Merge branch 'topic/asoc' into for-linus
[linux-2.6] / fs / btrfs / ctree.c
1 /*
2  * Copyright (C) 2007,2008 Oracle.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18
19 #include <linux/sched.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "transaction.h"
23 #include "print-tree.h"
24 #include "locking.h"
25
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);
40
41 inline void btrfs_init_path(struct btrfs_path *p)
42 {
43         memset(p, 0, sizeof(*p));
44 }
45
46 struct btrfs_path *btrfs_alloc_path(void)
47 {
48         struct btrfs_path *path;
49         path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
50         if (path) {
51                 btrfs_init_path(path);
52                 path->reada = 1;
53         }
54         return path;
55 }
56
57 /* this also releases the path */
58 void btrfs_free_path(struct btrfs_path *p)
59 {
60         btrfs_release_path(NULL, p);
61         kmem_cache_free(btrfs_path_cachep, p);
62 }
63
64 /*
65  * path release drops references on the extent buffers in the path
66  * and it drops any locks held by this path
67  *
68  * It is safe to call this on paths that no locks or extent buffers held.
69  */
70 noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
71 {
72         int i;
73
74         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
75                 p->slots[i] = 0;
76                 if (!p->nodes[i])
77                         continue;
78                 if (p->locks[i]) {
79                         btrfs_tree_unlock(p->nodes[i]);
80                         p->locks[i] = 0;
81                 }
82                 free_extent_buffer(p->nodes[i]);
83                 p->nodes[i] = NULL;
84         }
85 }
86
87 /*
88  * safely gets a reference on the root node of a tree.  A lock
89  * is not taken, so a concurrent writer may put a different node
90  * at the root of the tree.  See btrfs_lock_root_node for the
91  * looping required.
92  *
93  * The extent buffer returned by this has a reference taken, so
94  * it won't disappear.  It may stop being the root of the tree
95  * at any time because there are no locks held.
96  */
97 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
98 {
99         struct extent_buffer *eb;
100         spin_lock(&root->node_lock);
101         eb = root->node;
102         extent_buffer_get(eb);
103         spin_unlock(&root->node_lock);
104         return eb;
105 }
106
107 /* loop around taking references on and locking the root node of the
108  * tree until you end up with a lock on the root.  A locked buffer
109  * is returned, with a reference held.
110  */
111 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
112 {
113         struct extent_buffer *eb;
114
115         while (1) {
116                 eb = btrfs_root_node(root);
117                 btrfs_tree_lock(eb);
118
119                 spin_lock(&root->node_lock);
120                 if (eb == root->node) {
121                         spin_unlock(&root->node_lock);
122                         break;
123                 }
124                 spin_unlock(&root->node_lock);
125
126                 btrfs_tree_unlock(eb);
127                 free_extent_buffer(eb);
128         }
129         return eb;
130 }
131
132 /* cowonly root (everything not a reference counted cow subvolume), just get
133  * put onto a simple dirty list.  transaction.c walks this to make sure they
134  * get properly updated on disk.
135  */
136 static void add_root_to_dirty_list(struct btrfs_root *root)
137 {
138         if (root->track_dirty && list_empty(&root->dirty_list)) {
139                 list_add(&root->dirty_list,
140                          &root->fs_info->dirty_cowonly_roots);
141         }
142 }
143
144 /*
145  * used by snapshot creation to make a copy of a root for a tree with
146  * a given objectid.  The buffer with the new root node is returned in
147  * cow_ret, and this func returns zero on success or a negative error code.
148  */
149 int btrfs_copy_root(struct btrfs_trans_handle *trans,
150                       struct btrfs_root *root,
151                       struct extent_buffer *buf,
152                       struct extent_buffer **cow_ret, u64 new_root_objectid)
153 {
154         struct extent_buffer *cow;
155         u32 nritems;
156         int ret = 0;
157         int level;
158         struct btrfs_root *new_root;
159
160         new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
161         if (!new_root)
162                 return -ENOMEM;
163
164         memcpy(new_root, root, sizeof(*new_root));
165         new_root->root_key.objectid = new_root_objectid;
166
167         WARN_ON(root->ref_cows && trans->transid !=
168                 root->fs_info->running_transaction->transid);
169         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
170
171         level = btrfs_header_level(buf);
172         nritems = btrfs_header_nritems(buf);
173
174         cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
175                                      new_root_objectid, trans->transid,
176                                      level, buf->start, 0);
177         if (IS_ERR(cow)) {
178                 kfree(new_root);
179                 return PTR_ERR(cow);
180         }
181
182         copy_extent_buffer(cow, buf, 0, 0, cow->len);
183         btrfs_set_header_bytenr(cow, cow->start);
184         btrfs_set_header_generation(cow, trans->transid);
185         btrfs_set_header_owner(cow, new_root_objectid);
186         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
187
188         write_extent_buffer(cow, root->fs_info->fsid,
189                             (unsigned long)btrfs_header_fsid(cow),
190                             BTRFS_FSID_SIZE);
191
192         WARN_ON(btrfs_header_generation(buf) > trans->transid);
193         ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
194         kfree(new_root);
195
196         if (ret)
197                 return ret;
198
199         btrfs_mark_buffer_dirty(cow);
200         *cow_ret = cow;
201         return 0;
202 }
203
204 /*
205  * does the dirty work in cow of a single block.  The parent block (if
206  * supplied) is updated to point to the new cow copy.  The new buffer is marked
207  * dirty and returned locked.  If you modify the block it needs to be marked
208  * dirty again.
209  *
210  * search_start -- an allocation hint for the new block
211  *
212  * empty_size -- a hint that you plan on doing more cow.  This is the size in
213  * bytes the allocator should try to find free next to the block it returns.
214  * This is just a hint and may be ignored by the allocator.
215  *
216  * prealloc_dest -- if you have already reserved a destination for the cow,
217  * this uses that block instead of allocating a new one.
218  * btrfs_alloc_reserved_extent is used to finish the allocation.
219  */
220 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
221                              struct btrfs_root *root,
222                              struct extent_buffer *buf,
223                              struct extent_buffer *parent, int parent_slot,
224                              struct extent_buffer **cow_ret,
225                              u64 search_start, u64 empty_size,
226                              u64 prealloc_dest)
227 {
228         u64 parent_start;
229         struct extent_buffer *cow;
230         u32 nritems;
231         int ret = 0;
232         int level;
233         int unlock_orig = 0;
234
235         if (*cow_ret == buf)
236                 unlock_orig = 1;
237
238         WARN_ON(!btrfs_tree_locked(buf));
239
240         if (parent)
241                 parent_start = parent->start;
242         else
243                 parent_start = 0;
244
245         WARN_ON(root->ref_cows && trans->transid !=
246                 root->fs_info->running_transaction->transid);
247         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
248
249         level = btrfs_header_level(buf);
250         nritems = btrfs_header_nritems(buf);
251
252         if (prealloc_dest) {
253                 struct btrfs_key ins;
254
255                 ins.objectid = prealloc_dest;
256                 ins.offset = buf->len;
257                 ins.type = BTRFS_EXTENT_ITEM_KEY;
258
259                 ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
260                                                   root->root_key.objectid,
261                                                   trans->transid, level, &ins);
262                 BUG_ON(ret);
263                 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
264                                             buf->len);
265         } else {
266                 cow = btrfs_alloc_free_block(trans, root, buf->len,
267                                              parent_start,
268                                              root->root_key.objectid,
269                                              trans->transid, level,
270                                              search_start, empty_size);
271         }
272         if (IS_ERR(cow))
273                 return PTR_ERR(cow);
274
275         copy_extent_buffer(cow, buf, 0, 0, cow->len);
276         btrfs_set_header_bytenr(cow, cow->start);
277         btrfs_set_header_generation(cow, trans->transid);
278         btrfs_set_header_owner(cow, root->root_key.objectid);
279         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
280
281         write_extent_buffer(cow, root->fs_info->fsid,
282                             (unsigned long)btrfs_header_fsid(cow),
283                             BTRFS_FSID_SIZE);
284
285         WARN_ON(btrfs_header_generation(buf) > trans->transid);
286         if (btrfs_header_generation(buf) != trans->transid) {
287                 u32 nr_extents;
288                 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
289                 if (ret)
290                         return ret;
291
292                 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
293                 WARN_ON(ret);
294         } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
295                 /*
296                  * There are only two places that can drop reference to
297                  * tree blocks owned by living reloc trees, one is here,
298                  * the other place is btrfs_drop_subtree. In both places,
299                  * we check reference count while tree block is locked.
300                  * Furthermore, if reference count is one, it won't get
301                  * increased by someone else.
302                  */
303                 u32 refs;
304                 ret = btrfs_lookup_extent_ref(trans, root, buf->start,
305                                               buf->len, &refs);
306                 BUG_ON(ret);
307                 if (refs == 1) {
308                         ret = btrfs_update_ref(trans, root, buf, cow,
309                                                0, nritems);
310                         clean_tree_block(trans, root, buf);
311                 } else {
312                         ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
313                 }
314                 BUG_ON(ret);
315         } else {
316                 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
317                 if (ret)
318                         return ret;
319                 clean_tree_block(trans, root, buf);
320         }
321
322         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
323                 ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
324                 WARN_ON(ret);
325         }
326
327         if (buf == root->node) {
328                 WARN_ON(parent && parent != buf);
329
330                 spin_lock(&root->node_lock);
331                 root->node = cow;
332                 extent_buffer_get(cow);
333                 spin_unlock(&root->node_lock);
334
335                 if (buf != root->commit_root) {
336                         btrfs_free_extent(trans, root, buf->start,
337                                           buf->len, buf->start,
338                                           root->root_key.objectid,
339                                           btrfs_header_generation(buf),
340                                           level, 1);
341                 }
342                 free_extent_buffer(buf);
343                 add_root_to_dirty_list(root);
344         } else {
345                 btrfs_set_node_blockptr(parent, parent_slot,
346                                         cow->start);
347                 WARN_ON(trans->transid == 0);
348                 btrfs_set_node_ptr_generation(parent, parent_slot,
349                                               trans->transid);
350                 btrfs_mark_buffer_dirty(parent);
351                 WARN_ON(btrfs_header_generation(parent) != trans->transid);
352                 btrfs_free_extent(trans, root, buf->start, buf->len,
353                                   parent_start, btrfs_header_owner(parent),
354                                   btrfs_header_generation(parent), level, 1);
355         }
356         if (unlock_orig)
357                 btrfs_tree_unlock(buf);
358         free_extent_buffer(buf);
359         btrfs_mark_buffer_dirty(cow);
360         *cow_ret = cow;
361         return 0;
362 }
363
364 /*
365  * cows a single block, see __btrfs_cow_block for the real work.
366  * This version of it has extra checks so that a block isn't cow'd more than
367  * once per transaction, as long as it hasn't been written yet
368  */
369 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
370                     struct btrfs_root *root, struct extent_buffer *buf,
371                     struct extent_buffer *parent, int parent_slot,
372                     struct extent_buffer **cow_ret, u64 prealloc_dest)
373 {
374         u64 search_start;
375         int ret;
376
377         if (trans->transaction != root->fs_info->running_transaction) {
378                 printk(KERN_CRIT "trans %llu running %llu\n",
379                        (unsigned long long)trans->transid,
380                        (unsigned long long)
381                        root->fs_info->running_transaction->transid);
382                 WARN_ON(1);
383         }
384         if (trans->transid != root->fs_info->generation) {
385                 printk(KERN_CRIT "trans %llu running %llu\n",
386                        (unsigned long long)trans->transid,
387                        (unsigned long long)root->fs_info->generation);
388                 WARN_ON(1);
389         }
390
391         spin_lock(&root->fs_info->hash_lock);
392         if (btrfs_header_generation(buf) == trans->transid &&
393             btrfs_header_owner(buf) == root->root_key.objectid &&
394             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
395                 *cow_ret = buf;
396                 spin_unlock(&root->fs_info->hash_lock);
397                 WARN_ON(prealloc_dest);
398                 return 0;
399         }
400         spin_unlock(&root->fs_info->hash_lock);
401         search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
402         ret = __btrfs_cow_block(trans, root, buf, parent,
403                                  parent_slot, cow_ret, search_start, 0,
404                                  prealloc_dest);
405         return ret;
406 }
407
408 /*
409  * helper function for defrag to decide if two blocks pointed to by a
410  * node are actually close by
411  */
412 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
413 {
414         if (blocknr < other && other - (blocknr + blocksize) < 32768)
415                 return 1;
416         if (blocknr > other && blocknr - (other + blocksize) < 32768)
417                 return 1;
418         return 0;
419 }
420
421 /*
422  * compare two keys in a memcmp fashion
423  */
424 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
425 {
426         struct btrfs_key k1;
427
428         btrfs_disk_key_to_cpu(&k1, disk);
429
430         if (k1.objectid > k2->objectid)
431                 return 1;
432         if (k1.objectid < k2->objectid)
433                 return -1;
434         if (k1.type > k2->type)
435                 return 1;
436         if (k1.type < k2->type)
437                 return -1;
438         if (k1.offset > k2->offset)
439                 return 1;
440         if (k1.offset < k2->offset)
441                 return -1;
442         return 0;
443 }
444
445 /*
446  * same as comp_keys only with two btrfs_key's
447  */
448 static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
449 {
450         if (k1->objectid > k2->objectid)
451                 return 1;
452         if (k1->objectid < k2->objectid)
453                 return -1;
454         if (k1->type > k2->type)
455                 return 1;
456         if (k1->type < k2->type)
457                 return -1;
458         if (k1->offset > k2->offset)
459                 return 1;
460         if (k1->offset < k2->offset)
461                 return -1;
462         return 0;
463 }
464
465 /*
466  * this is used by the defrag code to go through all the
467  * leaves pointed to by a node and reallocate them so that
468  * disk order is close to key order
469  */
470 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
471                        struct btrfs_root *root, struct extent_buffer *parent,
472                        int start_slot, int cache_only, u64 *last_ret,
473                        struct btrfs_key *progress)
474 {
475         struct extent_buffer *cur;
476         u64 blocknr;
477         u64 gen;
478         u64 search_start = *last_ret;
479         u64 last_block = 0;
480         u64 other;
481         u32 parent_nritems;
482         int end_slot;
483         int i;
484         int err = 0;
485         int parent_level;
486         int uptodate;
487         u32 blocksize;
488         int progress_passed = 0;
489         struct btrfs_disk_key disk_key;
490
491         parent_level = btrfs_header_level(parent);
492         if (cache_only && parent_level != 1)
493                 return 0;
494
495         if (trans->transaction != root->fs_info->running_transaction)
496                 WARN_ON(1);
497         if (trans->transid != root->fs_info->generation)
498                 WARN_ON(1);
499
500         parent_nritems = btrfs_header_nritems(parent);
501         blocksize = btrfs_level_size(root, parent_level - 1);
502         end_slot = parent_nritems;
503
504         if (parent_nritems == 1)
505                 return 0;
506
507         for (i = start_slot; i < end_slot; i++) {
508                 int close = 1;
509
510                 if (!parent->map_token) {
511                         map_extent_buffer(parent,
512                                         btrfs_node_key_ptr_offset(i),
513                                         sizeof(struct btrfs_key_ptr),
514                                         &parent->map_token, &parent->kaddr,
515                                         &parent->map_start, &parent->map_len,
516                                         KM_USER1);
517                 }
518                 btrfs_node_key(parent, &disk_key, i);
519                 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
520                         continue;
521
522                 progress_passed = 1;
523                 blocknr = btrfs_node_blockptr(parent, i);
524                 gen = btrfs_node_ptr_generation(parent, i);
525                 if (last_block == 0)
526                         last_block = blocknr;
527
528                 if (i > 0) {
529                         other = btrfs_node_blockptr(parent, i - 1);
530                         close = close_blocks(blocknr, other, blocksize);
531                 }
532                 if (!close && i < end_slot - 2) {
533                         other = btrfs_node_blockptr(parent, i + 1);
534                         close = close_blocks(blocknr, other, blocksize);
535                 }
536                 if (close) {
537                         last_block = blocknr;
538                         continue;
539                 }
540                 if (parent->map_token) {
541                         unmap_extent_buffer(parent, parent->map_token,
542                                             KM_USER1);
543                         parent->map_token = NULL;
544                 }
545
546                 cur = btrfs_find_tree_block(root, blocknr, blocksize);
547                 if (cur)
548                         uptodate = btrfs_buffer_uptodate(cur, gen);
549                 else
550                         uptodate = 0;
551                 if (!cur || !uptodate) {
552                         if (cache_only) {
553                                 free_extent_buffer(cur);
554                                 continue;
555                         }
556                         if (!cur) {
557                                 cur = read_tree_block(root, blocknr,
558                                                          blocksize, gen);
559                         } else if (!uptodate) {
560                                 btrfs_read_buffer(cur, gen);
561                         }
562                 }
563                 if (search_start == 0)
564                         search_start = last_block;
565
566                 btrfs_tree_lock(cur);
567                 err = __btrfs_cow_block(trans, root, cur, parent, i,
568                                         &cur, search_start,
569                                         min(16 * blocksize,
570                                             (end_slot - i) * blocksize), 0);
571                 if (err) {
572                         btrfs_tree_unlock(cur);
573                         free_extent_buffer(cur);
574                         break;
575                 }
576                 search_start = cur->start;
577                 last_block = cur->start;
578                 *last_ret = search_start;
579                 btrfs_tree_unlock(cur);
580                 free_extent_buffer(cur);
581         }
582         if (parent->map_token) {
583                 unmap_extent_buffer(parent, parent->map_token,
584                                     KM_USER1);
585                 parent->map_token = NULL;
586         }
587         return err;
588 }
589
590 /*
591  * The leaf data grows from end-to-front in the node.
