]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
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 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
4b4e25f2 | 24 | #include "compat.h" |
74493f7a | 25 | #include "hash.h" |
fec577fb CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "print-tree.h" | |
e089f05c | 29 | #include "transaction.h" |
0b86a832 | 30 | #include "volumes.h" |
925baedd | 31 | #include "locking.h" |
fa9c0d79 | 32 | #include "free-space-cache.h" |
fec577fb | 33 | |
56bec294 CM |
34 | static int update_reserved_extents(struct btrfs_root *root, |
35 | u64 bytenr, u64 num, int reserve); | |
f3465ca4 JB |
36 | static int update_block_group(struct btrfs_trans_handle *trans, |
37 | struct btrfs_root *root, | |
38 | u64 bytenr, u64 num_bytes, int alloc, | |
39 | int mark_free); | |
5d4f98a2 YZ |
40 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
41 | struct btrfs_root *root, | |
42 | u64 bytenr, u64 num_bytes, u64 parent, | |
43 | u64 root_objectid, u64 owner_objectid, | |
44 | u64 owner_offset, int refs_to_drop, | |
45 | struct btrfs_delayed_extent_op *extra_op); | |
46 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
47 | struct extent_buffer *leaf, | |
48 | struct btrfs_extent_item *ei); | |
49 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
50 | struct btrfs_root *root, | |
51 | u64 parent, u64 root_objectid, | |
52 | u64 flags, u64 owner, u64 offset, | |
53 | struct btrfs_key *ins, int ref_mod); | |
54 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
55 | struct btrfs_root *root, | |
56 | u64 parent, u64 root_objectid, | |
57 | u64 flags, struct btrfs_disk_key *key, | |
58 | int level, struct btrfs_key *ins); | |
d548ee51 | 59 | |
6a63209f JB |
60 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
61 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
62 | u64 flags, int force); | |
63 | ||
0f9dd46c JB |
64 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
65 | { | |
66 | return (cache->flags & bits) == bits; | |
67 | } | |
68 | ||
69 | /* | |
70 | * this adds the block group to the fs_info rb tree for the block group | |
71 | * cache | |
72 | */ | |
b2950863 | 73 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
74 | struct btrfs_block_group_cache *block_group) |
75 | { | |
76 | struct rb_node **p; | |
77 | struct rb_node *parent = NULL; | |
78 | struct btrfs_block_group_cache *cache; | |
79 | ||
80 | spin_lock(&info->block_group_cache_lock); | |
81 | p = &info->block_group_cache_tree.rb_node; | |
82 | ||
83 | while (*p) { | |
84 | parent = *p; | |
85 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
86 | cache_node); | |
87 | if (block_group->key.objectid < cache->key.objectid) { | |
88 | p = &(*p)->rb_left; | |
89 | } else if (block_group->key.objectid > cache->key.objectid) { | |
90 | p = &(*p)->rb_right; | |
91 | } else { | |
92 | spin_unlock(&info->block_group_cache_lock); | |
93 | return -EEXIST; | |
94 | } | |
95 | } | |
96 | ||
97 | rb_link_node(&block_group->cache_node, parent, p); | |
98 | rb_insert_color(&block_group->cache_node, | |
99 | &info->block_group_cache_tree); | |
100 | spin_unlock(&info->block_group_cache_lock); | |
101 | ||
102 | return 0; | |
103 | } | |
104 | ||
105 | /* | |
106 | * This will return the block group at or after bytenr if contains is 0, else | |
107 | * it will return the block group that contains the bytenr | |
108 | */ | |
109 | static struct btrfs_block_group_cache * | |
110 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
111 | int contains) | |
112 | { | |
113 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
114 | struct rb_node *n; | |
115 | u64 end, start; | |
116 | ||
117 | spin_lock(&info->block_group_cache_lock); | |
118 | n = info->block_group_cache_tree.rb_node; | |
119 | ||
120 | while (n) { | |
121 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
122 | cache_node); | |
123 | end = cache->key.objectid + cache->key.offset - 1; | |
124 | start = cache->key.objectid; | |
125 | ||
126 | if (bytenr < start) { | |
127 | if (!contains && (!ret || start < ret->key.objectid)) | |
128 | ret = cache; | |
129 | n = n->rb_left; | |
130 | } else if (bytenr > start) { | |
131 | if (contains && bytenr <= end) { | |
132 | ret = cache; | |
133 | break; | |
134 | } | |
135 | n = n->rb_right; | |
136 | } else { | |
137 | ret = cache; | |
138 | break; | |
139 | } | |
140 | } | |
d2fb3437 YZ |
141 | if (ret) |
142 | atomic_inc(&ret->count); | |
0f9dd46c JB |
143 | spin_unlock(&info->block_group_cache_lock); |
144 | ||
145 | return ret; | |
146 | } | |
147 | ||
148 | /* | |
149 | * this is only called by cache_block_group, since we could have freed extents | |
150 | * we need to check the pinned_extents for any extents that can't be used yet | |
151 | * since their free space will be released as soon as the transaction commits. | |
152 | */ | |
153 | static int add_new_free_space(struct btrfs_block_group_cache *block_group, | |
154 | struct btrfs_fs_info *info, u64 start, u64 end) | |
155 | { | |
156 | u64 extent_start, extent_end, size; | |
157 | int ret; | |
158 | ||
159 | while (start < end) { | |
160 | ret = find_first_extent_bit(&info->pinned_extents, start, | |
161 | &extent_start, &extent_end, | |
162 | EXTENT_DIRTY); | |
163 | if (ret) | |
164 | break; | |
165 | ||
166 | if (extent_start == start) { | |
167 | start = extent_end + 1; | |
168 | } else if (extent_start > start && extent_start < end) { | |
169 | size = extent_start - start; | |
ea6a478e JB |
170 | ret = btrfs_add_free_space(block_group, start, |
171 | size); | |
0f9dd46c JB |
172 | BUG_ON(ret); |
173 | start = extent_end + 1; | |
174 | } else { | |
175 | break; | |
176 | } | |
177 | } | |
178 | ||
179 | if (start < end) { | |
180 | size = end - start; | |
ea6a478e | 181 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
182 | BUG_ON(ret); |
183 | } | |
184 | ||
185 | return 0; | |
186 | } | |
187 | ||
a512bbf8 YZ |
188 | static int remove_sb_from_cache(struct btrfs_root *root, |
189 | struct btrfs_block_group_cache *cache) | |
190 | { | |
191 | u64 bytenr; | |
192 | u64 *logical; | |
193 | int stripe_len; | |
194 | int i, nr, ret; | |
195 | ||
196 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
197 | bytenr = btrfs_sb_offset(i); | |
198 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
199 | cache->key.objectid, bytenr, 0, | |
200 | &logical, &nr, &stripe_len); | |
201 | BUG_ON(ret); | |
202 | while (nr--) { | |
203 | btrfs_remove_free_space(cache, logical[nr], | |
204 | stripe_len); | |
205 | } | |
206 | kfree(logical); | |
207 | } | |
208 | return 0; | |
209 | } | |
210 | ||
e37c9e69 CM |
211 | static int cache_block_group(struct btrfs_root *root, |
212 | struct btrfs_block_group_cache *block_group) | |
213 | { | |
214 | struct btrfs_path *path; | |
ef8bbdfe | 215 | int ret = 0; |
e37c9e69 | 216 | struct btrfs_key key; |
5f39d397 | 217 | struct extent_buffer *leaf; |
e37c9e69 | 218 | int slot; |
e4404d6e | 219 | u64 last; |
e37c9e69 | 220 | |
00f5c795 CM |
221 | if (!block_group) |
222 | return 0; | |
223 | ||
e37c9e69 | 224 | root = root->fs_info->extent_root; |
e37c9e69 CM |
225 | |
226 | if (block_group->cached) | |
227 | return 0; | |
f510cfec | 228 | |
e37c9e69 CM |
229 | path = btrfs_alloc_path(); |
230 | if (!path) | |
231 | return -ENOMEM; | |
7d7d6068 | 232 | |
2cc58cf2 | 233 | path->reada = 2; |
5cd57b2c CM |
234 | /* |
235 | * we get into deadlocks with paths held by callers of this function. | |
236 | * since the alloc_mutex is protecting things right now, just | |
237 | * skip the locking here | |
238 | */ | |
239 | path->skip_locking = 1; | |
e4404d6e YZ |
240 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
241 | key.objectid = last; | |
e37c9e69 CM |
242 | key.offset = 0; |
243 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
244 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
245 | if (ret < 0) | |
ef8bbdfe | 246 | goto err; |
a512bbf8 | 247 | |
d397712b | 248 | while (1) { |
5f39d397 | 249 | leaf = path->nodes[0]; |
e37c9e69 | 250 | slot = path->slots[0]; |
5f39d397 | 251 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 252 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
253 | if (ret < 0) |
254 | goto err; | |
0f9dd46c | 255 | if (ret == 0) |
e37c9e69 | 256 | continue; |
0f9dd46c | 257 | else |
e37c9e69 | 258 | break; |
e37c9e69 | 259 | } |
5f39d397 | 260 | btrfs_item_key_to_cpu(leaf, &key, slot); |
0f9dd46c | 261 | if (key.objectid < block_group->key.objectid) |
7d7d6068 | 262 | goto next; |
0f9dd46c | 263 | |
e37c9e69 | 264 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 265 | block_group->key.offset) |
e37c9e69 | 266 | break; |
7d7d6068 | 267 | |
e37c9e69 | 268 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
0f9dd46c JB |
269 | add_new_free_space(block_group, root->fs_info, last, |
270 | key.objectid); | |
271 | ||
7d7d6068 | 272 | last = key.objectid + key.offset; |
e37c9e69 | 273 | } |
7d7d6068 | 274 | next: |
e37c9e69 CM |
275 | path->slots[0]++; |
276 | } | |
277 | ||
0f9dd46c JB |
278 | add_new_free_space(block_group, root->fs_info, last, |
279 | block_group->key.objectid + | |
280 | block_group->key.offset); | |
281 | ||
e37c9e69 | 282 | block_group->cached = 1; |
70cb0743 | 283 | remove_sb_from_cache(root, block_group); |
ef8bbdfe | 284 | ret = 0; |
54aa1f4d | 285 | err: |
e37c9e69 | 286 | btrfs_free_path(path); |
ef8bbdfe | 287 | return ret; |
e37c9e69 CM |
288 | } |
289 | ||
0f9dd46c JB |
290 | /* |
291 | * return the block group that starts at or after bytenr | |
292 | */ | |
d397712b CM |
293 | static struct btrfs_block_group_cache * |
294 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 295 | { |
0f9dd46c | 296 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 297 | |
0f9dd46c | 298 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 299 | |
0f9dd46c | 300 | return cache; |
0ef3e66b CM |
301 | } |
302 | ||
0f9dd46c | 303 | /* |
9f55684c | 304 | * return the block group that contains the given bytenr |
0f9dd46c | 305 | */ |
d397712b CM |
306 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
307 | struct btrfs_fs_info *info, | |
308 | u64 bytenr) | |
be744175 | 309 | { |
0f9dd46c | 310 | struct btrfs_block_group_cache *cache; |
be744175 | 311 | |
0f9dd46c | 312 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 313 | |
0f9dd46c | 314 | return cache; |
be744175 | 315 | } |
0b86a832 | 316 | |
fa9c0d79 | 317 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) |
d2fb3437 YZ |
318 | { |
319 | if (atomic_dec_and_test(&cache->count)) | |
320 | kfree(cache); | |
321 | } | |
322 | ||
0f9dd46c JB |
323 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
324 | u64 flags) | |
6324fbf3 | 325 | { |
0f9dd46c | 326 | struct list_head *head = &info->space_info; |
0f9dd46c | 327 | struct btrfs_space_info *found; |
4184ea7f CM |
328 | |
329 | rcu_read_lock(); | |
330 | list_for_each_entry_rcu(found, head, list) { | |
331 | if (found->flags == flags) { | |
332 | rcu_read_unlock(); | |
0f9dd46c | 333 | return found; |
4184ea7f | 334 | } |
0f9dd46c | 335 | } |
4184ea7f | 336 | rcu_read_unlock(); |
0f9dd46c | 337 | return NULL; |
6324fbf3 CM |
338 | } |
339 | ||
4184ea7f CM |
340 | /* |
341 | * after adding space to the filesystem, we need to clear the full flags | |
342 | * on all the space infos. | |
343 | */ | |
344 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
345 | { | |
346 | struct list_head *head = &info->space_info; | |
347 | struct btrfs_space_info *found; | |
348 | ||
349 | rcu_read_lock(); | |
350 | list_for_each_entry_rcu(found, head, list) | |
351 | found->full = 0; | |
352 | rcu_read_unlock(); | |
353 | } | |
354 | ||
80eb234a JB |
355 | static u64 div_factor(u64 num, int factor) |
356 | { | |
357 | if (factor == 10) | |
358 | return num; | |
359 | num *= factor; | |
360 | do_div(num, 10); | |
361 | return num; | |
362 | } | |
363 | ||
d2fb3437 YZ |
364 | u64 btrfs_find_block_group(struct btrfs_root *root, |
365 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 366 | { |
96b5179d | 367 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 368 | u64 used; |
d2fb3437 YZ |
369 | u64 last = max(search_hint, search_start); |
370 | u64 group_start = 0; | |
31f3c99b | 371 | int full_search = 0; |
d2fb3437 | 372 | int factor = 9; |
0ef3e66b | 373 | int wrapped = 0; |
31f3c99b | 374 | again: |
e8569813 ZY |
375 | while (1) { |
376 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
377 | if (!cache) |
378 | break; | |
96b5179d | 379 | |
c286ac48 | 380 | spin_lock(&cache->lock); |
96b5179d CM |
381 | last = cache->key.objectid + cache->key.offset; |
382 | used = btrfs_block_group_used(&cache->item); | |
383 | ||
d2fb3437 YZ |
384 | if ((full_search || !cache->ro) && |
385 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 386 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
387 | div_factor(cache->key.offset, factor)) { |
388 | group_start = cache->key.objectid; | |
c286ac48 | 389 | spin_unlock(&cache->lock); |
fa9c0d79 | 390 | btrfs_put_block_group(cache); |
8790d502 CM |
391 | goto found; |
392 | } | |
6324fbf3 | 393 | } |
c286ac48 | 394 | spin_unlock(&cache->lock); |
fa9c0d79 | 395 | btrfs_put_block_group(cache); |
de428b63 | 396 | cond_resched(); |
cd1bc465 | 397 | } |
0ef3e66b CM |
398 | if (!wrapped) { |
399 | last = search_start; | |
400 | wrapped = 1; | |
401 | goto again; | |
402 | } | |
403 | if (!full_search && factor < 10) { | |
be744175 | 404 | last = search_start; |
31f3c99b | 405 | full_search = 1; |
0ef3e66b | 406 | factor = 10; |
31f3c99b CM |
407 | goto again; |
408 | } | |
be744175 | 409 | found: |
d2fb3437 | 410 | return group_start; |
925baedd | 411 | } |
0f9dd46c | 412 | |
e02119d5 | 413 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 414 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
415 | { |
416 | int ret; | |
417 | struct btrfs_key key; | |
31840ae1 | 418 | struct btrfs_path *path; |
e02119d5 | 419 | |
31840ae1 ZY |
420 | path = btrfs_alloc_path(); |
421 | BUG_ON(!path); | |
e02119d5 CM |
422 | key.objectid = start; |
423 | key.offset = len; | |
424 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
425 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
426 | 0, 0); | |
31840ae1 | 427 | btrfs_free_path(path); |
7bb86316 CM |
428 | return ret; |
429 | } | |
430 | ||
d8d5f3e1 CM |
431 | /* |
432 | * Back reference rules. Back refs have three main goals: | |
433 | * | |
434 | * 1) differentiate between all holders of references to an extent so that | |
435 | * when a reference is dropped we can make sure it was a valid reference | |
436 | * before freeing the extent. | |
437 | * | |
438 | * 2) Provide enough information to quickly find the holders of an extent | |
439 | * if we notice a given block is corrupted or bad. | |
440 | * | |
441 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
442 | * maintenance. This is actually the same as #2, but with a slightly | |
443 | * different use case. | |
444 | * | |
5d4f98a2 YZ |
445 | * There are two kinds of back refs. The implicit back refs is optimized |
446 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
447 | * back refs of this kind provide information about the block's owner tree | |
448 | * and the pointer's key. These information allow us to find the block by | |
449 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
450 | * referenced by their owner trees. The location of tree block is recorded | |
451 | * in the back refs. Actually the full back refs is generic, and can be | |
452 | * used in all cases the implicit back refs is used. The major shortcoming | |
453 | * of the full back refs is its overhead. Every time a tree block gets | |
454 | * COWed, we have to update back refs entry for all pointers in it. | |
455 | * | |
456 | * For a newly allocated tree block, we use implicit back refs for | |
457 | * pointers in it. This means most tree related operations only involve | |
458 | * implicit back refs. For a tree block created in old transaction, the | |
459 | * only way to drop a reference to it is COW it. So we can detect the | |
460 | * event that tree block loses its owner tree's reference and do the | |
461 | * back refs conversion. | |
462 | * | |
463 | * When a tree block is COW'd through a tree, there are four cases: | |
464 | * | |
465 | * The reference count of the block is one and the tree is the block's | |
466 | * owner tree. Nothing to do in this case. | |
467 | * | |
468 | * The reference count of the block is one and the tree is not the | |
469 | * block's owner tree. In this case, full back refs is used for pointers | |
470 | * in the block. Remove these full back refs, add implicit back refs for | |
471 | * every pointers in the new block. | |
472 | * | |
473 | * The reference count of the block is greater than one and the tree is | |
474 | * the block's owner tree. In this case, implicit back refs is used for | |
475 | * pointers in the block. Add full back refs for every pointers in the | |
476 | * block, increase lower level extents' reference counts. The original | |
477 | * implicit back refs are entailed to the new block. | |
478 | * | |
479 | * The reference count of the block is greater than one and the tree is | |
480 | * not the block's owner tree. Add implicit back refs for every pointer in | |
481 | * the new block, increase lower level extents' reference count. | |
482 | * | |
483 | * Back Reference Key composing: | |
484 | * | |
485 | * The key objectid corresponds to the first byte in the extent, | |
486 | * The key type is used to differentiate between types of back refs. | |
487 | * There are different meanings of the key offset for different types | |
488 | * of back refs. | |
489 | * | |
d8d5f3e1 CM |
490 | * File extents can be referenced by: |
491 | * | |
492 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 493 | * - different files inside a single subvolume |
d8d5f3e1 CM |
494 | * - different offsets inside a file (bookend extents in file.c) |
495 | * | |
5d4f98a2 | 496 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
497 | * |
498 | * - Objectid of the subvolume root | |
d8d5f3e1 | 499 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
500 | * - original offset in the file |
501 | * - how many bookend extents | |
d8d5f3e1 | 502 | * |
5d4f98a2 YZ |
503 | * The key offset for the implicit back refs is hash of the first |
504 | * three fields. | |
d8d5f3e1 | 505 | * |
5d4f98a2 | 506 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 507 | * |
5d4f98a2 | 508 | * - number of pointers in the tree leaf |
d8d5f3e1 | 509 | * |
5d4f98a2 YZ |
510 | * The key offset for the implicit back refs is the first byte of |
511 | * the tree leaf | |
d8d5f3e1 | 512 | * |
5d4f98a2 YZ |
513 | * When a file extent is allocated, The implicit back refs is used. |
514 | * the fields are filled in: | |
d8d5f3e1 | 515 | * |
5d4f98a2 | 516 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 517 | * |
5d4f98a2 YZ |
518 | * When a file extent is removed file truncation, we find the |
519 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 520 | * |
5d4f98a2 | 521 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 522 | * |
5d4f98a2 | 523 | * Btree extents can be referenced by: |
d8d5f3e1 | 524 | * |
5d4f98a2 | 525 | * - Different subvolumes |
d8d5f3e1 | 526 | * |
5d4f98a2 YZ |
527 | * Both the implicit back refs and the full back refs for tree blocks |
528 | * only consist of key. The key offset for the implicit back refs is | |
529 | * objectid of block's owner tree. The key offset for the full back refs | |
530 | * is the first byte of parent block. | |
d8d5f3e1 | 531 | * |
5d4f98a2 YZ |
532 | * When implicit back refs is used, information about the lowest key and |
533 | * level of the tree block are required. These information are stored in | |
534 | * tree block info structure. | |
d8d5f3e1 | 535 | */ |
31840ae1 | 536 | |
5d4f98a2 YZ |
537 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
538 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
539 | struct btrfs_root *root, | |
540 | struct btrfs_path *path, | |
541 | u64 owner, u32 extra_size) | |
7bb86316 | 542 | { |
5d4f98a2 YZ |
543 | struct btrfs_extent_item *item; |
544 | struct btrfs_extent_item_v0 *ei0; | |
545 | struct btrfs_extent_ref_v0 *ref0; | |
546 | struct btrfs_tree_block_info *bi; | |
547 | struct extent_buffer *leaf; | |
7bb86316 | 548 | struct btrfs_key key; |
5d4f98a2 YZ |
549 | struct btrfs_key found_key; |
550 | u32 new_size = sizeof(*item); | |
551 | u64 refs; | |
552 | int ret; | |
553 | ||
554 | leaf = path->nodes[0]; | |
555 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
556 | ||
557 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
558 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
559 | struct btrfs_extent_item_v0); | |
560 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
561 | ||
562 | if (owner == (u64)-1) { | |
563 | while (1) { | |
564 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
565 | ret = btrfs_next_leaf(root, path); | |
566 | if (ret < 0) | |
567 | return ret; | |
568 | BUG_ON(ret > 0); | |
569 | leaf = path->nodes[0]; | |
570 | } | |
571 | btrfs_item_key_to_cpu(leaf, &found_key, | |
572 | path->slots[0]); | |
573 | BUG_ON(key.objectid != found_key.objectid); | |
574 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
575 | path->slots[0]++; | |
576 | continue; | |
577 | } | |
578 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
579 | struct btrfs_extent_ref_v0); | |
580 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
581 | break; | |
582 | } | |
583 | } | |
584 | btrfs_release_path(root, path); | |
585 | ||
586 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
587 | new_size += sizeof(*bi); | |
588 | ||
589 | new_size -= sizeof(*ei0); | |
590 | ret = btrfs_search_slot(trans, root, &key, path, | |
591 | new_size + extra_size, 1); | |
592 | if (ret < 0) | |
593 | return ret; | |
594 | BUG_ON(ret); | |
595 | ||
596 | ret = btrfs_extend_item(trans, root, path, new_size); | |
597 | BUG_ON(ret); | |
598 | ||
599 | leaf = path->nodes[0]; | |
600 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
601 | btrfs_set_extent_refs(leaf, item, refs); | |
602 | /* FIXME: get real generation */ | |
603 | btrfs_set_extent_generation(leaf, item, 0); | |
604 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
605 | btrfs_set_extent_flags(leaf, item, | |
606 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
607 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
608 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
609 | /* FIXME: get first key of the block */ | |
610 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
611 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
612 | } else { | |
613 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
614 | } | |
615 | btrfs_mark_buffer_dirty(leaf); | |
616 | return 0; | |
617 | } | |
618 | #endif | |
619 | ||
620 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
621 | { | |
622 | u32 high_crc = ~(u32)0; | |
623 | u32 low_crc = ~(u32)0; | |
624 | __le64 lenum; | |
625 | ||
626 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 627 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 628 | lenum = cpu_to_le64(owner); |
163e783e | 629 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 630 | lenum = cpu_to_le64(offset); |
163e783e | 631 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
632 | |
633 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
634 | } | |
635 | ||
636 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
637 | struct btrfs_extent_data_ref *ref) | |
638 | { | |
639 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
640 | btrfs_extent_data_ref_objectid(leaf, ref), | |
641 | btrfs_extent_data_ref_offset(leaf, ref)); | |
642 | } | |
643 | ||
644 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
645 | struct btrfs_extent_data_ref *ref, | |
646 | u64 root_objectid, u64 owner, u64 offset) | |
647 | { | |
648 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
649 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
650 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
651 | return 0; | |
652 | return 1; | |
653 | } | |
654 | ||
655 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
656 | struct btrfs_root *root, | |
657 | struct btrfs_path *path, | |
658 | u64 bytenr, u64 parent, | |
659 | u64 root_objectid, | |
660 | u64 owner, u64 offset) | |
661 | { | |
662 | struct btrfs_key key; | |
663 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 664 | struct extent_buffer *leaf; |
5d4f98a2 | 665 | u32 nritems; |
74493f7a | 666 | int ret; |
5d4f98a2 YZ |
667 | int recow; |
668 | int err = -ENOENT; | |
74493f7a | 669 | |
31840ae1 | 670 | key.objectid = bytenr; |
5d4f98a2 YZ |
671 | if (parent) { |
672 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
673 | key.offset = parent; | |
674 | } else { | |
675 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
676 | key.offset = hash_extent_data_ref(root_objectid, | |
677 | owner, offset); | |
678 | } | |
679 | again: | |
680 | recow = 0; | |
681 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
682 | if (ret < 0) { | |
683 | err = ret; | |
684 | goto fail; | |
685 | } | |
31840ae1 | 686 | |
5d4f98a2 YZ |
687 | if (parent) { |
688 | if (!ret) | |
689 | return 0; | |
690 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
691 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
692 | btrfs_release_path(root, path); | |
693 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
694 | if (ret < 0) { | |
695 | err = ret; | |
696 | goto fail; | |
697 | } | |
698 | if (!ret) | |
699 | return 0; | |
700 | #endif | |
701 | goto fail; | |
31840ae1 ZY |
702 | } |
703 | ||
704 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
705 | nritems = btrfs_header_nritems(leaf); |
706 | while (1) { | |
707 | if (path->slots[0] >= nritems) { | |
708 | ret = btrfs_next_leaf(root, path); | |
709 | if (ret < 0) | |
710 | err = ret; | |
711 | if (ret) | |
712 | goto fail; | |
713 | ||
714 | leaf = path->nodes[0]; | |
715 | nritems = btrfs_header_nritems(leaf); | |
716 | recow = 1; | |
717 | } | |
718 | ||
719 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
720 | if (key.objectid != bytenr || | |
721 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
722 | goto fail; | |
723 | ||
724 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
725 | struct btrfs_extent_data_ref); | |
726 | ||
727 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
728 | owner, offset)) { | |
729 | if (recow) { | |
730 | btrfs_release_path(root, path); | |
731 | goto again; | |
732 | } | |
733 | err = 0; | |
734 | break; | |
735 | } | |
736 | path->slots[0]++; | |
31840ae1 | 737 | } |
5d4f98a2 YZ |
738 | fail: |
739 | return err; | |
31840ae1 ZY |
740 | } |
741 | ||
5d4f98a2 YZ |
742 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
743 | struct btrfs_root *root, | |
744 | struct btrfs_path *path, | |
745 | u64 bytenr, u64 parent, | |
746 | u64 root_objectid, u64 owner, | |
747 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
748 | { |
749 | struct btrfs_key key; | |
750 | struct extent_buffer *leaf; | |
5d4f98a2 | 751 | u32 size; |
31840ae1 ZY |
752 | u32 num_refs; |
753 | int ret; | |
74493f7a | 754 | |
74493f7a | 755 | key.objectid = bytenr; |
5d4f98a2 YZ |
756 | if (parent) { |
757 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
758 | key.offset = parent; | |
759 | size = sizeof(struct btrfs_shared_data_ref); | |
760 | } else { | |
761 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
762 | key.offset = hash_extent_data_ref(root_objectid, | |
763 | owner, offset); | |
764 | size = sizeof(struct btrfs_extent_data_ref); | |
765 | } | |
74493f7a | 766 | |
5d4f98a2 YZ |
767 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
768 | if (ret && ret != -EEXIST) | |
769 | goto fail; | |
770 | ||
771 | leaf = path->nodes[0]; | |
772 | if (parent) { | |
773 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 774 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
775 | struct btrfs_shared_data_ref); |
776 | if (ret == 0) { | |
777 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
778 | } else { | |
779 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
780 | num_refs += refs_to_add; | |
781 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 782 | } |
5d4f98a2 YZ |
783 | } else { |
784 | struct btrfs_extent_data_ref *ref; | |
785 | while (ret == -EEXIST) { | |
786 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
787 | struct btrfs_extent_data_ref); | |
788 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
789 | owner, offset)) | |
790 | break; | |
791 | btrfs_release_path(root, path); | |
792 | key.offset++; | |
793 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
794 | size); | |
795 | if (ret && ret != -EEXIST) | |
796 | goto fail; | |
31840ae1 | 797 | |
5d4f98a2 YZ |
798 | leaf = path->nodes[0]; |
799 | } | |
800 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
801 | struct btrfs_extent_data_ref); | |
802 | if (ret == 0) { | |
803 | btrfs_set_extent_data_ref_root(leaf, ref, | |
804 | root_objectid); | |
805 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
806 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
807 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
808 | } else { | |
809 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
810 | num_refs += refs_to_add; | |
811 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 812 | } |
31840ae1 | 813 | } |
5d4f98a2 YZ |
814 | btrfs_mark_buffer_dirty(leaf); |
815 | ret = 0; | |
816 | fail: | |
7bb86316 CM |
817 | btrfs_release_path(root, path); |
818 | return ret; | |
74493f7a CM |
819 | } |
820 | ||
5d4f98a2 YZ |
821 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
822 | struct btrfs_root *root, | |
823 | struct btrfs_path *path, | |
824 | int refs_to_drop) | |
31840ae1 | 825 | { |
5d4f98a2 YZ |
826 | struct btrfs_key key; |
827 | struct btrfs_extent_data_ref *ref1 = NULL; | |
828 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 829 | struct extent_buffer *leaf; |
5d4f98a2 | 830 | u32 num_refs = 0; |
31840ae1 ZY |
831 | int ret = 0; |
832 | ||
833 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
834 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
835 | ||
836 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
837 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
838 | struct btrfs_extent_data_ref); | |
839 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
840 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
841 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
842 | struct btrfs_shared_data_ref); | |
843 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
844 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
845 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
846 | struct btrfs_extent_ref_v0 *ref0; | |
847 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
848 | struct btrfs_extent_ref_v0); | |
849 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
850 | #endif | |
851 | } else { | |
852 | BUG(); | |
853 | } | |
854 | ||
56bec294 CM |
855 | BUG_ON(num_refs < refs_to_drop); |
856 | num_refs -= refs_to_drop; | |
5d4f98a2 | 857 | |
31840ae1 ZY |
858 | if (num_refs == 0) { |
859 | ret = btrfs_del_item(trans, root, path); | |
860 | } else { | |
5d4f98a2 YZ |
861 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
862 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
863 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
864 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
865 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
866 | else { | |
867 | struct btrfs_extent_ref_v0 *ref0; | |
868 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
869 | struct btrfs_extent_ref_v0); | |
870 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
871 | } | |
872 | #endif | |
31840ae1 ZY |
873 | btrfs_mark_buffer_dirty(leaf); |
874 | } | |
31840ae1 ZY |
875 | return ret; |
876 | } | |
877 | ||
5d4f98a2 YZ |
878 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
879 | struct btrfs_path *path, | |
880 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 881 | { |
5d4f98a2 YZ |
882 | struct btrfs_key key; |
883 | struct extent_buffer *leaf; | |
