blob: 8686a3d2ab3a06bccefba2978836aae2db701252 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
Chris Masond352ac62008-09-29 15:18:18 -04002 * Copyright (C) 2007,2008 Oracle. All rights reserved.
Chris Mason6cbd5572007-06-12 09:07:21 -04003 *
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
Chris Masona6b6e752007-10-15 16:22:39 -040019#include <linux/sched.h>
Chris Masoneb60cea2007-02-02 09:18:22 -050020#include "ctree.h"
21#include "disk-io.h"
Chris Mason7f5c1512007-03-23 15:56:19 -040022#include "transaction.h"
Chris Mason5f39d392007-10-15 16:14:19 -040023#include "print-tree.h"
Chris Mason925baed2008-06-25 16:01:30 -040024#include "locking.h"
Chris Mason9a8dd152007-02-23 08:38:36 -050025
Chris Masone089f052007-03-16 16:20:31 -040026static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
Chris Masond4dbff92007-04-04 14:08:15 -040029 *root, struct btrfs_key *ins_key,
Chris Masoncc0c5532007-10-25 15:42:57 -040030 struct btrfs_path *path, int data_size, int extend);
Chris Mason5f39d392007-10-15 16:14:19 -040031static int push_node_left(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root, struct extent_buffer *dst,
Chris Mason971a1f62008-04-24 10:54:32 -040033 struct extent_buffer *src, int empty);
Chris Mason5f39d392007-10-15 16:14:19 -040034static int balance_node_right(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 struct extent_buffer *dst_buf,
37 struct extent_buffer *src_buf);
Chris Masone089f052007-03-16 16:20:31 -040038static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 struct btrfs_path *path, int level, int slot);
Chris Masond97e63b2007-02-20 16:40:44 -050040
Chris Mason2c90e5d2007-04-02 10:50:19 -040041struct btrfs_path *btrfs_alloc_path(void)
42{
Chris Masondf24a2b2007-04-04 09:36:31 -040043 struct btrfs_path *path;
Jeff Mahoneye00f7302009-02-12 14:11:25 -050044 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
45 if (path)
Chris Mason2cc58cf2007-08-27 16:49:44 -040046 path->reada = 1;
Chris Masondf24a2b2007-04-04 09:36:31 -040047 return path;
Chris Mason2c90e5d2007-04-02 10:50:19 -040048}
49
Chris Masonb4ce94d2009-02-04 09:25:08 -050050/*
51 * set all locked nodes in the path to blocking locks. This should
52 * be done before scheduling
53 */
54noinline void btrfs_set_path_blocking(struct btrfs_path *p)
55{
56 int i;
57 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
58 if (p->nodes[i] && p->locks[i])
59 btrfs_set_lock_blocking(p->nodes[i]);
60 }
61}
62
63/*
64 * reset all the locked nodes in the patch to spinning locks.
Chris Mason4008c042009-02-12 14:09:45 -050065 *
66 * held is used to keep lockdep happy, when lockdep is enabled
67 * we set held to a blocking lock before we go around and
68 * retake all the spinlocks in the path. You can safely use NULL
69 * for held
Chris Masonb4ce94d2009-02-04 09:25:08 -050070 */
Chris Mason4008c042009-02-12 14:09:45 -050071noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
72 struct extent_buffer *held)
Chris Masonb4ce94d2009-02-04 09:25:08 -050073{
74 int i;
Chris Mason4008c042009-02-12 14:09:45 -050075
76#ifdef CONFIG_DEBUG_LOCK_ALLOC
77 /* lockdep really cares that we take all of these spinlocks
78 * in the right order. If any of the locks in the path are not
79 * currently blocking, it is going to complain. So, make really
80 * really sure by forcing the path to blocking before we clear
81 * the path blocking.
82 */
83 if (held)
84 btrfs_set_lock_blocking(held);
85 btrfs_set_path_blocking(p);
86#endif
87
88 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
Chris Masonb4ce94d2009-02-04 09:25:08 -050089 if (p->nodes[i] && p->locks[i])
90 btrfs_clear_lock_blocking(p->nodes[i]);
91 }
Chris Mason4008c042009-02-12 14:09:45 -050092
93#ifdef CONFIG_DEBUG_LOCK_ALLOC
94 if (held)
95 btrfs_clear_lock_blocking(held);
96#endif
Chris Masonb4ce94d2009-02-04 09:25:08 -050097}
98
Chris Masond352ac62008-09-29 15:18:18 -040099/* this also releases the path */
Chris Mason2c90e5d2007-04-02 10:50:19 -0400100void btrfs_free_path(struct btrfs_path *p)
101{
Chris Masondf24a2b2007-04-04 09:36:31 -0400102 btrfs_release_path(NULL, p);
Chris Mason2c90e5d2007-04-02 10:50:19 -0400103 kmem_cache_free(btrfs_path_cachep, p);
104}
105
Chris Masond352ac62008-09-29 15:18:18 -0400106/*
107 * path release drops references on the extent buffers in the path
108 * and it drops any locks held by this path
109 *
110 * It is safe to call this on paths that no locks or extent buffers held.
111 */
Chris Masond3977122009-01-05 21:25:51 -0500112noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
Chris Masoneb60cea2007-02-02 09:18:22 -0500113{
114 int i;
Chris Masona2135012008-06-25 16:01:30 -0400115
Chris Mason234b63a2007-03-13 10:46:10 -0400116 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
Chris Mason3f157a22008-06-25 16:01:31 -0400117 p->slots[i] = 0;
Chris Masoneb60cea2007-02-02 09:18:22 -0500118 if (!p->nodes[i])
Chris Mason925baed2008-06-25 16:01:30 -0400119 continue;
120 if (p->locks[i]) {
121 btrfs_tree_unlock(p->nodes[i]);
122 p->locks[i] = 0;
123 }
Chris Mason5f39d392007-10-15 16:14:19 -0400124 free_extent_buffer(p->nodes[i]);
Chris Mason3f157a22008-06-25 16:01:31 -0400125 p->nodes[i] = NULL;
Chris Masoneb60cea2007-02-02 09:18:22 -0500126 }
127}
128
Chris Masond352ac62008-09-29 15:18:18 -0400129/*
130 * safely gets a reference on the root node of a tree. A lock
131 * is not taken, so a concurrent writer may put a different node
132 * at the root of the tree. See btrfs_lock_root_node for the
133 * looping required.
134 *
135 * The extent buffer returned by this has a reference taken, so
136 * it won't disappear. It may stop being the root of the tree
137 * at any time because there are no locks held.
138 */
Chris Mason925baed2008-06-25 16:01:30 -0400139struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
140{
141 struct extent_buffer *eb;
142 spin_lock(&root->node_lock);
143 eb = root->node;
144 extent_buffer_get(eb);
145 spin_unlock(&root->node_lock);
146 return eb;
147}
148
Chris Masond352ac62008-09-29 15:18:18 -0400149/* loop around taking references on and locking the root node of the
150 * tree until you end up with a lock on the root. A locked buffer
151 * is returned, with a reference held.
152 */
Chris Mason925baed2008-06-25 16:01:30 -0400153struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
154{
155 struct extent_buffer *eb;
156
Chris Masond3977122009-01-05 21:25:51 -0500157 while (1) {
Chris Mason925baed2008-06-25 16:01:30 -0400158 eb = btrfs_root_node(root);
159 btrfs_tree_lock(eb);
160
161 spin_lock(&root->node_lock);
162 if (eb == root->node) {
163 spin_unlock(&root->node_lock);
164 break;
165 }
166 spin_unlock(&root->node_lock);
167
168 btrfs_tree_unlock(eb);
169 free_extent_buffer(eb);
170 }
171 return eb;
172}
173
Chris Masond352ac62008-09-29 15:18:18 -0400174/* cowonly root (everything not a reference counted cow subvolume), just get
175 * put onto a simple dirty list. transaction.c walks this to make sure they
176 * get properly updated on disk.
177 */
Chris Mason0b86a832008-03-24 15:01:56 -0400178static void add_root_to_dirty_list(struct btrfs_root *root)
179{
180 if (root->track_dirty && list_empty(&root->dirty_list)) {
181 list_add(&root->dirty_list,
182 &root->fs_info->dirty_cowonly_roots);
183 }
184}
185
Chris Masond352ac62008-09-29 15:18:18 -0400186/*
187 * used by snapshot creation to make a copy of a root for a tree with
188 * a given objectid. The buffer with the new root node is returned in
189 * cow_ret, and this func returns zero on success or a negative error code.
190 */
Chris Masonbe20aa92007-12-17 20:14:01 -0500191int btrfs_copy_root(struct btrfs_trans_handle *trans,
192 struct btrfs_root *root,
193 struct extent_buffer *buf,
194 struct extent_buffer **cow_ret, u64 new_root_objectid)
195{
196 struct extent_buffer *cow;
197 u32 nritems;
198 int ret = 0;
199 int level;
Chris Mason4aec2b52007-12-18 16:25:45 -0500200 struct btrfs_root *new_root;
Chris Masonbe20aa92007-12-17 20:14:01 -0500201
Chris Mason4aec2b52007-12-18 16:25:45 -0500202 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
203 if (!new_root)
204 return -ENOMEM;
205
206 memcpy(new_root, root, sizeof(*new_root));
207 new_root->root_key.objectid = new_root_objectid;
Chris Masonbe20aa92007-12-17 20:14:01 -0500208
209 WARN_ON(root->ref_cows && trans->transid !=
210 root->fs_info->running_transaction->transid);
211 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
212
213 level = btrfs_header_level(buf);
214 nritems = btrfs_header_nritems(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -0400215
216 cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
217 new_root_objectid, trans->transid,
218 level, buf->start, 0);
Chris Mason4aec2b52007-12-18 16:25:45 -0500219 if (IS_ERR(cow)) {
220 kfree(new_root);
Chris Masonbe20aa92007-12-17 20:14:01 -0500221 return PTR_ERR(cow);
Chris Mason4aec2b52007-12-18 16:25:45 -0500222 }
Chris Masonbe20aa92007-12-17 20:14:01 -0500223
224 copy_extent_buffer(cow, buf, 0, 0, cow->len);
225 btrfs_set_header_bytenr(cow, cow->start);
226 btrfs_set_header_generation(cow, trans->transid);
227 btrfs_set_header_owner(cow, new_root_objectid);
Chris Mason63b10fc2008-04-01 11:21:32 -0400228 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
Chris Masonbe20aa92007-12-17 20:14:01 -0500229
Yan Zheng2b820322008-11-17 21:11:30 -0500230 write_extent_buffer(cow, root->fs_info->fsid,
231 (unsigned long)btrfs_header_fsid(cow),
232 BTRFS_FSID_SIZE);
233
Chris Masonbe20aa92007-12-17 20:14:01 -0500234 WARN_ON(btrfs_header_generation(buf) > trans->transid);
Zheng Yan31840ae2008-09-23 13:14:14 -0400235 ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
Chris Mason4aec2b52007-12-18 16:25:45 -0500236 kfree(new_root);
237
Chris Masonbe20aa92007-12-17 20:14:01 -0500238 if (ret)
239 return ret;
240
241 btrfs_mark_buffer_dirty(cow);
242 *cow_ret = cow;
243 return 0;
244}
245
Chris Masond352ac62008-09-29 15:18:18 -0400246/*
Chris Masond3977122009-01-05 21:25:51 -0500247 * does the dirty work in cow of a single block. The parent block (if
248 * supplied) is updated to point to the new cow copy. The new buffer is marked
249 * dirty and returned locked. If you modify the block it needs to be marked
250 * dirty again.
Chris Masond352ac62008-09-29 15:18:18 -0400251 *
252 * search_start -- an allocation hint for the new block
253 *
Chris Masond3977122009-01-05 21:25:51 -0500254 * empty_size -- a hint that you plan on doing more cow. This is the size in
255 * bytes the allocator should try to find free next to the block it returns.
256 * This is just a hint and may be ignored by the allocator.
Chris Masond352ac62008-09-29 15:18:18 -0400257 */
Chris Masond3977122009-01-05 21:25:51 -0500258static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
Chris Mason5f39d392007-10-15 16:14:19 -0400259 struct btrfs_root *root,
260 struct extent_buffer *buf,
261 struct extent_buffer *parent, int parent_slot,
262 struct extent_buffer **cow_ret,
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400263 u64 search_start, u64 empty_size)
Chris Mason6702ed42007-08-07 16:15:09 -0400264{
Zheng Yan31840ae2008-09-23 13:14:14 -0400265 u64 parent_start;
Chris Mason5f39d392007-10-15 16:14:19 -0400266 struct extent_buffer *cow;
Chris Mason7bb86312007-12-11 09:25:06 -0500267 u32 nritems;
Chris Mason6702ed42007-08-07 16:15:09 -0400268 int ret = 0;
Chris Mason7bb86312007-12-11 09:25:06 -0500269 int level;
Chris Mason925baed2008-06-25 16:01:30 -0400270 int unlock_orig = 0;
Chris Mason6702ed42007-08-07 16:15:09 -0400271
Chris Mason925baed2008-06-25 16:01:30 -0400272 if (*cow_ret == buf)
273 unlock_orig = 1;
274
Chris Masonb9447ef2009-03-09 11:45:38 -0400275 btrfs_assert_tree_locked(buf);
Chris Mason925baed2008-06-25 16:01:30 -0400276
Zheng Yan31840ae2008-09-23 13:14:14 -0400277 if (parent)
278 parent_start = parent->start;
279 else
280 parent_start = 0;
281
Chris Mason7bb86312007-12-11 09:25:06 -0500282 WARN_ON(root->ref_cows && trans->transid !=
283 root->fs_info->running_transaction->transid);
Chris Mason6702ed42007-08-07 16:15:09 -0400284 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
Chris Mason5f39d392007-10-15 16:14:19 -0400285
Chris Mason7bb86312007-12-11 09:25:06 -0500286 level = btrfs_header_level(buf);
287 nritems = btrfs_header_nritems(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -0400288
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400289 cow = btrfs_alloc_free_block(trans, root, buf->len,
290 parent_start, root->root_key.objectid,
291 trans->transid, level,
292 search_start, empty_size);
Chris Mason6702ed42007-08-07 16:15:09 -0400293 if (IS_ERR(cow))
294 return PTR_ERR(cow);
295
Chris Masonb4ce94d2009-02-04 09:25:08 -0500296 /* cow is set to blocking by btrfs_init_new_buffer */
297
Chris Mason5f39d392007-10-15 16:14:19 -0400298 copy_extent_buffer(cow, buf, 0, 0, cow->len);
Chris Masondb945352007-10-15 16:15:53 -0400299 btrfs_set_header_bytenr(cow, cow->start);
Chris Mason5f39d392007-10-15 16:14:19 -0400300 btrfs_set_header_generation(cow, trans->transid);
301 btrfs_set_header_owner(cow, root->root_key.objectid);
Chris Mason63b10fc2008-04-01 11:21:32 -0400302 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
Chris Mason6702ed42007-08-07 16:15:09 -0400303
Yan Zheng2b820322008-11-17 21:11:30 -0500304 write_extent_buffer(cow, root->fs_info->fsid,
305 (unsigned long)btrfs_header_fsid(cow),
306 BTRFS_FSID_SIZE);
307
Chris Mason5f39d392007-10-15 16:14:19 -0400308 WARN_ON(btrfs_header_generation(buf) > trans->transid);
309 if (btrfs_header_generation(buf) != trans->transid) {
Zheng Yan31840ae2008-09-23 13:14:14 -0400310 u32 nr_extents;
Zheng Yan31840ae2008-09-23 13:14:14 -0400311 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
Chris Mason6702ed42007-08-07 16:15:09 -0400312 if (ret)
313 return ret;
Zheng Yan31840ae2008-09-23 13:14:14 -0400314
315 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
316 WARN_ON(ret);
Zheng Yan1a40e232008-09-26 10:09:34 -0400317 } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
318 /*
319 * There are only two places that can drop reference to
320 * tree blocks owned by living reloc trees, one is here,
Yan Zhengf82d02d2008-10-29 14:49:05 -0400321 * the other place is btrfs_drop_subtree. In both places,
Zheng Yan1a40e232008-09-26 10:09:34 -0400322 * we check reference count while tree block is locked.
323 * Furthermore, if reference count is one, it won't get
324 * increased by someone else.
325 */
326 u32 refs;
327 ret = btrfs_lookup_extent_ref(trans, root, buf->start,
328 buf->len, &refs);
329 BUG_ON(ret);
330 if (refs == 1) {
331 ret = btrfs_update_ref(trans, root, buf, cow,
332 0, nritems);
333 clean_tree_block(trans, root, buf);
334 } else {
335 ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
336 }
337 BUG_ON(ret);
Chris Mason6702ed42007-08-07 16:15:09 -0400338 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -0400339 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
340 if (ret)
341 return ret;
Chris Mason6702ed42007-08-07 16:15:09 -0400342 clean_tree_block(trans, root, buf);
343 }
344
Zheng Yan1a40e232008-09-26 10:09:34 -0400345 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
Zheng Yan1a40e232008-09-26 10:09:34 -0400346 ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
347 WARN_ON(ret);
348 }
349
Chris Mason6702ed42007-08-07 16:15:09 -0400350 if (buf == root->node) {
Chris Mason925baed2008-06-25 16:01:30 -0400351 WARN_ON(parent && parent != buf);
Chris Mason925baed2008-06-25 16:01:30 -0400352
353 spin_lock(&root->node_lock);
Chris Mason6702ed42007-08-07 16:15:09 -0400354 root->node = cow;
Chris Mason5f39d392007-10-15 16:14:19 -0400355 extent_buffer_get(cow);
Chris Mason925baed2008-06-25 16:01:30 -0400356 spin_unlock(&root->node_lock);
357
Chris Mason6702ed42007-08-07 16:15:09 -0400358 if (buf != root->commit_root) {
Chris Masondb945352007-10-15 16:15:53 -0400359 btrfs_free_extent(trans, root, buf->start,
Zheng Yan31840ae2008-09-23 13:14:14 -0400360 buf->len, buf->start,
361 root->root_key.objectid,
362 btrfs_header_generation(buf),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -0400363 level, 1);
Chris Mason6702ed42007-08-07 16:15:09 -0400364 }
Chris Mason5f39d392007-10-15 16:14:19 -0400365 free_extent_buffer(buf);
Chris Mason0b86a832008-03-24 15:01:56 -0400366 add_root_to_dirty_list(root);
Chris Mason6702ed42007-08-07 16:15:09 -0400367 } else {
Chris Mason5f39d392007-10-15 16:14:19 -0400368 btrfs_set_node_blockptr(parent, parent_slot,
Chris Masondb945352007-10-15 16:15:53 -0400369 cow->start);
Chris Mason74493f72007-12-11 09:25:06 -0500370 WARN_ON(trans->transid == 0);
371 btrfs_set_node_ptr_generation(parent, parent_slot,
372 trans->transid);
Chris Mason6702ed42007-08-07 16:15:09 -0400373 btrfs_mark_buffer_dirty(parent);
Chris Mason5f39d392007-10-15 16:14:19 -0400374 WARN_ON(btrfs_header_generation(parent) != trans->transid);
Chris Mason7bb86312007-12-11 09:25:06 -0500375 btrfs_free_extent(trans, root, buf->start, buf->len,
Zheng Yan31840ae2008-09-23 13:14:14 -0400376 parent_start, btrfs_header_owner(parent),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -0400377 btrfs_header_generation(parent), level, 1);
Chris Mason6702ed42007-08-07 16:15:09 -0400378 }
Chris Mason925baed2008-06-25 16:01:30 -0400379 if (unlock_orig)
380 btrfs_tree_unlock(buf);
Chris Mason5f39d392007-10-15 16:14:19 -0400381 free_extent_buffer(buf);
Chris Mason6702ed42007-08-07 16:15:09 -0400382 btrfs_mark_buffer_dirty(cow);
383 *cow_ret = cow;
384 return 0;
385}
386
Chris Masond352ac62008-09-29 15:18:18 -0400387/*
388 * cows a single block, see __btrfs_cow_block for the real work.
389 * This version of it has extra checks so that a block isn't cow'd more than
390 * once per transaction, as long as it hasn't been written yet
391 */
Chris Masond3977122009-01-05 21:25:51 -0500392noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
Chris Mason5f39d392007-10-15 16:14:19 -0400393 struct btrfs_root *root, struct extent_buffer *buf,
394 struct extent_buffer *parent, int parent_slot,
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400395 struct extent_buffer **cow_ret)
Chris Mason02217ed2007-03-02 16:08:05 -0500396{
Chris Mason6702ed42007-08-07 16:15:09 -0400397 u64 search_start;
Chris Masonf510cfe2007-10-15 16:14:48 -0400398 int ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500399
Chris Masonccd467d2007-06-28 15:57:36 -0400400 if (trans->transaction != root->fs_info->running_transaction) {
Chris Masond3977122009-01-05 21:25:51 -0500401 printk(KERN_CRIT "trans %llu running %llu\n",
402 (unsigned long long)trans->transid,
403 (unsigned long long)
Chris Masonccd467d2007-06-28 15:57:36 -0400404 root->fs_info->running_transaction->transid);
405 WARN_ON(1);
406 }
407 if (trans->transid != root->fs_info->generation) {
Chris Masond3977122009-01-05 21:25:51 -0500408 printk(KERN_CRIT "trans %llu running %llu\n",
409 (unsigned long long)trans->transid,
410 (unsigned long long)root->fs_info->generation);
Chris Masonccd467d2007-06-28 15:57:36 -0400411 WARN_ON(1);
412 }
Chris Masondc17ff82008-01-08 15:46:30 -0500413
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400414 if (btrfs_header_generation(buf) == trans->transid &&
415 btrfs_header_owner(buf) == root->root_key.objectid &&
Chris Mason63b10fc2008-04-01 11:21:32 -0400416 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
Chris Mason02217ed2007-03-02 16:08:05 -0500417 *cow_ret = buf;
418 return 0;
419 }
Chris Masonc4876852009-02-04 09:24:25 -0500420
Chris Mason0b86a832008-03-24 15:01:56 -0400421 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
Chris Masonb4ce94d2009-02-04 09:25:08 -0500422
423 if (parent)
424 btrfs_set_lock_blocking(parent);
425 btrfs_set_lock_blocking(buf);
426
Chris Masonf510cfe2007-10-15 16:14:48 -0400427 ret = __btrfs_cow_block(trans, root, buf, parent,
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400428 parent_slot, cow_ret, search_start, 0);
Chris Masonf510cfe2007-10-15 16:14:48 -0400429 return ret;
Chris Mason6702ed42007-08-07 16:15:09 -0400430}
431
Chris Masond352ac62008-09-29 15:18:18 -0400432/*
433 * helper function for defrag to decide if two blocks pointed to by a
434 * node are actually close by
435 */
Chris Mason6b800532007-10-15 16:17:34 -0400436static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
Chris Mason6702ed42007-08-07 16:15:09 -0400437{
Chris Mason6b800532007-10-15 16:17:34 -0400438 if (blocknr < other && other - (blocknr + blocksize) < 32768)
Chris Mason6702ed42007-08-07 16:15:09 -0400439 return 1;
Chris Mason6b800532007-10-15 16:17:34 -0400440 if (blocknr > other && blocknr - (other + blocksize) < 32768)
Chris Mason6702ed42007-08-07 16:15:09 -0400441 return 1;
Chris Mason02217ed2007-03-02 16:08:05 -0500442 return 0;
443}
444
Chris Mason081e9572007-11-06 10:26:24 -0500445/*
446 * compare two keys in a memcmp fashion
447 */
448static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
449{
450 struct btrfs_key k1;
451
452 btrfs_disk_key_to_cpu(&k1, disk);
453
454 if (k1.objectid > k2->objectid)
455 return 1;
456 if (k1.objectid < k2->objectid)
457 return -1;
458 if (k1.type > k2->type)
459 return 1;
460 if (k1.type < k2->type)
461 return -1;
462 if (k1.offset > k2->offset)
463 return 1;
464 if (k1.offset < k2->offset)
465 return -1;
466 return 0;
467}
468
Josef Bacikf3465ca2008-11-12 14:19:50 -0500469/*
470 * same as comp_keys only with two btrfs_key's
471 */
472static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
473{
474 if (k1->objectid > k2->objectid)
475 return 1;
476 if (k1->objectid < k2->objectid)
477 return -1;
478 if (k1->type > k2->type)
479 return 1;
480 if (k1->type < k2->type)
481 return -1;
482 if (k1->offset > k2->offset)
483 return 1;
484 if (k1->offset < k2->offset)
485 return -1;
486 return 0;
487}
Chris Mason081e9572007-11-06 10:26:24 -0500488
Chris Masond352ac62008-09-29 15:18:18 -0400489/*
490 * this is used by the defrag code to go through all the
491 * leaves pointed to by a node and reallocate them so that
492 * disk order is close to key order
493 */
Chris Mason6702ed42007-08-07 16:15:09 -0400494int btrfs_realloc_node(struct btrfs_trans_handle *trans,
Chris Mason5f39d392007-10-15 16:14:19 -0400495 struct btrfs_root *root, struct extent_buffer *parent,
Chris Masona6b6e752007-10-15 16:22:39 -0400496 int start_slot, int cache_only, u64 *last_ret,
497 struct btrfs_key *progress)
Chris Mason6702ed42007-08-07 16:15:09 -0400498{
Chris Mason6b800532007-10-15 16:17:34 -0400499 struct extent_buffer *cur;
Chris Mason6702ed42007-08-07 16:15:09 -0400500 u64 blocknr;
Chris Masonca7a79a2008-05-12 12:59:19 -0400501 u64 gen;
Chris Masone9d0b132007-08-10 14:06:19 -0400502 u64 search_start = *last_ret;
503 u64 last_block = 0;
Chris Mason6702ed42007-08-07 16:15:09 -0400504 u64 other;
505 u32 parent_nritems;
Chris Mason6702ed42007-08-07 16:15:09 -0400506 int end_slot;
507 int i;
508 int err = 0;
Chris Masonf2183bd2007-08-10 14:42:37 -0400509 int parent_level;
Chris Mason6b800532007-10-15 16:17:34 -0400510 int uptodate;
511 u32 blocksize;
Chris Mason081e9572007-11-06 10:26:24 -0500512 int progress_passed = 0;
513 struct btrfs_disk_key disk_key;
Chris Mason6702ed42007-08-07 16:15:09 -0400514
Chris Mason5708b952007-10-25 15:43:18 -0400515 parent_level = btrfs_header_level(parent);
516 if (cache_only && parent_level != 1)
517 return 0;
518
Chris Masond3977122009-01-05 21:25:51 -0500519 if (trans->transaction != root->fs_info->running_transaction)
Chris Mason6702ed42007-08-07 16:15:09 -0400520 WARN_ON(1);
Chris Masond3977122009-01-05 21:25:51 -0500521 if (trans->transid != root->fs_info->generation)
Chris Mason6702ed42007-08-07 16:15:09 -0400522 WARN_ON(1);
Chris Mason86479a02007-09-10 19:58:16 -0400523
Chris Mason6b800532007-10-15 16:17:34 -0400524 parent_nritems = btrfs_header_nritems(parent);
Chris Mason6b800532007-10-15 16:17:34 -0400525 blocksize = btrfs_level_size(root, parent_level - 1);
Chris Mason6702ed42007-08-07 16:15:09 -0400526 end_slot = parent_nritems;
527
528 if (parent_nritems == 1)
529 return 0;
530
Chris Masonb4ce94d2009-02-04 09:25:08 -0500531 btrfs_set_lock_blocking(parent);
532
Chris Mason6702ed42007-08-07 16:15:09 -0400533 for (i = start_slot; i < end_slot; i++) {
534 int close = 1;
Chris Masona6b6e752007-10-15 16:22:39 -0400535
Chris Mason5708b952007-10-25 15:43:18 -0400536 if (!parent->map_token) {
537 map_extent_buffer(parent,
538 btrfs_node_key_ptr_offset(i),
539 sizeof(struct btrfs_key_ptr),
540 &parent->map_token, &parent->kaddr,
541 &parent->map_start, &parent->map_len,
542 KM_USER1);
543 }
Chris Mason081e9572007-11-06 10:26:24 -0500544 btrfs_node_key(parent, &disk_key, i);
545 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
546 continue;
547
548 progress_passed = 1;
Chris Mason6b800532007-10-15 16:17:34 -0400549 blocknr = btrfs_node_blockptr(parent, i);
Chris Masonca7a79a2008-05-12 12:59:19 -0400550 gen = btrfs_node_ptr_generation(parent, i);
Chris Masone9d0b132007-08-10 14:06:19 -0400551 if (last_block == 0)
552 last_block = blocknr;
Chris Mason5708b952007-10-25 15:43:18 -0400553
Chris Mason6702ed42007-08-07 16:15:09 -0400554 if (i > 0) {
Chris Mason6b800532007-10-15 16:17:34 -0400555 other = btrfs_node_blockptr(parent, i - 1);
556 close = close_blocks(blocknr, other, blocksize);
Chris Mason6702ed42007-08-07 16:15:09 -0400557 }
Chris Mason0ef3e662008-05-24 14:04:53 -0400558 if (!close && i < end_slot - 2) {
Chris Mason6b800532007-10-15 16:17:34 -0400559 other = btrfs_node_blockptr(parent, i + 1);
560 close = close_blocks(blocknr, other, blocksize);
Chris Mason6702ed42007-08-07 16:15:09 -0400561 }
Chris Masone9d0b132007-08-10 14:06:19 -0400562 if (close) {
563 last_block = blocknr;
Chris Mason6702ed42007-08-07 16:15:09 -0400564 continue;
Chris Masone9d0b132007-08-10 14:06:19 -0400565 }
Chris Mason5708b952007-10-25 15:43:18 -0400566 if (parent->map_token) {
567 unmap_extent_buffer(parent, parent->map_token,
568 KM_USER1);
569 parent->map_token = NULL;
570 }
Chris Mason6702ed42007-08-07 16:15:09 -0400571
Chris Mason6b800532007-10-15 16:17:34 -0400572 cur = btrfs_find_tree_block(root, blocknr, blocksize);
573 if (cur)
Chris Mason1259ab72008-05-12 13:39:03 -0400574 uptodate = btrfs_buffer_uptodate(cur, gen);
Chris Mason6b800532007-10-15 16:17:34 -0400575 else
576 uptodate = 0;
Chris Mason5708b952007-10-25 15:43:18 -0400577 if (!cur || !uptodate) {
Chris Mason6702ed42007-08-07 16:15:09 -0400578 if (cache_only) {
Chris Mason6b800532007-10-15 16:17:34 -0400579 free_extent_buffer(cur);
Chris Mason6702ed42007-08-07 16:15:09 -0400580 continue;
581 }
Chris Mason6b800532007-10-15 16:17:34 -0400582 if (!cur) {
583 cur = read_tree_block(root, blocknr,
Chris Masonca7a79a2008-05-12 12:59:19 -0400584 blocksize, gen);
Chris Mason6b800532007-10-15 16:17:34 -0400585 } else if (!uptodate) {
Chris Masonca7a79a2008-05-12 12:59:19 -0400586 btrfs_read_buffer(cur, gen);
Chris Masonf2183bd2007-08-10 14:42:37 -0400587 }
Chris Mason6702ed42007-08-07 16:15:09 -0400588 }
Chris Masone9d0b132007-08-10 14:06:19 -0400589 if (search_start == 0)
Chris Mason6b800532007-10-15 16:17:34 -0400590 search_start = last_block;
Chris Masone9d0b132007-08-10 14:06:19 -0400591
Chris Masone7a84562008-06-25 16:01:31 -0400592 btrfs_tree_lock(cur);
Chris Masonb4ce94d2009-02-04 09:25:08 -0500593 btrfs_set_lock_blocking(cur);
Chris Mason6b800532007-10-15 16:17:34 -0400594 err = __btrfs_cow_block(trans, root, cur, parent, i,
Chris Masone7a84562008-06-25 16:01:31 -0400595 &cur, search_start,
Chris Mason6b800532007-10-15 16:17:34 -0400596 min(16 * blocksize,
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400597 (end_slot - i) * blocksize));
Yan252c38f2007-08-29 09:11:44 -0400598 if (err) {
Chris Masone7a84562008-06-25 16:01:31 -0400599 btrfs_tree_unlock(cur);
Chris Mason6b800532007-10-15 16:17:34 -0400600 free_extent_buffer(cur);
Chris Mason6702ed42007-08-07 16:15:09 -0400601 break;
Yan252c38f2007-08-29 09:11:44 -0400602 }
Chris Masone7a84562008-06-25 16:01:31 -0400603 search_start = cur->start;
604 last_block = cur->start;
Chris Masonf2183bd2007-08-10 14:42:37 -0400605 *last_ret = search_start;
Chris Masone7a84562008-06-25 16:01:31 -0400606 btrfs_tree_unlock(cur);
607 free_extent_buffer(cur);
Chris Mason6702ed42007-08-07 16:15:09 -0400608 }
Chris Mason5708b952007-10-25 15:43:18 -0400609 if (parent->map_token) {
610 unmap_extent_buffer(parent, parent->map_token,
611 KM_USER1);
612 parent->map_token = NULL;
613 }
Chris Mason6702ed42007-08-07 16:15:09 -0400614 return err;
615}
616
Chris Mason74123bd2007-02-02 11:05:29 -0500617/*
618 * The leaf data grows from end-to-front in the node.
