptr(trans, root, path, &disk_key, right->start,
3041 btrfs_tree_unlock(path->nodes[0]);
3042 free_extent_buffer(path->nodes[0]);
3043 path->nodes[0] = right;
3045 if (path->slots[1] == 0)
3046 fixup_low_keys(trans, root, path,
3049 btrfs_mark_buffer_dirty(right);
3053 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
3056 BUG_ON(num_doubles != 0);
3064 push_for_double_split(trans, root, path, data_size);
3065 tried_avoid_double = 1;
3066 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3071 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3072 struct btrfs_root *root,
3073 struct btrfs_path *path, int ins_len)
3075 struct btrfs_key key;
3076 struct extent_buffer *leaf;
3077 struct btrfs_file_extent_item *fi;
3082 leaf = path->nodes[0];
3083 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3085 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3086 key.type != BTRFS_EXTENT_CSUM_KEY);
3088 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3091 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3092 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3093 fi = btrfs_item_ptr(leaf, path->slots[0],
3094 struct btrfs_file_extent_item);
3095 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3097 btrfs_release_path(path);
3099 path->keep_locks = 1;
3100 path->search_for_split = 1;
3101 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3102 path->search_for_split = 0;
3107 leaf = path->nodes[0];
3108 /* if our item isn't there or got smaller, return now */
3109 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3112 /* the leaf has changed, it now has room. return now */
3113 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3116 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3117 fi = btrfs_item_ptr(leaf, path->slots[0],
3118 struct btrfs_file_extent_item);
3119 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3123 btrfs_set_path_blocking(path);
3124 ret = split_leaf(trans, root, &key, path, ins_len, 1);
3128 path->keep_locks = 0;
3129 btrfs_unlock_up_safe(path, 1);
3132 path->keep_locks = 0;
3136 static noinline int split_item(struct btrfs_trans_handle *trans,
3137 struct btrfs_root *root,
3138 struct btrfs_path *path,
3139 struct btrfs_key *new_key,
3140 unsigned long split_offset)
3142 struct extent_buffer *leaf;
3143 struct btrfs_item *item;
3144 struct btrfs_item *new_item;
3150 struct btrfs_disk_key disk_key;
3152 leaf = path->nodes[0];
3153 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3155 btrfs_set_path_blocking(path);
3157 item = btrfs_item_nr(leaf, path->slots[0]);
3158 orig_offset = btrfs_item_offset(leaf, item);
3159 item_size = btrfs_item_size(leaf, item);
3161 buf = kmalloc(item_size, GFP_NOFS);
3165 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3166 path->slots[0]), item_size);
3168 slot = path->slots[0] + 1;
3169 nritems = btrfs_header_nritems(leaf);
3170 if (slot != nritems) {
3171 /* shift the items */
3172 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3173 btrfs_item_nr_offset(slot),
3174 (nritems - slot) * sizeof(struct btrfs_item));
3177 btrfs_cpu_key_to_disk(&disk_key, new_key);
3178 btrfs_set_item_key(leaf, &disk_key, slot);
3180 new_item = btrfs_item_nr(leaf, slot);
3182 btrfs_set_item_offset(leaf, new_item, orig_offset);
3183 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3185 btrfs_set_item_offset(leaf, item,
3186 orig_offset + item_size - split_offset);
3187 btrfs_set_item_size(leaf, item, split_offset);
3189 btrfs_set_header_nritems(leaf, nritems + 1);
3191 /* write the data for the start of the original item */
3192 write_extent_buffer(leaf, buf,
3193 btrfs_item_ptr_offset(leaf, path->slots[0]),
3196 /* write the data for the new item */
3197 write_extent_buffer(leaf, buf + split_offset,
3198 btrfs_item_ptr_offset(leaf, slot),
3199 item_size - split_offset);
3200 btrfs_mark_buffer_dirty(leaf);
3202 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
3208 * This function splits a single item into two items,
3209 * giving 'new_key' to the new item and splitting the
3210 * old one at split_offset (from the start of the item).
3212 * The path may be released by this operation. After
3213 * the split, the path is pointing to the old item. The
3214 * new item is going to be in the same node as the old one.
