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 | inline void btrfs_init_path(struct btrfs_path *p) |
2c90e5d6 | 42 | { |
df24a2b9 | 43 | memset(p, 0, sizeof(*p)); |
2c90e5d6 CM |
44 | } |
45 | ||
df24a2b9 | 46 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 47 | { |
df24a2b9 CM |
48 | struct btrfs_path *path; |
49 | path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS); | |
2cc58cf2 | 50 | if (path) { |
df24a2b9 | 51 | btrfs_init_path(path); |
2cc58cf2 CM |
52 | path->reada = 1; |
53 | } | |
df24a2b9 | 54 | return path; |
2c90e5d6 CM |
55 | } |
56 | ||
d352ac68 | 57 | /* this also releases the path */ |
df24a2b9 | 58 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 59 | { |
df24a2b9 CM |
60 | btrfs_release_path(NULL, p); |
61 | kmem_cache_free(btrfs_path_cachep, p); | |
be0e5c09 CM |
62 | } |
63 | ||
d352ac68 CM |
64 | /* |
65 | * path release drops references on the extent buffers in the path | |
66 | * and it drops any locks held by this path | |
67 | * | |
68 | * It is safe to call this on paths that no locks or extent buffers held. | |
69 | */ | |
e02119d5 | 70 | void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p) |
eb60ceac CM |
71 | { |
72 | int i; | |
a2135011 | 73 | |
234b63a0 | 74 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 75 | p->slots[i] = 0; |
eb60ceac | 76 | if (!p->nodes[i]) |
925baedd CM |
77 | continue; |
78 | if (p->locks[i]) { | |
79 | btrfs_tree_unlock(p->nodes[i]); | |
80 | p->locks[i] = 0; | |
81 | } | |
5f39d397 | 82 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 83 | p->nodes[i] = NULL; |
eb60ceac CM |
84 | } |
85 | } | |
86 | ||
d352ac68 CM |
87 | /* |
88 | * safely gets a reference on the root node of a tree. A lock | |
89 | * is not taken, so a concurrent writer may put a different node | |
90 | * at the root of the tree. See btrfs_lock_root_node for the | |
91 | * looping required. | |
92 | * | |
93 | * The extent buffer returned by this has a reference taken, so | |
94 | * it won't disappear. It may stop being the root of the tree | |
95 | * at any time because there are no locks held. | |
96 | */ | |
925baedd CM |
97 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
98 | { | |
99 | struct extent_buffer *eb; | |
100 | spin_lock(&root->node_lock); | |
101 | eb = root->node; | |
102 | extent_buffer_get(eb); | |
103 | spin_unlock(&root->node_lock); | |
104 | return eb; | |
105 | } | |
106 | ||
d352ac68 CM |
107 | /* loop around taking references on and locking the root node of the |
108 | * tree until you end up with a lock on the root. A locked buffer | |
109 | * is returned, with a reference held. | |
110 | */ | |
925baedd CM |
111 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
112 | { | |
113 | struct extent_buffer *eb; | |
114 | ||
115 | while(1) { | |
116 | eb = btrfs_root_node(root); | |
117 | btrfs_tree_lock(eb); | |
118 | ||
119 | spin_lock(&root->node_lock); | |
120 | if (eb == root->node) { | |
121 | spin_unlock(&root->node_lock); | |
122 | break; | |
123 | } | |
124 | spin_unlock(&root->node_lock); | |
125 | ||
126 | btrfs_tree_unlock(eb); | |
127 | free_extent_buffer(eb); | |
128 | } | |
129 | return eb; | |
130 | } | |
131 | ||
d352ac68 CM |
132 | /* cowonly root (everything not a reference counted cow subvolume), just get |
133 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
134 | * get properly updated on disk. | |
135 | */ | |
0b86a832 CM |
136 | static void add_root_to_dirty_list(struct btrfs_root *root) |
137 | { | |
138 | if (root->track_dirty && list_empty(&root->dirty_list)) { | |
139 | list_add(&root->dirty_list, | |
140 | &root->fs_info->dirty_cowonly_roots); | |
141 | } | |
142 | } | |
143 | ||
d352ac68 CM |
144 | /* |
145 | * used by snapshot creation to make a copy of a root for a tree with | |
146 | * a given objectid. The buffer with the new root node is returned in | |
147 | * cow_ret, and this func returns zero on success or a negative error code. | |
148 | */ | |
be20aa9d CM |
149 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
150 | struct btrfs_root *root, | |
151 | struct extent_buffer *buf, | |
152 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
153 | { | |
154 | struct extent_buffer *cow; | |
155 | u32 nritems; | |
156 | int ret = 0; | |
157 | int level; | |
4aec2b52 | 158 | struct btrfs_root *new_root; |
be20aa9d | 159 | |
4aec2b52 CM |
160 | new_root = kmalloc(sizeof(*new_root), GFP_NOFS); |
161 | if (!new_root) | |
162 | return -ENOMEM; | |
163 | ||
164 | memcpy(new_root, root, sizeof(*new_root)); | |
165 | new_root->root_key.objectid = new_root_objectid; | |
be20aa9d CM |
166 | |
167 | WARN_ON(root->ref_cows && trans->transid != | |
168 | root->fs_info->running_transaction->transid); | |
169 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
170 | ||
171 | level = btrfs_header_level(buf); | |
172 | nritems = btrfs_header_nritems(buf); | |
31840ae1 ZY |
173 | |
174 | cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0, | |
175 | new_root_objectid, trans->transid, | |
176 | level, buf->start, 0); | |
4aec2b52 CM |
177 | if (IS_ERR(cow)) { |
178 | kfree(new_root); | |
be20aa9d | 179 | return PTR_ERR(cow); |
4aec2b52 | 180 | } |
be20aa9d CM |
181 | |
182 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
183 | btrfs_set_header_bytenr(cow, cow->start); | |
184 | btrfs_set_header_generation(cow, trans->transid); | |
185 | btrfs_set_header_owner(cow, new_root_objectid); | |
63b10fc4 | 186 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN); |
be20aa9d CM |
187 | |
188 | WARN_ON(btrfs_header_generation(buf) > trans->transid); | |
31840ae1 | 189 | ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL); |
4aec2b52 CM |
190 | kfree(new_root); |
191 | ||
be20aa9d CM |
192 | if (ret) |
193 | return ret; | |
194 | ||
195 | btrfs_mark_buffer_dirty(cow); | |
196 | *cow_ret = cow; | |
197 | return 0; | |
198 | } | |
199 | ||
d352ac68 CM |
200 | /* |
201 | * does the dirty work in cow of a single block. The parent block | |
202 | * (if supplied) is updated to point to the new cow copy. The new | |
203 | * buffer is marked dirty and returned locked. If you modify the block | |
204 | * it needs to be marked dirty again. | |
205 | * | |
206 | * search_start -- an allocation hint for the new block | |
207 | * | |
208 | * empty_size -- a hint that you plan on doing more cow. This is the size in bytes | |
209 | * the allocator should try to find free next to the block it returns. This is | |
210 | * just a hint and may be ignored by the allocator. | |
211 | * | |
212 | * prealloc_dest -- if you have already reserved a destination for the cow, | |
213 | * this uses that block instead of allocating a new one. btrfs_alloc_reserved_extent | |
214 | * is used to finish the allocation. | |
215 | */ | |
e02119d5 | 216 | int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
217 | struct btrfs_root *root, |
218 | struct extent_buffer *buf, | |
219 | struct extent_buffer *parent, int parent_slot, | |
220 | struct extent_buffer **cow_ret, | |
65b51a00 CM |
221 | u64 search_start, u64 empty_size, |
222 | u64 prealloc_dest) | |
02217ed2 | 223 | { |
31840ae1 | 224 | u64 parent_start; |
5f39d397 | 225 | struct extent_buffer *cow; |
7bb86316 | 226 | u32 nritems; |
6702ed49 | 227 | int ret = 0; |
7bb86316 | 228 | int level; |
925baedd | 229 | int unlock_orig = 0; |
7bb86316 | 230 | |
925baedd CM |
231 | if (*cow_ret == buf) |
232 | unlock_orig = 1; | |
233 | ||
234 | WARN_ON(!btrfs_tree_locked(buf)); | |
235 | ||
31840ae1 ZY |
236 | if (parent) |
237 | parent_start = parent->start; | |
238 | else | |
239 | parent_start = 0; | |
240 | ||
7bb86316 CM |
241 | WARN_ON(root->ref_cows && trans->transid != |
242 | root->fs_info->running_transaction->transid); | |
6702ed49 | 243 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 244 | |
7bb86316 CM |
245 | level = btrfs_header_level(buf); |
246 | nritems = btrfs_header_nritems(buf); | |
31840ae1 | 247 | |
65b51a00 CM |
248 | if (prealloc_dest) { |
249 | struct btrfs_key ins; | |
250 | ||
251 | ins.objectid = prealloc_dest; | |
252 | ins.offset = buf->len; | |
253 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
254 | ||
31840ae1 | 255 | ret = btrfs_alloc_reserved_extent(trans, root, parent_start, |
65b51a00 | 256 | root->root_key.objectid, |
3bb1a1bc | 257 | trans->transid, level, &ins); |
65b51a00 CM |
258 | BUG_ON(ret); |
259 | cow = btrfs_init_new_buffer(trans, root, prealloc_dest, | |
260 | buf->len); | |
261 | } else { | |
262 | cow = btrfs_alloc_free_block(trans, root, buf->len, | |
31840ae1 | 263 | parent_start, |
65b51a00 | 264 | root->root_key.objectid, |
31840ae1 ZY |
265 | trans->transid, level, |
266 | search_start, empty_size); | |
65b51a00 | 267 | } |
54aa1f4d CM |
268 | if (IS_ERR(cow)) |
269 | return PTR_ERR(cow); | |
6702ed49 | 270 | |
5f39d397 | 271 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 272 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 CM |
273 | btrfs_set_header_generation(cow, trans->transid); |
274 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
63b10fc4 | 275 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN); |
6702ed49 | 276 | |
5f39d397 CM |
277 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
278 | if (btrfs_header_generation(buf) != trans->transid) { | |
31840ae1 | 279 | u32 nr_extents; |
31840ae1 | 280 | ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents); |
6702ed49 CM |
281 | if (ret) |
282 | return ret; | |
31840ae1 ZY |
283 | |
284 | ret = btrfs_cache_ref(trans, root, buf, nr_extents); | |
285 | WARN_ON(ret); | |
1a40e23b ZY |
286 | } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) { |
287 | /* | |
288 | * There are only two places that can drop reference to | |
289 | * tree blocks owned by living reloc trees, one is here, | |
f82d02d9 | 290 | * the other place is btrfs_drop_subtree. In both places, |
1a40e23b ZY |
291 | * we check reference count while tree block is locked. |
292 | * Furthermore, if reference count is one, it won't get | |
293 | * increased by someone else. | |
294 | */ | |
295 | u32 refs; | |
296 | ret = btrfs_lookup_extent_ref(trans, root, buf->start, | |
297 | buf->len, &refs); | |
298 | BUG_ON(ret); | |
299 | if (refs == 1) { | |
300 | ret = btrfs_update_ref(trans, root, buf, cow, | |
301 | 0, nritems); | |
302 | clean_tree_block(trans, root, buf); | |
303 | } else { | |
304 | ret = btrfs_inc_ref(trans, root, buf, cow, NULL); | |
305 | } | |
306 | BUG_ON(ret); | |
6702ed49 | 307 | } else { |
31840ae1 ZY |
308 | ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems); |
309 | if (ret) | |
310 | return ret; | |
6702ed49 CM |
311 | clean_tree_block(trans, root, buf); |
312 | } | |
313 | ||
1a40e23b | 314 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { |
1a40e23b ZY |
315 | ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start); |
316 | WARN_ON(ret); | |
317 | } | |
318 | ||
02217ed2 | 319 | if (buf == root->node) { |
925baedd | 320 | WARN_ON(parent && parent != buf); |
925baedd CM |
321 | |
322 | spin_lock(&root->node_lock); | |
02217ed2 | 323 | root->node = cow; |
5f39d397 | 324 | extent_buffer_get(cow); |
925baedd CM |
325 | spin_unlock(&root->node_lock); |
326 | ||
2c90e5d6 | 327 | if (buf != root->commit_root) { |
db94535d | 328 | btrfs_free_extent(trans, root, buf->start, |
31840ae1 ZY |
329 | buf->len, buf->start, |
330 | root->root_key.objectid, | |
331 | btrfs_header_generation(buf), | |
3bb1a1bc | 332 | level, 1); |
2c90e5d6 | 333 | } |
5f39d397 | 334 | free_extent_buffer(buf); |
0b86a832 | 335 | add_root_to_dirty_list(root); |
02217ed2 | 336 | } else { |
5f39d397 | 337 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 338 | cow->start); |
74493f7a CM |
339 | WARN_ON(trans->transid == 0); |
340 | btrfs_set_node_ptr_generation(parent, parent_slot, | |
341 | trans->transid); | |
d6025579 | 342 | btrfs_mark_buffer_dirty(parent); |
5f39d397 | 343 | WARN_ON(btrfs_header_generation(parent) != trans->transid); |
7bb86316 | 344 | btrfs_free_extent(trans, root, buf->start, buf->len, |
31840ae1 | 345 | parent_start, btrfs_header_owner(parent), |
3bb1a1bc | 346 | btrfs_header_generation(parent), level, 1); |
02217ed2 | 347 | } |
925baedd CM |
348 | if (unlock_orig) |
349 | btrfs_tree_unlock(buf); | |
5f39d397 | 350 | free_extent_buffer(buf); |
ccd467d6 | 351 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 352 | *cow_ret = cow; |
02217ed2 CM |
353 | return 0; |
354 | } | |
355 | ||
d352ac68 CM |
356 | /* |
357 | * cows a single block, see __btrfs_cow_block for the real work. | |
358 | * This version of it has extra checks so that a block isn't cow'd more than | |
359 | * once per transaction, as long as it hasn't been written yet | |
360 | */ | |
e02119d5 | 361 | int noinline btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
362 | struct btrfs_root *root, struct extent_buffer *buf, |
363 | struct extent_buffer *parent, int parent_slot, | |
65b51a00 | 364 | struct extent_buffer **cow_ret, u64 prealloc_dest) |
6702ed49 CM |
365 | { |
366 | u64 search_start; | |
f510cfec | 367 | int ret; |
dc17ff8f | 368 | |
6702ed49 CM |
369 | if (trans->transaction != root->fs_info->running_transaction) { |
370 | printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid, | |
371 | root->fs_info->running_transaction->transid); | |
372 | WARN_ON(1); | |
373 | } | |
374 | if (trans->transid != root->fs_info->generation) { | |
375 | printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid, | |
376 | root->fs_info->generation); | |
377 | WARN_ON(1); | |
378 | } | |
dc17ff8f | 379 | |
63b10fc4 | 380 | spin_lock(&root->fs_info->hash_lock); |
5b21f2ed ZY |
381 | if (btrfs_header_generation(buf) == trans->transid && |
382 | btrfs_header_owner(buf) == root->root_key.objectid && | |
63b10fc4 | 383 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6702ed49 | 384 | *cow_ret = buf; |
63b10fc4 | 385 | spin_unlock(&root->fs_info->hash_lock); |
65b51a00 | 386 | WARN_ON(prealloc_dest); |
6702ed49 CM |
387 | return 0; |
388 | } | |
63b10fc4 | 389 | spin_unlock(&root->fs_info->hash_lock); |
0b86a832 | 390 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
f510cfec | 391 | ret = __btrfs_cow_block(trans, root, buf, parent, |
65b51a00 CM |
392 | parent_slot, cow_ret, search_start, 0, |
393 | prealloc_dest); | |
f510cfec | 394 | return ret; |
6702ed49 CM |
395 | } |
396 | ||
d352ac68 CM |
397 | /* |
398 | * helper function for defrag to decide if two blocks pointed to by a | |
399 | * node are actually close by | |
400 | */ | |
6b80053d | 401 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 402 | { |
6b80053d | 403 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 404 | return 1; |
6b80053d | 405 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
406 | return 1; |
407 | return 0; | |
408 | } | |
409 | ||
081e9573 CM |
410 | /* |
411 | * compare two keys in a memcmp fashion | |
412 | */ | |
413 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
414 | { | |
415 | struct btrfs_key k1; | |
416 | ||
417 | btrfs_disk_key_to_cpu(&k1, disk); | |
418 | ||
419 | if (k1.objectid > k2->objectid) | |
420 | return 1; | |
421 | if (k1.objectid < k2->objectid) | |
422 | return -1; | |
423 | if (k1.type > k2->type) | |
424 | return 1; | |
425 | if (k1.type < k2->type) | |
426 | return -1; | |
427 | if (k1.offset > k2->offset) | |
428 | return 1; | |
429 | if (k1.offset < k2->offset) | |
430 | return -1; | |
431 | return 0; | |
432 | } | |
433 | ||
434 | ||
d352ac68 CM |
435 | /* |
436 | * this is used by the defrag code to go through all the | |
437 | * leaves pointed to by a node and reallocate them so that | |
438 | * disk order is close to key order | |
439 | */ | |
6702ed49 | 440 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 441 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
442 | int start_slot, int cache_only, u64 *last_ret, |
443 | struct btrfs_key *progress) | |
6702ed49 | 444 | { |
6b80053d | 445 | struct extent_buffer *cur; |
6702ed49 | 446 | u64 blocknr; |
ca7a79ad | 447 | u64 gen; |
e9d0b13b CM |
448 | u64 search_start = *last_ret; |
449 | u64 last_block = 0; | |
6702ed49 CM |
450 | u64 other; |
451 | u32 parent_nritems; | |
6702ed49 CM |
452 | int end_slot; |
453 | int i; | |
454 | int err = 0; | |
f2183bde | 455 | int parent_level; |
6b80053d CM |
456 | int uptodate; |
457 | u32 blocksize; | |
081e9573 CM |
458 | int progress_passed = 0; |
459 | struct btrfs_disk_key disk_key; | |
6702ed49 | 460 | |
5708b959 CM |
461 | parent_level = btrfs_header_level(parent); |
462 | if (cache_only && parent_level != 1) | |
463 | return 0; | |
464 | ||
6702ed49 CM |
465 | if (trans->transaction != root->fs_info->running_transaction) { |
466 | printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid, | |
467 | root->fs_info->running_transaction->transid); | |
468 | WARN_ON(1); | |
469 | } | |
470 | if (trans->transid != root->fs_info->generation) { | |
471 | printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid, | |
472 | root->fs_info->generation); | |
473 | WARN_ON(1); | |
474 | } | |
86479a04 | 475 | |
6b80053d | 476 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 477 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
478 | end_slot = parent_nritems; |
479 | ||
480 | if (parent_nritems == 1) | |
481 | return 0; | |
482 | ||
483 | for (i = start_slot; i < end_slot; i++) { | |
484 | int close = 1; | |
a6b6e75e | 485 | |
5708b959 CM |
486 | if (!parent->map_token) { |
487 | map_extent_buffer(parent, | |
488 | btrfs_node_key_ptr_offset(i), | |
489 | sizeof(struct btrfs_key_ptr), | |
