]>
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> |
817d52f8 | 24 | #include <linux/kthread.h> |
4b4e25f2 | 25 | #include "compat.h" |
74493f7a | 26 | #include "hash.h" |
fec577fb CM |
27 | #include "ctree.h" |
28 | #include "disk-io.h" | |
29 | #include "print-tree.h" | |
e089f05c | 30 | #include "transaction.h" |
0b86a832 | 31 | #include "volumes.h" |
925baedd | 32 | #include "locking.h" |
fa9c0d79 | 33 | #include "free-space-cache.h" |
fec577fb | 34 | |
f3465ca4 JB |
35 | static int update_block_group(struct btrfs_trans_handle *trans, |
36 | struct btrfs_root *root, | |
37 | u64 bytenr, u64 num_bytes, int alloc, | |
38 | int mark_free); | |
11833d66 YZ |
39 | static int update_reserved_extents(struct btrfs_block_group_cache *cache, |
40 | u64 num_bytes, int reserve); | |
5d4f98a2 YZ |
41 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
42 | struct btrfs_root *root, | |
43 | u64 bytenr, u64 num_bytes, u64 parent, | |
44 | u64 root_objectid, u64 owner_objectid, | |
45 | u64 owner_offset, int refs_to_drop, | |
46 | struct btrfs_delayed_extent_op *extra_op); | |
47 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
48 | struct extent_buffer *leaf, | |
49 | struct btrfs_extent_item *ei); | |
50 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
51 | struct btrfs_root *root, | |
52 | u64 parent, u64 root_objectid, | |
53 | u64 flags, u64 owner, u64 offset, | |
54 | struct btrfs_key *ins, int ref_mod); | |
55 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
56 | struct btrfs_root *root, | |
57 | u64 parent, u64 root_objectid, | |
58 | u64 flags, struct btrfs_disk_key *key, | |
59 | int level, struct btrfs_key *ins); | |
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); | |
11833d66 YZ |
63 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
64 | struct btrfs_root *root, | |
65 | struct btrfs_path *path, | |
66 | u64 bytenr, u64 num_bytes, | |
67 | int is_data, int reserved, | |
68 | struct extent_buffer **must_clean); | |
69 | static int find_next_key(struct btrfs_path *path, int level, | |
70 | struct btrfs_key *key); | |
9ed74f2d JB |
71 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
72 | int dump_block_groups); | |
6a63209f | 73 | |
817d52f8 JB |
74 | static noinline int |
75 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
76 | { | |
77 | smp_mb(); | |
78 | return cache->cached == BTRFS_CACHE_FINISHED; | |
79 | } | |
80 | ||
0f9dd46c JB |
81 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
82 | { | |
83 | return (cache->flags & bits) == bits; | |
84 | } | |
85 | ||
86 | /* | |
87 | * this adds the block group to the fs_info rb tree for the block group | |
88 | * cache | |
89 | */ | |
b2950863 | 90 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
91 | struct btrfs_block_group_cache *block_group) |
92 | { | |
93 | struct rb_node **p; | |
94 | struct rb_node *parent = NULL; | |
95 | struct btrfs_block_group_cache *cache; | |
96 | ||
97 | spin_lock(&info->block_group_cache_lock); | |
98 | p = &info->block_group_cache_tree.rb_node; | |
99 | ||
100 | while (*p) { | |
101 | parent = *p; | |
102 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
103 | cache_node); | |
104 | if (block_group->key.objectid < cache->key.objectid) { | |
105 | p = &(*p)->rb_left; | |
106 | } else if (block_group->key.objectid > cache->key.objectid) { | |
107 | p = &(*p)->rb_right; | |
108 | } else { | |
109 | spin_unlock(&info->block_group_cache_lock); | |
110 | return -EEXIST; | |
111 | } | |
112 | } | |
113 | ||
114 | rb_link_node(&block_group->cache_node, parent, p); | |
115 | rb_insert_color(&block_group->cache_node, | |
116 | &info->block_group_cache_tree); | |
117 | spin_unlock(&info->block_group_cache_lock); | |
118 | ||
119 | return 0; | |
120 | } | |
121 | ||
122 | /* | |
123 | * This will return the block group at or after bytenr if contains is 0, else | |
124 | * it will return the block group that contains the bytenr | |
125 | */ | |
126 | static struct btrfs_block_group_cache * | |
127 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
128 | int contains) | |
129 | { | |
130 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
131 | struct rb_node *n; | |
132 | u64 end, start; | |
133 | ||
134 | spin_lock(&info->block_group_cache_lock); | |
135 | n = info->block_group_cache_tree.rb_node; | |
136 | ||
137 | while (n) { | |
138 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
139 | cache_node); | |
140 | end = cache->key.objectid + cache->key.offset - 1; | |
141 | start = cache->key.objectid; | |
142 | ||
143 | if (bytenr < start) { | |
144 | if (!contains && (!ret || start < ret->key.objectid)) | |
145 | ret = cache; | |
146 | n = n->rb_left; | |
147 | } else if (bytenr > start) { | |
148 | if (contains && bytenr <= end) { | |
149 | ret = cache; | |
150 | break; | |
151 | } | |
152 | n = n->rb_right; | |
153 | } else { | |
154 | ret = cache; | |
155 | break; | |
156 | } | |
157 | } | |
d2fb3437 YZ |
158 | if (ret) |
159 | atomic_inc(&ret->count); | |
0f9dd46c JB |
160 | spin_unlock(&info->block_group_cache_lock); |
161 | ||
162 | return ret; | |
163 | } | |
164 | ||
11833d66 YZ |
165 | static int add_excluded_extent(struct btrfs_root *root, |
166 | u64 start, u64 num_bytes) | |
817d52f8 | 167 | { |
11833d66 YZ |
168 | u64 end = start + num_bytes - 1; |
169 | set_extent_bits(&root->fs_info->freed_extents[0], | |
170 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
171 | set_extent_bits(&root->fs_info->freed_extents[1], | |
172 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
173 | return 0; | |
174 | } | |
817d52f8 | 175 | |
11833d66 YZ |
176 | static void free_excluded_extents(struct btrfs_root *root, |
177 | struct btrfs_block_group_cache *cache) | |
178 | { | |
179 | u64 start, end; | |
817d52f8 | 180 | |
11833d66 YZ |
181 | start = cache->key.objectid; |
182 | end = start + cache->key.offset - 1; | |
183 | ||
184 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
185 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
186 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
187 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
188 | } |
189 | ||
11833d66 YZ |
190 | static int exclude_super_stripes(struct btrfs_root *root, |
191 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 192 | { |
817d52f8 JB |
193 | u64 bytenr; |
194 | u64 *logical; | |
195 | int stripe_len; | |
196 | int i, nr, ret; | |
197 | ||
198 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
199 | bytenr = btrfs_sb_offset(i); | |
200 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
201 | cache->key.objectid, bytenr, | |
202 | 0, &logical, &nr, &stripe_len); | |
203 | BUG_ON(ret); | |
11833d66 | 204 | |
817d52f8 | 205 | while (nr--) { |
1b2da372 | 206 | cache->bytes_super += stripe_len; |
11833d66 YZ |
207 | ret = add_excluded_extent(root, logical[nr], |
208 | stripe_len); | |
209 | BUG_ON(ret); | |
817d52f8 | 210 | } |
11833d66 | 211 | |
817d52f8 JB |
212 | kfree(logical); |
213 | } | |
817d52f8 JB |
214 | return 0; |
215 | } | |
216 | ||
11833d66 YZ |
217 | static struct btrfs_caching_control * |
218 | get_caching_control(struct btrfs_block_group_cache *cache) | |
219 | { | |
220 | struct btrfs_caching_control *ctl; | |
221 | ||
222 | spin_lock(&cache->lock); | |
223 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
224 | spin_unlock(&cache->lock); | |
225 | return NULL; | |
226 | } | |
227 | ||
228 | ctl = cache->caching_ctl; | |
229 | atomic_inc(&ctl->count); | |
230 | spin_unlock(&cache->lock); | |
231 | return ctl; | |
232 | } | |
233 | ||
234 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
235 | { | |
236 | if (atomic_dec_and_test(&ctl->count)) | |
237 | kfree(ctl); | |
238 | } | |
239 | ||
0f9dd46c JB |
240 | /* |
241 | * this is only called by cache_block_group, since we could have freed extents | |
242 | * we need to check the pinned_extents for any extents that can't be used yet | |
243 | * since their free space will be released as soon as the transaction commits. | |
244 | */ | |
817d52f8 | 245 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
246 | struct btrfs_fs_info *info, u64 start, u64 end) |
247 | { | |
817d52f8 | 248 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
249 | int ret; |
250 | ||
251 | while (start < end) { | |
11833d66 | 252 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 253 | &extent_start, &extent_end, |
11833d66 | 254 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
255 | if (ret) |
256 | break; | |
257 | ||
258 | if (extent_start == start) { | |
259 | start = extent_end + 1; | |
260 | } else if (extent_start > start && extent_start < end) { | |
261 | size = extent_start - start; | |
817d52f8 | 262 | total_added += size; |
ea6a478e JB |
263 | ret = btrfs_add_free_space(block_group, start, |
264 | size); | |
0f9dd46c JB |
265 | BUG_ON(ret); |
266 | start = extent_end + 1; | |
267 | } else { | |
268 | break; | |
269 | } | |
270 | } | |
271 | ||
272 | if (start < end) { | |
273 | size = end - start; | |
817d52f8 | 274 | total_added += size; |
ea6a478e | 275 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
276 | BUG_ON(ret); |
277 | } | |
278 | ||
817d52f8 | 279 | return total_added; |
0f9dd46c JB |
280 | } |
281 | ||
817d52f8 | 282 | static int caching_kthread(void *data) |
e37c9e69 | 283 | { |
817d52f8 JB |
284 | struct btrfs_block_group_cache *block_group = data; |
285 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
11833d66 YZ |
286 | struct btrfs_caching_control *caching_ctl = block_group->caching_ctl; |
287 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 288 | struct btrfs_path *path; |
5f39d397 | 289 | struct extent_buffer *leaf; |
11833d66 | 290 | struct btrfs_key key; |
817d52f8 | 291 | u64 total_found = 0; |
11833d66 YZ |
292 | u64 last = 0; |
293 | u32 nritems; | |
294 | int ret = 0; | |
f510cfec | 295 | |
e37c9e69 CM |
296 | path = btrfs_alloc_path(); |
297 | if (!path) | |
298 | return -ENOMEM; | |
7d7d6068 | 299 | |
11833d66 | 300 | exclude_super_stripes(extent_root, block_group); |
1b2da372 JB |
301 | spin_lock(&block_group->space_info->lock); |
302 | block_group->space_info->bytes_super += block_group->bytes_super; | |
303 | spin_unlock(&block_group->space_info->lock); | |
11833d66 | 304 | |
817d52f8 | 305 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 306 | |
5cd57b2c | 307 | /* |
817d52f8 JB |
308 | * We don't want to deadlock with somebody trying to allocate a new |
309 | * extent for the extent root while also trying to search the extent | |
310 | * root to add free space. So we skip locking and search the commit | |
311 | * root, since its read-only | |
5cd57b2c CM |
312 | */ |
313 | path->skip_locking = 1; | |
817d52f8 JB |
314 | path->search_commit_root = 1; |
315 | path->reada = 2; | |
316 | ||
e4404d6e | 317 | key.objectid = last; |
e37c9e69 | 318 | key.offset = 0; |
11833d66 | 319 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 320 | again: |
11833d66 | 321 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
322 | /* need to make sure the commit_root doesn't disappear */ |
323 | down_read(&fs_info->extent_commit_sem); | |
324 | ||
11833d66 | 325 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 326 | if (ret < 0) |
ef8bbdfe | 327 | goto err; |
a512bbf8 | 328 | |
11833d66 YZ |
329 | leaf = path->nodes[0]; |
330 | nritems = btrfs_header_nritems(leaf); | |
331 | ||
d397712b | 332 | while (1) { |
817d52f8 | 333 | smp_mb(); |
11833d66 | 334 | if (fs_info->closing > 1) { |
f25784b3 | 335 | last = (u64)-1; |
817d52f8 | 336 | break; |
f25784b3 | 337 | } |
817d52f8 | 338 | |
11833d66 YZ |
339 | if (path->slots[0] < nritems) { |
340 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
341 | } else { | |
342 | ret = find_next_key(path, 0, &key); | |
343 | if (ret) | |
e37c9e69 | 344 | break; |
817d52f8 | 345 | |
11833d66 YZ |
346 | caching_ctl->progress = last; |
347 | btrfs_release_path(extent_root, path); | |
348 | up_read(&fs_info->extent_commit_sem); | |
349 | mutex_unlock(&caching_ctl->mutex); | |
350 | if (btrfs_transaction_in_commit(fs_info)) | |
f36f3042 | 351 | schedule_timeout(1); |
11833d66 YZ |
352 | else |
353 | cond_resched(); | |
354 | goto again; | |
355 | } | |
817d52f8 | 356 | |
11833d66 YZ |
357 | if (key.objectid < block_group->key.objectid) { |
358 | path->slots[0]++; | |
817d52f8 | 359 | continue; |
e37c9e69 | 360 | } |
0f9dd46c | 361 | |
e37c9e69 | 362 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 363 | block_group->key.offset) |
e37c9e69 | 364 | break; |
7d7d6068 | 365 | |
11833d66 | 366 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
367 | total_found += add_new_free_space(block_group, |
368 | fs_info, last, | |
369 | key.objectid); | |
7d7d6068 | 370 | last = key.objectid + key.offset; |
817d52f8 | 371 | |
11833d66 YZ |
372 | if (total_found > (1024 * 1024 * 2)) { |
373 | total_found = 0; | |
374 | wake_up(&caching_ctl->wait); | |
375 | } | |
817d52f8 | 376 | } |
e37c9e69 CM |
377 | path->slots[0]++; |
378 | } | |
817d52f8 | 379 | ret = 0; |
e37c9e69 | 380 | |
817d52f8 JB |
381 | total_found += add_new_free_space(block_group, fs_info, last, |
382 | block_group->key.objectid + | |
383 | block_group->key.offset); | |
11833d66 | 384 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
385 | |
386 | spin_lock(&block_group->lock); | |
11833d66 | 387 | block_group->caching_ctl = NULL; |
817d52f8 JB |
388 | block_group->cached = BTRFS_CACHE_FINISHED; |
389 | spin_unlock(&block_group->lock); | |
0f9dd46c | 390 | |
54aa1f4d | 391 | err: |
e37c9e69 | 392 | btrfs_free_path(path); |
276e680d | 393 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 394 | |
11833d66 YZ |
395 | free_excluded_extents(extent_root, block_group); |
396 | ||
397 | mutex_unlock(&caching_ctl->mutex); | |
398 | wake_up(&caching_ctl->wait); | |
399 | ||
400 | put_caching_control(caching_ctl); | |
401 | atomic_dec(&block_group->space_info->caching_threads); | |
817d52f8 JB |
402 | return 0; |
403 | } | |
404 | ||
405 | static int cache_block_group(struct btrfs_block_group_cache *cache) | |
406 | { | |
11833d66 YZ |
407 | struct btrfs_fs_info *fs_info = cache->fs_info; |
408 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
409 | struct task_struct *tsk; |
410 | int ret = 0; | |
411 | ||
11833d66 YZ |
412 | smp_mb(); |
413 | if (cache->cached != BTRFS_CACHE_NO) | |
414 | return 0; | |
415 | ||
416 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL); | |
417 | BUG_ON(!caching_ctl); | |
418 | ||
419 | INIT_LIST_HEAD(&caching_ctl->list); | |
420 | mutex_init(&caching_ctl->mutex); | |
421 | init_waitqueue_head(&caching_ctl->wait); | |
422 | caching_ctl->block_group = cache; | |
423 | caching_ctl->progress = cache->key.objectid; | |
424 | /* one for caching kthread, one for caching block group list */ | |
425 | atomic_set(&caching_ctl->count, 2); | |
426 | ||
817d52f8 JB |
427 | spin_lock(&cache->lock); |
428 | if (cache->cached != BTRFS_CACHE_NO) { | |
429 | spin_unlock(&cache->lock); | |
11833d66 YZ |
430 | kfree(caching_ctl); |
431 | return 0; | |
817d52f8 | 432 | } |
11833d66 | 433 | cache->caching_ctl = caching_ctl; |
817d52f8 JB |
434 | cache->cached = BTRFS_CACHE_STARTED; |
435 | spin_unlock(&cache->lock); | |
436 | ||
11833d66 YZ |
437 | down_write(&fs_info->extent_commit_sem); |
438 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
439 | up_write(&fs_info->extent_commit_sem); | |
440 | ||
441 | atomic_inc(&cache->space_info->caching_threads); | |
442 | ||
817d52f8 JB |
443 | tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", |
444 | cache->key.objectid); | |
445 | if (IS_ERR(tsk)) { | |
446 | ret = PTR_ERR(tsk); | |
447 | printk(KERN_ERR "error running thread %d\n", ret); | |
448 | BUG(); | |
449 | } | |
450 | ||
ef8bbdfe | 451 | return ret; |
e37c9e69 CM |
452 | } |
453 | ||
0f9dd46c JB |
454 | /* |
455 | * return the block group that starts at or after bytenr | |
456 | */ | |
d397712b CM |
457 | static struct btrfs_block_group_cache * |
458 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 459 | { |
0f9dd46c | 460 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 461 | |
0f9dd46c | 462 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 463 | |
0f9dd46c | 464 | return cache; |
0ef3e66b CM |
465 | } |
466 | ||
0f9dd46c | 467 | /* |
9f55684c | 468 | * return the block group that contains the given bytenr |
0f9dd46c | 469 | */ |
d397712b CM |
470 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
471 | struct btrfs_fs_info *info, | |
472 | u64 bytenr) | |
be744175 | 473 | { |
0f9dd46c | 474 | struct btrfs_block_group_cache *cache; |
be744175 | 475 | |
0f9dd46c | 476 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 477 | |
0f9dd46c | 478 | return cache; |
be744175 | 479 | } |
0b86a832 | 480 | |
fa9c0d79 | 481 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) |
d2fb3437 YZ |
482 | { |
483 | if (atomic_dec_and_test(&cache->count)) | |
484 | kfree(cache); | |
485 | } | |
486 | ||
0f9dd46c JB |
487 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
488 | u64 flags) | |
6324fbf3 | 489 | { |
0f9dd46c | 490 | struct list_head *head = &info->space_info; |
0f9dd46c | 491 | struct btrfs_space_info *found; |
4184ea7f CM |
492 | |
493 | rcu_read_lock(); | |
494 | list_for_each_entry_rcu(found, head, list) { | |
495 | if (found->flags == flags) { | |
496 | rcu_read_unlock(); | |
0f9dd46c | 497 | return found; |
4184ea7f | 498 | } |
0f9dd46c | 499 | } |
4184ea7f | 500 | rcu_read_unlock(); |
0f9dd46c | 501 | return NULL; |
6324fbf3 CM |
502 | } |
503 | ||
4184ea7f CM |
504 | /* |
505 | * after adding space to the filesystem, we need to clear the full flags | |
506 | * on all the space infos. | |
507 | */ | |
508 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
509 | { | |
510 | struct list_head *head = &info->space_info; | |
511 | struct btrfs_space_info *found; | |
512 | ||
513 | rcu_read_lock(); | |
514 | list_for_each_entry_rcu(found, head, list) | |
515 | found->full = 0; | |
516 | rcu_read_unlock(); | |
517 | } | |
518 | ||
80eb234a JB |
519 | static u64 div_factor(u64 num, int factor) |
520 | { | |
521 | if (factor == 10) | |
522 | return num; | |
523 | num *= factor; | |
524 | do_div(num, 10); | |
525 | return num; | |
526 | } | |
527 | ||
d2fb3437 YZ |
528 | u64 btrfs_find_block_group(struct btrfs_root *root, |
529 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 530 | { |
96b5179d | 531 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 532 | u64 used; |
d2fb3437 YZ |
533 | u64 last = max(search_hint, search_start); |
534 | u64 group_start = 0; | |
31f3c99b | 535 | int full_search = 0; |
d2fb3437 | 536 | int factor = 9; |
0ef3e66b | 537 | int wrapped = 0; |
31f3c99b | 538 | again: |
e8569813 ZY |
539 | while (1) { |
540 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
541 | if (!cache) |
542 | break; | |
96b5179d | 543 | |
c286ac48 | 544 | spin_lock(&cache->lock); |
96b5179d CM |
545 | last = cache->key.objectid + cache->key.offset; |
546 | used = btrfs_block_group_used(&cache->item); | |
547 | ||
d2fb3437 YZ |
548 | if ((full_search || !cache->ro) && |
549 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 550 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
551 | div_factor(cache->key.offset, factor)) { |
552 | group_start = cache->key.objectid; | |
c286ac48 | 553 | spin_unlock(&cache->lock); |
fa9c0d79 | 554 | btrfs_put_block_group(cache); |
8790d502 CM |
555 | goto found; |
556 | } | |
6324fbf3 | 557 | } |
c286ac48 | 558 | spin_unlock(&cache->lock); |
fa9c0d79 | 559 | btrfs_put_block_group(cache); |
de428b63 | 560 | cond_resched(); |
cd1bc465 | 561 | } |
0ef3e66b CM |
562 | if (!wrapped) { |
563 | last = search_start; | |
564 | wrapped = 1; | |
565 | goto again; | |
566 | } | |
567 | if (!full_search && factor < 10) { | |
be744175 | 568 | last = search_start; |
31f3c99b | 569 | full_search = 1; |
0ef3e66b | 570 | factor = 10; |
31f3c99b CM |
571 | goto again; |
572 | } | |
be744175 | 573 | found: |
d2fb3437 | 574 | return group_start; |
925baedd | 575 | } |
0f9dd46c | 576 | |
e02119d5 | 577 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 578 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
579 | { |
580 | int ret; | |
581 | struct btrfs_key key; | |
31840ae1 | 582 | struct btrfs_path *path; |
e02119d5 | 583 | |
31840ae1 ZY |
584 | path = btrfs_alloc_path(); |
585 | BUG_ON(!path); | |
e02119d5 CM |
586 | key.objectid = start; |
587 | key.offset = len; | |
588 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
589 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
590 | 0, 0); | |
31840ae1 | 591 | btrfs_free_path(path); |
7bb86316 CM |
592 | return ret; |
593 | } | |
594 | ||
d8d5f3e1 CM |
595 | /* |
596 | * Back reference rules. Back refs have three main goals: | |
597 | * | |
598 | * 1) differentiate between all holders of references to an extent so that | |
599 | * when a reference is dropped we can make sure it was a valid reference | |
600 | * before freeing the extent. | |
601 | * | |
602 | * 2) Provide enough information to quickly find the holders of an extent | |
603 | * if we notice a given block is corrupted or bad. | |
604 | * | |
605 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
606 | * maintenance. This is actually the same as #2, but with a slightly | |
607 | * different use case. | |
608 | * | |
5d4f98a2 YZ |
609 | * There are two kinds of back refs. The implicit back refs is optimized |
610 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
611 | * back refs of this kind provide information about the block's owner tree | |
612 | * and the pointer's key. These information allow us to find the block by | |
613 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
614 | * referenced by their owner trees. The location of tree block is recorded | |
615 | * in the back refs. Actually the full back refs is generic, and can be | |
616 | * used in all cases the implicit back refs is used. The major shortcoming | |
617 | * of the full back refs is its overhead. Every time a tree block gets | |
618 | * COWed, we have to update back refs entry for all pointers in it. | |
619 | * | |
620 | * For a newly allocated tree block, we use implicit back refs for | |
621 | * pointers in it. This means most tree related operations only involve | |
622 | * implicit back refs. For a tree block created in old transaction, the | |
623 | * only way to drop a reference to it is COW it. So we can detect the | |
624 | * event that tree block loses its owner tree's reference and do the | |
625 | * back refs conversion. | |
626 | * | |
627 | * When a tree block is COW'd through a tree, there are four cases: | |
628 | * | |
629 | * The reference count of the block is one and the tree is the block's | |
630 | * owner tree. Nothing to do in this case. | |
631 | * | |
632 | * The reference count of the block is one and the tree is not the | |
633 | * block's owner tree. In this case, full back refs is used for pointers | |
634 | * in the block. Remove these full back refs, add implicit back refs for | |
635 | * every pointers in the new block. | |
636 | * | |
637 | * The reference count of the block is greater than one and the tree is | |
638 | * the block's owner tree. In this case, implicit back refs is used for | |
639 | * pointers in the block. Add full back refs for every pointers in the | |
640 | * block, increase lower level extents' reference counts. The original | |
641 | * implicit back refs are entailed to the new block. | |
642 | * | |
643 | * The reference count of the block is greater than one and the tree is | |
644 | * not the block's owner tree. Add implicit back refs for every pointer in | |
645 | * the new block, increase lower level extents' reference count. | |
646 | * | |
647 | * Back Reference Key composing: | |
648 | * | |
649 | * The key objectid corresponds to the first byte in the extent, | |
650 | * The key type is used to differentiate between types of back refs. | |
651 | * There are different meanings of the key offset for different types | |
652 | * of back refs. | |
653 | * | |
d8d5f3e1 CM |
654 | * File extents can be referenced by: |
655 | * | |
656 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 657 | * - different files inside a single subvolume |
d8d5f3e1 CM |
658 | * - different offsets inside a file (bookend extents in file.c) |
659 | * | |
5d4f98a2 | 660 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
661 | * |
662 | * - Objectid of the subvolume root | |
d8d5f3e1 | 663 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
664 | * - original offset in the file |
665 | * - how many bookend extents | |
d8d5f3e1 | 666 | * |
5d4f98a2 YZ |
667 | * The key offset for the implicit back refs is hash of the first |
668 | * three fields. | |
d8d5f3e1 | 669 | * |
5d4f98a2 | 670 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 671 | * |
5d4f98a2 | 672 | * - number of pointers in the tree leaf |
d8d5f3e1 | 673 | * |
5d4f98a2 YZ |
674 | * The key offset for the implicit back refs is the first byte of |
675 | * the tree leaf | |
d8d5f3e1 | 676 | * |
5d4f98a2 YZ |
677 | * When a file extent is allocated, The implicit back refs is used. |
678 | * the fields are filled in: | |
d8d5f3e1 | 679 | * |
5d4f98a2 | 680 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 681 | * |
5d4f98a2 YZ |
682 | * When a file extent is removed file truncation, we find the |
683 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 684 | * |
5d4f98a2 | 685 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 686 | * |
5d4f98a2 | 687 | * Btree extents can be referenced by: |
d8d5f3e1 | 688 | * |
5d4f98a2 | 689 | * - Different subvolumes |
d8d5f3e1 | 690 | * |
5d4f98a2 YZ |
691 | * Both the implicit back refs and the full back refs for tree blocks |
692 | * only consist of key. The key offset for the implicit back refs is | |
693 | * objectid of block's owner tree. The key offset for the full back refs | |
694 | * is the first byte of parent block. | |
d8d5f3e1 | 695 | * |
5d4f98a2 YZ |
696 | * When implicit back refs is used, information about the lowest key and |
697 | * level of the tree block are required. These information are stored in | |
698 | * tree block info structure. | |
d8d5f3e1 | 699 | */ |
31840ae1 | 700 | |
5d4f98a2 YZ |
701 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
702 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
703 | struct btrfs_root *root, | |
704 | struct btrfs_path *path, | |
705 | u64 owner, u32 extra_size) | |
7bb86316 | 706 | { |
5d4f98a2 YZ |
707 | struct btrfs_extent_item *item; |
708 | struct btrfs_extent_item_v0 *ei0; | |
709 | struct btrfs_extent_ref_v0 *ref0; | |
710 | struct btrfs_tree_block_info *bi; | |
711 | struct extent_buffer *leaf; | |
7bb86316 | 712 | struct btrfs_key key; |
5d4f98a2 YZ |
713 | struct btrfs_key found_key; |
714 | u32 new_size = sizeof(*item); | |
715 | u64 refs; | |
716 | int ret; | |
717 | ||
718 | leaf = path->nodes[0]; | |
719 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
720 | ||
721 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
722 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
723 | struct btrfs_extent_item_v0); | |
724 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
725 | ||
726 | if (owner == (u64)-1) { | |
727 | while (1) { | |
728 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
729 | ret = btrfs_next_leaf(root, path); | |
730 | if (ret < 0) | |
731 | return ret; | |
732 | BUG_ON(ret > 0); | |
733 | leaf = path->nodes[0]; | |
734 | } | |
735 | btrfs_item_key_to_cpu(leaf, &found_key, | |
736 | path->slots[0]); | |
737 | BUG_ON(key.objectid != found_key.objectid); | |
738 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
739 | path->slots[0]++; | |
740 | continue; | |
741 | } | |
742 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
743 | struct btrfs_extent_ref_v0); | |
744 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
745 | break; | |
746 | } | |
747 | } | |
748 | btrfs_release_path(root, path); | |
749 | ||
750 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
751 | new_size += sizeof(*bi); | |
752 | ||
753 | new_size -= sizeof(*ei0); | |
754 | ret = btrfs_search_slot(trans, root, &key, path, | |
755 | new_size + extra_size, 1); | |
756 | if (ret < 0) | |
757 | return ret; | |
758 | BUG_ON(ret); | |
759 | ||
760 | ret = btrfs_extend_item(trans, root, path, new_size); | |
761 | BUG_ON(ret); | |
762 | ||
763 | leaf = path->nodes[0]; | |
764 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
765 | btrfs_set_extent_refs(leaf, item, refs); | |
766 | /* FIXME: get real generation */ | |
767 | btrfs_set_extent_generation(leaf, item, 0); | |
768 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
769 | btrfs_set_extent_flags(leaf, item, | |
770 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
771 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
772 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
773 | /* FIXME: get first key of the block */ | |
774 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
775 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
776 | } else { | |
777 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
778 | } | |
779 | btrfs_mark_buffer_dirty(leaf); | |
780 | return 0; | |
781 | } | |
782 | #endif | |
783 | ||
784 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
785 | { | |
786 | u32 high_crc = ~(u32)0; | |
787 | u32 low_crc = ~(u32)0; | |
788 | __le64 lenum; | |
789 | ||
790 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 791 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 792 | lenum = cpu_to_le64(owner); |
163e783e | 793 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 794 | lenum = cpu_to_le64(offset); |
163e783e | 795 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
796 | |
797 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
798 | } | |
799 | ||
800 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
801 | struct btrfs_extent_data_ref *ref) | |
802 | { | |
803 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
804 | btrfs_extent_data_ref_objectid(leaf, ref), | |
805 | btrfs_extent_data_ref_offset(leaf, ref)); | |
806 | } | |
807 | ||
808 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
809 | struct btrfs_extent_data_ref *ref, | |
810 | u64 root_objectid, u64 owner, u64 offset) | |
811 | { | |
812 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
813 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
814 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
815 | return 0; | |
816 | return 1; | |
817 | } | |
818 | ||
819 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
820 | struct btrfs_root *root, | |
821 | struct btrfs_path *path, | |
822 | u64 bytenr, u64 parent, | |
823 | u64 root_objectid, | |
824 | u64 owner, u64 offset) | |
825 | { | |
826 | struct btrfs_key key; | |
827 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 828 | struct extent_buffer *leaf; |
5d4f98a2 | 829 | u32 nritems; |
74493f7a | 830 | int ret; |
5d4f98a2 YZ |
831 | int recow; |
832 | int err = -ENOENT; | |
74493f7a | 833 | |
31840ae1 | 834 | key.objectid = bytenr; |
5d4f98a2 YZ |
835 | if (parent) { |
836 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
837 | key.offset = parent; | |
838 | } else { | |
839 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
840 | key.offset = hash_extent_data_ref(root_objectid, | |
841 | owner, offset); | |
842 | } | |
843 | again: | |
844 | recow = 0; | |
845 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
846 | if (ret < 0) { | |
847 | err = ret; | |
848 | goto fail; | |
849 | } | |
31840ae1 | 850 | |
5d4f98a2 YZ |
851 | if (parent) { |
852 | if (!ret) | |
853 | return 0; | |
854 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
855 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
856 | btrfs_release_path(root, path); | |
857 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
858 | if (ret < 0) { | |
859 | err = ret; | |
860 | goto fail; | |
861 | } | |
862 | if (!ret) | |
863 | return 0; | |
864 | #endif | |
865 | goto fail; | |
31840ae1 ZY |
866 | } |
867 | ||
868 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
869 | nritems = btrfs_header_nritems(leaf); |
870 | while (1) { | |
871 | if (path->slots[0] >= nritems) { | |
872 | ret = btrfs_next_leaf(root, path); | |
873 | if (ret < 0) | |
874 | err = ret; | |
875 | if (ret) | |
876 | goto fail; | |
877 | ||
878 | leaf = path->nodes[0]; | |
879 | nritems = btrfs_header_nritems(leaf); | |
880 | recow = 1; | |
881 | } | |
882 | ||
883 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
884 | if (key.objectid != bytenr || | |
885 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
886 | goto fail; | |
887 | ||
888 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
889 | struct btrfs_extent_data_ref); | |
890 | ||
891 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
892 | owner, offset)) { | |
893 | if (recow) { | |
894 | btrfs_release_path(root, path); | |
895 | goto again; | |
896 | } | |
897 | err = 0; | |
898 | break; | |
899 | } | |
900 | path->slots[0]++; | |
31840ae1 | 901 | } |
5d4f98a2 YZ |
902 | fail: |
903 | return err; | |
31840ae1 ZY |
904 | } |
905 | ||
5d4f98a2 YZ |
906 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
907 | struct btrfs_root *root, | |
908 | struct btrfs_path *path, | |
909 | u64 bytenr, u64 parent, | |
910 | u64 root_objectid, u64 owner, | |
911 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
912 | { |
913 | struct btrfs_key key; | |
914 | struct extent_buffer *leaf; | |
5d4f98a2 | 915 | u32 size; |
31840ae1 ZY |
916 | u32 num_refs; |
917 | int ret; | |
74493f7a | 918 | |
74493f7a | 919 | key.objectid = bytenr; |
5d4f98a2 YZ |
920 | if (parent) { |
921 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
922 | key.offset = parent; | |
923 | size = sizeof(struct btrfs_shared_data_ref); | |
924 | } else { | |
925 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
926 | key.offset = hash_extent_data_ref(root_objectid, | |
927 | owner, offset); | |
928 | size = sizeof(struct btrfs_extent_data_ref); | |
929 | } | |
74493f7a | 930 | |
5d4f98a2 YZ |
931 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
932 | if (ret && ret != -EEXIST) | |
933 | goto fail; | |
934 | ||
935 | leaf = path->nodes[0]; | |
936 | if (parent) { | |
937 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 938 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
939 | struct btrfs_shared_data_ref); |
940 | if (ret == 0) { | |
941 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
942 | } else { | |
943 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
944 | num_refs += refs_to_add; | |
945 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 946 | } |
5d4f98a2 YZ |
947 | } else { |
948 | struct btrfs_extent_data_ref *ref; | |
949 | while (ret == -EEXIST) { | |
950 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
951 | struct btrfs_extent_data_ref); | |
952 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
953 | owner, offset)) | |
954 | break; | |
955 | btrfs_release_path(root, path); | |
956 | key.offset++; | |
957 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
958 | size); | |
959 | if (ret && ret != -EEXIST) | |
960 | goto fail; | |
31840ae1 | 961 | |
5d4f98a2 YZ |
962 | leaf = path->nodes[0]; |
963 | } | |
964 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
965 | struct btrfs_extent_data_ref); | |
966 | if (ret == 0) { | |
967 | btrfs_set_extent_data_ref_root(leaf, ref, | |
968 | root_objectid); | |
969 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
970 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
971 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