619 * this returns the address of the start of the last item,
620 * which is the stop of the leaf data stack
621 */
Chris Mason123abc82007-03-14 14:14:43 -0400622static inline unsigned int leaf_data_end(struct btrfs_root *root,
Chris Mason5f39d392007-10-15 16:14:19 -0400623 struct extent_buffer *leaf)
Chris Masonbe0e5c02007-01-26 15:51:26 -0500624{
Chris Mason5f39d392007-10-15 16:14:19 -0400625 u32 nr = btrfs_header_nritems(leaf);
Chris Masonbe0e5c02007-01-26 15:51:26 -0500626 if (nr == 0)
Chris Mason123abc82007-03-14 14:14:43 -0400627 return BTRFS_LEAF_DATA_SIZE(root);
Chris Mason5f39d392007-10-15 16:14:19 -0400628 return btrfs_item_offset_nr(leaf, nr - 1);
Chris Masonbe0e5c02007-01-26 15:51:26 -0500629}
630
Chris Masond352ac62008-09-29 15:18:18 -0400631/*
632 * extra debugging checks to make sure all the items in a key are
633 * well formed and in the proper order
634 */
Chris Mason123abc82007-03-14 14:14:43 -0400635static int check_node(struct btrfs_root *root, struct btrfs_path *path,
636 int level)
Chris Masonaa5d6be2007-02-28 16:35:06 -0500637{
Chris Mason5f39d392007-10-15 16:14:19 -0400638 struct extent_buffer *parent = NULL;
639 struct extent_buffer *node = path->nodes[level];
640 struct btrfs_disk_key parent_key;
641 struct btrfs_disk_key node_key;
Chris Masonaa5d6be2007-02-28 16:35:06 -0500642 int parent_slot;
Chris Mason8d7be552007-05-10 11:24:42 -0400643 int slot;
644 struct btrfs_key cpukey;
Chris Mason5f39d392007-10-15 16:14:19 -0400645 u32 nritems = btrfs_header_nritems(node);
Chris Masonaa5d6be2007-02-28 16:35:06 -0500646
647 if (path->nodes[level + 1])
Chris Mason5f39d392007-10-15 16:14:19 -0400648 parent = path->nodes[level + 1];
Aneesha1f39632007-07-11 10:03:27 -0400649
Chris Mason8d7be552007-05-10 11:24:42 -0400650 slot = path->slots[level];
Chris Mason7518a232007-03-12 12:01:18 -0400651 BUG_ON(nritems == 0);
652 if (parent) {
Aneesha1f39632007-07-11 10:03:27 -0400653 parent_slot = path->slots[level + 1];
Chris Mason5f39d392007-10-15 16:14:19 -0400654 btrfs_node_key(parent, &parent_key, parent_slot);
655 btrfs_node_key(node, &node_key, 0);
656 BUG_ON(memcmp(&parent_key, &node_key,
Chris Masone2fa7222007-03-12 16:22:34 -0400657 sizeof(struct btrfs_disk_key)));
Chris Mason1d4f8a02007-03-13 09:28:32 -0400658 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
Chris Masondb945352007-10-15 16:15:53 -0400659 btrfs_header_bytenr(node));
Chris Masonaa5d6be2007-02-28 16:35:06 -0500660 }
Chris Mason123abc82007-03-14 14:14:43 -0400661 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
Chris Mason8d7be552007-05-10 11:24:42 -0400662 if (slot != 0) {
Chris Mason5f39d392007-10-15 16:14:19 -0400663 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
664 btrfs_node_key(node, &node_key, slot);
665 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
Chris Mason8d7be552007-05-10 11:24:42 -0400666 }
667 if (slot < nritems - 1) {
Chris Mason5f39d392007-10-15 16:14:19 -0400668 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
669 btrfs_node_key(node, &node_key, slot);
670 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
Chris Masonaa5d6be2007-02-28 16:35:06 -0500671 }
672 return 0;
673}
674
Chris Masond352ac62008-09-29 15:18:18 -0400675/*
676 * extra checking to make sure all the items in a leaf are
677 * well formed and in the proper order
678 */
Chris Mason123abc82007-03-14 14:14:43 -0400679static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
680 int level)
Chris Masonaa5d6be2007-02-28 16:35:06 -0500681{
Chris Mason5f39d392007-10-15 16:14:19 -0400682 struct extent_buffer *leaf = path->nodes[level];
683 struct extent_buffer *parent = NULL;
Chris Masonaa5d6be2007-02-28 16:35:06 -0500684 int parent_slot;
Chris Mason8d7be552007-05-10 11:24:42 -0400685 struct btrfs_key cpukey;
Chris Mason5f39d392007-10-15 16:14:19 -0400686 struct btrfs_disk_key parent_key;
687 struct btrfs_disk_key leaf_key;
688 int slot = path->slots[0];
Chris Mason8d7be552007-05-10 11:24:42 -0400689
Chris Mason5f39d392007-10-15 16:14:19 -0400690 u32 nritems = btrfs_header_nritems(leaf);
Chris Masonaa5d6be2007-02-28 16:35:06 -0500691
692 if (path->nodes[level + 1])
Chris Mason5f39d392007-10-15 16:14:19 -0400693 parent = path->nodes[level + 1];
Chris Mason7518a232007-03-12 12:01:18 -0400694
695 if (nritems == 0)
696 return 0;
697
698 if (parent) {
Aneesha1f39632007-07-11 10:03:27 -0400699 parent_slot = path->slots[level + 1];
Chris Mason5f39d392007-10-15 16:14:19 -0400700 btrfs_node_key(parent, &parent_key, parent_slot);
701 btrfs_item_key(leaf, &leaf_key, 0);
Chris Mason6702ed42007-08-07 16:15:09 -0400702
Chris Mason5f39d392007-10-15 16:14:19 -0400703 BUG_ON(memcmp(&parent_key, &leaf_key,
Chris Masone2fa7222007-03-12 16:22:34 -0400704 sizeof(struct btrfs_disk_key)));
Chris Mason1d4f8a02007-03-13 09:28:32 -0400705 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
Chris Masondb945352007-10-15 16:15:53 -0400706 btrfs_header_bytenr(leaf));
Chris Masonaa5d6be2007-02-28 16:35:06 -0500707 }
Chris Mason5f39d392007-10-15 16:14:19 -0400708 if (slot != 0 && slot < nritems - 1) {
709 btrfs_item_key(leaf, &leaf_key, slot);
710 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
711 if (comp_keys(&leaf_key, &cpukey) <= 0) {
712 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -0500713 printk(KERN_CRIT "slot %d offset bad key\n", slot);
Chris Mason5f39d392007-10-15 16:14:19 -0400714 BUG_ON(1);
715 }
716 if (btrfs_item_offset_nr(leaf, slot - 1) !=
717 btrfs_item_end_nr(leaf, slot)) {
718 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -0500719 printk(KERN_CRIT "slot %d offset bad\n", slot);
Chris Mason5f39d392007-10-15 16:14:19 -0400720 BUG_ON(1);
721 }
Chris Masonaa5d6be2007-02-28 16:35:06 -0500722 }
Chris Mason8d7be552007-05-10 11:24:42 -0400723 if (slot < nritems - 1) {
Chris Mason5f39d392007-10-15 16:14:19 -0400724 btrfs_item_key(leaf, &leaf_key, slot);
725 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
726 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
727 if (btrfs_item_offset_nr(leaf, slot) !=
728 btrfs_item_end_nr(leaf, slot + 1)) {
729 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -0500730 printk(KERN_CRIT "slot %d offset bad\n", slot);
Chris Mason5f39d392007-10-15 16:14:19 -0400731 BUG_ON(1);
732 }
Chris Mason8d7be552007-05-10 11:24:42 -0400733 }
Chris Mason5f39d392007-10-15 16:14:19 -0400734 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
735 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
Chris Masonaa5d6be2007-02-28 16:35:06 -0500736 return 0;
737}
738
Chris Masond3977122009-01-05 21:25:51 -0500739static noinline int check_block(struct btrfs_root *root,
Chris Mason98ed5172008-01-03 10:01:48 -0500740 struct btrfs_path *path, int level)
Chris Masonaa5d6be2007-02-28 16:35:06 -0500741{
Chris Mason85d824c2008-04-10 10:23:19 -0400742 return 0;
Chris Masonaa5d6be2007-02-28 16:35:06 -0500743 if (level == 0)
Chris Mason123abc82007-03-14 14:14:43 -0400744 return check_leaf(root, path, level);
745 return check_node(root, path, level);
Chris Masonaa5d6be2007-02-28 16:35:06 -0500746}
747
Chris Mason74123bd2007-02-02 11:05:29 -0500748/*
Chris Mason5f39d392007-10-15 16:14:19 -0400749 * search for key in the extent_buffer. The items start at offset p,
750 * and they are item_size apart. There are 'max' items in p.
751 *
Chris Mason74123bd2007-02-02 11:05:29 -0500752 * the slot in the array is returned via slot, and it points to
753 * the place where you would insert key if it is not found in
754 * the array.
755 *
756 * slot may point to max if the key is bigger than all of the keys
757 */
Chris Masone02119d2008-09-05 16:13:11 -0400758static noinline int generic_bin_search(struct extent_buffer *eb,
759 unsigned long p,
760 int item_size, struct btrfs_key *key,
761 int max, int *slot)
Chris Masonbe0e5c02007-01-26 15:51:26 -0500762{
763 int low = 0;
764 int high = max;
765 int mid;
766 int ret;
Chris Mason479965d2007-10-15 16:14:27 -0400767 struct btrfs_disk_key *tmp = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -0400768 struct btrfs_disk_key unaligned;
769 unsigned long offset;
770 char *map_token = NULL;
771 char *kaddr = NULL;
772 unsigned long map_start = 0;
773 unsigned long map_len = 0;
Chris Mason479965d2007-10-15 16:14:27 -0400774 int err;
Chris Masonbe0e5c02007-01-26 15:51:26 -0500775
Chris Masond3977122009-01-05 21:25:51 -0500776 while (low < high) {
Chris Masonbe0e5c02007-01-26 15:51:26 -0500777 mid = (low + high) / 2;
Chris Mason5f39d392007-10-15 16:14:19 -0400778 offset = p + mid * item_size;
779
780 if (!map_token || offset < map_start ||
781 (offset + sizeof(struct btrfs_disk_key)) >
782 map_start + map_len) {
Chris Mason479965d2007-10-15 16:14:27 -0400783 if (map_token) {
Chris Mason5f39d392007-10-15 16:14:19 -0400784 unmap_extent_buffer(eb, map_token, KM_USER0);
Chris Mason479965d2007-10-15 16:14:27 -0400785 map_token = NULL;
786 }
Chris Mason934d3752008-12-08 16:43:10 -0500787
788 err = map_private_extent_buffer(eb, offset,
Chris Mason479965d2007-10-15 16:14:27 -0400789 sizeof(struct btrfs_disk_key),
790 &map_token, &kaddr,
791 &map_start, &map_len, KM_USER0);
Chris Mason5f39d392007-10-15 16:14:19 -0400792
Chris Mason479965d2007-10-15 16:14:27 -0400793 if (!err) {
794 tmp = (struct btrfs_disk_key *)(kaddr + offset -
795 map_start);
796 } else {
797 read_extent_buffer(eb, &unaligned,
798 offset, sizeof(unaligned));
799 tmp = &unaligned;
800 }
801
Chris Mason5f39d392007-10-15 16:14:19 -0400802 } else {
803 tmp = (struct btrfs_disk_key *)(kaddr + offset -
804 map_start);
805 }
Chris Masonbe0e5c02007-01-26 15:51:26 -0500806 ret = comp_keys(tmp, key);
807
808 if (ret < 0)
809 low = mid + 1;
810 else if (ret > 0)
811 high = mid;
812 else {
813 *slot = mid;
Chris Mason479965d2007-10-15 16:14:27 -0400814 if (map_token)
815 unmap_extent_buffer(eb, map_token, KM_USER0);
Chris Masonbe0e5c02007-01-26 15:51:26 -0500816 return 0;
817 }
818 }
819 *slot = low;
Chris Mason5f39d392007-10-15 16:14:19 -0400820 if (map_token)
821 unmap_extent_buffer(eb, map_token, KM_USER0);
Chris Masonbe0e5c02007-01-26 15:51:26 -0500822 return 1;
823}
824
Chris Mason97571fd2007-02-24 13:39:08 -0500825/*
826 * simple bin_search frontend that does the right thing for
827 * leaves vs nodes
828 */
Chris Mason5f39d392007-10-15 16:14:19 -0400829static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
830 int level, int *slot)
Chris Masonbe0e5c02007-01-26 15:51:26 -0500831{
Chris Mason5f39d392007-10-15 16:14:19 -0400832 if (level == 0) {
833 return generic_bin_search(eb,
834 offsetof(struct btrfs_leaf, items),
Chris Mason0783fcf2007-03-12 20:12:07 -0400835 sizeof(struct btrfs_item),
Chris Mason5f39d392007-10-15 16:14:19 -0400836 key, btrfs_header_nritems(eb),
Chris Mason7518a232007-03-12 12:01:18 -0400837 slot);
Chris Masonbe0e5c02007-01-26 15:51:26 -0500838 } else {
Chris Mason5f39d392007-10-15 16:14:19 -0400839 return generic_bin_search(eb,
840 offsetof(struct btrfs_node, ptrs),
Chris Mason123abc82007-03-14 14:14:43 -0400841 sizeof(struct btrfs_key_ptr),
Chris Mason5f39d392007-10-15 16:14:19 -0400842 key, btrfs_header_nritems(eb),
Chris Mason7518a232007-03-12 12:01:18 -0400843 slot);
Chris Masonbe0e5c02007-01-26 15:51:26 -0500844 }
845 return -1;
846}
847
Chris Masond352ac62008-09-29 15:18:18 -0400848/* given a node and slot number, this reads the blocks it points to. The
849 * extent buffer is returned with a reference taken (but unlocked).
850 * NULL is returned on error.
851 */
Chris Masone02119d2008-09-05 16:13:11 -0400852static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
Chris Mason5f39d392007-10-15 16:14:19 -0400853 struct extent_buffer *parent, int slot)
Chris Masonbb803952007-03-01 12:04:21 -0500854{
Chris Masonca7a79a2008-05-12 12:59:19 -0400855 int level = btrfs_header_level(parent);
Chris Masonbb803952007-03-01 12:04:21 -0500856 if (slot < 0)
857 return NULL;
Chris Mason5f39d392007-10-15 16:14:19 -0400858 if (slot >= btrfs_header_nritems(parent))
Chris Masonbb803952007-03-01 12:04:21 -0500859 return NULL;
Chris Masonca7a79a2008-05-12 12:59:19 -0400860
861 BUG_ON(level == 0);
862
Chris Masondb945352007-10-15 16:15:53 -0400863 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
Chris Masonca7a79a2008-05-12 12:59:19 -0400864 btrfs_level_size(root, level - 1),
865 btrfs_node_ptr_generation(parent, slot));
Chris Masonbb803952007-03-01 12:04:21 -0500866}
867
Chris Masond352ac62008-09-29 15:18:18 -0400868/*
869 * node level balancing, used to make sure nodes are in proper order for
870 * item deletion. We balance from the top down, so we have to make sure
871 * that a deletion won't leave an node completely empty later on.
872 */
Chris Masone02119d2008-09-05 16:13:11 -0400873static noinline int balance_level(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -0500874 struct btrfs_root *root,
875 struct btrfs_path *path, int level)
Chris Masonbb803952007-03-01 12:04:21 -0500876{
Chris Mason5f39d392007-10-15 16:14:19 -0400877 struct extent_buffer *right = NULL;
878 struct extent_buffer *mid;
879 struct extent_buffer *left = NULL;
880 struct extent_buffer *parent = NULL;
Chris Masonbb803952007-03-01 12:04:21 -0500881 int ret = 0;
882 int wret;
883 int pslot;
Chris Masonbb803952007-03-01 12:04:21 -0500884 int orig_slot = path->slots[level];
Chris Mason54aa1f42007-06-22 14:16:25 -0400885 int err_on_enospc = 0;
Chris Mason79f95c82007-03-01 15:16:26 -0500886 u64 orig_ptr;
Chris Masonbb803952007-03-01 12:04:21 -0500887
888 if (level == 0)
889 return 0;
890
Chris Mason5f39d392007-10-15 16:14:19 -0400891 mid = path->nodes[level];
Chris Masonb4ce94d2009-02-04 09:25:08 -0500892
Chris Mason925baed2008-06-25 16:01:30 -0400893 WARN_ON(!path->locks[level]);
Chris Mason7bb86312007-12-11 09:25:06 -0500894 WARN_ON(btrfs_header_generation(mid) != trans->transid);
895
Chris Mason1d4f8a02007-03-13 09:28:32 -0400896 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
Chris Mason79f95c82007-03-01 15:16:26 -0500897
Chris Mason234b63a2007-03-13 10:46:10 -0400898 if (level < BTRFS_MAX_LEVEL - 1)
Chris Mason5f39d392007-10-15 16:14:19 -0400899 parent = path->nodes[level + 1];
Chris Masonbb803952007-03-01 12:04:21 -0500900 pslot = path->slots[level + 1];
901
Chris Mason40689472007-03-17 14:29:23 -0400902 /*
903 * deal with the case where there is only one pointer in the root
904 * by promoting the node below to a root
905 */
Chris Mason5f39d392007-10-15 16:14:19 -0400906 if (!parent) {
907 struct extent_buffer *child;
Chris Masonbb803952007-03-01 12:04:21 -0500908
Chris Mason5f39d392007-10-15 16:14:19 -0400909 if (btrfs_header_nritems(mid) != 1)
Chris Masonbb803952007-03-01 12:04:21 -0500910 return 0;
911
912 /* promote the child to a root */
Chris Mason5f39d392007-10-15 16:14:19 -0400913 child = read_node_slot(root, mid, 0);
Jeff Mahoney7951f3c2009-02-12 10:06:15 -0500914 BUG_ON(!child);
Chris Mason925baed2008-06-25 16:01:30 -0400915 btrfs_tree_lock(child);
Chris Masonb4ce94d2009-02-04 09:25:08 -0500916 btrfs_set_lock_blocking(child);
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400917 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
Yan2f375ab2008-02-01 14:58:07 -0500918 BUG_ON(ret);
919
Chris Mason925baed2008-06-25 16:01:30 -0400920 spin_lock(&root->node_lock);
Chris Masonbb803952007-03-01 12:04:21 -0500921 root->node = child;
Chris Mason925baed2008-06-25 16:01:30 -0400922 spin_unlock(&root->node_lock);
923
Zheng Yan31840ae2008-09-23 13:14:14 -0400924 ret = btrfs_update_extent_ref(trans, root, child->start,
Chris Mason56bec292009-03-13 10:10:06 -0400925 child->len,
Zheng Yan31840ae2008-09-23 13:14:14 -0400926 mid->start, child->start,
927 root->root_key.objectid,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -0400928 trans->transid, level - 1);
Zheng Yan31840ae2008-09-23 13:14:14 -0400929 BUG_ON(ret);
930
Chris Mason0b86a832008-03-24 15:01:56 -0400931 add_root_to_dirty_list(root);
Chris Mason925baed2008-06-25 16:01:30 -0400932 btrfs_tree_unlock(child);
Chris Masonb4ce94d2009-02-04 09:25:08 -0500933
Chris Mason925baed2008-06-25 16:01:30 -0400934 path->locks[level] = 0;
Chris Masonbb803952007-03-01 12:04:21 -0500935 path->nodes[level] = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -0400936 clean_tree_block(trans, root, mid);
Chris Mason925baed2008-06-25 16:01:30 -0400937 btrfs_tree_unlock(mid);
Chris Masonbb803952007-03-01 12:04:21 -0500938 /* once for the path */
Chris Mason5f39d392007-10-15 16:14:19 -0400939 free_extent_buffer(mid);
Chris Mason7bb86312007-12-11 09:25:06 -0500940 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
Zheng Yan31840ae2008-09-23 13:14:14 -0400941 mid->start, root->root_key.objectid,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -0400942 btrfs_header_generation(mid),
943 level, 1);
Chris Masonbb803952007-03-01 12:04:21 -0500944 /* once for the root ptr */
Chris Mason5f39d392007-10-15 16:14:19 -0400945 free_extent_buffer(mid);
Chris Masondb945352007-10-15 16:15:53 -0400946 return ret;
Chris Masonbb803952007-03-01 12:04:21 -0500947 }
Chris Mason5f39d392007-10-15 16:14:19 -0400948 if (btrfs_header_nritems(mid) >
Chris Mason123abc82007-03-14 14:14:43 -0400949 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
Chris Masonbb803952007-03-01 12:04:21 -0500950 return 0;
951
Chris Mason5f39d392007-10-15 16:14:19 -0400952 if (btrfs_header_nritems(mid) < 2)
Chris Mason54aa1f42007-06-22 14:16:25 -0400953 err_on_enospc = 1;
954
Chris Mason5f39d392007-10-15 16:14:19 -0400955 left = read_node_slot(root, parent, pslot - 1);
956 if (left) {
Chris Mason925baed2008-06-25 16:01:30 -0400957 btrfs_tree_lock(left);
Chris Masonb4ce94d2009-02-04 09:25:08 -0500958 btrfs_set_lock_blocking(left);
Chris Mason5f39d392007-10-15 16:14:19 -0400959 wret = btrfs_cow_block(trans, root, left,
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400960 parent, pslot - 1, &left);
Chris Mason54aa1f42007-06-22 14:16:25 -0400961 if (wret) {
962 ret = wret;
963 goto enospc;
964 }
Chris Mason2cc58cf2007-08-27 16:49:44 -0400965 }
Chris Mason5f39d392007-10-15 16:14:19 -0400966 right = read_node_slot(root, parent, pslot + 1);
967 if (right) {
Chris Mason925baed2008-06-25 16:01:30 -0400968 btrfs_tree_lock(right);
Chris Masonb4ce94d2009-02-04 09:25:08 -0500969 btrfs_set_lock_blocking(right);
Chris Mason5f39d392007-10-15 16:14:19 -0400970 wret = btrfs_cow_block(trans, root, right,
Chris Mason9fa8cfe2009-03-13 10:24:59 -0400971 parent, pslot + 1, &right);
Chris Mason2cc58cf2007-08-27 16:49:44 -0400972 if (wret) {
973 ret = wret;
974 goto enospc;
975 }
976 }
977
978 /* first, try to make some room in the middle buffer */
Chris Mason5f39d392007-10-15 16:14:19 -0400979 if (left) {
980 orig_slot += btrfs_header_nritems(left);
Chris Masonbce4eae2008-04-24 14:42:46 -0400981 wret = push_node_left(trans, root, left, mid, 1);
Chris Mason79f95c82007-03-01 15:16:26 -0500982 if (wret < 0)
983 ret = wret;
Chris Mason5f39d392007-10-15 16:14:19 -0400984 if (btrfs_header_nritems(mid) < 2)
Chris Mason54aa1f42007-06-22 14:16:25 -0400985 err_on_enospc = 1;
Chris Masonbb803952007-03-01 12:04:21 -0500986 }
Chris Mason79f95c82007-03-01 15:16:26 -0500987
988 /*
989 * then try to empty the right most buffer into the middle
990 */
Chris Mason5f39d392007-10-15 16:14:19 -0400991 if (right) {
Chris Mason971a1f62008-04-24 10:54:32 -0400992 wret = push_node_left(trans, root, mid, right, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -0400993 if (wret < 0 && wret != -ENOSPC)
Chris Mason79f95c82007-03-01 15:16:26 -0500994 ret = wret;
Chris Mason5f39d392007-10-15 16:14:19 -0400995 if (btrfs_header_nritems(right) == 0) {
Chris Masondb945352007-10-15 16:15:53 -0400996 u64 bytenr = right->start;
Chris Mason7bb86312007-12-11 09:25:06 -0500997 u64 generation = btrfs_header_generation(parent);
Chris Masondb945352007-10-15 16:15:53 -0400998 u32 blocksize = right->len;
999
Chris Mason5f39d392007-10-15 16:14:19 -04001000 clean_tree_block(trans, root, right);
Chris Mason925baed2008-06-25 16:01:30 -04001001 btrfs_tree_unlock(right);
Chris Mason5f39d392007-10-15 16:14:19 -04001002 free_extent_buffer(right);
Chris Masonbb803952007-03-01 12:04:21 -05001003 right = NULL;
Chris Masone089f052007-03-16 16:20:31 -04001004 wret = del_ptr(trans, root, path, level + 1, pslot +
1005 1);
Chris Masonbb803952007-03-01 12:04:21 -05001006 if (wret)
1007 ret = wret;
Chris Masondb945352007-10-15 16:15:53 -04001008 wret = btrfs_free_extent(trans, root, bytenr,
Zheng Yan31840ae2008-09-23 13:14:14 -04001009 blocksize, parent->start,
Chris Mason7bb86312007-12-11 09:25:06 -05001010 btrfs_header_owner(parent),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04001011 generation, level, 1);
Chris Masonbb803952007-03-01 12:04:21 -05001012 if (wret)
1013 ret = wret;
1014 } else {
Chris Mason5f39d392007-10-15 16:14:19 -04001015 struct btrfs_disk_key right_key;
1016 btrfs_node_key(right, &right_key, 0);
1017 btrfs_set_node_key(parent, &right_key, pslot + 1);
1018 btrfs_mark_buffer_dirty(parent);
Chris Masonbb803952007-03-01 12:04:21 -05001019 }
1020 }
Chris Mason5f39d392007-10-15 16:14:19 -04001021 if (btrfs_header_nritems(mid) == 1) {
Chris Mason79f95c82007-03-01 15:16:26 -05001022 /*
1023 * we're not allowed to leave a node with one item in the
1024 * tree during a delete. A deletion from lower in the tree
1025 * could try to delete the only pointer in this node.