490 | &parent->map_token, &parent->kaddr, | |
491 | &parent->map_start, &parent->map_len, | |
492 | KM_USER1); | |
493 | } | |
081e9573 CM |
494 | btrfs_node_key(parent, &disk_key, i); |
495 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
496 | continue; | |
497 | ||
498 | progress_passed = 1; | |
6b80053d | 499 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 500 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
501 | if (last_block == 0) |
502 | last_block = blocknr; | |
5708b959 | 503 | |
6702ed49 | 504 | if (i > 0) { |
6b80053d CM |
505 | other = btrfs_node_blockptr(parent, i - 1); |
506 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 507 | } |
0ef3e66b | 508 | if (!close && i < end_slot - 2) { |
6b80053d CM |
509 | other = btrfs_node_blockptr(parent, i + 1); |
510 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 511 | } |
e9d0b13b CM |
512 | if (close) { |
513 | last_block = blocknr; | |
6702ed49 | 514 | continue; |
e9d0b13b | 515 | } |
5708b959 CM |
516 | if (parent->map_token) { |
517 | unmap_extent_buffer(parent, parent->map_token, | |
518 | KM_USER1); | |
519 | parent->map_token = NULL; | |
520 | } | |
6702ed49 | 521 | |
6b80053d CM |
522 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
523 | if (cur) | |
1259ab75 | 524 | uptodate = btrfs_buffer_uptodate(cur, gen); |
6b80053d CM |
525 | else |
526 | uptodate = 0; | |
5708b959 | 527 | if (!cur || !uptodate) { |
6702ed49 | 528 | if (cache_only) { |
6b80053d | 529 | free_extent_buffer(cur); |
6702ed49 CM |
530 | continue; |
531 | } | |
6b80053d CM |
532 | if (!cur) { |
533 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 534 | blocksize, gen); |
6b80053d | 535 | } else if (!uptodate) { |
ca7a79ad | 536 | btrfs_read_buffer(cur, gen); |
f2183bde | 537 | } |
6702ed49 | 538 | } |
e9d0b13b | 539 | if (search_start == 0) |
6b80053d | 540 | search_start = last_block; |
e9d0b13b | 541 | |
e7a84565 | 542 | btrfs_tree_lock(cur); |
6b80053d | 543 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 544 | &cur, search_start, |
6b80053d | 545 | min(16 * blocksize, |
65b51a00 | 546 | (end_slot - i) * blocksize), 0); |
252c38f0 | 547 | if (err) { |
e7a84565 | 548 | btrfs_tree_unlock(cur); |
6b80053d | 549 | free_extent_buffer(cur); |
6702ed49 | 550 | break; |
252c38f0 | 551 | } |
e7a84565 CM |
552 | search_start = cur->start; |
553 | last_block = cur->start; | |
f2183bde | 554 | *last_ret = search_start; |
e7a84565 CM |
555 | btrfs_tree_unlock(cur); |
556 | free_extent_buffer(cur); | |
6702ed49 | 557 | } |
5708b959 CM |
558 | if (parent->map_token) { |
559 | unmap_extent_buffer(parent, parent->map_token, | |
560 | KM_USER1); | |
561 | parent->map_token = NULL; | |
562 | } | |
6702ed49 CM |
563 | return err; |
564 | } | |
565 | ||
74123bd7 CM |
566 | /* |
567 | * The leaf data grows from end-to-front in the node. | |
568 | * this returns the address of the start of the last item, | |
569 | * which is the stop of the leaf data stack | |
570 | */ | |
123abc88 | 571 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 572 | struct extent_buffer *leaf) |
be0e5c09 | 573 | { |
5f39d397 | 574 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 575 | if (nr == 0) |
123abc88 | 576 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 577 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
578 | } |
579 | ||
d352ac68 CM |
580 | /* |
581 | * extra debugging checks to make sure all the items in a key are | |
582 | * well formed and in the proper order | |
583 | */ | |
123abc88 CM |
584 | static int check_node(struct btrfs_root *root, struct btrfs_path *path, |
585 | int level) | |
aa5d6bed | 586 | { |
5f39d397 CM |
587 | struct extent_buffer *parent = NULL; |
588 | struct extent_buffer *node = path->nodes[level]; | |
589 | struct btrfs_disk_key parent_key; | |
590 | struct btrfs_disk_key node_key; | |
aa5d6bed | 591 | int parent_slot; |
8d7be552 CM |
592 | int slot; |
593 | struct btrfs_key cpukey; | |
5f39d397 | 594 | u32 nritems = btrfs_header_nritems(node); |
aa5d6bed CM |
595 | |
596 | if (path->nodes[level + 1]) | |
5f39d397 | 597 | parent = path->nodes[level + 1]; |
a1f39630 | 598 | |
8d7be552 | 599 | slot = path->slots[level]; |
7518a238 CM |
600 | BUG_ON(nritems == 0); |
601 | if (parent) { | |
a1f39630 | 602 | parent_slot = path->slots[level + 1]; |
5f39d397 CM |
603 | btrfs_node_key(parent, &parent_key, parent_slot); |
604 | btrfs_node_key(node, &node_key, 0); | |
605 | BUG_ON(memcmp(&parent_key, &node_key, | |
e2fa7227 | 606 | sizeof(struct btrfs_disk_key))); |
1d4f8a0c | 607 | BUG_ON(btrfs_node_blockptr(parent, parent_slot) != |
db94535d | 608 | btrfs_header_bytenr(node)); |
aa5d6bed | 609 | } |
123abc88 | 610 | BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root)); |
8d7be552 | 611 | if (slot != 0) { |
5f39d397 CM |
612 | btrfs_node_key_to_cpu(node, &cpukey, slot - 1); |
613 | btrfs_node_key(node, &node_key, slot); | |
614 | BUG_ON(comp_keys(&node_key, &cpukey) <= 0); | |
8d7be552 CM |
615 | } |
616 | if (slot < nritems - 1) { | |
5f39d397 CM |
617 | btrfs_node_key_to_cpu(node, &cpukey, slot + 1); |
618 | btrfs_node_key(node, &node_key, slot); | |
619 | BUG_ON(comp_keys(&node_key, &cpukey) >= 0); | |
aa5d6bed CM |
620 | } |
621 | return 0; | |
622 | } | |
623 | ||
d352ac68 CM |
624 | /* |
625 | * extra checking to make sure all the items in a leaf are | |
626 | * well formed and in the proper order | |
627 | */ | |
123abc88 CM |
628 | static int check_leaf(struct btrfs_root *root, struct btrfs_path *path, |
629 | int level) | |
aa5d6bed | 630 | { |
5f39d397 CM |
631 | struct extent_buffer *leaf = path->nodes[level]; |
632 | struct extent_buffer *parent = NULL; | |
aa5d6bed | 633 | int parent_slot; |
8d7be552 | 634 | struct btrfs_key cpukey; |
5f39d397 CM |
635 | struct btrfs_disk_key parent_key; |
636 | struct btrfs_disk_key leaf_key; | |
637 | int slot = path->slots[0]; | |
8d7be552 | 638 | |
5f39d397 | 639 | u32 nritems = btrfs_header_nritems(leaf); |
aa5d6bed CM |
640 | |
641 | if (path->nodes[level + 1]) | |
5f39d397 | 642 | parent = path->nodes[level + 1]; |
7518a238 CM |
643 | |
644 | if (nritems == 0) | |
645 | return 0; | |
646 | ||
647 | if (parent) { | |
a1f39630 | 648 | parent_slot = path->slots[level + 1]; |
5f39d397 CM |
649 | btrfs_node_key(parent, &parent_key, parent_slot); |
650 | btrfs_item_key(leaf, &leaf_key, 0); | |
6702ed49 | 651 | |
5f39d397 | 652 | BUG_ON(memcmp(&parent_key, &leaf_key, |
e2fa7227 | 653 | sizeof(struct btrfs_disk_key))); |
1d4f8a0c | 654 | BUG_ON(btrfs_node_blockptr(parent, parent_slot) != |
db94535d | 655 | btrfs_header_bytenr(leaf)); |
5f39d397 CM |
656 | } |
657 | #if 0 | |
658 | for (i = 0; nritems > 1 && i < nritems - 2; i++) { | |
659 | btrfs_item_key_to_cpu(leaf, &cpukey, i + 1); | |
660 | btrfs_item_key(leaf, &leaf_key, i); | |
661 | if (comp_keys(&leaf_key, &cpukey) >= 0) { | |
662 | btrfs_print_leaf(root, leaf); | |
663 | printk("slot %d offset bad key\n", i); | |
664 | BUG_ON(1); | |
665 | } | |
666 | if (btrfs_item_offset_nr(leaf, i) != | |
667 | btrfs_item_end_nr(leaf, i + 1)) { | |
668 | btrfs_print_leaf(root, leaf); | |
669 | printk("slot %d offset bad\n", i); | |
670 | BUG_ON(1); | |
671 | } | |
672 | if (i == 0) { | |
673 | if (btrfs_item_offset_nr(leaf, i) + | |
674 | btrfs_item_size_nr(leaf, i) != | |
675 | BTRFS_LEAF_DATA_SIZE(root)) { | |
676 | btrfs_print_leaf(root, leaf); | |
677 | printk("slot %d first offset bad\n", i); | |
678 | BUG_ON(1); | |
679 | } | |
680 | } | |
aa5d6bed | 681 | } |
5f39d397 CM |
682 | if (nritems > 0) { |
683 | if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) { | |
684 | btrfs_print_leaf(root, leaf); | |
685 | printk("slot %d bad size \n", nritems - 1); | |
686 | BUG_ON(1); | |
687 | } | |
688 | } | |
689 | #endif | |
690 | if (slot != 0 && slot < nritems - 1) { | |
691 | btrfs_item_key(leaf, &leaf_key, slot); | |
692 | btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1); | |
693 | if (comp_keys(&leaf_key, &cpukey) <= 0) { | |
694 | btrfs_print_leaf(root, leaf); | |
695 | printk("slot %d offset bad key\n", slot); | |
696 | BUG_ON(1); | |
697 | } | |
698 | if (btrfs_item_offset_nr(leaf, slot - 1) != | |
699 | btrfs_item_end_nr(leaf, slot)) { | |
700 | btrfs_print_leaf(root, leaf); | |
701 | printk("slot %d offset bad\n", slot); | |
702 | BUG_ON(1); | |
703 | } | |
8d7be552 CM |
704 | } |
705 | if (slot < nritems - 1) { | |
5f39d397 CM |
706 | btrfs_item_key(leaf, &leaf_key, slot); |
707 | btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1); | |
708 | BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0); | |
709 | if (btrfs_item_offset_nr(leaf, slot) != | |
710 | btrfs_item_end_nr(leaf, slot + 1)) { | |
711 | btrfs_print_leaf(root, leaf); | |
712 | printk("slot %d offset bad\n", slot); | |
713 | BUG_ON(1); | |
714 | } | |
aa5d6bed | 715 | } |
5f39d397 CM |
716 | BUG_ON(btrfs_item_offset_nr(leaf, 0) + |
717 | btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root)); | |
aa5d6bed CM |
718 | return 0; |
719 | } | |
720 | ||
98ed5174 CM |
721 | static int noinline check_block(struct btrfs_root *root, |
722 | struct btrfs_path *path, int level) | |
aa5d6bed | 723 | { |
f188591e | 724 | u64 found_start; |
85d824c4 | 725 | return 0; |
f188591e CM |
726 | if (btrfs_header_level(path->nodes[level]) != level) |
727 | printk("warning: bad level %Lu wanted %d found %d\n", | |
728 | path->nodes[level]->start, level, | |
729 | btrfs_header_level(path->nodes[level])); | |
730 | found_start = btrfs_header_bytenr(path->nodes[level]); | |
731 | if (found_start != path->nodes[level]->start) { | |
732 | printk("warning: bad bytentr %Lu found %Lu\n", | |
733 | path->nodes[level]->start, found_start); | |
734 | } | |
db94535d | 735 | #if 0 |
5f39d397 | 736 | struct extent_buffer *buf = path->nodes[level]; |
5f39d397 | 737 | |
479965d6 CM |
738 | if (memcmp_extent_buffer(buf, root->fs_info->fsid, |
739 | (unsigned long)btrfs_header_fsid(buf), | |
740 | BTRFS_FSID_SIZE)) { | |
5f39d397 | 741 | printk("warning bad block %Lu\n", buf->start); |
db94535d | 742 | return 1; |
5f39d397 | 743 | } |
db94535d | 744 | #endif |
aa5d6bed | 745 | if (level == 0) |
123abc88 CM |
746 | return check_leaf(root, path, level); |
747 | return check_node(root, path, level); | |
aa5d6bed CM |
748 | } |
749 | ||
74123bd7 | 750 | /* |
5f39d397 CM |
751 | * search for key in the extent_buffer. The items start at offset p, |
752 | * and they are item_size apart. There are 'max' items in p. | |
753 | * | |
74123bd7 CM |
754 | * the slot in the array is returned via slot, and it points to |
755 | * the place where you would insert key if it is not found in | |
756 | * the array. | |
757 | * | |
758 | * slot may point to max if the key is bigger than all of the keys | |
759 | */ | |
e02119d5 CM |
760 | static noinline int generic_bin_search(struct extent_buffer *eb, |
761 | unsigned long p, | |
762 | int item_size, struct btrfs_key *key, | |
763 | int max, int *slot) | |
be0e5c09 CM |
764 | { |
765 | int low = 0; | |
766 | int high = max; | |
767 | int mid; | |
768 | int ret; | |
479965d6 | 769 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
770 | struct btrfs_disk_key unaligned; |
771 | unsigned long offset; | |
772 | char *map_token = NULL; | |
773 | char *kaddr = NULL; | |
774 | unsigned long map_start = 0; | |
775 | unsigned long map_len = 0; | |
479965d6 | 776 | int err; |
be0e5c09 CM |
777 | |
778 | while(low < high) { | |
779 | mid = (low + high) / 2; | |
5f39d397 CM |
780 | offset = p + mid * item_size; |
781 | ||
782 | if (!map_token || offset < map_start || | |
783 | (offset + sizeof(struct btrfs_disk_key)) > | |
784 | map_start + map_len) { | |
479965d6 | 785 | if (map_token) { |
5f39d397 | 786 | unmap_extent_buffer(eb, map_token, KM_USER0); |
479965d6 CM |
787 | map_token = NULL; |
788 | } | |
789 | err = map_extent_buffer(eb, offset, | |
790 | sizeof(struct btrfs_disk_key), | |
791 | &map_token, &kaddr, | |
792 | &map_start, &map_len, KM_USER0); | |
793 | ||
794 | if (!err) { | |
795 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
796 | map_start); | |
797 | } else { | |
798 | read_extent_buffer(eb, &unaligned, | |
799 | offset, sizeof(unaligned)); | |
800 | tmp = &unaligned; | |
801 | } | |
5f39d397 | 802 | |
5f39d397 CM |
803 | } else { |
804 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
805 | map_start); | |
806 | } | |
be0e5c09 CM |
807 | ret = comp_keys(tmp, key); |
808 | ||
809 | if (ret < 0) | |
810 | low = mid + 1; | |
811 | else if (ret > 0) | |
812 | high = mid; | |
813 | else { | |
814 | *slot = mid; | |
479965d6 CM |
815 | if (map_token) |
816 | unmap_extent_buffer(eb, map_token, KM_USER0); | |
be0e5c09 CM |
817 | return 0; |
818 | } | |
819 | } | |
820 | *slot = low; | |
5f39d397 CM |
821 | if (map_token) |
822 | unmap_extent_buffer(eb, map_token, KM_USER0); | |
be0e5c09 CM |
823 | return 1; |
824 | } | |
825 | ||
97571fd0 CM |
826 | /* |
827 | * simple bin_search frontend that does the right thing for | |
828 | * leaves vs nodes | |
829 | */ | |
5f39d397 CM |
830 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
831 | int level, int *slot) | |
be0e5c09 | 832 | { |
5f39d397 CM |
833 | if (level == 0) { |
834 | return generic_bin_search(eb, | |
835 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 836 | sizeof(struct btrfs_item), |
5f39d397 | 837 | key, btrfs_header_nritems(eb), |
7518a238 | 838 | slot); |
be0e5c09 | 839 | } else { |
5f39d397 CM |
840 | return generic_bin_search(eb, |
841 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 842 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 843 | key, btrfs_header_nritems(eb), |
7518a238 | 844 | slot); |
be0e5c09 CM |
845 | } |
846 | return -1; | |
847 | } | |
848 | ||
d352ac68 CM |
849 | /* given a node and slot number, this reads the blocks it points to. The |
850 | * extent buffer is returned with a reference taken (but unlocked). | |
851 | * NULL is returned on error. | |
852 | */ | |
e02119d5 | 853 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 854 | struct extent_buffer *parent, int slot) |
bb803951 | 855 | { |
ca7a79ad | 856 | int level = btrfs_header_level(parent); |
bb803951 CM |
857 | if (slot < 0) |
858 | return NULL; | |
5f39d397 | 859 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 860 | return NULL; |
ca7a79ad CM |
861 | |
862 | BUG_ON(level == 0); | |
863 | ||
db94535d | 864 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
865 | btrfs_level_size(root, level - 1), |
866 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
867 | } |
868 | ||
d352ac68 CM |
869 | /* |
870 | * node level balancing, used to make sure nodes are in proper order for | |
871 | * item deletion. We balance from the top down, so we have to make sure | |
872 | * that a deletion won't leave an node completely empty later on. | |
873 | */ | |
e02119d5 | 874 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
875 | struct btrfs_root *root, |
876 | struct btrfs_path *path, int level) | |
bb803951 | 877 | { |
5f39d397 CM |
878 | struct extent_buffer *right = NULL; |
879 | struct extent_buffer *mid; | |
880 | struct extent_buffer *left = NULL; | |
881 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
882 | int ret = 0; |
883 | int wret; | |
884 | int pslot; | |
bb803951 | 885 | int orig_slot = path->slots[level]; |
54aa1f4d | 886 | int err_on_enospc = 0; |
79f95c82 | 887 | u64 orig_ptr; |
bb803951 CM |
888 | |
889 | if (level == 0) | |
890 | return 0; | |
891 | ||
5f39d397 | 892 | mid = path->nodes[level]; |
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); |
925baedd | 914 | btrfs_tree_lock(child); |
bb803951 | 915 | BUG_ON(!child); |
65b51a00 | 916 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0); |
2f375ab9 Y |
917 | BUG_ON(ret); |
918 | ||
925baedd | 919 | spin_lock(&root->node_lock); |
bb803951 | 920 | root->node = child; |
925baedd CM |
921 | spin_unlock(&root->node_lock); |
922 | ||
31840ae1 ZY |
923 | ret = btrfs_update_extent_ref(trans, root, child->start, |
924 | mid->start, child->start, | |
925 | root->root_key.objectid, | |
3bb1a1bc | 926 | trans->transid, level - 1); |
31840ae1 ZY |
927 | BUG_ON(ret); |
928 | ||
0b86a832 | 929 | add_root_to_dirty_list(root); |
925baedd CM |
930 | btrfs_tree_unlock(child); |
931 | path->locks[level] = 0; | |
bb803951 | 932 | path->nodes[level] = NULL; |
5f39d397 | 933 | clean_tree_block(trans, root, mid); |
925baedd | 934 | btrfs_tree_unlock(mid); |
bb803951 | 935 | /* once for the path */ |
5f39d397 | 936 | free_extent_buffer(mid); |
7bb86316 | 937 | ret = btrfs_free_extent(trans, root, mid->start, mid->len, |
31840ae1 | 938 | mid->start, root->root_key.objectid, |
3bb1a1bc YZ |
939 | btrfs_header_generation(mid), |
940 | level, 1); | |
bb803951 | 941 | /* once for the root ptr */ |
5f39d397 | 942 | free_extent_buffer(mid); |
db94535d | 943 | return ret; |
bb803951 | 944 | } |
5f39d397 | 945 | if (btrfs_header_nritems(mid) > |
123abc88 | 946 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
947 | return 0; |
948 | ||
5f39d397 | 949 | if (btrfs_header_nritems(mid) < 2) |
54aa1f4d CM |
950 | err_on_enospc = 1; |
951 | ||
5f39d397 CM |
952 | left = read_node_slot(root, parent, pslot - 1); |
953 | if (left) { | |
925baedd | 954 | btrfs_tree_lock(left); |
5f39d397 | 955 | wret = btrfs_cow_block(trans, root, left, |
65b51a00 | 956 | parent, pslot - 1, &left, 0); |
54aa1f4d CM |
957 | if (wret) { |
958 | ret = wret; | |
959 | goto enospc; | |
960 | } | |
2cc58cf2 | 961 | } |
5f39d397 CM |
962 | right = read_node_slot(root, parent, pslot + 1); |
963 | if (right) { | |
925baedd | 964 | btrfs_tree_lock(right); |
