]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/btrfs/ctree.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Kill init_btrfs_i()
[net-next-2.6.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{
df24a2b9
CM		 108 btrfs_release_path(NULL, p);
109 kmem_cache_free(btrfs_path_cachep, p);
be0e5c09
CM		 110}
111
d352ac68
CM		 112/*
113 * path release drops references on the extent buffers in the path
114 * and it drops any locks held by this path
115 *
116 * It is safe to call this on paths that no locks or extent buffers held.
117 */
d397712b 118noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
eb60ceac
CM		 119{
120 int i;
a2135011 121
234b63a0 122 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3f157a2f 123 p->slots[i] = 0;
eb60ceac 124 if (!p->nodes[i])
925baedd
CM		 125 continue;
126 if (p->locks[i]) {
127 btrfs_tree_unlock(p->nodes[i]);
128 p->locks[i] = 0;
129 }
5f39d397 130 free_extent_buffer(p->nodes[i]);
3f157a2f 131 p->nodes[i] = NULL;
eb60ceac
CM		 132 }
133}
134
d352ac68
CM		 135/*
136 * safely gets a reference on the root node of a tree. A lock
137 * is not taken, so a concurrent writer may put a different node
138 * at the root of the tree. See btrfs_lock_root_node for the
139 * looping required.
140 *
141 * The extent buffer returned by this has a reference taken, so
142 * it won't disappear. It may stop being the root of the tree
143 * at any time because there are no locks held.
144 */
925baedd
CM		 145struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
146{
147 struct extent_buffer *eb;
148 spin_lock(&root->node_lock);
149 eb = root->node;
150 extent_buffer_get(eb);
151 spin_unlock(&root->node_lock);
152 return eb;
153}
154
d352ac68
CM		 155/* loop around taking references on and locking the root node of the
156 * tree until you end up with a lock on the root. A locked buffer
157 * is returned, with a reference held.
158 */
925baedd
CM		 159struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
160{
161 struct extent_buffer *eb;
162
d397712b 163 while (1) {
925baedd
CM		 164 eb = btrfs_root_node(root);
165 btrfs_tree_lock(eb);
166
167 spin_lock(&root->node_lock);
168 if (eb == root->node) {
169 spin_unlock(&root->node_lock);
170 break;
171 }
172 spin_unlock(&root->node_lock);
173
174 btrfs_tree_unlock(eb);
175 free_extent_buffer(eb);
176 }
177 return eb;
178}
179
d352ac68
CM		 180/* cowonly root (everything not a reference counted cow subvolume), just get
181 * put onto a simple dirty list. transaction.c walks this to make sure they
182 * get properly updated on disk.
183 */
0b86a832
CM		 184static void add_root_to_dirty_list(struct btrfs_root *root)
185{
186 if (root->track_dirty && list_empty(&root->dirty_list)) {
187 list_add(&root->dirty_list,
188 &root->fs_info->dirty_cowonly_roots);
189 }
190}
191
d352ac68
CM		 192/*
193 * used by snapshot creation to make a copy of a root for a tree with
194 * a given objectid. The buffer with the new root node is returned in
195 * cow_ret, and this func returns zero on success or a negative error code.
196 */
be20aa9d
CM		 197int btrfs_copy_root(struct btrfs_trans_handle *trans,
198 struct btrfs_root *root,
199 struct extent_buffer *buf,
200 struct extent_buffer **cow_ret, u64 new_root_objectid)
201{
202 struct extent_buffer *cow;
203 u32 nritems;
204 int ret = 0;
205 int level;
5d4f98a2 206 struct btrfs_disk_key disk_key;
be20aa9d
CM		 207
208 WARN_ON(root->ref_cows && trans->transid !=
209 root->fs_info->running_transaction->transid);
210 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
211
212 level = btrfs_header_level(buf);
213 nritems = btrfs_header_nritems(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,
283 struct extent_buffer *cow)
284{
285 u64 refs;
286 u64 owner;
287 u64 flags;
288 u64 new_flags = 0;
289 int ret;
290
291 /*
292 * Backrefs update rules:
293 *
294 * Always use full backrefs for extent pointers in tree block
295 * allocated by tree relocation.
296 *
297 * If a shared tree block is no longer referenced by its owner
298 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
299 * use full backrefs for extent pointers in tree block.
300 *
301 * If a tree block is been relocating
302 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
303 * use full backrefs for extent pointers in tree block.
304 * The reason for this is some operations (such as drop tree)
305 * are only allowed for blocks use full backrefs.
306 */
307
308 if (btrfs_block_can_be_shared(root, buf)) {
309 ret = btrfs_lookup_extent_info(trans, root, buf->start,
310 buf->len, &refs, &flags);
311 BUG_ON(ret);
312 BUG_ON(refs == 0);
313 } else {
314 refs = 1;
315 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
316 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
317 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
318 else
319 flags = 0;
320 }
321
322 owner = btrfs_header_owner(buf);
323 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
324 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
325
326 if (refs > 1) {
327 if ((owner == root->root_key.objectid ||
328 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
329 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
330 ret = btrfs_inc_ref(trans, root, buf, 1);
331 BUG_ON(ret);
332
333 if (root->root_key.objectid ==
334 BTRFS_TREE_RELOC_OBJECTID) {
335 ret = btrfs_dec_ref(trans, root, buf, 0);
336 BUG_ON(ret);
337 ret = btrfs_inc_ref(trans, root, cow, 1);
338 BUG_ON(ret);
339 }
340 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
341 } else {
342
343 if (root->root_key.objectid ==
344 BTRFS_TREE_RELOC_OBJECTID)
345 ret = btrfs_inc_ref(trans, root, cow, 1);
346 else
347 ret = btrfs_inc_ref(trans, root, cow, 0);
348 BUG_ON(ret);
349 }
350 if (new_flags != 0) {
351 ret = btrfs_set_disk_extent_flags(trans, root,
352 buf->start,
353 buf->len,
354 new_flags, 0);
355 BUG_ON(ret);
356 }
357 } else {
358 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
359 if (root->root_key.objectid ==
360 BTRFS_TREE_RELOC_OBJECTID)
361 ret = btrfs_inc_ref(trans, root, cow, 1);
362 else
363 ret = btrfs_inc_ref(trans, root, cow, 0);
364 BUG_ON(ret);
365 ret = btrfs_dec_ref(trans, root, buf, 1);
366 BUG_ON(ret);
367 }
368 clean_tree_block(trans, root, buf);
369 }
370 return 0;
371}
372
d352ac68 373/*
d397712b
CM		 374 * does the dirty work in cow of a single block. The parent block (if
375 * supplied) is updated to point to the new cow copy. The new buffer is marked
376 * dirty and returned locked. If you modify the block it needs to be marked
377 * dirty again.
d352ac68
CM		 378 *
379 * search_start -- an allocation hint for the new block
380 *
d397712b
CM		 381 * empty_size -- a hint that you plan on doing more cow. This is the size in
382 * bytes the allocator should try to find free next to the block it returns.
383 * This is just a hint and may be ignored by the allocator.
d352ac68 384 */
d397712b 385static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
5f39d397
CM		 386 struct btrfs_root *root,
387 struct extent_buffer *buf,
388 struct extent_buffer *parent, int parent_slot,
389 struct extent_buffer **cow_ret,
9fa8cfe7 390 u64 search_start, u64 empty_size)
02217ed2 391{
5d4f98a2 392 struct btrfs_disk_key disk_key;
5f39d397 393 struct extent_buffer *cow;
7bb86316 394 int level;
925baedd 395 int unlock_orig = 0;
5d4f98a2 396 u64 parent_start;
7bb86316 397
925baedd
CM		 398 if (*cow_ret == buf)
399 unlock_orig = 1;
400
b9447ef8 401 btrfs_assert_tree_locked(buf);
925baedd 402
7bb86316
CM		 403 WARN_ON(root->ref_cows && trans->transid !=
404 root->fs_info->running_transaction->transid);
6702ed49 405 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
5f39d397 406
7bb86316 407 level = btrfs_header_level(buf);
31840ae1 408
5d4f98a2
YZ		 409 if (level == 0)
410 btrfs_item_key(buf, &disk_key, 0);
411 else
412 btrfs_node_key(buf, &disk_key, 0);
413
414 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
415 if (parent)
416 parent_start = parent->start;
417 else
418 parent_start = 0;
419 } else
420 parent_start = 0;
421
422 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
423 root->root_key.objectid, &disk_key,
424 level, search_start, empty_size);
54aa1f4d
CM		 425 if (IS_ERR(cow))
426 return PTR_ERR(cow);
6702ed49 427
b4ce94de
CM		 428 /* cow is set to blocking by btrfs_init_new_buffer */
429
5f39d397 430 copy_extent_buffer(cow, buf, 0, 0, cow->len);
db94535d 431 btrfs_set_header_bytenr(cow, cow->start);
5f39d397 432 btrfs_set_header_generation(cow, trans->transid);
5d4f98a2
YZ		 433 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
434 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
435 BTRFS_HEADER_FLAG_RELOC);
436 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
437 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
438 else
439 btrfs_set_header_owner(cow, root->root_key.objectid);
6702ed49 440
2b82032c
YZ		 441 write_extent_buffer(cow, root->fs_info->fsid,
442 (unsigned long)btrfs_header_fsid(cow),
443 BTRFS_FSID_SIZE);
444
5d4f98a2 445 update_ref_for_cow(trans, root, buf, cow);
1a40e23b 446
02217ed2 447 if (buf == root->node) {
925baedd 448 WARN_ON(parent && parent != buf);
5d4f98a2
YZ		 449 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
450 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
451 parent_start = buf->start;
452 else
453 parent_start = 0;
925baedd
CM		 454
455 spin_lock(&root->node_lock);
02217ed2 456 root->node = cow;
5f39d397 457 extent_buffer_get(cow);
925baedd
CM		 458 spin_unlock(&root->node_lock);
459
86b9f2ec
YZ		 460 btrfs_free_tree_block(trans, root, buf->start, buf->len,
461 parent_start, root->root_key.objectid, level);
5f39d397 462 free_extent_buffer(buf);
0b86a832 463 add_root_to_dirty_list(root);
02217ed2 464 } else {
5d4f98a2
YZ		 465 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
466 parent_start = parent->start;
467 else
468 parent_start = 0;
469
470 WARN_ON(trans->transid != btrfs_header_generation(parent));
5f39d397 471 btrfs_set_node_blockptr(parent, parent_slot,
db94535d 472 cow->start);
74493f7a
CM		 473 btrfs_set_node_ptr_generation(parent, parent_slot,
474 trans->transid);
d6025579 475 btrfs_mark_buffer_dirty(parent);
86b9f2ec
YZ		 476 btrfs_free_tree_block(trans, root, buf->start, buf->len,
477 parent_start, root->root_key.objectid, level);
02217ed2 478 }
925baedd
CM		 479 if (unlock_orig)
480 btrfs_tree_unlock(buf);
5f39d397 481 free_extent_buffer(buf);
ccd467d6 482 btrfs_mark_buffer_dirty(cow);
2c90e5d6 483 *cow_ret = cow;
02217ed2
CM		 484 return 0;
485}
486
5d4f98a2
YZ		 487static inline int should_cow_block(struct btrfs_trans_handle *trans,
488 struct btrfs_root *root,
489 struct extent_buffer *buf)
490{
491 if (btrfs_header_generation(buf) == trans->transid &&
492 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
493 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
494 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
495 return 0;
496 return 1;
497}
498
d352ac68
CM		 499/*
500 * cows a single block, see __btrfs_cow_block for the real work.
501 * This version of it has extra checks so that a block isn't cow'd more than
502 * once per transaction, as long as it hasn't been written yet
503 */
d397712b 504noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
5f39d397
CM		 505 struct btrfs_root *root, struct extent_buffer *buf,
506 struct extent_buffer *parent, int parent_slot,
9fa8cfe7 507 struct extent_buffer **cow_ret)
6702ed49
CM		 508{
509 u64 search_start;
f510cfec 510 int ret;
dc17ff8f 511
6702ed49 512 if (trans->transaction != root->fs_info->running_transaction) {
d397712b
CM		 513 printk(KERN_CRIT "trans %llu running %llu\n",
514 (unsigned long long)trans->transid,
515 (unsigned long long)
6702ed49
CM		 516 root->fs_info->running_transaction->transid);
517 WARN_ON(1);
518 }
519 if (trans->transid != root->fs_info->generation) {
d397712b
CM		 520 printk(KERN_CRIT "trans %llu running %llu\n",
521 (unsigned long long)trans->transid,
522 (unsigned long long)root->fs_info->generation);
6702ed49
CM		 523 WARN_ON(1);
524 }
dc17ff8f 525
5d4f98a2 526 if (!should_cow_block(trans, root, buf)) {
6702ed49
CM		 527 *cow_ret = buf;
528 return 0;
529 }
c487685d 530
0b86a832 531 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
b4ce94de
CM		 532
533 if (parent)
534 btrfs_set_lock_blocking(parent);
535 btrfs_set_lock_blocking(buf);
536
f510cfec 537 ret = __btrfs_cow_block(trans, root, buf, parent,
9fa8cfe7 538 parent_slot, cow_ret, search_start, 0);
f510cfec 539 return ret;
6702ed49
CM		 540}
541
d352ac68
CM		 542/*
543 * helper function for defrag to decide if two blocks pointed to by a
544 * node are actually close by
545 */
6b80053d 546static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
6702ed49 547{
6b80053d 548 if (blocknr < other && other - (blocknr + blocksize) < 32768)
6702ed49 549 return 1;
6b80053d 550 if (blocknr > other && blocknr - (other + blocksize) < 32768)
6702ed49
CM		 551 return 1;
552 return 0;
553}
554
081e9573
CM		 555/*
556 * compare two keys in a memcmp fashion
557 */
558static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
559{
560 struct btrfs_key k1;
561
562 btrfs_disk_key_to_cpu(&k1, disk);
563
20736aba 564 return btrfs_comp_cpu_keys(&k1, k2);
081e9573
CM		 565}
566
f3465ca4
JB		 567/*
568 * same as comp_keys only with two btrfs_key's
569 */
5d4f98a2 570int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
f3465ca4
JB		 571{
572 if (k1->objectid > k2->objectid)
573 return 1;
574 if (k1->objectid < k2->objectid)
575 return -1;
576 if (k1->type > k2->type)
577 return 1;
578 if (k1->type < k2->type)
579 return -1;
580 if (k1->offset > k2->offset)
581 return 1;
582 if (k1->offset < k2->offset)
583 return -1;
584 return 0;
585}
081e9573 586
d352ac68
CM		 587/*
588 * this is used by the defrag code to go through all the
589 * leaves pointed to by a node and reallocate them so that
590 * disk order is close to key order
591 */
6702ed49 592int btrfs_realloc_node(struct btrfs_trans_handle *trans,
5f39d397 593 struct btrfs_root *root, struct extent_buffer *parent,
a6b6e75e
CM		 594 int start_slot, int cache_only, u64 *last_ret,
595 struct btrfs_key *progress)
6702ed49 596{
6b80053d 597 struct extent_buffer *cur;
6702ed49 598 u64 blocknr;
ca7a79ad 599 u64 gen;
e9d0b13b
CM		 600 u64 search_start = *last_ret;
601 u64 last_block = 0;
6702ed49
CM		 602 u64 other;
603 u32 parent_nritems;
6702ed49
CM		 604 int end_slot;
605 int i;
606 int err = 0;
f2183bde 607 int parent_level;
6b80053d
CM		 608 int uptodate;
609 u32 blocksize;
081e9573
CM		 610 int progress_passed = 0;
611 struct btrfs_disk_key disk_key;
6702ed49 612
5708b959
CM		 613 parent_level = btrfs_header_level(parent);
614 if (cache_only && parent_level != 1)
615 return 0;
616
d397712b 617 if (trans->transaction != root->fs_info->running_transaction)
6702ed49 618 WARN_ON(1);
d397712b 619 if (trans->transid != root->fs_info->generation)
6702ed49 620 WARN_ON(1);
86479a04 621
6b80053d 622 parent_nritems = btrfs_header_nritems(parent);
6b80053d 623 blocksize = btrfs_level_size(root, parent_level - 1);
6702ed49
CM		 624 end_slot = parent_nritems;
625
626 if (parent_nritems == 1)
627 return 0;
628
b4ce94de
CM		 629 btrfs_set_lock_blocking(parent);
630
6702ed49
CM		 631 for (i = start_slot; i < end_slot; i++) {
632 int close = 1;
a6b6e75e 633
5708b959
CM		 634 if (!parent->map_token) {
635 map_extent_buffer(parent,
636 btrfs_node_key_ptr_offset(i),
637 sizeof(struct btrfs_key_ptr),
638 &parent->map_token, &parent->kaddr,
639 &parent->map_start, &parent->map_len,
640 KM_USER1);
641 }
081e9573
CM		 642 btrfs_node_key(parent, &disk_key, i);
643 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
644 continue;
645
646 progress_passed = 1;
6b80053d 647 blocknr = btrfs_node_blockptr(parent, i);
ca7a79ad 648 gen = btrfs_node_ptr_generation(parent, i);
e9d0b13b
CM		 649 if (last_block == 0)
650 last_block = blocknr;
5708b959 651
6702ed49 652 if (i > 0) {
6b80053d
CM		 653 other = btrfs_node_blockptr(parent, i - 1);
654 close = close_blocks(blocknr, other, blocksize);
6702ed49 655 }
0ef3e66b 656 if (!close && i < end_slot - 2) {
6b80053d
CM		 657 other = btrfs_node_blockptr(parent, i + 1);
658 close = close_blocks(blocknr, other, blocksize);
6702ed49 659 }
e9d0b13b
CM		 660 if (close) {
661 last_block = blocknr;
6702ed49 662 continue;
e9d0b13b 663 }
5708b959
CM		 664 if (parent->map_token) {
665 unmap_extent_buffer(parent, parent->map_token,
666 KM_USER1);
667 parent->map_token = NULL;
668 }
6702ed49 669
6b80053d
CM		 670 cur = btrfs_find_tree_block(root, blocknr, blocksize);
671 if (cur)
1259ab75 672 uptodate = btrfs_buffer_uptodate(cur, gen);
6b80053d
CM		 673 else
674 uptodate = 0;
5708b959 675 if (!cur || !uptodate) {
6702ed49 676 if (cache_only) {
6b80053d 677 free_extent_buffer(cur);
6702ed49
CM		 678 continue;
679 }
6b80053d
CM		 680 if (!cur) {
681 cur = read_tree_block(root, blocknr,
ca7a79ad 682 blocksize, gen);
6b80053d 683 } else if (!uptodate) {
ca7a79ad 684 btrfs_read_buffer(cur, gen);
f2183bde 685 }
6702ed49 686 }
e9d0b13b 687 if (search_start == 0)
6b80053d 688 search_start = last_block;
e9d0b13b 689
e7a84565 690 btrfs_tree_lock(cur);
b4ce94de 691 btrfs_set_lock_blocking(cur);
6b80053d 692 err = __btrfs_cow_block(trans, root, cur, parent, i,
e7a84565 693 &cur, search_start,
6b80053d 694 min(16 * blocksize,
9fa8cfe7 695 (end_slot - i) * blocksize));
252c38f0 696 if (err) {
e7a84565 697 btrfs_tree_unlock(cur);
6b80053d 698 free_extent_buffer(cur);
6702ed49 699 break;
252c38f0 700 }
e7a84565
CM		 701 search_start = cur->start;
702 last_block = cur->start;
f2183bde 703 *last_ret = search_start;
e7a84565
CM		 704 btrfs_tree_unlock(cur);
705 free_extent_buffer(cur);
6702ed49 706 }
5708b959
CM		 707 if (parent->map_token) {
708 unmap_extent_buffer(parent, parent->map_token,
709 KM_USER1);
710 parent->map_token = NULL;
711 }
6702ed49
CM		 712 return err;
713}
714
74123bd7
CM		 715/*
716 * The leaf data grows from end-to-front in the node.
