]>
Commit | Line | Data |
---|---|---|
6cbd5570 | 1 | /* |
d352ac68 | 2 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
a6b6e75e | 19 | #include <linux/sched.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
eb60ceac CM |
21 | #include "ctree.h" |
22 | #include "disk-io.h" | |
7f5c1516 | 23 | #include "transaction.h" |
5f39d397 | 24 | #include "print-tree.h" |
925baedd | 25 | #include "locking.h" |
9a8dd150 | 26 | |
e089f05c CM |
27 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
28 | *root, struct btrfs_path *path, int level); | |
29 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 30 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 31 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
32 | static int push_node_left(struct btrfs_trans_handle *trans, |
33 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 34 | struct extent_buffer *src, int empty); |
5f39d397 CM |
35 | static int balance_node_right(struct btrfs_trans_handle *trans, |
36 | struct btrfs_root *root, | |
37 | struct extent_buffer *dst_buf, | |
38 | struct extent_buffer *src_buf); | |
143bede5 | 39 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
e089f05c | 40 | struct btrfs_path *path, int level, int slot); |
d97e63b6 | 41 | |
df24a2b9 | 42 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 43 | { |
df24a2b9 | 44 | struct btrfs_path *path; |
e00f7308 | 45 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 46 | return path; |
2c90e5d6 CM |
47 | } |
48 | ||
b4ce94de CM |
49 | /* |
50 | * set all locked nodes in the path to blocking locks. This should | |
51 | * be done before scheduling | |
52 | */ | |
53 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
54 | { | |
55 | int i; | |
56 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
57 | if (!p->nodes[i] || !p->locks[i]) |
58 | continue; | |
59 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
60 | if (p->locks[i] == BTRFS_READ_LOCK) | |
61 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
62 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
63 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
64 | } |
65 | } | |
66 | ||
67 | /* | |
68 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
69 | * |
70 | * held is used to keep lockdep happy, when lockdep is enabled | |
71 | * we set held to a blocking lock before we go around and | |
72 | * retake all the spinlocks in the path. You can safely use NULL | |
73 | * for held | |
b4ce94de | 74 | */ |
4008c04a | 75 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 76 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
77 | { |
78 | int i; | |
4008c04a CM |
79 | |
80 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
81 | /* lockdep really cares that we take all of these spinlocks | |
82 | * in the right order. If any of the locks in the path are not | |
83 | * currently blocking, it is going to complain. So, make really | |
84 | * really sure by forcing the path to blocking before we clear | |
85 | * the path blocking. | |
86 | */ | |
bd681513 CM |
87 | if (held) { |
88 | btrfs_set_lock_blocking_rw(held, held_rw); | |
89 | if (held_rw == BTRFS_WRITE_LOCK) | |
90 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
91 | else if (held_rw == BTRFS_READ_LOCK) | |
92 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
93 | } | |
4008c04a CM |
94 | btrfs_set_path_blocking(p); |
95 | #endif | |
96 | ||
97 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
98 | if (p->nodes[i] && p->locks[i]) { |
99 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
100 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
101 | p->locks[i] = BTRFS_WRITE_LOCK; | |
102 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
103 | p->locks[i] = BTRFS_READ_LOCK; | |
104 | } | |
b4ce94de | 105 | } |
4008c04a CM |
106 | |
107 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
108 | if (held) | |
bd681513 | 109 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 110 | #endif |
b4ce94de CM |
111 | } |
112 | ||
d352ac68 | 113 | /* this also releases the path */ |
df24a2b9 | 114 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 115 | { |
ff175d57 JJ |
116 | if (!p) |
117 | return; | |
b3b4aa74 | 118 | btrfs_release_path(p); |
df24a2b9 | 119 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
120 | } |
121 | ||
d352ac68 CM |
122 | /* |
123 | * path release drops references on the extent buffers in the path | |
124 | * and it drops any locks held by this path | |
125 | * | |
126 | * It is safe to call this on paths that no locks or extent buffers held. | |
127 | */ | |
b3b4aa74 | 128 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
129 | { |
130 | int i; | |
a2135011 | 131 | |
234b63a0 | 132 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 133 | p->slots[i] = 0; |
eb60ceac | 134 | if (!p->nodes[i]) |
925baedd CM |
135 | continue; |
136 | if (p->locks[i]) { | |
bd681513 | 137 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
138 | p->locks[i] = 0; |
139 | } | |
5f39d397 | 140 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 141 | p->nodes[i] = NULL; |
eb60ceac CM |
142 | } |
143 | } | |
144 | ||
d352ac68 CM |
145 | /* |
146 | * safely gets a reference on the root node of a tree. A lock | |
147 | * is not taken, so a concurrent writer may put a different node | |
148 | * at the root of the tree. See btrfs_lock_root_node for the | |
149 | * looping required. | |
150 | * | |
151 | * The extent buffer returned by this has a reference taken, so | |
152 | * it won't disappear. It may stop being the root of the tree | |
153 | * at any time because there are no locks held. | |
154 | */ | |
925baedd CM |
155 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
156 | { | |
157 | struct extent_buffer *eb; | |
240f62c8 CM |
158 | |
159 | rcu_read_lock(); | |
160 | eb = rcu_dereference(root->node); | |
925baedd | 161 | extent_buffer_get(eb); |
240f62c8 | 162 | rcu_read_unlock(); |
925baedd CM |
163 | return eb; |
164 | } | |
165 | ||
d352ac68 CM |
166 | /* loop around taking references on and locking the root node of the |
167 | * tree until you end up with a lock on the root. A locked buffer | |
168 | * is returned, with a reference held. | |
169 | */ | |
925baedd CM |
170 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
171 | { | |
172 | struct extent_buffer *eb; | |
173 | ||
d397712b | 174 | while (1) { |
925baedd CM |
175 | eb = btrfs_root_node(root); |
176 | btrfs_tree_lock(eb); | |
240f62c8 | 177 | if (eb == root->node) |
925baedd | 178 | break; |
925baedd CM |
179 | btrfs_tree_unlock(eb); |
180 | free_extent_buffer(eb); | |
181 | } | |
182 | return eb; | |
183 | } | |
184 | ||
bd681513 CM |
185 | /* loop around taking references on and locking the root node of the |
186 | * tree until you end up with a lock on the root. A locked buffer | |
187 | * is returned, with a reference held. | |
188 | */ | |
189 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) | |
190 | { | |
191 | struct extent_buffer *eb; | |
192 | ||
193 | while (1) { | |
194 | eb = btrfs_root_node(root); | |
195 | btrfs_tree_read_lock(eb); | |
196 | if (eb == root->node) | |
197 | break; | |
198 | btrfs_tree_read_unlock(eb); | |
199 | free_extent_buffer(eb); | |
200 | } | |
201 | return eb; | |
202 | } | |
203 | ||
d352ac68 CM |
204 | /* cowonly root (everything not a reference counted cow subvolume), just get |
205 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
206 | * get properly updated on disk. | |
207 | */ | |
0b86a832 CM |
208 | static void add_root_to_dirty_list(struct btrfs_root *root) |
209 | { | |
210 | if (root->track_dirty && list_empty(&root->dirty_list)) { | |
211 | list_add(&root->dirty_list, | |
212 | &root->fs_info->dirty_cowonly_roots); | |
213 | } | |
214 | } | |
215 | ||
d352ac68 CM |
216 | /* |
217 | * used by snapshot creation to make a copy of a root for a tree with | |
218 | * a given objectid. The buffer with the new root node is returned in | |
219 | * cow_ret, and this func returns zero on success or a negative error code. | |
220 | */ | |
be20aa9d CM |
221 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
222 | struct btrfs_root *root, | |
223 | struct extent_buffer *buf, | |
224 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
225 | { | |
226 | struct extent_buffer *cow; | |
be20aa9d CM |
227 | int ret = 0; |
228 | int level; | |
5d4f98a2 | 229 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
230 | |
231 | WARN_ON(root->ref_cows && trans->transid != | |
232 | root->fs_info->running_transaction->transid); | |
233 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
234 | ||
235 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
236 | if (level == 0) |
237 | btrfs_item_key(buf, &disk_key, 0); | |
238 | else | |
239 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 240 | |
5d4f98a2 YZ |
241 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
242 | new_root_objectid, &disk_key, level, | |
66d7e7f0 | 243 | buf->start, 0, 1); |
5d4f98a2 | 244 | if (IS_ERR(cow)) |
be20aa9d CM |
245 | return PTR_ERR(cow); |
246 | ||
247 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
248 | btrfs_set_header_bytenr(cow, cow->start); | |
249 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
250 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
251 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
252 | BTRFS_HEADER_FLAG_RELOC); | |
253 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
254 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
255 | else | |
256 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 257 | |
2b82032c YZ |
258 | write_extent_buffer(cow, root->fs_info->fsid, |
259 | (unsigned long)btrfs_header_fsid(cow), | |
260 | BTRFS_FSID_SIZE); | |
261 | ||
be20aa9d | 262 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 263 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 264 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 265 | else |
66d7e7f0 | 266 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 267 | |
be20aa9d CM |
268 | if (ret) |
269 | return ret; | |
270 | ||
271 | btrfs_mark_buffer_dirty(cow); | |
272 | *cow_ret = cow; | |
273 | return 0; | |
274 | } | |
275 | ||
5d4f98a2 YZ |
276 | /* |
277 | * check if the tree block can be shared by multiple trees | |
278 | */ | |
279 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
280 | struct extent_buffer *buf) | |
281 | { | |
282 | /* | |
283 | * Tree blocks not in refernece counted trees and tree roots | |
284 | * are never shared. If a block was allocated after the last | |
285 | * snapshot and the block was not allocated by tree relocation, | |
286 | * we know the block is not shared. | |
287 | */ | |
288 | if (root->ref_cows && | |
289 | buf != root->node && buf != root->commit_root && | |
290 | (btrfs_header_generation(buf) <= | |
291 | btrfs_root_last_snapshot(&root->root_item) || | |
292 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
293 | return 1; | |
294 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
295 | if (root->ref_cows && | |
296 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
297 | return 1; | |
298 | #endif | |
299 | return 0; | |
300 | } | |
301 | ||
302 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
303 | struct btrfs_root *root, | |
304 | struct extent_buffer *buf, | |
f0486c68 YZ |
305 | struct extent_buffer *cow, |
306 | int *last_ref) | |
5d4f98a2 YZ |
307 | { |
308 | u64 refs; | |
309 | u64 owner; | |
310 | u64 flags; | |
311 | u64 new_flags = 0; | |
312 | int ret; | |
313 | ||
314 | /* | |
315 | * Backrefs update rules: | |
316 | * | |
317 | * Always use full backrefs for extent pointers in tree block | |
318 | * allocated by tree relocation. | |
319 | * | |
320 | * If a shared tree block is no longer referenced by its owner | |
321 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
322 | * use full backrefs for extent pointers in tree block. | |
323 | * | |
324 | * If a tree block is been relocating | |
325 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
326 | * use full backrefs for extent pointers in tree block. | |
327 | * The reason for this is some operations (such as drop tree) | |
328 | * are only allowed for blocks use full backrefs. | |
329 | */ | |
330 | ||
331 | if (btrfs_block_can_be_shared(root, buf)) { | |
332 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
333 | buf->len, &refs, &flags); | |
be1a5564 MF |
334 | if (ret) |
335 | return ret; | |
5d4f98a2 YZ |
336 | BUG_ON(refs == 0); |
337 | } else { | |
338 | refs = 1; | |
339 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
340 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
341 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
342 | else | |
343 | flags = 0; | |
344 | } | |
345 | ||
346 | owner = btrfs_header_owner(buf); | |
347 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
348 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
349 | ||
350 | if (refs > 1) { | |
351 | if ((owner == root->root_key.objectid || | |
352 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
353 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 354 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
5d4f98a2 YZ |
355 | BUG_ON(ret); |
356 | ||
357 | if (root->root_key.objectid == | |
358 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 359 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
5d4f98a2 | 360 | BUG_ON(ret); |
66d7e7f0 | 361 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 YZ |
362 | BUG_ON(ret); |
363 | } | |
364 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
365 | } else { | |
366 | ||
367 | if (root->root_key.objectid == | |
368 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 369 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 370 | else |
66d7e7f0 | 371 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
5d4f98a2 YZ |
372 | BUG_ON(ret); |
373 | } | |
374 | if (new_flags != 0) { | |
375 | ret = btrfs_set_disk_extent_flags(trans, root, | |
376 | buf->start, | |
377 | buf->len, | |
378 | new_flags, 0); | |
be1a5564 MF |
379 | if (ret) |
380 | return ret; | |
5d4f98a2 YZ |
381 | } |
382 | } else { | |
383 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
384 | if (root->root_key.objectid == | |
385 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 386 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 387 | else |
66d7e7f0 | 388 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
5d4f98a2 | 389 | BUG_ON(ret); |
66d7e7f0 | 390 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
5d4f98a2 YZ |
391 | BUG_ON(ret); |
392 | } | |
393 | clean_tree_block(trans, root, buf); | |
f0486c68 | 394 | *last_ref = 1; |
5d4f98a2 YZ |
395 | } |
396 | return 0; | |
397 | } | |
398 | ||
d352ac68 | 399 | /* |
d397712b CM |
400 | * does the dirty work in cow of a single block. The parent block (if |
401 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
402 | * dirty and returned locked. If you modify the block it needs to be marked | |
403 | * dirty again. | |
d352ac68 CM |
404 | * |
405 | * search_start -- an allocation hint for the new block | |
406 | * | |
d397712b CM |
407 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
408 | * bytes the allocator should try to find free next to the block it returns. | |
409 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 410 | */ |
d397712b | 411 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
412 | struct btrfs_root *root, |
413 | struct extent_buffer *buf, | |
414 | struct extent_buffer *parent, int parent_slot, | |
415 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 416 | u64 search_start, u64 empty_size) |
02217ed2 | 417 | { |
5d4f98a2 | 418 | struct btrfs_disk_key disk_key; |
5f39d397 | 419 | struct extent_buffer *cow; |
be1a5564 | 420 | int level, ret; |
f0486c68 | 421 | int last_ref = 0; |
925baedd | 422 | int unlock_orig = 0; |
5d4f98a2 | 423 | u64 parent_start; |
7bb86316 | 424 | |
925baedd CM |
425 | if (*cow_ret == buf) |
426 | unlock_orig = 1; | |
427 | ||
b9447ef8 | 428 | btrfs_assert_tree_locked(buf); |
925baedd | 429 | |
7bb86316 CM |
430 | WARN_ON(root->ref_cows && trans->transid != |
431 | root->fs_info->running_transaction->transid); | |
6702ed49 | 432 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 433 | |
7bb86316 | 434 | level = btrfs_header_level(buf); |
31840ae1 | 435 | |
5d4f98a2 YZ |
436 | if (level == 0) |
437 | btrfs_item_key(buf, &disk_key, 0); | |
438 | else | |
439 | btrfs_node_key(buf, &disk_key, 0); | |
440 | ||
441 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
442 | if (parent) | |
443 | parent_start = parent->start; | |
444 | else | |
445 | parent_start = 0; | |
446 | } else | |
447 | parent_start = 0; | |
448 | ||
449 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
450 | root->root_key.objectid, &disk_key, | |
66d7e7f0 | 451 | level, search_start, empty_size, 1); |
54aa1f4d CM |
452 | if (IS_ERR(cow)) |
453 | return PTR_ERR(cow); | |
6702ed49 | 454 | |
b4ce94de CM |
455 | /* cow is set to blocking by btrfs_init_new_buffer */ |
456 | ||
5f39d397 | 457 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 458 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 459 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
460 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
461 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
462 | BTRFS_HEADER_FLAG_RELOC); | |
463 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
464 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
465 | else | |
466 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 467 | |
2b82032c YZ |
468 | write_extent_buffer(cow, root->fs_info->fsid, |
469 | (unsigned long)btrfs_header_fsid(cow), | |
470 | BTRFS_FSID_SIZE); | |
471 | ||
be1a5564 MF |
472 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
473 | BUG_ON(ret); | |
1a40e23b | 474 | |
3fd0a558 YZ |
475 | if (root->ref_cows) |
476 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
477 | ||
02217ed2 | 478 | if (buf == root->node) { |
925baedd | 479 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
480 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
481 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
482 | parent_start = buf->start; | |
483 | else | |
484 | parent_start = 0; | |
925baedd | 485 | |
5f39d397 | 486 | extent_buffer_get(cow); |
240f62c8 | 487 | rcu_assign_pointer(root->node, cow); |
925baedd | 488 | |
f0486c68 | 489 | btrfs_free_tree_block(trans, root, buf, parent_start, |
66d7e7f0 | 490 | last_ref, 1); |
5f39d397 | 491 | free_extent_buffer(buf); |
0b86a832 | 492 | add_root_to_dirty_list(root); |
02217ed2 | 493 | } else { |
5d4f98a2 YZ |
494 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
495 | parent_start = parent->start; | |
496 | else | |
497 | parent_start = 0; | |
498 | ||
499 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
5f39d397 | 500 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 501 | cow->start); |
74493f7a CM |
502 | btrfs_set_node_ptr_generation(parent, parent_slot, |
503 | trans->transid); | |
d6025579 | 504 | btrfs_mark_buffer_dirty(parent); |
f0486c68 | 505 | btrfs_free_tree_block(trans, root, buf, parent_start, |
66d7e7f0 | 506 | last_ref, 1); |
02217ed2 | 507 | } |
925baedd CM |
508 | if (unlock_orig) |
509 | btrfs_tree_unlock(buf); | |
5f39d397 | 510 | free_extent_buffer(buf); |
ccd467d6 | 511 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 512 | *cow_ret = cow; |
02217ed2 CM |
513 | return 0; |
514 | } | |
515 | ||
5d4f98a2 YZ |
516 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
517 | struct btrfs_root *root, | |
518 | struct extent_buffer *buf) | |
519 | { | |
f1ebcc74 LB |
520 | /* ensure we can see the force_cow */ |
521 | smp_rmb(); | |
522 | ||
523 | /* | |
524 | * We do not need to cow a block if | |
525 | * 1) this block is not created or changed in this transaction; | |
526 | * 2) this block does not belong to TREE_RELOC tree; | |
527 | * 3) the root is not forced COW. | |
528 | * | |
529 | * What is forced COW: | |
530 | * when we create snapshot during commiting the transaction, | |