5f39d397 | 965 | wret = btrfs_cow_block(trans, root, right, |
65b51a00 | 966 | parent, pslot + 1, &right, 0); |
2cc58cf2 CM |
967 | if (wret) { |
968 | ret = wret; | |
969 | goto enospc; | |
970 | } | |
971 | } | |
972 | ||
973 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
974 | if (left) { |
975 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 976 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
977 | if (wret < 0) |
978 | ret = wret; | |
5f39d397 | 979 | if (btrfs_header_nritems(mid) < 2) |
54aa1f4d | 980 | err_on_enospc = 1; |
bb803951 | 981 | } |
79f95c82 CM |
982 | |
983 | /* | |
984 | * then try to empty the right most buffer into the middle | |
985 | */ | |
5f39d397 | 986 | if (right) { |
971a1f66 | 987 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 988 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 989 | ret = wret; |
5f39d397 | 990 | if (btrfs_header_nritems(right) == 0) { |
db94535d | 991 | u64 bytenr = right->start; |
7bb86316 | 992 | u64 generation = btrfs_header_generation(parent); |
db94535d CM |
993 | u32 blocksize = right->len; |
994 | ||
5f39d397 | 995 | clean_tree_block(trans, root, right); |
925baedd | 996 | btrfs_tree_unlock(right); |
5f39d397 | 997 | free_extent_buffer(right); |
bb803951 | 998 | right = NULL; |
e089f05c CM |
999 | wret = del_ptr(trans, root, path, level + 1, pslot + |
1000 | 1); | |
bb803951 CM |
1001 | if (wret) |
1002 | ret = wret; | |
db94535d | 1003 | wret = btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 1004 | blocksize, parent->start, |
7bb86316 | 1005 | btrfs_header_owner(parent), |
3bb1a1bc | 1006 | generation, level, 1); |
bb803951 CM |
1007 | if (wret) |
1008 | ret = wret; | |
1009 | } else { | |
5f39d397 CM |
1010 | struct btrfs_disk_key right_key; |
1011 | btrfs_node_key(right, &right_key, 0); | |
1012 | btrfs_set_node_key(parent, &right_key, pslot + 1); | |
1013 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1014 | } |
1015 | } | |
5f39d397 | 1016 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1017 | /* |
1018 | * we're not allowed to leave a node with one item in the | |
1019 | * tree during a delete. A deletion from lower in the tree | |
1020 | * could try to delete the only pointer in this node. | |
1021 | * So, pull some keys from the left. | |
1022 | * There has to be a left pointer at this point because | |
1023 | * otherwise we would have pulled some pointers from the | |
1024 | * right | |
1025 | */ | |
5f39d397 CM |
1026 | BUG_ON(!left); |
1027 | wret = balance_node_right(trans, root, mid, left); | |
54aa1f4d | 1028 | if (wret < 0) { |
79f95c82 | 1029 | ret = wret; |
54aa1f4d CM |
1030 | goto enospc; |
1031 | } | |
bce4eae9 CM |
1032 | if (wret == 1) { |
1033 | wret = push_node_left(trans, root, left, mid, 1); | |
1034 | if (wret < 0) | |
1035 | ret = wret; | |
1036 | } | |
79f95c82 CM |
1037 | BUG_ON(wret == 1); |
1038 | } | |
5f39d397 | 1039 | if (btrfs_header_nritems(mid) == 0) { |
79f95c82 | 1040 | /* we've managed to empty the middle node, drop it */ |
7bb86316 | 1041 | u64 root_gen = btrfs_header_generation(parent); |
db94535d CM |
1042 | u64 bytenr = mid->start; |
1043 | u32 blocksize = mid->len; | |
925baedd | 1044 | |
5f39d397 | 1045 | clean_tree_block(trans, root, mid); |
925baedd | 1046 | btrfs_tree_unlock(mid); |
5f39d397 | 1047 | free_extent_buffer(mid); |
bb803951 | 1048 | mid = NULL; |
e089f05c | 1049 | wret = del_ptr(trans, root, path, level + 1, pslot); |
bb803951 CM |
1050 | if (wret) |
1051 | ret = wret; | |
7bb86316 | 1052 | wret = btrfs_free_extent(trans, root, bytenr, blocksize, |
31840ae1 | 1053 | parent->start, |
7bb86316 | 1054 | btrfs_header_owner(parent), |
3bb1a1bc | 1055 | root_gen, level, 1); |
bb803951 CM |
1056 | if (wret) |
1057 | ret = wret; | |
79f95c82 CM |
1058 | } else { |
1059 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1060 | struct btrfs_disk_key mid_key; |
1061 | btrfs_node_key(mid, &mid_key, 0); | |
1062 | btrfs_set_node_key(parent, &mid_key, pslot); | |
1063 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1064 | } |
bb803951 | 1065 | |
79f95c82 | 1066 | /* update the path */ |
5f39d397 CM |
1067 | if (left) { |
1068 | if (btrfs_header_nritems(left) > orig_slot) { | |
1069 | extent_buffer_get(left); | |
925baedd | 1070 | /* left was locked after cow */ |
5f39d397 | 1071 | path->nodes[level] = left; |
bb803951 CM |
1072 | path->slots[level + 1] -= 1; |
1073 | path->slots[level] = orig_slot; | |
925baedd CM |
1074 | if (mid) { |
1075 | btrfs_tree_unlock(mid); | |
5f39d397 | 1076 | free_extent_buffer(mid); |
925baedd | 1077 | } |
bb803951 | 1078 | } else { |
5f39d397 | 1079 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1080 | path->slots[level] = orig_slot; |
1081 | } | |
1082 | } | |
79f95c82 | 1083 | /* double check we haven't messed things up */ |
123abc88 | 1084 | check_block(root, path, level); |
e20d96d6 | 1085 | if (orig_ptr != |
5f39d397 | 1086 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1087 | BUG(); |
54aa1f4d | 1088 | enospc: |
925baedd CM |
1089 | if (right) { |
1090 | btrfs_tree_unlock(right); | |
5f39d397 | 1091 | free_extent_buffer(right); |
925baedd CM |
1092 | } |
1093 | if (left) { | |
1094 | if (path->nodes[level] != left) | |
1095 | btrfs_tree_unlock(left); | |
5f39d397 | 1096 | free_extent_buffer(left); |
925baedd | 1097 | } |
bb803951 CM |
1098 | return ret; |
1099 | } | |
1100 | ||
d352ac68 CM |
1101 | /* Node balancing for insertion. Here we only split or push nodes around |
1102 | * when they are completely full. This is also done top down, so we | |
1103 | * have to be pessimistic. | |
1104 | */ | |
98ed5174 CM |
1105 | static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans, |
1106 | struct btrfs_root *root, | |
1107 | struct btrfs_path *path, int level) | |
e66f709b | 1108 | { |
5f39d397 CM |
1109 | struct extent_buffer *right = NULL; |
1110 | struct extent_buffer *mid; | |
1111 | struct extent_buffer *left = NULL; | |
1112 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1113 | int ret = 0; |
1114 | int wret; | |
1115 | int pslot; | |
1116 | int orig_slot = path->slots[level]; | |
1117 | u64 orig_ptr; | |
1118 | ||
1119 | if (level == 0) | |
1120 | return 1; | |
1121 | ||
5f39d397 | 1122 | mid = path->nodes[level]; |
7bb86316 | 1123 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b CM |
1124 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
1125 | ||
1126 | if (level < BTRFS_MAX_LEVEL - 1) | |
5f39d397 | 1127 | parent = path->nodes[level + 1]; |
e66f709b CM |
1128 | pslot = path->slots[level + 1]; |
1129 | ||
5f39d397 | 1130 | if (!parent) |
e66f709b | 1131 | return 1; |
e66f709b | 1132 | |
5f39d397 | 1133 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1134 | |
1135 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1136 | if (left) { |
e66f709b | 1137 | u32 left_nr; |
925baedd CM |
1138 | |
1139 | btrfs_tree_lock(left); | |
5f39d397 | 1140 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1141 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1142 | wret = 1; | |
1143 | } else { | |
5f39d397 | 1144 | ret = btrfs_cow_block(trans, root, left, parent, |
65b51a00 | 1145 | pslot - 1, &left, 0); |
54aa1f4d CM |
1146 | if (ret) |
1147 | wret = 1; | |
1148 | else { | |
54aa1f4d | 1149 | wret = push_node_left(trans, root, |
971a1f66 | 1150 | left, mid, 0); |
54aa1f4d | 1151 | } |
33ade1f8 | 1152 | } |
e66f709b CM |
1153 | if (wret < 0) |
1154 | ret = wret; | |
1155 | if (wret == 0) { | |
5f39d397 | 1156 | struct btrfs_disk_key disk_key; |
e66f709b | 1157 | orig_slot += left_nr; |
5f39d397 CM |
1158 | btrfs_node_key(mid, &disk_key, 0); |
1159 | btrfs_set_node_key(parent, &disk_key, pslot); | |
1160 | btrfs_mark_buffer_dirty(parent); | |
1161 | if (btrfs_header_nritems(left) > orig_slot) { | |
1162 | path->nodes[level] = left; | |
e66f709b CM |
1163 | path->slots[level + 1] -= 1; |
1164 | path->slots[level] = orig_slot; | |
925baedd | 1165 | btrfs_tree_unlock(mid); |
5f39d397 | 1166 | free_extent_buffer(mid); |
e66f709b CM |
1167 | } else { |
1168 | orig_slot -= | |
5f39d397 | 1169 | btrfs_header_nritems(left); |
e66f709b | 1170 | path->slots[level] = orig_slot; |
925baedd | 1171 | btrfs_tree_unlock(left); |
5f39d397 | 1172 | free_extent_buffer(left); |
e66f709b | 1173 | } |
e66f709b CM |
1174 | return 0; |
1175 | } | |
925baedd | 1176 | btrfs_tree_unlock(left); |
5f39d397 | 1177 | free_extent_buffer(left); |
e66f709b | 1178 | } |
925baedd | 1179 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1180 | |
1181 | /* | |
1182 | * then try to empty the right most buffer into the middle | |
1183 | */ | |
5f39d397 | 1184 | if (right) { |
33ade1f8 | 1185 | u32 right_nr; |
925baedd | 1186 | btrfs_tree_lock(right); |
5f39d397 | 1187 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1188 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1189 | wret = 1; | |
1190 | } else { | |
5f39d397 CM |
1191 | ret = btrfs_cow_block(trans, root, right, |
1192 | parent, pslot + 1, | |
65b51a00 | 1193 | &right, 0); |
54aa1f4d CM |
1194 | if (ret) |
1195 | wret = 1; | |
1196 | else { | |
54aa1f4d | 1197 | wret = balance_node_right(trans, root, |
5f39d397 | 1198 | right, mid); |
54aa1f4d | 1199 | } |
33ade1f8 | 1200 | } |
e66f709b CM |
1201 | if (wret < 0) |
1202 | ret = wret; | |
1203 | if (wret == 0) { | |
5f39d397 CM |
1204 | struct btrfs_disk_key disk_key; |
1205 | ||
1206 | btrfs_node_key(right, &disk_key, 0); | |
1207 | btrfs_set_node_key(parent, &disk_key, pslot + 1); | |
1208 | btrfs_mark_buffer_dirty(parent); | |
1209 | ||
1210 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1211 | path->nodes[level] = right; | |
e66f709b CM |
1212 | path->slots[level + 1] += 1; |
1213 | path->slots[level] = orig_slot - | |
5f39d397 | 1214 | btrfs_header_nritems(mid); |
925baedd | 1215 | btrfs_tree_unlock(mid); |
5f39d397 | 1216 | free_extent_buffer(mid); |
e66f709b | 1217 | } else { |
925baedd | 1218 | btrfs_tree_unlock(right); |
5f39d397 | 1219 | free_extent_buffer(right); |
e66f709b | 1220 | } |
e66f709b CM |
1221 | return 0; |
1222 | } | |
925baedd | 1223 | btrfs_tree_unlock(right); |
5f39d397 | 1224 | free_extent_buffer(right); |
e66f709b | 1225 | } |
e66f709b CM |
1226 | return 1; |
1227 | } | |
1228 | ||
3c69faec | 1229 | /* |
d352ac68 CM |
1230 | * readahead one full node of leaves, finding things that are close |
1231 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1232 | */ |
e02119d5 CM |
1233 | static noinline void reada_for_search(struct btrfs_root *root, |
1234 | struct btrfs_path *path, | |
1235 | int level, int slot, u64 objectid) | |
3c69faec | 1236 | { |
5f39d397 | 1237 | struct extent_buffer *node; |
01f46658 | 1238 | struct btrfs_disk_key disk_key; |
3c69faec | 1239 | u32 nritems; |
3c69faec | 1240 | u64 search; |
6b80053d CM |
1241 | u64 lowest_read; |
1242 | u64 highest_read; | |
1243 | u64 nread = 0; | |
3c69faec | 1244 | int direction = path->reada; |
5f39d397 | 1245 | struct extent_buffer *eb; |
6b80053d CM |
1246 | u32 nr; |
1247 | u32 blocksize; | |
1248 | u32 nscan = 0; | |
db94535d | 1249 | |
a6b6e75e | 1250 | if (level != 1) |
6702ed49 CM |
1251 | return; |
1252 | ||
1253 | if (!path->nodes[level]) | |
3c69faec CM |
1254 | return; |
1255 | ||
5f39d397 | 1256 | node = path->nodes[level]; |
925baedd | 1257 | |
3c69faec | 1258 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1259 | blocksize = btrfs_level_size(root, level - 1); |
1260 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1261 | if (eb) { |
1262 | free_extent_buffer(eb); | |
3c69faec CM |
1263 | return; |
1264 | } | |
1265 | ||
6b80053d CM |
1266 | highest_read = search; |
1267 | lowest_read = search; | |
1268 | ||
5f39d397 | 1269 | nritems = btrfs_header_nritems(node); |
6b80053d | 1270 | nr = slot; |
3c69faec | 1271 | while(1) { |
6b80053d CM |
1272 | if (direction < 0) { |
1273 | if (nr == 0) | |
1274 | break; | |
1275 | nr--; | |
1276 | } else if (direction > 0) { | |
1277 | nr++; | |
1278 | if (nr >= nritems) | |
1279 | break; | |
3c69faec | 1280 | } |
01f46658 CM |
1281 | if (path->reada < 0 && objectid) { |
1282 | btrfs_node_key(node, &disk_key, nr); | |
1283 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1284 | break; | |
1285 | } | |
6b80053d CM |
1286 | search = btrfs_node_blockptr(node, nr); |
1287 | if ((search >= lowest_read && search <= highest_read) || | |
6f3577bd CM |
1288 | (search < lowest_read && lowest_read - search <= 16384) || |
1289 | (search > highest_read && search - highest_read <= 16384)) { | |
ca7a79ad CM |
1290 | readahead_tree_block(root, search, blocksize, |
1291 | btrfs_node_ptr_generation(node, nr)); | |
6b80053d CM |
1292 | nread += blocksize; |
1293 | } | |
1294 | nscan++; | |
6f3577bd | 1295 | if (path->reada < 2 && (nread > (64 * 1024) || nscan > 32)) |
6b80053d | 1296 | break; |
6f3577bd | 1297 | if(nread > (256 * 1024) || nscan > 128) |
6b80053d CM |
1298 | break; |
1299 | ||
1300 | if (search < lowest_read) | |
1301 | lowest_read = search; | |
1302 | if (search > highest_read) | |
1303 | highest_read = search; | |
3c69faec CM |
1304 | } |
1305 | } | |
925baedd | 1306 | |
d352ac68 CM |
1307 | /* |
1308 | * when we walk down the tree, it is usually safe to unlock the higher layers in | |
1309 | * the tree. The exceptions are when our path goes through slot 0, because operations | |
1310 | * on the tree might require changing key pointers higher up in the tree. | |
1311 | * | |
1312 | * callers might also have set path->keep_locks, which tells this code to | |
1313 | * keep the lock if the path points to the last slot in the block. This is | |
1314 | * part of walking through the tree, and selecting the next slot in the higher | |
1315 | * block. | |
1316 | * | |
1317 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. | |
1318 | * so if lowest_unlock is 1, level 0 won't be unlocked | |
1319 | */ | |
e02119d5 CM |
1320 | static noinline void unlock_up(struct btrfs_path *path, int level, |
1321 | int lowest_unlock) | |
925baedd CM |
1322 | { |
1323 | int i; | |
1324 | int skip_level = level; | |
051e1b9f | 1325 | int no_skips = 0; |
925baedd CM |
1326 | struct extent_buffer *t; |
1327 | ||
1328 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1329 | if (!path->nodes[i]) | |
1330 | break; | |
1331 | if (!path->locks[i]) | |
1332 | break; | |
051e1b9f | 1333 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1334 | skip_level = i + 1; |
1335 | continue; | |
1336 | } | |
051e1b9f | 1337 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1338 | u32 nritems; |
1339 | t = path->nodes[i]; | |
1340 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1341 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1342 | skip_level = i + 1; |
1343 | continue; | |
1344 | } | |
1345 | } | |
051e1b9f CM |
1346 | if (skip_level < i && i >= lowest_unlock) |
1347 | no_skips = 1; | |
1348 | ||
925baedd CM |
1349 | t = path->nodes[i]; |
1350 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
1351 | btrfs_tree_unlock(t); | |
1352 | path->locks[i] = 0; | |
1353 | } | |
1354 | } | |
1355 | } | |
1356 | ||
74123bd7 CM |
1357 | /* |
1358 | * look for key in the tree. path is filled in with nodes along the way | |
1359 | * if key is found, we return zero and you can find the item in the leaf | |
1360 | * level of the path (level 0) | |
1361 | * | |
1362 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
1363 | * be inserted, and 1 is returned. If there are other errors during the |
1364 | * search a negative error number is returned. | |
97571fd0 CM |
1365 | * |
1366 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
1367 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
1368 | * possible) | |
74123bd7 | 1369 | */ |
e089f05c CM |
1370 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
1371 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
1372 | ins_len, int cow) | |
be0e5c09 | 1373 | { |
5f39d397 | 1374 | struct extent_buffer *b; |
051e1b9f | 1375 | struct extent_buffer *tmp; |
be0e5c09 CM |
1376 | int slot; |
1377 | int ret; | |
1378 | int level; | |
3c69faec | 1379 | int should_reada = p->reada; |
925baedd | 1380 | int lowest_unlock = 1; |
594a24eb | 1381 | int blocksize; |
9f3a7427 | 1382 | u8 lowest_level = 0; |
594a24eb CM |
1383 | u64 blocknr; |
1384 | u64 gen; | |
65b51a00 | 1385 | struct btrfs_key prealloc_block; |
9f3a7427 | 1386 | |
6702ed49 | 1387 | lowest_level = p->lowest_level; |
323ac95b | 1388 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1389 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 1390 | |
925baedd CM |
1391 | if (ins_len < 0) |
1392 | lowest_unlock = 2; | |
65b51a00 CM |
1393 | |
1394 | prealloc_block.objectid = 0; | |
1395 | ||
bb803951 | 1396 | again: |
5cd57b2c CM |
1397 | if (p->skip_locking) |
1398 | b = btrfs_root_node(root); | |
1399 | else | |
1400 | b = btrfs_lock_root_node(root); | |
925baedd | 1401 | |
eb60ceac | 1402 | while (b) { |
5f39d397 | 1403 | level = btrfs_header_level(b); |
65b51a00 CM |
1404 | |
1405 | /* | |
1406 | * setup the path here so we can release it under lock | |
1407 | * contention with the cow code | |
1408 | */ | |
1409 | p->nodes[level] = b; | |
1410 | if (!p->skip_locking) | |
1411 | p->locks[level] = 1; | |
1412 | ||
02217ed2 CM |
1413 | if (cow) { |
1414 | int wret; | |
65b51a00 CM |
1415 | |
1416 | /* is a cow on this block not required */ | |
1417 | spin_lock(&root->fs_info->hash_lock); | |
1418 | if (btrfs_header_generation(b) == trans->transid && | |
5b21f2ed | 1419 | btrfs_header_owner(b) == root->root_key.objectid && |
65b51a00 CM |
