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; |
097b8a7c | 2220 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2221 | int ret; |
c3e69d58 | 2222 | int count = 0; |
56bec294 | 2223 | int must_insert_reserved = 0; |
56bec294 CM |
2224 | |
2225 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2226 | while (1) { |
2227 | if (!locked_ref) { | |
c3e69d58 CM |
2228 | /* pick a new head ref from the cluster list */ |
2229 | if (list_empty(cluster)) | |
56bec294 | 2230 | break; |
56bec294 | 2231 | |
c3e69d58 CM |
2232 | locked_ref = list_entry(cluster->next, |
2233 | struct btrfs_delayed_ref_head, cluster); | |
2234 | ||
2235 | /* grab the lock that says we are going to process | |
2236 | * all the refs for this head */ | |
2237 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2238 | ||
2239 | /* | |
2240 | * we may have dropped the spin lock to get the head | |
2241 | * mutex lock, and that might have given someone else | |
2242 | * time to free the head. If that's true, it has been | |
2243 | * removed from our list and we can move on. | |
2244 | */ | |
2245 | if (ret == -EAGAIN) { | |
2246 | locked_ref = NULL; | |
2247 | count++; | |
2248 | continue; | |
56bec294 CM |
2249 | } |
2250 | } | |
a28ec197 | 2251 | |
d1270cd9 AJ |
2252 | /* |
2253 | * locked_ref is the head node, so we have to go one | |
2254 | * node back for any delayed ref updates | |
2255 | */ | |
2256 | ref = select_delayed_ref(locked_ref); | |
2257 | ||
2258 | if (ref && ref->seq && | |
097b8a7c | 2259 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2260 | /* |
2261 | * there are still refs with lower seq numbers in the | |
2262 | * process of being added. Don't run this ref yet. | |
2263 | */ | |
2264 | list_del_init(&locked_ref->cluster); | |
2265 | mutex_unlock(&locked_ref->mutex); | |
2266 | locked_ref = NULL; | |
2267 | delayed_refs->num_heads_ready++; | |
2268 | spin_unlock(&delayed_refs->lock); | |
2269 | cond_resched(); | |
2270 | spin_lock(&delayed_refs->lock); | |
2271 | continue; | |
2272 | } | |
2273 | ||
56bec294 CM |
2274 | /* |
2275 | * record the must insert reserved flag before we | |
2276 | * drop the spin lock. | |
2277 | */ | |
2278 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2279 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2280 | |
5d4f98a2 YZ |
2281 | extent_op = locked_ref->extent_op; |
2282 | locked_ref->extent_op = NULL; | |
2283 | ||
56bec294 CM |
2284 | if (!ref) { |
2285 | /* All delayed refs have been processed, Go ahead | |
2286 | * and send the head node to run_one_delayed_ref, | |
2287 | * so that any accounting fixes can happen | |
2288 | */ | |
2289 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2290 | |
2291 | if (extent_op && must_insert_reserved) { | |
2292 | kfree(extent_op); | |
2293 | extent_op = NULL; | |
2294 | } | |
2295 | ||
2296 | if (extent_op) { | |
2297 | spin_unlock(&delayed_refs->lock); | |
2298 | ||
2299 | ret = run_delayed_extent_op(trans, root, | |
2300 | ref, extent_op); | |
5d4f98a2 YZ |
2301 | kfree(extent_op); |
2302 | ||
79787eaa JM |
2303 | if (ret) { |
2304 | printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret); | |
253beebd | 2305 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2306 | return ret; |
2307 | } | |
2308 | ||
203bf287 | 2309 | goto next; |
5d4f98a2 YZ |
2310 | } |
2311 | ||
c3e69d58 | 2312 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2313 | locked_ref = NULL; |
2314 | } | |
02217ed2 | 2315 | |
56bec294 CM |
2316 | ref->in_tree = 0; |
2317 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2318 | delayed_refs->num_entries--; | |
a168650c JS |
2319 | /* |
2320 | * we modified num_entries, but as we're currently running | |
2321 | * delayed refs, skip | |
2322 | * wake_up(&delayed_refs->seq_wait); | |
2323 | * here. | |
2324 | */ | |
56bec294 | 2325 | spin_unlock(&delayed_refs->lock); |
925baedd | 2326 | |
5d4f98a2 | 2327 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2328 | must_insert_reserved); |
eb099670 | 2329 | |
5d4f98a2 YZ |
2330 | btrfs_put_delayed_ref(ref); |
2331 | kfree(extent_op); | |
c3e69d58 | 2332 | count++; |
79787eaa JM |
2333 | |
2334 | if (ret) { | |
2335 | printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2336 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2337 | return ret; |
2338 | } | |
2339 | ||
203bf287 | 2340 | next: |
097b8a7c | 2341 | do_chunk_alloc(trans, fs_info->extent_root, |
203bf287 CM |
2342 | 2 * 1024 * 1024, |
2343 | btrfs_get_alloc_profile(root, 0), | |
2344 | CHUNK_ALLOC_NO_FORCE); | |
c3e69d58 CM |
2345 | cond_resched(); |
2346 | spin_lock(&delayed_refs->lock); | |
2347 | } | |
2348 | return count; | |
2349 | } | |
2350 | ||
097b8a7c JS |
2351 | static void wait_for_more_refs(struct btrfs_fs_info *fs_info, |
2352 | struct btrfs_delayed_ref_root *delayed_refs, | |
8ca78f3e JS |
2353 | unsigned long num_refs, |
2354 | struct list_head *first_seq) | |
a168650c | 2355 | { |
a168650c JS |
2356 | spin_unlock(&delayed_refs->lock); |
2357 | pr_debug("waiting for more refs (num %ld, first %p)\n", | |
2358 | num_refs, first_seq); | |
097b8a7c | 2359 | wait_event(fs_info->tree_mod_seq_wait, |
a168650c | 2360 | num_refs != delayed_refs->num_entries || |
097b8a7c | 2361 | fs_info->tree_mod_seq_list.next != first_seq); |
a168650c | 2362 | pr_debug("done waiting for more refs (num %ld, first %p)\n", |
097b8a7c | 2363 | delayed_refs->num_entries, fs_info->tree_mod_seq_list.next); |
a168650c JS |
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; | |
8ca78f3e | 2384 | struct list_head *first_seq = NULL; |
c3e69d58 | 2385 | int ret; |
a168650c | 2386 | u64 delayed_start; |
c3e69d58 CM |
2387 | int run_all = count == (unsigned long)-1; |
2388 | int run_most = 0; | |
a168650c JS |
2389 | unsigned long num_refs = 0; |
2390 | int consider_waiting; | |
c3e69d58 | 2391 | |
79787eaa JM |
2392 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2393 | if (trans->aborted) | |
2394 | return 0; | |
2395 | ||
c3e69d58 CM |
2396 | if (root == root->fs_info->extent_root) |
2397 | root = root->fs_info->tree_root; | |
2398 | ||
203bf287 CM |
2399 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2400 | 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0), | |
2401 | CHUNK_ALLOC_NO_FORCE); | |
2402 | ||
c3e69d58 CM |
2403 | delayed_refs = &trans->transaction->delayed_refs; |
2404 | INIT_LIST_HEAD(&cluster); | |
2405 | again: | |
a168650c | 2406 | consider_waiting = 0; |
c3e69d58 | 2407 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2408 | |
c3e69d58 CM |
2409 | if (count == 0) { |
2410 | count = delayed_refs->num_entries * 2; | |
2411 | run_most = 1; | |
2412 | } | |
2413 | while (1) { | |
2414 | if (!(run_all || run_most) && | |
2415 | delayed_refs->num_heads_ready < 64) | |
2416 | break; | |
eb099670 | 2417 | |
56bec294 | 2418 | /* |
c3e69d58 CM |
2419 | * go find something we can process in the rbtree. We start at |
2420 | * the beginning of the tree, and then build a cluster | |
2421 | * of refs to process starting at the first one we are able to | |
2422 | * lock | |
56bec294 | 2423 | */ |
a168650c | 2424 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2425 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2426 | delayed_refs->run_delayed_start); | |
2427 | if (ret) | |
56bec294 CM |
2428 | break; |
2429 | ||
a168650c JS |
2430 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2431 | if (consider_waiting == 0) { | |
2432 | /* | |
2433 | * btrfs_find_ref_cluster looped. let's do one | |