717 * this returns the address of the start of the last item,
718 * which is the stop of the leaf data stack
719 */
123abc88 720static inline unsigned int leaf_data_end(struct btrfs_root *root,
5f39d397 721 struct extent_buffer *leaf)
be0e5c09 722{
5f39d397 723 u32 nr = btrfs_header_nritems(leaf);
be0e5c09 724 if (nr == 0)
123abc88 725 return BTRFS_LEAF_DATA_SIZE(root);
5f39d397 726 return btrfs_item_offset_nr(leaf, nr - 1);
be0e5c09
CM		 727}
728
d352ac68
CM		 729/*
730 * extra debugging checks to make sure all the items in a key are
731 * well formed and in the proper order
732 */
123abc88
CM		 733static int check_node(struct btrfs_root *root, struct btrfs_path *path,
734 int level)
aa5d6bed 735{
5f39d397
CM		 736 struct extent_buffer *parent = NULL;
737 struct extent_buffer *node = path->nodes[level];
738 struct btrfs_disk_key parent_key;
739 struct btrfs_disk_key node_key;
aa5d6bed 740 int parent_slot;
8d7be552
CM		 741 int slot;
742 struct btrfs_key cpukey;
5f39d397 743 u32 nritems = btrfs_header_nritems(node);
aa5d6bed
CM		 744
745 if (path->nodes[level + 1])
5f39d397 746 parent = path->nodes[level + 1];
a1f39630 747
8d7be552 748 slot = path->slots[level];
7518a238
CM		 749 BUG_ON(nritems == 0);
750 if (parent) {
a1f39630 751 parent_slot = path->slots[level + 1];
5f39d397
CM		 752 btrfs_node_key(parent, &parent_key, parent_slot);
753 btrfs_node_key(node, &node_key, 0);
754 BUG_ON(memcmp(&parent_key, &node_key,
e2fa7227 755 sizeof(struct btrfs_disk_key)));
1d4f8a0c 756 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
db94535d 757 btrfs_header_bytenr(node));
aa5d6bed 758 }
123abc88 759 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
8d7be552 760 if (slot != 0) {
5f39d397
CM		 761 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
762 btrfs_node_key(node, &node_key, slot);
763 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
8d7be552
CM		 764 }
765 if (slot < nritems - 1) {
5f39d397
CM		 766 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
767 btrfs_node_key(node, &node_key, slot);
768 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
aa5d6bed
CM		 769 }
770 return 0;
771}
772
d352ac68
CM		 773/*
774 * extra checking to make sure all the items in a leaf are
775 * well formed and in the proper order
776 */
123abc88
CM		 777static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
778 int level)
aa5d6bed 779{
5f39d397
CM		 780 struct extent_buffer *leaf = path->nodes[level];
781 struct extent_buffer *parent = NULL;
aa5d6bed 782 int parent_slot;
8d7be552 783 struct btrfs_key cpukey;
5f39d397
CM		 784 struct btrfs_disk_key parent_key;
785 struct btrfs_disk_key leaf_key;
786 int slot = path->slots[0];
8d7be552 787
5f39d397 788 u32 nritems = btrfs_header_nritems(leaf);
aa5d6bed
CM		 789
790 if (path->nodes[level + 1])
5f39d397 791 parent = path->nodes[level + 1];
7518a238
CM		 792
793 if (nritems == 0)
794 return 0;
795
796 if (parent) {
a1f39630 797 parent_slot = path->slots[level + 1];
5f39d397
CM		 798 btrfs_node_key(parent, &parent_key, parent_slot);
799 btrfs_item_key(leaf, &leaf_key, 0);
6702ed49 800
5f39d397 801 BUG_ON(memcmp(&parent_key, &leaf_key,
e2fa7227 802 sizeof(struct btrfs_disk_key)));
1d4f8a0c 803 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
db94535d 804 btrfs_header_bytenr(leaf));
5f39d397 805 }
5f39d397
CM		 806 if (slot != 0 && slot < nritems - 1) {
807 btrfs_item_key(leaf, &leaf_key, slot);
808 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
809 if (comp_keys(&leaf_key, &cpukey) <= 0) {
810 btrfs_print_leaf(root, leaf);
d397712b 811 printk(KERN_CRIT "slot %d offset bad key\n", slot);
5f39d397
CM		 812 BUG_ON(1);
813 }
814 if (btrfs_item_offset_nr(leaf, slot - 1) !=
815 btrfs_item_end_nr(leaf, slot)) {
816 btrfs_print_leaf(root, leaf);
d397712b 817 printk(KERN_CRIT "slot %d offset bad\n", slot);
5f39d397
CM		 818 BUG_ON(1);
819 }
8d7be552
CM		 820 }
821 if (slot < nritems - 1) {
5f39d397
CM		 822 btrfs_item_key(leaf, &leaf_key, slot);
823 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
824 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
825 if (btrfs_item_offset_nr(leaf, slot) !=
826 btrfs_item_end_nr(leaf, slot + 1)) {
827 btrfs_print_leaf(root, leaf);
d397712b 828 printk(KERN_CRIT "slot %d offset bad\n", slot);
5f39d397
CM		 829 BUG_ON(1);
830 }
aa5d6bed 831 }
5f39d397
CM		 832 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
833 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
aa5d6bed
CM		 834 return 0;
835}
836
d397712b 837static noinline int check_block(struct btrfs_root *root,
98ed5174 838 struct btrfs_path *path, int level)
aa5d6bed 839{
85d824c4 840 return 0;
aa5d6bed 841 if (level == 0)
123abc88
CM		 842 return check_leaf(root, path, level);
843 return check_node(root, path, level);
aa5d6bed
CM		 844}
845
74123bd7 846/*
5f39d397
CM		 847 * search for key in the extent_buffer. The items start at offset p,
848 * and they are item_size apart. There are 'max' items in p.
849 *
74123bd7
CM		 850 * the slot in the array is returned via slot, and it points to
851 * the place where you would insert key if it is not found in
852 * the array.
853 *
854 * slot may point to max if the key is bigger than all of the keys
855 */
e02119d5
CM		 856static noinline int generic_bin_search(struct extent_buffer *eb,
857 unsigned long p,
858 int item_size, struct btrfs_key *key,
859 int max, int *slot)
be0e5c09
CM		 860{
861 int low = 0;
862 int high = max;
863 int mid;
864 int ret;
479965d6 865 struct btrfs_disk_key *tmp = NULL;
5f39d397
CM		 866 struct btrfs_disk_key unaligned;
867 unsigned long offset;
868 char *map_token = NULL;
869 char *kaddr = NULL;
870 unsigned long map_start = 0;
871 unsigned long map_len = 0;
479965d6 872 int err;
be0e5c09 873
d397712b 874 while (low < high) {
be0e5c09 875 mid = (low + high) / 2;
5f39d397
CM		 876 offset = p + mid * item_size;
877
878 if (!map_token || offset < map_start ||
879 (offset + sizeof(struct btrfs_disk_key)) >
880 map_start + map_len) {
479965d6 881 if (map_token) {
5f39d397 882 unmap_extent_buffer(eb, map_token, KM_USER0);
479965d6
CM		 883 map_token = NULL;
884 }
934d375b
CM		 885
886 err = map_private_extent_buffer(eb, offset,
479965d6
CM		 887 sizeof(struct btrfs_disk_key),
888 &map_token, &kaddr,
889 &map_start, &map_len, KM_USER0);
890
891 if (!err) {
892 tmp = (struct btrfs_disk_key *)(kaddr + offset -
893 map_start);
894 } else {
895 read_extent_buffer(eb, &unaligned,
896 offset, sizeof(unaligned));
897 tmp = &unaligned;
898 }
5f39d397 899
5f39d397
CM		 900 } else {
901 tmp = (struct btrfs_disk_key *)(kaddr + offset -
902 map_start);
903 }
be0e5c09
CM		 904 ret = comp_keys(tmp, key);
905
906 if (ret < 0)
907 low = mid + 1;
908 else if (ret > 0)
909 high = mid;
910 else {
911 *slot = mid;
479965d6
CM		 912 if (map_token)
913 unmap_extent_buffer(eb, map_token, KM_USER0);
be0e5c09
CM		 914 return 0;
915 }
916 }
917 *slot = low;
5f39d397
CM		 918 if (map_token)
919 unmap_extent_buffer(eb, map_token, KM_USER0);
be0e5c09
CM		 920 return 1;
921}
922
97571fd0
CM		 923/*
924 * simple bin_search frontend that does the right thing for
925 * leaves vs nodes
926 */
5f39d397
CM		 927static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
928 int level, int *slot)
be0e5c09 929{
5f39d397
CM		 930 if (level == 0) {
931 return generic_bin_search(eb,
932 offsetof(struct btrfs_leaf, items),
0783fcfc 933 sizeof(struct btrfs_item),
5f39d397 934 key, btrfs_header_nritems(eb),
7518a238 935 slot);
be0e5c09 936 } else {
5f39d397
CM		 937 return generic_bin_search(eb,
938 offsetof(struct btrfs_node, ptrs),
123abc88 939 sizeof(struct btrfs_key_ptr),
5f39d397 940 key, btrfs_header_nritems(eb),
7518a238 941 slot);
be0e5c09
CM		 942 }
943 return -1;
944}
945
5d4f98a2
YZ		 946int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
947 int level, int *slot)
948{
949 return bin_search(eb, key, level, slot);
950}
951
d352ac68
CM		 952/* given a node and slot number, this reads the blocks it points to. The
953 * extent buffer is returned with a reference taken (but unlocked).
954 * NULL is returned on error.
955 */
e02119d5 956static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
5f39d397 957 struct extent_buffer *parent, int slot)
bb803951 958{
ca7a79ad 959 int level = btrfs_header_level(parent);
bb803951
CM		 960 if (slot < 0)
961 return NULL;
5f39d397 962 if (slot >= btrfs_header_nritems(parent))
bb803951 963 return NULL;
ca7a79ad
CM		 964
965 BUG_ON(level == 0);
966
db94535d 967 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
ca7a79ad
CM		 968 btrfs_level_size(root, level - 1),
969 btrfs_node_ptr_generation(parent, slot));
bb803951
CM		 970}
971
d352ac68
CM		 972/*
973 * node level balancing, used to make sure nodes are in proper order for
974 * item deletion. We balance from the top down, so we have to make sure
975 * that a deletion won't leave an node completely empty later on.
976 */
e02119d5 977static noinline int balance_level(struct btrfs_trans_handle *trans,
98ed5174
CM		 978 struct btrfs_root *root,
979 struct btrfs_path *path, int level)
bb803951 980{
5f39d397
CM		 981 struct extent_buffer *right = NULL;
982 struct extent_buffer *mid;
983 struct extent_buffer *left = NULL;
984 struct extent_buffer *parent = NULL;
bb803951
CM		 985 int ret = 0;
986 int wret;
987 int pslot;
bb803951 988 int orig_slot = path->slots[level];
54aa1f4d 989 int err_on_enospc = 0;
79f95c82 990 u64 orig_ptr;
bb803951
CM		 991
992 if (level == 0)
993 return 0;
994
5f39d397 995 mid = path->nodes[level];
b4ce94de 996
925baedd 997 WARN_ON(!path->locks[level]);
7bb86316
CM		 998 WARN_ON(btrfs_header_generation(mid) != trans->transid);
999
1d4f8a0c 1000 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
79f95c82 1001
234b63a0 1002 if (level < BTRFS_MAX_LEVEL - 1)
5f39d397 1003 parent = path->nodes[level + 1];
bb803951
CM		 1004 pslot = path->slots[level + 1];
1005
40689478
CM		 1006 /*
1007 * deal with the case where there is only one pointer in the root
1008 * by promoting the node below to a root
1009 */
5f39d397
CM		 1010 if (!parent) {
1011 struct extent_buffer *child;
bb803951 1012
5f39d397 1013 if (btrfs_header_nritems(mid) != 1)
bb803951
CM		 1014 return 0;
1015
1016 /* promote the child to a root */
5f39d397 1017 child = read_node_slot(root, mid, 0);
7951f3ce 1018 BUG_ON(!child);
925baedd 1019 btrfs_tree_lock(child);
b4ce94de 1020 btrfs_set_lock_blocking(child);
9fa8cfe7 1021 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
2f375ab9
Y		 1022 BUG_ON(ret);
1023
925baedd 1024 spin_lock(&root->node_lock);
bb803951 1025 root->node = child;
925baedd
CM		 1026 spin_unlock(&root->node_lock);
1027
0b86a832 1028 add_root_to_dirty_list(root);
925baedd 1029 btrfs_tree_unlock(child);
b4ce94de 1030
925baedd 1031 path->locks[level] = 0;
bb803951 1032 path->nodes[level] = NULL;
5f39d397 1033 clean_tree_block(trans, root, mid);
925baedd 1034 btrfs_tree_unlock(mid);
bb803951 1035 /* once for the path */
5f39d397 1036 free_extent_buffer(mid);
86b9f2ec
YZ		 1037 ret = btrfs_free_tree_block(trans, root, mid->start, mid->len,
1038 0, root->root_key.objectid, level);
bb803951 1039 /* once for the root ptr */
5f39d397 1040 free_extent_buffer(mid);
db94535d 1041 return ret;
bb803951 1042 }
5f39d397 1043 if (btrfs_header_nritems(mid) >
123abc88 1044 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
bb803951
CM		 1045 return 0;
1046
5f39d397 1047 if (btrfs_header_nritems(mid) < 2)
54aa1f4d
CM		 1048 err_on_enospc = 1;
1049
5f39d397
CM		 1050 left = read_node_slot(root, parent, pslot - 1);
1051 if (left) {
925baedd 1052 btrfs_tree_lock(left);
b4ce94de 1053 btrfs_set_lock_blocking(left);
5f39d397 1054 wret = btrfs_cow_block(trans, root, left,
9fa8cfe7 1055 parent, pslot - 1, &left);
54aa1f4d
CM		 1056 if (wret) {
1057 ret = wret;
1058 goto enospc;
1059 }
2cc58cf2 1060 }
5f39d397
CM		 1061 right = read_node_slot(root, parent, pslot + 1);
1062 if (right) {
925baedd 1063 btrfs_tree_lock(right);
b4ce94de 1064 btrfs_set_lock_blocking(right);
5f39d397 1065 wret = btrfs_cow_block(trans, root, right,
9fa8cfe7 1066 parent, pslot + 1, &right);
2cc58cf2
CM		 1067 if (wret) {
1068 ret = wret;
1069 goto enospc;
1070 }
1071 }
1072
1073 /* first, try to make some room in the middle buffer */
5f39d397
CM		 1074 if (left) {
1075 orig_slot += btrfs_header_nritems(left);
bce4eae9 1076 wret = push_node_left(trans, root, left, mid, 1);
79f95c82
CM		 1077 if (wret < 0)
1078 ret = wret;
5f39d397 1079 if (btrfs_header_nritems(mid) < 2)
54aa1f4d 1080 err_on_enospc = 1;
bb803951 1081 }
79f95c82
CM		 1082
1083 /*
1084 * then try to empty the right most buffer into the middle
1085 */
5f39d397 1086 if (right) {
971a1f66 1087 wret = push_node_left(trans, root, mid, right, 1);
54aa1f4d 1088 if (wret < 0 && wret != -ENOSPC)
79f95c82 1089 ret = wret;
5f39d397 1090 if (btrfs_header_nritems(right) == 0) {
db94535d
CM		 1091 u64 bytenr = right->start;
1092 u32 blocksize = right->len;
1093
5f39d397 1094 clean_tree_block(trans, root, right);
925baedd 1095 btrfs_tree_unlock(right);
5f39d397 1096 free_extent_buffer(right);
bb803951 1097 right = NULL;
e089f05c
CM		 1098 wret = del_ptr(trans, root, path, level + 1, pslot +
1099 1);
bb803951
CM		 1100 if (wret)
1101 ret = wret;
86b9f2ec
YZ		 1102 wret = btrfs_free_tree_block(trans, root,
1103 bytenr, blocksize, 0,
1104 root->root_key.objectid,
1105 level);
bb803951
CM		 1106 if (wret)
1107 ret = wret;
1108 } else {
5f39d397
CM		 1109 struct btrfs_disk_key right_key;
1110 btrfs_node_key(right, &right_key, 0);
1111 btrfs_set_node_key(parent, &right_key, pslot + 1);
1112 btrfs_mark_buffer_dirty(parent);
bb803951
CM		 1113 }
1114 }
5f39d397 1115 if (btrfs_header_nritems(mid) == 1) {
79f95c82
CM		 1116 /*
1117 * we're not allowed to leave a node with one item in the
1118 * tree during a delete. A deletion from lower in the tree
1119 * could try to delete the only pointer in this node.
1120 * So, pull some keys from the left.
1121 * There has to be a left pointer at this point because
1122 * otherwise we would have pulled some pointers from the
1123 * right
1124 */
5f39d397
CM		 1125 BUG_ON(!left);
1126 wret = balance_node_right(trans, root, mid, left);
54aa1f4d 1127 if (wret < 0) {
79f95c82 1128 ret = wret;
54aa1f4d
CM		 1129 goto enospc;
1130 }
bce4eae9
CM		 1131 if (wret == 1) {
1132 wret = push_node_left(trans, root, left, mid, 1);
1133 if (wret < 0)
1134 ret = wret;
1135 }
79f95c82
CM		 1136 BUG_ON(wret == 1);
1137 }
5f39d397 1138 if (btrfs_header_nritems(mid) == 0) {
79f95c82 1139 /* we've managed to empty the middle node, drop it */
db94535d
CM		 1140 u64 bytenr = mid->start;
1141 u32 blocksize = mid->len;
925baedd 1142
5f39d397 1143 clean_tree_block(trans, root, mid);
925baedd 1144 btrfs_tree_unlock(mid);
5f39d397 1145 free_extent_buffer(mid);
bb803951 1146 mid = NULL;
e089f05c 1147 wret = del_ptr(trans, root, path, level + 1, pslot);
bb803951
CM		 1148 if (wret)
1149 ret = wret;
86b9f2ec
YZ		 1150 wret = btrfs_free_tree_block(trans, root, bytenr, blocksize,
1151 0, root->root_key.objectid, level);
bb803951
CM		 1152 if (wret)
1153 ret = wret;
79f95c82
CM		 1154 } else {
1155 /* update the parent key to reflect our changes */
5f39d397
CM		 1156 struct btrfs_disk_key mid_key;
1157 btrfs_node_key(mid, &mid_key, 0);
1158 btrfs_set_node_key(parent, &mid_key, pslot);
1159 btrfs_mark_buffer_dirty(parent);
79f95c82 1160 }
bb803951 1161
79f95c82 1162 /* update the path */
5f39d397
CM		 1163 if (left) {
1164 if (btrfs_header_nritems(left) > orig_slot) {
1165 extent_buffer_get(left);
925baedd 1166 /* left was locked after cow */
5f39d397 1167 path->nodes[level] = left;
bb803951
CM		 1168 path->slots[level + 1] -= 1;
1169 path->slots[level] = orig_slot;
925baedd
CM		 1170 if (mid) {
1171 btrfs_tree_unlock(mid);
5f39d397 1172 free_extent_buffer(mid);
925baedd 1173 }
bb803951 1174 } else {
5f39d397 1175 orig_slot -= btrfs_header_nritems(left);
bb803951
CM		 1176 path->slots[level] = orig_slot;
1177 }
1178 }
79f95c82 1179 /* double check we haven't messed things up */
123abc88 1180 check_block(root, path, level);
e20d96d6 1181 if (orig_ptr !=
5f39d397 1182 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
79f95c82 1183 BUG();
54aa1f4d 1184enospc:
925baedd
CM		 1185 if (right) {
1186 btrfs_tree_unlock(right);
5f39d397 1187 free_extent_buffer(right);
925baedd
CM		 1188 }
1189 if (left) {
1190 if (path->nodes[level] != left)
1191 btrfs_tree_unlock(left);
5f39d397 1192 free_extent_buffer(left);
925baedd 1193 }
bb803951
CM		 1194 return ret;
1195}
1196
d352ac68
CM		 1197/* Node balancing for insertion. Here we only split or push nodes around
1198 * when they are completely full. This is also done top down, so we
1199 * have to be pessimistic.