1420 | !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) { |
1421 | spin_unlock(&root->fs_info->hash_lock); | |
1422 | goto cow_done; | |
1423 | } | |
1424 | spin_unlock(&root->fs_info->hash_lock); | |
1425 | ||
1426 | /* ok, we have to cow, is our old prealloc the right | |
1427 | * size? | |
1428 | */ | |
1429 | if (prealloc_block.objectid && | |
1430 | prealloc_block.offset != b->len) { | |
1431 | btrfs_free_reserved_extent(root, | |
1432 | prealloc_block.objectid, | |
1433 | prealloc_block.offset); | |
1434 | prealloc_block.objectid = 0; | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * for higher level blocks, try not to allocate blocks | |
1439 | * with the block and the parent locks held. | |
1440 | */ | |
1441 | if (level > 1 && !prealloc_block.objectid && | |
1442 | btrfs_path_lock_waiting(p, level)) { | |
1443 | u32 size = b->len; | |
1444 | u64 hint = b->start; | |
1445 | ||
1446 | btrfs_release_path(root, p); | |
1447 | ret = btrfs_reserve_extent(trans, root, | |
1448 | size, size, 0, | |
1449 | hint, (u64)-1, | |
1450 | &prealloc_block, 0); | |
1451 | BUG_ON(ret); | |
1452 | goto again; | |
1453 | } | |
1454 | ||
e20d96d6 CM |
1455 | wret = btrfs_cow_block(trans, root, b, |
1456 | p->nodes[level + 1], | |
1457 | p->slots[level + 1], | |
65b51a00 CM |
1458 | &b, prealloc_block.objectid); |
1459 | prealloc_block.objectid = 0; | |
54aa1f4d | 1460 | if (wret) { |
5f39d397 | 1461 | free_extent_buffer(b); |
65b51a00 CM |
1462 | ret = wret; |
1463 | goto done; | |
54aa1f4d | 1464 | } |
02217ed2 | 1465 | } |
65b51a00 | 1466 | cow_done: |
02217ed2 | 1467 | BUG_ON(!cow && ins_len); |
5f39d397 | 1468 | if (level != btrfs_header_level(b)) |
2c90e5d6 | 1469 | WARN_ON(1); |
5f39d397 | 1470 | level = btrfs_header_level(b); |
65b51a00 | 1471 | |
eb60ceac | 1472 | p->nodes[level] = b; |
5cd57b2c CM |
1473 | if (!p->skip_locking) |
1474 | p->locks[level] = 1; | |
65b51a00 | 1475 | |
123abc88 | 1476 | ret = check_block(root, p, level); |
65b51a00 CM |
1477 | if (ret) { |
1478 | ret = -1; | |
1479 | goto done; | |
1480 | } | |
925baedd | 1481 | |
5f39d397 CM |
1482 | ret = bin_search(b, key, level, &slot); |
1483 | if (level != 0) { | |
be0e5c09 CM |
1484 | if (ret && slot > 0) |
1485 | slot -= 1; | |
1486 | p->slots[level] = slot; | |
5f39d397 | 1487 | if (ins_len > 0 && btrfs_header_nritems(b) >= |
1514794e | 1488 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { |
e089f05c | 1489 | int sret = split_node(trans, root, p, level); |
5c680ed6 | 1490 | BUG_ON(sret > 0); |
65b51a00 CM |
1491 | if (sret) { |
1492 | ret = sret; | |
1493 | goto done; | |
1494 | } | |
5c680ed6 | 1495 | b = p->nodes[level]; |
5c680ed6 | 1496 | slot = p->slots[level]; |
bb803951 | 1497 | } else if (ins_len < 0) { |
e089f05c CM |
1498 | int sret = balance_level(trans, root, p, |
1499 | level); | |
65b51a00 CM |
1500 | if (sret) { |
1501 | ret = sret; | |
1502 | goto done; | |
1503 | } | |
bb803951 | 1504 | b = p->nodes[level]; |
f510cfec CM |
1505 | if (!b) { |
1506 | btrfs_release_path(NULL, p); | |
bb803951 | 1507 | goto again; |
f510cfec | 1508 | } |
bb803951 | 1509 | slot = p->slots[level]; |
5f39d397 | 1510 | BUG_ON(btrfs_header_nritems(b) == 1); |
5c680ed6 | 1511 | } |
f9efa9c7 CM |
1512 | unlock_up(p, level, lowest_unlock); |
1513 | ||
9f3a7427 | 1514 | /* this is only true while dropping a snapshot */ |
925baedd | 1515 | if (level == lowest_level) { |
5b21f2ed ZY |
1516 | ret = 0; |
1517 | goto done; | |
925baedd | 1518 | } |
ca7a79ad | 1519 | |
594a24eb CM |
1520 | blocknr = btrfs_node_blockptr(b, slot); |
1521 | gen = btrfs_node_ptr_generation(b, slot); | |
1522 | blocksize = btrfs_level_size(root, level - 1); | |
1523 | ||
1524 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
1525 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
051e1b9f CM |
1526 | b = tmp; |
1527 | } else { | |
1528 | /* | |
1529 | * reduce lock contention at high levels | |
1530 | * of the btree by dropping locks before | |
1531 | * we read. | |
1532 | */ | |
1533 | if (level > 1) { | |
1534 | btrfs_release_path(NULL, p); | |
594a24eb CM |
1535 | if (tmp) |
1536 | free_extent_buffer(tmp); | |
f9efa9c7 CM |
1537 | if (should_reada) |
1538 | reada_for_search(root, p, | |
1539 | level, slot, | |
1540 | key->objectid); | |
1541 | ||
594a24eb CM |
1542 | tmp = read_tree_block(root, blocknr, |
1543 | blocksize, gen); | |
051e1b9f CM |
1544 | if (tmp) |
1545 | free_extent_buffer(tmp); | |
1546 | goto again; | |
1547 | } else { | |
a74a4b97 CM |
1548 | if (tmp) |
1549 | free_extent_buffer(tmp); | |
f9efa9c7 CM |
1550 | if (should_reada) |
1551 | reada_for_search(root, p, | |
1552 | level, slot, | |
1553 | key->objectid); | |
051e1b9f CM |
1554 | b = read_node_slot(root, b, slot); |
1555 | } | |
1556 | } | |
5cd57b2c CM |
1557 | if (!p->skip_locking) |
1558 | btrfs_tree_lock(b); | |
be0e5c09 CM |
1559 | } else { |
1560 | p->slots[level] = slot; | |
5f39d397 | 1561 | if (ins_len > 0 && btrfs_leaf_free_space(root, b) < |
0783fcfc | 1562 | sizeof(struct btrfs_item) + ins_len) { |
d4dbff95 | 1563 | int sret = split_leaf(trans, root, key, |
cc0c5538 | 1564 | p, ins_len, ret == 0); |
5c680ed6 | 1565 | BUG_ON(sret > 0); |
65b51a00 CM |
1566 | if (sret) { |
1567 | ret = sret; | |
1568 | goto done; | |
1569 | } | |
5c680ed6 | 1570 | } |
925baedd | 1571 | unlock_up(p, level, lowest_unlock); |
65b51a00 | 1572 | goto done; |
be0e5c09 CM |
1573 | } |
1574 | } | |
65b51a00 CM |
1575 | ret = 1; |
1576 | done: | |
1577 | if (prealloc_block.objectid) { | |
1578 | btrfs_free_reserved_extent(root, | |
1579 | prealloc_block.objectid, | |
1580 | prealloc_block.offset); | |
1581 | } | |
1582 | ||
1583 | return ret; | |
be0e5c09 CM |
1584 | } |
1585 | ||
1a40e23b ZY |
1586 | int btrfs_merge_path(struct btrfs_trans_handle *trans, |
1587 | struct btrfs_root *root, | |
1588 | struct btrfs_key *node_keys, | |
1589 | u64 *nodes, int lowest_level) | |
1590 | { | |
1591 | struct extent_buffer *eb; | |
1592 | struct extent_buffer *parent; | |
1593 | struct btrfs_key key; | |
1594 | u64 bytenr; | |
1595 | u64 generation; | |
1596 | u32 blocksize; | |
1597 | int level; | |
1598 | int slot; | |
1599 | int key_match; | |
1600 | int ret; | |
1601 | ||
1602 | eb = btrfs_lock_root_node(root); | |
1603 | ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0); | |
1604 | BUG_ON(ret); | |
1605 | ||
1606 | parent = eb; | |
1607 | while (1) { | |
1608 | level = btrfs_header_level(parent); | |
1609 | if (level == 0 || level <= lowest_level) | |
1610 | break; | |
1611 | ||
1612 | ret = bin_search(parent, &node_keys[lowest_level], level, | |
1613 | &slot); | |
1614 | if (ret && slot > 0) | |
1615 | slot--; | |
1616 | ||
1617 | bytenr = btrfs_node_blockptr(parent, slot); | |
1618 | if (nodes[level - 1] == bytenr) | |
1619 | break; | |
1620 | ||
1621 | blocksize = btrfs_level_size(root, level - 1); | |
1622 | generation = btrfs_node_ptr_generation(parent, slot); | |
1623 | btrfs_node_key_to_cpu(eb, &key, slot); | |
1624 | key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key)); | |
1625 | ||
f82d02d9 YZ |
1626 | if (generation == trans->transid) { |
1627 | eb = read_tree_block(root, bytenr, blocksize, | |
1628 | generation); | |
1629 | btrfs_tree_lock(eb); | |
1630 | } | |
1631 | ||
1a40e23b ZY |
1632 | /* |
1633 | * if node keys match and node pointer hasn't been modified | |
1634 | * in the running transaction, we can merge the path. for | |
f82d02d9 YZ |
1635 | * blocks owened by reloc trees, the node pointer check is |
1636 | * skipped, this is because these blocks are fully controlled | |
1637 | * by the space balance code, no one else can modify them. | |
1a40e23b ZY |
1638 | */ |
1639 | if (!nodes[level - 1] || !key_match || | |
1640 | (generation == trans->transid && | |
f82d02d9 YZ |
1641 | btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) { |
1642 | if (level == 1 || level == lowest_level + 1) { | |
1643 | if (generation == trans->transid) { | |
1644 | btrfs_tree_unlock(eb); | |
1645 | free_extent_buffer(eb); | |
1646 | } | |
1a40e23b | 1647 | break; |
f82d02d9 | 1648 | } |
1a40e23b | 1649 | |
f82d02d9 YZ |
1650 | if (generation != trans->transid) { |
1651 | eb = read_tree_block(root, bytenr, blocksize, | |
1652 | generation); | |
1653 | btrfs_tree_lock(eb); | |
1654 | } | |
1a40e23b ZY |
1655 | |
1656 | ret = btrfs_cow_block(trans, root, eb, parent, slot, | |
1657 | &eb, 0); | |
1658 | BUG_ON(ret); | |
1659 | ||
f82d02d9 YZ |
1660 | if (root->root_key.objectid == |
1661 | BTRFS_TREE_RELOC_OBJECTID) { | |
1662 | if (!nodes[level - 1]) { | |
1663 | nodes[level - 1] = eb->start; | |
1664 | memcpy(&node_keys[level - 1], &key, | |
1665 | sizeof(node_keys[0])); | |
1666 | } else { | |
1667 | WARN_ON(1); | |
1668 | } | |
1669 | } | |
1670 | ||
1a40e23b ZY |
1671 | btrfs_tree_unlock(parent); |
1672 | free_extent_buffer(parent); | |
1673 | parent = eb; | |
1674 | continue; | |
1675 | } | |
1676 | ||
1a40e23b ZY |
1677 | btrfs_set_node_blockptr(parent, slot, nodes[level - 1]); |
1678 | btrfs_set_node_ptr_generation(parent, slot, trans->transid); | |
1679 | btrfs_mark_buffer_dirty(parent); | |
1680 | ||
1681 | ret = btrfs_inc_extent_ref(trans, root, | |
1682 | nodes[level - 1], | |
1683 | blocksize, parent->start, | |
1684 | btrfs_header_owner(parent), | |
1685 | btrfs_header_generation(parent), | |
3bb1a1bc | 1686 | level - 1); |
1a40e23b | 1687 | BUG_ON(ret); |
1a40e23b | 1688 | |
f82d02d9 YZ |
1689 | /* |
1690 | * If the block was created in the running transaction, | |
1691 | * it's possible this is the last reference to it, so we | |
1692 | * should drop the subtree. | |
1693 | */ | |
1a40e23b | 1694 | if (generation == trans->transid) { |
f82d02d9 YZ |
1695 | ret = btrfs_drop_subtree(trans, root, eb, parent); |
1696 | BUG_ON(ret); | |
1a40e23b ZY |
1697 | btrfs_tree_unlock(eb); |
1698 | free_extent_buffer(eb); | |
f82d02d9 YZ |
1699 | } else { |
1700 | ret = btrfs_free_extent(trans, root, bytenr, | |
1701 | blocksize, parent->start, | |
1702 | btrfs_header_owner(parent), | |
1703 | btrfs_header_generation(parent), | |
1704 | level - 1, 1); | |
1705 | BUG_ON(ret); | |
1a40e23b ZY |
1706 | } |
1707 | break; | |
1708 | } | |
1709 | btrfs_tree_unlock(parent); | |
1710 | free_extent_buffer(parent); | |
1711 | return 0; | |
1712 | } | |
1713 | ||
74123bd7 CM |
1714 | /* |
1715 | * adjust the pointers going up the tree, starting at level | |
1716 | * making sure the right key of each node is points to 'key'. | |
1717 | * This is used after shifting pointers to the left, so it stops | |
1718 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
1719 | * higher levels | |
aa5d6bed CM |
1720 | * |
1721 | * If this fails to write a tree block, it returns -1, but continues | |
1722 | * fixing up the blocks in ram so the tree is consistent. | |
74123bd7 | 1723 | */ |
5f39d397 CM |
1724 | static int fixup_low_keys(struct btrfs_trans_handle *trans, |
1725 | struct btrfs_root *root, struct btrfs_path *path, | |
1726 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
1727 | { |
1728 | int i; | |
aa5d6bed | 1729 | int ret = 0; |
5f39d397 CM |
1730 | struct extent_buffer *t; |
1731 | ||
234b63a0 | 1732 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 1733 | int tslot = path->slots[i]; |
eb60ceac | 1734 | if (!path->nodes[i]) |
be0e5c09 | 1735 | break; |
5f39d397 CM |
1736 | t = path->nodes[i]; |
1737 | btrfs_set_node_key(t, key, tslot); | |
d6025579 | 1738 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
1739 | if (tslot != 0) |
1740 | break; | |
1741 | } | |
aa5d6bed | 1742 | return ret; |
be0e5c09 CM |
1743 | } |
1744 | ||
31840ae1 ZY |
1745 | /* |
1746 | * update item key. | |
1747 | * | |
1748 | * This function isn't completely safe. It's the caller's responsibility | |
1749 | * that the new key won't break the order | |
1750 | */ | |
1751 | int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, | |
1752 | struct btrfs_root *root, struct btrfs_path *path, | |
1753 | struct btrfs_key *new_key) | |
1754 | { | |
1755 | struct btrfs_disk_key disk_key; | |
1756 | struct extent_buffer *eb; | |
1757 | int slot; | |
1758 | ||
1759 | eb = path->nodes[0]; | |
1760 | slot = path->slots[0]; | |
1761 | if (slot > 0) { | |
1762 | btrfs_item_key(eb, &disk_key, slot - 1); | |
1763 | if (comp_keys(&disk_key, new_key) >= 0) | |
1764 | return -1; | |
1765 | } | |
1766 | if (slot < btrfs_header_nritems(eb) - 1) { | |
1767 | btrfs_item_key(eb, &disk_key, slot + 1); | |
1768 | if (comp_keys(&disk_key, new_key) <= 0) | |
1769 | return -1; | |
1770 | } | |
1771 | ||
1772 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
1773 | btrfs_set_item_key(eb, &disk_key, slot); | |
1774 | btrfs_mark_buffer_dirty(eb); | |
1775 | if (slot == 0) | |
1776 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
1777 | return 0; | |
1778 | } | |
1779 | ||
74123bd7 CM |
1780 | /* |
1781 | * try to push data from one node into the next node left in the | |
79f95c82 | 1782 | * tree. |
aa5d6bed CM |
1783 | * |
1784 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
1785 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 1786 | */ |
98ed5174 CM |
1787 | static int push_node_left(struct btrfs_trans_handle *trans, |
1788 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 1789 | struct extent_buffer *src, int empty) |
be0e5c09 | 1790 | { |
be0e5c09 | 1791 | int push_items = 0; |
bb803951 CM |
1792 | int src_nritems; |
1793 | int dst_nritems; | |
aa5d6bed | 1794 | int ret = 0; |
be0e5c09 | 1795 | |
5f39d397 CM |
1796 | src_nritems = btrfs_header_nritems(src); |
1797 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 1798 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
1799 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
1800 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 1801 | |
bce4eae9 | 1802 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
1803 | return 1; |
1804 | ||
eb60ceac | 1805 | if (push_items <= 0) { |
be0e5c09 | 1806 | return 1; |
eb60ceac | 1807 | } |
be0e5c09 | 1808 | |
bce4eae9 | 1809 | if (empty) { |
971a1f66 | 1810 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
1811 | if (push_items < src_nritems) { |
1812 | /* leave at least 8 pointers in the node if | |
1813 | * we aren't going to empty it | |
1814 | */ | |
1815 | if (src_nritems - push_items < 8) { | |
1816 | if (push_items <= 8) | |
1817 | return 1; | |
1818 | push_items -= 8; | |
1819 | } | |
1820 | } | |
1821 | } else | |
1822 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 1823 | |
5f39d397 CM |
1824 | copy_extent_buffer(dst, src, |
1825 | btrfs_node_key_ptr_offset(dst_nritems), | |
1826 | btrfs_node_key_ptr_offset(0), | |
1827 | push_items * sizeof(struct btrfs_key_ptr)); | |
1828 | ||
bb803951 | 1829 | if (push_items < src_nritems) { |
5f39d397 CM |
1830 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
1831 | btrfs_node_key_ptr_offset(push_items), | |
1832 | (src_nritems - push_items) * | |
1833 | sizeof(struct btrfs_key_ptr)); | |
1834 | } | |
1835 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
1836 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
1837 | btrfs_mark_buffer_dirty(src); | |
1838 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 ZY |
1839 | |
1840 | ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items); | |
1841 | BUG_ON(ret); | |
1842 | ||
79f95c82 CM |
1843 | return ret; |
1844 | } | |
1845 | ||
1846 | /* | |
1847 | * try to push data from one node into the next node right in the | |
1848 | * tree. | |
1849 | * | |
1850 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
1851 | * error, and > 0 if there was no room in the right hand block. | |
1852 | * | |
1853 | * this will only push up to 1/2 the contents of the left node over | |
1854 | */ | |
5f39d397 CM |
1855 | static int balance_node_right(struct btrfs_trans_handle *trans, |
1856 | struct btrfs_root *root, | |
1857 | struct extent_buffer *dst, | |
1858 | struct extent_buffer *src) | |
79f95c82 | 1859 | { |
79f95c82 CM |
1860 | int push_items = 0; |
1861 | int max_push; | |
1862 | int src_nritems; | |
1863 | int dst_nritems; | |
1864 | int ret = 0; | |
79f95c82 | 1865 | |
7bb86316 CM |
1866 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
1867 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
1868 | ||
5f39d397 CM |
1869 | src_nritems = btrfs_header_nritems(src); |
1870 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 1871 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
bce4eae9 | 1872 | if (push_items <= 0) { |
79f95c82 | 1873 | return 1; |
bce4eae9 CM |
1874 | } |
1875 | ||
1876 | if (src_nritems < 4) { | |
1877 | return 1; | |
1878 | } | |
79f95c82 CM |
1879 | |
1880 | max_push = src_nritems / 2 + 1; | |
1881 | /* don't try to empty the node */ | |
bce4eae9 | 1882 | if (max_push >= src_nritems) { |
79f95c82 | 1883 | return 1; |
bce4eae9 | 1884 | } |
252c38f0 | 1885 | |
79f95c82 CM |
1886 | if (max_push < push_items) |
1887 | push_items = max_push; | |
1888 | ||
5f39d397 CM |
1889 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
1890 | btrfs_node_key_ptr_offset(0), | |
1891 | (dst_nritems) * | |
1892 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 1893 | |
5f39d397 CM |
1894 | copy_extent_buffer(dst, src, |
1895 | btrfs_node_key_ptr_offset(0), | |
1896 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
1897 | push_items * sizeof(struct btrfs_key_ptr)); | |
79f95c82 | 1898 | |
5f39d397 CM |
1899 | btrfs_set_header_nritems(src, src_nritems - push_items); |
1900 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 1901 | |
5f39d397 CM |
1902 | btrfs_mark_buffer_dirty(src); |