3216 * Note, the item being split must be smaller enough to live alone on
3217 * a tree block with room for one extra struct btrfs_item
3219 * This allows us to split the item in place, keeping a lock on the
3220 * leaf the entire time.
3222 int btrfs_split_item(struct btrfs_trans_handle *trans,
3223 struct btrfs_root *root,
3224 struct btrfs_path *path,
3225 struct btrfs_key *new_key,
3226 unsigned long split_offset)
3229 ret = setup_leaf_for_split(trans, root, path,
3230 sizeof(struct btrfs_item));
3234 ret = split_item(trans, root, path, new_key, split_offset);
3239 * This function duplicate a item, giving 'new_key' to the new item.
3240 * It guarantees both items live in the same tree leaf and the new item
3241 * is contiguous with the original item.
3243 * This allows us to split file extent in place, keeping a lock on the
3244 * leaf the entire time.
3246 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3247 struct btrfs_root *root,
3248 struct btrfs_path *path,
3249 struct btrfs_key *new_key)
3251 struct extent_buffer *leaf;
3255 leaf = path->nodes[0];
3256 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3257 ret = setup_leaf_for_split(trans, root, path,
3258 item_size + sizeof(struct btrfs_item));
3263 setup_items_for_insert(trans, root, path, new_key, &item_size,
3264 item_size, item_size +
3265 sizeof(struct btrfs_item), 1);
3266 leaf = path->nodes[0];
3267 memcpy_extent_buffer(leaf,
3268 btrfs_item_ptr_offset(leaf, path->slots[0]),
3269 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3275 * make the item pointed to by the path smaller. new_size indicates
3276 * how small to make it, and from_end tells us if we just chop bytes
3277 * off the end of the item or if we shift the item to chop bytes off
3280 void btrfs_truncate_item(struct btrfs_trans_handle *trans,
3281 struct btrfs_root *root,
3282 struct btrfs_path *path,
3283 u32 new_size, int from_end)
3286 struct extent_buffer *leaf;
3287 struct btrfs_item *item;
3289 unsigned int data_end;
3290 unsigned int old_data_start;
3291 unsigned int old_size;
3292 unsigned int size_diff;
3294 struct btrfs_map_token token;
3296 btrfs_init_map_token(&token);
3298 leaf = path->nodes[0];
3299 slot = path->slots[0];
3301 old_size = btrfs_item_size_nr(leaf, slot);
3302 if (old_size == new_size)
3305 nritems = btrfs_header_nritems(leaf);
3306 data_end = leaf_data_end(root, leaf);
3308 old_data_start = btrfs_item_offset_nr(leaf, slot);
3310 size_diff = old_size - new_size;
3313 BUG_ON(slot >= nritems);
3316 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3318 /* first correct the data pointers */
3319 for (i = slot; i < nritems; i++) {
3321 item = btrfs_item_nr(leaf, i);
3323 ioff = btrfs_token_item_offset(leaf, item, &token);
3324 btrfs_set_token_item_offset(leaf, item,
3325 ioff + size_diff, &token);
3328 /* shift the data */
3330 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3331 data_end + size_diff, btrfs_leaf_data(leaf) +
3332 data_end, old_data_start + new_size - data_end);
3334 struct btrfs_disk_key disk_key;
3337 btrfs_item_key(leaf, &disk_key, slot);
3339 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3341 struct btrfs_file_extent_item *fi;
3343 fi = btrfs_item_ptr(leaf, slot,
3344 struct btrfs_file_extent_item);
3345 fi = (struct btrfs_file_extent_item *)(
3346 (unsigned long)fi - size_diff);
3348 if (btrfs_file_extent_type(leaf, fi) ==
3349 BTRFS_FILE_EXTENT_INLINE) {
3350 ptr = btrfs_item_ptr_offset(leaf, slot);
3351 memmove_extent_buffer(leaf, ptr,
3353 offsetof(struct btrfs_file_extent_item,
3358 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3359 data_end + size_diff, btrfs_leaf_data(leaf) +
3360 data_end, old_data_start - data_end);
3362 offset = btrfs_disk_key_offset(&disk_key);
3363 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3364 btrfs_set_item_key(leaf, &disk_key, slot);
3366 fixup_low_keys(trans, root, path, &disk_key, 1);
3369 item = btrfs_item_nr(leaf, slot);
3370 btrfs_set_item_size(leaf, item, new_size);
3371 btrfs_mark_buffer_dirty(leaf);
3373 if (btrfs_leaf_free_space(root, leaf) < 0) {
3374 btrfs_print_leaf(root, leaf);
3380 * make the item pointed to by the path bigger, data_size is the new size.
