projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: check NULL or not
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / btrfs / ctree.c
CommitLineData
6cbd5570 1/*
d352ac68 2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
6cbd5570
CM		 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
a6b6e75e 19#include <linux/sched.h>
5a0e3ad6 20#include <linux/slab.h>
eb60ceac
CM		 21#include "ctree.h"
22#include "disk-io.h"
7f5c1516 23#include "transaction.h"
5f39d397 24#include "print-tree.h"
925baedd 25#include "locking.h"
9a8dd150 26
e089f05c
CM		 27static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
28 *root, struct btrfs_path *path, int level);
29static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
d4dbff95 30 *root, struct btrfs_key *ins_key,
cc0c5538 31 struct btrfs_path *path, int data_size, int extend);
5f39d397
CM		 32static int push_node_left(struct btrfs_trans_handle *trans,
33 struct btrfs_root *root, struct extent_buffer *dst,
971a1f66 34 struct extent_buffer *src, int empty);
5f39d397
CM		 35static 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);
e089f05c
CM		 39static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
40 struct btrfs_path *path, int level, int slot);
ad48fd75
YZ		 41static int setup_items_for_insert(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root, struct btrfs_path *path,
43 struct btrfs_key *cpu_key, u32 *data_size,
44 u32 total_data, u32 total_size, int nr);
45
d97e63b6 46
df24a2b9 47struct btrfs_path *btrfs_alloc_path(void)
2c90e5d6 48{
df24a2b9 49 struct btrfs_path *path;
e00f7308
JM		 50 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
51 if (path)
2cc58cf2 52 path->reada = 1;
df24a2b9 53 return path;
2c90e5d6
CM		 54}
55
b4ce94de
CM		 56/*
57 * set all locked nodes in the path to blocking locks. This should
58 * be done before scheduling
59 */
60noinline void btrfs_set_path_blocking(struct btrfs_path *p)
61{
62 int i;
63 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
64 if (p->nodes[i] && p->locks[i])
65 btrfs_set_lock_blocking(p->nodes[i]);
66 }
67}
68
69/*
70 * reset all the locked nodes in the patch to spinning locks.
4008c04a
CM		 71 *
72 * held is used to keep lockdep happy, when lockdep is enabled
73 * we set held to a blocking lock before we go around and
74 * retake all the spinlocks in the path. You can safely use NULL
75 * for held
b4ce94de 76 */
4008c04a
CM		 77noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
78 struct extent_buffer *held)
b4ce94de
CM		 79{
80 int i;
4008c04a
CM		 81
82#ifdef CONFIG_DEBUG_LOCK_ALLOC
83 /* lockdep really cares that we take all of these spinlocks
84 * in the right order. If any of the locks in the path are not
85 * currently blocking, it is going to complain. So, make really
86 * really sure by forcing the path to blocking before we clear
87 * the path blocking.
88 */
89 if (held)
90 btrfs_set_lock_blocking(held);
91 btrfs_set_path_blocking(p);
92#endif
93
94 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
b4ce94de
CM		 95 if (p->nodes[i] && p->locks[i])
96 btrfs_clear_lock_blocking(p->nodes[i]);
97 }
4008c04a
CM		 98
99#ifdef CONFIG_DEBUG_LOCK_ALLOC
100 if (held)
101 btrfs_clear_lock_blocking(held);
102#endif
b4ce94de
CM		 103}
104
d352ac68 105/* this also releases the path */
df24a2b9 106void btrfs_free_path(struct btrfs_path *p)
be0e5c09 107{
ff175d57
JJ		 108 if (!p)
109 return;
df24a2b9
CM		 110 btrfs_release_path(NULL, p);
111 kmem_cache_free(btrfs_path_cachep, p);
be0e5c09
CM		 112}
113
d352ac68
CM		 114/*
115 * path release drops references on the extent buffers in the path
116 * and it drops any locks held by this path
117 *
118 * It is safe to call this on paths that no locks or extent buffers held.
119 */
d397712b 120noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
eb60ceac
CM		 121{
122 int i;
a2135011 123
234b63a0 124 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3f157a2f 125 p->slots[i] = 0;
eb60ceac 126 if (!p->nodes[i])
925baedd
CM		 127 continue;
128 if (p->locks[i]) {
129 btrfs_tree_unlock(p->nodes[i]);
130 p->locks[i] = 0;
131 }
5f39d397 132 free_extent_buffer(p->nodes[i]);
3f157a2f 133 p->nodes[i] = NULL;
eb60ceac
CM		 134 }
135}
136
d352ac68
CM		 137/*
138 * safely gets a reference on the root node of a tree. A lock
139 * is not taken, so a concurrent writer may put a different node
140 * at the root of the tree. See btrfs_lock_root_node for the
141 * looping required.
142 *
143 * The extent buffer returned by this has a reference taken, so
144 * it won't disappear. It may stop being the root of the tree
145 * at any time because there are no locks held.
146 */
925baedd
CM		 147struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
148{
149 struct extent_buffer *eb;
150 spin_lock(&root->node_lock);
151 eb = root->node;
152 extent_buffer_get(eb);
153 spin_unlock(&root->node_lock);
154 return eb;
155}
156
d352ac68
CM		 157/* loop around taking references on and locking the root node of the
158 * tree until you end up with a lock on the root. A locked buffer
159 * is returned, with a reference held.
160 */
925baedd
CM		 161struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
162{
163 struct extent_buffer *eb;
164
d397712b 165 while (1) {
925baedd
CM		 166 eb = btrfs_root_node(root);
167 btrfs_tree_lock(eb);
168
169 spin_lock(&root->node_lock);
170 if (eb == root->node) {
171 spin_unlock(&root->node_lock);
172 break;
173 }
174 spin_unlock(&root->node_lock);
175
176 btrfs_tree_unlock(eb);
177 free_extent_buffer(eb);
178 }
179 return eb;
180}
181
d352ac68
CM		 182/* cowonly root (everything not a reference counted cow subvolume), just get
183 * put onto a simple dirty list. transaction.c walks this to make sure they
184 * get properly updated on disk.
185 */
0b86a832
CM		 186static void add_root_to_dirty_list(struct btrfs_root *root)
187{
188 if (root->track_dirty && list_empty(&root->dirty_list)) {
189 list_add(&root->dirty_list,
190 &root->fs_info->dirty_cowonly_roots);
191 }
192}
193
d352ac68
CM		 194/*
195 * used by snapshot creation to make a copy of a root for a tree with
196 * a given objectid. The buffer with the new root node is returned in
197 * cow_ret, and this func returns zero on success or a negative error code.
198 */
be20aa9d
CM		 199int btrfs_copy_root(struct btrfs_trans_handle *trans,
200 struct btrfs_root *root,
201 struct extent_buffer *buf,
202 struct extent_buffer **cow_ret, u64 new_root_objectid)
203{
204 struct extent_buffer *cow;
be20aa9d
CM		 205 int ret = 0;
206 int level;
5d4f98a2 207 struct btrfs_disk_key disk_key;
be20aa9d
CM		 208
209 WARN_ON(root->ref_cows && trans->transid !=
210 root->fs_info->running_transaction->transid);
211 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
212
213 level = btrfs_header_level(buf);
5d4f98a2
YZ		 214 if (level == 0)
215 btrfs_item_key(buf, &disk_key, 0);
216 else
217 btrfs_node_key(buf, &disk_key, 0);
31840ae1 218
5d4f98a2
YZ		 219 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
220 new_root_objectid, &disk_key, level,
221 buf->start, 0);
222 if (IS_ERR(cow))
be20aa9d
CM		 223 return PTR_ERR(cow);
224
225 copy_extent_buffer(cow, buf, 0, 0, cow->len);
226 btrfs_set_header_bytenr(cow, cow->start);
227 btrfs_set_header_generation(cow, trans->transid);
5d4f98a2
YZ		 228 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
229 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
230 BTRFS_HEADER_FLAG_RELOC);
231 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
232 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
233 else
234 btrfs_set_header_owner(cow, new_root_objectid);
be20aa9d 235
2b82032c
YZ		 236 write_extent_buffer(cow, root->fs_info->fsid,
237 (unsigned long)btrfs_header_fsid(cow),
238 BTRFS_FSID_SIZE);
239
be20aa9d 240 WARN_ON(btrfs_header_generation(buf) > trans->transid);
5d4f98a2
YZ		 241 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
242 ret = btrfs_inc_ref(trans, root, cow, 1);
243 else
244 ret = btrfs_inc_ref(trans, root, cow, 0);
4aec2b52 245
be20aa9d
CM		 246 if (ret)
247 return ret;
248
249 btrfs_mark_buffer_dirty(cow);
250 *cow_ret = cow;
251 return 0;
252}
253
5d4f98a2
YZ		 254/*
255 * check if the tree block can be shared by multiple trees
256 */
257int btrfs_block_can_be_shared(struct btrfs_root *root,
258 struct extent_buffer *buf)
259{
260 /*
261 * Tree blocks not in refernece counted trees and tree roots
262 * are never shared. If a block was allocated after the last
263 * snapshot and the block was not allocated by tree relocation,
264 * we know the block is not shared.
265 */
266 if (root->ref_cows &&
267 buf != root->node && buf != root->commit_root &&
268 (btrfs_header_generation(buf) <=
269 btrfs_root_last_snapshot(&root->root_item) ||
270 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
271 return 1;
272#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
273 if (root->ref_cows &&
274 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
275 return 1;
276#endif
277 return 0;
278}
279
280static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
281 struct btrfs_root *root,
282 struct extent_buffer *buf,
f0486c68
YZ		 283 struct extent_buffer *cow,
284 int *last_ref)
5d4f98a2
YZ		 285{
286 u64 refs;
287 u64 owner;
288 u64 flags;
289 u64 new_flags = 0;
290 int ret;
291
292 /*
293 * Backrefs update rules:
294 *
295 * Always use full backrefs for extent pointers in tree block
296 * allocated by tree relocation.
297 *
298 * If a shared tree block is no longer referenced by its owner
299 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
300 * use full backrefs for extent pointers in tree block.
301 *
302 * If a tree block is been relocating
303 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
304 * use full backrefs for extent pointers in tree block.
305 * The reason for this is some operations (such as drop tree)
306 * are only allowed for blocks use full backrefs.
307 */
308
309 if (btrfs_block_can_be_shared(root, buf)) {
310 ret = btrfs_lookup_extent_info(trans, root, buf->start,
311 buf->len, &refs, &flags);
312 BUG_ON(ret);
313 BUG_ON(refs == 0);
314 } else {
315 refs = 1;
316 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
317 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
318 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
319 else
320 flags = 0;
321 }
322
323 owner = btrfs_header_owner(buf);
324 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
325 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
326
327 if (refs > 1) {
328 if ((owner == root->root_key.objectid ||
329 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
330 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
331 ret = btrfs_inc_ref(trans, root, buf, 1);
332 BUG_ON(ret);
333
334 if (root->root_key.objectid ==
335 BTRFS_TREE_RELOC_OBJECTID) {
336 ret = btrfs_dec_ref(trans, root, buf, 0);
337 BUG_ON(ret);
338 ret = btrfs_inc_ref(trans, root, cow, 1);
339 BUG_ON(ret);
340 }
341 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
342 } else {
343
344 if (root->root_key.objectid ==
345 BTRFS_TREE_RELOC_OBJECTID)
346 ret = btrfs_inc_ref(trans, root, cow, 1);
347 else
348 ret = btrfs_inc_ref(trans, root, cow, 0);
349 BUG_ON(ret);
350 }
351 if (new_flags != 0) {
352 ret = btrfs_set_disk_extent_flags(trans, root,
353 buf->start,
354 buf->len,
355 new_flags, 0);
356 BUG_ON(ret);
357 }
358 } else {
359 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
360 if (root->root_key.objectid ==
361 BTRFS_TREE_RELOC_OBJECTID)
362 ret = btrfs_inc_ref(trans, root, cow, 1);
363 else
364 ret = btrfs_inc_ref(trans, root, cow, 0);
365 BUG_ON(ret);
366 ret = btrfs_dec_ref(trans, root, buf, 1);
367 BUG_ON(ret);
368 }
369 clean_tree_block(trans, root, buf);
f0486c68 370 *last_ref = 1;
5d4f98a2
YZ		 371 }
372 return 0;
373}
374
d352ac68 375/*
d397712b
CM		 376 * does the dirty work in cow of a single block. The parent block (if
377 * supplied) is updated to point to the new cow copy. The new buffer is marked
378 * dirty and returned locked. If you modify the block it needs to be marked
379 * dirty again.
d352ac68
CM		 380 *
381 * search_start -- an allocation hint for the new block
382 *
d397712b
CM		 383 * empty_size -- a hint that you plan on doing more cow. This is the size in
384 * bytes the allocator should try to find free next to the block it returns.
385 * This is just a hint and may be ignored by the allocator.
d352ac68 386 */
d397712b 387static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
5f39d397
CM		 388 struct btrfs_root *root,
389 struct extent_buffer *buf,
390 struct extent_buffer *parent, int parent_slot,
391 struct extent_buffer **cow_ret,
9fa8cfe7 392 u64 search_start, u64 empty_size)
02217ed2 393{
5d4f98a2 394 struct btrfs_disk_key disk_key;
5f39d397 395 struct extent_buffer *cow;
7bb86316 396 int level;
f0486c68 397 int last_ref = 0;
925baedd 398 int unlock_orig = 0;
5d4f98a2 399 u64 parent_start;
7bb86316 400
925baedd
CM		 401 if (*cow_ret == buf)
402 unlock_orig = 1;
403
b9447ef8 404 btrfs_assert_tree_locked(buf);
925baedd 405
7bb86316
CM		 406 WARN_ON(root->ref_cows && trans->transid !=
407 root->fs_info->running_transaction->transid);
6702ed49 408 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
5f39d397 409
7bb86316 410 level = btrfs_header_level(buf);
31840ae1 411
5d4f98a2
YZ		 412 if (level == 0)
413 btrfs_item_key(buf, &disk_key, 0);
414 else
415 btrfs_node_key(buf, &disk_key, 0);
416
417 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
418 if (parent)
419 parent_start = parent->start;
420 else
421 parent_start = 0;
422 } else
423 parent_start = 0;
424
425 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
426 root->root_key.objectid, &disk_key,
427 level, search_start, empty_size);
54aa1f4d
CM		 428 if (IS_ERR(cow))
429 return PTR_ERR(cow);
6702ed49 430
b4ce94de
CM		 431 /* cow is set to blocking by btrfs_init_new_buffer */
432
5f39d397 433 copy_extent_buffer(cow, buf, 0, 0, cow->len);
db94535d 434 btrfs_set_header_bytenr(cow, cow->start);
5f39d397 435 btrfs_set_header_generation(cow, trans->transid);
5d4f98a2
YZ		 436 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
437 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
438 BTRFS_HEADER_FLAG_RELOC);
439 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
440 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
441 else
442 btrfs_set_header_owner(cow, root->root_key.objectid);
6702ed49 443
2b82032c
YZ		 444 write_extent_buffer(cow, root->fs_info->fsid,
445 (unsigned long)btrfs_header_fsid(cow),
446 BTRFS_FSID_SIZE);
447
f0486c68 448 update_ref_for_cow(trans, root, buf, cow, &last_ref);
1a40e23b 449
3fd0a558
YZ		 450 if (root->ref_cows)
451 btrfs_reloc_cow_block(trans, root, buf, cow);
452
02217ed2 453 if (buf == root->node) {
925baedd 454 WARN_ON(parent && parent != buf);
5d4f98a2
YZ		 455 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
456 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
457 parent_start = buf->start;
458 else
459 parent_start = 0;
925baedd
CM		 460
461 spin_lock(&root->node_lock);
02217ed2 462 root->node = cow;
5f39d397 463 extent_buffer_get(cow);
925baedd
CM		 464 spin_unlock(&root->node_lock);
465
f0486c68
YZ		 466 btrfs_free_tree_block(trans, root, buf, parent_start,
467 last_ref);
5f39d397 468 free_extent_buffer(buf);
0b86a832 469 add_root_to_dirty_list(root);
02217ed2 470 } else {
5d4f98a2
YZ		 471 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
472 parent_start = parent->start;
473 else
474 parent_start = 0;
475
476 WARN_ON(trans->transid != btrfs_header_generation(parent));
5f39d397 477 btrfs_set_node_blockptr(parent, parent_slot,
db94535d 478 cow->start);
74493f7a
CM		 479 btrfs_set_node_ptr_generation(parent, parent_slot,
480 trans->transid);
d6025579 481 btrfs_mark_buffer_dirty(parent);
f0486c68
YZ		 482 btrfs_free_tree_block(trans, root, buf, parent_start,
483 last_ref);
02217ed2 484 }
925baedd
CM		 485 if (unlock_orig)
486 btrfs_tree_unlock(buf);
5f39d397 487 free_extent_buffer(buf);
ccd467d6 488 btrfs_mark_buffer_dirty(cow);
2c90e5d6 489 *cow_ret = cow;
02217ed2
CM		 490 return 0;
491}
492
5d4f98a2
YZ		 493static inline int should_cow_block(struct btrfs_trans_handle *trans,
494 struct btrfs_root *root,
495 struct extent_buffer *buf)
496{
497 if (btrfs_header_generation(buf) == trans->transid &&
498 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
499 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
500 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
501 return 0;
502 return 1;
503}
504
d352ac68
CM		 505/*
506 * cows a single block, see __btrfs_cow_block for the real work.
507 * This version of it has extra checks so that a block isn't cow'd more than
508 * once per transaction, as long as it hasn't been written yet
509 */
d397712b 510noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
5f39d397
CM		 511 struct btrfs_root *root, struct extent_buffer *buf,
512 struct extent_buffer *parent, int parent_slot,
9fa8cfe7 513 struct extent_buffer **cow_ret)
6702ed49
CM		 514{
515 u64 search_start;
f510cfec 516 int ret;
dc17ff8f 517
6702ed49 518 if (trans->transaction != root->fs_info->running_transaction) {
d397712b
CM		 519 printk(KERN_CRIT "trans %llu running %llu\n",
520 (unsigned long long)trans->transid,
521 (unsigned long long)
6702ed49
CM		 522 root->fs_info->running_transaction->transid);
523 WARN_ON(1);
524 }
525 if (trans->transid != root->fs_info->generation) {
d397712b
CM		 526 printk(KERN_CRIT "trans %llu running %llu\n",
527 (unsigned long long)trans->transid,
528 (unsigned long long)root->fs_info->generation);
6702ed49
CM		 529 WARN_ON(1);
530 }
dc17ff8f 531
5d4f98a2 532 if (!should_cow_block(trans, root, buf)) {
6702ed49
CM		 533 *cow_ret = buf;
534 return 0;
535 }
c487685d 536
0b86a832 537 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
b4ce94de
CM		 538
539 if (parent)
540 btrfs_set_lock_blocking(parent);
541 btrfs_set_lock_blocking(buf);
542
f510cfec 543 ret = __btrfs_cow_block(trans, root, buf, parent,
9fa8cfe7 544 parent_slot, cow_ret, search_start, 0);
f510cfec 545 return ret;
6702ed49
CM		 546}
547
d352ac68
CM		 548/*
549 * helper function for defrag to decide if two blocks pointed to by a
550 * node are actually close by
551 */
6b80053d 552static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
6702ed49 553{
6b80053d 554 if (blocknr < other && other - (blocknr + blocksize) < 32768)
6702ed49 555 return 1;
6b80053d 556 if (blocknr > other && blocknr - (other + blocksize) < 32768)
6702ed49
CM		 557 return 1;
558 return 0;
559}
560
081e9573
CM		 561/*
562 * compare two keys in a memcmp fashion
563 */
564static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
565{
566 struct btrfs_key k1;
567
568 btrfs_disk_key_to_cpu(&k1, disk);
569
20736aba 570 return btrfs_comp_cpu_keys(&k1, k2);
081e9573
CM		 571}
572
f3465ca4
JB		 573/*
574 * same as comp_keys only with two btrfs_key's
575 */
5d4f98a2 576int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
f3465ca4
JB		 577{
578 if (k1->objectid > k2->objectid)
579 return 1;
580 if (k1->objectid < k2->objectid)
581 return -1;
582 if (k1->type > k2->type)
583 return 1;
584 if (k1->type < k2->type)
585 return -1;
586 if (k1->offset > k2->offset)
587 return 1;
588 if (k1->offset < k2->offset)
589 return -1;
590 return 0;
591}
081e9573 592
d352ac68
CM		 593/*
594 * this is used by the defrag code to go through all the
595 * leaves pointed to by a node and reallocate them so that
596 * disk order is close to key order
597 */
6702ed49 598int btrfs_realloc_node(struct btrfs_trans_handle *trans,
5f39d397 599 struct btrfs_root *root, struct extent_buffer *parent,
a6b6e75e
CM		 600 int start_slot, int cache_only, u64 *last_ret,
601 struct btrfs_key *progress)
6702ed49 602{
6b80053d 603 struct extent_buffer *cur;
6702ed49 604 u64 blocknr;
ca7a79ad 605 u64 gen;
e9d0b13b
CM		 606 u64 search_start = *last_ret;
607 u64 last_block = 0;
6702ed49
CM		 608 u64 other;
609 u32 parent_nritems;
6702ed49
CM		 610 int end_slot;
611 int i;
612 int err = 0;
f2183bde 613 int parent_level;
6b80053d
CM		 614 int uptodate;
615 u32 blocksize;
081e9573
CM		 616 int progress_passed = 0;
617 struct btrfs_disk_key disk_key;
6702ed49 618
5708b959
CM		 619 parent_level = btrfs_header_level(parent);
620 if (cache_only && parent_level != 1)
621 return 0;
622
d397712b 623 if (trans->transaction != root->fs_info->running_transaction)
6702ed49 624 WARN_ON(1);
d397712b 625 if (trans->transid != root->fs_info->generation)
6702ed49 626 WARN_ON(1);
86479a04 627
6b80053d 628 parent_nritems = btrfs_header_nritems(parent);
6b80053d 629 blocksize = btrfs_level_size(root, parent_level - 1);
6702ed49
CM		 630 end_slot = parent_nritems;
631
632 if (parent_nritems == 1)
633 return 0;
634
b4ce94de
CM		 635 btrfs_set_lock_blocking(parent);
636
6702ed49
CM		 637 for (i = start_slot; i < end_slot; i++) {
638 int close = 1;
a6b6e75e 639
5708b959
CM		 640 if (!parent->map_token) {
641 map_extent_buffer(parent,
642 btrfs_node_key_ptr_offset(i),
643 sizeof(struct btrfs_key_ptr),
644 &parent->map_token, &parent->kaddr,
645 &parent->map_start, &parent->map_len,
646 KM_USER1);
647 }
081e9573
CM		 648 btrfs_node_key(parent, &disk_key, i);
649 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
650 continue;
651
652 progress_passed = 1;
6b80053d 653 blocknr = btrfs_node_blockptr(parent, i);
ca7a79ad 654 gen = btrfs_node_ptr_generation(parent, i);
e9d0b13b
CM		 655 if (last_block == 0)
656 last_block = blocknr;
5708b959 657
6702ed49 658 if (i > 0) {
6b80053d
CM		 659 other = btrfs_node_blockptr(parent, i - 1);
660 close = close_blocks(blocknr, other, blocksize);
6702ed49 661 }
0ef3e66b 662 if (!close && i < end_slot - 2) {
6b80053d
CM		 663 other = btrfs_node_blockptr(parent, i + 1);
664 close = close_blocks(blocknr, other, blocksize);
6702ed49 665 }
e9d0b13b
CM		 666 if (close) {
667 last_block = blocknr;
6702ed49 668 continue;
e9d0b13b 669 }
5708b959
CM		 670 if (parent->map_token) {
671 unmap_extent_buffer(parent, parent->map_token,
672 KM_USER1);
673 parent->map_token = NULL;
674 }
6702ed49 675
6b80053d
CM		 676 cur = btrfs_find_tree_block(root, blocknr, blocksize);
677 if (cur)
1259ab75 678 uptodate = btrfs_buffer_uptodate(cur, gen);
6b80053d
CM		 679 else
680 uptodate = 0;
5708b959 681 if (!cur || !uptodate) {
6702ed49 682 if (cache_only) {
6b80053d 683 free_extent_buffer(cur);
6702ed49
CM		 684 continue;
685 }
6b80053d
CM		 686 if (!cur) {
687 cur = read_tree_block(root, blocknr,
ca7a79ad 688 blocksize, gen);
6b80053d 689 } else if (!uptodate) {
ca7a79ad 690 btrfs_read_buffer(cur, gen);
f2183bde 691 }
6702ed49 692 }
e9d0b13b 693 if (search_start == 0)
6b80053d 694 search_start = last_block;
e9d0b13b 695
e7a84565 696 btrfs_tree_lock(cur);
b4ce94de 697 btrfs_set_lock_blocking(cur);
6b80053d 698 err = __btrfs_cow_block(trans, root, cur, parent, i,
e7a84565 699 &cur, search_start,
6b80053d 700 min(16 * blocksize,
9fa8cfe7 701 (end_slot - i) * blocksize));
252c38f0 702 if (err) {
e7a84565 703 btrfs_tree_unlock(cur);
6b80053d 704 free_extent_buffer(cur);
6702ed49 705 break;
252c38f0 706 }
e7a84565
CM		 707 search_start = cur->start;
708 last_block = cur->start;
f2183bde 709 *last_ret = search_start;
e7a84565
CM		 710 btrfs_tree_unlock(cur);
711 free_extent_buffer(cur);
6702ed49 712 }
5708b959
CM		 713 if (parent->map_token) {
714 unmap_extent_buffer(parent, parent->map_token,
715 KM_USER1);
716 parent->map_token = NULL;
717 }
6702ed49
CM		 718 return err;
719}
720
74123bd7
CM		 721/*
722 * The leaf data grows from end-to-front in the node.
