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); | |
79787eaa | 248 | BUG_ON(ret); /* -ENOMEM */ |
06b2331f YZ |
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); | |
79787eaa | 256 | BUG_ON(ret); /* -ENOMEM */ |
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); | |
79787eaa | 262 | BUG_ON(ret); /* -ENOMEM */ |
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); | |
79787eaa | 324 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
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); |
79787eaa | 335 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
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 | 476 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
477 | if (!caching_ctl) |
478 | return -ENOMEM; | |
291c7d2f JB |
479 | |
480 | INIT_LIST_HEAD(&caching_ctl->list); | |
481 | mutex_init(&caching_ctl->mutex); | |
482 | init_waitqueue_head(&caching_ctl->wait); | |
483 | caching_ctl->block_group = cache; | |
484 | caching_ctl->progress = cache->key.objectid; | |
485 | atomic_set(&caching_ctl->count, 1); | |
486 | caching_ctl->work.func = caching_thread; | |
487 | ||
488 | spin_lock(&cache->lock); | |
489 | /* | |
490 | * This should be a rare occasion, but this could happen I think in the | |
491 | * case where one thread starts to load the space cache info, and then | |
492 | * some other thread starts a transaction commit which tries to do an | |
493 | * allocation while the other thread is still loading the space cache | |
494 | * info. The previous loop should have kept us from choosing this block | |
495 | * group, but if we've moved to the state where we will wait on caching | |
496 | * block groups we need to first check if we're doing a fast load here, | |
497 | * so we can wait for it to finish, otherwise we could end up allocating | |
498 | * from a block group who's cache gets evicted for one reason or | |
499 | * another. | |
500 | */ | |
501 | while (cache->cached == BTRFS_CACHE_FAST) { | |
502 | struct btrfs_caching_control *ctl; | |
503 | ||
504 | ctl = cache->caching_ctl; | |
505 | atomic_inc(&ctl->count); | |
506 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
507 | spin_unlock(&cache->lock); | |
508 | ||
509 | schedule(); | |
510 | ||
511 | finish_wait(&ctl->wait, &wait); | |
512 | put_caching_control(ctl); | |
513 | spin_lock(&cache->lock); | |
514 | } | |
515 | ||
516 | if (cache->cached != BTRFS_CACHE_NO) { | |
517 | spin_unlock(&cache->lock); | |
518 | kfree(caching_ctl); | |
11833d66 | 519 | return 0; |
291c7d2f JB |
520 | } |
521 | WARN_ON(cache->caching_ctl); | |
522 | cache->caching_ctl = caching_ctl; | |
523 | cache->cached = BTRFS_CACHE_FAST; | |
524 | spin_unlock(&cache->lock); | |
11833d66 | 525 | |
9d66e233 JB |
526 | /* |
527 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
528 | * to have the normal tree locking. Also if we are currently trying to |
529 | * allocate blocks for the tree root we can't do the fast caching since | |
530 | * we likely hold important locks. | |
9d66e233 | 531 | */ |
d53ba474 | 532 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
533 | ret = load_free_space_cache(fs_info, cache); |
534 | ||
535 | spin_lock(&cache->lock); | |
536 | if (ret == 1) { | |
291c7d2f | 537 | cache->caching_ctl = NULL; |
9d66e233 JB |
538 | cache->cached = BTRFS_CACHE_FINISHED; |
539 | cache->last_byte_to_unpin = (u64)-1; | |
540 | } else { | |
291c7d2f JB |
541 | if (load_cache_only) { |
542 | cache->caching_ctl = NULL; | |
543 | cache->cached = BTRFS_CACHE_NO; | |
544 | } else { | |
545 | cache->cached = BTRFS_CACHE_STARTED; | |
546 | } | |
9d66e233 JB |
547 | } |
548 | spin_unlock(&cache->lock); | |
291c7d2f | 549 | wake_up(&caching_ctl->wait); |
3c14874a | 550 | if (ret == 1) { |
291c7d2f | 551 | put_caching_control(caching_ctl); |
3c14874a | 552 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 553 | return 0; |
3c14874a | 554 | } |
291c7d2f JB |
555 | } else { |
556 | /* | |
557 | * We are not going to do the fast caching, set cached to the | |
558 | * appropriate value and wakeup any waiters. | |
559 | */ | |
560 | spin_lock(&cache->lock); | |
561 | if (load_cache_only) { | |
562 | cache->caching_ctl = NULL; | |
563 | cache->cached = BTRFS_CACHE_NO; | |
564 | } else { | |
565 | cache->cached = BTRFS_CACHE_STARTED; | |
566 | } | |
567 | spin_unlock(&cache->lock); | |
568 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
569 | } |
570 | ||
291c7d2f JB |
571 | if (load_cache_only) { |
572 | put_caching_control(caching_ctl); | |
11833d66 | 573 | return 0; |
817d52f8 | 574 | } |
817d52f8 | 575 | |
11833d66 | 576 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 577 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
578 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
579 | up_write(&fs_info->extent_commit_sem); | |
580 | ||
11dfe35a | 581 | btrfs_get_block_group(cache); |
11833d66 | 582 | |
bab39bf9 | 583 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 584 | |
ef8bbdfe | 585 | return ret; |
e37c9e69 CM |
586 | } |
587 | ||
0f9dd46c JB |
588 | /* |
589 | * return the block group that starts at or after bytenr | |
590 | */ | |
d397712b CM |
591 | static struct btrfs_block_group_cache * |
592 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 593 | { |
0f9dd46c | 594 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 595 | |
0f9dd46c | 596 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 597 | |
0f9dd46c | 598 | return cache; |
0ef3e66b CM |
599 | } |
600 | ||
0f9dd46c | 601 | /* |
9f55684c | 602 | * return the block group that contains the given bytenr |
0f9dd46c | 603 | */ |
d397712b CM |
604 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
605 | struct btrfs_fs_info *info, | |
606 | u64 bytenr) | |
be744175 | 607 | { |
0f9dd46c | 608 | struct btrfs_block_group_cache *cache; |
be744175 | 609 | |
0f9dd46c | 610 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 611 | |
0f9dd46c | 612 | return cache; |
be744175 | 613 | } |
0b86a832 | 614 | |
0f9dd46c JB |
615 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
616 | u64 flags) | |
6324fbf3 | 617 | { |
0f9dd46c | 618 | struct list_head *head = &info->space_info; |
0f9dd46c | 619 | struct btrfs_space_info *found; |
4184ea7f | 620 | |
52ba6929 | 621 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 622 | |
4184ea7f CM |
623 | rcu_read_lock(); |
624 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 625 | if (found->flags & flags) { |
4184ea7f | 626 | rcu_read_unlock(); |
0f9dd46c | 627 | return found; |
4184ea7f | 628 | } |
0f9dd46c | 629 | } |
4184ea7f | 630 | rcu_read_unlock(); |
0f9dd46c | 631 | return NULL; |
6324fbf3 CM |
632 | } |
633 | ||
4184ea7f CM |
634 | /* |
635 | * after adding space to the filesystem, we need to clear the full flags | |
636 | * on all the space infos. | |
637 | */ | |
638 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
639 | { | |
640 | struct list_head *head = &info->space_info; | |
641 | struct btrfs_space_info *found; | |
642 | ||
643 | rcu_read_lock(); | |
644 | list_for_each_entry_rcu(found, head, list) | |
645 | found->full = 0; | |
646 | rcu_read_unlock(); | |
647 | } | |
648 | ||
80eb234a JB |
649 | static u64 div_factor(u64 num, int factor) |
650 | { | |
651 | if (factor == 10) | |
652 | return num; | |
653 | num *= factor; | |
654 | do_div(num, 10); | |
655 | return num; | |
656 | } | |
657 | ||
e5bc2458 CM |
658 | static u64 div_factor_fine(u64 num, int factor) |
659 | { | |
660 | if (factor == 100) | |
661 | return num; | |
662 | num *= factor; | |
663 | do_div(num, 100); | |
664 | return num; | |
665 | } | |
666 | ||
d2fb3437 YZ |
667 | u64 btrfs_find_block_group(struct btrfs_root *root, |
668 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 669 | { |
96b5179d | 670 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 671 | u64 used; |
d2fb3437 YZ |
672 | u64 last = max(search_hint, search_start); |
673 | u64 group_start = 0; | |
31f3c99b | 674 | int full_search = 0; |
d2fb3437 | 675 | int factor = 9; |
0ef3e66b | 676 | int wrapped = 0; |
31f3c99b | 677 | again: |
e8569813 ZY |
678 | while (1) { |
679 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
680 | if (!cache) |
681 | break; | |
96b5179d | 682 | |
c286ac48 | 683 | spin_lock(&cache->lock); |
96b5179d CM |
684 | last = cache->key.objectid + cache->key.offset; |
685 | used = btrfs_block_group_used(&cache->item); | |
686 | ||
d2fb3437 YZ |
687 | if ((full_search || !cache->ro) && |
688 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 689 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
690 | div_factor(cache->key.offset, factor)) { |
691 | group_start = cache->key.objectid; | |
c286ac48 | 692 | spin_unlock(&cache->lock); |
fa9c0d79 | 693 | btrfs_put_block_group(cache); |
8790d502 CM |
694 | goto found; |
695 | } | |
6324fbf3 | 696 | } |
c286ac48 | 697 | spin_unlock(&cache->lock); |
fa9c0d79 | 698 | btrfs_put_block_group(cache); |
de428b63 | 699 | cond_resched(); |
cd1bc465 | 700 | } |
0ef3e66b CM |
701 | if (!wrapped) { |
702 | last = search_start; | |
703 | wrapped = 1; | |
704 | goto again; | |
705 | } | |
706 | if (!full_search && factor < 10) { | |
be744175 | 707 | last = search_start; |
31f3c99b | 708 | full_search = 1; |
0ef3e66b | 709 | factor = 10; |
31f3c99b CM |
710 | goto again; |
711 | } | |
be744175 | 712 | found: |
d2fb3437 | 713 | return group_start; |
925baedd | 714 | } |
0f9dd46c | 715 | |
e02119d5 | 716 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 717 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
718 | { |
719 | int ret; | |
720 | struct btrfs_key key; | |
31840ae1 | 721 | struct btrfs_path *path; |
e02119d5 | 722 | |
31840ae1 | 723 | path = btrfs_alloc_path(); |
d8926bb3 MF |
724 | if (!path) |
725 | return -ENOMEM; | |
726 | ||
e02119d5 CM |
727 | key.objectid = start; |
728 | key.offset = len; | |
729 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
730 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
731 | 0, 0); | |
31840ae1 | 732 | btrfs_free_path(path); |
7bb86316 CM |
733 | return ret; |
734 | } | |
735 | ||
a22285a6 YZ |
736 | /* |
737 | * helper function to lookup reference count and flags of extent. | |
738 | * | |
739 | * the head node for delayed ref is used to store the sum of all the | |
740 | * reference count modifications queued up in the rbtree. the head | |
741 | * node may also store the extent flags to set. This way you can check | |
742 | * to see what the reference count and extent flags would be if all of | |
743 | * the delayed refs are not processed. | |
744 | */ | |
745 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
746 | struct btrfs_root *root, u64 bytenr, | |
747 | u64 num_bytes, u64 *refs, u64 *flags) | |
748 | { | |
749 | struct btrfs_delayed_ref_head *head; | |
750 | struct btrfs_delayed_ref_root *delayed_refs; | |
751 | struct btrfs_path *path; | |
752 | struct btrfs_extent_item *ei; | |
753 | struct extent_buffer *leaf; | |
754 | struct btrfs_key key; | |
755 | u32 item_size; | |
756 | u64 num_refs; | |
757 | u64 extent_flags; | |
758 | int ret; | |
759 | ||
760 | path = btrfs_alloc_path(); | |
761 | if (!path) | |
762 | return -ENOMEM; | |
763 | ||
764 | key.objectid = bytenr; | |
765 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
766 | key.offset = num_bytes; | |
767 | if (!trans) { | |
768 | path->skip_locking = 1; | |
769 | path->search_commit_root = 1; | |
770 | } | |
771 | again: | |
772 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
773 | &key, path, 0, 0); | |
774 | if (ret < 0) | |
775 | goto out_free; | |
776 | ||
777 | if (ret == 0) { | |
778 | leaf = path->nodes[0]; | |
779 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
780 | if (item_size >= sizeof(*ei)) { | |
781 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
782 | struct btrfs_extent_item); | |
783 | num_refs = btrfs_extent_refs(leaf, ei); | |
784 | extent_flags = btrfs_extent_flags(leaf, ei); | |
785 | } else { | |
786 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
787 | struct btrfs_extent_item_v0 *ei0; | |
788 | BUG_ON(item_size != sizeof(*ei0)); | |
789 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
790 | struct btrfs_extent_item_v0); | |
791 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
792 | /* FIXME: this isn't correct for data */ | |
793 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
794 | #else | |
795 | BUG(); | |
796 | #endif | |
797 | } | |
798 | BUG_ON(num_refs == 0); | |
799 | } else { | |
800 | num_refs = 0; | |
801 | extent_flags = 0; | |
802 | ret = 0; | |
803 | } | |
804 | ||
805 | if (!trans) | |
806 | goto out; | |
807 | ||
808 | delayed_refs = &trans->transaction->delayed_refs; | |
809 | spin_lock(&delayed_refs->lock); | |
810 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
811 | if (head) { | |
812 | if (!mutex_trylock(&head->mutex)) { | |
813 | atomic_inc(&head->node.refs); | |
814 | spin_unlock(&delayed_refs->lock); | |
815 | ||
b3b4aa74 | 816 | btrfs_release_path(path); |
a22285a6 | 817 | |
8cc33e5c DS |
818 | /* |
819 | * Mutex was contended, block until it's released and try | |
820 | * again | |
821 | */ | |
a22285a6 YZ |
822 | mutex_lock(&head->mutex); |
823 | mutex_unlock(&head->mutex); | |
824 | btrfs_put_delayed_ref(&head->node); | |
825 | goto again; | |
826 | } | |
827 | if (head->extent_op && head->extent_op->update_flags) | |
828 | extent_flags |= head->extent_op->flags_to_set; | |
829 | else | |
830 | BUG_ON(num_refs == 0); | |
831 | ||
832 | num_refs += head->node.ref_mod; | |
833 | mutex_unlock(&head->mutex); | |
834 | } | |
835 | spin_unlock(&delayed_refs->lock); | |
836 | out: | |
837 | WARN_ON(num_refs == 0); | |
838 | if (refs) | |
839 | *refs = num_refs; | |
840 | if (flags) | |
841 | *flags = extent_flags; | |
842 | out_free: | |
843 | btrfs_free_path(path); | |
844 | return ret; | |
845 | } | |
846 | ||
d8d5f3e1 CM |
847 | /* |
848 | * Back reference rules. Back refs have three main goals: | |
849 | * | |
850 | * 1) differentiate between all holders of references to an extent so that | |
851 | * when a reference is dropped we can make sure it was a valid reference | |
852 | * before freeing the extent. | |
853 | * | |
854 | * 2) Provide enough information to quickly find the holders of an extent | |
855 | * if we notice a given block is corrupted or bad. | |
856 | * | |
857 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
858 | * maintenance. This is actually the same as #2, but with a slightly | |
859 | * different use case. | |
860 | * | |
5d4f98a2 YZ |
861 | * There are two kinds of back refs. The implicit back refs is optimized |
862 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
863 | * back refs of this kind provide information about the block's owner tree | |
864 | * and the pointer's key. These information allow us to find the block by | |
865 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
866 | * referenced by their owner trees. The location of tree block is recorded | |
867 | * in the back refs. Actually the full back refs is generic, and can be | |
868 | * used in all cases the implicit back refs is used. The major shortcoming | |
869 | * of the full back refs is its overhead. Every time a tree block gets | |
870 | * COWed, we have to update back refs entry for all pointers in it. | |
871 | * | |
872 | * For a newly allocated tree block, we use implicit back refs for | |
873 | * pointers in it. This means most tree related operations only involve | |
874 | * implicit back refs. For a tree block created in old transaction, the | |
875 | * only way to drop a reference to it is COW it. So we can detect the | |
876 | * event that tree block loses its owner tree's reference and do the | |
877 | * back refs conversion. | |
878 | * | |
879 | * When a tree block is COW'd through a tree, there are four cases: | |
880 | * | |
881 | * The reference count of the block is one and the tree is the block's | |
882 | * owner tree. Nothing to do in this case. | |
883 | * | |
884 | * The reference count of the block is one and the tree is not the | |
885 | * block's owner tree. In this case, full back refs is used for pointers | |
886 | * in the block. Remove these full back refs, add implicit back refs for | |
887 | * every pointers in the new block. | |
888 | * | |
889 | * The reference count of the block is greater than one and the tree is | |
890 | * the block's owner tree. In this case, implicit back refs is used for | |
891 | * pointers in the block. Add full back refs for every pointers in the | |
892 | * block, increase lower level extents' reference counts. The original | |
893 | * implicit back refs are entailed to the new block. | |
894 | * | |
895 | * The reference count of the block is greater than one and the tree is | |
896 | * not the block's owner tree. Add implicit back refs for every pointer in | |
897 | * the new block, increase lower level extents' reference count. | |
898 | * | |
899 | * Back Reference Key composing: | |
900 | * | |
901 | * The key objectid corresponds to the first byte in the extent, | |
902 | * The key type is used to differentiate between types of back refs. | |
903 | * There are different meanings of the key offset for different types | |
904 | * of back refs. | |
905 | * | |
d8d5f3e1 CM |
906 | * File extents can be referenced by: |
907 | * | |
908 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 909 | * - different files inside a single subvolume |
d8d5f3e1 CM |
910 | * - different offsets inside a file (bookend extents in file.c) |
911 | * | |
5d4f98a2 | 912 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
913 | * |
914 | * - Objectid of the subvolume root | |
d8d5f3e1 | 915 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
916 | * - original offset in the file |
917 | * - how many bookend extents | |
d8d5f3e1 | 918 | * |
5d4f98a2 YZ |
919 | * The key offset for the implicit back refs is hash of the first |
920 | * three fields. | |
d8d5f3e1 | 921 | * |
5d4f98a2 | 922 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 923 | * |
5d4f98a2 | 924 | * - number of pointers in the tree leaf |
d8d5f3e1 | 925 | * |
5d4f98a2 YZ |
926 | * The key offset for the implicit back refs is the first byte of |
927 | * the tree leaf | |
d8d5f3e1 | 928 | * |
5d4f98a2 YZ |
929 | * When a file extent is allocated, The implicit back refs is used. |
930 | * the fields are filled in: | |
d8d5f3e1 | 931 | * |
5d4f98a2 | 932 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 933 | * |
5d4f98a2 YZ |
934 | * When a file extent is removed file truncation, we find the |
935 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 936 | * |
5d4f98a2 | 937 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 938 | * |
5d4f98a2 | 939 | * Btree extents can be referenced by: |
d8d5f3e1 | 940 | * |
5d4f98a2 | 941 | * - Different subvolumes |
d8d5f3e1 | 942 | * |
5d4f98a2 YZ |
943 | * Both the implicit back refs and the full back refs for tree blocks |
944 | * only consist of key. The key offset for the implicit back refs is | |
945 | * objectid of block's owner tree. The key offset for the full back refs | |
946 | * is the first byte of parent block. | |
d8d5f3e1 | 947 | * |
5d4f98a2 YZ |
948 | * When implicit back refs is used, information about the lowest key and |
949 | * level of the tree block are required. These information are stored in | |
950 | * tree block info structure. | |
d8d5f3e1 | 951 | */ |
31840ae1 | 952 | |
5d4f98a2 YZ |
953 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
954 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
955 | struct btrfs_root *root, | |
956 | struct btrfs_path *path, | |
957 | u64 owner, u32 extra_size) | |
7bb86316 | 958 | { |
5d4f98a2 YZ |
959 | struct btrfs_extent_item *item; |
960 | struct btrfs_extent_item_v0 *ei0; | |
961 | struct btrfs_extent_ref_v0 *ref0; | |
962 | struct btrfs_tree_block_info *bi; | |
963 | struct extent_buffer *leaf; | |
7bb86316 | 964 | struct btrfs_key key; |
5d4f98a2 YZ |
965 | struct btrfs_key found_key; |
966 | u32 new_size = sizeof(*item); | |
967 | u64 refs; | |
968 | int ret; | |
969 | ||
970 | leaf = path->nodes[0]; | |
971 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
972 | ||
973 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
974 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
975 | struct btrfs_extent_item_v0); | |
976 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
977 | ||
978 | if (owner == (u64)-1) { | |
979 | while (1) { | |
980 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
981 | ret = btrfs_next_leaf(root, path); | |
982 | if (ret < 0) | |
983 | return ret; | |
79787eaa | 984 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
985 | leaf = path->nodes[0]; |
986 | } | |
987 | btrfs_item_key_to_cpu(leaf, &found_key, | |
988 | path->slots[0]); | |
989 | BUG_ON(key.objectid != found_key.objectid); | |
990 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
991 | path->slots[0]++; | |
992 | continue; | |
993 | } | |
994 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
995 | struct btrfs_extent_ref_v0); | |
996 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
997 | break; | |
998 | } | |
999 | } | |
b3b4aa74 | 1000 | btrfs_release_path(path); |
5d4f98a2 YZ |
1001 | |
1002 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1003 | new_size += sizeof(*bi); | |
1004 | ||
1005 | new_size -= sizeof(*ei0); | |
1006 | ret = btrfs_search_slot(trans, root, &key, path, | |
1007 | new_size + extra_size, 1); | |
1008 | if (ret < 0) | |
1009 | return ret; | |
79787eaa | 1010 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1011 | |
143bede5 | 1012 | btrfs_extend_item(trans, root, path, new_size); |
5d4f98a2 YZ |
1013 | |
1014 | leaf = path->nodes[0]; | |
1015 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1016 | btrfs_set_extent_refs(leaf, item, refs); | |
1017 | /* FIXME: get real generation */ | |
1018 | btrfs_set_extent_generation(leaf, item, 0); | |
1019 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1020 | btrfs_set_extent_flags(leaf, item, | |
1021 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1022 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1023 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1024 | /* FIXME: get first key of the block */ | |
1025 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1026 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1027 | } else { | |
1028 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1029 | } | |
1030 | btrfs_mark_buffer_dirty(leaf); | |
1031 | return 0; | |
1032 | } | |
1033 | #endif | |
1034 | ||
1035 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1036 | { | |
1037 | u32 high_crc = ~(u32)0; | |
1038 | u32 low_crc = ~(u32)0; | |
1039 | __le64 lenum; | |
1040 | ||
1041 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1042 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1043 | lenum = cpu_to_le64(owner); |
163e783e | 1044 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1045 | lenum = cpu_to_le64(offset); |
163e783e | 1046 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1047 | |
1048 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1049 | } | |
1050 | ||
1051 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1052 | struct btrfs_extent_data_ref *ref) | |
1053 | { | |
1054 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1055 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1056 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1057 | } | |
1058 | ||
1059 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1060 | struct btrfs_extent_data_ref *ref, | |
1061 | u64 root_objectid, u64 owner, u64 offset) | |
1062 | { | |
1063 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1064 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1065 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1066 | return 0; | |
1067 | return 1; | |
1068 | } | |
1069 | ||
1070 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1071 | struct btrfs_root *root, | |
1072 | struct btrfs_path *path, | |
1073 | u64 bytenr, u64 parent, | |
1074 | u64 root_objectid, | |
1075 | u64 owner, u64 offset) | |
1076 | { | |
1077 | struct btrfs_key key; | |
1078 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1079 | struct extent_buffer *leaf; |
5d4f98a2 | 1080 | u32 nritems; |
74493f7a | 1081 | int ret; |
5d4f98a2 YZ |
1082 | int recow; |
1083 | int err = -ENOENT; | |
74493f7a | 1084 | |
31840ae1 | 1085 | key.objectid = bytenr; |
5d4f98a2 YZ |
1086 | if (parent) { |
1087 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1088 | key.offset = parent; | |
1089 | } else { | |
1090 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1091 | key.offset = hash_extent_data_ref(root_objectid, | |
1092 | owner, offset); | |
1093 | } | |
1094 | again: | |
1095 | recow = 0; | |
1096 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1097 | if (ret < 0) { | |
1098 | err = ret; | |
1099 | goto fail; | |
1100 | } | |
31840ae1 | 1101 | |
5d4f98a2 YZ |
1102 | if (parent) { |
1103 | if (!ret) | |
1104 | return 0; | |
1105 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1106 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1107 | btrfs_release_path(path); |
5d4f98a2 YZ |
1108 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1109 | if (ret < 0) { | |
1110 | err = ret; | |
1111 | goto fail; | |
1112 | } | |
1113 | if (!ret) | |
1114 | return 0; | |
1115 | #endif | |
1116 | goto fail; | |
31840ae1 ZY |
1117 | } |
1118 | ||
1119 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1120 | nritems = btrfs_header_nritems(leaf); |
1121 | while (1) { | |
1122 | if (path->slots[0] >= nritems) { | |
1123 | ret = btrfs_next_leaf(root, path); | |
1124 | if (ret < 0) | |
1125 | err = ret; | |
1126 | if (ret) | |
1127 | goto fail; | |
1128 | ||
1129 | leaf = path->nodes[0]; | |
1130 | nritems = btrfs_header_nritems(leaf); | |
1131 | recow = 1; | |
1132 | } | |
1133 | ||
1134 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1135 | if (key.objectid != bytenr || | |
1136 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1137 | goto fail; | |
1138 | ||
1139 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1140 | struct btrfs_extent_data_ref); | |
1141 | ||
1142 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1143 | owner, offset)) { | |
1144 | if (recow) { | |
b3b4aa74 | 1145 | btrfs_release_path(path); |
5d4f98a2 YZ |
1146 | goto again; |
1147 | } | |
1148 | err = 0; | |
1149 | break; | |
1150 | } | |
1151 | path->slots[0]++; | |
31840ae1 | 1152 | } |
5d4f98a2 YZ |
1153 | fail: |
1154 | return err; | |
31840ae1 ZY |
1155 | } |
1156 | ||
5d4f98a2 YZ |
1157 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1158 | struct btrfs_root *root, | |
1159 | struct btrfs_path *path, | |
1160 | u64 bytenr, u64 parent, | |
1161 | u64 root_objectid, u64 owner, | |
1162 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1163 | { |
1164 | struct btrfs_key key; | |
1165 | struct extent_buffer *leaf; | |
5d4f98a2 | 1166 | u32 size; |
31840ae1 ZY |
1167 | u32 num_refs; |
1168 | int ret; | |
74493f7a | 1169 | |
74493f7a | 1170 | key.objectid = bytenr; |
5d4f98a2 YZ |
1171 | if (parent) { |
1172 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1173 | key.offset = parent; | |
1174 | size = sizeof(struct btrfs_shared_data_ref); | |
1175 | } else { | |
1176 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1177 | key.offset = hash_extent_data_ref(root_objectid, | |
1178 | owner, offset); | |
1179 | size = sizeof(struct btrfs_extent_data_ref); | |
1180 | } | |
74493f7a | 1181 | |
5d4f98a2 YZ |
1182 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1183 | if (ret && ret != -EEXIST) | |
1184 | goto fail; | |
1185 | ||
1186 | leaf = path->nodes[0]; | |
1187 | if (parent) { | |
1188 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1189 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1190 | struct btrfs_shared_data_ref); |
1191 | if (ret == 0) { | |
1192 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1193 | } else { | |
1194 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1195 | num_refs += refs_to_add; | |
1196 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1197 | } |
5d4f98a2 YZ |
1198 | } else { |
1199 | struct btrfs_extent_data_ref *ref; | |
1200 | while (ret == -EEXIST) { | |
1201 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1202 | struct btrfs_extent_data_ref); | |
1203 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1204 | owner, offset)) | |
1205 | break; | |
b3b4aa74 | 1206 | btrfs_release_path(path); |
5d4f98a2 YZ |
1207 | key.offset++; |
1208 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1209 | size); | |
1210 | if (ret && ret != -EEXIST) | |
1211 | goto fail; | |
31840ae1 | 1212 | |
5d4f98a2 YZ |
1213 | leaf = path->nodes[0]; |
1214 | } | |
1215 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1216 | struct btrfs_extent_data_ref); | |
1217 | if (ret == 0) { | |
1218 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1219 | root_objectid); | |
1220 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1221 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1222 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1223 | } else { | |
1224 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1225 | num_refs += refs_to_add; | |
1226 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1227 | } |
31840ae1 | 1228 | } |
5d4f98a2 YZ |
1229 | btrfs_mark_buffer_dirty(leaf); |
1230 | ret = 0; | |
1231 | fail: | |
b3b4aa74 | 1232 | btrfs_release_path(path); |
7bb86316 | 1233 | return ret; |
74493f7a CM |
1234 | } |
1235 | ||
5d4f98a2 YZ |
1236 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1237 | struct btrfs_root *root, | |
1238 | struct btrfs_path *path, | |
1239 | int refs_to_drop) | |
31840ae1 | 1240 | { |
5d4f98a2 YZ |
1241 | struct btrfs_key key; |
1242 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1243 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1244 | struct extent_buffer *leaf; |
5d4f98a2 | 1245 | u32 num_refs = 0; |
31840ae1 ZY |
1246 | int ret = 0; |
1247 | ||
1248 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1249 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1250 | ||
1251 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1252 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1253 | struct btrfs_extent_data_ref); | |
1254 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1255 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1256 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1257 | struct btrfs_shared_data_ref); | |
1258 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1259 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1260 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1261 | struct btrfs_extent_ref_v0 *ref0; | |
1262 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1263 | struct btrfs_extent_ref_v0); | |
1264 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1265 | #endif | |
1266 | } else { | |
1267 | BUG(); | |
1268 | } | |
1269 | ||
56bec294 CM |
1270 | BUG_ON(num_refs < refs_to_drop); |
1271 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1272 | |
31840ae1 ZY |
1273 | if (num_refs == 0) { |
1274 | ret = btrfs_del_item(trans, root, path); | |
1275 | } else { | |
5d4f98a2 YZ |
1276 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1277 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1278 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1279 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1280 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1281 | else { | |
1282 | struct btrfs_extent_ref_v0 *ref0; | |
1283 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1284 | struct btrfs_extent_ref_v0); | |
1285 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1286 | } | |
1287 | #endif | |
31840ae1 ZY |
1288 | btrfs_mark_buffer_dirty(leaf); |
1289 | } | |
31840ae1 ZY |
1290 | return ret; |
1291 | } | |
1292 | ||
5d4f98a2 YZ |
1293 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1294 | struct btrfs_path *path, | |
1295 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1296 | { |
5d4f98a2 YZ |
1297 | struct btrfs_key key; |
1298 | struct extent_buffer *leaf; | |
1299 | struct btrfs_extent_data_ref *ref1; | |
1300 | struct btrfs_shared_data_ref *ref2; | |
1301 | u32 num_refs = 0; | |
1302 | ||
1303 | leaf = path->nodes[0]; | |
1304 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1305 | if (iref) { | |
1306 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1307 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1308 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1309 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1310 | } else { | |
1311 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1312 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1313 | } | |
1314 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1315 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1316 | struct btrfs_extent_data_ref); | |
1317 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1318 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1319 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1320 | struct btrfs_shared_data_ref); | |
1321 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1322 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1323 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1324 | struct btrfs_extent_ref_v0 *ref0; | |
1325 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1326 | struct btrfs_extent_ref_v0); | |
1327 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1328 | #endif |
5d4f98a2 YZ |
1329 | } else { |
1330 | WARN_ON(1); | |
1331 | } | |
1332 | return num_refs; | |
1333 | } | |
15916de8 | 1334 | |
5d4f98a2 YZ |
1335 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1336 | struct btrfs_root *root, | |
1337 | struct btrfs_path *path, | |
1338 | u64 bytenr, u64 parent, | |
1339 | u64 root_objectid) | |
1f3c79a2 | 1340 | { |
5d4f98a2 | 1341 | struct btrfs_key key; |
1f3c79a2 | 1342 | int ret; |
1f3c79a2 | 1343 | |
5d4f98a2 YZ |
1344 | key.objectid = bytenr; |
1345 | if (parent) { | |
1346 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1347 | key.offset = parent; | |
1348 | } else { | |
1349 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1350 | key.offset = root_objectid; | |
1f3c79a2 LH |
1351 | } |
1352 | ||
5d4f98a2 YZ |
1353 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1354 | if (ret > 0) | |
1355 | ret = -ENOENT; | |
1356 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1357 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1358 | btrfs_release_path(path); |
5d4f98a2 YZ |
1359 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1360 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1361 | if (ret > 0) | |
1362 | ret = -ENOENT; | |
1363 | } | |
1f3c79a2 | 1364 | #endif |
5d4f98a2 | 1365 | return ret; |
1f3c79a2 LH |
1366 | } |
1367 | ||
5d4f98a2 YZ |
1368 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1369 | struct btrfs_root *root, | |
1370 | struct btrfs_path *path, | |
1371 | u64 bytenr, u64 parent, | |
1372 | u64 root_objectid) | |
31840ae1 | 1373 | { |
5d4f98a2 | 1374 | struct btrfs_key key; |
31840ae1 | 1375 | int ret; |
31840ae1 | 1376 | |
5d4f98a2 YZ |
1377 | key.objectid = bytenr; |
1378 | if (parent) { | |
1379 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1380 | key.offset = parent; | |
1381 | } else { | |
1382 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1383 | key.offset = root_objectid; | |
1384 | } | |
1385 | ||
1386 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1387 | btrfs_release_path(path); |
31840ae1 ZY |
1388 | return ret; |
1389 | } | |
1390 | ||