3382 void btrfs_extend_item(struct btrfs_trans_handle *trans,
3383 struct btrfs_root *root, struct btrfs_path *path,
3387 struct extent_buffer *leaf;
3388 struct btrfs_item *item;
3390 unsigned int data_end;
3391 unsigned int old_data;
3392 unsigned int old_size;
3394 struct btrfs_map_token token;
3396 btrfs_init_map_token(&token);
3398 leaf = path->nodes[0];
3400 nritems = btrfs_header_nritems(leaf);
3401 data_end = leaf_data_end(root, leaf);
3403 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3404 btrfs_print_leaf(root, leaf);
3407 slot = path->slots[0];
3408 old_data = btrfs_item_end_nr(leaf, slot);
3411 if (slot >= nritems) {
3412 btrfs_print_leaf(root, leaf);
3413 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3419 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3421 /* first correct the data pointers */
3422 for (i = slot; i < nritems; i++) {
3424 item = btrfs_item_nr(leaf, i);
3426 ioff = btrfs_token_item_offset(leaf, item, &token);
3427 btrfs_set_token_item_offset(leaf, item,
3428 ioff - data_size, &token);
3431 /* shift the data */
3432 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3433 data_end - data_size, btrfs_leaf_data(leaf) +
3434 data_end, old_data - data_end);
3436 data_end = old_data;
3437 old_size = btrfs_item_size_nr(leaf, slot);
3438 item = btrfs_item_nr(leaf, slot);
3439 btrfs_set_item_size(leaf, item, old_size + data_size);
3440 btrfs_mark_buffer_dirty(leaf);
3442 if (btrfs_leaf_free_space(root, leaf) < 0) {
3443 btrfs_print_leaf(root, leaf);
3449 * Given a key and some data, insert items into the tree.
3450 * This does all the path init required, making room in the tree if needed.
3451 * Returns the number of keys that were inserted.
3453 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3454 struct btrfs_root *root,
3455 struct btrfs_path *path,
3456 struct btrfs_key *cpu_key, u32 *data_size,
3459 struct extent_buffer *leaf;
3460 struct btrfs_item *item;
3467 unsigned int data_end;
3468 struct btrfs_disk_key disk_key;
3469 struct btrfs_key found_key;
3470 struct btrfs_map_token token;
3472 btrfs_init_map_token(&token);
3474 for (i = 0; i < nr; i++) {
3475 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3476 BTRFS_LEAF_DATA_SIZE(root)) {
3480 total_data += data_size[i];
3481 total_size += data_size[i] + sizeof(struct btrfs_item);
3485 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3491 leaf = path->nodes[0];
3493 nritems = btrfs_header_nritems(leaf);
3494 data_end = leaf_data_end(root, leaf);
3496 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3497 for (i = nr; i >= 0; i--) {
3498 total_data -= data_size[i];
3499 total_size -= data_size[i] + sizeof(struct btrfs_item);
3500 if (total_size < btrfs_leaf_free_space(root, leaf))
3506 slot = path->slots[0];
3509 if (slot != nritems) {
3510 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3512 item = btrfs_item_nr(leaf, slot);
3513 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3515 /* figure out how many keys we can insert in here */
3516 total_data = data_size[0];
3517 for (i = 1; i < nr; i++) {
3518 if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3520 total_data += data_size[i];
3524 if (old_data < data_end) {
3525 btrfs_print_leaf(root, leaf);
3526 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3527 slot, old_data, data_end);
3531 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3533 /* first correct the data pointers */
3534 for (i = slot; i < nritems; i++) {
3537 item = btrfs_item_nr(leaf, i);
3538 ioff = btrfs_token_item_offset(leaf, item, &token);
3539 btrfs_set_token_item_offset(leaf, item,
3540 ioff - total_data, &token);
3542 /* shift the items */
3543 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3544 btrfs_item_nr_offset(slot),
3545 (nritems - slot) * sizeof(struct btrfs_item));
