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> |
eb60ceac CM |
20 | #include "ctree.h" |
21 | #include "disk-io.h" | |
7f5c1516 | 22 | #include "transaction.h" |
5f39d397 | 23 | #include "print-tree.h" |
925baedd | 24 | #include "locking.h" |
9a8dd150 | 25 | |
e089f05c CM |
26 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
27 | *root, struct btrfs_path *path, int level); | |
28 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 29 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 30 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
31 | static int push_node_left(struct btrfs_trans_handle *trans, |
32 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 33 | struct extent_buffer *src, int empty); |
5f39d397 CM |
34 | static int balance_node_right(struct btrfs_trans_handle *trans, |
35 | struct btrfs_root *root, | |
36 | struct extent_buffer *dst_buf, | |
37 | struct extent_buffer *src_buf); | |
e089f05c CM |
38 | static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
39 | struct btrfs_path *path, int level, int slot); | |
d97e63b6 | 40 | |
df24a2b9 | 41 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 42 | { |
df24a2b9 | 43 | struct btrfs_path *path; |
e00f7308 JM |
44 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
45 | if (path) | |
2cc58cf2 | 46 | path->reada = 1; |
df24a2b9 | 47 | return path; |
2c90e5d6 CM |
48 | } |
49 | ||
b4ce94de CM |
50 | /* |
51 | * set all locked nodes in the path to blocking locks. This should | |
52 | * be done before scheduling | |
53 | */ | |
54 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
55 | { | |
56 | int i; | |
57 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
58 | if (p->nodes[i] && p->locks[i]) | |
59 | btrfs_set_lock_blocking(p->nodes[i]); | |
60 | } | |
61 | } | |
62 | ||
63 | /* | |
64 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
65 | * |
66 | * held is used to keep lockdep happy, when lockdep is enabled | |
67 | * we set held to a blocking lock before we go around and | |
68 | * retake all the spinlocks in the path. You can safely use NULL | |
69 | * for held | |
b4ce94de | 70 | */ |
4008c04a CM |
71 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
72 | struct extent_buffer *held) | |
b4ce94de CM |
73 | { |
74 | int i; | |
4008c04a CM |
75 | |
76 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
77 | /* lockdep really cares that we take all of these spinlocks | |
78 | * in the right order. If any of the locks in the path are not | |
79 | * currently blocking, it is going to complain. So, make really | |
80 | * really sure by forcing the path to blocking before we clear | |
81 | * the path blocking. | |
82 | */ | |
83 | if (held) | |
84 | btrfs_set_lock_blocking(held); | |
85 | btrfs_set_path_blocking(p); | |
86 | #endif | |
87 | ||
88 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
b4ce94de CM |
89 | if (p->nodes[i] && p->locks[i]) |
90 | btrfs_clear_lock_blocking(p->nodes[i]); | |
91 | } | |
4008c04a CM |
92 | |
93 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
94 | if (held) | |
95 | btrfs_clear_lock_blocking(held); | |
96 | #endif | |
b4ce94de CM |
97 | } |
98 | ||
d352ac68 | 99 | /* this also releases the path */ |
df24a2b9 | 100 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 101 | { |
df24a2b9 CM |
102 | btrfs_release_path(NULL, p); |
103 | kmem_cache_free(btrfs_path_cachep, p); | |
be0e5c09 CM |
104 | } |
105 | ||
d352ac68 CM |
106 | /* |
107 | * path release drops references on the extent buffers in the path | |
108 | * and it drops any locks held by this path | |
109 | * | |
110 | * It is safe to call this on paths that no locks or extent buffers held. | |
111 | */ | |
d397712b | 112 | noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p) |
eb60ceac CM |
113 | { |
114 | int i; | |
a2135011 | 115 | |
234b63a0 | 116 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 117 | p->slots[i] = 0; |
eb60ceac | 118 | if (!p->nodes[i]) |
925baedd CM |
119 | continue; |
120 | if (p->locks[i]) { | |
121 | btrfs_tree_unlock(p->nodes[i]); | |
122 | p->locks[i] = 0; | |
123 | } | |
5f39d397 | 124 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 125 | p->nodes[i] = NULL; |
eb60ceac CM |
126 | } |
127 | } | |
128 | ||
d352ac68 CM |
129 | /* |
130 | * safely gets a reference on the root node of a tree. A lock | |
131 | * is not taken, so a concurrent writer may put a different node | |
132 | * at the root of the tree. See btrfs_lock_root_node for the | |
133 | * looping required. | |
134 | * | |
135 | * The extent buffer returned by this has a reference taken, so | |
136 | * it won't disappear. It may stop being the root of the tree | |
137 | * at any time because there are no locks held. | |
138 | */ | |
925baedd CM |
139 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
140 | { | |
141 | struct extent_buffer *eb; | |
142 | spin_lock(&root->node_lock); | |
143 | eb = root->node; | |
144 | extent_buffer_get(eb); | |
145 | spin_unlock(&root->node_lock); | |
146 | return eb; | |
147 | } | |
148 | ||
d352ac68 CM |
149 | /* loop around taking references on and locking the root node of the |
150 | * tree until you end up with a lock on the root. A locked buffer | |
151 | * is returned, with a reference held. | |
152 | */ | |
925baedd CM |
153 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
154 | { | |
155 | struct extent_buffer *eb; | |
156 | ||
d397712b | 157 | while (1) { |
925baedd CM |
158 | eb = btrfs_root_node(root); |
159 | btrfs_tree_lock(eb); | |
160 | ||
161 | spin_lock(&root->node_lock); | |
162 | if (eb == root->node) { | |
163 | spin_unlock(&root->node_lock); | |
164 | break; | |
165 | } | |
166 | spin_unlock(&root->node_lock); | |
167 | ||
168 | btrfs_tree_unlock(eb); | |
169 | free_extent_buffer(eb); | |
170 | } | |
171 | return eb; | |
172 | } | |
173 | ||
d352ac68 CM |
174 | /* cowonly root (everything not a reference counted cow subvolume), just get |
175 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
176 | * get properly updated on disk. | |
177 | */ | |
0b86a832 CM |
178 | static void add_root_to_dirty_list(struct btrfs_root *root) |
179 | { | |
180 | if (root->track_dirty && list_empty(&root->dirty_list)) { | |
181 | list_add(&root->dirty_list, | |
182 | &root->fs_info->dirty_cowonly_roots); | |
183 | } | |
184 | } | |
185 | ||
d352ac68 CM |
186 | /* |
187 | * used by snapshot creation to make a copy of a root for a tree with | |
188 | * a given objectid. The buffer with the new root node is returned in | |
189 | * cow_ret, and this func returns zero on success or a negative error code. | |
190 | */ | |
be20aa9d CM |
191 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
192 | struct btrfs_root *root, | |
193 | struct extent_buffer *buf, | |
194 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
195 | { | |
196 | struct extent_buffer *cow; | |
197 | u32 nritems; | |
198 | int ret = 0; | |
199 | int level; | |
4aec2b52 | 200 | struct btrfs_root *new_root; |
be20aa9d | 201 | |
4aec2b52 CM |
202 | new_root = kmalloc(sizeof(*new_root), GFP_NOFS); |
203 | if (!new_root) | |
204 | return -ENOMEM; | |
205 | ||
206 | memcpy(new_root, root, sizeof(*new_root)); | |
207 | new_root->root_key.objectid = new_root_objectid; | |
be20aa9d CM |
208 | |
209 | WARN_ON(root->ref_cows && trans->transid != | |
210 | root->fs_info->running_transaction->transid); | |
211 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
212 | ||
213 | level = btrfs_header_level(buf); | |
214 | nritems = btrfs_header_nritems(buf); | |
31840ae1 ZY |
215 | |
216 | cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0, | |
217 | new_root_objectid, trans->transid, | |
218 | level, buf->start, 0); | |
4aec2b52 CM |
219 | if (IS_ERR(cow)) { |
220 | kfree(new_root); | |
be20aa9d | 221 | return PTR_ERR(cow); |
4aec2b52 | 222 | } |
be20aa9d CM |
223 | |
224 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
225 | btrfs_set_header_bytenr(cow, cow->start); | |
226 | btrfs_set_header_generation(cow, trans->transid); | |
227 | btrfs_set_header_owner(cow, new_root_objectid); | |
63b10fc4 | 228 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN); |
be20aa9d | 229 | |
2b82032c YZ |
230 | write_extent_buffer(cow, root->fs_info->fsid, |
231 | (unsigned long)btrfs_header_fsid(cow), | |
232 | BTRFS_FSID_SIZE); | |
233 | ||
be20aa9d | 234 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
31840ae1 | 235 | ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL); |
4aec2b52 CM |
236 | kfree(new_root); |
237 | ||
be20aa9d CM |
238 | if (ret) |
239 | return ret; | |
240 | ||
241 | btrfs_mark_buffer_dirty(cow); | |
242 | *cow_ret = cow; | |
243 | return 0; | |
244 | } | |
245 | ||
d352ac68 | 246 | /* |
d397712b CM |
247 | * does the dirty work in cow of a single block. The parent block (if |
248 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
249 | * dirty and returned locked. If you modify the block it needs to be marked | |
250 | * dirty again. | |
d352ac68 CM |
251 | * |
252 | * search_start -- an allocation hint for the new block | |
253 | * | |
d397712b CM |
254 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
255 | * bytes the allocator should try to find free next to the block it returns. | |
256 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 257 | */ |
d397712b | 258 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
259 | struct btrfs_root *root, |
260 | struct extent_buffer *buf, | |
261 | struct extent_buffer *parent, int parent_slot, | |
262 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 263 | u64 search_start, u64 empty_size) |
02217ed2 | 264 | { |
31840ae1 | 265 | u64 parent_start; |
5f39d397 | 266 | struct extent_buffer *cow; |
7bb86316 | 267 | u32 nritems; |
6702ed49 | 268 | int ret = 0; |
7bb86316 | 269 | int level; |
925baedd | 270 | int unlock_orig = 0; |
7bb86316 | 271 | |
925baedd CM |
272 | if (*cow_ret == buf) |
273 | unlock_orig = 1; | |
274 | ||
b9447ef8 | 275 | btrfs_assert_tree_locked(buf); |
925baedd | 276 | |
31840ae1 ZY |
277 | if (parent) |
278 | parent_start = parent->start; | |
279 | else | |
280 | parent_start = 0; | |
281 | ||
7bb86316 CM |
282 | WARN_ON(root->ref_cows && trans->transid != |
283 | root->fs_info->running_transaction->transid); | |
6702ed49 | 284 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 285 | |
7bb86316 CM |
286 | level = btrfs_header_level(buf); |
287 | nritems = btrfs_header_nritems(buf); | |
31840ae1 | 288 | |
9fa8cfe7 CM |
289 | cow = btrfs_alloc_free_block(trans, root, buf->len, |
290 | parent_start, root->root_key.objectid, | |
291 | trans->transid, level, | |
292 | search_start, empty_size); | |
54aa1f4d CM |
293 | if (IS_ERR(cow)) |
294 | return PTR_ERR(cow); | |
6702ed49 | 295 | |
b4ce94de CM |
296 | /* cow is set to blocking by btrfs_init_new_buffer */ |
297 | ||
5f39d397 | 298 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 299 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 CM |
300 | btrfs_set_header_generation(cow, trans->transid); |
301 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
63b10fc4 | 302 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN); |
6702ed49 | 303 | |
2b82032c YZ |
304 | write_extent_buffer(cow, root->fs_info->fsid, |
305 | (unsigned long)btrfs_header_fsid(cow), | |
306 | BTRFS_FSID_SIZE); | |
307 | ||
5f39d397 CM |
308 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
309 | if (btrfs_header_generation(buf) != trans->transid) { | |
31840ae1 | 310 | u32 nr_extents; |
31840ae1 | 311 | ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents); |
6702ed49 CM |
312 | if (ret) |
313 | return ret; | |
31840ae1 ZY |
314 | |
315 | ret = btrfs_cache_ref(trans, root, buf, nr_extents); | |
316 | WARN_ON(ret); | |
1a40e23b ZY |
317 | } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) { |
318 | /* | |
319 | * There are only two places that can drop reference to | |
320 | * tree blocks owned by living reloc trees, one is here, | |
f82d02d9 | 321 | * the other place is btrfs_drop_subtree. In both places, |
1a40e23b ZY |
322 | * we check reference count while tree block is locked. |
323 | * Furthermore, if reference count is one, it won't get | |
324 | * increased by someone else. | |
325 | */ | |
326 | u32 refs; | |
327 | ret = btrfs_lookup_extent_ref(trans, root, buf->start, | |
328 | buf->len, &refs); | |
329 | BUG_ON(ret); | |
330 | if (refs == 1) { | |
331 | ret = btrfs_update_ref(trans, root, buf, cow, | |
332 | 0, nritems); | |
333 | clean_tree_block(trans, root, buf); | |
334 | } else { | |
335 | ret = btrfs_inc_ref(trans, root, buf, cow, NULL); | |
336 | } | |
337 | BUG_ON(ret); | |
6702ed49 | 338 | } else { |
31840ae1 ZY |
339 | ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems); |
340 | if (ret) | |
341 | return ret; | |
6702ed49 CM |
342 | clean_tree_block(trans, root, buf); |
343 | } | |
344 | ||
1a40e23b | 345 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { |
1a40e23b ZY |
346 | ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start); |
347 | WARN_ON(ret); | |
348 | } | |
349 | ||
02217ed2 | 350 | if (buf == root->node) { |
925baedd | 351 | WARN_ON(parent && parent != buf); |
925baedd CM |
352 | |
353 | spin_lock(&root->node_lock); | |
02217ed2 | 354 | root->node = cow; |
5f39d397 | 355 | extent_buffer_get(cow); |
925baedd CM |
356 | spin_unlock(&root->node_lock); |
357 | ||
2c90e5d6 | 358 | if (buf != root->commit_root) { |
db94535d | 359 | btrfs_free_extent(trans, root, buf->start, |
31840ae1 ZY |
360 | buf->len, buf->start, |
361 | root->root_key.objectid, | |
362 | btrfs_header_generation(buf), | |
3bb1a1bc | 363 | level, 1); |
2c90e5d6 | 364 | } |
5f39d397 | 365 | free_extent_buffer(buf); |
0b86a832 | 366 | add_root_to_dirty_list(root); |
02217ed2 | 367 | } else { |
5f39d397 | 368 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 369 | cow->start); |
74493f7a CM |
370 | WARN_ON(trans->transid == 0); |
371 | btrfs_set_node_ptr_generation(parent, parent_slot, | |
372 | trans->transid); | |
d6025579 | 373 | btrfs_mark_buffer_dirty(parent); |
5f39d397 | 374 | WARN_ON(btrfs_header_generation(parent) != trans->transid); |
7bb86316 | 375 | btrfs_free_extent(trans, root, buf->start, buf->len, |
31840ae1 | 376 | parent_start, btrfs_header_owner(parent), |
3bb1a1bc | 377 | btrfs_header_generation(parent), level, 1); |
02217ed2 | 378 | } |
925baedd CM |
379 | if (unlock_orig) |
380 | btrfs_tree_unlock(buf); | |
5f39d397 | 381 | free_extent_buffer(buf); |
ccd467d6 | 382 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 383 | *cow_ret = cow; |
02217ed2 CM |
384 | return 0; |
385 | } | |
386 | ||
d352ac68 CM |
387 | /* |
388 | * cows a single block, see __btrfs_cow_block for the real work. | |
389 | * This version of it has extra checks so that a block isn't cow'd more than | |
390 | * once per transaction, as long as it hasn't been written yet | |
391 | */ | |
d397712b | 392 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
393 | struct btrfs_root *root, struct extent_buffer *buf, |
394 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 395 | struct extent_buffer **cow_ret) |
6702ed49 CM |
396 | { |
397 | u64 search_start; | |
f510cfec | 398 | int ret; |
dc17ff8f | 399 | |
6702ed49 | 400 | if (trans->transaction != root->fs_info->running_transaction) { |
d397712b CM |
401 | printk(KERN_CRIT "trans %llu running %llu\n", |
402 | (unsigned long long)trans->transid, | |
403 | (unsigned long long) | |
6702ed49 CM |
404 | root->fs_info->running_transaction->transid); |
405 | WARN_ON(1); | |
406 | } | |
407 | if (trans->transid != root->fs_info->generation) { | |
d397712b CM |
408 | printk(KERN_CRIT "trans %llu running %llu\n", |
409 | (unsigned long long)trans->transid, | |
410 | (unsigned long long)root->fs_info->generation); | |
6702ed49 CM |
411 | WARN_ON(1); |
412 | } | |
dc17ff8f | 413 | |
5b21f2ed ZY |
414 | if (btrfs_header_generation(buf) == trans->transid && |
415 | btrfs_header_owner(buf) == root->root_key.objectid && | |
63b10fc4 | 416 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6702ed49 CM |
417 | *cow_ret = buf; |
418 | return 0; | |
419 | } | |
c487685d | 420 | |
0b86a832 | 421 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
422 | |
423 | if (parent) | |
424 | btrfs_set_lock_blocking(parent); | |
425 | btrfs_set_lock_blocking(buf); | |
426 | ||
f510cfec | 427 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 428 | parent_slot, cow_ret, search_start, 0); |
f510cfec | 429 | return ret; |
6702ed49 CM |
430 | } |
431 | ||
d352ac68 CM |
432 | /* |
433 | * helper function for defrag to decide if two blocks pointed to by a | |
434 | * node are actually close by | |
435 | */ | |
6b80053d | 436 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 437 | { |
6b80053d | 438 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 439 | return 1; |
6b80053d | 440 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
441 | return 1; |
442 | return 0; | |
443 | } | |
444 | ||
081e9573 CM |
445 | /* |
446 | * compare two keys in a memcmp fashion | |
447 | */ | |
448 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
449 | { | |
450 | struct btrfs_key k1; | |
451 | ||
452 | btrfs_disk_key_to_cpu(&k1, disk); | |
453 | ||
454 | if (k1.objectid > k2->objectid) | |
455 | return 1; | |
456 | if (k1.objectid < k2->objectid) | |
457 | return -1; | |
458 | if (k1.type > k2->type) | |
459 | return 1; | |
460 | if (k1.type < k2->type) | |
461 | return -1; | |
462 | if (k1.offset > k2->offset) | |
463 | return 1; | |
464 | if (k1.offset < k2->offset) | |
465 | return -1; | |
466 | return 0; | |
467 | } | |
468 | ||
f3465ca4 JB |
469 | /* |
470 | * same as comp_keys only with two btrfs_key's | |
471 | */ | |
472 | static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) | |
473 | { | |
474 | if (k1->objectid > k2->objectid) | |
475 | return 1; | |
476 | if (k1->objectid < k2->objectid) | |
477 | return -1; | |
478 | if (k1->type > k2->type) | |
479 | return 1; | |
480 | if (k1->type < k2->type) | |
481 | return -1; | |
482 | if (k1->offset > k2->offset) | |
483 | return 1; | |
484 | if (k1->offset < k2->offset) | |
485 | return -1; | |
486 | return 0; | |
487 | } | |
081e9573 | 488 | |
d352ac68 CM |
489 | /* |
490 | * this is used by the defrag code to go through all the | |
491 | * leaves pointed to by a node and reallocate them so that | |
492 | * disk order is close to key order | |
493 | */ | |
6702ed49 | 494 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 495 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
496 | int start_slot, int cache_only, u64 *last_ret, |
497 | struct btrfs_key *progress) | |
6702ed49 | 498 | { |
6b80053d | 499 | struct extent_buffer *cur; |
6702ed49 | 500 | u64 blocknr; |
ca7a79ad | 501 | u64 gen; |
e9d0b13b CM |
502 | u64 search_start = *last_ret; |
503 | u64 last_block = 0; | |
6702ed49 CM |
504 | u64 other; |
505 | u32 parent_nritems; | |
6702ed49 CM |
506 | int end_slot; |
507 | int i; | |
508 | int err = 0; | |
f2183bde | 509 | int parent_level; |
6b80053d CM |
510 | int uptodate; |
511 | u32 blocksize; | |
081e9573 CM |
512 | int progress_passed = 0; |
513 | struct btrfs_disk_key disk_key; | |
6702ed49 | 514 | |
5708b959 CM |
515 | parent_level = btrfs_header_level(parent); |
516 | if (cache_only && parent_level != 1) | |
517 | return 0; | |
518 | ||
d397712b | 519 | if (trans->transaction != root->fs_info->running_transaction) |
6702ed49 | 520 | WARN_ON(1); |
d397712b | 521 | if (trans->transid != root->fs_info->generation) |
6702ed49 | 522 | WARN_ON(1); |
86479a04 | 523 | |
6b80053d | 524 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 525 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
526 | end_slot = parent_nritems; |
527 | ||
528 | if (parent_nritems == 1) | |
529 | return 0; | |
530 | ||
b4ce94de CM |
531 | btrfs_set_lock_blocking(parent); |
532 | ||
6702ed49 CM |
533 | for (i = start_slot; i < end_slot; i++) { |
534 | int close = 1; | |
a6b6e75e | 535 | |
5708b959 CM |
536 | if (!parent->map_token) { |
537 | map_extent_buffer(parent, | |
538 | btrfs_node_key_ptr_offset(i), | |
539 | sizeof(struct btrfs_key_ptr), | |
540 | &parent->map_token, &parent->kaddr, | |
541 | &parent->map_start, &parent->map_len, | |
542 | KM_USER1); | |
543 | } | |
081e9573 CM |
544 | btrfs_node_key(parent, &disk_key, i); |
545 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
546 | continue; | |
547 | ||
548 | progress_passed = 1; | |