1200 */
d397712b 1201static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
98ed5174
CM		 1202 struct btrfs_root *root,
1203 struct btrfs_path *path, int level)
e66f709b 1204{
5f39d397
CM		 1205 struct extent_buffer *right = NULL;
1206 struct extent_buffer *mid;
1207 struct extent_buffer *left = NULL;
1208 struct extent_buffer *parent = NULL;
e66f709b
CM		 1209 int ret = 0;
1210 int wret;
1211 int pslot;
1212 int orig_slot = path->slots[level];
1213 u64 orig_ptr;
1214
1215 if (level == 0)
1216 return 1;
1217
5f39d397 1218 mid = path->nodes[level];
7bb86316 1219 WARN_ON(btrfs_header_generation(mid) != trans->transid);
e66f709b
CM		 1220 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1221
1222 if (level < BTRFS_MAX_LEVEL - 1)
5f39d397 1223 parent = path->nodes[level + 1];
e66f709b
CM		 1224 pslot = path->slots[level + 1];
1225
5f39d397 1226 if (!parent)
e66f709b 1227 return 1;
e66f709b 1228
5f39d397 1229 left = read_node_slot(root, parent, pslot - 1);
e66f709b
CM		 1230
1231 /* first, try to make some room in the middle buffer */
5f39d397 1232 if (left) {
e66f709b 1233 u32 left_nr;
925baedd
CM		 1234
1235 btrfs_tree_lock(left);
b4ce94de
CM		 1236 btrfs_set_lock_blocking(left);
1237
5f39d397 1238 left_nr = btrfs_header_nritems(left);
33ade1f8
CM		 1239 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1240 wret = 1;
1241 } else {
5f39d397 1242 ret = btrfs_cow_block(trans, root, left, parent,
9fa8cfe7 1243 pslot - 1, &left);
54aa1f4d
CM		 1244 if (ret)
1245 wret = 1;
1246 else {
54aa1f4d 1247 wret = push_node_left(trans, root,
971a1f66 1248 left, mid, 0);
54aa1f4d 1249 }
33ade1f8 1250 }
e66f709b
CM		 1251 if (wret < 0)
1252 ret = wret;
1253 if (wret == 0) {
5f39d397 1254 struct btrfs_disk_key disk_key;
e66f709b 1255 orig_slot += left_nr;
5f39d397
CM		 1256 btrfs_node_key(mid, &disk_key, 0);
1257 btrfs_set_node_key(parent, &disk_key, pslot);
1258 btrfs_mark_buffer_dirty(parent);
1259 if (btrfs_header_nritems(left) > orig_slot) {
1260 path->nodes[level] = left;
e66f709b
CM		 1261 path->slots[level + 1] -= 1;
1262 path->slots[level] = orig_slot;
925baedd 1263 btrfs_tree_unlock(mid);
5f39d397 1264 free_extent_buffer(mid);
e66f709b
CM		 1265 } else {
1266 orig_slot -=
5f39d397 1267 btrfs_header_nritems(left);
e66f709b 1268 path->slots[level] = orig_slot;
925baedd 1269 btrfs_tree_unlock(left);
5f39d397 1270 free_extent_buffer(left);
e66f709b 1271 }
e66f709b
CM		 1272 return 0;
1273 }
925baedd 1274 btrfs_tree_unlock(left);
5f39d397 1275 free_extent_buffer(left);
e66f709b 1276 }
925baedd 1277 right = read_node_slot(root, parent, pslot + 1);
e66f709b
CM		 1278
1279 /*
1280 * then try to empty the right most buffer into the middle
1281 */
5f39d397 1282 if (right) {
33ade1f8 1283 u32 right_nr;
b4ce94de 1284
925baedd 1285 btrfs_tree_lock(right);
b4ce94de
CM		 1286 btrfs_set_lock_blocking(right);
1287
5f39d397 1288 right_nr = btrfs_header_nritems(right);
33ade1f8
CM		 1289 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1290 wret = 1;
1291 } else {
5f39d397
CM		 1292 ret = btrfs_cow_block(trans, root, right,
1293 parent, pslot + 1,
9fa8cfe7 1294 &right);
54aa1f4d
CM		 1295 if (ret)
1296 wret = 1;
1297 else {
54aa1f4d 1298 wret = balance_node_right(trans, root,
5f39d397 1299 right, mid);
54aa1f4d 1300 }
33ade1f8 1301 }
e66f709b
CM		 1302 if (wret < 0)
1303 ret = wret;
1304 if (wret == 0) {
5f39d397
CM		 1305 struct btrfs_disk_key disk_key;
1306
1307 btrfs_node_key(right, &disk_key, 0);
1308 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1309 btrfs_mark_buffer_dirty(parent);
1310
1311 if (btrfs_header_nritems(mid) <= orig_slot) {
1312 path->nodes[level] = right;
e66f709b
CM		 1313 path->slots[level + 1] += 1;
1314 path->slots[level] = orig_slot -
5f39d397 1315 btrfs_header_nritems(mid);
925baedd 1316 btrfs_tree_unlock(mid);
5f39d397 1317 free_extent_buffer(mid);
e66f709b 1318 } else {
925baedd 1319 btrfs_tree_unlock(right);
5f39d397 1320 free_extent_buffer(right);
e66f709b 1321 }
e66f709b
CM		 1322 return 0;
1323 }
925baedd 1324 btrfs_tree_unlock(right);
5f39d397 1325 free_extent_buffer(right);
e66f709b 1326 }
e66f709b
CM		 1327 return 1;
1328}
1329
3c69faec 1330/*
d352ac68
CM		 1331 * readahead one full node of leaves, finding things that are close
1332 * to the block in 'slot', and triggering ra on them.
3c69faec 1333 */
c8c42864
CM		 1334static void reada_for_search(struct btrfs_root *root,
1335 struct btrfs_path *path,
1336 int level, int slot, u64 objectid)
3c69faec 1337{
5f39d397 1338 struct extent_buffer *node;
01f46658 1339 struct btrfs_disk_key disk_key;
3c69faec 1340 u32 nritems;
3c69faec 1341 u64 search;
a7175319 1342 u64 target;
6b80053d 1343 u64 nread = 0;
3c69faec 1344 int direction = path->reada;
5f39d397 1345 struct extent_buffer *eb;
6b80053d
CM		 1346 u32 nr;
1347 u32 blocksize;
1348 u32 nscan = 0;
db94535d 1349
a6b6e75e 1350 if (level != 1)
6702ed49
CM		 1351 return;
1352
1353 if (!path->nodes[level])
3c69faec
CM		 1354 return;
1355
5f39d397 1356 node = path->nodes[level];
925baedd 1357
3c69faec 1358 search = btrfs_node_blockptr(node, slot);
6b80053d
CM		 1359 blocksize = btrfs_level_size(root, level - 1);
1360 eb = btrfs_find_tree_block(root, search, blocksize);
5f39d397
CM		 1361 if (eb) {
1362 free_extent_buffer(eb);
3c69faec
CM		 1363 return;
1364 }
1365
a7175319 1366 target = search;
6b80053d 1367
5f39d397 1368 nritems = btrfs_header_nritems(node);
6b80053d 1369 nr = slot;
d397712b 1370 while (1) {
6b80053d
CM		 1371 if (direction < 0) {
1372 if (nr == 0)
1373 break;
1374 nr--;
1375 } else if (direction > 0) {
1376 nr++;
1377 if (nr >= nritems)
1378 break;
3c69faec 1379 }
01f46658
CM		 1380 if (path->reada < 0 && objectid) {
1381 btrfs_node_key(node, &disk_key, nr);
1382 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1383 break;
1384 }
6b80053d 1385 search = btrfs_node_blockptr(node, nr);
a7175319
CM		 1386 if ((search <= target && target - search <= 65536) ||
1387 (search > target && search - target <= 65536)) {
ca7a79ad
CM		 1388 readahead_tree_block(root, search, blocksize,
1389 btrfs_node_ptr_generation(node, nr));
6b80053d
CM		 1390 nread += blocksize;
1391 }
1392 nscan++;
a7175319 1393 if ((nread > 65536 || nscan > 32))
6b80053d 1394 break;
3c69faec
CM		 1395 }
1396}
925baedd 1397
b4ce94de
CM		 1398/*
1399 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1400 * cache
1401 */
1402static noinline int reada_for_balance(struct btrfs_root *root,
1403 struct btrfs_path *path, int level)
1404{
1405 int slot;
1406 int nritems;
1407 struct extent_buffer *parent;
1408 struct extent_buffer *eb;
1409 u64 gen;
1410 u64 block1 = 0;
1411 u64 block2 = 0;
1412 int ret = 0;
1413 int blocksize;
1414
8c594ea8 1415 parent = path->nodes[level + 1];
b4ce94de
CM		 1416 if (!parent)
1417 return 0;
1418
1419 nritems = btrfs_header_nritems(parent);
8c594ea8 1420 slot = path->slots[level + 1];
b4ce94de
CM		 1421 blocksize = btrfs_level_size(root, level);
1422
1423 if (slot > 0) {
1424 block1 = btrfs_node_blockptr(parent, slot - 1);
1425 gen = btrfs_node_ptr_generation(parent, slot - 1);
1426 eb = btrfs_find_tree_block(root, block1, blocksize);
1427 if (eb && btrfs_buffer_uptodate(eb, gen))
1428 block1 = 0;
1429 free_extent_buffer(eb);
1430 }
8c594ea8 1431 if (slot + 1 < nritems) {
b4ce94de
CM		 1432 block2 = btrfs_node_blockptr(parent, slot + 1);
1433 gen = btrfs_node_ptr_generation(parent, slot + 1);
1434 eb = btrfs_find_tree_block(root, block2, blocksize);
1435 if (eb && btrfs_buffer_uptodate(eb, gen))
1436 block2 = 0;
1437 free_extent_buffer(eb);
1438 }
1439 if (block1 || block2) {
1440 ret = -EAGAIN;
8c594ea8
CM		 1441
1442 /* release the whole path */
b4ce94de 1443 btrfs_release_path(root, path);
8c594ea8
CM		 1444
1445 /* read the blocks */
b4ce94de
CM		 1446 if (block1)
1447 readahead_tree_block(root, block1, blocksize, 0);
1448 if (block2)
1449 readahead_tree_block(root, block2, blocksize, 0);
1450
1451 if (block1) {
1452 eb = read_tree_block(root, block1, blocksize, 0);
1453 free_extent_buffer(eb);
1454 }
8c594ea8 1455 if (block2) {
b4ce94de
CM		 1456 eb = read_tree_block(root, block2, blocksize, 0);
1457 free_extent_buffer(eb);
1458 }
1459 }
1460 return ret;
1461}
1462
1463
d352ac68 1464/*
d397712b
CM		 1465 * when we walk down the tree, it is usually safe to unlock the higher layers
1466 * in the tree. The exceptions are when our path goes through slot 0, because
1467 * operations on the tree might require changing key pointers higher up in the
1468 * tree.
d352ac68 1469 *
d397712b
CM		 1470 * callers might also have set path->keep_locks, which tells this code to keep
1471 * the lock if the path points to the last slot in the block. This is part of
1472 * walking through the tree, and selecting the next slot in the higher block.
d352ac68 1473 *
d397712b
CM		 1474 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1475 * if lowest_unlock is 1, level 0 won't be unlocked
d352ac68 1476 */
e02119d5
CM		 1477static noinline void unlock_up(struct btrfs_path *path, int level,
1478 int lowest_unlock)
925baedd
CM		 1479{
1480 int i;
1481 int skip_level = level;
051e1b9f 1482 int no_skips = 0;
925baedd
CM		 1483 struct extent_buffer *t;
1484
1485 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1486 if (!path->nodes[i])
1487 break;
1488 if (!path->locks[i])
1489 break;
051e1b9f 1490 if (!no_skips && path->slots[i] == 0) {
925baedd
CM		 1491 skip_level = i + 1;
1492 continue;
1493 }
051e1b9f 1494 if (!no_skips && path->keep_locks) {
925baedd
CM		 1495 u32 nritems;
1496 t = path->nodes[i];
1497 nritems = btrfs_header_nritems(t);
051e1b9f 1498 if (nritems < 1 || path->slots[i] >= nritems - 1) {
925baedd
CM		 1499 skip_level = i + 1;
1500 continue;
1501 }
1502 }
051e1b9f
CM		 1503 if (skip_level < i && i >= lowest_unlock)
1504 no_skips = 1;
1505
925baedd
CM		 1506 t = path->nodes[i];
1507 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1508 btrfs_tree_unlock(t);
1509 path->locks[i] = 0;
1510 }
1511 }
1512}
1513
b4ce94de
CM		 1514/*
1515 * This releases any locks held in the path starting at level and
1516 * going all the way up to the root.
1517 *
1518 * btrfs_search_slot will keep the lock held on higher nodes in a few
1519 * corner cases, such as COW of the block at slot zero in the node. This
1520 * ignores those rules, and it should only be called when there are no
1521 * more updates to be done higher up in the tree.
1522 */
1523noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1524{
1525 int i;
1526
5d4f98a2 1527 if (path->keep_locks)
b4ce94de
CM		 1528 return;
1529
1530 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1531 if (!path->nodes[i])
12f4dacc 1532 continue;
b4ce94de 1533 if (!path->locks[i])
12f4dacc 1534 continue;
b4ce94de
CM		 1535 btrfs_tree_unlock(path->nodes[i]);
1536 path->locks[i] = 0;
1537 }
1538}
1539
c8c42864
CM		 1540/*
1541 * helper function for btrfs_search_slot. The goal is to find a block
1542 * in cache without setting the path to blocking. If we find the block
1543 * we return zero and the path is unchanged.
1544 *
1545 * If we can't find the block, we set the path blocking and do some
1546 * reada. -EAGAIN is returned and the search must be repeated.
1547 */
1548static int
1549read_block_for_search(struct btrfs_trans_handle *trans,
1550 struct btrfs_root *root, struct btrfs_path *p,
1551 struct extent_buffer **eb_ret, int level, int slot,
1552 struct btrfs_key *key)
1553{
1554 u64 blocknr;
1555 u64 gen;
1556 u32 blocksize;
1557 struct extent_buffer *b = *eb_ret;
1558 struct extent_buffer *tmp;
76a05b35 1559 int ret;
c8c42864
CM		 1560
1561 blocknr = btrfs_node_blockptr(b, slot);
1562 gen = btrfs_node_ptr_generation(b, slot);
1563 blocksize = btrfs_level_size(root, level - 1);
1564
1565 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1566 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
76a05b35
CM		 1567 /*
1568 * we found an up to date block without sleeping, return
1569 * right away
1570 */
c8c42864
CM		 1571 *eb_ret = tmp;
1572 return 0;
1573 }
1574
1575 /*
1576 * reduce lock contention at high levels
1577 * of the btree by dropping locks before
76a05b35
CM		 1578 * we read. Don't release the lock on the current
1579 * level because we need to walk this node to figure
1580 * out which blocks to read.
c8c42864 1581 */
8c594ea8
CM		 1582 btrfs_unlock_up_safe(p, level + 1);
1583 btrfs_set_path_blocking(p);
1584
c8c42864
CM		 1585 if (tmp)
1586 free_extent_buffer(tmp);
1587 if (p->reada)
1588 reada_for_search(root, p, level, slot, key->objectid);
1589
8c594ea8 1590 btrfs_release_path(NULL, p);
76a05b35
CM		 1591
1592 ret = -EAGAIN;
c8c42864 1593 tmp = read_tree_block(root, blocknr, blocksize, gen);
76a05b35
CM		 1594 if (tmp) {
1595 /*
1596 * If the read above didn't mark this buffer up to date,
1597 * it will never end up being up to date. Set ret to EIO now
1598 * and give up so that our caller doesn't loop forever
1599 * on our EAGAINs.
1600 */
1601 if (!btrfs_buffer_uptodate(tmp, 0))
1602 ret = -EIO;
c8c42864 1603 free_extent_buffer(tmp);
76a05b35
CM		 1604 }
1605 return ret;
c8c42864
CM		 1606}
1607
1608/*
1609 * helper function for btrfs_search_slot. This does all of the checks
1610 * for node-level blocks and does any balancing required based on
1611 * the ins_len.
1612 *
1613 * If no extra work was required, zero is returned. If we had to
1614 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1615 * start over
1616 */
1617static int
1618setup_nodes_for_search(struct btrfs_trans_handle *trans,
1619 struct btrfs_root *root, struct btrfs_path *p,
1620 struct extent_buffer *b, int level, int ins_len)
1621{
1622 int ret;
1623 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1624 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1625 int sret;
1626
1627 sret = reada_for_balance(root, p, level);
1628 if (sret)
1629 goto again;
1630
1631 btrfs_set_path_blocking(p);
1632 sret = split_node(trans, root, p, level);
1633 btrfs_clear_path_blocking(p, NULL);
1634
1635 BUG_ON(sret > 0);
1636 if (sret) {
1637 ret = sret;
1638 goto done;
1639 }
1640 b = p->nodes[level];
1641 } else if (ins_len < 0 && btrfs_header_nritems(b) <
cfbb9308 1642 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
c8c42864
CM		 1643 int sret;
1644
1645 sret = reada_for_balance(root, p, level);
1646 if (sret)
1647 goto again;
1648
1649 btrfs_set_path_blocking(p);
1650 sret = balance_level(trans, root, p, level);
1651 btrfs_clear_path_blocking(p, NULL);
1652
1653 if (sret) {
1654 ret = sret;
1655 goto done;
1656 }
1657 b = p->nodes[level];
1658 if (!b) {
1659 btrfs_release_path(NULL, p);
1660 goto again;
1661 }
1662 BUG_ON(btrfs_header_nritems(b) == 1);
1663 }
1664 return 0;
1665
1666again:
1667 ret = -EAGAIN;
1668done:
1669 return ret;
1670}
1671
74123bd7
CM		 1672/*
1673 * look for key in the tree. path is filled in with nodes along the way
1674 * if key is found, we return zero and you can find the item in the leaf
1675 * level of the path (level 0)
1676 *
1677 * If the key isn't found, the path points to the slot where it should
aa5d6bed
CM		 1678 * be inserted, and 1 is returned. If there are other errors during the
1679 * search a negative error number is returned.
97571fd0
CM		 1680 *
1681 * if ins_len > 0, nodes and leaves will be split as we walk down the
1682 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1683 * possible)
74123bd7 1684 */
e089f05c
CM		 1685int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1686 *root, struct btrfs_key *key, struct btrfs_path *p, int
1687 ins_len, int cow)
be0e5c09 1688{
5f39d397 1689 struct extent_buffer *b;
be0e5c09
CM		 1690 int slot;
1691 int ret;
33c66f43 1692 int err;
be0e5c09 1693 int level;
925baedd 1694 int lowest_unlock = 1;
9f3a7427
CM		 1695 u8 lowest_level = 0;
1696
6702ed49 1697 lowest_level = p->lowest_level;
323ac95b 1698 WARN_ON(lowest_level && ins_len > 0);
22b0ebda 1699 WARN_ON(p->nodes[0] != NULL);
25179201 1700
925baedd
CM		 1701 if (ins_len < 0)
1702 lowest_unlock = 2;
65b51a00 1703
bb803951 1704again:
5d4f98a2
YZ		 1705 if (p->search_commit_root) {
1706 b = root->commit_root;
1707 extent_buffer_get(b);
1708 if (!p->skip_locking)
1709 btrfs_tree_lock(b);
1710 } else {
1711 if (p->skip_locking)
1712 b = btrfs_root_node(root);
1713 else
1714 b = btrfs_lock_root_node(root);
1715 }
925baedd 1716
eb60ceac 1717 while (b) {
5f39d397 1718 level = btrfs_header_level(b);
65b51a00
CM		 1719
1720 /*
1721 * setup the path here so we can release it under lock
1722 * contention with the cow code
1723 */
1724 p->nodes[level] = b;
1725 if (!p->skip_locking)
1726 p->locks[level] = 1;
1727
02217ed2 1728 if (cow) {
c8c42864
CM		 1729 /*
1730 * if we don't really need to cow this block
1731 * then we don't want to set the path blocking,
1732 * so we test it here
1733 */
5d4f98a2 1734 if (!should_cow_block(trans, root, b))
65b51a00 1735 goto cow_done;
5d4f98a2 1736
b4ce94de
CM		 1737 btrfs_set_path_blocking(p);
1738
33c66f43
YZ		 1739 err = btrfs_cow_block(trans, root, b,
1740 p->nodes[level + 1],
1741 p->slots[level + 1], &b);
1742 if (err) {
5f39d397 1743 free_extent_buffer(b);
33c66f43 1744 ret = err;
65b51a00 1745 goto done;
54aa1f4d 1746 }
02217ed2 1747 }
65b51a00 1748cow_done:
02217ed2 1749 BUG_ON(!cow && ins_len);
5f39d397 1750 if (level != btrfs_header_level(b))
2c90e5d6 1751 WARN_ON(1);
5f39d397 1752 level = btrfs_header_level(b);
65b51a00 1753
eb60ceac 1754 p->nodes[level] = b;
5cd57b2c
CM		 1755 if (!p->skip_locking)
1756 p->locks[level] = 1;
65b51a00 1757
4008c04a 1758 btrfs_clear_path_blocking(p, NULL);
b4ce94de
CM		 1759
1760 /*
1761 * we have a lock on b and as long as we aren't changing
1762 * the tree, there is no way to for the items in b to change.