3547 /* shift the data */
3548 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3549 data_end - total_data, btrfs_leaf_data(leaf) +
3550 data_end, old_data - data_end);
3551 data_end = old_data;
3554 * this sucks but it has to be done, if we are inserting at
3555 * the end of the leaf only insert 1 of the items, since we
3556 * have no way of knowing whats on the next leaf and we'd have
3557 * to drop our current locks to figure it out
3562 /* setup the item for the new data */
3563 for (i = 0; i < nr; i++) {
3564 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3565 btrfs_set_item_key(leaf, &disk_key, slot + i);
3566 item = btrfs_item_nr(leaf, slot + i);
3567 btrfs_set_token_item_offset(leaf, item,
3568 data_end - data_size[i], &token);
3569 data_end -= data_size[i];
3570 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
3572 btrfs_set_header_nritems(leaf, nritems + nr);
3573 btrfs_mark_buffer_dirty(leaf);
3577 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3578 fixup_low_keys(trans, root, path, &disk_key, 1);
3581 if (btrfs_leaf_free_space(root, leaf) < 0) {
3582 btrfs_print_leaf(root, leaf);
3592 * this is a helper for btrfs_insert_empty_items, the main goal here is
3593 * to save stack depth by doing the bulk of the work in a function
3594 * that doesn't call btrfs_search_slot
3596 void setup_items_for_insert(struct btrfs_trans_handle *trans,
3597 struct btrfs_root *root, struct btrfs_path *path,
3598 struct btrfs_key *cpu_key, u32 *data_size,
3599 u32 total_data, u32 total_size, int nr)
3601 struct btrfs_item *item;
3604 unsigned int data_end;
3605 struct btrfs_disk_key disk_key;
3606 struct extent_buffer *leaf;
3608 struct btrfs_map_token token;
3610 btrfs_init_map_token(&token);
3612 leaf = path->nodes[0];
3613 slot = path->slots[0];
3615 nritems = btrfs_header_nritems(leaf);
3616 data_end = leaf_data_end(root, leaf);
3618 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3619 btrfs_print_leaf(root, leaf);
3620 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3621 total_size, btrfs_leaf_free_space(root, leaf));
3625 if (slot != nritems) {
3626 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3628 if (old_data < data_end) {
3629 btrfs_print_leaf(root, leaf);
3630 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3631 slot, old_data, data_end);
3635 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3637 /* first correct the data pointers */
3638 for (i = slot; i < nritems; i++) {
3641 item = btrfs_item_nr(leaf, i);
3642 ioff = btrfs_token_item_offset(leaf, item, &token);
3643 btrfs_set_token_item_offset(leaf, item,
3644 ioff - total_data, &token);
3646 /* shift the items */
3647 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3648 btrfs_item_nr_offset(slot),
3649 (nritems - slot) * sizeof(struct btrfs_item));
3651 /* shift the data */
3652 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3653 data_end - total_data, btrfs_leaf_data(leaf) +
3654 data_end, old_data - data_end);
3655 data_end = old_data;
3658 /* setup the item for the new data */
3659 for (i = 0; i < nr; i++) {
3660 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3661 btrfs_set_item_key(leaf, &disk_key, slot + i);
3662 item = btrfs_item_nr(leaf, slot + i);
3663 btrfs_set_token_item_offset(leaf, item,
3664 data_end - data_size[i], &token);
3665 data_end -= data_size[i];
3666 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
3669 btrfs_set_header_nritems(leaf, nritems + nr);
3672 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3673 fixup_low_keys(trans, root, path, &disk_key, 1);
3675 btrfs_unlock_up_safe(path, 1);
3676 btrfs_mark_buffer_dirty(leaf);
3678 if (btrfs_leaf_free_space(root, leaf) < 0) {
3679 btrfs_print_leaf(root, leaf);
3685 * Given a key and some data, insert items into the tree.