723 * this returns the address of the start of the last item,
724 * which is the stop of the leaf data stack
725 */
123abc88 726static inline unsigned int leaf_data_end(struct btrfs_root *root,
5f39d397 727 struct extent_buffer *leaf)
be0e5c09 728{
5f39d397 729 u32 nr = btrfs_header_nritems(leaf);
be0e5c09 730 if (nr == 0)
123abc88 731 return BTRFS_LEAF_DATA_SIZE(root);
5f39d397 732 return btrfs_item_offset_nr(leaf, nr - 1);
be0e5c09
CM		 733}
734
d352ac68
CM		 735/*
736 * extra debugging checks to make sure all the items in a key are
737 * well formed and in the proper order
738 */
123abc88
CM		 739static int check_node(struct btrfs_root *root, struct btrfs_path *path,
740 int level)
aa5d6bed 741{
5f39d397
CM		 742 struct extent_buffer *parent = NULL;
743 struct extent_buffer *node = path->nodes[level];
744 struct btrfs_disk_key parent_key;
745 struct btrfs_disk_key node_key;
aa5d6bed 746 int parent_slot;
8d7be552
CM		 747 int slot;
748 struct btrfs_key cpukey;
5f39d397 749 u32 nritems = btrfs_header_nritems(node);
aa5d6bed
CM		 750
751 if (path->nodes[level + 1])
5f39d397 752 parent = path->nodes[level + 1];
a1f39630 753
8d7be552 754 slot = path->slots[level];
7518a238
CM		 755 BUG_ON(nritems == 0);
756 if (parent) {
a1f39630 757 parent_slot = path->slots[level + 1];
5f39d397
CM		 758 btrfs_node_key(parent, &parent_key, parent_slot);
759 btrfs_node_key(node, &node_key, 0);
760 BUG_ON(memcmp(&parent_key, &node_key,
e2fa7227 761 sizeof(struct btrfs_disk_key)));
1d4f8a0c 762 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
db94535d 763 btrfs_header_bytenr(node));
aa5d6bed 764 }
123abc88 765 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
8d7be552 766 if (slot != 0) {
5f39d397
CM		 767 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
768 btrfs_node_key(node, &node_key, slot);
769 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
8d7be552
CM		 770 }
771 if (slot < nritems - 1) {
5f39d397
CM		 772 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
773 btrfs_node_key(node, &node_key, slot);
774 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
aa5d6bed
CM		 775 }
776 return 0;
777}
778
d352ac68
CM		 779/*
780 * extra checking to make sure all the items in a leaf are
781 * well formed and in the proper order
782 */
123abc88
CM		 783static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
784 int level)
aa5d6bed 785{
5f39d397
CM		 786 struct extent_buffer *leaf = path->nodes[level];
787 struct extent_buffer *parent = NULL;
aa5d6bed 788 int parent_slot;
8d7be552 789 struct btrfs_key cpukey;
5f39d397
CM		 790 struct btrfs_disk_key parent_key;
791 struct btrfs_disk_key leaf_key;
792 int slot = path->slots[0];
8d7be552 793
5f39d397 794 u32 nritems = btrfs_header_nritems(leaf);
aa5d6bed
CM		 795
796 if (path->nodes[level + 1])
5f39d397 797 parent = path->nodes[level + 1];
7518a238
CM		 798
799 if (nritems == 0)
800 return 0;
801
802 if (parent) {
a1f39630 803 parent_slot = path->slots[level + 1];
5f39d397
CM		 804 btrfs_node_key(parent, &parent_key, parent_slot);
805 btrfs_item_key(leaf, &leaf_key, 0);
6702ed49 806
5f39d397 807 BUG_ON(memcmp(&parent_key, &leaf_key,
e2fa7227 808 sizeof(struct btrfs_disk_key)));
1d4f8a0c 809 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
db94535d 810 btrfs_header_bytenr(leaf));
5f39d397 811 }
5f39d397
CM		 812 if (slot != 0 && slot < nritems - 1) {
813 btrfs_item_key(leaf, &leaf_key, slot);
814 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
815 if (comp_keys(&leaf_key, &cpukey) <= 0) {
816 btrfs_print_leaf(root, leaf);
d397712b 817 printk(KERN_CRIT "slot %d offset bad key\n", slot);
5f39d397
CM		 818 BUG_ON(1);
819 }
820 if (btrfs_item_offset_nr(leaf, slot - 1) !=
821 btrfs_item_end_nr(leaf, slot)) {
822 btrfs_print_leaf(root, leaf);
d397712b 823 printk(KERN_CRIT "slot %d offset bad\n", slot);
5f39d397
CM		 824 BUG_ON(1);
825 }
8d7be552
CM		 826 }
827 if (slot < nritems - 1) {
5f39d397
CM		 828 btrfs_item_key(leaf, &leaf_key, slot);
829 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
830 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
831 if (btrfs_item_offset_nr(leaf, slot) !=
832 btrfs_item_end_nr(leaf, slot + 1)) {
833 btrfs_print_leaf(root, leaf);
d397712b 834 printk(KERN_CRIT "slot %d offset bad\n", slot);
5f39d397
CM		 835 BUG_ON(1);
836 }
aa5d6bed 837 }
5f39d397
CM		 838 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
839 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
aa5d6bed
CM		 840 return 0;
841}
842
d397712b 843static noinline int check_block(struct btrfs_root *root,
98ed5174 844 struct btrfs_path *path, int level)
aa5d6bed 845{
85d824c4 846 return 0;
aa5d6bed 847 if (level == 0)
123abc88
CM		 848 return check_leaf(root, path, level);
849 return check_node(root, path, level);
aa5d6bed
CM		 850}
851
74123bd7 852/*
5f39d397
CM		 853 * search for key in the extent_buffer. The items start at offset p,
854 * and they are item_size apart. There are 'max' items in p.
855 *
74123bd7
CM		 856 * the slot in the array is returned via slot, and it points to
857 * the place where you would insert key if it is not found in
858 * the array.
859 *
860 * slot may point to max if the key is bigger than all of the keys
861 */
e02119d5
CM		 862static noinline int generic_bin_search(struct extent_buffer *eb,
863 unsigned long p,
864 int item_size, struct btrfs_key *key,
865 int max, int *slot)
be0e5c09
CM		 866{
867 int low = 0;
868 int high = max;
869 int mid;
870 int ret;
479965d6 871 struct btrfs_disk_key *tmp = NULL;
5f39d397
CM		 872 struct btrfs_disk_key unaligned;
873 unsigned long offset;
874 char *map_token = NULL;
875 char *kaddr = NULL;
876 unsigned long map_start = 0;
877 unsigned long map_len = 0;
479965d6 878 int err;
be0e5c09 879
d397712b 880 while (low < high) {
be0e5c09 881 mid = (low + high) / 2;
5f39d397
CM		 882 offset = p + mid * item_size;
883
884 if (!map_token || offset < map_start ||
885 (offset + sizeof(struct btrfs_disk_key)) >
886 map_start + map_len) {
479965d6 887 if (map_token) {
5f39d397 888 unmap_extent_buffer(eb, map_token, KM_USER0);
479965d6
CM		 889 map_token = NULL;
890 }
934d375b
CM		 891
892 err = map_private_extent_buffer(eb, offset,
479965d6
CM		 893 sizeof(struct btrfs_disk_key),
894 &map_token, &kaddr,
895 &map_start, &map_len, KM_USER0);
896
897 if (!err) {
898 tmp = (struct btrfs_disk_key *)(kaddr + offset -
899 map_start);
900 } else {
901 read_extent_buffer(eb, &unaligned,
902 offset, sizeof(unaligned));
903 tmp = &unaligned;
904 }
5f39d397 905
5f39d397
CM		 906 } else {
907 tmp = (struct btrfs_disk_key *)(kaddr + offset -
908 map_start);
909 }
be0e5c09
CM		 910 ret = comp_keys(tmp, key);
911
912 if (ret < 0)
913 low = mid + 1;
914 else if (ret > 0)
915 high = mid;
916 else {
917 *slot = mid;
479965d6
CM		 918 if (map_token)
919 unmap_extent_buffer(eb, map_token, KM_USER0);
be0e5c09
CM		 920 return 0;
921 }
922 }
923 *slot = low;
5f39d397
CM		 924 if (map_token)
925 unmap_extent_buffer(eb, map_token, KM_USER0);
be0e5c09
CM		 926 return 1;
927}
928
97571fd0
CM		 929/*
930 * simple bin_search frontend that does the right thing for
931 * leaves vs nodes
932 */
5f39d397
CM		 933static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
934 int level, int *slot)
be0e5c09 935{
5f39d397
CM		 936 if (level == 0) {
937 return generic_bin_search(eb,
938 offsetof(struct btrfs_leaf, items),
0783fcfc 939 sizeof(struct btrfs_item),
5f39d397 940 key, btrfs_header_nritems(eb),
7518a238 941 slot);
be0e5c09 942 } else {
5f39d397
CM		 943 return generic_bin_search(eb,
944 offsetof(struct btrfs_node, ptrs),
123abc88 945 sizeof(struct btrfs_key_ptr),
5f39d397 946 key, btrfs_header_nritems(eb),
7518a238 947 slot);
be0e5c09
CM		 948 }
949 return -1;
950}
951
5d4f98a2
YZ		 952int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
953 int level, int *slot)
954{
955 return bin_search(eb, key, level, slot);
956}
957
f0486c68
YZ		 958static void root_add_used(struct btrfs_root *root, u32 size)
959{
960 spin_lock(&root->accounting_lock);
961 btrfs_set_root_used(&root->root_item,
962 btrfs_root_used(&root->root_item) + size);
963 spin_unlock(&root->accounting_lock);
964}
965
966static void root_sub_used(struct btrfs_root *root, u32 size)
967{
968 spin_lock(&root->accounting_lock);
969 btrfs_set_root_used(&root->root_item,
970 btrfs_root_used(&root->root_item) - size);
971 spin_unlock(&root->accounting_lock);
972}
973
d352ac68
CM		 974/* given a node and slot number, this reads the blocks it points to. The
975 * extent buffer is returned with a reference taken (but unlocked).
976 * NULL is returned on error.
977 */
e02119d5 978static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
5f39d397 979 struct extent_buffer *parent, int slot)
bb803951 980{
ca7a79ad 981 int level = btrfs_header_level(parent);
bb803951
CM		 982 if (slot < 0)
983 return NULL;
5f39d397 984 if (slot >= btrfs_header_nritems(parent))
bb803951 985 return NULL;
ca7a79ad
CM		 986
987 BUG_ON(level == 0);
988
db94535d 989 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
ca7a79ad
CM		 990 btrfs_level_size(root, level - 1),
991 btrfs_node_ptr_generation(parent, slot));
bb803951
CM		 992}
993
d352ac68
CM		 994/*
995 * node level balancing, used to make sure nodes are in proper order for
996 * item deletion. We balance from the top down, so we have to make sure
997 * that a deletion won't leave an node completely empty later on.
998 */
e02119d5 999static noinline int balance_level(struct btrfs_trans_handle *trans,
98ed5174
CM		 1000 struct btrfs_root *root,
1001 struct btrfs_path *path, int level)
bb803951 1002{
5f39d397
CM		 1003 struct extent_buffer *right = NULL;
1004 struct extent_buffer *mid;
1005 struct extent_buffer *left = NULL;
1006 struct extent_buffer *parent = NULL;
bb803951
CM		 1007 int ret = 0;
1008 int wret;
1009 int pslot;
bb803951 1010 int orig_slot = path->slots[level];
79f95c82 1011 u64 orig_ptr;
bb803951
CM		 1012
1013 if (level == 0)
1014 return 0;
1015
5f39d397 1016 mid = path->nodes[level];
b4ce94de 1017
925baedd 1018 WARN_ON(!path->locks[level]);
7bb86316
CM		 1019 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1020
1d4f8a0c 1021 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
79f95c82 1022
234b63a0 1023 if (level < BTRFS_MAX_LEVEL - 1)
5f39d397 1024 parent = path->nodes[level + 1];
bb803951
CM		 1025 pslot = path->slots[level + 1];
1026
40689478
CM		 1027 /*
1028 * deal with the case where there is only one pointer in the root
1029 * by promoting the node below to a root
1030 */
5f39d397
CM		 1031 if (!parent) {
1032 struct extent_buffer *child;
bb803951 1033
5f39d397 1034 if (btrfs_header_nritems(mid) != 1)
bb803951
CM		 1035 return 0;
1036
1037 /* promote the child to a root */
5f39d397 1038 child = read_node_slot(root, mid, 0);
7951f3ce 1039 BUG_ON(!child);
925baedd 1040 btrfs_tree_lock(child);
b4ce94de 1041 btrfs_set_lock_blocking(child);
9fa8cfe7 1042 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
f0486c68
YZ		 1043 if (ret) {
1044 btrfs_tree_unlock(child);
1045 free_extent_buffer(child);
1046 goto enospc;
1047 }
2f375ab9 1048
925baedd 1049 spin_lock(&root->node_lock);
bb803951 1050 root->node = child;
925baedd
CM		 1051 spin_unlock(&root->node_lock);
1052
0b86a832 1053 add_root_to_dirty_list(root);
925baedd 1054 btrfs_tree_unlock(child);
b4ce94de 1055
925baedd 1056 path->locks[level] = 0;
bb803951 1057 path->nodes[level] = NULL;
5f39d397 1058 clean_tree_block(trans, root, mid);
925baedd 1059 btrfs_tree_unlock(mid);
bb803951 1060 /* once for the path */
5f39d397 1061 free_extent_buffer(mid);
f0486c68
YZ		 1062
1063 root_sub_used(root, mid->len);
1064 btrfs_free_tree_block(trans, root, mid, 0, 1);
bb803951 1065 /* once for the root ptr */
5f39d397 1066 free_extent_buffer(mid);
f0486c68 1067 return 0;
bb803951 1068 }
5f39d397 1069 if (btrfs_header_nritems(mid) >
123abc88 1070 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
bb803951
CM		 1071 return 0;
1072
559af821 1073 btrfs_header_nritems(mid);
54aa1f4d 1074
5f39d397
CM		 1075 left = read_node_slot(root, parent, pslot - 1);
1076 if (left) {
925baedd 1077 btrfs_tree_lock(left);
b4ce94de 1078 btrfs_set_lock_blocking(left);
5f39d397 1079 wret = btrfs_cow_block(trans, root, left,
9fa8cfe7 1080 parent, pslot - 1, &left);
54aa1f4d
CM		 1081 if (wret) {
1082 ret = wret;
1083 goto enospc;
1084 }
2cc58cf2 1085 }
5f39d397
CM		 1086 right = read_node_slot(root, parent, pslot + 1);
1087 if (right) {
925baedd 1088 btrfs_tree_lock(right);
b4ce94de 1089 btrfs_set_lock_blocking(right);
5f39d397 1090 wret = btrfs_cow_block(trans, root, right,
9fa8cfe7 1091 parent, pslot + 1, &right);
2cc58cf2
CM		 1092 if (wret) {
1093 ret = wret;
1094 goto enospc;
1095 }
1096 }
1097
1098 /* first, try to make some room in the middle buffer */
5f39d397
CM		 1099 if (left) {
1100 orig_slot += btrfs_header_nritems(left);
bce4eae9 1101 wret = push_node_left(trans, root, left, mid, 1);
79f95c82
CM		 1102 if (wret < 0)
1103 ret = wret;
559af821 1104 btrfs_header_nritems(mid);
bb803951 1105 }
79f95c82
CM		 1106
1107 /*
1108 * then try to empty the right most buffer into the middle
1109 */
5f39d397 1110 if (right) {
971a1f66 1111 wret = push_node_left(trans, root, mid, right, 1);
54aa1f4d 1112 if (wret < 0 && wret != -ENOSPC)
79f95c82 1113 ret = wret;
5f39d397 1114 if (btrfs_header_nritems(right) == 0) {
5f39d397 1115 clean_tree_block(trans, root, right);
925baedd 1116 btrfs_tree_unlock(right);
e089f05c
CM		 1117 wret = del_ptr(trans, root, path, level + 1, pslot +
1118 1);
bb803951
CM		 1119 if (wret)
1120 ret = wret;
f0486c68
YZ		 1121 root_sub_used(root, right->len);
1122 btrfs_free_tree_block(trans, root, right, 0, 1);
1123 free_extent_buffer(right);
1124 right = NULL;
bb803951 1125 } else {
5f39d397
CM		 1126 struct btrfs_disk_key right_key;
1127 btrfs_node_key(right, &right_key, 0);
1128 btrfs_set_node_key(parent, &right_key, pslot + 1);
1129 btrfs_mark_buffer_dirty(parent);
bb803951
CM		 1130 }
1131 }
5f39d397 1132 if (btrfs_header_nritems(mid) == 1) {
79f95c82
CM		 1133 /*
1134 * we're not allowed to leave a node with one item in the
1135 * tree during a delete. A deletion from lower in the tree
1136 * could try to delete the only pointer in this node.
1137 * So, pull some keys from the left.
