2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
22 #include "transaction.h"
23 #include "print-tree.h"
26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_key *ins_key,
30 struct btrfs_path *path, int data_size, int extend);
31 static int push_node_left(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root, struct extent_buffer *dst,
33 struct extent_buffer *src, int empty);
34 static int balance_node_right(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 struct extent_buffer *dst_buf,
37 struct extent_buffer *src_buf);
38 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 struct btrfs_path *path, int level, int slot);
41 inline void btrfs_init_path(struct btrfs_path *p)
43 memset(p, 0, sizeof(*p));
46 struct btrfs_path *btrfs_alloc_path(void)
48 struct btrfs_path *path;
49 path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
51 btrfs_init_path(path);
57 void btrfs_free_path(struct btrfs_path *p)
59 btrfs_release_path(NULL, p);
60 kmem_cache_free(btrfs_path_cachep, p);
63 void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
67 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
72 btrfs_tree_unlock(p->nodes[i]);
75 free_extent_buffer(p->nodes[i]);
80 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
82 struct extent_buffer *eb;
83 spin_lock(&root->node_lock);
85 extent_buffer_get(eb);
86 spin_unlock(&root->node_lock);
90 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
92 struct extent_buffer *eb;
95 eb = btrfs_root_node(root);
98 spin_lock(&root->node_lock);
99 if (eb == root->node) {
100 spin_unlock(&root->node_lock);
103 spin_unlock(&root->node_lock);
105 btrfs_tree_unlock(eb);
106 free_extent_buffer(eb);
111 static void add_root_to_dirty_list(struct btrfs_root *root)
113 if (root->track_dirty && list_empty(&root->dirty_list)) {
114 list_add(&root->dirty_list,
115 &root->fs_info->dirty_cowonly_roots);
119 int btrfs_copy_root(struct btrfs_trans_handle *trans,
120 struct btrfs_root *root,
121 struct extent_buffer *buf,
122 struct extent_buffer **cow_ret, u64 new_root_objectid)
124 struct extent_buffer *cow;
128 struct btrfs_key first_key;
129 struct btrfs_root *new_root;
131 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
135 memcpy(new_root, root, sizeof(*new_root));
136 new_root->root_key.objectid = new_root_objectid;
138 WARN_ON(root->ref_cows && trans->transid !=
139 root->fs_info->running_transaction->transid);
140 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
142 level = btrfs_header_level(buf);
143 nritems = btrfs_header_nritems(buf);
146 btrfs_item_key_to_cpu(buf, &first_key, 0);
148 btrfs_node_key_to_cpu(buf, &first_key, 0);
150 first_key.objectid = 0;
152 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
154 trans->transid, first_key.objectid,
155 level, buf->start, 0);
161 copy_extent_buffer(cow, buf, 0, 0, cow->len);
162 btrfs_set_header_bytenr(cow, cow->start);
163 btrfs_set_header_generation(cow, trans->transid);
164 btrfs_set_header_owner(cow, new_root_objectid);
165 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
167 WARN_ON(btrfs_header_generation(buf) > trans->transid);
168 ret = btrfs_inc_ref(trans, new_root, buf, 0);
174 btrfs_mark_buffer_dirty(cow);
179 int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
180 struct btrfs_root *root,
181 struct extent_buffer *buf,
182 struct extent_buffer *parent, int parent_slot,
183 struct extent_buffer **cow_ret,
184 u64 search_start, u64 empty_size,
188 struct extent_buffer *cow;
191 int different_trans = 0;
194 struct btrfs_key first_key;
199 WARN_ON(!btrfs_tree_locked(buf));
201 if (root->ref_cows) {
202 root_gen = trans->transid;
206 WARN_ON(root->ref_cows && trans->transid !=
207 root->fs_info->running_transaction->transid);
208 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
210 level = btrfs_header_level(buf);
211 nritems = btrfs_header_nritems(buf);
214 btrfs_item_key_to_cpu(buf, &first_key, 0);
216 btrfs_node_key_to_cpu(buf, &first_key, 0);
218 first_key.objectid = 0;
221 struct btrfs_key ins;
223 ins.objectid = prealloc_dest;
224 ins.offset = buf->len;
225 ins.type = BTRFS_EXTENT_ITEM_KEY;
227 ret = btrfs_alloc_reserved_extent(trans, root,
228 root->root_key.objectid,
233 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
236 cow = btrfs_alloc_free_block(trans, root, buf->len,
237 root->root_key.objectid,
238 root_gen, first_key.objectid,
239 level, search_start, empty_size);
244 copy_extent_buffer(cow, buf, 0, 0, cow->len);
245 btrfs_set_header_bytenr(cow, cow->start);
246 btrfs_set_header_generation(cow, trans->transid);
247 btrfs_set_header_owner(cow, root->root_key.objectid);
248 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
250 WARN_ON(btrfs_header_generation(buf) > trans->transid);
251 if (btrfs_header_generation(buf) != trans->transid) {
253 ret = btrfs_inc_ref(trans, root, buf, 1);
257 clean_tree_block(trans, root, buf);
260 if (buf == root->node) {
261 WARN_ON(parent && parent != buf);
262 root_gen = btrfs_header_generation(buf);
264 spin_lock(&root->node_lock);
266 extent_buffer_get(cow);
267 spin_unlock(&root->node_lock);
269 if (buf != root->commit_root) {
270 btrfs_free_extent(trans, root, buf->start,
271 buf->len, root->root_key.objectid,
274 free_extent_buffer(buf);
275 add_root_to_dirty_list(root);
277 root_gen = btrfs_header_generation(parent);
278 btrfs_set_node_blockptr(parent, parent_slot,
280 WARN_ON(trans->transid == 0);
281 btrfs_set_node_ptr_generation(parent, parent_slot,
283 btrfs_mark_buffer_dirty(parent);
284 WARN_ON(btrfs_header_generation(parent) != trans->transid);
285 btrfs_free_extent(trans, root, buf->start, buf->len,
286 btrfs_header_owner(parent), root_gen,
290 btrfs_tree_unlock(buf);
291 free_extent_buffer(buf);
292 btrfs_mark_buffer_dirty(cow);
297 int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
298 struct btrfs_root *root, struct extent_buffer *buf,
299 struct extent_buffer *parent, int parent_slot,
300 struct extent_buffer **cow_ret, u64 prealloc_dest)
306 if (trans->transaction != root->fs_info->running_transaction) {
307 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
308 root->fs_info->running_transaction->transid);
311 if (trans->transid != root->fs_info->generation) {
312 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
313 root->fs_info->generation);
317 header_trans = btrfs_header_generation(buf);
318 spin_lock(&root->fs_info->hash_lock);
319 if (header_trans == trans->transid &&
320 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
322 spin_unlock(&root->fs_info->hash_lock);
323 WARN_ON(prealloc_dest);
326 spin_unlock(&root->fs_info->hash_lock);
327 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
328 ret = __btrfs_cow_block(trans, root, buf, parent,
329 parent_slot, cow_ret, search_start, 0,
334 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
336 if (blocknr < other && other - (blocknr + blocksize) < 32768)
338 if (blocknr > other && blocknr - (other + blocksize) < 32768)
344 * compare two keys in a memcmp fashion
346 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
350 btrfs_disk_key_to_cpu(&k1, disk);
352 if (k1.objectid > k2->objectid)
354 if (k1.objectid < k2->objectid)
356 if (k1.type > k2->type)
358 if (k1.type < k2->type)
360 if (k1.offset > k2->offset)
362 if (k1.offset < k2->offset)
368 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
369 struct btrfs_root *root, struct extent_buffer *parent,
370 int start_slot, int cache_only, u64 *last_ret,
371 struct btrfs_key *progress)
373 struct extent_buffer *cur;
376 u64 search_start = *last_ret;
386 int progress_passed = 0;
387 struct btrfs_disk_key disk_key;
389 parent_level = btrfs_header_level(parent);
390 if (cache_only && parent_level != 1)
393 if (trans->transaction != root->fs_info->running_transaction) {
394 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
395 root->fs_info->running_transaction->transid);
398 if (trans->transid != root->fs_info->generation) {
399 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
400 root->fs_info->generation);
404 parent_nritems = btrfs_header_nritems(parent);
405 blocksize = btrfs_level_size(root, parent_level - 1);
406 end_slot = parent_nritems;
408 if (parent_nritems == 1)
411 for (i = start_slot; i < end_slot; i++) {
414 if (!parent->map_token) {
415 map_extent_buffer(parent,
416 btrfs_node_key_ptr_offset(i),
417 sizeof(struct btrfs_key_ptr),
418 &parent->map_token, &parent->kaddr,
419 &parent->map_start, &parent->map_len,
422 btrfs_node_key(parent, &disk_key, i);
423 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
427 blocknr = btrfs_node_blockptr(parent, i);
428 gen = btrfs_node_ptr_generation(parent, i);
430 last_block = blocknr;
433 other = btrfs_node_blockptr(parent, i - 1);
434 close = close_blocks(blocknr, other, blocksize);
436 if (!close && i < end_slot - 2) {
437 other = btrfs_node_blockptr(parent, i + 1);
438 close = close_blocks(blocknr, other, blocksize);
441 last_block = blocknr;
444 if (parent->map_token) {
445 unmap_extent_buffer(parent, parent->map_token,
447 parent->map_token = NULL;
450 cur = btrfs_find_tree_block(root, blocknr, blocksize);
452 uptodate = btrfs_buffer_uptodate(cur, gen);
455 if (!cur || !uptodate) {
457 free_extent_buffer(cur);
461 cur = read_tree_block(root, blocknr,
463 } else if (!uptodate) {
464 btrfs_read_buffer(cur, gen);
467 if (search_start == 0)
468 search_start = last_block;
470 btrfs_tree_lock(cur);
471 err = __btrfs_cow_block(trans, root, cur, parent, i,
474 (end_slot - i) * blocksize), 0);
476 btrfs_tree_unlock(cur);
477 free_extent_buffer(cur);
480 search_start = cur->start;
481 last_block = cur->start;
482 *last_ret = search_start;
483 btrfs_tree_unlock(cur);
484 free_extent_buffer(cur);
486 if (parent->map_token) {
487 unmap_extent_buffer(parent, parent->map_token,
489 parent->map_token = NULL;
495 * The leaf data grows from end-to-front in the node.