1903 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 ZY |
1904 | |
1905 | ret = btrfs_update_ref(trans, root, src, dst, 0, push_items); | |
1906 | BUG_ON(ret); | |
1907 | ||
aa5d6bed | 1908 | return ret; |
be0e5c09 CM |
1909 | } |
1910 | ||
97571fd0 CM |
1911 | /* |
1912 | * helper function to insert a new root level in the tree. | |
1913 | * A new node is allocated, and a single item is inserted to | |
1914 | * point to the existing root | |
aa5d6bed CM |
1915 | * |
1916 | * returns zero on success or < 0 on failure. | |
97571fd0 | 1917 | */ |
98ed5174 | 1918 | static int noinline insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1919 | struct btrfs_root *root, |
1920 | struct btrfs_path *path, int level) | |
5c680ed6 | 1921 | { |
7bb86316 | 1922 | u64 lower_gen; |
5f39d397 CM |
1923 | struct extent_buffer *lower; |
1924 | struct extent_buffer *c; | |
925baedd | 1925 | struct extent_buffer *old; |
5f39d397 | 1926 | struct btrfs_disk_key lower_key; |
31840ae1 | 1927 | int ret; |
5c680ed6 CM |
1928 | |
1929 | BUG_ON(path->nodes[level]); | |
1930 | BUG_ON(path->nodes[level-1] != root->node); | |
1931 | ||
7bb86316 CM |
1932 | lower = path->nodes[level-1]; |
1933 | if (level == 1) | |
1934 | btrfs_item_key(lower, &lower_key, 0); | |
1935 | else | |
1936 | btrfs_node_key(lower, &lower_key, 0); | |
1937 | ||
31840ae1 ZY |
1938 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
1939 | root->root_key.objectid, trans->transid, | |
ad3d81ba | 1940 | level, root->node->start, 0); |
5f39d397 CM |
1941 | if (IS_ERR(c)) |
1942 | return PTR_ERR(c); | |
925baedd | 1943 | |
5f39d397 CM |
1944 | memset_extent_buffer(c, 0, 0, root->nodesize); |
1945 | btrfs_set_header_nritems(c, 1); | |
1946 | btrfs_set_header_level(c, level); | |
db94535d | 1947 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 CM |
1948 | btrfs_set_header_generation(c, trans->transid); |
1949 | btrfs_set_header_owner(c, root->root_key.objectid); | |
5f39d397 CM |
1950 | |
1951 | write_extent_buffer(c, root->fs_info->fsid, | |
1952 | (unsigned long)btrfs_header_fsid(c), | |
1953 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
1954 | |
1955 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
1956 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
1957 | BTRFS_UUID_SIZE); | |
1958 | ||
5f39d397 | 1959 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 1960 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 1961 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 1962 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
1963 | |
1964 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 1965 | |
5f39d397 | 1966 | btrfs_mark_buffer_dirty(c); |
d5719762 | 1967 | |
925baedd CM |
1968 | spin_lock(&root->node_lock); |
1969 | old = root->node; | |
5f39d397 | 1970 | root->node = c; |
925baedd CM |
1971 | spin_unlock(&root->node_lock); |
1972 | ||
31840ae1 ZY |
1973 | ret = btrfs_update_extent_ref(trans, root, lower->start, |
1974 | lower->start, c->start, | |
1975 | root->root_key.objectid, | |
3bb1a1bc | 1976 | trans->transid, level - 1); |
31840ae1 ZY |
1977 | BUG_ON(ret); |
1978 | ||
925baedd CM |
1979 | /* the super has an extra ref to root->node */ |
1980 | free_extent_buffer(old); | |
1981 | ||
0b86a832 | 1982 | add_root_to_dirty_list(root); |
5f39d397 CM |
1983 | extent_buffer_get(c); |
1984 | path->nodes[level] = c; | |
925baedd | 1985 | path->locks[level] = 1; |
5c680ed6 CM |
1986 | path->slots[level] = 0; |
1987 | return 0; | |
1988 | } | |
1989 | ||
74123bd7 CM |
1990 | /* |
1991 | * worker function to insert a single pointer in a node. | |
1992 | * the node should have enough room for the pointer already | |
97571fd0 | 1993 | * |
74123bd7 CM |
1994 | * slot and level indicate where you want the key to go, and |
1995 | * blocknr is the block the key points to. | |
aa5d6bed CM |
1996 | * |
1997 | * returns zero on success and < 0 on any error | |
74123bd7 | 1998 | */ |
e089f05c CM |
1999 | static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root |
2000 | *root, struct btrfs_path *path, struct btrfs_disk_key | |
db94535d | 2001 | *key, u64 bytenr, int slot, int level) |
74123bd7 | 2002 | { |
5f39d397 | 2003 | struct extent_buffer *lower; |
74123bd7 | 2004 | int nritems; |
5c680ed6 CM |
2005 | |
2006 | BUG_ON(!path->nodes[level]); | |
5f39d397 CM |
2007 | lower = path->nodes[level]; |
2008 | nritems = btrfs_header_nritems(lower); | |
74123bd7 CM |
2009 | if (slot > nritems) |
2010 | BUG(); | |
123abc88 | 2011 | if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root)) |
74123bd7 CM |
2012 | BUG(); |
2013 | if (slot != nritems) { | |
5f39d397 CM |
2014 | memmove_extent_buffer(lower, |
2015 | btrfs_node_key_ptr_offset(slot + 1), | |
2016 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2017 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2018 | } |
5f39d397 | 2019 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2020 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2021 | WARN_ON(trans->transid == 0); |
2022 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2023 | btrfs_set_header_nritems(lower, nritems + 1); |
2024 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2025 | return 0; |
2026 | } | |
2027 | ||
97571fd0 CM |
2028 | /* |
2029 | * split the node at the specified level in path in two. | |
2030 | * The path is corrected to point to the appropriate node after the split | |
2031 | * | |
2032 | * Before splitting this tries to make some room in the node by pushing | |
2033 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2034 | * |
2035 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2036 | */ |
e02119d5 CM |
2037 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2038 | struct btrfs_root *root, | |
2039 | struct btrfs_path *path, int level) | |
be0e5c09 | 2040 | { |
5f39d397 CM |
2041 | struct extent_buffer *c; |
2042 | struct extent_buffer *split; | |
2043 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2044 | int mid; |
5c680ed6 | 2045 | int ret; |
aa5d6bed | 2046 | int wret; |
7518a238 | 2047 | u32 c_nritems; |
eb60ceac | 2048 | |
5f39d397 | 2049 | c = path->nodes[level]; |
7bb86316 | 2050 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2051 | if (c == root->node) { |
5c680ed6 | 2052 | /* trying to split the root, lets make a new one */ |
e089f05c | 2053 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2054 | if (ret) |
2055 | return ret; | |
e66f709b CM |
2056 | } else { |
2057 | ret = push_nodes_for_insert(trans, root, path, level); | |
5f39d397 CM |
2058 | c = path->nodes[level]; |
2059 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2060 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2061 | return 0; |
54aa1f4d CM |
2062 | if (ret < 0) |
2063 | return ret; | |
be0e5c09 | 2064 | } |
e66f709b | 2065 | |
5f39d397 | 2066 | c_nritems = btrfs_header_nritems(c); |
7bb86316 | 2067 | |
925baedd | 2068 | split = btrfs_alloc_free_block(trans, root, root->nodesize, |
31840ae1 ZY |
2069 | path->nodes[level + 1]->start, |
2070 | root->root_key.objectid, | |
2071 | trans->transid, level, c->start, 0); | |
5f39d397 CM |
2072 | if (IS_ERR(split)) |
2073 | return PTR_ERR(split); | |
2074 | ||
2075 | btrfs_set_header_flags(split, btrfs_header_flags(c)); | |
2076 | btrfs_set_header_level(split, btrfs_header_level(c)); | |
db94535d | 2077 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 CM |
2078 | btrfs_set_header_generation(split, trans->transid); |
2079 | btrfs_set_header_owner(split, root->root_key.objectid); | |
63b10fc4 | 2080 | btrfs_set_header_flags(split, 0); |
5f39d397 CM |
2081 | write_extent_buffer(split, root->fs_info->fsid, |
2082 | (unsigned long)btrfs_header_fsid(split), | |
2083 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2084 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2085 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2086 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2087 | |
7518a238 | 2088 | mid = (c_nritems + 1) / 2; |
5f39d397 CM |
2089 | |
2090 | copy_extent_buffer(split, c, | |
2091 | btrfs_node_key_ptr_offset(0), | |
2092 | btrfs_node_key_ptr_offset(mid), | |
2093 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2094 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2095 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2096 | ret = 0; |
2097 | ||
5f39d397 CM |
2098 | btrfs_mark_buffer_dirty(c); |
2099 | btrfs_mark_buffer_dirty(split); | |
2100 | ||
2101 | btrfs_node_key(split, &disk_key, 0); | |
db94535d | 2102 | wret = insert_ptr(trans, root, path, &disk_key, split->start, |
5f39d397 | 2103 | path->slots[level + 1] + 1, |
123abc88 | 2104 | level + 1); |
aa5d6bed CM |
2105 | if (wret) |
2106 | ret = wret; | |
2107 | ||
31840ae1 ZY |
2108 | ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid); |
2109 | BUG_ON(ret); | |
2110 | ||
5de08d7d | 2111 | if (path->slots[level] >= mid) { |
5c680ed6 | 2112 | path->slots[level] -= mid; |
925baedd | 2113 | btrfs_tree_unlock(c); |
5f39d397 CM |
2114 | free_extent_buffer(c); |
2115 | path->nodes[level] = split; | |
5c680ed6 CM |
2116 | path->slots[level + 1] += 1; |
2117 | } else { | |
925baedd | 2118 | btrfs_tree_unlock(split); |
5f39d397 | 2119 | free_extent_buffer(split); |
be0e5c09 | 2120 | } |
aa5d6bed | 2121 | return ret; |
be0e5c09 CM |
2122 | } |
2123 | ||
74123bd7 CM |
2124 | /* |
2125 | * how many bytes are required to store the items in a leaf. start | |
2126 | * and nr indicate which items in the leaf to check. This totals up the | |
2127 | * space used both by the item structs and the item data | |
2128 | */ | |
5f39d397 | 2129 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2130 | { |
2131 | int data_len; | |
5f39d397 | 2132 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2133 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2134 | |
2135 | if (!nr) | |
2136 | return 0; | |
5f39d397 CM |
2137 | data_len = btrfs_item_end_nr(l, start); |
2138 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2139 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2140 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2141 | return data_len; |
2142 | } | |
2143 | ||
d4dbff95 CM |
2144 | /* |
2145 | * The space between the end of the leaf items and | |
2146 | * the start of the leaf data. IOW, how much room | |
2147 | * the leaf has left for both items and data | |
2148 | */ | |
e02119d5 CM |
2149 | int noinline btrfs_leaf_free_space(struct btrfs_root *root, |
2150 | struct extent_buffer *leaf) | |
d4dbff95 | 2151 | { |
5f39d397 CM |
2152 | int nritems = btrfs_header_nritems(leaf); |
2153 | int ret; | |
2154 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2155 | if (ret < 0) { | |
2156 | printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n", | |
ae2f5411 | 2157 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2158 | leaf_space_used(leaf, 0, nritems), nritems); |
2159 | } | |
2160 | return ret; | |
d4dbff95 CM |
2161 | } |
2162 | ||
00ec4c51 CM |
2163 | /* |
2164 | * push some data in the path leaf to the right, trying to free up at | |
2165 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
aa5d6bed CM |
2166 | * |
2167 | * returns 1 if the push failed because the other node didn't have enough | |
2168 | * room, 0 if everything worked out and < 0 if there were major errors. | |
00ec4c51 | 2169 | */ |
e089f05c | 2170 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root |
34a38218 CM |
2171 | *root, struct btrfs_path *path, int data_size, |
2172 | int empty) | |
00ec4c51 | 2173 | { |
5f39d397 CM |
2174 | struct extent_buffer *left = path->nodes[0]; |
2175 | struct extent_buffer *right; | |
2176 | struct extent_buffer *upper; | |
2177 | struct btrfs_disk_key disk_key; | |
00ec4c51 | 2178 | int slot; |
34a38218 | 2179 | u32 i; |
00ec4c51 CM |
2180 | int free_space; |
2181 | int push_space = 0; | |
2182 | int push_items = 0; | |
0783fcfc | 2183 | struct btrfs_item *item; |
7518a238 | 2184 | u32 left_nritems; |
34a38218 | 2185 | u32 nr; |
7518a238 | 2186 | u32 right_nritems; |
5f39d397 | 2187 | u32 data_end; |
db94535d | 2188 | u32 this_item_size; |
54aa1f4d | 2189 | int ret; |
00ec4c51 CM |
2190 | |
2191 | slot = path->slots[1]; | |
2192 | if (!path->nodes[1]) { | |
2193 | return 1; | |
2194 | } | |
2195 | upper = path->nodes[1]; | |
5f39d397 | 2196 | if (slot >= btrfs_header_nritems(upper) - 1) |
00ec4c51 | 2197 | return 1; |
5f39d397 | 2198 | |
a2135011 CM |
2199 | WARN_ON(!btrfs_tree_locked(path->nodes[1])); |
2200 | ||
ca7a79ad | 2201 | right = read_node_slot(root, upper, slot + 1); |
925baedd | 2202 | btrfs_tree_lock(right); |
123abc88 | 2203 | free_space = btrfs_leaf_free_space(root, right); |
925baedd CM |
2204 | if (free_space < data_size + sizeof(struct btrfs_item)) |
2205 | goto out_unlock; | |
5f39d397 | 2206 | |
02217ed2 | 2207 | /* cow and double check */ |
5f39d397 | 2208 | ret = btrfs_cow_block(trans, root, right, upper, |
65b51a00 | 2209 | slot + 1, &right, 0); |
925baedd CM |
2210 | if (ret) |
2211 | goto out_unlock; | |
2212 | ||
123abc88 | 2213 | free_space = btrfs_leaf_free_space(root, right); |
925baedd CM |
2214 | if (free_space < data_size + sizeof(struct btrfs_item)) |
2215 | goto out_unlock; | |
02217ed2 | 2216 | |
5f39d397 | 2217 | left_nritems = btrfs_header_nritems(left); |
925baedd CM |
2218 | if (left_nritems == 0) |
2219 | goto out_unlock; | |
5f39d397 | 2220 | |
34a38218 CM |
2221 | if (empty) |
2222 | nr = 0; | |
2223 | else | |
2224 | nr = 1; | |
2225 | ||
31840ae1 ZY |
2226 | if (path->slots[0] >= left_nritems) |
2227 | push_space += data_size + sizeof(*item); | |
2228 | ||
34a38218 CM |
2229 | i = left_nritems - 1; |
2230 | while (i >= nr) { | |
5f39d397 | 2231 | item = btrfs_item_nr(left, i); |
db94535d | 2232 | |
31840ae1 ZY |
2233 | if (!empty && push_items > 0) { |
2234 | if (path->slots[0] > i) | |
2235 | break; | |
2236 | if (path->slots[0] == i) { | |
2237 | int space = btrfs_leaf_free_space(root, left); | |
2238 | if (space + push_space * 2 > free_space) | |
2239 | break; | |
2240 | } | |
2241 | } | |
2242 | ||
00ec4c51 CM |
2243 | if (path->slots[0] == i) |
2244 | push_space += data_size + sizeof(*item); | |
db94535d CM |
2245 | |
2246 | if (!left->map_token) { | |
2247 | map_extent_buffer(left, (unsigned long)item, | |
2248 | sizeof(struct btrfs_item), | |
2249 | &left->map_token, &left->kaddr, | |
2250 | &left->map_start, &left->map_len, | |
2251 | KM_USER1); | |
2252 | } | |
2253 | ||
2254 | this_item_size = btrfs_item_size(left, item); | |
2255 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2256 | break; |
31840ae1 | 2257 | |
00ec4c51 | 2258 | push_items++; |
db94535d | 2259 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2260 | if (i == 0) |
2261 | break; | |
2262 | i--; | |
db94535d CM |
2263 | } |
2264 | if (left->map_token) { | |
2265 | unmap_extent_buffer(left, left->map_token, KM_USER1); | |
2266 | left->map_token = NULL; | |
00ec4c51 | 2267 | } |
5f39d397 | 2268 | |
925baedd CM |
2269 | if (push_items == 0) |
2270 | goto out_unlock; | |
5f39d397 | 2271 | |
34a38218 | 2272 | if (!empty && push_items == left_nritems) |
a429e513 | 2273 | WARN_ON(1); |
5f39d397 | 2274 | |
00ec4c51 | 2275 | /* push left to right */ |
5f39d397 | 2276 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2277 | |
5f39d397 | 2278 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2279 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2280 | |
00ec4c51 | 2281 | /* make room in the right data area */ |
5f39d397 CM |
2282 | data_end = leaf_data_end(root, right); |
2283 | memmove_extent_buffer(right, | |
2284 | btrfs_leaf_data(right) + data_end - push_space, | |
2285 | btrfs_leaf_data(right) + data_end, | |
2286 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2287 | ||
00ec4c51 | 2288 | /* copy from the left data area */ |
5f39d397 | 2289 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2290 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2291 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2292 | push_space); | |
5f39d397 CM |
2293 | |
2294 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2295 | btrfs_item_nr_offset(0), | |
2296 | right_nritems * sizeof(struct btrfs_item)); | |
2297 | ||
00ec4c51 | 2298 | /* copy the items from left to right */ |
5f39d397 CM |
2299 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2300 | btrfs_item_nr_offset(left_nritems - push_items), | |
2301 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2302 | |
2303 | /* update the item pointers */ | |
7518a238 | 2304 | right_nritems += push_items; |
5f39d397 | 2305 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2306 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2307 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2308 | item = btrfs_item_nr(right, i); |
db94535d CM |
2309 | if (!right->map_token) { |
2310 | map_extent_buffer(right, (unsigned long)item, | |
2311 | sizeof(struct btrfs_item), | |
2312 | &right->map_token, &right->kaddr, | |
2313 | &right->map_start, &right->map_len, | |
2314 | KM_USER1); | |
2315 | } | |
2316 | push_space -= btrfs_item_size(right, item); | |
2317 | btrfs_set_item_offset(right, item, push_space); | |
2318 | } | |
2319 | ||
2320 | if (right->map_token) { | |
2321 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2322 | right->map_token = NULL; | |
00ec4c51 | 2323 | } |
7518a238 | 2324 | left_nritems -= push_items; |
5f39d397 | 2325 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2326 | |
34a38218 CM |
2327 | if (left_nritems) |
2328 | btrfs_mark_buffer_dirty(left); | |
5f39d397 | 2329 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2330 | |