1138 * There has to be a left pointer at this point because
1139 * otherwise we would have pulled some pointers from the
1140 * right
1141 */
5f39d397
CM		 1142 BUG_ON(!left);
1143 wret = balance_node_right(trans, root, mid, left);
54aa1f4d 1144 if (wret < 0) {
79f95c82 1145 ret = wret;
54aa1f4d
CM		 1146 goto enospc;
1147 }
bce4eae9
CM		 1148 if (wret == 1) {
1149 wret = push_node_left(trans, root, left, mid, 1);
1150 if (wret < 0)
1151 ret = wret;
1152 }
79f95c82
CM		 1153 BUG_ON(wret == 1);
1154 }
5f39d397 1155 if (btrfs_header_nritems(mid) == 0) {
5f39d397 1156 clean_tree_block(trans, root, mid);
925baedd 1157 btrfs_tree_unlock(mid);
e089f05c 1158 wret = del_ptr(trans, root, path, level + 1, pslot);
bb803951
CM		 1159 if (wret)
1160 ret = wret;
f0486c68
YZ		 1161 root_sub_used(root, mid->len);
1162 btrfs_free_tree_block(trans, root, mid, 0, 1);
1163 free_extent_buffer(mid);
1164 mid = NULL;
79f95c82
CM		 1165 } else {
1166 /* update the parent key to reflect our changes */
5f39d397
CM		 1167 struct btrfs_disk_key mid_key;
1168 btrfs_node_key(mid, &mid_key, 0);
1169 btrfs_set_node_key(parent, &mid_key, pslot);
1170 btrfs_mark_buffer_dirty(parent);
79f95c82 1171 }
bb803951 1172
79f95c82 1173 /* update the path */
5f39d397
CM		 1174 if (left) {
1175 if (btrfs_header_nritems(left) > orig_slot) {
1176 extent_buffer_get(left);
925baedd 1177 /* left was locked after cow */
5f39d397 1178 path->nodes[level] = left;
bb803951
CM		 1179 path->slots[level + 1] -= 1;
1180 path->slots[level] = orig_slot;
925baedd
CM		 1181 if (mid) {
1182 btrfs_tree_unlock(mid);
5f39d397 1183 free_extent_buffer(mid);
925baedd 1184 }
bb803951 1185 } else {
5f39d397 1186 orig_slot -= btrfs_header_nritems(left);
bb803951
CM		 1187 path->slots[level] = orig_slot;
1188 }
1189 }
79f95c82 1190 /* double check we haven't messed things up */
123abc88 1191 check_block(root, path, level);
e20d96d6 1192 if (orig_ptr !=
5f39d397 1193 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
79f95c82 1194 BUG();
54aa1f4d 1195enospc:
925baedd
CM		 1196 if (right) {
1197 btrfs_tree_unlock(right);
5f39d397 1198 free_extent_buffer(right);
925baedd
CM		 1199 }
1200 if (left) {
1201 if (path->nodes[level] != left)
1202 btrfs_tree_unlock(left);
5f39d397 1203 free_extent_buffer(left);
925baedd 1204 }
bb803951
CM		 1205 return ret;
1206}
1207
d352ac68
CM		 1208/* Node balancing for insertion. Here we only split or push nodes around
1209 * when they are completely full. This is also done top down, so we
1210 * have to be pessimistic.
1211 */
d397712b 1212static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
98ed5174
CM		 1213 struct btrfs_root *root,
1214 struct btrfs_path *path, int level)
e66f709b 1215{
5f39d397
CM		 1216 struct extent_buffer *right = NULL;
1217 struct extent_buffer *mid;
1218 struct extent_buffer *left = NULL;
1219 struct extent_buffer *parent = NULL;
e66f709b
CM		 1220 int ret = 0;
1221 int wret;
1222 int pslot;
1223 int orig_slot = path->slots[level];
e66f709b
CM		 1224
1225 if (level == 0)
1226 return 1;
1227
5f39d397 1228 mid = path->nodes[level];
7bb86316 1229 WARN_ON(btrfs_header_generation(mid) != trans->transid);
e66f709b
CM		 1230
1231 if (level < BTRFS_MAX_LEVEL - 1)
5f39d397 1232 parent = path->nodes[level + 1];
e66f709b
CM		 1233 pslot = path->slots[level + 1];
1234
5f39d397 1235 if (!parent)
e66f709b 1236 return 1;
e66f709b 1237
5f39d397 1238 left = read_node_slot(root, parent, pslot - 1);
e66f709b
CM		 1239
1240 /* first, try to make some room in the middle buffer */
5f39d397 1241 if (left) {
e66f709b 1242 u32 left_nr;
925baedd
CM		 1243
1244 btrfs_tree_lock(left);
b4ce94de
CM		 1245 btrfs_set_lock_blocking(left);
1246
5f39d397 1247 left_nr = btrfs_header_nritems(left);
33ade1f8
CM		 1248 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1249 wret = 1;
1250 } else {
5f39d397 1251 ret = btrfs_cow_block(trans, root, left, parent,
9fa8cfe7 1252 pslot - 1, &left);
54aa1f4d
CM		 1253 if (ret)
1254 wret = 1;
1255 else {
54aa1f4d 1256 wret = push_node_left(trans, root,
971a1f66 1257 left, mid, 0);
54aa1f4d 1258 }
33ade1f8 1259 }
e66f709b
CM		 1260 if (wret < 0)
1261 ret = wret;
1262 if (wret == 0) {
5f39d397 1263 struct btrfs_disk_key disk_key;
e66f709b 1264 orig_slot += left_nr;
5f39d397
CM		 1265 btrfs_node_key(mid, &disk_key, 0);
1266 btrfs_set_node_key(parent, &disk_key, pslot);
1267 btrfs_mark_buffer_dirty(parent);
1268 if (btrfs_header_nritems(left) > orig_slot) {
1269 path->nodes[level] = left;
e66f709b
CM		 1270 path->slots[level + 1] -= 1;
1271 path->slots[level] = orig_slot;
925baedd 1272 btrfs_tree_unlock(mid);
5f39d397 1273 free_extent_buffer(mid);
e66f709b
CM		 1274 } else {
1275 orig_slot -=
5f39d397 1276 btrfs_header_nritems(left);
e66f709b 1277 path->slots[level] = orig_slot;
925baedd 1278 btrfs_tree_unlock(left);
5f39d397 1279 free_extent_buffer(left);
e66f709b 1280 }
e66f709b
CM		 1281 return 0;
1282 }
925baedd 1283 btrfs_tree_unlock(left);
5f39d397 1284 free_extent_buffer(left);
e66f709b 1285 }
925baedd 1286 right = read_node_slot(root, parent, pslot + 1);
e66f709b
CM		 1287
1288 /*
1289 * then try to empty the right most buffer into the middle
1290 */
5f39d397 1291 if (right) {
33ade1f8 1292 u32 right_nr;
b4ce94de 1293
925baedd 1294 btrfs_tree_lock(right);
b4ce94de
CM		 1295 btrfs_set_lock_blocking(right);
1296
5f39d397 1297 right_nr = btrfs_header_nritems(right);
33ade1f8
CM		 1298 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1299 wret = 1;
1300 } else {
5f39d397
CM		 1301 ret = btrfs_cow_block(trans, root, right,
1302 parent, pslot + 1,
9fa8cfe7 1303 &right);
54aa1f4d
CM		 1304 if (ret)
1305 wret = 1;
1306 else {
54aa1f4d 1307 wret = balance_node_right(trans, root,
5f39d397 1308 right, mid);
54aa1f4d 1309 }
33ade1f8 1310 }
e66f709b
CM		 1311 if (wret < 0)
1312 ret = wret;
1313 if (wret == 0) {
5f39d397
CM		 1314 struct btrfs_disk_key disk_key;
1315
1316 btrfs_node_key(right, &disk_key, 0);
1317 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1318 btrfs_mark_buffer_dirty(parent);
1319
1320 if (btrfs_header_nritems(mid) <= orig_slot) {
1321 path->nodes[level] = right;
e66f709b
CM		 1322 path->slots[level + 1] += 1;
1323 path->slots[level] = orig_slot -
5f39d397 1324 btrfs_header_nritems(mid);
925baedd 1325 btrfs_tree_unlock(mid);
5f39d397 1326 free_extent_buffer(mid);
e66f709b 1327 } else {
925baedd 1328 btrfs_tree_unlock(right);
5f39d397 1329 free_extent_buffer(right);
e66f709b 1330 }
e66f709b
CM		 1331 return 0;
1332 }
925baedd 1333 btrfs_tree_unlock(right);
5f39d397 1334 free_extent_buffer(right);
e66f709b 1335 }
e66f709b
CM		 1336 return 1;
1337}
1338
3c69faec 1339/*
d352ac68
CM		 1340 * readahead one full node of leaves, finding things that are close
1341 * to the block in 'slot', and triggering ra on them.
3c69faec 1342 */
c8c42864
CM		 1343static void reada_for_search(struct btrfs_root *root,
1344 struct btrfs_path *path,
1345 int level, int slot, u64 objectid)
3c69faec 1346{
5f39d397 1347 struct extent_buffer *node;
01f46658 1348 struct btrfs_disk_key disk_key;
3c69faec 1349 u32 nritems;
3c69faec 1350 u64 search;
a7175319 1351 u64 target;
6b80053d 1352 u64 nread = 0;
3c69faec 1353 int direction = path->reada;
5f39d397 1354 struct extent_buffer *eb;
6b80053d
CM		 1355 u32 nr;
1356 u32 blocksize;
1357 u32 nscan = 0;
db94535d 1358
a6b6e75e 1359 if (level != 1)
6702ed49
CM		 1360 return;
1361
1362 if (!path->nodes[level])
3c69faec
CM		 1363 return;
1364
5f39d397 1365 node = path->nodes[level];
925baedd 1366
3c69faec 1367 search = btrfs_node_blockptr(node, slot);
6b80053d
CM		 1368 blocksize = btrfs_level_size(root, level - 1);
1369 eb = btrfs_find_tree_block(root, search, blocksize);
5f39d397
CM		 1370 if (eb) {
1371 free_extent_buffer(eb);
3c69faec
CM		 1372 return;
1373 }
1374
a7175319 1375 target = search;
6b80053d 1376
5f39d397 1377 nritems = btrfs_header_nritems(node);
6b80053d 1378 nr = slot;
d397712b 1379 while (1) {
6b80053d
CM		 1380 if (direction < 0) {
1381 if (nr == 0)
1382 break;
1383 nr--;
1384 } else if (direction > 0) {
1385 nr++;
1386 if (nr >= nritems)
1387 break;
3c69faec 1388 }
01f46658
CM		 1389 if (path->reada < 0 && objectid) {
1390 btrfs_node_key(node, &disk_key, nr);
1391 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1392 break;
1393 }
6b80053d 1394 search = btrfs_node_blockptr(node, nr);
a7175319
CM		 1395 if ((search <= target && target - search <= 65536) ||
1396 (search > target && search - target <= 65536)) {
ca7a79ad
CM		 1397 readahead_tree_block(root, search, blocksize,
1398 btrfs_node_ptr_generation(node, nr));
6b80053d
CM		 1399 nread += blocksize;
1400 }
1401 nscan++;
a7175319 1402 if ((nread > 65536 || nscan > 32))
6b80053d 1403 break;
3c69faec
CM		 1404 }
1405}
925baedd 1406
b4ce94de
CM		 1407/*
1408 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1409 * cache
1410 */
1411static noinline int reada_for_balance(struct btrfs_root *root,
1412 struct btrfs_path *path, int level)
1413{
1414 int slot;
1415 int nritems;
1416 struct extent_buffer *parent;
1417 struct extent_buffer *eb;
1418 u64 gen;
1419 u64 block1 = 0;
1420 u64 block2 = 0;
1421 int ret = 0;
1422 int blocksize;
1423
8c594ea8 1424 parent = path->nodes[level + 1];
b4ce94de
CM		 1425 if (!parent)
1426 return 0;
1427
1428 nritems = btrfs_header_nritems(parent);
8c594ea8 1429 slot = path->slots[level + 1];
b4ce94de
CM		 1430 blocksize = btrfs_level_size(root, level);
1431
1432 if (slot > 0) {
1433 block1 = btrfs_node_blockptr(parent, slot - 1);
1434 gen = btrfs_node_ptr_generation(parent, slot - 1);
1435 eb = btrfs_find_tree_block(root, block1, blocksize);
1436 if (eb && btrfs_buffer_uptodate(eb, gen))
1437 block1 = 0;
1438 free_extent_buffer(eb);
1439 }
8c594ea8 1440 if (slot + 1 < nritems) {
b4ce94de
CM		 1441 block2 = btrfs_node_blockptr(parent, slot + 1);
1442 gen = btrfs_node_ptr_generation(parent, slot + 1);
1443 eb = btrfs_find_tree_block(root, block2, blocksize);
1444 if (eb && btrfs_buffer_uptodate(eb, gen))
1445 block2 = 0;
1446 free_extent_buffer(eb);
1447 }
1448 if (block1 || block2) {
1449 ret = -EAGAIN;
8c594ea8
CM		 1450
1451 /* release the whole path */
b4ce94de 1452 btrfs_release_path(root, path);
8c594ea8
CM		 1453
1454 /* read the blocks */
b4ce94de
CM		 1455 if (block1)
1456 readahead_tree_block(root, block1, blocksize, 0);
1457 if (block2)
1458 readahead_tree_block(root, block2, blocksize, 0);
1459
1460 if (block1) {
1461 eb = read_tree_block(root, block1, blocksize, 0);
1462 free_extent_buffer(eb);
1463 }
8c594ea8 1464 if (block2) {
b4ce94de
CM		 1465 eb = read_tree_block(root, block2, blocksize, 0);
1466 free_extent_buffer(eb);
1467 }
1468 }
1469 return ret;
1470}
1471
1472
d352ac68 1473/*
d397712b
CM		 1474 * when we walk down the tree, it is usually safe to unlock the higher layers
1475 * in the tree. The exceptions are when our path goes through slot 0, because
1476 * operations on the tree might require changing key pointers higher up in the
1477 * tree.
d352ac68 1478 *
d397712b
CM		 1479 * callers might also have set path->keep_locks, which tells this code to keep
1480 * the lock if the path points to the last slot in the block. This is part of
1481 * walking through the tree, and selecting the next slot in the higher block.
d352ac68 1482 *
d397712b
CM		 1483 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1484 * if lowest_unlock is 1, level 0 won't be unlocked
d352ac68 1485 */
e02119d5
CM		 1486static noinline void unlock_up(struct btrfs_path *path, int level,
1487 int lowest_unlock)
925baedd
CM		 1488{
1489 int i;
1490 int skip_level = level;
051e1b9f 1491 int no_skips = 0;
925baedd
CM		 1492 struct extent_buffer *t;
1493
1494 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1495 if (!path->nodes[i])
1496 break;
1497 if (!path->locks[i])
1498 break;
051e1b9f 1499 if (!no_skips && path->slots[i] == 0) {
925baedd
CM		 1500 skip_level = i + 1;
1501 continue;
1502 }
051e1b9f 1503 if (!no_skips && path->keep_locks) {
925baedd
CM		 1504 u32 nritems;
1505 t = path->nodes[i];
1506 nritems = btrfs_header_nritems(t);
051e1b9f 1507 if (nritems < 1 || path->slots[i] >= nritems - 1) {
925baedd
CM		 1508 skip_level = i + 1;
1509 continue;
1510 }
1511 }
051e1b9f
CM		 1512 if (skip_level < i && i >= lowest_unlock)
1513 no_skips = 1;
1514
925baedd
CM		 1515 t = path->nodes[i];
1516 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1517 btrfs_tree_unlock(t);
1518 path->locks[i] = 0;
1519 }
1520 }
1521}
1522
b4ce94de
CM		 1523/*
1524 * This releases any locks held in the path starting at level and
1525 * going all the way up to the root.
1526 *
1527 * btrfs_search_slot will keep the lock held on higher nodes in a few
1528 * corner cases, such as COW of the block at slot zero in the node. This
1529 * ignores those rules, and it should only be called when there are no
1530 * more updates to be done higher up in the tree.
1531 */
1532noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1533{
1534 int i;
1535
5d4f98a2 1536 if (path->keep_locks)
b4ce94de
CM		 1537 return;
1538
1539 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1540 if (!path->nodes[i])
12f4dacc 1541 continue;
b4ce94de 1542 if (!path->locks[i])
12f4dacc 1543 continue;
b4ce94de
CM		 1544 btrfs_tree_unlock(path->nodes[i]);
1545 path->locks[i] = 0;
1546 }
1547}
1548
c8c42864
CM		 1549/*
1550 * helper function for btrfs_search_slot. The goal is to find a block
1551 * in cache without setting the path to blocking. If we find the block
1552 * we return zero and the path is unchanged.
1553 *
1554 * If we can't find the block, we set the path blocking and do some
1555 * reada. -EAGAIN is returned and the search must be repeated.
1556 */
1557static int
1558read_block_for_search(struct btrfs_trans_handle *trans,
1559 struct btrfs_root *root, struct btrfs_path *p,
1560 struct extent_buffer **eb_ret, int level, int slot,
1561 struct btrfs_key *key)
1562{
1563 u64 blocknr;
1564 u64 gen;
1565 u32 blocksize;
1566 struct extent_buffer *b = *eb_ret;
1567 struct extent_buffer *tmp;
76a05b35 1568 int ret;
c8c42864
CM		 1569
1570 blocknr = btrfs_node_blockptr(b, slot);
1571 gen = btrfs_node_ptr_generation(b, slot);
1572 blocksize = btrfs_level_size(root, level - 1);
1573
1574 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
cb44921a
CM		 1575 if (tmp) {
1576 if (btrfs_buffer_uptodate(tmp, 0)) {
1577 if (btrfs_buffer_uptodate(tmp, gen)) {
1578 /*
1579 * we found an up to date block without
1580 * sleeping, return
1581 * right away
1582 */
1583 *eb_ret = tmp;
1584 return 0;
1585 }
1586 /* the pages were up to date, but we failed
1587 * the generation number check. Do a full
1588 * read for the generation number that is correct.
1589 * We must do this without dropping locks so
1590 * we can trust our generation number
1591 */
1592 free_extent_buffer(tmp);
1593 tmp = read_tree_block(root, blocknr, blocksize, gen);
1594 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1595 *eb_ret = tmp;
1596 return 0;
1597 }
1598 free_extent_buffer(tmp);
1599 btrfs_release_path(NULL, p);
1600 return -EIO;
1601 }
c8c42864
CM		 1602 }
1603
1604 /*
1605 * reduce lock contention at high levels
1606 * of the btree by dropping locks before
76a05b35
CM		 1607 * we read. Don't release the lock on the current
1608 * level because we need to walk this node to figure
1609 * out which blocks to read.
c8c42864 1610 */
8c594ea8
CM		 1611 btrfs_unlock_up_safe(p, level + 1);
1612 btrfs_set_path_blocking(p);
1613
cb44921a 1614 free_extent_buffer(tmp);
c8c42864
CM		 1615 if (p->reada)
1616 reada_for_search(root, p, level, slot, key->objectid);
1617
8c594ea8 1618 btrfs_release_path(NULL, p);
76a05b35
CM		 1619
1620 ret = -EAGAIN;
5bdd3536 1621 tmp = read_tree_block(root, blocknr, blocksize, 0);
76a05b35
CM		 1622 if (tmp) {
1623 /*
1624 * If the read above didn't mark this buffer up to date,
1625 * it will never end up being up to date. Set ret to EIO now
1626 * and give up so that our caller doesn't loop forever
1627 * on our EAGAINs.
1628 */
1629 if (!btrfs_buffer_uptodate(tmp, 0))
1630 ret = -EIO;
c8c42864 1631 free_extent_buffer(tmp);
76a05b35
CM		 1632 }
1633 return ret;
c8c42864
CM		 1634}
1635
1636/*
1637 * helper function for btrfs_search_slot. This does all of the checks
1638 * for node-level blocks and does any balancing required based on
1639 * the ins_len.
1640 *
1641 * If no extra work was required, zero is returned. If we had to
1642 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1643 * start over
1644 */
1645static int
1646setup_nodes_for_search(struct btrfs_trans_handle *trans,
1647 struct btrfs_root *root, struct btrfs_path *p,
1648 struct extent_buffer *b, int level, int ins_len)
1649{
1650 int ret;
1651 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1652 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1653 int sret;
1654
1655 sret = reada_for_balance(root, p, level);
1656 if (sret)
1657 goto again;
1658
1659 btrfs_set_path_blocking(p);
1660 sret = split_node(trans, root, p, level);
1661 btrfs_clear_path_blocking(p, NULL);
1662
1663 BUG_ON(sret > 0);
1664 if (sret) {
1665 ret = sret;
1666 goto done;
1667 }
1668 b = p->nodes[level];
1669 } else if (ins_len < 0 && btrfs_header_nritems(b) <
cfbb9308 1670 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
c8c42864
CM		 1671 int sret;
1672
1673 sret = reada_for_balance(root, p, level);
1674 if (sret)
1675 goto again;
1676
1677 btrfs_set_path_blocking(p);
1678 sret = balance_level(trans, root, p, level);
1679 btrfs_clear_path_blocking(p, NULL);
1680
1681 if (sret) {
1682 ret = sret;
1683 goto done;
1684 }
1685 b = p->nodes[level];
1686 if (!b) {
1687 btrfs_release_path(NULL, p);
1688 goto again;
1689 }
1690 BUG_ON(btrfs_header_nritems(b) == 1);
1691 }
1692 return 0;
1693
1694again:
1695 ret = -EAGAIN;
1696done:
1697 return ret;
1698}
1699
74123bd7
CM		 1700/*
1701 * look for key in the tree. path is filled in with nodes along the way
1702 * if key is found, we return zero and you can find the item in the leaf
1703 * level of the path (level 0)
1704 *
1705 * If the key isn't found, the path points to the slot where it should
aa5d6bed
CM		 1706 * be inserted, and 1 is returned. If there are other errors during the
1707 * search a negative error number is returned.