531 | * after we've finished coping src root, we must COW the shared | |
532 | * block to ensure the metadata consistency. | |
533 | */ | |
5d4f98a2 YZ |
534 | if (btrfs_header_generation(buf) == trans->transid && |
535 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
536 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
537 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
538 | !root->force_cow) | |
5d4f98a2 YZ |
539 | return 0; |
540 | return 1; | |
541 | } | |
542 | ||
d352ac68 CM |
543 | /* |
544 | * cows a single block, see __btrfs_cow_block for the real work. | |
545 | * This version of it has extra checks so that a block isn't cow'd more than | |
546 | * once per transaction, as long as it hasn't been written yet | |
547 | */ | |
d397712b | 548 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
549 | struct btrfs_root *root, struct extent_buffer *buf, |
550 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 551 | struct extent_buffer **cow_ret) |
6702ed49 CM |
552 | { |
553 | u64 search_start; | |
f510cfec | 554 | int ret; |
dc17ff8f | 555 | |
6702ed49 | 556 | if (trans->transaction != root->fs_info->running_transaction) { |
d397712b CM |
557 | printk(KERN_CRIT "trans %llu running %llu\n", |
558 | (unsigned long long)trans->transid, | |
559 | (unsigned long long) | |
6702ed49 CM |
560 | root->fs_info->running_transaction->transid); |
561 | WARN_ON(1); | |
562 | } | |
563 | if (trans->transid != root->fs_info->generation) { | |
d397712b CM |
564 | printk(KERN_CRIT "trans %llu running %llu\n", |
565 | (unsigned long long)trans->transid, | |
566 | (unsigned long long)root->fs_info->generation); | |
6702ed49 CM |
567 | WARN_ON(1); |
568 | } | |
dc17ff8f | 569 | |
5d4f98a2 | 570 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
571 | *cow_ret = buf; |
572 | return 0; | |
573 | } | |
c487685d | 574 | |
0b86a832 | 575 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
576 | |
577 | if (parent) | |
578 | btrfs_set_lock_blocking(parent); | |
579 | btrfs_set_lock_blocking(buf); | |
580 | ||
f510cfec | 581 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 582 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 583 | |
584 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
585 | ||
f510cfec | 586 | return ret; |
6702ed49 CM |
587 | } |
588 | ||
d352ac68 CM |
589 | /* |
590 | * helper function for defrag to decide if two blocks pointed to by a | |
591 | * node are actually close by | |
592 | */ | |
6b80053d | 593 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 594 | { |
6b80053d | 595 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 596 | return 1; |
6b80053d | 597 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
598 | return 1; |
599 | return 0; | |
600 | } | |
601 | ||
081e9573 CM |
602 | /* |
603 | * compare two keys in a memcmp fashion | |
604 | */ | |
605 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
606 | { | |
607 | struct btrfs_key k1; | |
608 | ||
609 | btrfs_disk_key_to_cpu(&k1, disk); | |
610 | ||
20736aba | 611 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
612 | } |
613 | ||
f3465ca4 JB |
614 | /* |
615 | * same as comp_keys only with two btrfs_key's | |
616 | */ | |
5d4f98a2 | 617 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
618 | { |
619 | if (k1->objectid > k2->objectid) | |
620 | return 1; | |
621 | if (k1->objectid < k2->objectid) | |
622 | return -1; | |
623 | if (k1->type > k2->type) | |
624 | return 1; | |
625 | if (k1->type < k2->type) | |
626 | return -1; | |
627 | if (k1->offset > k2->offset) | |
628 | return 1; | |
629 | if (k1->offset < k2->offset) | |
630 | return -1; | |
631 | return 0; | |
632 | } | |
081e9573 | 633 | |
d352ac68 CM |
634 | /* |
635 | * this is used by the defrag code to go through all the | |
636 | * leaves pointed to by a node and reallocate them so that | |
637 | * disk order is close to key order | |
638 | */ | |
6702ed49 | 639 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 640 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
641 | int start_slot, int cache_only, u64 *last_ret, |
642 | struct btrfs_key *progress) | |
6702ed49 | 643 | { |
6b80053d | 644 | struct extent_buffer *cur; |
6702ed49 | 645 | u64 blocknr; |
ca7a79ad | 646 | u64 gen; |
e9d0b13b CM |
647 | u64 search_start = *last_ret; |
648 | u64 last_block = 0; | |
6702ed49 CM |
649 | u64 other; |
650 | u32 parent_nritems; | |
6702ed49 CM |
651 | int end_slot; |
652 | int i; | |
653 | int err = 0; | |
f2183bde | 654 | int parent_level; |
6b80053d CM |
655 | int uptodate; |
656 | u32 blocksize; | |
081e9573 CM |
657 | int progress_passed = 0; |
658 | struct btrfs_disk_key disk_key; | |
6702ed49 | 659 | |
5708b959 CM |
660 | parent_level = btrfs_header_level(parent); |
661 | if (cache_only && parent_level != 1) | |
662 | return 0; | |
663 | ||
d397712b | 664 | if (trans->transaction != root->fs_info->running_transaction) |
6702ed49 | 665 | WARN_ON(1); |
d397712b | 666 | if (trans->transid != root->fs_info->generation) |
6702ed49 | 667 | WARN_ON(1); |
86479a04 | 668 | |
6b80053d | 669 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 670 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
671 | end_slot = parent_nritems; |
672 | ||
673 | if (parent_nritems == 1) | |
674 | return 0; | |
675 | ||
b4ce94de CM |
676 | btrfs_set_lock_blocking(parent); |
677 | ||
6702ed49 CM |
678 | for (i = start_slot; i < end_slot; i++) { |
679 | int close = 1; | |
a6b6e75e | 680 | |
081e9573 CM |
681 | btrfs_node_key(parent, &disk_key, i); |
682 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
683 | continue; | |
684 | ||
685 | progress_passed = 1; | |
6b80053d | 686 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 687 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
688 | if (last_block == 0) |
689 | last_block = blocknr; | |
5708b959 | 690 | |
6702ed49 | 691 | if (i > 0) { |
6b80053d CM |
692 | other = btrfs_node_blockptr(parent, i - 1); |
693 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 694 | } |
0ef3e66b | 695 | if (!close && i < end_slot - 2) { |
6b80053d CM |
696 | other = btrfs_node_blockptr(parent, i + 1); |
697 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 698 | } |
e9d0b13b CM |
699 | if (close) { |
700 | last_block = blocknr; | |
6702ed49 | 701 | continue; |
e9d0b13b | 702 | } |
6702ed49 | 703 | |
6b80053d CM |
704 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
705 | if (cur) | |
1259ab75 | 706 | uptodate = btrfs_buffer_uptodate(cur, gen); |
6b80053d CM |
707 | else |
708 | uptodate = 0; | |
5708b959 | 709 | if (!cur || !uptodate) { |
6702ed49 | 710 | if (cache_only) { |
6b80053d | 711 | free_extent_buffer(cur); |
6702ed49 CM |
712 | continue; |
713 | } | |
6b80053d CM |
714 | if (!cur) { |
715 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 716 | blocksize, gen); |
97d9a8a4 TI |
717 | if (!cur) |
718 | return -EIO; | |
6b80053d | 719 | } else if (!uptodate) { |
ca7a79ad | 720 | btrfs_read_buffer(cur, gen); |
f2183bde | 721 | } |
6702ed49 | 722 | } |
e9d0b13b | 723 | if (search_start == 0) |
6b80053d | 724 | search_start = last_block; |
e9d0b13b | 725 | |
e7a84565 | 726 | btrfs_tree_lock(cur); |
b4ce94de | 727 | btrfs_set_lock_blocking(cur); |
6b80053d | 728 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 729 | &cur, search_start, |
6b80053d | 730 | min(16 * blocksize, |
9fa8cfe7 | 731 | (end_slot - i) * blocksize)); |
252c38f0 | 732 | if (err) { |
e7a84565 | 733 | btrfs_tree_unlock(cur); |
6b80053d | 734 | free_extent_buffer(cur); |
6702ed49 | 735 | break; |
252c38f0 | 736 | } |
e7a84565 CM |
737 | search_start = cur->start; |
738 | last_block = cur->start; | |
f2183bde | 739 | *last_ret = search_start; |
e7a84565 CM |
740 | btrfs_tree_unlock(cur); |
741 | free_extent_buffer(cur); | |
6702ed49 CM |
742 | } |
743 | return err; | |
744 | } | |
745 | ||
74123bd7 CM |
746 | /* |
747 | * The leaf data grows from end-to-front in the node. | |
748 | * this returns the address of the start of the last item, | |
749 | * which is the stop of the leaf data stack | |
750 | */ | |
123abc88 | 751 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 752 | struct extent_buffer *leaf) |
be0e5c09 | 753 | { |
5f39d397 | 754 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 755 | if (nr == 0) |
123abc88 | 756 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 757 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
758 | } |
759 | ||
aa5d6bed | 760 | |
74123bd7 | 761 | /* |
5f39d397 CM |
762 | * search for key in the extent_buffer. The items start at offset p, |
763 | * and they are item_size apart. There are 'max' items in p. | |
764 | * | |
74123bd7 CM |
765 | * the slot in the array is returned via slot, and it points to |
766 | * the place where you would insert key if it is not found in | |
767 | * the array. | |
768 | * | |
769 | * slot may point to max if the key is bigger than all of the keys | |
770 | */ | |
e02119d5 CM |
771 | static noinline int generic_bin_search(struct extent_buffer *eb, |
772 | unsigned long p, | |
773 | int item_size, struct btrfs_key *key, | |
774 | int max, int *slot) | |
be0e5c09 CM |
775 | { |
776 | int low = 0; | |
777 | int high = max; | |
778 | int mid; | |
779 | int ret; | |
479965d6 | 780 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
781 | struct btrfs_disk_key unaligned; |
782 | unsigned long offset; | |
5f39d397 CM |
783 | char *kaddr = NULL; |
784 | unsigned long map_start = 0; | |
785 | unsigned long map_len = 0; | |
479965d6 | 786 | int err; |
be0e5c09 | 787 | |
d397712b | 788 | while (low < high) { |
be0e5c09 | 789 | mid = (low + high) / 2; |
5f39d397 CM |
790 | offset = p + mid * item_size; |
791 | ||
a6591715 | 792 | if (!kaddr || offset < map_start || |
5f39d397 CM |
793 | (offset + sizeof(struct btrfs_disk_key)) > |
794 | map_start + map_len) { | |
934d375b CM |
795 | |
796 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 797 | sizeof(struct btrfs_disk_key), |
a6591715 | 798 | &kaddr, &map_start, &map_len); |
479965d6 CM |
799 | |
800 | if (!err) { | |
801 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
802 | map_start); | |
803 | } else { | |
804 | read_extent_buffer(eb, &unaligned, | |
805 | offset, sizeof(unaligned)); | |
806 | tmp = &unaligned; | |
807 | } | |
5f39d397 | 808 | |
5f39d397 CM |
809 | } else { |
810 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
811 | map_start); | |
812 | } | |
be0e5c09 CM |
813 | ret = comp_keys(tmp, key); |
814 | ||
815 | if (ret < 0) | |
816 | low = mid + 1; | |
817 | else if (ret > 0) | |
818 | high = mid; | |
819 | else { | |
820 | *slot = mid; | |
821 | return 0; | |
822 | } | |
823 | } | |
824 | *slot = low; | |
825 | return 1; | |
826 | } | |
827 | ||
97571fd0 CM |
828 | /* |
829 | * simple bin_search frontend that does the right thing for | |
830 | * leaves vs nodes | |
831 | */ | |
5f39d397 CM |
832 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
833 | int level, int *slot) | |
be0e5c09 | 834 | { |
5f39d397 CM |
835 | if (level == 0) { |
836 | return generic_bin_search(eb, | |
837 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 838 | sizeof(struct btrfs_item), |
5f39d397 | 839 | key, btrfs_header_nritems(eb), |
7518a238 | 840 | slot); |
be0e5c09 | 841 | } else { |
5f39d397 CM |
842 | return generic_bin_search(eb, |
843 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 844 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 845 | key, btrfs_header_nritems(eb), |
7518a238 | 846 | slot); |
be0e5c09 CM |
847 | } |
848 | return -1; | |
849 | } | |
850 | ||
5d4f98a2 YZ |
851 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
852 | int level, int *slot) | |
853 | { | |
854 | return bin_search(eb, key, level, slot); | |
855 | } | |
856 | ||
f0486c68 YZ |
857 | static void root_add_used(struct btrfs_root *root, u32 size) |
858 | { | |
859 | spin_lock(&root->accounting_lock); | |
860 | btrfs_set_root_used(&root->root_item, | |
861 | btrfs_root_used(&root->root_item) + size); | |
862 | spin_unlock(&root->accounting_lock); | |
863 | } | |
864 | ||
865 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
866 | { | |
867 | spin_lock(&root->accounting_lock); | |
868 | btrfs_set_root_used(&root->root_item, | |
869 | btrfs_root_used(&root->root_item) - size); | |
870 | spin_unlock(&root->accounting_lock); | |
871 | } | |
872 | ||
d352ac68 CM |
873 | /* given a node and slot number, this reads the blocks it points to. The |
874 | * extent buffer is returned with a reference taken (but unlocked). | |
875 | * NULL is returned on error. | |
876 | */ | |
e02119d5 | 877 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 878 | struct extent_buffer *parent, int slot) |
bb803951 | 879 | { |
ca7a79ad | 880 | int level = btrfs_header_level(parent); |
bb803951 CM |
881 | if (slot < 0) |
882 | return NULL; | |
5f39d397 | 883 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 884 | return NULL; |
ca7a79ad CM |
885 | |
886 | BUG_ON(level == 0); | |
887 | ||
db94535d | 888 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
889 | btrfs_level_size(root, level - 1), |
890 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
891 | } |
892 | ||
d352ac68 CM |
893 | /* |
894 | * node level balancing, used to make sure nodes are in proper order for | |
895 | * item deletion. We balance from the top down, so we have to make sure | |
896 | * that a deletion won't leave an node completely empty later on. | |
897 | */ | |
e02119d5 | 898 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
899 | struct btrfs_root *root, |
900 | struct btrfs_path *path, int level) | |
bb803951 | 901 | { |
5f39d397 CM |
902 | struct extent_buffer *right = NULL; |
903 | struct extent_buffer *mid; | |
904 | struct extent_buffer *left = NULL; | |
905 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
906 | int ret = 0; |
907 | int wret; | |
908 | int pslot; | |
bb803951 | 909 | int orig_slot = path->slots[level]; |
79f95c82 | 910 | u64 orig_ptr; |
bb803951 CM |
911 | |
912 | if (level == 0) | |
913 | return 0; | |
914 | ||
5f39d397 | 915 | mid = path->nodes[level]; |
b4ce94de | 916 | |
bd681513 CM |
917 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
918 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
919 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
920 | ||
1d4f8a0c | 921 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 922 | |
a05a9bb1 | 923 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 924 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
925 | pslot = path->slots[level + 1]; |
926 | } | |
bb803951 | 927 | |
40689478 CM |
928 | /* |
929 | * deal with the case where there is only one pointer in the root | |
930 | * by promoting the node below to a root | |
931 | */ | |
5f39d397 CM |
932 | if (!parent) { |
933 | struct extent_buffer *child; | |
bb803951 | 934 | |
5f39d397 | 935 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
936 | return 0; |
937 | ||
938 | /* promote the child to a root */ | |
5f39d397 | 939 | child = read_node_slot(root, mid, 0); |
7951f3ce | 940 | BUG_ON(!child); |
925baedd | 941 | btrfs_tree_lock(child); |
b4ce94de | 942 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 943 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
944 | if (ret) { |
945 | btrfs_tree_unlock(child); | |
946 | free_extent_buffer(child); | |
947 | goto enospc; | |
948 | } | |
2f375ab9 | 949 | |
240f62c8 | 950 | rcu_assign_pointer(root->node, child); |
925baedd | 951 | |
0b86a832 | 952 | add_root_to_dirty_list(root); |
925baedd | 953 | btrfs_tree_unlock(child); |
b4ce94de | 954 | |
925baedd | 955 | path->locks[level] = 0; |
bb803951 | 956 | path->nodes[level] = NULL; |
5f39d397 | 957 | clean_tree_block(trans, root, mid); |
925baedd | 958 | btrfs_tree_unlock(mid); |
bb803951 | 959 | /* once for the path */ |
5f39d397 | 960 | free_extent_buffer(mid); |
f0486c68 YZ |
961 | |
962 | root_sub_used(root, mid->len); | |
66d7e7f0 | 963 | btrfs_free_tree_block(trans, root, mid, 0, 1, 0); |
bb803951 | 964 | /* once for the root ptr */ |
5f39d397 | 965 | free_extent_buffer(mid); |
f0486c68 | 966 | return 0; |
bb803951 | 967 | } |
5f39d397 | 968 | if (btrfs_header_nritems(mid) > |
123abc88 | 969 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
970 | return 0; |
971 | ||
559af821 | 972 | btrfs_header_nritems(mid); |
54aa1f4d | 973 | |
5f39d397 CM |
974 | left = read_node_slot(root, parent, pslot - 1); |
975 | if (left) { | |
925baedd | 976 | btrfs_tree_lock(left); |
b4ce94de | 977 | btrfs_set_lock_blocking(left); |
5f39d397 | 978 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 979 | parent, pslot - 1, &left); |
54aa1f4d CM |
980 | if (wret) { |
981 | ret = wret; | |
982 | goto enospc; | |
983 | } | |
2cc58cf2 | 984 | } |
5f39d397 CM |
985 | right = read_node_slot(root, parent, pslot + 1); |
986 | if (right) { | |
925baedd | 987 | btrfs_tree_lock(right); |
b4ce94de | 988 | btrfs_set_lock_blocking(right); |
5f39d397 | 989 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 990 | parent, pslot + 1, &right); |
2cc58cf2 CM |
991 | if (wret) { |
992 | ret = wret; | |
993 | goto enospc; | |
994 | } | |
995 | } | |
996 | ||
997 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
998 | if (left) { |
999 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1000 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1001 | if (wret < 0) |
1002 | ret = wret; | |
559af821 | 1003 | btrfs_header_nritems(mid); |
bb803951 | 1004 | } |
79f95c82 CM |
1005 | |
1006 | /* | |
1007 | * then try to empty the right most buffer into the middle | |
1008 | */ | |
5f39d397 | 1009 | if (right) { |
971a1f66 | 1010 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1011 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1012 | ret = wret; |
5f39d397 | 1013 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1014 | clean_tree_block(trans, root, right); |
925baedd | 1015 | btrfs_tree_unlock(right); |
143bede5 | 1016 | del_ptr(trans, root, path, level + 1, pslot + 1); |
f0486c68 | 1017 | root_sub_used(root, right->len); |
66d7e7f0 | 1018 | btrfs_free_tree_block(trans, root, right, 0, 1, 0); |
f0486c68 YZ |
1019 | free_extent_buffer(right); |
1020 | right = NULL; | |
bb803951 | 1021 | } else { |
5f39d397 CM |
1022 | struct btrfs_disk_key right_key; |
1023 | btrfs_node_key(right, &right_key, 0); | |
1024 | btrfs_set_node_key(parent, &right_key, pslot + 1); | |
1025 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1026 | } |
1027 | } | |
5f39d397 | 1028 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1029 | /* |
1030 | * we're not allowed to leave a node with one item in the | |
1031 | * tree during a delete. A deletion from lower in the tree | |
1032 | * could try to delete the only pointer in this node. | |
1033 | * So, pull some keys from the left. | |
1034 | * There has to be a left pointer at this point because | |
1035 | * otherwise we would have pulled some pointers from the | |
1036 | * right | |
1037 | */ | |
5f39d397 CM |
1038 | BUG_ON(!left); |
1039 | wret = balance_node_right(trans, root, mid, left); | |
54aa1f4d | 1040 | if (wret < 0) { |
79f95c82 | 1041 | ret = wret; |
54aa1f4d CM |
1042 | goto enospc; |
1043 | } | |
bce4eae9 CM |
1044 | if (wret == 1) { |
1045 | wret = push_node_left(trans, root, left, mid, 1); | |
1046 | if (wret < 0) | |
1047 | ret = wret; | |
1048 | } | |
79f95c82 CM |
1049 | BUG_ON(wret == 1); |
1050 | } | |
5f39d397 | 1051 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1052 | clean_tree_block(trans, root, mid); |
925baedd | 1053 | btrfs_tree_unlock(mid); |
143bede5 | 1054 | del_ptr(trans, root, path, level + 1, pslot); |
f0486c68 | 1055 | root_sub_used(root, mid->len); |
66d7e7f0 | 1056 | btrfs_free_tree_block(trans, root, mid, 0, 1, 0); |
f0486c68 YZ |
1057 | free_extent_buffer(mid); |
1058 | mid = NULL; | |
79f95c82 CM |
1059 | } else { |
1060 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1061 | struct btrfs_disk_key mid_key; |
1062 | btrfs_node_key(mid, &mid_key, 0); | |
1063 | btrfs_set_node_key(parent, &mid_key, pslot); | |