3686 * This does all the path init required, making room in the tree if needed.
3688 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3689 struct btrfs_root *root,
3690 struct btrfs_path *path,
3691 struct btrfs_key *cpu_key, u32 *data_size,
3700 for (i = 0; i < nr; i++)
3701 total_data += data_size[i];
3703 total_size = total_data + (nr * sizeof(struct btrfs_item));
3704 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3710 slot = path->slots[0];
3713 setup_items_for_insert(trans, root, path, cpu_key, data_size,
3714 total_data, total_size, nr);
3719 * Given a key and some data, insert an item into the tree.
3720 * This does all the path init required, making room in the tree if needed.
3722 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3723 *root, struct btrfs_key *cpu_key, void *data, u32
3727 struct btrfs_path *path;
3728 struct extent_buffer *leaf;
3731 path = btrfs_alloc_path();
3734 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3736 leaf = path->nodes[0];
3737 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3738 write_extent_buffer(leaf, data, ptr, data_size);
3739 btrfs_mark_buffer_dirty(leaf);
3741 btrfs_free_path(path);
3746 * delete the pointer from a given node.
3748 * the tree should have been previously balanced so the deletion does not
3751 static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3752 struct btrfs_path *path, int level, int slot)
3754 struct extent_buffer *parent = path->nodes[level];
3757 nritems = btrfs_header_nritems(parent);
3758 if (slot != nritems - 1) {
3759 memmove_extent_buffer(parent,
3760 btrfs_node_key_ptr_offset(slot),
3761 btrfs_node_key_ptr_offset(slot + 1),
3762 sizeof(struct btrfs_key_ptr) *
3763 (nritems - slot - 1));
3766 btrfs_set_header_nritems(parent, nritems);
3767 if (nritems == 0 && parent == root->node) {
3768 BUG_ON(btrfs_header_level(root->node) != 1);
3769 /* just turn the root into a leaf and break */
3770 btrfs_set_header_level(root->node, 0);
3771 } else if (slot == 0) {
3772 struct btrfs_disk_key disk_key;
3774 btrfs_node_key(parent, &disk_key, 0);
3775 fixup_low_keys(trans, root, path, &disk_key, level + 1);
3777 btrfs_mark_buffer_dirty(parent);
3781 * a helper function to delete the leaf pointed to by path->slots[1] and
3784 * This deletes the pointer in path->nodes[1] and frees the leaf
3785 * block extent. zero is returned if it all worked out, < 0 otherwise.
3787 * The path must have already been setup for deleting the leaf, including
3788 * all the proper balancing. path->nodes[1] must be locked.
3790 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
3791 struct btrfs_root *root,
3792 struct btrfs_path *path,
3793 struct extent_buffer *leaf)
3795 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3796 del_ptr(trans, root, path, 1, path->slots[1]);
3799 * btrfs_free_extent is expensive, we want to make sure we
3800 * aren't holding any locks when we call it
3802 btrfs_unlock_up_safe(path, 0);
3804 root_sub_used(root, leaf->len);
3806 extent_buffer_get(leaf);
3807 btrfs_free_tree_block(trans, root, leaf, 0, 1, 0);
3808 free_extent_buffer_stale(leaf);
3811 * delete the item at the leaf level in path. If that empties
3812 * the leaf, remove it from the tree
3814 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3815 struct btrfs_path *path, int slot, int nr)
3817 struct extent_buffer *leaf;
3818 struct btrfs_item *item;
3825 struct btrfs_map_token token;
3827 btrfs_init_map_token(&token);
3829 leaf = path->nodes[0];
3830 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3832 for (i = 0; i < nr; i++)
3833 dsize += btrfs_item_size_nr(leaf, slot + i);
3835 nritems = btrfs_header_nritems(leaf);
3837 if (slot + nr != nritems) {
3838 int data_end = leaf_data_end(root, leaf);
3840 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3842 btrfs_leaf_data(leaf) + data_end,
3843 last_off - data_end);
3845 for (i = slot + nr; i < nritems; i++) {
3848 item = btrfs_item_nr(leaf, i);
3849 ioff = btrfs_token_item_offset(leaf, item, &token);
3850 btrfs_set_token_item_offset(leaf, item,
3851 ioff + dsize, &token);
3854 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3855 btrfs_item_nr_offset(slot + nr),
3856 sizeof(struct btrfs_item) *
3857 (nritems - slot - nr));
3859 btrfs_set_header_nritems(leaf, nritems - nr);
3862 /* delete the leaf if we've emptied it */
3864 if (leaf == root->node) {
3865 btrfs_set_header_level(leaf, 0);
3867 btrfs_set_path_blocking(path);
3868 clean_tree_block(trans, root, leaf);
3869 btrfs_del_leaf(trans, root, path, leaf);
3872 int used = leaf_space_used(leaf, 0, nritems);
3874 struct btrfs_disk_key disk_key;
3876 btrfs_item_key(leaf, &disk_key, 0);
3877 fixup_low_keys(trans, root, path, &disk_key, 1);
3880 /* delete the leaf if it is mostly empty */
3881 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3882 /* push_leaf_left fixes the path.
