]>
Commit | Line | Data |
---|---|---|
6cbd5570 | 1 | /* |
d352ac68 | 2 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
a6b6e75e | 19 | #include <linux/sched.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
eb60ceac CM |
21 | #include "ctree.h" |
22 | #include "disk-io.h" | |
7f5c1516 | 23 | #include "transaction.h" |
5f39d397 | 24 | #include "print-tree.h" |
925baedd | 25 | #include "locking.h" |
9a8dd150 | 26 | |
e089f05c CM |
27 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
28 | *root, struct btrfs_path *path, int level); | |
29 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 30 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 31 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
32 | static int push_node_left(struct btrfs_trans_handle *trans, |
33 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 34 | struct extent_buffer *src, int empty); |
5f39d397 CM |
35 | static int balance_node_right(struct btrfs_trans_handle *trans, |
36 | struct btrfs_root *root, | |
37 | struct extent_buffer *dst_buf, | |
38 | struct extent_buffer *src_buf); | |
143bede5 | 39 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
e089f05c | 40 | struct btrfs_path *path, int level, int slot); |
d97e63b6 | 41 | |
df24a2b9 | 42 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 43 | { |
df24a2b9 | 44 | struct btrfs_path *path; |
e00f7308 | 45 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 46 | return path; |
2c90e5d6 CM |
47 | } |
48 | ||
b4ce94de CM |
49 | /* |
50 | * set all locked nodes in the path to blocking locks. This should | |
51 | * be done before scheduling | |
52 | */ | |
53 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
54 | { | |
55 | int i; | |
56 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
57 | if (!p->nodes[i] || !p->locks[i]) |
58 | continue; | |
59 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
60 | if (p->locks[i] == BTRFS_READ_LOCK) | |
61 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
62 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
63 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
64 | } |
65 | } | |
66 | ||
67 | /* | |
68 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
69 | * |
70 | * held is used to keep lockdep happy, when lockdep is enabled | |
71 | * we set held to a blocking lock before we go around and | |
72 | * retake all the spinlocks in the path. You can safely use NULL | |
73 | * for held | |
b4ce94de | 74 | */ |
4008c04a | 75 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 76 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
77 | { |
78 | int i; | |
4008c04a CM |
79 | |
80 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
81 | /* lockdep really cares that we take all of these spinlocks | |
82 | * in the right order. If any of the locks in the path are not | |
83 | * currently blocking, it is going to complain. So, make really | |
84 | * really sure by forcing the path to blocking before we clear | |
85 | * the path blocking. | |
86 | */ | |
bd681513 CM |
87 | if (held) { |
88 | btrfs_set_lock_blocking_rw(held, held_rw); | |
89 | if (held_rw == BTRFS_WRITE_LOCK) | |
90 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
91 | else if (held_rw == BTRFS_READ_LOCK) | |
92 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
93 | } | |
4008c04a CM |
94 | btrfs_set_path_blocking(p); |
95 | #endif | |
96 | ||
97 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
98 | if (p->nodes[i] && p->locks[i]) { |
99 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
100 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
101 | p->locks[i] = BTRFS_WRITE_LOCK; | |
102 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
103 | p->locks[i] = BTRFS_READ_LOCK; | |
104 | } | |
b4ce94de | 105 | } |
4008c04a CM |
106 | |
107 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
108 | if (held) | |
bd681513 | 109 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 110 | #endif |
b4ce94de CM |
111 | } |
112 | ||
d352ac68 | 113 | /* this also releases the path */ |
df24a2b9 | 114 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 115 | { |
ff175d57 JJ |
116 | if (!p) |
117 | return; | |
b3b4aa74 | 118 | btrfs_release_path(p); |
df24a2b9 | 119 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
120 | } |
121 | ||
d352ac68 CM |
122 | /* |
123 | * path release drops references on the extent buffers in the path | |
124 | * and it drops any locks held by this path | |
125 | * | |
126 | * It is safe to call this on paths that no locks or extent buffers held. | |
127 | */ | |
b3b4aa74 | 128 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
129 | { |
130 | int i; | |
a2135011 | 131 | |
234b63a0 | 132 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 133 | p->slots[i] = 0; |
eb60ceac | 134 | if (!p->nodes[i]) |
925baedd CM |
135 | continue; |
136 | if (p->locks[i]) { | |
bd681513 | 137 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
138 | p->locks[i] = 0; |
139 | } | |
5f39d397 | 140 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 141 | p->nodes[i] = NULL; |
eb60ceac CM |
142 | } |
143 | } | |
144 | ||
d352ac68 CM |
145 | /* |
146 | * safely gets a reference on the root node of a tree. A lock | |
147 | * is not taken, so a concurrent writer may put a different node | |
148 | * at the root of the tree. See btrfs_lock_root_node for the | |
149 | * looping required. | |
150 | * | |
151 | * The extent buffer returned by this has a reference taken, so | |
152 | * it won't disappear. It may stop being the root of the tree | |
153 | * at any time because there are no locks held. | |
154 | */ | |
925baedd CM |
155 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
156 | { | |
157 | struct extent_buffer *eb; | |
240f62c8 CM |
158 | |
159 | rcu_read_lock(); | |
160 | eb = rcu_dereference(root->node); | |
925baedd | 161 | extent_buffer_get(eb); |
240f62c8 | 162 | rcu_read_unlock(); |
925baedd CM |
163 | return eb; |
164 | } | |
165 | ||
d352ac68 CM |
166 | /* loop around taking references on and locking the root node of the |
167 | * tree until you end up with a lock on the root. A locked buffer | |
168 | * is returned, with a reference held. | |
169 | */ | |
925baedd CM |
170 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
171 | { | |
172 | struct extent_buffer *eb; | |
173 | ||
d397712b | 174 | while (1) { |
925baedd CM |
175 | eb = btrfs_root_node(root); |
176 | btrfs_tree_lock(eb); | |
240f62c8 | 177 | if (eb == root->node) |
925baedd | 178 | break; |
925baedd CM |
179 | btrfs_tree_unlock(eb); |
180 | free_extent_buffer(eb); | |
181 | } | |
182 | return eb; | |
183 | } | |
184 | ||
bd681513 CM |
185 | /* loop around taking references on and locking the root node of the |
186 | * tree until you end up with a lock on the root. A locked buffer | |
187 | * is returned, with a reference held. | |
188 | */ | |
189 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) | |
190 | { | |
191 | struct extent_buffer *eb; | |
192 | ||
193 | while (1) { | |
194 | eb = btrfs_root_node(root); | |
195 | btrfs_tree_read_lock(eb); | |
196 | if (eb == root->node) | |
197 | break; | |
198 | btrfs_tree_read_unlock(eb); | |
199 | free_extent_buffer(eb); | |
200 | } | |
201 | return eb; | |
202 | } | |
203 | ||
d352ac68 CM |
204 | /* cowonly root (everything not a reference counted cow subvolume), just get |
205 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
206 | * get properly updated on disk. | |
207 | */ | |
0b86a832 CM |
208 | static void add_root_to_dirty_list(struct btrfs_root *root) |
209 | { | |
210 | if (root->track_dirty && list_empty(&root->dirty_list)) { | |
211 | list_add(&root->dirty_list, | |
212 | &root->fs_info->dirty_cowonly_roots); | |
213 | } | |
214 | } | |
215 | ||
d352ac68 CM |
216 | /* |
217 | * used by snapshot creation to make a copy of a root for a tree with | |
218 | * a given objectid. The buffer with the new root node is returned in | |
219 | * cow_ret, and this func returns zero on success or a negative error code. | |
220 | */ | |
be20aa9d CM |
221 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
222 | struct btrfs_root *root, | |
223 | struct extent_buffer *buf, | |
224 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
225 | { | |
226 | struct extent_buffer *cow; | |
be20aa9d CM |
227 | int ret = 0; |
228 | int level; | |
5d4f98a2 | 229 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
230 | |
231 | WARN_ON(root->ref_cows && trans->transid != | |
232 | root->fs_info->running_transaction->transid); | |
233 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
234 | ||
235 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
236 | if (level == 0) |
237 | btrfs_item_key(buf, &disk_key, 0); | |
238 | else | |
239 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 240 | |
5d4f98a2 YZ |
241 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
242 | new_root_objectid, &disk_key, level, | |
66d7e7f0 | 243 | buf->start, 0, 1); |
5d4f98a2 | 244 | if (IS_ERR(cow)) |
be20aa9d CM |
245 | return PTR_ERR(cow); |
246 | ||
247 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
248 | btrfs_set_header_bytenr(cow, cow->start); | |
249 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
250 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
251 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
252 | BTRFS_HEADER_FLAG_RELOC); | |
253 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
254 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
255 | else | |
256 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 257 | |
2b82032c YZ |
258 | write_extent_buffer(cow, root->fs_info->fsid, |
259 | (unsigned long)btrfs_header_fsid(cow), | |
260 | BTRFS_FSID_SIZE); | |
261 | ||
be20aa9d | 262 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 263 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 264 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 265 | else |
66d7e7f0 | 266 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 267 | |
be20aa9d CM |
268 | if (ret) |
269 | return ret; | |
270 | ||
271 | btrfs_mark_buffer_dirty(cow); | |
272 | *cow_ret = cow; | |
273 | return 0; | |
274 | } | |
275 | ||
5d4f98a2 YZ |
276 | /* |
277 | * check if the tree block can be shared by multiple trees | |
278 | */ | |
279 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
280 | struct extent_buffer *buf) | |
281 | { | |
282 | /* | |
283 | * Tree blocks not in refernece counted trees and tree roots | |
284 | * are never shared. If a block was allocated after the last | |
285 | * snapshot and the block was not allocated by tree relocation, | |
286 | * we know the block is not shared. | |
287 | */ | |
288 | if (root->ref_cows && | |
289 | buf != root->node && buf != root->commit_root && | |
290 | (btrfs_header_generation(buf) <= | |
291 | btrfs_root_last_snapshot(&root->root_item) || | |
292 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
293 | return 1; | |
294 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
295 | if (root->ref_cows && | |
296 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
297 | return 1; | |
298 | #endif | |
299 | return 0; | |
300 | } | |
301 | ||
302 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
303 | struct btrfs_root *root, | |
304 | struct extent_buffer *buf, | |
f0486c68 YZ |
305 | struct extent_buffer *cow, |
306 | int *last_ref) | |
5d4f98a2 YZ |
307 | { |
308 | u64 refs; | |
309 | u64 owner; | |
310 | u64 flags; | |
311 | u64 new_flags = 0; | |
312 | int ret; | |
313 | ||
314 | /* | |
315 | * Backrefs update rules: | |
316 | * | |
317 | * Always use full backrefs for extent pointers in tree block | |
318 | * allocated by tree relocation. | |
319 | * | |
320 | * If a shared tree block is no longer referenced by its owner | |
321 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
322 | * use full backrefs for extent pointers in tree block. | |
323 | * | |
324 | * If a tree block is been relocating | |
325 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
326 | * use full backrefs for extent pointers in tree block. | |
327 | * The reason for this is some operations (such as drop tree) | |
328 | * are only allowed for blocks use full backrefs. | |
329 | */ | |
330 | ||
331 | if (btrfs_block_can_be_shared(root, buf)) { | |
332 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
333 | buf->len, &refs, &flags); | |
be1a5564 MF |
334 | if (ret) |
335 | return ret; | |
e5df9573 MF |
336 | if (refs == 0) { |
337 | ret = -EROFS; | |
338 | btrfs_std_error(root->fs_info, ret); | |
339 | return ret; | |
340 | } | |
5d4f98a2 YZ |
341 | } else { |
342 | refs = 1; | |
343 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
344 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
345 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
346 | else | |
347 | flags = 0; | |
348 | } | |
349 | ||
350 | owner = btrfs_header_owner(buf); | |
351 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
352 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
353 | ||
354 | if (refs > 1) { | |
355 | if ((owner == root->root_key.objectid || | |
356 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
357 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 358 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
5d4f98a2 YZ |
359 | BUG_ON(ret); |
360 | ||
361 | if (root->root_key.objectid == | |
362 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 363 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
5d4f98a2 | 364 | BUG_ON(ret); |
66d7e7f0 | 365 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 YZ |
366 | BUG_ON(ret); |
367 | } | |
368 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
369 | } else { | |
370 | ||
371 | if (root->root_key.objectid == | |
372 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 373 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 374 | else |
66d7e7f0 | 375 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
5d4f98a2 YZ |
376 | BUG_ON(ret); |
377 | } | |
378 | if (new_flags != 0) { | |
379 | ret = btrfs_set_disk_extent_flags(trans, root, | |
380 | buf->start, | |
381 | buf->len, | |
382 | new_flags, 0); | |
be1a5564 MF |
383 | if (ret) |
384 | return ret; | |
5d4f98a2 YZ |
385 | } |
386 | } else { | |
387 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
388 | if (root->root_key.objectid == | |
389 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 390 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 391 | else |
66d7e7f0 | 392 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
5d4f98a2 | 393 | BUG_ON(ret); |
66d7e7f0 | 394 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
5d4f98a2 YZ |
395 | BUG_ON(ret); |
396 | } | |
397 | clean_tree_block(trans, root, buf); | |
f0486c68 | 398 | *last_ref = 1; |
5d4f98a2 YZ |
399 | } |
400 | return 0; | |
401 | } | |
402 | ||
d352ac68 | 403 | /* |
d397712b CM |
404 | * does the dirty work in cow of a single block. The parent block (if |
405 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
406 | * dirty and returned locked. If you modify the block it needs to be marked | |
407 | * dirty again. | |
d352ac68 CM |
408 | * |
409 | * search_start -- an allocation hint for the new block | |
410 | * | |
d397712b CM |
411 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
412 | * bytes the allocator should try to find free next to the block it returns. | |
413 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 414 | */ |
d397712b | 415 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
416 | struct btrfs_root *root, |
417 | struct extent_buffer *buf, | |
418 | struct extent_buffer *parent, int parent_slot, | |
419 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 420 | u64 search_start, u64 empty_size) |
02217ed2 | 421 | { |
5d4f98a2 | 422 | struct btrfs_disk_key disk_key; |
5f39d397 | 423 | struct extent_buffer *cow; |
be1a5564 | 424 | int level, ret; |
f0486c68 | 425 | int last_ref = 0; |
925baedd | 426 | int unlock_orig = 0; |
5d4f98a2 | 427 | u64 parent_start; |
7bb86316 | 428 | |
925baedd CM |
429 | if (*cow_ret == buf) |
430 | unlock_orig = 1; | |
431 | ||
b9447ef8 | 432 | btrfs_assert_tree_locked(buf); |
925baedd | 433 | |
7bb86316 CM |
434 | WARN_ON(root->ref_cows && trans->transid != |
435 | root->fs_info->running_transaction->transid); | |
6702ed49 | 436 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 437 | |
7bb86316 | 438 | level = btrfs_header_level(buf); |
31840ae1 | 439 | |
5d4f98a2 YZ |
440 | if (level == 0) |
441 | btrfs_item_key(buf, &disk_key, 0); | |
442 | else | |
443 | btrfs_node_key(buf, &disk_key, 0); | |
444 | ||
445 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
446 | if (parent) | |
447 | parent_start = parent->start; | |
448 | else | |
449 | parent_start = 0; | |
450 | } else | |
451 | parent_start = 0; | |
452 | ||
453 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
454 | root->root_key.objectid, &disk_key, | |
66d7e7f0 | 455 | level, search_start, empty_size, 1); |
54aa1f4d CM |
456 | if (IS_ERR(cow)) |
457 | return PTR_ERR(cow); | |
6702ed49 | 458 | |
b4ce94de CM |
459 | /* cow is set to blocking by btrfs_init_new_buffer */ |
460 | ||
5f39d397 | 461 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 462 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 463 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
464 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
465 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
466 | BTRFS_HEADER_FLAG_RELOC); | |
467 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
468 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
469 | else | |
470 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 471 | |
2b82032c YZ |
472 | write_extent_buffer(cow, root->fs_info->fsid, |
473 | (unsigned long)btrfs_header_fsid(cow), | |
474 | BTRFS_FSID_SIZE); | |
475 | ||
be1a5564 MF |
476 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
477 | BUG_ON(ret); | |
1a40e23b | 478 | |
3fd0a558 YZ |
479 | if (root->ref_cows) |
480 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
481 | ||
02217ed2 | 482 | if (buf == root->node) { |
925baedd | 483 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
484 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
485 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
486 | parent_start = buf->start; | |
487 | else | |
488 | parent_start = 0; | |
925baedd | 489 | |
5f39d397 | 490 | extent_buffer_get(cow); |
240f62c8 | 491 | rcu_assign_pointer(root->node, cow); |
925baedd | 492 | |
f0486c68 | 493 | btrfs_free_tree_block(trans, root, buf, parent_start, |
66d7e7f0 | 494 | last_ref, 1); |
5f39d397 | 495 | free_extent_buffer(buf); |
0b86a832 | 496 | add_root_to_dirty_list(root); |
02217ed2 | 497 | } else { |
5d4f98a2 YZ |
498 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
499 | parent_start = parent->start; | |
500 | else | |
501 | parent_start = 0; | |
502 | ||
503 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
5f39d397 | 504 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 505 | cow->start); |
74493f7a CM |
506 | btrfs_set_node_ptr_generation(parent, parent_slot, |
507 | trans->transid); | |
d6025579 | 508 | btrfs_mark_buffer_dirty(parent); |
f0486c68 | 509 | btrfs_free_tree_block(trans, root, buf, parent_start, |
66d7e7f0 | 510 | last_ref, 1); |
02217ed2 | 511 | } |
925baedd CM |
512 | if (unlock_orig) |
513 | btrfs_tree_unlock(buf); | |
5f39d397 | 514 | free_extent_buffer(buf); |
ccd467d6 | 515 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 516 | *cow_ret = cow; |
02217ed2 CM |
517 | return 0; |
518 | } | |
519 | ||
5d4f98a2 YZ |
520 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
521 | struct btrfs_root *root, | |
522 | struct extent_buffer *buf) | |
523 | { | |
f1ebcc74 LB |
524 | /* ensure we can see the force_cow */ |
525 | smp_rmb(); | |
526 | ||
527 | /* | |
528 | * We do not need to cow a block if | |
529 | * 1) this block is not created or changed in this transaction; | |
530 | * 2) this block does not belong to TREE_RELOC tree; | |
531 | * 3) the root is not forced COW. | |
532 | * | |
533 | * What is forced COW: | |
534 | * when we create snapshot during commiting the transaction, | |
535 | * after we've finished coping src root, we must COW the shared | |