592  * this returns the address of the start of the last item,
593  * which is the stop of the leaf data stack
594  */
595 static inline unsigned int leaf_data_end(struct btrfs_root *root,
596                                          struct extent_buffer *leaf)
597 {
598         u32 nr = btrfs_header_nritems(leaf);
599         if (nr == 0)
600                 return BTRFS_LEAF_DATA_SIZE(root);
601         return btrfs_item_offset_nr(leaf, nr - 1);
602 }
603
604 /*
605  * extra debugging checks to make sure all the items in a key are
606  * well formed and in the proper order
607  */
608 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
609                       int level)
610 {
611         struct extent_buffer *parent = NULL;
612         struct extent_buffer *node = path->nodes[level];
613         struct btrfs_disk_key parent_key;
614         struct btrfs_disk_key node_key;
615         int parent_slot;
616         int slot;
617         struct btrfs_key cpukey;
618         u32 nritems = btrfs_header_nritems(node);
619
620         if (path->nodes[level + 1])
621                 parent = path->nodes[level + 1];
622
623         slot = path->slots[level];
624         BUG_ON(nritems == 0);
625         if (parent) {
626                 parent_slot = path->slots[level + 1];
627                 btrfs_node_key(parent, &parent_key, parent_slot);
628                 btrfs_node_key(node, &node_key, 0);
629                 BUG_ON(memcmp(&parent_key, &node_key,
630                               sizeof(struct btrfs_disk_key)));
631                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
632                        btrfs_header_bytenr(node));
633         }
634         BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
635         if (slot != 0) {
636                 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
637                 btrfs_node_key(node, &node_key, slot);
638                 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
639         }
640         if (slot < nritems - 1) {
641                 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
642                 btrfs_node_key(node, &node_key, slot);
643                 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
644         }
645         return 0;
646 }
647
648 /*
649  * extra checking to make sure all the items in a leaf are
650  * well formed and in the proper order
651  */
652 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
653                       int level)
654 {
655         struct extent_buffer *leaf = path->nodes[level];
656         struct extent_buffer *parent = NULL;
657         int parent_slot;
658         struct btrfs_key cpukey;
659         struct btrfs_disk_key parent_key;
660         struct btrfs_disk_key leaf_key;
661         int slot = path->slots[0];
662
663         u32 nritems = btrfs_header_nritems(leaf);
664
665         if (path->nodes[level + 1])
666                 parent = path->nodes[level + 1];
667
668         if (nritems == 0)
669                 return 0;
670
671         if (parent) {
672                 parent_slot = path->slots[level + 1];
673                 btrfs_node_key(parent, &parent_key, parent_slot);
674                 btrfs_item_key(leaf, &leaf_key, 0);
675
676                 BUG_ON(memcmp(&parent_key, &leaf_key,
677                        sizeof(struct btrfs_disk_key)));
678                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
679                        btrfs_header_bytenr(leaf));
680         }
681         if (slot != 0 && slot < nritems - 1) {
682                 btrfs_item_key(leaf, &leaf_key, slot);
683                 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
684                 if (comp_keys(&leaf_key, &cpukey) <= 0) {
685                         btrfs_print_leaf(root, leaf);
686                         printk(KERN_CRIT "slot %d offset bad key\n", slot);
687                         BUG_ON(1);
688                 }
689                 if (btrfs_item_offset_nr(leaf, slot - 1) !=
690                        btrfs_item_end_nr(leaf, slot)) {
691                         btrfs_print_leaf(root, leaf);
692                         printk(KERN_CRIT "slot %d offset bad\n", slot);
693                         BUG_ON(1);
694                 }
695         }
696         if (slot < nritems - 1) {
697                 btrfs_item_key(leaf, &leaf_key, slot);
698                 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
699                 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
700                 if (btrfs_item_offset_nr(leaf, slot) !=
701                         btrfs_item_end_nr(leaf, slot + 1)) {
702                         btrfs_print_leaf(root, leaf);
703                         printk(KERN_CRIT "slot %d offset bad\n", slot);
704                         BUG_ON(1);
705                 }
706         }
707         BUG_ON(btrfs_item_offset_nr(leaf, 0) +
708                btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
709         return 0;
710 }
711
712 static noinline int check_block(struct btrfs_root *root,
713                                 struct btrfs_path *path, int level)
714 {
715         return 0;
716         if (level == 0)
717                 return check_leaf(root, path, level);
718         return check_node(root, path, level);
719 }
720
721 /*
722  * search for key in the extent_buffer.  The items start at offset p,
723  * and they are item_size apart.  There are 'max' items in p.
724  *
725  * the slot in the array is returned via slot, and it points to
726  * the place where you would insert key if it is not found in
727  * the array.
728  *
729  * slot may point to max if the key is bigger than all of the keys
730  */
731 static noinline int generic_bin_search(struct extent_buffer *eb,
732                                        unsigned long p,
733                                        int item_size, struct btrfs_key *key,
734                                        int max, int *slot)
735 {
736         int low = 0;
737         int high = max;
738         int mid;
739         int ret;
740         struct btrfs_disk_key *tmp = NULL;
741         struct btrfs_disk_key unaligned;
742         unsigned long offset;
743         char *map_token = NULL;
744         char *kaddr = NULL;
745         unsigned long map_start = 0;
746         unsigned long map_len = 0;
747         int err;
748
749         while (low < high) {
750                 mid = (low + high) / 2;
751                 offset = p + mid * item_size;
752
753                 if (!map_token || offset < map_start ||
754                     (offset + sizeof(struct btrfs_disk_key)) >
755                     map_start + map_len) {
756                         if (map_token) {
757                                 unmap_extent_buffer(eb, map_token, KM_USER0);
758                                 map_token = NULL;
759                         }
760
761                         err = map_private_extent_buffer(eb, offset,
762                                                 sizeof(struct btrfs_disk_key),
763                                                 &map_token, &kaddr,
764                                                 &map_start, &map_len, KM_USER0);
765
766                         if (!err) {
767                                 tmp = (struct btrfs_disk_key *)(kaddr + offset -
768                                                         map_start);
769                         } else {
770                                 read_extent_buffer(eb, &unaligned,
771                                                    offset, sizeof(unaligned));
772                                 tmp = &unaligned;
773                         }
774
775                 } else {
776                         tmp = (struct btrfs_disk_key *)(kaddr + offset -
777                                                         map_start);
778                 }
779                 ret = comp_keys(tmp, key);
780
781                 if (ret < 0)
782                         low = mid + 1;
783                 else if (ret > 0)
784                         high = mid;
785                 else {
786                         *slot = mid;
787                         if (map_token)
788                                 unmap_extent_buffer(eb, map_token, KM_USER0);
789                         return 0;
790                 }
791         }
792         *slot = low;
793         if (map_token)
794                 unmap_extent_buffer(eb, map_token, KM_USER0);
795         return 1;
796 }
797
798 /*
799  * simple bin_search frontend that does the right thing for
800  * leaves vs nodes
801  */
802 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
803                       int level, int *slot)
804 {
805         if (level == 0) {
806                 return generic_bin_search(eb,
807                                           offsetof(struct btrfs_leaf, items),
808                                           sizeof(struct btrfs_item),
809                                           key, btrfs_header_nritems(eb),
810                                           slot);
811         } else {
812                 return generic_bin_search(eb,
813                                           offsetof(struct btrfs_node, ptrs),
814                                           sizeof(struct btrfs_key_ptr),
815                                           key, btrfs_header_nritems(eb),
816                                           slot);
817         }
818         return -1;
819 }
820
821 /* given a node and slot number, this reads the blocks it points to.  The
822  * extent buffer is returned with a reference taken (but unlocked).
823  * NULL is returned on error.
824  */
825 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
826                                    struct extent_buffer *parent, int slot)
827 {
828         int level = btrfs_header_level(parent);
829         if (slot < 0)
830                 return NULL;
831         if (slot >= btrfs_header_nritems(parent))
832                 return NULL;
833
834         BUG_ON(level == 0);
835
836         return read_tree_block(root, btrfs_node_blockptr(parent, slot),
837                        btrfs_level_size(root, level - 1),
838                        btrfs_node_ptr_generation(parent, slot));
839 }
840
841 /*
842  * node level balancing, used to make sure nodes are in proper order for
843  * item deletion.  We balance from the top down, so we have to make sure
844  * that a deletion won't leave an node completely empty later on.
845  */
846 static noinline int balance_level(struct btrfs_trans_handle *trans,
847                          struct btrfs_root *root,
848                          struct btrfs_path *path, int level)
849 {
850         struct extent_buffer *right = NULL;
851         struct extent_buffer *mid;
852         struct extent_buffer *left = NULL;
853         struct extent_buffer *parent = NULL;
854         int ret = 0;
855         int wret;
856         int pslot;
857         int orig_slot = path->slots[level];
858         int err_on_enospc = 0;
859         u64 orig_ptr;
860
861         if (level == 0)
862                 return 0;
863
864         mid = path->nodes[level];
865         WARN_ON(!path->locks[level]);
866         WARN_ON(btrfs_header_generation(mid) != trans->transid);
867
868         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
869
870         if (level < BTRFS_MAX_LEVEL - 1)
871                 parent = path->nodes[level + 1];
872         pslot = path->slots[level + 1];
873
874         /*
875          * deal with the case where there is only one pointer in the root
876          * by promoting the node below to a root
877          */
878         if (!parent) {
879                 struct extent_buffer *child;
880
881                 if (btrfs_header_nritems(mid) != 1)
882                         return 0;
883
884                 /* promote the child to a root */
885                 child = read_node_slot(root, mid, 0);
886                 btrfs_tree_lock(child);
887                 BUG_ON(!child);
888                 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
889                 BUG_ON(ret);
890
891                 spin_lock(&root->node_lock);
892                 root->node = child;
893                 spin_unlock(&root->node_lock);
894
895                 ret = btrfs_update_extent_ref(trans, root, child->start,
896                                               mid->start, child->start,
897                                               root->root_key.objectid,
898                                               trans->transid, level - 1);
899                 BUG_ON(ret);
900
901                 add_root_to_dirty_list(root);
902                 btrfs_tree_unlock(child);
903                 path->locks[level] = 0;
904                 path->nodes[level] = NULL;
905                 clean_tree_block(trans, root, mid);
906                 btrfs_tree_unlock(mid);
907                 /* once for the path */
908                 free_extent_buffer(mid);
909                 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
910                                         mid->start, root->root_key.objectid,
911                                         btrfs_header_generation(mid),
912                                         level, 1);
913                 /* once for the root ptr */
914                 free_extent_buffer(mid);
915                 return ret;
916         }
917         if (btrfs_header_nritems(mid) >
918             BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
919                 return 0;
920
921         if (btrfs_header_nritems(mid) < 2)
922                 err_on_enospc = 1;
923
924         left = read_node_slot(root, parent, pslot - 1);
925         if (left) {
926                 btrfs_tree_lock(left);
927                 wret = btrfs_cow_block(trans, root, left,
928                                        parent, pslot - 1, &left, 0);
929                 if (wret) {
930                         ret = wret;
931                         goto enospc;
932                 }
933         }
934         right = read_node_slot(root, parent, pslot + 1);
935         if (right) {
936                 btrfs_tree_lock(right);
937                 wret = btrfs_cow_block(trans, root, right,
938                                        parent, pslot + 1, &right, 0);
939                 if (wret) {
940                         ret = wret;
941                         goto enospc;
942                 }
943         }
944
945         /* first, try to make some room in the middle buffer */
946         if (left) {
947                 orig_slot += btrfs_header_nritems(left);
948                 wret = push_node_left(trans, root, left, mid, 1);
949                 if (wret < 0)
950                         ret = wret;
951                 if (btrfs_header_nritems(mid) < 2)
952                         err_on_enospc = 1;
953         }
954
955         /*
956          * then try to empty the right most buffer into the middle
957          */
958         if (right) {
959                 wret = push_node_left(trans, root, mid, right, 1);
960                 if (wret < 0 && wret != -ENOSPC)
961                         ret = wret;
962                 if (btrfs_header_nritems(right) == 0) {
963                         u64 bytenr = right->start;
964                         u64 generation = btrfs_header_generation(parent);
965                         u32 blocksize = right->len;
966
967                         clean_tree_block(trans, root, right);
968                         btrfs_tree_unlock(right);
969                         free_extent_buffer(right);
970                         right = NULL;
971                         wret = del_ptr(trans, root, path, level + 1, pslot +
972                                        1);
973                         if (wret)
974                                 ret = wret;
975                         wret = btrfs_free_extent(trans, root, bytenr,
976                                                  blocksize, parent->start,
977                                                  btrfs_header_owner(parent),
978                                                  generation, level, 1);
979                         if (wret)
980                                 ret = wret;
981                 } else {
982                         struct btrfs_disk_key right_key;
983                         btrfs_node_key(right, &right_key, 0);
984                         btrfs_set_node_key(parent, &right_key, pslot + 1);
985                         btrfs_mark_buffer_dirty(parent);
986                 }
987         }
988         if (btrfs_header_nritems(mid) == 1) {
989                 /*
990                  * we're not allowed to leave a node with one item in the
991                  * tree during a delete.  A deletion from lower in the tree
992                  * could try to delete the only pointer in this node.
993                  * So, pull some keys from the left.