884 | struct btrfs_extent_data_ref *ref1; | |
885 | struct btrfs_shared_data_ref *ref2; | |
886 | u32 num_refs = 0; | |
887 | ||
888 | leaf = path->nodes[0]; | |
889 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
890 | if (iref) { | |
891 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
892 | BTRFS_EXTENT_DATA_REF_KEY) { | |
893 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
894 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
895 | } else { | |
896 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
897 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
898 | } | |
899 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
900 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
901 | struct btrfs_extent_data_ref); | |
902 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
903 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
904 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
905 | struct btrfs_shared_data_ref); | |
906 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
907 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
908 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
909 | struct btrfs_extent_ref_v0 *ref0; | |
910 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
911 | struct btrfs_extent_ref_v0); | |
912 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 913 | #endif |
5d4f98a2 YZ |
914 | } else { |
915 | WARN_ON(1); | |
916 | } | |
917 | return num_refs; | |
918 | } | |
15916de8 | 919 | |
5d4f98a2 YZ |
920 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
921 | struct btrfs_root *root, | |
922 | struct btrfs_path *path, | |
923 | u64 bytenr, u64 parent, | |
924 | u64 root_objectid) | |
1f3c79a2 | 925 | { |
5d4f98a2 | 926 | struct btrfs_key key; |
1f3c79a2 | 927 | int ret; |
1f3c79a2 | 928 | |
5d4f98a2 YZ |
929 | key.objectid = bytenr; |
930 | if (parent) { | |
931 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
932 | key.offset = parent; | |
933 | } else { | |
934 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
935 | key.offset = root_objectid; | |
1f3c79a2 LH |
936 | } |
937 | ||
5d4f98a2 YZ |
938 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
939 | if (ret > 0) | |
940 | ret = -ENOENT; | |
941 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
942 | if (ret == -ENOENT && parent) { | |
943 | btrfs_release_path(root, path); | |
944 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
945 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
946 | if (ret > 0) | |
947 | ret = -ENOENT; | |
948 | } | |
1f3c79a2 | 949 | #endif |
5d4f98a2 | 950 | return ret; |
1f3c79a2 LH |
951 | } |
952 | ||
5d4f98a2 YZ |
953 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
954 | struct btrfs_root *root, | |
955 | struct btrfs_path *path, | |
956 | u64 bytenr, u64 parent, | |
957 | u64 root_objectid) | |
31840ae1 | 958 | { |
5d4f98a2 | 959 | struct btrfs_key key; |
31840ae1 | 960 | int ret; |
31840ae1 | 961 | |
5d4f98a2 YZ |
962 | key.objectid = bytenr; |
963 | if (parent) { | |
964 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
965 | key.offset = parent; | |
966 | } else { | |
967 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
968 | key.offset = root_objectid; | |
969 | } | |
970 | ||
971 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
972 | btrfs_release_path(root, path); | |
31840ae1 ZY |
973 | return ret; |
974 | } | |
975 | ||
5d4f98a2 | 976 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 977 | { |
5d4f98a2 YZ |
978 | int type; |
979 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
980 | if (parent > 0) | |
981 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
982 | else | |
983 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
984 | } else { | |
985 | if (parent > 0) | |
986 | type = BTRFS_SHARED_DATA_REF_KEY; | |
987 | else | |
988 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
989 | } | |
990 | return type; | |
31840ae1 | 991 | } |
56bec294 | 992 | |
5d4f98a2 | 993 | static int find_next_key(struct btrfs_path *path, struct btrfs_key *key) |
56bec294 | 994 | |
02217ed2 | 995 | { |
5d4f98a2 YZ |
996 | int level; |
997 | BUG_ON(!path->keep_locks); | |
998 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
999 | if (!path->nodes[level]) | |
1000 | break; | |
1001 | btrfs_assert_tree_locked(path->nodes[level]); | |
1002 | if (path->slots[level] + 1 >= | |
1003 | btrfs_header_nritems(path->nodes[level])) | |
1004 | continue; | |
1005 | if (level == 0) | |
1006 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1007 | path->slots[level] + 1); | |
1008 | else | |
1009 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1010 | path->slots[level] + 1); | |
1011 | return 0; | |
1012 | } | |
1013 | return 1; | |
1014 | } | |
037e6390 | 1015 | |
5d4f98a2 YZ |
1016 | /* |
1017 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1018 | * to the address of inline back ref, and 0 is returned. | |
1019 | * | |
1020 | * if back ref isn't found, *ref_ret is set to the address where it | |
1021 | * should be inserted, and -ENOENT is returned. | |
1022 | * | |
1023 | * if insert is true and there are too many inline back refs, the path | |
1024 | * points to the extent item, and -EAGAIN is returned. | |
1025 | * | |
1026 | * NOTE: inline back refs are ordered in the same way that back ref | |
1027 | * items in the tree are ordered. | |
1028 | */ | |
1029 | static noinline_for_stack | |
1030 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1031 | struct btrfs_root *root, | |
1032 | struct btrfs_path *path, | |
1033 | struct btrfs_extent_inline_ref **ref_ret, | |
1034 | u64 bytenr, u64 num_bytes, | |
1035 | u64 parent, u64 root_objectid, | |
1036 | u64 owner, u64 offset, int insert) | |
1037 | { | |
1038 | struct btrfs_key key; | |
1039 | struct extent_buffer *leaf; | |
1040 | struct btrfs_extent_item *ei; | |
1041 | struct btrfs_extent_inline_ref *iref; | |
1042 | u64 flags; | |
1043 | u64 item_size; | |
1044 | unsigned long ptr; | |
1045 | unsigned long end; | |
1046 | int extra_size; | |
1047 | int type; | |
1048 | int want; | |
1049 | int ret; | |
1050 | int err = 0; | |
26b8003f | 1051 | |
db94535d | 1052 | key.objectid = bytenr; |
31840ae1 | 1053 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1054 | key.offset = num_bytes; |
31840ae1 | 1055 | |
5d4f98a2 YZ |
1056 | want = extent_ref_type(parent, owner); |
1057 | if (insert) { | |
1058 | extra_size = btrfs_extent_inline_ref_size(want); | |
1059 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) | |
1060 | path->keep_locks = 1; | |
1061 | } else | |
1062 | extra_size = -1; | |
1063 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1064 | if (ret < 0) { |
5d4f98a2 YZ |
1065 | err = ret; |
1066 | goto out; | |
1067 | } | |
1068 | BUG_ON(ret); | |
1069 | ||
1070 | leaf = path->nodes[0]; | |
1071 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1072 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1073 | if (item_size < sizeof(*ei)) { | |
1074 | if (!insert) { | |
1075 | err = -ENOENT; | |
1076 | goto out; | |
1077 | } | |
1078 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1079 | extra_size); | |
1080 | if (ret < 0) { | |
1081 | err = ret; | |
1082 | goto out; | |
1083 | } | |
1084 | leaf = path->nodes[0]; | |
1085 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1086 | } | |
1087 | #endif | |
1088 | BUG_ON(item_size < sizeof(*ei)); | |
1089 | ||
1090 | if (owner < BTRFS_FIRST_FREE_OBJECTID && insert && | |
1091 | item_size + extra_size >= BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1092 | err = -EAGAIN; | |
1093 | goto out; | |
1094 | } | |
1095 | ||
1096 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1097 | flags = btrfs_extent_flags(leaf, ei); | |
1098 | ||
1099 | ptr = (unsigned long)(ei + 1); | |
1100 | end = (unsigned long)ei + item_size; | |
1101 | ||
1102 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1103 | ptr += sizeof(struct btrfs_tree_block_info); | |
1104 | BUG_ON(ptr > end); | |
1105 | } else { | |
1106 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1107 | } | |
1108 | ||
1109 | err = -ENOENT; | |
1110 | while (1) { | |
1111 | if (ptr >= end) { | |
1112 | WARN_ON(ptr > end); | |
1113 | break; | |
1114 | } | |
1115 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1116 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1117 | if (want < type) | |
1118 | break; | |
1119 | if (want > type) { | |
1120 | ptr += btrfs_extent_inline_ref_size(type); | |
1121 | continue; | |
1122 | } | |
1123 | ||
1124 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1125 | struct btrfs_extent_data_ref *dref; | |
1126 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1127 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1128 | owner, offset)) { | |
1129 | err = 0; | |
1130 | break; | |
1131 | } | |
1132 | if (hash_extent_data_ref_item(leaf, dref) < | |
1133 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1134 | break; | |
1135 | } else { | |
1136 | u64 ref_offset; | |
1137 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1138 | if (parent > 0) { | |
1139 | if (parent == ref_offset) { | |
1140 | err = 0; | |
1141 | break; | |
1142 | } | |
1143 | if (ref_offset < parent) | |
1144 | break; | |
1145 | } else { | |
1146 | if (root_objectid == ref_offset) { | |
1147 | err = 0; | |
1148 | break; | |
1149 | } | |
1150 | if (ref_offset < root_objectid) | |
1151 | break; | |
1152 | } | |
1153 | } | |
1154 | ptr += btrfs_extent_inline_ref_size(type); | |
1155 | } | |
1156 | if (err == -ENOENT && insert) { | |
1157 | if (item_size + extra_size >= | |
1158 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1159 | err = -EAGAIN; | |
1160 | goto out; | |
1161 | } | |
1162 | /* | |
1163 | * To add new inline back ref, we have to make sure | |
1164 | * there is no corresponding back ref item. | |
1165 | * For simplicity, we just do not add new inline back | |
1166 | * ref if there is any kind of item for this block | |
1167 | */ | |
1168 | if (owner >= BTRFS_FIRST_FREE_OBJECTID && | |
1169 | find_next_key(path, &key) == 0 && key.objectid == bytenr) { | |
1170 | err = -EAGAIN; | |
1171 | goto out; | |
1172 | } | |
1173 | } | |
1174 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1175 | out: | |
1176 | if (insert && owner >= BTRFS_FIRST_FREE_OBJECTID) { | |
1177 | path->keep_locks = 0; | |
1178 | btrfs_unlock_up_safe(path, 1); | |
1179 | } | |
1180 | return err; | |
1181 | } | |
1182 | ||
1183 | /* | |
1184 | * helper to add new inline back ref | |
1185 | */ | |
1186 | static noinline_for_stack | |
1187 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1188 | struct btrfs_root *root, | |
1189 | struct btrfs_path *path, | |
1190 | struct btrfs_extent_inline_ref *iref, | |
1191 | u64 parent, u64 root_objectid, | |
1192 | u64 owner, u64 offset, int refs_to_add, | |
1193 | struct btrfs_delayed_extent_op *extent_op) | |
1194 | { | |
1195 | struct extent_buffer *leaf; | |
1196 | struct btrfs_extent_item *ei; | |
1197 | unsigned long ptr; | |
1198 | unsigned long end; | |
1199 | unsigned long item_offset; | |
1200 | u64 refs; | |
1201 | int size; | |
1202 | int type; | |
1203 | int ret; | |
1204 | ||
1205 | leaf = path->nodes[0]; | |
1206 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1207 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1208 | ||
1209 | type = extent_ref_type(parent, owner); | |
1210 | size = btrfs_extent_inline_ref_size(type); | |
1211 | ||
1212 | ret = btrfs_extend_item(trans, root, path, size); | |
1213 | BUG_ON(ret); | |
1214 | ||
1215 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1216 | refs = btrfs_extent_refs(leaf, ei); | |
1217 | refs += refs_to_add; | |
1218 | btrfs_set_extent_refs(leaf, ei, refs); | |
1219 | if (extent_op) | |
1220 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1221 | ||
1222 | ptr = (unsigned long)ei + item_offset; | |
1223 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1224 | if (ptr < end - size) | |
1225 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1226 | end - size - ptr); | |
1227 | ||
1228 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1229 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1230 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1231 | struct btrfs_extent_data_ref *dref; | |
1232 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1233 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1234 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1235 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1236 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1237 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1238 | struct btrfs_shared_data_ref *sref; | |
1239 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1240 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1241 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1242 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1243 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1244 | } else { | |
1245 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1246 | } | |
1247 | btrfs_mark_buffer_dirty(leaf); | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1252 | struct btrfs_root *root, | |
1253 | struct btrfs_path *path, | |
1254 | struct btrfs_extent_inline_ref **ref_ret, | |
1255 | u64 bytenr, u64 num_bytes, u64 parent, | |
1256 | u64 root_objectid, u64 owner, u64 offset) | |
1257 | { | |
1258 | int ret; | |
1259 | ||
1260 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1261 | bytenr, num_bytes, parent, | |
1262 | root_objectid, owner, offset, 0); | |
1263 | if (ret != -ENOENT) | |
54aa1f4d | 1264 | return ret; |
5d4f98a2 YZ |
1265 | |
1266 | btrfs_release_path(root, path); | |
1267 | *ref_ret = NULL; | |
1268 | ||
1269 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1270 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1271 | root_objectid); | |
1272 | } else { | |
1273 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1274 | root_objectid, owner, offset); | |
b9473439 | 1275 | } |
5d4f98a2 YZ |
1276 | return ret; |
1277 | } | |
31840ae1 | 1278 | |
5d4f98a2 YZ |
1279 | /* |
1280 | * helper to update/remove inline back ref | |
1281 | */ | |
1282 | static noinline_for_stack | |
1283 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1284 | struct btrfs_root *root, | |
1285 | struct btrfs_path *path, | |
1286 | struct btrfs_extent_inline_ref *iref, | |
1287 | int refs_to_mod, | |
1288 | struct btrfs_delayed_extent_op *extent_op) | |
1289 | { | |
1290 | struct extent_buffer *leaf; | |
1291 | struct btrfs_extent_item *ei; | |
1292 | struct btrfs_extent_data_ref *dref = NULL; | |
1293 | struct btrfs_shared_data_ref *sref = NULL; | |
1294 | unsigned long ptr; | |
1295 | unsigned long end; | |
1296 | u32 item_size; | |
1297 | int size; | |
1298 | int type; | |
1299 | int ret; | |
1300 | u64 refs; | |
1301 | ||
1302 | leaf = path->nodes[0]; | |
1303 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1304 | refs = btrfs_extent_refs(leaf, ei); | |
1305 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1306 | refs += refs_to_mod; | |
1307 | btrfs_set_extent_refs(leaf, ei, refs); | |
1308 | if (extent_op) | |
1309 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1310 | ||
1311 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1312 | ||
1313 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1314 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1315 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1316 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1317 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1318 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1319 | } else { | |
1320 | refs = 1; | |
1321 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1322 | } |
31840ae1 | 1323 | |
5d4f98a2 YZ |
1324 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1325 | refs += refs_to_mod; | |
1326 | ||
1327 | if (refs > 0) { | |
1328 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1329 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1330 | else | |
1331 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1332 | } else { | |
1333 | size = btrfs_extent_inline_ref_size(type); | |
1334 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1335 | ptr = (unsigned long)iref; | |
1336 | end = (unsigned long)ei + item_size; | |
1337 | if (ptr + size < end) | |
1338 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1339 | end - ptr - size); | |
1340 | item_size -= size; | |
1341 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
1342 | BUG_ON(ret); | |
1343 | } | |
1344 | btrfs_mark_buffer_dirty(leaf); | |
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | static noinline_for_stack | |
1349 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1350 | struct btrfs_root *root, | |
1351 | struct btrfs_path *path, | |
1352 | u64 bytenr, u64 num_bytes, u64 parent, | |
1353 | u64 root_objectid, u64 owner, | |
1354 | u64 offset, int refs_to_add, | |
1355 | struct btrfs_delayed_extent_op *extent_op) | |
1356 | { | |
1357 | struct btrfs_extent_inline_ref *iref; | |
1358 | int ret; | |
1359 | ||
1360 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1361 | bytenr, num_bytes, parent, | |
1362 | root_objectid, owner, offset, 1); | |
1363 | if (ret == 0) { | |
1364 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1365 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1366 | refs_to_add, extent_op); | |
1367 | } else if (ret == -ENOENT) { | |
1368 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1369 | parent, root_objectid, | |
1370 | owner, offset, refs_to_add, | |
1371 | extent_op); | |
771ed689 | 1372 | } |
5d4f98a2 YZ |
1373 | return ret; |
1374 | } | |
31840ae1 | 1375 | |
5d4f98a2 YZ |
1376 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1377 | struct btrfs_root *root, | |
1378 | struct btrfs_path *path, | |
1379 | u64 bytenr, u64 parent, u64 root_objectid, | |
1380 | u64 owner, u64 offset, int refs_to_add) | |
1381 | { | |
1382 | int ret; | |
1383 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1384 | BUG_ON(refs_to_add != 1); | |
1385 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1386 | parent, root_objectid); | |
1387 | } else { | |
1388 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1389 | parent, root_objectid, | |
1390 | owner, offset, refs_to_add); | |
1391 | } | |
1392 | return ret; | |
1393 | } | |
56bec294 | 1394 | |
5d4f98a2 YZ |
1395 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1396 | struct btrfs_root *root, | |
1397 | struct btrfs_path *path, | |
1398 | struct btrfs_extent_inline_ref *iref, | |
1399 | int refs_to_drop, int is_data) | |
1400 | { | |
1401 | int ret; | |
b9473439 | 1402 | |
5d4f98a2 YZ |
1403 | BUG_ON(!is_data && refs_to_drop != 1); |
1404 | if (iref) { | |
1405 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1406 | -refs_to_drop, NULL); | |
1407 | } else if (is_data) { | |
1408 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1409 | } else { | |
1410 | ret = btrfs_del_item(trans, root, path); | |
1411 | } | |
1412 | return ret; | |
1413 | } | |
1414 | ||
1415 | #ifdef BIO_RW_DISCARD | |
1416 | static void btrfs_issue_discard(struct block_device *bdev, | |
1417 | u64 start, u64 len) | |
1418 | { | |
1419 | blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL); | |
1420 | } | |
1421 | #endif | |
1422 | ||
1423 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
1424 | u64 num_bytes) | |
1425 | { | |
1426 | #ifdef BIO_RW_DISCARD | |
1427 | int ret; | |
1428 | u64 map_length = num_bytes; | |
1429 | struct btrfs_multi_bio *multi = NULL; | |
1430 | ||
1431 | /* Tell the block device(s) that the sectors can be discarded */ | |
1432 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1433 | bytenr, &map_length, &multi, 0); | |
1434 | if (!ret) { | |
1435 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1436 | int i; | |
1437 | ||
1438 | if (map_length > num_bytes) | |
1439 | map_length = num_bytes; | |
1440 | ||
1441 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1442 | btrfs_issue_discard(stripe->dev->bdev, | |
1443 | stripe->physical, | |
1444 | map_length); | |
1445 | } | |
1446 | kfree(multi); | |
1447 | } | |
1448 | ||
1449 | return ret; | |
1450 | #else | |
1451 | return 0; | |
1452 | #endif | |
1453 | } | |
1454 | ||
1455 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1456 | struct btrfs_root *root, | |
1457 | u64 bytenr, u64 num_bytes, u64 parent, | |
1458 | u64 root_objectid, u64 owner, u64 offset) | |
1459 | { | |
1460 | int ret; | |
1461 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1462 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1463 | ||
1464 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1465 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1466 | parent, root_objectid, (int)owner, | |
1467 | BTRFS_ADD_DELAYED_REF, NULL); | |
1468 | } else { | |
1469 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1470 | parent, root_objectid, owner, offset, | |
1471 | BTRFS_ADD_DELAYED_REF, NULL); | |
1472 | } | |
1473 | return ret; | |
1474 | } | |
1475 | ||
1476 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1477 | struct btrfs_root *root, | |
1478 | u64 bytenr, u64 num_bytes, | |
1479 | u64 parent, u64 root_objectid, | |
1480 | u64 owner, u64 offset, int refs_to_add, | |
1481 | struct btrfs_delayed_extent_op *extent_op) | |
1482 | { | |
1483 | struct btrfs_path *path; | |
1484 | struct extent_buffer *leaf; | |
1485 | struct btrfs_extent_item *item; | |
1486 | u64 refs; | |
1487 | int ret; | |
1488 | int err = 0; | |
1489 | ||
1490 | path = btrfs_alloc_path(); | |
1491 | if (!path) | |
1492 | return -ENOMEM; | |
1493 | ||
1494 | path->reada = 1; | |
1495 | path->leave_spinning = 1; | |
1496 | /* this will setup the path even if it fails to insert the back ref */ | |
1497 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1498 | path, bytenr, num_bytes, parent, | |
1499 | root_objectid, owner, offset, | |
1500 | refs_to_add, extent_op); | |
1501 | if (ret == 0) | |
1502 | goto out; | |
1503 | ||
1504 | if (ret != -EAGAIN) { | |
1505 | err = ret; | |
1506 | goto out; | |
1507 | } | |
1508 | ||
1509 | leaf = path->nodes[0]; | |
1510 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1511 | refs = btrfs_extent_refs(leaf, item); | |
1512 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1513 | if (extent_op) | |
1514 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1515 | |
5d4f98a2 | 1516 | btrfs_mark_buffer_dirty(leaf); |
56bec294 CM |
1517 | btrfs_release_path(root->fs_info->extent_root, path); |
1518 | ||
1519 | path->reada = 1; | |
b9473439 CM |
1520 | path->leave_spinning = 1; |
1521 | ||
56bec294 CM |
1522 | /* now insert the actual backref */ |
1523 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1524 | path, bytenr, parent, root_objectid, |
1525 | owner, offset, refs_to_add); | |
56bec294 | 1526 | BUG_ON(ret); |
5d4f98a2 | 1527 | out: |
56bec294 | 1528 | btrfs_free_path(path); |
5d4f98a2 | 1529 | return err; |
56bec294 CM |
1530 | } |
1531 | ||
5d4f98a2 YZ |
1532 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1533 | struct btrfs_root *root, | |
1534 | struct btrfs_delayed_ref_node *node, | |
1535 | struct btrfs_delayed_extent_op *extent_op, | |
1536 | int insert_reserved) | |
56bec294 | 1537 | { |
5d4f98a2 YZ |
1538 | int ret = 0; |
1539 | struct btrfs_delayed_data_ref *ref; | |
1540 | struct btrfs_key ins; | |
1541 | u64 parent = 0; | |
1542 | u64 ref_root = 0; | |
1543 | u64 flags = 0; | |
1544 | ||
1545 | ins.objectid = node->bytenr; | |
1546 | ins.offset = node->num_bytes; | |
1547 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1548 | ||
1549 | ref = btrfs_delayed_node_to_data_ref(node); | |
1550 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1551 | parent = ref->parent; | |
1552 | else | |
1553 | ref_root = ref->root; | |
1554 | ||
1555 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1556 | if (extent_op) { | |
1557 | BUG_ON(extent_op->update_key); | |
1558 | flags |= extent_op->flags_to_set; | |
1559 | } | |
1560 | ret = alloc_reserved_file_extent(trans, root, | |
1561 | parent, ref_root, flags, | |
1562 | ref->objectid, ref->offset, | |
1563 | &ins, node->ref_mod); | |
1564 | update_reserved_extents(root, ins.objectid, ins.offset, 0); | |
1565 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { | |
1566 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1567 | node->num_bytes, parent, | |
1568 | ref_root, ref->objectid, | |
1569 | ref->offset, node->ref_mod, | |
1570 | extent_op); | |
1571 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1572 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1573 | node->num_bytes, parent, | |
1574 | ref_root, ref->objectid, | |
1575 | ref->offset, node->ref_mod, | |
1576 | extent_op); | |
1577 | } else { | |
1578 | BUG(); | |
1579 | } | |
1580 | return ret; | |
1581 | } | |
1582 | ||
1583 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1584 | struct extent_buffer *leaf, | |
1585 | struct btrfs_extent_item *ei) | |
1586 | { | |
1587 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1588 | if (extent_op->update_flags) { | |
1589 | flags |= extent_op->flags_to_set; | |
1590 | btrfs_set_extent_flags(leaf, ei, flags); | |
1591 | } | |
1592 | ||
1593 | if (extent_op->update_key) { | |
1594 | struct btrfs_tree_block_info *bi; | |
1595 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1596 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1597 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1598 | } | |
1599 | } | |
1600 | ||
1601 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1602 | struct btrfs_root *root, | |
1603 | struct btrfs_delayed_ref_node *node, | |
1604 | struct btrfs_delayed_extent_op *extent_op) | |
1605 | { | |
1606 | struct btrfs_key key; | |
1607 | struct btrfs_path *path; | |
1608 | struct btrfs_extent_item *ei; | |
1609 | struct extent_buffer *leaf; | |
1610 | u32 item_size; | |
56bec294 | 1611 | int ret; |
5d4f98a2 YZ |
1612 | int err = 0; |
1613 | ||
1614 | path = btrfs_alloc_path(); | |
1615 | if (!path) | |
1616 | return -ENOMEM; | |
1617 | ||
1618 | key.objectid = node->bytenr; | |
1619 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1620 | key.offset = node->num_bytes; | |
1621 | ||
1622 | path->reada = 1; | |
1623 | path->leave_spinning = 1; | |
1624 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1625 | path, 0, 1); | |
1626 | if (ret < 0) { | |
1627 | err = ret; | |
1628 | goto out; | |
1629 | } | |
1630 | if (ret > 0) { | |
1631 | err = -EIO; | |
1632 | goto out; | |
1633 | } | |
1634 | ||
1635 | leaf = path->nodes[0]; | |
1636 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1637 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1638 | if (item_size < sizeof(*ei)) { | |
1639 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1640 | path, (u64)-1, 0); | |
1641 | if (ret < 0) { | |
1642 | err = ret; | |
1643 | goto out; | |
1644 | } | |
1645 | leaf = path->nodes[0]; | |
1646 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1647 | } | |
1648 | #endif | |
1649 | BUG_ON(item_size < sizeof(*ei)); | |
1650 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1651 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1652 | |
5d4f98a2 YZ |
1653 | btrfs_mark_buffer_dirty(leaf); |
1654 | out: | |
1655 | btrfs_free_path(path); | |
1656 | return err; | |
56bec294 CM |
1657 | } |
1658 | ||
5d4f98a2 YZ |
1659 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
1660 | struct btrfs_root *root, | |
1661 | struct btrfs_delayed_ref_node *node, | |
1662 | struct btrfs_delayed_extent_op *extent_op, | |
1663 | int insert_reserved) | |
56bec294 CM |
1664 | { |
1665 | int ret = 0; | |
5d4f98a2 YZ |
1666 | struct btrfs_delayed_tree_ref *ref; |
1667 | struct btrfs_key ins; | |
1668 | u64 parent = 0; | |
1669 | u64 ref_root = 0; | |
56bec294 | 1670 | |
5d4f98a2 YZ |
1671 | ins.objectid = node->bytenr; |
1672 | ins.offset = node->num_bytes; | |
1673 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 1674 | |
5d4f98a2 YZ |
1675 | ref = btrfs_delayed_node_to_tree_ref(node); |
1676 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1677 | parent = ref->parent; | |
1678 | else | |
1679 | ref_root = ref->root; | |
1680 | ||
1681 | BUG_ON(node->ref_mod != 1); | |
1682 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1683 | BUG_ON(!extent_op || !extent_op->update_flags || | |
1684 | !extent_op->update_key); | |
1685 | ret = alloc_reserved_tree_block(trans, root, | |
1686 | parent, ref_root, | |
1687 | extent_op->flags_to_set, | |
1688 | &extent_op->key, | |
1689 | ref->level, &ins); | |
1690 | update_reserved_extents(root, ins.objectid, ins.offset, 0); | |
1691 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { | |
1692 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1693 | node->num_bytes, parent, ref_root, | |
1694 | ref->level, 0, 1, extent_op); | |
1695 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1696 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1697 | node->num_bytes, parent, ref_root, | |
1698 | ref->level, 0, 1, extent_op); | |
1699 | } else { | |
1700 | BUG(); | |
1701 | } | |
56bec294 CM |
1702 | return ret; |
1703 | } | |
1704 | ||
5d4f98a2 | 1705 | |
56bec294 | 1706 | /* helper function to actually process a single delayed ref entry */ |
5d4f98a2 YZ |
1707 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
1708 | struct btrfs_root *root, | |
1709 | struct btrfs_delayed_ref_node *node, | |
1710 | struct btrfs_delayed_extent_op *extent_op, | |
1711 | int insert_reserved) | |
56bec294 CM |
1712 | { |
1713 | int ret; | |
5d4f98a2 | 1714 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
1715 | struct btrfs_delayed_ref_head *head; |
1716 | /* | |
1717 | * we've hit the end of the chain and we were supposed | |
1718 | * to insert this extent into the tree. But, it got | |
1719 | * deleted before we ever needed to insert it, so all | |
1720 | * we have to do is clean up the accounting | |
1721 | */ | |
5d4f98a2 YZ |
1722 | BUG_ON(extent_op); |
1723 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 1724 | if (insert_reserved) { |
5d4f98a2 YZ |
1725 | if (head->is_data) { |
1726 | ret = btrfs_del_csums(trans, root, | |
1727 | node->bytenr, | |
1728 | node->num_bytes); | |
1729 | BUG_ON(ret); | |
1730 | } | |
1731 | btrfs_update_pinned_extents(root, node->bytenr, | |
1732 | node->num_bytes, 1); | |
56bec294 CM |
1733 | update_reserved_extents(root, node->bytenr, |
1734 | node->num_bytes, 0); | |
1735 | } | |
56bec294 CM |
1736 | mutex_unlock(&head->mutex); |
1737 | return 0; | |
1738 | } | |
1739 | ||
5d4f98a2 YZ |
1740 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1741 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1742 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
1743 | insert_reserved); | |
1744 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1745 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1746 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
1747 | insert_reserved); | |
1748 | else | |
1749 | BUG(); | |
1750 | return ret; | |
56bec294 CM |
1751 | } |
1752 | ||
1753 | static noinline struct btrfs_delayed_ref_node * | |
1754 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
1755 | { | |
1756 | struct rb_node *node; | |
1757 | struct btrfs_delayed_ref_node *ref; | |
1758 | int action = BTRFS_ADD_DELAYED_REF; | |
1759 | again: | |
1760 | /* | |
1761 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1762 | * this prevents ref count from going down to zero when | |
1763 | * there still are pending delayed ref. | |
1764 | */ | |
1765 | node = rb_prev(&head->node.rb_node); | |
1766 | while (1) { | |
1767 | if (!node) | |
1768 | break; | |
1769 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
1770 | rb_node); | |
1771 | if (ref->bytenr != head->node.bytenr) | |
1772 | break; | |
5d4f98a2 | 1773 | if (ref->action == action) |
56bec294 CM |
1774 | return ref; |
1775 | node = rb_prev(node); | |
1776 | } | |
1777 | if (action == BTRFS_ADD_DELAYED_REF) { | |
1778 | action = BTRFS_DROP_DELAYED_REF; | |
1779 | goto again; | |
1780 | } | |
1781 | return NULL; | |
1782 | } | |
1783 | ||
c3e69d58 CM |
1784 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
1785 | struct btrfs_root *root, | |
