]>
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 | |
9e622d6b MX |
37 | /* |
38 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
40 | * if we really need one. | |
41 | * | |
0e4f8f88 CM |
42 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
43 | * if we have very few chunks already allocated. This is | |
44 | * used as part of the clustering code to help make sure | |
45 | * we have a good pool of storage to cluster in, without | |
46 | * filling the FS with empty chunks | |
47 | * | |
9e622d6b MX |
48 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
49 | * | |
0e4f8f88 CM |
50 | */ |
51 | enum { | |
52 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
53 | CHUNK_ALLOC_LIMITED = 1, |
54 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
55 | }; |
56 | ||
fb25e914 JB |
57 | /* |
58 | * Control how reservations are dealt with. | |
59 | * | |
60 | * RESERVE_FREE - freeing a reservation. | |
61 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
62 | * ENOSPC accounting | |
63 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
64 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
65 | */ | |
66 | enum { | |
67 | RESERVE_FREE = 0, | |
68 | RESERVE_ALLOC = 1, | |
69 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
70 | }; | |
71 | ||
f3465ca4 JB |
72 | static int update_block_group(struct btrfs_trans_handle *trans, |
73 | struct btrfs_root *root, | |
f0486c68 | 74 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
75 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
76 | struct btrfs_root *root, | |
77 | u64 bytenr, u64 num_bytes, u64 parent, | |
78 | u64 root_objectid, u64 owner_objectid, | |
79 | u64 owner_offset, int refs_to_drop, | |
80 | struct btrfs_delayed_extent_op *extra_op); | |
81 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
82 | struct extent_buffer *leaf, | |
83 | struct btrfs_extent_item *ei); | |
84 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
85 | struct btrfs_root *root, | |
86 | u64 parent, u64 root_objectid, | |
87 | u64 flags, u64 owner, u64 offset, | |
88 | struct btrfs_key *ins, int ref_mod); | |
89 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
90 | struct btrfs_root *root, | |
91 | u64 parent, u64 root_objectid, | |
92 | u64 flags, struct btrfs_disk_key *key, | |
93 | int level, struct btrfs_key *ins); | |
6a63209f JB |
94 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
95 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
96 | u64 flags, int force); | |
11833d66 YZ |
97 | static int find_next_key(struct btrfs_path *path, int level, |
98 | struct btrfs_key *key); | |
9ed74f2d JB |
99 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
100 | int dump_block_groups); | |
fb25e914 JB |
101 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
102 | u64 num_bytes, int reserve); | |
6a63209f | 103 | |
817d52f8 JB |
104 | static noinline int |
105 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
106 | { | |
107 | smp_mb(); | |
108 | return cache->cached == BTRFS_CACHE_FINISHED; | |
109 | } | |
110 | ||
0f9dd46c JB |
111 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
112 | { | |
113 | return (cache->flags & bits) == bits; | |
114 | } | |
115 | ||
62a45b60 | 116 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
117 | { |
118 | atomic_inc(&cache->count); | |
119 | } | |
120 | ||
121 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
122 | { | |
f0486c68 YZ |
123 | if (atomic_dec_and_test(&cache->count)) { |
124 | WARN_ON(cache->pinned > 0); | |
125 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 126 | kfree(cache->free_space_ctl); |
11dfe35a | 127 | kfree(cache); |
f0486c68 | 128 | } |
11dfe35a JB |
129 | } |
130 | ||
0f9dd46c JB |
131 | /* |
132 | * this adds the block group to the fs_info rb tree for the block group | |
133 | * cache | |
134 | */ | |
b2950863 | 135 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
136 | struct btrfs_block_group_cache *block_group) |
137 | { | |
138 | struct rb_node **p; | |
139 | struct rb_node *parent = NULL; | |
140 | struct btrfs_block_group_cache *cache; | |
141 | ||
142 | spin_lock(&info->block_group_cache_lock); | |
143 | p = &info->block_group_cache_tree.rb_node; | |
144 | ||
145 | while (*p) { | |
146 | parent = *p; | |
147 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
148 | cache_node); | |
149 | if (block_group->key.objectid < cache->key.objectid) { | |
150 | p = &(*p)->rb_left; | |
151 | } else if (block_group->key.objectid > cache->key.objectid) { | |
152 | p = &(*p)->rb_right; | |
153 | } else { | |
154 | spin_unlock(&info->block_group_cache_lock); | |
155 | return -EEXIST; | |
156 | } | |
157 | } | |
158 | ||
159 | rb_link_node(&block_group->cache_node, parent, p); | |
160 | rb_insert_color(&block_group->cache_node, | |
161 | &info->block_group_cache_tree); | |
162 | spin_unlock(&info->block_group_cache_lock); | |
163 | ||
164 | return 0; | |
165 | } | |
166 | ||
167 | /* | |
168 | * This will return the block group at or after bytenr if contains is 0, else | |
169 | * it will return the block group that contains the bytenr | |
170 | */ | |
171 | static struct btrfs_block_group_cache * | |
172 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
173 | int contains) | |
174 | { | |
175 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
176 | struct rb_node *n; | |
177 | u64 end, start; | |
178 | ||
179 | spin_lock(&info->block_group_cache_lock); | |
180 | n = info->block_group_cache_tree.rb_node; | |
181 | ||
182 | while (n) { | |
183 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
184 | cache_node); | |
185 | end = cache->key.objectid + cache->key.offset - 1; | |
186 | start = cache->key.objectid; | |
187 | ||
188 | if (bytenr < start) { | |
189 | if (!contains && (!ret || start < ret->key.objectid)) | |
190 | ret = cache; | |
191 | n = n->rb_left; | |
192 | } else if (bytenr > start) { | |
193 | if (contains && bytenr <= end) { | |
194 | ret = cache; | |
195 | break; | |
196 | } | |
197 | n = n->rb_right; | |
198 | } else { | |
199 | ret = cache; | |
200 | break; | |
201 | } | |
202 | } | |
d2fb3437 | 203 | if (ret) |
11dfe35a | 204 | btrfs_get_block_group(ret); |
0f9dd46c JB |
205 | spin_unlock(&info->block_group_cache_lock); |
206 | ||
207 | return ret; | |
208 | } | |
209 | ||
11833d66 YZ |
210 | static int add_excluded_extent(struct btrfs_root *root, |
211 | u64 start, u64 num_bytes) | |
817d52f8 | 212 | { |
11833d66 YZ |
213 | u64 end = start + num_bytes - 1; |
214 | set_extent_bits(&root->fs_info->freed_extents[0], | |
215 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
216 | set_extent_bits(&root->fs_info->freed_extents[1], | |
217 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
218 | return 0; | |
219 | } | |
817d52f8 | 220 | |
11833d66 YZ |
221 | static void free_excluded_extents(struct btrfs_root *root, |
222 | struct btrfs_block_group_cache *cache) | |
223 | { | |
224 | u64 start, end; | |
817d52f8 | 225 | |
11833d66 YZ |
226 | start = cache->key.objectid; |
227 | end = start + cache->key.offset - 1; | |
228 | ||
229 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
230 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
231 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
232 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
233 | } |
234 | ||
11833d66 YZ |
235 | static int exclude_super_stripes(struct btrfs_root *root, |
236 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 237 | { |
817d52f8 JB |
238 | u64 bytenr; |
239 | u64 *logical; | |
240 | int stripe_len; | |
241 | int i, nr, ret; | |
242 | ||
06b2331f YZ |
243 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
244 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
245 | cache->bytes_super += stripe_len; | |
246 | ret = add_excluded_extent(root, cache->key.objectid, | |
247 | stripe_len); | |
248 | BUG_ON(ret); | |
249 | } | |
250 | ||
817d52f8 JB |
251 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
252 | bytenr = btrfs_sb_offset(i); | |
253 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
254 | cache->key.objectid, bytenr, | |
255 | 0, &logical, &nr, &stripe_len); | |
256 | BUG_ON(ret); | |
11833d66 | 257 | |
817d52f8 | 258 | while (nr--) { |
1b2da372 | 259 | cache->bytes_super += stripe_len; |
11833d66 YZ |
260 | ret = add_excluded_extent(root, logical[nr], |
261 | stripe_len); | |
262 | BUG_ON(ret); | |
817d52f8 | 263 | } |
11833d66 | 264 | |
817d52f8 JB |
265 | kfree(logical); |
266 | } | |
817d52f8 JB |
267 | return 0; |
268 | } | |
269 | ||
11833d66 YZ |
270 | static struct btrfs_caching_control * |
271 | get_caching_control(struct btrfs_block_group_cache *cache) | |
272 | { | |
273 | struct btrfs_caching_control *ctl; | |
274 | ||
275 | spin_lock(&cache->lock); | |
276 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
277 | spin_unlock(&cache->lock); | |
278 | return NULL; | |
279 | } | |
280 | ||
dde5abee JB |
281 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
282 | if (!cache->caching_ctl) { | |
283 | spin_unlock(&cache->lock); | |
11833d66 YZ |
284 | return NULL; |
285 | } | |
286 | ||
287 | ctl = cache->caching_ctl; | |
288 | atomic_inc(&ctl->count); | |
289 | spin_unlock(&cache->lock); | |
290 | return ctl; | |
291 | } | |
292 | ||
293 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
294 | { | |
295 | if (atomic_dec_and_test(&ctl->count)) | |
296 | kfree(ctl); | |
297 | } | |
298 | ||
0f9dd46c JB |
299 | /* |
300 | * this is only called by cache_block_group, since we could have freed extents | |
301 | * we need to check the pinned_extents for any extents that can't be used yet | |
302 | * since their free space will be released as soon as the transaction commits. | |
303 | */ | |
817d52f8 | 304 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
305 | struct btrfs_fs_info *info, u64 start, u64 end) |
306 | { | |
817d52f8 | 307 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
308 | int ret; |
309 | ||
310 | while (start < end) { | |
11833d66 | 311 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 312 | &extent_start, &extent_end, |
11833d66 | 313 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
314 | if (ret) |
315 | break; | |
316 | ||
06b2331f | 317 | if (extent_start <= start) { |
0f9dd46c JB |
318 | start = extent_end + 1; |
319 | } else if (extent_start > start && extent_start < end) { | |
320 | size = extent_start - start; | |
817d52f8 | 321 | total_added += size; |
ea6a478e JB |
322 | ret = btrfs_add_free_space(block_group, start, |
323 | size); | |
0f9dd46c JB |
324 | BUG_ON(ret); |
325 | start = extent_end + 1; | |
326 | } else { | |
327 | break; | |
328 | } | |
329 | } | |
330 | ||
331 | if (start < end) { | |
332 | size = end - start; | |
817d52f8 | 333 | total_added += size; |
ea6a478e | 334 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
335 | BUG_ON(ret); |
336 | } | |
337 | ||
817d52f8 | 338 | return total_added; |
0f9dd46c JB |
339 | } |
340 | ||
bab39bf9 | 341 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 342 | { |
bab39bf9 JB |
343 | struct btrfs_block_group_cache *block_group; |
344 | struct btrfs_fs_info *fs_info; | |
345 | struct btrfs_caching_control *caching_ctl; | |
346 | struct btrfs_root *extent_root; | |
e37c9e69 | 347 | struct btrfs_path *path; |
5f39d397 | 348 | struct extent_buffer *leaf; |
11833d66 | 349 | struct btrfs_key key; |
817d52f8 | 350 | u64 total_found = 0; |
11833d66 YZ |
351 | u64 last = 0; |
352 | u32 nritems; | |
353 | int ret = 0; | |
f510cfec | 354 | |
bab39bf9 JB |
355 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
356 | block_group = caching_ctl->block_group; | |
357 | fs_info = block_group->fs_info; | |
358 | extent_root = fs_info->extent_root; | |
359 | ||
e37c9e69 CM |
360 | path = btrfs_alloc_path(); |
361 | if (!path) | |
bab39bf9 | 362 | goto out; |
7d7d6068 | 363 | |
817d52f8 | 364 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 365 | |
5cd57b2c | 366 | /* |
817d52f8 JB |
367 | * We don't want to deadlock with somebody trying to allocate a new |
368 | * extent for the extent root while also trying to search the extent | |
369 | * root to add free space. So we skip locking and search the commit | |
370 | * root, since its read-only | |
5cd57b2c CM |
371 | */ |
372 | path->skip_locking = 1; | |
817d52f8 | 373 | path->search_commit_root = 1; |
026fd317 | 374 | path->reada = 1; |
817d52f8 | 375 | |
e4404d6e | 376 | key.objectid = last; |
e37c9e69 | 377 | key.offset = 0; |
11833d66 | 378 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 379 | again: |
11833d66 | 380 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
381 | /* need to make sure the commit_root doesn't disappear */ |
382 | down_read(&fs_info->extent_commit_sem); | |
383 | ||
11833d66 | 384 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 385 | if (ret < 0) |
ef8bbdfe | 386 | goto err; |
a512bbf8 | 387 | |
11833d66 YZ |
388 | leaf = path->nodes[0]; |
389 | nritems = btrfs_header_nritems(leaf); | |
390 | ||
d397712b | 391 | while (1) { |
7841cb28 | 392 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 393 | last = (u64)-1; |
817d52f8 | 394 | break; |
f25784b3 | 395 | } |
817d52f8 | 396 | |
11833d66 YZ |
397 | if (path->slots[0] < nritems) { |
398 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
399 | } else { | |
400 | ret = find_next_key(path, 0, &key); | |
401 | if (ret) | |
e37c9e69 | 402 | break; |
817d52f8 | 403 | |
589d8ade JB |
404 | if (need_resched() || |
405 | btrfs_next_leaf(extent_root, path)) { | |
406 | caching_ctl->progress = last; | |
ff5714cc | 407 | btrfs_release_path(path); |
589d8ade JB |
408 | up_read(&fs_info->extent_commit_sem); |
409 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 410 | cond_resched(); |
589d8ade JB |
411 | goto again; |
412 | } | |
413 | leaf = path->nodes[0]; | |
414 | nritems = btrfs_header_nritems(leaf); | |
415 | continue; | |
11833d66 | 416 | } |
817d52f8 | 417 | |
11833d66 YZ |
418 | if (key.objectid < block_group->key.objectid) { |
419 | path->slots[0]++; | |
817d52f8 | 420 | continue; |
e37c9e69 | 421 | } |
0f9dd46c | 422 | |
e37c9e69 | 423 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 424 | block_group->key.offset) |
e37c9e69 | 425 | break; |
7d7d6068 | 426 | |
11833d66 | 427 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
428 | total_found += add_new_free_space(block_group, |
429 | fs_info, last, | |
430 | key.objectid); | |
7d7d6068 | 431 | last = key.objectid + key.offset; |
817d52f8 | 432 | |
11833d66 YZ |
433 | if (total_found > (1024 * 1024 * 2)) { |
434 | total_found = 0; | |
435 | wake_up(&caching_ctl->wait); | |
436 | } | |
817d52f8 | 437 | } |
e37c9e69 CM |
438 | path->slots[0]++; |
439 | } | |
817d52f8 | 440 | ret = 0; |
e37c9e69 | 441 | |
817d52f8 JB |
442 | total_found += add_new_free_space(block_group, fs_info, last, |
443 | block_group->key.objectid + | |
444 | block_group->key.offset); | |
11833d66 | 445 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
446 | |
447 | spin_lock(&block_group->lock); | |
11833d66 | 448 | block_group->caching_ctl = NULL; |
817d52f8 JB |
449 | block_group->cached = BTRFS_CACHE_FINISHED; |
450 | spin_unlock(&block_group->lock); | |
0f9dd46c | 451 | |
54aa1f4d | 452 | err: |
e37c9e69 | 453 | btrfs_free_path(path); |
276e680d | 454 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 455 | |
11833d66 YZ |
456 | free_excluded_extents(extent_root, block_group); |
457 | ||
458 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 459 | out: |
11833d66 YZ |
460 | wake_up(&caching_ctl->wait); |
461 | ||
462 | put_caching_control(caching_ctl); | |
11dfe35a | 463 | btrfs_put_block_group(block_group); |
817d52f8 JB |
464 | } |
465 | ||
9d66e233 JB |
466 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
467 | struct btrfs_trans_handle *trans, | |
b8399dee | 468 | struct btrfs_root *root, |
9d66e233 | 469 | int load_cache_only) |
817d52f8 | 470 | { |
291c7d2f | 471 | DEFINE_WAIT(wait); |
11833d66 YZ |
472 | struct btrfs_fs_info *fs_info = cache->fs_info; |
473 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
474 | int ret = 0; |
475 | ||
291c7d2f JB |
476 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
477 | BUG_ON(!caching_ctl); | |
478 | ||
479 | INIT_LIST_HEAD(&caching_ctl->list); | |
480 | mutex_init(&caching_ctl->mutex); | |
481 | init_waitqueue_head(&caching_ctl->wait); | |
482 | caching_ctl->block_group = cache; | |
483 | caching_ctl->progress = cache->key.objectid; | |
484 | atomic_set(&caching_ctl->count, 1); | |
485 | caching_ctl->work.func = caching_thread; | |
486 | ||
487 | spin_lock(&cache->lock); | |
488 | /* | |
489 | * This should be a rare occasion, but this could happen I think in the | |
490 | * case where one thread starts to load the space cache info, and then | |
491 | * some other thread starts a transaction commit which tries to do an | |
492 | * allocation while the other thread is still loading the space cache | |
493 | * info. The previous loop should have kept us from choosing this block | |
494 | * group, but if we've moved to the state where we will wait on caching | |
495 | * block groups we need to first check if we're doing a fast load here, | |
496 | * so we can wait for it to finish, otherwise we could end up allocating | |
497 | * from a block group who's cache gets evicted for one reason or | |
498 | * another. | |
499 | */ | |
500 | while (cache->cached == BTRFS_CACHE_FAST) { | |
501 | struct btrfs_caching_control *ctl; | |
502 | ||
503 | ctl = cache->caching_ctl; | |
504 | atomic_inc(&ctl->count); | |
505 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
506 | spin_unlock(&cache->lock); | |
507 | ||
508 | schedule(); | |
509 | ||
510 | finish_wait(&ctl->wait, &wait); | |
511 | put_caching_control(ctl); | |
512 | spin_lock(&cache->lock); | |
513 | } | |
514 | ||
515 | if (cache->cached != BTRFS_CACHE_NO) { | |
516 | spin_unlock(&cache->lock); | |
517 | kfree(caching_ctl); | |
11833d66 | 518 | return 0; |
291c7d2f JB |
519 | } |
520 | WARN_ON(cache->caching_ctl); | |
521 | cache->caching_ctl = caching_ctl; | |
522 | cache->cached = BTRFS_CACHE_FAST; | |
523 | spin_unlock(&cache->lock); | |
11833d66 | 524 | |
9d66e233 JB |
525 | /* |
526 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
527 | * to have the normal tree locking. Also if we are currently trying to |
528 | * allocate blocks for the tree root we can't do the fast caching since | |
529 | * we likely hold important locks. | |
9d66e233 | 530 | */ |
f7039b1d | 531 | if (trans && (!trans->transaction->in_commit) && |
73bc1876 JB |
532 | (root && root != root->fs_info->tree_root) && |
533 | btrfs_test_opt(root, SPACE_CACHE)) { | |
9d66e233 JB |
534 | ret = load_free_space_cache(fs_info, cache); |
535 | ||
536 | spin_lock(&cache->lock); | |
537 | if (ret == 1) { | |
291c7d2f | 538 | cache->caching_ctl = NULL; |
9d66e233 JB |
539 | cache->cached = BTRFS_CACHE_FINISHED; |
540 | cache->last_byte_to_unpin = (u64)-1; | |
541 | } else { | |
291c7d2f JB |
542 | if (load_cache_only) { |
543 | cache->caching_ctl = NULL; | |
544 | cache->cached = BTRFS_CACHE_NO; | |
545 | } else { | |
546 | cache->cached = BTRFS_CACHE_STARTED; | |
547 | } | |
9d66e233 JB |
548 | } |
549 | spin_unlock(&cache->lock); | |
291c7d2f | 550 | wake_up(&caching_ctl->wait); |
3c14874a | 551 | if (ret == 1) { |
291c7d2f | 552 | put_caching_control(caching_ctl); |
3c14874a | 553 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 554 | return 0; |
3c14874a | 555 | } |
291c7d2f JB |
556 | } else { |
557 | /* | |
558 | * We are not going to do the fast caching, set cached to the | |
559 | * appropriate value and wakeup any waiters. | |
560 | */ | |
561 | spin_lock(&cache->lock); | |
562 | if (load_cache_only) { | |
563 | cache->caching_ctl = NULL; | |
564 | cache->cached = BTRFS_CACHE_NO; | |
565 | } else { | |
566 | cache->cached = BTRFS_CACHE_STARTED; | |
567 | } | |
568 | spin_unlock(&cache->lock); | |
569 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
570 | } |
571 | ||
291c7d2f JB |
572 | if (load_cache_only) { |
573 | put_caching_control(caching_ctl); | |
11833d66 | 574 | return 0; |
817d52f8 | 575 | } |
817d52f8 | 576 | |
11833d66 | 577 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 578 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
579 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
580 | up_write(&fs_info->extent_commit_sem); | |
581 | ||
11dfe35a | 582 | btrfs_get_block_group(cache); |
11833d66 | 583 | |
bab39bf9 | 584 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 585 | |
ef8bbdfe | 586 | return ret; |
e37c9e69 CM |
587 | } |
588 | ||
0f9dd46c JB |
589 | /* |
590 | * return the block group that starts at or after bytenr | |
591 | */ | |
d397712b CM |
592 | static struct btrfs_block_group_cache * |
593 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 594 | { |
0f9dd46c | 595 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 596 | |
0f9dd46c | 597 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 598 | |
0f9dd46c | 599 | return cache; |
0ef3e66b CM |
600 | } |
601 | ||
0f9dd46c | 602 | /* |
9f55684c | 603 | * return the block group that contains the given bytenr |
0f9dd46c | 604 | */ |
d397712b CM |
605 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
606 | struct btrfs_fs_info *info, | |
607 | u64 bytenr) | |
be744175 | 608 | { |
0f9dd46c | 609 | struct btrfs_block_group_cache *cache; |
be744175 | 610 | |
0f9dd46c | 611 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 612 | |
0f9dd46c | 613 | return cache; |
be744175 | 614 | } |
0b86a832 | 615 | |
0f9dd46c JB |
616 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
617 | u64 flags) | |
6324fbf3 | 618 | { |
0f9dd46c | 619 | struct list_head *head = &info->space_info; |
0f9dd46c | 620 | struct btrfs_space_info *found; |
4184ea7f | 621 | |
52ba6929 | 622 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 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 | |
d1270cd9 AJ |
2239 | /* |
2240 | * locked_ref is the head node, so we have to go one | |
2241 | * node back for any delayed ref updates | |
2242 | */ | |
2243 | ref = select_delayed_ref(locked_ref); | |
2244 | ||
2245 | if (ref && ref->seq && | |
2246 | btrfs_check_delayed_seq(delayed_refs, ref->seq)) { | |
2247 | /* | |
2248 | * there are still refs with lower seq numbers in the | |
2249 | * process of being added. Don't run this ref yet. | |
2250 | */ | |
2251 | list_del_init(&locked_ref->cluster); | |
2252 | mutex_unlock(&locked_ref->mutex); | |
2253 | locked_ref = NULL; | |
2254 | delayed_refs->num_heads_ready++; | |
2255 | spin_unlock(&delayed_refs->lock); | |
2256 | cond_resched(); | |
2257 | spin_lock(&delayed_refs->lock); | |
2258 | continue; | |
2259 | } | |
2260 | ||
56bec294 CM |
2261 | /* |
2262 | * record the must insert reserved flag before we | |
2263 | * drop the spin lock. | |
2264 | */ | |
2265 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2266 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2267 | |
5d4f98a2 YZ |
2268 | extent_op = locked_ref->extent_op; |
2269 | locked_ref->extent_op = NULL; | |
2270 | ||
56bec294 CM |
2271 | if (!ref) { |
2272 | /* All delayed refs have been processed, Go ahead | |
2273 | * and send the head node to run_one_delayed_ref, | |
2274 | * so that any accounting fixes can happen | |
2275 | */ | |
2276 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2277 | |
2278 | if (extent_op && must_insert_reserved) { | |
2279 | kfree(extent_op); | |
2280 | extent_op = NULL; | |
2281 | } | |
2282 | ||
2283 | if (extent_op) { | |
2284 | spin_unlock(&delayed_refs->lock); | |
2285 | ||
2286 | ret = run_delayed_extent_op(trans, root, | |
2287 | ref, extent_op); | |
2288 | BUG_ON(ret); | |
2289 | kfree(extent_op); | |
2290 | ||
203bf287 | 2291 | goto next; |
5d4f98a2 YZ |
2292 | } |
2293 | ||
c3e69d58 | 2294 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2295 | locked_ref = NULL; |
2296 | } | |
02217ed2 | 2297 | |
56bec294 CM |
2298 | ref->in_tree = 0; |
2299 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2300 | delayed_refs->num_entries--; | |
a168650c JS |
2301 | /* |
2302 | * we modified num_entries, but as we're currently running | |
2303 | * delayed refs, skip | |
2304 | * wake_up(&delayed_refs->seq_wait); | |
2305 | * here. | |
2306 | */ | |
56bec294 | 2307 | spin_unlock(&delayed_refs->lock); |
925baedd | 2308 | |
5d4f98a2 | 2309 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2310 | must_insert_reserved); |
2311 | BUG_ON(ret); | |
eb099670 | 2312 | |
5d4f98a2 YZ |
2313 | btrfs_put_delayed_ref(ref); |
2314 | kfree(extent_op); | |
c3e69d58 | 2315 | count++; |
203bf287 CM |
2316 | next: |
2317 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
2318 | 2 * 1024 * 1024, | |
2319 | btrfs_get_alloc_profile(root, 0), | |
2320 | CHUNK_ALLOC_NO_FORCE); | |
c3e69d58 CM |
2321 | cond_resched(); |
2322 | spin_lock(&delayed_refs->lock); | |
2323 | } | |
2324 | return count; | |
2325 | } | |
2326 | ||
a168650c JS |
2327 | |
2328 | static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs, | |
2329 | unsigned long num_refs) | |
2330 | { | |
2331 | struct list_head *first_seq = delayed_refs->seq_head.next; | |
2332 | ||
2333 | spin_unlock(&delayed_refs->lock); | |
2334 | pr_debug("waiting for more refs (num %ld, first %p)\n", | |
2335 | num_refs, first_seq); | |
2336 | wait_event(delayed_refs->seq_wait, | |
2337 | num_refs != delayed_refs->num_entries || | |
2338 | delayed_refs->seq_head.next != first_seq); | |
2339 | pr_debug("done waiting for more refs (num %ld, first %p)\n", | |
2340 | delayed_refs->num_entries, delayed_refs->seq_head.next); | |
2341 | spin_lock(&delayed_refs->lock); | |
2342 | } | |
2343 | ||
c3e69d58 CM |
2344 | /* |
2345 | * this starts processing the delayed reference count updates and | |
2346 | * extent insertions we have queued up so far. count can be | |
2347 | * 0, which means to process everything in the tree at the start | |
2348 | * of the run (but not newly added entries), or it can be some target | |
2349 | * number you'd like to process. | |
2350 | */ | |
2351 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2352 | struct btrfs_root *root, unsigned long count) | |
2353 | { | |
2354 | struct rb_node *node; | |
2355 | struct btrfs_delayed_ref_root *delayed_refs; | |
2356 | struct btrfs_delayed_ref_node *ref; | |
2357 | struct list_head cluster; | |
2358 | int ret; | |
a168650c | 2359 | u64 delayed_start; |
c3e69d58 CM |
2360 | int run_all = count == (unsigned long)-1; |
2361 | int run_most = 0; | |
a168650c JS |
2362 | unsigned long num_refs = 0; |
2363 | int consider_waiting; | |
c3e69d58 CM |
2364 | |
2365 | if (root == root->fs_info->extent_root) | |
2366 | root = root->fs_info->tree_root; | |
2367 | ||
203bf287 CM |
2368 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2369 | 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0), | |
2370 | CHUNK_ALLOC_NO_FORCE); | |
2371 | ||
c3e69d58 CM |
2372 | delayed_refs = &trans->transaction->delayed_refs; |
2373 | INIT_LIST_HEAD(&cluster); | |
2374 | again: | |
a168650c | 2375 | consider_waiting = 0; |
c3e69d58 CM |
2376 | spin_lock(&delayed_refs->lock); |
2377 | if (count == 0) { | |
2378 | count = delayed_refs->num_entries * 2; | |
2379 | run_most = 1; | |
2380 | } | |
2381 | while (1) { | |
2382 | if (!(run_all || run_most) && | |
2383 | delayed_refs->num_heads_ready < 64) | |
2384 | break; | |
eb099670 | 2385 | |
56bec294 | 2386 | /* |
c3e69d58 CM |
2387 | * go find something we can process in the rbtree. We start at |
2388 | * the beginning of the tree, and then build a cluster | |
2389 | * of refs to process starting at the first one we are able to | |
2390 | * lock | |
56bec294 | 2391 | */ |
a168650c | 2392 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2393 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2394 | delayed_refs->run_delayed_start); | |
2395 | if (ret) | |
56bec294 CM |
2396 | break; |
2397 | ||
a168650c JS |
2398 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2399 | if (consider_waiting == 0) { | |
2400 | /* | |
2401 | * btrfs_find_ref_cluster looped. let's do one | |
2402 | * more cycle. if we don't run any delayed ref | |
2403 | * during that cycle (because we can't because | |
2404 | * all of them are blocked) and if the number of | |
2405 | * refs doesn't change, we avoid busy waiting. | |
2406 | */ | |
2407 | consider_waiting = 1; | |
2408 | num_refs = delayed_refs->num_entries; | |
2409 | } else { | |
2410 | wait_for_more_refs(delayed_refs, num_refs); | |
2411 | /* | |
2412 | * after waiting, things have changed. we | |
2413 | * dropped the lock and someone else might have | |
2414 | * run some refs, built new clusters and so on. | |
2415 | * therefore, we restart staleness detection. | |
2416 | */ | |
2417 | consider_waiting = 0; | |
2418 | } | |
2419 | } | |
2420 | ||
c3e69d58 CM |
2421 | ret = run_clustered_refs(trans, root, &cluster); |
2422 | BUG_ON(ret < 0); | |
2423 | ||
2424 | count -= min_t(unsigned long, ret, count); | |
2425 | ||
2426 | if (count == 0) | |
2427 | break; | |
a168650c JS |
2428 | |
2429 | if (ret || delayed_refs->run_delayed_start == 0) { | |
2430 | /* refs were run, let's reset staleness detection */ | |
2431 | consider_waiting = 0; | |
2432 | } | |
eb099670 | 2433 | } |
c3e69d58 | 2434 | |
56bec294 | 2435 | if (run_all) { |
56bec294 | 2436 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2437 | if (!node) |
56bec294 | 2438 | goto out; |
c3e69d58 | 2439 | count = (unsigned long)-1; |