994                  * There has to be a left pointer at this point because
995                  * otherwise we would have pulled some pointers from the
996                  * right
997                  */
998                 BUG_ON(!left);
999                 wret = balance_node_right(trans, root, mid, left);
1000                 if (wret < 0) {
1001                         ret = wret;
1002                         goto enospc;
1003                 }
1004                 if (wret == 1) {
1005                         wret = push_node_left(trans, root, left, mid, 1);
1006                         if (wret < 0)
1007                                 ret = wret;
1008                 }
1009                 BUG_ON(wret == 1);
1010         }
1011         if (btrfs_header_nritems(mid) == 0) {
1012                 /* we've managed to empty the middle node, drop it */
1013                 u64 root_gen = btrfs_header_generation(parent);
1014                 u64 bytenr = mid->start;
1015                 u32 blocksize = mid->len;
1016
1017                 clean_tree_block(trans, root, mid);
1018                 btrfs_tree_unlock(mid);
1019                 free_extent_buffer(mid);
1020                 mid = NULL;
1021                 wret = del_ptr(trans, root, path, level + 1, pslot);
1022                 if (wret)
1023                         ret = wret;
1024                 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1025                                          parent->start,
1026                                          btrfs_header_owner(parent),
1027                                          root_gen, level, 1);
1028                 if (wret)
1029                         ret = wret;
1030         } else {
1031                 /* update the parent key to reflect our changes */
1032                 struct btrfs_disk_key mid_key;
1033                 btrfs_node_key(mid, &mid_key, 0);
1034                 btrfs_set_node_key(parent, &mid_key, pslot);
1035                 btrfs_mark_buffer_dirty(parent);
1036         }
1037
1038         /* update the path */
1039         if (left) {
1040                 if (btrfs_header_nritems(left) > orig_slot) {
1041                         extent_buffer_get(left);
1042                         /* left was locked after cow */
1043                         path->nodes[level] = left;
1044                         path->slots[level + 1] -= 1;
1045                         path->slots[level] = orig_slot;
1046                         if (mid) {
1047                                 btrfs_tree_unlock(mid);
1048                                 free_extent_buffer(mid);
1049                         }
1050                 } else {
1051                         orig_slot -= btrfs_header_nritems(left);
1052                         path->slots[level] = orig_slot;
1053                 }
1054         }
1055         /* double check we haven't messed things up */
1056         check_block(root, path, level);
1057         if (orig_ptr !=
1058             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1059                 BUG();
1060 enospc:
1061         if (right) {
1062                 btrfs_tree_unlock(right);
1063                 free_extent_buffer(right);
1064         }
1065         if (left) {
1066                 if (path->nodes[level] != left)
1067                         btrfs_tree_unlock(left);
1068                 free_extent_buffer(left);
1069         }
1070         return ret;
1071 }
1072
1073 /* Node balancing for insertion.  Here we only split or push nodes around
1074  * when they are completely full.  This is also done top down, so we
1075  * have to be pessimistic.
1076  */
1077 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1078                                           struct btrfs_root *root,
1079                                           struct btrfs_path *path, int level)
1080 {
1081         struct extent_buffer *right = NULL;
1082         struct extent_buffer *mid;
1083         struct extent_buffer *left = NULL;
1084         struct extent_buffer *parent = NULL;
1085         int ret = 0;
1086         int wret;
1087         int pslot;
1088         int orig_slot = path->slots[level];
1089         u64 orig_ptr;
1090
1091         if (level == 0)
1092                 return 1;
1093
1094         mid = path->nodes[level];
1095         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1096         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1097
1098         if (level < BTRFS_MAX_LEVEL - 1)
1099                 parent = path->nodes[level + 1];
1100         pslot = path->slots[level + 1];
1101
1102         if (!parent)
1103                 return 1;
1104
1105         left = read_node_slot(root, parent, pslot - 1);
1106
1107         /* first, try to make some room in the middle buffer */
1108         if (left) {
1109                 u32 left_nr;
1110
1111                 btrfs_tree_lock(left);
1112                 left_nr = btrfs_header_nritems(left);
1113                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1114                         wret = 1;
1115                 } else {
1116                         ret = btrfs_cow_block(trans, root, left, parent,
1117                                               pslot - 1, &left, 0);
1118                         if (ret)
1119                                 wret = 1;
1120                         else {
1121                                 wret = push_node_left(trans, root,
1122                                                       left, mid, 0);
1123                         }
1124                 }
1125                 if (wret < 0)
1126                         ret = wret;
1127                 if (wret == 0) {
1128                         struct btrfs_disk_key disk_key;
1129                         orig_slot += left_nr;
1130                         btrfs_node_key(mid, &disk_key, 0);
1131                         btrfs_set_node_key(parent, &disk_key, pslot);
1132                         btrfs_mark_buffer_dirty(parent);
1133                         if (btrfs_header_nritems(left) > orig_slot) {
1134                                 path->nodes[level] = left;
1135                                 path->slots[level + 1] -= 1;
1136                                 path->slots[level] = orig_slot;
1137                                 btrfs_tree_unlock(mid);
1138                                 free_extent_buffer(mid);
1139                         } else {
1140                                 orig_slot -=
1141                                         btrfs_header_nritems(left);
1142                                 path->slots[level] = orig_slot;
1143                                 btrfs_tree_unlock(left);
1144                                 free_extent_buffer(left);
1145                         }
1146                         return 0;
1147                 }
1148                 btrfs_tree_unlock(left);
1149                 free_extent_buffer(left);
1150         }
1151         right = read_node_slot(root, parent, pslot + 1);
1152
1153         /*
1154          * then try to empty the right most buffer into the middle
1155          */
1156         if (right) {
1157                 u32 right_nr;
1158                 btrfs_tree_lock(right);
1159                 right_nr = btrfs_header_nritems(right);
1160                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1161                         wret = 1;
1162                 } else {
1163                         ret = btrfs_cow_block(trans, root, right,
1164                                               parent, pslot + 1,
1165                                               &right, 0);
1166                         if (ret)
1167                                 wret = 1;
1168                         else {
1169                                 wret = balance_node_right(trans, root,
1170                                                           right, mid);
1171                         }
1172                 }
1173                 if (wret < 0)
1174                         ret = wret;
1175                 if (wret == 0) {
1176                         struct btrfs_disk_key disk_key;
1177
1178                         btrfs_node_key(right, &disk_key, 0);
1179                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1180                         btrfs_mark_buffer_dirty(parent);
1181
1182                         if (btrfs_header_nritems(mid) <= orig_slot) {
1183                                 path->nodes[level] = right;
1184                                 path->slots[level + 1] += 1;
1185                                 path->slots[level] = orig_slot -
1186                                         btrfs_header_nritems(mid);
1187                                 btrfs_tree_unlock(mid);
1188                                 free_extent_buffer(mid);
1189                         } else {
1190                                 btrfs_tree_unlock(right);
1191                                 free_extent_buffer(right);
1192                         }
1193                         return 0;
1194                 }
1195                 btrfs_tree_unlock(right);
1196                 free_extent_buffer(right);
1197         }
1198         return 1;
1199 }
1200
1201 /*
1202  * readahead one full node of leaves, finding things that are close
1203  * to the block in 'slot', and triggering ra on them.
1204  */
1205 static noinline void reada_for_search(struct btrfs_root *root,
1206                                       struct btrfs_path *path,
1207                                       int level, int slot, u64 objectid)
1208 {
1209         struct extent_buffer *node;
1210         struct btrfs_disk_key disk_key;
1211         u32 nritems;
1212         u64 search;
1213         u64 lowest_read;
1214         u64 highest_read;
1215         u64 nread = 0;
1216         int direction = path->reada;
1217         struct extent_buffer *eb;
1218         u32 nr;
1219         u32 blocksize;
1220         u32 nscan = 0;
1221
1222         if (level != 1)
1223                 return;
1224
1225         if (!path->nodes[level])
1226                 return;
1227
1228         node = path->nodes[level];
1229
1230         search = btrfs_node_blockptr(node, slot);
1231         blocksize = btrfs_level_size(root, level - 1);
1232         eb = btrfs_find_tree_block(root, search, blocksize);
1233         if (eb) {
1234                 free_extent_buffer(eb);
1235                 return;
1236         }
1237
1238         highest_read = search;
1239         lowest_read = search;
1240
1241         nritems = btrfs_header_nritems(node);
1242         nr = slot;
1243         while (1) {
1244                 if (direction < 0) {
1245                         if (nr == 0)
1246                                 break;
1247                         nr--;
1248                 } else if (direction > 0) {
1249                         nr++;
1250                         if (nr >= nritems)
1251                                 break;
1252                 }
1253                 if (path->reada < 0 && objectid) {
1254                         btrfs_node_key(node, &disk_key, nr);
1255                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1256                                 break;
1257                 }
1258                 search = btrfs_node_blockptr(node, nr);
1259                 if ((search >= lowest_read && search <= highest_read) ||
1260                     (search < lowest_read && lowest_read - search <= 16384) ||
1261                     (search > highest_read && search - highest_read <= 16384)) {
1262                         readahead_tree_block(root, search, blocksize,
1263                                      btrfs_node_ptr_generation(node, nr));
1264                         nread += blocksize;
1265                 }
1266                 nscan++;
1267                 if (path->reada < 2 && (nread > (64 * 1024) || nscan > 32))
1268                         break;
1269
1270                 if (nread > (256 * 1024) || nscan > 128)
1271                         break;
1272
1273                 if (search < lowest_read)
1274                         lowest_read = search;
1275                 if (search > highest_read)
1276                         highest_read = search;
1277         }
1278 }
1279
1280 /*
1281  * when we walk down the tree, it is usually safe to unlock the higher layers
1282  * in the tree.  The exceptions are when our path goes through slot 0, because
1283  * operations on the tree might require changing key pointers higher up in the
1284  * tree.
1285  *
1286  * callers might also have set path->keep_locks, which tells this code to keep
1287  * the lock if the path points to the last slot in the block.  This is part of
1288  * walking through the tree, and selecting the next slot in the higher block.
1289  *
1290  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1291  * if lowest_unlock is 1, level 0 won't be unlocked
1292  */
1293 static noinline void unlock_up(struct btrfs_path *path, int level,
1294                                int lowest_unlock)
1295 {
1296         int i;
1297         int skip_level = level;
1298         int no_skips = 0;
1299         struct extent_buffer *t;
1300
1301         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1302                 if (!path->nodes[i])
1303                         break;
1304                 if (!path->locks[i])
1305                         break;
1306                 if (!no_skips && path->slots[i] == 0) {
1307                         skip_level = i + 1;
1308                         continue;
1309                 }
1310                 if (!no_skips && path->keep_locks) {
1311                         u32 nritems;
1312                         t = path->nodes[i];
1313                         nritems = btrfs_header_nritems(t);
1314                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1315                                 skip_level = i + 1;
1316                                 continue;
1317                         }
1318                 }
1319                 if (skip_level < i && i >= lowest_unlock)
1320                         no_skips = 1;
1321
1322                 t = path->nodes[i];
1323                 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1324                         btrfs_tree_unlock(t);
1325                         path->locks[i] = 0;
1326                 }
1327         }
1328 }
1329
1330 /*
1331  * look for key in the tree.  path is filled in with nodes along the way
1332  * if key is found, we return zero and you can find the item in the leaf
1333  * level of the path (level 0)
1334  *
1335  * If the key isn't found, the path points to the slot where it should
1336  * be inserted, and 1 is returned.  If there are other errors during the
1337  * search a negative error number is returned.
1338  *
1339  * if ins_len > 0, nodes and leaves will be split as we walk down the
1340  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1341  * possible)
1342  */
1343 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1344                       *root, struct btrfs_key *key, struct btrfs_path *p, int
1345                       ins_len, int cow)
1346 {
1347         struct extent_buffer *b;
1348         struct extent_buffer *tmp;
1349         int slot;
1350         int ret;
1351         int level;
1352         int should_reada = p->reada;
1353         int lowest_unlock = 1;
1354         int blocksize;
1355         u8 lowest_level = 0;
1356         u64 blocknr;
1357         u64 gen;
1358         struct btrfs_key prealloc_block;
1359
1360         lowest_level = p->lowest_level;
1361         WARN_ON(lowest_level && ins_len > 0);
1362         WARN_ON(p->nodes[0] != NULL);
1363
1364         if (ins_len < 0)
1365                 lowest_unlock = 2;
1366
1367         prealloc_block.objectid = 0;
1368
1369 again:
1370         if (p->skip_locking)
1371                 b = btrfs_root_node(root);
1372         else
1373                 b = btrfs_lock_root_node(root);
1374
1375         while (b) {
1376                 level = btrfs_header_level(b);
1377
1378                 /*
1379                  * setup the path here so we can release it under lock
1380                  * contention with the cow code
1381                  */
1382                 p->nodes[level] = b;
1383                 if (!p->skip_locking)
1384                         p->locks[level] = 1;
1385
1386                 if (cow) {
1387                         int wret;
1388
1389                         /* is a cow on this block not required */
1390                         spin_lock(&root->fs_info->hash_lock);
1391                         if (btrfs_header_generation(b) == trans->transid &&
1392                             btrfs_header_owner(b) == root->root_key.objectid &&
1393                             !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1394                                 spin_unlock(&root->fs_info->hash_lock);
1395                                 goto cow_done;
1396                         }
1397                         spin_unlock(&root->fs_info->hash_lock);
1398
1399                         /* ok, we have to cow, is our old prealloc the right
1400                          * size?
1401                          */
1402                         if (prealloc_block.objectid &&
1403                             prealloc_block.offset != b->len) {
1404                                 btrfs_free_reserved_extent(root,
1405                                            prealloc_block.objectid,
1406                                            prealloc_block.offset);
1407                                 prealloc_block.objectid = 0;
1408                         }
1409
1410                         /*
1411                          * for higher level blocks, try not to allocate blocks
1412                          * with the block and the parent locks held.
1413                          */
1414                         if (level > 1 && !prealloc_block.objectid &&
1415                             btrfs_path_lock_waiting(p, level)) {
1416                                 u32 size = b->len;
1417                                 u64 hint = b->start;
1418
1419                                 btrfs_release_path(root, p);
1420                                 ret = btrfs_reserve_extent(trans, root,
1421                                                            size, size, 0,
1422                                                            hint, (u64)-1,
1423                                                            &prealloc_block, 0);
1424                                 BUG_ON(ret);
1425                                 goto again;
1426                         }
1427
1428                         wret = btrfs_cow_block(trans, root, b,
1429                                                p->nodes[level + 1],
1430                                                p->slots[level + 1],
1431                                                &b, prealloc_block.objectid);
1432                         prealloc_block.objectid = 0;
1433                         if (wret) {
1434                                 free_extent_buffer(b);
1435                                 ret = wret;
1436                                 goto done;
1437                         }
1438                 }
1439 cow_done:
1440                 BUG_ON(!cow && ins_len);
1441                 if (level != btrfs_header_level(b))
1442                         WARN_ON(1);
1443                 level = btrfs_header_level(b);
1444
1445                 p->nodes[level] = b;
1446                 if (!p->skip_locking)
1447                         p->locks[level] = 1;
1448
1449                 ret = check_block(root, p, level);
1450                 if (ret) {
1451                         ret = -1;
1452                         goto done;
1453                 }
1454
1455                 ret = bin_search(b, key, level, &slot);
1456                 if (level != 0) {
1457                         if (ret && slot > 0)
1458                                 slot -= 1;
1459                         p->slots[level] = slot;
1460                         if ((p->search_for_split || ins_len > 0) &&
1461                             btrfs_header_nritems(b) >=
1462                             BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1463                                 int sret = split_node(trans, root, p, level);
1464                                 BUG_ON(sret > 0);
1465                                 if (sret) {
1466                                         ret = sret;
1467                                         goto done;
1468                                 }
1469                                 b = p->nodes[level];
1470                                 slot = p->slots[level];
1471                         } else if (ins_len < 0) {
1472                                 int sret = balance_level(trans, root, p,
1473                                                          level);
1474                                 if (sret) {
1475                                         ret = sret;
1476                                         goto done;
1477                                 }
1478                                 b = p->nodes[level];
1479                                 if (!b) {
1480                                         btrfs_release_path(NULL, p);
1481                                         goto again;
1482                                 }
1483                                 slot = p->slots[level];
1484                                 BUG_ON(btrfs_header_nritems(b) == 1);
1485                         }
1486                         unlock_up(p, level, lowest_unlock);
1487
1488                         /* this is only true while dropping a snapshot */
1489                         if (level == lowest_level) {
1490                                 ret = 0;
1491                                 goto done;
1492                         }
1493
1494                         blocknr = btrfs_node_blockptr(b, slot);
1495                         gen = btrfs_node_ptr_generation(b, slot);
1496                         blocksize = btrfs_level_size(root, level - 1);
1497
1498                         tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1499                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1500                                 b = tmp;
1501                         } else {
1502                                 /*
1503                                  * reduce lock contention at high levels
1504                                  * of the btree by dropping locks before
1505                                  * we read.