1026 * So, pull some keys from the left.
1027 * There has to be a left pointer at this point because
1028 * otherwise we would have pulled some pointers from the
1029 * right
1030 */
Chris Mason5f39d392007-10-15 16:14:19 -04001031 BUG_ON(!left);
1032 wret = balance_node_right(trans, root, mid, left);
Chris Mason54aa1f42007-06-22 14:16:25 -04001033 if (wret < 0) {
Chris Mason79f95c82007-03-01 15:16:26 -05001034 ret = wret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001035 goto enospc;
1036 }
Chris Masonbce4eae2008-04-24 14:42:46 -04001037 if (wret == 1) {
1038 wret = push_node_left(trans, root, left, mid, 1);
1039 if (wret < 0)
1040 ret = wret;
1041 }
Chris Mason79f95c82007-03-01 15:16:26 -05001042 BUG_ON(wret == 1);
1043 }
Chris Mason5f39d392007-10-15 16:14:19 -04001044 if (btrfs_header_nritems(mid) == 0) {
Chris Mason79f95c82007-03-01 15:16:26 -05001045 /* we've managed to empty the middle node, drop it */
Chris Mason7bb86312007-12-11 09:25:06 -05001046 u64 root_gen = btrfs_header_generation(parent);
Chris Masondb945352007-10-15 16:15:53 -04001047 u64 bytenr = mid->start;
1048 u32 blocksize = mid->len;
Chris Mason925baed2008-06-25 16:01:30 -04001049
Chris Mason5f39d392007-10-15 16:14:19 -04001050 clean_tree_block(trans, root, mid);
Chris Mason925baed2008-06-25 16:01:30 -04001051 btrfs_tree_unlock(mid);
Chris Mason5f39d392007-10-15 16:14:19 -04001052 free_extent_buffer(mid);
Chris Masonbb803952007-03-01 12:04:21 -05001053 mid = NULL;
Chris Masone089f052007-03-16 16:20:31 -04001054 wret = del_ptr(trans, root, path, level + 1, pslot);
Chris Masonbb803952007-03-01 12:04:21 -05001055 if (wret)
1056 ret = wret;
Chris Mason7bb86312007-12-11 09:25:06 -05001057 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
Zheng Yan31840ae2008-09-23 13:14:14 -04001058 parent->start,
Chris Mason7bb86312007-12-11 09:25:06 -05001059 btrfs_header_owner(parent),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04001060 root_gen, level, 1);
Chris Masonbb803952007-03-01 12:04:21 -05001061 if (wret)
1062 ret = wret;
Chris Mason79f95c82007-03-01 15:16:26 -05001063 } else {
1064 /* update the parent key to reflect our changes */
Chris Mason5f39d392007-10-15 16:14:19 -04001065 struct btrfs_disk_key mid_key;
1066 btrfs_node_key(mid, &mid_key, 0);
1067 btrfs_set_node_key(parent, &mid_key, pslot);
1068 btrfs_mark_buffer_dirty(parent);
Chris Mason79f95c82007-03-01 15:16:26 -05001069 }
Chris Masonbb803952007-03-01 12:04:21 -05001070
Chris Mason79f95c82007-03-01 15:16:26 -05001071 /* update the path */
Chris Mason5f39d392007-10-15 16:14:19 -04001072 if (left) {
1073 if (btrfs_header_nritems(left) > orig_slot) {
1074 extent_buffer_get(left);
Chris Mason925baed2008-06-25 16:01:30 -04001075 /* left was locked after cow */
Chris Mason5f39d392007-10-15 16:14:19 -04001076 path->nodes[level] = left;
Chris Masonbb803952007-03-01 12:04:21 -05001077 path->slots[level + 1] -= 1;
1078 path->slots[level] = orig_slot;
Chris Mason925baed2008-06-25 16:01:30 -04001079 if (mid) {
1080 btrfs_tree_unlock(mid);
Chris Mason5f39d392007-10-15 16:14:19 -04001081 free_extent_buffer(mid);
Chris Mason925baed2008-06-25 16:01:30 -04001082 }
Chris Masonbb803952007-03-01 12:04:21 -05001083 } else {
Chris Mason5f39d392007-10-15 16:14:19 -04001084 orig_slot -= btrfs_header_nritems(left);
Chris Masonbb803952007-03-01 12:04:21 -05001085 path->slots[level] = orig_slot;
1086 }
1087 }
Chris Mason79f95c82007-03-01 15:16:26 -05001088 /* double check we haven't messed things up */
Chris Mason123abc82007-03-14 14:14:43 -04001089 check_block(root, path, level);
Chris Masone20d96d2007-03-22 12:13:20 -04001090 if (orig_ptr !=
Chris Mason5f39d392007-10-15 16:14:19 -04001091 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
Chris Mason79f95c82007-03-01 15:16:26 -05001092 BUG();
Chris Mason54aa1f42007-06-22 14:16:25 -04001093enospc:
Chris Mason925baed2008-06-25 16:01:30 -04001094 if (right) {
1095 btrfs_tree_unlock(right);
Chris Mason5f39d392007-10-15 16:14:19 -04001096 free_extent_buffer(right);
Chris Mason925baed2008-06-25 16:01:30 -04001097 }
1098 if (left) {
1099 if (path->nodes[level] != left)
1100 btrfs_tree_unlock(left);
Chris Mason5f39d392007-10-15 16:14:19 -04001101 free_extent_buffer(left);
Chris Mason925baed2008-06-25 16:01:30 -04001102 }
Chris Masonbb803952007-03-01 12:04:21 -05001103 return ret;
1104}
1105
Chris Masond352ac62008-09-29 15:18:18 -04001106/* Node balancing for insertion. Here we only split or push nodes around
1107 * when they are completely full. This is also done top down, so we
1108 * have to be pessimistic.
1109 */
Chris Masond3977122009-01-05 21:25:51 -05001110static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05001111 struct btrfs_root *root,
1112 struct btrfs_path *path, int level)
Chris Masone66f7092007-04-20 13:16:02 -04001113{
Chris Mason5f39d392007-10-15 16:14:19 -04001114 struct extent_buffer *right = NULL;
1115 struct extent_buffer *mid;
1116 struct extent_buffer *left = NULL;
1117 struct extent_buffer *parent = NULL;
Chris Masone66f7092007-04-20 13:16:02 -04001118 int ret = 0;
1119 int wret;
1120 int pslot;
1121 int orig_slot = path->slots[level];
1122 u64 orig_ptr;
1123
1124 if (level == 0)
1125 return 1;
1126
Chris Mason5f39d392007-10-15 16:14:19 -04001127 mid = path->nodes[level];
Chris Mason7bb86312007-12-11 09:25:06 -05001128 WARN_ON(btrfs_header_generation(mid) != trans->transid);
Chris Masone66f7092007-04-20 13:16:02 -04001129 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1130
1131 if (level < BTRFS_MAX_LEVEL - 1)
Chris Mason5f39d392007-10-15 16:14:19 -04001132 parent = path->nodes[level + 1];
Chris Masone66f7092007-04-20 13:16:02 -04001133 pslot = path->slots[level + 1];
1134
Chris Mason5f39d392007-10-15 16:14:19 -04001135 if (!parent)
Chris Masone66f7092007-04-20 13:16:02 -04001136 return 1;
Chris Masone66f7092007-04-20 13:16:02 -04001137
Chris Mason5f39d392007-10-15 16:14:19 -04001138 left = read_node_slot(root, parent, pslot - 1);
Chris Masone66f7092007-04-20 13:16:02 -04001139
1140 /* first, try to make some room in the middle buffer */
Chris Mason5f39d392007-10-15 16:14:19 -04001141 if (left) {
Chris Masone66f7092007-04-20 13:16:02 -04001142 u32 left_nr;
Chris Mason925baed2008-06-25 16:01:30 -04001143
1144 btrfs_tree_lock(left);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001145 btrfs_set_lock_blocking(left);
1146
Chris Mason5f39d392007-10-15 16:14:19 -04001147 left_nr = btrfs_header_nritems(left);
Chris Mason33ade1f2007-04-20 13:48:57 -04001148 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1149 wret = 1;
1150 } else {
Chris Mason5f39d392007-10-15 16:14:19 -04001151 ret = btrfs_cow_block(trans, root, left, parent,
Chris Mason9fa8cfe2009-03-13 10:24:59 -04001152 pslot - 1, &left);
Chris Mason54aa1f42007-06-22 14:16:25 -04001153 if (ret)
1154 wret = 1;
1155 else {
Chris Mason54aa1f42007-06-22 14:16:25 -04001156 wret = push_node_left(trans, root,
Chris Mason971a1f62008-04-24 10:54:32 -04001157 left, mid, 0);
Chris Mason54aa1f42007-06-22 14:16:25 -04001158 }
Chris Mason33ade1f2007-04-20 13:48:57 -04001159 }
Chris Masone66f7092007-04-20 13:16:02 -04001160 if (wret < 0)
1161 ret = wret;
1162 if (wret == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04001163 struct btrfs_disk_key disk_key;
Chris Masone66f7092007-04-20 13:16:02 -04001164 orig_slot += left_nr;
Chris Mason5f39d392007-10-15 16:14:19 -04001165 btrfs_node_key(mid, &disk_key, 0);
1166 btrfs_set_node_key(parent, &disk_key, pslot);
1167 btrfs_mark_buffer_dirty(parent);
1168 if (btrfs_header_nritems(left) > orig_slot) {
1169 path->nodes[level] = left;
Chris Masone66f7092007-04-20 13:16:02 -04001170 path->slots[level + 1] -= 1;
1171 path->slots[level] = orig_slot;
Chris Mason925baed2008-06-25 16:01:30 -04001172 btrfs_tree_unlock(mid);
Chris Mason5f39d392007-10-15 16:14:19 -04001173 free_extent_buffer(mid);
Chris Masone66f7092007-04-20 13:16:02 -04001174 } else {
1175 orig_slot -=
Chris Mason5f39d392007-10-15 16:14:19 -04001176 btrfs_header_nritems(left);
Chris Masone66f7092007-04-20 13:16:02 -04001177 path->slots[level] = orig_slot;
Chris Mason925baed2008-06-25 16:01:30 -04001178 btrfs_tree_unlock(left);
Chris Mason5f39d392007-10-15 16:14:19 -04001179 free_extent_buffer(left);
Chris Masone66f7092007-04-20 13:16:02 -04001180 }
Chris Masone66f7092007-04-20 13:16:02 -04001181 return 0;
1182 }
Chris Mason925baed2008-06-25 16:01:30 -04001183 btrfs_tree_unlock(left);
Chris Mason5f39d392007-10-15 16:14:19 -04001184 free_extent_buffer(left);
Chris Masone66f7092007-04-20 13:16:02 -04001185 }
Chris Mason925baed2008-06-25 16:01:30 -04001186 right = read_node_slot(root, parent, pslot + 1);
Chris Masone66f7092007-04-20 13:16:02 -04001187
1188 /*
1189 * then try to empty the right most buffer into the middle
1190 */
Chris Mason5f39d392007-10-15 16:14:19 -04001191 if (right) {
Chris Mason33ade1f2007-04-20 13:48:57 -04001192 u32 right_nr;
Chris Masonb4ce94d2009-02-04 09:25:08 -05001193
Chris Mason925baed2008-06-25 16:01:30 -04001194 btrfs_tree_lock(right);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001195 btrfs_set_lock_blocking(right);
1196
Chris Mason5f39d392007-10-15 16:14:19 -04001197 right_nr = btrfs_header_nritems(right);
Chris Mason33ade1f2007-04-20 13:48:57 -04001198 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1199 wret = 1;
1200 } else {
Chris Mason5f39d392007-10-15 16:14:19 -04001201 ret = btrfs_cow_block(trans, root, right,
1202 parent, pslot + 1,
Chris Mason9fa8cfe2009-03-13 10:24:59 -04001203 &right);
Chris Mason54aa1f42007-06-22 14:16:25 -04001204 if (ret)
1205 wret = 1;
1206 else {
Chris Mason54aa1f42007-06-22 14:16:25 -04001207 wret = balance_node_right(trans, root,
Chris Mason5f39d392007-10-15 16:14:19 -04001208 right, mid);
Chris Mason54aa1f42007-06-22 14:16:25 -04001209 }
Chris Mason33ade1f2007-04-20 13:48:57 -04001210 }
Chris Masone66f7092007-04-20 13:16:02 -04001211 if (wret < 0)
1212 ret = wret;
1213 if (wret == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04001214 struct btrfs_disk_key disk_key;
1215
1216 btrfs_node_key(right, &disk_key, 0);
1217 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1218 btrfs_mark_buffer_dirty(parent);
1219
1220 if (btrfs_header_nritems(mid) <= orig_slot) {
1221 path->nodes[level] = right;
Chris Masone66f7092007-04-20 13:16:02 -04001222 path->slots[level + 1] += 1;
1223 path->slots[level] = orig_slot -
Chris Mason5f39d392007-10-15 16:14:19 -04001224 btrfs_header_nritems(mid);
Chris Mason925baed2008-06-25 16:01:30 -04001225 btrfs_tree_unlock(mid);
Chris Mason5f39d392007-10-15 16:14:19 -04001226 free_extent_buffer(mid);
Chris Masone66f7092007-04-20 13:16:02 -04001227 } else {
Chris Mason925baed2008-06-25 16:01:30 -04001228 btrfs_tree_unlock(right);
Chris Mason5f39d392007-10-15 16:14:19 -04001229 free_extent_buffer(right);
Chris Masone66f7092007-04-20 13:16:02 -04001230 }
Chris Masone66f7092007-04-20 13:16:02 -04001231 return 0;
1232 }
Chris Mason925baed2008-06-25 16:01:30 -04001233 btrfs_tree_unlock(right);
Chris Mason5f39d392007-10-15 16:14:19 -04001234 free_extent_buffer(right);
Chris Masone66f7092007-04-20 13:16:02 -04001235 }
Chris Masone66f7092007-04-20 13:16:02 -04001236 return 1;
1237}
1238
Chris Mason74123bd2007-02-02 11:05:29 -05001239/*
Chris Masond352ac62008-09-29 15:18:18 -04001240 * readahead one full node of leaves, finding things that are close
1241 * to the block in 'slot', and triggering ra on them.
Chris Mason3c69fae2007-08-07 15:52:22 -04001242 */
Chris Masone02119d2008-09-05 16:13:11 -04001243static noinline void reada_for_search(struct btrfs_root *root,
1244 struct btrfs_path *path,
1245 int level, int slot, u64 objectid)
Chris Mason3c69fae2007-08-07 15:52:22 -04001246{
Chris Mason5f39d392007-10-15 16:14:19 -04001247 struct extent_buffer *node;
Chris Mason01f46652007-12-21 16:24:26 -05001248 struct btrfs_disk_key disk_key;
Chris Mason3c69fae2007-08-07 15:52:22 -04001249 u32 nritems;
Chris Mason3c69fae2007-08-07 15:52:22 -04001250 u64 search;
Chris Masona7175312009-01-22 09:23:10 -05001251 u64 target;
Chris Mason6b800532007-10-15 16:17:34 -04001252 u64 nread = 0;
Chris Mason3c69fae2007-08-07 15:52:22 -04001253 int direction = path->reada;
Chris Mason5f39d392007-10-15 16:14:19 -04001254 struct extent_buffer *eb;
Chris Mason6b800532007-10-15 16:17:34 -04001255 u32 nr;
1256 u32 blocksize;
1257 u32 nscan = 0;
Chris Masondb945352007-10-15 16:15:53 -04001258
Chris Masona6b6e752007-10-15 16:22:39 -04001259 if (level != 1)
Chris Mason3c69fae2007-08-07 15:52:22 -04001260 return;
1261
Chris Mason6702ed42007-08-07 16:15:09 -04001262 if (!path->nodes[level])
1263 return;
1264
Chris Mason5f39d392007-10-15 16:14:19 -04001265 node = path->nodes[level];
Chris Mason925baed2008-06-25 16:01:30 -04001266
Chris Mason3c69fae2007-08-07 15:52:22 -04001267 search = btrfs_node_blockptr(node, slot);
Chris Mason6b800532007-10-15 16:17:34 -04001268 blocksize = btrfs_level_size(root, level - 1);
1269 eb = btrfs_find_tree_block(root, search, blocksize);
Chris Mason5f39d392007-10-15 16:14:19 -04001270 if (eb) {
1271 free_extent_buffer(eb);
Chris Mason3c69fae2007-08-07 15:52:22 -04001272 return;
1273 }
1274
Chris Masona7175312009-01-22 09:23:10 -05001275 target = search;
Chris Mason6b800532007-10-15 16:17:34 -04001276
Chris Mason5f39d392007-10-15 16:14:19 -04001277 nritems = btrfs_header_nritems(node);
Chris Mason6b800532007-10-15 16:17:34 -04001278 nr = slot;
Chris Masond3977122009-01-05 21:25:51 -05001279 while (1) {
Chris Mason6b800532007-10-15 16:17:34 -04001280 if (direction < 0) {
1281 if (nr == 0)
1282 break;
1283 nr--;
1284 } else if (direction > 0) {
1285 nr++;
1286 if (nr >= nritems)
1287 break;
Chris Mason3c69fae2007-08-07 15:52:22 -04001288 }
Chris Mason01f46652007-12-21 16:24:26 -05001289 if (path->reada < 0 && objectid) {
1290 btrfs_node_key(node, &disk_key, nr);
1291 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1292 break;
1293 }
Chris Mason6b800532007-10-15 16:17:34 -04001294 search = btrfs_node_blockptr(node, nr);
Chris Masona7175312009-01-22 09:23:10 -05001295 if ((search <= target && target - search <= 65536) ||
1296 (search > target && search - target <= 65536)) {
Chris Masonca7a79a2008-05-12 12:59:19 -04001297 readahead_tree_block(root, search, blocksize,
1298 btrfs_node_ptr_generation(node, nr));
Chris Mason6b800532007-10-15 16:17:34 -04001299 nread += blocksize;
1300 }
1301 nscan++;
Chris Masona7175312009-01-22 09:23:10 -05001302 if ((nread > 65536 || nscan > 32))
Chris Mason6b800532007-10-15 16:17:34 -04001303 break;
Chris Mason3c69fae2007-08-07 15:52:22 -04001304 }
1305}
Chris Mason925baed2008-06-25 16:01:30 -04001306
Chris Masond352ac62008-09-29 15:18:18 -04001307/*
Chris Masonb4ce94d2009-02-04 09:25:08 -05001308 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1309 * cache
1310 */
1311static noinline int reada_for_balance(struct btrfs_root *root,
1312 struct btrfs_path *path, int level)
1313{
1314 int slot;
1315 int nritems;
1316 struct extent_buffer *parent;
1317 struct extent_buffer *eb;
1318 u64 gen;
1319 u64 block1 = 0;
1320 u64 block2 = 0;
1321 int ret = 0;
1322 int blocksize;
1323
1324 parent = path->nodes[level - 1];
1325 if (!parent)
1326 return 0;
1327
1328 nritems = btrfs_header_nritems(parent);
1329 slot = path->slots[level];
1330 blocksize = btrfs_level_size(root, level);
1331
1332 if (slot > 0) {
1333 block1 = btrfs_node_blockptr(parent, slot - 1);
1334 gen = btrfs_node_ptr_generation(parent, slot - 1);
1335 eb = btrfs_find_tree_block(root, block1, blocksize);
1336 if (eb && btrfs_buffer_uptodate(eb, gen))
1337 block1 = 0;
1338 free_extent_buffer(eb);
1339 }
1340 if (slot < nritems) {
1341 block2 = btrfs_node_blockptr(parent, slot + 1);
1342 gen = btrfs_node_ptr_generation(parent, slot + 1);
1343 eb = btrfs_find_tree_block(root, block2, blocksize);
1344 if (eb && btrfs_buffer_uptodate(eb, gen))
1345 block2 = 0;
1346 free_extent_buffer(eb);
1347 }
1348 if (block1 || block2) {
1349 ret = -EAGAIN;
1350 btrfs_release_path(root, path);
1351 if (block1)
1352 readahead_tree_block(root, block1, blocksize, 0);
1353 if (block2)
1354 readahead_tree_block(root, block2, blocksize, 0);
1355
1356 if (block1) {
1357 eb = read_tree_block(root, block1, blocksize, 0);
1358 free_extent_buffer(eb);
1359 }
1360 if (block1) {
1361 eb = read_tree_block(root, block2, blocksize, 0);
1362 free_extent_buffer(eb);
1363 }
1364 }
1365 return ret;
1366}
1367
1368
1369/*
Chris Masond3977122009-01-05 21:25:51 -05001370 * when we walk down the tree, it is usually safe to unlock the higher layers
1371 * in the tree. The exceptions are when our path goes through slot 0, because
1372 * operations on the tree might require changing key pointers higher up in the
1373 * tree.
Chris Masond352ac62008-09-29 15:18:18 -04001374 *
Chris Masond3977122009-01-05 21:25:51 -05001375 * callers might also have set path->keep_locks, which tells this code to keep
1376 * the lock if the path points to the last slot in the block. This is part of
1377 * walking through the tree, and selecting the next slot in the higher block.
Chris Masond352ac62008-09-29 15:18:18 -04001378 *
Chris Masond3977122009-01-05 21:25:51 -05001379 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1380 * if lowest_unlock is 1, level 0 won't be unlocked
Chris Masond352ac62008-09-29 15:18:18 -04001381 */
Chris Masone02119d2008-09-05 16:13:11 -04001382static noinline void unlock_up(struct btrfs_path *path, int level,
1383 int lowest_unlock)
Chris Mason925baed2008-06-25 16:01:30 -04001384{
1385 int i;
1386 int skip_level = level;
Chris Mason051e1b92008-06-25 16:01:30 -04001387 int no_skips = 0;
Chris Mason925baed2008-06-25 16:01:30 -04001388 struct extent_buffer *t;
1389
1390 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1391 if (!path->nodes[i])
1392 break;
1393 if (!path->locks[i])
1394 break;
Chris Mason051e1b92008-06-25 16:01:30 -04001395 if (!no_skips && path->slots[i] == 0) {
Chris Mason925baed2008-06-25 16:01:30 -04001396 skip_level = i + 1;
1397 continue;
1398 }
Chris Mason051e1b92008-06-25 16:01:30 -04001399 if (!no_skips && path->keep_locks) {
Chris Mason925baed2008-06-25 16:01:30 -04001400 u32 nritems;
1401 t = path->nodes[i];
1402 nritems = btrfs_header_nritems(t);
Chris Mason051e1b92008-06-25 16:01:30 -04001403 if (nritems < 1 || path->slots[i] >= nritems - 1) {
Chris Mason925baed2008-06-25 16:01:30 -04001404 skip_level = i + 1;
1405 continue;
1406 }
1407 }
Chris Mason051e1b92008-06-25 16:01:30 -04001408 if (skip_level < i && i >= lowest_unlock)
1409 no_skips = 1;
1410
Chris Mason925baed2008-06-25 16:01:30 -04001411 t = path->nodes[i];
1412 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1413 btrfs_tree_unlock(t);
1414 path->locks[i] = 0;
1415 }
1416 }
1417}
1418
Chris Mason3c69fae2007-08-07 15:52:22 -04001419/*
Chris Masonb4ce94d2009-02-04 09:25:08 -05001420 * This releases any locks held in the path starting at level and
1421 * going all the way up to the root.
1422 *
1423 * btrfs_search_slot will keep the lock held on higher nodes in a few
1424 * corner cases, such as COW of the block at slot zero in the node. This
1425 * ignores those rules, and it should only be called when there are no
1426 * more updates to be done higher up in the tree.
1427 */
1428noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1429{
1430 int i;
1431
1432 if (path->keep_locks || path->lowest_level)
1433 return;
1434
1435 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1436 if (!path->nodes[i])
Chris Mason12f4dac2009-02-04 09:31:42 -05001437 continue;
Chris Masonb4ce94d2009-02-04 09:25:08 -05001438 if (!path->locks[i])
Chris Mason12f4dac2009-02-04 09:31:42 -05001439 continue;
Chris Masonb4ce94d2009-02-04 09:25:08 -05001440 btrfs_tree_unlock(path->nodes[i]);
1441 path->locks[i] = 0;
1442 }
1443}
1444
1445/*
Chris Mason74123bd2007-02-02 11:05:29 -05001446 * look for key in the tree. path is filled in with nodes along the way
1447 * if key is found, we return zero and you can find the item in the leaf
1448 * level of the path (level 0)
1449 *
1450 * If the key isn't found, the path points to the slot where it should
Chris Masonaa5d6be2007-02-28 16:35:06 -05001451 * be inserted, and 1 is returned. If there are other errors during the
1452 * search a negative error number is returned.