1786 | struct list_head *cluster) | |
56bec294 | 1787 | { |
56bec294 CM |
1788 | struct btrfs_delayed_ref_root *delayed_refs; |
1789 | struct btrfs_delayed_ref_node *ref; | |
1790 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 1791 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 1792 | int ret; |
c3e69d58 | 1793 | int count = 0; |
56bec294 | 1794 | int must_insert_reserved = 0; |
56bec294 CM |
1795 | |
1796 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
1797 | while (1) { |
1798 | if (!locked_ref) { | |
c3e69d58 CM |
1799 | /* pick a new head ref from the cluster list */ |
1800 | if (list_empty(cluster)) | |
56bec294 | 1801 | break; |
56bec294 | 1802 | |
c3e69d58 CM |
1803 | locked_ref = list_entry(cluster->next, |
1804 | struct btrfs_delayed_ref_head, cluster); | |
1805 | ||
1806 | /* grab the lock that says we are going to process | |
1807 | * all the refs for this head */ | |
1808 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
1809 | ||
1810 | /* | |
1811 | * we may have dropped the spin lock to get the head | |
1812 | * mutex lock, and that might have given someone else | |
1813 | * time to free the head. If that's true, it has been | |
1814 | * removed from our list and we can move on. | |
1815 | */ | |
1816 | if (ret == -EAGAIN) { | |
1817 | locked_ref = NULL; | |
1818 | count++; | |
1819 | continue; | |
56bec294 CM |
1820 | } |
1821 | } | |
a28ec197 | 1822 | |
56bec294 CM |
1823 | /* |
1824 | * record the must insert reserved flag before we | |
1825 | * drop the spin lock. | |
1826 | */ | |
1827 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1828 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 1829 | |
5d4f98a2 YZ |
1830 | extent_op = locked_ref->extent_op; |
1831 | locked_ref->extent_op = NULL; | |
1832 | ||
56bec294 CM |
1833 | /* |
1834 | * locked_ref is the head node, so we have to go one | |
1835 | * node back for any delayed ref updates | |
1836 | */ | |
56bec294 CM |
1837 | ref = select_delayed_ref(locked_ref); |
1838 | if (!ref) { | |
1839 | /* All delayed refs have been processed, Go ahead | |
1840 | * and send the head node to run_one_delayed_ref, | |
1841 | * so that any accounting fixes can happen | |
1842 | */ | |
1843 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
1844 | |
1845 | if (extent_op && must_insert_reserved) { | |
1846 | kfree(extent_op); | |
1847 | extent_op = NULL; | |
1848 | } | |
1849 | ||
1850 | if (extent_op) { | |
1851 | spin_unlock(&delayed_refs->lock); | |
1852 | ||
1853 | ret = run_delayed_extent_op(trans, root, | |
1854 | ref, extent_op); | |
1855 | BUG_ON(ret); | |
1856 | kfree(extent_op); | |
1857 | ||
1858 | cond_resched(); | |
1859 | spin_lock(&delayed_refs->lock); | |
1860 | continue; | |
1861 | } | |
1862 | ||
c3e69d58 | 1863 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
1864 | locked_ref = NULL; |
1865 | } | |
02217ed2 | 1866 | |
56bec294 CM |
1867 | ref->in_tree = 0; |
1868 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
1869 | delayed_refs->num_entries--; | |
5d4f98a2 | 1870 | |
56bec294 | 1871 | spin_unlock(&delayed_refs->lock); |
925baedd | 1872 | |
5d4f98a2 | 1873 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
1874 | must_insert_reserved); |
1875 | BUG_ON(ret); | |
eb099670 | 1876 | |
5d4f98a2 YZ |
1877 | btrfs_put_delayed_ref(ref); |
1878 | kfree(extent_op); | |
c3e69d58 | 1879 | count++; |
5d4f98a2 | 1880 | |
c3e69d58 CM |
1881 | cond_resched(); |
1882 | spin_lock(&delayed_refs->lock); | |
1883 | } | |
1884 | return count; | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * this starts processing the delayed reference count updates and | |
1889 | * extent insertions we have queued up so far. count can be | |
1890 | * 0, which means to process everything in the tree at the start | |
1891 | * of the run (but not newly added entries), or it can be some target | |
1892 | * number you'd like to process. | |
1893 | */ | |
1894 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
1895 | struct btrfs_root *root, unsigned long count) | |
1896 | { | |
1897 | struct rb_node *node; | |
1898 | struct btrfs_delayed_ref_root *delayed_refs; | |
1899 | struct btrfs_delayed_ref_node *ref; | |
1900 | struct list_head cluster; | |
1901 | int ret; | |
1902 | int run_all = count == (unsigned long)-1; | |
1903 | int run_most = 0; | |
1904 | ||
1905 | if (root == root->fs_info->extent_root) | |
1906 | root = root->fs_info->tree_root; | |
1907 | ||
1908 | delayed_refs = &trans->transaction->delayed_refs; | |
1909 | INIT_LIST_HEAD(&cluster); | |
1910 | again: | |
1911 | spin_lock(&delayed_refs->lock); | |
1912 | if (count == 0) { | |
1913 | count = delayed_refs->num_entries * 2; | |
1914 | run_most = 1; | |
1915 | } | |
1916 | while (1) { | |
1917 | if (!(run_all || run_most) && | |
1918 | delayed_refs->num_heads_ready < 64) | |
1919 | break; | |
eb099670 | 1920 | |
56bec294 | 1921 | /* |
c3e69d58 CM |
1922 | * go find something we can process in the rbtree. We start at |
1923 | * the beginning of the tree, and then build a cluster | |
1924 | * of refs to process starting at the first one we are able to | |
1925 | * lock | |
56bec294 | 1926 | */ |
c3e69d58 CM |
1927 | ret = btrfs_find_ref_cluster(trans, &cluster, |
1928 | delayed_refs->run_delayed_start); | |
1929 | if (ret) | |
56bec294 CM |
1930 | break; |
1931 | ||
c3e69d58 CM |
1932 | ret = run_clustered_refs(trans, root, &cluster); |
1933 | BUG_ON(ret < 0); | |
1934 | ||
1935 | count -= min_t(unsigned long, ret, count); | |
1936 | ||
1937 | if (count == 0) | |
1938 | break; | |
eb099670 | 1939 | } |
c3e69d58 | 1940 | |
56bec294 | 1941 | if (run_all) { |
56bec294 | 1942 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 1943 | if (!node) |
56bec294 | 1944 | goto out; |
c3e69d58 | 1945 | count = (unsigned long)-1; |
e9d0b13b | 1946 | |
56bec294 CM |
1947 | while (node) { |
1948 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
1949 | rb_node); | |
1950 | if (btrfs_delayed_ref_is_head(ref)) { | |
1951 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 1952 | |
56bec294 CM |
1953 | head = btrfs_delayed_node_to_head(ref); |
1954 | atomic_inc(&ref->refs); | |
1955 | ||
1956 | spin_unlock(&delayed_refs->lock); | |
1957 | mutex_lock(&head->mutex); | |
1958 | mutex_unlock(&head->mutex); | |
1959 | ||
1960 | btrfs_put_delayed_ref(ref); | |
1887be66 | 1961 | cond_resched(); |
56bec294 CM |
1962 | goto again; |
1963 | } | |
1964 | node = rb_next(node); | |
1965 | } | |
1966 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
1967 | schedule_timeout(1); |
1968 | goto again; | |
5f39d397 | 1969 | } |
54aa1f4d | 1970 | out: |
c3e69d58 | 1971 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
1972 | return 0; |
1973 | } | |
1974 | ||
5d4f98a2 YZ |
1975 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
1976 | struct btrfs_root *root, | |
1977 | u64 bytenr, u64 num_bytes, u64 flags, | |
1978 | int is_data) | |
1979 | { | |
1980 | struct btrfs_delayed_extent_op *extent_op; | |
1981 | int ret; | |
1982 | ||
1983 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
1984 | if (!extent_op) | |
1985 | return -ENOMEM; | |
1986 | ||
1987 | extent_op->flags_to_set = flags; | |
1988 | extent_op->update_flags = 1; | |
1989 | extent_op->update_key = 0; | |
1990 | extent_op->is_data = is_data ? 1 : 0; | |
1991 | ||
1992 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
1993 | if (ret) | |
1994 | kfree(extent_op); | |
1995 | return ret; | |
1996 | } | |
1997 | ||
1998 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
1999 | struct btrfs_root *root, | |
2000 | struct btrfs_path *path, | |
2001 | u64 objectid, u64 offset, u64 bytenr) | |
2002 | { | |
2003 | struct btrfs_delayed_ref_head *head; | |
2004 | struct btrfs_delayed_ref_node *ref; | |
2005 | struct btrfs_delayed_data_ref *data_ref; | |
2006 | struct btrfs_delayed_ref_root *delayed_refs; | |
2007 | struct rb_node *node; | |
2008 | int ret = 0; | |
2009 | ||
2010 | ret = -ENOENT; | |
2011 | delayed_refs = &trans->transaction->delayed_refs; | |
2012 | spin_lock(&delayed_refs->lock); | |
2013 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2014 | if (!head) | |
2015 | goto out; | |
2016 | ||
2017 | if (!mutex_trylock(&head->mutex)) { | |
2018 | atomic_inc(&head->node.refs); | |
2019 | spin_unlock(&delayed_refs->lock); | |
2020 | ||
2021 | btrfs_release_path(root->fs_info->extent_root, path); | |
2022 | ||
2023 | mutex_lock(&head->mutex); | |
2024 | mutex_unlock(&head->mutex); | |
2025 | btrfs_put_delayed_ref(&head->node); | |
2026 | return -EAGAIN; | |
2027 | } | |
2028 | ||
2029 | node = rb_prev(&head->node.rb_node); | |
2030 | if (!node) | |
2031 | goto out_unlock; | |
2032 | ||
2033 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2034 | ||
2035 | if (ref->bytenr != bytenr) | |
2036 | goto out_unlock; | |
2037 | ||
2038 | ret = 1; | |
2039 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2040 | goto out_unlock; | |
2041 | ||
2042 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2043 | ||
2044 | node = rb_prev(node); | |
2045 | if (node) { | |
2046 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2047 | if (ref->bytenr == bytenr) | |
2048 | goto out_unlock; | |
2049 | } | |
2050 | ||
2051 | if (data_ref->root != root->root_key.objectid || | |
2052 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2053 | goto out_unlock; | |
2054 | ||
2055 | ret = 0; | |
2056 | out_unlock: | |
2057 | mutex_unlock(&head->mutex); | |
2058 | out: | |
2059 | spin_unlock(&delayed_refs->lock); | |
2060 | return ret; | |
2061 | } | |
2062 | ||
2063 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2064 | struct btrfs_root *root, | |
2065 | struct btrfs_path *path, | |
2066 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2067 | { |
2068 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2069 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2070 | struct btrfs_extent_data_ref *ref; |
2071 | struct btrfs_extent_inline_ref *iref; | |
2072 | struct btrfs_extent_item *ei; | |
f321e491 | 2073 | struct btrfs_key key; |
5d4f98a2 | 2074 | u32 item_size; |
be20aa9d | 2075 | int ret; |
925baedd | 2076 | |
be20aa9d | 2077 | key.objectid = bytenr; |
31840ae1 | 2078 | key.offset = (u64)-1; |
f321e491 | 2079 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2080 | |
be20aa9d CM |
2081 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2082 | if (ret < 0) | |
2083 | goto out; | |
2084 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2085 | |
2086 | ret = -ENOENT; | |
2087 | if (path->slots[0] == 0) | |
31840ae1 | 2088 | goto out; |
be20aa9d | 2089 | |
31840ae1 | 2090 | path->slots[0]--; |
f321e491 | 2091 | leaf = path->nodes[0]; |
5d4f98a2 | 2092 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2093 | |
5d4f98a2 | 2094 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2095 | goto out; |
f321e491 | 2096 | |
5d4f98a2 YZ |
2097 | ret = 1; |
2098 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2099 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2100 | if (item_size < sizeof(*ei)) { | |
2101 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2102 | goto out; | |
2103 | } | |
2104 | #endif | |
2105 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2106 | |
5d4f98a2 YZ |
2107 | if (item_size != sizeof(*ei) + |
2108 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2109 | goto out; | |
be20aa9d | 2110 | |
5d4f98a2 YZ |
2111 | if (btrfs_extent_generation(leaf, ei) <= |
2112 | btrfs_root_last_snapshot(&root->root_item)) | |
2113 | goto out; | |
2114 | ||
2115 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2116 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2117 | BTRFS_EXTENT_DATA_REF_KEY) | |
2118 | goto out; | |
2119 | ||
2120 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2121 | if (btrfs_extent_refs(leaf, ei) != | |
2122 | btrfs_extent_data_ref_count(leaf, ref) || | |
2123 | btrfs_extent_data_ref_root(leaf, ref) != | |
2124 | root->root_key.objectid || | |
2125 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2126 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2127 | goto out; | |
2128 | ||
2129 | ret = 0; | |
2130 | out: | |
2131 | return ret; | |
2132 | } | |
2133 | ||
2134 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2135 | struct btrfs_root *root, | |
2136 | u64 objectid, u64 offset, u64 bytenr) | |
2137 | { | |
2138 | struct btrfs_path *path; | |
2139 | int ret; | |
2140 | int ret2; | |
2141 | ||
2142 | path = btrfs_alloc_path(); | |
2143 | if (!path) | |
2144 | return -ENOENT; | |
2145 | ||
2146 | do { | |
2147 | ret = check_committed_ref(trans, root, path, objectid, | |
2148 | offset, bytenr); | |
2149 | if (ret && ret != -ENOENT) | |
f321e491 | 2150 | goto out; |
80ff3856 | 2151 | |
5d4f98a2 YZ |
2152 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2153 | offset, bytenr); | |
2154 | } while (ret2 == -EAGAIN); | |
2155 | ||
2156 | if (ret2 && ret2 != -ENOENT) { | |
2157 | ret = ret2; | |
2158 | goto out; | |
f321e491 | 2159 | } |
5d4f98a2 YZ |
2160 | |
2161 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2162 | ret = 0; | |
be20aa9d | 2163 | out: |
80ff3856 | 2164 | btrfs_free_path(path); |
f321e491 | 2165 | return ret; |
be20aa9d | 2166 | } |
c5739bba | 2167 | |
5d4f98a2 | 2168 | #if 0 |
31840ae1 ZY |
2169 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2170 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 2171 | { |
5f39d397 | 2172 | struct btrfs_key key; |
6407bf6d | 2173 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
2174 | u64 root_gen; |
2175 | u32 nritems; | |
02217ed2 | 2176 | int i; |
db94535d | 2177 | int level; |
31840ae1 | 2178 | int ret = 0; |
e4657689 | 2179 | int shared = 0; |
a28ec197 | 2180 | |
3768f368 | 2181 | if (!root->ref_cows) |
a28ec197 | 2182 | return 0; |
5f39d397 | 2183 | |
e4657689 ZY |
2184 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
2185 | shared = 0; | |
2186 | root_gen = root->root_key.offset; | |
2187 | } else { | |
2188 | shared = 1; | |
2189 | root_gen = trans->transid - 1; | |
2190 | } | |
2191 | ||
db94535d | 2192 | level = btrfs_header_level(buf); |
5f39d397 | 2193 | nritems = btrfs_header_nritems(buf); |
4a096752 | 2194 | |
31840ae1 | 2195 | if (level == 0) { |
31153d81 YZ |
2196 | struct btrfs_leaf_ref *ref; |
2197 | struct btrfs_extent_info *info; | |
2198 | ||
31840ae1 | 2199 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 2200 | if (!ref) { |
31840ae1 | 2201 | ret = -ENOMEM; |
31153d81 YZ |
2202 | goto out; |
2203 | } | |
2204 | ||
e4657689 | 2205 | ref->root_gen = root_gen; |
31153d81 YZ |
2206 | ref->bytenr = buf->start; |
2207 | ref->owner = btrfs_header_owner(buf); | |
2208 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 2209 | ref->nritems = nr_extents; |
31153d81 | 2210 | info = ref->extents; |
bcc63abb | 2211 | |
31840ae1 | 2212 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
2213 | u64 disk_bytenr; |
2214 | btrfs_item_key_to_cpu(buf, &key, i); | |
2215 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2216 | continue; | |
2217 | fi = btrfs_item_ptr(buf, i, | |
2218 | struct btrfs_file_extent_item); | |
2219 | if (btrfs_file_extent_type(buf, fi) == | |
2220 | BTRFS_FILE_EXTENT_INLINE) | |
2221 | continue; | |
2222 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2223 | if (disk_bytenr == 0) | |
2224 | continue; | |
2225 | ||
2226 | info->bytenr = disk_bytenr; | |
2227 | info->num_bytes = | |
2228 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
2229 | info->objectid = key.objectid; | |
2230 | info->offset = key.offset; | |
2231 | info++; | |
2232 | } | |
2233 | ||
e4657689 | 2234 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
2235 | if (ret == -EEXIST && shared) { |
2236 | struct btrfs_leaf_ref *old; | |
2237 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
2238 | BUG_ON(!old); | |
2239 | btrfs_remove_leaf_ref(root, old); | |
2240 | btrfs_free_leaf_ref(root, old); | |
2241 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
2242 | } | |
31153d81 | 2243 | WARN_ON(ret); |
bcc63abb | 2244 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2245 | } |
2246 | out: | |
31840ae1 ZY |
2247 | return ret; |
2248 | } | |
2249 | ||
b7a9f29f CM |
2250 | /* when a block goes through cow, we update the reference counts of |
2251 | * everything that block points to. The internal pointers of the block | |
2252 | * can be in just about any order, and it is likely to have clusters of | |
2253 | * things that are close together and clusters of things that are not. | |
2254 | * | |
2255 | * To help reduce the seeks that come with updating all of these reference | |
2256 | * counts, sort them by byte number before actual updates are done. | |
2257 | * | |
2258 | * struct refsort is used to match byte number to slot in the btree block. | |
2259 | * we sort based on the byte number and then use the slot to actually | |
2260 | * find the item. | |
bd56b302 CM |
2261 | * |
2262 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
2263 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
2264 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
2265 | * single node to be bigger than a page. | |
b7a9f29f CM |
2266 | */ |
2267 | struct refsort { | |
2268 | u64 bytenr; | |
2269 | u32 slot; | |
2270 | }; | |
2271 | ||
2272 | /* | |
2273 | * for passing into sort() | |
2274 | */ | |
2275 | static int refsort_cmp(const void *a_void, const void *b_void) | |
2276 | { | |
2277 | const struct refsort *a = a_void; | |
2278 | const struct refsort *b = b_void; | |
2279 | ||
2280 | if (a->bytenr < b->bytenr) | |
2281 | return -1; | |
2282 | if (a->bytenr > b->bytenr) | |
2283 | return 1; | |
2284 | return 0; | |
2285 | } | |
5d4f98a2 | 2286 | #endif |
b7a9f29f | 2287 | |
5d4f98a2 | 2288 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2289 | struct btrfs_root *root, |
5d4f98a2 YZ |
2290 | struct extent_buffer *buf, |
2291 | int full_backref, int inc) | |
31840ae1 ZY |
2292 | { |
2293 | u64 bytenr; | |
5d4f98a2 YZ |
2294 | u64 num_bytes; |
2295 | u64 parent; | |
31840ae1 | 2296 | u64 ref_root; |
31840ae1 | 2297 | u32 nritems; |
31840ae1 ZY |
2298 | struct btrfs_key key; |
2299 | struct btrfs_file_extent_item *fi; | |
2300 | int i; | |
2301 | int level; | |
2302 | int ret = 0; | |
31840ae1 | 2303 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2304 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2305 | |
2306 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2307 | nritems = btrfs_header_nritems(buf); |
2308 | level = btrfs_header_level(buf); | |
2309 | ||
5d4f98a2 YZ |
2310 | if (!root->ref_cows && level == 0) |
2311 | return 0; | |
31840ae1 | 2312 | |
5d4f98a2 YZ |
2313 | if (inc) |
2314 | process_func = btrfs_inc_extent_ref; | |
2315 | else | |
2316 | process_func = btrfs_free_extent; | |
31840ae1 | 2317 | |
5d4f98a2 YZ |
2318 | if (full_backref) |
2319 | parent = buf->start; | |
2320 | else | |
2321 | parent = 0; | |
2322 | ||
2323 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2324 | if (level == 0) { |
5d4f98a2 | 2325 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2326 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2327 | continue; | |
5d4f98a2 | 2328 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2329 | struct btrfs_file_extent_item); |
2330 | if (btrfs_file_extent_type(buf, fi) == | |
2331 | BTRFS_FILE_EXTENT_INLINE) | |
2332 | continue; | |
2333 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2334 | if (bytenr == 0) | |
2335 | continue; | |
5d4f98a2 YZ |
2336 | |
2337 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2338 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2339 | ret = process_func(trans, root, bytenr, num_bytes, | |
2340 | parent, ref_root, key.objectid, | |
2341 | key.offset); | |
31840ae1 ZY |
2342 | if (ret) |
2343 | goto fail; | |
2344 | } else { | |
5d4f98a2 YZ |
2345 | bytenr = btrfs_node_blockptr(buf, i); |
2346 | num_bytes = btrfs_level_size(root, level - 1); | |
2347 | ret = process_func(trans, root, bytenr, num_bytes, | |
2348 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2349 | if (ret) |
2350 | goto fail; | |
2351 | } | |
2352 | } | |
2353 | return 0; | |
2354 | fail: | |
5d4f98a2 YZ |
2355 | BUG(); |
2356 | return ret; | |
2357 | } | |
2358 | ||
2359 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2360 | struct extent_buffer *buf, int full_backref) | |
2361 | { | |
2362 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2363 | } | |
2364 | ||
2365 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2366 | struct extent_buffer *buf, int full_backref) | |
2367 | { | |
2368 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2369 | } |
2370 | ||
9078a3e1 CM |
2371 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2372 | struct btrfs_root *root, | |
2373 | struct btrfs_path *path, | |
2374 | struct btrfs_block_group_cache *cache) | |
2375 | { | |
2376 | int ret; | |
9078a3e1 | 2377 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2378 | unsigned long bi; |
2379 | struct extent_buffer *leaf; | |
9078a3e1 | 2380 | |
9078a3e1 | 2381 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2382 | if (ret < 0) |
2383 | goto fail; | |
9078a3e1 | 2384 | BUG_ON(ret); |
5f39d397 CM |
2385 | |
2386 | leaf = path->nodes[0]; | |
2387 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2388 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2389 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 2390 | btrfs_release_path(extent_root, path); |
54aa1f4d | 2391 | fail: |
9078a3e1 CM |
2392 | if (ret) |
2393 | return ret; | |
9078a3e1 CM |
2394 | return 0; |
2395 | ||
2396 | } | |
2397 | ||
96b5179d CM |
2398 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2399 | struct btrfs_root *root) | |
9078a3e1 | 2400 | { |
0f9dd46c JB |
2401 | struct btrfs_block_group_cache *cache, *entry; |
2402 | struct rb_node *n; | |
9078a3e1 CM |
2403 | int err = 0; |
2404 | int werr = 0; | |
9078a3e1 | 2405 | struct btrfs_path *path; |
96b5179d | 2406 | u64 last = 0; |
9078a3e1 CM |
2407 | |
2408 | path = btrfs_alloc_path(); | |
2409 | if (!path) | |
2410 | return -ENOMEM; | |
2411 | ||
d397712b | 2412 | while (1) { |
0f9dd46c JB |
2413 | cache = NULL; |
2414 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2415 | for (n = rb_first(&root->fs_info->block_group_cache_tree); | |
2416 | n; n = rb_next(n)) { | |
2417 | entry = rb_entry(n, struct btrfs_block_group_cache, | |
2418 | cache_node); | |
2419 | if (entry->dirty) { | |
2420 | cache = entry; | |
2421 | break; | |
2422 | } | |
2423 | } | |
2424 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
54aa1f4d | 2425 | |
0f9dd46c | 2426 | if (!cache) |
96b5179d | 2427 | break; |
0f9dd46c | 2428 | |
e8569813 | 2429 | cache->dirty = 0; |
0f9dd46c JB |
2430 | last += cache->key.offset; |
2431 | ||
96b5179d CM |
2432 | err = write_one_cache_group(trans, root, |
2433 | path, cache); | |
2434 | /* | |
2435 | * if we fail to write the cache group, we want | |
2436 | * to keep it marked dirty in hopes that a later | |
2437 | * write will work | |
2438 | */ | |
2439 | if (err) { | |
2440 | werr = err; | |
2441 | continue; | |
9078a3e1 CM |
2442 | } |
2443 | } | |
2444 | btrfs_free_path(path); | |
2445 | return werr; | |
2446 | } | |
2447 | ||
d2fb3437 YZ |
2448 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2449 | { | |
2450 | struct btrfs_block_group_cache *block_group; | |
2451 | int readonly = 0; | |
2452 | ||
2453 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2454 | if (!block_group || block_group->ro) | |
2455 | readonly = 1; | |
2456 | if (block_group) | |
fa9c0d79 | 2457 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2458 | return readonly; |
2459 | } | |
2460 | ||
593060d7 CM |
2461 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2462 | u64 total_bytes, u64 bytes_used, | |
2463 | struct btrfs_space_info **space_info) | |
2464 | { | |
2465 | struct btrfs_space_info *found; | |
2466 | ||
2467 | found = __find_space_info(info, flags); | |
2468 | if (found) { | |
25179201 | 2469 | spin_lock(&found->lock); |
593060d7 CM |
2470 | found->total_bytes += total_bytes; |
2471 | found->bytes_used += bytes_used; | |
8f18cf13 | 2472 | found->full = 0; |
25179201 | 2473 | spin_unlock(&found->lock); |
593060d7 CM |
2474 | *space_info = found; |
2475 | return 0; | |
2476 | } | |
c146afad | 2477 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2478 | if (!found) |
2479 | return -ENOMEM; | |
2480 | ||
0f9dd46c | 2481 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 2482 | init_rwsem(&found->groups_sem); |
0f9dd46c | 2483 | spin_lock_init(&found->lock); |
593060d7 CM |
2484 | found->flags = flags; |
2485 | found->total_bytes = total_bytes; | |
2486 | found->bytes_used = bytes_used; | |
2487 | found->bytes_pinned = 0; | |
e8569813 | 2488 | found->bytes_reserved = 0; |
c146afad | 2489 | found->bytes_readonly = 0; |
6a63209f | 2490 | found->bytes_delalloc = 0; |
593060d7 | 2491 | found->full = 0; |
0ef3e66b | 2492 | found->force_alloc = 0; |
593060d7 | 2493 | *space_info = found; |
4184ea7f | 2494 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
2495 | return 0; |
2496 | } | |
2497 | ||
8790d502 CM |
2498 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
2499 | { | |
2500 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 2501 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 2502 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 2503 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
2504 | if (extra_flags) { |
2505 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
2506 | fs_info->avail_data_alloc_bits |= extra_flags; | |
2507 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2508 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
2509 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2510 | fs_info->avail_system_alloc_bits |= extra_flags; | |
2511 | } | |
2512 | } | |
593060d7 | 2513 | |
c146afad YZ |
2514 | static void set_block_group_readonly(struct btrfs_block_group_cache *cache) |
2515 | { | |
2516 | spin_lock(&cache->space_info->lock); | |
2517 | spin_lock(&cache->lock); | |
2518 | if (!cache->ro) { | |
2519 | cache->space_info->bytes_readonly += cache->key.offset - | |
2520 | btrfs_block_group_used(&cache->item); | |
2521 | cache->ro = 1; | |
2522 | } | |
2523 | spin_unlock(&cache->lock); | |
2524 | spin_unlock(&cache->space_info->lock); | |
2525 | } | |
2526 | ||
2b82032c | 2527 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 2528 | { |
2b82032c | 2529 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
a061fc8d CM |
2530 | |
2531 | if (num_devices == 1) | |
2532 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
2533 | if (num_devices < 4) | |
2534 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
2535 | ||
ec44a35c CM |
2536 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
2537 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 2538 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 2539 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 2540 | } |
ec44a35c CM |
2541 | |
2542 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 2543 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 2544 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 2545 | } |
ec44a35c CM |
2546 | |
2547 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
2548 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
2549 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
2550 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
2551 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
2552 | return flags; | |
2553 | } | |
2554 | ||
6a63209f JB |
2555 | static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) |
2556 | { | |
2557 | struct btrfs_fs_info *info = root->fs_info; | |
2558 | u64 alloc_profile; | |
2559 | ||
2560 | if (data) { | |
2561 | alloc_profile = info->avail_data_alloc_bits & | |
2562 | info->data_alloc_profile; | |
2563 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
2564 | } else if (root == root->fs_info->chunk_root) { | |
2565 | alloc_profile = info->avail_system_alloc_bits & | |
2566 | info->system_alloc_profile; | |
2567 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
2568 | } else { | |
2569 | alloc_profile = info->avail_metadata_alloc_bits & | |
2570 | info->metadata_alloc_profile; | |
2571 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
2572 | } | |
2573 | ||
2574 | return btrfs_reduce_alloc_profile(root, data); | |
2575 | } | |
2576 | ||
2577 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) | |
2578 | { | |
2579 | u64 alloc_target; | |
2580 | ||
2581 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
2582 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, | |
2583 | alloc_target); | |
2584 | } | |
2585 | ||
2586 | /* | |
2587 | * for now this just makes sure we have at least 5% of our metadata space free | |
2588 | * for use. | |
2589 | */ | |
2590 | int btrfs_check_metadata_free_space(struct btrfs_root *root) | |
2591 | { | |
2592 | struct btrfs_fs_info *info = root->fs_info; | |
2593 | struct btrfs_space_info *meta_sinfo; | |
2594 | u64 alloc_target, thresh; | |
4e06bdd6 | 2595 | int committed = 0, ret; |
6a63209f JB |
2596 | |
2597 | /* get the space info for where the metadata will live */ | |
2598 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
2599 | meta_sinfo = __find_space_info(info, alloc_target); | |
2600 | ||
4e06bdd6 | 2601 | again: |
6a63209f | 2602 | spin_lock(&meta_sinfo->lock); |
4e06bdd6 JB |
2603 | if (!meta_sinfo->full) |
2604 | thresh = meta_sinfo->total_bytes * 80; | |
2605 | else | |
2606 | thresh = meta_sinfo->total_bytes * 95; | |
6a63209f JB |
2607 | |
2608 | do_div(thresh, 100); | |
2609 | ||
2610 | if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
2611 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { | |
4e06bdd6 JB |
2612 | struct btrfs_trans_handle *trans; |
2613 | if (!meta_sinfo->full) { | |
2614 | meta_sinfo->force_alloc = 1; | |
2615 | spin_unlock(&meta_sinfo->lock); | |
2616 | ||
2617 | trans = btrfs_start_transaction(root, 1); | |
2618 | if (!trans) | |
2619 | return -ENOMEM; | |
2620 | ||
2621 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
2622 | 2 * 1024 * 1024, alloc_target, 0); | |
2623 | btrfs_end_transaction(trans, root); | |
2624 | goto again; | |
2625 | } | |
6a63209f | 2626 | spin_unlock(&meta_sinfo->lock); |
4e06bdd6 JB |
2627 | |
2628 | if (!committed) { | |
2629 | committed = 1; | |
2630 | trans = btrfs_join_transaction(root, 1); | |
2631 | if (!trans) | |
2632 | return -ENOMEM; | |
2633 | ret = btrfs_commit_transaction(trans, root); | |
2634 | if (ret) | |
2635 | return ret; | |
2636 | goto again; | |
2637 | } | |
6a63209f JB |
2638 | return -ENOSPC; |
2639 | } | |
2640 | spin_unlock(&meta_sinfo->lock); | |
2641 | ||
2642 | return 0; | |
2643 | } | |
2644 | ||
2645 | /* | |
2646 | * This will check the space that the inode allocates from to make sure we have | |
2647 | * enough space for bytes. | |
2648 | */ | |
2649 | int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, | |
2650 | u64 bytes) | |
2651 | { | |
2652 | struct btrfs_space_info *data_sinfo; | |
4e06bdd6 | 2653 | int ret = 0, committed = 0; |
6a63209f JB |
2654 | |
2655 | /* make sure bytes are sectorsize aligned */ | |
2656 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
2657 | ||
2658 | data_sinfo = BTRFS_I(inode)->space_info; | |
2659 | again: | |
2660 | /* make sure we have enough space to handle the data first */ | |
2661 | spin_lock(&data_sinfo->lock); | |
2662 | if (data_sinfo->total_bytes - data_sinfo->bytes_used - | |
2663 | data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - | |
2664 | data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - | |
2665 | data_sinfo->bytes_may_use < bytes) { | |
4e06bdd6 JB |
2666 | struct btrfs_trans_handle *trans; |
2667 | ||
6a63209f JB |
2668 | /* |
2669 | * if we don't have enough free bytes in this space then we need | |
2670 | * to alloc a new chunk. | |
2671 | */ | |
2672 | if (!data_sinfo->full) { | |
2673 | u64 alloc_target; | |
6a63209f JB |
2674 | |
2675 | data_sinfo->force_alloc = 1; | |
2676 | spin_unlock(&data_sinfo->lock); | |
2677 | ||
2678 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
2679 | trans = btrfs_start_transaction(root, 1); | |
2680 | if (!trans) | |