1064 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1065 | } |
bb803951 | 1066 | |
79f95c82 | 1067 | /* update the path */ |
5f39d397 CM |
1068 | if (left) { |
1069 | if (btrfs_header_nritems(left) > orig_slot) { | |
1070 | extent_buffer_get(left); | |
925baedd | 1071 | /* left was locked after cow */ |
5f39d397 | 1072 | path->nodes[level] = left; |
bb803951 CM |
1073 | path->slots[level + 1] -= 1; |
1074 | path->slots[level] = orig_slot; | |
925baedd CM |
1075 | if (mid) { |
1076 | btrfs_tree_unlock(mid); | |
5f39d397 | 1077 | free_extent_buffer(mid); |
925baedd | 1078 | } |
bb803951 | 1079 | } else { |
5f39d397 | 1080 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1081 | path->slots[level] = orig_slot; |
1082 | } | |
1083 | } | |
79f95c82 | 1084 | /* double check we haven't messed things up */ |
e20d96d6 | 1085 | if (orig_ptr != |
5f39d397 | 1086 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1087 | BUG(); |
54aa1f4d | 1088 | enospc: |
925baedd CM |
1089 | if (right) { |
1090 | btrfs_tree_unlock(right); | |
5f39d397 | 1091 | free_extent_buffer(right); |
925baedd CM |
1092 | } |
1093 | if (left) { | |
1094 | if (path->nodes[level] != left) | |
1095 | btrfs_tree_unlock(left); | |
5f39d397 | 1096 | free_extent_buffer(left); |
925baedd | 1097 | } |
bb803951 CM |
1098 | return ret; |
1099 | } | |
1100 | ||
d352ac68 CM |
1101 | /* Node balancing for insertion. Here we only split or push nodes around |
1102 | * when they are completely full. This is also done top down, so we | |
1103 | * have to be pessimistic. | |
1104 | */ | |
d397712b | 1105 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1106 | struct btrfs_root *root, |
1107 | struct btrfs_path *path, int level) | |
e66f709b | 1108 | { |
5f39d397 CM |
1109 | struct extent_buffer *right = NULL; |
1110 | struct extent_buffer *mid; | |
1111 | struct extent_buffer *left = NULL; | |
1112 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1113 | int ret = 0; |
1114 | int wret; | |
1115 | int pslot; | |
1116 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1117 | |
1118 | if (level == 0) | |
1119 | return 1; | |
1120 | ||
5f39d397 | 1121 | mid = path->nodes[level]; |
7bb86316 | 1122 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1123 | |
a05a9bb1 | 1124 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1125 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1126 | pslot = path->slots[level + 1]; |
1127 | } | |
e66f709b | 1128 | |
5f39d397 | 1129 | if (!parent) |
e66f709b | 1130 | return 1; |
e66f709b | 1131 | |
5f39d397 | 1132 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1133 | |
1134 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1135 | if (left) { |
e66f709b | 1136 | u32 left_nr; |
925baedd CM |
1137 | |
1138 | btrfs_tree_lock(left); | |
b4ce94de CM |
1139 | btrfs_set_lock_blocking(left); |
1140 | ||
5f39d397 | 1141 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1142 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1143 | wret = 1; | |
1144 | } else { | |
5f39d397 | 1145 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1146 | pslot - 1, &left); |
54aa1f4d CM |
1147 | if (ret) |
1148 | wret = 1; | |
1149 | else { | |
54aa1f4d | 1150 | wret = push_node_left(trans, root, |
971a1f66 | 1151 | left, mid, 0); |
54aa1f4d | 1152 | } |
33ade1f8 | 1153 | } |
e66f709b CM |
1154 | if (wret < 0) |
1155 | ret = wret; | |
1156 | if (wret == 0) { | |
5f39d397 | 1157 | struct btrfs_disk_key disk_key; |
e66f709b | 1158 | orig_slot += left_nr; |
5f39d397 CM |
1159 | btrfs_node_key(mid, &disk_key, 0); |
1160 | btrfs_set_node_key(parent, &disk_key, pslot); | |
1161 | btrfs_mark_buffer_dirty(parent); | |
1162 | if (btrfs_header_nritems(left) > orig_slot) { | |
1163 | path->nodes[level] = left; | |
e66f709b CM |
1164 | path->slots[level + 1] -= 1; |
1165 | path->slots[level] = orig_slot; | |
925baedd | 1166 | btrfs_tree_unlock(mid); |
5f39d397 | 1167 | free_extent_buffer(mid); |
e66f709b CM |
1168 | } else { |
1169 | orig_slot -= | |
5f39d397 | 1170 | btrfs_header_nritems(left); |
e66f709b | 1171 | path->slots[level] = orig_slot; |
925baedd | 1172 | btrfs_tree_unlock(left); |
5f39d397 | 1173 | free_extent_buffer(left); |
e66f709b | 1174 | } |
e66f709b CM |
1175 | return 0; |
1176 | } | |
925baedd | 1177 | btrfs_tree_unlock(left); |
5f39d397 | 1178 | free_extent_buffer(left); |
e66f709b | 1179 | } |
925baedd | 1180 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1181 | |
1182 | /* | |
1183 | * then try to empty the right most buffer into the middle | |
1184 | */ | |
5f39d397 | 1185 | if (right) { |
33ade1f8 | 1186 | u32 right_nr; |
b4ce94de | 1187 | |
925baedd | 1188 | btrfs_tree_lock(right); |
b4ce94de CM |
1189 | btrfs_set_lock_blocking(right); |
1190 | ||
5f39d397 | 1191 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1192 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1193 | wret = 1; | |
1194 | } else { | |
5f39d397 CM |
1195 | ret = btrfs_cow_block(trans, root, right, |
1196 | parent, pslot + 1, | |
9fa8cfe7 | 1197 | &right); |
54aa1f4d CM |
1198 | if (ret) |
1199 | wret = 1; | |
1200 | else { | |
54aa1f4d | 1201 | wret = balance_node_right(trans, root, |
5f39d397 | 1202 | right, mid); |
54aa1f4d | 1203 | } |
33ade1f8 | 1204 | } |
e66f709b CM |
1205 | if (wret < 0) |
1206 | ret = wret; | |
1207 | if (wret == 0) { | |
5f39d397 CM |
1208 | struct btrfs_disk_key disk_key; |
1209 | ||
1210 | btrfs_node_key(right, &disk_key, 0); | |
1211 | btrfs_set_node_key(parent, &disk_key, pslot + 1); | |
1212 | btrfs_mark_buffer_dirty(parent); | |
1213 | ||
1214 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1215 | path->nodes[level] = right; | |
e66f709b CM |
1216 | path->slots[level + 1] += 1; |
1217 | path->slots[level] = orig_slot - | |
5f39d397 | 1218 | btrfs_header_nritems(mid); |
925baedd | 1219 | btrfs_tree_unlock(mid); |
5f39d397 | 1220 | free_extent_buffer(mid); |
e66f709b | 1221 | } else { |
925baedd | 1222 | btrfs_tree_unlock(right); |
5f39d397 | 1223 | free_extent_buffer(right); |
e66f709b | 1224 | } |
e66f709b CM |
1225 | return 0; |
1226 | } | |
925baedd | 1227 | btrfs_tree_unlock(right); |
5f39d397 | 1228 | free_extent_buffer(right); |
e66f709b | 1229 | } |
e66f709b CM |
1230 | return 1; |
1231 | } | |
1232 | ||
3c69faec | 1233 | /* |
d352ac68 CM |
1234 | * readahead one full node of leaves, finding things that are close |
1235 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1236 | */ |
c8c42864 CM |
1237 | static void reada_for_search(struct btrfs_root *root, |
1238 | struct btrfs_path *path, | |
1239 | int level, int slot, u64 objectid) | |
3c69faec | 1240 | { |
5f39d397 | 1241 | struct extent_buffer *node; |
01f46658 | 1242 | struct btrfs_disk_key disk_key; |
3c69faec | 1243 | u32 nritems; |
3c69faec | 1244 | u64 search; |
a7175319 | 1245 | u64 target; |
6b80053d | 1246 | u64 nread = 0; |
cb25c2ea | 1247 | u64 gen; |
3c69faec | 1248 | int direction = path->reada; |
5f39d397 | 1249 | struct extent_buffer *eb; |
6b80053d CM |
1250 | u32 nr; |
1251 | u32 blocksize; | |
1252 | u32 nscan = 0; | |
db94535d | 1253 | |
a6b6e75e | 1254 | if (level != 1) |
6702ed49 CM |
1255 | return; |
1256 | ||
1257 | if (!path->nodes[level]) | |
3c69faec CM |
1258 | return; |
1259 | ||
5f39d397 | 1260 | node = path->nodes[level]; |
925baedd | 1261 | |
3c69faec | 1262 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1263 | blocksize = btrfs_level_size(root, level - 1); |
1264 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1265 | if (eb) { |
1266 | free_extent_buffer(eb); | |
3c69faec CM |
1267 | return; |
1268 | } | |
1269 | ||
a7175319 | 1270 | target = search; |
6b80053d | 1271 | |
5f39d397 | 1272 | nritems = btrfs_header_nritems(node); |
6b80053d | 1273 | nr = slot; |
25b8b936 | 1274 | |
d397712b | 1275 | while (1) { |
6b80053d CM |
1276 | if (direction < 0) { |
1277 | if (nr == 0) | |
1278 | break; | |
1279 | nr--; | |
1280 | } else if (direction > 0) { | |
1281 | nr++; | |
1282 | if (nr >= nritems) | |
1283 | break; | |
3c69faec | 1284 | } |
01f46658 CM |
1285 | if (path->reada < 0 && objectid) { |
1286 | btrfs_node_key(node, &disk_key, nr); | |
1287 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1288 | break; | |
1289 | } | |
6b80053d | 1290 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
1291 | if ((search <= target && target - search <= 65536) || |
1292 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 1293 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 1294 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
1295 | nread += blocksize; |
1296 | } | |
1297 | nscan++; | |
a7175319 | 1298 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 1299 | break; |
3c69faec CM |
1300 | } |
1301 | } | |
925baedd | 1302 | |
b4ce94de CM |
1303 | /* |
1304 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
1305 | * cache | |
1306 | */ | |
1307 | static noinline int reada_for_balance(struct btrfs_root *root, | |
1308 | struct btrfs_path *path, int level) | |
1309 | { | |
1310 | int slot; | |
1311 | int nritems; | |
1312 | struct extent_buffer *parent; | |
1313 | struct extent_buffer *eb; | |
1314 | u64 gen; | |
1315 | u64 block1 = 0; | |
1316 | u64 block2 = 0; | |
1317 | int ret = 0; | |
1318 | int blocksize; | |
1319 | ||
8c594ea8 | 1320 | parent = path->nodes[level + 1]; |
b4ce94de CM |
1321 | if (!parent) |
1322 | return 0; | |
1323 | ||
1324 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1325 | slot = path->slots[level + 1]; |
b4ce94de CM |
1326 | blocksize = btrfs_level_size(root, level); |
1327 | ||
1328 | if (slot > 0) { | |
1329 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
1330 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
1331 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
1332 | if (eb && btrfs_buffer_uptodate(eb, gen)) | |
1333 | block1 = 0; | |
1334 | free_extent_buffer(eb); | |
1335 | } | |
8c594ea8 | 1336 | if (slot + 1 < nritems) { |
b4ce94de CM |
1337 | block2 = btrfs_node_blockptr(parent, slot + 1); |
1338 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
1339 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
1340 | if (eb && btrfs_buffer_uptodate(eb, gen)) | |
1341 | block2 = 0; | |
1342 | free_extent_buffer(eb); | |
1343 | } | |
1344 | if (block1 || block2) { | |
1345 | ret = -EAGAIN; | |
8c594ea8 CM |
1346 | |
1347 | /* release the whole path */ | |
b3b4aa74 | 1348 | btrfs_release_path(path); |
8c594ea8 CM |
1349 | |
1350 | /* read the blocks */ | |
b4ce94de CM |
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 | } | |
8c594ea8 | 1360 | if (block2) { |
b4ce94de CM |
1361 | eb = read_tree_block(root, block2, blocksize, 0); |
1362 | free_extent_buffer(eb); | |
1363 | } | |
1364 | } | |
1365 | return ret; | |
1366 | } | |
1367 | ||
1368 | ||
d352ac68 | 1369 | /* |
d397712b CM |
1370 | * 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. | |
d352ac68 | 1374 | * |
d397712b CM |
1375 | * 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. | |
d352ac68 | 1378 | * |
d397712b CM |
1379 | * 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 | |
d352ac68 | 1381 | */ |
e02119d5 CM |
1382 | static noinline void unlock_up(struct btrfs_path *path, int level, |
1383 | int lowest_unlock) | |
925baedd CM |
1384 | { |
1385 | int i; | |
1386 | int skip_level = level; | |
051e1b9f | 1387 | int no_skips = 0; |
925baedd CM |
1388 | 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; | |
051e1b9f | 1395 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1396 | skip_level = i + 1; |
1397 | continue; | |
1398 | } | |
051e1b9f | 1399 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1400 | u32 nritems; |
1401 | t = path->nodes[i]; | |
1402 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1403 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1404 | skip_level = i + 1; |
1405 | continue; | |
1406 | } | |
1407 | } | |
051e1b9f CM |
1408 | if (skip_level < i && i >= lowest_unlock) |
1409 | no_skips = 1; | |
1410 | ||
925baedd CM |
1411 | t = path->nodes[i]; |
1412 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 1413 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd CM |
1414 | path->locks[i] = 0; |
1415 | } | |
1416 | } | |
1417 | } | |
1418 | ||
b4ce94de CM |
1419 | /* |
1420 | * 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 | */ | |
1428 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
1429 | { | |
1430 | int i; | |
1431 | ||
5d4f98a2 | 1432 | if (path->keep_locks) |
b4ce94de CM |
1433 | return; |
1434 | ||
1435 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1436 | if (!path->nodes[i]) | |
12f4dacc | 1437 | continue; |
b4ce94de | 1438 | if (!path->locks[i]) |
12f4dacc | 1439 | continue; |
bd681513 | 1440 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
1441 | path->locks[i] = 0; |
1442 | } | |
1443 | } | |
1444 | ||
c8c42864 CM |
1445 | /* |
1446 | * helper function for btrfs_search_slot. The goal is to find a block | |
1447 | * in cache without setting the path to blocking. If we find the block | |
1448 | * we return zero and the path is unchanged. | |
1449 | * | |
1450 | * If we can't find the block, we set the path blocking and do some | |
1451 | * reada. -EAGAIN is returned and the search must be repeated. | |
1452 | */ | |
1453 | static int | |
1454 | read_block_for_search(struct btrfs_trans_handle *trans, | |
1455 | struct btrfs_root *root, struct btrfs_path *p, | |
1456 | struct extent_buffer **eb_ret, int level, int slot, | |
1457 | struct btrfs_key *key) | |
1458 | { | |
1459 | u64 blocknr; | |
1460 | u64 gen; | |
1461 | u32 blocksize; | |
1462 | struct extent_buffer *b = *eb_ret; | |
1463 | struct extent_buffer *tmp; | |
76a05b35 | 1464 | int ret; |
c8c42864 CM |
1465 | |
1466 | blocknr = btrfs_node_blockptr(b, slot); | |
1467 | gen = btrfs_node_ptr_generation(b, slot); | |
1468 | blocksize = btrfs_level_size(root, level - 1); | |
1469 | ||
1470 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a CM |
1471 | if (tmp) { |
1472 | if (btrfs_buffer_uptodate(tmp, 0)) { | |
1473 | if (btrfs_buffer_uptodate(tmp, gen)) { | |
1474 | /* | |
1475 | * we found an up to date block without | |
1476 | * sleeping, return | |
1477 | * right away | |
1478 | */ | |
1479 | *eb_ret = tmp; | |
1480 | return 0; | |
1481 | } | |
1482 | /* the pages were up to date, but we failed | |
1483 | * the generation number check. Do a full | |
1484 | * read for the generation number that is correct. | |
1485 | * We must do this without dropping locks so | |
1486 | * we can trust our generation number | |
1487 | */ | |
1488 | free_extent_buffer(tmp); | |
bd681513 CM |
1489 | btrfs_set_path_blocking(p); |
1490 | ||
cb44921a CM |
1491 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
1492 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
1493 | *eb_ret = tmp; | |
1494 | return 0; | |
1495 | } | |
1496 | free_extent_buffer(tmp); | |
b3b4aa74 | 1497 | btrfs_release_path(p); |
cb44921a CM |
1498 | return -EIO; |
1499 | } | |
c8c42864 CM |
1500 | } |
1501 | ||
1502 | /* | |
1503 | * reduce lock contention at high levels | |
1504 | * of the btree by dropping locks before | |
76a05b35 CM |
1505 | * we read. Don't release the lock on the current |
1506 | * level because we need to walk this node to figure | |
1507 | * out which blocks to read. | |
c8c42864 | 1508 | */ |
8c594ea8 CM |
1509 | btrfs_unlock_up_safe(p, level + 1); |
1510 | btrfs_set_path_blocking(p); | |
1511 | ||
cb44921a | 1512 | free_extent_buffer(tmp); |
c8c42864 CM |
1513 | if (p->reada) |
1514 | reada_for_search(root, p, level, slot, key->objectid); | |
1515 | ||
b3b4aa74 | 1516 | btrfs_release_path(p); |
76a05b35 CM |
1517 | |
1518 | ret = -EAGAIN; | |
5bdd3536 | 1519 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
1520 | if (tmp) { |
1521 | /* | |
1522 | * If the read above didn't mark this buffer up to date, | |
1523 | * it will never end up being up to date. Set ret to EIO now | |
1524 | * and give up so that our caller doesn't loop forever | |
1525 | * on our EAGAINs. | |
1526 | */ | |
1527 | if (!btrfs_buffer_uptodate(tmp, 0)) | |
1528 | ret = -EIO; | |
c8c42864 | 1529 | free_extent_buffer(tmp); |
76a05b35 CM |
1530 | } |
1531 | return ret; | |
c8c42864 CM |
1532 | } |
1533 | ||
1534 | /* | |
1535 | * helper function for btrfs_search_slot. This does all of the checks | |
1536 | * for node-level blocks and does any balancing required based on | |
1537 | * the ins_len. | |
1538 | * | |
1539 | * If no extra work was required, zero is returned. If we had to | |
1540 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1541 | * start over | |
1542 | */ | |
1543 | static int | |
1544 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1545 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1546 | struct extent_buffer *b, int level, int ins_len, |
1547 | int *write_lock_level) | |
c8c42864 CM |
1548 | { |
1549 | int ret; | |
1550 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
1551 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
1552 | int sret; | |
1553 | ||
bd681513 CM |
1554 | if (*write_lock_level < level + 1) { |
1555 | *write_lock_level = level + 1; | |
1556 | btrfs_release_path(p); | |
1557 | goto again; | |
1558 | } | |
1559 | ||
c8c42864 CM |
1560 | sret = reada_for_balance(root, p, level); |
1561 | if (sret) | |
1562 | goto again; | |
1563 | ||
1564 | btrfs_set_path_blocking(p); | |
1565 | sret = split_node(trans, root, p, level); | |
bd681513 | 1566 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
1567 | |
1568 | BUG_ON(sret > 0); | |
1569 | if (sret) { | |
1570 | ret = sret; | |
1571 | goto done; | |
1572 | } | |
1573 | b = p->nodes[level]; | |
1574 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 1575 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
1576 | int sret; |
1577 | ||
bd681513 CM |
1578 | if (*write_lock_level < level + 1) { |
1579 | *write_lock_level = level + 1; | |
1580 | btrfs_release_path(p); | |
1581 | goto again; | |
1582 | } | |
1583 | ||
c8c42864 CM |
1584 | sret = reada_for_balance(root, p, level); |
1585 | if (sret) | |
1586 | goto again; | |
1587 | ||
1588 | btrfs_set_path_blocking(p); | |
1589 | sret = balance_level(trans, root, p, level); | |
bd681513 | 1590 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
1591 | |
1592 | if (sret) { | |
1593 | ret = sret; | |
1594 | goto done; | |
1595 | } | |
1596 | b = p->nodes[level]; | |
1597 | if (!b) { | |
b3b4aa74 | 1598 | btrfs_release_path(p); |
c8c42864 CM |
1599 | goto again; |
1600 | } | |
1601 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1602 | } | |
1603 | return 0; | |
1604 | ||
1605 | again: | |
1606 | ret = -EAGAIN; | |
1607 | done: | |
1608 | return ret; | |
1609 | } | |
1610 | ||
74123bd7 CM |
1611 | /* |
1612 | * look for key in the tree. path is filled in with nodes along the way | |
1613 | * if key is found, we return zero and you can find the item in the leaf | |
1614 | * level of the path (level 0) | |
1615 | * | |
1616 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
1617 | * be inserted, and 1 is returned. If there are other errors during the |
1618 | * search a negative error number is returned. | |
97571fd0 CM |
1619 | * |
1620 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
1621 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
1622 | * possible) | |
74123bd7 | 1623 | */ |
e089f05c CM |
1624 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
1625 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
1626 | ins_len, int cow) | |
be0e5c09 | 1627 | { |
5f39d397 | 1628 | struct extent_buffer *b; |
be0e5c09 CM |
1629 | int slot; |
1630 | int ret; | |
33c66f43 | 1631 | int err; |
be0e5c09 | 1632 | int level; |
925baedd | 1633 | int lowest_unlock = 1; |
bd681513 CM |
1634 | int root_lock; |
1635 | /* everything at write_lock_level or lower must be write locked */ | |
1636 | int write_lock_level = 0; | |
9f3a7427 CM |
1637 | u8 lowest_level = 0; |
1638 | ||
6702ed49 | 1639 | lowest_level = p->lowest_level; |
323ac95b | 1640 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1641 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 1642 | |