6b80053d | 549 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 550 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
551 | if (last_block == 0) |
552 | last_block = blocknr; | |
5708b959 | 553 | |
6702ed49 | 554 | if (i > 0) { |
6b80053d CM |
555 | other = btrfs_node_blockptr(parent, i - 1); |
556 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 557 | } |
0ef3e66b | 558 | if (!close && i < end_slot - 2) { |
6b80053d CM |
559 | other = btrfs_node_blockptr(parent, i + 1); |
560 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 561 | } |
e9d0b13b CM |
562 | if (close) { |
563 | last_block = blocknr; | |
6702ed49 | 564 | continue; |
e9d0b13b | 565 | } |
5708b959 CM |
566 | if (parent->map_token) { |
567 | unmap_extent_buffer(parent, parent->map_token, | |
568 | KM_USER1); | |
569 | parent->map_token = NULL; | |
570 | } | |
6702ed49 | 571 | |
6b80053d CM |
572 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
573 | if (cur) | |
1259ab75 | 574 | uptodate = btrfs_buffer_uptodate(cur, gen); |
6b80053d CM |
575 | else |
576 | uptodate = 0; | |
5708b959 | 577 | if (!cur || !uptodate) { |
6702ed49 | 578 | if (cache_only) { |
6b80053d | 579 | free_extent_buffer(cur); |
6702ed49 CM |
580 | continue; |
581 | } | |
6b80053d CM |
582 | if (!cur) { |
583 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 584 | blocksize, gen); |
6b80053d | 585 | } else if (!uptodate) { |
ca7a79ad | 586 | btrfs_read_buffer(cur, gen); |
f2183bde | 587 | } |
6702ed49 | 588 | } |
e9d0b13b | 589 | if (search_start == 0) |
6b80053d | 590 | search_start = last_block; |
e9d0b13b | 591 | |
e7a84565 | 592 | btrfs_tree_lock(cur); |
b4ce94de | 593 | btrfs_set_lock_blocking(cur); |
6b80053d | 594 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 595 | &cur, search_start, |
6b80053d | 596 | min(16 * blocksize, |
9fa8cfe7 | 597 | (end_slot - i) * blocksize)); |
252c38f0 | 598 | if (err) { |
e7a84565 | 599 | btrfs_tree_unlock(cur); |
6b80053d | 600 | free_extent_buffer(cur); |
6702ed49 | 601 | break; |
252c38f0 | 602 | } |
e7a84565 CM |
603 | search_start = cur->start; |
604 | last_block = cur->start; | |
f2183bde | 605 | *last_ret = search_start; |
e7a84565 CM |
606 | btrfs_tree_unlock(cur); |
607 | free_extent_buffer(cur); | |
6702ed49 | 608 | } |
5708b959 CM |
609 | if (parent->map_token) { |
610 | unmap_extent_buffer(parent, parent->map_token, | |
611 | KM_USER1); | |
612 | parent->map_token = NULL; | |
613 | } | |
6702ed49 CM |
614 | return err; |
615 | } | |
616 | ||
74123bd7 CM |
617 | /* |
618 | * The leaf data grows from end-to-front in the node. | |
619 | * this returns the address of the start of the last item, | |
620 | * which is the stop of the leaf data stack | |
621 | */ | |
123abc88 | 622 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 623 | struct extent_buffer *leaf) |
be0e5c09 | 624 | { |
5f39d397 | 625 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 626 | if (nr == 0) |
123abc88 | 627 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 628 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
629 | } |
630 | ||
d352ac68 CM |
631 | /* |
632 | * extra debugging checks to make sure all the items in a key are | |
633 | * well formed and in the proper order | |
634 | */ | |
123abc88 CM |
635 | static int check_node(struct btrfs_root *root, struct btrfs_path *path, |
636 | int level) | |
aa5d6bed | 637 | { |
5f39d397 CM |
638 | struct extent_buffer *parent = NULL; |
639 | struct extent_buffer *node = path->nodes[level]; | |
640 | struct btrfs_disk_key parent_key; | |
641 | struct btrfs_disk_key node_key; | |
aa5d6bed | 642 | int parent_slot; |
8d7be552 CM |
643 | int slot; |
644 | struct btrfs_key cpukey; | |
5f39d397 | 645 | u32 nritems = btrfs_header_nritems(node); |
aa5d6bed CM |
646 | |
647 | if (path->nodes[level + 1]) | |
5f39d397 | 648 | parent = path->nodes[level + 1]; |
a1f39630 | 649 | |
8d7be552 | 650 | slot = path->slots[level]; |
7518a238 CM |
651 | BUG_ON(nritems == 0); |
652 | if (parent) { | |
a1f39630 | 653 | parent_slot = path->slots[level + 1]; |
5f39d397 CM |
654 | btrfs_node_key(parent, &parent_key, parent_slot); |
655 | btrfs_node_key(node, &node_key, 0); | |
656 | BUG_ON(memcmp(&parent_key, &node_key, | |
e2fa7227 | 657 | sizeof(struct btrfs_disk_key))); |
1d4f8a0c | 658 | BUG_ON(btrfs_node_blockptr(parent, parent_slot) != |
db94535d | 659 | btrfs_header_bytenr(node)); |
aa5d6bed | 660 | } |
123abc88 | 661 | BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root)); |
8d7be552 | 662 | if (slot != 0) { |
5f39d397 CM |
663 | btrfs_node_key_to_cpu(node, &cpukey, slot - 1); |
664 | btrfs_node_key(node, &node_key, slot); | |
665 | BUG_ON(comp_keys(&node_key, &cpukey) <= 0); | |
8d7be552 CM |
666 | } |
667 | if (slot < nritems - 1) { | |
5f39d397 CM |
668 | btrfs_node_key_to_cpu(node, &cpukey, slot + 1); |
669 | btrfs_node_key(node, &node_key, slot); | |
670 | BUG_ON(comp_keys(&node_key, &cpukey) >= 0); | |
aa5d6bed CM |
671 | } |
672 | return 0; | |
673 | } | |
674 | ||
d352ac68 CM |
675 | /* |
676 | * extra checking to make sure all the items in a leaf are | |
677 | * well formed and in the proper order | |
678 | */ | |
123abc88 CM |
679 | static int check_leaf(struct btrfs_root *root, struct btrfs_path *path, |
680 | int level) | |
aa5d6bed | 681 | { |
5f39d397 CM |
682 | struct extent_buffer *leaf = path->nodes[level]; |
683 | struct extent_buffer *parent = NULL; | |
aa5d6bed | 684 | int parent_slot; |
8d7be552 | 685 | struct btrfs_key cpukey; |
5f39d397 CM |
686 | struct btrfs_disk_key parent_key; |
687 | struct btrfs_disk_key leaf_key; | |
688 | int slot = path->slots[0]; | |
8d7be552 | 689 | |
5f39d397 | 690 | u32 nritems = btrfs_header_nritems(leaf); |
aa5d6bed CM |
691 | |
692 | if (path->nodes[level + 1]) | |
5f39d397 | 693 | parent = path->nodes[level + 1]; |
7518a238 CM |
694 | |
695 | if (nritems == 0) | |
696 | return 0; | |
697 | ||
698 | if (parent) { | |
a1f39630 | 699 | parent_slot = path->slots[level + 1]; |
5f39d397 CM |
700 | btrfs_node_key(parent, &parent_key, parent_slot); |
701 | btrfs_item_key(leaf, &leaf_key, 0); | |
6702ed49 | 702 | |
5f39d397 | 703 | BUG_ON(memcmp(&parent_key, &leaf_key, |
e2fa7227 | 704 | sizeof(struct btrfs_disk_key))); |
1d4f8a0c | 705 | BUG_ON(btrfs_node_blockptr(parent, parent_slot) != |
db94535d | 706 | btrfs_header_bytenr(leaf)); |
5f39d397 | 707 | } |
5f39d397 CM |
708 | if (slot != 0 && slot < nritems - 1) { |
709 | btrfs_item_key(leaf, &leaf_key, slot); | |
710 | btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1); | |
711 | if (comp_keys(&leaf_key, &cpukey) <= 0) { | |
712 | btrfs_print_leaf(root, leaf); | |
d397712b | 713 | printk(KERN_CRIT "slot %d offset bad key\n", slot); |
5f39d397 CM |
714 | BUG_ON(1); |
715 | } | |
716 | if (btrfs_item_offset_nr(leaf, slot - 1) != | |
717 | btrfs_item_end_nr(leaf, slot)) { | |
718 | btrfs_print_leaf(root, leaf); | |
d397712b | 719 | printk(KERN_CRIT "slot %d offset bad\n", slot); |
5f39d397 CM |
720 | BUG_ON(1); |
721 | } | |
8d7be552 CM |
722 | } |
723 | if (slot < nritems - 1) { | |
5f39d397 CM |
724 | btrfs_item_key(leaf, &leaf_key, slot); |
725 | btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1); | |
726 | BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0); | |
727 | if (btrfs_item_offset_nr(leaf, slot) != | |
728 | btrfs_item_end_nr(leaf, slot + 1)) { | |
729 | btrfs_print_leaf(root, leaf); | |
d397712b | 730 | printk(KERN_CRIT "slot %d offset bad\n", slot); |
5f39d397 CM |
731 | BUG_ON(1); |
732 | } | |
aa5d6bed | 733 | } |
5f39d397 CM |
734 | BUG_ON(btrfs_item_offset_nr(leaf, 0) + |
735 | btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root)); | |
aa5d6bed CM |
736 | return 0; |
737 | } | |
738 | ||
d397712b | 739 | static noinline int check_block(struct btrfs_root *root, |
98ed5174 | 740 | struct btrfs_path *path, int level) |
aa5d6bed | 741 | { |
85d824c4 | 742 | return 0; |
aa5d6bed | 743 | if (level == 0) |
123abc88 CM |
744 | return check_leaf(root, path, level); |
745 | return check_node(root, path, level); | |
aa5d6bed CM |
746 | } |
747 | ||
74123bd7 | 748 | /* |
5f39d397 CM |
749 | * search for key in the extent_buffer. The items start at offset p, |
750 | * and they are item_size apart. There are 'max' items in p. | |
751 | * | |
74123bd7 CM |
752 | * the slot in the array is returned via slot, and it points to |
753 | * the place where you would insert key if it is not found in | |
754 | * the array. | |
755 | * | |
756 | * slot may point to max if the key is bigger than all of the keys | |
757 | */ | |
e02119d5 CM |
758 | static noinline int generic_bin_search(struct extent_buffer *eb, |
759 | unsigned long p, | |
760 | int item_size, struct btrfs_key *key, | |
761 | int max, int *slot) | |
be0e5c09 CM |
762 | { |
763 | int low = 0; | |
764 | int high = max; | |
765 | int mid; | |
766 | int ret; | |
479965d6 | 767 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
768 | struct btrfs_disk_key unaligned; |
769 | unsigned long offset; | |
770 | char *map_token = NULL; | |
771 | char *kaddr = NULL; | |
772 | unsigned long map_start = 0; | |
773 | unsigned long map_len = 0; | |
479965d6 | 774 | int err; |
be0e5c09 | 775 | |
d397712b | 776 | while (low < high) { |
be0e5c09 | 777 | mid = (low + high) / 2; |
5f39d397 CM |
778 | offset = p + mid * item_size; |
779 | ||
780 | if (!map_token || offset < map_start || | |
781 | (offset + sizeof(struct btrfs_disk_key)) > | |
782 | map_start + map_len) { | |
479965d6 | 783 | if (map_token) { |
5f39d397 | 784 | unmap_extent_buffer(eb, map_token, KM_USER0); |
479965d6 CM |
785 | map_token = NULL; |
786 | } | |
934d375b CM |
787 | |
788 | err = map_private_extent_buffer(eb, offset, | |
479965d6 CM |
789 | sizeof(struct btrfs_disk_key), |
790 | &map_token, &kaddr, | |
791 | &map_start, &map_len, KM_USER0); | |
792 | ||
793 | if (!err) { | |
794 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
795 | map_start); | |
796 | } else { | |
797 | read_extent_buffer(eb, &unaligned, | |
798 | offset, sizeof(unaligned)); | |
799 | tmp = &unaligned; | |
800 | } | |
5f39d397 | 801 | |
5f39d397 CM |
802 | } else { |
803 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
804 | map_start); | |
805 | } | |
be0e5c09 CM |
806 | ret = comp_keys(tmp, key); |
807 | ||
808 | if (ret < 0) | |
809 | low = mid + 1; | |
810 | else if (ret > 0) | |
811 | high = mid; | |
812 | else { | |
813 | *slot = mid; | |
479965d6 CM |
814 | if (map_token) |
815 | unmap_extent_buffer(eb, map_token, KM_USER0); | |
be0e5c09 CM |
816 | return 0; |
817 | } | |
818 | } | |
819 | *slot = low; | |
5f39d397 CM |
820 | if (map_token) |
821 | unmap_extent_buffer(eb, map_token, KM_USER0); | |
be0e5c09 CM |
822 | return 1; |
823 | } | |
824 | ||
97571fd0 CM |
825 | /* |
826 | * simple bin_search frontend that does the right thing for | |
827 | * leaves vs nodes | |
828 | */ | |
5f39d397 CM |
829 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
830 | int level, int *slot) | |
be0e5c09 | 831 | { |
5f39d397 CM |
832 | if (level == 0) { |
833 | return generic_bin_search(eb, | |
834 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 835 | sizeof(struct btrfs_item), |
5f39d397 | 836 | key, btrfs_header_nritems(eb), |
7518a238 | 837 | slot); |
be0e5c09 | 838 | } else { |
5f39d397 CM |
839 | return generic_bin_search(eb, |
840 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 841 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 842 | key, btrfs_header_nritems(eb), |
7518a238 | 843 | slot); |
be0e5c09 CM |
844 | } |
845 | return -1; | |
846 | } | |
847 | ||
d352ac68 CM |
848 | /* given a node and slot number, this reads the blocks it points to. The |
849 | * extent buffer is returned with a reference taken (but unlocked). | |
850 | * NULL is returned on error. | |
851 | */ | |
e02119d5 | 852 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 853 | struct extent_buffer *parent, int slot) |
bb803951 | 854 | { |
ca7a79ad | 855 | int level = btrfs_header_level(parent); |
bb803951 CM |
856 | if (slot < 0) |
857 | return NULL; | |
5f39d397 | 858 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 859 | return NULL; |
ca7a79ad CM |
860 | |
861 | BUG_ON(level == 0); | |
862 | ||
db94535d | 863 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
864 | btrfs_level_size(root, level - 1), |
865 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
866 | } |
867 | ||
d352ac68 CM |
868 | /* |
869 | * node level balancing, used to make sure nodes are in proper order for | |
870 | * item deletion. We balance from the top down, so we have to make sure | |
871 | * that a deletion won't leave an node completely empty later on. | |
872 | */ | |
e02119d5 | 873 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
874 | struct btrfs_root *root, |
875 | struct btrfs_path *path, int level) | |
bb803951 | 876 | { |
5f39d397 CM |
877 | struct extent_buffer *right = NULL; |
878 | struct extent_buffer *mid; | |
879 | struct extent_buffer *left = NULL; | |
880 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
881 | int ret = 0; |
882 | int wret; | |
883 | int pslot; | |
bb803951 | 884 | int orig_slot = path->slots[level]; |
54aa1f4d | 885 | int err_on_enospc = 0; |
79f95c82 | 886 | u64 orig_ptr; |
bb803951 CM |
887 | |
888 | if (level == 0) | |
889 | return 0; | |
890 | ||
5f39d397 | 891 | mid = path->nodes[level]; |
b4ce94de | 892 | |
925baedd | 893 | WARN_ON(!path->locks[level]); |
7bb86316 CM |
894 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
895 | ||
1d4f8a0c | 896 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 897 | |
234b63a0 | 898 | if (level < BTRFS_MAX_LEVEL - 1) |
5f39d397 | 899 | parent = path->nodes[level + 1]; |
bb803951 CM |
900 | pslot = path->slots[level + 1]; |
901 | ||
40689478 CM |
902 | /* |
903 | * deal with the case where there is only one pointer in the root | |
904 | * by promoting the node below to a root | |
905 | */ | |
5f39d397 CM |
906 | if (!parent) { |
907 | struct extent_buffer *child; | |
bb803951 | 908 | |
5f39d397 | 909 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
910 | return 0; |
911 | ||
912 | /* promote the child to a root */ | |
5f39d397 | 913 | child = read_node_slot(root, mid, 0); |
7951f3ce | 914 | BUG_ON(!child); |
925baedd | 915 | btrfs_tree_lock(child); |
b4ce94de | 916 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 917 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
2f375ab9 Y |
918 | BUG_ON(ret); |
919 | ||
925baedd | 920 | spin_lock(&root->node_lock); |
bb803951 | 921 | root->node = child; |
925baedd CM |
922 | spin_unlock(&root->node_lock); |
923 | ||
31840ae1 | 924 | ret = btrfs_update_extent_ref(trans, root, child->start, |
56bec294 | 925 | child->len, |
31840ae1 ZY |
926 | mid->start, child->start, |
927 | root->root_key.objectid, | |
3bb1a1bc | 928 | trans->transid, level - 1); |
31840ae1 ZY |
929 | BUG_ON(ret); |
930 | ||
0b86a832 | 931 | add_root_to_dirty_list(root); |
925baedd | 932 | btrfs_tree_unlock(child); |
b4ce94de | 933 | |
925baedd | 934 | path->locks[level] = 0; |
bb803951 | 935 | path->nodes[level] = NULL; |
5f39d397 | 936 | clean_tree_block(trans, root, mid); |
925baedd | 937 | btrfs_tree_unlock(mid); |
bb803951 | 938 | /* once for the path */ |
5f39d397 | 939 | free_extent_buffer(mid); |
7bb86316 | 940 | ret = btrfs_free_extent(trans, root, mid->start, mid->len, |
31840ae1 | 941 | mid->start, root->root_key.objectid, |
3bb1a1bc YZ |
942 | btrfs_header_generation(mid), |
943 | level, 1); | |
bb803951 | 944 | /* once for the root ptr */ |
5f39d397 | 945 | free_extent_buffer(mid); |
db94535d | 946 | return ret; |
bb803951 | 947 | } |
5f39d397 | 948 | if (btrfs_header_nritems(mid) > |
123abc88 | 949 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
950 | return 0; |
951 | ||
a4b6e07d CM |
952 | if (trans->transaction->delayed_refs.flushing && |
953 | btrfs_header_nritems(mid) > 2) | |
954 | return 0; | |
955 | ||
5f39d397 | 956 | if (btrfs_header_nritems(mid) < 2) |
54aa1f4d CM |
957 | err_on_enospc = 1; |
958 | ||
5f39d397 CM |
959 | left = read_node_slot(root, parent, pslot - 1); |
960 | if (left) { | |
925baedd | 961 | btrfs_tree_lock(left); |
b4ce94de | 962 | btrfs_set_lock_blocking(left); |
5f39d397 | 963 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 964 | parent, pslot - 1, &left); |
54aa1f4d CM |
965 | if (wret) { |
966 | ret = wret; | |
967 | goto enospc; | |
968 | } | |
2cc58cf2 | 969 | } |
5f39d397 CM |
970 | right = read_node_slot(root, parent, pslot + 1); |
971 | if (right) { | |
925baedd | 972 | btrfs_tree_lock(right); |
b4ce94de | 973 | btrfs_set_lock_blocking(right); |
5f39d397 | 974 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 975 | parent, pslot + 1, &right); |
2cc58cf2 CM |
976 | if (wret) { |
977 | ret = wret; | |
978 | goto enospc; | |
979 | } | |
980 | } | |
981 | ||
982 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
983 | if (left) { |
984 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 985 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
986 | if (wret < 0) |
987 | ret = wret; | |
5f39d397 | 988 | if (btrfs_header_nritems(mid) < 2) |
54aa1f4d | 989 | err_on_enospc = 1; |
bb803951 | 990 | } |
79f95c82 CM |
991 | |
992 | /* | |
993 | * then try to empty the right most buffer into the middle | |
994 | */ | |
5f39d397 | 995 | if (right) { |
971a1f66 | 996 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 997 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 998 | ret = wret; |
5f39d397 | 999 | if (btrfs_header_nritems(right) == 0) { |
db94535d | 1000 | u64 bytenr = right->start; |
7bb86316 | 1001 | u64 generation = btrfs_header_generation(parent); |
db94535d CM |
1002 | u32 blocksize = right->len; |
1003 | ||
5f39d397 | 1004 | clean_tree_block(trans, root, right); |
925baedd | 1005 | btrfs_tree_unlock(right); |
5f39d397 | 1006 | free_extent_buffer(right); |
bb803951 | 1007 | right = NULL; |
e089f05c CM |
1008 | wret = del_ptr(trans, root, path, level + 1, pslot + |
1009 | 1); | |
bb803951 CM |
1010 | if (wret) |
1011 | ret = wret; | |
db94535d | 1012 | wret = btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 1013 | blocksize, parent->start, |
7bb86316 | 1014 | btrfs_header_owner(parent), |
3bb1a1bc | 1015 | generation, level, 1); |
bb803951 CM |
1016 | if (wret) |
1017 | ret = wret; | |
1018 | } else { | |
5f39d397 CM |
1019 | struct btrfs_disk_key right_key; |
1020 | btrfs_node_key(right, &right_key, 0); | |
1021 | btrfs_set_node_key(parent, &right_key, pslot + 1); | |
1022 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1023 | } |
1024 | } | |
5f39d397 | 1025 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1026 | /* |
1027 | * we're not allowed to leave a node with one item in the | |
1028 | * tree during a delete. A deletion from lower in the tree | |
1029 | * could try to delete the only pointer in this node. | |
1030 | * So, pull some keys from the left. | |
1031 | * There has to be a left pointer at this point because | |
1032 | * otherwise we would have pulled some pointers from the | |
1033 | * right | |
1034 | */ | |
5f39d397 CM |
1035 | BUG_ON(!left); |
1036 | wret = balance_node_right(trans, root, mid, left); | |
54aa1f4d | 1037 | if (wret < 0) { |
79f95c82 | 1038 | ret = wret; |
54aa1f4d CM |
1039 | goto enospc; |
1040 | } | |
bce4eae9 CM |
1041 | if (wret == 1) { |
1042 | wret = push_node_left(trans, root, left, mid, 1); | |
1043 | if (wret < 0) | |
1044 | ret = wret; | |
1045 | } | |
79f95c82 CM |
1046 | BUG_ON(wret == 1); |
1047 | } | |
5f39d397 | 1048 | if (btrfs_header_nritems(mid) == 0) { |
79f95c82 | 1049 | /* we've managed to empty the middle node, drop it */ |
7bb86316 | 1050 | u64 root_gen = btrfs_header_generation(parent); |
db94535d CM |
1051 | u64 bytenr = mid->start; |
1052 | u32 blocksize = mid->len; | |
925baedd | 1053 | |
5f39d397 | 1054 | clean_tree_block(trans, root, mid); |
925baedd | 1055 | btrfs_tree_unlock(mid); |
5f39d397 | 1056 | free_extent_buffer(mid); |
bb803951 | 1057 | mid = NULL; |
e089f05c | 1058 | wret = del_ptr(trans, root, path, level + 1, pslot); |
bb803951 CM |
1059 | if (wret) |
1060 | ret = wret; | |
7bb86316 | 1061 | wret = btrfs_free_extent(trans, root, bytenr, blocksize, |
31840ae1 | 1062 | parent->start, |
7bb86316 | 1063 | btrfs_header_owner(parent), |
3bb1a1bc | 1064 | root_gen, level, 1); |
bb803951 CM |
1065 | if (wret) |
1066 | ret = wret; | |
79f95c82 CM |
1067 | } else { |
1068 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1069 | struct btrfs_disk_key mid_key; |