496 * this returns the address of the start of the last item,
497 * which is the stop of the leaf data stack
499 static inline unsigned int leaf_data_end(struct btrfs_root *root,
500 struct extent_buffer *leaf)
502 u32 nr = btrfs_header_nritems(leaf);
504 return BTRFS_LEAF_DATA_SIZE(root);
505 return btrfs_item_offset_nr(leaf, nr - 1);
508 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
511 struct extent_buffer *parent = NULL;
512 struct extent_buffer *node = path->nodes[level];
513 struct btrfs_disk_key parent_key;
514 struct btrfs_disk_key node_key;
517 struct btrfs_key cpukey;
518 u32 nritems = btrfs_header_nritems(node);
520 if (path->nodes[level + 1])
521 parent = path->nodes[level + 1];
523 slot = path->slots[level];
524 BUG_ON(nritems == 0);
526 parent_slot = path->slots[level + 1];
527 btrfs_node_key(parent, &parent_key, parent_slot);
528 btrfs_node_key(node, &node_key, 0);
529 BUG_ON(memcmp(&parent_key, &node_key,
530 sizeof(struct btrfs_disk_key)));
531 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
532 btrfs_header_bytenr(node));
534 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
536 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
537 btrfs_node_key(node, &node_key, slot);
538 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
540 if (slot < nritems - 1) {
541 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
542 btrfs_node_key(node, &node_key, slot);
543 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
548 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
551 struct extent_buffer *leaf = path->nodes[level];
552 struct extent_buffer *parent = NULL;
554 struct btrfs_key cpukey;
555 struct btrfs_disk_key parent_key;
556 struct btrfs_disk_key leaf_key;
557 int slot = path->slots[0];
559 u32 nritems = btrfs_header_nritems(leaf);
561 if (path->nodes[level + 1])
562 parent = path->nodes[level + 1];
568 parent_slot = path->slots[level + 1];
569 btrfs_node_key(parent, &parent_key, parent_slot);
570 btrfs_item_key(leaf, &leaf_key, 0);
572 BUG_ON(memcmp(&parent_key, &leaf_key,
573 sizeof(struct btrfs_disk_key)));
574 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
575 btrfs_header_bytenr(leaf));
578 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
579 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
580 btrfs_item_key(leaf, &leaf_key, i);
581 if (comp_keys(&leaf_key, &cpukey) >= 0) {
582 btrfs_print_leaf(root, leaf);
583 printk("slot %d offset bad key\n", i);
586 if (btrfs_item_offset_nr(leaf, i) !=
587 btrfs_item_end_nr(leaf, i + 1)) {
588 btrfs_print_leaf(root, leaf);
589 printk("slot %d offset bad\n", i);
593 if (btrfs_item_offset_nr(leaf, i) +
594 btrfs_item_size_nr(leaf, i) !=
595 BTRFS_LEAF_DATA_SIZE(root)) {
596 btrfs_print_leaf(root, leaf);
597 printk("slot %d first offset bad\n", i);
603 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
604 btrfs_print_leaf(root, leaf);
605 printk("slot %d bad size \n", nritems - 1);
610 if (slot != 0 && slot < nritems - 1) {
611 btrfs_item_key(leaf, &leaf_key, slot);
612 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
613 if (comp_keys(&leaf_key, &cpukey) <= 0) {
614 btrfs_print_leaf(root, leaf);
615 printk("slot %d offset bad key\n", slot);
618 if (btrfs_item_offset_nr(leaf, slot - 1) !=
619 btrfs_item_end_nr(leaf, slot)) {
620 btrfs_print_leaf(root, leaf);
621 printk("slot %d offset bad\n", slot);
625 if (slot < nritems - 1) {
626 btrfs_item_key(leaf, &leaf_key, slot);
627 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
628 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
629 if (btrfs_item_offset_nr(leaf, slot) !=
630 btrfs_item_end_nr(leaf, slot + 1)) {
631 btrfs_print_leaf(root, leaf);
632 printk("slot %d offset bad\n", slot);
636 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
637 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
641 static int noinline check_block(struct btrfs_root *root,
642 struct btrfs_path *path, int level)
646 if (btrfs_header_level(path->nodes[level]) != level)
647 printk("warning: bad level %Lu wanted %d found %d\n",
648 path->nodes[level]->start, level,
649 btrfs_header_level(path->nodes[level]));
650 found_start = btrfs_header_bytenr(path->nodes[level]);
651 if (found_start != path->nodes[level]->start) {
652 printk("warning: bad bytentr %Lu found %Lu\n",
653 path->nodes[level]->start, found_start);
656 struct extent_buffer *buf = path->nodes[level];
658 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
659 (unsigned long)btrfs_header_fsid(buf),
661 printk("warning bad block %Lu\n", buf->start);
666 return check_leaf(root, path, level);
667 return check_node(root, path, level);
671 * search for key in the extent_buffer. The items start at offset p,
672 * and they are item_size apart. There are 'max' items in p.
674 * the slot in the array is returned via slot, and it points to
675 * the place where you would insert key if it is not found in
678 * slot may point to max if the key is bigger than all of the keys
680 static noinline int generic_bin_search(struct extent_buffer *eb,
682 int item_size, struct btrfs_key *key,
689 struct btrfs_disk_key *tmp = NULL;
690 struct btrfs_disk_key unaligned;
691 unsigned long offset;
692 char *map_token = NULL;
694 unsigned long map_start = 0;
695 unsigned long map_len = 0;
699 mid = (low + high) / 2;
700 offset = p + mid * item_size;
702 if (!map_token || offset < map_start ||
703 (offset + sizeof(struct btrfs_disk_key)) >
704 map_start + map_len) {
706 unmap_extent_buffer(eb, map_token, KM_USER0);
709 err = map_extent_buffer(eb, offset,
710 sizeof(struct btrfs_disk_key),
712 &map_start, &map_len, KM_USER0);
715 tmp = (struct btrfs_disk_key *)(kaddr + offset -
718 read_extent_buffer(eb, &unaligned,
719 offset, sizeof(unaligned));
724 tmp = (struct btrfs_disk_key *)(kaddr + offset -
727 ret = comp_keys(tmp, key);
736 unmap_extent_buffer(eb, map_token, KM_USER0);
742 unmap_extent_buffer(eb, map_token, KM_USER0);
747 * simple bin_search frontend that does the right thing for
750 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
751 int level, int *slot)
754 return generic_bin_search(eb,
755 offsetof(struct btrfs_leaf, items),
756 sizeof(struct btrfs_item),
757 key, btrfs_header_nritems(eb),
760 return generic_bin_search(eb,
761 offsetof(struct btrfs_node, ptrs),
762 sizeof(struct btrfs_key_ptr),
763 key, btrfs_header_nritems(eb),
769 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
770 struct extent_buffer *parent, int slot)
772 int level = btrfs_header_level(parent);
775 if (slot >= btrfs_header_nritems(parent))
780 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
781 btrfs_level_size(root, level - 1),
782 btrfs_node_ptr_generation(parent, slot));
785 static noinline int balance_level(struct btrfs_trans_handle *trans,
786 struct btrfs_root *root,
787 struct btrfs_path *path, int level)
789 struct extent_buffer *right = NULL;
790 struct extent_buffer *mid;
791 struct extent_buffer *left = NULL;
792 struct extent_buffer *parent = NULL;
796 int orig_slot = path->slots[level];
797 int err_on_enospc = 0;
803 mid = path->nodes[level];
804 WARN_ON(!path->locks[level]);
805 WARN_ON(btrfs_header_generation(mid) != trans->transid);
807 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
809 if (level < BTRFS_MAX_LEVEL - 1)
810 parent = path->nodes[level + 1];
811 pslot = path->slots[level + 1];
814 * deal with the case where there is only one pointer in the root
815 * by promoting the node below to a root
818 struct extent_buffer *child;
820 if (btrfs_header_nritems(mid) != 1)
823 /* promote the child to a root */
824 child = read_node_slot(root, mid, 0);
825 btrfs_tree_lock(child);
827 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
830 spin_lock(&root->node_lock);
832 spin_unlock(&root->node_lock);
834 add_root_to_dirty_list(root);
835 btrfs_tree_unlock(child);
836 path->locks[level] = 0;
837 path->nodes[level] = NULL;
838 clean_tree_block(trans, root, mid);
839 btrfs_tree_unlock(mid);
840 /* once for the path */
841 free_extent_buffer(mid);
842 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
843 root->root_key.objectid,
844 btrfs_header_generation(mid), 0, 0, 1);
845 /* once for the root ptr */
846 free_extent_buffer(mid);
849 if (btrfs_header_nritems(mid) >
850 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
853 if (btrfs_header_nritems(mid) < 2)
856 left = read_node_slot(root, parent, pslot - 1);
858 btrfs_tree_lock(left);
859 wret = btrfs_cow_block(trans, root, left,
860 parent, pslot - 1, &left, 0);
866 right = read_node_slot(root, parent, pslot + 1);
868 btrfs_tree_lock(right);
869 wret = btrfs_cow_block(trans, root, right,
870 parent, pslot + 1, &right, 0);
877 /* first, try to make some room in the middle buffer */
879 orig_slot += btrfs_header_nritems(left);
880 wret = push_node_left(trans, root, left, mid, 1);
883 if (btrfs_header_nritems(mid) < 2)
888 * then try to empty the right most buffer into the middle
891 wret = push_node_left(trans, root, mid, right, 1);
892 if (wret < 0 && wret != -ENOSPC)
894 if (btrfs_header_nritems(right) == 0) {
895 u64 bytenr = right->start;
896 u64 generation = btrfs_header_generation(parent);
897 u32 blocksize = right->len;
899 clean_tree_block(trans, root, right);
900 btrfs_tree_unlock(right);
901 free_extent_buffer(right);
903 wret = del_ptr(trans, root, path, level + 1, pslot +
907 wret = btrfs_free_extent(trans, root, bytenr,
909 btrfs_header_owner(parent),
910 generation, 0, 0, 1);
914 struct btrfs_disk_key right_key;
915 btrfs_node_key(right, &right_key, 0);
916 btrfs_set_node_key(parent, &right_key, pslot + 1);
917 btrfs_mark_buffer_dirty(parent);
920 if (btrfs_header_nritems(mid) == 1) {
922 * we're not allowed to leave a node with one item in the
923 * tree during a delete. A deletion from lower in the tree
924 * could try to delete the only pointer in this node.