972 | } else { | |
973 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
974 | num_refs += refs_to_add; | |
975 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 976 | } |
31840ae1 | 977 | } |
5d4f98a2 YZ |
978 | btrfs_mark_buffer_dirty(leaf); |
979 | ret = 0; | |
980 | fail: | |
7bb86316 CM |
981 | btrfs_release_path(root, path); |
982 | return ret; | |
74493f7a CM |
983 | } |
984 | ||
5d4f98a2 YZ |
985 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
986 | struct btrfs_root *root, | |
987 | struct btrfs_path *path, | |
988 | int refs_to_drop) | |
31840ae1 | 989 | { |
5d4f98a2 YZ |
990 | struct btrfs_key key; |
991 | struct btrfs_extent_data_ref *ref1 = NULL; | |
992 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 993 | struct extent_buffer *leaf; |
5d4f98a2 | 994 | u32 num_refs = 0; |
31840ae1 ZY |
995 | int ret = 0; |
996 | ||
997 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
998 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
999 | ||
1000 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1001 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1002 | struct btrfs_extent_data_ref); | |
1003 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1004 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1005 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1006 | struct btrfs_shared_data_ref); | |
1007 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1008 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1009 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1010 | struct btrfs_extent_ref_v0 *ref0; | |
1011 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1012 | struct btrfs_extent_ref_v0); | |
1013 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1014 | #endif | |
1015 | } else { | |
1016 | BUG(); | |
1017 | } | |
1018 | ||
56bec294 CM |
1019 | BUG_ON(num_refs < refs_to_drop); |
1020 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1021 | |
31840ae1 ZY |
1022 | if (num_refs == 0) { |
1023 | ret = btrfs_del_item(trans, root, path); | |
1024 | } else { | |
5d4f98a2 YZ |
1025 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1026 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1027 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1028 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1029 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1030 | else { | |
1031 | struct btrfs_extent_ref_v0 *ref0; | |
1032 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1033 | struct btrfs_extent_ref_v0); | |
1034 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1035 | } | |
1036 | #endif | |
31840ae1 ZY |
1037 | btrfs_mark_buffer_dirty(leaf); |
1038 | } | |
31840ae1 ZY |
1039 | return ret; |
1040 | } | |
1041 | ||
5d4f98a2 YZ |
1042 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1043 | struct btrfs_path *path, | |
1044 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1045 | { |
5d4f98a2 YZ |
1046 | struct btrfs_key key; |
1047 | struct extent_buffer *leaf; | |
1048 | struct btrfs_extent_data_ref *ref1; | |
1049 | struct btrfs_shared_data_ref *ref2; | |
1050 | u32 num_refs = 0; | |
1051 | ||
1052 | leaf = path->nodes[0]; | |
1053 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1054 | if (iref) { | |
1055 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1056 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1057 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1058 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1059 | } else { | |
1060 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1061 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1062 | } | |
1063 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1064 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1065 | struct btrfs_extent_data_ref); | |
1066 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1067 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1068 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1069 | struct btrfs_shared_data_ref); | |
1070 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1071 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1072 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1073 | struct btrfs_extent_ref_v0 *ref0; | |
1074 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1075 | struct btrfs_extent_ref_v0); | |
1076 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1077 | #endif |
5d4f98a2 YZ |
1078 | } else { |
1079 | WARN_ON(1); | |
1080 | } | |
1081 | return num_refs; | |
1082 | } | |
15916de8 | 1083 | |
5d4f98a2 YZ |
1084 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1085 | struct btrfs_root *root, | |
1086 | struct btrfs_path *path, | |
1087 | u64 bytenr, u64 parent, | |
1088 | u64 root_objectid) | |
1f3c79a2 | 1089 | { |
5d4f98a2 | 1090 | struct btrfs_key key; |
1f3c79a2 | 1091 | int ret; |
1f3c79a2 | 1092 | |
5d4f98a2 YZ |
1093 | key.objectid = bytenr; |
1094 | if (parent) { | |
1095 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1096 | key.offset = parent; | |
1097 | } else { | |
1098 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1099 | key.offset = root_objectid; | |
1f3c79a2 LH |
1100 | } |
1101 | ||
5d4f98a2 YZ |
1102 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1103 | if (ret > 0) | |
1104 | ret = -ENOENT; | |
1105 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1106 | if (ret == -ENOENT && parent) { | |
1107 | btrfs_release_path(root, path); | |
1108 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1109 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1110 | if (ret > 0) | |
1111 | ret = -ENOENT; | |
1112 | } | |
1f3c79a2 | 1113 | #endif |
5d4f98a2 | 1114 | return ret; |
1f3c79a2 LH |
1115 | } |
1116 | ||
5d4f98a2 YZ |
1117 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1118 | struct btrfs_root *root, | |
1119 | struct btrfs_path *path, | |
1120 | u64 bytenr, u64 parent, | |
1121 | u64 root_objectid) | |
31840ae1 | 1122 | { |
5d4f98a2 | 1123 | struct btrfs_key key; |
31840ae1 | 1124 | int ret; |
31840ae1 | 1125 | |
5d4f98a2 YZ |
1126 | key.objectid = bytenr; |
1127 | if (parent) { | |
1128 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1129 | key.offset = parent; | |
1130 | } else { | |
1131 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1132 | key.offset = root_objectid; | |
1133 | } | |
1134 | ||
1135 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1136 | btrfs_release_path(root, path); | |
31840ae1 ZY |
1137 | return ret; |
1138 | } | |
1139 | ||
5d4f98a2 | 1140 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1141 | { |
5d4f98a2 YZ |
1142 | int type; |
1143 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1144 | if (parent > 0) | |
1145 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1146 | else | |
1147 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1148 | } else { | |
1149 | if (parent > 0) | |
1150 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1151 | else | |
1152 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1153 | } | |
1154 | return type; | |
31840ae1 | 1155 | } |
56bec294 | 1156 | |
2c47e605 YZ |
1157 | static int find_next_key(struct btrfs_path *path, int level, |
1158 | struct btrfs_key *key) | |
56bec294 | 1159 | |
02217ed2 | 1160 | { |
2c47e605 | 1161 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1162 | if (!path->nodes[level]) |
1163 | break; | |
5d4f98a2 YZ |
1164 | if (path->slots[level] + 1 >= |
1165 | btrfs_header_nritems(path->nodes[level])) | |
1166 | continue; | |
1167 | if (level == 0) | |
1168 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1169 | path->slots[level] + 1); | |
1170 | else | |
1171 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1172 | path->slots[level] + 1); | |
1173 | return 0; | |
1174 | } | |
1175 | return 1; | |
1176 | } | |
037e6390 | 1177 | |
5d4f98a2 YZ |
1178 | /* |
1179 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1180 | * to the address of inline back ref, and 0 is returned. | |
1181 | * | |
1182 | * if back ref isn't found, *ref_ret is set to the address where it | |
1183 | * should be inserted, and -ENOENT is returned. | |
1184 | * | |
1185 | * if insert is true and there are too many inline back refs, the path | |
1186 | * points to the extent item, and -EAGAIN is returned. | |
1187 | * | |
1188 | * NOTE: inline back refs are ordered in the same way that back ref | |
1189 | * items in the tree are ordered. | |
1190 | */ | |
1191 | static noinline_for_stack | |
1192 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1193 | struct btrfs_root *root, | |
1194 | struct btrfs_path *path, | |
1195 | struct btrfs_extent_inline_ref **ref_ret, | |
1196 | u64 bytenr, u64 num_bytes, | |
1197 | u64 parent, u64 root_objectid, | |
1198 | u64 owner, u64 offset, int insert) | |
1199 | { | |
1200 | struct btrfs_key key; | |
1201 | struct extent_buffer *leaf; | |
1202 | struct btrfs_extent_item *ei; | |
1203 | struct btrfs_extent_inline_ref *iref; | |
1204 | u64 flags; | |
1205 | u64 item_size; | |
1206 | unsigned long ptr; | |
1207 | unsigned long end; | |
1208 | int extra_size; | |
1209 | int type; | |
1210 | int want; | |
1211 | int ret; | |
1212 | int err = 0; | |
26b8003f | 1213 | |
db94535d | 1214 | key.objectid = bytenr; |
31840ae1 | 1215 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1216 | key.offset = num_bytes; |
31840ae1 | 1217 | |
5d4f98a2 YZ |
1218 | want = extent_ref_type(parent, owner); |
1219 | if (insert) { | |
1220 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1221 | path->keep_locks = 1; |
5d4f98a2 YZ |
1222 | } else |
1223 | extra_size = -1; | |
1224 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1225 | if (ret < 0) { |
5d4f98a2 YZ |
1226 | err = ret; |
1227 | goto out; | |
1228 | } | |
1229 | BUG_ON(ret); | |
1230 | ||
1231 | leaf = path->nodes[0]; | |
1232 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1233 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1234 | if (item_size < sizeof(*ei)) { | |
1235 | if (!insert) { | |
1236 | err = -ENOENT; | |
1237 | goto out; | |
1238 | } | |
1239 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1240 | extra_size); | |
1241 | if (ret < 0) { | |
1242 | err = ret; | |
1243 | goto out; | |
1244 | } | |
1245 | leaf = path->nodes[0]; | |
1246 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1247 | } | |
1248 | #endif | |
1249 | BUG_ON(item_size < sizeof(*ei)); | |
1250 | ||
5d4f98a2 YZ |
1251 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1252 | flags = btrfs_extent_flags(leaf, ei); | |
1253 | ||
1254 | ptr = (unsigned long)(ei + 1); | |
1255 | end = (unsigned long)ei + item_size; | |
1256 | ||
1257 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1258 | ptr += sizeof(struct btrfs_tree_block_info); | |
1259 | BUG_ON(ptr > end); | |
1260 | } else { | |
1261 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1262 | } | |
1263 | ||
1264 | err = -ENOENT; | |
1265 | while (1) { | |
1266 | if (ptr >= end) { | |
1267 | WARN_ON(ptr > end); | |
1268 | break; | |
1269 | } | |
1270 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1271 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1272 | if (want < type) | |
1273 | break; | |
1274 | if (want > type) { | |
1275 | ptr += btrfs_extent_inline_ref_size(type); | |
1276 | continue; | |
1277 | } | |
1278 | ||
1279 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1280 | struct btrfs_extent_data_ref *dref; | |
1281 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1282 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1283 | owner, offset)) { | |
1284 | err = 0; | |
1285 | break; | |
1286 | } | |
1287 | if (hash_extent_data_ref_item(leaf, dref) < | |
1288 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1289 | break; | |
1290 | } else { | |
1291 | u64 ref_offset; | |
1292 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1293 | if (parent > 0) { | |
1294 | if (parent == ref_offset) { | |
1295 | err = 0; | |
1296 | break; | |
1297 | } | |
1298 | if (ref_offset < parent) | |
1299 | break; | |
1300 | } else { | |
1301 | if (root_objectid == ref_offset) { | |
1302 | err = 0; | |
1303 | break; | |
1304 | } | |
1305 | if (ref_offset < root_objectid) | |
1306 | break; | |
1307 | } | |
1308 | } | |
1309 | ptr += btrfs_extent_inline_ref_size(type); | |
1310 | } | |
1311 | if (err == -ENOENT && insert) { | |
1312 | if (item_size + extra_size >= | |
1313 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1314 | err = -EAGAIN; | |
1315 | goto out; | |
1316 | } | |
1317 | /* | |
1318 | * To add new inline back ref, we have to make sure | |
1319 | * there is no corresponding back ref item. | |
1320 | * For simplicity, we just do not add new inline back | |
1321 | * ref if there is any kind of item for this block | |
1322 | */ | |
2c47e605 YZ |
1323 | if (find_next_key(path, 0, &key) == 0 && |
1324 | key.objectid == bytenr && | |
85d4198e | 1325 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1326 | err = -EAGAIN; |
1327 | goto out; | |
1328 | } | |
1329 | } | |
1330 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1331 | out: | |
85d4198e | 1332 | if (insert) { |
5d4f98a2 YZ |
1333 | path->keep_locks = 0; |
1334 | btrfs_unlock_up_safe(path, 1); | |
1335 | } | |
1336 | return err; | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * helper to add new inline back ref | |
1341 | */ | |
1342 | static noinline_for_stack | |
1343 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1344 | struct btrfs_root *root, | |
1345 | struct btrfs_path *path, | |
1346 | struct btrfs_extent_inline_ref *iref, | |
1347 | u64 parent, u64 root_objectid, | |
1348 | u64 owner, u64 offset, int refs_to_add, | |
1349 | struct btrfs_delayed_extent_op *extent_op) | |
1350 | { | |
1351 | struct extent_buffer *leaf; | |
1352 | struct btrfs_extent_item *ei; | |
1353 | unsigned long ptr; | |
1354 | unsigned long end; | |
1355 | unsigned long item_offset; | |
1356 | u64 refs; | |
1357 | int size; | |
1358 | int type; | |
1359 | int ret; | |
1360 | ||
1361 | leaf = path->nodes[0]; | |
1362 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1363 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1364 | ||
1365 | type = extent_ref_type(parent, owner); | |
1366 | size = btrfs_extent_inline_ref_size(type); | |
1367 | ||
1368 | ret = btrfs_extend_item(trans, root, path, size); | |
1369 | BUG_ON(ret); | |
1370 | ||
1371 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1372 | refs = btrfs_extent_refs(leaf, ei); | |
1373 | refs += refs_to_add; | |
1374 | btrfs_set_extent_refs(leaf, ei, refs); | |
1375 | if (extent_op) | |
1376 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1377 | ||
1378 | ptr = (unsigned long)ei + item_offset; | |
1379 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1380 | if (ptr < end - size) | |
1381 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1382 | end - size - ptr); | |
1383 | ||
1384 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1385 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1386 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1387 | struct btrfs_extent_data_ref *dref; | |
1388 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1389 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1390 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1391 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1392 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1393 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1394 | struct btrfs_shared_data_ref *sref; | |
1395 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1396 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1397 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1398 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1399 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1400 | } else { | |
1401 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1402 | } | |
1403 | btrfs_mark_buffer_dirty(leaf); | |
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1408 | struct btrfs_root *root, | |
1409 | struct btrfs_path *path, | |
1410 | struct btrfs_extent_inline_ref **ref_ret, | |
1411 | u64 bytenr, u64 num_bytes, u64 parent, | |
1412 | u64 root_objectid, u64 owner, u64 offset) | |
1413 | { | |
1414 | int ret; | |
1415 | ||
1416 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1417 | bytenr, num_bytes, parent, | |
1418 | root_objectid, owner, offset, 0); | |
1419 | if (ret != -ENOENT) | |
54aa1f4d | 1420 | return ret; |
5d4f98a2 YZ |
1421 | |
1422 | btrfs_release_path(root, path); | |
1423 | *ref_ret = NULL; | |
1424 | ||
1425 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1426 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1427 | root_objectid); | |
1428 | } else { | |
1429 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1430 | root_objectid, owner, offset); | |
b9473439 | 1431 | } |
5d4f98a2 YZ |
1432 | return ret; |
1433 | } | |
31840ae1 | 1434 | |
5d4f98a2 YZ |
1435 | /* |
1436 | * helper to update/remove inline back ref | |
1437 | */ | |
1438 | static noinline_for_stack | |
1439 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1440 | struct btrfs_root *root, | |
1441 | struct btrfs_path *path, | |
1442 | struct btrfs_extent_inline_ref *iref, | |
1443 | int refs_to_mod, | |
1444 | struct btrfs_delayed_extent_op *extent_op) | |
1445 | { | |
1446 | struct extent_buffer *leaf; | |
1447 | struct btrfs_extent_item *ei; | |
1448 | struct btrfs_extent_data_ref *dref = NULL; | |
1449 | struct btrfs_shared_data_ref *sref = NULL; | |
1450 | unsigned long ptr; | |
1451 | unsigned long end; | |
1452 | u32 item_size; | |
1453 | int size; | |
1454 | int type; | |
1455 | int ret; | |
1456 | u64 refs; | |
1457 | ||
1458 | leaf = path->nodes[0]; | |
1459 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1460 | refs = btrfs_extent_refs(leaf, ei); | |
1461 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1462 | refs += refs_to_mod; | |
1463 | btrfs_set_extent_refs(leaf, ei, refs); | |
1464 | if (extent_op) | |
1465 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1466 | ||
1467 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1468 | ||
1469 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1470 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1471 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1472 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1473 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1474 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1475 | } else { | |
1476 | refs = 1; | |
1477 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1478 | } |
31840ae1 | 1479 | |
5d4f98a2 YZ |
1480 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1481 | refs += refs_to_mod; | |
1482 | ||
1483 | if (refs > 0) { | |
1484 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1485 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1486 | else | |
1487 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1488 | } else { | |
1489 | size = btrfs_extent_inline_ref_size(type); | |
1490 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1491 | ptr = (unsigned long)iref; | |
1492 | end = (unsigned long)ei + item_size; | |
1493 | if (ptr + size < end) | |
1494 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1495 | end - ptr - size); | |
1496 | item_size -= size; | |
1497 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
1498 | BUG_ON(ret); | |
1499 | } | |
1500 | btrfs_mark_buffer_dirty(leaf); | |
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | static noinline_for_stack | |
1505 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1506 | struct btrfs_root *root, | |
1507 | struct btrfs_path *path, | |
1508 | u64 bytenr, u64 num_bytes, u64 parent, | |
1509 | u64 root_objectid, u64 owner, | |
1510 | u64 offset, int refs_to_add, | |
1511 | struct btrfs_delayed_extent_op *extent_op) | |
1512 | { | |
1513 | struct btrfs_extent_inline_ref *iref; | |
1514 | int ret; | |
1515 | ||
1516 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1517 | bytenr, num_bytes, parent, | |
1518 | root_objectid, owner, offset, 1); | |
1519 | if (ret == 0) { | |
1520 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1521 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1522 | refs_to_add, extent_op); | |
1523 | } else if (ret == -ENOENT) { | |
1524 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1525 | parent, root_objectid, | |
1526 | owner, offset, refs_to_add, | |
1527 | extent_op); | |
771ed689 | 1528 | } |
5d4f98a2 YZ |
1529 | return ret; |
1530 | } | |
31840ae1 | 1531 | |
5d4f98a2 YZ |
1532 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1533 | struct btrfs_root *root, | |
1534 | struct btrfs_path *path, | |
1535 | u64 bytenr, u64 parent, u64 root_objectid, | |
1536 | u64 owner, u64 offset, int refs_to_add) | |
1537 | { | |
1538 | int ret; | |
1539 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1540 | BUG_ON(refs_to_add != 1); | |
1541 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1542 | parent, root_objectid); | |
1543 | } else { | |
1544 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1545 | parent, root_objectid, | |
1546 | owner, offset, refs_to_add); | |
1547 | } | |
1548 | return ret; | |
1549 | } | |
56bec294 | 1550 | |
5d4f98a2 YZ |
1551 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1552 | struct btrfs_root *root, | |
1553 | struct btrfs_path *path, | |
1554 | struct btrfs_extent_inline_ref *iref, | |
1555 | int refs_to_drop, int is_data) | |
1556 | { | |
1557 | int ret; | |
b9473439 | 1558 | |
5d4f98a2 YZ |
1559 | BUG_ON(!is_data && refs_to_drop != 1); |
1560 | if (iref) { | |
1561 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1562 | -refs_to_drop, NULL); | |
1563 | } else if (is_data) { | |
1564 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1565 | } else { | |
1566 | ret = btrfs_del_item(trans, root, path); | |
1567 | } | |
1568 | return ret; | |
1569 | } | |
1570 | ||
5d4f98a2 YZ |
1571 | static void btrfs_issue_discard(struct block_device *bdev, |
1572 | u64 start, u64 len) | |
1573 | { | |
746cd1e7 CH |
1574 | blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, |
1575 | DISCARD_FL_BARRIER); | |
5d4f98a2 | 1576 | } |
5d4f98a2 YZ |
1577 | |
1578 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
1579 | u64 num_bytes) | |
1580 | { | |
5d4f98a2 YZ |
1581 | int ret; |
1582 | u64 map_length = num_bytes; | |
1583 | struct btrfs_multi_bio *multi = NULL; | |
1584 | ||
e244a0ae CH |
1585 | if (!btrfs_test_opt(root, DISCARD)) |
1586 | return 0; | |
1587 | ||
5d4f98a2 YZ |
1588 | /* Tell the block device(s) that the sectors can be discarded */ |
1589 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1590 | bytenr, &map_length, &multi, 0); | |
1591 | if (!ret) { | |
1592 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1593 | int i; | |
1594 | ||
1595 | if (map_length > num_bytes) | |
1596 | map_length = num_bytes; | |
1597 | ||
1598 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1599 | btrfs_issue_discard(stripe->dev->bdev, | |
1600 | stripe->physical, | |
1601 | map_length); | |
1602 | } | |
1603 | kfree(multi); | |
1604 | } | |
1605 | ||
1606 | return ret; | |
5d4f98a2 YZ |
1607 | } |
1608 | ||
1609 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1610 | struct btrfs_root *root, | |
1611 | u64 bytenr, u64 num_bytes, u64 parent, | |
1612 | u64 root_objectid, u64 owner, u64 offset) | |
1613 | { | |
1614 | int ret; | |
1615 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1616 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1617 | ||
1618 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1619 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1620 | parent, root_objectid, (int)owner, | |
1621 | BTRFS_ADD_DELAYED_REF, NULL); | |
1622 | } else { | |
1623 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1624 | parent, root_objectid, owner, offset, | |
1625 | BTRFS_ADD_DELAYED_REF, NULL); | |
1626 | } | |
1627 | return ret; | |
1628 | } | |
1629 | ||
1630 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1631 | struct btrfs_root *root, | |
1632 | u64 bytenr, u64 num_bytes, | |
1633 | u64 parent, u64 root_objectid, | |
1634 | u64 owner, u64 offset, int refs_to_add, | |
1635 | struct btrfs_delayed_extent_op *extent_op) | |
1636 | { | |
1637 | struct btrfs_path *path; | |
1638 | struct extent_buffer *leaf; | |
1639 | struct btrfs_extent_item *item; | |
1640 | u64 refs; | |
1641 | int ret; | |
1642 | int err = 0; | |
1643 | ||
1644 | path = btrfs_alloc_path(); | |
1645 | if (!path) | |
1646 | return -ENOMEM; | |
1647 | ||
1648 | path->reada = 1; | |
1649 | path->leave_spinning = 1; | |
1650 | /* this will setup the path even if it fails to insert the back ref */ | |
1651 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1652 | path, bytenr, num_bytes, parent, | |
1653 | root_objectid, owner, offset, | |
1654 | refs_to_add, extent_op); | |
1655 | if (ret == 0) | |
1656 | goto out; | |
1657 | ||
1658 | if (ret != -EAGAIN) { | |
1659 | err = ret; | |
1660 | goto out; | |
1661 | } | |
1662 | ||
1663 | leaf = path->nodes[0]; | |
1664 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1665 | refs = btrfs_extent_refs(leaf, item); | |
1666 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1667 | if (extent_op) | |
1668 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1669 | |
5d4f98a2 | 1670 | btrfs_mark_buffer_dirty(leaf); |
56bec294 CM |
1671 | btrfs_release_path(root->fs_info->extent_root, path); |
1672 | ||
1673 | path->reada = 1; | |
b9473439 CM |
1674 | path->leave_spinning = 1; |
1675 | ||
56bec294 CM |
1676 | /* now insert the actual backref */ |
1677 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1678 | path, bytenr, parent, root_objectid, |
1679 | owner, offset, refs_to_add); | |
56bec294 | 1680 | BUG_ON(ret); |
5d4f98a2 | 1681 | out: |
56bec294 | 1682 | btrfs_free_path(path); |
5d4f98a2 | 1683 | return err; |
56bec294 CM |
1684 | } |
1685 | ||
5d4f98a2 YZ |
1686 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1687 | struct btrfs_root *root, | |
1688 | struct btrfs_delayed_ref_node *node, | |
1689 | struct btrfs_delayed_extent_op *extent_op, | |
1690 | int insert_reserved) | |
56bec294 | 1691 | { |
5d4f98a2 YZ |
1692 | int ret = 0; |
1693 | struct btrfs_delayed_data_ref *ref; | |
1694 | struct btrfs_key ins; | |
1695 | u64 parent = 0; | |
1696 | u64 ref_root = 0; | |
1697 | u64 flags = 0; | |
1698 | ||
1699 | ins.objectid = node->bytenr; | |
1700 | ins.offset = node->num_bytes; | |
1701 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1702 | ||
1703 | ref = btrfs_delayed_node_to_data_ref(node); | |
1704 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1705 | parent = ref->parent; | |
1706 | else | |
1707 | ref_root = ref->root; | |
1708 | ||
1709 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1710 | if (extent_op) { | |
1711 | BUG_ON(extent_op->update_key); | |
1712 | flags |= extent_op->flags_to_set; | |
1713 | } | |
1714 | ret = alloc_reserved_file_extent(trans, root, | |
1715 | parent, ref_root, flags, | |
1716 | ref->objectid, ref->offset, | |
1717 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1718 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1719 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1720 | node->num_bytes, parent, | |
1721 | ref_root, ref->objectid, | |
1722 | ref->offset, node->ref_mod, | |
1723 | extent_op); | |
1724 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1725 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1726 | node->num_bytes, parent, | |
1727 | ref_root, ref->objectid, | |
1728 | ref->offset, node->ref_mod, | |
1729 | extent_op); | |
1730 | } else { | |
1731 | BUG(); | |
1732 | } | |
1733 | return ret; | |
1734 | } | |
1735 | ||
1736 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1737 | struct extent_buffer *leaf, | |
1738 | struct btrfs_extent_item *ei) | |
1739 | { | |
1740 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1741 | if (extent_op->update_flags) { | |
1742 | flags |= extent_op->flags_to_set; | |
1743 | btrfs_set_extent_flags(leaf, ei, flags); | |
1744 | } | |
1745 | ||
1746 | if (extent_op->update_key) { | |
1747 | struct btrfs_tree_block_info *bi; | |
1748 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1749 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1750 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1755 | struct btrfs_root *root, | |
1756 | struct btrfs_delayed_ref_node *node, | |
1757 | struct btrfs_delayed_extent_op *extent_op) | |
1758 | { | |
1759 | struct btrfs_key key; | |
1760 | struct btrfs_path *path; | |
1761 | struct btrfs_extent_item *ei; | |
1762 | struct extent_buffer *leaf; | |
1763 | u32 item_size; | |
56bec294 | 1764 | int ret; |
5d4f98a2 YZ |
1765 | int err = 0; |
1766 | ||
1767 | path = btrfs_alloc_path(); | |
1768 | if (!path) | |
1769 | return -ENOMEM; | |
1770 | ||
1771 | key.objectid = node->bytenr; | |
1772 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1773 | key.offset = node->num_bytes; | |
1774 | ||
1775 | path->reada = 1; | |
1776 | path->leave_spinning = 1; | |
1777 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1778 | path, 0, 1); | |
1779 | if (ret < 0) { | |
1780 | err = ret; | |
1781 | goto out; | |
1782 | } | |
1783 | if (ret > 0) { | |
1784 | err = -EIO; | |
1785 | goto out; | |
1786 | } | |
1787 | ||
1788 | leaf = path->nodes[0]; | |
1789 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1790 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1791 | if (item_size < sizeof(*ei)) { | |
1792 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1793 | path, (u64)-1, 0); | |
1794 | if (ret < 0) { | |
1795 | err = ret; | |
1796 | goto out; | |
1797 | } | |
1798 | leaf = path->nodes[0]; | |
1799 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1800 | } | |
1801 | #endif | |
1802 | BUG_ON(item_size < sizeof(*ei)); | |
1803 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1804 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1805 | |
5d4f98a2 YZ |
1806 | btrfs_mark_buffer_dirty(leaf); |
1807 | out: | |
1808 | btrfs_free_path(path); | |
1809 | return err; | |
56bec294 CM |
1810 | } |
1811 | ||
5d4f98a2 YZ |
1812 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
1813 | struct btrfs_root *root, | |
1814 | struct btrfs_delayed_ref_node *node, | |
1815 | struct btrfs_delayed_extent_op *extent_op, | |
1816 | int insert_reserved) | |
56bec294 CM |
1817 | { |
1818 | int ret = 0; | |
5d4f98a2 YZ |
1819 | struct btrfs_delayed_tree_ref *ref; |
1820 | struct btrfs_key ins; | |
1821 | u64 parent = 0; | |
1822 | u64 ref_root = 0; | |
56bec294 | 1823 | |
5d4f98a2 YZ |
1824 | ins.objectid = node->bytenr; |
1825 | ins.offset = node->num_bytes; | |
1826 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 1827 | |
5d4f98a2 YZ |
1828 | ref = btrfs_delayed_node_to_tree_ref(node); |
1829 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1830 | parent = ref->parent; | |
1831 | else | |
1832 | ref_root = ref->root; | |
1833 | ||
1834 | BUG_ON(node->ref_mod != 1); | |
1835 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1836 | BUG_ON(!extent_op || !extent_op->update_flags || | |
1837 | !extent_op->update_key); | |
1838 | ret = alloc_reserved_tree_block(trans, root, | |
1839 | parent, ref_root, | |
1840 | extent_op->flags_to_set, | |
1841 | &extent_op->key, | |
1842 | ref->level, &ins); | |
5d4f98a2 YZ |
1843 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1844 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1845 | node->num_bytes, parent, ref_root, | |
1846 | ref->level, 0, 1, extent_op); | |
1847 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1848 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1849 | node->num_bytes, parent, ref_root, | |
1850 | ref->level, 0, 1, extent_op); | |
1851 | } else { | |
1852 | BUG(); | |
1853 | } | |
56bec294 CM |
1854 | return ret; |
1855 | } | |
1856 | ||
5d4f98a2 | 1857 | |
56bec294 | 1858 | /* helper function to actually process a single delayed ref entry */ |
5d4f98a2 YZ |
1859 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
1860 | struct btrfs_root *root, | |
1861 | struct btrfs_delayed_ref_node *node, | |
1862 | struct btrfs_delayed_extent_op *extent_op, | |
1863 | int insert_reserved) | |
56bec294 CM |
1864 | { |
1865 | int ret; | |
5d4f98a2 | 1866 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
1867 | struct btrfs_delayed_ref_head *head; |
1868 | /* | |
1869 | * we've hit the end of the chain and we were supposed | |
1870 | * to insert this extent into the tree. But, it got | |
1871 | * deleted before we ever needed to insert it, so all | |
1872 | * we have to do is clean up the accounting | |
1873 | */ | |
5d4f98a2 YZ |
1874 | BUG_ON(extent_op); |
1875 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 1876 | if (insert_reserved) { |
11833d66 YZ |
1877 | int mark_free = 0; |
1878 | struct extent_buffer *must_clean = NULL; | |
1879 | ||
1880 | ret = pin_down_bytes(trans, root, NULL, | |
1881 | node->bytenr, node->num_bytes, | |
1882 | head->is_data, 1, &must_clean); | |
1883 | if (ret > 0) | |
1884 | mark_free = 1; | |
1885 | ||
1886 | if (must_clean) { | |
1887 | clean_tree_block(NULL, root, must_clean); | |
1888 | btrfs_tree_unlock(must_clean); | |
1889 | free_extent_buffer(must_clean); | |
1890 | } | |
5d4f98a2 YZ |
1891 | if (head->is_data) { |
1892 | ret = btrfs_del_csums(trans, root, | |
1893 | node->bytenr, | |
1894 | node->num_bytes); | |
1895 | BUG_ON(ret); | |
1896 | } | |
11833d66 YZ |
1897 | if (mark_free) { |
1898 | ret = btrfs_free_reserved_extent(root, | |
1899 | node->bytenr, | |
1900 | node->num_bytes); | |
1901 | BUG_ON(ret); | |
1902 | } | |
56bec294 | 1903 | } |
56bec294 CM |
1904 | mutex_unlock(&head->mutex); |
1905 | return 0; | |
1906 | } | |
1907 | ||
5d4f98a2 YZ |
1908 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1909 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1910 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
1911 | insert_reserved); | |
1912 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1913 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1914 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
1915 | insert_reserved); | |
1916 | else | |
1917 | BUG(); | |
1918 | return ret; | |
56bec294 CM |
1919 | } |
1920 | ||
1921 | static noinline struct btrfs_delayed_ref_node * | |
1922 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
1923 | { | |
1924 | struct rb_node *node; | |
1925 | struct btrfs_delayed_ref_node *ref; | |
1926 | int action = BTRFS_ADD_DELAYED_REF; | |
1927 | again: | |
1928 | /* | |
1929 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1930 | * this prevents ref count from going down to zero when | |
1931 | * there still are pending delayed ref. | |
1932 | */ | |
1933 | node = rb_prev(&head->node.rb_node); | |
1934 | while (1) { | |
1935 | if (!node) | |
1936 | break; | |
1937 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
1938 | rb_node); | |
1939 | if (ref->bytenr != head->node.bytenr) | |
1940 | break; | |
5d4f98a2 | 1941 | if (ref->action == action) |
56bec294 CM |
1942 | return ref; |
1943 | node = rb_prev(node); | |
1944 | } | |
1945 | if (action == BTRFS_ADD_DELAYED_REF) { | |
1946 | action = BTRFS_DROP_DELAYED_REF; | |
1947 | goto again; | |
1948 | } | |
1949 | return NULL; | |
1950 | } | |
1951 | ||
c3e69d58 CM |
1952 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
1953 | struct btrfs_root *root, | |
1954 | struct list_head *cluster) | |
56bec294 | 1955 | { |
56bec294 CM |
1956 | struct btrfs_delayed_ref_root *delayed_refs; |
1957 | struct btrfs_delayed_ref_node *ref; | |
