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 | 158 | |
3083ee2e JB |
159 | while (1) { |
160 | rcu_read_lock(); | |
161 | eb = rcu_dereference(root->node); | |
162 | ||
163 | /* | |
164 | * RCU really hurts here, we could free up the root node because | |
165 | * it was cow'ed but we may not get the new root node yet so do | |
166 | * the inc_not_zero dance and if it doesn't work then | |
167 | * synchronize_rcu and try again. | |
168 | */ | |
169 | if (atomic_inc_not_zero(&eb->refs)) { | |
170 | rcu_read_unlock(); | |
171 | break; | |
172 | } | |
173 | rcu_read_unlock(); | |
174 | synchronize_rcu(); | |
175 | } | |
925baedd CM |
176 | return eb; |
177 | } | |
178 | ||
d352ac68 CM |
179 | /* loop around taking references on and locking the root node of the |
180 | * tree until you end up with a lock on the root. A locked buffer | |
181 | * is returned, with a reference held. | |
182 | */ | |
925baedd CM |
183 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
184 | { | |
185 | struct extent_buffer *eb; | |
186 | ||
d397712b | 187 | while (1) { |
925baedd CM |
188 | eb = btrfs_root_node(root); |
189 | btrfs_tree_lock(eb); | |
240f62c8 | 190 | if (eb == root->node) |
925baedd | 191 | break; |
925baedd CM |
192 | btrfs_tree_unlock(eb); |
193 | free_extent_buffer(eb); | |
194 | } | |
195 | return eb; | |
196 | } | |
197 | ||
bd681513 CM |
198 | /* loop around taking references on and locking the root node of the |
199 | * tree until you end up with a lock on the root. A locked buffer | |
200 | * is returned, with a reference held. | |
201 | */ | |
202 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) | |
203 | { | |
204 | struct extent_buffer *eb; | |
205 | ||
206 | while (1) { | |
207 | eb = btrfs_root_node(root); | |
208 | btrfs_tree_read_lock(eb); | |
209 | if (eb == root->node) | |
210 | break; | |
211 | btrfs_tree_read_unlock(eb); | |
212 | free_extent_buffer(eb); | |
213 | } | |
214 | return eb; | |
215 | } | |
216 | ||
d352ac68 CM |
217 | /* cowonly root (everything not a reference counted cow subvolume), just get |
218 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
219 | * get properly updated on disk. | |
220 | */ | |
0b86a832 CM |
221 | static void add_root_to_dirty_list(struct btrfs_root *root) |
222 | { | |
e5846fc6 | 223 | spin_lock(&root->fs_info->trans_lock); |
0b86a832 CM |
224 | if (root->track_dirty && list_empty(&root->dirty_list)) { |
225 | list_add(&root->dirty_list, | |
226 | &root->fs_info->dirty_cowonly_roots); | |
227 | } | |
e5846fc6 | 228 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
229 | } |
230 | ||
d352ac68 CM |
231 | /* |
232 | * used by snapshot creation to make a copy of a root for a tree with | |
233 | * a given objectid. The buffer with the new root node is returned in | |
234 | * cow_ret, and this func returns zero on success or a negative error code. | |
235 | */ | |
be20aa9d CM |
236 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
237 | struct btrfs_root *root, | |
238 | struct extent_buffer *buf, | |
239 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
240 | { | |
241 | struct extent_buffer *cow; | |
be20aa9d CM |
242 | int ret = 0; |
243 | int level; | |
5d4f98a2 | 244 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
245 | |
246 | WARN_ON(root->ref_cows && trans->transid != | |
247 | root->fs_info->running_transaction->transid); | |
248 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
249 | ||
250 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
251 | if (level == 0) |
252 | btrfs_item_key(buf, &disk_key, 0); | |
253 | else | |
254 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 255 | |
5d4f98a2 YZ |
256 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
257 | new_root_objectid, &disk_key, level, | |
66d7e7f0 | 258 | buf->start, 0, 1); |
5d4f98a2 | 259 | if (IS_ERR(cow)) |
be20aa9d CM |
260 | return PTR_ERR(cow); |
261 | ||
262 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
263 | btrfs_set_header_bytenr(cow, cow->start); | |
264 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
265 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
266 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
267 | BTRFS_HEADER_FLAG_RELOC); | |
268 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
269 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
270 | else | |
271 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 272 | |
2b82032c YZ |
273 | write_extent_buffer(cow, root->fs_info->fsid, |
274 | (unsigned long)btrfs_header_fsid(cow), | |
275 | BTRFS_FSID_SIZE); | |
276 | ||
be20aa9d | 277 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 278 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 279 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 280 | else |
66d7e7f0 | 281 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 282 | |
be20aa9d CM |
283 | if (ret) |
284 | return ret; | |
285 | ||
286 | btrfs_mark_buffer_dirty(cow); | |
287 | *cow_ret = cow; | |
288 | return 0; | |
289 | } | |
290 | ||
5d4f98a2 YZ |
291 | /* |
292 | * check if the tree block can be shared by multiple trees | |
293 | */ | |
294 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
295 | struct extent_buffer *buf) | |
296 | { | |
297 | /* | |
298 | * Tree blocks not in refernece counted trees and tree roots | |
299 | * are never shared. If a block was allocated after the last | |
300 | * snapshot and the block was not allocated by tree relocation, | |
301 | * we know the block is not shared. | |
302 | */ | |
303 | if (root->ref_cows && | |
304 | buf != root->node && buf != root->commit_root && | |
305 | (btrfs_header_generation(buf) <= | |
306 | btrfs_root_last_snapshot(&root->root_item) || | |
307 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
308 | return 1; | |
309 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
310 | if (root->ref_cows && | |
311 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
312 | return 1; | |
313 | #endif | |
314 | return 0; | |
315 | } | |
316 | ||
317 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
318 | struct btrfs_root *root, | |
319 | struct extent_buffer *buf, | |
f0486c68 YZ |
320 | struct extent_buffer *cow, |
321 | int *last_ref) | |
5d4f98a2 YZ |
322 | { |
323 | u64 refs; | |
324 | u64 owner; | |
325 | u64 flags; | |
326 | u64 new_flags = 0; | |
327 | int ret; | |
328 | ||
329 | /* | |
330 | * Backrefs update rules: | |
331 | * | |
332 | * Always use full backrefs for extent pointers in tree block | |
333 | * allocated by tree relocation. | |
334 | * | |
335 | * If a shared tree block is no longer referenced by its owner | |
336 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
337 | * use full backrefs for extent pointers in tree block. | |
338 | * | |
339 | * If a tree block is been relocating | |
340 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
341 | * use full backrefs for extent pointers in tree block. | |
342 | * The reason for this is some operations (such as drop tree) | |
343 | * are only allowed for blocks use full backrefs. | |
344 | */ | |
345 | ||
346 | if (btrfs_block_can_be_shared(root, buf)) { | |
347 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
348 | buf->len, &refs, &flags); | |
be1a5564 MF |
349 | if (ret) |
350 | return ret; | |
e5df9573 MF |
351 | if (refs == 0) { |
352 | ret = -EROFS; | |
353 | btrfs_std_error(root->fs_info, ret); | |
354 | return ret; | |
355 | } | |
5d4f98a2 YZ |
356 | } else { |
357 | refs = 1; | |
358 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
359 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
360 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
361 | else | |
362 | flags = 0; | |
363 | } | |
364 | ||
365 | owner = btrfs_header_owner(buf); | |
366 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
367 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
368 | ||
369 | if (refs > 1) { | |
370 | if ((owner == root->root_key.objectid || | |
371 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
372 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 373 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 374 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
375 | |
376 | if (root->root_key.objectid == | |
377 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 378 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 379 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 380 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 381 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
382 | } |
383 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
384 | } else { | |
385 | ||
386 | if (root->root_key.objectid == | |
387 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 388 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 389 | else |
66d7e7f0 | 390 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 391 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
392 | } |
393 | if (new_flags != 0) { | |
394 | ret = btrfs_set_disk_extent_flags(trans, root, | |
395 | buf->start, | |
396 | buf->len, | |
397 | new_flags, 0); | |
be1a5564 MF |
398 | if (ret) |
399 | return ret; | |
5d4f98a2 YZ |
400 | } |
401 | } else { | |
402 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
403 | if (root->root_key.objectid == | |
404 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 405 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 406 | else |
66d7e7f0 | 407 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 408 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 409 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 410 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
411 | } |
412 | clean_tree_block(trans, root, buf); | |
f0486c68 | 413 | *last_ref = 1; |
5d4f98a2 YZ |
414 | } |
415 | return 0; | |
416 | } | |
417 | ||
d352ac68 | 418 | /* |
d397712b CM |
419 | * does the dirty work in cow of a single block. The parent block (if |
420 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
421 | * dirty and returned locked. If you modify the block it needs to be marked | |
422 | * dirty again. | |
d352ac68 CM |
423 | * |
424 | * search_start -- an allocation hint for the new block | |
425 | * | |
d397712b CM |
426 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
427 | * bytes the allocator should try to find free next to the block it returns. | |
428 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 429 | */ |
d397712b | 430 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
431 | struct btrfs_root *root, |
432 | struct extent_buffer *buf, | |
433 | struct extent_buffer *parent, int parent_slot, | |
434 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 435 | u64 search_start, u64 empty_size) |
02217ed2 | 436 | { |
5d4f98a2 | 437 | struct btrfs_disk_key disk_key; |
5f39d397 | 438 | struct extent_buffer *cow; |
be1a5564 | 439 | int level, ret; |
f0486c68 | 440 | int last_ref = 0; |
925baedd | 441 | int unlock_orig = 0; |
5d4f98a2 | 442 | u64 parent_start; |
7bb86316 | 443 | |
925baedd CM |
444 | if (*cow_ret == buf) |
445 | unlock_orig = 1; | |
446 | ||
b9447ef8 | 447 | btrfs_assert_tree_locked(buf); |
925baedd | 448 | |
7bb86316 CM |
449 | WARN_ON(root->ref_cows && trans->transid != |
450 | root->fs_info->running_transaction->transid); | |
6702ed49 | 451 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 452 | |
7bb86316 | 453 | level = btrfs_header_level(buf); |
31840ae1 | 454 | |
5d4f98a2 YZ |
455 | if (level == 0) |
456 | btrfs_item_key(buf, &disk_key, 0); | |
457 | else | |
458 | btrfs_node_key(buf, &disk_key, 0); | |
459 | ||
460 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
461 | if (parent) | |
462 | parent_start = parent->start; | |
463 | else | |
464 | parent_start = 0; | |
465 | } else | |
466 | parent_start = 0; | |
467 | ||
468 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
469 | root->root_key.objectid, &disk_key, | |
66d7e7f0 | 470 | level, search_start, empty_size, 1); |
54aa1f4d CM |
471 | if (IS_ERR(cow)) |
472 | return PTR_ERR(cow); | |
6702ed49 | 473 | |
b4ce94de CM |
474 | /* cow is set to blocking by btrfs_init_new_buffer */ |
475 | ||
5f39d397 | 476 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 477 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 478 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
479 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
480 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
481 | BTRFS_HEADER_FLAG_RELOC); | |
482 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
483 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
484 | else | |
485 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 486 | |
2b82032c YZ |
487 | write_extent_buffer(cow, root->fs_info->fsid, |
488 | (unsigned long)btrfs_header_fsid(cow), | |
489 | BTRFS_FSID_SIZE); | |
490 | ||
be1a5564 | 491 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 492 | if (ret) { |
79787eaa | 493 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
494 | return ret; |
495 | } | |
1a40e23b | 496 | |
3fd0a558 YZ |
497 | if (root->ref_cows) |
498 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
499 | ||
02217ed2 | 500 | if (buf == root->node) { |
925baedd | 501 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
502 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
503 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
504 | parent_start = buf->start; | |
505 | else | |
506 | parent_start = 0; | |
925baedd | 507 | |
5f39d397 | 508 | extent_buffer_get(cow); |
240f62c8 | 509 | rcu_assign_pointer(root->node, cow); |
925baedd | 510 | |
f0486c68 | 511 | btrfs_free_tree_block(trans, root, buf, parent_start, |
66d7e7f0 | 512 | last_ref, 1); |
5f39d397 | 513 | free_extent_buffer(buf); |
0b86a832 | 514 | add_root_to_dirty_list(root); |
02217ed2 | 515 | } else { |
5d4f98a2 YZ |
516 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
517 | parent_start = parent->start; | |
518 | else | |
519 | parent_start = 0; | |
520 | ||
521 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
5f39d397 | 522 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 523 | cow->start); |
74493f7a CM |
524 | btrfs_set_node_ptr_generation(parent, parent_slot, |
525 | trans->transid); | |
d6025579 | 526 | btrfs_mark_buffer_dirty(parent); |
f0486c68 | 527 | btrfs_free_tree_block(trans, root, buf, parent_start, |
66d7e7f0 | 528 | last_ref, 1); |
02217ed2 | 529 | } |
925baedd CM |
530 | if (unlock_orig) |
531 | btrfs_tree_unlock(buf); | |
3083ee2e | 532 | free_extent_buffer_stale(buf); |
ccd467d6 | 533 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 534 | *cow_ret = cow; |
02217ed2 CM |
535 | return 0; |
536 | } | |
537 | ||
5d4f98a2 YZ |
538 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
539 | struct btrfs_root *root, | |
540 | struct extent_buffer *buf) | |
541 | { | |
f1ebcc74 LB |
542 | /* ensure we can see the force_cow */ |
543 | smp_rmb(); | |
544 | ||
545 | /* | |
546 | * We do not need to cow a block if | |
547 | * 1) this block is not created or changed in this transaction; | |
548 | * 2) this block does not belong to TREE_RELOC tree; | |
549 | * 3) the root is not forced COW. | |
550 | * | |
551 | * What is forced COW: | |
552 | * when we create snapshot during commiting the transaction, | |
553 | * after we've finished coping src root, we must COW the shared | |
554 | * block to ensure the metadata consistency. | |
555 | */ | |
5d4f98a2 YZ |
556 | if (btrfs_header_generation(buf) == trans->transid && |
557 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
558 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
559 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
560 | !root->force_cow) | |
5d4f98a2 YZ |
561 | return 0; |
562 | return 1; | |
563 | } | |
564 | ||
d352ac68 CM |
565 | /* |
566 | * cows a single block, see __btrfs_cow_block for the real work. | |
567 | * This version of it has extra checks so that a block isn't cow'd more than | |
568 | * once per transaction, as long as it hasn't been written yet | |
569 | */ | |
d397712b | 570 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
571 | struct btrfs_root *root, struct extent_buffer *buf, |
572 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 573 | struct extent_buffer **cow_ret) |
6702ed49 CM |
574 | { |
575 | u64 search_start; | |
f510cfec | 576 | int ret; |
dc17ff8f | 577 | |
6702ed49 | 578 | if (trans->transaction != root->fs_info->running_transaction) { |
d397712b CM |
579 | printk(KERN_CRIT "trans %llu running %llu\n", |
580 | (unsigned long long)trans->transid, | |
581 | (unsigned long long) | |
6702ed49 CM |
582 | root->fs_info->running_transaction->transid); |
583 | WARN_ON(1); | |
584 | } | |
585 | if (trans->transid != root->fs_info->generation) { | |
d397712b CM |
586 | printk(KERN_CRIT "trans %llu running %llu\n", |
587 | (unsigned long long)trans->transid, | |
588 | (unsigned long long)root->fs_info->generation); | |
6702ed49 CM |
589 | WARN_ON(1); |
590 | } | |
dc17ff8f | 591 | |
5d4f98a2 | 592 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
593 | *cow_ret = buf; |
594 | return 0; | |
595 | } | |
c487685d | 596 | |
0b86a832 | 597 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
598 | |
599 | if (parent) | |
600 | btrfs_set_lock_blocking(parent); | |
601 | btrfs_set_lock_blocking(buf); | |
602 | ||
f510cfec | 603 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 604 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 605 | |
606 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
607 | ||
f510cfec | 608 | return ret; |
6702ed49 CM |
609 | } |
610 | ||
d352ac68 CM |
611 | /* |
612 | * helper function for defrag to decide if two blocks pointed to by a | |
613 | * node are actually close by | |
614 | */ | |
6b80053d | 615 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 616 | { |
6b80053d | 617 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 618 | return 1; |
6b80053d | 619 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
620 | return 1; |
621 | return 0; | |
622 | } | |
623 | ||
081e9573 CM |
624 | /* |
625 | * compare two keys in a memcmp fashion | |
626 | */ | |
627 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
628 | { | |
629 | struct btrfs_key k1; | |
630 | ||
631 | btrfs_disk_key_to_cpu(&k1, disk); | |
632 | ||
20736aba | 633 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
634 | } |
635 | ||
f3465ca4 JB |
636 | /* |
637 | * same as comp_keys only with two btrfs_key's | |
638 | */ | |
5d4f98a2 | 639 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
640 | { |
641 | if (k1->objectid > k2->objectid) | |
642 | return 1; | |
643 | if (k1->objectid < k2->objectid) | |
644 | return -1; | |
645 | if (k1->type > k2->type) | |
646 | return 1; | |
647 | if (k1->type < k2->type) | |
648 | return -1; | |
649 | if (k1->offset > k2->offset) | |
650 | return 1; | |
651 | if (k1->offset < k2->offset) | |
652 | return -1; | |
653 | return 0; | |
654 | } | |
081e9573 | 655 | |
d352ac68 CM |
656 | /* |
657 | * this is used by the defrag code to go through all the | |
658 | * leaves pointed to by a node and reallocate them so that | |
659 | * disk order is close to key order | |
660 | */ | |
6702ed49 | 661 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 662 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
663 | int start_slot, int cache_only, u64 *last_ret, |
664 | struct btrfs_key *progress) | |
6702ed49 | 665 | { |
6b80053d | 666 | struct extent_buffer *cur; |
6702ed49 | 667 | u64 blocknr; |
ca7a79ad | 668 | u64 gen; |
e9d0b13b CM |
669 | u64 search_start = *last_ret; |
670 | u64 last_block = 0; | |
6702ed49 CM |
671 | u64 other; |
672 | u32 parent_nritems; | |
6702ed49 CM |
673 | int end_slot; |
674 | int i; | |
675 | int err = 0; | |
f2183bde | 676 | int parent_level; |
6b80053d CM |
677 | int uptodate; |
678 | u32 blocksize; | |
081e9573 CM |
679 | int progress_passed = 0; |
680 | struct btrfs_disk_key disk_key; | |
6702ed49 | 681 | |