2681 | return -ENOMEM; | |
2682 | ||
2683 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
2684 | bytes + 2 * 1024 * 1024, | |
2685 | alloc_target, 0); | |
2686 | btrfs_end_transaction(trans, root); | |
2687 | if (ret) | |
2688 | return ret; | |
2689 | goto again; | |
2690 | } | |
2691 | spin_unlock(&data_sinfo->lock); | |
4e06bdd6 JB |
2692 | |
2693 | /* commit the current transaction and try again */ | |
2694 | if (!committed) { | |
2695 | committed = 1; | |
2696 | trans = btrfs_join_transaction(root, 1); | |
2697 | if (!trans) | |
2698 | return -ENOMEM; | |
2699 | ret = btrfs_commit_transaction(trans, root); | |
2700 | if (ret) | |
2701 | return ret; | |
2702 | goto again; | |
2703 | } | |
2704 | ||
6a63209f JB |
2705 | printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" |
2706 | ", %llu bytes_used, %llu bytes_reserved, " | |
2707 | "%llu bytes_pinned, %llu bytes_readonly, %llu may use" | |
21380931 JB |
2708 | "%llu total\n", (unsigned long long)bytes, |
2709 | (unsigned long long)data_sinfo->bytes_delalloc, | |
2710 | (unsigned long long)data_sinfo->bytes_used, | |
2711 | (unsigned long long)data_sinfo->bytes_reserved, | |
2712 | (unsigned long long)data_sinfo->bytes_pinned, | |
2713 | (unsigned long long)data_sinfo->bytes_readonly, | |
2714 | (unsigned long long)data_sinfo->bytes_may_use, | |
2715 | (unsigned long long)data_sinfo->total_bytes); | |
6a63209f JB |
2716 | return -ENOSPC; |
2717 | } | |
2718 | data_sinfo->bytes_may_use += bytes; | |
2719 | BTRFS_I(inode)->reserved_bytes += bytes; | |
2720 | spin_unlock(&data_sinfo->lock); | |
2721 | ||
2722 | return btrfs_check_metadata_free_space(root); | |
2723 | } | |
2724 | ||
2725 | /* | |
2726 | * if there was an error for whatever reason after calling | |
2727 | * btrfs_check_data_free_space, call this so we can cleanup the counters. | |
2728 | */ | |
2729 | void btrfs_free_reserved_data_space(struct btrfs_root *root, | |
2730 | struct inode *inode, u64 bytes) | |
2731 | { | |
2732 | struct btrfs_space_info *data_sinfo; | |
2733 | ||
2734 | /* make sure bytes are sectorsize aligned */ | |
2735 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
2736 | ||
2737 | data_sinfo = BTRFS_I(inode)->space_info; | |
2738 | spin_lock(&data_sinfo->lock); | |
2739 | data_sinfo->bytes_may_use -= bytes; | |
2740 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
2741 | spin_unlock(&data_sinfo->lock); | |
2742 | } | |
2743 | ||
2744 | /* called when we are adding a delalloc extent to the inode's io_tree */ | |
2745 | void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, | |
2746 | u64 bytes) | |
2747 | { | |
2748 | struct btrfs_space_info *data_sinfo; | |
2749 | ||
2750 | /* get the space info for where this inode will be storing its data */ | |
2751 | data_sinfo = BTRFS_I(inode)->space_info; | |
2752 | ||
2753 | /* make sure we have enough space to handle the data first */ | |
2754 | spin_lock(&data_sinfo->lock); | |
2755 | data_sinfo->bytes_delalloc += bytes; | |
2756 | ||
2757 | /* | |
2758 | * we are adding a delalloc extent without calling | |
2759 | * btrfs_check_data_free_space first. This happens on a weird | |
2760 | * writepage condition, but shouldn't hurt our accounting | |
2761 | */ | |
2762 | if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { | |
2763 | data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; | |
2764 | BTRFS_I(inode)->reserved_bytes = 0; | |
2765 | } else { | |
2766 | data_sinfo->bytes_may_use -= bytes; | |
2767 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
2768 | } | |
2769 | ||
2770 | spin_unlock(&data_sinfo->lock); | |
2771 | } | |
2772 | ||
2773 | /* called when we are clearing an delalloc extent from the inode's io_tree */ | |
2774 | void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, | |
2775 | u64 bytes) | |
2776 | { | |
2777 | struct btrfs_space_info *info; | |
2778 | ||
2779 | info = BTRFS_I(inode)->space_info; | |
2780 | ||
2781 | spin_lock(&info->lock); | |
2782 | info->bytes_delalloc -= bytes; | |
2783 | spin_unlock(&info->lock); | |
2784 | } | |
2785 | ||
97e728d4 JB |
2786 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
2787 | { | |
2788 | struct list_head *head = &info->space_info; | |
2789 | struct btrfs_space_info *found; | |
2790 | ||
2791 | rcu_read_lock(); | |
2792 | list_for_each_entry_rcu(found, head, list) { | |
2793 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
2794 | found->force_alloc = 1; | |
2795 | } | |
2796 | rcu_read_unlock(); | |
2797 | } | |
2798 | ||
6324fbf3 CM |
2799 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2800 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 2801 | u64 flags, int force) |
6324fbf3 CM |
2802 | { |
2803 | struct btrfs_space_info *space_info; | |
97e728d4 | 2804 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6324fbf3 | 2805 | u64 thresh; |
c146afad YZ |
2806 | int ret = 0; |
2807 | ||
97e728d4 | 2808 | mutex_lock(&fs_info->chunk_mutex); |
6324fbf3 | 2809 | |
2b82032c | 2810 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 2811 | |
6324fbf3 | 2812 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
2813 | if (!space_info) { |
2814 | ret = update_space_info(extent_root->fs_info, flags, | |
2815 | 0, 0, &space_info); | |
2816 | BUG_ON(ret); | |
2817 | } | |
6324fbf3 CM |
2818 | BUG_ON(!space_info); |
2819 | ||
25179201 | 2820 | spin_lock(&space_info->lock); |
0ef3e66b CM |
2821 | if (space_info->force_alloc) { |
2822 | force = 1; | |
2823 | space_info->force_alloc = 0; | |
2824 | } | |
25179201 JB |
2825 | if (space_info->full) { |
2826 | spin_unlock(&space_info->lock); | |
925baedd | 2827 | goto out; |
25179201 | 2828 | } |
6324fbf3 | 2829 | |
c146afad YZ |
2830 | thresh = space_info->total_bytes - space_info->bytes_readonly; |
2831 | thresh = div_factor(thresh, 6); | |
0ef3e66b | 2832 | if (!force && |
e8569813 | 2833 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
2834 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
2835 | spin_unlock(&space_info->lock); | |
925baedd | 2836 | goto out; |
25179201 | 2837 | } |
25179201 JB |
2838 | spin_unlock(&space_info->lock); |
2839 | ||
97e728d4 JB |
2840 | /* |
2841 | * if we're doing a data chunk, go ahead and make sure that | |
2842 | * we keep a reasonable number of metadata chunks allocated in the | |
2843 | * FS as well. | |
2844 | */ | |
2845 | if (flags & BTRFS_BLOCK_GROUP_DATA) { | |
2846 | fs_info->data_chunk_allocations++; | |
2847 | if (!(fs_info->data_chunk_allocations % | |
2848 | fs_info->metadata_ratio)) | |
2849 | force_metadata_allocation(fs_info); | |
2850 | } | |
2851 | ||
2b82032c | 2852 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
d397712b | 2853 | if (ret) |
6324fbf3 | 2854 | space_info->full = 1; |
a74a4b97 | 2855 | out: |
c146afad | 2856 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 2857 | return ret; |
6324fbf3 CM |
2858 | } |
2859 | ||
9078a3e1 CM |
2860 | static int update_block_group(struct btrfs_trans_handle *trans, |
2861 | struct btrfs_root *root, | |
db94535d | 2862 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 2863 | int mark_free) |
9078a3e1 CM |
2864 | { |
2865 | struct btrfs_block_group_cache *cache; | |
2866 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 2867 | u64 total = num_bytes; |
9078a3e1 | 2868 | u64 old_val; |
db94535d | 2869 | u64 byte_in_group; |
3e1ad54f | 2870 | |
5d4f98a2 YZ |
2871 | /* block accounting for super block */ |
2872 | spin_lock(&info->delalloc_lock); | |
2873 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
2874 | if (alloc) | |
2875 | old_val += num_bytes; | |
2876 | else | |
2877 | old_val -= num_bytes; | |
2878 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
2879 | ||
2880 | /* block accounting for root item */ | |
2881 | old_val = btrfs_root_used(&root->root_item); | |
2882 | if (alloc) | |
2883 | old_val += num_bytes; | |
2884 | else | |
2885 | old_val -= num_bytes; | |
2886 | btrfs_set_root_used(&root->root_item, old_val); | |
2887 | spin_unlock(&info->delalloc_lock); | |
2888 | ||
d397712b | 2889 | while (total) { |
db94535d | 2890 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 2891 | if (!cache) |
9078a3e1 | 2892 | return -1; |
db94535d CM |
2893 | byte_in_group = bytenr - cache->key.objectid; |
2894 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 2895 | |
25179201 | 2896 | spin_lock(&cache->space_info->lock); |
c286ac48 | 2897 | spin_lock(&cache->lock); |
0f9dd46c | 2898 | cache->dirty = 1; |
9078a3e1 | 2899 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 2900 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 2901 | if (alloc) { |
db94535d | 2902 | old_val += num_bytes; |
6324fbf3 | 2903 | cache->space_info->bytes_used += num_bytes; |
a512bbf8 | 2904 | if (cache->ro) |
c146afad | 2905 | cache->space_info->bytes_readonly -= num_bytes; |
c286ac48 CM |
2906 | btrfs_set_block_group_used(&cache->item, old_val); |
2907 | spin_unlock(&cache->lock); | |
25179201 | 2908 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 2909 | } else { |
db94535d | 2910 | old_val -= num_bytes; |
6324fbf3 | 2911 | cache->space_info->bytes_used -= num_bytes; |
c146afad YZ |
2912 | if (cache->ro) |
2913 | cache->space_info->bytes_readonly += num_bytes; | |
c286ac48 CM |
2914 | btrfs_set_block_group_used(&cache->item, old_val); |
2915 | spin_unlock(&cache->lock); | |
25179201 | 2916 | spin_unlock(&cache->space_info->lock); |
f510cfec | 2917 | if (mark_free) { |
0f9dd46c | 2918 | int ret; |
1f3c79a2 LH |
2919 | |
2920 | ret = btrfs_discard_extent(root, bytenr, | |
2921 | num_bytes); | |
2922 | WARN_ON(ret); | |
2923 | ||
0f9dd46c JB |
2924 | ret = btrfs_add_free_space(cache, bytenr, |
2925 | num_bytes); | |
d2fb3437 | 2926 | WARN_ON(ret); |
e37c9e69 | 2927 | } |
cd1bc465 | 2928 | } |
fa9c0d79 | 2929 | btrfs_put_block_group(cache); |
db94535d CM |
2930 | total -= num_bytes; |
2931 | bytenr += num_bytes; | |
9078a3e1 CM |
2932 | } |
2933 | return 0; | |
2934 | } | |
6324fbf3 | 2935 | |
a061fc8d CM |
2936 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
2937 | { | |
0f9dd46c | 2938 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 2939 | u64 bytenr; |
0f9dd46c JB |
2940 | |
2941 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
2942 | if (!cache) | |
a061fc8d | 2943 | return 0; |
0f9dd46c | 2944 | |
d2fb3437 | 2945 | bytenr = cache->key.objectid; |
fa9c0d79 | 2946 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2947 | |
2948 | return bytenr; | |
a061fc8d CM |
2949 | } |
2950 | ||
e02119d5 | 2951 | int btrfs_update_pinned_extents(struct btrfs_root *root, |
324ae4df Y |
2952 | u64 bytenr, u64 num, int pin) |
2953 | { | |
2954 | u64 len; | |
2955 | struct btrfs_block_group_cache *cache; | |
2956 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2957 | ||
2958 | if (pin) { | |
2959 | set_extent_dirty(&fs_info->pinned_extents, | |
2960 | bytenr, bytenr + num - 1, GFP_NOFS); | |
2961 | } else { | |
2962 | clear_extent_dirty(&fs_info->pinned_extents, | |
2963 | bytenr, bytenr + num - 1, GFP_NOFS); | |
2964 | } | |
b9473439 | 2965 | |
324ae4df Y |
2966 | while (num > 0) { |
2967 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
e8569813 ZY |
2968 | BUG_ON(!cache); |
2969 | len = min(num, cache->key.offset - | |
2970 | (bytenr - cache->key.objectid)); | |
324ae4df | 2971 | if (pin) { |
25179201 | 2972 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
2973 | spin_lock(&cache->lock); |
2974 | cache->pinned += len; | |
2975 | cache->space_info->bytes_pinned += len; | |
2976 | spin_unlock(&cache->lock); | |
25179201 | 2977 | spin_unlock(&cache->space_info->lock); |
324ae4df Y |
2978 | fs_info->total_pinned += len; |
2979 | } else { | |
25179201 | 2980 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
2981 | spin_lock(&cache->lock); |
2982 | cache->pinned -= len; | |
2983 | cache->space_info->bytes_pinned -= len; | |
2984 | spin_unlock(&cache->lock); | |
25179201 | 2985 | spin_unlock(&cache->space_info->lock); |
324ae4df | 2986 | fs_info->total_pinned -= len; |
07103a3c JB |
2987 | if (cache->cached) |
2988 | btrfs_add_free_space(cache, bytenr, len); | |
324ae4df | 2989 | } |
fa9c0d79 | 2990 | btrfs_put_block_group(cache); |
324ae4df Y |
2991 | bytenr += len; |
2992 | num -= len; | |
2993 | } | |
2994 | return 0; | |
2995 | } | |
9078a3e1 | 2996 | |
e8569813 ZY |
2997 | static int update_reserved_extents(struct btrfs_root *root, |
2998 | u64 bytenr, u64 num, int reserve) | |
2999 | { | |
3000 | u64 len; | |
3001 | struct btrfs_block_group_cache *cache; | |
3002 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3003 | ||
e8569813 ZY |
3004 | while (num > 0) { |
3005 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
3006 | BUG_ON(!cache); | |
3007 | len = min(num, cache->key.offset - | |
3008 | (bytenr - cache->key.objectid)); | |
25179201 JB |
3009 | |
3010 | spin_lock(&cache->space_info->lock); | |
3011 | spin_lock(&cache->lock); | |
e8569813 | 3012 | if (reserve) { |
e8569813 ZY |
3013 | cache->reserved += len; |
3014 | cache->space_info->bytes_reserved += len; | |
e8569813 | 3015 | } else { |
e8569813 ZY |
3016 | cache->reserved -= len; |
3017 | cache->space_info->bytes_reserved -= len; | |
e8569813 | 3018 | } |
25179201 JB |
3019 | spin_unlock(&cache->lock); |
3020 | spin_unlock(&cache->space_info->lock); | |
fa9c0d79 | 3021 | btrfs_put_block_group(cache); |
e8569813 ZY |
3022 | bytenr += len; |
3023 | num -= len; | |
3024 | } | |
3025 | return 0; | |
3026 | } | |
3027 | ||
d1310b2e | 3028 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 3029 | { |
ccd467d6 | 3030 | u64 last = 0; |
1a5bc167 CM |
3031 | u64 start; |
3032 | u64 end; | |
d1310b2e | 3033 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 3034 | int ret; |
ccd467d6 | 3035 | |
d397712b | 3036 | while (1) { |
1a5bc167 CM |
3037 | ret = find_first_extent_bit(pinned_extents, last, |
3038 | &start, &end, EXTENT_DIRTY); | |
3039 | if (ret) | |
ccd467d6 | 3040 | break; |
1a5bc167 CM |
3041 | set_extent_dirty(copy, start, end, GFP_NOFS); |
3042 | last = end + 1; | |
ccd467d6 CM |
3043 | } |
3044 | return 0; | |
3045 | } | |
3046 | ||
3047 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
3048 | struct btrfs_root *root, | |
d1310b2e | 3049 | struct extent_io_tree *unpin) |
a28ec197 | 3050 | { |
1a5bc167 CM |
3051 | u64 start; |
3052 | u64 end; | |
a28ec197 | 3053 | int ret; |
a28ec197 | 3054 | |
d397712b | 3055 | while (1) { |
1a5bc167 CM |
3056 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
3057 | EXTENT_DIRTY); | |
3058 | if (ret) | |
a28ec197 | 3059 | break; |
1f3c79a2 LH |
3060 | |
3061 | ret = btrfs_discard_extent(root, start, end + 1 - start); | |
3062 | ||
b9473439 | 3063 | /* unlocks the pinned mutex */ |
e02119d5 | 3064 | btrfs_update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 | 3065 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1f3c79a2 | 3066 | |
b9473439 | 3067 | cond_resched(); |
a28ec197 | 3068 | } |
1f3c79a2 | 3069 | return ret; |
a28ec197 CM |
3070 | } |
3071 | ||
31840ae1 ZY |
3072 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
3073 | struct btrfs_root *root, | |
b9473439 CM |
3074 | struct btrfs_path *path, |
3075 | u64 bytenr, u64 num_bytes, int is_data, | |
3076 | struct extent_buffer **must_clean) | |
e20d96d6 | 3077 | { |
1a5bc167 | 3078 | int err = 0; |
31840ae1 | 3079 | struct extent_buffer *buf; |
8ef97622 | 3080 | |
31840ae1 ZY |
3081 | if (is_data) |
3082 | goto pinit; | |
3083 | ||
3084 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); | |
3085 | if (!buf) | |
3086 | goto pinit; | |
3087 | ||
3088 | /* we can reuse a block if it hasn't been written | |
3089 | * and it is from this transaction. We can't | |
3090 | * reuse anything from the tree log root because | |
3091 | * it has tiny sub-transactions. | |
3092 | */ | |
3093 | if (btrfs_buffer_uptodate(buf, 0) && | |
3094 | btrfs_try_tree_lock(buf)) { | |
3095 | u64 header_owner = btrfs_header_owner(buf); | |
3096 | u64 header_transid = btrfs_header_generation(buf); | |
3097 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
3098 | header_transid == trans->transid && | |
3099 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
b9473439 | 3100 | *must_clean = buf; |
31840ae1 | 3101 | return 1; |
8ef97622 | 3102 | } |
31840ae1 | 3103 | btrfs_tree_unlock(buf); |
f4b9aa8d | 3104 | } |
31840ae1 ZY |
3105 | free_extent_buffer(buf); |
3106 | pinit: | |
b9473439 | 3107 | btrfs_set_path_blocking(path); |
b9473439 | 3108 | /* unlocks the pinned mutex */ |
31840ae1 ZY |
3109 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); |
3110 | ||
be744175 | 3111 | BUG_ON(err < 0); |
e20d96d6 CM |
3112 | return 0; |
3113 | } | |
3114 | ||
5d4f98a2 YZ |
3115 | |
3116 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, | |
3117 | struct btrfs_root *root, | |
3118 | u64 bytenr, u64 num_bytes, u64 parent, | |
3119 | u64 root_objectid, u64 owner_objectid, | |
3120 | u64 owner_offset, int refs_to_drop, | |
3121 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 3122 | { |
e2fa7227 | 3123 | struct btrfs_key key; |
5d4f98a2 | 3124 | struct btrfs_path *path; |
1261ec42 CM |
3125 | struct btrfs_fs_info *info = root->fs_info; |
3126 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 3127 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3128 | struct btrfs_extent_item *ei; |
3129 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3130 | int ret; |
5d4f98a2 | 3131 | int is_data; |
952fccac CM |
3132 | int extent_slot = 0; |
3133 | int found_extent = 0; | |
3134 | int num_to_del = 1; | |
5d4f98a2 YZ |
3135 | u32 item_size; |
3136 | u64 refs; | |
037e6390 | 3137 | |
5caf2a00 | 3138 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3139 | if (!path) |
3140 | return -ENOMEM; | |
5f26f772 | 3141 | |
3c12ac72 | 3142 | path->reada = 1; |
b9473439 | 3143 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3144 | |
3145 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
3146 | BUG_ON(!is_data && refs_to_drop != 1); | |
3147 | ||
3148 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
3149 | bytenr, num_bytes, parent, | |
3150 | root_objectid, owner_objectid, | |
3151 | owner_offset); | |
7bb86316 | 3152 | if (ret == 0) { |
952fccac | 3153 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3154 | while (extent_slot >= 0) { |
3155 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3156 | extent_slot); |
5d4f98a2 | 3157 | if (key.objectid != bytenr) |
952fccac | 3158 | break; |
5d4f98a2 YZ |
3159 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3160 | key.offset == num_bytes) { | |
952fccac CM |
3161 | found_extent = 1; |
3162 | break; | |
3163 | } | |
3164 | if (path->slots[0] - extent_slot > 5) | |
3165 | break; | |
5d4f98a2 | 3166 | extent_slot--; |
952fccac | 3167 | } |
5d4f98a2 YZ |
3168 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
3169 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
3170 | if (found_extent && item_size < sizeof(*ei)) | |
3171 | found_extent = 0; | |
3172 | #endif | |
31840ae1 | 3173 | if (!found_extent) { |
5d4f98a2 | 3174 | BUG_ON(iref); |
56bec294 | 3175 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
3176 | NULL, refs_to_drop, |
3177 | is_data); | |
31840ae1 ZY |
3178 | BUG_ON(ret); |
3179 | btrfs_release_path(extent_root, path); | |
b9473439 | 3180 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3181 | |
3182 | key.objectid = bytenr; | |
3183 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3184 | key.offset = num_bytes; | |
3185 | ||
31840ae1 ZY |
3186 | ret = btrfs_search_slot(trans, extent_root, |
3187 | &key, path, -1, 1); | |
f3465ca4 JB |
3188 | if (ret) { |
3189 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
3190 | ", was looking for %llu\n", ret, |
3191 | (unsigned long long)bytenr); | |
f3465ca4 JB |
3192 | btrfs_print_leaf(extent_root, path->nodes[0]); |
3193 | } | |
31840ae1 ZY |
3194 | BUG_ON(ret); |
3195 | extent_slot = path->slots[0]; | |
3196 | } | |
7bb86316 CM |
3197 | } else { |
3198 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3199 | WARN_ON(1); | |
d397712b | 3200 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 3201 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 3202 | (unsigned long long)bytenr, |
56bec294 | 3203 | (unsigned long long)parent, |
d397712b | 3204 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
3205 | (unsigned long long)owner_objectid, |
3206 | (unsigned long long)owner_offset); | |
7bb86316 | 3207 | } |
5f39d397 CM |
3208 | |
3209 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
3210 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
3211 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3212 | if (item_size < sizeof(*ei)) { | |
3213 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
3214 | ret = convert_extent_item_v0(trans, extent_root, path, | |
3215 | owner_objectid, 0); | |
3216 | BUG_ON(ret < 0); | |
3217 | ||
3218 | btrfs_release_path(extent_root, path); | |
3219 | path->leave_spinning = 1; | |
3220 | ||
3221 | key.objectid = bytenr; | |
3222 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3223 | key.offset = num_bytes; | |
3224 | ||
3225 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
3226 | -1, 1); | |
3227 | if (ret) { | |
3228 | printk(KERN_ERR "umm, got %d back from search" | |
3229 | ", was looking for %llu\n", ret, | |
3230 | (unsigned long long)bytenr); | |
3231 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3232 | } | |
3233 | BUG_ON(ret); | |
3234 | extent_slot = path->slots[0]; | |
3235 | leaf = path->nodes[0]; | |
3236 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
3237 | } | |
3238 | #endif | |
3239 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 3240 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3241 | struct btrfs_extent_item); |
5d4f98a2 YZ |
3242 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
3243 | struct btrfs_tree_block_info *bi; | |
3244 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3245 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3246 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3247 | } | |
56bec294 | 3248 | |
5d4f98a2 | 3249 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
3250 | BUG_ON(refs < refs_to_drop); |
3251 | refs -= refs_to_drop; | |
5f39d397 | 3252 | |
5d4f98a2 YZ |
3253 | if (refs > 0) { |
3254 | if (extent_op) | |
3255 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3256 | /* | |
3257 | * In the case of inline back ref, reference count will | |
3258 | * be updated by remove_extent_backref | |
952fccac | 3259 | */ |
5d4f98a2 YZ |
3260 | if (iref) { |
3261 | BUG_ON(!found_extent); | |
3262 | } else { | |
3263 | btrfs_set_extent_refs(leaf, ei, refs); | |
3264 | btrfs_mark_buffer_dirty(leaf); | |
3265 | } | |
3266 | if (found_extent) { | |
3267 | ret = remove_extent_backref(trans, extent_root, path, | |
3268 | iref, refs_to_drop, | |
3269 | is_data); | |
952fccac CM |
3270 | BUG_ON(ret); |
3271 | } | |
5d4f98a2 YZ |
3272 | } else { |
3273 | int mark_free = 0; | |
b9473439 | 3274 | struct extent_buffer *must_clean = NULL; |
78fae27e | 3275 | |
5d4f98a2 YZ |
3276 | if (found_extent) { |
3277 | BUG_ON(is_data && refs_to_drop != | |
3278 | extent_data_ref_count(root, path, iref)); | |
3279 | if (iref) { | |
3280 | BUG_ON(path->slots[0] != extent_slot); | |
3281 | } else { | |
3282 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3283 | path->slots[0] = extent_slot; | |
3284 | num_to_del = 2; | |
3285 | } | |
78fae27e | 3286 | } |
b9473439 | 3287 | |
5d4f98a2 YZ |
3288 | ret = pin_down_bytes(trans, root, path, bytenr, |
3289 | num_bytes, is_data, &must_clean); | |
3290 | if (ret > 0) | |
3291 | mark_free = 1; | |
3292 | BUG_ON(ret < 0); | |
b9473439 CM |
3293 | /* |
3294 | * it is going to be very rare for someone to be waiting | |
3295 | * on the block we're freeing. del_items might need to | |
3296 | * schedule, so rather than get fancy, just force it | |
3297 | * to blocking here | |
3298 | */ | |
3299 | if (must_clean) | |
3300 | btrfs_set_lock_blocking(must_clean); | |
3301 | ||
952fccac CM |
3302 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3303 | num_to_del); | |
31840ae1 | 3304 | BUG_ON(ret); |
25179201 | 3305 | btrfs_release_path(extent_root, path); |
21af804c | 3306 | |
b9473439 CM |
3307 | if (must_clean) { |
3308 | clean_tree_block(NULL, root, must_clean); | |
3309 | btrfs_tree_unlock(must_clean); | |
3310 | free_extent_buffer(must_clean); | |
3311 | } | |
3312 | ||
5d4f98a2 | 3313 | if (is_data) { |
459931ec CM |
3314 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
3315 | BUG_ON(ret); | |
d57e62b8 CM |
3316 | } else { |
3317 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
3318 | bytenr >> PAGE_CACHE_SHIFT, | |
3319 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
3320 | } |
3321 | ||
dcbdd4dc CM |
3322 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
3323 | mark_free); | |
3324 | BUG_ON(ret); | |
a28ec197 | 3325 | } |
5caf2a00 | 3326 | btrfs_free_path(path); |
a28ec197 CM |
3327 | return ret; |
3328 | } | |
3329 | ||
1887be66 CM |
3330 | /* |
3331 | * when we free an extent, it is possible (and likely) that we free the last | |
3332 | * delayed ref for that extent as well. This searches the delayed ref tree for | |
3333 | * a given extent, and if there are no other delayed refs to be processed, it | |
3334 | * removes it from the tree. | |
3335 | */ | |
3336 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
3337 | struct btrfs_root *root, u64 bytenr) | |
3338 | { | |
3339 | struct btrfs_delayed_ref_head *head; | |
3340 | struct btrfs_delayed_ref_root *delayed_refs; | |
3341 | struct btrfs_delayed_ref_node *ref; | |
3342 | struct rb_node *node; | |
3343 | int ret; | |
3344 | ||
3345 | delayed_refs = &trans->transaction->delayed_refs; | |
3346 | spin_lock(&delayed_refs->lock); | |
3347 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
3348 | if (!head) | |
3349 | goto out; | |
3350 | ||
3351 | node = rb_prev(&head->node.rb_node); | |
3352 | if (!node) | |
3353 | goto out; | |
3354 | ||
3355 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
3356 | ||
3357 | /* there are still entries for this ref, we can't drop it */ | |
3358 | if (ref->bytenr == bytenr) | |
3359 | goto out; | |
3360 | ||
5d4f98a2 YZ |
3361 | if (head->extent_op) { |
3362 | if (!head->must_insert_reserved) | |
3363 | goto out; | |
3364 | kfree(head->extent_op); | |
3365 | head->extent_op = NULL; | |
3366 | } | |
3367 | ||
1887be66 CM |
3368 | /* |
3369 | * waiting for the lock here would deadlock. If someone else has it | |
3370 | * locked they are already in the process of dropping it anyway | |
3371 | */ | |
3372 | if (!mutex_trylock(&head->mutex)) | |
3373 | goto out; | |
3374 | ||
3375 | /* | |
3376 | * at this point we have a head with no other entries. Go | |
3377 | * ahead and process it. | |
3378 | */ | |
3379 | head->node.in_tree = 0; | |
3380 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 3381 | |
1887be66 CM |
3382 | delayed_refs->num_entries--; |
3383 | ||
3384 | /* | |
3385 | * we don't take a ref on the node because we're removing it from the | |
3386 | * tree, so we just steal the ref the tree was holding. | |
3387 | */ | |
c3e69d58 CM |
3388 | delayed_refs->num_heads--; |
3389 | if (list_empty(&head->cluster)) | |
3390 | delayed_refs->num_heads_ready--; | |
3391 | ||
3392 | list_del_init(&head->cluster); | |
1887be66 CM |
3393 | spin_unlock(&delayed_refs->lock); |
3394 | ||
3395 | ret = run_one_delayed_ref(trans, root->fs_info->tree_root, | |
5d4f98a2 YZ |
3396 | &head->node, head->extent_op, |
3397 | head->must_insert_reserved); | |
1887be66 CM |
3398 | BUG_ON(ret); |
3399 | btrfs_put_delayed_ref(&head->node); | |
3400 | return 0; | |
3401 | out: | |
3402 | spin_unlock(&delayed_refs->lock); | |
3403 | return 0; | |
3404 | } | |
3405 | ||
925baedd | 3406 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
3407 | struct btrfs_root *root, |
3408 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 3409 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
3410 | { |
3411 | int ret; | |
3412 | ||
56bec294 CM |
3413 | /* |
3414 | * tree log blocks never actually go into the extent allocation | |
3415 | * tree, just update pinning info and exit early. | |
56bec294 | 3416 | */ |
5d4f98a2 YZ |
3417 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
3418 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 3419 | /* unlocks the pinned mutex */ |
56bec294 | 3420 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); |
56bec294 CM |
3421 | update_reserved_extents(root, bytenr, num_bytes, 0); |
3422 | ret = 0; | |
5d4f98a2 YZ |
3423 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
3424 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
3425 | parent, root_objectid, (int)owner, | |
3426 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 CM |
3427 | BUG_ON(ret); |
3428 | ret = check_ref_cleanup(trans, root, bytenr); | |
3429 | BUG_ON(ret); | |
5d4f98a2 YZ |
3430 | } else { |
3431 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
3432 | parent, root_objectid, owner, | |
3433 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
3434 | BUG_ON(ret); | |
56bec294 | 3435 | } |
925baedd CM |
3436 | return ret; |
3437 | } | |
3438 | ||
87ee04eb CM |
3439 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
3440 | { | |
3441 | u64 mask = ((u64)root->stripesize - 1); | |
3442 | u64 ret = (val + mask) & ~mask; | |
3443 | return ret; | |
3444 | } | |
3445 | ||
fec577fb CM |
3446 | /* |
3447 | * walks the btree of allocated extents and find a hole of a given size. | |
3448 | * The key ins is changed to record the hole: | |
3449 | * ins->objectid == block start | |
62e2749e | 3450 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
3451 | * ins->offset == number of blocks |
3452 | * Any available blocks before search_start are skipped. | |
3453 | */ | |
d397712b | 3454 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
3455 | struct btrfs_root *orig_root, |
3456 | u64 num_bytes, u64 empty_size, | |
3457 | u64 search_start, u64 search_end, | |
3458 | u64 hint_byte, struct btrfs_key *ins, | |
3459 | u64 exclude_start, u64 exclude_nr, | |
3460 | int data) | |
fec577fb | 3461 | { |
80eb234a | 3462 | int ret = 0; |
d397712b | 3463 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 3464 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 3465 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 3466 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 3467 | int allowed_chunk_alloc = 0; |
80eb234a | 3468 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
3469 | int last_ptr_loop = 0; |
3470 | int loop = 0; | |
fec577fb | 3471 | |
db94535d | 3472 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 3473 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
3474 | ins->objectid = 0; |
3475 | ins->offset = 0; | |
b1a4d965 | 3476 | |
2552d17e JB |
3477 | space_info = __find_space_info(root->fs_info, data); |
3478 | ||
0ef3e66b CM |
3479 | if (orig_root->ref_cows || empty_size) |
3480 | allowed_chunk_alloc = 1; | |
3481 | ||
239b14b3 | 3482 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
fa9c0d79 | 3483 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
3484 | if (!btrfs_test_opt(root, SSD)) |
3485 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
3486 | } |
3487 | ||
fa9c0d79 CM |
3488 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { |
3489 | last_ptr = &root->fs_info->data_alloc_cluster; | |
3490 | } | |
0f9dd46c | 3491 | |
239b14b3 | 3492 | if (last_ptr) { |
fa9c0d79 CM |
3493 | spin_lock(&last_ptr->lock); |
3494 | if (last_ptr->block_group) | |
3495 | hint_byte = last_ptr->window_start; | |
3496 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 3497 | } |
fa9c0d79 | 3498 | |
a061fc8d | 3499 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 3500 | search_start = max(search_start, hint_byte); |
0b86a832 | 3501 | |
fa9c0d79 CM |
3502 | if (!last_ptr) { |
3503 | empty_cluster = 0; | |
3504 | loop = 1; | |
3505 | } | |
3506 | ||
2552d17e | 3507 | if (search_start == hint_byte) { |
2552d17e JB |
3508 | block_group = btrfs_lookup_block_group(root->fs_info, |
3509 | search_start); | |
3510 | if (block_group && block_group_bits(block_group, data)) { | |
2552d17e | 3511 | down_read(&space_info->groups_sem); |