e9d0b13b | 2440 | |
56bec294 CM |
2441 | while (node) { |
2442 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2443 | rb_node); | |
2444 | if (btrfs_delayed_ref_is_head(ref)) { | |
2445 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2446 | |
56bec294 CM |
2447 | head = btrfs_delayed_node_to_head(ref); |
2448 | atomic_inc(&ref->refs); | |
2449 | ||
2450 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2451 | /* |
2452 | * Mutex was contended, block until it's | |
2453 | * released and try again | |
2454 | */ | |
56bec294 CM |
2455 | mutex_lock(&head->mutex); |
2456 | mutex_unlock(&head->mutex); | |
2457 | ||
2458 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2459 | cond_resched(); |
56bec294 CM |
2460 | goto again; |
2461 | } | |
2462 | node = rb_next(node); | |
2463 | } | |
2464 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2465 | schedule_timeout(1); |
2466 | goto again; | |
5f39d397 | 2467 | } |
54aa1f4d | 2468 | out: |
c3e69d58 | 2469 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2470 | return 0; |
2471 | } | |
2472 | ||
5d4f98a2 YZ |
2473 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2474 | struct btrfs_root *root, | |
2475 | u64 bytenr, u64 num_bytes, u64 flags, | |
2476 | int is_data) | |
2477 | { | |
2478 | struct btrfs_delayed_extent_op *extent_op; | |
2479 | int ret; | |
2480 | ||
2481 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2482 | if (!extent_op) | |
2483 | return -ENOMEM; | |
2484 | ||
2485 | extent_op->flags_to_set = flags; | |
2486 | extent_op->update_flags = 1; | |
2487 | extent_op->update_key = 0; | |
2488 | extent_op->is_data = is_data ? 1 : 0; | |
2489 | ||
66d7e7f0 AJ |
2490 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2491 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2492 | if (ret) |
2493 | kfree(extent_op); | |
2494 | return ret; | |
2495 | } | |
2496 | ||
2497 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2498 | struct btrfs_root *root, | |
2499 | struct btrfs_path *path, | |
2500 | u64 objectid, u64 offset, u64 bytenr) | |
2501 | { | |
2502 | struct btrfs_delayed_ref_head *head; | |
2503 | struct btrfs_delayed_ref_node *ref; | |
2504 | struct btrfs_delayed_data_ref *data_ref; | |
2505 | struct btrfs_delayed_ref_root *delayed_refs; | |
2506 | struct rb_node *node; | |
2507 | int ret = 0; | |
2508 | ||
2509 | ret = -ENOENT; | |
2510 | delayed_refs = &trans->transaction->delayed_refs; | |
2511 | spin_lock(&delayed_refs->lock); | |
2512 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2513 | if (!head) | |
2514 | goto out; | |
2515 | ||
2516 | if (!mutex_trylock(&head->mutex)) { | |
2517 | atomic_inc(&head->node.refs); | |
2518 | spin_unlock(&delayed_refs->lock); | |
2519 | ||
b3b4aa74 | 2520 | btrfs_release_path(path); |
5d4f98a2 | 2521 | |
8cc33e5c DS |
2522 | /* |
2523 | * Mutex was contended, block until it's released and let | |
2524 | * caller try again | |
2525 | */ | |
5d4f98a2 YZ |
2526 | mutex_lock(&head->mutex); |
2527 | mutex_unlock(&head->mutex); | |
2528 | btrfs_put_delayed_ref(&head->node); | |
2529 | return -EAGAIN; | |
2530 | } | |
2531 | ||
2532 | node = rb_prev(&head->node.rb_node); | |
2533 | if (!node) | |
2534 | goto out_unlock; | |
2535 | ||
2536 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2537 | ||
2538 | if (ref->bytenr != bytenr) | |
2539 | goto out_unlock; | |
2540 | ||
2541 | ret = 1; | |
2542 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2543 | goto out_unlock; | |
2544 | ||
2545 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2546 | ||
2547 | node = rb_prev(node); | |
2548 | if (node) { | |
2549 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2550 | if (ref->bytenr == bytenr) | |
2551 | goto out_unlock; | |
2552 | } | |
2553 | ||
2554 | if (data_ref->root != root->root_key.objectid || | |
2555 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2556 | goto out_unlock; | |
2557 | ||
2558 | ret = 0; | |
2559 | out_unlock: | |
2560 | mutex_unlock(&head->mutex); | |
2561 | out: | |
2562 | spin_unlock(&delayed_refs->lock); | |
2563 | return ret; | |
2564 | } | |
2565 | ||
2566 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2567 | struct btrfs_root *root, | |
2568 | struct btrfs_path *path, | |
2569 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2570 | { |
2571 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2572 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2573 | struct btrfs_extent_data_ref *ref; |
2574 | struct btrfs_extent_inline_ref *iref; | |
2575 | struct btrfs_extent_item *ei; | |
f321e491 | 2576 | struct btrfs_key key; |
5d4f98a2 | 2577 | u32 item_size; |
be20aa9d | 2578 | int ret; |
925baedd | 2579 | |
be20aa9d | 2580 | key.objectid = bytenr; |
31840ae1 | 2581 | key.offset = (u64)-1; |
f321e491 | 2582 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2583 | |
be20aa9d CM |
2584 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2585 | if (ret < 0) | |
2586 | goto out; | |
2587 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2588 | |
2589 | ret = -ENOENT; | |
2590 | if (path->slots[0] == 0) | |
31840ae1 | 2591 | goto out; |
be20aa9d | 2592 | |
31840ae1 | 2593 | path->slots[0]--; |
f321e491 | 2594 | leaf = path->nodes[0]; |
5d4f98a2 | 2595 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2596 | |
5d4f98a2 | 2597 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2598 | goto out; |
f321e491 | 2599 | |
5d4f98a2 YZ |
2600 | ret = 1; |
2601 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2602 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2603 | if (item_size < sizeof(*ei)) { | |
2604 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2605 | goto out; | |
2606 | } | |
2607 | #endif | |
2608 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2609 | |
5d4f98a2 YZ |
2610 | if (item_size != sizeof(*ei) + |
2611 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2612 | goto out; | |
be20aa9d | 2613 | |
5d4f98a2 YZ |
2614 | if (btrfs_extent_generation(leaf, ei) <= |
2615 | btrfs_root_last_snapshot(&root->root_item)) | |
2616 | goto out; | |
2617 | ||
2618 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2619 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2620 | BTRFS_EXTENT_DATA_REF_KEY) | |
2621 | goto out; | |
2622 | ||
2623 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2624 | if (btrfs_extent_refs(leaf, ei) != | |
2625 | btrfs_extent_data_ref_count(leaf, ref) || | |
2626 | btrfs_extent_data_ref_root(leaf, ref) != | |
2627 | root->root_key.objectid || | |
2628 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2629 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2630 | goto out; | |
2631 | ||
2632 | ret = 0; | |
2633 | out: | |
2634 | return ret; | |
2635 | } | |
2636 | ||
2637 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2638 | struct btrfs_root *root, | |
2639 | u64 objectid, u64 offset, u64 bytenr) | |
2640 | { | |
2641 | struct btrfs_path *path; | |
2642 | int ret; | |
2643 | int ret2; | |
2644 | ||
2645 | path = btrfs_alloc_path(); | |
2646 | if (!path) | |
2647 | return -ENOENT; | |
2648 | ||
2649 | do { | |
2650 | ret = check_committed_ref(trans, root, path, objectid, | |
2651 | offset, bytenr); | |
2652 | if (ret && ret != -ENOENT) | |
f321e491 | 2653 | goto out; |
80ff3856 | 2654 | |
5d4f98a2 YZ |
2655 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2656 | offset, bytenr); | |
2657 | } while (ret2 == -EAGAIN); | |
2658 | ||
2659 | if (ret2 && ret2 != -ENOENT) { | |
2660 | ret = ret2; | |
2661 | goto out; | |
f321e491 | 2662 | } |
5d4f98a2 YZ |
2663 | |
2664 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2665 | ret = 0; | |
be20aa9d | 2666 | out: |
80ff3856 | 2667 | btrfs_free_path(path); |
f0486c68 YZ |
2668 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2669 | WARN_ON(ret > 0); | |
f321e491 | 2670 | return ret; |
be20aa9d | 2671 | } |
c5739bba | 2672 | |
5d4f98a2 | 2673 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2674 | struct btrfs_root *root, |
5d4f98a2 | 2675 | struct extent_buffer *buf, |
66d7e7f0 | 2676 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2677 | { |
2678 | u64 bytenr; | |
5d4f98a2 YZ |
2679 | u64 num_bytes; |
2680 | u64 parent; | |
31840ae1 | 2681 | u64 ref_root; |
31840ae1 | 2682 | u32 nritems; |
31840ae1 ZY |
2683 | struct btrfs_key key; |
2684 | struct btrfs_file_extent_item *fi; | |
2685 | int i; | |
2686 | int level; | |
2687 | int ret = 0; | |
31840ae1 | 2688 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2689 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2690 | |
2691 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2692 | nritems = btrfs_header_nritems(buf); |
2693 | level = btrfs_header_level(buf); | |
2694 | ||
5d4f98a2 YZ |
2695 | if (!root->ref_cows && level == 0) |
2696 | return 0; | |
31840ae1 | 2697 | |
5d4f98a2 YZ |
2698 | if (inc) |
2699 | process_func = btrfs_inc_extent_ref; | |
2700 | else | |
2701 | process_func = btrfs_free_extent; | |
31840ae1 | 2702 | |
5d4f98a2 YZ |
2703 | if (full_backref) |
2704 | parent = buf->start; | |
2705 | else | |
2706 | parent = 0; | |
2707 | ||
2708 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2709 | if (level == 0) { |
5d4f98a2 | 2710 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2711 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2712 | continue; | |
5d4f98a2 | 2713 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2714 | struct btrfs_file_extent_item); |
2715 | if (btrfs_file_extent_type(buf, fi) == | |
2716 | BTRFS_FILE_EXTENT_INLINE) | |
2717 | continue; | |
2718 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2719 | if (bytenr == 0) | |
2720 | continue; | |
5d4f98a2 YZ |
2721 | |
2722 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2723 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2724 | ret = process_func(trans, root, bytenr, num_bytes, | |
2725 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2726 | key.offset, for_cow); |
31840ae1 ZY |
2727 | if (ret) |
2728 | goto fail; | |
2729 | } else { | |
5d4f98a2 YZ |
2730 | bytenr = btrfs_node_blockptr(buf, i); |
2731 | num_bytes = btrfs_level_size(root, level - 1); | |
2732 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2733 | parent, ref_root, level - 1, 0, |
2734 | for_cow); | |
31840ae1 ZY |
2735 | if (ret) |
2736 | goto fail; | |
2737 | } | |
2738 | } | |
2739 | return 0; | |
2740 | fail: | |
5d4f98a2 YZ |
2741 | BUG(); |
2742 | return ret; | |
2743 | } | |
2744 | ||
2745 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2746 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2747 | { |
66d7e7f0 | 2748 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2749 | } |
2750 | ||
2751 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2752 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2753 | { |
66d7e7f0 | 2754 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2755 | } |
2756 | ||
9078a3e1 CM |
2757 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2758 | struct btrfs_root *root, | |
2759 | struct btrfs_path *path, | |
2760 | struct btrfs_block_group_cache *cache) | |
2761 | { | |
2762 | int ret; | |
9078a3e1 | 2763 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2764 | unsigned long bi; |
2765 | struct extent_buffer *leaf; | |
9078a3e1 | 2766 | |
9078a3e1 | 2767 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2768 | if (ret < 0) |
2769 | goto fail; | |
9078a3e1 | 2770 | BUG_ON(ret); |
5f39d397 CM |
2771 | |
2772 | leaf = path->nodes[0]; | |
2773 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2774 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2775 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2776 | btrfs_release_path(path); |
54aa1f4d | 2777 | fail: |
9078a3e1 CM |
2778 | if (ret) |
2779 | return ret; | |
9078a3e1 CM |
2780 | return 0; |
2781 | ||
2782 | } | |
2783 | ||
4a8c9a62 YZ |
2784 | static struct btrfs_block_group_cache * |
2785 | next_block_group(struct btrfs_root *root, | |
2786 | struct btrfs_block_group_cache *cache) | |
2787 | { | |
2788 | struct rb_node *node; | |
2789 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2790 | node = rb_next(&cache->cache_node); | |
2791 | btrfs_put_block_group(cache); | |
2792 | if (node) { | |
2793 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2794 | cache_node); | |
11dfe35a | 2795 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2796 | } else |
2797 | cache = NULL; | |
2798 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2799 | return cache; | |
2800 | } | |
2801 | ||
0af3d00b JB |
2802 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2803 | struct btrfs_trans_handle *trans, | |
2804 | struct btrfs_path *path) | |
2805 | { | |
2806 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2807 | struct inode *inode = NULL; | |
2808 | u64 alloc_hint = 0; | |
2b20982e | 2809 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2810 | int num_pages = 0; |
2811 | int retries = 0; | |
2812 | int ret = 0; | |
2813 | ||
2814 | /* | |
2815 | * If this block group is smaller than 100 megs don't bother caching the | |
2816 | * block group. | |
2817 | */ | |
2818 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2819 | spin_lock(&block_group->lock); | |
2820 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2821 | spin_unlock(&block_group->lock); | |
2822 | return 0; | |
2823 | } | |
2824 | ||
2825 | again: | |
2826 | inode = lookup_free_space_inode(root, block_group, path); | |
2827 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2828 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2829 | btrfs_release_path(path); |
0af3d00b JB |
2830 | goto out; |
2831 | } | |
2832 | ||
2833 | if (IS_ERR(inode)) { | |
2834 | BUG_ON(retries); | |
2835 | retries++; | |
2836 | ||
2837 | if (block_group->ro) | |
2838 | goto out_free; | |
2839 | ||
2840 | ret = create_free_space_inode(root, trans, block_group, path); | |
2841 | if (ret) | |
2842 | goto out_free; | |
2843 | goto again; | |
2844 | } | |
2845 | ||
5b0e95bf JB |
2846 | /* We've already setup this transaction, go ahead and exit */ |
2847 | if (block_group->cache_generation == trans->transid && | |
2848 | i_size_read(inode)) { | |
2849 | dcs = BTRFS_DC_SETUP; | |
2850 | goto out_put; | |
2851 | } | |
2852 | ||
0af3d00b JB |
2853 | /* |
2854 | * We want to set the generation to 0, that way if anything goes wrong | |
2855 | * from here on out we know not to trust this cache when we load up next | |
2856 | * time. | |
2857 | */ | |
2858 | BTRFS_I(inode)->generation = 0; | |
2859 | ret = btrfs_update_inode(trans, root, inode); | |
2860 | WARN_ON(ret); | |
2861 | ||
2862 | if (i_size_read(inode) > 0) { | |
2863 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2864 | inode); | |
2865 | if (ret) | |
2866 | goto out_put; | |
2867 | } | |
2868 | ||
2869 | spin_lock(&block_group->lock); | |
2870 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2871 | /* We're not cached, don't bother trying to write stuff out */ |
2872 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2873 | spin_unlock(&block_group->lock); |
2874 | goto out_put; | |
2875 | } | |
2876 | spin_unlock(&block_group->lock); | |
2877 | ||
2878 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2879 | if (!num_pages) | |
2880 | num_pages = 1; | |
2881 | ||
2882 | /* | |
2883 | * Just to make absolutely sure we have enough space, we're going to | |
2884 | * preallocate 12 pages worth of space for each block group. In | |
2885 | * practice we ought to use at most 8, but we need extra space so we can | |
2886 | * add our header and have a terminator between the extents and the | |
2887 | * bitmaps. | |
2888 | */ | |
2889 | num_pages *= 16; | |
2890 | num_pages *= PAGE_CACHE_SIZE; | |
2891 | ||
2892 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2893 | if (ret) | |
2894 | goto out_put; | |
2895 | ||
2896 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2897 | num_pages, num_pages, | |
2898 | &alloc_hint); | |
2b20982e JB |
2899 | if (!ret) |
2900 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 2901 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 2902 | |
0af3d00b JB |
2903 | out_put: |
2904 | iput(inode); | |
2905 | out_free: | |
b3b4aa74 | 2906 | btrfs_release_path(path); |
0af3d00b JB |
2907 | out: |
2908 | spin_lock(&block_group->lock); | |
e65cbb94 | 2909 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 2910 | block_group->cache_generation = trans->transid; |
2b20982e | 2911 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2912 | spin_unlock(&block_group->lock); |
2913 | ||
2914 | return ret; | |
2915 | } | |
2916 | ||
96b5179d CM |
2917 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2918 | struct btrfs_root *root) | |
9078a3e1 | 2919 | { |
4a8c9a62 | 2920 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2921 | int err = 0; |
9078a3e1 | 2922 | struct btrfs_path *path; |
96b5179d | 2923 | u64 last = 0; |
9078a3e1 CM |
2924 | |
2925 | path = btrfs_alloc_path(); | |
2926 | if (!path) | |
2927 | return -ENOMEM; | |
2928 | ||
0af3d00b JB |
2929 | again: |
2930 | while (1) { | |
2931 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2932 | while (cache) { | |
2933 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2934 | break; | |
2935 | cache = next_block_group(root, cache); | |
2936 | } | |
2937 | if (!cache) { | |
2938 | if (last == 0) | |
2939 | break; | |
2940 | last = 0; | |
2941 | continue; | |
2942 | } | |
2943 | err = cache_save_setup(cache, trans, path); | |
2944 | last = cache->key.objectid + cache->key.offset; | |
2945 | btrfs_put_block_group(cache); | |
2946 | } | |
2947 | ||
d397712b | 2948 | while (1) { |
4a8c9a62 YZ |
2949 | if (last == 0) { |
2950 | err = btrfs_run_delayed_refs(trans, root, | |
2951 | (unsigned long)-1); | |
2952 | BUG_ON(err); | |
0f9dd46c | 2953 | } |
54aa1f4d | 2954 | |
4a8c9a62 YZ |
2955 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2956 | while (cache) { | |
0af3d00b JB |
2957 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2958 | btrfs_put_block_group(cache); | |
2959 | goto again; | |
2960 | } | |
2961 | ||
4a8c9a62 YZ |
2962 | if (cache->dirty) |
2963 | break; | |
2964 | cache = next_block_group(root, cache); | |
2965 | } | |
2966 | if (!cache) { | |
2967 | if (last == 0) | |
2968 | break; | |
2969 | last = 0; | |
2970 | continue; | |
2971 | } | |
0f9dd46c | 2972 | |
0cb59c99 JB |
2973 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2974 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2975 | cache->dirty = 0; |
4a8c9a62 | 2976 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2977 | |
4a8c9a62 YZ |
2978 | err = write_one_cache_group(trans, root, path, cache); |
2979 | BUG_ON(err); | |
2980 | btrfs_put_block_group(cache); | |
9078a3e1 | 2981 | } |
4a8c9a62 | 2982 | |
0cb59c99 JB |
2983 | while (1) { |
2984 | /* | |
2985 | * I don't think this is needed since we're just marking our | |
2986 | * preallocated extent as written, but just in case it can't | |
2987 | * hurt. | |
2988 | */ | |
2989 | if (last == 0) { | |
2990 | err = btrfs_run_delayed_refs(trans, root, | |
2991 | (unsigned long)-1); | |
2992 | BUG_ON(err); | |
2993 | } | |
2994 | ||
2995 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2996 | while (cache) { | |
2997 | /* | |
2998 | * Really this shouldn't happen, but it could if we | |
2999 | * couldn't write the entire preallocated extent and | |
3000 | * splitting the extent resulted in a new block. | |
3001 | */ | |
3002 | if (cache->dirty) { | |
3003 | btrfs_put_block_group(cache); | |
3004 | goto again; | |
3005 | } | |
3006 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3007 | break; | |
3008 | cache = next_block_group(root, cache); | |
3009 | } | |
3010 | if (!cache) { | |
3011 | if (last == 0) | |
3012 | break; | |
3013 | last = 0; | |
3014 | continue; | |
3015 | } | |
3016 | ||
3017 | btrfs_write_out_cache(root, trans, cache, path); | |
3018 | ||
3019 | /* | |
3020 | * If we didn't have an error then the cache state is still | |
3021 | * NEED_WRITE, so we can set it to WRITTEN. | |
3022 | */ | |
3023 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3024 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
3025 | last = cache->key.objectid + cache->key.offset; | |
3026 | btrfs_put_block_group(cache); | |
3027 | } | |
3028 | ||
9078a3e1 | 3029 | btrfs_free_path(path); |
4a8c9a62 | 3030 | return 0; |
9078a3e1 CM |
3031 | } |
3032 | ||
d2fb3437 YZ |
3033 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3034 | { | |
3035 | struct btrfs_block_group_cache *block_group; | |
3036 | int readonly = 0; | |
3037 | ||
3038 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3039 | if (!block_group || block_group->ro) | |
3040 | readonly = 1; | |
3041 | if (block_group) | |
fa9c0d79 | 3042 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3043 | return readonly; |
3044 | } | |
3045 | ||
593060d7 CM |
3046 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3047 | u64 total_bytes, u64 bytes_used, | |
3048 | struct btrfs_space_info **space_info) | |
3049 | { | |
3050 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3051 | int i; |
3052 | int factor; | |
3053 | ||
3054 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3055 | BTRFS_BLOCK_GROUP_RAID10)) | |
3056 | factor = 2; | |
3057 | else | |
3058 | factor = 1; | |
593060d7 CM |
3059 | |
3060 | found = __find_space_info(info, flags); | |
3061 | if (found) { | |
25179201 | 3062 | spin_lock(&found->lock); |
593060d7 | 3063 | found->total_bytes += total_bytes; |
89a55897 | 3064 | found->disk_total += total_bytes * factor; |
593060d7 | 3065 | found->bytes_used += bytes_used; |
b742bb82 | 3066 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3067 | found->full = 0; |
25179201 | 3068 | spin_unlock(&found->lock); |
593060d7 CM |
3069 | *space_info = found; |
3070 | return 0; | |
3071 | } | |
c146afad | 3072 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3073 | if (!found) |
3074 | return -ENOMEM; | |
3075 | ||
b742bb82 YZ |
3076 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3077 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3078 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3079 | spin_lock_init(&found->lock); |
52ba6929 | 3080 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3081 | found->total_bytes = total_bytes; |
89a55897 | 3082 | found->disk_total = total_bytes * factor; |
593060d7 | 3083 | found->bytes_used = bytes_used; |
b742bb82 | 3084 | found->disk_used = bytes_used * factor; |
593060d7 | 3085 | found->bytes_pinned = 0; |
e8569813 | 3086 | found->bytes_reserved = 0; |
c146afad | 3087 | found->bytes_readonly = 0; |
f0486c68 | 3088 | found->bytes_may_use = 0; |
593060d7 | 3089 | found->full = 0; |
0e4f8f88 | 3090 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3091 | found->chunk_alloc = 0; |
fdb5effd JB |
3092 | found->flush = 0; |
3093 | init_waitqueue_head(&found->wait); | |
593060d7 | 3094 | *space_info = found; |
4184ea7f | 3095 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
3096 | return 0; |
3097 | } | |
3098 | ||
8790d502 CM |
3099 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3100 | { | |
899c81ea ID |
3101 | u64 extra_flags = chunk_to_extended(flags) & |
3102 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 ID |
3103 | |
3104 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3105 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3106 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3107 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3108 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3109 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3110 | } |
593060d7 | 3111 | |
a46d11a8 ID |
3112 | /* |
3113 | * @flags: available profiles in extended format (see ctree.h) | |
3114 | * | |
e4d8ec0f ID |
3115 | * Returns reduced profile in chunk format. If profile changing is in |
3116 | * progress (either running or paused) picks the target profile (if it's | |
3117 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3118 | */ |
2b82032c | 3119 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3120 | { |
cd02dca5 CM |
3121 | /* |
3122 | * we add in the count of missing devices because we want | |
3123 | * to make sure that any RAID levels on a degraded FS | |
3124 | * continue to be honored. | |
3125 | */ | |
3126 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3127 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d | 3128 | |
e4d8ec0f ID |
3129 | /* pick restriper's target profile if it's available */ |
3130 | spin_lock(&root->fs_info->balance_lock); | |
3131 | if (root->fs_info->balance_ctl) { | |
3132 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3133 | u64 tgt = 0; | |
3134 | ||
3135 | if ((flags & BTRFS_BLOCK_GROUP_DATA) && | |
3136 | (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3137 | (flags & bctl->data.target)) { | |
3138 | tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3139 | } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3140 | (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3141 | (flags & bctl->sys.target)) { | |
3142 | tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3143 | } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) && | |
3144 | (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3145 | (flags & bctl->meta.target)) { | |
3146 | tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3147 | } | |
3148 | ||
3149 | if (tgt) { | |
3150 | spin_unlock(&root->fs_info->balance_lock); | |
3151 | flags = tgt; | |
3152 | goto out; | |
3153 | } | |
3154 | } | |
3155 | spin_unlock(&root->fs_info->balance_lock); | |
3156 | ||
a061fc8d CM |
3157 | if (num_devices == 1) |
3158 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3159 | if (num_devices < 4) | |
3160 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3161 | ||
ec44a35c CM |
3162 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3163 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3164 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3165 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3166 | } |
ec44a35c CM |
3167 | |
3168 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3169 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3170 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3171 | } |
ec44a35c CM |
3172 | |
3173 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3174 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3175 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3176 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3177 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3178 | } |
3179 | ||
e4d8ec0f | 3180 | out: |
899c81ea | 3181 | return extended_to_chunk(flags); |
ec44a35c CM |
3182 | } |
3183 | ||
b742bb82 | 3184 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3185 | { |
b742bb82 | 3186 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3187 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3188 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3189 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3190 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3191 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3192 | ||
b742bb82 | 3193 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3194 | } |
3195 | ||
6d07bcec | 3196 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3197 | { |
b742bb82 | 3198 | u64 flags; |
9ed74f2d | 3199 | |
b742bb82 YZ |
3200 | if (data) |
3201 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3202 | else if (root == root->fs_info->chunk_root) | |
3203 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3204 | else |
b742bb82 | 3205 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3206 | |
b742bb82 | 3207 | return get_alloc_profile(root, flags); |
6a63209f | 3208 | } |
9ed74f2d | 3209 | |
6a63209f JB |
3210 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3211 | { | |
6a63209f | 3212 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3213 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3214 | } |
3215 | ||
6a63209f | 3216 | /* |
6a63209f JB |
3217 | * This will check the space that the inode allocates from to make sure we have |
3218 | * enough space for bytes. | |
6a63209f | 3219 | */ |
0ca1f7ce | 3220 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3221 | { |
6a63209f | 3222 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3223 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3224 | u64 used; |
0af3d00b | 3225 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3226 | |
6a63209f JB |
3227 | /* make sure bytes are sectorsize aligned */ |
3228 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3229 | |
82d5902d LZ |
3230 | if (root == root->fs_info->tree_root || |
3231 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3232 | alloc_chunk = 0; |
3233 | committed = 1; | |
3234 | } | |
3235 | ||
6a63209f | 3236 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3237 | if (!data_sinfo) |
3238 | goto alloc; | |
9ed74f2d | 3239 | |
6a63209f JB |
3240 | again: |
3241 | /* make sure we have enough space to handle the data first */ | |
3242 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3243 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3244 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3245 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3246 | |
3247 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3248 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3249 | |
6a63209f JB |
3250 | /* |