925 * So, pull some keys from the left.
926 * There has to be a left pointer at this point because
927 * otherwise we would have pulled some pointers from the
931 wret = balance_node_right(trans, root, mid, left);
937 wret = push_node_left(trans, root, left, mid, 1);
943 if (btrfs_header_nritems(mid) == 0) {
944 /* we've managed to empty the middle node, drop it */
945 u64 root_gen = btrfs_header_generation(parent);
946 u64 bytenr = mid->start;
947 u32 blocksize = mid->len;
949 clean_tree_block(trans, root, mid);
950 btrfs_tree_unlock(mid);
951 free_extent_buffer(mid);
953 wret = del_ptr(trans, root, path, level + 1, pslot);
956 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
957 btrfs_header_owner(parent),
962 /* update the parent key to reflect our changes */
963 struct btrfs_disk_key mid_key;
964 btrfs_node_key(mid, &mid_key, 0);
965 btrfs_set_node_key(parent, &mid_key, pslot);
966 btrfs_mark_buffer_dirty(parent);
969 /* update the path */
971 if (btrfs_header_nritems(left) > orig_slot) {
972 extent_buffer_get(left);
973 /* left was locked after cow */
974 path->nodes[level] = left;
975 path->slots[level + 1] -= 1;
976 path->slots[level] = orig_slot;
978 btrfs_tree_unlock(mid);
979 free_extent_buffer(mid);
982 orig_slot -= btrfs_header_nritems(left);
983 path->slots[level] = orig_slot;
986 /* double check we haven't messed things up */
987 check_block(root, path, level);
989 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
993 btrfs_tree_unlock(right);
994 free_extent_buffer(right);
997 if (path->nodes[level] != left)
998 btrfs_tree_unlock(left);
999 free_extent_buffer(left);
1004 /* returns zero if the push worked, non-zero otherwise */
1005 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
1006 struct btrfs_root *root,
1007 struct btrfs_path *path, int level)
1009 struct extent_buffer *right = NULL;
1010 struct extent_buffer *mid;
1011 struct extent_buffer *left = NULL;
1012 struct extent_buffer *parent = NULL;
1016 int orig_slot = path->slots[level];
1022 mid = path->nodes[level];
1023 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1024 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1026 if (level < BTRFS_MAX_LEVEL - 1)
1027 parent = path->nodes[level + 1];
1028 pslot = path->slots[level + 1];
1033 left = read_node_slot(root, parent, pslot - 1);
1035 /* first, try to make some room in the middle buffer */
1039 btrfs_tree_lock(left);
1040 left_nr = btrfs_header_nritems(left);
1041 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1044 ret = btrfs_cow_block(trans, root, left, parent,
1045 pslot - 1, &left, 0);
1049 wret = push_node_left(trans, root,
1056 struct btrfs_disk_key disk_key;
1057 orig_slot += left_nr;
1058 btrfs_node_key(mid, &disk_key, 0);
1059 btrfs_set_node_key(parent, &disk_key, pslot);
1060 btrfs_mark_buffer_dirty(parent);
1061 if (btrfs_header_nritems(left) > orig_slot) {
1062 path->nodes[level] = left;
1063 path->slots[level + 1] -= 1;
1064 path->slots[level] = orig_slot;
1065 btrfs_tree_unlock(mid);
1066 free_extent_buffer(mid);
1069 btrfs_header_nritems(left);
1070 path->slots[level] = orig_slot;
1071 btrfs_tree_unlock(left);
1072 free_extent_buffer(left);
1076 btrfs_tree_unlock(left);
1077 free_extent_buffer(left);
1079 right = read_node_slot(root, parent, pslot + 1);
1082 * then try to empty the right most buffer into the middle
1086 btrfs_tree_lock(right);
1087 right_nr = btrfs_header_nritems(right);
1088 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1091 ret = btrfs_cow_block(trans, root, right,
1097 wret = balance_node_right(trans, root,
1104 struct btrfs_disk_key disk_key;
1106 btrfs_node_key(right, &disk_key, 0);
1107 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1108 btrfs_mark_buffer_dirty(parent);
1110 if (btrfs_header_nritems(mid) <= orig_slot) {
1111 path->nodes[level] = right;
1112 path->slots[level + 1] += 1;
1113 path->slots[level] = orig_slot -
1114 btrfs_header_nritems(mid);
1115 btrfs_tree_unlock(mid);
1116 free_extent_buffer(mid);
1118 btrfs_tree_unlock(right);
1119 free_extent_buffer(right);
1123 btrfs_tree_unlock(right);
1124 free_extent_buffer(right);
1130 * readahead one full node of leaves
1132 static noinline void reada_for_search(struct btrfs_root *root,
1133 struct btrfs_path *path,
1134 int level, int slot, u64 objectid)
1136 struct extent_buffer *node;
1137 struct btrfs_disk_key disk_key;
1143 int direction = path->reada;
1144 struct extent_buffer *eb;
1152 if (!path->nodes[level])
1155 node = path->nodes[level];
1157 search = btrfs_node_blockptr(node, slot);
1158 blocksize = btrfs_level_size(root, level - 1);
1159 eb = btrfs_find_tree_block(root, search, blocksize);
1161 free_extent_buffer(eb);
1165 highest_read = search;
1166 lowest_read = search;
1168 nritems = btrfs_header_nritems(node);
1171 if (direction < 0) {
1175 } else if (direction > 0) {
1180 if (path->reada < 0 && objectid) {
1181 btrfs_node_key(node, &disk_key, nr);
1182 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1185 search = btrfs_node_blockptr(node, nr);
1186 if ((search >= lowest_read && search <= highest_read) ||
1187 (search < lowest_read && lowest_read - search <= 32768) ||
1188 (search > highest_read && search - highest_read <= 32768)) {
1189 readahead_tree_block(root, search, blocksize,
1190 btrfs_node_ptr_generation(node, nr));
1194 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1196 if(nread > (1024 * 1024) || nscan > 128)
1199 if (search < lowest_read)
1200 lowest_read = search;
1201 if (search > highest_read)
1202 highest_read = search;
1206 static noinline void unlock_up(struct btrfs_path *path, int level,
1210 int skip_level = level;
1212 struct extent_buffer *t;
1214 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1215 if (!path->nodes[i])
1217 if (!path->locks[i])
1219 if (!no_skips && path->slots[i] == 0) {
1223 if (!no_skips && path->keep_locks) {
1226 nritems = btrfs_header_nritems(t);
1227 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1232 if (skip_level < i && i >= lowest_unlock)
1236 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1237 btrfs_tree_unlock(t);
1244 * look for key in the tree. path is filled in with nodes along the way
1245 * if key is found, we return zero and you can find the item in the leaf
1246 * level of the path (level 0)
1248 * If the key isn't found, the path points to the slot where it should
1249 * be inserted, and 1 is returned. If there are other errors during the
1250 * search a negative error number is returned.
1252 * if ins_len > 0, nodes and leaves will be split as we walk down the
1253 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1256 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1257 *root, struct btrfs_key *key, struct btrfs_path *p, int
1260 struct extent_buffer *b;
1261 struct extent_buffer *tmp;
1265 int should_reada = p->reada;
1266 int lowest_unlock = 1;
1268 u8 lowest_level = 0;
1271 struct btrfs_key prealloc_block;
1273 lowest_level = p->lowest_level;
1274 WARN_ON(lowest_level && ins_len);
1275 WARN_ON(p->nodes[0] != NULL);
1276 WARN_ON(cow && root == root->fs_info->extent_root &&
1277 !mutex_is_locked(&root->fs_info->alloc_mutex));
1281 prealloc_block.objectid = 0;
1284 if (p->skip_locking)
1285 b = btrfs_root_node(root);
1287 b = btrfs_lock_root_node(root);
1290 level = btrfs_header_level(b);
1293 * setup the path here so we can release it under lock
1294 * contention with the cow code
1296 p->nodes[level] = b;
1297 if (!p->skip_locking)
1298 p->locks[level] = 1;
1303 /* is a cow on this block not required */
1304 spin_lock(&root->fs_info->hash_lock);
1305 if (btrfs_header_generation(b) == trans->transid &&
1306 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1307 spin_unlock(&root->fs_info->hash_lock);
1310 spin_unlock(&root->fs_info->hash_lock);
1312 /* ok, we have to cow, is our old prealloc the right
1315 if (prealloc_block.objectid &&
1316 prealloc_block.offset != b->len) {
1317 btrfs_free_reserved_extent(root,
1318 prealloc_block.objectid,
1319 prealloc_block.offset);
1320 prealloc_block.objectid = 0;
1324 * for higher level blocks, try not to allocate blocks
1325 * with the block and the parent locks held.