97571fd0
CM		 1708 *
1709 * if ins_len > 0, nodes and leaves will be split as we walk down the
1710 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1711 * possible)
74123bd7 1712 */
e089f05c
CM		 1713int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1714 *root, struct btrfs_key *key, struct btrfs_path *p, int
1715 ins_len, int cow)
be0e5c09 1716{
5f39d397 1717 struct extent_buffer *b;
be0e5c09
CM		 1718 int slot;
1719 int ret;
33c66f43 1720 int err;
be0e5c09 1721 int level;
925baedd 1722 int lowest_unlock = 1;
9f3a7427
CM		 1723 u8 lowest_level = 0;
1724
6702ed49 1725 lowest_level = p->lowest_level;
323ac95b 1726 WARN_ON(lowest_level && ins_len > 0);
22b0ebda 1727 WARN_ON(p->nodes[0] != NULL);
25179201 1728
925baedd
CM		 1729 if (ins_len < 0)
1730 lowest_unlock = 2;
65b51a00 1731
bb803951 1732again:
5d4f98a2
YZ		 1733 if (p->search_commit_root) {
1734 b = root->commit_root;
1735 extent_buffer_get(b);
1736 if (!p->skip_locking)
1737 btrfs_tree_lock(b);
1738 } else {
1739 if (p->skip_locking)
1740 b = btrfs_root_node(root);
1741 else
1742 b = btrfs_lock_root_node(root);
1743 }
925baedd 1744
eb60ceac 1745 while (b) {
5f39d397 1746 level = btrfs_header_level(b);
65b51a00
CM		 1747
1748 /*
1749 * setup the path here so we can release it under lock
1750 * contention with the cow code
1751 */
1752 p->nodes[level] = b;
1753 if (!p->skip_locking)
1754 p->locks[level] = 1;
1755
02217ed2 1756 if (cow) {
c8c42864
CM		 1757 /*
1758 * if we don't really need to cow this block
1759 * then we don't want to set the path blocking,
1760 * so we test it here
1761 */
5d4f98a2 1762 if (!should_cow_block(trans, root, b))
65b51a00 1763 goto cow_done;
5d4f98a2 1764
b4ce94de
CM		 1765 btrfs_set_path_blocking(p);
1766
33c66f43
YZ		 1767 err = btrfs_cow_block(trans, root, b,
1768 p->nodes[level + 1],
1769 p->slots[level + 1], &b);
1770 if (err) {
33c66f43 1771 ret = err;
65b51a00 1772 goto done;
54aa1f4d 1773 }
02217ed2 1774 }
65b51a00 1775cow_done:
02217ed2 1776 BUG_ON(!cow && ins_len);
5f39d397 1777 if (level != btrfs_header_level(b))
2c90e5d6 1778 WARN_ON(1);
5f39d397 1779 level = btrfs_header_level(b);
65b51a00 1780
eb60ceac 1781 p->nodes[level] = b;
5cd57b2c
CM		 1782 if (!p->skip_locking)
1783 p->locks[level] = 1;
65b51a00 1784
4008c04a 1785 btrfs_clear_path_blocking(p, NULL);
b4ce94de
CM		 1786
1787 /*
1788 * we have a lock on b and as long as we aren't changing
1789 * the tree, there is no way to for the items in b to change.
1790 * It is safe to drop the lock on our parent before we
1791 * go through the expensive btree search on b.
1792 *
1793 * If cow is true, then we might be changing slot zero,
1794 * which may require changing the parent. So, we can't
1795 * drop the lock until after we know which slot we're
1796 * operating on.
1797 */
1798 if (!cow)
1799 btrfs_unlock_up_safe(p, level + 1);
1800
123abc88 1801 ret = check_block(root, p, level);
65b51a00
CM		 1802 if (ret) {
1803 ret = -1;
1804 goto done;
1805 }
925baedd 1806
5f39d397 1807 ret = bin_search(b, key, level, &slot);
b4ce94de 1808
5f39d397 1809 if (level != 0) {
33c66f43
YZ		 1810 int dec = 0;
1811 if (ret && slot > 0) {
1812 dec = 1;
be0e5c09 1813 slot -= 1;
33c66f43 1814 }
be0e5c09 1815 p->slots[level] = slot;
33c66f43 1816 err = setup_nodes_for_search(trans, root, p, b, level,
c8c42864 1817 ins_len);
33c66f43 1818 if (err == -EAGAIN)
c8c42864 1819 goto again;
33c66f43
YZ		 1820 if (err) {
1821 ret = err;
c8c42864 1822 goto done;
33c66f43 1823 }
c8c42864
CM		 1824 b = p->nodes[level];
1825 slot = p->slots[level];
b4ce94de 1826
f9efa9c7
CM		 1827 unlock_up(p, level, lowest_unlock);
1828
925baedd 1829 if (level == lowest_level) {
33c66f43
YZ		 1830 if (dec)
1831 p->slots[level]++;
5b21f2ed 1832 goto done;
925baedd 1833 }
ca7a79ad 1834
33c66f43 1835 err = read_block_for_search(trans, root, p,
c8c42864 1836 &b, level, slot, key);
33c66f43 1837 if (err == -EAGAIN)
c8c42864 1838 goto again;
33c66f43
YZ		 1839 if (err) {
1840 ret = err;
76a05b35 1841 goto done;
33c66f43 1842 }
76a05b35 1843
b4ce94de 1844 if (!p->skip_locking) {
4008c04a 1845 btrfs_clear_path_blocking(p, NULL);
33c66f43 1846 err = btrfs_try_spin_lock(b);
b4ce94de 1847
33c66f43 1848 if (!err) {
b4ce94de
CM		 1849 btrfs_set_path_blocking(p);
1850 btrfs_tree_lock(b);
4008c04a 1851 btrfs_clear_path_blocking(p, b);
b4ce94de
CM		 1852 }
1853 }
be0e5c09
CM		 1854 } else {
1855 p->slots[level] = slot;
87b29b20
YZ		 1856 if (ins_len > 0 &&
1857 btrfs_leaf_free_space(root, b) < ins_len) {
b4ce94de 1858 btrfs_set_path_blocking(p);
33c66f43
YZ		 1859 err = split_leaf(trans, root, key,
1860 p, ins_len, ret == 0);
4008c04a 1861 btrfs_clear_path_blocking(p, NULL);
b4ce94de 1862
33c66f43
YZ		 1863 BUG_ON(err > 0);
1864 if (err) {
1865 ret = err;
65b51a00
CM		 1866 goto done;
1867 }
5c680ed6 1868 }
459931ec
CM		 1869 if (!p->search_for_split)
1870 unlock_up(p, level, lowest_unlock);
65b51a00 1871 goto done;
be0e5c09
CM		 1872 }
1873 }
65b51a00
CM		 1874 ret = 1;
1875done:
b4ce94de
CM		 1876 /*
1877 * we don't really know what they plan on doing with the path
1878 * from here on, so for now just mark it as blocking
1879 */
b9473439
CM		 1880 if (!p->leave_spinning)
1881 btrfs_set_path_blocking(p);
76a05b35
CM		 1882 if (ret < 0)
1883 btrfs_release_path(root, p);
65b51a00 1884 return ret;
be0e5c09
CM		 1885}
1886
74123bd7
CM		 1887/*
1888 * adjust the pointers going up the tree, starting at level
1889 * making sure the right key of each node is points to 'key'.
1890 * This is used after shifting pointers to the left, so it stops
1891 * fixing up pointers when a given leaf/node is not in slot 0 of the
1892 * higher levels
aa5d6bed
CM		 1893 *
1894 * If this fails to write a tree block, it returns -1, but continues
1895 * fixing up the blocks in ram so the tree is consistent.
74123bd7 1896 */
5f39d397
CM		 1897static int fixup_low_keys(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root, struct btrfs_path *path,
1899 struct btrfs_disk_key *key, int level)
be0e5c09
CM		 1900{
1901 int i;
aa5d6bed 1902 int ret = 0;
5f39d397
CM		 1903 struct extent_buffer *t;
1904
234b63a0 1905 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
be0e5c09 1906 int tslot = path->slots[i];
eb60ceac 1907 if (!path->nodes[i])
be0e5c09 1908 break;
5f39d397
CM		 1909 t = path->nodes[i];
1910 btrfs_set_node_key(t, key, tslot);
d6025579 1911 btrfs_mark_buffer_dirty(path->nodes[i]);
be0e5c09
CM		 1912 if (tslot != 0)
1913 break;
1914 }
aa5d6bed 1915 return ret;
be0e5c09
CM		 1916}
1917
31840ae1
ZY		 1918/*
1919 * update item key.
1920 *
1921 * This function isn't completely safe. It's the caller's responsibility
1922 * that the new key won't break the order
1923 */
1924int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1925 struct btrfs_root *root, struct btrfs_path *path,
1926 struct btrfs_key *new_key)
1927{
1928 struct btrfs_disk_key disk_key;
1929 struct extent_buffer *eb;
1930 int slot;
1931
1932 eb = path->nodes[0];
1933 slot = path->slots[0];
1934 if (slot > 0) {
1935 btrfs_item_key(eb, &disk_key, slot - 1);
1936 if (comp_keys(&disk_key, new_key) >= 0)
1937 return -1;
1938 }
1939 if (slot < btrfs_header_nritems(eb) - 1) {
1940 btrfs_item_key(eb, &disk_key, slot + 1);
1941 if (comp_keys(&disk_key, new_key) <= 0)
1942 return -1;
1943 }
1944
1945 btrfs_cpu_key_to_disk(&disk_key, new_key);
1946 btrfs_set_item_key(eb, &disk_key, slot);
1947 btrfs_mark_buffer_dirty(eb);
1948 if (slot == 0)
1949 fixup_low_keys(trans, root, path, &disk_key, 1);
1950 return 0;
1951}
1952
74123bd7
CM		 1953/*
1954 * try to push data from one node into the next node left in the
79f95c82 1955 * tree.
aa5d6bed
CM		 1956 *
1957 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1958 * error, and > 0 if there was no room in the left hand block.
74123bd7 1959 */
98ed5174
CM		 1960static int push_node_left(struct btrfs_trans_handle *trans,
1961 struct btrfs_root *root, struct extent_buffer *dst,
971a1f66 1962 struct extent_buffer *src, int empty)
be0e5c09 1963{
be0e5c09 1964 int push_items = 0;
bb803951
CM		 1965 int src_nritems;
1966 int dst_nritems;
aa5d6bed 1967 int ret = 0;
be0e5c09 1968
5f39d397
CM		 1969 src_nritems = btrfs_header_nritems(src);
1970 dst_nritems = btrfs_header_nritems(dst);
123abc88 1971 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
7bb86316
CM		 1972 WARN_ON(btrfs_header_generation(src) != trans->transid);
1973 WARN_ON(btrfs_header_generation(dst) != trans->transid);
54aa1f4d 1974
bce4eae9 1975 if (!empty && src_nritems <= 8)
971a1f66
CM		 1976 return 1;
1977
d397712b 1978 if (push_items <= 0)
be0e5c09
CM		 1979 return 1;
1980
bce4eae9 1981 if (empty) {
971a1f66 1982 push_items = min(src_nritems, push_items);
bce4eae9
CM		 1983 if (push_items < src_nritems) {
1984 /* leave at least 8 pointers in the node if
1985 * we aren't going to empty it
1986 */
1987 if (src_nritems - push_items < 8) {
1988 if (push_items <= 8)
1989 return 1;
1990 push_items -= 8;
1991 }
1992 }
1993 } else
1994 push_items = min(src_nritems - 8, push_items);
79f95c82 1995
5f39d397
CM		 1996 copy_extent_buffer(dst, src,
1997 btrfs_node_key_ptr_offset(dst_nritems),
1998 btrfs_node_key_ptr_offset(0),
d397712b 1999 push_items * sizeof(struct btrfs_key_ptr));
5f39d397 2000
bb803951 2001 if (push_items < src_nritems) {
5f39d397
CM		 2002 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
2003 btrfs_node_key_ptr_offset(push_items),
2004 (src_nritems - push_items) *
2005 sizeof(struct btrfs_key_ptr));
2006 }
2007 btrfs_set_header_nritems(src, src_nritems - push_items);
2008 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2009 btrfs_mark_buffer_dirty(src);
2010 btrfs_mark_buffer_dirty(dst);
31840ae1 2011
79f95c82
CM		 2012 return ret;
2013}
2014
2015/*
2016 * try to push data from one node into the next node right in the
2017 * tree.
2018 *
2019 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2020 * error, and > 0 if there was no room in the right hand block.
2021 *
2022 * this will only push up to 1/2 the contents of the left node over
2023 */
5f39d397
CM		 2024static int balance_node_right(struct btrfs_trans_handle *trans,
2025 struct btrfs_root *root,
2026 struct extent_buffer *dst,
2027 struct extent_buffer *src)
79f95c82 2028{
79f95c82
CM		 2029 int push_items = 0;
2030 int max_push;
2031 int src_nritems;
2032 int dst_nritems;
2033 int ret = 0;
79f95c82 2034
7bb86316
CM		 2035 WARN_ON(btrfs_header_generation(src) != trans->transid);
2036 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2037
5f39d397
CM		 2038 src_nritems = btrfs_header_nritems(src);
2039 dst_nritems = btrfs_header_nritems(dst);
123abc88 2040 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
d397712b 2041 if (push_items <= 0)
79f95c82 2042 return 1;
bce4eae9 2043
d397712b 2044 if (src_nritems < 4)
bce4eae9 2045 return 1;
79f95c82
CM		 2046
2047 max_push = src_nritems / 2 + 1;
2048 /* don't try to empty the node */
d397712b 2049 if (max_push >= src_nritems)
79f95c82 2050 return 1;
252c38f0 2051
79f95c82
CM		 2052 if (max_push < push_items)
2053 push_items = max_push;
2054
5f39d397
CM		 2055 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2056 btrfs_node_key_ptr_offset(0),
2057 (dst_nritems) *
2058 sizeof(struct btrfs_key_ptr));
d6025579 2059
5f39d397
CM		 2060 copy_extent_buffer(dst, src,
2061 btrfs_node_key_ptr_offset(0),
2062 btrfs_node_key_ptr_offset(src_nritems - push_items),
d397712b 2063 push_items * sizeof(struct btrfs_key_ptr));
79f95c82 2064
5f39d397
CM		 2065 btrfs_set_header_nritems(src, src_nritems - push_items);
2066 btrfs_set_header_nritems(dst, dst_nritems + push_items);
79f95c82 2067
5f39d397
CM		 2068 btrfs_mark_buffer_dirty(src);
2069 btrfs_mark_buffer_dirty(dst);
31840ae1 2070
aa5d6bed 2071 return ret;
be0e5c09
CM		 2072}
2073
97571fd0
CM		 2074/*
2075 * helper function to insert a new root level in the tree.
2076 * A new node is allocated, and a single item is inserted to
2077 * point to the existing root
aa5d6bed
CM		 2078 *
2079 * returns zero on success or < 0 on failure.
97571fd0 2080 */
d397712b 2081static noinline int insert_new_root(struct btrfs_trans_handle *trans,
5f39d397
CM		 2082 struct btrfs_root *root,
2083 struct btrfs_path *path, int level)
5c680ed6 2084{
7bb86316 2085 u64 lower_gen;
5f39d397
CM		 2086 struct extent_buffer *lower;
2087 struct extent_buffer *c;
925baedd 2088 struct extent_buffer *old;
5f39d397 2089 struct btrfs_disk_key lower_key;
5c680ed6
CM		 2090
2091 BUG_ON(path->nodes[level]);
2092 BUG_ON(path->nodes[level-1] != root->node);
2093
7bb86316
CM		 2094 lower = path->nodes[level-1];
2095 if (level == 1)
2096 btrfs_item_key(lower, &lower_key, 0);
2097 else
2098 btrfs_node_key(lower, &lower_key, 0);
2099
31840ae1 2100 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
5d4f98a2 2101 root->root_key.objectid, &lower_key,
ad3d81ba 2102 level, root->node->start, 0);
5f39d397
CM		 2103 if (IS_ERR(c))
2104 return PTR_ERR(c);
925baedd 2105
f0486c68
YZ		 2106 root_add_used(root, root->nodesize);
2107
5d4f98a2 2108 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
5f39d397
CM		 2109 btrfs_set_header_nritems(c, 1);
2110 btrfs_set_header_level(c, level);
db94535d 2111 btrfs_set_header_bytenr(c, c->start);
5f39d397 2112 btrfs_set_header_generation(c, trans->transid);
5d4f98a2 2113 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
5f39d397 2114 btrfs_set_header_owner(c, root->root_key.objectid);
5f39d397
CM		 2115
2116 write_extent_buffer(c, root->fs_info->fsid,
2117 (unsigned long)btrfs_header_fsid(c),
2118 BTRFS_FSID_SIZE);
e17cade2
CM		 2119
2120 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2121 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2122 BTRFS_UUID_SIZE);
2123
5f39d397 2124 btrfs_set_node_key(c, &lower_key, 0);
db94535d 2125 btrfs_set_node_blockptr(c, 0, lower->start);
7bb86316 2126 lower_gen = btrfs_header_generation(lower);
31840ae1 2127 WARN_ON(lower_gen != trans->transid);
7bb86316
CM		 2128
2129 btrfs_set_node_ptr_generation(c, 0, lower_gen);
d5719762 2130
5f39d397 2131 btrfs_mark_buffer_dirty(c);
d5719762 2132
925baedd
CM		 2133 spin_lock(&root->node_lock);
2134 old = root->node;
5f39d397 2135 root->node = c;
925baedd
CM		 2136 spin_unlock(&root->node_lock);
2137
2138 /* the super has an extra ref to root->node */
2139 free_extent_buffer(old);
2140
0b86a832 2141 add_root_to_dirty_list(root);
5f39d397
CM		 2142 extent_buffer_get(c);
2143 path->nodes[level] = c;
925baedd 2144 path->locks[level] = 1;
5c680ed6
CM		 2145 path->slots[level] = 0;
2146 return 0;
2147}
2148
74123bd7
CM		 2149/*
2150 * worker function to insert a single pointer in a node.
2151 * the node should have enough room for the pointer already
97571fd0 2152 *
74123bd7
CM		 2153 * slot and level indicate where you want the key to go, and
2154 * blocknr is the block the key points to.
aa5d6bed
CM		 2155 *
2156 * returns zero on success and < 0 on any error
74123bd7 2157 */
e089f05c
CM		 2158static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2159 *root, struct btrfs_path *path, struct btrfs_disk_key
db94535d 2160 *key, u64 bytenr, int slot, int level)
74123bd7 2161{
5f39d397 2162 struct extent_buffer *lower;
74123bd7 2163 int nritems;
5c680ed6
CM		 2164
2165 BUG_ON(!path->nodes[level]);
f0486c68 2166 btrfs_assert_tree_locked(path->nodes[level]);
5f39d397
CM		 2167 lower = path->nodes[level];
2168 nritems = btrfs_header_nritems(lower);
c293498b 2169 BUG_ON(slot > nritems);
123abc88 2170 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
74123bd7
CM		 2171 BUG();
2172 if (slot != nritems) {
5f39d397
CM		 2173 memmove_extent_buffer(lower,
2174 btrfs_node_key_ptr_offset(slot + 1),
2175 btrfs_node_key_ptr_offset(slot),
d6025579 2176 (nritems - slot) * sizeof(struct btrfs_key_ptr));
74123bd7 2177 }
5f39d397 2178 btrfs_set_node_key(lower, key, slot);
db94535d 2179 btrfs_set_node_blockptr(lower, slot, bytenr);
74493f7a
CM		 2180 WARN_ON(trans->transid == 0);
2181 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
5f39d397
CM		 2182 btrfs_set_header_nritems(lower, nritems + 1);
2183 btrfs_mark_buffer_dirty(lower);
74123bd7
CM		 2184 return 0;
2185}
2186
97571fd0
CM		 2187/*
2188 * split the node at the specified level in path in two.