Chris Mason97571fd2007-02-24 13:39:08 -05001453 *
1454 * if ins_len > 0, nodes and leaves will be split as we walk down the
1455 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1456 * possible)
Chris Mason74123bd2007-02-02 11:05:29 -05001457 */
Chris Masone089f052007-03-16 16:20:31 -04001458int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1459 *root, struct btrfs_key *key, struct btrfs_path *p, int
1460 ins_len, int cow)
Chris Masonbe0e5c02007-01-26 15:51:26 -05001461{
Chris Mason5f39d392007-10-15 16:14:19 -04001462 struct extent_buffer *b;
Chris Mason051e1b92008-06-25 16:01:30 -04001463 struct extent_buffer *tmp;
Chris Masonbe0e5c02007-01-26 15:51:26 -05001464 int slot;
1465 int ret;
1466 int level;
Chris Mason3c69fae2007-08-07 15:52:22 -04001467 int should_reada = p->reada;
Chris Mason925baed2008-06-25 16:01:30 -04001468 int lowest_unlock = 1;
Chris Mason594a24e2008-06-25 16:01:30 -04001469 int blocksize;
Chris Mason9f3a7422007-08-07 15:52:19 -04001470 u8 lowest_level = 0;
Chris Mason594a24e2008-06-25 16:01:30 -04001471 u64 blocknr;
1472 u64 gen;
Chris Mason9f3a7422007-08-07 15:52:19 -04001473
Chris Mason6702ed42007-08-07 16:15:09 -04001474 lowest_level = p->lowest_level;
Chris Mason323ac952008-10-01 19:05:46 -04001475 WARN_ON(lowest_level && ins_len > 0);
Chris Mason22b0ebd2007-03-30 08:47:31 -04001476 WARN_ON(p->nodes[0] != NULL);
Josef Bacik25179202008-10-29 14:49:05 -04001477
Chris Mason925baed2008-06-25 16:01:30 -04001478 if (ins_len < 0)
1479 lowest_unlock = 2;
Chris Mason65b51a02008-08-01 15:11:20 -04001480
Chris Masonbb803952007-03-01 12:04:21 -05001481again:
Chris Mason5cd57b22008-06-25 16:01:30 -04001482 if (p->skip_locking)
1483 b = btrfs_root_node(root);
1484 else
1485 b = btrfs_lock_root_node(root);
Chris Mason925baed2008-06-25 16:01:30 -04001486
Chris Masoneb60cea2007-02-02 09:18:22 -05001487 while (b) {
Chris Mason5f39d392007-10-15 16:14:19 -04001488 level = btrfs_header_level(b);
Chris Mason65b51a02008-08-01 15:11:20 -04001489
1490 /*
1491 * setup the path here so we can release it under lock
1492 * contention with the cow code
1493 */
1494 p->nodes[level] = b;
1495 if (!p->skip_locking)
1496 p->locks[level] = 1;
1497
Chris Mason02217ed2007-03-02 16:08:05 -05001498 if (cow) {
1499 int wret;
Chris Mason65b51a02008-08-01 15:11:20 -04001500
1501 /* is a cow on this block not required */
Chris Mason65b51a02008-08-01 15:11:20 -04001502 if (btrfs_header_generation(b) == trans->transid &&
Zheng Yan5b21f2e2008-09-26 10:05:38 -04001503 btrfs_header_owner(b) == root->root_key.objectid &&
Chris Mason65b51a02008-08-01 15:11:20 -04001504 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
Chris Mason65b51a02008-08-01 15:11:20 -04001505 goto cow_done;
1506 }
Chris Masonb4ce94d2009-02-04 09:25:08 -05001507 btrfs_set_path_blocking(p);
1508
Chris Masone20d96d2007-03-22 12:13:20 -04001509 wret = btrfs_cow_block(trans, root, b,
1510 p->nodes[level + 1],
Chris Mason9fa8cfe2009-03-13 10:24:59 -04001511 p->slots[level + 1], &b);
Chris Mason54aa1f42007-06-22 14:16:25 -04001512 if (wret) {
Chris Mason5f39d392007-10-15 16:14:19 -04001513 free_extent_buffer(b);
Chris Mason65b51a02008-08-01 15:11:20 -04001514 ret = wret;
1515 goto done;
Chris Mason54aa1f42007-06-22 14:16:25 -04001516 }
Chris Mason02217ed2007-03-02 16:08:05 -05001517 }
Chris Mason65b51a02008-08-01 15:11:20 -04001518cow_done:
Chris Mason02217ed2007-03-02 16:08:05 -05001519 BUG_ON(!cow && ins_len);
Chris Mason5f39d392007-10-15 16:14:19 -04001520 if (level != btrfs_header_level(b))
Chris Mason2c90e5d2007-04-02 10:50:19 -04001521 WARN_ON(1);
Chris Mason5f39d392007-10-15 16:14:19 -04001522 level = btrfs_header_level(b);
Chris Mason65b51a02008-08-01 15:11:20 -04001523
Chris Masoneb60cea2007-02-02 09:18:22 -05001524 p->nodes[level] = b;
Chris Mason5cd57b22008-06-25 16:01:30 -04001525 if (!p->skip_locking)
1526 p->locks[level] = 1;
Chris Mason65b51a02008-08-01 15:11:20 -04001527
Chris Mason4008c042009-02-12 14:09:45 -05001528 btrfs_clear_path_blocking(p, NULL);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001529
1530 /*
1531 * we have a lock on b and as long as we aren't changing
1532 * the tree, there is no way to for the items in b to change.
1533 * It is safe to drop the lock on our parent before we
1534 * go through the expensive btree search on b.
1535 *
1536 * If cow is true, then we might be changing slot zero,
1537 * which may require changing the parent. So, we can't
1538 * drop the lock until after we know which slot we're
1539 * operating on.
1540 */
1541 if (!cow)
1542 btrfs_unlock_up_safe(p, level + 1);
1543
Chris Mason123abc82007-03-14 14:14:43 -04001544 ret = check_block(root, p, level);
Chris Mason65b51a02008-08-01 15:11:20 -04001545 if (ret) {
1546 ret = -1;
1547 goto done;
1548 }
Chris Mason925baed2008-06-25 16:01:30 -04001549
Chris Mason5f39d392007-10-15 16:14:19 -04001550 ret = bin_search(b, key, level, &slot);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001551
Chris Mason5f39d392007-10-15 16:14:19 -04001552 if (level != 0) {
Chris Masonbe0e5c02007-01-26 15:51:26 -05001553 if (ret && slot > 0)
1554 slot -= 1;
1555 p->slots[level] = slot;
Chris Mason459931e2008-12-10 09:10:46 -05001556 if ((p->search_for_split || ins_len > 0) &&
1557 btrfs_header_nritems(b) >=
Chris Mason15147942008-04-24 09:22:51 -04001558 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
Chris Masonb4ce94d2009-02-04 09:25:08 -05001559 int sret;
1560
1561 sret = reada_for_balance(root, p, level);
1562 if (sret)
1563 goto again;
1564
1565 btrfs_set_path_blocking(p);
1566 sret = split_node(trans, root, p, level);
Chris Mason4008c042009-02-12 14:09:45 -05001567 btrfs_clear_path_blocking(p, NULL);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001568
Chris Mason5c680ed2007-02-22 11:39:13 -05001569 BUG_ON(sret > 0);
Chris Mason65b51a02008-08-01 15:11:20 -04001570 if (sret) {
1571 ret = sret;
1572 goto done;
1573 }
Chris Mason5c680ed2007-02-22 11:39:13 -05001574 b = p->nodes[level];
Chris Mason5c680ed2007-02-22 11:39:13 -05001575 slot = p->slots[level];
Chris Mason7b78c172009-02-04 09:12:46 -05001576 } else if (ins_len < 0 &&
1577 btrfs_header_nritems(b) <
1578 BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
Chris Masonb4ce94d2009-02-04 09:25:08 -05001579 int sret;
1580
1581 sret = reada_for_balance(root, p, level);
1582 if (sret)
1583 goto again;
1584
1585 btrfs_set_path_blocking(p);
1586 sret = balance_level(trans, root, p, level);
Chris Mason4008c042009-02-12 14:09:45 -05001587 btrfs_clear_path_blocking(p, NULL);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001588
Chris Mason65b51a02008-08-01 15:11:20 -04001589 if (sret) {
1590 ret = sret;
1591 goto done;
1592 }
Chris Masonbb803952007-03-01 12:04:21 -05001593 b = p->nodes[level];
Chris Masonf510cfe2007-10-15 16:14:48 -04001594 if (!b) {
1595 btrfs_release_path(NULL, p);
Chris Masonbb803952007-03-01 12:04:21 -05001596 goto again;
Chris Masonf510cfe2007-10-15 16:14:48 -04001597 }
Chris Masonbb803952007-03-01 12:04:21 -05001598 slot = p->slots[level];
Chris Mason5f39d392007-10-15 16:14:19 -04001599 BUG_ON(btrfs_header_nritems(b) == 1);
Chris Mason5c680ed2007-02-22 11:39:13 -05001600 }
Chris Masonf9efa9c2008-06-25 16:14:04 -04001601 unlock_up(p, level, lowest_unlock);
1602
Chris Mason9f3a7422007-08-07 15:52:19 -04001603 /* this is only true while dropping a snapshot */
Chris Mason925baed2008-06-25 16:01:30 -04001604 if (level == lowest_level) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -04001605 ret = 0;
1606 goto done;
Chris Mason925baed2008-06-25 16:01:30 -04001607 }
Chris Masonca7a79a2008-05-12 12:59:19 -04001608
Chris Mason594a24e2008-06-25 16:01:30 -04001609 blocknr = btrfs_node_blockptr(b, slot);
1610 gen = btrfs_node_ptr_generation(b, slot);
1611 blocksize = btrfs_level_size(root, level - 1);
1612
1613 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1614 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
Chris Mason051e1b92008-06-25 16:01:30 -04001615 b = tmp;
1616 } else {
1617 /*
1618 * reduce lock contention at high levels
1619 * of the btree by dropping locks before
1620 * we read.
1621 */
Chris Masonb4ce94d2009-02-04 09:25:08 -05001622 if (level > 0) {
Chris Mason051e1b92008-06-25 16:01:30 -04001623 btrfs_release_path(NULL, p);
1624 if (tmp)
1625 free_extent_buffer(tmp);
Chris Masonf9efa9c2008-06-25 16:14:04 -04001626 if (should_reada)
1627 reada_for_search(root, p,
1628 level, slot,
1629 key->objectid);
1630
Chris Mason594a24e2008-06-25 16:01:30 -04001631 tmp = read_tree_block(root, blocknr,
1632 blocksize, gen);
1633 if (tmp)
1634 free_extent_buffer(tmp);
Chris Mason051e1b92008-06-25 16:01:30 -04001635 goto again;
1636 } else {
Chris Masonb4ce94d2009-02-04 09:25:08 -05001637 btrfs_set_path_blocking(p);
Chris Masona74a4b92008-06-25 16:01:31 -04001638 if (tmp)
1639 free_extent_buffer(tmp);
Chris Masonf9efa9c2008-06-25 16:14:04 -04001640 if (should_reada)
1641 reada_for_search(root, p,
1642 level, slot,
1643 key->objectid);
Chris Mason051e1b92008-06-25 16:01:30 -04001644 b = read_node_slot(root, b, slot);
1645 }
1646 }
Chris Masonb4ce94d2009-02-04 09:25:08 -05001647 if (!p->skip_locking) {
1648 int lret;
1649
Chris Mason4008c042009-02-12 14:09:45 -05001650 btrfs_clear_path_blocking(p, NULL);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001651 lret = btrfs_try_spin_lock(b);
1652
1653 if (!lret) {
1654 btrfs_set_path_blocking(p);
1655 btrfs_tree_lock(b);
Chris Mason4008c042009-02-12 14:09:45 -05001656 btrfs_clear_path_blocking(p, b);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001657 }
1658 }
Chris Masonbe0e5c02007-01-26 15:51:26 -05001659 } else {
1660 p->slots[level] = slot;
Yan Zheng87b29b22008-12-17 10:21:48 -05001661 if (ins_len > 0 &&
1662 btrfs_leaf_free_space(root, b) < ins_len) {
Chris Masonb4ce94d2009-02-04 09:25:08 -05001663 int sret;
1664
1665 btrfs_set_path_blocking(p);
1666 sret = split_leaf(trans, root, key,
Chris Masoncc0c5532007-10-25 15:42:57 -04001667 p, ins_len, ret == 0);
Chris Mason4008c042009-02-12 14:09:45 -05001668 btrfs_clear_path_blocking(p, NULL);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001669
Chris Mason5c680ed2007-02-22 11:39:13 -05001670 BUG_ON(sret > 0);
Chris Mason65b51a02008-08-01 15:11:20 -04001671 if (sret) {
1672 ret = sret;
1673 goto done;
1674 }
Chris Mason5c680ed2007-02-22 11:39:13 -05001675 }
Chris Mason459931e2008-12-10 09:10:46 -05001676 if (!p->search_for_split)
1677 unlock_up(p, level, lowest_unlock);
Chris Mason65b51a02008-08-01 15:11:20 -04001678 goto done;
Chris Masonbe0e5c02007-01-26 15:51:26 -05001679 }
1680 }
Chris Mason65b51a02008-08-01 15:11:20 -04001681 ret = 1;
1682done:
Chris Masonb4ce94d2009-02-04 09:25:08 -05001683 /*
1684 * we don't really know what they plan on doing with the path
1685 * from here on, so for now just mark it as blocking
1686 */
Chris Masonb9473432009-03-13 11:00:37 -04001687 if (!p->leave_spinning)
1688 btrfs_set_path_blocking(p);
Chris Mason65b51a02008-08-01 15:11:20 -04001689 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05001690}
1691
Zheng Yan1a40e232008-09-26 10:09:34 -04001692int btrfs_merge_path(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_key *node_keys,
1695 u64 *nodes, int lowest_level)
1696{
1697 struct extent_buffer *eb;
1698 struct extent_buffer *parent;
1699 struct btrfs_key key;
1700 u64 bytenr;
1701 u64 generation;
1702 u32 blocksize;
1703 int level;
1704 int slot;
1705 int key_match;
1706 int ret;
1707
1708 eb = btrfs_lock_root_node(root);
Chris Mason9fa8cfe2009-03-13 10:24:59 -04001709 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
Zheng Yan1a40e232008-09-26 10:09:34 -04001710 BUG_ON(ret);
1711
Chris Masonb4ce94d2009-02-04 09:25:08 -05001712 btrfs_set_lock_blocking(eb);
1713
Zheng Yan1a40e232008-09-26 10:09:34 -04001714 parent = eb;
1715 while (1) {
1716 level = btrfs_header_level(parent);
1717 if (level == 0 || level <= lowest_level)
1718 break;
1719
1720 ret = bin_search(parent, &node_keys[lowest_level], level,
1721 &slot);
1722 if (ret && slot > 0)
1723 slot--;
1724
1725 bytenr = btrfs_node_blockptr(parent, slot);
1726 if (nodes[level - 1] == bytenr)
1727 break;
1728
1729 blocksize = btrfs_level_size(root, level - 1);
1730 generation = btrfs_node_ptr_generation(parent, slot);
1731 btrfs_node_key_to_cpu(eb, &key, slot);
1732 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1733
Yan Zhengf82d02d2008-10-29 14:49:05 -04001734 if (generation == trans->transid) {
Zheng Yan1a40e232008-09-26 10:09:34 -04001735 eb = read_tree_block(root, bytenr, blocksize,
1736 generation);
1737 btrfs_tree_lock(eb);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001738 btrfs_set_lock_blocking(eb);
Yan Zhengf82d02d2008-10-29 14:49:05 -04001739 }
1740
1741 /*
1742 * if node keys match and node pointer hasn't been modified
1743 * in the running transaction, we can merge the path. for
1744 * blocks owened by reloc trees, the node pointer check is
1745 * skipped, this is because these blocks are fully controlled
1746 * by the space balance code, no one else can modify them.
1747 */
1748 if (!nodes[level - 1] || !key_match ||
1749 (generation == trans->transid &&
1750 btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1751 if (level == 1 || level == lowest_level + 1) {
1752 if (generation == trans->transid) {
1753 btrfs_tree_unlock(eb);
1754 free_extent_buffer(eb);
1755 }
1756 break;
1757 }
1758
1759 if (generation != trans->transid) {
1760 eb = read_tree_block(root, bytenr, blocksize,
1761 generation);
1762 btrfs_tree_lock(eb);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001763 btrfs_set_lock_blocking(eb);
Yan Zhengf82d02d2008-10-29 14:49:05 -04001764 }
Zheng Yan1a40e232008-09-26 10:09:34 -04001765
1766 ret = btrfs_cow_block(trans, root, eb, parent, slot,
Chris Mason9fa8cfe2009-03-13 10:24:59 -04001767 &eb);
Zheng Yan1a40e232008-09-26 10:09:34 -04001768 BUG_ON(ret);
1769
Yan Zhengf82d02d2008-10-29 14:49:05 -04001770 if (root->root_key.objectid ==
1771 BTRFS_TREE_RELOC_OBJECTID) {
1772 if (!nodes[level - 1]) {
1773 nodes[level - 1] = eb->start;
1774 memcpy(&node_keys[level - 1], &key,
1775 sizeof(node_keys[0]));
1776 } else {
1777 WARN_ON(1);
1778 }
1779 }
1780
Zheng Yan1a40e232008-09-26 10:09:34 -04001781 btrfs_tree_unlock(parent);
1782 free_extent_buffer(parent);
1783 parent = eb;
1784 continue;
1785 }
1786
Zheng Yan1a40e232008-09-26 10:09:34 -04001787 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1788 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1789 btrfs_mark_buffer_dirty(parent);
1790
1791 ret = btrfs_inc_extent_ref(trans, root,
1792 nodes[level - 1],
1793 blocksize, parent->start,
1794 btrfs_header_owner(parent),
1795 btrfs_header_generation(parent),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04001796 level - 1);
Zheng Yan1a40e232008-09-26 10:09:34 -04001797 BUG_ON(ret);
Yan Zhengf82d02d2008-10-29 14:49:05 -04001798
1799 /*
1800 * If the block was created in the running transaction,
1801 * it's possible this is the last reference to it, so we
1802 * should drop the subtree.
1803 */
1804 if (generation == trans->transid) {
1805 ret = btrfs_drop_subtree(trans, root, eb, parent);
1806 BUG_ON(ret);
1807 btrfs_tree_unlock(eb);
1808 free_extent_buffer(eb);
1809 } else {
1810 ret = btrfs_free_extent(trans, root, bytenr,
Zheng Yan1a40e232008-09-26 10:09:34 -04001811 blocksize, parent->start,
1812 btrfs_header_owner(parent),
1813 btrfs_header_generation(parent),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04001814 level - 1, 1);
Yan Zhengf82d02d2008-10-29 14:49:05 -04001815 BUG_ON(ret);
Zheng Yan1a40e232008-09-26 10:09:34 -04001816 }
1817 break;
1818 }
1819 btrfs_tree_unlock(parent);
1820 free_extent_buffer(parent);
1821 return 0;
1822}
1823
Chris Mason74123bd2007-02-02 11:05:29 -05001824/*
1825 * adjust the pointers going up the tree, starting at level
1826 * making sure the right key of each node is points to 'key'.
1827 * This is used after shifting pointers to the left, so it stops
1828 * fixing up pointers when a given leaf/node is not in slot 0 of the
1829 * higher levels
Chris Masonaa5d6be2007-02-28 16:35:06 -05001830 *
1831 * If this fails to write a tree block, it returns -1, but continues
1832 * fixing up the blocks in ram so the tree is consistent.
Chris Mason74123bd2007-02-02 11:05:29 -05001833 */
Chris Mason5f39d392007-10-15 16:14:19 -04001834static int fixup_low_keys(struct btrfs_trans_handle *trans,
1835 struct btrfs_root *root, struct btrfs_path *path,
1836 struct btrfs_disk_key *key, int level)
Chris Masonbe0e5c02007-01-26 15:51:26 -05001837{
1838 int i;
Chris Masonaa5d6be2007-02-28 16:35:06 -05001839 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04001840 struct extent_buffer *t;
1841
Chris Mason234b63a2007-03-13 10:46:10 -04001842 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
Chris Masonbe0e5c02007-01-26 15:51:26 -05001843 int tslot = path->slots[i];
Chris Masoneb60cea2007-02-02 09:18:22 -05001844 if (!path->nodes[i])
Chris Masonbe0e5c02007-01-26 15:51:26 -05001845 break;
Chris Mason5f39d392007-10-15 16:14:19 -04001846 t = path->nodes[i];
1847 btrfs_set_node_key(t, key, tslot);
Chris Masond6025572007-03-30 14:27:56 -04001848 btrfs_mark_buffer_dirty(path->nodes[i]);
Chris Masonbe0e5c02007-01-26 15:51:26 -05001849 if (tslot != 0)
1850 break;
1851 }
Chris Masonaa5d6be2007-02-28 16:35:06 -05001852 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05001853}
1854
Chris Mason74123bd2007-02-02 11:05:29 -05001855/*
Zheng Yan31840ae2008-09-23 13:14:14 -04001856 * update item key.
1857 *
1858 * This function isn't completely safe. It's the caller's responsibility
1859 * that the new key won't break the order
1860 */
1861int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1862 struct btrfs_root *root, struct btrfs_path *path,
1863 struct btrfs_key *new_key)
1864{
1865 struct btrfs_disk_key disk_key;
1866 struct extent_buffer *eb;
1867 int slot;
1868
1869 eb = path->nodes[0];
1870 slot = path->slots[0];
1871 if (slot > 0) {
1872 btrfs_item_key(eb, &disk_key, slot - 1);
1873 if (comp_keys(&disk_key, new_key) >= 0)
1874 return -1;
1875 }
1876 if (slot < btrfs_header_nritems(eb) - 1) {
1877 btrfs_item_key(eb, &disk_key, slot + 1);
1878 if (comp_keys(&disk_key, new_key) <= 0)
1879 return -1;
1880 }
1881
1882 btrfs_cpu_key_to_disk(&disk_key, new_key);
1883 btrfs_set_item_key(eb, &disk_key, slot);
1884 btrfs_mark_buffer_dirty(eb);
1885 if (slot == 0)
1886 fixup_low_keys(trans, root, path, &disk_key, 1);
1887 return 0;
1888}
1889
1890/*
Chris Mason74123bd2007-02-02 11:05:29 -05001891 * try to push data from one node into the next node left in the
Chris Mason79f95c82007-03-01 15:16:26 -05001892 * tree.
Chris Masonaa5d6be2007-02-28 16:35:06 -05001893 *
1894 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1895 * error, and > 0 if there was no room in the left hand block.
Chris Mason74123bd2007-02-02 11:05:29 -05001896 */
Chris Mason98ed5172008-01-03 10:01:48 -05001897static int push_node_left(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root, struct extent_buffer *dst,
Chris Mason971a1f62008-04-24 10:54:32 -04001899 struct extent_buffer *src, int empty)
Chris Masonbe0e5c02007-01-26 15:51:26 -05001900{
Chris Masonbe0e5c02007-01-26 15:51:26 -05001901 int push_items = 0;
Chris Masonbb803952007-03-01 12:04:21 -05001902 int src_nritems;
1903 int dst_nritems;
Chris Masonaa5d6be2007-02-28 16:35:06 -05001904 int ret = 0;
Chris Masonbe0e5c02007-01-26 15:51:26 -05001905
Chris Mason5f39d392007-10-15 16:14:19 -04001906 src_nritems = btrfs_header_nritems(src);
1907 dst_nritems = btrfs_header_nritems(dst);
Chris Mason123abc82007-03-14 14:14:43 -04001908 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
Chris Mason7bb86312007-12-11 09:25:06 -05001909 WARN_ON(btrfs_header_generation(src) != trans->transid);
1910 WARN_ON(btrfs_header_generation(dst) != trans->transid);
Chris Mason54aa1f42007-06-22 14:16:25 -04001911
Chris Masonbce4eae2008-04-24 14:42:46 -04001912 if (!empty && src_nritems <= 8)
Chris Mason971a1f62008-04-24 10:54:32 -04001913 return 1;
1914
Chris Masond3977122009-01-05 21:25:51 -05001915 if (push_items <= 0)
Chris Masonbe0e5c02007-01-26 15:51:26 -05001916 return 1;
1917
Chris Masonbce4eae2008-04-24 14:42:46 -04001918 if (empty) {
Chris Mason971a1f62008-04-24 10:54:32 -04001919 push_items = min(src_nritems, push_items);
Chris Masonbce4eae2008-04-24 14:42:46 -04001920 if (push_items < src_nritems) {
1921 /* leave at least 8 pointers in the node if
1922 * we aren't going to empty it
1923 */
1924 if (src_nritems - push_items < 8) {
1925 if (push_items <= 8)
1926 return 1;
1927 push_items -= 8;
1928 }
1929 }
1930 } else
1931 push_items = min(src_nritems - 8, push_items);
Chris Mason79f95c82007-03-01 15:16:26 -05001932
Chris Mason5f39d392007-10-15 16:14:19 -04001933 copy_extent_buffer(dst, src,
1934 btrfs_node_key_ptr_offset(dst_nritems),
1935 btrfs_node_key_ptr_offset(0),
Chris Masond3977122009-01-05 21:25:51 -05001936 push_items * sizeof(struct btrfs_key_ptr));
Chris Mason5f39d392007-10-15 16:14:19 -04001937
Chris Masonbb803952007-03-01 12:04:21 -05001938 if (push_items < src_nritems) {
Chris Mason5f39d392007-10-15 16:14:19 -04001939 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1940 btrfs_node_key_ptr_offset(push_items),
1941 (src_nritems - push_items) *
1942 sizeof(struct btrfs_key_ptr));
Chris Masonbb803952007-03-01 12:04:21 -05001943 }
Chris Mason5f39d392007-10-15 16:14:19 -04001944 btrfs_set_header_nritems(src, src_nritems - push_items);
1945 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1946 btrfs_mark_buffer_dirty(src);
1947 btrfs_mark_buffer_dirty(dst);
Zheng Yan31840ae2008-09-23 13:14:14 -04001948
1949 ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1950 BUG_ON(ret);
1951
Chris Masonbb803952007-03-01 12:04:21 -05001952 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05001953}
1954
Chris Mason97571fd2007-02-24 13:39:08 -05001955/*
Chris Mason79f95c82007-03-01 15:16:26 -05001956 * try to push data from one node into the next node right in the
1957 * tree.
1958 *
1959 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1960 * error, and > 0 if there was no room in the right hand block.
1961 *
1962 * this will only push up to 1/2 the contents of the left node over
1963 */
Chris Mason5f39d392007-10-15 16:14:19 -04001964static int balance_node_right(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root,
1966 struct extent_buffer *dst,
1967 struct extent_buffer *src)
Chris Mason79f95c82007-03-01 15:16:26 -05001968{
Chris Mason79f95c82007-03-01 15:16:26 -05001969 int push_items = 0;
1970 int max_push;
1971 int src_nritems;
1972 int dst_nritems;
1973 int ret = 0;
Chris Mason79f95c82007-03-01 15:16:26 -05001974
Chris Mason7bb86312007-12-11 09:25:06 -05001975 WARN_ON(btrfs_header_generation(src) != trans->transid);
1976 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1977
Chris Mason5f39d392007-10-15 16:14:19 -04001978 src_nritems = btrfs_header_nritems(src);
1979 dst_nritems = btrfs_header_nritems(dst);
Chris Mason123abc82007-03-14 14:14:43 -04001980 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
Chris Masond3977122009-01-05 21:25:51 -05001981 if (push_items <= 0)
Chris Mason79f95c82007-03-01 15:16:26 -05001982 return 1;
Chris Masonbce4eae2008-04-24 14:42:46 -04001983
Chris Masond3977122009-01-05 21:25:51 -05001984 if (src_nritems < 4)
Chris Masonbce4eae2008-04-24 14:42:46 -04001985 return 1;
Chris Mason79f95c82007-03-01 15:16:26 -05001986
1987 max_push = src_nritems / 2 + 1;
1988 /* don't try to empty the node */
Chris Masond3977122009-01-05 21:25:51 -05001989 if (max_push >= src_nritems)
Chris Mason79f95c82007-03-01 15:16:26 -05001990 return 1;
Yan252c38f2007-08-29 09:11:44 -04001991
Chris Mason79f95c82007-03-01 15:16:26 -05001992 if (max_push < push_items)
1993 push_items = max_push;
1994
Chris Mason5f39d392007-10-15 16:14:19 -04001995 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1996 btrfs_node_key_ptr_offset(0),
1997 (dst_nritems) *
1998 sizeof(struct btrfs_key_ptr));
Chris Masond6025572007-03-30 14:27:56 -04001999
Chris Mason5f39d392007-10-15 16:14:19 -04002000 copy_extent_buffer(dst, src,
2001 btrfs_node_key_ptr_offset(0),
2002 btrfs_node_key_ptr_offset(src_nritems - push_items),
Chris Masond3977122009-01-05 21:25:51 -05002003 push_items * sizeof(struct btrfs_key_ptr));
Chris Mason79f95c82007-03-01 15:16:26 -05002004
Chris Mason5f39d392007-10-15 16:14:19 -04002005 btrfs_set_header_nritems(src, src_nritems - push_items);
2006 btrfs_set_header_nritems(dst, dst_nritems + push_items);
Chris Mason79f95c82007-03-01 15:16:26 -05002007
Chris Mason5f39d392007-10-15 16:14:19 -04002008 btrfs_mark_buffer_dirty(src);
2009 btrfs_mark_buffer_dirty(dst);
Zheng Yan31840ae2008-09-23 13:14:14 -04002010
2011 ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
2012 BUG_ON(ret);
2013
Chris Mason79f95c82007-03-01 15:16:26 -05002014 return ret;
2015}
2016
2017/*
Chris Mason97571fd2007-02-24 13:39:08 -05002018 * helper function to insert a new root level in the tree.
2019 * A new node is allocated, and a single item is inserted to
2020 * point to the existing root
Chris Masonaa5d6be2007-02-28 16:35:06 -05002021 *
2022 * returns zero on success or < 0 on failure.