536 | * block to ensure the metadata consistency. | |
537 | */ | |
5d4f98a2 YZ |
538 | if (btrfs_header_generation(buf) == trans->transid && |
539 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
540 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
541 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
542 | !root->force_cow) | |
5d4f98a2 YZ |
543 | return 0; |
544 | return 1; | |
545 | } | |
546 | ||
d352ac68 CM |
547 | /* |
548 | * cows a single block, see __btrfs_cow_block for the real work. | |
549 | * This version of it has extra checks so that a block isn't cow'd more than | |
550 | * once per transaction, as long as it hasn't been written yet | |
551 | */ | |
d397712b | 552 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
553 | struct btrfs_root *root, struct extent_buffer *buf, |
554 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 555 | struct extent_buffer **cow_ret) |
6702ed49 CM |
556 | { |
557 | u64 search_start; | |
f510cfec | 558 | int ret; |
dc17ff8f | 559 | |
6702ed49 | 560 | if (trans->transaction != root->fs_info->running_transaction) { |
d397712b CM |
561 | printk(KERN_CRIT "trans %llu running %llu\n", |
562 | (unsigned long long)trans->transid, | |
563 | (unsigned long long) | |
6702ed49 CM |
564 | root->fs_info->running_transaction->transid); |
565 | WARN_ON(1); | |
566 | } | |
567 | if (trans->transid != root->fs_info->generation) { | |
d397712b CM |
568 | printk(KERN_CRIT "trans %llu running %llu\n", |
569 | (unsigned long long)trans->transid, | |
570 | (unsigned long long)root->fs_info->generation); | |
6702ed49 CM |
571 | WARN_ON(1); |
572 | } | |
dc17ff8f | 573 | |
5d4f98a2 | 574 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
575 | *cow_ret = buf; |
576 | return 0; | |
577 | } | |
c487685d | 578 | |
0b86a832 | 579 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
580 | |
581 | if (parent) | |
582 | btrfs_set_lock_blocking(parent); | |
583 | btrfs_set_lock_blocking(buf); | |
584 | ||
f510cfec | 585 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 586 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 587 | |
588 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
589 | ||
f510cfec | 590 | return ret; |
6702ed49 CM |
591 | } |
592 | ||
d352ac68 CM |
593 | /* |
594 | * helper function for defrag to decide if two blocks pointed to by a | |
595 | * node are actually close by | |
596 | */ | |
6b80053d | 597 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 598 | { |
6b80053d | 599 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 600 | return 1; |
6b80053d | 601 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
602 | return 1; |
603 | return 0; | |
604 | } | |
605 | ||
081e9573 CM |
606 | /* |
607 | * compare two keys in a memcmp fashion | |
608 | */ | |
609 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
610 | { | |
611 | struct btrfs_key k1; | |
612 | ||
613 | btrfs_disk_key_to_cpu(&k1, disk); | |
614 | ||
20736aba | 615 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
616 | } |
617 | ||
f3465ca4 JB |
618 | /* |
619 | * same as comp_keys only with two btrfs_key's | |
620 | */ | |
5d4f98a2 | 621 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
622 | { |
623 | if (k1->objectid > k2->objectid) | |
624 | return 1; | |
625 | if (k1->objectid < k2->objectid) | |
626 | return -1; | |
627 | if (k1->type > k2->type) | |
628 | return 1; | |
629 | if (k1->type < k2->type) | |
630 | return -1; | |
631 | if (k1->offset > k2->offset) | |
632 | return 1; | |
633 | if (k1->offset < k2->offset) | |
634 | return -1; | |
635 | return 0; | |
636 | } | |
081e9573 | 637 | |
d352ac68 CM |
638 | /* |
639 | * this is used by the defrag code to go through all the | |
640 | * leaves pointed to by a node and reallocate them so that | |
641 | * disk order is close to key order | |
642 | */ | |
6702ed49 | 643 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 644 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
645 | int start_slot, int cache_only, u64 *last_ret, |
646 | struct btrfs_key *progress) | |
6702ed49 | 647 | { |
6b80053d | 648 | struct extent_buffer *cur; |
6702ed49 | 649 | u64 blocknr; |
ca7a79ad | 650 | u64 gen; |
e9d0b13b CM |
651 | u64 search_start = *last_ret; |
652 | u64 last_block = 0; | |
6702ed49 CM |
653 | u64 other; |
654 | u32 parent_nritems; | |
6702ed49 CM |
655 | int end_slot; |
656 | int i; | |
657 | int err = 0; | |
f2183bde | 658 | int parent_level; |
6b80053d CM |
659 | int uptodate; |
660 | u32 blocksize; | |
081e9573 CM |
661 | int progress_passed = 0; |
662 | struct btrfs_disk_key disk_key; | |
6702ed49 | 663 | |
5708b959 CM |
664 | parent_level = btrfs_header_level(parent); |
665 | if (cache_only && parent_level != 1) | |
666 | return 0; | |
667 | ||
d397712b | 668 | if (trans->transaction != root->fs_info->running_transaction) |
6702ed49 | 669 | WARN_ON(1); |
d397712b | 670 | if (trans->transid != root->fs_info->generation) |
6702ed49 | 671 | WARN_ON(1); |
86479a04 | 672 | |
6b80053d | 673 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 674 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
675 | end_slot = parent_nritems; |
676 | ||
677 | if (parent_nritems == 1) | |
678 | return 0; | |
679 | ||
b4ce94de CM |
680 | btrfs_set_lock_blocking(parent); |
681 | ||
6702ed49 CM |
682 | for (i = start_slot; i < end_slot; i++) { |
683 | int close = 1; | |
a6b6e75e | 684 | |
081e9573 CM |
685 | btrfs_node_key(parent, &disk_key, i); |
686 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
687 | continue; | |
688 | ||
689 | progress_passed = 1; | |
6b80053d | 690 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 691 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
692 | if (last_block == 0) |
693 | last_block = blocknr; | |
5708b959 | 694 | |
6702ed49 | 695 | if (i > 0) { |
6b80053d CM |
696 | other = btrfs_node_blockptr(parent, i - 1); |
697 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 698 | } |
0ef3e66b | 699 | if (!close && i < end_slot - 2) { |
6b80053d CM |
700 | other = btrfs_node_blockptr(parent, i + 1); |
701 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 702 | } |
e9d0b13b CM |
703 | if (close) { |
704 | last_block = blocknr; | |
6702ed49 | 705 | continue; |
e9d0b13b | 706 | } |
6702ed49 | 707 | |
6b80053d CM |
708 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
709 | if (cur) | |
1259ab75 | 710 | uptodate = btrfs_buffer_uptodate(cur, gen); |
6b80053d CM |
711 | else |
712 | uptodate = 0; | |
5708b959 | 713 | if (!cur || !uptodate) { |
6702ed49 | 714 | if (cache_only) { |
6b80053d | 715 | free_extent_buffer(cur); |
6702ed49 CM |
716 | continue; |
717 | } | |
6b80053d CM |
718 | if (!cur) { |
719 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 720 | blocksize, gen); |
97d9a8a4 TI |
721 | if (!cur) |
722 | return -EIO; | |
6b80053d | 723 | } else if (!uptodate) { |
ca7a79ad | 724 | btrfs_read_buffer(cur, gen); |
f2183bde | 725 | } |
6702ed49 | 726 | } |
e9d0b13b | 727 | if (search_start == 0) |
6b80053d | 728 | search_start = last_block; |
e9d0b13b | 729 | |
e7a84565 | 730 | btrfs_tree_lock(cur); |
b4ce94de | 731 | btrfs_set_lock_blocking(cur); |
6b80053d | 732 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 733 | &cur, search_start, |
6b80053d | 734 | min(16 * blocksize, |
9fa8cfe7 | 735 | (end_slot - i) * blocksize)); |
252c38f0 | 736 | if (err) { |
e7a84565 | 737 | btrfs_tree_unlock(cur); |
6b80053d | 738 | free_extent_buffer(cur); |
6702ed49 | 739 | break; |
252c38f0 | 740 | } |
e7a84565 CM |
741 | search_start = cur->start; |
742 | last_block = cur->start; | |
f2183bde | 743 | *last_ret = search_start; |
e7a84565 CM |
744 | btrfs_tree_unlock(cur); |
745 | free_extent_buffer(cur); | |
6702ed49 CM |
746 | } |
747 | return err; | |
748 | } | |
749 | ||
74123bd7 CM |
750 | /* |
751 | * The leaf data grows from end-to-front in the node. | |
752 | * this returns the address of the start of the last item, | |
753 | * which is the stop of the leaf data stack | |
754 | */ | |
123abc88 | 755 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 756 | struct extent_buffer *leaf) |
be0e5c09 | 757 | { |
5f39d397 | 758 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 759 | if (nr == 0) |
123abc88 | 760 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 761 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
762 | } |
763 | ||
aa5d6bed | 764 | |
74123bd7 | 765 | /* |
5f39d397 CM |
766 | * search for key in the extent_buffer. The items start at offset p, |
767 | * and they are item_size apart. There are 'max' items in p. | |
768 | * | |
74123bd7 CM |
769 | * the slot in the array is returned via slot, and it points to |
770 | * the place where you would insert key if it is not found in | |
771 | * the array. | |
772 | * | |
773 | * slot may point to max if the key is bigger than all of the keys | |
774 | */ | |
e02119d5 CM |
775 | static noinline int generic_bin_search(struct extent_buffer *eb, |
776 | unsigned long p, | |
777 | int item_size, struct btrfs_key *key, | |
778 | int max, int *slot) | |
be0e5c09 CM |
779 | { |
780 | int low = 0; | |
781 | int high = max; | |
782 | int mid; | |
783 | int ret; | |
479965d6 | 784 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
785 | struct btrfs_disk_key unaligned; |
786 | unsigned long offset; | |
5f39d397 CM |
787 | char *kaddr = NULL; |
788 | unsigned long map_start = 0; | |
789 | unsigned long map_len = 0; | |
479965d6 | 790 | int err; |
be0e5c09 | 791 | |
d397712b | 792 | while (low < high) { |
be0e5c09 | 793 | mid = (low + high) / 2; |
5f39d397 CM |
794 | offset = p + mid * item_size; |
795 | ||
a6591715 | 796 | if (!kaddr || offset < map_start || |
5f39d397 CM |
797 | (offset + sizeof(struct btrfs_disk_key)) > |
798 | map_start + map_len) { | |
934d375b CM |
799 | |
800 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 801 | sizeof(struct btrfs_disk_key), |
a6591715 | 802 | &kaddr, &map_start, &map_len); |
479965d6 CM |
803 | |
804 | if (!err) { | |
805 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
806 | map_start); | |
807 | } else { | |
808 | read_extent_buffer(eb, &unaligned, | |
809 | offset, sizeof(unaligned)); | |
810 | tmp = &unaligned; | |
811 | } | |
5f39d397 | 812 | |
5f39d397 CM |
813 | } else { |
814 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
815 | map_start); | |
816 | } | |
be0e5c09 CM |
817 | ret = comp_keys(tmp, key); |
818 | ||
819 | if (ret < 0) | |
820 | low = mid + 1; | |
821 | else if (ret > 0) | |
822 | high = mid; | |
823 | else { | |
824 | *slot = mid; | |
825 | return 0; | |
826 | } | |
827 | } | |
828 | *slot = low; | |
829 | return 1; | |
830 | } | |
831 | ||
97571fd0 CM |
832 | /* |
833 | * simple bin_search frontend that does the right thing for | |
834 | * leaves vs nodes | |
835 | */ | |
5f39d397 CM |
836 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
837 | int level, int *slot) | |
be0e5c09 | 838 | { |
5f39d397 CM |
839 | if (level == 0) { |
840 | return generic_bin_search(eb, | |
841 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 842 | sizeof(struct btrfs_item), |
5f39d397 | 843 | key, btrfs_header_nritems(eb), |
7518a238 | 844 | slot); |
be0e5c09 | 845 | } else { |
5f39d397 CM |
846 | return generic_bin_search(eb, |
847 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 848 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 849 | key, btrfs_header_nritems(eb), |
7518a238 | 850 | slot); |
be0e5c09 CM |
851 | } |
852 | return -1; | |
853 | } | |
854 | ||
5d4f98a2 YZ |
855 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
856 | int level, int *slot) | |
857 | { | |
858 | return bin_search(eb, key, level, slot); | |
859 | } | |
860 | ||
f0486c68 YZ |
861 | static void root_add_used(struct btrfs_root *root, u32 size) |
862 | { | |
863 | spin_lock(&root->accounting_lock); | |
864 | btrfs_set_root_used(&root->root_item, | |
865 | btrfs_root_used(&root->root_item) + size); | |
866 | spin_unlock(&root->accounting_lock); | |
867 | } | |
868 | ||
869 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
870 | { | |
871 | spin_lock(&root->accounting_lock); | |
872 | btrfs_set_root_used(&root->root_item, | |
873 | btrfs_root_used(&root->root_item) - size); | |
874 | spin_unlock(&root->accounting_lock); | |
875 | } | |
876 | ||
d352ac68 CM |
877 | /* given a node and slot number, this reads the blocks it points to. The |
878 | * extent buffer is returned with a reference taken (but unlocked). | |
879 | * NULL is returned on error. | |
880 | */ | |
e02119d5 | 881 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 882 | struct extent_buffer *parent, int slot) |
bb803951 | 883 | { |
ca7a79ad | 884 | int level = btrfs_header_level(parent); |
bb803951 CM |
885 | if (slot < 0) |
886 | return NULL; | |
5f39d397 | 887 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 888 | return NULL; |
ca7a79ad CM |
889 | |
890 | BUG_ON(level == 0); | |
891 | ||
db94535d | 892 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
893 | btrfs_level_size(root, level - 1), |
894 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
895 | } |
896 | ||
d352ac68 CM |
897 | /* |
898 | * node level balancing, used to make sure nodes are in proper order for | |
899 | * item deletion. We balance from the top down, so we have to make sure | |
900 | * that a deletion won't leave an node completely empty later on. | |
901 | */ | |
e02119d5 | 902 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
903 | struct btrfs_root *root, |
904 | struct btrfs_path *path, int level) | |
bb803951 | 905 | { |
5f39d397 CM |
906 | struct extent_buffer *right = NULL; |
907 | struct extent_buffer *mid; | |
908 | struct extent_buffer *left = NULL; | |
909 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
910 | int ret = 0; |
911 | int wret; | |
912 | int pslot; | |
bb803951 | 913 | int orig_slot = path->slots[level]; |
79f95c82 | 914 | u64 orig_ptr; |
bb803951 CM |
915 | |
916 | if (level == 0) | |
917 | return 0; | |
918 | ||
5f39d397 | 919 | mid = path->nodes[level]; |
b4ce94de | 920 | |
bd681513 CM |
921 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
922 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
923 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
924 | ||
1d4f8a0c | 925 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 926 | |
a05a9bb1 | 927 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 928 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
929 | pslot = path->slots[level + 1]; |
930 | } | |
bb803951 | 931 | |
40689478 CM |
932 | /* |
933 | * deal with the case where there is only one pointer in the root | |
934 | * by promoting the node below to a root | |
935 | */ | |
5f39d397 CM |
936 | if (!parent) { |
937 | struct extent_buffer *child; | |
bb803951 | 938 | |
5f39d397 | 939 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
940 | return 0; |
941 | ||
942 | /* promote the child to a root */ | |
5f39d397 | 943 | child = read_node_slot(root, mid, 0); |
7951f3ce | 944 | BUG_ON(!child); |
925baedd | 945 | btrfs_tree_lock(child); |
b4ce94de | 946 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 947 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
948 | if (ret) { |
949 | btrfs_tree_unlock(child); | |
950 | free_extent_buffer(child); | |
951 | goto enospc; | |
952 | } | |
2f375ab9 | 953 | |
240f62c8 | 954 | rcu_assign_pointer(root->node, child); |
925baedd | 955 | |
0b86a832 | 956 | add_root_to_dirty_list(root); |
925baedd | 957 | btrfs_tree_unlock(child); |
b4ce94de | 958 | |
925baedd | 959 | path->locks[level] = 0; |
bb803951 | 960 | path->nodes[level] = NULL; |
5f39d397 | 961 | clean_tree_block(trans, root, mid); |
925baedd | 962 | btrfs_tree_unlock(mid); |
bb803951 | 963 | /* once for the path */ |
5f39d397 | 964 | free_extent_buffer(mid); |
f0486c68 YZ |
965 | |
966 | root_sub_used(root, mid->len); | |
66d7e7f0 | 967 | btrfs_free_tree_block(trans, root, mid, 0, 1, 0); |
bb803951 | 968 | /* once for the root ptr */ |
5f39d397 | 969 | free_extent_buffer(mid); |
f0486c68 | 970 | return 0; |
bb803951 | 971 | } |
5f39d397 | 972 | if (btrfs_header_nritems(mid) > |
123abc88 | 973 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
974 | return 0; |
975 | ||
559af821 | 976 | btrfs_header_nritems(mid); |
54aa1f4d | 977 | |
5f39d397 CM |
978 | left = read_node_slot(root, parent, pslot - 1); |
979 | if (left) { | |
925baedd | 980 | btrfs_tree_lock(left); |
b4ce94de | 981 | btrfs_set_lock_blocking(left); |
5f39d397 | 982 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 983 | parent, pslot - 1, &left); |
54aa1f4d CM |
984 | if (wret) { |
985 | ret = wret; | |
986 | goto enospc; | |
987 | } | |
2cc58cf2 | 988 | } |
5f39d397 CM |
989 | right = read_node_slot(root, parent, pslot + 1); |
990 | if (right) { | |
925baedd | 991 | btrfs_tree_lock(right); |
b4ce94de | 992 | btrfs_set_lock_blocking(right); |
5f39d397 | 993 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 994 | parent, pslot + 1, &right); |
2cc58cf2 CM |
995 | if (wret) { |
996 | ret = wret; | |
997 | goto enospc; | |
998 | } | |
999 | } | |
1000 | ||
1001 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1002 | if (left) { |
1003 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1004 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1005 | if (wret < 0) |
1006 | ret = wret; | |
559af821 | 1007 | btrfs_header_nritems(mid); |
bb803951 | 1008 | } |
79f95c82 CM |
1009 | |
1010 | /* | |
1011 | * then try to empty the right most buffer into the middle | |
1012 | */ | |
5f39d397 | 1013 | if (right) { |
971a1f66 | 1014 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1015 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1016 | ret = wret; |
5f39d397 | 1017 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1018 | clean_tree_block(trans, root, right); |
925baedd | 1019 | btrfs_tree_unlock(right); |
143bede5 | 1020 | del_ptr(trans, root, path, level + 1, pslot + 1); |
f0486c68 | 1021 | root_sub_used(root, right->len); |
66d7e7f0 | 1022 | btrfs_free_tree_block(trans, root, right, 0, 1, 0); |
f0486c68 YZ |
1023 | free_extent_buffer(right); |
1024 | right = NULL; | |
bb803951 | 1025 | } else { |
5f39d397 CM |
1026 | struct btrfs_disk_key right_key; |
1027 | btrfs_node_key(right, &right_key, 0); | |
1028 | btrfs_set_node_key(parent, &right_key, pslot + 1); | |
1029 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1030 | } |
1031 | } | |
5f39d397 | 1032 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1033 | /* |
1034 | * we're not allowed to leave a node with one item in the | |
1035 | * tree during a delete. A deletion from lower in the tree | |
1036 | * could try to delete the only pointer in this node. | |
1037 | * So, pull some keys from the left. | |
1038 | * There has to be a left pointer at this point because | |
1039 | * otherwise we would have pulled some pointers from the | |
1040 | * right | |
1041 | */ | |
5f39d397 CM |
1042 | BUG_ON(!left); |
1043 | wret = balance_node_right(trans, root, mid, left); | |
54aa1f4d | 1044 | if (wret < 0) { |
79f95c82 | 1045 | ret = wret; |
54aa1f4d CM |
1046 | goto enospc; |
1047 | } | |
bce4eae9 CM |
1048 | if (wret == 1) { |
1049 | wret = push_node_left(trans, root, left, mid, 1); | |
1050 | if (wret < 0) | |
1051 | ret = wret; | |
1052 | } | |
79f95c82 CM |
1053 | BUG_ON(wret == 1); |
1054 | } | |
5f39d397 | 1055 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1056 | clean_tree_block(trans, root, mid); |
925baedd | 1057 | btrfs_tree_unlock(mid); |
143bede5 | 1058 | del_ptr(trans, root, path, level + 1, pslot); |
f0486c68 | 1059 | root_sub_used(root, mid->len); |
66d7e7f0 | 1060 | btrfs_free_tree_block(trans, root, mid, 0, 1, 0); |
f0486c68 YZ |
1061 | free_extent_buffer(mid); |
1062 | mid = NULL; | |
79f95c82 CM |
1063 | } else { |
1064 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1065 | struct btrfs_disk_key mid_key; |
1066 | btrfs_node_key(mid, &mid_key, 0); | |
1067 | btrfs_set_node_key(parent, &mid_key, pslot); | |
1068 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1069 | } |
bb803951 | 1070 | |
79f95c82 | 1071 | /* update the path */ |
5f39d397 CM |
1072 | if (left) { |
1073 | if (btrfs_header_nritems(left) > orig_slot) { | |