1506                                  */
1507                                 if (level > 1) {
1508                                         btrfs_release_path(NULL, p);
1509                                         if (tmp)
1510                                                 free_extent_buffer(tmp);
1511                                         if (should_reada)
1512                                                 reada_for_search(root, p,
1513                                                                  level, slot,
1514                                                                  key->objectid);
1515
1516                                         tmp = read_tree_block(root, blocknr,
1517                                                          blocksize, gen);
1518                                         if (tmp)
1519                                                 free_extent_buffer(tmp);
1520                                         goto again;
1521                                 } else {
1522                                         if (tmp)
1523                                                 free_extent_buffer(tmp);
1524                                         if (should_reada)
1525                                                 reada_for_search(root, p,
1526                                                                  level, slot,
1527                                                                  key->objectid);
1528                                         b = read_node_slot(root, b, slot);
1529                                 }
1530                         }
1531                         if (!p->skip_locking)
1532                                 btrfs_tree_lock(b);
1533                 } else {
1534                         p->slots[level] = slot;
1535                         if (ins_len > 0 &&
1536                             btrfs_leaf_free_space(root, b) < ins_len) {
1537                                 int sret = split_leaf(trans, root, key,
1538                                                       p, ins_len, ret == 0);
1539                                 BUG_ON(sret > 0);
1540                                 if (sret) {
1541                                         ret = sret;
1542                                         goto done;
1543                                 }
1544                         }
1545                         if (!p->search_for_split)
1546                                 unlock_up(p, level, lowest_unlock);
1547                         goto done;
1548                 }
1549         }
1550         ret = 1;
1551 done:
1552         if (prealloc_block.objectid) {
1553                 btrfs_free_reserved_extent(root,
1554                            prealloc_block.objectid,
1555                            prealloc_block.offset);
1556         }
1557
1558         return ret;
1559 }
1560
1561 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1562                      struct btrfs_root *root,
1563                      struct btrfs_key *node_keys,
1564                      u64 *nodes, int lowest_level)
1565 {
1566         struct extent_buffer *eb;
1567         struct extent_buffer *parent;
1568         struct btrfs_key key;
1569         u64 bytenr;
1570         u64 generation;
1571         u32 blocksize;
1572         int level;
1573         int slot;
1574         int key_match;
1575         int ret;
1576
1577         eb = btrfs_lock_root_node(root);
1578         ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0);
1579         BUG_ON(ret);
1580
1581         parent = eb;
1582         while (1) {
1583                 level = btrfs_header_level(parent);
1584                 if (level == 0 || level <= lowest_level)
1585                         break;
1586
1587                 ret = bin_search(parent, &node_keys[lowest_level], level,
1588                                  &slot);
1589                 if (ret && slot > 0)
1590                         slot--;
1591
1592                 bytenr = btrfs_node_blockptr(parent, slot);
1593                 if (nodes[level - 1] == bytenr)
1594                         break;
1595
1596                 blocksize = btrfs_level_size(root, level - 1);
1597                 generation = btrfs_node_ptr_generation(parent, slot);
1598                 btrfs_node_key_to_cpu(eb, &key, slot);
1599                 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1600
1601                 if (generation == trans->transid) {
1602                         eb = read_tree_block(root, bytenr, blocksize,
1603                                              generation);
1604                         btrfs_tree_lock(eb);
1605                 }
1606
1607                 /*
1608                  * if node keys match and node pointer hasn't been modified
1609                  * in the running transaction, we can merge the path. for
1610                  * blocks owened by reloc trees, the node pointer check is
1611                  * skipped, this is because these blocks are fully controlled
1612                  * by the space balance code, no one else can modify them.
1613                  */
1614                 if (!nodes[level - 1] || !key_match ||
1615                     (generation == trans->transid &&
1616                      btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1617                         if (level == 1 || level == lowest_level + 1) {
1618                                 if (generation == trans->transid) {
1619                                         btrfs_tree_unlock(eb);
1620                                         free_extent_buffer(eb);
1621                                 }
1622                                 break;
1623                         }
1624
1625                         if (generation != trans->transid) {
1626                                 eb = read_tree_block(root, bytenr, blocksize,
1627                                                 generation);
1628                                 btrfs_tree_lock(eb);
1629                         }
1630
1631                         ret = btrfs_cow_block(trans, root, eb, parent, slot,
1632                                               &eb, 0);
1633                         BUG_ON(ret);
1634
1635                         if (root->root_key.objectid ==
1636                             BTRFS_TREE_RELOC_OBJECTID) {
1637                                 if (!nodes[level - 1]) {
1638                                         nodes[level - 1] = eb->start;
1639                                         memcpy(&node_keys[level - 1], &key,
1640                                                sizeof(node_keys[0]));
1641                                 } else {
1642                                         WARN_ON(1);
1643                                 }
1644                         }
1645
1646                         btrfs_tree_unlock(parent);
1647                         free_extent_buffer(parent);
1648                         parent = eb;
1649                         continue;
1650                 }
1651
1652                 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1653                 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1654                 btrfs_mark_buffer_dirty(parent);
1655
1656                 ret = btrfs_inc_extent_ref(trans, root,
1657                                         nodes[level - 1],
1658                                         blocksize, parent->start,
1659                                         btrfs_header_owner(parent),
1660                                         btrfs_header_generation(parent),
1661                                         level - 1);
1662                 BUG_ON(ret);
1663
1664                 /*
1665                  * If the block was created in the running transaction,
1666                  * it's possible this is the last reference to it, so we
1667                  * should drop the subtree.
1668                  */
1669                 if (generation == trans->transid) {
1670                         ret = btrfs_drop_subtree(trans, root, eb, parent);
1671                         BUG_ON(ret);
1672                         btrfs_tree_unlock(eb);
1673                         free_extent_buffer(eb);
1674                 } else {
1675                         ret = btrfs_free_extent(trans, root, bytenr,
1676                                         blocksize, parent->start,
1677                                         btrfs_header_owner(parent),
1678                                         btrfs_header_generation(parent),
1679                                         level - 1, 1);
1680                         BUG_ON(ret);
1681                 }
1682                 break;
1683         }
1684         btrfs_tree_unlock(parent);
1685         free_extent_buffer(parent);
1686         return 0;
1687 }
1688
1689 /*
1690  * adjust the pointers going up the tree, starting at level
1691  * making sure the right key of each node is points to 'key'.
1692  * This is used after shifting pointers to the left, so it stops
1693  * fixing up pointers when a given leaf/node is not in slot 0 of the
1694  * higher levels
1695  *
1696  * If this fails to write a tree block, it returns -1, but continues
1697  * fixing up the blocks in ram so the tree is consistent.
1698  */
1699 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1700                           struct btrfs_root *root, struct btrfs_path *path,
1701                           struct btrfs_disk_key *key, int level)
1702 {
1703         int i;
1704         int ret = 0;
1705         struct extent_buffer *t;
1706
1707         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1708                 int tslot = path->slots[i];
1709                 if (!path->nodes[i])
1710                         break;
1711                 t = path->nodes[i];
1712                 btrfs_set_node_key(t, key, tslot);
1713                 btrfs_mark_buffer_dirty(path->nodes[i]);
1714                 if (tslot != 0)
1715                         break;
1716         }
1717         return ret;
1718 }
1719
1720 /*
1721  * update item key.
1722  *
1723  * This function isn't completely safe. It's the caller's responsibility
1724  * that the new key won't break the order
1725  */
1726 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1727                             struct btrfs_root *root, struct btrfs_path *path,
1728                             struct btrfs_key *new_key)
1729 {
1730         struct btrfs_disk_key disk_key;
1731         struct extent_buffer *eb;
1732         int slot;
1733
1734         eb = path->nodes[0];
1735         slot = path->slots[0];
1736         if (slot > 0) {
1737                 btrfs_item_key(eb, &disk_key, slot - 1);
1738                 if (comp_keys(&disk_key, new_key) >= 0)
1739                         return -1;
1740         }
1741         if (slot < btrfs_header_nritems(eb) - 1) {
1742                 btrfs_item_key(eb, &disk_key, slot + 1);
1743                 if (comp_keys(&disk_key, new_key) <= 0)
1744                         return -1;
1745         }
1746
1747         btrfs_cpu_key_to_disk(&disk_key, new_key);
1748         btrfs_set_item_key(eb, &disk_key, slot);
1749         btrfs_mark_buffer_dirty(eb);
1750         if (slot == 0)
1751                 fixup_low_keys(trans, root, path, &disk_key, 1);
1752         return 0;
1753 }
1754
1755 /*
1756  * try to push data from one node into the next node left in the
1757  * tree.
1758  *
1759  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1760  * error, and > 0 if there was no room in the left hand block.
1761  */
1762 static int push_node_left(struct btrfs_trans_handle *trans,
1763                           struct btrfs_root *root, struct extent_buffer *dst,
1764                           struct extent_buffer *src, int empty)
1765 {
1766         int push_items = 0;
1767         int src_nritems;
1768         int dst_nritems;
1769         int ret = 0;
1770
1771         src_nritems = btrfs_header_nritems(src);
1772         dst_nritems = btrfs_header_nritems(dst);
1773         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1774         WARN_ON(btrfs_header_generation(src) != trans->transid);
1775         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1776
1777         if (!empty && src_nritems <= 8)
1778                 return 1;
1779
1780         if (push_items <= 0)
1781                 return 1;
1782
1783         if (empty) {
1784                 push_items = min(src_nritems, push_items);
1785                 if (push_items < src_nritems) {
1786                         /* leave at least 8 pointers in the node if
1787                          * we aren't going to empty it
1788                          */
1789                         if (src_nritems - push_items < 8) {
1790                                 if (push_items <= 8)
1791                                         return 1;
1792                                 push_items -= 8;
1793                         }
1794                 }
1795         } else
1796                 push_items = min(src_nritems - 8, push_items);
1797
1798         copy_extent_buffer(dst, src,
1799                            btrfs_node_key_ptr_offset(dst_nritems),
1800                            btrfs_node_key_ptr_offset(0),
1801                            push_items * sizeof(struct btrfs_key_ptr));
1802
1803         if (push_items < src_nritems) {
1804                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1805                                       btrfs_node_key_ptr_offset(push_items),
1806                                       (src_nritems - push_items) *
1807                                       sizeof(struct btrfs_key_ptr));
1808         }
1809         btrfs_set_header_nritems(src, src_nritems - push_items);
1810         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1811         btrfs_mark_buffer_dirty(src);
1812         btrfs_mark_buffer_dirty(dst);
1813
1814         ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1815         BUG_ON(ret);
1816
1817         return ret;
1818 }
1819
1820 /*
1821  * try to push data from one node into the next node right in the
1822  * tree.
1823  *
1824  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1825  * error, and > 0 if there was no room in the right hand block.
1826  *
1827  * this will  only push up to 1/2 the contents of the left node over
1828  */
1829 static int balance_node_right(struct btrfs_trans_handle *trans,
1830                               struct btrfs_root *root,
1831                               struct extent_buffer *dst,
1832                               struct extent_buffer *src)
1833 {
1834         int push_items = 0;
1835         int max_push;
1836         int src_nritems;
1837         int dst_nritems;
1838         int ret = 0;
1839
1840         WARN_ON(btrfs_header_generation(src) != trans->transid);
1841         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1842
1843         src_nritems = btrfs_header_nritems(src);
1844         dst_nritems = btrfs_header_nritems(dst);
1845         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1846         if (push_items <= 0)
1847                 return 1;
1848
1849         if (src_nritems < 4)
1850                 return 1;
1851
1852         max_push = src_nritems / 2 + 1;
1853         /* don't try to empty the node */
1854         if (max_push >= src_nritems)
1855                 return 1;
1856
1857         if (max_push < push_items)
1858                 push_items = max_push;
1859
1860         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1861                                       btrfs_node_key_ptr_offset(0),
1862                                       (dst_nritems) *
1863                                       sizeof(struct btrfs_key_ptr));
1864
1865         copy_extent_buffer(dst, src,
1866                            btrfs_node_key_ptr_offset(0),
1867                            btrfs_node_key_ptr_offset(src_nritems - push_items),
1868                            push_items * sizeof(struct btrfs_key_ptr));
1869
1870         btrfs_set_header_nritems(src, src_nritems - push_items);
1871         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1872
1873         btrfs_mark_buffer_dirty(src);
1874         btrfs_mark_buffer_dirty(dst);
1875
1876         ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
1877         BUG_ON(ret);
1878
1879         return ret;
1880 }
1881
1882 /*
1883  * helper function to insert a new root level in the tree.
1884  * A new node is allocated, and a single item is inserted to
1885  * point to the existing root
1886  *
1887  * returns zero on success or < 0 on failure.
1888  */
1889 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
1890                            struct btrfs_root *root,
1891                            struct btrfs_path *path, int level)
1892 {
1893         u64 lower_gen;
1894         struct extent_buffer *lower;
1895         struct extent_buffer *c;
1896         struct extent_buffer *old;
1897         struct btrfs_disk_key lower_key;
1898         int ret;
1899
1900         BUG_ON(path->nodes[level]);
1901         BUG_ON(path->nodes[level-1] != root->node);
1902
1903         lower = path->nodes[level-1];
1904         if (level == 1)
1905                 btrfs_item_key(lower, &lower_key, 0);
1906         else
1907                 btrfs_node_key(lower, &lower_key, 0);
1908
1909         c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1910                                    root->root_key.objectid, trans->transid,
1911                                    level, root->node->start, 0);
1912         if (IS_ERR(c))
1913                 return PTR_ERR(c);
1914
1915         memset_extent_buffer(c, 0, 0, root->nodesize);
1916         btrfs_set_header_nritems(c, 1);
1917         btrfs_set_header_level(c, level);
1918         btrfs_set_header_bytenr(c, c->start);
1919         btrfs_set_header_generation(c, trans->transid);
1920         btrfs_set_header_owner(c, root->root_key.objectid);
1921
1922         write_extent_buffer(c, root->fs_info->fsid,
1923                             (unsigned long)btrfs_header_fsid(c),
1924                             BTRFS_FSID_SIZE);
1925
1926         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1927                             (unsigned long)btrfs_header_chunk_tree_uuid(c),
1928                             BTRFS_UUID_SIZE);
1929
1930         btrfs_set_node_key(c, &lower_key, 0);
1931         btrfs_set_node_blockptr(c, 0, lower->start);
1932         lower_gen = btrfs_header_generation(lower);
1933         WARN_ON(lower_gen != trans->transid);
1934
1935         btrfs_set_node_ptr_generation(c, 0, lower_gen);
1936
1937         btrfs_mark_buffer_dirty(c);
1938
1939         spin_lock(&root->node_lock);
1940         old = root->node;
1941         root->node = c;
1942         spin_unlock(&root->node_lock);
1943
1944         ret = btrfs_update_extent_ref(trans, root, lower->start,
1945                                       lower->start, c->start,
1946                                       root->root_key.objectid,
1947                                       trans->transid, level - 1);
1948         BUG_ON(ret);
1949
1950         /* the super has an extra ref to root->node */
1951         free_extent_buffer(old);
1952
1953         add_root_to_dirty_list(root);
1954         extent_buffer_get(c);
1955         path->nodes[level] = c;
1956         path->locks[level] = 1;
1957         path->slots[level] = 0;
1958         return 0;
1959 }
1960
1961 /*
1962  * worker function to insert a single pointer in a node.
1963  * the node should have enough room for the pointer already
1964  *
1965  * slot and level indicate where you want the key to go, and
1966  * blocknr is the block the key points to.
1967  *
1968  * returns zero on success and < 0 on any error
1969  */
1970 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1971                       *root, struct btrfs_path *path, struct btrfs_disk_key
1972                       *key, u64 bytenr, int slot, int level)
1973 {
1974         struct extent_buffer *lower;
1975         int nritems;
1976
1977         BUG_ON(!path->nodes[level]);
1978         lower = path->nodes[level];
1979         nritems = btrfs_header_nritems(lower);
1980         if (slot > nritems)
1981                 BUG();
1982         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1983                 BUG();
1984         if (slot != nritems) {
1985                 memmove_extent_buffer(lower,
1986                               btrfs_node_key_ptr_offset(slot + 1),
1987                               btrfs_node_key_ptr_offset(slot),
1988                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
1989         }
1990         btrfs_set_node_key(lower, key, slot);
1991         btrfs_set_node_blockptr(lower, slot, bytenr);
1992         WARN_ON(trans->transid == 0);
1993         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1994         btrfs_set_header_nritems(lower, nritems + 1);
1995         btrfs_mark_buffer_dirty(lower);
1996         return 0;
1997 }
1998
1999 /*
2000  * split the node at the specified level in path in two.