1070 | btrfs_node_key(mid, &mid_key, 0); | |
1071 | btrfs_set_node_key(parent, &mid_key, pslot); | |
1072 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1073 | } |
bb803951 | 1074 | |
79f95c82 | 1075 | /* update the path */ |
5f39d397 CM |
1076 | if (left) { |
1077 | if (btrfs_header_nritems(left) > orig_slot) { | |
1078 | extent_buffer_get(left); | |
925baedd | 1079 | /* left was locked after cow */ |
5f39d397 | 1080 | path->nodes[level] = left; |
bb803951 CM |
1081 | path->slots[level + 1] -= 1; |
1082 | path->slots[level] = orig_slot; | |
925baedd CM |
1083 | if (mid) { |
1084 | btrfs_tree_unlock(mid); | |
5f39d397 | 1085 | free_extent_buffer(mid); |
925baedd | 1086 | } |
bb803951 | 1087 | } else { |
5f39d397 | 1088 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1089 | path->slots[level] = orig_slot; |
1090 | } | |
1091 | } | |
79f95c82 | 1092 | /* double check we haven't messed things up */ |
123abc88 | 1093 | check_block(root, path, level); |
e20d96d6 | 1094 | if (orig_ptr != |
5f39d397 | 1095 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1096 | BUG(); |
54aa1f4d | 1097 | enospc: |
925baedd CM |
1098 | if (right) { |
1099 | btrfs_tree_unlock(right); | |
5f39d397 | 1100 | free_extent_buffer(right); |
925baedd CM |
1101 | } |
1102 | if (left) { | |
1103 | if (path->nodes[level] != left) | |
1104 | btrfs_tree_unlock(left); | |
5f39d397 | 1105 | free_extent_buffer(left); |
925baedd | 1106 | } |
bb803951 CM |
1107 | return ret; |
1108 | } | |
1109 | ||
d352ac68 CM |
1110 | /* Node balancing for insertion. Here we only split or push nodes around |
1111 | * when they are completely full. This is also done top down, so we | |
1112 | * have to be pessimistic. | |
1113 | */ | |
d397712b | 1114 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1115 | struct btrfs_root *root, |
1116 | struct btrfs_path *path, int level) | |
e66f709b | 1117 | { |
5f39d397 CM |
1118 | struct extent_buffer *right = NULL; |
1119 | struct extent_buffer *mid; | |
1120 | struct extent_buffer *left = NULL; | |
1121 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1122 | int ret = 0; |
1123 | int wret; | |
1124 | int pslot; | |
1125 | int orig_slot = path->slots[level]; | |
1126 | u64 orig_ptr; | |
1127 | ||
1128 | if (level == 0) | |
1129 | return 1; | |
1130 | ||
5f39d397 | 1131 | mid = path->nodes[level]; |
7bb86316 | 1132 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b CM |
1133 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
1134 | ||
1135 | if (level < BTRFS_MAX_LEVEL - 1) | |
5f39d397 | 1136 | parent = path->nodes[level + 1]; |
e66f709b CM |
1137 | pslot = path->slots[level + 1]; |
1138 | ||
5f39d397 | 1139 | if (!parent) |
e66f709b | 1140 | return 1; |
e66f709b | 1141 | |
5f39d397 | 1142 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1143 | |
1144 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1145 | if (left) { |
e66f709b | 1146 | u32 left_nr; |
925baedd CM |
1147 | |
1148 | btrfs_tree_lock(left); | |
b4ce94de CM |
1149 | btrfs_set_lock_blocking(left); |
1150 | ||
5f39d397 | 1151 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1152 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1153 | wret = 1; | |
1154 | } else { | |
5f39d397 | 1155 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1156 | pslot - 1, &left); |
54aa1f4d CM |
1157 | if (ret) |
1158 | wret = 1; | |
1159 | else { | |
54aa1f4d | 1160 | wret = push_node_left(trans, root, |
971a1f66 | 1161 | left, mid, 0); |
54aa1f4d | 1162 | } |
33ade1f8 | 1163 | } |
e66f709b CM |
1164 | if (wret < 0) |
1165 | ret = wret; | |
1166 | if (wret == 0) { | |
5f39d397 | 1167 | struct btrfs_disk_key disk_key; |
e66f709b | 1168 | orig_slot += left_nr; |
5f39d397 CM |
1169 | btrfs_node_key(mid, &disk_key, 0); |
1170 | btrfs_set_node_key(parent, &disk_key, pslot); | |
1171 | btrfs_mark_buffer_dirty(parent); | |
1172 | if (btrfs_header_nritems(left) > orig_slot) { | |
1173 | path->nodes[level] = left; | |
e66f709b CM |
1174 | path->slots[level + 1] -= 1; |
1175 | path->slots[level] = orig_slot; | |
925baedd | 1176 | btrfs_tree_unlock(mid); |
5f39d397 | 1177 | free_extent_buffer(mid); |
e66f709b CM |
1178 | } else { |
1179 | orig_slot -= | |
5f39d397 | 1180 | btrfs_header_nritems(left); |
e66f709b | 1181 | path->slots[level] = orig_slot; |
925baedd | 1182 | btrfs_tree_unlock(left); |
5f39d397 | 1183 | free_extent_buffer(left); |
e66f709b | 1184 | } |
e66f709b CM |
1185 | return 0; |
1186 | } | |
925baedd | 1187 | btrfs_tree_unlock(left); |
5f39d397 | 1188 | free_extent_buffer(left); |
e66f709b | 1189 | } |
925baedd | 1190 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1191 | |
1192 | /* | |
1193 | * then try to empty the right most buffer into the middle | |
1194 | */ | |
5f39d397 | 1195 | if (right) { |
33ade1f8 | 1196 | u32 right_nr; |
b4ce94de | 1197 | |
925baedd | 1198 | btrfs_tree_lock(right); |
b4ce94de CM |
1199 | btrfs_set_lock_blocking(right); |
1200 | ||
5f39d397 | 1201 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1202 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1203 | wret = 1; | |
1204 | } else { | |
5f39d397 CM |
1205 | ret = btrfs_cow_block(trans, root, right, |
1206 | parent, pslot + 1, | |
9fa8cfe7 | 1207 | &right); |
54aa1f4d CM |
1208 | if (ret) |
1209 | wret = 1; | |
1210 | else { | |
54aa1f4d | 1211 | wret = balance_node_right(trans, root, |
5f39d397 | 1212 | right, mid); |
54aa1f4d | 1213 | } |
33ade1f8 | 1214 | } |
e66f709b CM |
1215 | if (wret < 0) |
1216 | ret = wret; | |
1217 | if (wret == 0) { | |
5f39d397 CM |
1218 | struct btrfs_disk_key disk_key; |
1219 | ||
1220 | btrfs_node_key(right, &disk_key, 0); | |
1221 | btrfs_set_node_key(parent, &disk_key, pslot + 1); | |
1222 | btrfs_mark_buffer_dirty(parent); | |
1223 | ||
1224 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1225 | path->nodes[level] = right; | |
e66f709b CM |
1226 | path->slots[level + 1] += 1; |
1227 | path->slots[level] = orig_slot - | |
5f39d397 | 1228 | btrfs_header_nritems(mid); |
925baedd | 1229 | btrfs_tree_unlock(mid); |
5f39d397 | 1230 | free_extent_buffer(mid); |
e66f709b | 1231 | } else { |
925baedd | 1232 | btrfs_tree_unlock(right); |
5f39d397 | 1233 | free_extent_buffer(right); |
e66f709b | 1234 | } |
e66f709b CM |
1235 | return 0; |
1236 | } | |
925baedd | 1237 | btrfs_tree_unlock(right); |
5f39d397 | 1238 | free_extent_buffer(right); |
e66f709b | 1239 | } |
e66f709b CM |
1240 | return 1; |
1241 | } | |
1242 | ||
3c69faec | 1243 | /* |
d352ac68 CM |
1244 | * readahead one full node of leaves, finding things that are close |
1245 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1246 | */ |
c8c42864 CM |
1247 | static void reada_for_search(struct btrfs_root *root, |
1248 | struct btrfs_path *path, | |
1249 | int level, int slot, u64 objectid) | |
3c69faec | 1250 | { |
5f39d397 | 1251 | struct extent_buffer *node; |
01f46658 | 1252 | struct btrfs_disk_key disk_key; |
3c69faec | 1253 | u32 nritems; |
3c69faec | 1254 | u64 search; |
a7175319 | 1255 | u64 target; |
6b80053d | 1256 | u64 nread = 0; |
3c69faec | 1257 | int direction = path->reada; |
5f39d397 | 1258 | struct extent_buffer *eb; |
6b80053d CM |
1259 | u32 nr; |
1260 | u32 blocksize; | |
1261 | u32 nscan = 0; | |
db94535d | 1262 | |
a6b6e75e | 1263 | if (level != 1) |
6702ed49 CM |
1264 | return; |
1265 | ||
1266 | if (!path->nodes[level]) | |
3c69faec CM |
1267 | return; |
1268 | ||
5f39d397 | 1269 | node = path->nodes[level]; |
925baedd | 1270 | |
3c69faec | 1271 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1272 | blocksize = btrfs_level_size(root, level - 1); |
1273 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1274 | if (eb) { |
1275 | free_extent_buffer(eb); | |
3c69faec CM |
1276 | return; |
1277 | } | |
1278 | ||
a7175319 | 1279 | target = search; |
6b80053d | 1280 | |
5f39d397 | 1281 | nritems = btrfs_header_nritems(node); |
6b80053d | 1282 | nr = slot; |
d397712b | 1283 | while (1) { |
6b80053d CM |
1284 | if (direction < 0) { |
1285 | if (nr == 0) | |
1286 | break; | |
1287 | nr--; | |
1288 | } else if (direction > 0) { | |
1289 | nr++; | |
1290 | if (nr >= nritems) | |
1291 | break; | |
3c69faec | 1292 | } |
01f46658 CM |
1293 | if (path->reada < 0 && objectid) { |
1294 | btrfs_node_key(node, &disk_key, nr); | |
1295 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1296 | break; | |
1297 | } | |
6b80053d | 1298 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
1299 | if ((search <= target && target - search <= 65536) || |
1300 | (search > target && search - target <= 65536)) { | |
ca7a79ad CM |
1301 | readahead_tree_block(root, search, blocksize, |
1302 | btrfs_node_ptr_generation(node, nr)); | |
6b80053d CM |
1303 | nread += blocksize; |
1304 | } | |
1305 | nscan++; | |
a7175319 | 1306 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 1307 | break; |
3c69faec CM |
1308 | } |
1309 | } | |
925baedd | 1310 | |
b4ce94de CM |
1311 | /* |
1312 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
1313 | * cache | |
1314 | */ | |
1315 | static noinline int reada_for_balance(struct btrfs_root *root, | |
1316 | struct btrfs_path *path, int level) | |
1317 | { | |
1318 | int slot; | |
1319 | int nritems; | |
1320 | struct extent_buffer *parent; | |
1321 | struct extent_buffer *eb; | |
1322 | u64 gen; | |
1323 | u64 block1 = 0; | |
1324 | u64 block2 = 0; | |
1325 | int ret = 0; | |
1326 | int blocksize; | |
1327 | ||
8c594ea8 | 1328 | parent = path->nodes[level + 1]; |
b4ce94de CM |
1329 | if (!parent) |
1330 | return 0; | |
1331 | ||
1332 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1333 | slot = path->slots[level + 1]; |
b4ce94de CM |
1334 | blocksize = btrfs_level_size(root, level); |
1335 | ||
1336 | if (slot > 0) { | |
1337 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
1338 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
1339 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
1340 | if (eb && btrfs_buffer_uptodate(eb, gen)) | |
1341 | block1 = 0; | |
1342 | free_extent_buffer(eb); | |
1343 | } | |
8c594ea8 | 1344 | if (slot + 1 < nritems) { |
b4ce94de CM |
1345 | block2 = btrfs_node_blockptr(parent, slot + 1); |
1346 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
1347 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
1348 | if (eb && btrfs_buffer_uptodate(eb, gen)) | |
1349 | block2 = 0; | |
1350 | free_extent_buffer(eb); | |
1351 | } | |
1352 | if (block1 || block2) { | |
1353 | ret = -EAGAIN; | |
8c594ea8 CM |
1354 | |
1355 | /* release the whole path */ | |
b4ce94de | 1356 | btrfs_release_path(root, path); |
8c594ea8 CM |
1357 | |
1358 | /* read the blocks */ | |
b4ce94de CM |
1359 | if (block1) |
1360 | readahead_tree_block(root, block1, blocksize, 0); | |
1361 | if (block2) | |
1362 | readahead_tree_block(root, block2, blocksize, 0); | |
1363 | ||
1364 | if (block1) { | |
1365 | eb = read_tree_block(root, block1, blocksize, 0); | |
1366 | free_extent_buffer(eb); | |
1367 | } | |
8c594ea8 | 1368 | if (block2) { |
b4ce94de CM |
1369 | eb = read_tree_block(root, block2, blocksize, 0); |
1370 | free_extent_buffer(eb); | |
1371 | } | |
1372 | } | |
1373 | return ret; | |
1374 | } | |
1375 | ||
1376 | ||
d352ac68 | 1377 | /* |
d397712b CM |
1378 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1379 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1380 | * operations on the tree might require changing key pointers higher up in the | |
1381 | * tree. | |
d352ac68 | 1382 | * |
d397712b CM |
1383 | * callers might also have set path->keep_locks, which tells this code to keep |
1384 | * the lock if the path points to the last slot in the block. This is part of | |
1385 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1386 | * |
d397712b CM |
1387 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1388 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1389 | */ |
e02119d5 CM |
1390 | static noinline void unlock_up(struct btrfs_path *path, int level, |
1391 | int lowest_unlock) | |
925baedd CM |
1392 | { |
1393 | int i; | |
1394 | int skip_level = level; | |
051e1b9f | 1395 | int no_skips = 0; |
925baedd CM |
1396 | struct extent_buffer *t; |
1397 | ||
1398 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1399 | if (!path->nodes[i]) | |
1400 | break; | |
1401 | if (!path->locks[i]) | |
1402 | break; | |
051e1b9f | 1403 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1404 | skip_level = i + 1; |
1405 | continue; | |
1406 | } | |
051e1b9f | 1407 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1408 | u32 nritems; |
1409 | t = path->nodes[i]; | |
1410 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1411 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1412 | skip_level = i + 1; |
1413 | continue; | |
1414 | } | |
1415 | } | |
051e1b9f CM |
1416 | if (skip_level < i && i >= lowest_unlock) |
1417 | no_skips = 1; | |
1418 | ||
925baedd CM |
1419 | t = path->nodes[i]; |
1420 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
1421 | btrfs_tree_unlock(t); | |
1422 | path->locks[i] = 0; | |
1423 | } | |
1424 | } | |
1425 | } | |
1426 | ||
b4ce94de CM |
1427 | /* |
1428 | * This releases any locks held in the path starting at level and | |
1429 | * going all the way up to the root. | |
1430 | * | |
1431 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
1432 | * corner cases, such as COW of the block at slot zero in the node. This | |
1433 | * ignores those rules, and it should only be called when there are no | |
1434 | * more updates to be done higher up in the tree. | |
1435 | */ | |
1436 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
1437 | { | |
1438 | int i; | |
1439 | ||
1440 | if (path->keep_locks || path->lowest_level) | |
1441 | return; | |
1442 | ||
1443 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1444 | if (!path->nodes[i]) | |
12f4dacc | 1445 | continue; |
b4ce94de | 1446 | if (!path->locks[i]) |
12f4dacc | 1447 | continue; |
b4ce94de CM |
1448 | btrfs_tree_unlock(path->nodes[i]); |
1449 | path->locks[i] = 0; | |
1450 | } | |
1451 | } | |
1452 | ||
c8c42864 CM |
1453 | /* |
1454 | * helper function for btrfs_search_slot. The goal is to find a block | |
1455 | * in cache without setting the path to blocking. If we find the block | |
1456 | * we return zero and the path is unchanged. | |
1457 | * | |
1458 | * If we can't find the block, we set the path blocking and do some | |
1459 | * reada. -EAGAIN is returned and the search must be repeated. | |
1460 | */ | |
1461 | static int | |
1462 | read_block_for_search(struct btrfs_trans_handle *trans, | |
1463 | struct btrfs_root *root, struct btrfs_path *p, | |
1464 | struct extent_buffer **eb_ret, int level, int slot, | |
1465 | struct btrfs_key *key) | |
1466 | { | |
1467 | u64 blocknr; | |
1468 | u64 gen; | |
1469 | u32 blocksize; | |
1470 | struct extent_buffer *b = *eb_ret; | |
1471 | struct extent_buffer *tmp; | |
76a05b35 | 1472 | int ret; |
c8c42864 CM |
1473 | |
1474 | blocknr = btrfs_node_blockptr(b, slot); | |
1475 | gen = btrfs_node_ptr_generation(b, slot); | |
1476 | blocksize = btrfs_level_size(root, level - 1); | |
1477 | ||
1478 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
1479 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
76a05b35 CM |
1480 | /* |
1481 | * we found an up to date block without sleeping, return | |
1482 | * right away | |
1483 | */ | |
c8c42864 CM |
1484 | *eb_ret = tmp; |
1485 | return 0; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * reduce lock contention at high levels | |
1490 | * of the btree by dropping locks before | |
76a05b35 CM |
1491 | * we read. Don't release the lock on the current |
1492 | * level because we need to walk this node to figure | |
1493 | * out which blocks to read. | |
c8c42864 | 1494 | */ |
8c594ea8 CM |
1495 | btrfs_unlock_up_safe(p, level + 1); |
1496 | btrfs_set_path_blocking(p); | |
1497 | ||
c8c42864 CM |
1498 | if (tmp) |
1499 | free_extent_buffer(tmp); | |
1500 | if (p->reada) | |
1501 | reada_for_search(root, p, level, slot, key->objectid); | |
1502 | ||
8c594ea8 | 1503 | btrfs_release_path(NULL, p); |
76a05b35 CM |
1504 | |
1505 | ret = -EAGAIN; | |
c8c42864 | 1506 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
76a05b35 CM |
1507 | if (tmp) { |
1508 | /* | |
1509 | * If the read above didn't mark this buffer up to date, | |
1510 | * it will never end up being up to date. Set ret to EIO now | |
1511 | * and give up so that our caller doesn't loop forever | |
1512 | * on our EAGAINs. | |
1513 | */ | |
1514 | if (!btrfs_buffer_uptodate(tmp, 0)) | |
1515 | ret = -EIO; | |
c8c42864 | 1516 | free_extent_buffer(tmp); |
76a05b35 CM |
1517 | } |
1518 | return ret; | |
c8c42864 CM |
1519 | } |
1520 | ||
1521 | /* | |
1522 | * helper function for btrfs_search_slot. This does all of the checks | |
1523 | * for node-level blocks and does any balancing required based on | |
1524 | * the ins_len. | |
1525 | * | |
1526 | * If no extra work was required, zero is returned. If we had to | |
1527 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1528 | * start over | |
1529 | */ | |
1530 | static int | |
1531 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1532 | struct btrfs_root *root, struct btrfs_path *p, | |
1533 | struct extent_buffer *b, int level, int ins_len) | |
1534 | { | |
1535 | int ret; | |
1536 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
1537 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
1538 | int sret; | |
1539 | ||
1540 | sret = reada_for_balance(root, p, level); | |
1541 | if (sret) | |
1542 | goto again; | |
1543 | ||
1544 | btrfs_set_path_blocking(p); | |
1545 | sret = split_node(trans, root, p, level); | |
1546 | btrfs_clear_path_blocking(p, NULL); | |
1547 | ||
1548 | BUG_ON(sret > 0); | |
1549 | if (sret) { | |
1550 | ret = sret; | |
1551 | goto done; | |
1552 | } | |
1553 | b = p->nodes[level]; | |
1554 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
1555 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) { | |
1556 | int sret; | |
1557 | ||
1558 | sret = reada_for_balance(root, p, level); | |
1559 | if (sret) | |
1560 | goto again; | |
1561 | ||
1562 | btrfs_set_path_blocking(p); | |
1563 | sret = balance_level(trans, root, p, level); | |
1564 | btrfs_clear_path_blocking(p, NULL); | |
1565 | ||
1566 | if (sret) { | |
1567 | ret = sret; | |
1568 | goto done; | |
1569 | } | |
1570 | b = p->nodes[level]; | |
1571 | if (!b) { | |
1572 | btrfs_release_path(NULL, p); | |
1573 | goto again; | |
1574 | } | |
1575 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1576 | } | |
1577 | return 0; | |
1578 | ||
1579 | again: | |
1580 | ret = -EAGAIN; | |
1581 | done: | |
1582 | return ret; | |
1583 | } | |
1584 | ||
74123bd7 CM |
1585 | /* |
1586 | * look for key in the tree. path is filled in with nodes along the way | |
1587 | * if key is found, we return zero and you can find the item in the leaf | |
1588 | * level of the path (level 0) | |
1589 | * | |
1590 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
1591 | * be inserted, and 1 is returned. If there are other errors during the |
1592 | * search a negative error number is returned. | |
97571fd0 CM |
1593 | * |
1594 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
1595 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
1596 | * possible) | |
74123bd7 | 1597 | */ |
e089f05c CM |
1598 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
1599 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
1600 | ins_len, int cow) | |
be0e5c09 | 1601 | { |
5f39d397 | 1602 | struct extent_buffer *b; |
be0e5c09 CM |
1603 | int slot; |
1604 | int ret; | |
1605 | int level; | |
925baedd | 1606 | int lowest_unlock = 1; |
9f3a7427 CM |
1607 | u8 lowest_level = 0; |
1608 | ||
6702ed49 | 1609 | lowest_level = p->lowest_level; |
323ac95b | 1610 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1611 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 1612 | |
925baedd CM |
1613 | if (ins_len < 0) |
1614 | lowest_unlock = 2; | |
65b51a00 | 1615 | |
bb803951 | 1616 | again: |
5cd57b2c CM |
1617 | if (p->skip_locking) |
1618 | b = btrfs_root_node(root); | |
1619 | else | |
1620 | b = btrfs_lock_root_node(root); | |
925baedd | 1621 | |
eb60ceac | 1622 | while (b) { |
5f39d397 | 1623 | level = btrfs_header_level(b); |
65b51a00 CM |
1624 | |
1625 | /* | |
1626 | * setup the path here so we can release it under lock | |
1627 | * contention with the cow code | |
1628 | */ | |
1629 | p->nodes[level] = b; | |