3883 * make sure the path still points to our leaf
3884 * for possible call to del_ptr below
3886 slot = path->slots[1];
3887 extent_buffer_get(leaf);
3889 btrfs_set_path_blocking(path);
3890 wret = push_leaf_left(trans, root, path, 1, 1,
3892 if (wret < 0 && wret != -ENOSPC)
3895 if (path->nodes[0] == leaf &&
3896 btrfs_header_nritems(leaf)) {
3897 wret = push_leaf_right(trans, root, path, 1,
3899 if (wret < 0 && wret != -ENOSPC)
3903 if (btrfs_header_nritems(leaf) == 0) {
3904 path->slots[1] = slot;
3905 btrfs_del_leaf(trans, root, path, leaf);
3906 free_extent_buffer(leaf);
3909 /* if we're still in the path, make sure
3910 * we're dirty. Otherwise, one of the
3911 * push_leaf functions must have already
3912 * dirtied this buffer
3914 if (path->nodes[0] == leaf)
3915 btrfs_mark_buffer_dirty(leaf);
3916 free_extent_buffer(leaf);
3919 btrfs_mark_buffer_dirty(leaf);
3926 * search the tree again to find a leaf with lesser keys
3927 * returns 0 if it found something or 1 if there are no lesser leaves.
3928 * returns < 0 on io errors.
3930 * This may release the path, and so you may lose any locks held at the
3933 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3935 struct btrfs_key key;
3936 struct btrfs_disk_key found_key;
3939 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3943 else if (key.type > 0)
3945 else if (key.objectid > 0)
3950 btrfs_release_path(path);
3951 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3954 btrfs_item_key(path->nodes[0], &found_key, 0);
3955 ret = comp_keys(&found_key, &key);
3962 * A helper function to walk down the tree starting at min_key, and looking
3963 * for nodes or leaves that are either in cache or have a minimum
3964 * transaction id. This is used by the btree defrag code, and tree logging
3966 * This does not cow, but it does stuff the starting key it finds back
3967 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3968 * key and get a writable path.
3970 * This does lock as it descends, and path->keep_locks should be set
3971 * to 1 by the caller.
3973 * This honors path->lowest_level to prevent descent past a given level
3976 * min_trans indicates the oldest transaction that you are interested
3977 * in walking through. Any nodes or leaves older than min_trans are
3978 * skipped over (without reading them).
3980 * returns zero if something useful was found, < 0 on error and 1 if there
3981 * was nothing in the tree that matched the search criteria.
