]>
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> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
4b4e25f2 | 27 | #include "compat.h" |
74493f7a | 28 | #include "hash.h" |
fec577fb CM |
29 | #include "ctree.h" |
30 | #include "disk-io.h" | |
31 | #include "print-tree.h" | |
e089f05c | 32 | #include "transaction.h" |
0b86a832 | 33 | #include "volumes.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
fec577fb | 36 | |
0e4f8f88 CM |
37 | /* control flags for do_chunk_alloc's force field |
38 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk | |
39 | * if we really need one. | |
40 | * | |
41 | * CHUNK_ALLOC_FORCE means it must try to allocate one | |
42 | * | |
43 | * CHUNK_ALLOC_LIMITED means to only try and allocate one | |
44 | * if we have very few chunks already allocated. This is | |
45 | * used as part of the clustering code to help make sure | |
46 | * we have a good pool of storage to cluster in, without | |
47 | * filling the FS with empty chunks | |
48 | * | |
49 | */ | |
50 | enum { | |
51 | CHUNK_ALLOC_NO_FORCE = 0, | |
52 | CHUNK_ALLOC_FORCE = 1, | |
53 | CHUNK_ALLOC_LIMITED = 2, | |
54 | }; | |
55 | ||
fb25e914 JB |
56 | /* |
57 | * Control how reservations are dealt with. | |
58 | * | |
59 | * RESERVE_FREE - freeing a reservation. | |
60 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
61 | * ENOSPC accounting | |
62 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
63 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
64 | */ | |
65 | enum { | |
66 | RESERVE_FREE = 0, | |
67 | RESERVE_ALLOC = 1, | |
68 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
69 | }; | |
70 | ||
f3465ca4 JB |
71 | static int update_block_group(struct btrfs_trans_handle *trans, |
72 | struct btrfs_root *root, | |
f0486c68 | 73 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
74 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
75 | struct btrfs_root *root, | |
76 | u64 bytenr, u64 num_bytes, u64 parent, | |
77 | u64 root_objectid, u64 owner_objectid, | |
78 | u64 owner_offset, int refs_to_drop, | |
79 | struct btrfs_delayed_extent_op *extra_op); | |
80 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
81 | struct extent_buffer *leaf, | |
82 | struct btrfs_extent_item *ei); | |
83 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84 | struct btrfs_root *root, | |
85 | u64 parent, u64 root_objectid, | |
86 | u64 flags, u64 owner, u64 offset, | |
87 | struct btrfs_key *ins, int ref_mod); | |
88 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
89 | struct btrfs_root *root, | |
90 | u64 parent, u64 root_objectid, | |
91 | u64 flags, struct btrfs_disk_key *key, | |
92 | int level, struct btrfs_key *ins); | |
6a63209f JB |
93 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
94 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
95 | u64 flags, int force); | |
11833d66 YZ |
96 | static int find_next_key(struct btrfs_path *path, int level, |
97 | struct btrfs_key *key); | |
9ed74f2d JB |
98 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
99 | int dump_block_groups); | |
fb25e914 JB |
100 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
101 | u64 num_bytes, int reserve); | |
6a63209f | 102 | |
817d52f8 JB |
103 | static noinline int |
104 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
105 | { | |
106 | smp_mb(); | |
107 | return cache->cached == BTRFS_CACHE_FINISHED; | |
108 | } | |
109 | ||
0f9dd46c JB |
110 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
111 | { | |
112 | return (cache->flags & bits) == bits; | |
113 | } | |
114 | ||
62a45b60 | 115 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
116 | { |
117 | atomic_inc(&cache->count); | |
118 | } | |
119 | ||
120 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
121 | { | |
f0486c68 YZ |
122 | if (atomic_dec_and_test(&cache->count)) { |
123 | WARN_ON(cache->pinned > 0); | |
124 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 125 | kfree(cache->free_space_ctl); |
11dfe35a | 126 | kfree(cache); |
f0486c68 | 127 | } |
11dfe35a JB |
128 | } |
129 | ||
0f9dd46c JB |
130 | /* |
131 | * this adds the block group to the fs_info rb tree for the block group | |
132 | * cache | |
133 | */ | |
b2950863 | 134 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
135 | struct btrfs_block_group_cache *block_group) |
136 | { | |
137 | struct rb_node **p; | |
138 | struct rb_node *parent = NULL; | |
139 | struct btrfs_block_group_cache *cache; | |
140 | ||
141 | spin_lock(&info->block_group_cache_lock); | |
142 | p = &info->block_group_cache_tree.rb_node; | |
143 | ||
144 | while (*p) { | |
145 | parent = *p; | |
146 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
147 | cache_node); | |
148 | if (block_group->key.objectid < cache->key.objectid) { | |
149 | p = &(*p)->rb_left; | |
150 | } else if (block_group->key.objectid > cache->key.objectid) { | |
151 | p = &(*p)->rb_right; | |
152 | } else { | |
153 | spin_unlock(&info->block_group_cache_lock); | |
154 | return -EEXIST; | |
155 | } | |
156 | } | |
157 | ||
158 | rb_link_node(&block_group->cache_node, parent, p); | |
159 | rb_insert_color(&block_group->cache_node, | |
160 | &info->block_group_cache_tree); | |
161 | spin_unlock(&info->block_group_cache_lock); | |
162 | ||
163 | return 0; | |
164 | } | |
165 | ||
166 | /* | |
167 | * This will return the block group at or after bytenr if contains is 0, else | |
168 | * it will return the block group that contains the bytenr | |
169 | */ | |
170 | static struct btrfs_block_group_cache * | |
171 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
172 | int contains) | |
173 | { | |
174 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
175 | struct rb_node *n; | |
176 | u64 end, start; | |
177 | ||
178 | spin_lock(&info->block_group_cache_lock); | |
179 | n = info->block_group_cache_tree.rb_node; | |
180 | ||
181 | while (n) { | |
182 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
183 | cache_node); | |
184 | end = cache->key.objectid + cache->key.offset - 1; | |
185 | start = cache->key.objectid; | |
186 | ||
187 | if (bytenr < start) { | |
188 | if (!contains && (!ret || start < ret->key.objectid)) | |
189 | ret = cache; | |
190 | n = n->rb_left; | |
191 | } else if (bytenr > start) { | |
192 | if (contains && bytenr <= end) { | |
193 | ret = cache; | |
194 | break; | |
195 | } | |
196 | n = n->rb_right; | |
197 | } else { | |
198 | ret = cache; | |
199 | break; | |
200 | } | |
201 | } | |
d2fb3437 | 202 | if (ret) |
11dfe35a | 203 | btrfs_get_block_group(ret); |
0f9dd46c JB |
204 | spin_unlock(&info->block_group_cache_lock); |
205 | ||
206 | return ret; | |
207 | } | |
208 | ||
11833d66 YZ |
209 | static int add_excluded_extent(struct btrfs_root *root, |
210 | u64 start, u64 num_bytes) | |
817d52f8 | 211 | { |
11833d66 YZ |
212 | u64 end = start + num_bytes - 1; |
213 | set_extent_bits(&root->fs_info->freed_extents[0], | |
214 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
215 | set_extent_bits(&root->fs_info->freed_extents[1], | |
216 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
217 | return 0; | |
218 | } | |
817d52f8 | 219 | |
11833d66 YZ |
220 | static void free_excluded_extents(struct btrfs_root *root, |
221 | struct btrfs_block_group_cache *cache) | |
222 | { | |
223 | u64 start, end; | |
817d52f8 | 224 | |
11833d66 YZ |
225 | start = cache->key.objectid; |
226 | end = start + cache->key.offset - 1; | |
227 | ||
228 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
229 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
230 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
231 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
232 | } |
233 | ||
11833d66 YZ |
234 | static int exclude_super_stripes(struct btrfs_root *root, |
235 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 236 | { |
817d52f8 JB |
237 | u64 bytenr; |
238 | u64 *logical; | |
239 | int stripe_len; | |
240 | int i, nr, ret; | |
241 | ||
06b2331f YZ |
242 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
243 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
244 | cache->bytes_super += stripe_len; | |
245 | ret = add_excluded_extent(root, cache->key.objectid, | |
246 | stripe_len); | |
247 | BUG_ON(ret); | |
248 | } | |
249 | ||
817d52f8 JB |
250 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
251 | bytenr = btrfs_sb_offset(i); | |
252 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
253 | cache->key.objectid, bytenr, | |
254 | 0, &logical, &nr, &stripe_len); | |
255 | BUG_ON(ret); | |
11833d66 | 256 | |
817d52f8 | 257 | while (nr--) { |
1b2da372 | 258 | cache->bytes_super += stripe_len; |
11833d66 YZ |
259 | ret = add_excluded_extent(root, logical[nr], |
260 | stripe_len); | |
261 | BUG_ON(ret); | |
817d52f8 | 262 | } |
11833d66 | 263 | |
817d52f8 JB |
264 | kfree(logical); |
265 | } | |
817d52f8 JB |
266 | return 0; |
267 | } | |
268 | ||
11833d66 YZ |
269 | static struct btrfs_caching_control * |
270 | get_caching_control(struct btrfs_block_group_cache *cache) | |
271 | { | |
272 | struct btrfs_caching_control *ctl; | |
273 | ||
274 | spin_lock(&cache->lock); | |
275 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
276 | spin_unlock(&cache->lock); | |
277 | return NULL; | |
278 | } | |
279 | ||
dde5abee JB |
280 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
281 | if (!cache->caching_ctl) { | |
282 | spin_unlock(&cache->lock); | |
11833d66 YZ |
283 | return NULL; |
284 | } | |
285 | ||
286 | ctl = cache->caching_ctl; | |
287 | atomic_inc(&ctl->count); | |
288 | spin_unlock(&cache->lock); | |
289 | return ctl; | |
290 | } | |
291 | ||
292 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
293 | { | |
294 | if (atomic_dec_and_test(&ctl->count)) | |
295 | kfree(ctl); | |
296 | } | |
297 | ||
0f9dd46c JB |
298 | /* |
299 | * this is only called by cache_block_group, since we could have freed extents | |
300 | * we need to check the pinned_extents for any extents that can't be used yet | |
301 | * since their free space will be released as soon as the transaction commits. | |
302 | */ | |
817d52f8 | 303 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
304 | struct btrfs_fs_info *info, u64 start, u64 end) |
305 | { | |
817d52f8 | 306 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
307 | int ret; |
308 | ||
309 | while (start < end) { | |
11833d66 | 310 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 311 | &extent_start, &extent_end, |
11833d66 | 312 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
313 | if (ret) |
314 | break; | |
315 | ||
06b2331f | 316 | if (extent_start <= start) { |
0f9dd46c JB |
317 | start = extent_end + 1; |
318 | } else if (extent_start > start && extent_start < end) { | |
319 | size = extent_start - start; | |
817d52f8 | 320 | total_added += size; |
ea6a478e JB |
321 | ret = btrfs_add_free_space(block_group, start, |
322 | size); | |
0f9dd46c JB |
323 | BUG_ON(ret); |
324 | start = extent_end + 1; | |
325 | } else { | |
326 | break; | |
327 | } | |
328 | } | |
329 | ||
330 | if (start < end) { | |
331 | size = end - start; | |
817d52f8 | 332 | total_added += size; |
ea6a478e | 333 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
334 | BUG_ON(ret); |
335 | } | |
336 | ||
817d52f8 | 337 | return total_added; |
0f9dd46c JB |
338 | } |
339 | ||
bab39bf9 | 340 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 341 | { |
bab39bf9 JB |
342 | struct btrfs_block_group_cache *block_group; |
343 | struct btrfs_fs_info *fs_info; | |
344 | struct btrfs_caching_control *caching_ctl; | |
345 | struct btrfs_root *extent_root; | |
e37c9e69 | 346 | struct btrfs_path *path; |
5f39d397 | 347 | struct extent_buffer *leaf; |
11833d66 | 348 | struct btrfs_key key; |
817d52f8 | 349 | u64 total_found = 0; |
11833d66 YZ |
350 | u64 last = 0; |
351 | u32 nritems; | |
352 | int ret = 0; | |
f510cfec | 353 | |
bab39bf9 JB |
354 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
355 | block_group = caching_ctl->block_group; | |
356 | fs_info = block_group->fs_info; | |
357 | extent_root = fs_info->extent_root; | |
358 | ||
e37c9e69 CM |
359 | path = btrfs_alloc_path(); |
360 | if (!path) | |
bab39bf9 | 361 | goto out; |
7d7d6068 | 362 | |
817d52f8 | 363 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 364 | |
5cd57b2c | 365 | /* |
817d52f8 JB |
366 | * We don't want to deadlock with somebody trying to allocate a new |
367 | * extent for the extent root while also trying to search the extent | |
368 | * root to add free space. So we skip locking and search the commit | |
369 | * root, since its read-only | |
5cd57b2c CM |
370 | */ |
371 | path->skip_locking = 1; | |
817d52f8 | 372 | path->search_commit_root = 1; |
026fd317 | 373 | path->reada = 1; |
817d52f8 | 374 | |
e4404d6e | 375 | key.objectid = last; |
e37c9e69 | 376 | key.offset = 0; |
11833d66 | 377 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 378 | again: |
11833d66 | 379 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
380 | /* need to make sure the commit_root doesn't disappear */ |
381 | down_read(&fs_info->extent_commit_sem); | |
382 | ||
11833d66 | 383 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 384 | if (ret < 0) |
ef8bbdfe | 385 | goto err; |
a512bbf8 | 386 | |
11833d66 YZ |
387 | leaf = path->nodes[0]; |
388 | nritems = btrfs_header_nritems(leaf); | |
389 | ||
d397712b | 390 | while (1) { |
7841cb28 | 391 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 392 | last = (u64)-1; |
817d52f8 | 393 | break; |
f25784b3 | 394 | } |
817d52f8 | 395 | |
11833d66 YZ |
396 | if (path->slots[0] < nritems) { |
397 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
398 | } else { | |
399 | ret = find_next_key(path, 0, &key); | |
400 | if (ret) | |
e37c9e69 | 401 | break; |
817d52f8 | 402 | |
589d8ade JB |
403 | if (need_resched() || |
404 | btrfs_next_leaf(extent_root, path)) { | |
405 | caching_ctl->progress = last; | |
ff5714cc | 406 | btrfs_release_path(path); |
589d8ade JB |
407 | up_read(&fs_info->extent_commit_sem); |
408 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 409 | cond_resched(); |
589d8ade JB |
410 | goto again; |
411 | } | |
412 | leaf = path->nodes[0]; | |
413 | nritems = btrfs_header_nritems(leaf); | |
414 | continue; | |
11833d66 | 415 | } |
817d52f8 | 416 | |
11833d66 YZ |
417 | if (key.objectid < block_group->key.objectid) { |
418 | path->slots[0]++; | |
817d52f8 | 419 | continue; |
e37c9e69 | 420 | } |
0f9dd46c | 421 | |
e37c9e69 | 422 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 423 | block_group->key.offset) |
e37c9e69 | 424 | break; |
7d7d6068 | 425 | |
11833d66 | 426 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
427 | total_found += add_new_free_space(block_group, |
428 | fs_info, last, | |
429 | key.objectid); | |
7d7d6068 | 430 | last = key.objectid + key.offset; |
817d52f8 | 431 | |
11833d66 YZ |
432 | if (total_found > (1024 * 1024 * 2)) { |
433 | total_found = 0; | |
434 | wake_up(&caching_ctl->wait); | |
435 | } | |
817d52f8 | 436 | } |
e37c9e69 CM |
437 | path->slots[0]++; |
438 | } | |
817d52f8 | 439 | ret = 0; |
e37c9e69 | 440 | |
817d52f8 JB |
441 | total_found += add_new_free_space(block_group, fs_info, last, |
442 | block_group->key.objectid + | |
443 | block_group->key.offset); | |
11833d66 | 444 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
445 | |
446 | spin_lock(&block_group->lock); | |
11833d66 | 447 | block_group->caching_ctl = NULL; |
817d52f8 JB |
448 | block_group->cached = BTRFS_CACHE_FINISHED; |
449 | spin_unlock(&block_group->lock); | |
0f9dd46c | 450 | |
54aa1f4d | 451 | err: |
e37c9e69 | 452 | btrfs_free_path(path); |
276e680d | 453 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 454 | |
11833d66 YZ |
455 | free_excluded_extents(extent_root, block_group); |
456 | ||
457 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 458 | out: |
11833d66 YZ |
459 | wake_up(&caching_ctl->wait); |
460 | ||
461 | put_caching_control(caching_ctl); | |
11dfe35a | 462 | btrfs_put_block_group(block_group); |
817d52f8 JB |
463 | } |
464 | ||
9d66e233 JB |
465 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
466 | struct btrfs_trans_handle *trans, | |
b8399dee | 467 | struct btrfs_root *root, |
9d66e233 | 468 | int load_cache_only) |
817d52f8 | 469 | { |
291c7d2f | 470 | DEFINE_WAIT(wait); |
11833d66 YZ |
471 | struct btrfs_fs_info *fs_info = cache->fs_info; |
472 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
473 | int ret = 0; |
474 | ||
291c7d2f JB |
475 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
476 | BUG_ON(!caching_ctl); | |
477 | ||
478 | INIT_LIST_HEAD(&caching_ctl->list); | |
479 | mutex_init(&caching_ctl->mutex); | |
480 | init_waitqueue_head(&caching_ctl->wait); | |
481 | caching_ctl->block_group = cache; | |
482 | caching_ctl->progress = cache->key.objectid; | |
483 | atomic_set(&caching_ctl->count, 1); | |
484 | caching_ctl->work.func = caching_thread; | |
485 | ||
486 | spin_lock(&cache->lock); | |
487 | /* | |
488 | * This should be a rare occasion, but this could happen I think in the | |
489 | * case where one thread starts to load the space cache info, and then | |
490 | * some other thread starts a transaction commit which tries to do an | |
491 | * allocation while the other thread is still loading the space cache | |
492 | * info. The previous loop should have kept us from choosing this block | |
493 | * group, but if we've moved to the state where we will wait on caching | |
494 | * block groups we need to first check if we're doing a fast load here, | |
495 | * so we can wait for it to finish, otherwise we could end up allocating | |
496 | * from a block group who's cache gets evicted for one reason or | |
497 | * another. | |
498 | */ | |
499 | while (cache->cached == BTRFS_CACHE_FAST) { | |
500 | struct btrfs_caching_control *ctl; | |
501 | ||
502 | ctl = cache->caching_ctl; | |
503 | atomic_inc(&ctl->count); | |
504 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
505 | spin_unlock(&cache->lock); | |
506 | ||
507 | schedule(); | |
508 | ||
509 | finish_wait(&ctl->wait, &wait); | |
510 | put_caching_control(ctl); | |
511 | spin_lock(&cache->lock); | |
512 | } | |
513 | ||
514 | if (cache->cached != BTRFS_CACHE_NO) { | |
515 | spin_unlock(&cache->lock); | |
516 | kfree(caching_ctl); | |
11833d66 | 517 | return 0; |
291c7d2f JB |
518 | } |
519 | WARN_ON(cache->caching_ctl); | |
520 | cache->caching_ctl = caching_ctl; | |
521 | cache->cached = BTRFS_CACHE_FAST; | |
522 | spin_unlock(&cache->lock); | |
11833d66 | 523 | |
9d66e233 JB |
524 | /* |
525 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
526 | * to have the normal tree locking. Also if we are currently trying to |
527 | * allocate blocks for the tree root we can't do the fast caching since | |
528 | * we likely hold important locks. | |
9d66e233 | 529 | */ |
f7039b1d | 530 | if (trans && (!trans->transaction->in_commit) && |
73bc1876 JB |
531 | (root && root != root->fs_info->tree_root) && |
532 | btrfs_test_opt(root, SPACE_CACHE)) { | |
9d66e233 JB |
533 | ret = load_free_space_cache(fs_info, cache); |
534 | ||
535 | spin_lock(&cache->lock); | |
536 | if (ret == 1) { | |
291c7d2f | 537 | cache->caching_ctl = NULL; |
9d66e233 JB |
538 | cache->cached = BTRFS_CACHE_FINISHED; |
539 | cache->last_byte_to_unpin = (u64)-1; | |
540 | } else { | |
291c7d2f JB |
541 | if (load_cache_only) { |
542 | cache->caching_ctl = NULL; | |
543 | cache->cached = BTRFS_CACHE_NO; | |
544 | } else { | |
545 | cache->cached = BTRFS_CACHE_STARTED; | |
546 | } | |
9d66e233 JB |
547 | } |
548 | spin_unlock(&cache->lock); | |
291c7d2f | 549 | wake_up(&caching_ctl->wait); |
3c14874a | 550 | if (ret == 1) { |
291c7d2f | 551 | put_caching_control(caching_ctl); |
3c14874a | 552 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 553 | return 0; |
3c14874a | 554 | } |
291c7d2f JB |
555 | } else { |
556 | /* | |
557 | * We are not going to do the fast caching, set cached to the | |
558 | * appropriate value and wakeup any waiters. | |
559 | */ | |
560 | spin_lock(&cache->lock); | |
561 | if (load_cache_only) { | |
562 | cache->caching_ctl = NULL; | |
563 | cache->cached = BTRFS_CACHE_NO; | |
564 | } else { | |
565 | cache->cached = BTRFS_CACHE_STARTED; | |
566 | } | |
567 | spin_unlock(&cache->lock); | |
568 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
569 | } |
570 | ||
291c7d2f JB |
571 | if (load_cache_only) { |
572 | put_caching_control(caching_ctl); | |
11833d66 | 573 | return 0; |
817d52f8 | 574 | } |
817d52f8 | 575 | |
11833d66 | 576 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 577 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
578 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
579 | up_write(&fs_info->extent_commit_sem); | |
580 | ||
11dfe35a | 581 | btrfs_get_block_group(cache); |
11833d66 | 582 | |
bab39bf9 | 583 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 584 | |
ef8bbdfe | 585 | return ret; |
e37c9e69 CM |
586 | } |
587 | ||
0f9dd46c JB |
588 | /* |
589 | * return the block group that starts at or after bytenr | |
590 | */ | |
d397712b CM |
591 | static struct btrfs_block_group_cache * |
592 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 593 | { |
0f9dd46c | 594 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 595 | |
0f9dd46c | 596 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 597 | |
0f9dd46c | 598 | return cache; |
0ef3e66b CM |
599 | } |
600 | ||
0f9dd46c | 601 | /* |
9f55684c | 602 | * return the block group that contains the given bytenr |
0f9dd46c | 603 | */ |
d397712b CM |
604 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
605 | struct btrfs_fs_info *info, | |
606 | u64 bytenr) | |
be744175 | 607 | { |
0f9dd46c | 608 | struct btrfs_block_group_cache *cache; |
be744175 | 609 | |
0f9dd46c | 610 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 611 | |
0f9dd46c | 612 | return cache; |
be744175 | 613 | } |
0b86a832 | 614 | |
0f9dd46c JB |
615 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
616 | u64 flags) | |
6324fbf3 | 617 | { |
0f9dd46c | 618 | struct list_head *head = &info->space_info; |
0f9dd46c | 619 | struct btrfs_space_info *found; |
4184ea7f | 620 | |
b742bb82 YZ |
621 | flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM | |
622 | BTRFS_BLOCK_GROUP_METADATA; | |
623 | ||
4184ea7f CM |
624 | rcu_read_lock(); |
625 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 626 | if (found->flags & flags) { |
4184ea7f | 627 | rcu_read_unlock(); |
0f9dd46c | 628 | return found; |
4184ea7f | 629 | } |
0f9dd46c | 630 | } |
4184ea7f | 631 | rcu_read_unlock(); |
0f9dd46c | 632 | return NULL; |
6324fbf3 CM |
633 | } |
634 | ||
4184ea7f CM |
635 | /* |
636 | * after adding space to the filesystem, we need to clear the full flags | |
637 | * on all the space infos. | |
638 | */ | |
639 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
640 | { | |
641 | struct list_head *head = &info->space_info; | |
642 | struct btrfs_space_info *found; | |
643 | ||
644 | rcu_read_lock(); | |
645 | list_for_each_entry_rcu(found, head, list) | |
646 | found->full = 0; | |
647 | rcu_read_unlock(); | |
648 | } | |
649 | ||
80eb234a JB |
650 | static u64 div_factor(u64 num, int factor) |
651 | { | |
652 | if (factor == 10) | |
653 | return num; | |
654 | num *= factor; | |
655 | do_div(num, 10); | |
656 | return num; | |
657 | } | |
658 | ||
e5bc2458 CM |
659 | static u64 div_factor_fine(u64 num, int factor) |
660 | { | |
661 | if (factor == 100) | |
662 | return num; | |
663 | num *= factor; | |
664 | do_div(num, 100); | |
665 | return num; | |
666 | } | |
667 | ||
d2fb3437 YZ |
668 | u64 btrfs_find_block_group(struct btrfs_root *root, |
669 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 670 | { |
96b5179d | 671 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 672 | u64 used; |
d2fb3437 YZ |
673 | u64 last = max(search_hint, search_start); |
674 | u64 group_start = 0; | |
31f3c99b | 675 | int full_search = 0; |
d2fb3437 | 676 | int factor = 9; |
0ef3e66b | 677 | int wrapped = 0; |
31f3c99b | 678 | again: |
e8569813 ZY |
679 | while (1) { |
680 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
681 | if (!cache) |
682 | break; | |
96b5179d | 683 | |
c286ac48 | 684 | spin_lock(&cache->lock); |
96b5179d CM |
685 | last = cache->key.objectid + cache->key.offset; |
686 | used = btrfs_block_group_used(&cache->item); | |
687 | ||
d2fb3437 YZ |
688 | if ((full_search || !cache->ro) && |
689 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 690 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
691 | div_factor(cache->key.offset, factor)) { |
692 | group_start = cache->key.objectid; | |
c286ac48 | 693 | spin_unlock(&cache->lock); |
fa9c0d79 | 694 | btrfs_put_block_group(cache); |
8790d502 CM |
695 | goto found; |
696 | } | |
6324fbf3 | 697 | } |
c286ac48 | 698 | spin_unlock(&cache->lock); |
fa9c0d79 | 699 | btrfs_put_block_group(cache); |
de428b63 | 700 | cond_resched(); |
cd1bc465 | 701 | } |
0ef3e66b CM |
702 | if (!wrapped) { |
703 | last = search_start; | |
704 | wrapped = 1; | |
705 | goto again; | |
706 | } | |
707 | if (!full_search && factor < 10) { | |
be744175 | 708 | last = search_start; |
31f3c99b | 709 | full_search = 1; |
0ef3e66b | 710 | factor = 10; |
31f3c99b CM |
711 | goto again; |
712 | } | |
be744175 | 713 | found: |
d2fb3437 | 714 | return group_start; |
925baedd | 715 | } |
0f9dd46c | 716 | |
e02119d5 | 717 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 718 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
719 | { |
720 | int ret; | |
721 | struct btrfs_key key; | |
31840ae1 | 722 | struct btrfs_path *path; |
e02119d5 | 723 | |
31840ae1 | 724 | path = btrfs_alloc_path(); |
d8926bb3 MF |
725 | if (!path) |
726 | return -ENOMEM; | |
727 | ||
e02119d5 CM |
728 | key.objectid = start; |
729 | key.offset = len; | |
730 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
731 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
732 | 0, 0); | |
31840ae1 | 733 | btrfs_free_path(path); |
7bb86316 CM |
734 | return ret; |
735 | } | |
736 | ||
a22285a6 YZ |
737 | /* |
738 | * helper function to lookup reference count and flags of extent. | |
739 | * | |
740 | * the head node for delayed ref is used to store the sum of all the | |
741 | * reference count modifications queued up in the rbtree. the head | |
742 | * node may also store the extent flags to set. This way you can check | |
743 | * to see what the reference count and extent flags would be if all of | |
744 | * the delayed refs are not processed. | |
745 | */ | |
746 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
747 | struct btrfs_root *root, u64 bytenr, | |
748 | u64 num_bytes, u64 *refs, u64 *flags) | |
749 | { | |
750 | struct btrfs_delayed_ref_head *head; | |
751 | struct btrfs_delayed_ref_root *delayed_refs; | |
752 | struct btrfs_path *path; | |
753 | struct btrfs_extent_item *ei; | |
754 | struct extent_buffer *leaf; | |
755 | struct btrfs_key key; | |
756 | u32 item_size; | |
757 | u64 num_refs; | |
758 | u64 extent_flags; | |
759 | int ret; | |
760 | ||
761 | path = btrfs_alloc_path(); | |
762 | if (!path) | |
763 | return -ENOMEM; | |
764 | ||
765 | key.objectid = bytenr; | |
766 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
767 | key.offset = num_bytes; | |
768 | if (!trans) { | |
769 | path->skip_locking = 1; | |
770 | path->search_commit_root = 1; | |
771 | } | |
772 | again: | |
773 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
774 | &key, path, 0, 0); | |
775 | if (ret < 0) | |
776 | goto out_free; | |
777 | ||
778 | if (ret == 0) { | |
779 | leaf = path->nodes[0]; | |
780 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
781 | if (item_size >= sizeof(*ei)) { | |
782 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
783 | struct btrfs_extent_item); | |
784 | num_refs = btrfs_extent_refs(leaf, ei); | |
785 | extent_flags = btrfs_extent_flags(leaf, ei); | |
786 | } else { | |
787 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
788 | struct btrfs_extent_item_v0 *ei0; | |
789 | BUG_ON(item_size != sizeof(*ei0)); | |
790 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
791 | struct btrfs_extent_item_v0); | |
792 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
793 | /* FIXME: this isn't correct for data */ | |
794 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
795 | #else | |
796 | BUG(); | |
797 | #endif | |
798 | } | |
799 | BUG_ON(num_refs == 0); | |
800 | } else { | |
801 | num_refs = 0; | |
802 | extent_flags = 0; | |
803 | ret = 0; | |
804 | } | |
805 | ||
806 | if (!trans) | |
807 | goto out; | |
808 | ||
809 | delayed_refs = &trans->transaction->delayed_refs; | |
810 | spin_lock(&delayed_refs->lock); | |
811 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
812 | if (head) { | |
813 | if (!mutex_trylock(&head->mutex)) { | |
814 | atomic_inc(&head->node.refs); | |
815 | spin_unlock(&delayed_refs->lock); | |
816 | ||
b3b4aa74 | 817 | btrfs_release_path(path); |
a22285a6 | 818 | |
8cc33e5c DS |
819 | /* |
820 | * Mutex was contended, block until it's released and try | |
821 | * again | |
822 | */ | |
a22285a6 YZ |
823 | mutex_lock(&head->mutex); |
824 | mutex_unlock(&head->mutex); | |
825 | btrfs_put_delayed_ref(&head->node); | |
826 | goto again; | |
827 | } | |
828 | if (head->extent_op && head->extent_op->update_flags) | |
829 | extent_flags |= head->extent_op->flags_to_set; | |
830 | else | |
831 | BUG_ON(num_refs == 0); | |
832 | ||
833 | num_refs += head->node.ref_mod; | |
834 | mutex_unlock(&head->mutex); | |
835 | } | |
836 | spin_unlock(&delayed_refs->lock); | |
837 | out: | |
838 | WARN_ON(num_refs == 0); | |
839 | if (refs) | |
840 | *refs = num_refs; | |
841 | if (flags) | |
842 | *flags = extent_flags; | |
843 | out_free: | |
844 | btrfs_free_path(path); | |
845 | return ret; | |
846 | } | |
847 | ||
d8d5f3e1 CM |
848 | /* |
849 | * Back reference rules. Back refs have three main goals: | |
850 | * | |
851 | * 1) differentiate between all holders of references to an extent so that | |
852 | * when a reference is dropped we can make sure it was a valid reference | |
853 | * before freeing the extent. | |
854 | * | |
855 | * 2) Provide enough information to quickly find the holders of an extent | |
856 | * if we notice a given block is corrupted or bad. | |
857 | * | |
858 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
859 | * maintenance. This is actually the same as #2, but with a slightly | |
860 | * different use case. | |
861 | * | |
5d4f98a2 YZ |
862 | * There are two kinds of back refs. The implicit back refs is optimized |
863 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
864 | * back refs of this kind provide information about the block's owner tree | |
865 | * and the pointer's key. These information allow us to find the block by | |
866 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
867 | * referenced by their owner trees. The location of tree block is recorded | |
868 | * in the back refs. Actually the full back refs is generic, and can be | |
869 | * used in all cases the implicit back refs is used. The major shortcoming | |
870 | * of the full back refs is its overhead. Every time a tree block gets | |
871 | * COWed, we have to update back refs entry for all pointers in it. | |
872 | * | |
873 | * For a newly allocated tree block, we use implicit back refs for | |
874 | * pointers in it. This means most tree related operations only involve | |
875 | * implicit back refs. For a tree block created in old transaction, the | |
876 | * only way to drop a reference to it is COW it. So we can detect the | |
877 | * event that tree block loses its owner tree's reference and do the | |
878 | * back refs conversion. | |
879 | * | |
880 | * When a tree block is COW'd through a tree, there are four cases: | |
881 | * | |
882 | * The reference count of the block is one and the tree is the block's | |
883 | * owner tree. Nothing to do in this case. | |
884 | * | |
885 | * The reference count of the block is one and the tree is not the | |
886 | * block's owner tree. In this case, full back refs is used for pointers | |
887 | * in the block. Remove these full back refs, add implicit back refs for | |
888 | * every pointers in the new block. | |
889 | * | |
890 | * The reference count of the block is greater than one and the tree is | |
891 | * the block's owner tree. In this case, implicit back refs is used for | |
892 | * pointers in the block. Add full back refs for every pointers in the | |
893 | * block, increase lower level extents' reference counts. The original | |
894 | * implicit back refs are entailed to the new block. | |
895 | * | |
896 | * The reference count of the block is greater than one and the tree is | |
897 | * not the block's owner tree. Add implicit back refs for every pointer in | |
898 | * the new block, increase lower level extents' reference count. | |
899 | * | |
900 | * Back Reference Key composing: | |
901 | * | |
902 | * The key objectid corresponds to the first byte in the extent, | |
903 | * The key type is used to differentiate between types of back refs. | |
904 | * There are different meanings of the key offset for different types | |
905 | * of back refs. | |
906 | * | |
d8d5f3e1 CM |
907 | * File extents can be referenced by: |
908 | * | |
909 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 910 | * - different files inside a single subvolume |
d8d5f3e1 CM |
911 | * - different offsets inside a file (bookend extents in file.c) |
912 | * | |
5d4f98a2 | 913 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
914 | * |
915 | * - Objectid of the subvolume root | |
d8d5f3e1 | 916 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
917 | * - original offset in the file |
918 | * - how many bookend extents | |
d8d5f3e1 | 919 | * |
5d4f98a2 YZ |
920 | * The key offset for the implicit back refs is hash of the first |
921 | * three fields. | |
d8d5f3e1 | 922 | * |
5d4f98a2 | 923 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 924 | * |
5d4f98a2 | 925 | * - number of pointers in the tree leaf |
d8d5f3e1 | 926 | * |
5d4f98a2 YZ |
927 | * The key offset for the implicit back refs is the first byte of |
928 | * the tree leaf | |
d8d5f3e1 | 929 | * |
5d4f98a2 YZ |
930 | * When a file extent is allocated, The implicit back refs is used. |
931 | * the fields are filled in: | |
d8d5f3e1 | 932 | * |
5d4f98a2 | 933 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 934 | * |
5d4f98a2 YZ |
935 | * When a file extent is removed file truncation, we find the |
936 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 937 | * |
5d4f98a2 | 938 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 939 | * |
5d4f98a2 | 940 | * Btree extents can be referenced by: |
d8d5f3e1 | 941 | * |
5d4f98a2 | 942 | * - Different subvolumes |
d8d5f3e1 | 943 | * |
5d4f98a2 YZ |
944 | * Both the implicit back refs and the full back refs for tree blocks |
945 | * only consist of key. The key offset for the implicit back refs is | |
946 | * objectid of block's owner tree. The key offset for the full back refs | |
947 | * is the first byte of parent block. | |
d8d5f3e1 | 948 | * |
5d4f98a2 YZ |
949 | * When implicit back refs is used, information about the lowest key and |
950 | * level of the tree block are required. These information are stored in | |
951 | * tree block info structure. | |
d8d5f3e1 | 952 | */ |
31840ae1 | 953 | |
5d4f98a2 YZ |
954 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
955 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
956 | struct btrfs_root *root, | |
957 | struct btrfs_path *path, | |
958 | u64 owner, u32 extra_size) | |
7bb86316 | 959 | { |
5d4f98a2 YZ |
960 | struct btrfs_extent_item *item; |
961 | struct btrfs_extent_item_v0 *ei0; | |
962 | struct btrfs_extent_ref_v0 *ref0; | |
963 | struct btrfs_tree_block_info *bi; | |
964 | struct extent_buffer *leaf; | |
7bb86316 | 965 | struct btrfs_key key; |
5d4f98a2 YZ |
966 | struct btrfs_key found_key; |
967 | u32 new_size = sizeof(*item); | |
968 | u64 refs; | |
969 | int ret; | |
970 | ||
971 | leaf = path->nodes[0]; | |
972 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
973 | ||