1958 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 1959 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 1960 | int ret; |
c3e69d58 | 1961 | int count = 0; |
56bec294 | 1962 | int must_insert_reserved = 0; |
56bec294 CM |
1963 | |
1964 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
1965 | while (1) { |
1966 | if (!locked_ref) { | |
c3e69d58 CM |
1967 | /* pick a new head ref from the cluster list */ |
1968 | if (list_empty(cluster)) | |
56bec294 | 1969 | break; |
56bec294 | 1970 | |
c3e69d58 CM |
1971 | locked_ref = list_entry(cluster->next, |
1972 | struct btrfs_delayed_ref_head, cluster); | |
1973 | ||
1974 | /* grab the lock that says we are going to process | |
1975 | * all the refs for this head */ | |
1976 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
1977 | ||
1978 | /* | |
1979 | * we may have dropped the spin lock to get the head | |
1980 | * mutex lock, and that might have given someone else | |
1981 | * time to free the head. If that's true, it has been | |
1982 | * removed from our list and we can move on. | |
1983 | */ | |
1984 | if (ret == -EAGAIN) { | |
1985 | locked_ref = NULL; | |
1986 | count++; | |
1987 | continue; | |
56bec294 CM |
1988 | } |
1989 | } | |
a28ec197 | 1990 | |
56bec294 CM |
1991 | /* |
1992 | * record the must insert reserved flag before we | |
1993 | * drop the spin lock. | |
1994 | */ | |
1995 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1996 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 1997 | |
5d4f98a2 YZ |
1998 | extent_op = locked_ref->extent_op; |
1999 | locked_ref->extent_op = NULL; | |
2000 | ||
56bec294 CM |
2001 | /* |
2002 | * locked_ref is the head node, so we have to go one | |
2003 | * node back for any delayed ref updates | |
2004 | */ | |
56bec294 CM |
2005 | ref = select_delayed_ref(locked_ref); |
2006 | if (!ref) { | |
2007 | /* All delayed refs have been processed, Go ahead | |
2008 | * and send the head node to run_one_delayed_ref, | |
2009 | * so that any accounting fixes can happen | |
2010 | */ | |
2011 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2012 | |
2013 | if (extent_op && must_insert_reserved) { | |
2014 | kfree(extent_op); | |
2015 | extent_op = NULL; | |
2016 | } | |
2017 | ||
2018 | if (extent_op) { | |
2019 | spin_unlock(&delayed_refs->lock); | |
2020 | ||
2021 | ret = run_delayed_extent_op(trans, root, | |
2022 | ref, extent_op); | |
2023 | BUG_ON(ret); | |
2024 | kfree(extent_op); | |
2025 | ||
2026 | cond_resched(); | |
2027 | spin_lock(&delayed_refs->lock); | |
2028 | continue; | |
2029 | } | |
2030 | ||
c3e69d58 | 2031 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2032 | locked_ref = NULL; |
2033 | } | |
02217ed2 | 2034 | |
56bec294 CM |
2035 | ref->in_tree = 0; |
2036 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2037 | delayed_refs->num_entries--; | |
5d4f98a2 | 2038 | |
56bec294 | 2039 | spin_unlock(&delayed_refs->lock); |
925baedd | 2040 | |
5d4f98a2 | 2041 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2042 | must_insert_reserved); |
2043 | BUG_ON(ret); | |
eb099670 | 2044 | |
5d4f98a2 YZ |
2045 | btrfs_put_delayed_ref(ref); |
2046 | kfree(extent_op); | |
c3e69d58 | 2047 | count++; |
5d4f98a2 | 2048 | |
c3e69d58 CM |
2049 | cond_resched(); |
2050 | spin_lock(&delayed_refs->lock); | |
2051 | } | |
2052 | return count; | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * this starts processing the delayed reference count updates and | |
2057 | * extent insertions we have queued up so far. count can be | |
2058 | * 0, which means to process everything in the tree at the start | |
2059 | * of the run (but not newly added entries), or it can be some target | |
2060 | * number you'd like to process. | |
2061 | */ | |
2062 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2063 | struct btrfs_root *root, unsigned long count) | |
2064 | { | |
2065 | struct rb_node *node; | |
2066 | struct btrfs_delayed_ref_root *delayed_refs; | |
2067 | struct btrfs_delayed_ref_node *ref; | |
2068 | struct list_head cluster; | |
2069 | int ret; | |
2070 | int run_all = count == (unsigned long)-1; | |
2071 | int run_most = 0; | |
2072 | ||
2073 | if (root == root->fs_info->extent_root) | |
2074 | root = root->fs_info->tree_root; | |
2075 | ||
2076 | delayed_refs = &trans->transaction->delayed_refs; | |
2077 | INIT_LIST_HEAD(&cluster); | |
2078 | again: | |
2079 | spin_lock(&delayed_refs->lock); | |
2080 | if (count == 0) { | |
2081 | count = delayed_refs->num_entries * 2; | |
2082 | run_most = 1; | |
2083 | } | |
2084 | while (1) { | |
2085 | if (!(run_all || run_most) && | |
2086 | delayed_refs->num_heads_ready < 64) | |
2087 | break; | |
eb099670 | 2088 | |
56bec294 | 2089 | /* |
c3e69d58 CM |
2090 | * go find something we can process in the rbtree. We start at |
2091 | * the beginning of the tree, and then build a cluster | |
2092 | * of refs to process starting at the first one we are able to | |
2093 | * lock | |
56bec294 | 2094 | */ |
c3e69d58 CM |
2095 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2096 | delayed_refs->run_delayed_start); | |
2097 | if (ret) | |
56bec294 CM |
2098 | break; |
2099 | ||
c3e69d58 CM |
2100 | ret = run_clustered_refs(trans, root, &cluster); |
2101 | BUG_ON(ret < 0); | |
2102 | ||
2103 | count -= min_t(unsigned long, ret, count); | |
2104 | ||
2105 | if (count == 0) | |
2106 | break; | |
eb099670 | 2107 | } |
c3e69d58 | 2108 | |
56bec294 | 2109 | if (run_all) { |
56bec294 | 2110 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2111 | if (!node) |
56bec294 | 2112 | goto out; |
c3e69d58 | 2113 | count = (unsigned long)-1; |
e9d0b13b | 2114 | |
56bec294 CM |
2115 | while (node) { |
2116 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2117 | rb_node); | |
2118 | if (btrfs_delayed_ref_is_head(ref)) { | |
2119 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2120 | |
56bec294 CM |
2121 | head = btrfs_delayed_node_to_head(ref); |
2122 | atomic_inc(&ref->refs); | |
2123 | ||
2124 | spin_unlock(&delayed_refs->lock); | |
2125 | mutex_lock(&head->mutex); | |
2126 | mutex_unlock(&head->mutex); | |
2127 | ||
2128 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2129 | cond_resched(); |
56bec294 CM |
2130 | goto again; |
2131 | } | |
2132 | node = rb_next(node); | |
2133 | } | |
2134 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2135 | schedule_timeout(1); |
2136 | goto again; | |
5f39d397 | 2137 | } |
54aa1f4d | 2138 | out: |
c3e69d58 | 2139 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2140 | return 0; |
2141 | } | |
2142 | ||
5d4f98a2 YZ |
2143 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2144 | struct btrfs_root *root, | |
2145 | u64 bytenr, u64 num_bytes, u64 flags, | |
2146 | int is_data) | |
2147 | { | |
2148 | struct btrfs_delayed_extent_op *extent_op; | |
2149 | int ret; | |
2150 | ||
2151 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2152 | if (!extent_op) | |
2153 | return -ENOMEM; | |
2154 | ||
2155 | extent_op->flags_to_set = flags; | |
2156 | extent_op->update_flags = 1; | |
2157 | extent_op->update_key = 0; | |
2158 | extent_op->is_data = is_data ? 1 : 0; | |
2159 | ||
2160 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2161 | if (ret) | |
2162 | kfree(extent_op); | |
2163 | return ret; | |
2164 | } | |
2165 | ||
2166 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2167 | struct btrfs_root *root, | |
2168 | struct btrfs_path *path, | |
2169 | u64 objectid, u64 offset, u64 bytenr) | |
2170 | { | |
2171 | struct btrfs_delayed_ref_head *head; | |
2172 | struct btrfs_delayed_ref_node *ref; | |
2173 | struct btrfs_delayed_data_ref *data_ref; | |
2174 | struct btrfs_delayed_ref_root *delayed_refs; | |
2175 | struct rb_node *node; | |
2176 | int ret = 0; | |
2177 | ||
2178 | ret = -ENOENT; | |
2179 | delayed_refs = &trans->transaction->delayed_refs; | |
2180 | spin_lock(&delayed_refs->lock); | |
2181 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2182 | if (!head) | |
2183 | goto out; | |
2184 | ||
2185 | if (!mutex_trylock(&head->mutex)) { | |
2186 | atomic_inc(&head->node.refs); | |
2187 | spin_unlock(&delayed_refs->lock); | |
2188 | ||
2189 | btrfs_release_path(root->fs_info->extent_root, path); | |
2190 | ||
2191 | mutex_lock(&head->mutex); | |
2192 | mutex_unlock(&head->mutex); | |
2193 | btrfs_put_delayed_ref(&head->node); | |
2194 | return -EAGAIN; | |
2195 | } | |
2196 | ||
2197 | node = rb_prev(&head->node.rb_node); | |
2198 | if (!node) | |
2199 | goto out_unlock; | |
2200 | ||
2201 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2202 | ||
2203 | if (ref->bytenr != bytenr) | |
2204 | goto out_unlock; | |
2205 | ||
2206 | ret = 1; | |
2207 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2208 | goto out_unlock; | |
2209 | ||
2210 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2211 | ||
2212 | node = rb_prev(node); | |
2213 | if (node) { | |
2214 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2215 | if (ref->bytenr == bytenr) | |
2216 | goto out_unlock; | |
2217 | } | |
2218 | ||
2219 | if (data_ref->root != root->root_key.objectid || | |
2220 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2221 | goto out_unlock; | |
2222 | ||
2223 | ret = 0; | |
2224 | out_unlock: | |
2225 | mutex_unlock(&head->mutex); | |
2226 | out: | |
2227 | spin_unlock(&delayed_refs->lock); | |
2228 | return ret; | |
2229 | } | |
2230 | ||
2231 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2232 | struct btrfs_root *root, | |
2233 | struct btrfs_path *path, | |
2234 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2235 | { |
2236 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2237 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2238 | struct btrfs_extent_data_ref *ref; |
2239 | struct btrfs_extent_inline_ref *iref; | |
2240 | struct btrfs_extent_item *ei; | |
f321e491 | 2241 | struct btrfs_key key; |
5d4f98a2 | 2242 | u32 item_size; |
be20aa9d | 2243 | int ret; |
925baedd | 2244 | |
be20aa9d | 2245 | key.objectid = bytenr; |
31840ae1 | 2246 | key.offset = (u64)-1; |
f321e491 | 2247 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2248 | |
be20aa9d CM |
2249 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2250 | if (ret < 0) | |
2251 | goto out; | |
2252 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2253 | |
2254 | ret = -ENOENT; | |
2255 | if (path->slots[0] == 0) | |
31840ae1 | 2256 | goto out; |
be20aa9d | 2257 | |
31840ae1 | 2258 | path->slots[0]--; |
f321e491 | 2259 | leaf = path->nodes[0]; |
5d4f98a2 | 2260 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2261 | |
5d4f98a2 | 2262 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2263 | goto out; |
f321e491 | 2264 | |
5d4f98a2 YZ |
2265 | ret = 1; |
2266 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2267 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2268 | if (item_size < sizeof(*ei)) { | |
2269 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2270 | goto out; | |
2271 | } | |
2272 | #endif | |
2273 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2274 | |
5d4f98a2 YZ |
2275 | if (item_size != sizeof(*ei) + |
2276 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2277 | goto out; | |
be20aa9d | 2278 | |
5d4f98a2 YZ |
2279 | if (btrfs_extent_generation(leaf, ei) <= |
2280 | btrfs_root_last_snapshot(&root->root_item)) | |
2281 | goto out; | |
2282 | ||
2283 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2284 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2285 | BTRFS_EXTENT_DATA_REF_KEY) | |
2286 | goto out; | |
2287 | ||
2288 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2289 | if (btrfs_extent_refs(leaf, ei) != | |
2290 | btrfs_extent_data_ref_count(leaf, ref) || | |
2291 | btrfs_extent_data_ref_root(leaf, ref) != | |
2292 | root->root_key.objectid || | |
2293 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2294 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2295 | goto out; | |
2296 | ||
2297 | ret = 0; | |
2298 | out: | |
2299 | return ret; | |
2300 | } | |
2301 | ||
2302 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2303 | struct btrfs_root *root, | |
2304 | u64 objectid, u64 offset, u64 bytenr) | |
2305 | { | |
2306 | struct btrfs_path *path; | |
2307 | int ret; | |
2308 | int ret2; | |
2309 | ||
2310 | path = btrfs_alloc_path(); | |
2311 | if (!path) | |
2312 | return -ENOENT; | |
2313 | ||
2314 | do { | |
2315 | ret = check_committed_ref(trans, root, path, objectid, | |
2316 | offset, bytenr); | |
2317 | if (ret && ret != -ENOENT) | |
f321e491 | 2318 | goto out; |
80ff3856 | 2319 | |
5d4f98a2 YZ |
2320 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2321 | offset, bytenr); | |
2322 | } while (ret2 == -EAGAIN); | |
2323 | ||
2324 | if (ret2 && ret2 != -ENOENT) { | |
2325 | ret = ret2; | |
2326 | goto out; | |
f321e491 | 2327 | } |
5d4f98a2 YZ |
2328 | |
2329 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2330 | ret = 0; | |
be20aa9d | 2331 | out: |
80ff3856 | 2332 | btrfs_free_path(path); |
f321e491 | 2333 | return ret; |
be20aa9d | 2334 | } |
c5739bba | 2335 | |
5d4f98a2 | 2336 | #if 0 |
31840ae1 ZY |
2337 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2338 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 2339 | { |
5f39d397 | 2340 | struct btrfs_key key; |
6407bf6d | 2341 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
2342 | u64 root_gen; |
2343 | u32 nritems; | |
02217ed2 | 2344 | int i; |
db94535d | 2345 | int level; |
31840ae1 | 2346 | int ret = 0; |
e4657689 | 2347 | int shared = 0; |
a28ec197 | 2348 | |
3768f368 | 2349 | if (!root->ref_cows) |
a28ec197 | 2350 | return 0; |
5f39d397 | 2351 | |
e4657689 ZY |
2352 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
2353 | shared = 0; | |
2354 | root_gen = root->root_key.offset; | |
2355 | } else { | |
2356 | shared = 1; | |
2357 | root_gen = trans->transid - 1; | |
2358 | } | |
2359 | ||
db94535d | 2360 | level = btrfs_header_level(buf); |
5f39d397 | 2361 | nritems = btrfs_header_nritems(buf); |
4a096752 | 2362 | |
31840ae1 | 2363 | if (level == 0) { |
31153d81 YZ |
2364 | struct btrfs_leaf_ref *ref; |
2365 | struct btrfs_extent_info *info; | |
2366 | ||
31840ae1 | 2367 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 2368 | if (!ref) { |
31840ae1 | 2369 | ret = -ENOMEM; |
31153d81 YZ |
2370 | goto out; |
2371 | } | |
2372 | ||
e4657689 | 2373 | ref->root_gen = root_gen; |
31153d81 YZ |
2374 | ref->bytenr = buf->start; |
2375 | ref->owner = btrfs_header_owner(buf); | |
2376 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 2377 | ref->nritems = nr_extents; |
31153d81 | 2378 | info = ref->extents; |
bcc63abb | 2379 | |
31840ae1 | 2380 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
2381 | u64 disk_bytenr; |
2382 | btrfs_item_key_to_cpu(buf, &key, i); | |
2383 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2384 | continue; | |
2385 | fi = btrfs_item_ptr(buf, i, | |
2386 | struct btrfs_file_extent_item); | |
2387 | if (btrfs_file_extent_type(buf, fi) == | |
2388 | BTRFS_FILE_EXTENT_INLINE) | |
2389 | continue; | |
2390 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2391 | if (disk_bytenr == 0) | |
2392 | continue; | |
2393 | ||
2394 | info->bytenr = disk_bytenr; | |
2395 | info->num_bytes = | |
2396 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
2397 | info->objectid = key.objectid; | |
2398 | info->offset = key.offset; | |
2399 | info++; | |
2400 | } | |
2401 | ||
e4657689 | 2402 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
2403 | if (ret == -EEXIST && shared) { |
2404 | struct btrfs_leaf_ref *old; | |
2405 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
2406 | BUG_ON(!old); | |
2407 | btrfs_remove_leaf_ref(root, old); | |
2408 | btrfs_free_leaf_ref(root, old); | |
2409 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
2410 | } | |
31153d81 | 2411 | WARN_ON(ret); |
bcc63abb | 2412 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2413 | } |
2414 | out: | |
31840ae1 ZY |
2415 | return ret; |
2416 | } | |
2417 | ||
b7a9f29f CM |
2418 | /* when a block goes through cow, we update the reference counts of |
2419 | * everything that block points to. The internal pointers of the block | |
2420 | * can be in just about any order, and it is likely to have clusters of | |
2421 | * things that are close together and clusters of things that are not. | |
2422 | * | |
2423 | * To help reduce the seeks that come with updating all of these reference | |
2424 | * counts, sort them by byte number before actual updates are done. | |
2425 | * | |
2426 | * struct refsort is used to match byte number to slot in the btree block. | |
2427 | * we sort based on the byte number and then use the slot to actually | |
2428 | * find the item. | |
bd56b302 CM |
2429 | * |
2430 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
2431 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
2432 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
2433 | * single node to be bigger than a page. | |
b7a9f29f CM |
2434 | */ |
2435 | struct refsort { | |
2436 | u64 bytenr; | |
2437 | u32 slot; | |
2438 | }; | |
2439 | ||
2440 | /* | |
2441 | * for passing into sort() | |
2442 | */ | |
2443 | static int refsort_cmp(const void *a_void, const void *b_void) | |
2444 | { | |
2445 | const struct refsort *a = a_void; | |
2446 | const struct refsort *b = b_void; | |
2447 | ||
2448 | if (a->bytenr < b->bytenr) | |
2449 | return -1; | |
2450 | if (a->bytenr > b->bytenr) | |
2451 | return 1; | |
2452 | return 0; | |
2453 | } | |
5d4f98a2 | 2454 | #endif |
b7a9f29f | 2455 | |
5d4f98a2 | 2456 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2457 | struct btrfs_root *root, |
5d4f98a2 YZ |
2458 | struct extent_buffer *buf, |
2459 | int full_backref, int inc) | |
31840ae1 ZY |
2460 | { |
2461 | u64 bytenr; | |
5d4f98a2 YZ |
2462 | u64 num_bytes; |
2463 | u64 parent; | |
31840ae1 | 2464 | u64 ref_root; |
31840ae1 | 2465 | u32 nritems; |
31840ae1 ZY |
2466 | struct btrfs_key key; |
2467 | struct btrfs_file_extent_item *fi; | |
2468 | int i; | |
2469 | int level; | |
2470 | int ret = 0; | |
31840ae1 | 2471 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2472 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2473 | |
2474 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2475 | nritems = btrfs_header_nritems(buf); |
2476 | level = btrfs_header_level(buf); | |
2477 | ||
5d4f98a2 YZ |
2478 | if (!root->ref_cows && level == 0) |
2479 | return 0; | |
31840ae1 | 2480 | |
5d4f98a2 YZ |
2481 | if (inc) |
2482 | process_func = btrfs_inc_extent_ref; | |
2483 | else | |
2484 | process_func = btrfs_free_extent; | |
31840ae1 | 2485 | |
5d4f98a2 YZ |
2486 | if (full_backref) |
2487 | parent = buf->start; | |
2488 | else | |
2489 | parent = 0; | |
2490 | ||
2491 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2492 | if (level == 0) { |
5d4f98a2 | 2493 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2494 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2495 | continue; | |
5d4f98a2 | 2496 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2497 | struct btrfs_file_extent_item); |
2498 | if (btrfs_file_extent_type(buf, fi) == | |
2499 | BTRFS_FILE_EXTENT_INLINE) | |
2500 | continue; | |
2501 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2502 | if (bytenr == 0) | |
2503 | continue; | |
5d4f98a2 YZ |
2504 | |
2505 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2506 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2507 | ret = process_func(trans, root, bytenr, num_bytes, | |
2508 | parent, ref_root, key.objectid, | |
2509 | key.offset); | |
31840ae1 ZY |
2510 | if (ret) |
2511 | goto fail; | |
2512 | } else { | |
5d4f98a2 YZ |
2513 | bytenr = btrfs_node_blockptr(buf, i); |
2514 | num_bytes = btrfs_level_size(root, level - 1); | |
2515 | ret = process_func(trans, root, bytenr, num_bytes, | |
2516 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2517 | if (ret) |
2518 | goto fail; | |
2519 | } | |
2520 | } | |
2521 | return 0; | |
2522 | fail: | |
5d4f98a2 YZ |
2523 | BUG(); |
2524 | return ret; | |
2525 | } | |
2526 | ||
2527 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2528 | struct extent_buffer *buf, int full_backref) | |
2529 | { | |
2530 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2531 | } | |
2532 | ||
2533 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2534 | struct extent_buffer *buf, int full_backref) | |
2535 | { | |
2536 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2537 | } |
2538 | ||
9078a3e1 CM |
2539 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2540 | struct btrfs_root *root, | |
2541 | struct btrfs_path *path, | |
2542 | struct btrfs_block_group_cache *cache) | |
2543 | { | |
2544 | int ret; | |
9078a3e1 | 2545 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2546 | unsigned long bi; |
2547 | struct extent_buffer *leaf; | |
9078a3e1 | 2548 | |
9078a3e1 | 2549 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2550 | if (ret < 0) |
2551 | goto fail; | |
9078a3e1 | 2552 | BUG_ON(ret); |
5f39d397 CM |
2553 | |
2554 | leaf = path->nodes[0]; | |
2555 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2556 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2557 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 2558 | btrfs_release_path(extent_root, path); |
54aa1f4d | 2559 | fail: |
9078a3e1 CM |
2560 | if (ret) |
2561 | return ret; | |
9078a3e1 CM |
2562 | return 0; |
2563 | ||
2564 | } | |
2565 | ||
4a8c9a62 YZ |
2566 | static struct btrfs_block_group_cache * |
2567 | next_block_group(struct btrfs_root *root, | |
2568 | struct btrfs_block_group_cache *cache) | |
2569 | { | |
2570 | struct rb_node *node; | |
2571 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2572 | node = rb_next(&cache->cache_node); | |
2573 | btrfs_put_block_group(cache); | |
2574 | if (node) { | |
2575 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2576 | cache_node); | |
2577 | atomic_inc(&cache->count); | |
2578 | } else | |
2579 | cache = NULL; | |
2580 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2581 | return cache; | |
2582 | } | |
2583 | ||
96b5179d CM |
2584 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2585 | struct btrfs_root *root) | |
9078a3e1 | 2586 | { |
4a8c9a62 | 2587 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2588 | int err = 0; |
9078a3e1 | 2589 | struct btrfs_path *path; |
96b5179d | 2590 | u64 last = 0; |
9078a3e1 CM |
2591 | |
2592 | path = btrfs_alloc_path(); | |
2593 | if (!path) | |
2594 | return -ENOMEM; | |
2595 | ||
d397712b | 2596 | while (1) { |
4a8c9a62 YZ |
2597 | if (last == 0) { |
2598 | err = btrfs_run_delayed_refs(trans, root, | |
2599 | (unsigned long)-1); | |
2600 | BUG_ON(err); | |
0f9dd46c | 2601 | } |
54aa1f4d | 2602 | |
4a8c9a62 YZ |
2603 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2604 | while (cache) { | |
2605 | if (cache->dirty) | |
2606 | break; | |
2607 | cache = next_block_group(root, cache); | |
2608 | } | |
2609 | if (!cache) { | |
2610 | if (last == 0) | |
2611 | break; | |
2612 | last = 0; | |
2613 | continue; | |
2614 | } | |
0f9dd46c | 2615 | |
e8569813 | 2616 | cache->dirty = 0; |
4a8c9a62 | 2617 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2618 | |
4a8c9a62 YZ |
2619 | err = write_one_cache_group(trans, root, path, cache); |
2620 | BUG_ON(err); | |
2621 | btrfs_put_block_group(cache); | |
9078a3e1 | 2622 | } |
4a8c9a62 | 2623 | |
9078a3e1 | 2624 | btrfs_free_path(path); |
4a8c9a62 | 2625 | return 0; |
9078a3e1 CM |
2626 | } |
2627 | ||
d2fb3437 YZ |
2628 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2629 | { | |
2630 | struct btrfs_block_group_cache *block_group; | |
2631 | int readonly = 0; | |
2632 | ||
2633 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2634 | if (!block_group || block_group->ro) | |
2635 | readonly = 1; | |
2636 | if (block_group) | |
fa9c0d79 | 2637 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2638 | return readonly; |
2639 | } | |
2640 | ||
593060d7 CM |
2641 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2642 | u64 total_bytes, u64 bytes_used, | |
2643 | struct btrfs_space_info **space_info) | |
2644 | { | |
2645 | struct btrfs_space_info *found; | |
2646 | ||
2647 | found = __find_space_info(info, flags); | |
2648 | if (found) { | |
25179201 | 2649 | spin_lock(&found->lock); |
593060d7 CM |
2650 | found->total_bytes += total_bytes; |
2651 | found->bytes_used += bytes_used; | |
8f18cf13 | 2652 | found->full = 0; |
25179201 | 2653 | spin_unlock(&found->lock); |
593060d7 CM |
2654 | *space_info = found; |
2655 | return 0; | |
2656 | } | |
c146afad | 2657 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2658 | if (!found) |
2659 | return -ENOMEM; | |
2660 | ||
0f9dd46c | 2661 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 2662 | init_rwsem(&found->groups_sem); |
0f9dd46c | 2663 | spin_lock_init(&found->lock); |
593060d7 CM |
2664 | found->flags = flags; |
2665 | found->total_bytes = total_bytes; | |
2666 | found->bytes_used = bytes_used; | |
2667 | found->bytes_pinned = 0; | |
e8569813 | 2668 | found->bytes_reserved = 0; |
c146afad | 2669 | found->bytes_readonly = 0; |
6a63209f | 2670 | found->bytes_delalloc = 0; |
593060d7 | 2671 | found->full = 0; |
0ef3e66b | 2672 | found->force_alloc = 0; |
593060d7 | 2673 | *space_info = found; |
4184ea7f | 2674 | list_add_rcu(&found->list, &info->space_info); |
817d52f8 | 2675 | atomic_set(&found->caching_threads, 0); |
593060d7 CM |
2676 | return 0; |
2677 | } | |
2678 | ||
8790d502 CM |
2679 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
2680 | { | |
2681 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 2682 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 2683 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 2684 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
2685 | if (extra_flags) { |
2686 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
2687 | fs_info->avail_data_alloc_bits |= extra_flags; | |
2688 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2689 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
2690 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2691 | fs_info->avail_system_alloc_bits |= extra_flags; | |
2692 | } | |
2693 | } | |
593060d7 | 2694 | |
c146afad YZ |
2695 | static void set_block_group_readonly(struct btrfs_block_group_cache *cache) |
2696 | { | |
2697 | spin_lock(&cache->space_info->lock); | |
2698 | spin_lock(&cache->lock); | |
2699 | if (!cache->ro) { | |
2700 | cache->space_info->bytes_readonly += cache->key.offset - | |
2701 | btrfs_block_group_used(&cache->item); | |
2702 | cache->ro = 1; | |
2703 | } | |
2704 | spin_unlock(&cache->lock); | |
2705 | spin_unlock(&cache->space_info->lock); | |
2706 | } | |
2707 | ||
2b82032c | 2708 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 2709 | { |
2b82032c | 2710 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
a061fc8d CM |
2711 | |
2712 | if (num_devices == 1) | |
2713 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
2714 | if (num_devices < 4) | |
2715 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
2716 | ||
ec44a35c CM |
2717 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
2718 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 2719 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 2720 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 2721 | } |
ec44a35c CM |
2722 | |
2723 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 2724 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 2725 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 2726 | } |
ec44a35c CM |
2727 | |
2728 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
2729 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
2730 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
2731 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
2732 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
2733 | return flags; | |
2734 | } | |
2735 | ||
6a63209f JB |
2736 | static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) |
2737 | { | |
2738 | struct btrfs_fs_info *info = root->fs_info; | |
2739 | u64 alloc_profile; | |
2740 | ||
2741 | if (data) { | |
2742 | alloc_profile = info->avail_data_alloc_bits & | |
2743 | info->data_alloc_profile; | |
2744 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
2745 | } else if (root == root->fs_info->chunk_root) { | |
2746 | alloc_profile = info->avail_system_alloc_bits & | |
2747 | info->system_alloc_profile; | |
2748 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
2749 | } else { | |
2750 | alloc_profile = info->avail_metadata_alloc_bits & | |
2751 | info->metadata_alloc_profile; | |
2752 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
2753 | } | |
2754 | ||
2755 | return btrfs_reduce_alloc_profile(root, data); | |
2756 | } | |
2757 | ||
2758 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) | |
2759 | { | |
2760 | u64 alloc_target; | |
2761 | ||
2762 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
2763 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, | |
2764 | alloc_target); | |
2765 | } | |
2766 | ||
9ed74f2d JB |
2767 | static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items) |
2768 | { | |
2769 | u64 num_bytes; | |
2770 | int level; | |
2771 | ||
2772 | level = BTRFS_MAX_LEVEL - 2; | |
2773 | /* | |
2774 | * NOTE: these calculations are absolutely the worst possible case. | |
2775 | * This assumes that _every_ item we insert will require a new leaf, and | |
2776 | * that the tree has grown to its maximum level size. | |
2777 | */ | |
2778 | ||
2779 | /* | |
2780 | * for every item we insert we could insert both an extent item and a | |
2781 | * extent ref item. Then for ever item we insert, we will need to cow | |
2782 | * both the original leaf, plus the leaf to the left and right of it. | |
2783 | * | |
2784 | * Unless we are talking about the extent root, then we just want the | |
2785 | * number of items * 2, since we just need the extent item plus its ref. | |
2786 | */ | |
2787 | if (root == root->fs_info->extent_root) | |
2788 | num_bytes = num_items * 2; | |
2789 | else | |
2790 | num_bytes = (num_items + (2 * num_items)) * 3; | |
2791 | ||
2792 | /* | |
2793 | * num_bytes is total number of leaves we could need times the leaf | |
2794 | * size, and then for every leaf we could end up cow'ing 2 nodes per | |
2795 | * level, down to the leaf level. | |
2796 | */ | |
2797 | num_bytes = (num_bytes * root->leafsize) + | |
2798 | (num_bytes * (level * 2)) * root->nodesize; | |
2799 | ||
2800 | return num_bytes; | |
2801 | } | |
2802 | ||
6a63209f | 2803 | /* |
9ed74f2d JB |
2804 | * Unreserve metadata space for delalloc. If we have less reserved credits than |
2805 | * we have extents, this function does nothing. | |
6a63209f | 2806 | */ |
9ed74f2d JB |
2807 | int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root, |
2808 | struct inode *inode, int num_items) | |
6a63209f JB |
2809 | { |
2810 | struct btrfs_fs_info *info = root->fs_info; | |
2811 | struct btrfs_space_info *meta_sinfo; | |
9ed74f2d JB |
2812 | u64 num_bytes; |
2813 | u64 alloc_target; | |
2814 | bool bug = false; | |
6a63209f JB |
2815 | |
2816 | /* get the space info for where the metadata will live */ | |
2817 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
2818 | meta_sinfo = __find_space_info(info, alloc_target); | |
2819 | ||
9ed74f2d JB |
2820 | num_bytes = calculate_bytes_needed(root->fs_info->extent_root, |
2821 | num_items); | |
2822 | ||
6a63209f | 2823 | spin_lock(&meta_sinfo->lock); |
32c00aff JB |
2824 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
2825 | if (BTRFS_I(inode)->reserved_extents <= | |
2826 | BTRFS_I(inode)->outstanding_extents) { | |
2827 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
2828 | spin_unlock(&meta_sinfo->lock); |
2829 | return 0; | |
2830 | } | |
32c00aff | 2831 | spin_unlock(&BTRFS_I(inode)->accounting_lock); |
9ed74f2d | 2832 | |
32c00aff JB |
2833 | BTRFS_I(inode)->reserved_extents--; |
2834 | BUG_ON(BTRFS_I(inode)->reserved_extents < 0); | |
9ed74f2d JB |
2835 | |
2836 | if (meta_sinfo->bytes_delalloc < num_bytes) { | |
2837 | bug = true; | |
2838 | meta_sinfo->bytes_delalloc = 0; | |
2839 | } else { | |
2840 | meta_sinfo->bytes_delalloc -= num_bytes; | |
2841 | } | |
2842 | spin_unlock(&meta_sinfo->lock); | |
2843 | ||
2844 | BUG_ON(bug); | |
2845 | ||
2846 | return 0; | |
2847 | } | |
6a63209f | 2848 | |
9ed74f2d JB |
2849 | static void check_force_delalloc(struct btrfs_space_info *meta_sinfo) |
2850 | { | |
2851 | u64 thresh; | |
2852 | ||
2853 | thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
2854 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly + | |
2855 | meta_sinfo->bytes_super + meta_sinfo->bytes_root + | |
2856 | meta_sinfo->bytes_may_use; | |
6a63209f | 2857 | |
9ed74f2d JB |
2858 | thresh = meta_sinfo->total_bytes - thresh; |
2859 | thresh *= 80; | |
6a63209f | 2860 | do_div(thresh, 100); |
9ed74f2d JB |
2861 | if (thresh <= meta_sinfo->bytes_delalloc) |
2862 | meta_sinfo->force_delalloc = 1; | |
2863 | else | |
2864 | meta_sinfo->force_delalloc = 0; | |
2865 | } | |
6a63209f | 2866 | |
e3ccfa98 JB |
2867 | struct async_flush { |
2868 | struct btrfs_root *root; | |
2869 | struct btrfs_space_info *info; | |
2870 | struct btrfs_work work; | |
2871 | }; | |
2872 | ||
2873 | static noinline void flush_delalloc_async(struct btrfs_work *work) | |
2874 | { | |
2875 | struct async_flush *async; | |
2876 | struct btrfs_root *root; | |
2877 | struct btrfs_space_info *info; | |
2878 | ||
2879 | async = container_of(work, struct async_flush, work); | |
2880 | root = async->root; | |
2881 | info = async->info; | |
2882 | ||
24bbcf04 | 2883 | btrfs_start_delalloc_inodes(root, 0); |
e3ccfa98 | 2884 | wake_up(&info->flush_wait); |
24bbcf04 | 2885 | btrfs_wait_ordered_extents(root, 0, 0); |
e3ccfa98 JB |
2886 | |
2887 | spin_lock(&info->lock); | |
2888 | info->flushing = 0; | |
2889 | spin_unlock(&info->lock); | |
2890 | wake_up(&info->flush_wait); | |
2891 | ||
2892 | kfree(async); | |
2893 | } | |
2894 | ||
2895 | static void wait_on_flush(struct btrfs_space_info *info) | |
2896 | { | |
2897 | DEFINE_WAIT(wait); | |
2898 | u64 used; | |
2899 | ||
2900 | while (1) { | |
2901 | prepare_to_wait(&info->flush_wait, &wait, | |
2902 | TASK_UNINTERRUPTIBLE); | |
2903 | spin_lock(&info->lock); | |
2904 | if (!info->flushing) { | |
2905 | spin_unlock(&info->lock); | |
2906 | break; | |
2907 | } | |
2908 | ||
2909 | used = info->bytes_used + info->bytes_reserved + | |
2910 | info->bytes_pinned + info->bytes_readonly + | |
2911 | info->bytes_super + info->bytes_root + | |
2912 | info->bytes_may_use + info->bytes_delalloc; | |
2913 | if (used < info->total_bytes) { | |
2914 | spin_unlock(&info->lock); | |
2915 | break; | |
2916 | } | |
2917 | spin_unlock(&info->lock); | |
2918 | schedule(); | |
2919 | } | |
2920 | finish_wait(&info->flush_wait, &wait); | |
2921 | } | |
2922 | ||
32c00aff JB |
2923 | static void flush_delalloc(struct btrfs_root *root, |
2924 | struct btrfs_space_info *info) | |
2925 | { | |
e3ccfa98 | 2926 | struct async_flush *async; |
32c00aff JB |
2927 | bool wait = false; |
2928 | ||
2929 | spin_lock(&info->lock); | |
2930 | ||
2931 | if (!info->flushing) { | |
2932 | info->flushing = 1; | |
2933 | init_waitqueue_head(&info->flush_wait); | |
2934 | } else { | |
2935 | wait = true; | |
2936 | } | |
2937 | ||
2938 | spin_unlock(&info->lock); | |
2939 | ||
2940 | if (wait) { | |
e3ccfa98 | 2941 | wait_on_flush(info); |
32c00aff JB |
2942 | return; |
2943 | } | |
2944 | ||
e3ccfa98 JB |
2945 | async = kzalloc(sizeof(*async), GFP_NOFS); |
2946 | if (!async) | |
2947 | goto flush; | |
2948 | ||
2949 | async->root = root; | |
2950 | async->info = info; | |
