Btrfs: switch the btrfs tree locks to reader/writer
[pandora-kernel.git] / 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 <linux/slab.h>
21 #include "ctree.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "print-tree.h"
25 #include "locking.h"
26
27 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
28                       *root, struct btrfs_path *path, int level);
29 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
30                       *root, struct btrfs_key *ins_key,
31                       struct btrfs_path *path, int data_size, int extend);
32 static int push_node_left(struct btrfs_trans_handle *trans,
33                           struct btrfs_root *root, struct extent_buffer *dst,
34                           struct extent_buffer *src, int empty);
35 static int balance_node_right(struct btrfs_trans_handle *trans,
36                               struct btrfs_root *root,
37                               struct extent_buffer *dst_buf,
38                               struct extent_buffer *src_buf);
39 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
40                    struct btrfs_path *path, int level, int slot);
41
42 struct btrfs_path *btrfs_alloc_path(void)
43 {
44         struct btrfs_path *path;
45         path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
46         return path;
47 }
48
49 /*
50  * set all locked nodes in the path to blocking locks.  This should
51  * be done before scheduling
52  */
53 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
54 {
55         int i;
56         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
57                 if (!p->nodes[i] || !p->locks[i])
58                         continue;
59                 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
60                 if (p->locks[i] == BTRFS_READ_LOCK)
61                         p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
62                 else if (p->locks[i] == BTRFS_WRITE_LOCK)
63                         p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
64         }
65 }
66
67 /*
68  * reset all the locked nodes in the patch to spinning locks.
69  *
70  * held is used to keep lockdep happy, when lockdep is enabled
71  * we set held to a blocking lock before we go around and
72  * retake all the spinlocks in the path.  You can safely use NULL
73  * for held
74  */
75 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
76                                         struct extent_buffer *held, int held_rw)
77 {
78         int i;
79
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81         /* lockdep really cares that we take all of these spinlocks
82          * in the right order.  If any of the locks in the path are not
83          * currently blocking, it is going to complain.  So, make really
84          * really sure by forcing the path to blocking before we clear
85          * the path blocking.
86          */
87         if (held) {
88                 btrfs_set_lock_blocking_rw(held, held_rw);
89                 if (held_rw == BTRFS_WRITE_LOCK)
90                         held_rw = BTRFS_WRITE_LOCK_BLOCKING;
91                 else if (held_rw == BTRFS_READ_LOCK)
92                         held_rw = BTRFS_READ_LOCK_BLOCKING;
93         }
94         btrfs_set_path_blocking(p);
95 #endif
96
97         for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
98                 if (p->nodes[i] && p->locks[i]) {
99                         btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
100                         if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
101                                 p->locks[i] = BTRFS_WRITE_LOCK;
102                         else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
103                                 p->locks[i] = BTRFS_READ_LOCK;
104                 }
105         }
106
107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
108         if (held)
109                 btrfs_clear_lock_blocking_rw(held, held_rw);
110 #endif
111 }
112
113 /* this also releases the path */
114 void btrfs_free_path(struct btrfs_path *p)
115 {
116         if (!p)
117                 return;
118         btrfs_release_path(p);
119         kmem_cache_free(btrfs_path_cachep, p);
120 }
121
122 /*
123  * path release drops references on the extent buffers in the path
124  * and it drops any locks held by this path
125  *
126  * It is safe to call this on paths that no locks or extent buffers held.
127  */
128 noinline void btrfs_release_path(struct btrfs_path *p)
129 {
130         int i;
131
132         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
133                 p->slots[i] = 0;
134                 if (!p->nodes[i])
135                         continue;
136                 if (p->locks[i]) {
137                         btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
138                         p->locks[i] = 0;
139                 }
140                 free_extent_buffer(p->nodes[i]);
141                 p->nodes[i] = NULL;
142         }
143 }
144
145 /*
146  * safely gets a reference on the root node of a tree.  A lock
147  * is not taken, so a concurrent writer may put a different node
148  * at the root of the tree.  See btrfs_lock_root_node for the
149  * looping required.
150  *
151  * The extent buffer returned by this has a reference taken, so
152  * it won't disappear.  It may stop being the root of the tree
153  * at any time because there are no locks held.
154  */
155 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
156 {
157         struct extent_buffer *eb;
158
159         rcu_read_lock();
160         eb = rcu_dereference(root->node);
161         extent_buffer_get(eb);
162         rcu_read_unlock();
163         return eb;
164 }
165
166 /* loop around taking references on and locking the root node of the
167  * tree until you end up with a lock on the root.  A locked buffer
168  * is returned, with a reference held.
169  */
170 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
171 {
172         struct extent_buffer *eb;
173
174         while (1) {
175                 eb = btrfs_root_node(root);
176                 btrfs_tree_lock(eb);
177                 if (eb == root->node)
178                         break;
179                 btrfs_tree_unlock(eb);
180                 free_extent_buffer(eb);
181         }
182         return eb;
183 }
184
185 /* loop around taking references on and locking the root node of the
186  * tree until you end up with a lock on the root.  A locked buffer
187  * is returned, with a reference held.
188  */
189 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
190 {
191         struct extent_buffer *eb;
192
193         while (1) {
194                 eb = btrfs_root_node(root);
195                 btrfs_tree_read_lock(eb);
196                 if (eb == root->node)
197                         break;
198                 btrfs_tree_read_unlock(eb);
199                 free_extent_buffer(eb);
200         }
201         return eb;
202 }
203
204 /* cowonly root (everything not a reference counted cow subvolume), just get
205  * put onto a simple dirty list.  transaction.c walks this to make sure they
206  * get properly updated on disk.
207  */
208 static void add_root_to_dirty_list(struct btrfs_root *root)
209 {
210         if (root->track_dirty && list_empty(&root->dirty_list)) {
211                 list_add(&root->dirty_list,
212                          &root->fs_info->dirty_cowonly_roots);
213         }
214 }
215
216 /*
217  * used by snapshot creation to make a copy of a root for a tree with
218  * a given objectid.  The buffer with the new root node is returned in
219  * cow_ret, and this func returns zero on success or a negative error code.
220  */
221 int btrfs_copy_root(struct btrfs_trans_handle *trans,
222                       struct btrfs_root *root,
223                       struct extent_buffer *buf,
224                       struct extent_buffer **cow_ret, u64 new_root_objectid)
225 {
226         struct extent_buffer *cow;
227         int ret = 0;
228         int level;
229         struct btrfs_disk_key disk_key;
230
231         WARN_ON(root->ref_cows && trans->transid !=
232                 root->fs_info->running_transaction->transid);
233         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
234
235         level = btrfs_header_level(buf);
236         if (level == 0)
237                 btrfs_item_key(buf, &disk_key, 0);
238         else
239                 btrfs_node_key(buf, &disk_key, 0);
240
241         cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
242                                      new_root_objectid, &disk_key, level,
243                                      buf->start, 0);
244         if (IS_ERR(cow))
245                 return PTR_ERR(cow);
246
247         copy_extent_buffer(cow, buf, 0, 0, cow->len);
248         btrfs_set_header_bytenr(cow, cow->start);
249         btrfs_set_header_generation(cow, trans->transid);
250         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
251         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
252                                      BTRFS_HEADER_FLAG_RELOC);
253         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
254                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
255         else
256                 btrfs_set_header_owner(cow, new_root_objectid);
257
258         write_extent_buffer(cow, root->fs_info->fsid,
259                             (unsigned long)btrfs_header_fsid(cow),
260                             BTRFS_FSID_SIZE);
261
262         WARN_ON(btrfs_header_generation(buf) > trans->transid);
263         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
264                 ret = btrfs_inc_ref(trans, root, cow, 1);
265         else
266                 ret = btrfs_inc_ref(trans, root, cow, 0);
267
268         if (ret)
269                 return ret;
270
271         btrfs_mark_buffer_dirty(cow);
272         *cow_ret = cow;
273         return 0;
274 }
275
276 /*
277  * check if the tree block can be shared by multiple trees
278  */
279 int btrfs_block_can_be_shared(struct btrfs_root *root,
280                               struct extent_buffer *buf)
281 {
282         /*
283          * Tree blocks not in refernece counted trees and tree roots
284          * are never shared. If a block was allocated after the last
285          * snapshot and the block was not allocated by tree relocation,
286          * we know the block is not shared.
287          */
288         if (root->ref_cows &&
289             buf != root->node && buf != root->commit_root &&
290             (btrfs_header_generation(buf) <=
291              btrfs_root_last_snapshot(&root->root_item) ||
292              btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
293                 return 1;
294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
295         if (root->ref_cows &&
296             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
297                 return 1;
298 #endif
299         return 0;
300 }
301
302 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
303                                        struct btrfs_root *root,
304                                        struct extent_buffer *buf,
305                                        struct extent_buffer *cow,
306                                        int *last_ref)
307 {
308         u64 refs;
309         u64 owner;
310         u64 flags;
311         u64 new_flags = 0;
312         int ret;
313
314         /*
315          * Backrefs update rules:
316          *
317          * Always use full backrefs for extent pointers in tree block
318          * allocated by tree relocation.
319          *
320          * If a shared tree block is no longer referenced by its owner
321          * tree (btrfs_header_owner(buf) == root->root_key.objectid),
322          * use full backrefs for extent pointers in tree block.
323          *
324          * If a tree block is been relocating
325          * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
326          * use full backrefs for extent pointers in tree block.
327          * The reason for this is some operations (such as drop tree)
328          * are only allowed for blocks use full backrefs.
329          */
330
331         if (btrfs_block_can_be_shared(root, buf)) {
332                 ret = btrfs_lookup_extent_info(trans, root, buf->start,
333                                                buf->len, &refs, &flags);
334                 BUG_ON(ret);
335                 BUG_ON(refs == 0);
336         } else {
337                 refs = 1;
338                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
339                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
340                         flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
341                 else
342                         flags = 0;
343         }
344
345         owner = btrfs_header_owner(buf);
346         BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
347                !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
348
349         if (refs > 1) {
350                 if ((owner == root->root_key.objectid ||
351                      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
352                     !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
353                         ret = btrfs_inc_ref(trans, root, buf, 1);
354                         BUG_ON(ret);
355
356                         if (root->root_key.objectid ==
357                             BTRFS_TREE_RELOC_OBJECTID) {
358                                 ret = btrfs_dec_ref(trans, root, buf, 0);
359                                 BUG_ON(ret);
360                                 ret = btrfs_inc_ref(trans, root, cow, 1);
361                                 BUG_ON(ret);
362                         }
363                         new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
364                 } else {
365
366                         if (root->root_key.objectid ==
367                             BTRFS_TREE_RELOC_OBJECTID)
368                                 ret = btrfs_inc_ref(trans, root, cow, 1);
369                         else
370                                 ret = btrfs_inc_ref(trans, root, cow, 0);
371                         BUG_ON(ret);
372                 }
373                 if (new_flags != 0) {
374                         ret = btrfs_set_disk_extent_flags(trans, root,
375                                                           buf->start,
376                                                           buf->len,
377                                                           new_flags, 0);
378                         BUG_ON(ret);
379                 }
380         } else {
381                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
382                         if (root->root_key.objectid ==
383                             BTRFS_TREE_RELOC_OBJECTID)
384                                 ret = btrfs_inc_ref(trans, root, cow, 1);
385                         else
386                                 ret = btrfs_inc_ref(trans, root, cow, 0);
387                         BUG_ON(ret);
388                         ret = btrfs_dec_ref(trans, root, buf, 1);
389                         BUG_ON(ret);
390                 }
391                 clean_tree_block(trans, root, buf);
392                 *last_ref = 1;
393         }
394         return 0;
395 }
396
397 /*
398  * does the dirty work in cow of a single block.  The parent block (if
399  * supplied) is updated to point to the new cow copy.  The new buffer is marked
400  * dirty and returned locked.  If you modify the block it needs to be marked
401  * dirty again.
402  *
403  * search_start -- an allocation hint for the new block
404  *
405  * empty_size -- a hint that you plan on doing more cow.  This is the size in
406  * bytes the allocator should try to find free next to the block it returns.
407  * This is just a hint and may be ignored by the allocator.