44fb5511 CM |
3512 | if (list_empty(&block_group->list) || |
3513 | block_group->ro) { | |
3514 | /* | |
3515 | * someone is removing this block group, | |
3516 | * we can't jump into the have_block_group | |
3517 | * target because our list pointers are not | |
3518 | * valid | |
3519 | */ | |
3520 | btrfs_put_block_group(block_group); | |
3521 | up_read(&space_info->groups_sem); | |
3522 | } else | |
3523 | goto have_block_group; | |
2552d17e | 3524 | } else if (block_group) { |
fa9c0d79 | 3525 | btrfs_put_block_group(block_group); |
2552d17e | 3526 | } |
42e70e7a | 3527 | } |
4366211c | 3528 | |
2552d17e | 3529 | search: |
80eb234a | 3530 | down_read(&space_info->groups_sem); |
2552d17e | 3531 | list_for_each_entry(block_group, &space_info->block_groups, list) { |
6226cb0a | 3532 | u64 offset; |
8a1413a2 | 3533 | |
2552d17e JB |
3534 | atomic_inc(&block_group->count); |
3535 | search_start = block_group->key.objectid; | |
42e70e7a | 3536 | |
2552d17e | 3537 | have_block_group: |
ea6a478e JB |
3538 | if (unlikely(!block_group->cached)) { |
3539 | mutex_lock(&block_group->cache_mutex); | |
3540 | ret = cache_block_group(root, block_group); | |
3541 | mutex_unlock(&block_group->cache_mutex); | |
2552d17e | 3542 | if (ret) { |
fa9c0d79 | 3543 | btrfs_put_block_group(block_group); |
ea6a478e | 3544 | break; |
2552d17e | 3545 | } |
ea6a478e JB |
3546 | } |
3547 | ||
ea6a478e | 3548 | if (unlikely(block_group->ro)) |
2552d17e | 3549 | goto loop; |
0f9dd46c | 3550 | |
fa9c0d79 CM |
3551 | if (last_ptr) { |
3552 | /* | |
3553 | * the refill lock keeps out other | |
3554 | * people trying to start a new cluster | |
3555 | */ | |
3556 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
3557 | if (last_ptr->block_group && |
3558 | (last_ptr->block_group->ro || | |
3559 | !block_group_bits(last_ptr->block_group, data))) { | |
3560 | offset = 0; | |
3561 | goto refill_cluster; | |
3562 | } | |
3563 | ||
fa9c0d79 CM |
3564 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
3565 | num_bytes, search_start); | |
3566 | if (offset) { | |
3567 | /* we have a block, we're done */ | |
3568 | spin_unlock(&last_ptr->refill_lock); | |
3569 | goto checks; | |
3570 | } | |
3571 | ||
3572 | spin_lock(&last_ptr->lock); | |
3573 | /* | |
3574 | * whoops, this cluster doesn't actually point to | |
3575 | * this block group. Get a ref on the block | |
3576 | * group is does point to and try again | |
3577 | */ | |
3578 | if (!last_ptr_loop && last_ptr->block_group && | |
3579 | last_ptr->block_group != block_group) { | |
3580 | ||
3581 | btrfs_put_block_group(block_group); | |
3582 | block_group = last_ptr->block_group; | |
3583 | atomic_inc(&block_group->count); | |
3584 | spin_unlock(&last_ptr->lock); | |
3585 | spin_unlock(&last_ptr->refill_lock); | |
3586 | ||
3587 | last_ptr_loop = 1; | |
3588 | search_start = block_group->key.objectid; | |
44fb5511 CM |
3589 | /* |
3590 | * we know this block group is properly | |
3591 | * in the list because | |
3592 | * btrfs_remove_block_group, drops the | |
3593 | * cluster before it removes the block | |
3594 | * group from the list | |
3595 | */ | |
fa9c0d79 CM |
3596 | goto have_block_group; |
3597 | } | |
3598 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 3599 | refill_cluster: |
fa9c0d79 CM |
3600 | /* |
3601 | * this cluster didn't work out, free it and | |
3602 | * start over | |
3603 | */ | |
3604 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3605 | ||
3606 | last_ptr_loop = 0; | |
3607 | ||
3608 | /* allocate a cluster in this block group */ | |
451d7585 | 3609 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
3610 | block_group, last_ptr, |
3611 | offset, num_bytes, | |
3612 | empty_cluster + empty_size); | |
3613 | if (ret == 0) { | |
3614 | /* | |
3615 | * now pull our allocation out of this | |
3616 | * cluster | |
3617 | */ | |
3618 | offset = btrfs_alloc_from_cluster(block_group, | |
3619 | last_ptr, num_bytes, | |
3620 | search_start); | |
3621 | if (offset) { | |
3622 | /* we found one, proceed */ | |
3623 | spin_unlock(&last_ptr->refill_lock); | |
3624 | goto checks; | |
3625 | } | |
3626 | } | |
3627 | /* | |
3628 | * at this point we either didn't find a cluster | |
3629 | * or we weren't able to allocate a block from our | |
3630 | * cluster. Free the cluster we've been trying | |
3631 | * to use, and go to the next block group | |
3632 | */ | |
3633 | if (loop < 2) { | |
3634 | btrfs_return_cluster_to_free_space(NULL, | |
3635 | last_ptr); | |
3636 | spin_unlock(&last_ptr->refill_lock); | |
3637 | goto loop; | |
3638 | } | |
3639 | spin_unlock(&last_ptr->refill_lock); | |
3640 | } | |
3641 | ||
6226cb0a JB |
3642 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
3643 | num_bytes, empty_size); | |
3644 | if (!offset) | |
2552d17e | 3645 | goto loop; |
fa9c0d79 | 3646 | checks: |
6226cb0a | 3647 | search_start = stripe_align(root, offset); |
80eb234a | 3648 | |
2552d17e | 3649 | /* move on to the next group */ |
6226cb0a JB |
3650 | if (search_start + num_bytes >= search_end) { |
3651 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 3652 | goto loop; |
6226cb0a | 3653 | } |
25179201 | 3654 | |
2552d17e JB |
3655 | /* move on to the next group */ |
3656 | if (search_start + num_bytes > | |
6226cb0a JB |
3657 | block_group->key.objectid + block_group->key.offset) { |
3658 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 3659 | goto loop; |
6226cb0a | 3660 | } |
f5a31e16 | 3661 | |
2552d17e JB |
3662 | if (exclude_nr > 0 && |
3663 | (search_start + num_bytes > exclude_start && | |
3664 | search_start < exclude_start + exclude_nr)) { | |
3665 | search_start = exclude_start + exclude_nr; | |
3666 | ||
6226cb0a | 3667 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e JB |
3668 | /* |
3669 | * if search_start is still in this block group | |
3670 | * then we just re-search this block group | |
f5a31e16 | 3671 | */ |
2552d17e JB |
3672 | if (search_start >= block_group->key.objectid && |
3673 | search_start < (block_group->key.objectid + | |
6226cb0a | 3674 | block_group->key.offset)) |
2552d17e | 3675 | goto have_block_group; |
2552d17e | 3676 | goto loop; |
0f9dd46c | 3677 | } |
0b86a832 | 3678 | |
2552d17e JB |
3679 | ins->objectid = search_start; |
3680 | ins->offset = num_bytes; | |
d2fb3437 | 3681 | |
6226cb0a JB |
3682 | if (offset < search_start) |
3683 | btrfs_add_free_space(block_group, offset, | |
3684 | search_start - offset); | |
3685 | BUG_ON(offset > search_start); | |
3686 | ||
2552d17e | 3687 | /* we are all good, lets return */ |
2552d17e JB |
3688 | break; |
3689 | loop: | |
fa9c0d79 | 3690 | btrfs_put_block_group(block_group); |
2552d17e JB |
3691 | } |
3692 | up_read(&space_info->groups_sem); | |
3693 | ||
fa9c0d79 CM |
3694 | /* loop == 0, try to find a clustered alloc in every block group |
3695 | * loop == 1, try again after forcing a chunk allocation | |
3696 | * loop == 2, set empty_size and empty_cluster to 0 and try again | |
3697 | */ | |
3698 | if (!ins->objectid && loop < 3 && | |
3699 | (empty_size || empty_cluster || allowed_chunk_alloc)) { | |
3700 | if (loop >= 2) { | |
3701 | empty_size = 0; | |
3702 | empty_cluster = 0; | |
3703 | } | |
2552d17e JB |
3704 | |
3705 | if (allowed_chunk_alloc) { | |
3706 | ret = do_chunk_alloc(trans, root, num_bytes + | |
3707 | 2 * 1024 * 1024, data, 1); | |
2552d17e JB |
3708 | allowed_chunk_alloc = 0; |
3709 | } else { | |
3710 | space_info->force_alloc = 1; | |
3711 | } | |
3712 | ||
fa9c0d79 CM |
3713 | if (loop < 3) { |
3714 | loop++; | |
2552d17e | 3715 | goto search; |
fa9c0d79 | 3716 | } |
2552d17e JB |
3717 | ret = -ENOSPC; |
3718 | } else if (!ins->objectid) { | |
3719 | ret = -ENOSPC; | |
f2654de4 | 3720 | } |
0b86a832 | 3721 | |
80eb234a JB |
3722 | /* we found what we needed */ |
3723 | if (ins->objectid) { | |
3724 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 3725 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 3726 | |
fa9c0d79 | 3727 | btrfs_put_block_group(block_group); |
80eb234a | 3728 | ret = 0; |
be744175 | 3729 | } |
be744175 | 3730 | |
0f70abe2 | 3731 | return ret; |
fec577fb | 3732 | } |
ec44a35c | 3733 | |
0f9dd46c JB |
3734 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes) |
3735 | { | |
3736 | struct btrfs_block_group_cache *cache; | |
0f9dd46c | 3737 | |
d397712b CM |
3738 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
3739 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
3740 | info->bytes_pinned - info->bytes_reserved), | |
3741 | (info->full) ? "" : "not "); | |
6a63209f | 3742 | printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," |
21380931 JB |
3743 | " may_use=%llu, used=%llu\n", |
3744 | (unsigned long long)info->total_bytes, | |
3745 | (unsigned long long)info->bytes_pinned, | |
3746 | (unsigned long long)info->bytes_delalloc, | |
3747 | (unsigned long long)info->bytes_may_use, | |
3748 | (unsigned long long)info->bytes_used); | |
0f9dd46c | 3749 | |
80eb234a | 3750 | down_read(&info->groups_sem); |
c6e30871 | 3751 | list_for_each_entry(cache, &info->block_groups, list) { |
0f9dd46c | 3752 | spin_lock(&cache->lock); |
d397712b CM |
3753 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
3754 | "%llu pinned %llu reserved\n", | |
3755 | (unsigned long long)cache->key.objectid, | |
3756 | (unsigned long long)cache->key.offset, | |
3757 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
3758 | (unsigned long long)cache->pinned, | |
3759 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
3760 | btrfs_dump_free_space(cache, bytes); |
3761 | spin_unlock(&cache->lock); | |
3762 | } | |
80eb234a | 3763 | up_read(&info->groups_sem); |
0f9dd46c | 3764 | } |
e8569813 | 3765 | |
e6dcd2dc CM |
3766 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
3767 | struct btrfs_root *root, | |
3768 | u64 num_bytes, u64 min_alloc_size, | |
3769 | u64 empty_size, u64 hint_byte, | |
3770 | u64 search_end, struct btrfs_key *ins, | |
3771 | u64 data) | |
fec577fb CM |
3772 | { |
3773 | int ret; | |
fbdc762b | 3774 | u64 search_start = 0; |
1261ec42 | 3775 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 3776 | |
6a63209f | 3777 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 3778 | again: |
0ef3e66b CM |
3779 | /* |
3780 | * the only place that sets empty_size is btrfs_realloc_node, which | |
3781 | * is not called recursively on allocations | |
3782 | */ | |
3783 | if (empty_size || root->ref_cows) { | |
593060d7 | 3784 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 3785 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
3786 | 2 * 1024 * 1024, |
3787 | BTRFS_BLOCK_GROUP_METADATA | | |
3788 | (info->metadata_alloc_profile & | |
3789 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
3790 | } |
3791 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 3792 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 3793 | } |
0b86a832 | 3794 | |
db94535d CM |
3795 | WARN_ON(num_bytes < root->sectorsize); |
3796 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
3797 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
3798 | trans->alloc_exclude_start, |
3799 | trans->alloc_exclude_nr, data); | |
3b951516 | 3800 | |
98d20f67 CM |
3801 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
3802 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 3803 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 3804 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
3805 | do_chunk_alloc(trans, root->fs_info->extent_root, |
3806 | num_bytes, data, 1); | |
98d20f67 CM |
3807 | goto again; |
3808 | } | |
ec44a35c | 3809 | if (ret) { |
0f9dd46c JB |
3810 | struct btrfs_space_info *sinfo; |
3811 | ||
3812 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
3813 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
3814 | "wanted %llu\n", (unsigned long long)data, | |
3815 | (unsigned long long)num_bytes); | |
0f9dd46c | 3816 | dump_space_info(sinfo, num_bytes); |
925baedd | 3817 | BUG(); |
925baedd | 3818 | } |
0f9dd46c JB |
3819 | |
3820 | return ret; | |
e6dcd2dc CM |
3821 | } |
3822 | ||
65b51a00 CM |
3823 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
3824 | { | |
0f9dd46c | 3825 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 3826 | int ret = 0; |
0f9dd46c | 3827 | |
0f9dd46c JB |
3828 | cache = btrfs_lookup_block_group(root->fs_info, start); |
3829 | if (!cache) { | |
d397712b CM |
3830 | printk(KERN_ERR "Unable to find block group for %llu\n", |
3831 | (unsigned long long)start); | |
0f9dd46c JB |
3832 | return -ENOSPC; |
3833 | } | |
1f3c79a2 LH |
3834 | |
3835 | ret = btrfs_discard_extent(root, start, len); | |
3836 | ||
0f9dd46c | 3837 | btrfs_add_free_space(cache, start, len); |
fa9c0d79 | 3838 | btrfs_put_block_group(cache); |
1a40e23b | 3839 | update_reserved_extents(root, start, len, 0); |
1f3c79a2 LH |
3840 | |
3841 | return ret; | |
65b51a00 CM |
3842 | } |
3843 | ||
e6dcd2dc CM |
3844 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
3845 | struct btrfs_root *root, | |
3846 | u64 num_bytes, u64 min_alloc_size, | |
3847 | u64 empty_size, u64 hint_byte, | |
3848 | u64 search_end, struct btrfs_key *ins, | |
3849 | u64 data) | |
3850 | { | |
3851 | int ret; | |
e6dcd2dc CM |
3852 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, |
3853 | empty_size, hint_byte, search_end, ins, | |
3854 | data); | |
e8569813 | 3855 | update_reserved_extents(root, ins->objectid, ins->offset, 1); |
e6dcd2dc CM |
3856 | return ret; |
3857 | } | |
3858 | ||
5d4f98a2 YZ |
3859 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
3860 | struct btrfs_root *root, | |
3861 | u64 parent, u64 root_objectid, | |
3862 | u64 flags, u64 owner, u64 offset, | |
3863 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
3864 | { |
3865 | int ret; | |
5d4f98a2 | 3866 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 3867 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 3868 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 3869 | struct btrfs_path *path; |
5d4f98a2 YZ |
3870 | struct extent_buffer *leaf; |
3871 | int type; | |
3872 | u32 size; | |
26b8003f | 3873 | |
5d4f98a2 YZ |
3874 | if (parent > 0) |
3875 | type = BTRFS_SHARED_DATA_REF_KEY; | |
3876 | else | |
3877 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 3878 | |
5d4f98a2 | 3879 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
3880 | |
3881 | path = btrfs_alloc_path(); | |
3882 | BUG_ON(!path); | |
47e4bb98 | 3883 | |
b9473439 | 3884 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3885 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
3886 | ins, size); | |
ccd467d6 | 3887 | BUG_ON(ret); |
0f9dd46c | 3888 | |
5d4f98a2 YZ |
3889 | leaf = path->nodes[0]; |
3890 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 3891 | struct btrfs_extent_item); |
5d4f98a2 YZ |
3892 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
3893 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
3894 | btrfs_set_extent_flags(leaf, extent_item, | |
3895 | flags | BTRFS_EXTENT_FLAG_DATA); | |
3896 | ||
3897 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
3898 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
3899 | if (parent > 0) { | |
3900 | struct btrfs_shared_data_ref *ref; | |
3901 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
3902 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
3903 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
3904 | } else { | |
3905 | struct btrfs_extent_data_ref *ref; | |
3906 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3907 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
3908 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
3909 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
3910 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
3911 | } | |
47e4bb98 CM |
3912 | |
3913 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 3914 | btrfs_free_path(path); |
f510cfec | 3915 | |
5d4f98a2 YZ |
3916 | ret = update_block_group(trans, root, ins->objectid, ins->offset, |
3917 | 1, 0); | |
f5947066 | 3918 | if (ret) { |
d397712b CM |
3919 | printk(KERN_ERR "btrfs update block group failed for %llu " |
3920 | "%llu\n", (unsigned long long)ins->objectid, | |
3921 | (unsigned long long)ins->offset); | |
f5947066 CM |
3922 | BUG(); |
3923 | } | |
e6dcd2dc CM |
3924 | return ret; |
3925 | } | |
3926 | ||
5d4f98a2 YZ |
3927 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
3928 | struct btrfs_root *root, | |
3929 | u64 parent, u64 root_objectid, | |
3930 | u64 flags, struct btrfs_disk_key *key, | |
3931 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
3932 | { |
3933 | int ret; | |
5d4f98a2 YZ |
3934 | struct btrfs_fs_info *fs_info = root->fs_info; |
3935 | struct btrfs_extent_item *extent_item; | |
3936 | struct btrfs_tree_block_info *block_info; | |
3937 | struct btrfs_extent_inline_ref *iref; | |
3938 | struct btrfs_path *path; | |
3939 | struct extent_buffer *leaf; | |
3940 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 3941 | |
5d4f98a2 YZ |
3942 | path = btrfs_alloc_path(); |
3943 | BUG_ON(!path); | |
56bec294 | 3944 | |
5d4f98a2 YZ |
3945 | path->leave_spinning = 1; |
3946 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
3947 | ins, size); | |
56bec294 | 3948 | BUG_ON(ret); |
5d4f98a2 YZ |
3949 | |
3950 | leaf = path->nodes[0]; | |
3951 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
3952 | struct btrfs_extent_item); | |
3953 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
3954 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
3955 | btrfs_set_extent_flags(leaf, extent_item, | |
3956 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
3957 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
3958 | ||
3959 | btrfs_set_tree_block_key(leaf, block_info, key); | |
3960 | btrfs_set_tree_block_level(leaf, block_info, level); | |
3961 | ||
3962 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
3963 | if (parent > 0) { | |
3964 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
3965 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
3966 | BTRFS_SHARED_BLOCK_REF_KEY); | |
3967 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
3968 | } else { | |
3969 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
3970 | BTRFS_TREE_BLOCK_REF_KEY); | |
3971 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
3972 | } | |
3973 | ||
3974 | btrfs_mark_buffer_dirty(leaf); | |
3975 | btrfs_free_path(path); | |
3976 | ||
3977 | ret = update_block_group(trans, root, ins->objectid, ins->offset, | |
3978 | 1, 0); | |
3979 | if (ret) { | |
3980 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
3981 | "%llu\n", (unsigned long long)ins->objectid, | |
3982 | (unsigned long long)ins->offset); | |
3983 | BUG(); | |
3984 | } | |
3985 | return ret; | |
3986 | } | |
3987 | ||
3988 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
3989 | struct btrfs_root *root, | |
3990 | u64 root_objectid, u64 owner, | |
3991 | u64 offset, struct btrfs_key *ins) | |
3992 | { | |
3993 | int ret; | |
3994 | ||
3995 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
3996 | ||
3997 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
3998 | 0, root_objectid, owner, offset, | |
3999 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
4000 | return ret; |
4001 | } | |
e02119d5 CM |
4002 | |
4003 | /* | |
4004 | * this is used by the tree logging recovery code. It records that | |
4005 | * an extent has been allocated and makes sure to clear the free | |
4006 | * space cache bits as well | |
4007 | */ | |
5d4f98a2 YZ |
4008 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
4009 | struct btrfs_root *root, | |
4010 | u64 root_objectid, u64 owner, u64 offset, | |
4011 | struct btrfs_key *ins) | |
e02119d5 CM |
4012 | { |
4013 | int ret; | |
4014 | struct btrfs_block_group_cache *block_group; | |
4015 | ||
e02119d5 | 4016 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
ea6a478e | 4017 | mutex_lock(&block_group->cache_mutex); |
e02119d5 | 4018 | cache_block_group(root, block_group); |
ea6a478e | 4019 | mutex_unlock(&block_group->cache_mutex); |
e02119d5 | 4020 | |
ea6a478e JB |
4021 | ret = btrfs_remove_free_space(block_group, ins->objectid, |
4022 | ins->offset); | |
0f9dd46c | 4023 | BUG_ON(ret); |
fa9c0d79 | 4024 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
4025 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
4026 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
4027 | return ret; |
4028 | } | |
4029 | ||
e6dcd2dc CM |
4030 | /* |
4031 | * finds a free extent and does all the dirty work required for allocation | |
4032 | * returns the key for the extent through ins, and a tree buffer for | |
4033 | * the first block of the extent through buf. | |
4034 | * | |
4035 | * returns 0 if everything worked, non-zero otherwise. | |
4036 | */ | |
5d4f98a2 YZ |
4037 | static int alloc_tree_block(struct btrfs_trans_handle *trans, |
4038 | struct btrfs_root *root, | |
4039 | u64 num_bytes, u64 parent, u64 root_objectid, | |
4040 | struct btrfs_disk_key *key, int level, | |
4041 | u64 empty_size, u64 hint_byte, u64 search_end, | |
4042 | struct btrfs_key *ins) | |
e6dcd2dc CM |
4043 | { |
4044 | int ret; | |
5d4f98a2 YZ |
4045 | u64 flags = 0; |
4046 | ||
4047 | ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes, | |
4048 | empty_size, hint_byte, search_end, | |
4049 | ins, 0); | |
e6dcd2dc | 4050 | BUG_ON(ret); |
5d4f98a2 YZ |
4051 | |
4052 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4053 | if (parent == 0) | |
4054 | parent = ins->objectid; | |
4055 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4056 | } else | |
4057 | BUG_ON(parent > 0); | |
4058 | ||
4059 | update_reserved_extents(root, ins->objectid, ins->offset, 1); | |
d00aff00 | 4060 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
5d4f98a2 YZ |
4061 | struct btrfs_delayed_extent_op *extent_op; |
4062 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
4063 | BUG_ON(!extent_op); | |
4064 | if (key) | |
4065 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4066 | else | |
4067 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4068 | extent_op->flags_to_set = flags; | |
4069 | extent_op->update_key = 1; | |
4070 | extent_op->update_flags = 1; | |
4071 | extent_op->is_data = 0; | |
4072 | ||
4073 | ret = btrfs_add_delayed_tree_ref(trans, ins->objectid, | |
4074 | ins->offset, parent, root_objectid, | |
4075 | level, BTRFS_ADD_DELAYED_EXTENT, | |
4076 | extent_op); | |
d00aff00 | 4077 | BUG_ON(ret); |
d00aff00 | 4078 | } |
925baedd | 4079 | return ret; |
fec577fb | 4080 | } |
65b51a00 CM |
4081 | |
4082 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
4083 | struct btrfs_root *root, | |
4008c04a CM |
4084 | u64 bytenr, u32 blocksize, |
4085 | int level) | |
65b51a00 CM |
4086 | { |
4087 | struct extent_buffer *buf; | |
4088 | ||
4089 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
4090 | if (!buf) | |
4091 | return ERR_PTR(-ENOMEM); | |
4092 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 4093 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
4094 | btrfs_tree_lock(buf); |
4095 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
4096 | |
4097 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 4098 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 4099 | |
d0c803c4 CM |
4100 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
4101 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
4102 | buf->start + buf->len - 1, GFP_NOFS); | |
4103 | } else { | |
4104 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 4105 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4106 | } |
65b51a00 | 4107 | trans->blocks_used++; |
b4ce94de | 4108 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4109 | return buf; |
4110 | } | |
4111 | ||
fec577fb CM |
4112 | /* |
4113 | * helper function to allocate a block for a given tree | |
4114 | * returns the tree buffer or NULL. | |
4115 | */ | |
5f39d397 | 4116 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4117 | struct btrfs_root *root, u32 blocksize, |
4118 | u64 parent, u64 root_objectid, | |
4119 | struct btrfs_disk_key *key, int level, | |
4120 | u64 hint, u64 empty_size) | |
fec577fb | 4121 | { |
e2fa7227 | 4122 | struct btrfs_key ins; |
fec577fb | 4123 | int ret; |
5f39d397 | 4124 | struct extent_buffer *buf; |
fec577fb | 4125 | |
5d4f98a2 YZ |
4126 | ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid, |
4127 | key, level, empty_size, hint, (u64)-1, &ins); | |
fec577fb | 4128 | if (ret) { |
54aa1f4d CM |
4129 | BUG_ON(ret > 0); |
4130 | return ERR_PTR(ret); | |
fec577fb | 4131 | } |
55c69072 | 4132 | |
4008c04a CM |
4133 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
4134 | blocksize, level); | |
fec577fb CM |
4135 | return buf; |
4136 | } | |
a28ec197 | 4137 | |
e02119d5 CM |
4138 | int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, |
4139 | struct btrfs_root *root, struct extent_buffer *leaf) | |
6407bf6d | 4140 | { |
5d4f98a2 YZ |
4141 | u64 disk_bytenr; |
4142 | u64 num_bytes; | |
5f39d397 | 4143 | struct btrfs_key key; |
6407bf6d | 4144 | struct btrfs_file_extent_item *fi; |
5d4f98a2 | 4145 | u32 nritems; |
6407bf6d | 4146 | int i; |
6407bf6d CM |
4147 | int ret; |
4148 | ||
5f39d397 CM |
4149 | BUG_ON(!btrfs_is_leaf(leaf)); |
4150 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 | 4151 | |
6407bf6d | 4152 | for (i = 0; i < nritems; i++) { |
e34a5b4f | 4153 | cond_resched(); |
5f39d397 | 4154 | btrfs_item_key_to_cpu(leaf, &key, i); |
bd56b302 CM |
4155 | |
4156 | /* only extents have references, skip everything else */ | |
5f39d397 | 4157 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
6407bf6d | 4158 | continue; |
bd56b302 | 4159 | |
6407bf6d | 4160 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); |
bd56b302 CM |
4161 | |
4162 | /* inline extents live in the btree, they don't have refs */ | |
5f39d397 CM |
4163 | if (btrfs_file_extent_type(leaf, fi) == |
4164 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 4165 | continue; |
bd56b302 | 4166 | |
db94535d | 4167 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
bd56b302 CM |
4168 | |
4169 | /* holes don't have refs */ | |
db94535d | 4170 | if (disk_bytenr == 0) |
3a686375 | 4171 | continue; |
4a096752 | 4172 | |
5d4f98a2 YZ |
4173 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); |
4174 | ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes, | |
4175 | leaf->start, 0, key.objectid, 0); | |
31840ae1 | 4176 | BUG_ON(ret); |
6407bf6d CM |
4177 | } |
4178 | return 0; | |
4179 | } | |
4180 | ||
5d4f98a2 YZ |
4181 | #if 0 |
4182 | ||
d397712b | 4183 | static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, |
e02119d5 CM |
4184 | struct btrfs_root *root, |
4185 | struct btrfs_leaf_ref *ref) | |
31153d81 YZ |
4186 | { |
4187 | int i; | |
4188 | int ret; | |
bd56b302 CM |
4189 | struct btrfs_extent_info *info; |
4190 | struct refsort *sorted; | |
4191 | ||
4192 | if (ref->nritems == 0) | |
4193 | return 0; | |
31153d81 | 4194 | |
bd56b302 CM |
4195 | sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS); |
4196 | for (i = 0; i < ref->nritems; i++) { | |
4197 | sorted[i].bytenr = ref->extents[i].bytenr; | |
4198 | sorted[i].slot = i; | |
4199 | } | |
4200 | sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL); | |
4201 | ||
4202 | /* | |
4203 | * the items in the ref were sorted when the ref was inserted | |
4204 | * into the ref cache, so this is already in order | |
4205 | */ | |
31153d81 | 4206 | for (i = 0; i < ref->nritems; i++) { |
bd56b302 | 4207 | info = ref->extents + sorted[i].slot; |
56bec294 | 4208 | ret = btrfs_free_extent(trans, root, info->bytenr, |
31840ae1 ZY |
4209 | info->num_bytes, ref->bytenr, |
4210 | ref->owner, ref->generation, | |
3bb1a1bc | 4211 | info->objectid, 0); |
2dd3e67b CM |
4212 | |
4213 | atomic_inc(&root->fs_info->throttle_gen); | |
4214 | wake_up(&root->fs_info->transaction_throttle); | |
4215 | cond_resched(); | |
4216 | ||
31153d81 YZ |
4217 | BUG_ON(ret); |
4218 | info++; | |
4219 | } | |
31153d81 | 4220 | |
806638bc | 4221 | kfree(sorted); |
31153d81 YZ |
4222 | return 0; |
4223 | } | |
4224 | ||
5d4f98a2 | 4225 | |
56bec294 CM |
4226 | static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans, |
4227 | struct btrfs_root *root, u64 start, | |
d397712b | 4228 | u64 len, u32 *refs) |
333db94c | 4229 | { |
017e5369 | 4230 | int ret; |
f87f057b | 4231 | |
5d4f98a2 | 4232 | ret = btrfs_lookup_extent_refs(trans, root, start, len, refs); |
f87f057b CM |
4233 | BUG_ON(ret); |
4234 | ||
d397712b | 4235 | #if 0 /* some debugging code in case we see problems here */ |
f87f057b CM |
4236 | /* if the refs count is one, it won't get increased again. But |
4237 | * if the ref count is > 1, someone may be decreasing it at | |
4238 | * the same time we are. | |
4239 | */ | |
4240 | if (*refs != 1) { | |
4241 | struct extent_buffer *eb = NULL; | |
4242 | eb = btrfs_find_create_tree_block(root, start, len); | |
4243 | if (eb) | |
4244 | btrfs_tree_lock(eb); | |
4245 | ||
4246 | mutex_lock(&root->fs_info->alloc_mutex); | |
4247 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
4248 | BUG_ON(ret); | |
4249 | mutex_unlock(&root->fs_info->alloc_mutex); | |
4250 | ||
4251 | if (eb) { | |
4252 | btrfs_tree_unlock(eb); | |
4253 | free_extent_buffer(eb); | |
4254 | } | |
4255 | if (*refs == 1) { | |
d397712b CM |
4256 | printk(KERN_ERR "btrfs block %llu went down to one " |
4257 | "during drop_snap\n", (unsigned long long)start); | |
f87f057b CM |
4258 | } |
4259 | ||
4260 | } | |
4261 | #endif | |
4262 | ||
e7a84565 | 4263 | cond_resched(); |
017e5369 | 4264 | return ret; |
333db94c CM |
4265 | } |
4266 | ||
5d4f98a2 | 4267 | |
bd56b302 CM |
4268 | /* |
4269 | * this is used while deleting old snapshots, and it drops the refs | |
4270 | * on a whole subtree starting from a level 1 node. | |
4271 | * | |
4272 | * The idea is to sort all the leaf pointers, and then drop the | |
4273 | * ref on all the leaves in order. Most of the time the leaves | |
4274 | * will have ref cache entries, so no leaf IOs will be required to | |
4275 | * find the extents they have references on. | |
4276 | * | |
4277 | * For each leaf, any references it has are also dropped in order | |
4278 | * | |
4279 | * This ends up dropping the references in something close to optimal | |
4280 | * order for reading and modifying the extent allocation tree. | |
4281 | */ | |
4282 | static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, | |
4283 | struct btrfs_root *root, | |
4284 | struct btrfs_path *path) | |
4285 | { | |
4286 | u64 bytenr; | |
4287 | u64 root_owner; | |
4288 | u64 root_gen; | |
4289 | struct extent_buffer *eb = path->nodes[1]; | |
4290 | struct extent_buffer *leaf; | |
4291 | struct btrfs_leaf_ref *ref; | |
4292 | struct refsort *sorted = NULL; | |
4293 | int nritems = btrfs_header_nritems(eb); | |
4294 | int ret; | |
4295 | int i; | |
4296 | int refi = 0; | |
4297 | int slot = path->slots[1]; | |
4298 | u32 blocksize = btrfs_level_size(root, 0); | |
4299 | u32 refs; | |
4300 | ||
4301 | if (nritems == 0) | |
4302 | goto out; | |
4303 | ||
4304 | root_owner = btrfs_header_owner(eb); | |
4305 | root_gen = btrfs_header_generation(eb); | |
4306 | sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); | |
4307 | ||
4308 | /* | |
4309 | * step one, sort all the leaf pointers so we don't scribble | |
4310 | * randomly into the extent allocation tree | |
4311 | */ | |
4312 | for (i = slot; i < nritems; i++) { | |
4313 | sorted[refi].bytenr = btrfs_node_blockptr(eb, i); | |
4314 | sorted[refi].slot = i; | |
4315 | refi++; | |
4316 | } | |
4317 | ||
4318 | /* | |
4319 | * nritems won't be zero, but if we're picking up drop_snapshot | |
4320 | * after a crash, slot might be > 0, so double check things | |
4321 | * just in case. | |
4322 | */ | |
4323 | if (refi == 0) | |
4324 | goto out; | |
4325 | ||
4326 | sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); | |
4327 | ||
4328 | /* | |
4329 | * the first loop frees everything the leaves point to | |
4330 | */ | |
4331 | for (i = 0; i < refi; i++) { | |
4332 | u64 ptr_gen; | |
4333 | ||
4334 | bytenr = sorted[i].bytenr; | |
4335 | ||
4336 | /* | |
4337 | * check the reference count on this leaf. If it is > 1 | |
4338 | * we just decrement it below and don't update any | |