2189 * The path is corrected to point to the appropriate node after the split
2190 *
2191 * Before splitting this tries to make some room in the node by pushing
2192 * left and right, if either one works, it returns right away.
aa5d6bed
CM		 2193 *
2194 * returns 0 on success and < 0 on failure
97571fd0 2195 */
e02119d5
CM		 2196static noinline int split_node(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *root,
2198 struct btrfs_path *path, int level)
be0e5c09 2199{
5f39d397
CM		 2200 struct extent_buffer *c;
2201 struct extent_buffer *split;
2202 struct btrfs_disk_key disk_key;
be0e5c09 2203 int mid;
5c680ed6 2204 int ret;
aa5d6bed 2205 int wret;
7518a238 2206 u32 c_nritems;
eb60ceac 2207
5f39d397 2208 c = path->nodes[level];
7bb86316 2209 WARN_ON(btrfs_header_generation(c) != trans->transid);
5f39d397 2210 if (c == root->node) {
5c680ed6 2211 /* trying to split the root, lets make a new one */
e089f05c 2212 ret = insert_new_root(trans, root, path, level + 1);
5c680ed6
CM		 2213 if (ret)
2214 return ret;
b3612421 2215 } else {
e66f709b 2216 ret = push_nodes_for_insert(trans, root, path, level);
5f39d397
CM		 2217 c = path->nodes[level];
2218 if (!ret && btrfs_header_nritems(c) <
c448acf0 2219 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
e66f709b 2220 return 0;
54aa1f4d
CM		 2221 if (ret < 0)
2222 return ret;
be0e5c09 2223 }
e66f709b 2224
5f39d397 2225 c_nritems = btrfs_header_nritems(c);
5d4f98a2
YZ		 2226 mid = (c_nritems + 1) / 2;
2227 btrfs_node_key(c, &disk_key, mid);
7bb86316 2228
5d4f98a2 2229 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
31840ae1 2230 root->root_key.objectid,
5d4f98a2 2231 &disk_key, level, c->start, 0);
5f39d397
CM		 2232 if (IS_ERR(split))
2233 return PTR_ERR(split);
2234
f0486c68
YZ		 2235 root_add_used(root, root->nodesize);
2236
5d4f98a2 2237 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
5f39d397 2238 btrfs_set_header_level(split, btrfs_header_level(c));
db94535d 2239 btrfs_set_header_bytenr(split, split->start);
5f39d397 2240 btrfs_set_header_generation(split, trans->transid);
5d4f98a2 2241 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
5f39d397
CM		 2242 btrfs_set_header_owner(split, root->root_key.objectid);
2243 write_extent_buffer(split, root->fs_info->fsid,
2244 (unsigned long)btrfs_header_fsid(split),
2245 BTRFS_FSID_SIZE);
e17cade2
CM		 2246 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2247 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2248 BTRFS_UUID_SIZE);
54aa1f4d 2249
5f39d397
CM		 2250
2251 copy_extent_buffer(split, c,
2252 btrfs_node_key_ptr_offset(0),
2253 btrfs_node_key_ptr_offset(mid),
2254 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2255 btrfs_set_header_nritems(split, c_nritems - mid);
2256 btrfs_set_header_nritems(c, mid);
aa5d6bed
CM		 2257 ret = 0;
2258
5f39d397
CM		 2259 btrfs_mark_buffer_dirty(c);
2260 btrfs_mark_buffer_dirty(split);
2261
db94535d 2262 wret = insert_ptr(trans, root, path, &disk_key, split->start,
5f39d397 2263 path->slots[level + 1] + 1,
123abc88 2264 level + 1);
aa5d6bed
CM		 2265 if (wret)
2266 ret = wret;
2267
5de08d7d 2268 if (path->slots[level] >= mid) {
5c680ed6 2269 path->slots[level] -= mid;
925baedd 2270 btrfs_tree_unlock(c);
5f39d397
CM		 2271 free_extent_buffer(c);
2272 path->nodes[level] = split;
5c680ed6
CM		 2273 path->slots[level + 1] += 1;
2274 } else {
925baedd 2275 btrfs_tree_unlock(split);
5f39d397 2276 free_extent_buffer(split);
be0e5c09 2277 }
aa5d6bed 2278 return ret;
be0e5c09
CM		 2279}
2280
74123bd7
CM		 2281/*
2282 * how many bytes are required to store the items in a leaf. start
2283 * and nr indicate which items in the leaf to check. This totals up the
2284 * space used both by the item structs and the item data
2285 */
5f39d397 2286static int leaf_space_used(struct extent_buffer *l, int start, int nr)
be0e5c09
CM		 2287{
2288 int data_len;
5f39d397 2289 int nritems = btrfs_header_nritems(l);
d4dbff95 2290 int end = min(nritems, start + nr) - 1;
be0e5c09
CM		 2291
2292 if (!nr)
2293 return 0;
5f39d397
CM		 2294 data_len = btrfs_item_end_nr(l, start);
2295 data_len = data_len - btrfs_item_offset_nr(l, end);
0783fcfc 2296 data_len += sizeof(struct btrfs_item) * nr;
d4dbff95 2297 WARN_ON(data_len < 0);
be0e5c09
CM		 2298 return data_len;
2299}
2300
d4dbff95
CM		 2301/*
2302 * The space between the end of the leaf items and
2303 * the start of the leaf data. IOW, how much room
2304 * the leaf has left for both items and data
2305 */
d397712b 2306noinline int btrfs_leaf_free_space(struct btrfs_root *root,
e02119d5 2307 struct extent_buffer *leaf)
d4dbff95 2308{
5f39d397
CM		 2309 int nritems = btrfs_header_nritems(leaf);
2310 int ret;
2311 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2312 if (ret < 0) {
d397712b
CM		 2313 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2314 "used %d nritems %d\n",
ae2f5411 2315 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
5f39d397
CM		 2316 leaf_space_used(leaf, 0, nritems), nritems);
2317 }
2318 return ret;
d4dbff95
CM		 2319}
2320
99d8f83c
CM		 2321/*
2322 * min slot controls the lowest index we're willing to push to the
2323 * right. We'll push up to and including min_slot, but no lower
2324 */
44871b1b
CM		 2325static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2326 struct btrfs_root *root,
2327 struct btrfs_path *path,
2328 int data_size, int empty,
2329 struct extent_buffer *right,
99d8f83c
CM		 2330 int free_space, u32 left_nritems,
2331 u32 min_slot)
00ec4c51 2332{
5f39d397 2333 struct extent_buffer *left = path->nodes[0];
44871b1b 2334 struct extent_buffer *upper = path->nodes[1];
5f39d397 2335 struct btrfs_disk_key disk_key;
00ec4c51 2336 int slot;
34a38218 2337 u32 i;
00ec4c51
CM		 2338 int push_space = 0;
2339 int push_items = 0;
0783fcfc 2340 struct btrfs_item *item;
34a38218 2341 u32 nr;
7518a238 2342 u32 right_nritems;
5f39d397 2343 u32 data_end;
db94535d 2344 u32 this_item_size;
00ec4c51 2345
34a38218
CM		 2346 if (empty)
2347 nr = 0;
2348 else
99d8f83c 2349 nr = max_t(u32, 1, min_slot);
34a38218 2350
31840ae1 2351 if (path->slots[0] >= left_nritems)
87b29b20 2352 push_space += data_size;
31840ae1 2353
44871b1b 2354 slot = path->slots[1];
34a38218
CM		 2355 i = left_nritems - 1;
2356 while (i >= nr) {
5f39d397 2357 item = btrfs_item_nr(left, i);
db94535d 2358
31840ae1
ZY		 2359 if (!empty && push_items > 0) {
2360 if (path->slots[0] > i)
2361 break;
2362 if (path->slots[0] == i) {
2363 int space = btrfs_leaf_free_space(root, left);
2364 if (space + push_space * 2 > free_space)
2365 break;
2366 }
2367 }
2368
00ec4c51 2369 if (path->slots[0] == i)
87b29b20 2370 push_space += data_size;
db94535d
CM		 2371
2372 if (!left->map_token) {
2373 map_extent_buffer(left, (unsigned long)item,
2374 sizeof(struct btrfs_item),
2375 &left->map_token, &left->kaddr,
2376 &left->map_start, &left->map_len,
2377 KM_USER1);
2378 }
2379
2380 this_item_size = btrfs_item_size(left, item);
2381 if (this_item_size + sizeof(*item) + push_space > free_space)
00ec4c51 2382 break;
31840ae1 2383
00ec4c51 2384 push_items++;
db94535d 2385 push_space += this_item_size + sizeof(*item);
34a38218
CM		 2386 if (i == 0)
2387 break;
2388 i--;
db94535d
CM		 2389 }
2390 if (left->map_token) {
2391 unmap_extent_buffer(left, left->map_token, KM_USER1);
2392 left->map_token = NULL;
00ec4c51 2393 }
5f39d397 2394
925baedd
CM		 2395 if (push_items == 0)
2396 goto out_unlock;
5f39d397 2397
34a38218 2398 if (!empty && push_items == left_nritems)
a429e513 2399 WARN_ON(1);
5f39d397 2400
00ec4c51 2401 /* push left to right */
5f39d397 2402 right_nritems = btrfs_header_nritems(right);
34a38218 2403
5f39d397 2404 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
123abc88 2405 push_space -= leaf_data_end(root, left);
5f39d397 2406
00ec4c51 2407 /* make room in the right data area */
5f39d397
CM		 2408 data_end = leaf_data_end(root, right);
2409 memmove_extent_buffer(right,
2410 btrfs_leaf_data(right) + data_end - push_space,
2411 btrfs_leaf_data(right) + data_end,
2412 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2413
00ec4c51 2414 /* copy from the left data area */
5f39d397 2415 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
d6025579
CM		 2416 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2417 btrfs_leaf_data(left) + leaf_data_end(root, left),
2418 push_space);
5f39d397
CM		 2419
2420 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2421 btrfs_item_nr_offset(0),
2422 right_nritems * sizeof(struct btrfs_item));
2423
00ec4c51 2424 /* copy the items from left to right */
5f39d397
CM		 2425 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2426 btrfs_item_nr_offset(left_nritems - push_items),
2427 push_items * sizeof(struct btrfs_item));
00ec4c51
CM		 2428
2429 /* update the item pointers */
7518a238 2430 right_nritems += push_items;
5f39d397 2431 btrfs_set_header_nritems(right, right_nritems);
123abc88 2432 push_space = BTRFS_LEAF_DATA_SIZE(root);
7518a238 2433 for (i = 0; i < right_nritems; i++) {
5f39d397 2434 item = btrfs_item_nr(right, i);
db94535d
CM		 2435 if (!right->map_token) {
2436 map_extent_buffer(right, (unsigned long)item,
2437 sizeof(struct btrfs_item),
2438 &right->map_token, &right->kaddr,
2439 &right->map_start, &right->map_len,
2440 KM_USER1);
2441 }
2442 push_space -= btrfs_item_size(right, item);
2443 btrfs_set_item_offset(right, item, push_space);
2444 }
2445
2446 if (right->map_token) {
2447 unmap_extent_buffer(right, right->map_token, KM_USER1);
2448 right->map_token = NULL;
00ec4c51 2449 }
7518a238 2450 left_nritems -= push_items;
5f39d397 2451 btrfs_set_header_nritems(left, left_nritems);
00ec4c51 2452
34a38218
CM		 2453 if (left_nritems)
2454 btrfs_mark_buffer_dirty(left);
f0486c68
YZ		 2455 else
2456 clean_tree_block(trans, root, left);
2457
5f39d397 2458 btrfs_mark_buffer_dirty(right);
a429e513 2459
5f39d397
CM		 2460 btrfs_item_key(right, &disk_key, 0);
2461 btrfs_set_node_key(upper, &disk_key, slot + 1);
d6025579 2462 btrfs_mark_buffer_dirty(upper);
02217ed2 2463
00ec4c51 2464 /* then fixup the leaf pointer in the path */
7518a238
CM		 2465 if (path->slots[0] >= left_nritems) {
2466 path->slots[0] -= left_nritems;
925baedd
CM		 2467 if (btrfs_header_nritems(path->nodes[0]) == 0)
2468 clean_tree_block(trans, root, path->nodes[0]);
2469 btrfs_tree_unlock(path->nodes[0]);
5f39d397
CM		 2470 free_extent_buffer(path->nodes[0]);
2471 path->nodes[0] = right;
00ec4c51
CM		 2472 path->slots[1] += 1;
2473 } else {
925baedd 2474 btrfs_tree_unlock(right);
5f39d397 2475 free_extent_buffer(right);
00ec4c51
CM		 2476 }
2477 return 0;
925baedd
CM		 2478
2479out_unlock:
2480 btrfs_tree_unlock(right);
2481 free_extent_buffer(right);
2482 return 1;
00ec4c51 2483}
925baedd 2484
44871b1b
CM		 2485/*
2486 * push some data in the path leaf to the right, trying to free up at
2487 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2488 *
2489 * returns 1 if the push failed because the other node didn't have enough
2490 * room, 0 if everything worked out and < 0 if there were major errors.
99d8f83c
CM		 2491 *
2492 * this will push starting from min_slot to the end of the leaf. It won't
2493 * push any slot lower than min_slot
44871b1b
CM		 2494 */
2495static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
99d8f83c
CM		 2496 *root, struct btrfs_path *path,
2497 int min_data_size, int data_size,
2498 int empty, u32 min_slot)
44871b1b
CM		 2499{
2500 struct extent_buffer *left = path->nodes[0];
2501 struct extent_buffer *right;
2502 struct extent_buffer *upper;
2503 int slot;
2504 int free_space;
2505 u32 left_nritems;
2506 int ret;
2507
2508 if (!path->nodes[1])
2509 return 1;
2510
2511 slot = path->slots[1];
2512 upper = path->nodes[1];
2513 if (slot >= btrfs_header_nritems(upper) - 1)
2514 return 1;
2515
2516 btrfs_assert_tree_locked(path->nodes[1]);
2517
2518 right = read_node_slot(root, upper, slot + 1);
91ca338d
TI		 2519 if (right == NULL)
2520 return 1;
2521
44871b1b
CM		 2522 btrfs_tree_lock(right);
2523 btrfs_set_lock_blocking(right);
2524
2525 free_space = btrfs_leaf_free_space(root, right);
2526 if (free_space < data_size)
2527 goto out_unlock;
2528
2529 /* cow and double check */
2530 ret = btrfs_cow_block(trans, root, right, upper,
2531 slot + 1, &right);
2532 if (ret)
2533 goto out_unlock;
2534
2535 free_space = btrfs_leaf_free_space(root, right);
2536 if (free_space < data_size)
2537 goto out_unlock;
2538
2539 left_nritems = btrfs_header_nritems(left);
2540 if (left_nritems == 0)
2541 goto out_unlock;
2542
99d8f83c
CM		 2543 return __push_leaf_right(trans, root, path, min_data_size, empty,
2544 right, free_space, left_nritems, min_slot);
44871b1b
CM		 2545out_unlock:
2546 btrfs_tree_unlock(right);
2547 free_extent_buffer(right);
2548 return 1;
2549}
2550
74123bd7
CM		 2551/*
2552 * push some data in the path leaf to the left, trying to free up at
2553 * least data_size bytes. returns zero if the push worked, nonzero otherwise
99d8f83c
CM		 2554 *
2555 * max_slot can put a limit on how far into the leaf we'll push items. The
2556 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
2557 * items
74123bd7 2558 */
44871b1b
CM		 2559static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2560 struct btrfs_root *root,
2561 struct btrfs_path *path, int data_size,
2562 int empty, struct extent_buffer *left,
99d8f83c
CM		 2563 int free_space, u32 right_nritems,
2564 u32 max_slot)
be0e5c09 2565{
5f39d397
CM		 2566 struct btrfs_disk_key disk_key;
2567 struct extent_buffer *right = path->nodes[0];
be0e5c09 2568 int i;
be0e5c09
CM		 2569 int push_space = 0;
2570 int push_items = 0;
0783fcfc 2571 struct btrfs_item *item;
7518a238 2572 u32 old_left_nritems;
34a38218 2573 u32 nr;
aa5d6bed
CM		 2574 int ret = 0;
2575 int wret;
db94535d
CM		 2576 u32 this_item_size;
2577 u32 old_left_item_size;
be0e5c09 2578
34a38218 2579 if (empty)
99d8f83c 2580 nr = min(right_nritems, max_slot);
34a38218 2581 else
99d8f83c 2582 nr = min(right_nritems - 1, max_slot);
34a38218
CM		 2583
2584 for (i = 0; i < nr; i++) {
5f39d397 2585 item = btrfs_item_nr(right, i);
db94535d
CM		 2586 if (!right->map_token) {
2587 map_extent_buffer(right, (unsigned long)item,
2588 sizeof(struct btrfs_item),
2589 &right->map_token, &right->kaddr,
2590 &right->map_start, &right->map_len,
2591 KM_USER1);
2592 }
2593
31840ae1
ZY		 2594 if (!empty && push_items > 0) {
2595 if (path->slots[0] < i)
2596 break;
2597 if (path->slots[0] == i) {
2598 int space = btrfs_leaf_free_space(root, right);
2599 if (space + push_space * 2 > free_space)
2600 break;
2601 }
2602 }
2603
be0e5c09 2604 if (path->slots[0] == i)
87b29b20 2605 push_space += data_size;
db94535d
CM		 2606
2607 this_item_size = btrfs_item_size(right, item);
2608 if (this_item_size + sizeof(*item) + push_space > free_space)
be0e5c09 2609 break;
db94535d 2610
be0e5c09 2611 push_items++;
db94535d
CM		 2612 push_space += this_item_size + sizeof(*item);
2613 }
2614
2615 if (right->map_token) {
2616 unmap_extent_buffer(right, right->map_token, KM_USER1);
2617 right->map_token = NULL;
be0e5c09 2618 }
db94535d 2619
be0e5c09 2620 if (push_items == 0) {
925baedd
CM		 2621 ret = 1;
2622 goto out;
be0e5c09 2623 }
34a38218 2624 if (!empty && push_items == btrfs_header_nritems(right))
a429e513 2625 WARN_ON(1);
5f39d397 2626
be0e5c09 2627 /* push data from right to left */
5f39d397
CM		 2628 copy_extent_buffer(left, right,
2629 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2630 btrfs_item_nr_offset(0),
2631 push_items * sizeof(struct btrfs_item));
2632
123abc88 2633 push_space = BTRFS_LEAF_DATA_SIZE(root) -
d397712b 2634 btrfs_item_offset_nr(right, push_items - 1);
5f39d397
CM		 2635
2636 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
d6025579
CM		 2637 leaf_data_end(root, left) - push_space,
2638 btrfs_leaf_data(right) +
5f39d397 2639 btrfs_item_offset_nr(right, push_items - 1),
d6025579 2640 push_space);
5f39d397 2641 old_left_nritems = btrfs_header_nritems(left);
87b29b20 2642 BUG_ON(old_left_nritems <= 0);
eb60ceac 2643
db94535d 2644 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
0783fcfc 2645 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
5f39d397 2646 u32 ioff;
db94535d 2647
5f39d397 2648 item = btrfs_item_nr(left, i);
db94535d
CM		 2649 if (!left->map_token) {
2650 map_extent_buffer(left, (unsigned long)item,
2651 sizeof(struct btrfs_item),
2652 &left->map_token, &left->kaddr,
2653 &left->map_start, &left->map_len,
2654 KM_USER1);
2655 }
2656
5f39d397
CM		 2657 ioff = btrfs_item_offset(left, item);
2658 btrfs_set_item_offset(left, item,
db94535d 2659 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
be0e5c09 2660 }
5f39d397 2661 btrfs_set_header_nritems(left, old_left_nritems + push_items);
db94535d
CM		 2662 if (left->map_token) {
2663 unmap_extent_buffer(left, left->map_token, KM_USER1);
2664 left->map_token = NULL;
2665 }
be0e5c09
CM		 2666
2667 /* fixup right node */
34a38218 2668 if (push_items > right_nritems) {
d397712b
CM		 2669 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2670 right_nritems);
34a38218
CM		 2671 WARN_ON(1);
2672 }
2673
2674 if (push_items < right_nritems) {
2675 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2676 leaf_data_end(root, right);
2677 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2678 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2679 btrfs_leaf_data(right) +
2680 leaf_data_end(root, right), push_space);
2681
2682 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
5f39d397
CM		 2683 btrfs_item_nr_offset(push_items),
2684 (btrfs_header_nritems(right) - push_items) *
2685 sizeof(struct btrfs_item));
34a38218 2686 }
eef1c494
Y		 2687 right_nritems -= push_items;
2688 btrfs_set_header_nritems(right, right_nritems);
123abc88 2689 push_space = BTRFS_LEAF_DATA_SIZE(root);
5f39d397
CM		 2690 for (i = 0; i < right_nritems; i++) {
2691 item = btrfs_item_nr(right, i);
db94535d
CM		 2692
2693 if (!right->map_token) {
2694 map_extent_buffer(right, (unsigned long)item,
2695 sizeof(struct btrfs_item),
2696 &right->map_token, &right->kaddr,
2697 &right->map_start, &right->map_len,
2698 KM_USER1);
2699 }
2700
2701 push_space = push_space - btrfs_item_size(right, item);
2702 btrfs_set_item_offset(right, item, push_space);
2703 }
2704 if (right->map_token) {
2705 unmap_extent_buffer(right, right->map_token, KM_USER1);
2706 right->map_token = NULL;
be0e5c09 2707 }
eb60ceac 2708
5f39d397 2709 btrfs_mark_buffer_dirty(left);
34a38218
CM		 2710 if (right_nritems)
2711 btrfs_mark_buffer_dirty(right);
f0486c68
YZ		 2712 else
2713 clean_tree_block(trans, root, right);
098f59c2 2714
5f39d397
CM		 2715 btrfs_item_key(right, &disk_key, 0);
2716 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
aa5d6bed
CM		 2717 if (wret)
2718 ret = wret;
be0e5c09
CM		 2719
2720 /* then fixup the leaf pointer in the path */
2721 if (path->slots[0] < push_items) {
2722 path->slots[0] += old_left_nritems;
925baedd 2723 btrfs_tree_unlock(path->nodes[0]);
5f39d397
CM		 2724 free_extent_buffer(path->nodes[0]);
2725 path->nodes[0] = left;
be0e5c09
CM		 2726 path->slots[1] -= 1;
2727 } else {
925baedd 2728 btrfs_tree_unlock(left);
5f39d397 2729 free_extent_buffer(left);
be0e5c09
CM		 2730 path->slots[0] -= push_items;
2731 }
eb60ceac 2732 BUG_ON(path->slots[0] < 0);
aa5d6bed 2733 return ret;
925baedd
CM		 2734out:
2735 btrfs_tree_unlock(left);
2736 free_extent_buffer(left);
2737 return ret;
be0e5c09
CM		 2738}
2739
44871b1b
CM		 2740/*
2741 * push some data in the path leaf to the left, trying to free up at
2742 * least data_size bytes. returns zero if the push worked, nonzero otherwise
99d8f83c
CM		 2743 *
2744 * max_slot can put a limit on how far into the leaf we'll push items. The
2745 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
2746 * items
44871b1b
CM		 2747 */
2748static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
99d8f83c
CM		 2749 *root, struct btrfs_path *path, int min_data_size,
2750 int data_size, int empty, u32 max_slot)
44871b1b
CM		 2751{
2752 struct extent_buffer *right = path->nodes[0];
2753 struct extent_buffer *left;
2754 int slot;
2755 int free_space;
2756 u32 right_nritems;
2757 int ret = 0;
2758
2759 slot = path->slots[1];
2760 if (slot == 0)
2761 return 1;
2762 if (!path->nodes[1])
2763 return 1;
2764
2765 right_nritems = btrfs_header_nritems(right);
2766 if (right_nritems == 0)
2767 return 1;
2768
2769 btrfs_assert_tree_locked(path->nodes[1]);
2770
2771 left = read_node_slot(root, path->nodes[1], slot - 1);
91ca338d
TI		 2772 if (left == NULL)
2773 return 1;
2774
44871b1b
CM		 2775 btrfs_tree_lock(left);
2776 btrfs_set_lock_blocking(left);
2777
2778 free_space = btrfs_leaf_free_space(root, left);
2779 if (free_space < data_size) {
2780 ret = 1;
2781 goto out;
2782 }
2783
2784 /* cow and double check */
2785 ret = btrfs_cow_block(trans, root, left,
2786 path->nodes[1], slot - 1, &left);
2787 if (ret) {
2788 /* we hit -ENOSPC, but it isn't fatal here */
2789 ret = 1;
2790 goto out;
2791 }
2792
2793 free_space = btrfs_leaf_free_space(root, left);
2794 if (free_space < data_size) {
2795 ret = 1;
2796 goto out;
2797 }
2798
99d8f83c
CM		 2799 return __push_leaf_left(trans, root, path, min_data_size,
2800 empty, left, free_space, right_nritems,
2801 max_slot);
44871b1b
CM		 2802out:
2803 btrfs_tree_unlock(left);
2804 free_extent_buffer(left);
2805 return ret;
2806}
2807
2808/*
2809 * split the path's leaf in two, making sure there is at least data_size
2810 * available for the resulting leaf level of the path.