3251 | * if we don't have enough free bytes in this space then we need | |
3252 | * to alloc a new chunk. | |
3253 | */ | |
0af3d00b | 3254 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3255 | u64 alloc_target; |
9ed74f2d | 3256 | |
0e4f8f88 | 3257 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3258 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3259 | alloc: |
6a63209f | 3260 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3261 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3262 | if (IS_ERR(trans)) |
3263 | return PTR_ERR(trans); | |
9ed74f2d | 3264 | |
6a63209f JB |
3265 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3266 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3267 | alloc_target, |
3268 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3269 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3270 | if (ret < 0) { |
3271 | if (ret != -ENOSPC) | |
3272 | return ret; | |
3273 | else | |
3274 | goto commit_trans; | |
3275 | } | |
9ed74f2d | 3276 | |
33b4d47f CM |
3277 | if (!data_sinfo) { |
3278 | btrfs_set_inode_space_info(root, inode); | |
3279 | data_sinfo = BTRFS_I(inode)->space_info; | |
3280 | } | |
6a63209f JB |
3281 | goto again; |
3282 | } | |
f2bb8f5c JB |
3283 | |
3284 | /* | |
3285 | * If we have less pinned bytes than we want to allocate then | |
3286 | * don't bother committing the transaction, it won't help us. | |
3287 | */ | |
3288 | if (data_sinfo->bytes_pinned < bytes) | |
3289 | committed = 1; | |
6a63209f | 3290 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3291 | |
4e06bdd6 | 3292 | /* commit the current transaction and try again */ |
d52a5b5f | 3293 | commit_trans: |
a4abeea4 JB |
3294 | if (!committed && |
3295 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3296 | committed = 1; |
7a7eaa40 | 3297 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3298 | if (IS_ERR(trans)) |
3299 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3300 | ret = btrfs_commit_transaction(trans, root); |
3301 | if (ret) | |
3302 | return ret; | |
3303 | goto again; | |
3304 | } | |
9ed74f2d | 3305 | |
6a63209f JB |
3306 | return -ENOSPC; |
3307 | } | |
3308 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3309 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
e77266e4 CM |
3310 | (u64)(unsigned long)data_sinfo, |
3311 | bytes, 1); | |
6a63209f | 3312 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3313 | |
9ed74f2d | 3314 | return 0; |
9ed74f2d | 3315 | } |
6a63209f | 3316 | |
6a63209f | 3317 | /* |
fb25e914 | 3318 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3319 | */ |
0ca1f7ce | 3320 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3321 | { |
0ca1f7ce | 3322 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3323 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3324 | |
6a63209f JB |
3325 | /* make sure bytes are sectorsize aligned */ |
3326 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3327 | |
6a63209f JB |
3328 | data_sinfo = BTRFS_I(inode)->space_info; |
3329 | spin_lock(&data_sinfo->lock); | |
3330 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3331 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
e77266e4 CM |
3332 | (u64)(unsigned long)data_sinfo, |
3333 | bytes, 0); | |
6a63209f | 3334 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3335 | } |
3336 | ||
97e728d4 | 3337 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3338 | { |
97e728d4 JB |
3339 | struct list_head *head = &info->space_info; |
3340 | struct btrfs_space_info *found; | |
e3ccfa98 | 3341 | |
97e728d4 JB |
3342 | rcu_read_lock(); |
3343 | list_for_each_entry_rcu(found, head, list) { | |
3344 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3345 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3346 | } |
97e728d4 | 3347 | rcu_read_unlock(); |
e3ccfa98 JB |
3348 | } |
3349 | ||
e5bc2458 | 3350 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3351 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3352 | int force) | |
32c00aff | 3353 | { |
fb25e914 | 3354 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3355 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3356 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3357 | u64 thresh; |
e3ccfa98 | 3358 | |
0e4f8f88 CM |
3359 | if (force == CHUNK_ALLOC_FORCE) |
3360 | return 1; | |
3361 | ||
fb25e914 JB |
3362 | /* |
3363 | * We need to take into account the global rsv because for all intents | |
3364 | * and purposes it's used space. Don't worry about locking the | |
3365 | * global_rsv, it doesn't change except when the transaction commits. | |
3366 | */ | |
3367 | num_allocated += global_rsv->size; | |
3368 | ||
0e4f8f88 CM |
3369 | /* |
3370 | * in limited mode, we want to have some free space up to | |
3371 | * about 1% of the FS size. | |
3372 | */ | |
3373 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3374 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3375 | thresh = max_t(u64, 64 * 1024 * 1024, |
3376 | div_factor_fine(thresh, 1)); | |
3377 | ||
3378 | if (num_bytes - num_allocated < thresh) | |
3379 | return 1; | |
3380 | } | |
6c41761f | 3381 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 | 3382 | |
cf1d72c9 CM |
3383 | /* 256MB or 2% of the FS */ |
3384 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2)); | |
96bdc7dc CM |
3385 | /* system chunks need a much small threshold */ |
3386 | if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3387 | thresh = 32 * 1024 * 1024; | |
e5bc2458 | 3388 | |
cf1d72c9 | 3389 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8)) |
14ed0ca6 | 3390 | return 0; |
424499db | 3391 | return 1; |
32c00aff JB |
3392 | } |
3393 | ||
6324fbf3 CM |
3394 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3395 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3396 | u64 flags, int force) |
9ed74f2d | 3397 | { |
6324fbf3 | 3398 | struct btrfs_space_info *space_info; |
97e728d4 | 3399 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3400 | int wait_for_alloc = 0; |
9ed74f2d | 3401 | int ret = 0; |
9ed74f2d | 3402 | |
6324fbf3 | 3403 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3404 | if (!space_info) { |
3405 | ret = update_space_info(extent_root->fs_info, flags, | |
3406 | 0, 0, &space_info); | |
3407 | BUG_ON(ret); | |
9ed74f2d | 3408 | } |
6324fbf3 | 3409 | BUG_ON(!space_info); |
9ed74f2d | 3410 | |
6d74119f | 3411 | again: |
25179201 | 3412 | spin_lock(&space_info->lock); |
9e622d6b | 3413 | if (force < space_info->force_alloc) |
0e4f8f88 | 3414 | force = space_info->force_alloc; |
25179201 JB |
3415 | if (space_info->full) { |
3416 | spin_unlock(&space_info->lock); | |
6d74119f | 3417 | return 0; |
9ed74f2d JB |
3418 | } |
3419 | ||
0e4f8f88 | 3420 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3421 | spin_unlock(&space_info->lock); |
6d74119f JB |
3422 | return 0; |
3423 | } else if (space_info->chunk_alloc) { | |
3424 | wait_for_alloc = 1; | |
3425 | } else { | |
3426 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3427 | } |
0e4f8f88 | 3428 | |
25179201 | 3429 | spin_unlock(&space_info->lock); |
9ed74f2d | 3430 | |
6d74119f JB |
3431 | mutex_lock(&fs_info->chunk_mutex); |
3432 | ||
3433 | /* | |
3434 | * The chunk_mutex is held throughout the entirety of a chunk | |
3435 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3436 | * other guy is done and we need to recheck and see if we should | |
3437 | * allocate. | |
3438 | */ | |
3439 | if (wait_for_alloc) { | |
3440 | mutex_unlock(&fs_info->chunk_mutex); | |
3441 | wait_for_alloc = 0; | |
3442 | goto again; | |
3443 | } | |
3444 | ||
67377734 JB |
3445 | /* |
3446 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3447 | * allocating mixed chunks instead of individual chunks. | |
3448 | */ | |
3449 | if (btrfs_mixed_space_info(space_info)) | |
3450 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3451 | ||
97e728d4 JB |
3452 | /* |
3453 | * if we're doing a data chunk, go ahead and make sure that | |
3454 | * we keep a reasonable number of metadata chunks allocated in the | |
3455 | * FS as well. | |
3456 | */ | |
9ed74f2d | 3457 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3458 | fs_info->data_chunk_allocations++; |
3459 | if (!(fs_info->data_chunk_allocations % | |
3460 | fs_info->metadata_ratio)) | |
3461 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3462 | } |
3463 | ||
2b82032c | 3464 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3465 | if (ret < 0 && ret != -ENOSPC) |
3466 | goto out; | |
3467 | ||
9ed74f2d | 3468 | spin_lock(&space_info->lock); |
9ed74f2d | 3469 | if (ret) |
6324fbf3 | 3470 | space_info->full = 1; |
424499db YZ |
3471 | else |
3472 | ret = 1; | |
6d74119f | 3473 | |
0e4f8f88 | 3474 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3475 | space_info->chunk_alloc = 0; |
9ed74f2d | 3476 | spin_unlock(&space_info->lock); |
92b8e897 | 3477 | out: |
c146afad | 3478 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3479 | return ret; |
6324fbf3 | 3480 | } |
9ed74f2d | 3481 | |
9ed74f2d | 3482 | /* |
5da9d01b | 3483 | * shrink metadata reservation for delalloc |
9ed74f2d | 3484 | */ |
663350ac | 3485 | static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, |
f104d044 | 3486 | bool wait_ordered) |
5da9d01b | 3487 | { |
0ca1f7ce | 3488 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3489 | struct btrfs_space_info *space_info; |
663350ac | 3490 | struct btrfs_trans_handle *trans; |
5da9d01b YZ |
3491 | u64 reserved; |
3492 | u64 max_reclaim; | |
3493 | u64 reclaimed = 0; | |
b1953bce | 3494 | long time_left; |
877da174 | 3495 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3496 | int loops = 0; |
36e39c40 | 3497 | unsigned long progress; |
5da9d01b | 3498 | |
663350ac | 3499 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3500 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3501 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3502 | |
3503 | smp_mb(); | |
fb25e914 | 3504 | reserved = space_info->bytes_may_use; |
36e39c40 | 3505 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3506 | |
3507 | if (reserved == 0) | |
3508 | return 0; | |
c4f675cd | 3509 | |
fdb5effd JB |
3510 | smp_mb(); |
3511 | if (root->fs_info->delalloc_bytes == 0) { | |
3512 | if (trans) | |
3513 | return 0; | |
3514 | btrfs_wait_ordered_extents(root, 0, 0); | |
3515 | return 0; | |
3516 | } | |
3517 | ||
5da9d01b | 3518 | max_reclaim = min(reserved, to_reclaim); |
877da174 JB |
3519 | nr_pages = max_t(unsigned long, nr_pages, |
3520 | max_reclaim >> PAGE_CACHE_SHIFT); | |
b1953bce | 3521 | while (loops < 1024) { |
bf9022e0 CM |
3522 | /* have the flusher threads jump in and do some IO */ |
3523 | smp_mb(); | |
3524 | nr_pages = min_t(unsigned long, nr_pages, | |
3525 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
0e175a18 CW |
3526 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages, |
3527 | WB_REASON_FS_FREE_SPACE); | |
5da9d01b | 3528 | |
0019f10d | 3529 | spin_lock(&space_info->lock); |
fb25e914 JB |
3530 | if (reserved > space_info->bytes_may_use) |
3531 | reclaimed += reserved - space_info->bytes_may_use; | |
3532 | reserved = space_info->bytes_may_use; | |
0019f10d | 3533 | spin_unlock(&space_info->lock); |
5da9d01b | 3534 | |
36e39c40 CM |
3535 | loops++; |
3536 | ||
5da9d01b YZ |
3537 | if (reserved == 0 || reclaimed >= max_reclaim) |
3538 | break; | |
3539 | ||
3540 | if (trans && trans->transaction->blocked) | |
3541 | return -EAGAIN; | |
bf9022e0 | 3542 | |
f104d044 JB |
3543 | if (wait_ordered && !trans) { |
3544 | btrfs_wait_ordered_extents(root, 0, 0); | |
3545 | } else { | |
3546 | time_left = schedule_timeout_interruptible(1); | |
b1953bce | 3547 | |
f104d044 JB |
3548 | /* We were interrupted, exit */ |
3549 | if (time_left) | |
3550 | break; | |
3551 | } | |
b1953bce | 3552 | |
36e39c40 CM |
3553 | /* we've kicked the IO a few times, if anything has been freed, |
3554 | * exit. There is no sense in looping here for a long time | |
3555 | * when we really need to commit the transaction, or there are | |
3556 | * just too many writers without enough free space | |
3557 | */ | |
3558 | ||
3559 | if (loops > 3) { | |
3560 | smp_mb(); | |
3561 | if (progress != space_info->reservation_progress) | |
3562 | break; | |
3563 | } | |
bf9022e0 | 3564 | |
5da9d01b | 3565 | } |
f104d044 | 3566 | |
5da9d01b YZ |
3567 | return reclaimed >= to_reclaim; |
3568 | } | |
3569 | ||
663350ac JB |
3570 | /** |
3571 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3572 | * @root - the root we're allocating for | |
3573 | * @bytes - the number of bytes we want to reserve | |
3574 | * @force - force the commit | |
8bb8ab2e | 3575 | * |
663350ac JB |
3576 | * This will check to make sure that committing the transaction will actually |
3577 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3578 | * will return -ENOSPC. | |
8bb8ab2e | 3579 | */ |
663350ac JB |
3580 | static int may_commit_transaction(struct btrfs_root *root, |
3581 | struct btrfs_space_info *space_info, | |
3582 | u64 bytes, int force) | |
3583 | { | |
3584 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3585 | struct btrfs_trans_handle *trans; | |
3586 | ||
3587 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3588 | if (trans) | |
3589 | return -EAGAIN; | |
3590 | ||
3591 | if (force) | |
3592 | goto commit; | |
3593 | ||
3594 | /* See if there is enough pinned space to make this reservation */ | |
3595 | spin_lock(&space_info->lock); | |
3596 | if (space_info->bytes_pinned >= bytes) { | |
3597 | spin_unlock(&space_info->lock); | |
3598 | goto commit; | |
3599 | } | |
3600 | spin_unlock(&space_info->lock); | |
3601 | ||
3602 | /* | |
3603 | * See if there is some space in the delayed insertion reservation for | |
3604 | * this reservation. | |
3605 | */ | |
3606 | if (space_info != delayed_rsv->space_info) | |
3607 | return -ENOSPC; | |
3608 | ||
d9b0218f | 3609 | spin_lock(&space_info->lock); |
663350ac | 3610 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3611 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3612 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3613 | spin_unlock(&space_info->lock); |
663350ac JB |
3614 | return -ENOSPC; |
3615 | } | |
3616 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3617 | spin_unlock(&space_info->lock); |
663350ac JB |
3618 | |
3619 | commit: | |
3620 | trans = btrfs_join_transaction(root); | |
3621 | if (IS_ERR(trans)) | |
3622 | return -ENOSPC; | |
3623 | ||
3624 | return btrfs_commit_transaction(trans, root); | |
3625 | } | |
3626 | ||
4a92b1b8 JB |
3627 | /** |
3628 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3629 | * @root - the root we're allocating for | |
3630 | * @block_rsv - the block_rsv we're allocating for | |
3631 | * @orig_bytes - the number of bytes we want | |
3632 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3633 | * |
4a92b1b8 JB |
3634 | * This will reserve orgi_bytes number of bytes from the space info associated |
3635 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3636 | * flush out space to make room. It will do this by flushing delalloc if | |
3637 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3638 | * regain reservations will be made and this will fail if there is not enough | |
3639 | * space already. | |
8bb8ab2e | 3640 | */ |
4a92b1b8 | 3641 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e JB |
3642 | struct btrfs_block_rsv *block_rsv, |
3643 | u64 orig_bytes, int flush) | |
9ed74f2d | 3644 | { |
f0486c68 | 3645 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3646 | u64 used; |
8bb8ab2e JB |
3647 | u64 num_bytes = orig_bytes; |
3648 | int retries = 0; | |
3649 | int ret = 0; | |
38227933 | 3650 | bool committed = false; |
fdb5effd | 3651 | bool flushing = false; |
f104d044 | 3652 | bool wait_ordered = false; |
9ed74f2d | 3653 | |
8bb8ab2e | 3654 | again: |
fdb5effd | 3655 | ret = 0; |
8bb8ab2e | 3656 | spin_lock(&space_info->lock); |
fdb5effd JB |
3657 | /* |
3658 | * We only want to wait if somebody other than us is flushing and we are | |
3659 | * actually alloed to flush. | |
3660 | */ | |
3661 | while (flush && !flushing && space_info->flush) { | |
3662 | spin_unlock(&space_info->lock); | |
3663 | /* | |
3664 | * If we have a trans handle we can't wait because the flusher | |
3665 | * may have to commit the transaction, which would mean we would | |
3666 | * deadlock since we are waiting for the flusher to finish, but | |
3667 | * hold the current transaction open. | |
3668 | */ | |
663350ac | 3669 | if (current->journal_info) |
fdb5effd JB |
3670 | return -EAGAIN; |
3671 | ret = wait_event_interruptible(space_info->wait, | |
3672 | !space_info->flush); | |
3673 | /* Must have been interrupted, return */ | |
3674 | if (ret) | |
3675 | return -EINTR; | |
3676 | ||
3677 | spin_lock(&space_info->lock); | |
3678 | } | |
3679 | ||
3680 | ret = -ENOSPC; | |
2bf64758 JB |
3681 | used = space_info->bytes_used + space_info->bytes_reserved + |
3682 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3683 | space_info->bytes_may_use; | |
9ed74f2d | 3684 | |
8bb8ab2e JB |
3685 | /* |
3686 | * The idea here is that we've not already over-reserved the block group | |
3687 | * then we can go ahead and save our reservation first and then start | |
3688 | * flushing if we need to. Otherwise if we've already overcommitted | |
3689 | * lets start flushing stuff first and then come back and try to make | |
3690 | * our reservation. | |
3691 | */ | |
2bf64758 JB |
3692 | if (used <= space_info->total_bytes) { |
3693 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3694 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3695 | trace_btrfs_space_reservation(root->fs_info, |
e77266e4 CM |
3696 | "space_info", |
3697 | (u64)(unsigned long)space_info, | |
3698 | orig_bytes, 1); | |
8bb8ab2e JB |
3699 | ret = 0; |
3700 | } else { | |
3701 | /* | |
3702 | * Ok set num_bytes to orig_bytes since we aren't | |
3703 | * overocmmitted, this way we only try and reclaim what | |
3704 | * we need. | |
3705 | */ | |
3706 | num_bytes = orig_bytes; | |
3707 | } | |
3708 | } else { | |
3709 | /* | |
3710 | * Ok we're over committed, set num_bytes to the overcommitted | |
3711 | * amount plus the amount of bytes that we need for this | |
3712 | * reservation. | |
3713 | */ | |
f104d044 | 3714 | wait_ordered = true; |
2bf64758 | 3715 | num_bytes = used - space_info->total_bytes + |
8bb8ab2e JB |
3716 | (orig_bytes * (retries + 1)); |
3717 | } | |
9ed74f2d | 3718 | |
36ba022a | 3719 | if (ret) { |
2bf64758 JB |
3720 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3721 | u64 avail; | |
3722 | ||
7e355b83 JB |
3723 | /* |
3724 | * If we have a lot of space that's pinned, don't bother doing | |
3725 | * the overcommit dance yet and just commit the transaction. | |
3726 | */ | |
3727 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3728 | do_div(avail, 10); | |
663350ac | 3729 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3730 | space_info->flush = 1; |
3731 | flushing = true; | |
3732 | spin_unlock(&space_info->lock); | |
663350ac JB |
3733 | ret = may_commit_transaction(root, space_info, |
3734 | orig_bytes, 1); | |
3735 | if (ret) | |
3736 | goto out; | |
3737 | committed = true; | |
3738 | goto again; | |
7e355b83 JB |
3739 | } |
3740 | ||
2bf64758 JB |
3741 | spin_lock(&root->fs_info->free_chunk_lock); |
3742 | avail = root->fs_info->free_chunk_space; | |
3743 | ||
3744 | /* | |
3745 | * If we have dup, raid1 or raid10 then only half of the free | |
3746 | * space is actually useable. | |
3747 | */ | |
3748 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3749 | BTRFS_BLOCK_GROUP_RAID1 | | |
3750 | BTRFS_BLOCK_GROUP_RAID10)) | |
3751 | avail >>= 1; | |
3752 | ||
3753 | /* | |
3754 | * If we aren't flushing don't let us overcommit too much, say | |
3755 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3756 | * 1/2 of the space. | |
3757 | */ | |
3758 | if (flush) | |
3759 | avail >>= 3; | |
3760 | else | |
3761 | avail >>= 1; | |
3762 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3763 | ||
9a82ca65 | 3764 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 | 3765 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3766 | trace_btrfs_space_reservation(root->fs_info, |
e77266e4 CM |
3767 | "space_info", |
3768 | (u64)(unsigned long)space_info, | |
3769 | orig_bytes, 1); | |
2bf64758 | 3770 | ret = 0; |
f104d044 JB |
3771 | } else { |
3772 | wait_ordered = true; | |
2bf64758 JB |
3773 | } |
3774 | } | |
3775 | ||
8bb8ab2e JB |
3776 | /* |
3777 | * Couldn't make our reservation, save our place so while we're trying | |
3778 | * to reclaim space we can actually use it instead of somebody else | |
3779 | * stealing it from us. | |
3780 | */ | |
fdb5effd JB |
3781 | if (ret && flush) { |
3782 | flushing = true; | |
3783 | space_info->flush = 1; | |
8bb8ab2e | 3784 | } |
9ed74f2d | 3785 | |
f0486c68 | 3786 | spin_unlock(&space_info->lock); |
9ed74f2d | 3787 | |
fdb5effd | 3788 | if (!ret || !flush) |
8bb8ab2e | 3789 | goto out; |
f0486c68 | 3790 | |
8bb8ab2e JB |
3791 | /* |
3792 | * We do synchronous shrinking since we don't actually unreserve | |
3793 | * metadata until after the IO is completed. | |
3794 | */ | |
663350ac | 3795 | ret = shrink_delalloc(root, num_bytes, wait_ordered); |
fdb5effd | 3796 | if (ret < 0) |
8bb8ab2e | 3797 | goto out; |
f0486c68 | 3798 | |
75c195a2 CM |
3799 | ret = 0; |
3800 | ||
8bb8ab2e JB |
3801 | /* |
3802 | * So if we were overcommitted it's possible that somebody else flushed | |
3803 | * out enough space and we simply didn't have enough space to reclaim, | |
3804 | * so go back around and try again. | |
3805 | */ | |
3806 | if (retries < 2) { | |
f104d044 | 3807 | wait_ordered = true; |
8bb8ab2e JB |
3808 | retries++; |
3809 | goto again; | |
3810 | } | |
f0486c68 | 3811 | |
8bb8ab2e | 3812 | ret = -ENOSPC; |
75c195a2 CM |
3813 | if (committed) |
3814 | goto out; | |
3815 | ||
663350ac | 3816 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); |
38227933 | 3817 | if (!ret) { |
38227933 | 3818 | committed = true; |
8bb8ab2e | 3819 | goto again; |
38227933 | 3820 | } |
8bb8ab2e JB |
3821 | |
3822 | out: | |
fdb5effd | 3823 | if (flushing) { |
8bb8ab2e | 3824 | spin_lock(&space_info->lock); |
fdb5effd JB |
3825 | space_info->flush = 0; |
3826 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3827 | spin_unlock(&space_info->lock); |
f0486c68 | 3828 | } |
f0486c68 YZ |
3829 | return ret; |
3830 | } | |
3831 | ||
3832 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3833 | struct btrfs_root *root) | |
3834 | { | |
4c13d758 JB |
3835 | struct btrfs_block_rsv *block_rsv = NULL; |
3836 | ||
3837 | if (root->ref_cows || root == root->fs_info->csum_root) | |
f0486c68 | 3838 | block_rsv = trans->block_rsv; |
4c13d758 JB |
3839 | |
3840 | if (!block_rsv) | |
f0486c68 YZ |
3841 | block_rsv = root->block_rsv; |
3842 | ||
3843 | if (!block_rsv) | |
3844 | block_rsv = &root->fs_info->empty_block_rsv; | |
3845 | ||
3846 | return block_rsv; | |
3847 | } | |
3848 | ||
3849 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3850 | u64 num_bytes) | |
3851 | { | |
3852 | int ret = -ENOSPC; | |
3853 | spin_lock(&block_rsv->lock); | |
3854 | if (block_rsv->reserved >= num_bytes) { | |
3855 | block_rsv->reserved -= num_bytes; | |
3856 | if (block_rsv->reserved < block_rsv->size) | |
3857 | block_rsv->full = 0; | |
3858 | ret = 0; | |
3859 | } | |
3860 | spin_unlock(&block_rsv->lock); | |
3861 | return ret; | |
3862 | } | |
3863 | ||
3864 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3865 | u64 num_bytes, int update_size) | |
3866 | { | |
3867 | spin_lock(&block_rsv->lock); | |
3868 | block_rsv->reserved += num_bytes; | |
3869 | if (update_size) | |
3870 | block_rsv->size += num_bytes; | |
3871 | else if (block_rsv->reserved >= block_rsv->size) | |
3872 | block_rsv->full = 1; | |
3873 | spin_unlock(&block_rsv->lock); | |
3874 | } | |
3875 | ||
8c2a3ca2 JB |
3876 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
3877 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 3878 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
3879 | { |
3880 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3881 | ||
3882 | spin_lock(&block_rsv->lock); | |
3883 | if (num_bytes == (u64)-1) | |
3884 | num_bytes = block_rsv->size; | |
3885 | block_rsv->size -= num_bytes; | |
3886 | if (block_rsv->reserved >= block_rsv->size) { | |
3887 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3888 | block_rsv->reserved = block_rsv->size; | |
3889 | block_rsv->full = 1; | |
3890 | } else { | |
3891 | num_bytes = 0; | |
3892 | } | |
3893 | spin_unlock(&block_rsv->lock); | |
3894 | ||
3895 | if (num_bytes > 0) { | |
3896 | if (dest) { | |
e9e22899 JB |
3897 | spin_lock(&dest->lock); |
3898 | if (!dest->full) { | |
3899 | u64 bytes_to_add; | |
3900 | ||
3901 | bytes_to_add = dest->size - dest->reserved; | |
3902 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3903 | dest->reserved += bytes_to_add; | |
3904 | if (dest->reserved >= dest->size) | |
3905 | dest->full = 1; | |
3906 | num_bytes -= bytes_to_add; | |
3907 | } | |
3908 | spin_unlock(&dest->lock); | |
3909 | } | |
3910 | if (num_bytes) { | |
f0486c68 | 3911 | spin_lock(&space_info->lock); |
fb25e914 | 3912 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 3913 | trace_btrfs_space_reservation(fs_info, "space_info", |
e77266e4 CM |
3914 | (u64)(unsigned long)space_info, |
3915 | num_bytes, 0); | |
36e39c40 | 3916 | space_info->reservation_progress++; |
f0486c68 | 3917 | spin_unlock(&space_info->lock); |
4e06bdd6 | 3918 | } |
9ed74f2d | 3919 | } |
f0486c68 | 3920 | } |
4e06bdd6 | 3921 | |
f0486c68 YZ |
3922 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3923 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3924 | { | |
3925 | int ret; | |
9ed74f2d | 3926 | |
f0486c68 YZ |
3927 | ret = block_rsv_use_bytes(src, num_bytes); |
3928 | if (ret) | |
3929 | return ret; | |
9ed74f2d | 3930 | |
f0486c68 | 3931 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3932 | return 0; |
3933 | } | |
3934 | ||
f0486c68 | 3935 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3936 | { |
f0486c68 YZ |
3937 | memset(rsv, 0, sizeof(*rsv)); |
3938 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
3939 | } |
3940 | ||
3941 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3942 | { | |
3943 | struct btrfs_block_rsv *block_rsv; | |
3944 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3945 | |
f0486c68 YZ |
3946 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3947 | if (!block_rsv) | |
3948 | return NULL; | |
9ed74f2d | 3949 | |
f0486c68 | 3950 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3951 | block_rsv->space_info = __find_space_info(fs_info, |
3952 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3953 | return block_rsv; |
3954 | } | |
9ed74f2d | 3955 | |
f0486c68 YZ |
3956 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3957 | struct btrfs_block_rsv *rsv) | |
3958 | { | |
dabdb640 JB |
3959 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
3960 | kfree(rsv); | |
9ed74f2d JB |
3961 | } |
3962 | ||
61b520a9 MX |
3963 | static inline int __block_rsv_add(struct btrfs_root *root, |
3964 | struct btrfs_block_rsv *block_rsv, | |
3965 | u64 num_bytes, int flush) | |
9ed74f2d | 3966 | { |
f0486c68 | 3967 | int ret; |
9ed74f2d | 3968 | |
f0486c68 YZ |
3969 | if (num_bytes == 0) |
3970 | return 0; | |
8bb8ab2e | 3971 | |
61b520a9 | 3972 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
3973 | if (!ret) { |
3974 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3975 | return 0; | |
3976 | } | |
9ed74f2d | 3977 | |
f0486c68 | 3978 | return ret; |
f0486c68 | 3979 | } |
9ed74f2d | 3980 | |
61b520a9 MX |
3981 | int btrfs_block_rsv_add(struct btrfs_root *root, |
3982 | struct btrfs_block_rsv *block_rsv, | |
3983 | u64 num_bytes) | |
3984 | { | |
3985 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
3986 | } | |
3987 | ||
c06a0e12 JB |
3988 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
3989 | struct btrfs_block_rsv *block_rsv, | |
3990 | u64 num_bytes) | |
f0486c68 | 3991 | { |
61b520a9 | 3992 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
f0486c68 | 3993 | } |
9ed74f2d | 3994 | |
4a92b1b8 | 3995 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 3996 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
3997 | { |
3998 | u64 num_bytes = 0; | |
f0486c68 | 3999 | int ret = -ENOSPC; |
9ed74f2d | 4000 | |
f0486c68 YZ |
4001 | if (!block_rsv) |
4002 | return 0; | |
9ed74f2d | 4003 | |
f0486c68 | 4004 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4005 | num_bytes = div_factor(block_rsv->size, min_factor); |
4006 | if (block_rsv->reserved >= num_bytes) | |
4007 | ret = 0; | |
4008 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4009 | |
36ba022a JB |
4010 | return ret; |
4011 | } | |
4012 | ||
aa38a711 MX |
4013 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
4014 | struct btrfs_block_rsv *block_rsv, | |
4015 | u64 min_reserved, int flush) | |
36ba022a JB |
4016 | { |
4017 | u64 num_bytes = 0; | |
4018 | int ret = -ENOSPC; | |
4019 | ||
4020 | if (!block_rsv) | |
4021 | return 0; | |
4022 | ||
4023 | spin_lock(&block_rsv->lock); | |
4024 | num_bytes = min_reserved; | |
13553e52 | 4025 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4026 | ret = 0; |
13553e52 | 4027 | else |
f0486c68 | 4028 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4029 | spin_unlock(&block_rsv->lock); |
13553e52 | 4030 | |