2951 | async->work.func = flush_delalloc_async; | |
2952 | ||
2953 | btrfs_queue_worker(&root->fs_info->enospc_workers, | |
2954 | &async->work); | |
2955 | wait_on_flush(info); | |
2956 | return; | |
2957 | ||
2958 | flush: | |
24bbcf04 YZ |
2959 | btrfs_start_delalloc_inodes(root, 0); |
2960 | btrfs_wait_ordered_extents(root, 0, 0); | |
32c00aff JB |
2961 | |
2962 | spin_lock(&info->lock); | |
2963 | info->flushing = 0; | |
2964 | spin_unlock(&info->lock); | |
2965 | wake_up(&info->flush_wait); | |
2966 | } | |
2967 | ||
9ed74f2d JB |
2968 | static int maybe_allocate_chunk(struct btrfs_root *root, |
2969 | struct btrfs_space_info *info) | |
2970 | { | |
2971 | struct btrfs_super_block *disk_super = &root->fs_info->super_copy; | |
2972 | struct btrfs_trans_handle *trans; | |
2973 | bool wait = false; | |
2974 | int ret = 0; | |
2975 | u64 min_metadata; | |
2976 | u64 free_space; | |
2977 | ||
2978 | free_space = btrfs_super_total_bytes(disk_super); | |
2979 | /* | |
33b25808 | 2980 | * we allow the metadata to grow to a max of either 10gb or 5% of the |
9ed74f2d JB |
2981 | * space in the volume. |
2982 | */ | |
33b25808 | 2983 | min_metadata = min((u64)10 * 1024 * 1024 * 1024, |
9ed74f2d JB |
2984 | div64_u64(free_space * 5, 100)); |
2985 | if (info->total_bytes >= min_metadata) { | |
2986 | spin_unlock(&info->lock); | |
2987 | return 0; | |
2988 | } | |
2989 | ||
2990 | if (info->full) { | |
2991 | spin_unlock(&info->lock); | |
2992 | return 0; | |
2993 | } | |
2994 | ||
2995 | if (!info->allocating_chunk) { | |
2996 | info->force_alloc = 1; | |
2997 | info->allocating_chunk = 1; | |
e3ccfa98 | 2998 | init_waitqueue_head(&info->allocate_wait); |
9ed74f2d JB |
2999 | } else { |
3000 | wait = true; | |
3001 | } | |
3002 | ||
3003 | spin_unlock(&info->lock); | |
3004 | ||
3005 | if (wait) { | |
e3ccfa98 | 3006 | wait_event(info->allocate_wait, |
9ed74f2d JB |
3007 | !info->allocating_chunk); |
3008 | return 1; | |
3009 | } | |
3010 | ||
3011 | trans = btrfs_start_transaction(root, 1); | |
3012 | if (!trans) { | |
3013 | ret = -ENOMEM; | |
3014 | goto out; | |
3015 | } | |
3016 | ||
3017 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
3018 | 4096 + 2 * 1024 * 1024, | |
3019 | info->flags, 0); | |
3020 | btrfs_end_transaction(trans, root); | |
3021 | if (ret) | |
3022 | goto out; | |
3023 | out: | |
3024 | spin_lock(&info->lock); | |
3025 | info->allocating_chunk = 0; | |
3026 | spin_unlock(&info->lock); | |
e3ccfa98 | 3027 | wake_up(&info->allocate_wait); |
9ed74f2d JB |
3028 | |
3029 | if (ret) | |
3030 | return 0; | |
3031 | return 1; | |
3032 | } | |
3033 | ||
3034 | /* | |
3035 | * Reserve metadata space for delalloc. | |
3036 | */ | |
3037 | int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root, | |
3038 | struct inode *inode, int num_items) | |
3039 | { | |
3040 | struct btrfs_fs_info *info = root->fs_info; | |
3041 | struct btrfs_space_info *meta_sinfo; | |
3042 | u64 num_bytes; | |
3043 | u64 used; | |
3044 | u64 alloc_target; | |
3045 | int flushed = 0; | |
3046 | int force_delalloc; | |
3047 | ||
3048 | /* get the space info for where the metadata will live */ | |
3049 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
3050 | meta_sinfo = __find_space_info(info, alloc_target); | |
3051 | ||
3052 | num_bytes = calculate_bytes_needed(root->fs_info->extent_root, | |
3053 | num_items); | |
3054 | again: | |
3055 | spin_lock(&meta_sinfo->lock); | |
3056 | ||
3057 | force_delalloc = meta_sinfo->force_delalloc; | |
3058 | ||
3059 | if (unlikely(!meta_sinfo->bytes_root)) | |
3060 | meta_sinfo->bytes_root = calculate_bytes_needed(root, 6); | |
3061 | ||
3062 | if (!flushed) | |
3063 | meta_sinfo->bytes_delalloc += num_bytes; | |
3064 | ||
3065 | used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
3066 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly + | |
3067 | meta_sinfo->bytes_super + meta_sinfo->bytes_root + | |
3068 | meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc; | |
3069 | ||
3070 | if (used > meta_sinfo->total_bytes) { | |
3071 | flushed++; | |
3072 | ||
3073 | if (flushed == 1) { | |
3074 | if (maybe_allocate_chunk(root, meta_sinfo)) | |
3075 | goto again; | |
3076 | flushed++; | |
3077 | } else { | |
4e06bdd6 | 3078 | spin_unlock(&meta_sinfo->lock); |
9ed74f2d | 3079 | } |
4e06bdd6 | 3080 | |
9ed74f2d JB |
3081 | if (flushed == 2) { |
3082 | filemap_flush(inode->i_mapping); | |
3083 | goto again; | |
3084 | } else if (flushed == 3) { | |
32c00aff | 3085 | flush_delalloc(root, meta_sinfo); |
4e06bdd6 JB |
3086 | goto again; |
3087 | } | |
9ed74f2d JB |
3088 | spin_lock(&meta_sinfo->lock); |
3089 | meta_sinfo->bytes_delalloc -= num_bytes; | |
6a63209f | 3090 | spin_unlock(&meta_sinfo->lock); |
9ed74f2d | 3091 | printk(KERN_ERR "enospc, has %d, reserved %d\n", |
32c00aff JB |
3092 | BTRFS_I(inode)->outstanding_extents, |
3093 | BTRFS_I(inode)->reserved_extents); | |
9ed74f2d JB |
3094 | dump_space_info(meta_sinfo, 0, 0); |
3095 | return -ENOSPC; | |
3096 | } | |
4e06bdd6 | 3097 | |
32c00aff | 3098 | BTRFS_I(inode)->reserved_extents++; |
9ed74f2d JB |
3099 | check_force_delalloc(meta_sinfo); |
3100 | spin_unlock(&meta_sinfo->lock); | |
3101 | ||
3102 | if (!flushed && force_delalloc) | |
3103 | filemap_flush(inode->i_mapping); | |
3104 | ||
3105 | return 0; | |
3106 | } | |
3107 | ||
3108 | /* | |
3109 | * unreserve num_items number of items worth of metadata space. This needs to | |
3110 | * be paired with btrfs_reserve_metadata_space. | |
3111 | * | |
3112 | * NOTE: if you have the option, run this _AFTER_ you do a | |
3113 | * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref | |
3114 | * oprations which will result in more used metadata, so we want to make sure we | |
3115 | * can do that without issue. | |
3116 | */ | |
3117 | int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items) | |
3118 | { | |
3119 | struct btrfs_fs_info *info = root->fs_info; | |
3120 | struct btrfs_space_info *meta_sinfo; | |
3121 | u64 num_bytes; | |
3122 | u64 alloc_target; | |
3123 | bool bug = false; | |
3124 | ||
3125 | /* get the space info for where the metadata will live */ | |
3126 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
3127 | meta_sinfo = __find_space_info(info, alloc_target); | |
3128 | ||
3129 | num_bytes = calculate_bytes_needed(root, num_items); | |
3130 | ||
3131 | spin_lock(&meta_sinfo->lock); | |
3132 | if (meta_sinfo->bytes_may_use < num_bytes) { | |
3133 | bug = true; | |
3134 | meta_sinfo->bytes_may_use = 0; | |
3135 | } else { | |
3136 | meta_sinfo->bytes_may_use -= num_bytes; | |
3137 | } | |
3138 | spin_unlock(&meta_sinfo->lock); | |
3139 | ||
3140 | BUG_ON(bug); | |
3141 | ||
3142 | return 0; | |
3143 | } | |
3144 | ||
3145 | /* | |
3146 | * Reserve some metadata space for use. We'll calculate the worste case number | |
3147 | * of bytes that would be needed to modify num_items number of items. If we | |
3148 | * have space, fantastic, if not, you get -ENOSPC. Please call | |
3149 | * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of | |
3150 | * items you reserved, since whatever metadata you needed should have already | |
3151 | * been allocated. | |
3152 | * | |
3153 | * This will commit the transaction to make more space if we don't have enough | |
3154 | * metadata space. THe only time we don't do this is if we're reserving space | |
3155 | * inside of a transaction, then we will just return -ENOSPC and it is the | |
3156 | * callers responsibility to handle it properly. | |
3157 | */ | |
3158 | int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items) | |
3159 | { | |
3160 | struct btrfs_fs_info *info = root->fs_info; | |
3161 | struct btrfs_space_info *meta_sinfo; | |
3162 | u64 num_bytes; | |
3163 | u64 used; | |
3164 | u64 alloc_target; | |
3165 | int retries = 0; | |
3166 | ||
3167 | /* get the space info for where the metadata will live */ | |
3168 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
3169 | meta_sinfo = __find_space_info(info, alloc_target); | |
3170 | ||
3171 | num_bytes = calculate_bytes_needed(root, num_items); | |
3172 | again: | |
3173 | spin_lock(&meta_sinfo->lock); | |
3174 | ||
3175 | if (unlikely(!meta_sinfo->bytes_root)) | |
3176 | meta_sinfo->bytes_root = calculate_bytes_needed(root, 6); | |
3177 | ||
3178 | if (!retries) | |
3179 | meta_sinfo->bytes_may_use += num_bytes; | |
3180 | ||
3181 | used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
3182 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly + | |
3183 | meta_sinfo->bytes_super + meta_sinfo->bytes_root + | |
3184 | meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc; | |
3185 | ||
3186 | if (used > meta_sinfo->total_bytes) { | |
3187 | retries++; | |
3188 | if (retries == 1) { | |
3189 | if (maybe_allocate_chunk(root, meta_sinfo)) | |
3190 | goto again; | |
3191 | retries++; | |
3192 | } else { | |
3193 | spin_unlock(&meta_sinfo->lock); | |
3194 | } | |
3195 | ||
3196 | if (retries == 2) { | |
32c00aff | 3197 | flush_delalloc(root, meta_sinfo); |
4e06bdd6 JB |
3198 | goto again; |
3199 | } | |
9ed74f2d JB |
3200 | spin_lock(&meta_sinfo->lock); |
3201 | meta_sinfo->bytes_may_use -= num_bytes; | |
3202 | spin_unlock(&meta_sinfo->lock); | |
3203 | ||
3204 | dump_space_info(meta_sinfo, 0, 0); | |
6a63209f JB |
3205 | return -ENOSPC; |
3206 | } | |
9ed74f2d JB |
3207 | |
3208 | check_force_delalloc(meta_sinfo); | |
6a63209f JB |
3209 | spin_unlock(&meta_sinfo->lock); |
3210 | ||
3211 | return 0; | |
3212 | } | |
3213 | ||
3214 | /* | |
3215 | * This will check the space that the inode allocates from to make sure we have | |
3216 | * enough space for bytes. | |
3217 | */ | |
3218 | int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, | |
3219 | u64 bytes) | |
3220 | { | |
3221 | struct btrfs_space_info *data_sinfo; | |
4e06bdd6 | 3222 | int ret = 0, committed = 0; |
6a63209f JB |
3223 | |
3224 | /* make sure bytes are sectorsize aligned */ | |
3225 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
3226 | ||
3227 | data_sinfo = BTRFS_I(inode)->space_info; | |
33b4d47f CM |
3228 | if (!data_sinfo) |
3229 | goto alloc; | |
3230 | ||
6a63209f JB |
3231 | again: |
3232 | /* make sure we have enough space to handle the data first */ | |
3233 | spin_lock(&data_sinfo->lock); | |
3234 | if (data_sinfo->total_bytes - data_sinfo->bytes_used - | |
3235 | data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - | |
3236 | data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - | |
1b2da372 | 3237 | data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) { |
4e06bdd6 JB |
3238 | struct btrfs_trans_handle *trans; |
3239 | ||
6a63209f JB |
3240 | /* |
3241 | * if we don't have enough free bytes in this space then we need | |
3242 | * to alloc a new chunk. | |
3243 | */ | |
3244 | if (!data_sinfo->full) { | |
3245 | u64 alloc_target; | |
6a63209f JB |
3246 | |
3247 | data_sinfo->force_alloc = 1; | |
3248 | spin_unlock(&data_sinfo->lock); | |
33b4d47f | 3249 | alloc: |
6a63209f JB |
3250 | alloc_target = btrfs_get_alloc_profile(root, 1); |
3251 | trans = btrfs_start_transaction(root, 1); | |
3252 | if (!trans) | |
3253 | return -ENOMEM; | |
3254 | ||
3255 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
3256 | bytes + 2 * 1024 * 1024, | |
3257 | alloc_target, 0); | |
3258 | btrfs_end_transaction(trans, root); | |
3259 | if (ret) | |
3260 | return ret; | |
33b4d47f CM |
3261 | |
3262 | if (!data_sinfo) { | |
3263 | btrfs_set_inode_space_info(root, inode); | |
3264 | data_sinfo = BTRFS_I(inode)->space_info; | |
3265 | } | |
6a63209f JB |
3266 | goto again; |
3267 | } | |
3268 | spin_unlock(&data_sinfo->lock); | |
4e06bdd6 JB |
3269 | |
3270 | /* commit the current transaction and try again */ | |
dd7e0b7b | 3271 | if (!committed && !root->fs_info->open_ioctl_trans) { |
4e06bdd6 JB |
3272 | committed = 1; |
3273 | trans = btrfs_join_transaction(root, 1); | |
3274 | if (!trans) | |
3275 | return -ENOMEM; | |
3276 | ret = btrfs_commit_transaction(trans, root); | |
3277 | if (ret) | |
3278 | return ret; | |
3279 | goto again; | |
3280 | } | |
3281 | ||
6a63209f JB |
3282 | printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" |
3283 | ", %llu bytes_used, %llu bytes_reserved, " | |
68f5a38c | 3284 | "%llu bytes_pinned, %llu bytes_readonly, %llu may use " |
21380931 JB |
3285 | "%llu total\n", (unsigned long long)bytes, |
3286 | (unsigned long long)data_sinfo->bytes_delalloc, | |
3287 | (unsigned long long)data_sinfo->bytes_used, | |
3288 | (unsigned long long)data_sinfo->bytes_reserved, | |
3289 | (unsigned long long)data_sinfo->bytes_pinned, | |
3290 | (unsigned long long)data_sinfo->bytes_readonly, | |
3291 | (unsigned long long)data_sinfo->bytes_may_use, | |
3292 | (unsigned long long)data_sinfo->total_bytes); | |
6a63209f JB |
3293 | return -ENOSPC; |
3294 | } | |
3295 | data_sinfo->bytes_may_use += bytes; | |
3296 | BTRFS_I(inode)->reserved_bytes += bytes; | |
3297 | spin_unlock(&data_sinfo->lock); | |
3298 | ||
9ed74f2d | 3299 | return 0; |
6a63209f JB |
3300 | } |
3301 | ||
3302 | /* | |
3303 | * if there was an error for whatever reason after calling | |
3304 | * btrfs_check_data_free_space, call this so we can cleanup the counters. | |
3305 | */ | |
3306 | void btrfs_free_reserved_data_space(struct btrfs_root *root, | |
3307 | struct inode *inode, u64 bytes) | |
3308 | { | |
3309 | struct btrfs_space_info *data_sinfo; | |
3310 | ||
3311 | /* make sure bytes are sectorsize aligned */ | |
3312 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
3313 | ||
3314 | data_sinfo = BTRFS_I(inode)->space_info; | |
3315 | spin_lock(&data_sinfo->lock); | |
3316 | data_sinfo->bytes_may_use -= bytes; | |
3317 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3318 | spin_unlock(&data_sinfo->lock); | |
3319 | } | |
3320 | ||
3321 | /* called when we are adding a delalloc extent to the inode's io_tree */ | |
3322 | void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, | |
3323 | u64 bytes) | |
3324 | { | |
3325 | struct btrfs_space_info *data_sinfo; | |
3326 | ||
3327 | /* get the space info for where this inode will be storing its data */ | |
3328 | data_sinfo = BTRFS_I(inode)->space_info; | |
3329 | ||
3330 | /* make sure we have enough space to handle the data first */ | |
3331 | spin_lock(&data_sinfo->lock); | |
3332 | data_sinfo->bytes_delalloc += bytes; | |
3333 | ||
3334 | /* | |
3335 | * we are adding a delalloc extent without calling | |
3336 | * btrfs_check_data_free_space first. This happens on a weird | |
3337 | * writepage condition, but shouldn't hurt our accounting | |
3338 | */ | |
3339 | if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { | |
3340 | data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; | |
3341 | BTRFS_I(inode)->reserved_bytes = 0; | |
3342 | } else { | |
3343 | data_sinfo->bytes_may_use -= bytes; | |
3344 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3345 | } | |
3346 | ||
3347 | spin_unlock(&data_sinfo->lock); | |
3348 | } | |
3349 | ||
3350 | /* called when we are clearing an delalloc extent from the inode's io_tree */ | |
3351 | void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, | |
3352 | u64 bytes) | |
3353 | { | |
3354 | struct btrfs_space_info *info; | |
3355 | ||
3356 | info = BTRFS_I(inode)->space_info; | |
3357 | ||
3358 | spin_lock(&info->lock); | |
3359 | info->bytes_delalloc -= bytes; | |
3360 | spin_unlock(&info->lock); | |
3361 | } | |
3362 | ||
97e728d4 JB |
3363 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
3364 | { | |
3365 | struct list_head *head = &info->space_info; | |
3366 | struct btrfs_space_info *found; | |
3367 | ||
3368 | rcu_read_lock(); | |
3369 | list_for_each_entry_rcu(found, head, list) { | |
3370 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
3371 | found->force_alloc = 1; | |
3372 | } | |
3373 | rcu_read_unlock(); | |
3374 | } | |
3375 | ||
6324fbf3 CM |
3376 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3377 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3378 | u64 flags, int force) |
6324fbf3 CM |
3379 | { |
3380 | struct btrfs_space_info *space_info; | |
97e728d4 | 3381 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6324fbf3 | 3382 | u64 thresh; |
c146afad YZ |
3383 | int ret = 0; |
3384 | ||
97e728d4 | 3385 | mutex_lock(&fs_info->chunk_mutex); |
6324fbf3 | 3386 | |
2b82032c | 3387 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3388 | |
6324fbf3 | 3389 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3390 | if (!space_info) { |
3391 | ret = update_space_info(extent_root->fs_info, flags, | |
3392 | 0, 0, &space_info); | |
3393 | BUG_ON(ret); | |
3394 | } | |
6324fbf3 CM |
3395 | BUG_ON(!space_info); |
3396 | ||
25179201 | 3397 | spin_lock(&space_info->lock); |
9ed74f2d | 3398 | if (space_info->force_alloc) |
0ef3e66b | 3399 | force = 1; |
25179201 JB |
3400 | if (space_info->full) { |
3401 | spin_unlock(&space_info->lock); | |
925baedd | 3402 | goto out; |
25179201 | 3403 | } |
6324fbf3 | 3404 | |
c146afad | 3405 | thresh = space_info->total_bytes - space_info->bytes_readonly; |
9ed74f2d | 3406 | thresh = div_factor(thresh, 8); |
0ef3e66b | 3407 | if (!force && |
e8569813 | 3408 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
3409 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
3410 | spin_unlock(&space_info->lock); | |
925baedd | 3411 | goto out; |
25179201 | 3412 | } |
25179201 JB |
3413 | spin_unlock(&space_info->lock); |
3414 | ||
97e728d4 JB |
3415 | /* |
3416 | * if we're doing a data chunk, go ahead and make sure that | |
3417 | * we keep a reasonable number of metadata chunks allocated in the | |
3418 | * FS as well. | |
3419 | */ | |
9ed74f2d | 3420 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3421 | fs_info->data_chunk_allocations++; |
3422 | if (!(fs_info->data_chunk_allocations % | |
3423 | fs_info->metadata_ratio)) | |
3424 | force_metadata_allocation(fs_info); | |
3425 | } | |
3426 | ||
2b82032c | 3427 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
9ed74f2d | 3428 | spin_lock(&space_info->lock); |
d397712b | 3429 | if (ret) |
6324fbf3 | 3430 | space_info->full = 1; |
9ed74f2d JB |
3431 | space_info->force_alloc = 0; |
3432 | spin_unlock(&space_info->lock); | |
a74a4b97 | 3433 | out: |
c146afad | 3434 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3435 | return ret; |
6324fbf3 CM |
3436 | } |
3437 | ||
9078a3e1 CM |
3438 | static int update_block_group(struct btrfs_trans_handle *trans, |
3439 | struct btrfs_root *root, | |
db94535d | 3440 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 3441 | int mark_free) |
9078a3e1 CM |
3442 | { |
3443 | struct btrfs_block_group_cache *cache; | |
3444 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 3445 | u64 total = num_bytes; |
9078a3e1 | 3446 | u64 old_val; |
db94535d | 3447 | u64 byte_in_group; |
3e1ad54f | 3448 | |
5d4f98a2 YZ |
3449 | /* block accounting for super block */ |
3450 | spin_lock(&info->delalloc_lock); | |
3451 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
3452 | if (alloc) | |
3453 | old_val += num_bytes; | |
3454 | else | |
3455 | old_val -= num_bytes; | |
3456 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
5d4f98a2 YZ |
3457 | spin_unlock(&info->delalloc_lock); |
3458 | ||
d397712b | 3459 | while (total) { |
db94535d | 3460 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 3461 | if (!cache) |
9078a3e1 | 3462 | return -1; |
db94535d CM |
3463 | byte_in_group = bytenr - cache->key.objectid; |
3464 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 3465 | |
25179201 | 3466 | spin_lock(&cache->space_info->lock); |
c286ac48 | 3467 | spin_lock(&cache->lock); |
0f9dd46c | 3468 | cache->dirty = 1; |
9078a3e1 | 3469 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 3470 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 3471 | if (alloc) { |
db94535d | 3472 | old_val += num_bytes; |
11833d66 YZ |
3473 | btrfs_set_block_group_used(&cache->item, old_val); |
3474 | cache->reserved -= num_bytes; | |
6324fbf3 | 3475 | cache->space_info->bytes_used += num_bytes; |
11833d66 | 3476 | cache->space_info->bytes_reserved -= num_bytes; |
a512bbf8 | 3477 | if (cache->ro) |
c146afad | 3478 | cache->space_info->bytes_readonly -= num_bytes; |
c286ac48 | 3479 | spin_unlock(&cache->lock); |
25179201 | 3480 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 3481 | } else { |
db94535d | 3482 | old_val -= num_bytes; |
6324fbf3 | 3483 | cache->space_info->bytes_used -= num_bytes; |
c146afad YZ |
3484 | if (cache->ro) |
3485 | cache->space_info->bytes_readonly += num_bytes; | |
c286ac48 CM |
3486 | btrfs_set_block_group_used(&cache->item, old_val); |
3487 | spin_unlock(&cache->lock); | |
25179201 | 3488 | spin_unlock(&cache->space_info->lock); |
f510cfec | 3489 | if (mark_free) { |
0f9dd46c | 3490 | int ret; |
1f3c79a2 LH |
3491 | |
3492 | ret = btrfs_discard_extent(root, bytenr, | |
3493 | num_bytes); | |
3494 | WARN_ON(ret); | |
3495 | ||
0f9dd46c JB |
3496 | ret = btrfs_add_free_space(cache, bytenr, |
3497 | num_bytes); | |
d2fb3437 | 3498 | WARN_ON(ret); |
e37c9e69 | 3499 | } |
cd1bc465 | 3500 | } |
fa9c0d79 | 3501 | btrfs_put_block_group(cache); |
db94535d CM |
3502 | total -= num_bytes; |
3503 | bytenr += num_bytes; | |
9078a3e1 CM |
3504 | } |
3505 | return 0; | |
3506 | } | |
6324fbf3 | 3507 | |
a061fc8d CM |
3508 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
3509 | { | |
0f9dd46c | 3510 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 3511 | u64 bytenr; |
0f9dd46c JB |
3512 | |
3513 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
3514 | if (!cache) | |
a061fc8d | 3515 | return 0; |
0f9dd46c | 3516 | |
d2fb3437 | 3517 | bytenr = cache->key.objectid; |
fa9c0d79 | 3518 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
3519 | |
3520 | return bytenr; | |
a061fc8d CM |
3521 | } |
3522 | ||
11833d66 YZ |
3523 | /* |
3524 | * this function must be called within transaction | |
3525 | */ | |
3526 | int btrfs_pin_extent(struct btrfs_root *root, | |
3527 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 3528 | { |
324ae4df | 3529 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 | 3530 | struct btrfs_block_group_cache *cache; |
324ae4df | 3531 | |
11833d66 YZ |
3532 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
3533 | BUG_ON(!cache); | |
68b38550 | 3534 | |
11833d66 YZ |
3535 | spin_lock(&cache->space_info->lock); |
3536 | spin_lock(&cache->lock); | |
3537 | cache->pinned += num_bytes; | |
3538 | cache->space_info->bytes_pinned += num_bytes; | |
3539 | if (reserved) { | |
3540 | cache->reserved -= num_bytes; | |
3541 | cache->space_info->bytes_reserved -= num_bytes; | |
3542 | } | |
3543 | spin_unlock(&cache->lock); | |
3544 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 3545 | |
11833d66 | 3546 | btrfs_put_block_group(cache); |
68b38550 | 3547 | |
11833d66 YZ |
3548 | set_extent_dirty(fs_info->pinned_extents, |
3549 | bytenr, bytenr + num_bytes - 1, GFP_NOFS); | |
3550 | return 0; | |
3551 | } | |
3552 | ||
3553 | static int update_reserved_extents(struct btrfs_block_group_cache *cache, | |
3554 | u64 num_bytes, int reserve) | |
3555 | { | |
3556 | spin_lock(&cache->space_info->lock); | |
3557 | spin_lock(&cache->lock); | |
3558 | if (reserve) { | |
3559 | cache->reserved += num_bytes; | |
3560 | cache->space_info->bytes_reserved += num_bytes; | |
3561 | } else { | |
3562 | cache->reserved -= num_bytes; | |
3563 | cache->space_info->bytes_reserved -= num_bytes; | |
324ae4df | 3564 | } |
11833d66 YZ |
3565 | spin_unlock(&cache->lock); |
3566 | spin_unlock(&cache->space_info->lock); | |
324ae4df Y |
3567 | return 0; |
3568 | } | |
9078a3e1 | 3569 | |
11833d66 YZ |
3570 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
3571 | struct btrfs_root *root) | |
e8569813 | 3572 | { |
e8569813 | 3573 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
3574 | struct btrfs_caching_control *next; |
3575 | struct btrfs_caching_control *caching_ctl; | |
3576 | struct btrfs_block_group_cache *cache; | |
e8569813 | 3577 | |
11833d66 | 3578 | down_write(&fs_info->extent_commit_sem); |
25179201 | 3579 | |
11833d66 YZ |
3580 | list_for_each_entry_safe(caching_ctl, next, |
3581 | &fs_info->caching_block_groups, list) { | |
3582 | cache = caching_ctl->block_group; | |
3583 | if (block_group_cache_done(cache)) { | |
3584 | cache->last_byte_to_unpin = (u64)-1; | |
3585 | list_del_init(&caching_ctl->list); | |
3586 | put_caching_control(caching_ctl); | |
e8569813 | 3587 | } else { |
11833d66 | 3588 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 3589 | } |
e8569813 | 3590 | } |
11833d66 YZ |
3591 | |
3592 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
3593 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
3594 | else | |
3595 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
3596 | ||
3597 | up_write(&fs_info->extent_commit_sem); | |
e8569813 ZY |
3598 | return 0; |
3599 | } | |
3600 | ||
11833d66 | 3601 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 3602 | { |
11833d66 YZ |
3603 | struct btrfs_fs_info *fs_info = root->fs_info; |
3604 | struct btrfs_block_group_cache *cache = NULL; | |
3605 | u64 len; | |
ccd467d6 | 3606 | |
11833d66 YZ |
3607 | while (start <= end) { |
3608 | if (!cache || | |
3609 | start >= cache->key.objectid + cache->key.offset) { | |
3610 | if (cache) | |
3611 | btrfs_put_block_group(cache); | |
3612 | cache = btrfs_lookup_block_group(fs_info, start); | |
3613 | BUG_ON(!cache); | |
3614 | } | |
3615 | ||
3616 | len = cache->key.objectid + cache->key.offset - start; | |
3617 | len = min(len, end + 1 - start); | |
3618 | ||
3619 | if (start < cache->last_byte_to_unpin) { | |
3620 | len = min(len, cache->last_byte_to_unpin - start); | |
3621 | btrfs_add_free_space(cache, start, len); | |
3622 | } | |
3623 | ||
3624 | spin_lock(&cache->space_info->lock); | |
3625 | spin_lock(&cache->lock); | |
3626 | cache->pinned -= len; | |
3627 | cache->space_info->bytes_pinned -= len; | |
3628 | spin_unlock(&cache->lock); | |
3629 | spin_unlock(&cache->space_info->lock); | |
817d52f8 | 3630 | |
11833d66 | 3631 | start += len; |
ccd467d6 | 3632 | } |
11833d66 YZ |
3633 | |
3634 | if (cache) | |
3635 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
3636 | return 0; |
3637 | } | |
3638 | ||
3639 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 3640 | struct btrfs_root *root) |
a28ec197 | 3641 | { |
11833d66 YZ |
3642 | struct btrfs_fs_info *fs_info = root->fs_info; |
3643 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
3644 | u64 start; |
3645 | u64 end; | |
a28ec197 | 3646 | int ret; |
a28ec197 | 3647 | |
11833d66 YZ |
3648 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
3649 | unpin = &fs_info->freed_extents[1]; | |
3650 | else | |
3651 | unpin = &fs_info->freed_extents[0]; | |
3652 | ||
d397712b | 3653 | while (1) { |
1a5bc167 CM |
3654 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
3655 | EXTENT_DIRTY); | |
3656 | if (ret) | |
a28ec197 | 3657 | break; |
1f3c79a2 LH |
3658 | |
3659 | ret = btrfs_discard_extent(root, start, end + 1 - start); | |
3660 | ||
1a5bc167 | 3661 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 3662 | unpin_extent_range(root, start, end); |
b9473439 | 3663 | cond_resched(); |
a28ec197 | 3664 | } |
817d52f8 | 3665 | |
1f3c79a2 | 3666 | return ret; |
a28ec197 CM |
3667 | } |
3668 | ||
31840ae1 ZY |
3669 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
3670 | struct btrfs_root *root, | |
b9473439 | 3671 | struct btrfs_path *path, |
11833d66 YZ |
3672 | u64 bytenr, u64 num_bytes, |
3673 | int is_data, int reserved, | |
b9473439 | 3674 | struct extent_buffer **must_clean) |
e20d96d6 | 3675 | { |
1a5bc167 | 3676 | int err = 0; |
31840ae1 | 3677 | struct extent_buffer *buf; |
8ef97622 | 3678 | |
31840ae1 ZY |
3679 | if (is_data) |
3680 | goto pinit; | |
3681 | ||
444528b3 CM |
3682 | /* |
3683 | * discard is sloooow, and so triggering discards on | |
3684 | * individual btree blocks isn't a good plan. Just | |
3685 | * pin everything in discard mode. | |
3686 | */ | |
3687 | if (btrfs_test_opt(root, DISCARD)) | |
3688 | goto pinit; | |
3689 | ||
31840ae1 ZY |
3690 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
3691 | if (!buf) | |
3692 | goto pinit; | |
3693 | ||
3694 | /* we can reuse a block if it hasn't been written | |
3695 | * and it is from this transaction. We can't | |
3696 | * reuse anything from the tree log root because | |
3697 | * it has tiny sub-transactions. | |
3698 | */ | |
3699 | if (btrfs_buffer_uptodate(buf, 0) && | |
3700 | btrfs_try_tree_lock(buf)) { | |
3701 | u64 header_owner = btrfs_header_owner(buf); | |
3702 | u64 header_transid = btrfs_header_generation(buf); | |
3703 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
3704 | header_transid == trans->transid && | |
3705 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
b9473439 | 3706 | *must_clean = buf; |
31840ae1 | 3707 | return 1; |
8ef97622 | 3708 | } |
31840ae1 | 3709 | btrfs_tree_unlock(buf); |
f4b9aa8d | 3710 | } |
31840ae1 ZY |
3711 | free_extent_buffer(buf); |
3712 | pinit: | |
11833d66 YZ |
3713 | if (path) |
3714 | btrfs_set_path_blocking(path); | |
b9473439 | 3715 | /* unlocks the pinned mutex */ |
11833d66 | 3716 | btrfs_pin_extent(root, bytenr, num_bytes, reserved); |
31840ae1 | 3717 | |
be744175 | 3718 | BUG_ON(err < 0); |
e20d96d6 CM |
3719 | return 0; |
3720 | } | |
3721 | ||
5d4f98a2 YZ |
3722 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
3723 | struct btrfs_root *root, | |
3724 | u64 bytenr, u64 num_bytes, u64 parent, | |
3725 | u64 root_objectid, u64 owner_objectid, | |
3726 | u64 owner_offset, int refs_to_drop, | |
3727 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 3728 | { |
e2fa7227 | 3729 | struct btrfs_key key; |
5d4f98a2 | 3730 | struct btrfs_path *path; |
1261ec42 CM |
3731 | struct btrfs_fs_info *info = root->fs_info; |
3732 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 3733 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3734 | struct btrfs_extent_item *ei; |
3735 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3736 | int ret; |
5d4f98a2 | 3737 | int is_data; |
952fccac CM |
3738 | int extent_slot = 0; |
3739 | int found_extent = 0; | |
3740 | int num_to_del = 1; | |
5d4f98a2 YZ |
3741 | u32 item_size; |
3742 | u64 refs; | |
037e6390 | 3743 | |
5caf2a00 | 3744 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3745 | if (!path) |
3746 | return -ENOMEM; | |
5f26f772 | 3747 | |
3c12ac72 | 3748 | path->reada = 1; |
b9473439 | 3749 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3750 | |
3751 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
3752 | BUG_ON(!is_data && refs_to_drop != 1); | |
3753 | ||
3754 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
3755 | bytenr, num_bytes, parent, | |
3756 | root_objectid, owner_objectid, | |
3757 | owner_offset); | |
7bb86316 | 3758 | if (ret == 0) { |
952fccac | 3759 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3760 | while (extent_slot >= 0) { |
3761 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3762 | extent_slot); |
5d4f98a2 | 3763 | if (key.objectid != bytenr) |
952fccac | 3764 | break; |
5d4f98a2 YZ |
3765 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3766 | key.offset == num_bytes) { | |
952fccac CM |
3767 | found_extent = 1; |
3768 | break; | |
3769 | } | |
3770 | if (path->slots[0] - extent_slot > 5) | |
3771 | break; | |
5d4f98a2 | 3772 | extent_slot--; |
952fccac | 3773 | } |
5d4f98a2 YZ |
3774 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
3775 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
3776 | if (found_extent && item_size < sizeof(*ei)) | |
3777 | found_extent = 0; | |
3778 | #endif | |
31840ae1 | 3779 | if (!found_extent) { |
5d4f98a2 | 3780 | BUG_ON(iref); |
56bec294 | 3781 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
3782 | NULL, refs_to_drop, |
3783 | is_data); | |
31840ae1 ZY |
3784 | BUG_ON(ret); |
3785 | btrfs_release_path(extent_root, path); | |
b9473439 | 3786 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3787 | |
3788 | key.objectid = bytenr; | |
3789 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3790 | key.offset = num_bytes; | |
3791 | ||
31840ae1 ZY |
3792 | ret = btrfs_search_slot(trans, extent_root, |
3793 | &key, path, -1, 1); | |
f3465ca4 JB |
3794 | if (ret) { |
3795 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
3796 | ", was looking for %llu\n", ret, |
3797 | (unsigned long long)bytenr); | |
f3465ca4 JB |
3798 | btrfs_print_leaf(extent_root, path->nodes[0]); |
3799 | } | |
31840ae1 ZY |
3800 | BUG_ON(ret); |
3801 | extent_slot = path->slots[0]; | |
3802 | } | |
7bb86316 CM |
3803 | } else { |
3804 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3805 | WARN_ON(1); | |
d397712b | 3806 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 3807 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 3808 | (unsigned long long)bytenr, |
56bec294 | 3809 | (unsigned long long)parent, |
d397712b | 3810 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
3811 | (unsigned long long)owner_objectid, |
3812 | (unsigned long long)owner_offset); | |
7bb86316 | 3813 | } |
5f39d397 CM |
3814 | |
3815 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
3816 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
3817 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3818 | if (item_size < sizeof(*ei)) { | |
3819 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
3820 | ret = convert_extent_item_v0(trans, extent_root, path, | |
3821 | owner_objectid, 0); | |
3822 | BUG_ON(ret < 0); | |
3823 | ||
3824 | btrfs_release_path(extent_root, path); | |
3825 | path->leave_spinning = 1; | |
3826 | ||
3827 | key.objectid = bytenr; | |
3828 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3829 | key.offset = num_bytes; | |
3830 | ||
3831 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
3832 | -1, 1); | |
3833 | if (ret) { | |
3834 | printk(KERN_ERR "umm, got %d back from search" | |
3835 | ", was looking for %llu\n", ret, | |
3836 | (unsigned long long)bytenr); | |
3837 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3838 | } | |
3839 | BUG_ON(ret); | |
3840 | extent_slot = path->slots[0]; | |
3841 | leaf = path->nodes[0]; | |
3842 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
3843 | } | |
3844 | #endif | |
3845 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 3846 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3847 | struct btrfs_extent_item); |
5d4f98a2 YZ |
3848 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
3849 | struct btrfs_tree_block_info *bi; | |
3850 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3851 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3852 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3853 | } | |
56bec294 | 3854 | |
5d4f98a2 | 3855 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
3856 | BUG_ON(refs < refs_to_drop); |
3857 | refs -= refs_to_drop; | |
5f39d397 | 3858 | |
5d4f98a2 YZ |
3859 | if (refs > 0) { |
3860 | if (extent_op) | |
3861 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3862 | /* | |
3863 | * In the case of inline back ref, reference count will | |
3864 | * be updated by remove_extent_backref | |
952fccac | 3865 | */ |
5d4f98a2 YZ |
3866 | if (iref) { |
3867 | BUG_ON(!found_extent); | |
3868 | } else { | |
3869 | btrfs_set_extent_refs(leaf, ei, refs); | |
3870 | btrfs_mark_buffer_dirty(leaf); | |
3871 | } | |
3872 | if (found_extent) { | |
3873 | ret = remove_extent_backref(trans, extent_root, path, | |
3874 | iref, refs_to_drop, | |
3875 | is_data); | |
952fccac CM |
3876 | BUG_ON(ret); |
3877 | } | |
5d4f98a2 YZ |
3878 | } else { |
3879 | int mark_free = 0; | |