bd681513 | 1643 | if (ins_len < 0) { |
925baedd | 1644 | lowest_unlock = 2; |
65b51a00 | 1645 | |
bd681513 CM |
1646 | /* when we are removing items, we might have to go up to level |
1647 | * two as we update tree pointers Make sure we keep write | |
1648 | * for those levels as well | |
1649 | */ | |
1650 | write_lock_level = 2; | |
1651 | } else if (ins_len > 0) { | |
1652 | /* | |
1653 | * for inserting items, make sure we have a write lock on | |
1654 | * level 1 so we can update keys | |
1655 | */ | |
1656 | write_lock_level = 1; | |
1657 | } | |
1658 | ||
1659 | if (!cow) | |
1660 | write_lock_level = -1; | |
1661 | ||
1662 | if (cow && (p->keep_locks || p->lowest_level)) | |
1663 | write_lock_level = BTRFS_MAX_LEVEL; | |
1664 | ||
bb803951 | 1665 | again: |
bd681513 CM |
1666 | /* |
1667 | * we try very hard to do read locks on the root | |
1668 | */ | |
1669 | root_lock = BTRFS_READ_LOCK; | |
1670 | level = 0; | |
5d4f98a2 | 1671 | if (p->search_commit_root) { |
bd681513 CM |
1672 | /* |
1673 | * the commit roots are read only | |
1674 | * so we always do read locks | |
1675 | */ | |
5d4f98a2 YZ |
1676 | b = root->commit_root; |
1677 | extent_buffer_get(b); | |
bd681513 | 1678 | level = btrfs_header_level(b); |
5d4f98a2 | 1679 | if (!p->skip_locking) |
bd681513 | 1680 | btrfs_tree_read_lock(b); |
5d4f98a2 | 1681 | } else { |
bd681513 | 1682 | if (p->skip_locking) { |
5d4f98a2 | 1683 | b = btrfs_root_node(root); |
bd681513 CM |
1684 | level = btrfs_header_level(b); |
1685 | } else { | |
1686 | /* we don't know the level of the root node | |
1687 | * until we actually have it read locked | |
1688 | */ | |
1689 | b = btrfs_read_lock_root_node(root); | |
1690 | level = btrfs_header_level(b); | |
1691 | if (level <= write_lock_level) { | |
1692 | /* whoops, must trade for write lock */ | |
1693 | btrfs_tree_read_unlock(b); | |
1694 | free_extent_buffer(b); | |
1695 | b = btrfs_lock_root_node(root); | |
1696 | root_lock = BTRFS_WRITE_LOCK; | |
1697 | ||
1698 | /* the level might have changed, check again */ | |
1699 | level = btrfs_header_level(b); | |
1700 | } | |
1701 | } | |
5d4f98a2 | 1702 | } |
bd681513 CM |
1703 | p->nodes[level] = b; |
1704 | if (!p->skip_locking) | |
1705 | p->locks[level] = root_lock; | |
925baedd | 1706 | |
eb60ceac | 1707 | while (b) { |
5f39d397 | 1708 | level = btrfs_header_level(b); |
65b51a00 CM |
1709 | |
1710 | /* | |
1711 | * setup the path here so we can release it under lock | |
1712 | * contention with the cow code | |
1713 | */ | |
02217ed2 | 1714 | if (cow) { |
c8c42864 CM |
1715 | /* |
1716 | * if we don't really need to cow this block | |
1717 | * then we don't want to set the path blocking, | |
1718 | * so we test it here | |
1719 | */ | |
5d4f98a2 | 1720 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 1721 | goto cow_done; |
5d4f98a2 | 1722 | |
b4ce94de CM |
1723 | btrfs_set_path_blocking(p); |
1724 | ||
bd681513 CM |
1725 | /* |
1726 | * must have write locks on this node and the | |
1727 | * parent | |
1728 | */ | |
1729 | if (level + 1 > write_lock_level) { | |
1730 | write_lock_level = level + 1; | |
1731 | btrfs_release_path(p); | |
1732 | goto again; | |
1733 | } | |
1734 | ||
33c66f43 YZ |
1735 | err = btrfs_cow_block(trans, root, b, |
1736 | p->nodes[level + 1], | |
1737 | p->slots[level + 1], &b); | |
1738 | if (err) { | |
33c66f43 | 1739 | ret = err; |
65b51a00 | 1740 | goto done; |
54aa1f4d | 1741 | } |
02217ed2 | 1742 | } |
65b51a00 | 1743 | cow_done: |
02217ed2 | 1744 | BUG_ON(!cow && ins_len); |
65b51a00 | 1745 | |
eb60ceac | 1746 | p->nodes[level] = b; |
bd681513 | 1747 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
1748 | |
1749 | /* | |
1750 | * we have a lock on b and as long as we aren't changing | |
1751 | * the tree, there is no way to for the items in b to change. | |
1752 | * It is safe to drop the lock on our parent before we | |
1753 | * go through the expensive btree search on b. | |
1754 | * | |
1755 | * If cow is true, then we might be changing slot zero, | |
1756 | * which may require changing the parent. So, we can't | |
1757 | * drop the lock until after we know which slot we're | |
1758 | * operating on. | |
1759 | */ | |
1760 | if (!cow) | |
1761 | btrfs_unlock_up_safe(p, level + 1); | |
1762 | ||
5f39d397 | 1763 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 1764 | |
5f39d397 | 1765 | if (level != 0) { |
33c66f43 YZ |
1766 | int dec = 0; |
1767 | if (ret && slot > 0) { | |
1768 | dec = 1; | |
be0e5c09 | 1769 | slot -= 1; |
33c66f43 | 1770 | } |
be0e5c09 | 1771 | p->slots[level] = slot; |
33c66f43 | 1772 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 1773 | ins_len, &write_lock_level); |
33c66f43 | 1774 | if (err == -EAGAIN) |
c8c42864 | 1775 | goto again; |
33c66f43 YZ |
1776 | if (err) { |
1777 | ret = err; | |
c8c42864 | 1778 | goto done; |
33c66f43 | 1779 | } |
c8c42864 CM |
1780 | b = p->nodes[level]; |
1781 | slot = p->slots[level]; | |
b4ce94de | 1782 | |
bd681513 CM |
1783 | /* |
1784 | * slot 0 is special, if we change the key | |
1785 | * we have to update the parent pointer | |
1786 | * which means we must have a write lock | |
1787 | * on the parent | |
1788 | */ | |
1789 | if (slot == 0 && cow && | |
1790 | write_lock_level < level + 1) { | |
1791 | write_lock_level = level + 1; | |
1792 | btrfs_release_path(p); | |
1793 | goto again; | |
1794 | } | |
1795 | ||
f9efa9c7 CM |
1796 | unlock_up(p, level, lowest_unlock); |
1797 | ||
925baedd | 1798 | if (level == lowest_level) { |
33c66f43 YZ |
1799 | if (dec) |
1800 | p->slots[level]++; | |
5b21f2ed | 1801 | goto done; |
925baedd | 1802 | } |
ca7a79ad | 1803 | |
33c66f43 | 1804 | err = read_block_for_search(trans, root, p, |
c8c42864 | 1805 | &b, level, slot, key); |
33c66f43 | 1806 | if (err == -EAGAIN) |
c8c42864 | 1807 | goto again; |
33c66f43 YZ |
1808 | if (err) { |
1809 | ret = err; | |
76a05b35 | 1810 | goto done; |
33c66f43 | 1811 | } |
76a05b35 | 1812 | |
b4ce94de | 1813 | if (!p->skip_locking) { |
bd681513 CM |
1814 | level = btrfs_header_level(b); |
1815 | if (level <= write_lock_level) { | |
1816 | err = btrfs_try_tree_write_lock(b); | |
1817 | if (!err) { | |
1818 | btrfs_set_path_blocking(p); | |
1819 | btrfs_tree_lock(b); | |
1820 | btrfs_clear_path_blocking(p, b, | |
1821 | BTRFS_WRITE_LOCK); | |
1822 | } | |
1823 | p->locks[level] = BTRFS_WRITE_LOCK; | |
1824 | } else { | |
1825 | err = btrfs_try_tree_read_lock(b); | |
1826 | if (!err) { | |
1827 | btrfs_set_path_blocking(p); | |
1828 | btrfs_tree_read_lock(b); | |
1829 | btrfs_clear_path_blocking(p, b, | |
1830 | BTRFS_READ_LOCK); | |
1831 | } | |
1832 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 1833 | } |
bd681513 | 1834 | p->nodes[level] = b; |
b4ce94de | 1835 | } |
be0e5c09 CM |
1836 | } else { |
1837 | p->slots[level] = slot; | |
87b29b20 YZ |
1838 | if (ins_len > 0 && |
1839 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
1840 | if (write_lock_level < 1) { |
1841 | write_lock_level = 1; | |
1842 | btrfs_release_path(p); | |
1843 | goto again; | |
1844 | } | |
1845 | ||
b4ce94de | 1846 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
1847 | err = split_leaf(trans, root, key, |
1848 | p, ins_len, ret == 0); | |
bd681513 | 1849 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 1850 | |
33c66f43 YZ |
1851 | BUG_ON(err > 0); |
1852 | if (err) { | |
1853 | ret = err; | |
65b51a00 CM |
1854 | goto done; |
1855 | } | |
5c680ed6 | 1856 | } |
459931ec CM |
1857 | if (!p->search_for_split) |
1858 | unlock_up(p, level, lowest_unlock); | |
65b51a00 | 1859 | goto done; |
be0e5c09 CM |
1860 | } |
1861 | } | |
65b51a00 CM |
1862 | ret = 1; |
1863 | done: | |
b4ce94de CM |
1864 | /* |
1865 | * we don't really know what they plan on doing with the path | |
1866 | * from here on, so for now just mark it as blocking | |
1867 | */ | |
b9473439 CM |
1868 | if (!p->leave_spinning) |
1869 | btrfs_set_path_blocking(p); | |
76a05b35 | 1870 | if (ret < 0) |
b3b4aa74 | 1871 | btrfs_release_path(p); |
65b51a00 | 1872 | return ret; |
be0e5c09 CM |
1873 | } |
1874 | ||
74123bd7 CM |
1875 | /* |
1876 | * adjust the pointers going up the tree, starting at level | |
1877 | * making sure the right key of each node is points to 'key'. | |
1878 | * This is used after shifting pointers to the left, so it stops | |
1879 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
1880 | * higher levels | |
aa5d6bed | 1881 | * |
74123bd7 | 1882 | */ |
143bede5 JM |
1883 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
1884 | struct btrfs_root *root, struct btrfs_path *path, | |
1885 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
1886 | { |
1887 | int i; | |
5f39d397 CM |
1888 | struct extent_buffer *t; |
1889 | ||
234b63a0 | 1890 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 1891 | int tslot = path->slots[i]; |
eb60ceac | 1892 | if (!path->nodes[i]) |
be0e5c09 | 1893 | break; |
5f39d397 CM |
1894 | t = path->nodes[i]; |
1895 | btrfs_set_node_key(t, key, tslot); | |
d6025579 | 1896 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
1897 | if (tslot != 0) |
1898 | break; | |
1899 | } | |
1900 | } | |
1901 | ||
31840ae1 ZY |
1902 | /* |
1903 | * update item key. | |
1904 | * | |
1905 | * This function isn't completely safe. It's the caller's responsibility | |
1906 | * that the new key won't break the order | |
1907 | */ | |
143bede5 JM |
1908 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
1909 | struct btrfs_root *root, struct btrfs_path *path, | |
1910 | struct btrfs_key *new_key) | |
31840ae1 ZY |
1911 | { |
1912 | struct btrfs_disk_key disk_key; | |
1913 | struct extent_buffer *eb; | |
1914 | int slot; | |
1915 | ||
1916 | eb = path->nodes[0]; | |
1917 | slot = path->slots[0]; | |
1918 | if (slot > 0) { | |
1919 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 1920 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
1921 | } |
1922 | if (slot < btrfs_header_nritems(eb) - 1) { | |
1923 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 1924 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
1925 | } |
1926 | ||
1927 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
1928 | btrfs_set_item_key(eb, &disk_key, slot); | |
1929 | btrfs_mark_buffer_dirty(eb); | |
1930 | if (slot == 0) | |
1931 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
31840ae1 ZY |
1932 | } |
1933 | ||
74123bd7 CM |
1934 | /* |
1935 | * try to push data from one node into the next node left in the | |
79f95c82 | 1936 | * tree. |
aa5d6bed CM |
1937 | * |
1938 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
1939 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 1940 | */ |
98ed5174 CM |
1941 | static int push_node_left(struct btrfs_trans_handle *trans, |
1942 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 1943 | struct extent_buffer *src, int empty) |
be0e5c09 | 1944 | { |
be0e5c09 | 1945 | int push_items = 0; |
bb803951 CM |
1946 | int src_nritems; |
1947 | int dst_nritems; | |
aa5d6bed | 1948 | int ret = 0; |
be0e5c09 | 1949 | |
5f39d397 CM |
1950 | src_nritems = btrfs_header_nritems(src); |
1951 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 1952 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
1953 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
1954 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 1955 | |
bce4eae9 | 1956 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
1957 | return 1; |
1958 | ||
d397712b | 1959 | if (push_items <= 0) |
be0e5c09 CM |
1960 | return 1; |
1961 | ||
bce4eae9 | 1962 | if (empty) { |
971a1f66 | 1963 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
1964 | if (push_items < src_nritems) { |
1965 | /* leave at least 8 pointers in the node if | |
1966 | * we aren't going to empty it | |
1967 | */ | |
1968 | if (src_nritems - push_items < 8) { | |
1969 | if (push_items <= 8) | |
1970 | return 1; | |
1971 | push_items -= 8; | |
1972 | } | |
1973 | } | |
1974 | } else | |
1975 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 1976 | |
5f39d397 CM |
1977 | copy_extent_buffer(dst, src, |
1978 | btrfs_node_key_ptr_offset(dst_nritems), | |
1979 | btrfs_node_key_ptr_offset(0), | |
d397712b | 1980 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 1981 | |
bb803951 | 1982 | if (push_items < src_nritems) { |
5f39d397 CM |
1983 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
1984 | btrfs_node_key_ptr_offset(push_items), | |
1985 | (src_nritems - push_items) * | |
1986 | sizeof(struct btrfs_key_ptr)); | |
1987 | } | |
1988 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
1989 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
1990 | btrfs_mark_buffer_dirty(src); | |
1991 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 1992 | |
79f95c82 CM |
1993 | return ret; |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * try to push data from one node into the next node right in the | |
1998 | * tree. | |
1999 | * | |
2000 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2001 | * error, and > 0 if there was no room in the right hand block. | |
2002 | * | |
2003 | * this will only push up to 1/2 the contents of the left node over | |
2004 | */ | |
5f39d397 CM |
2005 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2006 | struct btrfs_root *root, | |
2007 | struct extent_buffer *dst, | |
2008 | struct extent_buffer *src) | |
79f95c82 | 2009 | { |
79f95c82 CM |
2010 | int push_items = 0; |
2011 | int max_push; | |
2012 | int src_nritems; | |
2013 | int dst_nritems; | |
2014 | int ret = 0; | |
79f95c82 | 2015 | |
7bb86316 CM |
2016 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2017 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2018 | ||
5f39d397 CM |
2019 | src_nritems = btrfs_header_nritems(src); |
2020 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2021 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 2022 | if (push_items <= 0) |
79f95c82 | 2023 | return 1; |
bce4eae9 | 2024 | |
d397712b | 2025 | if (src_nritems < 4) |
bce4eae9 | 2026 | return 1; |
79f95c82 CM |
2027 | |
2028 | max_push = src_nritems / 2 + 1; | |
2029 | /* don't try to empty the node */ | |
d397712b | 2030 | if (max_push >= src_nritems) |
79f95c82 | 2031 | return 1; |
252c38f0 | 2032 | |
79f95c82 CM |
2033 | if (max_push < push_items) |
2034 | push_items = max_push; | |
2035 | ||
5f39d397 CM |
2036 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2037 | btrfs_node_key_ptr_offset(0), | |
2038 | (dst_nritems) * | |
2039 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2040 | |
5f39d397 CM |
2041 | copy_extent_buffer(dst, src, |
2042 | btrfs_node_key_ptr_offset(0), | |
2043 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2044 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2045 | |
5f39d397 CM |
2046 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2047 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2048 | |
5f39d397 CM |
2049 | btrfs_mark_buffer_dirty(src); |
2050 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2051 | |
aa5d6bed | 2052 | return ret; |
be0e5c09 CM |
2053 | } |
2054 | ||
97571fd0 CM |
2055 | /* |
2056 | * helper function to insert a new root level in the tree. | |
2057 | * A new node is allocated, and a single item is inserted to | |
2058 | * point to the existing root | |
aa5d6bed CM |
2059 | * |
2060 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2061 | */ |
d397712b | 2062 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2063 | struct btrfs_root *root, |
2064 | struct btrfs_path *path, int level) | |
5c680ed6 | 2065 | { |
7bb86316 | 2066 | u64 lower_gen; |
5f39d397 CM |
2067 | struct extent_buffer *lower; |
2068 | struct extent_buffer *c; | |
925baedd | 2069 | struct extent_buffer *old; |
5f39d397 | 2070 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
2071 | |
2072 | BUG_ON(path->nodes[level]); | |
2073 | BUG_ON(path->nodes[level-1] != root->node); | |
2074 | ||
7bb86316 CM |
2075 | lower = path->nodes[level-1]; |
2076 | if (level == 1) | |
2077 | btrfs_item_key(lower, &lower_key, 0); | |
2078 | else | |
2079 | btrfs_node_key(lower, &lower_key, 0); | |
2080 | ||
31840ae1 | 2081 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 2082 | root->root_key.objectid, &lower_key, |
66d7e7f0 | 2083 | level, root->node->start, 0, 0); |
5f39d397 CM |
2084 | if (IS_ERR(c)) |
2085 | return PTR_ERR(c); | |
925baedd | 2086 | |
f0486c68 YZ |
2087 | root_add_used(root, root->nodesize); |
2088 | ||
5d4f98a2 | 2089 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
2090 | btrfs_set_header_nritems(c, 1); |
2091 | btrfs_set_header_level(c, level); | |
db94535d | 2092 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 2093 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 2094 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 2095 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
2096 | |
2097 | write_extent_buffer(c, root->fs_info->fsid, | |
2098 | (unsigned long)btrfs_header_fsid(c), | |
2099 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2100 | |
2101 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
2102 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
2103 | BTRFS_UUID_SIZE); | |
2104 | ||
5f39d397 | 2105 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2106 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2107 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2108 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2109 | |
2110 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2111 | |
5f39d397 | 2112 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2113 | |
925baedd | 2114 | old = root->node; |
240f62c8 | 2115 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2116 | |
2117 | /* the super has an extra ref to root->node */ | |
2118 | free_extent_buffer(old); | |
2119 | ||
0b86a832 | 2120 | add_root_to_dirty_list(root); |
5f39d397 CM |
2121 | extent_buffer_get(c); |
2122 | path->nodes[level] = c; | |
bd681513 | 2123 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2124 | path->slots[level] = 0; |
2125 | return 0; | |
2126 | } | |
2127 | ||
74123bd7 CM |
2128 | /* |
2129 | * worker function to insert a single pointer in a node. | |
2130 | * the node should have enough room for the pointer already | |
97571fd0 | 2131 | * |
74123bd7 CM |
2132 | * slot and level indicate where you want the key to go, and |
2133 | * blocknr is the block the key points to. | |
2134 | */ | |
143bede5 JM |
2135 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2136 | struct btrfs_root *root, struct btrfs_path *path, | |
2137 | struct btrfs_disk_key *key, u64 bytenr, | |
2138 | int slot, int level) | |
74123bd7 | 2139 | { |
5f39d397 | 2140 | struct extent_buffer *lower; |
74123bd7 | 2141 | int nritems; |
5c680ed6 CM |
2142 | |
2143 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 2144 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
2145 | lower = path->nodes[level]; |
2146 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2147 | BUG_ON(slot > nritems); |
143bede5 | 2148 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 2149 | if (slot != nritems) { |
5f39d397 CM |
2150 | memmove_extent_buffer(lower, |
2151 | btrfs_node_key_ptr_offset(slot + 1), | |
2152 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2153 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2154 | } |
5f39d397 | 2155 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2156 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2157 | WARN_ON(trans->transid == 0); |
2158 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2159 | btrfs_set_header_nritems(lower, nritems + 1); |
2160 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2161 | } |
2162 | ||
97571fd0 CM |
2163 | /* |
2164 | * split the node at the specified level in path in two. | |
2165 | * The path is corrected to point to the appropriate node after the split | |
2166 | * | |
2167 | * Before splitting this tries to make some room in the node by pushing | |
2168 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2169 | * |
2170 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2171 | */ |