2001  * The path is corrected to point to the appropriate node after the split
2002  *
2003  * Before splitting this tries to make some room in the node by pushing
2004  * left and right, if either one works, it returns right away.
2005  *
2006  * returns 0 on success and < 0 on failure
2007  */
2008 static noinline int split_node(struct btrfs_trans_handle *trans,
2009                                struct btrfs_root *root,
2010                                struct btrfs_path *path, int level)
2011 {
2012         struct extent_buffer *c;
2013         struct extent_buffer *split;
2014         struct btrfs_disk_key disk_key;
2015         int mid;
2016         int ret;
2017         int wret;
2018         u32 c_nritems;
2019
2020         c = path->nodes[level];
2021         WARN_ON(btrfs_header_generation(c) != trans->transid);
2022         if (c == root->node) {
2023                 /* trying to split the root, lets make a new one */
2024                 ret = insert_new_root(trans, root, path, level + 1);
2025                 if (ret)
2026                         return ret;
2027         } else {
2028                 ret = push_nodes_for_insert(trans, root, path, level);
2029                 c = path->nodes[level];
2030                 if (!ret && btrfs_header_nritems(c) <
2031                     BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2032                         return 0;
2033                 if (ret < 0)
2034                         return ret;
2035         }
2036
2037         c_nritems = btrfs_header_nritems(c);
2038
2039         split = btrfs_alloc_free_block(trans, root, root->nodesize,
2040                                         path->nodes[level + 1]->start,
2041                                         root->root_key.objectid,
2042                                         trans->transid, level, c->start, 0);
2043         if (IS_ERR(split))
2044                 return PTR_ERR(split);
2045
2046         btrfs_set_header_flags(split, btrfs_header_flags(c));
2047         btrfs_set_header_level(split, btrfs_header_level(c));
2048         btrfs_set_header_bytenr(split, split->start);
2049         btrfs_set_header_generation(split, trans->transid);
2050         btrfs_set_header_owner(split, root->root_key.objectid);
2051         btrfs_set_header_flags(split, 0);
2052         write_extent_buffer(split, root->fs_info->fsid,
2053                             (unsigned long)btrfs_header_fsid(split),
2054                             BTRFS_FSID_SIZE);
2055         write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2056                             (unsigned long)btrfs_header_chunk_tree_uuid(split),
2057                             BTRFS_UUID_SIZE);
2058
2059         mid = (c_nritems + 1) / 2;
2060
2061         copy_extent_buffer(split, c,
2062                            btrfs_node_key_ptr_offset(0),
2063                            btrfs_node_key_ptr_offset(mid),
2064                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2065         btrfs_set_header_nritems(split, c_nritems - mid);
2066         btrfs_set_header_nritems(c, mid);
2067         ret = 0;
2068
2069         btrfs_mark_buffer_dirty(c);
2070         btrfs_mark_buffer_dirty(split);
2071
2072         btrfs_node_key(split, &disk_key, 0);
2073         wret = insert_ptr(trans, root, path, &disk_key, split->start,
2074                           path->slots[level + 1] + 1,
2075                           level + 1);
2076         if (wret)
2077                 ret = wret;
2078
2079         ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2080         BUG_ON(ret);
2081
2082         if (path->slots[level] >= mid) {
2083                 path->slots[level] -= mid;
2084                 btrfs_tree_unlock(c);
2085                 free_extent_buffer(c);
2086                 path->nodes[level] = split;
2087                 path->slots[level + 1] += 1;
2088         } else {
2089                 btrfs_tree_unlock(split);
2090                 free_extent_buffer(split);
2091         }
2092         return ret;
2093 }
2094
2095 /*
2096  * how many bytes are required to store the items in a leaf.  start
2097  * and nr indicate which items in the leaf to check.  This totals up the
2098  * space used both by the item structs and the item data
2099  */
2100 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2101 {
2102         int data_len;
2103         int nritems = btrfs_header_nritems(l);
2104         int end = min(nritems, start + nr) - 1;
2105
2106         if (!nr)
2107                 return 0;
2108         data_len = btrfs_item_end_nr(l, start);
2109         data_len = data_len - btrfs_item_offset_nr(l, end);
2110         data_len += sizeof(struct btrfs_item) * nr;
2111         WARN_ON(data_len < 0);
2112         return data_len;
2113 }
2114
2115 /*
2116  * The space between the end of the leaf items and
2117  * the start of the leaf data.  IOW, how much room
2118  * the leaf has left for both items and data
2119  */
2120 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2121                                    struct extent_buffer *leaf)
2122 {
2123         int nritems = btrfs_header_nritems(leaf);
2124         int ret;
2125         ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2126         if (ret < 0) {
2127                 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2128                        "used %d nritems %d\n",
2129                        ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2130                        leaf_space_used(leaf, 0, nritems), nritems);
2131         }
2132         return ret;
2133 }
2134
2135 /*
2136  * push some data in the path leaf to the right, trying to free up at
2137  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2138  *
2139  * returns 1 if the push failed because the other node didn't have enough
2140  * room, 0 if everything worked out and < 0 if there were major errors.
2141  */
2142 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2143                            *root, struct btrfs_path *path, int data_size,
2144                            int empty)
2145 {
2146         struct extent_buffer *left = path->nodes[0];
2147         struct extent_buffer *right;
2148         struct extent_buffer *upper;
2149         struct btrfs_disk_key disk_key;
2150         int slot;
2151         u32 i;
2152         int free_space;
2153         int push_space = 0;
2154         int push_items = 0;
2155         struct btrfs_item *item;
2156         u32 left_nritems;
2157         u32 nr;
2158         u32 right_nritems;
2159         u32 data_end;
2160         u32 this_item_size;
2161         int ret;
2162
2163         slot = path->slots[1];
2164         if (!path->nodes[1])
2165                 return 1;
2166
2167         upper = path->nodes[1];
2168         if (slot >= btrfs_header_nritems(upper) - 1)
2169                 return 1;
2170
2171         WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2172
2173         right = read_node_slot(root, upper, slot + 1);
2174         btrfs_tree_lock(right);
2175         free_space = btrfs_leaf_free_space(root, right);
2176         if (free_space < data_size)
2177                 goto out_unlock;
2178
2179         /* cow and double check */
2180         ret = btrfs_cow_block(trans, root, right, upper,
2181                               slot + 1, &right, 0);
2182         if (ret)
2183                 goto out_unlock;
2184
2185         free_space = btrfs_leaf_free_space(root, right);
2186         if (free_space < data_size)
2187                 goto out_unlock;
2188
2189         left_nritems = btrfs_header_nritems(left);
2190         if (left_nritems == 0)
2191                 goto out_unlock;
2192
2193         if (empty)
2194                 nr = 0;
2195         else
2196                 nr = 1;
2197
2198         if (path->slots[0] >= left_nritems)
2199                 push_space += data_size;
2200
2201         i = left_nritems - 1;
2202         while (i >= nr) {
2203                 item = btrfs_item_nr(left, i);
2204
2205                 if (!empty && push_items > 0) {
2206                         if (path->slots[0] > i)
2207                                 break;
2208                         if (path->slots[0] == i) {
2209                                 int space = btrfs_leaf_free_space(root, left);
2210                                 if (space + push_space * 2 > free_space)
2211                                         break;
2212                         }
2213                 }
2214
2215                 if (path->slots[0] == i)
2216                         push_space += data_size;
2217
2218                 if (!left->map_token) {
2219                         map_extent_buffer(left, (unsigned long)item,
2220                                         sizeof(struct btrfs_item),
2221                                         &left->map_token, &left->kaddr,
2222                                         &left->map_start, &left->map_len,
2223                                         KM_USER1);
2224                 }
2225
2226                 this_item_size = btrfs_item_size(left, item);
2227                 if (this_item_size + sizeof(*item) + push_space > free_space)
2228                         break;
2229
2230                 push_items++;
2231                 push_space += this_item_size + sizeof(*item);
2232                 if (i == 0)
2233                         break;
2234                 i--;
2235         }
2236         if (left->map_token) {
2237                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2238                 left->map_token = NULL;
2239         }
2240
2241         if (push_items == 0)
2242                 goto out_unlock;
2243
2244         if (!empty && push_items == left_nritems)
2245                 WARN_ON(1);
2246
2247         /* push left to right */
2248         right_nritems = btrfs_header_nritems(right);
2249
2250         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2251         push_space -= leaf_data_end(root, left);
2252
2253         /* make room in the right data area */
2254         data_end = leaf_data_end(root, right);
2255         memmove_extent_buffer(right,
2256                               btrfs_leaf_data(right) + data_end - push_space,
2257                               btrfs_leaf_data(right) + data_end,
2258                               BTRFS_LEAF_DATA_SIZE(root) - data_end);
2259
2260         /* copy from the left data area */
2261         copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2262                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2263                      btrfs_leaf_data(left) + leaf_data_end(root, left),
2264                      push_space);
2265
2266         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2267                               btrfs_item_nr_offset(0),
2268                               right_nritems * sizeof(struct btrfs_item));
2269
2270         /* copy the items from left to right */
2271         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2272                    btrfs_item_nr_offset(left_nritems - push_items),
2273                    push_items * sizeof(struct btrfs_item));
2274
2275         /* update the item pointers */
2276         right_nritems += push_items;
2277         btrfs_set_header_nritems(right, right_nritems);
2278         push_space = BTRFS_LEAF_DATA_SIZE(root);
2279         for (i = 0; i < right_nritems; i++) {
2280                 item = btrfs_item_nr(right, i);
2281                 if (!right->map_token) {
2282                         map_extent_buffer(right, (unsigned long)item,
2283                                         sizeof(struct btrfs_item),
2284                                         &right->map_token, &right->kaddr,
2285                                         &right->map_start, &right->map_len,
2286                                         KM_USER1);
2287                 }
2288                 push_space -= btrfs_item_size(right, item);
2289                 btrfs_set_item_offset(right, item, push_space);
2290         }
2291
2292         if (right->map_token) {
2293                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2294                 right->map_token = NULL;
2295         }
2296         left_nritems -= push_items;
2297         btrfs_set_header_nritems(left, left_nritems);
2298
2299         if (left_nritems)
2300                 btrfs_mark_buffer_dirty(left);
2301         btrfs_mark_buffer_dirty(right);
2302
2303         ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2304         BUG_ON(ret);
2305
2306         btrfs_item_key(right, &disk_key, 0);
2307         btrfs_set_node_key(upper, &disk_key, slot + 1);
2308         btrfs_mark_buffer_dirty(upper);
2309
2310         /* then fixup the leaf pointer in the path */
2311         if (path->slots[0] >= left_nritems) {
2312                 path->slots[0] -= left_nritems;
2313                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2314                         clean_tree_block(trans, root, path->nodes[0]);
2315                 btrfs_tree_unlock(path->nodes[0]);
2316                 free_extent_buffer(path->nodes[0]);
2317                 path->nodes[0] = right;
2318                 path->slots[1] += 1;
2319         } else {
2320                 btrfs_tree_unlock(right);
2321                 free_extent_buffer(right);
2322         }
2323         return 0;
2324
2325 out_unlock:
2326         btrfs_tree_unlock(right);
2327         free_extent_buffer(right);
2328         return 1;
2329 }
2330
2331 /*
2332  * push some data in the path leaf to the left, trying to free up at
2333  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2334  */
2335 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2336                           *root, struct btrfs_path *path, int data_size,
2337                           int empty)
2338 {
2339         struct btrfs_disk_key disk_key;
2340         struct extent_buffer *right = path->nodes[0];
2341         struct extent_buffer *left;
2342         int slot;
2343         int i;
2344         int free_space;
2345         int push_space = 0;
2346         int push_items = 0;
2347         struct btrfs_item *item;
2348         u32 old_left_nritems;
2349         u32 right_nritems;
2350         u32 nr;
2351         int ret = 0;
2352         int wret;
2353         u32 this_item_size;
2354         u32 old_left_item_size;
2355
2356         slot = path->slots[1];
2357         if (slot == 0)
2358                 return 1;
2359         if (!path->nodes[1])
2360                 return 1;
2361
2362         right_nritems = btrfs_header_nritems(right);
2363         if (right_nritems == 0)
2364                 return 1;
2365
2366         WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2367
2368         left = read_node_slot(root, path->nodes[1], slot - 1);
2369         btrfs_tree_lock(left);
2370         free_space = btrfs_leaf_free_space(root, left);
2371         if (free_space < data_size) {
2372                 ret = 1;
2373                 goto out;
2374         }
2375
2376         /* cow and double check */
2377         ret = btrfs_cow_block(trans, root, left,
2378                               path->nodes[1], slot - 1, &left, 0);
2379         if (ret) {
2380                 /* we hit -ENOSPC, but it isn't fatal here */
2381                 ret = 1;
2382                 goto out;
2383         }
2384
2385         free_space = btrfs_leaf_free_space(root, left);
2386         if (free_space < data_size) {
2387                 ret = 1;
2388                 goto out;
2389         }
2390
2391         if (empty)
2392                 nr = right_nritems;
2393         else
2394                 nr = right_nritems - 1;
2395
2396         for (i = 0; i < nr; i++) {
2397                 item = btrfs_item_nr(right, i);
2398                 if (!right->map_token) {
2399                         map_extent_buffer(right, (unsigned long)item,
2400                                         sizeof(struct btrfs_item),
2401                                         &right->map_token, &right->kaddr,
2402                                         &right->map_start, &right->map_len,
2403                                         KM_USER1);
2404                 }
2405
2406                 if (!empty && push_items > 0) {
2407                         if (path->slots[0] < i)
2408                                 break;
2409                         if (path->slots[0] == i) {
2410                                 int space = btrfs_leaf_free_space(root, right);
2411                                 if (space + push_space * 2 > free_space)
2412                                         break;
2413                         }
2414                 }
2415
2416                 if (path->slots[0] == i)
2417                         push_space += data_size;
2418
2419                 this_item_size = btrfs_item_size(right, item);
2420                 if (this_item_size + sizeof(*item) + push_space > free_space)
2421                         break;
2422
2423                 push_items++;
2424                 push_space += this_item_size + sizeof(*item);
2425         }
2426
2427         if (right->map_token) {
2428                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2429                 right->map_token = NULL;
2430         }
2431
2432         if (push_items == 0) {
2433                 ret = 1;
2434                 goto out;
2435         }
2436         if (!empty && push_items == btrfs_header_nritems(right))
2437                 WARN_ON(1);
2438
2439         /* push data from right to left */
2440         copy_extent_buffer(left, right,
2441                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2442                            btrfs_item_nr_offset(0),
2443                            push_items * sizeof(struct btrfs_item));
2444
2445         push_space = BTRFS_LEAF_DATA_SIZE(root) -
2446                      btrfs_item_offset_nr(right, push_items - 1);
2447
2448         copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2449                      leaf_data_end(root, left) - push_space,
2450                      btrfs_leaf_data(right) +
2451                      btrfs_item_offset_nr(right, push_items - 1),
2452                      push_space);
2453         old_left_nritems = btrfs_header_nritems(left);
2454         BUG_ON(old_left_nritems <= 0);
2455
2456         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2457         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2458                 u32 ioff;
2459
2460                 item = btrfs_item_nr(left, i);
2461                 if (!left->map_token) {
2462                         map_extent_buffer(left, (unsigned long)item,
2463                                         sizeof(struct btrfs_item),
2464                                         &left->map_token, &left->kaddr,
2465                                         &left->map_start, &left->map_len,
2466                                         KM_USER1);
2467                 }
2468
2469                 ioff = btrfs_item_offset(left, item);
2470                 btrfs_set_item_offset(left, item,
2471                       ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2472         }
2473         btrfs_set_header_nritems(left, old_left_nritems + push_items);
2474         if (left->map_token) {
2475                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2476                 left->map_token = NULL;
2477         }
2478
2479         /* fixup right node */
2480         if (push_items > right_nritems) {
2481                 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2482                        right_nritems);
2483                 WARN_ON(1);
2484         }
2485
2486         if (push_items < right_nritems) {
2487                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2488                                                   leaf_data_end(root, right);
2489                 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2490                                       BTRFS_LEAF_DATA_SIZE(root) - push_space,
2491                                       btrfs_leaf_data(right) +
2492                                       leaf_data_end(root, right), push_space);
2493
2494                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2495                               btrfs_item_nr_offset(push_items),
2496                              (btrfs_header_nritems(right) - push_items) *
2497                              sizeof(struct btrfs_item));
2498         }
2499         right_nritems -= push_items;
2500         btrfs_set_header_nritems(right, right_nritems);
2501         push_space = BTRFS_LEAF_DATA_SIZE(root);
2502         for (i = 0; i < right_nritems; i++) {
2503                 item = btrfs_item_nr(right, i);
2504
2505                 if (!right->map_token) {
2506                         map_extent_buffer(right, (unsigned long)item,
2507                                         sizeof(struct btrfs_item),
2508                                         &right->map_token, &right->kaddr,
2509                                         &right->map_start, &right->map_len,
2510                                         KM_USER1);
2511                 }
2512
2513                 push_space = push_space - btrfs_item_size(right, item);
2514                 btrfs_set_item_offset(right, item, push_space);
2515         }
2516         if (right->map_token) {
2517                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2518                 right->map_token = NULL;
2519         }
2520
2521         btrfs_mark_buffer_dirty(left);
2522         if (right_nritems)
2523                 btrfs_mark_buffer_dirty(right);
2524
2525         ret = btrfs_update_ref(trans, root, right, left,
2526                                old_left_nritems, push_items);
2527         BUG_ON(ret);
2528
2529         btrfs_item_key(right, &disk_key, 0);
2530         wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2531         if (wret)
2532                 ret = wret;
2533
2534         /* then fixup the leaf pointer in the path */
2535         if (path->slots[0] < push_items) {
2536                 path->slots[0] += old_left_nritems;
2537                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2538                         clean_tree_block(trans, root, path->nodes[0]);
2539                 btrfs_tree_unlock(path->nodes[0]);
2540                 free_extent_buffer(path->nodes[0]);
2541                 path->nodes[0] = left;
2542                 path->slots[1] -= 1;
2543         } else {
2544                 btrfs_tree_unlock(left);
2545                 free_extent_buffer(left);
2546                 path->slots[0] -= push_items;
2547         }
2548         BUG_ON(path->slots[0] < 0);
2549         return ret;
2550 out:
2551         btrfs_tree_unlock(left);
2552         free_extent_buffer(left);
2553         return ret;
2554 }
2555
2556 /*
2557  * split the path's leaf in two, making sure there is at least data_size
2558  * available for the resulting leaf level of the path.