1763 * It is safe to drop the lock on our parent before we
1764 * go through the expensive btree search on b.
1765 *
1766 * If cow is true, then we might be changing slot zero,
1767 * which may require changing the parent. So, we can't
1768 * drop the lock until after we know which slot we're
1769 * operating on.
1770 */
1771 if (!cow)
1772 btrfs_unlock_up_safe(p, level + 1);
1773
123abc88 1774 ret = check_block(root, p, level);
65b51a00
CM		 1775 if (ret) {
1776 ret = -1;
1777 goto done;
1778 }
925baedd 1779
5f39d397 1780 ret = bin_search(b, key, level, &slot);
b4ce94de 1781
5f39d397 1782 if (level != 0) {
33c66f43
YZ		 1783 int dec = 0;
1784 if (ret && slot > 0) {
1785 dec = 1;
be0e5c09 1786 slot -= 1;
33c66f43 1787 }
be0e5c09 1788 p->slots[level] = slot;
33c66f43 1789 err = setup_nodes_for_search(trans, root, p, b, level,
c8c42864 1790 ins_len);
33c66f43 1791 if (err == -EAGAIN)
c8c42864 1792 goto again;
33c66f43
YZ		 1793 if (err) {
1794 ret = err;
c8c42864 1795 goto done;
33c66f43 1796 }
c8c42864
CM		 1797 b = p->nodes[level];
1798 slot = p->slots[level];
b4ce94de 1799
f9efa9c7
CM		 1800 unlock_up(p, level, lowest_unlock);
1801
925baedd 1802 if (level == lowest_level) {
33c66f43
YZ		 1803 if (dec)
1804 p->slots[level]++;
5b21f2ed 1805 goto done;
925baedd 1806 }
ca7a79ad 1807
33c66f43 1808 err = read_block_for_search(trans, root, p,
c8c42864 1809 &b, level, slot, key);
33c66f43 1810 if (err == -EAGAIN)
c8c42864 1811 goto again;
33c66f43
YZ		 1812 if (err) {
1813 ret = err;
76a05b35 1814 goto done;
33c66f43 1815 }
76a05b35 1816
b4ce94de 1817 if (!p->skip_locking) {
4008c04a 1818 btrfs_clear_path_blocking(p, NULL);
33c66f43 1819 err = btrfs_try_spin_lock(b);
b4ce94de 1820
33c66f43 1821 if (!err) {
b4ce94de
CM		 1822 btrfs_set_path_blocking(p);
1823 btrfs_tree_lock(b);
4008c04a 1824 btrfs_clear_path_blocking(p, b);
b4ce94de
CM		 1825 }
1826 }
be0e5c09
CM		 1827 } else {
1828 p->slots[level] = slot;
87b29b20
YZ		 1829 if (ins_len > 0 &&
1830 btrfs_leaf_free_space(root, b) < ins_len) {
b4ce94de 1831 btrfs_set_path_blocking(p);
33c66f43
YZ		 1832 err = split_leaf(trans, root, key,
1833 p, ins_len, ret == 0);
4008c04a 1834 btrfs_clear_path_blocking(p, NULL);
b4ce94de 1835
33c66f43
YZ		 1836 BUG_ON(err > 0);
1837 if (err) {
1838 ret = err;
65b51a00
CM		 1839 goto done;
1840 }
5c680ed6 1841 }
459931ec
CM		 1842 if (!p->search_for_split)
1843 unlock_up(p, level, lowest_unlock);
65b51a00 1844 goto done;
be0e5c09
CM		 1845 }
1846 }
65b51a00
CM		 1847 ret = 1;
1848done:
b4ce94de
CM		 1849 /*
1850 * we don't really know what they plan on doing with the path
1851 * from here on, so for now just mark it as blocking
1852 */
b9473439
CM		 1853 if (!p->leave_spinning)
1854 btrfs_set_path_blocking(p);
76a05b35
CM		 1855 if (ret < 0)
1856 btrfs_release_path(root, p);
65b51a00 1857 return ret;
be0e5c09
CM		 1858}
1859
74123bd7
CM		 1860/*
1861 * adjust the pointers going up the tree, starting at level
1862 * making sure the right key of each node is points to 'key'.
1863 * This is used after shifting pointers to the left, so it stops
1864 * fixing up pointers when a given leaf/node is not in slot 0 of the
1865 * higher levels
aa5d6bed
CM		 1866 *
1867 * If this fails to write a tree block, it returns -1, but continues
1868 * fixing up the blocks in ram so the tree is consistent.
74123bd7 1869 */
5f39d397
CM		 1870static int fixup_low_keys(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root, struct btrfs_path *path,
1872 struct btrfs_disk_key *key, int level)
be0e5c09
CM		 1873{
1874 int i;
aa5d6bed 1875 int ret = 0;
5f39d397
CM		 1876 struct extent_buffer *t;
1877
234b63a0 1878 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
be0e5c09 1879 int tslot = path->slots[i];
eb60ceac 1880 if (!path->nodes[i])
be0e5c09 1881 break;
5f39d397
CM		 1882 t = path->nodes[i];
1883 btrfs_set_node_key(t, key, tslot);
d6025579 1884 btrfs_mark_buffer_dirty(path->nodes[i]);
be0e5c09
CM		 1885 if (tslot != 0)
1886 break;
1887 }
aa5d6bed 1888 return ret;
be0e5c09
CM		 1889}
1890
31840ae1
ZY		 1891/*
1892 * update item key.
1893 *
1894 * This function isn't completely safe. It's the caller's responsibility
1895 * that the new key won't break the order
1896 */
1897int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root, struct btrfs_path *path,
1899 struct btrfs_key *new_key)
1900{
1901 struct btrfs_disk_key disk_key;
1902 struct extent_buffer *eb;
1903 int slot;
1904
1905 eb = path->nodes[0];
1906 slot = path->slots[0];
1907 if (slot > 0) {
1908 btrfs_item_key(eb, &disk_key, slot - 1);
1909 if (comp_keys(&disk_key, new_key) >= 0)
1910 return -1;
1911 }
1912 if (slot < btrfs_header_nritems(eb) - 1) {
1913 btrfs_item_key(eb, &disk_key, slot + 1);
1914 if (comp_keys(&disk_key, new_key) <= 0)
1915 return -1;
1916 }
1917
1918 btrfs_cpu_key_to_disk(&disk_key, new_key);
1919 btrfs_set_item_key(eb, &disk_key, slot);
1920 btrfs_mark_buffer_dirty(eb);
1921 if (slot == 0)
1922 fixup_low_keys(trans, root, path, &disk_key, 1);
1923 return 0;
1924}
1925
74123bd7
CM		 1926/*
1927 * try to push data from one node into the next node left in the
79f95c82 1928 * tree.
aa5d6bed
CM		 1929 *
1930 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1931 * error, and > 0 if there was no room in the left hand block.
74123bd7 1932 */
98ed5174
CM		 1933static int push_node_left(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root, struct extent_buffer *dst,
971a1f66 1935 struct extent_buffer *src, int empty)
be0e5c09 1936{
be0e5c09 1937 int push_items = 0;
bb803951
CM		 1938 int src_nritems;
1939 int dst_nritems;
aa5d6bed 1940 int ret = 0;
be0e5c09 1941
5f39d397
CM		 1942 src_nritems = btrfs_header_nritems(src);
1943 dst_nritems = btrfs_header_nritems(dst);
123abc88 1944 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
7bb86316
CM		 1945 WARN_ON(btrfs_header_generation(src) != trans->transid);
1946 WARN_ON(btrfs_header_generation(dst) != trans->transid);
54aa1f4d 1947
bce4eae9 1948 if (!empty && src_nritems <= 8)
971a1f66
CM		 1949 return 1;
1950
d397712b 1951 if (push_items <= 0)
be0e5c09
CM		 1952 return 1;
1953
bce4eae9 1954 if (empty) {
971a1f66 1955 push_items = min(src_nritems, push_items);
bce4eae9
CM		 1956 if (push_items < src_nritems) {
1957 /* leave at least 8 pointers in the node if
1958 * we aren't going to empty it
1959 */
1960 if (src_nritems - push_items < 8) {
1961 if (push_items <= 8)
1962 return 1;
1963 push_items -= 8;
1964 }
1965 }
1966 } else
1967 push_items = min(src_nritems - 8, push_items);
79f95c82 1968
5f39d397
CM		 1969 copy_extent_buffer(dst, src,
1970 btrfs_node_key_ptr_offset(dst_nritems),
1971 btrfs_node_key_ptr_offset(0),
d397712b 1972 push_items * sizeof(struct btrfs_key_ptr));
5f39d397 1973
bb803951 1974 if (push_items < src_nritems) {
5f39d397
CM		 1975 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1976 btrfs_node_key_ptr_offset(push_items),
1977 (src_nritems - push_items) *
1978 sizeof(struct btrfs_key_ptr));
1979 }
1980 btrfs_set_header_nritems(src, src_nritems - push_items);
1981 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1982 btrfs_mark_buffer_dirty(src);
1983 btrfs_mark_buffer_dirty(dst);
31840ae1 1984
79f95c82
CM		 1985 return ret;
1986}
1987
1988/*
1989 * try to push data from one node into the next node right in the
1990 * tree.
1991 *
1992 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1993 * error, and > 0 if there was no room in the right hand block.
1994 *
1995 * this will only push up to 1/2 the contents of the left node over
1996 */
5f39d397
CM		 1997static int balance_node_right(struct btrfs_trans_handle *trans,
1998 struct btrfs_root *root,
1999 struct extent_buffer *dst,
2000 struct extent_buffer *src)
79f95c82 2001{
79f95c82
CM		 2002 int push_items = 0;
2003 int max_push;
2004 int src_nritems;
2005 int dst_nritems;
2006 int ret = 0;
79f95c82 2007
7bb86316
CM		 2008 WARN_ON(btrfs_header_generation(src) != trans->transid);
2009 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2010
5f39d397
CM		 2011 src_nritems = btrfs_header_nritems(src);
2012 dst_nritems = btrfs_header_nritems(dst);
123abc88 2013 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
d397712b 2014 if (push_items <= 0)
79f95c82 2015 return 1;
bce4eae9 2016
d397712b 2017 if (src_nritems < 4)
bce4eae9 2018 return 1;
79f95c82
CM		 2019
2020 max_push = src_nritems / 2 + 1;
2021 /* don't try to empty the node */
d397712b 2022 if (max_push >= src_nritems)
79f95c82 2023 return 1;
252c38f0 2024
79f95c82
CM		 2025 if (max_push < push_items)
2026 push_items = max_push;
2027
5f39d397
CM		 2028 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2029 btrfs_node_key_ptr_offset(0),
2030 (dst_nritems) *
2031 sizeof(struct btrfs_key_ptr));
d6025579 2032
5f39d397
CM		 2033 copy_extent_buffer(dst, src,
2034 btrfs_node_key_ptr_offset(0),
2035 btrfs_node_key_ptr_offset(src_nritems - push_items),
d397712b 2036 push_items * sizeof(struct btrfs_key_ptr));
79f95c82 2037
5f39d397
CM		 2038 btrfs_set_header_nritems(src, src_nritems - push_items);
2039 btrfs_set_header_nritems(dst, dst_nritems + push_items);
79f95c82 2040
5f39d397
CM		 2041 btrfs_mark_buffer_dirty(src);
2042 btrfs_mark_buffer_dirty(dst);
31840ae1 2043
aa5d6bed 2044 return ret;
be0e5c09
CM		 2045}
2046
97571fd0
CM		 2047/*
2048 * helper function to insert a new root level in the tree.
2049 * A new node is allocated, and a single item is inserted to
2050 * point to the existing root
aa5d6bed
CM		 2051 *
2052 * returns zero on success or < 0 on failure.
97571fd0 2053 */
d397712b 2054static noinline int insert_new_root(struct btrfs_trans_handle *trans,
5f39d397
CM		 2055 struct btrfs_root *root,
2056 struct btrfs_path *path, int level)
5c680ed6 2057{
7bb86316 2058 u64 lower_gen;
5f39d397
CM		 2059 struct extent_buffer *lower;
2060 struct extent_buffer *c;
925baedd 2061 struct extent_buffer *old;
5f39d397 2062 struct btrfs_disk_key lower_key;
5c680ed6
CM		 2063
2064 BUG_ON(path->nodes[level]);
2065 BUG_ON(path->nodes[level-1] != root->node);
2066
7bb86316
CM		 2067 lower = path->nodes[level-1];
2068 if (level == 1)
2069 btrfs_item_key(lower, &lower_key, 0);
2070 else
2071 btrfs_node_key(lower, &lower_key, 0);
2072
31840ae1 2073 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
5d4f98a2 2074 root->root_key.objectid, &lower_key,
ad3d81ba 2075 level, root->node->start, 0);
5f39d397
CM		 2076 if (IS_ERR(c))
2077 return PTR_ERR(c);
925baedd 2078
5d4f98a2 2079 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
5f39d397
CM		 2080 btrfs_set_header_nritems(c, 1);
2081 btrfs_set_header_level(c, level);
db94535d 2082 btrfs_set_header_bytenr(c, c->start);
5f39d397 2083 btrfs_set_header_generation(c, trans->transid);
5d4f98a2 2084 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
5f39d397 2085 btrfs_set_header_owner(c, root->root_key.objectid);
5f39d397
CM		 2086
2087 write_extent_buffer(c, root->fs_info->fsid,
2088 (unsigned long)btrfs_header_fsid(c),
2089 BTRFS_FSID_SIZE);
e17cade2
CM		 2090
2091 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2092 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2093 BTRFS_UUID_SIZE);
2094
5f39d397 2095 btrfs_set_node_key(c, &lower_key, 0);
db94535d 2096 btrfs_set_node_blockptr(c, 0, lower->start);
7bb86316 2097 lower_gen = btrfs_header_generation(lower);
31840ae1 2098 WARN_ON(lower_gen != trans->transid);
7bb86316
CM		 2099
2100 btrfs_set_node_ptr_generation(c, 0, lower_gen);
d5719762 2101
5f39d397 2102 btrfs_mark_buffer_dirty(c);
d5719762 2103
925baedd
CM		 2104 spin_lock(&root->node_lock);
2105 old = root->node;
5f39d397 2106 root->node = c;
925baedd
CM		 2107 spin_unlock(&root->node_lock);
2108
2109 /* the super has an extra ref to root->node */
2110 free_extent_buffer(old);
2111
0b86a832 2112 add_root_to_dirty_list(root);
5f39d397
CM		 2113 extent_buffer_get(c);
2114 path->nodes[level] = c;
925baedd 2115 path->locks[level] = 1;
5c680ed6
CM		 2116 path->slots[level] = 0;
2117 return 0;
2118}
2119
74123bd7
CM		 2120/*
2121 * worker function to insert a single pointer in a node.
2122 * the node should have enough room for the pointer already
97571fd0 2123 *
74123bd7
CM		 2124 * slot and level indicate where you want the key to go, and
2125 * blocknr is the block the key points to.
aa5d6bed
CM		 2126 *
2127 * returns zero on success and < 0 on any error
74123bd7 2128 */
e089f05c
CM		 2129static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2130 *root, struct btrfs_path *path, struct btrfs_disk_key
db94535d 2131 *key, u64 bytenr, int slot, int level)
74123bd7 2132{
5f39d397 2133 struct extent_buffer *lower;
74123bd7 2134 int nritems;
5c680ed6
CM		 2135
2136 BUG_ON(!path->nodes[level]);
5f39d397
CM		 2137 lower = path->nodes[level];
2138 nritems = btrfs_header_nritems(lower);
c293498b 2139 BUG_ON(slot > nritems);
123abc88 2140 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
74123bd7
CM		 2141 BUG();
2142 if (slot != nritems) {
5f39d397
CM		 2143 memmove_extent_buffer(lower,
2144 btrfs_node_key_ptr_offset(slot + 1),
2145 btrfs_node_key_ptr_offset(slot),
d6025579 2146 (nritems - slot) * sizeof(struct btrfs_key_ptr));
74123bd7 2147 }
5f39d397 2148 btrfs_set_node_key(lower, key, slot);
db94535d 2149 btrfs_set_node_blockptr(lower, slot, bytenr);
74493f7a
CM		 2150 WARN_ON(trans->transid == 0);
2151 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
5f39d397
CM		 2152 btrfs_set_header_nritems(lower, nritems + 1);
2153 btrfs_mark_buffer_dirty(lower);
74123bd7
CM		 2154 return 0;
2155}
2156
97571fd0
CM		 2157/*
2158 * split the node at the specified level in path in two.
2159 * The path is corrected to point to the appropriate node after the split
2160 *
2161 * Before splitting this tries to make some room in the node by pushing
2162 * left and right, if either one works, it returns right away.