1630 | if (!p->skip_locking) | |
1631 | p->locks[level] = 1; | |
1632 | ||
02217ed2 CM |
1633 | if (cow) { |
1634 | int wret; | |
65b51a00 | 1635 | |
c8c42864 CM |
1636 | /* |
1637 | * if we don't really need to cow this block | |
1638 | * then we don't want to set the path blocking, | |
1639 | * so we test it here | |
1640 | */ | |
65b51a00 | 1641 | if (btrfs_header_generation(b) == trans->transid && |
5b21f2ed | 1642 | btrfs_header_owner(b) == root->root_key.objectid && |
65b51a00 | 1643 | !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) { |
65b51a00 CM |
1644 | goto cow_done; |
1645 | } | |
b4ce94de CM |
1646 | btrfs_set_path_blocking(p); |
1647 | ||
e20d96d6 CM |
1648 | wret = btrfs_cow_block(trans, root, b, |
1649 | p->nodes[level + 1], | |
9fa8cfe7 | 1650 | p->slots[level + 1], &b); |
54aa1f4d | 1651 | if (wret) { |
5f39d397 | 1652 | free_extent_buffer(b); |
65b51a00 CM |
1653 | ret = wret; |
1654 | goto done; | |
54aa1f4d | 1655 | } |
02217ed2 | 1656 | } |
65b51a00 | 1657 | cow_done: |
02217ed2 | 1658 | BUG_ON(!cow && ins_len); |
5f39d397 | 1659 | if (level != btrfs_header_level(b)) |
2c90e5d6 | 1660 | WARN_ON(1); |
5f39d397 | 1661 | level = btrfs_header_level(b); |
65b51a00 | 1662 | |
eb60ceac | 1663 | p->nodes[level] = b; |
5cd57b2c CM |
1664 | if (!p->skip_locking) |
1665 | p->locks[level] = 1; | |
65b51a00 | 1666 | |
4008c04a | 1667 | btrfs_clear_path_blocking(p, NULL); |
b4ce94de CM |
1668 | |
1669 | /* | |
1670 | * we have a lock on b and as long as we aren't changing | |
1671 | * the tree, there is no way to for the items in b to change. | |
1672 | * It is safe to drop the lock on our parent before we | |
1673 | * go through the expensive btree search on b. | |
1674 | * | |
1675 | * If cow is true, then we might be changing slot zero, | |
1676 | * which may require changing the parent. So, we can't | |
1677 | * drop the lock until after we know which slot we're | |
1678 | * operating on. | |
1679 | */ | |
1680 | if (!cow) | |
1681 | btrfs_unlock_up_safe(p, level + 1); | |
1682 | ||
123abc88 | 1683 | ret = check_block(root, p, level); |
65b51a00 CM |
1684 | if (ret) { |
1685 | ret = -1; | |
1686 | goto done; | |
1687 | } | |
925baedd | 1688 | |
5f39d397 | 1689 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 1690 | |
5f39d397 | 1691 | if (level != 0) { |
be0e5c09 CM |
1692 | if (ret && slot > 0) |
1693 | slot -= 1; | |
1694 | p->slots[level] = slot; | |
c8c42864 CM |
1695 | ret = setup_nodes_for_search(trans, root, p, b, level, |
1696 | ins_len); | |
1697 | if (ret == -EAGAIN) | |
1698 | goto again; | |
1699 | else if (ret) | |
1700 | goto done; | |
1701 | b = p->nodes[level]; | |
1702 | slot = p->slots[level]; | |
b4ce94de | 1703 | |
f9efa9c7 CM |
1704 | unlock_up(p, level, lowest_unlock); |
1705 | ||
9f3a7427 | 1706 | /* this is only true while dropping a snapshot */ |
925baedd | 1707 | if (level == lowest_level) { |
5b21f2ed ZY |
1708 | ret = 0; |
1709 | goto done; | |
925baedd | 1710 | } |
ca7a79ad | 1711 | |
c8c42864 CM |
1712 | ret = read_block_for_search(trans, root, p, |
1713 | &b, level, slot, key); | |
1714 | if (ret == -EAGAIN) | |
1715 | goto again; | |
594a24eb | 1716 | |
76a05b35 CM |
1717 | if (ret == -EIO) |
1718 | goto done; | |
1719 | ||
b4ce94de CM |
1720 | if (!p->skip_locking) { |
1721 | int lret; | |
1722 | ||
4008c04a | 1723 | btrfs_clear_path_blocking(p, NULL); |
b4ce94de CM |
1724 | lret = btrfs_try_spin_lock(b); |
1725 | ||
1726 | if (!lret) { | |
1727 | btrfs_set_path_blocking(p); | |
1728 | btrfs_tree_lock(b); | |
4008c04a | 1729 | btrfs_clear_path_blocking(p, b); |
b4ce94de CM |
1730 | } |
1731 | } | |
be0e5c09 CM |
1732 | } else { |
1733 | p->slots[level] = slot; | |
87b29b20 YZ |
1734 | if (ins_len > 0 && |
1735 | btrfs_leaf_free_space(root, b) < ins_len) { | |
b4ce94de CM |
1736 | int sret; |
1737 | ||
1738 | btrfs_set_path_blocking(p); | |
1739 | sret = split_leaf(trans, root, key, | |
cc0c5538 | 1740 | p, ins_len, ret == 0); |
4008c04a | 1741 | btrfs_clear_path_blocking(p, NULL); |
b4ce94de | 1742 | |
5c680ed6 | 1743 | BUG_ON(sret > 0); |
65b51a00 CM |
1744 | if (sret) { |
1745 | ret = sret; | |
1746 | goto done; | |
1747 | } | |
5c680ed6 | 1748 | } |
459931ec CM |
1749 | if (!p->search_for_split) |
1750 | unlock_up(p, level, lowest_unlock); | |
65b51a00 | 1751 | goto done; |
be0e5c09 CM |
1752 | } |
1753 | } | |
65b51a00 CM |
1754 | ret = 1; |
1755 | done: | |
b4ce94de CM |
1756 | /* |
1757 | * we don't really know what they plan on doing with the path | |
1758 | * from here on, so for now just mark it as blocking | |
1759 | */ | |
b9473439 CM |
1760 | if (!p->leave_spinning) |
1761 | btrfs_set_path_blocking(p); | |
76a05b35 CM |
1762 | if (ret < 0) |
1763 | btrfs_release_path(root, p); | |
65b51a00 | 1764 | return ret; |
be0e5c09 CM |
1765 | } |
1766 | ||
1a40e23b ZY |
1767 | int btrfs_merge_path(struct btrfs_trans_handle *trans, |
1768 | struct btrfs_root *root, | |
1769 | struct btrfs_key *node_keys, | |
1770 | u64 *nodes, int lowest_level) | |
1771 | { | |
1772 | struct extent_buffer *eb; | |
1773 | struct extent_buffer *parent; | |
1774 | struct btrfs_key key; | |
1775 | u64 bytenr; | |
1776 | u64 generation; | |
1777 | u32 blocksize; | |
1778 | int level; | |
1779 | int slot; | |
1780 | int key_match; | |
1781 | int ret; | |
1782 | ||
1783 | eb = btrfs_lock_root_node(root); | |
9fa8cfe7 | 1784 | ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb); |
1a40e23b ZY |
1785 | BUG_ON(ret); |
1786 | ||
b4ce94de CM |
1787 | btrfs_set_lock_blocking(eb); |
1788 | ||
1a40e23b ZY |
1789 | parent = eb; |
1790 | while (1) { | |
1791 | level = btrfs_header_level(parent); | |
1792 | if (level == 0 || level <= lowest_level) | |
1793 | break; | |
1794 | ||
1795 | ret = bin_search(parent, &node_keys[lowest_level], level, | |
1796 | &slot); | |
1797 | if (ret && slot > 0) | |
1798 | slot--; | |
1799 | ||
1800 | bytenr = btrfs_node_blockptr(parent, slot); | |
1801 | if (nodes[level - 1] == bytenr) | |
1802 | break; | |
1803 | ||
1804 | blocksize = btrfs_level_size(root, level - 1); | |
1805 | generation = btrfs_node_ptr_generation(parent, slot); | |
1806 | btrfs_node_key_to_cpu(eb, &key, slot); | |
1807 | key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key)); | |
1808 | ||
f82d02d9 YZ |
1809 | if (generation == trans->transid) { |
1810 | eb = read_tree_block(root, bytenr, blocksize, | |
1811 | generation); | |
1812 | btrfs_tree_lock(eb); | |
b4ce94de | 1813 | btrfs_set_lock_blocking(eb); |
f82d02d9 YZ |
1814 | } |
1815 | ||
1a40e23b ZY |
1816 | /* |
1817 | * if node keys match and node pointer hasn't been modified | |
1818 | * in the running transaction, we can merge the path. for | |
f82d02d9 YZ |
1819 | * blocks owened by reloc trees, the node pointer check is |
1820 | * skipped, this is because these blocks are fully controlled | |
1821 | * by the space balance code, no one else can modify them. | |
1a40e23b ZY |
1822 | */ |
1823 | if (!nodes[level - 1] || !key_match || | |
1824 | (generation == trans->transid && | |
f82d02d9 YZ |
1825 | btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) { |
1826 | if (level == 1 || level == lowest_level + 1) { | |
1827 | if (generation == trans->transid) { | |
1828 | btrfs_tree_unlock(eb); | |
1829 | free_extent_buffer(eb); | |
1830 | } | |
1a40e23b | 1831 | break; |
f82d02d9 | 1832 | } |
1a40e23b | 1833 | |
f82d02d9 YZ |
1834 | if (generation != trans->transid) { |
1835 | eb = read_tree_block(root, bytenr, blocksize, | |
1836 | generation); | |
1837 | btrfs_tree_lock(eb); | |
b4ce94de | 1838 | btrfs_set_lock_blocking(eb); |
f82d02d9 | 1839 | } |
1a40e23b ZY |
1840 | |
1841 | ret = btrfs_cow_block(trans, root, eb, parent, slot, | |
9fa8cfe7 | 1842 | &eb); |
1a40e23b ZY |
1843 | BUG_ON(ret); |
1844 | ||
f82d02d9 YZ |
1845 | if (root->root_key.objectid == |
1846 | BTRFS_TREE_RELOC_OBJECTID) { | |
1847 | if (!nodes[level - 1]) { | |
1848 | nodes[level - 1] = eb->start; | |
1849 | memcpy(&node_keys[level - 1], &key, | |
1850 | sizeof(node_keys[0])); | |
1851 | } else { | |
1852 | WARN_ON(1); | |
1853 | } | |
1854 | } | |
1855 | ||
1a40e23b ZY |
1856 | btrfs_tree_unlock(parent); |
1857 | free_extent_buffer(parent); | |
1858 | parent = eb; | |
1859 | continue; | |
1860 | } | |
1861 | ||
1a40e23b ZY |
1862 | btrfs_set_node_blockptr(parent, slot, nodes[level - 1]); |
1863 | btrfs_set_node_ptr_generation(parent, slot, trans->transid); | |
1864 | btrfs_mark_buffer_dirty(parent); | |
1865 | ||
1866 | ret = btrfs_inc_extent_ref(trans, root, | |
1867 | nodes[level - 1], | |
1868 | blocksize, parent->start, | |
1869 | btrfs_header_owner(parent), | |
1870 | btrfs_header_generation(parent), | |
3bb1a1bc | 1871 | level - 1); |
1a40e23b | 1872 | BUG_ON(ret); |
1a40e23b | 1873 | |
f82d02d9 YZ |
1874 | /* |
1875 | * If the block was created in the running transaction, | |
1876 | * it's possible this is the last reference to it, so we | |
1877 | * should drop the subtree. | |
1878 | */ | |
1a40e23b | 1879 | if (generation == trans->transid) { |
f82d02d9 YZ |
1880 | ret = btrfs_drop_subtree(trans, root, eb, parent); |
1881 | BUG_ON(ret); | |
1a40e23b ZY |
1882 | btrfs_tree_unlock(eb); |
1883 | free_extent_buffer(eb); | |
f82d02d9 YZ |
1884 | } else { |
1885 | ret = btrfs_free_extent(trans, root, bytenr, | |
1886 | blocksize, parent->start, | |
1887 | btrfs_header_owner(parent), | |
1888 | btrfs_header_generation(parent), | |
1889 | level - 1, 1); | |
1890 | BUG_ON(ret); | |
1a40e23b ZY |
1891 | } |
1892 | break; | |
1893 | } | |
1894 | btrfs_tree_unlock(parent); | |
1895 | free_extent_buffer(parent); | |
1896 | return 0; | |
1897 | } | |
1898 | ||
74123bd7 CM |
1899 | /* |
1900 | * adjust the pointers going up the tree, starting at level | |
1901 | * making sure the right key of each node is points to 'key'. | |
1902 | * This is used after shifting pointers to the left, so it stops | |
1903 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
1904 | * higher levels | |
aa5d6bed CM |
1905 | * |
1906 | * If this fails to write a tree block, it returns -1, but continues | |
1907 | * fixing up the blocks in ram so the tree is consistent. | |
74123bd7 | 1908 | */ |
5f39d397 CM |
1909 | static int fixup_low_keys(struct btrfs_trans_handle *trans, |
1910 | struct btrfs_root *root, struct btrfs_path *path, | |
1911 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
1912 | { |
1913 | int i; | |
aa5d6bed | 1914 | int ret = 0; |
5f39d397 CM |
1915 | struct extent_buffer *t; |
1916 | ||
234b63a0 | 1917 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 1918 | int tslot = path->slots[i]; |
eb60ceac | 1919 | if (!path->nodes[i]) |
be0e5c09 | 1920 | break; |
5f39d397 CM |
1921 | t = path->nodes[i]; |
1922 | btrfs_set_node_key(t, key, tslot); | |
d6025579 | 1923 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
1924 | if (tslot != 0) |
1925 | break; | |
1926 | } | |
aa5d6bed | 1927 | return ret; |
be0e5c09 CM |
1928 | } |
1929 | ||
31840ae1 ZY |
1930 | /* |
1931 | * update item key. | |
1932 | * | |
1933 | * This function isn't completely safe. It's the caller's responsibility | |
1934 | * that the new key won't break the order | |
1935 | */ | |
1936 | int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, | |
1937 | struct btrfs_root *root, struct btrfs_path *path, | |
1938 | struct btrfs_key *new_key) | |
1939 | { | |
1940 | struct btrfs_disk_key disk_key; | |
1941 | struct extent_buffer *eb; | |
1942 | int slot; | |
1943 | ||
1944 | eb = path->nodes[0]; | |
1945 | slot = path->slots[0]; | |
1946 | if (slot > 0) { | |
1947 | btrfs_item_key(eb, &disk_key, slot - 1); | |
1948 | if (comp_keys(&disk_key, new_key) >= 0) | |
1949 | return -1; | |
1950 | } | |
1951 | if (slot < btrfs_header_nritems(eb) - 1) { | |
1952 | btrfs_item_key(eb, &disk_key, slot + 1); | |
1953 | if (comp_keys(&disk_key, new_key) <= 0) | |
1954 | return -1; | |
1955 | } | |
1956 | ||
1957 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
1958 | btrfs_set_item_key(eb, &disk_key, slot); | |
1959 | btrfs_mark_buffer_dirty(eb); | |
1960 | if (slot == 0) | |
1961 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
1962 | return 0; | |
1963 | } | |
1964 | ||
74123bd7 CM |
1965 | /* |
1966 | * try to push data from one node into the next node left in the | |
79f95c82 | 1967 | * tree. |
aa5d6bed CM |
1968 | * |
1969 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
1970 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 1971 | */ |
98ed5174 CM |
1972 | static int push_node_left(struct btrfs_trans_handle *trans, |
1973 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 1974 | struct extent_buffer *src, int empty) |
be0e5c09 | 1975 | { |
be0e5c09 | 1976 | int push_items = 0; |
bb803951 CM |
1977 | int src_nritems; |
1978 | int dst_nritems; | |
aa5d6bed | 1979 | int ret = 0; |
be0e5c09 | 1980 | |
5f39d397 CM |
1981 | src_nritems = btrfs_header_nritems(src); |
1982 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 1983 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
1984 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
1985 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 1986 | |
bce4eae9 | 1987 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
1988 | return 1; |
1989 | ||
d397712b | 1990 | if (push_items <= 0) |
be0e5c09 CM |
1991 | return 1; |
1992 | ||
bce4eae9 | 1993 | if (empty) { |
971a1f66 | 1994 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
1995 | if (push_items < src_nritems) { |
1996 | /* leave at least 8 pointers in the node if | |
1997 | * we aren't going to empty it | |
1998 | */ | |
1999 | if (src_nritems - push_items < 8) { | |
2000 | if (push_items <= 8) | |
2001 | return 1; | |
2002 | push_items -= 8; | |
2003 | } | |
2004 | } | |
2005 | } else | |
2006 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2007 | |
5f39d397 CM |
2008 | copy_extent_buffer(dst, src, |
2009 | btrfs_node_key_ptr_offset(dst_nritems), | |
2010 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2011 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2012 | |
bb803951 | 2013 | if (push_items < src_nritems) { |
5f39d397 CM |
2014 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2015 | btrfs_node_key_ptr_offset(push_items), | |
2016 | (src_nritems - push_items) * | |
2017 | sizeof(struct btrfs_key_ptr)); | |
2018 | } | |
2019 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2020 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2021 | btrfs_mark_buffer_dirty(src); | |
2022 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 ZY |
2023 | |
2024 | ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items); | |
2025 | BUG_ON(ret); | |
2026 | ||
79f95c82 CM |
2027 | return ret; |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * try to push data from one node into the next node right in the | |
2032 | * tree. | |
2033 | * | |
2034 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2035 | * error, and > 0 if there was no room in the right hand block. | |
2036 | * | |
2037 | * this will only push up to 1/2 the contents of the left node over | |
2038 | */ | |
5f39d397 CM |
2039 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2040 | struct btrfs_root *root, | |
2041 | struct extent_buffer *dst, | |
2042 | struct extent_buffer *src) | |
79f95c82 | 2043 | { |
79f95c82 CM |
2044 | int push_items = 0; |
2045 | int max_push; | |
2046 | int src_nritems; | |
2047 | int dst_nritems; | |
2048 | int ret = 0; | |
79f95c82 | 2049 | |
7bb86316 CM |
2050 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2051 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2052 | ||
5f39d397 CM |
2053 | src_nritems = btrfs_header_nritems(src); |
2054 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2055 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 2056 | if (push_items <= 0) |
79f95c82 | 2057 | return 1; |
bce4eae9 | 2058 | |
d397712b | 2059 | if (src_nritems < 4) |
bce4eae9 | 2060 | return 1; |
79f95c82 CM |
2061 | |
2062 | max_push = src_nritems / 2 + 1; | |
2063 | /* don't try to empty the node */ | |
d397712b | 2064 | if (max_push >= src_nritems) |
79f95c82 | 2065 | return 1; |
252c38f0 | 2066 | |
79f95c82 CM |
2067 | if (max_push < push_items) |
2068 | push_items = max_push; | |
2069 | ||
5f39d397 CM |
2070 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2071 | btrfs_node_key_ptr_offset(0), | |
2072 | (dst_nritems) * | |
2073 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2074 | |
5f39d397 CM |
2075 | copy_extent_buffer(dst, src, |
2076 | btrfs_node_key_ptr_offset(0), | |
2077 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2078 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2079 | |
5f39d397 CM |
2080 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2081 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2082 | |
5f39d397 CM |
2083 | btrfs_mark_buffer_dirty(src); |
2084 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 ZY |
2085 | |
2086 | ret = btrfs_update_ref(trans, root, src, dst, 0, push_items); | |
2087 | BUG_ON(ret); | |
2088 | ||
aa5d6bed | 2089 | return ret; |
be0e5c09 CM |
2090 | } |
2091 | ||
97571fd0 CM |
2092 | /* |
2093 | * helper function to insert a new root level in the tree. | |
2094 | * A new node is allocated, and a single item is inserted to | |
2095 | * point to the existing root | |
aa5d6bed CM |
2096 | * |
2097 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2098 | */ |
d397712b | 2099 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2100 | struct btrfs_root *root, |
2101 | struct btrfs_path *path, int level) | |
5c680ed6 | 2102 | { |
7bb86316 | 2103 | u64 lower_gen; |
5f39d397 CM |
2104 | struct extent_buffer *lower; |
2105 | struct extent_buffer *c; | |
925baedd | 2106 | struct extent_buffer *old; |
5f39d397 | 2107 | struct btrfs_disk_key lower_key; |
31840ae1 | 2108 | int ret; |
5c680ed6 CM |
2109 | |
2110 | BUG_ON(path->nodes[level]); | |
2111 | BUG_ON(path->nodes[level-1] != root->node); | |
2112 | ||
7bb86316 CM |
2113 | lower = path->nodes[level-1]; |
2114 | if (level == 1) | |
2115 | btrfs_item_key(lower, &lower_key, 0); | |
2116 | else | |
2117 | btrfs_node_key(lower, &lower_key, 0); | |
2118 | ||
31840ae1 ZY |
2119 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
2120 | root->root_key.objectid, trans->transid, | |
ad3d81ba | 2121 | level, root->node->start, 0); |
5f39d397 CM |
2122 | if (IS_ERR(c)) |
2123 | return PTR_ERR(c); | |
925baedd | 2124 | |
5f39d397 CM |
2125 | memset_extent_buffer(c, 0, 0, root->nodesize); |
2126 | btrfs_set_header_nritems(c, 1); | |
2127 | btrfs_set_header_level(c, level); | |
db94535d | 2128 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 CM |
2129 | btrfs_set_header_generation(c, trans->transid); |
2130 | btrfs_set_header_owner(c, root->root_key.objectid); | |
5f39d397 CM |
2131 | |
2132 | write_extent_buffer(c, root->fs_info->fsid, | |
2133 | (unsigned long)btrfs_header_fsid(c), | |
2134 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2135 | |
2136 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
2137 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
2138 | BTRFS_UUID_SIZE); | |
2139 | ||
5f39d397 | 2140 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2141 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2142 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2143 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2144 | |
2145 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2146 | |
5f39d397 | 2147 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2148 | |
925baedd CM |
2149 | spin_lock(&root->node_lock); |
2150 | old = root->node; | |
5f39d397 | 2151 | root->node = c; |
925baedd CM |
2152 | spin_unlock(&root->node_lock); |
2153 | ||
31840ae1 | 2154 | ret = btrfs_update_extent_ref(trans, root, lower->start, |
56bec294 | 2155 | lower->len, lower->start, c->start, |
31840ae1 | 2156 | root->root_key.objectid, |
3bb1a1bc | 2157 | trans->transid, level - 1); |
31840ae1 ZY |
2158 | BUG_ON(ret); |
2159 | ||
925baedd CM |
2160 | /* the super has an extra ref to root->node */ |
2161 | free_extent_buffer(old); | |
2162 | ||
0b86a832 | 2163 | add_root_to_dirty_list(root); |
5f39d397 CM |
2164 | extent_buffer_get(c); |
2165 | path->nodes[level] = c; | |
925baedd | 2166 | path->locks[level] = 1; |
5c680ed6 CM |
2167 | path->slots[level] = 0; |
2168 | return 0; | |
2169 | } | |
2170 | ||
74123bd7 CM |
2171 | /* |
2172 | * worker function to insert a single pointer in a node. | |