2434 | * more cycle. if we don't run any delayed ref | |
2435 | * during that cycle (because we can't because | |
2436 | * all of them are blocked) and if the number of | |
2437 | * refs doesn't change, we avoid busy waiting. | |
2438 | */ | |
2439 | consider_waiting = 1; | |
2440 | num_refs = delayed_refs->num_entries; | |
8ca78f3e | 2441 | first_seq = root->fs_info->tree_mod_seq_list.next; |
a168650c | 2442 | } else { |
097b8a7c | 2443 | wait_for_more_refs(root->fs_info, delayed_refs, |
8ca78f3e | 2444 | num_refs, first_seq); |
a168650c JS |
2445 | /* |
2446 | * after waiting, things have changed. we | |
2447 | * dropped the lock and someone else might have | |
2448 | * run some refs, built new clusters and so on. | |
2449 | * therefore, we restart staleness detection. | |
2450 | */ | |
2451 | consider_waiting = 0; | |
2452 | } | |
2453 | } | |
2454 | ||
c3e69d58 | 2455 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa JM |
2456 | if (ret < 0) { |
2457 | spin_unlock(&delayed_refs->lock); | |
2458 | btrfs_abort_transaction(trans, root, ret); | |
2459 | return ret; | |
2460 | } | |
c3e69d58 CM |
2461 | |
2462 | count -= min_t(unsigned long, ret, count); | |
2463 | ||
2464 | if (count == 0) | |
2465 | break; | |
a168650c JS |
2466 | |
2467 | if (ret || delayed_refs->run_delayed_start == 0) { | |
2468 | /* refs were run, let's reset staleness detection */ | |
2469 | consider_waiting = 0; | |
2470 | } | |
eb099670 | 2471 | } |
c3e69d58 | 2472 | |
56bec294 | 2473 | if (run_all) { |
56bec294 | 2474 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2475 | if (!node) |
56bec294 | 2476 | goto out; |
c3e69d58 | 2477 | count = (unsigned long)-1; |
e9d0b13b | 2478 | |
56bec294 CM |
2479 | while (node) { |
2480 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2481 | rb_node); | |
2482 | if (btrfs_delayed_ref_is_head(ref)) { | |
2483 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2484 | |
56bec294 CM |
2485 | head = btrfs_delayed_node_to_head(ref); |
2486 | atomic_inc(&ref->refs); | |
2487 | ||
2488 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2489 | /* |
2490 | * Mutex was contended, block until it's | |
2491 | * released and try again | |
2492 | */ | |
56bec294 CM |
2493 | mutex_lock(&head->mutex); |
2494 | mutex_unlock(&head->mutex); | |
2495 | ||
2496 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2497 | cond_resched(); |
56bec294 CM |
2498 | goto again; |
2499 | } | |
2500 | node = rb_next(node); | |
2501 | } | |
2502 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2503 | schedule_timeout(1); |
2504 | goto again; | |
5f39d397 | 2505 | } |
54aa1f4d | 2506 | out: |
c3e69d58 | 2507 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2508 | return 0; |
2509 | } | |
2510 | ||
5d4f98a2 YZ |
2511 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2512 | struct btrfs_root *root, | |
2513 | u64 bytenr, u64 num_bytes, u64 flags, | |
2514 | int is_data) | |
2515 | { | |
2516 | struct btrfs_delayed_extent_op *extent_op; | |
2517 | int ret; | |
2518 | ||
2519 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2520 | if (!extent_op) | |
2521 | return -ENOMEM; | |
2522 | ||
2523 | extent_op->flags_to_set = flags; | |
2524 | extent_op->update_flags = 1; | |
2525 | extent_op->update_key = 0; | |
2526 | extent_op->is_data = is_data ? 1 : 0; | |
2527 | ||
66d7e7f0 AJ |
2528 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2529 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2530 | if (ret) |
2531 | kfree(extent_op); | |
2532 | return ret; | |
2533 | } | |
2534 | ||
2535 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2536 | struct btrfs_root *root, | |
2537 | struct btrfs_path *path, | |
2538 | u64 objectid, u64 offset, u64 bytenr) | |
2539 | { | |
2540 | struct btrfs_delayed_ref_head *head; | |
2541 | struct btrfs_delayed_ref_node *ref; | |
2542 | struct btrfs_delayed_data_ref *data_ref; | |
2543 | struct btrfs_delayed_ref_root *delayed_refs; | |
2544 | struct rb_node *node; | |
2545 | int ret = 0; | |
2546 | ||
2547 | ret = -ENOENT; | |
2548 | delayed_refs = &trans->transaction->delayed_refs; | |
2549 | spin_lock(&delayed_refs->lock); | |
2550 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2551 | if (!head) | |
2552 | goto out; | |
2553 | ||
2554 | if (!mutex_trylock(&head->mutex)) { | |
2555 | atomic_inc(&head->node.refs); | |
2556 | spin_unlock(&delayed_refs->lock); | |
2557 | ||
b3b4aa74 | 2558 | btrfs_release_path(path); |
5d4f98a2 | 2559 | |
8cc33e5c DS |
2560 | /* |
2561 | * Mutex was contended, block until it's released and let | |
2562 | * caller try again | |
2563 | */ | |
5d4f98a2 YZ |
2564 | mutex_lock(&head->mutex); |
2565 | mutex_unlock(&head->mutex); | |
2566 | btrfs_put_delayed_ref(&head->node); | |
2567 | return -EAGAIN; | |
2568 | } | |
2569 | ||
2570 | node = rb_prev(&head->node.rb_node); | |
2571 | if (!node) | |
2572 | goto out_unlock; | |
2573 | ||
2574 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2575 | ||
2576 | if (ref->bytenr != bytenr) | |
2577 | goto out_unlock; | |
2578 | ||
2579 | ret = 1; | |
2580 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2581 | goto out_unlock; | |
2582 | ||
2583 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2584 | ||
2585 | node = rb_prev(node); | |
2586 | if (node) { | |
2587 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2588 | if (ref->bytenr == bytenr) | |
2589 | goto out_unlock; | |
2590 | } | |
2591 | ||
2592 | if (data_ref->root != root->root_key.objectid || | |
2593 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2594 | goto out_unlock; | |
2595 | ||
2596 | ret = 0; | |
2597 | out_unlock: | |
2598 | mutex_unlock(&head->mutex); | |
2599 | out: | |
2600 | spin_unlock(&delayed_refs->lock); | |
2601 | return ret; | |
2602 | } | |
2603 | ||
2604 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2605 | struct btrfs_root *root, | |
2606 | struct btrfs_path *path, | |
2607 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2608 | { |
2609 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2610 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2611 | struct btrfs_extent_data_ref *ref; |
2612 | struct btrfs_extent_inline_ref *iref; | |
2613 | struct btrfs_extent_item *ei; | |
f321e491 | 2614 | struct btrfs_key key; |
5d4f98a2 | 2615 | u32 item_size; |
be20aa9d | 2616 | int ret; |
925baedd | 2617 | |
be20aa9d | 2618 | key.objectid = bytenr; |
31840ae1 | 2619 | key.offset = (u64)-1; |
f321e491 | 2620 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2621 | |
be20aa9d CM |
2622 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2623 | if (ret < 0) | |
2624 | goto out; | |
79787eaa | 2625 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2626 | |
2627 | ret = -ENOENT; | |
2628 | if (path->slots[0] == 0) | |
31840ae1 | 2629 | goto out; |
be20aa9d | 2630 | |
31840ae1 | 2631 | path->slots[0]--; |
f321e491 | 2632 | leaf = path->nodes[0]; |
5d4f98a2 | 2633 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2634 | |
5d4f98a2 | 2635 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2636 | goto out; |
f321e491 | 2637 | |
5d4f98a2 YZ |
2638 | ret = 1; |
2639 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2640 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2641 | if (item_size < sizeof(*ei)) { | |
2642 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2643 | goto out; | |
2644 | } | |
2645 | #endif | |
2646 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2647 | |
5d4f98a2 YZ |
2648 | if (item_size != sizeof(*ei) + |