974 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
975 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
976 | struct btrfs_extent_item_v0); | |
977 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
978 | ||
979 | if (owner == (u64)-1) { | |
980 | while (1) { | |
981 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
982 | ret = btrfs_next_leaf(root, path); | |
983 | if (ret < 0) | |
984 | return ret; | |
985 | BUG_ON(ret > 0); | |
986 | leaf = path->nodes[0]; | |
987 | } | |
988 | btrfs_item_key_to_cpu(leaf, &found_key, | |
989 | path->slots[0]); | |
990 | BUG_ON(key.objectid != found_key.objectid); | |
991 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
992 | path->slots[0]++; | |
993 | continue; | |
994 | } | |
995 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
996 | struct btrfs_extent_ref_v0); | |
997 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
998 | break; | |
999 | } | |
1000 | } | |
b3b4aa74 | 1001 | btrfs_release_path(path); |
5d4f98a2 YZ |
1002 | |
1003 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1004 | new_size += sizeof(*bi); | |
1005 | ||
1006 | new_size -= sizeof(*ei0); | |
1007 | ret = btrfs_search_slot(trans, root, &key, path, | |
1008 | new_size + extra_size, 1); | |
1009 | if (ret < 0) | |
1010 | return ret; | |
1011 | BUG_ON(ret); | |
1012 | ||
1013 | ret = btrfs_extend_item(trans, root, path, new_size); | |
5d4f98a2 YZ |
1014 | |
1015 | leaf = path->nodes[0]; | |
1016 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1017 | btrfs_set_extent_refs(leaf, item, refs); | |
1018 | /* FIXME: get real generation */ | |
1019 | btrfs_set_extent_generation(leaf, item, 0); | |
1020 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1021 | btrfs_set_extent_flags(leaf, item, | |
1022 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1023 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1024 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1025 | /* FIXME: get first key of the block */ | |
1026 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1027 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1028 | } else { | |
1029 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1030 | } | |
1031 | btrfs_mark_buffer_dirty(leaf); | |
1032 | return 0; | |
1033 | } | |
1034 | #endif | |
1035 | ||
1036 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1037 | { | |
1038 | u32 high_crc = ~(u32)0; | |
1039 | u32 low_crc = ~(u32)0; | |
1040 | __le64 lenum; | |
1041 | ||
1042 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1043 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1044 | lenum = cpu_to_le64(owner); |
163e783e | 1045 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1046 | lenum = cpu_to_le64(offset); |
163e783e | 1047 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1048 | |
1049 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1050 | } | |
1051 | ||
1052 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1053 | struct btrfs_extent_data_ref *ref) | |
1054 | { | |
1055 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1056 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1057 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1058 | } | |
1059 | ||
1060 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1061 | struct btrfs_extent_data_ref *ref, | |
1062 | u64 root_objectid, u64 owner, u64 offset) | |
1063 | { | |
1064 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1065 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1066 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1067 | return 0; | |
1068 | return 1; | |
1069 | } | |
1070 | ||
1071 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1072 | struct btrfs_root *root, | |
1073 | struct btrfs_path *path, | |
1074 | u64 bytenr, u64 parent, | |
1075 | u64 root_objectid, | |
1076 | u64 owner, u64 offset) | |
1077 | { | |
1078 | struct btrfs_key key; | |
1079 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1080 | struct extent_buffer *leaf; |
5d4f98a2 | 1081 | u32 nritems; |
74493f7a | 1082 | int ret; |
5d4f98a2 YZ |
1083 | int recow; |
1084 | int err = -ENOENT; | |
74493f7a | 1085 | |
31840ae1 | 1086 | key.objectid = bytenr; |
5d4f98a2 YZ |
1087 | if (parent) { |
1088 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1089 | key.offset = parent; | |
1090 | } else { | |
1091 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1092 | key.offset = hash_extent_data_ref(root_objectid, | |
1093 | owner, offset); | |
1094 | } | |
1095 | again: | |
1096 | recow = 0; | |
1097 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1098 | if (ret < 0) { | |
1099 | err = ret; | |
1100 | goto fail; | |
1101 | } | |
31840ae1 | 1102 | |
5d4f98a2 YZ |
1103 | if (parent) { |
1104 | if (!ret) | |
1105 | return 0; | |
1106 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1107 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1108 | btrfs_release_path(path); |
5d4f98a2 YZ |
1109 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1110 | if (ret < 0) { | |
1111 | err = ret; | |
1112 | goto fail; | |
1113 | } | |
1114 | if (!ret) | |
1115 | return 0; | |
1116 | #endif | |
1117 | goto fail; | |
31840ae1 ZY |
1118 | } |
1119 | ||
1120 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1121 | nritems = btrfs_header_nritems(leaf); |
1122 | while (1) { | |
1123 | if (path->slots[0] >= nritems) { | |
1124 | ret = btrfs_next_leaf(root, path); | |
1125 | if (ret < 0) | |
1126 | err = ret; | |
1127 | if (ret) | |
1128 | goto fail; | |
1129 | ||
1130 | leaf = path->nodes[0]; | |
1131 | nritems = btrfs_header_nritems(leaf); | |
1132 | recow = 1; | |
1133 | } | |
1134 | ||
1135 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1136 | if (key.objectid != bytenr || | |
1137 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1138 | goto fail; | |
1139 | ||
1140 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1141 | struct btrfs_extent_data_ref); | |
1142 | ||
1143 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1144 | owner, offset)) { | |
1145 | if (recow) { | |
b3b4aa74 | 1146 | btrfs_release_path(path); |
5d4f98a2 YZ |
1147 | goto again; |
1148 | } | |
1149 | err = 0; | |
1150 | break; | |
1151 | } | |
1152 | path->slots[0]++; | |
31840ae1 | 1153 | } |
5d4f98a2 YZ |
1154 | fail: |
1155 | return err; | |
31840ae1 ZY |
1156 | } |
1157 | ||
5d4f98a2 YZ |
1158 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1159 | struct btrfs_root *root, | |
1160 | struct btrfs_path *path, | |
1161 | u64 bytenr, u64 parent, | |
1162 | u64 root_objectid, u64 owner, | |
1163 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1164 | { |
1165 | struct btrfs_key key; | |
1166 | struct extent_buffer *leaf; | |
5d4f98a2 | 1167 | u32 size; |
31840ae1 ZY |
1168 | u32 num_refs; |
1169 | int ret; | |
74493f7a | 1170 | |
74493f7a | 1171 | key.objectid = bytenr; |
5d4f98a2 YZ |
1172 | if (parent) { |
1173 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1174 | key.offset = parent; | |
1175 | size = sizeof(struct btrfs_shared_data_ref); | |
1176 | } else { | |
1177 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1178 | key.offset = hash_extent_data_ref(root_objectid, | |
1179 | owner, offset); | |
1180 | size = sizeof(struct btrfs_extent_data_ref); | |
1181 | } | |
74493f7a | 1182 | |
5d4f98a2 YZ |
1183 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1184 | if (ret && ret != -EEXIST) | |
1185 | goto fail; | |
1186 | ||
1187 | leaf = path->nodes[0]; | |
1188 | if (parent) { | |
1189 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1190 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1191 | struct btrfs_shared_data_ref); |
1192 | if (ret == 0) { | |
1193 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1194 | } else { | |
1195 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1196 | num_refs += refs_to_add; | |
1197 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1198 | } |
5d4f98a2 YZ |
1199 | } else { |
1200 | struct btrfs_extent_data_ref *ref; | |
1201 | while (ret == -EEXIST) { | |
1202 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1203 | struct btrfs_extent_data_ref); | |
1204 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1205 | owner, offset)) | |
1206 | break; | |
b3b4aa74 | 1207 | btrfs_release_path(path); |
5d4f98a2 YZ |
1208 | key.offset++; |
1209 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1210 | size); | |
1211 | if (ret && ret != -EEXIST) | |
1212 | goto fail; | |
31840ae1 | 1213 | |
5d4f98a2 YZ |
1214 | leaf = path->nodes[0]; |
1215 | } | |
1216 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1217 | struct btrfs_extent_data_ref); | |
1218 | if (ret == 0) { | |
1219 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1220 | root_objectid); | |
1221 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1222 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1223 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1224 | } else { | |
1225 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1226 | num_refs += refs_to_add; | |
1227 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1228 | } |
31840ae1 | 1229 | } |
5d4f98a2 YZ |
1230 | btrfs_mark_buffer_dirty(leaf); |
1231 | ret = 0; | |
1232 | fail: | |
b3b4aa74 | 1233 | btrfs_release_path(path); |
7bb86316 | 1234 | return ret; |
74493f7a CM |
1235 | } |
1236 | ||
5d4f98a2 YZ |
1237 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1238 | struct btrfs_root *root, | |
1239 | struct btrfs_path *path, | |
1240 | int refs_to_drop) | |
31840ae1 | 1241 | { |
5d4f98a2 YZ |
1242 | struct btrfs_key key; |
1243 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1244 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1245 | struct extent_buffer *leaf; |
5d4f98a2 | 1246 | u32 num_refs = 0; |
31840ae1 ZY |
1247 | int ret = 0; |
1248 | ||
1249 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1250 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1251 | ||
1252 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1253 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1254 | struct btrfs_extent_data_ref); | |
1255 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1256 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1257 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_shared_data_ref); | |
1259 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1260 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1261 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1262 | struct btrfs_extent_ref_v0 *ref0; | |
1263 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1264 | struct btrfs_extent_ref_v0); | |
1265 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1266 | #endif | |
1267 | } else { | |
1268 | BUG(); | |
1269 | } | |
1270 | ||
56bec294 CM |
1271 | BUG_ON(num_refs < refs_to_drop); |
1272 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1273 | |
31840ae1 ZY |
1274 | if (num_refs == 0) { |
1275 | ret = btrfs_del_item(trans, root, path); | |
1276 | } else { | |
5d4f98a2 YZ |
1277 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1278 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1279 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1280 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1281 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1282 | else { | |
1283 | struct btrfs_extent_ref_v0 *ref0; | |
1284 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1285 | struct btrfs_extent_ref_v0); | |
1286 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1287 | } | |
1288 | #endif | |
31840ae1 ZY |
1289 | btrfs_mark_buffer_dirty(leaf); |
1290 | } | |
31840ae1 ZY |
1291 | return ret; |
1292 | } | |
1293 | ||
5d4f98a2 YZ |
1294 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1295 | struct btrfs_path *path, | |
1296 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1297 | { |
5d4f98a2 YZ |
1298 | struct btrfs_key key; |
1299 | struct extent_buffer *leaf; | |
1300 | struct btrfs_extent_data_ref *ref1; | |
1301 | struct btrfs_shared_data_ref *ref2; | |
1302 | u32 num_refs = 0; | |
1303 | ||
1304 | leaf = path->nodes[0]; | |
1305 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1306 | if (iref) { | |
1307 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1308 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1309 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1310 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1311 | } else { | |
1312 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1313 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1314 | } | |
1315 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1316 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1317 | struct btrfs_extent_data_ref); | |
1318 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1319 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1320 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1321 | struct btrfs_shared_data_ref); | |
1322 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1323 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1324 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1325 | struct btrfs_extent_ref_v0 *ref0; | |
1326 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1327 | struct btrfs_extent_ref_v0); | |
1328 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1329 | #endif |
5d4f98a2 YZ |
1330 | } else { |
1331 | WARN_ON(1); | |
1332 | } | |
1333 | return num_refs; | |
1334 | } | |
15916de8 | 1335 | |
5d4f98a2 YZ |
1336 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1337 | struct btrfs_root *root, | |
1338 | struct btrfs_path *path, | |
1339 | u64 bytenr, u64 parent, | |
1340 | u64 root_objectid) | |
1f3c79a2 | 1341 | { |
5d4f98a2 | 1342 | struct btrfs_key key; |
1f3c79a2 | 1343 | int ret; |
1f3c79a2 | 1344 | |
5d4f98a2 YZ |
1345 | key.objectid = bytenr; |
1346 | if (parent) { | |
1347 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1348 | key.offset = parent; | |
1349 | } else { | |
1350 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1351 | key.offset = root_objectid; | |
1f3c79a2 LH |
1352 | } |
1353 | ||
5d4f98a2 YZ |
1354 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1355 | if (ret > 0) | |
1356 | ret = -ENOENT; | |
1357 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1358 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1359 | btrfs_release_path(path); |
5d4f98a2 YZ |
1360 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1361 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1362 | if (ret > 0) | |
1363 | ret = -ENOENT; | |
1364 | } | |
1f3c79a2 | 1365 | #endif |
5d4f98a2 | 1366 | return ret; |
1f3c79a2 LH |
1367 | } |
1368 | ||
5d4f98a2 YZ |
1369 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1370 | struct btrfs_root *root, | |
1371 | struct btrfs_path *path, | |
1372 | u64 bytenr, u64 parent, | |
1373 | u64 root_objectid) | |
31840ae1 | 1374 | { |
5d4f98a2 | 1375 | struct btrfs_key key; |
31840ae1 | 1376 | int ret; |
31840ae1 | 1377 | |
5d4f98a2 YZ |
1378 | key.objectid = bytenr; |
1379 | if (parent) { | |
1380 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1381 | key.offset = parent; | |
1382 | } else { | |
1383 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1384 | key.offset = root_objectid; | |
1385 | } | |
1386 | ||
1387 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1388 | btrfs_release_path(path); |
31840ae1 ZY |
1389 | return ret; |
1390 | } | |
1391 | ||
5d4f98a2 | 1392 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1393 | { |
5d4f98a2 YZ |
1394 | int type; |
1395 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1396 | if (parent > 0) | |
1397 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1398 | else | |
1399 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1400 | } else { | |
1401 | if (parent > 0) | |
1402 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1403 | else | |
1404 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1405 | } | |
1406 | return type; | |
31840ae1 | 1407 | } |
56bec294 | 1408 | |
2c47e605 YZ |
1409 | static int find_next_key(struct btrfs_path *path, int level, |
1410 | struct btrfs_key *key) | |
56bec294 | 1411 | |
02217ed2 | 1412 | { |
2c47e605 | 1413 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1414 | if (!path->nodes[level]) |
1415 | break; | |
5d4f98a2 YZ |
1416 | if (path->slots[level] + 1 >= |
1417 | btrfs_header_nritems(path->nodes[level])) | |
1418 | continue; | |
1419 | if (level == 0) | |
1420 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1421 | path->slots[level] + 1); | |
1422 | else | |
1423 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1424 | path->slots[level] + 1); | |
1425 | return 0; | |
1426 | } | |
1427 | return 1; | |
1428 | } | |
037e6390 | 1429 | |
5d4f98a2 YZ |
1430 | /* |
1431 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1432 | * to the address of inline back ref, and 0 is returned. | |
1433 | * | |
1434 | * if back ref isn't found, *ref_ret is set to the address where it | |
1435 | * should be inserted, and -ENOENT is returned. | |
1436 | * | |
1437 | * if insert is true and there are too many inline back refs, the path | |
1438 | * points to the extent item, and -EAGAIN is returned. | |
1439 | * | |
1440 | * NOTE: inline back refs are ordered in the same way that back ref | |
1441 | * items in the tree are ordered. | |
1442 | */ | |
1443 | static noinline_for_stack | |
1444 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1445 | struct btrfs_root *root, | |
1446 | struct btrfs_path *path, | |
1447 | struct btrfs_extent_inline_ref **ref_ret, | |
1448 | u64 bytenr, u64 num_bytes, | |
1449 | u64 parent, u64 root_objectid, | |
1450 | u64 owner, u64 offset, int insert) | |
1451 | { | |
1452 | struct btrfs_key key; | |
1453 | struct extent_buffer *leaf; | |
1454 | struct btrfs_extent_item *ei; | |
1455 | struct btrfs_extent_inline_ref *iref; | |
1456 | u64 flags; | |
1457 | u64 item_size; | |
1458 | unsigned long ptr; | |
1459 | unsigned long end; | |
1460 | int extra_size; | |
1461 | int type; | |
1462 | int want; | |
1463 | int ret; | |
1464 | int err = 0; | |
26b8003f | 1465 | |
db94535d | 1466 | key.objectid = bytenr; |
31840ae1 | 1467 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1468 | key.offset = num_bytes; |
31840ae1 | 1469 | |
5d4f98a2 YZ |
1470 | want = extent_ref_type(parent, owner); |
1471 | if (insert) { | |
1472 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1473 | path->keep_locks = 1; |
5d4f98a2 YZ |
1474 | } else |
1475 | extra_size = -1; | |
1476 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1477 | if (ret < 0) { |
5d4f98a2 YZ |
1478 | err = ret; |
1479 | goto out; | |
1480 | } | |
1481 | BUG_ON(ret); | |
1482 | ||
1483 | leaf = path->nodes[0]; | |
1484 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1485 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1486 | if (item_size < sizeof(*ei)) { | |
1487 | if (!insert) { | |
1488 | err = -ENOENT; | |
1489 | goto out; | |
1490 | } | |
1491 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1492 | extra_size); | |
1493 | if (ret < 0) { | |
1494 | err = ret; | |
1495 | goto out; | |
1496 | } | |
1497 | leaf = path->nodes[0]; | |
1498 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1499 | } | |
1500 | #endif | |
1501 | BUG_ON(item_size < sizeof(*ei)); | |
1502 | ||
5d4f98a2 YZ |
1503 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1504 | flags = btrfs_extent_flags(leaf, ei); | |
1505 | ||
1506 | ptr = (unsigned long)(ei + 1); | |
1507 | end = (unsigned long)ei + item_size; | |
1508 | ||
1509 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1510 | ptr += sizeof(struct btrfs_tree_block_info); | |
1511 | BUG_ON(ptr > end); | |
1512 | } else { | |
1513 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1514 | } | |
1515 | ||
1516 | err = -ENOENT; | |
1517 | while (1) { | |
1518 | if (ptr >= end) { | |
1519 | WARN_ON(ptr > end); | |
1520 | break; | |
1521 | } | |
1522 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1523 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1524 | if (want < type) | |
1525 | break; | |
1526 | if (want > type) { | |
1527 | ptr += btrfs_extent_inline_ref_size(type); | |
1528 | continue; | |
1529 | } | |
1530 | ||
1531 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1532 | struct btrfs_extent_data_ref *dref; | |
1533 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1534 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1535 | owner, offset)) { | |
1536 | err = 0; | |
1537 | break; | |
1538 | } | |
1539 | if (hash_extent_data_ref_item(leaf, dref) < | |
1540 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1541 | break; | |
1542 | } else { | |
1543 | u64 ref_offset; | |
1544 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1545 | if (parent > 0) { | |
1546 | if (parent == ref_offset) { | |
1547 | err = 0; | |
1548 | break; | |
1549 | } | |
1550 | if (ref_offset < parent) | |
1551 | break; | |
1552 | } else { | |
1553 | if (root_objectid == ref_offset) { | |
1554 | err = 0; | |
1555 | break; | |
1556 | } | |
1557 | if (ref_offset < root_objectid) | |
1558 | break; | |
1559 | } | |
1560 | } | |
1561 | ptr += btrfs_extent_inline_ref_size(type); | |
1562 | } | |
1563 | if (err == -ENOENT && insert) { | |
1564 | if (item_size + extra_size >= | |
1565 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1566 | err = -EAGAIN; | |
1567 | goto out; | |
1568 | } | |
1569 | /* | |
1570 | * To add new inline back ref, we have to make sure | |
1571 | * there is no corresponding back ref item. | |
1572 | * For simplicity, we just do not add new inline back | |
1573 | * ref if there is any kind of item for this block | |
1574 | */ | |
2c47e605 YZ |
1575 | if (find_next_key(path, 0, &key) == 0 && |
1576 | key.objectid == bytenr && | |
85d4198e | 1577 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1578 | err = -EAGAIN; |
1579 | goto out; | |
1580 | } | |
1581 | } | |
1582 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1583 | out: | |
85d4198e | 1584 | if (insert) { |
5d4f98a2 YZ |
1585 | path->keep_locks = 0; |
1586 | btrfs_unlock_up_safe(path, 1); | |
1587 | } | |
1588 | return err; | |
1589 | } | |
1590 | ||
1591 | /* | |
1592 | * helper to add new inline back ref | |
1593 | */ | |
1594 | static noinline_for_stack | |
1595 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1596 | struct btrfs_root *root, | |
1597 | struct btrfs_path *path, | |
1598 | struct btrfs_extent_inline_ref *iref, | |
1599 | u64 parent, u64 root_objectid, | |
1600 | u64 owner, u64 offset, int refs_to_add, | |
1601 | struct btrfs_delayed_extent_op *extent_op) | |
1602 | { | |
1603 | struct extent_buffer *leaf; | |
1604 | struct btrfs_extent_item *ei; | |
1605 | unsigned long ptr; | |
1606 | unsigned long end; | |
1607 | unsigned long item_offset; | |
1608 | u64 refs; | |
1609 | int size; | |
1610 | int type; | |
1611 | int ret; | |
1612 | ||
1613 | leaf = path->nodes[0]; | |
1614 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1615 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1616 | ||
1617 | type = extent_ref_type(parent, owner); | |
1618 | size = btrfs_extent_inline_ref_size(type); | |
1619 | ||
1620 | ret = btrfs_extend_item(trans, root, path, size); | |
5d4f98a2 YZ |
1621 | |
1622 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1623 | refs = btrfs_extent_refs(leaf, ei); | |
1624 | refs += refs_to_add; | |
1625 | btrfs_set_extent_refs(leaf, ei, refs); | |
1626 | if (extent_op) | |
1627 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1628 | ||
1629 | ptr = (unsigned long)ei + item_offset; | |
1630 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1631 | if (ptr < end - size) | |
1632 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1633 | end - size - ptr); | |
1634 | ||
1635 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1636 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1637 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1638 | struct btrfs_extent_data_ref *dref; | |
1639 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1640 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1641 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1642 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1643 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1644 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1645 | struct btrfs_shared_data_ref *sref; | |
1646 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1647 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1648 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1649 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1650 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1651 | } else { | |
1652 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1653 | } | |
1654 | btrfs_mark_buffer_dirty(leaf); | |
1655 | return 0; | |
1656 | } | |
1657 | ||
1658 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1659 | struct btrfs_root *root, | |
1660 | struct btrfs_path *path, | |
1661 | struct btrfs_extent_inline_ref **ref_ret, | |
1662 | u64 bytenr, u64 num_bytes, u64 parent, | |
1663 | u64 root_objectid, u64 owner, u64 offset) | |
1664 | { | |
1665 | int ret; | |
1666 | ||
1667 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1668 | bytenr, num_bytes, parent, | |
1669 | root_objectid, owner, offset, 0); | |
1670 | if (ret != -ENOENT) | |
54aa1f4d | 1671 | return ret; |
5d4f98a2 | 1672 | |
b3b4aa74 | 1673 | btrfs_release_path(path); |
5d4f98a2 YZ |
1674 | *ref_ret = NULL; |
1675 | ||
1676 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1677 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1678 | root_objectid); | |
1679 | } else { | |
1680 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1681 | root_objectid, owner, offset); | |
b9473439 | 1682 | } |
5d4f98a2 YZ |
1683 | return ret; |
1684 | } | |
31840ae1 | 1685 | |
5d4f98a2 YZ |
1686 | /* |
1687 | * helper to update/remove inline back ref | |
1688 | */ | |
1689 | static noinline_for_stack | |
1690 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1691 | struct btrfs_root *root, | |
1692 | struct btrfs_path *path, | |
1693 | struct btrfs_extent_inline_ref *iref, | |
1694 | int refs_to_mod, | |
1695 | struct btrfs_delayed_extent_op *extent_op) | |
1696 | { | |
1697 | struct extent_buffer *leaf; | |
1698 | struct btrfs_extent_item *ei; | |
1699 | struct btrfs_extent_data_ref *dref = NULL; | |
1700 | struct btrfs_shared_data_ref *sref = NULL; | |
1701 | unsigned long ptr; | |
1702 | unsigned long end; | |
1703 | u32 item_size; | |
1704 | int size; | |
1705 | int type; | |
1706 | int ret; | |
1707 | u64 refs; | |
1708 | ||
1709 | leaf = path->nodes[0]; | |
1710 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1711 | refs = btrfs_extent_refs(leaf, ei); | |
1712 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1713 | refs += refs_to_mod; | |
1714 | btrfs_set_extent_refs(leaf, ei, refs); | |
1715 | if (extent_op) | |
1716 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1717 | ||
1718 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1719 | ||
1720 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1721 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1722 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1723 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1724 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1725 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1726 | } else { | |
1727 | refs = 1; | |
1728 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1729 | } |
31840ae1 | 1730 | |
5d4f98a2 YZ |
1731 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1732 | refs += refs_to_mod; | |
1733 | ||
1734 | if (refs > 0) { | |
1735 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1736 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1737 | else | |
1738 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1739 | } else { | |
1740 | size = btrfs_extent_inline_ref_size(type); | |
1741 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1742 | ptr = (unsigned long)iref; | |
1743 | end = (unsigned long)ei + item_size; | |
1744 | if (ptr + size < end) | |
1745 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1746 | end - ptr - size); | |
1747 | item_size -= size; | |
1748 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
5d4f98a2 YZ |
1749 | } |
1750 | btrfs_mark_buffer_dirty(leaf); | |
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | static noinline_for_stack | |
1755 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1756 | struct btrfs_root *root, | |
1757 | struct btrfs_path *path, | |
1758 | u64 bytenr, u64 num_bytes, u64 parent, | |
1759 | u64 root_objectid, u64 owner, | |
1760 | u64 offset, int refs_to_add, | |
1761 | struct btrfs_delayed_extent_op *extent_op) | |
1762 | { | |
1763 | struct btrfs_extent_inline_ref *iref; | |
1764 | int ret; | |
1765 | ||
1766 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1767 | bytenr, num_bytes, parent, | |
1768 | root_objectid, owner, offset, 1); | |
1769 | if (ret == 0) { | |
1770 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1771 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1772 | refs_to_add, extent_op); | |
1773 | } else if (ret == -ENOENT) { | |
1774 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1775 | parent, root_objectid, | |
1776 | owner, offset, refs_to_add, | |
1777 | extent_op); | |
771ed689 | 1778 | } |
5d4f98a2 YZ |
1779 | return ret; |
1780 | } | |
31840ae1 | 1781 | |
5d4f98a2 YZ |
1782 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1783 | struct btrfs_root *root, | |
1784 | struct btrfs_path *path, | |
1785 | u64 bytenr, u64 parent, u64 root_objectid, | |
1786 | u64 owner, u64 offset, int refs_to_add) | |
1787 | { | |
1788 | int ret; | |
1789 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1790 | BUG_ON(refs_to_add != 1); | |
1791 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1792 | parent, root_objectid); | |
1793 | } else { | |
1794 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1795 | parent, root_objectid, | |
1796 | owner, offset, refs_to_add); | |
1797 | } | |
1798 | return ret; | |
1799 | } | |
56bec294 | 1800 | |
5d4f98a2 YZ |
1801 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1802 | struct btrfs_root *root, | |
1803 | struct btrfs_path *path, | |
1804 | struct btrfs_extent_inline_ref *iref, | |
1805 | int refs_to_drop, int is_data) | |
1806 | { | |
1807 | int ret; | |
b9473439 | 1808 | |
5d4f98a2 YZ |
1809 | BUG_ON(!is_data && refs_to_drop != 1); |
1810 | if (iref) { | |
1811 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1812 | -refs_to_drop, NULL); | |
1813 | } else if (is_data) { | |
1814 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1815 | } else { | |
1816 | ret = btrfs_del_item(trans, root, path); | |
1817 | } | |
1818 | return ret; | |
1819 | } | |
1820 | ||
5378e607 | 1821 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1822 | u64 start, u64 len) |
1823 | { | |
5378e607 | 1824 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1825 | } |
5d4f98a2 YZ |
1826 | |
1827 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1828 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1829 | { |
5d4f98a2 | 1830 | int ret; |
5378e607 | 1831 | u64 discarded_bytes = 0; |
a1d3c478 | 1832 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1833 | |
e244a0ae | 1834 | |
5d4f98a2 | 1835 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 | 1836 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
a1d3c478 | 1837 | bytenr, &num_bytes, &bbio, 0); |
5d4f98a2 | 1838 | if (!ret) { |
a1d3c478 | 1839 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1840 | int i; |
1841 | ||
5d4f98a2 | 1842 | |
a1d3c478 | 1843 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1844 | if (!stripe->dev->can_discard) |
1845 | continue; | |
1846 | ||
5378e607 LD |
1847 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1848 | stripe->physical, | |
1849 | stripe->length); | |
1850 | if (!ret) | |
1851 | discarded_bytes += stripe->length; | |
1852 | else if (ret != -EOPNOTSUPP) | |
1853 | break; | |
d5e2003c JB |
1854 | |
1855 | /* | |
1856 | * Just in case we get back EOPNOTSUPP for some reason, | |
1857 | * just ignore the return value so we don't screw up | |
1858 | * people calling discard_extent. | |
1859 | */ | |
1860 | ret = 0; | |
5d4f98a2 | 1861 | } |
a1d3c478 | 1862 | kfree(bbio); |
5d4f98a2 | 1863 | } |
5378e607 LD |
1864 | |
1865 | if (actual_bytes) | |
1866 | *actual_bytes = discarded_bytes; | |
1867 | ||
5d4f98a2 YZ |
1868 | |
1869 | return ret; | |
5d4f98a2 YZ |
1870 | } |
1871 | ||
1872 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1873 | struct btrfs_root *root, | |
1874 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1875 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1876 | { |
1877 | int ret; | |
66d7e7f0 AJ |
1878 | struct btrfs_fs_info *fs_info = root->fs_info; |
1879 | ||
5d4f98a2 YZ |
1880 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1881 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1882 | ||
1883 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1884 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1885 | num_bytes, | |
5d4f98a2 | 1886 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1887 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1888 | } else { |
66d7e7f0 AJ |
1889 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1890 | num_bytes, | |
5d4f98a2 | 1891 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1892 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1893 | } |
1894 | return ret; | |
1895 | } | |
1896 | ||
1897 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1898 | struct btrfs_root *root, | |
1899 | u64 bytenr, u64 num_bytes, | |
1900 | u64 parent, u64 root_objectid, | |
1901 | u64 owner, u64 offset, int refs_to_add, | |
1902 | struct btrfs_delayed_extent_op *extent_op) | |
1903 | { | |
1904 | struct btrfs_path *path; | |
1905 | struct extent_buffer *leaf; | |
1906 | struct btrfs_extent_item *item; | |
1907 | u64 refs; | |
1908 | int ret; | |
1909 | int err = 0; | |
1910 | ||
1911 | path = btrfs_alloc_path(); | |
1912 | if (!path) | |
1913 | return -ENOMEM; | |
1914 | ||
1915 | path->reada = 1; | |
1916 | path->leave_spinning = 1; | |
1917 | /* this will setup the path even if it fails to insert the back ref */ | |
1918 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1919 | path, bytenr, num_bytes, parent, | |
1920 | root_objectid, owner, offset, | |
1921 | refs_to_add, extent_op); | |
1922 | if (ret == 0) | |
1923 | goto out; | |
1924 | ||
1925 | if (ret != -EAGAIN) { | |
1926 | err = ret; | |
1927 | goto out; | |
1928 | } | |
1929 | ||
1930 | leaf = path->nodes[0]; | |
1931 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1932 | refs = btrfs_extent_refs(leaf, item); | |
1933 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1934 | if (extent_op) | |
1935 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1936 | |
5d4f98a2 | 1937 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1938 | btrfs_release_path(path); |
56bec294 CM |
1939 | |
1940 | path->reada = 1; | |
b9473439 CM |
1941 | path->leave_spinning = 1; |
1942 | ||
56bec294 CM |
1943 | /* now insert the actual backref */ |
1944 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1945 | path, bytenr, parent, root_objectid, |
1946 | owner, offset, refs_to_add); | |
56bec294 | 1947 | BUG_ON(ret); |
5d4f98a2 | 1948 | out: |
56bec294 | 1949 | btrfs_free_path(path); |
5d4f98a2 | 1950 | return err; |
56bec294 CM |
1951 | } |
1952 | ||
5d4f98a2 YZ |
1953 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1954 | struct btrfs_root *root, | |
1955 | struct btrfs_delayed_ref_node *node, | |
1956 | struct btrfs_delayed_extent_op *extent_op, | |
1957 | int insert_reserved) | |
56bec294 | 1958 | { |
5d4f98a2 YZ |
1959 | int ret = 0; |
1960 | struct btrfs_delayed_data_ref *ref; | |
1961 | struct btrfs_key ins; | |
1962 | u64 parent = 0; | |
1963 | u64 ref_root = 0; | |
1964 | u64 flags = 0; | |
1965 | ||
1966 | ins.objectid = node->bytenr; | |
1967 | ins.offset = node->num_bytes; | |
1968 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1969 | ||