f0486c68 YZ |
4031 | if (!ret) |
4032 | return 0; | |
4033 | ||
aa38a711 | 4034 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4035 | if (!ret) { |
4036 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4037 | return 0; |
6a63209f | 4038 | } |
9ed74f2d | 4039 | |
13553e52 | 4040 | return ret; |
f0486c68 YZ |
4041 | } |
4042 | ||
aa38a711 MX |
4043 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4044 | struct btrfs_block_rsv *block_rsv, | |
4045 | u64 min_reserved) | |
4046 | { | |
4047 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
4048 | } | |
4049 | ||
4050 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
4051 | struct btrfs_block_rsv *block_rsv, | |
4052 | u64 min_reserved) | |
4053 | { | |
4054 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
4055 | } | |
4056 | ||
f0486c68 YZ |
4057 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
4058 | struct btrfs_block_rsv *dst_rsv, | |
4059 | u64 num_bytes) | |
4060 | { | |
4061 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4062 | } | |
4063 | ||
4064 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4065 | struct btrfs_block_rsv *block_rsv, | |
4066 | u64 num_bytes) | |
4067 | { | |
4068 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4069 | if (global_rsv->full || global_rsv == block_rsv || | |
4070 | block_rsv->space_info != global_rsv->space_info) | |
4071 | global_rsv = NULL; | |
8c2a3ca2 JB |
4072 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4073 | num_bytes); | |
6a63209f JB |
4074 | } |
4075 | ||
4076 | /* | |
8929ecfa YZ |
4077 | * helper to calculate size of global block reservation. |
4078 | * the desired value is sum of space used by extent tree, | |
4079 | * checksum tree and root tree | |
6a63209f | 4080 | */ |
8929ecfa | 4081 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4082 | { |
8929ecfa YZ |
4083 | struct btrfs_space_info *sinfo; |
4084 | u64 num_bytes; | |
4085 | u64 meta_used; | |
4086 | u64 data_used; | |
6c41761f | 4087 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4088 | |
8929ecfa YZ |
4089 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4090 | spin_lock(&sinfo->lock); | |
4091 | data_used = sinfo->bytes_used; | |
4092 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4093 | |
8929ecfa YZ |
4094 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4095 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4096 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4097 | data_used = 0; | |
8929ecfa YZ |
4098 | meta_used = sinfo->bytes_used; |
4099 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4100 | |
8929ecfa YZ |
4101 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4102 | csum_size * 2; | |
4103 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4104 | |
8929ecfa | 4105 | if (num_bytes * 3 > meta_used) |
5500cdbe | 4106 | num_bytes = div64_u64(meta_used, 3) * 2; |
ab6e2410 | 4107 | |
8929ecfa YZ |
4108 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4109 | } | |
6a63209f | 4110 | |
8929ecfa YZ |
4111 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4112 | { | |
4113 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4114 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4115 | u64 num_bytes; | |
6a63209f | 4116 | |
8929ecfa | 4117 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4118 | |
8929ecfa YZ |
4119 | spin_lock(&block_rsv->lock); |
4120 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 4121 | |
8929ecfa | 4122 | block_rsv->size = num_bytes; |
4e06bdd6 | 4123 | |
8929ecfa | 4124 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4125 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4126 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4127 | |
4128 | if (sinfo->total_bytes > num_bytes) { | |
4129 | num_bytes = sinfo->total_bytes - num_bytes; | |
4130 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4131 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4132 | trace_btrfs_space_reservation(fs_info, "space_info", |
e77266e4 | 4133 | (u64)(unsigned long)sinfo, num_bytes, 1); |
6a63209f | 4134 | } |
6a63209f | 4135 | |
8929ecfa YZ |
4136 | if (block_rsv->reserved >= block_rsv->size) { |
4137 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4138 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4139 | trace_btrfs_space_reservation(fs_info, "space_info", |
e77266e4 | 4140 | (u64)(unsigned long)sinfo, num_bytes, 0); |
36e39c40 | 4141 | sinfo->reservation_progress++; |
8929ecfa YZ |
4142 | block_rsv->reserved = block_rsv->size; |
4143 | block_rsv->full = 1; | |
4144 | } | |
182608c8 | 4145 | |
8929ecfa YZ |
4146 | spin_unlock(&sinfo->lock); |
4147 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
4148 | } |
4149 | ||
f0486c68 | 4150 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4151 | { |
f0486c68 | 4152 | struct btrfs_space_info *space_info; |
6a63209f | 4153 | |
f0486c68 YZ |
4154 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4155 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4156 | |
f0486c68 | 4157 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4158 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4159 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4160 | fs_info->trans_block_rsv.space_info = space_info; |
4161 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4162 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4163 | |
8929ecfa YZ |
4164 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4165 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4166 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4167 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4168 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4169 | |
8929ecfa | 4170 | update_global_block_rsv(fs_info); |
6a63209f JB |
4171 | } |
4172 | ||
8929ecfa | 4173 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4174 | { |
8c2a3ca2 JB |
4175 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4176 | (u64)-1); | |
8929ecfa YZ |
4177 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4178 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4179 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4180 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4181 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4182 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4183 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4184 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4185 | } |
4186 | ||
a22285a6 YZ |
4187 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4188 | struct btrfs_root *root) | |
6a63209f | 4189 | { |
a22285a6 YZ |
4190 | if (!trans->bytes_reserved) |
4191 | return; | |
6a63209f | 4192 | |
e77266e4 CM |
4193 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
4194 | (u64)(unsigned long)trans, | |
8c2a3ca2 | 4195 | trans->bytes_reserved, 0); |
b24e03db | 4196 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4197 | trans->bytes_reserved = 0; |
4198 | } | |
6a63209f | 4199 | |
d68fc57b YZ |
4200 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4201 | struct inode *inode) | |
4202 | { | |
4203 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4204 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4205 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4206 | ||
4207 | /* | |
fcb80c2a JB |
4208 | * We need to hold space in order to delete our orphan item once we've |
4209 | * added it, so this takes the reservation so we can release it later | |
4210 | * when we are truly done with the orphan item. | |
d68fc57b | 4211 | */ |
ff5714cc | 4212 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4213 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4214 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4215 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4216 | } |
4217 | ||
d68fc57b | 4218 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4219 | { |
d68fc57b | 4220 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4221 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4222 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4223 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4224 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4225 | } | |
97e728d4 | 4226 | |
a22285a6 YZ |
4227 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4228 | struct btrfs_pending_snapshot *pending) | |
4229 | { | |
4230 | struct btrfs_root *root = pending->root; | |
4231 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4232 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4233 | /* | |
4234 | * two for root back/forward refs, two for directory entries | |
4235 | * and one for root of the snapshot. | |
4236 | */ | |
16cdcec7 | 4237 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4238 | dst_rsv->space_info = src_rsv->space_info; |
4239 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4240 | } |
4241 | ||
7709cde3 JB |
4242 | /** |
4243 | * drop_outstanding_extent - drop an outstanding extent | |
4244 | * @inode: the inode we're dropping the extent for | |
4245 | * | |
4246 | * This is called when we are freeing up an outstanding extent, either called | |
4247 | * after an error or after an extent is written. This will return the number of | |
4248 | * reserved extents that need to be freed. This must be called with | |
4249 | * BTRFS_I(inode)->lock held. | |
4250 | */ | |
9e0baf60 JB |
4251 | static unsigned drop_outstanding_extent(struct inode *inode) |
4252 | { | |
7fd2ae21 | 4253 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4254 | unsigned dropped_extents = 0; |
4255 | ||
9e0baf60 JB |
4256 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4257 | BTRFS_I(inode)->outstanding_extents--; | |
4258 | ||
7fd2ae21 JB |
4259 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
4260 | BTRFS_I(inode)->delalloc_meta_reserved) { | |
4261 | drop_inode_space = 1; | |
4262 | BTRFS_I(inode)->delalloc_meta_reserved = 0; | |
4263 | } | |
4264 | ||
9e0baf60 JB |
4265 | /* |
4266 | * If we have more or the same amount of outsanding extents than we have | |
4267 | * reserved then we need to leave the reserved extents count alone. | |
4268 | */ | |
4269 | if (BTRFS_I(inode)->outstanding_extents >= | |
4270 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4271 | return drop_inode_space; |
9e0baf60 JB |
4272 | |
4273 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4274 | BTRFS_I(inode)->outstanding_extents; | |
4275 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4276 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4277 | } |
4278 | ||
7709cde3 JB |
4279 | /** |
4280 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4281 | * reserved/free'd for the given bytes. | |
4282 | * @inode: the inode we're manipulating | |
4283 | * @num_bytes: the number of bytes in question | |
4284 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4285 | * | |
4286 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4287 | * correct amount of metadata that must either be reserved or freed. We | |
4288 | * calculate how many checksums we can fit into one leaf and then divide the | |
4289 | * number of bytes that will need to be checksumed by this value to figure out | |
4290 | * how many checksums will be required. If we are adding bytes then the number | |
4291 | * may go up and we will return the number of additional bytes that must be | |
4292 | * reserved. If it is going down we will return the number of bytes that must | |
4293 | * be freed. | |
4294 | * | |
4295 | * This must be called with BTRFS_I(inode)->lock held. | |
4296 | */ | |
4297 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4298 | int reserve) | |
6324fbf3 | 4299 | { |
7709cde3 JB |
4300 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4301 | u64 csum_size; | |
4302 | int num_csums_per_leaf; | |
4303 | int num_csums; | |
4304 | int old_csums; | |
4305 | ||
4306 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4307 | BTRFS_I(inode)->csum_bytes == 0) | |
4308 | return 0; | |
4309 | ||
4310 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4311 | if (reserve) | |
4312 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4313 | else | |
4314 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4315 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4316 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4317 | sizeof(struct btrfs_csum_item) + | |
4318 | sizeof(struct btrfs_disk_key)); | |
4319 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4320 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4321 | num_csums = num_csums / num_csums_per_leaf; | |
4322 | ||
4323 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4324 | old_csums = old_csums / num_csums_per_leaf; | |
4325 | ||
4326 | /* No change, no need to reserve more */ | |
4327 | if (old_csums == num_csums) | |
4328 | return 0; | |
4329 | ||
4330 | if (reserve) | |
4331 | return btrfs_calc_trans_metadata_size(root, | |
4332 | num_csums - old_csums); | |
4333 | ||
4334 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4335 | } |
c146afad | 4336 | |
0ca1f7ce YZ |
4337 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4338 | { | |
4339 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4340 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4341 | u64 to_reserve = 0; |
660d3f6c | 4342 | u64 csum_bytes; |
9e0baf60 | 4343 | unsigned nr_extents = 0; |
660d3f6c | 4344 | int extra_reserve = 0; |
c09544e0 | 4345 | int flush = 1; |
0ca1f7ce | 4346 | int ret; |
6324fbf3 | 4347 | |
660d3f6c | 4348 | /* Need to be holding the i_mutex here if we aren't free space cache */ |
c09544e0 JB |
4349 | if (btrfs_is_free_space_inode(root, inode)) |
4350 | flush = 0; | |
4351 | ||
4352 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4353 | schedule_timeout(1); |
ec44a35c | 4354 | |
f248679e | 4355 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4356 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4357 | |
9e0baf60 JB |
4358 | spin_lock(&BTRFS_I(inode)->lock); |
4359 | BTRFS_I(inode)->outstanding_extents++; | |
4360 | ||
4361 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4362 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4363 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4364 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4365 | |
7fd2ae21 JB |
4366 | /* |
4367 | * Add an item to reserve for updating the inode when we complete the | |
4368 | * delalloc io. | |
4369 | */ | |
4370 | if (!BTRFS_I(inode)->delalloc_meta_reserved) { | |
4371 | nr_extents++; | |
660d3f6c | 4372 | extra_reserve = 1; |
593060d7 | 4373 | } |
7fd2ae21 JB |
4374 | |
4375 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4376 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4377 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4378 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4379 | |
36ba022a | 4380 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4381 | if (ret) { |
7ed49f18 | 4382 | u64 to_free = 0; |
9e0baf60 | 4383 | unsigned dropped; |
7ed49f18 | 4384 | |
7709cde3 | 4385 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4386 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4387 | /* |
660d3f6c JB |
4388 | * If the inodes csum_bytes is the same as the original |
4389 | * csum_bytes then we know we haven't raced with any free()ers | |
4390 | * so we can just reduce our inodes csum bytes and carry on. | |
4391 | * Otherwise we have to do the normal free thing to account for | |
4392 | * the case that the free side didn't free up its reserve | |
4393 | * because of this outstanding reservation. | |
9e0baf60 | 4394 | */ |
660d3f6c JB |
4395 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4396 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4397 | else | |
4398 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4399 | spin_unlock(&BTRFS_I(inode)->lock); | |
4400 | if (dropped) | |
4401 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4402 | ||
8c2a3ca2 | 4403 | if (to_free) { |
7ed49f18 | 4404 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4405 | trace_btrfs_space_reservation(root->fs_info, |
4406 | "delalloc", | |
4407 | btrfs_ino(inode), | |
4408 | to_free, 0); | |
4409 | } | |
f248679e | 4410 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4411 | return ret; |
9e0baf60 | 4412 | } |
25179201 | 4413 | |
660d3f6c JB |
4414 | spin_lock(&BTRFS_I(inode)->lock); |
4415 | if (extra_reserve) { | |
4416 | BTRFS_I(inode)->delalloc_meta_reserved = 1; | |
4417 | nr_extents--; | |
4418 | } | |
4419 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4420 | spin_unlock(&BTRFS_I(inode)->lock); | |
f248679e | 4421 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
660d3f6c | 4422 | |
8c2a3ca2 JB |
4423 | if (to_reserve) |
4424 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4425 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4426 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4427 | ||
0ca1f7ce YZ |
4428 | return 0; |
4429 | } | |
4430 | ||
7709cde3 JB |
4431 | /** |
4432 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4433 | * @inode: the inode to release the reservation for | |
4434 | * @num_bytes: the number of bytes we're releasing | |
4435 | * | |
4436 | * This will release the metadata reservation for an inode. This can be called | |
4437 | * once we complete IO for a given set of bytes to release their metadata | |
4438 | * reservations. | |
4439 | */ | |
0ca1f7ce YZ |
4440 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4441 | { | |
4442 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4443 | u64 to_free = 0; |
4444 | unsigned dropped; | |
0ca1f7ce YZ |
4445 | |
4446 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4447 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4448 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4449 | |
7709cde3 JB |
4450 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4451 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4452 | if (dropped > 0) |
4453 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4454 | |
8c2a3ca2 JB |
4455 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4456 | btrfs_ino(inode), to_free, 0); | |
0ca1f7ce YZ |
4457 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4458 | to_free); | |
4459 | } | |
4460 | ||
7709cde3 JB |
4461 | /** |
4462 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4463 | * @inode: inode we're writing to | |
4464 | * @num_bytes: the number of bytes we want to allocate | |
4465 | * | |
4466 | * This will do the following things | |
4467 | * | |
4468 | * o reserve space in the data space info for num_bytes | |
4469 | * o reserve space in the metadata space info based on number of outstanding | |
4470 | * extents and how much csums will be needed | |
4471 | * o add to the inodes ->delalloc_bytes | |
4472 | * o add it to the fs_info's delalloc inodes list. | |
4473 | * | |
4474 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4475 | */ | |
0ca1f7ce YZ |
4476 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4477 | { | |
4478 | int ret; | |
4479 | ||
4480 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4481 | if (ret) |
0ca1f7ce YZ |
4482 | return ret; |
4483 | ||
4484 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4485 | if (ret) { | |
4486 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4487 | return ret; | |
4488 | } | |
4489 | ||
4490 | return 0; | |
4491 | } | |
4492 | ||
7709cde3 JB |
4493 | /** |
4494 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4495 | * @inode: inode we're releasing space for | |
4496 | * @num_bytes: the number of bytes we want to free up | |
4497 | * | |
4498 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4499 | * called in the case that we don't need the metadata AND data reservations | |
4500 | * anymore. So if there is an error or we insert an inline extent. | |
4501 | * | |
4502 | * This function will release the metadata space that was not used and will | |
4503 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4504 | * list if there are no delalloc bytes left. | |
4505 | */ | |
0ca1f7ce YZ |
4506 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4507 | { | |
4508 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4509 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4510 | } |
4511 | ||
9078a3e1 CM |
4512 | static int update_block_group(struct btrfs_trans_handle *trans, |
4513 | struct btrfs_root *root, | |
f0486c68 | 4514 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4515 | { |
0af3d00b | 4516 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4517 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4518 | u64 total = num_bytes; |
9078a3e1 | 4519 | u64 old_val; |
db94535d | 4520 | u64 byte_in_group; |
0af3d00b | 4521 | int factor; |
3e1ad54f | 4522 | |
5d4f98a2 YZ |
4523 | /* block accounting for super block */ |
4524 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4525 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4526 | if (alloc) |
4527 | old_val += num_bytes; | |
4528 | else | |
4529 | old_val -= num_bytes; | |
6c41761f | 4530 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4531 | spin_unlock(&info->delalloc_lock); |
4532 | ||
d397712b | 4533 | while (total) { |
db94535d | 4534 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4535 | if (!cache) |
9078a3e1 | 4536 | return -1; |
b742bb82 YZ |
4537 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4538 | BTRFS_BLOCK_GROUP_RAID1 | | |
4539 | BTRFS_BLOCK_GROUP_RAID10)) | |
4540 | factor = 2; | |
4541 | else | |
4542 | factor = 1; | |
9d66e233 JB |
4543 | /* |
4544 | * If this block group has free space cache written out, we | |
4545 | * need to make sure to load it if we are removing space. This | |
4546 | * is because we need the unpinning stage to actually add the | |
4547 | * space back to the block group, otherwise we will leak space. | |
4548 | */ | |
4549 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4550 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4551 | |
db94535d CM |
4552 | byte_in_group = bytenr - cache->key.objectid; |
4553 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4554 | |
25179201 | 4555 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4556 | spin_lock(&cache->lock); |
0af3d00b | 4557 | |
73bc1876 | 4558 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4559 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4560 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4561 | ||
0f9dd46c | 4562 | cache->dirty = 1; |
9078a3e1 | 4563 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4564 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4565 | if (alloc) { |
db94535d | 4566 | old_val += num_bytes; |
11833d66 YZ |
4567 | btrfs_set_block_group_used(&cache->item, old_val); |
4568 | cache->reserved -= num_bytes; | |
11833d66 | 4569 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4570 | cache->space_info->bytes_used += num_bytes; |
4571 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4572 | spin_unlock(&cache->lock); |
25179201 | 4573 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4574 | } else { |
db94535d | 4575 | old_val -= num_bytes; |
c286ac48 | 4576 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4577 | cache->pinned += num_bytes; |
4578 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4579 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4580 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4581 | spin_unlock(&cache->lock); |
25179201 | 4582 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4583 | |
f0486c68 YZ |
4584 | set_extent_dirty(info->pinned_extents, |
4585 | bytenr, bytenr + num_bytes - 1, | |
4586 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4587 | } |
fa9c0d79 | 4588 | btrfs_put_block_group(cache); |
db94535d CM |
4589 | total -= num_bytes; |
4590 | bytenr += num_bytes; | |
9078a3e1 CM |
4591 | } |
4592 | return 0; | |
4593 | } | |
6324fbf3 | 4594 | |
a061fc8d CM |
4595 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4596 | { | |
0f9dd46c | 4597 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4598 | u64 bytenr; |
0f9dd46c JB |
4599 | |
4600 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4601 | if (!cache) | |
a061fc8d | 4602 | return 0; |
0f9dd46c | 4603 | |
d2fb3437 | 4604 | bytenr = cache->key.objectid; |
fa9c0d79 | 4605 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4606 | |
4607 | return bytenr; | |
a061fc8d CM |
4608 | } |
4609 | ||
f0486c68 YZ |
4610 | static int pin_down_extent(struct btrfs_root *root, |
4611 | struct btrfs_block_group_cache *cache, | |
4612 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4613 | { |
11833d66 YZ |
4614 | spin_lock(&cache->space_info->lock); |
4615 | spin_lock(&cache->lock); | |
4616 | cache->pinned += num_bytes; | |
4617 | cache->space_info->bytes_pinned += num_bytes; | |
4618 | if (reserved) { | |
4619 | cache->reserved -= num_bytes; | |
4620 | cache->space_info->bytes_reserved -= num_bytes; | |
4621 | } | |
4622 | spin_unlock(&cache->lock); | |
4623 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4624 | |
f0486c68 YZ |
4625 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4626 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4627 | return 0; | |
4628 | } | |
68b38550 | 4629 | |
f0486c68 YZ |
4630 | /* |
4631 | * this function must be called within transaction | |
4632 | */ | |
4633 | int btrfs_pin_extent(struct btrfs_root *root, | |
4634 | u64 bytenr, u64 num_bytes, int reserved) | |
4635 | { | |
4636 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4637 | |
f0486c68 YZ |
4638 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4639 | BUG_ON(!cache); | |
4640 | ||
4641 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4642 | ||
4643 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4644 | return 0; |
4645 | } | |
4646 | ||
f0486c68 | 4647 | /* |
e688b725 CM |
4648 | * this function must be called within transaction |
4649 | */ | |
4650 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4651 | struct btrfs_root *root, | |
4652 | u64 bytenr, u64 num_bytes) | |
4653 | { | |
4654 | struct btrfs_block_group_cache *cache; | |
4655 | ||
4656 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
4657 | BUG_ON(!cache); | |
4658 | ||
4659 | /* | |
4660 | * pull in the free space cache (if any) so that our pin | |
4661 | * removes the free space from the cache. We have load_only set | |
4662 | * to one because the slow code to read in the free extents does check | |
4663 | * the pinned extents. | |
4664 | */ | |
4665 | cache_block_group(cache, trans, root, 1); | |
4666 | ||
4667 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4668 | ||
4669 | /* remove us from the free space cache (if we're there at all) */ | |
4670 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4671 | btrfs_put_block_group(cache); | |
4672 | return 0; | |
4673 | } | |
4674 | ||
fb25e914 JB |
4675 | /** |
4676 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4677 | * @cache: The cache we are manipulating | |
4678 | * @num_bytes: The number of bytes in question | |
4679 | * @reserve: One of the reservation enums | |
4680 | * | |
4681 | * This is called by the allocator when it reserves space, or by somebody who is | |
4682 | * freeing space that was never actually used on disk. For example if you | |
4683 | * reserve some space for a new leaf in transaction A and before transaction A | |
4684 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4685 | * clear the reservation. | |
4686 | * | |
4687 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4688 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4689 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4690 | * allocating less disk space for the amount of data we have reserved in the | |
4691 | * case of compression. | |
4692 | * | |
4693 | * If this is a reservation and the block group has become read only we cannot | |
4694 | * make the reservation and return -EAGAIN, otherwise this function always | |
4695 | * succeeds. | |
f0486c68 | 4696 | */ |
fb25e914 JB |
4697 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4698 | u64 num_bytes, int reserve) | |
11833d66 | 4699 | { |
fb25e914 | 4700 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4701 | int ret = 0; |
fb25e914 JB |
4702 | spin_lock(&space_info->lock); |
4703 | spin_lock(&cache->lock); | |
4704 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4705 | if (cache->ro) { |
4706 | ret = -EAGAIN; | |
4707 | } else { | |
fb25e914 JB |
4708 | cache->reserved += num_bytes; |
4709 | space_info->bytes_reserved += num_bytes; | |
4710 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4711 | trace_btrfs_space_reservation(cache->fs_info, |
e77266e4 CM |
4712 | "space_info", |
4713 | (u64)(unsigned long)space_info, | |
4714 | num_bytes, 0); | |
fb25e914 JB |
4715 | space_info->bytes_may_use -= num_bytes; |
4716 | } | |
f0486c68 | 4717 | } |
fb25e914 JB |
4718 | } else { |
4719 | if (cache->ro) | |
4720 | space_info->bytes_readonly += num_bytes; | |
4721 | cache->reserved -= num_bytes; | |
4722 | space_info->bytes_reserved -= num_bytes; | |
4723 | space_info->reservation_progress++; | |
324ae4df | 4724 | } |
fb25e914 JB |
4725 | spin_unlock(&cache->lock); |
4726 | spin_unlock(&space_info->lock); | |
f0486c68 | 4727 | return ret; |
324ae4df | 4728 | } |
9078a3e1 | 4729 | |
11833d66 YZ |
4730 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4731 | struct btrfs_root *root) | |
e8569813 | 4732 | { |