5d4f98a2 | 1391 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1392 | { |
5d4f98a2 YZ |
1393 | int type; |
1394 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1395 | if (parent > 0) | |
1396 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1397 | else | |
1398 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1399 | } else { | |
1400 | if (parent > 0) | |
1401 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1402 | else | |
1403 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1404 | } | |
1405 | return type; | |
31840ae1 | 1406 | } |
56bec294 | 1407 | |
2c47e605 YZ |
1408 | static int find_next_key(struct btrfs_path *path, int level, |
1409 | struct btrfs_key *key) | |
56bec294 | 1410 | |
02217ed2 | 1411 | { |
2c47e605 | 1412 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1413 | if (!path->nodes[level]) |
1414 | break; | |
5d4f98a2 YZ |
1415 | if (path->slots[level] + 1 >= |
1416 | btrfs_header_nritems(path->nodes[level])) | |
1417 | continue; | |
1418 | if (level == 0) | |
1419 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1420 | path->slots[level] + 1); | |
1421 | else | |
1422 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1423 | path->slots[level] + 1); | |
1424 | return 0; | |
1425 | } | |
1426 | return 1; | |
1427 | } | |
037e6390 | 1428 | |
5d4f98a2 YZ |
1429 | /* |
1430 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1431 | * to the address of inline back ref, and 0 is returned. | |
1432 | * | |
1433 | * if back ref isn't found, *ref_ret is set to the address where it | |
1434 | * should be inserted, and -ENOENT is returned. | |
1435 | * | |
1436 | * if insert is true and there are too many inline back refs, the path | |
1437 | * points to the extent item, and -EAGAIN is returned. | |
1438 | * | |
1439 | * NOTE: inline back refs are ordered in the same way that back ref | |
1440 | * items in the tree are ordered. | |
1441 | */ | |
1442 | static noinline_for_stack | |
1443 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1444 | struct btrfs_root *root, | |
1445 | struct btrfs_path *path, | |
1446 | struct btrfs_extent_inline_ref **ref_ret, | |
1447 | u64 bytenr, u64 num_bytes, | |
1448 | u64 parent, u64 root_objectid, | |
1449 | u64 owner, u64 offset, int insert) | |
1450 | { | |
1451 | struct btrfs_key key; | |
1452 | struct extent_buffer *leaf; | |
1453 | struct btrfs_extent_item *ei; | |
1454 | struct btrfs_extent_inline_ref *iref; | |
1455 | u64 flags; | |
1456 | u64 item_size; | |
1457 | unsigned long ptr; | |
1458 | unsigned long end; | |
1459 | int extra_size; | |
1460 | int type; | |
1461 | int want; | |
1462 | int ret; | |
1463 | int err = 0; | |
26b8003f | 1464 | |
db94535d | 1465 | key.objectid = bytenr; |
31840ae1 | 1466 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1467 | key.offset = num_bytes; |
31840ae1 | 1468 | |
5d4f98a2 YZ |
1469 | want = extent_ref_type(parent, owner); |
1470 | if (insert) { | |
1471 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1472 | path->keep_locks = 1; |
5d4f98a2 YZ |
1473 | } else |
1474 | extra_size = -1; | |
1475 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1476 | if (ret < 0) { |
5d4f98a2 YZ |
1477 | err = ret; |
1478 | goto out; | |
1479 | } | |
79787eaa JM |
1480 | if (ret && !insert) { |
1481 | err = -ENOENT; | |
1482 | goto out; | |
1483 | } | |
1484 | BUG_ON(ret); /* Corruption */ | |
5d4f98a2 YZ |
1485 | |
1486 | leaf = path->nodes[0]; | |
1487 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1488 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1489 | if (item_size < sizeof(*ei)) { | |
1490 | if (!insert) { | |
1491 | err = -ENOENT; | |
1492 | goto out; | |
1493 | } | |
1494 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1495 | extra_size); | |
1496 | if (ret < 0) { | |
1497 | err = ret; | |
1498 | goto out; | |
1499 | } | |
1500 | leaf = path->nodes[0]; | |
1501 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1502 | } | |
1503 | #endif | |
1504 | BUG_ON(item_size < sizeof(*ei)); | |
1505 | ||
5d4f98a2 YZ |
1506 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1507 | flags = btrfs_extent_flags(leaf, ei); | |
1508 | ||
1509 | ptr = (unsigned long)(ei + 1); | |
1510 | end = (unsigned long)ei + item_size; | |
1511 | ||
1512 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1513 | ptr += sizeof(struct btrfs_tree_block_info); | |
1514 | BUG_ON(ptr > end); | |
1515 | } else { | |
1516 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1517 | } | |
1518 | ||
1519 | err = -ENOENT; | |
1520 | while (1) { | |
1521 | if (ptr >= end) { | |
1522 | WARN_ON(ptr > end); | |
1523 | break; | |
1524 | } | |
1525 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1526 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1527 | if (want < type) | |
1528 | break; | |
1529 | if (want > type) { | |
1530 | ptr += btrfs_extent_inline_ref_size(type); | |
1531 | continue; | |
1532 | } | |
1533 | ||
1534 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1535 | struct btrfs_extent_data_ref *dref; | |
1536 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1537 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1538 | owner, offset)) { | |
1539 | err = 0; | |
1540 | break; | |
1541 | } | |
1542 | if (hash_extent_data_ref_item(leaf, dref) < | |
1543 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1544 | break; | |
1545 | } else { | |
1546 | u64 ref_offset; | |
1547 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1548 | if (parent > 0) { | |
1549 | if (parent == ref_offset) { | |
1550 | err = 0; | |
1551 | break; | |
1552 | } | |
1553 | if (ref_offset < parent) | |
1554 | break; | |
1555 | } else { | |
1556 | if (root_objectid == ref_offset) { | |
1557 | err = 0; | |
1558 | break; | |
1559 | } | |
1560 | if (ref_offset < root_objectid) | |
1561 | break; | |
1562 | } | |
1563 | } | |
1564 | ptr += btrfs_extent_inline_ref_size(type); | |
1565 | } | |
1566 | if (err == -ENOENT && insert) { | |
1567 | if (item_size + extra_size >= | |
1568 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1569 | err = -EAGAIN; | |
1570 | goto out; | |
1571 | } | |
1572 | /* | |
1573 | * To add new inline back ref, we have to make sure | |
1574 | * there is no corresponding back ref item. | |
1575 | * For simplicity, we just do not add new inline back | |
1576 | * ref if there is any kind of item for this block | |
1577 | */ | |
2c47e605 YZ |
1578 | if (find_next_key(path, 0, &key) == 0 && |
1579 | key.objectid == bytenr && | |
85d4198e | 1580 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1581 | err = -EAGAIN; |
1582 | goto out; | |
1583 | } | |
1584 | } | |
1585 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1586 | out: | |
85d4198e | 1587 | if (insert) { |
5d4f98a2 YZ |
1588 | path->keep_locks = 0; |
1589 | btrfs_unlock_up_safe(path, 1); | |
1590 | } | |
1591 | return err; | |
1592 | } | |
1593 | ||
1594 | /* | |
1595 | * helper to add new inline back ref | |
1596 | */ | |
1597 | static noinline_for_stack | |
143bede5 JM |
1598 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
1599 | struct btrfs_root *root, | |
1600 | struct btrfs_path *path, | |
1601 | struct btrfs_extent_inline_ref *iref, | |
1602 | u64 parent, u64 root_objectid, | |
1603 | u64 owner, u64 offset, int refs_to_add, | |
1604 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1605 | { |
1606 | struct extent_buffer *leaf; | |
1607 | struct btrfs_extent_item *ei; | |
1608 | unsigned long ptr; | |
1609 | unsigned long end; | |
1610 | unsigned long item_offset; | |
1611 | u64 refs; | |
1612 | int size; | |
1613 | int type; | |
5d4f98a2 YZ |
1614 | |
1615 | leaf = path->nodes[0]; | |
1616 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1617 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1618 | ||
1619 | type = extent_ref_type(parent, owner); | |
1620 | size = btrfs_extent_inline_ref_size(type); | |
1621 | ||
143bede5 | 1622 | btrfs_extend_item(trans, root, path, size); |
5d4f98a2 YZ |
1623 | |
1624 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1625 | refs = btrfs_extent_refs(leaf, ei); | |
1626 | refs += refs_to_add; | |
1627 | btrfs_set_extent_refs(leaf, ei, refs); | |
1628 | if (extent_op) | |
1629 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1630 | ||
1631 | ptr = (unsigned long)ei + item_offset; | |
1632 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1633 | if (ptr < end - size) | |
1634 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1635 | end - size - ptr); | |
1636 | ||
1637 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1638 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1639 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1640 | struct btrfs_extent_data_ref *dref; | |
1641 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1642 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1643 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1644 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1645 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1646 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1647 | struct btrfs_shared_data_ref *sref; | |
1648 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1649 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1650 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1651 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1652 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1653 | } else { | |
1654 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1655 | } | |
1656 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1657 | } |
1658 | ||
1659 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1660 | struct btrfs_root *root, | |
1661 | struct btrfs_path *path, | |
1662 | struct btrfs_extent_inline_ref **ref_ret, | |
1663 | u64 bytenr, u64 num_bytes, u64 parent, | |
1664 | u64 root_objectid, u64 owner, u64 offset) | |
1665 | { | |
1666 | int ret; | |
1667 | ||
1668 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1669 | bytenr, num_bytes, parent, | |
1670 | root_objectid, owner, offset, 0); | |
1671 | if (ret != -ENOENT) | |
54aa1f4d | 1672 | return ret; |
5d4f98a2 | 1673 | |
b3b4aa74 | 1674 | btrfs_release_path(path); |
5d4f98a2 YZ |
1675 | *ref_ret = NULL; |
1676 | ||
1677 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1678 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1679 | root_objectid); | |
1680 | } else { | |
1681 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1682 | root_objectid, owner, offset); | |
b9473439 | 1683 | } |
5d4f98a2 YZ |
1684 | return ret; |
1685 | } | |
31840ae1 | 1686 | |
5d4f98a2 YZ |
1687 | /* |
1688 | * helper to update/remove inline back ref | |
1689 | */ | |
1690 | static noinline_for_stack | |
143bede5 JM |
1691 | void update_inline_extent_backref(struct btrfs_trans_handle *trans, |
1692 | struct btrfs_root *root, | |
1693 | struct btrfs_path *path, | |
1694 | struct btrfs_extent_inline_ref *iref, | |
1695 | int refs_to_mod, | |
1696 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1697 | { |
1698 | struct extent_buffer *leaf; | |
1699 | struct btrfs_extent_item *ei; | |
1700 | struct btrfs_extent_data_ref *dref = NULL; | |
1701 | struct btrfs_shared_data_ref *sref = NULL; | |
1702 | unsigned long ptr; | |
1703 | unsigned long end; | |
1704 | u32 item_size; | |
1705 | int size; | |
1706 | int type; | |
5d4f98a2 YZ |
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; | |
143bede5 | 1748 | btrfs_truncate_item(trans, root, path, item_size, 1); |
5d4f98a2 YZ |
1749 | } |
1750 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1751 | } |
1752 | ||
1753 | static noinline_for_stack | |
1754 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1755 | struct btrfs_root *root, | |
1756 | struct btrfs_path *path, | |
1757 | u64 bytenr, u64 num_bytes, u64 parent, | |
1758 | u64 root_objectid, u64 owner, | |
1759 | u64 offset, int refs_to_add, | |
1760 | struct btrfs_delayed_extent_op *extent_op) | |
1761 | { | |
1762 | struct btrfs_extent_inline_ref *iref; | |
1763 | int ret; | |
1764 | ||
1765 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1766 | bytenr, num_bytes, parent, | |
1767 | root_objectid, owner, offset, 1); | |
1768 | if (ret == 0) { | |
1769 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
143bede5 JM |
1770 | update_inline_extent_backref(trans, root, path, iref, |
1771 | refs_to_add, extent_op); | |
5d4f98a2 | 1772 | } else if (ret == -ENOENT) { |
143bede5 JM |
1773 | setup_inline_extent_backref(trans, root, path, iref, parent, |
1774 | root_objectid, owner, offset, | |
1775 | refs_to_add, extent_op); | |
1776 | ret = 0; | |
771ed689 | 1777 | } |
5d4f98a2 YZ |
1778 | return ret; |
1779 | } | |
31840ae1 | 1780 | |
5d4f98a2 YZ |
1781 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1782 | struct btrfs_root *root, | |
1783 | struct btrfs_path *path, | |
1784 | u64 bytenr, u64 parent, u64 root_objectid, | |
1785 | u64 owner, u64 offset, int refs_to_add) | |
1786 | { | |
1787 | int ret; | |
1788 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1789 | BUG_ON(refs_to_add != 1); | |
1790 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1791 | parent, root_objectid); | |
1792 | } else { | |
1793 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1794 | parent, root_objectid, | |
1795 | owner, offset, refs_to_add); | |
1796 | } | |
1797 | return ret; | |
1798 | } | |
56bec294 | 1799 | |
5d4f98a2 YZ |
1800 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1801 | struct btrfs_root *root, | |
1802 | struct btrfs_path *path, | |
1803 | struct btrfs_extent_inline_ref *iref, | |
1804 | int refs_to_drop, int is_data) | |
1805 | { | |
143bede5 | 1806 | int ret = 0; |
b9473439 | 1807 | |
5d4f98a2 YZ |
1808 | BUG_ON(!is_data && refs_to_drop != 1); |
1809 | if (iref) { | |
143bede5 JM |
1810 | update_inline_extent_backref(trans, root, path, iref, |
1811 | -refs_to_drop, NULL); | |
5d4f98a2 YZ |
1812 | } else if (is_data) { |
1813 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1814 | } else { | |
1815 | ret = btrfs_del_item(trans, root, path); | |
1816 | } | |
1817 | return ret; | |
1818 | } | |
1819 | ||
5378e607 | 1820 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1821 | u64 start, u64 len) |
1822 | { | |
5378e607 | 1823 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1824 | } |
5d4f98a2 YZ |
1825 | |
1826 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1827 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1828 | { |
5d4f98a2 | 1829 | int ret; |
5378e607 | 1830 | u64 discarded_bytes = 0; |
a1d3c478 | 1831 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1832 | |
e244a0ae | 1833 | |
5d4f98a2 | 1834 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 | 1835 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
a1d3c478 | 1836 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1837 | /* Error condition is -ENOMEM */ |
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) | |
79787eaa | 1853 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
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 | ||
79787eaa | 1872 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1873 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1874 | struct btrfs_root *root, | |
1875 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1876 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1877 | { |
1878 | int ret; | |
66d7e7f0 AJ |
1879 | struct btrfs_fs_info *fs_info = root->fs_info; |
1880 | ||
5d4f98a2 YZ |
1881 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1882 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1883 | ||
1884 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1885 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1886 | num_bytes, | |
5d4f98a2 | 1887 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1888 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1889 | } else { |
66d7e7f0 AJ |
1890 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1891 | num_bytes, | |
5d4f98a2 | 1892 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1893 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1894 | } |
1895 | return ret; | |
1896 | } | |
1897 | ||
1898 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1899 | struct btrfs_root *root, | |
1900 | u64 bytenr, u64 num_bytes, | |
1901 | u64 parent, u64 root_objectid, | |
1902 | u64 owner, u64 offset, int refs_to_add, | |
1903 | struct btrfs_delayed_extent_op *extent_op) | |
1904 | { | |
1905 | struct btrfs_path *path; | |
1906 | struct extent_buffer *leaf; | |
1907 | struct btrfs_extent_item *item; | |
1908 | u64 refs; | |
1909 | int ret; | |
1910 | int err = 0; | |
1911 | ||
1912 | path = btrfs_alloc_path(); | |
1913 | if (!path) | |
1914 | return -ENOMEM; | |
1915 | ||
1916 | path->reada = 1; | |
1917 | path->leave_spinning = 1; | |
1918 | /* this will setup the path even if it fails to insert the back ref */ | |
1919 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1920 | path, bytenr, num_bytes, parent, | |
1921 | root_objectid, owner, offset, | |
1922 | refs_to_add, extent_op); | |
1923 | if (ret == 0) | |
1924 | goto out; | |
1925 | ||
1926 | if (ret != -EAGAIN) { | |
1927 | err = ret; | |
1928 | goto out; | |
1929 | } | |
1930 | ||
1931 | leaf = path->nodes[0]; | |
1932 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1933 | refs = btrfs_extent_refs(leaf, item); | |
1934 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1935 | if (extent_op) | |
1936 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1937 | |
5d4f98a2 | 1938 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1939 | btrfs_release_path(path); |
56bec294 CM |
1940 | |
1941 | path->reada = 1; | |
b9473439 CM |
1942 | path->leave_spinning = 1; |
1943 | ||
56bec294 CM |
1944 | /* now insert the actual backref */ |
1945 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1946 | path, bytenr, parent, root_objectid, |
1947 | owner, offset, refs_to_add); | |
79787eaa JM |
1948 | if (ret) |
1949 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 1950 | out: |
56bec294 | 1951 | btrfs_free_path(path); |
5d4f98a2 | 1952 | return err; |
56bec294 CM |
1953 | } |
1954 | ||
5d4f98a2 YZ |
1955 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1956 | struct btrfs_root *root, | |
1957 | struct btrfs_delayed_ref_node *node, | |
1958 | struct btrfs_delayed_extent_op *extent_op, | |
1959 | int insert_reserved) | |
56bec294 | 1960 | { |
5d4f98a2 YZ |
1961 | int ret = 0; |
1962 | struct btrfs_delayed_data_ref *ref; | |
1963 | struct btrfs_key ins; | |
1964 | u64 parent = 0; | |
1965 | u64 ref_root = 0; | |
1966 | u64 flags = 0; | |
1967 | ||
1968 | ins.objectid = node->bytenr; | |
1969 | ins.offset = node->num_bytes; | |
1970 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1971 | ||
1972 | ref = btrfs_delayed_node_to_data_ref(node); | |
1973 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1974 | parent = ref->parent; | |
1975 | else | |
1976 | ref_root = ref->root; | |
1977 | ||
1978 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1979 | if (extent_op) { | |
1980 | BUG_ON(extent_op->update_key); | |
1981 | flags |= extent_op->flags_to_set; | |
1982 | } | |
1983 | ret = alloc_reserved_file_extent(trans, root, | |
1984 | parent, ref_root, flags, | |
1985 | ref->objectid, ref->offset, | |
1986 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1987 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1988 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1989 | node->num_bytes, parent, | |
1990 | ref_root, ref->objectid, | |
1991 | ref->offset, node->ref_mod, | |
1992 | extent_op); | |
1993 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1994 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1995 | node->num_bytes, parent, | |
1996 | ref_root, ref->objectid, | |
1997 | ref->offset, node->ref_mod, | |
1998 | extent_op); | |
1999 | } else { | |
2000 | BUG(); | |
2001 | } | |
2002 | return ret; | |
2003 | } | |
2004 | ||
2005 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2006 | struct extent_buffer *leaf, | |
2007 | struct btrfs_extent_item *ei) | |
2008 | { | |
2009 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2010 | if (extent_op->update_flags) { | |
2011 | flags |= extent_op->flags_to_set; | |
2012 | btrfs_set_extent_flags(leaf, ei, flags); | |
2013 | } | |
2014 | ||
2015 | if (extent_op->update_key) { | |
2016 | struct btrfs_tree_block_info *bi; | |
2017 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2018 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2019 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2020 | } | |
2021 | } | |
2022 | ||
2023 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2024 | struct btrfs_root *root, | |
2025 | struct btrfs_delayed_ref_node *node, | |
2026 | struct btrfs_delayed_extent_op *extent_op) | |
2027 | { | |
2028 | struct btrfs_key key; | |
2029 | struct btrfs_path *path; | |
2030 | struct btrfs_extent_item *ei; | |
2031 | struct extent_buffer *leaf; | |
2032 | u32 item_size; | |
56bec294 | 2033 | int ret; |
5d4f98a2 YZ |
2034 | int err = 0; |
2035 | ||
79787eaa JM |
2036 | if (trans->aborted) |
2037 | return 0; | |
2038 | ||
5d4f98a2 YZ |
2039 | path = btrfs_alloc_path(); |
2040 | if (!path) | |
2041 | return -ENOMEM; | |
2042 | ||
2043 | key.objectid = node->bytenr; | |
2044 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2045 | key.offset = node->num_bytes; | |
2046 | ||
2047 | path->reada = 1; | |
2048 | path->leave_spinning = 1; | |
2049 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2050 | path, 0, 1); | |
2051 | if (ret < 0) { | |
2052 | err = ret; | |
2053 | goto out; | |
2054 | } | |
2055 | if (ret > 0) { | |
2056 | err = -EIO; | |
2057 | goto out; | |
2058 | } | |
2059 | ||
2060 | leaf = path->nodes[0]; | |
2061 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2062 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2063 | if (item_size < sizeof(*ei)) { | |
2064 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2065 | path, (u64)-1, 0); | |
2066 | if (ret < 0) { | |
2067 | err = ret; | |
2068 | goto out; | |
2069 | } | |
2070 | leaf = path->nodes[0]; | |
2071 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2072 | } | |
2073 | #endif | |
2074 | BUG_ON(item_size < sizeof(*ei)); | |
2075 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2076 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2077 | |
5d4f98a2 YZ |
2078 | btrfs_mark_buffer_dirty(leaf); |
2079 | out: | |
2080 | btrfs_free_path(path); | |
2081 | return err; | |
56bec294 CM |
2082 | } |
2083 | ||
5d4f98a2 YZ |
2084 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2085 | struct btrfs_root *root, | |
2086 | struct btrfs_delayed_ref_node *node, | |
2087 | struct btrfs_delayed_extent_op *extent_op, | |
2088 | int insert_reserved) | |
56bec294 CM |
2089 | { |
2090 | int ret = 0; | |
5d4f98a2 YZ |
2091 | struct btrfs_delayed_tree_ref *ref; |
2092 | struct btrfs_key ins; | |
2093 | u64 parent = 0; | |
2094 | u64 ref_root = 0; | |
56bec294 | 2095 | |
5d4f98a2 YZ |
2096 | ins.objectid = node->bytenr; |
2097 | ins.offset = node->num_bytes; | |
2098 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2099 | |
5d4f98a2 YZ |
2100 | ref = btrfs_delayed_node_to_tree_ref(node); |
2101 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2102 | parent = ref->parent; | |
2103 | else | |
2104 | ref_root = ref->root; | |
2105 | ||
2106 | BUG_ON(node->ref_mod != 1); | |
2107 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2108 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2109 | !extent_op->update_key); | |
2110 | ret = alloc_reserved_tree_block(trans, root, | |
2111 | parent, ref_root, | |
2112 | extent_op->flags_to_set, | |
2113 | &extent_op->key, | |
2114 | ref->level, &ins); | |
5d4f98a2 YZ |
2115 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2116 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2117 | node->num_bytes, parent, ref_root, | |
2118 | ref->level, 0, 1, extent_op); | |
2119 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2120 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2121 | node->num_bytes, parent, ref_root, | |
2122 | ref->level, 0, 1, extent_op); | |
2123 | } else { | |
2124 | BUG(); | |
2125 | } | |
56bec294 CM |
2126 | return ret; |
2127 | } | |
2128 | ||
2129 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2130 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2131 | struct btrfs_root *root, | |
2132 | struct btrfs_delayed_ref_node *node, | |
2133 | struct btrfs_delayed_extent_op *extent_op, | |
2134 | int insert_reserved) | |
56bec294 | 2135 | { |
79787eaa JM |
2136 | int ret = 0; |
2137 | ||
2138 | if (trans->aborted) | |
2139 | return 0; | |
2140 | ||
5d4f98a2 | 2141 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2142 | struct btrfs_delayed_ref_head *head; |
2143 | /* | |
2144 | * we've hit the end of the chain and we were supposed | |
2145 | * to insert this extent into the tree. But, it got | |
2146 | * deleted before we ever needed to insert it, so all | |
2147 | * we have to do is clean up the accounting | |
2148 | */ | |
5d4f98a2 YZ |
2149 | BUG_ON(extent_op); |
2150 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2151 | if (insert_reserved) { |
f0486c68 YZ |
2152 | btrfs_pin_extent(root, node->bytenr, |
2153 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2154 | if (head->is_data) { |
2155 | ret = btrfs_del_csums(trans, root, | |
2156 | node->bytenr, | |
2157 | node->num_bytes); | |
5d4f98a2 | 2158 | } |
56bec294 | 2159 | } |
56bec294 | 2160 | mutex_unlock(&head->mutex); |
79787eaa | 2161 | return ret; |
56bec294 CM |
2162 | } |
2163 | ||
5d4f98a2 YZ |
2164 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2165 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2166 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2167 | insert_reserved); | |
2168 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2169 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2170 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2171 | insert_reserved); | |
2172 | else | |
2173 | BUG(); | |
2174 | return ret; | |
56bec294 CM |
2175 | } |
2176 | ||
2177 | static noinline struct btrfs_delayed_ref_node * | |
2178 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2179 | { | |
2180 | struct rb_node *node; | |
2181 | struct btrfs_delayed_ref_node *ref; | |
2182 | int action = BTRFS_ADD_DELAYED_REF; | |
2183 | again: | |
2184 | /* | |
2185 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2186 | * this prevents ref count from going down to zero when | |
2187 | * there still are pending delayed ref. | |
2188 | */ | |
2189 | node = rb_prev(&head->node.rb_node); | |
2190 | while (1) { | |
2191 | if (!node) | |
2192 | break; | |
2193 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2194 | rb_node); | |
2195 | if (ref->bytenr != head->node.bytenr) | |
2196 | break; | |
5d4f98a2 | 2197 | if (ref->action == action) |
56bec294 CM |
2198 | return ref; |
2199 | node = rb_prev(node); | |
2200 | } | |
2201 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2202 | action = BTRFS_DROP_DELAYED_REF; | |
2203 | goto again; | |
2204 | } | |
2205 | return NULL; | |
2206 | } | |
2207 | ||
79787eaa JM |
2208 | /* |
2209 | * Returns 0 on success or if called with an already aborted transaction. | |
2210 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2211 | */ | |
c3e69d58 CM |
2212 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2213 | struct btrfs_root *root, | |
2214 | struct list_head *cluster) | |
56bec294 | 2215 | { |
56bec294 CM |
2216 | struct btrfs_delayed_ref_root *delayed_refs; |
2217 | struct btrfs_delayed_ref_node *ref; | |
2218 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2219 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2220 | int ret; |
c3e69d58 | 2221 | int count = 0; |
56bec294 | 2222 | int must_insert_reserved = 0; |
56bec294 CM |
2223 | |
2224 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2225 | while (1) { |
2226 | if (!locked_ref) { | |
c3e69d58 CM |
2227 | /* pick a new head ref from the cluster list */ |
2228 | if (list_empty(cluster)) | |
56bec294 | 2229 | break; |
56bec294 | 2230 | |
c3e69d58 CM |
2231 | locked_ref = list_entry(cluster->next, |
2232 | struct btrfs_delayed_ref_head, cluster); | |
2233 | ||
2234 | /* grab the lock that says we are going to process | |
2235 | * all the refs for this head */ | |
2236 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2237 | ||
2238 | /* | |
2239 | * we may have dropped the spin lock to get the head | |
2240 | * mutex lock, and that might have given someone else | |
2241 | * time to free the head. If that's true, it has been | |
2242 | * removed from our list and we can move on. | |
2243 | */ | |
2244 | if (ret == -EAGAIN) { | |
2245 | locked_ref = NULL; | |
2246 | count++; | |
2247 | continue; | |
56bec294 CM |
2248 | } |
2249 | } | |
a28ec197 | 2250 | |
d1270cd9 AJ |
2251 | /* |
2252 | * locked_ref is the head node, so we have to go one | |
2253 | * node back for any delayed ref updates | |
2254 | */ | |
2255 | ref = select_delayed_ref(locked_ref); | |
2256 | ||
2257 | if (ref && ref->seq && | |
2258 | btrfs_check_delayed_seq(delayed_refs, ref->seq)) { | |
2259 | /* | |
2260 | * there are still refs with lower seq numbers in the | |
2261 | * process of being added. Don't run this ref yet. | |
2262 | */ | |
2263 | list_del_init(&locked_ref->cluster); | |
2264 | mutex_unlock(&locked_ref->mutex); | |
2265 | locked_ref = NULL; | |
2266 | delayed_refs->num_heads_ready++; | |
2267 | spin_unlock(&delayed_refs->lock); | |
2268 | cond_resched(); | |
2269 | spin_lock(&delayed_refs->lock); | |
2270 | continue; | |
2271 | } | |
2272 | ||
56bec294 CM |
2273 | /* |
2274 | * record the must insert reserved flag before we | |
2275 | * drop the spin lock. | |
2276 | */ | |
2277 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2278 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2279 | |
5d4f98a2 YZ |
2280 | extent_op = locked_ref->extent_op; |
2281 | locked_ref->extent_op = NULL; | |
2282 | ||
56bec294 CM |
2283 | if (!ref) { |
2284 | /* All delayed refs have been processed, Go ahead | |
2285 | * and send the head node to run_one_delayed_ref, | |
2286 | * so that any accounting fixes can happen | |
2287 | */ | |
2288 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2289 | |
2290 | if (extent_op && must_insert_reserved) { | |
2291 | kfree(extent_op); | |
2292 | extent_op = NULL; | |
2293 | } | |
2294 | ||
2295 | if (extent_op) { | |
2296 | spin_unlock(&delayed_refs->lock); | |
2297 | ||
2298 | ret = run_delayed_extent_op(trans, root, | |
2299 | ref, extent_op); | |
5d4f98a2 YZ |
2300 | kfree(extent_op); |
2301 | ||
79787eaa JM |
2302 | if (ret) { |
2303 | printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret); | |
253beebd | 2304 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2305 | return ret; |
2306 | } | |
2307 | ||
203bf287 | 2308 | goto next; |
5d4f98a2 YZ |
2309 | } |
2310 | ||
c3e69d58 | 2311 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2312 | locked_ref = NULL; |
2313 | } | |
02217ed2 | 2314 | |
56bec294 CM |
2315 | ref->in_tree = 0; |
2316 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2317 | delayed_refs->num_entries--; | |
a168650c JS |
2318 | /* |
2319 | * we modified num_entries, but as we're currently running | |
2320 | * delayed refs, skip | |
2321 | * wake_up(&delayed_refs->seq_wait); | |
2322 | * here. | |
2323 | */ | |
56bec294 | 2324 | spin_unlock(&delayed_refs->lock); |
925baedd | 2325 | |
5d4f98a2 | 2326 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2327 | must_insert_reserved); |
eb099670 | 2328 | |
5d4f98a2 YZ |
2329 | btrfs_put_delayed_ref(ref); |
2330 | kfree(extent_op); | |
c3e69d58 | 2331 | count++; |
79787eaa JM |
2332 | |
2333 | if (ret) { | |
2334 | printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2335 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2336 | return ret; |
2337 | } | |
2338 | ||
203bf287 CM |
2339 | next: |
2340 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
2341 | 2 * 1024 * 1024, | |
2342 | btrfs_get_alloc_profile(root, 0), | |
2343 | CHUNK_ALLOC_NO_FORCE); | |
c3e69d58 CM |
2344 | cond_resched(); |
2345 | spin_lock(&delayed_refs->lock); | |
2346 | } | |
2347 | return count; | |
2348 | } | |
2349 | ||
a168650c JS |
2350 | |
2351 | static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs, | |
2352 | unsigned long num_refs) | |
2353 | { | |
2354 | struct list_head *first_seq = delayed_refs->seq_head.next; | |
2355 | ||
2356 | spin_unlock(&delayed_refs->lock); | |
2357 | pr_debug("waiting for more refs (num %ld, first %p)\n", | |
2358 | num_refs, first_seq); | |
2359 | wait_event(delayed_refs->seq_wait, | |
2360 | num_refs != delayed_refs->num_entries || | |
2361 | delayed_refs->seq_head.next != first_seq); | |
2362 | pr_debug("done waiting for more refs (num %ld, first %p)\n", | |
2363 | delayed_refs->num_entries, delayed_refs->seq_head.next); | |
2364 | spin_lock(&delayed_refs->lock); | |
2365 | } | |
2366 | ||
c3e69d58 CM |
2367 | /* |
2368 | * this starts processing the delayed reference count updates and | |
2369 | * extent insertions we have queued up so far. count can be | |
2370 | * 0, which means to process everything in the tree at the start | |
2371 | * of the run (but not newly added entries), or it can be some target | |
2372 | * number you'd like to process. | |
79787eaa JM |
2373 | * |
2374 | * Returns 0 on success or if called with an aborted transaction | |
2375 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2376 | */ |
2377 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2378 | struct btrfs_root *root, unsigned long count) | |
2379 | { | |
2380 | struct rb_node *node; | |
2381 | struct btrfs_delayed_ref_root *delayed_refs; | |
2382 | struct btrfs_delayed_ref_node *ref; | |
2383 | struct list_head cluster; | |
2384 | int ret; | |
a168650c | 2385 | u64 delayed_start; |
c3e69d58 CM |
2386 | int run_all = count == (unsigned long)-1; |
2387 | int run_most = 0; | |
a168650c JS |
2388 | unsigned long num_refs = 0; |
2389 | int consider_waiting; | |
c3e69d58 | 2390 | |
79787eaa JM |
2391 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2392 | if (trans->aborted) | |
2393 | return 0; | |
2394 | ||
c3e69d58 CM |
2395 | if (root == root->fs_info->extent_root) |
2396 | root = root->fs_info->tree_root; | |
2397 | ||
203bf287 CM |
2398 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2399 | 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0), | |
2400 | CHUNK_ALLOC_NO_FORCE); | |
2401 | ||
c3e69d58 CM |
2402 | delayed_refs = &trans->transaction->delayed_refs; |
2403 | INIT_LIST_HEAD(&cluster); | |
2404 | again: | |
a168650c | 2405 | consider_waiting = 0; |
c3e69d58 CM |
2406 | spin_lock(&delayed_refs->lock); |
2407 | if (count == 0) { | |
2408 | count = delayed_refs->num_entries * 2; | |
2409 | run_most = 1; | |
2410 | } | |
2411 | while (1) { | |
2412 | if (!(run_all || run_most) && | |
2413 | delayed_refs->num_heads_ready < 64) | |
2414 | break; | |
eb099670 | 2415 | |
56bec294 | 2416 | /* |
c3e69d58 CM |
2417 | * go find something we can process in the rbtree. We start at |
2418 | * the beginning of the tree, and then build a cluster | |
2419 | * of refs to process starting at the first one we are able to | |
2420 | * lock | |
56bec294 | 2421 | */ |
a168650c | 2422 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2423 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2424 | delayed_refs->run_delayed_start); | |
2425 | if (ret) | |
56bec294 CM |
2426 | break; |
2427 | ||
a168650c JS |
2428 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2429 | if (consider_waiting == 0) { | |
2430 | /* | |