2559  *
2560  * returns 0 if all went well and < 0 on failure.
2561  */
2562 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2563                                struct btrfs_root *root,
2564                                struct btrfs_key *ins_key,
2565                                struct btrfs_path *path, int data_size,
2566                                int extend)
2567 {
2568         struct extent_buffer *l;
2569         u32 nritems;
2570         int mid;
2571         int slot;
2572         struct extent_buffer *right;
2573         int data_copy_size;
2574         int rt_data_off;
2575         int i;
2576         int ret = 0;
2577         int wret;
2578         int double_split;
2579         int num_doubles = 0;
2580         struct btrfs_disk_key disk_key;
2581
2582         /* first try to make some room by pushing left and right */
2583         if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2584                 wret = push_leaf_right(trans, root, path, data_size, 0);
2585                 if (wret < 0)
2586                         return wret;
2587                 if (wret) {
2588                         wret = push_leaf_left(trans, root, path, data_size, 0);
2589                         if (wret < 0)
2590                                 return wret;
2591                 }
2592                 l = path->nodes[0];
2593
2594                 /* did the pushes work? */
2595                 if (btrfs_leaf_free_space(root, l) >= data_size)
2596                         return 0;
2597         }
2598
2599         if (!path->nodes[1]) {
2600                 ret = insert_new_root(trans, root, path, 1);
2601                 if (ret)
2602                         return ret;
2603         }
2604 again:
2605         double_split = 0;
2606         l = path->nodes[0];
2607         slot = path->slots[0];
2608         nritems = btrfs_header_nritems(l);
2609         mid = (nritems + 1) / 2;
2610
2611         right = btrfs_alloc_free_block(trans, root, root->leafsize,
2612                                         path->nodes[1]->start,
2613                                         root->root_key.objectid,
2614                                         trans->transid, 0, l->start, 0);
2615         if (IS_ERR(right)) {
2616                 BUG_ON(1);
2617                 return PTR_ERR(right);
2618         }
2619
2620         memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2621         btrfs_set_header_bytenr(right, right->start);
2622         btrfs_set_header_generation(right, trans->transid);
2623         btrfs_set_header_owner(right, root->root_key.objectid);
2624         btrfs_set_header_level(right, 0);
2625         write_extent_buffer(right, root->fs_info->fsid,
2626                             (unsigned long)btrfs_header_fsid(right),
2627                             BTRFS_FSID_SIZE);
2628
2629         write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2630                             (unsigned long)btrfs_header_chunk_tree_uuid(right),
2631                             BTRFS_UUID_SIZE);
2632         if (mid <= slot) {
2633                 if (nritems == 1 ||
2634                     leaf_space_used(l, mid, nritems - mid) + data_size >
2635                         BTRFS_LEAF_DATA_SIZE(root)) {
2636                         if (slot >= nritems) {
2637                                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2638                                 btrfs_set_header_nritems(right, 0);
2639                                 wret = insert_ptr(trans, root, path,
2640                                                   &disk_key, right->start,
2641                                                   path->slots[1] + 1, 1);
2642                                 if (wret)
2643                                         ret = wret;
2644
2645                                 btrfs_tree_unlock(path->nodes[0]);
2646                                 free_extent_buffer(path->nodes[0]);
2647                                 path->nodes[0] = right;
2648                                 path->slots[0] = 0;
2649                                 path->slots[1] += 1;
2650                                 btrfs_mark_buffer_dirty(right);
2651                                 return ret;
2652                         }
2653                         mid = slot;
2654                         if (mid != nritems &&
2655                             leaf_space_used(l, mid, nritems - mid) +
2656                             data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2657                                 double_split = 1;
2658                         }
2659                 }
2660         } else {
2661                 if (leaf_space_used(l, 0, mid) + data_size >
2662                         BTRFS_LEAF_DATA_SIZE(root)) {
2663                         if (!extend && data_size && slot == 0) {
2664                                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2665                                 btrfs_set_header_nritems(right, 0);
2666                                 wret = insert_ptr(trans, root, path,
2667                                                   &disk_key,
2668                                                   right->start,
2669                                                   path->slots[1], 1);
2670                                 if (wret)
2671                                         ret = wret;
2672                                 btrfs_tree_unlock(path->nodes[0]);
2673                                 free_extent_buffer(path->nodes[0]);
2674                                 path->nodes[0] = right;
2675                                 path->slots[0] = 0;
2676                                 if (path->slots[1] == 0) {
2677                                         wret = fixup_low_keys(trans, root,
2678                                                       path, &disk_key, 1);
2679                                         if (wret)
2680                                                 ret = wret;
2681                                 }
2682                                 btrfs_mark_buffer_dirty(right);
2683                                 return ret;
2684                         } else if ((extend || !data_size) && slot == 0) {
2685                                 mid = 1;
2686                         } else {
2687                                 mid = slot;
2688                                 if (mid != nritems &&
2689                                     leaf_space_used(l, mid, nritems - mid) +
2690                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2691                                         double_split = 1;
2692                                 }
2693                         }
2694                 }
2695         }
2696         nritems = nritems - mid;
2697         btrfs_set_header_nritems(right, nritems);
2698         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2699
2700         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2701                            btrfs_item_nr_offset(mid),
2702                            nritems * sizeof(struct btrfs_item));
2703
2704         copy_extent_buffer(right, l,
2705                      btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2706                      data_copy_size, btrfs_leaf_data(l) +
2707                      leaf_data_end(root, l), data_copy_size);
2708
2709         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2710                       btrfs_item_end_nr(l, mid);
2711
2712         for (i = 0; i < nritems; i++) {
2713                 struct btrfs_item *item = btrfs_item_nr(right, i);
2714                 u32 ioff;
2715
2716                 if (!right->map_token) {
2717                         map_extent_buffer(right, (unsigned long)item,
2718                                         sizeof(struct btrfs_item),
2719                                         &right->map_token, &right->kaddr,
2720                                         &right->map_start, &right->map_len,
2721                                         KM_USER1);
2722                 }
2723
2724                 ioff = btrfs_item_offset(right, item);
2725                 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2726         }
2727
2728         if (right->map_token) {
2729                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2730                 right->map_token = NULL;
2731         }
2732
2733         btrfs_set_header_nritems(l, mid);
2734         ret = 0;
2735         btrfs_item_key(right, &disk_key, 0);
2736         wret = insert_ptr(trans, root, path, &disk_key, right->start,
2737                           path->slots[1] + 1, 1);
2738         if (wret)
2739                 ret = wret;
2740
2741         btrfs_mark_buffer_dirty(right);
2742         btrfs_mark_buffer_dirty(l);
2743         BUG_ON(path->slots[0] != slot);
2744
2745         ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2746         BUG_ON(ret);
2747
2748         if (mid <= slot) {
2749                 btrfs_tree_unlock(path->nodes[0]);
2750                 free_extent_buffer(path->nodes[0]);
2751                 path->nodes[0] = right;
2752                 path->slots[0] -= mid;
2753                 path->slots[1] += 1;
2754         } else {
2755                 btrfs_tree_unlock(right);
2756                 free_extent_buffer(right);
2757         }
2758
2759         BUG_ON(path->slots[0] < 0);
2760
2761         if (double_split) {
2762                 BUG_ON(num_doubles != 0);
2763                 num_doubles++;
2764                 goto again;
2765         }
2766         return ret;
2767 }
2768
2769 /*
2770  * This function splits a single item into two items,
2771  * giving 'new_key' to the new item and splitting the
2772  * old one at split_offset (from the start of the item).
2773  *
2774  * The path may be released by this operation.  After
2775  * the split, the path is pointing to the old item.  The
2776  * new item is going to be in the same node as the old one.
2777  *
2778  * Note, the item being split must be smaller enough to live alone on
2779  * a tree block with room for one extra struct btrfs_item
2780  *
2781  * This allows us to split the item in place, keeping a lock on the
2782  * leaf the entire time.
2783  */
2784 int btrfs_split_item(struct btrfs_trans_handle *trans,
2785                      struct btrfs_root *root,
2786                      struct btrfs_path *path,
2787                      struct btrfs_key *new_key,
2788                      unsigned long split_offset)
2789 {
2790         u32 item_size;
2791         struct extent_buffer *leaf;
2792         struct btrfs_key orig_key;
2793         struct btrfs_item *item;
2794         struct btrfs_item *new_item;
2795         int ret = 0;
2796         int slot;
2797         u32 nritems;
2798         u32 orig_offset;
2799         struct btrfs_disk_key disk_key;
2800         char *buf;
2801
2802         leaf = path->nodes[0];
2803         btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2804         if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2805                 goto split;
2806
2807         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2808         btrfs_release_path(root, path);
2809
2810         path->search_for_split = 1;
2811         path->keep_locks = 1;
2812
2813         ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2814         path->search_for_split = 0;
2815
2816         /* if our item isn't there or got smaller, return now */
2817         if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2818                                                         path->slots[0])) {
2819                 path->keep_locks = 0;
2820                 return -EAGAIN;
2821         }
2822
2823         ret = split_leaf(trans, root, &orig_key, path,
2824                          sizeof(struct btrfs_item), 1);
2825         path->keep_locks = 0;
2826         BUG_ON(ret);
2827
2828         leaf = path->nodes[0];
2829         BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2830
2831 split:
2832         item = btrfs_item_nr(leaf, path->slots[0]);
2833         orig_offset = btrfs_item_offset(leaf, item);
2834         item_size = btrfs_item_size(leaf, item);
2835
2836
2837         buf = kmalloc(item_size, GFP_NOFS);
2838         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2839                             path->slots[0]), item_size);
2840         slot = path->slots[0] + 1;
2841         leaf = path->nodes[0];
2842
2843         nritems = btrfs_header_nritems(leaf);
2844
2845         if (slot != nritems) {
2846                 /* shift the items */
2847                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2848                               btrfs_item_nr_offset(slot),
2849                               (nritems - slot) * sizeof(struct btrfs_item));
2850
2851         }
2852
2853         btrfs_cpu_key_to_disk(&disk_key, new_key);
2854         btrfs_set_item_key(leaf, &disk_key, slot);
2855
2856         new_item = btrfs_item_nr(leaf, slot);
2857
2858         btrfs_set_item_offset(leaf, new_item, orig_offset);
2859         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
2860
2861         btrfs_set_item_offset(leaf, item,
2862                               orig_offset + item_size - split_offset);
2863         btrfs_set_item_size(leaf, item, split_offset);
2864
2865         btrfs_set_header_nritems(leaf, nritems + 1);
2866
2867         /* write the data for the start of the original item */
2868         write_extent_buffer(leaf, buf,
2869                             btrfs_item_ptr_offset(leaf, path->slots[0]),
2870                             split_offset);
2871
2872         /* write the data for the new item */
2873         write_extent_buffer(leaf, buf + split_offset,
2874                             btrfs_item_ptr_offset(leaf, slot),
2875                             item_size - split_offset);
2876         btrfs_mark_buffer_dirty(leaf);
2877
2878         ret = 0;
2879         if (btrfs_leaf_free_space(root, leaf) < 0) {
2880                 btrfs_print_leaf(root, leaf);
2881                 BUG();
2882         }
2883         kfree(buf);
2884         return ret;
2885 }
2886
2887 /*
2888  * make the item pointed to by the path smaller.  new_size indicates
2889  * how small to make it, and from_end tells us if we just chop bytes
2890  * off the end of the item or if we shift the item to chop bytes off
2891  * the front.
2892  */
2893 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2894                         struct btrfs_root *root,
2895                         struct btrfs_path *path,
2896                         u32 new_size, int from_end)
2897 {
2898         int ret = 0;
2899         int slot;
2900         int slot_orig;
2901         struct extent_buffer *leaf;
2902         struct btrfs_item *item;
2903         u32 nritems;
2904         unsigned int data_end;
2905         unsigned int old_data_start;
2906         unsigned int old_size;
2907         unsigned int size_diff;
2908         int i;
2909
2910         slot_orig = path->slots[0];
2911         leaf = path->nodes[0];
2912         slot = path->slots[0];
2913
2914         old_size = btrfs_item_size_nr(leaf, slot);
2915         if (old_size == new_size)
2916                 return 0;
2917
2918         nritems = btrfs_header_nritems(leaf);
2919         data_end = leaf_data_end(root, leaf);
2920
2921         old_data_start = btrfs_item_offset_nr(leaf, slot);
2922
2923         size_diff = old_size - new_size;
2924
2925         BUG_ON(slot < 0);
2926         BUG_ON(slot >= nritems);
2927
2928         /*
2929          * item0..itemN ... dataN.offset..dataN.size .. data0.size
2930          */
2931         /* first correct the data pointers */
2932         for (i = slot; i < nritems; i++) {
2933                 u32 ioff;
2934                 item = btrfs_item_nr(leaf, i);
2935
2936                 if (!leaf->map_token) {
2937                         map_extent_buffer(leaf, (unsigned long)item,
2938                                         sizeof(struct btrfs_item),
2939                                         &leaf->map_token, &leaf->kaddr,
2940                                         &leaf->map_start, &leaf->map_len,
2941                                         KM_USER1);
2942                 }
2943
2944                 ioff = btrfs_item_offset(leaf, item);
2945                 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2946         }
2947
2948         if (leaf->map_token) {
2949                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2950                 leaf->map_token = NULL;
2951         }
2952
2953         /* shift the data */
2954         if (from_end) {
2955                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2956                               data_end + size_diff, btrfs_leaf_data(leaf) +
2957                               data_end, old_data_start + new_size - data_end);
2958         } else {
2959                 struct btrfs_disk_key disk_key;
2960                 u64 offset;
2961
2962                 btrfs_item_key(leaf, &disk_key, slot);
2963
2964                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2965                         unsigned long ptr;
2966                         struct btrfs_file_extent_item *fi;
2967
2968                         fi = btrfs_item_ptr(leaf, slot,
2969                                             struct btrfs_file_extent_item);
2970                         fi = (struct btrfs_file_extent_item *)(
2971                              (unsigned long)fi - size_diff);
2972
2973                         if (btrfs_file_extent_type(leaf, fi) ==
2974                             BTRFS_FILE_EXTENT_INLINE) {
2975                                 ptr = btrfs_item_ptr_offset(leaf, slot);
2976                                 memmove_extent_buffer(leaf, ptr,
2977                                       (unsigned long)fi,
2978                                       offsetof(struct btrfs_file_extent_item,
2979                                                  disk_bytenr));
2980                         }
2981                 }
2982
2983                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2984                               data_end + size_diff, btrfs_leaf_data(leaf) +
2985                               data_end, old_data_start - data_end);
2986
2987                 offset = btrfs_disk_key_offset(&disk_key);
2988                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2989                 btrfs_set_item_key(leaf, &disk_key, slot);
2990                 if (slot == 0)
2991                         fixup_low_keys(trans, root, path, &disk_key, 1);
2992         }
2993
2994         item = btrfs_item_nr(leaf, slot);
2995         btrfs_set_item_size(leaf, item, new_size);
2996         btrfs_mark_buffer_dirty(leaf);
2997
2998         ret = 0;
2999         if (btrfs_leaf_free_space(root, leaf) < 0) {
3000                 btrfs_print_leaf(root, leaf);
3001                 BUG();
3002         }
3003         return ret;
3004 }
3005
3006 /*
3007  * make the item pointed to by the path bigger, data_size is the new size.