31840ae1 ZY |
2331 | ret = btrfs_update_ref(trans, root, left, right, 0, push_items); |
2332 | BUG_ON(ret); | |
2333 | ||
5f39d397 CM |
2334 | btrfs_item_key(right, &disk_key, 0); |
2335 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2336 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2337 | |
00ec4c51 | 2338 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2339 | if (path->slots[0] >= left_nritems) { |
2340 | path->slots[0] -= left_nritems; | |
925baedd CM |
2341 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2342 | clean_tree_block(trans, root, path->nodes[0]); | |
2343 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2344 | free_extent_buffer(path->nodes[0]); |
2345 | path->nodes[0] = right; | |
00ec4c51 CM |
2346 | path->slots[1] += 1; |
2347 | } else { | |
925baedd | 2348 | btrfs_tree_unlock(right); |
5f39d397 | 2349 | free_extent_buffer(right); |
00ec4c51 CM |
2350 | } |
2351 | return 0; | |
925baedd CM |
2352 | |
2353 | out_unlock: | |
2354 | btrfs_tree_unlock(right); | |
2355 | free_extent_buffer(right); | |
2356 | return 1; | |
00ec4c51 | 2357 | } |
925baedd | 2358 | |
74123bd7 CM |
2359 | /* |
2360 | * push some data in the path leaf to the left, trying to free up at | |
2361 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2362 | */ | |
e089f05c | 2363 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root |
34a38218 CM |
2364 | *root, struct btrfs_path *path, int data_size, |
2365 | int empty) | |
be0e5c09 | 2366 | { |
5f39d397 CM |
2367 | struct btrfs_disk_key disk_key; |
2368 | struct extent_buffer *right = path->nodes[0]; | |
2369 | struct extent_buffer *left; | |
be0e5c09 CM |
2370 | int slot; |
2371 | int i; | |
2372 | int free_space; | |
2373 | int push_space = 0; | |
2374 | int push_items = 0; | |
0783fcfc | 2375 | struct btrfs_item *item; |
7518a238 | 2376 | u32 old_left_nritems; |
5f39d397 | 2377 | u32 right_nritems; |
34a38218 | 2378 | u32 nr; |
aa5d6bed CM |
2379 | int ret = 0; |
2380 | int wret; | |
db94535d CM |
2381 | u32 this_item_size; |
2382 | u32 old_left_item_size; | |
be0e5c09 CM |
2383 | |
2384 | slot = path->slots[1]; | |
5f39d397 | 2385 | if (slot == 0) |
be0e5c09 | 2386 | return 1; |
5f39d397 | 2387 | if (!path->nodes[1]) |
be0e5c09 | 2388 | return 1; |
5f39d397 | 2389 | |
3685f791 CM |
2390 | right_nritems = btrfs_header_nritems(right); |
2391 | if (right_nritems == 0) { | |
2392 | return 1; | |
2393 | } | |
2394 | ||
a2135011 CM |
2395 | WARN_ON(!btrfs_tree_locked(path->nodes[1])); |
2396 | ||
ca7a79ad | 2397 | left = read_node_slot(root, path->nodes[1], slot - 1); |
925baedd | 2398 | btrfs_tree_lock(left); |
123abc88 | 2399 | free_space = btrfs_leaf_free_space(root, left); |
0783fcfc | 2400 | if (free_space < data_size + sizeof(struct btrfs_item)) { |
925baedd CM |
2401 | ret = 1; |
2402 | goto out; | |
be0e5c09 | 2403 | } |
02217ed2 CM |
2404 | |
2405 | /* cow and double check */ | |
5f39d397 | 2406 | ret = btrfs_cow_block(trans, root, left, |
65b51a00 | 2407 | path->nodes[1], slot - 1, &left, 0); |
54aa1f4d CM |
2408 | if (ret) { |
2409 | /* we hit -ENOSPC, but it isn't fatal here */ | |
925baedd CM |
2410 | ret = 1; |
2411 | goto out; | |
54aa1f4d | 2412 | } |
3685f791 | 2413 | |
123abc88 | 2414 | free_space = btrfs_leaf_free_space(root, left); |
0783fcfc | 2415 | if (free_space < data_size + sizeof(struct btrfs_item)) { |
925baedd CM |
2416 | ret = 1; |
2417 | goto out; | |
02217ed2 CM |
2418 | } |
2419 | ||
34a38218 CM |
2420 | if (empty) |
2421 | nr = right_nritems; | |
2422 | else | |
2423 | nr = right_nritems - 1; | |
2424 | ||
2425 | for (i = 0; i < nr; i++) { | |
5f39d397 | 2426 | item = btrfs_item_nr(right, i); |
db94535d CM |
2427 | if (!right->map_token) { |
2428 | map_extent_buffer(right, (unsigned long)item, | |
2429 | sizeof(struct btrfs_item), | |
2430 | &right->map_token, &right->kaddr, | |
2431 | &right->map_start, &right->map_len, | |
2432 | KM_USER1); | |
2433 | } | |
2434 | ||
31840ae1 ZY |
2435 | if (!empty && push_items > 0) { |
2436 | if (path->slots[0] < i) | |
2437 | break; | |
2438 | if (path->slots[0] == i) { | |
2439 | int space = btrfs_leaf_free_space(root, right); | |
2440 | if (space + push_space * 2 > free_space) | |
2441 | break; | |
2442 | } | |
2443 | } | |
2444 | ||
be0e5c09 CM |
2445 | if (path->slots[0] == i) |
2446 | push_space += data_size + sizeof(*item); | |
db94535d CM |
2447 | |
2448 | this_item_size = btrfs_item_size(right, item); | |
2449 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 2450 | break; |
db94535d | 2451 | |
be0e5c09 | 2452 | push_items++; |
db94535d CM |
2453 | push_space += this_item_size + sizeof(*item); |
2454 | } | |
2455 | ||
2456 | if (right->map_token) { | |
2457 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2458 | right->map_token = NULL; | |
be0e5c09 | 2459 | } |
db94535d | 2460 | |
be0e5c09 | 2461 | if (push_items == 0) { |
925baedd CM |
2462 | ret = 1; |
2463 | goto out; | |
be0e5c09 | 2464 | } |
34a38218 | 2465 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 2466 | WARN_ON(1); |
5f39d397 | 2467 | |
be0e5c09 | 2468 | /* push data from right to left */ |
5f39d397 CM |
2469 | copy_extent_buffer(left, right, |
2470 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
2471 | btrfs_item_nr_offset(0), | |
2472 | push_items * sizeof(struct btrfs_item)); | |
2473 | ||
123abc88 | 2474 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
5f39d397 CM |
2475 | btrfs_item_offset_nr(right, push_items -1); |
2476 | ||
2477 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
2478 | leaf_data_end(root, left) - push_space, |
2479 | btrfs_leaf_data(right) + | |
5f39d397 | 2480 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 2481 | push_space); |
5f39d397 | 2482 | old_left_nritems = btrfs_header_nritems(left); |
eb60ceac CM |
2483 | BUG_ON(old_left_nritems < 0); |
2484 | ||
db94535d | 2485 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 2486 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 2487 | u32 ioff; |
db94535d | 2488 | |
5f39d397 | 2489 | item = btrfs_item_nr(left, i); |
db94535d CM |
2490 | if (!left->map_token) { |
2491 | map_extent_buffer(left, (unsigned long)item, | |
2492 | sizeof(struct btrfs_item), | |
2493 | &left->map_token, &left->kaddr, | |
2494 | &left->map_start, &left->map_len, | |
2495 | KM_USER1); | |
2496 | } | |
2497 | ||
5f39d397 CM |
2498 | ioff = btrfs_item_offset(left, item); |
2499 | btrfs_set_item_offset(left, item, | |
db94535d | 2500 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size)); |
be0e5c09 | 2501 | } |
5f39d397 | 2502 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
db94535d CM |
2503 | if (left->map_token) { |
2504 | unmap_extent_buffer(left, left->map_token, KM_USER1); | |
2505 | left->map_token = NULL; | |
2506 | } | |
be0e5c09 CM |
2507 | |
2508 | /* fixup right node */ | |
34a38218 CM |
2509 | if (push_items > right_nritems) { |
2510 | printk("push items %d nr %u\n", push_items, right_nritems); | |
2511 | WARN_ON(1); | |
2512 | } | |
2513 | ||
2514 | if (push_items < right_nritems) { | |
2515 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2516 | leaf_data_end(root, right); | |
2517 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
2518 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
2519 | btrfs_leaf_data(right) + | |
2520 | leaf_data_end(root, right), push_space); | |
2521 | ||
2522 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
2523 | btrfs_item_nr_offset(push_items), |
2524 | (btrfs_header_nritems(right) - push_items) * | |
2525 | sizeof(struct btrfs_item)); | |
34a38218 | 2526 | } |
eef1c494 Y |
2527 | right_nritems -= push_items; |
2528 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 2529 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
2530 | for (i = 0; i < right_nritems; i++) { |
2531 | item = btrfs_item_nr(right, i); | |
db94535d CM |
2532 | |
2533 | if (!right->map_token) { | |
2534 | map_extent_buffer(right, (unsigned long)item, | |
2535 | sizeof(struct btrfs_item), | |
2536 | &right->map_token, &right->kaddr, | |
2537 | &right->map_start, &right->map_len, | |
2538 | KM_USER1); | |
2539 | } | |
2540 | ||
2541 | push_space = push_space - btrfs_item_size(right, item); | |
2542 | btrfs_set_item_offset(right, item, push_space); | |
2543 | } | |
2544 | if (right->map_token) { | |
2545 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2546 | right->map_token = NULL; | |
be0e5c09 | 2547 | } |
eb60ceac | 2548 | |
5f39d397 | 2549 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
2550 | if (right_nritems) |
2551 | btrfs_mark_buffer_dirty(right); | |
098f59c2 | 2552 | |
31840ae1 ZY |
2553 | ret = btrfs_update_ref(trans, root, right, left, |
2554 | old_left_nritems, push_items); | |
2555 | BUG_ON(ret); | |
2556 | ||
5f39d397 CM |
2557 | btrfs_item_key(right, &disk_key, 0); |
2558 | wret = fixup_low_keys(trans, root, path, &disk_key, 1); | |
aa5d6bed CM |
2559 | if (wret) |
2560 | ret = wret; | |
be0e5c09 CM |
2561 | |
2562 | /* then fixup the leaf pointer in the path */ | |
2563 | if (path->slots[0] < push_items) { | |
2564 | path->slots[0] += old_left_nritems; | |
925baedd CM |
2565 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2566 | clean_tree_block(trans, root, path->nodes[0]); | |
2567 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2568 | free_extent_buffer(path->nodes[0]); |
2569 | path->nodes[0] = left; | |
be0e5c09 CM |
2570 | path->slots[1] -= 1; |
2571 | } else { | |
925baedd | 2572 | btrfs_tree_unlock(left); |
5f39d397 | 2573 | free_extent_buffer(left); |
be0e5c09 CM |
2574 | path->slots[0] -= push_items; |
2575 | } | |
eb60ceac | 2576 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 2577 | return ret; |
925baedd CM |
2578 | out: |
2579 | btrfs_tree_unlock(left); | |
2580 | free_extent_buffer(left); | |
2581 | return ret; | |
be0e5c09 CM |
2582 | } |
2583 | ||
74123bd7 CM |
2584 | /* |
2585 | * split the path's leaf in two, making sure there is at least data_size | |
2586 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
2587 | * |
2588 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 2589 | */ |
e02119d5 CM |
2590 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
2591 | struct btrfs_root *root, | |
2592 | struct btrfs_key *ins_key, | |
2593 | struct btrfs_path *path, int data_size, | |
2594 | int extend) | |
be0e5c09 | 2595 | { |
5f39d397 | 2596 | struct extent_buffer *l; |
7518a238 | 2597 | u32 nritems; |
eb60ceac CM |
2598 | int mid; |
2599 | int slot; | |
5f39d397 | 2600 | struct extent_buffer *right; |
0783fcfc | 2601 | int space_needed = data_size + sizeof(struct btrfs_item); |
be0e5c09 CM |
2602 | int data_copy_size; |
2603 | int rt_data_off; | |
2604 | int i; | |
d4dbff95 | 2605 | int ret = 0; |
aa5d6bed | 2606 | int wret; |
cc0c5538 CM |
2607 | int double_split; |
2608 | int num_doubles = 0; | |
d4dbff95 | 2609 | struct btrfs_disk_key disk_key; |
aa5d6bed | 2610 | |
cc0c5538 CM |
2611 | if (extend) |
2612 | space_needed = data_size; | |
2613 | ||
40689478 | 2614 | /* first try to make some room by pushing left and right */ |
3685f791 | 2615 | if (ins_key->type != BTRFS_DIR_ITEM_KEY) { |
34a38218 | 2616 | wret = push_leaf_right(trans, root, path, data_size, 0); |
3685f791 | 2617 | if (wret < 0) { |
eaee50e8 | 2618 | return wret; |
3685f791 CM |
2619 | } |
2620 | if (wret) { | |
34a38218 | 2621 | wret = push_leaf_left(trans, root, path, data_size, 0); |
3685f791 CM |
2622 | if (wret < 0) |
2623 | return wret; | |
2624 | } | |
2625 | l = path->nodes[0]; | |
aa5d6bed | 2626 | |
3685f791 | 2627 | /* did the pushes work? */ |
cc0c5538 | 2628 | if (btrfs_leaf_free_space(root, l) >= space_needed) |
3685f791 | 2629 | return 0; |
3326d1b0 | 2630 | } |
aa5d6bed | 2631 | |
5c680ed6 | 2632 | if (!path->nodes[1]) { |
e089f05c | 2633 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
2634 | if (ret) |
2635 | return ret; | |
2636 | } | |
cc0c5538 CM |
2637 | again: |
2638 | double_split = 0; | |
2639 | l = path->nodes[0]; | |
eb60ceac | 2640 | slot = path->slots[0]; |
5f39d397 | 2641 | nritems = btrfs_header_nritems(l); |
eb60ceac | 2642 | mid = (nritems + 1)/ 2; |
54aa1f4d | 2643 | |
925baedd | 2644 | right = btrfs_alloc_free_block(trans, root, root->leafsize, |
31840ae1 ZY |
2645 | path->nodes[1]->start, |
2646 | root->root_key.objectid, | |
2647 | trans->transid, 0, l->start, 0); | |
cea9e445 CM |
2648 | if (IS_ERR(right)) { |
2649 | BUG_ON(1); | |
5f39d397 | 2650 | return PTR_ERR(right); |
cea9e445 | 2651 | } |
5f39d397 CM |
2652 | |
2653 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 2654 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 CM |
2655 | btrfs_set_header_generation(right, trans->transid); |
2656 | btrfs_set_header_owner(right, root->root_key.objectid); | |
2657 | btrfs_set_header_level(right, 0); | |
2658 | write_extent_buffer(right, root->fs_info->fsid, | |
2659 | (unsigned long)btrfs_header_fsid(right), | |
2660 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2661 | |
2662 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
2663 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
2664 | BTRFS_UUID_SIZE); | |
d4dbff95 CM |
2665 | if (mid <= slot) { |
2666 | if (nritems == 1 || | |
2667 | leaf_space_used(l, mid, nritems - mid) + space_needed > | |
2668 | BTRFS_LEAF_DATA_SIZE(root)) { | |
2669 | if (slot >= nritems) { | |
2670 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
5f39d397 | 2671 | btrfs_set_header_nritems(right, 0); |
d4dbff95 | 2672 | wret = insert_ptr(trans, root, path, |
db94535d | 2673 | &disk_key, right->start, |
d4dbff95 CM |
2674 | path->slots[1] + 1, 1); |
2675 | if (wret) | |
2676 | ret = wret; | |
925baedd CM |
2677 | |
2678 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2679 | free_extent_buffer(path->nodes[0]); |
2680 | path->nodes[0] = right; | |
d4dbff95 CM |
2681 | path->slots[0] = 0; |
2682 | path->slots[1] += 1; | |
0ef8b242 | 2683 | btrfs_mark_buffer_dirty(right); |
d4dbff95 CM |
2684 | return ret; |
2685 | } | |
2686 | mid = slot; | |
3326d1b0 CM |
2687 | if (mid != nritems && |
2688 | leaf_space_used(l, mid, nritems - mid) + | |
2689 | space_needed > BTRFS_LEAF_DATA_SIZE(root)) { | |
2690 | double_split = 1; | |
2691 | } | |
d4dbff95 CM |
2692 | } |
2693 | } else { | |
2694 | if (leaf_space_used(l, 0, mid + 1) + space_needed > | |
2695 | BTRFS_LEAF_DATA_SIZE(root)) { | |
cc0c5538 | 2696 | if (!extend && slot == 0) { |
d4dbff95 | 2697 | btrfs_cpu_key_to_disk(&disk_key, ins_key); |
5f39d397 | 2698 | btrfs_set_header_nritems(right, 0); |
d4dbff95 CM |
2699 | wret = insert_ptr(trans, root, path, |
2700 | &disk_key, | |
db94535d | 2701 | right->start, |
098f59c2 | 2702 | path->slots[1], 1); |
d4dbff95 CM |
2703 | if (wret) |
2704 | ret = wret; | |
925baedd | 2705 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2706 | free_extent_buffer(path->nodes[0]); |
2707 | path->nodes[0] = right; | |
d4dbff95 | 2708 | path->slots[0] = 0; |
a429e513 CM |
2709 | if (path->slots[1] == 0) { |
2710 | wret = fixup_low_keys(trans, root, | |
2711 | path, &disk_key, 1); | |
2712 | if (wret) | |
2713 | ret = wret; | |
2714 | } | |
0ef8b242 | 2715 | btrfs_mark_buffer_dirty(right); |
d4dbff95 | 2716 | return ret; |
cc0c5538 CM |
2717 | } else if (extend && slot == 0) { |
2718 | mid = 1; | |
2719 | } else { | |
2720 | mid = slot; | |
2721 | if (mid != nritems && | |
2722 | leaf_space_used(l, mid, nritems - mid) + | |
2723 | space_needed > BTRFS_LEAF_DATA_SIZE(root)) { | |
2724 | double_split = 1; | |
2725 | } | |
5ee78ac7 | 2726 | } |
d4dbff95 CM |
2727 | } |
2728 | } | |
5f39d397 CM |
2729 | nritems = nritems - mid; |
2730 | btrfs_set_header_nritems(right, nritems); | |
2731 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
2732 | ||
2733 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
2734 | btrfs_item_nr_offset(mid), | |
2735 | nritems * sizeof(struct btrfs_item)); | |
2736 | ||
2737 | copy_extent_buffer(right, l, | |
d6025579 CM |
2738 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - |
2739 | data_copy_size, btrfs_leaf_data(l) + | |
2740 | leaf_data_end(root, l), data_copy_size); | |
5f39d397 | 2741 | |
123abc88 | 2742 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - |
5f39d397 | 2743 | btrfs_item_end_nr(l, mid); |
74123bd7 | 2744 | |
5f39d397 CM |
2745 | for (i = 0; i < nritems; i++) { |
2746 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
db94535d CM |
2747 | u32 ioff; |
2748 | ||
2749 | if (!right->map_token) { | |
2750 | map_extent_buffer(right, (unsigned long)item, | |
2751 | sizeof(struct btrfs_item), | |
2752 | &right->map_token, &right->kaddr, | |
2753 | &right->map_start, &right->map_len, | |
2754 | KM_USER1); | |
2755 | } | |
2756 | ||
2757 | ioff = btrfs_item_offset(right, item); | |
5f39d397 | 2758 | btrfs_set_item_offset(right, item, ioff + rt_data_off); |
0783fcfc | 2759 | } |
74123bd7 | 2760 | |
db94535d CM |
2761 | if (right->map_token) { |
2762 | unmap_extent_buffer(right, right->map_token, KM_USER1); | |
2763 | right->map_token = NULL; | |
2764 | } | |
2765 | ||
5f39d397 | 2766 | btrfs_set_header_nritems(l, mid); |
aa5d6bed | 2767 | ret = 0; |
5f39d397 | 2768 | btrfs_item_key(right, &disk_key, 0); |
db94535d CM |
2769 | wret = insert_ptr(trans, root, path, &disk_key, right->start, |
2770 | path->slots[1] + 1, 1); | |
aa5d6bed CM |
2771 | if (wret) |
2772 | ret = wret; | |
5f39d397 CM |
2773 | |
2774 | btrfs_mark_buffer_dirty(right); | |
2775 | btrfs_mark_buffer_dirty(l); | |
eb60ceac | 2776 | BUG_ON(path->slots[0] != slot); |
5f39d397 | 2777 | |
31840ae1 ZY |