2649 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2650 | goto out; | |
be20aa9d | 2651 | |
5d4f98a2 YZ |
2652 | if (btrfs_extent_generation(leaf, ei) <= |
2653 | btrfs_root_last_snapshot(&root->root_item)) | |
2654 | goto out; | |
2655 | ||
2656 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2657 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2658 | BTRFS_EXTENT_DATA_REF_KEY) | |
2659 | goto out; | |
2660 | ||
2661 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2662 | if (btrfs_extent_refs(leaf, ei) != | |
2663 | btrfs_extent_data_ref_count(leaf, ref) || | |
2664 | btrfs_extent_data_ref_root(leaf, ref) != | |
2665 | root->root_key.objectid || | |
2666 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2667 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2668 | goto out; | |
2669 | ||
2670 | ret = 0; | |
2671 | out: | |
2672 | return ret; | |
2673 | } | |
2674 | ||
2675 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2676 | struct btrfs_root *root, | |
2677 | u64 objectid, u64 offset, u64 bytenr) | |
2678 | { | |
2679 | struct btrfs_path *path; | |
2680 | int ret; | |
2681 | int ret2; | |
2682 | ||
2683 | path = btrfs_alloc_path(); | |
2684 | if (!path) | |
2685 | return -ENOENT; | |
2686 | ||
2687 | do { | |
2688 | ret = check_committed_ref(trans, root, path, objectid, | |
2689 | offset, bytenr); | |
2690 | if (ret && ret != -ENOENT) | |
f321e491 | 2691 | goto out; |
80ff3856 | 2692 | |
5d4f98a2 YZ |
2693 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2694 | offset, bytenr); | |
2695 | } while (ret2 == -EAGAIN); | |
2696 | ||
2697 | if (ret2 && ret2 != -ENOENT) { | |
2698 | ret = ret2; | |
2699 | goto out; | |
f321e491 | 2700 | } |
5d4f98a2 YZ |
2701 | |
2702 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2703 | ret = 0; | |
be20aa9d | 2704 | out: |
80ff3856 | 2705 | btrfs_free_path(path); |
f0486c68 YZ |
2706 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2707 | WARN_ON(ret > 0); | |
f321e491 | 2708 | return ret; |
be20aa9d | 2709 | } |
c5739bba | 2710 | |
5d4f98a2 | 2711 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2712 | struct btrfs_root *root, |
5d4f98a2 | 2713 | struct extent_buffer *buf, |
66d7e7f0 | 2714 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2715 | { |
2716 | u64 bytenr; | |
5d4f98a2 YZ |
2717 | u64 num_bytes; |
2718 | u64 parent; | |
31840ae1 | 2719 | u64 ref_root; |
31840ae1 | 2720 | u32 nritems; |
31840ae1 ZY |
2721 | struct btrfs_key key; |
2722 | struct btrfs_file_extent_item *fi; | |
2723 | int i; | |
2724 | int level; | |
2725 | int ret = 0; | |
31840ae1 | 2726 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2727 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2728 | |
2729 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2730 | nritems = btrfs_header_nritems(buf); |
2731 | level = btrfs_header_level(buf); | |
2732 | ||
5d4f98a2 YZ |
2733 | if (!root->ref_cows && level == 0) |
2734 | return 0; | |
31840ae1 | 2735 | |
5d4f98a2 YZ |
2736 | if (inc) |
2737 | process_func = btrfs_inc_extent_ref; | |
2738 | else | |
2739 | process_func = btrfs_free_extent; | |
31840ae1 | 2740 | |
5d4f98a2 YZ |
2741 | if (full_backref) |
2742 | parent = buf->start; | |
2743 | else | |
2744 | parent = 0; | |
2745 | ||
2746 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2747 | if (level == 0) { |
5d4f98a2 | 2748 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2749 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2750 | continue; | |
5d4f98a2 | 2751 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2752 | struct btrfs_file_extent_item); |
2753 | if (btrfs_file_extent_type(buf, fi) == | |
2754 | BTRFS_FILE_EXTENT_INLINE) | |
2755 | continue; | |
2756 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2757 | if (bytenr == 0) | |
2758 | continue; | |
5d4f98a2 YZ |
2759 | |
2760 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2761 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2762 | ret = process_func(trans, root, bytenr, num_bytes, | |
2763 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2764 | key.offset, for_cow); |
31840ae1 ZY |
2765 | if (ret) |
2766 | goto fail; | |
2767 | } else { | |
5d4f98a2 YZ |
2768 | bytenr = btrfs_node_blockptr(buf, i); |
2769 | num_bytes = btrfs_level_size(root, level - 1); | |
2770 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2771 | parent, ref_root, level - 1, 0, |
2772 | for_cow); | |
31840ae1 ZY |
2773 | if (ret) |
2774 | goto fail; | |
2775 | } | |
2776 | } | |
2777 | return 0; | |
2778 | fail: | |
5d4f98a2 YZ |
2779 | return ret; |
2780 | } | |
2781 | ||
2782 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2783 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2784 | { |
66d7e7f0 | 2785 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2786 | } |
2787 | ||
2788 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2789 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2790 | { |
66d7e7f0 | 2791 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2792 | } |
2793 | ||
9078a3e1 CM |
2794 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2795 | struct btrfs_root *root, | |
2796 | struct btrfs_path *path, | |
2797 | struct btrfs_block_group_cache *cache) | |
2798 | { | |
2799 | int ret; | |
9078a3e1 | 2800 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2801 | unsigned long bi; |
2802 | struct extent_buffer *leaf; | |
9078a3e1 | 2803 | |
9078a3e1 | 2804 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2805 | if (ret < 0) |
2806 | goto fail; | |
79787eaa | 2807 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
2808 | |
2809 | leaf = path->nodes[0]; | |
2810 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2811 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2812 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2813 | btrfs_release_path(path); |
54aa1f4d | 2814 | fail: |
79787eaa JM |
2815 | if (ret) { |
2816 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 2817 | return ret; |
79787eaa | 2818 | } |
9078a3e1 CM |
2819 | return 0; |
2820 | ||
2821 | } | |
2822 | ||
4a8c9a62 YZ |
2823 | static struct btrfs_block_group_cache * |
2824 | next_block_group(struct btrfs_root *root, | |
2825 | struct btrfs_block_group_cache *cache) | |
2826 | { | |
2827 | struct rb_node *node; | |
2828 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2829 | node = rb_next(&cache->cache_node); | |
2830 | btrfs_put_block_group(cache); | |
2831 | if (node) { | |
2832 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2833 | cache_node); | |
11dfe35a | 2834 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2835 | } else |
2836 | cache = NULL; | |
2837 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2838 | return cache; | |
2839 | } | |
2840 | ||
0af3d00b JB |
2841 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2842 | struct btrfs_trans_handle *trans, | |
2843 | struct btrfs_path *path) | |
2844 | { | |
2845 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2846 | struct inode *inode = NULL; | |
2847 | u64 alloc_hint = 0; | |
2b20982e | 2848 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2849 | int num_pages = 0; |
2850 | int retries = 0; | |
2851 | int ret = 0; | |
2852 | ||
2853 | /* | |
2854 | * If this block group is smaller than 100 megs don't bother caching the | |
2855 | * block group. | |
2856 | */ | |
2857 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2858 | spin_lock(&block_group->lock); | |
2859 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2860 | spin_unlock(&block_group->lock); | |