aa5d6bed
CM		 2163 *
2164 * returns 0 on success and < 0 on failure
97571fd0 2165 */
e02119d5
CM		 2166static noinline int split_node(struct btrfs_trans_handle *trans,
2167 struct btrfs_root *root,
2168 struct btrfs_path *path, int level)
be0e5c09 2169{
5f39d397
CM		 2170 struct extent_buffer *c;
2171 struct extent_buffer *split;
2172 struct btrfs_disk_key disk_key;
be0e5c09 2173 int mid;
5c680ed6 2174 int ret;
aa5d6bed 2175 int wret;
7518a238 2176 u32 c_nritems;
eb60ceac 2177
5f39d397 2178 c = path->nodes[level];
7bb86316 2179 WARN_ON(btrfs_header_generation(c) != trans->transid);
5f39d397 2180 if (c == root->node) {
5c680ed6 2181 /* trying to split the root, lets make a new one */
e089f05c 2182 ret = insert_new_root(trans, root, path, level + 1);
5c680ed6
CM		 2183 if (ret)
2184 return ret;
b3612421 2185 } else {
e66f709b 2186 ret = push_nodes_for_insert(trans, root, path, level);
5f39d397
CM		 2187 c = path->nodes[level];
2188 if (!ret && btrfs_header_nritems(c) <
c448acf0 2189 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
e66f709b 2190 return 0;
54aa1f4d
CM		 2191 if (ret < 0)
2192 return ret;
be0e5c09 2193 }
e66f709b 2194
5f39d397 2195 c_nritems = btrfs_header_nritems(c);
5d4f98a2
YZ		 2196 mid = (c_nritems + 1) / 2;
2197 btrfs_node_key(c, &disk_key, mid);
7bb86316 2198
5d4f98a2 2199 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
31840ae1 2200 root->root_key.objectid,
5d4f98a2 2201 &disk_key, level, c->start, 0);
5f39d397
CM		 2202 if (IS_ERR(split))
2203 return PTR_ERR(split);
2204
5d4f98a2 2205 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
5f39d397 2206 btrfs_set_header_level(split, btrfs_header_level(c));
db94535d 2207 btrfs_set_header_bytenr(split, split->start);
5f39d397 2208 btrfs_set_header_generation(split, trans->transid);
5d4f98a2 2209 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
5f39d397
CM		 2210 btrfs_set_header_owner(split, root->root_key.objectid);
2211 write_extent_buffer(split, root->fs_info->fsid,
2212 (unsigned long)btrfs_header_fsid(split),
2213 BTRFS_FSID_SIZE);
e17cade2
CM		 2214 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2215 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2216 BTRFS_UUID_SIZE);
54aa1f4d 2217
5f39d397
CM		 2218
2219 copy_extent_buffer(split, c,
2220 btrfs_node_key_ptr_offset(0),
2221 btrfs_node_key_ptr_offset(mid),
2222 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2223 btrfs_set_header_nritems(split, c_nritems - mid);
2224 btrfs_set_header_nritems(c, mid);
aa5d6bed
CM		 2225 ret = 0;
2226
5f39d397
CM		 2227 btrfs_mark_buffer_dirty(c);
2228 btrfs_mark_buffer_dirty(split);
2229
db94535d 2230 wret = insert_ptr(trans, root, path, &disk_key, split->start,
5f39d397 2231 path->slots[level + 1] + 1,
123abc88 2232 level + 1);
aa5d6bed
CM		 2233 if (wret)
2234 ret = wret;
2235
5de08d7d 2236 if (path->slots[level] >= mid) {
5c680ed6 2237 path->slots[level] -= mid;
925baedd 2238 btrfs_tree_unlock(c);
5f39d397
CM		 2239 free_extent_buffer(c);
2240 path->nodes[level] = split;
5c680ed6
CM		 2241 path->slots[level + 1] += 1;
2242 } else {
925baedd 2243 btrfs_tree_unlock(split);
5f39d397 2244 free_extent_buffer(split);
be0e5c09 2245 }
aa5d6bed 2246 return ret;
be0e5c09
CM		 2247}
2248
74123bd7
CM		 2249/*
2250 * how many bytes are required to store the items in a leaf. start
2251 * and nr indicate which items in the leaf to check. This totals up the
2252 * space used both by the item structs and the item data
2253 */
5f39d397 2254static int leaf_space_used(struct extent_buffer *l, int start, int nr)
be0e5c09
CM		 2255{
2256 int data_len;
5f39d397 2257 int nritems = btrfs_header_nritems(l);
d4dbff95 2258 int end = min(nritems, start + nr) - 1;
be0e5c09
CM		 2259
2260 if (!nr)
2261 return 0;
5f39d397
CM		 2262 data_len = btrfs_item_end_nr(l, start);
2263 data_len = data_len - btrfs_item_offset_nr(l, end);
0783fcfc 2264 data_len += sizeof(struct btrfs_item) * nr;
d4dbff95 2265 WARN_ON(data_len < 0);
be0e5c09
CM		 2266 return data_len;
2267}
2268
d4dbff95
CM		 2269/*
2270 * The space between the end of the leaf items and
2271 * the start of the leaf data. IOW, how much room
2272 * the leaf has left for both items and data
2273 */
d397712b 2274noinline int btrfs_leaf_free_space(struct btrfs_root *root,
e02119d5 2275 struct extent_buffer *leaf)
d4dbff95 2276{
5f39d397
CM		 2277 int nritems = btrfs_header_nritems(leaf);
2278 int ret;
2279 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2280 if (ret < 0) {
d397712b
CM		 2281 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2282 "used %d nritems %d\n",
ae2f5411 2283 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
5f39d397
CM		 2284 leaf_space_used(leaf, 0, nritems), nritems);
2285 }
2286 return ret;
d4dbff95
CM		 2287}
2288
44871b1b
CM		 2289static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2290 struct btrfs_root *root,
2291 struct btrfs_path *path,
2292 int data_size, int empty,
2293 struct extent_buffer *right,
2294 int free_space, u32 left_nritems)
00ec4c51 2295{
5f39d397 2296 struct extent_buffer *left = path->nodes[0];
44871b1b 2297 struct extent_buffer *upper = path->nodes[1];
5f39d397 2298 struct btrfs_disk_key disk_key;
00ec4c51 2299 int slot;
34a38218 2300 u32 i;
00ec4c51
CM		 2301 int push_space = 0;
2302 int push_items = 0;
0783fcfc 2303 struct btrfs_item *item;
34a38218 2304 u32 nr;
7518a238 2305 u32 right_nritems;
5f39d397 2306 u32 data_end;
db94535d 2307 u32 this_item_size;
00ec4c51 2308
34a38218
CM		 2309 if (empty)
2310 nr = 0;
2311 else
2312 nr = 1;
2313
31840ae1 2314 if (path->slots[0] >= left_nritems)
87b29b20 2315 push_space += data_size;
31840ae1 2316
44871b1b 2317 slot = path->slots[1];
34a38218
CM		 2318 i = left_nritems - 1;
2319 while (i >= nr) {
5f39d397 2320 item = btrfs_item_nr(left, i);
db94535d 2321
31840ae1
ZY		 2322 if (!empty && push_items > 0) {
2323 if (path->slots[0] > i)
2324 break;
2325 if (path->slots[0] == i) {
2326 int space = btrfs_leaf_free_space(root, left);
2327 if (space + push_space * 2 > free_space)
2328 break;
2329 }
2330 }
2331
00ec4c51 2332 if (path->slots[0] == i)
87b29b20 2333 push_space += data_size;
db94535d
CM		 2334
2335 if (!left->map_token) {
2336 map_extent_buffer(left, (unsigned long)item,
2337 sizeof(struct btrfs_item),
2338 &left->map_token, &left->kaddr,
2339 &left->map_start, &left->map_len,
2340 KM_USER1);
2341 }
2342
2343 this_item_size = btrfs_item_size(left, item);
2344 if (this_item_size + sizeof(*item) + push_space > free_space)
00ec4c51 2345 break;
31840ae1 2346
00ec4c51 2347 push_items++;
db94535d 2348 push_space += this_item_size + sizeof(*item);
34a38218
CM		 2349 if (i == 0)
2350 break;
2351 i--;
db94535d
CM		 2352 }
2353 if (left->map_token) {
2354 unmap_extent_buffer(left, left->map_token, KM_USER1);
2355 left->map_token = NULL;
00ec4c51 2356 }
5f39d397 2357
925baedd
CM		 2358 if (push_items == 0)
2359 goto out_unlock;
5f39d397 2360
34a38218 2361 if (!empty && push_items == left_nritems)
a429e513 2362 WARN_ON(1);
5f39d397 2363
00ec4c51 2364 /* push left to right */
5f39d397 2365 right_nritems = btrfs_header_nritems(right);
34a38218 2366
5f39d397 2367 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
123abc88 2368 push_space -= leaf_data_end(root, left);
5f39d397 2369
00ec4c51 2370 /* make room in the right data area */
5f39d397
CM		 2371 data_end = leaf_data_end(root, right);
2372 memmove_extent_buffer(right,
2373 btrfs_leaf_data(right) + data_end - push_space,
2374 btrfs_leaf_data(right) + data_end,
2375 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2376
00ec4c51 2377 /* copy from the left data area */
5f39d397 2378 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
d6025579
CM		 2379 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2380 btrfs_leaf_data(left) + leaf_data_end(root, left),
2381 push_space);
5f39d397
CM		 2382
2383 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2384 btrfs_item_nr_offset(0),
2385 right_nritems * sizeof(struct btrfs_item));
2386
00ec4c51 2387 /* copy the items from left to right */
5f39d397
CM		 2388 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2389 btrfs_item_nr_offset(left_nritems - push_items),
2390 push_items * sizeof(struct btrfs_item));
00ec4c51
CM		 2391
2392 /* update the item pointers */
7518a238 2393 right_nritems += push_items;
5f39d397 2394 btrfs_set_header_nritems(right, right_nritems);
123abc88 2395 push_space = BTRFS_LEAF_DATA_SIZE(root);
7518a238 2396 for (i = 0; i < right_nritems; i++) {
5f39d397 2397 item = btrfs_item_nr(right, i);
db94535d
CM		 2398 if (!right->map_token) {
2399 map_extent_buffer(right, (unsigned long)item,
2400 sizeof(struct btrfs_item),
2401 &right->map_token, &right->kaddr,
2402 &right->map_start, &right->map_len,
2403 KM_USER1);
2404 }
2405 push_space -= btrfs_item_size(right, item);
2406 btrfs_set_item_offset(right, item, push_space);
2407 }
2408
2409 if (right->map_token) {
2410 unmap_extent_buffer(right, right->map_token, KM_USER1);
2411 right->map_token = NULL;
00ec4c51 2412 }
7518a238 2413 left_nritems -= push_items;
5f39d397 2414 btrfs_set_header_nritems(left, left_nritems);
00ec4c51 2415
34a38218
CM		 2416 if (left_nritems)
2417 btrfs_mark_buffer_dirty(left);
5f39d397 2418 btrfs_mark_buffer_dirty(right);
a429e513 2419
5f39d397
CM		 2420 btrfs_item_key(right, &disk_key, 0);
2421 btrfs_set_node_key(upper, &disk_key, slot + 1);
d6025579 2422 btrfs_mark_buffer_dirty(upper);
02217ed2 2423
00ec4c51 2424 /* then fixup the leaf pointer in the path */
7518a238
CM		 2425 if (path->slots[0] >= left_nritems) {
2426 path->slots[0] -= left_nritems;
925baedd
CM		 2427 if (btrfs_header_nritems(path->nodes[0]) == 0)
2428 clean_tree_block(trans, root, path->nodes[0]);
2429 btrfs_tree_unlock(path->nodes[0]);
5f39d397
CM		 2430 free_extent_buffer(path->nodes[0]);
2431 path->nodes[0] = right;
00ec4c51
CM		 2432 path->slots[1] += 1;
2433 } else {
925baedd 2434 btrfs_tree_unlock(right);
5f39d397 2435 free_extent_buffer(right);
00ec4c51
CM		 2436 }
2437 return 0;
925baedd
CM		 2438
2439out_unlock:
2440 btrfs_tree_unlock(right);
2441 free_extent_buffer(right);
2442 return 1;
00ec4c51 2443}
925baedd 2444
44871b1b
CM		 2445/*
2446 * push some data in the path leaf to the right, trying to free up at
2447 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2448 *
2449 * returns 1 if the push failed because the other node didn't have enough
2450 * room, 0 if everything worked out and < 0 if there were major errors.
2451 */
2452static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2453 *root, struct btrfs_path *path, int data_size,
2454 int empty)
2455{
2456 struct extent_buffer *left = path->nodes[0];
2457 struct extent_buffer *right;
2458 struct extent_buffer *upper;
2459 int slot;
2460 int free_space;
2461 u32 left_nritems;
2462 int ret;
2463
2464 if (!path->nodes[1])
2465 return 1;
2466
2467 slot = path->slots[1];
2468 upper = path->nodes[1];
2469 if (slot >= btrfs_header_nritems(upper) - 1)
2470 return 1;
2471
2472 btrfs_assert_tree_locked(path->nodes[1]);
2473
2474 right = read_node_slot(root, upper, slot + 1);
2475 btrfs_tree_lock(right);
2476 btrfs_set_lock_blocking(right);
2477
2478 free_space = btrfs_leaf_free_space(root, right);
2479 if (free_space < data_size)
2480 goto out_unlock;
2481
2482 /* cow and double check */
2483 ret = btrfs_cow_block(trans, root, right, upper,
2484 slot + 1, &right);
2485 if (ret)
2486 goto out_unlock;
2487
2488 free_space = btrfs_leaf_free_space(root, right);
2489 if (free_space < data_size)
2490 goto out_unlock;
2491
2492 left_nritems = btrfs_header_nritems(left);
2493 if (left_nritems == 0)
2494 goto out_unlock;
2495
2496 return __push_leaf_right(trans, root, path, data_size, empty,
2497 right, free_space, left_nritems);
2498out_unlock:
2499 btrfs_tree_unlock(right);
2500 free_extent_buffer(right);
2501 return 1;
2502}
2503
74123bd7
CM		 2504/*
2505 * push some data in the path leaf to the left, trying to free up at
2506 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2507 */
44871b1b
CM		 2508static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 struct btrfs_path *path, int data_size,
2511 int empty, struct extent_buffer *left,
2512 int free_space, int right_nritems)
be0e5c09 2513{
5f39d397
CM		 2514 struct btrfs_disk_key disk_key;
2515 struct extent_buffer *right = path->nodes[0];
be0e5c09
CM		 2516 int slot;
2517 int i;
be0e5c09
CM		 2518 int push_space = 0;
2519 int push_items = 0;
0783fcfc 2520 struct btrfs_item *item;
7518a238 2521 u32 old_left_nritems;
34a38218 2522 u32 nr;
aa5d6bed
CM		 2523 int ret = 0;
2524 int wret;
db94535d
CM		 2525 u32 this_item_size;
2526 u32 old_left_item_size;
be0e5c09
CM		 2527
2528 slot = path->slots[1];
02217ed2 2529
34a38218
CM		 2530 if (empty)
2531 nr = right_nritems;
2532 else
2533 nr = right_nritems - 1;
2534
2535 for (i = 0; i < nr; i++) {
5f39d397 2536 item = btrfs_item_nr(right, i);
db94535d
CM		 2537 if (!right->map_token) {
2538 map_extent_buffer(right, (unsigned long)item,
2539 sizeof(struct btrfs_item),
2540 &right->map_token, &right->kaddr,
2541 &right->map_start, &right->map_len,
2542 KM_USER1);
2543 }
2544
31840ae1
ZY		 2545 if (!empty && push_items > 0) {
2546 if (path->slots[0] < i)
2547 break;
2548 if (path->slots[0] == i) {
2549 int space = btrfs_leaf_free_space(root, right);
2550 if (space + push_space * 2 > free_space)
2551 break;
2552 }
2553 }
2554
be0e5c09 2555 if (path->slots[0] == i)
87b29b20 2556 push_space += data_size;
db94535d
CM		 2557
2558 this_item_size = btrfs_item_size(right, item);
2559 if (this_item_size + sizeof(*item) + push_space > free_space)
be0e5c09 2560 break;
db94535d 2561
be0e5c09 2562 push_items++;
db94535d
CM		 2563 push_space += this_item_size + sizeof(*item);
2564 }
2565
2566 if (right->map_token) {
2567 unmap_extent_buffer(right, right->map_token, KM_USER1);
2568 right->map_token = NULL;
be0e5c09 2569 }
db94535d 2570
be0e5c09 2571 if (push_items == 0) {
925baedd
CM		 2572 ret = 1;
2573 goto out;
be0e5c09 2574 }
34a38218 2575 if (!empty && push_items == btrfs_header_nritems(right))
a429e513 2576 WARN_ON(1);
5f39d397 2577
be0e5c09 2578 /* push data from right to left */
5f39d397
CM		 2579 copy_extent_buffer(left, right,
2580 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2581 btrfs_item_nr_offset(0),
2582 push_items * sizeof(struct btrfs_item));
2583
123abc88 2584 push_space = BTRFS_LEAF_DATA_SIZE(root) -
d397712b 2585 btrfs_item_offset_nr(right, push_items - 1);
5f39d397
CM		 2586
2587 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
d6025579
CM		 2588 leaf_data_end(root, left) - push_space,
2589 btrfs_leaf_data(right) +
5f39d397 2590 btrfs_item_offset_nr(right, push_items - 1),
d6025579 2591 push_space);
5f39d397 2592 old_left_nritems = btrfs_header_nritems(left);
87b29b20 2593 BUG_ON(old_left_nritems <= 0);
eb60ceac 2594
db94535d 2595 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
0783fcfc 2596 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
5f39d397 2597 u32 ioff;
db94535d 2598
5f39d397 2599 item = btrfs_item_nr(left, i);
db94535d
CM		 2600 if (!left->map_token) {
2601 map_extent_buffer(left, (unsigned long)item,
2602 sizeof(struct btrfs_item),
2603 &left->map_token, &left->kaddr,
2604 &left->map_start, &left->map_len,
2605 KM_USER1);
2606 }
2607
5f39d397
CM		 2608 ioff = btrfs_item_offset(left, item);
2609 btrfs_set_item_offset(left, item,
db94535d 2610 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
be0e5c09 2611 }
5f39d397 2612 btrfs_set_header_nritems(left, old_left_nritems + push_items);
db94535d
CM		 2613 if (left->map_token) {
2614 unmap_extent_buffer(left, left->map_token, KM_USER1);
2615 left->map_token = NULL;
2616 }
be0e5c09
CM		 2617
2618 /* fixup right node */
34a38218 2619 if (push_items > right_nritems) {
d397712b
CM		 2620 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2621 right_nritems);
34a38218
CM		 2622 WARN_ON(1);
2623 }
2624
2625 if (push_items < right_nritems) {
2626 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2627 leaf_data_end(root, right);
2628 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2629 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2630 btrfs_leaf_data(right) +
2631 leaf_data_end(root, right), push_space);
2632
2633 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
5f39d397
CM		 2634 btrfs_item_nr_offset(push_items),
2635 (btrfs_header_nritems(right) - push_items) *
2636 sizeof(struct btrfs_item));
34a38218 2637 }
eef1c494
Y		 2638 right_nritems -= push_items;
2639 btrfs_set_header_nritems(right, right_nritems);
123abc88 2640 push_space = BTRFS_LEAF_DATA_SIZE(root);
5f39d397
CM		 2641 for (i = 0; i < right_nritems; i++) {
2642 item = btrfs_item_nr(right, i);
db94535d
CM		 2643
2644 if (!right->map_token) {
2645 map_extent_buffer(right, (unsigned long)item,
2646 sizeof(struct btrfs_item),
2647 &right->map_token, &right->kaddr,
2648 &right->map_start, &right->map_len,
2649 KM_USER1);
2650 }
2651
2652 push_space = push_space - btrfs_item_size(right, item);
2653 btrfs_set_item_offset(right, item, push_space);
2654 }
2655 if (right->map_token) {
2656 unmap_extent_buffer(right, right->map_token, KM_USER1);
2657 right->map_token = NULL;
be0e5c09 2658 }
eb60ceac 2659
5f39d397 2660 btrfs_mark_buffer_dirty(left);
34a38218
CM		 2661 if (right_nritems)
2662 btrfs_mark_buffer_dirty(right);
098f59c2 2663
5f39d397
CM		 2664 btrfs_item_key(right, &disk_key, 0);
2665 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
aa5d6bed
CM		 2666 if (wret)
2667 ret = wret;
be0e5c09
CM		 2668
2669 /* then fixup the leaf pointer in the path */
2670 if (path->slots[0] < push_items) {
2671 path->slots[0] += old_left_nritems;
925baedd
CM		 2672 if (btrfs_header_nritems(path->nodes[0]) == 0)
2673 clean_tree_block(trans, root, path->nodes[0]);
2674 btrfs_tree_unlock(path->nodes[0]);
5f39d397
CM		 2675 free_extent_buffer(path->nodes[0]);
2676 path->nodes[0] = left;
be0e5c09
CM		 2677 path->slots[1] -= 1;
2678 } else {
925baedd 2679 btrfs_tree_unlock(left);
5f39d397 2680 free_extent_buffer(left);
be0e5c09
CM		 2681 path->slots[0] -= push_items;
2682 }
eb60ceac 2683 BUG_ON(path->slots[0] < 0);
aa5d6bed 2684 return ret;
925baedd
CM		 2685out:
2686 btrfs_tree_unlock(left);
2687 free_extent_buffer(left);
2688 return ret;
be0e5c09
CM		 2689}
2690
44871b1b
CM		 2691/*
2692 * push some data in the path leaf to the left, trying to free up at
2693 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2694 */
2695static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2696 *root, struct btrfs_path *path, int data_size,
2697 int empty)
2698{
2699 struct extent_buffer *right = path->nodes[0];
2700 struct extent_buffer *left;
2701 int slot;
2702 int free_space;
2703 u32 right_nritems;
2704 int ret = 0;
2705
2706 slot = path->slots[1];
2707 if (slot == 0)
2708 return 1;
2709 if (!path->nodes[1])
2710 return 1;
2711
2712 right_nritems = btrfs_header_nritems(right);
2713 if (right_nritems == 0)
2714 return 1;
2715
2716 btrfs_assert_tree_locked(path->nodes[1]);
2717
2718 left = read_node_slot(root, path->nodes[1], slot - 1);
2719 btrfs_tree_lock(left);
2720 btrfs_set_lock_blocking(left);
2721
2722 free_space = btrfs_leaf_free_space(root, left);
2723 if (free_space < data_size) {
2724 ret = 1;
2725 goto out;
2726 }
2727
2728 /* cow and double check */
2729 ret = btrfs_cow_block(trans, root, left,
2730 path->nodes[1], slot - 1, &left);
2731 if (ret) {
2732 /* we hit -ENOSPC, but it isn't fatal here */
2733 ret = 1;
2734 goto out;
2735 }
2736
2737 free_space = btrfs_leaf_free_space(root, left);
2738 if (free_space < data_size) {
2739 ret = 1;
2740 goto out;
2741 }
2742
2743 return __push_leaf_left(trans, root, path, data_size,
2744 empty, left, free_space, right_nritems);
2745out:
2746 btrfs_tree_unlock(left);
2747 free_extent_buffer(left);
2748 return ret;
2749}
2750
2751/*
2752 * split the path's leaf in two, making sure there is at least data_size
2753 * available for the resulting leaf level of the path.