1327 if (level > 1 && !prealloc_block.objectid &&
1328 btrfs_path_lock_waiting(p, level)) {
1330 u64 hint = b->start;
1332 btrfs_release_path(root, p);
1333 ret = btrfs_reserve_extent(trans, root,
1336 &prealloc_block, 0);
1341 wret = btrfs_cow_block(trans, root, b,
1342 p->nodes[level + 1],
1343 p->slots[level + 1],
1344 &b, prealloc_block.objectid);
1345 prealloc_block.objectid = 0;
1347 free_extent_buffer(b);
1353 BUG_ON(!cow && ins_len);
1354 if (level != btrfs_header_level(b))
1356 level = btrfs_header_level(b);
1358 p->nodes[level] = b;
1359 if (!p->skip_locking)
1360 p->locks[level] = 1;
1362 ret = check_block(root, p, level);
1368 ret = bin_search(b, key, level, &slot);
1370 if (ret && slot > 0)
1372 p->slots[level] = slot;
1373 if (ins_len > 0 && btrfs_header_nritems(b) >=
1374 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1375 int sret = split_node(trans, root, p, level);
1381 b = p->nodes[level];
1382 slot = p->slots[level];
1383 } else if (ins_len < 0) {
1384 int sret = balance_level(trans, root, p,
1390 b = p->nodes[level];
1392 btrfs_release_path(NULL, p);
1395 slot = p->slots[level];
1396 BUG_ON(btrfs_header_nritems(b) == 1);
1398 unlock_up(p, level, lowest_unlock);
1400 /* this is only true while dropping a snapshot */
1401 if (level == lowest_level) {
1405 blocknr = btrfs_node_blockptr(b, slot);
1406 gen = btrfs_node_ptr_generation(b, slot);
1407 blocksize = btrfs_level_size(root, level - 1);
1409 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1410 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1414 * reduce lock contention at high levels
1415 * of the btree by dropping locks before
1419 btrfs_release_path(NULL, p);
1421 free_extent_buffer(tmp);
1423 reada_for_search(root, p,
1427 tmp = read_tree_block(root, blocknr,
1430 free_extent_buffer(tmp);
1434 free_extent_buffer(tmp);
1436 reada_for_search(root, p,
1439 b = read_node_slot(root, b, slot);
1442 if (!p->skip_locking)
1445 p->slots[level] = slot;
1446 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1447 sizeof(struct btrfs_item) + ins_len) {
1448 int sret = split_leaf(trans, root, key,
1449 p, ins_len, ret == 0);
1456 unlock_up(p, level, lowest_unlock);
1462 if (prealloc_block.objectid) {
1463 btrfs_free_reserved_extent(root,
1464 prealloc_block.objectid,
1465 prealloc_block.offset);
1472 * adjust the pointers going up the tree, starting at level
1473 * making sure the right key of each node is points to 'key'.
1474 * This is used after shifting pointers to the left, so it stops
1475 * fixing up pointers when a given leaf/node is not in slot 0 of the
1478 * If this fails to write a tree block, it returns -1, but continues
1479 * fixing up the blocks in ram so the tree is consistent.
1481 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root, struct btrfs_path *path,
1483 struct btrfs_disk_key *key, int level)
1487 struct extent_buffer *t;
1489 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1490 int tslot = path->slots[i];
1491 if (!path->nodes[i])
1494 btrfs_set_node_key(t, key, tslot);
1495 btrfs_mark_buffer_dirty(path->nodes[i]);
1503 * try to push data from one node into the next node left in the
1506 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1507 * error, and > 0 if there was no room in the left hand block.
1509 static int push_node_left(struct btrfs_trans_handle *trans,
1510 struct btrfs_root *root, struct extent_buffer *dst,
1511 struct extent_buffer *src, int empty)
1518 src_nritems = btrfs_header_nritems(src);
1519 dst_nritems = btrfs_header_nritems(dst);
1520 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1521 WARN_ON(btrfs_header_generation(src) != trans->transid);
1522 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1524 if (!empty && src_nritems <= 8)
1527 if (push_items <= 0) {
1532 push_items = min(src_nritems, push_items);
1533 if (push_items < src_nritems) {
1534 /* leave at least 8 pointers in the node if
1535 * we aren't going to empty it
1537 if (src_nritems - push_items < 8) {
1538 if (push_items <= 8)
1544 push_items = min(src_nritems - 8, push_items);
1546 copy_extent_buffer(dst, src,
1547 btrfs_node_key_ptr_offset(dst_nritems),
1548 btrfs_node_key_ptr_offset(0),
1549 push_items * sizeof(struct btrfs_key_ptr));
1551 if (push_items < src_nritems) {
1552 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1553 btrfs_node_key_ptr_offset(push_items),
1554 (src_nritems - push_items) *
1555 sizeof(struct btrfs_key_ptr));
1557 btrfs_set_header_nritems(src, src_nritems - push_items);
1558 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1559 btrfs_mark_buffer_dirty(src);
1560 btrfs_mark_buffer_dirty(dst);
1565 * try to push data from one node into the next node right in the
1568 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1569 * error, and > 0 if there was no room in the right hand block.
1571 * this will only push up to 1/2 the contents of the left node over
1573 static int balance_node_right(struct btrfs_trans_handle *trans,
1574 struct btrfs_root *root,
1575 struct extent_buffer *dst,
1576 struct extent_buffer *src)
1584 WARN_ON(btrfs_header_generation(src) != trans->transid);
1585 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1587 src_nritems = btrfs_header_nritems(src);
1588 dst_nritems = btrfs_header_nritems(dst);
1589 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1590 if (push_items <= 0) {
1594 if (src_nritems < 4) {
1598 max_push = src_nritems / 2 + 1;
1599 /* don't try to empty the node */
1600 if (max_push >= src_nritems) {
1604 if (max_push < push_items)
1605 push_items = max_push;
1607 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1608 btrfs_node_key_ptr_offset(0),
1610 sizeof(struct btrfs_key_ptr));
1612 copy_extent_buffer(dst, src,
1613 btrfs_node_key_ptr_offset(0),
1614 btrfs_node_key_ptr_offset(src_nritems - push_items),
1615 push_items * sizeof(struct btrfs_key_ptr));
1617 btrfs_set_header_nritems(src, src_nritems - push_items);
1618 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1620 btrfs_mark_buffer_dirty(src);
1621 btrfs_mark_buffer_dirty(dst);
1626 * helper function to insert a new root level in the tree.
1627 * A new node is allocated, and a single item is inserted to
1628 * point to the existing root
1630 * returns zero on success or < 0 on failure.
1632 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1633 struct btrfs_root *root,
1634 struct btrfs_path *path, int level)
1638 struct extent_buffer *lower;
1639 struct extent_buffer *c;
1640 struct extent_buffer *old;
1641 struct btrfs_disk_key lower_key;
1643 BUG_ON(path->nodes[level]);
1644 BUG_ON(path->nodes[level-1] != root->node);
1647 root_gen = trans->transid;
1651 lower = path->nodes[level-1];
1653 btrfs_item_key(lower, &lower_key, 0);
1655 btrfs_node_key(lower, &lower_key, 0);
1657 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1658 root->root_key.objectid,
1659 root_gen, lower_key.objectid, level,
1660 root->node->start, 0);
1664 memset_extent_buffer(c, 0, 0, root->nodesize);
1665 btrfs_set_header_nritems(c, 1);
1666 btrfs_set_header_level(c, level);
1667 btrfs_set_header_bytenr(c, c->start);
1668 btrfs_set_header_generation(c, trans->transid);
1669 btrfs_set_header_owner(c, root->root_key.objectid);
1671 write_extent_buffer(c, root->fs_info->fsid,
1672 (unsigned long)btrfs_header_fsid(c),
1675 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1676 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1679 btrfs_set_node_key(c, &lower_key, 0);
1680 btrfs_set_node_blockptr(c, 0, lower->start);
1681 lower_gen = btrfs_header_generation(lower);
1682 WARN_ON(lower_gen == 0);
1684 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1686 btrfs_mark_buffer_dirty(c);
1688 spin_lock(&root->node_lock);
1691 spin_unlock(&root->node_lock);
1693 /* the super has an extra ref to root->node */
1694 free_extent_buffer(old);
1696 add_root_to_dirty_list(root);
1697 extent_buffer_get(c);
1698 path->nodes[level] = c;
1699 path->locks[level] = 1;
1700 path->slots[level] = 0;
1702 if (root->ref_cows && lower_gen != trans->transid) {
1703 struct btrfs_path *back_path = btrfs_alloc_path();
1705 mutex_lock(&root->fs_info->alloc_mutex);
1706 ret = btrfs_insert_extent_backref(trans,
1707 root->fs_info->extent_root,
1709 root->root_key.objectid,
1710 trans->transid, 0, 0);
1712 mutex_unlock(&root->fs_info->alloc_mutex);
1713 btrfs_free_path(back_path);
1719 * worker function to insert a single pointer in a node.
1720 * the node should have enough room for the pointer already
1722 * slot and level indicate where you want the key to go, and
1723 * blocknr is the block the key points to.
1725 * returns zero on success and < 0 on any error
1727 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1728 *root, struct btrfs_path *path, struct btrfs_disk_key
1729 *key, u64 bytenr, int slot, int level)
1731 struct extent_buffer *lower;
1734 BUG_ON(!path->nodes[level]);
1735 lower = path->nodes[level];
1736 nritems = btrfs_header_nritems(lower);
1739 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1741 if (slot != nritems) {
1742 memmove_extent_buffer(lower,
1743 btrfs_node_key_ptr_offset(slot + 1),
1744 btrfs_node_key_ptr_offset(slot),
1745 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1747 btrfs_set_node_key(lower, key, slot);
1748 btrfs_set_node_blockptr(lower, slot, bytenr);
1749 WARN_ON(trans->transid == 0);
1750 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1751 btrfs_set_header_nritems(lower, nritems + 1);
1752 btrfs_mark_buffer_dirty(lower);
1757 * split the node at the specified level in path in two.
1758 * The path is corrected to point to the appropriate node after the split
1760 * Before splitting this tries to make some room in the node by pushing
1761 * left and right, if either one works, it returns right away.