2861 | return 0; | |
2862 | } | |
2863 | ||
2864 | again: | |
2865 | inode = lookup_free_space_inode(root, block_group, path); | |
2866 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2867 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2868 | btrfs_release_path(path); |
0af3d00b JB |
2869 | goto out; |
2870 | } | |
2871 | ||
2872 | if (IS_ERR(inode)) { | |
2873 | BUG_ON(retries); | |
2874 | retries++; | |
2875 | ||
2876 | if (block_group->ro) | |
2877 | goto out_free; | |
2878 | ||
2879 | ret = create_free_space_inode(root, trans, block_group, path); | |
2880 | if (ret) | |
2881 | goto out_free; | |
2882 | goto again; | |
2883 | } | |
2884 | ||
5b0e95bf JB |
2885 | /* We've already setup this transaction, go ahead and exit */ |
2886 | if (block_group->cache_generation == trans->transid && | |
2887 | i_size_read(inode)) { | |
2888 | dcs = BTRFS_DC_SETUP; | |
2889 | goto out_put; | |
2890 | } | |
2891 | ||
0af3d00b JB |
2892 | /* |
2893 | * We want to set the generation to 0, that way if anything goes wrong | |
2894 | * from here on out we know not to trust this cache when we load up next | |
2895 | * time. | |
2896 | */ | |
2897 | BTRFS_I(inode)->generation = 0; | |
2898 | ret = btrfs_update_inode(trans, root, inode); | |
2899 | WARN_ON(ret); | |
2900 | ||
2901 | if (i_size_read(inode) > 0) { | |
2902 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2903 | inode); | |
2904 | if (ret) | |
2905 | goto out_put; | |
2906 | } | |
2907 | ||
2908 | spin_lock(&block_group->lock); | |
2909 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2910 | /* We're not cached, don't bother trying to write stuff out */ |
2911 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2912 | spin_unlock(&block_group->lock); |
2913 | goto out_put; | |
2914 | } | |
2915 | spin_unlock(&block_group->lock); | |
2916 | ||
2917 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2918 | if (!num_pages) | |
2919 | num_pages = 1; | |
2920 | ||
2921 | /* | |
2922 | * Just to make absolutely sure we have enough space, we're going to | |
2923 | * preallocate 12 pages worth of space for each block group. In | |
2924 | * practice we ought to use at most 8, but we need extra space so we can | |
2925 | * add our header and have a terminator between the extents and the | |
2926 | * bitmaps. | |
2927 | */ | |
2928 | num_pages *= 16; | |
2929 | num_pages *= PAGE_CACHE_SIZE; | |
2930 | ||
2931 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2932 | if (ret) | |
2933 | goto out_put; | |
2934 | ||
2935 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2936 | num_pages, num_pages, | |
2937 | &alloc_hint); | |
2b20982e JB |
2938 | if (!ret) |
2939 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 2940 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 2941 | |
0af3d00b JB |
2942 | out_put: |
2943 | iput(inode); | |
2944 | out_free: | |
b3b4aa74 | 2945 | btrfs_release_path(path); |
0af3d00b JB |
2946 | out: |
2947 | spin_lock(&block_group->lock); | |
e65cbb94 | 2948 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 2949 | block_group->cache_generation = trans->transid; |
2b20982e | 2950 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2951 | spin_unlock(&block_group->lock); |
2952 | ||
2953 | return ret; | |
2954 | } | |
2955 | ||
96b5179d CM |
2956 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2957 | struct btrfs_root *root) | |
9078a3e1 | 2958 | { |
4a8c9a62 | 2959 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2960 | int err = 0; |
9078a3e1 | 2961 | struct btrfs_path *path; |
96b5179d | 2962 | u64 last = 0; |
9078a3e1 CM |
2963 | |
2964 | path = btrfs_alloc_path(); | |
2965 | if (!path) | |
2966 | return -ENOMEM; | |
2967 | ||
0af3d00b JB |
2968 | again: |
2969 | while (1) { | |
2970 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2971 | while (cache) { | |
2972 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2973 | break; | |
2974 | cache = next_block_group(root, cache); | |
2975 | } | |
2976 | if (!cache) { | |
2977 | if (last == 0) | |
2978 | break; | |
2979 | last = 0; | |
2980 | continue; | |
2981 | } | |
2982 | err = cache_save_setup(cache, trans, path); | |
2983 | last = cache->key.objectid + cache->key.offset; | |
2984 | btrfs_put_block_group(cache); | |
2985 | } | |
2986 | ||
d397712b | 2987 | while (1) { |
4a8c9a62 YZ |
2988 | if (last == 0) { |
2989 | err = btrfs_run_delayed_refs(trans, root, | |
2990 | (unsigned long)-1); | |
79787eaa JM |
2991 | if (err) /* File system offline */ |
2992 | goto out; | |
0f9dd46c | 2993 | } |
54aa1f4d | 2994 | |
4a8c9a62 YZ |
2995 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2996 | while (cache) { | |
0af3d00b JB |
2997 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2998 | btrfs_put_block_group(cache); | |
2999 | goto again; | |
3000 | } | |
3001 | ||
4a8c9a62 YZ |
3002 | if (cache->dirty) |
3003 | break; | |
3004 | cache = next_block_group(root, cache); | |
3005 | } | |
3006 | if (!cache) { | |
3007 | if (last == 0) | |
3008 | break; | |
3009 | last = 0; | |
3010 | continue; | |
3011 | } | |
0f9dd46c | 3012 | |
0cb59c99 JB |
3013 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3014 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3015 | cache->dirty = 0; |
4a8c9a62 | 3016 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3017 | |
4a8c9a62 | 3018 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3019 | if (err) /* File system offline */ |
3020 | goto out; | |
3021 | ||
4a8c9a62 | 3022 | btrfs_put_block_group(cache); |
9078a3e1 | 3023 | } |
4a8c9a62 | 3024 | |
0cb59c99 JB |
3025 | while (1) { |
3026 | /* | |
3027 | * I don't think this is needed since we're just marking our | |
3028 | * preallocated extent as written, but just in case it can't | |
3029 | * hurt. | |
3030 | */ | |
3031 | if (last == 0) { | |
3032 | err = btrfs_run_delayed_refs(trans, root, | |
3033 | (unsigned long)-1); | |
79787eaa JM |
3034 | if (err) /* File system offline */ |
3035 | goto out; | |
0cb59c99 JB |
3036 | } |
3037 | ||
3038 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3039 | while (cache) { | |
3040 | /* | |
3041 | * Really this shouldn't happen, but it could if we | |
3042 | * couldn't write the entire preallocated extent and | |
3043 | * splitting the extent resulted in a new block. | |
3044 | */ | |
3045 | if (cache->dirty) { | |
3046 | btrfs_put_block_group(cache); | |
3047 | goto again; | |
3048 | } | |
3049 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3050 | break; | |
3051 | cache = next_block_group(root, cache); | |
3052 | } | |
3053 | if (!cache) { | |
3054 | if (last == 0) | |
3055 | break; | |
3056 | last = 0; | |
3057 | continue; | |
3058 | } | |
3059 | ||
79787eaa | 3060 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3061 | |
3062 | /* | |
3063 | * If we didn't have an error then the cache state is still | |
3064 | * NEED_WRITE, so we can set it to WRITTEN. | |
3065 | */ | |
79787eaa | 3066 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3067 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3068 | last = cache->key.objectid + cache->key.offset; | |
3069 | btrfs_put_block_group(cache); | |
3070 | } | |
79787eaa | 3071 | out: |
0cb59c99 | 3072 | |
9078a3e1 | 3073 | btrfs_free_path(path); |
79787eaa | 3074 | return err; |
9078a3e1 CM |
3075 | } |
3076 | ||
d2fb3437 YZ |
3077 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3078 | { | |
3079 | struct btrfs_block_group_cache *block_group; | |
3080 | int readonly = 0; | |
3081 | ||