Chris Mason97571fd2007-02-24 13:39:08 -05002023 */
Chris Masond3977122009-01-05 21:25:51 -05002024static noinline int insert_new_root(struct btrfs_trans_handle *trans,
Chris Mason5f39d392007-10-15 16:14:19 -04002025 struct btrfs_root *root,
2026 struct btrfs_path *path, int level)
Chris Mason5c680ed2007-02-22 11:39:13 -05002027{
Chris Mason7bb86312007-12-11 09:25:06 -05002028 u64 lower_gen;
Chris Mason5f39d392007-10-15 16:14:19 -04002029 struct extent_buffer *lower;
2030 struct extent_buffer *c;
Chris Mason925baed2008-06-25 16:01:30 -04002031 struct extent_buffer *old;
Chris Mason5f39d392007-10-15 16:14:19 -04002032 struct btrfs_disk_key lower_key;
Zheng Yan31840ae2008-09-23 13:14:14 -04002033 int ret;
Chris Mason5c680ed2007-02-22 11:39:13 -05002034
2035 BUG_ON(path->nodes[level]);
2036 BUG_ON(path->nodes[level-1] != root->node);
2037
Chris Mason7bb86312007-12-11 09:25:06 -05002038 lower = path->nodes[level-1];
2039 if (level == 1)
2040 btrfs_item_key(lower, &lower_key, 0);
2041 else
2042 btrfs_node_key(lower, &lower_key, 0);
2043
Zheng Yan31840ae2008-09-23 13:14:14 -04002044 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2045 root->root_key.objectid, trans->transid,
Christoph Hellwigad3d81b2008-09-05 16:43:28 -04002046 level, root->node->start, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002047 if (IS_ERR(c))
2048 return PTR_ERR(c);
Chris Mason925baed2008-06-25 16:01:30 -04002049
Chris Mason5f39d392007-10-15 16:14:19 -04002050 memset_extent_buffer(c, 0, 0, root->nodesize);
2051 btrfs_set_header_nritems(c, 1);
2052 btrfs_set_header_level(c, level);
Chris Masondb945352007-10-15 16:15:53 -04002053 btrfs_set_header_bytenr(c, c->start);
Chris Mason5f39d392007-10-15 16:14:19 -04002054 btrfs_set_header_generation(c, trans->transid);
2055 btrfs_set_header_owner(c, root->root_key.objectid);
Chris Masond5719762007-03-23 10:01:08 -04002056
Chris Mason5f39d392007-10-15 16:14:19 -04002057 write_extent_buffer(c, root->fs_info->fsid,
2058 (unsigned long)btrfs_header_fsid(c),
2059 BTRFS_FSID_SIZE);
Chris Masone17cade2008-04-15 15:41:47 -04002060
2061 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2062 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2063 BTRFS_UUID_SIZE);
2064
Chris Mason5f39d392007-10-15 16:14:19 -04002065 btrfs_set_node_key(c, &lower_key, 0);
Chris Masondb945352007-10-15 16:15:53 -04002066 btrfs_set_node_blockptr(c, 0, lower->start);
Chris Mason7bb86312007-12-11 09:25:06 -05002067 lower_gen = btrfs_header_generation(lower);
Zheng Yan31840ae2008-09-23 13:14:14 -04002068 WARN_ON(lower_gen != trans->transid);
Chris Mason7bb86312007-12-11 09:25:06 -05002069
2070 btrfs_set_node_ptr_generation(c, 0, lower_gen);
Chris Mason5f39d392007-10-15 16:14:19 -04002071
2072 btrfs_mark_buffer_dirty(c);
Chris Masond5719762007-03-23 10:01:08 -04002073
Chris Mason925baed2008-06-25 16:01:30 -04002074 spin_lock(&root->node_lock);
2075 old = root->node;
Chris Mason5f39d392007-10-15 16:14:19 -04002076 root->node = c;
Chris Mason925baed2008-06-25 16:01:30 -04002077 spin_unlock(&root->node_lock);
2078
Zheng Yan31840ae2008-09-23 13:14:14 -04002079 ret = btrfs_update_extent_ref(trans, root, lower->start,
Chris Mason56bec292009-03-13 10:10:06 -04002080 lower->len, lower->start, c->start,
Zheng Yan31840ae2008-09-23 13:14:14 -04002081 root->root_key.objectid,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04002082 trans->transid, level - 1);
Zheng Yan31840ae2008-09-23 13:14:14 -04002083 BUG_ON(ret);
2084
Chris Mason925baed2008-06-25 16:01:30 -04002085 /* the super has an extra ref to root->node */
2086 free_extent_buffer(old);
2087
Chris Mason0b86a832008-03-24 15:01:56 -04002088 add_root_to_dirty_list(root);
Chris Mason5f39d392007-10-15 16:14:19 -04002089 extent_buffer_get(c);
2090 path->nodes[level] = c;
Chris Mason925baed2008-06-25 16:01:30 -04002091 path->locks[level] = 1;
Chris Mason5c680ed2007-02-22 11:39:13 -05002092 path->slots[level] = 0;
2093 return 0;
2094}
2095
Chris Mason74123bd2007-02-02 11:05:29 -05002096/*
2097 * worker function to insert a single pointer in a node.
2098 * the node should have enough room for the pointer already
Chris Mason97571fd2007-02-24 13:39:08 -05002099 *
Chris Mason74123bd2007-02-02 11:05:29 -05002100 * slot and level indicate where you want the key to go, and
2101 * blocknr is the block the key points to.
Chris Masonaa5d6be2007-02-28 16:35:06 -05002102 *
2103 * returns zero on success and < 0 on any error
Chris Mason74123bd2007-02-02 11:05:29 -05002104 */
Chris Masone089f052007-03-16 16:20:31 -04002105static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2106 *root, struct btrfs_path *path, struct btrfs_disk_key
Chris Masondb945352007-10-15 16:15:53 -04002107 *key, u64 bytenr, int slot, int level)
Chris Mason74123bd2007-02-02 11:05:29 -05002108{
Chris Mason5f39d392007-10-15 16:14:19 -04002109 struct extent_buffer *lower;
Chris Mason74123bd2007-02-02 11:05:29 -05002110 int nritems;
Chris Mason5c680ed2007-02-22 11:39:13 -05002111
2112 BUG_ON(!path->nodes[level]);
Chris Mason5f39d392007-10-15 16:14:19 -04002113 lower = path->nodes[level];
2114 nritems = btrfs_header_nritems(lower);
Chris Mason74123bd2007-02-02 11:05:29 -05002115 if (slot > nritems)
2116 BUG();
Chris Mason123abc82007-03-14 14:14:43 -04002117 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
Chris Mason74123bd2007-02-02 11:05:29 -05002118 BUG();
2119 if (slot != nritems) {
Chris Mason5f39d392007-10-15 16:14:19 -04002120 memmove_extent_buffer(lower,
2121 btrfs_node_key_ptr_offset(slot + 1),
2122 btrfs_node_key_ptr_offset(slot),
Chris Masond6025572007-03-30 14:27:56 -04002123 (nritems - slot) * sizeof(struct btrfs_key_ptr));
Chris Mason74123bd2007-02-02 11:05:29 -05002124 }
Chris Mason5f39d392007-10-15 16:14:19 -04002125 btrfs_set_node_key(lower, key, slot);
Chris Masondb945352007-10-15 16:15:53 -04002126 btrfs_set_node_blockptr(lower, slot, bytenr);
Chris Mason74493f72007-12-11 09:25:06 -05002127 WARN_ON(trans->transid == 0);
2128 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002129 btrfs_set_header_nritems(lower, nritems + 1);
2130 btrfs_mark_buffer_dirty(lower);
Chris Mason74123bd2007-02-02 11:05:29 -05002131 return 0;
2132}
2133
Chris Mason97571fd2007-02-24 13:39:08 -05002134/*
2135 * split the node at the specified level in path in two.
2136 * The path is corrected to point to the appropriate node after the split
2137 *
2138 * Before splitting this tries to make some room in the node by pushing
2139 * left and right, if either one works, it returns right away.
Chris Masonaa5d6be2007-02-28 16:35:06 -05002140 *
2141 * returns 0 on success and < 0 on failure
Chris Mason97571fd2007-02-24 13:39:08 -05002142 */
Chris Masone02119d2008-09-05 16:13:11 -04002143static noinline int split_node(struct btrfs_trans_handle *trans,
2144 struct btrfs_root *root,
2145 struct btrfs_path *path, int level)
Chris Masonbe0e5c02007-01-26 15:51:26 -05002146{
Chris Mason5f39d392007-10-15 16:14:19 -04002147 struct extent_buffer *c;
2148 struct extent_buffer *split;
2149 struct btrfs_disk_key disk_key;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002150 int mid;
Chris Mason5c680ed2007-02-22 11:39:13 -05002151 int ret;
Chris Masonaa5d6be2007-02-28 16:35:06 -05002152 int wret;
Chris Mason7518a232007-03-12 12:01:18 -04002153 u32 c_nritems;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002154
Chris Mason5f39d392007-10-15 16:14:19 -04002155 c = path->nodes[level];
Chris Mason7bb86312007-12-11 09:25:06 -05002156 WARN_ON(btrfs_header_generation(c) != trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002157 if (c == root->node) {
Chris Mason5c680ed2007-02-22 11:39:13 -05002158 /* trying to split the root, lets make a new one */
Chris Masone089f052007-03-16 16:20:31 -04002159 ret = insert_new_root(trans, root, path, level + 1);
Chris Mason5c680ed2007-02-22 11:39:13 -05002160 if (ret)
2161 return ret;
Chris Masone66f7092007-04-20 13:16:02 -04002162 } else {
2163 ret = push_nodes_for_insert(trans, root, path, level);
Chris Mason5f39d392007-10-15 16:14:19 -04002164 c = path->nodes[level];
2165 if (!ret && btrfs_header_nritems(c) <
Chris Masonc448acf2008-04-24 09:34:34 -04002166 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
Chris Masone66f7092007-04-20 13:16:02 -04002167 return 0;
Chris Mason54aa1f42007-06-22 14:16:25 -04002168 if (ret < 0)
2169 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002170 }
Chris Masone66f7092007-04-20 13:16:02 -04002171
Chris Mason5f39d392007-10-15 16:14:19 -04002172 c_nritems = btrfs_header_nritems(c);
Chris Mason7bb86312007-12-11 09:25:06 -05002173
Chris Mason925baed2008-06-25 16:01:30 -04002174 split = btrfs_alloc_free_block(trans, root, root->nodesize,
Zheng Yan31840ae2008-09-23 13:14:14 -04002175 path->nodes[level + 1]->start,
2176 root->root_key.objectid,
2177 trans->transid, level, c->start, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002178 if (IS_ERR(split))
2179 return PTR_ERR(split);
Chris Mason54aa1f42007-06-22 14:16:25 -04002180
Chris Mason5f39d392007-10-15 16:14:19 -04002181 btrfs_set_header_flags(split, btrfs_header_flags(c));
2182 btrfs_set_header_level(split, btrfs_header_level(c));
Chris Masondb945352007-10-15 16:15:53 -04002183 btrfs_set_header_bytenr(split, split->start);
Chris Mason5f39d392007-10-15 16:14:19 -04002184 btrfs_set_header_generation(split, trans->transid);
2185 btrfs_set_header_owner(split, root->root_key.objectid);
Chris Mason63b10fc2008-04-01 11:21:32 -04002186 btrfs_set_header_flags(split, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002187 write_extent_buffer(split, root->fs_info->fsid,
2188 (unsigned long)btrfs_header_fsid(split),
2189 BTRFS_FSID_SIZE);
Chris Masone17cade2008-04-15 15:41:47 -04002190 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2191 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2192 BTRFS_UUID_SIZE);
Chris Mason5f39d392007-10-15 16:14:19 -04002193
Chris Mason7518a232007-03-12 12:01:18 -04002194 mid = (c_nritems + 1) / 2;
Chris Mason5f39d392007-10-15 16:14:19 -04002195
2196 copy_extent_buffer(split, c,
2197 btrfs_node_key_ptr_offset(0),
2198 btrfs_node_key_ptr_offset(mid),
2199 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2200 btrfs_set_header_nritems(split, c_nritems - mid);
2201 btrfs_set_header_nritems(c, mid);
Chris Masonaa5d6be2007-02-28 16:35:06 -05002202 ret = 0;
2203
Chris Mason5f39d392007-10-15 16:14:19 -04002204 btrfs_mark_buffer_dirty(c);
2205 btrfs_mark_buffer_dirty(split);
2206
2207 btrfs_node_key(split, &disk_key, 0);
Chris Masondb945352007-10-15 16:15:53 -04002208 wret = insert_ptr(trans, root, path, &disk_key, split->start,
Chris Mason5f39d392007-10-15 16:14:19 -04002209 path->slots[level + 1] + 1,
Chris Mason123abc82007-03-14 14:14:43 -04002210 level + 1);
Chris Masonaa5d6be2007-02-28 16:35:06 -05002211 if (wret)
2212 ret = wret;
2213
Zheng Yan31840ae2008-09-23 13:14:14 -04002214 ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2215 BUG_ON(ret);
2216
Chris Mason5de08d72007-02-24 06:24:44 -05002217 if (path->slots[level] >= mid) {
Chris Mason5c680ed2007-02-22 11:39:13 -05002218 path->slots[level] -= mid;
Chris Mason925baed2008-06-25 16:01:30 -04002219 btrfs_tree_unlock(c);
Chris Mason5f39d392007-10-15 16:14:19 -04002220 free_extent_buffer(c);
2221 path->nodes[level] = split;
Chris Mason5c680ed2007-02-22 11:39:13 -05002222 path->slots[level + 1] += 1;
2223 } else {
Chris Mason925baed2008-06-25 16:01:30 -04002224 btrfs_tree_unlock(split);
Chris Mason5f39d392007-10-15 16:14:19 -04002225 free_extent_buffer(split);
Chris Masonbe0e5c02007-01-26 15:51:26 -05002226 }
Chris Masonaa5d6be2007-02-28 16:35:06 -05002227 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002228}
2229
Chris Mason74123bd2007-02-02 11:05:29 -05002230/*
2231 * how many bytes are required to store the items in a leaf. start
2232 * and nr indicate which items in the leaf to check. This totals up the
2233 * space used both by the item structs and the item data
2234 */
Chris Mason5f39d392007-10-15 16:14:19 -04002235static int leaf_space_used(struct extent_buffer *l, int start, int nr)
Chris Masonbe0e5c02007-01-26 15:51:26 -05002236{
2237 int data_len;
Chris Mason5f39d392007-10-15 16:14:19 -04002238 int nritems = btrfs_header_nritems(l);
Chris Masond4dbff92007-04-04 14:08:15 -04002239 int end = min(nritems, start + nr) - 1;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002240
2241 if (!nr)
2242 return 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002243 data_len = btrfs_item_end_nr(l, start);
2244 data_len = data_len - btrfs_item_offset_nr(l, end);
Chris Mason0783fcf2007-03-12 20:12:07 -04002245 data_len += sizeof(struct btrfs_item) * nr;
Chris Masond4dbff92007-04-04 14:08:15 -04002246 WARN_ON(data_len < 0);
Chris Masonbe0e5c02007-01-26 15:51:26 -05002247 return data_len;
2248}
2249
Chris Mason74123bd2007-02-02 11:05:29 -05002250/*
Chris Masond4dbff92007-04-04 14:08:15 -04002251 * The space between the end of the leaf items and
2252 * the start of the leaf data. IOW, how much room
2253 * the leaf has left for both items and data
2254 */
Chris Masond3977122009-01-05 21:25:51 -05002255noinline int btrfs_leaf_free_space(struct btrfs_root *root,
Chris Masone02119d2008-09-05 16:13:11 -04002256 struct extent_buffer *leaf)
Chris Masond4dbff92007-04-04 14:08:15 -04002257{
Chris Mason5f39d392007-10-15 16:14:19 -04002258 int nritems = btrfs_header_nritems(leaf);
2259 int ret;
2260 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2261 if (ret < 0) {
Chris Masond3977122009-01-05 21:25:51 -05002262 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2263 "used %d nritems %d\n",
Jens Axboeae2f5412007-10-19 09:22:59 -04002264 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
Chris Mason5f39d392007-10-15 16:14:19 -04002265 leaf_space_used(leaf, 0, nritems), nritems);
2266 }
2267 return ret;
Chris Masond4dbff92007-04-04 14:08:15 -04002268}
2269
Chris Mason44871b12009-03-13 10:04:31 -04002270static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2271 struct btrfs_root *root,
2272 struct btrfs_path *path,
2273 int data_size, int empty,
2274 struct extent_buffer *right,
2275 int free_space, u32 left_nritems)
Chris Mason00ec4c52007-02-24 12:47:20 -05002276{
Chris Mason5f39d392007-10-15 16:14:19 -04002277 struct extent_buffer *left = path->nodes[0];
Chris Mason44871b12009-03-13 10:04:31 -04002278 struct extent_buffer *upper = path->nodes[1];
Chris Mason5f39d392007-10-15 16:14:19 -04002279 struct btrfs_disk_key disk_key;
Chris Mason00ec4c52007-02-24 12:47:20 -05002280 int slot;
Chris Mason34a38212007-11-07 13:31:03 -05002281 u32 i;
Chris Mason00ec4c52007-02-24 12:47:20 -05002282 int push_space = 0;
2283 int push_items = 0;
Chris Mason0783fcf2007-03-12 20:12:07 -04002284 struct btrfs_item *item;
Chris Mason34a38212007-11-07 13:31:03 -05002285 u32 nr;
Chris Mason7518a232007-03-12 12:01:18 -04002286 u32 right_nritems;
Chris Mason5f39d392007-10-15 16:14:19 -04002287 u32 data_end;
Chris Masondb945352007-10-15 16:15:53 -04002288 u32 this_item_size;
Chris Mason54aa1f42007-06-22 14:16:25 -04002289 int ret;
Chris Mason00ec4c52007-02-24 12:47:20 -05002290
Chris Mason34a38212007-11-07 13:31:03 -05002291 if (empty)
2292 nr = 0;
2293 else
2294 nr = 1;
2295
Zheng Yan31840ae2008-09-23 13:14:14 -04002296 if (path->slots[0] >= left_nritems)
Yan Zheng87b29b22008-12-17 10:21:48 -05002297 push_space += data_size;
Zheng Yan31840ae2008-09-23 13:14:14 -04002298
Chris Mason44871b12009-03-13 10:04:31 -04002299 slot = path->slots[1];
Chris Mason34a38212007-11-07 13:31:03 -05002300 i = left_nritems - 1;
2301 while (i >= nr) {
Chris Mason5f39d392007-10-15 16:14:19 -04002302 item = btrfs_item_nr(left, i);
Chris Masondb945352007-10-15 16:15:53 -04002303
Zheng Yan31840ae2008-09-23 13:14:14 -04002304 if (!empty && push_items > 0) {
2305 if (path->slots[0] > i)
2306 break;
2307 if (path->slots[0] == i) {
2308 int space = btrfs_leaf_free_space(root, left);
2309 if (space + push_space * 2 > free_space)
2310 break;
2311 }
2312 }
2313
Chris Mason00ec4c52007-02-24 12:47:20 -05002314 if (path->slots[0] == i)
Yan Zheng87b29b22008-12-17 10:21:48 -05002315 push_space += data_size;
Chris Masondb945352007-10-15 16:15:53 -04002316
2317 if (!left->map_token) {
2318 map_extent_buffer(left, (unsigned long)item,
2319 sizeof(struct btrfs_item),
2320 &left->map_token, &left->kaddr,
2321 &left->map_start, &left->map_len,
2322 KM_USER1);
2323 }
2324
2325 this_item_size = btrfs_item_size(left, item);
2326 if (this_item_size + sizeof(*item) + push_space > free_space)
Chris Mason00ec4c52007-02-24 12:47:20 -05002327 break;
Zheng Yan31840ae2008-09-23 13:14:14 -04002328
Chris Mason00ec4c52007-02-24 12:47:20 -05002329 push_items++;
Chris Masondb945352007-10-15 16:15:53 -04002330 push_space += this_item_size + sizeof(*item);
Chris Mason34a38212007-11-07 13:31:03 -05002331 if (i == 0)
2332 break;
2333 i--;
Chris Masondb945352007-10-15 16:15:53 -04002334 }
2335 if (left->map_token) {
2336 unmap_extent_buffer(left, left->map_token, KM_USER1);
2337 left->map_token = NULL;
Chris Mason00ec4c52007-02-24 12:47:20 -05002338 }
Chris Mason5f39d392007-10-15 16:14:19 -04002339
Chris Mason925baed2008-06-25 16:01:30 -04002340 if (push_items == 0)
2341 goto out_unlock;
Chris Mason5f39d392007-10-15 16:14:19 -04002342
Chris Mason34a38212007-11-07 13:31:03 -05002343 if (!empty && push_items == left_nritems)
Chris Masona429e512007-04-18 16:15:28 -04002344 WARN_ON(1);
Chris Mason5f39d392007-10-15 16:14:19 -04002345
Chris Mason00ec4c52007-02-24 12:47:20 -05002346 /* push left to right */
Chris Mason5f39d392007-10-15 16:14:19 -04002347 right_nritems = btrfs_header_nritems(right);
Chris Mason34a38212007-11-07 13:31:03 -05002348
Chris Mason5f39d392007-10-15 16:14:19 -04002349 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
Chris Mason123abc82007-03-14 14:14:43 -04002350 push_space -= leaf_data_end(root, left);
Chris Mason5f39d392007-10-15 16:14:19 -04002351
Chris Mason00ec4c52007-02-24 12:47:20 -05002352 /* make room in the right data area */
Chris Mason5f39d392007-10-15 16:14:19 -04002353 data_end = leaf_data_end(root, right);
2354 memmove_extent_buffer(right,
2355 btrfs_leaf_data(right) + data_end - push_space,
2356 btrfs_leaf_data(right) + data_end,
2357 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2358
Chris Mason00ec4c52007-02-24 12:47:20 -05002359 /* copy from the left data area */
Chris Mason5f39d392007-10-15 16:14:19 -04002360 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
Chris Masond6025572007-03-30 14:27:56 -04002361 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2362 btrfs_leaf_data(left) + leaf_data_end(root, left),
2363 push_space);
Chris Mason5f39d392007-10-15 16:14:19 -04002364
2365 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2366 btrfs_item_nr_offset(0),
2367 right_nritems * sizeof(struct btrfs_item));
2368
Chris Mason00ec4c52007-02-24 12:47:20 -05002369 /* copy the items from left to right */
Chris Mason5f39d392007-10-15 16:14:19 -04002370 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2371 btrfs_item_nr_offset(left_nritems - push_items),
2372 push_items * sizeof(struct btrfs_item));
Chris Mason00ec4c52007-02-24 12:47:20 -05002373
2374 /* update the item pointers */
Chris Mason7518a232007-03-12 12:01:18 -04002375 right_nritems += push_items;
Chris Mason5f39d392007-10-15 16:14:19 -04002376 btrfs_set_header_nritems(right, right_nritems);
Chris Mason123abc82007-03-14 14:14:43 -04002377 push_space = BTRFS_LEAF_DATA_SIZE(root);
Chris Mason7518a232007-03-12 12:01:18 -04002378 for (i = 0; i < right_nritems; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04002379 item = btrfs_item_nr(right, i);
Chris Masondb945352007-10-15 16:15:53 -04002380 if (!right->map_token) {
2381 map_extent_buffer(right, (unsigned long)item,
2382 sizeof(struct btrfs_item),
2383 &right->map_token, &right->kaddr,
2384 &right->map_start, &right->map_len,
2385 KM_USER1);
2386 }
2387 push_space -= btrfs_item_size(right, item);
2388 btrfs_set_item_offset(right, item, push_space);
2389 }
2390
2391 if (right->map_token) {
2392 unmap_extent_buffer(right, right->map_token, KM_USER1);
2393 right->map_token = NULL;
Chris Mason00ec4c52007-02-24 12:47:20 -05002394 }
Chris Mason7518a232007-03-12 12:01:18 -04002395 left_nritems -= push_items;
Chris Mason5f39d392007-10-15 16:14:19 -04002396 btrfs_set_header_nritems(left, left_nritems);
Chris Mason00ec4c52007-02-24 12:47:20 -05002397
Chris Mason34a38212007-11-07 13:31:03 -05002398 if (left_nritems)
2399 btrfs_mark_buffer_dirty(left);
Chris Mason5f39d392007-10-15 16:14:19 -04002400 btrfs_mark_buffer_dirty(right);
Chris Masona429e512007-04-18 16:15:28 -04002401
Zheng Yan31840ae2008-09-23 13:14:14 -04002402 ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2403 BUG_ON(ret);
2404
Chris Mason5f39d392007-10-15 16:14:19 -04002405 btrfs_item_key(right, &disk_key, 0);
2406 btrfs_set_node_key(upper, &disk_key, slot + 1);
Chris Masond6025572007-03-30 14:27:56 -04002407 btrfs_mark_buffer_dirty(upper);
Chris Mason02217ed2007-03-02 16:08:05 -05002408
Chris Mason00ec4c52007-02-24 12:47:20 -05002409 /* then fixup the leaf pointer in the path */
Chris Mason7518a232007-03-12 12:01:18 -04002410 if (path->slots[0] >= left_nritems) {
2411 path->slots[0] -= left_nritems;
Chris Mason925baed2008-06-25 16:01:30 -04002412 if (btrfs_header_nritems(path->nodes[0]) == 0)
2413 clean_tree_block(trans, root, path->nodes[0]);
2414 btrfs_tree_unlock(path->nodes[0]);
Chris Mason5f39d392007-10-15 16:14:19 -04002415 free_extent_buffer(path->nodes[0]);
2416 path->nodes[0] = right;
Chris Mason00ec4c52007-02-24 12:47:20 -05002417 path->slots[1] += 1;
2418 } else {
Chris Mason925baed2008-06-25 16:01:30 -04002419 btrfs_tree_unlock(right);
Chris Mason5f39d392007-10-15 16:14:19 -04002420 free_extent_buffer(right);
Chris Mason00ec4c52007-02-24 12:47:20 -05002421 }
2422 return 0;
Chris Mason925baed2008-06-25 16:01:30 -04002423
2424out_unlock:
2425 btrfs_tree_unlock(right);
2426 free_extent_buffer(right);
2427 return 1;
Chris Mason00ec4c52007-02-24 12:47:20 -05002428}
Chris Mason925baed2008-06-25 16:01:30 -04002429
Chris Mason00ec4c52007-02-24 12:47:20 -05002430/*
Chris Mason44871b12009-03-13 10:04:31 -04002431 * push some data in the path leaf to the right, trying to free up at
2432 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2433 *
2434 * returns 1 if the push failed because the other node didn't have enough
2435 * room, 0 if everything worked out and < 0 if there were major errors.