2173 | * the node should have enough room for the pointer already | |
97571fd0 | 2174 | * |
74123bd7 CM |
2175 | * slot and level indicate where you want the key to go, and |
2176 | * blocknr is the block the key points to. | |
aa5d6bed CM |
2177 | * |
2178 | * returns zero on success and < 0 on any error | |
74123bd7 | 2179 | */ |
e089f05c CM |
2180 | static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root |
2181 | *root, struct btrfs_path *path, struct btrfs_disk_key | |
db94535d | 2182 | *key, u64 bytenr, int slot, int level) |
74123bd7 | 2183 | { |
5f39d397 | 2184 | struct extent_buffer *lower; |
74123bd7 | 2185 | int nritems; |
5c680ed6 CM |
2186 | |
2187 | BUG_ON(!path->nodes[level]); | |
5f39d397 CM |
2188 | lower = path->nodes[level]; |
2189 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2190 | BUG_ON(slot > nritems); |
123abc88 | 2191 | if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root)) |
74123bd7 CM |
2192 | BUG(); |
2193 | if (slot != nritems) { | |
5f39d397 CM |
2194 | memmove_extent_buffer(lower, |
2195 | btrfs_node_key_ptr_offset(slot + 1), | |
2196 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2197 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2198 | } |
5f39d397 | 2199 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2200 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2201 | WARN_ON(trans->transid == 0); |
2202 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2203 | btrfs_set_header_nritems(lower, nritems + 1); |
2204 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2205 | return 0; |
2206 | } | |
2207 | ||
97571fd0 CM |
2208 | /* |
2209 | * split the node at the specified level in path in two. | |
2210 | * The path is corrected to point to the appropriate node after the split | |
2211 | * | |
2212 | * Before splitting this tries to make some room in the node by pushing | |
2213 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2214 | * |
2215 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2216 | */ |
e02119d5 CM |
2217 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2218 | struct btrfs_root *root, | |
2219 | struct btrfs_path *path, int level) | |
be0e5c09 | 2220 | { |
5f39d397 CM |
2221 | struct extent_buffer *c; |
2222 | struct extent_buffer *split; | |
2223 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2224 | int mid; |
5c680ed6 | 2225 | int ret; |
aa5d6bed | 2226 | int wret; |
7518a238 | 2227 | u32 c_nritems; |
eb60ceac | 2228 | |
5f39d397 | 2229 | c = path->nodes[level]; |
7bb86316 | 2230 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2231 | if (c == root->node) { |
5c680ed6 | 2232 | /* trying to split the root, lets make a new one */ |
e089f05c | 2233 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2234 | if (ret) |
2235 | return ret; | |
a4b6e07d | 2236 | } else if (!trans->transaction->delayed_refs.flushing) { |
e66f709b | 2237 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2238 | c = path->nodes[level]; |
2239 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2240 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2241 | return 0; |
54aa1f4d CM |
2242 | if (ret < 0) |
2243 | return ret; | |
be0e5c09 | 2244 | } |
e66f709b | 2245 | |
5f39d397 | 2246 | c_nritems = btrfs_header_nritems(c); |
7bb86316 | 2247 | |
925baedd | 2248 | split = btrfs_alloc_free_block(trans, root, root->nodesize, |
31840ae1 ZY |
2249 | path->nodes[level + 1]->start, |
2250 | root->root_key.objectid, | |
2251 | trans->transid, level, c->start, 0); | |
5f39d397 CM |
2252 | if (IS_ERR(split)) |
2253 | return PTR_ERR(split); | |
2254 | ||
2255 | btrfs_set_header_flags(split, btrfs_header_flags(c)); | |
2256 | btrfs_set_header_level(split, btrfs_header_level(c)); | |
db94535d | 2257 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 CM |
2258 | btrfs_set_header_generation(split, trans->transid); |
2259 | btrfs_set_header_owner(split, root->root_key.objectid); | |
63b10fc4 | 2260 | btrfs_set_header_flags(split, 0); |
5f39d397 CM |
2261 | write_extent_buffer(split, root->fs_info->fsid, |
2262 | (unsigned long)btrfs_header_fsid(split), | |
2263 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2264 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2265 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2266 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2267 | |
7518a238 | 2268 | mid = (c_nritems + 1) / 2; |
5f39d397 CM |
2269 | |
2270 | copy_extent_buffer(split, c, | |
2271 | btrfs_node_key_ptr_offset(0), | |
2272 | btrfs_node_key_ptr_offset(mid), | |
2273 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2274 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2275 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2276 | ret = 0; |
2277 | ||
5f39d397 CM |
2278 | btrfs_mark_buffer_dirty(c); |
2279 | btrfs_mark_buffer_dirty(split); | |
2280 | ||
2281 | btrfs_node_key(split, &disk_key, 0); | |
db94535d | 2282 | wret = insert_ptr(trans, root, path, &disk_key, split->start, |
5f39d397 | 2283 | path->slots[level + 1] + 1, |
123abc88 | 2284 | level + 1); |
aa5d6bed CM |
2285 | if (wret) |
2286 | ret = wret; | |
2287 | ||
31840ae1 ZY |
2288 | ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid); |
2289 | BUG_ON(ret); | |
2290 | ||
5de08d7d | 2291 | if (path->slots[level] >= mid) { |
5c680ed6 | 2292 | path->slots[level] -= mid; |
925baedd | 2293 | btrfs_tree_unlock(c); |
5f39d397 CM |
2294 | free_extent_buffer(c); |
2295 | path->nodes[level] = split; | |
5c680ed6 CM |
2296 | path->slots[level + 1] += 1; |
2297 | } else { | |
925baedd | 2298 | btrfs_tree_unlock(split); |
5f39d397 | 2299 | free_extent_buffer(split); |
be0e5c09 | 2300 | } |
aa5d6bed | 2301 | return ret; |
be0e5c09 CM |
2302 | } |
2303 | ||
74123bd7 CM |
2304 | /* |
2305 | * how many bytes are required to store the items in a leaf. start | |
2306 | * and nr indicate which items in the leaf to check. This totals up the | |
2307 | * space used both by the item structs and the item data | |
2308 | */ | |
5f39d397 | 2309 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2310 | { |
2311 | int data_len; | |
5f39d397 | 2312 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2313 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2314 | |
2315 | if (!nr) | |
2316 | return 0; | |
5f39d397 CM |
2317 | data_len = btrfs_item_end_nr(l, start); |
2318 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2319 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2320 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2321 | return data_len; |
2322 | } | |
2323 | ||
d4dbff95 CM |
2324 | /* |
2325 | * The space between the end of the leaf items and | |
2326 | * the start of the leaf data. IOW, how much room | |
2327 | * the leaf has left for both items and data | |
2328 | */ | |
d397712b | 2329 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 2330 | struct extent_buffer *leaf) |
d4dbff95 | 2331 | { |
5f39d397 CM |
2332 | int nritems = btrfs_header_nritems(leaf); |
2333 | int ret; | |
2334 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2335 | if (ret < 0) { | |
d397712b CM |
2336 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
2337 | "used %d nritems %d\n", | |
ae2f5411 | 2338 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2339 | leaf_space_used(leaf, 0, nritems), nritems); |
2340 | } | |
2341 | return ret; | |
d4dbff95 CM |
2342 | } |
2343 | ||
44871b1b CM |
2344 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
2345 | struct btrfs_root *root, | |
2346 | struct btrfs_path *path, | |
2347 | int data_size, int empty, | |
2348 | struct extent_buffer *right, | |
2349 | int free_space, u32 left_nritems) | |
00ec4c51 | 2350 | { |
5f39d397 | 2351 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2352 | struct extent_buffer *upper = path->nodes[1]; |
5f39d397 | 2353 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2354 | int slot; |
34a38218 | 2355 | u32 i; |
00ec4c51 CM |
2356 | int push_space = 0; |
2357 | int push_items = 0; | |
0783fcfc | 2358 | struct btrfs_item *item; |
34a38218 | 2359 | u32 nr; |
7518a238 | 2360 | u32 right_nritems; |
5f39d397 | 2361 | u32 data_end; |
db94535d | 2362 | u32 this_item_size; |
54aa1f4d | 2363 | int ret; |
00ec4c51 | 2364 | |
34a38218 CM |
2365 | if (empty) |
2366 | nr = 0; | |
2367 | else | |
2368 | nr = 1; | |
2369 | ||
31840ae1 | 2370 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2371 | push_space += data_size; |
31840ae1 | 2372 | |
44871b1b | 2373 | slot = path->slots[1]; |
34a38218 CM |
2374 | i = left_nritems - 1; |
2375 | while (i >= nr) { | |
5f39d397 | 2376 | item = btrfs_item_nr(left, i); |
db94535d | 2377 | |
31840ae1 ZY |
2378 | if (!empty && push_items > 0) { |
2379 | if (path->slots[0] > i) | |
2380 | break; | |
2381 | if (path->slots[0] == i) { | |
2382 | int space = btrfs_leaf_free_space(root, left); | |
2383 | if (space + push_space * 2 > free_space) | |
2384 | break; | |
2385 | } | |
2386 | } | |
2387 | ||
00ec4c51 | 2388 | if (path->slots[0] == i) |
87b29b20 | 2389 | push_space += data_size; |
db94535d CM |
2390 | |
2391 | if (!left->map_token) { | |
2392 | map_extent_buffer(left, (unsigned long)item, | |
2393 | sizeof(struct btrfs_item), | |
2394 | &left->map_token, &left->kaddr, | |
2395 | &left->map_start, &left->map_len, | |
2396 | KM_USER1); | |
2397 | } | |
2398 | ||
2399 | this_item_size = btrfs_item_size(left, item); | |
2400 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2401 | break; |
31840ae1 | 2402 | |
00ec4c51 | 2403 | push_items++; |
db94535d | 2404 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2405 | if (i == 0) |
2406 | break; | |
2407 | i--; | |
db94535d CM |
2408 | } |
2409 | if (left->map_token) { | |
2410 | unmap_extent_buffer(left, left->map_token, KM_USER1); | |
2411 | left->map_token = NULL; | |
00ec4c51 | 2412 | } |
5f39d397 | 2413 | |
925baedd CM |
2414 | if (push_items == 0) |
2415 | goto out_unlock; | |
5f39d397 | 2416 | |
34a38218 | 2417 | if (!empty && push_items == left_nritems) |
a429e513 | 2418 | WARN_ON(1); |
5f39d397 | 2419 | |
00ec4c51 | 2420 | /* push left to right */ |
5f39d397 | 2421 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2422 | |
5f39d397 | 2423 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2424 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2425 | |
00ec4c51 | 2426 | /* make room in the right data area */ |
5f39d397 CM |
2427 | data_end = leaf_data_end(root, right); |
2428 | memmove_extent_buffer(right, | |
2429 | btrfs_leaf_data(right) + data_end - push_space, | |
2430 | btrfs_leaf_data(right) + data_end, | |
2431 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2432 | ||
00ec4c51 | 2433 | /* copy from the left data area */ |
5f39d397 | 2434 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2435 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2436 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2437 | push_space); | |
5f39d397 CM |
2438 | |
2439 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2440 | btrfs_item_nr_offset(0), | |
2441 | right_nritems * sizeof(struct btrfs_item)); | |
2442 | ||
00ec4c51 | 2443 | /* copy the items from left to right */ |
5f39d397 CM |
2444 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2445 | btrfs_item_nr_offset(left_nritems - push_items), | |
2446 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2447 | |
2448 | /* update the item pointers */ | |
7518a238 | 2449 | right_nritems += push_items; |
5f39d397 | 2450 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2451 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2452 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2453 | item = btrfs_item_nr(right, i); |
db94535d CM |
2454 | if (!right->map_token) { |
2455 | map_extent_buffer(right, (unsigned long)item, | |
2456 | sizeof(struct btrfs_item), | |
2457 | &right->map_token, &right->kaddr, | |
2458 | &right->map_start, &right->map_len, | |
2459 | KM_USER1); | |
2460 | } | |
2461 | push_space -= btrfs_item_size(right, item); | |
2462 | btrfs_set_item_offset(right, item, push_space); | |
2463 | } | |
2464 | ||
2465 | if (right->map_token) { | |
2466 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2467 | right->map_token = NULL; | |
00ec4c51 | 2468 | } |
7518a238 | 2469 | left_nritems -= push_items; |
5f39d397 | 2470 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2471 | |
34a38218 CM |
2472 | if (left_nritems) |
2473 | btrfs_mark_buffer_dirty(left); | |
5f39d397 | 2474 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2475 | |
31840ae1 ZY |
2476 | ret = btrfs_update_ref(trans, root, left, right, 0, push_items); |
2477 | BUG_ON(ret); | |
2478 | ||
5f39d397 CM |
2479 | btrfs_item_key(right, &disk_key, 0); |
2480 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2481 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2482 | |
00ec4c51 | 2483 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2484 | if (path->slots[0] >= left_nritems) { |
2485 | path->slots[0] -= left_nritems; | |
925baedd CM |
2486 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2487 | clean_tree_block(trans, root, path->nodes[0]); | |
2488 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2489 | free_extent_buffer(path->nodes[0]); |
2490 | path->nodes[0] = right; | |
00ec4c51 CM |
2491 | path->slots[1] += 1; |
2492 | } else { | |
925baedd | 2493 | btrfs_tree_unlock(right); |
5f39d397 | 2494 | free_extent_buffer(right); |
00ec4c51 CM |
2495 | } |
2496 | return 0; | |
925baedd CM |
2497 | |
2498 | out_unlock: | |
2499 | btrfs_tree_unlock(right); | |
2500 | free_extent_buffer(right); | |
2501 | return 1; | |
00ec4c51 | 2502 | } |
925baedd | 2503 | |
44871b1b CM |
2504 | /* |
2505 | * push some data in the path leaf to the right, trying to free up at | |
2506 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2507 | * | |
2508 | * returns 1 if the push failed because the other node didn't have enough | |
2509 | * room, 0 if everything worked out and < 0 if there were major errors. | |
2510 | */ | |
2511 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
2512 | *root, struct btrfs_path *path, int data_size, | |
2513 | int empty) | |
2514 | { | |
2515 | struct extent_buffer *left = path->nodes[0]; | |
2516 | struct extent_buffer *right; | |
2517 | struct extent_buffer *upper; | |
2518 | int slot; | |
2519 | int free_space; | |
2520 | u32 left_nritems; | |
2521 | int ret; | |
2522 | ||
2523 | if (!path->nodes[1]) | |
2524 | return 1; | |
2525 | ||
2526 | slot = path->slots[1]; | |
2527 | upper = path->nodes[1]; | |
2528 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2529 | return 1; | |
2530 | ||
2531 | btrfs_assert_tree_locked(path->nodes[1]); | |
2532 | ||
2533 | right = read_node_slot(root, upper, slot + 1); | |
2534 | btrfs_tree_lock(right); | |
2535 | btrfs_set_lock_blocking(right); | |
2536 | ||
2537 | free_space = btrfs_leaf_free_space(root, right); | |
2538 | if (free_space < data_size) | |
2539 | goto out_unlock; | |
2540 | ||
2541 | /* cow and double check */ | |
2542 | ret = btrfs_cow_block(trans, root, right, upper, | |
2543 | slot + 1, &right); | |
2544 | if (ret) | |
2545 | goto out_unlock; | |
2546 | ||
2547 | free_space = btrfs_leaf_free_space(root, right); | |
2548 | if (free_space < data_size) | |
2549 | goto out_unlock; | |
2550 | ||
2551 | left_nritems = btrfs_header_nritems(left); | |
2552 | if (left_nritems == 0) | |
2553 | goto out_unlock; | |
2554 | ||
2555 | return __push_leaf_right(trans, root, path, data_size, empty, | |
2556 | right, free_space, left_nritems); | |
2557 | out_unlock: | |
2558 | btrfs_tree_unlock(right); | |
2559 | free_extent_buffer(right); | |
2560 | return 1; | |
2561 | } | |
2562 | ||
74123bd7 CM |
2563 | /* |
2564 | * push some data in the path leaf to the left, trying to free up at | |
2565 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2566 | */ | |
44871b1b CM |
2567 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
2568 | struct btrfs_root *root, | |
2569 | struct btrfs_path *path, int data_size, | |
2570 | int empty, struct extent_buffer *left, | |
2571 | int free_space, int right_nritems) | |
be0e5c09 | 2572 | { |
5f39d397 CM |
2573 | struct btrfs_disk_key disk_key; |
2574 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 CM |
2575 | int slot; |
2576 | int i; | |
be0e5c09 CM |
2577 | int push_space = 0; |
2578 | int push_items = 0; | |
0783fcfc | 2579 | struct btrfs_item *item; |
7518a238 | 2580 | u32 old_left_nritems; |
34a38218 | 2581 | u32 nr; |
aa5d6bed CM |
2582 | int ret = 0; |
2583 | int wret; | |
db94535d CM |
2584 | u32 this_item_size; |
2585 | u32 old_left_item_size; | |
be0e5c09 CM |
2586 | |
2587 | slot = path->slots[1]; | |
02217ed2 | 2588 | |
34a38218 CM |
2589 | if (empty) |
2590 | nr = right_nritems; | |
2591 | else | |
2592 | nr = right_nritems - 1; | |
2593 | ||
2594 | for (i = 0; i < nr; i++) { | |
5f39d397 | 2595 | item = btrfs_item_nr(right, i); |
db94535d CM |
2596 | if (!right->map_token) { |
2597 | map_extent_buffer(right, (unsigned long)item, | |
2598 | sizeof(struct btrfs_item), | |
2599 | &right->map_token, &right->kaddr, | |
2600 | &right->map_start, &right->map_len, | |
2601 | KM_USER1); | |
2602 | } | |
2603 | ||
31840ae1 ZY |
2604 | if (!empty && push_items > 0) { |
2605 | if (path->slots[0] < i) | |
2606 | break; | |
2607 | if (path->slots[0] == i) { | |
2608 | int space = btrfs_leaf_free_space(root, right); | |
2609 | if (space + push_space * 2 > free_space) | |
2610 | break; | |
2611 | } | |
2612 | } | |
2613 | ||
be0e5c09 | 2614 | if (path->slots[0] == i) |
87b29b20 | 2615 | push_space += data_size; |
db94535d CM |
2616 | |
2617 | this_item_size = btrfs_item_size(right, item); | |
2618 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 2619 | break; |
db94535d | 2620 | |
be0e5c09 | 2621 | push_items++; |
db94535d CM |
2622 | push_space += this_item_size + sizeof(*item); |
2623 | } | |
2624 | ||
2625 | if (right->map_token) { | |
2626 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2627 | right->map_token = NULL; | |
be0e5c09 | 2628 | } |
db94535d | 2629 | |
be0e5c09 | 2630 | if (push_items == 0) { |
925baedd CM |
2631 | ret = 1; |
2632 | goto out; | |
be0e5c09 | 2633 | } |
34a38218 | 2634 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 2635 | WARN_ON(1); |
5f39d397 | 2636 | |
be0e5c09 | 2637 | /* push data from right to left */ |
5f39d397 CM |
2638 | copy_extent_buffer(left, right, |
2639 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
2640 | btrfs_item_nr_offset(0), | |
2641 | push_items * sizeof(struct btrfs_item)); | |
2642 | ||
123abc88 | 2643 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 2644 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
2645 | |
2646 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
2647 | leaf_data_end(root, left) - push_space, |
2648 | btrfs_leaf_data(right) + | |
5f39d397 | 2649 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 2650 | push_space); |
5f39d397 | 2651 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 2652 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 2653 | |
db94535d | 2654 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 2655 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 2656 | u32 ioff; |
db94535d | 2657 | |
5f39d397 | 2658 | item = btrfs_item_nr(left, i); |
db94535d CM |
2659 | if (!left->map_token) { |
2660 | map_extent_buffer(left, (unsigned long)item, | |
2661 | sizeof(struct btrfs_item), | |
2662 | &left->map_token, &left->kaddr, | |
2663 | &left->map_start, &left->map_len, | |
2664 | KM_USER1); | |
2665 | } | |
2666 | ||
5f39d397 CM |
2667 | ioff = btrfs_item_offset(left, item); |
2668 | btrfs_set_item_offset(left, item, | |
db94535d | 2669 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size)); |
be0e5c09 | 2670 | } |
5f39d397 | 2671 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
db94535d CM |
2672 | if (left->map_token) { |
2673 | unmap_extent_buffer(left, left->map_token, KM_USER1); | |
2674 | left->map_token = NULL; | |
2675 | } | |
be0e5c09 CM |
2676 | |
2677 | /* fixup right node */ | |
34a38218 | 2678 | if (push_items > right_nritems) { |
d397712b CM |
2679 | printk(KERN_CRIT "push items %d nr %u\n", push_items, |
2680 | right_nritems); | |
34a38218 CM |
2681 | WARN_ON(1); |
2682 | } | |
2683 | ||
2684 | if (push_items < right_nritems) { | |
2685 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2686 | leaf_data_end(root, right); | |
2687 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