5708b959 CM |
682 | parent_level = btrfs_header_level(parent); |
683 | if (cache_only && parent_level != 1) | |
684 | return 0; | |
685 | ||
d397712b | 686 | if (trans->transaction != root->fs_info->running_transaction) |
6702ed49 | 687 | WARN_ON(1); |
d397712b | 688 | if (trans->transid != root->fs_info->generation) |
6702ed49 | 689 | WARN_ON(1); |
86479a04 | 690 | |
6b80053d | 691 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 692 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
693 | end_slot = parent_nritems; |
694 | ||
695 | if (parent_nritems == 1) | |
696 | return 0; | |
697 | ||
b4ce94de CM |
698 | btrfs_set_lock_blocking(parent); |
699 | ||
6702ed49 CM |
700 | for (i = start_slot; i < end_slot; i++) { |
701 | int close = 1; | |
a6b6e75e | 702 | |
081e9573 CM |
703 | btrfs_node_key(parent, &disk_key, i); |
704 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
705 | continue; | |
706 | ||
707 | progress_passed = 1; | |
6b80053d | 708 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 709 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
710 | if (last_block == 0) |
711 | last_block = blocknr; | |
5708b959 | 712 | |
6702ed49 | 713 | if (i > 0) { |
6b80053d CM |
714 | other = btrfs_node_blockptr(parent, i - 1); |
715 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 716 | } |
0ef3e66b | 717 | if (!close && i < end_slot - 2) { |
6b80053d CM |
718 | other = btrfs_node_blockptr(parent, i + 1); |
719 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 720 | } |
e9d0b13b CM |
721 | if (close) { |
722 | last_block = blocknr; | |
6702ed49 | 723 | continue; |
e9d0b13b | 724 | } |
6702ed49 | 725 | |
6b80053d CM |
726 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
727 | if (cur) | |
b9fab919 | 728 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
729 | else |
730 | uptodate = 0; | |
5708b959 | 731 | if (!cur || !uptodate) { |
6702ed49 | 732 | if (cache_only) { |
6b80053d | 733 | free_extent_buffer(cur); |
6702ed49 CM |
734 | continue; |
735 | } | |
6b80053d CM |
736 | if (!cur) { |
737 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 738 | blocksize, gen); |
97d9a8a4 TI |
739 | if (!cur) |
740 | return -EIO; | |
6b80053d | 741 | } else if (!uptodate) { |
ca7a79ad | 742 | btrfs_read_buffer(cur, gen); |
f2183bde | 743 | } |
6702ed49 | 744 | } |
e9d0b13b | 745 | if (search_start == 0) |
6b80053d | 746 | search_start = last_block; |
e9d0b13b | 747 | |
e7a84565 | 748 | btrfs_tree_lock(cur); |
b4ce94de | 749 | btrfs_set_lock_blocking(cur); |
6b80053d | 750 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 751 | &cur, search_start, |
6b80053d | 752 | min(16 * blocksize, |
9fa8cfe7 | 753 | (end_slot - i) * blocksize)); |
252c38f0 | 754 | if (err) { |
e7a84565 | 755 | btrfs_tree_unlock(cur); |
6b80053d | 756 | free_extent_buffer(cur); |
6702ed49 | 757 | break; |
252c38f0 | 758 | } |
e7a84565 CM |
759 | search_start = cur->start; |
760 | last_block = cur->start; | |
f2183bde | 761 | *last_ret = search_start; |
e7a84565 CM |
762 | btrfs_tree_unlock(cur); |
763 | free_extent_buffer(cur); | |
6702ed49 CM |
764 | } |
765 | return err; | |
766 | } | |
767 | ||
74123bd7 CM |
768 | /* |
769 | * The leaf data grows from end-to-front in the node. | |
770 | * this returns the address of the start of the last item, | |
771 | * which is the stop of the leaf data stack | |
772 | */ | |
123abc88 | 773 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 774 | struct extent_buffer *leaf) |
be0e5c09 | 775 | { |
5f39d397 | 776 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 777 | if (nr == 0) |
123abc88 | 778 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 779 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
780 | } |
781 | ||
aa5d6bed | 782 | |
74123bd7 | 783 | /* |
5f39d397 CM |
784 | * search for key in the extent_buffer. The items start at offset p, |
785 | * and they are item_size apart. There are 'max' items in p. | |
786 | * | |
74123bd7 CM |
787 | * the slot in the array is returned via slot, and it points to |
788 | * the place where you would insert key if it is not found in | |
789 | * the array. | |
790 | * | |
791 | * slot may point to max if the key is bigger than all of the keys | |
792 | */ | |
e02119d5 CM |
793 | static noinline int generic_bin_search(struct extent_buffer *eb, |
794 | unsigned long p, | |
795 | int item_size, struct btrfs_key *key, | |
796 | int max, int *slot) | |
be0e5c09 CM |
797 | { |
798 | int low = 0; | |
799 | int high = max; | |
800 | int mid; | |
801 | int ret; | |
479965d6 | 802 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
803 | struct btrfs_disk_key unaligned; |
804 | unsigned long offset; | |
5f39d397 CM |
805 | char *kaddr = NULL; |
806 | unsigned long map_start = 0; | |
807 | unsigned long map_len = 0; | |
479965d6 | 808 | int err; |
be0e5c09 | 809 | |
d397712b | 810 | while (low < high) { |
be0e5c09 | 811 | mid = (low + high) / 2; |
5f39d397 CM |
812 | offset = p + mid * item_size; |
813 | ||
a6591715 | 814 | if (!kaddr || offset < map_start || |
5f39d397 CM |
815 | (offset + sizeof(struct btrfs_disk_key)) > |
816 | map_start + map_len) { | |
934d375b CM |
817 | |
818 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 819 | sizeof(struct btrfs_disk_key), |
a6591715 | 820 | &kaddr, &map_start, &map_len); |
479965d6 CM |
821 | |
822 | if (!err) { | |
823 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
824 | map_start); | |
825 | } else { | |
826 | read_extent_buffer(eb, &unaligned, | |
827 | offset, sizeof(unaligned)); | |
828 | tmp = &unaligned; | |
829 | } | |
5f39d397 | 830 | |
5f39d397 CM |
831 | } else { |
832 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
833 | map_start); | |
834 | } | |
be0e5c09 CM |
835 | ret = comp_keys(tmp, key); |
836 | ||
837 | if (ret < 0) | |
838 | low = mid + 1; | |
839 | else if (ret > 0) | |
840 | high = mid; | |
841 | else { | |
842 | *slot = mid; | |
843 | return 0; | |
844 | } | |
845 | } | |
846 | *slot = low; | |
847 | return 1; | |
848 | } | |
849 | ||
97571fd0 CM |
850 | /* |
851 | * simple bin_search frontend that does the right thing for | |
852 | * leaves vs nodes | |
853 | */ | |
5f39d397 CM |
854 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
855 | int level, int *slot) | |
be0e5c09 | 856 | { |
f775738f | 857 | if (level == 0) |
5f39d397 CM |
858 | return generic_bin_search(eb, |
859 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 860 | sizeof(struct btrfs_item), |
5f39d397 | 861 | key, btrfs_header_nritems(eb), |
7518a238 | 862 | slot); |
f775738f | 863 | else |
5f39d397 CM |
864 | return generic_bin_search(eb, |
865 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 866 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 867 | key, btrfs_header_nritems(eb), |
7518a238 | 868 | slot); |
be0e5c09 CM |
869 | } |
870 | ||
5d4f98a2 YZ |
871 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
872 | int level, int *slot) | |
873 | { | |
874 | return bin_search(eb, key, level, slot); | |
875 | } | |
876 | ||
f0486c68 YZ |
877 | static void root_add_used(struct btrfs_root *root, u32 size) |
878 | { | |
879 | spin_lock(&root->accounting_lock); | |
880 | btrfs_set_root_used(&root->root_item, | |
881 | btrfs_root_used(&root->root_item) + size); | |
882 | spin_unlock(&root->accounting_lock); | |
883 | } | |
884 | ||
885 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
886 | { | |
887 | spin_lock(&root->accounting_lock); | |
888 | btrfs_set_root_used(&root->root_item, | |
889 | btrfs_root_used(&root->root_item) - size); | |
890 | spin_unlock(&root->accounting_lock); | |
891 | } | |
892 | ||
d352ac68 CM |
893 | /* given a node and slot number, this reads the blocks it points to. The |
894 | * extent buffer is returned with a reference taken (but unlocked). | |
895 | * NULL is returned on error. | |
896 | */ | |
e02119d5 | 897 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 898 | struct extent_buffer *parent, int slot) |
bb803951 | 899 | { |
ca7a79ad | 900 | int level = btrfs_header_level(parent); |
bb803951 CM |
901 | if (slot < 0) |
902 | return NULL; | |
5f39d397 | 903 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 904 | return NULL; |
ca7a79ad CM |
905 | |
906 | BUG_ON(level == 0); | |
907 | ||
db94535d | 908 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
909 | btrfs_level_size(root, level - 1), |
910 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
911 | } |
912 | ||
d352ac68 CM |
913 | /* |
914 | * node level balancing, used to make sure nodes are in proper order for | |
915 | * item deletion. We balance from the top down, so we have to make sure | |
916 | * that a deletion won't leave an node completely empty later on. | |
917 | */ | |
e02119d5 | 918 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
919 | struct btrfs_root *root, |
920 | struct btrfs_path *path, int level) | |
bb803951 | 921 | { |
5f39d397 CM |
922 | struct extent_buffer *right = NULL; |
923 | struct extent_buffer *mid; | |
924 | struct extent_buffer *left = NULL; | |
925 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
926 | int ret = 0; |
927 | int wret; | |
928 | int pslot; | |
bb803951 | 929 | int orig_slot = path->slots[level]; |
79f95c82 | 930 | u64 orig_ptr; |
bb803951 CM |
931 | |
932 | if (level == 0) | |
933 | return 0; | |
934 | ||
5f39d397 | 935 | mid = path->nodes[level]; |
b4ce94de | 936 | |
bd681513 CM |
937 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
938 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
939 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
940 | ||
1d4f8a0c | 941 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 942 | |
a05a9bb1 | 943 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 944 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
945 | pslot = path->slots[level + 1]; |
946 | } | |
bb803951 | 947 | |
40689478 CM |
948 | /* |
949 | * deal with the case where there is only one pointer in the root | |
950 | * by promoting the node below to a root | |
951 | */ | |
5f39d397 CM |
952 | if (!parent) { |
953 | struct extent_buffer *child; | |
bb803951 | 954 | |
5f39d397 | 955 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
956 | return 0; |
957 | ||
958 | /* promote the child to a root */ | |
5f39d397 | 959 | child = read_node_slot(root, mid, 0); |
305a26af MF |
960 | if (!child) { |
961 | ret = -EROFS; | |
962 | btrfs_std_error(root->fs_info, ret); | |
963 | goto enospc; | |
964 | } | |
965 | ||
925baedd | 966 | btrfs_tree_lock(child); |
b4ce94de | 967 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 968 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
969 | if (ret) { |
970 | btrfs_tree_unlock(child); | |
971 | free_extent_buffer(child); | |
972 | goto enospc; | |
973 | } | |
2f375ab9 | 974 | |
240f62c8 | 975 | rcu_assign_pointer(root->node, child); |
925baedd | 976 | |
0b86a832 | 977 | add_root_to_dirty_list(root); |
925baedd | 978 | btrfs_tree_unlock(child); |
b4ce94de | 979 | |
925baedd | 980 | path->locks[level] = 0; |
bb803951 | 981 | path->nodes[level] = NULL; |
5f39d397 | 982 | clean_tree_block(trans, root, mid); |
925baedd | 983 | btrfs_tree_unlock(mid); |
bb803951 | 984 | /* once for the path */ |
5f39d397 | 985 | free_extent_buffer(mid); |
f0486c68 YZ |
986 | |
987 | root_sub_used(root, mid->len); | |
66d7e7f0 | 988 | btrfs_free_tree_block(trans, root, mid, 0, 1, 0); |
bb803951 | 989 | /* once for the root ptr */ |
3083ee2e | 990 | free_extent_buffer_stale(mid); |
f0486c68 | 991 | return 0; |
bb803951 | 992 | } |
5f39d397 | 993 | if (btrfs_header_nritems(mid) > |
123abc88 | 994 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
995 | return 0; |
996 | ||
559af821 | 997 | btrfs_header_nritems(mid); |
54aa1f4d | 998 | |
5f39d397 CM |
999 | left = read_node_slot(root, parent, pslot - 1); |
1000 | if (left) { | |
925baedd | 1001 | btrfs_tree_lock(left); |
b4ce94de | 1002 | btrfs_set_lock_blocking(left); |
5f39d397 | 1003 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1004 | parent, pslot - 1, &left); |
54aa1f4d CM |
1005 | if (wret) { |
1006 | ret = wret; | |
1007 | goto enospc; | |
1008 | } | |
2cc58cf2 | 1009 | } |
5f39d397 CM |
1010 | right = read_node_slot(root, parent, pslot + 1); |
1011 | if (right) { | |
925baedd | 1012 | btrfs_tree_lock(right); |
b4ce94de | 1013 | btrfs_set_lock_blocking(right); |
5f39d397 | 1014 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1015 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1016 | if (wret) { |
1017 | ret = wret; | |
1018 | goto enospc; | |
1019 | } | |
1020 | } | |
1021 | ||
1022 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1023 | if (left) { |
1024 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1025 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1026 | if (wret < 0) |
1027 | ret = wret; | |
559af821 | 1028 | btrfs_header_nritems(mid); |
bb803951 | 1029 | } |
79f95c82 CM |
1030 | |
1031 | /* | |
1032 | * then try to empty the right most buffer into the middle | |
1033 | */ | |
5f39d397 | 1034 | if (right) { |
971a1f66 | 1035 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1036 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1037 | ret = wret; |
5f39d397 | 1038 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1039 | clean_tree_block(trans, root, right); |
925baedd | 1040 | btrfs_tree_unlock(right); |
143bede5 | 1041 | del_ptr(trans, root, path, level + 1, pslot + 1); |
f0486c68 | 1042 | root_sub_used(root, right->len); |
66d7e7f0 | 1043 | btrfs_free_tree_block(trans, root, right, 0, 1, 0); |
3083ee2e | 1044 | free_extent_buffer_stale(right); |
f0486c68 | 1045 | right = NULL; |
bb803951 | 1046 | } else { |
5f39d397 CM |
1047 | struct btrfs_disk_key right_key; |
1048 | btrfs_node_key(right, &right_key, 0); | |
1049 | btrfs_set_node_key(parent, &right_key, pslot + 1); | |
1050 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1051 | } |
1052 | } | |
5f39d397 | 1053 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1054 | /* |
1055 | * we're not allowed to leave a node with one item in the | |
1056 | * tree during a delete. A deletion from lower in the tree | |
1057 | * could try to delete the only pointer in this node. | |
1058 | * So, pull some keys from the left. | |
1059 | * There has to be a left pointer at this point because | |
1060 | * otherwise we would have pulled some pointers from the | |
1061 | * right | |
1062 | */ | |
305a26af MF |
1063 | if (!left) { |
1064 | ret = -EROFS; | |
1065 | btrfs_std_error(root->fs_info, ret); | |
1066 | goto enospc; | |
1067 | } | |
5f39d397 | 1068 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1069 | if (wret < 0) { |
79f95c82 | 1070 | ret = wret; |
54aa1f4d CM |
1071 | goto enospc; |
1072 | } | |
bce4eae9 CM |
1073 | if (wret == 1) { |
1074 | wret = push_node_left(trans, root, left, mid, 1); | |
1075 | if (wret < 0) | |
1076 | ret = wret; | |
1077 | } | |
79f95c82 CM |
1078 | BUG_ON(wret == 1); |
1079 | } | |
5f39d397 | 1080 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1081 | clean_tree_block(trans, root, mid); |
925baedd | 1082 | btrfs_tree_unlock(mid); |
143bede5 | 1083 | del_ptr(trans, root, path, level + 1, pslot); |
f0486c68 | 1084 | root_sub_used(root, mid->len); |
66d7e7f0 | 1085 | btrfs_free_tree_block(trans, root, mid, 0, 1, 0); |
3083ee2e | 1086 | free_extent_buffer_stale(mid); |
f0486c68 | 1087 | mid = NULL; |
79f95c82 CM |
1088 | } else { |
1089 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1090 | struct btrfs_disk_key mid_key; |
1091 | btrfs_node_key(mid, &mid_key, 0); | |
1092 | btrfs_set_node_key(parent, &mid_key, pslot); | |
1093 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1094 | } |
bb803951 | 1095 | |
79f95c82 | 1096 | /* update the path */ |
5f39d397 CM |
1097 | if (left) { |
1098 | if (btrfs_header_nritems(left) > orig_slot) { | |
1099 | extent_buffer_get(left); | |
925baedd | 1100 | /* left was locked after cow */ |
5f39d397 | 1101 | path->nodes[level] = left; |
bb803951 CM |
1102 | path->slots[level + 1] -= 1; |
1103 | path->slots[level] = orig_slot; | |
925baedd CM |
1104 | if (mid) { |
1105 | btrfs_tree_unlock(mid); | |
5f39d397 | 1106 | free_extent_buffer(mid); |
925baedd | 1107 | } |
bb803951 | 1108 | } else { |
5f39d397 | 1109 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1110 | path->slots[level] = orig_slot; |
1111 | } | |
1112 | } | |
79f95c82 | 1113 | /* double check we haven't messed things up */ |
e20d96d6 | 1114 | if (orig_ptr != |
5f39d397 | 1115 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1116 | BUG(); |
54aa1f4d | 1117 | enospc: |
925baedd CM |
1118 | if (right) { |
1119 | btrfs_tree_unlock(right); | |
5f39d397 | 1120 | free_extent_buffer(right); |
925baedd CM |
1121 | } |
1122 | if (left) { | |
1123 | if (path->nodes[level] != left) | |
1124 | btrfs_tree_unlock(left); | |
5f39d397 | 1125 | free_extent_buffer(left); |
925baedd | 1126 | } |
bb803951 CM |
1127 | return ret; |
1128 | } | |
1129 | ||
d352ac68 CM |
1130 | /* Node balancing for insertion. Here we only split or push nodes around |
1131 | * when they are completely full. This is also done top down, so we | |
1132 | * have to be pessimistic. | |
1133 | */ | |
d397712b | 1134 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1135 | struct btrfs_root *root, |
1136 | struct btrfs_path *path, int level) | |
e66f709b | 1137 | { |
5f39d397 CM |
1138 | struct extent_buffer *right = NULL; |
1139 | struct extent_buffer *mid; | |
1140 | struct extent_buffer *left = NULL; | |
1141 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1142 | int ret = 0; |
1143 | int wret; | |
1144 | int pslot; | |
1145 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1146 | |
1147 | if (level == 0) | |
1148 | return 1; | |
1149 | ||
5f39d397 | 1150 | mid = path->nodes[level]; |
7bb86316 | 1151 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1152 | |
a05a9bb1 | 1153 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1154 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1155 | pslot = path->slots[level + 1]; |
1156 | } | |
e66f709b | 1157 | |
5f39d397 | 1158 | if (!parent) |
e66f709b | 1159 | return 1; |
e66f709b | 1160 | |
5f39d397 | 1161 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1162 | |
1163 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1164 | if (left) { |
e66f709b | 1165 | u32 left_nr; |
925baedd CM |
1166 | |
1167 | btrfs_tree_lock(left); | |
b4ce94de CM |
1168 | btrfs_set_lock_blocking(left); |
1169 | ||
5f39d397 | 1170 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1171 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1172 | wret = 1; | |
1173 | } else { | |
5f39d397 | 1174 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1175 | pslot - 1, &left); |
54aa1f4d CM |
1176 | if (ret) |
1177 | wret = 1; | |
1178 | else { | |
54aa1f4d | 1179 | wret = push_node_left(trans, root, |
971a1f66 | 1180 | left, mid, 0); |
54aa1f4d | 1181 | } |
33ade1f8 | 1182 | } |
e66f709b CM |
1183 | if (wret < 0) |
1184 | ret = wret; | |
1185 | if (wret == 0) { | |
5f39d397 | 1186 | struct btrfs_disk_key disk_key; |
e66f709b | 1187 | orig_slot += left_nr; |
5f39d397 CM |
1188 | btrfs_node_key(mid, &disk_key, 0); |
1189 | btrfs_set_node_key(parent, &disk_key, pslot); | |
1190 | btrfs_mark_buffer_dirty(parent); | |
1191 | if (btrfs_header_nritems(left) > orig_slot) { | |
1192 | path->nodes[level] = left; | |
e66f709b CM |
1193 | path->slots[level + 1] -= 1; |
1194 | path->slots[level] = orig_slot; | |
925baedd | 1195 | btrfs_tree_unlock(mid); |
5f39d397 | 1196 | free_extent_buffer(mid); |
e66f709b CM |
1197 | } else { |
1198 | orig_slot -= | |
5f39d397 | 1199 | btrfs_header_nritems(left); |
e66f709b | 1200 | path->slots[level] = orig_slot; |
925baedd | 1201 | btrfs_tree_unlock(left); |
5f39d397 | 1202 | free_extent_buffer(left); |
e66f709b | 1203 | } |
e66f709b CM |
1204 | return 0; |
1205 | } | |
925baedd | 1206 | btrfs_tree_unlock(left); |
5f39d397 | 1207 | free_extent_buffer(left); |
e66f709b | 1208 | } |