1763 * returns 0 on success and < 0 on failure
1765 static noinline int split_node(struct btrfs_trans_handle *trans,
1766 struct btrfs_root *root,
1767 struct btrfs_path *path, int level)
1770 struct extent_buffer *c;
1771 struct extent_buffer *split;
1772 struct btrfs_disk_key disk_key;
1778 c = path->nodes[level];
1779 WARN_ON(btrfs_header_generation(c) != trans->transid);
1780 if (c == root->node) {
1781 /* trying to split the root, lets make a new one */
1782 ret = insert_new_root(trans, root, path, level + 1);
1786 ret = push_nodes_for_insert(trans, root, path, level);
1787 c = path->nodes[level];
1788 if (!ret && btrfs_header_nritems(c) <
1789 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1795 c_nritems = btrfs_header_nritems(c);
1797 root_gen = trans->transid;
1801 btrfs_node_key(c, &disk_key, 0);
1802 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1803 root->root_key.objectid,
1805 btrfs_disk_key_objectid(&disk_key),
1806 level, c->start, 0);
1808 return PTR_ERR(split);
1810 btrfs_set_header_flags(split, btrfs_header_flags(c));
1811 btrfs_set_header_level(split, btrfs_header_level(c));
1812 btrfs_set_header_bytenr(split, split->start);
1813 btrfs_set_header_generation(split, trans->transid);
1814 btrfs_set_header_owner(split, root->root_key.objectid);
1815 btrfs_set_header_flags(split, 0);
1816 write_extent_buffer(split, root->fs_info->fsid,
1817 (unsigned long)btrfs_header_fsid(split),
1819 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1820 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1823 mid = (c_nritems + 1) / 2;
1825 copy_extent_buffer(split, c,
1826 btrfs_node_key_ptr_offset(0),
1827 btrfs_node_key_ptr_offset(mid),
1828 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1829 btrfs_set_header_nritems(split, c_nritems - mid);
1830 btrfs_set_header_nritems(c, mid);
1833 btrfs_mark_buffer_dirty(c);
1834 btrfs_mark_buffer_dirty(split);
1836 btrfs_node_key(split, &disk_key, 0);
1837 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1838 path->slots[level + 1] + 1,
1843 if (path->slots[level] >= mid) {
1844 path->slots[level] -= mid;
1845 btrfs_tree_unlock(c);
1846 free_extent_buffer(c);
1847 path->nodes[level] = split;
1848 path->slots[level + 1] += 1;
1850 btrfs_tree_unlock(split);
1851 free_extent_buffer(split);
1857 * how many bytes are required to store the items in a leaf. start
1858 * and nr indicate which items in the leaf to check. This totals up the
1859 * space used both by the item structs and the item data
1861 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1864 int nritems = btrfs_header_nritems(l);
1865 int end = min(nritems, start + nr) - 1;
1869 data_len = btrfs_item_end_nr(l, start);
1870 data_len = data_len - btrfs_item_offset_nr(l, end);
1871 data_len += sizeof(struct btrfs_item) * nr;
1872 WARN_ON(data_len < 0);
1877 * The space between the end of the leaf items and
1878 * the start of the leaf data. IOW, how much room
1879 * the leaf has left for both items and data
1881 int noinline btrfs_leaf_free_space(struct btrfs_root *root,
1882 struct extent_buffer *leaf)
1884 int nritems = btrfs_header_nritems(leaf);
1886 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1888 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1889 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1890 leaf_space_used(leaf, 0, nritems), nritems);
1896 * push some data in the path leaf to the right, trying to free up at
1897 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1899 * returns 1 if the push failed because the other node didn't have enough
1900 * room, 0 if everything worked out and < 0 if there were major errors.
1902 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1903 *root, struct btrfs_path *path, int data_size,
1906 struct extent_buffer *left = path->nodes[0];
1907 struct extent_buffer *right;
1908 struct extent_buffer *upper;
1909 struct btrfs_disk_key disk_key;
1915 struct btrfs_item *item;
1923 slot = path->slots[1];
1924 if (!path->nodes[1]) {
1927 upper = path->nodes[1];
1928 if (slot >= btrfs_header_nritems(upper) - 1)
1931 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1933 right = read_node_slot(root, upper, slot + 1);
1934 btrfs_tree_lock(right);
1935 free_space = btrfs_leaf_free_space(root, right);
1936 if (free_space < data_size + sizeof(struct btrfs_item))
1939 /* cow and double check */
1940 ret = btrfs_cow_block(trans, root, right, upper,
1941 slot + 1, &right, 0);
1945 free_space = btrfs_leaf_free_space(root, right);
1946 if (free_space < data_size + sizeof(struct btrfs_item))
1949 left_nritems = btrfs_header_nritems(left);
1950 if (left_nritems == 0)
1958 i = left_nritems - 1;
1960 item = btrfs_item_nr(left, i);
1962 if (path->slots[0] == i)
1963 push_space += data_size + sizeof(*item);
1965 if (!left->map_token) {
1966 map_extent_buffer(left, (unsigned long)item,
1967 sizeof(struct btrfs_item),
1968 &left->map_token, &left->kaddr,
1969 &left->map_start, &left->map_len,
1973 this_item_size = btrfs_item_size(left, item);
1974 if (this_item_size + sizeof(*item) + push_space > free_space)
1977 push_space += this_item_size + sizeof(*item);
1982 if (left->map_token) {
1983 unmap_extent_buffer(left, left->map_token, KM_USER1);
1984 left->map_token = NULL;
1987 if (push_items == 0)
1990 if (!empty && push_items == left_nritems)
1993 /* push left to right */
1994 right_nritems = btrfs_header_nritems(right);
1996 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1997 push_space -= leaf_data_end(root, left);
1999 /* make room in the right data area */
2000 data_end = leaf_data_end(root, right);
2001 memmove_extent_buffer(right,
2002 btrfs_leaf_data(right) + data_end - push_space,
2003 btrfs_leaf_data(right) + data_end,
2004 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2006 /* copy from the left data area */
2007 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2008 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2009 btrfs_leaf_data(left) + leaf_data_end(root, left),
2012 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2013 btrfs_item_nr_offset(0),
2014 right_nritems * sizeof(struct btrfs_item));
2016 /* copy the items from left to right */
2017 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2018 btrfs_item_nr_offset(left_nritems - push_items),
2019 push_items * sizeof(struct btrfs_item));
2021 /* update the item pointers */
2022 right_nritems += push_items;
2023 btrfs_set_header_nritems(right, right_nritems);
2024 push_space = BTRFS_LEAF_DATA_SIZE(root);
2025 for (i = 0; i < right_nritems; i++) {
2026 item = btrfs_item_nr(right, i);
2027 if (!right->map_token) {
2028 map_extent_buffer(right, (unsigned long)item,
2029 sizeof(struct btrfs_item),
2030 &right->map_token, &right->kaddr,
2031 &right->map_start, &right->map_len,
2034 push_space -= btrfs_item_size(right, item);
2035 btrfs_set_item_offset(right, item, push_space);
2038 if (right->map_token) {
2039 unmap_extent_buffer(right, right->map_token, KM_USER1);
2040 right->map_token = NULL;
2042 left_nritems -= push_items;
2043 btrfs_set_header_nritems(left, left_nritems);
2046 btrfs_mark_buffer_dirty(left);
2047 btrfs_mark_buffer_dirty(right);
2049 btrfs_item_key(right, &disk_key, 0);
2050 btrfs_set_node_key(upper, &disk_key, slot + 1);
2051 btrfs_mark_buffer_dirty(upper);
2053 /* then fixup the leaf pointer in the path */
2054 if (path->slots[0] >= left_nritems) {
2055 path->slots[0] -= left_nritems;
2056 if (btrfs_header_nritems(path->nodes[0]) == 0)
2057 clean_tree_block(trans, root, path->nodes[0]);
2058 btrfs_tree_unlock(path->nodes[0]);
2059 free_extent_buffer(path->nodes[0]);
2060 path->nodes[0] = right;
2061 path->slots[1] += 1;
2063 btrfs_tree_unlock(right);
2064 free_extent_buffer(right);
2069 btrfs_tree_unlock(right);
2070 free_extent_buffer(right);
2075 * push some data in the path leaf to the left, trying to free up at
2076 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2078 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2079 *root, struct btrfs_path *path, int data_size,
2082 struct btrfs_disk_key disk_key;
2083 struct extent_buffer *right = path->nodes[0];
2084 struct extent_buffer *left;
2090 struct btrfs_item *item;
2091 u32 old_left_nritems;
2097 u32 old_left_item_size;
2099 slot = path->slots[1];
2102 if (!path->nodes[1])
2105 right_nritems = btrfs_header_nritems(right);
2106 if (right_nritems == 0) {
2110 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2112 left = read_node_slot(root, path->nodes[1], slot - 1);
2113 btrfs_tree_lock(left);
2114 free_space = btrfs_leaf_free_space(root, left);
2115 if (free_space < data_size + sizeof(struct btrfs_item)) {
2120 /* cow and double check */
2121 ret = btrfs_cow_block(trans, root, left,
2122 path->nodes[1], slot - 1, &left, 0);
2124 /* we hit -ENOSPC, but it isn't fatal here */
2129 free_space = btrfs_leaf_free_space(root, left);
2130 if (free_space < data_size + sizeof(struct btrfs_item)) {
2138 nr = right_nritems - 1;
2140 for (i = 0; i < nr; i++) {
2141 item = btrfs_item_nr(right, i);
2142 if (!right->map_token) {
2143 map_extent_buffer(right, (unsigned long)item,
2144 sizeof(struct btrfs_item),
2145 &right->map_token, &right->kaddr,
2146 &right->map_start, &right->map_len,
2150 if (path->slots[0] == i)
2151 push_space += data_size + sizeof(*item);
2153 this_item_size = btrfs_item_size(right, item);
2154 if (this_item_size + sizeof(*item) + push_space > free_space)
2158 push_space += this_item_size + sizeof(*item);
2161 if (right->map_token) {
2162 unmap_extent_buffer(right, right->map_token, KM_USER1);