2688 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
2689 | btrfs_leaf_data(right) + | |
2690 | leaf_data_end(root, right), push_space); | |
2691 | ||
2692 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
2693 | btrfs_item_nr_offset(push_items), |
2694 | (btrfs_header_nritems(right) - push_items) * | |
2695 | sizeof(struct btrfs_item)); | |
34a38218 | 2696 | } |
eef1c494 Y |
2697 | right_nritems -= push_items; |
2698 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 2699 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
2700 | for (i = 0; i < right_nritems; i++) { |
2701 | item = btrfs_item_nr(right, i); | |
db94535d CM |
2702 | |
2703 | if (!right->map_token) { | |
2704 | map_extent_buffer(right, (unsigned long)item, | |
2705 | sizeof(struct btrfs_item), | |
2706 | &right->map_token, &right->kaddr, | |
2707 | &right->map_start, &right->map_len, | |
2708 | KM_USER1); | |
2709 | } | |
2710 | ||
2711 | push_space = push_space - btrfs_item_size(right, item); | |
2712 | btrfs_set_item_offset(right, item, push_space); | |
2713 | } | |
2714 | if (right->map_token) { | |
2715 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2716 | right->map_token = NULL; | |
be0e5c09 | 2717 | } |
eb60ceac | 2718 | |
5f39d397 | 2719 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
2720 | if (right_nritems) |
2721 | btrfs_mark_buffer_dirty(right); | |
098f59c2 | 2722 | |
31840ae1 ZY |
2723 | ret = btrfs_update_ref(trans, root, right, left, |
2724 | old_left_nritems, push_items); | |
2725 | BUG_ON(ret); | |
2726 | ||
5f39d397 CM |
2727 | btrfs_item_key(right, &disk_key, 0); |
2728 | wret = fixup_low_keys(trans, root, path, &disk_key, 1); | |
aa5d6bed CM |
2729 | if (wret) |
2730 | ret = wret; | |
be0e5c09 CM |
2731 | |
2732 | /* then fixup the leaf pointer in the path */ | |
2733 | if (path->slots[0] < push_items) { | |
2734 | path->slots[0] += old_left_nritems; | |
925baedd CM |
2735 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2736 | clean_tree_block(trans, root, path->nodes[0]); | |
2737 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2738 | free_extent_buffer(path->nodes[0]); |
2739 | path->nodes[0] = left; | |
be0e5c09 CM |
2740 | path->slots[1] -= 1; |
2741 | } else { | |
925baedd | 2742 | btrfs_tree_unlock(left); |
5f39d397 | 2743 | free_extent_buffer(left); |
be0e5c09 CM |
2744 | path->slots[0] -= push_items; |
2745 | } | |
eb60ceac | 2746 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 2747 | return ret; |
925baedd CM |
2748 | out: |
2749 | btrfs_tree_unlock(left); | |
2750 | free_extent_buffer(left); | |
2751 | return ret; | |
be0e5c09 CM |
2752 | } |
2753 | ||
44871b1b CM |
2754 | /* |
2755 | * push some data in the path leaf to the left, trying to free up at | |
2756 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2757 | */ | |
2758 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
2759 | *root, struct btrfs_path *path, int data_size, | |
2760 | int empty) | |
2761 | { | |
2762 | struct extent_buffer *right = path->nodes[0]; | |
2763 | struct extent_buffer *left; | |
2764 | int slot; | |
2765 | int free_space; | |
2766 | u32 right_nritems; | |
2767 | int ret = 0; | |
2768 | ||
2769 | slot = path->slots[1]; | |
2770 | if (slot == 0) | |
2771 | return 1; | |
2772 | if (!path->nodes[1]) | |
2773 | return 1; | |
2774 | ||
2775 | right_nritems = btrfs_header_nritems(right); | |
2776 | if (right_nritems == 0) | |
2777 | return 1; | |
2778 | ||
2779 | btrfs_assert_tree_locked(path->nodes[1]); | |
2780 | ||
2781 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
2782 | btrfs_tree_lock(left); | |
2783 | btrfs_set_lock_blocking(left); | |
2784 | ||
2785 | free_space = btrfs_leaf_free_space(root, left); | |
2786 | if (free_space < data_size) { | |
2787 | ret = 1; | |
2788 | goto out; | |
2789 | } | |
2790 | ||
2791 | /* cow and double check */ | |
2792 | ret = btrfs_cow_block(trans, root, left, | |
2793 | path->nodes[1], slot - 1, &left); | |
2794 | if (ret) { | |
2795 | /* we hit -ENOSPC, but it isn't fatal here */ | |
2796 | ret = 1; | |
2797 | goto out; | |
2798 | } | |
2799 | ||
2800 | free_space = btrfs_leaf_free_space(root, left); | |
2801 | if (free_space < data_size) { | |
2802 | ret = 1; | |
2803 | goto out; | |
2804 | } | |
2805 | ||
2806 | return __push_leaf_left(trans, root, path, data_size, | |
2807 | empty, left, free_space, right_nritems); | |
2808 | out: | |
2809 | btrfs_tree_unlock(left); | |
2810 | free_extent_buffer(left); | |
2811 | return ret; | |
2812 | } | |
2813 | ||
2814 | /* | |
2815 | * split the path's leaf in two, making sure there is at least data_size | |
2816 | * available for the resulting leaf level of the path. | |
2817 | * | |
2818 | * returns 0 if all went well and < 0 on failure. | |
2819 | */ | |
2820 | static noinline int copy_for_split(struct btrfs_trans_handle *trans, | |
2821 | struct btrfs_root *root, | |
2822 | struct btrfs_path *path, | |
2823 | struct extent_buffer *l, | |
2824 | struct extent_buffer *right, | |
2825 | int slot, int mid, int nritems) | |
2826 | { | |
2827 | int data_copy_size; | |
2828 | int rt_data_off; | |
2829 | int i; | |
2830 | int ret = 0; | |
2831 | int wret; | |
2832 | struct btrfs_disk_key disk_key; | |
2833 | ||
2834 | nritems = nritems - mid; | |
2835 | btrfs_set_header_nritems(right, nritems); | |
2836 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
2837 | ||
2838 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
2839 | btrfs_item_nr_offset(mid), | |
2840 | nritems * sizeof(struct btrfs_item)); | |
2841 | ||
2842 | copy_extent_buffer(right, l, | |
2843 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
2844 | data_copy_size, btrfs_leaf_data(l) + | |
2845 | leaf_data_end(root, l), data_copy_size); | |
2846 | ||
2847 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
2848 | btrfs_item_end_nr(l, mid); | |
2849 | ||
2850 | for (i = 0; i < nritems; i++) { | |
2851 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
2852 | u32 ioff; | |
2853 | ||
2854 | if (!right->map_token) { | |
2855 | map_extent_buffer(right, (unsigned long)item, | |
2856 | sizeof(struct btrfs_item), | |
2857 | &right->map_token, &right->kaddr, | |
2858 | &right->map_start, &right->map_len, | |
2859 | KM_USER1); | |
2860 | } | |
2861 | ||
2862 | ioff = btrfs_item_offset(right, item); | |
2863 | btrfs_set_item_offset(right, item, ioff + rt_data_off); | |
2864 | } | |
2865 | ||
2866 | if (right->map_token) { | |
2867 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2868 | right->map_token = NULL; | |
2869 | } | |
2870 | ||
2871 | btrfs_set_header_nritems(l, mid); | |
2872 | ret = 0; | |
2873 | btrfs_item_key(right, &disk_key, 0); | |
2874 | wret = insert_ptr(trans, root, path, &disk_key, right->start, | |
2875 | path->slots[1] + 1, 1); | |
2876 | if (wret) | |
2877 | ret = wret; | |
2878 | ||
2879 | btrfs_mark_buffer_dirty(right); | |
2880 | btrfs_mark_buffer_dirty(l); | |
2881 | BUG_ON(path->slots[0] != slot); | |
2882 | ||
2883 | ret = btrfs_update_ref(trans, root, l, right, 0, nritems); | |
2884 | BUG_ON(ret); | |
2885 | ||
2886 | if (mid <= slot) { | |
2887 | btrfs_tree_unlock(path->nodes[0]); | |
2888 | free_extent_buffer(path->nodes[0]); | |
2889 | path->nodes[0] = right; | |
2890 | path->slots[0] -= mid; | |
2891 | path->slots[1] += 1; | |
2892 | } else { | |
2893 | btrfs_tree_unlock(right); | |
2894 | free_extent_buffer(right); | |
2895 | } | |
2896 | ||
2897 | BUG_ON(path->slots[0] < 0); | |
2898 | ||
2899 | return ret; | |
2900 | } | |
2901 | ||
74123bd7 CM |
2902 | /* |
2903 | * split the path's leaf in two, making sure there is at least data_size | |
2904 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
2905 | * |
2906 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 2907 | */ |
e02119d5 CM |
2908 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
2909 | struct btrfs_root *root, | |
2910 | struct btrfs_key *ins_key, | |
2911 | struct btrfs_path *path, int data_size, | |
2912 | int extend) | |
be0e5c09 | 2913 | { |
5f39d397 | 2914 | struct extent_buffer *l; |
7518a238 | 2915 | u32 nritems; |
eb60ceac CM |
2916 | int mid; |
2917 | int slot; | |
5f39d397 | 2918 | struct extent_buffer *right; |
d4dbff95 | 2919 | int ret = 0; |
aa5d6bed | 2920 | int wret; |
cc0c5538 CM |
2921 | int double_split; |
2922 | int num_doubles = 0; | |
aa5d6bed | 2923 | |
40689478 | 2924 | /* first try to make some room by pushing left and right */ |
a4b6e07d CM |
2925 | if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY && |
2926 | !trans->transaction->delayed_refs.flushing) { | |
34a38218 | 2927 | wret = push_leaf_right(trans, root, path, data_size, 0); |
d397712b | 2928 | if (wret < 0) |
eaee50e8 | 2929 | return wret; |
3685f791 | 2930 | if (wret) { |
34a38218 | 2931 | wret = push_leaf_left(trans, root, path, data_size, 0); |
3685f791 CM |
2932 | if (wret < 0) |
2933 | return wret; | |
2934 | } | |
2935 | l = path->nodes[0]; | |
aa5d6bed | 2936 | |
3685f791 | 2937 | /* did the pushes work? */ |
87b29b20 | 2938 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 2939 | return 0; |
3326d1b0 | 2940 | } |
aa5d6bed | 2941 | |
5c680ed6 | 2942 | if (!path->nodes[1]) { |
e089f05c | 2943 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
2944 | if (ret) |
2945 | return ret; | |
2946 | } | |
cc0c5538 CM |
2947 | again: |
2948 | double_split = 0; | |
2949 | l = path->nodes[0]; | |
eb60ceac | 2950 | slot = path->slots[0]; |
5f39d397 | 2951 | nritems = btrfs_header_nritems(l); |
d397712b | 2952 | mid = (nritems + 1) / 2; |
54aa1f4d | 2953 | |
925baedd | 2954 | right = btrfs_alloc_free_block(trans, root, root->leafsize, |
31840ae1 ZY |
2955 | path->nodes[1]->start, |
2956 | root->root_key.objectid, | |
2957 | trans->transid, 0, l->start, 0); | |
cea9e445 CM |
2958 | if (IS_ERR(right)) { |
2959 | BUG_ON(1); | |
5f39d397 | 2960 | return PTR_ERR(right); |
cea9e445 | 2961 | } |
5f39d397 CM |
2962 | |
2963 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 2964 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 CM |
2965 | btrfs_set_header_generation(right, trans->transid); |
2966 | btrfs_set_header_owner(right, root->root_key.objectid); | |
2967 | btrfs_set_header_level(right, 0); | |
2968 | write_extent_buffer(right, root->fs_info->fsid, | |
2969 | (unsigned long)btrfs_header_fsid(right), | |
2970 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2971 | |
2972 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
2973 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
2974 | BTRFS_UUID_SIZE); | |
44871b1b | 2975 | |
d4dbff95 CM |
2976 | if (mid <= slot) { |
2977 | if (nritems == 1 || | |
87b29b20 | 2978 | leaf_space_used(l, mid, nritems - mid) + data_size > |
d4dbff95 CM |
2979 | BTRFS_LEAF_DATA_SIZE(root)) { |
2980 | if (slot >= nritems) { | |
44871b1b CM |
2981 | struct btrfs_disk_key disk_key; |
2982 | ||
d4dbff95 | 2983 | btrfs_cpu_key_to_disk(&disk_key, ins_key); |
5f39d397 | 2984 | btrfs_set_header_nritems(right, 0); |
d4dbff95 | 2985 | wret = insert_ptr(trans, root, path, |
db94535d | 2986 | &disk_key, right->start, |
d4dbff95 CM |
2987 | path->slots[1] + 1, 1); |
2988 | if (wret) | |
2989 | ret = wret; | |
925baedd CM |
2990 | |
2991 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2992 | free_extent_buffer(path->nodes[0]); |
2993 | path->nodes[0] = right; | |
d4dbff95 CM |
2994 | path->slots[0] = 0; |
2995 | path->slots[1] += 1; | |
0ef8b242 | 2996 | btrfs_mark_buffer_dirty(right); |
d4dbff95 CM |
2997 | return ret; |
2998 | } | |
2999 | mid = slot; | |
3326d1b0 CM |
3000 | if (mid != nritems && |
3001 | leaf_space_used(l, mid, nritems - mid) + | |
87b29b20 | 3002 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { |
3326d1b0 CM |
3003 | double_split = 1; |
3004 | } | |
d4dbff95 CM |
3005 | } |
3006 | } else { | |
87b29b20 | 3007 | if (leaf_space_used(l, 0, mid) + data_size > |
d4dbff95 | 3008 | BTRFS_LEAF_DATA_SIZE(root)) { |
459931ec | 3009 | if (!extend && data_size && slot == 0) { |
44871b1b CM |
3010 | struct btrfs_disk_key disk_key; |
3011 | ||
d4dbff95 | 3012 | btrfs_cpu_key_to_disk(&disk_key, ins_key); |
5f39d397 | 3013 | btrfs_set_header_nritems(right, 0); |
d4dbff95 CM |
3014 | wret = insert_ptr(trans, root, path, |
3015 | &disk_key, | |
db94535d | 3016 | right->start, |
098f59c2 | 3017 | path->slots[1], 1); |
d4dbff95 CM |
3018 | if (wret) |
3019 | ret = wret; | |
925baedd | 3020 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3021 | free_extent_buffer(path->nodes[0]); |
3022 | path->nodes[0] = right; | |
d4dbff95 | 3023 | path->slots[0] = 0; |
a429e513 CM |
3024 | if (path->slots[1] == 0) { |
3025 | wret = fixup_low_keys(trans, root, | |
d397712b | 3026 | path, &disk_key, 1); |
a429e513 CM |
3027 | if (wret) |
3028 | ret = wret; | |
3029 | } | |
0ef8b242 | 3030 | btrfs_mark_buffer_dirty(right); |
d4dbff95 | 3031 | return ret; |
459931ec | 3032 | } else if ((extend || !data_size) && slot == 0) { |
cc0c5538 CM |
3033 | mid = 1; |
3034 | } else { | |
3035 | mid = slot; | |
3036 | if (mid != nritems && | |
3037 | leaf_space_used(l, mid, nritems - mid) + | |
87b29b20 | 3038 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { |
cc0c5538 CM |
3039 | double_split = 1; |
3040 | } | |
5ee78ac7 | 3041 | } |
d4dbff95 CM |
3042 | } |
3043 | } | |
74123bd7 | 3044 | |
44871b1b | 3045 | ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 ZY |
3046 | BUG_ON(ret); |
3047 | ||
cc0c5538 CM |
3048 | if (double_split) { |
3049 | BUG_ON(num_doubles != 0); | |
3050 | num_doubles++; | |
3051 | goto again; | |
a429e513 | 3052 | } |
44871b1b | 3053 | |
be0e5c09 CM |
3054 | return ret; |
3055 | } | |
3056 | ||
459931ec CM |
3057 | /* |
3058 | * This function splits a single item into two items, | |
3059 | * giving 'new_key' to the new item and splitting the | |
3060 | * old one at split_offset (from the start of the item). | |
3061 | * | |
3062 | * The path may be released by this operation. After | |
3063 | * the split, the path is pointing to the old item. The | |
3064 | * new item is going to be in the same node as the old one. | |
3065 | * | |
3066 | * Note, the item being split must be smaller enough to live alone on | |
3067 | * a tree block with room for one extra struct btrfs_item | |
3068 | * | |
3069 | * This allows us to split the item in place, keeping a lock on the | |
3070 | * leaf the entire time. | |
3071 | */ | |
3072 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3073 | struct btrfs_root *root, | |
3074 | struct btrfs_path *path, | |
3075 | struct btrfs_key *new_key, | |
3076 | unsigned long split_offset) | |
3077 | { | |
3078 | u32 item_size; | |
3079 | struct extent_buffer *leaf; | |
3080 | struct btrfs_key orig_key; | |
3081 | struct btrfs_item *item; | |
3082 | struct btrfs_item *new_item; | |
3083 | int ret = 0; | |
3084 | int slot; | |
3085 | u32 nritems; | |
3086 | u32 orig_offset; | |
3087 | struct btrfs_disk_key disk_key; | |
3088 | char *buf; | |
3089 | ||
3090 | leaf = path->nodes[0]; | |
3091 | btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]); | |
3092 | if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item)) | |
3093 | goto split; | |
3094 | ||
3095 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3096 | btrfs_release_path(root, path); | |
3097 | ||
3098 | path->search_for_split = 1; | |
3099 | path->keep_locks = 1; | |
3100 | ||
3101 | ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1); | |
3102 | path->search_for_split = 0; | |
3103 | ||
3104 | /* if our item isn't there or got smaller, return now */ | |
3105 | if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0], | |
3106 | path->slots[0])) { | |
3107 | path->keep_locks = 0; | |
3108 | return -EAGAIN; | |
3109 | } | |
3110 | ||
b9473439 | 3111 | btrfs_set_path_blocking(path); |
87b29b20 YZ |
3112 | ret = split_leaf(trans, root, &orig_key, path, |
3113 | sizeof(struct btrfs_item), 1); | |
459931ec CM |
3114 | path->keep_locks = 0; |
3115 | BUG_ON(ret); | |
3116 | ||
b9473439 CM |
3117 | btrfs_unlock_up_safe(path, 1); |
3118 | leaf = path->nodes[0]; | |
3119 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
3120 | ||
3121 | split: | |
b4ce94de CM |
3122 | /* |
3123 | * make sure any changes to the path from split_leaf leave it | |
3124 | * in a blocking state | |
3125 | */ | |
3126 | btrfs_set_path_blocking(path); | |
3127 | ||
459931ec CM |
3128 | item = btrfs_item_nr(leaf, path->slots[0]); |
3129 | orig_offset = btrfs_item_offset(leaf, item); | |
3130 | item_size = btrfs_item_size(leaf, item); | |
3131 | ||
459931ec CM |
3132 | buf = kmalloc(item_size, GFP_NOFS); |
3133 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, | |
3134 | path->slots[0]), item_size); | |
3135 | slot = path->slots[0] + 1; | |
3136 | leaf = path->nodes[0]; | |
3137 | ||
3138 | nritems = btrfs_header_nritems(leaf); | |
3139 | ||
3140 | if (slot != nritems) { | |
3141 | /* shift the items */ | |
3142 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
3143 | btrfs_item_nr_offset(slot), | |
3144 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3145 | ||
3146 | } | |
3147 | ||
3148 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3149 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3150 | ||
3151 | new_item = btrfs_item_nr(leaf, slot); | |
3152 | ||
3153 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3154 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3155 | ||
3156 | btrfs_set_item_offset(leaf, item, | |
3157 | orig_offset + item_size - split_offset); | |
3158 | btrfs_set_item_size(leaf, item, split_offset); | |
3159 | ||
3160 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3161 | ||
3162 | /* write the data for the start of the original item */ | |
3163 | write_extent_buffer(leaf, buf, | |
3164 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3165 | split_offset); | |
3166 | ||
3167 | /* write the data for the new item */ | |
3168 | write_extent_buffer(leaf, buf + split_offset, | |
3169 | btrfs_item_ptr_offset(leaf, slot), | |
3170 | item_size - split_offset); | |
3171 | btrfs_mark_buffer_dirty(leaf); | |
3172 | ||
3173 | ret = 0; | |
3174 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
3175 | btrfs_print_leaf(root, leaf); | |
3176 | BUG(); | |
3177 | } | |
3178 | kfree(buf); | |
3179 | return ret; | |
3180 | } | |
3181 | ||
d352ac68 CM |
3182 | /* |
3183 | * make the item pointed to by the path smaller. new_size indicates | |
3184 | * how small to make it, and from_end tells us if we just chop bytes | |
3185 | * off the end of the item or if we shift the item to chop bytes off | |
3186 | * the front. | |
3187 | */ | |
b18c6685 CM |
3188 | int btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3189 | struct btrfs_root *root, | |
3190 | struct btrfs_path *path, | |
179e29e4 | 3191 | u32 new_size, int from_end) |
b18c6685 CM |
3192 | { |
3193 | int ret = 0; | |
3194 | int slot; | |
3195 | int slot_orig; | |
5f39d397 CM |
3196 | struct extent_buffer *leaf; |
3197 | struct btrfs_item *item; | |
b18c6685 CM |
3198 | u32 nritems; |
3199 | unsigned int data_end; | |
3200 | unsigned int old_data_start; | |
3201 | unsigned int old_size; | |
3202 | unsigned int size_diff; | |
3203 | int i; | |
3204 | ||
3205 | slot_orig = path->slots[0]; | |
5f39d397 | 3206 | leaf = path->nodes[0]; |
179e29e4 CM |
3207 | slot = path->slots[0]; |
3208 | ||
3209 | old_size = btrfs_item_size_nr(leaf, slot); | |
3210 | if (old_size == new_size) | |
3211 | return 0; | |
b18c6685 | 3212 | |
5f39d397 | 3213 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
3214 | data_end = leaf_data_end(root, leaf); |
3215 | ||
5f39d397 | 3216 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3217 | |
b18c6685 CM |
3218 | size_diff = old_size - new_size; |
3219 | ||
3220 | BUG_ON(slot < 0); | |
3221 | BUG_ON(slot >= nritems); | |
3222 | ||
3223 | /* | |
3224 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3225 | */ | |
3226 | /* first correct the data pointers */ | |
3227 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3228 | u32 ioff; |
3229 | item = btrfs_item_nr(leaf, i); | |
db94535d CM |
3230 | |
3231 | if (!leaf->map_token) { | |
3232 | map_extent_buffer(leaf, (unsigned long)item, | |
3233 | sizeof(struct btrfs_item), | |
3234 | &leaf->map_token, &leaf->kaddr, | |
3235 | &leaf->map_start, &leaf->map_len, | |
3236 | KM_USER1); | |
3237 | } | |
3238 | ||
5f39d397 CM |
3239 | ioff = btrfs_item_offset(leaf, item); |
3240 | btrfs_set_item_offset(leaf, item, ioff + size_diff); | |
b18c6685 | 3241 | } |