408  */
409 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
410                              struct btrfs_root *root,
411                              struct extent_buffer *buf,
412                              struct extent_buffer *parent, int parent_slot,
413                              struct extent_buffer **cow_ret,
414                              u64 search_start, u64 empty_size)
415 {
416         struct btrfs_disk_key disk_key;
417         struct extent_buffer *cow;
418         int level;
419         int last_ref = 0;
420         int unlock_orig = 0;
421         u64 parent_start;
422
423         if (*cow_ret == buf)
424                 unlock_orig = 1;
425
426         btrfs_assert_tree_locked(buf);
427
428         WARN_ON(root->ref_cows && trans->transid !=
429                 root->fs_info->running_transaction->transid);
430         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
431
432         level = btrfs_header_level(buf);
433
434         if (level == 0)
435                 btrfs_item_key(buf, &disk_key, 0);
436         else
437                 btrfs_node_key(buf, &disk_key, 0);
438
439         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
440                 if (parent)
441                         parent_start = parent->start;
442                 else
443                         parent_start = 0;
444         } else
445                 parent_start = 0;
446
447         cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
448                                      root->root_key.objectid, &disk_key,
449                                      level, search_start, empty_size);
450         if (IS_ERR(cow))
451                 return PTR_ERR(cow);
452
453         /* cow is set to blocking by btrfs_init_new_buffer */
454
455         copy_extent_buffer(cow, buf, 0, 0, cow->len);
456         btrfs_set_header_bytenr(cow, cow->start);
457         btrfs_set_header_generation(cow, trans->transid);
458         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
459         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
460                                      BTRFS_HEADER_FLAG_RELOC);
461         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
462                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
463         else
464                 btrfs_set_header_owner(cow, root->root_key.objectid);
465
466         write_extent_buffer(cow, root->fs_info->fsid,
467                             (unsigned long)btrfs_header_fsid(cow),
468                             BTRFS_FSID_SIZE);
469
470         update_ref_for_cow(trans, root, buf, cow, &last_ref);
471
472         if (root->ref_cows)
473                 btrfs_reloc_cow_block(trans, root, buf, cow);
474
475         if (buf == root->node) {
476                 WARN_ON(parent && parent != buf);
477                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
478                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
479                         parent_start = buf->start;
480                 else
481                         parent_start = 0;
482
483                 extent_buffer_get(cow);
484                 rcu_assign_pointer(root->node, cow);
485
486                 btrfs_free_tree_block(trans, root, buf, parent_start,
487                                       last_ref);
488                 free_extent_buffer(buf);
489                 add_root_to_dirty_list(root);
490         } else {
491                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
492                         parent_start = parent->start;
493                 else
494                         parent_start = 0;
495
496                 WARN_ON(trans->transid != btrfs_header_generation(parent));
497                 btrfs_set_node_blockptr(parent, parent_slot,
498                                         cow->start);
499                 btrfs_set_node_ptr_generation(parent, parent_slot,
500                                               trans->transid);
501                 btrfs_mark_buffer_dirty(parent);
502                 btrfs_free_tree_block(trans, root, buf, parent_start,
503                                       last_ref);
504         }
505         if (unlock_orig)
506                 btrfs_tree_unlock(buf);
507         free_extent_buffer(buf);
508         btrfs_mark_buffer_dirty(cow);
509         *cow_ret = cow;
510         return 0;
511 }
512
513 static inline int should_cow_block(struct btrfs_trans_handle *trans,
514                                    struct btrfs_root *root,
515                                    struct extent_buffer *buf)
516 {
517         if (btrfs_header_generation(buf) == trans->transid &&
518             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
519             !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
520               btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
521                 return 0;
522         return 1;
523 }
524
525 /*
526  * cows a single block, see __btrfs_cow_block for the real work.
527  * This version of it has extra checks so that a block isn't cow'd more than
528  * once per transaction, as long as it hasn't been written yet
529  */
530 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
531                     struct btrfs_root *root, struct extent_buffer *buf,
532                     struct extent_buffer *parent, int parent_slot,
533                     struct extent_buffer **cow_ret)
534 {
535         u64 search_start;
536         int ret;
537
538         if (trans->transaction != root->fs_info->running_transaction) {
539                 printk(KERN_CRIT "trans %llu running %llu\n",
540                        (unsigned long long)trans->transid,
541                        (unsigned long long)
542                        root->fs_info->running_transaction->transid);
543                 WARN_ON(1);
544         }
545         if (trans->transid != root->fs_info->generation) {
546                 printk(KERN_CRIT "trans %llu running %llu\n",
547                        (unsigned long long)trans->transid,
548                        (unsigned long long)root->fs_info->generation);
549                 WARN_ON(1);
550         }
551
552         if (!should_cow_block(trans, root, buf)) {
553                 *cow_ret = buf;
554                 return 0;
555         }
556
557         search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
558
559         if (parent)
560                 btrfs_set_lock_blocking(parent);
561         btrfs_set_lock_blocking(buf);
562
563         ret = __btrfs_cow_block(trans, root, buf, parent,
564                                  parent_slot, cow_ret, search_start, 0);
565
566         trace_btrfs_cow_block(root, buf, *cow_ret);
567
568         return ret;
569 }
570
571 /*
572  * helper function for defrag to decide if two blocks pointed to by a
573  * node are actually close by
574  */
575 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
576 {
577         if (blocknr < other && other - (blocknr + blocksize) < 32768)
578                 return 1;
579         if (blocknr > other && blocknr - (other + blocksize) < 32768)
580                 return 1;
581         return 0;
582 }
583
584 /*
585  * compare two keys in a memcmp fashion
586  */
587 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
588 {
589         struct btrfs_key k1;
590
591         btrfs_disk_key_to_cpu(&k1, disk);
592
593         return btrfs_comp_cpu_keys(&k1, k2);
594 }
595
596 /*
597  * same as comp_keys only with two btrfs_key's
598  */
599 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
600 {
601         if (k1->objectid > k2->objectid)
602                 return 1;
603         if (k1->objectid < k2->objectid)
604                 return -1;
605         if (k1->type > k2->type)
606                 return 1;
607         if (k1->type < k2->type)
608                 return -1;
609         if (k1->offset > k2->offset)
610                 return 1;
611         if (k1->offset < k2->offset)
612                 return -1;
613         return 0;
614 }
615
616 /*
617  * this is used by the defrag code to go through all the
618  * leaves pointed to by a node and reallocate them so that
619  * disk order is close to key order
620  */
621 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
622                        struct btrfs_root *root, struct extent_buffer *parent,
623                        int start_slot, int cache_only, u64 *last_ret,
624                        struct btrfs_key *progress)
625 {
626         struct extent_buffer *cur;
627         u64 blocknr;
628         u64 gen;
629         u64 search_start = *last_ret;
630         u64 last_block = 0;
631         u64 other;
632         u32 parent_nritems;
633         int end_slot;
634         int i;
635         int err = 0;
636         int parent_level;
637         int uptodate;
638         u32 blocksize;
639         int progress_passed = 0;
640         struct btrfs_disk_key disk_key;
641
642         parent_level = btrfs_header_level(parent);
643         if (cache_only && parent_level != 1)
644                 return 0;
645
646         if (trans->transaction != root->fs_info->running_transaction)
647                 WARN_ON(1);
648         if (trans->transid != root->fs_info->generation)
649                 WARN_ON(1);
650
651         parent_nritems = btrfs_header_nritems(parent);
652         blocksize = btrfs_level_size(root, parent_level - 1);
653         end_slot = parent_nritems;
654
655         if (parent_nritems == 1)
656                 return 0;
657
658         btrfs_set_lock_blocking(parent);
659
660         for (i = start_slot; i < end_slot; i++) {
661                 int close = 1;
662
663                 btrfs_node_key(parent, &disk_key, i);
664                 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
665                         continue;
666
667                 progress_passed = 1;
668                 blocknr = btrfs_node_blockptr(parent, i);
669                 gen = btrfs_node_ptr_generation(parent, i);
670                 if (last_block == 0)
671                         last_block = blocknr;
672
673                 if (i > 0) {
674                         other = btrfs_node_blockptr(parent, i - 1);
675                         close = close_blocks(blocknr, other, blocksize);
676                 }
677                 if (!close && i < end_slot - 2) {
678                         other = btrfs_node_blockptr(parent, i + 1);
679                         close = close_blocks(blocknr, other, blocksize);
680                 }
681                 if (close) {
682                         last_block = blocknr;
683                         continue;
684                 }
685
686                 cur = btrfs_find_tree_block(root, blocknr, blocksize);
687                 if (cur)
688                         uptodate = btrfs_buffer_uptodate(cur, gen);
689                 else
690                         uptodate = 0;
691                 if (!cur || !uptodate) {
692                         if (cache_only) {
693                                 free_extent_buffer(cur);
694                                 continue;
695                         }
696                         if (!cur) {
697                                 cur = read_tree_block(root, blocknr,
698                                                          blocksize, gen);
699                                 if (!cur)
700                                         return -EIO;
701                         } else if (!uptodate) {
702                                 btrfs_read_buffer(cur, gen);
703                         }
704                 }
705                 if (search_start == 0)
706                         search_start = last_block;
707
708                 btrfs_tree_lock(cur);
709                 btrfs_set_lock_blocking(cur);
710                 err = __btrfs_cow_block(trans, root, cur, parent, i,
711                                         &cur, search_start,
712                                         min(16 * blocksize,
713                                             (end_slot - i) * blocksize));
714                 if (err) {
715                         btrfs_tree_unlock(cur);
716                         free_extent_buffer(cur);
717                         break;
718                 }
719                 search_start = cur->start;
720                 last_block = cur->start;
721                 *last_ret = search_start;
722                 btrfs_tree_unlock(cur);
723                 free_extent_buffer(cur);
724         }
725         return err;
726 }
727
728 /*
729  * The leaf data grows from end-to-front in the node.
730  * this returns the address of the start of the last item,
731  * which is the stop of the leaf data stack
732  */
733 static inline unsigned int leaf_data_end(struct btrfs_root *root,
734                                          struct extent_buffer *leaf)
735 {
736         u32 nr = btrfs_header_nritems(leaf);
737         if (nr == 0)
738                 return BTRFS_LEAF_DATA_SIZE(root);
739         return btrfs_item_offset_nr(leaf, nr - 1);
740 }
741
742
743 /*
744  * search for key in the extent_buffer.  The items start at offset p,
745  * and they are item_size apart.  There are 'max' items in p.
746  *
747  * the slot in the array is returned via slot, and it points to
748  * the place where you would insert key if it is not found in
749  * the array.
750  *
751  * slot may point to max if the key is bigger than all of the keys
752  */
753 static noinline int generic_bin_search(struct extent_buffer *eb,
754                                        unsigned long p,
755                                        int item_size, struct btrfs_key *key,
756                                        int max, int *slot)
757 {
758         int low = 0;
759         int high = max;
760         int mid;
761         int ret;
762         struct btrfs_disk_key *tmp = NULL;
763         struct btrfs_disk_key unaligned;
764         unsigned long offset;
765         char *kaddr = NULL;
766         unsigned long map_start = 0;
767         unsigned long map_len = 0;
768         int err;
769
770         while (low < high) {
771                 mid = (low + high) / 2;
772                 offset = p + mid * item_size;
773
774                 if (!kaddr || offset < map_start ||
775                     (offset + sizeof(struct btrfs_disk_key)) >
776                     map_start + map_len) {
777
778                         err = map_private_extent_buffer(eb, offset,
779                                                 sizeof(struct btrfs_disk_key),
780                                                 &kaddr, &map_start, &map_len);
781
782                         if (!err) {
783                                 tmp = (struct btrfs_disk_key *)(kaddr + offset -
784                                                         map_start);
785                         } else {
786                                 read_extent_buffer(eb, &unaligned,
787                                                    offset, sizeof(unaligned));
788                                 tmp = &unaligned;
789                         }
790
791                 } else {
792                         tmp = (struct btrfs_disk_key *)(kaddr + offset -
793                                                         map_start);
794                 }
795                 ret = comp_keys(tmp, key);
796
797                 if (ret < 0)
798                         low = mid + 1;
799                 else if (ret > 0)
800                         high = mid;
801                 else {
802                         *slot = mid;
803                         return 0;
804                 }
805         }
806         *slot = low;
807         return 1;
808 }
809
810 /*
811  * simple bin_search frontend that does the right thing for
812  * leaves vs nodes
813  */
814 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
815                       int level, int *slot)
816 {
817         if (level == 0) {
818                 return generic_bin_search(eb,
819                                           offsetof(struct btrfs_leaf, items),
820                                           sizeof(struct btrfs_item),
821                                           key, btrfs_header_nritems(eb),
822                                           slot);
823         } else {
824                 return generic_bin_search(eb,
825                                           offsetof(struct btrfs_node, ptrs),
826                                           sizeof(struct btrfs_key_ptr),
827                                           key, btrfs_header_nritems(eb),
828                                           slot);
829         }
830         return -1;
831 }
832
833 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
834                      int level, int *slot)
835 {
836         return bin_search(eb, key, level, slot);
837 }
838
839 static void root_add_used(struct btrfs_root *root, u32 size)
840 {
841         spin_lock(&root->accounting_lock);
842         btrfs_set_root_used(&root->root_item,
843                             btrfs_root_used(&root->root_item) + size);
844         spin_unlock(&root->accounting_lock);
845 }
846
847 static void root_sub_used(struct btrfs_root *root, u32 size)
848 {
849         spin_lock(&root->accounting_lock);
850         btrfs_set_root_used(&root->root_item,
851                             btrfs_root_used(&root->root_item) - size);
852         spin_unlock(&root->accounting_lock);
853 }
854
855 /* given a node and slot number, this reads the blocks it points to.  The
856  * extent buffer is returned with a reference taken (but unlocked).