2163 right->map_token = NULL;
2166 if (push_items == 0) {
2170 if (!empty && push_items == btrfs_header_nritems(right))
2173 /* push data from right to left */
2174 copy_extent_buffer(left, right,
2175 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2176 btrfs_item_nr_offset(0),
2177 push_items * sizeof(struct btrfs_item));
2179 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2180 btrfs_item_offset_nr(right, push_items -1);
2182 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2183 leaf_data_end(root, left) - push_space,
2184 btrfs_leaf_data(right) +
2185 btrfs_item_offset_nr(right, push_items - 1),
2187 old_left_nritems = btrfs_header_nritems(left);
2188 BUG_ON(old_left_nritems < 0);
2190 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2191 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2194 item = btrfs_item_nr(left, i);
2195 if (!left->map_token) {
2196 map_extent_buffer(left, (unsigned long)item,
2197 sizeof(struct btrfs_item),
2198 &left->map_token, &left->kaddr,
2199 &left->map_start, &left->map_len,
2203 ioff = btrfs_item_offset(left, item);
2204 btrfs_set_item_offset(left, item,
2205 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2207 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2208 if (left->map_token) {
2209 unmap_extent_buffer(left, left->map_token, KM_USER1);
2210 left->map_token = NULL;
2213 /* fixup right node */
2214 if (push_items > right_nritems) {
2215 printk("push items %d nr %u\n", push_items, right_nritems);
2219 if (push_items < right_nritems) {
2220 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2221 leaf_data_end(root, right);
2222 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2223 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2224 btrfs_leaf_data(right) +
2225 leaf_data_end(root, right), push_space);
2227 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2228 btrfs_item_nr_offset(push_items),
2229 (btrfs_header_nritems(right) - push_items) *
2230 sizeof(struct btrfs_item));
2232 right_nritems -= push_items;
2233 btrfs_set_header_nritems(right, right_nritems);
2234 push_space = BTRFS_LEAF_DATA_SIZE(root);
2235 for (i = 0; i < right_nritems; i++) {
2236 item = btrfs_item_nr(right, i);
2238 if (!right->map_token) {
2239 map_extent_buffer(right, (unsigned long)item,
2240 sizeof(struct btrfs_item),
2241 &right->map_token, &right->kaddr,
2242 &right->map_start, &right->map_len,
2246 push_space = push_space - btrfs_item_size(right, item);
2247 btrfs_set_item_offset(right, item, push_space);
2249 if (right->map_token) {
2250 unmap_extent_buffer(right, right->map_token, KM_USER1);
2251 right->map_token = NULL;
2254 btrfs_mark_buffer_dirty(left);
2256 btrfs_mark_buffer_dirty(right);
2258 btrfs_item_key(right, &disk_key, 0);
2259 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2263 /* then fixup the leaf pointer in the path */
2264 if (path->slots[0] < push_items) {
2265 path->slots[0] += old_left_nritems;
2266 if (btrfs_header_nritems(path->nodes[0]) == 0)
2267 clean_tree_block(trans, root, path->nodes[0]);
2268 btrfs_tree_unlock(path->nodes[0]);
2269 free_extent_buffer(path->nodes[0]);
2270 path->nodes[0] = left;
2271 path->slots[1] -= 1;
2273 btrfs_tree_unlock(left);
2274 free_extent_buffer(left);
2275 path->slots[0] -= push_items;
2277 BUG_ON(path->slots[0] < 0);
2280 btrfs_tree_unlock(left);
2281 free_extent_buffer(left);
2286 * split the path's leaf in two, making sure there is at least data_size
2287 * available for the resulting leaf level of the path.
2289 * returns 0 if all went well and < 0 on failure.
2291 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2292 struct btrfs_root *root,
2293 struct btrfs_key *ins_key,
2294 struct btrfs_path *path, int data_size,
2298 struct extent_buffer *l;
2302 struct extent_buffer *right;
2303 int space_needed = data_size + sizeof(struct btrfs_item);
2310 int num_doubles = 0;
2311 struct btrfs_disk_key disk_key;
2314 space_needed = data_size;
2317 root_gen = trans->transid;
2321 /* first try to make some room by pushing left and right */
2322 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2323 wret = push_leaf_right(trans, root, path, data_size, 0);
2328 wret = push_leaf_left(trans, root, path, data_size, 0);
2334 /* did the pushes work? */
2335 if (btrfs_leaf_free_space(root, l) >= space_needed)
2339 if (!path->nodes[1]) {
2340 ret = insert_new_root(trans, root, path, 1);
2347 slot = path->slots[0];
2348 nritems = btrfs_header_nritems(l);
2349 mid = (nritems + 1)/ 2;
2351 btrfs_item_key(l, &disk_key, 0);
2353 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2354 root->root_key.objectid,
2355 root_gen, disk_key.objectid, 0,
2357 if (IS_ERR(right)) {
2359 return PTR_ERR(right);
2362 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2363 btrfs_set_header_bytenr(right, right->start);
2364 btrfs_set_header_generation(right, trans->transid);
2365 btrfs_set_header_owner(right, root->root_key.objectid);
2366 btrfs_set_header_level(right, 0);
2367 write_extent_buffer(right, root->fs_info->fsid,
2368 (unsigned long)btrfs_header_fsid(right),
2371 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2372 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2376 leaf_space_used(l, mid, nritems - mid) + space_needed >
2377 BTRFS_LEAF_DATA_SIZE(root)) {
2378 if (slot >= nritems) {
2379 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2380 btrfs_set_header_nritems(right, 0);
2381 wret = insert_ptr(trans, root, path,
2382 &disk_key, right->start,
2383 path->slots[1] + 1, 1);
2387 btrfs_tree_unlock(path->nodes[0]);
2388 free_extent_buffer(path->nodes[0]);
2389 path->nodes[0] = right;
2391 path->slots[1] += 1;
2392 btrfs_mark_buffer_dirty(right);
2396 if (mid != nritems &&
2397 leaf_space_used(l, mid, nritems - mid) +
2398 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2403 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2404 BTRFS_LEAF_DATA_SIZE(root)) {
2405 if (!extend && slot == 0) {
2406 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2407 btrfs_set_header_nritems(right, 0);
2408 wret = insert_ptr(trans, root, path,
2414 btrfs_tree_unlock(path->nodes[0]);
2415 free_extent_buffer(path->nodes[0]);
2416 path->nodes[0] = right;
2418 if (path->slots[1] == 0) {
2419 wret = fixup_low_keys(trans, root,
2420 path, &disk_key, 1);
2424 btrfs_mark_buffer_dirty(right);
2426 } else if (extend && slot == 0) {
2430 if (mid != nritems &&
2431 leaf_space_used(l, mid, nritems - mid) +
2432 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2438 nritems = nritems - mid;
2439 btrfs_set_header_nritems(right, nritems);
2440 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2442 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2443 btrfs_item_nr_offset(mid),
2444 nritems * sizeof(struct btrfs_item));
2446 copy_extent_buffer(right, l,
2447 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2448 data_copy_size, btrfs_leaf_data(l) +
2449 leaf_data_end(root, l), data_copy_size);
2451 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2452 btrfs_item_end_nr(l, mid);
2454 for (i = 0; i < nritems; i++) {
2455 struct btrfs_item *item = btrfs_item_nr(right, i);
2458 if (!right->map_token) {
2459 map_extent_buffer(right, (unsigned long)item,
2460 sizeof(struct btrfs_item),
2461 &right->map_token, &right->kaddr,
2462 &right->map_start, &right->map_len,
2466 ioff = btrfs_item_offset(right, item);
2467 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2470 if (right->map_token) {
2471 unmap_extent_buffer(right, right->map_token, KM_USER1);
2472 right->map_token = NULL;
2475 btrfs_set_header_nritems(l, mid);
2477 btrfs_item_key(right, &disk_key, 0);
2478 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2479 path->slots[1] + 1, 1);
2483 btrfs_mark_buffer_dirty(right);
2484 btrfs_mark_buffer_dirty(l);
2485 BUG_ON(path->slots[0] != slot);
2488 btrfs_tree_unlock(path->nodes[0]);
2489 free_extent_buffer(path->nodes[0]);
2490 path->nodes[0] = right;
2491 path->slots[0] -= mid;
2492 path->slots[1] += 1;
2494 btrfs_tree_unlock(right);
2495 free_extent_buffer(right);
2498 BUG_ON(path->slots[0] < 0);
2501 BUG_ON(num_doubles != 0);
2508 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 struct btrfs_path *path,
2511 u32 new_size, int from_end)
2516 struct extent_buffer *leaf;
2517 struct btrfs_item *item;
2519 unsigned int data_end;
2520 unsigned int old_data_start;
2521 unsigned int old_size;
2522 unsigned int size_diff;
2525 slot_orig = path->slots[0];
2526 leaf = path->nodes[0];
2527 slot = path->slots[0];
2529 old_size = btrfs_item_size_nr(leaf, slot);
2530 if (old_size == new_size)
2533 nritems = btrfs_header_nritems(leaf);
2534 data_end = leaf_data_end(root, leaf);
2536 old_data_start = btrfs_item_offset_nr(leaf, slot);
2538 size_diff = old_size - new_size;
2541 BUG_ON(slot >= nritems);
2544 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2546 /* first correct the data pointers */
2547 for (i = slot; i < nritems; i++) {
2549 item = btrfs_item_nr(leaf, i);
2551 if (!leaf->map_token) {
2552 map_extent_buffer(leaf, (unsigned long)item,
2553 sizeof(struct btrfs_item),
2554 &leaf->map_token, &leaf->kaddr,
2555 &leaf->map_start, &leaf->map_len,
2559 ioff = btrfs_item_offset(leaf, item);
2560 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2563 if (leaf->map_token) {
2564 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2565 leaf->map_token = NULL;
2568 /* shift the data */
2570 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2571 data_end + size_diff, btrfs_leaf_data(leaf) +
2572 data_end, old_data_start + new_size - data_end);