1074 | extent_buffer_get(left); | |
925baedd | 1075 | /* left was locked after cow */ |
5f39d397 | 1076 | path->nodes[level] = left; |
bb803951 CM |
1077 | path->slots[level + 1] -= 1; |
1078 | path->slots[level] = orig_slot; | |
925baedd CM |
1079 | if (mid) { |
1080 | btrfs_tree_unlock(mid); | |
5f39d397 | 1081 | free_extent_buffer(mid); |
925baedd | 1082 | } |
bb803951 | 1083 | } else { |
5f39d397 | 1084 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1085 | path->slots[level] = orig_slot; |
1086 | } | |
1087 | } | |
79f95c82 | 1088 | /* double check we haven't messed things up */ |
e20d96d6 | 1089 | if (orig_ptr != |
5f39d397 | 1090 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1091 | BUG(); |
54aa1f4d | 1092 | enospc: |
925baedd CM |
1093 | if (right) { |
1094 | btrfs_tree_unlock(right); | |
5f39d397 | 1095 | free_extent_buffer(right); |
925baedd CM |
1096 | } |
1097 | if (left) { | |
1098 | if (path->nodes[level] != left) | |
1099 | btrfs_tree_unlock(left); | |
5f39d397 | 1100 | free_extent_buffer(left); |
925baedd | 1101 | } |
bb803951 CM |
1102 | return ret; |
1103 | } | |
1104 | ||
d352ac68 CM |
1105 | /* Node balancing for insertion. Here we only split or push nodes around |
1106 | * when they are completely full. This is also done top down, so we | |
1107 | * have to be pessimistic. | |
1108 | */ | |
d397712b | 1109 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1110 | struct btrfs_root *root, |
1111 | struct btrfs_path *path, int level) | |
e66f709b | 1112 | { |
5f39d397 CM |
1113 | struct extent_buffer *right = NULL; |
1114 | struct extent_buffer *mid; | |
1115 | struct extent_buffer *left = NULL; | |
1116 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1117 | int ret = 0; |
1118 | int wret; | |
1119 | int pslot; | |
1120 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1121 | |
1122 | if (level == 0) | |
1123 | return 1; | |
1124 | ||
5f39d397 | 1125 | mid = path->nodes[level]; |
7bb86316 | 1126 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1127 | |
a05a9bb1 | 1128 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1129 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1130 | pslot = path->slots[level + 1]; |
1131 | } | |
e66f709b | 1132 | |
5f39d397 | 1133 | if (!parent) |
e66f709b | 1134 | return 1; |
e66f709b | 1135 | |
5f39d397 | 1136 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1137 | |
1138 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1139 | if (left) { |
e66f709b | 1140 | u32 left_nr; |
925baedd CM |
1141 | |
1142 | btrfs_tree_lock(left); | |
b4ce94de CM |
1143 | btrfs_set_lock_blocking(left); |
1144 | ||
5f39d397 | 1145 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1146 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1147 | wret = 1; | |
1148 | } else { | |
5f39d397 | 1149 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1150 | pslot - 1, &left); |
54aa1f4d CM |
1151 | if (ret) |
1152 | wret = 1; | |
1153 | else { | |
54aa1f4d | 1154 | wret = push_node_left(trans, root, |
971a1f66 | 1155 | left, mid, 0); |
54aa1f4d | 1156 | } |
33ade1f8 | 1157 | } |
e66f709b CM |
1158 | if (wret < 0) |
1159 | ret = wret; | |
1160 | if (wret == 0) { | |
5f39d397 | 1161 | struct btrfs_disk_key disk_key; |
e66f709b | 1162 | orig_slot += left_nr; |
5f39d397 CM |
1163 | btrfs_node_key(mid, &disk_key, 0); |
1164 | btrfs_set_node_key(parent, &disk_key, pslot); | |
1165 | btrfs_mark_buffer_dirty(parent); | |
1166 | if (btrfs_header_nritems(left) > orig_slot) { | |
1167 | path->nodes[level] = left; | |
e66f709b CM |
1168 | path->slots[level + 1] -= 1; |
1169 | path->slots[level] = orig_slot; | |
925baedd | 1170 | btrfs_tree_unlock(mid); |
5f39d397 | 1171 | free_extent_buffer(mid); |
e66f709b CM |
1172 | } else { |
1173 | orig_slot -= | |
5f39d397 | 1174 | btrfs_header_nritems(left); |
e66f709b | 1175 | path->slots[level] = orig_slot; |
925baedd | 1176 | btrfs_tree_unlock(left); |
5f39d397 | 1177 | free_extent_buffer(left); |
e66f709b | 1178 | } |
e66f709b CM |
1179 | return 0; |
1180 | } | |
925baedd | 1181 | btrfs_tree_unlock(left); |
5f39d397 | 1182 | free_extent_buffer(left); |
e66f709b | 1183 | } |
925baedd | 1184 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1185 | |
1186 | /* | |
1187 | * then try to empty the right most buffer into the middle | |
1188 | */ | |
5f39d397 | 1189 | if (right) { |
33ade1f8 | 1190 | u32 right_nr; |
b4ce94de | 1191 | |
925baedd | 1192 | btrfs_tree_lock(right); |
b4ce94de CM |
1193 | btrfs_set_lock_blocking(right); |
1194 | ||
5f39d397 | 1195 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1196 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1197 | wret = 1; | |
1198 | } else { | |
5f39d397 CM |
1199 | ret = btrfs_cow_block(trans, root, right, |
1200 | parent, pslot + 1, | |
9fa8cfe7 | 1201 | &right); |
54aa1f4d CM |
1202 | if (ret) |
1203 | wret = 1; | |
1204 | else { | |
54aa1f4d | 1205 | wret = balance_node_right(trans, root, |
5f39d397 | 1206 | right, mid); |
54aa1f4d | 1207 | } |
33ade1f8 | 1208 | } |
e66f709b CM |
1209 | if (wret < 0) |
1210 | ret = wret; | |
1211 | if (wret == 0) { | |
5f39d397 CM |
1212 | struct btrfs_disk_key disk_key; |
1213 | ||
1214 | btrfs_node_key(right, &disk_key, 0); | |
1215 | btrfs_set_node_key(parent, &disk_key, pslot + 1); | |
1216 | btrfs_mark_buffer_dirty(parent); | |
1217 | ||
1218 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1219 | path->nodes[level] = right; | |
e66f709b CM |
1220 | path->slots[level + 1] += 1; |
1221 | path->slots[level] = orig_slot - | |
5f39d397 | 1222 | btrfs_header_nritems(mid); |
925baedd | 1223 | btrfs_tree_unlock(mid); |
5f39d397 | 1224 | free_extent_buffer(mid); |
e66f709b | 1225 | } else { |
925baedd | 1226 | btrfs_tree_unlock(right); |
5f39d397 | 1227 | free_extent_buffer(right); |
e66f709b | 1228 | } |
e66f709b CM |
1229 | return 0; |
1230 | } | |
925baedd | 1231 | btrfs_tree_unlock(right); |
5f39d397 | 1232 | free_extent_buffer(right); |
e66f709b | 1233 | } |
e66f709b CM |
1234 | return 1; |
1235 | } | |
1236 | ||
3c69faec | 1237 | /* |
d352ac68 CM |
1238 | * readahead one full node of leaves, finding things that are close |
1239 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1240 | */ |
c8c42864 CM |
1241 | static void reada_for_search(struct btrfs_root *root, |
1242 | struct btrfs_path *path, | |
1243 | int level, int slot, u64 objectid) | |
3c69faec | 1244 | { |
5f39d397 | 1245 | struct extent_buffer *node; |
01f46658 | 1246 | struct btrfs_disk_key disk_key; |
3c69faec | 1247 | u32 nritems; |
3c69faec | 1248 | u64 search; |
a7175319 | 1249 | u64 target; |
6b80053d | 1250 | u64 nread = 0; |
cb25c2ea | 1251 | u64 gen; |
3c69faec | 1252 | int direction = path->reada; |
5f39d397 | 1253 | struct extent_buffer *eb; |
6b80053d CM |
1254 | u32 nr; |
1255 | u32 blocksize; | |
1256 | u32 nscan = 0; | |
db94535d | 1257 | |
a6b6e75e | 1258 | if (level != 1) |
6702ed49 CM |
1259 | return; |
1260 | ||
1261 | if (!path->nodes[level]) | |
3c69faec CM |
1262 | return; |
1263 | ||
5f39d397 | 1264 | node = path->nodes[level]; |
925baedd | 1265 | |
3c69faec | 1266 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1267 | blocksize = btrfs_level_size(root, level - 1); |
1268 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1269 | if (eb) { |
1270 | free_extent_buffer(eb); | |
3c69faec CM |
1271 | return; |
1272 | } | |
1273 | ||
a7175319 | 1274 | target = search; |
6b80053d | 1275 | |
5f39d397 | 1276 | nritems = btrfs_header_nritems(node); |
6b80053d | 1277 | nr = slot; |
25b8b936 | 1278 | |
d397712b | 1279 | while (1) { |
6b80053d CM |
1280 | if (direction < 0) { |
1281 | if (nr == 0) | |
1282 | break; | |
1283 | nr--; | |
1284 | } else if (direction > 0) { | |
1285 | nr++; | |
1286 | if (nr >= nritems) | |
1287 | break; | |
3c69faec | 1288 | } |
01f46658 CM |
1289 | if (path->reada < 0 && objectid) { |
1290 | btrfs_node_key(node, &disk_key, nr); | |
1291 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1292 | break; | |
1293 | } | |
6b80053d | 1294 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
1295 | if ((search <= target && target - search <= 65536) || |
1296 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 1297 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 1298 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
1299 | nread += blocksize; |
1300 | } | |
1301 | nscan++; | |
a7175319 | 1302 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 1303 | break; |
3c69faec CM |
1304 | } |
1305 | } | |
925baedd | 1306 | |
b4ce94de CM |
1307 | /* |
1308 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
1309 | * cache | |
1310 | */ | |
1311 | static noinline int reada_for_balance(struct btrfs_root *root, | |
1312 | struct btrfs_path *path, int level) | |
1313 | { | |
1314 | int slot; | |
1315 | int nritems; | |
1316 | struct extent_buffer *parent; | |
1317 | struct extent_buffer *eb; | |
1318 | u64 gen; | |
1319 | u64 block1 = 0; | |
1320 | u64 block2 = 0; | |
1321 | int ret = 0; | |
1322 | int blocksize; | |
1323 | ||
8c594ea8 | 1324 | parent = path->nodes[level + 1]; |
b4ce94de CM |
1325 | if (!parent) |
1326 | return 0; | |
1327 | ||
1328 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1329 | slot = path->slots[level + 1]; |
b4ce94de CM |
1330 | blocksize = btrfs_level_size(root, level); |
1331 | ||
1332 | if (slot > 0) { | |
1333 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
1334 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
1335 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
1336 | if (eb && btrfs_buffer_uptodate(eb, gen)) | |
1337 | block1 = 0; | |
1338 | free_extent_buffer(eb); | |
1339 | } | |
8c594ea8 | 1340 | if (slot + 1 < nritems) { |
b4ce94de CM |
1341 | block2 = btrfs_node_blockptr(parent, slot + 1); |
1342 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
1343 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
1344 | if (eb && btrfs_buffer_uptodate(eb, gen)) | |
1345 | block2 = 0; | |
1346 | free_extent_buffer(eb); | |
1347 | } | |
1348 | if (block1 || block2) { | |
1349 | ret = -EAGAIN; | |
8c594ea8 CM |
1350 | |
1351 | /* release the whole path */ | |
b3b4aa74 | 1352 | btrfs_release_path(path); |
8c594ea8 CM |
1353 | |
1354 | /* read the blocks */ | |
b4ce94de CM |
1355 | if (block1) |
1356 | readahead_tree_block(root, block1, blocksize, 0); | |
1357 | if (block2) | |
1358 | readahead_tree_block(root, block2, blocksize, 0); | |
1359 | ||
1360 | if (block1) { | |
1361 | eb = read_tree_block(root, block1, blocksize, 0); | |
1362 | free_extent_buffer(eb); | |
1363 | } | |
8c594ea8 | 1364 | if (block2) { |
b4ce94de CM |
1365 | eb = read_tree_block(root, block2, blocksize, 0); |
1366 | free_extent_buffer(eb); | |
1367 | } | |
1368 | } | |
1369 | return ret; | |
1370 | } | |
1371 | ||
1372 | ||
d352ac68 | 1373 | /* |
d397712b CM |
1374 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1375 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1376 | * operations on the tree might require changing key pointers higher up in the | |
1377 | * tree. | |
d352ac68 | 1378 | * |
d397712b CM |
1379 | * callers might also have set path->keep_locks, which tells this code to keep |
1380 | * the lock if the path points to the last slot in the block. This is part of | |
1381 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1382 | * |
d397712b CM |
1383 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1384 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1385 | */ |
e02119d5 CM |
1386 | static noinline void unlock_up(struct btrfs_path *path, int level, |
1387 | int lowest_unlock) | |
925baedd CM |
1388 | { |
1389 | int i; | |
1390 | int skip_level = level; | |
051e1b9f | 1391 | int no_skips = 0; |
925baedd CM |
1392 | struct extent_buffer *t; |
1393 | ||
1394 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1395 | if (!path->nodes[i]) | |
1396 | break; | |
1397 | if (!path->locks[i]) | |
1398 | break; | |
051e1b9f | 1399 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1400 | skip_level = i + 1; |
1401 | continue; | |
1402 | } | |
051e1b9f | 1403 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1404 | u32 nritems; |
1405 | t = path->nodes[i]; | |
1406 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1407 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1408 | skip_level = i + 1; |
1409 | continue; | |
1410 | } | |
1411 | } | |
051e1b9f CM |
1412 | if (skip_level < i && i >= lowest_unlock) |
1413 | no_skips = 1; | |
1414 | ||
925baedd CM |
1415 | t = path->nodes[i]; |
1416 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 1417 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd CM |
1418 | path->locks[i] = 0; |
1419 | } | |
1420 | } | |
1421 | } | |
1422 | ||
b4ce94de CM |
1423 | /* |
1424 | * This releases any locks held in the path starting at level and | |
1425 | * going all the way up to the root. | |
1426 | * | |
1427 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
1428 | * corner cases, such as COW of the block at slot zero in the node. This | |
1429 | * ignores those rules, and it should only be called when there are no | |
1430 | * more updates to be done higher up in the tree. | |
1431 | */ | |
1432 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
1433 | { | |
1434 | int i; | |
1435 | ||
5d4f98a2 | 1436 | if (path->keep_locks) |
b4ce94de CM |
1437 | return; |
1438 | ||
1439 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1440 | if (!path->nodes[i]) | |
12f4dacc | 1441 | continue; |
b4ce94de | 1442 | if (!path->locks[i]) |
12f4dacc | 1443 | continue; |
bd681513 | 1444 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
1445 | path->locks[i] = 0; |
1446 | } | |
1447 | } | |
1448 | ||
c8c42864 CM |
1449 | /* |
1450 | * helper function for btrfs_search_slot. The goal is to find a block | |
1451 | * in cache without setting the path to blocking. If we find the block | |
1452 | * we return zero and the path is unchanged. | |
1453 | * | |
1454 | * If we can't find the block, we set the path blocking and do some | |
1455 | * reada. -EAGAIN is returned and the search must be repeated. | |
1456 | */ | |
1457 | static int | |
1458 | read_block_for_search(struct btrfs_trans_handle *trans, | |
1459 | struct btrfs_root *root, struct btrfs_path *p, | |
1460 | struct extent_buffer **eb_ret, int level, int slot, | |
1461 | struct btrfs_key *key) | |
1462 | { | |
1463 | u64 blocknr; | |
1464 | u64 gen; | |
1465 | u32 blocksize; | |
1466 | struct extent_buffer *b = *eb_ret; | |
1467 | struct extent_buffer *tmp; | |
76a05b35 | 1468 | int ret; |
c8c42864 CM |
1469 | |
1470 | blocknr = btrfs_node_blockptr(b, slot); | |
1471 | gen = btrfs_node_ptr_generation(b, slot); | |
1472 | blocksize = btrfs_level_size(root, level - 1); | |
1473 | ||
1474 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a CM |
1475 | if (tmp) { |
1476 | if (btrfs_buffer_uptodate(tmp, 0)) { | |
1477 | if (btrfs_buffer_uptodate(tmp, gen)) { | |
1478 | /* | |
1479 | * we found an up to date block without | |
1480 | * sleeping, return | |
1481 | * right away | |
1482 | */ | |
1483 | *eb_ret = tmp; | |
1484 | return 0; | |
1485 | } | |
1486 | /* the pages were up to date, but we failed | |
1487 | * the generation number check. Do a full | |
1488 | * read for the generation number that is correct. | |
1489 | * We must do this without dropping locks so | |
1490 | * we can trust our generation number | |
1491 | */ | |
1492 | free_extent_buffer(tmp); | |
bd681513 CM |
1493 | btrfs_set_path_blocking(p); |
1494 | ||
cb44921a CM |
1495 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
1496 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
1497 | *eb_ret = tmp; | |
1498 | return 0; | |
1499 | } | |
1500 | free_extent_buffer(tmp); | |
b3b4aa74 | 1501 | btrfs_release_path(p); |
cb44921a CM |
1502 | return -EIO; |
1503 | } | |
c8c42864 CM |
1504 | } |
1505 | ||
1506 | /* | |
1507 | * reduce lock contention at high levels | |
1508 | * of the btree by dropping locks before | |
76a05b35 CM |
1509 | * we read. Don't release the lock on the current |
1510 | * level because we need to walk this node to figure | |
1511 | * out which blocks to read. | |
c8c42864 | 1512 | */ |
8c594ea8 CM |
1513 | btrfs_unlock_up_safe(p, level + 1); |
1514 | btrfs_set_path_blocking(p); | |
1515 | ||
cb44921a | 1516 | free_extent_buffer(tmp); |
c8c42864 CM |
1517 | if (p->reada) |
1518 | reada_for_search(root, p, level, slot, key->objectid); | |
1519 | ||
b3b4aa74 | 1520 | btrfs_release_path(p); |
76a05b35 CM |
1521 | |
1522 | ret = -EAGAIN; | |
5bdd3536 | 1523 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
1524 | if (tmp) { |
1525 | /* | |
1526 | * If the read above didn't mark this buffer up to date, | |
1527 | * it will never end up being up to date. Set ret to EIO now | |
1528 | * and give up so that our caller doesn't loop forever | |
1529 | * on our EAGAINs. | |
1530 | */ | |
1531 | if (!btrfs_buffer_uptodate(tmp, 0)) | |
1532 | ret = -EIO; | |
c8c42864 | 1533 | free_extent_buffer(tmp); |
76a05b35 CM |
1534 | } |
1535 | return ret; | |
c8c42864 CM |
1536 | } |
1537 | ||
1538 | /* | |
1539 | * helper function for btrfs_search_slot. This does all of the checks | |
1540 | * for node-level blocks and does any balancing required based on | |
1541 | * the ins_len. | |
1542 | * | |
1543 | * If no extra work was required, zero is returned. If we had to | |
1544 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1545 | * start over | |
1546 | */ | |
1547 | static int | |
1548 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1549 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1550 | struct extent_buffer *b, int level, int ins_len, |
1551 | int *write_lock_level) | |
c8c42864 CM |
1552 | { |
1553 | int ret; | |
1554 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
1555 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
1556 | int sret; | |
1557 | ||
bd681513 CM |
1558 | if (*write_lock_level < level + 1) { |
1559 | *write_lock_level = level + 1; | |
1560 | btrfs_release_path(p); | |
1561 | goto again; | |
1562 | } | |
1563 | ||
c8c42864 CM |
1564 | sret = reada_for_balance(root, p, level); |
1565 | if (sret) | |
1566 | goto again; | |
1567 | ||
1568 | btrfs_set_path_blocking(p); | |
1569 | sret = split_node(trans, root, p, level); | |
bd681513 | 1570 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
1571 | |
1572 | BUG_ON(sret > 0); | |
1573 | if (sret) { | |
1574 | ret = sret; | |
1575 | goto done; | |
1576 | } | |
1577 | b = p->nodes[level]; | |
1578 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 1579 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
1580 | int sret; |
1581 | ||
bd681513 CM |
1582 | if (*write_lock_level < level + 1) { |
1583 | *write_lock_level = level + 1; | |
1584 | btrfs_release_path(p); | |
1585 | goto again; | |
1586 | } | |
1587 | ||
c8c42864 CM |
1588 | sret = reada_for_balance(root, p, level); |
1589 | if (sret) | |
1590 | goto again; | |
1591 | ||
1592 | btrfs_set_path_blocking(p); | |
1593 | sret = balance_level(trans, root, p, level); | |
bd681513 | 1594 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
1595 | |
1596 | if (sret) { | |
1597 | ret = sret; | |
1598 | goto done; | |
1599 | } | |
1600 | b = p->nodes[level]; | |
1601 | if (!b) { | |
b3b4aa74 | 1602 | btrfs_release_path(p); |
c8c42864 CM |
1603 | goto again; |
1604 | } | |
1605 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1606 | } | |
1607 | return 0; | |
1608 | ||
1609 | again: | |
1610 | ret = -EAGAIN; | |
1611 | done: | |
1612 | return ret; | |
1613 | } | |
1614 | ||
74123bd7 CM |
1615 | /* |
1616 | * look for key in the tree. path is filled in with nodes along the way | |
1617 | * if key is found, we return zero and you can find the item in the leaf | |
1618 | * level of the path (level 0) | |
1619 | * | |
1620 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
1621 | * be inserted, and 1 is returned. If there are other errors during the |
1622 | * search a negative error number is returned. | |
97571fd0 CM |
1623 | * |
1624 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
1625 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
1626 | * possible) | |
74123bd7 | 1627 | */ |
e089f05c CM |
1628 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