2778 | ret = btrfs_update_ref(trans, root, l, right, 0, nritems); |
2779 | BUG_ON(ret); | |
2780 | ||
be0e5c09 | 2781 | if (mid <= slot) { |
925baedd | 2782 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2783 | free_extent_buffer(path->nodes[0]); |
2784 | path->nodes[0] = right; | |
be0e5c09 CM |
2785 | path->slots[0] -= mid; |
2786 | path->slots[1] += 1; | |
925baedd CM |
2787 | } else { |
2788 | btrfs_tree_unlock(right); | |
5f39d397 | 2789 | free_extent_buffer(right); |
925baedd | 2790 | } |
5f39d397 | 2791 | |
eb60ceac | 2792 | BUG_ON(path->slots[0] < 0); |
d4dbff95 | 2793 | |
cc0c5538 CM |
2794 | if (double_split) { |
2795 | BUG_ON(num_doubles != 0); | |
2796 | num_doubles++; | |
2797 | goto again; | |
a429e513 | 2798 | } |
be0e5c09 CM |
2799 | return ret; |
2800 | } | |
2801 | ||
d352ac68 CM |
2802 | /* |
2803 | * make the item pointed to by the path smaller. new_size indicates | |
2804 | * how small to make it, and from_end tells us if we just chop bytes | |
2805 | * off the end of the item or if we shift the item to chop bytes off | |
2806 | * the front. | |
2807 | */ | |
b18c6685 CM |
2808 | int btrfs_truncate_item(struct btrfs_trans_handle *trans, |
2809 | struct btrfs_root *root, | |
2810 | struct btrfs_path *path, | |
179e29e4 | 2811 | u32 new_size, int from_end) |
b18c6685 CM |
2812 | { |
2813 | int ret = 0; | |
2814 | int slot; | |
2815 | int slot_orig; | |
5f39d397 CM |
2816 | struct extent_buffer *leaf; |
2817 | struct btrfs_item *item; | |
b18c6685 CM |
2818 | u32 nritems; |
2819 | unsigned int data_end; | |
2820 | unsigned int old_data_start; | |
2821 | unsigned int old_size; | |
2822 | unsigned int size_diff; | |
2823 | int i; | |
2824 | ||
2825 | slot_orig = path->slots[0]; | |
5f39d397 | 2826 | leaf = path->nodes[0]; |
179e29e4 CM |
2827 | slot = path->slots[0]; |
2828 | ||
2829 | old_size = btrfs_item_size_nr(leaf, slot); | |
2830 | if (old_size == new_size) | |
2831 | return 0; | |
b18c6685 | 2832 | |
5f39d397 | 2833 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
2834 | data_end = leaf_data_end(root, leaf); |
2835 | ||
5f39d397 | 2836 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 2837 | |
b18c6685 CM |
2838 | size_diff = old_size - new_size; |
2839 | ||
2840 | BUG_ON(slot < 0); | |
2841 | BUG_ON(slot >= nritems); | |
2842 | ||
2843 | /* | |
2844 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
2845 | */ | |
2846 | /* first correct the data pointers */ | |
2847 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
2848 | u32 ioff; |
2849 | item = btrfs_item_nr(leaf, i); | |
db94535d CM |
2850 | |
2851 | if (!leaf->map_token) { | |
2852 | map_extent_buffer(leaf, (unsigned long)item, | |
2853 | sizeof(struct btrfs_item), | |
2854 | &leaf->map_token, &leaf->kaddr, | |
2855 | &leaf->map_start, &leaf->map_len, | |
2856 | KM_USER1); | |
2857 | } | |
2858 | ||
5f39d397 CM |
2859 | ioff = btrfs_item_offset(leaf, item); |
2860 | btrfs_set_item_offset(leaf, item, ioff + size_diff); | |
b18c6685 | 2861 | } |
db94535d CM |
2862 | |
2863 | if (leaf->map_token) { | |
2864 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
2865 | leaf->map_token = NULL; | |
2866 | } | |
2867 | ||
b18c6685 | 2868 | /* shift the data */ |
179e29e4 CM |
2869 | if (from_end) { |
2870 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
2871 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
2872 | data_end, old_data_start + new_size - data_end); | |
2873 | } else { | |
2874 | struct btrfs_disk_key disk_key; | |
2875 | u64 offset; | |
2876 | ||
2877 | btrfs_item_key(leaf, &disk_key, slot); | |
2878 | ||
2879 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
2880 | unsigned long ptr; | |
2881 | struct btrfs_file_extent_item *fi; | |
2882 | ||
2883 | fi = btrfs_item_ptr(leaf, slot, | |
2884 | struct btrfs_file_extent_item); | |
2885 | fi = (struct btrfs_file_extent_item *)( | |
2886 | (unsigned long)fi - size_diff); | |
2887 | ||
2888 | if (btrfs_file_extent_type(leaf, fi) == | |
2889 | BTRFS_FILE_EXTENT_INLINE) { | |
2890 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
2891 | memmove_extent_buffer(leaf, ptr, | |
2892 | (unsigned long)fi, | |
2893 | offsetof(struct btrfs_file_extent_item, | |
2894 | disk_bytenr)); | |
2895 | } | |
2896 | } | |
2897 | ||
2898 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
2899 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
2900 | data_end, old_data_start - data_end); | |
2901 | ||
2902 | offset = btrfs_disk_key_offset(&disk_key); | |
2903 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
2904 | btrfs_set_item_key(leaf, &disk_key, slot); | |
2905 | if (slot == 0) | |
2906 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
2907 | } | |
5f39d397 CM |
2908 | |
2909 | item = btrfs_item_nr(leaf, slot); | |
2910 | btrfs_set_item_size(leaf, item, new_size); | |
2911 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 CM |
2912 | |
2913 | ret = 0; | |
5f39d397 CM |
2914 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
2915 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 2916 | BUG(); |
5f39d397 | 2917 | } |
b18c6685 CM |
2918 | return ret; |
2919 | } | |
2920 | ||
d352ac68 CM |
2921 | /* |
2922 | * make the item pointed to by the path bigger, data_size is the new size. | |
2923 | */ | |
5f39d397 CM |
2924 | int btrfs_extend_item(struct btrfs_trans_handle *trans, |
2925 | struct btrfs_root *root, struct btrfs_path *path, | |
2926 | u32 data_size) | |
6567e837 CM |
2927 | { |
2928 | int ret = 0; | |
2929 | int slot; | |
2930 | int slot_orig; | |
5f39d397 CM |
2931 | struct extent_buffer *leaf; |
2932 | struct btrfs_item *item; | |
6567e837 CM |
2933 | u32 nritems; |
2934 | unsigned int data_end; | |
2935 | unsigned int old_data; | |
2936 | unsigned int old_size; | |
2937 | int i; | |
2938 | ||
2939 | slot_orig = path->slots[0]; | |
5f39d397 | 2940 | leaf = path->nodes[0]; |
6567e837 | 2941 | |
5f39d397 | 2942 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
2943 | data_end = leaf_data_end(root, leaf); |
2944 | ||
5f39d397 CM |
2945 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
2946 | btrfs_print_leaf(root, leaf); | |
6567e837 | 2947 | BUG(); |
5f39d397 | 2948 | } |
6567e837 | 2949 | slot = path->slots[0]; |
5f39d397 | 2950 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
2951 | |
2952 | BUG_ON(slot < 0); | |
3326d1b0 CM |
2953 | if (slot >= nritems) { |
2954 | btrfs_print_leaf(root, leaf); | |
2955 | printk("slot %d too large, nritems %d\n", slot, nritems); | |
2956 | BUG_ON(1); | |
2957 | } | |
6567e837 CM |
2958 | |
2959 | /* | |
2960 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
2961 | */ | |
2962 | /* first correct the data pointers */ | |
2963 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
2964 | u32 ioff; |
2965 | item = btrfs_item_nr(leaf, i); | |
db94535d CM |
2966 | |
2967 | if (!leaf->map_token) { | |
2968 | map_extent_buffer(leaf, (unsigned long)item, | |
2969 | sizeof(struct btrfs_item), | |
2970 | &leaf->map_token, &leaf->kaddr, | |
2971 | &leaf->map_start, &leaf->map_len, | |
2972 | KM_USER1); | |
2973 | } | |
5f39d397 CM |
2974 | ioff = btrfs_item_offset(leaf, item); |
2975 | btrfs_set_item_offset(leaf, item, ioff - data_size); | |
6567e837 | 2976 | } |
5f39d397 | 2977 | |
db94535d CM |
2978 | if (leaf->map_token) { |
2979 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
2980 | leaf->map_token = NULL; | |
2981 | } | |
2982 | ||
6567e837 | 2983 | /* shift the data */ |
5f39d397 | 2984 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
2985 | data_end - data_size, btrfs_leaf_data(leaf) + |
2986 | data_end, old_data - data_end); | |
5f39d397 | 2987 | |
6567e837 | 2988 | data_end = old_data; |
5f39d397 CM |
2989 | old_size = btrfs_item_size_nr(leaf, slot); |
2990 | item = btrfs_item_nr(leaf, slot); | |
2991 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
2992 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 CM |
2993 | |
2994 | ret = 0; | |
5f39d397 CM |
2995 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
2996 | btrfs_print_leaf(root, leaf); | |
6567e837 | 2997 | BUG(); |
5f39d397 | 2998 | } |
6567e837 CM |
2999 | return ret; |
3000 | } | |
3001 | ||
74123bd7 | 3002 | /* |
d352ac68 | 3003 | * Given a key and some data, insert items into the tree. |
74123bd7 CM |
3004 | * This does all the path init required, making room in the tree if needed. |
3005 | */ | |
9c58309d | 3006 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3007 | struct btrfs_root *root, |
3008 | struct btrfs_path *path, | |
9c58309d CM |
3009 | struct btrfs_key *cpu_key, u32 *data_size, |
3010 | int nr) | |
be0e5c09 | 3011 | { |
5f39d397 CM |
3012 | struct extent_buffer *leaf; |
3013 | struct btrfs_item *item; | |
aa5d6bed | 3014 | int ret = 0; |
be0e5c09 | 3015 | int slot; |
eb60ceac | 3016 | int slot_orig; |
9c58309d | 3017 | int i; |
7518a238 | 3018 | u32 nritems; |
9c58309d CM |
3019 | u32 total_size = 0; |
3020 | u32 total_data = 0; | |
be0e5c09 | 3021 | unsigned int data_end; |
e2fa7227 CM |
3022 | struct btrfs_disk_key disk_key; |
3023 | ||
9c58309d CM |
3024 | for (i = 0; i < nr; i++) { |
3025 | total_data += data_size[i]; | |
3026 | } | |
be0e5c09 | 3027 | |
7b128766 | 3028 | total_size = total_data + (nr * sizeof(struct btrfs_item)); |
9c58309d | 3029 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
0f9dd46c | 3030 | if (ret == 0) |
f0930a37 | 3031 | return -EEXIST; |
ed2ff2cb CM |
3032 | if (ret < 0) |
3033 | goto out; | |
be0e5c09 | 3034 | |
62e2749e | 3035 | slot_orig = path->slots[0]; |
5f39d397 | 3036 | leaf = path->nodes[0]; |
74123bd7 | 3037 | |
5f39d397 | 3038 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 3039 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 3040 | |
f25956cc | 3041 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 CM |
3042 | btrfs_print_leaf(root, leaf); |
3043 | printk("not enough freespace need %u have %d\n", | |
9c58309d | 3044 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 3045 | BUG(); |
d4dbff95 | 3046 | } |
5f39d397 | 3047 | |
62e2749e | 3048 | slot = path->slots[0]; |
eb60ceac | 3049 | BUG_ON(slot < 0); |
5f39d397 | 3050 | |
be0e5c09 | 3051 | if (slot != nritems) { |
5f39d397 | 3052 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 3053 | |
5f39d397 CM |
3054 | if (old_data < data_end) { |
3055 | btrfs_print_leaf(root, leaf); | |
3056 | printk("slot %d old_data %d data_end %d\n", | |
3057 | slot, old_data, data_end); | |
3058 | BUG_ON(1); | |
3059 | } | |
be0e5c09 CM |
3060 | /* |
3061 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3062 | */ | |
3063 | /* first correct the data pointers */ | |
db94535d | 3064 | WARN_ON(leaf->map_token); |
0783fcfc | 3065 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3066 | u32 ioff; |
db94535d | 3067 | |
5f39d397 | 3068 | item = btrfs_item_nr(leaf, i); |
db94535d CM |
3069 | if (!leaf->map_token) { |
3070 | map_extent_buffer(leaf, (unsigned long)item, | |
3071 | sizeof(struct btrfs_item), | |
3072 | &leaf->map_token, &leaf->kaddr, | |
3073 | &leaf->map_start, &leaf->map_len, | |
3074 | KM_USER1); | |
3075 | } | |
3076 | ||
5f39d397 | 3077 | ioff = btrfs_item_offset(leaf, item); |
9c58309d | 3078 | btrfs_set_item_offset(leaf, item, ioff - total_data); |
0783fcfc | 3079 | } |
db94535d CM |
3080 | if (leaf->map_token) { |
3081 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3082 | leaf->map_token = NULL; | |
3083 | } | |
be0e5c09 CM |
3084 | |
3085 | /* shift the items */ | |
9c58309d | 3086 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 3087 | btrfs_item_nr_offset(slot), |
d6025579 | 3088 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
3089 | |
3090 | /* shift the data */ | |
5f39d397 | 3091 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 3092 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 3093 | data_end, old_data - data_end); |
be0e5c09 CM |
3094 | data_end = old_data; |
3095 | } | |
5f39d397 | 3096 | |
62e2749e | 3097 | /* setup the item for the new data */ |
9c58309d CM |
3098 | for (i = 0; i < nr; i++) { |
3099 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3100 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3101 | item = btrfs_item_nr(leaf, slot + i); | |
3102 | btrfs_set_item_offset(leaf, item, data_end - data_size[i]); | |
3103 | data_end -= data_size[i]; | |
3104 | btrfs_set_item_size(leaf, item, data_size[i]); | |
3105 | } | |
3106 | btrfs_set_header_nritems(leaf, nritems + nr); | |
5f39d397 | 3107 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed CM |
3108 | |
3109 | ret = 0; | |
5a01a2e3 CM |
3110 | if (slot == 0) { |
3111 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
e089f05c | 3112 | ret = fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 3113 | } |
aa5d6bed | 3114 | |
5f39d397 CM |
3115 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3116 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 3117 | BUG(); |
5f39d397 | 3118 | } |
ed2ff2cb | 3119 | out: |
62e2749e CM |
3120 | return ret; |
3121 | } | |
3122 | ||
3123 | /* | |
3124 | * Given a key and some data, insert an item into the tree. | |
3125 | * This does all the path init required, making room in the tree if needed. | |
3126 | */ | |
e089f05c CM |
3127 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
3128 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
3129 | data_size) | |
62e2749e CM |
3130 | { |
3131 | int ret = 0; | |
2c90e5d6 | 3132 | struct btrfs_path *path; |
5f39d397 CM |
3133 | struct extent_buffer *leaf; |
3134 | unsigned long ptr; | |
62e2749e | 3135 | |
2c90e5d6 CM |
3136 | path = btrfs_alloc_path(); |
3137 | BUG_ON(!path); | |
2c90e5d6 | 3138 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 3139 | if (!ret) { |
5f39d397 CM |
3140 | leaf = path->nodes[0]; |
3141 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3142 | write_extent_buffer(leaf, data, ptr, data_size); | |
3143 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 3144 | } |
2c90e5d6 | 3145 | btrfs_free_path(path); |
aa5d6bed | 3146 | return ret; |
be0e5c09 CM |
3147 | } |
3148 | ||
74123bd7 | 3149 | /* |
5de08d7d | 3150 | * delete the pointer from a given node. |
74123bd7 | 3151 | * |
d352ac68 CM |
3152 | * the tree should have been previously balanced so the deletion does not |
3153 | * empty a node. | |
74123bd7 | 3154 | */ |
e089f05c CM |
3155 | static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3156 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 3157 | { |
5f39d397 | 3158 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 3159 | u32 nritems; |
aa5d6bed | 3160 | int ret = 0; |
bb803951 | 3161 | int wret; |
be0e5c09 | 3162 | |
5f39d397 | 3163 | nritems = btrfs_header_nritems(parent); |
bb803951 | 3164 | if (slot != nritems -1) { |
5f39d397 CM |
3165 | memmove_extent_buffer(parent, |
3166 | btrfs_node_key_ptr_offset(slot), | |
3167 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
3168 | sizeof(struct btrfs_key_ptr) * |
3169 | (nritems - slot - 1)); | |
bb803951 | 3170 | } |
7518a238 | 3171 | nritems--; |
5f39d397 | 3172 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 3173 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 3174 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 3175 | /* just turn the root into a leaf and break */ |
5f39d397 | 3176 | btrfs_set_header_level(root->node, 0); |
bb803951 | 3177 | } else if (slot == 0) { |
5f39d397 CM |
3178 | struct btrfs_disk_key disk_key; |
3179 | ||
3180 | btrfs_node_key(parent, &disk_key, 0); | |
3181 | wret = fixup_low_keys(trans, root, path, &disk_key, level + 1); | |
0f70abe2 CM |
3182 | if (wret) |
3183 | ret = wret; | |
be0e5c09 | 3184 | } |
d6025579 | 3185 | btrfs_mark_buffer_dirty(parent); |
aa5d6bed | 3186 | return ret; |
be0e5c09 CM |
3187 | } |
3188 | ||
323ac95b CM |
3189 | /* |
3190 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
3191 | * path->nodes[1]. bytenr is the node block pointer, but since the callers | |
3192 | * already know it, it is faster to have them pass it down than to | |
3193 | * read it out of the node again. | |
3194 | * | |
3195 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
3196 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
3197 | * | |
3198 | * The path must have already been setup for deleting the leaf, including | |
3199 | * all the proper balancing. path->nodes[1] must be locked. | |
3200 | */ | |
3201 | noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans, | |
3202 | struct btrfs_root *root, | |
3203 | struct btrfs_path *path, u64 bytenr) | |
3204 | { | |
3205 | int ret; | |
3206 | u64 root_gen = btrfs_header_generation(path->nodes[1]); | |
3207 | ||
3208 | ret = del_ptr(trans, root, path, 1, path->slots[1]); | |
3209 | if (ret) | |
3210 | return ret; | |
3211 | ||
3212 | ret = btrfs_free_extent(trans, root, bytenr, | |
3213 | btrfs_level_size(root, 0), | |
3214 | path->nodes[1]->start, | |
3215 | btrfs_header_owner(path->nodes[1]), | |
3bb1a1bc | 3216 | root_gen, 0, 1); |
323ac95b CM |
3217 | return ret; |
3218 | } | |
74123bd7 CM |
3219 | /* |
3220 | * delete the item at the leaf level in path. If that empties | |
3221 | * the leaf, remove it from the tree | |
3222 | */ | |
85e21bac CM |
3223 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3224 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 3225 | { |
5f39d397 CM |
3226 | struct extent_buffer *leaf; |
3227 | struct btrfs_item *item; | |
85e21bac CM |
3228 | int last_off; |
3229 | int dsize = 0; | |
aa5d6bed CM |
3230 | int ret = 0; |
3231 | int wret; | |
85e21bac | 3232 | int i; |
7518a238 | 3233 | u32 nritems; |
be0e5c09 | 3234 | |