2811 *
2812 * returns 0 if all went well and < 0 on failure.
2813 */
2814static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2815 struct btrfs_root *root,
2816 struct btrfs_path *path,
2817 struct extent_buffer *l,
2818 struct extent_buffer *right,
2819 int slot, int mid, int nritems)
2820{
2821 int data_copy_size;
2822 int rt_data_off;
2823 int i;
2824 int ret = 0;
2825 int wret;
2826 struct btrfs_disk_key disk_key;
2827
2828 nritems = nritems - mid;
2829 btrfs_set_header_nritems(right, nritems);
2830 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2831
2832 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2833 btrfs_item_nr_offset(mid),
2834 nritems * sizeof(struct btrfs_item));
2835
2836 copy_extent_buffer(right, l,
2837 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2838 data_copy_size, btrfs_leaf_data(l) +
2839 leaf_data_end(root, l), data_copy_size);
2840
2841 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2842 btrfs_item_end_nr(l, mid);
2843
2844 for (i = 0; i < nritems; i++) {
2845 struct btrfs_item *item = btrfs_item_nr(right, i);
2846 u32 ioff;
2847
2848 if (!right->map_token) {
2849 map_extent_buffer(right, (unsigned long)item,
2850 sizeof(struct btrfs_item),
2851 &right->map_token, &right->kaddr,
2852 &right->map_start, &right->map_len,
2853 KM_USER1);
2854 }
2855
2856 ioff = btrfs_item_offset(right, item);
2857 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2858 }
2859
2860 if (right->map_token) {
2861 unmap_extent_buffer(right, right->map_token, KM_USER1);
2862 right->map_token = NULL;
2863 }
2864
2865 btrfs_set_header_nritems(l, mid);
2866 ret = 0;
2867 btrfs_item_key(right, &disk_key, 0);
2868 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2869 path->slots[1] + 1, 1);
2870 if (wret)
2871 ret = wret;
2872
2873 btrfs_mark_buffer_dirty(right);
2874 btrfs_mark_buffer_dirty(l);
2875 BUG_ON(path->slots[0] != slot);
2876
44871b1b
CM		 2877 if (mid <= slot) {
2878 btrfs_tree_unlock(path->nodes[0]);
2879 free_extent_buffer(path->nodes[0]);
2880 path->nodes[0] = right;
2881 path->slots[0] -= mid;
2882 path->slots[1] += 1;
2883 } else {
2884 btrfs_tree_unlock(right);
2885 free_extent_buffer(right);
2886 }
2887
2888 BUG_ON(path->slots[0] < 0);
2889
2890 return ret;
2891}
2892
99d8f83c
CM		 2893/*
2894 * double splits happen when we need to insert a big item in the middle
2895 * of a leaf. A double split can leave us with 3 mostly empty leaves:
2896 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
2897 * A B C
2898 *
2899 * We avoid this by trying to push the items on either side of our target
2900 * into the adjacent leaves. If all goes well we can avoid the double split
2901 * completely.
2902 */
2903static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
2904 struct btrfs_root *root,
2905 struct btrfs_path *path,
2906 int data_size)
2907{
2908 int ret;
2909 int progress = 0;
2910 int slot;
2911 u32 nritems;
2912
2913 slot = path->slots[0];
2914
2915 /*
2916 * try to push all the items after our slot into the
2917 * right leaf
2918 */
2919 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
2920 if (ret < 0)
2921 return ret;
2922
2923 if (ret == 0)
2924 progress++;
2925
2926 nritems = btrfs_header_nritems(path->nodes[0]);
2927 /*
2928 * our goal is to get our slot at the start or end of a leaf. If
2929 * we've done so we're done
2930 */
2931 if (path->slots[0] == 0 || path->slots[0] == nritems)
2932 return 0;
2933
2934 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
2935 return 0;
2936
2937 /* try to push all the items before our slot into the next leaf */
2938 slot = path->slots[0];
2939 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
2940 if (ret < 0)
2941 return ret;
2942
2943 if (ret == 0)
2944 progress++;
2945
2946 if (progress)
2947 return 0;
2948 return 1;
2949}
2950
74123bd7
CM		 2951/*
2952 * split the path's leaf in two, making sure there is at least data_size
2953 * available for the resulting leaf level of the path.
aa5d6bed
CM		 2954 *
2955 * returns 0 if all went well and < 0 on failure.
74123bd7 2956 */
e02119d5
CM		 2957static noinline int split_leaf(struct btrfs_trans_handle *trans,
2958 struct btrfs_root *root,
2959 struct btrfs_key *ins_key,
2960 struct btrfs_path *path, int data_size,
2961 int extend)
be0e5c09 2962{
5d4f98a2 2963 struct btrfs_disk_key disk_key;
5f39d397 2964 struct extent_buffer *l;
7518a238 2965 u32 nritems;
eb60ceac
CM		 2966 int mid;
2967 int slot;
5f39d397 2968 struct extent_buffer *right;
d4dbff95 2969 int ret = 0;
aa5d6bed 2970 int wret;
5d4f98a2 2971 int split;
cc0c5538 2972 int num_doubles = 0;
99d8f83c 2973 int tried_avoid_double = 0;
aa5d6bed 2974
a5719521
YZ		 2975 l = path->nodes[0];
2976 slot = path->slots[0];
2977 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2978 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2979 return -EOVERFLOW;
2980
40689478 2981 /* first try to make some room by pushing left and right */
99d8f83c
CM		 2982 if (data_size) {
2983 wret = push_leaf_right(trans, root, path, data_size,
2984 data_size, 0, 0);
d397712b 2985 if (wret < 0)
eaee50e8 2986 return wret;
3685f791 2987 if (wret) {
99d8f83c
CM		 2988 wret = push_leaf_left(trans, root, path, data_size,
2989 data_size, 0, (u32)-1);
3685f791
CM		 2990 if (wret < 0)
2991 return wret;
2992 }
2993 l = path->nodes[0];
aa5d6bed 2994
3685f791 2995 /* did the pushes work? */
87b29b20 2996 if (btrfs_leaf_free_space(root, l) >= data_size)
3685f791 2997 return 0;
3326d1b0 2998 }
aa5d6bed 2999
5c680ed6 3000 if (!path->nodes[1]) {
e089f05c 3001 ret = insert_new_root(trans, root, path, 1);
5c680ed6
CM		 3002 if (ret)
3003 return ret;
3004 }
cc0c5538 3005again:
5d4f98a2 3006 split = 1;
cc0c5538 3007 l = path->nodes[0];
eb60ceac 3008 slot = path->slots[0];
5f39d397 3009 nritems = btrfs_header_nritems(l);
d397712b 3010 mid = (nritems + 1) / 2;
54aa1f4d 3011
5d4f98a2
YZ		 3012 if (mid <= slot) {
3013 if (nritems == 1 ||
3014 leaf_space_used(l, mid, nritems - mid) + data_size >
3015 BTRFS_LEAF_DATA_SIZE(root)) {
3016 if (slot >= nritems) {
3017 split = 0;
3018 } else {
3019 mid = slot;
3020 if (mid != nritems &&
3021 leaf_space_used(l, mid, nritems - mid) +
3022 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
99d8f83c
CM		 3023 if (data_size && !tried_avoid_double)
3024 goto push_for_double;
5d4f98a2
YZ		 3025 split = 2;
3026 }
3027 }
3028 }
3029 } else {
3030 if (leaf_space_used(l, 0, mid) + data_size >
3031 BTRFS_LEAF_DATA_SIZE(root)) {
3032 if (!extend && data_size && slot == 0) {
3033 split = 0;
3034 } else if ((extend || !data_size) && slot == 0) {
3035 mid = 1;
3036 } else {
3037 mid = slot;
3038 if (mid != nritems &&
3039 leaf_space_used(l, mid, nritems - mid) +
3040 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
99d8f83c
CM		 3041 if (data_size && !tried_avoid_double)
3042 goto push_for_double;
5d4f98a2
YZ		 3043 split = 2 ;
3044 }
3045 }
3046 }
3047 }
3048
3049 if (split == 0)
3050 btrfs_cpu_key_to_disk(&disk_key, ins_key);
3051 else
3052 btrfs_item_key(l, &disk_key, mid);
3053
3054 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
31840ae1 3055 root->root_key.objectid,
5d4f98a2 3056 &disk_key, 0, l->start, 0);
f0486c68 3057 if (IS_ERR(right))
5f39d397 3058 return PTR_ERR(right);
f0486c68
YZ		 3059
3060 root_add_used(root, root->leafsize);
5f39d397
CM		 3061
3062 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
db94535d 3063 btrfs_set_header_bytenr(right, right->start);
5f39d397 3064 btrfs_set_header_generation(right, trans->transid);
5d4f98a2 3065 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
5f39d397
CM		 3066 btrfs_set_header_owner(right, root->root_key.objectid);
3067 btrfs_set_header_level(right, 0);
3068 write_extent_buffer(right, root->fs_info->fsid,
3069 (unsigned long)btrfs_header_fsid(right),
3070 BTRFS_FSID_SIZE);
e17cade2
CM		 3071
3072 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
3073 (unsigned long)btrfs_header_chunk_tree_uuid(right),
3074 BTRFS_UUID_SIZE);
44871b1b 3075
5d4f98a2
YZ		 3076 if (split == 0) {
3077 if (mid <= slot) {
3078 btrfs_set_header_nritems(right, 0);
3079 wret = insert_ptr(trans, root, path,
3080 &disk_key, right->start,
3081 path->slots[1] + 1, 1);
3082 if (wret)
3083 ret = wret;
925baedd 3084
5d4f98a2
YZ		 3085 btrfs_tree_unlock(path->nodes[0]);
3086 free_extent_buffer(path->nodes[0]);
3087 path->nodes[0] = right;
3088 path->slots[0] = 0;
3089 path->slots[1] += 1;
3090 } else {
3091 btrfs_set_header_nritems(right, 0);
3092 wret = insert_ptr(trans, root, path,
3093 &disk_key,
3094 right->start,
3095 path->slots[1], 1);
3096 if (wret)
3097 ret = wret;
3098 btrfs_tree_unlock(path->nodes[0]);
3099 free_extent_buffer(path->nodes[0]);
3100 path->nodes[0] = right;
3101 path->slots[0] = 0;
3102 if (path->slots[1] == 0) {
3103 wret = fixup_low_keys(trans, root,
3104 path, &disk_key, 1);
d4dbff95
CM		 3105 if (wret)
3106 ret = wret;
5ee78ac7 3107 }
d4dbff95 3108 }
5d4f98a2
YZ		 3109 btrfs_mark_buffer_dirty(right);
3110 return ret;
d4dbff95 3111 }
74123bd7 3112
44871b1b 3113 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
31840ae1
ZY		 3114 BUG_ON(ret);
3115
5d4f98a2 3116 if (split == 2) {
cc0c5538
CM		 3117 BUG_ON(num_doubles != 0);
3118 num_doubles++;
3119 goto again;
a429e513 3120 }
44871b1b 3121
be0e5c09 3122 return ret;
99d8f83c
CM		 3123
3124push_for_double:
3125 push_for_double_split(trans, root, path, data_size);
3126 tried_avoid_double = 1;
3127 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3128 return 0;
3129 goto again;
be0e5c09
CM		 3130}
3131
ad48fd75
YZ		 3132static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3133 struct btrfs_root *root,
3134 struct btrfs_path *path, int ins_len)
459931ec 3135{
ad48fd75 3136 struct btrfs_key key;
459931ec 3137 struct extent_buffer *leaf;
ad48fd75
YZ		 3138 struct btrfs_file_extent_item *fi;
3139 u64 extent_len = 0;
3140 u32 item_size;
3141 int ret;
459931ec
CM		 3142
3143 leaf = path->nodes[0];
ad48fd75
YZ		 3144 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3145
3146 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3147 key.type != BTRFS_EXTENT_CSUM_KEY);
3148
3149 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3150 return 0;
459931ec
CM		 3151
3152 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
ad48fd75
YZ		 3153 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3154 fi = btrfs_item_ptr(leaf, path->slots[0],
3155 struct btrfs_file_extent_item);
3156 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3157 }
459931ec
CM		 3158 btrfs_release_path(root, path);
3159
459931ec 3160 path->keep_locks = 1;
ad48fd75
YZ		 3161 path->search_for_split = 1;
3162 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
459931ec 3163 path->search_for_split = 0;
ad48fd75
YZ		 3164 if (ret < 0)
3165 goto err;
459931ec 3166
ad48fd75
YZ		 3167 ret = -EAGAIN;
3168 leaf = path->nodes[0];
459931ec 3169 /* if our item isn't there or got smaller, return now */
ad48fd75
YZ		 3170 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3171 goto err;
3172
109f6aef
CM		 3173 /* the leaf has changed, it now has room. return now */
3174 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3175 goto err;
3176
ad48fd75
YZ		 3177 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3178 fi = btrfs_item_ptr(leaf, path->slots[0],
3179 struct btrfs_file_extent_item);
3180 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3181 goto err;
459931ec
CM		 3182 }
3183
b9473439 3184 btrfs_set_path_blocking(path);
ad48fd75 3185 ret = split_leaf(trans, root, &key, path, ins_len, 1);
f0486c68
YZ		 3186 if (ret)
3187 goto err;
459931ec 3188
ad48fd75 3189 path->keep_locks = 0;
b9473439 3190 btrfs_unlock_up_safe(path, 1);
ad48fd75
YZ		 3191 return 0;
3192err:
3193 path->keep_locks = 0;
3194 return ret;
3195}
3196
3197static noinline int split_item(struct btrfs_trans_handle *trans,
3198 struct btrfs_root *root,
3199 struct btrfs_path *path,
3200 struct btrfs_key *new_key,
3201 unsigned long split_offset)
3202{
3203 struct extent_buffer *leaf;
3204 struct btrfs_item *item;
3205 struct btrfs_item *new_item;
3206 int slot;
3207 char *buf;
3208 u32 nritems;
3209 u32 item_size;
3210 u32 orig_offset;
3211 struct btrfs_disk_key disk_key;
3212
b9473439
CM		 3213 leaf = path->nodes[0];
3214 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3215
b4ce94de
CM		 3216 btrfs_set_path_blocking(path);
3217
459931ec
CM		 3218 item = btrfs_item_nr(leaf, path->slots[0]);
3219 orig_offset = btrfs_item_offset(leaf, item);
3220 item_size = btrfs_item_size(leaf, item);
3221
459931ec 3222 buf = kmalloc(item_size, GFP_NOFS);
ad48fd75
YZ		 3223 if (!buf)
3224 return -ENOMEM;
3225
459931ec
CM		 3226 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3227 path->slots[0]), item_size);
459931ec 3228
ad48fd75 3229 slot = path->slots[0] + 1;
459931ec 3230 nritems = btrfs_header_nritems(leaf);
459931ec
CM		 3231 if (slot != nritems) {
3232 /* shift the items */
3233 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
ad48fd75
YZ		 3234 btrfs_item_nr_offset(slot),
3235 (nritems - slot) * sizeof(struct btrfs_item));
459931ec
CM		 3236 }
3237
3238 btrfs_cpu_key_to_disk(&disk_key, new_key);
3239 btrfs_set_item_key(leaf, &disk_key, slot);
3240
3241 new_item = btrfs_item_nr(leaf, slot);
3242
3243 btrfs_set_item_offset(leaf, new_item, orig_offset);
3244 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3245
3246 btrfs_set_item_offset(leaf, item,
3247 orig_offset + item_size - split_offset);
3248 btrfs_set_item_size(leaf, item, split_offset);
3249
3250 btrfs_set_header_nritems(leaf, nritems + 1);
3251
3252 /* write the data for the start of the original item */
3253 write_extent_buffer(leaf, buf,
3254 btrfs_item_ptr_offset(leaf, path->slots[0]),
3255 split_offset);
3256
3257 /* write the data for the new item */
3258 write_extent_buffer(leaf, buf + split_offset,
3259 btrfs_item_ptr_offset(leaf, slot),
3260 item_size - split_offset);
3261 btrfs_mark_buffer_dirty(leaf);
3262
ad48fd75 3263 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
459931ec 3264 kfree(buf);
ad48fd75
YZ		 3265 return 0;
3266}
3267
3268/*
3269 * This function splits a single item into two items,
3270 * giving 'new_key' to the new item and splitting the
3271 * old one at split_offset (from the start of the item).
3272 *
3273 * The path may be released by this operation. After
3274 * the split, the path is pointing to the old item. The
3275 * new item is going to be in the same node as the old one.
3276 *
3277 * Note, the item being split must be smaller enough to live alone on
3278 * a tree block with room for one extra struct btrfs_item
3279 *
3280 * This allows us to split the item in place, keeping a lock on the
3281 * leaf the entire time.
3282 */
3283int btrfs_split_item(struct btrfs_trans_handle *trans,
3284 struct btrfs_root *root,
3285 struct btrfs_path *path,
3286 struct btrfs_key *new_key,
3287 unsigned long split_offset)
3288{
3289 int ret;
3290 ret = setup_leaf_for_split(trans, root, path,
3291 sizeof(struct btrfs_item));
3292 if (ret)
3293 return ret;
3294
3295 ret = split_item(trans, root, path, new_key, split_offset);
459931ec
CM		 3296 return ret;
3297}
3298
ad48fd75
YZ		 3299/*
3300 * This function duplicate a item, giving 'new_key' to the new item.
3301 * It guarantees both items live in the same tree leaf and the new item
3302 * is contiguous with the original item.
3303 *
3304 * This allows us to split file extent in place, keeping a lock on the
3305 * leaf the entire time.
3306 */
3307int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3308 struct btrfs_root *root,
3309 struct btrfs_path *path,
3310 struct btrfs_key *new_key)
3311{
3312 struct extent_buffer *leaf;
3313 int ret;
3314 u32 item_size;
3315
3316 leaf = path->nodes[0];
3317 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3318 ret = setup_leaf_for_split(trans, root, path,
3319 item_size + sizeof(struct btrfs_item));
3320 if (ret)
3321 return ret;
3322
3323 path->slots[0]++;
3324 ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3325 item_size, item_size +
3326 sizeof(struct btrfs_item), 1);
3327 BUG_ON(ret);
3328
3329 leaf = path->nodes[0];
3330 memcpy_extent_buffer(leaf,
3331 btrfs_item_ptr_offset(leaf, path->slots[0]),
3332 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3333 item_size);
3334 return 0;
3335}
3336
d352ac68
CM		 3337/*
3338 * make the item pointed to by the path smaller. new_size indicates
3339 * how small to make it, and from_end tells us if we just chop bytes
3340 * off the end of the item or if we shift the item to chop bytes off
3341 * the front.