3082 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3083 | if (!block_group || block_group->ro) | |
3084 | readonly = 1; | |
3085 | if (block_group) | |
fa9c0d79 | 3086 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3087 | return readonly; |
3088 | } | |
3089 | ||
593060d7 CM |
3090 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3091 | u64 total_bytes, u64 bytes_used, | |
3092 | struct btrfs_space_info **space_info) | |
3093 | { | |
3094 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3095 | int i; |
3096 | int factor; | |
3097 | ||
3098 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3099 | BTRFS_BLOCK_GROUP_RAID10)) | |
3100 | factor = 2; | |
3101 | else | |
3102 | factor = 1; | |
593060d7 CM |
3103 | |
3104 | found = __find_space_info(info, flags); | |
3105 | if (found) { | |
25179201 | 3106 | spin_lock(&found->lock); |
593060d7 | 3107 | found->total_bytes += total_bytes; |
89a55897 | 3108 | found->disk_total += total_bytes * factor; |
593060d7 | 3109 | found->bytes_used += bytes_used; |
b742bb82 | 3110 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3111 | found->full = 0; |
25179201 | 3112 | spin_unlock(&found->lock); |
593060d7 CM |
3113 | *space_info = found; |
3114 | return 0; | |
3115 | } | |
c146afad | 3116 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3117 | if (!found) |
3118 | return -ENOMEM; | |
3119 | ||
b742bb82 YZ |
3120 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3121 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3122 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3123 | spin_lock_init(&found->lock); |
52ba6929 | 3124 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3125 | found->total_bytes = total_bytes; |
89a55897 | 3126 | found->disk_total = total_bytes * factor; |
593060d7 | 3127 | found->bytes_used = bytes_used; |
b742bb82 | 3128 | found->disk_used = bytes_used * factor; |
593060d7 | 3129 | found->bytes_pinned = 0; |
e8569813 | 3130 | found->bytes_reserved = 0; |
c146afad | 3131 | found->bytes_readonly = 0; |
f0486c68 | 3132 | found->bytes_may_use = 0; |
593060d7 | 3133 | found->full = 0; |
0e4f8f88 | 3134 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3135 | found->chunk_alloc = 0; |
fdb5effd JB |
3136 | found->flush = 0; |
3137 | init_waitqueue_head(&found->wait); | |
593060d7 | 3138 | *space_info = found; |
4184ea7f | 3139 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
3140 | return 0; |
3141 | } | |
3142 | ||
8790d502 CM |
3143 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3144 | { | |
899c81ea ID |
3145 | u64 extra_flags = chunk_to_extended(flags) & |
3146 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 ID |
3147 | |
3148 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3149 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3150 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3151 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3152 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3153 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3154 | } |
593060d7 | 3155 | |
fc67c450 ID |
3156 | /* |
3157 | * returns target flags in extended format or 0 if restripe for this | |
3158 | * chunk_type is not in progress | |
c6664b42 ID |
3159 | * |
3160 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3161 | */ |
3162 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3163 | { | |
3164 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3165 | u64 target = 0; | |
3166 | ||
fc67c450 ID |
3167 | if (!bctl) |
3168 | return 0; | |
3169 | ||
3170 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3171 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3172 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3173 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3174 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3175 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3176 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3177 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3178 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3179 | } | |
3180 | ||
3181 | return target; | |
3182 | } | |
3183 | ||
a46d11a8 ID |
3184 | /* |
3185 | * @flags: available profiles in extended format (see ctree.h) | |
3186 | * | |
e4d8ec0f ID |
3187 | * Returns reduced profile in chunk format. If profile changing is in |
3188 | * progress (either running or paused) picks the target profile (if it's | |
3189 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3190 | */ |
2b82032c | 3191 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3192 | { |
cd02dca5 CM |
3193 | /* |
3194 | * we add in the count of missing devices because we want | |
3195 | * to make sure that any RAID levels on a degraded FS | |
3196 | * continue to be honored. | |
3197 | */ | |
3198 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3199 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3200 | u64 target; |
a061fc8d | 3201 | |
fc67c450 ID |
3202 | /* |
3203 | * see if restripe for this chunk_type is in progress, if so | |
3204 | * try to reduce to the target profile | |
3205 | */ | |
e4d8ec0f | 3206 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3207 | target = get_restripe_target(root->fs_info, flags); |
3208 | if (target) { | |
3209 | /* pick target profile only if it's already available */ | |
3210 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3211 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3212 | return extended_to_chunk(target); |
e4d8ec0f ID |
3213 | } |
3214 | } | |
3215 | spin_unlock(&root->fs_info->balance_lock); | |
3216 | ||
a061fc8d CM |
3217 | if (num_devices == 1) |
3218 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3219 | if (num_devices < 4) | |
3220 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3221 | ||
ec44a35c CM |
3222 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3223 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3224 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3225 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3226 | } |
ec44a35c CM |
3227 | |
3228 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3229 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3230 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3231 | } |
ec44a35c CM |
3232 | |
3233 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3234 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3235 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3236 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3237 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3238 | } |
3239 | ||
899c81ea | 3240 | return extended_to_chunk(flags); |
ec44a35c CM |
3241 | } |
3242 | ||
b742bb82 | 3243 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3244 | { |
b742bb82 | 3245 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3246 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3247 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3248 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3249 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3250 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3251 | ||
b742bb82 | 3252 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3253 | } |
3254 | ||
6d07bcec | 3255 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3256 | { |
b742bb82 | 3257 | u64 flags; |
9ed74f2d | 3258 | |
b742bb82 YZ |
3259 | if (data) |
3260 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3261 | else if (root == root->fs_info->chunk_root) | |
3262 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3263 | else |