2431 | * btrfs_find_ref_cluster looped. let's do one | |
2432 | * more cycle. if we don't run any delayed ref | |
2433 | * during that cycle (because we can't because | |
2434 | * all of them are blocked) and if the number of | |
2435 | * refs doesn't change, we avoid busy waiting. | |
2436 | */ | |
2437 | consider_waiting = 1; | |
2438 | num_refs = delayed_refs->num_entries; | |
2439 | } else { | |
2440 | wait_for_more_refs(delayed_refs, num_refs); | |
2441 | /* | |
2442 | * after waiting, things have changed. we | |
2443 | * dropped the lock and someone else might have | |
2444 | * run some refs, built new clusters and so on. | |
2445 | * therefore, we restart staleness detection. | |
2446 | */ | |
2447 | consider_waiting = 0; | |
2448 | } | |
2449 | } | |
2450 | ||
c3e69d58 | 2451 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa JM |
2452 | if (ret < 0) { |
2453 | spin_unlock(&delayed_refs->lock); | |
2454 | btrfs_abort_transaction(trans, root, ret); | |
2455 | return ret; | |
2456 | } | |
c3e69d58 CM |
2457 | |
2458 | count -= min_t(unsigned long, ret, count); | |
2459 | ||
2460 | if (count == 0) | |
2461 | break; | |
a168650c JS |
2462 | |
2463 | if (ret || delayed_refs->run_delayed_start == 0) { | |
2464 | /* refs were run, let's reset staleness detection */ | |
2465 | consider_waiting = 0; | |
2466 | } | |
eb099670 | 2467 | } |
c3e69d58 | 2468 | |
56bec294 | 2469 | if (run_all) { |
56bec294 | 2470 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2471 | if (!node) |
56bec294 | 2472 | goto out; |
c3e69d58 | 2473 | count = (unsigned long)-1; |
e9d0b13b | 2474 | |
56bec294 CM |
2475 | while (node) { |
2476 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2477 | rb_node); | |
2478 | if (btrfs_delayed_ref_is_head(ref)) { | |
2479 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2480 | |
56bec294 CM |
2481 | head = btrfs_delayed_node_to_head(ref); |
2482 | atomic_inc(&ref->refs); | |
2483 | ||
2484 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2485 | /* |
2486 | * Mutex was contended, block until it's | |
2487 | * released and try again | |
2488 | */ | |
56bec294 CM |
2489 | mutex_lock(&head->mutex); |
2490 | mutex_unlock(&head->mutex); | |
2491 | ||
2492 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2493 | cond_resched(); |
56bec294 CM |
2494 | goto again; |
2495 | } | |
2496 | node = rb_next(node); | |
2497 | } | |
2498 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2499 | schedule_timeout(1); |
2500 | goto again; | |
5f39d397 | 2501 | } |
54aa1f4d | 2502 | out: |
c3e69d58 | 2503 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2504 | return 0; |
2505 | } | |
2506 | ||
5d4f98a2 YZ |
2507 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2508 | struct btrfs_root *root, | |
2509 | u64 bytenr, u64 num_bytes, u64 flags, | |
2510 | int is_data) | |
2511 | { | |
2512 | struct btrfs_delayed_extent_op *extent_op; | |
2513 | int ret; | |
2514 | ||
2515 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2516 | if (!extent_op) | |
2517 | return -ENOMEM; | |
2518 | ||
2519 | extent_op->flags_to_set = flags; | |
2520 | extent_op->update_flags = 1; | |
2521 | extent_op->update_key = 0; | |
2522 | extent_op->is_data = is_data ? 1 : 0; | |
2523 | ||
66d7e7f0 AJ |
2524 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2525 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2526 | if (ret) |
2527 | kfree(extent_op); | |
2528 | return ret; | |
2529 | } | |
2530 | ||
2531 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2532 | struct btrfs_root *root, | |
2533 | struct btrfs_path *path, | |
2534 | u64 objectid, u64 offset, u64 bytenr) | |
2535 | { | |
2536 | struct btrfs_delayed_ref_head *head; | |
2537 | struct btrfs_delayed_ref_node *ref; | |
2538 | struct btrfs_delayed_data_ref *data_ref; | |
2539 | struct btrfs_delayed_ref_root *delayed_refs; | |
2540 | struct rb_node *node; | |
2541 | int ret = 0; | |
2542 | ||
2543 | ret = -ENOENT; | |
2544 | delayed_refs = &trans->transaction->delayed_refs; | |
2545 | spin_lock(&delayed_refs->lock); | |
2546 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2547 | if (!head) | |
2548 | goto out; | |
2549 | ||
2550 | if (!mutex_trylock(&head->mutex)) { | |
2551 | atomic_inc(&head->node.refs); | |
2552 | spin_unlock(&delayed_refs->lock); | |
2553 | ||
b3b4aa74 | 2554 | btrfs_release_path(path); |
5d4f98a2 | 2555 | |
8cc33e5c DS |
2556 | /* |
2557 | * Mutex was contended, block until it's released and let | |
2558 | * caller try again | |
2559 | */ | |
5d4f98a2 YZ |
2560 | mutex_lock(&head->mutex); |
2561 | mutex_unlock(&head->mutex); | |
2562 | btrfs_put_delayed_ref(&head->node); | |
2563 | return -EAGAIN; | |
2564 | } | |
2565 | ||
2566 | node = rb_prev(&head->node.rb_node); | |
2567 | if (!node) | |
2568 | goto out_unlock; | |
2569 | ||
2570 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2571 | ||
2572 | if (ref->bytenr != bytenr) | |
2573 | goto out_unlock; | |
2574 | ||
2575 | ret = 1; | |
2576 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2577 | goto out_unlock; | |
2578 | ||
2579 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2580 | ||
2581 | node = rb_prev(node); | |
2582 | if (node) { | |
2583 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2584 | if (ref->bytenr == bytenr) | |
2585 | goto out_unlock; | |
2586 | } | |
2587 | ||
2588 | if (data_ref->root != root->root_key.objectid || | |
2589 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2590 | goto out_unlock; | |
2591 | ||
2592 | ret = 0; | |
2593 | out_unlock: | |
2594 | mutex_unlock(&head->mutex); | |
2595 | out: | |
2596 | spin_unlock(&delayed_refs->lock); | |
2597 | return ret; | |
2598 | } | |
2599 | ||
2600 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2601 | struct btrfs_root *root, | |
2602 | struct btrfs_path *path, | |
2603 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2604 | { |
2605 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2606 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2607 | struct btrfs_extent_data_ref *ref; |
2608 | struct btrfs_extent_inline_ref *iref; | |
2609 | struct btrfs_extent_item *ei; | |
f321e491 | 2610 | struct btrfs_key key; |
5d4f98a2 | 2611 | u32 item_size; |
be20aa9d | 2612 | int ret; |
925baedd | 2613 | |
be20aa9d | 2614 | key.objectid = bytenr; |
31840ae1 | 2615 | key.offset = (u64)-1; |
f321e491 | 2616 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2617 | |
be20aa9d CM |
2618 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2619 | if (ret < 0) | |
2620 | goto out; | |
79787eaa | 2621 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2622 | |
2623 | ret = -ENOENT; | |
2624 | if (path->slots[0] == 0) | |
31840ae1 | 2625 | goto out; |
be20aa9d | 2626 | |
31840ae1 | 2627 | path->slots[0]--; |
f321e491 | 2628 | leaf = path->nodes[0]; |
5d4f98a2 | 2629 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2630 | |
5d4f98a2 | 2631 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2632 | goto out; |
f321e491 | 2633 | |
5d4f98a2 YZ |
2634 | ret = 1; |
2635 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2636 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2637 | if (item_size < sizeof(*ei)) { | |
2638 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2639 | goto out; | |
2640 | } | |
2641 | #endif | |
2642 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2643 | |
5d4f98a2 YZ |
2644 | if (item_size != sizeof(*ei) + |
2645 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2646 | goto out; | |
be20aa9d | 2647 | |
5d4f98a2 YZ |
2648 | if (btrfs_extent_generation(leaf, ei) <= |
2649 | btrfs_root_last_snapshot(&root->root_item)) | |
2650 | goto out; | |
2651 | ||
2652 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2653 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2654 | BTRFS_EXTENT_DATA_REF_KEY) | |
2655 | goto out; | |
2656 | ||
2657 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2658 | if (btrfs_extent_refs(leaf, ei) != | |
2659 | btrfs_extent_data_ref_count(leaf, ref) || | |
2660 | btrfs_extent_data_ref_root(leaf, ref) != | |
2661 | root->root_key.objectid || | |
2662 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2663 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2664 | goto out; | |
2665 | ||
2666 | ret = 0; | |
2667 | out: | |
2668 | return ret; | |
2669 | } | |
2670 | ||
2671 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2672 | struct btrfs_root *root, | |
2673 | u64 objectid, u64 offset, u64 bytenr) | |
2674 | { | |
2675 | struct btrfs_path *path; | |
2676 | int ret; | |
2677 | int ret2; | |
2678 | ||
2679 | path = btrfs_alloc_path(); | |
2680 | if (!path) | |
2681 | return -ENOENT; | |
2682 | ||
2683 | do { | |
2684 | ret = check_committed_ref(trans, root, path, objectid, | |
2685 | offset, bytenr); | |
2686 | if (ret && ret != -ENOENT) | |
f321e491 | 2687 | goto out; |
80ff3856 | 2688 | |
5d4f98a2 YZ |
2689 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2690 | offset, bytenr); | |
2691 | } while (ret2 == -EAGAIN); | |
2692 | ||
2693 | if (ret2 && ret2 != -ENOENT) { | |
2694 | ret = ret2; | |
2695 | goto out; | |
f321e491 | 2696 | } |
5d4f98a2 YZ |
2697 | |
2698 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2699 | ret = 0; | |
be20aa9d | 2700 | out: |
80ff3856 | 2701 | btrfs_free_path(path); |
f0486c68 YZ |
2702 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2703 | WARN_ON(ret > 0); | |
f321e491 | 2704 | return ret; |
be20aa9d | 2705 | } |
c5739bba | 2706 | |
5d4f98a2 | 2707 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2708 | struct btrfs_root *root, |
5d4f98a2 | 2709 | struct extent_buffer *buf, |
66d7e7f0 | 2710 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2711 | { |
2712 | u64 bytenr; | |
5d4f98a2 YZ |
2713 | u64 num_bytes; |
2714 | u64 parent; | |
31840ae1 | 2715 | u64 ref_root; |
31840ae1 | 2716 | u32 nritems; |
31840ae1 ZY |
2717 | struct btrfs_key key; |
2718 | struct btrfs_file_extent_item *fi; | |
2719 | int i; | |
2720 | int level; | |
2721 | int ret = 0; | |
31840ae1 | 2722 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2723 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2724 | |
2725 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2726 | nritems = btrfs_header_nritems(buf); |
2727 | level = btrfs_header_level(buf); | |
2728 | ||
5d4f98a2 YZ |
2729 | if (!root->ref_cows && level == 0) |
2730 | return 0; | |
31840ae1 | 2731 | |
5d4f98a2 YZ |
2732 | if (inc) |
2733 | process_func = btrfs_inc_extent_ref; | |
2734 | else | |
2735 | process_func = btrfs_free_extent; | |
31840ae1 | 2736 | |
5d4f98a2 YZ |
2737 | if (full_backref) |
2738 | parent = buf->start; | |
2739 | else | |
2740 | parent = 0; | |
2741 | ||
2742 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2743 | if (level == 0) { |
5d4f98a2 | 2744 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2745 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2746 | continue; | |
5d4f98a2 | 2747 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2748 | struct btrfs_file_extent_item); |
2749 | if (btrfs_file_extent_type(buf, fi) == | |
2750 | BTRFS_FILE_EXTENT_INLINE) | |
2751 | continue; | |
2752 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2753 | if (bytenr == 0) | |
2754 | continue; | |
5d4f98a2 YZ |
2755 | |
2756 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2757 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2758 | ret = process_func(trans, root, bytenr, num_bytes, | |
2759 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2760 | key.offset, for_cow); |
31840ae1 ZY |
2761 | if (ret) |
2762 | goto fail; | |
2763 | } else { | |
5d4f98a2 YZ |
2764 | bytenr = btrfs_node_blockptr(buf, i); |
2765 | num_bytes = btrfs_level_size(root, level - 1); | |
2766 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2767 | parent, ref_root, level - 1, 0, |
2768 | for_cow); | |
31840ae1 ZY |
2769 | if (ret) |
2770 | goto fail; | |
2771 | } | |
2772 | } | |
2773 | return 0; | |
2774 | fail: | |
5d4f98a2 YZ |
2775 | return ret; |
2776 | } | |
2777 | ||
2778 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2779 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2780 | { |
66d7e7f0 | 2781 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2782 | } |
2783 | ||
2784 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2785 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2786 | { |
66d7e7f0 | 2787 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2788 | } |
2789 | ||
9078a3e1 CM |
2790 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2791 | struct btrfs_root *root, | |
2792 | struct btrfs_path *path, | |
2793 | struct btrfs_block_group_cache *cache) | |
2794 | { | |
2795 | int ret; | |
9078a3e1 | 2796 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2797 | unsigned long bi; |
2798 | struct extent_buffer *leaf; | |
9078a3e1 | 2799 | |
9078a3e1 | 2800 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2801 | if (ret < 0) |
2802 | goto fail; | |
79787eaa | 2803 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
2804 | |
2805 | leaf = path->nodes[0]; | |
2806 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2807 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2808 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2809 | btrfs_release_path(path); |
54aa1f4d | 2810 | fail: |
79787eaa JM |
2811 | if (ret) { |
2812 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 2813 | return ret; |
79787eaa | 2814 | } |
9078a3e1 CM |
2815 | return 0; |
2816 | ||
2817 | } | |
2818 | ||
4a8c9a62 YZ |
2819 | static struct btrfs_block_group_cache * |
2820 | next_block_group(struct btrfs_root *root, | |
2821 | struct btrfs_block_group_cache *cache) | |
2822 | { | |
2823 | struct rb_node *node; | |
2824 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2825 | node = rb_next(&cache->cache_node); | |
2826 | btrfs_put_block_group(cache); | |
2827 | if (node) { | |
2828 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2829 | cache_node); | |
11dfe35a | 2830 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2831 | } else |
2832 | cache = NULL; | |
2833 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2834 | return cache; | |
2835 | } | |
2836 | ||
0af3d00b JB |
2837 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2838 | struct btrfs_trans_handle *trans, | |
2839 | struct btrfs_path *path) | |
2840 | { | |
2841 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2842 | struct inode *inode = NULL; | |
2843 | u64 alloc_hint = 0; | |
2b20982e | 2844 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2845 | int num_pages = 0; |
2846 | int retries = 0; | |
2847 | int ret = 0; | |
2848 | ||
2849 | /* | |
2850 | * If this block group is smaller than 100 megs don't bother caching the | |
2851 | * block group. | |
2852 | */ | |
2853 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2854 | spin_lock(&block_group->lock); | |
2855 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2856 | spin_unlock(&block_group->lock); | |
2857 | return 0; | |
2858 | } | |
2859 | ||
2860 | again: | |
2861 | inode = lookup_free_space_inode(root, block_group, path); | |
2862 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2863 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2864 | btrfs_release_path(path); |
0af3d00b JB |
2865 | goto out; |
2866 | } | |
2867 | ||
2868 | if (IS_ERR(inode)) { | |
2869 | BUG_ON(retries); | |
2870 | retries++; | |
2871 | ||
2872 | if (block_group->ro) | |
2873 | goto out_free; | |
2874 | ||
2875 | ret = create_free_space_inode(root, trans, block_group, path); | |
2876 | if (ret) | |
2877 | goto out_free; | |
2878 | goto again; | |
2879 | } | |
2880 | ||
5b0e95bf JB |
2881 | /* We've already setup this transaction, go ahead and exit */ |
2882 | if (block_group->cache_generation == trans->transid && | |
2883 | i_size_read(inode)) { | |
2884 | dcs = BTRFS_DC_SETUP; | |
2885 | goto out_put; | |
2886 | } | |
2887 | ||
0af3d00b JB |
2888 | /* |
2889 | * We want to set the generation to 0, that way if anything goes wrong | |
2890 | * from here on out we know not to trust this cache when we load up next | |
2891 | * time. | |
2892 | */ | |
2893 | BTRFS_I(inode)->generation = 0; | |
2894 | ret = btrfs_update_inode(trans, root, inode); | |
2895 | WARN_ON(ret); | |
2896 | ||
2897 | if (i_size_read(inode) > 0) { | |
2898 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2899 | inode); | |
2900 | if (ret) | |
2901 | goto out_put; | |
2902 | } | |
2903 | ||
2904 | spin_lock(&block_group->lock); | |
2905 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2906 | /* We're not cached, don't bother trying to write stuff out */ |
2907 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2908 | spin_unlock(&block_group->lock); |
2909 | goto out_put; | |
2910 | } | |
2911 | spin_unlock(&block_group->lock); | |
2912 | ||
2913 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2914 | if (!num_pages) | |
2915 | num_pages = 1; | |
2916 | ||
2917 | /* | |
2918 | * Just to make absolutely sure we have enough space, we're going to | |
2919 | * preallocate 12 pages worth of space for each block group. In | |
2920 | * practice we ought to use at most 8, but we need extra space so we can | |
2921 | * add our header and have a terminator between the extents and the | |
2922 | * bitmaps. | |
2923 | */ | |
2924 | num_pages *= 16; | |
2925 | num_pages *= PAGE_CACHE_SIZE; | |
2926 | ||
2927 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2928 | if (ret) | |
2929 | goto out_put; | |
2930 | ||
2931 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2932 | num_pages, num_pages, | |
2933 | &alloc_hint); | |
2b20982e JB |
2934 | if (!ret) |
2935 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 2936 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 2937 | |
0af3d00b JB |
2938 | out_put: |
2939 | iput(inode); | |
2940 | out_free: | |
b3b4aa74 | 2941 | btrfs_release_path(path); |
0af3d00b JB |
2942 | out: |
2943 | spin_lock(&block_group->lock); | |
e65cbb94 | 2944 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 2945 | block_group->cache_generation = trans->transid; |
2b20982e | 2946 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2947 | spin_unlock(&block_group->lock); |
2948 | ||
2949 | return ret; | |
2950 | } | |
2951 | ||
96b5179d CM |
2952 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2953 | struct btrfs_root *root) | |
9078a3e1 | 2954 | { |
4a8c9a62 | 2955 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2956 | int err = 0; |
9078a3e1 | 2957 | struct btrfs_path *path; |
96b5179d | 2958 | u64 last = 0; |
9078a3e1 CM |
2959 | |
2960 | path = btrfs_alloc_path(); | |
2961 | if (!path) | |
2962 | return -ENOMEM; | |
2963 | ||
0af3d00b JB |
2964 | again: |
2965 | while (1) { | |
2966 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2967 | while (cache) { | |
2968 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2969 | break; | |
2970 | cache = next_block_group(root, cache); | |
2971 | } | |
2972 | if (!cache) { | |
2973 | if (last == 0) | |
2974 | break; | |
2975 | last = 0; | |
2976 | continue; | |
2977 | } | |
2978 | err = cache_save_setup(cache, trans, path); | |
2979 | last = cache->key.objectid + cache->key.offset; | |
2980 | btrfs_put_block_group(cache); | |
2981 | } | |
2982 | ||
d397712b | 2983 | while (1) { |
4a8c9a62 YZ |
2984 | if (last == 0) { |
2985 | err = btrfs_run_delayed_refs(trans, root, | |
2986 | (unsigned long)-1); | |
79787eaa JM |
2987 | if (err) /* File system offline */ |
2988 | goto out; | |
0f9dd46c | 2989 | } |
54aa1f4d | 2990 | |
4a8c9a62 YZ |
2991 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2992 | while (cache) { | |
0af3d00b JB |
2993 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2994 | btrfs_put_block_group(cache); | |
2995 | goto again; | |
2996 | } | |
2997 | ||
4a8c9a62 YZ |
2998 | if (cache->dirty) |
2999 | break; | |
3000 | cache = next_block_group(root, cache); | |
3001 | } | |
3002 | if (!cache) { | |
3003 | if (last == 0) | |
3004 | break; | |
3005 | last = 0; | |
3006 | continue; | |
3007 | } | |
0f9dd46c | 3008 | |
0cb59c99 JB |
3009 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3010 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3011 | cache->dirty = 0; |
4a8c9a62 | 3012 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3013 | |
4a8c9a62 | 3014 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3015 | if (err) /* File system offline */ |
3016 | goto out; | |
3017 | ||
4a8c9a62 | 3018 | btrfs_put_block_group(cache); |
9078a3e1 | 3019 | } |
4a8c9a62 | 3020 | |
0cb59c99 JB |
3021 | while (1) { |
3022 | /* | |
3023 | * I don't think this is needed since we're just marking our | |
3024 | * preallocated extent as written, but just in case it can't | |
3025 | * hurt. | |
3026 | */ | |
3027 | if (last == 0) { | |
3028 | err = btrfs_run_delayed_refs(trans, root, | |
3029 | (unsigned long)-1); | |
79787eaa JM |
3030 | if (err) /* File system offline */ |
3031 | goto out; | |
0cb59c99 JB |
3032 | } |
3033 | ||
3034 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3035 | while (cache) { | |
3036 | /* | |
3037 | * Really this shouldn't happen, but it could if we | |
3038 | * couldn't write the entire preallocated extent and | |
3039 | * splitting the extent resulted in a new block. | |
3040 | */ | |
3041 | if (cache->dirty) { | |
3042 | btrfs_put_block_group(cache); | |
3043 | goto again; | |
3044 | } | |
3045 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3046 | break; | |
3047 | cache = next_block_group(root, cache); | |
3048 | } | |
3049 | if (!cache) { | |
3050 | if (last == 0) | |
3051 | break; | |
3052 | last = 0; | |
3053 | continue; | |
3054 | } | |
3055 | ||
79787eaa | 3056 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3057 | |
3058 | /* | |
3059 | * If we didn't have an error then the cache state is still | |
3060 | * NEED_WRITE, so we can set it to WRITTEN. | |
3061 | */ | |
79787eaa | 3062 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3063 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3064 | last = cache->key.objectid + cache->key.offset; | |
3065 | btrfs_put_block_group(cache); | |
3066 | } | |
79787eaa | 3067 | out: |
0cb59c99 | 3068 | |
9078a3e1 | 3069 | btrfs_free_path(path); |
79787eaa | 3070 | return err; |
9078a3e1 CM |
3071 | } |
3072 | ||
d2fb3437 YZ |
3073 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3074 | { | |
3075 | struct btrfs_block_group_cache *block_group; | |
3076 | int readonly = 0; | |
3077 | ||
3078 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3079 | if (!block_group || block_group->ro) | |
3080 | readonly = 1; | |
3081 | if (block_group) | |
fa9c0d79 | 3082 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3083 | return readonly; |
3084 | } | |
3085 | ||
593060d7 CM |
3086 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3087 | u64 total_bytes, u64 bytes_used, | |
3088 | struct btrfs_space_info **space_info) | |
3089 | { | |
3090 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3091 | int i; |
3092 | int factor; | |
3093 | ||
3094 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3095 | BTRFS_BLOCK_GROUP_RAID10)) | |
3096 | factor = 2; | |
3097 | else | |
3098 | factor = 1; | |
593060d7 CM |
3099 | |
3100 | found = __find_space_info(info, flags); | |
3101 | if (found) { | |
25179201 | 3102 | spin_lock(&found->lock); |
593060d7 | 3103 | found->total_bytes += total_bytes; |
89a55897 | 3104 | found->disk_total += total_bytes * factor; |
593060d7 | 3105 | found->bytes_used += bytes_used; |
b742bb82 | 3106 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3107 | found->full = 0; |
25179201 | 3108 | spin_unlock(&found->lock); |
593060d7 CM |
3109 | *space_info = found; |
3110 | return 0; | |
3111 | } | |
c146afad | 3112 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3113 | if (!found) |
3114 | return -ENOMEM; | |
3115 | ||
b742bb82 YZ |
3116 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3117 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3118 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3119 | spin_lock_init(&found->lock); |
52ba6929 | 3120 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3121 | found->total_bytes = total_bytes; |
89a55897 | 3122 | found->disk_total = total_bytes * factor; |
593060d7 | 3123 | found->bytes_used = bytes_used; |
b742bb82 | 3124 | found->disk_used = bytes_used * factor; |
593060d7 | 3125 | found->bytes_pinned = 0; |
e8569813 | 3126 | found->bytes_reserved = 0; |
c146afad | 3127 | found->bytes_readonly = 0; |
f0486c68 | 3128 | found->bytes_may_use = 0; |
593060d7 | 3129 | found->full = 0; |
0e4f8f88 | 3130 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3131 | found->chunk_alloc = 0; |
fdb5effd JB |
3132 | found->flush = 0; |
3133 | init_waitqueue_head(&found->wait); | |
593060d7 | 3134 | *space_info = found; |
4184ea7f | 3135 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
3136 | return 0; |
3137 | } | |
3138 | ||
8790d502 CM |
3139 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3140 | { | |
899c81ea ID |
3141 | u64 extra_flags = chunk_to_extended(flags) & |
3142 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 ID |
3143 | |
3144 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3145 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3146 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3147 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3148 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3149 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3150 | } |
593060d7 | 3151 | |
fc67c450 ID |
3152 | /* |
3153 | * returns target flags in extended format or 0 if restripe for this | |
3154 | * chunk_type is not in progress | |
c6664b42 ID |
3155 | * |
3156 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3157 | */ |
3158 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3159 | { | |
3160 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3161 | u64 target = 0; | |
3162 | ||
fc67c450 ID |
3163 | if (!bctl) |
3164 | return 0; | |
3165 | ||
3166 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3167 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3168 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3169 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3170 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3171 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3172 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3173 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3174 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3175 | } | |
3176 | ||
3177 | return target; | |
3178 | } | |
3179 | ||
a46d11a8 ID |
3180 | /* |
3181 | * @flags: available profiles in extended format (see ctree.h) | |
3182 | * | |
e4d8ec0f ID |
3183 | * Returns reduced profile in chunk format. If profile changing is in |
3184 | * progress (either running or paused) picks the target profile (if it's | |
3185 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3186 | */ |
2b82032c | 3187 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3188 | { |
cd02dca5 CM |
3189 | /* |
3190 | * we add in the count of missing devices because we want | |
3191 | * to make sure that any RAID levels on a degraded FS | |
3192 | * continue to be honored. | |
3193 | */ | |
3194 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3195 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3196 | u64 target; |
a061fc8d | 3197 | |
fc67c450 ID |
3198 | /* |
3199 | * see if restripe for this chunk_type is in progress, if so | |
3200 | * try to reduce to the target profile | |
3201 | */ | |
e4d8ec0f | 3202 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3203 | target = get_restripe_target(root->fs_info, flags); |
3204 | if (target) { | |
3205 | /* pick target profile only if it's already available */ | |
3206 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3207 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3208 | return extended_to_chunk(target); |
e4d8ec0f ID |
3209 | } |
3210 | } | |
3211 | spin_unlock(&root->fs_info->balance_lock); | |
3212 | ||
a061fc8d CM |
3213 | if (num_devices == 1) |
3214 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3215 | if (num_devices < 4) | |
3216 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3217 | ||
ec44a35c CM |
3218 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3219 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3220 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3221 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3222 | } |
ec44a35c CM |
3223 | |
3224 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3225 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3226 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3227 | } |
ec44a35c CM |
3228 | |
3229 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3230 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3231 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3232 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3233 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3234 | } |
3235 | ||
899c81ea | 3236 | return extended_to_chunk(flags); |
ec44a35c CM |
3237 | } |
3238 | ||
b742bb82 | 3239 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3240 | { |
b742bb82 | 3241 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3242 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3243 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3244 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3245 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3246 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3247 | ||
b742bb82 | 3248 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3249 | } |
3250 | ||
6d07bcec | 3251 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3252 | { |
b742bb82 | 3253 | u64 flags; |
9ed74f2d | 3254 | |
b742bb82 YZ |
3255 | if (data) |
3256 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3257 | else if (root == root->fs_info->chunk_root) | |
3258 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3259 | else |
b742bb82 | 3260 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3261 | |
b742bb82 | 3262 | return get_alloc_profile(root, flags); |
6a63209f | 3263 | } |
9ed74f2d | 3264 | |
6a63209f JB |
3265 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3266 | { | |
6a63209f | 3267 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3268 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3269 | } |
3270 | ||
6a63209f | 3271 | /* |
6a63209f JB |
3272 | * This will check the space that the inode allocates from to make sure we have |
3273 | * enough space for bytes. | |
6a63209f | 3274 | */ |
0ca1f7ce | 3275 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3276 | { |
6a63209f | 3277 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3278 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3279 | u64 used; |
0af3d00b | 3280 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3281 | |
6a63209f JB |
3282 | /* make sure bytes are sectorsize aligned */ |
3283 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3284 | |
82d5902d LZ |
3285 | if (root == root->fs_info->tree_root || |
3286 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3287 | alloc_chunk = 0; |
3288 | committed = 1; | |
3289 | } | |
3290 | ||
6a63209f | 3291 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3292 | if (!data_sinfo) |
3293 | goto alloc; | |
9ed74f2d | 3294 | |
6a63209f JB |
3295 | again: |
3296 | /* make sure we have enough space to handle the data first */ | |
3297 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3298 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3299 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3300 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3301 | |
3302 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3303 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3304 | |
6a63209f JB |
3305 | /* |
3306 | * if we don't have enough free bytes in this space then we need | |
3307 | * to alloc a new chunk. | |
3308 | */ | |
0af3d00b | 3309 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3310 | u64 alloc_target; |
9ed74f2d | 3311 | |
0e4f8f88 | 3312 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3313 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3314 | alloc: |
6a63209f | 3315 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3316 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3317 | if (IS_ERR(trans)) |
3318 | return PTR_ERR(trans); | |
9ed74f2d | 3319 | |
6a63209f JB |
3320 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3321 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3322 | alloc_target, |
3323 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3324 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3325 | if (ret < 0) { |
3326 | if (ret != -ENOSPC) | |
3327 | return ret; | |
3328 | else | |
3329 | goto commit_trans; | |
3330 | } | |
9ed74f2d | 3331 | |
33b4d47f CM |
3332 | if (!data_sinfo) { |
3333 | btrfs_set_inode_space_info(root, inode); | |
3334 | data_sinfo = BTRFS_I(inode)->space_info; | |
3335 | } | |
6a63209f JB |
3336 | goto again; |
3337 | } | |
f2bb8f5c JB |
3338 | |
3339 | /* | |
3340 | * If we have less pinned bytes than we want to allocate then | |
3341 | * don't bother committing the transaction, it won't help us. | |
3342 | */ | |
3343 | if (data_sinfo->bytes_pinned < bytes) | |
3344 | committed = 1; | |
6a63209f | 3345 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3346 | |
4e06bdd6 | 3347 | /* commit the current transaction and try again */ |
d52a5b5f | 3348 | commit_trans: |
a4abeea4 JB |
3349 | if (!committed && |
3350 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3351 | committed = 1; |
7a7eaa40 | 3352 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3353 | if (IS_ERR(trans)) |
3354 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3355 | ret = btrfs_commit_transaction(trans, root); |
3356 | if (ret) | |
3357 | return ret; | |
3358 | goto again; | |
3359 | } | |
9ed74f2d | 3360 | |
6a63209f JB |
3361 | return -ENOSPC; |
3362 | } | |
3363 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3364 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3365 | data_sinfo->flags, bytes, 1); |
6a63209f | 3366 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3367 | |
9ed74f2d | 3368 | return 0; |
9ed74f2d | 3369 | } |
6a63209f | 3370 | |
6a63209f | 3371 | /* |
fb25e914 | 3372 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3373 | */ |
0ca1f7ce | 3374 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3375 | { |
0ca1f7ce | 3376 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3377 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3378 | |
6a63209f JB |
3379 | /* make sure bytes are sectorsize aligned */ |
3380 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3381 | |
6a63209f JB |
3382 | data_sinfo = BTRFS_I(inode)->space_info; |
3383 | spin_lock(&data_sinfo->lock); | |
3384 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3385 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3386 | data_sinfo->flags, bytes, 0); |
6a63209f | 3387 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3388 | } |
3389 | ||
97e728d4 | 3390 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3391 | { |
97e728d4 JB |
3392 | struct list_head *head = &info->space_info; |
3393 | struct btrfs_space_info *found; | |
e3ccfa98 | 3394 | |
97e728d4 JB |
3395 | rcu_read_lock(); |
3396 | list_for_each_entry_rcu(found, head, list) { | |
3397 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3398 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3399 | } |
97e728d4 | 3400 | rcu_read_unlock(); |
e3ccfa98 JB |
3401 | } |
3402 | ||