3983 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3984 struct btrfs_key *max_key,
3985 struct btrfs_path *path, int cache_only,
3988 struct extent_buffer *cur;
3989 struct btrfs_key found_key;
3996 WARN_ON(!path->keep_locks);
3998 cur = btrfs_read_lock_root_node(root);
3999 level = btrfs_header_level(cur);
4000 WARN_ON(path->nodes[level]);
4001 path->nodes[level] = cur;
4002 path->locks[level] = BTRFS_READ_LOCK;
4004 if (btrfs_header_generation(cur) < min_trans) {
4009 nritems = btrfs_header_nritems(cur);
4010 level = btrfs_header_level(cur);
4011 sret = bin_search(cur, min_key, level, &slot);
4013 /* at the lowest level, we're done, setup the path and exit */
4014 if (level == path->lowest_level) {
4015 if (slot >= nritems)
4018 path->slots[level] = slot;
4019 btrfs_item_key_to_cpu(cur, &found_key, slot);
4022 if (sret && slot > 0)
4025 * check this node pointer against the cache_only and
4026 * min_trans parameters. If it isn't in cache or is too
4027 * old, skip to the next one.
4029 while (slot < nritems) {
4032 struct extent_buffer *tmp;
4033 struct btrfs_disk_key disk_key;
4035 blockptr = btrfs_node_blockptr(cur, slot);
4036 gen = btrfs_node_ptr_generation(cur, slot);
4037 if (gen < min_trans) {
4045 btrfs_node_key(cur, &disk_key, slot);
4046 if (comp_keys(&disk_key, max_key) >= 0) {
4052 tmp = btrfs_find_tree_block(root, blockptr,
4053 btrfs_level_size(root, level - 1));
4055 if (tmp && btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
4056 free_extent_buffer(tmp);
4060 free_extent_buffer(tmp);
4065 * we didn't find a candidate key in this node, walk forward
4066 * and find another one
4068 if (slot >= nritems) {
4069 path->slots[level] = slot;
4070 btrfs_set_path_blocking(path);
4071 sret = btrfs_find_next_key(root, path, min_key, level,
4072 cache_only, min_trans);
4074 btrfs_release_path(path);
4080 /* save our key for returning back */
4081 btrfs_node_key_to_cpu(cur, &found_key, slot);
4082 path->slots[level] = slot;
4083 if (level == path->lowest_level) {
4085 unlock_up(path, level, 1, 0, NULL);
4088 btrfs_set_path_blocking(path);
4089 cur = read_node_slot(root, cur, slot);
4090 BUG_ON(!cur); /* -ENOMEM */
4092 btrfs_tree_read_lock(cur);
4094 path->locks[level - 1] = BTRFS_READ_LOCK;
4095 path->nodes[level - 1] = cur;
4096 unlock_up(path, level, 1, 0, NULL);
4097 btrfs_clear_path_blocking(path, NULL, 0);
4101 memcpy(min_key, &found_key, sizeof(found_key));
4102 btrfs_set_path_blocking(path);
4107 * this is similar to btrfs_next_leaf, but does not try to preserve
4108 * and fixup the path. It looks for and returns the next key in the
4109 * tree based on the current path and the cache_only and min_trans
4112 * 0 is returned if another key is found, < 0 if there are any errors
4113 * and 1 is returned if there are no higher keys in the tree
4115 * path->keep_locks should be set to 1 on the search made before
4116 * calling this function.
4118 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4119 struct btrfs_key *key, int level,
4120 int cache_only, u64 min_trans)
4123 struct extent_buffer *c;
4125 WARN_ON(!path->keep_locks);
4126 while (level < BTRFS_MAX_LEVEL) {
4127 if (!path->nodes[level])
4130 slot = path->slots[level] + 1;
4131 c = path->nodes[level];
4133 if (slot >= btrfs_header_nritems(c)) {
4136 struct btrfs_key cur_key;
4137 if (level + 1 >= BTRFS_MAX_LEVEL ||
4138 !path->nodes[level + 1])
4141 if (path->locks[level + 1]) {
4146 slot = btrfs_header_nritems(c) - 1;
4148 btrfs_item_key_to_cpu(c, &cur_key, slot);
4150 btrfs_node_key_to_cpu(c, &cur_key, slot);
4152 orig_lowest = path->lowest_level;
4153 btrfs_release_path(path);
4154 path->lowest_level = level;
4155 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4157 path->lowest_level = orig_lowest;
4161 c = path->nodes[level];
4162 slot = path->slots[level];
4169 btrfs_item_key_to_cpu(c, key, slot);
4171 u64 blockptr = btrfs_node_blockptr(c, slot);
4172 u64 gen = btrfs_node_ptr_generation(c, slot);
4175 struct extent_buffer *cur;
4176 cur = btrfs_find_tree_block(root, blockptr,
4177 btrfs_level_size(root, level - 1));
4179 btrfs_buffer_uptodate(cur, gen, 1) <= 0) {
4182 free_extent_buffer(cur);
4185 free_extent_buffer(cur);
4187 if (gen < min_trans) {
4191 btrfs_node_key_to_cpu(c, key, slot);
4199 * search the tree again to find a leaf with greater keys
4200 * returns 0 if it found something or 1 if there are no greater leaves.