b742bb82 | 3264 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3265 | |
b742bb82 | 3266 | return get_alloc_profile(root, flags); |
6a63209f | 3267 | } |
9ed74f2d | 3268 | |
6a63209f JB |
3269 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3270 | { | |
6a63209f | 3271 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3272 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3273 | } |
3274 | ||
6a63209f | 3275 | /* |
6a63209f JB |
3276 | * This will check the space that the inode allocates from to make sure we have |
3277 | * enough space for bytes. | |
6a63209f | 3278 | */ |
0ca1f7ce | 3279 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3280 | { |
6a63209f | 3281 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3282 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3283 | u64 used; |
0af3d00b | 3284 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3285 | |
6a63209f JB |
3286 | /* make sure bytes are sectorsize aligned */ |
3287 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3288 | |
82d5902d LZ |
3289 | if (root == root->fs_info->tree_root || |
3290 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3291 | alloc_chunk = 0; |
3292 | committed = 1; | |
3293 | } | |
3294 | ||
6a63209f | 3295 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3296 | if (!data_sinfo) |
3297 | goto alloc; | |
9ed74f2d | 3298 | |
6a63209f JB |
3299 | again: |
3300 | /* make sure we have enough space to handle the data first */ | |
3301 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3302 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3303 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3304 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3305 | |
3306 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3307 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3308 | |
6a63209f JB |
3309 | /* |
3310 | * if we don't have enough free bytes in this space then we need | |
3311 | * to alloc a new chunk. | |
3312 | */ | |
0af3d00b | 3313 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3314 | u64 alloc_target; |
9ed74f2d | 3315 | |
0e4f8f88 | 3316 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3317 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3318 | alloc: |
6a63209f | 3319 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3320 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3321 | if (IS_ERR(trans)) |
3322 | return PTR_ERR(trans); | |
9ed74f2d | 3323 | |
6a63209f JB |
3324 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3325 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3326 | alloc_target, |
3327 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3328 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3329 | if (ret < 0) { |
3330 | if (ret != -ENOSPC) | |
3331 | return ret; | |
3332 | else | |
3333 | goto commit_trans; | |
3334 | } | |
9ed74f2d | 3335 | |
33b4d47f CM |
3336 | if (!data_sinfo) { |
3337 | btrfs_set_inode_space_info(root, inode); | |
3338 | data_sinfo = BTRFS_I(inode)->space_info; | |
3339 | } | |
6a63209f JB |
3340 | goto again; |
3341 | } | |
f2bb8f5c JB |
3342 | |
3343 | /* | |
3344 | * If we have less pinned bytes than we want to allocate then | |
3345 | * don't bother committing the transaction, it won't help us. | |
3346 | */ | |
3347 | if (data_sinfo->bytes_pinned < bytes) | |
3348 | committed = 1; | |
6a63209f | 3349 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3350 | |
4e06bdd6 | 3351 | /* commit the current transaction and try again */ |
d52a5b5f | 3352 | commit_trans: |
a4abeea4 JB |
3353 | if (!committed && |
3354 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3355 | committed = 1; |
7a7eaa40 | 3356 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3357 | if (IS_ERR(trans)) |
3358 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3359 | ret = btrfs_commit_transaction(trans, root); |
3360 | if (ret) | |
3361 | return ret; | |
3362 | goto again; | |
3363 | } | |
9ed74f2d | 3364 | |
6a63209f JB |
3365 | return -ENOSPC; |
3366 | } | |
3367 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3368 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3369 | data_sinfo->flags, bytes, 1); |
6a63209f | 3370 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3371 | |
9ed74f2d | 3372 | return 0; |
9ed74f2d | 3373 | } |
6a63209f | 3374 | |
6a63209f | 3375 | /* |
fb25e914 | 3376 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3377 | */ |
0ca1f7ce | 3378 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3379 | { |
0ca1f7ce | 3380 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3381 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3382 | |
6a63209f JB |
3383 | /* make sure bytes are sectorsize aligned */ |
3384 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3385 | |
6a63209f JB |
3386 | data_sinfo = BTRFS_I(inode)->space_info; |
3387 | spin_lock(&data_sinfo->lock); | |
3388 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3389 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3390 | data_sinfo->flags, bytes, 0); |
6a63209f | 3391 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3392 | } |
3393 | ||
97e728d4 | 3394 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3395 | { |
97e728d4 JB |
3396 | struct list_head *head = &info->space_info; |
3397 | struct btrfs_space_info *found; | |
e3ccfa98 | 3398 | |
97e728d4 JB |
3399 | rcu_read_lock(); |
3400 | list_for_each_entry_rcu(found, head, list) { | |
3401 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3402 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3403 | } |
97e728d4 | 3404 | rcu_read_unlock(); |
e3ccfa98 JB |
3405 | } |
3406 | ||
e5bc2458 | 3407 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3408 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3409 | int force) | |
32c00aff | 3410 | { |
fb25e914 | 3411 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3412 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3413 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3414 | u64 thresh; |
e3ccfa98 | 3415 | |
0e4f8f88 CM |
3416 | if (force == CHUNK_ALLOC_FORCE) |
3417 | return 1; | |
3418 | ||
fb25e914 JB |
3419 | /* |
3420 | * We need to take into account the global rsv because for all intents | |
3421 | * and purposes it's used space. Don't worry about locking the | |
3422 | * global_rsv, it doesn't change except when the transaction commits. | |
3423 | */ | |
3424 | num_allocated += global_rsv->size; | |
3425 | ||
0e4f8f88 CM |
3426 | /* |
3427 | * in limited mode, we want to have some free space up to | |
3428 | * about 1% of the FS size. | |
3429 | */ | |
3430 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3431 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3432 | thresh = max_t(u64, 64 * 1024 * 1024, |
3433 | div_factor_fine(thresh, 1)); | |
3434 | ||
3435 | if (num_bytes - num_allocated < thresh) | |
3436 | return 1; | |
3437 | } | |
6c41761f | 3438 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 | 3439 | |
cf1d72c9 CM |
3440 | /* 256MB or 2% of the FS */ |
3441 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2)); | |
96bdc7dc CM |
3442 | /* system chunks need a much small threshold */ |
3443 | if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3444 | thresh = 32 * 1024 * 1024; | |
e5bc2458 | 3445 | |
cf1d72c9 | 3446 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8)) |
14ed0ca6 | 3447 | return 0; |
424499db | 3448 | return 1; |
32c00aff JB |
3449 | } |
3450 | ||
15d1ff81 LB |