925baedd | 1209 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1210 | |
1211 | /* | |
1212 | * then try to empty the right most buffer into the middle | |
1213 | */ | |
5f39d397 | 1214 | if (right) { |
33ade1f8 | 1215 | u32 right_nr; |
b4ce94de | 1216 | |
925baedd | 1217 | btrfs_tree_lock(right); |
b4ce94de CM |
1218 | btrfs_set_lock_blocking(right); |
1219 | ||
5f39d397 | 1220 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1221 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1222 | wret = 1; | |
1223 | } else { | |
5f39d397 CM |
1224 | ret = btrfs_cow_block(trans, root, right, |
1225 | parent, pslot + 1, | |
9fa8cfe7 | 1226 | &right); |
54aa1f4d CM |
1227 | if (ret) |
1228 | wret = 1; | |
1229 | else { | |
54aa1f4d | 1230 | wret = balance_node_right(trans, root, |
5f39d397 | 1231 | right, mid); |
54aa1f4d | 1232 | } |
33ade1f8 | 1233 | } |
e66f709b CM |
1234 | if (wret < 0) |
1235 | ret = wret; | |
1236 | if (wret == 0) { | |
5f39d397 CM |
1237 | struct btrfs_disk_key disk_key; |
1238 | ||
1239 | btrfs_node_key(right, &disk_key, 0); | |
1240 | btrfs_set_node_key(parent, &disk_key, pslot + 1); | |
1241 | btrfs_mark_buffer_dirty(parent); | |
1242 | ||
1243 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1244 | path->nodes[level] = right; | |
e66f709b CM |
1245 | path->slots[level + 1] += 1; |
1246 | path->slots[level] = orig_slot - | |
5f39d397 | 1247 | btrfs_header_nritems(mid); |
925baedd | 1248 | btrfs_tree_unlock(mid); |
5f39d397 | 1249 | free_extent_buffer(mid); |
e66f709b | 1250 | } else { |
925baedd | 1251 | btrfs_tree_unlock(right); |
5f39d397 | 1252 | free_extent_buffer(right); |
e66f709b | 1253 | } |
e66f709b CM |
1254 | return 0; |
1255 | } | |
925baedd | 1256 | btrfs_tree_unlock(right); |
5f39d397 | 1257 | free_extent_buffer(right); |
e66f709b | 1258 | } |
e66f709b CM |
1259 | return 1; |
1260 | } | |
1261 | ||
3c69faec | 1262 | /* |
d352ac68 CM |
1263 | * readahead one full node of leaves, finding things that are close |
1264 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1265 | */ |
c8c42864 CM |
1266 | static void reada_for_search(struct btrfs_root *root, |
1267 | struct btrfs_path *path, | |
1268 | int level, int slot, u64 objectid) | |
3c69faec | 1269 | { |
5f39d397 | 1270 | struct extent_buffer *node; |
01f46658 | 1271 | struct btrfs_disk_key disk_key; |
3c69faec | 1272 | u32 nritems; |
3c69faec | 1273 | u64 search; |
a7175319 | 1274 | u64 target; |
6b80053d | 1275 | u64 nread = 0; |
cb25c2ea | 1276 | u64 gen; |
3c69faec | 1277 | int direction = path->reada; |
5f39d397 | 1278 | struct extent_buffer *eb; |
6b80053d CM |
1279 | u32 nr; |
1280 | u32 blocksize; | |
1281 | u32 nscan = 0; | |
db94535d | 1282 | |
a6b6e75e | 1283 | if (level != 1) |
6702ed49 CM |
1284 | return; |
1285 | ||
1286 | if (!path->nodes[level]) | |
3c69faec CM |
1287 | return; |
1288 | ||
5f39d397 | 1289 | node = path->nodes[level]; |
925baedd | 1290 | |
3c69faec | 1291 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1292 | blocksize = btrfs_level_size(root, level - 1); |
1293 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1294 | if (eb) { |
1295 | free_extent_buffer(eb); | |
3c69faec CM |
1296 | return; |
1297 | } | |
1298 | ||
a7175319 | 1299 | target = search; |
6b80053d | 1300 | |
5f39d397 | 1301 | nritems = btrfs_header_nritems(node); |
6b80053d | 1302 | nr = slot; |
25b8b936 | 1303 | |
d397712b | 1304 | while (1) { |
6b80053d CM |
1305 | if (direction < 0) { |
1306 | if (nr == 0) | |
1307 | break; | |
1308 | nr--; | |
1309 | } else if (direction > 0) { | |
1310 | nr++; | |
1311 | if (nr >= nritems) | |
1312 | break; | |
3c69faec | 1313 | } |
01f46658 CM |
1314 | if (path->reada < 0 && objectid) { |
1315 | btrfs_node_key(node, &disk_key, nr); | |
1316 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1317 | break; | |
1318 | } | |
6b80053d | 1319 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
1320 | if ((search <= target && target - search <= 65536) || |
1321 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 1322 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 1323 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
1324 | nread += blocksize; |
1325 | } | |
1326 | nscan++; | |
a7175319 | 1327 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 1328 | break; |
3c69faec CM |
1329 | } |
1330 | } | |
925baedd | 1331 | |
b4ce94de CM |
1332 | /* |
1333 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
1334 | * cache | |
1335 | */ | |
1336 | static noinline int reada_for_balance(struct btrfs_root *root, | |
1337 | struct btrfs_path *path, int level) | |
1338 | { | |
1339 | int slot; | |
1340 | int nritems; | |
1341 | struct extent_buffer *parent; | |
1342 | struct extent_buffer *eb; | |
1343 | u64 gen; | |
1344 | u64 block1 = 0; | |
1345 | u64 block2 = 0; | |
1346 | int ret = 0; | |
1347 | int blocksize; | |
1348 | ||
8c594ea8 | 1349 | parent = path->nodes[level + 1]; |
b4ce94de CM |
1350 | if (!parent) |
1351 | return 0; | |
1352 | ||
1353 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1354 | slot = path->slots[level + 1]; |
b4ce94de CM |
1355 | blocksize = btrfs_level_size(root, level); |
1356 | ||
1357 | if (slot > 0) { | |
1358 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
1359 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
1360 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
1361 | /* |
1362 | * if we get -eagain from btrfs_buffer_uptodate, we | |
1363 | * don't want to return eagain here. That will loop | |
1364 | * forever | |
1365 | */ | |
1366 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
1367 | block1 = 0; |
1368 | free_extent_buffer(eb); | |
1369 | } | |
8c594ea8 | 1370 | if (slot + 1 < nritems) { |
b4ce94de CM |
1371 | block2 = btrfs_node_blockptr(parent, slot + 1); |
1372 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
1373 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 1374 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
1375 | block2 = 0; |
1376 | free_extent_buffer(eb); | |
1377 | } | |
1378 | if (block1 || block2) { | |
1379 | ret = -EAGAIN; | |
8c594ea8 CM |
1380 | |
1381 | /* release the whole path */ | |
b3b4aa74 | 1382 | btrfs_release_path(path); |
8c594ea8 CM |
1383 | |
1384 | /* read the blocks */ | |
b4ce94de CM |
1385 | if (block1) |
1386 | readahead_tree_block(root, block1, blocksize, 0); | |
1387 | if (block2) | |
1388 | readahead_tree_block(root, block2, blocksize, 0); | |
1389 | ||
1390 | if (block1) { | |
1391 | eb = read_tree_block(root, block1, blocksize, 0); | |
1392 | free_extent_buffer(eb); | |
1393 | } | |
8c594ea8 | 1394 | if (block2) { |
b4ce94de CM |
1395 | eb = read_tree_block(root, block2, blocksize, 0); |
1396 | free_extent_buffer(eb); | |
1397 | } | |
1398 | } | |
1399 | return ret; | |
1400 | } | |
1401 | ||
1402 | ||
d352ac68 | 1403 | /* |
d397712b CM |
1404 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1405 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1406 | * operations on the tree might require changing key pointers higher up in the | |
1407 | * tree. | |
d352ac68 | 1408 | * |
d397712b CM |
1409 | * callers might also have set path->keep_locks, which tells this code to keep |
1410 | * the lock if the path points to the last slot in the block. This is part of | |
1411 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1412 | * |
d397712b CM |
1413 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1414 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1415 | */ |
e02119d5 | 1416 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1417 | int lowest_unlock, int min_write_lock_level, |
1418 | int *write_lock_level) | |
925baedd CM |
1419 | { |
1420 | int i; | |
1421 | int skip_level = level; | |
051e1b9f | 1422 | int no_skips = 0; |
925baedd CM |
1423 | struct extent_buffer *t; |
1424 | ||
1425 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1426 | if (!path->nodes[i]) | |
1427 | break; | |
1428 | if (!path->locks[i]) | |
1429 | break; | |
051e1b9f | 1430 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1431 | skip_level = i + 1; |
1432 | continue; | |
1433 | } | |
051e1b9f | 1434 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1435 | u32 nritems; |
1436 | t = path->nodes[i]; | |
1437 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1438 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1439 | skip_level = i + 1; |
1440 | continue; | |
1441 | } | |
1442 | } | |
051e1b9f CM |
1443 | if (skip_level < i && i >= lowest_unlock) |
1444 | no_skips = 1; | |
1445 | ||
925baedd CM |
1446 | t = path->nodes[i]; |
1447 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 1448 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 1449 | path->locks[i] = 0; |
f7c79f30 CM |
1450 | if (write_lock_level && |
1451 | i > min_write_lock_level && | |
1452 | i <= *write_lock_level) { | |
1453 | *write_lock_level = i - 1; | |
1454 | } | |
925baedd CM |
1455 | } |
1456 | } | |
1457 | } | |
1458 | ||
b4ce94de CM |
1459 | /* |
1460 | * This releases any locks held in the path starting at level and | |
1461 | * going all the way up to the root. | |
1462 | * | |
1463 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
1464 | * corner cases, such as COW of the block at slot zero in the node. This | |
1465 | * ignores those rules, and it should only be called when there are no | |
1466 | * more updates to be done higher up in the tree. | |
1467 | */ | |
1468 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
1469 | { | |
1470 | int i; | |
1471 | ||
5d4f98a2 | 1472 | if (path->keep_locks) |
b4ce94de CM |
1473 | return; |
1474 | ||
1475 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1476 | if (!path->nodes[i]) | |
12f4dacc | 1477 | continue; |
b4ce94de | 1478 | if (!path->locks[i]) |
12f4dacc | 1479 | continue; |
bd681513 | 1480 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
1481 | path->locks[i] = 0; |
1482 | } | |
1483 | } | |
1484 | ||
c8c42864 CM |
1485 | /* |
1486 | * helper function for btrfs_search_slot. The goal is to find a block | |
1487 | * in cache without setting the path to blocking. If we find the block | |
1488 | * we return zero and the path is unchanged. | |
1489 | * | |
1490 | * If we can't find the block, we set the path blocking and do some | |
1491 | * reada. -EAGAIN is returned and the search must be repeated. | |
1492 | */ | |
1493 | static int | |
1494 | read_block_for_search(struct btrfs_trans_handle *trans, | |
1495 | struct btrfs_root *root, struct btrfs_path *p, | |
1496 | struct extent_buffer **eb_ret, int level, int slot, | |
1497 | struct btrfs_key *key) | |
1498 | { | |
1499 | u64 blocknr; | |
1500 | u64 gen; | |
1501 | u32 blocksize; | |
1502 | struct extent_buffer *b = *eb_ret; | |
1503 | struct extent_buffer *tmp; | |
76a05b35 | 1504 | int ret; |
c8c42864 CM |
1505 | |
1506 | blocknr = btrfs_node_blockptr(b, slot); | |
1507 | gen = btrfs_node_ptr_generation(b, slot); | |
1508 | blocksize = btrfs_level_size(root, level - 1); | |
1509 | ||
1510 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 1511 | if (tmp) { |
b9fab919 CM |
1512 | /* first we do an atomic uptodate check */ |
1513 | if (btrfs_buffer_uptodate(tmp, 0, 1) > 0) { | |
1514 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { | |
cb44921a CM |
1515 | /* |
1516 | * we found an up to date block without | |
1517 | * sleeping, return | |
1518 | * right away | |
1519 | */ | |
1520 | *eb_ret = tmp; | |
1521 | return 0; | |
1522 | } | |
1523 | /* the pages were up to date, but we failed | |
1524 | * the generation number check. Do a full | |
1525 | * read for the generation number that is correct. | |
1526 | * We must do this without dropping locks so | |
1527 | * we can trust our generation number | |
1528 | */ | |
1529 | free_extent_buffer(tmp); | |
bd681513 CM |
1530 | btrfs_set_path_blocking(p); |
1531 | ||
b9fab919 | 1532 | /* now we're allowed to do a blocking uptodate check */ |
cb44921a | 1533 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
b9fab919 | 1534 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 0) > 0) { |
cb44921a CM |
1535 | *eb_ret = tmp; |
1536 | return 0; | |
1537 | } | |
1538 | free_extent_buffer(tmp); | |
b3b4aa74 | 1539 | btrfs_release_path(p); |
cb44921a CM |
1540 | return -EIO; |
1541 | } | |
c8c42864 CM |
1542 | } |
1543 | ||
1544 | /* | |
1545 | * reduce lock contention at high levels | |
1546 | * of the btree by dropping locks before | |
76a05b35 CM |
1547 | * we read. Don't release the lock on the current |
1548 | * level because we need to walk this node to figure | |
1549 | * out which blocks to read. | |
c8c42864 | 1550 | */ |
8c594ea8 CM |
1551 | btrfs_unlock_up_safe(p, level + 1); |
1552 | btrfs_set_path_blocking(p); | |
1553 | ||
cb44921a | 1554 | free_extent_buffer(tmp); |
c8c42864 CM |
1555 | if (p->reada) |
1556 | reada_for_search(root, p, level, slot, key->objectid); | |
1557 | ||
b3b4aa74 | 1558 | btrfs_release_path(p); |
76a05b35 CM |
1559 | |
1560 | ret = -EAGAIN; | |
5bdd3536 | 1561 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
1562 | if (tmp) { |
1563 | /* | |
1564 | * If the read above didn't mark this buffer up to date, | |
1565 | * it will never end up being up to date. Set ret to EIO now | |
1566 | * and give up so that our caller doesn't loop forever | |
1567 | * on our EAGAINs. | |
1568 | */ | |
b9fab919 | 1569 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 1570 | ret = -EIO; |
c8c42864 | 1571 | free_extent_buffer(tmp); |
76a05b35 CM |
1572 | } |
1573 | return ret; | |
c8c42864 CM |
1574 | } |
1575 | ||
1576 | /* | |
1577 | * helper function for btrfs_search_slot. This does all of the checks | |
1578 | * for node-level blocks and does any balancing required based on | |
1579 | * the ins_len. | |
1580 | * | |
1581 | * If no extra work was required, zero is returned. If we had to | |
1582 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1583 | * start over | |
1584 | */ | |
1585 | static int | |
1586 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1587 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1588 | struct extent_buffer *b, int level, int ins_len, |
1589 | int *write_lock_level) | |
c8c42864 CM |
1590 | { |
1591 | int ret; | |
1592 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
1593 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
1594 | int sret; | |
1595 | ||
bd681513 CM |
1596 | if (*write_lock_level < level + 1) { |
1597 | *write_lock_level = level + 1; | |
1598 | btrfs_release_path(p); | |
1599 | goto again; | |
1600 | } | |
1601 | ||
c8c42864 CM |
1602 | sret = reada_for_balance(root, p, level); |
1603 | if (sret) | |
1604 | goto again; | |
1605 | ||
1606 | btrfs_set_path_blocking(p); | |
1607 | sret = split_node(trans, root, p, level); | |
bd681513 | 1608 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
1609 | |
1610 | BUG_ON(sret > 0); | |
1611 | if (sret) { | |
1612 | ret = sret; | |
1613 | goto done; | |
1614 | } | |
1615 | b = p->nodes[level]; | |
1616 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 1617 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
1618 | int sret; |
1619 | ||
bd681513 CM |
1620 | if (*write_lock_level < level + 1) { |
1621 | *write_lock_level = level + 1; | |
1622 | btrfs_release_path(p); | |
1623 | goto again; | |
1624 | } | |
1625 | ||
c8c42864 CM |
1626 | sret = reada_for_balance(root, p, level); |
1627 | if (sret) | |
1628 | goto again; | |
1629 | ||
1630 | btrfs_set_path_blocking(p); | |
1631 | sret = balance_level(trans, root, p, level); | |
bd681513 | 1632 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
1633 | |
1634 | if (sret) { | |
1635 | ret = sret; | |
1636 | goto done; | |
1637 | } | |
1638 | b = p->nodes[level]; | |
1639 | if (!b) { | |
b3b4aa74 | 1640 | btrfs_release_path(p); |
c8c42864 CM |
1641 | goto again; |
1642 | } | |
1643 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1644 | } | |
1645 | return 0; | |
1646 | ||
1647 | again: | |
1648 | ret = -EAGAIN; | |
1649 | done: | |
1650 | return ret; | |
1651 | } | |
1652 | ||
74123bd7 CM |
1653 | /* |
1654 | * look for key in the tree. path is filled in with nodes along the way | |
1655 | * if key is found, we return zero and you can find the item in the leaf | |
1656 | * level of the path (level 0) | |
1657 | * | |
1658 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
1659 | * be inserted, and 1 is returned. If there are other errors during the |
1660 | * search a negative error number is returned. | |
97571fd0 CM |
1661 | * |
1662 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
1663 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
1664 | * possible) | |
74123bd7 | 1665 | */ |
e089f05c CM |
1666 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
1667 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
1668 | ins_len, int cow) | |
be0e5c09 | 1669 | { |
5f39d397 | 1670 | struct extent_buffer *b; |
be0e5c09 CM |
1671 | int slot; |
1672 | int ret; | |
33c66f43 | 1673 | int err; |
be0e5c09 | 1674 | int level; |
925baedd | 1675 | int lowest_unlock = 1; |
bd681513 CM |
1676 | int root_lock; |
1677 | /* everything at write_lock_level or lower must be write locked */ | |
1678 | int write_lock_level = 0; | |
9f3a7427 | 1679 | u8 lowest_level = 0; |
f7c79f30 | 1680 | int min_write_lock_level; |
9f3a7427 | 1681 | |
6702ed49 | 1682 | lowest_level = p->lowest_level; |
323ac95b | 1683 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1684 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 1685 | |
bd681513 | 1686 | if (ins_len < 0) { |
925baedd | 1687 | lowest_unlock = 2; |
65b51a00 | 1688 | |
bd681513 CM |
1689 | /* when we are removing items, we might have to go up to level |
1690 | * two as we update tree pointers Make sure we keep write | |
1691 | * for those levels as well | |
1692 | */ | |
1693 | write_lock_level = 2; | |
1694 | } else if (ins_len > 0) { | |
1695 | /* | |
1696 | * for inserting items, make sure we have a write lock on | |
1697 | * level 1 so we can update keys | |
1698 | */ | |
1699 | write_lock_level = 1; | |
1700 | } | |
1701 | ||
1702 | if (!cow) | |
1703 | write_lock_level = -1; | |
1704 | ||
1705 | if (cow && (p->keep_locks || p->lowest_level)) | |
1706 | write_lock_level = BTRFS_MAX_LEVEL; | |
1707 | ||
f7c79f30 CM |
1708 | min_write_lock_level = write_lock_level; |
1709 | ||
bb803951 | 1710 | again: |
bd681513 CM |
1711 | /* |
1712 | * we try very hard to do read locks on the root | |
1713 | */ | |
1714 | root_lock = BTRFS_READ_LOCK; | |
1715 | level = 0; | |
5d4f98a2 | 1716 | if (p->search_commit_root) { |
bd681513 CM |
1717 | /* |
1718 | * the commit roots are read only | |
1719 | * so we always do read locks | |
1720 | */ | |
5d4f98a2 YZ |
1721 | b = root->commit_root; |
1722 | extent_buffer_get(b); | |
bd681513 | 1723 | level = btrfs_header_level(b); |
5d4f98a2 | 1724 | if (!p->skip_locking) |
bd681513 | 1725 | btrfs_tree_read_lock(b); |
5d4f98a2 | 1726 | } else { |
bd681513 | 1727 | if (p->skip_locking) { |
5d4f98a2 | 1728 | b = btrfs_root_node(root); |
bd681513 CM |
1729 | level = btrfs_header_level(b); |
1730 | } else { | |
1731 | /* we don't know the level of the root node | |
1732 | * until we actually have it read locked | |
1733 | */ | |
1734 | b = btrfs_read_lock_root_node(root); | |
1735 | level = btrfs_header_level(b); | |
1736 | if (level <= write_lock_level) { | |
1737 | /* whoops, must trade for write lock */ | |
1738 | btrfs_tree_read_unlock(b); | |
1739 | free_extent_buffer(b); | |
1740 | b = btrfs_lock_root_node(root); | |
1741 | root_lock = BTRFS_WRITE_LOCK; | |
1742 | ||
1743 | /* the level might have changed, check again */ | |
1744 | level = btrfs_header_level(b); | |
1745 | } | |
1746 | } | |
5d4f98a2 | 1747 | } |
bd681513 CM |
1748 | p->nodes[level] = b; |
1749 | if (!p->skip_locking) | |
1750 | p->locks[level] = root_lock; | |
925baedd | 1751 | |
eb60ceac | 1752 | while (b) { |
5f39d397 | 1753 | level = btrfs_header_level(b); |
65b51a00 CM |
1754 | |