db94535d CM |
3242 | |
3243 | if (leaf->map_token) { | |
3244 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3245 | leaf->map_token = NULL; | |
3246 | } | |
3247 | ||
b18c6685 | 3248 | /* shift the data */ |
179e29e4 CM |
3249 | if (from_end) { |
3250 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3251 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3252 | data_end, old_data_start + new_size - data_end); | |
3253 | } else { | |
3254 | struct btrfs_disk_key disk_key; | |
3255 | u64 offset; | |
3256 | ||
3257 | btrfs_item_key(leaf, &disk_key, slot); | |
3258 | ||
3259 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3260 | unsigned long ptr; | |
3261 | struct btrfs_file_extent_item *fi; | |
3262 | ||
3263 | fi = btrfs_item_ptr(leaf, slot, | |
3264 | struct btrfs_file_extent_item); | |
3265 | fi = (struct btrfs_file_extent_item *)( | |
3266 | (unsigned long)fi - size_diff); | |
3267 | ||
3268 | if (btrfs_file_extent_type(leaf, fi) == | |
3269 | BTRFS_FILE_EXTENT_INLINE) { | |
3270 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3271 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
3272 | (unsigned long)fi, |
3273 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
3274 | disk_bytenr)); |
3275 | } | |
3276 | } | |
3277 | ||
3278 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3279 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3280 | data_end, old_data_start - data_end); | |
3281 | ||
3282 | offset = btrfs_disk_key_offset(&disk_key); | |
3283 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3284 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3285 | if (slot == 0) | |
3286 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
3287 | } | |
5f39d397 CM |
3288 | |
3289 | item = btrfs_item_nr(leaf, slot); | |
3290 | btrfs_set_item_size(leaf, item, new_size); | |
3291 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 CM |
3292 | |
3293 | ret = 0; | |
5f39d397 CM |
3294 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3295 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 3296 | BUG(); |
5f39d397 | 3297 | } |
b18c6685 CM |
3298 | return ret; |
3299 | } | |
3300 | ||
d352ac68 CM |
3301 | /* |
3302 | * make the item pointed to by the path bigger, data_size is the new size. | |
3303 | */ | |
5f39d397 CM |
3304 | int btrfs_extend_item(struct btrfs_trans_handle *trans, |
3305 | struct btrfs_root *root, struct btrfs_path *path, | |
3306 | u32 data_size) | |
6567e837 CM |
3307 | { |
3308 | int ret = 0; | |
3309 | int slot; | |
3310 | int slot_orig; | |
5f39d397 CM |
3311 | struct extent_buffer *leaf; |
3312 | struct btrfs_item *item; | |
6567e837 CM |
3313 | u32 nritems; |
3314 | unsigned int data_end; | |
3315 | unsigned int old_data; | |
3316 | unsigned int old_size; | |
3317 | int i; | |
3318 | ||
3319 | slot_orig = path->slots[0]; | |
5f39d397 | 3320 | leaf = path->nodes[0]; |
6567e837 | 3321 | |
5f39d397 | 3322 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
3323 | data_end = leaf_data_end(root, leaf); |
3324 | ||
5f39d397 CM |
3325 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
3326 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3327 | BUG(); |
5f39d397 | 3328 | } |
6567e837 | 3329 | slot = path->slots[0]; |
5f39d397 | 3330 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3331 | |
3332 | BUG_ON(slot < 0); | |
3326d1b0 CM |
3333 | if (slot >= nritems) { |
3334 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
3335 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
3336 | slot, nritems); | |
3326d1b0 CM |
3337 | BUG_ON(1); |
3338 | } | |
6567e837 CM |
3339 | |
3340 | /* | |
3341 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3342 | */ | |
3343 | /* first correct the data pointers */ | |
3344 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3345 | u32 ioff; |
3346 | item = btrfs_item_nr(leaf, i); | |
db94535d CM |
3347 | |
3348 | if (!leaf->map_token) { | |
3349 | map_extent_buffer(leaf, (unsigned long)item, | |
3350 | sizeof(struct btrfs_item), | |
3351 | &leaf->map_token, &leaf->kaddr, | |
3352 | &leaf->map_start, &leaf->map_len, | |
3353 | KM_USER1); | |
3354 | } | |
5f39d397 CM |
3355 | ioff = btrfs_item_offset(leaf, item); |
3356 | btrfs_set_item_offset(leaf, item, ioff - data_size); | |
6567e837 | 3357 | } |
5f39d397 | 3358 | |
db94535d CM |
3359 | if (leaf->map_token) { |
3360 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3361 | leaf->map_token = NULL; | |
3362 | } | |
3363 | ||
6567e837 | 3364 | /* shift the data */ |
5f39d397 | 3365 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
3366 | data_end - data_size, btrfs_leaf_data(leaf) + |
3367 | data_end, old_data - data_end); | |
5f39d397 | 3368 | |
6567e837 | 3369 | data_end = old_data; |
5f39d397 CM |
3370 | old_size = btrfs_item_size_nr(leaf, slot); |
3371 | item = btrfs_item_nr(leaf, slot); | |
3372 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
3373 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 CM |
3374 | |
3375 | ret = 0; | |
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 | return ret; |
3381 | } | |
3382 | ||
f3465ca4 JB |
3383 | /* |
3384 | * Given a key and some data, insert items into the tree. | |
3385 | * This does all the path init required, making room in the tree if needed. | |
3386 | * Returns the number of keys that were inserted. | |
3387 | */ | |
3388 | int btrfs_insert_some_items(struct btrfs_trans_handle *trans, | |
3389 | struct btrfs_root *root, | |
3390 | struct btrfs_path *path, | |
3391 | struct btrfs_key *cpu_key, u32 *data_size, | |
3392 | int nr) | |
3393 | { | |
3394 | struct extent_buffer *leaf; | |
3395 | struct btrfs_item *item; | |
3396 | int ret = 0; | |
3397 | int slot; | |
f3465ca4 JB |
3398 | int i; |
3399 | u32 nritems; | |
3400 | u32 total_data = 0; | |
3401 | u32 total_size = 0; | |
3402 | unsigned int data_end; | |
3403 | struct btrfs_disk_key disk_key; | |
3404 | struct btrfs_key found_key; | |
3405 | ||
87b29b20 YZ |
3406 | for (i = 0; i < nr; i++) { |
3407 | if (total_size + data_size[i] + sizeof(struct btrfs_item) > | |
3408 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3409 | break; | |
3410 | nr = i; | |
3411 | } | |
f3465ca4 | 3412 | total_data += data_size[i]; |
87b29b20 YZ |
3413 | total_size += data_size[i] + sizeof(struct btrfs_item); |
3414 | } | |
3415 | BUG_ON(nr == 0); | |
f3465ca4 | 3416 | |
f3465ca4 JB |
3417 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
3418 | if (ret == 0) | |
3419 | return -EEXIST; | |
3420 | if (ret < 0) | |
3421 | goto out; | |
3422 | ||
f3465ca4 JB |
3423 | leaf = path->nodes[0]; |
3424 | ||
3425 | nritems = btrfs_header_nritems(leaf); | |
3426 | data_end = leaf_data_end(root, leaf); | |
3427 | ||
3428 | if (btrfs_leaf_free_space(root, leaf) < total_size) { | |
3429 | for (i = nr; i >= 0; i--) { | |
3430 | total_data -= data_size[i]; | |
3431 | total_size -= data_size[i] + sizeof(struct btrfs_item); | |
3432 | if (total_size < btrfs_leaf_free_space(root, leaf)) | |
3433 | break; | |
3434 | } | |
3435 | nr = i; | |
3436 | } | |
3437 | ||
3438 | slot = path->slots[0]; | |
3439 | BUG_ON(slot < 0); | |
3440 | ||
3441 | if (slot != nritems) { | |
3442 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); | |
3443 | ||
3444 | item = btrfs_item_nr(leaf, slot); | |
3445 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3446 | ||
3447 | /* figure out how many keys we can insert in here */ | |
3448 | total_data = data_size[0]; | |
3449 | for (i = 1; i < nr; i++) { | |
3450 | if (comp_cpu_keys(&found_key, cpu_key + i) <= 0) | |
3451 | break; | |
3452 | total_data += data_size[i]; | |
3453 | } | |
3454 | nr = i; | |
3455 | ||
3456 | if (old_data < data_end) { | |
3457 | btrfs_print_leaf(root, leaf); | |
d397712b | 3458 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
f3465ca4 JB |
3459 | slot, old_data, data_end); |
3460 | BUG_ON(1); | |
3461 | } | |
3462 | /* | |
3463 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3464 | */ | |
3465 | /* first correct the data pointers */ | |
3466 | WARN_ON(leaf->map_token); | |
3467 | for (i = slot; i < nritems; i++) { | |
3468 | u32 ioff; | |
3469 | ||
3470 | item = btrfs_item_nr(leaf, i); | |
3471 | if (!leaf->map_token) { | |
3472 | map_extent_buffer(leaf, (unsigned long)item, | |
3473 | sizeof(struct btrfs_item), | |
3474 | &leaf->map_token, &leaf->kaddr, | |
3475 | &leaf->map_start, &leaf->map_len, | |
3476 | KM_USER1); | |
3477 | } | |
3478 | ||
3479 | ioff = btrfs_item_offset(leaf, item); | |
3480 | btrfs_set_item_offset(leaf, item, ioff - total_data); | |
3481 | } | |
3482 | if (leaf->map_token) { | |
3483 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3484 | leaf->map_token = NULL; | |
3485 | } | |
3486 | ||
3487 | /* shift the items */ | |
3488 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), | |
3489 | btrfs_item_nr_offset(slot), | |
3490 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3491 | ||
3492 | /* shift the data */ | |
3493 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3494 | data_end - total_data, btrfs_leaf_data(leaf) + | |
3495 | data_end, old_data - data_end); | |
3496 | data_end = old_data; | |
3497 | } else { | |
3498 | /* | |
3499 | * this sucks but it has to be done, if we are inserting at | |
3500 | * the end of the leaf only insert 1 of the items, since we | |
3501 | * have no way of knowing whats on the next leaf and we'd have | |
3502 | * to drop our current locks to figure it out | |
3503 | */ | |
3504 | nr = 1; | |
3505 | } | |
3506 | ||
3507 | /* setup the item for the new data */ | |
3508 | for (i = 0; i < nr; i++) { | |
3509 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3510 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3511 | item = btrfs_item_nr(leaf, slot + i); | |
3512 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3513 | data_end -= data_size[i]; | |
3514 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3515 | } | |
3516 | btrfs_set_header_nritems(leaf, nritems + nr); | |
3517 | btrfs_mark_buffer_dirty(leaf); | |
3518 | ||
3519 | ret = 0; | |
3520 | if (slot == 0) { | |
3521 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
3522 | ret = fixup_low_keys(trans, root, path, &disk_key, 1); | |
3523 | } | |
3524 | ||
3525 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
3526 | btrfs_print_leaf(root, leaf); | |
3527 | BUG(); | |
3528 | } | |
3529 | out: | |
3530 | if (!ret) | |
3531 | ret = nr; | |
3532 | return ret; | |
3533 | } | |
3534 | ||
74123bd7 | 3535 | /* |
44871b1b CM |
3536 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
3537 | * to save stack depth by doing the bulk of the work in a function | |
3538 | * that doesn't call btrfs_search_slot | |
74123bd7 | 3539 | */ |
44871b1b CM |
3540 | static noinline_for_stack int |
3541 | setup_items_for_insert(struct btrfs_trans_handle *trans, | |
3542 | struct btrfs_root *root, struct btrfs_path *path, | |
3543 | struct btrfs_key *cpu_key, u32 *data_size, | |
3544 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 3545 | { |
5f39d397 | 3546 | struct btrfs_item *item; |
9c58309d | 3547 | int i; |
7518a238 | 3548 | u32 nritems; |
be0e5c09 | 3549 | unsigned int data_end; |
e2fa7227 | 3550 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3551 | int ret; |
3552 | struct extent_buffer *leaf; | |
3553 | int slot; | |
e2fa7227 | 3554 | |
5f39d397 | 3555 | leaf = path->nodes[0]; |
44871b1b | 3556 | slot = path->slots[0]; |
74123bd7 | 3557 | |
5f39d397 | 3558 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 3559 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 3560 | |
f25956cc | 3561 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 3562 | btrfs_print_leaf(root, leaf); |
d397712b | 3563 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 3564 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 3565 | BUG(); |
d4dbff95 | 3566 | } |
5f39d397 | 3567 | |
be0e5c09 | 3568 | if (slot != nritems) { |
5f39d397 | 3569 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 3570 | |
5f39d397 CM |
3571 | if (old_data < data_end) { |
3572 | btrfs_print_leaf(root, leaf); | |
d397712b | 3573 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
3574 | slot, old_data, data_end); |
3575 | BUG_ON(1); | |
3576 | } | |
be0e5c09 CM |
3577 | /* |
3578 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3579 | */ | |
3580 | /* first correct the data pointers */ | |
db94535d | 3581 | WARN_ON(leaf->map_token); |
0783fcfc | 3582 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3583 | u32 ioff; |
db94535d | 3584 | |
5f39d397 | 3585 | item = btrfs_item_nr(leaf, i); |
db94535d CM |
3586 | if (!leaf->map_token) { |
3587 | map_extent_buffer(leaf, (unsigned long)item, | |
3588 | sizeof(struct btrfs_item), | |
3589 | &leaf->map_token, &leaf->kaddr, | |
3590 | &leaf->map_start, &leaf->map_len, | |
3591 | KM_USER1); | |
3592 | } | |
3593 | ||
5f39d397 | 3594 | ioff = btrfs_item_offset(leaf, item); |
9c58309d | 3595 | btrfs_set_item_offset(leaf, item, ioff - total_data); |
0783fcfc | 3596 | } |
db94535d CM |
3597 | if (leaf->map_token) { |
3598 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3599 | leaf->map_token = NULL; | |
3600 | } | |
be0e5c09 CM |
3601 | |
3602 | /* shift the items */ | |
9c58309d | 3603 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 3604 | btrfs_item_nr_offset(slot), |
d6025579 | 3605 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
3606 | |
3607 | /* shift the data */ | |
5f39d397 | 3608 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 3609 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 3610 | data_end, old_data - data_end); |
be0e5c09 CM |
3611 | data_end = old_data; |
3612 | } | |
5f39d397 | 3613 | |
62e2749e | 3614 | /* setup the item for the new data */ |
9c58309d CM |
3615 | for (i = 0; i < nr; i++) { |
3616 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3617 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3618 | item = btrfs_item_nr(leaf, slot + i); | |
3619 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3620 | data_end -= data_size[i]; | |
3621 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3622 | } | |
44871b1b | 3623 | |
9c58309d | 3624 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed CM |
3625 | |
3626 | ret = 0; | |
5a01a2e3 | 3627 | if (slot == 0) { |
44871b1b | 3628 | struct btrfs_disk_key disk_key; |
5a01a2e3 | 3629 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); |
e089f05c | 3630 | ret = fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 3631 | } |
b9473439 CM |
3632 | btrfs_unlock_up_safe(path, 1); |
3633 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 3634 | |
5f39d397 CM |
3635 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3636 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 3637 | BUG(); |
5f39d397 | 3638 | } |
44871b1b CM |
3639 | return ret; |
3640 | } | |
3641 | ||
3642 | /* | |
3643 | * Given a key and some data, insert items into the tree. | |
3644 | * This does all the path init required, making room in the tree if needed. | |
3645 | */ | |
3646 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
3647 | struct btrfs_root *root, | |
3648 | struct btrfs_path *path, | |
3649 | struct btrfs_key *cpu_key, u32 *data_size, | |
3650 | int nr) | |
3651 | { | |
3652 | struct extent_buffer *leaf; | |
3653 | int ret = 0; | |
3654 | int slot; | |
3655 | int i; | |
3656 | u32 total_size = 0; | |
3657 | u32 total_data = 0; | |
3658 | ||
3659 | for (i = 0; i < nr; i++) | |
3660 | total_data += data_size[i]; | |
3661 | ||
3662 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
3663 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
3664 | if (ret == 0) | |
3665 | return -EEXIST; | |
3666 | if (ret < 0) | |
3667 | goto out; | |
3668 | ||
3669 | leaf = path->nodes[0]; | |
3670 | slot = path->slots[0]; | |
3671 | BUG_ON(slot < 0); | |
3672 | ||
3673 | ret = setup_items_for_insert(trans, root, path, cpu_key, data_size, | |
3674 | total_data, total_size, nr); | |
3675 | ||
ed2ff2cb | 3676 | out: |
62e2749e CM |
3677 | return ret; |
3678 | } | |
3679 | ||
3680 | /* | |
3681 | * Given a key and some data, insert an item into the tree. | |
3682 | * This does all the path init required, making room in the tree if needed. | |
3683 | */ | |
e089f05c CM |
3684 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
3685 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
3686 | data_size) | |
62e2749e CM |
3687 | { |
3688 | int ret = 0; | |
2c90e5d6 | 3689 | struct btrfs_path *path; |
5f39d397 CM |
3690 | struct extent_buffer *leaf; |
3691 | unsigned long ptr; | |
62e2749e | 3692 | |
2c90e5d6 CM |
3693 | path = btrfs_alloc_path(); |
3694 | BUG_ON(!path); | |
2c90e5d6 | 3695 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 3696 | if (!ret) { |
5f39d397 CM |
3697 | leaf = path->nodes[0]; |
3698 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3699 | write_extent_buffer(leaf, data, ptr, data_size); | |
3700 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 3701 | } |
2c90e5d6 | 3702 | btrfs_free_path(path); |
aa5d6bed | 3703 | return ret; |
be0e5c09 CM |
3704 | } |
3705 | ||
74123bd7 | 3706 | /* |
5de08d7d | 3707 | * delete the pointer from a given node. |
74123bd7 | 3708 | * |
d352ac68 CM |
3709 | * the tree should have been previously balanced so the deletion does not |
3710 | * empty a node. | |
74123bd7 | 3711 | */ |
e089f05c CM |
3712 | static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3713 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 3714 | { |
5f39d397 | 3715 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 3716 | u32 nritems; |
aa5d6bed | 3717 | int ret = 0; |
bb803951 | 3718 | int wret; |
be0e5c09 | 3719 | |
5f39d397 | 3720 | nritems = btrfs_header_nritems(parent); |
d397712b | 3721 | if (slot != nritems - 1) { |
5f39d397 CM |
3722 | memmove_extent_buffer(parent, |
3723 | btrfs_node_key_ptr_offset(slot), | |
3724 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
3725 | sizeof(struct btrfs_key_ptr) * |
3726 | (nritems - slot - 1)); | |
bb803951 | 3727 | } |
7518a238 | 3728 | nritems--; |
5f39d397 | 3729 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 3730 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 3731 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 3732 | /* just turn the root into a leaf and break */ |
5f39d397 | 3733 | btrfs_set_header_level(root->node, 0); |
bb803951 | 3734 | } else if (slot == 0) { |
5f39d397 CM |
3735 | struct btrfs_disk_key disk_key; |
3736 | ||
3737 | btrfs_node_key(parent, &disk_key, 0); | |
3738 | wret = fixup_low_keys(trans, root, path, &disk_key, level + 1); | |
0f70abe2 CM |
3739 | if (wret) |
3740 | ret = wret; | |
be0e5c09 | 3741 | } |
d6025579 | 3742 | btrfs_mark_buffer_dirty(parent); |
aa5d6bed | 3743 | return ret; |
be0e5c09 CM |
3744 | } |
3745 | ||
323ac95b CM |
3746 | /* |
3747 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
3748 | * path->nodes[1]. bytenr is the node block pointer, but since the callers | |
3749 | * already know it, it is faster to have them pass it down than to | |
3750 | * read it out of the node again. | |
3751 | * | |
3752 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
3753 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
3754 | * | |
3755 | * The path must have already been setup for deleting the leaf, including | |
3756 | * all the proper balancing. path->nodes[1] must be locked. | |
3757 | */ | |
3758 | noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans, | |
3759 | struct btrfs_root *root, | |
3760 | struct btrfs_path *path, u64 bytenr) | |
3761 | { | |
3762 | int ret; | |
3763 | u64 root_gen = btrfs_header_generation(path->nodes[1]); | |
4d081c41 CM |
3764 | u64 parent_start = path->nodes[1]->start; |
3765 | u64 parent_owner = btrfs_header_owner(path->nodes[1]); | |
323ac95b CM |
3766 | |
3767 | ret = del_ptr(trans, root, path, 1, path->slots[1]); | |
3768 | if (ret) | |
3769 | return ret; | |
3770 | ||
4d081c41 CM |
3771 | /* |
3772 | * btrfs_free_extent is expensive, we want to make sure we | |
3773 | * aren't holding any locks when we call it | |
3774 | */ | |
3775 | btrfs_unlock_up_safe(path, 0); | |
3776 | ||
323ac95b CM |
3777 | ret = btrfs_free_extent(trans, root, bytenr, |
3778 | btrfs_level_size(root, 0), | |
4d081c41 | 3779 | parent_start, parent_owner, |
3bb1a1bc | 3780 | root_gen, 0, 1); |
323ac95b CM |
3781 | return ret; |
3782 | } | |
74123bd7 CM |
3783 | /* |