1629 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
1630 | ins_len, int cow) | |
be0e5c09 | 1631 | { |
5f39d397 | 1632 | struct extent_buffer *b; |
be0e5c09 CM |
1633 | int slot; |
1634 | int ret; | |
33c66f43 | 1635 | int err; |
be0e5c09 | 1636 | int level; |
925baedd | 1637 | int lowest_unlock = 1; |
bd681513 CM |
1638 | int root_lock; |
1639 | /* everything at write_lock_level or lower must be write locked */ | |
1640 | int write_lock_level = 0; | |
9f3a7427 CM |
1641 | u8 lowest_level = 0; |
1642 | ||
6702ed49 | 1643 | lowest_level = p->lowest_level; |
323ac95b | 1644 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1645 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 1646 | |
bd681513 | 1647 | if (ins_len < 0) { |
925baedd | 1648 | lowest_unlock = 2; |
65b51a00 | 1649 | |
bd681513 CM |
1650 | /* when we are removing items, we might have to go up to level |
1651 | * two as we update tree pointers Make sure we keep write | |
1652 | * for those levels as well | |
1653 | */ | |
1654 | write_lock_level = 2; | |
1655 | } else if (ins_len > 0) { | |
1656 | /* | |
1657 | * for inserting items, make sure we have a write lock on | |
1658 | * level 1 so we can update keys | |
1659 | */ | |
1660 | write_lock_level = 1; | |
1661 | } | |
1662 | ||
1663 | if (!cow) | |
1664 | write_lock_level = -1; | |
1665 | ||
1666 | if (cow && (p->keep_locks || p->lowest_level)) | |
1667 | write_lock_level = BTRFS_MAX_LEVEL; | |
1668 | ||
bb803951 | 1669 | again: |
bd681513 CM |
1670 | /* |
1671 | * we try very hard to do read locks on the root | |
1672 | */ | |
1673 | root_lock = BTRFS_READ_LOCK; | |
1674 | level = 0; | |
5d4f98a2 | 1675 | if (p->search_commit_root) { |
bd681513 CM |
1676 | /* |
1677 | * the commit roots are read only | |
1678 | * so we always do read locks | |
1679 | */ | |
5d4f98a2 YZ |
1680 | b = root->commit_root; |
1681 | extent_buffer_get(b); | |
bd681513 | 1682 | level = btrfs_header_level(b); |
5d4f98a2 | 1683 | if (!p->skip_locking) |
bd681513 | 1684 | btrfs_tree_read_lock(b); |
5d4f98a2 | 1685 | } else { |
bd681513 | 1686 | if (p->skip_locking) { |
5d4f98a2 | 1687 | b = btrfs_root_node(root); |
bd681513 CM |
1688 | level = btrfs_header_level(b); |
1689 | } else { | |
1690 | /* we don't know the level of the root node | |
1691 | * until we actually have it read locked | |
1692 | */ | |
1693 | b = btrfs_read_lock_root_node(root); | |
1694 | level = btrfs_header_level(b); | |
1695 | if (level <= write_lock_level) { | |
1696 | /* whoops, must trade for write lock */ | |
1697 | btrfs_tree_read_unlock(b); | |
1698 | free_extent_buffer(b); | |
1699 | b = btrfs_lock_root_node(root); | |
1700 | root_lock = BTRFS_WRITE_LOCK; | |
1701 | ||
1702 | /* the level might have changed, check again */ | |
1703 | level = btrfs_header_level(b); | |
1704 | } | |
1705 | } | |
5d4f98a2 | 1706 | } |
bd681513 CM |
1707 | p->nodes[level] = b; |
1708 | if (!p->skip_locking) | |
1709 | p->locks[level] = root_lock; | |
925baedd | 1710 | |
eb60ceac | 1711 | while (b) { |
5f39d397 | 1712 | level = btrfs_header_level(b); |
65b51a00 CM |
1713 | |
1714 | /* | |
1715 | * setup the path here so we can release it under lock | |
1716 | * contention with the cow code | |
1717 | */ | |
02217ed2 | 1718 | if (cow) { |
c8c42864 CM |
1719 | /* |
1720 | * if we don't really need to cow this block | |
1721 | * then we don't want to set the path blocking, | |
1722 | * so we test it here | |
1723 | */ | |
5d4f98a2 | 1724 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 1725 | goto cow_done; |
5d4f98a2 | 1726 | |
b4ce94de CM |
1727 | btrfs_set_path_blocking(p); |
1728 | ||
bd681513 CM |
1729 | /* |
1730 | * must have write locks on this node and the | |
1731 | * parent | |
1732 | */ | |
1733 | if (level + 1 > write_lock_level) { | |
1734 | write_lock_level = level + 1; | |
1735 | btrfs_release_path(p); | |
1736 | goto again; | |
1737 | } | |
1738 | ||
33c66f43 YZ |
1739 | err = btrfs_cow_block(trans, root, b, |
1740 | p->nodes[level + 1], | |
1741 | p->slots[level + 1], &b); | |
1742 | if (err) { | |
33c66f43 | 1743 | ret = err; |
65b51a00 | 1744 | goto done; |
54aa1f4d | 1745 | } |
02217ed2 | 1746 | } |
65b51a00 | 1747 | cow_done: |
02217ed2 | 1748 | BUG_ON(!cow && ins_len); |
65b51a00 | 1749 | |
eb60ceac | 1750 | p->nodes[level] = b; |
bd681513 | 1751 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
1752 | |
1753 | /* | |
1754 | * we have a lock on b and as long as we aren't changing | |
1755 | * the tree, there is no way to for the items in b to change. | |
1756 | * It is safe to drop the lock on our parent before we | |
1757 | * go through the expensive btree search on b. | |
1758 | * | |
1759 | * If cow is true, then we might be changing slot zero, | |
1760 | * which may require changing the parent. So, we can't | |
1761 | * drop the lock until after we know which slot we're | |
1762 | * operating on. | |
1763 | */ | |
1764 | if (!cow) | |
1765 | btrfs_unlock_up_safe(p, level + 1); | |
1766 | ||
5f39d397 | 1767 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 1768 | |
5f39d397 | 1769 | if (level != 0) { |
33c66f43 YZ |
1770 | int dec = 0; |
1771 | if (ret && slot > 0) { | |
1772 | dec = 1; | |
be0e5c09 | 1773 | slot -= 1; |
33c66f43 | 1774 | } |
be0e5c09 | 1775 | p->slots[level] = slot; |
33c66f43 | 1776 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 1777 | ins_len, &write_lock_level); |
33c66f43 | 1778 | if (err == -EAGAIN) |
c8c42864 | 1779 | goto again; |
33c66f43 YZ |
1780 | if (err) { |
1781 | ret = err; | |
c8c42864 | 1782 | goto done; |
33c66f43 | 1783 | } |
c8c42864 CM |
1784 | b = p->nodes[level]; |
1785 | slot = p->slots[level]; | |
b4ce94de | 1786 | |
bd681513 CM |
1787 | /* |
1788 | * slot 0 is special, if we change the key | |
1789 | * we have to update the parent pointer | |
1790 | * which means we must have a write lock | |
1791 | * on the parent | |
1792 | */ | |
1793 | if (slot == 0 && cow && | |
1794 | write_lock_level < level + 1) { | |
1795 | write_lock_level = level + 1; | |
1796 | btrfs_release_path(p); | |
1797 | goto again; | |
1798 | } | |
1799 | ||
f9efa9c7 CM |
1800 | unlock_up(p, level, lowest_unlock); |
1801 | ||
925baedd | 1802 | if (level == lowest_level) { |
33c66f43 YZ |
1803 | if (dec) |
1804 | p->slots[level]++; | |
5b21f2ed | 1805 | goto done; |
925baedd | 1806 | } |
ca7a79ad | 1807 | |
33c66f43 | 1808 | err = read_block_for_search(trans, root, p, |
c8c42864 | 1809 | &b, level, slot, key); |
33c66f43 | 1810 | if (err == -EAGAIN) |
c8c42864 | 1811 | goto again; |
33c66f43 YZ |
1812 | if (err) { |
1813 | ret = err; | |
76a05b35 | 1814 | goto done; |
33c66f43 | 1815 | } |
76a05b35 | 1816 | |
b4ce94de | 1817 | if (!p->skip_locking) { |
bd681513 CM |
1818 | level = btrfs_header_level(b); |
1819 | if (level <= write_lock_level) { | |
1820 | err = btrfs_try_tree_write_lock(b); | |
1821 | if (!err) { | |
1822 | btrfs_set_path_blocking(p); | |
1823 | btrfs_tree_lock(b); | |
1824 | btrfs_clear_path_blocking(p, b, | |
1825 | BTRFS_WRITE_LOCK); | |
1826 | } | |
1827 | p->locks[level] = BTRFS_WRITE_LOCK; | |
1828 | } else { | |
1829 | err = btrfs_try_tree_read_lock(b); | |
1830 | if (!err) { | |
1831 | btrfs_set_path_blocking(p); | |
1832 | btrfs_tree_read_lock(b); | |
1833 | btrfs_clear_path_blocking(p, b, | |
1834 | BTRFS_READ_LOCK); | |
1835 | } | |
1836 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 1837 | } |
bd681513 | 1838 | p->nodes[level] = b; |
b4ce94de | 1839 | } |
be0e5c09 CM |
1840 | } else { |
1841 | p->slots[level] = slot; | |
87b29b20 YZ |
1842 | if (ins_len > 0 && |
1843 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
1844 | if (write_lock_level < 1) { |
1845 | write_lock_level = 1; | |
1846 | btrfs_release_path(p); | |
1847 | goto again; | |
1848 | } | |
1849 | ||
b4ce94de | 1850 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
1851 | err = split_leaf(trans, root, key, |
1852 | p, ins_len, ret == 0); | |
bd681513 | 1853 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 1854 | |
33c66f43 YZ |
1855 | BUG_ON(err > 0); |
1856 | if (err) { | |
1857 | ret = err; | |
65b51a00 CM |
1858 | goto done; |
1859 | } | |
5c680ed6 | 1860 | } |
459931ec CM |
1861 | if (!p->search_for_split) |
1862 | unlock_up(p, level, lowest_unlock); | |
65b51a00 | 1863 | goto done; |
be0e5c09 CM |
1864 | } |
1865 | } | |
65b51a00 CM |
1866 | ret = 1; |
1867 | done: | |
b4ce94de CM |
1868 | /* |
1869 | * we don't really know what they plan on doing with the path | |
1870 | * from here on, so for now just mark it as blocking | |
1871 | */ | |
b9473439 CM |
1872 | if (!p->leave_spinning) |
1873 | btrfs_set_path_blocking(p); | |
76a05b35 | 1874 | if (ret < 0) |
b3b4aa74 | 1875 | btrfs_release_path(p); |
65b51a00 | 1876 | return ret; |
be0e5c09 CM |
1877 | } |
1878 | ||
74123bd7 CM |
1879 | /* |
1880 | * adjust the pointers going up the tree, starting at level | |
1881 | * making sure the right key of each node is points to 'key'. | |
1882 | * This is used after shifting pointers to the left, so it stops | |
1883 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
1884 | * higher levels | |
aa5d6bed | 1885 | * |
74123bd7 | 1886 | */ |
143bede5 JM |
1887 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
1888 | struct btrfs_root *root, struct btrfs_path *path, | |
1889 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
1890 | { |
1891 | int i; | |
5f39d397 CM |
1892 | struct extent_buffer *t; |
1893 | ||
234b63a0 | 1894 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 1895 | int tslot = path->slots[i]; |
eb60ceac | 1896 | if (!path->nodes[i]) |
be0e5c09 | 1897 | break; |
5f39d397 CM |
1898 | t = path->nodes[i]; |
1899 | btrfs_set_node_key(t, key, tslot); | |
d6025579 | 1900 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
1901 | if (tslot != 0) |
1902 | break; | |
1903 | } | |
1904 | } | |
1905 | ||
31840ae1 ZY |
1906 | /* |
1907 | * update item key. | |
1908 | * | |
1909 | * This function isn't completely safe. It's the caller's responsibility | |
1910 | * that the new key won't break the order | |
1911 | */ | |
143bede5 JM |
1912 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
1913 | struct btrfs_root *root, struct btrfs_path *path, | |
1914 | struct btrfs_key *new_key) | |
31840ae1 ZY |
1915 | { |
1916 | struct btrfs_disk_key disk_key; | |
1917 | struct extent_buffer *eb; | |
1918 | int slot; | |
1919 | ||
1920 | eb = path->nodes[0]; | |
1921 | slot = path->slots[0]; | |
1922 | if (slot > 0) { | |
1923 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 1924 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
1925 | } |
1926 | if (slot < btrfs_header_nritems(eb) - 1) { | |
1927 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 1928 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
1929 | } |
1930 | ||
1931 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
1932 | btrfs_set_item_key(eb, &disk_key, slot); | |
1933 | btrfs_mark_buffer_dirty(eb); | |
1934 | if (slot == 0) | |
1935 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
31840ae1 ZY |
1936 | } |
1937 | ||
74123bd7 CM |
1938 | /* |
1939 | * try to push data from one node into the next node left in the | |
79f95c82 | 1940 | * tree. |
aa5d6bed CM |
1941 | * |
1942 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
1943 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 1944 | */ |
98ed5174 CM |
1945 | static int push_node_left(struct btrfs_trans_handle *trans, |
1946 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 1947 | struct extent_buffer *src, int empty) |
be0e5c09 | 1948 | { |
be0e5c09 | 1949 | int push_items = 0; |
bb803951 CM |
1950 | int src_nritems; |
1951 | int dst_nritems; | |
aa5d6bed | 1952 | int ret = 0; |
be0e5c09 | 1953 | |
5f39d397 CM |
1954 | src_nritems = btrfs_header_nritems(src); |
1955 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 1956 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
1957 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
1958 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 1959 | |
bce4eae9 | 1960 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
1961 | return 1; |
1962 | ||
d397712b | 1963 | if (push_items <= 0) |
be0e5c09 CM |
1964 | return 1; |
1965 | ||
bce4eae9 | 1966 | if (empty) { |
971a1f66 | 1967 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
1968 | if (push_items < src_nritems) { |
1969 | /* leave at least 8 pointers in the node if | |
1970 | * we aren't going to empty it | |
1971 | */ | |
1972 | if (src_nritems - push_items < 8) { | |
1973 | if (push_items <= 8) | |
1974 | return 1; | |
1975 | push_items -= 8; | |
1976 | } | |
1977 | } | |
1978 | } else | |
1979 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 1980 | |
5f39d397 CM |
1981 | copy_extent_buffer(dst, src, |
1982 | btrfs_node_key_ptr_offset(dst_nritems), | |
1983 | btrfs_node_key_ptr_offset(0), | |
d397712b | 1984 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 1985 | |
bb803951 | 1986 | if (push_items < src_nritems) { |
5f39d397 CM |
1987 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
1988 | btrfs_node_key_ptr_offset(push_items), | |
1989 | (src_nritems - push_items) * | |
1990 | sizeof(struct btrfs_key_ptr)); | |
1991 | } | |
1992 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
1993 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
1994 | btrfs_mark_buffer_dirty(src); | |
1995 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 1996 | |
79f95c82 CM |
1997 | return ret; |
1998 | } | |
1999 | ||
2000 | /* | |
2001 | * try to push data from one node into the next node right in the | |
2002 | * tree. | |
2003 | * | |
2004 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2005 | * error, and > 0 if there was no room in the right hand block. | |
2006 | * | |
2007 | * this will only push up to 1/2 the contents of the left node over | |
2008 | */ | |
5f39d397 CM |
2009 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2010 | struct btrfs_root *root, | |
2011 | struct extent_buffer *dst, | |
2012 | struct extent_buffer *src) | |
79f95c82 | 2013 | { |
79f95c82 CM |
2014 | int push_items = 0; |
2015 | int max_push; | |
2016 | int src_nritems; | |
2017 | int dst_nritems; | |
2018 | int ret = 0; | |
79f95c82 | 2019 | |
7bb86316 CM |
2020 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2021 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2022 | ||
5f39d397 CM |
2023 | src_nritems = btrfs_header_nritems(src); |
2024 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2025 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 2026 | if (push_items <= 0) |
79f95c82 | 2027 | return 1; |
bce4eae9 | 2028 | |
d397712b | 2029 | if (src_nritems < 4) |
bce4eae9 | 2030 | return 1; |
79f95c82 CM |
2031 | |
2032 | max_push = src_nritems / 2 + 1; | |
2033 | /* don't try to empty the node */ | |
d397712b | 2034 | if (max_push >= src_nritems) |
79f95c82 | 2035 | return 1; |
252c38f0 | 2036 | |
79f95c82 CM |
2037 | if (max_push < push_items) |
2038 | push_items = max_push; | |
2039 | ||
5f39d397 CM |
2040 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2041 | btrfs_node_key_ptr_offset(0), | |
2042 | (dst_nritems) * | |
2043 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2044 | |
5f39d397 CM |
2045 | copy_extent_buffer(dst, src, |
2046 | btrfs_node_key_ptr_offset(0), | |
2047 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2048 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2049 | |
5f39d397 CM |
2050 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2051 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2052 | |
5f39d397 CM |
2053 | btrfs_mark_buffer_dirty(src); |
2054 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2055 | |
aa5d6bed | 2056 | return ret; |
be0e5c09 CM |
2057 | } |
2058 | ||
97571fd0 CM |
2059 | /* |
2060 | * helper function to insert a new root level in the tree. | |
2061 | * A new node is allocated, and a single item is inserted to | |
2062 | * point to the existing root | |
aa5d6bed CM |
2063 | * |
2064 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2065 | */ |
d397712b | 2066 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2067 | struct btrfs_root *root, |
2068 | struct btrfs_path *path, int level) | |
5c680ed6 | 2069 | { |
7bb86316 | 2070 | u64 lower_gen; |
5f39d397 CM |
2071 | struct extent_buffer *lower; |
2072 | struct extent_buffer *c; | |
925baedd | 2073 | struct extent_buffer *old; |
5f39d397 | 2074 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
2075 | |
2076 | BUG_ON(path->nodes[level]); | |
2077 | BUG_ON(path->nodes[level-1] != root->node); | |
2078 | ||
7bb86316 CM |
2079 | lower = path->nodes[level-1]; |
2080 | if (level == 1) | |
2081 | btrfs_item_key(lower, &lower_key, 0); | |
2082 | else | |
2083 | btrfs_node_key(lower, &lower_key, 0); | |
2084 | ||
31840ae1 | 2085 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 2086 | root->root_key.objectid, &lower_key, |
66d7e7f0 | 2087 | level, root->node->start, 0, 0); |
5f39d397 CM |
2088 | if (IS_ERR(c)) |
2089 | return PTR_ERR(c); | |
925baedd | 2090 | |
f0486c68 YZ |
2091 | root_add_used(root, root->nodesize); |
2092 | ||
5d4f98a2 | 2093 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
2094 | btrfs_set_header_nritems(c, 1); |
2095 | btrfs_set_header_level(c, level); | |
db94535d | 2096 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 2097 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 2098 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 2099 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
2100 | |
2101 | write_extent_buffer(c, root->fs_info->fsid, | |
2102 | (unsigned long)btrfs_header_fsid(c), | |
2103 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2104 | |
2105 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
2106 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
2107 | BTRFS_UUID_SIZE); | |
2108 | ||
5f39d397 | 2109 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2110 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2111 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2112 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2113 | |
2114 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2115 | |
5f39d397 | 2116 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2117 | |
925baedd | 2118 | old = root->node; |
240f62c8 | 2119 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2120 | |
2121 | /* the super has an extra ref to root->node */ | |
2122 | free_extent_buffer(old); | |
2123 | ||
0b86a832 | 2124 | add_root_to_dirty_list(root); |
5f39d397 CM |
2125 | extent_buffer_get(c); |
2126 | path->nodes[level] = c; | |
bd681513 | 2127 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2128 | path->slots[level] = 0; |
2129 | return 0; | |
2130 | } | |
2131 | ||
74123bd7 CM |
2132 | /* |
2133 | * worker function to insert a single pointer in a node. | |
2134 | * the node should have enough room for the pointer already | |
97571fd0 | 2135 | * |
74123bd7 CM |
2136 | * slot and level indicate where you want the key to go, and |
2137 | * blocknr is the block the key points to. | |
2138 | */ | |
143bede5 JM |
2139 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2140 | struct btrfs_root *root, struct btrfs_path *path, | |
2141 | struct btrfs_disk_key *key, u64 bytenr, | |
2142 | int slot, int level) | |
74123bd7 | 2143 | { |
5f39d397 | 2144 | struct extent_buffer *lower; |
74123bd7 | 2145 | int nritems; |
5c680ed6 CM |
2146 | |
2147 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 2148 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
2149 | lower = path->nodes[level]; |
2150 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2151 | BUG_ON(slot > nritems); |
143bede5 | 2152 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 2153 | if (slot != nritems) { |
5f39d397 CM |
2154 | memmove_extent_buffer(lower, |
2155 | btrfs_node_key_ptr_offset(slot + 1), | |
2156 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2157 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2158 | } |
5f39d397 | 2159 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2160 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2161 | WARN_ON(trans->transid == 0); |