4339 | * of the refs the leaf points to. | |
4340 | */ | |
56bec294 CM |
4341 | ret = drop_snap_lookup_refcount(trans, root, bytenr, |
4342 | blocksize, &refs); | |
bd56b302 CM |
4343 | BUG_ON(ret); |
4344 | if (refs != 1) | |
4345 | continue; | |
4346 | ||
4347 | ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); | |
4348 | ||
4349 | /* | |
4350 | * the leaf only had one reference, which means the | |
4351 | * only thing pointing to this leaf is the snapshot | |
4352 | * we're deleting. It isn't possible for the reference | |
4353 | * count to increase again later | |
4354 | * | |
4355 | * The reference cache is checked for the leaf, | |
4356 | * and if found we'll be able to drop any refs held by | |
4357 | * the leaf without needing to read it in. | |
4358 | */ | |
4359 | ref = btrfs_lookup_leaf_ref(root, bytenr); | |
4360 | if (ref && ref->generation != ptr_gen) { | |
4361 | btrfs_free_leaf_ref(root, ref); | |
4362 | ref = NULL; | |
4363 | } | |
4364 | if (ref) { | |
4365 | ret = cache_drop_leaf_ref(trans, root, ref); | |
4366 | BUG_ON(ret); | |
4367 | btrfs_remove_leaf_ref(root, ref); | |
4368 | btrfs_free_leaf_ref(root, ref); | |
4369 | } else { | |
4370 | /* | |
4371 | * the leaf wasn't in the reference cache, so | |
4372 | * we have to read it. | |
4373 | */ | |
4374 | leaf = read_tree_block(root, bytenr, blocksize, | |
4375 | ptr_gen); | |
4376 | ret = btrfs_drop_leaf_ref(trans, root, leaf); | |
4377 | BUG_ON(ret); | |
4378 | free_extent_buffer(leaf); | |
4379 | } | |
4380 | atomic_inc(&root->fs_info->throttle_gen); | |
4381 | wake_up(&root->fs_info->transaction_throttle); | |
4382 | cond_resched(); | |
4383 | } | |
4384 | ||
4385 | /* | |
4386 | * run through the loop again to free the refs on the leaves. | |
4387 | * This is faster than doing it in the loop above because | |
4388 | * the leaves are likely to be clustered together. We end up | |
4389 | * working in nice chunks on the extent allocation tree. | |
4390 | */ | |
4391 | for (i = 0; i < refi; i++) { | |
4392 | bytenr = sorted[i].bytenr; | |
56bec294 | 4393 | ret = btrfs_free_extent(trans, root, bytenr, |
bd56b302 CM |
4394 | blocksize, eb->start, |
4395 | root_owner, root_gen, 0, 1); | |
4396 | BUG_ON(ret); | |
4397 | ||
4398 | atomic_inc(&root->fs_info->throttle_gen); | |
4399 | wake_up(&root->fs_info->transaction_throttle); | |
4400 | cond_resched(); | |
4401 | } | |
4402 | out: | |
4403 | kfree(sorted); | |
4404 | ||
4405 | /* | |
4406 | * update the path to show we've processed the entire level 1 | |
4407 | * node. This will get saved into the root's drop_snapshot_progress | |
4408 | * field so these drops are not repeated again if this transaction | |
4409 | * commits. | |
4410 | */ | |
4411 | path->slots[1] = nritems; | |
4412 | return 0; | |
4413 | } | |
4414 | ||
9aca1d51 CM |
4415 | /* |
4416 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
4417 | * counts as it goes. | |
4418 | */ | |
d397712b | 4419 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4420 | struct btrfs_root *root, |
4421 | struct btrfs_path *path, int *level) | |
20524f02 | 4422 | { |
7bb86316 CM |
4423 | u64 root_owner; |
4424 | u64 root_gen; | |
4425 | u64 bytenr; | |
ca7a79ad | 4426 | u64 ptr_gen; |
5f39d397 CM |
4427 | struct extent_buffer *next; |
4428 | struct extent_buffer *cur; | |
7bb86316 | 4429 | struct extent_buffer *parent; |
db94535d | 4430 | u32 blocksize; |
20524f02 CM |
4431 | int ret; |
4432 | u32 refs; | |
4433 | ||
5caf2a00 CM |
4434 | WARN_ON(*level < 0); |
4435 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
56bec294 | 4436 | ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start, |
db94535d | 4437 | path->nodes[*level]->len, &refs); |
20524f02 CM |
4438 | BUG_ON(ret); |
4439 | if (refs > 1) | |
4440 | goto out; | |
e011599b | 4441 | |
9aca1d51 CM |
4442 | /* |
4443 | * walk down to the last node level and free all the leaves | |
4444 | */ | |
d397712b | 4445 | while (*level >= 0) { |
5caf2a00 CM |
4446 | WARN_ON(*level < 0); |
4447 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 4448 | cur = path->nodes[*level]; |
e011599b | 4449 | |
5f39d397 | 4450 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 4451 | WARN_ON(1); |
e011599b | 4452 | |
7518a238 | 4453 | if (path->slots[*level] >= |
5f39d397 | 4454 | btrfs_header_nritems(cur)) |
20524f02 | 4455 | break; |
bd56b302 CM |
4456 | |
4457 | /* the new code goes down to level 1 and does all the | |
4458 | * leaves pointed to that node in bulk. So, this check | |
4459 | * for level 0 will always be false. | |
4460 | * | |
4461 | * But, the disk format allows the drop_snapshot_progress | |
4462 | * field in the root to leave things in a state where | |
4463 | * a leaf will need cleaning up here. If someone crashes | |
4464 | * with the old code and then boots with the new code, | |
4465 | * we might find a leaf here. | |
4466 | */ | |
6407bf6d | 4467 | if (*level == 0) { |
e02119d5 | 4468 | ret = btrfs_drop_leaf_ref(trans, root, cur); |
6407bf6d CM |
4469 | BUG_ON(ret); |
4470 | break; | |
4471 | } | |
bd56b302 CM |
4472 | |
4473 | /* | |
4474 | * once we get to level one, process the whole node | |
4475 | * at once, including everything below it. | |
4476 | */ | |
4477 | if (*level == 1) { | |
4478 | ret = drop_level_one_refs(trans, root, path); | |
4479 | BUG_ON(ret); | |
4480 | break; | |
4481 | } | |
4482 | ||
db94535d | 4483 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 4484 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 4485 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 4486 | |
56bec294 CM |
4487 | ret = drop_snap_lookup_refcount(trans, root, bytenr, |
4488 | blocksize, &refs); | |
6407bf6d | 4489 | BUG_ON(ret); |
bd56b302 CM |
4490 | |
4491 | /* | |
4492 | * if there is more than one reference, we don't need | |
4493 | * to read that node to drop any references it has. We | |
4494 | * just drop the ref we hold on that node and move on to the | |
4495 | * next slot in this level. | |
4496 | */ | |
6407bf6d | 4497 | if (refs != 1) { |
7bb86316 CM |
4498 | parent = path->nodes[*level]; |
4499 | root_owner = btrfs_header_owner(parent); | |
4500 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 4501 | path->slots[*level]++; |
f87f057b | 4502 | |
56bec294 | 4503 | ret = btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 4504 | blocksize, parent->start, |
3bb1a1bc YZ |
4505 | root_owner, root_gen, |
4506 | *level - 1, 1); | |
20524f02 | 4507 | BUG_ON(ret); |
18e35e0a CM |
4508 | |
4509 | atomic_inc(&root->fs_info->throttle_gen); | |
4510 | wake_up(&root->fs_info->transaction_throttle); | |
2dd3e67b | 4511 | cond_resched(); |
18e35e0a | 4512 | |
20524f02 CM |
4513 | continue; |
4514 | } | |
bd56b302 | 4515 | |
f87f057b | 4516 | /* |
bd56b302 CM |
4517 | * we need to keep freeing things in the next level down. |
4518 | * read the block and loop around to process it | |
f87f057b | 4519 | */ |
bd56b302 | 4520 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); |
5caf2a00 | 4521 | WARN_ON(*level <= 0); |
83e15a28 | 4522 | if (path->nodes[*level-1]) |
5f39d397 | 4523 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 4524 | path->nodes[*level-1] = next; |
5f39d397 | 4525 | *level = btrfs_header_level(next); |
20524f02 | 4526 | path->slots[*level] = 0; |
2dd3e67b | 4527 | cond_resched(); |
20524f02 CM |
4528 | } |
4529 | out: | |
5caf2a00 CM |
4530 | WARN_ON(*level < 0); |
4531 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
4532 | |
4533 | if (path->nodes[*level] == root->node) { | |
7bb86316 | 4534 | parent = path->nodes[*level]; |
31153d81 | 4535 | bytenr = path->nodes[*level]->start; |
7bb86316 CM |
4536 | } else { |
4537 | parent = path->nodes[*level + 1]; | |
31153d81 | 4538 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); |
7bb86316 CM |
4539 | } |
4540 | ||
31153d81 YZ |
4541 | blocksize = btrfs_level_size(root, *level); |
4542 | root_owner = btrfs_header_owner(parent); | |
7bb86316 | 4543 | root_gen = btrfs_header_generation(parent); |
31153d81 | 4544 | |
bd56b302 CM |
4545 | /* |
4546 | * cleanup and free the reference on the last node | |
4547 | * we processed | |
4548 | */ | |
56bec294 | 4549 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
31840ae1 | 4550 | parent->start, root_owner, root_gen, |
3bb1a1bc | 4551 | *level, 1); |
5f39d397 | 4552 | free_extent_buffer(path->nodes[*level]); |
20524f02 | 4553 | path->nodes[*level] = NULL; |
bd56b302 | 4554 | |
20524f02 CM |
4555 | *level += 1; |
4556 | BUG_ON(ret); | |
f87f057b | 4557 | |
e7a84565 | 4558 | cond_resched(); |
20524f02 CM |
4559 | return 0; |
4560 | } | |
5d4f98a2 | 4561 | #endif |
20524f02 | 4562 | |
f82d02d9 YZ |
4563 | /* |
4564 | * helper function for drop_subtree, this function is similar to | |
4565 | * walk_down_tree. The main difference is that it checks reference | |
4566 | * counts while tree blocks are locked. | |
4567 | */ | |
5d4f98a2 YZ |
4568 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, |
4569 | struct btrfs_root *root, | |
4570 | struct btrfs_path *path, int *level) | |
f82d02d9 YZ |
4571 | { |
4572 | struct extent_buffer *next; | |
4573 | struct extent_buffer *cur; | |
4574 | struct extent_buffer *parent; | |
4575 | u64 bytenr; | |
4576 | u64 ptr_gen; | |
5d4f98a2 YZ |
4577 | u64 refs; |
4578 | u64 flags; | |
f82d02d9 | 4579 | u32 blocksize; |
f82d02d9 YZ |
4580 | int ret; |
4581 | ||
4582 | cur = path->nodes[*level]; | |
5d4f98a2 YZ |
4583 | ret = btrfs_lookup_extent_info(trans, root, cur->start, cur->len, |
4584 | &refs, &flags); | |
f82d02d9 YZ |
4585 | BUG_ON(ret); |
4586 | if (refs > 1) | |
4587 | goto out; | |
4588 | ||
5d4f98a2 YZ |
4589 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4590 | ||
f82d02d9 YZ |
4591 | while (*level >= 0) { |
4592 | cur = path->nodes[*level]; | |
4593 | if (*level == 0) { | |
4594 | ret = btrfs_drop_leaf_ref(trans, root, cur); | |
4595 | BUG_ON(ret); | |
4596 | clean_tree_block(trans, root, cur); | |
4597 | break; | |
4598 | } | |
4599 | if (path->slots[*level] >= btrfs_header_nritems(cur)) { | |
4600 | clean_tree_block(trans, root, cur); | |
4601 | break; | |
4602 | } | |
4603 | ||
4604 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
4605 | blocksize = btrfs_level_size(root, *level - 1); | |
4606 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
4607 | ||
4608 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); | |
4609 | btrfs_tree_lock(next); | |
b4ce94de | 4610 | btrfs_set_lock_blocking(next); |
f82d02d9 | 4611 | |
5d4f98a2 YZ |
4612 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
4613 | &refs, &flags); | |
f82d02d9 YZ |
4614 | BUG_ON(ret); |
4615 | if (refs > 1) { | |
4616 | parent = path->nodes[*level]; | |
4617 | ret = btrfs_free_extent(trans, root, bytenr, | |
5d4f98a2 YZ |
4618 | blocksize, parent->start, |
4619 | btrfs_header_owner(parent), | |
4620 | *level - 1, 0); | |
f82d02d9 YZ |
4621 | BUG_ON(ret); |
4622 | path->slots[*level]++; | |
4623 | btrfs_tree_unlock(next); | |
4624 | free_extent_buffer(next); | |
4625 | continue; | |
4626 | } | |
4627 | ||
5d4f98a2 YZ |
4628 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4629 | ||
f82d02d9 YZ |
4630 | *level = btrfs_header_level(next); |
4631 | path->nodes[*level] = next; | |
4632 | path->slots[*level] = 0; | |
4633 | path->locks[*level] = 1; | |
4634 | cond_resched(); | |
4635 | } | |
4636 | out: | |
5d4f98a2 YZ |
4637 | if (path->nodes[*level] == root->node) |
4638 | parent = path->nodes[*level]; | |
4639 | else | |
4640 | parent = path->nodes[*level + 1]; | |
f82d02d9 YZ |
4641 | bytenr = path->nodes[*level]->start; |
4642 | blocksize = path->nodes[*level]->len; | |
4643 | ||
5d4f98a2 YZ |
4644 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent->start, |
4645 | btrfs_header_owner(parent), *level, 0); | |
f82d02d9 YZ |
4646 | BUG_ON(ret); |
4647 | ||
4648 | if (path->locks[*level]) { | |
4649 | btrfs_tree_unlock(path->nodes[*level]); | |
4650 | path->locks[*level] = 0; | |
4651 | } | |
4652 | free_extent_buffer(path->nodes[*level]); | |
4653 | path->nodes[*level] = NULL; | |
4654 | *level += 1; | |
4655 | cond_resched(); | |
4656 | return 0; | |
4657 | } | |
4658 | ||
9aca1d51 CM |
4659 | /* |
4660 | * helper for dropping snapshots. This walks back up the tree in the path | |
4661 | * to find the first node higher up where we haven't yet gone through | |
4662 | * all the slots | |
4663 | */ | |
d397712b | 4664 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 4665 | struct btrfs_root *root, |
f82d02d9 YZ |
4666 | struct btrfs_path *path, |
4667 | int *level, int max_level) | |
20524f02 | 4668 | { |
7bb86316 | 4669 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 CM |
4670 | int i; |
4671 | int slot; | |
4672 | int ret; | |
9f3a7427 | 4673 | |
f82d02d9 | 4674 | for (i = *level; i < max_level && path->nodes[i]; i++) { |
20524f02 | 4675 | slot = path->slots[i]; |
5d4f98a2 | 4676 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
bd56b302 CM |
4677 | /* |
4678 | * there is more work to do in this level. | |
4679 | * Update the drop_progress marker to reflect | |
4680 | * the work we've done so far, and then bump | |
4681 | * the slot number | |
4682 | */ | |
20524f02 | 4683 | path->slots[i]++; |
9f3a7427 | 4684 | WARN_ON(*level == 0); |
5d4f98a2 YZ |
4685 | if (max_level == BTRFS_MAX_LEVEL) { |
4686 | btrfs_node_key(path->nodes[i], | |
4687 | &root_item->drop_progress, | |
4688 | path->slots[i]); | |
4689 | root_item->drop_level = i; | |
4690 | } | |
4691 | *level = i; | |
20524f02 CM |
4692 | return 0; |
4693 | } else { | |
31840ae1 | 4694 | struct extent_buffer *parent; |
bd56b302 CM |
4695 | |
4696 | /* | |
4697 | * this whole node is done, free our reference | |
4698 | * on it and go up one level | |
4699 | */ | |
31840ae1 ZY |
4700 | if (path->nodes[*level] == root->node) |
4701 | parent = path->nodes[*level]; | |
4702 | else | |
4703 | parent = path->nodes[*level + 1]; | |
4704 | ||
5d4f98a2 | 4705 | clean_tree_block(trans, root, path->nodes[i]); |
e089f05c | 4706 | ret = btrfs_free_extent(trans, root, |
5d4f98a2 YZ |
4707 | path->nodes[i]->start, |
4708 | path->nodes[i]->len, | |
4709 | parent->start, | |
4710 | btrfs_header_owner(parent), | |
4711 | *level, 0); | |
6407bf6d | 4712 | BUG_ON(ret); |
f82d02d9 | 4713 | if (path->locks[*level]) { |
5d4f98a2 YZ |
4714 | btrfs_tree_unlock(path->nodes[i]); |
4715 | path->locks[i] = 0; | |
f82d02d9 | 4716 | } |
5d4f98a2 YZ |
4717 | free_extent_buffer(path->nodes[i]); |
4718 | path->nodes[i] = NULL; | |
20524f02 | 4719 | *level = i + 1; |
20524f02 CM |
4720 | } |
4721 | } | |
4722 | return 1; | |
4723 | } | |
4724 | ||
9aca1d51 CM |
4725 | /* |
4726 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
4727 | * the tree freeing any blocks that have a ref count of zero after being | |
4728 | * decremented. | |
4729 | */ | |
e089f05c | 4730 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 4731 | *root) |
20524f02 | 4732 | { |
3768f368 | 4733 | int ret = 0; |
9aca1d51 | 4734 | int wret; |
20524f02 | 4735 | int level; |
5caf2a00 | 4736 | struct btrfs_path *path; |
c3e69d58 | 4737 | int update_count; |
9f3a7427 | 4738 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 4739 | |
5caf2a00 CM |
4740 | path = btrfs_alloc_path(); |
4741 | BUG_ON(!path); | |
20524f02 | 4742 | |
5f39d397 | 4743 | level = btrfs_header_level(root->node); |
9f3a7427 | 4744 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
5d4f98a2 YZ |
4745 | path->nodes[level] = btrfs_lock_root_node(root); |
4746 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 4747 | path->slots[level] = 0; |
5d4f98a2 | 4748 | path->locks[level] = 1; |
9f3a7427 CM |
4749 | } else { |
4750 | struct btrfs_key key; | |
5f39d397 CM |
4751 | struct btrfs_disk_key found_key; |
4752 | struct extent_buffer *node; | |
6702ed49 | 4753 | |
9f3a7427 | 4754 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
4755 | level = root_item->drop_level; |
4756 | path->lowest_level = level; | |
9f3a7427 | 4757 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 4758 | if (wret < 0) { |
9f3a7427 CM |
4759 | ret = wret; |
4760 | goto out; | |
4761 | } | |
5f39d397 CM |
4762 | node = path->nodes[level]; |
4763 | btrfs_node_key(node, &found_key, path->slots[level]); | |
4764 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
4765 | sizeof(found_key))); | |
7d9eb12c CM |
4766 | /* |
4767 | * unlock our path, this is safe because only this | |
4768 | * function is allowed to delete this snapshot | |
4769 | */ | |
5d4f98a2 | 4770 | btrfs_unlock_up_safe(path, 0); |
9f3a7427 | 4771 | } |
d397712b | 4772 | while (1) { |
c3e69d58 | 4773 | unsigned long update; |
5caf2a00 | 4774 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 4775 | if (wret > 0) |
20524f02 | 4776 | break; |
9aca1d51 CM |
4777 | if (wret < 0) |
4778 | ret = wret; | |
4779 | ||
f82d02d9 YZ |
4780 | wret = walk_up_tree(trans, root, path, &level, |
4781 | BTRFS_MAX_LEVEL); | |
9aca1d51 | 4782 | if (wret > 0) |
20524f02 | 4783 | break; |
9aca1d51 CM |
4784 | if (wret < 0) |
4785 | ret = wret; | |
b7ec40d7 CM |
4786 | if (trans->transaction->in_commit || |
4787 | trans->transaction->delayed_refs.flushing) { | |
e7a84565 CM |
4788 | ret = -EAGAIN; |
4789 | break; | |
4790 | } | |
c3e69d58 CM |
4791 | for (update_count = 0; update_count < 16; update_count++) { |
4792 | update = trans->delayed_ref_updates; | |
4793 | trans->delayed_ref_updates = 0; | |
4794 | if (update) | |
4795 | btrfs_run_delayed_refs(trans, root, update); | |
4796 | else | |
4797 | break; | |
4798 | } | |
20524f02 | 4799 | } |
9f3a7427 | 4800 | out: |
5caf2a00 | 4801 | btrfs_free_path(path); |
9aca1d51 | 4802 | return ret; |
20524f02 | 4803 | } |
9078a3e1 | 4804 | |
f82d02d9 YZ |
4805 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
4806 | struct btrfs_root *root, | |
4807 | struct extent_buffer *node, | |
4808 | struct extent_buffer *parent) | |
4809 | { | |
4810 | struct btrfs_path *path; | |
4811 | int level; | |
4812 | int parent_level; | |
4813 | int ret = 0; | |
4814 | int wret; | |
4815 | ||
4816 | path = btrfs_alloc_path(); | |
4817 | BUG_ON(!path); | |
4818 | ||
b9447ef8 | 4819 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
4820 | parent_level = btrfs_header_level(parent); |
4821 | extent_buffer_get(parent); | |
4822 | path->nodes[parent_level] = parent; | |
4823 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
4824 | ||
b9447ef8 | 4825 | btrfs_assert_tree_locked(node); |
f82d02d9 YZ |
4826 | level = btrfs_header_level(node); |
4827 | extent_buffer_get(node); | |
4828 | path->nodes[level] = node; | |
4829 | path->slots[level] = 0; | |
4830 | ||
4831 | while (1) { | |
5d4f98a2 | 4832 | wret = walk_down_tree(trans, root, path, &level); |
f82d02d9 YZ |
4833 | if (wret < 0) |
4834 | ret = wret; | |
4835 | if (wret != 0) | |
4836 | break; | |
4837 | ||
4838 | wret = walk_up_tree(trans, root, path, &level, parent_level); | |
4839 | if (wret < 0) | |
4840 | ret = wret; | |
4841 | if (wret != 0) | |
4842 | break; | |
4843 | } | |
4844 | ||
4845 | btrfs_free_path(path); | |
4846 | return ret; | |
4847 | } | |
4848 | ||
5d4f98a2 | 4849 | #if 0 |
8e7bf94f CM |
4850 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
4851 | unsigned long nr) | |
4852 | { | |
4853 | return min(last, start + nr - 1); | |
4854 | } | |
4855 | ||
d397712b | 4856 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 4857 | u64 len) |
edbd8d4e CM |
4858 | { |
4859 | u64 page_start; | |
4860 | u64 page_end; | |
1a40e23b | 4861 | unsigned long first_index; |
edbd8d4e | 4862 | unsigned long last_index; |
edbd8d4e CM |
4863 | unsigned long i; |
4864 | struct page *page; | |
d1310b2e | 4865 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 4866 | struct file_ra_state *ra; |
3eaa2885 | 4867 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
4868 | unsigned int total_read = 0; |
4869 | unsigned int total_dirty = 0; | |
4870 | int ret = 0; | |
4313b399 CM |
4871 | |
4872 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
4873 | |
4874 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 4875 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
4876 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
4877 | ||
1a40e23b ZY |
4878 | /* make sure the dirty trick played by the caller work */ |
4879 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
4880 | first_index, last_index); | |
4881 | if (ret) | |
4882 | goto out_unlock; | |
8e7bf94f | 4883 | |
4313b399 | 4884 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 4885 | |
1a40e23b ZY |
4886 | for (i = first_index ; i <= last_index; i++) { |
4887 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 4888 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 4889 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
4890 | } |
4891 | total_read++; | |
3eaa2885 CM |
4892 | again: |
4893 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 4894 | BUG_ON(1); |
edbd8d4e | 4895 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 4896 | if (!page) { |
1a40e23b | 4897 | ret = -ENOMEM; |
edbd8d4e | 4898 | goto out_unlock; |
a061fc8d | 4899 | } |
edbd8d4e CM |
4900 | if (!PageUptodate(page)) { |
4901 | btrfs_readpage(NULL, page); | |
4902 | lock_page(page); | |
4903 | if (!PageUptodate(page)) { | |
4904 | unlock_page(page); | |
4905 | page_cache_release(page); | |
1a40e23b | 4906 | ret = -EIO; |
edbd8d4e CM |
4907 | goto out_unlock; |
4908 | } | |
4909 | } | |
ec44a35c | 4910 | wait_on_page_writeback(page); |
3eaa2885 | 4911 | |
edbd8d4e CM |
4912 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
4913 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 4914 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 4915 | |
3eaa2885 CM |
4916 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4917 | if (ordered) { | |
4918 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4919 | unlock_page(page); | |
4920 | page_cache_release(page); | |
4921 | btrfs_start_ordered_extent(inode, ordered, 1); | |
4922 | btrfs_put_ordered_extent(ordered); | |
4923 | goto again; | |
4924 | } | |
4925 | set_page_extent_mapped(page); | |
4926 | ||
1a40e23b ZY |
4927 | if (i == first_index) |
4928 | set_extent_bits(io_tree, page_start, page_end, | |
4929 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 4930 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 4931 | |
a061fc8d | 4932 | set_page_dirty(page); |
1a40e23b | 4933 | total_dirty++; |
edbd8d4e | 4934 | |
d1310b2e | 4935 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
4936 | unlock_page(page); |
4937 | page_cache_release(page); | |
4938 | } | |
4939 | ||
4940 | out_unlock: | |
ec44a35c | 4941 | kfree(ra); |
edbd8d4e | 4942 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
4943 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
4944 | return ret; | |
edbd8d4e CM |
4945 | } |
4946 | ||
d397712b | 4947 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
4948 | struct btrfs_key *extent_key, |
4949 | u64 offset) | |
4950 | { | |
4951 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
4952 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
4953 | struct extent_map *em; | |
6643558d YZ |
4954 | u64 start = extent_key->objectid - offset; |
4955 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 4956 | |
1a40e23b ZY |
4957 | em = alloc_extent_map(GFP_NOFS); |
4958 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 4959 | |
6643558d | 4960 | em->start = start; |
1a40e23b | 4961 | em->len = extent_key->offset; |
c8b97818 | 4962 | em->block_len = extent_key->offset; |
1a40e23b ZY |
4963 | em->block_start = extent_key->objectid; |
4964 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
4965 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
4966 | ||
4967 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 4968 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
4969 | while (1) { |
4970 | int ret; | |
4971 | spin_lock(&em_tree->lock); | |
4972 | ret = add_extent_mapping(em_tree, em); | |
4973 | spin_unlock(&em_tree->lock); | |
4974 | if (ret != -EEXIST) { | |
4975 | free_extent_map(em); | |
bf4ef679 CM |
4976 | break; |
4977 | } | |
6643558d | 4978 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 4979 | } |
6643558d | 4980 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 4981 | |
6643558d | 4982 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 4983 | } |
edbd8d4e | 4984 | |
1a40e23b ZY |
4985 | struct btrfs_ref_path { |
4986 | u64 extent_start; | |
4987 | u64 nodes[BTRFS_MAX_LEVEL]; | |
4988 | u64 root_objectid; | |
4989 | u64 root_generation; | |
4990 | u64 owner_objectid; | |
1a40e23b ZY |
4991 | u32 num_refs; |
4992 | int lowest_level; | |
4993 | int current_level; | |
f82d02d9 YZ |
4994 | int shared_level; |
4995 | ||
4996 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
4997 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 4998 | }; |
7d9eb12c | 4999 | |
1a40e23b | 5000 | struct disk_extent { |
c8b97818 | 5001 | u64 ram_bytes; |
1a40e23b ZY |
5002 | u64 disk_bytenr; |
5003 | u64 disk_num_bytes; | |
5004 | u64 offset; | |
5005 | u64 num_bytes; | |
c8b97818 CM |
5006 | u8 compression; |
5007 | u8 encryption; | |
5008 | u16 other_encoding; | |
1a40e23b | 5009 | }; |
4313b399 | 5010 | |
1a40e23b ZY |
5011 | static int is_cowonly_root(u64 root_objectid) |
5012 | { | |
5013 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
5014 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
5015 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
5016 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
5017 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
5018 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
5019 | return 1; |
5020 | return 0; | |
5021 | } | |
edbd8d4e | 5022 | |
d397712b | 5023 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5024 | struct btrfs_root *extent_root, |
5025 | struct btrfs_ref_path *ref_path, | |
5026 | int first_time) | |
5027 | { | |
5028 | struct extent_buffer *leaf; | |
5029 | struct btrfs_path *path; | |
5030 | struct btrfs_extent_ref *ref; | |
5031 | struct btrfs_key key; | |
5032 | struct btrfs_key found_key; | |
5033 | u64 bytenr; | |
5034 | u32 nritems; | |
5035 | int level; | |
5036 | int ret = 1; | |
edbd8d4e | 5037 | |
1a40e23b ZY |
5038 | path = btrfs_alloc_path(); |
5039 | if (!path) | |
5040 | return -ENOMEM; | |
bf4ef679 | 5041 | |
1a40e23b ZY |
5042 | if (first_time) { |
5043 | ref_path->lowest_level = -1; | |
5044 | ref_path->current_level = -1; | |
f82d02d9 | 5045 | ref_path->shared_level = -1; |
1a40e23b ZY |
5046 | goto walk_up; |
5047 | } | |
5048 | walk_down: | |
5049 | level = ref_path->current_level - 1; | |
5050 | while (level >= -1) { | |
5051 | u64 parent; | |
5052 | if (level < ref_path->lowest_level) | |
5053 | break; | |
bf4ef679 | 5054 | |
d397712b | 5055 | if (level >= 0) |
1a40e23b | 5056 | bytenr = ref_path->nodes[level]; |
d397712b | 5057 | else |
1a40e23b | 5058 | bytenr = ref_path->extent_start; |
1a40e23b | 5059 | BUG_ON(bytenr == 0); |
bf4ef679 | 5060 | |
1a40e23b ZY |
5061 | parent = ref_path->nodes[level + 1]; |
5062 | ref_path->nodes[level + 1] = 0; | |
5063 | ref_path->current_level = level; | |
5064 | BUG_ON(parent == 0); | |
0ef3e66b | 5065 | |
1a40e23b ZY |
5066 | key.objectid = bytenr; |
5067 | key.offset = parent + 1; | |
5068 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5069 | |
1a40e23b ZY |
5070 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5071 | if (ret < 0) | |
edbd8d4e | 5072 | goto out; |
1a40e23b | 5073 | BUG_ON(ret == 0); |
7d9eb12c | 5074 | |
1a40e23b ZY |
5075 | leaf = path->nodes[0]; |
5076 | nritems = btrfs_header_nritems(leaf); | |
5077 | if (path->slots[0] >= nritems) { | |
5078 | ret = btrfs_next_leaf(extent_root, path); | |
5079 | if (ret < 0) | |
5080 | goto out; | |
5081 | if (ret > 0) | |
5082 | goto next; | |
5083 | leaf = path->nodes[0]; | |
5084 | } | |
0ef3e66b | 5085 | |
1a40e23b ZY |
5086 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5087 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
5088 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
5089 | if (level < ref_path->shared_level) | |
5090 | ref_path->shared_level = level; | |
1a40e23b | 5091 | goto found; |
f82d02d9 | 5092 | } |
1a40e23b ZY |
5093 | next: |
5094 | level--; | |
5095 | btrfs_release_path(extent_root, path); | |
d899e052 | 5096 | cond_resched(); |
1a40e23b ZY |
5097 | } |
5098 | /* reached lowest level */ | |
5099 | ret = 1; | |
5100 | goto out; | |
5101 | walk_up: | |
5102 | level = ref_path->current_level; | |
5103 | while (level < BTRFS_MAX_LEVEL - 1) { | |
5104 | u64 ref_objectid; | |
d397712b CM |
5105 | |
5106 | if (level >= 0) | |
1a40e23b | 5107 | bytenr = ref_path->nodes[level]; |
d397712b | 5108 | else |
1a40e23b | 5109 | bytenr = ref_path->extent_start; |
d397712b | 5110 | |
1a40e23b | 5111 | BUG_ON(bytenr == 0); |
edbd8d4e | 5112 | |
1a40e23b ZY |
5113 | key.objectid = bytenr; |
5114 | key.offset = 0; | |
5115 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5116 | |
1a40e23b ZY |
5117 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5118 | if (ret < 0) | |
5119 | goto out; | |
edbd8d4e | 5120 | |
1a40e23b ZY |
5121 | leaf = path->nodes[0]; |
5122 | nritems = btrfs_header_nritems(leaf); | |
5123 | if (path->slots[0] >= nritems) { | |
5124 | ret = btrfs_next_leaf(extent_root, path); | |
5125 | if (ret < 0) | |
5126 | goto out; | |
5127 | if (ret > 0) { | |
5128 | /* the extent was freed by someone */ | |
5129 | if (ref_path->lowest_level == level) | |
5130 | goto out; | |
5131 | btrfs_release_path(extent_root, path); | |
5132 | goto walk_down; | |
5133 | } | |
5134 | leaf = path->nodes[0]; | |
5135 | } | |
edbd8d4e | 5136 | |
1a40e23b ZY |
5137 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5138 | if (found_key.objectid != bytenr || | |
5139 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
5140 | /* the extent was freed by someone */ | |
5141 | if (ref_path->lowest_level == level) { | |
5142 | ret = 1; | |
5143 | goto out; | |
5144 | } | |
5145 | btrfs_release_path(extent_root, path); | |
5146 | goto walk_down; | |
5147 | } | |
5148 | found: | |
5149 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
5150 | struct btrfs_extent_ref); | |
5151 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
5152 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
5153 | if (first_time) { | |
5154 | level = (int)ref_objectid; | |
5155 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
5156 | ref_path->lowest_level = level; | |
5157 | ref_path->current_level = level; | |
5158 | ref_path->nodes[level] = bytenr; | |
5159 | } else { | |
5160 | WARN_ON(ref_objectid != level); | |
5161 | } | |
5162 | } else { | |
5163 | WARN_ON(level != -1); | |
5164 | } | |
5165 | first_time = 0; | |
bf4ef679 | 5166 | |
1a40e23b ZY |
5167 | if (ref_path->lowest_level == level) { |
5168 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
5169 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
5170 | } | |
bf4ef679 | 5171 | |
7d9eb12c | 5172 | /* |
1a40e23b ZY |
5173 | * the block is tree root or the block isn't in reference |
5174 | * counted tree. | |
7d9eb12c | 5175 | */ |
1a40e23b ZY |
5176 | if (found_key.objectid == found_key.offset || |
5177 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
5178 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
5179 | ref_path->root_generation = | |
5180 | btrfs_ref_generation(leaf, ref); | |
5181 | if (level < 0) { | |
5182 | /* special reference from the tree log */ | |
5183 | ref_path->nodes[0] = found_key.offset; | |
5184 | ref_path->current_level = 0; | |
5185 | } | |
5186 | ret = 0; | |
5187 | goto out; | |
5188 | } | |
7d9eb12c | 5189 | |
1a40e23b ZY |
5190 | level++; |
5191 | BUG_ON(ref_path->nodes[level] != 0); | |
5192 | ref_path->nodes[level] = found_key.offset; | |
5193 | ref_path->current_level = level; | |
bf4ef679 | 5194 | |
1a40e23b ZY |
5195 | /* |
5196 | * the reference was created in the running transaction, | |
5197 | * no need to continue walking up. | |
5198 | */ | |
5199 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
5200 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
5201 | ref_path->root_generation = | |
5202 | btrfs_ref_generation(leaf, ref); | |
5203 | ret = 0; | |
5204 | goto out; | |
7d9eb12c CM |