1755 | /* | |
1756 | * setup the path here so we can release it under lock | |
1757 | * contention with the cow code | |
1758 | */ | |
02217ed2 | 1759 | if (cow) { |
c8c42864 CM |
1760 | /* |
1761 | * if we don't really need to cow this block | |
1762 | * then we don't want to set the path blocking, | |
1763 | * so we test it here | |
1764 | */ | |
5d4f98a2 | 1765 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 1766 | goto cow_done; |
5d4f98a2 | 1767 | |
b4ce94de CM |
1768 | btrfs_set_path_blocking(p); |
1769 | ||
bd681513 CM |
1770 | /* |
1771 | * must have write locks on this node and the | |
1772 | * parent | |
1773 | */ | |
1774 | if (level + 1 > write_lock_level) { | |
1775 | write_lock_level = level + 1; | |
1776 | btrfs_release_path(p); | |
1777 | goto again; | |
1778 | } | |
1779 | ||
33c66f43 YZ |
1780 | err = btrfs_cow_block(trans, root, b, |
1781 | p->nodes[level + 1], | |
1782 | p->slots[level + 1], &b); | |
1783 | if (err) { | |
33c66f43 | 1784 | ret = err; |
65b51a00 | 1785 | goto done; |
54aa1f4d | 1786 | } |
02217ed2 | 1787 | } |
65b51a00 | 1788 | cow_done: |
02217ed2 | 1789 | BUG_ON(!cow && ins_len); |
65b51a00 | 1790 | |
eb60ceac | 1791 | p->nodes[level] = b; |
bd681513 | 1792 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
1793 | |
1794 | /* | |
1795 | * we have a lock on b and as long as we aren't changing | |
1796 | * the tree, there is no way to for the items in b to change. | |
1797 | * It is safe to drop the lock on our parent before we | |
1798 | * go through the expensive btree search on b. | |
1799 | * | |
1800 | * If cow is true, then we might be changing slot zero, | |
1801 | * which may require changing the parent. So, we can't | |
1802 | * drop the lock until after we know which slot we're | |
1803 | * operating on. | |
1804 | */ | |
1805 | if (!cow) | |
1806 | btrfs_unlock_up_safe(p, level + 1); | |
1807 | ||
5f39d397 | 1808 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 1809 | |
5f39d397 | 1810 | if (level != 0) { |
33c66f43 YZ |
1811 | int dec = 0; |
1812 | if (ret && slot > 0) { | |
1813 | dec = 1; | |
be0e5c09 | 1814 | slot -= 1; |
33c66f43 | 1815 | } |
be0e5c09 | 1816 | p->slots[level] = slot; |
33c66f43 | 1817 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 1818 | ins_len, &write_lock_level); |
33c66f43 | 1819 | if (err == -EAGAIN) |
c8c42864 | 1820 | goto again; |
33c66f43 YZ |
1821 | if (err) { |
1822 | ret = err; | |
c8c42864 | 1823 | goto done; |
33c66f43 | 1824 | } |
c8c42864 CM |
1825 | b = p->nodes[level]; |
1826 | slot = p->slots[level]; | |
b4ce94de | 1827 | |
bd681513 CM |
1828 | /* |
1829 | * slot 0 is special, if we change the key | |
1830 | * we have to update the parent pointer | |
1831 | * which means we must have a write lock | |
1832 | * on the parent | |
1833 | */ | |
1834 | if (slot == 0 && cow && | |
1835 | write_lock_level < level + 1) { | |
1836 | write_lock_level = level + 1; | |
1837 | btrfs_release_path(p); | |
1838 | goto again; | |
1839 | } | |
1840 | ||
f7c79f30 CM |
1841 | unlock_up(p, level, lowest_unlock, |
1842 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 1843 | |
925baedd | 1844 | if (level == lowest_level) { |
33c66f43 YZ |
1845 | if (dec) |
1846 | p->slots[level]++; | |
5b21f2ed | 1847 | goto done; |
925baedd | 1848 | } |
ca7a79ad | 1849 | |
33c66f43 | 1850 | err = read_block_for_search(trans, root, p, |
c8c42864 | 1851 | &b, level, slot, key); |
33c66f43 | 1852 | if (err == -EAGAIN) |
c8c42864 | 1853 | goto again; |
33c66f43 YZ |
1854 | if (err) { |
1855 | ret = err; | |
76a05b35 | 1856 | goto done; |
33c66f43 | 1857 | } |
76a05b35 | 1858 | |
b4ce94de | 1859 | if (!p->skip_locking) { |
bd681513 CM |
1860 | level = btrfs_header_level(b); |
1861 | if (level <= write_lock_level) { | |
1862 | err = btrfs_try_tree_write_lock(b); | |
1863 | if (!err) { | |
1864 | btrfs_set_path_blocking(p); | |
1865 | btrfs_tree_lock(b); | |
1866 | btrfs_clear_path_blocking(p, b, | |
1867 | BTRFS_WRITE_LOCK); | |
1868 | } | |
1869 | p->locks[level] = BTRFS_WRITE_LOCK; | |
1870 | } else { | |
1871 | err = btrfs_try_tree_read_lock(b); | |
1872 | if (!err) { | |
1873 | btrfs_set_path_blocking(p); | |
1874 | btrfs_tree_read_lock(b); | |
1875 | btrfs_clear_path_blocking(p, b, | |
1876 | BTRFS_READ_LOCK); | |
1877 | } | |
1878 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 1879 | } |
bd681513 | 1880 | p->nodes[level] = b; |
b4ce94de | 1881 | } |
be0e5c09 CM |
1882 | } else { |
1883 | p->slots[level] = slot; | |
87b29b20 YZ |
1884 | if (ins_len > 0 && |
1885 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
1886 | if (write_lock_level < 1) { |
1887 | write_lock_level = 1; | |
1888 | btrfs_release_path(p); | |
1889 | goto again; | |
1890 | } | |
1891 | ||
b4ce94de | 1892 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
1893 | err = split_leaf(trans, root, key, |
1894 | p, ins_len, ret == 0); | |
bd681513 | 1895 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 1896 | |
33c66f43 YZ |
1897 | BUG_ON(err > 0); |
1898 | if (err) { | |
1899 | ret = err; | |
65b51a00 CM |
1900 | goto done; |
1901 | } | |
5c680ed6 | 1902 | } |
459931ec | 1903 | if (!p->search_for_split) |
f7c79f30 CM |
1904 | unlock_up(p, level, lowest_unlock, |
1905 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 1906 | goto done; |
be0e5c09 CM |
1907 | } |
1908 | } | |
65b51a00 CM |
1909 | ret = 1; |
1910 | done: | |
b4ce94de CM |
1911 | /* |
1912 | * we don't really know what they plan on doing with the path | |
1913 | * from here on, so for now just mark it as blocking | |
1914 | */ | |
b9473439 CM |
1915 | if (!p->leave_spinning) |
1916 | btrfs_set_path_blocking(p); | |
76a05b35 | 1917 | if (ret < 0) |
b3b4aa74 | 1918 | btrfs_release_path(p); |
65b51a00 | 1919 | return ret; |
be0e5c09 CM |
1920 | } |
1921 | ||
74123bd7 CM |
1922 | /* |
1923 | * adjust the pointers going up the tree, starting at level | |
1924 | * making sure the right key of each node is points to 'key'. | |
1925 | * This is used after shifting pointers to the left, so it stops | |
1926 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
1927 | * higher levels | |
aa5d6bed | 1928 | * |
74123bd7 | 1929 | */ |
143bede5 JM |
1930 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
1931 | struct btrfs_root *root, struct btrfs_path *path, | |
1932 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
1933 | { |
1934 | int i; | |
5f39d397 CM |
1935 | struct extent_buffer *t; |
1936 | ||
234b63a0 | 1937 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 1938 | int tslot = path->slots[i]; |
eb60ceac | 1939 | if (!path->nodes[i]) |
be0e5c09 | 1940 | break; |
5f39d397 CM |
1941 | t = path->nodes[i]; |
1942 | btrfs_set_node_key(t, key, tslot); | |
d6025579 | 1943 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
1944 | if (tslot != 0) |
1945 | break; | |
1946 | } | |
1947 | } | |
1948 | ||
31840ae1 ZY |
1949 | /* |
1950 | * update item key. | |
1951 | * | |
1952 | * This function isn't completely safe. It's the caller's responsibility | |
1953 | * that the new key won't break the order | |
1954 | */ | |
143bede5 JM |
1955 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
1956 | struct btrfs_root *root, struct btrfs_path *path, | |
1957 | struct btrfs_key *new_key) | |
31840ae1 ZY |
1958 | { |
1959 | struct btrfs_disk_key disk_key; | |
1960 | struct extent_buffer *eb; | |
1961 | int slot; | |
1962 | ||
1963 | eb = path->nodes[0]; | |
1964 | slot = path->slots[0]; | |
1965 | if (slot > 0) { | |
1966 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 1967 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
1968 | } |
1969 | if (slot < btrfs_header_nritems(eb) - 1) { | |
1970 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 1971 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
1972 | } |
1973 | ||
1974 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
1975 | btrfs_set_item_key(eb, &disk_key, slot); | |
1976 | btrfs_mark_buffer_dirty(eb); | |
1977 | if (slot == 0) | |
1978 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
31840ae1 ZY |
1979 | } |
1980 | ||
74123bd7 CM |
1981 | /* |
1982 | * try to push data from one node into the next node left in the | |
79f95c82 | 1983 | * tree. |
aa5d6bed CM |
1984 | * |
1985 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
1986 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 1987 | */ |
98ed5174 CM |
1988 | static int push_node_left(struct btrfs_trans_handle *trans, |
1989 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 1990 | struct extent_buffer *src, int empty) |
be0e5c09 | 1991 | { |
be0e5c09 | 1992 | int push_items = 0; |
bb803951 CM |
1993 | int src_nritems; |
1994 | int dst_nritems; | |
aa5d6bed | 1995 | int ret = 0; |
be0e5c09 | 1996 | |
5f39d397 CM |
1997 | src_nritems = btrfs_header_nritems(src); |
1998 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 1999 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
2000 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2001 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2002 | |
bce4eae9 | 2003 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2004 | return 1; |
2005 | ||
d397712b | 2006 | if (push_items <= 0) |
be0e5c09 CM |
2007 | return 1; |
2008 | ||
bce4eae9 | 2009 | if (empty) { |
971a1f66 | 2010 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2011 | if (push_items < src_nritems) { |
2012 | /* leave at least 8 pointers in the node if | |
2013 | * we aren't going to empty it | |
2014 | */ | |
2015 | if (src_nritems - push_items < 8) { | |
2016 | if (push_items <= 8) | |
2017 | return 1; | |
2018 | push_items -= 8; | |
2019 | } | |
2020 | } | |
2021 | } else | |
2022 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2023 | |
5f39d397 CM |
2024 | copy_extent_buffer(dst, src, |
2025 | btrfs_node_key_ptr_offset(dst_nritems), | |
2026 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2027 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2028 | |
bb803951 | 2029 | if (push_items < src_nritems) { |
5f39d397 CM |
2030 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2031 | btrfs_node_key_ptr_offset(push_items), | |
2032 | (src_nritems - push_items) * | |
2033 | sizeof(struct btrfs_key_ptr)); | |
2034 | } | |
2035 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2036 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2037 | btrfs_mark_buffer_dirty(src); | |
2038 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2039 | |
79f95c82 CM |
2040 | return ret; |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * try to push data from one node into the next node right in the | |
2045 | * tree. | |
2046 | * | |
2047 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2048 | * error, and > 0 if there was no room in the right hand block. | |
2049 | * | |
2050 | * this will only push up to 1/2 the contents of the left node over | |
2051 | */ | |
5f39d397 CM |
2052 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2053 | struct btrfs_root *root, | |
2054 | struct extent_buffer *dst, | |
2055 | struct extent_buffer *src) | |
79f95c82 | 2056 | { |
79f95c82 CM |
2057 | int push_items = 0; |
2058 | int max_push; | |
2059 | int src_nritems; | |
2060 | int dst_nritems; | |
2061 | int ret = 0; | |
79f95c82 | 2062 | |
7bb86316 CM |
2063 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2064 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2065 | ||
5f39d397 CM |
2066 | src_nritems = btrfs_header_nritems(src); |
2067 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2068 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 2069 | if (push_items <= 0) |
79f95c82 | 2070 | return 1; |
bce4eae9 | 2071 | |
d397712b | 2072 | if (src_nritems < 4) |
bce4eae9 | 2073 | return 1; |
79f95c82 CM |
2074 | |
2075 | max_push = src_nritems / 2 + 1; | |
2076 | /* don't try to empty the node */ | |
d397712b | 2077 | if (max_push >= src_nritems) |
79f95c82 | 2078 | return 1; |
252c38f0 | 2079 | |
79f95c82 CM |
2080 | if (max_push < push_items) |
2081 | push_items = max_push; | |
2082 | ||
5f39d397 CM |
2083 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2084 | btrfs_node_key_ptr_offset(0), | |
2085 | (dst_nritems) * | |
2086 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2087 | |
5f39d397 CM |
2088 | copy_extent_buffer(dst, src, |
2089 | btrfs_node_key_ptr_offset(0), | |
2090 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2091 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2092 | |
5f39d397 CM |
2093 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2094 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2095 | |
5f39d397 CM |
2096 | btrfs_mark_buffer_dirty(src); |
2097 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2098 | |
aa5d6bed | 2099 | return ret; |
be0e5c09 CM |
2100 | } |
2101 | ||
97571fd0 CM |
2102 | /* |
2103 | * helper function to insert a new root level in the tree. | |
2104 | * A new node is allocated, and a single item is inserted to | |
2105 | * point to the existing root | |
aa5d6bed CM |
2106 | * |
2107 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2108 | */ |
d397712b | 2109 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2110 | struct btrfs_root *root, |
2111 | struct btrfs_path *path, int level) | |
5c680ed6 | 2112 | { |
7bb86316 | 2113 | u64 lower_gen; |
5f39d397 CM |
2114 | struct extent_buffer *lower; |
2115 | struct extent_buffer *c; | |
925baedd | 2116 | struct extent_buffer *old; |
5f39d397 | 2117 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
2118 | |
2119 | BUG_ON(path->nodes[level]); | |
2120 | BUG_ON(path->nodes[level-1] != root->node); | |
2121 | ||
7bb86316 CM |
2122 | lower = path->nodes[level-1]; |
2123 | if (level == 1) | |
2124 | btrfs_item_key(lower, &lower_key, 0); | |
2125 | else | |
2126 | btrfs_node_key(lower, &lower_key, 0); | |
2127 | ||
31840ae1 | 2128 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 2129 | root->root_key.objectid, &lower_key, |
66d7e7f0 | 2130 | level, root->node->start, 0, 0); |
5f39d397 CM |
2131 | if (IS_ERR(c)) |
2132 | return PTR_ERR(c); | |
925baedd | 2133 | |
f0486c68 YZ |
2134 | root_add_used(root, root->nodesize); |
2135 | ||
5d4f98a2 | 2136 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
2137 | btrfs_set_header_nritems(c, 1); |
2138 | btrfs_set_header_level(c, level); | |
db94535d | 2139 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 2140 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 2141 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 2142 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
2143 | |
2144 | write_extent_buffer(c, root->fs_info->fsid, | |
2145 | (unsigned long)btrfs_header_fsid(c), | |
2146 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2147 | |
2148 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
2149 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
2150 | BTRFS_UUID_SIZE); | |
2151 | ||
5f39d397 | 2152 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2153 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2154 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2155 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2156 | |
2157 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2158 | |
5f39d397 | 2159 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2160 | |
925baedd | 2161 | old = root->node; |
240f62c8 | 2162 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2163 | |
2164 | /* the super has an extra ref to root->node */ | |
2165 | free_extent_buffer(old); | |
2166 | ||
0b86a832 | 2167 | add_root_to_dirty_list(root); |
5f39d397 CM |
2168 | extent_buffer_get(c); |
2169 | path->nodes[level] = c; | |
bd681513 | 2170 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2171 | path->slots[level] = 0; |
2172 | return 0; | |
2173 | } | |
2174 | ||
74123bd7 CM |
2175 | /* |
2176 | * worker function to insert a single pointer in a node. | |
2177 | * the node should have enough room for the pointer already | |
97571fd0 | 2178 | * |
74123bd7 CM |
2179 | * slot and level indicate where you want the key to go, and |
2180 | * blocknr is the block the key points to. | |
2181 | */ | |
143bede5 JM |
2182 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2183 | struct btrfs_root *root, struct btrfs_path *path, | |
2184 | struct btrfs_disk_key *key, u64 bytenr, | |
2185 | int slot, int level) | |
74123bd7 | 2186 | { |
5f39d397 | 2187 | struct extent_buffer *lower; |
74123bd7 | 2188 | int nritems; |
5c680ed6 CM |
2189 | |
2190 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 2191 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
2192 | lower = path->nodes[level]; |
2193 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2194 | BUG_ON(slot > nritems); |
143bede5 | 2195 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 2196 | if (slot != nritems) { |
5f39d397 CM |
2197 | memmove_extent_buffer(lower, |
2198 | btrfs_node_key_ptr_offset(slot + 1), | |
2199 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2200 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2201 | } |
5f39d397 | 2202 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2203 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2204 | WARN_ON(trans->transid == 0); |
2205 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2206 | btrfs_set_header_nritems(lower, nritems + 1); |
2207 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2208 | } |
2209 | ||
97571fd0 CM |
2210 | /* |
2211 | * split the node at the specified level in path in two. | |
2212 | * The path is corrected to point to the appropriate node after the split | |
2213 | * | |
2214 | * Before splitting this tries to make some room in the node by pushing | |
2215 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2216 | * |
2217 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2218 | */ |
e02119d5 CM |
2219 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2220 | struct btrfs_root *root, | |
2221 | struct btrfs_path *path, int level) | |
be0e5c09 | 2222 | { |
5f39d397 CM |
2223 | struct extent_buffer *c; |
2224 | struct extent_buffer *split; | |
2225 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2226 | int mid; |
5c680ed6 | 2227 | int ret; |
7518a238 | 2228 | u32 c_nritems; |
eb60ceac | 2229 | |
5f39d397 | 2230 | c = path->nodes[level]; |
7bb86316 | 2231 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2232 | if (c == root->node) { |
5c680ed6 | 2233 | /* trying to split the root, lets make a new one */ |
e089f05c | 2234 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2235 | if (ret) |
2236 | return ret; | |
b3612421 | 2237 | } else { |
e66f709b | 2238 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2239 | c = path->nodes[level]; |
2240 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2241 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2242 | return 0; |
54aa1f4d CM |
2243 | if (ret < 0) |
2244 | return ret; | |
be0e5c09 | 2245 | } |
e66f709b | 2246 | |
5f39d397 | 2247 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2248 | mid = (c_nritems + 1) / 2; |
2249 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2250 | |
5d4f98a2 | 2251 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 2252 | root->root_key.objectid, |
66d7e7f0 | 2253 | &disk_key, level, c->start, 0, 0); |
5f39d397 CM |
2254 | if (IS_ERR(split)) |
2255 | return PTR_ERR(split); | |
2256 | ||
f0486c68 YZ |
2257 | root_add_used(root, root->nodesize); |
2258 | ||
5d4f98a2 | 2259 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 2260 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 2261 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 2262 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 2263 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2264 | btrfs_set_header_owner(split, root->root_key.objectid); |
2265 | write_extent_buffer(split, root->fs_info->fsid, | |
2266 | (unsigned long)btrfs_header_fsid(split), | |