2162 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2163 | btrfs_set_header_nritems(lower, nritems + 1); |
2164 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2165 | } |
2166 | ||
97571fd0 CM |
2167 | /* |
2168 | * split the node at the specified level in path in two. | |
2169 | * The path is corrected to point to the appropriate node after the split | |
2170 | * | |
2171 | * Before splitting this tries to make some room in the node by pushing | |
2172 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2173 | * |
2174 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2175 | */ |
e02119d5 CM |
2176 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2177 | struct btrfs_root *root, | |
2178 | struct btrfs_path *path, int level) | |
be0e5c09 | 2179 | { |
5f39d397 CM |
2180 | struct extent_buffer *c; |
2181 | struct extent_buffer *split; | |
2182 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2183 | int mid; |
5c680ed6 | 2184 | int ret; |
7518a238 | 2185 | u32 c_nritems; |
eb60ceac | 2186 | |
5f39d397 | 2187 | c = path->nodes[level]; |
7bb86316 | 2188 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2189 | if (c == root->node) { |
5c680ed6 | 2190 | /* trying to split the root, lets make a new one */ |
e089f05c | 2191 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2192 | if (ret) |
2193 | return ret; | |
b3612421 | 2194 | } else { |
e66f709b | 2195 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2196 | c = path->nodes[level]; |
2197 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2198 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2199 | return 0; |
54aa1f4d CM |
2200 | if (ret < 0) |
2201 | return ret; | |
be0e5c09 | 2202 | } |
e66f709b | 2203 | |
5f39d397 | 2204 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2205 | mid = (c_nritems + 1) / 2; |
2206 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2207 | |
5d4f98a2 | 2208 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 2209 | root->root_key.objectid, |
66d7e7f0 | 2210 | &disk_key, level, c->start, 0, 0); |
5f39d397 CM |
2211 | if (IS_ERR(split)) |
2212 | return PTR_ERR(split); | |
2213 | ||
f0486c68 YZ |
2214 | root_add_used(root, root->nodesize); |
2215 | ||
5d4f98a2 | 2216 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 2217 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 2218 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 2219 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 2220 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2221 | btrfs_set_header_owner(split, root->root_key.objectid); |
2222 | write_extent_buffer(split, root->fs_info->fsid, | |
2223 | (unsigned long)btrfs_header_fsid(split), | |
2224 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2225 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2226 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2227 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2228 | |
5f39d397 CM |
2229 | |
2230 | copy_extent_buffer(split, c, | |
2231 | btrfs_node_key_ptr_offset(0), | |
2232 | btrfs_node_key_ptr_offset(mid), | |
2233 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2234 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2235 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2236 | ret = 0; |
2237 | ||
5f39d397 CM |
2238 | btrfs_mark_buffer_dirty(c); |
2239 | btrfs_mark_buffer_dirty(split); | |
2240 | ||
143bede5 JM |
2241 | insert_ptr(trans, root, path, &disk_key, split->start, |
2242 | path->slots[level + 1] + 1, level + 1); | |
aa5d6bed | 2243 | |
5de08d7d | 2244 | if (path->slots[level] >= mid) { |
5c680ed6 | 2245 | path->slots[level] -= mid; |
925baedd | 2246 | btrfs_tree_unlock(c); |
5f39d397 CM |
2247 | free_extent_buffer(c); |
2248 | path->nodes[level] = split; | |
5c680ed6 CM |
2249 | path->slots[level + 1] += 1; |
2250 | } else { | |
925baedd | 2251 | btrfs_tree_unlock(split); |
5f39d397 | 2252 | free_extent_buffer(split); |
be0e5c09 | 2253 | } |
aa5d6bed | 2254 | return ret; |
be0e5c09 CM |
2255 | } |
2256 | ||
74123bd7 CM |
2257 | /* |
2258 | * how many bytes are required to store the items in a leaf. start | |
2259 | * and nr indicate which items in the leaf to check. This totals up the | |
2260 | * space used both by the item structs and the item data | |
2261 | */ | |
5f39d397 | 2262 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2263 | { |
2264 | int data_len; | |
5f39d397 | 2265 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2266 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2267 | |
2268 | if (!nr) | |
2269 | return 0; | |
5f39d397 CM |
2270 | data_len = btrfs_item_end_nr(l, start); |
2271 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2272 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2273 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2274 | return data_len; |
2275 | } | |
2276 | ||
d4dbff95 CM |
2277 | /* |
2278 | * The space between the end of the leaf items and | |
2279 | * the start of the leaf data. IOW, how much room | |
2280 | * the leaf has left for both items and data | |
2281 | */ | |
d397712b | 2282 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 2283 | struct extent_buffer *leaf) |
d4dbff95 | 2284 | { |
5f39d397 CM |
2285 | int nritems = btrfs_header_nritems(leaf); |
2286 | int ret; | |
2287 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2288 | if (ret < 0) { | |
d397712b CM |
2289 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
2290 | "used %d nritems %d\n", | |
ae2f5411 | 2291 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2292 | leaf_space_used(leaf, 0, nritems), nritems); |
2293 | } | |
2294 | return ret; | |
d4dbff95 CM |
2295 | } |
2296 | ||
99d8f83c CM |
2297 | /* |
2298 | * min slot controls the lowest index we're willing to push to the | |
2299 | * right. We'll push up to and including min_slot, but no lower | |
2300 | */ | |
44871b1b CM |
2301 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
2302 | struct btrfs_root *root, | |
2303 | struct btrfs_path *path, | |
2304 | int data_size, int empty, | |
2305 | struct extent_buffer *right, | |
99d8f83c CM |
2306 | int free_space, u32 left_nritems, |
2307 | u32 min_slot) | |
00ec4c51 | 2308 | { |
5f39d397 | 2309 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2310 | struct extent_buffer *upper = path->nodes[1]; |
5f39d397 | 2311 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2312 | int slot; |
34a38218 | 2313 | u32 i; |
00ec4c51 CM |
2314 | int push_space = 0; |
2315 | int push_items = 0; | |
0783fcfc | 2316 | struct btrfs_item *item; |
34a38218 | 2317 | u32 nr; |
7518a238 | 2318 | u32 right_nritems; |
5f39d397 | 2319 | u32 data_end; |
db94535d | 2320 | u32 this_item_size; |
00ec4c51 | 2321 | |
34a38218 CM |
2322 | if (empty) |
2323 | nr = 0; | |
2324 | else | |
99d8f83c | 2325 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2326 | |
31840ae1 | 2327 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2328 | push_space += data_size; |
31840ae1 | 2329 | |
44871b1b | 2330 | slot = path->slots[1]; |
34a38218 CM |
2331 | i = left_nritems - 1; |
2332 | while (i >= nr) { | |
5f39d397 | 2333 | item = btrfs_item_nr(left, i); |
db94535d | 2334 | |
31840ae1 ZY |
2335 | if (!empty && push_items > 0) { |
2336 | if (path->slots[0] > i) | |
2337 | break; | |
2338 | if (path->slots[0] == i) { | |
2339 | int space = btrfs_leaf_free_space(root, left); | |
2340 | if (space + push_space * 2 > free_space) | |
2341 | break; | |
2342 | } | |
2343 | } | |
2344 | ||
00ec4c51 | 2345 | if (path->slots[0] == i) |
87b29b20 | 2346 | push_space += data_size; |
db94535d | 2347 | |
db94535d CM |
2348 | this_item_size = btrfs_item_size(left, item); |
2349 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2350 | break; |
31840ae1 | 2351 | |
00ec4c51 | 2352 | push_items++; |
db94535d | 2353 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2354 | if (i == 0) |
2355 | break; | |
2356 | i--; | |
db94535d | 2357 | } |
5f39d397 | 2358 | |
925baedd CM |
2359 | if (push_items == 0) |
2360 | goto out_unlock; | |
5f39d397 | 2361 | |
34a38218 | 2362 | if (!empty && push_items == left_nritems) |
a429e513 | 2363 | WARN_ON(1); |
5f39d397 | 2364 | |
00ec4c51 | 2365 | /* push left to right */ |
5f39d397 | 2366 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2367 | |
5f39d397 | 2368 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2369 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2370 | |
00ec4c51 | 2371 | /* make room in the right data area */ |
5f39d397 CM |
2372 | data_end = leaf_data_end(root, right); |
2373 | memmove_extent_buffer(right, | |
2374 | btrfs_leaf_data(right) + data_end - push_space, | |
2375 | btrfs_leaf_data(right) + data_end, | |
2376 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2377 | ||
00ec4c51 | 2378 | /* copy from the left data area */ |
5f39d397 | 2379 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2380 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2381 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2382 | push_space); | |
5f39d397 CM |
2383 | |
2384 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2385 | btrfs_item_nr_offset(0), | |
2386 | right_nritems * sizeof(struct btrfs_item)); | |
2387 | ||
00ec4c51 | 2388 | /* copy the items from left to right */ |
5f39d397 CM |
2389 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2390 | btrfs_item_nr_offset(left_nritems - push_items), | |
2391 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2392 | |
2393 | /* update the item pointers */ | |
7518a238 | 2394 | right_nritems += push_items; |
5f39d397 | 2395 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2396 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2397 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2398 | item = btrfs_item_nr(right, i); |
db94535d CM |
2399 | push_space -= btrfs_item_size(right, item); |
2400 | btrfs_set_item_offset(right, item, push_space); | |
2401 | } | |
2402 | ||
7518a238 | 2403 | left_nritems -= push_items; |
5f39d397 | 2404 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2405 | |
34a38218 CM |
2406 | if (left_nritems) |
2407 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
2408 | else |
2409 | clean_tree_block(trans, root, left); | |
2410 | ||
5f39d397 | 2411 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2412 | |
5f39d397 CM |
2413 | btrfs_item_key(right, &disk_key, 0); |
2414 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2415 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2416 | |
00ec4c51 | 2417 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2418 | if (path->slots[0] >= left_nritems) { |
2419 | path->slots[0] -= left_nritems; | |
925baedd CM |
2420 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2421 | clean_tree_block(trans, root, path->nodes[0]); | |
2422 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2423 | free_extent_buffer(path->nodes[0]); |
2424 | path->nodes[0] = right; | |
00ec4c51 CM |
2425 | path->slots[1] += 1; |
2426 | } else { | |
925baedd | 2427 | btrfs_tree_unlock(right); |
5f39d397 | 2428 | free_extent_buffer(right); |
00ec4c51 CM |
2429 | } |
2430 | return 0; | |
925baedd CM |
2431 | |
2432 | out_unlock: | |
2433 | btrfs_tree_unlock(right); | |
2434 | free_extent_buffer(right); | |
2435 | return 1; | |
00ec4c51 | 2436 | } |
925baedd | 2437 | |
44871b1b CM |
2438 | /* |
2439 | * push some data in the path leaf to the right, trying to free up at | |
2440 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2441 | * | |
2442 | * returns 1 if the push failed because the other node didn't have enough | |
2443 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
2444 | * |
2445 | * this will push starting from min_slot to the end of the leaf. It won't | |
2446 | * push any slot lower than min_slot | |
44871b1b CM |
2447 | */ |
2448 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2449 | *root, struct btrfs_path *path, |
2450 | int min_data_size, int data_size, | |
2451 | int empty, u32 min_slot) | |
44871b1b CM |
2452 | { |
2453 | struct extent_buffer *left = path->nodes[0]; | |
2454 | struct extent_buffer *right; | |
2455 | struct extent_buffer *upper; | |
2456 | int slot; | |
2457 | int free_space; | |
2458 | u32 left_nritems; | |
2459 | int ret; | |
2460 | ||
2461 | if (!path->nodes[1]) | |
2462 | return 1; | |
2463 | ||
2464 | slot = path->slots[1]; | |
2465 | upper = path->nodes[1]; | |
2466 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2467 | return 1; | |
2468 | ||
2469 | btrfs_assert_tree_locked(path->nodes[1]); | |
2470 | ||
2471 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
2472 | if (right == NULL) |
2473 | return 1; | |
2474 | ||
44871b1b CM |
2475 | btrfs_tree_lock(right); |
2476 | btrfs_set_lock_blocking(right); | |
2477 | ||
2478 | free_space = btrfs_leaf_free_space(root, right); | |
2479 | if (free_space < data_size) | |
2480 | goto out_unlock; | |
2481 | ||
2482 | /* cow and double check */ | |
2483 | ret = btrfs_cow_block(trans, root, right, upper, | |
2484 | slot + 1, &right); | |
2485 | if (ret) | |
2486 | goto out_unlock; | |
2487 | ||
2488 | free_space = btrfs_leaf_free_space(root, right); | |
2489 | if (free_space < data_size) | |
2490 | goto out_unlock; | |
2491 | ||
2492 | left_nritems = btrfs_header_nritems(left); | |
2493 | if (left_nritems == 0) | |
2494 | goto out_unlock; | |
2495 | ||
99d8f83c CM |
2496 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
2497 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
2498 | out_unlock: |
2499 | btrfs_tree_unlock(right); | |
2500 | free_extent_buffer(right); | |
2501 | return 1; | |
2502 | } | |
2503 | ||
74123bd7 CM |
2504 | /* |
2505 | * push some data in the path leaf to the left, trying to free up at | |
2506 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2507 | * |
2508 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2509 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
2510 | * items | |
74123bd7 | 2511 | */ |
44871b1b CM |
2512 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
2513 | struct btrfs_root *root, | |
2514 | struct btrfs_path *path, int data_size, | |
2515 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
2516 | int free_space, u32 right_nritems, |
2517 | u32 max_slot) | |
be0e5c09 | 2518 | { |
5f39d397 CM |
2519 | struct btrfs_disk_key disk_key; |
2520 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 2521 | int i; |
be0e5c09 CM |
2522 | int push_space = 0; |
2523 | int push_items = 0; | |
0783fcfc | 2524 | struct btrfs_item *item; |
7518a238 | 2525 | u32 old_left_nritems; |
34a38218 | 2526 | u32 nr; |
aa5d6bed | 2527 | int ret = 0; |
db94535d CM |
2528 | u32 this_item_size; |
2529 | u32 old_left_item_size; | |
be0e5c09 | 2530 | |
34a38218 | 2531 | if (empty) |
99d8f83c | 2532 | nr = min(right_nritems, max_slot); |
34a38218 | 2533 | else |
99d8f83c | 2534 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
2535 | |
2536 | for (i = 0; i < nr; i++) { | |
5f39d397 | 2537 | item = btrfs_item_nr(right, i); |
db94535d | 2538 | |
31840ae1 ZY |
2539 | if (!empty && push_items > 0) { |
2540 | if (path->slots[0] < i) | |
2541 | break; | |
2542 | if (path->slots[0] == i) { | |
2543 | int space = btrfs_leaf_free_space(root, right); | |
2544 | if (space + push_space * 2 > free_space) | |
2545 | break; | |
2546 | } | |
2547 | } | |
2548 | ||
be0e5c09 | 2549 | if (path->slots[0] == i) |
87b29b20 | 2550 | push_space += data_size; |
db94535d CM |
2551 | |
2552 | this_item_size = btrfs_item_size(right, item); | |
2553 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 2554 | break; |
db94535d | 2555 | |
be0e5c09 | 2556 | push_items++; |
db94535d CM |
2557 | push_space += this_item_size + sizeof(*item); |
2558 | } | |
2559 | ||
be0e5c09 | 2560 | if (push_items == 0) { |
925baedd CM |
2561 | ret = 1; |
2562 | goto out; | |
be0e5c09 | 2563 | } |
34a38218 | 2564 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 2565 | WARN_ON(1); |
5f39d397 | 2566 | |
be0e5c09 | 2567 | /* push data from right to left */ |
5f39d397 CM |
2568 | copy_extent_buffer(left, right, |
2569 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
2570 | btrfs_item_nr_offset(0), | |
2571 | push_items * sizeof(struct btrfs_item)); | |
2572 | ||
123abc88 | 2573 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 2574 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
2575 | |
2576 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
2577 | leaf_data_end(root, left) - push_space, |
2578 | btrfs_leaf_data(right) + | |
5f39d397 | 2579 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 2580 | push_space); |
5f39d397 | 2581 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 2582 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 2583 | |
db94535d | 2584 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 2585 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 2586 | u32 ioff; |
db94535d | 2587 | |
5f39d397 | 2588 | item = btrfs_item_nr(left, i); |
db94535d | 2589 | |
5f39d397 CM |
2590 | ioff = btrfs_item_offset(left, item); |
2591 | btrfs_set_item_offset(left, item, | |
db94535d | 2592 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size)); |
be0e5c09 | 2593 | } |
5f39d397 | 2594 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
2595 | |
2596 | /* fixup right node */ | |
34a38218 | 2597 | if (push_items > right_nritems) { |
d397712b CM |
2598 | printk(KERN_CRIT "push items %d nr %u\n", push_items, |
2599 | right_nritems); | |
34a38218 CM |
2600 | WARN_ON(1); |
2601 | } | |
2602 | ||
2603 | if (push_items < right_nritems) { | |
2604 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2605 | leaf_data_end(root, right); | |
2606 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
2607 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
2608 | btrfs_leaf_data(right) + | |
2609 | leaf_data_end(root, right), push_space); | |
2610 | ||
2611 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
2612 | btrfs_item_nr_offset(push_items), |
2613 | (btrfs_header_nritems(right) - push_items) * | |
2614 | sizeof(struct btrfs_item)); | |
34a38218 | 2615 | } |
eef1c494 Y |
2616 | right_nritems -= push_items; |
2617 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 2618 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
2619 | for (i = 0; i < right_nritems; i++) { |
2620 | item = btrfs_item_nr(right, i); | |
db94535d | 2621 | |
db94535d CM |
2622 | push_space = push_space - btrfs_item_size(right, item); |
2623 | btrfs_set_item_offset(right, item, push_space); | |
2624 | } | |
eb60ceac | 2625 | |
5f39d397 | 2626 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
2627 | if (right_nritems) |
2628 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
2629 | else |
2630 | clean_tree_block(trans, root, right); | |
098f59c2 | 2631 | |
5f39d397 | 2632 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 2633 | fixup_low_keys(trans, root, path, &disk_key, 1); |
be0e5c09 CM |
2634 | |
2635 | /* then fixup the leaf pointer in the path */ | |
2636 | if (path->slots[0] < push_items) { | |
2637 | path->slots[0] += old_left_nritems; | |
925baedd | 2638 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2639 | free_extent_buffer(path->nodes[0]); |
2640 | path->nodes[0] = left; | |
be0e5c09 CM |
2641 | path->slots[1] -= 1; |
2642 | } else { | |
925baedd | 2643 | btrfs_tree_unlock(left); |
5f39d397 | 2644 | free_extent_buffer(left); |
be0e5c09 CM |
2645 | path->slots[0] -= push_items; |
2646 | } | |
eb60ceac | 2647 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 2648 | return ret; |
925baedd CM |
2649 | out: |
2650 | btrfs_tree_unlock(left); | |
2651 | free_extent_buffer(left); | |
2652 | return ret; | |
be0e5c09 CM |
2653 | } |
2654 | ||
44871b1b CM |
2655 | /* |
2656 | * push some data in the path leaf to the left, trying to free up at | |
2657 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2658 | * |
2659 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2660 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
2661 | * items | |
44871b1b CM |
2662 | */ |
2663 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2664 | *root, struct btrfs_path *path, int min_data_size, |
2665 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
2666 | { |
2667 | struct extent_buffer *right = path->nodes[0]; | |