e5bc2458 | 3403 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3404 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3405 | int force) | |
32c00aff | 3406 | { |
fb25e914 | 3407 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3408 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3409 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3410 | u64 thresh; |
e3ccfa98 | 3411 | |
0e4f8f88 CM |
3412 | if (force == CHUNK_ALLOC_FORCE) |
3413 | return 1; | |
3414 | ||
fb25e914 JB |
3415 | /* |
3416 | * We need to take into account the global rsv because for all intents | |
3417 | * and purposes it's used space. Don't worry about locking the | |
3418 | * global_rsv, it doesn't change except when the transaction commits. | |
3419 | */ | |
3420 | num_allocated += global_rsv->size; | |
3421 | ||
0e4f8f88 CM |
3422 | /* |
3423 | * in limited mode, we want to have some free space up to | |
3424 | * about 1% of the FS size. | |
3425 | */ | |
3426 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3427 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3428 | thresh = max_t(u64, 64 * 1024 * 1024, |
3429 | div_factor_fine(thresh, 1)); | |
3430 | ||
3431 | if (num_bytes - num_allocated < thresh) | |
3432 | return 1; | |
3433 | } | |
6c41761f | 3434 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 | 3435 | |
cf1d72c9 CM |
3436 | /* 256MB or 2% of the FS */ |
3437 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2)); | |
96bdc7dc CM |
3438 | /* system chunks need a much small threshold */ |
3439 | if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3440 | thresh = 32 * 1024 * 1024; | |
e5bc2458 | 3441 | |
cf1d72c9 | 3442 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8)) |
14ed0ca6 | 3443 | return 0; |
424499db | 3444 | return 1; |
32c00aff JB |
3445 | } |
3446 | ||
15d1ff81 LB |
3447 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3448 | { | |
3449 | u64 num_dev; | |
3450 | ||
3451 | if (type & BTRFS_BLOCK_GROUP_RAID10 || | |
3452 | type & BTRFS_BLOCK_GROUP_RAID0) | |
3453 | num_dev = root->fs_info->fs_devices->rw_devices; | |
3454 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3455 | num_dev = 2; | |
3456 | else | |
3457 | num_dev = 1; /* DUP or single */ | |
3458 | ||
3459 | /* metadata for updaing devices and chunk tree */ | |
3460 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3461 | } | |
3462 | ||
3463 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3464 | struct btrfs_root *root, u64 type) | |
3465 | { | |
3466 | struct btrfs_space_info *info; | |
3467 | u64 left; | |
3468 | u64 thresh; | |
3469 | ||
3470 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3471 | spin_lock(&info->lock); | |
3472 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3473 | info->bytes_reserved - info->bytes_readonly; | |
3474 | spin_unlock(&info->lock); | |
3475 | ||
3476 | thresh = get_system_chunk_thresh(root, type); | |
3477 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3478 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3479 | left, thresh, type); | |
3480 | dump_space_info(info, 0, 0); | |
3481 | } | |
3482 | ||
3483 | if (left < thresh) { | |
3484 | u64 flags; | |
3485 | ||
3486 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3487 | btrfs_alloc_chunk(trans, root, flags); | |
3488 | } | |
3489 | } | |
3490 | ||
6324fbf3 CM |
3491 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3492 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3493 | u64 flags, int force) |
9ed74f2d | 3494 | { |
6324fbf3 | 3495 | struct btrfs_space_info *space_info; |
97e728d4 | 3496 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3497 | int wait_for_alloc = 0; |
9ed74f2d | 3498 | int ret = 0; |
9ed74f2d | 3499 | |
6324fbf3 | 3500 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3501 | if (!space_info) { |
3502 | ret = update_space_info(extent_root->fs_info, flags, | |
3503 | 0, 0, &space_info); | |
79787eaa | 3504 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3505 | } |
79787eaa | 3506 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3507 | |
6d74119f | 3508 | again: |
25179201 | 3509 | spin_lock(&space_info->lock); |
9e622d6b | 3510 | if (force < space_info->force_alloc) |
0e4f8f88 | 3511 | force = space_info->force_alloc; |
25179201 JB |
3512 | if (space_info->full) { |
3513 | spin_unlock(&space_info->lock); | |
6d74119f | 3514 | return 0; |
9ed74f2d JB |
3515 | } |
3516 | ||
0e4f8f88 | 3517 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3518 | spin_unlock(&space_info->lock); |
6d74119f JB |
3519 | return 0; |
3520 | } else if (space_info->chunk_alloc) { | |
3521 | wait_for_alloc = 1; | |
3522 | } else { | |
3523 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3524 | } |
0e4f8f88 | 3525 | |
25179201 | 3526 | spin_unlock(&space_info->lock); |
9ed74f2d | 3527 | |
6d74119f JB |
3528 | mutex_lock(&fs_info->chunk_mutex); |
3529 | ||
3530 | /* | |
3531 | * The chunk_mutex is held throughout the entirety of a chunk | |
3532 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3533 | * other guy is done and we need to recheck and see if we should | |
3534 | * allocate. | |
3535 | */ | |
3536 | if (wait_for_alloc) { | |
3537 | mutex_unlock(&fs_info->chunk_mutex); | |
3538 | wait_for_alloc = 0; | |
3539 | goto again; | |
3540 | } | |
3541 | ||
67377734 JB |
3542 | /* |
3543 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3544 | * allocating mixed chunks instead of individual chunks. | |
3545 | */ | |
3546 | if (btrfs_mixed_space_info(space_info)) | |
3547 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3548 | ||
97e728d4 JB |
3549 | /* |
3550 | * if we're doing a data chunk, go ahead and make sure that | |
3551 | * we keep a reasonable number of metadata chunks allocated in the | |
3552 | * FS as well. | |
3553 | */ | |
9ed74f2d | 3554 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3555 | fs_info->data_chunk_allocations++; |
3556 | if (!(fs_info->data_chunk_allocations % | |
3557 | fs_info->metadata_ratio)) | |
3558 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3559 | } |
3560 | ||
15d1ff81 LB |
3561 | /* |
3562 | * Check if we have enough space in SYSTEM chunk because we may need | |
3563 | * to update devices. | |
3564 | */ | |
3565 | check_system_chunk(trans, extent_root, flags); | |
3566 | ||
2b82032c | 3567 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3568 | if (ret < 0 && ret != -ENOSPC) |
3569 | goto out; | |
3570 | ||
9ed74f2d | 3571 | spin_lock(&space_info->lock); |
9ed74f2d | 3572 | if (ret) |
6324fbf3 | 3573 | space_info->full = 1; |
424499db YZ |
3574 | else |
3575 | ret = 1; | |
6d74119f | 3576 | |
0e4f8f88 | 3577 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3578 | space_info->chunk_alloc = 0; |
9ed74f2d | 3579 | spin_unlock(&space_info->lock); |
92b8e897 | 3580 | out: |
a25c75d5 | 3581 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3582 | return ret; |
6324fbf3 | 3583 | } |
9ed74f2d | 3584 | |
9ed74f2d | 3585 | /* |
5da9d01b | 3586 | * shrink metadata reservation for delalloc |
9ed74f2d | 3587 | */ |
663350ac | 3588 | static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, |
f104d044 | 3589 | bool wait_ordered) |
5da9d01b | 3590 | { |
0ca1f7ce | 3591 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3592 | struct btrfs_space_info *space_info; |
663350ac | 3593 | struct btrfs_trans_handle *trans; |
5da9d01b YZ |
3594 | u64 reserved; |
3595 | u64 max_reclaim; | |
3596 | u64 reclaimed = 0; | |
b1953bce | 3597 | long time_left; |
877da174 | 3598 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3599 | int loops = 0; |
36e39c40 | 3600 | unsigned long progress; |
5da9d01b | 3601 | |
663350ac | 3602 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3603 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3604 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3605 | |
3606 | smp_mb(); | |
fb25e914 | 3607 | reserved = space_info->bytes_may_use; |
36e39c40 | 3608 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3609 | |
3610 | if (reserved == 0) | |
3611 | return 0; | |
c4f675cd | 3612 | |
fdb5effd JB |
3613 | smp_mb(); |
3614 | if (root->fs_info->delalloc_bytes == 0) { | |
3615 | if (trans) | |
3616 | return 0; | |
3617 | btrfs_wait_ordered_extents(root, 0, 0); | |
3618 | return 0; | |
3619 | } | |
3620 | ||
5da9d01b | 3621 | max_reclaim = min(reserved, to_reclaim); |
877da174 JB |
3622 | nr_pages = max_t(unsigned long, nr_pages, |
3623 | max_reclaim >> PAGE_CACHE_SHIFT); | |
b1953bce | 3624 | while (loops < 1024) { |
bf9022e0 CM |
3625 | /* have the flusher threads jump in and do some IO */ |
3626 | smp_mb(); | |
3627 | nr_pages = min_t(unsigned long, nr_pages, | |
3628 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
0e175a18 CW |
3629 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages, |
3630 | WB_REASON_FS_FREE_SPACE); | |
5da9d01b | 3631 | |
0019f10d | 3632 | spin_lock(&space_info->lock); |
fb25e914 JB |
3633 | if (reserved > space_info->bytes_may_use) |
3634 | reclaimed += reserved - space_info->bytes_may_use; | |
3635 | reserved = space_info->bytes_may_use; | |
0019f10d | 3636 | spin_unlock(&space_info->lock); |
5da9d01b | 3637 | |
36e39c40 CM |
3638 | loops++; |
3639 | ||
5da9d01b YZ |
3640 | if (reserved == 0 || reclaimed >= max_reclaim) |
3641 | break; | |
3642 | ||
3643 | if (trans && trans->transaction->blocked) | |
3644 | return -EAGAIN; | |
bf9022e0 | 3645 | |
f104d044 JB |
3646 | if (wait_ordered && !trans) { |
3647 | btrfs_wait_ordered_extents(root, 0, 0); | |
3648 | } else { | |
3649 | time_left = schedule_timeout_interruptible(1); | |
b1953bce | 3650 | |
f104d044 JB |
3651 | /* We were interrupted, exit */ |
3652 | if (time_left) | |
3653 | break; | |
3654 | } | |
b1953bce | 3655 | |
36e39c40 CM |
3656 | /* we've kicked the IO a few times, if anything has been freed, |
3657 | * exit. There is no sense in looping here for a long time | |
3658 | * when we really need to commit the transaction, or there are | |
3659 | * just too many writers without enough free space | |
3660 | */ | |
3661 | ||
3662 | if (loops > 3) { | |
3663 | smp_mb(); | |
3664 | if (progress != space_info->reservation_progress) | |
3665 | break; | |
3666 | } | |
bf9022e0 | 3667 | |
5da9d01b | 3668 | } |
f104d044 | 3669 | |
5da9d01b YZ |
3670 | return reclaimed >= to_reclaim; |
3671 | } | |
3672 | ||
663350ac JB |
3673 | /** |
3674 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3675 | * @root - the root we're allocating for | |
3676 | * @bytes - the number of bytes we want to reserve | |
3677 | * @force - force the commit | |
8bb8ab2e | 3678 | * |
663350ac JB |
3679 | * This will check to make sure that committing the transaction will actually |
3680 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3681 | * will return -ENOSPC. | |
8bb8ab2e | 3682 | */ |
663350ac JB |
3683 | static int may_commit_transaction(struct btrfs_root *root, |
3684 | struct btrfs_space_info *space_info, | |
3685 | u64 bytes, int force) | |
3686 | { | |
3687 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3688 | struct btrfs_trans_handle *trans; | |
3689 | ||
3690 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3691 | if (trans) | |
3692 | return -EAGAIN; | |
3693 | ||
3694 | if (force) | |
3695 | goto commit; | |
3696 | ||
3697 | /* See if there is enough pinned space to make this reservation */ | |
3698 | spin_lock(&space_info->lock); | |
3699 | if (space_info->bytes_pinned >= bytes) { | |
3700 | spin_unlock(&space_info->lock); | |
3701 | goto commit; | |
3702 | } | |
3703 | spin_unlock(&space_info->lock); | |
3704 | ||
3705 | /* | |
3706 | * See if there is some space in the delayed insertion reservation for | |
3707 | * this reservation. | |
3708 | */ | |
3709 | if (space_info != delayed_rsv->space_info) | |
3710 | return -ENOSPC; | |
3711 | ||
d9b0218f | 3712 | spin_lock(&space_info->lock); |
663350ac | 3713 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3714 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3715 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3716 | spin_unlock(&space_info->lock); |
663350ac JB |
3717 | return -ENOSPC; |
3718 | } | |
3719 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3720 | spin_unlock(&space_info->lock); |
663350ac JB |
3721 | |
3722 | commit: | |
3723 | trans = btrfs_join_transaction(root); | |
3724 | if (IS_ERR(trans)) | |
3725 | return -ENOSPC; | |
3726 | ||
3727 | return btrfs_commit_transaction(trans, root); | |
3728 | } | |
3729 | ||
4a92b1b8 JB |
3730 | /** |
3731 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3732 | * @root - the root we're allocating for | |
3733 | * @block_rsv - the block_rsv we're allocating for | |
3734 | * @orig_bytes - the number of bytes we want | |
3735 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3736 | * |
4a92b1b8 JB |
3737 | * This will reserve orgi_bytes number of bytes from the space info associated |
3738 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3739 | * flush out space to make room. It will do this by flushing delalloc if | |
3740 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3741 | * regain reservations will be made and this will fail if there is not enough | |
3742 | * space already. | |
8bb8ab2e | 3743 | */ |
4a92b1b8 | 3744 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e JB |
3745 | struct btrfs_block_rsv *block_rsv, |
3746 | u64 orig_bytes, int flush) | |
9ed74f2d | 3747 | { |
f0486c68 | 3748 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3749 | u64 used; |
8bb8ab2e JB |
3750 | u64 num_bytes = orig_bytes; |
3751 | int retries = 0; | |
3752 | int ret = 0; | |
38227933 | 3753 | bool committed = false; |
fdb5effd | 3754 | bool flushing = false; |
f104d044 | 3755 | bool wait_ordered = false; |
9ed74f2d | 3756 | |
8bb8ab2e | 3757 | again: |
fdb5effd | 3758 | ret = 0; |
8bb8ab2e | 3759 | spin_lock(&space_info->lock); |
fdb5effd JB |
3760 | /* |
3761 | * We only want to wait if somebody other than us is flushing and we are | |
3762 | * actually alloed to flush. | |
3763 | */ | |
3764 | while (flush && !flushing && space_info->flush) { | |
3765 | spin_unlock(&space_info->lock); | |
3766 | /* | |
3767 | * If we have a trans handle we can't wait because the flusher | |
3768 | * may have to commit the transaction, which would mean we would | |
3769 | * deadlock since we are waiting for the flusher to finish, but | |
3770 | * hold the current transaction open. | |
3771 | */ | |
663350ac | 3772 | if (current->journal_info) |
fdb5effd | 3773 | return -EAGAIN; |
b9688bb8 AJ |
3774 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
3775 | /* Must have been killed, return */ | |
3776 | if (ret) | |
fdb5effd JB |
3777 | return -EINTR; |
3778 | ||
3779 | spin_lock(&space_info->lock); | |
3780 | } | |
3781 | ||
3782 | ret = -ENOSPC; | |
2bf64758 JB |
3783 | used = space_info->bytes_used + space_info->bytes_reserved + |
3784 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3785 | space_info->bytes_may_use; | |
9ed74f2d | 3786 | |
8bb8ab2e JB |
3787 | /* |
3788 | * The idea here is that we've not already over-reserved the block group | |
3789 | * then we can go ahead and save our reservation first and then start | |
3790 | * flushing if we need to. Otherwise if we've already overcommitted | |
3791 | * lets start flushing stuff first and then come back and try to make | |
3792 | * our reservation. | |
3793 | */ | |
2bf64758 JB |
3794 | if (used <= space_info->total_bytes) { |
3795 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3796 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3797 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3798 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
3799 | ret = 0; |
3800 | } else { | |
3801 | /* | |
3802 | * Ok set num_bytes to orig_bytes since we aren't | |
3803 | * overocmmitted, this way we only try and reclaim what | |
3804 | * we need. | |
3805 | */ | |
3806 | num_bytes = orig_bytes; | |
3807 | } | |
3808 | } else { | |
3809 | /* | |
3810 | * Ok we're over committed, set num_bytes to the overcommitted | |
3811 | * amount plus the amount of bytes that we need for this | |
3812 | * reservation. | |
3813 | */ | |
f104d044 | 3814 | wait_ordered = true; |
2bf64758 | 3815 | num_bytes = used - space_info->total_bytes + |
8bb8ab2e JB |
3816 | (orig_bytes * (retries + 1)); |
3817 | } | |
9ed74f2d | 3818 | |
36ba022a | 3819 | if (ret) { |
2bf64758 JB |
3820 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3821 | u64 avail; | |
3822 | ||
7e355b83 JB |
3823 | /* |
3824 | * If we have a lot of space that's pinned, don't bother doing | |
3825 | * the overcommit dance yet and just commit the transaction. | |
3826 | */ | |
3827 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3828 | do_div(avail, 10); | |
663350ac | 3829 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3830 | space_info->flush = 1; |
3831 | flushing = true; | |
3832 | spin_unlock(&space_info->lock); | |
663350ac JB |
3833 | ret = may_commit_transaction(root, space_info, |
3834 | orig_bytes, 1); | |
3835 | if (ret) | |
3836 | goto out; | |
3837 | committed = true; | |
3838 | goto again; | |
7e355b83 JB |
3839 | } |
3840 | ||
2bf64758 JB |
3841 | spin_lock(&root->fs_info->free_chunk_lock); |
3842 | avail = root->fs_info->free_chunk_space; | |
3843 | ||
3844 | /* | |
3845 | * If we have dup, raid1 or raid10 then only half of the free | |
3846 | * space is actually useable. | |
3847 | */ | |
3848 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3849 | BTRFS_BLOCK_GROUP_RAID1 | | |
3850 | BTRFS_BLOCK_GROUP_RAID10)) | |
3851 | avail >>= 1; | |
3852 | ||
3853 | /* | |
3854 | * If we aren't flushing don't let us overcommit too much, say | |
3855 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3856 | * 1/2 of the space. | |
3857 | */ | |
3858 | if (flush) | |
3859 | avail >>= 3; | |
3860 | else | |
3861 | avail >>= 1; | |
3862 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3863 | ||
9a82ca65 | 3864 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 | 3865 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3866 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3867 | "space_info", space_info->flags, orig_bytes, 1); |
2bf64758 | 3868 | ret = 0; |
f104d044 JB |
3869 | } else { |
3870 | wait_ordered = true; | |
2bf64758 JB |
3871 | } |
3872 | } | |
3873 | ||
8bb8ab2e JB |
3874 | /* |
3875 | * Couldn't make our reservation, save our place so while we're trying | |
3876 | * to reclaim space we can actually use it instead of somebody else | |
3877 | * stealing it from us. | |
3878 | */ | |
fdb5effd JB |
3879 | if (ret && flush) { |
3880 | flushing = true; | |
3881 | space_info->flush = 1; | |
8bb8ab2e | 3882 | } |
9ed74f2d | 3883 | |
f0486c68 | 3884 | spin_unlock(&space_info->lock); |
9ed74f2d | 3885 | |
fdb5effd | 3886 | if (!ret || !flush) |
8bb8ab2e | 3887 | goto out; |
f0486c68 | 3888 | |
8bb8ab2e JB |
3889 | /* |
3890 | * We do synchronous shrinking since we don't actually unreserve | |
3891 | * metadata until after the IO is completed. | |
3892 | */ | |
663350ac | 3893 | ret = shrink_delalloc(root, num_bytes, wait_ordered); |
fdb5effd | 3894 | if (ret < 0) |
8bb8ab2e | 3895 | goto out; |
f0486c68 | 3896 | |
75c195a2 CM |
3897 | ret = 0; |
3898 | ||
8bb8ab2e JB |
3899 | /* |
3900 | * So if we were overcommitted it's possible that somebody else flushed | |
3901 | * out enough space and we simply didn't have enough space to reclaim, | |
3902 | * so go back around and try again. | |
3903 | */ | |
3904 | if (retries < 2) { | |
f104d044 | 3905 | wait_ordered = true; |
8bb8ab2e JB |
3906 | retries++; |
3907 | goto again; | |
3908 | } | |
f0486c68 | 3909 | |
8bb8ab2e | 3910 | ret = -ENOSPC; |
75c195a2 CM |
3911 | if (committed) |
3912 | goto out; | |
3913 | ||
663350ac | 3914 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); |
38227933 | 3915 | if (!ret) { |
38227933 | 3916 | committed = true; |
8bb8ab2e | 3917 | goto again; |
38227933 | 3918 | } |
8bb8ab2e JB |
3919 | |
3920 | out: | |
fdb5effd | 3921 | if (flushing) { |
8bb8ab2e | 3922 | spin_lock(&space_info->lock); |
fdb5effd JB |
3923 | space_info->flush = 0; |
3924 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3925 | spin_unlock(&space_info->lock); |
f0486c68 | 3926 | } |
f0486c68 YZ |
3927 | return ret; |
3928 | } | |
3929 | ||
79787eaa JM |
3930 | static struct btrfs_block_rsv *get_block_rsv( |
3931 | const struct btrfs_trans_handle *trans, | |
3932 | const struct btrfs_root *root) | |
f0486c68 | 3933 | { |
4c13d758 JB |
3934 | struct btrfs_block_rsv *block_rsv = NULL; |
3935 | ||
3936 | if (root->ref_cows || root == root->fs_info->csum_root) | |
f0486c68 | 3937 | block_rsv = trans->block_rsv; |
4c13d758 JB |
3938 | |
3939 | if (!block_rsv) | |
f0486c68 YZ |
3940 | block_rsv = root->block_rsv; |
3941 | ||
3942 | if (!block_rsv) | |
3943 | block_rsv = &root->fs_info->empty_block_rsv; | |
3944 | ||
3945 | return block_rsv; | |
3946 | } | |
3947 | ||
3948 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3949 | u64 num_bytes) | |
3950 | { | |
3951 | int ret = -ENOSPC; | |
3952 | spin_lock(&block_rsv->lock); | |
3953 | if (block_rsv->reserved >= num_bytes) { | |
3954 | block_rsv->reserved -= num_bytes; | |
3955 | if (block_rsv->reserved < block_rsv->size) | |
3956 | block_rsv->full = 0; | |
3957 | ret = 0; | |
3958 | } | |
3959 | spin_unlock(&block_rsv->lock); | |
3960 | return ret; | |
3961 | } | |
3962 | ||
3963 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3964 | u64 num_bytes, int update_size) | |
3965 | { | |
3966 | spin_lock(&block_rsv->lock); | |
3967 | block_rsv->reserved += num_bytes; | |
3968 | if (update_size) | |
3969 | block_rsv->size += num_bytes; | |
3970 | else if (block_rsv->reserved >= block_rsv->size) | |
3971 | block_rsv->full = 1; | |
3972 | spin_unlock(&block_rsv->lock); | |
3973 | } | |
3974 | ||
8c2a3ca2 JB |
3975 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
3976 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 3977 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
3978 | { |
3979 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3980 | ||
3981 | spin_lock(&block_rsv->lock); | |
3982 | if (num_bytes == (u64)-1) | |
3983 | num_bytes = block_rsv->size; | |
3984 | block_rsv->size -= num_bytes; | |
3985 | if (block_rsv->reserved >= block_rsv->size) { | |
3986 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3987 | block_rsv->reserved = block_rsv->size; | |
3988 | block_rsv->full = 1; | |
3989 | } else { | |
3990 | num_bytes = 0; | |
3991 | } | |
3992 | spin_unlock(&block_rsv->lock); | |
3993 | ||
3994 | if (num_bytes > 0) { | |
3995 | if (dest) { | |
e9e22899 JB |
3996 | spin_lock(&dest->lock); |
3997 | if (!dest->full) { | |
3998 | u64 bytes_to_add; | |
3999 | ||
4000 | bytes_to_add = dest->size - dest->reserved; | |
4001 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4002 | dest->reserved += bytes_to_add; | |
4003 | if (dest->reserved >= dest->size) | |
4004 | dest->full = 1; | |
4005 | num_bytes -= bytes_to_add; | |
4006 | } | |
4007 | spin_unlock(&dest->lock); | |
4008 | } | |
4009 | if (num_bytes) { | |
f0486c68 | 4010 | spin_lock(&space_info->lock); |
fb25e914 | 4011 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4012 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4013 | space_info->flags, num_bytes, 0); |
36e39c40 | 4014 | space_info->reservation_progress++; |
f0486c68 | 4015 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4016 | } |
9ed74f2d | 4017 | } |
f0486c68 | 4018 | } |
4e06bdd6 | 4019 | |
f0486c68 YZ |
4020 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4021 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4022 | { | |
4023 | int ret; | |
9ed74f2d | 4024 | |
f0486c68 YZ |
4025 | ret = block_rsv_use_bytes(src, num_bytes); |
4026 | if (ret) | |
4027 | return ret; | |
9ed74f2d | 4028 | |
f0486c68 | 4029 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4030 | return 0; |
4031 | } | |
4032 | ||
f0486c68 | 4033 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 4034 | { |
f0486c68 YZ |
4035 | memset(rsv, 0, sizeof(*rsv)); |
4036 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
4037 | } |
4038 | ||
4039 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
4040 | { | |
4041 | struct btrfs_block_rsv *block_rsv; | |
4042 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4043 | |
f0486c68 YZ |
4044 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4045 | if (!block_rsv) | |
4046 | return NULL; | |
9ed74f2d | 4047 | |
f0486c68 | 4048 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
4049 | block_rsv->space_info = __find_space_info(fs_info, |
4050 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4051 | return block_rsv; |
4052 | } | |
9ed74f2d | 4053 | |
f0486c68 YZ |
4054 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4055 | struct btrfs_block_rsv *rsv) | |
4056 | { | |
dabdb640 JB |
4057 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4058 | kfree(rsv); | |
9ed74f2d JB |
4059 | } |
4060 | ||
61b520a9 MX |
4061 | static inline int __block_rsv_add(struct btrfs_root *root, |
4062 | struct btrfs_block_rsv *block_rsv, | |
4063 | u64 num_bytes, int flush) | |
9ed74f2d | 4064 | { |
f0486c68 | 4065 | int ret; |
9ed74f2d | 4066 | |
f0486c68 YZ |
4067 | if (num_bytes == 0) |
4068 | return 0; | |
8bb8ab2e | 4069 | |
61b520a9 | 4070 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4071 | if (!ret) { |
4072 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4073 | return 0; | |
4074 | } | |
9ed74f2d | 4075 | |
f0486c68 | 4076 | return ret; |
f0486c68 | 4077 | } |
9ed74f2d | 4078 | |
61b520a9 MX |
4079 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4080 | struct btrfs_block_rsv *block_rsv, | |
4081 | u64 num_bytes) | |
4082 | { | |
4083 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
4084 | } | |
4085 | ||
c06a0e12 JB |
4086 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
4087 | struct btrfs_block_rsv *block_rsv, | |
4088 | u64 num_bytes) | |
f0486c68 | 4089 | { |
61b520a9 | 4090 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
f0486c68 | 4091 | } |
9ed74f2d | 4092 | |
4a92b1b8 | 4093 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4094 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4095 | { |
4096 | u64 num_bytes = 0; | |
f0486c68 | 4097 | int ret = -ENOSPC; |
9ed74f2d | 4098 | |
f0486c68 YZ |
4099 | if (!block_rsv) |
4100 | return 0; | |
9ed74f2d | 4101 | |
f0486c68 | 4102 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4103 | num_bytes = div_factor(block_rsv->size, min_factor); |
4104 | if (block_rsv->reserved >= num_bytes) | |
4105 | ret = 0; | |
4106 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4107 | |
36ba022a JB |
4108 | return ret; |
4109 | } | |
4110 | ||
aa38a711 MX |
4111 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
4112 | struct btrfs_block_rsv *block_rsv, | |
4113 | u64 min_reserved, int flush) | |
36ba022a JB |
4114 | { |
4115 | u64 num_bytes = 0; | |
4116 | int ret = -ENOSPC; | |
4117 | ||
4118 | if (!block_rsv) | |
4119 | return 0; | |
4120 | ||
4121 | spin_lock(&block_rsv->lock); | |
4122 | num_bytes = min_reserved; | |
13553e52 | 4123 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4124 | ret = 0; |
13553e52 | 4125 | else |
f0486c68 | 4126 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4127 | spin_unlock(&block_rsv->lock); |
13553e52 | 4128 | |
f0486c68 YZ |
4129 | if (!ret) |
4130 | return 0; | |
4131 | ||
aa38a711 | 4132 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4133 | if (!ret) { |
4134 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4135 | return 0; |
6a63209f | 4136 | } |
9ed74f2d | 4137 | |
13553e52 | 4138 | return ret; |
f0486c68 YZ |
4139 | } |
4140 | ||
aa38a711 MX |
4141 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4142 | struct btrfs_block_rsv *block_rsv, | |
4143 | u64 min_reserved) | |
4144 | { | |
4145 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
4146 | } | |
4147 | ||
4148 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
4149 | struct btrfs_block_rsv *block_rsv, | |
4150 | u64 min_reserved) | |
4151 | { | |
4152 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
4153 | } | |
4154 | ||
f0486c68 YZ |
4155 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
4156 | struct btrfs_block_rsv *dst_rsv, | |
4157 | u64 num_bytes) | |
4158 | { | |
4159 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4160 | } | |
4161 | ||
4162 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4163 | struct btrfs_block_rsv *block_rsv, | |
4164 | u64 num_bytes) | |
4165 | { | |
4166 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4167 | if (global_rsv->full || global_rsv == block_rsv || | |
4168 | block_rsv->space_info != global_rsv->space_info) | |
4169 | global_rsv = NULL; | |
8c2a3ca2 JB |
4170 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4171 | num_bytes); | |
6a63209f JB |
4172 | } |
4173 | ||
4174 | /* | |
8929ecfa YZ |
4175 | * helper to calculate size of global block reservation. |
4176 | * the desired value is sum of space used by extent tree, | |
4177 | * checksum tree and root tree | |
6a63209f | 4178 | */ |
8929ecfa | 4179 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4180 | { |
8929ecfa YZ |
4181 | struct btrfs_space_info *sinfo; |
4182 | u64 num_bytes; | |
4183 | u64 meta_used; | |
4184 | u64 data_used; | |
6c41761f | 4185 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4186 | |
8929ecfa YZ |
4187 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4188 | spin_lock(&sinfo->lock); | |
4189 | data_used = sinfo->bytes_used; | |
4190 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4191 | |
8929ecfa YZ |
4192 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4193 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4194 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4195 | data_used = 0; | |
8929ecfa YZ |
4196 | meta_used = sinfo->bytes_used; |
4197 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4198 | |
8929ecfa YZ |
4199 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4200 | csum_size * 2; | |
4201 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4202 | |
8929ecfa | 4203 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4204 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4205 | |
8929ecfa YZ |
4206 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4207 | } | |
6a63209f | 4208 | |
8929ecfa YZ |
4209 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4210 | { | |
4211 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4212 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4213 | u64 num_bytes; | |
6a63209f | 4214 | |
8929ecfa | 4215 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4216 | |
8929ecfa | 4217 | spin_lock(&sinfo->lock); |
1f699d38 | 4218 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4219 | |
8929ecfa | 4220 | block_rsv->size = num_bytes; |
4e06bdd6 | 4221 | |
8929ecfa | 4222 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4223 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4224 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4225 | |
4226 | if (sinfo->total_bytes > num_bytes) { | |
4227 | num_bytes = sinfo->total_bytes - num_bytes; | |
4228 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4229 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4230 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4231 | sinfo->flags, num_bytes, 1); |
6a63209f | 4232 | } |
6a63209f | 4233 | |
8929ecfa YZ |
4234 | if (block_rsv->reserved >= block_rsv->size) { |
4235 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4236 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4237 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4238 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4239 | sinfo->reservation_progress++; |
8929ecfa YZ |
4240 | block_rsv->reserved = block_rsv->size; |
4241 | block_rsv->full = 1; | |
4242 | } | |
182608c8 | 4243 | |
8929ecfa | 4244 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4245 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4246 | } |
4247 | ||
f0486c68 | 4248 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4249 | { |
f0486c68 | 4250 | struct btrfs_space_info *space_info; |
6a63209f | 4251 | |
f0486c68 YZ |
4252 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4253 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4254 | |
f0486c68 | 4255 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4256 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4257 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4258 | fs_info->trans_block_rsv.space_info = space_info; |
4259 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4260 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4261 | |
8929ecfa YZ |
4262 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4263 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4264 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4265 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4266 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4267 | |
8929ecfa | 4268 | update_global_block_rsv(fs_info); |
6a63209f JB |
4269 | } |
4270 | ||
8929ecfa | 4271 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4272 | { |
8c2a3ca2 JB |
4273 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4274 | (u64)-1); | |
8929ecfa YZ |
4275 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4276 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4277 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4278 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4279 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4280 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4281 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4282 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4283 | } |
4284 | ||
a22285a6 YZ |
4285 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4286 | struct btrfs_root *root) | |
6a63209f | 4287 | { |
a22285a6 YZ |
4288 | if (!trans->bytes_reserved) |
4289 | return; | |
6a63209f | 4290 | |
e77266e4 | 4291 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4292 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4293 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4294 | trans->bytes_reserved = 0; |
4295 | } | |
6a63209f | 4296 | |
79787eaa | 4297 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4298 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4299 | struct inode *inode) | |
4300 | { | |
4301 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4302 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4303 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4304 | ||
4305 | /* | |
fcb80c2a JB |
4306 | * We need to hold space in order to delete our orphan item once we've |
4307 | * added it, so this takes the reservation so we can release it later | |
4308 | * when we are truly done with the orphan item. | |
d68fc57b | 4309 | */ |
ff5714cc | 4310 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4311 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4312 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4313 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4314 | } |
4315 | ||
d68fc57b | 4316 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4317 | { |
d68fc57b | 4318 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4319 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4320 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4321 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4322 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4323 | } | |
97e728d4 | 4324 | |
a22285a6 YZ |
4325 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4326 | struct btrfs_pending_snapshot *pending) | |
4327 | { | |