b9473439 | 3880 | struct extent_buffer *must_clean = NULL; |
78fae27e | 3881 | |
5d4f98a2 YZ |
3882 | if (found_extent) { |
3883 | BUG_ON(is_data && refs_to_drop != | |
3884 | extent_data_ref_count(root, path, iref)); | |
3885 | if (iref) { | |
3886 | BUG_ON(path->slots[0] != extent_slot); | |
3887 | } else { | |
3888 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3889 | path->slots[0] = extent_slot; | |
3890 | num_to_del = 2; | |
3891 | } | |
78fae27e | 3892 | } |
b9473439 | 3893 | |
5d4f98a2 | 3894 | ret = pin_down_bytes(trans, root, path, bytenr, |
11833d66 | 3895 | num_bytes, is_data, 0, &must_clean); |
5d4f98a2 YZ |
3896 | if (ret > 0) |
3897 | mark_free = 1; | |
3898 | BUG_ON(ret < 0); | |
b9473439 CM |
3899 | /* |
3900 | * it is going to be very rare for someone to be waiting | |
3901 | * on the block we're freeing. del_items might need to | |
3902 | * schedule, so rather than get fancy, just force it | |
3903 | * to blocking here | |
3904 | */ | |
3905 | if (must_clean) | |
3906 | btrfs_set_lock_blocking(must_clean); | |
3907 | ||
952fccac CM |
3908 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3909 | num_to_del); | |
31840ae1 | 3910 | BUG_ON(ret); |
25179201 | 3911 | btrfs_release_path(extent_root, path); |
21af804c | 3912 | |
b9473439 CM |
3913 | if (must_clean) { |
3914 | clean_tree_block(NULL, root, must_clean); | |
3915 | btrfs_tree_unlock(must_clean); | |
3916 | free_extent_buffer(must_clean); | |
3917 | } | |
3918 | ||
5d4f98a2 | 3919 | if (is_data) { |
459931ec CM |
3920 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
3921 | BUG_ON(ret); | |
d57e62b8 CM |
3922 | } else { |
3923 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
3924 | bytenr >> PAGE_CACHE_SHIFT, | |
3925 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
3926 | } |
3927 | ||
dcbdd4dc CM |
3928 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
3929 | mark_free); | |
3930 | BUG_ON(ret); | |
a28ec197 | 3931 | } |
5caf2a00 | 3932 | btrfs_free_path(path); |
a28ec197 CM |
3933 | return ret; |
3934 | } | |
3935 | ||
1887be66 CM |
3936 | /* |
3937 | * when we free an extent, it is possible (and likely) that we free the last | |
3938 | * delayed ref for that extent as well. This searches the delayed ref tree for | |
3939 | * a given extent, and if there are no other delayed refs to be processed, it | |
3940 | * removes it from the tree. | |
3941 | */ | |
3942 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
3943 | struct btrfs_root *root, u64 bytenr) | |
3944 | { | |
3945 | struct btrfs_delayed_ref_head *head; | |
3946 | struct btrfs_delayed_ref_root *delayed_refs; | |
3947 | struct btrfs_delayed_ref_node *ref; | |
3948 | struct rb_node *node; | |
3949 | int ret; | |
3950 | ||
3951 | delayed_refs = &trans->transaction->delayed_refs; | |
3952 | spin_lock(&delayed_refs->lock); | |
3953 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
3954 | if (!head) | |
3955 | goto out; | |
3956 | ||
3957 | node = rb_prev(&head->node.rb_node); | |
3958 | if (!node) | |
3959 | goto out; | |
3960 | ||
3961 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
3962 | ||
3963 | /* there are still entries for this ref, we can't drop it */ | |
3964 | if (ref->bytenr == bytenr) | |
3965 | goto out; | |
3966 | ||
5d4f98a2 YZ |
3967 | if (head->extent_op) { |
3968 | if (!head->must_insert_reserved) | |
3969 | goto out; | |
3970 | kfree(head->extent_op); | |
3971 | head->extent_op = NULL; | |
3972 | } | |
3973 | ||
1887be66 CM |
3974 | /* |
3975 | * waiting for the lock here would deadlock. If someone else has it | |
3976 | * locked they are already in the process of dropping it anyway | |
3977 | */ | |
3978 | if (!mutex_trylock(&head->mutex)) | |
3979 | goto out; | |
3980 | ||
3981 | /* | |
3982 | * at this point we have a head with no other entries. Go | |
3983 | * ahead and process it. | |
3984 | */ | |
3985 | head->node.in_tree = 0; | |
3986 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 3987 | |
1887be66 CM |
3988 | delayed_refs->num_entries--; |
3989 | ||
3990 | /* | |
3991 | * we don't take a ref on the node because we're removing it from the | |
3992 | * tree, so we just steal the ref the tree was holding. | |
3993 | */ | |
c3e69d58 CM |
3994 | delayed_refs->num_heads--; |
3995 | if (list_empty(&head->cluster)) | |
3996 | delayed_refs->num_heads_ready--; | |
3997 | ||
3998 | list_del_init(&head->cluster); | |
1887be66 CM |
3999 | spin_unlock(&delayed_refs->lock); |
4000 | ||
4001 | ret = run_one_delayed_ref(trans, root->fs_info->tree_root, | |
5d4f98a2 YZ |
4002 | &head->node, head->extent_op, |
4003 | head->must_insert_reserved); | |
1887be66 CM |
4004 | BUG_ON(ret); |
4005 | btrfs_put_delayed_ref(&head->node); | |
4006 | return 0; | |
4007 | out: | |
4008 | spin_unlock(&delayed_refs->lock); | |
4009 | return 0; | |
4010 | } | |
4011 | ||
925baedd | 4012 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
4013 | struct btrfs_root *root, |
4014 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 4015 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
4016 | { |
4017 | int ret; | |
4018 | ||
56bec294 CM |
4019 | /* |
4020 | * tree log blocks never actually go into the extent allocation | |
4021 | * tree, just update pinning info and exit early. | |
56bec294 | 4022 | */ |
5d4f98a2 YZ |
4023 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
4024 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 4025 | /* unlocks the pinned mutex */ |
11833d66 | 4026 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 4027 | ret = 0; |
5d4f98a2 YZ |
4028 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
4029 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
4030 | parent, root_objectid, (int)owner, | |
4031 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 CM |
4032 | BUG_ON(ret); |
4033 | ret = check_ref_cleanup(trans, root, bytenr); | |
4034 | BUG_ON(ret); | |
5d4f98a2 YZ |
4035 | } else { |
4036 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
4037 | parent, root_objectid, owner, | |
4038 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
4039 | BUG_ON(ret); | |
56bec294 | 4040 | } |
925baedd CM |
4041 | return ret; |
4042 | } | |
4043 | ||
86b9f2ec YZ |
4044 | int btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4045 | struct btrfs_root *root, | |
4046 | u64 bytenr, u32 blocksize, | |
4047 | u64 parent, u64 root_objectid, int level) | |
4048 | { | |
4049 | u64 used; | |
4050 | spin_lock(&root->node_lock); | |
4051 | used = btrfs_root_used(&root->root_item) - blocksize; | |
4052 | btrfs_set_root_used(&root->root_item, used); | |
4053 | spin_unlock(&root->node_lock); | |
4054 | ||
4055 | return btrfs_free_extent(trans, root, bytenr, blocksize, | |
4056 | parent, root_objectid, level, 0); | |
4057 | } | |
4058 | ||
87ee04eb CM |
4059 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
4060 | { | |
4061 | u64 mask = ((u64)root->stripesize - 1); | |
4062 | u64 ret = (val + mask) & ~mask; | |
4063 | return ret; | |
4064 | } | |
4065 | ||
817d52f8 JB |
4066 | /* |
4067 | * when we wait for progress in the block group caching, its because | |
4068 | * our allocation attempt failed at least once. So, we must sleep | |
4069 | * and let some progress happen before we try again. | |
4070 | * | |
4071 | * This function will sleep at least once waiting for new free space to | |
4072 | * show up, and then it will check the block group free space numbers | |
4073 | * for our min num_bytes. Another option is to have it go ahead | |
4074 | * and look in the rbtree for a free extent of a given size, but this | |
4075 | * is a good start. | |
4076 | */ | |
4077 | static noinline int | |
4078 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
4079 | u64 num_bytes) | |
4080 | { | |
11833d66 | 4081 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
4082 | DEFINE_WAIT(wait); |
4083 | ||
11833d66 YZ |
4084 | caching_ctl = get_caching_control(cache); |
4085 | if (!caching_ctl) | |
817d52f8 | 4086 | return 0; |
817d52f8 | 4087 | |
11833d66 | 4088 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
817d52f8 | 4089 | (cache->free_space >= num_bytes)); |
11833d66 YZ |
4090 | |
4091 | put_caching_control(caching_ctl); | |
4092 | return 0; | |
4093 | } | |
4094 | ||
4095 | static noinline int | |
4096 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
4097 | { | |
4098 | struct btrfs_caching_control *caching_ctl; | |
4099 | DEFINE_WAIT(wait); | |
4100 | ||
4101 | caching_ctl = get_caching_control(cache); | |
4102 | if (!caching_ctl) | |
4103 | return 0; | |
4104 | ||
4105 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
4106 | ||
4107 | put_caching_control(caching_ctl); | |
817d52f8 JB |
4108 | return 0; |
4109 | } | |
4110 | ||
4111 | enum btrfs_loop_type { | |
ccf0e725 | 4112 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
4113 | LOOP_CACHING_NOWAIT = 1, |
4114 | LOOP_CACHING_WAIT = 2, | |
4115 | LOOP_ALLOC_CHUNK = 3, | |
4116 | LOOP_NO_EMPTY_SIZE = 4, | |
4117 | }; | |
4118 | ||
fec577fb CM |
4119 | /* |
4120 | * walks the btree of allocated extents and find a hole of a given size. | |
4121 | * The key ins is changed to record the hole: | |
4122 | * ins->objectid == block start | |
62e2749e | 4123 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
4124 | * ins->offset == number of blocks |
4125 | * Any available blocks before search_start are skipped. | |
4126 | */ | |
d397712b | 4127 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4128 | struct btrfs_root *orig_root, |
4129 | u64 num_bytes, u64 empty_size, | |
4130 | u64 search_start, u64 search_end, | |
4131 | u64 hint_byte, struct btrfs_key *ins, | |
4132 | u64 exclude_start, u64 exclude_nr, | |
4133 | int data) | |
fec577fb | 4134 | { |
80eb234a | 4135 | int ret = 0; |
d397712b | 4136 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 4137 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 4138 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 4139 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 4140 | int allowed_chunk_alloc = 0; |
ccf0e725 | 4141 | int done_chunk_alloc = 0; |
80eb234a | 4142 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
4143 | int last_ptr_loop = 0; |
4144 | int loop = 0; | |
817d52f8 | 4145 | bool found_uncached_bg = false; |
0a24325e | 4146 | bool failed_cluster_refill = false; |
1cdda9b8 | 4147 | bool failed_alloc = false; |
ccf0e725 JB |
4148 | u64 ideal_cache_percent = 0; |
4149 | u64 ideal_cache_offset = 0; | |
fec577fb | 4150 | |
db94535d | 4151 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 4152 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
4153 | ins->objectid = 0; |
4154 | ins->offset = 0; | |
b1a4d965 | 4155 | |
2552d17e JB |
4156 | space_info = __find_space_info(root->fs_info, data); |
4157 | ||
0ef3e66b CM |
4158 | if (orig_root->ref_cows || empty_size) |
4159 | allowed_chunk_alloc = 1; | |
4160 | ||
239b14b3 | 4161 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
fa9c0d79 | 4162 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
4163 | if (!btrfs_test_opt(root, SSD)) |
4164 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
4165 | } |
4166 | ||
fa9c0d79 CM |
4167 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { |
4168 | last_ptr = &root->fs_info->data_alloc_cluster; | |
4169 | } | |
0f9dd46c | 4170 | |
239b14b3 | 4171 | if (last_ptr) { |
fa9c0d79 CM |
4172 | spin_lock(&last_ptr->lock); |
4173 | if (last_ptr->block_group) | |
4174 | hint_byte = last_ptr->window_start; | |
4175 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 4176 | } |
fa9c0d79 | 4177 | |
a061fc8d | 4178 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 4179 | search_start = max(search_start, hint_byte); |
0b86a832 | 4180 | |
817d52f8 | 4181 | if (!last_ptr) |
fa9c0d79 | 4182 | empty_cluster = 0; |
fa9c0d79 | 4183 | |
2552d17e | 4184 | if (search_start == hint_byte) { |
ccf0e725 | 4185 | ideal_cache: |
2552d17e JB |
4186 | block_group = btrfs_lookup_block_group(root->fs_info, |
4187 | search_start); | |
817d52f8 JB |
4188 | /* |
4189 | * we don't want to use the block group if it doesn't match our | |
4190 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4191 | * |
4192 | * However if we are re-searching with an ideal block group | |
4193 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
4194 | */ |
4195 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
4196 | (block_group->cached != BTRFS_CACHE_NO || |
4197 | search_start == ideal_cache_offset)) { | |
2552d17e | 4198 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4199 | if (list_empty(&block_group->list) || |
4200 | block_group->ro) { | |
4201 | /* | |
4202 | * someone is removing this block group, | |
4203 | * we can't jump into the have_block_group | |
4204 | * target because our list pointers are not | |
4205 | * valid | |
4206 | */ | |
4207 | btrfs_put_block_group(block_group); | |
4208 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4209 | } else { |
44fb5511 | 4210 | goto have_block_group; |
ccf0e725 | 4211 | } |
2552d17e | 4212 | } else if (block_group) { |
fa9c0d79 | 4213 | btrfs_put_block_group(block_group); |
2552d17e | 4214 | } |
42e70e7a | 4215 | } |
2552d17e | 4216 | search: |
80eb234a | 4217 | down_read(&space_info->groups_sem); |
2552d17e | 4218 | list_for_each_entry(block_group, &space_info->block_groups, list) { |
6226cb0a | 4219 | u64 offset; |
817d52f8 | 4220 | int cached; |
8a1413a2 | 4221 | |
2552d17e JB |
4222 | atomic_inc(&block_group->count); |
4223 | search_start = block_group->key.objectid; | |
42e70e7a | 4224 | |
2552d17e | 4225 | have_block_group: |
817d52f8 | 4226 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
ccf0e725 JB |
4227 | u64 free_percent; |
4228 | ||
4229 | free_percent = btrfs_block_group_used(&block_group->item); | |
4230 | free_percent *= 100; | |
4231 | free_percent = div64_u64(free_percent, | |
4232 | block_group->key.offset); | |
4233 | free_percent = 100 - free_percent; | |
4234 | if (free_percent > ideal_cache_percent && | |
4235 | likely(!block_group->ro)) { | |
4236 | ideal_cache_offset = block_group->key.objectid; | |
4237 | ideal_cache_percent = free_percent; | |
4238 | } | |
4239 | ||
817d52f8 | 4240 | /* |
ccf0e725 JB |
4241 | * We only want to start kthread caching if we are at |
4242 | * the point where we will wait for caching to make | |
4243 | * progress, or if our ideal search is over and we've | |
4244 | * found somebody to start caching. | |
817d52f8 JB |
4245 | */ |
4246 | if (loop > LOOP_CACHING_NOWAIT || | |
ccf0e725 JB |
4247 | (loop > LOOP_FIND_IDEAL && |
4248 | atomic_read(&space_info->caching_threads) < 2)) { | |
817d52f8 JB |
4249 | ret = cache_block_group(block_group); |
4250 | BUG_ON(ret); | |
2552d17e | 4251 | } |
817d52f8 JB |
4252 | found_uncached_bg = true; |
4253 | ||
ccf0e725 JB |
4254 | /* |
4255 | * If loop is set for cached only, try the next block | |
4256 | * group. | |
4257 | */ | |
4258 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
4259 | goto loop; |
4260 | } | |
4261 | ||
ccf0e725 JB |
4262 | cached = block_group_cache_done(block_group); |
4263 | if (unlikely(!cached)) | |
4264 | found_uncached_bg = true; | |
4265 | ||
ea6a478e | 4266 | if (unlikely(block_group->ro)) |
2552d17e | 4267 | goto loop; |
0f9dd46c | 4268 | |
0a24325e JB |
4269 | /* |
4270 | * Ok we want to try and use the cluster allocator, so lets look | |
4271 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
4272 | * have tried the cluster allocator plenty of times at this | |
4273 | * point and not have found anything, so we are likely way too | |
4274 | * fragmented for the clustering stuff to find anything, so lets | |
4275 | * just skip it and let the allocator find whatever block it can | |
4276 | * find | |
4277 | */ | |
4278 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
4279 | /* |
4280 | * the refill lock keeps out other | |
4281 | * people trying to start a new cluster | |
4282 | */ | |
4283 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
4284 | if (last_ptr->block_group && |
4285 | (last_ptr->block_group->ro || | |
4286 | !block_group_bits(last_ptr->block_group, data))) { | |
4287 | offset = 0; | |
4288 | goto refill_cluster; | |
4289 | } | |
4290 | ||
fa9c0d79 CM |
4291 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
4292 | num_bytes, search_start); | |
4293 | if (offset) { | |
4294 | /* we have a block, we're done */ | |
4295 | spin_unlock(&last_ptr->refill_lock); | |
4296 | goto checks; | |
4297 | } | |
4298 | ||
4299 | spin_lock(&last_ptr->lock); | |
4300 | /* | |
4301 | * whoops, this cluster doesn't actually point to | |
4302 | * this block group. Get a ref on the block | |
4303 | * group is does point to and try again | |
4304 | */ | |
4305 | if (!last_ptr_loop && last_ptr->block_group && | |
4306 | last_ptr->block_group != block_group) { | |
4307 | ||
4308 | btrfs_put_block_group(block_group); | |
4309 | block_group = last_ptr->block_group; | |
4310 | atomic_inc(&block_group->count); | |
4311 | spin_unlock(&last_ptr->lock); | |
4312 | spin_unlock(&last_ptr->refill_lock); | |
4313 | ||
4314 | last_ptr_loop = 1; | |
4315 | search_start = block_group->key.objectid; | |
44fb5511 CM |
4316 | /* |
4317 | * we know this block group is properly | |
4318 | * in the list because | |
4319 | * btrfs_remove_block_group, drops the | |
4320 | * cluster before it removes the block | |
4321 | * group from the list | |
4322 | */ | |
fa9c0d79 CM |
4323 | goto have_block_group; |
4324 | } | |
4325 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 4326 | refill_cluster: |
fa9c0d79 CM |
4327 | /* |
4328 | * this cluster didn't work out, free it and | |
4329 | * start over | |
4330 | */ | |
4331 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
4332 | ||
4333 | last_ptr_loop = 0; | |
4334 | ||
4335 | /* allocate a cluster in this block group */ | |
451d7585 | 4336 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
4337 | block_group, last_ptr, |
4338 | offset, num_bytes, | |
4339 | empty_cluster + empty_size); | |
4340 | if (ret == 0) { | |
4341 | /* | |
4342 | * now pull our allocation out of this | |
4343 | * cluster | |
4344 | */ | |
4345 | offset = btrfs_alloc_from_cluster(block_group, | |
4346 | last_ptr, num_bytes, | |
4347 | search_start); | |
4348 | if (offset) { | |
4349 | /* we found one, proceed */ | |
4350 | spin_unlock(&last_ptr->refill_lock); | |
4351 | goto checks; | |
4352 | } | |
0a24325e JB |
4353 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
4354 | && !failed_cluster_refill) { | |
817d52f8 JB |
4355 | spin_unlock(&last_ptr->refill_lock); |
4356 | ||
0a24325e | 4357 | failed_cluster_refill = true; |
817d52f8 JB |
4358 | wait_block_group_cache_progress(block_group, |
4359 | num_bytes + empty_cluster + empty_size); | |
4360 | goto have_block_group; | |
fa9c0d79 | 4361 | } |
817d52f8 | 4362 | |
fa9c0d79 CM |
4363 | /* |
4364 | * at this point we either didn't find a cluster | |
4365 | * or we weren't able to allocate a block from our | |
4366 | * cluster. Free the cluster we've been trying | |
4367 | * to use, and go to the next block group | |
4368 | */ | |
0a24325e | 4369 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 4370 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 4371 | goto loop; |
fa9c0d79 CM |
4372 | } |
4373 | ||
6226cb0a JB |
4374 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
4375 | num_bytes, empty_size); | |
1cdda9b8 JB |
4376 | /* |
4377 | * If we didn't find a chunk, and we haven't failed on this | |
4378 | * block group before, and this block group is in the middle of | |
4379 | * caching and we are ok with waiting, then go ahead and wait | |
4380 | * for progress to be made, and set failed_alloc to true. | |
4381 | * | |
4382 | * If failed_alloc is true then we've already waited on this | |
4383 | * block group once and should move on to the next block group. | |
4384 | */ | |
4385 | if (!offset && !failed_alloc && !cached && | |
4386 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 4387 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
4388 | num_bytes + empty_size); |
4389 | failed_alloc = true; | |
817d52f8 | 4390 | goto have_block_group; |
1cdda9b8 JB |
4391 | } else if (!offset) { |
4392 | goto loop; | |
817d52f8 | 4393 | } |
fa9c0d79 | 4394 | checks: |
6226cb0a | 4395 | search_start = stripe_align(root, offset); |
2552d17e | 4396 | /* move on to the next group */ |
6226cb0a JB |
4397 | if (search_start + num_bytes >= search_end) { |
4398 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 4399 | goto loop; |
6226cb0a | 4400 | } |
25179201 | 4401 | |
2552d17e JB |
4402 | /* move on to the next group */ |
4403 | if (search_start + num_bytes > | |
6226cb0a JB |
4404 | block_group->key.objectid + block_group->key.offset) { |
4405 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 4406 | goto loop; |
6226cb0a | 4407 | } |
f5a31e16 | 4408 | |
2552d17e JB |
4409 | if (exclude_nr > 0 && |
4410 | (search_start + num_bytes > exclude_start && | |
4411 | search_start < exclude_start + exclude_nr)) { | |
4412 | search_start = exclude_start + exclude_nr; | |
4413 | ||
6226cb0a | 4414 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e JB |
4415 | /* |
4416 | * if search_start is still in this block group | |
4417 | * then we just re-search this block group | |
f5a31e16 | 4418 | */ |
2552d17e JB |
4419 | if (search_start >= block_group->key.objectid && |
4420 | search_start < (block_group->key.objectid + | |
6226cb0a | 4421 | block_group->key.offset)) |
2552d17e | 4422 | goto have_block_group; |
2552d17e | 4423 | goto loop; |
0f9dd46c | 4424 | } |
0b86a832 | 4425 | |
2552d17e JB |
4426 | ins->objectid = search_start; |
4427 | ins->offset = num_bytes; | |
d2fb3437 | 4428 | |
6226cb0a JB |
4429 | if (offset < search_start) |
4430 | btrfs_add_free_space(block_group, offset, | |
4431 | search_start - offset); | |
4432 | BUG_ON(offset > search_start); | |
4433 | ||
11833d66 YZ |
4434 | update_reserved_extents(block_group, num_bytes, 1); |
4435 | ||
2552d17e | 4436 | /* we are all good, lets return */ |
2552d17e JB |
4437 | break; |
4438 | loop: | |
0a24325e | 4439 | failed_cluster_refill = false; |
1cdda9b8 | 4440 | failed_alloc = false; |
fa9c0d79 | 4441 | btrfs_put_block_group(block_group); |
2552d17e JB |
4442 | } |
4443 | up_read(&space_info->groups_sem); | |
4444 | ||
ccf0e725 JB |
4445 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
4446 | * for them to make caching progress. Also | |
4447 | * determine the best possible bg to cache | |
4448 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
4449 | * caching kthreads as we move along | |
817d52f8 JB |
4450 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
4451 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
4452 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
4453 | * again | |
fa9c0d79 | 4454 | */ |
817d52f8 JB |
4455 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && |
4456 | (found_uncached_bg || empty_size || empty_cluster || | |
4457 | allowed_chunk_alloc)) { | |
ccf0e725 | 4458 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 4459 | found_uncached_bg = false; |
ccf0e725 JB |
4460 | loop++; |
4461 | if (!ideal_cache_percent && | |
4462 | atomic_read(&space_info->caching_threads)) | |
817d52f8 | 4463 | goto search; |
ccf0e725 JB |
4464 | |
4465 | /* | |
4466 | * 1 of the following 2 things have happened so far | |
4467 | * | |
4468 | * 1) We found an ideal block group for caching that | |
4469 | * is mostly full and will cache quickly, so we might | |
4470 | * as well wait for it. | |
4471 | * | |
4472 | * 2) We searched for cached only and we didn't find | |
4473 | * anything, and we didn't start any caching kthreads | |
4474 | * either, so chances are we will loop through and | |
4475 | * start a couple caching kthreads, and then come back | |
4476 | * around and just wait for them. This will be slower | |
4477 | * because we will have 2 caching kthreads reading at | |
4478 | * the same time when we could have just started one | |
4479 | * and waited for it to get far enough to give us an | |
4480 | * allocation, so go ahead and go to the wait caching | |
4481 | * loop. | |
4482 | */ | |
4483 | loop = LOOP_CACHING_WAIT; | |
4484 | search_start = ideal_cache_offset; | |
4485 | ideal_cache_percent = 0; | |
4486 | goto ideal_cache; | |
4487 | } else if (loop == LOOP_FIND_IDEAL) { | |
4488 | /* | |
4489 | * Didn't find a uncached bg, wait on anything we find | |
4490 | * next. | |
4491 | */ | |
4492 | loop = LOOP_CACHING_WAIT; | |
4493 | goto search; | |
4494 | } | |
4495 | ||
4496 | if (loop < LOOP_CACHING_WAIT) { | |
4497 | loop++; | |
4498 | goto search; | |
817d52f8 JB |
4499 | } |
4500 | ||
4501 | if (loop == LOOP_ALLOC_CHUNK) { | |
fa9c0d79 CM |
4502 | empty_size = 0; |
4503 | empty_cluster = 0; | |
4504 | } | |
2552d17e JB |
4505 | |
4506 | if (allowed_chunk_alloc) { | |
4507 | ret = do_chunk_alloc(trans, root, num_bytes + | |
4508 | 2 * 1024 * 1024, data, 1); | |
2552d17e | 4509 | allowed_chunk_alloc = 0; |
ccf0e725 JB |
4510 | done_chunk_alloc = 1; |
4511 | } else if (!done_chunk_alloc) { | |
2552d17e JB |
4512 | space_info->force_alloc = 1; |
4513 | } | |
4514 | ||
817d52f8 | 4515 | if (loop < LOOP_NO_EMPTY_SIZE) { |
fa9c0d79 | 4516 | loop++; |
2552d17e | 4517 | goto search; |
fa9c0d79 | 4518 | } |
2552d17e JB |
4519 | ret = -ENOSPC; |
4520 | } else if (!ins->objectid) { | |
4521 | ret = -ENOSPC; | |
f2654de4 | 4522 | } |
0b86a832 | 4523 | |
80eb234a JB |
4524 | /* we found what we needed */ |
4525 | if (ins->objectid) { | |
4526 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 4527 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 4528 | |
fa9c0d79 | 4529 | btrfs_put_block_group(block_group); |
80eb234a | 4530 | ret = 0; |
be744175 | 4531 | } |
be744175 | 4532 | |
0f70abe2 | 4533 | return ret; |
fec577fb | 4534 | } |
ec44a35c | 4535 | |
9ed74f2d JB |
4536 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
4537 | int dump_block_groups) | |
0f9dd46c JB |
4538 | { |
4539 | struct btrfs_block_group_cache *cache; | |
0f9dd46c | 4540 | |
9ed74f2d | 4541 | spin_lock(&info->lock); |
d397712b CM |
4542 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
4543 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
9ed74f2d JB |
4544 | info->bytes_pinned - info->bytes_reserved - |
4545 | info->bytes_super), | |
d397712b | 4546 | (info->full) ? "" : "not "); |
6a63209f | 4547 | printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," |
9ed74f2d JB |
4548 | " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu" |
4549 | "\n", | |
21380931 JB |
4550 | (unsigned long long)info->total_bytes, |
4551 | (unsigned long long)info->bytes_pinned, | |
4552 | (unsigned long long)info->bytes_delalloc, | |
4553 | (unsigned long long)info->bytes_may_use, | |
9ed74f2d JB |
4554 | (unsigned long long)info->bytes_used, |
4555 | (unsigned long long)info->bytes_root, | |
4556 | (unsigned long long)info->bytes_super, | |
4557 | (unsigned long long)info->bytes_reserved); | |
4558 | spin_unlock(&info->lock); | |
4559 | ||
4560 | if (!dump_block_groups) | |
4561 | return; | |
0f9dd46c | 4562 | |
80eb234a | 4563 | down_read(&info->groups_sem); |
c6e30871 | 4564 | list_for_each_entry(cache, &info->block_groups, list) { |
0f9dd46c | 4565 | spin_lock(&cache->lock); |
d397712b CM |
4566 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
4567 | "%llu pinned %llu reserved\n", | |
4568 | (unsigned long long)cache->key.objectid, | |
4569 | (unsigned long long)cache->key.offset, | |
4570 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
4571 | (unsigned long long)cache->pinned, | |
4572 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
4573 | btrfs_dump_free_space(cache, bytes); |
4574 | spin_unlock(&cache->lock); | |
4575 | } | |
80eb234a | 4576 | up_read(&info->groups_sem); |
0f9dd46c | 4577 | } |
e8569813 | 4578 | |
11833d66 YZ |
4579 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
4580 | struct btrfs_root *root, | |
4581 | u64 num_bytes, u64 min_alloc_size, | |
4582 | u64 empty_size, u64 hint_byte, | |
4583 | u64 search_end, struct btrfs_key *ins, | |
4584 | u64 data) | |
fec577fb CM |
4585 | { |
4586 | int ret; | |
fbdc762b | 4587 | u64 search_start = 0; |
1261ec42 | 4588 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 4589 | |
6a63209f | 4590 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 4591 | again: |
0ef3e66b CM |
4592 | /* |
4593 | * the only place that sets empty_size is btrfs_realloc_node, which | |
4594 | * is not called recursively on allocations | |
4595 | */ | |
4596 | if (empty_size || root->ref_cows) { | |
593060d7 | 4597 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 4598 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
4599 | 2 * 1024 * 1024, |
4600 | BTRFS_BLOCK_GROUP_METADATA | | |
4601 | (info->metadata_alloc_profile & | |
4602 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
4603 | } |
4604 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 4605 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 4606 | } |
0b86a832 | 4607 | |
db94535d CM |
4608 | WARN_ON(num_bytes < root->sectorsize); |
4609 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
4610 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
4611 | trans->alloc_exclude_start, |
4612 | trans->alloc_exclude_nr, data); | |
3b951516 | 4613 | |
98d20f67 CM |
4614 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
4615 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 4616 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 4617 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
4618 | do_chunk_alloc(trans, root->fs_info->extent_root, |
4619 | num_bytes, data, 1); | |
98d20f67 CM |
4620 | goto again; |
4621 | } | |
817d52f8 | 4622 | if (ret == -ENOSPC) { |
0f9dd46c JB |
4623 | struct btrfs_space_info *sinfo; |
4624 | ||
4625 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
4626 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
4627 | "wanted %llu\n", (unsigned long long)data, | |
4628 | (unsigned long long)num_bytes); | |
9ed74f2d | 4629 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 4630 | } |
0f9dd46c JB |
4631 | |
4632 | return ret; | |
e6dcd2dc CM |
4633 | } |
4634 | ||
65b51a00 CM |
4635 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
4636 | { | |
0f9dd46c | 4637 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 4638 | int ret = 0; |
0f9dd46c | 4639 | |
0f9dd46c JB |
4640 | cache = btrfs_lookup_block_group(root->fs_info, start); |
4641 | if (!cache) { | |
d397712b CM |
4642 | printk(KERN_ERR "Unable to find block group for %llu\n", |
4643 | (unsigned long long)start); | |
0f9dd46c JB |
4644 | return -ENOSPC; |
4645 | } | |
1f3c79a2 LH |
4646 | |
4647 | ret = btrfs_discard_extent(root, start, len); | |
4648 | ||
0f9dd46c | 4649 | btrfs_add_free_space(cache, start, len); |
11833d66 | 4650 | update_reserved_extents(cache, len, 0); |
fa9c0d79 | 4651 | btrfs_put_block_group(cache); |
817d52f8 | 4652 | |
e6dcd2dc CM |
4653 | return ret; |
4654 | } | |
4655 | ||
5d4f98a2 YZ |
4656 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
4657 | struct btrfs_root *root, | |
4658 | u64 parent, u64 root_objectid, | |
4659 | u64 flags, u64 owner, u64 offset, | |
4660 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
4661 | { |
4662 | int ret; | |
5d4f98a2 | 4663 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 4664 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4665 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4666 | struct btrfs_path *path; |
5d4f98a2 YZ |
4667 | struct extent_buffer *leaf; |
4668 | int type; | |
4669 | u32 size; | |
26b8003f | 4670 | |
5d4f98a2 YZ |
4671 | if (parent > 0) |
4672 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4673 | else | |
4674 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4675 | |
5d4f98a2 | 4676 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4677 | |
4678 | path = btrfs_alloc_path(); | |
4679 | BUG_ON(!path); | |
47e4bb98 | 4680 | |
b9473439 | 4681 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4682 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4683 | ins, size); | |
ccd467d6 | 4684 | BUG_ON(ret); |
0f9dd46c | 4685 | |
5d4f98a2 YZ |
4686 | leaf = path->nodes[0]; |
4687 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4688 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4689 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4690 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4691 | btrfs_set_extent_flags(leaf, extent_item, | |
4692 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4693 | ||
4694 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4695 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4696 | if (parent > 0) { | |
4697 | struct btrfs_shared_data_ref *ref; | |
4698 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4699 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4700 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4701 | } else { | |
4702 | struct btrfs_extent_data_ref *ref; | |
4703 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4704 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4705 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4706 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4707 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4708 | } | |
47e4bb98 CM |
4709 | |
4710 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4711 | btrfs_free_path(path); |
f510cfec | 4712 | |
5d4f98a2 YZ |
4713 | ret = update_block_group(trans, root, ins->objectid, ins->offset, |
4714 | 1, 0); | |
f5947066 | 4715 | if (ret) { |
d397712b CM |
4716 | printk(KERN_ERR "btrfs update block group failed for %llu " |
4717 | "%llu\n", (unsigned long long)ins->objectid, | |
4718 | (unsigned long long)ins->offset); | |
f5947066 CM |
4719 | BUG(); |
4720 | } | |
e6dcd2dc CM |
4721 | return ret; |
4722 | } | |
4723 | ||
5d4f98a2 YZ |
4724 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4725 | struct btrfs_root *root, | |
4726 | u64 parent, u64 root_objectid, | |
4727 | u64 flags, struct btrfs_disk_key *key, | |
4728 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
4729 | { |
4730 | int ret; | |
5d4f98a2 YZ |
4731 | struct btrfs_fs_info *fs_info = root->fs_info; |
4732 | struct btrfs_extent_item *extent_item; | |
4733 | struct btrfs_tree_block_info *block_info; | |
4734 | struct btrfs_extent_inline_ref *iref; | |
4735 | struct btrfs_path *path; | |
4736 | struct extent_buffer *leaf; | |
4737 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 4738 | |
5d4f98a2 YZ |
4739 | path = btrfs_alloc_path(); |
4740 | BUG_ON(!path); | |
56bec294 | 4741 | |
5d4f98a2 YZ |
4742 | path->leave_spinning = 1; |
4743 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4744 | ins, size); | |
56bec294 | 4745 | BUG_ON(ret); |
5d4f98a2 YZ |
4746 | |
4747 | leaf = path->nodes[0]; | |
4748 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4749 | struct btrfs_extent_item); | |
4750 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4751 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4752 | btrfs_set_extent_flags(leaf, extent_item, | |
4753 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
4754 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
4755 | ||
4756 | btrfs_set_tree_block_key(leaf, block_info, key); | |
4757 | btrfs_set_tree_block_level(leaf, block_info, level); | |
4758 | ||
4759 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
4760 | if (parent > 0) { | |
4761 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
4762 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4763 | BTRFS_SHARED_BLOCK_REF_KEY); | |