2574 struct btrfs_disk_key disk_key;
2577 btrfs_item_key(leaf, &disk_key, slot);
2579 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2581 struct btrfs_file_extent_item *fi;
2583 fi = btrfs_item_ptr(leaf, slot,
2584 struct btrfs_file_extent_item);
2585 fi = (struct btrfs_file_extent_item *)(
2586 (unsigned long)fi - size_diff);
2588 if (btrfs_file_extent_type(leaf, fi) ==
2589 BTRFS_FILE_EXTENT_INLINE) {
2590 ptr = btrfs_item_ptr_offset(leaf, slot);
2591 memmove_extent_buffer(leaf, ptr,
2593 offsetof(struct btrfs_file_extent_item,
2598 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2599 data_end + size_diff, btrfs_leaf_data(leaf) +
2600 data_end, old_data_start - data_end);
2602 offset = btrfs_disk_key_offset(&disk_key);
2603 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2604 btrfs_set_item_key(leaf, &disk_key, slot);
2606 fixup_low_keys(trans, root, path, &disk_key, 1);
2609 item = btrfs_item_nr(leaf, slot);
2610 btrfs_set_item_size(leaf, item, new_size);
2611 btrfs_mark_buffer_dirty(leaf);
2614 if (btrfs_leaf_free_space(root, leaf) < 0) {
2615 btrfs_print_leaf(root, leaf);
2621 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2622 struct btrfs_root *root, struct btrfs_path *path,
2628 struct extent_buffer *leaf;
2629 struct btrfs_item *item;
2631 unsigned int data_end;
2632 unsigned int old_data;
2633 unsigned int old_size;
2636 slot_orig = path->slots[0];
2637 leaf = path->nodes[0];
2639 nritems = btrfs_header_nritems(leaf);
2640 data_end = leaf_data_end(root, leaf);
2642 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2643 btrfs_print_leaf(root, leaf);
2646 slot = path->slots[0];
2647 old_data = btrfs_item_end_nr(leaf, slot);
2650 if (slot >= nritems) {
2651 btrfs_print_leaf(root, leaf);
2652 printk("slot %d too large, nritems %d\n", slot, nritems);
2657 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2659 /* first correct the data pointers */
2660 for (i = slot; i < nritems; i++) {
2662 item = btrfs_item_nr(leaf, i);
2664 if (!leaf->map_token) {
2665 map_extent_buffer(leaf, (unsigned long)item,
2666 sizeof(struct btrfs_item),
2667 &leaf->map_token, &leaf->kaddr,
2668 &leaf->map_start, &leaf->map_len,
2671 ioff = btrfs_item_offset(leaf, item);
2672 btrfs_set_item_offset(leaf, item, ioff - data_size);
2675 if (leaf->map_token) {
2676 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2677 leaf->map_token = NULL;
2680 /* shift the data */
2681 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2682 data_end - data_size, btrfs_leaf_data(leaf) +
2683 data_end, old_data - data_end);
2685 data_end = old_data;
2686 old_size = btrfs_item_size_nr(leaf, slot);
2687 item = btrfs_item_nr(leaf, slot);
2688 btrfs_set_item_size(leaf, item, old_size + data_size);
2689 btrfs_mark_buffer_dirty(leaf);
2692 if (btrfs_leaf_free_space(root, leaf) < 0) {
2693 btrfs_print_leaf(root, leaf);
2700 * Given a key and some data, insert an item into the tree.
2701 * This does all the path init required, making room in the tree if needed.
2703 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2704 struct btrfs_root *root,
2705 struct btrfs_path *path,
2706 struct btrfs_key *cpu_key, u32 *data_size,
2709 struct extent_buffer *leaf;
2710 struct btrfs_item *item;
2718 unsigned int data_end;
2719 struct btrfs_disk_key disk_key;
2721 for (i = 0; i < nr; i++) {
2722 total_data += data_size[i];
2725 total_size = total_data + (nr * sizeof(struct btrfs_item));
2726 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2733 slot_orig = path->slots[0];
2734 leaf = path->nodes[0];
2736 nritems = btrfs_header_nritems(leaf);
2737 data_end = leaf_data_end(root, leaf);
2739 if (btrfs_leaf_free_space(root, leaf) <
2740 sizeof(struct btrfs_item) + total_size) {
2741 btrfs_print_leaf(root, leaf);
2742 printk("not enough freespace need %u have %d\n",
2743 total_size, btrfs_leaf_free_space(root, leaf));
2747 slot = path->slots[0];
2750 if (slot != nritems) {
2752 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2754 if (old_data < data_end) {
2755 btrfs_print_leaf(root, leaf);
2756 printk("slot %d old_data %d data_end %d\n",
2757 slot, old_data, data_end);
2761 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2763 /* first correct the data pointers */
2764 WARN_ON(leaf->map_token);
2765 for (i = slot; i < nritems; i++) {
2768 item = btrfs_item_nr(leaf, i);
2769 if (!leaf->map_token) {
2770 map_extent_buffer(leaf, (unsigned long)item,
2771 sizeof(struct btrfs_item),
2772 &leaf->map_token, &leaf->kaddr,
2773 &leaf->map_start, &leaf->map_len,
2777 ioff = btrfs_item_offset(leaf, item);
2778 btrfs_set_item_offset(leaf, item, ioff - total_data);
2780 if (leaf->map_token) {
2781 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2782 leaf->map_token = NULL;
2785 /* shift the items */
2786 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2787 btrfs_item_nr_offset(slot),
2788 (nritems - slot) * sizeof(struct btrfs_item));
2790 /* shift the data */
2791 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2792 data_end - total_data, btrfs_leaf_data(leaf) +
2793 data_end, old_data - data_end);
2794 data_end = old_data;
2797 /* setup the item for the new data */
2798 for (i = 0; i < nr; i++) {
2799 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2800 btrfs_set_item_key(leaf, &disk_key, slot + i);
2801 item = btrfs_item_nr(leaf, slot + i);
2802 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2803 data_end -= data_size[i];
2804 btrfs_set_item_size(leaf, item, data_size[i]);
2806 btrfs_set_header_nritems(leaf, nritems + nr);
2807 btrfs_mark_buffer_dirty(leaf);
2811 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2812 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2815 if (btrfs_leaf_free_space(root, leaf) < 0) {
2816 btrfs_print_leaf(root, leaf);
2824 * Given a key and some data, insert an item into the tree.
2825 * This does all the path init required, making room in the tree if needed.
2827 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2828 *root, struct btrfs_key *cpu_key, void *data, u32
2832 struct btrfs_path *path;
2833 struct extent_buffer *leaf;
2836 path = btrfs_alloc_path();
2838 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2840 leaf = path->nodes[0];
2841 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2842 write_extent_buffer(leaf, data, ptr, data_size);
2843 btrfs_mark_buffer_dirty(leaf);
2845 btrfs_free_path(path);
2850 * delete the pointer from a given node.
2852 * If the delete empties a node, the node is removed from the tree,
2853 * continuing all the way the root if required. The root is converted into
2854 * a leaf if all the nodes are emptied.
2856 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2857 struct btrfs_path *path, int level, int slot)
2859 struct extent_buffer *parent = path->nodes[level];
2864 nritems = btrfs_header_nritems(parent);
2865 if (slot != nritems -1) {
2866 memmove_extent_buffer(parent,
2867 btrfs_node_key_ptr_offset(slot),
2868 btrfs_node_key_ptr_offset(slot + 1),
2869 sizeof(struct btrfs_key_ptr) *
2870 (nritems - slot - 1));
2873 btrfs_set_header_nritems(parent, nritems);
2874 if (nritems == 0 && parent == root->node) {
2875 BUG_ON(btrfs_header_level(root->node) != 1);
2876 /* just turn the root into a leaf and break */
2877 btrfs_set_header_level(root->node, 0);
2878 } else if (slot == 0) {
2879 struct btrfs_disk_key disk_key;
2881 btrfs_node_key(parent, &disk_key, 0);
2882 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2886 btrfs_mark_buffer_dirty(parent);
2891 * delete the item at the leaf level in path. If that empties
2892 * the leaf, remove it from the tree
2894 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2895 struct btrfs_path *path, int slot, int nr)
2897 struct extent_buffer *leaf;
2898 struct btrfs_item *item;
2906 leaf = path->nodes[0];
2907 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2909 for (i = 0; i < nr; i++)
2910 dsize += btrfs_item_size_nr(leaf, slot + i);
2912 nritems = btrfs_header_nritems(leaf);
2914 if (slot + nr != nritems) {
2916 int data_end = leaf_data_end(root, leaf);
2918 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2920 btrfs_leaf_data(leaf) + data_end,
2921 last_off - data_end);
2923 for (i = slot + nr; i < nritems; i++) {
2926 item = btrfs_item_nr(leaf, i);
2927 if (!leaf->map_token) {
2928 map_extent_buffer(leaf, (unsigned long)item,
2929 sizeof(struct btrfs_item),
2930 &leaf->map_token, &leaf->kaddr,
2931 &leaf->map_start, &leaf->map_len,
2934 ioff = btrfs_item_offset(leaf, item);
2935 btrfs_set_item_offset(leaf, item, ioff + dsize);
2938 if (leaf->map_token) {
2939 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2940 leaf->map_token = NULL;
2943 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2944 btrfs_item_nr_offset(slot + nr),
2945 sizeof(struct btrfs_item) *
2946 (nritems - slot - nr));
2948 btrfs_set_header_nritems(leaf, nritems - nr);
2951 /* delete the leaf if we've emptied it */
2953 if (leaf == root->node) {
2954 btrfs_set_header_level(leaf, 0);
2956 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2957 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2960 wret = btrfs_free_extent(trans, root,
2961 leaf->start, leaf->len,
2962 btrfs_header_owner(path->nodes[1]),
2968 int used = leaf_space_used(leaf, 0, nritems);
2970 struct btrfs_disk_key disk_key;
2972 btrfs_item_key(leaf, &disk_key, 0);
2973 wret = fixup_low_keys(trans, root, path,
2979 /* delete the leaf if it is mostly empty */
2980 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2981 /* push_leaf_left fixes the path.