857  * NULL is returned on error.
858  */
859 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
860                                    struct extent_buffer *parent, int slot)
861 {
862         int level = btrfs_header_level(parent);
863         if (slot < 0)
864                 return NULL;
865         if (slot >= btrfs_header_nritems(parent))
866                 return NULL;
867
868         BUG_ON(level == 0);
869
870         return read_tree_block(root, btrfs_node_blockptr(parent, slot),
871                        btrfs_level_size(root, level - 1),
872                        btrfs_node_ptr_generation(parent, slot));
873 }
874
875 /*
876  * node level balancing, used to make sure nodes are in proper order for
877  * item deletion.  We balance from the top down, so we have to make sure
878  * that a deletion won't leave an node completely empty later on.
879  */
880 static noinline int balance_level(struct btrfs_trans_handle *trans,
881                          struct btrfs_root *root,
882                          struct btrfs_path *path, int level)
883 {
884         struct extent_buffer *right = NULL;
885         struct extent_buffer *mid;
886         struct extent_buffer *left = NULL;
887         struct extent_buffer *parent = NULL;
888         int ret = 0;
889         int wret;
890         int pslot;
891         int orig_slot = path->slots[level];
892         u64 orig_ptr;
893
894         if (level == 0)
895                 return 0;
896
897         mid = path->nodes[level];
898
899         WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
900                 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
901         WARN_ON(btrfs_header_generation(mid) != trans->transid);
902
903         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
904
905         if (level < BTRFS_MAX_LEVEL - 1)
906                 parent = path->nodes[level + 1];
907         pslot = path->slots[level + 1];
908
909         /*
910          * deal with the case where there is only one pointer in the root
911          * by promoting the node below to a root
912          */
913         if (!parent) {
914                 struct extent_buffer *child;
915
916                 if (btrfs_header_nritems(mid) != 1)
917                         return 0;
918
919                 /* promote the child to a root */
920                 child = read_node_slot(root, mid, 0);
921                 BUG_ON(!child);
922                 btrfs_tree_lock(child);
923                 btrfs_set_lock_blocking(child);
924                 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
925                 if (ret) {
926                         btrfs_tree_unlock(child);
927                         free_extent_buffer(child);
928                         goto enospc;
929                 }
930
931                 rcu_assign_pointer(root->node, child);
932
933                 add_root_to_dirty_list(root);
934                 btrfs_tree_unlock(child);
935
936                 path->locks[level] = 0;
937                 path->nodes[level] = NULL;
938                 clean_tree_block(trans, root, mid);
939                 btrfs_tree_unlock(mid);
940                 /* once for the path */
941                 free_extent_buffer(mid);
942
943                 root_sub_used(root, mid->len);
944                 btrfs_free_tree_block(trans, root, mid, 0, 1);
945                 /* once for the root ptr */
946                 free_extent_buffer(mid);
947                 return 0;
948         }
949         if (btrfs_header_nritems(mid) >
950             BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
951                 return 0;
952
953         btrfs_header_nritems(mid);
954
955         left = read_node_slot(root, parent, pslot - 1);
956         if (left) {
957                 btrfs_tree_lock(left);
958                 btrfs_set_lock_blocking(left);
959                 wret = btrfs_cow_block(trans, root, left,
960                                        parent, pslot - 1, &left);
961                 if (wret) {
962                         ret = wret;
963                         goto enospc;
964                 }
965         }
966         right = read_node_slot(root, parent, pslot + 1);
967         if (right) {
968                 btrfs_tree_lock(right);
969                 btrfs_set_lock_blocking(right);
970                 wret = btrfs_cow_block(trans, root, right,
971                                        parent, pslot + 1, &right);
972                 if (wret) {
973                         ret = wret;
974                         goto enospc;
975                 }
976         }
977
978         /* first, try to make some room in the middle buffer */
979         if (left) {
980                 orig_slot += btrfs_header_nritems(left);
981                 wret = push_node_left(trans, root, left, mid, 1);
982                 if (wret < 0)
983                         ret = wret;
984                 btrfs_header_nritems(mid);
985         }
986
987         /*
988          * then try to empty the right most buffer into the middle
989          */
990         if (right) {
991                 wret = push_node_left(trans, root, mid, right, 1);
992                 if (wret < 0 && wret != -ENOSPC)
993                         ret = wret;
994                 if (btrfs_header_nritems(right) == 0) {
995                         clean_tree_block(trans, root, right);
996                         btrfs_tree_unlock(right);
997                         wret = del_ptr(trans, root, path, level + 1, pslot +
998                                        1);
999                         if (wret)
1000                                 ret = wret;
1001                         root_sub_used(root, right->len);
1002                         btrfs_free_tree_block(trans, root, right, 0, 1);
1003                         free_extent_buffer(right);
1004                         right = NULL;
1005                 } else {
1006                         struct btrfs_disk_key right_key;
1007                         btrfs_node_key(right, &right_key, 0);
1008                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1009                         btrfs_mark_buffer_dirty(parent);
1010                 }
1011         }
1012         if (btrfs_header_nritems(mid) == 1) {
1013                 /*
1014                  * we're not allowed to leave a node with one item in the
1015                  * tree during a delete.  A deletion from lower in the tree
1016                  * could try to delete the only pointer in this node.
1017                  * So, pull some keys from the left.
1018                  * There has to be a left pointer at this point because
1019                  * otherwise we would have pulled some pointers from the
1020                  * right
1021                  */
1022                 BUG_ON(!left);
1023                 wret = balance_node_right(trans, root, mid, left);
1024                 if (wret < 0) {
1025                         ret = wret;
1026                         goto enospc;
1027                 }
1028                 if (wret == 1) {
1029                         wret = push_node_left(trans, root, left, mid, 1);
1030                         if (wret < 0)
1031                                 ret = wret;
1032                 }
1033                 BUG_ON(wret == 1);
1034         }
1035         if (btrfs_header_nritems(mid) == 0) {
1036                 clean_tree_block(trans, root, mid);
1037                 btrfs_tree_unlock(mid);
1038                 wret = del_ptr(trans, root, path, level + 1, pslot);
1039                 if (wret)
1040                         ret = wret;
1041                 root_sub_used(root, mid->len);
1042                 btrfs_free_tree_block(trans, root, mid, 0, 1);
1043                 free_extent_buffer(mid);
1044                 mid = NULL;
1045         } else {
1046                 /* update the parent key to reflect our changes */
1047                 struct btrfs_disk_key mid_key;
1048                 btrfs_node_key(mid, &mid_key, 0);
1049                 btrfs_set_node_key(parent, &mid_key, pslot);
1050                 btrfs_mark_buffer_dirty(parent);
1051         }
1052
1053         /* update the path */
1054         if (left) {
1055                 if (btrfs_header_nritems(left) > orig_slot) {
1056                         extent_buffer_get(left);
1057                         /* left was locked after cow */
1058                         path->nodes[level] = left;
1059                         path->slots[level + 1] -= 1;
1060                         path->slots[level] = orig_slot;
1061                         if (mid) {
1062                                 btrfs_tree_unlock(mid);
1063                                 free_extent_buffer(mid);
1064                         }
1065                 } else {
1066                         orig_slot -= btrfs_header_nritems(left);
1067                         path->slots[level] = orig_slot;
1068                 }
1069         }
1070         /* double check we haven't messed things up */
1071         if (orig_ptr !=
1072             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1073                 BUG();
1074 enospc:
1075         if (right) {
1076                 btrfs_tree_unlock(right);
1077                 free_extent_buffer(right);
1078         }
1079         if (left) {
1080                 if (path->nodes[level] != left)
1081                         btrfs_tree_unlock(left);
1082                 free_extent_buffer(left);
1083         }
1084         return ret;
1085 }
1086
1087 /* Node balancing for insertion.  Here we only split or push nodes around
1088  * when they are completely full.  This is also done top down, so we
1089  * have to be pessimistic.
1090  */
1091 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1092                                           struct btrfs_root *root,
1093                                           struct btrfs_path *path, int level)
1094 {
1095         struct extent_buffer *right = NULL;
1096         struct extent_buffer *mid;
1097         struct extent_buffer *left = NULL;
1098         struct extent_buffer *parent = NULL;
1099         int ret = 0;
1100         int wret;
1101         int pslot;
1102         int orig_slot = path->slots[level];
1103
1104         if (level == 0)
1105                 return 1;
1106
1107         mid = path->nodes[level];
1108         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1109
1110         if (level < BTRFS_MAX_LEVEL - 1)
1111                 parent = path->nodes[level + 1];
1112         pslot = path->slots[level + 1];
1113
1114         if (!parent)
1115                 return 1;
1116
1117         left = read_node_slot(root, parent, pslot - 1);
1118
1119         /* first, try to make some room in the middle buffer */
1120         if (left) {
1121                 u32 left_nr;
1122
1123                 btrfs_tree_lock(left);
1124                 btrfs_set_lock_blocking(left);
1125
1126                 left_nr = btrfs_header_nritems(left);
1127                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1128                         wret = 1;
1129                 } else {
1130                         ret = btrfs_cow_block(trans, root, left, parent,
1131                                               pslot - 1, &left);
1132                         if (ret)
1133                                 wret = 1;
1134                         else {
1135                                 wret = push_node_left(trans, root,
1136                                                       left, mid, 0);
1137                         }
1138                 }
1139                 if (wret < 0)
1140                         ret = wret;
1141                 if (wret == 0) {
1142                         struct btrfs_disk_key disk_key;
1143                         orig_slot += left_nr;
1144                         btrfs_node_key(mid, &disk_key, 0);
1145                         btrfs_set_node_key(parent, &disk_key, pslot);
1146                         btrfs_mark_buffer_dirty(parent);
1147                         if (btrfs_header_nritems(left) > orig_slot) {
1148                                 path->nodes[level] = left;
1149                                 path->slots[level + 1] -= 1;
1150                                 path->slots[level] = orig_slot;
1151                                 btrfs_tree_unlock(mid);
1152                                 free_extent_buffer(mid);
1153                         } else {
1154                                 orig_slot -=
1155                                         btrfs_header_nritems(left);
1156                                 path->slots[level] = orig_slot;
1157                                 btrfs_tree_unlock(left);
1158                                 free_extent_buffer(left);
1159                         }
1160                         return 0;
1161                 }
1162                 btrfs_tree_unlock(left);
1163                 free_extent_buffer(left);
1164         }
1165         right = read_node_slot(root, parent, pslot + 1);
1166
1167         /*
1168          * then try to empty the right most buffer into the middle
1169          */
1170         if (right) {
1171                 u32 right_nr;
1172
1173                 btrfs_tree_lock(right);
1174                 btrfs_set_lock_blocking(right);
1175
1176                 right_nr = btrfs_header_nritems(right);
1177                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1178                         wret = 1;
1179                 } else {
1180                         ret = btrfs_cow_block(trans, root, right,
1181                                               parent, pslot + 1,
1182                                               &right);
1183                         if (ret)
1184                                 wret = 1;
1185                         else {
1186                                 wret = balance_node_right(trans, root,
1187                                                           right, mid);
1188                         }
1189                 }
1190                 if (wret < 0)
1191                         ret = wret;
1192                 if (wret == 0) {
1193                         struct btrfs_disk_key disk_key;
1194
1195                         btrfs_node_key(right, &disk_key, 0);
1196                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1197                         btrfs_mark_buffer_dirty(parent);
1198
1199                         if (btrfs_header_nritems(mid) <= orig_slot) {
1200                                 path->nodes[level] = right;
1201                                 path->slots[level + 1] += 1;
1202                                 path->slots[level] = orig_slot -
1203                                         btrfs_header_nritems(mid);
1204                                 btrfs_tree_unlock(mid);
1205                                 free_extent_buffer(mid);
1206                         } else {
1207                                 btrfs_tree_unlock(right);
1208                                 free_extent_buffer(right);
1209                         }
1210                         return 0;
1211                 }
1212                 btrfs_tree_unlock(right);
1213                 free_extent_buffer(right);
1214         }
1215         return 1;
1216 }
1217
1218 /*
1219  * readahead one full node of leaves, finding things that are close
1220  * to the block in 'slot', and triggering ra on them.