1970 | ref = btrfs_delayed_node_to_data_ref(node); | |
1971 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1972 | parent = ref->parent; | |
1973 | else | |
1974 | ref_root = ref->root; | |
1975 | ||
1976 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1977 | if (extent_op) { | |
1978 | BUG_ON(extent_op->update_key); | |
1979 | flags |= extent_op->flags_to_set; | |
1980 | } | |
1981 | ret = alloc_reserved_file_extent(trans, root, | |
1982 | parent, ref_root, flags, | |
1983 | ref->objectid, ref->offset, | |
1984 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1985 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1986 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1987 | node->num_bytes, parent, | |
1988 | ref_root, ref->objectid, | |
1989 | ref->offset, node->ref_mod, | |
1990 | extent_op); | |
1991 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1992 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1993 | node->num_bytes, parent, | |
1994 | ref_root, ref->objectid, | |
1995 | ref->offset, node->ref_mod, | |
1996 | extent_op); | |
1997 | } else { | |
1998 | BUG(); | |
1999 | } | |
2000 | return ret; | |
2001 | } | |
2002 | ||
2003 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2004 | struct extent_buffer *leaf, | |
2005 | struct btrfs_extent_item *ei) | |
2006 | { | |
2007 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2008 | if (extent_op->update_flags) { | |
2009 | flags |= extent_op->flags_to_set; | |
2010 | btrfs_set_extent_flags(leaf, ei, flags); | |
2011 | } | |
2012 | ||
2013 | if (extent_op->update_key) { | |
2014 | struct btrfs_tree_block_info *bi; | |
2015 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2016 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2017 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2018 | } | |
2019 | } | |
2020 | ||
2021 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2022 | struct btrfs_root *root, | |
2023 | struct btrfs_delayed_ref_node *node, | |
2024 | struct btrfs_delayed_extent_op *extent_op) | |
2025 | { | |
2026 | struct btrfs_key key; | |
2027 | struct btrfs_path *path; | |
2028 | struct btrfs_extent_item *ei; | |
2029 | struct extent_buffer *leaf; | |
2030 | u32 item_size; | |
56bec294 | 2031 | int ret; |
5d4f98a2 YZ |
2032 | int err = 0; |
2033 | ||
2034 | path = btrfs_alloc_path(); | |
2035 | if (!path) | |
2036 | return -ENOMEM; | |
2037 | ||
2038 | key.objectid = node->bytenr; | |
2039 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2040 | key.offset = node->num_bytes; | |
2041 | ||
2042 | path->reada = 1; | |
2043 | path->leave_spinning = 1; | |
2044 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2045 | path, 0, 1); | |
2046 | if (ret < 0) { | |
2047 | err = ret; | |
2048 | goto out; | |
2049 | } | |
2050 | if (ret > 0) { | |
2051 | err = -EIO; | |
2052 | goto out; | |
2053 | } | |
2054 | ||
2055 | leaf = path->nodes[0]; | |
2056 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2057 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2058 | if (item_size < sizeof(*ei)) { | |
2059 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2060 | path, (u64)-1, 0); | |
2061 | if (ret < 0) { | |
2062 | err = ret; | |
2063 | goto out; | |
2064 | } | |
2065 | leaf = path->nodes[0]; | |
2066 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2067 | } | |
2068 | #endif | |
2069 | BUG_ON(item_size < sizeof(*ei)); | |
2070 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2071 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2072 | |
5d4f98a2 YZ |
2073 | btrfs_mark_buffer_dirty(leaf); |
2074 | out: | |
2075 | btrfs_free_path(path); | |
2076 | return err; | |
56bec294 CM |
2077 | } |
2078 | ||
5d4f98a2 YZ |
2079 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2080 | struct btrfs_root *root, | |
2081 | struct btrfs_delayed_ref_node *node, | |
2082 | struct btrfs_delayed_extent_op *extent_op, | |
2083 | int insert_reserved) | |
56bec294 CM |
2084 | { |
2085 | int ret = 0; | |
5d4f98a2 YZ |
2086 | struct btrfs_delayed_tree_ref *ref; |
2087 | struct btrfs_key ins; | |
2088 | u64 parent = 0; | |
2089 | u64 ref_root = 0; | |
56bec294 | 2090 | |
5d4f98a2 YZ |
2091 | ins.objectid = node->bytenr; |
2092 | ins.offset = node->num_bytes; | |
2093 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2094 | |
5d4f98a2 YZ |
2095 | ref = btrfs_delayed_node_to_tree_ref(node); |
2096 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2097 | parent = ref->parent; | |
2098 | else | |
2099 | ref_root = ref->root; | |
2100 | ||
2101 | BUG_ON(node->ref_mod != 1); | |
2102 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2103 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2104 | !extent_op->update_key); | |
2105 | ret = alloc_reserved_tree_block(trans, root, | |
2106 | parent, ref_root, | |
2107 | extent_op->flags_to_set, | |
2108 | &extent_op->key, | |
2109 | ref->level, &ins); | |
5d4f98a2 YZ |
2110 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2111 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2112 | node->num_bytes, parent, ref_root, | |
2113 | ref->level, 0, 1, extent_op); | |
2114 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2115 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2116 | node->num_bytes, parent, ref_root, | |
2117 | ref->level, 0, 1, extent_op); | |
2118 | } else { | |
2119 | BUG(); | |
2120 | } | |
56bec294 CM |
2121 | return ret; |
2122 | } | |
2123 | ||
2124 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2125 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2126 | struct btrfs_root *root, | |
2127 | struct btrfs_delayed_ref_node *node, | |
2128 | struct btrfs_delayed_extent_op *extent_op, | |
2129 | int insert_reserved) | |
56bec294 CM |
2130 | { |
2131 | int ret; | |
5d4f98a2 | 2132 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2133 | struct btrfs_delayed_ref_head *head; |
2134 | /* | |
2135 | * we've hit the end of the chain and we were supposed | |
2136 | * to insert this extent into the tree. But, it got | |
2137 | * deleted before we ever needed to insert it, so all | |
2138 | * we have to do is clean up the accounting | |
2139 | */ | |
5d4f98a2 YZ |
2140 | BUG_ON(extent_op); |
2141 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2142 | if (insert_reserved) { |
f0486c68 YZ |
2143 | btrfs_pin_extent(root, node->bytenr, |
2144 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2145 | if (head->is_data) { |
2146 | ret = btrfs_del_csums(trans, root, | |
2147 | node->bytenr, | |
2148 | node->num_bytes); | |
2149 | BUG_ON(ret); | |
2150 | } | |
56bec294 | 2151 | } |
56bec294 CM |
2152 | mutex_unlock(&head->mutex); |
2153 | return 0; | |
2154 | } | |
2155 | ||
5d4f98a2 YZ |
2156 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2157 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2158 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2159 | insert_reserved); | |
2160 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2161 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2162 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2163 | insert_reserved); | |
2164 | else | |
2165 | BUG(); | |
2166 | return ret; | |
56bec294 CM |
2167 | } |
2168 | ||
2169 | static noinline struct btrfs_delayed_ref_node * | |
2170 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2171 | { | |
2172 | struct rb_node *node; | |
2173 | struct btrfs_delayed_ref_node *ref; | |
2174 | int action = BTRFS_ADD_DELAYED_REF; | |
2175 | again: | |
2176 | /* | |
2177 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2178 | * this prevents ref count from going down to zero when | |
2179 | * there still are pending delayed ref. | |
2180 | */ | |
2181 | node = rb_prev(&head->node.rb_node); | |
2182 | while (1) { | |
2183 | if (!node) | |
2184 | break; | |
2185 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2186 | rb_node); | |
2187 | if (ref->bytenr != head->node.bytenr) | |
2188 | break; | |
5d4f98a2 | 2189 | if (ref->action == action) |
56bec294 CM |
2190 | return ref; |
2191 | node = rb_prev(node); | |
2192 | } | |
2193 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2194 | action = BTRFS_DROP_DELAYED_REF; | |
2195 | goto again; | |
2196 | } | |
2197 | return NULL; | |
2198 | } | |
2199 | ||
c3e69d58 CM |
2200 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2201 | struct btrfs_root *root, | |
2202 | struct list_head *cluster) | |
56bec294 | 2203 | { |
56bec294 CM |
2204 | struct btrfs_delayed_ref_root *delayed_refs; |
2205 | struct btrfs_delayed_ref_node *ref; | |
2206 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2207 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2208 | int ret; |
c3e69d58 | 2209 | int count = 0; |
56bec294 | 2210 | int must_insert_reserved = 0; |
56bec294 CM |
2211 | |
2212 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2213 | while (1) { |
2214 | if (!locked_ref) { | |
c3e69d58 CM |
2215 | /* pick a new head ref from the cluster list */ |
2216 | if (list_empty(cluster)) | |
56bec294 | 2217 | break; |
56bec294 | 2218 | |
c3e69d58 CM |
2219 | locked_ref = list_entry(cluster->next, |
2220 | struct btrfs_delayed_ref_head, cluster); | |
2221 | ||
2222 | /* grab the lock that says we are going to process | |
2223 | * all the refs for this head */ | |
2224 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2225 | ||
2226 | /* | |
2227 | * we may have dropped the spin lock to get the head | |
2228 | * mutex lock, and that might have given someone else | |
2229 | * time to free the head. If that's true, it has been | |
2230 | * removed from our list and we can move on. | |
2231 | */ | |
2232 | if (ret == -EAGAIN) { | |
2233 | locked_ref = NULL; | |
2234 | count++; | |
2235 | continue; | |
56bec294 CM |
2236 | } |
2237 | } | |
a28ec197 | 2238 | |
56bec294 CM |
2239 | /* |
2240 | * record the must insert reserved flag before we | |
2241 | * drop the spin lock. | |
2242 | */ | |
2243 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2244 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2245 | |
5d4f98a2 YZ |
2246 | extent_op = locked_ref->extent_op; |
2247 | locked_ref->extent_op = NULL; | |
2248 | ||
56bec294 CM |
2249 | /* |
2250 | * locked_ref is the head node, so we have to go one | |
2251 | * node back for any delayed ref updates | |
2252 | */ | |
56bec294 CM |
2253 | ref = select_delayed_ref(locked_ref); |
2254 | if (!ref) { | |
2255 | /* All delayed refs have been processed, Go ahead | |
2256 | * and send the head node to run_one_delayed_ref, | |
2257 | * so that any accounting fixes can happen | |
2258 | */ | |
2259 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2260 | |
2261 | if (extent_op && must_insert_reserved) { | |
2262 | kfree(extent_op); | |
2263 | extent_op = NULL; | |
2264 | } | |
2265 | ||
2266 | if (extent_op) { | |
2267 | spin_unlock(&delayed_refs->lock); | |
2268 | ||
2269 | ret = run_delayed_extent_op(trans, root, | |
2270 | ref, extent_op); | |
2271 | BUG_ON(ret); | |
2272 | kfree(extent_op); | |
2273 | ||
2274 | cond_resched(); | |
2275 | spin_lock(&delayed_refs->lock); | |
2276 | continue; | |
2277 | } | |
2278 | ||
c3e69d58 | 2279 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2280 | locked_ref = NULL; |
2281 | } | |
02217ed2 | 2282 | |
56bec294 CM |
2283 | ref->in_tree = 0; |
2284 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2285 | delayed_refs->num_entries--; | |
5d4f98a2 | 2286 | |
56bec294 | 2287 | spin_unlock(&delayed_refs->lock); |
925baedd | 2288 | |
5d4f98a2 | 2289 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2290 | must_insert_reserved); |
2291 | BUG_ON(ret); | |
eb099670 | 2292 | |
5d4f98a2 YZ |
2293 | btrfs_put_delayed_ref(ref); |
2294 | kfree(extent_op); | |
c3e69d58 | 2295 | count++; |
5d4f98a2 | 2296 | |
c3e69d58 CM |
2297 | cond_resched(); |
2298 | spin_lock(&delayed_refs->lock); | |
2299 | } | |
2300 | return count; | |
2301 | } | |
2302 | ||
2303 | /* | |
2304 | * this starts processing the delayed reference count updates and | |
2305 | * extent insertions we have queued up so far. count can be | |
2306 | * 0, which means to process everything in the tree at the start | |
2307 | * of the run (but not newly added entries), or it can be some target | |
2308 | * number you'd like to process. | |
2309 | */ | |
2310 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2311 | struct btrfs_root *root, unsigned long count) | |
2312 | { | |
2313 | struct rb_node *node; | |
2314 | struct btrfs_delayed_ref_root *delayed_refs; | |
2315 | struct btrfs_delayed_ref_node *ref; | |
2316 | struct list_head cluster; | |
2317 | int ret; | |
2318 | int run_all = count == (unsigned long)-1; | |
2319 | int run_most = 0; | |
2320 | ||
2321 | if (root == root->fs_info->extent_root) | |
2322 | root = root->fs_info->tree_root; | |
2323 | ||
2324 | delayed_refs = &trans->transaction->delayed_refs; | |
2325 | INIT_LIST_HEAD(&cluster); | |
2326 | again: | |
2327 | spin_lock(&delayed_refs->lock); | |
2328 | if (count == 0) { | |
2329 | count = delayed_refs->num_entries * 2; | |
2330 | run_most = 1; | |
2331 | } | |
2332 | while (1) { | |
2333 | if (!(run_all || run_most) && | |
2334 | delayed_refs->num_heads_ready < 64) | |
2335 | break; | |
eb099670 | 2336 | |
56bec294 | 2337 | /* |
c3e69d58 CM |
2338 | * go find something we can process in the rbtree. We start at |
2339 | * the beginning of the tree, and then build a cluster | |
2340 | * of refs to process starting at the first one we are able to | |
2341 | * lock | |
56bec294 | 2342 | */ |
c3e69d58 CM |
2343 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2344 | delayed_refs->run_delayed_start); | |
2345 | if (ret) | |
56bec294 CM |
2346 | break; |
2347 | ||
c3e69d58 CM |
2348 | ret = run_clustered_refs(trans, root, &cluster); |
2349 | BUG_ON(ret < 0); | |
2350 | ||
2351 | count -= min_t(unsigned long, ret, count); | |
2352 | ||
2353 | if (count == 0) | |
2354 | break; | |
eb099670 | 2355 | } |
c3e69d58 | 2356 | |
56bec294 | 2357 | if (run_all) { |
56bec294 | 2358 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2359 | if (!node) |
56bec294 | 2360 | goto out; |
c3e69d58 | 2361 | count = (unsigned long)-1; |
e9d0b13b | 2362 | |
56bec294 CM |
2363 | while (node) { |
2364 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2365 | rb_node); | |
2366 | if (btrfs_delayed_ref_is_head(ref)) { | |
2367 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2368 | |
56bec294 CM |
2369 | head = btrfs_delayed_node_to_head(ref); |
2370 | atomic_inc(&ref->refs); | |
2371 | ||
2372 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2373 | /* |
2374 | * Mutex was contended, block until it's | |
2375 | * released and try again | |
2376 | */ | |
56bec294 CM |
2377 | mutex_lock(&head->mutex); |
2378 | mutex_unlock(&head->mutex); | |
2379 | ||
2380 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2381 | cond_resched(); |
56bec294 CM |
2382 | goto again; |
2383 | } | |
2384 | node = rb_next(node); | |
2385 | } | |
2386 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2387 | schedule_timeout(1); |
2388 | goto again; | |
5f39d397 | 2389 | } |
54aa1f4d | 2390 | out: |
c3e69d58 | 2391 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2392 | return 0; |
2393 | } | |
2394 | ||
5d4f98a2 YZ |
2395 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2396 | struct btrfs_root *root, | |
2397 | u64 bytenr, u64 num_bytes, u64 flags, | |
2398 | int is_data) | |
2399 | { | |
2400 | struct btrfs_delayed_extent_op *extent_op; | |
2401 | int ret; | |
2402 | ||
2403 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2404 | if (!extent_op) | |
2405 | return -ENOMEM; | |
2406 | ||
2407 | extent_op->flags_to_set = flags; | |
2408 | extent_op->update_flags = 1; | |
2409 | extent_op->update_key = 0; | |
2410 | extent_op->is_data = is_data ? 1 : 0; | |
2411 | ||
66d7e7f0 AJ |
2412 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2413 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2414 | if (ret) |
2415 | kfree(extent_op); | |
2416 | return ret; | |
2417 | } | |
2418 | ||
2419 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2420 | struct btrfs_root *root, | |
2421 | struct btrfs_path *path, | |
2422 | u64 objectid, u64 offset, u64 bytenr) | |
2423 | { | |
2424 | struct btrfs_delayed_ref_head *head; | |
2425 | struct btrfs_delayed_ref_node *ref; | |
2426 | struct btrfs_delayed_data_ref *data_ref; | |
2427 | struct btrfs_delayed_ref_root *delayed_refs; | |
2428 | struct rb_node *node; | |
2429 | int ret = 0; | |
2430 | ||
2431 | ret = -ENOENT; | |
2432 | delayed_refs = &trans->transaction->delayed_refs; | |
2433 | spin_lock(&delayed_refs->lock); | |
2434 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2435 | if (!head) | |
2436 | goto out; | |
2437 | ||
2438 | if (!mutex_trylock(&head->mutex)) { | |
2439 | atomic_inc(&head->node.refs); | |
2440 | spin_unlock(&delayed_refs->lock); | |
2441 | ||
b3b4aa74 | 2442 | btrfs_release_path(path); |
5d4f98a2 | 2443 | |
8cc33e5c DS |
2444 | /* |
2445 | * Mutex was contended, block until it's released and let | |
2446 | * caller try again | |
2447 | */ | |
5d4f98a2 YZ |
2448 | mutex_lock(&head->mutex); |
2449 | mutex_unlock(&head->mutex); | |
2450 | btrfs_put_delayed_ref(&head->node); | |
2451 | return -EAGAIN; | |
2452 | } | |
2453 | ||
2454 | node = rb_prev(&head->node.rb_node); | |
2455 | if (!node) | |
2456 | goto out_unlock; | |
2457 | ||
2458 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2459 | ||
2460 | if (ref->bytenr != bytenr) | |
2461 | goto out_unlock; | |
2462 | ||
2463 | ret = 1; | |
2464 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2465 | goto out_unlock; | |
2466 | ||
2467 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2468 | ||
2469 | node = rb_prev(node); | |
2470 | if (node) { | |
2471 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2472 | if (ref->bytenr == bytenr) | |
2473 | goto out_unlock; | |
2474 | } | |
2475 | ||
2476 | if (data_ref->root != root->root_key.objectid || | |
2477 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2478 | goto out_unlock; | |
2479 | ||
2480 | ret = 0; | |
2481 | out_unlock: | |
2482 | mutex_unlock(&head->mutex); | |
2483 | out: | |
2484 | spin_unlock(&delayed_refs->lock); | |
2485 | return ret; | |
2486 | } | |
2487 | ||
2488 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2489 | struct btrfs_root *root, | |
2490 | struct btrfs_path *path, | |
2491 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2492 | { |
2493 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2494 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2495 | struct btrfs_extent_data_ref *ref; |
2496 | struct btrfs_extent_inline_ref *iref; | |
2497 | struct btrfs_extent_item *ei; | |
f321e491 | 2498 | struct btrfs_key key; |
5d4f98a2 | 2499 | u32 item_size; |
be20aa9d | 2500 | int ret; |
925baedd | 2501 | |
be20aa9d | 2502 | key.objectid = bytenr; |
31840ae1 | 2503 | key.offset = (u64)-1; |
f321e491 | 2504 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2505 | |
be20aa9d CM |
2506 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2507 | if (ret < 0) | |
2508 | goto out; | |
2509 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2510 | |
2511 | ret = -ENOENT; | |
2512 | if (path->slots[0] == 0) | |
31840ae1 | 2513 | goto out; |
be20aa9d | 2514 | |
31840ae1 | 2515 | path->slots[0]--; |
f321e491 | 2516 | leaf = path->nodes[0]; |
5d4f98a2 | 2517 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2518 | |
5d4f98a2 | 2519 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2520 | goto out; |
f321e491 | 2521 | |
5d4f98a2 YZ |
2522 | ret = 1; |
2523 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2524 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2525 | if (item_size < sizeof(*ei)) { | |
2526 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2527 | goto out; | |
2528 | } | |
2529 | #endif | |
2530 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2531 | |
5d4f98a2 YZ |
2532 | if (item_size != sizeof(*ei) + |
2533 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2534 | goto out; | |
be20aa9d | 2535 | |
5d4f98a2 YZ |
2536 | if (btrfs_extent_generation(leaf, ei) <= |
2537 | btrfs_root_last_snapshot(&root->root_item)) | |
2538 | goto out; | |
2539 | ||
2540 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2541 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2542 | BTRFS_EXTENT_DATA_REF_KEY) | |
2543 | goto out; | |
2544 | ||
2545 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2546 | if (btrfs_extent_refs(leaf, ei) != | |
2547 | btrfs_extent_data_ref_count(leaf, ref) || | |
2548 | btrfs_extent_data_ref_root(leaf, ref) != | |
2549 | root->root_key.objectid || | |
2550 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2551 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2552 | goto out; | |
2553 | ||
2554 | ret = 0; | |
2555 | out: | |
2556 | return ret; | |
2557 | } | |
2558 | ||
2559 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2560 | struct btrfs_root *root, | |
2561 | u64 objectid, u64 offset, u64 bytenr) | |
2562 | { | |
2563 | struct btrfs_path *path; | |
2564 | int ret; | |
2565 | int ret2; | |
2566 | ||
2567 | path = btrfs_alloc_path(); | |
2568 | if (!path) | |
2569 | return -ENOENT; | |
2570 | ||
2571 | do { | |
2572 | ret = check_committed_ref(trans, root, path, objectid, | |
2573 | offset, bytenr); | |
2574 | if (ret && ret != -ENOENT) | |
f321e491 | 2575 | goto out; |
80ff3856 | 2576 | |
5d4f98a2 YZ |
2577 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2578 | offset, bytenr); | |
2579 | } while (ret2 == -EAGAIN); | |
2580 | ||
2581 | if (ret2 && ret2 != -ENOENT) { | |
2582 | ret = ret2; | |
2583 | goto out; | |
f321e491 | 2584 | } |
5d4f98a2 YZ |
2585 | |
2586 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2587 | ret = 0; | |
be20aa9d | 2588 | out: |
80ff3856 | 2589 | btrfs_free_path(path); |
f0486c68 YZ |
2590 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2591 | WARN_ON(ret > 0); | |
f321e491 | 2592 | return ret; |
be20aa9d | 2593 | } |
c5739bba | 2594 | |
5d4f98a2 | 2595 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2596 | struct btrfs_root *root, |
5d4f98a2 | 2597 | struct extent_buffer *buf, |
66d7e7f0 | 2598 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2599 | { |
2600 | u64 bytenr; | |
5d4f98a2 YZ |
2601 | u64 num_bytes; |
2602 | u64 parent; | |
31840ae1 | 2603 | u64 ref_root; |
31840ae1 | 2604 | u32 nritems; |
31840ae1 ZY |
2605 | struct btrfs_key key; |
2606 | struct btrfs_file_extent_item *fi; | |
2607 | int i; | |
2608 | int level; | |
2609 | int ret = 0; | |
31840ae1 | 2610 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2611 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2612 | |
2613 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2614 | nritems = btrfs_header_nritems(buf); |
2615 | level = btrfs_header_level(buf); | |
2616 | ||
5d4f98a2 YZ |
2617 | if (!root->ref_cows && level == 0) |
2618 | return 0; | |
31840ae1 | 2619 | |
5d4f98a2 YZ |
2620 | if (inc) |
2621 | process_func = btrfs_inc_extent_ref; | |
2622 | else | |
2623 | process_func = btrfs_free_extent; | |
31840ae1 | 2624 | |
5d4f98a2 YZ |
2625 | if (full_backref) |
2626 | parent = buf->start; | |
2627 | else | |
2628 | parent = 0; | |
2629 | ||
2630 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2631 | if (level == 0) { |
5d4f98a2 | 2632 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2633 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2634 | continue; | |
5d4f98a2 | 2635 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2636 | struct btrfs_file_extent_item); |
2637 | if (btrfs_file_extent_type(buf, fi) == | |
2638 | BTRFS_FILE_EXTENT_INLINE) | |
2639 | continue; | |
2640 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2641 | if (bytenr == 0) | |
2642 | continue; | |
5d4f98a2 YZ |
2643 | |
2644 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2645 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2646 | ret = process_func(trans, root, bytenr, num_bytes, | |
2647 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2648 | key.offset, for_cow); |
31840ae1 ZY |
2649 | if (ret) |
2650 | goto fail; | |
2651 | } else { | |
5d4f98a2 YZ |
2652 | bytenr = btrfs_node_blockptr(buf, i); |
2653 | num_bytes = btrfs_level_size(root, level - 1); | |
2654 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2655 | parent, ref_root, level - 1, 0, |
2656 | for_cow); | |
31840ae1 ZY |
2657 | if (ret) |
2658 | goto fail; | |
2659 | } | |
2660 | } | |
2661 | return 0; | |
2662 | fail: | |
5d4f98a2 YZ |
2663 | BUG(); |
2664 | return ret; | |
2665 | } | |
2666 | ||
2667 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2668 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2669 | { |
66d7e7f0 | 2670 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2671 | } |
2672 | ||
2673 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2674 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2675 | { |
66d7e7f0 | 2676 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2677 | } |
2678 | ||
9078a3e1 CM |
2679 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2680 | struct btrfs_root *root, | |
2681 | struct btrfs_path *path, | |
2682 | struct btrfs_block_group_cache *cache) | |
2683 | { | |
2684 | int ret; | |
9078a3e1 | 2685 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2686 | unsigned long bi; |
2687 | struct extent_buffer *leaf; | |
9078a3e1 | 2688 | |
9078a3e1 | 2689 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2690 | if (ret < 0) |
2691 | goto fail; | |
9078a3e1 | 2692 | BUG_ON(ret); |
5f39d397 CM |
2693 | |
2694 | leaf = path->nodes[0]; | |
2695 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2696 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2697 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2698 | btrfs_release_path(path); |
54aa1f4d | 2699 | fail: |
9078a3e1 CM |
2700 | if (ret) |
2701 | return ret; | |
9078a3e1 CM |
2702 | return 0; |
2703 | ||
2704 | } | |
2705 | ||
4a8c9a62 YZ |
2706 | static struct btrfs_block_group_cache * |
2707 | next_block_group(struct btrfs_root *root, | |
2708 | struct btrfs_block_group_cache *cache) | |
2709 | { | |
2710 | struct rb_node *node; | |
2711 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2712 | node = rb_next(&cache->cache_node); | |
2713 | btrfs_put_block_group(cache); | |
2714 | if (node) { | |
2715 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2716 | cache_node); | |
11dfe35a | 2717 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2718 | } else |
2719 | cache = NULL; | |
2720 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2721 | return cache; | |
2722 | } | |
2723 | ||
0af3d00b JB |
2724 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2725 | struct btrfs_trans_handle *trans, | |
2726 | struct btrfs_path *path) | |
2727 | { | |
2728 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2729 | struct inode *inode = NULL; | |
2730 | u64 alloc_hint = 0; | |
2b20982e | 2731 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2732 | int num_pages = 0; |
2733 | int retries = 0; | |
2734 | int ret = 0; | |
2735 | ||
2736 | /* | |
2737 | * If this block group is smaller than 100 megs don't bother caching the | |
2738 | * block group. | |
2739 | */ | |
2740 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2741 | spin_lock(&block_group->lock); | |
2742 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2743 | spin_unlock(&block_group->lock); | |
2744 | return 0; | |
2745 | } | |
2746 | ||
2747 | again: | |
2748 | inode = lookup_free_space_inode(root, block_group, path); | |
2749 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2750 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2751 | btrfs_release_path(path); |
0af3d00b JB |
2752 | goto out; |
2753 | } | |
2754 | ||
2755 | if (IS_ERR(inode)) { | |
2756 | BUG_ON(retries); | |
2757 | retries++; | |
2758 | ||
2759 | if (block_group->ro) | |
2760 | goto out_free; | |
2761 | ||
2762 | ret = create_free_space_inode(root, trans, block_group, path); | |
2763 | if (ret) | |
2764 | goto out_free; | |
2765 | goto again; | |
2766 | } | |
2767 | ||
5b0e95bf JB |
2768 | /* We've already setup this transaction, go ahead and exit */ |
2769 | if (block_group->cache_generation == trans->transid && | |
2770 | i_size_read(inode)) { | |
2771 | dcs = BTRFS_DC_SETUP; | |
2772 | goto out_put; | |
2773 | } | |
2774 | ||
0af3d00b JB |
2775 | /* |
2776 | * We want to set the generation to 0, that way if anything goes wrong | |
2777 | * from here on out we know not to trust this cache when we load up next | |
2778 | * time. | |
2779 | */ | |
2780 | BTRFS_I(inode)->generation = 0; | |
2781 | ret = btrfs_update_inode(trans, root, inode); | |
2782 | WARN_ON(ret); | |
2783 | ||
2784 | if (i_size_read(inode) > 0) { | |
2785 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2786 | inode); | |
2787 | if (ret) | |
2788 | goto out_put; | |
2789 | } | |
2790 | ||
2791 | spin_lock(&block_group->lock); | |
2792 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2793 | /* We're not cached, don't bother trying to write stuff out */ |
2794 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2795 | spin_unlock(&block_group->lock); |
2796 | goto out_put; | |
2797 | } | |
2798 | spin_unlock(&block_group->lock); | |
2799 | ||
2800 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2801 | if (!num_pages) | |
2802 | num_pages = 1; | |
2803 | ||
2804 | /* | |
2805 | * Just to make absolutely sure we have enough space, we're going to | |
2806 | * preallocate 12 pages worth of space for each block group. In | |
2807 | * practice we ought to use at most 8, but we need extra space so we can | |
2808 | * add our header and have a terminator between the extents and the | |
2809 | * bitmaps. | |
2810 | */ | |
2811 | num_pages *= 16; | |
2812 | num_pages *= PAGE_CACHE_SIZE; | |
2813 | ||
5b0e95bf | 2814 | ret = btrfs_check_data_free_space(inode, num_pages); |
0af3d00b JB |
2815 | if (ret) |
2816 | goto out_put; | |
2817 | ||
2818 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2819 | num_pages, num_pages, | |
2820 | &alloc_hint); | |
5b0e95bf | 2821 | if (!ret) |
2b20982e | 2822 | dcs = BTRFS_DC_SETUP; |
5b0e95bf | 2823 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 2824 | |
0af3d00b JB |
2825 | out_put: |
2826 | iput(inode); | |
2827 | out_free: | |
b3b4aa74 | 2828 | btrfs_release_path(path); |
0af3d00b JB |
2829 | out: |
2830 | spin_lock(&block_group->lock); | |
5b0e95bf JB |
2831 | if (!ret) |
2832 | block_group->cache_generation = trans->transid; | |
2b20982e | 2833 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2834 | spin_unlock(&block_group->lock); |
2835 | ||
2836 | return ret; | |
2837 | } | |
2838 | ||
96b5179d CM |
2839 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2840 | struct btrfs_root *root) | |
9078a3e1 | 2841 | { |
4a8c9a62 | 2842 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2843 | int err = 0; |
9078a3e1 | 2844 | struct btrfs_path *path; |
96b5179d | 2845 | u64 last = 0; |
9078a3e1 CM |
2846 | |
2847 | path = btrfs_alloc_path(); | |
2848 | if (!path) | |
2849 | return -ENOMEM; | |
2850 | ||
0af3d00b JB |
2851 | again: |
2852 | while (1) { | |
2853 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2854 | while (cache) { | |
2855 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2856 | break; | |
2857 | cache = next_block_group(root, cache); | |
2858 | } | |
2859 | if (!cache) { | |
2860 | if (last == 0) | |
2861 | break; | |
2862 | last = 0; | |
2863 | continue; | |
2864 | } | |
2865 | err = cache_save_setup(cache, trans, path); | |
2866 | last = cache->key.objectid + cache->key.offset; | |
2867 | btrfs_put_block_group(cache); | |
2868 | } | |
2869 | ||
d397712b | 2870 | while (1) { |
4a8c9a62 YZ |
2871 | if (last == 0) { |
2872 | err = btrfs_run_delayed_refs(trans, root, | |
2873 | (unsigned long)-1); | |
2874 | BUG_ON(err); | |
0f9dd46c | 2875 | } |
54aa1f4d | 2876 | |
4a8c9a62 YZ |
2877 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2878 | while (cache) { | |
0af3d00b JB |
2879 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2880 | btrfs_put_block_group(cache); | |
2881 | goto again; | |
2882 | } | |
2883 | ||
4a8c9a62 YZ |
2884 | if (cache->dirty) |
2885 | break; | |
2886 | cache = next_block_group(root, cache); | |
2887 | } | |
2888 | if (!cache) { | |
2889 | if (last == 0) | |
2890 | break; | |
2891 | last = 0; | |
2892 | continue; | |
2893 | } | |
0f9dd46c | 2894 | |
0cb59c99 JB |
2895 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2896 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2897 | cache->dirty = 0; |
4a8c9a62 | 2898 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2899 | |
4a8c9a62 YZ |
2900 | err = write_one_cache_group(trans, root, path, cache); |
2901 | BUG_ON(err); | |
2902 | btrfs_put_block_group(cache); | |
9078a3e1 | 2903 | } |
4a8c9a62 | 2904 | |
0cb59c99 JB |
2905 | while (1) { |
2906 | /* | |
2907 | * I don't think this is needed since we're just marking our | |
2908 | * preallocated extent as written, but just in case it can't | |
2909 | * hurt. | |
2910 | */ | |
2911 | if (last == 0) { | |
2912 | err = btrfs_run_delayed_refs(trans, root, | |
2913 | (unsigned long)-1); | |
2914 | BUG_ON(err); | |
2915 | } | |
2916 | ||
2917 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2918 | while (cache) { | |
2919 | /* | |
2920 | * Really this shouldn't happen, but it could if we | |
2921 | * couldn't write the entire preallocated extent and | |
2922 | * splitting the extent resulted in a new block. | |
2923 | */ | |
2924 | if (cache->dirty) { | |
2925 | btrfs_put_block_group(cache); | |
2926 | goto again; | |
2927 | } | |
2928 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2929 | break; | |
2930 | cache = next_block_group(root, cache); | |
2931 | } | |
2932 | if (!cache) { | |
2933 | if (last == 0) | |
2934 | break; | |
2935 | last = 0; | |
2936 | continue; | |
2937 | } | |
2938 | ||
2939 | btrfs_write_out_cache(root, trans, cache, path); | |
2940 | ||
2941 | /* | |
2942 | * If we didn't have an error then the cache state is still | |
2943 | * NEED_WRITE, so we can set it to WRITTEN. | |
2944 | */ | |
2945 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2946 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
2947 | last = cache->key.objectid + cache->key.offset; | |
2948 | btrfs_put_block_group(cache); | |
2949 | } | |
2950 | ||
9078a3e1 | 2951 | btrfs_free_path(path); |
4a8c9a62 | 2952 | return 0; |
9078a3e1 CM |
2953 | } |
2954 | ||
d2fb3437 YZ |
2955 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2956 | { | |
2957 | struct btrfs_block_group_cache *block_group; | |
2958 | int readonly = 0; | |
2959 | ||
2960 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2961 | if (!block_group || block_group->ro) | |
2962 | readonly = 1; | |
2963 | if (block_group) | |
fa9c0d79 | 2964 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2965 | return readonly; |
2966 | } | |
2967 | ||
593060d7 CM |
2968 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2969 | u64 total_bytes, u64 bytes_used, | |
2970 | struct btrfs_space_info **space_info) | |
2971 | { | |
2972 | struct btrfs_space_info *found; | |
b742bb82 YZ |
2973 | int i; |
2974 | int factor; | |
2975 | ||
2976 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2977 | BTRFS_BLOCK_GROUP_RAID10)) | |
2978 | factor = 2; | |
2979 | else | |
2980 | factor = 1; | |
593060d7 CM |
2981 | |
2982 | found = __find_space_info(info, flags); | |
2983 | if (found) { | |
25179201 | 2984 | spin_lock(&found->lock); |
593060d7 | 2985 | found->total_bytes += total_bytes; |
89a55897 | 2986 | found->disk_total += total_bytes * factor; |
593060d7 | 2987 | found->bytes_used += bytes_used; |
b742bb82 | 2988 | found->disk_used += bytes_used * factor; |
8f18cf13 | 2989 | found->full = 0; |
25179201 | 2990 | spin_unlock(&found->lock); |
593060d7 CM |
2991 | *space_info = found; |
2992 | return 0; | |
2993 | } | |
c146afad | 2994 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2995 | if (!found) |
2996 | return -ENOMEM; | |
2997 | ||
b742bb82 YZ |
2998 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