4764 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4765 | } else { | |
4766 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4767 | BTRFS_TREE_BLOCK_REF_KEY); | |
4768 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
4769 | } | |
4770 | ||
4771 | btrfs_mark_buffer_dirty(leaf); | |
4772 | btrfs_free_path(path); | |
4773 | ||
4774 | ret = update_block_group(trans, root, ins->objectid, ins->offset, | |
4775 | 1, 0); | |
4776 | if (ret) { | |
4777 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
4778 | "%llu\n", (unsigned long long)ins->objectid, | |
4779 | (unsigned long long)ins->offset); | |
4780 | BUG(); | |
4781 | } | |
4782 | return ret; | |
4783 | } | |
4784 | ||
4785 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
4786 | struct btrfs_root *root, | |
4787 | u64 root_objectid, u64 owner, | |
4788 | u64 offset, struct btrfs_key *ins) | |
4789 | { | |
4790 | int ret; | |
4791 | ||
4792 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
4793 | ||
4794 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
4795 | 0, root_objectid, owner, offset, | |
4796 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
4797 | return ret; |
4798 | } | |
e02119d5 CM |
4799 | |
4800 | /* | |
4801 | * this is used by the tree logging recovery code. It records that | |
4802 | * an extent has been allocated and makes sure to clear the free | |
4803 | * space cache bits as well | |
4804 | */ | |
5d4f98a2 YZ |
4805 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
4806 | struct btrfs_root *root, | |
4807 | u64 root_objectid, u64 owner, u64 offset, | |
4808 | struct btrfs_key *ins) | |
e02119d5 CM |
4809 | { |
4810 | int ret; | |
4811 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
4812 | struct btrfs_caching_control *caching_ctl; |
4813 | u64 start = ins->objectid; | |
4814 | u64 num_bytes = ins->offset; | |
e02119d5 | 4815 | |
e02119d5 | 4816 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
817d52f8 | 4817 | cache_block_group(block_group); |
11833d66 | 4818 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 4819 | |
11833d66 YZ |
4820 | if (!caching_ctl) { |
4821 | BUG_ON(!block_group_cache_done(block_group)); | |
4822 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
4823 | BUG_ON(ret); | |
4824 | } else { | |
4825 | mutex_lock(&caching_ctl->mutex); | |
4826 | ||
4827 | if (start >= caching_ctl->progress) { | |
4828 | ret = add_excluded_extent(root, start, num_bytes); | |
4829 | BUG_ON(ret); | |
4830 | } else if (start + num_bytes <= caching_ctl->progress) { | |
4831 | ret = btrfs_remove_free_space(block_group, | |
4832 | start, num_bytes); | |
4833 | BUG_ON(ret); | |
4834 | } else { | |
4835 | num_bytes = caching_ctl->progress - start; | |
4836 | ret = btrfs_remove_free_space(block_group, | |
4837 | start, num_bytes); | |
4838 | BUG_ON(ret); | |
4839 | ||
4840 | start = caching_ctl->progress; | |
4841 | num_bytes = ins->objectid + ins->offset - | |
4842 | caching_ctl->progress; | |
4843 | ret = add_excluded_extent(root, start, num_bytes); | |
4844 | BUG_ON(ret); | |
4845 | } | |
4846 | ||
4847 | mutex_unlock(&caching_ctl->mutex); | |
4848 | put_caching_control(caching_ctl); | |
4849 | } | |
4850 | ||
4851 | update_reserved_extents(block_group, ins->offset, 1); | |
fa9c0d79 | 4852 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
4853 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
4854 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
4855 | return ret; |
4856 | } | |
4857 | ||
e6dcd2dc CM |
4858 | /* |
4859 | * finds a free extent and does all the dirty work required for allocation | |
4860 | * returns the key for the extent through ins, and a tree buffer for | |
4861 | * the first block of the extent through buf. | |
4862 | * | |
4863 | * returns 0 if everything worked, non-zero otherwise. | |
4864 | */ | |
5d4f98a2 YZ |
4865 | static int alloc_tree_block(struct btrfs_trans_handle *trans, |
4866 | struct btrfs_root *root, | |
4867 | u64 num_bytes, u64 parent, u64 root_objectid, | |
4868 | struct btrfs_disk_key *key, int level, | |
4869 | u64 empty_size, u64 hint_byte, u64 search_end, | |
4870 | struct btrfs_key *ins) | |
e6dcd2dc CM |
4871 | { |
4872 | int ret; | |
5d4f98a2 YZ |
4873 | u64 flags = 0; |
4874 | ||
11833d66 YZ |
4875 | ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes, |
4876 | empty_size, hint_byte, search_end, | |
4877 | ins, 0); | |
817d52f8 JB |
4878 | if (ret) |
4879 | return ret; | |
5d4f98a2 YZ |
4880 | |
4881 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4882 | if (parent == 0) | |
4883 | parent = ins->objectid; | |
4884 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4885 | } else | |
4886 | BUG_ON(parent > 0); | |
4887 | ||
d00aff00 | 4888 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
5d4f98a2 YZ |
4889 | struct btrfs_delayed_extent_op *extent_op; |
4890 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
4891 | BUG_ON(!extent_op); | |
4892 | if (key) | |
4893 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4894 | else | |
4895 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4896 | extent_op->flags_to_set = flags; | |
4897 | extent_op->update_key = 1; | |
4898 | extent_op->update_flags = 1; | |
4899 | extent_op->is_data = 0; | |
4900 | ||
4901 | ret = btrfs_add_delayed_tree_ref(trans, ins->objectid, | |
4902 | ins->offset, parent, root_objectid, | |
4903 | level, BTRFS_ADD_DELAYED_EXTENT, | |
4904 | extent_op); | |
d00aff00 | 4905 | BUG_ON(ret); |
d00aff00 | 4906 | } |
86b9f2ec YZ |
4907 | |
4908 | if (root_objectid == root->root_key.objectid) { | |
4909 | u64 used; | |
4910 | spin_lock(&root->node_lock); | |
4911 | used = btrfs_root_used(&root->root_item) + num_bytes; | |
4912 | btrfs_set_root_used(&root->root_item, used); | |
4913 | spin_unlock(&root->node_lock); | |
4914 | } | |
925baedd | 4915 | return ret; |
fec577fb | 4916 | } |
65b51a00 CM |
4917 | |
4918 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
4919 | struct btrfs_root *root, | |
4008c04a CM |
4920 | u64 bytenr, u32 blocksize, |
4921 | int level) | |
65b51a00 CM |
4922 | { |
4923 | struct extent_buffer *buf; | |
4924 | ||
4925 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
4926 | if (!buf) | |
4927 | return ERR_PTR(-ENOMEM); | |
4928 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 4929 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
4930 | btrfs_tree_lock(buf); |
4931 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
4932 | |
4933 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 4934 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 4935 | |
d0c803c4 | 4936 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
4937 | /* |
4938 | * we allow two log transactions at a time, use different | |
4939 | * EXENT bit to differentiate dirty pages. | |
4940 | */ | |
4941 | if (root->log_transid % 2 == 0) | |
4942 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
4943 | buf->start + buf->len - 1, GFP_NOFS); | |
4944 | else | |
4945 | set_extent_new(&root->dirty_log_pages, buf->start, | |
4946 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
4947 | } else { |
4948 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 4949 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4950 | } |
65b51a00 | 4951 | trans->blocks_used++; |
b4ce94de | 4952 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4953 | return buf; |
4954 | } | |
4955 | ||
fec577fb CM |
4956 | /* |
4957 | * helper function to allocate a block for a given tree | |
4958 | * returns the tree buffer or NULL. | |
4959 | */ | |
5f39d397 | 4960 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4961 | struct btrfs_root *root, u32 blocksize, |
4962 | u64 parent, u64 root_objectid, | |
4963 | struct btrfs_disk_key *key, int level, | |
4964 | u64 hint, u64 empty_size) | |
fec577fb | 4965 | { |
e2fa7227 | 4966 | struct btrfs_key ins; |
fec577fb | 4967 | int ret; |
5f39d397 | 4968 | struct extent_buffer *buf; |
fec577fb | 4969 | |
5d4f98a2 YZ |
4970 | ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid, |
4971 | key, level, empty_size, hint, (u64)-1, &ins); | |
fec577fb | 4972 | if (ret) { |
54aa1f4d CM |
4973 | BUG_ON(ret > 0); |
4974 | return ERR_PTR(ret); | |
fec577fb | 4975 | } |
55c69072 | 4976 | |
4008c04a CM |
4977 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
4978 | blocksize, level); | |
fec577fb CM |
4979 | return buf; |
4980 | } | |
a28ec197 | 4981 | |
2c47e605 YZ |
4982 | struct walk_control { |
4983 | u64 refs[BTRFS_MAX_LEVEL]; | |
4984 | u64 flags[BTRFS_MAX_LEVEL]; | |
4985 | struct btrfs_key update_progress; | |
4986 | int stage; | |
4987 | int level; | |
4988 | int shared_level; | |
4989 | int update_ref; | |
4990 | int keep_locks; | |
1c4850e2 YZ |
4991 | int reada_slot; |
4992 | int reada_count; | |
2c47e605 YZ |
4993 | }; |
4994 | ||
4995 | #define DROP_REFERENCE 1 | |
4996 | #define UPDATE_BACKREF 2 | |
4997 | ||
1c4850e2 YZ |
4998 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4999 | struct btrfs_root *root, | |
5000 | struct walk_control *wc, | |
5001 | struct btrfs_path *path) | |
6407bf6d | 5002 | { |
1c4850e2 YZ |
5003 | u64 bytenr; |
5004 | u64 generation; | |
5005 | u64 refs; | |
94fcca9f | 5006 | u64 flags; |
1c4850e2 | 5007 | u64 last = 0; |
5d4f98a2 | 5008 | u32 nritems; |
1c4850e2 YZ |
5009 | u32 blocksize; |
5010 | struct btrfs_key key; | |
5011 | struct extent_buffer *eb; | |
6407bf6d | 5012 | int ret; |
1c4850e2 YZ |
5013 | int slot; |
5014 | int nread = 0; | |
6407bf6d | 5015 | |
1c4850e2 YZ |
5016 | if (path->slots[wc->level] < wc->reada_slot) { |
5017 | wc->reada_count = wc->reada_count * 2 / 3; | |
5018 | wc->reada_count = max(wc->reada_count, 2); | |
5019 | } else { | |
5020 | wc->reada_count = wc->reada_count * 3 / 2; | |
5021 | wc->reada_count = min_t(int, wc->reada_count, | |
5022 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
5023 | } | |
7bb86316 | 5024 | |
1c4850e2 YZ |
5025 | eb = path->nodes[wc->level]; |
5026 | nritems = btrfs_header_nritems(eb); | |
5027 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 5028 | |
1c4850e2 YZ |
5029 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5030 | if (nread >= wc->reada_count) | |
5031 | break; | |
bd56b302 | 5032 | |
2dd3e67b | 5033 | cond_resched(); |
1c4850e2 YZ |
5034 | bytenr = btrfs_node_blockptr(eb, slot); |
5035 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5036 | |
1c4850e2 YZ |
5037 | if (slot == path->slots[wc->level]) |
5038 | goto reada; | |
5d4f98a2 | 5039 | |
1c4850e2 YZ |
5040 | if (wc->stage == UPDATE_BACKREF && |
5041 | generation <= root->root_key.offset) | |
bd56b302 CM |
5042 | continue; |
5043 | ||
94fcca9f YZ |
5044 | /* We don't lock the tree block, it's OK to be racy here */ |
5045 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
5046 | &refs, &flags); | |
5047 | BUG_ON(ret); | |
5048 | BUG_ON(refs == 0); | |
5049 | ||
1c4850e2 | 5050 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5051 | if (refs == 1) |
5052 | goto reada; | |
bd56b302 | 5053 | |
94fcca9f YZ |
5054 | if (wc->level == 1 && |
5055 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5056 | continue; | |
1c4850e2 YZ |
5057 | if (!wc->update_ref || |
5058 | generation <= root->root_key.offset) | |
5059 | continue; | |
5060 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5061 | ret = btrfs_comp_cpu_keys(&key, | |
5062 | &wc->update_progress); | |
5063 | if (ret < 0) | |
5064 | continue; | |
94fcca9f YZ |
5065 | } else { |
5066 | if (wc->level == 1 && | |
5067 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5068 | continue; | |
6407bf6d | 5069 | } |
1c4850e2 YZ |
5070 | reada: |
5071 | ret = readahead_tree_block(root, bytenr, blocksize, | |
5072 | generation); | |
5073 | if (ret) | |
bd56b302 | 5074 | break; |
1c4850e2 YZ |
5075 | last = bytenr + blocksize; |
5076 | nread++; | |
20524f02 | 5077 | } |
1c4850e2 | 5078 | wc->reada_slot = slot; |
20524f02 | 5079 | } |
2c47e605 | 5080 | |
f82d02d9 | 5081 | /* |
2c47e605 YZ |
5082 | * hepler to process tree block while walking down the tree. |
5083 | * | |
2c47e605 YZ |
5084 | * when wc->stage == UPDATE_BACKREF, this function updates |
5085 | * back refs for pointers in the block. | |
5086 | * | |
5087 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5088 | */ |
2c47e605 | 5089 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5090 | struct btrfs_root *root, |
2c47e605 | 5091 | struct btrfs_path *path, |
94fcca9f | 5092 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5093 | { |
2c47e605 YZ |
5094 | int level = wc->level; |
5095 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5096 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5097 | int ret; |
5098 | ||
2c47e605 YZ |
5099 | if (wc->stage == UPDATE_BACKREF && |
5100 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5101 | return 1; | |
f82d02d9 | 5102 | |
2c47e605 YZ |
5103 | /* |
5104 | * when reference count of tree block is 1, it won't increase | |
5105 | * again. once full backref flag is set, we never clear it. | |
5106 | */ | |
94fcca9f YZ |
5107 | if (lookup_info && |
5108 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5109 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
5110 | BUG_ON(!path->locks[level]); |
5111 | ret = btrfs_lookup_extent_info(trans, root, | |
5112 | eb->start, eb->len, | |
5113 | &wc->refs[level], | |
5114 | &wc->flags[level]); | |
5115 | BUG_ON(ret); | |
5116 | BUG_ON(wc->refs[level] == 0); | |
5117 | } | |
5d4f98a2 | 5118 | |
2c47e605 YZ |
5119 | if (wc->stage == DROP_REFERENCE) { |
5120 | if (wc->refs[level] > 1) | |
5121 | return 1; | |
f82d02d9 | 5122 | |
2c47e605 YZ |
5123 | if (path->locks[level] && !wc->keep_locks) { |
5124 | btrfs_tree_unlock(eb); | |
5125 | path->locks[level] = 0; | |
5126 | } | |
5127 | return 0; | |
5128 | } | |
f82d02d9 | 5129 | |
2c47e605 YZ |
5130 | /* wc->stage == UPDATE_BACKREF */ |
5131 | if (!(wc->flags[level] & flag)) { | |
5132 | BUG_ON(!path->locks[level]); | |
5133 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 5134 | BUG_ON(ret); |
2c47e605 YZ |
5135 | ret = btrfs_dec_ref(trans, root, eb, 0); |
5136 | BUG_ON(ret); | |
5137 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
5138 | eb->len, flag, 0); | |
5139 | BUG_ON(ret); | |
5140 | wc->flags[level] |= flag; | |
5141 | } | |
5142 | ||
5143 | /* | |
5144 | * the block is shared by multiple trees, so it's not good to | |
5145 | * keep the tree lock | |
5146 | */ | |
5147 | if (path->locks[level] && level > 0) { | |
5148 | btrfs_tree_unlock(eb); | |
5149 | path->locks[level] = 0; | |
5150 | } | |
5151 | return 0; | |
5152 | } | |
5153 | ||
1c4850e2 YZ |
5154 | /* |
5155 | * hepler to process tree block pointer. | |
5156 | * | |
5157 | * when wc->stage == DROP_REFERENCE, this function checks | |
5158 | * reference count of the block pointed to. if the block | |
5159 | * is shared and we need update back refs for the subtree | |
5160 | * rooted at the block, this function changes wc->stage to | |
5161 | * UPDATE_BACKREF. if the block is shared and there is no | |
5162 | * need to update back, this function drops the reference | |
5163 | * to the block. | |
5164 | * | |
5165 | * NOTE: return value 1 means we should stop walking down. | |
5166 | */ | |
5167 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5168 | struct btrfs_root *root, | |
5169 | struct btrfs_path *path, | |
94fcca9f | 5170 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
5171 | { |
5172 | u64 bytenr; | |
5173 | u64 generation; | |
5174 | u64 parent; | |
5175 | u32 blocksize; | |
5176 | struct btrfs_key key; | |
5177 | struct extent_buffer *next; | |
5178 | int level = wc->level; | |
5179 | int reada = 0; | |
5180 | int ret = 0; | |
5181 | ||
5182 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5183 | path->slots[level]); | |
5184 | /* | |
5185 | * if the lower level block was created before the snapshot | |
5186 | * was created, we know there is no need to update back refs | |
5187 | * for the subtree | |
5188 | */ | |
5189 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5190 | generation <= root->root_key.offset) { |
5191 | *lookup_info = 1; | |
1c4850e2 | 5192 | return 1; |
94fcca9f | 5193 | } |
1c4850e2 YZ |
5194 | |
5195 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
5196 | blocksize = btrfs_level_size(root, level - 1); | |
5197 | ||
5198 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
5199 | if (!next) { | |
5200 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5201 | reada = 1; | |
5202 | } | |
5203 | btrfs_tree_lock(next); | |
5204 | btrfs_set_lock_blocking(next); | |
5205 | ||
94fcca9f YZ |
5206 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
5207 | &wc->refs[level - 1], | |
5208 | &wc->flags[level - 1]); | |
5209 | BUG_ON(ret); | |
5210 | BUG_ON(wc->refs[level - 1] == 0); | |
5211 | *lookup_info = 0; | |
1c4850e2 | 5212 | |
94fcca9f | 5213 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5214 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
5215 | if (level == 1 && |
5216 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5217 | goto skip; | |
5218 | ||
1c4850e2 YZ |
5219 | if (!wc->update_ref || |
5220 | generation <= root->root_key.offset) | |
5221 | goto skip; | |
5222 | ||
5223 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5224 | path->slots[level]); | |
5225 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5226 | if (ret < 0) | |
5227 | goto skip; | |
5228 | ||
5229 | wc->stage = UPDATE_BACKREF; | |
5230 | wc->shared_level = level - 1; | |
5231 | } | |
94fcca9f YZ |
5232 | } else { |
5233 | if (level == 1 && | |
5234 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5235 | goto skip; | |
1c4850e2 YZ |
5236 | } |
5237 | ||
5238 | if (!btrfs_buffer_uptodate(next, generation)) { | |
5239 | btrfs_tree_unlock(next); | |
5240 | free_extent_buffer(next); | |
5241 | next = NULL; | |
94fcca9f | 5242 | *lookup_info = 1; |
1c4850e2 YZ |
5243 | } |
5244 | ||
5245 | if (!next) { | |
5246 | if (reada && level == 1) | |
5247 | reada_walk_down(trans, root, wc, path); | |
5248 | next = read_tree_block(root, bytenr, blocksize, generation); | |
5249 | btrfs_tree_lock(next); | |
5250 | btrfs_set_lock_blocking(next); | |
5251 | } | |
5252 | ||
5253 | level--; | |
5254 | BUG_ON(level != btrfs_header_level(next)); | |
5255 | path->nodes[level] = next; | |
5256 | path->slots[level] = 0; | |
5257 | path->locks[level] = 1; | |
5258 | wc->level = level; | |
5259 | if (wc->level == 1) | |
5260 | wc->reada_slot = 0; | |
5261 | return 0; | |
5262 | skip: | |
5263 | wc->refs[level - 1] = 0; | |
5264 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5265 | if (wc->stage == DROP_REFERENCE) { |
5266 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5267 | parent = path->nodes[level]->start; | |
5268 | } else { | |
5269 | BUG_ON(root->root_key.objectid != | |
5270 | btrfs_header_owner(path->nodes[level])); | |
5271 | parent = 0; | |
5272 | } | |
1c4850e2 | 5273 | |
94fcca9f YZ |
5274 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
5275 | root->root_key.objectid, level - 1, 0); | |
5276 | BUG_ON(ret); | |
1c4850e2 | 5277 | } |
1c4850e2 YZ |
5278 | btrfs_tree_unlock(next); |
5279 | free_extent_buffer(next); | |
94fcca9f | 5280 | *lookup_info = 1; |
1c4850e2 YZ |
5281 | return 1; |
5282 | } | |
5283 | ||
2c47e605 YZ |
5284 | /* |
5285 | * hepler to process tree block while walking up the tree. | |
5286 | * | |
5287 | * when wc->stage == DROP_REFERENCE, this function drops | |
5288 | * reference count on the block. | |
5289 | * | |
5290 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5291 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5292 | * to UPDATE_BACKREF previously while processing the block. | |
5293 | * | |
5294 | * NOTE: return value 1 means we should stop walking up. | |
5295 | */ | |
5296 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5297 | struct btrfs_root *root, | |
5298 | struct btrfs_path *path, | |
5299 | struct walk_control *wc) | |
5300 | { | |
5301 | int ret = 0; | |
5302 | int level = wc->level; | |
5303 | struct extent_buffer *eb = path->nodes[level]; | |
5304 | u64 parent = 0; | |
5305 | ||
5306 | if (wc->stage == UPDATE_BACKREF) { | |
5307 | BUG_ON(wc->shared_level < level); | |
5308 | if (level < wc->shared_level) | |
5309 | goto out; | |
5310 | ||
2c47e605 YZ |
5311 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5312 | if (ret > 0) | |
5313 | wc->update_ref = 0; | |
5314 | ||
5315 | wc->stage = DROP_REFERENCE; | |
5316 | wc->shared_level = -1; | |
5317 | path->slots[level] = 0; | |
5318 | ||
5319 | /* | |
5320 | * check reference count again if the block isn't locked. | |
5321 | * we should start walking down the tree again if reference | |
5322 | * count is one. | |
5323 | */ | |
5324 | if (!path->locks[level]) { | |
5325 | BUG_ON(level == 0); | |
5326 | btrfs_tree_lock(eb); | |
5327 | btrfs_set_lock_blocking(eb); | |
5328 | path->locks[level] = 1; | |
5329 | ||
5330 | ret = btrfs_lookup_extent_info(trans, root, | |
5331 | eb->start, eb->len, | |
5332 | &wc->refs[level], | |
5333 | &wc->flags[level]); | |
f82d02d9 | 5334 | BUG_ON(ret); |
2c47e605 YZ |
5335 | BUG_ON(wc->refs[level] == 0); |
5336 | if (wc->refs[level] == 1) { | |
5337 | btrfs_tree_unlock(eb); | |
5338 | path->locks[level] = 0; | |
5339 | return 1; | |
5340 | } | |
f82d02d9 | 5341 | } |
2c47e605 | 5342 | } |
f82d02d9 | 5343 | |
2c47e605 YZ |
5344 | /* wc->stage == DROP_REFERENCE */ |
5345 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5346 | |
2c47e605 YZ |
5347 | if (wc->refs[level] == 1) { |
5348 | if (level == 0) { | |
5349 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5350 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
5351 | else | |
5352 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
5353 | BUG_ON(ret); | |
5354 | } | |
5355 | /* make block locked assertion in clean_tree_block happy */ | |
5356 | if (!path->locks[level] && | |
5357 | btrfs_header_generation(eb) == trans->transid) { | |
5358 | btrfs_tree_lock(eb); | |
5359 | btrfs_set_lock_blocking(eb); | |
5360 | path->locks[level] = 1; | |
5361 | } | |
5362 | clean_tree_block(trans, root, eb); | |
5363 | } | |
5364 | ||
5365 | if (eb == root->node) { | |
5366 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5367 | parent = eb->start; | |
5368 | else | |
5369 | BUG_ON(root->root_key.objectid != | |
5370 | btrfs_header_owner(eb)); | |
5371 | } else { | |
5372 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5373 | parent = path->nodes[level + 1]->start; | |
5374 | else | |
5375 | BUG_ON(root->root_key.objectid != | |
5376 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 5377 | } |
f82d02d9 | 5378 | |
2c47e605 YZ |
5379 | ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent, |
5380 | root->root_key.objectid, level, 0); | |
f82d02d9 | 5381 | BUG_ON(ret); |
2c47e605 YZ |
5382 | out: |
5383 | wc->refs[level] = 0; | |
5384 | wc->flags[level] = 0; | |
5385 | return ret; | |
5386 | } | |
5387 | ||
5388 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5389 | struct btrfs_root *root, | |
5390 | struct btrfs_path *path, | |
5391 | struct walk_control *wc) | |
5392 | { | |
2c47e605 | 5393 | int level = wc->level; |
94fcca9f | 5394 | int lookup_info = 1; |
2c47e605 YZ |
5395 | int ret; |
5396 | ||
5397 | while (level >= 0) { | |
1c4850e2 YZ |
5398 | if (path->slots[level] >= |
5399 | btrfs_header_nritems(path->nodes[level])) | |
5400 | break; | |
f82d02d9 | 5401 | |
94fcca9f | 5402 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5403 | if (ret > 0) |
5404 | break; | |
5405 | ||
5406 | if (level == 0) | |
5407 | break; | |
5408 | ||
94fcca9f | 5409 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5410 | if (ret > 0) { |
5411 | path->slots[level]++; | |
5412 | continue; | |
5413 | } | |
5414 | level = wc->level; | |
f82d02d9 | 5415 | } |
f82d02d9 YZ |
5416 | return 0; |
5417 | } | |
5418 | ||
d397712b | 5419 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5420 | struct btrfs_root *root, |
f82d02d9 | 5421 | struct btrfs_path *path, |
2c47e605 | 5422 | struct walk_control *wc, int max_level) |
20524f02 | 5423 | { |
2c47e605 | 5424 | int level = wc->level; |
20524f02 | 5425 | int ret; |
9f3a7427 | 5426 | |
2c47e605 YZ |
5427 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5428 | while (level < max_level && path->nodes[level]) { | |
5429 | wc->level = level; | |
5430 | if (path->slots[level] + 1 < | |
5431 | btrfs_header_nritems(path->nodes[level])) { | |
5432 | path->slots[level]++; | |
20524f02 CM |
5433 | return 0; |
5434 | } else { | |
2c47e605 YZ |
5435 | ret = walk_up_proc(trans, root, path, wc); |
5436 | if (ret > 0) | |
5437 | return 0; | |
bd56b302 | 5438 | |
2c47e605 YZ |
5439 | if (path->locks[level]) { |
5440 | btrfs_tree_unlock(path->nodes[level]); | |
5441 | path->locks[level] = 0; | |
f82d02d9 | 5442 | } |
2c47e605 YZ |
5443 | free_extent_buffer(path->nodes[level]); |
5444 | path->nodes[level] = NULL; | |
5445 | level++; | |
20524f02 CM |
5446 | } |
5447 | } | |
5448 | return 1; | |
5449 | } | |
5450 | ||
9aca1d51 | 5451 | /* |
2c47e605 YZ |
5452 | * drop a subvolume tree. |
5453 | * | |
5454 | * this function traverses the tree freeing any blocks that only | |
5455 | * referenced by the tree. | |
5456 | * | |
5457 | * when a shared tree block is found. this function decreases its | |
5458 | * reference count by one. if update_ref is true, this function | |
5459 | * also make sure backrefs for the shared block and all lower level | |
5460 | * blocks are properly updated. | |
9aca1d51 | 5461 | */ |
2c47e605 | 5462 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref) |
20524f02 | 5463 | { |
5caf2a00 | 5464 | struct btrfs_path *path; |
2c47e605 YZ |
5465 | struct btrfs_trans_handle *trans; |
5466 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 5467 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5468 | struct walk_control *wc; |
5469 | struct btrfs_key key; | |
5470 | int err = 0; | |
5471 | int ret; | |
5472 | int level; | |
20524f02 | 5473 | |
5caf2a00 CM |
5474 | path = btrfs_alloc_path(); |
5475 | BUG_ON(!path); | |
20524f02 | 5476 | |
2c47e605 YZ |
5477 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
5478 | BUG_ON(!wc); | |
5479 | ||
5480 | trans = btrfs_start_transaction(tree_root, 1); | |
5481 | ||
9f3a7427 | 5482 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5483 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
5484 | path->nodes[level] = btrfs_lock_root_node(root); |
5485 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 5486 | path->slots[level] = 0; |
5d4f98a2 | 5487 | path->locks[level] = 1; |
2c47e605 YZ |
5488 | memset(&wc->update_progress, 0, |
5489 | sizeof(wc->update_progress)); | |
9f3a7427 | 5490 | } else { |
9f3a7427 | 5491 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5492 | memcpy(&wc->update_progress, &key, |
5493 | sizeof(wc->update_progress)); | |
5494 | ||
6702ed49 | 5495 | level = root_item->drop_level; |
2c47e605 | 5496 | BUG_ON(level == 0); |
6702ed49 | 5497 | path->lowest_level = level; |
2c47e605 YZ |
5498 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5499 | path->lowest_level = 0; | |
5500 | if (ret < 0) { | |
5501 | err = ret; | |
9f3a7427 CM |
5502 | goto out; |
5503 | } | |
1c4850e2 | 5504 | WARN_ON(ret > 0); |
2c47e605 | 5505 | |
7d9eb12c CM |
5506 | /* |
5507 | * unlock our path, this is safe because only this | |
5508 | * function is allowed to delete this snapshot | |
5509 | */ | |
5d4f98a2 | 5510 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5511 | |
5512 | level = btrfs_header_level(root->node); | |
5513 | while (1) { | |
5514 | btrfs_tree_lock(path->nodes[level]); | |
5515 | btrfs_set_lock_blocking(path->nodes[level]); | |
5516 | ||
5517 | ret = btrfs_lookup_extent_info(trans, root, | |
5518 | path->nodes[level]->start, | |
5519 | path->nodes[level]->len, | |
5520 | &wc->refs[level], | |
5521 | &wc->flags[level]); | |
5522 | BUG_ON(ret); | |
5523 | BUG_ON(wc->refs[level] == 0); | |
5524 | ||
5525 | if (level == root_item->drop_level) | |
5526 | break; | |
5527 | ||
5528 | btrfs_tree_unlock(path->nodes[level]); | |
5529 | WARN_ON(wc->refs[level] != 1); | |
5530 | level--; | |
5531 | } | |
9f3a7427 | 5532 | } |
2c47e605 YZ |
5533 | |
5534 | wc->level = level; | |
5535 | wc->shared_level = -1; | |
5536 | wc->stage = DROP_REFERENCE; | |
5537 | wc->update_ref = update_ref; | |
5538 | wc->keep_locks = 0; | |
1c4850e2 | 5539 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 5540 | |
d397712b | 5541 | while (1) { |
2c47e605 YZ |
5542 | ret = walk_down_tree(trans, root, path, wc); |
5543 | if (ret < 0) { | |
5544 | err = ret; | |
20524f02 | 5545 | break; |
2c47e605 | 5546 | } |
9aca1d51 | 5547 | |
2c47e605 YZ |
5548 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5549 | if (ret < 0) { | |
5550 | err = ret; | |
20524f02 | 5551 | break; |
2c47e605 YZ |
5552 | } |
5553 | ||
5554 | if (ret > 0) { | |
5555 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5556 | break; |
5557 | } | |
2c47e605 YZ |
5558 | |
5559 | if (wc->stage == DROP_REFERENCE) { | |
5560 | level = wc->level; | |
5561 | btrfs_node_key(path->nodes[level], | |
5562 | &root_item->drop_progress, | |
5563 | path->slots[level]); | |
5564 | root_item->drop_level = level; | |
5565 | } | |
5566 | ||
5567 | BUG_ON(wc->level == 0); | |
5568 | if (trans->transaction->in_commit || | |
5569 | trans->transaction->delayed_refs.flushing) { | |
5570 | ret = btrfs_update_root(trans, tree_root, | |
5571 | &root->root_key, | |
5572 | root_item); | |
5573 | BUG_ON(ret); | |
5574 | ||
5575 | btrfs_end_transaction(trans, tree_root); | |
5576 | trans = btrfs_start_transaction(tree_root, 1); | |
5577 | } else { | |
5578 | unsigned long update; | |
c3e69d58 CM |
5579 | update = trans->delayed_ref_updates; |
5580 | trans->delayed_ref_updates = 0; | |
5581 | if (update) | |
2c47e605 YZ |
5582 | btrfs_run_delayed_refs(trans, tree_root, |
5583 | update); | |
c3e69d58 | 5584 | } |
20524f02 | 5585 | } |
2c47e605 YZ |
5586 | btrfs_release_path(root, path); |
5587 | BUG_ON(err); | |
5588 | ||
5589 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
5590 | BUG_ON(ret); | |
5591 | ||
76dda93c YZ |
5592 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
5593 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
5594 | NULL, NULL); | |
5595 | BUG_ON(ret < 0); | |
5596 | if (ret > 0) { | |
5597 | ret = btrfs_del_orphan_item(trans, tree_root, | |
5598 | root->root_key.objectid); | |
5599 | BUG_ON(ret); | |
5600 | } | |
5601 | } | |
5602 | ||
5603 | if (root->in_radix) { | |
5604 | btrfs_free_fs_root(tree_root->fs_info, root); | |
5605 | } else { | |
5606 | free_extent_buffer(root->node); | |
5607 | free_extent_buffer(root->commit_root); | |
5608 | kfree(root); | |
5609 | } | |
9f3a7427 | 5610 | out: |
2c47e605 YZ |
5611 | btrfs_end_transaction(trans, tree_root); |
5612 | kfree(wc); | |
5caf2a00 | 5613 | btrfs_free_path(path); |
2c47e605 | 5614 | return err; |
20524f02 | 5615 | } |
9078a3e1 | 5616 | |
2c47e605 YZ |
5617 | /* |
5618 | * drop subtree rooted at tree block 'node'. | |
5619 | * | |
5620 | * NOTE: this function will unlock and release tree block 'node' | |
5621 | */ | |
f82d02d9 YZ |
5622 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5623 | struct btrfs_root *root, | |
5624 | struct extent_buffer *node, | |
5625 | struct extent_buffer *parent) | |
5626 | { | |
5627 | struct btrfs_path *path; | |
2c47e605 | 5628 | struct walk_control *wc; |
f82d02d9 YZ |
5629 | int level; |
5630 | int parent_level; | |
5631 | int ret = 0; | |
5632 | int wret; | |
5633 | ||
2c47e605 YZ |
5634 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5635 | ||
f82d02d9 YZ |
5636 | path = btrfs_alloc_path(); |
5637 | BUG_ON(!path); | |
5638 | ||
2c47e605 YZ |
5639 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
5640 | BUG_ON(!wc); | |
5641 | ||
b9447ef8 | 5642 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
5643 | parent_level = btrfs_header_level(parent); |
5644 | extent_buffer_get(parent); | |
5645 | path->nodes[parent_level] = parent; | |
5646 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5647 | ||
b9447ef8 | 5648 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5649 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5650 | path->nodes[level] = node; |
5651 | path->slots[level] = 0; | |
2c47e605 YZ |
5652 | path->locks[level] = 1; |
5653 | ||
5654 | wc->refs[parent_level] = 1; | |
5655 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5656 | wc->level = level; | |
5657 | wc->shared_level = -1; | |
5658 | wc->stage = DROP_REFERENCE; | |
5659 | wc->update_ref = 0; | |
5660 | wc->keep_locks = 1; | |
1c4850e2 | 5661 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
5662 | |
5663 | while (1) { | |
2c47e605 YZ |
5664 | wret = walk_down_tree(trans, root, path, wc); |
5665 | if (wret < 0) { | |
f82d02d9 | 5666 | ret = wret; |
f82d02d9 | 5667 | break; |
2c47e605 | 5668 | } |
f82d02d9 | 5669 | |
2c47e605 | 5670 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5671 | if (wret < 0) |
5672 | ret = wret; | |
5673 | if (wret != 0) | |
5674 | break; | |
5675 | } | |
5676 | ||
2c47e605 | 5677 | kfree(wc); |
f82d02d9 YZ |
5678 | btrfs_free_path(path); |
5679 | return ret; | |
5680 | } | |
5681 | ||
5d4f98a2 | 5682 | #if 0 |
8e7bf94f CM |
5683 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
5684 | unsigned long nr) | |
5685 | { | |
5686 | return min(last, start + nr - 1); | |
5687 | } | |
5688 | ||
d397712b | 5689 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 5690 | u64 len) |
edbd8d4e CM |
5691 | { |
5692 | u64 page_start; | |
5693 | u64 page_end; | |
1a40e23b | 5694 | unsigned long first_index; |
edbd8d4e | 5695 | unsigned long last_index; |
edbd8d4e CM |
5696 | unsigned long i; |
5697 | struct page *page; | |
d1310b2e | 5698 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 5699 | struct file_ra_state *ra; |
3eaa2885 | 5700 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
5701 | unsigned int total_read = 0; |
5702 | unsigned int total_dirty = 0; | |
5703 | int ret = 0; | |
4313b399 CM |
5704 | |
5705 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
5706 | |
5707 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 5708 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
5709 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
5710 | ||
1a40e23b ZY |
5711 | /* make sure the dirty trick played by the caller work */ |
5712 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5713 | first_index, last_index); | |
5714 | if (ret) | |
5715 | goto out_unlock; | |
8e7bf94f | 5716 | |
4313b399 | 5717 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 5718 | |
1a40e23b ZY |
5719 | for (i = first_index ; i <= last_index; i++) { |
5720 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 5721 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 5722 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
5723 | } |
5724 | total_read++; | |
3eaa2885 CM |
5725 | again: |
5726 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 5727 | BUG_ON(1); |
edbd8d4e | 5728 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 5729 | if (!page) { |
1a40e23b | 5730 | ret = -ENOMEM; |
edbd8d4e | 5731 | goto out_unlock; |
a061fc8d | 5732 | } |
edbd8d4e CM |
5733 | if (!PageUptodate(page)) { |
5734 | btrfs_readpage(NULL, page); | |
5735 | lock_page(page); | |
5736 | if (!PageUptodate(page)) { | |
5737 | unlock_page(page); | |
5738 | page_cache_release(page); | |
1a40e23b | 5739 | ret = -EIO; |
edbd8d4e CM |
5740 | goto out_unlock; |
5741 | } | |
5742 | } | |
ec44a35c | 5743 | wait_on_page_writeback(page); |
3eaa2885 | 5744 | |
edbd8d4e CM |
5745 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