1221  */
1222 static void reada_for_search(struct btrfs_root *root,
1223                              struct btrfs_path *path,
1224                              int level, int slot, u64 objectid)
1225 {
1226         struct extent_buffer *node;
1227         struct btrfs_disk_key disk_key;
1228         u32 nritems;
1229         u64 search;
1230         u64 target;
1231         u64 nread = 0;
1232         u64 gen;
1233         int direction = path->reada;
1234         struct extent_buffer *eb;
1235         u32 nr;
1236         u32 blocksize;
1237         u32 nscan = 0;
1238
1239         if (level != 1)
1240                 return;
1241
1242         if (!path->nodes[level])
1243                 return;
1244
1245         node = path->nodes[level];
1246
1247         search = btrfs_node_blockptr(node, slot);
1248         blocksize = btrfs_level_size(root, level - 1);
1249         eb = btrfs_find_tree_block(root, search, blocksize);
1250         if (eb) {
1251                 free_extent_buffer(eb);
1252                 return;
1253         }
1254
1255         target = search;
1256
1257         nritems = btrfs_header_nritems(node);
1258         nr = slot;
1259
1260         while (1) {
1261                 if (direction < 0) {
1262                         if (nr == 0)
1263                                 break;
1264                         nr--;
1265                 } else if (direction > 0) {
1266                         nr++;
1267                         if (nr >= nritems)
1268                                 break;
1269                 }
1270                 if (path->reada < 0 && objectid) {
1271                         btrfs_node_key(node, &disk_key, nr);
1272                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1273                                 break;
1274                 }
1275                 search = btrfs_node_blockptr(node, nr);
1276                 if ((search <= target && target - search <= 65536) ||
1277                     (search > target && search - target <= 65536)) {
1278                         gen = btrfs_node_ptr_generation(node, nr);
1279                         readahead_tree_block(root, search, blocksize, gen);
1280                         nread += blocksize;
1281                 }
1282                 nscan++;
1283                 if ((nread > 65536 || nscan > 32))
1284                         break;
1285         }
1286 }
1287
1288 /*
1289  * returns -EAGAIN if it had to drop the path, or zero if everything was in
1290  * cache
1291  */
1292 static noinline int reada_for_balance(struct btrfs_root *root,
1293                                       struct btrfs_path *path, int level)
1294 {
1295         int slot;
1296         int nritems;
1297         struct extent_buffer *parent;
1298         struct extent_buffer *eb;
1299         u64 gen;
1300         u64 block1 = 0;
1301         u64 block2 = 0;
1302         int ret = 0;
1303         int blocksize;
1304
1305         parent = path->nodes[level + 1];
1306         if (!parent)
1307                 return 0;
1308
1309         nritems = btrfs_header_nritems(parent);
1310         slot = path->slots[level + 1];
1311         blocksize = btrfs_level_size(root, level);
1312
1313         if (slot > 0) {
1314                 block1 = btrfs_node_blockptr(parent, slot - 1);
1315                 gen = btrfs_node_ptr_generation(parent, slot - 1);
1316                 eb = btrfs_find_tree_block(root, block1, blocksize);
1317                 if (eb && btrfs_buffer_uptodate(eb, gen))
1318                         block1 = 0;
1319                 free_extent_buffer(eb);
1320         }
1321         if (slot + 1 < nritems) {
1322                 block2 = btrfs_node_blockptr(parent, slot + 1);
1323                 gen = btrfs_node_ptr_generation(parent, slot + 1);
1324                 eb = btrfs_find_tree_block(root, block2, blocksize);
1325                 if (eb && btrfs_buffer_uptodate(eb, gen))
1326                         block2 = 0;
1327                 free_extent_buffer(eb);
1328         }
1329         if (block1 || block2) {
1330                 ret = -EAGAIN;
1331
1332                 /* release the whole path */
1333                 btrfs_release_path(path);
1334
1335                 /* read the blocks */
1336                 if (block1)
1337                         readahead_tree_block(root, block1, blocksize, 0);
1338                 if (block2)
1339                         readahead_tree_block(root, block2, blocksize, 0);
1340
1341                 if (block1) {
1342                         eb = read_tree_block(root, block1, blocksize, 0);
1343                         free_extent_buffer(eb);
1344                 }
1345                 if (block2) {
1346                         eb = read_tree_block(root, block2, blocksize, 0);
1347                         free_extent_buffer(eb);
1348                 }
1349         }
1350         return ret;
1351 }
1352
1353
1354 /*
1355  * when we walk down the tree, it is usually safe to unlock the higher layers
1356  * in the tree.  The exceptions are when our path goes through slot 0, because
1357  * operations on the tree might require changing key pointers higher up in the
1358  * tree.
1359  *
1360  * callers might also have set path->keep_locks, which tells this code to keep
1361  * the lock if the path points to the last slot in the block.  This is part of
1362  * walking through the tree, and selecting the next slot in the higher block.
1363  *
1364  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1365  * if lowest_unlock is 1, level 0 won't be unlocked
1366  */
1367 static noinline void unlock_up(struct btrfs_path *path, int level,
1368                                int lowest_unlock)
1369 {
1370         int i;
1371         int skip_level = level;
1372         int no_skips = 0;
1373         struct extent_buffer *t;
1374
1375         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1376                 if (!path->nodes[i])
1377                         break;
1378                 if (!path->locks[i])
1379                         break;
1380                 if (!no_skips && path->slots[i] == 0) {
1381                         skip_level = i + 1;
1382                         continue;
1383                 }
1384                 if (!no_skips && path->keep_locks) {
1385                         u32 nritems;
1386                         t = path->nodes[i];
1387                         nritems = btrfs_header_nritems(t);
1388                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1389                                 skip_level = i + 1;
1390                                 continue;
1391                         }
1392                 }
1393                 if (skip_level < i && i >= lowest_unlock)
1394                         no_skips = 1;
1395
1396                 t = path->nodes[i];
1397                 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1398                         btrfs_tree_unlock_rw(t, path->locks[i]);
1399                         path->locks[i] = 0;
1400                 }
1401         }
1402 }
1403
1404 /*
1405  * This releases any locks held in the path starting at level and
1406  * going all the way up to the root.
1407  *
1408  * btrfs_search_slot will keep the lock held on higher nodes in a few
1409  * corner cases, such as COW of the block at slot zero in the node.  This
1410  * ignores those rules, and it should only be called when there are no
1411  * more updates to be done higher up in the tree.
1412  */
1413 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1414 {
1415         int i;
1416
1417         if (path->keep_locks)
1418                 return;
1419
1420         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1421                 if (!path->nodes[i])
1422                         continue;
1423                 if (!path->locks[i])
1424                         continue;
1425                 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
1426                 path->locks[i] = 0;
1427         }
1428 }
1429
1430 /*
1431  * helper function for btrfs_search_slot.  The goal is to find a block
1432  * in cache without setting the path to blocking.  If we find the block
1433  * we return zero and the path is unchanged.
1434  *
1435  * If we can't find the block, we set the path blocking and do some
1436  * reada.  -EAGAIN is returned and the search must be repeated.
1437  */
1438 static int
1439 read_block_for_search(struct btrfs_trans_handle *trans,
1440                        struct btrfs_root *root, struct btrfs_path *p,
1441                        struct extent_buffer **eb_ret, int level, int slot,
1442                        struct btrfs_key *key)
1443 {
1444         u64 blocknr;
1445         u64 gen;
1446         u32 blocksize;
1447         struct extent_buffer *b = *eb_ret;
1448         struct extent_buffer *tmp;
1449         int ret;
1450
1451         blocknr = btrfs_node_blockptr(b, slot);
1452         gen = btrfs_node_ptr_generation(b, slot);
1453         blocksize = btrfs_level_size(root, level - 1);
1454
1455         tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1456         if (tmp) {
1457                 if (btrfs_buffer_uptodate(tmp, 0)) {
1458                         if (btrfs_buffer_uptodate(tmp, gen)) {
1459                                 /*
1460                                  * we found an up to date block without
1461                                  * sleeping, return
1462                                  * right away
1463                                  */
1464                                 *eb_ret = tmp;
1465                                 return 0;
1466                         }
1467                         /* the pages were up to date, but we failed
1468                          * the generation number check.  Do a full
1469                          * read for the generation number that is correct.
1470                          * We must do this without dropping locks so
1471                          * we can trust our generation number
1472                          */
1473                         free_extent_buffer(tmp);
1474                         btrfs_set_path_blocking(p);
1475
1476                         tmp = read_tree_block(root, blocknr, blocksize, gen);
1477                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1478                                 *eb_ret = tmp;
1479                                 return 0;
1480                         }
1481                         free_extent_buffer(tmp);
1482                         btrfs_release_path(p);
1483                         return -EIO;
1484                 }
1485         }
1486
1487         /*
1488          * reduce lock contention at high levels
1489          * of the btree by dropping locks before
1490          * we read.  Don't release the lock on the current
1491          * level because we need to walk this node to figure
1492          * out which blocks to read.
1493          */
1494         btrfs_unlock_up_safe(p, level + 1);
1495         btrfs_set_path_blocking(p);
1496
1497         free_extent_buffer(tmp);
1498         if (p->reada)
1499                 reada_for_search(root, p, level, slot, key->objectid);
1500
1501         btrfs_release_path(p);
1502
1503         ret = -EAGAIN;
1504         tmp = read_tree_block(root, blocknr, blocksize, 0);
1505         if (tmp) {
1506                 /*
1507                  * If the read above didn't mark this buffer up to date,
1508                  * it will never end up being up to date.  Set ret to EIO now
1509                  * and give up so that our caller doesn't loop forever
1510                  * on our EAGAINs.
1511                  */
1512                 if (!btrfs_buffer_uptodate(tmp, 0))
1513                         ret = -EIO;
1514                 free_extent_buffer(tmp);
1515         }
1516         return ret;
1517 }
1518
1519 /*
1520  * helper function for btrfs_search_slot.  This does all of the checks
1521  * for node-level blocks and does any balancing required based on
1522  * the ins_len.
1523  *
1524  * If no extra work was required, zero is returned.  If we had to
1525  * drop the path, -EAGAIN is returned and btrfs_search_slot must
1526  * start over
1527  */
1528 static int
1529 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1530                        struct btrfs_root *root, struct btrfs_path *p,
1531                        struct extent_buffer *b, int level, int ins_len,
1532                        int *write_lock_level)
1533 {
1534         int ret;
1535         if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1536             BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1537                 int sret;
1538
1539                 if (*write_lock_level < level + 1) {
1540                         *write_lock_level = level + 1;
1541                         btrfs_release_path(p);
1542                         goto again;
1543                 }
1544
1545                 sret = reada_for_balance(root, p, level);
1546                 if (sret)
1547                         goto again;
1548
1549                 btrfs_set_path_blocking(p);
1550                 sret = split_node(trans, root, p, level);
1551                 btrfs_clear_path_blocking(p, NULL, 0);
1552
1553                 BUG_ON(sret > 0);
1554                 if (sret) {
1555                         ret = sret;
1556                         goto done;
1557                 }
1558                 b = p->nodes[level];
1559         } else if (ins_len < 0 && btrfs_header_nritems(b) <
1560                    BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1561                 int sret;
1562
1563                 if (*write_lock_level < level + 1) {
1564                         *write_lock_level = level + 1;
1565                         btrfs_release_path(p);
1566                         goto again;
1567                 }
1568
1569                 sret = reada_for_balance(root, p, level);
1570                 if (sret)
1571                         goto again;
1572
1573                 btrfs_set_path_blocking(p);
1574                 sret = balance_level(trans, root, p, level);
1575                 btrfs_clear_path_blocking(p, NULL, 0);
1576
1577                 if (sret) {
1578                         ret = sret;
1579                         goto done;
1580                 }
1581                 b = p->nodes[level];
1582                 if (!b) {
1583                         btrfs_release_path(p);
1584                         goto again;
1585                 }
1586                 BUG_ON(btrfs_header_nritems(b) == 1);
1587         }
1588         return 0;
1589
1590 again:
1591         ret = -EAGAIN;
1592 done:
1593         return ret;
1594 }
1595
1596 /*
1597  * look for key in the tree.  path is filled in with nodes along the way
1598  * if key is found, we return zero and you can find the item in the leaf
1599  * level of the path (level 0)
1600  *
1601  * If the key isn't found, the path points to the slot where it should
1602  * be inserted, and 1 is returned.  If there are other errors during the
1603  * search a negative error number is returned.