3342 */
b18c6685
CM		 3343int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3344 struct btrfs_root *root,
3345 struct btrfs_path *path,
179e29e4 3346 u32 new_size, int from_end)
b18c6685
CM		 3347{
3348 int ret = 0;
3349 int slot;
5f39d397
CM		 3350 struct extent_buffer *leaf;
3351 struct btrfs_item *item;
b18c6685
CM		 3352 u32 nritems;
3353 unsigned int data_end;
3354 unsigned int old_data_start;
3355 unsigned int old_size;
3356 unsigned int size_diff;
3357 int i;
3358
5f39d397 3359 leaf = path->nodes[0];
179e29e4
CM		 3360 slot = path->slots[0];
3361
3362 old_size = btrfs_item_size_nr(leaf, slot);
3363 if (old_size == new_size)
3364 return 0;
b18c6685 3365
5f39d397 3366 nritems = btrfs_header_nritems(leaf);
b18c6685
CM		 3367 data_end = leaf_data_end(root, leaf);
3368
5f39d397 3369 old_data_start = btrfs_item_offset_nr(leaf, slot);
179e29e4 3370
b18c6685
CM		 3371 size_diff = old_size - new_size;
3372
3373 BUG_ON(slot < 0);
3374 BUG_ON(slot >= nritems);
3375
3376 /*
3377 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3378 */
3379 /* first correct the data pointers */
3380 for (i = slot; i < nritems; i++) {
5f39d397
CM		 3381 u32 ioff;
3382 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3383
3384 if (!leaf->map_token) {
3385 map_extent_buffer(leaf, (unsigned long)item,
3386 sizeof(struct btrfs_item),
3387 &leaf->map_token, &leaf->kaddr,
3388 &leaf->map_start, &leaf->map_len,
3389 KM_USER1);
3390 }
3391
5f39d397
CM		 3392 ioff = btrfs_item_offset(leaf, item);
3393 btrfs_set_item_offset(leaf, item, ioff + size_diff);
b18c6685 3394 }
db94535d
CM		 3395
3396 if (leaf->map_token) {
3397 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3398 leaf->map_token = NULL;
3399 }
3400
b18c6685 3401 /* shift the data */
179e29e4
CM		 3402 if (from_end) {
3403 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3404 data_end + size_diff, btrfs_leaf_data(leaf) +
3405 data_end, old_data_start + new_size - data_end);
3406 } else {
3407 struct btrfs_disk_key disk_key;
3408 u64 offset;
3409
3410 btrfs_item_key(leaf, &disk_key, slot);
3411
3412 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3413 unsigned long ptr;
3414 struct btrfs_file_extent_item *fi;
3415
3416 fi = btrfs_item_ptr(leaf, slot,
3417 struct btrfs_file_extent_item);
3418 fi = (struct btrfs_file_extent_item *)(
3419 (unsigned long)fi - size_diff);
3420
3421 if (btrfs_file_extent_type(leaf, fi) ==
3422 BTRFS_FILE_EXTENT_INLINE) {
3423 ptr = btrfs_item_ptr_offset(leaf, slot);
3424 memmove_extent_buffer(leaf, ptr,
d397712b
CM		 3425 (unsigned long)fi,
3426 offsetof(struct btrfs_file_extent_item,
179e29e4
CM		 3427 disk_bytenr));
3428 }
3429 }
3430
3431 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3432 data_end + size_diff, btrfs_leaf_data(leaf) +
3433 data_end, old_data_start - data_end);
3434
3435 offset = btrfs_disk_key_offset(&disk_key);
3436 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3437 btrfs_set_item_key(leaf, &disk_key, slot);
3438 if (slot == 0)
3439 fixup_low_keys(trans, root, path, &disk_key, 1);
3440 }
5f39d397
CM		 3441
3442 item = btrfs_item_nr(leaf, slot);
3443 btrfs_set_item_size(leaf, item, new_size);
3444 btrfs_mark_buffer_dirty(leaf);
b18c6685
CM		 3445
3446 ret = 0;
5f39d397
CM		 3447 if (btrfs_leaf_free_space(root, leaf) < 0) {
3448 btrfs_print_leaf(root, leaf);
b18c6685 3449 BUG();
5f39d397 3450 }
b18c6685
CM		 3451 return ret;
3452}
3453
d352ac68
CM		 3454/*
3455 * make the item pointed to by the path bigger, data_size is the new size.
3456 */
5f39d397
CM		 3457int btrfs_extend_item(struct btrfs_trans_handle *trans,
3458 struct btrfs_root *root, struct btrfs_path *path,
3459 u32 data_size)
6567e837
CM		 3460{
3461 int ret = 0;
3462 int slot;
5f39d397
CM		 3463 struct extent_buffer *leaf;
3464 struct btrfs_item *item;
6567e837
CM		 3465 u32 nritems;
3466 unsigned int data_end;
3467 unsigned int old_data;
3468 unsigned int old_size;
3469 int i;
3470
5f39d397 3471 leaf = path->nodes[0];
6567e837 3472
5f39d397 3473 nritems = btrfs_header_nritems(leaf);
6567e837
CM		 3474 data_end = leaf_data_end(root, leaf);
3475
5f39d397
CM		 3476 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3477 btrfs_print_leaf(root, leaf);
6567e837 3478 BUG();
5f39d397 3479 }
6567e837 3480 slot = path->slots[0];
5f39d397 3481 old_data = btrfs_item_end_nr(leaf, slot);
6567e837
CM		 3482
3483 BUG_ON(slot < 0);
3326d1b0
CM		 3484 if (slot >= nritems) {
3485 btrfs_print_leaf(root, leaf);
d397712b
CM		 3486 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3487 slot, nritems);
3326d1b0
CM		 3488 BUG_ON(1);
3489 }
6567e837
CM		 3490
3491 /*
3492 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3493 */
3494 /* first correct the data pointers */
3495 for (i = slot; i < nritems; i++) {
5f39d397
CM		 3496 u32 ioff;
3497 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3498
3499 if (!leaf->map_token) {
3500 map_extent_buffer(leaf, (unsigned long)item,
3501 sizeof(struct btrfs_item),
3502 &leaf->map_token, &leaf->kaddr,
3503 &leaf->map_start, &leaf->map_len,
3504 KM_USER1);
3505 }
5f39d397
CM		 3506 ioff = btrfs_item_offset(leaf, item);
3507 btrfs_set_item_offset(leaf, item, ioff - data_size);
6567e837 3508 }
5f39d397 3509
db94535d
CM		 3510 if (leaf->map_token) {
3511 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3512 leaf->map_token = NULL;
3513 }
3514
6567e837 3515 /* shift the data */
5f39d397 3516 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
6567e837
CM		 3517 data_end - data_size, btrfs_leaf_data(leaf) +
3518 data_end, old_data - data_end);
5f39d397 3519
6567e837 3520 data_end = old_data;
5f39d397
CM		 3521 old_size = btrfs_item_size_nr(leaf, slot);
3522 item = btrfs_item_nr(leaf, slot);
3523 btrfs_set_item_size(leaf, item, old_size + data_size);
3524 btrfs_mark_buffer_dirty(leaf);
6567e837
CM		 3525
3526 ret = 0;
5f39d397
CM		 3527 if (btrfs_leaf_free_space(root, leaf) < 0) {
3528 btrfs_print_leaf(root, leaf);
6567e837 3529 BUG();
5f39d397 3530 }
6567e837
CM		 3531 return ret;
3532}
3533
f3465ca4
JB		 3534/*
3535 * Given a key and some data, insert items into the tree.
3536 * This does all the path init required, making room in the tree if needed.
3537 * Returns the number of keys that were inserted.
3538 */
3539int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3540 struct btrfs_root *root,
3541 struct btrfs_path *path,
3542 struct btrfs_key *cpu_key, u32 *data_size,
3543 int nr)
3544{
3545 struct extent_buffer *leaf;
3546 struct btrfs_item *item;
3547 int ret = 0;
3548 int slot;
f3465ca4
JB		 3549 int i;
3550 u32 nritems;
3551 u32 total_data = 0;
3552 u32 total_size = 0;
3553 unsigned int data_end;
3554 struct btrfs_disk_key disk_key;
3555 struct btrfs_key found_key;
3556
87b29b20
YZ		 3557 for (i = 0; i < nr; i++) {
3558 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3559 BTRFS_LEAF_DATA_SIZE(root)) {
3560 break;
3561 nr = i;
3562 }
f3465ca4 3563 total_data += data_size[i];
87b29b20
YZ		 3564 total_size += data_size[i] + sizeof(struct btrfs_item);
3565 }
3566 BUG_ON(nr == 0);
f3465ca4 3567
f3465ca4
JB		 3568 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3569 if (ret == 0)
3570 return -EEXIST;
3571 if (ret < 0)
3572 goto out;
3573
f3465ca4
JB		 3574 leaf = path->nodes[0];
3575
3576 nritems = btrfs_header_nritems(leaf);
3577 data_end = leaf_data_end(root, leaf);
3578
3579 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3580 for (i = nr; i >= 0; i--) {
3581 total_data -= data_size[i];
3582 total_size -= data_size[i] + sizeof(struct btrfs_item);
3583 if (total_size < btrfs_leaf_free_space(root, leaf))
3584 break;
3585 }
3586 nr = i;
3587 }
3588
3589 slot = path->slots[0];
3590 BUG_ON(slot < 0);
3591
3592 if (slot != nritems) {
3593 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3594
3595 item = btrfs_item_nr(leaf, slot);
3596 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3597
3598 /* figure out how many keys we can insert in here */
3599 total_data = data_size[0];
3600 for (i = 1; i < nr; i++) {
5d4f98a2 3601 if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
f3465ca4
JB		 3602 break;
3603 total_data += data_size[i];
3604 }
3605 nr = i;
3606
3607 if (old_data < data_end) {
3608 btrfs_print_leaf(root, leaf);
d397712b 3609 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
f3465ca4
JB		 3610 slot, old_data, data_end);
3611 BUG_ON(1);
3612 }
3613 /*
3614 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3615 */
3616 /* first correct the data pointers */
3617 WARN_ON(leaf->map_token);
3618 for (i = slot; i < nritems; i++) {
3619 u32 ioff;
3620
3621 item = btrfs_item_nr(leaf, i);
3622 if (!leaf->map_token) {
3623 map_extent_buffer(leaf, (unsigned long)item,
3624 sizeof(struct btrfs_item),
3625 &leaf->map_token, &leaf->kaddr,
3626 &leaf->map_start, &leaf->map_len,
3627 KM_USER1);
3628 }
3629
3630 ioff = btrfs_item_offset(leaf, item);
3631 btrfs_set_item_offset(leaf, item, ioff - total_data);
3632 }
3633 if (leaf->map_token) {
3634 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3635 leaf->map_token = NULL;
3636 }
3637
3638 /* shift the items */
3639 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3640 btrfs_item_nr_offset(slot),
3641 (nritems - slot) * sizeof(struct btrfs_item));
3642
3643 /* shift the data */
3644 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3645 data_end - total_data, btrfs_leaf_data(leaf) +
3646 data_end, old_data - data_end);
3647 data_end = old_data;
3648 } else {
3649 /*
3650 * this sucks but it has to be done, if we are inserting at
3651 * the end of the leaf only insert 1 of the items, since we
3652 * have no way of knowing whats on the next leaf and we'd have
3653 * to drop our current locks to figure it out
3654 */
3655 nr = 1;
3656 }
3657
3658 /* setup the item for the new data */
3659 for (i = 0; i < nr; i++) {
3660 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3661 btrfs_set_item_key(leaf, &disk_key, slot + i);
3662 item = btrfs_item_nr(leaf, slot + i);
3663 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3664 data_end -= data_size[i];
3665 btrfs_set_item_size(leaf, item, data_size[i]);
3666 }
3667 btrfs_set_header_nritems(leaf, nritems + nr);
3668 btrfs_mark_buffer_dirty(leaf);
3669
3670 ret = 0;
3671 if (slot == 0) {
3672 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3673 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3674 }
3675
3676 if (btrfs_leaf_free_space(root, leaf) < 0) {
3677 btrfs_print_leaf(root, leaf);
3678 BUG();
3679 }
3680out:
3681 if (!ret)
3682 ret = nr;
3683 return ret;
3684}
3685
74123bd7 3686/*
44871b1b
CM		 3687 * this is a helper for btrfs_insert_empty_items, the main goal here is
3688 * to save stack depth by doing the bulk of the work in a function
3689 * that doesn't call btrfs_search_slot
74123bd7 3690 */
44871b1b
CM		 3691static noinline_for_stack int
3692setup_items_for_insert(struct btrfs_trans_handle *trans,
3693 struct btrfs_root *root, struct btrfs_path *path,
3694 struct btrfs_key *cpu_key, u32 *data_size,
3695 u32 total_data, u32 total_size, int nr)
be0e5c09 3696{
5f39d397 3697 struct btrfs_item *item;
9c58309d 3698 int i;
7518a238 3699 u32 nritems;
be0e5c09 3700 unsigned int data_end;
e2fa7227 3701 struct btrfs_disk_key disk_key;
44871b1b
CM		 3702 int ret;
3703 struct extent_buffer *leaf;
3704 int slot;
e2fa7227 3705
5f39d397 3706 leaf = path->nodes[0];
44871b1b 3707 slot = path->slots[0];
74123bd7 3708
5f39d397 3709 nritems = btrfs_header_nritems(leaf);
123abc88 3710 data_end = leaf_data_end(root, leaf);
eb60ceac 3711
f25956cc 3712 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3326d1b0 3713 btrfs_print_leaf(root, leaf);
d397712b 3714 printk(KERN_CRIT "not enough freespace need %u have %d\n",
9c58309d 3715 total_size, btrfs_leaf_free_space(root, leaf));
be0e5c09 3716 BUG();
d4dbff95 3717 }
5f39d397 3718
be0e5c09 3719 if (slot != nritems) {
5f39d397 3720 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
be0e5c09 3721
5f39d397
CM		 3722 if (old_data < data_end) {
3723 btrfs_print_leaf(root, leaf);
d397712b 3724 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
5f39d397
CM		 3725 slot, old_data, data_end);
3726 BUG_ON(1);
3727 }
be0e5c09
CM		 3728 /*
3729 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3730 */
3731 /* first correct the data pointers */
db94535d 3732 WARN_ON(leaf->map_token);
0783fcfc 3733 for (i = slot; i < nritems; i++) {
5f39d397 3734 u32 ioff;
db94535d 3735
5f39d397 3736 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3737 if (!leaf->map_token) {
3738 map_extent_buffer(leaf, (unsigned long)item,
3739 sizeof(struct btrfs_item),
3740 &leaf->map_token, &leaf->kaddr,
3741 &leaf->map_start, &leaf->map_len,
3742 KM_USER1);
3743 }
3744
5f39d397 3745 ioff = btrfs_item_offset(leaf, item);
9c58309d 3746 btrfs_set_item_offset(leaf, item, ioff - total_data);
0783fcfc 3747 }
db94535d
CM		 3748 if (leaf->map_token) {
3749 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3750 leaf->map_token = NULL;
3751 }
be0e5c09
CM		 3752
3753 /* shift the items */
9c58309d 3754 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
5f39d397 3755 btrfs_item_nr_offset(slot),
d6025579 3756 (nritems - slot) * sizeof(struct btrfs_item));
be0e5c09
CM		 3757
3758 /* shift the data */
5f39d397 3759 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
9c58309d 3760 data_end - total_data, btrfs_leaf_data(leaf) +
d6025579 3761 data_end, old_data - data_end);
be0e5c09
CM		 3762 data_end = old_data;
3763 }
5f39d397 3764
62e2749e 3765 /* setup the item for the new data */
9c58309d
CM		 3766 for (i = 0; i < nr; i++) {
3767 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3768 btrfs_set_item_key(leaf, &disk_key, slot + i);
3769 item = btrfs_item_nr(leaf, slot + i);
3770 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3771 data_end -= data_size[i];
3772 btrfs_set_item_size(leaf, item, data_size[i]);
3773 }
44871b1b 3774
9c58309d 3775 btrfs_set_header_nritems(leaf, nritems + nr);
aa5d6bed
CM		 3776
3777 ret = 0;
5a01a2e3 3778 if (slot == 0) {
44871b1b 3779 struct btrfs_disk_key disk_key;
5a01a2e3 3780 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
e089f05c 3781 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
5a01a2e3 3782 }
b9473439
CM		 3783 btrfs_unlock_up_safe(path, 1);
3784 btrfs_mark_buffer_dirty(leaf);
aa5d6bed 3785
5f39d397
CM		 3786 if (btrfs_leaf_free_space(root, leaf) < 0) {
3787 btrfs_print_leaf(root, leaf);
be0e5c09 3788 BUG();
5f39d397 3789 }
44871b1b
CM		 3790 return ret;
3791}
3792
3793/*
3794 * Given a key and some data, insert items into the tree.
3795 * This does all the path init required, making room in the tree if needed.
3796 */
3797int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3798 struct btrfs_root *root,
3799 struct btrfs_path *path,
3800 struct btrfs_key *cpu_key, u32 *data_size,
3801 int nr)
3802{
44871b1b
CM		 3803 int ret = 0;
3804 int slot;
3805 int i;
3806 u32 total_size = 0;
3807 u32 total_data = 0;
3808
3809 for (i = 0; i < nr; i++)
3810 total_data += data_size[i];
3811
3812 total_size = total_data + (nr * sizeof(struct btrfs_item));
3813 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3814 if (ret == 0)
3815 return -EEXIST;
3816 if (ret < 0)
3817 goto out;
3818
44871b1b
CM		 3819 slot = path->slots[0];
3820 BUG_ON(slot < 0);
3821
3822 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3823 total_data, total_size, nr);
3824
ed2ff2cb 3825out:
62e2749e
CM		 3826 return ret;
3827}
3828
3829/*
3830 * Given a key and some data, insert an item into the tree.
3831 * This does all the path init required, making room in the tree if needed.
3832 */
e089f05c
CM		 3833int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3834 *root, struct btrfs_key *cpu_key, void *data, u32
3835 data_size)
62e2749e
CM		 3836{
3837 int ret = 0;
2c90e5d6 3838 struct btrfs_path *path;
5f39d397
CM		 3839 struct extent_buffer *leaf;
3840 unsigned long ptr;
62e2749e 3841
2c90e5d6
CM		 3842 path = btrfs_alloc_path();
3843 BUG_ON(!path);
2c90e5d6 3844 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
62e2749e 3845 if (!ret) {
5f39d397
CM		 3846 leaf = path->nodes[0];
3847 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3848 write_extent_buffer(leaf, data, ptr, data_size);
3849 btrfs_mark_buffer_dirty(leaf);
62e2749e 3850 }
2c90e5d6 3851 btrfs_free_path(path);
aa5d6bed 3852 return ret;
be0e5c09
CM		 3853}
3854
74123bd7 3855/*
5de08d7d 3856 * delete the pointer from a given node.
74123bd7 3857 *
d352ac68
CM		 3858 * the tree should have been previously balanced so the deletion does not
3859 * empty a node.
74123bd7 3860 */
e089f05c
CM		 3861static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3862 struct btrfs_path *path, int level, int slot)
be0e5c09 3863{
5f39d397 3864 struct extent_buffer *parent = path->nodes[level];
7518a238 3865 u32 nritems;
aa5d6bed 3866 int ret = 0;
bb803951 3867 int wret;
be0e5c09 3868
5f39d397 3869 nritems = btrfs_header_nritems(parent);
d397712b 3870 if (slot != nritems - 1) {
5f39d397
CM		 3871 memmove_extent_buffer(parent,
3872 btrfs_node_key_ptr_offset(slot),
3873 btrfs_node_key_ptr_offset(slot + 1),
d6025579
CM		 3874 sizeof(struct btrfs_key_ptr) *
3875 (nritems - slot - 1));
bb803951 3876 }
7518a238 3877 nritems--;
5f39d397 3878 btrfs_set_header_nritems(parent, nritems);
7518a238 3879 if (nritems == 0 && parent == root->node) {
5f39d397 3880 BUG_ON(btrfs_header_level(root->node) != 1);
bb803951 3881 /* just turn the root into a leaf and break */
5f39d397 3882 btrfs_set_header_level(root->node, 0);
bb803951 3883 } else if (slot == 0) {
5f39d397
CM		 3884 struct btrfs_disk_key disk_key;
3885
3886 btrfs_node_key(parent, &disk_key, 0);
3887 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
0f70abe2
CM		 3888 if (wret)
3889 ret = wret;
be0e5c09 3890 }
d6025579 3891 btrfs_mark_buffer_dirty(parent);
aa5d6bed 3892 return ret;
be0e5c09
CM		 3893}
3894
323ac95b
CM		 3895/*
3896 * a helper function to delete the leaf pointed to by path->slots[1] and
5d4f98a2 3897 * path->nodes[1].
323ac95b
CM		 3898 *
3899 * This deletes the pointer in path->nodes[1] and frees the leaf
3900 * block extent. zero is returned if it all worked out, < 0 otherwise.
3901 *
3902 * The path must have already been setup for deleting the leaf, including
3903 * all the proper balancing. path->nodes[1] must be locked.