2267 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2268 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2269 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2270 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2271 | |
5f39d397 CM |
2272 | |
2273 | copy_extent_buffer(split, c, | |
2274 | btrfs_node_key_ptr_offset(0), | |
2275 | btrfs_node_key_ptr_offset(mid), | |
2276 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2277 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2278 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2279 | ret = 0; |
2280 | ||
5f39d397 CM |
2281 | btrfs_mark_buffer_dirty(c); |
2282 | btrfs_mark_buffer_dirty(split); | |
2283 | ||
143bede5 JM |
2284 | insert_ptr(trans, root, path, &disk_key, split->start, |
2285 | path->slots[level + 1] + 1, level + 1); | |
aa5d6bed | 2286 | |
5de08d7d | 2287 | if (path->slots[level] >= mid) { |
5c680ed6 | 2288 | path->slots[level] -= mid; |
925baedd | 2289 | btrfs_tree_unlock(c); |
5f39d397 CM |
2290 | free_extent_buffer(c); |
2291 | path->nodes[level] = split; | |
5c680ed6 CM |
2292 | path->slots[level + 1] += 1; |
2293 | } else { | |
925baedd | 2294 | btrfs_tree_unlock(split); |
5f39d397 | 2295 | free_extent_buffer(split); |
be0e5c09 | 2296 | } |
aa5d6bed | 2297 | return ret; |
be0e5c09 CM |
2298 | } |
2299 | ||
74123bd7 CM |
2300 | /* |
2301 | * how many bytes are required to store the items in a leaf. start | |
2302 | * and nr indicate which items in the leaf to check. This totals up the | |
2303 | * space used both by the item structs and the item data | |
2304 | */ | |
5f39d397 | 2305 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2306 | { |
2307 | int data_len; | |
5f39d397 | 2308 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2309 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2310 | |
2311 | if (!nr) | |
2312 | return 0; | |
5f39d397 CM |
2313 | data_len = btrfs_item_end_nr(l, start); |
2314 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2315 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2316 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2317 | return data_len; |
2318 | } | |
2319 | ||
d4dbff95 CM |
2320 | /* |
2321 | * The space between the end of the leaf items and | |
2322 | * the start of the leaf data. IOW, how much room | |
2323 | * the leaf has left for both items and data | |
2324 | */ | |
d397712b | 2325 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 2326 | struct extent_buffer *leaf) |
d4dbff95 | 2327 | { |
5f39d397 CM |
2328 | int nritems = btrfs_header_nritems(leaf); |
2329 | int ret; | |
2330 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2331 | if (ret < 0) { | |
d397712b CM |
2332 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
2333 | "used %d nritems %d\n", | |
ae2f5411 | 2334 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2335 | leaf_space_used(leaf, 0, nritems), nritems); |
2336 | } | |
2337 | return ret; | |
d4dbff95 CM |
2338 | } |
2339 | ||
99d8f83c CM |
2340 | /* |
2341 | * min slot controls the lowest index we're willing to push to the | |
2342 | * right. We'll push up to and including min_slot, but no lower | |
2343 | */ | |
44871b1b CM |
2344 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
2345 | struct btrfs_root *root, | |
2346 | struct btrfs_path *path, | |
2347 | int data_size, int empty, | |
2348 | struct extent_buffer *right, | |
99d8f83c CM |
2349 | int free_space, u32 left_nritems, |
2350 | u32 min_slot) | |
00ec4c51 | 2351 | { |
5f39d397 | 2352 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2353 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 2354 | struct btrfs_map_token token; |
5f39d397 | 2355 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2356 | int slot; |
34a38218 | 2357 | u32 i; |
00ec4c51 CM |
2358 | int push_space = 0; |
2359 | int push_items = 0; | |
0783fcfc | 2360 | struct btrfs_item *item; |
34a38218 | 2361 | u32 nr; |
7518a238 | 2362 | u32 right_nritems; |
5f39d397 | 2363 | u32 data_end; |
db94535d | 2364 | u32 this_item_size; |
00ec4c51 | 2365 | |
cfed81a0 CM |
2366 | btrfs_init_map_token(&token); |
2367 | ||
34a38218 CM |
2368 | if (empty) |
2369 | nr = 0; | |
2370 | else | |
99d8f83c | 2371 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2372 | |
31840ae1 | 2373 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2374 | push_space += data_size; |
31840ae1 | 2375 | |
44871b1b | 2376 | slot = path->slots[1]; |
34a38218 CM |
2377 | i = left_nritems - 1; |
2378 | while (i >= nr) { | |
5f39d397 | 2379 | item = btrfs_item_nr(left, i); |
db94535d | 2380 | |
31840ae1 ZY |
2381 | if (!empty && push_items > 0) { |
2382 | if (path->slots[0] > i) | |
2383 | break; | |
2384 | if (path->slots[0] == i) { | |
2385 | int space = btrfs_leaf_free_space(root, left); | |
2386 | if (space + push_space * 2 > free_space) | |
2387 | break; | |
2388 | } | |
2389 | } | |
2390 | ||
00ec4c51 | 2391 | if (path->slots[0] == i) |
87b29b20 | 2392 | push_space += data_size; |
db94535d | 2393 | |
db94535d CM |
2394 | this_item_size = btrfs_item_size(left, item); |
2395 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2396 | break; |
31840ae1 | 2397 | |
00ec4c51 | 2398 | push_items++; |
db94535d | 2399 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2400 | if (i == 0) |
2401 | break; | |
2402 | i--; | |
db94535d | 2403 | } |
5f39d397 | 2404 | |
925baedd CM |
2405 | if (push_items == 0) |
2406 | goto out_unlock; | |
5f39d397 | 2407 | |
34a38218 | 2408 | if (!empty && push_items == left_nritems) |
a429e513 | 2409 | WARN_ON(1); |
5f39d397 | 2410 | |
00ec4c51 | 2411 | /* push left to right */ |
5f39d397 | 2412 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2413 | |
5f39d397 | 2414 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2415 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2416 | |
00ec4c51 | 2417 | /* make room in the right data area */ |
5f39d397 CM |
2418 | data_end = leaf_data_end(root, right); |
2419 | memmove_extent_buffer(right, | |
2420 | btrfs_leaf_data(right) + data_end - push_space, | |
2421 | btrfs_leaf_data(right) + data_end, | |
2422 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2423 | ||
00ec4c51 | 2424 | /* copy from the left data area */ |
5f39d397 | 2425 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2426 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2427 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2428 | push_space); | |
5f39d397 CM |
2429 | |
2430 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2431 | btrfs_item_nr_offset(0), | |
2432 | right_nritems * sizeof(struct btrfs_item)); | |
2433 | ||
00ec4c51 | 2434 | /* copy the items from left to right */ |
5f39d397 CM |
2435 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2436 | btrfs_item_nr_offset(left_nritems - push_items), | |
2437 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2438 | |
2439 | /* update the item pointers */ | |
7518a238 | 2440 | right_nritems += push_items; |
5f39d397 | 2441 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2442 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2443 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2444 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
2445 | push_space -= btrfs_token_item_size(right, item, &token); |
2446 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
2447 | } |
2448 | ||
7518a238 | 2449 | left_nritems -= push_items; |
5f39d397 | 2450 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2451 | |
34a38218 CM |
2452 | if (left_nritems) |
2453 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
2454 | else |
2455 | clean_tree_block(trans, root, left); | |
2456 | ||
5f39d397 | 2457 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2458 | |
5f39d397 CM |
2459 | btrfs_item_key(right, &disk_key, 0); |
2460 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2461 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2462 | |
00ec4c51 | 2463 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2464 | if (path->slots[0] >= left_nritems) { |
2465 | path->slots[0] -= left_nritems; | |
925baedd CM |
2466 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2467 | clean_tree_block(trans, root, path->nodes[0]); | |
2468 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2469 | free_extent_buffer(path->nodes[0]); |
2470 | path->nodes[0] = right; | |
00ec4c51 CM |
2471 | path->slots[1] += 1; |
2472 | } else { | |
925baedd | 2473 | btrfs_tree_unlock(right); |
5f39d397 | 2474 | free_extent_buffer(right); |
00ec4c51 CM |
2475 | } |
2476 | return 0; | |
925baedd CM |
2477 | |
2478 | out_unlock: | |
2479 | btrfs_tree_unlock(right); | |
2480 | free_extent_buffer(right); | |
2481 | return 1; | |
00ec4c51 | 2482 | } |
925baedd | 2483 | |
44871b1b CM |
2484 | /* |
2485 | * push some data in the path leaf to the right, trying to free up at | |
2486 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2487 | * | |
2488 | * returns 1 if the push failed because the other node didn't have enough | |
2489 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
2490 | * |
2491 | * this will push starting from min_slot to the end of the leaf. It won't | |
2492 | * push any slot lower than min_slot | |
44871b1b CM |
2493 | */ |
2494 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2495 | *root, struct btrfs_path *path, |
2496 | int min_data_size, int data_size, | |
2497 | int empty, u32 min_slot) | |
44871b1b CM |
2498 | { |
2499 | struct extent_buffer *left = path->nodes[0]; | |
2500 | struct extent_buffer *right; | |
2501 | struct extent_buffer *upper; | |
2502 | int slot; | |
2503 | int free_space; | |
2504 | u32 left_nritems; | |
2505 | int ret; | |
2506 | ||
2507 | if (!path->nodes[1]) | |
2508 | return 1; | |
2509 | ||
2510 | slot = path->slots[1]; | |
2511 | upper = path->nodes[1]; | |
2512 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2513 | return 1; | |
2514 | ||
2515 | btrfs_assert_tree_locked(path->nodes[1]); | |
2516 | ||
2517 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
2518 | if (right == NULL) |
2519 | return 1; | |
2520 | ||
44871b1b CM |
2521 | btrfs_tree_lock(right); |
2522 | btrfs_set_lock_blocking(right); | |
2523 | ||
2524 | free_space = btrfs_leaf_free_space(root, right); | |
2525 | if (free_space < data_size) | |
2526 | goto out_unlock; | |
2527 | ||
2528 | /* cow and double check */ | |
2529 | ret = btrfs_cow_block(trans, root, right, upper, | |
2530 | slot + 1, &right); | |
2531 | if (ret) | |
2532 | goto out_unlock; | |
2533 | ||
2534 | free_space = btrfs_leaf_free_space(root, right); | |
2535 | if (free_space < data_size) | |
2536 | goto out_unlock; | |
2537 | ||
2538 | left_nritems = btrfs_header_nritems(left); | |
2539 | if (left_nritems == 0) | |
2540 | goto out_unlock; | |
2541 | ||
99d8f83c CM |
2542 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
2543 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
2544 | out_unlock: |
2545 | btrfs_tree_unlock(right); | |
2546 | free_extent_buffer(right); | |
2547 | return 1; | |
2548 | } | |
2549 | ||
74123bd7 CM |
2550 | /* |
2551 | * push some data in the path leaf to the left, trying to free up at | |
2552 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2553 | * |
2554 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2555 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
2556 | * items | |
74123bd7 | 2557 | */ |
44871b1b CM |
2558 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
2559 | struct btrfs_root *root, | |
2560 | struct btrfs_path *path, int data_size, | |
2561 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
2562 | int free_space, u32 right_nritems, |
2563 | u32 max_slot) | |
be0e5c09 | 2564 | { |
5f39d397 CM |
2565 | struct btrfs_disk_key disk_key; |
2566 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 2567 | int i; |
be0e5c09 CM |
2568 | int push_space = 0; |
2569 | int push_items = 0; | |
0783fcfc | 2570 | struct btrfs_item *item; |
7518a238 | 2571 | u32 old_left_nritems; |
34a38218 | 2572 | u32 nr; |
aa5d6bed | 2573 | int ret = 0; |
db94535d CM |
2574 | u32 this_item_size; |
2575 | u32 old_left_item_size; | |
cfed81a0 CM |
2576 | struct btrfs_map_token token; |
2577 | ||
2578 | btrfs_init_map_token(&token); | |
be0e5c09 | 2579 | |
34a38218 | 2580 | if (empty) |
99d8f83c | 2581 | nr = min(right_nritems, max_slot); |
34a38218 | 2582 | else |
99d8f83c | 2583 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
2584 | |
2585 | for (i = 0; i < nr; i++) { | |
5f39d397 | 2586 | item = btrfs_item_nr(right, i); |
db94535d | 2587 | |
31840ae1 ZY |
2588 | if (!empty && push_items > 0) { |
2589 | if (path->slots[0] < i) | |
2590 | break; | |
2591 | if (path->slots[0] == i) { | |
2592 | int space = btrfs_leaf_free_space(root, right); | |
2593 | if (space + push_space * 2 > free_space) | |
2594 | break; | |
2595 | } | |
2596 | } | |
2597 | ||
be0e5c09 | 2598 | if (path->slots[0] == i) |
87b29b20 | 2599 | push_space += data_size; |
db94535d CM |
2600 | |
2601 | this_item_size = btrfs_item_size(right, item); | |
2602 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 2603 | break; |
db94535d | 2604 | |
be0e5c09 | 2605 | push_items++; |
db94535d CM |
2606 | push_space += this_item_size + sizeof(*item); |
2607 | } | |
2608 | ||
be0e5c09 | 2609 | if (push_items == 0) { |
925baedd CM |
2610 | ret = 1; |
2611 | goto out; | |
be0e5c09 | 2612 | } |
34a38218 | 2613 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 2614 | WARN_ON(1); |
5f39d397 | 2615 | |
be0e5c09 | 2616 | /* push data from right to left */ |
5f39d397 CM |
2617 | copy_extent_buffer(left, right, |
2618 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
2619 | btrfs_item_nr_offset(0), | |
2620 | push_items * sizeof(struct btrfs_item)); | |
2621 | ||
123abc88 | 2622 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 2623 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
2624 | |
2625 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
2626 | leaf_data_end(root, left) - push_space, |
2627 | btrfs_leaf_data(right) + | |
5f39d397 | 2628 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 2629 | push_space); |
5f39d397 | 2630 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 2631 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 2632 | |
db94535d | 2633 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 2634 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 2635 | u32 ioff; |
db94535d | 2636 | |
5f39d397 | 2637 | item = btrfs_item_nr(left, i); |
db94535d | 2638 | |
cfed81a0 CM |
2639 | ioff = btrfs_token_item_offset(left, item, &token); |
2640 | btrfs_set_token_item_offset(left, item, | |
2641 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
2642 | &token); | |
be0e5c09 | 2643 | } |
5f39d397 | 2644 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
2645 | |
2646 | /* fixup right node */ | |
34a38218 | 2647 | if (push_items > right_nritems) { |
d397712b CM |
2648 | printk(KERN_CRIT "push items %d nr %u\n", push_items, |
2649 | right_nritems); | |
34a38218 CM |
2650 | WARN_ON(1); |
2651 | } | |
2652 | ||
2653 | if (push_items < right_nritems) { | |
2654 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2655 | leaf_data_end(root, right); | |
2656 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
2657 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
2658 | btrfs_leaf_data(right) + | |
2659 | leaf_data_end(root, right), push_space); | |
2660 | ||
2661 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
2662 | btrfs_item_nr_offset(push_items), |
2663 | (btrfs_header_nritems(right) - push_items) * | |
2664 | sizeof(struct btrfs_item)); | |
34a38218 | 2665 | } |
eef1c494 Y |
2666 | right_nritems -= push_items; |
2667 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 2668 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
2669 | for (i = 0; i < right_nritems; i++) { |
2670 | item = btrfs_item_nr(right, i); | |
db94535d | 2671 | |
cfed81a0 CM |
2672 | push_space = push_space - btrfs_token_item_size(right, |
2673 | item, &token); | |
2674 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 2675 | } |
eb60ceac | 2676 | |
5f39d397 | 2677 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
2678 | if (right_nritems) |
2679 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
2680 | else |
2681 | clean_tree_block(trans, root, right); | |
098f59c2 | 2682 | |
5f39d397 | 2683 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 2684 | fixup_low_keys(trans, root, path, &disk_key, 1); |
be0e5c09 CM |
2685 | |
2686 | /* then fixup the leaf pointer in the path */ | |
2687 | if (path->slots[0] < push_items) { | |
2688 | path->slots[0] += old_left_nritems; | |
925baedd | 2689 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2690 | free_extent_buffer(path->nodes[0]); |
2691 | path->nodes[0] = left; | |
be0e5c09 CM |
2692 | path->slots[1] -= 1; |
2693 | } else { | |
925baedd | 2694 | btrfs_tree_unlock(left); |
5f39d397 | 2695 | free_extent_buffer(left); |
be0e5c09 CM |
2696 | path->slots[0] -= push_items; |
2697 | } | |
eb60ceac | 2698 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 2699 | return ret; |
925baedd CM |
2700 | out: |
2701 | btrfs_tree_unlock(left); | |
2702 | free_extent_buffer(left); | |
2703 | return ret; | |
be0e5c09 CM |
2704 | } |
2705 | ||
44871b1b CM |
2706 | /* |
2707 | * push some data in the path leaf to the left, trying to free up at | |
2708 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2709 | * |
2710 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2711 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
2712 | * items | |
44871b1b CM |
2713 | */ |
2714 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2715 | *root, struct btrfs_path *path, int min_data_size, |
2716 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
2717 | { |
2718 | struct extent_buffer *right = path->nodes[0]; | |
2719 | struct extent_buffer *left; | |
2720 | int slot; | |
2721 | int free_space; | |
2722 | u32 right_nritems; | |
2723 | int ret = 0; | |
2724 | ||
2725 | slot = path->slots[1]; | |
2726 | if (slot == 0) | |
2727 | return 1; | |
2728 | if (!path->nodes[1]) | |
2729 | return 1; | |
2730 | ||
2731 | right_nritems = btrfs_header_nritems(right); | |
2732 | if (right_nritems == 0) | |
2733 | return 1; | |
2734 | ||
2735 | btrfs_assert_tree_locked(path->nodes[1]); | |
2736 | ||
2737 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
2738 | if (left == NULL) |
2739 | return 1; | |
2740 | ||
44871b1b CM |
2741 | btrfs_tree_lock(left); |
2742 | btrfs_set_lock_blocking(left); | |
2743 | ||
2744 | free_space = btrfs_leaf_free_space(root, left); | |
2745 | if (free_space < data_size) { | |
2746 | ret = 1; | |
2747 | goto out; | |
2748 | } | |
2749 | ||
2750 | /* cow and double check */ | |
2751 | ret = btrfs_cow_block(trans, root, left, | |
2752 | path->nodes[1], slot - 1, &left); | |
2753 | if (ret) { | |
2754 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
2755 | if (ret == -ENOSPC) |
2756 | ret = 1; | |
44871b1b CM |
2757 | goto out; |
2758 | } | |
2759 | ||
2760 | free_space = btrfs_leaf_free_space(root, left); | |
2761 | if (free_space < data_size) { | |
2762 | ret = 1; | |
2763 | goto out; | |
2764 | } | |
2765 | ||
99d8f83c CM |
2766 | return __push_leaf_left(trans, root, path, min_data_size, |
2767 | empty, left, free_space, right_nritems, | |
2768 | max_slot); | |
44871b1b CM |
2769 | out: |
2770 | btrfs_tree_unlock(left); | |
2771 | free_extent_buffer(left); | |
2772 | return ret; | |
2773 | } | |
2774 | ||
2775 | /* | |
2776 | * split the path's leaf in two, making sure there is at least data_size | |
2777 | * available for the resulting leaf level of the path. | |
44871b1b | 2778 | */ |
143bede5 JM |
2779 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2780 | struct btrfs_root *root, | |
2781 | struct btrfs_path *path, | |
2782 | struct extent_buffer *l, | |
2783 | struct extent_buffer *right, | |
2784 | int slot, int mid, int nritems) | |
44871b1b CM |
2785 | { |
2786 | int data_copy_size; | |
2787 | int rt_data_off; | |