1604  *
1605  * if ins_len > 0, nodes and leaves will be split as we walk down the
1606  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1607  * possible)
1608  */
1609 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1610                       *root, struct btrfs_key *key, struct btrfs_path *p, int
1611                       ins_len, int cow)
1612 {
1613         struct extent_buffer *b;
1614         int slot;
1615         int ret;
1616         int err;
1617         int level;
1618         int lowest_unlock = 1;
1619         int root_lock;
1620         /* everything at write_lock_level or lower must be write locked */
1621         int write_lock_level = 0;
1622         u8 lowest_level = 0;
1623
1624         lowest_level = p->lowest_level;
1625         WARN_ON(lowest_level && ins_len > 0);
1626         WARN_ON(p->nodes[0] != NULL);
1627
1628         if (ins_len < 0) {
1629                 lowest_unlock = 2;
1630
1631                 /* when we are removing items, we might have to go up to level
1632                  * two as we update tree pointers  Make sure we keep write
1633                  * for those levels as well
1634                  */
1635                 write_lock_level = 2;
1636         } else if (ins_len > 0) {
1637                 /*
1638                  * for inserting items, make sure we have a write lock on
1639                  * level 1 so we can update keys
1640                  */
1641                 write_lock_level = 1;
1642         }
1643
1644         if (!cow)
1645                 write_lock_level = -1;
1646
1647         if (cow && (p->keep_locks || p->lowest_level))
1648                 write_lock_level = BTRFS_MAX_LEVEL;
1649
1650 again:
1651         /*
1652          * we try very hard to do read locks on the root
1653          */
1654         root_lock = BTRFS_READ_LOCK;
1655         level = 0;
1656         if (p->search_commit_root) {
1657                 /*
1658                  * the commit roots are read only
1659                  * so we always do read locks
1660                  */
1661                 b = root->commit_root;
1662                 extent_buffer_get(b);
1663                 level = btrfs_header_level(b);
1664                 if (!p->skip_locking)
1665                         btrfs_tree_read_lock(b);
1666         } else {
1667                 if (p->skip_locking) {
1668                         b = btrfs_root_node(root);
1669                         level = btrfs_header_level(b);
1670                 } else {
1671                         /* we don't know the level of the root node
1672                          * until we actually have it read locked
1673                          */
1674                         b = btrfs_read_lock_root_node(root);
1675                         level = btrfs_header_level(b);
1676                         if (level <= write_lock_level) {
1677                                 /* whoops, must trade for write lock */
1678                                 btrfs_tree_read_unlock(b);
1679                                 free_extent_buffer(b);
1680                                 b = btrfs_lock_root_node(root);
1681                                 root_lock = BTRFS_WRITE_LOCK;
1682
1683                                 /* the level might have changed, check again */
1684                                 level = btrfs_header_level(b);
1685                         }
1686                 }
1687         }
1688         p->nodes[level] = b;
1689         if (!p->skip_locking)
1690                 p->locks[level] = root_lock;
1691
1692         while (b) {
1693                 level = btrfs_header_level(b);
1694
1695                 /*
1696                  * setup the path here so we can release it under lock
1697                  * contention with the cow code
1698                  */
1699                 if (cow) {
1700                         /*
1701                          * if we don't really need to cow this block
1702                          * then we don't want to set the path blocking,
1703                          * so we test it here
1704                          */
1705                         if (!should_cow_block(trans, root, b))
1706                                 goto cow_done;
1707
1708                         btrfs_set_path_blocking(p);
1709
1710                         /*
1711                          * must have write locks on this node and the
1712                          * parent
1713                          */
1714                         if (level + 1 > write_lock_level) {
1715                                 write_lock_level = level + 1;
1716                                 btrfs_release_path(p);
1717                                 goto again;
1718                         }
1719
1720                         err = btrfs_cow_block(trans, root, b,
1721                                               p->nodes[level + 1],
1722                                               p->slots[level + 1], &b);
1723                         if (err) {
1724                                 ret = err;
1725                                 goto done;
1726                         }
1727                 }
1728 cow_done:
1729                 BUG_ON(!cow && ins_len);
1730
1731                 p->nodes[level] = b;
1732                 btrfs_clear_path_blocking(p, NULL, 0);
1733
1734                 /*
1735                  * we have a lock on b and as long as we aren't changing
1736                  * the tree, there is no way to for the items in b to change.
1737                  * It is safe to drop the lock on our parent before we
1738                  * go through the expensive btree search on b.
1739                  *
1740                  * If cow is true, then we might be changing slot zero,
1741                  * which may require changing the parent.  So, we can't
1742                  * drop the lock until after we know which slot we're
1743                  * operating on.
1744                  */
1745                 if (!cow)
1746                         btrfs_unlock_up_safe(p, level + 1);
1747
1748                 ret = bin_search(b, key, level, &slot);
1749
1750                 if (level != 0) {
1751                         int dec = 0;
1752                         if (ret && slot > 0) {
1753                                 dec = 1;
1754                                 slot -= 1;
1755                         }
1756                         p->slots[level] = slot;
1757                         err = setup_nodes_for_search(trans, root, p, b, level,
1758                                              ins_len, &write_lock_level);
1759                         if (err == -EAGAIN)
1760                                 goto again;
1761                         if (err) {
1762                                 ret = err;
1763                                 goto done;
1764                         }
1765                         b = p->nodes[level];
1766                         slot = p->slots[level];
1767
1768                         /*
1769                          * slot 0 is special, if we change the key
1770                          * we have to update the parent pointer
1771                          * which means we must have a write lock
1772                          * on the parent
1773                          */
1774                         if (slot == 0 && cow &&
1775                             write_lock_level < level + 1) {
1776                                 write_lock_level = level + 1;
1777                                 btrfs_release_path(p);
1778                                 goto again;
1779                         }
1780
1781                         unlock_up(p, level, lowest_unlock);
1782
1783                         if (level == lowest_level) {
1784                                 if (dec)
1785                                         p->slots[level]++;
1786                                 goto done;
1787                         }
1788
1789                         err = read_block_for_search(trans, root, p,
1790                                                     &b, level, slot, key);
1791                         if (err == -EAGAIN)
1792                                 goto again;
1793                         if (err) {
1794                                 ret = err;
1795                                 goto done;
1796                         }
1797
1798                         if (!p->skip_locking) {
1799                                 level = btrfs_header_level(b);
1800                                 if (level <= write_lock_level) {
1801                                         err = btrfs_try_tree_write_lock(b);
1802                                         if (!err) {
1803                                                 btrfs_set_path_blocking(p);
1804                                                 btrfs_tree_lock(b);
1805                                                 btrfs_clear_path_blocking(p, b,
1806                                                                   BTRFS_WRITE_LOCK);
1807                                         }
1808                                         p->locks[level] = BTRFS_WRITE_LOCK;
1809                                 } else {
1810                                         err = btrfs_try_tree_read_lock(b);
1811                                         if (!err) {
1812                                                 btrfs_set_path_blocking(p);
1813                                                 btrfs_tree_read_lock(b);
1814                                                 btrfs_clear_path_blocking(p, b,
1815                                                                   BTRFS_READ_LOCK);
1816                                         }
1817                                         p->locks[level] = BTRFS_READ_LOCK;
1818                                 }
1819                                 p->nodes[level] = b;
1820                         }
1821                 } else {
1822                         p->slots[level] = slot;
1823                         if (ins_len > 0 &&
1824                             btrfs_leaf_free_space(root, b) < ins_len) {
1825                                 if (write_lock_level < 1) {
1826                                         write_lock_level = 1;
1827                                         btrfs_release_path(p);
1828                                         goto again;
1829                                 }
1830
1831                                 btrfs_set_path_blocking(p);
1832                                 err = split_leaf(trans, root, key,
1833                                                  p, ins_len, ret == 0);
1834                                 btrfs_clear_path_blocking(p, NULL, 0);
1835
1836                                 BUG_ON(err > 0);
1837                                 if (err) {
1838                                         ret = err;
1839                                         goto done;
1840                                 }
1841                         }
1842                         if (!p->search_for_split)
1843                                 unlock_up(p, level, lowest_unlock);
1844                         goto done;
1845                 }
1846         }
1847         ret = 1;
1848 done:
1849         /*
1850          * we don't really know what they plan on doing with the path
1851          * from here on, so for now just mark it as blocking
1852          */
1853         if (!p->leave_spinning)
1854                 btrfs_set_path_blocking(p);
1855         if (ret < 0)
1856                 btrfs_release_path(p);
1857         return ret;
1858 }
1859
1860 /*
1861  * adjust the pointers going up the tree, starting at level
1862  * making sure the right key of each node is points to 'key'.
1863  * This is used after shifting pointers to the left, so it stops
1864  * fixing up pointers when a given leaf/node is not in slot 0 of the
1865  * higher levels
1866  *
1867  * If this fails to write a tree block, it returns -1, but continues
1868  * fixing up the blocks in ram so the tree is consistent.
1869  */
1870 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1871                           struct btrfs_root *root, struct btrfs_path *path,
1872                           struct btrfs_disk_key *key, int level)
1873 {
1874         int i;
1875         int ret = 0;
1876         struct extent_buffer *t;
1877
1878         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1879                 int tslot = path->slots[i];
1880                 if (!path->nodes[i])
1881                         break;
1882                 t = path->nodes[i];
1883                 btrfs_set_node_key(t, key, tslot);
1884                 btrfs_mark_buffer_dirty(path->nodes[i]);
1885                 if (tslot != 0)
1886                         break;
1887         }
1888         return ret;
1889 }
1890
1891 /*
1892  * update item key.
1893  *
1894  * This function isn't completely safe. It's the caller's responsibility
1895  * that the new key won't break the order
1896  */
1897 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1898                             struct btrfs_root *root, struct btrfs_path *path,
1899                             struct btrfs_key *new_key)
1900 {
1901         struct btrfs_disk_key disk_key;
1902         struct extent_buffer *eb;
1903         int slot;
1904
1905         eb = path->nodes[0];
1906         slot = path->slots[0];
1907         if (slot > 0) {
1908                 btrfs_item_key(eb, &disk_key, slot - 1);
1909                 if (comp_keys(&disk_key, new_key) >= 0)
1910                         return -1;
1911         }
1912         if (slot < btrfs_header_nritems(eb) - 1) {
1913                 btrfs_item_key(eb, &disk_key, slot + 1);
1914                 if (comp_keys(&disk_key, new_key) <= 0)
1915                         return -1;
1916         }
1917
1918         btrfs_cpu_key_to_disk(&disk_key, new_key);
1919         btrfs_set_item_key(eb, &disk_key, slot);
1920         btrfs_mark_buffer_dirty(eb);
1921         if (slot == 0)
1922                 fixup_low_keys(trans, root, path, &disk_key, 1);
1923         return 0;
1924 }
1925
1926 /*
1927  * try to push data from one node into the next node left in the
1928  * tree.
1929  *
1930  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1931  * error, and > 0 if there was no room in the left hand block.