2668 | struct extent_buffer *left; | |
2669 | int slot; | |
2670 | int free_space; | |
2671 | u32 right_nritems; | |
2672 | int ret = 0; | |
2673 | ||
2674 | slot = path->slots[1]; | |
2675 | if (slot == 0) | |
2676 | return 1; | |
2677 | if (!path->nodes[1]) | |
2678 | return 1; | |
2679 | ||
2680 | right_nritems = btrfs_header_nritems(right); | |
2681 | if (right_nritems == 0) | |
2682 | return 1; | |
2683 | ||
2684 | btrfs_assert_tree_locked(path->nodes[1]); | |
2685 | ||
2686 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
2687 | if (left == NULL) |
2688 | return 1; | |
2689 | ||
44871b1b CM |
2690 | btrfs_tree_lock(left); |
2691 | btrfs_set_lock_blocking(left); | |
2692 | ||
2693 | free_space = btrfs_leaf_free_space(root, left); | |
2694 | if (free_space < data_size) { | |
2695 | ret = 1; | |
2696 | goto out; | |
2697 | } | |
2698 | ||
2699 | /* cow and double check */ | |
2700 | ret = btrfs_cow_block(trans, root, left, | |
2701 | path->nodes[1], slot - 1, &left); | |
2702 | if (ret) { | |
2703 | /* we hit -ENOSPC, but it isn't fatal here */ | |
2704 | ret = 1; | |
2705 | goto out; | |
2706 | } | |
2707 | ||
2708 | free_space = btrfs_leaf_free_space(root, left); | |
2709 | if (free_space < data_size) { | |
2710 | ret = 1; | |
2711 | goto out; | |
2712 | } | |
2713 | ||
99d8f83c CM |
2714 | return __push_leaf_left(trans, root, path, min_data_size, |
2715 | empty, left, free_space, right_nritems, | |
2716 | max_slot); | |
44871b1b CM |
2717 | out: |
2718 | btrfs_tree_unlock(left); | |
2719 | free_extent_buffer(left); | |
2720 | return ret; | |
2721 | } | |
2722 | ||
2723 | /* | |
2724 | * split the path's leaf in two, making sure there is at least data_size | |
2725 | * available for the resulting leaf level of the path. | |
44871b1b | 2726 | */ |
143bede5 JM |
2727 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2728 | struct btrfs_root *root, | |
2729 | struct btrfs_path *path, | |
2730 | struct extent_buffer *l, | |
2731 | struct extent_buffer *right, | |
2732 | int slot, int mid, int nritems) | |
44871b1b CM |
2733 | { |
2734 | int data_copy_size; | |
2735 | int rt_data_off; | |
2736 | int i; | |
44871b1b CM |
2737 | struct btrfs_disk_key disk_key; |
2738 | ||
2739 | nritems = nritems - mid; | |
2740 | btrfs_set_header_nritems(right, nritems); | |
2741 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
2742 | ||
2743 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
2744 | btrfs_item_nr_offset(mid), | |
2745 | nritems * sizeof(struct btrfs_item)); | |
2746 | ||
2747 | copy_extent_buffer(right, l, | |
2748 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
2749 | data_copy_size, btrfs_leaf_data(l) + | |
2750 | leaf_data_end(root, l), data_copy_size); | |
2751 | ||
2752 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
2753 | btrfs_item_end_nr(l, mid); | |
2754 | ||
2755 | for (i = 0; i < nritems; i++) { | |
2756 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
2757 | u32 ioff; | |
2758 | ||
44871b1b CM |
2759 | ioff = btrfs_item_offset(right, item); |
2760 | btrfs_set_item_offset(right, item, ioff + rt_data_off); | |
2761 | } | |
2762 | ||
44871b1b | 2763 | btrfs_set_header_nritems(l, mid); |
44871b1b | 2764 | btrfs_item_key(right, &disk_key, 0); |
143bede5 JM |
2765 | insert_ptr(trans, root, path, &disk_key, right->start, |
2766 | path->slots[1] + 1, 1); | |
44871b1b CM |
2767 | |
2768 | btrfs_mark_buffer_dirty(right); | |
2769 | btrfs_mark_buffer_dirty(l); | |
2770 | BUG_ON(path->slots[0] != slot); | |
2771 | ||
44871b1b CM |
2772 | if (mid <= slot) { |
2773 | btrfs_tree_unlock(path->nodes[0]); | |
2774 | free_extent_buffer(path->nodes[0]); | |
2775 | path->nodes[0] = right; | |
2776 | path->slots[0] -= mid; | |
2777 | path->slots[1] += 1; | |
2778 | } else { | |
2779 | btrfs_tree_unlock(right); | |
2780 | free_extent_buffer(right); | |
2781 | } | |
2782 | ||
2783 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
2784 | } |
2785 | ||
99d8f83c CM |
2786 | /* |
2787 | * double splits happen when we need to insert a big item in the middle | |
2788 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
2789 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
2790 | * A B C | |
2791 | * | |
2792 | * We avoid this by trying to push the items on either side of our target | |
2793 | * into the adjacent leaves. If all goes well we can avoid the double split | |
2794 | * completely. | |
2795 | */ | |
2796 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
2797 | struct btrfs_root *root, | |
2798 | struct btrfs_path *path, | |
2799 | int data_size) | |
2800 | { | |
2801 | int ret; | |
2802 | int progress = 0; | |
2803 | int slot; | |
2804 | u32 nritems; | |
2805 | ||
2806 | slot = path->slots[0]; | |
2807 | ||
2808 | /* | |
2809 | * try to push all the items after our slot into the | |
2810 | * right leaf | |
2811 | */ | |
2812 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
2813 | if (ret < 0) | |
2814 | return ret; | |
2815 | ||
2816 | if (ret == 0) | |
2817 | progress++; | |
2818 | ||
2819 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2820 | /* | |
2821 | * our goal is to get our slot at the start or end of a leaf. If | |
2822 | * we've done so we're done | |
2823 | */ | |
2824 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
2825 | return 0; | |
2826 | ||
2827 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
2828 | return 0; | |
2829 | ||
2830 | /* try to push all the items before our slot into the next leaf */ | |
2831 | slot = path->slots[0]; | |
2832 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
2833 | if (ret < 0) | |
2834 | return ret; | |
2835 | ||
2836 | if (ret == 0) | |
2837 | progress++; | |
2838 | ||
2839 | if (progress) | |
2840 | return 0; | |
2841 | return 1; | |
2842 | } | |
2843 | ||
74123bd7 CM |
2844 | /* |
2845 | * split the path's leaf in two, making sure there is at least data_size | |
2846 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
2847 | * |
2848 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 2849 | */ |
e02119d5 CM |
2850 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
2851 | struct btrfs_root *root, | |
2852 | struct btrfs_key *ins_key, | |
2853 | struct btrfs_path *path, int data_size, | |
2854 | int extend) | |
be0e5c09 | 2855 | { |
5d4f98a2 | 2856 | struct btrfs_disk_key disk_key; |
5f39d397 | 2857 | struct extent_buffer *l; |
7518a238 | 2858 | u32 nritems; |
eb60ceac CM |
2859 | int mid; |
2860 | int slot; | |
5f39d397 | 2861 | struct extent_buffer *right; |
d4dbff95 | 2862 | int ret = 0; |
aa5d6bed | 2863 | int wret; |
5d4f98a2 | 2864 | int split; |
cc0c5538 | 2865 | int num_doubles = 0; |
99d8f83c | 2866 | int tried_avoid_double = 0; |
aa5d6bed | 2867 | |
a5719521 YZ |
2868 | l = path->nodes[0]; |
2869 | slot = path->slots[0]; | |
2870 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
2871 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
2872 | return -EOVERFLOW; | |
2873 | ||
40689478 | 2874 | /* first try to make some room by pushing left and right */ |
99d8f83c CM |
2875 | if (data_size) { |
2876 | wret = push_leaf_right(trans, root, path, data_size, | |
2877 | data_size, 0, 0); | |
d397712b | 2878 | if (wret < 0) |
eaee50e8 | 2879 | return wret; |
3685f791 | 2880 | if (wret) { |
99d8f83c CM |
2881 | wret = push_leaf_left(trans, root, path, data_size, |
2882 | data_size, 0, (u32)-1); | |
3685f791 CM |
2883 | if (wret < 0) |
2884 | return wret; | |
2885 | } | |
2886 | l = path->nodes[0]; | |
aa5d6bed | 2887 | |
3685f791 | 2888 | /* did the pushes work? */ |
87b29b20 | 2889 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 2890 | return 0; |
3326d1b0 | 2891 | } |
aa5d6bed | 2892 | |
5c680ed6 | 2893 | if (!path->nodes[1]) { |
e089f05c | 2894 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
2895 | if (ret) |
2896 | return ret; | |
2897 | } | |
cc0c5538 | 2898 | again: |
5d4f98a2 | 2899 | split = 1; |
cc0c5538 | 2900 | l = path->nodes[0]; |
eb60ceac | 2901 | slot = path->slots[0]; |
5f39d397 | 2902 | nritems = btrfs_header_nritems(l); |
d397712b | 2903 | mid = (nritems + 1) / 2; |
54aa1f4d | 2904 | |
5d4f98a2 YZ |
2905 | if (mid <= slot) { |
2906 | if (nritems == 1 || | |
2907 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
2908 | BTRFS_LEAF_DATA_SIZE(root)) { | |
2909 | if (slot >= nritems) { | |
2910 | split = 0; | |
2911 | } else { | |
2912 | mid = slot; | |
2913 | if (mid != nritems && | |
2914 | leaf_space_used(l, mid, nritems - mid) + | |
2915 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
2916 | if (data_size && !tried_avoid_double) |
2917 | goto push_for_double; | |
5d4f98a2 YZ |
2918 | split = 2; |
2919 | } | |
2920 | } | |
2921 | } | |
2922 | } else { | |
2923 | if (leaf_space_used(l, 0, mid) + data_size > | |
2924 | BTRFS_LEAF_DATA_SIZE(root)) { | |
2925 | if (!extend && data_size && slot == 0) { | |
2926 | split = 0; | |
2927 | } else if ((extend || !data_size) && slot == 0) { | |
2928 | mid = 1; | |
2929 | } else { | |
2930 | mid = slot; | |
2931 | if (mid != nritems && | |
2932 | leaf_space_used(l, mid, nritems - mid) + | |
2933 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
2934 | if (data_size && !tried_avoid_double) |
2935 | goto push_for_double; | |
5d4f98a2 YZ |
2936 | split = 2 ; |
2937 | } | |
2938 | } | |
2939 | } | |
2940 | } | |
2941 | ||
2942 | if (split == 0) | |
2943 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
2944 | else | |
2945 | btrfs_item_key(l, &disk_key, mid); | |
2946 | ||
2947 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 2948 | root->root_key.objectid, |
66d7e7f0 | 2949 | &disk_key, 0, l->start, 0, 0); |
f0486c68 | 2950 | if (IS_ERR(right)) |
5f39d397 | 2951 | return PTR_ERR(right); |
f0486c68 YZ |
2952 | |
2953 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
2954 | |
2955 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 2956 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 2957 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 2958 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2959 | btrfs_set_header_owner(right, root->root_key.objectid); |
2960 | btrfs_set_header_level(right, 0); | |
2961 | write_extent_buffer(right, root->fs_info->fsid, | |
2962 | (unsigned long)btrfs_header_fsid(right), | |
2963 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2964 | |
2965 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
2966 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
2967 | BTRFS_UUID_SIZE); | |
44871b1b | 2968 | |
5d4f98a2 YZ |
2969 | if (split == 0) { |
2970 | if (mid <= slot) { | |
2971 | btrfs_set_header_nritems(right, 0); | |
143bede5 JM |
2972 | insert_ptr(trans, root, path, &disk_key, right->start, |
2973 | path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
2974 | btrfs_tree_unlock(path->nodes[0]); |
2975 | free_extent_buffer(path->nodes[0]); | |
2976 | path->nodes[0] = right; | |
2977 | path->slots[0] = 0; | |
2978 | path->slots[1] += 1; | |
2979 | } else { | |
2980 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 2981 | insert_ptr(trans, root, path, &disk_key, right->start, |
5d4f98a2 | 2982 | path->slots[1], 1); |
5d4f98a2 YZ |
2983 | btrfs_tree_unlock(path->nodes[0]); |
2984 | free_extent_buffer(path->nodes[0]); | |
2985 | path->nodes[0] = right; | |
2986 | path->slots[0] = 0; | |
143bede5 JM |
2987 | if (path->slots[1] == 0) |
2988 | fixup_low_keys(trans, root, path, | |
2989 | &disk_key, 1); | |
d4dbff95 | 2990 | } |
5d4f98a2 YZ |
2991 | btrfs_mark_buffer_dirty(right); |
2992 | return ret; | |
d4dbff95 | 2993 | } |
74123bd7 | 2994 | |
143bede5 | 2995 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 2996 | |
5d4f98a2 | 2997 | if (split == 2) { |
cc0c5538 CM |
2998 | BUG_ON(num_doubles != 0); |
2999 | num_doubles++; | |
3000 | goto again; | |
a429e513 | 3001 | } |
44871b1b | 3002 | |
143bede5 | 3003 | return 0; |
99d8f83c CM |
3004 | |
3005 | push_for_double: | |
3006 | push_for_double_split(trans, root, path, data_size); | |
3007 | tried_avoid_double = 1; | |
3008 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3009 | return 0; | |
3010 | goto again; | |
be0e5c09 CM |
3011 | } |
3012 | ||
ad48fd75 YZ |
3013 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3014 | struct btrfs_root *root, | |
3015 | struct btrfs_path *path, int ins_len) | |
459931ec | 3016 | { |
ad48fd75 | 3017 | struct btrfs_key key; |
459931ec | 3018 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3019 | struct btrfs_file_extent_item *fi; |
3020 | u64 extent_len = 0; | |
3021 | u32 item_size; | |
3022 | int ret; | |
459931ec CM |
3023 | |
3024 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3025 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3026 | ||
3027 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3028 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3029 | ||
3030 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
3031 | return 0; | |
459931ec CM |
3032 | |
3033 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
3034 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3035 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3036 | struct btrfs_file_extent_item); | |
3037 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3038 | } | |
b3b4aa74 | 3039 | btrfs_release_path(path); |
459931ec | 3040 | |
459931ec | 3041 | path->keep_locks = 1; |
ad48fd75 YZ |
3042 | path->search_for_split = 1; |
3043 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3044 | path->search_for_split = 0; |
ad48fd75 YZ |
3045 | if (ret < 0) |
3046 | goto err; | |
459931ec | 3047 | |
ad48fd75 YZ |
3048 | ret = -EAGAIN; |
3049 | leaf = path->nodes[0]; | |
459931ec | 3050 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
3051 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
3052 | goto err; | |
3053 | ||
109f6aef CM |
3054 | /* the leaf has changed, it now has room. return now */ |
3055 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
3056 | goto err; | |
3057 | ||
ad48fd75 YZ |
3058 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3059 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3060 | struct btrfs_file_extent_item); | |
3061 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3062 | goto err; | |
459931ec CM |
3063 | } |
3064 | ||
b9473439 | 3065 | btrfs_set_path_blocking(path); |
ad48fd75 | 3066 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3067 | if (ret) |
3068 | goto err; | |
459931ec | 3069 | |
ad48fd75 | 3070 | path->keep_locks = 0; |
b9473439 | 3071 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3072 | return 0; |
3073 | err: | |
3074 | path->keep_locks = 0; | |
3075 | return ret; | |
3076 | } | |
3077 | ||
3078 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
3079 | struct btrfs_root *root, | |
3080 | struct btrfs_path *path, | |
3081 | struct btrfs_key *new_key, | |
3082 | unsigned long split_offset) | |
3083 | { | |
3084 | struct extent_buffer *leaf; | |
3085 | struct btrfs_item *item; | |
3086 | struct btrfs_item *new_item; | |
3087 | int slot; | |
3088 | char *buf; | |
3089 | u32 nritems; | |
3090 | u32 item_size; | |
3091 | u32 orig_offset; | |
3092 | struct btrfs_disk_key disk_key; | |
3093 | ||
b9473439 CM |
3094 | leaf = path->nodes[0]; |
3095 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
3096 | ||
b4ce94de CM |
3097 | btrfs_set_path_blocking(path); |
3098 | ||
459931ec CM |
3099 | item = btrfs_item_nr(leaf, path->slots[0]); |
3100 | orig_offset = btrfs_item_offset(leaf, item); | |
3101 | item_size = btrfs_item_size(leaf, item); | |
3102 | ||
459931ec | 3103 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3104 | if (!buf) |
3105 | return -ENOMEM; | |
3106 | ||
459931ec CM |
3107 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3108 | path->slots[0]), item_size); | |
459931ec | 3109 | |
ad48fd75 | 3110 | slot = path->slots[0] + 1; |
459931ec | 3111 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3112 | if (slot != nritems) { |
3113 | /* shift the items */ | |
3114 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3115 | btrfs_item_nr_offset(slot), |
3116 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3117 | } |
3118 | ||
3119 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3120 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3121 | ||
3122 | new_item = btrfs_item_nr(leaf, slot); | |
3123 | ||
3124 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3125 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3126 | ||
3127 | btrfs_set_item_offset(leaf, item, | |
3128 | orig_offset + item_size - split_offset); | |
3129 | btrfs_set_item_size(leaf, item, split_offset); | |
3130 | ||
3131 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3132 | ||
3133 | /* write the data for the start of the original item */ | |
3134 | write_extent_buffer(leaf, buf, | |
3135 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3136 | split_offset); | |
3137 | ||
3138 | /* write the data for the new item */ | |
3139 | write_extent_buffer(leaf, buf + split_offset, | |
3140 | btrfs_item_ptr_offset(leaf, slot), | |
3141 | item_size - split_offset); | |
3142 | btrfs_mark_buffer_dirty(leaf); | |
3143 | ||
ad48fd75 | 3144 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 3145 | kfree(buf); |
ad48fd75 YZ |
3146 | return 0; |
3147 | } | |
3148 | ||
3149 | /* | |
3150 | * This function splits a single item into two items, | |
3151 | * giving 'new_key' to the new item and splitting the | |
3152 | * old one at split_offset (from the start of the item). | |
3153 | * | |
3154 | * The path may be released by this operation. After | |
3155 | * the split, the path is pointing to the old item. The | |
3156 | * new item is going to be in the same node as the old one. | |
3157 | * | |
3158 | * Note, the item being split must be smaller enough to live alone on | |
3159 | * a tree block with room for one extra struct btrfs_item | |
3160 | * | |
3161 | * This allows us to split the item in place, keeping a lock on the | |
3162 | * leaf the entire time. | |
3163 | */ | |
3164 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3165 | struct btrfs_root *root, | |
3166 | struct btrfs_path *path, | |
3167 | struct btrfs_key *new_key, | |
3168 | unsigned long split_offset) | |
3169 | { | |
3170 | int ret; | |
3171 | ret = setup_leaf_for_split(trans, root, path, | |
3172 | sizeof(struct btrfs_item)); | |
3173 | if (ret) | |
3174 | return ret; | |
3175 | ||
3176 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
3177 | return ret; |
3178 | } | |
3179 | ||
ad48fd75 YZ |
3180 | /* |
3181 | * This function duplicate a item, giving 'new_key' to the new item. | |
3182 | * It guarantees both items live in the same tree leaf and the new item | |
3183 | * is contiguous with the original item. | |
3184 | * | |
3185 | * This allows us to split file extent in place, keeping a lock on the | |
3186 | * leaf the entire time. | |
3187 | */ | |
3188 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
3189 | struct btrfs_root *root, | |
3190 | struct btrfs_path *path, | |
3191 | struct btrfs_key *new_key) | |
3192 | { | |
3193 | struct extent_buffer *leaf; | |
3194 | int ret; | |
3195 | u32 item_size; | |
3196 | ||
3197 | leaf = path->nodes[0]; | |
3198 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3199 | ret = setup_leaf_for_split(trans, root, path, | |
3200 | item_size + sizeof(struct btrfs_item)); | |
3201 | if (ret) | |
3202 | return ret; | |
3203 | ||
3204 | path->slots[0]++; | |
143bede5 JM |
3205 | setup_items_for_insert(trans, root, path, new_key, &item_size, |
3206 | item_size, item_size + | |
3207 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
3208 | leaf = path->nodes[0]; |
3209 | memcpy_extent_buffer(leaf, | |
3210 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3211 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
3212 | item_size); | |
3213 | return 0; | |
3214 | } | |
3215 | ||
d352ac68 CM |
3216 | /* |
3217 | * make the item pointed to by the path smaller. new_size indicates | |
3218 | * how small to make it, and from_end tells us if we just chop bytes | |
3219 | * off the end of the item or if we shift the item to chop bytes off | |
3220 | * the front. | |
3221 | */ | |
143bede5 JM |
3222 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3223 | struct btrfs_root *root, | |