3451 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3452 | { | |
3453 | u64 num_dev; | |
3454 | ||
3455 | if (type & BTRFS_BLOCK_GROUP_RAID10 || | |
3456 | type & BTRFS_BLOCK_GROUP_RAID0) | |
3457 | num_dev = root->fs_info->fs_devices->rw_devices; | |
3458 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3459 | num_dev = 2; | |
3460 | else | |
3461 | num_dev = 1; /* DUP or single */ | |
3462 | ||
3463 | /* metadata for updaing devices and chunk tree */ | |
3464 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3465 | } | |
3466 | ||
3467 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3468 | struct btrfs_root *root, u64 type) | |
3469 | { | |
3470 | struct btrfs_space_info *info; | |
3471 | u64 left; | |
3472 | u64 thresh; | |
3473 | ||
3474 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3475 | spin_lock(&info->lock); | |
3476 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3477 | info->bytes_reserved - info->bytes_readonly; | |
3478 | spin_unlock(&info->lock); | |
3479 | ||
3480 | thresh = get_system_chunk_thresh(root, type); | |
3481 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3482 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3483 | left, thresh, type); | |
3484 | dump_space_info(info, 0, 0); | |
3485 | } | |
3486 | ||
3487 | if (left < thresh) { | |
3488 | u64 flags; | |
3489 | ||
3490 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3491 | btrfs_alloc_chunk(trans, root, flags); | |
3492 | } | |
3493 | } | |
3494 | ||
6324fbf3 CM |
3495 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3496 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3497 | u64 flags, int force) |
9ed74f2d | 3498 | { |
6324fbf3 | 3499 | struct btrfs_space_info *space_info; |
97e728d4 | 3500 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3501 | int wait_for_alloc = 0; |
9ed74f2d | 3502 | int ret = 0; |
9ed74f2d | 3503 | |
6324fbf3 | 3504 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3505 | if (!space_info) { |
3506 | ret = update_space_info(extent_root->fs_info, flags, | |
3507 | 0, 0, &space_info); | |
79787eaa | 3508 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3509 | } |
79787eaa | 3510 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3511 | |
6d74119f | 3512 | again: |
25179201 | 3513 | spin_lock(&space_info->lock); |
9e622d6b | 3514 | if (force < space_info->force_alloc) |
0e4f8f88 | 3515 | force = space_info->force_alloc; |
25179201 JB |
3516 | if (space_info->full) { |
3517 | spin_unlock(&space_info->lock); | |
6d74119f | 3518 | return 0; |
9ed74f2d JB |
3519 | } |
3520 | ||
0e4f8f88 | 3521 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3522 | spin_unlock(&space_info->lock); |
6d74119f JB |
3523 | return 0; |
3524 | } else if (space_info->chunk_alloc) { | |
3525 | wait_for_alloc = 1; | |
3526 | } else { | |
3527 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3528 | } |
0e4f8f88 | 3529 | |
25179201 | 3530 | spin_unlock(&space_info->lock); |
9ed74f2d | 3531 | |
6d74119f JB |
3532 | mutex_lock(&fs_info->chunk_mutex); |
3533 | ||
3534 | /* | |
3535 | * The chunk_mutex is held throughout the entirety of a chunk | |
3536 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3537 | * other guy is done and we need to recheck and see if we should | |
3538 | * allocate. | |
3539 | */ | |
3540 | if (wait_for_alloc) { | |
3541 | mutex_unlock(&fs_info->chunk_mutex); | |
3542 | wait_for_alloc = 0; | |
3543 | goto again; | |
3544 | } | |
3545 | ||
67377734 JB |
3546 | /* |
3547 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3548 | * allocating mixed chunks instead of individual chunks. | |
3549 | */ | |
3550 | if (btrfs_mixed_space_info(space_info)) | |
3551 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3552 | ||
97e728d4 JB |
3553 | /* |
3554 | * if we're doing a data chunk, go ahead and make sure that | |
3555 | * we keep a reasonable number of metadata chunks allocated in the | |
3556 | * FS as well. | |
3557 | */ | |
9ed74f2d | 3558 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3559 | fs_info->data_chunk_allocations++; |
3560 | if (!(fs_info->data_chunk_allocations % | |
3561 | fs_info->metadata_ratio)) | |
3562 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3563 | } |
3564 | ||
15d1ff81 LB |
3565 | /* |
3566 | * Check if we have enough space in SYSTEM chunk because we may need | |
3567 | * to update devices. | |
3568 | */ | |
3569 | check_system_chunk(trans, extent_root, flags); | |
3570 | ||
2b82032c | 3571 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3572 | if (ret < 0 && ret != -ENOSPC) |
3573 | goto out; | |
3574 | ||
9ed74f2d | 3575 | spin_lock(&space_info->lock); |
9ed74f2d | 3576 | if (ret) |
6324fbf3 | 3577 | space_info->full = 1; |
424499db YZ |
3578 | else |
3579 | ret = 1; | |
6d74119f | 3580 | |
0e4f8f88 | 3581 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3582 | space_info->chunk_alloc = 0; |
9ed74f2d | 3583 | spin_unlock(&space_info->lock); |
92b8e897 | 3584 | out: |
a25c75d5 | 3585 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3586 | return ret; |
6324fbf3 | 3587 | } |
9ed74f2d | 3588 | |
9ed74f2d | 3589 | /* |
5da9d01b | 3590 | * shrink metadata reservation for delalloc |
9ed74f2d | 3591 | */ |
663350ac | 3592 | static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, |
f104d044 | 3593 | bool wait_ordered) |
5da9d01b | 3594 | { |
0ca1f7ce | 3595 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3596 | struct btrfs_space_info *space_info; |
663350ac | 3597 | struct btrfs_trans_handle *trans; |
5da9d01b YZ |
3598 | u64 reserved; |
3599 | u64 max_reclaim; | |
3600 | u64 reclaimed = 0; | |
b1953bce | 3601 | long time_left; |
877da174 | 3602 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3603 | int loops = 0; |
36e39c40 | 3604 | unsigned long progress; |
5da9d01b | 3605 | |
663350ac | 3606 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3607 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3608 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3609 | |
3610 | smp_mb(); | |
fb25e914 | 3611 | reserved = space_info->bytes_may_use; |
36e39c40 | 3612 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3613 | |
3614 | if (reserved == 0) | |
3615 | return 0; | |
c4f675cd | 3616 | |
fdb5effd JB |
3617 | smp_mb(); |
3618 | if (root->fs_info->delalloc_bytes == 0) { | |
3619 | if (trans) | |
3620 | return 0; | |
3621 | btrfs_wait_ordered_extents(root, 0, 0); | |
3622 | return 0; | |
3623 | } | |
3624 | ||
5da9d01b | 3625 | max_reclaim = min(reserved, to_reclaim); |
877da174 JB |
3626 | nr_pages = max_t(unsigned long, nr_pages, |
3627 | max_reclaim >> PAGE_CACHE_SHIFT); | |
b1953bce | 3628 | while (loops < 1024) { |
bf9022e0 CM |
3629 | /* have the flusher threads jump in and do some IO */ |
3630 | smp_mb(); | |
3631 | nr_pages = min_t(unsigned long, nr_pages, | |
3632 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
0e175a18 CW |
3633 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages, |
3634 | WB_REASON_FS_FREE_SPACE); | |
5da9d01b | 3635 | |
0019f10d | 3636 | spin_lock(&space_info->lock); |
fb25e914 JB |
3637 | if (reserved > space_info->bytes_may_use) |
3638 | reclaimed += reserved - space_info->bytes_may_use; | |
3639 | reserved = space_info->bytes_may_use; | |
0019f10d | 3640 | spin_unlock(&space_info->lock); |
5da9d01b | 3641 | |
36e39c40 CM |
3642 | loops++; |
3643 | ||
5da9d01b YZ |
3644 | if (reserved == 0 || reclaimed >= max_reclaim) |
3645 | break; | |
3646 | ||
3647 | if (trans && trans->transaction->blocked) | |