3784 | * delete the item at the leaf level in path. If that empties | |
3785 | * the leaf, remove it from the tree | |
3786 | */ | |
85e21bac CM |
3787 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3788 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 3789 | { |
5f39d397 CM |
3790 | struct extent_buffer *leaf; |
3791 | struct btrfs_item *item; | |
85e21bac CM |
3792 | int last_off; |
3793 | int dsize = 0; | |
aa5d6bed CM |
3794 | int ret = 0; |
3795 | int wret; | |
85e21bac | 3796 | int i; |
7518a238 | 3797 | u32 nritems; |
be0e5c09 | 3798 | |
5f39d397 | 3799 | leaf = path->nodes[0]; |
85e21bac CM |
3800 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
3801 | ||
3802 | for (i = 0; i < nr; i++) | |
3803 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
3804 | ||
5f39d397 | 3805 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 3806 | |
85e21bac | 3807 | if (slot + nr != nritems) { |
123abc88 | 3808 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
3809 | |
3810 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
3811 | data_end + dsize, |
3812 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 3813 | last_off - data_end); |
5f39d397 | 3814 | |
85e21bac | 3815 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 3816 | u32 ioff; |
db94535d | 3817 | |
5f39d397 | 3818 | item = btrfs_item_nr(leaf, i); |
db94535d CM |
3819 | if (!leaf->map_token) { |
3820 | map_extent_buffer(leaf, (unsigned long)item, | |
3821 | sizeof(struct btrfs_item), | |
3822 | &leaf->map_token, &leaf->kaddr, | |
3823 | &leaf->map_start, &leaf->map_len, | |
3824 | KM_USER1); | |
3825 | } | |
5f39d397 CM |
3826 | ioff = btrfs_item_offset(leaf, item); |
3827 | btrfs_set_item_offset(leaf, item, ioff + dsize); | |
0783fcfc | 3828 | } |
db94535d CM |
3829 | |
3830 | if (leaf->map_token) { | |
3831 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3832 | leaf->map_token = NULL; | |
3833 | } | |
3834 | ||
5f39d397 | 3835 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 3836 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 3837 | sizeof(struct btrfs_item) * |
85e21bac | 3838 | (nritems - slot - nr)); |
be0e5c09 | 3839 | } |
85e21bac CM |
3840 | btrfs_set_header_nritems(leaf, nritems - nr); |
3841 | nritems -= nr; | |
5f39d397 | 3842 | |
74123bd7 | 3843 | /* delete the leaf if we've emptied it */ |
7518a238 | 3844 | if (nritems == 0) { |
5f39d397 CM |
3845 | if (leaf == root->node) { |
3846 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 3847 | } else { |
323ac95b CM |
3848 | ret = btrfs_del_leaf(trans, root, path, leaf->start); |
3849 | BUG_ON(ret); | |
9a8dd150 | 3850 | } |
be0e5c09 | 3851 | } else { |
7518a238 | 3852 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 3853 | if (slot == 0) { |
5f39d397 CM |
3854 | struct btrfs_disk_key disk_key; |
3855 | ||
3856 | btrfs_item_key(leaf, &disk_key, 0); | |
e089f05c | 3857 | wret = fixup_low_keys(trans, root, path, |
5f39d397 | 3858 | &disk_key, 1); |
aa5d6bed CM |
3859 | if (wret) |
3860 | ret = wret; | |
3861 | } | |
aa5d6bed | 3862 | |
74123bd7 | 3863 | /* delete the leaf if it is mostly empty */ |
a4b6e07d CM |
3864 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 4 && |
3865 | !trans->transaction->delayed_refs.flushing) { | |
be0e5c09 CM |
3866 | /* push_leaf_left fixes the path. |
3867 | * make sure the path still points to our leaf | |
3868 | * for possible call to del_ptr below | |
3869 | */ | |
4920c9ac | 3870 | slot = path->slots[1]; |
5f39d397 CM |
3871 | extent_buffer_get(leaf); |
3872 | ||
b9473439 | 3873 | btrfs_set_path_blocking(path); |
85e21bac | 3874 | wret = push_leaf_left(trans, root, path, 1, 1); |
54aa1f4d | 3875 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 3876 | ret = wret; |
5f39d397 CM |
3877 | |
3878 | if (path->nodes[0] == leaf && | |
3879 | btrfs_header_nritems(leaf)) { | |
85e21bac | 3880 | wret = push_leaf_right(trans, root, path, 1, 1); |
54aa1f4d | 3881 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
3882 | ret = wret; |
3883 | } | |
5f39d397 CM |
3884 | |
3885 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 3886 | path->slots[1] = slot; |
d397712b CM |
3887 | ret = btrfs_del_leaf(trans, root, path, |
3888 | leaf->start); | |
323ac95b | 3889 | BUG_ON(ret); |
5f39d397 | 3890 | free_extent_buffer(leaf); |
5de08d7d | 3891 | } else { |
925baedd CM |
3892 | /* if we're still in the path, make sure |
3893 | * we're dirty. Otherwise, one of the | |
3894 | * push_leaf functions must have already | |
3895 | * dirtied this buffer | |
3896 | */ | |
3897 | if (path->nodes[0] == leaf) | |
3898 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 3899 | free_extent_buffer(leaf); |
be0e5c09 | 3900 | } |
d5719762 | 3901 | } else { |
5f39d397 | 3902 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
3903 | } |
3904 | } | |
aa5d6bed | 3905 | return ret; |
be0e5c09 CM |
3906 | } |
3907 | ||
7bb86316 | 3908 | /* |
925baedd | 3909 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
3910 | * returns 0 if it found something or 1 if there are no lesser leaves. |
3911 | * returns < 0 on io errors. | |
d352ac68 CM |
3912 | * |
3913 | * This may release the path, and so you may lose any locks held at the | |
3914 | * time you call it. | |
7bb86316 CM |
3915 | */ |
3916 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
3917 | { | |
925baedd CM |
3918 | struct btrfs_key key; |
3919 | struct btrfs_disk_key found_key; | |
3920 | int ret; | |
7bb86316 | 3921 | |
925baedd | 3922 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 3923 | |
925baedd CM |
3924 | if (key.offset > 0) |
3925 | key.offset--; | |
3926 | else if (key.type > 0) | |
3927 | key.type--; | |
3928 | else if (key.objectid > 0) | |
3929 | key.objectid--; | |
3930 | else | |
3931 | return 1; | |
7bb86316 | 3932 | |
925baedd CM |
3933 | btrfs_release_path(root, path); |
3934 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3935 | if (ret < 0) | |
3936 | return ret; | |
3937 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
3938 | ret = comp_keys(&found_key, &key); | |
3939 | if (ret < 0) | |
3940 | return 0; | |
3941 | return 1; | |
7bb86316 CM |
3942 | } |
3943 | ||
3f157a2f CM |
3944 | /* |
3945 | * A helper function to walk down the tree starting at min_key, and looking | |
3946 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 3947 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
3948 | * |
3949 | * This does not cow, but it does stuff the starting key it finds back | |
3950 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
3951 | * key and get a writable path. | |
3952 | * | |
3953 | * This does lock as it descends, and path->keep_locks should be set | |
3954 | * to 1 by the caller. | |
3955 | * | |
3956 | * This honors path->lowest_level to prevent descent past a given level | |
3957 | * of the tree. | |
3958 | * | |
d352ac68 CM |
3959 | * min_trans indicates the oldest transaction that you are interested |
3960 | * in walking through. Any nodes or leaves older than min_trans are | |
3961 | * skipped over (without reading them). | |
3962 | * | |
3f157a2f CM |
3963 | * returns zero if something useful was found, < 0 on error and 1 if there |
3964 | * was nothing in the tree that matched the search criteria. | |
3965 | */ | |
3966 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 3967 | struct btrfs_key *max_key, |
3f157a2f CM |
3968 | struct btrfs_path *path, int cache_only, |
3969 | u64 min_trans) | |
3970 | { | |
3971 | struct extent_buffer *cur; | |
3972 | struct btrfs_key found_key; | |
3973 | int slot; | |
9652480b | 3974 | int sret; |
3f157a2f CM |
3975 | u32 nritems; |
3976 | int level; | |
3977 | int ret = 1; | |
3978 | ||
934d375b | 3979 | WARN_ON(!path->keep_locks); |
3f157a2f CM |
3980 | again: |
3981 | cur = btrfs_lock_root_node(root); | |
3982 | level = btrfs_header_level(cur); | |
e02119d5 | 3983 | WARN_ON(path->nodes[level]); |
3f157a2f CM |
3984 | path->nodes[level] = cur; |
3985 | path->locks[level] = 1; | |
3986 | ||
3987 | if (btrfs_header_generation(cur) < min_trans) { | |
3988 | ret = 1; | |
3989 | goto out; | |
3990 | } | |
d397712b | 3991 | while (1) { |
3f157a2f CM |
3992 | nritems = btrfs_header_nritems(cur); |
3993 | level = btrfs_header_level(cur); | |
9652480b | 3994 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 3995 | |
323ac95b CM |
3996 | /* at the lowest level, we're done, setup the path and exit */ |
3997 | if (level == path->lowest_level) { | |
e02119d5 CM |
3998 | if (slot >= nritems) |
3999 | goto find_next_key; | |
3f157a2f CM |
4000 | ret = 0; |
4001 | path->slots[level] = slot; | |
4002 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4003 | goto out; | |
4004 | } | |
9652480b Y |
4005 | if (sret && slot > 0) |
4006 | slot--; | |
3f157a2f CM |
4007 | /* |
4008 | * check this node pointer against the cache_only and | |
4009 | * min_trans parameters. If it isn't in cache or is too | |
4010 | * old, skip to the next one. | |
4011 | */ | |
d397712b | 4012 | while (slot < nritems) { |
3f157a2f CM |
4013 | u64 blockptr; |
4014 | u64 gen; | |
4015 | struct extent_buffer *tmp; | |
e02119d5 CM |
4016 | struct btrfs_disk_key disk_key; |
4017 | ||
3f157a2f CM |
4018 | blockptr = btrfs_node_blockptr(cur, slot); |
4019 | gen = btrfs_node_ptr_generation(cur, slot); | |
4020 | if (gen < min_trans) { | |
4021 | slot++; | |
4022 | continue; | |
4023 | } | |
4024 | if (!cache_only) | |
4025 | break; | |
4026 | ||
e02119d5 CM |
4027 | if (max_key) { |
4028 | btrfs_node_key(cur, &disk_key, slot); | |
4029 | if (comp_keys(&disk_key, max_key) >= 0) { | |
4030 | ret = 1; | |
4031 | goto out; | |
4032 | } | |
4033 | } | |
4034 | ||
3f157a2f CM |
4035 | tmp = btrfs_find_tree_block(root, blockptr, |
4036 | btrfs_level_size(root, level - 1)); | |
4037 | ||
4038 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
4039 | free_extent_buffer(tmp); | |
4040 | break; | |
4041 | } | |
4042 | if (tmp) | |
4043 | free_extent_buffer(tmp); | |
4044 | slot++; | |
4045 | } | |
e02119d5 | 4046 | find_next_key: |
3f157a2f CM |
4047 | /* |
4048 | * we didn't find a candidate key in this node, walk forward | |
4049 | * and find another one | |
4050 | */ | |
4051 | if (slot >= nritems) { | |
e02119d5 | 4052 | path->slots[level] = slot; |
b4ce94de | 4053 | btrfs_set_path_blocking(path); |
e02119d5 | 4054 | sret = btrfs_find_next_key(root, path, min_key, level, |
3f157a2f | 4055 | cache_only, min_trans); |
e02119d5 | 4056 | if (sret == 0) { |
3f157a2f CM |
4057 | btrfs_release_path(root, path); |
4058 | goto again; | |
4059 | } else { | |
4060 | goto out; | |
4061 | } | |
4062 | } | |
4063 | /* save our key for returning back */ | |
4064 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4065 | path->slots[level] = slot; | |
4066 | if (level == path->lowest_level) { | |
4067 | ret = 0; | |
4068 | unlock_up(path, level, 1); | |
4069 | goto out; | |
4070 | } | |
b4ce94de | 4071 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4072 | cur = read_node_slot(root, cur, slot); |
4073 | ||
4074 | btrfs_tree_lock(cur); | |
b4ce94de | 4075 | |
3f157a2f CM |
4076 | path->locks[level - 1] = 1; |
4077 | path->nodes[level - 1] = cur; | |
4078 | unlock_up(path, level, 1); | |
4008c04a | 4079 | btrfs_clear_path_blocking(path, NULL); |
3f157a2f CM |
4080 | } |
4081 | out: | |
4082 | if (ret == 0) | |
4083 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4084 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4085 | return ret; |
4086 | } | |
4087 | ||
4088 | /* | |
4089 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4090 | * and fixup the path. It looks for and returns the next key in the | |
4091 | * tree based on the current path and the cache_only and min_trans | |
4092 | * parameters. | |
4093 | * | |
4094 | * 0 is returned if another key is found, < 0 if there are any errors | |
4095 | * and 1 is returned if there are no higher keys in the tree | |
4096 | * | |
4097 | * path->keep_locks should be set to 1 on the search made before | |
4098 | * calling this function. | |
4099 | */ | |
e7a84565 | 4100 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
3f157a2f CM |
4101 | struct btrfs_key *key, int lowest_level, |
4102 | int cache_only, u64 min_trans) | |
e7a84565 CM |
4103 | { |
4104 | int level = lowest_level; | |
4105 | int slot; | |
4106 | struct extent_buffer *c; | |
4107 | ||
934d375b | 4108 | WARN_ON(!path->keep_locks); |
d397712b | 4109 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4110 | if (!path->nodes[level]) |
4111 | return 1; | |
4112 | ||
4113 | slot = path->slots[level] + 1; | |
4114 | c = path->nodes[level]; | |
3f157a2f | 4115 | next: |
e7a84565 CM |
4116 | if (slot >= btrfs_header_nritems(c)) { |
4117 | level++; | |
d397712b | 4118 | if (level == BTRFS_MAX_LEVEL) |
e7a84565 | 4119 | return 1; |
e7a84565 CM |
4120 | continue; |
4121 | } | |
4122 | if (level == 0) | |
4123 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
4124 | else { |
4125 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
4126 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
4127 | ||
4128 | if (cache_only) { | |
4129 | struct extent_buffer *cur; | |
4130 | cur = btrfs_find_tree_block(root, blockptr, | |
4131 | btrfs_level_size(root, level - 1)); | |
4132 | if (!cur || !btrfs_buffer_uptodate(cur, gen)) { | |
4133 | slot++; | |
4134 | if (cur) | |
4135 | free_extent_buffer(cur); | |
4136 | goto next; | |
4137 | } | |
4138 | free_extent_buffer(cur); | |
4139 | } | |
4140 | if (gen < min_trans) { | |
4141 | slot++; | |
4142 | goto next; | |
4143 | } | |
e7a84565 | 4144 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4145 | } |
e7a84565 CM |
4146 | return 0; |
4147 | } | |
4148 | return 1; | |
4149 | } | |
4150 | ||
97571fd0 | 4151 | /* |
925baedd | 4152 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
4153 | * returns 0 if it found something or 1 if there are no greater leaves. |
4154 | * returns < 0 on io errors. | |
97571fd0 | 4155 | */ |
234b63a0 | 4156 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
4157 | { |
4158 | int slot; | |
8e73f275 | 4159 | int level; |
5f39d397 | 4160 | struct extent_buffer *c; |
8e73f275 | 4161 | struct extent_buffer *next; |
925baedd CM |
4162 | struct btrfs_key key; |
4163 | u32 nritems; | |
4164 | int ret; | |
8e73f275 CM |
4165 | int old_spinning = path->leave_spinning; |
4166 | int force_blocking = 0; | |
925baedd CM |
4167 | |
4168 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4169 | if (nritems == 0) |
925baedd | 4170 | return 1; |
925baedd | 4171 | |
8e73f275 CM |
4172 | /* |
4173 | * we take the blocks in an order that upsets lockdep. Using | |
4174 | * blocking mode is the only way around it. | |
4175 | */ | |
4176 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
4177 | force_blocking = 1; | |
4178 | #endif | |
925baedd | 4179 | |
8e73f275 CM |
4180 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4181 | again: | |
4182 | level = 1; | |
4183 | next = NULL; | |
925baedd | 4184 | btrfs_release_path(root, path); |
8e73f275 | 4185 | |
a2135011 | 4186 | path->keep_locks = 1; |
8e73f275 CM |
4187 | |
4188 | if (!force_blocking) | |
4189 | path->leave_spinning = 1; | |
4190 | ||
925baedd CM |
4191 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4192 | path->keep_locks = 0; | |
4193 | ||
4194 | if (ret < 0) | |
4195 | return ret; | |
4196 | ||
a2135011 | 4197 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4198 | /* |
4199 | * by releasing the path above we dropped all our locks. A balance | |
4200 | * could have added more items next to the key that used to be | |
4201 | * at the very end of the block. So, check again here and | |
4202 | * advance the path if there are now more items available. | |
4203 | */ | |
a2135011 | 4204 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
168fd7d2 | 4205 | path->slots[0]++; |
8e73f275 | 4206 | ret = 0; |
925baedd CM |
4207 | goto done; |
4208 | } | |
d97e63b6 | 4209 | |
d397712b | 4210 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4211 | if (!path->nodes[level]) { |
4212 | ret = 1; | |
4213 | goto done; | |
4214 | } | |
5f39d397 | 4215 | |
d97e63b6 CM |
4216 | slot = path->slots[level] + 1; |
4217 | c = path->nodes[level]; | |
5f39d397 | 4218 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4219 | level++; |
8e73f275 CM |
4220 | if (level == BTRFS_MAX_LEVEL) { |
4221 | ret = 1; | |
4222 | goto done; | |
4223 | } | |
d97e63b6 CM |
4224 | continue; |
4225 | } | |
5f39d397 | 4226 | |
925baedd CM |
4227 | if (next) { |
4228 | btrfs_tree_unlock(next); | |
5f39d397 | 4229 | free_extent_buffer(next); |
925baedd | 4230 | } |
5f39d397 | 4231 | |
8e73f275 CM |
4232 | next = c; |
4233 | ret = read_block_for_search(NULL, root, path, &next, level, | |
4234 | slot, &key); | |
4235 | if (ret == -EAGAIN) | |
4236 | goto again; | |
5f39d397 | 4237 | |
76a05b35 CM |
4238 | if (ret < 0) { |
4239 | btrfs_release_path(root, path); | |
4240 | goto done; | |
4241 | } | |
4242 | ||
5cd57b2c | 4243 | if (!path->skip_locking) { |
8e73f275 CM |
4244 | ret = btrfs_try_spin_lock(next); |
4245 | if (!ret) { | |
4246 | btrfs_set_path_blocking(path); | |
4247 | btrfs_tree_lock(next); | |
4248 | if (!force_blocking) | |
4249 | btrfs_clear_path_blocking(path, next); | |
4250 | } | |
4251 | if (force_blocking) | |
4252 | btrfs_set_lock_blocking(next); | |
5cd57b2c | 4253 | } |
d97e63b6 CM |
4254 | break; |
4255 | } | |
4256 | path->slots[level] = slot; | |
d397712b | 4257 | while (1) { |
d97e63b6 CM |
4258 | level--; |
4259 | c = path->nodes[level]; | |
925baedd CM |
4260 | if (path->locks[level]) |
4261 | btrfs_tree_unlock(c); | |
8e73f275 | 4262 | |
5f39d397 | 4263 | free_extent_buffer(c); |
d97e63b6 CM |
4264 | path->nodes[level] = next; |
4265 | path->slots[level] = 0; | |
a74a4b97 CM |
4266 | if (!path->skip_locking) |
4267 | path->locks[level] = 1; | |
8e73f275 | 4268 | |
d97e63b6 CM |
4269 | if (!level) |
4270 | break; | |
b4ce94de | 4271 | |
8e73f275 CM |
4272 | ret = read_block_for_search(NULL, root, path, &next, level, |
4273 | 0, &key); | |
4274 | if (ret == -EAGAIN) | |
4275 | goto again; | |
4276 | ||
76a05b35 CM |
4277 | if (ret < 0) { |
4278 | btrfs_release_path(root, path); | |
4279 | goto done; | |
4280 | } | |
4281 | ||
5cd57b2c | 4282 | if (!path->skip_locking) { |
b9447ef8 | 4283 | btrfs_assert_tree_locked(path->nodes[level]); |
8e73f275 CM |
4284 | ret = btrfs_try_spin_lock(next); |
4285 | if (!ret) { | |
4286 | btrfs_set_path_blocking(path); | |
4287 | btrfs_tree_lock(next); | |
4288 | if (!force_blocking) | |
4289 | btrfs_clear_path_blocking(path, next); | |
4290 | } | |
4291 | if (force_blocking) | |
4292 | btrfs_set_lock_blocking(next); | |
5cd57b2c | 4293 | } |
d97e63b6 | 4294 | } |
8e73f275 | 4295 | ret = 0; |
925baedd CM |
4296 | done: |
4297 | unlock_up(path, 0, 1); | |
8e73f275 CM |
4298 | path->leave_spinning = old_spinning; |
4299 | if (!old_spinning) | |
4300 | btrfs_set_path_blocking(path); | |
4301 | ||
4302 | return ret; | |
d97e63b6 | 4303 | } |
0b86a832 | 4304 | |
3f157a2f CM |
4305 | /* |
4306 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4307 | * searching until it gets past min_objectid or finds an item of 'type' | |
4308 | * | |
4309 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4310 | */ | |
0b86a832 CM |
4311 | int btrfs_previous_item(struct btrfs_root *root, |
4312 | struct btrfs_path *path, u64 min_objectid, | |
4313 | int type) | |
4314 | { | |
4315 | struct btrfs_key found_key; | |
4316 | struct extent_buffer *leaf; | |
e02119d5 | 4317 | u32 nritems; |
0b86a832 CM |
4318 | int ret; |
4319 | ||
d397712b | 4320 | while (1) { |
0b86a832 | 4321 | if (path->slots[0] == 0) { |
b4ce94de | 4322 | btrfs_set_path_blocking(path); |
0b86a832 CM |
4323 | ret = btrfs_prev_leaf(root, path); |
4324 | if (ret != 0) | |
4325 | return ret; | |
4326 | } else { | |
4327 | path->slots[0]--; | |
4328 | } | |
4329 | leaf = path->nodes[0]; | |
e02119d5 CM |
4330 | nritems = btrfs_header_nritems(leaf); |
4331 | if (nritems == 0) | |
4332 | return 1; | |
4333 | if (path->slots[0] == nritems) | |
4334 | path->slots[0]--; | |
4335 | ||
0b86a832 CM |
4336 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4337 | if (found_key.type == type) | |
4338 | return 0; | |
e02119d5 CM |
4339 | if (found_key.objectid < min_objectid) |
4340 | break; | |
4341 | if (found_key.objectid == min_objectid && | |
4342 | found_key.type < type) | |
4343 | break; | |
0b86a832 CM |
4344 | } |
4345 | return 1; | |
4346 | } |