3224 | struct btrfs_path *path, | |
3225 | u32 new_size, int from_end) | |
b18c6685 | 3226 | { |
b18c6685 | 3227 | int slot; |
5f39d397 CM |
3228 | struct extent_buffer *leaf; |
3229 | struct btrfs_item *item; | |
b18c6685 CM |
3230 | u32 nritems; |
3231 | unsigned int data_end; | |
3232 | unsigned int old_data_start; | |
3233 | unsigned int old_size; | |
3234 | unsigned int size_diff; | |
3235 | int i; | |
3236 | ||
5f39d397 | 3237 | leaf = path->nodes[0]; |
179e29e4 CM |
3238 | slot = path->slots[0]; |
3239 | ||
3240 | old_size = btrfs_item_size_nr(leaf, slot); | |
3241 | if (old_size == new_size) | |
143bede5 | 3242 | return; |
b18c6685 | 3243 | |
5f39d397 | 3244 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
3245 | data_end = leaf_data_end(root, leaf); |
3246 | ||
5f39d397 | 3247 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3248 | |
b18c6685 CM |
3249 | size_diff = old_size - new_size; |
3250 | ||
3251 | BUG_ON(slot < 0); | |
3252 | BUG_ON(slot >= nritems); | |
3253 | ||
3254 | /* | |
3255 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3256 | */ | |
3257 | /* first correct the data pointers */ | |
3258 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3259 | u32 ioff; |
3260 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3261 | |
5f39d397 CM |
3262 | ioff = btrfs_item_offset(leaf, item); |
3263 | btrfs_set_item_offset(leaf, item, ioff + size_diff); | |
b18c6685 | 3264 | } |
db94535d | 3265 | |
b18c6685 | 3266 | /* shift the data */ |
179e29e4 CM |
3267 | if (from_end) { |
3268 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3269 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3270 | data_end, old_data_start + new_size - data_end); | |
3271 | } else { | |
3272 | struct btrfs_disk_key disk_key; | |
3273 | u64 offset; | |
3274 | ||
3275 | btrfs_item_key(leaf, &disk_key, slot); | |
3276 | ||
3277 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3278 | unsigned long ptr; | |
3279 | struct btrfs_file_extent_item *fi; | |
3280 | ||
3281 | fi = btrfs_item_ptr(leaf, slot, | |
3282 | struct btrfs_file_extent_item); | |
3283 | fi = (struct btrfs_file_extent_item *)( | |
3284 | (unsigned long)fi - size_diff); | |
3285 | ||
3286 | if (btrfs_file_extent_type(leaf, fi) == | |
3287 | BTRFS_FILE_EXTENT_INLINE) { | |
3288 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3289 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
3290 | (unsigned long)fi, |
3291 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
3292 | disk_bytenr)); |
3293 | } | |
3294 | } | |
3295 | ||
3296 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3297 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3298 | data_end, old_data_start - data_end); | |
3299 | ||
3300 | offset = btrfs_disk_key_offset(&disk_key); | |
3301 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3302 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3303 | if (slot == 0) | |
3304 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
3305 | } | |
5f39d397 CM |
3306 | |
3307 | item = btrfs_item_nr(leaf, slot); | |
3308 | btrfs_set_item_size(leaf, item, new_size); | |
3309 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 3310 | |
5f39d397 CM |
3311 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3312 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 3313 | BUG(); |
5f39d397 | 3314 | } |
b18c6685 CM |
3315 | } |
3316 | ||
d352ac68 CM |
3317 | /* |
3318 | * make the item pointed to by the path bigger, data_size is the new size. | |
3319 | */ | |
143bede5 JM |
3320 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
3321 | struct btrfs_root *root, struct btrfs_path *path, | |
3322 | u32 data_size) | |
6567e837 | 3323 | { |
6567e837 | 3324 | int slot; |
5f39d397 CM |
3325 | struct extent_buffer *leaf; |
3326 | struct btrfs_item *item; | |
6567e837 CM |
3327 | u32 nritems; |
3328 | unsigned int data_end; | |
3329 | unsigned int old_data; | |
3330 | unsigned int old_size; | |
3331 | int i; | |
3332 | ||
5f39d397 | 3333 | leaf = path->nodes[0]; |
6567e837 | 3334 | |
5f39d397 | 3335 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
3336 | data_end = leaf_data_end(root, leaf); |
3337 | ||
5f39d397 CM |
3338 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
3339 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3340 | BUG(); |
5f39d397 | 3341 | } |
6567e837 | 3342 | slot = path->slots[0]; |
5f39d397 | 3343 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3344 | |
3345 | BUG_ON(slot < 0); | |
3326d1b0 CM |
3346 | if (slot >= nritems) { |
3347 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
3348 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
3349 | slot, nritems); | |
3326d1b0 CM |
3350 | BUG_ON(1); |
3351 | } | |
6567e837 CM |
3352 | |
3353 | /* | |
3354 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3355 | */ | |
3356 | /* first correct the data pointers */ | |
3357 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3358 | u32 ioff; |
3359 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3360 | |
5f39d397 CM |
3361 | ioff = btrfs_item_offset(leaf, item); |
3362 | btrfs_set_item_offset(leaf, item, ioff - data_size); | |
6567e837 | 3363 | } |
5f39d397 | 3364 | |
6567e837 | 3365 | /* shift the data */ |
5f39d397 | 3366 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
3367 | data_end - data_size, btrfs_leaf_data(leaf) + |
3368 | data_end, old_data - data_end); | |
5f39d397 | 3369 | |
6567e837 | 3370 | data_end = old_data; |
5f39d397 CM |
3371 | old_size = btrfs_item_size_nr(leaf, slot); |
3372 | item = btrfs_item_nr(leaf, slot); | |
3373 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
3374 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 3375 | |
5f39d397 CM |
3376 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3377 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3378 | BUG(); |
5f39d397 | 3379 | } |
6567e837 CM |
3380 | } |
3381 | ||
f3465ca4 JB |
3382 | /* |
3383 | * Given a key and some data, insert items into the tree. | |
3384 | * This does all the path init required, making room in the tree if needed. | |
3385 | * Returns the number of keys that were inserted. | |
3386 | */ | |
3387 | int btrfs_insert_some_items(struct btrfs_trans_handle *trans, | |
3388 | struct btrfs_root *root, | |
3389 | struct btrfs_path *path, | |
3390 | struct btrfs_key *cpu_key, u32 *data_size, | |
3391 | int nr) | |
3392 | { | |
3393 | struct extent_buffer *leaf; | |
3394 | struct btrfs_item *item; | |
3395 | int ret = 0; | |
3396 | int slot; | |
f3465ca4 JB |
3397 | int i; |
3398 | u32 nritems; | |
3399 | u32 total_data = 0; | |
3400 | u32 total_size = 0; | |
3401 | unsigned int data_end; | |
3402 | struct btrfs_disk_key disk_key; | |
3403 | struct btrfs_key found_key; | |
3404 | ||
87b29b20 YZ |
3405 | for (i = 0; i < nr; i++) { |
3406 | if (total_size + data_size[i] + sizeof(struct btrfs_item) > | |
3407 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3408 | break; | |
3409 | nr = i; | |
3410 | } | |
f3465ca4 | 3411 | total_data += data_size[i]; |
87b29b20 YZ |
3412 | total_size += data_size[i] + sizeof(struct btrfs_item); |
3413 | } | |
3414 | BUG_ON(nr == 0); | |
f3465ca4 | 3415 | |
f3465ca4 JB |
3416 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
3417 | if (ret == 0) | |
3418 | return -EEXIST; | |
3419 | if (ret < 0) | |
3420 | goto out; | |
3421 | ||
f3465ca4 JB |
3422 | leaf = path->nodes[0]; |
3423 | ||
3424 | nritems = btrfs_header_nritems(leaf); | |
3425 | data_end = leaf_data_end(root, leaf); | |
3426 | ||
3427 | if (btrfs_leaf_free_space(root, leaf) < total_size) { | |
3428 | for (i = nr; i >= 0; i--) { | |
3429 | total_data -= data_size[i]; | |
3430 | total_size -= data_size[i] + sizeof(struct btrfs_item); | |
3431 | if (total_size < btrfs_leaf_free_space(root, leaf)) | |
3432 | break; | |
3433 | } | |
3434 | nr = i; | |
3435 | } | |
3436 | ||
3437 | slot = path->slots[0]; | |
3438 | BUG_ON(slot < 0); | |
3439 | ||
3440 | if (slot != nritems) { | |
3441 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); | |
3442 | ||
3443 | item = btrfs_item_nr(leaf, slot); | |
3444 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3445 | ||
3446 | /* figure out how many keys we can insert in here */ | |
3447 | total_data = data_size[0]; | |
3448 | for (i = 1; i < nr; i++) { | |
5d4f98a2 | 3449 | if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0) |
f3465ca4 JB |
3450 | break; |
3451 | total_data += data_size[i]; | |
3452 | } | |
3453 | nr = i; | |
3454 | ||
3455 | if (old_data < data_end) { | |
3456 | btrfs_print_leaf(root, leaf); | |
d397712b | 3457 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
f3465ca4 JB |
3458 | slot, old_data, data_end); |
3459 | BUG_ON(1); | |
3460 | } | |
3461 | /* | |
3462 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3463 | */ | |
3464 | /* first correct the data pointers */ | |
f3465ca4 JB |
3465 | for (i = slot; i < nritems; i++) { |
3466 | u32 ioff; | |
3467 | ||
3468 | item = btrfs_item_nr(leaf, i); | |
f3465ca4 JB |
3469 | ioff = btrfs_item_offset(leaf, item); |
3470 | btrfs_set_item_offset(leaf, item, ioff - total_data); | |
3471 | } | |
f3465ca4 JB |
3472 | /* shift the items */ |
3473 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), | |
3474 | btrfs_item_nr_offset(slot), | |
3475 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3476 | ||
3477 | /* shift the data */ | |
3478 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3479 | data_end - total_data, btrfs_leaf_data(leaf) + | |
3480 | data_end, old_data - data_end); | |
3481 | data_end = old_data; | |
3482 | } else { | |
3483 | /* | |
3484 | * this sucks but it has to be done, if we are inserting at | |
3485 | * the end of the leaf only insert 1 of the items, since we | |
3486 | * have no way of knowing whats on the next leaf and we'd have | |
3487 | * to drop our current locks to figure it out | |
3488 | */ | |
3489 | nr = 1; | |
3490 | } | |
3491 | ||
3492 | /* setup the item for the new data */ | |
3493 | for (i = 0; i < nr; i++) { | |
3494 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3495 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3496 | item = btrfs_item_nr(leaf, slot + i); | |
3497 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3498 | data_end -= data_size[i]; | |
3499 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3500 | } | |
3501 | btrfs_set_header_nritems(leaf, nritems + nr); | |
3502 | btrfs_mark_buffer_dirty(leaf); | |
3503 | ||
3504 | ret = 0; | |
3505 | if (slot == 0) { | |
3506 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3507 | fixup_low_keys(trans, root, path, &disk_key, 1); |
f3465ca4 JB |
3508 | } |
3509 | ||
3510 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
3511 | btrfs_print_leaf(root, leaf); | |
3512 | BUG(); | |
3513 | } | |
3514 | out: | |
3515 | if (!ret) | |
3516 | ret = nr; | |
3517 | return ret; | |
3518 | } | |
3519 | ||
74123bd7 | 3520 | /* |
44871b1b CM |
3521 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
3522 | * to save stack depth by doing the bulk of the work in a function | |
3523 | * that doesn't call btrfs_search_slot | |
74123bd7 | 3524 | */ |
143bede5 JM |
3525 | void setup_items_for_insert(struct btrfs_trans_handle *trans, |
3526 | struct btrfs_root *root, struct btrfs_path *path, | |
3527 | struct btrfs_key *cpu_key, u32 *data_size, | |
3528 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 3529 | { |
5f39d397 | 3530 | struct btrfs_item *item; |
9c58309d | 3531 | int i; |
7518a238 | 3532 | u32 nritems; |
be0e5c09 | 3533 | unsigned int data_end; |
e2fa7227 | 3534 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3535 | struct extent_buffer *leaf; |
3536 | int slot; | |
e2fa7227 | 3537 | |
5f39d397 | 3538 | leaf = path->nodes[0]; |
44871b1b | 3539 | slot = path->slots[0]; |
74123bd7 | 3540 | |
5f39d397 | 3541 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 3542 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 3543 | |
f25956cc | 3544 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 3545 | btrfs_print_leaf(root, leaf); |
d397712b | 3546 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 3547 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 3548 | BUG(); |
d4dbff95 | 3549 | } |
5f39d397 | 3550 | |
be0e5c09 | 3551 | if (slot != nritems) { |
5f39d397 | 3552 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 3553 | |
5f39d397 CM |
3554 | if (old_data < data_end) { |
3555 | btrfs_print_leaf(root, leaf); | |
d397712b | 3556 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
3557 | slot, old_data, data_end); |
3558 | BUG_ON(1); | |
3559 | } | |
be0e5c09 CM |
3560 | /* |
3561 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3562 | */ | |
3563 | /* first correct the data pointers */ | |
0783fcfc | 3564 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3565 | u32 ioff; |
db94535d | 3566 | |
5f39d397 CM |
3567 | item = btrfs_item_nr(leaf, i); |
3568 | ioff = btrfs_item_offset(leaf, item); | |
9c58309d | 3569 | btrfs_set_item_offset(leaf, item, ioff - total_data); |
0783fcfc | 3570 | } |
be0e5c09 | 3571 | /* shift the items */ |
9c58309d | 3572 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 3573 | btrfs_item_nr_offset(slot), |
d6025579 | 3574 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
3575 | |
3576 | /* shift the data */ | |
5f39d397 | 3577 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 3578 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 3579 | data_end, old_data - data_end); |
be0e5c09 CM |
3580 | data_end = old_data; |
3581 | } | |
5f39d397 | 3582 | |
62e2749e | 3583 | /* setup the item for the new data */ |
9c58309d CM |
3584 | for (i = 0; i < nr; i++) { |
3585 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3586 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3587 | item = btrfs_item_nr(leaf, slot + i); | |
3588 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3589 | data_end -= data_size[i]; | |
3590 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3591 | } | |
44871b1b | 3592 | |
9c58309d | 3593 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 3594 | |
5a01a2e3 CM |
3595 | if (slot == 0) { |
3596 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3597 | fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 3598 | } |
b9473439 CM |
3599 | btrfs_unlock_up_safe(path, 1); |
3600 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 3601 | |
5f39d397 CM |
3602 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3603 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 3604 | BUG(); |
5f39d397 | 3605 | } |
44871b1b CM |
3606 | } |
3607 | ||
3608 | /* | |
3609 | * Given a key and some data, insert items into the tree. | |
3610 | * This does all the path init required, making room in the tree if needed. | |
3611 | */ | |
3612 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
3613 | struct btrfs_root *root, | |
3614 | struct btrfs_path *path, | |
3615 | struct btrfs_key *cpu_key, u32 *data_size, | |
3616 | int nr) | |
3617 | { | |
44871b1b CM |
3618 | int ret = 0; |
3619 | int slot; | |
3620 | int i; | |
3621 | u32 total_size = 0; | |
3622 | u32 total_data = 0; | |
3623 | ||
3624 | for (i = 0; i < nr; i++) | |
3625 | total_data += data_size[i]; | |
3626 | ||
3627 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
3628 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
3629 | if (ret == 0) | |
3630 | return -EEXIST; | |
3631 | if (ret < 0) | |
143bede5 | 3632 | return ret; |
44871b1b | 3633 | |
44871b1b CM |
3634 | slot = path->slots[0]; |
3635 | BUG_ON(slot < 0); | |
3636 | ||
143bede5 | 3637 | setup_items_for_insert(trans, root, path, cpu_key, data_size, |
44871b1b | 3638 | total_data, total_size, nr); |
143bede5 | 3639 | return 0; |
62e2749e CM |
3640 | } |
3641 | ||
3642 | /* | |
3643 | * Given a key and some data, insert an item into the tree. | |
3644 | * This does all the path init required, making room in the tree if needed. | |
3645 | */ | |
e089f05c CM |
3646 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
3647 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
3648 | data_size) | |
62e2749e CM |
3649 | { |
3650 | int ret = 0; | |
2c90e5d6 | 3651 | struct btrfs_path *path; |
5f39d397 CM |
3652 | struct extent_buffer *leaf; |
3653 | unsigned long ptr; | |
62e2749e | 3654 | |
2c90e5d6 | 3655 | path = btrfs_alloc_path(); |
db5b493a TI |
3656 | if (!path) |
3657 | return -ENOMEM; | |
2c90e5d6 | 3658 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 3659 | if (!ret) { |
5f39d397 CM |
3660 | leaf = path->nodes[0]; |
3661 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3662 | write_extent_buffer(leaf, data, ptr, data_size); | |
3663 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 3664 | } |
2c90e5d6 | 3665 | btrfs_free_path(path); |
aa5d6bed | 3666 | return ret; |
be0e5c09 CM |
3667 | } |
3668 | ||
74123bd7 | 3669 | /* |
5de08d7d | 3670 | * delete the pointer from a given node. |
74123bd7 | 3671 | * |
d352ac68 CM |
3672 | * the tree should have been previously balanced so the deletion does not |
3673 | * empty a node. | |
74123bd7 | 3674 | */ |
143bede5 JM |
3675 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3676 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 3677 | { |
5f39d397 | 3678 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 3679 | u32 nritems; |
be0e5c09 | 3680 | |
5f39d397 | 3681 | nritems = btrfs_header_nritems(parent); |
d397712b | 3682 | if (slot != nritems - 1) { |
5f39d397 CM |
3683 | memmove_extent_buffer(parent, |
3684 | btrfs_node_key_ptr_offset(slot), | |
3685 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
3686 | sizeof(struct btrfs_key_ptr) * |
3687 | (nritems - slot - 1)); | |
bb803951 | 3688 | } |
7518a238 | 3689 | nritems--; |
5f39d397 | 3690 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 3691 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 3692 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 3693 | /* just turn the root into a leaf and break */ |
5f39d397 | 3694 | btrfs_set_header_level(root->node, 0); |
bb803951 | 3695 | } else if (slot == 0) { |
5f39d397 CM |
3696 | struct btrfs_disk_key disk_key; |
3697 | ||
3698 | btrfs_node_key(parent, &disk_key, 0); | |
143bede5 | 3699 | fixup_low_keys(trans, root, path, &disk_key, level + 1); |
be0e5c09 | 3700 | } |
d6025579 | 3701 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
3702 | } |
3703 | ||
323ac95b CM |
3704 | /* |
3705 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 3706 | * path->nodes[1]. |
323ac95b CM |
3707 | * |
3708 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
3709 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
3710 | * | |
3711 | * The path must have already been setup for deleting the leaf, including | |
3712 | * all the proper balancing. path->nodes[1] must be locked. | |
3713 | */ | |
143bede5 JM |
3714 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
3715 | struct btrfs_root *root, | |
3716 | struct btrfs_path *path, | |
3717 | struct extent_buffer *leaf) | |
323ac95b | 3718 | { |
5d4f98a2 | 3719 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
143bede5 | 3720 | del_ptr(trans, root, path, 1, path->slots[1]); |
323ac95b | 3721 | |
4d081c41 CM |
3722 | /* |
3723 | * btrfs_free_extent is expensive, we want to make sure we | |
3724 | * aren't holding any locks when we call it | |
3725 | */ | |
3726 | btrfs_unlock_up_safe(path, 0); | |
3727 | ||
f0486c68 YZ |
3728 | root_sub_used(root, leaf->len); |
3729 | ||
66d7e7f0 | 3730 | btrfs_free_tree_block(trans, root, leaf, 0, 1, 0); |
323ac95b | 3731 | } |
74123bd7 CM |
3732 | /* |
3733 | * delete the item at the leaf level in path. If that empties | |
3734 | * the leaf, remove it from the tree | |
3735 | */ | |
85e21bac CM |
3736 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3737 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 3738 | { |
5f39d397 CM |
3739 | struct extent_buffer *leaf; |
3740 | struct btrfs_item *item; | |
85e21bac CM |
3741 | int last_off; |
3742 | int dsize = 0; | |
aa5d6bed CM |
3743 | int ret = 0; |
3744 | int wret; | |
85e21bac | 3745 | int i; |
7518a238 | 3746 | u32 nritems; |
be0e5c09 | 3747 | |
5f39d397 | 3748 | leaf = path->nodes[0]; |