e8569813 | 4733 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4734 | struct btrfs_caching_control *next; |
4735 | struct btrfs_caching_control *caching_ctl; | |
4736 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4737 | |
11833d66 | 4738 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4739 | |
11833d66 YZ |
4740 | list_for_each_entry_safe(caching_ctl, next, |
4741 | &fs_info->caching_block_groups, list) { | |
4742 | cache = caching_ctl->block_group; | |
4743 | if (block_group_cache_done(cache)) { | |
4744 | cache->last_byte_to_unpin = (u64)-1; | |
4745 | list_del_init(&caching_ctl->list); | |
4746 | put_caching_control(caching_ctl); | |
e8569813 | 4747 | } else { |
11833d66 | 4748 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4749 | } |
e8569813 | 4750 | } |
11833d66 YZ |
4751 | |
4752 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4753 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4754 | else | |
4755 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4756 | ||
4757 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4758 | |
4759 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4760 | return 0; |
4761 | } | |
4762 | ||
11833d66 | 4763 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4764 | { |
11833d66 YZ |
4765 | struct btrfs_fs_info *fs_info = root->fs_info; |
4766 | struct btrfs_block_group_cache *cache = NULL; | |
4767 | u64 len; | |
ccd467d6 | 4768 | |
11833d66 YZ |
4769 | while (start <= end) { |
4770 | if (!cache || | |
4771 | start >= cache->key.objectid + cache->key.offset) { | |
4772 | if (cache) | |
4773 | btrfs_put_block_group(cache); | |
4774 | cache = btrfs_lookup_block_group(fs_info, start); | |
4775 | BUG_ON(!cache); | |
4776 | } | |
4777 | ||
4778 | len = cache->key.objectid + cache->key.offset - start; | |
4779 | len = min(len, end + 1 - start); | |
4780 | ||
4781 | if (start < cache->last_byte_to_unpin) { | |
4782 | len = min(len, cache->last_byte_to_unpin - start); | |
4783 | btrfs_add_free_space(cache, start, len); | |
4784 | } | |
4785 | ||
f0486c68 YZ |
4786 | start += len; |
4787 | ||
11833d66 YZ |
4788 | spin_lock(&cache->space_info->lock); |
4789 | spin_lock(&cache->lock); | |
4790 | cache->pinned -= len; | |
4791 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4792 | if (cache->ro) |
f0486c68 | 4793 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4794 | spin_unlock(&cache->lock); |
4795 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4796 | } |
11833d66 YZ |
4797 | |
4798 | if (cache) | |
4799 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4800 | return 0; |
4801 | } | |
4802 | ||
4803 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4804 | struct btrfs_root *root) |
a28ec197 | 4805 | { |
11833d66 YZ |
4806 | struct btrfs_fs_info *fs_info = root->fs_info; |
4807 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
4808 | u64 start; |
4809 | u64 end; | |
a28ec197 | 4810 | int ret; |
a28ec197 | 4811 | |
11833d66 YZ |
4812 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4813 | unpin = &fs_info->freed_extents[1]; | |
4814 | else | |
4815 | unpin = &fs_info->freed_extents[0]; | |
4816 | ||
d397712b | 4817 | while (1) { |
1a5bc167 CM |
4818 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4819 | EXTENT_DIRTY); | |
4820 | if (ret) | |
a28ec197 | 4821 | break; |
1f3c79a2 | 4822 | |
5378e607 LD |
4823 | if (btrfs_test_opt(root, DISCARD)) |
4824 | ret = btrfs_discard_extent(root, start, | |
4825 | end + 1 - start, NULL); | |
1f3c79a2 | 4826 | |
1a5bc167 | 4827 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4828 | unpin_extent_range(root, start, end); |
b9473439 | 4829 | cond_resched(); |
a28ec197 | 4830 | } |
817d52f8 | 4831 | |
e20d96d6 CM |
4832 | return 0; |
4833 | } | |
4834 | ||
5d4f98a2 YZ |
4835 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4836 | struct btrfs_root *root, | |
4837 | u64 bytenr, u64 num_bytes, u64 parent, | |
4838 | u64 root_objectid, u64 owner_objectid, | |
4839 | u64 owner_offset, int refs_to_drop, | |
4840 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4841 | { |
e2fa7227 | 4842 | struct btrfs_key key; |
5d4f98a2 | 4843 | struct btrfs_path *path; |
1261ec42 CM |
4844 | struct btrfs_fs_info *info = root->fs_info; |
4845 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4846 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4847 | struct btrfs_extent_item *ei; |
4848 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4849 | int ret; |
5d4f98a2 | 4850 | int is_data; |
952fccac CM |
4851 | int extent_slot = 0; |
4852 | int found_extent = 0; | |
4853 | int num_to_del = 1; | |
5d4f98a2 YZ |
4854 | u32 item_size; |
4855 | u64 refs; | |
037e6390 | 4856 | |
5caf2a00 | 4857 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4858 | if (!path) |
4859 | return -ENOMEM; | |
5f26f772 | 4860 | |
3c12ac72 | 4861 | path->reada = 1; |
b9473439 | 4862 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4863 | |
4864 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4865 | BUG_ON(!is_data && refs_to_drop != 1); | |
4866 | ||
4867 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4868 | bytenr, num_bytes, parent, | |
4869 | root_objectid, owner_objectid, | |
4870 | owner_offset); | |
7bb86316 | 4871 | if (ret == 0) { |
952fccac | 4872 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4873 | while (extent_slot >= 0) { |
4874 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4875 | extent_slot); |
5d4f98a2 | 4876 | if (key.objectid != bytenr) |
952fccac | 4877 | break; |
5d4f98a2 YZ |
4878 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4879 | key.offset == num_bytes) { | |
952fccac CM |
4880 | found_extent = 1; |
4881 | break; | |
4882 | } | |
4883 | if (path->slots[0] - extent_slot > 5) | |
4884 | break; | |
5d4f98a2 | 4885 | extent_slot--; |
952fccac | 4886 | } |
5d4f98a2 YZ |
4887 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4888 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4889 | if (found_extent && item_size < sizeof(*ei)) | |
4890 | found_extent = 0; | |
4891 | #endif | |
31840ae1 | 4892 | if (!found_extent) { |
5d4f98a2 | 4893 | BUG_ON(iref); |
56bec294 | 4894 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4895 | NULL, refs_to_drop, |
4896 | is_data); | |
31840ae1 | 4897 | BUG_ON(ret); |
b3b4aa74 | 4898 | btrfs_release_path(path); |
b9473439 | 4899 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4900 | |
4901 | key.objectid = bytenr; | |
4902 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4903 | key.offset = num_bytes; | |
4904 | ||
31840ae1 ZY |
4905 | ret = btrfs_search_slot(trans, extent_root, |
4906 | &key, path, -1, 1); | |
f3465ca4 JB |
4907 | if (ret) { |
4908 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4909 | ", was looking for %llu\n", ret, |
4910 | (unsigned long long)bytenr); | |
b783e62d JB |
4911 | if (ret > 0) |
4912 | btrfs_print_leaf(extent_root, | |
4913 | path->nodes[0]); | |
f3465ca4 | 4914 | } |
31840ae1 ZY |
4915 | BUG_ON(ret); |
4916 | extent_slot = path->slots[0]; | |
4917 | } | |
7bb86316 CM |
4918 | } else { |
4919 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4920 | WARN_ON(1); | |
d397712b | 4921 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4922 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4923 | (unsigned long long)bytenr, |
56bec294 | 4924 | (unsigned long long)parent, |
d397712b | 4925 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4926 | (unsigned long long)owner_objectid, |
4927 | (unsigned long long)owner_offset); | |
7bb86316 | 4928 | } |
5f39d397 CM |
4929 | |
4930 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4931 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4932 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4933 | if (item_size < sizeof(*ei)) { | |
4934 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4935 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4936 | owner_objectid, 0); | |
4937 | BUG_ON(ret < 0); | |
4938 | ||
b3b4aa74 | 4939 | btrfs_release_path(path); |
5d4f98a2 YZ |
4940 | path->leave_spinning = 1; |
4941 | ||
4942 | key.objectid = bytenr; | |
4943 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4944 | key.offset = num_bytes; | |
4945 | ||
4946 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4947 | -1, 1); | |
4948 | if (ret) { | |
4949 | printk(KERN_ERR "umm, got %d back from search" | |
4950 | ", was looking for %llu\n", ret, | |
4951 | (unsigned long long)bytenr); | |
4952 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4953 | } | |
4954 | BUG_ON(ret); | |
4955 | extent_slot = path->slots[0]; | |
4956 | leaf = path->nodes[0]; | |
4957 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4958 | } | |
4959 | #endif | |
4960 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4961 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4962 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4963 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4964 | struct btrfs_tree_block_info *bi; | |
4965 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4966 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4967 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4968 | } | |
56bec294 | 4969 | |
5d4f98a2 | 4970 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4971 | BUG_ON(refs < refs_to_drop); |
4972 | refs -= refs_to_drop; | |
5f39d397 | 4973 | |
5d4f98a2 YZ |
4974 | if (refs > 0) { |
4975 | if (extent_op) | |
4976 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4977 | /* | |
4978 | * In the case of inline back ref, reference count will | |
4979 | * be updated by remove_extent_backref | |
952fccac | 4980 | */ |
5d4f98a2 YZ |
4981 | if (iref) { |
4982 | BUG_ON(!found_extent); | |
4983 | } else { | |
4984 | btrfs_set_extent_refs(leaf, ei, refs); | |
4985 | btrfs_mark_buffer_dirty(leaf); | |
4986 | } | |
4987 | if (found_extent) { | |
4988 | ret = remove_extent_backref(trans, extent_root, path, | |
4989 | iref, refs_to_drop, | |
4990 | is_data); | |
952fccac CM |
4991 | BUG_ON(ret); |
4992 | } | |
5d4f98a2 | 4993 | } else { |
5d4f98a2 YZ |
4994 | if (found_extent) { |
4995 | BUG_ON(is_data && refs_to_drop != | |
4996 | extent_data_ref_count(root, path, iref)); | |
4997 | if (iref) { | |
4998 | BUG_ON(path->slots[0] != extent_slot); | |
4999 | } else { | |
5000 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5001 | path->slots[0] = extent_slot; | |
5002 | num_to_del = 2; | |
5003 | } | |
78fae27e | 5004 | } |
b9473439 | 5005 | |
952fccac CM |
5006 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5007 | num_to_del); | |
31840ae1 | 5008 | BUG_ON(ret); |
b3b4aa74 | 5009 | btrfs_release_path(path); |
21af804c | 5010 | |
5d4f98a2 | 5011 | if (is_data) { |
459931ec CM |
5012 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
5013 | BUG_ON(ret); | |
5014 | } | |
5015 | ||
f0486c68 | 5016 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 5017 | BUG_ON(ret); |
a28ec197 | 5018 | } |
5caf2a00 | 5019 | btrfs_free_path(path); |
a28ec197 CM |
5020 | return ret; |
5021 | } | |
5022 | ||
1887be66 | 5023 | /* |
f0486c68 | 5024 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5025 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5026 | * a given extent, and if there are no other delayed refs to be processed, it | |
5027 | * removes it from the tree. | |
5028 | */ | |
5029 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5030 | struct btrfs_root *root, u64 bytenr) | |
5031 | { | |
5032 | struct btrfs_delayed_ref_head *head; | |
5033 | struct btrfs_delayed_ref_root *delayed_refs; | |
5034 | struct btrfs_delayed_ref_node *ref; | |
5035 | struct rb_node *node; | |
f0486c68 | 5036 | int ret = 0; |
1887be66 CM |
5037 | |
5038 | delayed_refs = &trans->transaction->delayed_refs; | |
5039 | spin_lock(&delayed_refs->lock); | |
5040 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5041 | if (!head) | |
5042 | goto out; | |
5043 | ||
5044 | node = rb_prev(&head->node.rb_node); | |
5045 | if (!node) | |
5046 | goto out; | |
5047 | ||
5048 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5049 | ||
5050 | /* there are still entries for this ref, we can't drop it */ | |
5051 | if (ref->bytenr == bytenr) | |
5052 | goto out; | |
5053 | ||
5d4f98a2 YZ |
5054 | if (head->extent_op) { |
5055 | if (!head->must_insert_reserved) | |
5056 | goto out; | |
5057 | kfree(head->extent_op); | |
5058 | head->extent_op = NULL; | |
5059 | } | |
5060 | ||
1887be66 CM |
5061 | /* |
5062 | * waiting for the lock here would deadlock. If someone else has it | |
5063 | * locked they are already in the process of dropping it anyway | |
5064 | */ | |
5065 | if (!mutex_trylock(&head->mutex)) | |
5066 | goto out; | |
5067 | ||
5068 | /* | |
5069 | * at this point we have a head with no other entries. Go | |
5070 | * ahead and process it. | |
5071 | */ | |
5072 | head->node.in_tree = 0; | |
5073 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5074 | |
1887be66 | 5075 | delayed_refs->num_entries--; |
a168650c JS |
5076 | if (waitqueue_active(&delayed_refs->seq_wait)) |
5077 | wake_up(&delayed_refs->seq_wait); | |
1887be66 CM |
5078 | |
5079 | /* | |
5080 | * we don't take a ref on the node because we're removing it from the | |
5081 | * tree, so we just steal the ref the tree was holding. | |
5082 | */ | |
c3e69d58 CM |
5083 | delayed_refs->num_heads--; |
5084 | if (list_empty(&head->cluster)) | |
5085 | delayed_refs->num_heads_ready--; | |
5086 | ||
5087 | list_del_init(&head->cluster); | |
1887be66 CM |
5088 | spin_unlock(&delayed_refs->lock); |
5089 | ||
f0486c68 YZ |
5090 | BUG_ON(head->extent_op); |
5091 | if (head->must_insert_reserved) | |
5092 | ret = 1; | |
5093 | ||
5094 | mutex_unlock(&head->mutex); | |
1887be66 | 5095 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5096 | return ret; |
1887be66 CM |
5097 | out: |
5098 | spin_unlock(&delayed_refs->lock); | |
5099 | return 0; | |
5100 | } | |
5101 | ||
f0486c68 YZ |
5102 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5103 | struct btrfs_root *root, | |
5104 | struct extent_buffer *buf, | |
66d7e7f0 | 5105 | u64 parent, int last_ref, int for_cow) |
f0486c68 | 5106 | { |
f0486c68 YZ |
5107 | struct btrfs_block_group_cache *cache = NULL; |
5108 | int ret; | |
5109 | ||
5110 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5111 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5112 | buf->start, buf->len, | |
5113 | parent, root->root_key.objectid, | |
5114 | btrfs_header_level(buf), | |
5115 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); | |
f0486c68 YZ |
5116 | BUG_ON(ret); |
5117 | } | |
5118 | ||
5119 | if (!last_ref) | |
5120 | return; | |
5121 | ||
f0486c68 | 5122 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5123 | |
5124 | if (btrfs_header_generation(buf) == trans->transid) { | |
5125 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5126 | ret = check_ref_cleanup(trans, root, buf->start); | |
5127 | if (!ret) | |
37be25bc | 5128 | goto out; |
f0486c68 YZ |
5129 | } |
5130 | ||
5131 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5132 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5133 | goto out; |
f0486c68 YZ |
5134 | } |
5135 | ||
5136 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5137 | ||
5138 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5139 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5140 | } |
5141 | out: | |
a826d6dc JB |
5142 | /* |
5143 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5144 | * anymore. | |
5145 | */ | |
5146 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5147 | btrfs_put_block_group(cache); |
5148 | } | |
5149 | ||
66d7e7f0 AJ |
5150 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5151 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5152 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5153 | { |
5154 | int ret; | |
66d7e7f0 | 5155 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5156 | |
56bec294 CM |
5157 | /* |
5158 | * tree log blocks never actually go into the extent allocation | |
5159 | * tree, just update pinning info and exit early. | |
56bec294 | 5160 | */ |
5d4f98a2 YZ |
5161 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5162 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5163 | /* unlocks the pinned mutex */ |
11833d66 | 5164 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5165 | ret = 0; |
5d4f98a2 | 5166 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5167 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5168 | num_bytes, | |
5d4f98a2 | 5169 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5170 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
1887be66 | 5171 | BUG_ON(ret); |
5d4f98a2 | 5172 | } else { |
66d7e7f0 AJ |
5173 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5174 | num_bytes, | |
5175 | parent, root_objectid, owner, | |
5176 | offset, BTRFS_DROP_DELAYED_REF, | |
5177 | NULL, for_cow); | |
5d4f98a2 | 5178 | BUG_ON(ret); |
56bec294 | 5179 | } |
925baedd CM |
5180 | return ret; |
5181 | } | |
5182 | ||
87ee04eb CM |
5183 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5184 | { | |
5185 | u64 mask = ((u64)root->stripesize - 1); | |
5186 | u64 ret = (val + mask) & ~mask; | |
5187 | return ret; | |
5188 | } | |
5189 | ||
817d52f8 JB |
5190 | /* |
5191 | * when we wait for progress in the block group caching, its because | |
5192 | * our allocation attempt failed at least once. So, we must sleep | |
5193 | * and let some progress happen before we try again. | |
5194 | * | |
5195 | * This function will sleep at least once waiting for new free space to | |
5196 | * show up, and then it will check the block group free space numbers | |
5197 | * for our min num_bytes. Another option is to have it go ahead | |
5198 | * and look in the rbtree for a free extent of a given size, but this | |
5199 | * is a good start. | |
5200 | */ | |
5201 | static noinline int | |
5202 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5203 | u64 num_bytes) | |
5204 | { | |
11833d66 | 5205 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5206 | DEFINE_WAIT(wait); |
5207 | ||
11833d66 YZ |
5208 | caching_ctl = get_caching_control(cache); |
5209 | if (!caching_ctl) | |
817d52f8 | 5210 | return 0; |
817d52f8 | 5211 | |
11833d66 | 5212 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5213 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5214 | |
5215 | put_caching_control(caching_ctl); | |
5216 | return 0; | |
5217 | } | |
5218 | ||
5219 | static noinline int | |
5220 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5221 | { | |
5222 | struct btrfs_caching_control *caching_ctl; | |
5223 | DEFINE_WAIT(wait); | |
5224 | ||
5225 | caching_ctl = get_caching_control(cache); | |
5226 | if (!caching_ctl) | |
5227 | return 0; | |
5228 | ||
5229 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5230 | ||
5231 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5232 | return 0; |
5233 | } | |
5234 | ||
b742bb82 YZ |
5235 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5236 | { | |
5237 | int index; | |
5238 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
5239 | index = 0; | |
5240 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
5241 | index = 1; | |
5242 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
5243 | index = 2; | |
5244 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
5245 | index = 3; | |
5246 | else | |
5247 | index = 4; | |
5248 | return index; | |
5249 | } | |
5250 | ||
817d52f8 | 5251 | enum btrfs_loop_type { |
285ff5af JB |
5252 | LOOP_CACHING_NOWAIT = 0, |
5253 | LOOP_CACHING_WAIT = 1, | |
5254 | LOOP_ALLOC_CHUNK = 2, | |
5255 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5256 | }; |
5257 | ||
fec577fb CM |
5258 | /* |
5259 | * walks the btree of allocated extents and find a hole of a given size. | |
5260 | * The key ins is changed to record the hole: | |
5261 | * ins->objectid == block start | |
62e2749e | 5262 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5263 | * ins->offset == number of blocks |
5264 | * Any available blocks before search_start are skipped. | |
5265 | */ | |
d397712b | 5266 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5267 | struct btrfs_root *orig_root, |
5268 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5269 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5270 | u64 data) |
fec577fb | 5271 | { |
80eb234a | 5272 | int ret = 0; |
d397712b | 5273 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5274 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5275 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5276 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5277 | u64 search_start = 0; |
239b14b3 | 5278 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5279 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5280 | int done_chunk_alloc = 0; |
80eb234a | 5281 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5282 | int loop = 0; |
f0486c68 | 5283 | int index = 0; |
fb25e914 JB |
5284 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5285 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5286 | bool found_uncached_bg = false; |
0a24325e | 5287 | bool failed_cluster_refill = false; |
1cdda9b8 | 5288 | bool failed_alloc = false; |
67377734 | 5289 | bool use_cluster = true; |
60d2adbb | 5290 | bool have_caching_bg = false; |
fec577fb | 5291 | |
db94535d | 5292 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5293 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5294 | ins->objectid = 0; |
5295 | ins->offset = 0; | |
b1a4d965 | 5296 | |
3f7de037 JB |
5297 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5298 | ||
2552d17e | 5299 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5300 | if (!space_info) { |
e0f54067 | 5301 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5302 | return -ENOSPC; |
5303 | } | |
2552d17e | 5304 | |
67377734 JB |
5305 | /* |
5306 | * If the space info is for both data and metadata it means we have a | |
5307 | * small filesystem and we can't use the clustering stuff. | |
5308 | */ | |
5309 | if (btrfs_mixed_space_info(space_info)) | |
5310 | use_cluster = false; | |
5311 | ||
0ef3e66b CM |
5312 | if (orig_root->ref_cows || empty_size) |
5313 | allowed_chunk_alloc = 1; | |
5314 | ||
67377734 | 5315 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5316 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5317 | if (!btrfs_test_opt(root, SSD)) |
5318 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5319 | } |
5320 | ||
67377734 JB |
5321 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5322 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5323 | last_ptr = &root->fs_info->data_alloc_cluster; |
5324 | } | |
0f9dd46c | 5325 | |
239b14b3 | 5326 | if (last_ptr) { |
fa9c0d79 CM |
5327 | spin_lock(&last_ptr->lock); |
5328 | if (last_ptr->block_group) | |
5329 | hint_byte = last_ptr->window_start; | |
5330 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5331 | } |
fa9c0d79 | 5332 | |
a061fc8d | 5333 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5334 | search_start = max(search_start, hint_byte); |
0b86a832 | 5335 | |
817d52f8 | 5336 | if (!last_ptr) |
fa9c0d79 | 5337 | empty_cluster = 0; |
fa9c0d79 | 5338 | |
2552d17e | 5339 | if (search_start == hint_byte) { |
2552d17e JB |
5340 | block_group = btrfs_lookup_block_group(root->fs_info, |
5341 | search_start); | |
274bd4fb | 5342 | used_block_group = block_group; |
817d52f8 JB |
5343 | /* |
5344 | * we don't want to use the block group if it doesn't match our | |
5345 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5346 | * |
5347 | * However if we are re-searching with an ideal block group | |
5348 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5349 | */ |
5350 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5351 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5352 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5353 | if (list_empty(&block_group->list) || |
5354 | block_group->ro) { | |
5355 | /* | |
5356 | * someone is removing this block group, | |
5357 | * we can't jump into the have_block_group | |
5358 | * target because our list pointers are not | |
5359 | * valid | |
5360 | */ | |
5361 | btrfs_put_block_group(block_group); | |
5362 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5363 | } else { |
b742bb82 | 5364 | index = get_block_group_index(block_group); |
44fb5511 | 5365 | goto have_block_group; |
ccf0e725 | 5366 | } |
2552d17e | 5367 | } else if (block_group) { |
fa9c0d79 | 5368 | btrfs_put_block_group(block_group); |
2552d17e | 5369 | } |
42e70e7a | 5370 | } |
2552d17e | 5371 | search: |
60d2adbb | 5372 | have_caching_bg = false; |
80eb234a | 5373 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5374 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5375 | list) { | |
6226cb0a | 5376 | u64 offset; |
817d52f8 | 5377 | int cached; |
8a1413a2 | 5378 | |
274bd4fb | 5379 | used_block_group = block_group; |
11dfe35a | 5380 | btrfs_get_block_group(block_group); |
2552d17e | 5381 | search_start = block_group->key.objectid; |
42e70e7a | 5382 | |
83a50de9 CM |
5383 | /* |
5384 | * this can happen if we end up cycling through all the | |
5385 | * raid types, but we want to make sure we only allocate | |
5386 | * for the proper type. | |
5387 | */ | |
5388 | if (!block_group_bits(block_group, data)) { | |
5389 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5390 | BTRFS_BLOCK_GROUP_RAID1 | | |
5391 | BTRFS_BLOCK_GROUP_RAID10; | |
5392 | ||
5393 | /* | |
5394 | * if they asked for extra copies and this block group | |
5395 | * doesn't provide them, bail. This does allow us to | |
5396 | * fill raid0 from raid1. | |
5397 | */ | |
5398 | if ((data & extra) && !(block_group->flags & extra)) | |
5399 | goto loop; | |
5400 | } | |
5401 | ||
2552d17e | 5402 | have_block_group: |
291c7d2f JB |
5403 | cached = block_group_cache_done(block_group); |
5404 | if (unlikely(!cached)) { | |
291c7d2f | 5405 | found_uncached_bg = true; |
b8399dee | 5406 | ret = cache_block_group(block_group, trans, |
285ff5af JB |
5407 | orig_root, 0); |
5408 | BUG_ON(ret); | |
817d52f8 JB |
5409 | } |
5410 | ||
ea6a478e | 5411 | if (unlikely(block_group->ro)) |
2552d17e | 5412 | goto loop; |
0f9dd46c | 5413 | |
0a24325e | 5414 | /* |
062c05c4 AO |
5415 | * Ok we want to try and use the cluster allocator, so |
5416 | * lets look there | |
0a24325e | 5417 | */ |
062c05c4 | 5418 | if (last_ptr) { |
fa9c0d79 CM |
5419 | /* |
5420 | * the refill lock keeps out other | |
5421 | * people trying to start a new cluster | |
5422 | */ | |
5423 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5424 | used_block_group = last_ptr->block_group; |
5425 | if (used_block_group != block_group && | |
5426 | (!used_block_group || | |
5427 | used_block_group->ro || | |
5428 | !block_group_bits(used_block_group, data))) { | |
5429 | used_block_group = block_group; | |
44fb5511 | 5430 | goto refill_cluster; |
274bd4fb AO |
5431 | } |
5432 | ||
5433 | if (used_block_group != block_group) | |
5434 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5435 | |
274bd4fb AO |
5436 | offset = btrfs_alloc_from_cluster(used_block_group, |
5437 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5438 | if (offset) { |
5439 | /* we have a block, we're done */ | |
5440 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5441 | trace_btrfs_reserve_extent_cluster(root, |
5442 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5443 | goto checks; |
5444 | } | |
5445 | ||
274bd4fb AO |
5446 | WARN_ON(last_ptr->block_group != used_block_group); |
5447 | if (used_block_group != block_group) { | |
5448 | btrfs_put_block_group(used_block_group); | |
5449 | used_block_group = block_group; | |
fa9c0d79 | 5450 | } |
44fb5511 | 5451 | refill_cluster: |
274bd4fb | 5452 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5453 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5454 | * set up a new clusters, so lets just skip it | |
5455 | * and let the allocator find whatever block | |
5456 | * it can find. If we reach this point, we | |
5457 | * will have tried the cluster allocator | |
5458 | * plenty of times and not have found | |
5459 | * anything, so we are likely way too | |
5460 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5461 | * anything. |
5462 | * | |
5463 | * However, if the cluster is taken from the | |
5464 | * current block group, release the cluster | |
5465 | * first, so that we stand a better chance of | |
5466 | * succeeding in the unclustered | |
5467 | * allocation. */ | |
5468 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5469 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5470 | spin_unlock(&last_ptr->refill_lock); |
5471 | goto unclustered_alloc; | |
5472 | } | |
5473 | ||
fa9c0d79 CM |
5474 | /* |
5475 | * this cluster didn't work out, free it and | |
5476 | * start over | |
5477 | */ | |
5478 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5479 | ||
a5f6f719 AO |
5480 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5481 | spin_unlock(&last_ptr->refill_lock); | |
5482 | goto unclustered_alloc; | |
5483 | } | |
5484 | ||
fa9c0d79 | 5485 | /* allocate a cluster in this block group */ |
451d7585 | 5486 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5487 | block_group, last_ptr, |
1b22bad7 | 5488 | search_start, num_bytes, |
fa9c0d79 CM |
5489 | empty_cluster + empty_size); |
5490 | if (ret == 0) { | |
5491 | /* | |
5492 | * now pull our allocation out of this | |
5493 | * cluster | |