4328 | struct btrfs_root *root = pending->root; | |
4329 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4330 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4331 | /* | |
4332 | * two for root back/forward refs, two for directory entries | |
4333 | * and one for root of the snapshot. | |
4334 | */ | |
16cdcec7 | 4335 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4336 | dst_rsv->space_info = src_rsv->space_info; |
4337 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4338 | } |
4339 | ||
7709cde3 JB |
4340 | /** |
4341 | * drop_outstanding_extent - drop an outstanding extent | |
4342 | * @inode: the inode we're dropping the extent for | |
4343 | * | |
4344 | * This is called when we are freeing up an outstanding extent, either called | |
4345 | * after an error or after an extent is written. This will return the number of | |
4346 | * reserved extents that need to be freed. This must be called with | |
4347 | * BTRFS_I(inode)->lock held. | |
4348 | */ | |
9e0baf60 JB |
4349 | static unsigned drop_outstanding_extent(struct inode *inode) |
4350 | { | |
7fd2ae21 | 4351 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4352 | unsigned dropped_extents = 0; |
4353 | ||
9e0baf60 JB |
4354 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4355 | BTRFS_I(inode)->outstanding_extents--; | |
4356 | ||
7fd2ae21 | 4357 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4358 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4359 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4360 | drop_inode_space = 1; |
7fd2ae21 | 4361 | |
9e0baf60 JB |
4362 | /* |
4363 | * If we have more or the same amount of outsanding extents than we have | |
4364 | * reserved then we need to leave the reserved extents count alone. | |
4365 | */ | |
4366 | if (BTRFS_I(inode)->outstanding_extents >= | |
4367 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4368 | return drop_inode_space; |
9e0baf60 JB |
4369 | |
4370 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4371 | BTRFS_I(inode)->outstanding_extents; | |
4372 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4373 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4374 | } |
4375 | ||
7709cde3 JB |
4376 | /** |
4377 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4378 | * reserved/free'd for the given bytes. | |
4379 | * @inode: the inode we're manipulating | |
4380 | * @num_bytes: the number of bytes in question | |
4381 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4382 | * | |
4383 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4384 | * correct amount of metadata that must either be reserved or freed. We | |
4385 | * calculate how many checksums we can fit into one leaf and then divide the | |
4386 | * number of bytes that will need to be checksumed by this value to figure out | |
4387 | * how many checksums will be required. If we are adding bytes then the number | |
4388 | * may go up and we will return the number of additional bytes that must be | |
4389 | * reserved. If it is going down we will return the number of bytes that must | |
4390 | * be freed. | |
4391 | * | |
4392 | * This must be called with BTRFS_I(inode)->lock held. | |
4393 | */ | |
4394 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4395 | int reserve) | |
6324fbf3 | 4396 | { |
7709cde3 JB |
4397 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4398 | u64 csum_size; | |
4399 | int num_csums_per_leaf; | |
4400 | int num_csums; | |
4401 | int old_csums; | |
4402 | ||
4403 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4404 | BTRFS_I(inode)->csum_bytes == 0) | |
4405 | return 0; | |
4406 | ||
4407 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4408 | if (reserve) | |
4409 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4410 | else | |
4411 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4412 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4413 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4414 | sizeof(struct btrfs_csum_item) + | |
4415 | sizeof(struct btrfs_disk_key)); | |
4416 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4417 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4418 | num_csums = num_csums / num_csums_per_leaf; | |
4419 | ||
4420 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4421 | old_csums = old_csums / num_csums_per_leaf; | |
4422 | ||
4423 | /* No change, no need to reserve more */ | |
4424 | if (old_csums == num_csums) | |
4425 | return 0; | |
4426 | ||
4427 | if (reserve) | |
4428 | return btrfs_calc_trans_metadata_size(root, | |
4429 | num_csums - old_csums); | |
4430 | ||
4431 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4432 | } |
c146afad | 4433 | |
0ca1f7ce YZ |
4434 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4435 | { | |
4436 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4437 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4438 | u64 to_reserve = 0; |
660d3f6c | 4439 | u64 csum_bytes; |
9e0baf60 | 4440 | unsigned nr_extents = 0; |
660d3f6c | 4441 | int extra_reserve = 0; |
c09544e0 | 4442 | int flush = 1; |
0ca1f7ce | 4443 | int ret; |
6324fbf3 | 4444 | |
660d3f6c | 4445 | /* Need to be holding the i_mutex here if we aren't free space cache */ |
c09544e0 JB |
4446 | if (btrfs_is_free_space_inode(root, inode)) |
4447 | flush = 0; | |
4448 | ||
4449 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4450 | schedule_timeout(1); |
ec44a35c | 4451 | |
f248679e | 4452 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4453 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4454 | |
9e0baf60 JB |
4455 | spin_lock(&BTRFS_I(inode)->lock); |
4456 | BTRFS_I(inode)->outstanding_extents++; | |
4457 | ||
4458 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4459 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4460 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4461 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4462 | |
7fd2ae21 JB |
4463 | /* |
4464 | * Add an item to reserve for updating the inode when we complete the | |
4465 | * delalloc io. | |
4466 | */ | |
72ac3c0d JB |
4467 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4468 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4469 | nr_extents++; |
660d3f6c | 4470 | extra_reserve = 1; |
593060d7 | 4471 | } |
7fd2ae21 JB |
4472 | |
4473 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4474 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4475 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4476 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4477 | |
36ba022a | 4478 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4479 | if (ret) { |
7ed49f18 | 4480 | u64 to_free = 0; |
9e0baf60 | 4481 | unsigned dropped; |
7ed49f18 | 4482 | |
7709cde3 | 4483 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4484 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4485 | /* |
660d3f6c JB |
4486 | * If the inodes csum_bytes is the same as the original |
4487 | * csum_bytes then we know we haven't raced with any free()ers | |
4488 | * so we can just reduce our inodes csum bytes and carry on. | |
4489 | * Otherwise we have to do the normal free thing to account for | |
4490 | * the case that the free side didn't free up its reserve | |
4491 | * because of this outstanding reservation. | |
9e0baf60 | 4492 | */ |
660d3f6c JB |
4493 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4494 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4495 | else | |
4496 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4497 | spin_unlock(&BTRFS_I(inode)->lock); | |
4498 | if (dropped) | |
4499 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4500 | ||
8c2a3ca2 | 4501 | if (to_free) { |
7ed49f18 | 4502 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4503 | trace_btrfs_space_reservation(root->fs_info, |
4504 | "delalloc", | |
4505 | btrfs_ino(inode), | |
4506 | to_free, 0); | |
4507 | } | |
f248679e | 4508 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4509 | return ret; |
9e0baf60 | 4510 | } |
25179201 | 4511 | |
660d3f6c JB |
4512 | spin_lock(&BTRFS_I(inode)->lock); |
4513 | if (extra_reserve) { | |
72ac3c0d JB |
4514 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4515 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4516 | nr_extents--; |
4517 | } | |
4518 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4519 | spin_unlock(&BTRFS_I(inode)->lock); | |
f248679e | 4520 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
660d3f6c | 4521 | |
8c2a3ca2 JB |
4522 | if (to_reserve) |
4523 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4524 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4525 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4526 | ||
0ca1f7ce YZ |
4527 | return 0; |
4528 | } | |
4529 | ||
7709cde3 JB |
4530 | /** |
4531 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4532 | * @inode: the inode to release the reservation for | |
4533 | * @num_bytes: the number of bytes we're releasing | |
4534 | * | |
4535 | * This will release the metadata reservation for an inode. This can be called | |
4536 | * once we complete IO for a given set of bytes to release their metadata | |
4537 | * reservations. | |
4538 | */ | |
0ca1f7ce YZ |
4539 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4540 | { | |
4541 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4542 | u64 to_free = 0; |
4543 | unsigned dropped; | |
0ca1f7ce YZ |
4544 | |
4545 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4546 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4547 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4548 | |
7709cde3 JB |
4549 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4550 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4551 | if (dropped > 0) |
4552 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4553 | |
8c2a3ca2 JB |
4554 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4555 | btrfs_ino(inode), to_free, 0); | |
0ca1f7ce YZ |
4556 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4557 | to_free); | |
4558 | } | |
4559 | ||
7709cde3 JB |
4560 | /** |
4561 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4562 | * @inode: inode we're writing to | |
4563 | * @num_bytes: the number of bytes we want to allocate | |
4564 | * | |
4565 | * This will do the following things | |
4566 | * | |
4567 | * o reserve space in the data space info for num_bytes | |
4568 | * o reserve space in the metadata space info based on number of outstanding | |
4569 | * extents and how much csums will be needed | |
4570 | * o add to the inodes ->delalloc_bytes | |
4571 | * o add it to the fs_info's delalloc inodes list. | |
4572 | * | |
4573 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4574 | */ | |
0ca1f7ce YZ |
4575 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4576 | { | |
4577 | int ret; | |
4578 | ||
4579 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4580 | if (ret) |
0ca1f7ce YZ |
4581 | return ret; |
4582 | ||
4583 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4584 | if (ret) { | |
4585 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4586 | return ret; | |
4587 | } | |
4588 | ||
4589 | return 0; | |
4590 | } | |
4591 | ||
7709cde3 JB |
4592 | /** |
4593 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4594 | * @inode: inode we're releasing space for | |
4595 | * @num_bytes: the number of bytes we want to free up | |
4596 | * | |
4597 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4598 | * called in the case that we don't need the metadata AND data reservations | |
4599 | * anymore. So if there is an error or we insert an inline extent. | |
4600 | * | |
4601 | * This function will release the metadata space that was not used and will | |
4602 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4603 | * list if there are no delalloc bytes left. | |
4604 | */ | |
0ca1f7ce YZ |
4605 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4606 | { | |
4607 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4608 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4609 | } |
4610 | ||
9078a3e1 CM |
4611 | static int update_block_group(struct btrfs_trans_handle *trans, |
4612 | struct btrfs_root *root, | |
f0486c68 | 4613 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4614 | { |
0af3d00b | 4615 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4616 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4617 | u64 total = num_bytes; |
9078a3e1 | 4618 | u64 old_val; |
db94535d | 4619 | u64 byte_in_group; |
0af3d00b | 4620 | int factor; |
3e1ad54f | 4621 | |
5d4f98a2 YZ |
4622 | /* block accounting for super block */ |
4623 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4624 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4625 | if (alloc) |
4626 | old_val += num_bytes; | |
4627 | else | |
4628 | old_val -= num_bytes; | |
6c41761f | 4629 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4630 | spin_unlock(&info->delalloc_lock); |
4631 | ||
d397712b | 4632 | while (total) { |
db94535d | 4633 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4634 | if (!cache) |
79787eaa | 4635 | return -ENOENT; |
b742bb82 YZ |
4636 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4637 | BTRFS_BLOCK_GROUP_RAID1 | | |
4638 | BTRFS_BLOCK_GROUP_RAID10)) | |
4639 | factor = 2; | |
4640 | else | |
4641 | factor = 1; | |
9d66e233 JB |
4642 | /* |
4643 | * If this block group has free space cache written out, we | |
4644 | * need to make sure to load it if we are removing space. This | |
4645 | * is because we need the unpinning stage to actually add the | |
4646 | * space back to the block group, otherwise we will leak space. | |
4647 | */ | |
4648 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4649 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4650 | |
db94535d CM |
4651 | byte_in_group = bytenr - cache->key.objectid; |
4652 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4653 | |
25179201 | 4654 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4655 | spin_lock(&cache->lock); |
0af3d00b | 4656 | |
73bc1876 | 4657 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4658 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4659 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4660 | ||
0f9dd46c | 4661 | cache->dirty = 1; |
9078a3e1 | 4662 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4663 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4664 | if (alloc) { |
db94535d | 4665 | old_val += num_bytes; |
11833d66 YZ |
4666 | btrfs_set_block_group_used(&cache->item, old_val); |
4667 | cache->reserved -= num_bytes; | |
11833d66 | 4668 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4669 | cache->space_info->bytes_used += num_bytes; |
4670 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4671 | spin_unlock(&cache->lock); |
25179201 | 4672 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4673 | } else { |
db94535d | 4674 | old_val -= num_bytes; |
c286ac48 | 4675 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4676 | cache->pinned += num_bytes; |
4677 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4678 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4679 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4680 | spin_unlock(&cache->lock); |
25179201 | 4681 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4682 | |
f0486c68 YZ |
4683 | set_extent_dirty(info->pinned_extents, |
4684 | bytenr, bytenr + num_bytes - 1, | |
4685 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4686 | } |
fa9c0d79 | 4687 | btrfs_put_block_group(cache); |
db94535d CM |
4688 | total -= num_bytes; |
4689 | bytenr += num_bytes; | |
9078a3e1 CM |
4690 | } |
4691 | return 0; | |
4692 | } | |
6324fbf3 | 4693 | |
a061fc8d CM |
4694 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4695 | { | |
0f9dd46c | 4696 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4697 | u64 bytenr; |
0f9dd46c JB |
4698 | |
4699 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4700 | if (!cache) | |
a061fc8d | 4701 | return 0; |
0f9dd46c | 4702 | |
d2fb3437 | 4703 | bytenr = cache->key.objectid; |
fa9c0d79 | 4704 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4705 | |
4706 | return bytenr; | |
a061fc8d CM |
4707 | } |
4708 | ||
f0486c68 YZ |
4709 | static int pin_down_extent(struct btrfs_root *root, |
4710 | struct btrfs_block_group_cache *cache, | |
4711 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4712 | { |
11833d66 YZ |
4713 | spin_lock(&cache->space_info->lock); |
4714 | spin_lock(&cache->lock); | |
4715 | cache->pinned += num_bytes; | |
4716 | cache->space_info->bytes_pinned += num_bytes; | |
4717 | if (reserved) { | |
4718 | cache->reserved -= num_bytes; | |
4719 | cache->space_info->bytes_reserved -= num_bytes; | |
4720 | } | |
4721 | spin_unlock(&cache->lock); | |
4722 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4723 | |
f0486c68 YZ |
4724 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4725 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4726 | return 0; | |
4727 | } | |
68b38550 | 4728 | |
f0486c68 YZ |
4729 | /* |
4730 | * this function must be called within transaction | |
4731 | */ | |
4732 | int btrfs_pin_extent(struct btrfs_root *root, | |
4733 | u64 bytenr, u64 num_bytes, int reserved) | |
4734 | { | |
4735 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4736 | |
f0486c68 | 4737 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 4738 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
4739 | |
4740 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4741 | ||
4742 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4743 | return 0; |
4744 | } | |
4745 | ||
f0486c68 | 4746 | /* |
e688b725 CM |
4747 | * this function must be called within transaction |
4748 | */ | |
4749 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4750 | struct btrfs_root *root, | |
4751 | u64 bytenr, u64 num_bytes) | |
4752 | { | |
4753 | struct btrfs_block_group_cache *cache; | |
4754 | ||
4755 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 4756 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
4757 | |
4758 | /* | |
4759 | * pull in the free space cache (if any) so that our pin | |
4760 | * removes the free space from the cache. We have load_only set | |
4761 | * to one because the slow code to read in the free extents does check | |
4762 | * the pinned extents. | |
4763 | */ | |
4764 | cache_block_group(cache, trans, root, 1); | |
4765 | ||
4766 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4767 | ||
4768 | /* remove us from the free space cache (if we're there at all) */ | |
4769 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4770 | btrfs_put_block_group(cache); | |
4771 | return 0; | |
4772 | } | |
4773 | ||
fb25e914 JB |
4774 | /** |
4775 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4776 | * @cache: The cache we are manipulating | |
4777 | * @num_bytes: The number of bytes in question | |
4778 | * @reserve: One of the reservation enums | |
4779 | * | |
4780 | * This is called by the allocator when it reserves space, or by somebody who is | |
4781 | * freeing space that was never actually used on disk. For example if you | |
4782 | * reserve some space for a new leaf in transaction A and before transaction A | |
4783 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4784 | * clear the reservation. | |
4785 | * | |
4786 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4787 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4788 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4789 | * allocating less disk space for the amount of data we have reserved in the | |
4790 | * case of compression. | |
4791 | * | |
4792 | * If this is a reservation and the block group has become read only we cannot | |
4793 | * make the reservation and return -EAGAIN, otherwise this function always | |
4794 | * succeeds. | |
f0486c68 | 4795 | */ |
fb25e914 JB |
4796 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4797 | u64 num_bytes, int reserve) | |
11833d66 | 4798 | { |
fb25e914 | 4799 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4800 | int ret = 0; |
79787eaa | 4801 | |
fb25e914 JB |
4802 | spin_lock(&space_info->lock); |
4803 | spin_lock(&cache->lock); | |
4804 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4805 | if (cache->ro) { |
4806 | ret = -EAGAIN; | |
4807 | } else { | |
fb25e914 JB |
4808 | cache->reserved += num_bytes; |
4809 | space_info->bytes_reserved += num_bytes; | |
4810 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4811 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
4812 | "space_info", space_info->flags, |
4813 | num_bytes, 0); | |
fb25e914 JB |
4814 | space_info->bytes_may_use -= num_bytes; |
4815 | } | |
f0486c68 | 4816 | } |
fb25e914 JB |
4817 | } else { |
4818 | if (cache->ro) | |
4819 | space_info->bytes_readonly += num_bytes; | |
4820 | cache->reserved -= num_bytes; | |
4821 | space_info->bytes_reserved -= num_bytes; | |
4822 | space_info->reservation_progress++; | |
324ae4df | 4823 | } |
fb25e914 JB |
4824 | spin_unlock(&cache->lock); |
4825 | spin_unlock(&space_info->lock); | |
f0486c68 | 4826 | return ret; |
324ae4df | 4827 | } |
9078a3e1 | 4828 | |
143bede5 | 4829 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 4830 | struct btrfs_root *root) |
e8569813 | 4831 | { |
e8569813 | 4832 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4833 | struct btrfs_caching_control *next; |
4834 | struct btrfs_caching_control *caching_ctl; | |
4835 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4836 | |
11833d66 | 4837 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4838 | |
11833d66 YZ |
4839 | list_for_each_entry_safe(caching_ctl, next, |
4840 | &fs_info->caching_block_groups, list) { | |
4841 | cache = caching_ctl->block_group; | |
4842 | if (block_group_cache_done(cache)) { | |
4843 | cache->last_byte_to_unpin = (u64)-1; | |
4844 | list_del_init(&caching_ctl->list); | |
4845 | put_caching_control(caching_ctl); | |
e8569813 | 4846 | } else { |
11833d66 | 4847 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4848 | } |
e8569813 | 4849 | } |
11833d66 YZ |
4850 | |
4851 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4852 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4853 | else | |
4854 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4855 | ||
4856 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4857 | |
4858 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4859 | } |
4860 | ||
11833d66 | 4861 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4862 | { |
11833d66 YZ |
4863 | struct btrfs_fs_info *fs_info = root->fs_info; |
4864 | struct btrfs_block_group_cache *cache = NULL; | |
4865 | u64 len; | |
ccd467d6 | 4866 | |
11833d66 YZ |
4867 | while (start <= end) { |
4868 | if (!cache || | |
4869 | start >= cache->key.objectid + cache->key.offset) { | |
4870 | if (cache) | |
4871 | btrfs_put_block_group(cache); | |
4872 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 4873 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
4874 | } |
4875 | ||
4876 | len = cache->key.objectid + cache->key.offset - start; | |
4877 | len = min(len, end + 1 - start); | |
4878 | ||
4879 | if (start < cache->last_byte_to_unpin) { | |
4880 | len = min(len, cache->last_byte_to_unpin - start); | |
4881 | btrfs_add_free_space(cache, start, len); | |
4882 | } | |
4883 | ||
f0486c68 YZ |
4884 | start += len; |
4885 | ||
11833d66 YZ |
4886 | spin_lock(&cache->space_info->lock); |
4887 | spin_lock(&cache->lock); | |
4888 | cache->pinned -= len; | |
4889 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4890 | if (cache->ro) |
f0486c68 | 4891 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4892 | spin_unlock(&cache->lock); |
4893 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4894 | } |
11833d66 YZ |
4895 | |
4896 | if (cache) | |
4897 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4898 | return 0; |
4899 | } | |
4900 | ||
4901 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4902 | struct btrfs_root *root) |
a28ec197 | 4903 | { |
11833d66 YZ |
4904 | struct btrfs_fs_info *fs_info = root->fs_info; |
4905 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
4906 | u64 start; |
4907 | u64 end; | |
a28ec197 | 4908 | int ret; |
a28ec197 | 4909 | |
79787eaa JM |
4910 | if (trans->aborted) |
4911 | return 0; | |
4912 | ||
11833d66 YZ |
4913 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4914 | unpin = &fs_info->freed_extents[1]; | |
4915 | else | |
4916 | unpin = &fs_info->freed_extents[0]; | |
4917 | ||
d397712b | 4918 | while (1) { |
1a5bc167 CM |
4919 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4920 | EXTENT_DIRTY); | |
4921 | if (ret) | |
a28ec197 | 4922 | break; |
1f3c79a2 | 4923 | |
5378e607 LD |
4924 | if (btrfs_test_opt(root, DISCARD)) |
4925 | ret = btrfs_discard_extent(root, start, | |
4926 | end + 1 - start, NULL); | |
1f3c79a2 | 4927 | |
1a5bc167 | 4928 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4929 | unpin_extent_range(root, start, end); |
b9473439 | 4930 | cond_resched(); |
a28ec197 | 4931 | } |
817d52f8 | 4932 | |
e20d96d6 CM |
4933 | return 0; |
4934 | } | |
4935 | ||
5d4f98a2 YZ |
4936 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4937 | struct btrfs_root *root, | |
4938 | u64 bytenr, u64 num_bytes, u64 parent, | |
4939 | u64 root_objectid, u64 owner_objectid, | |
4940 | u64 owner_offset, int refs_to_drop, | |
4941 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4942 | { |
e2fa7227 | 4943 | struct btrfs_key key; |
5d4f98a2 | 4944 | struct btrfs_path *path; |
1261ec42 CM |
4945 | struct btrfs_fs_info *info = root->fs_info; |
4946 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4947 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4948 | struct btrfs_extent_item *ei; |
4949 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4950 | int ret; |
5d4f98a2 | 4951 | int is_data; |
952fccac CM |
4952 | int extent_slot = 0; |
4953 | int found_extent = 0; | |
4954 | int num_to_del = 1; | |
5d4f98a2 YZ |
4955 | u32 item_size; |
4956 | u64 refs; | |
037e6390 | 4957 | |
5caf2a00 | 4958 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4959 | if (!path) |
4960 | return -ENOMEM; | |
5f26f772 | 4961 | |
3c12ac72 | 4962 | path->reada = 1; |
b9473439 | 4963 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4964 | |
4965 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4966 | BUG_ON(!is_data && refs_to_drop != 1); | |
4967 | ||
4968 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4969 | bytenr, num_bytes, parent, | |
4970 | root_objectid, owner_objectid, | |
4971 | owner_offset); | |
7bb86316 | 4972 | if (ret == 0) { |
952fccac | 4973 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4974 | while (extent_slot >= 0) { |
4975 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4976 | extent_slot); |
5d4f98a2 | 4977 | if (key.objectid != bytenr) |
952fccac | 4978 | break; |
5d4f98a2 YZ |
4979 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4980 | key.offset == num_bytes) { | |
952fccac CM |
4981 | found_extent = 1; |
4982 | break; | |
4983 | } | |
4984 | if (path->slots[0] - extent_slot > 5) | |
4985 | break; | |
5d4f98a2 | 4986 | extent_slot--; |
952fccac | 4987 | } |
5d4f98a2 YZ |
4988 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4989 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4990 | if (found_extent && item_size < sizeof(*ei)) | |
4991 | found_extent = 0; | |
4992 | #endif | |
31840ae1 | 4993 | if (!found_extent) { |
5d4f98a2 | 4994 | BUG_ON(iref); |
56bec294 | 4995 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4996 | NULL, refs_to_drop, |
4997 | is_data); | |
79787eaa JM |
4998 | if (ret) |
4999 | goto abort; | |
b3b4aa74 | 5000 | btrfs_release_path(path); |
b9473439 | 5001 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5002 | |
5003 | key.objectid = bytenr; | |
5004 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5005 | key.offset = num_bytes; | |
5006 | ||
31840ae1 ZY |
5007 | ret = btrfs_search_slot(trans, extent_root, |
5008 | &key, path, -1, 1); | |
f3465ca4 JB |
5009 | if (ret) { |
5010 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5011 | ", was looking for %llu\n", ret, |
5012 | (unsigned long long)bytenr); | |
b783e62d JB |
5013 | if (ret > 0) |
5014 | btrfs_print_leaf(extent_root, | |
5015 | path->nodes[0]); | |
f3465ca4 | 5016 | } |
79787eaa JM |
5017 | if (ret < 0) |
5018 | goto abort; | |
31840ae1 ZY |
5019 | extent_slot = path->slots[0]; |
5020 | } | |
79787eaa | 5021 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5022 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5023 | WARN_ON(1); | |
d397712b | 5024 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5025 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5026 | (unsigned long long)bytenr, |
56bec294 | 5027 | (unsigned long long)parent, |
d397712b | 5028 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5029 | (unsigned long long)owner_objectid, |
5030 | (unsigned long long)owner_offset); | |
79787eaa JM |
5031 | } else { |
5032 | goto abort; | |
7bb86316 | 5033 | } |
5f39d397 CM |
5034 | |
5035 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5036 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5037 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5038 | if (item_size < sizeof(*ei)) { | |
5039 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5040 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5041 | owner_objectid, 0); | |
79787eaa JM |
5042 | if (ret < 0) |
5043 | goto abort; | |
5d4f98a2 | 5044 | |
b3b4aa74 | 5045 | btrfs_release_path(path); |
5d4f98a2 YZ |
5046 | path->leave_spinning = 1; |
5047 | ||
5048 | key.objectid = bytenr; | |
5049 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5050 | key.offset = num_bytes; | |
5051 | ||
5052 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5053 | -1, 1); | |
5054 | if (ret) { | |
5055 | printk(KERN_ERR "umm, got %d back from search" | |
5056 | ", was looking for %llu\n", ret, | |
5057 | (unsigned long long)bytenr); | |
5058 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5059 | } | |
79787eaa JM |
5060 | if (ret < 0) |
5061 | goto abort; | |
5d4f98a2 YZ |
5062 | extent_slot = path->slots[0]; |
5063 | leaf = path->nodes[0]; | |
5064 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5065 | } | |
5066 | #endif | |
5067 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5068 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5069 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5070 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
5071 | struct btrfs_tree_block_info *bi; | |
5072 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5073 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5074 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5075 | } | |
56bec294 | 5076 | |
5d4f98a2 | 5077 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5078 | BUG_ON(refs < refs_to_drop); |
5079 | refs -= refs_to_drop; | |
5f39d397 | 5080 | |
5d4f98a2 YZ |
5081 | if (refs > 0) { |
5082 | if (extent_op) | |
5083 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5084 | /* | |
5085 | * In the case of inline back ref, reference count will | |
5086 | * be updated by remove_extent_backref | |
952fccac | 5087 | */ |
5d4f98a2 YZ |
5088 | if (iref) { |
5089 | BUG_ON(!found_extent); | |
5090 | } else { | |
5091 | btrfs_set_extent_refs(leaf, ei, refs); | |
5092 | btrfs_mark_buffer_dirty(leaf); | |
5093 | } | |
5094 | if (found_extent) { | |
5095 | ret = remove_extent_backref(trans, extent_root, path, | |
5096 | iref, refs_to_drop, | |
5097 | is_data); | |
79787eaa JM |
5098 | if (ret) |
5099 | goto abort; | |
952fccac | 5100 | } |
5d4f98a2 | 5101 | } else { |
5d4f98a2 YZ |
5102 | if (found_extent) { |
5103 | BUG_ON(is_data && refs_to_drop != | |
5104 | extent_data_ref_count(root, path, iref)); | |
5105 | if (iref) { | |
5106 | BUG_ON(path->slots[0] != extent_slot); | |
5107 | } else { | |
5108 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5109 | path->slots[0] = extent_slot; | |
5110 | num_to_del = 2; | |
5111 | } | |
78fae27e | 5112 | } |
b9473439 | 5113 | |
952fccac CM |
5114 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5115 | num_to_del); | |
79787eaa JM |
5116 | if (ret) |
5117 | goto abort; | |
b3b4aa74 | 5118 | btrfs_release_path(path); |
21af804c | 5119 | |
5d4f98a2 | 5120 | if (is_data) { |
459931ec | 5121 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
79787eaa JM |
5122 | if (ret) |
5123 | goto abort; | |
459931ec CM |
5124 | } |
5125 | ||
f0486c68 | 5126 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
79787eaa JM |
5127 | if (ret) |
5128 | goto abort; | |
a28ec197 | 5129 | } |
79787eaa | 5130 | out: |
5caf2a00 | 5131 | btrfs_free_path(path); |
a28ec197 | 5132 | return ret; |
79787eaa JM |
5133 | |
5134 | abort: | |
5135 | btrfs_abort_transaction(trans, extent_root, ret); | |
5136 | goto out; | |
a28ec197 CM |
5137 | } |
5138 | ||
1887be66 | 5139 | /* |
f0486c68 | 5140 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5141 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5142 | * a given extent, and if there are no other delayed refs to be processed, it | |
5143 | * removes it from the tree. | |
5144 | */ | |
5145 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5146 | struct btrfs_root *root, u64 bytenr) | |
5147 | { | |
5148 | struct btrfs_delayed_ref_head *head; | |
5149 | struct btrfs_delayed_ref_root *delayed_refs; | |
5150 | struct btrfs_delayed_ref_node *ref; | |
5151 | struct rb_node *node; | |
f0486c68 | 5152 | int ret = 0; |
1887be66 CM |
5153 | |
5154 | delayed_refs = &trans->transaction->delayed_refs; | |
5155 | spin_lock(&delayed_refs->lock); | |
5156 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5157 | if (!head) | |
5158 | goto out; | |
5159 | ||
5160 | node = rb_prev(&head->node.rb_node); | |
5161 | if (!node) | |
5162 | goto out; | |
5163 | ||
5164 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5165 | ||
5166 | /* there are still entries for this ref, we can't drop it */ | |
5167 | if (ref->bytenr == bytenr) | |
5168 | goto out; | |
5169 | ||
5d4f98a2 YZ |
5170 | if (head->extent_op) { |
5171 | if (!head->must_insert_reserved) | |
5172 | goto out; | |
5173 | kfree(head->extent_op); | |
5174 | head->extent_op = NULL; | |
5175 | } | |
5176 | ||
1887be66 CM |
5177 | /* |
5178 | * waiting for the lock here would deadlock. If someone else has it | |
5179 | * locked they are already in the process of dropping it anyway | |
5180 | */ | |
5181 | if (!mutex_trylock(&head->mutex)) | |
5182 | goto out; | |
5183 | ||
5184 | /* | |
5185 | * at this point we have a head with no other entries. Go | |
5186 | * ahead and process it. | |
5187 | */ | |
5188 | head->node.in_tree = 0; | |
5189 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5190 | |
1887be66 | 5191 | delayed_refs->num_entries--; |
a168650c JS |
5192 | if (waitqueue_active(&delayed_refs->seq_wait)) |
5193 | wake_up(&delayed_refs->seq_wait); | |
1887be66 CM |
5194 | |
5195 | /* | |
5196 | * we don't take a ref on the node because we're removing it from the | |
5197 | * tree, so we just steal the ref the tree was holding. | |
5198 | */ | |
c3e69d58 CM |
5199 | delayed_refs->num_heads--; |
5200 | if (list_empty(&head->cluster)) | |
5201 | delayed_refs->num_heads_ready--; | |
5202 | ||
5203 | list_del_init(&head->cluster); | |
1887be66 CM |
5204 | spin_unlock(&delayed_refs->lock); |
5205 | ||
f0486c68 YZ |
5206 | BUG_ON(head->extent_op); |
5207 | if (head->must_insert_reserved) | |
5208 | ret = 1; | |
5209 | ||
5210 | mutex_unlock(&head->mutex); | |
1887be66 | 5211 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5212 | return ret; |
1887be66 CM |