2436 */
2437static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2438 *root, struct btrfs_path *path, int data_size,
2439 int empty)
2440{
2441 struct extent_buffer *left = path->nodes[0];
2442 struct extent_buffer *right;
2443 struct extent_buffer *upper;
2444 int slot;
2445 int free_space;
2446 u32 left_nritems;
2447 int ret;
2448
2449 if (!path->nodes[1])
2450 return 1;
2451
2452 slot = path->slots[1];
2453 upper = path->nodes[1];
2454 if (slot >= btrfs_header_nritems(upper) - 1)
2455 return 1;
2456
2457 btrfs_assert_tree_locked(path->nodes[1]);
2458
2459 right = read_node_slot(root, upper, slot + 1);
2460 btrfs_tree_lock(right);
2461 btrfs_set_lock_blocking(right);
2462
2463 free_space = btrfs_leaf_free_space(root, right);
2464 if (free_space < data_size)
2465 goto out_unlock;
2466
2467 /* cow and double check */
2468 ret = btrfs_cow_block(trans, root, right, upper,
2469 slot + 1, &right);
2470 if (ret)
2471 goto out_unlock;
2472
2473 free_space = btrfs_leaf_free_space(root, right);
2474 if (free_space < data_size)
2475 goto out_unlock;
2476
2477 left_nritems = btrfs_header_nritems(left);
2478 if (left_nritems == 0)
2479 goto out_unlock;
2480
2481 return __push_leaf_right(trans, root, path, data_size, empty,
2482 right, free_space, left_nritems);
2483out_unlock:
2484 btrfs_tree_unlock(right);
2485 free_extent_buffer(right);
2486 return 1;
2487}
2488
2489/*
Chris Mason74123bd2007-02-02 11:05:29 -05002490 * push some data in the path leaf to the left, trying to free up at
2491 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2492 */
Chris Mason44871b12009-03-13 10:04:31 -04002493static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2494 struct btrfs_root *root,
2495 struct btrfs_path *path, int data_size,
2496 int empty, struct extent_buffer *left,
2497 int free_space, int right_nritems)
Chris Masonbe0e5c02007-01-26 15:51:26 -05002498{
Chris Mason5f39d392007-10-15 16:14:19 -04002499 struct btrfs_disk_key disk_key;
2500 struct extent_buffer *right = path->nodes[0];
Chris Masonbe0e5c02007-01-26 15:51:26 -05002501 int slot;
2502 int i;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002503 int push_space = 0;
2504 int push_items = 0;
Chris Mason0783fcf2007-03-12 20:12:07 -04002505 struct btrfs_item *item;
Chris Mason7518a232007-03-12 12:01:18 -04002506 u32 old_left_nritems;
Chris Mason34a38212007-11-07 13:31:03 -05002507 u32 nr;
Chris Masonaa5d6be2007-02-28 16:35:06 -05002508 int ret = 0;
2509 int wret;
Chris Masondb945352007-10-15 16:15:53 -04002510 u32 this_item_size;
2511 u32 old_left_item_size;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002512
2513 slot = path->slots[1];
Chris Mason02217ed2007-03-02 16:08:05 -05002514
Chris Mason34a38212007-11-07 13:31:03 -05002515 if (empty)
2516 nr = right_nritems;
2517 else
2518 nr = right_nritems - 1;
2519
2520 for (i = 0; i < nr; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04002521 item = btrfs_item_nr(right, i);
Chris Masondb945352007-10-15 16:15:53 -04002522 if (!right->map_token) {
2523 map_extent_buffer(right, (unsigned long)item,
2524 sizeof(struct btrfs_item),
2525 &right->map_token, &right->kaddr,
2526 &right->map_start, &right->map_len,
2527 KM_USER1);
2528 }
2529
Zheng Yan31840ae2008-09-23 13:14:14 -04002530 if (!empty && push_items > 0) {
2531 if (path->slots[0] < i)
2532 break;
2533 if (path->slots[0] == i) {
2534 int space = btrfs_leaf_free_space(root, right);
2535 if (space + push_space * 2 > free_space)
2536 break;
2537 }
2538 }
2539
Chris Masonbe0e5c02007-01-26 15:51:26 -05002540 if (path->slots[0] == i)
Yan Zheng87b29b22008-12-17 10:21:48 -05002541 push_space += data_size;
Chris Masondb945352007-10-15 16:15:53 -04002542
2543 this_item_size = btrfs_item_size(right, item);
2544 if (this_item_size + sizeof(*item) + push_space > free_space)
Chris Masonbe0e5c02007-01-26 15:51:26 -05002545 break;
Chris Masondb945352007-10-15 16:15:53 -04002546
Chris Masonbe0e5c02007-01-26 15:51:26 -05002547 push_items++;
Chris Masondb945352007-10-15 16:15:53 -04002548 push_space += this_item_size + sizeof(*item);
Chris Masonbe0e5c02007-01-26 15:51:26 -05002549 }
Chris Masondb945352007-10-15 16:15:53 -04002550
2551 if (right->map_token) {
2552 unmap_extent_buffer(right, right->map_token, KM_USER1);
2553 right->map_token = NULL;
2554 }
2555
Chris Masonbe0e5c02007-01-26 15:51:26 -05002556 if (push_items == 0) {
Chris Mason925baed2008-06-25 16:01:30 -04002557 ret = 1;
2558 goto out;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002559 }
Chris Mason34a38212007-11-07 13:31:03 -05002560 if (!empty && push_items == btrfs_header_nritems(right))
Chris Masona429e512007-04-18 16:15:28 -04002561 WARN_ON(1);
Chris Mason5f39d392007-10-15 16:14:19 -04002562
Chris Masonbe0e5c02007-01-26 15:51:26 -05002563 /* push data from right to left */
Chris Mason5f39d392007-10-15 16:14:19 -04002564 copy_extent_buffer(left, right,
2565 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2566 btrfs_item_nr_offset(0),
2567 push_items * sizeof(struct btrfs_item));
2568
Chris Mason123abc82007-03-14 14:14:43 -04002569 push_space = BTRFS_LEAF_DATA_SIZE(root) -
Chris Masond3977122009-01-05 21:25:51 -05002570 btrfs_item_offset_nr(right, push_items - 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002571
2572 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
Chris Masond6025572007-03-30 14:27:56 -04002573 leaf_data_end(root, left) - push_space,
2574 btrfs_leaf_data(right) +
Chris Mason5f39d392007-10-15 16:14:19 -04002575 btrfs_item_offset_nr(right, push_items - 1),
Chris Masond6025572007-03-30 14:27:56 -04002576 push_space);
Chris Mason5f39d392007-10-15 16:14:19 -04002577 old_left_nritems = btrfs_header_nritems(left);
Yan Zheng87b29b22008-12-17 10:21:48 -05002578 BUG_ON(old_left_nritems <= 0);
Chris Masoneb60cea2007-02-02 09:18:22 -05002579
Chris Masondb945352007-10-15 16:15:53 -04002580 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
Chris Mason0783fcf2007-03-12 20:12:07 -04002581 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04002582 u32 ioff;
Chris Masondb945352007-10-15 16:15:53 -04002583
Chris Mason5f39d392007-10-15 16:14:19 -04002584 item = btrfs_item_nr(left, i);
Chris Masondb945352007-10-15 16:15:53 -04002585 if (!left->map_token) {
2586 map_extent_buffer(left, (unsigned long)item,
2587 sizeof(struct btrfs_item),
2588 &left->map_token, &left->kaddr,
2589 &left->map_start, &left->map_len,
2590 KM_USER1);
2591 }
2592
Chris Mason5f39d392007-10-15 16:14:19 -04002593 ioff = btrfs_item_offset(left, item);
2594 btrfs_set_item_offset(left, item,
Chris Masondb945352007-10-15 16:15:53 -04002595 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
Chris Masonbe0e5c02007-01-26 15:51:26 -05002596 }
Chris Mason5f39d392007-10-15 16:14:19 -04002597 btrfs_set_header_nritems(left, old_left_nritems + push_items);
Chris Masondb945352007-10-15 16:15:53 -04002598 if (left->map_token) {
2599 unmap_extent_buffer(left, left->map_token, KM_USER1);
2600 left->map_token = NULL;
2601 }
Chris Masonbe0e5c02007-01-26 15:51:26 -05002602
2603 /* fixup right node */
Chris Mason34a38212007-11-07 13:31:03 -05002604 if (push_items > right_nritems) {
Chris Masond3977122009-01-05 21:25:51 -05002605 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2606 right_nritems);
Chris Mason34a38212007-11-07 13:31:03 -05002607 WARN_ON(1);
2608 }
Chris Mason5f39d392007-10-15 16:14:19 -04002609
Chris Mason34a38212007-11-07 13:31:03 -05002610 if (push_items < right_nritems) {
2611 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2612 leaf_data_end(root, right);
2613 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2614 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2615 btrfs_leaf_data(right) +
2616 leaf_data_end(root, right), push_space);
2617
2618 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
Chris Mason5f39d392007-10-15 16:14:19 -04002619 btrfs_item_nr_offset(push_items),
2620 (btrfs_header_nritems(right) - push_items) *
2621 sizeof(struct btrfs_item));
Chris Mason34a38212007-11-07 13:31:03 -05002622 }
Yaneef1c492007-11-26 10:58:13 -05002623 right_nritems -= push_items;
2624 btrfs_set_header_nritems(right, right_nritems);
Chris Mason123abc82007-03-14 14:14:43 -04002625 push_space = BTRFS_LEAF_DATA_SIZE(root);
Chris Mason5f39d392007-10-15 16:14:19 -04002626 for (i = 0; i < right_nritems; i++) {
2627 item = btrfs_item_nr(right, i);
Chris Masondb945352007-10-15 16:15:53 -04002628
2629 if (!right->map_token) {
2630 map_extent_buffer(right, (unsigned long)item,
2631 sizeof(struct btrfs_item),
2632 &right->map_token, &right->kaddr,
2633 &right->map_start, &right->map_len,
2634 KM_USER1);
2635 }
2636
2637 push_space = push_space - btrfs_item_size(right, item);
2638 btrfs_set_item_offset(right, item, push_space);
2639 }
2640 if (right->map_token) {
2641 unmap_extent_buffer(right, right->map_token, KM_USER1);
2642 right->map_token = NULL;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002643 }
Chris Masoneb60cea2007-02-02 09:18:22 -05002644
Chris Mason5f39d392007-10-15 16:14:19 -04002645 btrfs_mark_buffer_dirty(left);
Chris Mason34a38212007-11-07 13:31:03 -05002646 if (right_nritems)
2647 btrfs_mark_buffer_dirty(right);
Chris Mason098f59c2007-05-11 11:33:21 -04002648
Zheng Yan31840ae2008-09-23 13:14:14 -04002649 ret = btrfs_update_ref(trans, root, right, left,
2650 old_left_nritems, push_items);
2651 BUG_ON(ret);
2652
Chris Mason5f39d392007-10-15 16:14:19 -04002653 btrfs_item_key(right, &disk_key, 0);
2654 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
Chris Masonaa5d6be2007-02-28 16:35:06 -05002655 if (wret)
2656 ret = wret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002657
2658 /* then fixup the leaf pointer in the path */
2659 if (path->slots[0] < push_items) {
2660 path->slots[0] += old_left_nritems;
Chris Mason925baed2008-06-25 16:01:30 -04002661 if (btrfs_header_nritems(path->nodes[0]) == 0)
2662 clean_tree_block(trans, root, path->nodes[0]);
2663 btrfs_tree_unlock(path->nodes[0]);
Chris Mason5f39d392007-10-15 16:14:19 -04002664 free_extent_buffer(path->nodes[0]);
2665 path->nodes[0] = left;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002666 path->slots[1] -= 1;
2667 } else {
Chris Mason925baed2008-06-25 16:01:30 -04002668 btrfs_tree_unlock(left);
Chris Mason5f39d392007-10-15 16:14:19 -04002669 free_extent_buffer(left);
Chris Masonbe0e5c02007-01-26 15:51:26 -05002670 path->slots[0] -= push_items;
2671 }
Chris Masoneb60cea2007-02-02 09:18:22 -05002672 BUG_ON(path->slots[0] < 0);
Chris Masonaa5d6be2007-02-28 16:35:06 -05002673 return ret;
Chris Mason925baed2008-06-25 16:01:30 -04002674out:
2675 btrfs_tree_unlock(left);
2676 free_extent_buffer(left);
2677 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002678}
2679
Chris Mason74123bd2007-02-02 11:05:29 -05002680/*
Chris Mason44871b12009-03-13 10:04:31 -04002681 * push some data in the path leaf to the left, trying to free up at
2682 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2683 */
2684static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2685 *root, struct btrfs_path *path, int data_size,
2686 int empty)
2687{
2688 struct extent_buffer *right = path->nodes[0];
2689 struct extent_buffer *left;
2690 int slot;
2691 int free_space;
2692 u32 right_nritems;
2693 int ret = 0;
2694
2695 slot = path->slots[1];
2696 if (slot == 0)
2697 return 1;
2698 if (!path->nodes[1])
2699 return 1;
2700
2701 right_nritems = btrfs_header_nritems(right);
2702 if (right_nritems == 0)
2703 return 1;
2704
2705 btrfs_assert_tree_locked(path->nodes[1]);
2706
2707 left = read_node_slot(root, path->nodes[1], slot - 1);
2708 btrfs_tree_lock(left);
2709 btrfs_set_lock_blocking(left);
2710
2711 free_space = btrfs_leaf_free_space(root, left);
2712 if (free_space < data_size) {
2713 ret = 1;
2714 goto out;
2715 }
2716
2717 /* cow and double check */
2718 ret = btrfs_cow_block(trans, root, left,
2719 path->nodes[1], slot - 1, &left);
2720 if (ret) {
2721 /* we hit -ENOSPC, but it isn't fatal here */
2722 ret = 1;
2723 goto out;
2724 }
2725
2726 free_space = btrfs_leaf_free_space(root, left);
2727 if (free_space < data_size) {
2728 ret = 1;
2729 goto out;
2730 }
2731
2732 return __push_leaf_left(trans, root, path, data_size,
2733 empty, left, free_space, right_nritems);
2734out:
2735 btrfs_tree_unlock(left);
2736 free_extent_buffer(left);
2737 return ret;
2738}
2739
2740/*
Chris Mason74123bd2007-02-02 11:05:29 -05002741 * split the path's leaf in two, making sure there is at least data_size
2742 * available for the resulting leaf level of the path.
Chris Masonaa5d6be2007-02-28 16:35:06 -05002743 *
2744 * returns 0 if all went well and < 0 on failure.
Chris Mason74123bd2007-02-02 11:05:29 -05002745 */
Chris Mason44871b12009-03-13 10:04:31 -04002746static noinline int copy_for_split(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002747 struct btrfs_root *root,
Chris Mason44871b12009-03-13 10:04:31 -04002748 struct btrfs_path *path,
2749 struct extent_buffer *l,
2750 struct extent_buffer *right,
2751 int slot, int mid, int nritems)
Chris Masonbe0e5c02007-01-26 15:51:26 -05002752{
Chris Masonbe0e5c02007-01-26 15:51:26 -05002753 int data_copy_size;
2754 int rt_data_off;
2755 int i;
Chris Masond4dbff92007-04-04 14:08:15 -04002756 int ret = 0;
Chris Masonaa5d6be2007-02-28 16:35:06 -05002757 int wret;
Chris Masond4dbff92007-04-04 14:08:15 -04002758 struct btrfs_disk_key disk_key;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002759
Chris Mason5f39d392007-10-15 16:14:19 -04002760 nritems = nritems - mid;
2761 btrfs_set_header_nritems(right, nritems);
2762 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2763
2764 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2765 btrfs_item_nr_offset(mid),
2766 nritems * sizeof(struct btrfs_item));
2767
2768 copy_extent_buffer(right, l,
Chris Masond6025572007-03-30 14:27:56 -04002769 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2770 data_copy_size, btrfs_leaf_data(l) +
2771 leaf_data_end(root, l), data_copy_size);
Chris Mason74123bd2007-02-02 11:05:29 -05002772
Chris Mason5f39d392007-10-15 16:14:19 -04002773 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2774 btrfs_item_end_nr(l, mid);
2775
2776 for (i = 0; i < nritems; i++) {
2777 struct btrfs_item *item = btrfs_item_nr(right, i);
Chris Masondb945352007-10-15 16:15:53 -04002778 u32 ioff;
2779
2780 if (!right->map_token) {
2781 map_extent_buffer(right, (unsigned long)item,
2782 sizeof(struct btrfs_item),
2783 &right->map_token, &right->kaddr,
2784 &right->map_start, &right->map_len,
2785 KM_USER1);
2786 }
2787
2788 ioff = btrfs_item_offset(right, item);
Chris Mason5f39d392007-10-15 16:14:19 -04002789 btrfs_set_item_offset(right, item, ioff + rt_data_off);
Chris Mason0783fcf2007-03-12 20:12:07 -04002790 }
Chris Mason74123bd2007-02-02 11:05:29 -05002791
Chris Masondb945352007-10-15 16:15:53 -04002792 if (right->map_token) {
2793 unmap_extent_buffer(right, right->map_token, KM_USER1);
2794 right->map_token = NULL;
2795 }
2796
Chris Mason5f39d392007-10-15 16:14:19 -04002797 btrfs_set_header_nritems(l, mid);
Chris Masonaa5d6be2007-02-28 16:35:06 -05002798 ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002799 btrfs_item_key(right, &disk_key, 0);
Chris Masondb945352007-10-15 16:15:53 -04002800 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2801 path->slots[1] + 1, 1);
Chris Masonaa5d6be2007-02-28 16:35:06 -05002802 if (wret)
2803 ret = wret;
Chris Mason5f39d392007-10-15 16:14:19 -04002804
2805 btrfs_mark_buffer_dirty(right);
2806 btrfs_mark_buffer_dirty(l);
Chris Masoneb60cea2007-02-02 09:18:22 -05002807 BUG_ON(path->slots[0] != slot);
Chris Mason5f39d392007-10-15 16:14:19 -04002808
Zheng Yan31840ae2008-09-23 13:14:14 -04002809 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2810 BUG_ON(ret);
2811
Chris Masonbe0e5c02007-01-26 15:51:26 -05002812 if (mid <= slot) {
Chris Mason925baed2008-06-25 16:01:30 -04002813 btrfs_tree_unlock(path->nodes[0]);
Chris Mason5f39d392007-10-15 16:14:19 -04002814 free_extent_buffer(path->nodes[0]);
2815 path->nodes[0] = right;
Chris Masonbe0e5c02007-01-26 15:51:26 -05002816 path->slots[0] -= mid;
2817 path->slots[1] += 1;
Chris Mason925baed2008-06-25 16:01:30 -04002818 } else {
2819 btrfs_tree_unlock(right);
Chris Mason5f39d392007-10-15 16:14:19 -04002820 free_extent_buffer(right);
Chris Mason925baed2008-06-25 16:01:30 -04002821 }
Chris Mason5f39d392007-10-15 16:14:19 -04002822
Chris Masoneb60cea2007-02-02 09:18:22 -05002823 BUG_ON(path->slots[0] < 0);
Chris Masond4dbff92007-04-04 14:08:15 -04002824
Chris Mason44871b12009-03-13 10:04:31 -04002825 return ret;
2826}
2827
2828/*
2829 * split the path's leaf in two, making sure there is at least data_size
2830 * available for the resulting leaf level of the path.
2831 *
2832 * returns 0 if all went well and < 0 on failure.
2833 */
2834static noinline int split_leaf(struct btrfs_trans_handle *trans,
2835 struct btrfs_root *root,
2836 struct btrfs_key *ins_key,
2837 struct btrfs_path *path, int data_size,
2838 int extend)
2839{
2840 struct extent_buffer *l;
2841 u32 nritems;
2842 int mid;
2843 int slot;
2844 struct extent_buffer *right;
2845 int ret = 0;
2846 int wret;
2847 int double_split;
2848 int num_doubles = 0;
2849
2850 /* first try to make some room by pushing left and right */
2851 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2852 wret = push_leaf_right(trans, root, path, data_size, 0);
2853 if (wret < 0)
2854 return wret;
2855 if (wret) {
2856 wret = push_leaf_left(trans, root, path, data_size, 0);
2857 if (wret < 0)
2858 return wret;
2859 }
2860 l = path->nodes[0];
2861
2862 /* did the pushes work? */
2863 if (btrfs_leaf_free_space(root, l) >= data_size)
2864 return 0;
2865 }
2866
2867 if (!path->nodes[1]) {
2868 ret = insert_new_root(trans, root, path, 1);
2869 if (ret)
2870 return ret;
2871 }
2872again:
2873 double_split = 0;
2874 l = path->nodes[0];
2875 slot = path->slots[0];
2876 nritems = btrfs_header_nritems(l);
2877 mid = (nritems + 1) / 2;
2878
2879 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2880 path->nodes[1]->start,
2881 root->root_key.objectid,
2882 trans->transid, 0, l->start, 0);
2883 if (IS_ERR(right)) {
2884 BUG_ON(1);
2885 return PTR_ERR(right);
2886 }
2887
2888 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2889 btrfs_set_header_bytenr(right, right->start);
2890 btrfs_set_header_generation(right, trans->transid);
2891 btrfs_set_header_owner(right, root->root_key.objectid);
2892 btrfs_set_header_level(right, 0);
2893 write_extent_buffer(right, root->fs_info->fsid,
2894 (unsigned long)btrfs_header_fsid(right),
2895 BTRFS_FSID_SIZE);
2896
2897 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2898 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2899 BTRFS_UUID_SIZE);
2900
2901 if (mid <= slot) {
2902 if (nritems == 1 ||
2903 leaf_space_used(l, mid, nritems - mid) + data_size >
2904 BTRFS_LEAF_DATA_SIZE(root)) {
2905 if (slot >= nritems) {
2906 struct btrfs_disk_key disk_key;
2907
2908 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2909 btrfs_set_header_nritems(right, 0);
2910 wret = insert_ptr(trans, root, path,
2911 &disk_key, right->start,
2912 path->slots[1] + 1, 1);
2913 if (wret)
2914 ret = wret;
2915
2916 btrfs_tree_unlock(path->nodes[0]);
2917 free_extent_buffer(path->nodes[0]);
2918 path->nodes[0] = right;
2919 path->slots[0] = 0;
2920 path->slots[1] += 1;
2921 btrfs_mark_buffer_dirty(right);
2922 return ret;
2923 }
2924 mid = slot;
2925 if (mid != nritems &&
2926 leaf_space_used(l, mid, nritems - mid) +
2927 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2928 double_split = 1;
2929 }
2930 }
2931 } else {
2932 if (leaf_space_used(l, 0, mid) + data_size >
2933 BTRFS_LEAF_DATA_SIZE(root)) {
2934 if (!extend && data_size && slot == 0) {
2935 struct btrfs_disk_key disk_key;
2936
2937 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2938 btrfs_set_header_nritems(right, 0);
2939 wret = insert_ptr(trans, root, path,
2940 &disk_key,
2941 right->start,
2942 path->slots[1], 1);
2943 if (wret)
2944 ret = wret;
2945 btrfs_tree_unlock(path->nodes[0]);
2946 free_extent_buffer(path->nodes[0]);
2947 path->nodes[0] = right;
2948 path->slots[0] = 0;
2949 if (path->slots[1] == 0) {
2950 wret = fixup_low_keys(trans, root,
2951 path, &disk_key, 1);
2952 if (wret)
2953 ret = wret;
2954 }
2955 btrfs_mark_buffer_dirty(right);
2956 return ret;
2957 } else if ((extend || !data_size) && slot == 0) {
2958 mid = 1;
2959 } else {
2960 mid = slot;
2961 if (mid != nritems &&
2962 leaf_space_used(l, mid, nritems - mid) +
2963 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2964 double_split = 1;
2965 }
2966 }
2967 }
2968 }
2969
2970 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2971 BUG_ON(ret);
2972
Chris Masoncc0c5532007-10-25 15:42:57 -04002973 if (double_split) {
2974 BUG_ON(num_doubles != 0);
2975 num_doubles++;
2976 goto again;
Chris Mason3326d1b2007-10-15 16:18:25 -04002977 }
Chris Mason44871b12009-03-13 10:04:31 -04002978
Chris Masonbe0e5c02007-01-26 15:51:26 -05002979 return ret;
2980}
2981
Chris Masond352ac62008-09-29 15:18:18 -04002982/*
Chris Mason459931e2008-12-10 09:10:46 -05002983 * This function splits a single item into two items,
2984 * giving 'new_key' to the new item and splitting the
2985 * old one at split_offset (from the start of the item).
2986 *
2987 * The path may be released by this operation. After
2988 * the split, the path is pointing to the old item. The
2989 * new item is going to be in the same node as the old one.
2990 *
2991 * Note, the item being split must be smaller enough to live alone on
2992 * a tree block with room for one extra struct btrfs_item
2993 *
2994 * This allows us to split the item in place, keeping a lock on the
2995 * leaf the entire time.
2996 */
2997int btrfs_split_item(struct btrfs_trans_handle *trans,
2998 struct btrfs_root *root,
2999 struct btrfs_path *path,
3000 struct btrfs_key *new_key,
3001 unsigned long split_offset)
3002{
3003 u32 item_size;
3004 struct extent_buffer *leaf;
3005 struct btrfs_key orig_key;
3006 struct btrfs_item *item;
3007 struct btrfs_item *new_item;
3008 int ret = 0;
3009 int slot;
3010 u32 nritems;
3011 u32 orig_offset;
3012 struct btrfs_disk_key disk_key;
3013 char *buf;
3014
3015 leaf = path->nodes[0];
3016 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
3017 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
3018 goto split;
3019
3020 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3021 btrfs_release_path(root, path);
3022
3023 path->search_for_split = 1;
3024 path->keep_locks = 1;
3025
3026 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
3027 path->search_for_split = 0;
3028
3029 /* if our item isn't there or got smaller, return now */
3030 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
3031 path->slots[0])) {
3032 path->keep_locks = 0;
3033 return -EAGAIN;
3034 }
3035
Chris Masonb9473432009-03-13 11:00:37 -04003036 btrfs_set_path_blocking(path);
Yan Zheng87b29b22008-12-17 10:21:48 -05003037 ret = split_leaf(trans, root, &orig_key, path,
3038 sizeof(struct btrfs_item), 1);
Chris Mason459931e2008-12-10 09:10:46 -05003039 path->keep_locks = 0;
3040 BUG_ON(ret);
3041
Chris Masonb9473432009-03-13 11:00:37 -04003042 btrfs_unlock_up_safe(path, 1);
3043 leaf = path->nodes[0];
3044 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3045
3046split:
Chris Masonb4ce94d2009-02-04 09:25:08 -05003047 /*
3048 * make sure any changes to the path from split_leaf leave it
3049 * in a blocking state
3050 */
3051 btrfs_set_path_blocking(path);
3052
Chris Mason459931e2008-12-10 09:10:46 -05003053 item = btrfs_item_nr(leaf, path->slots[0]);
3054 orig_offset = btrfs_item_offset(leaf, item);
3055 item_size = btrfs_item_size(leaf, item);
3056
Chris Mason459931e2008-12-10 09:10:46 -05003057 buf = kmalloc(item_size, GFP_NOFS);
3058 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3059 path->slots[0]), item_size);
3060 slot = path->slots[0] + 1;
3061 leaf = path->nodes[0];
3062
3063 nritems = btrfs_header_nritems(leaf);
3064
3065 if (slot != nritems) {
3066 /* shift the items */
3067 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3068 btrfs_item_nr_offset(slot),
3069 (nritems - slot) * sizeof(struct btrfs_item));
3070
3071 }
3072
3073 btrfs_cpu_key_to_disk(&disk_key, new_key);
3074 btrfs_set_item_key(leaf, &disk_key, slot);
3075
3076 new_item = btrfs_item_nr(leaf, slot);
3077
3078 btrfs_set_item_offset(leaf, new_item, orig_offset);
3079 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3080
3081 btrfs_set_item_offset(leaf, item,
3082 orig_offset + item_size - split_offset);
3083 btrfs_set_item_size(leaf, item, split_offset);
3084
3085 btrfs_set_header_nritems(leaf, nritems + 1);
3086
3087 /* write the data for the start of the original item */
3088 write_extent_buffer(leaf, buf,
3089 btrfs_item_ptr_offset(leaf, path->slots[0]),
3090 split_offset);
3091
3092 /* write the data for the new item */
3093 write_extent_buffer(leaf, buf + split_offset,
3094 btrfs_item_ptr_offset(leaf, slot),
3095 item_size - split_offset);
3096 btrfs_mark_buffer_dirty(leaf);
3097
3098 ret = 0;
3099 if (btrfs_leaf_free_space(root, leaf) < 0) {
3100 btrfs_print_leaf(root, leaf);
3101 BUG();
3102 }
3103 kfree(buf);
3104 return ret;
3105}
3106
3107/*
Chris Masond352ac62008-09-29 15:18:18 -04003108 * make the item pointed to by the path smaller. new_size indicates
3109 * how small to make it, and from_end tells us if we just chop bytes
3110 * off the end of the item or if we shift the item to chop bytes off
3111 * the front.