2999 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3000 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3001 | spin_lock_init(&found->lock); |
b742bb82 YZ |
3002 | found->flags = flags & (BTRFS_BLOCK_GROUP_DATA | |
3003 | BTRFS_BLOCK_GROUP_SYSTEM | | |
3004 | BTRFS_BLOCK_GROUP_METADATA); | |
593060d7 | 3005 | found->total_bytes = total_bytes; |
89a55897 | 3006 | found->disk_total = total_bytes * factor; |
593060d7 | 3007 | found->bytes_used = bytes_used; |
b742bb82 | 3008 | found->disk_used = bytes_used * factor; |
593060d7 | 3009 | found->bytes_pinned = 0; |
e8569813 | 3010 | found->bytes_reserved = 0; |
c146afad | 3011 | found->bytes_readonly = 0; |
f0486c68 | 3012 | found->bytes_may_use = 0; |
593060d7 | 3013 | found->full = 0; |
0e4f8f88 | 3014 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3015 | found->chunk_alloc = 0; |
fdb5effd JB |
3016 | found->flush = 0; |
3017 | init_waitqueue_head(&found->wait); | |
593060d7 | 3018 | *space_info = found; |
4184ea7f | 3019 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
3020 | return 0; |
3021 | } | |
3022 | ||
8790d502 CM |
3023 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3024 | { | |
3025 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 3026 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 3027 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 3028 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
3029 | if (extra_flags) { |
3030 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3031 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3032 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3033 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3034 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3035 | fs_info->avail_system_alloc_bits |= extra_flags; | |
3036 | } | |
3037 | } | |
593060d7 | 3038 | |
2b82032c | 3039 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3040 | { |
cd02dca5 CM |
3041 | /* |
3042 | * we add in the count of missing devices because we want | |
3043 | * to make sure that any RAID levels on a degraded FS | |
3044 | * continue to be honored. | |
3045 | */ | |
3046 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3047 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d CM |
3048 | |
3049 | if (num_devices == 1) | |
3050 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3051 | if (num_devices < 4) | |
3052 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3053 | ||
ec44a35c CM |
3054 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3055 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3056 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3057 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3058 | } |
ec44a35c CM |
3059 | |
3060 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3061 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3062 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3063 | } |
ec44a35c CM |
3064 | |
3065 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3066 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3067 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
3068 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
3069 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
3070 | return flags; | |
3071 | } | |
3072 | ||
b742bb82 | 3073 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3074 | { |
b742bb82 YZ |
3075 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3076 | flags |= root->fs_info->avail_data_alloc_bits & | |
3077 | root->fs_info->data_alloc_profile; | |
3078 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3079 | flags |= root->fs_info->avail_system_alloc_bits & | |
3080 | root->fs_info->system_alloc_profile; | |
3081 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3082 | flags |= root->fs_info->avail_metadata_alloc_bits & | |
3083 | root->fs_info->metadata_alloc_profile; | |
3084 | return btrfs_reduce_alloc_profile(root, flags); | |
6a63209f JB |
3085 | } |
3086 | ||
6d07bcec | 3087 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3088 | { |
b742bb82 | 3089 | u64 flags; |
9ed74f2d | 3090 | |
b742bb82 YZ |
3091 | if (data) |
3092 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3093 | else if (root == root->fs_info->chunk_root) | |
3094 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3095 | else |
b742bb82 | 3096 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3097 | |
b742bb82 | 3098 | return get_alloc_profile(root, flags); |
6a63209f | 3099 | } |
9ed74f2d | 3100 | |
6a63209f JB |
3101 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3102 | { | |
6a63209f | 3103 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3104 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3105 | } |
3106 | ||
6a63209f | 3107 | /* |
6a63209f JB |
3108 | * This will check the space that the inode allocates from to make sure we have |
3109 | * enough space for bytes. | |
6a63209f | 3110 | */ |
0ca1f7ce | 3111 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3112 | { |
6a63209f | 3113 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3114 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3115 | u64 used; |
0af3d00b | 3116 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3117 | |
6a63209f JB |
3118 | /* make sure bytes are sectorsize aligned */ |
3119 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3120 | |
82d5902d LZ |
3121 | if (root == root->fs_info->tree_root || |
3122 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3123 | alloc_chunk = 0; |
3124 | committed = 1; | |
3125 | } | |
3126 | ||
6a63209f | 3127 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3128 | if (!data_sinfo) |
3129 | goto alloc; | |
9ed74f2d | 3130 | |
6a63209f JB |
3131 | again: |
3132 | /* make sure we have enough space to handle the data first */ | |
3133 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3134 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3135 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3136 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3137 | |
3138 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3139 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3140 | |
6a63209f JB |
3141 | /* |
3142 | * if we don't have enough free bytes in this space then we need | |
3143 | * to alloc a new chunk. | |
3144 | */ | |
0af3d00b | 3145 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3146 | u64 alloc_target; |
9ed74f2d | 3147 | |
0e4f8f88 | 3148 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3149 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3150 | alloc: |
6a63209f | 3151 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3152 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3153 | if (IS_ERR(trans)) |
3154 | return PTR_ERR(trans); | |
9ed74f2d | 3155 | |
6a63209f JB |
3156 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3157 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3158 | alloc_target, |
3159 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3160 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3161 | if (ret < 0) { |
3162 | if (ret != -ENOSPC) | |
3163 | return ret; | |
3164 | else | |
3165 | goto commit_trans; | |
3166 | } | |
9ed74f2d | 3167 | |
33b4d47f CM |
3168 | if (!data_sinfo) { |
3169 | btrfs_set_inode_space_info(root, inode); | |
3170 | data_sinfo = BTRFS_I(inode)->space_info; | |
3171 | } | |
6a63209f JB |
3172 | goto again; |
3173 | } | |
f2bb8f5c JB |
3174 | |
3175 | /* | |
3176 | * If we have less pinned bytes than we want to allocate then | |
3177 | * don't bother committing the transaction, it won't help us. | |
3178 | */ | |
3179 | if (data_sinfo->bytes_pinned < bytes) | |
3180 | committed = 1; | |
6a63209f | 3181 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3182 | |
4e06bdd6 | 3183 | /* commit the current transaction and try again */ |
d52a5b5f | 3184 | commit_trans: |
a4abeea4 JB |
3185 | if (!committed && |
3186 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3187 | committed = 1; |
7a7eaa40 | 3188 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3189 | if (IS_ERR(trans)) |
3190 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3191 | ret = btrfs_commit_transaction(trans, root); |
3192 | if (ret) | |
3193 | return ret; | |
3194 | goto again; | |
3195 | } | |
9ed74f2d | 3196 | |
6a63209f JB |
3197 | return -ENOSPC; |
3198 | } | |
3199 | data_sinfo->bytes_may_use += bytes; | |
6a63209f | 3200 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3201 | |
9ed74f2d | 3202 | return 0; |
9ed74f2d | 3203 | } |
6a63209f | 3204 | |
6a63209f | 3205 | /* |
fb25e914 | 3206 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3207 | */ |
0ca1f7ce | 3208 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3209 | { |
0ca1f7ce | 3210 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3211 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3212 | |
6a63209f JB |
3213 | /* make sure bytes are sectorsize aligned */ |
3214 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3215 | |
6a63209f JB |
3216 | data_sinfo = BTRFS_I(inode)->space_info; |
3217 | spin_lock(&data_sinfo->lock); | |
3218 | data_sinfo->bytes_may_use -= bytes; | |
6a63209f | 3219 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3220 | } |
3221 | ||
97e728d4 | 3222 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3223 | { |
97e728d4 JB |
3224 | struct list_head *head = &info->space_info; |
3225 | struct btrfs_space_info *found; | |
e3ccfa98 | 3226 | |
97e728d4 JB |
3227 | rcu_read_lock(); |
3228 | list_for_each_entry_rcu(found, head, list) { | |
3229 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3230 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3231 | } |
97e728d4 | 3232 | rcu_read_unlock(); |
e3ccfa98 JB |
3233 | } |
3234 | ||
e5bc2458 | 3235 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3236 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3237 | int force) | |
32c00aff | 3238 | { |
fb25e914 | 3239 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3240 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3241 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3242 | u64 thresh; |
e3ccfa98 | 3243 | |
0e4f8f88 CM |
3244 | if (force == CHUNK_ALLOC_FORCE) |
3245 | return 1; | |
3246 | ||
fb25e914 JB |
3247 | /* |
3248 | * We need to take into account the global rsv because for all intents | |
3249 | * and purposes it's used space. Don't worry about locking the | |
3250 | * global_rsv, it doesn't change except when the transaction commits. | |
3251 | */ | |
3252 | num_allocated += global_rsv->size; | |
3253 | ||
0e4f8f88 CM |
3254 | /* |
3255 | * in limited mode, we want to have some free space up to | |
3256 | * about 1% of the FS size. | |
3257 | */ | |
3258 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3259 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3260 | thresh = max_t(u64, 64 * 1024 * 1024, |
3261 | div_factor_fine(thresh, 1)); | |
3262 | ||
3263 | if (num_bytes - num_allocated < thresh) | |
3264 | return 1; | |
3265 | } | |
3266 | ||
3267 | /* | |
3268 | * we have two similar checks here, one based on percentage | |
3269 | * and once based on a hard number of 256MB. The idea | |
3270 | * is that if we have a good amount of free | |
3271 | * room, don't allocate a chunk. A good mount is | |
3272 | * less than 80% utilized of the chunks we have allocated, | |
3273 | * or more than 256MB free | |
3274 | */ | |
3275 | if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes) | |
424499db | 3276 | return 0; |
e3ccfa98 | 3277 | |
0e4f8f88 | 3278 | if (num_allocated + alloc_bytes < div_factor(num_bytes, 8)) |
424499db | 3279 | return 0; |
32c00aff | 3280 | |
6c41761f | 3281 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3282 | |
3283 | /* 256MB or 5% of the FS */ | |
e5bc2458 CM |
3284 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5)); |
3285 | ||
3286 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3)) | |
14ed0ca6 | 3287 | return 0; |
424499db | 3288 | return 1; |
32c00aff JB |
3289 | } |
3290 | ||
6324fbf3 CM |
3291 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3292 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3293 | u64 flags, int force) |
9ed74f2d | 3294 | { |
6324fbf3 | 3295 | struct btrfs_space_info *space_info; |
97e728d4 | 3296 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3297 | int wait_for_alloc = 0; |
9ed74f2d | 3298 | int ret = 0; |
9ed74f2d | 3299 | |
2b82032c | 3300 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3301 | |
6324fbf3 | 3302 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3303 | if (!space_info) { |
3304 | ret = update_space_info(extent_root->fs_info, flags, | |
3305 | 0, 0, &space_info); | |
3306 | BUG_ON(ret); | |
9ed74f2d | 3307 | } |
6324fbf3 | 3308 | BUG_ON(!space_info); |
9ed74f2d | 3309 | |
6d74119f | 3310 | again: |
25179201 | 3311 | spin_lock(&space_info->lock); |
9ed74f2d | 3312 | if (space_info->force_alloc) |
0e4f8f88 | 3313 | force = space_info->force_alloc; |
25179201 JB |
3314 | if (space_info->full) { |
3315 | spin_unlock(&space_info->lock); | |
6d74119f | 3316 | return 0; |
9ed74f2d JB |
3317 | } |
3318 | ||
0e4f8f88 | 3319 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3320 | spin_unlock(&space_info->lock); |
6d74119f JB |
3321 | return 0; |
3322 | } else if (space_info->chunk_alloc) { | |
3323 | wait_for_alloc = 1; | |
3324 | } else { | |
3325 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3326 | } |
0e4f8f88 | 3327 | |
25179201 | 3328 | spin_unlock(&space_info->lock); |
9ed74f2d | 3329 | |
6d74119f JB |
3330 | mutex_lock(&fs_info->chunk_mutex); |
3331 | ||
3332 | /* | |
3333 | * The chunk_mutex is held throughout the entirety of a chunk | |
3334 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3335 | * other guy is done and we need to recheck and see if we should | |
3336 | * allocate. | |
3337 | */ | |
3338 | if (wait_for_alloc) { | |
3339 | mutex_unlock(&fs_info->chunk_mutex); | |
3340 | wait_for_alloc = 0; | |
3341 | goto again; | |
3342 | } | |
3343 | ||
67377734 JB |
3344 | /* |
3345 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3346 | * allocating mixed chunks instead of individual chunks. | |
3347 | */ | |
3348 | if (btrfs_mixed_space_info(space_info)) | |
3349 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3350 | ||
97e728d4 JB |
3351 | /* |
3352 | * if we're doing a data chunk, go ahead and make sure that | |
3353 | * we keep a reasonable number of metadata chunks allocated in the | |
3354 | * FS as well. | |
3355 | */ | |
9ed74f2d | 3356 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3357 | fs_info->data_chunk_allocations++; |
3358 | if (!(fs_info->data_chunk_allocations % | |
3359 | fs_info->metadata_ratio)) | |
3360 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3361 | } |
3362 | ||
2b82032c | 3363 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3364 | if (ret < 0 && ret != -ENOSPC) |
3365 | goto out; | |
3366 | ||
9ed74f2d | 3367 | spin_lock(&space_info->lock); |
9ed74f2d | 3368 | if (ret) |
6324fbf3 | 3369 | space_info->full = 1; |
424499db YZ |
3370 | else |
3371 | ret = 1; | |
6d74119f | 3372 | |
0e4f8f88 | 3373 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3374 | space_info->chunk_alloc = 0; |
9ed74f2d | 3375 | spin_unlock(&space_info->lock); |
92b8e897 | 3376 | out: |
c146afad | 3377 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3378 | return ret; |
6324fbf3 | 3379 | } |
9ed74f2d | 3380 | |
9ed74f2d | 3381 | /* |
5da9d01b | 3382 | * shrink metadata reservation for delalloc |
9ed74f2d | 3383 | */ |
663350ac | 3384 | static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, |
f104d044 | 3385 | bool wait_ordered) |
5da9d01b | 3386 | { |
0ca1f7ce | 3387 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3388 | struct btrfs_space_info *space_info; |
663350ac | 3389 | struct btrfs_trans_handle *trans; |
5da9d01b YZ |
3390 | u64 reserved; |
3391 | u64 max_reclaim; | |
3392 | u64 reclaimed = 0; | |
b1953bce | 3393 | long time_left; |
877da174 | 3394 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3395 | int loops = 0; |
36e39c40 | 3396 | unsigned long progress; |
5da9d01b | 3397 | |
663350ac | 3398 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3399 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3400 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3401 | |
3402 | smp_mb(); | |
fb25e914 | 3403 | reserved = space_info->bytes_may_use; |
36e39c40 | 3404 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3405 | |
3406 | if (reserved == 0) | |
3407 | return 0; | |
c4f675cd | 3408 | |
fdb5effd JB |
3409 | smp_mb(); |
3410 | if (root->fs_info->delalloc_bytes == 0) { | |
3411 | if (trans) | |
3412 | return 0; | |
3413 | btrfs_wait_ordered_extents(root, 0, 0); | |
3414 | return 0; | |
3415 | } | |
3416 | ||
5da9d01b | 3417 | max_reclaim = min(reserved, to_reclaim); |
877da174 JB |
3418 | nr_pages = max_t(unsigned long, nr_pages, |
3419 | max_reclaim >> PAGE_CACHE_SHIFT); | |
b1953bce | 3420 | while (loops < 1024) { |
bf9022e0 CM |
3421 | /* have the flusher threads jump in and do some IO */ |
3422 | smp_mb(); | |
3423 | nr_pages = min_t(unsigned long, nr_pages, | |
3424 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
3425 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages); | |
5da9d01b | 3426 | |
0019f10d | 3427 | spin_lock(&space_info->lock); |
fb25e914 JB |
3428 | if (reserved > space_info->bytes_may_use) |
3429 | reclaimed += reserved - space_info->bytes_may_use; | |
3430 | reserved = space_info->bytes_may_use; | |
0019f10d | 3431 | spin_unlock(&space_info->lock); |
5da9d01b | 3432 | |
36e39c40 CM |
3433 | loops++; |
3434 | ||
5da9d01b YZ |
3435 | if (reserved == 0 || reclaimed >= max_reclaim) |
3436 | break; | |
3437 | ||
3438 | if (trans && trans->transaction->blocked) | |
3439 | return -EAGAIN; | |
bf9022e0 | 3440 | |
f104d044 JB |
3441 | if (wait_ordered && !trans) { |
3442 | btrfs_wait_ordered_extents(root, 0, 0); | |
3443 | } else { | |
3444 | time_left = schedule_timeout_interruptible(1); | |
b1953bce | 3445 | |
f104d044 JB |
3446 | /* We were interrupted, exit */ |
3447 | if (time_left) | |
3448 | break; | |
3449 | } | |
b1953bce | 3450 | |
36e39c40 CM |
3451 | /* we've kicked the IO a few times, if anything has been freed, |
3452 | * exit. There is no sense in looping here for a long time | |
3453 | * when we really need to commit the transaction, or there are | |
3454 | * just too many writers without enough free space | |
3455 | */ | |
3456 | ||
3457 | if (loops > 3) { | |
3458 | smp_mb(); | |
3459 | if (progress != space_info->reservation_progress) | |
3460 | break; | |
3461 | } | |
bf9022e0 | 3462 | |
5da9d01b | 3463 | } |
f104d044 | 3464 | |
5da9d01b YZ |
3465 | return reclaimed >= to_reclaim; |
3466 | } | |
3467 | ||
663350ac JB |
3468 | /** |
3469 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3470 | * @root - the root we're allocating for | |
3471 | * @bytes - the number of bytes we want to reserve | |
3472 | * @force - force the commit | |
3473 | * | |
3474 | * This will check to make sure that committing the transaction will actually | |
3475 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3476 | * will return -ENOSPC. | |
3477 | */ | |
3478 | static int may_commit_transaction(struct btrfs_root *root, | |
3479 | struct btrfs_space_info *space_info, | |
3480 | u64 bytes, int force) | |
3481 | { | |
3482 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3483 | struct btrfs_trans_handle *trans; | |
3484 | ||
3485 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3486 | if (trans) | |
3487 | return -EAGAIN; | |
3488 | ||
3489 | if (force) | |
3490 | goto commit; | |
3491 | ||
3492 | /* See if there is enough pinned space to make this reservation */ | |
3493 | spin_lock(&space_info->lock); | |
3494 | if (space_info->bytes_pinned >= bytes) { | |
3495 | spin_unlock(&space_info->lock); | |
3496 | goto commit; | |
3497 | } | |
3498 | spin_unlock(&space_info->lock); | |
3499 | ||
3500 | /* | |
3501 | * See if there is some space in the delayed insertion reservation for | |
3502 | * this reservation. | |
3503 | */ | |
3504 | if (space_info != delayed_rsv->space_info) | |
3505 | return -ENOSPC; | |
3506 | ||
3507 | spin_lock(&delayed_rsv->lock); | |
3508 | if (delayed_rsv->size < bytes) { | |
3509 | spin_unlock(&delayed_rsv->lock); | |
3510 | return -ENOSPC; | |
3511 | } | |
3512 | spin_unlock(&delayed_rsv->lock); | |
3513 | ||
3514 | commit: | |
3515 | trans = btrfs_join_transaction(root); | |
3516 | if (IS_ERR(trans)) | |
3517 | return -ENOSPC; | |
3518 | ||
3519 | return btrfs_commit_transaction(trans, root); | |
3520 | } | |
3521 | ||
4a92b1b8 JB |
3522 | /** |
3523 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3524 | * @root - the root we're allocating for | |
3525 | * @block_rsv - the block_rsv we're allocating for | |
3526 | * @orig_bytes - the number of bytes we want | |
3527 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3528 | * |
4a92b1b8 JB |
3529 | * This will reserve orgi_bytes number of bytes from the space info associated |
3530 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3531 | * flush out space to make room. It will do this by flushing delalloc if | |
3532 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3533 | * regain reservations will be made and this will fail if there is not enough | |
3534 | * space already. | |
8bb8ab2e | 3535 | */ |
4a92b1b8 | 3536 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 3537 | struct btrfs_block_rsv *block_rsv, |
36ba022a | 3538 | u64 orig_bytes, int flush) |
9ed74f2d | 3539 | { |
f0486c68 | 3540 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3541 | u64 used; |
8bb8ab2e JB |
3542 | u64 num_bytes = orig_bytes; |
3543 | int retries = 0; | |
3544 | int ret = 0; | |
38227933 | 3545 | bool committed = false; |
fdb5effd | 3546 | bool flushing = false; |
f104d044 | 3547 | bool wait_ordered = false; |
4a92b1b8 | 3548 | |
8bb8ab2e | 3549 | again: |
fdb5effd | 3550 | ret = 0; |
8bb8ab2e | 3551 | spin_lock(&space_info->lock); |
fdb5effd JB |
3552 | /* |
3553 | * We only want to wait if somebody other than us is flushing and we are | |
3554 | * actually alloed to flush. | |
3555 | */ | |
3556 | while (flush && !flushing && space_info->flush) { | |
3557 | spin_unlock(&space_info->lock); | |
3558 | /* | |
3559 | * If we have a trans handle we can't wait because the flusher | |
3560 | * may have to commit the transaction, which would mean we would | |
3561 | * deadlock since we are waiting for the flusher to finish, but | |
3562 | * hold the current transaction open. | |
3563 | */ | |
663350ac | 3564 | if (current->journal_info) |
fdb5effd JB |
3565 | return -EAGAIN; |
3566 | ret = wait_event_interruptible(space_info->wait, | |
3567 | !space_info->flush); | |
3568 | /* Must have been interrupted, return */ | |
3569 | if (ret) | |
3570 | return -EINTR; | |
3571 | ||
3572 | spin_lock(&space_info->lock); | |
3573 | } | |
3574 | ||
3575 | ret = -ENOSPC; | |
2bf64758 JB |
3576 | used = space_info->bytes_used + space_info->bytes_reserved + |
3577 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3578 | space_info->bytes_may_use; | |
9ed74f2d | 3579 | |
8bb8ab2e JB |
3580 | /* |
3581 | * The idea here is that we've not already over-reserved the block group | |
3582 | * then we can go ahead and save our reservation first and then start | |
3583 | * flushing if we need to. Otherwise if we've already overcommitted | |
3584 | * lets start flushing stuff first and then come back and try to make | |
3585 | * our reservation. | |
3586 | */ | |
2bf64758 JB |
3587 | if (used <= space_info->total_bytes) { |
3588 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3589 | space_info->bytes_may_use += orig_bytes; |
8bb8ab2e JB |
3590 | ret = 0; |
3591 | } else { | |
3592 | /* | |
3593 | * Ok set num_bytes to orig_bytes since we aren't | |
3594 | * overocmmitted, this way we only try and reclaim what | |
3595 | * we need. | |
3596 | */ | |
3597 | num_bytes = orig_bytes; | |
3598 | } | |
3599 | } else { | |
3600 | /* | |
3601 | * Ok we're over committed, set num_bytes to the overcommitted | |
3602 | * amount plus the amount of bytes that we need for this | |
3603 | * reservation. | |
3604 | */ | |
f104d044 | 3605 | wait_ordered = true; |
2bf64758 | 3606 | num_bytes = used - space_info->total_bytes + |
8bb8ab2e JB |
3607 | (orig_bytes * (retries + 1)); |
3608 | } | |
9ed74f2d | 3609 | |
36ba022a | 3610 | if (ret) { |
2bf64758 JB |
3611 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3612 | u64 avail; | |
3613 | ||
7e355b83 JB |
3614 | /* |
3615 | * If we have a lot of space that's pinned, don't bother doing | |
3616 | * the overcommit dance yet and just commit the transaction. | |
3617 | */ | |
3618 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3619 | do_div(avail, 10); | |
663350ac | 3620 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3621 | space_info->flush = 1; |
3622 | flushing = true; | |
3623 | spin_unlock(&space_info->lock); | |
663350ac JB |
3624 | ret = may_commit_transaction(root, space_info, |
3625 | orig_bytes, 1); | |
3626 | if (ret) | |
3627 | goto out; | |
3628 | committed = true; | |
3629 | goto again; | |
7e355b83 JB |
3630 | } |
3631 | ||
2bf64758 JB |
3632 | spin_lock(&root->fs_info->free_chunk_lock); |
3633 | avail = root->fs_info->free_chunk_space; | |
3634 | ||
3635 | /* | |
3636 | * If we have dup, raid1 or raid10 then only half of the free | |
3637 | * space is actually useable. | |
3638 | */ | |
3639 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3640 | BTRFS_BLOCK_GROUP_RAID1 | | |
3641 | BTRFS_BLOCK_GROUP_RAID10)) | |
3642 | avail >>= 1; | |
3643 | ||
3644 | /* | |
3645 | * If we aren't flushing don't let us overcommit too much, say | |
3646 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3647 | * 1/2 of the space. | |
3648 | */ | |
3649 | if (flush) | |
3650 | avail >>= 3; | |
3651 | else | |
3652 | avail >>= 1; | |
3653 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3654 | ||
9a82ca65 | 3655 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 JB |
3656 | space_info->bytes_may_use += orig_bytes; |
3657 | ret = 0; | |
f104d044 JB |
3658 | } else { |
3659 | wait_ordered = true; | |
2bf64758 JB |
3660 | } |
3661 | } | |
3662 | ||
8bb8ab2e JB |
3663 | /* |
3664 | * Couldn't make our reservation, save our place so while we're trying | |
3665 | * to reclaim space we can actually use it instead of somebody else | |
3666 | * stealing it from us. | |
3667 | */ | |
fdb5effd JB |
3668 | if (ret && flush) { |
3669 | flushing = true; | |
3670 | space_info->flush = 1; | |
8bb8ab2e | 3671 | } |
9ed74f2d | 3672 | |
f0486c68 | 3673 | spin_unlock(&space_info->lock); |
9ed74f2d | 3674 | |
fdb5effd | 3675 | if (!ret || !flush) |
8bb8ab2e | 3676 | goto out; |
f0486c68 | 3677 | |
8bb8ab2e JB |
3678 | /* |
3679 | * We do synchronous shrinking since we don't actually unreserve | |
3680 | * metadata until after the IO is completed. | |
3681 | */ | |
663350ac | 3682 | ret = shrink_delalloc(root, num_bytes, wait_ordered); |
fdb5effd | 3683 | if (ret < 0) |
8bb8ab2e | 3684 | goto out; |
f0486c68 | 3685 | |
75c195a2 CM |
3686 | ret = 0; |
3687 | ||
8bb8ab2e JB |
3688 | /* |
3689 | * So if we were overcommitted it's possible that somebody else flushed | |
3690 | * out enough space and we simply didn't have enough space to reclaim, | |
3691 | * so go back around and try again. | |
3692 | */ | |
3693 | if (retries < 2) { | |
f104d044 | 3694 | wait_ordered = true; |
8bb8ab2e JB |
3695 | retries++; |
3696 | goto again; | |
3697 | } | |
f0486c68 | 3698 | |
8bb8ab2e | 3699 | ret = -ENOSPC; |
75c195a2 CM |
3700 | if (committed) |
3701 | goto out; | |
3702 | ||
663350ac | 3703 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); |
38227933 | 3704 | if (!ret) { |
38227933 | 3705 | committed = true; |
8bb8ab2e | 3706 | goto again; |
38227933 | 3707 | } |
8bb8ab2e JB |
3708 | |
3709 | out: | |
fdb5effd | 3710 | if (flushing) { |
8bb8ab2e | 3711 | spin_lock(&space_info->lock); |
fdb5effd JB |
3712 | space_info->flush = 0; |
3713 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3714 | spin_unlock(&space_info->lock); |
f0486c68 | 3715 | } |
f0486c68 YZ |
3716 | return ret; |
3717 | } | |
3718 | ||
3719 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3720 | struct btrfs_root *root) | |
3721 | { | |
4c13d758 JB |
3722 | struct btrfs_block_rsv *block_rsv = NULL; |
3723 | ||
3724 | if (root->ref_cows || root == root->fs_info->csum_root) | |
f0486c68 | 3725 | block_rsv = trans->block_rsv; |
4c13d758 JB |
3726 | |
3727 | if (!block_rsv) | |
f0486c68 YZ |
3728 | block_rsv = root->block_rsv; |
3729 | ||
3730 | if (!block_rsv) | |
3731 | block_rsv = &root->fs_info->empty_block_rsv; | |
3732 | ||
3733 | return block_rsv; | |
3734 | } | |
3735 | ||
3736 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3737 | u64 num_bytes) | |
3738 | { | |
3739 | int ret = -ENOSPC; | |
3740 | spin_lock(&block_rsv->lock); | |
3741 | if (block_rsv->reserved >= num_bytes) { | |
3742 | block_rsv->reserved -= num_bytes; | |
3743 | if (block_rsv->reserved < block_rsv->size) | |
3744 | block_rsv->full = 0; | |
3745 | ret = 0; | |
3746 | } | |
3747 | spin_unlock(&block_rsv->lock); | |
3748 | return ret; | |
3749 | } | |
3750 | ||
3751 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3752 | u64 num_bytes, int update_size) | |
3753 | { | |
3754 | spin_lock(&block_rsv->lock); | |
3755 | block_rsv->reserved += num_bytes; | |
3756 | if (update_size) | |
3757 | block_rsv->size += num_bytes; | |
3758 | else if (block_rsv->reserved >= block_rsv->size) | |
3759 | block_rsv->full = 1; | |
3760 | spin_unlock(&block_rsv->lock); | |
3761 | } | |
3762 | ||
62a45b60 DS |
3763 | static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv, |
3764 | struct btrfs_block_rsv *dest, u64 num_bytes) | |
f0486c68 YZ |
3765 | { |
3766 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3767 | ||
3768 | spin_lock(&block_rsv->lock); | |
3769 | if (num_bytes == (u64)-1) | |
3770 | num_bytes = block_rsv->size; | |
3771 | block_rsv->size -= num_bytes; | |
3772 | if (block_rsv->reserved >= block_rsv->size) { | |
3773 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3774 | block_rsv->reserved = block_rsv->size; | |
3775 | block_rsv->full = 1; | |
3776 | } else { | |
3777 | num_bytes = 0; | |
3778 | } | |
3779 | spin_unlock(&block_rsv->lock); | |
3780 | ||
3781 | if (num_bytes > 0) { | |
3782 | if (dest) { | |
e9e22899 JB |
3783 | spin_lock(&dest->lock); |
3784 | if (!dest->full) { | |
3785 | u64 bytes_to_add; | |
3786 | ||
3787 | bytes_to_add = dest->size - dest->reserved; | |
3788 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3789 | dest->reserved += bytes_to_add; | |
3790 | if (dest->reserved >= dest->size) | |
3791 | dest->full = 1; | |
3792 | num_bytes -= bytes_to_add; | |
3793 | } | |
3794 | spin_unlock(&dest->lock); | |
3795 | } | |
3796 | if (num_bytes) { | |
f0486c68 | 3797 | spin_lock(&space_info->lock); |
fb25e914 | 3798 | space_info->bytes_may_use -= num_bytes; |
36e39c40 | 3799 | space_info->reservation_progress++; |
f0486c68 | 3800 | spin_unlock(&space_info->lock); |
4e06bdd6 | 3801 | } |
9ed74f2d | 3802 | } |
f0486c68 | 3803 | } |
4e06bdd6 | 3804 | |
f0486c68 YZ |
3805 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3806 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3807 | { | |
3808 | int ret; | |
9ed74f2d | 3809 | |
f0486c68 YZ |
3810 | ret = block_rsv_use_bytes(src, num_bytes); |
3811 | if (ret) | |
3812 | return ret; | |
9ed74f2d | 3813 | |
f0486c68 | 3814 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3815 | return 0; |
3816 | } | |
3817 | ||
f0486c68 | 3818 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3819 | { |
f0486c68 YZ |
3820 | memset(rsv, 0, sizeof(*rsv)); |
3821 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
3822 | } |
3823 | ||
3824 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3825 | { | |
3826 | struct btrfs_block_rsv *block_rsv; | |
3827 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3828 | |
f0486c68 YZ |
3829 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3830 | if (!block_rsv) | |
3831 | return NULL; | |
9ed74f2d | 3832 | |
f0486c68 | 3833 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3834 | block_rsv->space_info = __find_space_info(fs_info, |
3835 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3836 | return block_rsv; |
3837 | } | |
9ed74f2d | 3838 | |
f0486c68 YZ |
3839 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3840 | struct btrfs_block_rsv *rsv) | |
3841 | { | |
dabdb640 JB |
3842 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
3843 | kfree(rsv); | |
9ed74f2d JB |
3844 | } |
3845 | ||
61b520a9 MX |
3846 | static inline int __block_rsv_add(struct btrfs_root *root, |
3847 | struct btrfs_block_rsv *block_rsv, | |
3848 | u64 num_bytes, int flush) | |
f0486c68 YZ |
3849 | { |
3850 | int ret; | |
9ed74f2d | 3851 | |
f0486c68 YZ |
3852 | if (num_bytes == 0) |
3853 | return 0; | |
8bb8ab2e | 3854 | |
61b520a9 | 3855 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
3856 | if (!ret) { |
3857 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3858 | return 0; | |
3859 | } | |
9ed74f2d | 3860 | |
f0486c68 YZ |
3861 | return ret; |
3862 | } | |
9ed74f2d | 3863 | |
61b520a9 MX |
3864 | int btrfs_block_rsv_add(struct btrfs_root *root, |
3865 | struct btrfs_block_rsv *block_rsv, | |
3866 | u64 num_bytes) | |
3867 | { | |
3868 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
3869 | } | |
3870 | ||
c06a0e12 JB |
3871 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
3872 | struct btrfs_block_rsv *block_rsv, | |
3873 | u64 num_bytes) | |
3874 | { | |
61b520a9 | 3875 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
c06a0e12 JB |
3876 | } |
3877 | ||
4a92b1b8 | 3878 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 3879 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
3880 | { |
3881 | u64 num_bytes = 0; | |
f0486c68 | 3882 | int ret = -ENOSPC; |
9ed74f2d | 3883 | |
f0486c68 YZ |
3884 | if (!block_rsv) |
3885 | return 0; | |
9ed74f2d | 3886 | |
f0486c68 | 3887 | spin_lock(&block_rsv->lock); |
36ba022a JB |
3888 | num_bytes = div_factor(block_rsv->size, min_factor); |
3889 | if (block_rsv->reserved >= num_bytes) | |
3890 | ret = 0; | |
3891 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 3892 | |
36ba022a JB |
3893 | return ret; |
3894 | } | |
3895 | ||
aa38a711 MX |
3896 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
3897 | struct btrfs_block_rsv *block_rsv, | |
3898 | u64 min_reserved, int flush) | |
36ba022a JB |
3899 | { |
3900 | u64 num_bytes = 0; | |
3901 | int ret = -ENOSPC; | |
3902 | ||
3903 | if (!block_rsv) | |
3904 | return 0; | |
3905 | ||
3906 | spin_lock(&block_rsv->lock); | |
3907 | num_bytes = min_reserved; | |
13553e52 | 3908 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 3909 | ret = 0; |
13553e52 | 3910 | else |
f0486c68 | 3911 | num_bytes -= block_rsv->reserved; |
f0486c68 | 3912 | spin_unlock(&block_rsv->lock); |
13553e52 | 3913 | |
f0486c68 YZ |
3914 | if (!ret) |
3915 | return 0; | |
3916 | ||
aa38a711 | 3917 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
3918 | if (!ret) { |
3919 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
3920 | return 0; | |
f0486c68 | 3921 | } |
9ed74f2d | 3922 | |
13553e52 | 3923 | return ret; |
f0486c68 YZ |
3924 | } |
3925 | ||
aa38a711 MX |
3926 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
3927 | struct btrfs_block_rsv *block_rsv, | |
3928 | u64 min_reserved) | |
3929 | { | |
3930 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
3931 | } | |
3932 | ||
3933 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
3934 | struct btrfs_block_rsv *block_rsv, | |
3935 | u64 min_reserved) | |
3936 | { | |
3937 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
3938 | } | |
3939 | ||
f0486c68 YZ |
3940 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
3941 | struct btrfs_block_rsv *dst_rsv, | |
3942 | u64 num_bytes) | |
3943 | { | |
3944 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
3945 | } | |
3946 | ||
3947 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
3948 | struct btrfs_block_rsv *block_rsv, | |
3949 | u64 num_bytes) | |