e02119d5 CM |
2172 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2173 | struct btrfs_root *root, | |
2174 | struct btrfs_path *path, int level) | |
be0e5c09 | 2175 | { |
5f39d397 CM |
2176 | struct extent_buffer *c; |
2177 | struct extent_buffer *split; | |
2178 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2179 | int mid; |
5c680ed6 | 2180 | int ret; |
7518a238 | 2181 | u32 c_nritems; |
eb60ceac | 2182 | |
5f39d397 | 2183 | c = path->nodes[level]; |
7bb86316 | 2184 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2185 | if (c == root->node) { |
5c680ed6 | 2186 | /* trying to split the root, lets make a new one */ |
e089f05c | 2187 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2188 | if (ret) |
2189 | return ret; | |
b3612421 | 2190 | } else { |
e66f709b | 2191 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2192 | c = path->nodes[level]; |
2193 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2194 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2195 | return 0; |
54aa1f4d CM |
2196 | if (ret < 0) |
2197 | return ret; | |
be0e5c09 | 2198 | } |
e66f709b | 2199 | |
5f39d397 | 2200 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2201 | mid = (c_nritems + 1) / 2; |
2202 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2203 | |
5d4f98a2 | 2204 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 2205 | root->root_key.objectid, |
66d7e7f0 | 2206 | &disk_key, level, c->start, 0, 0); |
5f39d397 CM |
2207 | if (IS_ERR(split)) |
2208 | return PTR_ERR(split); | |
2209 | ||
f0486c68 YZ |
2210 | root_add_used(root, root->nodesize); |
2211 | ||
5d4f98a2 | 2212 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 2213 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 2214 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 2215 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 2216 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2217 | btrfs_set_header_owner(split, root->root_key.objectid); |
2218 | write_extent_buffer(split, root->fs_info->fsid, | |
2219 | (unsigned long)btrfs_header_fsid(split), | |
2220 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2221 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2222 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2223 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2224 | |
5f39d397 CM |
2225 | |
2226 | copy_extent_buffer(split, c, | |
2227 | btrfs_node_key_ptr_offset(0), | |
2228 | btrfs_node_key_ptr_offset(mid), | |
2229 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2230 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2231 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2232 | ret = 0; |
2233 | ||
5f39d397 CM |
2234 | btrfs_mark_buffer_dirty(c); |
2235 | btrfs_mark_buffer_dirty(split); | |
2236 | ||
143bede5 JM |
2237 | insert_ptr(trans, root, path, &disk_key, split->start, |
2238 | path->slots[level + 1] + 1, level + 1); | |
aa5d6bed | 2239 | |
5de08d7d | 2240 | if (path->slots[level] >= mid) { |
5c680ed6 | 2241 | path->slots[level] -= mid; |
925baedd | 2242 | btrfs_tree_unlock(c); |
5f39d397 CM |
2243 | free_extent_buffer(c); |
2244 | path->nodes[level] = split; | |
5c680ed6 CM |
2245 | path->slots[level + 1] += 1; |
2246 | } else { | |
925baedd | 2247 | btrfs_tree_unlock(split); |
5f39d397 | 2248 | free_extent_buffer(split); |
be0e5c09 | 2249 | } |
aa5d6bed | 2250 | return ret; |
be0e5c09 CM |
2251 | } |
2252 | ||
74123bd7 CM |
2253 | /* |
2254 | * how many bytes are required to store the items in a leaf. start | |
2255 | * and nr indicate which items in the leaf to check. This totals up the | |
2256 | * space used both by the item structs and the item data | |
2257 | */ | |
5f39d397 | 2258 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2259 | { |
2260 | int data_len; | |
5f39d397 | 2261 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2262 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2263 | |
2264 | if (!nr) | |
2265 | return 0; | |
5f39d397 CM |
2266 | data_len = btrfs_item_end_nr(l, start); |
2267 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2268 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2269 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2270 | return data_len; |
2271 | } | |
2272 | ||
d4dbff95 CM |
2273 | /* |
2274 | * The space between the end of the leaf items and | |
2275 | * the start of the leaf data. IOW, how much room | |
2276 | * the leaf has left for both items and data | |
2277 | */ | |
d397712b | 2278 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 2279 | struct extent_buffer *leaf) |
d4dbff95 | 2280 | { |
5f39d397 CM |
2281 | int nritems = btrfs_header_nritems(leaf); |
2282 | int ret; | |
2283 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2284 | if (ret < 0) { | |
d397712b CM |
2285 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
2286 | "used %d nritems %d\n", | |
ae2f5411 | 2287 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2288 | leaf_space_used(leaf, 0, nritems), nritems); |
2289 | } | |
2290 | return ret; | |
d4dbff95 CM |
2291 | } |
2292 | ||
99d8f83c CM |
2293 | /* |
2294 | * min slot controls the lowest index we're willing to push to the | |
2295 | * right. We'll push up to and including min_slot, but no lower | |
2296 | */ | |
44871b1b CM |
2297 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
2298 | struct btrfs_root *root, | |
2299 | struct btrfs_path *path, | |
2300 | int data_size, int empty, | |
2301 | struct extent_buffer *right, | |
99d8f83c CM |
2302 | int free_space, u32 left_nritems, |
2303 | u32 min_slot) | |
00ec4c51 | 2304 | { |
5f39d397 | 2305 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2306 | struct extent_buffer *upper = path->nodes[1]; |
5f39d397 | 2307 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2308 | int slot; |
34a38218 | 2309 | u32 i; |
00ec4c51 CM |
2310 | int push_space = 0; |
2311 | int push_items = 0; | |
0783fcfc | 2312 | struct btrfs_item *item; |
34a38218 | 2313 | u32 nr; |
7518a238 | 2314 | u32 right_nritems; |
5f39d397 | 2315 | u32 data_end; |
db94535d | 2316 | u32 this_item_size; |
00ec4c51 | 2317 | |
34a38218 CM |
2318 | if (empty) |
2319 | nr = 0; | |
2320 | else | |
99d8f83c | 2321 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2322 | |
31840ae1 | 2323 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2324 | push_space += data_size; |
31840ae1 | 2325 | |
44871b1b | 2326 | slot = path->slots[1]; |
34a38218 CM |
2327 | i = left_nritems - 1; |
2328 | while (i >= nr) { | |
5f39d397 | 2329 | item = btrfs_item_nr(left, i); |
db94535d | 2330 | |
31840ae1 ZY |
2331 | if (!empty && push_items > 0) { |
2332 | if (path->slots[0] > i) | |
2333 | break; | |
2334 | if (path->slots[0] == i) { | |
2335 | int space = btrfs_leaf_free_space(root, left); | |
2336 | if (space + push_space * 2 > free_space) | |
2337 | break; | |
2338 | } | |
2339 | } | |
2340 | ||
00ec4c51 | 2341 | if (path->slots[0] == i) |
87b29b20 | 2342 | push_space += data_size; |
db94535d | 2343 | |
db94535d CM |
2344 | this_item_size = btrfs_item_size(left, item); |
2345 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2346 | break; |
31840ae1 | 2347 | |
00ec4c51 | 2348 | push_items++; |
db94535d | 2349 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2350 | if (i == 0) |
2351 | break; | |
2352 | i--; | |
db94535d | 2353 | } |
5f39d397 | 2354 | |
925baedd CM |
2355 | if (push_items == 0) |
2356 | goto out_unlock; | |
5f39d397 | 2357 | |
34a38218 | 2358 | if (!empty && push_items == left_nritems) |
a429e513 | 2359 | WARN_ON(1); |
5f39d397 | 2360 | |
00ec4c51 | 2361 | /* push left to right */ |
5f39d397 | 2362 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2363 | |
5f39d397 | 2364 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2365 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2366 | |
00ec4c51 | 2367 | /* make room in the right data area */ |
5f39d397 CM |
2368 | data_end = leaf_data_end(root, right); |
2369 | memmove_extent_buffer(right, | |
2370 | btrfs_leaf_data(right) + data_end - push_space, | |
2371 | btrfs_leaf_data(right) + data_end, | |
2372 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2373 | ||
00ec4c51 | 2374 | /* copy from the left data area */ |
5f39d397 | 2375 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2376 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2377 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2378 | push_space); | |
5f39d397 CM |
2379 | |
2380 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2381 | btrfs_item_nr_offset(0), | |
2382 | right_nritems * sizeof(struct btrfs_item)); | |
2383 | ||
00ec4c51 | 2384 | /* copy the items from left to right */ |
5f39d397 CM |
2385 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2386 | btrfs_item_nr_offset(left_nritems - push_items), | |
2387 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2388 | |
2389 | /* update the item pointers */ | |
7518a238 | 2390 | right_nritems += push_items; |
5f39d397 | 2391 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2392 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2393 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2394 | item = btrfs_item_nr(right, i); |
db94535d CM |
2395 | push_space -= btrfs_item_size(right, item); |
2396 | btrfs_set_item_offset(right, item, push_space); | |
2397 | } | |
2398 | ||
7518a238 | 2399 | left_nritems -= push_items; |
5f39d397 | 2400 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2401 | |
34a38218 CM |
2402 | if (left_nritems) |
2403 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
2404 | else |
2405 | clean_tree_block(trans, root, left); | |
2406 | ||
5f39d397 | 2407 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2408 | |
5f39d397 CM |
2409 | btrfs_item_key(right, &disk_key, 0); |
2410 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2411 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2412 | |
00ec4c51 | 2413 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2414 | if (path->slots[0] >= left_nritems) { |
2415 | path->slots[0] -= left_nritems; | |
925baedd CM |
2416 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2417 | clean_tree_block(trans, root, path->nodes[0]); | |
2418 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2419 | free_extent_buffer(path->nodes[0]); |
2420 | path->nodes[0] = right; | |
00ec4c51 CM |
2421 | path->slots[1] += 1; |
2422 | } else { | |
925baedd | 2423 | btrfs_tree_unlock(right); |
5f39d397 | 2424 | free_extent_buffer(right); |
00ec4c51 CM |
2425 | } |
2426 | return 0; | |
925baedd CM |
2427 | |
2428 | out_unlock: | |
2429 | btrfs_tree_unlock(right); | |
2430 | free_extent_buffer(right); | |
2431 | return 1; | |
00ec4c51 | 2432 | } |
925baedd | 2433 | |
44871b1b CM |
2434 | /* |
2435 | * push some data in the path leaf to the right, trying to free up at | |
2436 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2437 | * | |
2438 | * returns 1 if the push failed because the other node didn't have enough | |
2439 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
2440 | * |
2441 | * this will push starting from min_slot to the end of the leaf. It won't | |
2442 | * push any slot lower than min_slot | |
44871b1b CM |
2443 | */ |
2444 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2445 | *root, struct btrfs_path *path, |
2446 | int min_data_size, int data_size, | |
2447 | int empty, u32 min_slot) | |
44871b1b CM |
2448 | { |
2449 | struct extent_buffer *left = path->nodes[0]; | |
2450 | struct extent_buffer *right; | |
2451 | struct extent_buffer *upper; | |
2452 | int slot; | |
2453 | int free_space; | |
2454 | u32 left_nritems; | |
2455 | int ret; | |
2456 | ||
2457 | if (!path->nodes[1]) | |
2458 | return 1; | |
2459 | ||
2460 | slot = path->slots[1]; | |
2461 | upper = path->nodes[1]; | |
2462 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2463 | return 1; | |
2464 | ||
2465 | btrfs_assert_tree_locked(path->nodes[1]); | |
2466 | ||
2467 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
2468 | if (right == NULL) |
2469 | return 1; | |
2470 | ||
44871b1b CM |
2471 | btrfs_tree_lock(right); |
2472 | btrfs_set_lock_blocking(right); | |
2473 | ||
2474 | free_space = btrfs_leaf_free_space(root, right); | |
2475 | if (free_space < data_size) | |
2476 | goto out_unlock; | |
2477 | ||
2478 | /* cow and double check */ | |
2479 | ret = btrfs_cow_block(trans, root, right, upper, | |
2480 | slot + 1, &right); | |
2481 | if (ret) | |
2482 | goto out_unlock; | |
2483 | ||
2484 | free_space = btrfs_leaf_free_space(root, right); | |
2485 | if (free_space < data_size) | |
2486 | goto out_unlock; | |
2487 | ||
2488 | left_nritems = btrfs_header_nritems(left); | |
2489 | if (left_nritems == 0) | |
2490 | goto out_unlock; | |
2491 | ||
99d8f83c CM |
2492 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
2493 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
2494 | out_unlock: |
2495 | btrfs_tree_unlock(right); | |
2496 | free_extent_buffer(right); | |
2497 | return 1; | |
2498 | } | |
2499 | ||
74123bd7 CM |
2500 | /* |
2501 | * push some data in the path leaf to the left, trying to free up at | |
2502 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2503 | * |
2504 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2505 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
2506 | * items | |
74123bd7 | 2507 | */ |
44871b1b CM |
2508 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
2509 | struct btrfs_root *root, | |
2510 | struct btrfs_path *path, int data_size, | |
2511 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
2512 | int free_space, u32 right_nritems, |
2513 | u32 max_slot) | |
be0e5c09 | 2514 | { |
5f39d397 CM |
2515 | struct btrfs_disk_key disk_key; |
2516 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 2517 | int i; |
be0e5c09 CM |
2518 | int push_space = 0; |
2519 | int push_items = 0; | |
0783fcfc | 2520 | struct btrfs_item *item; |
7518a238 | 2521 | u32 old_left_nritems; |
34a38218 | 2522 | u32 nr; |
aa5d6bed | 2523 | int ret = 0; |
db94535d CM |
2524 | u32 this_item_size; |
2525 | u32 old_left_item_size; | |
be0e5c09 | 2526 | |
34a38218 | 2527 | if (empty) |
99d8f83c | 2528 | nr = min(right_nritems, max_slot); |
34a38218 | 2529 | else |
99d8f83c | 2530 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
2531 | |
2532 | for (i = 0; i < nr; i++) { | |
5f39d397 | 2533 | item = btrfs_item_nr(right, i); |
db94535d | 2534 | |
31840ae1 ZY |
2535 | if (!empty && push_items > 0) { |
2536 | if (path->slots[0] < i) | |
2537 | break; | |
2538 | if (path->slots[0] == i) { | |
2539 | int space = btrfs_leaf_free_space(root, right); | |
2540 | if (space + push_space * 2 > free_space) | |
2541 | break; | |
2542 | } | |
2543 | } | |
2544 | ||
be0e5c09 | 2545 | if (path->slots[0] == i) |
87b29b20 | 2546 | push_space += data_size; |
db94535d CM |
2547 | |
2548 | this_item_size = btrfs_item_size(right, item); | |
2549 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 2550 | break; |
db94535d | 2551 | |
be0e5c09 | 2552 | push_items++; |
db94535d CM |
2553 | push_space += this_item_size + sizeof(*item); |
2554 | } | |
2555 | ||
be0e5c09 | 2556 | if (push_items == 0) { |
925baedd CM |
2557 | ret = 1; |
2558 | goto out; | |
be0e5c09 | 2559 | } |
34a38218 | 2560 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 2561 | WARN_ON(1); |
5f39d397 | 2562 | |
be0e5c09 | 2563 | /* push data from right to left */ |
5f39d397 CM |
2564 | copy_extent_buffer(left, right, |
2565 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
2566 | btrfs_item_nr_offset(0), | |
2567 | push_items * sizeof(struct btrfs_item)); | |
2568 | ||
123abc88 | 2569 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 2570 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
2571 | |
2572 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
2573 | leaf_data_end(root, left) - push_space, |
2574 | btrfs_leaf_data(right) + | |
5f39d397 | 2575 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 2576 | push_space); |
5f39d397 | 2577 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 2578 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 2579 | |
db94535d | 2580 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 2581 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 2582 | u32 ioff; |
db94535d | 2583 | |
5f39d397 | 2584 | item = btrfs_item_nr(left, i); |
db94535d | 2585 | |
5f39d397 CM |
2586 | ioff = btrfs_item_offset(left, item); |
2587 | btrfs_set_item_offset(left, item, | |
db94535d | 2588 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size)); |
be0e5c09 | 2589 | } |
5f39d397 | 2590 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
2591 | |
2592 | /* fixup right node */ | |
34a38218 | 2593 | if (push_items > right_nritems) { |
d397712b CM |
2594 | printk(KERN_CRIT "push items %d nr %u\n", push_items, |
2595 | right_nritems); | |
34a38218 CM |
2596 | WARN_ON(1); |
2597 | } | |
2598 | ||
2599 | if (push_items < right_nritems) { | |
2600 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2601 | leaf_data_end(root, right); | |
2602 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
2603 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
2604 | btrfs_leaf_data(right) + | |
2605 | leaf_data_end(root, right), push_space); | |
2606 | ||
2607 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
2608 | btrfs_item_nr_offset(push_items), |
2609 | (btrfs_header_nritems(right) - push_items) * | |
2610 | sizeof(struct btrfs_item)); | |
34a38218 | 2611 | } |
eef1c494 Y |
2612 | right_nritems -= push_items; |
2613 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 2614 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
2615 | for (i = 0; i < right_nritems; i++) { |
2616 | item = btrfs_item_nr(right, i); | |
db94535d | 2617 | |
db94535d CM |
2618 | push_space = push_space - btrfs_item_size(right, item); |
2619 | btrfs_set_item_offset(right, item, push_space); | |
2620 | } | |
eb60ceac | 2621 | |
5f39d397 | 2622 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
2623 | if (right_nritems) |
2624 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
2625 | else |
2626 | clean_tree_block(trans, root, right); | |
098f59c2 | 2627 | |
5f39d397 | 2628 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 2629 | fixup_low_keys(trans, root, path, &disk_key, 1); |
be0e5c09 CM |
2630 | |
2631 | /* then fixup the leaf pointer in the path */ | |
2632 | if (path->slots[0] < push_items) { | |
2633 | path->slots[0] += old_left_nritems; | |
925baedd | 2634 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2635 | free_extent_buffer(path->nodes[0]); |
2636 | path->nodes[0] = left; | |
be0e5c09 CM |
2637 | path->slots[1] -= 1; |
2638 | } else { | |
925baedd | 2639 | btrfs_tree_unlock(left); |
5f39d397 | 2640 | free_extent_buffer(left); |
be0e5c09 CM |
2641 | path->slots[0] -= push_items; |
2642 | } | |
eb60ceac | 2643 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 2644 | return ret; |
925baedd CM |
2645 | out: |
2646 | btrfs_tree_unlock(left); | |
2647 | free_extent_buffer(left); | |
2648 | return ret; | |
be0e5c09 CM |
2649 | } |
2650 | ||
44871b1b CM |
2651 | /* |
2652 | * push some data in the path leaf to the left, trying to free up at | |
2653 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2654 | * |
2655 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2656 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
2657 | * items | |
44871b1b CM |
2658 | */ |
2659 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2660 | *root, struct btrfs_path *path, int min_data_size, |
2661 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
2662 | { |
2663 | struct extent_buffer *right = path->nodes[0]; | |
2664 | struct extent_buffer *left; | |
2665 | int slot; | |
2666 | int free_space; | |
2667 | u32 right_nritems; | |
2668 | int ret = 0; | |
2669 | ||
2670 | slot = path->slots[1]; | |
2671 | if (slot == 0) | |
2672 | return 1; | |
2673 | if (!path->nodes[1]) | |
2674 | return 1; | |
2675 | ||
2676 | right_nritems = btrfs_header_nritems(right); | |
2677 | if (right_nritems == 0) | |