5494 | */ | |
5495 | offset = btrfs_alloc_from_cluster(block_group, | |
5496 | last_ptr, num_bytes, | |
5497 | search_start); | |
5498 | if (offset) { | |
5499 | /* we found one, proceed */ | |
5500 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5501 | trace_btrfs_reserve_extent_cluster(root, |
5502 | block_group, search_start, | |
5503 | num_bytes); | |
fa9c0d79 CM |
5504 | goto checks; |
5505 | } | |
0a24325e JB |
5506 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5507 | && !failed_cluster_refill) { | |
817d52f8 JB |
5508 | spin_unlock(&last_ptr->refill_lock); |
5509 | ||
0a24325e | 5510 | failed_cluster_refill = true; |
817d52f8 JB |
5511 | wait_block_group_cache_progress(block_group, |
5512 | num_bytes + empty_cluster + empty_size); | |
5513 | goto have_block_group; | |
fa9c0d79 | 5514 | } |
817d52f8 | 5515 | |
fa9c0d79 CM |
5516 | /* |
5517 | * at this point we either didn't find a cluster | |
5518 | * or we weren't able to allocate a block from our | |
5519 | * cluster. Free the cluster we've been trying | |
5520 | * to use, and go to the next block group | |
5521 | */ | |
0a24325e | 5522 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5523 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5524 | goto loop; |
fa9c0d79 CM |
5525 | } |
5526 | ||
062c05c4 | 5527 | unclustered_alloc: |
a5f6f719 AO |
5528 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5529 | if (cached && | |
5530 | block_group->free_space_ctl->free_space < | |
5531 | num_bytes + empty_cluster + empty_size) { | |
5532 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5533 | goto loop; | |
5534 | } | |
5535 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5536 | ||
6226cb0a JB |
5537 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5538 | num_bytes, empty_size); | |
1cdda9b8 JB |
5539 | /* |
5540 | * If we didn't find a chunk, and we haven't failed on this | |
5541 | * block group before, and this block group is in the middle of | |
5542 | * caching and we are ok with waiting, then go ahead and wait | |
5543 | * for progress to be made, and set failed_alloc to true. | |
5544 | * | |
5545 | * If failed_alloc is true then we've already waited on this | |
5546 | * block group once and should move on to the next block group. | |
5547 | */ | |
5548 | if (!offset && !failed_alloc && !cached && | |
5549 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5550 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5551 | num_bytes + empty_size); |
5552 | failed_alloc = true; | |
817d52f8 | 5553 | goto have_block_group; |
1cdda9b8 | 5554 | } else if (!offset) { |
60d2adbb MX |
5555 | if (!cached) |
5556 | have_caching_bg = true; | |
1cdda9b8 | 5557 | goto loop; |
817d52f8 | 5558 | } |
fa9c0d79 | 5559 | checks: |
6226cb0a | 5560 | search_start = stripe_align(root, offset); |
25179201 | 5561 | |
2552d17e JB |
5562 | /* move on to the next group */ |
5563 | if (search_start + num_bytes > | |
274bd4fb AO |
5564 | used_block_group->key.objectid + used_block_group->key.offset) { |
5565 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5566 | goto loop; |
6226cb0a | 5567 | } |
f5a31e16 | 5568 | |
f0486c68 | 5569 | if (offset < search_start) |
274bd4fb | 5570 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5571 | search_start - offset); |
5572 | BUG_ON(offset > search_start); | |
2552d17e | 5573 | |
274bd4fb | 5574 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5575 | alloc_type); |
f0486c68 | 5576 | if (ret == -EAGAIN) { |
274bd4fb | 5577 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5578 | goto loop; |
0f9dd46c | 5579 | } |
0b86a832 | 5580 | |
f0486c68 | 5581 | /* we are all good, lets return */ |
2552d17e JB |
5582 | ins->objectid = search_start; |
5583 | ins->offset = num_bytes; | |
d2fb3437 | 5584 | |
3f7de037 JB |
5585 | trace_btrfs_reserve_extent(orig_root, block_group, |
5586 | search_start, num_bytes); | |
6226cb0a | 5587 | if (offset < search_start) |
274bd4fb | 5588 | btrfs_add_free_space(used_block_group, offset, |
6226cb0a JB |
5589 | search_start - offset); |
5590 | BUG_ON(offset > search_start); | |
274bd4fb AO |
5591 | if (used_block_group != block_group) |
5592 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5593 | btrfs_put_block_group(block_group); |
2552d17e JB |
5594 | break; |
5595 | loop: | |
0a24325e | 5596 | failed_cluster_refill = false; |
1cdda9b8 | 5597 | failed_alloc = false; |
b742bb82 | 5598 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5599 | if (used_block_group != block_group) |
5600 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5601 | btrfs_put_block_group(block_group); |
2552d17e JB |
5602 | } |
5603 | up_read(&space_info->groups_sem); | |
5604 | ||
60d2adbb MX |
5605 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5606 | goto search; | |
5607 | ||
b742bb82 YZ |
5608 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5609 | goto search; | |
5610 | ||
285ff5af | 5611 | /* |
ccf0e725 JB |
5612 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5613 | * caching kthreads as we move along | |
817d52f8 JB |
5614 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5615 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5616 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5617 | * again | |
fa9c0d79 | 5618 | */ |
723bda20 | 5619 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5620 | index = 0; |
723bda20 | 5621 | loop++; |
817d52f8 | 5622 | if (loop == LOOP_ALLOC_CHUNK) { |
723bda20 JB |
5623 | if (allowed_chunk_alloc) { |
5624 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5625 | 2 * 1024 * 1024, data, | |
5626 | CHUNK_ALLOC_LIMITED); | |
5627 | allowed_chunk_alloc = 0; | |
5628 | if (ret == 1) | |
5629 | done_chunk_alloc = 1; | |
5630 | } else if (!done_chunk_alloc && | |
5631 | space_info->force_alloc == | |
5632 | CHUNK_ALLOC_NO_FORCE) { | |
5633 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5634 | } | |
2552d17e | 5635 | |
723bda20 JB |
5636 | /* |
5637 | * We didn't allocate a chunk, go ahead and drop the | |
5638 | * empty size and loop again. | |
5639 | */ | |
5640 | if (!done_chunk_alloc) | |
5641 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5642 | } |
5643 | ||
723bda20 JB |
5644 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5645 | empty_size = 0; | |
5646 | empty_cluster = 0; | |
fa9c0d79 | 5647 | } |
723bda20 JB |
5648 | |
5649 | goto search; | |
2552d17e JB |
5650 | } else if (!ins->objectid) { |
5651 | ret = -ENOSPC; | |
d82a6f1d | 5652 | } else if (ins->objectid) { |
80eb234a | 5653 | ret = 0; |
be744175 | 5654 | } |
be744175 | 5655 | |
0f70abe2 | 5656 | return ret; |
fec577fb | 5657 | } |
ec44a35c | 5658 | |
9ed74f2d JB |
5659 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5660 | int dump_block_groups) | |
0f9dd46c JB |
5661 | { |
5662 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5663 | int index = 0; |
0f9dd46c | 5664 | |
9ed74f2d | 5665 | spin_lock(&info->lock); |
fb25e914 JB |
5666 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5667 | (unsigned long long)info->flags, | |
d397712b | 5668 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5669 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5670 | info->bytes_readonly), |
d397712b | 5671 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5672 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5673 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5674 | (unsigned long long)info->total_bytes, |
8929ecfa | 5675 | (unsigned long long)info->bytes_used, |
21380931 | 5676 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5677 | (unsigned long long)info->bytes_reserved, |
21380931 | 5678 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5679 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5680 | spin_unlock(&info->lock); |
5681 | ||
5682 | if (!dump_block_groups) | |
5683 | return; | |
0f9dd46c | 5684 | |
80eb234a | 5685 | down_read(&info->groups_sem); |
b742bb82 YZ |
5686 | again: |
5687 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5688 | spin_lock(&cache->lock); |
d397712b CM |
5689 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5690 | "%llu pinned %llu reserved\n", | |
5691 | (unsigned long long)cache->key.objectid, | |
5692 | (unsigned long long)cache->key.offset, | |
5693 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5694 | (unsigned long long)cache->pinned, | |
5695 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5696 | btrfs_dump_free_space(cache, bytes); |
5697 | spin_unlock(&cache->lock); | |
5698 | } | |
b742bb82 YZ |
5699 | if (++index < BTRFS_NR_RAID_TYPES) |
5700 | goto again; | |
80eb234a | 5701 | up_read(&info->groups_sem); |
0f9dd46c | 5702 | } |
e8569813 | 5703 | |
11833d66 YZ |
5704 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5705 | struct btrfs_root *root, | |
5706 | u64 num_bytes, u64 min_alloc_size, | |
5707 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 5708 | struct btrfs_key *ins, u64 data) |
fec577fb | 5709 | { |
9e622d6b | 5710 | bool final_tried = false; |
fec577fb | 5711 | int ret; |
925baedd | 5712 | |
6a63209f | 5713 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5714 | again: |
0ef3e66b CM |
5715 | /* |
5716 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5717 | * is not called recursively on allocations | |
5718 | */ | |
83d3c969 | 5719 | if (empty_size || root->ref_cows) |
6324fbf3 | 5720 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5721 | num_bytes + 2 * 1024 * 1024, data, |
5722 | CHUNK_ALLOC_NO_FORCE); | |
0b86a832 | 5723 | |
db94535d CM |
5724 | WARN_ON(num_bytes < root->sectorsize); |
5725 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 5726 | hint_byte, ins, data); |
3b951516 | 5727 | |
9e622d6b MX |
5728 | if (ret == -ENOSPC) { |
5729 | if (!final_tried) { | |
5730 | num_bytes = num_bytes >> 1; | |
5731 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
5732 | num_bytes = max(num_bytes, min_alloc_size); | |
5733 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
5734 | num_bytes, data, CHUNK_ALLOC_FORCE); | |
5735 | if (num_bytes == min_alloc_size) | |
5736 | final_tried = true; | |
5737 | goto again; | |
5738 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
5739 | struct btrfs_space_info *sinfo; | |
5740 | ||
5741 | sinfo = __find_space_info(root->fs_info, data); | |
5742 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
5743 | "wanted %llu\n", (unsigned long long)data, | |
5744 | (unsigned long long)num_bytes); | |
5745 | dump_space_info(sinfo, num_bytes, 1); | |
5746 | } | |
925baedd | 5747 | } |
0f9dd46c | 5748 | |
1abe9b8a | 5749 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5750 | ||
0f9dd46c | 5751 | return ret; |
e6dcd2dc CM |
5752 | } |
5753 | ||
e688b725 CM |
5754 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
5755 | u64 start, u64 len, int pin) | |
65b51a00 | 5756 | { |
0f9dd46c | 5757 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5758 | int ret = 0; |
0f9dd46c | 5759 | |
0f9dd46c JB |
5760 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5761 | if (!cache) { | |
d397712b CM |
5762 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5763 | (unsigned long long)start); | |
0f9dd46c JB |
5764 | return -ENOSPC; |
5765 | } | |
1f3c79a2 | 5766 | |
5378e607 LD |
5767 | if (btrfs_test_opt(root, DISCARD)) |
5768 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5769 | |
e688b725 CM |
5770 | if (pin) |
5771 | pin_down_extent(root, cache, start, len, 1); | |
5772 | else { | |
5773 | btrfs_add_free_space(cache, start, len); | |
5774 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
5775 | } | |
fa9c0d79 | 5776 | btrfs_put_block_group(cache); |
817d52f8 | 5777 | |
1abe9b8a | 5778 | trace_btrfs_reserved_extent_free(root, start, len); |
5779 | ||
e6dcd2dc CM |
5780 | return ret; |
5781 | } | |
5782 | ||
e688b725 CM |
5783 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
5784 | u64 start, u64 len) | |
5785 | { | |
5786 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
5787 | } | |
5788 | ||
5789 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
5790 | u64 start, u64 len) | |
5791 | { | |
5792 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
5793 | } | |
5794 | ||
5d4f98a2 YZ |
5795 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5796 | struct btrfs_root *root, | |
5797 | u64 parent, u64 root_objectid, | |
5798 | u64 flags, u64 owner, u64 offset, | |
5799 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5800 | { |
5801 | int ret; | |
5d4f98a2 | 5802 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5803 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5804 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5805 | struct btrfs_path *path; |
5d4f98a2 YZ |
5806 | struct extent_buffer *leaf; |
5807 | int type; | |
5808 | u32 size; | |
26b8003f | 5809 | |
5d4f98a2 YZ |
5810 | if (parent > 0) |
5811 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5812 | else | |
5813 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5814 | |
5d4f98a2 | 5815 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5816 | |
5817 | path = btrfs_alloc_path(); | |
db5b493a TI |
5818 | if (!path) |
5819 | return -ENOMEM; | |
47e4bb98 | 5820 | |
b9473439 | 5821 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5822 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5823 | ins, size); | |
ccd467d6 | 5824 | BUG_ON(ret); |
0f9dd46c | 5825 | |
5d4f98a2 YZ |
5826 | leaf = path->nodes[0]; |
5827 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5828 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5829 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5830 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5831 | btrfs_set_extent_flags(leaf, extent_item, | |
5832 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5833 | ||
5834 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5835 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5836 | if (parent > 0) { | |
5837 | struct btrfs_shared_data_ref *ref; | |
5838 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5839 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5840 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5841 | } else { | |
5842 | struct btrfs_extent_data_ref *ref; | |
5843 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5844 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5845 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5846 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5847 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5848 | } | |
47e4bb98 CM |
5849 | |
5850 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5851 | btrfs_free_path(path); |
f510cfec | 5852 | |
f0486c68 | 5853 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5854 | if (ret) { |
d397712b CM |
5855 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5856 | "%llu\n", (unsigned long long)ins->objectid, | |
5857 | (unsigned long long)ins->offset); | |
f5947066 CM |
5858 | BUG(); |
5859 | } | |
e6dcd2dc CM |
5860 | return ret; |
5861 | } | |
5862 | ||
5d4f98a2 YZ |
5863 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5864 | struct btrfs_root *root, | |
5865 | u64 parent, u64 root_objectid, | |
5866 | u64 flags, struct btrfs_disk_key *key, | |
5867 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5868 | { |
5869 | int ret; | |
5d4f98a2 YZ |
5870 | struct btrfs_fs_info *fs_info = root->fs_info; |
5871 | struct btrfs_extent_item *extent_item; | |
5872 | struct btrfs_tree_block_info *block_info; | |
5873 | struct btrfs_extent_inline_ref *iref; | |
5874 | struct btrfs_path *path; | |
5875 | struct extent_buffer *leaf; | |
5876 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5877 | |
5d4f98a2 | 5878 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5879 | if (!path) |
5880 | return -ENOMEM; | |
56bec294 | 5881 | |
5d4f98a2 YZ |
5882 | path->leave_spinning = 1; |
5883 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5884 | ins, size); | |
56bec294 | 5885 | BUG_ON(ret); |
5d4f98a2 YZ |
5886 | |
5887 | leaf = path->nodes[0]; | |
5888 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5889 | struct btrfs_extent_item); | |
5890 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5891 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5892 | btrfs_set_extent_flags(leaf, extent_item, | |
5893 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5894 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5895 | ||
5896 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5897 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5898 | ||
5899 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5900 | if (parent > 0) { | |
5901 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5902 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5903 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5904 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5905 | } else { | |
5906 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5907 | BTRFS_TREE_BLOCK_REF_KEY); | |
5908 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5909 | } | |
5910 | ||
5911 | btrfs_mark_buffer_dirty(leaf); | |
5912 | btrfs_free_path(path); | |
5913 | ||
f0486c68 | 5914 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5915 | if (ret) { |
5916 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5917 | "%llu\n", (unsigned long long)ins->objectid, | |
5918 | (unsigned long long)ins->offset); | |
5919 | BUG(); | |
5920 | } | |
5921 | return ret; | |
5922 | } | |
5923 | ||
5924 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5925 | struct btrfs_root *root, | |
5926 | u64 root_objectid, u64 owner, | |
5927 | u64 offset, struct btrfs_key *ins) | |
5928 | { | |
5929 | int ret; | |
5930 | ||
5931 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5932 | ||
66d7e7f0 AJ |
5933 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
5934 | ins->offset, 0, | |
5935 | root_objectid, owner, offset, | |
5936 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
5937 | return ret; |
5938 | } | |
e02119d5 CM |
5939 | |
5940 | /* | |
5941 | * this is used by the tree logging recovery code. It records that | |
5942 | * an extent has been allocated and makes sure to clear the free | |
5943 | * space cache bits as well | |
5944 | */ | |
5d4f98a2 YZ |
5945 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5946 | struct btrfs_root *root, | |
5947 | u64 root_objectid, u64 owner, u64 offset, | |
5948 | struct btrfs_key *ins) | |
e02119d5 CM |
5949 | { |
5950 | int ret; | |
5951 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5952 | struct btrfs_caching_control *caching_ctl; |
5953 | u64 start = ins->objectid; | |
5954 | u64 num_bytes = ins->offset; | |
e02119d5 | 5955 | |
e02119d5 | 5956 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5957 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5958 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5959 | |
11833d66 YZ |
5960 | if (!caching_ctl) { |
5961 | BUG_ON(!block_group_cache_done(block_group)); | |
5962 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5963 | BUG_ON(ret); | |
5964 | } else { | |
5965 | mutex_lock(&caching_ctl->mutex); | |
5966 | ||
5967 | if (start >= caching_ctl->progress) { | |
5968 | ret = add_excluded_extent(root, start, num_bytes); | |
5969 | BUG_ON(ret); | |
5970 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5971 | ret = btrfs_remove_free_space(block_group, | |
5972 | start, num_bytes); | |
5973 | BUG_ON(ret); | |
5974 | } else { | |
5975 | num_bytes = caching_ctl->progress - start; | |
5976 | ret = btrfs_remove_free_space(block_group, | |
5977 | start, num_bytes); | |
5978 | BUG_ON(ret); | |
5979 | ||
5980 | start = caching_ctl->progress; | |
5981 | num_bytes = ins->objectid + ins->offset - | |
5982 | caching_ctl->progress; | |
5983 | ret = add_excluded_extent(root, start, num_bytes); | |
5984 | BUG_ON(ret); | |
5985 | } | |
5986 | ||
5987 | mutex_unlock(&caching_ctl->mutex); | |
5988 | put_caching_control(caching_ctl); | |
5989 | } | |
5990 | ||
fb25e914 JB |
5991 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
5992 | RESERVE_ALLOC_NO_ACCOUNT); | |
f0486c68 | 5993 | BUG_ON(ret); |
fa9c0d79 | 5994 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
5995 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
5996 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
5997 | return ret; |
5998 | } | |
5999 | ||
65b51a00 CM |
6000 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6001 | struct btrfs_root *root, | |
4008c04a CM |
6002 | u64 bytenr, u32 blocksize, |
6003 | int level) | |
65b51a00 CM |
6004 | { |
6005 | struct extent_buffer *buf; | |
6006 | ||
6007 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6008 | if (!buf) | |
6009 | return ERR_PTR(-ENOMEM); | |
6010 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6011 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6012 | btrfs_tree_lock(buf); |
6013 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6014 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6015 | |
6016 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6017 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6018 | |
d0c803c4 | 6019 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6020 | /* |
6021 | * we allow two log transactions at a time, use different | |
6022 | * EXENT bit to differentiate dirty pages. | |
6023 | */ | |
6024 | if (root->log_transid % 2 == 0) | |
6025 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6026 | buf->start + buf->len - 1, GFP_NOFS); | |
6027 | else | |
6028 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6029 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6030 | } else { |
6031 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6032 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6033 | } |
65b51a00 | 6034 | trans->blocks_used++; |
b4ce94de | 6035 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6036 | return buf; |
6037 | } | |
6038 | ||
f0486c68 YZ |
6039 | static struct btrfs_block_rsv * |
6040 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6041 | struct btrfs_root *root, u32 blocksize) | |
6042 | { | |
6043 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6044 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6045 | int ret; |
6046 | ||
6047 | block_rsv = get_block_rsv(trans, root); | |
6048 | ||
6049 | if (block_rsv->size == 0) { | |
36ba022a | 6050 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
6051 | /* |
6052 | * If we couldn't reserve metadata bytes try and use some from | |
6053 | * the global reserve. | |
6054 | */ | |
6055 | if (ret && block_rsv != global_rsv) { | |
6056 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6057 | if (!ret) | |
6058 | return global_rsv; | |
f0486c68 | 6059 | return ERR_PTR(ret); |
68a82277 | 6060 | } else if (ret) { |
f0486c68 | 6061 | return ERR_PTR(ret); |
68a82277 | 6062 | } |
f0486c68 YZ |
6063 | return block_rsv; |
6064 | } | |
6065 | ||
6066 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6067 | if (!ret) | |
6068 | return block_rsv; | |
68a82277 | 6069 | if (ret) { |
dff51cd1 DS |
6070 | static DEFINE_RATELIMIT_STATE(_rs, |
6071 | DEFAULT_RATELIMIT_INTERVAL, | |
6072 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
6073 | if (__ratelimit(&_rs)) { | |
6074 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
6075 | WARN_ON(1); | |
6076 | } | |
36ba022a | 6077 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 6078 | if (!ret) { |
68a82277 JB |
6079 | return block_rsv; |
6080 | } else if (ret && block_rsv != global_rsv) { | |
6081 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6082 | if (!ret) | |
6083 | return global_rsv; | |
6084 | } | |
6085 | } | |
f0486c68 | 6086 | |
f0486c68 YZ |
6087 | return ERR_PTR(-ENOSPC); |
6088 | } | |
6089 | ||
8c2a3ca2 JB |
6090 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6091 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6092 | { |
6093 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6094 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6095 | } |
6096 | ||
fec577fb | 6097 | /* |
f0486c68 YZ |
6098 | * finds a free extent and does all the dirty work required for allocation |
6099 | * returns the key for the extent through ins, and a tree buffer for | |
6100 | * the first block of the extent through buf. | |
6101 | * | |
fec577fb CM |
6102 | * returns the tree buffer or NULL. |
6103 | */ | |
5f39d397 | 6104 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6105 | struct btrfs_root *root, u32 blocksize, |
6106 | u64 parent, u64 root_objectid, | |
6107 | struct btrfs_disk_key *key, int level, | |
66d7e7f0 | 6108 | u64 hint, u64 empty_size, int for_cow) |
fec577fb | 6109 | { |
e2fa7227 | 6110 | struct btrfs_key ins; |
f0486c68 | 6111 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6112 | struct extent_buffer *buf; |
f0486c68 YZ |
6113 | u64 flags = 0; |
6114 | int ret; | |
6115 | ||
fec577fb | 6116 | |
f0486c68 YZ |
6117 | block_rsv = use_block_rsv(trans, root, blocksize); |
6118 | if (IS_ERR(block_rsv)) | |
6119 | return ERR_CAST(block_rsv); | |
6120 | ||
6121 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6122 | empty_size, hint, &ins, 0); |
fec577fb | 6123 | if (ret) { |
8c2a3ca2 | 6124 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6125 | return ERR_PTR(ret); |
fec577fb | 6126 | } |
55c69072 | 6127 | |
4008c04a CM |
6128 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6129 | blocksize, level); | |
f0486c68 YZ |
6130 | BUG_ON(IS_ERR(buf)); |
6131 | ||
6132 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6133 | if (parent == 0) | |
6134 | parent = ins.objectid; | |
6135 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6136 | } else | |
6137 | BUG_ON(parent > 0); | |
6138 | ||
6139 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6140 | struct btrfs_delayed_extent_op *extent_op; | |
6141 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
6142 | BUG_ON(!extent_op); | |
6143 | if (key) | |
6144 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6145 | else | |
6146 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6147 | extent_op->flags_to_set = flags; | |
6148 | extent_op->update_key = 1; | |
6149 | extent_op->update_flags = 1; | |
6150 | extent_op->is_data = 0; | |
6151 | ||
66d7e7f0 AJ |
6152 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6153 | ins.objectid, | |
f0486c68 YZ |
6154 | ins.offset, parent, root_objectid, |
6155 | level, BTRFS_ADD_DELAYED_EXTENT, | |
66d7e7f0 | 6156 | extent_op, for_cow); |
f0486c68 YZ |
6157 | BUG_ON(ret); |
6158 | } | |
fec577fb CM |
6159 | return buf; |
6160 | } | |
a28ec197 | 6161 | |
2c47e605 YZ |
6162 | struct walk_control { |
6163 | u64 refs[BTRFS_MAX_LEVEL]; | |
6164 | u64 flags[BTRFS_MAX_LEVEL]; | |
6165 | struct btrfs_key update_progress; | |
6166 | int stage; | |
6167 | int level; | |
6168 | int shared_level; | |
6169 | int update_ref; | |
6170 | int keep_locks; | |
1c4850e2 YZ |
6171 | int reada_slot; |
6172 | int reada_count; | |
66d7e7f0 | 6173 | int for_reloc; |
2c47e605 YZ |
6174 | }; |
6175 | ||
6176 | #define DROP_REFERENCE 1 | |
6177 | #define UPDATE_BACKREF 2 | |
6178 | ||
1c4850e2 YZ |
6179 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6180 | struct btrfs_root *root, | |
6181 | struct walk_control *wc, | |
6182 | struct btrfs_path *path) | |
6407bf6d | 6183 | { |
1c4850e2 YZ |
6184 | u64 bytenr; |
6185 | u64 generation; | |
6186 | u64 refs; | |
94fcca9f | 6187 | u64 flags; |
5d4f98a2 | 6188 | u32 nritems; |
1c4850e2 YZ |
6189 | u32 blocksize; |
6190 | struct btrfs_key key; | |
6191 | struct extent_buffer *eb; | |
6407bf6d | 6192 | int ret; |
1c4850e2 YZ |
6193 | int slot; |
6194 | int nread = 0; | |
6407bf6d | 6195 | |
1c4850e2 YZ |
6196 | if (path->slots[wc->level] < wc->reada_slot) { |
6197 | wc->reada_count = wc->reada_count * 2 / 3; | |
6198 | wc->reada_count = max(wc->reada_count, 2); | |
6199 | } else { | |
6200 | wc->reada_count = wc->reada_count * 3 / 2; | |
6201 | wc->reada_count = min_t(int, wc->reada_count, | |
6202 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6203 | } | |
7bb86316 | 6204 | |
1c4850e2 YZ |
6205 | eb = path->nodes[wc->level]; |
6206 | nritems = btrfs_header_nritems(eb); | |
6207 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6208 | |
1c4850e2 YZ |
6209 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6210 | if (nread >= wc->reada_count) | |
6211 | break; | |
bd56b302 | 6212 | |
2dd3e67b | 6213 | cond_resched(); |
1c4850e2 YZ |
6214 | bytenr = btrfs_node_blockptr(eb, slot); |
6215 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6216 | |
1c4850e2 YZ |
6217 | if (slot == path->slots[wc->level]) |
6218 | goto reada; | |
5d4f98a2 | 6219 | |
1c4850e2 YZ |
6220 | if (wc->stage == UPDATE_BACKREF && |
6221 | generation <= root->root_key.offset) | |
bd56b302 CM |
6222 | continue; |
6223 | ||
94fcca9f YZ |
6224 | /* We don't lock the tree block, it's OK to be racy here */ |
6225 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6226 | &refs, &flags); | |
6227 | BUG_ON(ret); | |
6228 | BUG_ON(refs == 0); | |
6229 | ||
1c4850e2 | 6230 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6231 | if (refs == 1) |
6232 | goto reada; | |
bd56b302 | 6233 | |
94fcca9f YZ |
6234 | if (wc->level == 1 && |
6235 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6236 | continue; | |
1c4850e2 YZ |
6237 | if (!wc->update_ref || |
6238 | generation <= root->root_key.offset) | |
6239 | continue; | |
6240 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6241 | ret = btrfs_comp_cpu_keys(&key, | |