5746 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 5747 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 5748 | |
3eaa2885 CM |
5749 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5750 | if (ordered) { | |
5751 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5752 | unlock_page(page); | |
5753 | page_cache_release(page); | |
5754 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5755 | btrfs_put_ordered_extent(ordered); | |
5756 | goto again; | |
5757 | } | |
5758 | set_page_extent_mapped(page); | |
5759 | ||
1a40e23b ZY |
5760 | if (i == first_index) |
5761 | set_extent_bits(io_tree, page_start, page_end, | |
5762 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 5763 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 5764 | |
a061fc8d | 5765 | set_page_dirty(page); |
1a40e23b | 5766 | total_dirty++; |
edbd8d4e | 5767 | |
d1310b2e | 5768 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
5769 | unlock_page(page); |
5770 | page_cache_release(page); | |
5771 | } | |
5772 | ||
5773 | out_unlock: | |
ec44a35c | 5774 | kfree(ra); |
edbd8d4e | 5775 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
5776 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
5777 | return ret; | |
edbd8d4e CM |
5778 | } |
5779 | ||
d397712b | 5780 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
5781 | struct btrfs_key *extent_key, |
5782 | u64 offset) | |
5783 | { | |
5784 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
5785 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
5786 | struct extent_map *em; | |
6643558d YZ |
5787 | u64 start = extent_key->objectid - offset; |
5788 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 5789 | |
1a40e23b ZY |
5790 | em = alloc_extent_map(GFP_NOFS); |
5791 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 5792 | |
6643558d | 5793 | em->start = start; |
1a40e23b | 5794 | em->len = extent_key->offset; |
c8b97818 | 5795 | em->block_len = extent_key->offset; |
1a40e23b ZY |
5796 | em->block_start = extent_key->objectid; |
5797 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
5798 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
5799 | ||
5800 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 5801 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
5802 | while (1) { |
5803 | int ret; | |
890871be | 5804 | write_lock(&em_tree->lock); |
1a40e23b | 5805 | ret = add_extent_mapping(em_tree, em); |
890871be | 5806 | write_unlock(&em_tree->lock); |
1a40e23b ZY |
5807 | if (ret != -EEXIST) { |
5808 | free_extent_map(em); | |
bf4ef679 CM |
5809 | break; |
5810 | } | |
6643558d | 5811 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 5812 | } |
6643558d | 5813 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 5814 | |
6643558d | 5815 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 5816 | } |
edbd8d4e | 5817 | |
1a40e23b ZY |
5818 | struct btrfs_ref_path { |
5819 | u64 extent_start; | |
5820 | u64 nodes[BTRFS_MAX_LEVEL]; | |
5821 | u64 root_objectid; | |
5822 | u64 root_generation; | |
5823 | u64 owner_objectid; | |
1a40e23b ZY |
5824 | u32 num_refs; |
5825 | int lowest_level; | |
5826 | int current_level; | |
f82d02d9 YZ |
5827 | int shared_level; |
5828 | ||
5829 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
5830 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 5831 | }; |
7d9eb12c | 5832 | |
1a40e23b | 5833 | struct disk_extent { |
c8b97818 | 5834 | u64 ram_bytes; |
1a40e23b ZY |
5835 | u64 disk_bytenr; |
5836 | u64 disk_num_bytes; | |
5837 | u64 offset; | |
5838 | u64 num_bytes; | |
c8b97818 CM |
5839 | u8 compression; |
5840 | u8 encryption; | |
5841 | u16 other_encoding; | |
1a40e23b | 5842 | }; |
4313b399 | 5843 | |
1a40e23b ZY |
5844 | static int is_cowonly_root(u64 root_objectid) |
5845 | { | |
5846 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
5847 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
5848 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
5849 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
5850 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
5851 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
5852 | return 1; |
5853 | return 0; | |
5854 | } | |
edbd8d4e | 5855 | |
d397712b | 5856 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5857 | struct btrfs_root *extent_root, |
5858 | struct btrfs_ref_path *ref_path, | |
5859 | int first_time) | |
5860 | { | |
5861 | struct extent_buffer *leaf; | |
5862 | struct btrfs_path *path; | |
5863 | struct btrfs_extent_ref *ref; | |
5864 | struct btrfs_key key; | |
5865 | struct btrfs_key found_key; | |
5866 | u64 bytenr; | |
5867 | u32 nritems; | |
5868 | int level; | |
5869 | int ret = 1; | |
edbd8d4e | 5870 | |
1a40e23b ZY |
5871 | path = btrfs_alloc_path(); |
5872 | if (!path) | |
5873 | return -ENOMEM; | |
bf4ef679 | 5874 | |
1a40e23b ZY |
5875 | if (first_time) { |
5876 | ref_path->lowest_level = -1; | |
5877 | ref_path->current_level = -1; | |
f82d02d9 | 5878 | ref_path->shared_level = -1; |
1a40e23b ZY |
5879 | goto walk_up; |
5880 | } | |
5881 | walk_down: | |
5882 | level = ref_path->current_level - 1; | |
5883 | while (level >= -1) { | |
5884 | u64 parent; | |
5885 | if (level < ref_path->lowest_level) | |
5886 | break; | |
bf4ef679 | 5887 | |
d397712b | 5888 | if (level >= 0) |
1a40e23b | 5889 | bytenr = ref_path->nodes[level]; |
d397712b | 5890 | else |
1a40e23b | 5891 | bytenr = ref_path->extent_start; |
1a40e23b | 5892 | BUG_ON(bytenr == 0); |
bf4ef679 | 5893 | |
1a40e23b ZY |
5894 | parent = ref_path->nodes[level + 1]; |
5895 | ref_path->nodes[level + 1] = 0; | |
5896 | ref_path->current_level = level; | |
5897 | BUG_ON(parent == 0); | |
0ef3e66b | 5898 | |
1a40e23b ZY |
5899 | key.objectid = bytenr; |
5900 | key.offset = parent + 1; | |
5901 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5902 | |
1a40e23b ZY |
5903 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5904 | if (ret < 0) | |
edbd8d4e | 5905 | goto out; |
1a40e23b | 5906 | BUG_ON(ret == 0); |
7d9eb12c | 5907 | |
1a40e23b ZY |
5908 | leaf = path->nodes[0]; |
5909 | nritems = btrfs_header_nritems(leaf); | |
5910 | if (path->slots[0] >= nritems) { | |
5911 | ret = btrfs_next_leaf(extent_root, path); | |
5912 | if (ret < 0) | |
5913 | goto out; | |
5914 | if (ret > 0) | |
5915 | goto next; | |
5916 | leaf = path->nodes[0]; | |
5917 | } | |
0ef3e66b | 5918 | |
1a40e23b ZY |
5919 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5920 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
5921 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
5922 | if (level < ref_path->shared_level) | |
5923 | ref_path->shared_level = level; | |
1a40e23b | 5924 | goto found; |
f82d02d9 | 5925 | } |
1a40e23b ZY |
5926 | next: |
5927 | level--; | |
5928 | btrfs_release_path(extent_root, path); | |
d899e052 | 5929 | cond_resched(); |
1a40e23b ZY |
5930 | } |
5931 | /* reached lowest level */ | |
5932 | ret = 1; | |
5933 | goto out; | |
5934 | walk_up: | |
5935 | level = ref_path->current_level; | |
5936 | while (level < BTRFS_MAX_LEVEL - 1) { | |
5937 | u64 ref_objectid; | |
d397712b CM |
5938 | |
5939 | if (level >= 0) | |
1a40e23b | 5940 | bytenr = ref_path->nodes[level]; |
d397712b | 5941 | else |
1a40e23b | 5942 | bytenr = ref_path->extent_start; |
d397712b | 5943 | |
1a40e23b | 5944 | BUG_ON(bytenr == 0); |
edbd8d4e | 5945 | |
1a40e23b ZY |
5946 | key.objectid = bytenr; |
5947 | key.offset = 0; | |
5948 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5949 | |
1a40e23b ZY |
5950 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5951 | if (ret < 0) | |
5952 | goto out; | |
edbd8d4e | 5953 | |
1a40e23b ZY |
5954 | leaf = path->nodes[0]; |
5955 | nritems = btrfs_header_nritems(leaf); | |
5956 | if (path->slots[0] >= nritems) { | |
5957 | ret = btrfs_next_leaf(extent_root, path); | |
5958 | if (ret < 0) | |
5959 | goto out; | |
5960 | if (ret > 0) { | |
5961 | /* the extent was freed by someone */ | |
5962 | if (ref_path->lowest_level == level) | |
5963 | goto out; | |
5964 | btrfs_release_path(extent_root, path); | |
5965 | goto walk_down; | |
5966 | } | |
5967 | leaf = path->nodes[0]; | |
5968 | } | |
edbd8d4e | 5969 | |
1a40e23b ZY |
5970 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5971 | if (found_key.objectid != bytenr || | |
5972 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
5973 | /* the extent was freed by someone */ | |
5974 | if (ref_path->lowest_level == level) { | |
5975 | ret = 1; | |
5976 | goto out; | |
5977 | } | |
5978 | btrfs_release_path(extent_root, path); | |
5979 | goto walk_down; | |
5980 | } | |
5981 | found: | |
5982 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
5983 | struct btrfs_extent_ref); | |
5984 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
5985 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
5986 | if (first_time) { | |
5987 | level = (int)ref_objectid; | |
5988 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
5989 | ref_path->lowest_level = level; | |
5990 | ref_path->current_level = level; | |
5991 | ref_path->nodes[level] = bytenr; | |
5992 | } else { | |
5993 | WARN_ON(ref_objectid != level); | |
5994 | } | |
5995 | } else { | |
5996 | WARN_ON(level != -1); | |
5997 | } | |
5998 | first_time = 0; | |
bf4ef679 | 5999 | |
1a40e23b ZY |
6000 | if (ref_path->lowest_level == level) { |
6001 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
6002 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
6003 | } | |
bf4ef679 | 6004 | |
7d9eb12c | 6005 | /* |
1a40e23b ZY |
6006 | * the block is tree root or the block isn't in reference |
6007 | * counted tree. | |
7d9eb12c | 6008 | */ |
1a40e23b ZY |
6009 | if (found_key.objectid == found_key.offset || |
6010 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
6011 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6012 | ref_path->root_generation = | |
6013 | btrfs_ref_generation(leaf, ref); | |
6014 | if (level < 0) { | |
6015 | /* special reference from the tree log */ | |
6016 | ref_path->nodes[0] = found_key.offset; | |
6017 | ref_path->current_level = 0; | |
6018 | } | |
6019 | ret = 0; | |
6020 | goto out; | |
6021 | } | |
7d9eb12c | 6022 | |
1a40e23b ZY |
6023 | level++; |
6024 | BUG_ON(ref_path->nodes[level] != 0); | |
6025 | ref_path->nodes[level] = found_key.offset; | |
6026 | ref_path->current_level = level; | |
bf4ef679 | 6027 | |
1a40e23b ZY |
6028 | /* |
6029 | * the reference was created in the running transaction, | |
6030 | * no need to continue walking up. | |
6031 | */ | |
6032 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
6033 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6034 | ref_path->root_generation = | |
6035 | btrfs_ref_generation(leaf, ref); | |
6036 | ret = 0; | |
6037 | goto out; | |
7d9eb12c CM |
6038 | } |
6039 | ||
1a40e23b | 6040 | btrfs_release_path(extent_root, path); |
d899e052 | 6041 | cond_resched(); |
7d9eb12c | 6042 | } |
1a40e23b ZY |
6043 | /* reached max tree level, but no tree root found. */ |
6044 | BUG(); | |
edbd8d4e | 6045 | out: |
1a40e23b ZY |
6046 | btrfs_free_path(path); |
6047 | return ret; | |
edbd8d4e CM |
6048 | } |
6049 | ||
1a40e23b ZY |
6050 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
6051 | struct btrfs_root *extent_root, | |
6052 | struct btrfs_ref_path *ref_path, | |
6053 | u64 extent_start) | |
a061fc8d | 6054 | { |
1a40e23b ZY |
6055 | memset(ref_path, 0, sizeof(*ref_path)); |
6056 | ref_path->extent_start = extent_start; | |
a061fc8d | 6057 | |
1a40e23b | 6058 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
6059 | } |
6060 | ||
1a40e23b ZY |
6061 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
6062 | struct btrfs_root *extent_root, | |
6063 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 6064 | { |
1a40e23b ZY |
6065 | return __next_ref_path(trans, extent_root, ref_path, 0); |
6066 | } | |
6067 | ||
d397712b | 6068 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
6069 | struct btrfs_key *extent_key, |
6070 | u64 offset, int no_fragment, | |
6071 | struct disk_extent **extents, | |
6072 | int *nr_extents) | |
6073 | { | |
6074 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
6075 | struct btrfs_path *path; | |
6076 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 6077 | struct extent_buffer *leaf; |
1a40e23b ZY |
6078 | struct disk_extent *exts = *extents; |
6079 | struct btrfs_key found_key; | |
6080 | u64 cur_pos; | |
6081 | u64 last_byte; | |
edbd8d4e | 6082 | u32 nritems; |
1a40e23b ZY |
6083 | int nr = 0; |
6084 | int max = *nr_extents; | |
6085 | int ret; | |
edbd8d4e | 6086 | |
1a40e23b ZY |
6087 | WARN_ON(!no_fragment && *extents); |
6088 | if (!exts) { | |
6089 | max = 1; | |
6090 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
6091 | if (!exts) | |
6092 | return -ENOMEM; | |
a061fc8d | 6093 | } |
edbd8d4e | 6094 | |
1a40e23b ZY |
6095 | path = btrfs_alloc_path(); |
6096 | BUG_ON(!path); | |
edbd8d4e | 6097 | |
1a40e23b ZY |
6098 | cur_pos = extent_key->objectid - offset; |
6099 | last_byte = extent_key->objectid + extent_key->offset; | |
6100 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
6101 | cur_pos, 0); | |
6102 | if (ret < 0) | |
6103 | goto out; | |
6104 | if (ret > 0) { | |
6105 | ret = -ENOENT; | |
6106 | goto out; | |
6107 | } | |
edbd8d4e | 6108 | |
1a40e23b | 6109 | while (1) { |
edbd8d4e CM |
6110 | leaf = path->nodes[0]; |
6111 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
6112 | if (path->slots[0] >= nritems) { |
6113 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
6114 | if (ret < 0) |
6115 | goto out; | |
1a40e23b ZY |
6116 | if (ret > 0) |
6117 | break; | |
bf4ef679 | 6118 | leaf = path->nodes[0]; |
a061fc8d | 6119 | } |
edbd8d4e CM |
6120 | |
6121 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
6122 | if (found_key.offset != cur_pos || |
6123 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
6124 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
6125 | break; |
6126 | ||
1a40e23b ZY |
6127 | fi = btrfs_item_ptr(leaf, path->slots[0], |
6128 | struct btrfs_file_extent_item); | |
6129 | if (btrfs_file_extent_type(leaf, fi) != | |
6130 | BTRFS_FILE_EXTENT_REG || | |
6131 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 6132 | break; |
1a40e23b ZY |
6133 | |
6134 | if (nr == max) { | |
6135 | struct disk_extent *old = exts; | |
6136 | max *= 2; | |
6137 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
6138 | memcpy(exts, old, sizeof(*exts) * nr); | |
6139 | if (old != *extents) | |
6140 | kfree(old); | |
a061fc8d | 6141 | } |
edbd8d4e | 6142 | |
1a40e23b ZY |
6143 | exts[nr].disk_bytenr = |
6144 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
6145 | exts[nr].disk_num_bytes = | |
6146 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6147 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
6148 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
6149 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
6150 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
6151 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
6152 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
6153 | fi); | |
d899e052 YZ |
6154 | BUG_ON(exts[nr].offset > 0); |
6155 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
6156 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 6157 | |
1a40e23b ZY |
6158 | cur_pos += exts[nr].num_bytes; |
6159 | nr++; | |
6160 | ||
6161 | if (cur_pos + offset >= last_byte) | |
6162 | break; | |
6163 | ||
6164 | if (no_fragment) { | |
6165 | ret = 1; | |
edbd8d4e | 6166 | goto out; |
1a40e23b ZY |
6167 | } |
6168 | path->slots[0]++; | |
6169 | } | |
6170 | ||
1f80e4db | 6171 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
6172 | if (cur_pos + offset < last_byte) { |
6173 | ret = -ENOENT; | |
6174 | goto out; | |
edbd8d4e CM |
6175 | } |
6176 | ret = 0; | |
6177 | out: | |
1a40e23b ZY |
6178 | btrfs_free_path(path); |
6179 | if (ret) { | |
6180 | if (exts != *extents) | |
6181 | kfree(exts); | |
6182 | } else { | |
6183 | *extents = exts; | |
6184 | *nr_extents = nr; | |
6185 | } | |
6186 | return ret; | |
6187 | } | |
6188 | ||
d397712b | 6189 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6190 | struct btrfs_root *root, |
6191 | struct btrfs_path *path, | |
6192 | struct btrfs_key *extent_key, | |
6193 | struct btrfs_key *leaf_key, | |
6194 | struct btrfs_ref_path *ref_path, | |
6195 | struct disk_extent *new_extents, | |
6196 | int nr_extents) | |
6197 | { | |
6198 | struct extent_buffer *leaf; | |
6199 | struct btrfs_file_extent_item *fi; | |
6200 | struct inode *inode = NULL; | |
6201 | struct btrfs_key key; | |
6202 | u64 lock_start = 0; | |
6203 | u64 lock_end = 0; | |
6204 | u64 num_bytes; | |
6205 | u64 ext_offset; | |
86288a19 | 6206 | u64 search_end = (u64)-1; |
1a40e23b | 6207 | u32 nritems; |
3bb1a1bc | 6208 | int nr_scaned = 0; |
1a40e23b | 6209 | int extent_locked = 0; |
d899e052 | 6210 | int extent_type; |
1a40e23b ZY |
6211 | int ret; |
6212 | ||
3bb1a1bc | 6213 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 6214 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
6215 | if (key.objectid < ref_path->owner_objectid || |
6216 | (key.objectid == ref_path->owner_objectid && | |
6217 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
6218 | key.objectid = ref_path->owner_objectid; | |
6219 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6220 | key.offset = 0; | |
6221 | } | |
1a40e23b ZY |
6222 | } |
6223 | ||
6224 | while (1) { | |
6225 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
6226 | if (ret < 0) | |
6227 | goto out; | |
6228 | ||
6229 | leaf = path->nodes[0]; | |
6230 | nritems = btrfs_header_nritems(leaf); | |
6231 | next: | |
6232 | if (extent_locked && ret > 0) { | |
6233 | /* | |
6234 | * the file extent item was modified by someone | |
6235 | * before the extent got locked. | |
6236 | */ | |
1a40e23b ZY |
6237 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6238 | lock_end, GFP_NOFS); | |
6239 | extent_locked = 0; | |
6240 | } | |
6241 | ||
6242 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 6243 | if (++nr_scaned > 2) |
1a40e23b ZY |
6244 | break; |
6245 | ||
6246 | BUG_ON(extent_locked); | |
6247 | ret = btrfs_next_leaf(root, path); | |
6248 | if (ret < 0) | |
6249 | goto out; | |
6250 | if (ret > 0) | |
6251 | break; | |
6252 | leaf = path->nodes[0]; | |
6253 | nritems = btrfs_header_nritems(leaf); | |
6254 | } | |
6255 | ||
6256 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
6257 | ||
6258 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
6259 | if ((key.objectid > ref_path->owner_objectid) || | |
6260 | (key.objectid == ref_path->owner_objectid && | |
6261 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 6262 | key.offset >= search_end) |
1a40e23b ZY |
6263 | break; |
6264 | } | |
6265 | ||
6266 | if (inode && key.objectid != inode->i_ino) { | |
6267 | BUG_ON(extent_locked); | |
6268 | btrfs_release_path(root, path); | |
6269 | mutex_unlock(&inode->i_mutex); | |
6270 | iput(inode); | |
6271 | inode = NULL; | |
6272 | continue; | |
6273 | } | |
6274 | ||
6275 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
6276 | path->slots[0]++; | |
6277 | ret = 1; | |
6278 | goto next; | |
6279 | } | |
6280 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
6281 | struct btrfs_file_extent_item); | |
d899e052 YZ |
6282 | extent_type = btrfs_file_extent_type(leaf, fi); |
6283 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
6284 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
6285 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
6286 | extent_key->objectid)) { | |
6287 | path->slots[0]++; | |
6288 | ret = 1; | |
6289 | goto next; | |
6290 | } | |
6291 | ||
6292 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
6293 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
6294 | ||
86288a19 YZ |
6295 | if (search_end == (u64)-1) { |
6296 | search_end = key.offset - ext_offset + | |
6297 | btrfs_file_extent_ram_bytes(leaf, fi); | |
6298 | } | |
1a40e23b ZY |
6299 | |
6300 | if (!extent_locked) { | |
6301 | lock_start = key.offset; | |
6302 | lock_end = lock_start + num_bytes - 1; | |
6303 | } else { | |
6643558d YZ |
6304 | if (lock_start > key.offset || |
6305 | lock_end + 1 < key.offset + num_bytes) { | |
6306 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
6307 | lock_start, lock_end, GFP_NOFS); | |
6308 | extent_locked = 0; | |
6309 | } | |
1a40e23b ZY |
6310 | } |
6311 | ||
6312 | if (!inode) { | |
6313 | btrfs_release_path(root, path); | |
6314 | ||
6315 | inode = btrfs_iget_locked(root->fs_info->sb, | |
6316 | key.objectid, root); | |
6317 | if (inode->i_state & I_NEW) { | |
6318 | BTRFS_I(inode)->root = root; | |
6319 | BTRFS_I(inode)->location.objectid = | |
6320 | key.objectid; | |
6321 | BTRFS_I(inode)->location.type = | |
6322 | BTRFS_INODE_ITEM_KEY; | |
6323 | BTRFS_I(inode)->location.offset = 0; | |
6324 | btrfs_read_locked_inode(inode); | |
6325 | unlock_new_inode(inode); | |
6326 | } | |
6327 | /* | |
6328 | * some code call btrfs_commit_transaction while | |
6329 | * holding the i_mutex, so we can't use mutex_lock | |
6330 | * here. | |
6331 | */ | |
6332 | if (is_bad_inode(inode) || | |
6333 | !mutex_trylock(&inode->i_mutex)) { | |
6334 | iput(inode); | |
6335 | inode = NULL; | |
6336 | key.offset = (u64)-1; | |
6337 | goto skip; | |
6338 | } | |
6339 | } | |
6340 | ||
6341 | if (!extent_locked) { | |
6342 | struct btrfs_ordered_extent *ordered; | |
6343 | ||
6344 | btrfs_release_path(root, path); | |
6345 | ||
6346 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
6347 | lock_end, GFP_NOFS); | |
6348 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
6349 | lock_end); | |
6350 | if (ordered && | |
6351 | ordered->file_offset <= lock_end && | |
6352 | ordered->file_offset + ordered->len > lock_start) { | |
6353 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
6354 | lock_start, lock_end, GFP_NOFS); | |
6355 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6356 | btrfs_put_ordered_extent(ordered); | |
6357 | key.offset += num_bytes; | |
6358 | goto skip; | |
6359 | } | |
6360 | if (ordered) | |
6361 | btrfs_put_ordered_extent(ordered); | |
6362 | ||
1a40e23b ZY |
6363 | extent_locked = 1; |
6364 | continue; | |
6365 | } | |
6366 | ||
6367 | if (nr_extents == 1) { | |
6368 | /* update extent pointer in place */ | |
1a40e23b ZY |
6369 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
6370 | new_extents[0].disk_bytenr); | |
6371 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6372 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
6373 | btrfs_mark_buffer_dirty(leaf); |
6374 | ||
6375 | btrfs_drop_extent_cache(inode, key.offset, | |
6376 | key.offset + num_bytes - 1, 0); | |
6377 | ||
6378 | ret = btrfs_inc_extent_ref(trans, root, | |
6379 | new_extents[0].disk_bytenr, | |
6380 | new_extents[0].disk_num_bytes, | |
6381 | leaf->start, | |
6382 | root->root_key.objectid, | |
6383 | trans->transid, | |
3bb1a1bc | 6384 | key.objectid); |
1a40e23b ZY |
6385 | BUG_ON(ret); |
6386 | ||
6387 | ret = btrfs_free_extent(trans, root, | |
6388 | extent_key->objectid, | |
6389 | extent_key->offset, | |
6390 | leaf->start, | |
6391 | btrfs_header_owner(leaf), | |
6392 | btrfs_header_generation(leaf), | |
3bb1a1bc | 6393 | key.objectid, 0); |
1a40e23b ZY |
6394 | BUG_ON(ret); |
6395 | ||
6396 | btrfs_release_path(root, path); | |
6397 | key.offset += num_bytes; | |
6398 | } else { | |
d899e052 YZ |
6399 | BUG_ON(1); |
6400 | #if 0 | |
1a40e23b ZY |
6401 | u64 alloc_hint; |
6402 | u64 extent_len; | |
6403 | int i; | |
6404 | /* | |
6405 | * drop old extent pointer at first, then insert the | |
6406 | * new pointers one bye one | |
6407 | */ | |
6408 | btrfs_release_path(root, path); | |
6409 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
6410 | key.offset + num_bytes, | |
6411 | key.offset, &alloc_hint); | |
6412 | BUG_ON(ret); | |
6413 | ||
6414 | for (i = 0; i < nr_extents; i++) { | |
6415 | if (ext_offset >= new_extents[i].num_bytes) { | |
6416 | ext_offset -= new_extents[i].num_bytes; | |
6417 | continue; | |
6418 | } | |
6419 | extent_len = min(new_extents[i].num_bytes - | |
6420 | ext_offset, num_bytes); | |
6421 | ||
6422 | ret = btrfs_insert_empty_item(trans, root, | |
6423 | path, &key, | |
6424 | sizeof(*fi)); | |
6425 | BUG_ON(ret); | |
6426 | ||
6427 | leaf = path->nodes[0]; | |
6428 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
6429 | struct btrfs_file_extent_item); | |
6430 | btrfs_set_file_extent_generation(leaf, fi, | |
6431 | trans->transid); | |
6432 | btrfs_set_file_extent_type(leaf, fi, | |
6433 | BTRFS_FILE_EXTENT_REG); | |
6434 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
6435 | new_extents[i].disk_bytenr); | |
6436 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6437 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
6438 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
6439 | new_extents[i].ram_bytes); | |
6440 | ||
6441 | btrfs_set_file_extent_compression(leaf, fi, | |
6442 | new_extents[i].compression); | |
6443 | btrfs_set_file_extent_encryption(leaf, fi, | |
6444 | new_extents[i].encryption); | |
6445 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
6446 | new_extents[i].other_encoding); | |
6447 | ||
1a40e23b ZY |
6448 | btrfs_set_file_extent_num_bytes(leaf, fi, |
6449 | extent_len); | |
6450 | ext_offset += new_extents[i].offset; | |
6451 | btrfs_set_file_extent_offset(leaf, fi, | |
6452 | ext_offset); | |
6453 | btrfs_mark_buffer_dirty(leaf); | |
6454 | ||
6455 | btrfs_drop_extent_cache(inode, key.offset, | |
6456 | key.offset + extent_len - 1, 0); | |
6457 | ||
6458 | ret = btrfs_inc_extent_ref(trans, root, | |
6459 | new_extents[i].disk_bytenr, | |
6460 | new_extents[i].disk_num_bytes, | |
6461 | leaf->start, | |
6462 | root->root_key.objectid, | |
3bb1a1bc | 6463 | trans->transid, key.objectid); |
1a40e23b ZY |
6464 | BUG_ON(ret); |
6465 | btrfs_release_path(root, path); | |
6466 | ||
a76a3cd4 | 6467 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
6468 | |
6469 | ext_offset = 0; | |
6470 | num_bytes -= extent_len; | |
6471 | key.offset += extent_len; | |
6472 | ||
6473 | if (num_bytes == 0) | |
6474 | break; | |
6475 | } | |
6476 | BUG_ON(i >= nr_extents); | |
d899e052 | 6477 | #endif |
1a40e23b ZY |
6478 | } |
6479 | ||
6480 | if (extent_locked) { | |
1a40e23b ZY |
6481 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6482 | lock_end, GFP_NOFS); | |
6483 | extent_locked = 0; | |
6484 | } | |
6485 | skip: | |
6486 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 6487 | key.offset >= search_end) |
1a40e23b ZY |
6488 | break; |
6489 | ||
6490 | cond_resched(); | |
6491 | } | |
6492 | ret = 0; | |
6493 | out: | |
6494 | btrfs_release_path(root, path); | |
6495 | if (inode) { | |
6496 | mutex_unlock(&inode->i_mutex); | |
6497 | if (extent_locked) { | |
1a40e23b ZY |
6498 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6499 | lock_end, GFP_NOFS); | |
6500 | } | |
6501 | iput(inode); | |
6502 | } | |
6503 | return ret; | |
6504 | } | |
6505 | ||
1a40e23b ZY |
6506 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
6507 | struct btrfs_root *root, | |
6508 | struct extent_buffer *buf, u64 orig_start) | |
6509 | { | |
6510 | int level; | |
6511 | int ret; | |
6512 | ||
6513 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
6514 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
6515 | ||
6516 | level = btrfs_header_level(buf); | |
6517 | if (level == 0) { | |
6518 | struct btrfs_leaf_ref *ref; | |
6519 | struct btrfs_leaf_ref *orig_ref; | |
6520 | ||
6521 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
6522 | if (!orig_ref) | |
6523 | return -ENOENT; | |
6524 | ||
6525 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
6526 | if (!ref) { | |
6527 | btrfs_free_leaf_ref(root, orig_ref); | |
6528 | return -ENOMEM; | |
6529 | } | |
6530 | ||
6531 | ref->nritems = orig_ref->nritems; | |
6532 | memcpy(ref->extents, orig_ref->extents, | |
6533 | sizeof(ref->extents[0]) * ref->nritems); | |
6534 | ||
6535 | btrfs_free_leaf_ref(root, orig_ref); | |
6536 | ||
6537 | ref->root_gen = trans->transid; | |
6538 | ref->bytenr = buf->start; | |
6539 | ref->owner = btrfs_header_owner(buf); | |
6540 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 6541 | |
1a40e23b ZY |
6542 | ret = btrfs_add_leaf_ref(root, ref, 0); |
6543 | WARN_ON(ret); | |
6544 | btrfs_free_leaf_ref(root, ref); | |
6545 | } | |
6546 | return 0; | |
6547 | } | |
6548 | ||
d397712b | 6549 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
6550 | struct extent_buffer *leaf, |
6551 | struct btrfs_block_group_cache *group, | |
6552 | struct btrfs_root *target_root) | |
6553 | { | |
6554 | struct btrfs_key key; | |
6555 | struct inode *inode = NULL; | |
6556 | struct btrfs_file_extent_item *fi; | |
6557 | u64 num_bytes; | |
6558 | u64 skip_objectid = 0; | |
6559 | u32 nritems; | |
6560 | u32 i; | |
6561 | ||
6562 | nritems = btrfs_header_nritems(leaf); | |
6563 | for (i = 0; i < nritems; i++) { | |
6564 | btrfs_item_key_to_cpu(leaf, &key, i); | |
6565 | if (key.objectid == skip_objectid || | |
6566 | key.type != BTRFS_EXTENT_DATA_KEY) | |
6567 | continue; | |
6568 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
6569 | if (btrfs_file_extent_type(leaf, fi) == | |
6570 | BTRFS_FILE_EXTENT_INLINE) | |
6571 | continue; | |
6572 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
6573 | continue; | |
6574 | if (!inode || inode->i_ino != key.objectid) { | |
6575 | iput(inode); | |
6576 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
6577 | key.objectid, target_root, 1); | |
6578 | } | |
6579 | if (!inode) { | |
6580 | skip_objectid = key.objectid; | |
6581 | continue; | |
6582 | } | |
6583 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
6584 | ||
6585 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
6586 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
6587 | btrfs_drop_extent_cache(inode, key.offset, |
6588 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
6589 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
6590 | key.offset + num_bytes - 1, GFP_NOFS); | |
6591 | cond_resched(); | |
6592 | } | |
6593 | iput(inode); | |
6594 | return 0; | |
6595 | } | |
6596 | ||
d397712b | 6597 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6598 | struct btrfs_root *root, |
6599 | struct extent_buffer *leaf, | |
6600 | struct btrfs_block_group_cache *group, | |
6601 | struct inode *reloc_inode) | |
6602 | { | |
6603 | struct btrfs_key key; | |
6604 | struct btrfs_key extent_key; | |
6605 | struct btrfs_file_extent_item *fi; | |
6606 | struct btrfs_leaf_ref *ref; | |
6607 | struct disk_extent *new_extent; | |
6608 | u64 bytenr; | |
6609 | u64 num_bytes; | |
6610 | u32 nritems; | |
6611 | u32 i; | |
6612 | int ext_index; | |
6613 | int nr_extent; | |
6614 | int ret; | |
6615 | ||
6616 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
6617 | BUG_ON(!new_extent); | |
6618 | ||
6619 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
6620 | BUG_ON(!ref); | |
6621 | ||
6622 | ext_index = -1; | |
6623 | nritems = btrfs_header_nritems(leaf); | |
6624 | for (i = 0; i < nritems; i++) { | |
6625 | btrfs_item_key_to_cpu(leaf, &key, i); | |
6626 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6627 | continue; | |
6628 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
6629 | if (btrfs_file_extent_type(leaf, fi) == | |
6630 | BTRFS_FILE_EXTENT_INLINE) | |
6631 | continue; | |
6632 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
6633 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6634 | if (bytenr == 0) | |
6635 | continue; | |
6636 | ||
6637 | ext_index++; | |
6638 | if (bytenr >= group->key.objectid + group->key.offset || | |
6639 | bytenr + num_bytes <= group->key.objectid) | |
6640 | continue; | |
6641 | ||
6642 | extent_key.objectid = bytenr; | |
6643 | extent_key.offset = num_bytes; | |
6644 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
6645 | nr_extent = 1; | |
6646 | ret = get_new_locations(reloc_inode, &extent_key, | |
6647 | group->key.objectid, 1, | |
6648 | &new_extent, &nr_extent); | |
6649 | if (ret > 0) | |
6650 | continue; | |
6651 | BUG_ON(ret < 0); | |
6652 | ||
6653 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
6654 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
6655 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
6656 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
6657 | ||
1a40e23b ZY |
6658 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
6659 | new_extent->disk_bytenr); | |
6660 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6661 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
6662 | btrfs_mark_buffer_dirty(leaf); |
6663 | ||
6664 | ret = btrfs_inc_extent_ref(trans, root, | |
6665 | new_extent->disk_bytenr, | |
6666 | new_extent->disk_num_bytes, | |
6667 | leaf->start, | |
6668 | root->root_key.objectid, | |
3bb1a1bc | 6669 | trans->transid, key.objectid); |
1a40e23b | 6670 | BUG_ON(ret); |
56bec294 | 6671 | |
1a40e23b ZY |
6672 | ret = btrfs_free_extent(trans, root, |
6673 | bytenr, num_bytes, leaf->start, | |
6674 | btrfs_header_owner(leaf), | |
6675 | btrfs_header_generation(leaf), | |
3bb1a1bc | 6676 | key.objectid, 0); |
1a40e23b ZY |
6677 | BUG_ON(ret); |
6678 | cond_resched(); | |
6679 | } | |
6680 | kfree(new_extent); | |
6681 | BUG_ON(ext_index + 1 != ref->nritems); | |
6682 | btrfs_free_leaf_ref(root, ref); | |
6683 | return 0; | |
6684 | } | |
6685 | ||
f82d02d9 YZ |
6686 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
6687 | struct btrfs_root *root) | |
1a40e23b ZY |
6688 | { |
6689 | struct btrfs_root *reloc_root; | |
f82d02d9 | 6690 | int ret; |
1a40e23b ZY |
6691 | |
6692 | if (root->reloc_root) { | |
6693 | reloc_root = root->reloc_root; | |
6694 | root->reloc_root = NULL; | |
6695 | list_add(&reloc_root->dead_list, | |
6696 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
6697 | |
6698 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
6699 | reloc_root->node->start); | |
6700 | btrfs_set_root_level(&root->root_item, | |
6701 | btrfs_header_level(reloc_root->node)); | |
6702 | memset(&reloc_root->root_item.drop_progress, 0, | |
6703 | sizeof(struct btrfs_disk_key)); | |
6704 | reloc_root->root_item.drop_level = 0; | |
6705 | ||
6706 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
6707 | &reloc_root->root_key, | |
6708 | &reloc_root->root_item); | |
6709 | BUG_ON(ret); | |
1a40e23b ZY |
6710 | } |
6711 | return 0; | |
6712 | } | |
6713 | ||
6714 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
6715 | { | |
6716 | struct btrfs_trans_handle *trans; | |
6717 | struct btrfs_root *reloc_root; | |
6718 | struct btrfs_root *prev_root = NULL; | |
6719 | struct list_head dead_roots; | |
6720 | int ret; | |
6721 | unsigned long nr; | |
6722 | ||
6723 | INIT_LIST_HEAD(&dead_roots); | |
6724 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
6725 | ||
6726 | while (!list_empty(&dead_roots)) { | |
6727 | reloc_root = list_entry(dead_roots.prev, | |
6728 | struct btrfs_root, dead_list); | |
6729 | list_del_init(&reloc_root->dead_list); | |
6730 | ||
6731 | BUG_ON(reloc_root->commit_root != NULL); | |
6732 | while (1) { | |
6733 | trans = btrfs_join_transaction(root, 1); | |
6734 | BUG_ON(!trans); | |
6735 | ||
6736 | mutex_lock(&root->fs_info->drop_mutex); | |
6737 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
6738 | if (ret != -EAGAIN) | |
6739 | break; | |
6740 | mutex_unlock(&root->fs_info->drop_mutex); | |
6741 | ||
6742 | nr = trans->blocks_used; | |
6743 | ret = btrfs_end_transaction(trans, root); | |
6744 | BUG_ON(ret); | |