1932  */
1933 static int push_node_left(struct btrfs_trans_handle *trans,
1934                           struct btrfs_root *root, struct extent_buffer *dst,
1935                           struct extent_buffer *src, int empty)
1936 {
1937         int push_items = 0;
1938         int src_nritems;
1939         int dst_nritems;
1940         int ret = 0;
1941
1942         src_nritems = btrfs_header_nritems(src);
1943         dst_nritems = btrfs_header_nritems(dst);
1944         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1945         WARN_ON(btrfs_header_generation(src) != trans->transid);
1946         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1947
1948         if (!empty && src_nritems <= 8)
1949                 return 1;
1950
1951         if (push_items <= 0)
1952                 return 1;
1953
1954         if (empty) {
1955                 push_items = min(src_nritems, push_items);
1956                 if (push_items < src_nritems) {
1957                         /* leave at least 8 pointers in the node if
1958                          * we aren't going to empty it
1959                          */
1960                         if (src_nritems - push_items < 8) {
1961                                 if (push_items <= 8)
1962                                         return 1;
1963                                 push_items -= 8;
1964                         }
1965                 }
1966         } else
1967                 push_items = min(src_nritems - 8, push_items);
1968
1969         copy_extent_buffer(dst, src,
1970                            btrfs_node_key_ptr_offset(dst_nritems),
1971                            btrfs_node_key_ptr_offset(0),
1972                            push_items * sizeof(struct btrfs_key_ptr));
1973
1974         if (push_items < src_nritems) {
1975                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1976                                       btrfs_node_key_ptr_offset(push_items),
1977                                       (src_nritems - push_items) *
1978                                       sizeof(struct btrfs_key_ptr));
1979         }
1980         btrfs_set_header_nritems(src, src_nritems - push_items);
1981         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1982         btrfs_mark_buffer_dirty(src);
1983         btrfs_mark_buffer_dirty(dst);
1984
1985         return ret;
1986 }
1987
1988 /*
1989  * try to push data from one node into the next node right in the
1990  * tree.
1991  *
1992  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1993  * error, and > 0 if there was no room in the right hand block.
1994  *
1995  * this will  only push up to 1/2 the contents of the left node over
1996  */
1997 static int balance_node_right(struct btrfs_trans_handle *trans,
1998                               struct btrfs_root *root,
1999                               struct extent_buffer *dst,
2000                               struct extent_buffer *src)
2001 {
2002         int push_items = 0;
2003         int max_push;
2004         int src_nritems;
2005         int dst_nritems;
2006         int ret = 0;
2007
2008         WARN_ON(btrfs_header_generation(src) != trans->transid);
2009         WARN_ON(btrfs_header_generation(dst) != trans->transid);
2010
2011         src_nritems = btrfs_header_nritems(src);
2012         dst_nritems = btrfs_header_nritems(dst);
2013         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2014         if (push_items <= 0)
2015                 return 1;
2016
2017         if (src_nritems < 4)
2018                 return 1;
2019
2020         max_push = src_nritems / 2 + 1;
2021         /* don't try to empty the node */
2022         if (max_push >= src_nritems)
2023                 return 1;
2024
2025         if (max_push < push_items)
2026                 push_items = max_push;
2027
2028         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2029                                       btrfs_node_key_ptr_offset(0),
2030                                       (dst_nritems) *
2031                                       sizeof(struct btrfs_key_ptr));
2032
2033         copy_extent_buffer(dst, src,
2034                            btrfs_node_key_ptr_offset(0),
2035                            btrfs_node_key_ptr_offset(src_nritems - push_items),
2036                            push_items * sizeof(struct btrfs_key_ptr));
2037
2038         btrfs_set_header_nritems(src, src_nritems - push_items);
2039         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2040
2041         btrfs_mark_buffer_dirty(src);
2042         btrfs_mark_buffer_dirty(dst);
2043
2044         return ret;
2045 }
2046
2047 /*
2048  * helper function to insert a new root level in the tree.
2049  * A new node is allocated, and a single item is inserted to
2050  * point to the existing root
2051  *
2052  * returns zero on success or < 0 on failure.
2053  */
2054 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2055                            struct btrfs_root *root,
2056                            struct btrfs_path *path, int level)
2057 {
2058         u64 lower_gen;
2059         struct extent_buffer *lower;
2060         struct extent_buffer *c;
2061         struct extent_buffer *old;
2062         struct btrfs_disk_key lower_key;
2063
2064         BUG_ON(path->nodes[level]);
2065         BUG_ON(path->nodes[level-1] != root->node);
2066
2067         lower = path->nodes[level-1];
2068         if (level == 1)
2069                 btrfs_item_key(lower, &lower_key, 0);
2070         else
2071                 btrfs_node_key(lower, &lower_key, 0);
2072
2073         c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2074                                    root->root_key.objectid, &lower_key,
2075                                    level, root->node->start, 0);
2076         if (IS_ERR(c))
2077                 return PTR_ERR(c);
2078
2079         root_add_used(root, root->nodesize);
2080
2081         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2082         btrfs_set_header_nritems(c, 1);
2083         btrfs_set_header_level(c, level);
2084         btrfs_set_header_bytenr(c, c->start);
2085         btrfs_set_header_generation(c, trans->transid);
2086         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2087         btrfs_set_header_owner(c, root->root_key.objectid);
2088
2089         write_extent_buffer(c, root->fs_info->fsid,
2090                             (unsigned long)btrfs_header_fsid(c),
2091                             BTRFS_FSID_SIZE);
2092
2093         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2094                             (unsigned long)btrfs_header_chunk_tree_uuid(c),
2095                             BTRFS_UUID_SIZE);
2096
2097         btrfs_set_node_key(c, &lower_key, 0);
2098         btrfs_set_node_blockptr(c, 0, lower->start);
2099         lower_gen = btrfs_header_generation(lower);
2100         WARN_ON(lower_gen != trans->transid);
2101
2102         btrfs_set_node_ptr_generation(c, 0, lower_gen);
2103
2104         btrfs_mark_buffer_dirty(c);
2105
2106         old = root->node;
2107         rcu_assign_pointer(root->node, c);
2108
2109         /* the super has an extra ref to root->node */
2110         free_extent_buffer(old);
2111
2112         add_root_to_dirty_list(root);
2113         extent_buffer_get(c);
2114         path->nodes[level] = c;
2115         path->locks[level] = BTRFS_WRITE_LOCK;
2116         path->slots[level] = 0;
2117         return 0;
2118 }
2119
2120 /*
2121  * worker function to insert a single pointer in a node.
2122  * the node should have enough room for the pointer already
2123  *
2124  * slot and level indicate where you want the key to go, and
2125  * blocknr is the block the key points to.
2126  *
2127  * returns zero on success and < 0 on any error
2128  */
2129 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2130                       *root, struct btrfs_path *path, struct btrfs_disk_key
2131                       *key, u64 bytenr, int slot, int level)
2132 {
2133         struct extent_buffer *lower;
2134         int nritems;
2135
2136         BUG_ON(!path->nodes[level]);
2137         btrfs_assert_tree_locked(path->nodes[level]);
2138         lower = path->nodes[level];
2139         nritems = btrfs_header_nritems(lower);
2140         BUG_ON(slot > nritems);
2141         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2142                 BUG();
2143         if (slot != nritems) {
2144                 memmove_extent_buffer(lower,
2145                               btrfs_node_key_ptr_offset(slot + 1),
2146                               btrfs_node_key_ptr_offset(slot),
2147                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
2148         }
2149         btrfs_set_node_key(lower, key, slot);
2150         btrfs_set_node_blockptr(lower, slot, bytenr);
2151         WARN_ON(trans->transid == 0);
2152         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2153         btrfs_set_header_nritems(lower, nritems + 1);
2154         btrfs_mark_buffer_dirty(lower);
2155         return 0;
2156 }
2157
2158 /*
2159  * split the node at the specified level in path in two.
2160  * The path is corrected to point to the appropriate node after the split
2161  *
2162  * Before splitting this tries to make some room in the node by pushing
2163  * left and right, if either one works, it returns right away.
2164  *
2165  * returns 0 on success and < 0 on failure
2166  */
2167 static noinline int split_node(struct btrfs_trans_handle *trans,
2168                                struct btrfs_root *root,
2169                                struct btrfs_path *path, int level)
2170 {
2171         struct extent_buffer *c;
2172         struct extent_buffer *split;
2173         struct btrfs_disk_key disk_key;
2174         int mid;
2175         int ret;
2176         int wret;
2177         u32 c_nritems;
2178
2179         c = path->nodes[level];
2180         WARN_ON(btrfs_header_generation(c) != trans->transid);
2181         if (c == root->node) {
2182                 /* trying to split the root, lets make a new one */
2183                 ret = insert_new_root(trans, root, path, level + 1);
2184                 if (ret)
2185                         return ret;
2186         } else {
2187                 ret = push_nodes_for_insert(trans, root, path, level);
2188                 c = path->nodes[level];
2189                 if (!ret && btrfs_header_nritems(c) <
2190                     BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2191                         return 0;
2192                 if (ret < 0)
2193                         return ret;
2194         }
2195
2196         c_nritems = btrfs_header_nritems(c);
2197         mid = (c_nritems + 1) / 2;
2198         btrfs_node_key(c, &disk_key, mid);
2199
2200         split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2201                                         root->root_key.objectid,
2202                                         &disk_key, level, c->start, 0);
2203         if (IS_ERR(split))
2204                 return PTR_ERR(split);
2205
2206         root_add_used(root, root->nodesize);
2207
2208         memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2209         btrfs_set_header_level(split, btrfs_header_level(c));
2210         btrfs_set_header_bytenr(split, split->start);
2211         btrfs_set_header_generation(split, trans->transid);
2212         btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2213         btrfs_set_header_owner(split, root->root_key.objectid);
2214         write_extent_buffer(split, root->fs_info->fsid,
2215                             (unsigned long)btrfs_header_fsid(split),
2216                             BTRFS_FSID_SIZE);
2217         write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2218                             (unsigned long)btrfs_header_chunk_tree_uuid(split),
2219                             BTRFS_UUID_SIZE);
2220
2221
2222         copy_extent_buffer(split, c,
2223                            btrfs_node_key_ptr_offset(0),
2224                            btrfs_node_key_ptr_offset(mid),
2225                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2226         btrfs_set_header_nritems(split, c_nritems - mid);
2227         btrfs_set_header_nritems(c, mid);
2228         ret = 0;
2229
2230         btrfs_mark_buffer_dirty(c);
2231         btrfs_mark_buffer_dirty(split);
2232
2233         wret = insert_ptr(trans, root, path, &disk_key, split->start,
2234                           path->slots[level + 1] + 1,
2235                           level + 1);
2236         if (wret)
2237                 ret = wret;
2238
2239         if (path->slots[level] >= mid) {
2240                 path->slots[level] -= mid;
2241                 btrfs_tree_unlock(c);
2242                 free_extent_buffer(c);
2243                 path->nodes[level] = split;
2244                 path->slots[level + 1] += 1;
2245         } else {
2246                 btrfs_tree_unlock(split);
2247                 free_extent_buffer(split);
2248         }
2249         return ret;
2250 }
2251
2252 /*
2253  * how many bytes are required to store the items in a leaf.  start
2254  * and nr indicate which items in the leaf to check.  This totals up the
2255  * space used both by the item structs and the item data
2256  */
2257 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2258 {
2259         int data_len;
2260         int nritems = btrfs_header_nritems(l);
2261         int end = min(nritems, start + nr) - 1;
2262
2263         if (!nr)
2264                 return 0;
2265         data_len = btrfs_item_end_nr(l, start);
2266         data_len = data_len - btrfs_item_offset_nr(l, end);
2267         data_len += sizeof(struct btrfs_item) * nr;
2268         WARN_ON(data_len < 0);
2269         return data_len;
2270 }
2271
2272 /*
2273  * The space between the end of the leaf items and
2274  * the start of the leaf data.  IOW, how much room
2275  * the leaf has left for both items and data
2276  */
2277 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2278                                    struct extent_buffer *leaf)
2279 {
2280         int nritems = btrfs_header_nritems(leaf);
2281         int ret;
2282         ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2283         if (ret < 0) {
2284                 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2285                        "used %d nritems %d\n",
2286                        ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2287                        leaf_space_used(leaf, 0, nritems), nritems);
2288         }
2289         return ret;
2290 }
2291
2292 /*
2293  * min slot controls the lowest index we're willing to push to the
2294  * right.  We'll push up to and including min_slot, but no lower
2295  */
2296 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2297                                       struct btrfs_root *root,
2298                                       struct btrfs_path *path,
2299                                       int data_size, int empty,
2300                                       struct extent_buffer *right,
2301                                       int free_space, u32 left_nritems,
2302                                       u32 min_slot)
2303 {
2304         struct extent_buffer *left = path->nodes[0];
2305         struct extent_buffer *upper = path->nodes[1];
2306         struct btrfs_disk_key disk_key;
2307         int slot;
2308         u32 i;
2309         int push_space = 0;
2310         int push_items = 0;
2311         struct btrfs_item *item;
2312         u32 nr;
2313         u32 right_nritems;
2314         u32 data_end;
2315         u32 this_item_size;
2316
2317         if (empty)
2318                 nr = 0;
2319         else
2320                 nr = max_t(u32, 1, min_slot);
2321
2322         if (path->slots[0] >= left_nritems)