2788 | int i; | |
44871b1b | 2789 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
2790 | struct btrfs_map_token token; |
2791 | ||
2792 | btrfs_init_map_token(&token); | |
44871b1b CM |
2793 | |
2794 | nritems = nritems - mid; | |
2795 | btrfs_set_header_nritems(right, nritems); | |
2796 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
2797 | ||
2798 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
2799 | btrfs_item_nr_offset(mid), | |
2800 | nritems * sizeof(struct btrfs_item)); | |
2801 | ||
2802 | copy_extent_buffer(right, l, | |
2803 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
2804 | data_copy_size, btrfs_leaf_data(l) + | |
2805 | leaf_data_end(root, l), data_copy_size); | |
2806 | ||
2807 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
2808 | btrfs_item_end_nr(l, mid); | |
2809 | ||
2810 | for (i = 0; i < nritems; i++) { | |
2811 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
2812 | u32 ioff; | |
2813 | ||
cfed81a0 CM |
2814 | ioff = btrfs_token_item_offset(right, item, &token); |
2815 | btrfs_set_token_item_offset(right, item, | |
2816 | ioff + rt_data_off, &token); | |
44871b1b CM |
2817 | } |
2818 | ||
44871b1b | 2819 | btrfs_set_header_nritems(l, mid); |
44871b1b | 2820 | btrfs_item_key(right, &disk_key, 0); |
143bede5 JM |
2821 | insert_ptr(trans, root, path, &disk_key, right->start, |
2822 | path->slots[1] + 1, 1); | |
44871b1b CM |
2823 | |
2824 | btrfs_mark_buffer_dirty(right); | |
2825 | btrfs_mark_buffer_dirty(l); | |
2826 | BUG_ON(path->slots[0] != slot); | |
2827 | ||
44871b1b CM |
2828 | if (mid <= slot) { |
2829 | btrfs_tree_unlock(path->nodes[0]); | |
2830 | free_extent_buffer(path->nodes[0]); | |
2831 | path->nodes[0] = right; | |
2832 | path->slots[0] -= mid; | |
2833 | path->slots[1] += 1; | |
2834 | } else { | |
2835 | btrfs_tree_unlock(right); | |
2836 | free_extent_buffer(right); | |
2837 | } | |
2838 | ||
2839 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
2840 | } |
2841 | ||
99d8f83c CM |
2842 | /* |
2843 | * double splits happen when we need to insert a big item in the middle | |
2844 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
2845 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
2846 | * A B C | |
2847 | * | |
2848 | * We avoid this by trying to push the items on either side of our target | |
2849 | * into the adjacent leaves. If all goes well we can avoid the double split | |
2850 | * completely. | |
2851 | */ | |
2852 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
2853 | struct btrfs_root *root, | |
2854 | struct btrfs_path *path, | |
2855 | int data_size) | |
2856 | { | |
2857 | int ret; | |
2858 | int progress = 0; | |
2859 | int slot; | |
2860 | u32 nritems; | |
2861 | ||
2862 | slot = path->slots[0]; | |
2863 | ||
2864 | /* | |
2865 | * try to push all the items after our slot into the | |
2866 | * right leaf | |
2867 | */ | |
2868 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
2869 | if (ret < 0) | |
2870 | return ret; | |
2871 | ||
2872 | if (ret == 0) | |
2873 | progress++; | |
2874 | ||
2875 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2876 | /* | |
2877 | * our goal is to get our slot at the start or end of a leaf. If | |
2878 | * we've done so we're done | |
2879 | */ | |
2880 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
2881 | return 0; | |
2882 | ||
2883 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
2884 | return 0; | |
2885 | ||
2886 | /* try to push all the items before our slot into the next leaf */ | |
2887 | slot = path->slots[0]; | |
2888 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
2889 | if (ret < 0) | |
2890 | return ret; | |
2891 | ||
2892 | if (ret == 0) | |
2893 | progress++; | |
2894 | ||
2895 | if (progress) | |
2896 | return 0; | |
2897 | return 1; | |
2898 | } | |
2899 | ||
74123bd7 CM |
2900 | /* |
2901 | * split the path's leaf in two, making sure there is at least data_size | |
2902 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
2903 | * |
2904 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 2905 | */ |
e02119d5 CM |
2906 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
2907 | struct btrfs_root *root, | |
2908 | struct btrfs_key *ins_key, | |
2909 | struct btrfs_path *path, int data_size, | |
2910 | int extend) | |
be0e5c09 | 2911 | { |
5d4f98a2 | 2912 | struct btrfs_disk_key disk_key; |
5f39d397 | 2913 | struct extent_buffer *l; |
7518a238 | 2914 | u32 nritems; |
eb60ceac CM |
2915 | int mid; |
2916 | int slot; | |
5f39d397 | 2917 | struct extent_buffer *right; |
d4dbff95 | 2918 | int ret = 0; |
aa5d6bed | 2919 | int wret; |
5d4f98a2 | 2920 | int split; |
cc0c5538 | 2921 | int num_doubles = 0; |
99d8f83c | 2922 | int tried_avoid_double = 0; |
aa5d6bed | 2923 | |
a5719521 YZ |
2924 | l = path->nodes[0]; |
2925 | slot = path->slots[0]; | |
2926 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
2927 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
2928 | return -EOVERFLOW; | |
2929 | ||
40689478 | 2930 | /* first try to make some room by pushing left and right */ |
99d8f83c CM |
2931 | if (data_size) { |
2932 | wret = push_leaf_right(trans, root, path, data_size, | |
2933 | data_size, 0, 0); | |
d397712b | 2934 | if (wret < 0) |
eaee50e8 | 2935 | return wret; |
3685f791 | 2936 | if (wret) { |
99d8f83c CM |
2937 | wret = push_leaf_left(trans, root, path, data_size, |
2938 | data_size, 0, (u32)-1); | |
3685f791 CM |
2939 | if (wret < 0) |
2940 | return wret; | |
2941 | } | |
2942 | l = path->nodes[0]; | |
aa5d6bed | 2943 | |
3685f791 | 2944 | /* did the pushes work? */ |
87b29b20 | 2945 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 2946 | return 0; |
3326d1b0 | 2947 | } |
aa5d6bed | 2948 | |
5c680ed6 | 2949 | if (!path->nodes[1]) { |
e089f05c | 2950 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
2951 | if (ret) |
2952 | return ret; | |
2953 | } | |
cc0c5538 | 2954 | again: |
5d4f98a2 | 2955 | split = 1; |
cc0c5538 | 2956 | l = path->nodes[0]; |
eb60ceac | 2957 | slot = path->slots[0]; |
5f39d397 | 2958 | nritems = btrfs_header_nritems(l); |
d397712b | 2959 | mid = (nritems + 1) / 2; |
54aa1f4d | 2960 | |
5d4f98a2 YZ |
2961 | if (mid <= slot) { |
2962 | if (nritems == 1 || | |
2963 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
2964 | BTRFS_LEAF_DATA_SIZE(root)) { | |
2965 | if (slot >= nritems) { | |
2966 | split = 0; | |
2967 | } else { | |
2968 | mid = slot; | |
2969 | if (mid != nritems && | |
2970 | leaf_space_used(l, mid, nritems - mid) + | |
2971 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
2972 | if (data_size && !tried_avoid_double) |
2973 | goto push_for_double; | |
5d4f98a2 YZ |
2974 | split = 2; |
2975 | } | |
2976 | } | |
2977 | } | |
2978 | } else { | |
2979 | if (leaf_space_used(l, 0, mid) + data_size > | |
2980 | BTRFS_LEAF_DATA_SIZE(root)) { | |
2981 | if (!extend && data_size && slot == 0) { | |
2982 | split = 0; | |
2983 | } else if ((extend || !data_size) && slot == 0) { | |
2984 | mid = 1; | |
2985 | } else { | |
2986 | mid = slot; | |
2987 | if (mid != nritems && | |
2988 | leaf_space_used(l, mid, nritems - mid) + | |
2989 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
2990 | if (data_size && !tried_avoid_double) |
2991 | goto push_for_double; | |
5d4f98a2 YZ |
2992 | split = 2 ; |
2993 | } | |
2994 | } | |
2995 | } | |
2996 | } | |
2997 | ||
2998 | if (split == 0) | |
2999 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3000 | else | |
3001 | btrfs_item_key(l, &disk_key, mid); | |
3002 | ||
3003 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 3004 | root->root_key.objectid, |
66d7e7f0 | 3005 | &disk_key, 0, l->start, 0, 0); |
f0486c68 | 3006 | if (IS_ERR(right)) |
5f39d397 | 3007 | return PTR_ERR(right); |
f0486c68 YZ |
3008 | |
3009 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
3010 | |
3011 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 3012 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 3013 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 3014 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3015 | btrfs_set_header_owner(right, root->root_key.objectid); |
3016 | btrfs_set_header_level(right, 0); | |
3017 | write_extent_buffer(right, root->fs_info->fsid, | |
3018 | (unsigned long)btrfs_header_fsid(right), | |
3019 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3020 | |
3021 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
3022 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
3023 | BTRFS_UUID_SIZE); | |
44871b1b | 3024 | |
5d4f98a2 YZ |
3025 | if (split == 0) { |
3026 | if (mid <= slot) { | |
3027 | btrfs_set_header_nritems(right, 0); | |
143bede5 JM |
3028 | insert_ptr(trans, root, path, &disk_key, right->start, |
3029 | path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
3030 | btrfs_tree_unlock(path->nodes[0]); |
3031 | free_extent_buffer(path->nodes[0]); | |
3032 | path->nodes[0] = right; | |
3033 | path->slots[0] = 0; | |
3034 | path->slots[1] += 1; | |
3035 | } else { | |
3036 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 3037 | insert_ptr(trans, root, path, &disk_key, right->start, |
5d4f98a2 | 3038 | path->slots[1], 1); |
5d4f98a2 YZ |
3039 | btrfs_tree_unlock(path->nodes[0]); |
3040 | free_extent_buffer(path->nodes[0]); | |
3041 | path->nodes[0] = right; | |
3042 | path->slots[0] = 0; | |
143bede5 JM |
3043 | if (path->slots[1] == 0) |
3044 | fixup_low_keys(trans, root, path, | |
3045 | &disk_key, 1); | |
d4dbff95 | 3046 | } |
5d4f98a2 YZ |
3047 | btrfs_mark_buffer_dirty(right); |
3048 | return ret; | |
d4dbff95 | 3049 | } |
74123bd7 | 3050 | |
143bede5 | 3051 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 3052 | |
5d4f98a2 | 3053 | if (split == 2) { |
cc0c5538 CM |
3054 | BUG_ON(num_doubles != 0); |
3055 | num_doubles++; | |
3056 | goto again; | |
a429e513 | 3057 | } |
44871b1b | 3058 | |
143bede5 | 3059 | return 0; |
99d8f83c CM |
3060 | |
3061 | push_for_double: | |
3062 | push_for_double_split(trans, root, path, data_size); | |
3063 | tried_avoid_double = 1; | |
3064 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3065 | return 0; | |
3066 | goto again; | |
be0e5c09 CM |
3067 | } |
3068 | ||
ad48fd75 YZ |
3069 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3070 | struct btrfs_root *root, | |
3071 | struct btrfs_path *path, int ins_len) | |
459931ec | 3072 | { |
ad48fd75 | 3073 | struct btrfs_key key; |
459931ec | 3074 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3075 | struct btrfs_file_extent_item *fi; |
3076 | u64 extent_len = 0; | |
3077 | u32 item_size; | |
3078 | int ret; | |
459931ec CM |
3079 | |
3080 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3081 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3082 | ||
3083 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3084 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3085 | ||
3086 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
3087 | return 0; | |
459931ec CM |
3088 | |
3089 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
3090 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3091 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3092 | struct btrfs_file_extent_item); | |
3093 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3094 | } | |
b3b4aa74 | 3095 | btrfs_release_path(path); |
459931ec | 3096 | |
459931ec | 3097 | path->keep_locks = 1; |
ad48fd75 YZ |
3098 | path->search_for_split = 1; |
3099 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3100 | path->search_for_split = 0; |
ad48fd75 YZ |
3101 | if (ret < 0) |
3102 | goto err; | |
459931ec | 3103 | |
ad48fd75 YZ |
3104 | ret = -EAGAIN; |
3105 | leaf = path->nodes[0]; | |
459931ec | 3106 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
3107 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
3108 | goto err; | |
3109 | ||
109f6aef CM |
3110 | /* the leaf has changed, it now has room. return now */ |
3111 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
3112 | goto err; | |
3113 | ||
ad48fd75 YZ |
3114 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3115 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3116 | struct btrfs_file_extent_item); | |
3117 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3118 | goto err; | |
459931ec CM |
3119 | } |
3120 | ||
b9473439 | 3121 | btrfs_set_path_blocking(path); |
ad48fd75 | 3122 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3123 | if (ret) |
3124 | goto err; | |
459931ec | 3125 | |
ad48fd75 | 3126 | path->keep_locks = 0; |
b9473439 | 3127 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3128 | return 0; |
3129 | err: | |
3130 | path->keep_locks = 0; | |
3131 | return ret; | |
3132 | } | |
3133 | ||
3134 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
3135 | struct btrfs_root *root, | |
3136 | struct btrfs_path *path, | |
3137 | struct btrfs_key *new_key, | |
3138 | unsigned long split_offset) | |
3139 | { | |
3140 | struct extent_buffer *leaf; | |
3141 | struct btrfs_item *item; | |
3142 | struct btrfs_item *new_item; | |
3143 | int slot; | |
3144 | char *buf; | |
3145 | u32 nritems; | |
3146 | u32 item_size; | |
3147 | u32 orig_offset; | |
3148 | struct btrfs_disk_key disk_key; | |
3149 | ||
b9473439 CM |
3150 | leaf = path->nodes[0]; |
3151 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
3152 | ||
b4ce94de CM |
3153 | btrfs_set_path_blocking(path); |
3154 | ||
459931ec CM |
3155 | item = btrfs_item_nr(leaf, path->slots[0]); |
3156 | orig_offset = btrfs_item_offset(leaf, item); | |
3157 | item_size = btrfs_item_size(leaf, item); | |
3158 | ||
459931ec | 3159 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3160 | if (!buf) |
3161 | return -ENOMEM; | |
3162 | ||
459931ec CM |
3163 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3164 | path->slots[0]), item_size); | |
459931ec | 3165 | |
ad48fd75 | 3166 | slot = path->slots[0] + 1; |
459931ec | 3167 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3168 | if (slot != nritems) { |
3169 | /* shift the items */ | |
3170 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3171 | btrfs_item_nr_offset(slot), |
3172 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3173 | } |
3174 | ||
3175 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3176 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3177 | ||
3178 | new_item = btrfs_item_nr(leaf, slot); | |
3179 | ||
3180 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3181 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3182 | ||
3183 | btrfs_set_item_offset(leaf, item, | |
3184 | orig_offset + item_size - split_offset); | |
3185 | btrfs_set_item_size(leaf, item, split_offset); | |
3186 | ||
3187 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3188 | ||
3189 | /* write the data for the start of the original item */ | |
3190 | write_extent_buffer(leaf, buf, | |
3191 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3192 | split_offset); | |
3193 | ||
3194 | /* write the data for the new item */ | |
3195 | write_extent_buffer(leaf, buf + split_offset, | |
3196 | btrfs_item_ptr_offset(leaf, slot), | |
3197 | item_size - split_offset); | |
3198 | btrfs_mark_buffer_dirty(leaf); | |
3199 | ||
ad48fd75 | 3200 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 3201 | kfree(buf); |
ad48fd75 YZ |
3202 | return 0; |
3203 | } | |
3204 | ||
3205 | /* | |
3206 | * This function splits a single item into two items, | |
3207 | * giving 'new_key' to the new item and splitting the | |
3208 | * old one at split_offset (from the start of the item). | |
3209 | * | |
3210 | * The path may be released by this operation. After | |
3211 | * the split, the path is pointing to the old item. The | |
3212 | * new item is going to be in the same node as the old one. | |
3213 | * | |
3214 | * Note, the item being split must be smaller enough to live alone on | |
3215 | * a tree block with room for one extra struct btrfs_item | |
3216 | * | |
3217 | * This allows us to split the item in place, keeping a lock on the | |
3218 | * leaf the entire time. | |
3219 | */ | |
3220 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3221 | struct btrfs_root *root, | |
3222 | struct btrfs_path *path, | |
3223 | struct btrfs_key *new_key, | |
3224 | unsigned long split_offset) | |
3225 | { | |
3226 | int ret; | |
3227 | ret = setup_leaf_for_split(trans, root, path, | |
3228 | sizeof(struct btrfs_item)); | |
3229 | if (ret) | |
3230 | return ret; | |
3231 | ||
3232 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
3233 | return ret; |
3234 | } | |
3235 | ||
ad48fd75 YZ |
3236 | /* |
3237 | * This function duplicate a item, giving 'new_key' to the new item. | |
3238 | * It guarantees both items live in the same tree leaf and the new item | |
3239 | * is contiguous with the original item. | |
3240 | * | |
3241 | * This allows us to split file extent in place, keeping a lock on the | |
3242 | * leaf the entire time. | |
3243 | */ | |
3244 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
3245 | struct btrfs_root *root, | |
3246 | struct btrfs_path *path, | |
3247 | struct btrfs_key *new_key) | |
3248 | { | |
3249 | struct extent_buffer *leaf; | |
3250 | int ret; | |
3251 | u32 item_size; | |
3252 | ||
3253 | leaf = path->nodes[0]; | |
3254 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3255 | ret = setup_leaf_for_split(trans, root, path, | |
3256 | item_size + sizeof(struct btrfs_item)); | |
3257 | if (ret) | |
3258 | return ret; | |
3259 | ||
3260 | path->slots[0]++; | |
143bede5 JM |
3261 | setup_items_for_insert(trans, root, path, new_key, &item_size, |
3262 | item_size, item_size + | |
3263 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
3264 | leaf = path->nodes[0]; |
3265 | memcpy_extent_buffer(leaf, | |
3266 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3267 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
3268 | item_size); | |
3269 | return 0; | |
3270 | } | |
3271 | ||
d352ac68 CM |
3272 | /* |
3273 | * make the item pointed to by the path smaller. new_size indicates | |
3274 | * how small to make it, and from_end tells us if we just chop bytes | |
3275 | * off the end of the item or if we shift the item to chop bytes off | |
3276 | * the front. | |
3277 | */ | |
143bede5 JM |
3278 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3279 | struct btrfs_root *root, | |
3280 | struct btrfs_path *path, | |
3281 | u32 new_size, int from_end) | |
b18c6685 | 3282 | { |
b18c6685 | 3283 | int slot; |
5f39d397 CM |
3284 | struct extent_buffer *leaf; |
3285 | struct btrfs_item *item; | |
b18c6685 CM |
3286 | u32 nritems; |
3287 | unsigned int data_end; | |
3288 | unsigned int old_data_start; | |
3289 | unsigned int old_size; | |
3290 | unsigned int size_diff; | |
3291 | int i; | |
cfed81a0 CM |
3292 | struct btrfs_map_token token; |
3293 | ||
3294 | btrfs_init_map_token(&token); | |
b18c6685 | 3295 | |
5f39d397 | 3296 | leaf = path->nodes[0]; |
179e29e4 CM |
3297 | slot = path->slots[0]; |
3298 | ||
3299 | old_size = btrfs_item_size_nr(leaf, slot); | |
3300 | if (old_size == new_size) | |
143bede5 | 3301 | return; |
b18c6685 | 3302 | |
5f39d397 | 3303 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
3304 | data_end = leaf_data_end(root, leaf); |
3305 | ||
5f39d397 | 3306 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3307 | |
b18c6685 CM |
3308 | size_diff = old_size - new_size; |
3309 | ||
3310 | BUG_ON(slot < 0); | |
3311 | BUG_ON(slot >= nritems); | |
3312 | ||
3313 | /* | |
3314 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3315 | */ | |
3316 | /* first correct the data pointers */ | |
3317 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3318 | u32 ioff; |
3319 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3320 | |
cfed81a0 CM |
3321 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3322 | btrfs_set_token_item_offset(leaf, item, | |
3323 | ioff + size_diff, &token); | |
b18c6685 | 3324 | } |
db94535d | 3325 | |
b18c6685 | 3326 | /* shift the data */ |
179e29e4 CM |
3327 | if (from_end) { |