5213 | out: |
5214 | spin_unlock(&delayed_refs->lock); | |
5215 | return 0; | |
5216 | } | |
5217 | ||
f0486c68 YZ |
5218 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5219 | struct btrfs_root *root, | |
5220 | struct extent_buffer *buf, | |
5581a51a | 5221 | u64 parent, int last_ref) |
f0486c68 | 5222 | { |
f0486c68 YZ |
5223 | struct btrfs_block_group_cache *cache = NULL; |
5224 | int ret; | |
5225 | ||
5226 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5227 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5228 | buf->start, buf->len, | |
5229 | parent, root->root_key.objectid, | |
5230 | btrfs_header_level(buf), | |
5581a51a | 5231 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5232 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5233 | } |
5234 | ||
5235 | if (!last_ref) | |
5236 | return; | |
5237 | ||
f0486c68 | 5238 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5239 | |
5240 | if (btrfs_header_generation(buf) == trans->transid) { | |
5241 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5242 | ret = check_ref_cleanup(trans, root, buf->start); | |
5243 | if (!ret) | |
37be25bc | 5244 | goto out; |
f0486c68 YZ |
5245 | } |
5246 | ||
5247 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5248 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5249 | goto out; |
f0486c68 YZ |
5250 | } |
5251 | ||
5252 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5253 | ||
5254 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5255 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5256 | } |
5257 | out: | |
a826d6dc JB |
5258 | /* |
5259 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5260 | * anymore. | |
5261 | */ | |
5262 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5263 | btrfs_put_block_group(cache); |
5264 | } | |
5265 | ||
79787eaa | 5266 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5267 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5268 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5269 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5270 | { |
5271 | int ret; | |
66d7e7f0 | 5272 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5273 | |
56bec294 CM |
5274 | /* |
5275 | * tree log blocks never actually go into the extent allocation | |
5276 | * tree, just update pinning info and exit early. | |
56bec294 | 5277 | */ |
5d4f98a2 YZ |
5278 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5279 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5280 | /* unlocks the pinned mutex */ |
11833d66 | 5281 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5282 | ret = 0; |
5d4f98a2 | 5283 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5284 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5285 | num_bytes, | |
5d4f98a2 | 5286 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5287 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5288 | } else { |
66d7e7f0 AJ |
5289 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5290 | num_bytes, | |
5291 | parent, root_objectid, owner, | |
5292 | offset, BTRFS_DROP_DELAYED_REF, | |
5293 | NULL, for_cow); | |
56bec294 | 5294 | } |
925baedd CM |
5295 | return ret; |
5296 | } | |
5297 | ||
87ee04eb CM |
5298 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5299 | { | |
5300 | u64 mask = ((u64)root->stripesize - 1); | |
5301 | u64 ret = (val + mask) & ~mask; | |
5302 | return ret; | |
5303 | } | |
5304 | ||
817d52f8 JB |
5305 | /* |
5306 | * when we wait for progress in the block group caching, its because | |
5307 | * our allocation attempt failed at least once. So, we must sleep | |
5308 | * and let some progress happen before we try again. | |
5309 | * | |
5310 | * This function will sleep at least once waiting for new free space to | |
5311 | * show up, and then it will check the block group free space numbers | |
5312 | * for our min num_bytes. Another option is to have it go ahead | |
5313 | * and look in the rbtree for a free extent of a given size, but this | |
5314 | * is a good start. | |
5315 | */ | |
5316 | static noinline int | |
5317 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5318 | u64 num_bytes) | |
5319 | { | |
11833d66 | 5320 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5321 | DEFINE_WAIT(wait); |
5322 | ||
11833d66 YZ |
5323 | caching_ctl = get_caching_control(cache); |
5324 | if (!caching_ctl) | |
817d52f8 | 5325 | return 0; |
817d52f8 | 5326 | |
11833d66 | 5327 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5328 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5329 | |
5330 | put_caching_control(caching_ctl); | |
5331 | return 0; | |
5332 | } | |
5333 | ||
5334 | static noinline int | |
5335 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5336 | { | |
5337 | struct btrfs_caching_control *caching_ctl; | |
5338 | DEFINE_WAIT(wait); | |
5339 | ||
5340 | caching_ctl = get_caching_control(cache); | |
5341 | if (!caching_ctl) | |
5342 | return 0; | |
5343 | ||
5344 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5345 | ||
5346 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5347 | return 0; |
5348 | } | |
5349 | ||
7738a53a | 5350 | static int __get_block_group_index(u64 flags) |
b742bb82 YZ |
5351 | { |
5352 | int index; | |
7738a53a ID |
5353 | |
5354 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
b742bb82 | 5355 | index = 0; |
7738a53a | 5356 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
b742bb82 | 5357 | index = 1; |
7738a53a | 5358 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
b742bb82 | 5359 | index = 2; |
7738a53a | 5360 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
b742bb82 YZ |
5361 | index = 3; |
5362 | else | |
5363 | index = 4; | |
7738a53a | 5364 | |
b742bb82 YZ |
5365 | return index; |
5366 | } | |
5367 | ||
7738a53a ID |
5368 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5369 | { | |
5370 | return __get_block_group_index(cache->flags); | |
5371 | } | |
5372 | ||
817d52f8 | 5373 | enum btrfs_loop_type { |
285ff5af JB |
5374 | LOOP_CACHING_NOWAIT = 0, |
5375 | LOOP_CACHING_WAIT = 1, | |
5376 | LOOP_ALLOC_CHUNK = 2, | |
5377 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5378 | }; |
5379 | ||
fec577fb CM |
5380 | /* |
5381 | * walks the btree of allocated extents and find a hole of a given size. | |
5382 | * The key ins is changed to record the hole: | |
5383 | * ins->objectid == block start | |
62e2749e | 5384 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5385 | * ins->offset == number of blocks |
5386 | * Any available blocks before search_start are skipped. | |
5387 | */ | |
d397712b | 5388 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5389 | struct btrfs_root *orig_root, |
5390 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5391 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5392 | u64 data) |
fec577fb | 5393 | { |
80eb234a | 5394 | int ret = 0; |
d397712b | 5395 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5396 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5397 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5398 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5399 | u64 search_start = 0; |
239b14b3 | 5400 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5401 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5402 | int done_chunk_alloc = 0; |
80eb234a | 5403 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5404 | int loop = 0; |
f0486c68 | 5405 | int index = 0; |
fb25e914 JB |
5406 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5407 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5408 | bool found_uncached_bg = false; |
0a24325e | 5409 | bool failed_cluster_refill = false; |
1cdda9b8 | 5410 | bool failed_alloc = false; |
67377734 | 5411 | bool use_cluster = true; |
60d2adbb | 5412 | bool have_caching_bg = false; |
fec577fb | 5413 | |
db94535d | 5414 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5415 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5416 | ins->objectid = 0; |
5417 | ins->offset = 0; | |
b1a4d965 | 5418 | |
3f7de037 JB |
5419 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5420 | ||
2552d17e | 5421 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5422 | if (!space_info) { |
e0f54067 | 5423 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5424 | return -ENOSPC; |
5425 | } | |
2552d17e | 5426 | |
67377734 JB |
5427 | /* |
5428 | * If the space info is for both data and metadata it means we have a | |
5429 | * small filesystem and we can't use the clustering stuff. | |
5430 | */ | |
5431 | if (btrfs_mixed_space_info(space_info)) | |
5432 | use_cluster = false; | |
5433 | ||
0ef3e66b CM |
5434 | if (orig_root->ref_cows || empty_size) |
5435 | allowed_chunk_alloc = 1; | |
5436 | ||
67377734 | 5437 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5438 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5439 | if (!btrfs_test_opt(root, SSD)) |
5440 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5441 | } |
5442 | ||
67377734 JB |
5443 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5444 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5445 | last_ptr = &root->fs_info->data_alloc_cluster; |
5446 | } | |
0f9dd46c | 5447 | |
239b14b3 | 5448 | if (last_ptr) { |
fa9c0d79 CM |
5449 | spin_lock(&last_ptr->lock); |
5450 | if (last_ptr->block_group) | |
5451 | hint_byte = last_ptr->window_start; | |
5452 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5453 | } |
fa9c0d79 | 5454 | |
a061fc8d | 5455 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5456 | search_start = max(search_start, hint_byte); |
0b86a832 | 5457 | |
817d52f8 | 5458 | if (!last_ptr) |
fa9c0d79 | 5459 | empty_cluster = 0; |
fa9c0d79 | 5460 | |
2552d17e | 5461 | if (search_start == hint_byte) { |
2552d17e JB |
5462 | block_group = btrfs_lookup_block_group(root->fs_info, |
5463 | search_start); | |
274bd4fb | 5464 | used_block_group = block_group; |
817d52f8 JB |
5465 | /* |
5466 | * we don't want to use the block group if it doesn't match our | |
5467 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5468 | * |
5469 | * However if we are re-searching with an ideal block group | |
5470 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5471 | */ |
5472 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5473 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5474 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5475 | if (list_empty(&block_group->list) || |
5476 | block_group->ro) { | |
5477 | /* | |
5478 | * someone is removing this block group, | |
5479 | * we can't jump into the have_block_group | |
5480 | * target because our list pointers are not | |
5481 | * valid | |
5482 | */ | |
5483 | btrfs_put_block_group(block_group); | |
5484 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5485 | } else { |
b742bb82 | 5486 | index = get_block_group_index(block_group); |
44fb5511 | 5487 | goto have_block_group; |
ccf0e725 | 5488 | } |
2552d17e | 5489 | } else if (block_group) { |
fa9c0d79 | 5490 | btrfs_put_block_group(block_group); |
2552d17e | 5491 | } |
42e70e7a | 5492 | } |
2552d17e | 5493 | search: |
60d2adbb | 5494 | have_caching_bg = false; |
80eb234a | 5495 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5496 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5497 | list) { | |
6226cb0a | 5498 | u64 offset; |
817d52f8 | 5499 | int cached; |
8a1413a2 | 5500 | |
274bd4fb | 5501 | used_block_group = block_group; |
11dfe35a | 5502 | btrfs_get_block_group(block_group); |
2552d17e | 5503 | search_start = block_group->key.objectid; |
42e70e7a | 5504 | |
83a50de9 CM |
5505 | /* |
5506 | * this can happen if we end up cycling through all the | |
5507 | * raid types, but we want to make sure we only allocate | |
5508 | * for the proper type. | |
5509 | */ | |
5510 | if (!block_group_bits(block_group, data)) { | |
5511 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5512 | BTRFS_BLOCK_GROUP_RAID1 | | |
5513 | BTRFS_BLOCK_GROUP_RAID10; | |
5514 | ||
5515 | /* | |
5516 | * if they asked for extra copies and this block group | |
5517 | * doesn't provide them, bail. This does allow us to | |
5518 | * fill raid0 from raid1. | |
5519 | */ | |
5520 | if ((data & extra) && !(block_group->flags & extra)) | |
5521 | goto loop; | |
5522 | } | |
5523 | ||
2552d17e | 5524 | have_block_group: |
291c7d2f JB |
5525 | cached = block_group_cache_done(block_group); |
5526 | if (unlikely(!cached)) { | |
291c7d2f | 5527 | found_uncached_bg = true; |
b8399dee | 5528 | ret = cache_block_group(block_group, trans, |
285ff5af | 5529 | orig_root, 0); |
1d4284bd CM |
5530 | BUG_ON(ret < 0); |
5531 | ret = 0; | |
817d52f8 JB |
5532 | } |
5533 | ||
ea6a478e | 5534 | if (unlikely(block_group->ro)) |
2552d17e | 5535 | goto loop; |
0f9dd46c | 5536 | |
0a24325e | 5537 | /* |
062c05c4 AO |
5538 | * Ok we want to try and use the cluster allocator, so |
5539 | * lets look there | |
0a24325e | 5540 | */ |
062c05c4 | 5541 | if (last_ptr) { |
fa9c0d79 CM |
5542 | /* |
5543 | * the refill lock keeps out other | |
5544 | * people trying to start a new cluster | |
5545 | */ | |
5546 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5547 | used_block_group = last_ptr->block_group; |
5548 | if (used_block_group != block_group && | |
5549 | (!used_block_group || | |
5550 | used_block_group->ro || | |
5551 | !block_group_bits(used_block_group, data))) { | |
5552 | used_block_group = block_group; | |
44fb5511 | 5553 | goto refill_cluster; |
274bd4fb AO |
5554 | } |
5555 | ||
5556 | if (used_block_group != block_group) | |
5557 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5558 | |
274bd4fb AO |
5559 | offset = btrfs_alloc_from_cluster(used_block_group, |
5560 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5561 | if (offset) { |
5562 | /* we have a block, we're done */ | |
5563 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5564 | trace_btrfs_reserve_extent_cluster(root, |
5565 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5566 | goto checks; |
5567 | } | |
5568 | ||
274bd4fb AO |
5569 | WARN_ON(last_ptr->block_group != used_block_group); |
5570 | if (used_block_group != block_group) { | |
5571 | btrfs_put_block_group(used_block_group); | |
5572 | used_block_group = block_group; | |
fa9c0d79 | 5573 | } |
44fb5511 | 5574 | refill_cluster: |
274bd4fb | 5575 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5576 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5577 | * set up a new clusters, so lets just skip it | |
5578 | * and let the allocator find whatever block | |
5579 | * it can find. If we reach this point, we | |
5580 | * will have tried the cluster allocator | |
5581 | * plenty of times and not have found | |
5582 | * anything, so we are likely way too | |
5583 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5584 | * anything. |
5585 | * | |
5586 | * However, if the cluster is taken from the | |
5587 | * current block group, release the cluster | |
5588 | * first, so that we stand a better chance of | |
5589 | * succeeding in the unclustered | |
5590 | * allocation. */ | |
5591 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5592 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5593 | spin_unlock(&last_ptr->refill_lock); |
5594 | goto unclustered_alloc; | |
5595 | } | |
5596 | ||
fa9c0d79 CM |
5597 | /* |
5598 | * this cluster didn't work out, free it and | |
5599 | * start over | |
5600 | */ | |
5601 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5602 | ||
a5f6f719 AO |
5603 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5604 | spin_unlock(&last_ptr->refill_lock); | |
5605 | goto unclustered_alloc; | |
5606 | } | |
5607 | ||
fa9c0d79 | 5608 | /* allocate a cluster in this block group */ |
451d7585 | 5609 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5610 | block_group, last_ptr, |
1b22bad7 | 5611 | search_start, num_bytes, |
fa9c0d79 CM |
5612 | empty_cluster + empty_size); |
5613 | if (ret == 0) { | |
5614 | /* | |
5615 | * now pull our allocation out of this | |
5616 | * cluster | |
5617 | */ | |
5618 | offset = btrfs_alloc_from_cluster(block_group, | |
5619 | last_ptr, num_bytes, | |
5620 | search_start); | |
5621 | if (offset) { | |
5622 | /* we found one, proceed */ | |
5623 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5624 | trace_btrfs_reserve_extent_cluster(root, |
5625 | block_group, search_start, | |
5626 | num_bytes); | |
fa9c0d79 CM |
5627 | goto checks; |
5628 | } | |
0a24325e JB |
5629 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5630 | && !failed_cluster_refill) { | |
817d52f8 JB |
5631 | spin_unlock(&last_ptr->refill_lock); |
5632 | ||
0a24325e | 5633 | failed_cluster_refill = true; |
817d52f8 JB |
5634 | wait_block_group_cache_progress(block_group, |
5635 | num_bytes + empty_cluster + empty_size); | |
5636 | goto have_block_group; | |
fa9c0d79 | 5637 | } |
817d52f8 | 5638 | |
fa9c0d79 CM |
5639 | /* |
5640 | * at this point we either didn't find a cluster | |
5641 | * or we weren't able to allocate a block from our | |
5642 | * cluster. Free the cluster we've been trying | |
5643 | * to use, and go to the next block group | |
5644 | */ | |
0a24325e | 5645 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5646 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5647 | goto loop; |
fa9c0d79 CM |
5648 | } |
5649 | ||
062c05c4 | 5650 | unclustered_alloc: |
a5f6f719 AO |
5651 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5652 | if (cached && | |
5653 | block_group->free_space_ctl->free_space < | |
5654 | num_bytes + empty_cluster + empty_size) { | |
5655 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5656 | goto loop; | |
5657 | } | |
5658 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5659 | ||
6226cb0a JB |
5660 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5661 | num_bytes, empty_size); | |
1cdda9b8 JB |
5662 | /* |
5663 | * If we didn't find a chunk, and we haven't failed on this | |
5664 | * block group before, and this block group is in the middle of | |
5665 | * caching and we are ok with waiting, then go ahead and wait | |
5666 | * for progress to be made, and set failed_alloc to true. | |
5667 | * | |
5668 | * If failed_alloc is true then we've already waited on this | |
5669 | * block group once and should move on to the next block group. | |
5670 | */ | |
5671 | if (!offset && !failed_alloc && !cached && | |
5672 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5673 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5674 | num_bytes + empty_size); |
5675 | failed_alloc = true; | |
817d52f8 | 5676 | goto have_block_group; |
1cdda9b8 | 5677 | } else if (!offset) { |
60d2adbb MX |
5678 | if (!cached) |
5679 | have_caching_bg = true; | |
1cdda9b8 | 5680 | goto loop; |
817d52f8 | 5681 | } |
fa9c0d79 | 5682 | checks: |
6226cb0a | 5683 | search_start = stripe_align(root, offset); |
25179201 | 5684 | |
2552d17e JB |
5685 | /* move on to the next group */ |
5686 | if (search_start + num_bytes > | |
274bd4fb AO |
5687 | used_block_group->key.objectid + used_block_group->key.offset) { |
5688 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5689 | goto loop; |
6226cb0a | 5690 | } |
f5a31e16 | 5691 | |
f0486c68 | 5692 | if (offset < search_start) |
274bd4fb | 5693 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5694 | search_start - offset); |
5695 | BUG_ON(offset > search_start); | |
2552d17e | 5696 | |
274bd4fb | 5697 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5698 | alloc_type); |
f0486c68 | 5699 | if (ret == -EAGAIN) { |
274bd4fb | 5700 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5701 | goto loop; |
0f9dd46c | 5702 | } |
0b86a832 | 5703 | |
f0486c68 | 5704 | /* we are all good, lets return */ |
2552d17e JB |
5705 | ins->objectid = search_start; |
5706 | ins->offset = num_bytes; | |
d2fb3437 | 5707 | |
3f7de037 JB |
5708 | trace_btrfs_reserve_extent(orig_root, block_group, |
5709 | search_start, num_bytes); | |
6226cb0a | 5710 | if (offset < search_start) |
274bd4fb | 5711 | btrfs_add_free_space(used_block_group, offset, |
6226cb0a JB |
5712 | search_start - offset); |
5713 | BUG_ON(offset > search_start); | |
274bd4fb AO |
5714 | if (used_block_group != block_group) |
5715 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5716 | btrfs_put_block_group(block_group); |
2552d17e JB |
5717 | break; |
5718 | loop: | |
0a24325e | 5719 | failed_cluster_refill = false; |
1cdda9b8 | 5720 | failed_alloc = false; |
b742bb82 | 5721 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5722 | if (used_block_group != block_group) |
5723 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5724 | btrfs_put_block_group(block_group); |
2552d17e JB |
5725 | } |
5726 | up_read(&space_info->groups_sem); | |
5727 | ||
60d2adbb MX |
5728 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5729 | goto search; | |
5730 | ||
b742bb82 YZ |
5731 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5732 | goto search; | |
5733 | ||
285ff5af | 5734 | /* |
ccf0e725 JB |
5735 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5736 | * caching kthreads as we move along | |
817d52f8 JB |
5737 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5738 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5739 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5740 | * again | |
fa9c0d79 | 5741 | */ |
723bda20 | 5742 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5743 | index = 0; |
723bda20 | 5744 | loop++; |
817d52f8 | 5745 | if (loop == LOOP_ALLOC_CHUNK) { |
723bda20 JB |
5746 | if (allowed_chunk_alloc) { |
5747 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5748 | 2 * 1024 * 1024, data, | |
5749 | CHUNK_ALLOC_LIMITED); | |
79787eaa JM |
5750 | if (ret < 0) { |
5751 | btrfs_abort_transaction(trans, | |
5752 | root, ret); | |
5753 | goto out; | |
5754 | } | |
723bda20 JB |
5755 | allowed_chunk_alloc = 0; |
5756 | if (ret == 1) | |
5757 | done_chunk_alloc = 1; | |
5758 | } else if (!done_chunk_alloc && | |
5759 | space_info->force_alloc == | |
5760 | CHUNK_ALLOC_NO_FORCE) { | |
5761 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5762 | } | |
2552d17e | 5763 | |
723bda20 JB |
5764 | /* |
5765 | * We didn't allocate a chunk, go ahead and drop the | |
5766 | * empty size and loop again. | |
5767 | */ | |
5768 | if (!done_chunk_alloc) | |
5769 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5770 | } |
5771 | ||
723bda20 JB |
5772 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5773 | empty_size = 0; | |
5774 | empty_cluster = 0; | |
fa9c0d79 | 5775 | } |
723bda20 JB |
5776 | |
5777 | goto search; | |
2552d17e JB |
5778 | } else if (!ins->objectid) { |
5779 | ret = -ENOSPC; | |
d82a6f1d | 5780 | } else if (ins->objectid) { |
80eb234a | 5781 | ret = 0; |
be744175 | 5782 | } |
79787eaa | 5783 | out: |
be744175 | 5784 | |
0f70abe2 | 5785 | return ret; |
fec577fb | 5786 | } |
ec44a35c | 5787 | |
9ed74f2d JB |
5788 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5789 | int dump_block_groups) | |
0f9dd46c JB |
5790 | { |
5791 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5792 | int index = 0; |
0f9dd46c | 5793 | |
9ed74f2d | 5794 | spin_lock(&info->lock); |
fb25e914 JB |
5795 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5796 | (unsigned long long)info->flags, | |
d397712b | 5797 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5798 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5799 | info->bytes_readonly), |
d397712b | 5800 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5801 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5802 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5803 | (unsigned long long)info->total_bytes, |
8929ecfa | 5804 | (unsigned long long)info->bytes_used, |
21380931 | 5805 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5806 | (unsigned long long)info->bytes_reserved, |
21380931 | 5807 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5808 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5809 | spin_unlock(&info->lock); |
5810 | ||
5811 | if (!dump_block_groups) | |
5812 | return; | |
0f9dd46c | 5813 | |
80eb234a | 5814 | down_read(&info->groups_sem); |
b742bb82 YZ |
5815 | again: |
5816 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5817 | spin_lock(&cache->lock); |
d397712b CM |
5818 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5819 | "%llu pinned %llu reserved\n", | |
5820 | (unsigned long long)cache->key.objectid, | |
5821 | (unsigned long long)cache->key.offset, | |
5822 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5823 | (unsigned long long)cache->pinned, | |
5824 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5825 | btrfs_dump_free_space(cache, bytes); |
5826 | spin_unlock(&cache->lock); | |
5827 | } | |
b742bb82 YZ |
5828 | if (++index < BTRFS_NR_RAID_TYPES) |
5829 | goto again; | |
80eb234a | 5830 | up_read(&info->groups_sem); |
0f9dd46c | 5831 | } |
e8569813 | 5832 | |
11833d66 YZ |
5833 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5834 | struct btrfs_root *root, | |
5835 | u64 num_bytes, u64 min_alloc_size, | |
5836 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 5837 | struct btrfs_key *ins, u64 data) |
fec577fb | 5838 | { |
9e622d6b | 5839 | bool final_tried = false; |
fec577fb | 5840 | int ret; |
925baedd | 5841 | |
6a63209f | 5842 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5843 | again: |
0ef3e66b CM |
5844 | /* |
5845 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5846 | * is not called recursively on allocations | |
5847 | */ | |
79787eaa | 5848 | if (empty_size || root->ref_cows) { |
6324fbf3 | 5849 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5850 | num_bytes + 2 * 1024 * 1024, data, |
5851 | CHUNK_ALLOC_NO_FORCE); | |
79787eaa JM |
5852 | if (ret < 0 && ret != -ENOSPC) { |
5853 | btrfs_abort_transaction(trans, root, ret); | |
5854 | return ret; | |
5855 | } | |
5856 | } | |
0b86a832 | 5857 | |
db94535d CM |
5858 | WARN_ON(num_bytes < root->sectorsize); |
5859 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 5860 | hint_byte, ins, data); |
3b951516 | 5861 | |
9e622d6b MX |
5862 | if (ret == -ENOSPC) { |
5863 | if (!final_tried) { | |
5864 | num_bytes = num_bytes >> 1; | |
5865 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
5866 | num_bytes = max(num_bytes, min_alloc_size); | |
79787eaa | 5867 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
9e622d6b | 5868 | num_bytes, data, CHUNK_ALLOC_FORCE); |
79787eaa JM |
5869 | if (ret < 0 && ret != -ENOSPC) { |
5870 | btrfs_abort_transaction(trans, root, ret); | |
5871 | return ret; | |
5872 | } | |
9e622d6b MX |
5873 | if (num_bytes == min_alloc_size) |
5874 | final_tried = true; | |
5875 | goto again; | |
5876 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
5877 | struct btrfs_space_info *sinfo; | |
5878 | ||
5879 | sinfo = __find_space_info(root->fs_info, data); | |
5880 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
5881 | "wanted %llu\n", (unsigned long long)data, | |
5882 | (unsigned long long)num_bytes); | |
53804280 JM |
5883 | if (sinfo) |
5884 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 5885 | } |
925baedd | 5886 | } |
0f9dd46c | 5887 | |
1abe9b8a | 5888 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5889 | ||
0f9dd46c | 5890 | return ret; |
e6dcd2dc CM |
5891 | } |
5892 | ||
e688b725 CM |
5893 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
5894 | u64 start, u64 len, int pin) | |
65b51a00 | 5895 | { |
0f9dd46c | 5896 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5897 | int ret = 0; |
0f9dd46c | 5898 | |
0f9dd46c JB |
5899 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5900 | if (!cache) { | |
d397712b CM |
5901 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5902 | (unsigned long long)start); | |
0f9dd46c JB |
5903 | return -ENOSPC; |
5904 | } | |
1f3c79a2 | 5905 | |
5378e607 LD |
5906 | if (btrfs_test_opt(root, DISCARD)) |
5907 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5908 | |
e688b725 CM |
5909 | if (pin) |
5910 | pin_down_extent(root, cache, start, len, 1); | |
5911 | else { | |
5912 | btrfs_add_free_space(cache, start, len); | |
5913 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
5914 | } | |
fa9c0d79 | 5915 | btrfs_put_block_group(cache); |
817d52f8 | 5916 | |
1abe9b8a | 5917 | trace_btrfs_reserved_extent_free(root, start, len); |
5918 | ||
e6dcd2dc CM |
5919 | return ret; |
5920 | } | |
5921 | ||
e688b725 CM |
5922 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
5923 | u64 start, u64 len) | |
5924 | { | |
5925 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
5926 | } | |
5927 | ||
5928 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
5929 | u64 start, u64 len) | |
5930 | { | |
5931 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
5932 | } | |
5933 | ||
5d4f98a2 YZ |
5934 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5935 | struct btrfs_root *root, | |
5936 | u64 parent, u64 root_objectid, | |
5937 | u64 flags, u64 owner, u64 offset, | |
5938 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5939 | { |
5940 | int ret; | |
5d4f98a2 | 5941 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5942 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5943 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5944 | struct btrfs_path *path; |
5d4f98a2 YZ |
5945 | struct extent_buffer *leaf; |
5946 | int type; | |
5947 | u32 size; | |
26b8003f | 5948 | |
5d4f98a2 YZ |
5949 | if (parent > 0) |
5950 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5951 | else | |
5952 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5953 | |
5d4f98a2 | 5954 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5955 | |
5956 | path = btrfs_alloc_path(); | |
db5b493a TI |
5957 | if (!path) |
5958 | return -ENOMEM; | |
47e4bb98 | 5959 | |
b9473439 | 5960 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5961 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5962 | ins, size); | |
79787eaa JM |
5963 | if (ret) { |
5964 | btrfs_free_path(path); | |
5965 | return ret; | |
5966 | } | |
0f9dd46c | 5967 | |
5d4f98a2 YZ |
5968 | leaf = path->nodes[0]; |
5969 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5970 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5971 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5972 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5973 | btrfs_set_extent_flags(leaf, extent_item, | |
5974 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5975 | ||
5976 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5977 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5978 | if (parent > 0) { | |
5979 | struct btrfs_shared_data_ref *ref; | |
5980 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5981 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5982 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5983 | } else { | |
5984 | struct btrfs_extent_data_ref *ref; | |
5985 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5986 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5987 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5988 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5989 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5990 | } | |
47e4bb98 CM |
5991 | |
5992 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5993 | btrfs_free_path(path); |
f510cfec | 5994 | |
f0486c68 | 5995 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 5996 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
5997 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5998 | "%llu\n", (unsigned long long)ins->objectid, | |
5999 | (unsigned long long)ins->offset); | |
f5947066 CM |
6000 | BUG(); |
6001 | } | |
e6dcd2dc CM |
6002 | return ret; |
6003 | } | |
6004 | ||
5d4f98a2 YZ |
6005 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6006 | struct btrfs_root *root, | |
6007 | u64 parent, u64 root_objectid, | |
6008 | u64 flags, struct btrfs_disk_key *key, | |
6009 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6010 | { |
6011 | int ret; | |
5d4f98a2 YZ |
6012 | struct btrfs_fs_info *fs_info = root->fs_info; |
6013 | struct btrfs_extent_item *extent_item; | |
6014 | struct btrfs_tree_block_info *block_info; | |
6015 | struct btrfs_extent_inline_ref *iref; | |
6016 | struct btrfs_path *path; | |
6017 | struct extent_buffer *leaf; | |
6018 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 6019 | |
5d4f98a2 | 6020 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6021 | if (!path) |
6022 | return -ENOMEM; | |
56bec294 | 6023 | |
5d4f98a2 YZ |
6024 | path->leave_spinning = 1; |
6025 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6026 | ins, size); | |
79787eaa JM |
6027 | if (ret) { |
6028 | btrfs_free_path(path); | |
6029 | return ret; | |
6030 | } | |
5d4f98a2 YZ |
6031 | |
6032 | leaf = path->nodes[0]; | |
6033 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6034 | struct btrfs_extent_item); | |
6035 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6036 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6037 | btrfs_set_extent_flags(leaf, extent_item, | |
6038 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
6039 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6040 | ||
6041 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6042 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6043 | ||
6044 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6045 | if (parent > 0) { | |
6046 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6047 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6048 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6049 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6050 | } else { | |
6051 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6052 | BTRFS_TREE_BLOCK_REF_KEY); | |
6053 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6054 | } | |
6055 | ||
6056 | btrfs_mark_buffer_dirty(leaf); | |
6057 | btrfs_free_path(path); | |
6058 | ||
f0486c68 | 6059 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6060 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6061 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6062 | "%llu\n", (unsigned long long)ins->objectid, | |
6063 | (unsigned long long)ins->offset); | |
6064 | BUG(); | |
6065 | } | |
6066 | return ret; | |
6067 | } | |
6068 | ||
6069 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6070 | struct btrfs_root *root, | |
6071 | u64 root_objectid, u64 owner, | |
6072 | u64 offset, struct btrfs_key *ins) | |
6073 | { | |
6074 | int ret; | |
6075 | ||
6076 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6077 | ||
66d7e7f0 AJ |
6078 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6079 | ins->offset, 0, | |
6080 | root_objectid, owner, offset, | |
6081 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6082 | return ret; |
6083 | } | |
e02119d5 CM |
6084 | |
6085 | /* | |
6086 | * this is used by the tree logging recovery code. It records that | |
6087 | * an extent has been allocated and makes sure to clear the free | |
6088 | * space cache bits as well | |
6089 | */ | |
5d4f98a2 YZ |
6090 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6091 | struct btrfs_root *root, | |
6092 | u64 root_objectid, u64 owner, u64 offset, | |
6093 | struct btrfs_key *ins) | |
e02119d5 CM |
6094 | { |
6095 | int ret; | |
6096 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6097 | struct btrfs_caching_control *caching_ctl; |
6098 | u64 start = ins->objectid; | |
6099 | u64 num_bytes = ins->offset; | |
e02119d5 | 6100 | |
e02119d5 | 6101 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 6102 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 6103 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6104 | |
11833d66 YZ |
6105 | if (!caching_ctl) { |