3008  */
3009 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3010                       struct btrfs_root *root, struct btrfs_path *path,
3011                       u32 data_size)
3012 {
3013         int ret = 0;
3014         int slot;
3015         int slot_orig;
3016         struct extent_buffer *leaf;
3017         struct btrfs_item *item;
3018         u32 nritems;
3019         unsigned int data_end;
3020         unsigned int old_data;
3021         unsigned int old_size;
3022         int i;
3023
3024         slot_orig = path->slots[0];
3025         leaf = path->nodes[0];
3026
3027         nritems = btrfs_header_nritems(leaf);
3028         data_end = leaf_data_end(root, leaf);
3029
3030         if (btrfs_leaf_free_space(root, leaf) < data_size) {
3031                 btrfs_print_leaf(root, leaf);
3032                 BUG();
3033         }
3034         slot = path->slots[0];
3035         old_data = btrfs_item_end_nr(leaf, slot);
3036
3037         BUG_ON(slot < 0);
3038         if (slot >= nritems) {
3039                 btrfs_print_leaf(root, leaf);
3040                 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3041                        slot, nritems);
3042                 BUG_ON(1);
3043         }
3044
3045         /*
3046          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3047          */
3048         /* first correct the data pointers */
3049         for (i = slot; i < nritems; i++) {
3050                 u32 ioff;
3051                 item = btrfs_item_nr(leaf, i);
3052
3053                 if (!leaf->map_token) {
3054                         map_extent_buffer(leaf, (unsigned long)item,
3055                                         sizeof(struct btrfs_item),
3056                                         &leaf->map_token, &leaf->kaddr,
3057                                         &leaf->map_start, &leaf->map_len,
3058                                         KM_USER1);
3059                 }
3060                 ioff = btrfs_item_offset(leaf, item);
3061                 btrfs_set_item_offset(leaf, item, ioff - data_size);
3062         }
3063
3064         if (leaf->map_token) {
3065                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3066                 leaf->map_token = NULL;
3067         }
3068
3069         /* shift the data */
3070         memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3071                       data_end - data_size, btrfs_leaf_data(leaf) +
3072                       data_end, old_data - data_end);
3073
3074         data_end = old_data;
3075         old_size = btrfs_item_size_nr(leaf, slot);
3076         item = btrfs_item_nr(leaf, slot);
3077         btrfs_set_item_size(leaf, item, old_size + data_size);
3078         btrfs_mark_buffer_dirty(leaf);
3079
3080         ret = 0;
3081         if (btrfs_leaf_free_space(root, leaf) < 0) {
3082                 btrfs_print_leaf(root, leaf);
3083                 BUG();
3084         }
3085         return ret;
3086 }
3087
3088 /*
3089  * Given a key and some data, insert items into the tree.
3090  * This does all the path init required, making room in the tree if needed.
3091  * Returns the number of keys that were inserted.
3092  */
3093 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3094                             struct btrfs_root *root,
3095                             struct btrfs_path *path,
3096                             struct btrfs_key *cpu_key, u32 *data_size,
3097                             int nr)
3098 {
3099         struct extent_buffer *leaf;
3100         struct btrfs_item *item;
3101         int ret = 0;
3102         int slot;
3103         int i;
3104         u32 nritems;
3105         u32 total_data = 0;
3106         u32 total_size = 0;
3107         unsigned int data_end;
3108         struct btrfs_disk_key disk_key;
3109         struct btrfs_key found_key;
3110
3111         for (i = 0; i < nr; i++) {
3112                 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3113                     BTRFS_LEAF_DATA_SIZE(root)) {
3114                         break;
3115                         nr = i;
3116                 }
3117                 total_data += data_size[i];
3118                 total_size += data_size[i] + sizeof(struct btrfs_item);
3119         }
3120         BUG_ON(nr == 0);
3121
3122         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3123         if (ret == 0)
3124                 return -EEXIST;
3125         if (ret < 0)
3126                 goto out;
3127
3128         leaf = path->nodes[0];
3129
3130         nritems = btrfs_header_nritems(leaf);
3131         data_end = leaf_data_end(root, leaf);
3132
3133         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3134                 for (i = nr; i >= 0; i--) {
3135                         total_data -= data_size[i];
3136                         total_size -= data_size[i] + sizeof(struct btrfs_item);
3137                         if (total_size < btrfs_leaf_free_space(root, leaf))
3138                                 break;
3139                 }
3140                 nr = i;
3141         }
3142
3143         slot = path->slots[0];
3144         BUG_ON(slot < 0);
3145
3146         if (slot != nritems) {
3147                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3148
3149                 item = btrfs_item_nr(leaf, slot);
3150                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3151
3152                 /* figure out how many keys we can insert in here */
3153                 total_data = data_size[0];
3154                 for (i = 1; i < nr; i++) {
3155                         if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3156                                 break;
3157                         total_data += data_size[i];
3158                 }
3159                 nr = i;
3160
3161                 if (old_data < data_end) {
3162                         btrfs_print_leaf(root, leaf);
3163                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3164                                slot, old_data, data_end);
3165                         BUG_ON(1);
3166                 }
3167                 /*
3168                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3169                  */
3170                 /* first correct the data pointers */
3171                 WARN_ON(leaf->map_token);
3172                 for (i = slot; i < nritems; i++) {
3173                         u32 ioff;
3174
3175                         item = btrfs_item_nr(leaf, i);
3176                         if (!leaf->map_token) {
3177                                 map_extent_buffer(leaf, (unsigned long)item,
3178                                         sizeof(struct btrfs_item),
3179                                         &leaf->map_token, &leaf->kaddr,
3180                                         &leaf->map_start, &leaf->map_len,
3181                                         KM_USER1);
3182                         }
3183
3184                         ioff = btrfs_item_offset(leaf, item);
3185                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3186                 }
3187                 if (leaf->map_token) {
3188                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3189                         leaf->map_token = NULL;
3190                 }
3191
3192                 /* shift the items */
3193                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3194                               btrfs_item_nr_offset(slot),
3195                               (nritems - slot) * sizeof(struct btrfs_item));
3196
3197                 /* shift the data */
3198                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3199                               data_end - total_data, btrfs_leaf_data(leaf) +
3200                               data_end, old_data - data_end);
3201                 data_end = old_data;
3202         } else {
3203                 /*
3204                  * this sucks but it has to be done, if we are inserting at
3205                  * the end of the leaf only insert 1 of the items, since we
3206                  * have no way of knowing whats on the next leaf and we'd have
3207                  * to drop our current locks to figure it out
3208                  */
3209                 nr = 1;
3210         }
3211
3212         /* setup the item for the new data */
3213         for (i = 0; i < nr; i++) {
3214                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3215                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3216                 item = btrfs_item_nr(leaf, slot + i);
3217                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3218                 data_end -= data_size[i];
3219                 btrfs_set_item_size(leaf, item, data_size[i]);
3220         }
3221         btrfs_set_header_nritems(leaf, nritems + nr);
3222         btrfs_mark_buffer_dirty(leaf);
3223
3224         ret = 0;
3225         if (slot == 0) {
3226                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3227                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3228         }
3229
3230         if (btrfs_leaf_free_space(root, leaf) < 0) {
3231                 btrfs_print_leaf(root, leaf);
3232                 BUG();
3233         }
3234 out:
3235         if (!ret)
3236                 ret = nr;
3237         return ret;
3238 }
3239
3240 /*
3241  * Given a key and some data, insert items into the tree.
3242  * This does all the path init required, making room in the tree if needed.
3243  */
3244 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3245                             struct btrfs_root *root,
3246                             struct btrfs_path *path,
3247                             struct btrfs_key *cpu_key, u32 *data_size,
3248                             int nr)
3249 {
3250         struct extent_buffer *leaf;
3251         struct btrfs_item *item;
3252         int ret = 0;
3253         int slot;
3254         int slot_orig;
3255         int i;
3256         u32 nritems;
3257         u32 total_size = 0;
3258         u32 total_data = 0;
3259         unsigned int data_end;
3260         struct btrfs_disk_key disk_key;
3261
3262         for (i = 0; i < nr; i++)
3263                 total_data += data_size[i];
3264
3265         total_size = total_data + (nr * sizeof(struct btrfs_item));
3266         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3267         if (ret == 0)
3268                 return -EEXIST;
3269         if (ret < 0)
3270                 goto out;
3271
3272         slot_orig = path->slots[0];
3273         leaf = path->nodes[0];
3274
3275         nritems = btrfs_header_nritems(leaf);
3276         data_end = leaf_data_end(root, leaf);
3277
3278         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3279                 btrfs_print_leaf(root, leaf);
3280                 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3281                        total_size, btrfs_leaf_free_space(root, leaf));
3282                 BUG();
3283         }
3284
3285         slot = path->slots[0];
3286         BUG_ON(slot < 0);
3287
3288         if (slot != nritems) {
3289                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3290
3291                 if (old_data < data_end) {
3292                         btrfs_print_leaf(root, leaf);
3293                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3294                                slot, old_data, data_end);
3295                         BUG_ON(1);
3296                 }
3297                 /*
3298                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3299                  */
3300                 /* first correct the data pointers */
3301                 WARN_ON(leaf->map_token);
3302                 for (i = slot; i < nritems; i++) {
3303                         u32 ioff;
3304
3305                         item = btrfs_item_nr(leaf, i);
3306                         if (!leaf->map_token) {
3307                                 map_extent_buffer(leaf, (unsigned long)item,
3308                                         sizeof(struct btrfs_item),
3309                                         &leaf->map_token, &leaf->kaddr,
3310                                         &leaf->map_start, &leaf->map_len,
3311                                         KM_USER1);
3312                         }
3313
3314                         ioff = btrfs_item_offset(leaf, item);
3315                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3316                 }
3317                 if (leaf->map_token) {
3318                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3319                         leaf->map_token = NULL;
3320                 }
3321
3322                 /* shift the items */
3323                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3324                               btrfs_item_nr_offset(slot),
3325                               (nritems - slot) * sizeof(struct btrfs_item));
3326
3327                 /* shift the data */
3328                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3329                               data_end - total_data, btrfs_leaf_data(leaf) +
3330                               data_end, old_data - data_end);
3331                 data_end = old_data;
3332         }
3333
3334         /* setup the item for the new data */
3335         for (i = 0; i < nr; i++) {
3336                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3337                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3338                 item = btrfs_item_nr(leaf, slot + i);
3339                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3340                 data_end -= data_size[i];
3341                 btrfs_set_item_size(leaf, item, data_size[i]);
3342         }
3343         btrfs_set_header_nritems(leaf, nritems + nr);
3344         btrfs_mark_buffer_dirty(leaf);
3345
3346         ret = 0;
3347         if (slot == 0) {
3348                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3349                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3350         }
3351
3352         if (btrfs_leaf_free_space(root, leaf) < 0) {
3353                 btrfs_print_leaf(root, leaf);
3354                 BUG();
3355         }
3356 out:
3357         return ret;
3358 }
3359
3360 /*
3361  * Given a key and some data, insert an item into the tree.
3362  * This does all the path init required, making room in the tree if needed.
3363  */
3364 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3365                       *root, struct btrfs_key *cpu_key, void *data, u32
3366                       data_size)
3367 {
3368         int ret = 0;
3369         struct btrfs_path *path;
3370         struct extent_buffer *leaf;
3371         unsigned long ptr;
3372
3373         path = btrfs_alloc_path();
3374         BUG_ON(!path);
3375         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3376         if (!ret) {
3377                 leaf = path->nodes[0];
3378                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3379                 write_extent_buffer(leaf, data, ptr, data_size);
3380                 btrfs_mark_buffer_dirty(leaf);
3381         }
3382         btrfs_free_path(path);
3383         return ret;
3384 }
3385
3386 /*
3387  * delete the pointer from a given node.
3388  *
3389  * the tree should have been previously balanced so the deletion does not
3390  * empty a node.
3391  */
3392 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3393                    struct btrfs_path *path, int level, int slot)
3394 {
3395         struct extent_buffer *parent = path->nodes[level];
3396         u32 nritems;
3397         int ret = 0;
3398         int wret;
3399
3400         nritems = btrfs_header_nritems(parent);
3401         if (slot != nritems - 1) {
3402                 memmove_extent_buffer(parent,
3403                               btrfs_node_key_ptr_offset(slot),
3404                               btrfs_node_key_ptr_offset(slot + 1),
3405                               sizeof(struct btrfs_key_ptr) *
3406                               (nritems - slot - 1));
3407         }
3408         nritems--;
3409         btrfs_set_header_nritems(parent, nritems);
3410         if (nritems == 0 && parent == root->node) {
3411                 BUG_ON(btrfs_header_level(root->node) != 1);
3412                 /* just turn the root into a leaf and break */
3413                 btrfs_set_header_level(root->node, 0);
3414         } else if (slot == 0) {
3415                 struct btrfs_disk_key disk_key;
3416
3417                 btrfs_node_key(parent, &disk_key, 0);
3418                 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3419                 if (wret)
3420                         ret = wret;
3421         }
3422         btrfs_mark_buffer_dirty(parent);
3423         return ret;
3424 }
3425
3426 /*
3427  * a helper function to delete the leaf pointed to by path->slots[1] and
3428  * path->nodes[1].  bytenr is the node block pointer, but since the callers
3429  * already know it, it is faster to have them pass it down than to
3430  * read it out of the node again.
3431  *
3432  * This deletes the pointer in path->nodes[1] and frees the leaf
3433  * block extent.  zero is returned if it all worked out, < 0 otherwise.
3434  *
3435  * The path must have already been setup for deleting the leaf, including
3436  * all the proper balancing.  path->nodes[1] must be locked.