2982 * make sure the path still points to our leaf
2983 * for possible call to del_ptr below
2985 slot = path->slots[1];
2986 extent_buffer_get(leaf);
2988 wret = push_leaf_left(trans, root, path, 1, 1);
2989 if (wret < 0 && wret != -ENOSPC)
2992 if (path->nodes[0] == leaf &&
2993 btrfs_header_nritems(leaf)) {
2994 wret = push_leaf_right(trans, root, path, 1, 1);
2995 if (wret < 0 && wret != -ENOSPC)
2999 if (btrfs_header_nritems(leaf) == 0) {
3001 u64 bytenr = leaf->start;
3002 u32 blocksize = leaf->len;
3004 root_gen = btrfs_header_generation(
3007 wret = del_ptr(trans, root, path, 1, slot);
3011 free_extent_buffer(leaf);
3012 wret = btrfs_free_extent(trans, root, bytenr,
3014 btrfs_header_owner(path->nodes[1]),
3019 /* if we're still in the path, make sure
3020 * we're dirty. Otherwise, one of the
3021 * push_leaf functions must have already
3022 * dirtied this buffer
3024 if (path->nodes[0] == leaf)
3025 btrfs_mark_buffer_dirty(leaf);
3026 free_extent_buffer(leaf);
3029 btrfs_mark_buffer_dirty(leaf);
3036 * search the tree again to find a leaf with lesser keys
3037 * returns 0 if it found something or 1 if there are no lesser leaves.
3038 * returns < 0 on io errors.
3040 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3042 struct btrfs_key key;
3043 struct btrfs_disk_key found_key;
3046 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3050 else if (key.type > 0)
3052 else if (key.objectid > 0)
3057 btrfs_release_path(root, path);
3058 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3061 btrfs_item_key(path->nodes[0], &found_key, 0);
3062 ret = comp_keys(&found_key, &key);
3069 * A helper function to walk down the tree starting at min_key, and looking
3070 * for nodes or leaves that are either in cache or have a minimum
3071 * transaction id. This is used by the btree defrag code, but could
3072 * also be used to search for blocks that have changed since a given
3075 * This does not cow, but it does stuff the starting key it finds back
3076 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3077 * key and get a writable path.
3079 * This does lock as it descends, and path->keep_locks should be set
3080 * to 1 by the caller.
3082 * This honors path->lowest_level to prevent descent past a given level
3085 * returns zero if something useful was found, < 0 on error and 1 if there
3086 * was nothing in the tree that matched the search criteria.
3088 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3089 struct btrfs_key *max_key,
3090 struct btrfs_path *path, int cache_only,
3093 struct extent_buffer *cur;
3094 struct btrfs_key found_key;
3102 cur = btrfs_lock_root_node(root);
3103 level = btrfs_header_level(cur);
3104 WARN_ON(path->nodes[level]);
3105 path->nodes[level] = cur;
3106 path->locks[level] = 1;
3108 if (btrfs_header_generation(cur) < min_trans) {
3113 nritems = btrfs_header_nritems(cur);
3114 level = btrfs_header_level(cur);
3115 sret = bin_search(cur, min_key, level, &slot);
3117 /* at level = 0, we're done, setup the path and exit */
3119 if (slot >= nritems)
3122 path->slots[level] = slot;
3123 btrfs_item_key_to_cpu(cur, &found_key, slot);
3126 if (sret && slot > 0)
3129 * check this node pointer against the cache_only and
3130 * min_trans parameters. If it isn't in cache or is too
3131 * old, skip to the next one.
3133 while(slot < nritems) {
3136 struct extent_buffer *tmp;
3137 struct btrfs_disk_key disk_key;
3139 blockptr = btrfs_node_blockptr(cur, slot);
3140 gen = btrfs_node_ptr_generation(cur, slot);
3141 if (gen < min_trans) {
3149 btrfs_node_key(cur, &disk_key, slot);
3150 if (comp_keys(&disk_key, max_key) >= 0) {
3156 tmp = btrfs_find_tree_block(root, blockptr,
3157 btrfs_level_size(root, level - 1));
3159 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3160 free_extent_buffer(tmp);
3164 free_extent_buffer(tmp);
3169 * we didn't find a candidate key in this node, walk forward
3170 * and find another one
3172 if (slot >= nritems) {
3173 path->slots[level] = slot;
3174 sret = btrfs_find_next_key(root, path, min_key, level,
3175 cache_only, min_trans);
3177 btrfs_release_path(root, path);
3183 /* save our key for returning back */
3184 btrfs_node_key_to_cpu(cur, &found_key, slot);
3185 path->slots[level] = slot;
3186 if (level == path->lowest_level) {
3188 unlock_up(path, level, 1);
3191 cur = read_node_slot(root, cur, slot);
3193 btrfs_tree_lock(cur);
3194 path->locks[level - 1] = 1;
3195 path->nodes[level - 1] = cur;
3196 unlock_up(path, level, 1);
3200 memcpy(min_key, &found_key, sizeof(found_key));
3205 * this is similar to btrfs_next_leaf, but does not try to preserve
3206 * and fixup the path. It looks for and returns the next key in the
3207 * tree based on the current path and the cache_only and min_trans
3210 * 0 is returned if another key is found, < 0 if there are any errors
3211 * and 1 is returned if there are no higher keys in the tree
3213 * path->keep_locks should be set to 1 on the search made before
3214 * calling this function.
3216 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3217 struct btrfs_key *key, int lowest_level,
3218 int cache_only, u64 min_trans)
3220 int level = lowest_level;
3222 struct extent_buffer *c;
3224 while(level < BTRFS_MAX_LEVEL) {
3225 if (!path->nodes[level])
3228 slot = path->slots[level] + 1;
3229 c = path->nodes[level];
3231 if (slot >= btrfs_header_nritems(c)) {
3233 if (level == BTRFS_MAX_LEVEL) {
3239 btrfs_item_key_to_cpu(c, key, slot);
3241 u64 blockptr = btrfs_node_blockptr(c, slot);
3242 u64 gen = btrfs_node_ptr_generation(c, slot);
3245 struct extent_buffer *cur;
3246 cur = btrfs_find_tree_block(root, blockptr,
3247 btrfs_level_size(root, level - 1));
3248 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3251 free_extent_buffer(cur);
3254 free_extent_buffer(cur);
3256 if (gen < min_trans) {
3260 btrfs_node_key_to_cpu(c, key, slot);
3268 * search the tree again to find a leaf with greater keys
3269 * returns 0 if it found something or 1 if there are no greater leaves.
3270 * returns < 0 on io errors.
3272 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3276 struct extent_buffer *c;
3277 struct extent_buffer *next = NULL;
3278 struct btrfs_key key;
3282 nritems = btrfs_header_nritems(path->nodes[0]);
3287 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3289 btrfs_release_path(root, path);
3290 path->keep_locks = 1;
3291 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3292 path->keep_locks = 0;
3297 nritems = btrfs_header_nritems(path->nodes[0]);
3299 * by releasing the path above we dropped all our locks. A balance
3300 * could have added more items next to the key that used to be
3301 * at the very end of the block. So, check again here and
3302 * advance the path if there are now more items available.
3304 if (nritems > 0 && path->slots[0] < nritems - 1) {
3309 while(level < BTRFS_MAX_LEVEL) {
3310 if (!path->nodes[level])
3313 slot = path->slots[level] + 1;
3314 c = path->nodes[level];
3315 if (slot >= btrfs_header_nritems(c)) {
3317 if (level == BTRFS_MAX_LEVEL) {
3324 btrfs_tree_unlock(next);
3325 free_extent_buffer(next);
3328 if (level == 1 && (path->locks[1] || path->skip_locking) &&
3330 reada_for_search(root, path, level, slot, 0);
3332 next = read_node_slot(root, c, slot);
3333 if (!path->skip_locking) {
3334 WARN_ON(!btrfs_tree_locked(c));
3335 btrfs_tree_lock(next);
3339 path->slots[level] = slot;
3342 c = path->nodes[level];
3343 if (path->locks[level])
3344 btrfs_tree_unlock(c);
3345 free_extent_buffer(c);
3346 path->nodes[level] = next;
3347 path->slots[level] = 0;
3348 if (!path->skip_locking)
3349 path->locks[level] = 1;
3352 if (level == 1 && path->locks[1] && path->reada)
3353 reada_for_search(root, path, level, slot, 0);
3354 next = read_node_slot(root, next, 0);
3355 if (!path->skip_locking) {
3356 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3357 btrfs_tree_lock(next);
3361 unlock_up(path, 0, 1);
3366 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3367 * searching until it gets past min_objectid or finds an item of 'type'
3369 * returns 0 if something is found, 1 if nothing was found and < 0 on error
3371 int btrfs_previous_item(struct btrfs_root *root,
3372 struct btrfs_path *path, u64 min_objectid,
3375 struct btrfs_key found_key;
3376 struct extent_buffer *leaf;
3381 if (path->slots[0] == 0) {
3382 ret = btrfs_prev_leaf(root, path);
3388 leaf = path->nodes[0];
3389 nritems = btrfs_header_nritems(leaf);
3392 if (path->slots[0] == nritems)
3395 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3396 if (found_key.type == type)
3398 if (found_key.objectid < min_objectid)
3400 if (found_key.objectid == min_objectid &&
3401 found_key.type < type)