2678 | return 1; | |
2679 | ||
2680 | btrfs_assert_tree_locked(path->nodes[1]); | |
2681 | ||
2682 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
2683 | if (left == NULL) |
2684 | return 1; | |
2685 | ||
44871b1b CM |
2686 | btrfs_tree_lock(left); |
2687 | btrfs_set_lock_blocking(left); | |
2688 | ||
2689 | free_space = btrfs_leaf_free_space(root, left); | |
2690 | if (free_space < data_size) { | |
2691 | ret = 1; | |
2692 | goto out; | |
2693 | } | |
2694 | ||
2695 | /* cow and double check */ | |
2696 | ret = btrfs_cow_block(trans, root, left, | |
2697 | path->nodes[1], slot - 1, &left); | |
2698 | if (ret) { | |
2699 | /* we hit -ENOSPC, but it isn't fatal here */ | |
2700 | ret = 1; | |
2701 | goto out; | |
2702 | } | |
2703 | ||
2704 | free_space = btrfs_leaf_free_space(root, left); | |
2705 | if (free_space < data_size) { | |
2706 | ret = 1; | |
2707 | goto out; | |
2708 | } | |
2709 | ||
99d8f83c CM |
2710 | return __push_leaf_left(trans, root, path, min_data_size, |
2711 | empty, left, free_space, right_nritems, | |
2712 | max_slot); | |
44871b1b CM |
2713 | out: |
2714 | btrfs_tree_unlock(left); | |
2715 | free_extent_buffer(left); | |
2716 | return ret; | |
2717 | } | |
2718 | ||
2719 | /* | |
2720 | * split the path's leaf in two, making sure there is at least data_size | |
2721 | * available for the resulting leaf level of the path. | |
44871b1b | 2722 | */ |
143bede5 JM |
2723 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2724 | struct btrfs_root *root, | |
2725 | struct btrfs_path *path, | |
2726 | struct extent_buffer *l, | |
2727 | struct extent_buffer *right, | |
2728 | int slot, int mid, int nritems) | |
44871b1b CM |
2729 | { |
2730 | int data_copy_size; | |
2731 | int rt_data_off; | |
2732 | int i; | |
44871b1b CM |
2733 | struct btrfs_disk_key disk_key; |
2734 | ||
2735 | nritems = nritems - mid; | |
2736 | btrfs_set_header_nritems(right, nritems); | |
2737 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
2738 | ||
2739 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
2740 | btrfs_item_nr_offset(mid), | |
2741 | nritems * sizeof(struct btrfs_item)); | |
2742 | ||
2743 | copy_extent_buffer(right, l, | |
2744 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
2745 | data_copy_size, btrfs_leaf_data(l) + | |
2746 | leaf_data_end(root, l), data_copy_size); | |
2747 | ||
2748 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
2749 | btrfs_item_end_nr(l, mid); | |
2750 | ||
2751 | for (i = 0; i < nritems; i++) { | |
2752 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
2753 | u32 ioff; | |
2754 | ||
44871b1b CM |
2755 | ioff = btrfs_item_offset(right, item); |
2756 | btrfs_set_item_offset(right, item, ioff + rt_data_off); | |
2757 | } | |
2758 | ||
44871b1b | 2759 | btrfs_set_header_nritems(l, mid); |
44871b1b | 2760 | btrfs_item_key(right, &disk_key, 0); |
143bede5 JM |
2761 | insert_ptr(trans, root, path, &disk_key, right->start, |
2762 | path->slots[1] + 1, 1); | |
44871b1b CM |
2763 | |
2764 | btrfs_mark_buffer_dirty(right); | |
2765 | btrfs_mark_buffer_dirty(l); | |
2766 | BUG_ON(path->slots[0] != slot); | |
2767 | ||
44871b1b CM |
2768 | if (mid <= slot) { |
2769 | btrfs_tree_unlock(path->nodes[0]); | |
2770 | free_extent_buffer(path->nodes[0]); | |
2771 | path->nodes[0] = right; | |
2772 | path->slots[0] -= mid; | |
2773 | path->slots[1] += 1; | |
2774 | } else { | |
2775 | btrfs_tree_unlock(right); | |
2776 | free_extent_buffer(right); | |
2777 | } | |
2778 | ||
2779 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
2780 | } |
2781 | ||
99d8f83c CM |
2782 | /* |
2783 | * double splits happen when we need to insert a big item in the middle | |
2784 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
2785 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
2786 | * A B C | |
2787 | * | |
2788 | * We avoid this by trying to push the items on either side of our target | |
2789 | * into the adjacent leaves. If all goes well we can avoid the double split | |
2790 | * completely. | |
2791 | */ | |
2792 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
2793 | struct btrfs_root *root, | |
2794 | struct btrfs_path *path, | |
2795 | int data_size) | |
2796 | { | |
2797 | int ret; | |
2798 | int progress = 0; | |
2799 | int slot; | |
2800 | u32 nritems; | |
2801 | ||
2802 | slot = path->slots[0]; | |
2803 | ||
2804 | /* | |
2805 | * try to push all the items after our slot into the | |
2806 | * right leaf | |
2807 | */ | |
2808 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
2809 | if (ret < 0) | |
2810 | return ret; | |
2811 | ||
2812 | if (ret == 0) | |
2813 | progress++; | |
2814 | ||
2815 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2816 | /* | |
2817 | * our goal is to get our slot at the start or end of a leaf. If | |
2818 | * we've done so we're done | |
2819 | */ | |
2820 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
2821 | return 0; | |
2822 | ||
2823 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
2824 | return 0; | |
2825 | ||
2826 | /* try to push all the items before our slot into the next leaf */ | |
2827 | slot = path->slots[0]; | |
2828 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
2829 | if (ret < 0) | |
2830 | return ret; | |
2831 | ||
2832 | if (ret == 0) | |
2833 | progress++; | |
2834 | ||
2835 | if (progress) | |
2836 | return 0; | |
2837 | return 1; | |
2838 | } | |
2839 | ||
74123bd7 CM |
2840 | /* |
2841 | * split the path's leaf in two, making sure there is at least data_size | |
2842 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
2843 | * |
2844 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 2845 | */ |
e02119d5 CM |
2846 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
2847 | struct btrfs_root *root, | |
2848 | struct btrfs_key *ins_key, | |
2849 | struct btrfs_path *path, int data_size, | |
2850 | int extend) | |
be0e5c09 | 2851 | { |
5d4f98a2 | 2852 | struct btrfs_disk_key disk_key; |
5f39d397 | 2853 | struct extent_buffer *l; |
7518a238 | 2854 | u32 nritems; |
eb60ceac CM |
2855 | int mid; |
2856 | int slot; | |
5f39d397 | 2857 | struct extent_buffer *right; |
d4dbff95 | 2858 | int ret = 0; |
aa5d6bed | 2859 | int wret; |
5d4f98a2 | 2860 | int split; |
cc0c5538 | 2861 | int num_doubles = 0; |
99d8f83c | 2862 | int tried_avoid_double = 0; |
aa5d6bed | 2863 | |
a5719521 YZ |
2864 | l = path->nodes[0]; |
2865 | slot = path->slots[0]; | |
2866 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
2867 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
2868 | return -EOVERFLOW; | |
2869 | ||
40689478 | 2870 | /* first try to make some room by pushing left and right */ |
99d8f83c CM |
2871 | if (data_size) { |
2872 | wret = push_leaf_right(trans, root, path, data_size, | |
2873 | data_size, 0, 0); | |
d397712b | 2874 | if (wret < 0) |
eaee50e8 | 2875 | return wret; |
3685f791 | 2876 | if (wret) { |
99d8f83c CM |
2877 | wret = push_leaf_left(trans, root, path, data_size, |
2878 | data_size, 0, (u32)-1); | |
3685f791 CM |
2879 | if (wret < 0) |
2880 | return wret; | |
2881 | } | |
2882 | l = path->nodes[0]; | |
aa5d6bed | 2883 | |
3685f791 | 2884 | /* did the pushes work? */ |
87b29b20 | 2885 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 2886 | return 0; |
3326d1b0 | 2887 | } |
aa5d6bed | 2888 | |
5c680ed6 | 2889 | if (!path->nodes[1]) { |
e089f05c | 2890 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
2891 | if (ret) |
2892 | return ret; | |
2893 | } | |
cc0c5538 | 2894 | again: |
5d4f98a2 | 2895 | split = 1; |
cc0c5538 | 2896 | l = path->nodes[0]; |
eb60ceac | 2897 | slot = path->slots[0]; |
5f39d397 | 2898 | nritems = btrfs_header_nritems(l); |
d397712b | 2899 | mid = (nritems + 1) / 2; |
54aa1f4d | 2900 | |
5d4f98a2 YZ |
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 | split = 0; | |
2907 | } else { | |
2908 | mid = slot; | |
2909 | if (mid != nritems && | |
2910 | leaf_space_used(l, mid, nritems - mid) + | |
2911 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
2912 | if (data_size && !tried_avoid_double) |
2913 | goto push_for_double; | |
5d4f98a2 YZ |
2914 | split = 2; |
2915 | } | |
2916 | } | |
2917 | } | |
2918 | } else { | |
2919 | if (leaf_space_used(l, 0, mid) + data_size > | |
2920 | BTRFS_LEAF_DATA_SIZE(root)) { | |
2921 | if (!extend && data_size && slot == 0) { | |
2922 | split = 0; | |
2923 | } else if ((extend || !data_size) && slot == 0) { | |
2924 | mid = 1; | |
2925 | } else { | |
2926 | mid = slot; | |
2927 | if (mid != nritems && | |
2928 | leaf_space_used(l, mid, nritems - mid) + | |
2929 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
2930 | if (data_size && !tried_avoid_double) |
2931 | goto push_for_double; | |
5d4f98a2 YZ |
2932 | split = 2 ; |
2933 | } | |
2934 | } | |
2935 | } | |
2936 | } | |
2937 | ||
2938 | if (split == 0) | |
2939 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
2940 | else | |
2941 | btrfs_item_key(l, &disk_key, mid); | |
2942 | ||
2943 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 2944 | root->root_key.objectid, |
66d7e7f0 | 2945 | &disk_key, 0, l->start, 0, 0); |
f0486c68 | 2946 | if (IS_ERR(right)) |
5f39d397 | 2947 | return PTR_ERR(right); |
f0486c68 YZ |
2948 | |
2949 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
2950 | |
2951 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 2952 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 2953 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 2954 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2955 | btrfs_set_header_owner(right, root->root_key.objectid); |
2956 | btrfs_set_header_level(right, 0); | |
2957 | write_extent_buffer(right, root->fs_info->fsid, | |
2958 | (unsigned long)btrfs_header_fsid(right), | |
2959 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2960 | |
2961 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
2962 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
2963 | BTRFS_UUID_SIZE); | |
44871b1b | 2964 | |
5d4f98a2 YZ |
2965 | if (split == 0) { |
2966 | if (mid <= slot) { | |
2967 | btrfs_set_header_nritems(right, 0); | |
143bede5 JM |
2968 | insert_ptr(trans, root, path, &disk_key, right->start, |
2969 | path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
2970 | btrfs_tree_unlock(path->nodes[0]); |
2971 | free_extent_buffer(path->nodes[0]); | |
2972 | path->nodes[0] = right; | |
2973 | path->slots[0] = 0; | |
2974 | path->slots[1] += 1; | |
2975 | } else { | |
2976 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 2977 | insert_ptr(trans, root, path, &disk_key, right->start, |
5d4f98a2 | 2978 | path->slots[1], 1); |
5d4f98a2 YZ |
2979 | btrfs_tree_unlock(path->nodes[0]); |
2980 | free_extent_buffer(path->nodes[0]); | |
2981 | path->nodes[0] = right; | |
2982 | path->slots[0] = 0; | |
143bede5 JM |
2983 | if (path->slots[1] == 0) |
2984 | fixup_low_keys(trans, root, path, | |
2985 | &disk_key, 1); | |
d4dbff95 | 2986 | } |
5d4f98a2 YZ |
2987 | btrfs_mark_buffer_dirty(right); |
2988 | return ret; | |
d4dbff95 | 2989 | } |
74123bd7 | 2990 | |
143bede5 | 2991 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 2992 | |
5d4f98a2 | 2993 | if (split == 2) { |
cc0c5538 CM |
2994 | BUG_ON(num_doubles != 0); |
2995 | num_doubles++; | |
2996 | goto again; | |
a429e513 | 2997 | } |
44871b1b | 2998 | |
143bede5 | 2999 | return 0; |
99d8f83c CM |
3000 | |
3001 | push_for_double: | |
3002 | push_for_double_split(trans, root, path, data_size); | |
3003 | tried_avoid_double = 1; | |
3004 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3005 | return 0; | |
3006 | goto again; | |
be0e5c09 CM |
3007 | } |
3008 | ||
ad48fd75 YZ |
3009 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3010 | struct btrfs_root *root, | |
3011 | struct btrfs_path *path, int ins_len) | |
459931ec | 3012 | { |
ad48fd75 | 3013 | struct btrfs_key key; |
459931ec | 3014 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3015 | struct btrfs_file_extent_item *fi; |
3016 | u64 extent_len = 0; | |
3017 | u32 item_size; | |
3018 | int ret; | |
459931ec CM |
3019 | |
3020 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3021 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3022 | ||
3023 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3024 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3025 | ||
3026 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
3027 | return 0; | |
459931ec CM |
3028 | |
3029 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
3030 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3031 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3032 | struct btrfs_file_extent_item); | |
3033 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3034 | } | |
b3b4aa74 | 3035 | btrfs_release_path(path); |
459931ec | 3036 | |
459931ec | 3037 | path->keep_locks = 1; |
ad48fd75 YZ |
3038 | path->search_for_split = 1; |
3039 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3040 | path->search_for_split = 0; |
ad48fd75 YZ |
3041 | if (ret < 0) |
3042 | goto err; | |
459931ec | 3043 | |
ad48fd75 YZ |
3044 | ret = -EAGAIN; |
3045 | leaf = path->nodes[0]; | |
459931ec | 3046 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
3047 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
3048 | goto err; | |
3049 | ||
109f6aef CM |
3050 | /* the leaf has changed, it now has room. return now */ |
3051 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
3052 | goto err; | |
3053 | ||
ad48fd75 YZ |
3054 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3055 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3056 | struct btrfs_file_extent_item); | |
3057 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3058 | goto err; | |
459931ec CM |
3059 | } |
3060 | ||
b9473439 | 3061 | btrfs_set_path_blocking(path); |
ad48fd75 | 3062 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3063 | if (ret) |
3064 | goto err; | |
459931ec | 3065 | |
ad48fd75 | 3066 | path->keep_locks = 0; |
b9473439 | 3067 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3068 | return 0; |
3069 | err: | |
3070 | path->keep_locks = 0; | |
3071 | return ret; | |
3072 | } | |
3073 | ||
3074 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
3075 | struct btrfs_root *root, | |
3076 | struct btrfs_path *path, | |
3077 | struct btrfs_key *new_key, | |
3078 | unsigned long split_offset) | |
3079 | { | |
3080 | struct extent_buffer *leaf; | |
3081 | struct btrfs_item *item; | |
3082 | struct btrfs_item *new_item; | |
3083 | int slot; | |
3084 | char *buf; | |
3085 | u32 nritems; | |
3086 | u32 item_size; | |
3087 | u32 orig_offset; | |
3088 | struct btrfs_disk_key disk_key; | |
3089 | ||
b9473439 CM |
3090 | leaf = path->nodes[0]; |
3091 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
3092 | ||
b4ce94de CM |
3093 | btrfs_set_path_blocking(path); |
3094 | ||
459931ec CM |
3095 | item = btrfs_item_nr(leaf, path->slots[0]); |
3096 | orig_offset = btrfs_item_offset(leaf, item); | |
3097 | item_size = btrfs_item_size(leaf, item); | |
3098 | ||
459931ec | 3099 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3100 | if (!buf) |
3101 | return -ENOMEM; | |
3102 | ||
459931ec CM |
3103 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3104 | path->slots[0]), item_size); | |
459931ec | 3105 | |
ad48fd75 | 3106 | slot = path->slots[0] + 1; |
459931ec | 3107 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3108 | if (slot != nritems) { |
3109 | /* shift the items */ | |
3110 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3111 | btrfs_item_nr_offset(slot), |
3112 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3113 | } |
3114 | ||
3115 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3116 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3117 | ||
3118 | new_item = btrfs_item_nr(leaf, slot); | |
3119 | ||
3120 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3121 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3122 | ||
3123 | btrfs_set_item_offset(leaf, item, | |
3124 | orig_offset + item_size - split_offset); | |
3125 | btrfs_set_item_size(leaf, item, split_offset); | |
3126 | ||
3127 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3128 | ||
3129 | /* write the data for the start of the original item */ | |
3130 | write_extent_buffer(leaf, buf, | |
3131 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3132 | split_offset); | |
3133 | ||
3134 | /* write the data for the new item */ | |
3135 | write_extent_buffer(leaf, buf + split_offset, | |
3136 | btrfs_item_ptr_offset(leaf, slot), | |
3137 | item_size - split_offset); | |
3138 | btrfs_mark_buffer_dirty(leaf); | |
3139 | ||
ad48fd75 | 3140 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 3141 | kfree(buf); |
ad48fd75 YZ |
3142 | return 0; |
3143 | } | |
3144 | ||
3145 | /* | |
3146 | * This function splits a single item into two items, | |
3147 | * giving 'new_key' to the new item and splitting the | |
3148 | * old one at split_offset (from the start of the item). | |
3149 | * | |
3150 | * The path may be released by this operation. After | |
3151 | * the split, the path is pointing to the old item. The | |
3152 | * new item is going to be in the same node as the old one. | |
3153 | * | |
3154 | * Note, the item being split must be smaller enough to live alone on | |
3155 | * a tree block with room for one extra struct btrfs_item | |
3156 | * | |
3157 | * This allows us to split the item in place, keeping a lock on the | |
3158 | * leaf the entire time. | |
3159 | */ | |
3160 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3161 | struct btrfs_root *root, | |
3162 | struct btrfs_path *path, | |
3163 | struct btrfs_key *new_key, | |
3164 | unsigned long split_offset) | |
3165 | { | |
3166 | int ret; | |
3167 | ret = setup_leaf_for_split(trans, root, path, | |
3168 | sizeof(struct btrfs_item)); | |
3169 | if (ret) | |
3170 | return ret; | |
3171 | ||
3172 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
3173 | return ret; |
3174 | } | |
3175 | ||
ad48fd75 YZ |
3176 | /* |
3177 | * This function duplicate a item, giving 'new_key' to the new item. | |
3178 | * It guarantees both items live in the same tree leaf and the new item | |
3179 | * is contiguous with the original item. | |
3180 | * | |
3181 | * This allows us to split file extent in place, keeping a lock on the | |
3182 | * leaf the entire time. | |
3183 | */ | |
3184 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
3185 | struct btrfs_root *root, | |
3186 | struct btrfs_path *path, | |
3187 | struct btrfs_key *new_key) | |
3188 | { | |
3189 | struct extent_buffer *leaf; | |
3190 | int ret; | |
3191 | u32 item_size; | |
3192 | ||
3193 | leaf = path->nodes[0]; | |
3194 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3195 | ret = setup_leaf_for_split(trans, root, path, | |
3196 | item_size + sizeof(struct btrfs_item)); | |
3197 | if (ret) | |
3198 | return ret; | |
3199 | ||
3200 | path->slots[0]++; | |
143bede5 JM |
3201 | setup_items_for_insert(trans, root, path, new_key, &item_size, |
3202 | item_size, item_size + | |
3203 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
3204 | leaf = path->nodes[0]; |
3205 | memcpy_extent_buffer(leaf, | |
3206 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3207 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
3208 | item_size); | |
3209 | return 0; | |
3210 | } | |
3211 | ||
d352ac68 CM |
3212 | /* |
3213 | * make the item pointed to by the path smaller. new_size indicates | |
3214 | * how small to make it, and from_end tells us if we just chop bytes | |
3215 | * off the end of the item or if we shift the item to chop bytes off | |
3216 | * the front. | |
3217 | */ | |
143bede5 JM |
3218 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3219 | struct btrfs_root *root, | |
3220 | struct btrfs_path *path, | |
3221 | u32 new_size, int from_end) | |
b18c6685 | 3222 | { |
b18c6685 | 3223 | int slot; |
5f39d397 CM |
3224 | struct extent_buffer *leaf; |
3225 | struct btrfs_item *item; | |
b18c6685 CM |
3226 | u32 nritems; |
3227 | unsigned int data_end; | |
3228 | unsigned int old_data_start; | |