3437  */
3438 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3439                             struct btrfs_root *root,
3440                             struct btrfs_path *path, u64 bytenr)
3441 {
3442         int ret;
3443         u64 root_gen = btrfs_header_generation(path->nodes[1]);
3444
3445         ret = del_ptr(trans, root, path, 1, path->slots[1]);
3446         if (ret)
3447                 return ret;
3448
3449         ret = btrfs_free_extent(trans, root, bytenr,
3450                                 btrfs_level_size(root, 0),
3451                                 path->nodes[1]->start,
3452                                 btrfs_header_owner(path->nodes[1]),
3453                                 root_gen, 0, 1);
3454         return ret;
3455 }
3456 /*
3457  * delete the item at the leaf level in path.  If that empties
3458  * the leaf, remove it from the tree
3459  */
3460 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3461                     struct btrfs_path *path, int slot, int nr)
3462 {
3463         struct extent_buffer *leaf;
3464         struct btrfs_item *item;
3465         int last_off;
3466         int dsize = 0;
3467         int ret = 0;
3468         int wret;
3469         int i;
3470         u32 nritems;
3471
3472         leaf = path->nodes[0];
3473         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3474
3475         for (i = 0; i < nr; i++)
3476                 dsize += btrfs_item_size_nr(leaf, slot + i);
3477
3478         nritems = btrfs_header_nritems(leaf);
3479
3480         if (slot + nr != nritems) {
3481                 int data_end = leaf_data_end(root, leaf);
3482
3483                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3484                               data_end + dsize,
3485                               btrfs_leaf_data(leaf) + data_end,
3486                               last_off - data_end);
3487
3488                 for (i = slot + nr; i < nritems; i++) {
3489                         u32 ioff;
3490
3491                         item = btrfs_item_nr(leaf, i);
3492                         if (!leaf->map_token) {
3493                                 map_extent_buffer(leaf, (unsigned long)item,
3494                                         sizeof(struct btrfs_item),
3495                                         &leaf->map_token, &leaf->kaddr,
3496                                         &leaf->map_start, &leaf->map_len,
3497                                         KM_USER1);
3498                         }
3499                         ioff = btrfs_item_offset(leaf, item);
3500                         btrfs_set_item_offset(leaf, item, ioff + dsize);
3501                 }
3502
3503                 if (leaf->map_token) {
3504                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3505                         leaf->map_token = NULL;
3506                 }
3507
3508                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3509                               btrfs_item_nr_offset(slot + nr),
3510                               sizeof(struct btrfs_item) *
3511                               (nritems - slot - nr));
3512         }
3513         btrfs_set_header_nritems(leaf, nritems - nr);
3514         nritems -= nr;
3515
3516         /* delete the leaf if we've emptied it */
3517         if (nritems == 0) {
3518                 if (leaf == root->node) {
3519                         btrfs_set_header_level(leaf, 0);
3520                 } else {
3521                         ret = btrfs_del_leaf(trans, root, path, leaf->start);
3522                         BUG_ON(ret);
3523                 }
3524         } else {
3525                 int used = leaf_space_used(leaf, 0, nritems);
3526                 if (slot == 0) {
3527                         struct btrfs_disk_key disk_key;
3528
3529                         btrfs_item_key(leaf, &disk_key, 0);
3530                         wret = fixup_low_keys(trans, root, path,
3531                                               &disk_key, 1);
3532                         if (wret)
3533                                 ret = wret;
3534                 }
3535
3536                 /* delete the leaf if it is mostly empty */
3537                 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3538                         /* push_leaf_left fixes the path.
3539                          * make sure the path still points to our leaf
3540                          * for possible call to del_ptr below
3541                          */
3542                         slot = path->slots[1];
3543                         extent_buffer_get(leaf);
3544
3545                         wret = push_leaf_left(trans, root, path, 1, 1);
3546                         if (wret < 0 && wret != -ENOSPC)
3547                                 ret = wret;
3548
3549                         if (path->nodes[0] == leaf &&
3550                             btrfs_header_nritems(leaf)) {
3551                                 wret = push_leaf_right(trans, root, path, 1, 1);
3552                                 if (wret < 0 && wret != -ENOSPC)
3553                                         ret = wret;
3554                         }
3555
3556                         if (btrfs_header_nritems(leaf) == 0) {
3557                                 path->slots[1] = slot;
3558                                 ret = btrfs_del_leaf(trans, root, path,
3559                                                      leaf->start);
3560                                 BUG_ON(ret);
3561                                 free_extent_buffer(leaf);
3562                         } else {
3563                                 /* if we're still in the path, make sure
3564                                  * we're dirty.  Otherwise, one of the
3565                                  * push_leaf functions must have already
3566                                  * dirtied this buffer
3567                                  */
3568                                 if (path->nodes[0] == leaf)
3569                                         btrfs_mark_buffer_dirty(leaf);
3570                                 free_extent_buffer(leaf);
3571                         }
3572                 } else {
3573                         btrfs_mark_buffer_dirty(leaf);
3574                 }
3575         }
3576         return ret;
3577 }
3578
3579 /*
3580  * search the tree again to find a leaf with lesser keys
3581  * returns 0 if it found something or 1 if there are no lesser leaves.
3582  * returns < 0 on io errors.
3583  *
3584  * This may release the path, and so you may lose any locks held at the
3585  * time you call it.
3586  */
3587 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3588 {
3589         struct btrfs_key key;
3590         struct btrfs_disk_key found_key;
3591         int ret;
3592
3593         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3594
3595         if (key.offset > 0)
3596                 key.offset--;
3597         else if (key.type > 0)
3598                 key.type--;
3599         else if (key.objectid > 0)
3600                 key.objectid--;
3601         else
3602                 return 1;
3603
3604         btrfs_release_path(root, path);
3605         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3606         if (ret < 0)
3607                 return ret;
3608         btrfs_item_key(path->nodes[0], &found_key, 0);
3609         ret = comp_keys(&found_key, &key);
3610         if (ret < 0)
3611                 return 0;
3612         return 1;
3613 }
3614
3615 /*
3616  * A helper function to walk down the tree starting at min_key, and looking
3617  * for nodes or leaves that are either in cache or have a minimum
3618  * transaction id.  This is used by the btree defrag code, and tree logging
3619  *
3620  * This does not cow, but it does stuff the starting key it finds back
3621  * into min_key, so you can call btrfs_search_slot with cow=1 on the
3622  * key and get a writable path.
3623  *
3624  * This does lock as it descends, and path->keep_locks should be set
3625  * to 1 by the caller.
3626  *
3627  * This honors path->lowest_level to prevent descent past a given level
3628  * of the tree.
3629  *
3630  * min_trans indicates the oldest transaction that you are interested
3631  * in walking through.  Any nodes or leaves older than min_trans are
3632  * skipped over (without reading them).
3633  *
3634  * returns zero if something useful was found, < 0 on error and 1 if there
3635  * was nothing in the tree that matched the search criteria.
3636  */
3637 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3638                          struct btrfs_key *max_key,
3639                          struct btrfs_path *path, int cache_only,
3640                          u64 min_trans)
3641 {
3642         struct extent_buffer *cur;
3643         struct btrfs_key found_key;
3644         int slot;
3645         int sret;
3646         u32 nritems;
3647         int level;
3648         int ret = 1;
3649
3650         WARN_ON(!path->keep_locks);
3651 again:
3652         cur = btrfs_lock_root_node(root);
3653         level = btrfs_header_level(cur);
3654         WARN_ON(path->nodes[level]);
3655         path->nodes[level] = cur;
3656         path->locks[level] = 1;
3657
3658         if (btrfs_header_generation(cur) < min_trans) {
3659                 ret = 1;
3660                 goto out;
3661         }
3662         while (1) {
3663                 nritems = btrfs_header_nritems(cur);
3664                 level = btrfs_header_level(cur);
3665                 sret = bin_search(cur, min_key, level, &slot);
3666
3667                 /* at the lowest level, we're done, setup the path and exit */
3668                 if (level == path->lowest_level) {
3669                         if (slot >= nritems)
3670                                 goto find_next_key;
3671                         ret = 0;
3672                         path->slots[level] = slot;
3673                         btrfs_item_key_to_cpu(cur, &found_key, slot);
3674                         goto out;
3675                 }
3676                 if (sret && slot > 0)
3677                         slot--;
3678                 /*
3679                  * check this node pointer against the cache_only and
3680                  * min_trans parameters.  If it isn't in cache or is too
3681                  * old, skip to the next one.
3682                  */
3683                 while (slot < nritems) {
3684                         u64 blockptr;
3685                         u64 gen;
3686                         struct extent_buffer *tmp;
3687                         struct btrfs_disk_key disk_key;
3688
3689                         blockptr = btrfs_node_blockptr(cur, slot);
3690                         gen = btrfs_node_ptr_generation(cur, slot);
3691                         if (gen < min_trans) {
3692                                 slot++;
3693                                 continue;
3694                         }
3695                         if (!cache_only)
3696                                 break;
3697
3698                         if (max_key) {
3699                                 btrfs_node_key(cur, &disk_key, slot);
3700                                 if (comp_keys(&disk_key, max_key) >= 0) {
3701                                         ret = 1;
3702                                         goto out;
3703                                 }
3704                         }
3705
3706                         tmp = btrfs_find_tree_block(root, blockptr,
3707                                             btrfs_level_size(root, level - 1));
3708
3709                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3710                                 free_extent_buffer(tmp);
3711                                 break;
3712                         }
3713                         if (tmp)
3714                                 free_extent_buffer(tmp);
3715                         slot++;
3716                 }
3717 find_next_key:
3718                 /*
3719                  * we didn't find a candidate key in this node, walk forward
3720                  * and find another one
3721                  */
3722                 if (slot >= nritems) {
3723                         path->slots[level] = slot;
3724                         sret = btrfs_find_next_key(root, path, min_key, level,
3725                                                   cache_only, min_trans);
3726                         if (sret == 0) {
3727                                 btrfs_release_path(root, path);
3728                                 goto again;
3729                         } else {
3730                                 goto out;
3731                         }
3732                 }
3733                 /* save our key for returning back */
3734                 btrfs_node_key_to_cpu(cur, &found_key, slot);
3735                 path->slots[level] = slot;
3736                 if (level == path->lowest_level) {
3737                         ret = 0;
3738                         unlock_up(path, level, 1);
3739                         goto out;
3740                 }
3741                 cur = read_node_slot(root, cur, slot);
3742
3743                 btrfs_tree_lock(cur);
3744                 path->locks[level - 1] = 1;
3745                 path->nodes[level - 1] = cur;
3746                 unlock_up(path, level, 1);
3747         }
3748 out:
3749         if (ret == 0)
3750                 memcpy(min_key, &found_key, sizeof(found_key));
3751         return ret;
3752 }
3753
3754 /*
3755  * this is similar to btrfs_next_leaf, but does not try to preserve
3756  * and fixup the path.  It looks for and returns the next key in the
3757  * tree based on the current path and the cache_only and min_trans
3758  * parameters.
3759  *
3760  * 0 is returned if another key is found, < 0 if there are any errors
3761  * and 1 is returned if there are no higher keys in the tree
3762  *
3763  * path->keep_locks should be set to 1 on the search made before
3764  * calling this function.
3765  */
3766 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3767                         struct btrfs_key *key, int lowest_level,
3768                         int cache_only, u64 min_trans)
3769 {
3770         int level = lowest_level;
3771         int slot;
3772         struct extent_buffer *c;
3773
3774         WARN_ON(!path->keep_locks);
3775         while (level < BTRFS_MAX_LEVEL) {
3776                 if (!path->nodes[level])
3777                         return 1;
3778
3779                 slot = path->slots[level] + 1;
3780                 c = path->nodes[level];
3781 next:
3782                 if (slot >= btrfs_header_nritems(c)) {
3783                         level++;
3784                         if (level == BTRFS_MAX_LEVEL)
3785                                 return 1;
3786                         continue;
3787                 }
3788                 if (level == 0)
3789                         btrfs_item_key_to_cpu(c, key, slot);
3790                 else {
3791                         u64 blockptr = btrfs_node_blockptr(c, slot);
3792                         u64 gen = btrfs_node_ptr_generation(c, slot);
3793
3794                         if (cache_only) {
3795                                 struct extent_buffer *cur;
3796                                 cur = btrfs_find_tree_block(root, blockptr,
3797                                             btrfs_level_size(root, level - 1));
3798                                 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3799                                         slot++;
3800                                         if (cur)
3801                                                 free_extent_buffer(cur);
3802                                         goto next;
3803                                 }
3804                                 free_extent_buffer(cur);
3805                         }
3806                         if (gen < min_trans) {
3807                                 slot++;
3808                                 goto next;
3809                         }
3810                         btrfs_node_key_to_cpu(c, key, slot);
3811                 }
3812                 return 0;
3813         }
3814         return 1;
3815 }
3816
3817 /*
3818  * search the tree again to find a leaf with greater keys
3819  * returns 0 if it found something or 1 if there are no greater leaves.
3820  * returns < 0 on io errors.
3821  */
3822 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3823 {
3824         int slot;
3825         int level = 1;
3826         struct extent_buffer *c;
3827         struct extent_buffer *next = NULL;
3828         struct btrfs_key key;
3829         u32 nritems;
3830         int ret;
3831
3832         nritems = btrfs_header_nritems(path->nodes[0]);
3833         if (nritems == 0)
3834                 return 1;
3835
3836         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3837
3838         btrfs_release_path(root, path);
3839         path->keep_locks = 1;
3840         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3841         path->keep_locks = 0;
3842
3843         if (ret < 0)
3844                 return ret;
3845
3846         nritems = btrfs_header_nritems(path->nodes[0]);
3847         /*
3848          * by releasing the path above we dropped all our locks.  A balance
3849          * could have added more items next to the key that used to be
3850          * at the very end of the block.  So, check again here and
3851          * advance the path if there are now more items available.
3852          */
3853         if (nritems > 0 && path->slots[0] < nritems - 1) {
3854                 path->slots[0]++;
3855                 goto done;
3856         }
3857
3858         while (level < BTRFS_MAX_LEVEL) {
3859                 if (!path->nodes[level])
3860                         return 1;
3861
3862                 slot = path->slots[level] + 1;
3863                 c = path->nodes[level];
3864                 if (slot >= btrfs_header_nritems(c)) {
3865                         level++;
3866                         if (level == BTRFS_MAX_LEVEL)
3867                                 return 1;
3868                         continue;
3869                 }
3870
3871                 if (next) {
3872                         btrfs_tree_unlock(next);
3873                         free_extent_buffer(next);
3874                 }
3875
3876                 if (level == 1 && (path->locks[1] || path->skip_locking) &&
3877                     path->reada)
3878                         reada_for_search(root, path, level, slot, 0);
3879
3880                 next = read_node_slot(root, c, slot);
3881                 if (!path->skip_locking) {
3882                         WARN_ON(!btrfs_tree_locked(c));
3883                         btrfs_tree_lock(next);
3884                 }
3885                 break;
3886         }
3887         path->slots[level] = slot;
3888         while (1) {
3889                 level--;
3890                 c = path->nodes[level];
3891                 if (path->locks[level])
3892                         btrfs_tree_unlock(c);
3893                 free_extent_buffer(c);
3894                 path->nodes[level] = next;
3895                 path->slots[level] = 0;
3896                 if (!path->skip_locking)
3897                         path->locks[level] = 1;
3898                 if (!level)
3899                         break;
3900                 if (level == 1 && path->locks[1] && path->reada)
3901                         reada_for_search(root, path, level, slot, 0);
3902                 next = read_node_slot(root, next, 0);
3903                 if (!path->skip_locking) {
3904                         WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3905                         btrfs_tree_lock(next);
3906                 }
3907         }
3908 done:
3909         unlock_up(path, 0, 1);
3910         return 0;
3911 }
3912
3913 /*
3914  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3915  * searching until it gets past min_objectid or finds an item of 'type'
3916  *
3917  * returns 0 if something is found, 1 if nothing was found and < 0 on error
3918  */
3919 int btrfs_previous_item(struct btrfs_root *root,
3920                         struct btrfs_path *path, u64 min_objectid,
3921                         int type)
3922 {
3923         struct btrfs_key found_key;
3924         struct extent_buffer *leaf;
3925         u32 nritems;
3926         int ret;
3927
3928         while (1) {
3929                 if (path->slots[0] == 0) {
3930                         ret = btrfs_prev_leaf(root, path);
3931                         if (ret != 0)
3932                                 return ret;
3933                 } else {
3934                         path->slots[0]--;
3935                 }
3936                 leaf = path->nodes[0];
3937                 nritems = btrfs_header_nritems(leaf);
3938                 if (nritems == 0)
3939                         return 1;
3940                 if (path->slots[0] == nritems)
3941                         path->slots[0]--;
3942
3943                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3944                 if (found_key.type == type)
3945                         return 0;
3946                 if (found_key.objectid < min_objectid)
3947                         break;
3948                 if (found_key.objectid == min_objectid &&
3949                     found_key.type < type)
3950                         break;
3951         }
3952         return 1;
3953 }