85e21bac CM |
3749 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
3750 | ||
3751 | for (i = 0; i < nr; i++) | |
3752 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
3753 | ||
5f39d397 | 3754 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 3755 | |
85e21bac | 3756 | if (slot + nr != nritems) { |
123abc88 | 3757 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
3758 | |
3759 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
3760 | data_end + dsize, |
3761 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 3762 | last_off - data_end); |
5f39d397 | 3763 | |
85e21bac | 3764 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 3765 | u32 ioff; |
db94535d | 3766 | |
5f39d397 CM |
3767 | item = btrfs_item_nr(leaf, i); |
3768 | ioff = btrfs_item_offset(leaf, item); | |
3769 | btrfs_set_item_offset(leaf, item, ioff + dsize); | |
0783fcfc | 3770 | } |
db94535d | 3771 | |
5f39d397 | 3772 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 3773 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 3774 | sizeof(struct btrfs_item) * |
85e21bac | 3775 | (nritems - slot - nr)); |
be0e5c09 | 3776 | } |
85e21bac CM |
3777 | btrfs_set_header_nritems(leaf, nritems - nr); |
3778 | nritems -= nr; | |
5f39d397 | 3779 | |
74123bd7 | 3780 | /* delete the leaf if we've emptied it */ |
7518a238 | 3781 | if (nritems == 0) { |
5f39d397 CM |
3782 | if (leaf == root->node) { |
3783 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 3784 | } else { |
f0486c68 YZ |
3785 | btrfs_set_path_blocking(path); |
3786 | clean_tree_block(trans, root, leaf); | |
143bede5 | 3787 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 3788 | } |
be0e5c09 | 3789 | } else { |
7518a238 | 3790 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 3791 | if (slot == 0) { |
5f39d397 CM |
3792 | struct btrfs_disk_key disk_key; |
3793 | ||
3794 | btrfs_item_key(leaf, &disk_key, 0); | |
143bede5 | 3795 | fixup_low_keys(trans, root, path, &disk_key, 1); |
aa5d6bed | 3796 | } |
aa5d6bed | 3797 | |
74123bd7 | 3798 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 3799 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
3800 | /* push_leaf_left fixes the path. |
3801 | * make sure the path still points to our leaf | |
3802 | * for possible call to del_ptr below | |
3803 | */ | |
4920c9ac | 3804 | slot = path->slots[1]; |
5f39d397 CM |
3805 | extent_buffer_get(leaf); |
3806 | ||
b9473439 | 3807 | btrfs_set_path_blocking(path); |
99d8f83c CM |
3808 | wret = push_leaf_left(trans, root, path, 1, 1, |
3809 | 1, (u32)-1); | |
54aa1f4d | 3810 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 3811 | ret = wret; |
5f39d397 CM |
3812 | |
3813 | if (path->nodes[0] == leaf && | |
3814 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
3815 | wret = push_leaf_right(trans, root, path, 1, |
3816 | 1, 1, 0); | |
54aa1f4d | 3817 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
3818 | ret = wret; |
3819 | } | |
5f39d397 CM |
3820 | |
3821 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 3822 | path->slots[1] = slot; |
143bede5 | 3823 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 3824 | free_extent_buffer(leaf); |
143bede5 | 3825 | ret = 0; |
5de08d7d | 3826 | } else { |
925baedd CM |
3827 | /* if we're still in the path, make sure |
3828 | * we're dirty. Otherwise, one of the | |
3829 | * push_leaf functions must have already | |
3830 | * dirtied this buffer | |
3831 | */ | |
3832 | if (path->nodes[0] == leaf) | |
3833 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 3834 | free_extent_buffer(leaf); |
be0e5c09 | 3835 | } |
d5719762 | 3836 | } else { |
5f39d397 | 3837 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
3838 | } |
3839 | } | |
aa5d6bed | 3840 | return ret; |
be0e5c09 CM |
3841 | } |
3842 | ||
7bb86316 | 3843 | /* |
925baedd | 3844 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
3845 | * returns 0 if it found something or 1 if there are no lesser leaves. |
3846 | * returns < 0 on io errors. | |
d352ac68 CM |
3847 | * |
3848 | * This may release the path, and so you may lose any locks held at the | |
3849 | * time you call it. | |
7bb86316 CM |
3850 | */ |
3851 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
3852 | { | |
925baedd CM |
3853 | struct btrfs_key key; |
3854 | struct btrfs_disk_key found_key; | |
3855 | int ret; | |
7bb86316 | 3856 | |
925baedd | 3857 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 3858 | |
925baedd CM |
3859 | if (key.offset > 0) |
3860 | key.offset--; | |
3861 | else if (key.type > 0) | |
3862 | key.type--; | |
3863 | else if (key.objectid > 0) | |
3864 | key.objectid--; | |
3865 | else | |
3866 | return 1; | |
7bb86316 | 3867 | |
b3b4aa74 | 3868 | btrfs_release_path(path); |
925baedd CM |
3869 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3870 | if (ret < 0) | |
3871 | return ret; | |
3872 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
3873 | ret = comp_keys(&found_key, &key); | |
3874 | if (ret < 0) | |
3875 | return 0; | |
3876 | return 1; | |
7bb86316 CM |
3877 | } |
3878 | ||
3f157a2f CM |
3879 | /* |
3880 | * A helper function to walk down the tree starting at min_key, and looking | |
3881 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 3882 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
3883 | * |
3884 | * This does not cow, but it does stuff the starting key it finds back | |
3885 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
3886 | * key and get a writable path. | |
3887 | * | |
3888 | * This does lock as it descends, and path->keep_locks should be set | |
3889 | * to 1 by the caller. | |
3890 | * | |
3891 | * This honors path->lowest_level to prevent descent past a given level | |
3892 | * of the tree. | |
3893 | * | |
d352ac68 CM |
3894 | * min_trans indicates the oldest transaction that you are interested |
3895 | * in walking through. Any nodes or leaves older than min_trans are | |
3896 | * skipped over (without reading them). | |
3897 | * | |
3f157a2f CM |
3898 | * returns zero if something useful was found, < 0 on error and 1 if there |
3899 | * was nothing in the tree that matched the search criteria. | |
3900 | */ | |
3901 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 3902 | struct btrfs_key *max_key, |
3f157a2f CM |
3903 | struct btrfs_path *path, int cache_only, |
3904 | u64 min_trans) | |
3905 | { | |
3906 | struct extent_buffer *cur; | |
3907 | struct btrfs_key found_key; | |
3908 | int slot; | |
9652480b | 3909 | int sret; |
3f157a2f CM |
3910 | u32 nritems; |
3911 | int level; | |
3912 | int ret = 1; | |
3913 | ||
934d375b | 3914 | WARN_ON(!path->keep_locks); |
3f157a2f | 3915 | again: |
bd681513 | 3916 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 3917 | level = btrfs_header_level(cur); |
e02119d5 | 3918 | WARN_ON(path->nodes[level]); |
3f157a2f | 3919 | path->nodes[level] = cur; |
bd681513 | 3920 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
3921 | |
3922 | if (btrfs_header_generation(cur) < min_trans) { | |
3923 | ret = 1; | |
3924 | goto out; | |
3925 | } | |
d397712b | 3926 | while (1) { |
3f157a2f CM |
3927 | nritems = btrfs_header_nritems(cur); |
3928 | level = btrfs_header_level(cur); | |
9652480b | 3929 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 3930 | |
323ac95b CM |
3931 | /* at the lowest level, we're done, setup the path and exit */ |
3932 | if (level == path->lowest_level) { | |
e02119d5 CM |
3933 | if (slot >= nritems) |
3934 | goto find_next_key; | |
3f157a2f CM |
3935 | ret = 0; |
3936 | path->slots[level] = slot; | |
3937 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
3938 | goto out; | |
3939 | } | |
9652480b Y |
3940 | if (sret && slot > 0) |
3941 | slot--; | |
3f157a2f CM |
3942 | /* |
3943 | * check this node pointer against the cache_only and | |
3944 | * min_trans parameters. If it isn't in cache or is too | |
3945 | * old, skip to the next one. | |
3946 | */ | |
d397712b | 3947 | while (slot < nritems) { |
3f157a2f CM |
3948 | u64 blockptr; |
3949 | u64 gen; | |
3950 | struct extent_buffer *tmp; | |
e02119d5 CM |
3951 | struct btrfs_disk_key disk_key; |
3952 | ||
3f157a2f CM |
3953 | blockptr = btrfs_node_blockptr(cur, slot); |
3954 | gen = btrfs_node_ptr_generation(cur, slot); | |
3955 | if (gen < min_trans) { | |
3956 | slot++; | |
3957 | continue; | |
3958 | } | |
3959 | if (!cache_only) | |
3960 | break; | |
3961 | ||
e02119d5 CM |
3962 | if (max_key) { |
3963 | btrfs_node_key(cur, &disk_key, slot); | |
3964 | if (comp_keys(&disk_key, max_key) >= 0) { | |
3965 | ret = 1; | |
3966 | goto out; | |
3967 | } | |
3968 | } | |
3969 | ||
3f157a2f CM |
3970 | tmp = btrfs_find_tree_block(root, blockptr, |
3971 | btrfs_level_size(root, level - 1)); | |
3972 | ||
3973 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
3974 | free_extent_buffer(tmp); | |
3975 | break; | |
3976 | } | |
3977 | if (tmp) | |
3978 | free_extent_buffer(tmp); | |
3979 | slot++; | |
3980 | } | |
e02119d5 | 3981 | find_next_key: |
3f157a2f CM |
3982 | /* |
3983 | * we didn't find a candidate key in this node, walk forward | |
3984 | * and find another one | |
3985 | */ | |
3986 | if (slot >= nritems) { | |
e02119d5 | 3987 | path->slots[level] = slot; |
b4ce94de | 3988 | btrfs_set_path_blocking(path); |
e02119d5 | 3989 | sret = btrfs_find_next_key(root, path, min_key, level, |
3f157a2f | 3990 | cache_only, min_trans); |
e02119d5 | 3991 | if (sret == 0) { |
b3b4aa74 | 3992 | btrfs_release_path(path); |
3f157a2f CM |
3993 | goto again; |
3994 | } else { | |
3995 | goto out; | |
3996 | } | |
3997 | } | |
3998 | /* save our key for returning back */ | |
3999 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4000 | path->slots[level] = slot; | |
4001 | if (level == path->lowest_level) { | |
4002 | ret = 0; | |
4003 | unlock_up(path, level, 1); | |
4004 | goto out; | |
4005 | } | |
b4ce94de | 4006 | btrfs_set_path_blocking(path); |
3f157a2f | 4007 | cur = read_node_slot(root, cur, slot); |
97d9a8a4 | 4008 | BUG_ON(!cur); |
3f157a2f | 4009 | |
bd681513 | 4010 | btrfs_tree_read_lock(cur); |
b4ce94de | 4011 | |
bd681513 | 4012 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f CM |
4013 | path->nodes[level - 1] = cur; |
4014 | unlock_up(path, level, 1); | |
bd681513 | 4015 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4016 | } |
4017 | out: | |
4018 | if (ret == 0) | |
4019 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4020 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4021 | return ret; |
4022 | } | |
4023 | ||
4024 | /* | |
4025 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4026 | * and fixup the path. It looks for and returns the next key in the | |
4027 | * tree based on the current path and the cache_only and min_trans | |
4028 | * parameters. | |
4029 | * | |
4030 | * 0 is returned if another key is found, < 0 if there are any errors | |
4031 | * and 1 is returned if there are no higher keys in the tree | |
4032 | * | |
4033 | * path->keep_locks should be set to 1 on the search made before | |
4034 | * calling this function. | |
4035 | */ | |
e7a84565 | 4036 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
33c66f43 | 4037 | struct btrfs_key *key, int level, |
3f157a2f | 4038 | int cache_only, u64 min_trans) |
e7a84565 | 4039 | { |
e7a84565 CM |
4040 | int slot; |
4041 | struct extent_buffer *c; | |
4042 | ||
934d375b | 4043 | WARN_ON(!path->keep_locks); |
d397712b | 4044 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4045 | if (!path->nodes[level]) |
4046 | return 1; | |
4047 | ||
4048 | slot = path->slots[level] + 1; | |
4049 | c = path->nodes[level]; | |
3f157a2f | 4050 | next: |
e7a84565 | 4051 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4052 | int ret; |
4053 | int orig_lowest; | |
4054 | struct btrfs_key cur_key; | |
4055 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4056 | !path->nodes[level + 1]) | |
e7a84565 | 4057 | return 1; |
33c66f43 YZ |
4058 | |
4059 | if (path->locks[level + 1]) { | |
4060 | level++; | |
4061 | continue; | |
4062 | } | |
4063 | ||
4064 | slot = btrfs_header_nritems(c) - 1; | |
4065 | if (level == 0) | |
4066 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4067 | else | |
4068 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4069 | ||
4070 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4071 | btrfs_release_path(path); |
33c66f43 YZ |
4072 | path->lowest_level = level; |
4073 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4074 | 0, 0); | |
4075 | path->lowest_level = orig_lowest; | |
4076 | if (ret < 0) | |
4077 | return ret; | |
4078 | ||
4079 | c = path->nodes[level]; | |
4080 | slot = path->slots[level]; | |
4081 | if (ret == 0) | |
4082 | slot++; | |
4083 | goto next; | |
e7a84565 | 4084 | } |
33c66f43 | 4085 | |
e7a84565 CM |
4086 | if (level == 0) |
4087 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
4088 | else { |
4089 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
4090 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
4091 | ||
4092 | if (cache_only) { | |
4093 | struct extent_buffer *cur; | |
4094 | cur = btrfs_find_tree_block(root, blockptr, | |
4095 | btrfs_level_size(root, level - 1)); | |
4096 | if (!cur || !btrfs_buffer_uptodate(cur, gen)) { | |
4097 | slot++; | |
4098 | if (cur) | |
4099 | free_extent_buffer(cur); | |
4100 | goto next; | |
4101 | } | |
4102 | free_extent_buffer(cur); | |
4103 | } | |
4104 | if (gen < min_trans) { | |
4105 | slot++; | |
4106 | goto next; | |
4107 | } | |
e7a84565 | 4108 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4109 | } |
e7a84565 CM |
4110 | return 0; |
4111 | } | |
4112 | return 1; | |
4113 | } | |
4114 | ||
97571fd0 | 4115 | /* |
925baedd | 4116 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
4117 | * returns 0 if it found something or 1 if there are no greater leaves. |
4118 | * returns < 0 on io errors. | |
97571fd0 | 4119 | */ |
234b63a0 | 4120 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
4121 | { |
4122 | int slot; | |
8e73f275 | 4123 | int level; |
5f39d397 | 4124 | struct extent_buffer *c; |
8e73f275 | 4125 | struct extent_buffer *next; |
925baedd CM |
4126 | struct btrfs_key key; |
4127 | u32 nritems; | |
4128 | int ret; | |
8e73f275 | 4129 | int old_spinning = path->leave_spinning; |
bd681513 | 4130 | int next_rw_lock = 0; |
925baedd CM |
4131 | |
4132 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4133 | if (nritems == 0) |
925baedd | 4134 | return 1; |
925baedd | 4135 | |
8e73f275 CM |
4136 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4137 | again: | |
4138 | level = 1; | |
4139 | next = NULL; | |
bd681513 | 4140 | next_rw_lock = 0; |
b3b4aa74 | 4141 | btrfs_release_path(path); |
8e73f275 | 4142 | |
a2135011 | 4143 | path->keep_locks = 1; |
31533fb2 | 4144 | path->leave_spinning = 1; |
8e73f275 | 4145 | |
925baedd CM |
4146 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4147 | path->keep_locks = 0; | |
4148 | ||
4149 | if (ret < 0) | |
4150 | return ret; | |
4151 | ||
a2135011 | 4152 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4153 | /* |
4154 | * by releasing the path above we dropped all our locks. A balance | |
4155 | * could have added more items next to the key that used to be | |
4156 | * at the very end of the block. So, check again here and | |
4157 | * advance the path if there are now more items available. | |
4158 | */ | |
a2135011 | 4159 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4160 | if (ret == 0) |
4161 | path->slots[0]++; | |
8e73f275 | 4162 | ret = 0; |
925baedd CM |
4163 | goto done; |
4164 | } | |
d97e63b6 | 4165 | |
d397712b | 4166 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4167 | if (!path->nodes[level]) { |
4168 | ret = 1; | |
4169 | goto done; | |
4170 | } | |
5f39d397 | 4171 | |
d97e63b6 CM |
4172 | slot = path->slots[level] + 1; |
4173 | c = path->nodes[level]; | |
5f39d397 | 4174 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4175 | level++; |
8e73f275 CM |
4176 | if (level == BTRFS_MAX_LEVEL) { |
4177 | ret = 1; | |
4178 | goto done; | |
4179 | } | |
d97e63b6 CM |
4180 | continue; |
4181 | } | |
5f39d397 | 4182 | |
925baedd | 4183 | if (next) { |
bd681513 | 4184 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 4185 | free_extent_buffer(next); |
925baedd | 4186 | } |
5f39d397 | 4187 | |
8e73f275 | 4188 | next = c; |
bd681513 | 4189 | next_rw_lock = path->locks[level]; |
8e73f275 CM |
4190 | ret = read_block_for_search(NULL, root, path, &next, level, |
4191 | slot, &key); | |
4192 | if (ret == -EAGAIN) | |
4193 | goto again; | |
5f39d397 | 4194 | |
76a05b35 | 4195 | if (ret < 0) { |
b3b4aa74 | 4196 | btrfs_release_path(path); |
76a05b35 CM |
4197 | goto done; |
4198 | } | |
4199 | ||
5cd57b2c | 4200 | if (!path->skip_locking) { |
bd681513 | 4201 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4202 | if (!ret) { |
4203 | btrfs_set_path_blocking(path); | |
bd681513 | 4204 | btrfs_tree_read_lock(next); |
31533fb2 | 4205 | btrfs_clear_path_blocking(path, next, |
bd681513 | 4206 | BTRFS_READ_LOCK); |
8e73f275 | 4207 | } |
31533fb2 | 4208 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4209 | } |
d97e63b6 CM |
4210 | break; |
4211 | } | |
4212 | path->slots[level] = slot; | |
d397712b | 4213 | while (1) { |
d97e63b6 CM |
4214 | level--; |
4215 | c = path->nodes[level]; | |
925baedd | 4216 | if (path->locks[level]) |
bd681513 | 4217 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 4218 | |
5f39d397 | 4219 | free_extent_buffer(c); |
d97e63b6 CM |
4220 | path->nodes[level] = next; |
4221 | path->slots[level] = 0; | |
a74a4b97 | 4222 | if (!path->skip_locking) |
bd681513 | 4223 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
4224 | if (!level) |
4225 | break; | |
b4ce94de | 4226 | |
8e73f275 CM |
4227 | ret = read_block_for_search(NULL, root, path, &next, level, |
4228 | 0, &key); | |
4229 | if (ret == -EAGAIN) | |
4230 | goto again; | |
4231 | ||
76a05b35 | 4232 | if (ret < 0) { |
b3b4aa74 | 4233 | btrfs_release_path(path); |
76a05b35 CM |
4234 | goto done; |
4235 | } | |
4236 | ||
5cd57b2c | 4237 | if (!path->skip_locking) { |
bd681513 | 4238 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4239 | if (!ret) { |
4240 | btrfs_set_path_blocking(path); | |
bd681513 | 4241 | btrfs_tree_read_lock(next); |
31533fb2 | 4242 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
4243 | BTRFS_READ_LOCK); |
4244 | } | |
31533fb2 | 4245 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4246 | } |
d97e63b6 | 4247 | } |
8e73f275 | 4248 | ret = 0; |
925baedd CM |
4249 | done: |
4250 | unlock_up(path, 0, 1); | |
8e73f275 CM |
4251 | path->leave_spinning = old_spinning; |
4252 | if (!old_spinning) | |
4253 | btrfs_set_path_blocking(path); | |
4254 | ||
4255 | return ret; | |
d97e63b6 | 4256 | } |
0b86a832 | 4257 | |
3f157a2f CM |
4258 | /* |
4259 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4260 | * searching until it gets past min_objectid or finds an item of 'type' | |
4261 | * | |
4262 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4263 | */ | |
0b86a832 CM |
4264 | int btrfs_previous_item(struct btrfs_root *root, |
4265 | struct btrfs_path *path, u64 min_objectid, | |
4266 | int type) | |
4267 | { | |
4268 | struct btrfs_key found_key; | |
4269 | struct extent_buffer *leaf; | |
e02119d5 | 4270 | u32 nritems; |
0b86a832 CM |
4271 | int ret; |
4272 | ||
d397712b | 4273 | while (1) { |
0b86a832 | 4274 | if (path->slots[0] == 0) { |
b4ce94de | 4275 | btrfs_set_path_blocking(path); |
0b86a832 CM |
4276 | ret = btrfs_prev_leaf(root, path); |
4277 | if (ret != 0) | |
4278 | return ret; | |
4279 | } else { | |
4280 | path->slots[0]--; | |
4281 | } | |
4282 | leaf = path->nodes[0]; | |
e02119d5 CM |
4283 | nritems = btrfs_header_nritems(leaf); |
4284 | if (nritems == 0) | |
4285 | return 1; | |
4286 | if (path->slots[0] == nritems) | |
4287 | path->slots[0]--; | |
4288 | ||
0b86a832 | 4289 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4290 | if (found_key.objectid < min_objectid) |
4291 | break; | |
0a4eefbb YZ |
4292 | if (found_key.type == type) |
4293 | return 0; | |
e02119d5 CM |
4294 | if (found_key.objectid == min_objectid && |
4295 | found_key.type < type) | |
4296 | break; | |
0b86a832 CM |
4297 | } |
4298 | return 1; | |
4299 | } |