4201 * returns < 0 on io errors.
4203 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4207 struct extent_buffer *c;
4208 struct extent_buffer *next;
4209 struct btrfs_key key;
4212 int old_spinning = path->leave_spinning;
4213 int next_rw_lock = 0;
4215 nritems = btrfs_header_nritems(path->nodes[0]);
4219 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4224 btrfs_release_path(path);
4226 path->keep_locks = 1;
4227 path->leave_spinning = 1;
4229 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4230 path->keep_locks = 0;
4235 nritems = btrfs_header_nritems(path->nodes[0]);
4237 * by releasing the path above we dropped all our locks. A balance
4238 * could have added more items next to the key that used to be
4239 * at the very end of the block. So, check again here and
4240 * advance the path if there are now more items available.
4242 if (nritems > 0 && path->slots[0] < nritems - 1) {
4249 while (level < BTRFS_MAX_LEVEL) {
4250 if (!path->nodes[level]) {
4255 slot = path->slots[level] + 1;
4256 c = path->nodes[level];
4257 if (slot >= btrfs_header_nritems(c)) {
4259 if (level == BTRFS_MAX_LEVEL) {
4267 btrfs_tree_unlock_rw(next, next_rw_lock);
4268 free_extent_buffer(next);
4272 next_rw_lock = path->locks[level];
4273 ret = read_block_for_search(NULL, root, path, &next, level,
4279 btrfs_release_path(path);
4283 if (!path->skip_locking) {
4284 ret = btrfs_try_tree_read_lock(next);
4286 btrfs_set_path_blocking(path);
4287 btrfs_tree_read_lock(next);
4288 btrfs_clear_path_blocking(path, next,
4291 next_rw_lock = BTRFS_READ_LOCK;
4295 path->slots[level] = slot;
4298 c = path->nodes[level];
4299 if (path->locks[level])
4300 btrfs_tree_unlock_rw(c, path->locks[level]);
4302 free_extent_buffer(c);
4303 path->nodes[level] = next;
4304 path->slots[level] = 0;
4305 if (!path->skip_locking)
4306 path->locks[level] = next_rw_lock;
4310 ret = read_block_for_search(NULL, root, path, &next, level,
4316 btrfs_release_path(path);
4320 if (!path->skip_locking) {
4321 ret = btrfs_try_tree_read_lock(next);
4323 btrfs_set_path_blocking(path);
4324 btrfs_tree_read_lock(next);
4325 btrfs_clear_path_blocking(path, next,
4328 next_rw_lock = BTRFS_READ_LOCK;
4333 unlock_up(path, 0, 1, 0, NULL);
4334 path->leave_spinning = old_spinning;
4336 btrfs_set_path_blocking(path);
4342 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4343 * searching until it gets past min_objectid or finds an item of 'type'
4345 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4347 int btrfs_previous_item(struct btrfs_root *root,
4348 struct btrfs_path *path, u64 min_objectid,
4351 struct btrfs_key found_key;
4352 struct extent_buffer *leaf;
4357 if (path->slots[0] == 0) {
4358 btrfs_set_path_blocking(path);
4359 ret = btrfs_prev_leaf(root, path);
4365 leaf = path->nodes[0];
4366 nritems = btrfs_header_nritems(leaf);
4369 if (path->slots[0] == nritems)
4372 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4373 if (found_key.objectid < min_objectid)
4375 if (found_key.type == type)
4377 if (found_key.objectid == min_objectid &&
4378 found_key.type < type)