2754 *
2755 * returns 0 if all went well and < 0 on failure.
2756 */
2757static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2758 struct btrfs_root *root,
2759 struct btrfs_path *path,
2760 struct extent_buffer *l,
2761 struct extent_buffer *right,
2762 int slot, int mid, int nritems)
2763{
2764 int data_copy_size;
2765 int rt_data_off;
2766 int i;
2767 int ret = 0;
2768 int wret;
2769 struct btrfs_disk_key disk_key;
2770
2771 nritems = nritems - mid;
2772 btrfs_set_header_nritems(right, nritems);
2773 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2774
2775 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2776 btrfs_item_nr_offset(mid),
2777 nritems * sizeof(struct btrfs_item));
2778
2779 copy_extent_buffer(right, l,
2780 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2781 data_copy_size, btrfs_leaf_data(l) +
2782 leaf_data_end(root, l), data_copy_size);
2783
2784 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2785 btrfs_item_end_nr(l, mid);
2786
2787 for (i = 0; i < nritems; i++) {
2788 struct btrfs_item *item = btrfs_item_nr(right, i);
2789 u32 ioff;
2790
2791 if (!right->map_token) {
2792 map_extent_buffer(right, (unsigned long)item,
2793 sizeof(struct btrfs_item),
2794 &right->map_token, &right->kaddr,
2795 &right->map_start, &right->map_len,
2796 KM_USER1);
2797 }
2798
2799 ioff = btrfs_item_offset(right, item);
2800 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2801 }
2802
2803 if (right->map_token) {
2804 unmap_extent_buffer(right, right->map_token, KM_USER1);
2805 right->map_token = NULL;
2806 }
2807
2808 btrfs_set_header_nritems(l, mid);
2809 ret = 0;
2810 btrfs_item_key(right, &disk_key, 0);
2811 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2812 path->slots[1] + 1, 1);
2813 if (wret)
2814 ret = wret;
2815
2816 btrfs_mark_buffer_dirty(right);
2817 btrfs_mark_buffer_dirty(l);
2818 BUG_ON(path->slots[0] != slot);
2819
44871b1b
CM		 2820 if (mid <= slot) {
2821 btrfs_tree_unlock(path->nodes[0]);
2822 free_extent_buffer(path->nodes[0]);
2823 path->nodes[0] = right;
2824 path->slots[0] -= mid;
2825 path->slots[1] += 1;
2826 } else {
2827 btrfs_tree_unlock(right);
2828 free_extent_buffer(right);
2829 }
2830
2831 BUG_ON(path->slots[0] < 0);
2832
2833 return ret;
2834}
2835
74123bd7
CM		 2836/*
2837 * split the path's leaf in two, making sure there is at least data_size
2838 * available for the resulting leaf level of the path.
aa5d6bed
CM		 2839 *
2840 * returns 0 if all went well and < 0 on failure.
74123bd7 2841 */
e02119d5
CM		 2842static noinline int split_leaf(struct btrfs_trans_handle *trans,
2843 struct btrfs_root *root,
2844 struct btrfs_key *ins_key,
2845 struct btrfs_path *path, int data_size,
2846 int extend)
be0e5c09 2847{
5d4f98a2 2848 struct btrfs_disk_key disk_key;
5f39d397 2849 struct extent_buffer *l;
7518a238 2850 u32 nritems;
eb60ceac
CM		 2851 int mid;
2852 int slot;
5f39d397 2853 struct extent_buffer *right;
d4dbff95 2854 int ret = 0;
aa5d6bed 2855 int wret;
5d4f98a2 2856 int split;
cc0c5538 2857 int num_doubles = 0;
aa5d6bed 2858
a5719521
YZ		 2859 l = path->nodes[0];
2860 slot = path->slots[0];
2861 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2862 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2863 return -EOVERFLOW;
2864
40689478 2865 /* first try to make some room by pushing left and right */
b3612421 2866 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
34a38218 2867 wret = push_leaf_right(trans, root, path, data_size, 0);
d397712b 2868 if (wret < 0)
eaee50e8 2869 return wret;
3685f791 2870 if (wret) {
34a38218 2871 wret = push_leaf_left(trans, root, path, data_size, 0);
3685f791
CM		 2872 if (wret < 0)
2873 return wret;
2874 }
2875 l = path->nodes[0];
aa5d6bed 2876
3685f791 2877 /* did the pushes work? */
87b29b20 2878 if (btrfs_leaf_free_space(root, l) >= data_size)
3685f791 2879 return 0;
3326d1b0 2880 }
aa5d6bed 2881
5c680ed6 2882 if (!path->nodes[1]) {
e089f05c 2883 ret = insert_new_root(trans, root, path, 1);
5c680ed6
CM		 2884 if (ret)
2885 return ret;
2886 }
cc0c5538 2887again:
5d4f98a2 2888 split = 1;
cc0c5538 2889 l = path->nodes[0];
eb60ceac 2890 slot = path->slots[0];
5f39d397 2891 nritems = btrfs_header_nritems(l);
d397712b 2892 mid = (nritems + 1) / 2;
54aa1f4d 2893
5d4f98a2
YZ		 2894 if (mid <= slot) {
2895 if (nritems == 1 ||
2896 leaf_space_used(l, mid, nritems - mid) + data_size >
2897 BTRFS_LEAF_DATA_SIZE(root)) {
2898 if (slot >= nritems) {
2899 split = 0;
2900 } else {
2901 mid = slot;
2902 if (mid != nritems &&
2903 leaf_space_used(l, mid, nritems - mid) +
2904 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2905 split = 2;
2906 }
2907 }
2908 }
2909 } else {
2910 if (leaf_space_used(l, 0, mid) + data_size >
2911 BTRFS_LEAF_DATA_SIZE(root)) {
2912 if (!extend && data_size && slot == 0) {
2913 split = 0;
2914 } else if ((extend || !data_size) && slot == 0) {
2915 mid = 1;
2916 } else {
2917 mid = slot;
2918 if (mid != nritems &&
2919 leaf_space_used(l, mid, nritems - mid) +
2920 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2921 split = 2 ;
2922 }
2923 }
2924 }
2925 }
2926
2927 if (split == 0)
2928 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2929 else
2930 btrfs_item_key(l, &disk_key, mid);
2931
2932 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
31840ae1 2933 root->root_key.objectid,
5d4f98a2 2934 &disk_key, 0, l->start, 0);
cea9e445
CM		 2935 if (IS_ERR(right)) {
2936 BUG_ON(1);
5f39d397 2937 return PTR_ERR(right);
cea9e445 2938 }
5f39d397
CM		 2939
2940 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
db94535d 2941 btrfs_set_header_bytenr(right, right->start);
5f39d397 2942 btrfs_set_header_generation(right, trans->transid);
5d4f98a2 2943 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
5f39d397
CM		 2944 btrfs_set_header_owner(right, root->root_key.objectid);
2945 btrfs_set_header_level(right, 0);
2946 write_extent_buffer(right, root->fs_info->fsid,
2947 (unsigned long)btrfs_header_fsid(right),
2948 BTRFS_FSID_SIZE);
e17cade2
CM		 2949
2950 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2951 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2952 BTRFS_UUID_SIZE);
44871b1b 2953
5d4f98a2
YZ		 2954 if (split == 0) {
2955 if (mid <= slot) {
2956 btrfs_set_header_nritems(right, 0);
2957 wret = insert_ptr(trans, root, path,
2958 &disk_key, right->start,
2959 path->slots[1] + 1, 1);
2960 if (wret)
2961 ret = wret;
925baedd 2962
5d4f98a2
YZ		 2963 btrfs_tree_unlock(path->nodes[0]);
2964 free_extent_buffer(path->nodes[0]);
2965 path->nodes[0] = right;
2966 path->slots[0] = 0;
2967 path->slots[1] += 1;
2968 } else {
2969 btrfs_set_header_nritems(right, 0);
2970 wret = insert_ptr(trans, root, path,
2971 &disk_key,
2972 right->start,
2973 path->slots[1], 1);
2974 if (wret)
2975 ret = wret;
2976 btrfs_tree_unlock(path->nodes[0]);
2977 free_extent_buffer(path->nodes[0]);
2978 path->nodes[0] = right;
2979 path->slots[0] = 0;
2980 if (path->slots[1] == 0) {
2981 wret = fixup_low_keys(trans, root,
2982 path, &disk_key, 1);
d4dbff95
CM		 2983 if (wret)
2984 ret = wret;
5ee78ac7 2985 }
d4dbff95 2986 }
5d4f98a2
YZ		 2987 btrfs_mark_buffer_dirty(right);
2988 return ret;
d4dbff95 2989 }
74123bd7 2990
44871b1b 2991 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
31840ae1
ZY		 2992 BUG_ON(ret);
2993
5d4f98a2 2994 if (split == 2) {
cc0c5538
CM		 2995 BUG_ON(num_doubles != 0);
2996 num_doubles++;
2997 goto again;
a429e513 2998 }
44871b1b 2999
be0e5c09
CM		 3000 return ret;
3001}
3002
ad48fd75
YZ		 3003static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3004 struct btrfs_root *root,
3005 struct btrfs_path *path, int ins_len)
459931ec 3006{
ad48fd75 3007 struct btrfs_key key;
459931ec 3008 struct extent_buffer *leaf;
ad48fd75
YZ		 3009 struct btrfs_file_extent_item *fi;
3010 u64 extent_len = 0;
3011 u32 item_size;
3012 int ret;
459931ec
CM		 3013
3014 leaf = path->nodes[0];
ad48fd75
YZ		 3015 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3016
3017 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3018 key.type != BTRFS_EXTENT_CSUM_KEY);
3019
3020 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3021 return 0;
459931ec
CM		 3022
3023 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
ad48fd75
YZ		 3024 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3025 fi = btrfs_item_ptr(leaf, path->slots[0],
3026 struct btrfs_file_extent_item);
3027 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3028 }
459931ec
CM		 3029 btrfs_release_path(root, path);
3030
459931ec 3031 path->keep_locks = 1;
ad48fd75
YZ		 3032 path->search_for_split = 1;
3033 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
459931ec 3034 path->search_for_split = 0;
ad48fd75
YZ		 3035 if (ret < 0)
3036 goto err;
459931ec 3037
ad48fd75
YZ		 3038 ret = -EAGAIN;
3039 leaf = path->nodes[0];
459931ec 3040 /* if our item isn't there or got smaller, return now */
ad48fd75
YZ		 3041 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3042 goto err;
3043
109f6aef
CM		 3044 /* the leaf has changed, it now has room. return now */
3045 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3046 goto err;
3047
ad48fd75
YZ		 3048 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3049 fi = btrfs_item_ptr(leaf, path->slots[0],
3050 struct btrfs_file_extent_item);
3051 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3052 goto err;
459931ec
CM		 3053 }
3054
b9473439 3055 btrfs_set_path_blocking(path);
ad48fd75 3056 ret = split_leaf(trans, root, &key, path, ins_len, 1);
459931ec
CM		 3057 BUG_ON(ret);
3058
ad48fd75 3059 path->keep_locks = 0;
b9473439 3060 btrfs_unlock_up_safe(path, 1);
ad48fd75
YZ		 3061 return 0;
3062err:
3063 path->keep_locks = 0;
3064 return ret;
3065}
3066
3067static noinline int split_item(struct btrfs_trans_handle *trans,
3068 struct btrfs_root *root,
3069 struct btrfs_path *path,
3070 struct btrfs_key *new_key,
3071 unsigned long split_offset)
3072{
3073 struct extent_buffer *leaf;
3074 struct btrfs_item *item;
3075 struct btrfs_item *new_item;
3076 int slot;
3077 char *buf;
3078 u32 nritems;
3079 u32 item_size;
3080 u32 orig_offset;
3081 struct btrfs_disk_key disk_key;
3082
b9473439
CM		 3083 leaf = path->nodes[0];
3084 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3085
b4ce94de
CM		 3086 btrfs_set_path_blocking(path);
3087
459931ec
CM		 3088 item = btrfs_item_nr(leaf, path->slots[0]);
3089 orig_offset = btrfs_item_offset(leaf, item);
3090 item_size = btrfs_item_size(leaf, item);
3091
459931ec 3092 buf = kmalloc(item_size, GFP_NOFS);
ad48fd75
YZ		 3093 if (!buf)
3094 return -ENOMEM;
3095
459931ec
CM		 3096 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3097 path->slots[0]), item_size);
459931ec 3098
ad48fd75 3099 slot = path->slots[0] + 1;
459931ec 3100 nritems = btrfs_header_nritems(leaf);
459931ec
CM		 3101 if (slot != nritems) {
3102 /* shift the items */
3103 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
ad48fd75
YZ		 3104 btrfs_item_nr_offset(slot),
3105 (nritems - slot) * sizeof(struct btrfs_item));
459931ec
CM		 3106 }
3107
3108 btrfs_cpu_key_to_disk(&disk_key, new_key);
3109 btrfs_set_item_key(leaf, &disk_key, slot);
3110
3111 new_item = btrfs_item_nr(leaf, slot);
3112
3113 btrfs_set_item_offset(leaf, new_item, orig_offset);
3114 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3115
3116 btrfs_set_item_offset(leaf, item,
3117 orig_offset + item_size - split_offset);
3118 btrfs_set_item_size(leaf, item, split_offset);
3119
3120 btrfs_set_header_nritems(leaf, nritems + 1);
3121
3122 /* write the data for the start of the original item */
3123 write_extent_buffer(leaf, buf,
3124 btrfs_item_ptr_offset(leaf, path->slots[0]),
3125 split_offset);
3126
3127 /* write the data for the new item */
3128 write_extent_buffer(leaf, buf + split_offset,
3129 btrfs_item_ptr_offset(leaf, slot),
3130 item_size - split_offset);
3131 btrfs_mark_buffer_dirty(leaf);
3132
ad48fd75 3133 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
459931ec 3134 kfree(buf);
ad48fd75
YZ		 3135 return 0;
3136}
3137
3138/*
3139 * This function splits a single item into two items,
3140 * giving 'new_key' to the new item and splitting the
3141 * old one at split_offset (from the start of the item).
3142 *
3143 * The path may be released by this operation. After
3144 * the split, the path is pointing to the old item. The
3145 * new item is going to be in the same node as the old one.
3146 *
3147 * Note, the item being split must be smaller enough to live alone on
3148 * a tree block with room for one extra struct btrfs_item
3149 *
3150 * This allows us to split the item in place, keeping a lock on the
3151 * leaf the entire time.
3152 */
3153int btrfs_split_item(struct btrfs_trans_handle *trans,
3154 struct btrfs_root *root,
3155 struct btrfs_path *path,
3156 struct btrfs_key *new_key,
3157 unsigned long split_offset)
3158{
3159 int ret;
3160 ret = setup_leaf_for_split(trans, root, path,
3161 sizeof(struct btrfs_item));
3162 if (ret)
3163 return ret;
3164
3165 ret = split_item(trans, root, path, new_key, split_offset);
459931ec
CM		 3166 return ret;
3167}
3168
ad48fd75
YZ		 3169/*
3170 * This function duplicate a item, giving 'new_key' to the new item.
3171 * It guarantees both items live in the same tree leaf and the new item
3172 * is contiguous with the original item.
3173 *
3174 * This allows us to split file extent in place, keeping a lock on the
3175 * leaf the entire time.
3176 */
3177int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3178 struct btrfs_root *root,
3179 struct btrfs_path *path,
3180 struct btrfs_key *new_key)
3181{
3182 struct extent_buffer *leaf;
3183 int ret;
3184 u32 item_size;
3185
3186 leaf = path->nodes[0];
3187 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3188 ret = setup_leaf_for_split(trans, root, path,
3189 item_size + sizeof(struct btrfs_item));
3190 if (ret)
3191 return ret;
3192
3193 path->slots[0]++;
3194 ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3195 item_size, item_size +
3196 sizeof(struct btrfs_item), 1);
3197 BUG_ON(ret);
3198
3199 leaf = path->nodes[0];
3200 memcpy_extent_buffer(leaf,
3201 btrfs_item_ptr_offset(leaf, path->slots[0]),
3202 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3203 item_size);
3204 return 0;
3205}
3206
d352ac68
CM		 3207/*
3208 * make the item pointed to by the path smaller. new_size indicates
3209 * how small to make it, and from_end tells us if we just chop bytes
3210 * off the end of the item or if we shift the item to chop bytes off
3211 * the front.
3212 */
b18c6685
CM		 3213int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3214 struct btrfs_root *root,
3215 struct btrfs_path *path,
179e29e4 3216 u32 new_size, int from_end)
b18c6685
CM		 3217{
3218 int ret = 0;
3219 int slot;
3220 int slot_orig;
5f39d397
CM		 3221 struct extent_buffer *leaf;
3222 struct btrfs_item *item;
b18c6685
CM		 3223 u32 nritems;
3224 unsigned int data_end;
3225 unsigned int old_data_start;
3226 unsigned int old_size;
3227 unsigned int size_diff;
3228 int i;
3229
3230 slot_orig = path->slots[0];
5f39d397 3231 leaf = path->nodes[0];
179e29e4
CM		 3232 slot = path->slots[0];
3233
3234 old_size = btrfs_item_size_nr(leaf, slot);
3235 if (old_size == new_size)
3236 return 0;
b18c6685 3237
5f39d397 3238 nritems = btrfs_header_nritems(leaf);
b18c6685
CM		 3239 data_end = leaf_data_end(root, leaf);
3240
5f39d397 3241 old_data_start = btrfs_item_offset_nr(leaf, slot);
179e29e4 3242
b18c6685
CM		 3243 size_diff = old_size - new_size;
3244
3245 BUG_ON(slot < 0);
3246 BUG_ON(slot >= nritems);
3247
3248 /*
3249 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3250 */
3251 /* first correct the data pointers */
3252 for (i = slot; i < nritems; i++) {
5f39d397
CM		 3253 u32 ioff;
3254 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3255
3256 if (!leaf->map_token) {
3257 map_extent_buffer(leaf, (unsigned long)item,
3258 sizeof(struct btrfs_item),
3259 &leaf->map_token, &leaf->kaddr,
3260 &leaf->map_start, &leaf->map_len,
3261 KM_USER1);
3262 }
3263
5f39d397
CM		 3264 ioff = btrfs_item_offset(leaf, item);
3265 btrfs_set_item_offset(leaf, item, ioff + size_diff);
b18c6685 3266 }
db94535d
CM		 3267
3268 if (leaf->map_token) {
3269 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3270 leaf->map_token = NULL;
3271 }
3272
b18c6685 3273 /* shift the data */
179e29e4
CM		 3274 if (from_end) {
3275 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3276 data_end + size_diff, btrfs_leaf_data(leaf) +
3277 data_end, old_data_start + new_size - data_end);
3278 } else {
3279 struct btrfs_disk_key disk_key;
3280 u64 offset;
3281
3282 btrfs_item_key(leaf, &disk_key, slot);
3283
3284 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3285 unsigned long ptr;
3286 struct btrfs_file_extent_item *fi;
3287
3288 fi = btrfs_item_ptr(leaf, slot,
3289 struct btrfs_file_extent_item);
3290 fi = (struct btrfs_file_extent_item *)(
3291 (unsigned long)fi - size_diff);
3292
3293 if (btrfs_file_extent_type(leaf, fi) ==
3294 BTRFS_FILE_EXTENT_INLINE) {
3295 ptr = btrfs_item_ptr_offset(leaf, slot);
3296 memmove_extent_buffer(leaf, ptr,
d397712b
CM		 3297 (unsigned long)fi,
3298 offsetof(struct btrfs_file_extent_item,
179e29e4
CM		 3299 disk_bytenr));
3300 }
3301 }
3302
3303 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3304 data_end + size_diff, btrfs_leaf_data(leaf) +
3305 data_end, old_data_start - data_end);
3306
3307 offset = btrfs_disk_key_offset(&disk_key);
3308 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3309 btrfs_set_item_key(leaf, &disk_key, slot);
3310 if (slot == 0)
3311 fixup_low_keys(trans, root, path, &disk_key, 1);
3312 }
5f39d397
CM		 3313
3314 item = btrfs_item_nr(leaf, slot);
3315 btrfs_set_item_size(leaf, item, new_size);
3316 btrfs_mark_buffer_dirty(leaf);
b18c6685
CM		 3317
3318 ret = 0;
5f39d397
CM		 3319 if (btrfs_leaf_free_space(root, leaf) < 0) {
3320 btrfs_print_leaf(root, leaf);
b18c6685 3321 BUG();
5f39d397 3322 }
b18c6685
CM		 3323 return ret;
3324}
3325
d352ac68
CM		 3326/*
3327 * make the item pointed to by the path bigger, data_size is the new size.