3328 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3329 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3330 | data_end, old_data_start + new_size - data_end); | |
3331 | } else { | |
3332 | struct btrfs_disk_key disk_key; | |
3333 | u64 offset; | |
3334 | ||
3335 | btrfs_item_key(leaf, &disk_key, slot); | |
3336 | ||
3337 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3338 | unsigned long ptr; | |
3339 | struct btrfs_file_extent_item *fi; | |
3340 | ||
3341 | fi = btrfs_item_ptr(leaf, slot, | |
3342 | struct btrfs_file_extent_item); | |
3343 | fi = (struct btrfs_file_extent_item *)( | |
3344 | (unsigned long)fi - size_diff); | |
3345 | ||
3346 | if (btrfs_file_extent_type(leaf, fi) == | |
3347 | BTRFS_FILE_EXTENT_INLINE) { | |
3348 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3349 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
3350 | (unsigned long)fi, |
3351 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
3352 | disk_bytenr)); |
3353 | } | |
3354 | } | |
3355 | ||
3356 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3357 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3358 | data_end, old_data_start - data_end); | |
3359 | ||
3360 | offset = btrfs_disk_key_offset(&disk_key); | |
3361 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3362 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3363 | if (slot == 0) | |
3364 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
3365 | } | |
5f39d397 CM |
3366 | |
3367 | item = btrfs_item_nr(leaf, slot); | |
3368 | btrfs_set_item_size(leaf, item, new_size); | |
3369 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 3370 | |
5f39d397 CM |
3371 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3372 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 3373 | BUG(); |
5f39d397 | 3374 | } |
b18c6685 CM |
3375 | } |
3376 | ||
d352ac68 CM |
3377 | /* |
3378 | * make the item pointed to by the path bigger, data_size is the new size. | |
3379 | */ | |
143bede5 JM |
3380 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
3381 | struct btrfs_root *root, struct btrfs_path *path, | |
3382 | u32 data_size) | |
6567e837 | 3383 | { |
6567e837 | 3384 | int slot; |
5f39d397 CM |
3385 | struct extent_buffer *leaf; |
3386 | struct btrfs_item *item; | |
6567e837 CM |
3387 | u32 nritems; |
3388 | unsigned int data_end; | |
3389 | unsigned int old_data; | |
3390 | unsigned int old_size; | |
3391 | int i; | |
cfed81a0 CM |
3392 | struct btrfs_map_token token; |
3393 | ||
3394 | btrfs_init_map_token(&token); | |
6567e837 | 3395 | |
5f39d397 | 3396 | leaf = path->nodes[0]; |
6567e837 | 3397 | |
5f39d397 | 3398 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
3399 | data_end = leaf_data_end(root, leaf); |
3400 | ||
5f39d397 CM |
3401 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
3402 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3403 | BUG(); |
5f39d397 | 3404 | } |
6567e837 | 3405 | slot = path->slots[0]; |
5f39d397 | 3406 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3407 | |
3408 | BUG_ON(slot < 0); | |
3326d1b0 CM |
3409 | if (slot >= nritems) { |
3410 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
3411 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
3412 | slot, nritems); | |
3326d1b0 CM |
3413 | BUG_ON(1); |
3414 | } | |
6567e837 CM |
3415 | |
3416 | /* | |
3417 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3418 | */ | |
3419 | /* first correct the data pointers */ | |
3420 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3421 | u32 ioff; |
3422 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3423 | |
cfed81a0 CM |
3424 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3425 | btrfs_set_token_item_offset(leaf, item, | |
3426 | ioff - data_size, &token); | |
6567e837 | 3427 | } |
5f39d397 | 3428 | |
6567e837 | 3429 | /* shift the data */ |
5f39d397 | 3430 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
3431 | data_end - data_size, btrfs_leaf_data(leaf) + |
3432 | data_end, old_data - data_end); | |
5f39d397 | 3433 | |
6567e837 | 3434 | data_end = old_data; |
5f39d397 CM |
3435 | old_size = btrfs_item_size_nr(leaf, slot); |
3436 | item = btrfs_item_nr(leaf, slot); | |
3437 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
3438 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 3439 | |
5f39d397 CM |
3440 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3441 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3442 | BUG(); |
5f39d397 | 3443 | } |
6567e837 CM |
3444 | } |
3445 | ||
f3465ca4 JB |
3446 | /* |
3447 | * Given a key and some data, insert items into the tree. | |
3448 | * This does all the path init required, making room in the tree if needed. | |
3449 | * Returns the number of keys that were inserted. | |
3450 | */ | |
3451 | int btrfs_insert_some_items(struct btrfs_trans_handle *trans, | |
3452 | struct btrfs_root *root, | |
3453 | struct btrfs_path *path, | |
3454 | struct btrfs_key *cpu_key, u32 *data_size, | |
3455 | int nr) | |
3456 | { | |
3457 | struct extent_buffer *leaf; | |
3458 | struct btrfs_item *item; | |
3459 | int ret = 0; | |
3460 | int slot; | |
f3465ca4 JB |
3461 | int i; |
3462 | u32 nritems; | |
3463 | u32 total_data = 0; | |
3464 | u32 total_size = 0; | |
3465 | unsigned int data_end; | |
3466 | struct btrfs_disk_key disk_key; | |
3467 | struct btrfs_key found_key; | |
cfed81a0 CM |
3468 | struct btrfs_map_token token; |
3469 | ||
3470 | btrfs_init_map_token(&token); | |
f3465ca4 | 3471 | |
87b29b20 YZ |
3472 | for (i = 0; i < nr; i++) { |
3473 | if (total_size + data_size[i] + sizeof(struct btrfs_item) > | |
3474 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3475 | break; | |
3476 | nr = i; | |
3477 | } | |
f3465ca4 | 3478 | total_data += data_size[i]; |
87b29b20 YZ |
3479 | total_size += data_size[i] + sizeof(struct btrfs_item); |
3480 | } | |
3481 | BUG_ON(nr == 0); | |
f3465ca4 | 3482 | |
f3465ca4 JB |
3483 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
3484 | if (ret == 0) | |
3485 | return -EEXIST; | |
3486 | if (ret < 0) | |
3487 | goto out; | |
3488 | ||
f3465ca4 JB |
3489 | leaf = path->nodes[0]; |
3490 | ||
3491 | nritems = btrfs_header_nritems(leaf); | |
3492 | data_end = leaf_data_end(root, leaf); | |
3493 | ||
3494 | if (btrfs_leaf_free_space(root, leaf) < total_size) { | |
3495 | for (i = nr; i >= 0; i--) { | |
3496 | total_data -= data_size[i]; | |
3497 | total_size -= data_size[i] + sizeof(struct btrfs_item); | |
3498 | if (total_size < btrfs_leaf_free_space(root, leaf)) | |
3499 | break; | |
3500 | } | |
3501 | nr = i; | |
3502 | } | |
3503 | ||
3504 | slot = path->slots[0]; | |
3505 | BUG_ON(slot < 0); | |
3506 | ||
3507 | if (slot != nritems) { | |
3508 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); | |
3509 | ||
3510 | item = btrfs_item_nr(leaf, slot); | |
3511 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3512 | ||
3513 | /* figure out how many keys we can insert in here */ | |
3514 | total_data = data_size[0]; | |
3515 | for (i = 1; i < nr; i++) { | |
5d4f98a2 | 3516 | if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0) |
f3465ca4 JB |
3517 | break; |
3518 | total_data += data_size[i]; | |
3519 | } | |
3520 | nr = i; | |
3521 | ||
3522 | if (old_data < data_end) { | |
3523 | btrfs_print_leaf(root, leaf); | |
d397712b | 3524 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
f3465ca4 JB |
3525 | slot, old_data, data_end); |
3526 | BUG_ON(1); | |
3527 | } | |
3528 | /* | |
3529 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3530 | */ | |
3531 | /* first correct the data pointers */ | |
f3465ca4 JB |
3532 | for (i = slot; i < nritems; i++) { |
3533 | u32 ioff; | |
3534 | ||
3535 | item = btrfs_item_nr(leaf, i); | |
cfed81a0 CM |
3536 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3537 | btrfs_set_token_item_offset(leaf, item, | |
3538 | ioff - total_data, &token); | |
f3465ca4 | 3539 | } |
f3465ca4 JB |
3540 | /* shift the items */ |
3541 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), | |
3542 | btrfs_item_nr_offset(slot), | |
3543 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3544 | ||
3545 | /* shift the data */ | |
3546 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3547 | data_end - total_data, btrfs_leaf_data(leaf) + | |
3548 | data_end, old_data - data_end); | |
3549 | data_end = old_data; | |
3550 | } else { | |
3551 | /* | |
3552 | * this sucks but it has to be done, if we are inserting at | |
3553 | * the end of the leaf only insert 1 of the items, since we | |
3554 | * have no way of knowing whats on the next leaf and we'd have | |
3555 | * to drop our current locks to figure it out | |
3556 | */ | |
3557 | nr = 1; | |
3558 | } | |
3559 | ||
3560 | /* setup the item for the new data */ | |
3561 | for (i = 0; i < nr; i++) { | |
3562 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3563 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3564 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
3565 | btrfs_set_token_item_offset(leaf, item, |
3566 | data_end - data_size[i], &token); | |
f3465ca4 | 3567 | data_end -= data_size[i]; |
cfed81a0 | 3568 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
f3465ca4 JB |
3569 | } |
3570 | btrfs_set_header_nritems(leaf, nritems + nr); | |
3571 | btrfs_mark_buffer_dirty(leaf); | |
3572 | ||
3573 | ret = 0; | |
3574 | if (slot == 0) { | |
3575 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3576 | fixup_low_keys(trans, root, path, &disk_key, 1); |
f3465ca4 JB |
3577 | } |
3578 | ||
3579 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
3580 | btrfs_print_leaf(root, leaf); | |
3581 | BUG(); | |
3582 | } | |
3583 | out: | |
3584 | if (!ret) | |
3585 | ret = nr; | |
3586 | return ret; | |
3587 | } | |
3588 | ||
74123bd7 | 3589 | /* |
44871b1b CM |
3590 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
3591 | * to save stack depth by doing the bulk of the work in a function | |
3592 | * that doesn't call btrfs_search_slot | |
74123bd7 | 3593 | */ |
143bede5 JM |
3594 | void setup_items_for_insert(struct btrfs_trans_handle *trans, |
3595 | struct btrfs_root *root, struct btrfs_path *path, | |
3596 | struct btrfs_key *cpu_key, u32 *data_size, | |
3597 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 3598 | { |
5f39d397 | 3599 | struct btrfs_item *item; |
9c58309d | 3600 | int i; |
7518a238 | 3601 | u32 nritems; |
be0e5c09 | 3602 | unsigned int data_end; |
e2fa7227 | 3603 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3604 | struct extent_buffer *leaf; |
3605 | int slot; | |
cfed81a0 CM |
3606 | struct btrfs_map_token token; |
3607 | ||
3608 | btrfs_init_map_token(&token); | |
e2fa7227 | 3609 | |
5f39d397 | 3610 | leaf = path->nodes[0]; |
44871b1b | 3611 | slot = path->slots[0]; |
74123bd7 | 3612 | |
5f39d397 | 3613 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 3614 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 3615 | |
f25956cc | 3616 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 3617 | btrfs_print_leaf(root, leaf); |
d397712b | 3618 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 3619 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 3620 | BUG(); |
d4dbff95 | 3621 | } |
5f39d397 | 3622 | |
be0e5c09 | 3623 | if (slot != nritems) { |
5f39d397 | 3624 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 3625 | |
5f39d397 CM |
3626 | if (old_data < data_end) { |
3627 | btrfs_print_leaf(root, leaf); | |
d397712b | 3628 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
3629 | slot, old_data, data_end); |
3630 | BUG_ON(1); | |
3631 | } | |
be0e5c09 CM |
3632 | /* |
3633 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3634 | */ | |
3635 | /* first correct the data pointers */ | |
0783fcfc | 3636 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3637 | u32 ioff; |
db94535d | 3638 | |
5f39d397 | 3639 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
3640 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3641 | btrfs_set_token_item_offset(leaf, item, | |
3642 | ioff - total_data, &token); | |
0783fcfc | 3643 | } |
be0e5c09 | 3644 | /* shift the items */ |
9c58309d | 3645 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 3646 | btrfs_item_nr_offset(slot), |
d6025579 | 3647 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
3648 | |
3649 | /* shift the data */ | |
5f39d397 | 3650 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 3651 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 3652 | data_end, old_data - data_end); |
be0e5c09 CM |
3653 | data_end = old_data; |
3654 | } | |
5f39d397 | 3655 | |
62e2749e | 3656 | /* setup the item for the new data */ |
9c58309d CM |
3657 | for (i = 0; i < nr; i++) { |
3658 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3659 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3660 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
3661 | btrfs_set_token_item_offset(leaf, item, |
3662 | data_end - data_size[i], &token); | |
9c58309d | 3663 | data_end -= data_size[i]; |
cfed81a0 | 3664 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 3665 | } |
44871b1b | 3666 | |
9c58309d | 3667 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 3668 | |
5a01a2e3 CM |
3669 | if (slot == 0) { |
3670 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3671 | fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 3672 | } |
b9473439 CM |
3673 | btrfs_unlock_up_safe(path, 1); |
3674 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 3675 | |
5f39d397 CM |
3676 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3677 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 3678 | BUG(); |
5f39d397 | 3679 | } |
44871b1b CM |
3680 | } |
3681 | ||
3682 | /* | |
3683 | * Given a key and some data, insert items into the tree. | |
3684 | * This does all the path init required, making room in the tree if needed. | |
3685 | */ | |
3686 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
3687 | struct btrfs_root *root, | |
3688 | struct btrfs_path *path, | |
3689 | struct btrfs_key *cpu_key, u32 *data_size, | |
3690 | int nr) | |
3691 | { | |
44871b1b CM |
3692 | int ret = 0; |
3693 | int slot; | |
3694 | int i; | |
3695 | u32 total_size = 0; | |
3696 | u32 total_data = 0; | |
3697 | ||
3698 | for (i = 0; i < nr; i++) | |
3699 | total_data += data_size[i]; | |
3700 | ||
3701 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
3702 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
3703 | if (ret == 0) | |
3704 | return -EEXIST; | |
3705 | if (ret < 0) | |
143bede5 | 3706 | return ret; |
44871b1b | 3707 | |
44871b1b CM |
3708 | slot = path->slots[0]; |
3709 | BUG_ON(slot < 0); | |
3710 | ||
143bede5 | 3711 | setup_items_for_insert(trans, root, path, cpu_key, data_size, |
44871b1b | 3712 | total_data, total_size, nr); |
143bede5 | 3713 | return 0; |
62e2749e CM |
3714 | } |
3715 | ||
3716 | /* | |
3717 | * Given a key and some data, insert an item into the tree. | |
3718 | * This does all the path init required, making room in the tree if needed. | |
3719 | */ | |
e089f05c CM |
3720 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
3721 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
3722 | data_size) | |
62e2749e CM |
3723 | { |
3724 | int ret = 0; | |
2c90e5d6 | 3725 | struct btrfs_path *path; |
5f39d397 CM |
3726 | struct extent_buffer *leaf; |
3727 | unsigned long ptr; | |
62e2749e | 3728 | |
2c90e5d6 | 3729 | path = btrfs_alloc_path(); |
db5b493a TI |
3730 | if (!path) |
3731 | return -ENOMEM; | |
2c90e5d6 | 3732 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 3733 | if (!ret) { |
5f39d397 CM |
3734 | leaf = path->nodes[0]; |
3735 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3736 | write_extent_buffer(leaf, data, ptr, data_size); | |
3737 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 3738 | } |
2c90e5d6 | 3739 | btrfs_free_path(path); |
aa5d6bed | 3740 | return ret; |
be0e5c09 CM |
3741 | } |
3742 | ||
74123bd7 | 3743 | /* |
5de08d7d | 3744 | * delete the pointer from a given node. |
74123bd7 | 3745 | * |
d352ac68 CM |
3746 | * the tree should have been previously balanced so the deletion does not |
3747 | * empty a node. | |
74123bd7 | 3748 | */ |
143bede5 JM |
3749 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3750 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 3751 | { |
5f39d397 | 3752 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 3753 | u32 nritems; |
be0e5c09 | 3754 | |
5f39d397 | 3755 | nritems = btrfs_header_nritems(parent); |
d397712b | 3756 | if (slot != nritems - 1) { |
5f39d397 CM |
3757 | memmove_extent_buffer(parent, |
3758 | btrfs_node_key_ptr_offset(slot), | |
3759 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
3760 | sizeof(struct btrfs_key_ptr) * |
3761 | (nritems - slot - 1)); | |
bb803951 | 3762 | } |
7518a238 | 3763 | nritems--; |
5f39d397 | 3764 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 3765 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 3766 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 3767 | /* just turn the root into a leaf and break */ |
5f39d397 | 3768 | btrfs_set_header_level(root->node, 0); |
bb803951 | 3769 | } else if (slot == 0) { |
5f39d397 CM |
3770 | struct btrfs_disk_key disk_key; |
3771 | ||
3772 | btrfs_node_key(parent, &disk_key, 0); | |
143bede5 | 3773 | fixup_low_keys(trans, root, path, &disk_key, level + 1); |
be0e5c09 | 3774 | } |
d6025579 | 3775 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
3776 | } |
3777 | ||
323ac95b CM |
3778 | /* |
3779 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 3780 | * path->nodes[1]. |
323ac95b CM |
3781 | * |
3782 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
3783 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
3784 | * | |
3785 | * The path must have already been setup for deleting the leaf, including | |
3786 | * all the proper balancing. path->nodes[1] must be locked. | |
3787 | */ | |
143bede5 JM |
3788 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
3789 | struct btrfs_root *root, | |
3790 | struct btrfs_path *path, | |
3791 | struct extent_buffer *leaf) | |
323ac95b | 3792 | { |
5d4f98a2 | 3793 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
143bede5 | 3794 | del_ptr(trans, root, path, 1, path->slots[1]); |
323ac95b | 3795 | |
4d081c41 CM |
3796 | /* |
3797 | * btrfs_free_extent is expensive, we want to make sure we | |
3798 | * aren't holding any locks when we call it | |
3799 | */ | |
3800 | btrfs_unlock_up_safe(path, 0); | |
3801 | ||
f0486c68 YZ |
3802 | root_sub_used(root, leaf->len); |
3803 | ||
3083ee2e | 3804 | extent_buffer_get(leaf); |
66d7e7f0 | 3805 | btrfs_free_tree_block(trans, root, leaf, 0, 1, 0); |
3083ee2e | 3806 | free_extent_buffer_stale(leaf); |
323ac95b | 3807 | } |
74123bd7 CM |
3808 | /* |
3809 | * delete the item at the leaf level in path. If that empties | |
3810 | * the leaf, remove it from the tree | |
3811 | */ | |
85e21bac CM |
3812 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3813 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 3814 | { |
5f39d397 CM |
3815 | struct extent_buffer *leaf; |
3816 | struct btrfs_item *item; | |
85e21bac CM |
3817 | int last_off; |
3818 | int dsize = 0; | |
aa5d6bed CM |
3819 | int ret = 0; |
3820 | int wret; | |
85e21bac | 3821 | int i; |
7518a238 | 3822 | u32 nritems; |
cfed81a0 CM |
3823 | struct btrfs_map_token token; |
3824 | ||
3825 | btrfs_init_map_token(&token); | |
be0e5c09 | 3826 | |
5f39d397 | 3827 | leaf = path->nodes[0]; |
85e21bac CM |
3828 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
3829 | ||
3830 | for (i = 0; i < nr; i++) | |
3831 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
3832 | ||
5f39d397 | 3833 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 3834 | |
85e21bac | 3835 | if (slot + nr != nritems) { |
123abc88 | 3836 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
3837 | |
3838 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
3839 | data_end + dsize, |