5f39d397 | 3235 | leaf = path->nodes[0]; |
85e21bac CM |
3236 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
3237 | ||
3238 | for (i = 0; i < nr; i++) | |
3239 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
3240 | ||
5f39d397 | 3241 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 3242 | |
85e21bac | 3243 | if (slot + nr != nritems) { |
123abc88 | 3244 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
3245 | |
3246 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
3247 | data_end + dsize, |
3248 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 3249 | last_off - data_end); |
5f39d397 | 3250 | |
85e21bac | 3251 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 3252 | u32 ioff; |
db94535d | 3253 | |
5f39d397 | 3254 | item = btrfs_item_nr(leaf, i); |
db94535d CM |
3255 | if (!leaf->map_token) { |
3256 | map_extent_buffer(leaf, (unsigned long)item, | |
3257 | sizeof(struct btrfs_item), | |
3258 | &leaf->map_token, &leaf->kaddr, | |
3259 | &leaf->map_start, &leaf->map_len, | |
3260 | KM_USER1); | |
3261 | } | |
5f39d397 CM |
3262 | ioff = btrfs_item_offset(leaf, item); |
3263 | btrfs_set_item_offset(leaf, item, ioff + dsize); | |
0783fcfc | 3264 | } |
db94535d CM |
3265 | |
3266 | if (leaf->map_token) { | |
3267 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
3268 | leaf->map_token = NULL; | |
3269 | } | |
3270 | ||
5f39d397 | 3271 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 3272 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 3273 | sizeof(struct btrfs_item) * |
85e21bac | 3274 | (nritems - slot - nr)); |
be0e5c09 | 3275 | } |
85e21bac CM |
3276 | btrfs_set_header_nritems(leaf, nritems - nr); |
3277 | nritems -= nr; | |
5f39d397 | 3278 | |
74123bd7 | 3279 | /* delete the leaf if we've emptied it */ |
7518a238 | 3280 | if (nritems == 0) { |
5f39d397 CM |
3281 | if (leaf == root->node) { |
3282 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 3283 | } else { |
323ac95b CM |
3284 | ret = btrfs_del_leaf(trans, root, path, leaf->start); |
3285 | BUG_ON(ret); | |
9a8dd150 | 3286 | } |
be0e5c09 | 3287 | } else { |
7518a238 | 3288 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 3289 | if (slot == 0) { |
5f39d397 CM |
3290 | struct btrfs_disk_key disk_key; |
3291 | ||
3292 | btrfs_item_key(leaf, &disk_key, 0); | |
e089f05c | 3293 | wret = fixup_low_keys(trans, root, path, |
5f39d397 | 3294 | &disk_key, 1); |
aa5d6bed CM |
3295 | if (wret) |
3296 | ret = wret; | |
3297 | } | |
aa5d6bed | 3298 | |
74123bd7 | 3299 | /* delete the leaf if it is mostly empty */ |
85e21bac | 3300 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) { |
be0e5c09 CM |
3301 | /* push_leaf_left fixes the path. |
3302 | * make sure the path still points to our leaf | |
3303 | * for possible call to del_ptr below | |
3304 | */ | |
4920c9ac | 3305 | slot = path->slots[1]; |
5f39d397 CM |
3306 | extent_buffer_get(leaf); |
3307 | ||
85e21bac | 3308 | wret = push_leaf_left(trans, root, path, 1, 1); |
54aa1f4d | 3309 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 3310 | ret = wret; |
5f39d397 CM |
3311 | |
3312 | if (path->nodes[0] == leaf && | |
3313 | btrfs_header_nritems(leaf)) { | |
85e21bac | 3314 | wret = push_leaf_right(trans, root, path, 1, 1); |
54aa1f4d | 3315 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
3316 | ret = wret; |
3317 | } | |
5f39d397 CM |
3318 | |
3319 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b CM |
3320 | path->slots[1] = slot; |
3321 | ret = btrfs_del_leaf(trans, root, path, leaf->start); | |
3322 | BUG_ON(ret); | |
5f39d397 | 3323 | free_extent_buffer(leaf); |
5de08d7d | 3324 | } else { |
925baedd CM |
3325 | /* if we're still in the path, make sure |
3326 | * we're dirty. Otherwise, one of the | |
3327 | * push_leaf functions must have already | |
3328 | * dirtied this buffer | |
3329 | */ | |
3330 | if (path->nodes[0] == leaf) | |
3331 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 3332 | free_extent_buffer(leaf); |
be0e5c09 | 3333 | } |
d5719762 | 3334 | } else { |
5f39d397 | 3335 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
3336 | } |
3337 | } | |
aa5d6bed | 3338 | return ret; |
be0e5c09 CM |
3339 | } |
3340 | ||
7bb86316 | 3341 | /* |
925baedd | 3342 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
3343 | * returns 0 if it found something or 1 if there are no lesser leaves. |
3344 | * returns < 0 on io errors. | |
d352ac68 CM |
3345 | * |
3346 | * This may release the path, and so you may lose any locks held at the | |
3347 | * time you call it. | |
7bb86316 CM |
3348 | */ |
3349 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
3350 | { | |
925baedd CM |
3351 | struct btrfs_key key; |
3352 | struct btrfs_disk_key found_key; | |
3353 | int ret; | |
7bb86316 | 3354 | |
925baedd | 3355 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 3356 | |
925baedd CM |
3357 | if (key.offset > 0) |
3358 | key.offset--; | |
3359 | else if (key.type > 0) | |
3360 | key.type--; | |
3361 | else if (key.objectid > 0) | |
3362 | key.objectid--; | |
3363 | else | |
3364 | return 1; | |
7bb86316 | 3365 | |
925baedd CM |
3366 | btrfs_release_path(root, path); |
3367 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3368 | if (ret < 0) | |
3369 | return ret; | |
3370 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
3371 | ret = comp_keys(&found_key, &key); | |
3372 | if (ret < 0) | |
3373 | return 0; | |
3374 | return 1; | |
7bb86316 CM |
3375 | } |
3376 | ||
3f157a2f CM |
3377 | /* |
3378 | * A helper function to walk down the tree starting at min_key, and looking | |
3379 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 3380 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
3381 | * |
3382 | * This does not cow, but it does stuff the starting key it finds back | |
3383 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
3384 | * key and get a writable path. | |
3385 | * | |
3386 | * This does lock as it descends, and path->keep_locks should be set | |
3387 | * to 1 by the caller. | |
3388 | * | |
3389 | * This honors path->lowest_level to prevent descent past a given level | |
3390 | * of the tree. | |
3391 | * | |
d352ac68 CM |
3392 | * min_trans indicates the oldest transaction that you are interested |
3393 | * in walking through. Any nodes or leaves older than min_trans are | |
3394 | * skipped over (without reading them). | |
3395 | * | |
3f157a2f CM |
3396 | * returns zero if something useful was found, < 0 on error and 1 if there |
3397 | * was nothing in the tree that matched the search criteria. | |
3398 | */ | |
3399 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 3400 | struct btrfs_key *max_key, |
3f157a2f CM |
3401 | struct btrfs_path *path, int cache_only, |
3402 | u64 min_trans) | |
3403 | { | |
3404 | struct extent_buffer *cur; | |
3405 | struct btrfs_key found_key; | |
3406 | int slot; | |
9652480b | 3407 | int sret; |
3f157a2f CM |
3408 | u32 nritems; |
3409 | int level; | |
3410 | int ret = 1; | |
3411 | ||
3412 | again: | |
3413 | cur = btrfs_lock_root_node(root); | |
3414 | level = btrfs_header_level(cur); | |
e02119d5 | 3415 | WARN_ON(path->nodes[level]); |
3f157a2f CM |
3416 | path->nodes[level] = cur; |
3417 | path->locks[level] = 1; | |
3418 | ||
3419 | if (btrfs_header_generation(cur) < min_trans) { | |
3420 | ret = 1; | |
3421 | goto out; | |
3422 | } | |
3423 | while(1) { | |
3424 | nritems = btrfs_header_nritems(cur); | |
3425 | level = btrfs_header_level(cur); | |
9652480b | 3426 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 3427 | |
323ac95b CM |
3428 | /* at the lowest level, we're done, setup the path and exit */ |
3429 | if (level == path->lowest_level) { | |
e02119d5 CM |
3430 | if (slot >= nritems) |
3431 | goto find_next_key; | |
3f157a2f CM |
3432 | ret = 0; |
3433 | path->slots[level] = slot; | |
3434 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
3435 | goto out; | |
3436 | } | |
9652480b Y |
3437 | if (sret && slot > 0) |
3438 | slot--; | |
3f157a2f CM |
3439 | /* |
3440 | * check this node pointer against the cache_only and | |
3441 | * min_trans parameters. If it isn't in cache or is too | |
3442 | * old, skip to the next one. | |
3443 | */ | |
3444 | while(slot < nritems) { | |
3445 | u64 blockptr; | |
3446 | u64 gen; | |
3447 | struct extent_buffer *tmp; | |
e02119d5 CM |
3448 | struct btrfs_disk_key disk_key; |
3449 | ||
3f157a2f CM |
3450 | blockptr = btrfs_node_blockptr(cur, slot); |
3451 | gen = btrfs_node_ptr_generation(cur, slot); | |
3452 | if (gen < min_trans) { | |
3453 | slot++; | |
3454 | continue; | |
3455 | } | |
3456 | if (!cache_only) | |
3457 | break; | |
3458 | ||
e02119d5 CM |
3459 | if (max_key) { |
3460 | btrfs_node_key(cur, &disk_key, slot); | |
3461 | if (comp_keys(&disk_key, max_key) >= 0) { | |
3462 | ret = 1; | |
3463 | goto out; | |
3464 | } | |
3465 | } | |
3466 | ||
3f157a2f CM |
3467 | tmp = btrfs_find_tree_block(root, blockptr, |
3468 | btrfs_level_size(root, level - 1)); | |
3469 | ||
3470 | if (tmp && btrfs_buffer_uptodate(tmp, gen)) { | |
3471 | free_extent_buffer(tmp); | |
3472 | break; | |
3473 | } | |
3474 | if (tmp) | |
3475 | free_extent_buffer(tmp); | |
3476 | slot++; | |
3477 | } | |
e02119d5 | 3478 | find_next_key: |
3f157a2f CM |
3479 | /* |
3480 | * we didn't find a candidate key in this node, walk forward | |
3481 | * and find another one | |
3482 | */ | |
3483 | if (slot >= nritems) { | |
e02119d5 CM |
3484 | path->slots[level] = slot; |
3485 | sret = btrfs_find_next_key(root, path, min_key, level, | |
3f157a2f | 3486 | cache_only, min_trans); |
e02119d5 | 3487 | if (sret == 0) { |
3f157a2f CM |
3488 | btrfs_release_path(root, path); |
3489 | goto again; | |
3490 | } else { | |
3491 | goto out; | |
3492 | } | |
3493 | } | |
3494 | /* save our key for returning back */ | |
3495 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
3496 | path->slots[level] = slot; | |
3497 | if (level == path->lowest_level) { | |
3498 | ret = 0; | |
3499 | unlock_up(path, level, 1); | |
3500 | goto out; | |
3501 | } | |
3502 | cur = read_node_slot(root, cur, slot); | |
3503 | ||
3504 | btrfs_tree_lock(cur); | |
3505 | path->locks[level - 1] = 1; | |
3506 | path->nodes[level - 1] = cur; | |
3507 | unlock_up(path, level, 1); | |
3508 | } | |
3509 | out: | |
3510 | if (ret == 0) | |
3511 | memcpy(min_key, &found_key, sizeof(found_key)); | |
3512 | return ret; | |
3513 | } | |
3514 | ||
3515 | /* | |
3516 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
3517 | * and fixup the path. It looks for and returns the next key in the | |
3518 | * tree based on the current path and the cache_only and min_trans | |
3519 | * parameters. | |
3520 | * | |
3521 | * 0 is returned if another key is found, < 0 if there are any errors | |
3522 | * and 1 is returned if there are no higher keys in the tree | |
3523 | * | |
3524 | * path->keep_locks should be set to 1 on the search made before | |
3525 | * calling this function. | |
3526 | */ | |
e7a84565 | 3527 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
3f157a2f CM |
3528 | struct btrfs_key *key, int lowest_level, |
3529 | int cache_only, u64 min_trans) | |
e7a84565 CM |
3530 | { |
3531 | int level = lowest_level; | |
3532 | int slot; | |
3533 | struct extent_buffer *c; | |
3534 | ||
3535 | while(level < BTRFS_MAX_LEVEL) { | |
3536 | if (!path->nodes[level]) | |
3537 | return 1; | |
3538 | ||
3539 | slot = path->slots[level] + 1; | |
3540 | c = path->nodes[level]; | |
3f157a2f | 3541 | next: |
e7a84565 CM |
3542 | if (slot >= btrfs_header_nritems(c)) { |
3543 | level++; | |
3544 | if (level == BTRFS_MAX_LEVEL) { | |
3545 | return 1; | |
3546 | } | |
3547 | continue; | |
3548 | } | |
3549 | if (level == 0) | |
3550 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
3551 | else { |
3552 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
3553 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
3554 | ||
3555 | if (cache_only) { | |
3556 | struct extent_buffer *cur; | |
3557 | cur = btrfs_find_tree_block(root, blockptr, | |
3558 | btrfs_level_size(root, level - 1)); | |
3559 | if (!cur || !btrfs_buffer_uptodate(cur, gen)) { | |
3560 | slot++; | |
3561 | if (cur) | |
3562 | free_extent_buffer(cur); | |
3563 | goto next; | |
3564 | } | |
3565 | free_extent_buffer(cur); | |
3566 | } | |
3567 | if (gen < min_trans) { | |
3568 | slot++; | |
3569 | goto next; | |
3570 | } | |
e7a84565 | 3571 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 3572 | } |
e7a84565 CM |
3573 | return 0; |
3574 | } | |
3575 | return 1; | |
3576 | } | |
3577 | ||
97571fd0 | 3578 | /* |
925baedd | 3579 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
3580 | * returns 0 if it found something or 1 if there are no greater leaves. |
3581 | * returns < 0 on io errors. | |
97571fd0 | 3582 | */ |
234b63a0 | 3583 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
3584 | { |
3585 | int slot; | |
3586 | int level = 1; | |
5f39d397 CM |
3587 | struct extent_buffer *c; |
3588 | struct extent_buffer *next = NULL; | |
925baedd CM |
3589 | struct btrfs_key key; |
3590 | u32 nritems; | |
3591 | int ret; | |
3592 | ||
3593 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3594 | if (nritems == 0) { | |
3595 | return 1; | |
3596 | } | |
3597 | ||
3598 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); | |
3599 | ||
925baedd | 3600 | btrfs_release_path(root, path); |
a2135011 | 3601 | path->keep_locks = 1; |
925baedd CM |
3602 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3603 | path->keep_locks = 0; | |
3604 | ||
3605 | if (ret < 0) | |
3606 | return ret; | |
3607 | ||
a2135011 | 3608 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
3609 | /* |
3610 | * by releasing the path above we dropped all our locks. A balance | |
3611 | * could have added more items next to the key that used to be | |
3612 | * at the very end of the block. So, check again here and | |
3613 | * advance the path if there are now more items available. | |
3614 | */ | |
a2135011 | 3615 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
168fd7d2 | 3616 | path->slots[0]++; |
925baedd CM |
3617 | goto done; |
3618 | } | |
d97e63b6 | 3619 | |
234b63a0 | 3620 | while(level < BTRFS_MAX_LEVEL) { |
d97e63b6 | 3621 | if (!path->nodes[level]) |
0f70abe2 | 3622 | return 1; |
5f39d397 | 3623 | |
d97e63b6 CM |
3624 | slot = path->slots[level] + 1; |
3625 | c = path->nodes[level]; | |
5f39d397 | 3626 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 3627 | level++; |
925baedd | 3628 | if (level == BTRFS_MAX_LEVEL) { |
7bb86316 | 3629 | return 1; |
925baedd | 3630 | } |
d97e63b6 CM |
3631 | continue; |
3632 | } | |
5f39d397 | 3633 | |
925baedd CM |
3634 | if (next) { |
3635 | btrfs_tree_unlock(next); | |
5f39d397 | 3636 | free_extent_buffer(next); |
925baedd | 3637 | } |
5f39d397 | 3638 | |
0bd40a71 CM |
3639 | if (level == 1 && (path->locks[1] || path->skip_locking) && |
3640 | path->reada) | |
01f46658 | 3641 | reada_for_search(root, path, level, slot, 0); |
5f39d397 | 3642 | |
ca7a79ad | 3643 | next = read_node_slot(root, c, slot); |
5cd57b2c CM |
3644 | if (!path->skip_locking) { |
3645 | WARN_ON(!btrfs_tree_locked(c)); | |
3646 | btrfs_tree_lock(next); | |
3647 | } | |
d97e63b6 CM |
3648 | break; |
3649 | } | |
3650 | path->slots[level] = slot; | |
3651 | while(1) { | |
3652 | level--; | |
3653 | c = path->nodes[level]; | |
925baedd CM |
3654 | if (path->locks[level]) |
3655 | btrfs_tree_unlock(c); | |
5f39d397 | 3656 | free_extent_buffer(c); |
d97e63b6 CM |
3657 | path->nodes[level] = next; |
3658 | path->slots[level] = 0; | |
a74a4b97 CM |
3659 | if (!path->skip_locking) |
3660 | path->locks[level] = 1; | |
d97e63b6 CM |
3661 | if (!level) |
3662 | break; | |
925baedd CM |
3663 | if (level == 1 && path->locks[1] && path->reada) |
3664 | reada_for_search(root, path, level, slot, 0); | |
ca7a79ad | 3665 | next = read_node_slot(root, next, 0); |
5cd57b2c CM |
3666 | if (!path->skip_locking) { |
3667 | WARN_ON(!btrfs_tree_locked(path->nodes[level])); | |
3668 | btrfs_tree_lock(next); | |
3669 | } | |
d97e63b6 | 3670 | } |
925baedd CM |
3671 | done: |
3672 | unlock_up(path, 0, 1); | |
d97e63b6 CM |
3673 | return 0; |
3674 | } | |
0b86a832 | 3675 | |
3f157a2f CM |
3676 | /* |
3677 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
3678 | * searching until it gets past min_objectid or finds an item of 'type' | |
3679 | * | |
3680 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
3681 | */ | |
0b86a832 CM |
3682 | int btrfs_previous_item(struct btrfs_root *root, |
3683 | struct btrfs_path *path, u64 min_objectid, | |
3684 | int type) | |
3685 | { | |
3686 | struct btrfs_key found_key; | |
3687 | struct extent_buffer *leaf; | |
e02119d5 | 3688 | u32 nritems; |
0b86a832 CM |
3689 | int ret; |
3690 | ||
3691 | while(1) { | |
3692 | if (path->slots[0] == 0) { | |
3693 | ret = btrfs_prev_leaf(root, path); | |
3694 | if (ret != 0) | |
3695 | return ret; | |
3696 | } else { | |
3697 | path->slots[0]--; | |
3698 | } | |
3699 | leaf = path->nodes[0]; | |
e02119d5 CM |
3700 | nritems = btrfs_header_nritems(leaf); |
3701 | if (nritems == 0) | |
3702 | return 1; | |
3703 | if (path->slots[0] == nritems) | |
3704 | path->slots[0]--; | |
3705 | ||
0b86a832 CM |
3706 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3707 | if (found_key.type == type) | |
3708 | return 0; | |
e02119d5 CM |
3709 | if (found_key.objectid < min_objectid) |
3710 | break; | |
3711 | if (found_key.objectid == min_objectid && | |
3712 | found_key.type < type) | |
3713 | break; | |
0b86a832 CM |
3714 | } |
3715 | return 1; | |
3716 | } |