3950 | { | |
3951 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
3952 | if (global_rsv->full || global_rsv == block_rsv || | |
3953 | block_rsv->space_info != global_rsv->space_info) | |
3954 | global_rsv = NULL; | |
3955 | block_rsv_release_bytes(block_rsv, global_rsv, num_bytes); | |
6a63209f JB |
3956 | } |
3957 | ||
3958 | /* | |
8929ecfa YZ |
3959 | * helper to calculate size of global block reservation. |
3960 | * the desired value is sum of space used by extent tree, | |
3961 | * checksum tree and root tree | |
6a63209f | 3962 | */ |
8929ecfa | 3963 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 3964 | { |
8929ecfa YZ |
3965 | struct btrfs_space_info *sinfo; |
3966 | u64 num_bytes; | |
3967 | u64 meta_used; | |
3968 | u64 data_used; | |
6c41761f | 3969 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 3970 | |
8929ecfa YZ |
3971 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
3972 | spin_lock(&sinfo->lock); | |
3973 | data_used = sinfo->bytes_used; | |
3974 | spin_unlock(&sinfo->lock); | |
33b4d47f | 3975 | |
8929ecfa YZ |
3976 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
3977 | spin_lock(&sinfo->lock); | |
6d48755d JB |
3978 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
3979 | data_used = 0; | |
8929ecfa YZ |
3980 | meta_used = sinfo->bytes_used; |
3981 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 3982 | |
8929ecfa YZ |
3983 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
3984 | csum_size * 2; | |
3985 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 3986 | |
8929ecfa YZ |
3987 | if (num_bytes * 3 > meta_used) |
3988 | num_bytes = div64_u64(meta_used, 3); | |
ab6e2410 | 3989 | |
8929ecfa YZ |
3990 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
3991 | } | |
6a63209f | 3992 | |
8929ecfa YZ |
3993 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
3994 | { | |
3995 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
3996 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
3997 | u64 num_bytes; | |
6a63209f | 3998 | |
8929ecfa | 3999 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4000 | |
8929ecfa YZ |
4001 | spin_lock(&block_rsv->lock); |
4002 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 4003 | |
8929ecfa | 4004 | block_rsv->size = num_bytes; |
4e06bdd6 | 4005 | |
8929ecfa | 4006 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4007 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4008 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4009 | |
4010 | if (sinfo->total_bytes > num_bytes) { | |
4011 | num_bytes = sinfo->total_bytes - num_bytes; | |
4012 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4013 | sinfo->bytes_may_use += num_bytes; |
6a63209f | 4014 | } |
6a63209f | 4015 | |
8929ecfa YZ |
4016 | if (block_rsv->reserved >= block_rsv->size) { |
4017 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4018 | sinfo->bytes_may_use -= num_bytes; |
36e39c40 | 4019 | sinfo->reservation_progress++; |
8929ecfa YZ |
4020 | block_rsv->reserved = block_rsv->size; |
4021 | block_rsv->full = 1; | |
4022 | } | |
182608c8 | 4023 | |
8929ecfa YZ |
4024 | spin_unlock(&sinfo->lock); |
4025 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
4026 | } |
4027 | ||
f0486c68 | 4028 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4029 | { |
f0486c68 | 4030 | struct btrfs_space_info *space_info; |
6a63209f | 4031 | |
f0486c68 YZ |
4032 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4033 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4034 | |
f0486c68 | 4035 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4036 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4037 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4038 | fs_info->trans_block_rsv.space_info = space_info; |
4039 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4040 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4041 | |
8929ecfa YZ |
4042 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4043 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4044 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4045 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4046 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4047 | |
8929ecfa | 4048 | update_global_block_rsv(fs_info); |
6a63209f JB |
4049 | } |
4050 | ||
8929ecfa | 4051 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4052 | { |
8929ecfa YZ |
4053 | block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1); |
4054 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); | |
4055 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4056 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4057 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4058 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4059 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4060 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4061 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
f0486c68 | 4062 | } |
6a63209f | 4063 | |
a22285a6 YZ |
4064 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4065 | struct btrfs_root *root) | |
6a63209f | 4066 | { |
a22285a6 YZ |
4067 | if (!trans->bytes_reserved) |
4068 | return; | |
6a63209f | 4069 | |
b24e03db | 4070 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4071 | trans->bytes_reserved = 0; |
4072 | } | |
6a63209f | 4073 | |
d68fc57b YZ |
4074 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4075 | struct inode *inode) | |
4076 | { | |
4077 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4078 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4079 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4080 | ||
4081 | /* | |
fcb80c2a JB |
4082 | * We need to hold space in order to delete our orphan item once we've |
4083 | * added it, so this takes the reservation so we can release it later | |
4084 | * when we are truly done with the orphan item. | |
d68fc57b | 4085 | */ |
ff5714cc | 4086 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
d68fc57b | 4087 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4088 | } |
4089 | ||
d68fc57b | 4090 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4091 | { |
d68fc57b | 4092 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4093 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
d68fc57b YZ |
4094 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4095 | } | |
97e728d4 | 4096 | |
a22285a6 YZ |
4097 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4098 | struct btrfs_pending_snapshot *pending) | |
4099 | { | |
4100 | struct btrfs_root *root = pending->root; | |
4101 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4102 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4103 | /* | |
4104 | * two for root back/forward refs, two for directory entries | |
4105 | * and one for root of the snapshot. | |
4106 | */ | |
16cdcec7 | 4107 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4108 | dst_rsv->space_info = src_rsv->space_info; |
4109 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4110 | } |
4111 | ||
7709cde3 JB |
4112 | /** |
4113 | * drop_outstanding_extent - drop an outstanding extent | |
4114 | * @inode: the inode we're dropping the extent for | |
4115 | * | |
4116 | * This is called when we are freeing up an outstanding extent, either called | |
4117 | * after an error or after an extent is written. This will return the number of | |
4118 | * reserved extents that need to be freed. This must be called with | |
4119 | * BTRFS_I(inode)->lock held. | |
4120 | */ | |
9e0baf60 JB |
4121 | static unsigned drop_outstanding_extent(struct inode *inode) |
4122 | { | |
7fd2ae21 | 4123 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4124 | unsigned dropped_extents = 0; |
4125 | ||
9e0baf60 JB |
4126 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4127 | BTRFS_I(inode)->outstanding_extents--; | |
4128 | ||
7fd2ae21 JB |
4129 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
4130 | BTRFS_I(inode)->delalloc_meta_reserved) { | |
4131 | drop_inode_space = 1; | |
4132 | BTRFS_I(inode)->delalloc_meta_reserved = 0; | |
4133 | } | |
4134 | ||
9e0baf60 JB |
4135 | /* |
4136 | * If we have more or the same amount of outsanding extents than we have | |
4137 | * reserved then we need to leave the reserved extents count alone. | |
4138 | */ | |
4139 | if (BTRFS_I(inode)->outstanding_extents >= | |
4140 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4141 | return drop_inode_space; |
9e0baf60 JB |
4142 | |
4143 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4144 | BTRFS_I(inode)->outstanding_extents; | |
4145 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4146 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4147 | } |
4148 | ||
7709cde3 JB |
4149 | /** |
4150 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4151 | * reserved/free'd for the given bytes. | |
4152 | * @inode: the inode we're manipulating | |
4153 | * @num_bytes: the number of bytes in question | |
4154 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4155 | * | |
4156 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4157 | * correct amount of metadata that must either be reserved or freed. We | |
4158 | * calculate how many checksums we can fit into one leaf and then divide the | |
4159 | * number of bytes that will need to be checksumed by this value to figure out | |
4160 | * how many checksums will be required. If we are adding bytes then the number | |
4161 | * may go up and we will return the number of additional bytes that must be | |
4162 | * reserved. If it is going down we will return the number of bytes that must | |
4163 | * be freed. | |
4164 | * | |
4165 | * This must be called with BTRFS_I(inode)->lock held. | |
4166 | */ | |
4167 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4168 | int reserve) | |
6324fbf3 | 4169 | { |
7709cde3 JB |
4170 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4171 | u64 csum_size; | |
4172 | int num_csums_per_leaf; | |
4173 | int num_csums; | |
4174 | int old_csums; | |
4175 | ||
4176 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4177 | BTRFS_I(inode)->csum_bytes == 0) | |
4178 | return 0; | |
4179 | ||
4180 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4181 | if (reserve) | |
4182 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4183 | else | |
4184 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4185 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4186 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4187 | sizeof(struct btrfs_csum_item) + | |
4188 | sizeof(struct btrfs_disk_key)); | |
4189 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4190 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4191 | num_csums = num_csums / num_csums_per_leaf; | |
4192 | ||
4193 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4194 | old_csums = old_csums / num_csums_per_leaf; | |
4195 | ||
4196 | /* No change, no need to reserve more */ | |
4197 | if (old_csums == num_csums) | |
4198 | return 0; | |
4199 | ||
4200 | if (reserve) | |
4201 | return btrfs_calc_trans_metadata_size(root, | |
4202 | num_csums - old_csums); | |
4203 | ||
4204 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4205 | } |
c146afad | 4206 | |
0ca1f7ce YZ |
4207 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4208 | { | |
4209 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4210 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 JB |
4211 | u64 to_reserve = 0; |
4212 | unsigned nr_extents = 0; | |
c09544e0 | 4213 | int flush = 1; |
0ca1f7ce | 4214 | int ret; |
6324fbf3 | 4215 | |
c09544e0 JB |
4216 | if (btrfs_is_free_space_inode(root, inode)) |
4217 | flush = 0; | |
4218 | ||
4219 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4220 | schedule_timeout(1); |
ec44a35c | 4221 | |
0ca1f7ce | 4222 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4223 | |
9e0baf60 JB |
4224 | spin_lock(&BTRFS_I(inode)->lock); |
4225 | BTRFS_I(inode)->outstanding_extents++; | |
4226 | ||
4227 | if (BTRFS_I(inode)->outstanding_extents > | |
4228 | BTRFS_I(inode)->reserved_extents) { | |
4229 | nr_extents = BTRFS_I(inode)->outstanding_extents - | |
4230 | BTRFS_I(inode)->reserved_extents; | |
4231 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
7fd2ae21 | 4232 | } |
57a45ced | 4233 | |
7fd2ae21 JB |
4234 | /* |
4235 | * Add an item to reserve for updating the inode when we complete the | |
4236 | * delalloc io. | |
4237 | */ | |
4238 | if (!BTRFS_I(inode)->delalloc_meta_reserved) { | |
4239 | nr_extents++; | |
4240 | BTRFS_I(inode)->delalloc_meta_reserved = 1; | |
593060d7 | 4241 | } |
7fd2ae21 JB |
4242 | |
4243 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4244 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
9e0baf60 | 4245 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4246 | |
36ba022a | 4247 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4248 | if (ret) { |
7ed49f18 | 4249 | u64 to_free = 0; |
9e0baf60 | 4250 | unsigned dropped; |
7ed49f18 | 4251 | |
7709cde3 | 4252 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4253 | dropped = drop_outstanding_extent(inode); |
7ed49f18 | 4254 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
7709cde3 | 4255 | spin_unlock(&BTRFS_I(inode)->lock); |
7ed49f18 JB |
4256 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
4257 | ||
4258 | /* | |
4259 | * Somebody could have come in and twiddled with the | |
4260 | * reservation, so if we have to free more than we would have | |
4261 | * reserved from this reservation go ahead and release those | |
4262 | * bytes. | |
4263 | */ | |
4264 | to_free -= to_reserve; | |
4265 | if (to_free) | |
4266 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
0ca1f7ce | 4267 | return ret; |
9e0baf60 | 4268 | } |
25179201 | 4269 | |
0ca1f7ce YZ |
4270 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4271 | ||
0ca1f7ce YZ |
4272 | return 0; |
4273 | } | |
4274 | ||
7709cde3 JB |
4275 | /** |
4276 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4277 | * @inode: the inode to release the reservation for | |
4278 | * @num_bytes: the number of bytes we're releasing | |
4279 | * | |
4280 | * This will release the metadata reservation for an inode. This can be called | |
4281 | * once we complete IO for a given set of bytes to release their metadata | |
4282 | * reservations. | |
4283 | */ | |
0ca1f7ce YZ |
4284 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4285 | { | |
4286 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4287 | u64 to_free = 0; |
4288 | unsigned dropped; | |
0ca1f7ce YZ |
4289 | |
4290 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4291 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4292 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4293 | |
7709cde3 JB |
4294 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4295 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4296 | if (dropped > 0) |
4297 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce YZ |
4298 | |
4299 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, | |
4300 | to_free); | |
4301 | } | |
4302 | ||
7709cde3 JB |
4303 | /** |
4304 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4305 | * @inode: inode we're writing to | |
4306 | * @num_bytes: the number of bytes we want to allocate | |
4307 | * | |
4308 | * This will do the following things | |
4309 | * | |
4310 | * o reserve space in the data space info for num_bytes | |
4311 | * o reserve space in the metadata space info based on number of outstanding | |
4312 | * extents and how much csums will be needed | |
4313 | * o add to the inodes ->delalloc_bytes | |
4314 | * o add it to the fs_info's delalloc inodes list. | |
4315 | * | |
4316 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4317 | */ | |
0ca1f7ce YZ |
4318 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4319 | { | |
4320 | int ret; | |
4321 | ||
4322 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4323 | if (ret) |
0ca1f7ce YZ |
4324 | return ret; |
4325 | ||
4326 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4327 | if (ret) { | |
4328 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4329 | return ret; | |
4330 | } | |
4331 | ||
4332 | return 0; | |
4333 | } | |
4334 | ||
7709cde3 JB |
4335 | /** |
4336 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4337 | * @inode: inode we're releasing space for | |
4338 | * @num_bytes: the number of bytes we want to free up | |
4339 | * | |
4340 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4341 | * called in the case that we don't need the metadata AND data reservations | |
4342 | * anymore. So if there is an error or we insert an inline extent. | |
4343 | * | |
4344 | * This function will release the metadata space that was not used and will | |
4345 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4346 | * list if there are no delalloc bytes left. | |
4347 | */ | |
0ca1f7ce YZ |
4348 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4349 | { | |
4350 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4351 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4352 | } |
4353 | ||
9078a3e1 CM |
4354 | static int update_block_group(struct btrfs_trans_handle *trans, |
4355 | struct btrfs_root *root, | |
f0486c68 | 4356 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4357 | { |
0af3d00b | 4358 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4359 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4360 | u64 total = num_bytes; |
9078a3e1 | 4361 | u64 old_val; |
db94535d | 4362 | u64 byte_in_group; |
0af3d00b | 4363 | int factor; |
3e1ad54f | 4364 | |
5d4f98a2 YZ |
4365 | /* block accounting for super block */ |
4366 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4367 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4368 | if (alloc) |
4369 | old_val += num_bytes; | |
4370 | else | |
4371 | old_val -= num_bytes; | |
6c41761f | 4372 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4373 | spin_unlock(&info->delalloc_lock); |
4374 | ||
d397712b | 4375 | while (total) { |
db94535d | 4376 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4377 | if (!cache) |
9078a3e1 | 4378 | return -1; |
b742bb82 YZ |
4379 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4380 | BTRFS_BLOCK_GROUP_RAID1 | | |
4381 | BTRFS_BLOCK_GROUP_RAID10)) | |
4382 | factor = 2; | |
4383 | else | |
4384 | factor = 1; | |
9d66e233 JB |
4385 | /* |
4386 | * If this block group has free space cache written out, we | |
4387 | * need to make sure to load it if we are removing space. This | |
4388 | * is because we need the unpinning stage to actually add the | |
4389 | * space back to the block group, otherwise we will leak space. | |
4390 | */ | |
4391 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4392 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4393 | |
db94535d CM |
4394 | byte_in_group = bytenr - cache->key.objectid; |
4395 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4396 | |
25179201 | 4397 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4398 | spin_lock(&cache->lock); |
0af3d00b | 4399 | |
73bc1876 | 4400 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4401 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4402 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4403 | ||
0f9dd46c | 4404 | cache->dirty = 1; |
9078a3e1 | 4405 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4406 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4407 | if (alloc) { |
db94535d | 4408 | old_val += num_bytes; |
11833d66 YZ |
4409 | btrfs_set_block_group_used(&cache->item, old_val); |
4410 | cache->reserved -= num_bytes; | |
11833d66 | 4411 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4412 | cache->space_info->bytes_used += num_bytes; |
4413 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4414 | spin_unlock(&cache->lock); |
25179201 | 4415 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4416 | } else { |
db94535d | 4417 | old_val -= num_bytes; |
c286ac48 | 4418 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4419 | cache->pinned += num_bytes; |
4420 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4421 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4422 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4423 | spin_unlock(&cache->lock); |
25179201 | 4424 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4425 | |
f0486c68 YZ |
4426 | set_extent_dirty(info->pinned_extents, |
4427 | bytenr, bytenr + num_bytes - 1, | |
4428 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4429 | } |
fa9c0d79 | 4430 | btrfs_put_block_group(cache); |
db94535d CM |
4431 | total -= num_bytes; |
4432 | bytenr += num_bytes; | |
9078a3e1 CM |
4433 | } |
4434 | return 0; | |
4435 | } | |
6324fbf3 | 4436 | |
a061fc8d CM |
4437 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4438 | { | |
0f9dd46c | 4439 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4440 | u64 bytenr; |
0f9dd46c JB |
4441 | |
4442 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4443 | if (!cache) | |
a061fc8d | 4444 | return 0; |
0f9dd46c | 4445 | |
d2fb3437 | 4446 | bytenr = cache->key.objectid; |
fa9c0d79 | 4447 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4448 | |
4449 | return bytenr; | |
a061fc8d CM |
4450 | } |
4451 | ||
f0486c68 YZ |
4452 | static int pin_down_extent(struct btrfs_root *root, |
4453 | struct btrfs_block_group_cache *cache, | |
4454 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4455 | { |
11833d66 YZ |
4456 | spin_lock(&cache->space_info->lock); |
4457 | spin_lock(&cache->lock); | |
4458 | cache->pinned += num_bytes; | |
4459 | cache->space_info->bytes_pinned += num_bytes; | |
4460 | if (reserved) { | |
4461 | cache->reserved -= num_bytes; | |
4462 | cache->space_info->bytes_reserved -= num_bytes; | |
4463 | } | |
4464 | spin_unlock(&cache->lock); | |
4465 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4466 | |
f0486c68 YZ |
4467 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4468 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4469 | return 0; | |
4470 | } | |
68b38550 | 4471 | |
f0486c68 YZ |
4472 | /* |
4473 | * this function must be called within transaction | |
4474 | */ | |
4475 | int btrfs_pin_extent(struct btrfs_root *root, | |
4476 | u64 bytenr, u64 num_bytes, int reserved) | |
4477 | { | |
4478 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4479 | |
f0486c68 YZ |
4480 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4481 | BUG_ON(!cache); | |
4482 | ||
4483 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4484 | ||
4485 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4486 | return 0; |
4487 | } | |
4488 | ||
e688b725 CM |
4489 | /* |
4490 | * this function must be called within transaction | |
4491 | */ | |
4492 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4493 | struct btrfs_root *root, | |
4494 | u64 bytenr, u64 num_bytes) | |
4495 | { | |
4496 | struct btrfs_block_group_cache *cache; | |
4497 | ||
4498 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
4499 | BUG_ON(!cache); | |
4500 | ||
4501 | /* | |
4502 | * pull in the free space cache (if any) so that our pin | |
4503 | * removes the free space from the cache. We have load_only set | |
4504 | * to one because the slow code to read in the free extents does check | |
4505 | * the pinned extents. | |
4506 | */ | |
4507 | cache_block_group(cache, trans, root, 1); | |
4508 | ||
4509 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4510 | ||
4511 | /* remove us from the free space cache (if we're there at all) */ | |
4512 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4513 | btrfs_put_block_group(cache); | |
4514 | return 0; | |
4515 | } | |
4516 | ||
fb25e914 JB |
4517 | /** |
4518 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4519 | * @cache: The cache we are manipulating | |
4520 | * @num_bytes: The number of bytes in question | |
4521 | * @reserve: One of the reservation enums | |
4522 | * | |
4523 | * This is called by the allocator when it reserves space, or by somebody who is | |
4524 | * freeing space that was never actually used on disk. For example if you | |
4525 | * reserve some space for a new leaf in transaction A and before transaction A | |
4526 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4527 | * clear the reservation. | |
4528 | * | |
4529 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4530 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4531 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4532 | * allocating less disk space for the amount of data we have reserved in the | |
4533 | * case of compression. | |
4534 | * | |
4535 | * If this is a reservation and the block group has become read only we cannot | |
4536 | * make the reservation and return -EAGAIN, otherwise this function always | |
4537 | * succeeds. | |
f0486c68 | 4538 | */ |
fb25e914 JB |
4539 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4540 | u64 num_bytes, int reserve) | |
11833d66 | 4541 | { |
fb25e914 | 4542 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4543 | int ret = 0; |
fb25e914 JB |
4544 | spin_lock(&space_info->lock); |
4545 | spin_lock(&cache->lock); | |
4546 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4547 | if (cache->ro) { |
4548 | ret = -EAGAIN; | |
4549 | } else { | |
fb25e914 JB |
4550 | cache->reserved += num_bytes; |
4551 | space_info->bytes_reserved += num_bytes; | |
4552 | if (reserve == RESERVE_ALLOC) { | |
4553 | BUG_ON(space_info->bytes_may_use < num_bytes); | |
4554 | space_info->bytes_may_use -= num_bytes; | |
4555 | } | |
f0486c68 | 4556 | } |
fb25e914 JB |
4557 | } else { |
4558 | if (cache->ro) | |
4559 | space_info->bytes_readonly += num_bytes; | |
4560 | cache->reserved -= num_bytes; | |
4561 | space_info->bytes_reserved -= num_bytes; | |
4562 | space_info->reservation_progress++; | |
324ae4df | 4563 | } |
fb25e914 JB |
4564 | spin_unlock(&cache->lock); |
4565 | spin_unlock(&space_info->lock); | |
f0486c68 | 4566 | return ret; |
324ae4df | 4567 | } |
9078a3e1 | 4568 | |
11833d66 YZ |
4569 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4570 | struct btrfs_root *root) | |
e8569813 | 4571 | { |
e8569813 | 4572 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4573 | struct btrfs_caching_control *next; |
4574 | struct btrfs_caching_control *caching_ctl; | |
4575 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4576 | |
11833d66 | 4577 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4578 | |
11833d66 YZ |
4579 | list_for_each_entry_safe(caching_ctl, next, |
4580 | &fs_info->caching_block_groups, list) { | |
4581 | cache = caching_ctl->block_group; | |
4582 | if (block_group_cache_done(cache)) { | |
4583 | cache->last_byte_to_unpin = (u64)-1; | |
4584 | list_del_init(&caching_ctl->list); | |
4585 | put_caching_control(caching_ctl); | |
e8569813 | 4586 | } else { |
11833d66 | 4587 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4588 | } |
e8569813 | 4589 | } |
11833d66 YZ |
4590 | |
4591 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4592 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4593 | else | |
4594 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4595 | ||
4596 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4597 | |
4598 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4599 | return 0; |
4600 | } | |
4601 | ||
11833d66 | 4602 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4603 | { |
11833d66 YZ |
4604 | struct btrfs_fs_info *fs_info = root->fs_info; |
4605 | struct btrfs_block_group_cache *cache = NULL; | |
4606 | u64 len; | |
ccd467d6 | 4607 | |
11833d66 YZ |
4608 | while (start <= end) { |
4609 | if (!cache || | |
4610 | start >= cache->key.objectid + cache->key.offset) { | |
4611 | if (cache) | |
4612 | btrfs_put_block_group(cache); | |
4613 | cache = btrfs_lookup_block_group(fs_info, start); | |
4614 | BUG_ON(!cache); | |
4615 | } | |
4616 | ||
4617 | len = cache->key.objectid + cache->key.offset - start; | |
4618 | len = min(len, end + 1 - start); | |
4619 | ||
4620 | if (start < cache->last_byte_to_unpin) { | |
4621 | len = min(len, cache->last_byte_to_unpin - start); | |
4622 | btrfs_add_free_space(cache, start, len); | |
4623 | } | |
4624 | ||
f0486c68 YZ |
4625 | start += len; |
4626 | ||
11833d66 YZ |
4627 | spin_lock(&cache->space_info->lock); |
4628 | spin_lock(&cache->lock); | |
4629 | cache->pinned -= len; | |
4630 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4631 | if (cache->ro) |
f0486c68 | 4632 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4633 | spin_unlock(&cache->lock); |
4634 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4635 | } |
11833d66 YZ |
4636 | |
4637 | if (cache) | |
4638 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4639 | return 0; |
4640 | } | |
4641 | ||
4642 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4643 | struct btrfs_root *root) |
a28ec197 | 4644 | { |
11833d66 YZ |
4645 | struct btrfs_fs_info *fs_info = root->fs_info; |
4646 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
4647 | u64 start; |
4648 | u64 end; | |
a28ec197 | 4649 | int ret; |
a28ec197 | 4650 | |
11833d66 YZ |
4651 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4652 | unpin = &fs_info->freed_extents[1]; | |
4653 | else | |
4654 | unpin = &fs_info->freed_extents[0]; | |
4655 | ||
d397712b | 4656 | while (1) { |
1a5bc167 CM |
4657 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4658 | EXTENT_DIRTY); | |
4659 | if (ret) | |
a28ec197 | 4660 | break; |
1f3c79a2 | 4661 | |
5378e607 LD |
4662 | if (btrfs_test_opt(root, DISCARD)) |
4663 | ret = btrfs_discard_extent(root, start, | |
4664 | end + 1 - start, NULL); | |
1f3c79a2 | 4665 | |
1a5bc167 | 4666 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4667 | unpin_extent_range(root, start, end); |
b9473439 | 4668 | cond_resched(); |
a28ec197 | 4669 | } |
817d52f8 | 4670 | |
e20d96d6 CM |
4671 | return 0; |
4672 | } | |
4673 | ||
5d4f98a2 YZ |
4674 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4675 | struct btrfs_root *root, | |
4676 | u64 bytenr, u64 num_bytes, u64 parent, | |
4677 | u64 root_objectid, u64 owner_objectid, | |
4678 | u64 owner_offset, int refs_to_drop, | |
4679 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4680 | { |
e2fa7227 | 4681 | struct btrfs_key key; |
5d4f98a2 | 4682 | struct btrfs_path *path; |
1261ec42 CM |
4683 | struct btrfs_fs_info *info = root->fs_info; |
4684 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4685 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4686 | struct btrfs_extent_item *ei; |
4687 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4688 | int ret; |
5d4f98a2 | 4689 | int is_data; |
952fccac CM |
4690 | int extent_slot = 0; |
4691 | int found_extent = 0; | |
4692 | int num_to_del = 1; | |
5d4f98a2 YZ |
4693 | u32 item_size; |
4694 | u64 refs; | |
037e6390 | 4695 | |
5caf2a00 | 4696 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4697 | if (!path) |
4698 | return -ENOMEM; | |
5f26f772 | 4699 | |
3c12ac72 | 4700 | path->reada = 1; |
b9473439 | 4701 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4702 | |
4703 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4704 | BUG_ON(!is_data && refs_to_drop != 1); | |
4705 | ||
4706 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4707 | bytenr, num_bytes, parent, | |
4708 | root_objectid, owner_objectid, | |
4709 | owner_offset); | |
7bb86316 | 4710 | if (ret == 0) { |
952fccac | 4711 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4712 | while (extent_slot >= 0) { |
4713 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4714 | extent_slot); |
5d4f98a2 | 4715 | if (key.objectid != bytenr) |
952fccac | 4716 | break; |
5d4f98a2 YZ |
4717 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4718 | key.offset == num_bytes) { | |
952fccac CM |
4719 | found_extent = 1; |
4720 | break; | |
4721 | } | |
4722 | if (path->slots[0] - extent_slot > 5) | |
4723 | break; | |
5d4f98a2 | 4724 | extent_slot--; |
952fccac | 4725 | } |
5d4f98a2 YZ |
4726 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4727 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4728 | if (found_extent && item_size < sizeof(*ei)) | |
4729 | found_extent = 0; | |
4730 | #endif | |
31840ae1 | 4731 | if (!found_extent) { |
5d4f98a2 | 4732 | BUG_ON(iref); |
56bec294 | 4733 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4734 | NULL, refs_to_drop, |
4735 | is_data); | |
31840ae1 | 4736 | BUG_ON(ret); |
b3b4aa74 | 4737 | btrfs_release_path(path); |
b9473439 | 4738 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4739 | |
4740 | key.objectid = bytenr; | |
4741 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4742 | key.offset = num_bytes; | |
4743 | ||
31840ae1 ZY |
4744 | ret = btrfs_search_slot(trans, extent_root, |
4745 | &key, path, -1, 1); | |
f3465ca4 JB |
4746 | if (ret) { |
4747 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4748 | ", was looking for %llu\n", ret, |
4749 | (unsigned long long)bytenr); | |
b783e62d JB |
4750 | if (ret > 0) |
4751 | btrfs_print_leaf(extent_root, | |
4752 | path->nodes[0]); | |
f3465ca4 | 4753 | } |
31840ae1 ZY |
4754 | BUG_ON(ret); |
4755 | extent_slot = path->slots[0]; | |
4756 | } | |
7bb86316 CM |
4757 | } else { |
4758 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4759 | WARN_ON(1); | |
d397712b | 4760 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4761 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4762 | (unsigned long long)bytenr, |
56bec294 | 4763 | (unsigned long long)parent, |
d397712b | 4764 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4765 | (unsigned long long)owner_objectid, |
4766 | (unsigned long long)owner_offset); | |
7bb86316 | 4767 | } |
5f39d397 CM |
4768 | |
4769 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4770 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4771 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4772 | if (item_size < sizeof(*ei)) { | |
4773 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4774 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4775 | owner_objectid, 0); | |
4776 | BUG_ON(ret < 0); | |
4777 | ||
b3b4aa74 | 4778 | btrfs_release_path(path); |
5d4f98a2 YZ |
4779 | path->leave_spinning = 1; |
4780 | ||
4781 | key.objectid = bytenr; | |
4782 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4783 | key.offset = num_bytes; | |
4784 | ||
4785 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4786 | -1, 1); | |
4787 | if (ret) { | |
4788 | printk(KERN_ERR "umm, got %d back from search" | |
4789 | ", was looking for %llu\n", ret, | |
4790 | (unsigned long long)bytenr); | |
4791 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4792 | } | |
4793 | BUG_ON(ret); | |
4794 | extent_slot = path->slots[0]; | |
4795 | leaf = path->nodes[0]; | |
4796 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4797 | } | |
4798 | #endif | |
4799 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4800 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4801 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4802 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4803 | struct btrfs_tree_block_info *bi; | |
4804 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4805 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4806 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4807 | } | |
56bec294 | 4808 | |
5d4f98a2 | 4809 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4810 | BUG_ON(refs < refs_to_drop); |
4811 | refs -= refs_to_drop; | |
5f39d397 | 4812 | |
5d4f98a2 YZ |
4813 | if (refs > 0) { |
4814 | if (extent_op) | |
4815 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4816 | /* | |
4817 | * In the case of inline back ref, reference count will | |
4818 | * be updated by remove_extent_backref | |
952fccac | 4819 | */ |
5d4f98a2 YZ |
4820 | if (iref) { |
4821 | BUG_ON(!found_extent); | |
4822 | } else { | |
4823 | btrfs_set_extent_refs(leaf, ei, refs); | |
4824 | btrfs_mark_buffer_dirty(leaf); | |
4825 | } | |
4826 | if (found_extent) { | |
4827 | ret = remove_extent_backref(trans, extent_root, path, | |
4828 | iref, refs_to_drop, | |
4829 | is_data); | |
952fccac CM |
4830 | BUG_ON(ret); |
4831 | } | |
5d4f98a2 | 4832 | } else { |