6242 | &wc->update_progress); | |
6243 | if (ret < 0) | |
6244 | continue; | |
94fcca9f YZ |
6245 | } else { |
6246 | if (wc->level == 1 && | |
6247 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6248 | continue; | |
6407bf6d | 6249 | } |
1c4850e2 YZ |
6250 | reada: |
6251 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6252 | generation); | |
6253 | if (ret) | |
bd56b302 | 6254 | break; |
1c4850e2 | 6255 | nread++; |
20524f02 | 6256 | } |
1c4850e2 | 6257 | wc->reada_slot = slot; |
20524f02 | 6258 | } |
2c47e605 | 6259 | |
f82d02d9 | 6260 | /* |
2c47e605 YZ |
6261 | * hepler to process tree block while walking down the tree. |
6262 | * | |
2c47e605 YZ |
6263 | * when wc->stage == UPDATE_BACKREF, this function updates |
6264 | * back refs for pointers in the block. | |
6265 | * | |
6266 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6267 | */ |
2c47e605 | 6268 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6269 | struct btrfs_root *root, |
2c47e605 | 6270 | struct btrfs_path *path, |
94fcca9f | 6271 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6272 | { |
2c47e605 YZ |
6273 | int level = wc->level; |
6274 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6275 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6276 | int ret; |
6277 | ||
2c47e605 YZ |
6278 | if (wc->stage == UPDATE_BACKREF && |
6279 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6280 | return 1; | |
f82d02d9 | 6281 | |
2c47e605 YZ |
6282 | /* |
6283 | * when reference count of tree block is 1, it won't increase | |
6284 | * again. once full backref flag is set, we never clear it. | |
6285 | */ | |
94fcca9f YZ |
6286 | if (lookup_info && |
6287 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6288 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6289 | BUG_ON(!path->locks[level]); |
6290 | ret = btrfs_lookup_extent_info(trans, root, | |
6291 | eb->start, eb->len, | |
6292 | &wc->refs[level], | |
6293 | &wc->flags[level]); | |
6294 | BUG_ON(ret); | |
6295 | BUG_ON(wc->refs[level] == 0); | |
6296 | } | |
5d4f98a2 | 6297 | |
2c47e605 YZ |
6298 | if (wc->stage == DROP_REFERENCE) { |
6299 | if (wc->refs[level] > 1) | |
6300 | return 1; | |
f82d02d9 | 6301 | |
2c47e605 | 6302 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6303 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6304 | path->locks[level] = 0; |
6305 | } | |
6306 | return 0; | |
6307 | } | |
f82d02d9 | 6308 | |
2c47e605 YZ |
6309 | /* wc->stage == UPDATE_BACKREF */ |
6310 | if (!(wc->flags[level] & flag)) { | |
6311 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6312 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
f82d02d9 | 6313 | BUG_ON(ret); |
66d7e7f0 | 6314 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
2c47e605 YZ |
6315 | BUG_ON(ret); |
6316 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
6317 | eb->len, flag, 0); | |
6318 | BUG_ON(ret); | |
6319 | wc->flags[level] |= flag; | |
6320 | } | |
6321 | ||
6322 | /* | |
6323 | * the block is shared by multiple trees, so it's not good to | |
6324 | * keep the tree lock | |
6325 | */ | |
6326 | if (path->locks[level] && level > 0) { | |
bd681513 | 6327 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6328 | path->locks[level] = 0; |
6329 | } | |
6330 | return 0; | |
6331 | } | |
6332 | ||
1c4850e2 YZ |
6333 | /* |
6334 | * hepler to process tree block pointer. | |
6335 | * | |
6336 | * when wc->stage == DROP_REFERENCE, this function checks | |
6337 | * reference count of the block pointed to. if the block | |
6338 | * is shared and we need update back refs for the subtree | |
6339 | * rooted at the block, this function changes wc->stage to | |
6340 | * UPDATE_BACKREF. if the block is shared and there is no | |
6341 | * need to update back, this function drops the reference | |
6342 | * to the block. | |
6343 | * | |
6344 | * NOTE: return value 1 means we should stop walking down. | |
6345 | */ | |
6346 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6347 | struct btrfs_root *root, | |
6348 | struct btrfs_path *path, | |
94fcca9f | 6349 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6350 | { |
6351 | u64 bytenr; | |
6352 | u64 generation; | |
6353 | u64 parent; | |
6354 | u32 blocksize; | |
6355 | struct btrfs_key key; | |
6356 | struct extent_buffer *next; | |
6357 | int level = wc->level; | |
6358 | int reada = 0; | |
6359 | int ret = 0; | |
6360 | ||
6361 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6362 | path->slots[level]); | |
6363 | /* | |
6364 | * if the lower level block was created before the snapshot | |
6365 | * was created, we know there is no need to update back refs | |
6366 | * for the subtree | |
6367 | */ | |
6368 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6369 | generation <= root->root_key.offset) { |
6370 | *lookup_info = 1; | |
1c4850e2 | 6371 | return 1; |
94fcca9f | 6372 | } |
1c4850e2 YZ |
6373 | |
6374 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6375 | blocksize = btrfs_level_size(root, level - 1); | |
6376 | ||
6377 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6378 | if (!next) { | |
6379 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6380 | if (!next) |
6381 | return -ENOMEM; | |
1c4850e2 YZ |
6382 | reada = 1; |
6383 | } | |
6384 | btrfs_tree_lock(next); | |
6385 | btrfs_set_lock_blocking(next); | |
6386 | ||
94fcca9f YZ |
6387 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6388 | &wc->refs[level - 1], | |
6389 | &wc->flags[level - 1]); | |
6390 | BUG_ON(ret); | |
6391 | BUG_ON(wc->refs[level - 1] == 0); | |
6392 | *lookup_info = 0; | |
1c4850e2 | 6393 | |
94fcca9f | 6394 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6395 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6396 | if (level == 1 && |
6397 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6398 | goto skip; | |
6399 | ||
1c4850e2 YZ |
6400 | if (!wc->update_ref || |
6401 | generation <= root->root_key.offset) | |
6402 | goto skip; | |
6403 | ||
6404 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6405 | path->slots[level]); | |
6406 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6407 | if (ret < 0) | |
6408 | goto skip; | |
6409 | ||
6410 | wc->stage = UPDATE_BACKREF; | |
6411 | wc->shared_level = level - 1; | |
6412 | } | |
94fcca9f YZ |
6413 | } else { |
6414 | if (level == 1 && | |
6415 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6416 | goto skip; | |
1c4850e2 YZ |
6417 | } |
6418 | ||
6419 | if (!btrfs_buffer_uptodate(next, generation)) { | |
6420 | btrfs_tree_unlock(next); | |
6421 | free_extent_buffer(next); | |
6422 | next = NULL; | |
94fcca9f | 6423 | *lookup_info = 1; |
1c4850e2 YZ |
6424 | } |
6425 | ||
6426 | if (!next) { | |
6427 | if (reada && level == 1) | |
6428 | reada_walk_down(trans, root, wc, path); | |
6429 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6430 | if (!next) |
6431 | return -EIO; | |
1c4850e2 YZ |
6432 | btrfs_tree_lock(next); |
6433 | btrfs_set_lock_blocking(next); | |
6434 | } | |
6435 | ||
6436 | level--; | |
6437 | BUG_ON(level != btrfs_header_level(next)); | |
6438 | path->nodes[level] = next; | |
6439 | path->slots[level] = 0; | |
bd681513 | 6440 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6441 | wc->level = level; |
6442 | if (wc->level == 1) | |
6443 | wc->reada_slot = 0; | |
6444 | return 0; | |
6445 | skip: | |
6446 | wc->refs[level - 1] = 0; | |
6447 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6448 | if (wc->stage == DROP_REFERENCE) { |
6449 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6450 | parent = path->nodes[level]->start; | |
6451 | } else { | |
6452 | BUG_ON(root->root_key.objectid != | |
6453 | btrfs_header_owner(path->nodes[level])); | |
6454 | parent = 0; | |
6455 | } | |
1c4850e2 | 6456 | |
94fcca9f | 6457 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6458 | root->root_key.objectid, level - 1, 0, 0); |
94fcca9f | 6459 | BUG_ON(ret); |
1c4850e2 | 6460 | } |
1c4850e2 YZ |
6461 | btrfs_tree_unlock(next); |
6462 | free_extent_buffer(next); | |
94fcca9f | 6463 | *lookup_info = 1; |
1c4850e2 YZ |
6464 | return 1; |
6465 | } | |
6466 | ||
2c47e605 YZ |
6467 | /* |
6468 | * hepler to process tree block while walking up the tree. | |
6469 | * | |
6470 | * when wc->stage == DROP_REFERENCE, this function drops | |
6471 | * reference count on the block. | |
6472 | * | |
6473 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6474 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6475 | * to UPDATE_BACKREF previously while processing the block. | |
6476 | * | |
6477 | * NOTE: return value 1 means we should stop walking up. | |
6478 | */ | |
6479 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6480 | struct btrfs_root *root, | |
6481 | struct btrfs_path *path, | |
6482 | struct walk_control *wc) | |
6483 | { | |
f0486c68 | 6484 | int ret; |
2c47e605 YZ |
6485 | int level = wc->level; |
6486 | struct extent_buffer *eb = path->nodes[level]; | |
6487 | u64 parent = 0; | |
6488 | ||
6489 | if (wc->stage == UPDATE_BACKREF) { | |
6490 | BUG_ON(wc->shared_level < level); | |
6491 | if (level < wc->shared_level) | |
6492 | goto out; | |
6493 | ||
2c47e605 YZ |
6494 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6495 | if (ret > 0) | |
6496 | wc->update_ref = 0; | |
6497 | ||
6498 | wc->stage = DROP_REFERENCE; | |
6499 | wc->shared_level = -1; | |
6500 | path->slots[level] = 0; | |
6501 | ||
6502 | /* | |
6503 | * check reference count again if the block isn't locked. | |
6504 | * we should start walking down the tree again if reference | |
6505 | * count is one. | |
6506 | */ | |
6507 | if (!path->locks[level]) { | |
6508 | BUG_ON(level == 0); | |
6509 | btrfs_tree_lock(eb); | |
6510 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6511 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6512 | |
6513 | ret = btrfs_lookup_extent_info(trans, root, | |
6514 | eb->start, eb->len, | |
6515 | &wc->refs[level], | |
6516 | &wc->flags[level]); | |
f82d02d9 | 6517 | BUG_ON(ret); |
2c47e605 YZ |
6518 | BUG_ON(wc->refs[level] == 0); |
6519 | if (wc->refs[level] == 1) { | |
bd681513 | 6520 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6521 | return 1; |
6522 | } | |
f82d02d9 | 6523 | } |
2c47e605 | 6524 | } |
f82d02d9 | 6525 | |
2c47e605 YZ |
6526 | /* wc->stage == DROP_REFERENCE */ |
6527 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6528 | |
2c47e605 YZ |
6529 | if (wc->refs[level] == 1) { |
6530 | if (level == 0) { | |
6531 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6532 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6533 | wc->for_reloc); | |
2c47e605 | 6534 | else |
66d7e7f0 AJ |
6535 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6536 | wc->for_reloc); | |
2c47e605 YZ |
6537 | BUG_ON(ret); |
6538 | } | |
6539 | /* make block locked assertion in clean_tree_block happy */ | |
6540 | if (!path->locks[level] && | |
6541 | btrfs_header_generation(eb) == trans->transid) { | |
6542 | btrfs_tree_lock(eb); | |
6543 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6544 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6545 | } |
6546 | clean_tree_block(trans, root, eb); | |
6547 | } | |
6548 | ||
6549 | if (eb == root->node) { | |
6550 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6551 | parent = eb->start; | |
6552 | else | |
6553 | BUG_ON(root->root_key.objectid != | |
6554 | btrfs_header_owner(eb)); | |
6555 | } else { | |
6556 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6557 | parent = path->nodes[level + 1]->start; | |
6558 | else | |
6559 | BUG_ON(root->root_key.objectid != | |
6560 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6561 | } |
f82d02d9 | 6562 | |
66d7e7f0 | 6563 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0); |
2c47e605 YZ |
6564 | out: |
6565 | wc->refs[level] = 0; | |
6566 | wc->flags[level] = 0; | |
f0486c68 | 6567 | return 0; |
2c47e605 YZ |
6568 | } |
6569 | ||
6570 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6571 | struct btrfs_root *root, | |
6572 | struct btrfs_path *path, | |
6573 | struct walk_control *wc) | |
6574 | { | |
2c47e605 | 6575 | int level = wc->level; |
94fcca9f | 6576 | int lookup_info = 1; |
2c47e605 YZ |
6577 | int ret; |
6578 | ||
6579 | while (level >= 0) { | |
94fcca9f | 6580 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6581 | if (ret > 0) |
6582 | break; | |
6583 | ||
6584 | if (level == 0) | |
6585 | break; | |
6586 | ||
7a7965f8 YZ |
6587 | if (path->slots[level] >= |
6588 | btrfs_header_nritems(path->nodes[level])) | |
6589 | break; | |
6590 | ||
94fcca9f | 6591 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6592 | if (ret > 0) { |
6593 | path->slots[level]++; | |
6594 | continue; | |
90d2c51d MX |
6595 | } else if (ret < 0) |
6596 | return ret; | |
1c4850e2 | 6597 | level = wc->level; |
f82d02d9 | 6598 | } |
f82d02d9 YZ |
6599 | return 0; |
6600 | } | |
6601 | ||
d397712b | 6602 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6603 | struct btrfs_root *root, |
f82d02d9 | 6604 | struct btrfs_path *path, |
2c47e605 | 6605 | struct walk_control *wc, int max_level) |
20524f02 | 6606 | { |
2c47e605 | 6607 | int level = wc->level; |
20524f02 | 6608 | int ret; |
9f3a7427 | 6609 | |
2c47e605 YZ |
6610 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6611 | while (level < max_level && path->nodes[level]) { | |
6612 | wc->level = level; | |
6613 | if (path->slots[level] + 1 < | |
6614 | btrfs_header_nritems(path->nodes[level])) { | |
6615 | path->slots[level]++; | |
20524f02 CM |
6616 | return 0; |
6617 | } else { | |
2c47e605 YZ |
6618 | ret = walk_up_proc(trans, root, path, wc); |
6619 | if (ret > 0) | |
6620 | return 0; | |
bd56b302 | 6621 | |
2c47e605 | 6622 | if (path->locks[level]) { |
bd681513 CM |
6623 | btrfs_tree_unlock_rw(path->nodes[level], |
6624 | path->locks[level]); | |
2c47e605 | 6625 | path->locks[level] = 0; |
f82d02d9 | 6626 | } |
2c47e605 YZ |
6627 | free_extent_buffer(path->nodes[level]); |
6628 | path->nodes[level] = NULL; | |
6629 | level++; | |
20524f02 CM |
6630 | } |
6631 | } | |
6632 | return 1; | |
6633 | } | |
6634 | ||
9aca1d51 | 6635 | /* |
2c47e605 YZ |
6636 | * drop a subvolume tree. |
6637 | * | |
6638 | * this function traverses the tree freeing any blocks that only | |
6639 | * referenced by the tree. | |
6640 | * | |
6641 | * when a shared tree block is found. this function decreases its | |
6642 | * reference count by one. if update_ref is true, this function | |
6643 | * also make sure backrefs for the shared block and all lower level | |
6644 | * blocks are properly updated. | |
9aca1d51 | 6645 | */ |
cb1b69f4 | 6646 | void btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6647 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6648 | int for_reloc) | |
20524f02 | 6649 | { |
5caf2a00 | 6650 | struct btrfs_path *path; |
2c47e605 YZ |
6651 | struct btrfs_trans_handle *trans; |
6652 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6653 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6654 | struct walk_control *wc; |
6655 | struct btrfs_key key; | |
6656 | int err = 0; | |
6657 | int ret; | |
6658 | int level; | |
20524f02 | 6659 | |
5caf2a00 | 6660 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6661 | if (!path) { |
6662 | err = -ENOMEM; | |
6663 | goto out; | |
6664 | } | |
20524f02 | 6665 | |
2c47e605 | 6666 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6667 | if (!wc) { |
6668 | btrfs_free_path(path); | |
cb1b69f4 TI |
6669 | err = -ENOMEM; |
6670 | goto out; | |
38a1a919 | 6671 | } |
2c47e605 | 6672 | |
a22285a6 | 6673 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 TI |
6674 | BUG_ON(IS_ERR(trans)); |
6675 | ||
3fd0a558 YZ |
6676 | if (block_rsv) |
6677 | trans->block_rsv = block_rsv; | |
2c47e605 | 6678 | |
9f3a7427 | 6679 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6680 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6681 | path->nodes[level] = btrfs_lock_root_node(root); |
6682 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6683 | path->slots[level] = 0; |
bd681513 | 6684 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6685 | memset(&wc->update_progress, 0, |
6686 | sizeof(wc->update_progress)); | |
9f3a7427 | 6687 | } else { |
9f3a7427 | 6688 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6689 | memcpy(&wc->update_progress, &key, |
6690 | sizeof(wc->update_progress)); | |
6691 | ||
6702ed49 | 6692 | level = root_item->drop_level; |
2c47e605 | 6693 | BUG_ON(level == 0); |
6702ed49 | 6694 | path->lowest_level = level; |
2c47e605 YZ |
6695 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6696 | path->lowest_level = 0; | |
6697 | if (ret < 0) { | |
6698 | err = ret; | |
cb1b69f4 | 6699 | goto out_free; |
9f3a7427 | 6700 | } |
1c4850e2 | 6701 | WARN_ON(ret > 0); |
2c47e605 | 6702 | |
7d9eb12c CM |
6703 | /* |
6704 | * unlock our path, this is safe because only this | |
6705 | * function is allowed to delete this snapshot | |
6706 | */ | |
5d4f98a2 | 6707 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6708 | |
6709 | level = btrfs_header_level(root->node); | |
6710 | while (1) { | |
6711 | btrfs_tree_lock(path->nodes[level]); | |
6712 | btrfs_set_lock_blocking(path->nodes[level]); | |
6713 | ||
6714 | ret = btrfs_lookup_extent_info(trans, root, | |
6715 | path->nodes[level]->start, | |
6716 | path->nodes[level]->len, | |
6717 | &wc->refs[level], | |
6718 | &wc->flags[level]); | |
6719 | BUG_ON(ret); | |
6720 | BUG_ON(wc->refs[level] == 0); | |
6721 | ||
6722 | if (level == root_item->drop_level) | |
6723 | break; | |
6724 | ||
6725 | btrfs_tree_unlock(path->nodes[level]); | |
6726 | WARN_ON(wc->refs[level] != 1); | |
6727 | level--; | |
6728 | } | |
9f3a7427 | 6729 | } |
2c47e605 YZ |
6730 | |
6731 | wc->level = level; | |
6732 | wc->shared_level = -1; | |
6733 | wc->stage = DROP_REFERENCE; | |
6734 | wc->update_ref = update_ref; | |
6735 | wc->keep_locks = 0; | |
66d7e7f0 | 6736 | wc->for_reloc = for_reloc; |
1c4850e2 | 6737 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6738 | |
d397712b | 6739 | while (1) { |
2c47e605 YZ |
6740 | ret = walk_down_tree(trans, root, path, wc); |
6741 | if (ret < 0) { | |
6742 | err = ret; | |
20524f02 | 6743 | break; |
2c47e605 | 6744 | } |
9aca1d51 | 6745 | |
2c47e605 YZ |
6746 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6747 | if (ret < 0) { | |
6748 | err = ret; | |
20524f02 | 6749 | break; |
2c47e605 YZ |
6750 | } |
6751 | ||
6752 | if (ret > 0) { | |
6753 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6754 | break; |
6755 | } | |
2c47e605 YZ |
6756 | |
6757 | if (wc->stage == DROP_REFERENCE) { | |
6758 | level = wc->level; | |
6759 | btrfs_node_key(path->nodes[level], | |
6760 | &root_item->drop_progress, | |
6761 | path->slots[level]); | |
6762 | root_item->drop_level = level; | |
6763 | } | |
6764 | ||
6765 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6766 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6767 | ret = btrfs_update_root(trans, tree_root, |
6768 | &root->root_key, | |
6769 | root_item); | |
6770 | BUG_ON(ret); | |
6771 | ||
3fd0a558 | 6772 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6773 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 | 6774 | BUG_ON(IS_ERR(trans)); |
3fd0a558 YZ |
6775 | if (block_rsv) |
6776 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6777 | } |
20524f02 | 6778 | } |
b3b4aa74 | 6779 | btrfs_release_path(path); |
2c47e605 YZ |
6780 | BUG_ON(err); |
6781 | ||
6782 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6783 | BUG_ON(ret); | |
6784 | ||
76dda93c YZ |
6785 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6786 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6787 | NULL, NULL); | |
6788 | BUG_ON(ret < 0); | |
6789 | if (ret > 0) { | |
84cd948c JB |
6790 | /* if we fail to delete the orphan item this time |
6791 | * around, it'll get picked up the next time. | |
6792 | * | |
6793 | * The most common failure here is just -ENOENT. | |
6794 | */ | |
6795 | btrfs_del_orphan_item(trans, tree_root, | |
6796 | root->root_key.objectid); | |
76dda93c YZ |
6797 | } |
6798 | } | |
6799 | ||
6800 | if (root->in_radix) { | |
6801 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6802 | } else { | |
6803 | free_extent_buffer(root->node); | |
6804 | free_extent_buffer(root->commit_root); | |
6805 | kfree(root); | |
6806 | } | |
cb1b69f4 | 6807 | out_free: |
3fd0a558 | 6808 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6809 | kfree(wc); |
5caf2a00 | 6810 | btrfs_free_path(path); |
cb1b69f4 TI |
6811 | out: |
6812 | if (err) | |
6813 | btrfs_std_error(root->fs_info, err); | |
6814 | return; | |
20524f02 | 6815 | } |
9078a3e1 | 6816 | |
2c47e605 YZ |
6817 | /* |
6818 | * drop subtree rooted at tree block 'node'. | |
6819 | * | |
6820 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 6821 | * only used by relocation code |
2c47e605 | 6822 | */ |
f82d02d9 YZ |
6823 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6824 | struct btrfs_root *root, | |
6825 | struct extent_buffer *node, | |
6826 | struct extent_buffer *parent) | |
6827 | { | |
6828 | struct btrfs_path *path; | |
2c47e605 | 6829 | struct walk_control *wc; |
f82d02d9 YZ |
6830 | int level; |
6831 | int parent_level; | |
6832 | int ret = 0; | |
6833 | int wret; | |
6834 | ||
2c47e605 YZ |
6835 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6836 | ||
f82d02d9 | 6837 | path = btrfs_alloc_path(); |
db5b493a TI |
6838 | if (!path) |
6839 | return -ENOMEM; | |
f82d02d9 | 6840 | |
2c47e605 | 6841 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6842 | if (!wc) { |
6843 | btrfs_free_path(path); | |
6844 | return -ENOMEM; | |
6845 | } | |
2c47e605 | 6846 | |
b9447ef8 | 6847 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6848 | parent_level = btrfs_header_level(parent); |
6849 | extent_buffer_get(parent); | |
6850 | path->nodes[parent_level] = parent; | |
6851 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6852 | ||
b9447ef8 | 6853 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6854 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6855 | path->nodes[level] = node; |
6856 | path->slots[level] = 0; | |
bd681513 | 6857 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6858 | |
6859 | wc->refs[parent_level] = 1; | |
6860 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6861 | wc->level = level; | |
6862 | wc->shared_level = -1; | |
6863 | wc->stage = DROP_REFERENCE; | |
6864 | wc->update_ref = 0; | |
6865 | wc->keep_locks = 1; | |
66d7e7f0 | 6866 | wc->for_reloc = 1; |
1c4850e2 | 6867 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6868 | |
6869 | while (1) { | |
2c47e605 YZ |
6870 | wret = walk_down_tree(trans, root, path, wc); |
6871 | if (wret < 0) { | |
f82d02d9 | 6872 | ret = wret; |
f82d02d9 | 6873 | break; |
2c47e605 | 6874 | } |
f82d02d9 | 6875 | |
2c47e605 | 6876 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6877 | if (wret < 0) |
6878 | ret = wret; | |
6879 | if (wret != 0) | |
6880 | break; | |
6881 | } | |
6882 | ||
2c47e605 | 6883 | kfree(wc); |
f82d02d9 YZ |
6884 | btrfs_free_path(path); |
6885 | return ret; | |
6886 | } | |
6887 | ||
ec44a35c CM |
6888 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6889 | { | |
6890 | u64 num_devices; | |
6891 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6892 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6893 | ||
e4d8ec0f ID |
6894 | if (root->fs_info->balance_ctl) { |
6895 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
6896 | u64 tgt = 0; | |
6897 | ||
6898 | /* pick restriper's target profile and return */ | |
6899 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
6900 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
6901 | tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
6902 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
6903 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
6904 | tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
6905 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
6906 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
6907 | tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
6908 | } | |
6909 | ||
899c81ea ID |
6910 | if (tgt) |
6911 | return extended_to_chunk(tgt); | |
e4d8ec0f ID |
6912 | } |
6913 | ||
cd02dca5 CM |
6914 | /* |
6915 | * we add in the count of missing devices because we want | |
6916 | * to make sure that any RAID levels on a degraded FS | |
6917 | * continue to be honored. | |
6918 | */ | |
6919 | num_devices = root->fs_info->fs_devices->rw_devices + | |
6920 | root->fs_info->fs_devices->missing_devices; | |
6921 | ||
ec44a35c CM |
6922 | if (num_devices == 1) { |
6923 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6924 | stripped = flags & ~stripped; | |
6925 | ||
6926 | /* turn raid0 into single device chunks */ | |
6927 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6928 | return stripped; | |
6929 | ||
6930 | /* turn mirroring into duplication */ | |
6931 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6932 | BTRFS_BLOCK_GROUP_RAID10)) | |
6933 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
6934 | } else { |
6935 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6936 | if (flags & stripped) |
6937 | return flags; | |
6938 | ||
6939 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6940 | stripped = flags & ~stripped; | |
6941 | ||
6942 | /* switch duplicated blocks with raid1 */ | |
6943 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6944 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6945 | ||
e3176ca2 | 6946 | /* this is drive concat, leave it alone */ |
ec44a35c | 6947 | } |
e3176ca2 | 6948 | |
ec44a35c CM |
6949 | return flags; |
6950 | } | |
6951 | ||
199c36ea | 6952 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 6953 | { |
f0486c68 YZ |
6954 | struct btrfs_space_info *sinfo = cache->space_info; |
6955 | u64 num_bytes; | |
199c36ea | 6956 | u64 min_allocable_bytes; |
f0486c68 | 6957 | int ret = -ENOSPC; |
0ef3e66b | 6958 | |
c286ac48 | 6959 | |
199c36ea MX |
6960 | /* |
6961 | * We need some metadata space and system metadata space for | |
6962 | * allocating chunks in some corner cases until we force to set | |
6963 | * it to be readonly. | |
6964 | */ | |
6965 | if ((sinfo->flags & | |
6966 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
6967 | !force) | |
6968 | min_allocable_bytes = 1 * 1024 * 1024; | |
6969 | else | |
6970 | min_allocable_bytes = 0; | |
6971 | ||
f0486c68 YZ |
6972 | spin_lock(&sinfo->lock); |
6973 | spin_lock(&cache->lock); | |
61cfea9b W |
6974 | |
6975 | if (cache->ro) { | |
6976 | ret = 0; | |
6977 | goto out; | |
6978 | } | |
6979 | ||
f0486c68 YZ |
6980 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
6981 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6982 | ||
6983 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
6984 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
6985 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 6986 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
6987 | cache->ro = 1; |
6988 | ret = 0; | |
6989 | } | |
61cfea9b | 6990 | out: |
f0486c68 YZ |
6991 | spin_unlock(&cache->lock); |
6992 | spin_unlock(&sinfo->lock); | |
6993 | return ret; | |
6994 | } | |
7d9eb12c | 6995 | |
f0486c68 YZ |
6996 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
6997 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 6998 | |
f0486c68 YZ |
6999 | { |
7000 | struct btrfs_trans_handle *trans; | |
7001 | u64 alloc_flags; | |
7002 | int ret; | |
7d9eb12c | 7003 | |
f0486c68 | 7004 | BUG_ON(cache->ro); |
0ef3e66b | 7005 | |
ff5714cc | 7006 | trans = btrfs_join_transaction(root); |
f0486c68 | 7007 | BUG_ON(IS_ERR(trans)); |
5d4f98a2 | 7008 | |
f0486c68 YZ |
7009 | alloc_flags = update_block_group_flags(root, cache->flags); |
7010 | if (alloc_flags != cache->flags) | |
0e4f8f88 CM |
7011 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7012 | CHUNK_ALLOC_FORCE); | |
5d4f98a2 | 7013 | |
199c36ea | 7014 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7015 | if (!ret) |
7016 | goto out; | |
7017 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
7018 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7019 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
7020 | if (ret < 0) |
7021 | goto out; | |
199c36ea | 7022 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7023 | out: |
7024 | btrfs_end_transaction(trans, root); | |
7025 | return ret; | |
7026 | } | |
5d4f98a2 | 7027 | |
c87f08ca CM |
7028 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7029 | struct btrfs_root *root, u64 type) | |
7030 | { | |
7031 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
7032 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7033 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
7034 | } |
7035 | ||
6d07bcec MX |
7036 | /* |
7037 | * helper to account the unused space of all the readonly block group in the | |
7038 | * list. takes mirrors into account. | |
7039 | */ | |
7040 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7041 | { | |
7042 | struct btrfs_block_group_cache *block_group; | |
7043 | u64 free_bytes = 0; | |
7044 | int factor; | |
7045 | ||
7046 | list_for_each_entry(block_group, groups_list, list) { | |
7047 | spin_lock(&block_group->lock); | |
7048 | ||
7049 | if (!block_group->ro) { | |
7050 | spin_unlock(&block_group->lock); | |
7051 | continue; | |
7052 | } | |
7053 | ||