6106 | BUG_ON(!block_group_cache_done(block_group)); | |
6107 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6108 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6109 | } else { |
6110 | mutex_lock(&caching_ctl->mutex); | |
6111 | ||
6112 | if (start >= caching_ctl->progress) { | |
6113 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6114 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6115 | } else if (start + num_bytes <= caching_ctl->progress) { |
6116 | ret = btrfs_remove_free_space(block_group, | |
6117 | start, num_bytes); | |
79787eaa | 6118 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6119 | } else { |
6120 | num_bytes = caching_ctl->progress - start; | |
6121 | ret = btrfs_remove_free_space(block_group, | |
6122 | start, num_bytes); | |
79787eaa | 6123 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6124 | |
6125 | start = caching_ctl->progress; | |
6126 | num_bytes = ins->objectid + ins->offset - | |
6127 | caching_ctl->progress; | |
6128 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6129 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6130 | } |
6131 | ||
6132 | mutex_unlock(&caching_ctl->mutex); | |
6133 | put_caching_control(caching_ctl); | |
6134 | } | |
6135 | ||
fb25e914 JB |
6136 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6137 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6138 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6139 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6140 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6141 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6142 | return ret; |
6143 | } | |
6144 | ||
65b51a00 CM |
6145 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6146 | struct btrfs_root *root, | |
4008c04a CM |
6147 | u64 bytenr, u32 blocksize, |
6148 | int level) | |
65b51a00 CM |
6149 | { |
6150 | struct extent_buffer *buf; | |
6151 | ||
6152 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6153 | if (!buf) | |
6154 | return ERR_PTR(-ENOMEM); | |
6155 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6156 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6157 | btrfs_tree_lock(buf); |
6158 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6159 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6160 | |
6161 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6162 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6163 | |
d0c803c4 | 6164 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6165 | /* |
6166 | * we allow two log transactions at a time, use different | |
6167 | * EXENT bit to differentiate dirty pages. | |
6168 | */ | |
6169 | if (root->log_transid % 2 == 0) | |
6170 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6171 | buf->start + buf->len - 1, GFP_NOFS); | |
6172 | else | |
6173 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6174 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6175 | } else { |
6176 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6177 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6178 | } |
65b51a00 | 6179 | trans->blocks_used++; |
b4ce94de | 6180 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6181 | return buf; |
6182 | } | |
6183 | ||
f0486c68 YZ |
6184 | static struct btrfs_block_rsv * |
6185 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6186 | struct btrfs_root *root, u32 blocksize) | |
6187 | { | |
6188 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6189 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6190 | int ret; |
6191 | ||
6192 | block_rsv = get_block_rsv(trans, root); | |
6193 | ||
6194 | if (block_rsv->size == 0) { | |
36ba022a | 6195 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
6196 | /* |
6197 | * If we couldn't reserve metadata bytes try and use some from | |
6198 | * the global reserve. | |
6199 | */ | |
6200 | if (ret && block_rsv != global_rsv) { | |
6201 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6202 | if (!ret) | |
6203 | return global_rsv; | |
f0486c68 | 6204 | return ERR_PTR(ret); |
68a82277 | 6205 | } else if (ret) { |
f0486c68 | 6206 | return ERR_PTR(ret); |
68a82277 | 6207 | } |
f0486c68 YZ |
6208 | return block_rsv; |
6209 | } | |
6210 | ||
6211 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6212 | if (!ret) | |
6213 | return block_rsv; | |
68a82277 | 6214 | if (ret) { |
dff51cd1 DS |
6215 | static DEFINE_RATELIMIT_STATE(_rs, |
6216 | DEFAULT_RATELIMIT_INTERVAL, | |
6217 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
6218 | if (__ratelimit(&_rs)) { | |
6219 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
6220 | WARN_ON(1); | |
6221 | } | |
36ba022a | 6222 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 6223 | if (!ret) { |
68a82277 JB |
6224 | return block_rsv; |
6225 | } else if (ret && block_rsv != global_rsv) { | |
6226 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6227 | if (!ret) | |
6228 | return global_rsv; | |
6229 | } | |
6230 | } | |
f0486c68 | 6231 | |
f0486c68 YZ |
6232 | return ERR_PTR(-ENOSPC); |
6233 | } | |
6234 | ||
8c2a3ca2 JB |
6235 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6236 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6237 | { |
6238 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6239 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6240 | } |
6241 | ||
fec577fb | 6242 | /* |
f0486c68 YZ |
6243 | * finds a free extent and does all the dirty work required for allocation |
6244 | * returns the key for the extent through ins, and a tree buffer for | |
6245 | * the first block of the extent through buf. | |
6246 | * | |
fec577fb CM |
6247 | * returns the tree buffer or NULL. |
6248 | */ | |
5f39d397 | 6249 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6250 | struct btrfs_root *root, u32 blocksize, |
6251 | u64 parent, u64 root_objectid, | |
6252 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6253 | u64 hint, u64 empty_size) |
fec577fb | 6254 | { |
e2fa7227 | 6255 | struct btrfs_key ins; |
f0486c68 | 6256 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6257 | struct extent_buffer *buf; |
f0486c68 YZ |
6258 | u64 flags = 0; |
6259 | int ret; | |
6260 | ||
fec577fb | 6261 | |
f0486c68 YZ |
6262 | block_rsv = use_block_rsv(trans, root, blocksize); |
6263 | if (IS_ERR(block_rsv)) | |
6264 | return ERR_CAST(block_rsv); | |
6265 | ||
6266 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6267 | empty_size, hint, &ins, 0); |
fec577fb | 6268 | if (ret) { |
8c2a3ca2 | 6269 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6270 | return ERR_PTR(ret); |
fec577fb | 6271 | } |
55c69072 | 6272 | |
4008c04a CM |
6273 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6274 | blocksize, level); | |
79787eaa | 6275 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6276 | |
6277 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6278 | if (parent == 0) | |
6279 | parent = ins.objectid; | |
6280 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6281 | } else | |
6282 | BUG_ON(parent > 0); | |
6283 | ||
6284 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6285 | struct btrfs_delayed_extent_op *extent_op; | |
6286 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
79787eaa | 6287 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6288 | if (key) |
6289 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6290 | else | |
6291 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6292 | extent_op->flags_to_set = flags; | |
6293 | extent_op->update_key = 1; | |
6294 | extent_op->update_flags = 1; | |
6295 | extent_op->is_data = 0; | |
6296 | ||
66d7e7f0 AJ |
6297 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6298 | ins.objectid, | |
f0486c68 YZ |
6299 | ins.offset, parent, root_objectid, |
6300 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6301 | extent_op, 0); |
79787eaa | 6302 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6303 | } |
fec577fb CM |
6304 | return buf; |
6305 | } | |
a28ec197 | 6306 | |
2c47e605 YZ |
6307 | struct walk_control { |
6308 | u64 refs[BTRFS_MAX_LEVEL]; | |
6309 | u64 flags[BTRFS_MAX_LEVEL]; | |
6310 | struct btrfs_key update_progress; | |
6311 | int stage; | |
6312 | int level; | |
6313 | int shared_level; | |
6314 | int update_ref; | |
6315 | int keep_locks; | |
1c4850e2 YZ |
6316 | int reada_slot; |
6317 | int reada_count; | |
66d7e7f0 | 6318 | int for_reloc; |
2c47e605 YZ |
6319 | }; |
6320 | ||
6321 | #define DROP_REFERENCE 1 | |
6322 | #define UPDATE_BACKREF 2 | |
6323 | ||
1c4850e2 YZ |
6324 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6325 | struct btrfs_root *root, | |
6326 | struct walk_control *wc, | |
6327 | struct btrfs_path *path) | |
6407bf6d | 6328 | { |
1c4850e2 YZ |
6329 | u64 bytenr; |
6330 | u64 generation; | |
6331 | u64 refs; | |
94fcca9f | 6332 | u64 flags; |
5d4f98a2 | 6333 | u32 nritems; |
1c4850e2 YZ |
6334 | u32 blocksize; |
6335 | struct btrfs_key key; | |
6336 | struct extent_buffer *eb; | |
6407bf6d | 6337 | int ret; |
1c4850e2 YZ |
6338 | int slot; |
6339 | int nread = 0; | |
6407bf6d | 6340 | |
1c4850e2 YZ |
6341 | if (path->slots[wc->level] < wc->reada_slot) { |
6342 | wc->reada_count = wc->reada_count * 2 / 3; | |
6343 | wc->reada_count = max(wc->reada_count, 2); | |
6344 | } else { | |
6345 | wc->reada_count = wc->reada_count * 3 / 2; | |
6346 | wc->reada_count = min_t(int, wc->reada_count, | |
6347 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6348 | } | |
7bb86316 | 6349 | |
1c4850e2 YZ |
6350 | eb = path->nodes[wc->level]; |
6351 | nritems = btrfs_header_nritems(eb); | |
6352 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6353 | |
1c4850e2 YZ |
6354 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6355 | if (nread >= wc->reada_count) | |
6356 | break; | |
bd56b302 | 6357 | |
2dd3e67b | 6358 | cond_resched(); |
1c4850e2 YZ |
6359 | bytenr = btrfs_node_blockptr(eb, slot); |
6360 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6361 | |
1c4850e2 YZ |
6362 | if (slot == path->slots[wc->level]) |
6363 | goto reada; | |
5d4f98a2 | 6364 | |
1c4850e2 YZ |
6365 | if (wc->stage == UPDATE_BACKREF && |
6366 | generation <= root->root_key.offset) | |
bd56b302 CM |
6367 | continue; |
6368 | ||
94fcca9f YZ |
6369 | /* We don't lock the tree block, it's OK to be racy here */ |
6370 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6371 | &refs, &flags); | |
79787eaa JM |
6372 | /* We don't care about errors in readahead. */ |
6373 | if (ret < 0) | |
6374 | continue; | |
94fcca9f YZ |
6375 | BUG_ON(refs == 0); |
6376 | ||
1c4850e2 | 6377 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6378 | if (refs == 1) |
6379 | goto reada; | |
bd56b302 | 6380 | |
94fcca9f YZ |
6381 | if (wc->level == 1 && |
6382 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6383 | continue; | |
1c4850e2 YZ |
6384 | if (!wc->update_ref || |
6385 | generation <= root->root_key.offset) | |
6386 | continue; | |
6387 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6388 | ret = btrfs_comp_cpu_keys(&key, | |
6389 | &wc->update_progress); | |
6390 | if (ret < 0) | |
6391 | continue; | |
94fcca9f YZ |
6392 | } else { |
6393 | if (wc->level == 1 && | |
6394 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6395 | continue; | |
6407bf6d | 6396 | } |
1c4850e2 YZ |
6397 | reada: |
6398 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6399 | generation); | |
6400 | if (ret) | |
bd56b302 | 6401 | break; |
1c4850e2 | 6402 | nread++; |
20524f02 | 6403 | } |
1c4850e2 | 6404 | wc->reada_slot = slot; |
20524f02 | 6405 | } |
2c47e605 | 6406 | |
f82d02d9 | 6407 | /* |
2c47e605 YZ |
6408 | * hepler to process tree block while walking down the tree. |
6409 | * | |
2c47e605 YZ |
6410 | * when wc->stage == UPDATE_BACKREF, this function updates |
6411 | * back refs for pointers in the block. | |
6412 | * | |
6413 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6414 | */ |
2c47e605 | 6415 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6416 | struct btrfs_root *root, |
2c47e605 | 6417 | struct btrfs_path *path, |
94fcca9f | 6418 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6419 | { |
2c47e605 YZ |
6420 | int level = wc->level; |
6421 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6422 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6423 | int ret; |
6424 | ||
2c47e605 YZ |
6425 | if (wc->stage == UPDATE_BACKREF && |
6426 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6427 | return 1; | |
f82d02d9 | 6428 | |
2c47e605 YZ |
6429 | /* |
6430 | * when reference count of tree block is 1, it won't increase | |
6431 | * again. once full backref flag is set, we never clear it. | |
6432 | */ | |
94fcca9f YZ |
6433 | if (lookup_info && |
6434 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6435 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6436 | BUG_ON(!path->locks[level]); |
6437 | ret = btrfs_lookup_extent_info(trans, root, | |
6438 | eb->start, eb->len, | |
6439 | &wc->refs[level], | |
6440 | &wc->flags[level]); | |
79787eaa JM |
6441 | BUG_ON(ret == -ENOMEM); |
6442 | if (ret) | |
6443 | return ret; | |
2c47e605 YZ |
6444 | BUG_ON(wc->refs[level] == 0); |
6445 | } | |
5d4f98a2 | 6446 | |
2c47e605 YZ |
6447 | if (wc->stage == DROP_REFERENCE) { |
6448 | if (wc->refs[level] > 1) | |
6449 | return 1; | |
f82d02d9 | 6450 | |
2c47e605 | 6451 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6452 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6453 | path->locks[level] = 0; |
6454 | } | |
6455 | return 0; | |
6456 | } | |
f82d02d9 | 6457 | |
2c47e605 YZ |
6458 | /* wc->stage == UPDATE_BACKREF */ |
6459 | if (!(wc->flags[level] & flag)) { | |
6460 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6461 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6462 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6463 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6464 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6465 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6466 | eb->len, flag, 0); | |
79787eaa | 6467 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6468 | wc->flags[level] |= flag; |
6469 | } | |
6470 | ||
6471 | /* | |
6472 | * the block is shared by multiple trees, so it's not good to | |
6473 | * keep the tree lock | |
6474 | */ | |
6475 | if (path->locks[level] && level > 0) { | |
bd681513 | 6476 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6477 | path->locks[level] = 0; |
6478 | } | |
6479 | return 0; | |
6480 | } | |
6481 | ||
1c4850e2 YZ |
6482 | /* |
6483 | * hepler to process tree block pointer. | |
6484 | * | |
6485 | * when wc->stage == DROP_REFERENCE, this function checks | |
6486 | * reference count of the block pointed to. if the block | |
6487 | * is shared and we need update back refs for the subtree | |
6488 | * rooted at the block, this function changes wc->stage to | |
6489 | * UPDATE_BACKREF. if the block is shared and there is no | |
6490 | * need to update back, this function drops the reference | |
6491 | * to the block. | |
6492 | * | |
6493 | * NOTE: return value 1 means we should stop walking down. | |
6494 | */ | |
6495 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6496 | struct btrfs_root *root, | |
6497 | struct btrfs_path *path, | |
94fcca9f | 6498 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6499 | { |
6500 | u64 bytenr; | |
6501 | u64 generation; | |
6502 | u64 parent; | |
6503 | u32 blocksize; | |
6504 | struct btrfs_key key; | |
6505 | struct extent_buffer *next; | |
6506 | int level = wc->level; | |
6507 | int reada = 0; | |
6508 | int ret = 0; | |
6509 | ||
6510 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6511 | path->slots[level]); | |
6512 | /* | |
6513 | * if the lower level block was created before the snapshot | |
6514 | * was created, we know there is no need to update back refs | |
6515 | * for the subtree | |
6516 | */ | |
6517 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6518 | generation <= root->root_key.offset) { |
6519 | *lookup_info = 1; | |
1c4850e2 | 6520 | return 1; |
94fcca9f | 6521 | } |
1c4850e2 YZ |
6522 | |
6523 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6524 | blocksize = btrfs_level_size(root, level - 1); | |
6525 | ||
6526 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6527 | if (!next) { | |
6528 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6529 | if (!next) |
6530 | return -ENOMEM; | |
1c4850e2 YZ |
6531 | reada = 1; |
6532 | } | |
6533 | btrfs_tree_lock(next); | |
6534 | btrfs_set_lock_blocking(next); | |
6535 | ||
94fcca9f YZ |
6536 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6537 | &wc->refs[level - 1], | |
6538 | &wc->flags[level - 1]); | |
79787eaa JM |
6539 | if (ret < 0) { |
6540 | btrfs_tree_unlock(next); | |
6541 | return ret; | |
6542 | } | |
6543 | ||
94fcca9f YZ |
6544 | BUG_ON(wc->refs[level - 1] == 0); |
6545 | *lookup_info = 0; | |
1c4850e2 | 6546 | |
94fcca9f | 6547 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6548 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6549 | if (level == 1 && |
6550 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6551 | goto skip; | |
6552 | ||
1c4850e2 YZ |
6553 | if (!wc->update_ref || |
6554 | generation <= root->root_key.offset) | |
6555 | goto skip; | |
6556 | ||
6557 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6558 | path->slots[level]); | |
6559 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6560 | if (ret < 0) | |
6561 | goto skip; | |
6562 | ||
6563 | wc->stage = UPDATE_BACKREF; | |
6564 | wc->shared_level = level - 1; | |
6565 | } | |
94fcca9f YZ |
6566 | } else { |
6567 | if (level == 1 && | |
6568 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6569 | goto skip; | |
1c4850e2 YZ |
6570 | } |
6571 | ||
b9fab919 | 6572 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
6573 | btrfs_tree_unlock(next); |
6574 | free_extent_buffer(next); | |
6575 | next = NULL; | |
94fcca9f | 6576 | *lookup_info = 1; |
1c4850e2 YZ |
6577 | } |
6578 | ||
6579 | if (!next) { | |
6580 | if (reada && level == 1) | |
6581 | reada_walk_down(trans, root, wc, path); | |
6582 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6583 | if (!next) |
6584 | return -EIO; | |
1c4850e2 YZ |
6585 | btrfs_tree_lock(next); |
6586 | btrfs_set_lock_blocking(next); | |
6587 | } | |
6588 | ||
6589 | level--; | |
6590 | BUG_ON(level != btrfs_header_level(next)); | |
6591 | path->nodes[level] = next; | |
6592 | path->slots[level] = 0; | |
bd681513 | 6593 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6594 | wc->level = level; |
6595 | if (wc->level == 1) | |
6596 | wc->reada_slot = 0; | |
6597 | return 0; | |
6598 | skip: | |
6599 | wc->refs[level - 1] = 0; | |
6600 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6601 | if (wc->stage == DROP_REFERENCE) { |
6602 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6603 | parent = path->nodes[level]->start; | |
6604 | } else { | |
6605 | BUG_ON(root->root_key.objectid != | |
6606 | btrfs_header_owner(path->nodes[level])); | |
6607 | parent = 0; | |
6608 | } | |
1c4850e2 | 6609 | |
94fcca9f | 6610 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6611 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 6612 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 6613 | } |
1c4850e2 YZ |
6614 | btrfs_tree_unlock(next); |
6615 | free_extent_buffer(next); | |
94fcca9f | 6616 | *lookup_info = 1; |
1c4850e2 YZ |
6617 | return 1; |
6618 | } | |
6619 | ||
2c47e605 YZ |
6620 | /* |
6621 | * hepler to process tree block while walking up the tree. | |
6622 | * | |
6623 | * when wc->stage == DROP_REFERENCE, this function drops | |
6624 | * reference count on the block. | |
6625 | * | |
6626 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6627 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6628 | * to UPDATE_BACKREF previously while processing the block. | |
6629 | * | |
6630 | * NOTE: return value 1 means we should stop walking up. | |
6631 | */ | |
6632 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6633 | struct btrfs_root *root, | |
6634 | struct btrfs_path *path, | |
6635 | struct walk_control *wc) | |
6636 | { | |
f0486c68 | 6637 | int ret; |
2c47e605 YZ |
6638 | int level = wc->level; |
6639 | struct extent_buffer *eb = path->nodes[level]; | |
6640 | u64 parent = 0; | |
6641 | ||
6642 | if (wc->stage == UPDATE_BACKREF) { | |
6643 | BUG_ON(wc->shared_level < level); | |
6644 | if (level < wc->shared_level) | |
6645 | goto out; | |
6646 | ||
2c47e605 YZ |
6647 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6648 | if (ret > 0) | |
6649 | wc->update_ref = 0; | |
6650 | ||
6651 | wc->stage = DROP_REFERENCE; | |
6652 | wc->shared_level = -1; | |
6653 | path->slots[level] = 0; | |
6654 | ||
6655 | /* | |
6656 | * check reference count again if the block isn't locked. | |
6657 | * we should start walking down the tree again if reference | |
6658 | * count is one. | |
6659 | */ | |
6660 | if (!path->locks[level]) { | |
6661 | BUG_ON(level == 0); | |
6662 | btrfs_tree_lock(eb); | |
6663 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6664 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6665 | |
6666 | ret = btrfs_lookup_extent_info(trans, root, | |
6667 | eb->start, eb->len, | |
6668 | &wc->refs[level], | |
6669 | &wc->flags[level]); | |
79787eaa JM |
6670 | if (ret < 0) { |
6671 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
6672 | return ret; | |
6673 | } | |
2c47e605 YZ |
6674 | BUG_ON(wc->refs[level] == 0); |
6675 | if (wc->refs[level] == 1) { | |
bd681513 | 6676 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6677 | return 1; |
6678 | } | |
f82d02d9 | 6679 | } |
2c47e605 | 6680 | } |
f82d02d9 | 6681 | |
2c47e605 YZ |
6682 | /* wc->stage == DROP_REFERENCE */ |
6683 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6684 | |
2c47e605 YZ |
6685 | if (wc->refs[level] == 1) { |
6686 | if (level == 0) { | |
6687 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6688 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6689 | wc->for_reloc); | |
2c47e605 | 6690 | else |
66d7e7f0 AJ |
6691 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6692 | wc->for_reloc); | |
79787eaa | 6693 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6694 | } |
6695 | /* make block locked assertion in clean_tree_block happy */ | |
6696 | if (!path->locks[level] && | |
6697 | btrfs_header_generation(eb) == trans->transid) { | |
6698 | btrfs_tree_lock(eb); | |
6699 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6700 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6701 | } |
6702 | clean_tree_block(trans, root, eb); | |
6703 | } | |
6704 | ||
6705 | if (eb == root->node) { | |
6706 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6707 | parent = eb->start; | |
6708 | else | |
6709 | BUG_ON(root->root_key.objectid != | |
6710 | btrfs_header_owner(eb)); | |
6711 | } else { | |
6712 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6713 | parent = path->nodes[level + 1]->start; | |
6714 | else | |
6715 | BUG_ON(root->root_key.objectid != | |
6716 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6717 | } |
f82d02d9 | 6718 | |
5581a51a | 6719 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6720 | out: |
6721 | wc->refs[level] = 0; | |
6722 | wc->flags[level] = 0; | |
f0486c68 | 6723 | return 0; |
2c47e605 YZ |
6724 | } |
6725 | ||
6726 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6727 | struct btrfs_root *root, | |
6728 | struct btrfs_path *path, | |
6729 | struct walk_control *wc) | |
6730 | { | |
2c47e605 | 6731 | int level = wc->level; |
94fcca9f | 6732 | int lookup_info = 1; |
2c47e605 YZ |
6733 | int ret; |
6734 | ||
6735 | while (level >= 0) { | |
94fcca9f | 6736 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6737 | if (ret > 0) |
6738 | break; | |
6739 | ||
6740 | if (level == 0) | |
6741 | break; | |
6742 | ||
7a7965f8 YZ |
6743 | if (path->slots[level] >= |
6744 | btrfs_header_nritems(path->nodes[level])) | |
6745 | break; | |
6746 | ||
94fcca9f | 6747 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6748 | if (ret > 0) { |
6749 | path->slots[level]++; | |
6750 | continue; | |
90d2c51d MX |
6751 | } else if (ret < 0) |
6752 | return ret; | |
1c4850e2 | 6753 | level = wc->level; |
f82d02d9 | 6754 | } |
f82d02d9 YZ |
6755 | return 0; |
6756 | } | |
6757 | ||
d397712b | 6758 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6759 | struct btrfs_root *root, |
f82d02d9 | 6760 | struct btrfs_path *path, |
2c47e605 | 6761 | struct walk_control *wc, int max_level) |
20524f02 | 6762 | { |
2c47e605 | 6763 | int level = wc->level; |
20524f02 | 6764 | int ret; |
9f3a7427 | 6765 | |
2c47e605 YZ |
6766 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6767 | while (level < max_level && path->nodes[level]) { | |
6768 | wc->level = level; | |
6769 | if (path->slots[level] + 1 < | |
6770 | btrfs_header_nritems(path->nodes[level])) { | |
6771 | path->slots[level]++; | |
20524f02 CM |
6772 | return 0; |
6773 | } else { | |
2c47e605 YZ |
6774 | ret = walk_up_proc(trans, root, path, wc); |
6775 | if (ret > 0) | |
6776 | return 0; | |
bd56b302 | 6777 | |
2c47e605 | 6778 | if (path->locks[level]) { |
bd681513 CM |
6779 | btrfs_tree_unlock_rw(path->nodes[level], |
6780 | path->locks[level]); | |
2c47e605 | 6781 | path->locks[level] = 0; |
f82d02d9 | 6782 | } |
2c47e605 YZ |
6783 | free_extent_buffer(path->nodes[level]); |
6784 | path->nodes[level] = NULL; | |
6785 | level++; | |
20524f02 CM |
6786 | } |
6787 | } | |
6788 | return 1; | |
6789 | } | |
6790 | ||
9aca1d51 | 6791 | /* |
2c47e605 YZ |
6792 | * drop a subvolume tree. |
6793 | * | |
6794 | * this function traverses the tree freeing any blocks that only | |
6795 | * referenced by the tree. | |
6796 | * | |
6797 | * when a shared tree block is found. this function decreases its | |
6798 | * reference count by one. if update_ref is true, this function | |
6799 | * also make sure backrefs for the shared block and all lower level | |
6800 | * blocks are properly updated. | |
9aca1d51 | 6801 | */ |
2c536799 | 6802 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6803 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6804 | int for_reloc) | |
20524f02 | 6805 | { |
5caf2a00 | 6806 | struct btrfs_path *path; |
2c47e605 YZ |
6807 | struct btrfs_trans_handle *trans; |
6808 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6809 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6810 | struct walk_control *wc; |
6811 | struct btrfs_key key; | |
6812 | int err = 0; | |
6813 | int ret; | |
6814 | int level; | |
20524f02 | 6815 | |
5caf2a00 | 6816 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6817 | if (!path) { |
6818 | err = -ENOMEM; | |
6819 | goto out; | |
6820 | } | |
20524f02 | 6821 | |
2c47e605 | 6822 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6823 | if (!wc) { |
6824 | btrfs_free_path(path); | |
cb1b69f4 TI |
6825 | err = -ENOMEM; |
6826 | goto out; | |
38a1a919 | 6827 | } |
2c47e605 | 6828 | |
a22285a6 | 6829 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6830 | if (IS_ERR(trans)) { |
6831 | err = PTR_ERR(trans); | |
6832 | goto out_free; | |
6833 | } | |
98d5dc13 | 6834 | |
3fd0a558 YZ |
6835 | if (block_rsv) |
6836 | trans->block_rsv = block_rsv; | |
2c47e605 | 6837 | |
9f3a7427 | 6838 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6839 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6840 | path->nodes[level] = btrfs_lock_root_node(root); |
6841 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6842 | path->slots[level] = 0; |
bd681513 | 6843 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6844 | memset(&wc->update_progress, 0, |
6845 | sizeof(wc->update_progress)); | |
9f3a7427 | 6846 | } else { |
9f3a7427 | 6847 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6848 | memcpy(&wc->update_progress, &key, |
6849 | sizeof(wc->update_progress)); | |
6850 | ||
6702ed49 | 6851 | level = root_item->drop_level; |
2c47e605 | 6852 | BUG_ON(level == 0); |
6702ed49 | 6853 | path->lowest_level = level; |
2c47e605 YZ |
6854 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6855 | path->lowest_level = 0; | |
6856 | if (ret < 0) { | |
6857 | err = ret; | |
79787eaa | 6858 | goto out_end_trans; |
9f3a7427 | 6859 | } |
1c4850e2 | 6860 | WARN_ON(ret > 0); |
2c47e605 | 6861 | |
7d9eb12c CM |
6862 | /* |
6863 | * unlock our path, this is safe because only this | |
6864 | * function is allowed to delete this snapshot | |
6865 | */ | |
5d4f98a2 | 6866 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6867 | |
6868 | level = btrfs_header_level(root->node); | |
6869 | while (1) { | |
6870 | btrfs_tree_lock(path->nodes[level]); | |
6871 | btrfs_set_lock_blocking(path->nodes[level]); | |
6872 | ||
6873 | ret = btrfs_lookup_extent_info(trans, root, | |
6874 | path->nodes[level]->start, | |
6875 | path->nodes[level]->len, | |
6876 | &wc->refs[level], | |
6877 | &wc->flags[level]); | |
79787eaa JM |
6878 | if (ret < 0) { |
6879 | err = ret; | |
6880 | goto out_end_trans; | |
6881 | } | |
2c47e605 YZ |
6882 | BUG_ON(wc->refs[level] == 0); |
6883 | ||
6884 | if (level == root_item->drop_level) | |
6885 | break; | |
6886 | ||
6887 | btrfs_tree_unlock(path->nodes[level]); | |
6888 | WARN_ON(wc->refs[level] != 1); | |
6889 | level--; | |
6890 | } | |
9f3a7427 | 6891 | } |
2c47e605 YZ |
6892 | |
6893 | wc->level = level; | |
6894 | wc->shared_level = -1; | |
6895 | wc->stage = DROP_REFERENCE; | |
6896 | wc->update_ref = update_ref; | |
6897 | wc->keep_locks = 0; | |
66d7e7f0 | 6898 | wc->for_reloc = for_reloc; |
1c4850e2 | 6899 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6900 | |
d397712b | 6901 | while (1) { |
2c47e605 YZ |
6902 | ret = walk_down_tree(trans, root, path, wc); |
6903 | if (ret < 0) { | |
6904 | err = ret; | |
20524f02 | 6905 | break; |
2c47e605 | 6906 | } |
9aca1d51 | 6907 | |
2c47e605 YZ |
6908 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6909 | if (ret < 0) { | |
6910 | err = ret; | |
20524f02 | 6911 | break; |
2c47e605 YZ |
6912 | } |
6913 | ||
6914 | if (ret > 0) { | |
6915 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6916 | break; |
6917 | } | |
2c47e605 YZ |
6918 | |
6919 | if (wc->stage == DROP_REFERENCE) { | |
6920 | level = wc->level; | |
6921 | btrfs_node_key(path->nodes[level], | |
6922 | &root_item->drop_progress, | |
6923 | path->slots[level]); | |
6924 | root_item->drop_level = level; | |
6925 | } | |
6926 | ||
6927 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6928 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6929 | ret = btrfs_update_root(trans, tree_root, |
6930 | &root->root_key, | |
6931 | root_item); | |
79787eaa JM |
6932 | if (ret) { |
6933 | btrfs_abort_transaction(trans, tree_root, ret); | |
6934 | err = ret; | |
6935 | goto out_end_trans; | |
6936 | } | |
2c47e605 | 6937 | |
3fd0a558 | 6938 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6939 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6940 | if (IS_ERR(trans)) { |
6941 | err = PTR_ERR(trans); | |
6942 | goto out_free; | |
6943 | } | |
3fd0a558 YZ |
6944 | if (block_rsv) |
6945 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6946 | } |
20524f02 | 6947 | } |
b3b4aa74 | 6948 | btrfs_release_path(path); |
79787eaa JM |
6949 | if (err) |
6950 | goto out_end_trans; | |
2c47e605 YZ |
6951 | |
6952 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
6953 | if (ret) { |
6954 | btrfs_abort_transaction(trans, tree_root, ret); | |
6955 | goto out_end_trans; | |
6956 | } | |
2c47e605 | 6957 | |
76dda93c YZ |
6958 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6959 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6960 | NULL, NULL); | |
79787eaa JM |
6961 | if (ret < 0) { |
6962 | btrfs_abort_transaction(trans, tree_root, ret); | |
6963 | err = ret; | |
6964 | goto out_end_trans; | |
6965 | } else if (ret > 0) { | |
84cd948c JB |
6966 | /* if we fail to delete the orphan item this time |
6967 | * around, it'll get picked up the next time. | |
6968 | * | |
6969 | * The most common failure here is just -ENOENT. | |
6970 | */ | |
6971 | btrfs_del_orphan_item(trans, tree_root, | |
6972 | root->root_key.objectid); | |
76dda93c YZ |
6973 | } |
6974 | } | |
6975 | ||
6976 | if (root->in_radix) { | |
6977 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6978 | } else { | |
6979 | free_extent_buffer(root->node); | |
6980 | free_extent_buffer(root->commit_root); | |
6981 | kfree(root); | |
6982 | } | |
79787eaa | 6983 | out_end_trans: |
3fd0a558 | 6984 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 6985 | out_free: |
2c47e605 | 6986 | kfree(wc); |
5caf2a00 | 6987 | btrfs_free_path(path); |
cb1b69f4 TI |
6988 | out: |
6989 | if (err) | |
6990 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 6991 | return err; |
20524f02 | 6992 | } |
9078a3e1 | 6993 | |
2c47e605 YZ |
6994 | /* |
6995 | * drop subtree rooted at tree block 'node'. | |
6996 | * | |
6997 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 6998 | * only used by relocation code |
2c47e605 | 6999 | */ |
f82d02d9 YZ |
7000 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7001 | struct btrfs_root *root, | |
7002 | struct extent_buffer *node, | |
7003 | struct extent_buffer *parent) | |
7004 | { | |
7005 | struct btrfs_path *path; | |
2c47e605 | 7006 | struct walk_control *wc; |
f82d02d9 YZ |
7007 | int level; |
7008 | int parent_level; | |
7009 | int ret = 0; | |
7010 | int wret; | |
7011 | ||
2c47e605 YZ |
7012 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7013 | ||
f82d02d9 | 7014 | path = btrfs_alloc_path(); |
db5b493a TI |
7015 | if (!path) |
7016 | return -ENOMEM; | |
f82d02d9 | 7017 | |
2c47e605 | 7018 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7019 | if (!wc) { |
7020 | btrfs_free_path(path); | |
7021 | return -ENOMEM; | |
7022 | } | |
2c47e605 | 7023 | |
b9447ef8 | 7024 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7025 | parent_level = btrfs_header_level(parent); |
7026 | extent_buffer_get(parent); | |
7027 | path->nodes[parent_level] = parent; | |
7028 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7029 | ||
b9447ef8 | 7030 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7031 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7032 | path->nodes[level] = node; |
7033 | path->slots[level] = 0; | |
bd681513 | 7034 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7035 | |
7036 | wc->refs[parent_level] = 1; | |
7037 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7038 | wc->level = level; | |
7039 | wc->shared_level = -1; | |
7040 | wc->stage = DROP_REFERENCE; | |
7041 | wc->update_ref = 0; | |
7042 | wc->keep_locks = 1; | |
66d7e7f0 | 7043 | wc->for_reloc = 1; |
1c4850e2 | 7044 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7045 | |
7046 | while (1) { | |
2c47e605 YZ |
7047 | wret = walk_down_tree(trans, root, path, wc); |
7048 | if (wret < 0) { | |
f82d02d9 | 7049 | ret = wret; |
f82d02d9 | 7050 | break; |
2c47e605 | 7051 | } |
f82d02d9 | 7052 | |
2c47e605 | 7053 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7054 | if (wret < 0) |
7055 | ret = wret; | |
7056 | if (wret != 0) | |
7057 | break; | |
7058 | } | |
7059 | ||
2c47e605 | 7060 | kfree(wc); |
f82d02d9 YZ |
7061 | btrfs_free_path(path); |
7062 | return ret; | |
7063 | } | |
7064 | ||
ec44a35c CM |
7065 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7066 | { | |
7067 | u64 num_devices; | |
fc67c450 | 7068 | u64 stripped; |
e4d8ec0f | 7069 | |
fc67c450 ID |
7070 | /* |
7071 | * if restripe for this chunk_type is on pick target profile and | |
7072 | * return, otherwise do the usual balance | |
7073 | */ | |
7074 | stripped = get_restripe_target(root->fs_info, flags); | |
7075 | if (stripped) | |
7076 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7077 | |