5d4f98a2 YZ |
4833 | if (found_extent) { |
4834 | BUG_ON(is_data && refs_to_drop != | |
4835 | extent_data_ref_count(root, path, iref)); | |
4836 | if (iref) { | |
4837 | BUG_ON(path->slots[0] != extent_slot); | |
4838 | } else { | |
4839 | BUG_ON(path->slots[0] != extent_slot + 1); | |
4840 | path->slots[0] = extent_slot; | |
4841 | num_to_del = 2; | |
4842 | } | |
78fae27e | 4843 | } |
b9473439 | 4844 | |
952fccac CM |
4845 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
4846 | num_to_del); | |
31840ae1 | 4847 | BUG_ON(ret); |
b3b4aa74 | 4848 | btrfs_release_path(path); |
21af804c | 4849 | |
5d4f98a2 | 4850 | if (is_data) { |
459931ec CM |
4851 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
4852 | BUG_ON(ret); | |
d57e62b8 CM |
4853 | } else { |
4854 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
4855 | bytenr >> PAGE_CACHE_SHIFT, | |
4856 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
4857 | } |
4858 | ||
f0486c68 | 4859 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 4860 | BUG_ON(ret); |
a28ec197 | 4861 | } |
5caf2a00 | 4862 | btrfs_free_path(path); |
a28ec197 CM |
4863 | return ret; |
4864 | } | |
4865 | ||
1887be66 | 4866 | /* |
f0486c68 | 4867 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
4868 | * delayed ref for that extent as well. This searches the delayed ref tree for |
4869 | * a given extent, and if there are no other delayed refs to be processed, it | |
4870 | * removes it from the tree. | |
4871 | */ | |
4872 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
4873 | struct btrfs_root *root, u64 bytenr) | |
4874 | { | |
4875 | struct btrfs_delayed_ref_head *head; | |
4876 | struct btrfs_delayed_ref_root *delayed_refs; | |
4877 | struct btrfs_delayed_ref_node *ref; | |
4878 | struct rb_node *node; | |
f0486c68 | 4879 | int ret = 0; |
1887be66 CM |
4880 | |
4881 | delayed_refs = &trans->transaction->delayed_refs; | |
4882 | spin_lock(&delayed_refs->lock); | |
4883 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
4884 | if (!head) | |
4885 | goto out; | |
4886 | ||
4887 | node = rb_prev(&head->node.rb_node); | |
4888 | if (!node) | |
4889 | goto out; | |
4890 | ||
4891 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
4892 | ||
4893 | /* there are still entries for this ref, we can't drop it */ | |
4894 | if (ref->bytenr == bytenr) | |
4895 | goto out; | |
4896 | ||
5d4f98a2 YZ |
4897 | if (head->extent_op) { |
4898 | if (!head->must_insert_reserved) | |
4899 | goto out; | |
4900 | kfree(head->extent_op); | |
4901 | head->extent_op = NULL; | |
4902 | } | |
4903 | ||
1887be66 CM |
4904 | /* |
4905 | * waiting for the lock here would deadlock. If someone else has it | |
4906 | * locked they are already in the process of dropping it anyway | |
4907 | */ | |
4908 | if (!mutex_trylock(&head->mutex)) | |
4909 | goto out; | |
4910 | ||
4911 | /* | |
4912 | * at this point we have a head with no other entries. Go | |
4913 | * ahead and process it. | |
4914 | */ | |
4915 | head->node.in_tree = 0; | |
4916 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 4917 | |
1887be66 CM |
4918 | delayed_refs->num_entries--; |
4919 | ||
4920 | /* | |
4921 | * we don't take a ref on the node because we're removing it from the | |
4922 | * tree, so we just steal the ref the tree was holding. | |
4923 | */ | |
c3e69d58 CM |
4924 | delayed_refs->num_heads--; |
4925 | if (list_empty(&head->cluster)) | |
4926 | delayed_refs->num_heads_ready--; | |
4927 | ||
4928 | list_del_init(&head->cluster); | |
1887be66 CM |
4929 | spin_unlock(&delayed_refs->lock); |
4930 | ||
f0486c68 YZ |
4931 | BUG_ON(head->extent_op); |
4932 | if (head->must_insert_reserved) | |
4933 | ret = 1; | |
4934 | ||
4935 | mutex_unlock(&head->mutex); | |
1887be66 | 4936 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 4937 | return ret; |
1887be66 CM |
4938 | out: |
4939 | spin_unlock(&delayed_refs->lock); | |
4940 | return 0; | |
4941 | } | |
4942 | ||
f0486c68 YZ |
4943 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4944 | struct btrfs_root *root, | |
4945 | struct extent_buffer *buf, | |
66d7e7f0 | 4946 | u64 parent, int last_ref, int for_cow) |
f0486c68 | 4947 | { |
f0486c68 YZ |
4948 | struct btrfs_block_group_cache *cache = NULL; |
4949 | int ret; | |
4950 | ||
4951 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
4952 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
4953 | buf->start, buf->len, | |
4954 | parent, root->root_key.objectid, | |
4955 | btrfs_header_level(buf), | |
4956 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); | |
f0486c68 YZ |
4957 | BUG_ON(ret); |
4958 | } | |
4959 | ||
4960 | if (!last_ref) | |
4961 | return; | |
4962 | ||
f0486c68 | 4963 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
4964 | |
4965 | if (btrfs_header_generation(buf) == trans->transid) { | |
4966 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4967 | ret = check_ref_cleanup(trans, root, buf->start); | |
4968 | if (!ret) | |
37be25bc | 4969 | goto out; |
f0486c68 YZ |
4970 | } |
4971 | ||
4972 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
4973 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 4974 | goto out; |
f0486c68 YZ |
4975 | } |
4976 | ||
4977 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
4978 | ||
4979 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 4980 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
4981 | } |
4982 | out: | |
a826d6dc JB |
4983 | /* |
4984 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
4985 | * anymore. | |
4986 | */ | |
4987 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
4988 | btrfs_put_block_group(cache); |
4989 | } | |
4990 | ||
66d7e7f0 AJ |
4991 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4992 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
4993 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
4994 | { |
4995 | int ret; | |
66d7e7f0 | 4996 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 4997 | |
56bec294 CM |
4998 | /* |
4999 | * tree log blocks never actually go into the extent allocation | |
5000 | * tree, just update pinning info and exit early. | |
56bec294 | 5001 | */ |
5d4f98a2 YZ |
5002 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5003 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5004 | /* unlocks the pinned mutex */ |
11833d66 | 5005 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5006 | ret = 0; |
5d4f98a2 | 5007 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5008 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5009 | num_bytes, | |
5d4f98a2 | 5010 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5011 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
1887be66 | 5012 | BUG_ON(ret); |
5d4f98a2 | 5013 | } else { |
66d7e7f0 AJ |
5014 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5015 | num_bytes, | |
5016 | parent, root_objectid, owner, | |
5017 | offset, BTRFS_DROP_DELAYED_REF, | |
5018 | NULL, for_cow); | |
5d4f98a2 | 5019 | BUG_ON(ret); |
56bec294 | 5020 | } |
925baedd CM |
5021 | return ret; |
5022 | } | |
5023 | ||
87ee04eb CM |
5024 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5025 | { | |
5026 | u64 mask = ((u64)root->stripesize - 1); | |
5027 | u64 ret = (val + mask) & ~mask; | |
5028 | return ret; | |
5029 | } | |
5030 | ||
817d52f8 JB |
5031 | /* |
5032 | * when we wait for progress in the block group caching, its because | |
5033 | * our allocation attempt failed at least once. So, we must sleep | |
5034 | * and let some progress happen before we try again. | |
5035 | * | |
5036 | * This function will sleep at least once waiting for new free space to | |
5037 | * show up, and then it will check the block group free space numbers | |
5038 | * for our min num_bytes. Another option is to have it go ahead | |
5039 | * and look in the rbtree for a free extent of a given size, but this | |
5040 | * is a good start. | |
5041 | */ | |
5042 | static noinline int | |
5043 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5044 | u64 num_bytes) | |
5045 | { | |
11833d66 | 5046 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5047 | DEFINE_WAIT(wait); |
5048 | ||
11833d66 YZ |
5049 | caching_ctl = get_caching_control(cache); |
5050 | if (!caching_ctl) | |
817d52f8 | 5051 | return 0; |
817d52f8 | 5052 | |
11833d66 | 5053 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5054 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5055 | |
5056 | put_caching_control(caching_ctl); | |
5057 | return 0; | |
5058 | } | |
5059 | ||
5060 | static noinline int | |
5061 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5062 | { | |
5063 | struct btrfs_caching_control *caching_ctl; | |
5064 | DEFINE_WAIT(wait); | |
5065 | ||
5066 | caching_ctl = get_caching_control(cache); | |
5067 | if (!caching_ctl) | |
5068 | return 0; | |
5069 | ||
5070 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5071 | ||
5072 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5073 | return 0; |
5074 | } | |
5075 | ||
b742bb82 YZ |
5076 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5077 | { | |
5078 | int index; | |
5079 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
5080 | index = 0; | |
5081 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
5082 | index = 1; | |
5083 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
5084 | index = 2; | |
5085 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
5086 | index = 3; | |
5087 | else | |
5088 | index = 4; | |
5089 | return index; | |
5090 | } | |
5091 | ||
817d52f8 | 5092 | enum btrfs_loop_type { |
ccf0e725 | 5093 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
5094 | LOOP_CACHING_NOWAIT = 1, |
5095 | LOOP_CACHING_WAIT = 2, | |
5096 | LOOP_ALLOC_CHUNK = 3, | |
5097 | LOOP_NO_EMPTY_SIZE = 4, | |
5098 | }; | |
5099 | ||
fec577fb CM |
5100 | /* |
5101 | * walks the btree of allocated extents and find a hole of a given size. | |
5102 | * The key ins is changed to record the hole: | |
5103 | * ins->objectid == block start | |
62e2749e | 5104 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5105 | * ins->offset == number of blocks |
5106 | * Any available blocks before search_start are skipped. | |
5107 | */ | |
d397712b | 5108 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5109 | struct btrfs_root *orig_root, |
5110 | u64 num_bytes, u64 empty_size, | |
5111 | u64 search_start, u64 search_end, | |
5112 | u64 hint_byte, struct btrfs_key *ins, | |
e0f54067 | 5113 | u64 data) |
fec577fb | 5114 | { |
80eb234a | 5115 | int ret = 0; |
d397712b | 5116 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5117 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5118 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 5119 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5120 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5121 | int done_chunk_alloc = 0; |
80eb234a | 5122 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
5123 | int last_ptr_loop = 0; |
5124 | int loop = 0; | |
f0486c68 | 5125 | int index = 0; |
fb25e914 JB |
5126 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5127 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5128 | bool found_uncached_bg = false; |
0a24325e | 5129 | bool failed_cluster_refill = false; |
1cdda9b8 | 5130 | bool failed_alloc = false; |
67377734 | 5131 | bool use_cluster = true; |
60d2adbb | 5132 | bool have_caching_bg = false; |
ccf0e725 JB |
5133 | u64 ideal_cache_percent = 0; |
5134 | u64 ideal_cache_offset = 0; | |
fec577fb | 5135 | |
db94535d | 5136 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5137 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5138 | ins->objectid = 0; |
5139 | ins->offset = 0; | |
b1a4d965 | 5140 | |
2552d17e | 5141 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5142 | if (!space_info) { |
e0f54067 | 5143 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5144 | return -ENOSPC; |
5145 | } | |
2552d17e | 5146 | |
67377734 JB |
5147 | /* |
5148 | * If the space info is for both data and metadata it means we have a | |
5149 | * small filesystem and we can't use the clustering stuff. | |
5150 | */ | |
5151 | if (btrfs_mixed_space_info(space_info)) | |
5152 | use_cluster = false; | |
5153 | ||
0ef3e66b CM |
5154 | if (orig_root->ref_cows || empty_size) |
5155 | allowed_chunk_alloc = 1; | |
5156 | ||
67377734 | 5157 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5158 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5159 | if (!btrfs_test_opt(root, SSD)) |
5160 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5161 | } |
5162 | ||
67377734 JB |
5163 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5164 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5165 | last_ptr = &root->fs_info->data_alloc_cluster; |
5166 | } | |
0f9dd46c | 5167 | |
239b14b3 | 5168 | if (last_ptr) { |
fa9c0d79 CM |
5169 | spin_lock(&last_ptr->lock); |
5170 | if (last_ptr->block_group) | |
5171 | hint_byte = last_ptr->window_start; | |
5172 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5173 | } |
fa9c0d79 | 5174 | |
a061fc8d | 5175 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5176 | search_start = max(search_start, hint_byte); |
0b86a832 | 5177 | |
817d52f8 | 5178 | if (!last_ptr) |
fa9c0d79 | 5179 | empty_cluster = 0; |
fa9c0d79 | 5180 | |
2552d17e | 5181 | if (search_start == hint_byte) { |
ccf0e725 | 5182 | ideal_cache: |
2552d17e JB |
5183 | block_group = btrfs_lookup_block_group(root->fs_info, |
5184 | search_start); | |
817d52f8 JB |
5185 | /* |
5186 | * we don't want to use the block group if it doesn't match our | |
5187 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5188 | * |
5189 | * However if we are re-searching with an ideal block group | |
5190 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5191 | */ |
5192 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
5193 | (block_group->cached != BTRFS_CACHE_NO || |
5194 | search_start == ideal_cache_offset)) { | |
2552d17e | 5195 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5196 | if (list_empty(&block_group->list) || |
5197 | block_group->ro) { | |
5198 | /* | |
5199 | * someone is removing this block group, | |
5200 | * we can't jump into the have_block_group | |
5201 | * target because our list pointers are not | |
5202 | * valid | |
5203 | */ | |
5204 | btrfs_put_block_group(block_group); | |
5205 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5206 | } else { |
b742bb82 | 5207 | index = get_block_group_index(block_group); |
44fb5511 | 5208 | goto have_block_group; |
ccf0e725 | 5209 | } |
2552d17e | 5210 | } else if (block_group) { |
fa9c0d79 | 5211 | btrfs_put_block_group(block_group); |
2552d17e | 5212 | } |
42e70e7a | 5213 | } |
2552d17e | 5214 | search: |
60d2adbb | 5215 | have_caching_bg = false; |
80eb234a | 5216 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5217 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5218 | list) { | |
6226cb0a | 5219 | u64 offset; |
817d52f8 | 5220 | int cached; |
8a1413a2 | 5221 | |
11dfe35a | 5222 | btrfs_get_block_group(block_group); |
2552d17e | 5223 | search_start = block_group->key.objectid; |
42e70e7a | 5224 | |
83a50de9 CM |
5225 | /* |
5226 | * this can happen if we end up cycling through all the | |
5227 | * raid types, but we want to make sure we only allocate | |
5228 | * for the proper type. | |
5229 | */ | |
5230 | if (!block_group_bits(block_group, data)) { | |
5231 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5232 | BTRFS_BLOCK_GROUP_RAID1 | | |
5233 | BTRFS_BLOCK_GROUP_RAID10; | |
5234 | ||
5235 | /* | |
5236 | * if they asked for extra copies and this block group | |
5237 | * doesn't provide them, bail. This does allow us to | |
5238 | * fill raid0 from raid1. | |
5239 | */ | |
5240 | if ((data & extra) && !(block_group->flags & extra)) | |
5241 | goto loop; | |
5242 | } | |
5243 | ||
2552d17e | 5244 | have_block_group: |
291c7d2f JB |
5245 | cached = block_group_cache_done(block_group); |
5246 | if (unlikely(!cached)) { | |
ccf0e725 JB |
5247 | u64 free_percent; |
5248 | ||
291c7d2f | 5249 | found_uncached_bg = true; |
b8399dee JB |
5250 | ret = cache_block_group(block_group, trans, |
5251 | orig_root, 1); | |
9d66e233 | 5252 | if (block_group->cached == BTRFS_CACHE_FINISHED) |
291c7d2f | 5253 | goto alloc; |
9d66e233 | 5254 | |
ccf0e725 JB |
5255 | free_percent = btrfs_block_group_used(&block_group->item); |
5256 | free_percent *= 100; | |
5257 | free_percent = div64_u64(free_percent, | |
5258 | block_group->key.offset); | |
5259 | free_percent = 100 - free_percent; | |
5260 | if (free_percent > ideal_cache_percent && | |
5261 | likely(!block_group->ro)) { | |
5262 | ideal_cache_offset = block_group->key.objectid; | |
5263 | ideal_cache_percent = free_percent; | |
5264 | } | |
5265 | ||
817d52f8 | 5266 | /* |
bab39bf9 JB |
5267 | * The caching workers are limited to 2 threads, so we |
5268 | * can queue as much work as we care to. | |
817d52f8 | 5269 | */ |
bab39bf9 | 5270 | if (loop > LOOP_FIND_IDEAL) { |
b8399dee JB |
5271 | ret = cache_block_group(block_group, trans, |
5272 | orig_root, 0); | |
817d52f8 | 5273 | BUG_ON(ret); |
2552d17e | 5274 | } |
817d52f8 | 5275 | |
ccf0e725 JB |
5276 | /* |
5277 | * If loop is set for cached only, try the next block | |
5278 | * group. | |
5279 | */ | |
5280 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
5281 | goto loop; |
5282 | } | |
5283 | ||
291c7d2f | 5284 | alloc: |
ea6a478e | 5285 | if (unlikely(block_group->ro)) |
2552d17e | 5286 | goto loop; |
0f9dd46c | 5287 | |
ff5714cc | 5288 | spin_lock(&block_group->free_space_ctl->tree_lock); |
cca1c81f | 5289 | if (cached && |
ff5714cc | 5290 | block_group->free_space_ctl->free_space < |
425d8315 | 5291 | num_bytes + empty_cluster + empty_size) { |
ff5714cc | 5292 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
cca1c81f JB |
5293 | goto loop; |
5294 | } | |
ff5714cc | 5295 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
cca1c81f | 5296 | |
0a24325e JB |
5297 | /* |
5298 | * Ok we want to try and use the cluster allocator, so lets look | |
5299 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
5300 | * have tried the cluster allocator plenty of times at this | |
5301 | * point and not have found anything, so we are likely way too | |
5302 | * fragmented for the clustering stuff to find anything, so lets | |
5303 | * just skip it and let the allocator find whatever block it can | |
5304 | * find | |
5305 | */ | |
5306 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
5307 | /* |
5308 | * the refill lock keeps out other | |
5309 | * people trying to start a new cluster | |
5310 | */ | |
5311 | spin_lock(&last_ptr->refill_lock); | |
be064d11 AO |
5312 | if (!last_ptr->block_group || |
5313 | last_ptr->block_group->ro || | |
5314 | !block_group_bits(last_ptr->block_group, data)) | |
44fb5511 | 5315 | goto refill_cluster; |
44fb5511 | 5316 | |
fa9c0d79 CM |
5317 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
5318 | num_bytes, search_start); | |
5319 | if (offset) { | |
5320 | /* we have a block, we're done */ | |
5321 | spin_unlock(&last_ptr->refill_lock); | |
5322 | goto checks; | |
5323 | } | |
5324 | ||
5325 | spin_lock(&last_ptr->lock); | |
5326 | /* | |
5327 | * whoops, this cluster doesn't actually point to | |
5328 | * this block group. Get a ref on the block | |
5329 | * group is does point to and try again | |
5330 | */ | |
5331 | if (!last_ptr_loop && last_ptr->block_group && | |
ff1f2b44 | 5332 | last_ptr->block_group != block_group && |
5333 | index <= | |
5334 | get_block_group_index(last_ptr->block_group)) { | |
fa9c0d79 CM |
5335 | |
5336 | btrfs_put_block_group(block_group); | |
5337 | block_group = last_ptr->block_group; | |
11dfe35a | 5338 | btrfs_get_block_group(block_group); |
fa9c0d79 CM |
5339 | spin_unlock(&last_ptr->lock); |
5340 | spin_unlock(&last_ptr->refill_lock); | |
5341 | ||
5342 | last_ptr_loop = 1; | |
5343 | search_start = block_group->key.objectid; | |
44fb5511 CM |
5344 | /* |
5345 | * we know this block group is properly | |
5346 | * in the list because | |
5347 | * btrfs_remove_block_group, drops the | |
5348 | * cluster before it removes the block | |
5349 | * group from the list | |
5350 | */ | |
fa9c0d79 CM |
5351 | goto have_block_group; |
5352 | } | |
5353 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 5354 | refill_cluster: |
fa9c0d79 CM |
5355 | /* |
5356 | * this cluster didn't work out, free it and | |
5357 | * start over | |
5358 | */ | |
5359 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5360 | ||
5361 | last_ptr_loop = 0; | |
5362 | ||
5363 | /* allocate a cluster in this block group */ | |
451d7585 | 5364 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5365 | block_group, last_ptr, |
1b22bad7 | 5366 | search_start, num_bytes, |
fa9c0d79 CM |
5367 | empty_cluster + empty_size); |
5368 | if (ret == 0) { | |
5369 | /* | |
5370 | * now pull our allocation out of this | |
5371 | * cluster | |
5372 | */ | |
5373 | offset = btrfs_alloc_from_cluster(block_group, | |
5374 | last_ptr, num_bytes, | |
5375 | search_start); | |
5376 | if (offset) { | |
5377 | /* we found one, proceed */ | |
5378 | spin_unlock(&last_ptr->refill_lock); | |
5379 | goto checks; | |
5380 | } | |
0a24325e JB |
5381 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5382 | && !failed_cluster_refill) { | |
817d52f8 JB |
5383 | spin_unlock(&last_ptr->refill_lock); |
5384 | ||
0a24325e | 5385 | failed_cluster_refill = true; |
817d52f8 JB |
5386 | wait_block_group_cache_progress(block_group, |
5387 | num_bytes + empty_cluster + empty_size); | |
5388 | goto have_block_group; | |
fa9c0d79 | 5389 | } |
817d52f8 | 5390 | |
fa9c0d79 CM |
5391 | /* |
5392 | * at this point we either didn't find a cluster | |
5393 | * or we weren't able to allocate a block from our | |
5394 | * cluster. Free the cluster we've been trying | |
5395 | * to use, and go to the next block group | |
5396 | */ | |
0a24325e | 5397 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5398 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5399 | goto loop; |
fa9c0d79 CM |
5400 | } |
5401 | ||
6226cb0a JB |
5402 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5403 | num_bytes, empty_size); | |
1cdda9b8 JB |
5404 | /* |
5405 | * If we didn't find a chunk, and we haven't failed on this | |
5406 | * block group before, and this block group is in the middle of | |
5407 | * caching and we are ok with waiting, then go ahead and wait | |
5408 | * for progress to be made, and set failed_alloc to true. | |
5409 | * | |
5410 | * If failed_alloc is true then we've already waited on this | |
5411 | * block group once and should move on to the next block group. | |
5412 | */ | |
5413 | if (!offset && !failed_alloc && !cached && | |
5414 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5415 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5416 | num_bytes + empty_size); |
5417 | failed_alloc = true; | |
817d52f8 | 5418 | goto have_block_group; |
1cdda9b8 | 5419 | } else if (!offset) { |
60d2adbb MX |
5420 | if (!cached) |
5421 | have_caching_bg = true; | |
1cdda9b8 | 5422 | goto loop; |
817d52f8 | 5423 | } |
fa9c0d79 | 5424 | checks: |
6226cb0a | 5425 | search_start = stripe_align(root, offset); |
2552d17e | 5426 | /* move on to the next group */ |
6226cb0a JB |
5427 | if (search_start + num_bytes >= search_end) { |
5428 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5429 | goto loop; |
6226cb0a | 5430 | } |
25179201 | 5431 | |
2552d17e JB |
5432 | /* move on to the next group */ |
5433 | if (search_start + num_bytes > | |
6226cb0a JB |
5434 | block_group->key.objectid + block_group->key.offset) { |
5435 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5436 | goto loop; |
6226cb0a | 5437 | } |
f5a31e16 | 5438 | |
f0486c68 YZ |
5439 | ins->objectid = search_start; |
5440 | ins->offset = num_bytes; | |
2552d17e | 5441 | |
f0486c68 YZ |
5442 | if (offset < search_start) |
5443 | btrfs_add_free_space(block_group, offset, | |
5444 | search_start - offset); | |
5445 | BUG_ON(offset > search_start); | |
2552d17e | 5446 | |
fb25e914 JB |
5447 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
5448 | alloc_type); | |
f0486c68 | 5449 | if (ret == -EAGAIN) { |
6226cb0a | 5450 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 5451 | goto loop; |
0f9dd46c | 5452 | } |
0b86a832 | 5453 | |
f0486c68 | 5454 | /* we are all good, lets return */ |
2552d17e JB |
5455 | ins->objectid = search_start; |
5456 | ins->offset = num_bytes; | |
d2fb3437 | 5457 | |
6226cb0a JB |
5458 | if (offset < search_start) |
5459 | btrfs_add_free_space(block_group, offset, | |
5460 | search_start - offset); | |
5461 | BUG_ON(offset > search_start); | |
d82a6f1d | 5462 | btrfs_put_block_group(block_group); |
2552d17e JB |
5463 | break; |
5464 | loop: | |
0a24325e | 5465 | failed_cluster_refill = false; |
1cdda9b8 | 5466 | failed_alloc = false; |
b742bb82 | 5467 | BUG_ON(index != get_block_group_index(block_group)); |
fa9c0d79 | 5468 | btrfs_put_block_group(block_group); |
2552d17e JB |
5469 | } |
5470 | up_read(&space_info->groups_sem); | |
5471 | ||
60d2adbb MX |
5472 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5473 | goto search; | |
5474 | ||
b742bb82 YZ |
5475 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5476 | goto search; | |
5477 | ||
ccf0e725 JB |
5478 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
5479 | * for them to make caching progress. Also | |
5480 | * determine the best possible bg to cache | |
5481 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
5482 | * caching kthreads as we move along | |
817d52f8 JB |
5483 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5484 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5485 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5486 | * again | |
fa9c0d79 | 5487 | */ |
723bda20 | 5488 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5489 | index = 0; |
ccf0e725 | 5490 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 5491 | found_uncached_bg = false; |
ccf0e725 | 5492 | loop++; |
bab39bf9 | 5493 | if (!ideal_cache_percent) |
817d52f8 | 5494 | goto search; |
ccf0e725 JB |
5495 | |
5496 | /* | |
5497 | * 1 of the following 2 things have happened so far | |
5498 | * | |
5499 | * 1) We found an ideal block group for caching that | |
5500 | * is mostly full and will cache quickly, so we might | |
5501 | * as well wait for it. | |
5502 | * | |
5503 | * 2) We searched for cached only and we didn't find | |
5504 | * anything, and we didn't start any caching kthreads | |
5505 | * either, so chances are we will loop through and | |
5506 | * start a couple caching kthreads, and then come back | |
5507 | * around and just wait for them. This will be slower | |
5508 | * because we will have 2 caching kthreads reading at | |
5509 | * the same time when we could have just started one | |
5510 | * and waited for it to get far enough to give us an | |
5511 | * allocation, so go ahead and go to the wait caching | |
5512 | * loop. | |
5513 | */ | |
5514 | loop = LOOP_CACHING_WAIT; | |
5515 | search_start = ideal_cache_offset; | |
5516 | ideal_cache_percent = 0; | |
5517 | goto ideal_cache; | |
5518 | } else if (loop == LOOP_FIND_IDEAL) { | |
5519 | /* | |
5520 | * Didn't find a uncached bg, wait on anything we find | |
5521 | * next. | |
5522 | */ | |
5523 | loop = LOOP_CACHING_WAIT; | |
5524 | goto search; | |
5525 | } | |
5526 | ||
723bda20 | 5527 | loop++; |
817d52f8 JB |
5528 | |
5529 | if (loop == LOOP_ALLOC_CHUNK) { | |
723bda20 JB |
5530 | if (allowed_chunk_alloc) { |
5531 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5532 | 2 * 1024 * 1024, data, | |
5533 | CHUNK_ALLOC_LIMITED); | |
5534 | allowed_chunk_alloc = 0; | |
5535 | if (ret == 1) | |
5536 | done_chunk_alloc = 1; | |
5537 | } else if (!done_chunk_alloc && | |
5538 | space_info->force_alloc == | |
5539 | CHUNK_ALLOC_NO_FORCE) { | |
5540 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5541 | } | |
2552d17e | 5542 | |
723bda20 JB |
5543 | /* |
5544 | * We didn't allocate a chunk, go ahead and drop the | |
5545 | * empty size and loop again. | |
5546 | */ | |
5547 | if (!done_chunk_alloc) | |
5548 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5549 | } |
5550 | ||
723bda20 JB |
5551 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5552 | empty_size = 0; | |
5553 | empty_cluster = 0; | |
fa9c0d79 | 5554 | } |
723bda20 JB |
5555 | |
5556 | goto search; | |
2552d17e JB |
5557 | } else if (!ins->objectid) { |
5558 | ret = -ENOSPC; | |
d82a6f1d | 5559 | } else if (ins->objectid) { |
80eb234a | 5560 | ret = 0; |
be744175 | 5561 | } |
be744175 | 5562 | |
0f70abe2 | 5563 | return ret; |
fec577fb | 5564 | } |
ec44a35c | 5565 | |
9ed74f2d JB |
5566 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5567 | int dump_block_groups) | |
0f9dd46c JB |
5568 | { |
5569 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5570 | int index = 0; |
0f9dd46c | 5571 | |
9ed74f2d | 5572 | spin_lock(&info->lock); |
fb25e914 JB |
5573 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5574 | (unsigned long long)info->flags, | |
d397712b | 5575 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5576 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5577 | info->bytes_readonly), |
d397712b | 5578 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5579 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5580 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5581 | (unsigned long long)info->total_bytes, |
8929ecfa | 5582 | (unsigned long long)info->bytes_used, |
21380931 | 5583 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5584 | (unsigned long long)info->bytes_reserved, |
21380931 | 5585 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5586 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5587 | spin_unlock(&info->lock); |
5588 | ||
5589 | if (!dump_block_groups) | |
5590 | return; | |
0f9dd46c | 5591 | |
80eb234a | 5592 | down_read(&info->groups_sem); |
b742bb82 YZ |
5593 | again: |
5594 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5595 | spin_lock(&cache->lock); |
d397712b CM |
5596 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5597 | "%llu pinned %llu reserved\n", | |
5598 | (unsigned long long)cache->key.objectid, | |
5599 | (unsigned long long)cache->key.offset, | |
5600 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5601 | (unsigned long long)cache->pinned, | |
5602 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5603 | btrfs_dump_free_space(cache, bytes); |
5604 | spin_unlock(&cache->lock); | |
5605 | } | |
b742bb82 YZ |
5606 | if (++index < BTRFS_NR_RAID_TYPES) |
5607 | goto again; | |
80eb234a | 5608 | up_read(&info->groups_sem); |
0f9dd46c | 5609 | } |
e8569813 | 5610 | |
11833d66 YZ |
5611 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5612 | struct btrfs_root *root, | |
5613 | u64 num_bytes, u64 min_alloc_size, | |
5614 | u64 empty_size, u64 hint_byte, | |
5615 | u64 search_end, struct btrfs_key *ins, | |
5616 | u64 data) | |
fec577fb CM |
5617 | { |
5618 | int ret; | |
fbdc762b | 5619 | u64 search_start = 0; |
925baedd | 5620 | |
6a63209f | 5621 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5622 | again: |
0ef3e66b CM |
5623 | /* |
5624 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5625 | * is not called recursively on allocations | |
5626 | */ | |
83d3c969 | 5627 | if (empty_size || root->ref_cows) |
6324fbf3 | 5628 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5629 | num_bytes + 2 * 1024 * 1024, data, |
5630 | CHUNK_ALLOC_NO_FORCE); | |
0b86a832 | 5631 | |
db94535d CM |
5632 | WARN_ON(num_bytes < root->sectorsize); |
5633 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
f0486c68 YZ |
5634 | search_start, search_end, hint_byte, |
5635 | ins, data); | |
3b951516 | 5636 | |
98d20f67 CM |
5637 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
5638 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 5639 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 5640 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b | 5641 | do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 | 5642 | num_bytes, data, CHUNK_ALLOC_FORCE); |
98d20f67 CM |
5643 | goto again; |
5644 | } | |
91435650 | 5645 | if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
0f9dd46c JB |
5646 | struct btrfs_space_info *sinfo; |
5647 | ||
5648 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
5649 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
5650 | "wanted %llu\n", (unsigned long long)data, | |
5651 | (unsigned long long)num_bytes); | |
9ed74f2d | 5652 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 5653 | } |
0f9dd46c | 5654 | |
1abe9b8a | 5655 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5656 | ||
0f9dd46c | 5657 | return ret; |
e6dcd2dc CM |
5658 | } |
5659 | ||
e688b725 CM |
5660 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
5661 | u64 start, u64 len, int pin) | |
65b51a00 | 5662 | { |
0f9dd46c | 5663 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5664 | int ret = 0; |
0f9dd46c | 5665 | |
0f9dd46c JB |
5666 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5667 | if (!cache) { | |
d397712b CM |
5668 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5669 | (unsigned long long)start); | |
0f9dd46c JB |
5670 | return -ENOSPC; |
5671 | } | |
1f3c79a2 | 5672 | |
5378e607 LD |
5673 | if (btrfs_test_opt(root, DISCARD)) |
5674 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5675 | |
e688b725 CM |
5676 | if (pin) |
5677 | pin_down_extent(root, cache, start, len, 1); | |
5678 | else { | |
5679 | btrfs_add_free_space(cache, start, len); | |
5680 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
5681 | } | |
fa9c0d79 | 5682 | btrfs_put_block_group(cache); |
817d52f8 | 5683 | |
1abe9b8a | 5684 | trace_btrfs_reserved_extent_free(root, start, len); |
5685 | ||
e6dcd2dc CM |
5686 | return ret; |
5687 | } | |
5688 | ||
e688b725 CM |
5689 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
5690 | u64 start, u64 len) | |
5691 | { | |
5692 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
5693 | } | |
5694 | ||
5695 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
5696 | u64 start, u64 len) | |
5697 | { | |
5698 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
5699 | } | |
5700 | ||
5d4f98a2 YZ |
5701 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5702 | struct btrfs_root *root, | |
5703 | u64 parent, u64 root_objectid, | |
5704 | u64 flags, u64 owner, u64 offset, | |
5705 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5706 | { |
5707 | int ret; | |
5d4f98a2 | 5708 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5709 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5710 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5711 | struct btrfs_path *path; |
5d4f98a2 YZ |
5712 | struct extent_buffer *leaf; |
5713 | int type; | |
5714 | u32 size; | |
26b8003f | 5715 | |
5d4f98a2 YZ |
5716 | if (parent > 0) |
5717 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5718 | else | |
5719 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5720 | |
5d4f98a2 | 5721 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5722 | |
5723 | path = btrfs_alloc_path(); | |
db5b493a TI |
5724 | if (!path) |
5725 | return -ENOMEM; | |
47e4bb98 | 5726 | |
b9473439 | 5727 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5728 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5729 | ins, size); | |
ccd467d6 | 5730 | BUG_ON(ret); |
0f9dd46c | 5731 | |
5d4f98a2 YZ |
5732 | leaf = path->nodes[0]; |
5733 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5734 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5735 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5736 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5737 | btrfs_set_extent_flags(leaf, extent_item, | |
5738 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5739 | ||
5740 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5741 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5742 | if (parent > 0) { | |
5743 | struct btrfs_shared_data_ref *ref; | |
5744 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5745 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5746 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5747 | } else { | |