2323                 push_space += data_size;
2324
2325         slot = path->slots[1];
2326         i = left_nritems - 1;
2327         while (i >= nr) {
2328                 item = btrfs_item_nr(left, i);
2329
2330                 if (!empty && push_items > 0) {
2331                         if (path->slots[0] > i)
2332                                 break;
2333                         if (path->slots[0] == i) {
2334                                 int space = btrfs_leaf_free_space(root, left);
2335                                 if (space + push_space * 2 > free_space)
2336                                         break;
2337                         }
2338                 }
2339
2340                 if (path->slots[0] == i)
2341                         push_space += data_size;
2342
2343                 this_item_size = btrfs_item_size(left, item);
2344                 if (this_item_size + sizeof(*item) + push_space > free_space)
2345                         break;
2346
2347                 push_items++;
2348                 push_space += this_item_size + sizeof(*item);
2349                 if (i == 0)
2350                         break;
2351                 i--;
2352         }
2353
2354         if (push_items == 0)
2355                 goto out_unlock;
2356
2357         if (!empty && push_items == left_nritems)
2358                 WARN_ON(1);
2359
2360         /* push left to right */
2361         right_nritems = btrfs_header_nritems(right);
2362
2363         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2364         push_space -= leaf_data_end(root, left);
2365
2366         /* make room in the right data area */
2367         data_end = leaf_data_end(root, right);
2368         memmove_extent_buffer(right,
2369                               btrfs_leaf_data(right) + data_end - push_space,
2370                               btrfs_leaf_data(right) + data_end,
2371                               BTRFS_LEAF_DATA_SIZE(root) - data_end);
2372
2373         /* copy from the left data area */
2374         copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2375                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2376                      btrfs_leaf_data(left) + leaf_data_end(root, left),
2377                      push_space);
2378
2379         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2380                               btrfs_item_nr_offset(0),
2381                               right_nritems * sizeof(struct btrfs_item));
2382
2383         /* copy the items from left to right */
2384         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2385                    btrfs_item_nr_offset(left_nritems - push_items),
2386                    push_items * sizeof(struct btrfs_item));
2387
2388         /* update the item pointers */
2389         right_nritems += push_items;
2390         btrfs_set_header_nritems(right, right_nritems);
2391         push_space = BTRFS_LEAF_DATA_SIZE(root);
2392         for (i = 0; i < right_nritems; i++) {
2393                 item = btrfs_item_nr(right, i);
2394                 push_space -= btrfs_item_size(right, item);
2395                 btrfs_set_item_offset(right, item, push_space);
2396         }
2397
2398         left_nritems -= push_items;
2399         btrfs_set_header_nritems(left, left_nritems);
2400
2401         if (left_nritems)
2402                 btrfs_mark_buffer_dirty(left);
2403         else
2404                 clean_tree_block(trans, root, left);
2405
2406         btrfs_mark_buffer_dirty(right);
2407
2408         btrfs_item_key(right, &disk_key, 0);
2409         btrfs_set_node_key(upper, &disk_key, slot + 1);
2410         btrfs_mark_buffer_dirty(upper);
2411
2412         /* then fixup the leaf pointer in the path */
2413         if (path->slots[0] >= left_nritems) {
2414                 path->slots[0] -= left_nritems;
2415                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2416                         clean_tree_block(trans, root, path->nodes[0]);
2417                 btrfs_tree_unlock(path->nodes[0]);
2418                 free_extent_buffer(path->nodes[0]);
2419                 path->nodes[0] = right;
2420                 path->slots[1] += 1;
2421         } else {
2422                 btrfs_tree_unlock(right);
2423                 free_extent_buffer(right);
2424         }
2425         return 0;
2426
2427 out_unlock:
2428         btrfs_tree_unlock(right);
2429         free_extent_buffer(right);
2430         return 1;
2431 }
2432
2433 /*
2434  * push some data in the path leaf to the right, trying to free up at
2435  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2436  *
2437  * returns 1 if the push failed because the other node didn't have enough
2438  * room, 0 if everything worked out and < 0 if there were major errors.
2439  *
2440  * this will push starting from min_slot to the end of the leaf.  It won't
2441  * push any slot lower than min_slot
2442  */
2443 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2444                            *root, struct btrfs_path *path,
2445                            int min_data_size, int data_size,
2446                            int empty, u32 min_slot)
2447 {
2448         struct extent_buffer *left = path->nodes[0];
2449         struct extent_buffer *right;
2450         struct extent_buffer *upper;
2451         int slot;
2452         int free_space;
2453         u32 left_nritems;
2454         int ret;
2455
2456         if (!path->nodes[1])
2457                 return 1;
2458
2459         slot = path->slots[1];
2460         upper = path->nodes[1];
2461         if (slot >= btrfs_header_nritems(upper) - 1)
2462                 return 1;
2463
2464         btrfs_assert_tree_locked(path->nodes[1]);
2465
2466         right = read_node_slot(root, upper, slot + 1);
2467         if (right == NULL)
2468                 return 1;
2469
2470         btrfs_tree_lock(right);
2471         btrfs_set_lock_blocking(right);
2472
2473         free_space = btrfs_leaf_free_space(root, right);
2474         if (free_space < data_size)
2475                 goto out_unlock;
2476
2477         /* cow and double check */
2478         ret = btrfs_cow_block(trans, root, right, upper,
2479                               slot + 1, &right);
2480         if (ret)
2481                 goto out_unlock;
2482
2483         free_space = btrfs_leaf_free_space(root, right);
2484         if (free_space < data_size)
2485                 goto out_unlock;
2486
2487         left_nritems = btrfs_header_nritems(left);
2488         if (left_nritems == 0)
2489                 goto out_unlock;
2490
2491         return __push_leaf_right(trans, root, path, min_data_size, empty,
2492                                 right, free_space, left_nritems, min_slot);
2493 out_unlock:
2494         btrfs_tree_unlock(right);
2495         free_extent_buffer(right);
2496         return 1;
2497 }
2498
2499 /*
2500  * push some data in the path leaf to the left, trying to free up at
2501  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2502  *
2503  * max_slot can put a limit on how far into the leaf we'll push items.  The
2504  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us do all the
2505  * items
2506  */
2507 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2508                                      struct btrfs_root *root,
2509                                      struct btrfs_path *path, int data_size,
2510                                      int empty, struct extent_buffer *left,
2511                                      int free_space, u32 right_nritems,
2512                                      u32 max_slot)
2513 {
2514         struct btrfs_disk_key disk_key;
2515         struct extent_buffer *right = path->nodes[0];
2516         int i;
2517         int push_space = 0;
2518         int push_items = 0;
2519         struct btrfs_item *item;
2520         u32 old_left_nritems;
2521         u32 nr;
2522         int ret = 0;
2523         int wret;
2524         u32 this_item_size;
2525         u32 old_left_item_size;
2526
2527         if (empty)
2528                 nr = min(right_nritems, max_slot);
2529         else
2530                 nr = min(right_nritems - 1, max_slot);
2531
2532         for (i = 0; i < nr; i++) {
2533                 item = btrfs_item_nr(right, i);
2534
2535                 if (!empty && push_items > 0) {
2536                         if (path->slots[0] < i)
2537                                 break;
2538                         if (path->slots[0] == i) {
2539                                 int space = btrfs_leaf_free_space(root, right);
2540                                 if (space + push_space * 2 > free_space)
2541                                         break;
2542                         }
2543                 }
2544
2545                 if (path->slots[0] == i)
2546                         push_space += data_size;
2547
2548                 this_item_size = btrfs_item_size(right, item);
2549                 if (this_item_size + sizeof(*item) + push_space > free_space)
2550                         break;
2551
2552                 push_items++;
2553                 push_space += this_item_size + sizeof(*item);
2554         }
2555
2556         if (push_items == 0) {
2557                 ret = 1;
2558                 goto out;
2559         }
2560         if (!empty && push_items == btrfs_header_nritems(right))
2561                 WARN_ON(1);
2562
2563         /* push data from right to left */
2564         copy_extent_buffer(left, right,
2565                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2566                            btrfs_item_nr_offset(0),
2567                            push_items * sizeof(struct btrfs_item));
2568
2569         push_space = BTRFS_LEAF_DATA_SIZE(root) -
2570                      btrfs_item_offset_nr(right, push_items - 1);
2571
2572         copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2573                      leaf_data_end(root, left) - push_space,
2574                      btrfs_leaf_data(right) +
2575                      btrfs_item_offset_nr(right, push_items - 1),
2576                      push_space);
2577         old_left_nritems = btrfs_header_nritems(left);
2578         BUG_ON(old_left_nritems <= 0);
2579
2580         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2581         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2582                 u32 ioff;
2583
2584                 item = btrfs_item_nr(left, i);
2585
2586                 ioff = btrfs_item_offset(left, item);
2587                 btrfs_set_item_offset(left, item,
2588                       ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2589         }
2590         btrfs_set_header_nritems(left, old_left_nritems + push_items);
2591
2592         /* fixup right node */
2593         if (push_items > right_nritems) {
2594                 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2595                        right_nritems);
2596                 WARN_ON(1);
2597         }
2598
2599         if (push_items < right_nritems) {
2600                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2601                                                   leaf_data_end(root, right);
2602                 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2603                                       BTRFS_LEAF_DATA_SIZE(root) - push_space,
2604                                       btrfs_leaf_data(right) +
2605                                       leaf_data_end(root, right), push_space);
2606
2607                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2608                               btrfs_item_nr_offset(push_items),
2609                              (btrfs_header_nritems(right) - push_items) *
2610                              sizeof(struct btrfs_item));
2611         }
2612         right_nritems -= push_items;
2613         btrfs_set_header_nritems(right, right_nritems);
2614         push_space = BTRFS_LEAF_DATA_SIZE(root);
2615         for (i = 0; i < right_nritems; i++) {
2616                 item = btrfs_item_nr(right, i);
2617
2618                 push_space = push_space - btrfs_item_size(right, item);
2619                 btrfs_set_item_offset(right, item, push_space);
2620         }
2621
2622         btrfs_mark_buffer_dirty(left);
2623         if (right_nritems)
2624                 btrfs_mark_buffer_dirty(right);
2625         else
2626                 clean_tree_block(trans, root, right);
2627
2628         btrfs_item_key(right, &disk_key, 0);
2629         wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2630         if (wret)
2631                 ret = wret;
2632
2633         /* then fixup the leaf pointer in the path */
2634         if (path->slots[0] < push_items) {
2635                 path->slots[0] += old_left_nritems;
2636                 btrfs_tree_unlock(path->nodes[0]);
2637                 free_extent_buffer(path->nodes[0]);
2638                 path->nodes[0] = left;
2639                 path->slots[1] -= 1;
2640         } else {
2641                 btrfs_tree_unlock(left);
2642                 free_extent_buffer(left);
2643                 path->slots[0] -= push_items;
2644         }
2645         BUG_ON(path->slots[0] < 0);
2646         return ret;
2647 out:
2648         btrfs_tree_unlock(left);
2649         free_extent_buffer(left);
2650         return ret;
2651 }
2652
2653 /*
2654  * push some data in the path leaf to the left, trying to free up at
2655  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2656  *
2657  * max_slot can put a limit on how far into the leaf we'll push items.  The
2658  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us push all the
2659  * items
2660  */
2661 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2662                           *root, struct btrfs_path *path, int min_data_size,
2663                           int data_size, int empty, u32 max_slot)
2664 {
2665         struct extent_buffer *right = path->nodes[0];
2666         struct extent_buffer *left;
2667         int slot;
2668         int free_space;
2669         u32 right_nritems;
2670         int ret = 0;
2671
2672         slot = path->slots[1];
2673         if (slot == 0)
2674                 return 1;
2675         if (!path->nodes[1])
2676                 return 1;
2677
2678         right_nritems = btrfs_header_nritems(right);
2679         if (right_nritems == 0)
2680                 return 1;
2681
2682         btrfs_assert_tree_locked(path->nodes[1]);
2683
2684         left = read_node_slot(root, path->nodes[1], slot - 1);
2685         if (left == NULL)
2686                 return 1;
2687
2688         btrfs_tree_lock(left);
2689         btrfs_set_lock_blocking(left);
2690
2691         free_space = btrfs_leaf_free_space(root, left);
2692         if (free_space < data_size) {
2693                 ret = 1;
2694                 goto out;
2695         }
2696
2697         /* cow and double check */
2698         ret = btrfs_cow_block(trans, root, left,
2699                               path->nodes[1], slot - 1, &left);
2700         if (ret) {
2701                 /* we hit -ENOSPC, but it isn't fatal here */
2702                 ret = 1;
2703                 goto out;
2704         }
2705
2706         free_space = btrfs_leaf_free_space(root, left);
2707         if (free_space < data_size) {
2708                 ret = 1;
2709                 goto out;
2710         }
2711
2712         return __push_leaf_left(trans, root, path, min_data_size,
2713                                empty, left, free_space, right_nritems,
2714                                max_slot);
2715 out:
2716         btrfs_tree_unlock(left);
2717         free_extent_buffer(left);
2718         return ret;
2719 }
2720
2721 /*
2722  * split the path's leaf in two, making sure there is at least data_size
2723  * available for the resulting leaf level of the path.