6745 | btrfs_btree_balance_dirty(root, nr); | |
6746 | } | |
6747 | ||
6748 | free_extent_buffer(reloc_root->node); | |
6749 | ||
6750 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
6751 | &reloc_root->root_key); | |
6752 | BUG_ON(ret); | |
6753 | mutex_unlock(&root->fs_info->drop_mutex); | |
6754 | ||
6755 | nr = trans->blocks_used; | |
6756 | ret = btrfs_end_transaction(trans, root); | |
6757 | BUG_ON(ret); | |
6758 | btrfs_btree_balance_dirty(root, nr); | |
6759 | ||
6760 | kfree(prev_root); | |
6761 | prev_root = reloc_root; | |
6762 | } | |
6763 | if (prev_root) { | |
6764 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
6765 | kfree(prev_root); | |
6766 | } | |
6767 | return 0; | |
6768 | } | |
6769 | ||
6770 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
6771 | { | |
6772 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
6773 | return 0; | |
6774 | } | |
6775 | ||
6776 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
6777 | { | |
6778 | struct btrfs_root *reloc_root; | |
6779 | struct btrfs_trans_handle *trans; | |
6780 | struct btrfs_key location; | |
6781 | int found; | |
6782 | int ret; | |
6783 | ||
6784 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
6785 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
6786 | BUG_ON(ret); | |
6787 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
6788 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
6789 | ||
6790 | if (found) { | |
6791 | trans = btrfs_start_transaction(root, 1); | |
6792 | BUG_ON(!trans); | |
6793 | ret = btrfs_commit_transaction(trans, root); | |
6794 | BUG_ON(ret); | |
6795 | } | |
6796 | ||
6797 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
6798 | location.offset = (u64)-1; | |
6799 | location.type = BTRFS_ROOT_ITEM_KEY; | |
6800 | ||
6801 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
6802 | BUG_ON(!reloc_root); | |
6803 | btrfs_orphan_cleanup(reloc_root); | |
6804 | return 0; | |
6805 | } | |
6806 | ||
d397712b | 6807 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6808 | struct btrfs_root *root) |
6809 | { | |
6810 | struct btrfs_root *reloc_root; | |
6811 | struct extent_buffer *eb; | |
6812 | struct btrfs_root_item *root_item; | |
6813 | struct btrfs_key root_key; | |
6814 | int ret; | |
6815 | ||
6816 | BUG_ON(!root->ref_cows); | |
6817 | if (root->reloc_root) | |
6818 | return 0; | |
6819 | ||
6820 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
6821 | BUG_ON(!root_item); | |
6822 | ||
6823 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
6824 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
6825 | BUG_ON(ret); | |
6826 | ||
6827 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
6828 | root_key.offset = root->root_key.objectid; | |
6829 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6830 | ||
6831 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
6832 | btrfs_set_root_refs(root_item, 0); | |
6833 | btrfs_set_root_bytenr(root_item, eb->start); | |
6834 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 6835 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
6836 | |
6837 | btrfs_tree_unlock(eb); | |
6838 | free_extent_buffer(eb); | |
6839 | ||
6840 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
6841 | &root_key, root_item); | |
6842 | BUG_ON(ret); | |
6843 | kfree(root_item); | |
6844 | ||
6845 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
6846 | &root_key); | |
6847 | BUG_ON(!reloc_root); | |
6848 | reloc_root->last_trans = trans->transid; | |
6849 | reloc_root->commit_root = NULL; | |
6850 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
6851 | ||
6852 | root->reloc_root = reloc_root; | |
6853 | return 0; | |
6854 | } | |
6855 | ||
6856 | /* | |
6857 | * Core function of space balance. | |
6858 | * | |
6859 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
6860 | * counted roots. There is one reloc tree for each subvol, and all |
6861 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
6862 | * of the latest committed roots of subvols (root->commit_root). | |
6863 | * | |
6864 | * To relocate a tree block referenced by a subvol, there are two steps. | |
6865 | * COW the block through subvol's reloc tree, then update block pointer | |
6866 | * in the subvol to point to the new block. Since all reloc trees share | |
6867 | * same root key objectid, doing special handing for tree blocks owned | |
6868 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
6869 | * we can use the resulting new block directly when the same block is | |
6870 | * required to COW again through other reloc trees. By this way, relocated | |
6871 | * tree blocks are shared between reloc trees, so they are also shared | |
6872 | * between subvols. | |
1a40e23b | 6873 | */ |
d397712b | 6874 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6875 | struct btrfs_root *root, |
6876 | struct btrfs_path *path, | |
6877 | struct btrfs_key *first_key, | |
6878 | struct btrfs_ref_path *ref_path, | |
6879 | struct btrfs_block_group_cache *group, | |
6880 | struct inode *reloc_inode) | |
6881 | { | |
6882 | struct btrfs_root *reloc_root; | |
6883 | struct extent_buffer *eb = NULL; | |
6884 | struct btrfs_key *keys; | |
6885 | u64 *nodes; | |
6886 | int level; | |
f82d02d9 | 6887 | int shared_level; |
1a40e23b | 6888 | int lowest_level = 0; |
1a40e23b ZY |
6889 | int ret; |
6890 | ||
6891 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
6892 | lowest_level = ref_path->owner_objectid; | |
6893 | ||
f82d02d9 | 6894 | if (!root->ref_cows) { |
1a40e23b ZY |
6895 | path->lowest_level = lowest_level; |
6896 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
6897 | BUG_ON(ret < 0); | |
6898 | path->lowest_level = 0; | |
6899 | btrfs_release_path(root, path); | |
6900 | return 0; | |
6901 | } | |
6902 | ||
1a40e23b ZY |
6903 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
6904 | ret = init_reloc_tree(trans, root); | |
6905 | BUG_ON(ret); | |
6906 | reloc_root = root->reloc_root; | |
6907 | ||
f82d02d9 YZ |
6908 | shared_level = ref_path->shared_level; |
6909 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 6910 | |
f82d02d9 YZ |
6911 | keys = ref_path->node_keys; |
6912 | nodes = ref_path->new_nodes; | |
6913 | memset(&keys[shared_level + 1], 0, | |
6914 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
6915 | memset(&nodes[shared_level + 1], 0, | |
6916 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 6917 | |
f82d02d9 YZ |
6918 | if (nodes[lowest_level] == 0) { |
6919 | path->lowest_level = lowest_level; | |
6920 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
6921 | 0, 1); | |
6922 | BUG_ON(ret); | |
6923 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
6924 | eb = path->nodes[level]; | |
6925 | if (!eb || eb == reloc_root->node) | |
6926 | break; | |
6927 | nodes[level] = eb->start; | |
6928 | if (level == 0) | |
6929 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
6930 | else | |
6931 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
6932 | } | |
2b82032c YZ |
6933 | if (nodes[0] && |
6934 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6935 | eb = path->nodes[0]; |
6936 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
6937 | group, reloc_inode); | |
6938 | BUG_ON(ret); | |
6939 | } | |
6940 | btrfs_release_path(reloc_root, path); | |
6941 | } else { | |
1a40e23b | 6942 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 6943 | lowest_level); |
1a40e23b ZY |
6944 | BUG_ON(ret); |
6945 | } | |
6946 | ||
1a40e23b ZY |
6947 | /* |
6948 | * replace tree blocks in the fs tree with tree blocks in | |
6949 | * the reloc tree. | |
6950 | */ | |
6951 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
6952 | BUG_ON(ret < 0); | |
6953 | ||
6954 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6955 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
6956 | 0, 0); | |
6957 | BUG_ON(ret); | |
6958 | extent_buffer_get(path->nodes[0]); | |
6959 | eb = path->nodes[0]; | |
6960 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
6961 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
6962 | BUG_ON(ret); | |
6963 | free_extent_buffer(eb); | |
6964 | } | |
1a40e23b | 6965 | |
f82d02d9 | 6966 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 6967 | path->lowest_level = 0; |
1a40e23b ZY |
6968 | return 0; |
6969 | } | |
6970 | ||
d397712b | 6971 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6972 | struct btrfs_root *root, |
6973 | struct btrfs_path *path, | |
6974 | struct btrfs_key *first_key, | |
6975 | struct btrfs_ref_path *ref_path) | |
6976 | { | |
6977 | int ret; | |
1a40e23b ZY |
6978 | |
6979 | ret = relocate_one_path(trans, root, path, first_key, | |
6980 | ref_path, NULL, NULL); | |
6981 | BUG_ON(ret); | |
6982 | ||
1a40e23b ZY |
6983 | return 0; |
6984 | } | |
6985 | ||
d397712b | 6986 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6987 | struct btrfs_root *extent_root, |
6988 | struct btrfs_path *path, | |
6989 | struct btrfs_key *extent_key) | |
6990 | { | |
6991 | int ret; | |
6992 | ||
1a40e23b ZY |
6993 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
6994 | if (ret) | |
6995 | goto out; | |
6996 | ret = btrfs_del_item(trans, extent_root, path); | |
6997 | out: | |
6998 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
6999 | return ret; |
7000 | } | |
7001 | ||
d397712b | 7002 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
7003 | struct btrfs_ref_path *ref_path) |
7004 | { | |
7005 | struct btrfs_key root_key; | |
7006 | ||
7007 | root_key.objectid = ref_path->root_objectid; | |
7008 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
7009 | if (is_cowonly_root(ref_path->root_objectid)) | |
7010 | root_key.offset = 0; | |
7011 | else | |
7012 | root_key.offset = (u64)-1; | |
7013 | ||
7014 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
7015 | } | |
7016 | ||
d397712b | 7017 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
7018 | struct btrfs_path *path, |
7019 | struct btrfs_key *extent_key, | |
7020 | struct btrfs_block_group_cache *group, | |
7021 | struct inode *reloc_inode, int pass) | |
7022 | { | |
7023 | struct btrfs_trans_handle *trans; | |
7024 | struct btrfs_root *found_root; | |
7025 | struct btrfs_ref_path *ref_path = NULL; | |
7026 | struct disk_extent *new_extents = NULL; | |
7027 | int nr_extents = 0; | |
7028 | int loops; | |
7029 | int ret; | |
7030 | int level; | |
7031 | struct btrfs_key first_key; | |
7032 | u64 prev_block = 0; | |
7033 | ||
1a40e23b ZY |
7034 | |
7035 | trans = btrfs_start_transaction(extent_root, 1); | |
7036 | BUG_ON(!trans); | |
7037 | ||
7038 | if (extent_key->objectid == 0) { | |
7039 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
7040 | goto out; | |
7041 | } | |
7042 | ||
7043 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
7044 | if (!ref_path) { | |
d397712b CM |
7045 | ret = -ENOMEM; |
7046 | goto out; | |
1a40e23b ZY |
7047 | } |
7048 | ||
7049 | for (loops = 0; ; loops++) { | |
7050 | if (loops == 0) { | |
7051 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
7052 | extent_key->objectid); | |
7053 | } else { | |
7054 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
7055 | } | |
7056 | if (ret < 0) | |
7057 | goto out; | |
7058 | if (ret > 0) | |
7059 | break; | |
7060 | ||
7061 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
7062 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
7063 | continue; | |
7064 | ||
7065 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
7066 | BUG_ON(!found_root); | |
7067 | /* | |
7068 | * for reference counted tree, only process reference paths | |
7069 | * rooted at the latest committed root. | |
7070 | */ | |
7071 | if (found_root->ref_cows && | |
7072 | ref_path->root_generation != found_root->root_key.offset) | |
7073 | continue; | |
7074 | ||
7075 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
7076 | if (pass == 0) { | |
7077 | /* | |
7078 | * copy data extents to new locations | |
7079 | */ | |
7080 | u64 group_start = group->key.objectid; | |
7081 | ret = relocate_data_extent(reloc_inode, | |
7082 | extent_key, | |
7083 | group_start); | |
7084 | if (ret < 0) | |
7085 | goto out; | |
7086 | break; | |
7087 | } | |
7088 | level = 0; | |
7089 | } else { | |
7090 | level = ref_path->owner_objectid; | |
7091 | } | |
7092 | ||
7093 | if (prev_block != ref_path->nodes[level]) { | |
7094 | struct extent_buffer *eb; | |
7095 | u64 block_start = ref_path->nodes[level]; | |
7096 | u64 block_size = btrfs_level_size(found_root, level); | |
7097 | ||
7098 | eb = read_tree_block(found_root, block_start, | |
7099 | block_size, 0); | |
7100 | btrfs_tree_lock(eb); | |
7101 | BUG_ON(level != btrfs_header_level(eb)); | |
7102 | ||
7103 | if (level == 0) | |
7104 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
7105 | else | |
7106 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
7107 | ||
7108 | btrfs_tree_unlock(eb); | |
7109 | free_extent_buffer(eb); | |
7110 | prev_block = block_start; | |
7111 | } | |
7112 | ||
24562425 | 7113 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 7114 | btrfs_record_root_in_trans(found_root); |
24562425 | 7115 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
7116 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
7117 | /* | |
7118 | * try to update data extent references while | |
7119 | * keeping metadata shared between snapshots. | |
7120 | */ | |
7121 | if (pass == 1) { | |
7122 | ret = relocate_one_path(trans, found_root, | |
7123 | path, &first_key, ref_path, | |
7124 | group, reloc_inode); | |
7125 | if (ret < 0) | |
7126 | goto out; | |
7127 | continue; | |
7128 | } | |
1a40e23b ZY |
7129 | /* |
7130 | * use fallback method to process the remaining | |
7131 | * references. | |
7132 | */ | |
7133 | if (!new_extents) { | |
7134 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
7135 | new_extents = kmalloc(sizeof(*new_extents), |
7136 | GFP_NOFS); | |
7137 | nr_extents = 1; | |
1a40e23b ZY |
7138 | ret = get_new_locations(reloc_inode, |
7139 | extent_key, | |
d899e052 | 7140 | group_start, 1, |
1a40e23b ZY |
7141 | &new_extents, |
7142 | &nr_extents); | |
d899e052 | 7143 | if (ret) |
1a40e23b ZY |
7144 | goto out; |
7145 | } | |
1a40e23b ZY |
7146 | ret = replace_one_extent(trans, found_root, |
7147 | path, extent_key, | |
7148 | &first_key, ref_path, | |
7149 | new_extents, nr_extents); | |
e4404d6e | 7150 | } else { |
1a40e23b ZY |
7151 | ret = relocate_tree_block(trans, found_root, path, |
7152 | &first_key, ref_path); | |
1a40e23b ZY |
7153 | } |
7154 | if (ret < 0) | |
7155 | goto out; | |
7156 | } | |
7157 | ret = 0; | |
7158 | out: | |
7159 | btrfs_end_transaction(trans, extent_root); | |
7160 | kfree(new_extents); | |
7161 | kfree(ref_path); | |
1a40e23b ZY |
7162 | return ret; |
7163 | } | |
5d4f98a2 | 7164 | #endif |
1a40e23b | 7165 | |
ec44a35c CM |
7166 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7167 | { | |
7168 | u64 num_devices; | |
7169 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
7170 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7171 | ||
2b82032c | 7172 | num_devices = root->fs_info->fs_devices->rw_devices; |
ec44a35c CM |
7173 | if (num_devices == 1) { |
7174 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7175 | stripped = flags & ~stripped; | |
7176 | ||
7177 | /* turn raid0 into single device chunks */ | |
7178 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7179 | return stripped; | |
7180 | ||
7181 | /* turn mirroring into duplication */ | |
7182 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7183 | BTRFS_BLOCK_GROUP_RAID10)) | |
7184 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
7185 | return flags; | |
7186 | } else { | |
7187 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7188 | if (flags & stripped) |
7189 | return flags; | |
7190 | ||
7191 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7192 | stripped = flags & ~stripped; | |
7193 | ||
7194 | /* switch duplicated blocks with raid1 */ | |
7195 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7196 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7197 | ||
7198 | /* turn single device chunks into raid0 */ | |
7199 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
7200 | } | |
7201 | return flags; | |
7202 | } | |
7203 | ||
b2950863 | 7204 | static int __alloc_chunk_for_shrink(struct btrfs_root *root, |
0ef3e66b CM |
7205 | struct btrfs_block_group_cache *shrink_block_group, |
7206 | int force) | |
7207 | { | |
7208 | struct btrfs_trans_handle *trans; | |
7209 | u64 new_alloc_flags; | |
7210 | u64 calc; | |
7211 | ||
c286ac48 | 7212 | spin_lock(&shrink_block_group->lock); |
5d4f98a2 YZ |
7213 | if (btrfs_block_group_used(&shrink_block_group->item) + |
7214 | shrink_block_group->reserved > 0) { | |
c286ac48 | 7215 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 7216 | |
0ef3e66b | 7217 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 7218 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 7219 | |
0ef3e66b CM |
7220 | new_alloc_flags = update_block_group_flags(root, |
7221 | shrink_block_group->flags); | |
7222 | if (new_alloc_flags != shrink_block_group->flags) { | |
7223 | calc = | |
7224 | btrfs_block_group_used(&shrink_block_group->item); | |
7225 | } else { | |
7226 | calc = shrink_block_group->key.offset; | |
7227 | } | |
c286ac48 CM |
7228 | spin_unlock(&shrink_block_group->lock); |
7229 | ||
0ef3e66b CM |
7230 | do_chunk_alloc(trans, root->fs_info->extent_root, |
7231 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 7232 | |
0ef3e66b | 7233 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
7234 | } else |
7235 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
7236 | return 0; |
7237 | } | |
7238 | ||
5d4f98a2 YZ |
7239 | |
7240 | int btrfs_prepare_block_group_relocation(struct btrfs_root *root, | |
7241 | struct btrfs_block_group_cache *group) | |
7242 | ||
7243 | { | |
7244 | __alloc_chunk_for_shrink(root, group, 1); | |
7245 | set_block_group_readonly(group); | |
7246 | return 0; | |
7247 | } | |
7248 | ||
ba1bf481 JB |
7249 | /* |
7250 | * checks to see if its even possible to relocate this block group. | |
7251 | * | |
7252 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7253 | * ok to go ahead and try. | |
7254 | */ | |
7255 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7256 | { |
ba1bf481 JB |
7257 | struct btrfs_block_group_cache *block_group; |
7258 | struct btrfs_space_info *space_info; | |
7259 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7260 | struct btrfs_device *device; | |
7261 | int full = 0; | |
7262 | int ret = 0; | |
1a40e23b | 7263 | |
ba1bf481 | 7264 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7265 | |
ba1bf481 JB |
7266 | /* odd, couldn't find the block group, leave it alone */ |
7267 | if (!block_group) | |
7268 | return -1; | |
1a40e23b | 7269 | |
ba1bf481 JB |
7270 | /* no bytes used, we're good */ |
7271 | if (!btrfs_block_group_used(&block_group->item)) | |
1a40e23b ZY |
7272 | goto out; |
7273 | ||
ba1bf481 JB |
7274 | space_info = block_group->space_info; |
7275 | spin_lock(&space_info->lock); | |
17d217fe | 7276 | |
ba1bf481 | 7277 | full = space_info->full; |
17d217fe | 7278 | |
ba1bf481 JB |
7279 | /* |
7280 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7281 | * relocate it unless we're able to allocate a new chunk below. |
7282 | * | |
7283 | * Otherwise, we need to make sure we have room in the space to handle | |
7284 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7285 | */ |
7ce618db CM |
7286 | if ((space_info->total_bytes != block_group->key.offset) && |
7287 | (space_info->bytes_used + space_info->bytes_reserved + | |
ba1bf481 JB |
7288 | space_info->bytes_pinned + space_info->bytes_readonly + |
7289 | btrfs_block_group_used(&block_group->item) < | |
7ce618db | 7290 | space_info->total_bytes)) { |
ba1bf481 JB |
7291 | spin_unlock(&space_info->lock); |
7292 | goto out; | |
17d217fe | 7293 | } |
ba1bf481 | 7294 | spin_unlock(&space_info->lock); |
ea8c2819 | 7295 | |
ba1bf481 JB |
7296 | /* |
7297 | * ok we don't have enough space, but maybe we have free space on our | |
7298 | * devices to allocate new chunks for relocation, so loop through our | |
7299 | * alloc devices and guess if we have enough space. However, if we | |
7300 | * were marked as full, then we know there aren't enough chunks, and we | |
7301 | * can just return. | |
7302 | */ | |
7303 | ret = -1; | |
7304 | if (full) | |
7305 | goto out; | |
ea8c2819 | 7306 | |
ba1bf481 JB |
7307 | mutex_lock(&root->fs_info->chunk_mutex); |
7308 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7309 | u64 min_free = btrfs_block_group_used(&block_group->item); | |
7310 | u64 dev_offset, max_avail; | |
56bec294 | 7311 | |
ba1bf481 JB |
7312 | /* |
7313 | * check to make sure we can actually find a chunk with enough | |
7314 | * space to fit our block group in. | |
7315 | */ | |
7316 | if (device->total_bytes > device->bytes_used + min_free) { | |
7317 | ret = find_free_dev_extent(NULL, device, min_free, | |
7318 | &dev_offset, &max_avail); | |
7319 | if (!ret) | |
73e48b27 | 7320 | break; |
ba1bf481 | 7321 | ret = -1; |
725c8463 | 7322 | } |
edbd8d4e | 7323 | } |
ba1bf481 | 7324 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7325 | out: |
ba1bf481 | 7326 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7327 | return ret; |
7328 | } | |
7329 | ||
b2950863 CH |
7330 | static int find_first_block_group(struct btrfs_root *root, |
7331 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7332 | { |
925baedd | 7333 | int ret = 0; |
0b86a832 CM |
7334 | struct btrfs_key found_key; |
7335 | struct extent_buffer *leaf; | |
7336 | int slot; | |
edbd8d4e | 7337 | |
0b86a832 CM |
7338 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7339 | if (ret < 0) | |
925baedd CM |
7340 | goto out; |
7341 | ||
d397712b | 7342 | while (1) { |
0b86a832 | 7343 | slot = path->slots[0]; |
edbd8d4e | 7344 | leaf = path->nodes[0]; |
0b86a832 CM |
7345 | if (slot >= btrfs_header_nritems(leaf)) { |
7346 | ret = btrfs_next_leaf(root, path); | |
7347 | if (ret == 0) | |
7348 | continue; | |
7349 | if (ret < 0) | |
925baedd | 7350 | goto out; |
0b86a832 | 7351 | break; |
edbd8d4e | 7352 | } |
0b86a832 | 7353 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7354 | |
0b86a832 | 7355 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7356 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7357 | ret = 0; | |
7358 | goto out; | |
7359 | } | |
0b86a832 | 7360 | path->slots[0]++; |
edbd8d4e | 7361 | } |
0b86a832 | 7362 | ret = -ENOENT; |
925baedd | 7363 | out: |
0b86a832 | 7364 | return ret; |
edbd8d4e CM |
7365 | } |
7366 | ||
1a40e23b ZY |
7367 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7368 | { | |
7369 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7370 | struct btrfs_space_info *space_info; |
11833d66 | 7371 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7372 | struct rb_node *n; |
7373 | ||
11833d66 YZ |
7374 | down_write(&info->extent_commit_sem); |
7375 | while (!list_empty(&info->caching_block_groups)) { | |
7376 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7377 | struct btrfs_caching_control, list); | |
7378 | list_del(&caching_ctl->list); | |
7379 | put_caching_control(caching_ctl); | |
7380 | } | |
7381 | up_write(&info->extent_commit_sem); | |
7382 | ||
1a40e23b ZY |
7383 | spin_lock(&info->block_group_cache_lock); |
7384 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7385 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7386 | cache_node); | |
1a40e23b ZY |
7387 | rb_erase(&block_group->cache_node, |
7388 | &info->block_group_cache_tree); | |
d899e052 YZ |
7389 | spin_unlock(&info->block_group_cache_lock); |
7390 | ||
80eb234a | 7391 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7392 | list_del(&block_group->list); |
80eb234a | 7393 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7394 | |
817d52f8 | 7395 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7396 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7397 | |
7398 | btrfs_remove_free_space_cache(block_group); | |
7399 | ||
d2fb3437 | 7400 | WARN_ON(atomic_read(&block_group->count) != 1); |
1a40e23b | 7401 | kfree(block_group); |
d899e052 YZ |
7402 | |
7403 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7404 | } |
7405 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7406 | |
7407 | /* now that all the block groups are freed, go through and | |
7408 | * free all the space_info structs. This is only called during | |
7409 | * the final stages of unmount, and so we know nobody is | |
7410 | * using them. We call synchronize_rcu() once before we start, | |
7411 | * just to be on the safe side. | |
7412 | */ | |
7413 | synchronize_rcu(); | |
7414 | ||
7415 | while(!list_empty(&info->space_info)) { | |
7416 | space_info = list_entry(info->space_info.next, | |
7417 | struct btrfs_space_info, | |
7418 | list); | |
7419 | ||
7420 | list_del(&space_info->list); | |
7421 | kfree(space_info); | |
7422 | } | |
1a40e23b ZY |
7423 | return 0; |
7424 | } | |
7425 | ||
9078a3e1 CM |
7426 | int btrfs_read_block_groups(struct btrfs_root *root) |
7427 | { | |
7428 | struct btrfs_path *path; | |
7429 | int ret; | |
9078a3e1 | 7430 | struct btrfs_block_group_cache *cache; |
be744175 | 7431 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7432 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7433 | struct btrfs_key key; |
7434 | struct btrfs_key found_key; | |
5f39d397 | 7435 | struct extent_buffer *leaf; |
96b5179d | 7436 | |
be744175 | 7437 | root = info->extent_root; |
9078a3e1 | 7438 | key.objectid = 0; |
0b86a832 | 7439 | key.offset = 0; |
9078a3e1 | 7440 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7441 | path = btrfs_alloc_path(); |
7442 | if (!path) | |
7443 | return -ENOMEM; | |
7444 | ||
d397712b | 7445 | while (1) { |
0b86a832 CM |
7446 | ret = find_first_block_group(root, path, &key); |
7447 | if (ret > 0) { | |
7448 | ret = 0; | |
7449 | goto error; | |
9078a3e1 | 7450 | } |
0b86a832 CM |
7451 | if (ret != 0) |
7452 | goto error; | |
7453 | ||
5f39d397 CM |
7454 | leaf = path->nodes[0]; |
7455 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7456 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7457 | if (!cache) { |
0b86a832 | 7458 | ret = -ENOMEM; |
9078a3e1 CM |
7459 | break; |
7460 | } | |
3e1ad54f | 7461 | |
d2fb3437 | 7462 | atomic_set(&cache->count, 1); |
c286ac48 | 7463 | spin_lock_init(&cache->lock); |
6226cb0a | 7464 | spin_lock_init(&cache->tree_lock); |
817d52f8 | 7465 | cache->fs_info = info; |
0f9dd46c | 7466 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7467 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 JB |
7468 | |
7469 | /* | |
7470 | * we only want to have 32k of ram per block group for keeping | |
7471 | * track of free space, and if we pass 1/2 of that we want to | |
7472 | * start converting things over to using bitmaps | |
7473 | */ | |
7474 | cache->extents_thresh = ((1024 * 32) / 2) / | |
7475 | sizeof(struct btrfs_free_space); | |
7476 | ||
5f39d397 CM |
7477 | read_extent_buffer(leaf, &cache->item, |
7478 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7479 | sizeof(cache->item)); | |
9078a3e1 | 7480 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7481 | |
9078a3e1 CM |
7482 | key.objectid = found_key.objectid + found_key.offset; |
7483 | btrfs_release_path(root, path); | |
0b86a832 | 7484 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7485 | cache->sectorsize = root->sectorsize; |
7486 | ||
817d52f8 JB |
7487 | /* |
7488 | * check for two cases, either we are full, and therefore | |
7489 | * don't need to bother with the caching work since we won't | |
7490 | * find any space, or we are empty, and we can just add all | |
7491 | * the space in and be done with it. This saves us _alot_ of | |
7492 | * time, particularly in the full case. | |
7493 | */ | |
7494 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
1b2da372 | 7495 | exclude_super_stripes(root, cache); |
11833d66 | 7496 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7497 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7498 | free_excluded_extents(root, cache); |
817d52f8 | 7499 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 YZ |
7500 | exclude_super_stripes(root, cache); |
7501 | cache->last_byte_to_unpin = (u64)-1; | |
817d52f8 JB |
7502 | cache->cached = BTRFS_CACHE_FINISHED; |
7503 | add_new_free_space(cache, root->fs_info, | |
7504 | found_key.objectid, | |
7505 | found_key.objectid + | |
7506 | found_key.offset); | |
11833d66 | 7507 | free_excluded_extents(root, cache); |
817d52f8 | 7508 | } |
96b5179d | 7509 | |
6324fbf3 CM |
7510 | ret = update_space_info(info, cache->flags, found_key.offset, |
7511 | btrfs_block_group_used(&cache->item), | |
7512 | &space_info); | |
7513 | BUG_ON(ret); | |
7514 | cache->space_info = space_info; | |
1b2da372 JB |
7515 | spin_lock(&cache->space_info->lock); |
7516 | cache->space_info->bytes_super += cache->bytes_super; | |
7517 | spin_unlock(&cache->space_info->lock); | |
7518 | ||
80eb234a JB |
7519 | down_write(&space_info->groups_sem); |
7520 | list_add_tail(&cache->list, &space_info->block_groups); | |
7521 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
7522 | |
7523 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7524 | BUG_ON(ret); | |
75ccf47d CM |
7525 | |
7526 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c YZ |
7527 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
7528 | set_block_group_readonly(cache); | |
9078a3e1 | 7529 | } |
0b86a832 CM |
7530 | ret = 0; |
7531 | error: | |
9078a3e1 | 7532 | btrfs_free_path(path); |
0b86a832 | 7533 | return ret; |
9078a3e1 | 7534 | } |
6324fbf3 CM |
7535 | |
7536 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7537 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7538 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7539 | u64 size) |
7540 | { | |
7541 | int ret; | |
6324fbf3 CM |
7542 | struct btrfs_root *extent_root; |
7543 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7544 | |
7545 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7546 | |
12fcfd22 | 7547 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7548 | |
8f18cf13 | 7549 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7550 | if (!cache) |
7551 | return -ENOMEM; | |
7552 | ||
e17cade2 | 7553 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7554 | cache->key.offset = size; |
d2fb3437 | 7555 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 JB |
7556 | cache->sectorsize = root->sectorsize; |
7557 | ||
7558 | /* | |
7559 | * we only want to have 32k of ram per block group for keeping track | |
7560 | * of free space, and if we pass 1/2 of that we want to start | |
7561 | * converting things over to using bitmaps | |
7562 | */ | |
7563 | cache->extents_thresh = ((1024 * 32) / 2) / | |
7564 | sizeof(struct btrfs_free_space); | |
d2fb3437 | 7565 | atomic_set(&cache->count, 1); |
c286ac48 | 7566 | spin_lock_init(&cache->lock); |
6226cb0a | 7567 | spin_lock_init(&cache->tree_lock); |
0f9dd46c | 7568 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7569 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7570 | |
6324fbf3 | 7571 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7572 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7573 | cache->flags = type; | |
7574 | btrfs_set_block_group_flags(&cache->item, type); | |
7575 | ||
11833d66 | 7576 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7577 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7578 | exclude_super_stripes(root, cache); |
96303081 | 7579 | |
817d52f8 JB |
7580 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7581 | chunk_offset + size); | |
7582 | ||
11833d66 YZ |
7583 | free_excluded_extents(root, cache); |
7584 | ||
6324fbf3 CM |
7585 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7586 | &cache->space_info); | |
7587 | BUG_ON(ret); | |
1b2da372 JB |
7588 | |
7589 | spin_lock(&cache->space_info->lock); | |
7590 | cache->space_info->bytes_super += cache->bytes_super; | |
7591 | spin_unlock(&cache->space_info->lock); | |
7592 | ||
80eb234a JB |
7593 | down_write(&cache->space_info->groups_sem); |
7594 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
7595 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 7596 | |
0f9dd46c JB |
7597 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7598 | BUG_ON(ret); | |
c286ac48 | 7599 | |
6324fbf3 CM |
7600 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7601 | sizeof(cache->item)); | |
7602 | BUG_ON(ret); | |
7603 | ||
d18a2c44 | 7604 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7605 | |
6324fbf3 CM |
7606 | return 0; |
7607 | } | |
1a40e23b ZY |
7608 | |
7609 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
7610 | struct btrfs_root *root, u64 group_start) | |
7611 | { | |
7612 | struct btrfs_path *path; | |
7613 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7614 | struct btrfs_free_cluster *cluster; |
1a40e23b ZY |
7615 | struct btrfs_key key; |
7616 | int ret; | |
7617 | ||
1a40e23b ZY |
7618 | root = root->fs_info->extent_root; |
7619 | ||
7620 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7621 | BUG_ON(!block_group); | |
c146afad | 7622 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
7623 | |
7624 | memcpy(&key, &block_group->key, sizeof(key)); | |
7625 | ||
44fb5511 CM |
7626 | /* make sure this block group isn't part of an allocation cluster */ |
7627 | cluster = &root->fs_info->data_alloc_cluster; | |
7628 | spin_lock(&cluster->refill_lock); | |
7629 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7630 | spin_unlock(&cluster->refill_lock); | |
7631 | ||
7632 | /* | |
7633 | * make sure this block group isn't part of a metadata | |
7634 | * allocation cluster | |
7635 | */ | |
7636 | cluster = &root->fs_info->meta_alloc_cluster; | |
7637 | spin_lock(&cluster->refill_lock); | |
7638 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7639 | spin_unlock(&cluster->refill_lock); | |
7640 | ||
1a40e23b ZY |
7641 | path = btrfs_alloc_path(); |
7642 | BUG_ON(!path); | |
7643 | ||
3dfdb934 | 7644 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7645 | rb_erase(&block_group->cache_node, |
7646 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7647 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7648 | |
80eb234a | 7649 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7650 | /* |
7651 | * we must use list_del_init so people can check to see if they | |
7652 | * are still on the list after taking the semaphore | |
7653 | */ | |
7654 | list_del_init(&block_group->list); | |
80eb234a | 7655 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7656 | |
817d52f8 | 7657 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7658 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7659 | |
7660 | btrfs_remove_free_space_cache(block_group); | |
7661 | ||
c146afad YZ |
7662 | spin_lock(&block_group->space_info->lock); |
7663 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7664 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
7665 | spin_unlock(&block_group->space_info->lock); | |
283bb197 CM |
7666 | |
7667 | btrfs_clear_space_info_full(root->fs_info); | |
c146afad | 7668 | |
fa9c0d79 CM |
7669 | btrfs_put_block_group(block_group); |
7670 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7671 | |
7672 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7673 | if (ret > 0) | |
7674 | ret = -EIO; | |
7675 | if (ret < 0) | |
7676 | goto out; | |
7677 | ||
7678 | ret = btrfs_del_item(trans, root, path); | |
7679 | out: | |
7680 | btrfs_free_path(path); | |
7681 | return ret; | |
7682 | } |