3229 | unsigned int old_size; | |
3230 | unsigned int size_diff; | |
3231 | int i; | |
3232 | ||
5f39d397 | 3233 | leaf = path->nodes[0]; |
179e29e4 CM |
3234 | slot = path->slots[0]; |
3235 | ||
3236 | old_size = btrfs_item_size_nr(leaf, slot); | |
3237 | if (old_size == new_size) | |
143bede5 | 3238 | return; |
b18c6685 | 3239 | |
5f39d397 | 3240 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
3241 | data_end = leaf_data_end(root, leaf); |
3242 | ||
5f39d397 | 3243 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3244 | |
b18c6685 CM |
3245 | size_diff = old_size - new_size; |
3246 | ||
3247 | BUG_ON(slot < 0); | |
3248 | BUG_ON(slot >= nritems); | |
3249 | ||
3250 | /* | |
3251 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3252 | */ | |
3253 | /* first correct the data pointers */ | |
3254 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3255 | u32 ioff; |
3256 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3257 | |
5f39d397 CM |
3258 | ioff = btrfs_item_offset(leaf, item); |
3259 | btrfs_set_item_offset(leaf, item, ioff + size_diff); | |
b18c6685 | 3260 | } |
db94535d | 3261 | |
b18c6685 | 3262 | /* shift the data */ |
179e29e4 CM |
3263 | if (from_end) { |
3264 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3265 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3266 | data_end, old_data_start + new_size - data_end); | |
3267 | } else { | |
3268 | struct btrfs_disk_key disk_key; | |
3269 | u64 offset; | |
3270 | ||
3271 | btrfs_item_key(leaf, &disk_key, slot); | |
3272 | ||
3273 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3274 | unsigned long ptr; | |
3275 | struct btrfs_file_extent_item *fi; | |
3276 | ||
3277 | fi = btrfs_item_ptr(leaf, slot, | |
3278 | struct btrfs_file_extent_item); | |
3279 | fi = (struct btrfs_file_extent_item *)( | |
3280 | (unsigned long)fi - size_diff); | |
3281 | ||
3282 | if (btrfs_file_extent_type(leaf, fi) == | |
3283 | BTRFS_FILE_EXTENT_INLINE) { | |
3284 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3285 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
3286 | (unsigned long)fi, |
3287 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
3288 | disk_bytenr)); |
3289 | } | |
3290 | } | |
3291 | ||
3292 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3293 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3294 | data_end, old_data_start - data_end); | |
3295 | ||
3296 | offset = btrfs_disk_key_offset(&disk_key); | |
3297 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3298 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3299 | if (slot == 0) | |
3300 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
3301 | } | |
5f39d397 CM |
3302 | |
3303 | item = btrfs_item_nr(leaf, slot); | |
3304 | btrfs_set_item_size(leaf, item, new_size); | |
3305 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 3306 | |
5f39d397 CM |
3307 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3308 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 3309 | BUG(); |
5f39d397 | 3310 | } |
b18c6685 CM |
3311 | } |
3312 | ||
d352ac68 CM |
3313 | /* |
3314 | * make the item pointed to by the path bigger, data_size is the new size. | |
3315 | */ | |
143bede5 JM |
3316 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
3317 | struct btrfs_root *root, struct btrfs_path *path, | |
3318 | u32 data_size) | |
6567e837 | 3319 | { |
6567e837 | 3320 | int slot; |
5f39d397 CM |
3321 | struct extent_buffer *leaf; |
3322 | struct btrfs_item *item; | |
6567e837 CM |
3323 | u32 nritems; |
3324 | unsigned int data_end; | |
3325 | unsigned int old_data; | |
3326 | unsigned int old_size; | |
3327 | int i; | |
3328 | ||
5f39d397 | 3329 | leaf = path->nodes[0]; |
6567e837 | 3330 | |
5f39d397 | 3331 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
3332 | data_end = leaf_data_end(root, leaf); |
3333 | ||
5f39d397 CM |
3334 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
3335 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3336 | BUG(); |
5f39d397 | 3337 | } |
6567e837 | 3338 | slot = path->slots[0]; |
5f39d397 | 3339 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3340 | |
3341 | BUG_ON(slot < 0); | |
3326d1b0 CM |
3342 | if (slot >= nritems) { |
3343 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
3344 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
3345 | slot, nritems); | |
3326d1b0 CM |
3346 | BUG_ON(1); |
3347 | } | |
6567e837 CM |
3348 | |
3349 | /* | |
3350 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3351 | */ | |
3352 | /* first correct the data pointers */ | |
3353 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3354 | u32 ioff; |
3355 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3356 | |
5f39d397 CM |
3357 | ioff = btrfs_item_offset(leaf, item); |
3358 | btrfs_set_item_offset(leaf, item, ioff - data_size); | |
6567e837 | 3359 | } |
5f39d397 | 3360 | |
6567e837 | 3361 | /* shift the data */ |
5f39d397 | 3362 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
3363 | data_end - data_size, btrfs_leaf_data(leaf) + |
3364 | data_end, old_data - data_end); | |
5f39d397 | 3365 | |
6567e837 | 3366 | data_end = old_data; |
5f39d397 CM |
3367 | old_size = btrfs_item_size_nr(leaf, slot); |
3368 | item = btrfs_item_nr(leaf, slot); | |
3369 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
3370 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 3371 | |
5f39d397 CM |
3372 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3373 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3374 | BUG(); |
5f39d397 | 3375 | } |
6567e837 CM |
3376 | } |
3377 | ||
f3465ca4 JB |
3378 | /* |
3379 | * Given a key and some data, insert items into the tree. | |
3380 | * This does all the path init required, making room in the tree if needed. | |
3381 | * Returns the number of keys that were inserted. | |
3382 | */ | |
3383 | int btrfs_insert_some_items(struct btrfs_trans_handle *trans, | |
3384 | struct btrfs_root *root, | |
3385 | struct btrfs_path *path, | |
3386 | struct btrfs_key *cpu_key, u32 *data_size, | |
3387 | int nr) | |
3388 | { | |
3389 | struct extent_buffer *leaf; | |
3390 | struct btrfs_item *item; | |
3391 | int ret = 0; | |
3392 | int slot; | |
f3465ca4 JB |
3393 | int i; |
3394 | u32 nritems; | |
3395 | u32 total_data = 0; | |
3396 | u32 total_size = 0; | |
3397 | unsigned int data_end; | |
3398 | struct btrfs_disk_key disk_key; | |
3399 | struct btrfs_key found_key; | |
3400 | ||
87b29b20 YZ |
3401 | for (i = 0; i < nr; i++) { |
3402 | if (total_size + data_size[i] + sizeof(struct btrfs_item) > | |
3403 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3404 | break; | |
3405 | nr = i; | |
3406 | } | |
f3465ca4 | 3407 | total_data += data_size[i]; |
87b29b20 YZ |
3408 | total_size += data_size[i] + sizeof(struct btrfs_item); |
3409 | } | |
3410 | BUG_ON(nr == 0); | |
f3465ca4 | 3411 | |
f3465ca4 JB |
3412 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
3413 | if (ret == 0) | |
3414 | return -EEXIST; | |
3415 | if (ret < 0) | |
3416 | goto out; | |
3417 | ||
f3465ca4 JB |
3418 | leaf = path->nodes[0]; |
3419 | ||
3420 | nritems = btrfs_header_nritems(leaf); | |
3421 | data_end = leaf_data_end(root, leaf); | |
3422 | ||
3423 | if (btrfs_leaf_free_space(root, leaf) < total_size) { | |
3424 | for (i = nr; i >= 0; i--) { | |
3425 | total_data -= data_size[i]; | |
3426 | total_size -= data_size[i] + sizeof(struct btrfs_item); | |
3427 | if (total_size < btrfs_leaf_free_space(root, leaf)) | |
3428 | break; | |
3429 | } | |
3430 | nr = i; | |
3431 | } | |
3432 | ||
3433 | slot = path->slots[0]; | |
3434 | BUG_ON(slot < 0); | |
3435 | ||
3436 | if (slot != nritems) { | |
3437 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); | |
3438 | ||
3439 | item = btrfs_item_nr(leaf, slot); | |
3440 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3441 | ||
3442 | /* figure out how many keys we can insert in here */ | |
3443 | total_data = data_size[0]; | |
3444 | for (i = 1; i < nr; i++) { | |
5d4f98a2 | 3445 | if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0) |
f3465ca4 JB |
3446 | break; |
3447 | total_data += data_size[i]; | |
3448 | } | |
3449 | nr = i; | |
3450 | ||
3451 | if (old_data < data_end) { | |
3452 | btrfs_print_leaf(root, leaf); | |
d397712b | 3453 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
f3465ca4 JB |
3454 | slot, old_data, data_end); |
3455 | BUG_ON(1); | |
3456 | } | |
3457 | /* | |
3458 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3459 | */ | |
3460 | /* first correct the data pointers */ | |
f3465ca4 JB |
3461 | for (i = slot; i < nritems; i++) { |
3462 | u32 ioff; | |
3463 | ||
3464 | item = btrfs_item_nr(leaf, i); | |
f3465ca4 JB |
3465 | ioff = btrfs_item_offset(leaf, item); |
3466 | btrfs_set_item_offset(leaf, item, ioff - total_data); | |
3467 | } | |
f3465ca4 JB |
3468 | /* shift the items */ |
3469 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), | |
3470 | btrfs_item_nr_offset(slot), | |
3471 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3472 | ||
3473 | /* shift the data */ | |
3474 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3475 | data_end - total_data, btrfs_leaf_data(leaf) + | |
3476 | data_end, old_data - data_end); | |
3477 | data_end = old_data; | |
3478 | } else { | |
3479 | /* | |
3480 | * this sucks but it has to be done, if we are inserting at | |
3481 | * the end of the leaf only insert 1 of the items, since we | |
3482 | * have no way of knowing whats on the next leaf and we'd have | |
3483 | * to drop our current locks to figure it out | |
3484 | */ | |
3485 | nr = 1; | |
3486 | } | |
3487 | ||
3488 | /* setup the item for the new data */ | |
3489 | for (i = 0; i < nr; i++) { | |
3490 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3491 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3492 | item = btrfs_item_nr(leaf, slot + i); | |
3493 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3494 | data_end -= data_size[i]; | |
3495 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3496 | } | |
3497 | btrfs_set_header_nritems(leaf, nritems + nr); | |
3498 | btrfs_mark_buffer_dirty(leaf); | |
3499 | ||
3500 | ret = 0; | |
3501 | if (slot == 0) { | |
3502 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3503 | fixup_low_keys(trans, root, path, &disk_key, 1); |
f3465ca4 JB |
3504 | } |
3505 | ||
3506 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
3507 | btrfs_print_leaf(root, leaf); | |
3508 | BUG(); | |
3509 | } | |
3510 | out: | |
3511 | if (!ret) | |
3512 | ret = nr; | |
3513 | return ret; | |
3514 | } | |
3515 | ||
74123bd7 | 3516 | /* |
44871b1b CM |
3517 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
3518 | * to save stack depth by doing the bulk of the work in a function | |
3519 | * that doesn't call btrfs_search_slot | |
74123bd7 | 3520 | */ |
143bede5 JM |
3521 | void setup_items_for_insert(struct btrfs_trans_handle *trans, |
3522 | struct btrfs_root *root, struct btrfs_path *path, | |
3523 | struct btrfs_key *cpu_key, u32 *data_size, | |
3524 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 3525 | { |
5f39d397 | 3526 | struct btrfs_item *item; |
9c58309d | 3527 | int i; |
7518a238 | 3528 | u32 nritems; |
be0e5c09 | 3529 | unsigned int data_end; |
e2fa7227 | 3530 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3531 | struct extent_buffer *leaf; |
3532 | int slot; | |
e2fa7227 | 3533 | |
5f39d397 | 3534 | leaf = path->nodes[0]; |
44871b1b | 3535 | slot = path->slots[0]; |
74123bd7 | 3536 | |
5f39d397 | 3537 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 3538 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 3539 | |
f25956cc | 3540 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 3541 | btrfs_print_leaf(root, leaf); |
d397712b | 3542 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 3543 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 3544 | BUG(); |
d4dbff95 | 3545 | } |
5f39d397 | 3546 | |
be0e5c09 | 3547 | if (slot != nritems) { |
5f39d397 | 3548 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 3549 | |
5f39d397 CM |
3550 | if (old_data < data_end) { |
3551 | btrfs_print_leaf(root, leaf); | |
d397712b | 3552 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
3553 | slot, old_data, data_end); |
3554 | BUG_ON(1); | |
3555 | } | |
be0e5c09 CM |
3556 | /* |
3557 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3558 | */ | |
3559 | /* first correct the data pointers */ | |
0783fcfc | 3560 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3561 | u32 ioff; |
db94535d | 3562 | |
5f39d397 CM |
3563 | item = btrfs_item_nr(leaf, i); |
3564 | ioff = btrfs_item_offset(leaf, item); | |
9c58309d | 3565 | btrfs_set_item_offset(leaf, item, ioff - total_data); |
0783fcfc | 3566 | } |
be0e5c09 | 3567 | /* shift the items */ |
9c58309d | 3568 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 3569 | btrfs_item_nr_offset(slot), |
d6025579 | 3570 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
3571 | |
3572 | /* shift the data */ | |
5f39d397 | 3573 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 3574 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 3575 | data_end, old_data - data_end); |
be0e5c09 CM |
3576 | data_end = old_data; |
3577 | } | |
5f39d397 | 3578 | |
62e2749e | 3579 | /* setup the item for the new data */ |
9c58309d CM |
3580 | for (i = 0; i < nr; i++) { |
3581 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3582 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3583 | item = btrfs_item_nr(leaf, slot + i); | |
3584 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3585 | data_end -= data_size[i]; | |
3586 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3587 | } | |
44871b1b | 3588 | |
9c58309d | 3589 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 3590 | |
5a01a2e3 CM |
3591 | if (slot == 0) { |
3592 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3593 | fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 3594 | } |
b9473439 CM |
3595 | btrfs_unlock_up_safe(path, 1); |
3596 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 3597 | |
5f39d397 CM |
3598 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3599 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 3600 | BUG(); |
5f39d397 | 3601 | } |
44871b1b CM |
3602 | } |
3603 | ||
3604 | /* | |
3605 | * Given a key and some data, insert items into the tree. | |
3606 | * This does all the path init required, making room in the tree if needed. | |
3607 | */ | |
3608 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
3609 | struct btrfs_root *root, | |
3610 | struct btrfs_path *path, | |
3611 | struct btrfs_key *cpu_key, u32 *data_size, | |
3612 | int nr) | |
3613 | { | |
44871b1b CM |
3614 | int ret = 0; |
3615 | int slot; | |
3616 | int i; | |
3617 | u32 total_size = 0; | |
3618 | u32 total_data = 0; | |
3619 | ||
3620 | for (i = 0; i < nr; i++) | |
3621 | total_data += data_size[i]; | |
3622 | ||
3623 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
3624 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
3625 | if (ret == 0) | |
3626 | return -EEXIST; | |
3627 | if (ret < 0) | |
143bede5 | 3628 | return ret; |
44871b1b | 3629 | |
44871b1b CM |
3630 | slot = path->slots[0]; |
3631 | BUG_ON(slot < 0); | |
3632 | ||
143bede5 | 3633 | setup_items_for_insert(trans, root, path, cpu_key, data_size, |
44871b1b | 3634 | total_data, total_size, nr); |
143bede5 | 3635 | return 0; |
62e2749e CM |
3636 | } |
3637 | ||
3638 | /* | |
3639 | * Given a key and some data, insert an item into the tree. | |
3640 | * This does all the path init required, making room in the tree if needed. | |
3641 | */ | |
e089f05c CM |
3642 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
3643 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
3644 | data_size) | |
62e2749e CM |
3645 | { |
3646 | int ret = 0; | |
2c90e5d6 | 3647 | struct btrfs_path *path; |
5f39d397 CM |
3648 | struct extent_buffer *leaf; |
3649 | unsigned long ptr; | |
62e2749e | 3650 | |
2c90e5d6 | 3651 | path = btrfs_alloc_path(); |
db5b493a TI |
3652 | if (!path) |
3653 | return -ENOMEM; | |
2c90e5d6 | 3654 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 3655 | if (!ret) { |
5f39d397 CM |
3656 | leaf = path->nodes[0]; |
3657 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3658 | write_extent_buffer(leaf, data, ptr, data_size); | |
3659 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 3660 | } |
2c90e5d6 | 3661 | btrfs_free_path(path); |
aa5d6bed | 3662 | return ret; |
be0e5c09 CM |
3663 | } |
3664 | ||
74123bd7 | 3665 | /* |
5de08d7d | 3666 | * delete the pointer from a given node. |
74123bd7 | 3667 | * |
d352ac68 CM |
3668 | * the tree should have been previously balanced so the deletion does not |
3669 | * empty a node. | |
74123bd7 | 3670 | */ |
143bede5 JM |
3671 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3672 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 3673 | { |
5f39d397 | 3674 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 3675 | u32 nritems; |
be0e5c09 | 3676 | |
5f39d397 | 3677 | nritems = btrfs_header_nritems(parent); |
d397712b | 3678 | if (slot != nritems - 1) { |
5f39d397 CM |
3679 | memmove_extent_buffer(parent, |
3680 | btrfs_node_key_ptr_offset(slot), | |
3681 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
3682 | sizeof(struct btrfs_key_ptr) * |
3683 | (nritems - slot - 1)); | |
bb803951 | 3684 | } |
7518a238 | 3685 | nritems--; |
5f39d397 | 3686 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 3687 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 3688 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 3689 | /* just turn the root into a leaf and break */ |
5f39d397 | 3690 | btrfs_set_header_level(root->node, 0); |
bb803951 | 3691 | } else if (slot == 0) { |
5f39d397 CM |
3692 | struct btrfs_disk_key disk_key; |
3693 | ||
3694 | btrfs_node_key(parent, &disk_key, 0); | |
143bede5 | 3695 | fixup_low_keys(trans, root, path, &disk_key, level + 1); |
be0e5c09 | 3696 | } |
d6025579 | 3697 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
3698 | } |
3699 | ||
323ac95b CM |
3700 | /* |
3701 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 3702 | * path->nodes[1]. |
323ac95b CM |
3703 | * |
3704 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
3705 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
3706 | * | |
3707 | * The path must have already been setup for deleting the leaf, including | |
3708 | * all the proper balancing. path->nodes[1] must be locked. | |
3709 | */ | |
143bede5 JM |
3710 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
3711 | struct btrfs_root *root, | |
3712 | struct btrfs_path *path, | |
3713 | struct extent_buffer *leaf) | |
323ac95b | 3714 | { |
5d4f98a2 | 3715 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
143bede5 | 3716 | del_ptr(trans, root, path, 1, path->slots[1]); |
323ac95b | 3717 | |
4d081c41 CM |
3718 | /* |
3719 | * btrfs_free_extent is expensive, we want to make sure we | |
3720 | * aren't holding any locks when we call it | |
3721 | */ | |
3722 | btrfs_unlock_up_safe(path, 0); | |
3723 | ||
f0486c68 YZ |
3724 | root_sub_used(root, leaf->len); |
3725 | ||
66d7e7f0 | 3726 | btrfs_free_tree_block(trans, root, leaf, 0, 1, 0); |
323ac95b | 3727 | } |
74123bd7 CM |
3728 | /* |
3729 | * delete the item at the leaf level in path. If that empties | |
3730 | * the leaf, remove it from the tree | |
3731 | */ | |
85e21bac CM |
3732 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3733 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 3734 | { |
5f39d397 CM |
3735 | struct extent_buffer *leaf; |
3736 | struct btrfs_item *item; | |
85e21bac CM |
3737 | int last_off; |
3738 | int dsize = 0; | |
aa5d6bed CM |
3739 | int ret = 0; |
3740 | int wret; | |
85e21bac | 3741 | int i; |
7518a238 | 3742 | u32 nritems; |
be0e5c09 | 3743 | |
5f39d397 | 3744 | leaf = path->nodes[0]; |
85e21bac CM |
3745 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
3746 | ||
3747 | for (i = 0; i < nr; i++) | |
3748 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