3112 */
Chris Masonb18c6682007-04-17 13:26:50 -04003113int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3114 struct btrfs_root *root,
3115 struct btrfs_path *path,
Chris Mason179e29e2007-11-01 11:28:41 -04003116 u32 new_size, int from_end)
Chris Masonb18c6682007-04-17 13:26:50 -04003117{
3118 int ret = 0;
3119 int slot;
3120 int slot_orig;
Chris Mason5f39d392007-10-15 16:14:19 -04003121 struct extent_buffer *leaf;
3122 struct btrfs_item *item;
Chris Masonb18c6682007-04-17 13:26:50 -04003123 u32 nritems;
3124 unsigned int data_end;
3125 unsigned int old_data_start;
3126 unsigned int old_size;
3127 unsigned int size_diff;
3128 int i;
3129
3130 slot_orig = path->slots[0];
Chris Mason5f39d392007-10-15 16:14:19 -04003131 leaf = path->nodes[0];
Chris Mason179e29e2007-11-01 11:28:41 -04003132 slot = path->slots[0];
3133
3134 old_size = btrfs_item_size_nr(leaf, slot);
3135 if (old_size == new_size)
3136 return 0;
Chris Masonb18c6682007-04-17 13:26:50 -04003137
Chris Mason5f39d392007-10-15 16:14:19 -04003138 nritems = btrfs_header_nritems(leaf);
Chris Masonb18c6682007-04-17 13:26:50 -04003139 data_end = leaf_data_end(root, leaf);
3140
Chris Mason5f39d392007-10-15 16:14:19 -04003141 old_data_start = btrfs_item_offset_nr(leaf, slot);
Chris Mason179e29e2007-11-01 11:28:41 -04003142
Chris Masonb18c6682007-04-17 13:26:50 -04003143 size_diff = old_size - new_size;
3144
3145 BUG_ON(slot < 0);
3146 BUG_ON(slot >= nritems);
3147
3148 /*
3149 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3150 */
3151 /* first correct the data pointers */
3152 for (i = slot; i < nritems; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04003153 u32 ioff;
3154 item = btrfs_item_nr(leaf, i);
Chris Masondb945352007-10-15 16:15:53 -04003155
3156 if (!leaf->map_token) {
3157 map_extent_buffer(leaf, (unsigned long)item,
3158 sizeof(struct btrfs_item),
3159 &leaf->map_token, &leaf->kaddr,
3160 &leaf->map_start, &leaf->map_len,
3161 KM_USER1);
3162 }
3163
Chris Mason5f39d392007-10-15 16:14:19 -04003164 ioff = btrfs_item_offset(leaf, item);
3165 btrfs_set_item_offset(leaf, item, ioff + size_diff);
Chris Masonb18c6682007-04-17 13:26:50 -04003166 }
Chris Masondb945352007-10-15 16:15:53 -04003167
3168 if (leaf->map_token) {
3169 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3170 leaf->map_token = NULL;
3171 }
3172
Chris Masonb18c6682007-04-17 13:26:50 -04003173 /* shift the data */
Chris Mason179e29e2007-11-01 11:28:41 -04003174 if (from_end) {
3175 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3176 data_end + size_diff, btrfs_leaf_data(leaf) +
3177 data_end, old_data_start + new_size - data_end);
3178 } else {
3179 struct btrfs_disk_key disk_key;
3180 u64 offset;
3181
3182 btrfs_item_key(leaf, &disk_key, slot);
3183
3184 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3185 unsigned long ptr;
3186 struct btrfs_file_extent_item *fi;
3187
3188 fi = btrfs_item_ptr(leaf, slot,
3189 struct btrfs_file_extent_item);
3190 fi = (struct btrfs_file_extent_item *)(
3191 (unsigned long)fi - size_diff);
3192
3193 if (btrfs_file_extent_type(leaf, fi) ==
3194 BTRFS_FILE_EXTENT_INLINE) {
3195 ptr = btrfs_item_ptr_offset(leaf, slot);
3196 memmove_extent_buffer(leaf, ptr,
Chris Masond3977122009-01-05 21:25:51 -05003197 (unsigned long)fi,
3198 offsetof(struct btrfs_file_extent_item,
Chris Mason179e29e2007-11-01 11:28:41 -04003199 disk_bytenr));
3200 }
3201 }
3202
3203 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3204 data_end + size_diff, btrfs_leaf_data(leaf) +
3205 data_end, old_data_start - data_end);
3206
3207 offset = btrfs_disk_key_offset(&disk_key);
3208 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3209 btrfs_set_item_key(leaf, &disk_key, slot);
3210 if (slot == 0)
3211 fixup_low_keys(trans, root, path, &disk_key, 1);
3212 }
Chris Mason5f39d392007-10-15 16:14:19 -04003213
3214 item = btrfs_item_nr(leaf, slot);
3215 btrfs_set_item_size(leaf, item, new_size);
3216 btrfs_mark_buffer_dirty(leaf);
Chris Masonb18c6682007-04-17 13:26:50 -04003217
3218 ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003219 if (btrfs_leaf_free_space(root, leaf) < 0) {
3220 btrfs_print_leaf(root, leaf);
Chris Masonb18c6682007-04-17 13:26:50 -04003221 BUG();
Chris Mason5f39d392007-10-15 16:14:19 -04003222 }
Chris Masonb18c6682007-04-17 13:26:50 -04003223 return ret;
3224}
3225
Chris Masond352ac62008-09-29 15:18:18 -04003226/*
3227 * make the item pointed to by the path bigger, data_size is the new size.
3228 */
Chris Mason5f39d392007-10-15 16:14:19 -04003229int btrfs_extend_item(struct btrfs_trans_handle *trans,
3230 struct btrfs_root *root, struct btrfs_path *path,
3231 u32 data_size)
Chris Mason6567e832007-04-16 09:22:45 -04003232{
3233 int ret = 0;
3234 int slot;
3235 int slot_orig;
Chris Mason5f39d392007-10-15 16:14:19 -04003236 struct extent_buffer *leaf;
3237 struct btrfs_item *item;
Chris Mason6567e832007-04-16 09:22:45 -04003238 u32 nritems;
3239 unsigned int data_end;
3240 unsigned int old_data;
3241 unsigned int old_size;
3242 int i;
3243
3244 slot_orig = path->slots[0];
Chris Mason5f39d392007-10-15 16:14:19 -04003245 leaf = path->nodes[0];
Chris Mason6567e832007-04-16 09:22:45 -04003246
Chris Mason5f39d392007-10-15 16:14:19 -04003247 nritems = btrfs_header_nritems(leaf);
Chris Mason6567e832007-04-16 09:22:45 -04003248 data_end = leaf_data_end(root, leaf);
3249
Chris Mason5f39d392007-10-15 16:14:19 -04003250 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3251 btrfs_print_leaf(root, leaf);
Chris Mason6567e832007-04-16 09:22:45 -04003252 BUG();
Chris Mason5f39d392007-10-15 16:14:19 -04003253 }
Chris Mason6567e832007-04-16 09:22:45 -04003254 slot = path->slots[0];
Chris Mason5f39d392007-10-15 16:14:19 -04003255 old_data = btrfs_item_end_nr(leaf, slot);
Chris Mason6567e832007-04-16 09:22:45 -04003256
3257 BUG_ON(slot < 0);
Chris Mason3326d1b2007-10-15 16:18:25 -04003258 if (slot >= nritems) {
3259 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -05003260 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3261 slot, nritems);
Chris Mason3326d1b2007-10-15 16:18:25 -04003262 BUG_ON(1);
3263 }
Chris Mason6567e832007-04-16 09:22:45 -04003264
3265 /*
3266 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3267 */
3268 /* first correct the data pointers */
3269 for (i = slot; i < nritems; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04003270 u32 ioff;
3271 item = btrfs_item_nr(leaf, i);
Chris Masondb945352007-10-15 16:15:53 -04003272
3273 if (!leaf->map_token) {
3274 map_extent_buffer(leaf, (unsigned long)item,
3275 sizeof(struct btrfs_item),
3276 &leaf->map_token, &leaf->kaddr,
3277 &leaf->map_start, &leaf->map_len,
3278 KM_USER1);
3279 }
Chris Mason5f39d392007-10-15 16:14:19 -04003280 ioff = btrfs_item_offset(leaf, item);
3281 btrfs_set_item_offset(leaf, item, ioff - data_size);
Chris Mason6567e832007-04-16 09:22:45 -04003282 }
Chris Mason5f39d392007-10-15 16:14:19 -04003283
Chris Masondb945352007-10-15 16:15:53 -04003284 if (leaf->map_token) {
3285 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3286 leaf->map_token = NULL;
3287 }
3288
Chris Mason6567e832007-04-16 09:22:45 -04003289 /* shift the data */
Chris Mason5f39d392007-10-15 16:14:19 -04003290 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
Chris Mason6567e832007-04-16 09:22:45 -04003291 data_end - data_size, btrfs_leaf_data(leaf) +
3292 data_end, old_data - data_end);
Chris Mason5f39d392007-10-15 16:14:19 -04003293
Chris Mason6567e832007-04-16 09:22:45 -04003294 data_end = old_data;
Chris Mason5f39d392007-10-15 16:14:19 -04003295 old_size = btrfs_item_size_nr(leaf, slot);
3296 item = btrfs_item_nr(leaf, slot);
3297 btrfs_set_item_size(leaf, item, old_size + data_size);
3298 btrfs_mark_buffer_dirty(leaf);
Chris Mason6567e832007-04-16 09:22:45 -04003299
3300 ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003301 if (btrfs_leaf_free_space(root, leaf) < 0) {
3302 btrfs_print_leaf(root, leaf);
Chris Mason6567e832007-04-16 09:22:45 -04003303 BUG();
Chris Mason5f39d392007-10-15 16:14:19 -04003304 }
Chris Mason6567e832007-04-16 09:22:45 -04003305 return ret;
3306}
3307
Chris Mason74123bd2007-02-02 11:05:29 -05003308/*
Chris Masond352ac62008-09-29 15:18:18 -04003309 * Given a key and some data, insert items into the tree.
Chris Mason74123bd2007-02-02 11:05:29 -05003310 * This does all the path init required, making room in the tree if needed.
Josef Bacikf3465ca2008-11-12 14:19:50 -05003311 * Returns the number of keys that were inserted.
3312 */
3313int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3314 struct btrfs_root *root,
3315 struct btrfs_path *path,
3316 struct btrfs_key *cpu_key, u32 *data_size,
3317 int nr)
3318{
3319 struct extent_buffer *leaf;
3320 struct btrfs_item *item;
3321 int ret = 0;
3322 int slot;
Josef Bacikf3465ca2008-11-12 14:19:50 -05003323 int i;
3324 u32 nritems;
3325 u32 total_data = 0;
3326 u32 total_size = 0;
3327 unsigned int data_end;
3328 struct btrfs_disk_key disk_key;
3329 struct btrfs_key found_key;
3330
Yan Zheng87b29b22008-12-17 10:21:48 -05003331 for (i = 0; i < nr; i++) {
3332 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3333 BTRFS_LEAF_DATA_SIZE(root)) {
3334 break;
3335 nr = i;
3336 }
Josef Bacikf3465ca2008-11-12 14:19:50 -05003337 total_data += data_size[i];
Yan Zheng87b29b22008-12-17 10:21:48 -05003338 total_size += data_size[i] + sizeof(struct btrfs_item);
3339 }
3340 BUG_ON(nr == 0);
Josef Bacikf3465ca2008-11-12 14:19:50 -05003341
Josef Bacikf3465ca2008-11-12 14:19:50 -05003342 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3343 if (ret == 0)
3344 return -EEXIST;
3345 if (ret < 0)
3346 goto out;
3347
Josef Bacikf3465ca2008-11-12 14:19:50 -05003348 leaf = path->nodes[0];
3349
3350 nritems = btrfs_header_nritems(leaf);
3351 data_end = leaf_data_end(root, leaf);
3352
3353 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3354 for (i = nr; i >= 0; i--) {
3355 total_data -= data_size[i];
3356 total_size -= data_size[i] + sizeof(struct btrfs_item);
3357 if (total_size < btrfs_leaf_free_space(root, leaf))
3358 break;
3359 }
3360 nr = i;
3361 }
3362
3363 slot = path->slots[0];
3364 BUG_ON(slot < 0);
3365
3366 if (slot != nritems) {
3367 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3368
3369 item = btrfs_item_nr(leaf, slot);
3370 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3371
3372 /* figure out how many keys we can insert in here */
3373 total_data = data_size[0];
3374 for (i = 1; i < nr; i++) {
3375 if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3376 break;
3377 total_data += data_size[i];
3378 }
3379 nr = i;
3380
3381 if (old_data < data_end) {
3382 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -05003383 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
Josef Bacikf3465ca2008-11-12 14:19:50 -05003384 slot, old_data, data_end);
3385 BUG_ON(1);
3386 }
3387 /*
3388 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3389 */
3390 /* first correct the data pointers */
3391 WARN_ON(leaf->map_token);
3392 for (i = slot; i < nritems; i++) {
3393 u32 ioff;
3394
3395 item = btrfs_item_nr(leaf, i);
3396 if (!leaf->map_token) {
3397 map_extent_buffer(leaf, (unsigned long)item,
3398 sizeof(struct btrfs_item),
3399 &leaf->map_token, &leaf->kaddr,
3400 &leaf->map_start, &leaf->map_len,
3401 KM_USER1);
3402 }
3403
3404 ioff = btrfs_item_offset(leaf, item);
3405 btrfs_set_item_offset(leaf, item, ioff - total_data);
3406 }
3407 if (leaf->map_token) {
3408 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3409 leaf->map_token = NULL;
3410 }
3411
3412 /* shift the items */
3413 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3414 btrfs_item_nr_offset(slot),
3415 (nritems - slot) * sizeof(struct btrfs_item));
3416
3417 /* shift the data */
3418 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3419 data_end - total_data, btrfs_leaf_data(leaf) +
3420 data_end, old_data - data_end);
3421 data_end = old_data;
3422 } else {
3423 /*
3424 * this sucks but it has to be done, if we are inserting at
3425 * the end of the leaf only insert 1 of the items, since we
3426 * have no way of knowing whats on the next leaf and we'd have
3427 * to drop our current locks to figure it out
3428 */
3429 nr = 1;
3430 }
3431
3432 /* setup the item for the new data */
3433 for (i = 0; i < nr; i++) {
3434 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3435 btrfs_set_item_key(leaf, &disk_key, slot + i);
3436 item = btrfs_item_nr(leaf, slot + i);
3437 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3438 data_end -= data_size[i];
3439 btrfs_set_item_size(leaf, item, data_size[i]);
3440 }
3441 btrfs_set_header_nritems(leaf, nritems + nr);
3442 btrfs_mark_buffer_dirty(leaf);
3443
3444 ret = 0;
3445 if (slot == 0) {
3446 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3447 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3448 }
3449
3450 if (btrfs_leaf_free_space(root, leaf) < 0) {
3451 btrfs_print_leaf(root, leaf);
3452 BUG();
3453 }
3454out:
3455 if (!ret)
3456 ret = nr;
3457 return ret;
3458}
3459
3460/*
Chris Mason44871b12009-03-13 10:04:31 -04003461 * this is a helper for btrfs_insert_empty_items, the main goal here is
3462 * to save stack depth by doing the bulk of the work in a function
3463 * that doesn't call btrfs_search_slot
Chris Mason74123bd2007-02-02 11:05:29 -05003464 */
Chris Mason44871b12009-03-13 10:04:31 -04003465static noinline_for_stack int
3466setup_items_for_insert(struct btrfs_trans_handle *trans,
3467 struct btrfs_root *root, struct btrfs_path *path,
3468 struct btrfs_key *cpu_key, u32 *data_size,
3469 u32 total_data, u32 total_size, int nr)
Chris Masonbe0e5c02007-01-26 15:51:26 -05003470{
Chris Mason5f39d392007-10-15 16:14:19 -04003471 struct btrfs_item *item;
Chris Mason9c583092008-01-29 15:15:18 -05003472 int i;
Chris Mason7518a232007-03-12 12:01:18 -04003473 u32 nritems;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003474 unsigned int data_end;
Chris Masone2fa7222007-03-12 16:22:34 -04003475 struct btrfs_disk_key disk_key;
Chris Mason44871b12009-03-13 10:04:31 -04003476 int ret;
3477 struct extent_buffer *leaf;
3478 int slot;
Chris Masone2fa7222007-03-12 16:22:34 -04003479
Chris Mason5f39d392007-10-15 16:14:19 -04003480 leaf = path->nodes[0];
Chris Mason44871b12009-03-13 10:04:31 -04003481 slot = path->slots[0];
Chris Mason74123bd2007-02-02 11:05:29 -05003482
Chris Mason5f39d392007-10-15 16:14:19 -04003483 nritems = btrfs_header_nritems(leaf);
Chris Mason123abc82007-03-14 14:14:43 -04003484 data_end = leaf_data_end(root, leaf);
Chris Masoneb60cea2007-02-02 09:18:22 -05003485
Chris Masonf25956c2008-09-12 15:32:53 -04003486 if (btrfs_leaf_free_space(root, leaf) < total_size) {
Chris Mason3326d1b2007-10-15 16:18:25 -04003487 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -05003488 printk(KERN_CRIT "not enough freespace need %u have %d\n",
Chris Mason9c583092008-01-29 15:15:18 -05003489 total_size, btrfs_leaf_free_space(root, leaf));
Chris Masonbe0e5c02007-01-26 15:51:26 -05003490 BUG();
Chris Masond4dbff92007-04-04 14:08:15 -04003491 }
Chris Mason5f39d392007-10-15 16:14:19 -04003492
Chris Masonbe0e5c02007-01-26 15:51:26 -05003493 if (slot != nritems) {
Chris Mason5f39d392007-10-15 16:14:19 -04003494 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
Chris Masonbe0e5c02007-01-26 15:51:26 -05003495
Chris Mason5f39d392007-10-15 16:14:19 -04003496 if (old_data < data_end) {
3497 btrfs_print_leaf(root, leaf);
Chris Masond3977122009-01-05 21:25:51 -05003498 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
Chris Mason5f39d392007-10-15 16:14:19 -04003499 slot, old_data, data_end);
3500 BUG_ON(1);
3501 }
Chris Masonbe0e5c02007-01-26 15:51:26 -05003502 /*
3503 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3504 */
3505 /* first correct the data pointers */
Chris Masondb945352007-10-15 16:15:53 -04003506 WARN_ON(leaf->map_token);
Chris Mason0783fcf2007-03-12 20:12:07 -04003507 for (i = slot; i < nritems; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04003508 u32 ioff;
Chris Masondb945352007-10-15 16:15:53 -04003509
Chris Mason5f39d392007-10-15 16:14:19 -04003510 item = btrfs_item_nr(leaf, i);
Chris Masondb945352007-10-15 16:15:53 -04003511 if (!leaf->map_token) {
3512 map_extent_buffer(leaf, (unsigned long)item,
3513 sizeof(struct btrfs_item),
3514 &leaf->map_token, &leaf->kaddr,
3515 &leaf->map_start, &leaf->map_len,
3516 KM_USER1);
3517 }
3518
Chris Mason5f39d392007-10-15 16:14:19 -04003519 ioff = btrfs_item_offset(leaf, item);
Chris Mason9c583092008-01-29 15:15:18 -05003520 btrfs_set_item_offset(leaf, item, ioff - total_data);
Chris Mason0783fcf2007-03-12 20:12:07 -04003521 }
Chris Masondb945352007-10-15 16:15:53 -04003522 if (leaf->map_token) {
3523 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3524 leaf->map_token = NULL;
3525 }
Chris Masonbe0e5c02007-01-26 15:51:26 -05003526
3527 /* shift the items */
Chris Mason9c583092008-01-29 15:15:18 -05003528 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
Chris Mason5f39d392007-10-15 16:14:19 -04003529 btrfs_item_nr_offset(slot),
Chris Masond6025572007-03-30 14:27:56 -04003530 (nritems - slot) * sizeof(struct btrfs_item));
Chris Masonbe0e5c02007-01-26 15:51:26 -05003531
3532 /* shift the data */
Chris Mason5f39d392007-10-15 16:14:19 -04003533 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
Chris Mason9c583092008-01-29 15:15:18 -05003534 data_end - total_data, btrfs_leaf_data(leaf) +
Chris Masond6025572007-03-30 14:27:56 -04003535 data_end, old_data - data_end);
Chris Masonbe0e5c02007-01-26 15:51:26 -05003536 data_end = old_data;
3537 }
Chris Mason5f39d392007-10-15 16:14:19 -04003538
Chris Mason62e27492007-03-15 12:56:47 -04003539 /* setup the item for the new data */
Chris Mason9c583092008-01-29 15:15:18 -05003540 for (i = 0; i < nr; i++) {
3541 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3542 btrfs_set_item_key(leaf, &disk_key, slot + i);
3543 item = btrfs_item_nr(leaf, slot + i);
3544 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3545 data_end -= data_size[i];
3546 btrfs_set_item_size(leaf, item, data_size[i]);
3547 }
Chris Mason44871b12009-03-13 10:04:31 -04003548
Chris Mason9c583092008-01-29 15:15:18 -05003549 btrfs_set_header_nritems(leaf, nritems + nr);
Chris Masonaa5d6be2007-02-28 16:35:06 -05003550
3551 ret = 0;
Chris Mason5a01a2e2008-01-30 11:43:54 -05003552 if (slot == 0) {
Chris Mason44871b12009-03-13 10:04:31 -04003553 struct btrfs_disk_key disk_key;
Chris Mason5a01a2e2008-01-30 11:43:54 -05003554 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
Chris Masone089f052007-03-16 16:20:31 -04003555 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
Chris Mason5a01a2e2008-01-30 11:43:54 -05003556 }
Chris Masonb9473432009-03-13 11:00:37 -04003557 btrfs_unlock_up_safe(path, 1);
3558 btrfs_mark_buffer_dirty(leaf);
Chris Masonaa5d6be2007-02-28 16:35:06 -05003559
Chris Mason5f39d392007-10-15 16:14:19 -04003560 if (btrfs_leaf_free_space(root, leaf) < 0) {
3561 btrfs_print_leaf(root, leaf);
Chris Masonbe0e5c02007-01-26 15:51:26 -05003562 BUG();
Chris Mason5f39d392007-10-15 16:14:19 -04003563 }
Chris Mason44871b12009-03-13 10:04:31 -04003564 return ret;
3565}
3566
3567/*
3568 * Given a key and some data, insert items into the tree.
3569 * This does all the path init required, making room in the tree if needed.
3570 */
3571int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3572 struct btrfs_root *root,
3573 struct btrfs_path *path,
3574 struct btrfs_key *cpu_key, u32 *data_size,
3575 int nr)
3576{
3577 struct extent_buffer *leaf;
3578 int ret = 0;
3579 int slot;
3580 int i;
3581 u32 total_size = 0;
3582 u32 total_data = 0;
3583
3584 for (i = 0; i < nr; i++)
3585 total_data += data_size[i];
3586
3587 total_size = total_data + (nr * sizeof(struct btrfs_item));
3588 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3589 if (ret == 0)
3590 return -EEXIST;
3591 if (ret < 0)
3592 goto out;
3593
3594 leaf = path->nodes[0];
3595 slot = path->slots[0];
3596 BUG_ON(slot < 0);
3597
3598 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3599 total_data, total_size, nr);
3600
Chris Masoned2ff2c2007-03-01 18:59:40 -05003601out:
Chris Mason62e27492007-03-15 12:56:47 -04003602 return ret;
3603}
3604
3605/*
3606 * Given a key and some data, insert an item into the tree.
3607 * This does all the path init required, making room in the tree if needed.
3608 */
Chris Masone089f052007-03-16 16:20:31 -04003609int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3610 *root, struct btrfs_key *cpu_key, void *data, u32
3611 data_size)
Chris Mason62e27492007-03-15 12:56:47 -04003612{
3613 int ret = 0;
Chris Mason2c90e5d2007-04-02 10:50:19 -04003614 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003615 struct extent_buffer *leaf;
3616 unsigned long ptr;
Chris Mason62e27492007-03-15 12:56:47 -04003617
Chris Mason2c90e5d2007-04-02 10:50:19 -04003618 path = btrfs_alloc_path();
3619 BUG_ON(!path);
Chris Mason2c90e5d2007-04-02 10:50:19 -04003620 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
Chris Mason62e27492007-03-15 12:56:47 -04003621 if (!ret) {
Chris Mason5f39d392007-10-15 16:14:19 -04003622 leaf = path->nodes[0];
3623 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3624 write_extent_buffer(leaf, data, ptr, data_size);
3625 btrfs_mark_buffer_dirty(leaf);
Chris Mason62e27492007-03-15 12:56:47 -04003626 }
Chris Mason2c90e5d2007-04-02 10:50:19 -04003627 btrfs_free_path(path);
Chris Masonaa5d6be2007-02-28 16:35:06 -05003628 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003629}
3630
Chris Mason74123bd2007-02-02 11:05:29 -05003631/*
Chris Mason5de08d72007-02-24 06:24:44 -05003632 * delete the pointer from a given node.
Chris Mason74123bd2007-02-02 11:05:29 -05003633 *
Chris Masond352ac62008-09-29 15:18:18 -04003634 * the tree should have been previously balanced so the deletion does not
3635 * empty a node.
Chris Mason74123bd2007-02-02 11:05:29 -05003636 */
Chris Masone089f052007-03-16 16:20:31 -04003637static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3638 struct btrfs_path *path, int level, int slot)
Chris Masonbe0e5c02007-01-26 15:51:26 -05003639{
Chris Mason5f39d392007-10-15 16:14:19 -04003640 struct extent_buffer *parent = path->nodes[level];
Chris Mason7518a232007-03-12 12:01:18 -04003641 u32 nritems;
Chris Masonaa5d6be2007-02-28 16:35:06 -05003642 int ret = 0;
Chris Masonbb803952007-03-01 12:04:21 -05003643 int wret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003644
Chris Mason5f39d392007-10-15 16:14:19 -04003645 nritems = btrfs_header_nritems(parent);
Chris Masond3977122009-01-05 21:25:51 -05003646 if (slot != nritems - 1) {
Chris Mason5f39d392007-10-15 16:14:19 -04003647 memmove_extent_buffer(parent,
3648 btrfs_node_key_ptr_offset(slot),
3649 btrfs_node_key_ptr_offset(slot + 1),
Chris Masond6025572007-03-30 14:27:56 -04003650 sizeof(struct btrfs_key_ptr) *
3651 (nritems - slot - 1));
Chris Masonbb803952007-03-01 12:04:21 -05003652 }
Chris Mason7518a232007-03-12 12:01:18 -04003653 nritems--;
Chris Mason5f39d392007-10-15 16:14:19 -04003654 btrfs_set_header_nritems(parent, nritems);
Chris Mason7518a232007-03-12 12:01:18 -04003655 if (nritems == 0 && parent == root->node) {
Chris Mason5f39d392007-10-15 16:14:19 -04003656 BUG_ON(btrfs_header_level(root->node) != 1);
Chris Masonbb803952007-03-01 12:04:21 -05003657 /* just turn the root into a leaf and break */
Chris Mason5f39d392007-10-15 16:14:19 -04003658 btrfs_set_header_level(root->node, 0);
Chris Masonbb803952007-03-01 12:04:21 -05003659 } else if (slot == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04003660 struct btrfs_disk_key disk_key;
3661
3662 btrfs_node_key(parent, &disk_key, 0);
3663 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
Chris Mason0f70abe2007-02-28 16:46:22 -05003664 if (wret)
3665 ret = wret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003666 }
Chris Masond6025572007-03-30 14:27:56 -04003667 btrfs_mark_buffer_dirty(parent);
Chris Masonaa5d6be2007-02-28 16:35:06 -05003668 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003669}
3670
Chris Mason74123bd2007-02-02 11:05:29 -05003671/*
Chris Mason323ac952008-10-01 19:05:46 -04003672 * a helper function to delete the leaf pointed to by path->slots[1] and
3673 * path->nodes[1]. bytenr is the node block pointer, but since the callers
3674 * already know it, it is faster to have them pass it down than to
3675 * read it out of the node again.
3676 *
3677 * This deletes the pointer in path->nodes[1] and frees the leaf
3678 * block extent. zero is returned if it all worked out, < 0 otherwise.
3679 *
3680 * The path must have already been setup for deleting the leaf, including
3681 * all the proper balancing. path->nodes[1] must be locked.