cd02dca5 CM |
7078 | /* |
7079 | * we add in the count of missing devices because we want | |
7080 | * to make sure that any RAID levels on a degraded FS | |
7081 | * continue to be honored. | |
7082 | */ | |
7083 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7084 | root->fs_info->fs_devices->missing_devices; | |
7085 | ||
fc67c450 ID |
7086 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
7087 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7088 | ||
ec44a35c CM |
7089 | if (num_devices == 1) { |
7090 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7091 | stripped = flags & ~stripped; | |
7092 | ||
7093 | /* turn raid0 into single device chunks */ | |
7094 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7095 | return stripped; | |
7096 | ||
7097 | /* turn mirroring into duplication */ | |
7098 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7099 | BTRFS_BLOCK_GROUP_RAID10)) | |
7100 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7101 | } else { |
7102 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7103 | if (flags & stripped) |
7104 | return flags; | |
7105 | ||
7106 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7107 | stripped = flags & ~stripped; | |
7108 | ||
7109 | /* switch duplicated blocks with raid1 */ | |
7110 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7111 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7112 | ||
e3176ca2 | 7113 | /* this is drive concat, leave it alone */ |
ec44a35c | 7114 | } |
e3176ca2 | 7115 | |
ec44a35c CM |
7116 | return flags; |
7117 | } | |
7118 | ||
199c36ea | 7119 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7120 | { |
f0486c68 YZ |
7121 | struct btrfs_space_info *sinfo = cache->space_info; |
7122 | u64 num_bytes; | |
199c36ea | 7123 | u64 min_allocable_bytes; |
f0486c68 | 7124 | int ret = -ENOSPC; |
0ef3e66b | 7125 | |
c286ac48 | 7126 | |
199c36ea MX |
7127 | /* |
7128 | * We need some metadata space and system metadata space for | |
7129 | * allocating chunks in some corner cases until we force to set | |
7130 | * it to be readonly. | |
7131 | */ | |
7132 | if ((sinfo->flags & | |
7133 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7134 | !force) | |
7135 | min_allocable_bytes = 1 * 1024 * 1024; | |
7136 | else | |
7137 | min_allocable_bytes = 0; | |
7138 | ||
f0486c68 YZ |
7139 | spin_lock(&sinfo->lock); |
7140 | spin_lock(&cache->lock); | |
61cfea9b W |
7141 | |
7142 | if (cache->ro) { | |
7143 | ret = 0; | |
7144 | goto out; | |
7145 | } | |
7146 | ||
f0486c68 YZ |
7147 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7148 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7149 | ||
7150 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7151 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7152 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7153 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7154 | cache->ro = 1; |
7155 | ret = 0; | |
7156 | } | |
61cfea9b | 7157 | out: |
f0486c68 YZ |
7158 | spin_unlock(&cache->lock); |
7159 | spin_unlock(&sinfo->lock); | |
7160 | return ret; | |
7161 | } | |
7d9eb12c | 7162 | |
f0486c68 YZ |
7163 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7164 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7165 | |
f0486c68 YZ |
7166 | { |
7167 | struct btrfs_trans_handle *trans; | |
7168 | u64 alloc_flags; | |
7169 | int ret; | |
7d9eb12c | 7170 | |
f0486c68 | 7171 | BUG_ON(cache->ro); |
0ef3e66b | 7172 | |
ff5714cc | 7173 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7174 | if (IS_ERR(trans)) |
7175 | return PTR_ERR(trans); | |
5d4f98a2 | 7176 | |
f0486c68 | 7177 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa JM |
7178 | if (alloc_flags != cache->flags) { |
7179 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, | |
7180 | CHUNK_ALLOC_FORCE); | |
7181 | if (ret < 0) | |
7182 | goto out; | |
7183 | } | |
5d4f98a2 | 7184 | |
199c36ea | 7185 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7186 | if (!ret) |
7187 | goto out; | |
7188 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
7189 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7190 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
7191 | if (ret < 0) |
7192 | goto out; | |
199c36ea | 7193 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7194 | out: |
7195 | btrfs_end_transaction(trans, root); | |
7196 | return ret; | |
7197 | } | |
5d4f98a2 | 7198 | |
c87f08ca CM |
7199 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7200 | struct btrfs_root *root, u64 type) | |
7201 | { | |
7202 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
7203 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7204 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
7205 | } |
7206 | ||
6d07bcec MX |
7207 | /* |
7208 | * helper to account the unused space of all the readonly block group in the | |
7209 | * list. takes mirrors into account. | |
7210 | */ | |
7211 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7212 | { | |
7213 | struct btrfs_block_group_cache *block_group; | |
7214 | u64 free_bytes = 0; | |
7215 | int factor; | |
7216 | ||
7217 | list_for_each_entry(block_group, groups_list, list) { | |
7218 | spin_lock(&block_group->lock); | |
7219 | ||
7220 | if (!block_group->ro) { | |
7221 | spin_unlock(&block_group->lock); | |
7222 | continue; | |
7223 | } | |
7224 | ||
7225 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7226 | BTRFS_BLOCK_GROUP_RAID10 | | |
7227 | BTRFS_BLOCK_GROUP_DUP)) | |
7228 | factor = 2; | |
7229 | else | |
7230 | factor = 1; | |
7231 | ||
7232 | free_bytes += (block_group->key.offset - | |
7233 | btrfs_block_group_used(&block_group->item)) * | |
7234 | factor; | |
7235 | ||
7236 | spin_unlock(&block_group->lock); | |
7237 | } | |
7238 | ||
7239 | return free_bytes; | |
7240 | } | |
7241 | ||
7242 | /* | |
7243 | * helper to account the unused space of all the readonly block group in the | |
7244 | * space_info. takes mirrors into account. | |
7245 | */ | |
7246 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7247 | { | |
7248 | int i; | |
7249 | u64 free_bytes = 0; | |
7250 | ||
7251 | spin_lock(&sinfo->lock); | |
7252 | ||
7253 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7254 | if (!list_empty(&sinfo->block_groups[i])) | |
7255 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7256 | &sinfo->block_groups[i]); | |
7257 | ||
7258 | spin_unlock(&sinfo->lock); | |
7259 | ||
7260 | return free_bytes; | |
7261 | } | |
7262 | ||
143bede5 | 7263 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7264 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7265 | { |
f0486c68 YZ |
7266 | struct btrfs_space_info *sinfo = cache->space_info; |
7267 | u64 num_bytes; | |
7268 | ||
7269 | BUG_ON(!cache->ro); | |
7270 | ||
7271 | spin_lock(&sinfo->lock); | |
7272 | spin_lock(&cache->lock); | |
7273 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7274 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7275 | sinfo->bytes_readonly -= num_bytes; | |
7276 | cache->ro = 0; | |
7277 | spin_unlock(&cache->lock); | |
7278 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7279 | } |
7280 | ||
ba1bf481 JB |
7281 | /* |
7282 | * checks to see if its even possible to relocate this block group. | |
7283 | * | |
7284 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7285 | * ok to go ahead and try. | |
7286 | */ | |
7287 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7288 | { |
ba1bf481 JB |
7289 | struct btrfs_block_group_cache *block_group; |
7290 | struct btrfs_space_info *space_info; | |
7291 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7292 | struct btrfs_device *device; | |
cdcb725c | 7293 | u64 min_free; |
6719db6a JB |
7294 | u64 dev_min = 1; |
7295 | u64 dev_nr = 0; | |
4a5e98f5 | 7296 | u64 target; |
cdcb725c | 7297 | int index; |
ba1bf481 JB |
7298 | int full = 0; |
7299 | int ret = 0; | |
1a40e23b | 7300 | |
ba1bf481 | 7301 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7302 | |
ba1bf481 JB |
7303 | /* odd, couldn't find the block group, leave it alone */ |
7304 | if (!block_group) | |
7305 | return -1; | |
1a40e23b | 7306 | |
cdcb725c | 7307 | min_free = btrfs_block_group_used(&block_group->item); |
7308 | ||
ba1bf481 | 7309 | /* no bytes used, we're good */ |
cdcb725c | 7310 | if (!min_free) |
1a40e23b ZY |
7311 | goto out; |
7312 | ||
ba1bf481 JB |
7313 | space_info = block_group->space_info; |
7314 | spin_lock(&space_info->lock); | |
17d217fe | 7315 | |
ba1bf481 | 7316 | full = space_info->full; |
17d217fe | 7317 | |
ba1bf481 JB |
7318 | /* |
7319 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7320 | * relocate it unless we're able to allocate a new chunk below. |
7321 | * | |
7322 | * Otherwise, we need to make sure we have room in the space to handle | |
7323 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7324 | */ |
7ce618db | 7325 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7326 | (space_info->bytes_used + space_info->bytes_reserved + |
7327 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7328 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7329 | spin_unlock(&space_info->lock); |
7330 | goto out; | |
17d217fe | 7331 | } |
ba1bf481 | 7332 | spin_unlock(&space_info->lock); |
ea8c2819 | 7333 | |
ba1bf481 JB |
7334 | /* |
7335 | * ok we don't have enough space, but maybe we have free space on our | |
7336 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7337 | * alloc devices and guess if we have enough space. if this block |
7338 | * group is going to be restriped, run checks against the target | |
7339 | * profile instead of the current one. | |
ba1bf481 JB |
7340 | */ |
7341 | ret = -1; | |
ea8c2819 | 7342 | |
cdcb725c | 7343 | /* |
7344 | * index: | |
7345 | * 0: raid10 | |
7346 | * 1: raid1 | |
7347 | * 2: dup | |
7348 | * 3: raid0 | |
7349 | * 4: single | |
7350 | */ | |
4a5e98f5 ID |
7351 | target = get_restripe_target(root->fs_info, block_group->flags); |
7352 | if (target) { | |
7353 | index = __get_block_group_index(extended_to_chunk(target)); | |
7354 | } else { | |
7355 | /* | |
7356 | * this is just a balance, so if we were marked as full | |
7357 | * we know there is no space for a new chunk | |
7358 | */ | |
7359 | if (full) | |
7360 | goto out; | |
7361 | ||
7362 | index = get_block_group_index(block_group); | |
7363 | } | |
7364 | ||
cdcb725c | 7365 | if (index == 0) { |
7366 | dev_min = 4; | |
6719db6a JB |
7367 | /* Divide by 2 */ |
7368 | min_free >>= 1; | |
cdcb725c | 7369 | } else if (index == 1) { |
7370 | dev_min = 2; | |
7371 | } else if (index == 2) { | |
6719db6a JB |
7372 | /* Multiply by 2 */ |
7373 | min_free <<= 1; | |
cdcb725c | 7374 | } else if (index == 3) { |
7375 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7376 | do_div(min_free, dev_min); |
cdcb725c | 7377 | } |
7378 | ||
ba1bf481 JB |
7379 | mutex_lock(&root->fs_info->chunk_mutex); |
7380 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7381 | u64 dev_offset; |
56bec294 | 7382 | |
ba1bf481 JB |
7383 | /* |
7384 | * check to make sure we can actually find a chunk with enough | |
7385 | * space to fit our block group in. | |
7386 | */ | |
7387 | if (device->total_bytes > device->bytes_used + min_free) { | |
125ccb0a | 7388 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7389 | &dev_offset, NULL); |
ba1bf481 | 7390 | if (!ret) |
cdcb725c | 7391 | dev_nr++; |
7392 | ||
7393 | if (dev_nr >= dev_min) | |
73e48b27 | 7394 | break; |
cdcb725c | 7395 | |
ba1bf481 | 7396 | ret = -1; |
725c8463 | 7397 | } |
edbd8d4e | 7398 | } |
ba1bf481 | 7399 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7400 | out: |
ba1bf481 | 7401 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7402 | return ret; |
7403 | } | |
7404 | ||
b2950863 CH |
7405 | static int find_first_block_group(struct btrfs_root *root, |
7406 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7407 | { |
925baedd | 7408 | int ret = 0; |
0b86a832 CM |
7409 | struct btrfs_key found_key; |
7410 | struct extent_buffer *leaf; | |
7411 | int slot; | |
edbd8d4e | 7412 | |
0b86a832 CM |
7413 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7414 | if (ret < 0) | |
925baedd CM |
7415 | goto out; |
7416 | ||
d397712b | 7417 | while (1) { |
0b86a832 | 7418 | slot = path->slots[0]; |
edbd8d4e | 7419 | leaf = path->nodes[0]; |
0b86a832 CM |
7420 | if (slot >= btrfs_header_nritems(leaf)) { |
7421 | ret = btrfs_next_leaf(root, path); | |
7422 | if (ret == 0) | |
7423 | continue; | |
7424 | if (ret < 0) | |
925baedd | 7425 | goto out; |
0b86a832 | 7426 | break; |
edbd8d4e | 7427 | } |
0b86a832 | 7428 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7429 | |
0b86a832 | 7430 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7431 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7432 | ret = 0; | |
7433 | goto out; | |
7434 | } | |
0b86a832 | 7435 | path->slots[0]++; |
edbd8d4e | 7436 | } |
925baedd | 7437 | out: |
0b86a832 | 7438 | return ret; |
edbd8d4e CM |
7439 | } |
7440 | ||
0af3d00b JB |
7441 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7442 | { | |
7443 | struct btrfs_block_group_cache *block_group; | |
7444 | u64 last = 0; | |
7445 | ||
7446 | while (1) { | |
7447 | struct inode *inode; | |
7448 | ||
7449 | block_group = btrfs_lookup_first_block_group(info, last); | |
7450 | while (block_group) { | |
7451 | spin_lock(&block_group->lock); | |
7452 | if (block_group->iref) | |
7453 | break; | |
7454 | spin_unlock(&block_group->lock); | |
7455 | block_group = next_block_group(info->tree_root, | |
7456 | block_group); | |
7457 | } | |
7458 | if (!block_group) { | |
7459 | if (last == 0) | |
7460 | break; | |
7461 | last = 0; | |
7462 | continue; | |
7463 | } | |
7464 | ||
7465 | inode = block_group->inode; | |
7466 | block_group->iref = 0; | |
7467 | block_group->inode = NULL; | |
7468 | spin_unlock(&block_group->lock); | |
7469 | iput(inode); | |
7470 | last = block_group->key.objectid + block_group->key.offset; | |
7471 | btrfs_put_block_group(block_group); | |
7472 | } | |
7473 | } | |
7474 | ||
1a40e23b ZY |
7475 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7476 | { | |
7477 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7478 | struct btrfs_space_info *space_info; |
11833d66 | 7479 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7480 | struct rb_node *n; |
7481 | ||
11833d66 YZ |
7482 | down_write(&info->extent_commit_sem); |
7483 | while (!list_empty(&info->caching_block_groups)) { | |
7484 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7485 | struct btrfs_caching_control, list); | |
7486 | list_del(&caching_ctl->list); | |
7487 | put_caching_control(caching_ctl); | |
7488 | } | |
7489 | up_write(&info->extent_commit_sem); | |
7490 | ||
1a40e23b ZY |
7491 | spin_lock(&info->block_group_cache_lock); |
7492 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7493 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7494 | cache_node); | |
1a40e23b ZY |
7495 | rb_erase(&block_group->cache_node, |
7496 | &info->block_group_cache_tree); | |
d899e052 YZ |
7497 | spin_unlock(&info->block_group_cache_lock); |
7498 | ||
80eb234a | 7499 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7500 | list_del(&block_group->list); |
80eb234a | 7501 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7502 | |
817d52f8 | 7503 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7504 | wait_block_group_cache_done(block_group); |
817d52f8 | 7505 | |
3c14874a JB |
7506 | /* |
7507 | * We haven't cached this block group, which means we could | |
7508 | * possibly have excluded extents on this block group. | |
7509 | */ | |
7510 | if (block_group->cached == BTRFS_CACHE_NO) | |
7511 | free_excluded_extents(info->extent_root, block_group); | |
7512 | ||
817d52f8 | 7513 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7514 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7515 | |
7516 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7517 | } |
7518 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7519 | |
7520 | /* now that all the block groups are freed, go through and | |
7521 | * free all the space_info structs. This is only called during | |
7522 | * the final stages of unmount, and so we know nobody is | |
7523 | * using them. We call synchronize_rcu() once before we start, | |
7524 | * just to be on the safe side. | |
7525 | */ | |
7526 | synchronize_rcu(); | |
7527 | ||
8929ecfa YZ |
7528 | release_global_block_rsv(info); |
7529 | ||
4184ea7f CM |
7530 | while(!list_empty(&info->space_info)) { |
7531 | space_info = list_entry(info->space_info.next, | |
7532 | struct btrfs_space_info, | |
7533 | list); | |
f0486c68 | 7534 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7535 | space_info->bytes_reserved > 0 || |
7536 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7537 | WARN_ON(1); |
7538 | dump_space_info(space_info, 0, 0); | |
7539 | } | |
4184ea7f CM |
7540 | list_del(&space_info->list); |
7541 | kfree(space_info); | |
7542 | } | |
1a40e23b ZY |
7543 | return 0; |
7544 | } | |
7545 | ||
b742bb82 YZ |
7546 | static void __link_block_group(struct btrfs_space_info *space_info, |
7547 | struct btrfs_block_group_cache *cache) | |
7548 | { | |
7549 | int index = get_block_group_index(cache); | |
7550 | ||
7551 | down_write(&space_info->groups_sem); | |
7552 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7553 | up_write(&space_info->groups_sem); | |
7554 | } | |
7555 | ||
9078a3e1 CM |
7556 | int btrfs_read_block_groups(struct btrfs_root *root) |
7557 | { | |
7558 | struct btrfs_path *path; | |
7559 | int ret; | |
9078a3e1 | 7560 | struct btrfs_block_group_cache *cache; |
be744175 | 7561 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7562 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7563 | struct btrfs_key key; |
7564 | struct btrfs_key found_key; | |
5f39d397 | 7565 | struct extent_buffer *leaf; |
0af3d00b JB |
7566 | int need_clear = 0; |
7567 | u64 cache_gen; | |
96b5179d | 7568 | |
be744175 | 7569 | root = info->extent_root; |
9078a3e1 | 7570 | key.objectid = 0; |
0b86a832 | 7571 | key.offset = 0; |
9078a3e1 | 7572 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7573 | path = btrfs_alloc_path(); |
7574 | if (!path) | |
7575 | return -ENOMEM; | |
026fd317 | 7576 | path->reada = 1; |
9078a3e1 | 7577 | |
6c41761f | 7578 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7579 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7580 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7581 | need_clear = 1; |
88c2ba3b JB |
7582 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7583 | need_clear = 1; | |
0af3d00b | 7584 | |
d397712b | 7585 | while (1) { |
0b86a832 | 7586 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7587 | if (ret > 0) |
7588 | break; | |
0b86a832 CM |
7589 | if (ret != 0) |
7590 | goto error; | |
5f39d397 CM |
7591 | leaf = path->nodes[0]; |
7592 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7593 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7594 | if (!cache) { |
0b86a832 | 7595 | ret = -ENOMEM; |
f0486c68 | 7596 | goto error; |
9078a3e1 | 7597 | } |
34d52cb6 LZ |
7598 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7599 | GFP_NOFS); | |
7600 | if (!cache->free_space_ctl) { | |
7601 | kfree(cache); | |
7602 | ret = -ENOMEM; | |
7603 | goto error; | |
7604 | } | |
3e1ad54f | 7605 | |
d2fb3437 | 7606 | atomic_set(&cache->count, 1); |
c286ac48 | 7607 | spin_lock_init(&cache->lock); |
817d52f8 | 7608 | cache->fs_info = info; |
0f9dd46c | 7609 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7610 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7611 | |
0af3d00b JB |
7612 | if (need_clear) |
7613 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
7614 | ||
5f39d397 CM |
7615 | read_extent_buffer(leaf, &cache->item, |
7616 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7617 | sizeof(cache->item)); | |
9078a3e1 | 7618 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7619 | |
9078a3e1 | 7620 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7621 | btrfs_release_path(path); |
0b86a832 | 7622 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7623 | cache->sectorsize = root->sectorsize; |
7624 | ||
34d52cb6 LZ |
7625 | btrfs_init_free_space_ctl(cache); |
7626 | ||
3c14874a JB |
7627 | /* |
7628 | * We need to exclude the super stripes now so that the space | |
7629 | * info has super bytes accounted for, otherwise we'll think | |
7630 | * we have more space than we actually do. | |
7631 | */ | |
7632 | exclude_super_stripes(root, cache); | |
7633 | ||
817d52f8 JB |
7634 | /* |
7635 | * check for two cases, either we are full, and therefore | |
7636 | * don't need to bother with the caching work since we won't | |
7637 | * find any space, or we are empty, and we can just add all | |
7638 | * the space in and be done with it. This saves us _alot_ of | |
7639 | * time, particularly in the full case. | |
7640 | */ | |
7641 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7642 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7643 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7644 | free_excluded_extents(root, cache); |
817d52f8 | 7645 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7646 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7647 | cache->cached = BTRFS_CACHE_FINISHED; |
7648 | add_new_free_space(cache, root->fs_info, | |
7649 | found_key.objectid, | |
7650 | found_key.objectid + | |
7651 | found_key.offset); | |
11833d66 | 7652 | free_excluded_extents(root, cache); |
817d52f8 | 7653 | } |
96b5179d | 7654 | |
6324fbf3 CM |
7655 | ret = update_space_info(info, cache->flags, found_key.offset, |
7656 | btrfs_block_group_used(&cache->item), | |
7657 | &space_info); | |
79787eaa | 7658 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 7659 | cache->space_info = space_info; |
1b2da372 | 7660 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7661 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7662 | spin_unlock(&cache->space_info->lock); |
7663 | ||
b742bb82 | 7664 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7665 | |
7666 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 7667 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
7668 | |
7669 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7670 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7671 | set_block_group_ro(cache, 1); |
9078a3e1 | 7672 | } |
b742bb82 YZ |
7673 | |
7674 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7675 | if (!(get_alloc_profile(root, space_info->flags) & | |
7676 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7677 | BTRFS_BLOCK_GROUP_RAID1 | | |
7678 | BTRFS_BLOCK_GROUP_DUP))) | |
7679 | continue; | |
7680 | /* | |
7681 | * avoid allocating from un-mirrored block group if there are | |
7682 | * mirrored block groups. | |
7683 | */ | |
7684 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7685 | set_block_group_ro(cache, 1); |
b742bb82 | 7686 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7687 | set_block_group_ro(cache, 1); |
9078a3e1 | 7688 | } |
f0486c68 YZ |
7689 | |
7690 | init_global_block_rsv(info); | |
0b86a832 CM |
7691 | ret = 0; |
7692 | error: | |
9078a3e1 | 7693 | btrfs_free_path(path); |
0b86a832 | 7694 | return ret; |
9078a3e1 | 7695 | } |
6324fbf3 CM |
7696 | |
7697 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7698 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7699 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7700 | u64 size) |
7701 | { | |
7702 | int ret; | |
6324fbf3 CM |
7703 | struct btrfs_root *extent_root; |
7704 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7705 | |
7706 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7707 | |
12fcfd22 | 7708 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7709 | |
8f18cf13 | 7710 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7711 | if (!cache) |
7712 | return -ENOMEM; | |
34d52cb6 LZ |
7713 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7714 | GFP_NOFS); | |
7715 | if (!cache->free_space_ctl) { | |
7716 | kfree(cache); | |
7717 | return -ENOMEM; | |
7718 | } | |
0f9dd46c | 7719 | |
e17cade2 | 7720 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7721 | cache->key.offset = size; |
d2fb3437 | 7722 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7723 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7724 | cache->fs_info = root->fs_info; |
96303081 | 7725 | |
d2fb3437 | 7726 | atomic_set(&cache->count, 1); |
c286ac48 | 7727 | spin_lock_init(&cache->lock); |
0f9dd46c | 7728 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7729 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7730 | |
34d52cb6 LZ |
7731 | btrfs_init_free_space_ctl(cache); |
7732 | ||
6324fbf3 | 7733 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7734 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7735 | cache->flags = type; | |
7736 | btrfs_set_block_group_flags(&cache->item, type); | |
7737 | ||
11833d66 | 7738 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7739 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7740 | exclude_super_stripes(root, cache); |
96303081 | 7741 | |
817d52f8 JB |
7742 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7743 | chunk_offset + size); | |
7744 | ||
11833d66 YZ |
7745 | free_excluded_extents(root, cache); |
7746 | ||
6324fbf3 CM |
7747 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7748 | &cache->space_info); | |
79787eaa | 7749 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 7750 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7751 | |
7752 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7753 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7754 | spin_unlock(&cache->space_info->lock); |
7755 | ||
b742bb82 | 7756 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7757 | |
0f9dd46c | 7758 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 7759 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 7760 | |
6324fbf3 CM |
7761 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7762 | sizeof(cache->item)); | |
79787eaa JM |
7763 | if (ret) { |
7764 | btrfs_abort_transaction(trans, extent_root, ret); | |
7765 | return ret; | |
7766 | } | |
6324fbf3 | 7767 | |
d18a2c44 | 7768 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7769 | |
6324fbf3 CM |
7770 | return 0; |
7771 | } | |
1a40e23b | 7772 | |
10ea00f5 ID |
7773 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7774 | { | |
899c81ea ID |
7775 | u64 extra_flags = chunk_to_extended(flags) & |
7776 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 ID |
7777 | |
7778 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7779 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7780 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7781 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7782 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7783 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7784 | } | |
7785 | ||
1a40e23b ZY |
7786 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7787 | struct btrfs_root *root, u64 group_start) | |
7788 | { | |
7789 | struct btrfs_path *path; | |
7790 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7791 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7792 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7793 | struct btrfs_key key; |
0af3d00b | 7794 | struct inode *inode; |
1a40e23b | 7795 | int ret; |
10ea00f5 | 7796 | int index; |
89a55897 | 7797 | int factor; |
1a40e23b | 7798 | |
1a40e23b ZY |
7799 | root = root->fs_info->extent_root; |
7800 | ||
7801 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7802 | BUG_ON(!block_group); | |
c146afad | 7803 | BUG_ON(!block_group->ro); |
1a40e23b | 7804 | |
9f7c43c9 | 7805 | /* |
7806 | * Free the reserved super bytes from this block group before | |
7807 | * remove it. | |
7808 | */ | |
7809 | free_excluded_extents(root, block_group); | |
7810 | ||
1a40e23b | 7811 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7812 | index = get_block_group_index(block_group); |
89a55897 JB |
7813 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7814 | BTRFS_BLOCK_GROUP_RAID1 | | |
7815 | BTRFS_BLOCK_GROUP_RAID10)) | |
7816 | factor = 2; | |
7817 | else | |
7818 | factor = 1; | |
1a40e23b | 7819 | |
44fb5511 CM |
7820 | /* make sure this block group isn't part of an allocation cluster */ |
7821 | cluster = &root->fs_info->data_alloc_cluster; | |
7822 | spin_lock(&cluster->refill_lock); | |
7823 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7824 | spin_unlock(&cluster->refill_lock); | |
7825 | ||
7826 | /* | |
7827 | * make sure this block group isn't part of a metadata | |
7828 | * allocation cluster | |
7829 | */ | |
7830 | cluster = &root->fs_info->meta_alloc_cluster; | |
7831 | spin_lock(&cluster->refill_lock); | |
7832 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7833 | spin_unlock(&cluster->refill_lock); | |
7834 | ||
1a40e23b | 7835 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7836 | if (!path) { |
7837 | ret = -ENOMEM; | |
7838 | goto out; | |
7839 | } | |
1a40e23b | 7840 | |
10b2f34d | 7841 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7842 | if (!IS_ERR(inode)) { |
b532402e | 7843 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
7844 | if (ret) { |
7845 | btrfs_add_delayed_iput(inode); | |
7846 | goto out; | |
7847 | } | |
0af3d00b JB |
7848 | clear_nlink(inode); |
7849 | /* One for the block groups ref */ | |
7850 | spin_lock(&block_group->lock); | |
7851 | if (block_group->iref) { | |
7852 | block_group->iref = 0; | |
7853 | block_group->inode = NULL; | |
7854 | spin_unlock(&block_group->lock); | |
7855 | iput(inode); | |
7856 | } else { | |
7857 | spin_unlock(&block_group->lock); | |
7858 | } | |
7859 | /* One for our lookup ref */ | |
455757c3 | 7860 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7861 | } |
7862 | ||
7863 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7864 | key.offset = block_group->key.objectid; | |
7865 | key.type = 0; | |
7866 | ||
7867 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7868 | if (ret < 0) | |
7869 | goto out; | |
7870 | if (ret > 0) | |
b3b4aa74 | 7871 | btrfs_release_path(path); |
0af3d00b JB |
7872 | if (ret == 0) { |
7873 | ret = btrfs_del_item(trans, tree_root, path); | |
7874 | if (ret) | |
7875 | goto out; | |
b3b4aa74 | 7876 | btrfs_release_path(path); |
0af3d00b JB |
7877 | } |
7878 | ||
3dfdb934 | 7879 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7880 | rb_erase(&block_group->cache_node, |
7881 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7882 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7883 | |
80eb234a | 7884 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7885 | /* |
7886 | * we must use list_del_init so people can check to see if they | |
7887 | * are still on the list after taking the semaphore | |
7888 | */ | |
7889 | list_del_init(&block_group->list); | |
10ea00f5 ID |
7890 | if (list_empty(&block_group->space_info->block_groups[index])) |
7891 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 7892 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7893 | |
817d52f8 | 7894 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7895 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7896 | |
7897 | btrfs_remove_free_space_cache(block_group); | |
7898 | ||
c146afad YZ |
7899 | spin_lock(&block_group->space_info->lock); |
7900 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7901 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7902 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7903 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7904 | |
0af3d00b JB |
7905 | memcpy(&key, &block_group->key, sizeof(key)); |
7906 | ||
283bb197 | 7907 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7908 | |
fa9c0d79 CM |
7909 | btrfs_put_block_group(block_group); |
7910 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7911 | |
7912 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7913 | if (ret > 0) | |
7914 | ret = -EIO; | |
7915 | if (ret < 0) | |
7916 | goto out; | |
7917 | ||
7918 | ret = btrfs_del_item(trans, root, path); | |
7919 | out: | |
7920 | btrfs_free_path(path); | |
7921 | return ret; | |
7922 | } | |
acce952b | 7923 | |
c59021f8 | 7924 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7925 | { | |
7926 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7927 | struct btrfs_super_block *disk_super; |
7928 | u64 features; | |
7929 | u64 flags; | |
7930 | int mixed = 0; | |
c59021f8 | 7931 | int ret; |
7932 | ||
6c41761f | 7933 | disk_super = fs_info->super_copy; |
1aba86d6 | 7934 | if (!btrfs_super_root(disk_super)) |
7935 | return 1; | |
c59021f8 | 7936 | |
1aba86d6 | 7937 | features = btrfs_super_incompat_flags(disk_super); |
7938 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7939 | mixed = 1; | |
c59021f8 | 7940 | |
1aba86d6 | 7941 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7942 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7943 | if (ret) |
1aba86d6 | 7944 | goto out; |
c59021f8 | 7945 | |
1aba86d6 | 7946 | if (mixed) { |
7947 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7948 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7949 | } else { | |
7950 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7951 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7952 | if (ret) | |
7953 | goto out; | |
7954 | ||
7955 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7956 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7957 | } | |
7958 | out: | |
c59021f8 | 7959 | return ret; |
7960 | } | |
7961 | ||
acce952b | 7962 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7963 | { | |
7964 | return unpin_extent_range(root, start, end); | |
7965 | } | |
7966 | ||
7967 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7968 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7969 | { |
5378e607 | 7970 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7971 | } |
f7039b1d LD |
7972 | |
7973 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7974 | { | |
7975 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7976 | struct btrfs_block_group_cache *cache = NULL; | |
7977 | u64 group_trimmed; | |
7978 | u64 start; | |
7979 | u64 end; | |
7980 | u64 trimmed = 0; | |
2cac13e4 | 7981 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
7982 | int ret = 0; |
7983 | ||
2cac13e4 LB |
7984 | /* |
7985 | * try to trim all FS space, our block group may start from non-zero. | |
7986 | */ | |
7987 | if (range->len == total_bytes) | |
7988 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
7989 | else | |
7990 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
7991 | |
7992 | while (cache) { | |
7993 | if (cache->key.objectid >= (range->start + range->len)) { | |
7994 | btrfs_put_block_group(cache); | |
7995 | break; | |
7996 | } | |
7997 | ||
7998 | start = max(range->start, cache->key.objectid); | |
7999 | end = min(range->start + range->len, | |
8000 | cache->key.objectid + cache->key.offset); | |
8001 | ||
8002 | if (end - start >= range->minlen) { | |
8003 | if (!block_group_cache_done(cache)) { | |
8004 | ret = cache_block_group(cache, NULL, root, 0); | |
8005 | if (!ret) | |
8006 | wait_block_group_cache_done(cache); | |
8007 | } | |
8008 | ret = btrfs_trim_block_group(cache, | |
8009 | &group_trimmed, | |
8010 | start, | |
8011 | end, | |
8012 | range->minlen); | |
8013 | ||
8014 | trimmed += group_trimmed; | |
8015 | if (ret) { | |
8016 | btrfs_put_block_group(cache); | |
8017 | break; | |
8018 | } | |
8019 | } | |
8020 | ||
8021 | cache = next_block_group(fs_info->tree_root, cache); | |
8022 | } | |
8023 | ||
8024 | range->len = trimmed; | |
8025 | return ret; | |
8026 | } |