3904 */
5d4f98a2
YZ		 3905static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root,
3907 struct btrfs_path *path,
3908 struct extent_buffer *leaf)
323ac95b
CM		 3909{
3910 int ret;
323ac95b 3911
5d4f98a2 3912 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
323ac95b
CM		 3913 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3914 if (ret)
3915 return ret;
3916
4d081c41
CM		 3917 /*
3918 * btrfs_free_extent is expensive, we want to make sure we
3919 * aren't holding any locks when we call it
3920 */
3921 btrfs_unlock_up_safe(path, 0);
3922
f0486c68
YZ		 3923 root_sub_used(root, leaf->len);
3924
3925 btrfs_free_tree_block(trans, root, leaf, 0, 1);
3926 return 0;
323ac95b 3927}
74123bd7
CM		 3928/*
3929 * delete the item at the leaf level in path. If that empties
3930 * the leaf, remove it from the tree
3931 */
85e21bac
CM		 3932int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3933 struct btrfs_path *path, int slot, int nr)
be0e5c09 3934{
5f39d397
CM		 3935 struct extent_buffer *leaf;
3936 struct btrfs_item *item;
85e21bac
CM		 3937 int last_off;
3938 int dsize = 0;
aa5d6bed
CM		 3939 int ret = 0;
3940 int wret;
85e21bac 3941 int i;
7518a238 3942 u32 nritems;
be0e5c09 3943
5f39d397 3944 leaf = path->nodes[0];
85e21bac
CM		 3945 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3946
3947 for (i = 0; i < nr; i++)
3948 dsize += btrfs_item_size_nr(leaf, slot + i);
3949
5f39d397 3950 nritems = btrfs_header_nritems(leaf);
be0e5c09 3951
85e21bac 3952 if (slot + nr != nritems) {
123abc88 3953 int data_end = leaf_data_end(root, leaf);
5f39d397
CM		 3954
3955 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
d6025579
CM		 3956 data_end + dsize,
3957 btrfs_leaf_data(leaf) + data_end,
85e21bac 3958 last_off - data_end);
5f39d397 3959
85e21bac 3960 for (i = slot + nr; i < nritems; i++) {
5f39d397 3961 u32 ioff;
db94535d 3962
5f39d397 3963 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3964 if (!leaf->map_token) {
3965 map_extent_buffer(leaf, (unsigned long)item,
3966 sizeof(struct btrfs_item),
3967 &leaf->map_token, &leaf->kaddr,
3968 &leaf->map_start, &leaf->map_len,
3969 KM_USER1);
3970 }
5f39d397
CM		 3971 ioff = btrfs_item_offset(leaf, item);
3972 btrfs_set_item_offset(leaf, item, ioff + dsize);
0783fcfc 3973 }
db94535d
CM		 3974
3975 if (leaf->map_token) {
3976 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3977 leaf->map_token = NULL;
3978 }
3979
5f39d397 3980 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
85e21bac 3981 btrfs_item_nr_offset(slot + nr),
d6025579 3982 sizeof(struct btrfs_item) *
85e21bac 3983 (nritems - slot - nr));
be0e5c09 3984 }
85e21bac
CM		 3985 btrfs_set_header_nritems(leaf, nritems - nr);
3986 nritems -= nr;
5f39d397 3987
74123bd7 3988 /* delete the leaf if we've emptied it */
7518a238 3989 if (nritems == 0) {
5f39d397
CM		 3990 if (leaf == root->node) {
3991 btrfs_set_header_level(leaf, 0);
9a8dd150 3992 } else {
f0486c68
YZ		 3993 btrfs_set_path_blocking(path);
3994 clean_tree_block(trans, root, leaf);
5d4f98a2 3995 ret = btrfs_del_leaf(trans, root, path, leaf);
323ac95b 3996 BUG_ON(ret);
9a8dd150 3997 }
be0e5c09 3998 } else {
7518a238 3999 int used = leaf_space_used(leaf, 0, nritems);
aa5d6bed 4000 if (slot == 0) {
5f39d397
CM		 4001 struct btrfs_disk_key disk_key;
4002
4003 btrfs_item_key(leaf, &disk_key, 0);
e089f05c 4004 wret = fixup_low_keys(trans, root, path,
5f39d397 4005 &disk_key, 1);
aa5d6bed
CM		 4006 if (wret)
4007 ret = wret;
4008 }
aa5d6bed 4009
74123bd7 4010 /* delete the leaf if it is mostly empty */
d717aa1d 4011 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
be0e5c09
CM		 4012 /* push_leaf_left fixes the path.
4013 * make sure the path still points to our leaf
4014 * for possible call to del_ptr below
4015 */
4920c9ac 4016 slot = path->slots[1];
5f39d397
CM		 4017 extent_buffer_get(leaf);
4018
b9473439 4019 btrfs_set_path_blocking(path);
99d8f83c
CM		 4020 wret = push_leaf_left(trans, root, path, 1, 1,
4021 1, (u32)-1);
54aa1f4d 4022 if (wret < 0 && wret != -ENOSPC)
aa5d6bed 4023 ret = wret;
5f39d397
CM		 4024
4025 if (path->nodes[0] == leaf &&
4026 btrfs_header_nritems(leaf)) {
99d8f83c
CM		 4027 wret = push_leaf_right(trans, root, path, 1,
4028 1, 1, 0);
54aa1f4d 4029 if (wret < 0 && wret != -ENOSPC)
aa5d6bed
CM		 4030 ret = wret;
4031 }
5f39d397
CM		 4032
4033 if (btrfs_header_nritems(leaf) == 0) {
323ac95b 4034 path->slots[1] = slot;
5d4f98a2 4035 ret = btrfs_del_leaf(trans, root, path, leaf);
323ac95b 4036 BUG_ON(ret);
5f39d397 4037 free_extent_buffer(leaf);
5de08d7d 4038 } else {
925baedd
CM		 4039 /* if we're still in the path, make sure
4040 * we're dirty. Otherwise, one of the
4041 * push_leaf functions must have already
4042 * dirtied this buffer
4043 */
4044 if (path->nodes[0] == leaf)
4045 btrfs_mark_buffer_dirty(leaf);
5f39d397 4046 free_extent_buffer(leaf);
be0e5c09 4047 }
d5719762 4048 } else {
5f39d397 4049 btrfs_mark_buffer_dirty(leaf);
be0e5c09
CM		 4050 }
4051 }
aa5d6bed 4052 return ret;
be0e5c09
CM		 4053}
4054
7bb86316 4055/*
925baedd 4056 * search the tree again to find a leaf with lesser keys
7bb86316
CM		 4057 * returns 0 if it found something or 1 if there are no lesser leaves.
4058 * returns < 0 on io errors.
d352ac68
CM		 4059 *
4060 * This may release the path, and so you may lose any locks held at the
4061 * time you call it.
7bb86316
CM		 4062 */
4063int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4064{
925baedd
CM		 4065 struct btrfs_key key;
4066 struct btrfs_disk_key found_key;
4067 int ret;
7bb86316 4068
925baedd 4069 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
7bb86316 4070
925baedd
CM		 4071 if (key.offset > 0)
4072 key.offset--;
4073 else if (key.type > 0)
4074 key.type--;
4075 else if (key.objectid > 0)
4076 key.objectid--;
4077 else
4078 return 1;
7bb86316 4079
925baedd
CM		 4080 btrfs_release_path(root, path);
4081 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4082 if (ret < 0)
4083 return ret;
4084 btrfs_item_key(path->nodes[0], &found_key, 0);
4085 ret = comp_keys(&found_key, &key);
4086 if (ret < 0)
4087 return 0;
4088 return 1;
7bb86316
CM		 4089}
4090
3f157a2f
CM		 4091/*
4092 * A helper function to walk down the tree starting at min_key, and looking
4093 * for nodes or leaves that are either in cache or have a minimum
d352ac68 4094 * transaction id. This is used by the btree defrag code, and tree logging
3f157a2f
CM		 4095 *
4096 * This does not cow, but it does stuff the starting key it finds back
4097 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4098 * key and get a writable path.
4099 *
4100 * This does lock as it descends, and path->keep_locks should be set
4101 * to 1 by the caller.
4102 *
4103 * This honors path->lowest_level to prevent descent past a given level
4104 * of the tree.
4105 *
d352ac68
CM		 4106 * min_trans indicates the oldest transaction that you are interested
4107 * in walking through. Any nodes or leaves older than min_trans are
4108 * skipped over (without reading them).
4109 *
3f157a2f
CM		 4110 * returns zero if something useful was found, < 0 on error and 1 if there
4111 * was nothing in the tree that matched the search criteria.
4112 */
4113int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
e02119d5 4114 struct btrfs_key *max_key,
3f157a2f
CM		 4115 struct btrfs_path *path, int cache_only,
4116 u64 min_trans)
4117{
4118 struct extent_buffer *cur;
4119 struct btrfs_key found_key;
4120 int slot;
9652480b 4121 int sret;
3f157a2f
CM		 4122 u32 nritems;
4123 int level;
4124 int ret = 1;
4125
934d375b 4126 WARN_ON(!path->keep_locks);
3f157a2f
CM		 4127again:
4128 cur = btrfs_lock_root_node(root);
4129 level = btrfs_header_level(cur);
e02119d5 4130 WARN_ON(path->nodes[level]);
3f157a2f
CM		 4131 path->nodes[level] = cur;
4132 path->locks[level] = 1;
4133
4134 if (btrfs_header_generation(cur) < min_trans) {
4135 ret = 1;
4136 goto out;
4137 }
d397712b 4138 while (1) {
3f157a2f
CM		 4139 nritems = btrfs_header_nritems(cur);
4140 level = btrfs_header_level(cur);
9652480b 4141 sret = bin_search(cur, min_key, level, &slot);
3f157a2f 4142
323ac95b
CM		 4143 /* at the lowest level, we're done, setup the path and exit */
4144 if (level == path->lowest_level) {
e02119d5
CM		 4145 if (slot >= nritems)
4146 goto find_next_key;
3f157a2f
CM		 4147 ret = 0;
4148 path->slots[level] = slot;
4149 btrfs_item_key_to_cpu(cur, &found_key, slot);
4150 goto out;
4151 }
9652480b
Y		 4152 if (sret && slot > 0)
4153 slot--;
3f157a2f
CM		 4154 /*
4155 * check this node pointer against the cache_only and
4156 * min_trans parameters. If it isn't in cache or is too
4157 * old, skip to the next one.
4158 */
d397712b 4159 while (slot < nritems) {
3f157a2f
CM		 4160 u64 blockptr;
4161 u64 gen;
4162 struct extent_buffer *tmp;
e02119d5
CM		 4163 struct btrfs_disk_key disk_key;
4164
3f157a2f
CM		 4165 blockptr = btrfs_node_blockptr(cur, slot);
4166 gen = btrfs_node_ptr_generation(cur, slot);
4167 if (gen < min_trans) {
4168 slot++;
4169 continue;
4170 }
4171 if (!cache_only)
4172 break;
4173
e02119d5
CM		 4174 if (max_key) {
4175 btrfs_node_key(cur, &disk_key, slot);
4176 if (comp_keys(&disk_key, max_key) >= 0) {
4177 ret = 1;
4178 goto out;
4179 }
4180 }
4181
3f157a2f
CM		 4182 tmp = btrfs_find_tree_block(root, blockptr,
4183 btrfs_level_size(root, level - 1));
4184
4185 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4186 free_extent_buffer(tmp);
4187 break;
4188 }
4189 if (tmp)
4190 free_extent_buffer(tmp);
4191 slot++;
4192 }
e02119d5 4193find_next_key:
3f157a2f
CM		 4194 /*
4195 * we didn't find a candidate key in this node, walk forward
4196 * and find another one
4197 */
4198 if (slot >= nritems) {
e02119d5 4199 path->slots[level] = slot;
b4ce94de 4200 btrfs_set_path_blocking(path);
e02119d5 4201 sret = btrfs_find_next_key(root, path, min_key, level,
3f157a2f 4202 cache_only, min_trans);
e02119d5 4203 if (sret == 0) {
3f157a2f
CM		 4204 btrfs_release_path(root, path);
4205 goto again;
4206 } else {
4207 goto out;
4208 }
4209 }
4210 /* save our key for returning back */
4211 btrfs_node_key_to_cpu(cur, &found_key, slot);
4212 path->slots[level] = slot;
4213 if (level == path->lowest_level) {
4214 ret = 0;
4215 unlock_up(path, level, 1);
4216 goto out;
4217 }
b4ce94de 4218 btrfs_set_path_blocking(path);
3f157a2f
CM		 4219 cur = read_node_slot(root, cur, slot);
4220
4221 btrfs_tree_lock(cur);
b4ce94de 4222
3f157a2f
CM		 4223 path->locks[level - 1] = 1;
4224 path->nodes[level - 1] = cur;
4225 unlock_up(path, level, 1);
4008c04a 4226 btrfs_clear_path_blocking(path, NULL);
3f157a2f
CM		 4227 }
4228out:
4229 if (ret == 0)
4230 memcpy(min_key, &found_key, sizeof(found_key));
b4ce94de 4231 btrfs_set_path_blocking(path);
3f157a2f
CM		 4232 return ret;
4233}
4234
4235/*
4236 * this is similar to btrfs_next_leaf, but does not try to preserve
4237 * and fixup the path. It looks for and returns the next key in the
4238 * tree based on the current path and the cache_only and min_trans
4239 * parameters.
4240 *
4241 * 0 is returned if another key is found, < 0 if there are any errors
4242 * and 1 is returned if there are no higher keys in the tree
4243 *
4244 * path->keep_locks should be set to 1 on the search made before
4245 * calling this function.
4246 */
e7a84565 4247int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
33c66f43 4248 struct btrfs_key *key, int level,
3f157a2f 4249 int cache_only, u64 min_trans)
e7a84565 4250{
e7a84565
CM		 4251 int slot;
4252 struct extent_buffer *c;
4253
934d375b 4254 WARN_ON(!path->keep_locks);
d397712b 4255 while (level < BTRFS_MAX_LEVEL) {
e7a84565
CM		 4256 if (!path->nodes[level])
4257 return 1;
4258
4259 slot = path->slots[level] + 1;
4260 c = path->nodes[level];
3f157a2f 4261next:
e7a84565 4262 if (slot >= btrfs_header_nritems(c)) {
33c66f43
YZ		 4263 int ret;
4264 int orig_lowest;
4265 struct btrfs_key cur_key;
4266 if (level + 1 >= BTRFS_MAX_LEVEL ||
4267 !path->nodes[level + 1])
e7a84565 4268 return 1;
33c66f43
YZ		 4269
4270 if (path->locks[level + 1]) {
4271 level++;
4272 continue;
4273 }
4274
4275 slot = btrfs_header_nritems(c) - 1;
4276 if (level == 0)
4277 btrfs_item_key_to_cpu(c, &cur_key, slot);
4278 else
4279 btrfs_node_key_to_cpu(c, &cur_key, slot);
4280
4281 orig_lowest = path->lowest_level;
4282 btrfs_release_path(root, path);
4283 path->lowest_level = level;
4284 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4285 0, 0);
4286 path->lowest_level = orig_lowest;
4287 if (ret < 0)
4288 return ret;
4289
4290 c = path->nodes[level];
4291 slot = path->slots[level];
4292 if (ret == 0)
4293 slot++;
4294 goto next;
e7a84565 4295 }
33c66f43 4296
e7a84565
CM		 4297 if (level == 0)
4298 btrfs_item_key_to_cpu(c, key, slot);
3f157a2f
CM		 4299 else {
4300 u64 blockptr = btrfs_node_blockptr(c, slot);
4301 u64 gen = btrfs_node_ptr_generation(c, slot);
4302
4303 if (cache_only) {
4304 struct extent_buffer *cur;
4305 cur = btrfs_find_tree_block(root, blockptr,
4306 btrfs_level_size(root, level - 1));
4307 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4308 slot++;
4309 if (cur)
4310 free_extent_buffer(cur);
4311 goto next;
4312 }
4313 free_extent_buffer(cur);
4314 }
4315 if (gen < min_trans) {
4316 slot++;
4317 goto next;
4318 }
e7a84565 4319 btrfs_node_key_to_cpu(c, key, slot);
3f157a2f 4320 }
e7a84565
CM		 4321 return 0;
4322 }
4323 return 1;
4324}
4325
97571fd0 4326/*
925baedd 4327 * search the tree again to find a leaf with greater keys
0f70abe2
CM		 4328 * returns 0 if it found something or 1 if there are no greater leaves.
4329 * returns < 0 on io errors.
97571fd0 4330 */
234b63a0 4331int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
d97e63b6
CM		 4332{
4333 int slot;
8e73f275 4334 int level;
5f39d397 4335 struct extent_buffer *c;
8e73f275 4336 struct extent_buffer *next;
925baedd
CM		 4337 struct btrfs_key key;
4338 u32 nritems;
4339 int ret;
8e73f275
CM		 4340 int old_spinning = path->leave_spinning;
4341 int force_blocking = 0;
925baedd
CM		 4342
4343 nritems = btrfs_header_nritems(path->nodes[0]);
d397712b 4344 if (nritems == 0)
925baedd 4345 return 1;
925baedd 4346
8e73f275
CM		 4347 /*
4348 * we take the blocks in an order that upsets lockdep. Using
4349 * blocking mode is the only way around it.
4350 */
4351#ifdef CONFIG_DEBUG_LOCK_ALLOC
4352 force_blocking = 1;
4353#endif
925baedd 4354
8e73f275
CM		 4355 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4356again:
4357 level = 1;
4358 next = NULL;
925baedd 4359 btrfs_release_path(root, path);
8e73f275 4360
a2135011 4361 path->keep_locks = 1;
8e73f275
CM		 4362
4363 if (!force_blocking)
4364 path->leave_spinning = 1;
4365
925baedd
CM		 4366 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4367 path->keep_locks = 0;
4368
4369 if (ret < 0)
4370 return ret;
4371
a2135011 4372 nritems = btrfs_header_nritems(path->nodes[0]);
168fd7d2
CM		 4373 /*
4374 * by releasing the path above we dropped all our locks. A balance
4375 * could have added more items next to the key that used to be
4376 * at the very end of the block. So, check again here and
4377 * advance the path if there are now more items available.
4378 */
a2135011 4379 if (nritems > 0 && path->slots[0] < nritems - 1) {
e457afec
YZ		 4380 if (ret == 0)
4381 path->slots[0]++;
8e73f275 4382 ret = 0;
925baedd
CM		 4383 goto done;
4384 }
d97e63b6 4385
d397712b 4386 while (level < BTRFS_MAX_LEVEL) {
8e73f275
CM		 4387 if (!path->nodes[level]) {
4388 ret = 1;
4389 goto done;
4390 }
5f39d397 4391
d97e63b6
CM		 4392 slot = path->slots[level] + 1;
4393 c = path->nodes[level];
5f39d397 4394 if (slot >= btrfs_header_nritems(c)) {
d97e63b6 4395 level++;
8e73f275
CM		 4396 if (level == BTRFS_MAX_LEVEL) {
4397 ret = 1;
4398 goto done;
4399 }
d97e63b6
CM		 4400 continue;
4401 }
5f39d397 4402
925baedd
CM		 4403 if (next) {
4404 btrfs_tree_unlock(next);
5f39d397 4405 free_extent_buffer(next);
925baedd 4406 }
5f39d397 4407
8e73f275
CM		 4408 next = c;
4409 ret = read_block_for_search(NULL, root, path, &next, level,
4410 slot, &key);
4411 if (ret == -EAGAIN)
4412 goto again;
5f39d397 4413
76a05b35
CM		 4414 if (ret < 0) {
4415 btrfs_release_path(root, path);
4416 goto done;
4417 }
4418
5cd57b2c 4419 if (!path->skip_locking) {
8e73f275
CM		 4420 ret = btrfs_try_spin_lock(next);
4421 if (!ret) {
4422 btrfs_set_path_blocking(path);
4423 btrfs_tree_lock(next);
4424 if (!force_blocking)
4425 btrfs_clear_path_blocking(path, next);
4426 }
4427 if (force_blocking)
4428 btrfs_set_lock_blocking(next);
5cd57b2c 4429 }
d97e63b6
CM		 4430 break;
4431 }
4432 path->slots[level] = slot;
d397712b 4433 while (1) {
d97e63b6
CM		 4434 level--;
4435 c = path->nodes[level];
925baedd
CM		 4436 if (path->locks[level])
4437 btrfs_tree_unlock(c);
8e73f275 4438
5f39d397 4439 free_extent_buffer(c);
d97e63b6
CM		 4440 path->nodes[level] = next;
4441 path->slots[level] = 0;
a74a4b97
CM		 4442 if (!path->skip_locking)
4443 path->locks[level] = 1;
8e73f275 4444
d97e63b6
CM		 4445 if (!level)
4446 break;
b4ce94de 4447
8e73f275
CM		 4448 ret = read_block_for_search(NULL, root, path, &next, level,
4449 0, &key);
4450 if (ret == -EAGAIN)
4451 goto again;
4452
76a05b35
CM		 4453 if (ret < 0) {
4454 btrfs_release_path(root, path);
4455 goto done;
4456 }
4457
5cd57b2c 4458 if (!path->skip_locking) {
b9447ef8 4459 btrfs_assert_tree_locked(path->nodes[level]);
8e73f275
CM		 4460 ret = btrfs_try_spin_lock(next);
4461 if (!ret) {
4462 btrfs_set_path_blocking(path);
4463 btrfs_tree_lock(next);
4464 if (!force_blocking)
4465 btrfs_clear_path_blocking(path, next);
4466 }
4467 if (force_blocking)
4468 btrfs_set_lock_blocking(next);
5cd57b2c 4469 }
d97e63b6 4470 }
8e73f275 4471 ret = 0;
925baedd
CM		 4472done:
4473 unlock_up(path, 0, 1);
8e73f275
CM		 4474 path->leave_spinning = old_spinning;
4475 if (!old_spinning)
4476 btrfs_set_path_blocking(path);
4477
4478 return ret;
d97e63b6 4479}
0b86a832 4480
3f157a2f
CM		 4481/*
4482 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4483 * searching until it gets past min_objectid or finds an item of 'type'
4484 *
4485 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4486 */
0b86a832
CM		 4487int btrfs_previous_item(struct btrfs_root *root,
4488 struct btrfs_path *path, u64 min_objectid,
4489 int type)
4490{
4491 struct btrfs_key found_key;
4492 struct extent_buffer *leaf;
e02119d5 4493 u32 nritems;
0b86a832
CM		 4494 int ret;
4495
d397712b 4496 while (1) {
0b86a832 4497 if (path->slots[0] == 0) {
b4ce94de 4498 btrfs_set_path_blocking(path);
0b86a832
CM		 4499 ret = btrfs_prev_leaf(root, path);
4500 if (ret != 0)
4501 return ret;
4502 } else {
4503 path->slots[0]--;
4504 }
4505 leaf = path->nodes[0];
e02119d5
CM		 4506 nritems = btrfs_header_nritems(leaf);
4507 if (nritems == 0)
4508 return 1;
4509 if (path->slots[0] == nritems)
4510 path->slots[0]--;
4511
0b86a832 4512 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
e02119d5
CM		 4513 if (found_key.objectid < min_objectid)
4514 break;
0a4eefbb
YZ		 4515 if (found_key.type == type)
4516 return 0;
e02119d5
CM		 4517 if (found_key.objectid == min_objectid &&
4518 found_key.type < type)
4519 break;
0b86a832
CM		 4520 }
4521 return 1;
4522}
kernel source for telekom puls (alcatel/mediatek)