3749 | ||
5f39d397 | 3750 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 3751 | |
85e21bac | 3752 | if (slot + nr != nritems) { |
123abc88 | 3753 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
3754 | |
3755 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
3756 | data_end + dsize, |
3757 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 3758 | last_off - data_end); |
5f39d397 | 3759 | |
85e21bac | 3760 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 3761 | u32 ioff; |
db94535d | 3762 | |
5f39d397 CM |
3763 | item = btrfs_item_nr(leaf, i); |
3764 | ioff = btrfs_item_offset(leaf, item); | |
3765 | btrfs_set_item_offset(leaf, item, ioff + dsize); | |
0783fcfc | 3766 | } |
db94535d | 3767 | |
5f39d397 | 3768 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 3769 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 3770 | sizeof(struct btrfs_item) * |
85e21bac | 3771 | (nritems - slot - nr)); |
be0e5c09 | 3772 | } |
85e21bac CM |
3773 | btrfs_set_header_nritems(leaf, nritems - nr); |
3774 | nritems -= nr; | |
5f39d397 | 3775 | |
74123bd7 | 3776 | /* delete the leaf if we've emptied it */ |
7518a238 | 3777 | if (nritems == 0) { |
5f39d397 CM |
3778 | if (leaf == root->node) { |
3779 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 3780 | } else { |
f0486c68 YZ |
3781 | btrfs_set_path_blocking(path); |
3782 | clean_tree_block(trans, root, leaf); | |
143bede5 | 3783 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 3784 | } |
be0e5c09 | 3785 | } else { |
7518a238 | 3786 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 3787 | if (slot == 0) { |
5f39d397 CM |
3788 | struct btrfs_disk_key disk_key; |
3789 | ||
3790 | btrfs_item_key(leaf, &disk_key, 0); | |
143bede5 | 3791 | fixup_low_keys(trans, root, path, &disk_key, 1); |
aa5d6bed | 3792 | } |
aa5d6bed | 3793 | |
74123bd7 | 3794 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 3795 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
3796 | /* push_leaf_left fixes the path. |
3797 | * make sure the path still points to our leaf | |
3798 | * for possible call to del_ptr below | |
3799 | */ | |
4920c9ac | 3800 | slot = path->slots[1]; |
5f39d397 CM |
3801 | extent_buffer_get(leaf); |
3802 | ||
b9473439 | 3803 | btrfs_set_path_blocking(path); |
99d8f83c CM |
3804 | wret = push_leaf_left(trans, root, path, 1, 1, |
3805 | 1, (u32)-1); | |
54aa1f4d | 3806 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 3807 | ret = wret; |
5f39d397 CM |
3808 | |
3809 | if (path->nodes[0] == leaf && | |
3810 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
3811 | wret = push_leaf_right(trans, root, path, 1, |
3812 | 1, 1, 0); | |
54aa1f4d | 3813 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
3814 | ret = wret; |
3815 | } | |
5f39d397 CM |
3816 | |
3817 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 3818 | path->slots[1] = slot; |
143bede5 | 3819 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 3820 | free_extent_buffer(leaf); |
143bede5 | 3821 | ret = 0; |
5de08d7d | 3822 | } else { |
925baedd CM |
3823 | /* if we're still in the path, make sure |
3824 | * we're dirty. Otherwise, one of the | |
3825 | * push_leaf functions must have already | |
3826 | * dirtied this buffer | |
3827 | */ | |
3828 | if (path->nodes[0] == leaf) | |
3829 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 3830 | free_extent_buffer(leaf); |
be0e5c09 | 3831 | } |
d5719762 | 3832 | } else { |
5f39d397 | 3833 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
3834 | } |
3835 | } | |
aa5d6bed | 3836 | return ret; |
be0e5c09 CM |
3837 | } |
3838 | ||
7bb86316 | 3839 | /* |
925baedd | 3840 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
3841 | * returns 0 if it found something or 1 if there are no lesser leaves. |
3842 | * returns < 0 on io errors. | |
d352ac68 CM |
3843 | * |
3844 | * This may release the path, and so you may lose any locks held at the | |
3845 | * time you call it. | |
7bb86316 CM |
3846 | */ |
3847 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
3848 | { | |
925baedd CM |
3849 | struct btrfs_key key; |
3850 | struct btrfs_disk_key found_key; | |
3851 | int ret; | |
7bb86316 | 3852 | |
925baedd | 3853 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 3854 | |
925baedd CM |
3855 | if (key.offset > 0) |
3856 | key.offset--; | |
3857 | else if (key.type > 0) | |
3858 | key.type--; | |
3859 | else if (key.objectid > 0) | |
3860 | key.objectid--; | |
3861 | else | |
3862 | return 1; | |
7bb86316 | 3863 | |
b3b4aa74 | 3864 | btrfs_release_path(path); |
925baedd CM |
3865 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3866 | if (ret < 0) | |
3867 | return ret; | |
3868 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
3869 | ret = comp_keys(&found_key, &key); | |
3870 | if (ret < 0) | |
3871 | return 0; | |
3872 | return 1; | |
7bb86316 CM |
3873 | } |
3874 | ||
3f157a2f CM |
3875 | /* |
3876 | * A helper function to walk down the tree starting at min_key, and looking | |
3877 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 3878 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
3879 | * |
3880 | * This does not cow, but it does stuff the starting key it finds back | |
3881 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
3882 | * key and get a writable path. | |
3883 | * | |
3884 | * This does lock as it descends, and path->keep_locks should be set | |
3885 | * to 1 by the caller. | |
3886 | * | |
3887 | * This honors path->lowest_level to prevent descent past a given level | |
3888 | * of the tree. | |
3889 | * | |
d352ac68 CM |
3890 | * min_trans indicates the oldest transaction that you are interested |
3891 | * in walking through. Any nodes or leaves older than min_trans are | |
3892 | * skipped over (without reading them). | |
3893 | * | |
3f157a2f CM |
3894 | * returns zero if something useful was found, < 0 on error and 1 if there |
3895 | * was nothing in the tree that matched the search criteria. | |
3896 | */ | |
3897 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 3898 | struct btrfs_key *max_key, |
3f157a2f CM |
3899 | struct btrfs_path *path, int cache_only, |
3900 | u64 min_trans) | |
3901 | { | |
3902 | struct extent_buffer *cur; | |
3903 | struct btrfs_key found_key; | |
3904 | int slot; | |
9652480b | 3905 | int sret; |
3f157a2f CM |
3906 | u32 nritems; |
3907 | int level; | |
3908 | int ret = 1; | |
3909 | ||
934d375b | 3910 | WARN_ON(!path->keep_locks); |
3f157a2f | 3911 | again: |
bd681513 | 3912 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 3913 | level = btrfs_header_level(cur); |
e02119d5 | 3914 | WARN_ON(path->nodes[level]); |
3f157a2f | 3915 | path->nodes[level] = cur; |
bd681513 | 3916 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
3917 | |
3918 | if (btrfs_header_generation(cur) < min_trans) { | |
3919 | ret = 1; | |
3920 | goto out; | |
3921 | } | |
d397712b | 3922 | while (1) { |
3f157a2f CM |
3923 | nritems = btrfs_header_nritems(cur); |
3924 | level = btrfs_header_level(cur); | |
9652480b | 3925 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 3926 | |
323ac95b CM |
3927 | /* at the lowest level, we're done, setup the path and exit */ |
3928 | if (level == path->lowest_level) { | |
e02119d5 CM |
3929 | if (slot >= nritems) |
3930 | goto find_next_key; | |
3f157a2f CM |
3931 | ret = 0; |
3932 | path->slots[level] = slot; | |
3933 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
3934 | goto out; | |
3935 | } | |
9652480b Y |
3936 | if (sret && slot > 0) |
3937 | slot--; | |
3f157a2f CM |
3938 | /* |
3939 | * check this node pointer against the cache_only and | |
3940 | * min_trans parameters. If it isn't in cache or is too | |
3941 | * old, skip to the next one. | |
3942 | */ | |
d397712b | 3943 | while (slot < nritems) { |
3f157a2f CM |
3944 | u64 blockptr; |
3945 | u64 gen; | |
3946 | struct extent_buffer *tmp; | |
e02119d5 CM |
3947 | struct btrfs_disk_key disk_key; |
3948 | ||
3f157a2f CM |
3949 | blockptr = btrfs_node_blockptr(cur, slot); |
3950 | gen = btrfs_node_ptr_generation(cur, slot); | |
3951 | if (gen < min_trans) { | |
3952 | slot++; | |
3953 | continue; | |
3954 | } | |
3955 | if (!cache_only) | |
3956 | break; | |
3957 | ||
e02119d5 CM |
3958 | if (max_key) { |
3959 | btrfs_node_key(cur, &disk_key, slot); | |
3960 | if (comp_keys(&disk_key, max_key) >= 0) { | |
3961 | ret = 1; | |
3962 | goto out; | |
3963 | } | |
3964 | } | |
3965 | ||
3f157a2f CM |
3966 | tmp = btrfs_find_tree_block(root, blockptr, |
3967 | btrfs_level_size(root, level - 1)); | |
3968 | ||
3969 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
3970 | free_extent_buffer(tmp); | |
3971 | break; | |
3972 | } | |
3973 | if (tmp) | |
3974 | free_extent_buffer(tmp); | |
3975 | slot++; | |
3976 | } | |
e02119d5 | 3977 | find_next_key: |
3f157a2f CM |
3978 | /* |
3979 | * we didn't find a candidate key in this node, walk forward | |
3980 | * and find another one | |
3981 | */ | |
3982 | if (slot >= nritems) { | |
e02119d5 | 3983 | path->slots[level] = slot; |
b4ce94de | 3984 | btrfs_set_path_blocking(path); |
e02119d5 | 3985 | sret = btrfs_find_next_key(root, path, min_key, level, |
3f157a2f | 3986 | cache_only, min_trans); |
e02119d5 | 3987 | if (sret == 0) { |
b3b4aa74 | 3988 | btrfs_release_path(path); |
3f157a2f CM |
3989 | goto again; |
3990 | } else { | |
3991 | goto out; | |
3992 | } | |
3993 | } | |
3994 | /* save our key for returning back */ | |
3995 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
3996 | path->slots[level] = slot; | |
3997 | if (level == path->lowest_level) { | |
3998 | ret = 0; | |
3999 | unlock_up(path, level, 1); | |
4000 | goto out; | |
4001 | } | |
b4ce94de | 4002 | btrfs_set_path_blocking(path); |
3f157a2f | 4003 | cur = read_node_slot(root, cur, slot); |
97d9a8a4 | 4004 | BUG_ON(!cur); |
3f157a2f | 4005 | |
bd681513 | 4006 | btrfs_tree_read_lock(cur); |
b4ce94de | 4007 | |
bd681513 | 4008 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f CM |
4009 | path->nodes[level - 1] = cur; |
4010 | unlock_up(path, level, 1); | |
bd681513 | 4011 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4012 | } |
4013 | out: | |
4014 | if (ret == 0) | |
4015 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4016 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4017 | return ret; |
4018 | } | |
4019 | ||
4020 | /* | |
4021 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4022 | * and fixup the path. It looks for and returns the next key in the | |
4023 | * tree based on the current path and the cache_only and min_trans | |
4024 | * parameters. | |
4025 | * | |
4026 | * 0 is returned if another key is found, < 0 if there are any errors | |
4027 | * and 1 is returned if there are no higher keys in the tree | |
4028 | * | |
4029 | * path->keep_locks should be set to 1 on the search made before | |
4030 | * calling this function. | |
4031 | */ | |
e7a84565 | 4032 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
33c66f43 | 4033 | struct btrfs_key *key, int level, |
3f157a2f | 4034 | int cache_only, u64 min_trans) |
e7a84565 | 4035 | { |
e7a84565 CM |
4036 | int slot; |
4037 | struct extent_buffer *c; | |
4038 | ||
934d375b | 4039 | WARN_ON(!path->keep_locks); |
d397712b | 4040 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4041 | if (!path->nodes[level]) |
4042 | return 1; | |
4043 | ||
4044 | slot = path->slots[level] + 1; | |
4045 | c = path->nodes[level]; | |
3f157a2f | 4046 | next: |
e7a84565 | 4047 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4048 | int ret; |
4049 | int orig_lowest; | |
4050 | struct btrfs_key cur_key; | |
4051 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4052 | !path->nodes[level + 1]) | |
e7a84565 | 4053 | return 1; |
33c66f43 YZ |
4054 | |
4055 | if (path->locks[level + 1]) { | |
4056 | level++; | |
4057 | continue; | |
4058 | } | |
4059 | ||
4060 | slot = btrfs_header_nritems(c) - 1; | |
4061 | if (level == 0) | |
4062 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4063 | else | |
4064 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4065 | ||
4066 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4067 | btrfs_release_path(path); |
33c66f43 YZ |
4068 | path->lowest_level = level; |
4069 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4070 | 0, 0); | |
4071 | path->lowest_level = orig_lowest; | |
4072 | if (ret < 0) | |
4073 | return ret; | |
4074 | ||
4075 | c = path->nodes[level]; | |
4076 | slot = path->slots[level]; | |
4077 | if (ret == 0) | |
4078 | slot++; | |
4079 | goto next; | |
e7a84565 | 4080 | } |
33c66f43 | 4081 | |
e7a84565 CM |
4082 | if (level == 0) |
4083 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
4084 | else { |
4085 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
4086 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
4087 | ||
4088 | if (cache_only) { | |
4089 | struct extent_buffer *cur; | |
4090 | cur = btrfs_find_tree_block(root, blockptr, | |
4091 | btrfs_level_size(root, level - 1)); | |
4092 | if (!cur || !btrfs_buffer_uptodate(cur, gen)) { | |
4093 | slot++; | |
4094 | if (cur) | |
4095 | free_extent_buffer(cur); | |
4096 | goto next; | |
4097 | } | |
4098 | free_extent_buffer(cur); | |
4099 | } | |
4100 | if (gen < min_trans) { | |
4101 | slot++; | |
4102 | goto next; | |
4103 | } | |
e7a84565 | 4104 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4105 | } |
e7a84565 CM |
4106 | return 0; |
4107 | } | |
4108 | return 1; | |
4109 | } | |
4110 | ||
97571fd0 | 4111 | /* |
925baedd | 4112 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
4113 | * returns 0 if it found something or 1 if there are no greater leaves. |
4114 | * returns < 0 on io errors. | |
97571fd0 | 4115 | */ |
234b63a0 | 4116 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
4117 | { |
4118 | int slot; | |
8e73f275 | 4119 | int level; |
5f39d397 | 4120 | struct extent_buffer *c; |
8e73f275 | 4121 | struct extent_buffer *next; |
925baedd CM |
4122 | struct btrfs_key key; |
4123 | u32 nritems; | |
4124 | int ret; | |
8e73f275 | 4125 | int old_spinning = path->leave_spinning; |
bd681513 | 4126 | int next_rw_lock = 0; |
925baedd CM |
4127 | |
4128 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4129 | if (nritems == 0) |
925baedd | 4130 | return 1; |
925baedd | 4131 | |
8e73f275 CM |
4132 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4133 | again: | |
4134 | level = 1; | |
4135 | next = NULL; | |
bd681513 | 4136 | next_rw_lock = 0; |
b3b4aa74 | 4137 | btrfs_release_path(path); |
8e73f275 | 4138 | |
a2135011 | 4139 | path->keep_locks = 1; |
31533fb2 | 4140 | path->leave_spinning = 1; |
8e73f275 | 4141 | |
925baedd CM |
4142 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4143 | path->keep_locks = 0; | |
4144 | ||
4145 | if (ret < 0) | |
4146 | return ret; | |
4147 | ||
a2135011 | 4148 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4149 | /* |
4150 | * by releasing the path above we dropped all our locks. A balance | |
4151 | * could have added more items next to the key that used to be | |
4152 | * at the very end of the block. So, check again here and | |
4153 | * advance the path if there are now more items available. | |
4154 | */ | |
a2135011 | 4155 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4156 | if (ret == 0) |
4157 | path->slots[0]++; | |
8e73f275 | 4158 | ret = 0; |
925baedd CM |
4159 | goto done; |
4160 | } | |
d97e63b6 | 4161 | |
d397712b | 4162 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4163 | if (!path->nodes[level]) { |
4164 | ret = 1; | |
4165 | goto done; | |
4166 | } | |
5f39d397 | 4167 | |
d97e63b6 CM |
4168 | slot = path->slots[level] + 1; |
4169 | c = path->nodes[level]; | |
5f39d397 | 4170 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4171 | level++; |
8e73f275 CM |
4172 | if (level == BTRFS_MAX_LEVEL) { |
4173 | ret = 1; | |
4174 | goto done; | |
4175 | } | |
d97e63b6 CM |
4176 | continue; |
4177 | } | |
5f39d397 | 4178 | |
925baedd | 4179 | if (next) { |
bd681513 | 4180 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 4181 | free_extent_buffer(next); |
925baedd | 4182 | } |
5f39d397 | 4183 | |
8e73f275 | 4184 | next = c; |
bd681513 | 4185 | next_rw_lock = path->locks[level]; |
8e73f275 CM |
4186 | ret = read_block_for_search(NULL, root, path, &next, level, |
4187 | slot, &key); | |
4188 | if (ret == -EAGAIN) | |
4189 | goto again; | |
5f39d397 | 4190 | |
76a05b35 | 4191 | if (ret < 0) { |
b3b4aa74 | 4192 | btrfs_release_path(path); |
76a05b35 CM |
4193 | goto done; |
4194 | } | |
4195 | ||
5cd57b2c | 4196 | if (!path->skip_locking) { |
bd681513 | 4197 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4198 | if (!ret) { |
4199 | btrfs_set_path_blocking(path); | |
bd681513 | 4200 | btrfs_tree_read_lock(next); |
31533fb2 | 4201 | btrfs_clear_path_blocking(path, next, |
bd681513 | 4202 | BTRFS_READ_LOCK); |
8e73f275 | 4203 | } |
31533fb2 | 4204 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4205 | } |
d97e63b6 CM |
4206 | break; |
4207 | } | |
4208 | path->slots[level] = slot; | |
d397712b | 4209 | while (1) { |
d97e63b6 CM |
4210 | level--; |
4211 | c = path->nodes[level]; | |
925baedd | 4212 | if (path->locks[level]) |
bd681513 | 4213 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 4214 | |
5f39d397 | 4215 | free_extent_buffer(c); |
d97e63b6 CM |
4216 | path->nodes[level] = next; |
4217 | path->slots[level] = 0; | |
a74a4b97 | 4218 | if (!path->skip_locking) |
bd681513 | 4219 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
4220 | if (!level) |
4221 | break; | |
b4ce94de | 4222 | |
8e73f275 CM |
4223 | ret = read_block_for_search(NULL, root, path, &next, level, |
4224 | 0, &key); | |
4225 | if (ret == -EAGAIN) | |
4226 | goto again; | |
4227 | ||
76a05b35 | 4228 | if (ret < 0) { |
b3b4aa74 | 4229 | btrfs_release_path(path); |
76a05b35 CM |
4230 | goto done; |
4231 | } | |
4232 | ||
5cd57b2c | 4233 | if (!path->skip_locking) { |
bd681513 | 4234 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4235 | if (!ret) { |
4236 | btrfs_set_path_blocking(path); | |
bd681513 | 4237 | btrfs_tree_read_lock(next); |
31533fb2 | 4238 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
4239 | BTRFS_READ_LOCK); |
4240 | } | |
31533fb2 | 4241 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4242 | } |
d97e63b6 | 4243 | } |
8e73f275 | 4244 | ret = 0; |
925baedd CM |
4245 | done: |
4246 | unlock_up(path, 0, 1); | |
8e73f275 CM |
4247 | path->leave_spinning = old_spinning; |
4248 | if (!old_spinning) | |
4249 | btrfs_set_path_blocking(path); | |
4250 | ||
4251 | return ret; | |
d97e63b6 | 4252 | } |
0b86a832 | 4253 | |
3f157a2f CM |
4254 | /* |
4255 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4256 | * searching until it gets past min_objectid or finds an item of 'type' | |
4257 | * | |
4258 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4259 | */ | |
0b86a832 CM |
4260 | int btrfs_previous_item(struct btrfs_root *root, |
4261 | struct btrfs_path *path, u64 min_objectid, | |
4262 | int type) | |
4263 | { | |
4264 | struct btrfs_key found_key; | |
4265 | struct extent_buffer *leaf; | |
e02119d5 | 4266 | u32 nritems; |
0b86a832 CM |
4267 | int ret; |
4268 | ||
d397712b | 4269 | while (1) { |
0b86a832 | 4270 | if (path->slots[0] == 0) { |
b4ce94de | 4271 | btrfs_set_path_blocking(path); |
0b86a832 CM |
4272 | ret = btrfs_prev_leaf(root, path); |
4273 | if (ret != 0) | |
4274 | return ret; | |
4275 | } else { | |
4276 | path->slots[0]--; | |
4277 | } | |
4278 | leaf = path->nodes[0]; | |
e02119d5 CM |
4279 | nritems = btrfs_header_nritems(leaf); |
4280 | if (nritems == 0) | |
4281 | return 1; | |
4282 | if (path->slots[0] == nritems) | |
4283 | path->slots[0]--; | |
4284 | ||
0b86a832 | 4285 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4286 | if (found_key.objectid < min_objectid) |
4287 | break; | |
0a4eefbb YZ |
4288 | if (found_key.type == type) |
4289 | return 0; | |
e02119d5 CM |
4290 | if (found_key.objectid == min_objectid && |
4291 | found_key.type < type) | |
4292 | break; | |
0b86a832 CM |
4293 | } |
4294 | return 1; | |
4295 | } |