5205 | } |
5206 | ||
1a40e23b | 5207 | btrfs_release_path(extent_root, path); |
d899e052 | 5208 | cond_resched(); |
7d9eb12c | 5209 | } |
1a40e23b ZY |
5210 | /* reached max tree level, but no tree root found. */ |
5211 | BUG(); | |
edbd8d4e | 5212 | out: |
1a40e23b ZY |
5213 | btrfs_free_path(path); |
5214 | return ret; | |
edbd8d4e CM |
5215 | } |
5216 | ||
1a40e23b ZY |
5217 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
5218 | struct btrfs_root *extent_root, | |
5219 | struct btrfs_ref_path *ref_path, | |
5220 | u64 extent_start) | |
a061fc8d | 5221 | { |
1a40e23b ZY |
5222 | memset(ref_path, 0, sizeof(*ref_path)); |
5223 | ref_path->extent_start = extent_start; | |
a061fc8d | 5224 | |
1a40e23b | 5225 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
5226 | } |
5227 | ||
1a40e23b ZY |
5228 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
5229 | struct btrfs_root *extent_root, | |
5230 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 5231 | { |
1a40e23b ZY |
5232 | return __next_ref_path(trans, extent_root, ref_path, 0); |
5233 | } | |
5234 | ||
d397712b | 5235 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
5236 | struct btrfs_key *extent_key, |
5237 | u64 offset, int no_fragment, | |
5238 | struct disk_extent **extents, | |
5239 | int *nr_extents) | |
5240 | { | |
5241 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
5242 | struct btrfs_path *path; | |
5243 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 5244 | struct extent_buffer *leaf; |
1a40e23b ZY |
5245 | struct disk_extent *exts = *extents; |
5246 | struct btrfs_key found_key; | |
5247 | u64 cur_pos; | |
5248 | u64 last_byte; | |
edbd8d4e | 5249 | u32 nritems; |
1a40e23b ZY |
5250 | int nr = 0; |
5251 | int max = *nr_extents; | |
5252 | int ret; | |
edbd8d4e | 5253 | |
1a40e23b ZY |
5254 | WARN_ON(!no_fragment && *extents); |
5255 | if (!exts) { | |
5256 | max = 1; | |
5257 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
5258 | if (!exts) | |
5259 | return -ENOMEM; | |
a061fc8d | 5260 | } |
edbd8d4e | 5261 | |
1a40e23b ZY |
5262 | path = btrfs_alloc_path(); |
5263 | BUG_ON(!path); | |
edbd8d4e | 5264 | |
1a40e23b ZY |
5265 | cur_pos = extent_key->objectid - offset; |
5266 | last_byte = extent_key->objectid + extent_key->offset; | |
5267 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
5268 | cur_pos, 0); | |
5269 | if (ret < 0) | |
5270 | goto out; | |
5271 | if (ret > 0) { | |
5272 | ret = -ENOENT; | |
5273 | goto out; | |
5274 | } | |
edbd8d4e | 5275 | |
1a40e23b | 5276 | while (1) { |
edbd8d4e CM |
5277 | leaf = path->nodes[0]; |
5278 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
5279 | if (path->slots[0] >= nritems) { |
5280 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
5281 | if (ret < 0) |
5282 | goto out; | |
1a40e23b ZY |
5283 | if (ret > 0) |
5284 | break; | |
bf4ef679 | 5285 | leaf = path->nodes[0]; |
a061fc8d | 5286 | } |
edbd8d4e CM |
5287 | |
5288 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
5289 | if (found_key.offset != cur_pos || |
5290 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
5291 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
5292 | break; |
5293 | ||
1a40e23b ZY |
5294 | fi = btrfs_item_ptr(leaf, path->slots[0], |
5295 | struct btrfs_file_extent_item); | |
5296 | if (btrfs_file_extent_type(leaf, fi) != | |
5297 | BTRFS_FILE_EXTENT_REG || | |
5298 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 5299 | break; |
1a40e23b ZY |
5300 | |
5301 | if (nr == max) { | |
5302 | struct disk_extent *old = exts; | |
5303 | max *= 2; | |
5304 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
5305 | memcpy(exts, old, sizeof(*exts) * nr); | |
5306 | if (old != *extents) | |
5307 | kfree(old); | |
a061fc8d | 5308 | } |
edbd8d4e | 5309 | |
1a40e23b ZY |
5310 | exts[nr].disk_bytenr = |
5311 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
5312 | exts[nr].disk_num_bytes = | |
5313 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5314 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
5315 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
5316 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
5317 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
5318 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
5319 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
5320 | fi); | |
d899e052 YZ |
5321 | BUG_ON(exts[nr].offset > 0); |
5322 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
5323 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 5324 | |
1a40e23b ZY |
5325 | cur_pos += exts[nr].num_bytes; |
5326 | nr++; | |
5327 | ||
5328 | if (cur_pos + offset >= last_byte) | |
5329 | break; | |
5330 | ||
5331 | if (no_fragment) { | |
5332 | ret = 1; | |
edbd8d4e | 5333 | goto out; |
1a40e23b ZY |
5334 | } |
5335 | path->slots[0]++; | |
5336 | } | |
5337 | ||
1f80e4db | 5338 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
5339 | if (cur_pos + offset < last_byte) { |
5340 | ret = -ENOENT; | |
5341 | goto out; | |
edbd8d4e CM |
5342 | } |
5343 | ret = 0; | |
5344 | out: | |
1a40e23b ZY |
5345 | btrfs_free_path(path); |
5346 | if (ret) { | |
5347 | if (exts != *extents) | |
5348 | kfree(exts); | |
5349 | } else { | |
5350 | *extents = exts; | |
5351 | *nr_extents = nr; | |
5352 | } | |
5353 | return ret; | |
5354 | } | |
5355 | ||
d397712b | 5356 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5357 | struct btrfs_root *root, |
5358 | struct btrfs_path *path, | |
5359 | struct btrfs_key *extent_key, | |
5360 | struct btrfs_key *leaf_key, | |
5361 | struct btrfs_ref_path *ref_path, | |
5362 | struct disk_extent *new_extents, | |
5363 | int nr_extents) | |
5364 | { | |
5365 | struct extent_buffer *leaf; | |
5366 | struct btrfs_file_extent_item *fi; | |
5367 | struct inode *inode = NULL; | |
5368 | struct btrfs_key key; | |
5369 | u64 lock_start = 0; | |
5370 | u64 lock_end = 0; | |
5371 | u64 num_bytes; | |
5372 | u64 ext_offset; | |
86288a19 | 5373 | u64 search_end = (u64)-1; |
1a40e23b | 5374 | u32 nritems; |
3bb1a1bc | 5375 | int nr_scaned = 0; |
1a40e23b | 5376 | int extent_locked = 0; |
d899e052 | 5377 | int extent_type; |
1a40e23b ZY |
5378 | int ret; |
5379 | ||
3bb1a1bc | 5380 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 5381 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
5382 | if (key.objectid < ref_path->owner_objectid || |
5383 | (key.objectid == ref_path->owner_objectid && | |
5384 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
5385 | key.objectid = ref_path->owner_objectid; | |
5386 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5387 | key.offset = 0; | |
5388 | } | |
1a40e23b ZY |
5389 | } |
5390 | ||
5391 | while (1) { | |
5392 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
5393 | if (ret < 0) | |
5394 | goto out; | |
5395 | ||
5396 | leaf = path->nodes[0]; | |
5397 | nritems = btrfs_header_nritems(leaf); | |
5398 | next: | |
5399 | if (extent_locked && ret > 0) { | |
5400 | /* | |
5401 | * the file extent item was modified by someone | |
5402 | * before the extent got locked. | |
5403 | */ | |
1a40e23b ZY |
5404 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
5405 | lock_end, GFP_NOFS); | |
5406 | extent_locked = 0; | |
5407 | } | |
5408 | ||
5409 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 5410 | if (++nr_scaned > 2) |
1a40e23b ZY |
5411 | break; |
5412 | ||
5413 | BUG_ON(extent_locked); | |
5414 | ret = btrfs_next_leaf(root, path); | |
5415 | if (ret < 0) | |
5416 | goto out; | |
5417 | if (ret > 0) | |
5418 | break; | |
5419 | leaf = path->nodes[0]; | |
5420 | nritems = btrfs_header_nritems(leaf); | |
5421 | } | |
5422 | ||
5423 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
5424 | ||
5425 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
5426 | if ((key.objectid > ref_path->owner_objectid) || | |
5427 | (key.objectid == ref_path->owner_objectid && | |
5428 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 5429 | key.offset >= search_end) |
1a40e23b ZY |
5430 | break; |
5431 | } | |
5432 | ||
5433 | if (inode && key.objectid != inode->i_ino) { | |
5434 | BUG_ON(extent_locked); | |
5435 | btrfs_release_path(root, path); | |
5436 | mutex_unlock(&inode->i_mutex); | |
5437 | iput(inode); | |
5438 | inode = NULL; | |
5439 | continue; | |
5440 | } | |
5441 | ||
5442 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
5443 | path->slots[0]++; | |
5444 | ret = 1; | |
5445 | goto next; | |
5446 | } | |
5447 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
5448 | struct btrfs_file_extent_item); | |
d899e052 YZ |
5449 | extent_type = btrfs_file_extent_type(leaf, fi); |
5450 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
5451 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
5452 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
5453 | extent_key->objectid)) { | |
5454 | path->slots[0]++; | |
5455 | ret = 1; | |
5456 | goto next; | |
5457 | } | |
5458 | ||
5459 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
5460 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
5461 | ||
86288a19 YZ |
5462 | if (search_end == (u64)-1) { |
5463 | search_end = key.offset - ext_offset + | |
5464 | btrfs_file_extent_ram_bytes(leaf, fi); | |
5465 | } | |
1a40e23b ZY |
5466 | |
5467 | if (!extent_locked) { | |
5468 | lock_start = key.offset; | |
5469 | lock_end = lock_start + num_bytes - 1; | |
5470 | } else { | |
6643558d YZ |
5471 | if (lock_start > key.offset || |
5472 | lock_end + 1 < key.offset + num_bytes) { | |
5473 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
5474 | lock_start, lock_end, GFP_NOFS); | |
5475 | extent_locked = 0; | |
5476 | } | |
1a40e23b ZY |
5477 | } |
5478 | ||
5479 | if (!inode) { | |
5480 | btrfs_release_path(root, path); | |
5481 | ||
5482 | inode = btrfs_iget_locked(root->fs_info->sb, | |
5483 | key.objectid, root); | |
5484 | if (inode->i_state & I_NEW) { | |
5485 | BTRFS_I(inode)->root = root; | |
5486 | BTRFS_I(inode)->location.objectid = | |
5487 | key.objectid; | |
5488 | BTRFS_I(inode)->location.type = | |
5489 | BTRFS_INODE_ITEM_KEY; | |
5490 | BTRFS_I(inode)->location.offset = 0; | |
5491 | btrfs_read_locked_inode(inode); | |
5492 | unlock_new_inode(inode); | |
5493 | } | |
5494 | /* | |
5495 | * some code call btrfs_commit_transaction while | |
5496 | * holding the i_mutex, so we can't use mutex_lock | |
5497 | * here. | |
5498 | */ | |
5499 | if (is_bad_inode(inode) || | |
5500 | !mutex_trylock(&inode->i_mutex)) { | |
5501 | iput(inode); | |
5502 | inode = NULL; | |
5503 | key.offset = (u64)-1; | |
5504 | goto skip; | |
5505 | } | |
5506 | } | |
5507 | ||
5508 | if (!extent_locked) { | |
5509 | struct btrfs_ordered_extent *ordered; | |
5510 | ||
5511 | btrfs_release_path(root, path); | |
5512 | ||
5513 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
5514 | lock_end, GFP_NOFS); | |
5515 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
5516 | lock_end); | |
5517 | if (ordered && | |
5518 | ordered->file_offset <= lock_end && | |
5519 | ordered->file_offset + ordered->len > lock_start) { | |
5520 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
5521 | lock_start, lock_end, GFP_NOFS); | |
5522 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5523 | btrfs_put_ordered_extent(ordered); | |
5524 | key.offset += num_bytes; | |
5525 | goto skip; | |
5526 | } | |
5527 | if (ordered) | |
5528 | btrfs_put_ordered_extent(ordered); | |
5529 | ||
1a40e23b ZY |
5530 | extent_locked = 1; |
5531 | continue; | |
5532 | } | |
5533 | ||
5534 | if (nr_extents == 1) { | |
5535 | /* update extent pointer in place */ | |
1a40e23b ZY |
5536 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
5537 | new_extents[0].disk_bytenr); | |
5538 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
5539 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
5540 | btrfs_mark_buffer_dirty(leaf); |
5541 | ||
5542 | btrfs_drop_extent_cache(inode, key.offset, | |
5543 | key.offset + num_bytes - 1, 0); | |
5544 | ||
5545 | ret = btrfs_inc_extent_ref(trans, root, | |
5546 | new_extents[0].disk_bytenr, | |
5547 | new_extents[0].disk_num_bytes, | |
5548 | leaf->start, | |
5549 | root->root_key.objectid, | |
5550 | trans->transid, | |
3bb1a1bc | 5551 | key.objectid); |
1a40e23b ZY |
5552 | BUG_ON(ret); |
5553 | ||
5554 | ret = btrfs_free_extent(trans, root, | |
5555 | extent_key->objectid, | |
5556 | extent_key->offset, | |
5557 | leaf->start, | |
5558 | btrfs_header_owner(leaf), | |
5559 | btrfs_header_generation(leaf), | |
3bb1a1bc | 5560 | key.objectid, 0); |
1a40e23b ZY |
5561 | BUG_ON(ret); |
5562 | ||
5563 | btrfs_release_path(root, path); | |
5564 | key.offset += num_bytes; | |
5565 | } else { | |
d899e052 YZ |
5566 | BUG_ON(1); |
5567 | #if 0 | |
1a40e23b ZY |
5568 | u64 alloc_hint; |
5569 | u64 extent_len; | |
5570 | int i; | |
5571 | /* | |
5572 | * drop old extent pointer at first, then insert the | |
5573 | * new pointers one bye one | |
5574 | */ | |
5575 | btrfs_release_path(root, path); | |
5576 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
5577 | key.offset + num_bytes, | |
5578 | key.offset, &alloc_hint); | |
5579 | BUG_ON(ret); | |
5580 | ||
5581 | for (i = 0; i < nr_extents; i++) { | |
5582 | if (ext_offset >= new_extents[i].num_bytes) { | |
5583 | ext_offset -= new_extents[i].num_bytes; | |
5584 | continue; | |
5585 | } | |
5586 | extent_len = min(new_extents[i].num_bytes - | |
5587 | ext_offset, num_bytes); | |
5588 | ||
5589 | ret = btrfs_insert_empty_item(trans, root, | |
5590 | path, &key, | |
5591 | sizeof(*fi)); | |
5592 | BUG_ON(ret); | |
5593 | ||
5594 | leaf = path->nodes[0]; | |
5595 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
5596 | struct btrfs_file_extent_item); | |
5597 | btrfs_set_file_extent_generation(leaf, fi, | |
5598 | trans->transid); | |
5599 | btrfs_set_file_extent_type(leaf, fi, | |
5600 | BTRFS_FILE_EXTENT_REG); | |
5601 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
5602 | new_extents[i].disk_bytenr); | |
5603 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
5604 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
5605 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
5606 | new_extents[i].ram_bytes); | |
5607 | ||
5608 | btrfs_set_file_extent_compression(leaf, fi, | |
5609 | new_extents[i].compression); | |
5610 | btrfs_set_file_extent_encryption(leaf, fi, | |
5611 | new_extents[i].encryption); | |
5612 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
5613 | new_extents[i].other_encoding); | |
5614 | ||
1a40e23b ZY |
5615 | btrfs_set_file_extent_num_bytes(leaf, fi, |
5616 | extent_len); | |
5617 | ext_offset += new_extents[i].offset; | |
5618 | btrfs_set_file_extent_offset(leaf, fi, | |
5619 | ext_offset); | |
5620 | btrfs_mark_buffer_dirty(leaf); | |
5621 | ||
5622 | btrfs_drop_extent_cache(inode, key.offset, | |
5623 | key.offset + extent_len - 1, 0); | |
5624 | ||
5625 | ret = btrfs_inc_extent_ref(trans, root, | |
5626 | new_extents[i].disk_bytenr, | |
5627 | new_extents[i].disk_num_bytes, | |
5628 | leaf->start, | |
5629 | root->root_key.objectid, | |
3bb1a1bc | 5630 | trans->transid, key.objectid); |
1a40e23b ZY |
5631 | BUG_ON(ret); |
5632 | btrfs_release_path(root, path); | |
5633 | ||
a76a3cd4 | 5634 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
5635 | |
5636 | ext_offset = 0; | |
5637 | num_bytes -= extent_len; | |
5638 | key.offset += extent_len; | |
5639 | ||
5640 | if (num_bytes == 0) | |
5641 | break; | |
5642 | } | |
5643 | BUG_ON(i >= nr_extents); | |
d899e052 | 5644 | #endif |
1a40e23b ZY |
5645 | } |
5646 | ||
5647 | if (extent_locked) { | |
1a40e23b ZY |
5648 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
5649 | lock_end, GFP_NOFS); | |
5650 | extent_locked = 0; | |
5651 | } | |
5652 | skip: | |
5653 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 5654 | key.offset >= search_end) |
1a40e23b ZY |
5655 | break; |
5656 | ||
5657 | cond_resched(); | |
5658 | } | |
5659 | ret = 0; | |
5660 | out: | |
5661 | btrfs_release_path(root, path); | |
5662 | if (inode) { | |
5663 | mutex_unlock(&inode->i_mutex); | |
5664 | if (extent_locked) { | |
1a40e23b ZY |
5665 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
5666 | lock_end, GFP_NOFS); | |
5667 | } | |
5668 | iput(inode); | |
5669 | } | |
5670 | return ret; | |
5671 | } | |
5672 | ||
1a40e23b ZY |
5673 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
5674 | struct btrfs_root *root, | |
5675 | struct extent_buffer *buf, u64 orig_start) | |
5676 | { | |
5677 | int level; | |
5678 | int ret; | |
5679 | ||
5680 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
5681 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
5682 | ||
5683 | level = btrfs_header_level(buf); | |
5684 | if (level == 0) { | |
5685 | struct btrfs_leaf_ref *ref; | |
5686 | struct btrfs_leaf_ref *orig_ref; | |
5687 | ||
5688 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
5689 | if (!orig_ref) | |
5690 | return -ENOENT; | |
5691 | ||
5692 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
5693 | if (!ref) { | |
5694 | btrfs_free_leaf_ref(root, orig_ref); | |
5695 | return -ENOMEM; | |
5696 | } | |
5697 | ||
5698 | ref->nritems = orig_ref->nritems; | |
5699 | memcpy(ref->extents, orig_ref->extents, | |
5700 | sizeof(ref->extents[0]) * ref->nritems); | |
5701 | ||
5702 | btrfs_free_leaf_ref(root, orig_ref); | |
5703 | ||
5704 | ref->root_gen = trans->transid; | |
5705 | ref->bytenr = buf->start; | |
5706 | ref->owner = btrfs_header_owner(buf); | |
5707 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 5708 | |
1a40e23b ZY |
5709 | ret = btrfs_add_leaf_ref(root, ref, 0); |
5710 | WARN_ON(ret); | |
5711 | btrfs_free_leaf_ref(root, ref); | |
5712 | } | |
5713 | return 0; | |
5714 | } | |
5715 | ||
d397712b | 5716 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
5717 | struct extent_buffer *leaf, |
5718 | struct btrfs_block_group_cache *group, | |
5719 | struct btrfs_root *target_root) | |
5720 | { | |
5721 | struct btrfs_key key; | |
5722 | struct inode *inode = NULL; | |
5723 | struct btrfs_file_extent_item *fi; | |
5724 | u64 num_bytes; | |
5725 | u64 skip_objectid = 0; | |
5726 | u32 nritems; | |
5727 | u32 i; | |
5728 | ||
5729 | nritems = btrfs_header_nritems(leaf); | |
5730 | for (i = 0; i < nritems; i++) { | |
5731 | btrfs_item_key_to_cpu(leaf, &key, i); | |
5732 | if (key.objectid == skip_objectid || | |
5733 | key.type != BTRFS_EXTENT_DATA_KEY) | |
5734 | continue; | |
5735 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5736 | if (btrfs_file_extent_type(leaf, fi) == | |
5737 | BTRFS_FILE_EXTENT_INLINE) | |
5738 | continue; | |
5739 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
5740 | continue; | |
5741 | if (!inode || inode->i_ino != key.objectid) { | |
5742 | iput(inode); | |
5743 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
5744 | key.objectid, target_root, 1); | |
5745 | } | |
5746 | if (!inode) { | |
5747 | skip_objectid = key.objectid; | |
5748 | continue; | |
5749 | } | |
5750 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
5751 | ||
5752 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
5753 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
5754 | btrfs_drop_extent_cache(inode, key.offset, |
5755 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
5756 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
5757 | key.offset + num_bytes - 1, GFP_NOFS); | |
5758 | cond_resched(); | |
5759 | } | |
5760 | iput(inode); | |
5761 | return 0; | |
5762 | } | |
5763 | ||
d397712b | 5764 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5765 | struct btrfs_root *root, |
5766 | struct extent_buffer *leaf, | |
5767 | struct btrfs_block_group_cache *group, | |
5768 | struct inode *reloc_inode) | |
5769 | { | |
5770 | struct btrfs_key key; | |
5771 | struct btrfs_key extent_key; | |
5772 | struct btrfs_file_extent_item *fi; | |
5773 | struct btrfs_leaf_ref *ref; | |
5774 | struct disk_extent *new_extent; | |
5775 | u64 bytenr; | |
5776 | u64 num_bytes; | |
5777 | u32 nritems; | |
5778 | u32 i; | |
5779 | int ext_index; | |
5780 | int nr_extent; | |
5781 | int ret; | |
5782 | ||
5783 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
5784 | BUG_ON(!new_extent); | |
5785 | ||
5786 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
5787 | BUG_ON(!ref); | |
5788 | ||
5789 | ext_index = -1; | |
5790 | nritems = btrfs_header_nritems(leaf); | |
5791 | for (i = 0; i < nritems; i++) { | |
5792 | btrfs_item_key_to_cpu(leaf, &key, i); | |
5793 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
5794 | continue; | |
5795 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5796 | if (btrfs_file_extent_type(leaf, fi) == | |
5797 | BTRFS_FILE_EXTENT_INLINE) | |
5798 | continue; | |
5799 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5800 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5801 | if (bytenr == 0) | |
5802 | continue; | |
5803 | ||
5804 | ext_index++; | |
5805 | if (bytenr >= group->key.objectid + group->key.offset || | |
5806 | bytenr + num_bytes <= group->key.objectid) | |
5807 | continue; | |
5808 | ||
5809 | extent_key.objectid = bytenr; | |
5810 | extent_key.offset = num_bytes; | |
5811 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
5812 | nr_extent = 1; | |
5813 | ret = get_new_locations(reloc_inode, &extent_key, | |
5814 | group->key.objectid, 1, | |
5815 | &new_extent, &nr_extent); | |
5816 | if (ret > 0) | |
5817 | continue; | |
5818 | BUG_ON(ret < 0); | |
5819 | ||
5820 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
5821 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
5822 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
5823 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
5824 | ||
1a40e23b ZY |
5825 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
5826 | new_extent->disk_bytenr); | |
5827 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
5828 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
5829 | btrfs_mark_buffer_dirty(leaf); |
5830 | ||
5831 | ret = btrfs_inc_extent_ref(trans, root, | |
5832 | new_extent->disk_bytenr, | |
5833 | new_extent->disk_num_bytes, | |
5834 | leaf->start, | |
5835 | root->root_key.objectid, | |
3bb1a1bc | 5836 | trans->transid, key.objectid); |
1a40e23b | 5837 | BUG_ON(ret); |
56bec294 | 5838 | |
1a40e23b ZY |
5839 | ret = btrfs_free_extent(trans, root, |
5840 | bytenr, num_bytes, leaf->start, | |
5841 | btrfs_header_owner(leaf), | |
5842 | btrfs_header_generation(leaf), | |
3bb1a1bc | 5843 | key.objectid, 0); |
1a40e23b ZY |
5844 | BUG_ON(ret); |
5845 | cond_resched(); | |
5846 | } | |
5847 | kfree(new_extent); | |
5848 | BUG_ON(ext_index + 1 != ref->nritems); | |
5849 | btrfs_free_leaf_ref(root, ref); | |
5850 | return 0; | |
5851 | } | |
5852 | ||
f82d02d9 YZ |
5853 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
5854 | struct btrfs_root *root) | |
1a40e23b ZY |
5855 | { |
5856 | struct btrfs_root *reloc_root; | |
f82d02d9 | 5857 | int ret; |
1a40e23b ZY |
5858 | |
5859 | if (root->reloc_root) { | |
5860 | reloc_root = root->reloc_root; | |
5861 | root->reloc_root = NULL; | |
5862 | list_add(&reloc_root->dead_list, | |
5863 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
5864 | |
5865 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
5866 | reloc_root->node->start); | |
5867 | btrfs_set_root_level(&root->root_item, | |
5868 | btrfs_header_level(reloc_root->node)); | |
5869 | memset(&reloc_root->root_item.drop_progress, 0, | |
5870 | sizeof(struct btrfs_disk_key)); | |
5871 | reloc_root->root_item.drop_level = 0; | |
5872 | ||
5873 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
5874 | &reloc_root->root_key, | |
5875 | &reloc_root->root_item); | |
5876 | BUG_ON(ret); | |
1a40e23b ZY |
5877 | } |
5878 | return 0; | |
5879 | } | |
5880 | ||
5881 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
5882 | { | |
5883 | struct btrfs_trans_handle *trans; | |
5884 | struct btrfs_root *reloc_root; | |
5885 | struct btrfs_root *prev_root = NULL; | |
5886 | struct list_head dead_roots; | |
5887 | int ret; | |
5888 | unsigned long nr; | |
5889 | ||
5890 | INIT_LIST_HEAD(&dead_roots); | |
5891 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
5892 | ||
5893 | while (!list_empty(&dead_roots)) { | |
5894 | reloc_root = list_entry(dead_roots.prev, | |
5895 | struct btrfs_root, dead_list); | |
5896 | list_del_init(&reloc_root->dead_list); | |
5897 | ||
5898 | BUG_ON(reloc_root->commit_root != NULL); | |
5899 | while (1) { | |
5900 | trans = btrfs_join_transaction(root, 1); | |
5901 | BUG_ON(!trans); | |
5902 | ||
5903 | mutex_lock(&root->fs_info->drop_mutex); | |
5904 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
5905 | if (ret != -EAGAIN) | |
5906 | break; | |
5907 | mutex_unlock(&root->fs_info->drop_mutex); | |
5908 | ||
5909 | nr = trans->blocks_used; | |
5910 | ret = btrfs_end_transaction(trans, root); | |
5911 | BUG_ON(ret); | |
5912 | btrfs_btree_balance_dirty(root, nr); | |
5913 | } | |
5914 | ||
5915 | free_extent_buffer(reloc_root->node); | |
5916 | ||
5917 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
5918 | &reloc_root->root_key); | |
5919 | BUG_ON(ret); | |
5920 | mutex_unlock(&root->fs_info->drop_mutex); | |
5921 | ||
5922 | nr = trans->blocks_used; | |
5923 | ret = btrfs_end_transaction(trans, root); | |
5924 | BUG_ON(ret); | |
5925 | btrfs_btree_balance_dirty(root, nr); | |
5926 | ||
5927 | kfree(prev_root); | |
5928 | prev_root = reloc_root; | |
5929 | } | |
5930 | if (prev_root) { | |
5931 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
5932 | kfree(prev_root); | |
5933 | } | |
5934 | return 0; | |
5935 | } | |
5936 | ||
5937 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
5938 | { | |
5939 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
5940 | return 0; | |
5941 | } | |
5942 | ||
5943 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
5944 | { | |
5945 | struct btrfs_root *reloc_root; | |
5946 | struct btrfs_trans_handle *trans; | |
5947 | struct btrfs_key location; | |
5948 | int found; | |
5949 | int ret; | |
5950 | ||
5951 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
5952 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
5953 | BUG_ON(ret); | |
5954 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
5955 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
5956 | ||
5957 | if (found) { | |
5958 | trans = btrfs_start_transaction(root, 1); | |
5959 | BUG_ON(!trans); | |
5960 | ret = btrfs_commit_transaction(trans, root); | |
5961 | BUG_ON(ret); | |
5962 | } | |
5963 | ||
5964 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
5965 | location.offset = (u64)-1; | |
5966 | location.type = BTRFS_ROOT_ITEM_KEY; | |
5967 | ||
5968 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
5969 | BUG_ON(!reloc_root); | |
5970 | btrfs_orphan_cleanup(reloc_root); | |
5971 | return 0; | |
5972 | } | |
5973 | ||
d397712b | 5974 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5975 | struct btrfs_root *root) |
5976 | { | |
5977 | struct btrfs_root *reloc_root; | |
5978 | struct extent_buffer *eb; | |
5979 | struct btrfs_root_item *root_item; | |
5980 | struct btrfs_key root_key; | |
5981 | int ret; | |
5982 | ||
5983 | BUG_ON(!root->ref_cows); | |
5984 | if (root->reloc_root) | |
5985 | return 0; | |
5986 | ||
5987 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
5988 | BUG_ON(!root_item); | |
5989 | ||
5990 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
5991 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
5992 | BUG_ON(ret); | |
5993 | ||
5994 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
5995 | root_key.offset = root->root_key.objectid; | |
5996 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
5997 | ||
5998 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
5999 | btrfs_set_root_refs(root_item, 0); | |
6000 | btrfs_set_root_bytenr(root_item, eb->start); | |
6001 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 6002 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
6003 | |
6004 | btrfs_tree_unlock(eb); | |
6005 | free_extent_buffer(eb); | |
6006 | ||
6007 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
6008 | &root_key, root_item); | |
6009 | BUG_ON(ret); | |
6010 | kfree(root_item); | |
6011 | ||
6012 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
6013 | &root_key); | |
6014 | BUG_ON(!reloc_root); | |
6015 | reloc_root->last_trans = trans->transid; | |
6016 | reloc_root->commit_root = NULL; | |
6017 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
6018 | ||
6019 | root->reloc_root = reloc_root; | |
6020 | return 0; | |
6021 | } | |
6022 | ||
6023 | /* | |
6024 | * Core function of space balance. | |
6025 | * | |
6026 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
6027 | * counted roots. There is one reloc tree for each subvol, and all |
6028 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
6029 | * of the latest committed roots of subvols (root->commit_root). | |
6030 | * | |
6031 | * To relocate a tree block referenced by a subvol, there are two steps. | |
6032 | * COW the block through subvol's reloc tree, then update block pointer | |
6033 | * in the subvol to point to the new block. Since all reloc trees share | |
6034 | * same root key objectid, doing special handing for tree blocks owned | |
6035 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
6036 | * we can use the resulting new block directly when the same block is | |
6037 | * required to COW again through other reloc trees. By this way, relocated | |
6038 | * tree blocks are shared between reloc trees, so they are also shared | |
6039 | * between subvols. | |
1a40e23b | 6040 | */ |
d397712b | 6041 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6042 | struct btrfs_root *root, |
6043 | struct btrfs_path *path, | |
6044 | struct btrfs_key *first_key, | |
6045 | struct btrfs_ref_path *ref_path, | |
6046 | struct btrfs_block_group_cache *group, | |
6047 | struct inode *reloc_inode) | |
6048 | { | |
6049 | struct btrfs_root *reloc_root; | |
6050 | struct extent_buffer *eb = NULL; | |
6051 | struct btrfs_key *keys; | |
6052 | u64 *nodes; | |
6053 | int level; | |
f82d02d9 | 6054 | int shared_level; |
1a40e23b | 6055 | int lowest_level = 0; |
1a40e23b ZY |
6056 | int ret; |
6057 | ||
6058 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
6059 | lowest_level = ref_path->owner_objectid; | |
6060 | ||
f82d02d9 | 6061 | if (!root->ref_cows) { |
1a40e23b ZY |
6062 | path->lowest_level = lowest_level; |
6063 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
6064 | BUG_ON(ret < 0); | |
6065 | path->lowest_level = 0; | |
6066 | btrfs_release_path(root, path); | |
6067 | return 0; | |
6068 | } | |
6069 | ||
1a40e23b ZY |
6070 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
6071 | ret = init_reloc_tree(trans, root); | |
6072 | BUG_ON(ret); | |
6073 | reloc_root = root->reloc_root; | |
6074 | ||
f82d02d9 YZ |
6075 | shared_level = ref_path->shared_level; |
6076 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 6077 | |
f82d02d9 YZ |
6078 | keys = ref_path->node_keys; |
6079 | nodes = ref_path->new_nodes; | |
6080 | memset(&keys[shared_level + 1], 0, | |
6081 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
6082 | memset(&nodes[shared_level + 1], 0, | |
6083 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 6084 | |
f82d02d9 YZ |
6085 | if (nodes[lowest_level] == 0) { |
6086 | path->lowest_level = lowest_level; | |
6087 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
6088 | 0, 1); | |
6089 | BUG_ON(ret); | |
6090 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
6091 | eb = path->nodes[level]; | |
6092 | if (!eb || eb == reloc_root->node) | |
6093 | break; | |
6094 | nodes[level] = eb->start; | |
6095 | if (level == 0) | |
6096 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
6097 | else | |
6098 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
6099 | } | |
2b82032c YZ |
6100 | if (nodes[0] && |
6101 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6102 | eb = path->nodes[0]; |
6103 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
6104 | group, reloc_inode); | |
6105 | BUG_ON(ret); | |
6106 | } | |
6107 | btrfs_release_path(reloc_root, path); | |
6108 | } else { | |
1a40e23b | 6109 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 6110 | lowest_level); |