3840 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 3841 | last_off - data_end); |
5f39d397 | 3842 | |
85e21bac | 3843 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 3844 | u32 ioff; |
db94535d | 3845 | |
5f39d397 | 3846 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
3847 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3848 | btrfs_set_token_item_offset(leaf, item, | |
3849 | ioff + dsize, &token); | |
0783fcfc | 3850 | } |
db94535d | 3851 | |
5f39d397 | 3852 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 3853 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 3854 | sizeof(struct btrfs_item) * |
85e21bac | 3855 | (nritems - slot - nr)); |
be0e5c09 | 3856 | } |
85e21bac CM |
3857 | btrfs_set_header_nritems(leaf, nritems - nr); |
3858 | nritems -= nr; | |
5f39d397 | 3859 | |
74123bd7 | 3860 | /* delete the leaf if we've emptied it */ |
7518a238 | 3861 | if (nritems == 0) { |
5f39d397 CM |
3862 | if (leaf == root->node) { |
3863 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 3864 | } else { |
f0486c68 YZ |
3865 | btrfs_set_path_blocking(path); |
3866 | clean_tree_block(trans, root, leaf); | |
143bede5 | 3867 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 3868 | } |
be0e5c09 | 3869 | } else { |
7518a238 | 3870 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 3871 | if (slot == 0) { |
5f39d397 CM |
3872 | struct btrfs_disk_key disk_key; |
3873 | ||
3874 | btrfs_item_key(leaf, &disk_key, 0); | |
143bede5 | 3875 | fixup_low_keys(trans, root, path, &disk_key, 1); |
aa5d6bed | 3876 | } |
aa5d6bed | 3877 | |
74123bd7 | 3878 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 3879 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
3880 | /* push_leaf_left fixes the path. |
3881 | * make sure the path still points to our leaf | |
3882 | * for possible call to del_ptr below | |
3883 | */ | |
4920c9ac | 3884 | slot = path->slots[1]; |
5f39d397 CM |
3885 | extent_buffer_get(leaf); |
3886 | ||
b9473439 | 3887 | btrfs_set_path_blocking(path); |
99d8f83c CM |
3888 | wret = push_leaf_left(trans, root, path, 1, 1, |
3889 | 1, (u32)-1); | |
54aa1f4d | 3890 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 3891 | ret = wret; |
5f39d397 CM |
3892 | |
3893 | if (path->nodes[0] == leaf && | |
3894 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
3895 | wret = push_leaf_right(trans, root, path, 1, |
3896 | 1, 1, 0); | |
54aa1f4d | 3897 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
3898 | ret = wret; |
3899 | } | |
5f39d397 CM |
3900 | |
3901 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 3902 | path->slots[1] = slot; |
143bede5 | 3903 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 3904 | free_extent_buffer(leaf); |
143bede5 | 3905 | ret = 0; |
5de08d7d | 3906 | } else { |
925baedd CM |
3907 | /* if we're still in the path, make sure |
3908 | * we're dirty. Otherwise, one of the | |
3909 | * push_leaf functions must have already | |
3910 | * dirtied this buffer | |
3911 | */ | |
3912 | if (path->nodes[0] == leaf) | |
3913 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 3914 | free_extent_buffer(leaf); |
be0e5c09 | 3915 | } |
d5719762 | 3916 | } else { |
5f39d397 | 3917 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
3918 | } |
3919 | } | |
aa5d6bed | 3920 | return ret; |
be0e5c09 CM |
3921 | } |
3922 | ||
7bb86316 | 3923 | /* |
925baedd | 3924 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
3925 | * returns 0 if it found something or 1 if there are no lesser leaves. |
3926 | * returns < 0 on io errors. | |
d352ac68 CM |
3927 | * |
3928 | * This may release the path, and so you may lose any locks held at the | |
3929 | * time you call it. | |
7bb86316 CM |
3930 | */ |
3931 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
3932 | { | |
925baedd CM |
3933 | struct btrfs_key key; |
3934 | struct btrfs_disk_key found_key; | |
3935 | int ret; | |
7bb86316 | 3936 | |
925baedd | 3937 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 3938 | |
925baedd CM |
3939 | if (key.offset > 0) |
3940 | key.offset--; | |
3941 | else if (key.type > 0) | |
3942 | key.type--; | |
3943 | else if (key.objectid > 0) | |
3944 | key.objectid--; | |
3945 | else | |
3946 | return 1; | |
7bb86316 | 3947 | |
b3b4aa74 | 3948 | btrfs_release_path(path); |
925baedd CM |
3949 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3950 | if (ret < 0) | |
3951 | return ret; | |
3952 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
3953 | ret = comp_keys(&found_key, &key); | |
3954 | if (ret < 0) | |
3955 | return 0; | |
3956 | return 1; | |
7bb86316 CM |
3957 | } |
3958 | ||
3f157a2f CM |
3959 | /* |
3960 | * A helper function to walk down the tree starting at min_key, and looking | |
3961 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 3962 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
3963 | * |
3964 | * This does not cow, but it does stuff the starting key it finds back | |
3965 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
3966 | * key and get a writable path. | |
3967 | * | |
3968 | * This does lock as it descends, and path->keep_locks should be set | |
3969 | * to 1 by the caller. | |
3970 | * | |
3971 | * This honors path->lowest_level to prevent descent past a given level | |
3972 | * of the tree. | |
3973 | * | |
d352ac68 CM |
3974 | * min_trans indicates the oldest transaction that you are interested |
3975 | * in walking through. Any nodes or leaves older than min_trans are | |
3976 | * skipped over (without reading them). | |
3977 | * | |
3f157a2f CM |
3978 | * returns zero if something useful was found, < 0 on error and 1 if there |
3979 | * was nothing in the tree that matched the search criteria. | |
3980 | */ | |
3981 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 3982 | struct btrfs_key *max_key, |
3f157a2f CM |
3983 | struct btrfs_path *path, int cache_only, |
3984 | u64 min_trans) | |
3985 | { | |
3986 | struct extent_buffer *cur; | |
3987 | struct btrfs_key found_key; | |
3988 | int slot; | |
9652480b | 3989 | int sret; |
3f157a2f CM |
3990 | u32 nritems; |
3991 | int level; | |
3992 | int ret = 1; | |
3993 | ||
934d375b | 3994 | WARN_ON(!path->keep_locks); |
3f157a2f | 3995 | again: |
bd681513 | 3996 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 3997 | level = btrfs_header_level(cur); |
e02119d5 | 3998 | WARN_ON(path->nodes[level]); |
3f157a2f | 3999 | path->nodes[level] = cur; |
bd681513 | 4000 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4001 | |
4002 | if (btrfs_header_generation(cur) < min_trans) { | |
4003 | ret = 1; | |
4004 | goto out; | |
4005 | } | |
d397712b | 4006 | while (1) { |
3f157a2f CM |
4007 | nritems = btrfs_header_nritems(cur); |
4008 | level = btrfs_header_level(cur); | |
9652480b | 4009 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4010 | |
323ac95b CM |
4011 | /* at the lowest level, we're done, setup the path and exit */ |
4012 | if (level == path->lowest_level) { | |
e02119d5 CM |
4013 | if (slot >= nritems) |
4014 | goto find_next_key; | |
3f157a2f CM |
4015 | ret = 0; |
4016 | path->slots[level] = slot; | |
4017 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4018 | goto out; | |
4019 | } | |
9652480b Y |
4020 | if (sret && slot > 0) |
4021 | slot--; | |
3f157a2f CM |
4022 | /* |
4023 | * check this node pointer against the cache_only and | |
4024 | * min_trans parameters. If it isn't in cache or is too | |
4025 | * old, skip to the next one. | |
4026 | */ | |
d397712b | 4027 | while (slot < nritems) { |
3f157a2f CM |
4028 | u64 blockptr; |
4029 | u64 gen; | |
4030 | struct extent_buffer *tmp; | |
e02119d5 CM |
4031 | struct btrfs_disk_key disk_key; |
4032 | ||
3f157a2f CM |
4033 | blockptr = btrfs_node_blockptr(cur, slot); |
4034 | gen = btrfs_node_ptr_generation(cur, slot); | |
4035 | if (gen < min_trans) { | |
4036 | slot++; | |
4037 | continue; | |
4038 | } | |
4039 | if (!cache_only) | |
4040 | break; | |
4041 | ||
e02119d5 CM |
4042 | if (max_key) { |
4043 | btrfs_node_key(cur, &disk_key, slot); | |
4044 | if (comp_keys(&disk_key, max_key) >= 0) { | |
4045 | ret = 1; | |
4046 | goto out; | |
4047 | } | |
4048 | } | |
4049 | ||
3f157a2f CM |
4050 | tmp = btrfs_find_tree_block(root, blockptr, |
4051 | btrfs_level_size(root, level - 1)); | |
4052 | ||
b9fab919 | 4053 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
3f157a2f CM |
4054 | free_extent_buffer(tmp); |
4055 | break; | |
4056 | } | |
4057 | if (tmp) | |
4058 | free_extent_buffer(tmp); | |
4059 | slot++; | |
4060 | } | |
e02119d5 | 4061 | find_next_key: |
3f157a2f CM |
4062 | /* |
4063 | * we didn't find a candidate key in this node, walk forward | |
4064 | * and find another one | |
4065 | */ | |
4066 | if (slot >= nritems) { | |
e02119d5 | 4067 | path->slots[level] = slot; |
b4ce94de | 4068 | btrfs_set_path_blocking(path); |
e02119d5 | 4069 | sret = btrfs_find_next_key(root, path, min_key, level, |
3f157a2f | 4070 | cache_only, min_trans); |
e02119d5 | 4071 | if (sret == 0) { |
b3b4aa74 | 4072 | btrfs_release_path(path); |
3f157a2f CM |
4073 | goto again; |
4074 | } else { | |
4075 | goto out; | |
4076 | } | |
4077 | } | |
4078 | /* save our key for returning back */ | |
4079 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4080 | path->slots[level] = slot; | |
4081 | if (level == path->lowest_level) { | |
4082 | ret = 0; | |
f7c79f30 | 4083 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4084 | goto out; |
4085 | } | |
b4ce94de | 4086 | btrfs_set_path_blocking(path); |
3f157a2f | 4087 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4088 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4089 | |
bd681513 | 4090 | btrfs_tree_read_lock(cur); |
b4ce94de | 4091 | |
bd681513 | 4092 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4093 | path->nodes[level - 1] = cur; |
f7c79f30 | 4094 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4095 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4096 | } |
4097 | out: | |
4098 | if (ret == 0) | |
4099 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4100 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4101 | return ret; |
4102 | } | |
4103 | ||
4104 | /* | |
4105 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4106 | * and fixup the path. It looks for and returns the next key in the | |
4107 | * tree based on the current path and the cache_only and min_trans | |
4108 | * parameters. | |
4109 | * | |
4110 | * 0 is returned if another key is found, < 0 if there are any errors | |
4111 | * and 1 is returned if there are no higher keys in the tree | |
4112 | * | |
4113 | * path->keep_locks should be set to 1 on the search made before | |
4114 | * calling this function. | |
4115 | */ | |
e7a84565 | 4116 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
33c66f43 | 4117 | struct btrfs_key *key, int level, |
3f157a2f | 4118 | int cache_only, u64 min_trans) |
e7a84565 | 4119 | { |
e7a84565 CM |
4120 | int slot; |
4121 | struct extent_buffer *c; | |
4122 | ||
934d375b | 4123 | WARN_ON(!path->keep_locks); |
d397712b | 4124 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4125 | if (!path->nodes[level]) |
4126 | return 1; | |
4127 | ||
4128 | slot = path->slots[level] + 1; | |
4129 | c = path->nodes[level]; | |
3f157a2f | 4130 | next: |
e7a84565 | 4131 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4132 | int ret; |
4133 | int orig_lowest; | |
4134 | struct btrfs_key cur_key; | |
4135 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4136 | !path->nodes[level + 1]) | |
e7a84565 | 4137 | return 1; |
33c66f43 YZ |
4138 | |
4139 | if (path->locks[level + 1]) { | |
4140 | level++; | |
4141 | continue; | |
4142 | } | |
4143 | ||
4144 | slot = btrfs_header_nritems(c) - 1; | |
4145 | if (level == 0) | |
4146 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4147 | else | |
4148 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4149 | ||
4150 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4151 | btrfs_release_path(path); |
33c66f43 YZ |
4152 | path->lowest_level = level; |
4153 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4154 | 0, 0); | |
4155 | path->lowest_level = orig_lowest; | |
4156 | if (ret < 0) | |
4157 | return ret; | |
4158 | ||
4159 | c = path->nodes[level]; | |
4160 | slot = path->slots[level]; | |
4161 | if (ret == 0) | |
4162 | slot++; | |
4163 | goto next; | |
e7a84565 | 4164 | } |
33c66f43 | 4165 | |
e7a84565 CM |
4166 | if (level == 0) |
4167 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
4168 | else { |
4169 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
4170 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
4171 | ||
4172 | if (cache_only) { | |
4173 | struct extent_buffer *cur; | |
4174 | cur = btrfs_find_tree_block(root, blockptr, | |
4175 | btrfs_level_size(root, level - 1)); | |
b9fab919 CM |
4176 | if (!cur || |
4177 | btrfs_buffer_uptodate(cur, gen, 1) <= 0) { | |
3f157a2f CM |
4178 | slot++; |
4179 | if (cur) | |
4180 | free_extent_buffer(cur); | |
4181 | goto next; | |
4182 | } | |
4183 | free_extent_buffer(cur); | |
4184 | } | |
4185 | if (gen < min_trans) { | |
4186 | slot++; | |
4187 | goto next; | |
4188 | } | |
e7a84565 | 4189 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4190 | } |
e7a84565 CM |
4191 | return 0; |
4192 | } | |
4193 | return 1; | |
4194 | } | |
4195 | ||
97571fd0 | 4196 | /* |
925baedd | 4197 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
4198 | * returns 0 if it found something or 1 if there are no greater leaves. |
4199 | * returns < 0 on io errors. | |
97571fd0 | 4200 | */ |
234b63a0 | 4201 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
4202 | { |
4203 | int slot; | |
8e73f275 | 4204 | int level; |
5f39d397 | 4205 | struct extent_buffer *c; |
8e73f275 | 4206 | struct extent_buffer *next; |
925baedd CM |
4207 | struct btrfs_key key; |
4208 | u32 nritems; | |
4209 | int ret; | |
8e73f275 | 4210 | int old_spinning = path->leave_spinning; |
bd681513 | 4211 | int next_rw_lock = 0; |
925baedd CM |
4212 | |
4213 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4214 | if (nritems == 0) |
925baedd | 4215 | return 1; |
925baedd | 4216 | |
8e73f275 CM |
4217 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4218 | again: | |
4219 | level = 1; | |
4220 | next = NULL; | |
bd681513 | 4221 | next_rw_lock = 0; |
b3b4aa74 | 4222 | btrfs_release_path(path); |
8e73f275 | 4223 | |
a2135011 | 4224 | path->keep_locks = 1; |
31533fb2 | 4225 | path->leave_spinning = 1; |
8e73f275 | 4226 | |
925baedd CM |
4227 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4228 | path->keep_locks = 0; | |
4229 | ||
4230 | if (ret < 0) | |
4231 | return ret; | |
4232 | ||
a2135011 | 4233 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4234 | /* |
4235 | * by releasing the path above we dropped all our locks. A balance | |
4236 | * could have added more items next to the key that used to be | |
4237 | * at the very end of the block. So, check again here and | |
4238 | * advance the path if there are now more items available. | |
4239 | */ | |
a2135011 | 4240 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4241 | if (ret == 0) |
4242 | path->slots[0]++; | |
8e73f275 | 4243 | ret = 0; |
925baedd CM |
4244 | goto done; |
4245 | } | |
d97e63b6 | 4246 | |
d397712b | 4247 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4248 | if (!path->nodes[level]) { |
4249 | ret = 1; | |
4250 | goto done; | |
4251 | } | |
5f39d397 | 4252 | |
d97e63b6 CM |
4253 | slot = path->slots[level] + 1; |
4254 | c = path->nodes[level]; | |
5f39d397 | 4255 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4256 | level++; |
8e73f275 CM |
4257 | if (level == BTRFS_MAX_LEVEL) { |
4258 | ret = 1; | |
4259 | goto done; | |
4260 | } | |
d97e63b6 CM |
4261 | continue; |
4262 | } | |
5f39d397 | 4263 | |
925baedd | 4264 | if (next) { |
bd681513 | 4265 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 4266 | free_extent_buffer(next); |
925baedd | 4267 | } |
5f39d397 | 4268 | |
8e73f275 | 4269 | next = c; |
bd681513 | 4270 | next_rw_lock = path->locks[level]; |
8e73f275 CM |
4271 | ret = read_block_for_search(NULL, root, path, &next, level, |
4272 | slot, &key); | |
4273 | if (ret == -EAGAIN) | |
4274 | goto again; | |
5f39d397 | 4275 | |
76a05b35 | 4276 | if (ret < 0) { |
b3b4aa74 | 4277 | btrfs_release_path(path); |
76a05b35 CM |
4278 | goto done; |
4279 | } | |
4280 | ||
5cd57b2c | 4281 | if (!path->skip_locking) { |
bd681513 | 4282 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4283 | if (!ret) { |
4284 | btrfs_set_path_blocking(path); | |
bd681513 | 4285 | btrfs_tree_read_lock(next); |
31533fb2 | 4286 | btrfs_clear_path_blocking(path, next, |
bd681513 | 4287 | BTRFS_READ_LOCK); |
8e73f275 | 4288 | } |
31533fb2 | 4289 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4290 | } |
d97e63b6 CM |
4291 | break; |
4292 | } | |
4293 | path->slots[level] = slot; | |
d397712b | 4294 | while (1) { |
d97e63b6 CM |
4295 | level--; |
4296 | c = path->nodes[level]; | |
925baedd | 4297 | if (path->locks[level]) |
bd681513 | 4298 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 4299 | |
5f39d397 | 4300 | free_extent_buffer(c); |
d97e63b6 CM |
4301 | path->nodes[level] = next; |
4302 | path->slots[level] = 0; | |
a74a4b97 | 4303 | if (!path->skip_locking) |
bd681513 | 4304 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
4305 | if (!level) |
4306 | break; | |
b4ce94de | 4307 | |
8e73f275 CM |
4308 | ret = read_block_for_search(NULL, root, path, &next, level, |
4309 | 0, &key); | |
4310 | if (ret == -EAGAIN) | |
4311 | goto again; | |
4312 | ||
76a05b35 | 4313 | if (ret < 0) { |
b3b4aa74 | 4314 | btrfs_release_path(path); |
76a05b35 CM |
4315 | goto done; |
4316 | } | |
4317 | ||
5cd57b2c | 4318 | if (!path->skip_locking) { |
bd681513 | 4319 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4320 | if (!ret) { |
4321 | btrfs_set_path_blocking(path); | |
bd681513 | 4322 | btrfs_tree_read_lock(next); |
31533fb2 | 4323 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
4324 | BTRFS_READ_LOCK); |
4325 | } | |
31533fb2 | 4326 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4327 | } |
d97e63b6 | 4328 | } |
8e73f275 | 4329 | ret = 0; |
925baedd | 4330 | done: |
f7c79f30 | 4331 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
4332 | path->leave_spinning = old_spinning; |
4333 | if (!old_spinning) | |
4334 | btrfs_set_path_blocking(path); | |
4335 | ||
4336 | return ret; | |
d97e63b6 | 4337 | } |
0b86a832 | 4338 | |
3f157a2f CM |
4339 | /* |
4340 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4341 | * searching until it gets past min_objectid or finds an item of 'type' | |
4342 | * | |
4343 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4344 | */ | |
0b86a832 CM |
4345 | int btrfs_previous_item(struct btrfs_root *root, |
4346 | struct btrfs_path *path, u64 min_objectid, | |
4347 | int type) | |
4348 | { | |
4349 | struct btrfs_key found_key; | |
4350 | struct extent_buffer *leaf; | |
e02119d5 | 4351 | u32 nritems; |
0b86a832 CM |
4352 | int ret; |
4353 | ||
d397712b | 4354 | while (1) { |
0b86a832 | 4355 | if (path->slots[0] == 0) { |
b4ce94de | 4356 | btrfs_set_path_blocking(path); |
0b86a832 CM |
4357 | ret = btrfs_prev_leaf(root, path); |
4358 | if (ret != 0) | |
4359 | return ret; | |
4360 | } else { | |
4361 | path->slots[0]--; | |
4362 | } | |
4363 | leaf = path->nodes[0]; | |
e02119d5 CM |
4364 | nritems = btrfs_header_nritems(leaf); |
4365 | if (nritems == 0) | |
4366 | return 1; | |
4367 | if (path->slots[0] == nritems) | |
4368 | path->slots[0]--; | |
4369 | ||
0b86a832 | 4370 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4371 | if (found_key.objectid < min_objectid) |
4372 | break; | |
0a4eefbb YZ |
4373 | if (found_key.type == type) |
4374 | return 0; | |
e02119d5 CM |
4375 | if (found_key.objectid == min_objectid && |
4376 | found_key.type < type) | |
4377 | break; | |
0b86a832 CM |
4378 | } |
4379 | return 1; | |
4380 | } |