3682 */
3683noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3684 struct btrfs_root *root,
3685 struct btrfs_path *path, u64 bytenr)
3686{
3687 int ret;
3688 u64 root_gen = btrfs_header_generation(path->nodes[1]);
Chris Mason4d081c42009-02-04 09:31:28 -05003689 u64 parent_start = path->nodes[1]->start;
3690 u64 parent_owner = btrfs_header_owner(path->nodes[1]);
Chris Mason323ac952008-10-01 19:05:46 -04003691
3692 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3693 if (ret)
3694 return ret;
3695
Chris Mason4d081c42009-02-04 09:31:28 -05003696 /*
3697 * btrfs_free_extent is expensive, we want to make sure we
3698 * aren't holding any locks when we call it
3699 */
3700 btrfs_unlock_up_safe(path, 0);
3701
Chris Mason323ac952008-10-01 19:05:46 -04003702 ret = btrfs_free_extent(trans, root, bytenr,
3703 btrfs_level_size(root, 0),
Chris Mason4d081c42009-02-04 09:31:28 -05003704 parent_start, parent_owner,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04003705 root_gen, 0, 1);
Chris Mason323ac952008-10-01 19:05:46 -04003706 return ret;
3707}
3708/*
Chris Mason74123bd2007-02-02 11:05:29 -05003709 * delete the item at the leaf level in path. If that empties
3710 * the leaf, remove it from the tree
3711 */
Chris Mason85e21ba2008-01-29 15:11:36 -05003712int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3713 struct btrfs_path *path, int slot, int nr)
Chris Masonbe0e5c02007-01-26 15:51:26 -05003714{
Chris Mason5f39d392007-10-15 16:14:19 -04003715 struct extent_buffer *leaf;
3716 struct btrfs_item *item;
Chris Mason85e21ba2008-01-29 15:11:36 -05003717 int last_off;
3718 int dsize = 0;
Chris Masonaa5d6be2007-02-28 16:35:06 -05003719 int ret = 0;
3720 int wret;
Chris Mason85e21ba2008-01-29 15:11:36 -05003721 int i;
Chris Mason7518a232007-03-12 12:01:18 -04003722 u32 nritems;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003723
Chris Mason5f39d392007-10-15 16:14:19 -04003724 leaf = path->nodes[0];
Chris Mason85e21ba2008-01-29 15:11:36 -05003725 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3726
3727 for (i = 0; i < nr; i++)
3728 dsize += btrfs_item_size_nr(leaf, slot + i);
3729
Chris Mason5f39d392007-10-15 16:14:19 -04003730 nritems = btrfs_header_nritems(leaf);
Chris Masonbe0e5c02007-01-26 15:51:26 -05003731
Chris Mason85e21ba2008-01-29 15:11:36 -05003732 if (slot + nr != nritems) {
Chris Mason123abc82007-03-14 14:14:43 -04003733 int data_end = leaf_data_end(root, leaf);
Chris Mason5f39d392007-10-15 16:14:19 -04003734
3735 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
Chris Masond6025572007-03-30 14:27:56 -04003736 data_end + dsize,
3737 btrfs_leaf_data(leaf) + data_end,
Chris Mason85e21ba2008-01-29 15:11:36 -05003738 last_off - data_end);
Chris Mason5f39d392007-10-15 16:14:19 -04003739
Chris Mason85e21ba2008-01-29 15:11:36 -05003740 for (i = slot + nr; i < nritems; i++) {
Chris Mason5f39d392007-10-15 16:14:19 -04003741 u32 ioff;
Chris Masondb945352007-10-15 16:15:53 -04003742
Chris Mason5f39d392007-10-15 16:14:19 -04003743 item = btrfs_item_nr(leaf, i);
Chris Masondb945352007-10-15 16:15:53 -04003744 if (!leaf->map_token) {
3745 map_extent_buffer(leaf, (unsigned long)item,
3746 sizeof(struct btrfs_item),
3747 &leaf->map_token, &leaf->kaddr,
3748 &leaf->map_start, &leaf->map_len,
3749 KM_USER1);
3750 }
Chris Mason5f39d392007-10-15 16:14:19 -04003751 ioff = btrfs_item_offset(leaf, item);
3752 btrfs_set_item_offset(leaf, item, ioff + dsize);
Chris Mason0783fcf2007-03-12 20:12:07 -04003753 }
Chris Masondb945352007-10-15 16:15:53 -04003754
3755 if (leaf->map_token) {
3756 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3757 leaf->map_token = NULL;
3758 }
3759
Chris Mason5f39d392007-10-15 16:14:19 -04003760 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
Chris Mason85e21ba2008-01-29 15:11:36 -05003761 btrfs_item_nr_offset(slot + nr),
Chris Masond6025572007-03-30 14:27:56 -04003762 sizeof(struct btrfs_item) *
Chris Mason85e21ba2008-01-29 15:11:36 -05003763 (nritems - slot - nr));
Chris Masonbe0e5c02007-01-26 15:51:26 -05003764 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003765 btrfs_set_header_nritems(leaf, nritems - nr);
3766 nritems -= nr;
Chris Mason5f39d392007-10-15 16:14:19 -04003767
Chris Mason74123bd2007-02-02 11:05:29 -05003768 /* delete the leaf if we've emptied it */
Chris Mason7518a232007-03-12 12:01:18 -04003769 if (nritems == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04003770 if (leaf == root->node) {
3771 btrfs_set_header_level(leaf, 0);
Chris Mason9a8dd152007-02-23 08:38:36 -05003772 } else {
Chris Mason323ac952008-10-01 19:05:46 -04003773 ret = btrfs_del_leaf(trans, root, path, leaf->start);
3774 BUG_ON(ret);
Chris Mason9a8dd152007-02-23 08:38:36 -05003775 }
Chris Masonbe0e5c02007-01-26 15:51:26 -05003776 } else {
Chris Mason7518a232007-03-12 12:01:18 -04003777 int used = leaf_space_used(leaf, 0, nritems);
Chris Masonaa5d6be2007-02-28 16:35:06 -05003778 if (slot == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04003779 struct btrfs_disk_key disk_key;
3780
3781 btrfs_item_key(leaf, &disk_key, 0);
Chris Masone089f052007-03-16 16:20:31 -04003782 wret = fixup_low_keys(trans, root, path,
Chris Mason5f39d392007-10-15 16:14:19 -04003783 &disk_key, 1);
Chris Masonaa5d6be2007-02-28 16:35:06 -05003784 if (wret)
3785 ret = wret;
3786 }
Chris Masonaa5d6be2007-02-28 16:35:06 -05003787
Chris Mason74123bd2007-02-02 11:05:29 -05003788 /* delete the leaf if it is mostly empty */
Chris Mason85e21ba2008-01-29 15:11:36 -05003789 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
Chris Masonbe0e5c02007-01-26 15:51:26 -05003790 /* push_leaf_left fixes the path.
3791 * make sure the path still points to our leaf
3792 * for possible call to del_ptr below
3793 */
Chris Mason4920c9a2007-01-26 16:38:42 -05003794 slot = path->slots[1];
Chris Mason5f39d392007-10-15 16:14:19 -04003795 extent_buffer_get(leaf);
3796
Chris Masonb9473432009-03-13 11:00:37 -04003797 btrfs_set_path_blocking(path);
Chris Mason85e21ba2008-01-29 15:11:36 -05003798 wret = push_leaf_left(trans, root, path, 1, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04003799 if (wret < 0 && wret != -ENOSPC)
Chris Masonaa5d6be2007-02-28 16:35:06 -05003800 ret = wret;
Chris Mason5f39d392007-10-15 16:14:19 -04003801
3802 if (path->nodes[0] == leaf &&
3803 btrfs_header_nritems(leaf)) {
Chris Mason85e21ba2008-01-29 15:11:36 -05003804 wret = push_leaf_right(trans, root, path, 1, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04003805 if (wret < 0 && wret != -ENOSPC)
Chris Masonaa5d6be2007-02-28 16:35:06 -05003806 ret = wret;
3807 }
Chris Mason5f39d392007-10-15 16:14:19 -04003808
3809 if (btrfs_header_nritems(leaf) == 0) {
Chris Mason323ac952008-10-01 19:05:46 -04003810 path->slots[1] = slot;
Chris Masond3977122009-01-05 21:25:51 -05003811 ret = btrfs_del_leaf(trans, root, path,
3812 leaf->start);
Chris Mason323ac952008-10-01 19:05:46 -04003813 BUG_ON(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04003814 free_extent_buffer(leaf);
Chris Mason5de08d72007-02-24 06:24:44 -05003815 } else {
Chris Mason925baed2008-06-25 16:01:30 -04003816 /* if we're still in the path, make sure
3817 * we're dirty. Otherwise, one of the
3818 * push_leaf functions must have already
3819 * dirtied this buffer
3820 */
3821 if (path->nodes[0] == leaf)
3822 btrfs_mark_buffer_dirty(leaf);
Chris Mason5f39d392007-10-15 16:14:19 -04003823 free_extent_buffer(leaf);
Chris Masonbe0e5c02007-01-26 15:51:26 -05003824 }
Chris Masond5719762007-03-23 10:01:08 -04003825 } else {
Chris Mason5f39d392007-10-15 16:14:19 -04003826 btrfs_mark_buffer_dirty(leaf);
Chris Masonbe0e5c02007-01-26 15:51:26 -05003827 }
3828 }
Chris Masonaa5d6be2007-02-28 16:35:06 -05003829 return ret;
Chris Masonbe0e5c02007-01-26 15:51:26 -05003830}
3831
Chris Mason97571fd2007-02-24 13:39:08 -05003832/*
Chris Mason925baed2008-06-25 16:01:30 -04003833 * search the tree again to find a leaf with lesser keys
Chris Mason7bb86312007-12-11 09:25:06 -05003834 * returns 0 if it found something or 1 if there are no lesser leaves.
3835 * returns < 0 on io errors.
Chris Masond352ac62008-09-29 15:18:18 -04003836 *
3837 * This may release the path, and so you may lose any locks held at the
3838 * time you call it.
Chris Mason7bb86312007-12-11 09:25:06 -05003839 */
3840int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3841{
Chris Mason925baed2008-06-25 16:01:30 -04003842 struct btrfs_key key;
3843 struct btrfs_disk_key found_key;
3844 int ret;
Chris Mason7bb86312007-12-11 09:25:06 -05003845
Chris Mason925baed2008-06-25 16:01:30 -04003846 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
Chris Mason7bb86312007-12-11 09:25:06 -05003847
Chris Mason925baed2008-06-25 16:01:30 -04003848 if (key.offset > 0)
3849 key.offset--;
3850 else if (key.type > 0)
3851 key.type--;
3852 else if (key.objectid > 0)
3853 key.objectid--;
3854 else
3855 return 1;
Chris Mason7bb86312007-12-11 09:25:06 -05003856
Chris Mason925baed2008-06-25 16:01:30 -04003857 btrfs_release_path(root, path);
3858 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3859 if (ret < 0)
3860 return ret;
3861 btrfs_item_key(path->nodes[0], &found_key, 0);
3862 ret = comp_keys(&found_key, &key);
3863 if (ret < 0)
3864 return 0;
3865 return 1;
Chris Mason7bb86312007-12-11 09:25:06 -05003866}
3867
Chris Mason3f157a22008-06-25 16:01:31 -04003868/*
3869 * A helper function to walk down the tree starting at min_key, and looking
3870 * for nodes or leaves that are either in cache or have a minimum
Chris Masond352ac62008-09-29 15:18:18 -04003871 * transaction id. This is used by the btree defrag code, and tree logging
Chris Mason3f157a22008-06-25 16:01:31 -04003872 *
3873 * This does not cow, but it does stuff the starting key it finds back
3874 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3875 * key and get a writable path.
3876 *
3877 * This does lock as it descends, and path->keep_locks should be set
3878 * to 1 by the caller.
3879 *
3880 * This honors path->lowest_level to prevent descent past a given level
3881 * of the tree.
3882 *
Chris Masond352ac62008-09-29 15:18:18 -04003883 * min_trans indicates the oldest transaction that you are interested
3884 * in walking through. Any nodes or leaves older than min_trans are
3885 * skipped over (without reading them).
3886 *
Chris Mason3f157a22008-06-25 16:01:31 -04003887 * returns zero if something useful was found, < 0 on error and 1 if there
3888 * was nothing in the tree that matched the search criteria.
3889 */
3890int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
Chris Masone02119d2008-09-05 16:13:11 -04003891 struct btrfs_key *max_key,
Chris Mason3f157a22008-06-25 16:01:31 -04003892 struct btrfs_path *path, int cache_only,
3893 u64 min_trans)
3894{
3895 struct extent_buffer *cur;
3896 struct btrfs_key found_key;
3897 int slot;
Yan96524802008-07-24 12:19:49 -04003898 int sret;
Chris Mason3f157a22008-06-25 16:01:31 -04003899 u32 nritems;
3900 int level;
3901 int ret = 1;
3902
Chris Mason934d3752008-12-08 16:43:10 -05003903 WARN_ON(!path->keep_locks);
Chris Mason3f157a22008-06-25 16:01:31 -04003904again:
3905 cur = btrfs_lock_root_node(root);
3906 level = btrfs_header_level(cur);
Chris Masone02119d2008-09-05 16:13:11 -04003907 WARN_ON(path->nodes[level]);
Chris Mason3f157a22008-06-25 16:01:31 -04003908 path->nodes[level] = cur;
3909 path->locks[level] = 1;
3910
3911 if (btrfs_header_generation(cur) < min_trans) {
3912 ret = 1;
3913 goto out;
3914 }
Chris Masond3977122009-01-05 21:25:51 -05003915 while (1) {
Chris Mason3f157a22008-06-25 16:01:31 -04003916 nritems = btrfs_header_nritems(cur);
3917 level = btrfs_header_level(cur);
Yan96524802008-07-24 12:19:49 -04003918 sret = bin_search(cur, min_key, level, &slot);
Chris Mason3f157a22008-06-25 16:01:31 -04003919
Chris Mason323ac952008-10-01 19:05:46 -04003920 /* at the lowest level, we're done, setup the path and exit */
3921 if (level == path->lowest_level) {
Chris Masone02119d2008-09-05 16:13:11 -04003922 if (slot >= nritems)
3923 goto find_next_key;
Chris Mason3f157a22008-06-25 16:01:31 -04003924 ret = 0;
3925 path->slots[level] = slot;
3926 btrfs_item_key_to_cpu(cur, &found_key, slot);
3927 goto out;
3928 }
Yan96524802008-07-24 12:19:49 -04003929 if (sret && slot > 0)
3930 slot--;
Chris Mason3f157a22008-06-25 16:01:31 -04003931 /*
3932 * check this node pointer against the cache_only and
3933 * min_trans parameters. If it isn't in cache or is too
3934 * old, skip to the next one.
3935 */
Chris Masond3977122009-01-05 21:25:51 -05003936 while (slot < nritems) {
Chris Mason3f157a22008-06-25 16:01:31 -04003937 u64 blockptr;
3938 u64 gen;
3939 struct extent_buffer *tmp;
Chris Masone02119d2008-09-05 16:13:11 -04003940 struct btrfs_disk_key disk_key;
3941
Chris Mason3f157a22008-06-25 16:01:31 -04003942 blockptr = btrfs_node_blockptr(cur, slot);
3943 gen = btrfs_node_ptr_generation(cur, slot);
3944 if (gen < min_trans) {
3945 slot++;
3946 continue;
3947 }
3948 if (!cache_only)
3949 break;
3950
Chris Masone02119d2008-09-05 16:13:11 -04003951 if (max_key) {
3952 btrfs_node_key(cur, &disk_key, slot);
3953 if (comp_keys(&disk_key, max_key) >= 0) {
3954 ret = 1;
3955 goto out;
3956 }
3957 }
3958
Chris Mason3f157a22008-06-25 16:01:31 -04003959 tmp = btrfs_find_tree_block(root, blockptr,
3960 btrfs_level_size(root, level - 1));
3961
3962 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3963 free_extent_buffer(tmp);
3964 break;
3965 }
3966 if (tmp)
3967 free_extent_buffer(tmp);
3968 slot++;
3969 }
Chris Masone02119d2008-09-05 16:13:11 -04003970find_next_key:
Chris Mason3f157a22008-06-25 16:01:31 -04003971 /*
3972 * we didn't find a candidate key in this node, walk forward
3973 * and find another one
3974 */
3975 if (slot >= nritems) {
Chris Masone02119d2008-09-05 16:13:11 -04003976 path->slots[level] = slot;
Chris Masonb4ce94d2009-02-04 09:25:08 -05003977 btrfs_set_path_blocking(path);
Chris Masone02119d2008-09-05 16:13:11 -04003978 sret = btrfs_find_next_key(root, path, min_key, level,
Chris Mason3f157a22008-06-25 16:01:31 -04003979 cache_only, min_trans);
Chris Masone02119d2008-09-05 16:13:11 -04003980 if (sret == 0) {
Chris Mason3f157a22008-06-25 16:01:31 -04003981 btrfs_release_path(root, path);
3982 goto again;
3983 } else {
3984 goto out;
3985 }
3986 }
3987 /* save our key for returning back */
3988 btrfs_node_key_to_cpu(cur, &found_key, slot);
3989 path->slots[level] = slot;
3990 if (level == path->lowest_level) {
3991 ret = 0;
3992 unlock_up(path, level, 1);
3993 goto out;
3994 }
Chris Masonb4ce94d2009-02-04 09:25:08 -05003995 btrfs_set_path_blocking(path);
Chris Mason3f157a22008-06-25 16:01:31 -04003996 cur = read_node_slot(root, cur, slot);
3997
3998 btrfs_tree_lock(cur);
Chris Masonb4ce94d2009-02-04 09:25:08 -05003999
Chris Mason3f157a22008-06-25 16:01:31 -04004000 path->locks[level - 1] = 1;
4001 path->nodes[level - 1] = cur;
4002 unlock_up(path, level, 1);
Chris Mason4008c042009-02-12 14:09:45 -05004003 btrfs_clear_path_blocking(path, NULL);
Chris Mason3f157a22008-06-25 16:01:31 -04004004 }
4005out:
4006 if (ret == 0)
4007 memcpy(min_key, &found_key, sizeof(found_key));
Chris Masonb4ce94d2009-02-04 09:25:08 -05004008 btrfs_set_path_blocking(path);
Chris Mason3f157a22008-06-25 16:01:31 -04004009 return ret;
4010}
4011
4012/*
4013 * this is similar to btrfs_next_leaf, but does not try to preserve
4014 * and fixup the path. It looks for and returns the next key in the
4015 * tree based on the current path and the cache_only and min_trans
4016 * parameters.
4017 *
4018 * 0 is returned if another key is found, < 0 if there are any errors
4019 * and 1 is returned if there are no higher keys in the tree
4020 *
4021 * path->keep_locks should be set to 1 on the search made before
4022 * calling this function.
4023 */
Chris Masone7a84562008-06-25 16:01:31 -04004024int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
Chris Mason3f157a22008-06-25 16:01:31 -04004025 struct btrfs_key *key, int lowest_level,
4026 int cache_only, u64 min_trans)
Chris Masone7a84562008-06-25 16:01:31 -04004027{
4028 int level = lowest_level;
4029 int slot;
4030 struct extent_buffer *c;
4031
Chris Mason934d3752008-12-08 16:43:10 -05004032 WARN_ON(!path->keep_locks);
Chris Masond3977122009-01-05 21:25:51 -05004033 while (level < BTRFS_MAX_LEVEL) {
Chris Masone7a84562008-06-25 16:01:31 -04004034 if (!path->nodes[level])
4035 return 1;
4036
4037 slot = path->slots[level] + 1;
4038 c = path->nodes[level];
Chris Mason3f157a22008-06-25 16:01:31 -04004039next:
Chris Masone7a84562008-06-25 16:01:31 -04004040 if (slot >= btrfs_header_nritems(c)) {
4041 level++;
Chris Masond3977122009-01-05 21:25:51 -05004042 if (level == BTRFS_MAX_LEVEL)
Chris Masone7a84562008-06-25 16:01:31 -04004043 return 1;
Chris Masone7a84562008-06-25 16:01:31 -04004044 continue;
4045 }
4046 if (level == 0)
4047 btrfs_item_key_to_cpu(c, key, slot);
Chris Mason3f157a22008-06-25 16:01:31 -04004048 else {
4049 u64 blockptr = btrfs_node_blockptr(c, slot);
4050 u64 gen = btrfs_node_ptr_generation(c, slot);
4051
4052 if (cache_only) {
4053 struct extent_buffer *cur;
4054 cur = btrfs_find_tree_block(root, blockptr,
4055 btrfs_level_size(root, level - 1));
4056 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4057 slot++;
4058 if (cur)
4059 free_extent_buffer(cur);
4060 goto next;
4061 }
4062 free_extent_buffer(cur);
4063 }
4064 if (gen < min_trans) {
4065 slot++;
4066 goto next;
4067 }
Chris Masone7a84562008-06-25 16:01:31 -04004068 btrfs_node_key_to_cpu(c, key, slot);
Chris Mason3f157a22008-06-25 16:01:31 -04004069 }
Chris Masone7a84562008-06-25 16:01:31 -04004070 return 0;
4071 }
4072 return 1;
4073}
4074
Chris Mason7bb86312007-12-11 09:25:06 -05004075/*
Chris Mason925baed2008-06-25 16:01:30 -04004076 * search the tree again to find a leaf with greater keys
Chris Mason0f70abe2007-02-28 16:46:22 -05004077 * returns 0 if it found something or 1 if there are no greater leaves.
4078 * returns < 0 on io errors.
Chris Mason97571fd2007-02-24 13:39:08 -05004079 */
Chris Mason234b63a2007-03-13 10:46:10 -04004080int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
Chris Masond97e63b2007-02-20 16:40:44 -05004081{
4082 int slot;
4083 int level = 1;
Chris Mason5f39d392007-10-15 16:14:19 -04004084 struct extent_buffer *c;
4085 struct extent_buffer *next = NULL;
Chris Mason925baed2008-06-25 16:01:30 -04004086 struct btrfs_key key;
4087 u32 nritems;
4088 int ret;
4089
4090 nritems = btrfs_header_nritems(path->nodes[0]);
Chris Masond3977122009-01-05 21:25:51 -05004091 if (nritems == 0)
Chris Mason925baed2008-06-25 16:01:30 -04004092 return 1;
Chris Mason925baed2008-06-25 16:01:30 -04004093
4094 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4095
Chris Mason925baed2008-06-25 16:01:30 -04004096 btrfs_release_path(root, path);
Chris Masona2135012008-06-25 16:01:30 -04004097 path->keep_locks = 1;
Chris Mason925baed2008-06-25 16:01:30 -04004098 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4099 path->keep_locks = 0;
4100
4101 if (ret < 0)
4102 return ret;
4103
Chris Masonb4ce94d2009-02-04 09:25:08 -05004104 btrfs_set_path_blocking(path);
Chris Masona2135012008-06-25 16:01:30 -04004105 nritems = btrfs_header_nritems(path->nodes[0]);
Chris Mason168fd7d2008-06-25 16:01:30 -04004106 /*
4107 * by releasing the path above we dropped all our locks. A balance
4108 * could have added more items next to the key that used to be
4109 * at the very end of the block. So, check again here and
4110 * advance the path if there are now more items available.
4111 */
Chris Masona2135012008-06-25 16:01:30 -04004112 if (nritems > 0 && path->slots[0] < nritems - 1) {
Chris Mason168fd7d2008-06-25 16:01:30 -04004113 path->slots[0]++;
Chris Mason925baed2008-06-25 16:01:30 -04004114 goto done;
4115 }
Chris Masond97e63b2007-02-20 16:40:44 -05004116
Chris Masond3977122009-01-05 21:25:51 -05004117 while (level < BTRFS_MAX_LEVEL) {
Chris Masond97e63b2007-02-20 16:40:44 -05004118 if (!path->nodes[level])
Chris Mason0f70abe2007-02-28 16:46:22 -05004119 return 1;
Chris Mason5f39d392007-10-15 16:14:19 -04004120
Chris Masond97e63b2007-02-20 16:40:44 -05004121 slot = path->slots[level] + 1;
4122 c = path->nodes[level];
Chris Mason5f39d392007-10-15 16:14:19 -04004123 if (slot >= btrfs_header_nritems(c)) {
Chris Masond97e63b2007-02-20 16:40:44 -05004124 level++;
Chris Masond3977122009-01-05 21:25:51 -05004125 if (level == BTRFS_MAX_LEVEL)
Chris Mason7bb86312007-12-11 09:25:06 -05004126 return 1;
Chris Masond97e63b2007-02-20 16:40:44 -05004127 continue;
4128 }
Chris Mason5f39d392007-10-15 16:14:19 -04004129
Chris Mason925baed2008-06-25 16:01:30 -04004130 if (next) {
4131 btrfs_tree_unlock(next);
Chris Mason5f39d392007-10-15 16:14:19 -04004132 free_extent_buffer(next);
Chris Mason925baed2008-06-25 16:01:30 -04004133 }
Chris Mason5f39d392007-10-15 16:14:19 -04004134
Chris Masonb4ce94d2009-02-04 09:25:08 -05004135 /* the path was set to blocking above */
Chris Mason0bd40a72008-07-17 12:54:43 -04004136 if (level == 1 && (path->locks[1] || path->skip_locking) &&
4137 path->reada)
Chris Mason01f46652007-12-21 16:24:26 -05004138 reada_for_search(root, path, level, slot, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04004139
Chris Masonca7a79a2008-05-12 12:59:19 -04004140 next = read_node_slot(root, c, slot);
Chris Mason5cd57b22008-06-25 16:01:30 -04004141 if (!path->skip_locking) {
Chris Masonb9447ef2009-03-09 11:45:38 -04004142 btrfs_assert_tree_locked(c);
Chris Mason5cd57b22008-06-25 16:01:30 -04004143 btrfs_tree_lock(next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05004144 btrfs_set_lock_blocking(next);
Chris Mason5cd57b22008-06-25 16:01:30 -04004145 }
Chris Masond97e63b2007-02-20 16:40:44 -05004146 break;
4147 }
4148 path->slots[level] = slot;
Chris Masond3977122009-01-05 21:25:51 -05004149 while (1) {
Chris Masond97e63b2007-02-20 16:40:44 -05004150 level--;
4151 c = path->nodes[level];
Chris Mason925baed2008-06-25 16:01:30 -04004152 if (path->locks[level])
4153 btrfs_tree_unlock(c);
Chris Mason5f39d392007-10-15 16:14:19 -04004154 free_extent_buffer(c);
Chris Masond97e63b2007-02-20 16:40:44 -05004155 path->nodes[level] = next;
4156 path->slots[level] = 0;
Chris Masona74a4b92008-06-25 16:01:31 -04004157 if (!path->skip_locking)
4158 path->locks[level] = 1;
Chris Masond97e63b2007-02-20 16:40:44 -05004159 if (!level)
4160 break;
Chris Masonb4ce94d2009-02-04 09:25:08 -05004161
4162 btrfs_set_path_blocking(path);
Chris Mason925baed2008-06-25 16:01:30 -04004163 if (level == 1 && path->locks[1] && path->reada)
4164 reada_for_search(root, path, level, slot, 0);
Chris Masonca7a79a2008-05-12 12:59:19 -04004165 next = read_node_slot(root, next, 0);
Chris Mason5cd57b22008-06-25 16:01:30 -04004166 if (!path->skip_locking) {
Chris Masonb9447ef2009-03-09 11:45:38 -04004167 btrfs_assert_tree_locked(path->nodes[level]);
Chris Mason5cd57b22008-06-25 16:01:30 -04004168 btrfs_tree_lock(next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05004169 btrfs_set_lock_blocking(next);
Chris Mason5cd57b22008-06-25 16:01:30 -04004170 }
Chris Masond97e63b2007-02-20 16:40:44 -05004171 }
Chris Mason925baed2008-06-25 16:01:30 -04004172done:
4173 unlock_up(path, 0, 1);
Chris Masond97e63b2007-02-20 16:40:44 -05004174 return 0;
4175}
Chris Mason0b86a832008-03-24 15:01:56 -04004176
Chris Mason3f157a22008-06-25 16:01:31 -04004177/*
4178 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4179 * searching until it gets past min_objectid or finds an item of 'type'
4180 *
4181 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4182 */
Chris Mason0b86a832008-03-24 15:01:56 -04004183int btrfs_previous_item(struct btrfs_root *root,
4184 struct btrfs_path *path, u64 min_objectid,
4185 int type)
4186{
4187 struct btrfs_key found_key;
4188 struct extent_buffer *leaf;
Chris Masone02119d2008-09-05 16:13:11 -04004189 u32 nritems;
Chris Mason0b86a832008-03-24 15:01:56 -04004190 int ret;
4191
Chris Masond3977122009-01-05 21:25:51 -05004192 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04004193 if (path->slots[0] == 0) {
Chris Masonb4ce94d2009-02-04 09:25:08 -05004194 btrfs_set_path_blocking(path);
Chris Mason0b86a832008-03-24 15:01:56 -04004195 ret = btrfs_prev_leaf(root, path);
4196 if (ret != 0)
4197 return ret;
4198 } else {
4199 path->slots[0]--;
4200 }
4201 leaf = path->nodes[0];
Chris Masone02119d2008-09-05 16:13:11 -04004202 nritems = btrfs_header_nritems(leaf);
4203 if (nritems == 0)
4204 return 1;
4205 if (path->slots[0] == nritems)
4206 path->slots[0]--;
4207
Chris Mason0b86a832008-03-24 15:01:56 -04004208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4209 if (found_key.type == type)
4210 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04004211 if (found_key.objectid < min_objectid)
4212 break;
4213 if (found_key.objectid == min_objectid &&
4214 found_key.type < type)
4215 break;
Chris Mason0b86a832008-03-24 15:01:56 -04004216 }
4217 return 1;
4218}