1a40e23b ZY |
6111 | BUG_ON(ret); |
6112 | } | |
6113 | ||
1a40e23b ZY |
6114 | /* |
6115 | * replace tree blocks in the fs tree with tree blocks in | |
6116 | * the reloc tree. | |
6117 | */ | |
6118 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
6119 | BUG_ON(ret < 0); | |
6120 | ||
6121 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6122 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
6123 | 0, 0); | |
6124 | BUG_ON(ret); | |
6125 | extent_buffer_get(path->nodes[0]); | |
6126 | eb = path->nodes[0]; | |
6127 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
6128 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
6129 | BUG_ON(ret); | |
6130 | free_extent_buffer(eb); | |
6131 | } | |
1a40e23b | 6132 | |
f82d02d9 | 6133 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 6134 | path->lowest_level = 0; |
1a40e23b ZY |
6135 | return 0; |
6136 | } | |
6137 | ||
d397712b | 6138 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6139 | struct btrfs_root *root, |
6140 | struct btrfs_path *path, | |
6141 | struct btrfs_key *first_key, | |
6142 | struct btrfs_ref_path *ref_path) | |
6143 | { | |
6144 | int ret; | |
1a40e23b ZY |
6145 | |
6146 | ret = relocate_one_path(trans, root, path, first_key, | |
6147 | ref_path, NULL, NULL); | |
6148 | BUG_ON(ret); | |
6149 | ||
1a40e23b ZY |
6150 | return 0; |
6151 | } | |
6152 | ||
d397712b | 6153 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6154 | struct btrfs_root *extent_root, |
6155 | struct btrfs_path *path, | |
6156 | struct btrfs_key *extent_key) | |
6157 | { | |
6158 | int ret; | |
6159 | ||
1a40e23b ZY |
6160 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
6161 | if (ret) | |
6162 | goto out; | |
6163 | ret = btrfs_del_item(trans, extent_root, path); | |
6164 | out: | |
6165 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
6166 | return ret; |
6167 | } | |
6168 | ||
d397712b | 6169 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
6170 | struct btrfs_ref_path *ref_path) |
6171 | { | |
6172 | struct btrfs_key root_key; | |
6173 | ||
6174 | root_key.objectid = ref_path->root_objectid; | |
6175 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6176 | if (is_cowonly_root(ref_path->root_objectid)) | |
6177 | root_key.offset = 0; | |
6178 | else | |
6179 | root_key.offset = (u64)-1; | |
6180 | ||
6181 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
6182 | } | |
6183 | ||
d397712b | 6184 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
6185 | struct btrfs_path *path, |
6186 | struct btrfs_key *extent_key, | |
6187 | struct btrfs_block_group_cache *group, | |
6188 | struct inode *reloc_inode, int pass) | |
6189 | { | |
6190 | struct btrfs_trans_handle *trans; | |
6191 | struct btrfs_root *found_root; | |
6192 | struct btrfs_ref_path *ref_path = NULL; | |
6193 | struct disk_extent *new_extents = NULL; | |
6194 | int nr_extents = 0; | |
6195 | int loops; | |
6196 | int ret; | |
6197 | int level; | |
6198 | struct btrfs_key first_key; | |
6199 | u64 prev_block = 0; | |
6200 | ||
1a40e23b ZY |
6201 | |
6202 | trans = btrfs_start_transaction(extent_root, 1); | |
6203 | BUG_ON(!trans); | |
6204 | ||
6205 | if (extent_key->objectid == 0) { | |
6206 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
6207 | goto out; | |
6208 | } | |
6209 | ||
6210 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
6211 | if (!ref_path) { | |
d397712b CM |
6212 | ret = -ENOMEM; |
6213 | goto out; | |
1a40e23b ZY |
6214 | } |
6215 | ||
6216 | for (loops = 0; ; loops++) { | |
6217 | if (loops == 0) { | |
6218 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
6219 | extent_key->objectid); | |
6220 | } else { | |
6221 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
6222 | } | |
6223 | if (ret < 0) | |
6224 | goto out; | |
6225 | if (ret > 0) | |
6226 | break; | |
6227 | ||
6228 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
6229 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
6230 | continue; | |
6231 | ||
6232 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
6233 | BUG_ON(!found_root); | |
6234 | /* | |
6235 | * for reference counted tree, only process reference paths | |
6236 | * rooted at the latest committed root. | |
6237 | */ | |
6238 | if (found_root->ref_cows && | |
6239 | ref_path->root_generation != found_root->root_key.offset) | |
6240 | continue; | |
6241 | ||
6242 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
6243 | if (pass == 0) { | |
6244 | /* | |
6245 | * copy data extents to new locations | |
6246 | */ | |
6247 | u64 group_start = group->key.objectid; | |
6248 | ret = relocate_data_extent(reloc_inode, | |
6249 | extent_key, | |
6250 | group_start); | |
6251 | if (ret < 0) | |
6252 | goto out; | |
6253 | break; | |
6254 | } | |
6255 | level = 0; | |
6256 | } else { | |
6257 | level = ref_path->owner_objectid; | |
6258 | } | |
6259 | ||
6260 | if (prev_block != ref_path->nodes[level]) { | |
6261 | struct extent_buffer *eb; | |
6262 | u64 block_start = ref_path->nodes[level]; | |
6263 | u64 block_size = btrfs_level_size(found_root, level); | |
6264 | ||
6265 | eb = read_tree_block(found_root, block_start, | |
6266 | block_size, 0); | |
6267 | btrfs_tree_lock(eb); | |
6268 | BUG_ON(level != btrfs_header_level(eb)); | |
6269 | ||
6270 | if (level == 0) | |
6271 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
6272 | else | |
6273 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
6274 | ||
6275 | btrfs_tree_unlock(eb); | |
6276 | free_extent_buffer(eb); | |
6277 | prev_block = block_start; | |
6278 | } | |
6279 | ||
24562425 | 6280 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 6281 | btrfs_record_root_in_trans(found_root); |
24562425 | 6282 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
6283 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
6284 | /* | |
6285 | * try to update data extent references while | |
6286 | * keeping metadata shared between snapshots. | |
6287 | */ | |
6288 | if (pass == 1) { | |
6289 | ret = relocate_one_path(trans, found_root, | |
6290 | path, &first_key, ref_path, | |
6291 | group, reloc_inode); | |
6292 | if (ret < 0) | |
6293 | goto out; | |
6294 | continue; | |
6295 | } | |
1a40e23b ZY |
6296 | /* |
6297 | * use fallback method to process the remaining | |
6298 | * references. | |
6299 | */ | |
6300 | if (!new_extents) { | |
6301 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
6302 | new_extents = kmalloc(sizeof(*new_extents), |
6303 | GFP_NOFS); | |
6304 | nr_extents = 1; | |
1a40e23b ZY |
6305 | ret = get_new_locations(reloc_inode, |
6306 | extent_key, | |
d899e052 | 6307 | group_start, 1, |
1a40e23b ZY |
6308 | &new_extents, |
6309 | &nr_extents); | |
d899e052 | 6310 | if (ret) |
1a40e23b ZY |
6311 | goto out; |
6312 | } | |
1a40e23b ZY |
6313 | ret = replace_one_extent(trans, found_root, |
6314 | path, extent_key, | |
6315 | &first_key, ref_path, | |
6316 | new_extents, nr_extents); | |
e4404d6e | 6317 | } else { |
1a40e23b ZY |
6318 | ret = relocate_tree_block(trans, found_root, path, |
6319 | &first_key, ref_path); | |
1a40e23b ZY |
6320 | } |
6321 | if (ret < 0) | |
6322 | goto out; | |
6323 | } | |
6324 | ret = 0; | |
6325 | out: | |
6326 | btrfs_end_transaction(trans, extent_root); | |
6327 | kfree(new_extents); | |
6328 | kfree(ref_path); | |
1a40e23b ZY |
6329 | return ret; |
6330 | } | |
5d4f98a2 | 6331 | #endif |
1a40e23b | 6332 | |
ec44a35c CM |
6333 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6334 | { | |
6335 | u64 num_devices; | |
6336 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6337 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6338 | ||
2b82032c | 6339 | num_devices = root->fs_info->fs_devices->rw_devices; |
ec44a35c CM |
6340 | if (num_devices == 1) { |
6341 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6342 | stripped = flags & ~stripped; | |
6343 | ||
6344 | /* turn raid0 into single device chunks */ | |
6345 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6346 | return stripped; | |
6347 | ||
6348 | /* turn mirroring into duplication */ | |
6349 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6350 | BTRFS_BLOCK_GROUP_RAID10)) | |
6351 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
6352 | return flags; | |
6353 | } else { | |
6354 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6355 | if (flags & stripped) |
6356 | return flags; | |
6357 | ||
6358 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6359 | stripped = flags & ~stripped; | |
6360 | ||
6361 | /* switch duplicated blocks with raid1 */ | |
6362 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6363 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6364 | ||
6365 | /* turn single device chunks into raid0 */ | |
6366 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
6367 | } | |
6368 | return flags; | |
6369 | } | |
6370 | ||
b2950863 | 6371 | static int __alloc_chunk_for_shrink(struct btrfs_root *root, |
0ef3e66b CM |
6372 | struct btrfs_block_group_cache *shrink_block_group, |
6373 | int force) | |
6374 | { | |
6375 | struct btrfs_trans_handle *trans; | |
6376 | u64 new_alloc_flags; | |
6377 | u64 calc; | |
6378 | ||
c286ac48 | 6379 | spin_lock(&shrink_block_group->lock); |
5d4f98a2 YZ |
6380 | if (btrfs_block_group_used(&shrink_block_group->item) + |
6381 | shrink_block_group->reserved > 0) { | |
c286ac48 | 6382 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 6383 | |
0ef3e66b | 6384 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 6385 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 6386 | |
0ef3e66b CM |
6387 | new_alloc_flags = update_block_group_flags(root, |
6388 | shrink_block_group->flags); | |
6389 | if (new_alloc_flags != shrink_block_group->flags) { | |
6390 | calc = | |
6391 | btrfs_block_group_used(&shrink_block_group->item); | |
6392 | } else { | |
6393 | calc = shrink_block_group->key.offset; | |
6394 | } | |
c286ac48 CM |
6395 | spin_unlock(&shrink_block_group->lock); |
6396 | ||
0ef3e66b CM |
6397 | do_chunk_alloc(trans, root->fs_info->extent_root, |
6398 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 6399 | |
0ef3e66b | 6400 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
6401 | } else |
6402 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
6403 | return 0; |
6404 | } | |
6405 | ||
5d4f98a2 YZ |
6406 | |
6407 | int btrfs_prepare_block_group_relocation(struct btrfs_root *root, | |
6408 | struct btrfs_block_group_cache *group) | |
6409 | ||
6410 | { | |
6411 | __alloc_chunk_for_shrink(root, group, 1); | |
6412 | set_block_group_readonly(group); | |
6413 | return 0; | |
6414 | } | |
6415 | ||
6416 | #if 0 | |
1a40e23b ZY |
6417 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, |
6418 | struct btrfs_root *root, | |
6419 | u64 objectid, u64 size) | |
6420 | { | |
6421 | struct btrfs_path *path; | |
6422 | struct btrfs_inode_item *item; | |
6423 | struct extent_buffer *leaf; | |
6424 | int ret; | |
6425 | ||
6426 | path = btrfs_alloc_path(); | |
6427 | if (!path) | |
6428 | return -ENOMEM; | |
6429 | ||
b9473439 | 6430 | path->leave_spinning = 1; |
1a40e23b ZY |
6431 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); |
6432 | if (ret) | |
6433 | goto out; | |
6434 | ||
6435 | leaf = path->nodes[0]; | |
6436 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
6437 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
6438 | btrfs_set_inode_generation(leaf, item, 1); | |
6439 | btrfs_set_inode_size(leaf, item, size); | |
6440 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); | |
0403e47e | 6441 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS); |
1a40e23b ZY |
6442 | btrfs_mark_buffer_dirty(leaf); |
6443 | btrfs_release_path(root, path); | |
6444 | out: | |
6445 | btrfs_free_path(path); | |
6446 | return ret; | |
6447 | } | |
6448 | ||
d397712b | 6449 | static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
6450 | struct btrfs_block_group_cache *group) |
6451 | { | |
6452 | struct inode *inode = NULL; | |
6453 | struct btrfs_trans_handle *trans; | |
6454 | struct btrfs_root *root; | |
6455 | struct btrfs_key root_key; | |
6456 | u64 objectid = BTRFS_FIRST_FREE_OBJECTID; | |
6457 | int err = 0; | |
6458 | ||
6459 | root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
6460 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6461 | root_key.offset = (u64)-1; | |
6462 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
6463 | if (IS_ERR(root)) | |
6464 | return ERR_CAST(root); | |
6465 | ||
6466 | trans = btrfs_start_transaction(root, 1); | |
6467 | BUG_ON(!trans); | |
6468 | ||
6469 | err = btrfs_find_free_objectid(trans, root, objectid, &objectid); | |
6470 | if (err) | |
6471 | goto out; | |
6472 | ||
6473 | err = __insert_orphan_inode(trans, root, objectid, group->key.offset); | |
6474 | BUG_ON(err); | |
6475 | ||
6476 | err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0, | |
c8b97818 CM |
6477 | group->key.offset, 0, group->key.offset, |
6478 | 0, 0, 0); | |
1a40e23b ZY |
6479 | BUG_ON(err); |
6480 | ||
6481 | inode = btrfs_iget_locked(root->fs_info->sb, objectid, root); | |
6482 | if (inode->i_state & I_NEW) { | |
6483 | BTRFS_I(inode)->root = root; | |
6484 | BTRFS_I(inode)->location.objectid = objectid; | |
6485 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
6486 | BTRFS_I(inode)->location.offset = 0; | |
6487 | btrfs_read_locked_inode(inode); | |
6488 | unlock_new_inode(inode); | |
6489 | BUG_ON(is_bad_inode(inode)); | |
6490 | } else { | |
6491 | BUG_ON(1); | |
6492 | } | |
17d217fe | 6493 | BTRFS_I(inode)->index_cnt = group->key.objectid; |
1a40e23b ZY |
6494 | |
6495 | err = btrfs_orphan_add(trans, inode); | |
6496 | out: | |
6497 | btrfs_end_transaction(trans, root); | |
6498 | if (err) { | |
6499 | if (inode) | |
6500 | iput(inode); | |
6501 | inode = ERR_PTR(err); | |
6502 | } | |
6503 | return inode; | |
6504 | } | |
6505 | ||
17d217fe YZ |
6506 | int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) |
6507 | { | |
6508 | ||
6509 | struct btrfs_ordered_sum *sums; | |
6510 | struct btrfs_sector_sum *sector_sum; | |
6511 | struct btrfs_ordered_extent *ordered; | |
6512 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6513 | struct list_head list; | |
6514 | size_t offset; | |
6515 | int ret; | |
6516 | u64 disk_bytenr; | |
6517 | ||
6518 | INIT_LIST_HEAD(&list); | |
6519 | ||
6520 | ordered = btrfs_lookup_ordered_extent(inode, file_pos); | |
6521 | BUG_ON(ordered->file_offset != file_pos || ordered->len != len); | |
6522 | ||
6523 | disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; | |
07d400a6 | 6524 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, |
17d217fe YZ |
6525 | disk_bytenr + len - 1, &list); |
6526 | ||
6527 | while (!list_empty(&list)) { | |
6528 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
6529 | list_del_init(&sums->list); | |
6530 | ||
6531 | sector_sum = sums->sums; | |
6532 | sums->bytenr = ordered->start; | |
6533 | ||
6534 | offset = 0; | |
6535 | while (offset < sums->len) { | |
6536 | sector_sum->bytenr += ordered->start - disk_bytenr; | |
6537 | sector_sum++; | |
6538 | offset += root->sectorsize; | |
6539 | } | |
6540 | ||
6541 | btrfs_add_ordered_sum(inode, ordered, sums); | |
6542 | } | |
6543 | btrfs_put_ordered_extent(ordered); | |
6544 | return 0; | |
6545 | } | |
6546 | ||
1a40e23b | 6547 | int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) |
edbd8d4e CM |
6548 | { |
6549 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 6550 | struct btrfs_path *path; |
1a40e23b ZY |
6551 | struct btrfs_fs_info *info = root->fs_info; |
6552 | struct extent_buffer *leaf; | |
6553 | struct inode *reloc_inode; | |
6554 | struct btrfs_block_group_cache *block_group; | |
6555 | struct btrfs_key key; | |
d899e052 | 6556 | u64 skipped; |
edbd8d4e CM |
6557 | u64 cur_byte; |
6558 | u64 total_found; | |
edbd8d4e CM |
6559 | u32 nritems; |
6560 | int ret; | |
a061fc8d | 6561 | int progress; |
1a40e23b | 6562 | int pass = 0; |
edbd8d4e | 6563 | |
1a40e23b ZY |
6564 | root = root->fs_info->extent_root; |
6565 | ||
6566 | block_group = btrfs_lookup_block_group(info, group_start); | |
6567 | BUG_ON(!block_group); | |
8f18cf13 | 6568 | |
d397712b | 6569 | printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n", |
1a40e23b ZY |
6570 | (unsigned long long)block_group->key.objectid, |
6571 | (unsigned long long)block_group->flags); | |
8f18cf13 | 6572 | |
edbd8d4e | 6573 | path = btrfs_alloc_path(); |
1a40e23b | 6574 | BUG_ON(!path); |
edbd8d4e | 6575 | |
1a40e23b ZY |
6576 | reloc_inode = create_reloc_inode(info, block_group); |
6577 | BUG_ON(IS_ERR(reloc_inode)); | |
323da79c | 6578 | |
1a40e23b | 6579 | __alloc_chunk_for_shrink(root, block_group, 1); |
c146afad | 6580 | set_block_group_readonly(block_group); |
323da79c | 6581 | |
1a40e23b ZY |
6582 | btrfs_start_delalloc_inodes(info->tree_root); |
6583 | btrfs_wait_ordered_extents(info->tree_root, 0); | |
6584 | again: | |
d899e052 | 6585 | skipped = 0; |
edbd8d4e | 6586 | total_found = 0; |
a061fc8d | 6587 | progress = 0; |
1a40e23b | 6588 | key.objectid = block_group->key.objectid; |
edbd8d4e CM |
6589 | key.offset = 0; |
6590 | key.type = 0; | |
73e48b27 | 6591 | cur_byte = key.objectid; |
4313b399 | 6592 | |
1a40e23b ZY |
6593 | trans = btrfs_start_transaction(info->tree_root, 1); |
6594 | btrfs_commit_transaction(trans, info->tree_root); | |
ea8c2819 | 6595 | |
1a40e23b ZY |
6596 | mutex_lock(&root->fs_info->cleaner_mutex); |
6597 | btrfs_clean_old_snapshots(info->tree_root); | |
6598 | btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); | |
6599 | mutex_unlock(&root->fs_info->cleaner_mutex); | |
ea8c2819 | 6600 | |
56bec294 CM |
6601 | trans = btrfs_start_transaction(info->tree_root, 1); |
6602 | btrfs_commit_transaction(trans, info->tree_root); | |
6603 | ||
d397712b | 6604 | while (1) { |
edbd8d4e CM |
6605 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6606 | if (ret < 0) | |
6607 | goto out; | |
7d9eb12c | 6608 | next: |
edbd8d4e | 6609 | leaf = path->nodes[0]; |
73e48b27 | 6610 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
6611 | if (path->slots[0] >= nritems) { |
6612 | ret = btrfs_next_leaf(root, path); | |
6613 | if (ret < 0) | |
6614 | goto out; | |
6615 | if (ret == 1) { | |
6616 | ret = 0; | |
6617 | break; | |
edbd8d4e | 6618 | } |
73e48b27 Y |
6619 | leaf = path->nodes[0]; |
6620 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 6621 | } |
73e48b27 | 6622 | |
1a40e23b | 6623 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
725c8463 | 6624 | |
1a40e23b ZY |
6625 | if (key.objectid >= block_group->key.objectid + |
6626 | block_group->key.offset) | |
8f18cf13 CM |
6627 | break; |
6628 | ||
725c8463 | 6629 | if (progress && need_resched()) { |
725c8463 | 6630 | btrfs_release_path(root, path); |
1a40e23b | 6631 | cond_resched(); |
725c8463 | 6632 | progress = 0; |
1a40e23b | 6633 | continue; |
725c8463 CM |
6634 | } |
6635 | progress = 1; | |
6636 | ||
1a40e23b ZY |
6637 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY || |
6638 | key.objectid + key.offset <= cur_byte) { | |
edbd8d4e | 6639 | path->slots[0]++; |
edbd8d4e CM |
6640 | goto next; |
6641 | } | |
73e48b27 | 6642 | |
edbd8d4e | 6643 | total_found++; |
1a40e23b | 6644 | cur_byte = key.objectid + key.offset; |
edbd8d4e | 6645 | btrfs_release_path(root, path); |
edbd8d4e | 6646 | |
1a40e23b ZY |
6647 | __alloc_chunk_for_shrink(root, block_group, 0); |
6648 | ret = relocate_one_extent(root, path, &key, block_group, | |
6649 | reloc_inode, pass); | |
6650 | BUG_ON(ret < 0); | |
d899e052 YZ |
6651 | if (ret > 0) |
6652 | skipped++; | |
edbd8d4e | 6653 | |
1a40e23b ZY |
6654 | key.objectid = cur_byte; |
6655 | key.type = 0; | |
6656 | key.offset = 0; | |
edbd8d4e CM |
6657 | } |
6658 | ||
1a40e23b | 6659 | btrfs_release_path(root, path); |
7d9eb12c | 6660 | |
1a40e23b ZY |
6661 | if (pass == 0) { |
6662 | btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); | |
6663 | invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); | |
8e8a1e31 | 6664 | } |
73e48b27 | 6665 | |
1a40e23b | 6666 | if (total_found > 0) { |
d397712b | 6667 | printk(KERN_INFO "btrfs found %llu extents in pass %d\n", |
1a40e23b ZY |
6668 | (unsigned long long)total_found, pass); |
6669 | pass++; | |
d899e052 YZ |
6670 | if (total_found == skipped && pass > 2) { |
6671 | iput(reloc_inode); | |
6672 | reloc_inode = create_reloc_inode(info, block_group); | |
6673 | pass = 0; | |
6674 | } | |
1a40e23b | 6675 | goto again; |
0f9dd46c | 6676 | } |
0ef3e66b | 6677 | |
1a40e23b ZY |
6678 | /* delete reloc_inode */ |
6679 | iput(reloc_inode); | |
6680 | ||
6681 | /* unpin extents in this range */ | |
6682 | trans = btrfs_start_transaction(info->tree_root, 1); | |
6683 | btrfs_commit_transaction(trans, info->tree_root); | |
0ef3e66b | 6684 | |
1a40e23b ZY |
6685 | spin_lock(&block_group->lock); |
6686 | WARN_ON(block_group->pinned > 0); | |
6687 | WARN_ON(block_group->reserved > 0); | |
6688 | WARN_ON(btrfs_block_group_used(&block_group->item) > 0); | |
6689 | spin_unlock(&block_group->lock); | |
fa9c0d79 | 6690 | btrfs_put_block_group(block_group); |
1a40e23b | 6691 | ret = 0; |
edbd8d4e | 6692 | out: |
1a40e23b | 6693 | btrfs_free_path(path); |
edbd8d4e CM |
6694 | return ret; |
6695 | } | |
5d4f98a2 | 6696 | #endif |
edbd8d4e | 6697 | |
b2950863 CH |
6698 | static int find_first_block_group(struct btrfs_root *root, |
6699 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 6700 | { |
925baedd | 6701 | int ret = 0; |
0b86a832 CM |
6702 | struct btrfs_key found_key; |
6703 | struct extent_buffer *leaf; | |
6704 | int slot; | |
edbd8d4e | 6705 | |
0b86a832 CM |
6706 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
6707 | if (ret < 0) | |
925baedd CM |
6708 | goto out; |
6709 | ||
d397712b | 6710 | while (1) { |
0b86a832 | 6711 | slot = path->slots[0]; |
edbd8d4e | 6712 | leaf = path->nodes[0]; |
0b86a832 CM |
6713 | if (slot >= btrfs_header_nritems(leaf)) { |
6714 | ret = btrfs_next_leaf(root, path); | |
6715 | if (ret == 0) | |
6716 | continue; | |
6717 | if (ret < 0) | |
925baedd | 6718 | goto out; |
0b86a832 | 6719 | break; |
edbd8d4e | 6720 | } |
0b86a832 | 6721 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 6722 | |
0b86a832 | 6723 | if (found_key.objectid >= key->objectid && |
925baedd CM |
6724 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6725 | ret = 0; | |
6726 | goto out; | |
6727 | } | |
0b86a832 | 6728 | path->slots[0]++; |
edbd8d4e | 6729 | } |
0b86a832 | 6730 | ret = -ENOENT; |
925baedd | 6731 | out: |
0b86a832 | 6732 | return ret; |
edbd8d4e CM |
6733 | } |
6734 | ||
1a40e23b ZY |
6735 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
6736 | { | |
6737 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 6738 | struct btrfs_space_info *space_info; |
1a40e23b ZY |
6739 | struct rb_node *n; |
6740 | ||
1a40e23b ZY |
6741 | spin_lock(&info->block_group_cache_lock); |
6742 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
6743 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
6744 | cache_node); | |
1a40e23b ZY |
6745 | rb_erase(&block_group->cache_node, |
6746 | &info->block_group_cache_tree); | |
d899e052 YZ |
6747 | spin_unlock(&info->block_group_cache_lock); |
6748 | ||
6749 | btrfs_remove_free_space_cache(block_group); | |
80eb234a | 6750 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 6751 | list_del(&block_group->list); |
80eb234a | 6752 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 YZ |
6753 | |
6754 | WARN_ON(atomic_read(&block_group->count) != 1); | |
1a40e23b | 6755 | kfree(block_group); |
d899e052 YZ |
6756 | |
6757 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
6758 | } |
6759 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
6760 | |
6761 | /* now that all the block groups are freed, go through and | |
6762 | * free all the space_info structs. This is only called during | |
6763 | * the final stages of unmount, and so we know nobody is | |
6764 | * using them. We call synchronize_rcu() once before we start, | |
6765 | * just to be on the safe side. | |
6766 | */ | |
6767 | synchronize_rcu(); | |
6768 | ||
6769 | while(!list_empty(&info->space_info)) { | |
6770 | space_info = list_entry(info->space_info.next, | |
6771 | struct btrfs_space_info, | |
6772 | list); | |
6773 | ||
6774 | list_del(&space_info->list); | |
6775 | kfree(space_info); | |
6776 | } | |
1a40e23b ZY |
6777 | return 0; |
6778 | } | |
6779 | ||
9078a3e1 CM |
6780 | int btrfs_read_block_groups(struct btrfs_root *root) |
6781 | { | |
6782 | struct btrfs_path *path; | |
6783 | int ret; | |
9078a3e1 | 6784 | struct btrfs_block_group_cache *cache; |
be744175 | 6785 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 6786 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
6787 | struct btrfs_key key; |
6788 | struct btrfs_key found_key; | |
5f39d397 | 6789 | struct extent_buffer *leaf; |
96b5179d | 6790 | |
be744175 | 6791 | root = info->extent_root; |
9078a3e1 | 6792 | key.objectid = 0; |
0b86a832 | 6793 | key.offset = 0; |
9078a3e1 | 6794 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
6795 | path = btrfs_alloc_path(); |
6796 | if (!path) | |
6797 | return -ENOMEM; | |
6798 | ||
d397712b | 6799 | while (1) { |
0b86a832 CM |
6800 | ret = find_first_block_group(root, path, &key); |
6801 | if (ret > 0) { | |
6802 | ret = 0; | |
6803 | goto error; | |
9078a3e1 | 6804 | } |
0b86a832 CM |
6805 | if (ret != 0) |
6806 | goto error; | |
6807 | ||
5f39d397 CM |
6808 | leaf = path->nodes[0]; |
6809 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 6810 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 6811 | if (!cache) { |
0b86a832 | 6812 | ret = -ENOMEM; |
9078a3e1 CM |
6813 | break; |
6814 | } | |
3e1ad54f | 6815 | |
d2fb3437 | 6816 | atomic_set(&cache->count, 1); |
c286ac48 | 6817 | spin_lock_init(&cache->lock); |
6226cb0a | 6818 | spin_lock_init(&cache->tree_lock); |
ea6a478e | 6819 | mutex_init(&cache->cache_mutex); |
0f9dd46c | 6820 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 6821 | INIT_LIST_HEAD(&cache->cluster_list); |
5f39d397 CM |
6822 | read_extent_buffer(leaf, &cache->item, |
6823 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
6824 | sizeof(cache->item)); | |
9078a3e1 | 6825 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 6826 | |
9078a3e1 CM |
6827 | key.objectid = found_key.objectid + found_key.offset; |
6828 | btrfs_release_path(root, path); | |
0b86a832 | 6829 | cache->flags = btrfs_block_group_flags(&cache->item); |
96b5179d | 6830 | |
6324fbf3 CM |
6831 | ret = update_space_info(info, cache->flags, found_key.offset, |
6832 | btrfs_block_group_used(&cache->item), | |
6833 | &space_info); | |
6834 | BUG_ON(ret); | |
6835 | cache->space_info = space_info; | |
80eb234a JB |
6836 | down_write(&space_info->groups_sem); |
6837 | list_add_tail(&cache->list, &space_info->block_groups); | |
6838 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
6839 | |
6840 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
6841 | BUG_ON(ret); | |
75ccf47d CM |
6842 | |
6843 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c YZ |
6844 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
6845 | set_block_group_readonly(cache); | |
9078a3e1 | 6846 | } |
0b86a832 CM |
6847 | ret = 0; |
6848 | error: | |
9078a3e1 | 6849 | btrfs_free_path(path); |
0b86a832 | 6850 | return ret; |
9078a3e1 | 6851 | } |
6324fbf3 CM |
6852 | |
6853 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
6854 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 6855 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
6856 | u64 size) |
6857 | { | |
6858 | int ret; | |
6324fbf3 CM |
6859 | struct btrfs_root *extent_root; |
6860 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
6861 | |
6862 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 6863 | |
12fcfd22 | 6864 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 6865 | |
8f18cf13 | 6866 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
6867 | if (!cache) |
6868 | return -ENOMEM; | |
6869 | ||
e17cade2 | 6870 | cache->key.objectid = chunk_offset; |
6324fbf3 | 6871 | cache->key.offset = size; |
d2fb3437 YZ |
6872 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
6873 | atomic_set(&cache->count, 1); | |
c286ac48 | 6874 | spin_lock_init(&cache->lock); |
6226cb0a | 6875 | spin_lock_init(&cache->tree_lock); |
ea6a478e | 6876 | mutex_init(&cache->cache_mutex); |
0f9dd46c | 6877 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 6878 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 6879 | |
6324fbf3 | 6880 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
6881 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6882 | cache->flags = type; | |
6883 | btrfs_set_block_group_flags(&cache->item, type); | |
6884 | ||
6885 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
6886 | &cache->space_info); | |
6887 | BUG_ON(ret); | |
80eb234a JB |
6888 | down_write(&cache->space_info->groups_sem); |
6889 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
6890 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 6891 | |
0f9dd46c JB |
6892 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
6893 | BUG_ON(ret); | |
c286ac48 | 6894 | |
6324fbf3 CM |
6895 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
6896 | sizeof(cache->item)); | |
6897 | BUG_ON(ret); | |
6898 | ||
d18a2c44 | 6899 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 6900 | |
6324fbf3 CM |
6901 | return 0; |
6902 | } | |
1a40e23b ZY |
6903 | |
6904 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
6905 | struct btrfs_root *root, u64 group_start) | |
6906 | { | |
6907 | struct btrfs_path *path; | |
6908 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 6909 | struct btrfs_free_cluster *cluster; |
1a40e23b ZY |
6910 | struct btrfs_key key; |
6911 | int ret; | |
6912 | ||
1a40e23b ZY |
6913 | root = root->fs_info->extent_root; |
6914 | ||
6915 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
6916 | BUG_ON(!block_group); | |
c146afad | 6917 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
6918 | |
6919 | memcpy(&key, &block_group->key, sizeof(key)); | |
6920 | ||
44fb5511 CM |
6921 | /* make sure this block group isn't part of an allocation cluster */ |
6922 | cluster = &root->fs_info->data_alloc_cluster; | |
6923 | spin_lock(&cluster->refill_lock); | |
6924 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
6925 | spin_unlock(&cluster->refill_lock); | |
6926 | ||
6927 | /* | |
6928 | * make sure this block group isn't part of a metadata | |
6929 | * allocation cluster | |
6930 | */ | |
6931 | cluster = &root->fs_info->meta_alloc_cluster; | |
6932 | spin_lock(&cluster->refill_lock); | |
6933 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
6934 | spin_unlock(&cluster->refill_lock); | |
6935 | ||
1a40e23b ZY |
6936 | path = btrfs_alloc_path(); |
6937 | BUG_ON(!path); | |
6938 | ||
3dfdb934 | 6939 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
6940 | rb_erase(&block_group->cache_node, |
6941 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 YZ |
6942 | spin_unlock(&root->fs_info->block_group_cache_lock); |
6943 | btrfs_remove_free_space_cache(block_group); | |
80eb234a | 6944 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
6945 | /* |
6946 | * we must use list_del_init so people can check to see if they | |
6947 | * are still on the list after taking the semaphore | |
6948 | */ | |
6949 | list_del_init(&block_group->list); | |
80eb234a | 6950 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 6951 | |
c146afad YZ |
6952 | spin_lock(&block_group->space_info->lock); |
6953 | block_group->space_info->total_bytes -= block_group->key.offset; | |
6954 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
6955 | spin_unlock(&block_group->space_info->lock); | |
2b82032c | 6956 | block_group->space_info->full = 0; |
c146afad | 6957 | |
fa9c0d79 CM |
6958 | btrfs_put_block_group(block_group); |
6959 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
6960 | |
6961 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
6962 | if (ret > 0) | |
6963 | ret = -EIO; | |
6964 | if (ret < 0) | |
6965 | goto out; | |
6966 | ||
6967 | ret = btrfs_del_item(trans, root, path); | |
6968 | out: | |
6969 | btrfs_free_path(path); | |
6970 | return ret; | |
6971 | } |