5748 | struct btrfs_extent_data_ref *ref; | |
5749 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5750 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5751 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5752 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5753 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5754 | } | |
47e4bb98 CM |
5755 | |
5756 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5757 | btrfs_free_path(path); |
f510cfec | 5758 | |
f0486c68 | 5759 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5760 | if (ret) { |
d397712b CM |
5761 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5762 | "%llu\n", (unsigned long long)ins->objectid, | |
5763 | (unsigned long long)ins->offset); | |
f5947066 CM |
5764 | BUG(); |
5765 | } | |
e6dcd2dc CM |
5766 | return ret; |
5767 | } | |
5768 | ||
5d4f98a2 YZ |
5769 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5770 | struct btrfs_root *root, | |
5771 | u64 parent, u64 root_objectid, | |
5772 | u64 flags, struct btrfs_disk_key *key, | |
5773 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5774 | { |
5775 | int ret; | |
5d4f98a2 YZ |
5776 | struct btrfs_fs_info *fs_info = root->fs_info; |
5777 | struct btrfs_extent_item *extent_item; | |
5778 | struct btrfs_tree_block_info *block_info; | |
5779 | struct btrfs_extent_inline_ref *iref; | |
5780 | struct btrfs_path *path; | |
5781 | struct extent_buffer *leaf; | |
5782 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5783 | |
5d4f98a2 | 5784 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5785 | if (!path) |
5786 | return -ENOMEM; | |
56bec294 | 5787 | |
5d4f98a2 YZ |
5788 | path->leave_spinning = 1; |
5789 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5790 | ins, size); | |
56bec294 | 5791 | BUG_ON(ret); |
5d4f98a2 YZ |
5792 | |
5793 | leaf = path->nodes[0]; | |
5794 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5795 | struct btrfs_extent_item); | |
5796 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5797 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5798 | btrfs_set_extent_flags(leaf, extent_item, | |
5799 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5800 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5801 | ||
5802 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5803 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5804 | ||
5805 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5806 | if (parent > 0) { | |
5807 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5808 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5809 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5810 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5811 | } else { | |
5812 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5813 | BTRFS_TREE_BLOCK_REF_KEY); | |
5814 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5815 | } | |
5816 | ||
5817 | btrfs_mark_buffer_dirty(leaf); | |
5818 | btrfs_free_path(path); | |
5819 | ||
f0486c68 | 5820 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5821 | if (ret) { |
5822 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5823 | "%llu\n", (unsigned long long)ins->objectid, | |
5824 | (unsigned long long)ins->offset); | |
5825 | BUG(); | |
5826 | } | |
5827 | return ret; | |
5828 | } | |
5829 | ||
5830 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5831 | struct btrfs_root *root, | |
5832 | u64 root_objectid, u64 owner, | |
5833 | u64 offset, struct btrfs_key *ins) | |
5834 | { | |
5835 | int ret; | |
5836 | ||
5837 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5838 | ||
66d7e7f0 AJ |
5839 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
5840 | ins->offset, 0, | |
5841 | root_objectid, owner, offset, | |
5842 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
5843 | return ret; |
5844 | } | |
e02119d5 CM |
5845 | |
5846 | /* | |
5847 | * this is used by the tree logging recovery code. It records that | |
5848 | * an extent has been allocated and makes sure to clear the free | |
5849 | * space cache bits as well | |
5850 | */ | |
5d4f98a2 YZ |
5851 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5852 | struct btrfs_root *root, | |
5853 | u64 root_objectid, u64 owner, u64 offset, | |
5854 | struct btrfs_key *ins) | |
e02119d5 CM |
5855 | { |
5856 | int ret; | |
5857 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5858 | struct btrfs_caching_control *caching_ctl; |
5859 | u64 start = ins->objectid; | |
5860 | u64 num_bytes = ins->offset; | |
e02119d5 | 5861 | |
e02119d5 | 5862 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5863 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5864 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5865 | |
11833d66 YZ |
5866 | if (!caching_ctl) { |
5867 | BUG_ON(!block_group_cache_done(block_group)); | |
5868 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5869 | BUG_ON(ret); | |
5870 | } else { | |
5871 | mutex_lock(&caching_ctl->mutex); | |
5872 | ||
5873 | if (start >= caching_ctl->progress) { | |
5874 | ret = add_excluded_extent(root, start, num_bytes); | |
5875 | BUG_ON(ret); | |
5876 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5877 | ret = btrfs_remove_free_space(block_group, | |
5878 | start, num_bytes); | |
5879 | BUG_ON(ret); | |
5880 | } else { | |
5881 | num_bytes = caching_ctl->progress - start; | |
5882 | ret = btrfs_remove_free_space(block_group, | |
5883 | start, num_bytes); | |
5884 | BUG_ON(ret); | |
5885 | ||
5886 | start = caching_ctl->progress; | |
5887 | num_bytes = ins->objectid + ins->offset - | |
5888 | caching_ctl->progress; | |
5889 | ret = add_excluded_extent(root, start, num_bytes); | |
5890 | BUG_ON(ret); | |
5891 | } | |
5892 | ||
5893 | mutex_unlock(&caching_ctl->mutex); | |
5894 | put_caching_control(caching_ctl); | |
5895 | } | |
5896 | ||
fb25e914 JB |
5897 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
5898 | RESERVE_ALLOC_NO_ACCOUNT); | |
f0486c68 | 5899 | BUG_ON(ret); |
fa9c0d79 | 5900 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
5901 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
5902 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
5903 | return ret; |
5904 | } | |
5905 | ||
65b51a00 CM |
5906 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
5907 | struct btrfs_root *root, | |
4008c04a CM |
5908 | u64 bytenr, u32 blocksize, |
5909 | int level) | |
65b51a00 CM |
5910 | { |
5911 | struct extent_buffer *buf; | |
5912 | ||
5913 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5914 | if (!buf) | |
5915 | return ERR_PTR(-ENOMEM); | |
5916 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 5917 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
5918 | btrfs_tree_lock(buf); |
5919 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
5920 | |
5921 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 5922 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 5923 | |
d0c803c4 | 5924 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
5925 | /* |
5926 | * we allow two log transactions at a time, use different | |
5927 | * EXENT bit to differentiate dirty pages. | |
5928 | */ | |
5929 | if (root->log_transid % 2 == 0) | |
5930 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
5931 | buf->start + buf->len - 1, GFP_NOFS); | |
5932 | else | |
5933 | set_extent_new(&root->dirty_log_pages, buf->start, | |
5934 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
5935 | } else { |
5936 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 5937 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 5938 | } |
65b51a00 | 5939 | trans->blocks_used++; |
b4ce94de | 5940 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5941 | return buf; |
5942 | } | |
5943 | ||
f0486c68 YZ |
5944 | static struct btrfs_block_rsv * |
5945 | use_block_rsv(struct btrfs_trans_handle *trans, | |
5946 | struct btrfs_root *root, u32 blocksize) | |
5947 | { | |
5948 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 5949 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
5950 | int ret; |
5951 | ||
5952 | block_rsv = get_block_rsv(trans, root); | |
5953 | ||
5954 | if (block_rsv->size == 0) { | |
36ba022a | 5955 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
5956 | /* |
5957 | * If we couldn't reserve metadata bytes try and use some from | |
5958 | * the global reserve. | |
5959 | */ | |
5960 | if (ret && block_rsv != global_rsv) { | |
5961 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5962 | if (!ret) | |
5963 | return global_rsv; | |
f0486c68 | 5964 | return ERR_PTR(ret); |
68a82277 | 5965 | } else if (ret) { |
f0486c68 | 5966 | return ERR_PTR(ret); |
68a82277 | 5967 | } |
f0486c68 YZ |
5968 | return block_rsv; |
5969 | } | |
5970 | ||
5971 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
5972 | if (!ret) | |
5973 | return block_rsv; | |
68a82277 | 5974 | if (ret) { |
dff51cd1 DS |
5975 | static DEFINE_RATELIMIT_STATE(_rs, |
5976 | DEFAULT_RATELIMIT_INTERVAL, | |
5977 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
5978 | if (__ratelimit(&_rs)) { | |
5979 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
5980 | WARN_ON(1); | |
5981 | } | |
36ba022a | 5982 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 5983 | if (!ret) { |
68a82277 JB |
5984 | return block_rsv; |
5985 | } else if (ret && block_rsv != global_rsv) { | |
5986 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5987 | if (!ret) | |
5988 | return global_rsv; | |
5989 | } | |
5990 | } | |
f0486c68 | 5991 | |
f0486c68 YZ |
5992 | return ERR_PTR(-ENOSPC); |
5993 | } | |
5994 | ||
5995 | static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
5996 | { | |
5997 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
5998 | block_rsv_release_bytes(block_rsv, NULL, 0); | |
5999 | } | |
6000 | ||
fec577fb | 6001 | /* |
f0486c68 YZ |
6002 | * finds a free extent and does all the dirty work required for allocation |
6003 | * returns the key for the extent through ins, and a tree buffer for | |
6004 | * the first block of the extent through buf. | |
6005 | * | |
fec577fb CM |
6006 | * returns the tree buffer or NULL. |
6007 | */ | |
5f39d397 | 6008 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6009 | struct btrfs_root *root, u32 blocksize, |
6010 | u64 parent, u64 root_objectid, | |
6011 | struct btrfs_disk_key *key, int level, | |
66d7e7f0 | 6012 | u64 hint, u64 empty_size, int for_cow) |
fec577fb | 6013 | { |
e2fa7227 | 6014 | struct btrfs_key ins; |
f0486c68 | 6015 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6016 | struct extent_buffer *buf; |
f0486c68 YZ |
6017 | u64 flags = 0; |
6018 | int ret; | |
6019 | ||
fec577fb | 6020 | |
f0486c68 YZ |
6021 | block_rsv = use_block_rsv(trans, root, blocksize); |
6022 | if (IS_ERR(block_rsv)) | |
6023 | return ERR_CAST(block_rsv); | |
6024 | ||
6025 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
6026 | empty_size, hint, (u64)-1, &ins, 0); | |
fec577fb | 6027 | if (ret) { |
f0486c68 | 6028 | unuse_block_rsv(block_rsv, blocksize); |
54aa1f4d | 6029 | return ERR_PTR(ret); |
fec577fb | 6030 | } |
55c69072 | 6031 | |
4008c04a CM |
6032 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6033 | blocksize, level); | |
f0486c68 YZ |
6034 | BUG_ON(IS_ERR(buf)); |
6035 | ||
6036 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6037 | if (parent == 0) | |
6038 | parent = ins.objectid; | |
6039 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6040 | } else | |
6041 | BUG_ON(parent > 0); | |
6042 | ||
6043 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6044 | struct btrfs_delayed_extent_op *extent_op; | |
6045 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
6046 | BUG_ON(!extent_op); | |
6047 | if (key) | |
6048 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6049 | else | |
6050 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6051 | extent_op->flags_to_set = flags; | |
6052 | extent_op->update_key = 1; | |
6053 | extent_op->update_flags = 1; | |
6054 | extent_op->is_data = 0; | |
6055 | ||
66d7e7f0 AJ |
6056 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6057 | ins.objectid, | |
f0486c68 YZ |
6058 | ins.offset, parent, root_objectid, |
6059 | level, BTRFS_ADD_DELAYED_EXTENT, | |
66d7e7f0 | 6060 | extent_op, for_cow); |
f0486c68 YZ |
6061 | BUG_ON(ret); |
6062 | } | |
fec577fb CM |
6063 | return buf; |
6064 | } | |
a28ec197 | 6065 | |
2c47e605 YZ |
6066 | struct walk_control { |
6067 | u64 refs[BTRFS_MAX_LEVEL]; | |
6068 | u64 flags[BTRFS_MAX_LEVEL]; | |
6069 | struct btrfs_key update_progress; | |
6070 | int stage; | |
6071 | int level; | |
6072 | int shared_level; | |
6073 | int update_ref; | |
6074 | int keep_locks; | |
1c4850e2 YZ |
6075 | int reada_slot; |
6076 | int reada_count; | |
66d7e7f0 | 6077 | int for_reloc; |
2c47e605 YZ |
6078 | }; |
6079 | ||
6080 | #define DROP_REFERENCE 1 | |
6081 | #define UPDATE_BACKREF 2 | |
6082 | ||
1c4850e2 YZ |
6083 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6084 | struct btrfs_root *root, | |
6085 | struct walk_control *wc, | |
6086 | struct btrfs_path *path) | |
6407bf6d | 6087 | { |
1c4850e2 YZ |
6088 | u64 bytenr; |
6089 | u64 generation; | |
6090 | u64 refs; | |
94fcca9f | 6091 | u64 flags; |
5d4f98a2 | 6092 | u32 nritems; |
1c4850e2 YZ |
6093 | u32 blocksize; |
6094 | struct btrfs_key key; | |
6095 | struct extent_buffer *eb; | |
6407bf6d | 6096 | int ret; |
1c4850e2 YZ |
6097 | int slot; |
6098 | int nread = 0; | |
6407bf6d | 6099 | |
1c4850e2 YZ |
6100 | if (path->slots[wc->level] < wc->reada_slot) { |
6101 | wc->reada_count = wc->reada_count * 2 / 3; | |
6102 | wc->reada_count = max(wc->reada_count, 2); | |
6103 | } else { | |
6104 | wc->reada_count = wc->reada_count * 3 / 2; | |
6105 | wc->reada_count = min_t(int, wc->reada_count, | |
6106 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6107 | } | |
7bb86316 | 6108 | |
1c4850e2 YZ |
6109 | eb = path->nodes[wc->level]; |
6110 | nritems = btrfs_header_nritems(eb); | |
6111 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6112 | |
1c4850e2 YZ |
6113 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6114 | if (nread >= wc->reada_count) | |
6115 | break; | |
bd56b302 | 6116 | |
2dd3e67b | 6117 | cond_resched(); |
1c4850e2 YZ |
6118 | bytenr = btrfs_node_blockptr(eb, slot); |
6119 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6120 | |
1c4850e2 YZ |
6121 | if (slot == path->slots[wc->level]) |
6122 | goto reada; | |
5d4f98a2 | 6123 | |
1c4850e2 YZ |
6124 | if (wc->stage == UPDATE_BACKREF && |
6125 | generation <= root->root_key.offset) | |
bd56b302 CM |
6126 | continue; |
6127 | ||
94fcca9f YZ |
6128 | /* We don't lock the tree block, it's OK to be racy here */ |
6129 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6130 | &refs, &flags); | |
6131 | BUG_ON(ret); | |
6132 | BUG_ON(refs == 0); | |
6133 | ||
1c4850e2 | 6134 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6135 | if (refs == 1) |
6136 | goto reada; | |
bd56b302 | 6137 | |
94fcca9f YZ |
6138 | if (wc->level == 1 && |
6139 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6140 | continue; | |
1c4850e2 YZ |
6141 | if (!wc->update_ref || |
6142 | generation <= root->root_key.offset) | |
6143 | continue; | |
6144 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6145 | ret = btrfs_comp_cpu_keys(&key, | |
6146 | &wc->update_progress); | |
6147 | if (ret < 0) | |
6148 | continue; | |
94fcca9f YZ |
6149 | } else { |
6150 | if (wc->level == 1 && | |
6151 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6152 | continue; | |
6407bf6d | 6153 | } |
1c4850e2 YZ |
6154 | reada: |
6155 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6156 | generation); | |
6157 | if (ret) | |
bd56b302 | 6158 | break; |
1c4850e2 | 6159 | nread++; |
20524f02 | 6160 | } |
1c4850e2 | 6161 | wc->reada_slot = slot; |
20524f02 | 6162 | } |
2c47e605 | 6163 | |
f82d02d9 | 6164 | /* |
2c47e605 YZ |
6165 | * hepler to process tree block while walking down the tree. |
6166 | * | |
2c47e605 YZ |
6167 | * when wc->stage == UPDATE_BACKREF, this function updates |
6168 | * back refs for pointers in the block. | |
6169 | * | |
6170 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6171 | */ |
2c47e605 | 6172 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6173 | struct btrfs_root *root, |
2c47e605 | 6174 | struct btrfs_path *path, |
94fcca9f | 6175 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6176 | { |
2c47e605 YZ |
6177 | int level = wc->level; |
6178 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6179 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6180 | int ret; |
6181 | ||
2c47e605 YZ |
6182 | if (wc->stage == UPDATE_BACKREF && |
6183 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6184 | return 1; | |
f82d02d9 | 6185 | |
2c47e605 YZ |
6186 | /* |
6187 | * when reference count of tree block is 1, it won't increase | |
6188 | * again. once full backref flag is set, we never clear it. | |
6189 | */ | |
94fcca9f YZ |
6190 | if (lookup_info && |
6191 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6192 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6193 | BUG_ON(!path->locks[level]); |
6194 | ret = btrfs_lookup_extent_info(trans, root, | |
6195 | eb->start, eb->len, | |
6196 | &wc->refs[level], | |
6197 | &wc->flags[level]); | |
6198 | BUG_ON(ret); | |
6199 | BUG_ON(wc->refs[level] == 0); | |
6200 | } | |
5d4f98a2 | 6201 | |
2c47e605 YZ |
6202 | if (wc->stage == DROP_REFERENCE) { |
6203 | if (wc->refs[level] > 1) | |
6204 | return 1; | |
f82d02d9 | 6205 | |
2c47e605 | 6206 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6207 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6208 | path->locks[level] = 0; |
6209 | } | |
6210 | return 0; | |
6211 | } | |
f82d02d9 | 6212 | |
2c47e605 YZ |
6213 | /* wc->stage == UPDATE_BACKREF */ |
6214 | if (!(wc->flags[level] & flag)) { | |
6215 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6216 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
f82d02d9 | 6217 | BUG_ON(ret); |
66d7e7f0 | 6218 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
2c47e605 YZ |
6219 | BUG_ON(ret); |
6220 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
6221 | eb->len, flag, 0); | |
6222 | BUG_ON(ret); | |
6223 | wc->flags[level] |= flag; | |
6224 | } | |
6225 | ||
6226 | /* | |
6227 | * the block is shared by multiple trees, so it's not good to | |
6228 | * keep the tree lock | |
6229 | */ | |
6230 | if (path->locks[level] && level > 0) { | |
bd681513 | 6231 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6232 | path->locks[level] = 0; |
6233 | } | |
6234 | return 0; | |
6235 | } | |
6236 | ||
1c4850e2 YZ |
6237 | /* |
6238 | * hepler to process tree block pointer. | |
6239 | * | |
6240 | * when wc->stage == DROP_REFERENCE, this function checks | |
6241 | * reference count of the block pointed to. if the block | |
6242 | * is shared and we need update back refs for the subtree | |
6243 | * rooted at the block, this function changes wc->stage to | |
6244 | * UPDATE_BACKREF. if the block is shared and there is no | |
6245 | * need to update back, this function drops the reference | |
6246 | * to the block. | |
6247 | * | |
6248 | * NOTE: return value 1 means we should stop walking down. | |
6249 | */ | |
6250 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6251 | struct btrfs_root *root, | |
6252 | struct btrfs_path *path, | |
94fcca9f | 6253 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6254 | { |
6255 | u64 bytenr; | |
6256 | u64 generation; | |
6257 | u64 parent; | |
6258 | u32 blocksize; | |
6259 | struct btrfs_key key; | |
6260 | struct extent_buffer *next; | |
6261 | int level = wc->level; | |
6262 | int reada = 0; | |
6263 | int ret = 0; | |
6264 | ||
6265 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6266 | path->slots[level]); | |
6267 | /* | |
6268 | * if the lower level block was created before the snapshot | |
6269 | * was created, we know there is no need to update back refs | |
6270 | * for the subtree | |
6271 | */ | |
6272 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6273 | generation <= root->root_key.offset) { |
6274 | *lookup_info = 1; | |
1c4850e2 | 6275 | return 1; |
94fcca9f | 6276 | } |
1c4850e2 YZ |
6277 | |
6278 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6279 | blocksize = btrfs_level_size(root, level - 1); | |
6280 | ||
6281 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6282 | if (!next) { | |
6283 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6284 | if (!next) |
6285 | return -ENOMEM; | |
1c4850e2 YZ |
6286 | reada = 1; |
6287 | } | |
6288 | btrfs_tree_lock(next); | |
6289 | btrfs_set_lock_blocking(next); | |
6290 | ||
94fcca9f YZ |
6291 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6292 | &wc->refs[level - 1], | |
6293 | &wc->flags[level - 1]); | |
6294 | BUG_ON(ret); | |
6295 | BUG_ON(wc->refs[level - 1] == 0); | |
6296 | *lookup_info = 0; | |
1c4850e2 | 6297 | |
94fcca9f | 6298 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6299 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6300 | if (level == 1 && |
6301 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6302 | goto skip; | |
6303 | ||
1c4850e2 YZ |
6304 | if (!wc->update_ref || |
6305 | generation <= root->root_key.offset) | |
6306 | goto skip; | |
6307 | ||
6308 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6309 | path->slots[level]); | |
6310 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6311 | if (ret < 0) | |
6312 | goto skip; | |
6313 | ||
6314 | wc->stage = UPDATE_BACKREF; | |
6315 | wc->shared_level = level - 1; | |
6316 | } | |
94fcca9f YZ |
6317 | } else { |
6318 | if (level == 1 && | |
6319 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6320 | goto skip; | |
1c4850e2 YZ |
6321 | } |
6322 | ||
6323 | if (!btrfs_buffer_uptodate(next, generation)) { | |
6324 | btrfs_tree_unlock(next); | |
6325 | free_extent_buffer(next); | |
6326 | next = NULL; | |
94fcca9f | 6327 | *lookup_info = 1; |
1c4850e2 YZ |
6328 | } |
6329 | ||
6330 | if (!next) { | |
6331 | if (reada && level == 1) | |
6332 | reada_walk_down(trans, root, wc, path); | |
6333 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6334 | if (!next) |
6335 | return -EIO; | |
1c4850e2 YZ |
6336 | btrfs_tree_lock(next); |
6337 | btrfs_set_lock_blocking(next); | |
6338 | } | |
6339 | ||
6340 | level--; | |
6341 | BUG_ON(level != btrfs_header_level(next)); | |
6342 | path->nodes[level] = next; | |
6343 | path->slots[level] = 0; | |
bd681513 | 6344 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6345 | wc->level = level; |
6346 | if (wc->level == 1) | |
6347 | wc->reada_slot = 0; | |
6348 | return 0; | |
6349 | skip: | |
6350 | wc->refs[level - 1] = 0; | |
6351 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6352 | if (wc->stage == DROP_REFERENCE) { |
6353 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6354 | parent = path->nodes[level]->start; | |
6355 | } else { | |
6356 | BUG_ON(root->root_key.objectid != | |
6357 | btrfs_header_owner(path->nodes[level])); | |
6358 | parent = 0; | |
6359 | } | |
1c4850e2 | 6360 | |
94fcca9f | 6361 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6362 | root->root_key.objectid, level - 1, 0, 0); |
94fcca9f | 6363 | BUG_ON(ret); |
1c4850e2 | 6364 | } |
1c4850e2 YZ |
6365 | btrfs_tree_unlock(next); |
6366 | free_extent_buffer(next); | |
94fcca9f | 6367 | *lookup_info = 1; |
1c4850e2 YZ |
6368 | return 1; |
6369 | } | |
6370 | ||
2c47e605 YZ |
6371 | /* |
6372 | * hepler to process tree block while walking up the tree. | |
6373 | * | |
6374 | * when wc->stage == DROP_REFERENCE, this function drops | |
6375 | * reference count on the block. | |
6376 | * | |
6377 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6378 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6379 | * to UPDATE_BACKREF previously while processing the block. | |
6380 | * | |
6381 | * NOTE: return value 1 means we should stop walking up. | |
6382 | */ | |
6383 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6384 | struct btrfs_root *root, | |
6385 | struct btrfs_path *path, | |
6386 | struct walk_control *wc) | |
6387 | { | |
f0486c68 | 6388 | int ret; |
2c47e605 YZ |
6389 | int level = wc->level; |
6390 | struct extent_buffer *eb = path->nodes[level]; | |
6391 | u64 parent = 0; | |
6392 | ||
6393 | if (wc->stage == UPDATE_BACKREF) { | |
6394 | BUG_ON(wc->shared_level < level); | |
6395 | if (level < wc->shared_level) | |
6396 | goto out; | |
6397 | ||
2c47e605 YZ |
6398 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6399 | if (ret > 0) | |
6400 | wc->update_ref = 0; | |
6401 | ||
6402 | wc->stage = DROP_REFERENCE; | |
6403 | wc->shared_level = -1; | |
6404 | path->slots[level] = 0; | |
6405 | ||
6406 | /* | |
6407 | * check reference count again if the block isn't locked. | |
6408 | * we should start walking down the tree again if reference | |
6409 | * count is one. | |
6410 | */ | |
6411 | if (!path->locks[level]) { | |
6412 | BUG_ON(level == 0); | |
6413 | btrfs_tree_lock(eb); | |
6414 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6415 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6416 | |
6417 | ret = btrfs_lookup_extent_info(trans, root, | |
6418 | eb->start, eb->len, | |
6419 | &wc->refs[level], | |
6420 | &wc->flags[level]); | |
f82d02d9 | 6421 | BUG_ON(ret); |
2c47e605 YZ |
6422 | BUG_ON(wc->refs[level] == 0); |
6423 | if (wc->refs[level] == 1) { | |
bd681513 | 6424 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6425 | return 1; |
6426 | } | |
f82d02d9 | 6427 | } |
2c47e605 | 6428 | } |
f82d02d9 | 6429 | |
2c47e605 YZ |
6430 | /* wc->stage == DROP_REFERENCE */ |
6431 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6432 | |
2c47e605 YZ |
6433 | if (wc->refs[level] == 1) { |
6434 | if (level == 0) { | |
6435 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6436 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6437 | wc->for_reloc); | |
2c47e605 | 6438 | else |
66d7e7f0 AJ |
6439 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6440 | wc->for_reloc); | |
2c47e605 YZ |
6441 | BUG_ON(ret); |
6442 | } | |
6443 | /* make block locked assertion in clean_tree_block happy */ | |
6444 | if (!path->locks[level] && | |
6445 | btrfs_header_generation(eb) == trans->transid) { | |
6446 | btrfs_tree_lock(eb); | |
6447 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6448 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6449 | } |
6450 | clean_tree_block(trans, root, eb); | |
6451 | } | |
6452 | ||
6453 | if (eb == root->node) { | |
6454 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6455 | parent = eb->start; | |
6456 | else | |
6457 | BUG_ON(root->root_key.objectid != | |
6458 | btrfs_header_owner(eb)); | |
6459 | } else { | |
6460 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6461 | parent = path->nodes[level + 1]->start; | |
6462 | else | |
6463 | BUG_ON(root->root_key.objectid != | |
6464 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6465 | } |
f82d02d9 | 6466 | |
66d7e7f0 | 6467 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0); |
2c47e605 YZ |
6468 | out: |
6469 | wc->refs[level] = 0; | |
6470 | wc->flags[level] = 0; | |
f0486c68 | 6471 | return 0; |
2c47e605 YZ |
6472 | } |
6473 | ||
6474 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6475 | struct btrfs_root *root, | |
6476 | struct btrfs_path *path, | |
6477 | struct walk_control *wc) | |
6478 | { | |
2c47e605 | 6479 | int level = wc->level; |
94fcca9f | 6480 | int lookup_info = 1; |
2c47e605 YZ |
6481 | int ret; |
6482 | ||
6483 | while (level >= 0) { | |
94fcca9f | 6484 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6485 | if (ret > 0) |
6486 | break; | |
6487 | ||
6488 | if (level == 0) | |
6489 | break; | |
6490 | ||
7a7965f8 YZ |
6491 | if (path->slots[level] >= |
6492 | btrfs_header_nritems(path->nodes[level])) | |
6493 | break; | |
6494 | ||
94fcca9f | 6495 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6496 | if (ret > 0) { |
6497 | path->slots[level]++; | |
6498 | continue; | |
90d2c51d MX |
6499 | } else if (ret < 0) |
6500 | return ret; | |
1c4850e2 | 6501 | level = wc->level; |
f82d02d9 | 6502 | } |
f82d02d9 YZ |
6503 | return 0; |
6504 | } | |
6505 | ||
d397712b | 6506 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6507 | struct btrfs_root *root, |
f82d02d9 | 6508 | struct btrfs_path *path, |
2c47e605 | 6509 | struct walk_control *wc, int max_level) |
20524f02 | 6510 | { |
2c47e605 | 6511 | int level = wc->level; |
20524f02 | 6512 | int ret; |
9f3a7427 | 6513 | |
2c47e605 YZ |
6514 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6515 | while (level < max_level && path->nodes[level]) { | |
6516 | wc->level = level; | |
6517 | if (path->slots[level] + 1 < | |
6518 | btrfs_header_nritems(path->nodes[level])) { | |
6519 | path->slots[level]++; | |
20524f02 CM |
6520 | return 0; |
6521 | } else { | |
2c47e605 YZ |
6522 | ret = walk_up_proc(trans, root, path, wc); |
6523 | if (ret > 0) | |
6524 | return 0; | |
bd56b302 | 6525 | |
2c47e605 | 6526 | if (path->locks[level]) { |
bd681513 CM |
6527 | btrfs_tree_unlock_rw(path->nodes[level], |
6528 | path->locks[level]); | |
2c47e605 | 6529 | path->locks[level] = 0; |
f82d02d9 | 6530 | } |
2c47e605 YZ |
6531 | free_extent_buffer(path->nodes[level]); |
6532 | path->nodes[level] = NULL; | |
6533 | level++; | |
20524f02 CM |
6534 | } |
6535 | } | |
6536 | return 1; | |
6537 | } | |
6538 | ||
9aca1d51 | 6539 | /* |
2c47e605 YZ |
6540 | * drop a subvolume tree. |
6541 | * | |
6542 | * this function traverses the tree freeing any blocks that only | |
6543 | * referenced by the tree. | |
6544 | * | |
6545 | * when a shared tree block is found. this function decreases its | |
6546 | * reference count by one. if update_ref is true, this function | |
6547 | * also make sure backrefs for the shared block and all lower level | |
6548 | * blocks are properly updated. | |
9aca1d51 | 6549 | */ |
cb1b69f4 | 6550 | void btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6551 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6552 | int for_reloc) | |
20524f02 | 6553 | { |
5caf2a00 | 6554 | struct btrfs_path *path; |
2c47e605 YZ |
6555 | struct btrfs_trans_handle *trans; |
6556 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6557 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6558 | struct walk_control *wc; |
6559 | struct btrfs_key key; | |
6560 | int err = 0; | |
6561 | int ret; | |
6562 | int level; | |
20524f02 | 6563 | |
5caf2a00 | 6564 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6565 | if (!path) { |
6566 | err = -ENOMEM; | |
6567 | goto out; | |
6568 | } | |
20524f02 | 6569 | |
2c47e605 | 6570 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6571 | if (!wc) { |
6572 | btrfs_free_path(path); | |
cb1b69f4 TI |
6573 | err = -ENOMEM; |
6574 | goto out; | |
38a1a919 | 6575 | } |
2c47e605 | 6576 | |
a22285a6 | 6577 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 TI |
6578 | BUG_ON(IS_ERR(trans)); |
6579 | ||
3fd0a558 YZ |
6580 | if (block_rsv) |
6581 | trans->block_rsv = block_rsv; | |
2c47e605 | 6582 | |
9f3a7427 | 6583 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6584 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6585 | path->nodes[level] = btrfs_lock_root_node(root); |
6586 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6587 | path->slots[level] = 0; |
bd681513 | 6588 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6589 | memset(&wc->update_progress, 0, |
6590 | sizeof(wc->update_progress)); | |
9f3a7427 | 6591 | } else { |
9f3a7427 | 6592 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6593 | memcpy(&wc->update_progress, &key, |
6594 | sizeof(wc->update_progress)); | |
6595 | ||
6702ed49 | 6596 | level = root_item->drop_level; |
2c47e605 | 6597 | BUG_ON(level == 0); |
6702ed49 | 6598 | path->lowest_level = level; |
2c47e605 YZ |
6599 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6600 | path->lowest_level = 0; | |
6601 | if (ret < 0) { | |
6602 | err = ret; | |
cb1b69f4 | 6603 | goto out_free; |
9f3a7427 | 6604 | } |
1c4850e2 | 6605 | WARN_ON(ret > 0); |
2c47e605 | 6606 | |
7d9eb12c CM |
6607 | /* |
6608 | * unlock our path, this is safe because only this | |
6609 | * function is allowed to delete this snapshot | |
6610 | */ | |
5d4f98a2 | 6611 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6612 | |
6613 | level = btrfs_header_level(root->node); | |
6614 | while (1) { | |
6615 | btrfs_tree_lock(path->nodes[level]); | |
6616 | btrfs_set_lock_blocking(path->nodes[level]); | |
6617 | ||
6618 | ret = btrfs_lookup_extent_info(trans, root, | |
6619 | path->nodes[level]->start, | |
6620 | path->nodes[level]->len, | |
6621 | &wc->refs[level], | |
6622 | &wc->flags[level]); | |
6623 | BUG_ON(ret); | |
6624 | BUG_ON(wc->refs[level] == 0); | |
6625 | ||
6626 | if (level == root_item->drop_level) | |
6627 | break; | |
6628 | ||
6629 | btrfs_tree_unlock(path->nodes[level]); | |
6630 | WARN_ON(wc->refs[level] != 1); | |
6631 | level--; | |
6632 | } | |
9f3a7427 | 6633 | } |
2c47e605 YZ |
6634 | |
6635 | wc->level = level; | |
6636 | wc->shared_level = -1; | |
6637 | wc->stage = DROP_REFERENCE; | |
6638 | wc->update_ref = update_ref; | |
6639 | wc->keep_locks = 0; | |
66d7e7f0 | 6640 | wc->for_reloc = for_reloc; |
1c4850e2 | 6641 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6642 | |
d397712b | 6643 | while (1) { |
2c47e605 YZ |
6644 | ret = walk_down_tree(trans, root, path, wc); |
6645 | if (ret < 0) { | |
6646 | err = ret; | |
20524f02 | 6647 | break; |
2c47e605 | 6648 | } |
9aca1d51 | 6649 | |
2c47e605 YZ |
6650 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6651 | if (ret < 0) { | |
6652 | err = ret; | |
20524f02 | 6653 | break; |
2c47e605 YZ |
6654 | } |
6655 | ||
6656 | if (ret > 0) { | |
6657 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6658 | break; |
6659 | } | |
2c47e605 YZ |
6660 | |
6661 | if (wc->stage == DROP_REFERENCE) { | |
6662 | level = wc->level; | |
6663 | btrfs_node_key(path->nodes[level], | |
6664 | &root_item->drop_progress, | |
6665 | path->slots[level]); | |
6666 | root_item->drop_level = level; | |
6667 | } | |
6668 | ||
6669 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6670 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6671 | ret = btrfs_update_root(trans, tree_root, |
6672 | &root->root_key, | |
6673 | root_item); | |
6674 | BUG_ON(ret); | |
6675 | ||
3fd0a558 | 6676 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6677 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 | 6678 | BUG_ON(IS_ERR(trans)); |
3fd0a558 YZ |
6679 | if (block_rsv) |
6680 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6681 | } |
20524f02 | 6682 | } |
b3b4aa74 | 6683 | btrfs_release_path(path); |
2c47e605 YZ |
6684 | BUG_ON(err); |
6685 | ||
6686 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6687 | BUG_ON(ret); | |
6688 | ||
76dda93c YZ |
6689 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6690 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6691 | NULL, NULL); | |
6692 | BUG_ON(ret < 0); | |
6693 | if (ret > 0) { | |
84cd948c JB |
6694 | /* if we fail to delete the orphan item this time |
6695 | * around, it'll get picked up the next time. | |
6696 | * | |
6697 | * The most common failure here is just -ENOENT. | |
6698 | */ | |
6699 | btrfs_del_orphan_item(trans, tree_root, | |
6700 | root->root_key.objectid); | |
76dda93c YZ |
6701 | } |
6702 | } | |
6703 | ||
6704 | if (root->in_radix) { | |
6705 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6706 | } else { | |
6707 | free_extent_buffer(root->node); | |
6708 | free_extent_buffer(root->commit_root); | |
6709 | kfree(root); | |
6710 | } | |
cb1b69f4 | 6711 | out_free: |
3fd0a558 | 6712 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6713 | kfree(wc); |
5caf2a00 | 6714 | btrfs_free_path(path); |
cb1b69f4 TI |
6715 | out: |
6716 | if (err) | |
6717 | btrfs_std_error(root->fs_info, err); | |
6718 | return; | |
20524f02 | 6719 | } |
9078a3e1 | 6720 | |
2c47e605 YZ |
6721 | /* |
6722 | * drop subtree rooted at tree block 'node'. | |
6723 | * | |
6724 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 6725 | * only used by relocation code |