3328 */
5f39d397
CM		 3329int btrfs_extend_item(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root, struct btrfs_path *path,
3331 u32 data_size)
6567e837
CM		 3332{
3333 int ret = 0;
3334 int slot;
3335 int slot_orig;
5f39d397
CM		 3336 struct extent_buffer *leaf;
3337 struct btrfs_item *item;
6567e837
CM		 3338 u32 nritems;
3339 unsigned int data_end;
3340 unsigned int old_data;
3341 unsigned int old_size;
3342 int i;
3343
3344 slot_orig = path->slots[0];
5f39d397 3345 leaf = path->nodes[0];
6567e837 3346
5f39d397 3347 nritems = btrfs_header_nritems(leaf);
6567e837
CM		 3348 data_end = leaf_data_end(root, leaf);
3349
5f39d397
CM		 3350 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3351 btrfs_print_leaf(root, leaf);
6567e837 3352 BUG();
5f39d397 3353 }
6567e837 3354 slot = path->slots[0];
5f39d397 3355 old_data = btrfs_item_end_nr(leaf, slot);
6567e837
CM		 3356
3357 BUG_ON(slot < 0);
3326d1b0
CM		 3358 if (slot >= nritems) {
3359 btrfs_print_leaf(root, leaf);
d397712b
CM		 3360 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3361 slot, nritems);
3326d1b0
CM		 3362 BUG_ON(1);
3363 }
6567e837
CM		 3364
3365 /*
3366 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3367 */
3368 /* first correct the data pointers */
3369 for (i = slot; i < nritems; i++) {
5f39d397
CM		 3370 u32 ioff;
3371 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3372
3373 if (!leaf->map_token) {
3374 map_extent_buffer(leaf, (unsigned long)item,
3375 sizeof(struct btrfs_item),
3376 &leaf->map_token, &leaf->kaddr,
3377 &leaf->map_start, &leaf->map_len,
3378 KM_USER1);
3379 }
5f39d397
CM		 3380 ioff = btrfs_item_offset(leaf, item);
3381 btrfs_set_item_offset(leaf, item, ioff - data_size);
6567e837 3382 }
5f39d397 3383
db94535d
CM		 3384 if (leaf->map_token) {
3385 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3386 leaf->map_token = NULL;
3387 }
3388
6567e837 3389 /* shift the data */
5f39d397 3390 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
6567e837
CM		 3391 data_end - data_size, btrfs_leaf_data(leaf) +
3392 data_end, old_data - data_end);
5f39d397 3393
6567e837 3394 data_end = old_data;
5f39d397
CM		 3395 old_size = btrfs_item_size_nr(leaf, slot);
3396 item = btrfs_item_nr(leaf, slot);
3397 btrfs_set_item_size(leaf, item, old_size + data_size);
3398 btrfs_mark_buffer_dirty(leaf);
6567e837
CM		 3399
3400 ret = 0;
5f39d397
CM		 3401 if (btrfs_leaf_free_space(root, leaf) < 0) {
3402 btrfs_print_leaf(root, leaf);
6567e837 3403 BUG();
5f39d397 3404 }
6567e837
CM		 3405 return ret;
3406}
3407
f3465ca4
JB		 3408/*
3409 * Given a key and some data, insert items into the tree.
3410 * This does all the path init required, making room in the tree if needed.
3411 * Returns the number of keys that were inserted.
3412 */
3413int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3414 struct btrfs_root *root,
3415 struct btrfs_path *path,
3416 struct btrfs_key *cpu_key, u32 *data_size,
3417 int nr)
3418{
3419 struct extent_buffer *leaf;
3420 struct btrfs_item *item;
3421 int ret = 0;
3422 int slot;
f3465ca4
JB		 3423 int i;
3424 u32 nritems;
3425 u32 total_data = 0;
3426 u32 total_size = 0;
3427 unsigned int data_end;
3428 struct btrfs_disk_key disk_key;
3429 struct btrfs_key found_key;
3430
87b29b20
YZ		 3431 for (i = 0; i < nr; i++) {
3432 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3433 BTRFS_LEAF_DATA_SIZE(root)) {
3434 break;
3435 nr = i;
3436 }
f3465ca4 3437 total_data += data_size[i];
87b29b20
YZ		 3438 total_size += data_size[i] + sizeof(struct btrfs_item);
3439 }
3440 BUG_ON(nr == 0);
f3465ca4 3441
f3465ca4
JB		 3442 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3443 if (ret == 0)
3444 return -EEXIST;
3445 if (ret < 0)
3446 goto out;
3447
f3465ca4
JB		 3448 leaf = path->nodes[0];
3449
3450 nritems = btrfs_header_nritems(leaf);
3451 data_end = leaf_data_end(root, leaf);
3452
3453 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3454 for (i = nr; i >= 0; i--) {
3455 total_data -= data_size[i];
3456 total_size -= data_size[i] + sizeof(struct btrfs_item);
3457 if (total_size < btrfs_leaf_free_space(root, leaf))
3458 break;
3459 }
3460 nr = i;
3461 }
3462
3463 slot = path->slots[0];
3464 BUG_ON(slot < 0);
3465
3466 if (slot != nritems) {
3467 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3468
3469 item = btrfs_item_nr(leaf, slot);
3470 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3471
3472 /* figure out how many keys we can insert in here */
3473 total_data = data_size[0];
3474 for (i = 1; i < nr; i++) {
5d4f98a2 3475 if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
f3465ca4
JB		 3476 break;
3477 total_data += data_size[i];
3478 }
3479 nr = i;
3480
3481 if (old_data < data_end) {
3482 btrfs_print_leaf(root, leaf);
d397712b 3483 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
f3465ca4
JB		 3484 slot, old_data, data_end);
3485 BUG_ON(1);
3486 }
3487 /*
3488 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3489 */
3490 /* first correct the data pointers */
3491 WARN_ON(leaf->map_token);
3492 for (i = slot; i < nritems; i++) {
3493 u32 ioff;
3494
3495 item = btrfs_item_nr(leaf, i);
3496 if (!leaf->map_token) {
3497 map_extent_buffer(leaf, (unsigned long)item,
3498 sizeof(struct btrfs_item),
3499 &leaf->map_token, &leaf->kaddr,
3500 &leaf->map_start, &leaf->map_len,
3501 KM_USER1);
3502 }
3503
3504 ioff = btrfs_item_offset(leaf, item);
3505 btrfs_set_item_offset(leaf, item, ioff - total_data);
3506 }
3507 if (leaf->map_token) {
3508 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3509 leaf->map_token = NULL;
3510 }
3511
3512 /* shift the items */
3513 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3514 btrfs_item_nr_offset(slot),
3515 (nritems - slot) * sizeof(struct btrfs_item));
3516
3517 /* shift the data */
3518 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3519 data_end - total_data, btrfs_leaf_data(leaf) +
3520 data_end, old_data - data_end);
3521 data_end = old_data;
3522 } else {
3523 /*
3524 * this sucks but it has to be done, if we are inserting at
3525 * the end of the leaf only insert 1 of the items, since we
3526 * have no way of knowing whats on the next leaf and we'd have
3527 * to drop our current locks to figure it out
3528 */
3529 nr = 1;
3530 }
3531
3532 /* setup the item for the new data */
3533 for (i = 0; i < nr; i++) {
3534 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3535 btrfs_set_item_key(leaf, &disk_key, slot + i);
3536 item = btrfs_item_nr(leaf, slot + i);
3537 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3538 data_end -= data_size[i];
3539 btrfs_set_item_size(leaf, item, data_size[i]);
3540 }
3541 btrfs_set_header_nritems(leaf, nritems + nr);
3542 btrfs_mark_buffer_dirty(leaf);
3543
3544 ret = 0;
3545 if (slot == 0) {
3546 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3547 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3548 }
3549
3550 if (btrfs_leaf_free_space(root, leaf) < 0) {
3551 btrfs_print_leaf(root, leaf);
3552 BUG();
3553 }
3554out:
3555 if (!ret)
3556 ret = nr;
3557 return ret;
3558}
3559
74123bd7 3560/*
44871b1b
CM		 3561 * this is a helper for btrfs_insert_empty_items, the main goal here is
3562 * to save stack depth by doing the bulk of the work in a function
3563 * that doesn't call btrfs_search_slot
74123bd7 3564 */
44871b1b
CM		 3565static noinline_for_stack int
3566setup_items_for_insert(struct btrfs_trans_handle *trans,
3567 struct btrfs_root *root, struct btrfs_path *path,
3568 struct btrfs_key *cpu_key, u32 *data_size,
3569 u32 total_data, u32 total_size, int nr)
be0e5c09 3570{
5f39d397 3571 struct btrfs_item *item;
9c58309d 3572 int i;
7518a238 3573 u32 nritems;
be0e5c09 3574 unsigned int data_end;
e2fa7227 3575 struct btrfs_disk_key disk_key;
44871b1b
CM		 3576 int ret;
3577 struct extent_buffer *leaf;
3578 int slot;
e2fa7227 3579
5f39d397 3580 leaf = path->nodes[0];
44871b1b 3581 slot = path->slots[0];
74123bd7 3582
5f39d397 3583 nritems = btrfs_header_nritems(leaf);
123abc88 3584 data_end = leaf_data_end(root, leaf);
eb60ceac 3585
f25956cc 3586 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3326d1b0 3587 btrfs_print_leaf(root, leaf);
d397712b 3588 printk(KERN_CRIT "not enough freespace need %u have %d\n",
9c58309d 3589 total_size, btrfs_leaf_free_space(root, leaf));
be0e5c09 3590 BUG();
d4dbff95 3591 }
5f39d397 3592
be0e5c09 3593 if (slot != nritems) {
5f39d397 3594 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
be0e5c09 3595
5f39d397
CM		 3596 if (old_data < data_end) {
3597 btrfs_print_leaf(root, leaf);
d397712b 3598 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
5f39d397
CM		 3599 slot, old_data, data_end);
3600 BUG_ON(1);
3601 }
be0e5c09
CM		 3602 /*
3603 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3604 */
3605 /* first correct the data pointers */
db94535d 3606 WARN_ON(leaf->map_token);
0783fcfc 3607 for (i = slot; i < nritems; i++) {
5f39d397 3608 u32 ioff;
db94535d 3609
5f39d397 3610 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3611 if (!leaf->map_token) {
3612 map_extent_buffer(leaf, (unsigned long)item,
3613 sizeof(struct btrfs_item),
3614 &leaf->map_token, &leaf->kaddr,
3615 &leaf->map_start, &leaf->map_len,
3616 KM_USER1);
3617 }
3618
5f39d397 3619 ioff = btrfs_item_offset(leaf, item);
9c58309d 3620 btrfs_set_item_offset(leaf, item, ioff - total_data);
0783fcfc 3621 }
db94535d
CM		 3622 if (leaf->map_token) {
3623 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3624 leaf->map_token = NULL;
3625 }
be0e5c09
CM		 3626
3627 /* shift the items */
9c58309d 3628 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
5f39d397 3629 btrfs_item_nr_offset(slot),
d6025579 3630 (nritems - slot) * sizeof(struct btrfs_item));
be0e5c09
CM		 3631
3632 /* shift the data */
5f39d397 3633 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
9c58309d 3634 data_end - total_data, btrfs_leaf_data(leaf) +
d6025579 3635 data_end, old_data - data_end);
be0e5c09
CM		 3636 data_end = old_data;
3637 }
5f39d397 3638
62e2749e 3639 /* setup the item for the new data */
9c58309d
CM		 3640 for (i = 0; i < nr; i++) {
3641 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3642 btrfs_set_item_key(leaf, &disk_key, slot + i);
3643 item = btrfs_item_nr(leaf, slot + i);
3644 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3645 data_end -= data_size[i];
3646 btrfs_set_item_size(leaf, item, data_size[i]);
3647 }
44871b1b 3648
9c58309d 3649 btrfs_set_header_nritems(leaf, nritems + nr);
aa5d6bed
CM		 3650
3651 ret = 0;
5a01a2e3 3652 if (slot == 0) {
44871b1b 3653 struct btrfs_disk_key disk_key;
5a01a2e3 3654 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
e089f05c 3655 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
5a01a2e3 3656 }
b9473439
CM		 3657 btrfs_unlock_up_safe(path, 1);
3658 btrfs_mark_buffer_dirty(leaf);
aa5d6bed 3659
5f39d397
CM		 3660 if (btrfs_leaf_free_space(root, leaf) < 0) {
3661 btrfs_print_leaf(root, leaf);
be0e5c09 3662 BUG();
5f39d397 3663 }
44871b1b
CM		 3664 return ret;
3665}
3666
3667/*
3668 * Given a key and some data, insert items into the tree.
3669 * This does all the path init required, making room in the tree if needed.
3670 */
3671int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3672 struct btrfs_root *root,
3673 struct btrfs_path *path,
3674 struct btrfs_key *cpu_key, u32 *data_size,
3675 int nr)
3676{
3677 struct extent_buffer *leaf;
3678 int ret = 0;
3679 int slot;
3680 int i;
3681 u32 total_size = 0;
3682 u32 total_data = 0;
3683
3684 for (i = 0; i < nr; i++)
3685 total_data += data_size[i];
3686
3687 total_size = total_data + (nr * sizeof(struct btrfs_item));
3688 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3689 if (ret == 0)
3690 return -EEXIST;
3691 if (ret < 0)
3692 goto out;
3693
3694 leaf = path->nodes[0];
3695 slot = path->slots[0];
3696 BUG_ON(slot < 0);
3697
3698 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3699 total_data, total_size, nr);
3700
ed2ff2cb 3701out:
62e2749e
CM		 3702 return ret;
3703}
3704
3705/*
3706 * Given a key and some data, insert an item into the tree.
3707 * This does all the path init required, making room in the tree if needed.
3708 */
e089f05c
CM		 3709int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3710 *root, struct btrfs_key *cpu_key, void *data, u32
3711 data_size)
62e2749e
CM		 3712{
3713 int ret = 0;
2c90e5d6 3714 struct btrfs_path *path;
5f39d397
CM		 3715 struct extent_buffer *leaf;
3716 unsigned long ptr;
62e2749e 3717
2c90e5d6
CM		 3718 path = btrfs_alloc_path();
3719 BUG_ON(!path);
2c90e5d6 3720 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
62e2749e 3721 if (!ret) {
5f39d397
CM		 3722 leaf = path->nodes[0];
3723 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724 write_extent_buffer(leaf, data, ptr, data_size);
3725 btrfs_mark_buffer_dirty(leaf);
62e2749e 3726 }
2c90e5d6 3727 btrfs_free_path(path);
aa5d6bed 3728 return ret;
be0e5c09
CM		 3729}
3730
74123bd7 3731/*
5de08d7d 3732 * delete the pointer from a given node.
74123bd7 3733 *
d352ac68
CM		 3734 * the tree should have been previously balanced so the deletion does not
3735 * empty a node.
74123bd7 3736 */
e089f05c
CM		 3737static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3738 struct btrfs_path *path, int level, int slot)
be0e5c09 3739{
5f39d397 3740 struct extent_buffer *parent = path->nodes[level];
7518a238 3741 u32 nritems;
aa5d6bed 3742 int ret = 0;
bb803951 3743 int wret;
be0e5c09 3744
5f39d397 3745 nritems = btrfs_header_nritems(parent);
d397712b 3746 if (slot != nritems - 1) {
5f39d397
CM		 3747 memmove_extent_buffer(parent,
3748 btrfs_node_key_ptr_offset(slot),
3749 btrfs_node_key_ptr_offset(slot + 1),
d6025579
CM		 3750 sizeof(struct btrfs_key_ptr) *
3751 (nritems - slot - 1));
bb803951 3752 }
7518a238 3753 nritems--;
5f39d397 3754 btrfs_set_header_nritems(parent, nritems);
7518a238 3755 if (nritems == 0 && parent == root->node) {
5f39d397 3756 BUG_ON(btrfs_header_level(root->node) != 1);
bb803951 3757 /* just turn the root into a leaf and break */
5f39d397 3758 btrfs_set_header_level(root->node, 0);
bb803951 3759 } else if (slot == 0) {
5f39d397
CM		 3760 struct btrfs_disk_key disk_key;
3761
3762 btrfs_node_key(parent, &disk_key, 0);
3763 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
0f70abe2
CM		 3764 if (wret)
3765 ret = wret;
be0e5c09 3766 }
d6025579 3767 btrfs_mark_buffer_dirty(parent);
aa5d6bed 3768 return ret;
be0e5c09
CM		 3769}
3770
323ac95b
CM		 3771/*
3772 * a helper function to delete the leaf pointed to by path->slots[1] and
5d4f98a2 3773 * path->nodes[1].
323ac95b
CM		 3774 *
3775 * This deletes the pointer in path->nodes[1] and frees the leaf
3776 * block extent. zero is returned if it all worked out, < 0 otherwise.
3777 *
3778 * The path must have already been setup for deleting the leaf, including
3779 * all the proper balancing. path->nodes[1] must be locked.
3780 */
5d4f98a2
YZ		 3781static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3782 struct btrfs_root *root,
3783 struct btrfs_path *path,
3784 struct extent_buffer *leaf)
323ac95b
CM		 3785{
3786 int ret;
323ac95b 3787
5d4f98a2 3788 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
323ac95b
CM		 3789 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3790 if (ret)
3791 return ret;
3792
4d081c41
CM		 3793 /*
3794 * btrfs_free_extent is expensive, we want to make sure we
3795 * aren't holding any locks when we call it
3796 */
3797 btrfs_unlock_up_safe(path, 0);
3798
86b9f2ec
YZ		 3799 ret = btrfs_free_tree_block(trans, root, leaf->start, leaf->len,
3800 0, root->root_key.objectid, 0);
323ac95b
CM		 3801 return ret;
3802}
74123bd7
CM		 3803/*
3804 * delete the item at the leaf level in path. If that empties
3805 * the leaf, remove it from the tree
3806 */
85e21bac
CM		 3807int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3808 struct btrfs_path *path, int slot, int nr)
be0e5c09 3809{
5f39d397
CM		 3810 struct extent_buffer *leaf;
3811 struct btrfs_item *item;
85e21bac
CM		 3812 int last_off;
3813 int dsize = 0;
aa5d6bed
CM		 3814 int ret = 0;
3815 int wret;
85e21bac 3816 int i;
7518a238 3817 u32 nritems;
be0e5c09 3818
5f39d397 3819 leaf = path->nodes[0];
85e21bac
CM		 3820 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3821
3822 for (i = 0; i < nr; i++)
3823 dsize += btrfs_item_size_nr(leaf, slot + i);
3824
5f39d397 3825 nritems = btrfs_header_nritems(leaf);
be0e5c09 3826
85e21bac 3827 if (slot + nr != nritems) {
123abc88 3828 int data_end = leaf_data_end(root, leaf);
5f39d397
CM		 3829
3830 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
d6025579
CM		 3831 data_end + dsize,
3832 btrfs_leaf_data(leaf) + data_end,
85e21bac 3833 last_off - data_end);
5f39d397 3834
85e21bac 3835 for (i = slot + nr; i < nritems; i++) {
5f39d397 3836 u32 ioff;
db94535d 3837
5f39d397 3838 item = btrfs_item_nr(leaf, i);
db94535d
CM		 3839 if (!leaf->map_token) {
3840 map_extent_buffer(leaf, (unsigned long)item,
3841 sizeof(struct btrfs_item),
3842 &leaf->map_token, &leaf->kaddr,
3843 &leaf->map_start, &leaf->map_len,
3844 KM_USER1);
3845 }
5f39d397
CM		 3846 ioff = btrfs_item_offset(leaf, item);
3847 btrfs_set_item_offset(leaf, item, ioff + dsize);
0783fcfc 3848 }
db94535d
CM		 3849
3850 if (leaf->map_token) {
3851 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3852 leaf->map_token = NULL;
3853 }
3854
5f39d397 3855 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
85e21bac 3856 btrfs_item_nr_offset(slot + nr),
d6025579 3857 sizeof(struct btrfs_item) *
85e21bac 3858 (nritems - slot - nr));
be0e5c09 3859 }
85e21bac
CM		 3860 btrfs_set_header_nritems(leaf, nritems - nr);
3861 nritems -= nr;
5f39d397 3862
74123bd7 3863 /* delete the leaf if we've emptied it */
7518a238 3864 if (nritems == 0) {
5f39d397
CM		 3865 if (leaf == root->node) {
3866 btrfs_set_header_level(leaf, 0);
9a8dd150 3867 } else {
5d4f98a2 3868 ret = btrfs_del_leaf(trans, root, path, leaf);
323ac95b 3869 BUG_ON(ret);
9a8dd150 3870 }
be0e5c09 3871 } else {
7518a238 3872 int used = leaf_space_used(leaf, 0, nritems);
aa5d6bed 3873 if (slot == 0) {
5f39d397
CM		 3874 struct btrfs_disk_key disk_key;
3875
3876 btrfs_item_key(leaf, &disk_key, 0);
e089f05c 3877 wret = fixup_low_keys(trans, root, path,
5f39d397 3878 &disk_key, 1);
aa5d6bed
CM		 3879 if (wret)
3880 ret = wret;
3881 }
aa5d6bed 3882
74123bd7 3883 /* delete the leaf if it is mostly empty */
d717aa1d 3884 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
be0e5c09
CM		 3885 /* push_leaf_left fixes the path.