7054 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7055 | BTRFS_BLOCK_GROUP_RAID10 | | |
7056 | BTRFS_BLOCK_GROUP_DUP)) | |
7057 | factor = 2; | |
7058 | else | |
7059 | factor = 1; | |
7060 | ||
7061 | free_bytes += (block_group->key.offset - | |
7062 | btrfs_block_group_used(&block_group->item)) * | |
7063 | factor; | |
7064 | ||
7065 | spin_unlock(&block_group->lock); | |
7066 | } | |
7067 | ||
7068 | return free_bytes; | |
7069 | } | |
7070 | ||
7071 | /* | |
7072 | * helper to account the unused space of all the readonly block group in the | |
7073 | * space_info. takes mirrors into account. | |
7074 | */ | |
7075 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7076 | { | |
7077 | int i; | |
7078 | u64 free_bytes = 0; | |
7079 | ||
7080 | spin_lock(&sinfo->lock); | |
7081 | ||
7082 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7083 | if (!list_empty(&sinfo->block_groups[i])) | |
7084 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7085 | &sinfo->block_groups[i]); | |
7086 | ||
7087 | spin_unlock(&sinfo->lock); | |
7088 | ||
7089 | return free_bytes; | |
7090 | } | |
7091 | ||
f0486c68 YZ |
7092 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
7093 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 7094 | { |
f0486c68 YZ |
7095 | struct btrfs_space_info *sinfo = cache->space_info; |
7096 | u64 num_bytes; | |
7097 | ||
7098 | BUG_ON(!cache->ro); | |
7099 | ||
7100 | spin_lock(&sinfo->lock); | |
7101 | spin_lock(&cache->lock); | |
7102 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7103 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7104 | sinfo->bytes_readonly -= num_bytes; | |
7105 | cache->ro = 0; | |
7106 | spin_unlock(&cache->lock); | |
7107 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7108 | return 0; |
7109 | } | |
7110 | ||
ba1bf481 JB |
7111 | /* |
7112 | * checks to see if its even possible to relocate this block group. | |
7113 | * | |
7114 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7115 | * ok to go ahead and try. | |
7116 | */ | |
7117 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7118 | { |
ba1bf481 JB |
7119 | struct btrfs_block_group_cache *block_group; |
7120 | struct btrfs_space_info *space_info; | |
7121 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7122 | struct btrfs_device *device; | |
cdcb725c | 7123 | u64 min_free; |
6719db6a JB |
7124 | u64 dev_min = 1; |
7125 | u64 dev_nr = 0; | |
cdcb725c | 7126 | int index; |
ba1bf481 JB |
7127 | int full = 0; |
7128 | int ret = 0; | |
1a40e23b | 7129 | |
ba1bf481 | 7130 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7131 | |
ba1bf481 JB |
7132 | /* odd, couldn't find the block group, leave it alone */ |
7133 | if (!block_group) | |
7134 | return -1; | |
1a40e23b | 7135 | |
cdcb725c | 7136 | min_free = btrfs_block_group_used(&block_group->item); |
7137 | ||
ba1bf481 | 7138 | /* no bytes used, we're good */ |
cdcb725c | 7139 | if (!min_free) |
1a40e23b ZY |
7140 | goto out; |
7141 | ||
ba1bf481 JB |
7142 | space_info = block_group->space_info; |
7143 | spin_lock(&space_info->lock); | |
17d217fe | 7144 | |
ba1bf481 | 7145 | full = space_info->full; |
17d217fe | 7146 | |
ba1bf481 JB |
7147 | /* |
7148 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7149 | * relocate it unless we're able to allocate a new chunk below. |
7150 | * | |
7151 | * Otherwise, we need to make sure we have room in the space to handle | |
7152 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7153 | */ |
7ce618db | 7154 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7155 | (space_info->bytes_used + space_info->bytes_reserved + |
7156 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7157 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7158 | spin_unlock(&space_info->lock); |
7159 | goto out; | |
17d217fe | 7160 | } |
ba1bf481 | 7161 | spin_unlock(&space_info->lock); |
ea8c2819 | 7162 | |
ba1bf481 JB |
7163 | /* |
7164 | * ok we don't have enough space, but maybe we have free space on our | |
7165 | * devices to allocate new chunks for relocation, so loop through our | |
7166 | * alloc devices and guess if we have enough space. However, if we | |
7167 | * were marked as full, then we know there aren't enough chunks, and we | |
7168 | * can just return. | |
7169 | */ | |
7170 | ret = -1; | |
7171 | if (full) | |
7172 | goto out; | |
ea8c2819 | 7173 | |
cdcb725c | 7174 | /* |
7175 | * index: | |
7176 | * 0: raid10 | |
7177 | * 1: raid1 | |
7178 | * 2: dup | |
7179 | * 3: raid0 | |
7180 | * 4: single | |
7181 | */ | |
7182 | index = get_block_group_index(block_group); | |
7183 | if (index == 0) { | |
7184 | dev_min = 4; | |
6719db6a JB |
7185 | /* Divide by 2 */ |
7186 | min_free >>= 1; | |
cdcb725c | 7187 | } else if (index == 1) { |
7188 | dev_min = 2; | |
7189 | } else if (index == 2) { | |
6719db6a JB |
7190 | /* Multiply by 2 */ |
7191 | min_free <<= 1; | |
cdcb725c | 7192 | } else if (index == 3) { |
7193 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7194 | do_div(min_free, dev_min); |
cdcb725c | 7195 | } |
7196 | ||
ba1bf481 JB |
7197 | mutex_lock(&root->fs_info->chunk_mutex); |
7198 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7199 | u64 dev_offset; |
56bec294 | 7200 | |
ba1bf481 JB |
7201 | /* |
7202 | * check to make sure we can actually find a chunk with enough | |
7203 | * space to fit our block group in. | |
7204 | */ | |
7205 | if (device->total_bytes > device->bytes_used + min_free) { | |
125ccb0a | 7206 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7207 | &dev_offset, NULL); |
ba1bf481 | 7208 | if (!ret) |
cdcb725c | 7209 | dev_nr++; |
7210 | ||
7211 | if (dev_nr >= dev_min) | |
73e48b27 | 7212 | break; |
cdcb725c | 7213 | |
ba1bf481 | 7214 | ret = -1; |
725c8463 | 7215 | } |
edbd8d4e | 7216 | } |
ba1bf481 | 7217 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7218 | out: |
ba1bf481 | 7219 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7220 | return ret; |
7221 | } | |
7222 | ||
b2950863 CH |
7223 | static int find_first_block_group(struct btrfs_root *root, |
7224 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7225 | { |
925baedd | 7226 | int ret = 0; |
0b86a832 CM |
7227 | struct btrfs_key found_key; |
7228 | struct extent_buffer *leaf; | |
7229 | int slot; | |
edbd8d4e | 7230 | |
0b86a832 CM |
7231 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7232 | if (ret < 0) | |
925baedd CM |
7233 | goto out; |
7234 | ||
d397712b | 7235 | while (1) { |
0b86a832 | 7236 | slot = path->slots[0]; |
edbd8d4e | 7237 | leaf = path->nodes[0]; |
0b86a832 CM |
7238 | if (slot >= btrfs_header_nritems(leaf)) { |
7239 | ret = btrfs_next_leaf(root, path); | |
7240 | if (ret == 0) | |
7241 | continue; | |
7242 | if (ret < 0) | |
925baedd | 7243 | goto out; |
0b86a832 | 7244 | break; |
edbd8d4e | 7245 | } |
0b86a832 | 7246 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7247 | |
0b86a832 | 7248 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7249 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7250 | ret = 0; | |
7251 | goto out; | |
7252 | } | |
0b86a832 | 7253 | path->slots[0]++; |
edbd8d4e | 7254 | } |
925baedd | 7255 | out: |
0b86a832 | 7256 | return ret; |
edbd8d4e CM |
7257 | } |
7258 | ||
0af3d00b JB |
7259 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7260 | { | |
7261 | struct btrfs_block_group_cache *block_group; | |
7262 | u64 last = 0; | |
7263 | ||
7264 | while (1) { | |
7265 | struct inode *inode; | |
7266 | ||
7267 | block_group = btrfs_lookup_first_block_group(info, last); | |
7268 | while (block_group) { | |
7269 | spin_lock(&block_group->lock); | |
7270 | if (block_group->iref) | |
7271 | break; | |
7272 | spin_unlock(&block_group->lock); | |
7273 | block_group = next_block_group(info->tree_root, | |
7274 | block_group); | |
7275 | } | |
7276 | if (!block_group) { | |
7277 | if (last == 0) | |
7278 | break; | |
7279 | last = 0; | |
7280 | continue; | |
7281 | } | |
7282 | ||
7283 | inode = block_group->inode; | |
7284 | block_group->iref = 0; | |
7285 | block_group->inode = NULL; | |
7286 | spin_unlock(&block_group->lock); | |
7287 | iput(inode); | |
7288 | last = block_group->key.objectid + block_group->key.offset; | |
7289 | btrfs_put_block_group(block_group); | |
7290 | } | |
7291 | } | |
7292 | ||
1a40e23b ZY |
7293 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7294 | { | |
7295 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7296 | struct btrfs_space_info *space_info; |
11833d66 | 7297 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7298 | struct rb_node *n; |
7299 | ||
11833d66 YZ |
7300 | down_write(&info->extent_commit_sem); |
7301 | while (!list_empty(&info->caching_block_groups)) { | |
7302 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7303 | struct btrfs_caching_control, list); | |
7304 | list_del(&caching_ctl->list); | |
7305 | put_caching_control(caching_ctl); | |
7306 | } | |
7307 | up_write(&info->extent_commit_sem); | |
7308 | ||
1a40e23b ZY |
7309 | spin_lock(&info->block_group_cache_lock); |
7310 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7311 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7312 | cache_node); | |
1a40e23b ZY |
7313 | rb_erase(&block_group->cache_node, |
7314 | &info->block_group_cache_tree); | |
d899e052 YZ |
7315 | spin_unlock(&info->block_group_cache_lock); |
7316 | ||
80eb234a | 7317 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7318 | list_del(&block_group->list); |
80eb234a | 7319 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7320 | |
817d52f8 | 7321 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7322 | wait_block_group_cache_done(block_group); |
817d52f8 | 7323 | |
3c14874a JB |
7324 | /* |
7325 | * We haven't cached this block group, which means we could | |
7326 | * possibly have excluded extents on this block group. | |
7327 | */ | |
7328 | if (block_group->cached == BTRFS_CACHE_NO) | |
7329 | free_excluded_extents(info->extent_root, block_group); | |
7330 | ||
817d52f8 | 7331 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7332 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7333 | |
7334 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7335 | } |
7336 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7337 | |
7338 | /* now that all the block groups are freed, go through and | |
7339 | * free all the space_info structs. This is only called during | |
7340 | * the final stages of unmount, and so we know nobody is | |
7341 | * using them. We call synchronize_rcu() once before we start, | |
7342 | * just to be on the safe side. | |
7343 | */ | |
7344 | synchronize_rcu(); | |
7345 | ||
8929ecfa YZ |
7346 | release_global_block_rsv(info); |
7347 | ||
4184ea7f CM |
7348 | while(!list_empty(&info->space_info)) { |
7349 | space_info = list_entry(info->space_info.next, | |
7350 | struct btrfs_space_info, | |
7351 | list); | |
f0486c68 | 7352 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7353 | space_info->bytes_reserved > 0 || |
7354 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7355 | WARN_ON(1); |
7356 | dump_space_info(space_info, 0, 0); | |
7357 | } | |
4184ea7f CM |
7358 | list_del(&space_info->list); |
7359 | kfree(space_info); | |
7360 | } | |
1a40e23b ZY |
7361 | return 0; |
7362 | } | |
7363 | ||
b742bb82 YZ |
7364 | static void __link_block_group(struct btrfs_space_info *space_info, |
7365 | struct btrfs_block_group_cache *cache) | |
7366 | { | |
7367 | int index = get_block_group_index(cache); | |
7368 | ||
7369 | down_write(&space_info->groups_sem); | |
7370 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7371 | up_write(&space_info->groups_sem); | |
7372 | } | |
7373 | ||
9078a3e1 CM |
7374 | int btrfs_read_block_groups(struct btrfs_root *root) |
7375 | { | |
7376 | struct btrfs_path *path; | |
7377 | int ret; | |
9078a3e1 | 7378 | struct btrfs_block_group_cache *cache; |
be744175 | 7379 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7380 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7381 | struct btrfs_key key; |
7382 | struct btrfs_key found_key; | |
5f39d397 | 7383 | struct extent_buffer *leaf; |
0af3d00b JB |
7384 | int need_clear = 0; |
7385 | u64 cache_gen; | |
96b5179d | 7386 | |
be744175 | 7387 | root = info->extent_root; |
9078a3e1 | 7388 | key.objectid = 0; |
0b86a832 | 7389 | key.offset = 0; |
9078a3e1 | 7390 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7391 | path = btrfs_alloc_path(); |
7392 | if (!path) | |
7393 | return -ENOMEM; | |
026fd317 | 7394 | path->reada = 1; |
9078a3e1 | 7395 | |
6c41761f | 7396 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7397 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7398 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7399 | need_clear = 1; |
88c2ba3b JB |
7400 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7401 | need_clear = 1; | |
0af3d00b | 7402 | |
d397712b | 7403 | while (1) { |
0b86a832 | 7404 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7405 | if (ret > 0) |
7406 | break; | |
0b86a832 CM |
7407 | if (ret != 0) |
7408 | goto error; | |
5f39d397 CM |
7409 | leaf = path->nodes[0]; |
7410 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7411 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7412 | if (!cache) { |
0b86a832 | 7413 | ret = -ENOMEM; |
f0486c68 | 7414 | goto error; |
9078a3e1 | 7415 | } |
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 | ret = -ENOMEM; | |
7421 | goto error; | |
7422 | } | |
3e1ad54f | 7423 | |
d2fb3437 | 7424 | atomic_set(&cache->count, 1); |
c286ac48 | 7425 | spin_lock_init(&cache->lock); |
817d52f8 | 7426 | cache->fs_info = info; |
0f9dd46c | 7427 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7428 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7429 | |
0af3d00b JB |
7430 | if (need_clear) |
7431 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
7432 | ||
5f39d397 CM |
7433 | read_extent_buffer(leaf, &cache->item, |
7434 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7435 | sizeof(cache->item)); | |
9078a3e1 | 7436 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7437 | |
9078a3e1 | 7438 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7439 | btrfs_release_path(path); |
0b86a832 | 7440 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7441 | cache->sectorsize = root->sectorsize; |
7442 | ||
34d52cb6 LZ |
7443 | btrfs_init_free_space_ctl(cache); |
7444 | ||
3c14874a JB |
7445 | /* |
7446 | * We need to exclude the super stripes now so that the space | |
7447 | * info has super bytes accounted for, otherwise we'll think | |
7448 | * we have more space than we actually do. | |
7449 | */ | |
7450 | exclude_super_stripes(root, cache); | |
7451 | ||
817d52f8 JB |
7452 | /* |
7453 | * check for two cases, either we are full, and therefore | |
7454 | * don't need to bother with the caching work since we won't | |
7455 | * find any space, or we are empty, and we can just add all | |
7456 | * the space in and be done with it. This saves us _alot_ of | |
7457 | * time, particularly in the full case. | |
7458 | */ | |
7459 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7460 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7461 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7462 | free_excluded_extents(root, cache); |
817d52f8 | 7463 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7464 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7465 | cache->cached = BTRFS_CACHE_FINISHED; |
7466 | add_new_free_space(cache, root->fs_info, | |
7467 | found_key.objectid, | |
7468 | found_key.objectid + | |
7469 | found_key.offset); | |
11833d66 | 7470 | free_excluded_extents(root, cache); |
817d52f8 | 7471 | } |
96b5179d | 7472 | |
6324fbf3 CM |
7473 | ret = update_space_info(info, cache->flags, found_key.offset, |
7474 | btrfs_block_group_used(&cache->item), | |
7475 | &space_info); | |
7476 | BUG_ON(ret); | |
7477 | cache->space_info = space_info; | |
1b2da372 | 7478 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7479 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7480 | spin_unlock(&cache->space_info->lock); |
7481 | ||
b742bb82 | 7482 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7483 | |
7484 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7485 | BUG_ON(ret); | |
75ccf47d CM |
7486 | |
7487 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7488 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7489 | set_block_group_ro(cache, 1); |
9078a3e1 | 7490 | } |
b742bb82 YZ |
7491 | |
7492 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7493 | if (!(get_alloc_profile(root, space_info->flags) & | |
7494 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7495 | BTRFS_BLOCK_GROUP_RAID1 | | |
7496 | BTRFS_BLOCK_GROUP_DUP))) | |
7497 | continue; | |
7498 | /* | |
7499 | * avoid allocating from un-mirrored block group if there are | |
7500 | * mirrored block groups. | |
7501 | */ | |
7502 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7503 | set_block_group_ro(cache, 1); |
b742bb82 | 7504 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7505 | set_block_group_ro(cache, 1); |
9078a3e1 | 7506 | } |
f0486c68 YZ |
7507 | |
7508 | init_global_block_rsv(info); | |
0b86a832 CM |
7509 | ret = 0; |
7510 | error: | |
9078a3e1 | 7511 | btrfs_free_path(path); |
0b86a832 | 7512 | return ret; |
9078a3e1 | 7513 | } |
6324fbf3 CM |
7514 | |
7515 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7516 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7517 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7518 | u64 size) |
7519 | { | |
7520 | int ret; | |
6324fbf3 CM |
7521 | struct btrfs_root *extent_root; |
7522 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7523 | |
7524 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7525 | |
12fcfd22 | 7526 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7527 | |
8f18cf13 | 7528 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7529 | if (!cache) |
7530 | return -ENOMEM; | |
34d52cb6 LZ |
7531 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7532 | GFP_NOFS); | |
7533 | if (!cache->free_space_ctl) { | |
7534 | kfree(cache); | |
7535 | return -ENOMEM; | |
7536 | } | |
0f9dd46c | 7537 | |
e17cade2 | 7538 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7539 | cache->key.offset = size; |
d2fb3437 | 7540 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7541 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7542 | cache->fs_info = root->fs_info; |
96303081 | 7543 | |
d2fb3437 | 7544 | atomic_set(&cache->count, 1); |
c286ac48 | 7545 | spin_lock_init(&cache->lock); |
0f9dd46c | 7546 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7547 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7548 | |
34d52cb6 LZ |
7549 | btrfs_init_free_space_ctl(cache); |
7550 | ||
6324fbf3 | 7551 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7552 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7553 | cache->flags = type; | |
7554 | btrfs_set_block_group_flags(&cache->item, type); | |
7555 | ||
11833d66 | 7556 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7557 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7558 | exclude_super_stripes(root, cache); |
96303081 | 7559 | |
817d52f8 JB |
7560 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7561 | chunk_offset + size); | |
7562 | ||
11833d66 YZ |
7563 | free_excluded_extents(root, cache); |
7564 | ||
6324fbf3 CM |
7565 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7566 | &cache->space_info); | |
7567 | BUG_ON(ret); | |
c7c144db | 7568 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7569 | |
7570 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7571 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7572 | spin_unlock(&cache->space_info->lock); |
7573 | ||
b742bb82 | 7574 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7575 | |
0f9dd46c JB |
7576 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7577 | BUG_ON(ret); | |
c286ac48 | 7578 | |
6324fbf3 CM |
7579 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7580 | sizeof(cache->item)); | |
7581 | BUG_ON(ret); | |
7582 | ||
d18a2c44 | 7583 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7584 | |
6324fbf3 CM |
7585 | return 0; |
7586 | } | |
1a40e23b | 7587 | |
10ea00f5 ID |
7588 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7589 | { | |
899c81ea ID |
7590 | u64 extra_flags = chunk_to_extended(flags) & |
7591 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 ID |
7592 | |
7593 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7594 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7595 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7596 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7597 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7598 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7599 | } | |
7600 | ||
1a40e23b ZY |
7601 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7602 | struct btrfs_root *root, u64 group_start) | |
7603 | { | |
7604 | struct btrfs_path *path; | |
7605 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7606 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7607 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7608 | struct btrfs_key key; |
0af3d00b | 7609 | struct inode *inode; |
1a40e23b | 7610 | int ret; |
10ea00f5 | 7611 | int index; |
89a55897 | 7612 | int factor; |
1a40e23b | 7613 | |
1a40e23b ZY |
7614 | root = root->fs_info->extent_root; |
7615 | ||
7616 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7617 | BUG_ON(!block_group); | |
c146afad | 7618 | BUG_ON(!block_group->ro); |
1a40e23b | 7619 | |
9f7c43c9 | 7620 | /* |
7621 | * Free the reserved super bytes from this block group before | |
7622 | * remove it. | |
7623 | */ | |
7624 | free_excluded_extents(root, block_group); | |
7625 | ||
1a40e23b | 7626 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7627 | index = get_block_group_index(block_group); |
89a55897 JB |
7628 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7629 | BTRFS_BLOCK_GROUP_RAID1 | | |
7630 | BTRFS_BLOCK_GROUP_RAID10)) | |
7631 | factor = 2; | |
7632 | else | |
7633 | factor = 1; | |
1a40e23b | 7634 | |
44fb5511 CM |
7635 | /* make sure this block group isn't part of an allocation cluster */ |
7636 | cluster = &root->fs_info->data_alloc_cluster; | |
7637 | spin_lock(&cluster->refill_lock); | |
7638 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7639 | spin_unlock(&cluster->refill_lock); | |
7640 | ||
7641 | /* | |
7642 | * make sure this block group isn't part of a metadata | |
7643 | * allocation cluster | |
7644 | */ | |
7645 | cluster = &root->fs_info->meta_alloc_cluster; | |
7646 | spin_lock(&cluster->refill_lock); | |
7647 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7648 | spin_unlock(&cluster->refill_lock); | |
7649 | ||
1a40e23b | 7650 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7651 | if (!path) { |
7652 | ret = -ENOMEM; | |
7653 | goto out; | |
7654 | } | |
1a40e23b | 7655 | |
10b2f34d | 7656 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7657 | if (!IS_ERR(inode)) { |
b532402e TI |
7658 | ret = btrfs_orphan_add(trans, inode); |
7659 | BUG_ON(ret); | |
0af3d00b JB |
7660 | clear_nlink(inode); |
7661 | /* One for the block groups ref */ | |
7662 | spin_lock(&block_group->lock); | |
7663 | if (block_group->iref) { | |
7664 | block_group->iref = 0; | |
7665 | block_group->inode = NULL; | |
7666 | spin_unlock(&block_group->lock); | |
7667 | iput(inode); | |
7668 | } else { | |
7669 | spin_unlock(&block_group->lock); | |
7670 | } | |
7671 | /* One for our lookup ref */ | |
455757c3 | 7672 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7673 | } |
7674 | ||
7675 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7676 | key.offset = block_group->key.objectid; | |
7677 | key.type = 0; | |
7678 | ||
7679 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7680 | if (ret < 0) | |
7681 | goto out; | |
7682 | if (ret > 0) | |
b3b4aa74 | 7683 | btrfs_release_path(path); |
0af3d00b JB |
7684 | if (ret == 0) { |
7685 | ret = btrfs_del_item(trans, tree_root, path); | |
7686 | if (ret) | |
7687 | goto out; | |
b3b4aa74 | 7688 | btrfs_release_path(path); |
0af3d00b JB |
7689 | } |
7690 | ||
3dfdb934 | 7691 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7692 | rb_erase(&block_group->cache_node, |
7693 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7694 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7695 | |
80eb234a | 7696 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7697 | /* |
7698 | * we must use list_del_init so people can check to see if they | |
7699 | * are still on the list after taking the semaphore | |
7700 | */ | |
7701 | list_del_init(&block_group->list); | |
10ea00f5 ID |
7702 | if (list_empty(&block_group->space_info->block_groups[index])) |
7703 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 7704 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7705 | |
817d52f8 | 7706 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7707 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7708 | |
7709 | btrfs_remove_free_space_cache(block_group); | |
7710 | ||
c146afad YZ |
7711 | spin_lock(&block_group->space_info->lock); |
7712 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7713 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7714 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7715 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7716 | |
0af3d00b JB |
7717 | memcpy(&key, &block_group->key, sizeof(key)); |
7718 | ||
283bb197 | 7719 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7720 | |
fa9c0d79 CM |
7721 | btrfs_put_block_group(block_group); |
7722 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7723 | |
7724 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7725 | if (ret > 0) | |
7726 | ret = -EIO; | |
7727 | if (ret < 0) | |
7728 | goto out; | |
7729 | ||
7730 | ret = btrfs_del_item(trans, root, path); | |
7731 | out: | |
7732 | btrfs_free_path(path); | |
7733 | return ret; | |
7734 | } | |
acce952b | 7735 | |
c59021f8 | 7736 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7737 | { | |
7738 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7739 | struct btrfs_super_block *disk_super; |
7740 | u64 features; | |
7741 | u64 flags; | |
7742 | int mixed = 0; | |
c59021f8 | 7743 | int ret; |
7744 | ||
6c41761f | 7745 | disk_super = fs_info->super_copy; |
1aba86d6 | 7746 | if (!btrfs_super_root(disk_super)) |
7747 | return 1; | |
c59021f8 | 7748 | |
1aba86d6 | 7749 | features = btrfs_super_incompat_flags(disk_super); |
7750 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7751 | mixed = 1; | |
c59021f8 | 7752 | |
1aba86d6 | 7753 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7754 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7755 | if (ret) |
1aba86d6 | 7756 | goto out; |
c59021f8 | 7757 | |
1aba86d6 | 7758 | if (mixed) { |
7759 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7760 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7761 | } else { | |
7762 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7763 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7764 | if (ret) | |
7765 | goto out; | |
7766 | ||
7767 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7768 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7769 | } | |
7770 | out: | |
c59021f8 | 7771 | return ret; |
7772 | } | |
7773 | ||
acce952b | 7774 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7775 | { | |
7776 | return unpin_extent_range(root, start, end); | |
7777 | } | |
7778 | ||
7779 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7780 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7781 | { |
5378e607 | 7782 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7783 | } |
f7039b1d LD |
7784 | |
7785 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7786 | { | |
7787 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7788 | struct btrfs_block_group_cache *cache = NULL; | |
7789 | u64 group_trimmed; | |
7790 | u64 start; | |
7791 | u64 end; | |
7792 | u64 trimmed = 0; | |
2cac13e4 | 7793 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
7794 | int ret = 0; |
7795 | ||
2cac13e4 LB |
7796 | /* |
7797 | * try to trim all FS space, our block group may start from non-zero. | |
7798 | */ | |
7799 | if (range->len == total_bytes) | |
7800 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
7801 | else | |
7802 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
7803 | |
7804 | while (cache) { | |
7805 | if (cache->key.objectid >= (range->start + range->len)) { | |
7806 | btrfs_put_block_group(cache); | |
7807 | break; | |
7808 | } | |
7809 | ||
7810 | start = max(range->start, cache->key.objectid); | |
7811 | end = min(range->start + range->len, | |
7812 | cache->key.objectid + cache->key.offset); | |
7813 | ||
7814 | if (end - start >= range->minlen) { | |
7815 | if (!block_group_cache_done(cache)) { | |
7816 | ret = cache_block_group(cache, NULL, root, 0); | |
7817 | if (!ret) | |
7818 | wait_block_group_cache_done(cache); | |
7819 | } | |
7820 | ret = btrfs_trim_block_group(cache, | |
7821 | &group_trimmed, | |
7822 | start, | |
7823 | end, | |
7824 | range->minlen); | |
7825 | ||
7826 | trimmed += group_trimmed; | |
7827 | if (ret) { | |
7828 | btrfs_put_block_group(cache); | |
7829 | break; | |
7830 | } | |
7831 | } | |
7832 | ||
7833 | cache = next_block_group(fs_info->tree_root, cache); | |
7834 | } | |
7835 | ||
7836 | range->len = trimmed; | |
7837 | return ret; | |
7838 | } |