2724  *
2725  * returns 0 if all went well and < 0 on failure.
2726  */
2727 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2728                                struct btrfs_root *root,
2729                                struct btrfs_path *path,
2730                                struct extent_buffer *l,
2731                                struct extent_buffer *right,
2732                                int slot, int mid, int nritems)
2733 {
2734         int data_copy_size;
2735         int rt_data_off;
2736         int i;
2737         int ret = 0;
2738         int wret;
2739         struct btrfs_disk_key disk_key;
2740
2741         nritems = nritems - mid;
2742         btrfs_set_header_nritems(right, nritems);
2743         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2744
2745         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2746                            btrfs_item_nr_offset(mid),
2747                            nritems * sizeof(struct btrfs_item));
2748
2749         copy_extent_buffer(right, l,
2750                      btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2751                      data_copy_size, btrfs_leaf_data(l) +
2752                      leaf_data_end(root, l), data_copy_size);
2753
2754         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2755                       btrfs_item_end_nr(l, mid);
2756
2757         for (i = 0; i < nritems; i++) {
2758                 struct btrfs_item *item = btrfs_item_nr(right, i);
2759                 u32 ioff;
2760
2761                 ioff = btrfs_item_offset(right, item);
2762                 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2763         }
2764
2765         btrfs_set_header_nritems(l, mid);
2766         ret = 0;
2767         btrfs_item_key(right, &disk_key, 0);
2768         wret = insert_ptr(trans, root, path, &disk_key, right->start,
2769                           path->slots[1] + 1, 1);
2770         if (wret)
2771                 ret = wret;
2772
2773         btrfs_mark_buffer_dirty(right);
2774         btrfs_mark_buffer_dirty(l);
2775         BUG_ON(path->slots[0] != slot);
2776
2777         if (mid <= slot) {
2778                 btrfs_tree_unlock(path->nodes[0]);
2779                 free_extent_buffer(path->nodes[0]);
2780                 path->nodes[0] = right;
2781                 path->slots[0] -= mid;
2782                 path->slots[1] += 1;
2783         } else {
2784                 btrfs_tree_unlock(right);
2785                 free_extent_buffer(right);
2786         }
2787
2788         BUG_ON(path->slots[0] < 0);
2789
2790         return ret;
2791 }
2792
2793 /*
2794  * double splits happen when we need to insert a big item in the middle
2795  * of a leaf.  A double split can leave us with 3 mostly empty leaves:
2796  * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
2797  *          A                 B                 C
2798  *
2799  * We avoid this by trying to push the items on either side of our target
2800  * into the adjacent leaves.  If all goes well we can avoid the double split
2801  * completely.
2802  */
2803 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
2804                                           struct btrfs_root *root,
2805                                           struct btrfs_path *path,
2806                                           int data_size)
2807 {
2808         int ret;
2809         int progress = 0;
2810         int slot;
2811         u32 nritems;
2812
2813         slot = path->slots[0];
2814
2815         /*
2816          * try to push all the items after our slot into the
2817          * right leaf
2818          */
2819         ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
2820         if (ret < 0)
2821                 return ret;
2822
2823         if (ret == 0)
2824                 progress++;
2825
2826         nritems = btrfs_header_nritems(path->nodes[0]);
2827         /*
2828          * our goal is to get our slot at the start or end of a leaf.  If
2829          * we've done so we're done
2830          */
2831         if (path->slots[0] == 0 || path->slots[0] == nritems)
2832                 return 0;
2833
2834         if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
2835                 return 0;
2836
2837         /* try to push all the items before our slot into the next leaf */
2838         slot = path->slots[0];
2839         ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
2840         if (ret < 0)
2841                 return ret;
2842
2843         if (ret == 0)
2844                 progress++;
2845
2846         if (progress)
2847                 return 0;
2848         return 1;
2849 }
2850
2851 /*
2852  * split the path's leaf in two, making sure there is at least data_size
2853  * available for the resulting leaf level of the path.
2854  *
2855  * returns 0 if all went well and < 0 on failure.
2856  */
2857 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2858                                struct btrfs_root *root,
2859                                struct btrfs_key *ins_key,
2860                                struct btrfs_path *path, int data_size,
2861                                int extend)
2862 {
2863         struct btrfs_disk_key disk_key;
2864         struct extent_buffer *l;
2865         u32 nritems;
2866         int mid;
2867         int slot;
2868         struct extent_buffer *right;
2869         int ret = 0;
2870         int wret;
2871         int split;
2872         int num_doubles = 0;
2873         int tried_avoid_double = 0;
2874
2875         l = path->nodes[0];
2876         slot = path->slots[0];
2877         if (extend && data_size + btrfs_item_size_nr(l, slot) +
2878             sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2879                 return -EOVERFLOW;
2880
2881         /* first try to make some room by pushing left and right */
2882         if (data_size) {
2883                 wret = push_leaf_right(trans, root, path, data_size,
2884                                        data_size, 0, 0);
2885                 if (wret < 0)
2886                         return wret;
2887                 if (wret) {
2888                         wret = push_leaf_left(trans, root, path, data_size,
2889                                               data_size, 0, (u32)-1);
2890                         if (wret < 0)
2891                                 return wret;
2892                 }
2893                 l = path->nodes[0];
2894
2895                 /* did the pushes work? */
2896                 if (btrfs_leaf_free_space(root, l) >= data_size)
2897                         return 0;
2898         }
2899
2900         if (!path->nodes[1]) {
2901                 ret = insert_new_root(trans, root, path, 1);
2902                 if (ret)
2903                         return ret;
2904         }
2905 again:
2906         split = 1;
2907         l = path->nodes[0];
2908         slot = path->slots[0];
2909         nritems = btrfs_header_nritems(l);
2910         mid = (nritems + 1) / 2;
2911
2912         if (mid <= slot) {
2913                 if (nritems == 1 ||
2914                     leaf_space_used(l, mid, nritems - mid) + data_size >
2915                         BTRFS_LEAF_DATA_SIZE(root)) {
2916                         if (slot >= nritems) {
2917                                 split = 0;
2918                         } else {
2919                                 mid = slot;
2920                                 if (mid != nritems &&
2921                                     leaf_space_used(l, mid, nritems - mid) +
2922                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2923                                         if (data_size && !tried_avoid_double)
2924                                                 goto push_for_double;
2925                                         split = 2;
2926                                 }
2927                         }
2928                 }
2929         } else {
2930                 if (leaf_space_used(l, 0, mid) + data_size >
2931                         BTRFS_LEAF_DATA_SIZE(root)) {
2932                         if (!extend && data_size && slot == 0) {
2933                                 split = 0;
2934                         } else if ((extend || !data_size) && slot == 0) {
2935                                 mid = 1;
2936                         } else {
2937                                 mid = slot;
2938                                 if (mid != nritems &&
2939                                     leaf_space_used(l, mid, nritems - mid) +
2940                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2941                                         if (data_size && !tried_avoid_double)
2942                                                 goto push_for_double;
2943                                         split = 2 ;
2944                                 }
2945                         }
2946                 }
2947         }
2948
2949         if (split == 0)
2950                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2951         else
2952                 btrfs_item_key(l, &disk_key, mid);
2953
2954         right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2955                                         root->root_key.objectid,
2956                                         &disk_key, 0, l->start, 0);
2957         if (IS_ERR(right))
2958                 return PTR_ERR(right);
2959
2960         root_add_used(root, root->leafsize);
2961
2962         memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2963         btrfs_set_header_bytenr(right, right->start);
2964         btrfs_set_header_generation(right, trans->transid);
2965         btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2966         btrfs_set_header_owner(right, root->root_key.objectid);
2967         btrfs_set_header_level(right, 0);
2968         write_extent_buffer(right, root->fs_info->fsid,
2969                             (unsigned long)btrfs_header_fsid(right),
2970                             BTRFS_FSID_SIZE);
2971
2972         write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2973                             (unsigned long)btrfs_header_chunk_tree_uuid(right),
2974                             BTRFS_UUID_SIZE);
2975
2976         if (split == 0) {
2977                 if (mid <= slot) {
2978                         btrfs_set_header_nritems(right, 0);
2979                         wret = insert_ptr(trans, root, path,
2980                                           &disk_key, right->start,
2981                                           path->slots[1] + 1, 1);
2982                         if (wret)
2983                                 ret = wret;
2984
2985                         btrfs_tree_unlock(path->nodes[0]);
2986                         free_extent_buffer(path->nodes[0]);
2987                         path->nodes[0] = right;
2988                         path->slots[0] = 0;
2989                         path->slots[1] += 1;
2990                 } else {
2991                         btrfs_set_header_nritems(right, 0);
2992                         wret = insert_ptr(trans, root, path,
2993                                           &disk_key,
2994                                           right->start,
2995                                           path->slots[1], 1);
2996                         if (wret)
2997                                 ret = wret;
2998                         btrfs_tree_unlock(path->nodes[0]);
2999                         free_extent_buffer(path->nodes[0]);
3000                         path->nodes[0] = right;
3001                         path->slots[0] = 0;
3002                         if (path->slots[1] == 0) {
3003                                 wret = fixup_low_keys(trans, root,
3004                                                 path, &disk_key, 1);
3005                                 if (wret)
3006                                         ret = wret;
3007                         }
3008                 }
3009                 btrfs_mark_buffer_dirty(right);
3010                 return ret;
3011         }
3012
3013         ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
3014         BUG_ON(ret);
3015
3016         if (split == 2) {
3017                 BUG_ON(num_doubles != 0);
3018                 num_doubles++;
3019                 goto again;
3020         }
3021
3022         return ret;
3023
3024 push_for_double:
3025         push_for_double_split(trans, root, path, data_size);
3026         tried_avoid_double = 1;
3027         if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3028                 return 0;
3029         goto again;
3030 }
3031
3032 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3033                                          struct btrfs_root *root,
3034                                          struct btrfs_path *path, int ins_len)
3035 {
3036         struct btrfs_key key;
3037         struct extent_buffer *leaf;
3038         struct btrfs_file_extent_item *fi;
3039         u64 extent_len = 0;
3040         u32 item_size;
3041         int ret;
3042
3043         leaf = path->nodes[0];
3044         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3045
3046         BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3047                key.type != BTRFS_EXTENT_CSUM_KEY);
3048
3049         if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3050                 return 0;
3051
3052         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3053         if (key.type == BTRFS_EXTENT_DATA_KEY) {
3054                 fi = btrfs_item_ptr(leaf, path->slots[0],
3055                                     struct btrfs_file_extent_item);
3056                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3057         }
3058         btrfs_release_path(path);
3059
3060         path->keep_locks = 1;
3061         path->search_for_split = 1;
3062         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3063         path->search_for_split = 0;
3064         if (ret < 0)
3065                 goto err;
3066
3067         ret = -EAGAIN;
3068         leaf = path->nodes[0];
3069         /* if our item isn't there or got smaller, return now */
3070         if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3071                 goto err;
3072
3073         /* the leaf has  changed, it now has room.  return now */
3074         if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3075                 goto err;
3076
3077         if (key.type == BTRFS_EXTENT_DATA_KEY) {
3078              &nbs