3886 * make sure the path still points to our leaf
3887 * for possible call to del_ptr below
3888 */
4920c9ac 3889 slot = path->slots[1];
5f39d397
CM		 3890 extent_buffer_get(leaf);
3891
b9473439 3892 btrfs_set_path_blocking(path);
85e21bac 3893 wret = push_leaf_left(trans, root, path, 1, 1);
54aa1f4d 3894 if (wret < 0 && wret != -ENOSPC)
aa5d6bed 3895 ret = wret;
5f39d397
CM		 3896
3897 if (path->nodes[0] == leaf &&
3898 btrfs_header_nritems(leaf)) {
85e21bac 3899 wret = push_leaf_right(trans, root, path, 1, 1);
54aa1f4d 3900 if (wret < 0 && wret != -ENOSPC)
aa5d6bed
CM		 3901 ret = wret;
3902 }
5f39d397
CM		 3903
3904 if (btrfs_header_nritems(leaf) == 0) {
323ac95b 3905 path->slots[1] = slot;
5d4f98a2 3906 ret = btrfs_del_leaf(trans, root, path, leaf);
323ac95b 3907 BUG_ON(ret);
5f39d397 3908 free_extent_buffer(leaf);
5de08d7d 3909 } else {
925baedd
CM		 3910 /* if we're still in the path, make sure
3911 * we're dirty. Otherwise, one of the
3912 * push_leaf functions must have already
3913 * dirtied this buffer
3914 */
3915 if (path->nodes[0] == leaf)
3916 btrfs_mark_buffer_dirty(leaf);
5f39d397 3917 free_extent_buffer(leaf);
be0e5c09 3918 }
d5719762 3919 } else {
5f39d397 3920 btrfs_mark_buffer_dirty(leaf);
be0e5c09
CM		 3921 }
3922 }
aa5d6bed 3923 return ret;
be0e5c09
CM		 3924}
3925
7bb86316 3926/*
925baedd 3927 * search the tree again to find a leaf with lesser keys
7bb86316
CM		 3928 * returns 0 if it found something or 1 if there are no lesser leaves.
3929 * returns < 0 on io errors.
d352ac68
CM		 3930 *
3931 * This may release the path, and so you may lose any locks held at the
3932 * time you call it.
7bb86316
CM		 3933 */
3934int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3935{
925baedd
CM		 3936 struct btrfs_key key;
3937 struct btrfs_disk_key found_key;
3938 int ret;
7bb86316 3939
925baedd 3940 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
7bb86316 3941
925baedd
CM		 3942 if (key.offset > 0)
3943 key.offset--;
3944 else if (key.type > 0)
3945 key.type--;
3946 else if (key.objectid > 0)
3947 key.objectid--;
3948 else
3949 return 1;
7bb86316 3950
925baedd
CM		 3951 btrfs_release_path(root, path);
3952 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3953 if (ret < 0)
3954 return ret;
3955 btrfs_item_key(path->nodes[0], &found_key, 0);
3956 ret = comp_keys(&found_key, &key);
3957 if (ret < 0)
3958 return 0;
3959 return 1;
7bb86316
CM		 3960}
3961
3f157a2f
CM		 3962/*
3963 * A helper function to walk down the tree starting at min_key, and looking
3964 * for nodes or leaves that are either in cache or have a minimum
d352ac68 3965 * transaction id. This is used by the btree defrag code, and tree logging
3f157a2f
CM		 3966 *
3967 * This does not cow, but it does stuff the starting key it finds back
3968 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3969 * key and get a writable path.
3970 *
3971 * This does lock as it descends, and path->keep_locks should be set
3972 * to 1 by the caller.
3973 *
3974 * This honors path->lowest_level to prevent descent past a given level
3975 * of the tree.
3976 *
d352ac68
CM		 3977 * min_trans indicates the oldest transaction that you are interested
3978 * in walking through. Any nodes or leaves older than min_trans are
3979 * skipped over (without reading them).
3980 *
3f157a2f
CM		 3981 * returns zero if something useful was found, < 0 on error and 1 if there
3982 * was nothing in the tree that matched the search criteria.
3983 */
3984int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
e02119d5 3985 struct btrfs_key *max_key,
3f157a2f
CM		 3986 struct btrfs_path *path, int cache_only,
3987 u64 min_trans)
3988{
3989 struct extent_buffer *cur;
3990 struct btrfs_key found_key;
3991 int slot;
9652480b 3992 int sret;
3f157a2f
CM		 3993 u32 nritems;
3994 int level;
3995 int ret = 1;
3996
934d375b 3997 WARN_ON(!path->keep_locks);
3f157a2f
CM		 3998again:
3999 cur = btrfs_lock_root_node(root);
4000 level = btrfs_header_level(cur);
e02119d5 4001 WARN_ON(path->nodes[level]);
3f157a2f
CM		 4002 path->nodes[level] = cur;
4003 path->locks[level] = 1;
4004
4005 if (btrfs_header_generation(cur) < min_trans) {
4006 ret = 1;
4007 goto out;
4008 }
d397712b 4009 while (1) {
3f157a2f
CM		 4010 nritems = btrfs_header_nritems(cur);
4011 level = btrfs_header_level(cur);
9652480b 4012 sret = bin_search(cur, min_key, level, &slot);
3f157a2f 4013
323ac95b
CM		 4014 /* at the lowest level, we're done, setup the path and exit */
4015 if (level == path->lowest_level) {
e02119d5
CM		 4016 if (slot >= nritems)
4017 goto find_next_key;
3f157a2f
CM		 4018 ret = 0;
4019 path->slots[level] = slot;
4020 btrfs_item_key_to_cpu(cur, &found_key, slot);
4021 goto out;
4022 }
9652480b
Y		 4023 if (sret && slot > 0)
4024 slot--;
3f157a2f
CM		 4025 /*
4026 * check this node pointer against the cache_only and
4027 * min_trans parameters. If it isn't in cache or is too
4028 * old, skip to the next one.
4029 */
d397712b 4030 while (slot < nritems) {
3f157a2f
CM		 4031 u64 blockptr;
4032 u64 gen;
4033 struct extent_buffer *tmp;
e02119d5
CM		 4034 struct btrfs_disk_key disk_key;
4035
3f157a2f
CM		 4036 blockptr = btrfs_node_blockptr(cur, slot);
4037 gen = btrfs_node_ptr_generation(cur, slot);
4038 if (gen < min_trans) {
4039 slot++;
4040 continue;
4041 }
4042 if (!cache_only)
4043 break;
4044
e02119d5
CM		 4045 if (max_key) {
4046 btrfs_node_key(cur, &disk_key, slot);
4047 if (comp_keys(&disk_key, max_key) >= 0) {
4048 ret = 1;
4049 goto out;
4050 }
4051 }
4052
3f157a2f
CM		 4053 tmp = btrfs_find_tree_block(root, blockptr,
4054 btrfs_level_size(root, level - 1));
4055
4056 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4057 free_extent_buffer(tmp);
4058 break;
4059 }
4060 if (tmp)
4061 free_extent_buffer(tmp);
4062 slot++;
4063 }
e02119d5 4064find_next_key:
3f157a2f
CM		 4065 /*
4066 * we didn't find a candidate key in this node, walk forward
4067 * and find another one
4068 */
4069 if (slot >= nritems) {
e02119d5 4070 path->slots[level] = slot;
b4ce94de 4071 btrfs_set_path_blocking(path);
e02119d5 4072 sret = btrfs_find_next_key(root, path, min_key, level,
3f157a2f 4073 cache_only, min_trans);
e02119d5 4074 if (sret == 0) {
3f157a2f
CM		 4075 btrfs_release_path(root, path);
4076 goto again;
4077 } else {
4078 goto out;
4079 }
4080 }
4081 /* save our key for returning back */
4082 btrfs_node_key_to_cpu(cur, &found_key, slot);
4083 path->slots[level] = slot;
4084 if (level == path->lowest_level) {
4085 ret = 0;
4086 unlock_up(path, level, 1);
4087 goto out;
4088 }
b4ce94de 4089 btrfs_set_path_blocking(path);
3f157a2f
CM		 4090 cur = read_node_slot(root, cur, slot);
4091
4092 btrfs_tree_lock(cur);
b4ce94de 4093
3f157a2f
CM		 4094 path->locks[level - 1] = 1;
4095 path->nodes[level - 1] = cur;
4096 unlock_up(path, level, 1);
4008c04a 4097 btrfs_clear_path_blocking(path, NULL);
3f157a2f
CM		 4098 }
4099out:
4100 if (ret == 0)
4101 memcpy(min_key, &found_key, sizeof(found_key));
b4ce94de 4102 btrfs_set_path_blocking(path);
3f157a2f
CM		 4103 return ret;
4104}
4105
4106/*
4107 * this is similar to btrfs_next_leaf, but does not try to preserve
4108 * and fixup the path. It looks for and returns the next key in the
4109 * tree based on the current path and the cache_only and min_trans
4110 * parameters.
4111 *
4112 * 0 is returned if another key is found, < 0 if there are any errors
4113 * and 1 is returned if there are no higher keys in the tree
4114 *
4115 * path->keep_locks should be set to 1 on the search made before
4116 * calling this function.
4117 */
e7a84565 4118int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
33c66f43 4119 struct btrfs_key *key, int level,
3f157a2f 4120 int cache_only, u64 min_trans)
e7a84565 4121{
e7a84565
CM		 4122 int slot;
4123 struct extent_buffer *c;
4124
934d375b 4125 WARN_ON(!path->keep_locks);
d397712b 4126 while (level < BTRFS_MAX_LEVEL) {
e7a84565
CM		 4127 if (!path->nodes[level])
4128 return 1;
4129
4130 slot = path->slots[level] + 1;
4131 c = path->nodes[level];
3f157a2f 4132next:
e7a84565 4133 if (slot >= btrfs_header_nritems(c)) {
33c66f43
YZ		 4134 int ret;
4135 int orig_lowest;
4136 struct btrfs_key cur_key;
4137 if (level + 1 >= BTRFS_MAX_LEVEL ||
4138 !path->nodes[level + 1])
e7a84565 4139 return 1;
33c66f43
YZ		 4140
4141 if (path->locks[level + 1]) {
4142 level++;
4143 continue;
4144 }
4145
4146 slot = btrfs_header_nritems(c) - 1;
4147 if (level == 0)
4148 btrfs_item_key_to_cpu(c, &cur_key, slot);
4149 else
4150 btrfs_node_key_to_cpu(c, &cur_key, slot);
4151
4152 orig_lowest = path->lowest_level;
4153 btrfs_release_path(root, path);
4154 path->lowest_level = level;
4155 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4156 0, 0);
4157 path->lowest_level = orig_lowest;
4158 if (ret < 0)
4159 return ret;
4160
4161 c = path->nodes[level];
4162 slot = path->slots[level];
4163 if (ret == 0)
4164 slot++;
4165 goto next;
e7a84565 4166 }
33c66f43 4167
e7a84565
CM		 4168 if (level == 0)
4169 btrfs_item_key_to_cpu(c, key, slot);
3f157a2f
CM		 4170 else {
4171 u64 blockptr = btrfs_node_blockptr(c, slot);
4172 u64 gen = btrfs_node_ptr_generation(c, slot);
4173
4174 if (cache_only) {
4175 struct extent_buffer *cur;
4176 cur = btrfs_find_tree_block(root, blockptr,
4177 btrfs_level_size(root, level - 1));
4178 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4179 slot++;
4180 if (cur)
4181 free_extent_buffer(cur);
4182 goto next;
4183 }
4184 free_extent_buffer(cur);
4185 }
4186 if (gen < min_trans) {
4187 slot++;
4188 goto next;
4189 }
e7a84565 4190 btrfs_node_key_to_cpu(c, key, slot);
3f157a2f 4191 }
e7a84565
CM		 4192 return 0;
4193 }
4194 return 1;
4195}
4196
97571fd0 4197/*
925baedd 4198 * search the tree again to find a leaf with greater keys
0f70abe2
CM		 4199 * returns 0 if it found something or 1 if there are no greater leaves.
4200 * returns < 0 on io errors.
97571fd0 4201 */
234b63a0 4202int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
d97e63b6
CM		 4203{
4204 int slot;
8e73f275 4205 int level;
5f39d397 4206 struct extent_buffer *c;
8e73f275 4207 struct extent_buffer *next;
925baedd
CM		 4208 struct btrfs_key key;
4209 u32 nritems;
4210 int ret;
8e73f275
CM		 4211 int old_spinning = path->leave_spinning;
4212 int force_blocking = 0;
925baedd
CM		 4213
4214 nritems = btrfs_header_nritems(path->nodes[0]);
d397712b 4215 if (nritems == 0)
925baedd 4216 return 1;
925baedd 4217
8e73f275
CM		 4218 /*
4219 * we take the blocks in an order that upsets lockdep. Using
4220 * blocking mode is the only way around it.
4221 */
4222#ifdef CONFIG_DEBUG_LOCK_ALLOC
4223 force_blocking = 1;
4224#endif
925baedd 4225
8e73f275
CM		 4226 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4227again:
4228 level = 1;
4229 next = NULL;
925baedd 4230 btrfs_release_path(root, path);
8e73f275 4231
a2135011 4232 path->keep_locks = 1;
8e73f275
CM		 4233
4234 if (!force_blocking)
4235 path->leave_spinning = 1;
4236
925baedd
CM		 4237 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4238 path->keep_locks = 0;
4239
4240 if (ret < 0)
4241 return ret;
4242
a2135011 4243 nritems = btrfs_header_nritems(path->nodes[0]);
168fd7d2
CM		 4244 /*
4245 * by releasing the path above we dropped all our locks. A balance
4246 * could have added more items next to the key that used to be
4247 * at the very end of the block. So, check again here and
4248 * advance the path if there are now more items available.
4249 */
a2135011 4250 if (nritems > 0 && path->slots[0] < nritems - 1) {
e457afec
YZ		 4251 if (ret == 0)
4252 path->slots[0]++;
8e73f275 4253 ret = 0;
925baedd
CM		 4254 goto done;
4255 }
d97e63b6 4256
d397712b 4257 while (level < BTRFS_MAX_LEVEL) {
8e73f275
CM		 4258 if (!path->nodes[level]) {
4259 ret = 1;
4260 goto done;
4261 }
5f39d397 4262
d97e63b6
CM		 4263 slot = path->slots[level] + 1;
4264 c = path->nodes[level];
5f39d397 4265 if (slot >= btrfs_header_nritems(c)) {
d97e63b6 4266 level++;
8e73f275
CM		 4267 if (level == BTRFS_MAX_LEVEL) {
4268 ret = 1;
4269 goto done;
4270 }
d97e63b6
CM		 4271 continue;
4272 }
5f39d397 4273
925baedd
CM		 4274 if (next) {
4275 btrfs_tree_unlock(next);
5f39d397 4276 free_extent_buffer(next);
925baedd 4277 }
5f39d397 4278
8e73f275
CM		 4279 next = c;
4280 ret = read_block_for_search(NULL, root, path, &next, level,
4281 slot, &key);
4282 if (ret == -EAGAIN)
4283 goto again;
5f39d397 4284
76a05b35
CM		 4285 if (ret < 0) {
4286 btrfs_release_path(root, path);
4287 goto done;
4288 }
4289
5cd57b2c 4290 if (!path->skip_locking) {
8e73f275
CM		 4291 ret = btrfs_try_spin_lock(next);
4292 if (!ret) {
4293 btrfs_set_path_blocking(path);
4294 btrfs_tree_lock(next);
4295 if (!force_blocking)
4296 btrfs_clear_path_blocking(path, next);
4297 }
4298 if (force_blocking)
4299 btrfs_set_lock_blocking(next);
5cd57b2c 4300 }
d97e63b6
CM		 4301 break;
4302 }
4303 path->slots[level] = slot;
d397712b 4304 while (1) {
d97e63b6
CM		 4305 level--;
4306 c = path->nodes[level];
925baedd
CM		 4307 if (path->locks[level])
4308 btrfs_tree_unlock(c);
8e73f275 4309
5f39d397 4310 free_extent_buffer(c);
d97e63b6
CM		 4311 path->nodes[level] = next;
4312 path->slots[level] = 0;
a74a4b97
CM		 4313 if (!path->skip_locking)
4314 path->locks[level] = 1;
8e73f275 4315
d97e63b6
CM		 4316 if (!level)
4317 break;
b4ce94de 4318
8e73f275
CM		 4319 ret = read_block_for_search(NULL, root, path, &next, level,
4320 0, &key);
4321 if (ret == -EAGAIN)
4322 goto again;
4323
76a05b35
CM		 4324 if (ret < 0) {
4325 btrfs_release_path(root, path);
4326 goto done;
4327 }
4328
5cd57b2c 4329 if (!path->skip_locking) {
b9447ef8 4330 btrfs_assert_tree_locked(path->nodes[level]);
8e73f275
CM		 4331 ret = btrfs_try_spin_lock(next);
4332 if (!ret) {
4333 btrfs_set_path_blocking(path);
4334 btrfs_tree_lock(next);
4335 if (!force_blocking)
4336 btrfs_clear_path_blocking(path, next);
4337 }
4338 if (force_blocking)
4339 btrfs_set_lock_blocking(next);
5cd57b2c 4340 }
d97e63b6 4341 }
8e73f275 4342 ret = 0;
925baedd
CM		 4343done:
4344 unlock_up(path, 0, 1);
8e73f275
CM		 4345 path->leave_spinning = old_spinning;
4346 if (!old_spinning)
4347 btrfs_set_path_blocking(path);
4348
4349 return ret;
d97e63b6 4350}
0b86a832 4351
3f157a2f
CM		 4352/*
4353 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4354 * searching until it gets past min_objectid or finds an item of 'type'
4355 *
4356 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4357 */
0b86a832
CM		 4358int btrfs_previous_item(struct btrfs_root *root,
4359 struct btrfs_path *path, u64 min_objectid,
4360 int type)
4361{
4362 struct btrfs_key found_key;
4363 struct extent_buffer *leaf;
e02119d5 4364 u32 nritems;
0b86a832
CM		 4365 int ret;
4366
d397712b 4367 while (1) {
0b86a832 4368 if (path->slots[0] == 0) {
b4ce94de 4369 btrfs_set_path_blocking(path);
0b86a832
CM		 4370 ret = btrfs_prev_leaf(root, path);
4371 if (ret != 0)
4372 return ret;
4373 } else {
4374 path->slots[0]--;
4375 }
4376 leaf = path->nodes[0];
e02119d5
CM		 4377 nritems = btrfs_header_nritems(leaf);
4378 if (nritems == 0)
4379 return 1;
4380 if (path->slots[0] == nritems)
4381 path->slots[0]--;
4382
0b86a832 4383 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
e02119d5
CM		 4384 if (found_key.objectid < min_objectid)
4385 break;
0a4eefbb
YZ		 4386 if (found_key.type == type)
4387 return 0;
e02119d5
CM		 4388 if (found_key.objectid == min_objectid &&
4389 found_key.type < type)
4390 break;
0b86a832
CM		 4391 }
4392 return 1;
4393}
Clone of davem's net-next-2.6 tree