2c47e605 | 6726 | */ |
f82d02d9 YZ |
6727 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6728 | struct btrfs_root *root, | |
6729 | struct extent_buffer *node, | |
6730 | struct extent_buffer *parent) | |
6731 | { | |
6732 | struct btrfs_path *path; | |
2c47e605 | 6733 | struct walk_control *wc; |
f82d02d9 YZ |
6734 | int level; |
6735 | int parent_level; | |
6736 | int ret = 0; | |
6737 | int wret; | |
6738 | ||
2c47e605 YZ |
6739 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6740 | ||
f82d02d9 | 6741 | path = btrfs_alloc_path(); |
db5b493a TI |
6742 | if (!path) |
6743 | return -ENOMEM; | |
f82d02d9 | 6744 | |
2c47e605 | 6745 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6746 | if (!wc) { |
6747 | btrfs_free_path(path); | |
6748 | return -ENOMEM; | |
6749 | } | |
2c47e605 | 6750 | |
b9447ef8 | 6751 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6752 | parent_level = btrfs_header_level(parent); |
6753 | extent_buffer_get(parent); | |
6754 | path->nodes[parent_level] = parent; | |
6755 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6756 | ||
b9447ef8 | 6757 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6758 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6759 | path->nodes[level] = node; |
6760 | path->slots[level] = 0; | |
bd681513 | 6761 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6762 | |
6763 | wc->refs[parent_level] = 1; | |
6764 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6765 | wc->level = level; | |
6766 | wc->shared_level = -1; | |
6767 | wc->stage = DROP_REFERENCE; | |
6768 | wc->update_ref = 0; | |
6769 | wc->keep_locks = 1; | |
66d7e7f0 | 6770 | wc->for_reloc = 1; |
1c4850e2 | 6771 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6772 | |
6773 | while (1) { | |
2c47e605 YZ |
6774 | wret = walk_down_tree(trans, root, path, wc); |
6775 | if (wret < 0) { | |
f82d02d9 | 6776 | ret = wret; |
f82d02d9 | 6777 | break; |
2c47e605 | 6778 | } |
f82d02d9 | 6779 | |
2c47e605 | 6780 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6781 | if (wret < 0) |
6782 | ret = wret; | |
6783 | if (wret != 0) | |
6784 | break; | |
6785 | } | |
6786 | ||
2c47e605 | 6787 | kfree(wc); |
f82d02d9 YZ |
6788 | btrfs_free_path(path); |
6789 | return ret; | |
6790 | } | |
6791 | ||
ec44a35c CM |
6792 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6793 | { | |
6794 | u64 num_devices; | |
6795 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6796 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6797 | ||
cd02dca5 CM |
6798 | /* |
6799 | * we add in the count of missing devices because we want | |
6800 | * to make sure that any RAID levels on a degraded FS | |
6801 | * continue to be honored. | |
6802 | */ | |
6803 | num_devices = root->fs_info->fs_devices->rw_devices + | |
6804 | root->fs_info->fs_devices->missing_devices; | |
6805 | ||
ec44a35c CM |
6806 | if (num_devices == 1) { |
6807 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6808 | stripped = flags & ~stripped; | |
6809 | ||
6810 | /* turn raid0 into single device chunks */ | |
6811 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6812 | return stripped; | |
6813 | ||
6814 | /* turn mirroring into duplication */ | |
6815 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6816 | BTRFS_BLOCK_GROUP_RAID10)) | |
6817 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
6818 | return flags; | |
6819 | } else { | |
6820 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6821 | if (flags & stripped) |
6822 | return flags; | |
6823 | ||
6824 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6825 | stripped = flags & ~stripped; | |
6826 | ||
6827 | /* switch duplicated blocks with raid1 */ | |
6828 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6829 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6830 | ||
6831 | /* turn single device chunks into raid0 */ | |
6832 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
6833 | } | |
6834 | return flags; | |
6835 | } | |
6836 | ||
199c36ea | 6837 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 6838 | { |
f0486c68 YZ |
6839 | struct btrfs_space_info *sinfo = cache->space_info; |
6840 | u64 num_bytes; | |
199c36ea | 6841 | u64 min_allocable_bytes; |
f0486c68 | 6842 | int ret = -ENOSPC; |
0ef3e66b | 6843 | |
c286ac48 | 6844 | |
199c36ea MX |
6845 | /* |
6846 | * We need some metadata space and system metadata space for | |
6847 | * allocating chunks in some corner cases until we force to set | |
6848 | * it to be readonly. | |
6849 | */ | |
6850 | if ((sinfo->flags & | |
6851 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
6852 | !force) | |
6853 | min_allocable_bytes = 1 * 1024 * 1024; | |
6854 | else | |
6855 | min_allocable_bytes = 0; | |
6856 | ||
f0486c68 YZ |
6857 | spin_lock(&sinfo->lock); |
6858 | spin_lock(&cache->lock); | |
61cfea9b W |
6859 | |
6860 | if (cache->ro) { | |
6861 | ret = 0; | |
6862 | goto out; | |
6863 | } | |
6864 | ||
f0486c68 YZ |
6865 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
6866 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6867 | ||
6868 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
6869 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
6870 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 6871 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
6872 | cache->ro = 1; |
6873 | ret = 0; | |
6874 | } | |
61cfea9b | 6875 | out: |
f0486c68 YZ |
6876 | spin_unlock(&cache->lock); |
6877 | spin_unlock(&sinfo->lock); | |
6878 | return ret; | |
6879 | } | |
7d9eb12c | 6880 | |
f0486c68 YZ |
6881 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
6882 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 6883 | |
f0486c68 YZ |
6884 | { |
6885 | struct btrfs_trans_handle *trans; | |
6886 | u64 alloc_flags; | |
6887 | int ret; | |
7d9eb12c | 6888 | |
f0486c68 | 6889 | BUG_ON(cache->ro); |
0ef3e66b | 6890 | |
ff5714cc | 6891 | trans = btrfs_join_transaction(root); |
f0486c68 | 6892 | BUG_ON(IS_ERR(trans)); |
5d4f98a2 | 6893 | |
f0486c68 YZ |
6894 | alloc_flags = update_block_group_flags(root, cache->flags); |
6895 | if (alloc_flags != cache->flags) | |
0e4f8f88 CM |
6896 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6897 | CHUNK_ALLOC_FORCE); | |
5d4f98a2 | 6898 | |
199c36ea | 6899 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
6900 | if (!ret) |
6901 | goto out; | |
6902 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
6903 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6904 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
6905 | if (ret < 0) |
6906 | goto out; | |
199c36ea | 6907 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
6908 | out: |
6909 | btrfs_end_transaction(trans, root); | |
6910 | return ret; | |
6911 | } | |
5d4f98a2 | 6912 | |
c87f08ca CM |
6913 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
6914 | struct btrfs_root *root, u64 type) | |
6915 | { | |
6916 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
6917 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6918 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
6919 | } |
6920 | ||
6d07bcec MX |
6921 | /* |
6922 | * helper to account the unused space of all the readonly block group in the | |
6923 | * list. takes mirrors into account. | |
6924 | */ | |
6925 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
6926 | { | |
6927 | struct btrfs_block_group_cache *block_group; | |
6928 | u64 free_bytes = 0; | |
6929 | int factor; | |
6930 | ||
6931 | list_for_each_entry(block_group, groups_list, list) { | |
6932 | spin_lock(&block_group->lock); | |
6933 | ||
6934 | if (!block_group->ro) { | |
6935 | spin_unlock(&block_group->lock); | |
6936 | continue; | |
6937 | } | |
6938 | ||
6939 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6940 | BTRFS_BLOCK_GROUP_RAID10 | | |
6941 | BTRFS_BLOCK_GROUP_DUP)) | |
6942 | factor = 2; | |
6943 | else | |
6944 | factor = 1; | |
6945 | ||
6946 | free_bytes += (block_group->key.offset - | |
6947 | btrfs_block_group_used(&block_group->item)) * | |
6948 | factor; | |
6949 | ||
6950 | spin_unlock(&block_group->lock); | |
6951 | } | |
6952 | ||
6953 | return free_bytes; | |
6954 | } | |
6955 | ||
6956 | /* | |
6957 | * helper to account the unused space of all the readonly block group in the | |
6958 | * space_info. takes mirrors into account. | |
6959 | */ | |
6960 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
6961 | { | |
6962 | int i; | |
6963 | u64 free_bytes = 0; | |
6964 | ||
6965 | spin_lock(&sinfo->lock); | |
6966 | ||
6967 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
6968 | if (!list_empty(&sinfo->block_groups[i])) | |
6969 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
6970 | &sinfo->block_groups[i]); | |
6971 | ||
6972 | spin_unlock(&sinfo->lock); | |
6973 | ||
6974 | return free_bytes; | |
6975 | } | |
6976 | ||
f0486c68 YZ |
6977 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
6978 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 6979 | { |
f0486c68 YZ |
6980 | struct btrfs_space_info *sinfo = cache->space_info; |
6981 | u64 num_bytes; | |
6982 | ||
6983 | BUG_ON(!cache->ro); | |
6984 | ||
6985 | spin_lock(&sinfo->lock); | |
6986 | spin_lock(&cache->lock); | |
6987 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
6988 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6989 | sinfo->bytes_readonly -= num_bytes; | |
6990 | cache->ro = 0; | |
6991 | spin_unlock(&cache->lock); | |
6992 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
6993 | return 0; |
6994 | } | |
6995 | ||
ba1bf481 JB |
6996 | /* |
6997 | * checks to see if its even possible to relocate this block group. | |
6998 | * | |
6999 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7000 | * ok to go ahead and try. | |
7001 | */ | |
7002 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7003 | { |
ba1bf481 JB |
7004 | struct btrfs_block_group_cache *block_group; |
7005 | struct btrfs_space_info *space_info; | |
7006 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7007 | struct btrfs_device *device; | |
cdcb725c | 7008 | u64 min_free; |
6719db6a JB |
7009 | u64 dev_min = 1; |
7010 | u64 dev_nr = 0; | |
cdcb725c | 7011 | int index; |
ba1bf481 JB |
7012 | int full = 0; |
7013 | int ret = 0; | |
1a40e23b | 7014 | |
ba1bf481 | 7015 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7016 | |
ba1bf481 JB |
7017 | /* odd, couldn't find the block group, leave it alone */ |
7018 | if (!block_group) | |
7019 | return -1; | |
1a40e23b | 7020 | |
cdcb725c | 7021 | min_free = btrfs_block_group_used(&block_group->item); |
7022 | ||
ba1bf481 | 7023 | /* no bytes used, we're good */ |
cdcb725c | 7024 | if (!min_free) |
1a40e23b ZY |
7025 | goto out; |
7026 | ||
ba1bf481 JB |
7027 | space_info = block_group->space_info; |
7028 | spin_lock(&space_info->lock); | |
17d217fe | 7029 | |
ba1bf481 | 7030 | full = space_info->full; |
17d217fe | 7031 | |
ba1bf481 JB |
7032 | /* |
7033 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7034 | * relocate it unless we're able to allocate a new chunk below. |
7035 | * | |
7036 | * Otherwise, we need to make sure we have room in the space to handle | |
7037 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7038 | */ |
7ce618db | 7039 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7040 | (space_info->bytes_used + space_info->bytes_reserved + |
7041 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7042 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7043 | spin_unlock(&space_info->lock); |
7044 | goto out; | |
17d217fe | 7045 | } |
ba1bf481 | 7046 | spin_unlock(&space_info->lock); |
ea8c2819 | 7047 | |
ba1bf481 JB |
7048 | /* |
7049 | * ok we don't have enough space, but maybe we have free space on our | |
7050 | * devices to allocate new chunks for relocation, so loop through our | |
7051 | * alloc devices and guess if we have enough space. However, if we | |
7052 | * were marked as full, then we know there aren't enough chunks, and we | |
7053 | * can just return. | |
7054 | */ | |
7055 | ret = -1; | |
7056 | if (full) | |
7057 | goto out; | |
ea8c2819 | 7058 | |
cdcb725c | 7059 | /* |
7060 | * index: | |
7061 | * 0: raid10 | |
7062 | * 1: raid1 | |
7063 | * 2: dup | |
7064 | * 3: raid0 | |
7065 | * 4: single | |
7066 | */ | |
7067 | index = get_block_group_index(block_group); | |
7068 | if (index == 0) { | |
7069 | dev_min = 4; | |
6719db6a JB |
7070 | /* Divide by 2 */ |
7071 | min_free >>= 1; | |
cdcb725c | 7072 | } else if (index == 1) { |
7073 | dev_min = 2; | |
7074 | } else if (index == 2) { | |
6719db6a JB |
7075 | /* Multiply by 2 */ |
7076 | min_free <<= 1; | |
cdcb725c | 7077 | } else if (index == 3) { |
7078 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7079 | do_div(min_free, dev_min); |
cdcb725c | 7080 | } |
7081 | ||
ba1bf481 JB |
7082 | mutex_lock(&root->fs_info->chunk_mutex); |
7083 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7084 | u64 dev_offset; |
56bec294 | 7085 | |
ba1bf481 JB |
7086 | /* |
7087 | * check to make sure we can actually find a chunk with enough | |
7088 | * space to fit our block group in. | |
7089 | */ | |
7090 | if (device->total_bytes > device->bytes_used + min_free) { | |
7091 | ret = find_free_dev_extent(NULL, device, min_free, | |
7bfc837d | 7092 | &dev_offset, NULL); |
ba1bf481 | 7093 | if (!ret) |
cdcb725c | 7094 | dev_nr++; |
7095 | ||
7096 | if (dev_nr >= dev_min) | |
73e48b27 | 7097 | break; |
cdcb725c | 7098 | |
ba1bf481 | 7099 | ret = -1; |
725c8463 | 7100 | } |
edbd8d4e | 7101 | } |
ba1bf481 | 7102 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7103 | out: |
ba1bf481 | 7104 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7105 | return ret; |
7106 | } | |
7107 | ||
b2950863 CH |
7108 | static int find_first_block_group(struct btrfs_root *root, |
7109 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7110 | { |
925baedd | 7111 | int ret = 0; |
0b86a832 CM |
7112 | struct btrfs_key found_key; |
7113 | struct extent_buffer *leaf; | |
7114 | int slot; | |
edbd8d4e | 7115 | |
0b86a832 CM |
7116 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7117 | if (ret < 0) | |
925baedd CM |
7118 | goto out; |
7119 | ||
d397712b | 7120 | while (1) { |
0b86a832 | 7121 | slot = path->slots[0]; |
edbd8d4e | 7122 | leaf = path->nodes[0]; |
0b86a832 CM |
7123 | if (slot >= btrfs_header_nritems(leaf)) { |
7124 | ret = btrfs_next_leaf(root, path); | |
7125 | if (ret == 0) | |
7126 | continue; | |
7127 | if (ret < 0) | |
925baedd | 7128 | goto out; |
0b86a832 | 7129 | break; |
edbd8d4e | 7130 | } |
0b86a832 | 7131 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7132 | |
0b86a832 | 7133 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7134 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7135 | ret = 0; | |
7136 | goto out; | |
7137 | } | |
0b86a832 | 7138 | path->slots[0]++; |
edbd8d4e | 7139 | } |
925baedd | 7140 | out: |
0b86a832 | 7141 | return ret; |
edbd8d4e CM |
7142 | } |
7143 | ||
0af3d00b JB |
7144 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7145 | { | |
7146 | struct btrfs_block_group_cache *block_group; | |
7147 | u64 last = 0; | |
7148 | ||
7149 | while (1) { | |
7150 | struct inode *inode; | |
7151 | ||
7152 | block_group = btrfs_lookup_first_block_group(info, last); | |
7153 | while (block_group) { | |
7154 | spin_lock(&block_group->lock); | |
7155 | if (block_group->iref) | |
7156 | break; | |
7157 | spin_unlock(&block_group->lock); | |
7158 | block_group = next_block_group(info->tree_root, | |
7159 | block_group); | |
7160 | } | |
7161 | if (!block_group) { | |
7162 | if (last == 0) | |
7163 | break; | |
7164 | last = 0; | |
7165 | continue; | |
7166 | } | |
7167 | ||
7168 | inode = block_group->inode; | |
7169 | block_group->iref = 0; | |
7170 | block_group->inode = NULL; | |
7171 | spin_unlock(&block_group->lock); | |
7172 | iput(inode); | |
7173 | last = block_group->key.objectid + block_group->key.offset; | |
7174 | btrfs_put_block_group(block_group); | |
7175 | } | |
7176 | } | |
7177 | ||
1a40e23b ZY |
7178 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7179 | { | |
7180 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7181 | struct btrfs_space_info *space_info; |
11833d66 | 7182 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7183 | struct rb_node *n; |
7184 | ||
11833d66 YZ |
7185 | down_write(&info->extent_commit_sem); |
7186 | while (!list_empty(&info->caching_block_groups)) { | |
7187 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7188 | struct btrfs_caching_control, list); | |
7189 | list_del(&caching_ctl->list); | |
7190 | put_caching_control(caching_ctl); | |
7191 | } | |
7192 | up_write(&info->extent_commit_sem); | |
7193 | ||
1a40e23b ZY |
7194 | spin_lock(&info->block_group_cache_lock); |
7195 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7196 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7197 | cache_node); | |
1a40e23b ZY |
7198 | rb_erase(&block_group->cache_node, |
7199 | &info->block_group_cache_tree); | |
d899e052 YZ |
7200 | spin_unlock(&info->block_group_cache_lock); |
7201 | ||
80eb234a | 7202 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7203 | list_del(&block_group->list); |
80eb234a | 7204 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7205 | |
817d52f8 | 7206 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7207 | wait_block_group_cache_done(block_group); |
817d52f8 | 7208 | |
3c14874a JB |
7209 | /* |
7210 | * We haven't cached this block group, which means we could | |
7211 | * possibly have excluded extents on this block group. | |
7212 | */ | |
7213 | if (block_group->cached == BTRFS_CACHE_NO) | |
7214 | free_excluded_extents(info->extent_root, block_group); | |
7215 | ||
817d52f8 | 7216 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7217 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7218 | |
7219 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7220 | } |
7221 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7222 | |
7223 | /* now that all the block groups are freed, go through and | |
7224 | * free all the space_info structs. This is only called during | |
7225 | * the final stages of unmount, and so we know nobody is | |
7226 | * using them. We call synchronize_rcu() once before we start, | |
7227 | * just to be on the safe side. | |
7228 | */ | |
7229 | synchronize_rcu(); | |
7230 | ||
8929ecfa YZ |
7231 | release_global_block_rsv(info); |
7232 | ||
4184ea7f CM |
7233 | while(!list_empty(&info->space_info)) { |
7234 | space_info = list_entry(info->space_info.next, | |
7235 | struct btrfs_space_info, | |
7236 | list); | |
f0486c68 | 7237 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7238 | space_info->bytes_reserved > 0 || |
7239 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7240 | WARN_ON(1); |
7241 | dump_space_info(space_info, 0, 0); | |
7242 | } | |
4184ea7f CM |
7243 | list_del(&space_info->list); |
7244 | kfree(space_info); | |
7245 | } | |
1a40e23b ZY |
7246 | return 0; |
7247 | } | |
7248 | ||
b742bb82 YZ |
7249 | static void __link_block_group(struct btrfs_space_info *space_info, |
7250 | struct btrfs_block_group_cache *cache) | |
7251 | { | |
7252 | int index = get_block_group_index(cache); | |
7253 | ||
7254 | down_write(&space_info->groups_sem); | |
7255 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7256 | up_write(&space_info->groups_sem); | |
7257 | } | |
7258 | ||
9078a3e1 CM |
7259 | int btrfs_read_block_groups(struct btrfs_root *root) |
7260 | { | |
7261 | struct btrfs_path *path; | |
7262 | int ret; | |
9078a3e1 | 7263 | struct btrfs_block_group_cache *cache; |
be744175 | 7264 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7265 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7266 | struct btrfs_key key; |
7267 | struct btrfs_key found_key; | |
5f39d397 | 7268 | struct extent_buffer *leaf; |
0af3d00b JB |
7269 | int need_clear = 0; |
7270 | u64 cache_gen; | |
96b5179d | 7271 | |
be744175 | 7272 | root = info->extent_root; |
9078a3e1 | 7273 | key.objectid = 0; |
0b86a832 | 7274 | key.offset = 0; |
9078a3e1 | 7275 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7276 | path = btrfs_alloc_path(); |
7277 | if (!path) | |
7278 | return -ENOMEM; | |
026fd317 | 7279 | path->reada = 1; |
9078a3e1 | 7280 | |
6c41761f | 7281 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7282 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7283 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7284 | need_clear = 1; |
88c2ba3b JB |
7285 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7286 | need_clear = 1; | |
0af3d00b | 7287 | |
d397712b | 7288 | while (1) { |
0b86a832 | 7289 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7290 | if (ret > 0) |
7291 | break; | |
0b86a832 CM |
7292 | if (ret != 0) |
7293 | goto error; | |
5f39d397 CM |
7294 | leaf = path->nodes[0]; |
7295 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7296 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7297 | if (!cache) { |
0b86a832 | 7298 | ret = -ENOMEM; |
f0486c68 | 7299 | goto error; |
9078a3e1 | 7300 | } |
34d52cb6 LZ |
7301 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7302 | GFP_NOFS); | |
7303 | if (!cache->free_space_ctl) { | |
7304 | kfree(cache); | |
7305 | ret = -ENOMEM; | |
7306 | goto error; | |
7307 | } | |
3e1ad54f | 7308 | |
d2fb3437 | 7309 | atomic_set(&cache->count, 1); |
c286ac48 | 7310 | spin_lock_init(&cache->lock); |
817d52f8 | 7311 | cache->fs_info = info; |
0f9dd46c | 7312 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7313 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7314 | |
0af3d00b JB |
7315 | if (need_clear) |
7316 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
7317 | ||
5f39d397 CM |
7318 | read_extent_buffer(leaf, &cache->item, |
7319 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7320 | sizeof(cache->item)); | |
9078a3e1 | 7321 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7322 | |
9078a3e1 | 7323 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7324 | btrfs_release_path(path); |
0b86a832 | 7325 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7326 | cache->sectorsize = root->sectorsize; |
7327 | ||
34d52cb6 LZ |
7328 | btrfs_init_free_space_ctl(cache); |
7329 | ||
3c14874a JB |
7330 | /* |
7331 | * We need to exclude the super stripes now so that the space | |
7332 | * info has super bytes accounted for, otherwise we'll think | |
7333 | * we have more space than we actually do. | |
7334 | */ | |
7335 | exclude_super_stripes(root, cache); | |
7336 | ||
817d52f8 JB |
7337 | /* |
7338 | * check for two cases, either we are full, and therefore | |
7339 | * don't need to bother with the caching work since we won't | |
7340 | * find any space, or we are empty, and we can just add all | |
7341 | * the space in and be done with it. This saves us _alot_ of | |
7342 | * time, particularly in the full case. | |
7343 | */ | |
7344 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7345 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7346 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7347 | free_excluded_extents(root, cache); |
817d52f8 | 7348 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7349 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7350 | cache->cached = BTRFS_CACHE_FINISHED; |
7351 | add_new_free_space(cache, root->fs_info, | |
7352 | found_key.objectid, | |
7353 | found_key.objectid + | |
7354 | found_key.offset); | |
11833d66 | 7355 | free_excluded_extents(root, cache); |
817d52f8 | 7356 | } |
96b5179d | 7357 | |
6324fbf3 CM |
7358 | ret = update_space_info(info, cache->flags, found_key.offset, |
7359 | btrfs_block_group_used(&cache->item), | |
7360 | &space_info); | |
7361 | BUG_ON(ret); | |
7362 | cache->space_info = space_info; | |
1b2da372 | 7363 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7364 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7365 | spin_unlock(&cache->space_info->lock); |
7366 | ||
b742bb82 | 7367 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7368 | |
7369 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7370 | BUG_ON(ret); | |
75ccf47d CM |
7371 | |
7372 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7373 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7374 | set_block_group_ro(cache, 1); |
9078a3e1 | 7375 | } |
b742bb82 YZ |
7376 | |
7377 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7378 | if (!(get_alloc_profile(root, space_info->flags) & | |
7379 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7380 | BTRFS_BLOCK_GROUP_RAID1 | | |
7381 | BTRFS_BLOCK_GROUP_DUP))) | |
7382 | continue; | |
7383 | /* | |
7384 | * avoid allocating from un-mirrored block group if there are | |
7385 | * mirrored block groups. | |
7386 | */ | |
7387 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7388 | set_block_group_ro(cache, 1); |
b742bb82 | 7389 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7390 | set_block_group_ro(cache, 1); |
9078a3e1 | 7391 | } |
f0486c68 YZ |
7392 | |
7393 | init_global_block_rsv(info); | |
0b86a832 CM |
7394 | ret = 0; |
7395 | error: | |
9078a3e1 | 7396 | btrfs_free_path(path); |
0b86a832 | 7397 | return ret; |
9078a3e1 | 7398 | } |
6324fbf3 CM |
7399 | |
7400 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7401 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7402 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7403 | u64 size) |
7404 | { | |
7405 | int ret; | |
6324fbf3 CM |
7406 | struct btrfs_root *extent_root; |
7407 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7408 | |
7409 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7410 | |
12fcfd22 | 7411 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7412 | |
8f18cf13 | 7413 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7414 | if (!cache) |
7415 | return -ENOMEM; | |
34d52cb6 LZ |
7416 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7417 | GFP_NOFS); | |
7418 | if (!cache->free_space_ctl) { | |
7419 | kfree(cache); | |
7420 | return -ENOMEM; | |
7421 | } | |
0f9dd46c | 7422 | |
e17cade2 | 7423 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7424 | cache->key.offset = size; |
d2fb3437 | 7425 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7426 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7427 | cache->fs_info = root->fs_info; |
96303081 | 7428 | |
d2fb3437 | 7429 | atomic_set(&cache->count, 1); |
c286ac48 | 7430 | spin_lock_init(&cache->lock); |
0f9dd46c | 7431 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7432 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7433 | |
34d52cb6 LZ |
7434 | btrfs_init_free_space_ctl(cache); |
7435 | ||
6324fbf3 | 7436 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7437 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7438 | cache->flags = type; | |
7439 | btrfs_set_block_group_flags(&cache->item, type); | |
7440 | ||
11833d66 | 7441 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7442 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7443 | exclude_super_stripes(root, cache); |
96303081 | 7444 | |
817d52f8 JB |
7445 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7446 | chunk_offset + size); | |
7447 | ||
11833d66 YZ |
7448 | free_excluded_extents(root, cache); |
7449 | ||
6324fbf3 CM |
7450 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7451 | &cache->space_info); | |
7452 | BUG_ON(ret); | |
1b2da372 JB |
7453 | |
7454 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7455 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7456 | spin_unlock(&cache->space_info->lock); |
7457 | ||
b742bb82 | 7458 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7459 | |
0f9dd46c JB |
7460 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7461 | BUG_ON(ret); | |
c286ac48 | 7462 | |
6324fbf3 CM |
7463 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7464 | sizeof(cache->item)); | |
7465 | BUG_ON(ret); | |
7466 | ||
d18a2c44 | 7467 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7468 | |
6324fbf3 CM |
7469 | return 0; |
7470 | } | |
1a40e23b ZY |
7471 | |
7472 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
7473 | struct btrfs_root *root, u64 group_start) | |
7474 | { | |
7475 | struct btrfs_path *path; | |
7476 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7477 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7478 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7479 | struct btrfs_key key; |
0af3d00b | 7480 | struct inode *inode; |
1a40e23b | 7481 | int ret; |
89a55897 | 7482 | int factor; |
1a40e23b | 7483 | |
1a40e23b ZY |
7484 | root = root->fs_info->extent_root; |
7485 | ||
7486 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7487 | BUG_ON(!block_group); | |
c146afad | 7488 | BUG_ON(!block_group->ro); |
1a40e23b | 7489 | |
9f7c43c9 | 7490 | /* |
7491 | * Free the reserved super bytes from this block group before | |
7492 | * remove it. | |
7493 | */ | |
7494 | free_excluded_extents(root, block_group); | |
7495 | ||
1a40e23b | 7496 | memcpy(&key, &block_group->key, sizeof(key)); |
89a55897 JB |
7497 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7498 | BTRFS_BLOCK_GROUP_RAID1 | | |
7499 | BTRFS_BLOCK_GROUP_RAID10)) | |
7500 | factor = 2; | |
7501 | else | |
7502 | factor = 1; | |
1a40e23b | 7503 | |
44fb5511 CM |
7504 | /* make sure this block group isn't part of an allocation cluster */ |
7505 | cluster = &root->fs_info->data_alloc_cluster; | |
7506 | spin_lock(&cluster->refill_lock); | |
7507 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7508 | spin_unlock(&cluster->refill_lock); | |
7509 | ||
7510 | /* | |
7511 | * make sure this block group isn't part of a metadata | |
7512 | * allocation cluster | |
7513 | */ | |
7514 | cluster = &root->fs_info->meta_alloc_cluster; | |
7515 | spin_lock(&cluster->refill_lock); | |
7516 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7517 | spin_unlock(&cluster->refill_lock); | |
7518 | ||
1a40e23b | 7519 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7520 | if (!path) { |
7521 | ret = -ENOMEM; | |
7522 | goto out; | |
7523 | } | |
1a40e23b | 7524 | |
10b2f34d | 7525 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7526 | if (!IS_ERR(inode)) { |
b532402e TI |
7527 | ret = btrfs_orphan_add(trans, inode); |
7528 | BUG_ON(ret); | |
0af3d00b JB |
7529 | clear_nlink(inode); |
7530 | /* One for the block groups ref */ | |
7531 | spin_lock(&block_group->lock); | |
7532 | if (block_group->iref) { | |
7533 | block_group->iref = 0; | |
7534 | block_group->inode = NULL; | |
7535 | spin_unlock(&block_group->lock); | |
7536 | iput(inode); | |
7537 | } else { | |
7538 | spin_unlock(&block_group->lock); | |
7539 | } | |
7540 | /* One for our lookup ref */ | |
455757c3 | 7541 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7542 | } |
7543 | ||
7544 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7545 | key.offset = block_group->key.objectid; | |
7546 | key.type = 0; | |
7547 | ||
7548 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7549 | if (ret < 0) | |
7550 | goto out; | |
7551 | if (ret > 0) | |
b3b4aa74 | 7552 | btrfs_release_path(path); |
0af3d00b JB |
7553 | if (ret == 0) { |
7554 | ret = btrfs_del_item(trans, tree_root, path); | |
7555 | if (ret) | |
7556 | goto out; | |
b3b4aa74 | 7557 | btrfs_release_path(path); |
0af3d00b JB |
7558 | } |
7559 | ||
3dfdb934 | 7560 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7561 | rb_erase(&block_group->cache_node, |
7562 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7563 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7564 | |
80eb234a | 7565 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7566 | /* |
7567 | * we must use list_del_init so people can check to see if they | |
7568 | * are still on the list after taking the semaphore | |
7569 | */ | |
7570 | list_del_init(&block_group->list); | |
80eb234a | 7571 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7572 | |
817d52f8 | 7573 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7574 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7575 | |
7576 | btrfs_remove_free_space_cache(block_group); | |
7577 | ||
c146afad YZ |
7578 | spin_lock(&block_group->space_info->lock); |
7579 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7580 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7581 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7582 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7583 | |
0af3d00b JB |
7584 | memcpy(&key, &block_group->key, sizeof(key)); |
7585 | ||
283bb197 | 7586 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7587 | |
fa9c0d79 CM |
7588 | btrfs_put_block_group(block_group); |
7589 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7590 | |
7591 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7592 | if (ret > 0) | |
7593 | ret = -EIO; | |
7594 | if (ret < 0) | |
7595 | goto out; | |
7596 | ||
7597 | ret = btrfs_del_item(trans, root, path); | |
7598 | out: | |
7599 | btrfs_free_path(path); | |
7600 | return ret; | |
7601 | } | |
acce952b | 7602 | |
c59021f8 | 7603 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7604 | { | |
7605 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7606 | struct btrfs_super_block *disk_super; |
7607 | u64 features; | |
7608 | u64 flags; | |
7609 | int mixed = 0; | |
c59021f8 | 7610 | int ret; |
7611 | ||
6c41761f | 7612 | disk_super = fs_info->super_copy; |
1aba86d6 | 7613 | if (!btrfs_super_root(disk_super)) |
7614 | return 1; | |
c59021f8 | 7615 | |
1aba86d6 | 7616 | features = btrfs_super_incompat_flags(disk_super); |
7617 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7618 | mixed = 1; | |
c59021f8 | 7619 | |
1aba86d6 | 7620 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7621 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7622 | if (ret) |
1aba86d6 | 7623 | goto out; |
c59021f8 | 7624 | |
1aba86d6 | 7625 | if (mixed) { |
7626 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7627 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7628 | } else { | |
7629 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7630 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7631 | if (ret) | |
7632 | goto out; | |
7633 | ||
7634 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7635 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7636 | } | |
7637 | out: | |
c59021f8 | 7638 | return ret; |
7639 | } | |
7640 | ||
acce952b | 7641 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7642 | { | |
7643 | return unpin_extent_range(root, start, end); | |
7644 | } | |
7645 | ||
7646 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7647 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7648 | { |
5378e607 | 7649 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7650 | } |
f7039b1d LD |
7651 | |
7652 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7653 | { | |
7654 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7655 | struct btrfs_block_group_cache *cache = NULL; | |
7656 | u64 group_trimmed; | |
7657 | u64 start; | |
7658 | u64 end; | |
7659 | u64 trimmed = 0; | |
7660 | int ret = 0; | |
7661 | ||
7662 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
7663 | ||
7664 | while (cache) { | |
7665 | if (cache->key.objectid >= (range->start + range->len)) { | |
7666 | btrfs_put_block_group(cache); | |
7667 | break; | |
7668 | } | |
7669 | ||
7670 | start = max(range->start, cache->key.objectid); | |
7671 | end = min(range->start + range->len, | |
7672 | cache->key.objectid + cache->key.offset); | |
7673 | ||
7674 | if (end - start >= range->minlen) { | |
7675 | if (!block_group_cache_done(cache)) { | |
7676 | ret = cache_block_group(cache, NULL, root, 0); | |
7677 | if (!ret) | |
7678 | wait_block_group_cache_done(cache); | |
7679 | } | |
7680 | ret = btrfs_trim_block_group(cache, | |
7681 | &group_trimmed, | |
7682 | start, | |
7683 | end, | |
7684 | range->minlen); | |
7685 | ||
7686 | trimmed += group_trimmed; | |
7687 | if (ret) { | |
7688 | btrfs_put_block_group(cache); | |
7689 | break; | |
7690 | } | |
7691 | } | |
7692 | ||
7693 | cache = next_block_group(fs_info->tree_root, cache); | |
7694 | } | |
7695 | ||
7696 | range->len = trimmed; | |
7697 | return ret; | |
7698 | } |