3648 | return -EAGAIN; | |
bf9022e0 | 3649 | |
f104d044 JB |
3650 | if (wait_ordered && !trans) { |
3651 | btrfs_wait_ordered_extents(root, 0, 0); | |
3652 | } else { | |
3653 | time_left = schedule_timeout_interruptible(1); | |
b1953bce | 3654 | |
f104d044 JB |
3655 | /* We were interrupted, exit */ |
3656 | if (time_left) | |
3657 | break; | |
3658 | } | |
b1953bce | 3659 | |
36e39c40 CM |
3660 | /* we've kicked the IO a few times, if anything has been freed, |
3661 | * exit. There is no sense in looping here for a long time | |
3662 | * when we really need to commit the transaction, or there are | |
3663 | * just too many writers without enough free space | |
3664 | */ | |
3665 | ||
3666 | if (loops > 3) { | |
3667 | smp_mb(); | |
3668 | if (progress != space_info->reservation_progress) | |
3669 | break; | |
3670 | } | |
bf9022e0 | 3671 | |
5da9d01b | 3672 | } |
f104d044 | 3673 | |
5da9d01b YZ |
3674 | return reclaimed >= to_reclaim; |
3675 | } | |
3676 | ||
663350ac JB |
3677 | /** |
3678 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3679 | * @root - the root we're allocating for | |
3680 | * @bytes - the number of bytes we want to reserve | |
3681 | * @force - force the commit | |
8bb8ab2e | 3682 | * |
663350ac JB |
3683 | * This will check to make sure that committing the transaction will actually |
3684 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3685 | * will return -ENOSPC. | |
8bb8ab2e | 3686 | */ |
663350ac JB |
3687 | static int may_commit_transaction(struct btrfs_root *root, |
3688 | struct btrfs_space_info *space_info, | |
3689 | u64 bytes, int force) | |
3690 | { | |
3691 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3692 | struct btrfs_trans_handle *trans; | |
3693 | ||
3694 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3695 | if (trans) | |
3696 | return -EAGAIN; | |
3697 | ||
3698 | if (force) | |
3699 | goto commit; | |
3700 | ||
3701 | /* See if there is enough pinned space to make this reservation */ | |
3702 | spin_lock(&space_info->lock); | |
3703 | if (space_info->bytes_pinned >= bytes) { | |
3704 | spin_unlock(&space_info->lock); | |
3705 | goto commit; | |
3706 | } | |
3707 | spin_unlock(&space_info->lock); | |
3708 | ||
3709 | /* | |
3710 | * See if there is some space in the delayed insertion reservation for | |
3711 | * this reservation. | |
3712 | */ | |
3713 | if (space_info != delayed_rsv->space_info) | |
3714 | return -ENOSPC; | |
3715 | ||
d9b0218f | 3716 | spin_lock(&space_info->lock); |
663350ac | 3717 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3718 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3719 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3720 | spin_unlock(&space_info->lock); |
663350ac JB |
3721 | return -ENOSPC; |
3722 | } | |
3723 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3724 | spin_unlock(&space_info->lock); |
663350ac JB |
3725 | |
3726 | commit: | |
3727 | trans = btrfs_join_transaction(root); | |
3728 | if (IS_ERR(trans)) | |
3729 | return -ENOSPC; | |
3730 | ||
3731 | return btrfs_commit_transaction(trans, root); | |
3732 | } | |
3733 | ||
4a92b1b8 JB |
3734 | /** |
3735 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3736 | * @root - the root we're allocating for | |
3737 | * @block_rsv - the block_rsv we're allocating for | |
3738 | * @orig_bytes - the number of bytes we want | |
3739 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3740 | * |
4a92b1b8 JB |
3741 | * This will reserve orgi_bytes number of bytes from the space info associated |
3742 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3743 | * flush out space to make room. It will do this by flushing delalloc if | |
3744 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3745 | * regain reservations will be made and this will fail if there is not enough | |
3746 | * space already. | |
8bb8ab2e | 3747 | */ |
4a92b1b8 | 3748 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e JB |
3749 | struct btrfs_block_rsv *block_rsv, |
3750 | u64 orig_bytes, int flush) | |
9ed74f2d | 3751 | { |
f0486c68 | 3752 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3753 | u64 used; |
8bb8ab2e JB |
3754 | u64 num_bytes = orig_bytes; |
3755 | int retries = 0; | |
3756 | int ret = 0; | |
38227933 | 3757 | bool committed = false; |
fdb5effd | 3758 | bool flushing = false; |
f104d044 | 3759 | bool wait_ordered = false; |
9ed74f2d | 3760 | |
8bb8ab2e | 3761 | again: |
fdb5effd | 3762 | ret = 0; |
8bb8ab2e | 3763 | spin_lock(&space_info->lock); |
fdb5effd JB |
3764 | /* |
3765 | * We only want to wait if somebody other than us is flushing and we are | |
3766 | * actually alloed to flush. | |
3767 | */ | |
3768 | while (flush && !flushing && space_info->flush) { | |
3769 | spin_unlock(&space_info->lock); | |
3770 | /* | |
3771 | * If we have a trans handle we can't wait because the flusher | |
3772 | * may have to commit the transaction, which would mean we would | |
3773 | * deadlock since we are waiting for the flusher to finish, but | |
3774 | * hold the current transaction open. | |
3775 | */ | |
663350ac | 3776 | if (current->journal_info) |
fdb5effd | 3777 | return -EAGAIN; |
b9688bb8 AJ |
3778 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
3779 | /* Must have been killed, return */ | |
3780 | if (ret) | |
fdb5effd JB |
3781 | return -EINTR; |
3782 | ||
3783 | spin_lock(&space_info->lock); | |
3784 | } | |
3785 | ||
3786 | ret = -ENOSPC; | |
2bf64758 JB |
3787 | used = space_info->bytes_used + space_info->bytes_reserved + |
3788 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3789 | space_info->bytes_may_use; | |
9ed74f2d | 3790 | |
8bb8ab2e JB |
3791 | /* |
3792 | * The idea here is that we've not already over-reserved the block group | |
3793 | * then we can go ahead and save our reservation first and then start | |
3794 | * flushing if we need to. Otherwise if we've already overcommitted | |
3795 | * lets start flushing stuff first and then come back and try to make | |
3796 | * our reservation. | |
3797 | */ | |
2bf64758 JB |
3798 | if (used <= space_info->total_bytes) { |
3799 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3800 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3801 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3802 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
3803 | ret = 0; |
3804 | } else { | |
3805 | /* | |
3806 | * Ok set num_bytes to orig_bytes since we aren't | |
3807 | * overocmmitted, this way we only try and reclaim what | |
3808 | * we need. | |
3809 | */ | |
3810 | num_bytes = orig_bytes; | |
3811 | } | |
3812 | } else { | |
3813 | /* | |
3814 | * Ok we're over committed, set num_bytes to the overcommitted | |
3815 | * amount plus the amount of bytes that we need for this | |
3816 | * reservation. | |
3817 | */ | |
f104d044 | 3818 | wait_ordered = true; |
2bf64758 | 3819 | num_bytes = used - space_info->total_bytes + |
8bb8ab2e JB |
3820 | (orig_bytes * (retries + 1)); |
3821 | } | |
9ed74f2d | 3822 | |
36ba022a | 3823 | if (ret) { |
2bf64758 JB |
3824 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3825 | u64 avail; | |
3826 | ||
7e355b83 JB |
3827 | /* |
3828 | * If we have a lot of space that's pinned, don't bother doing | |
3829 | * the overcommit dance yet and just commit the transaction. | |
3830 | */ | |
3831 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3832 | do_div(avail, 10); | |
663350ac | 3833 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3834 | space_info->flush = 1; |
3835 | flushing = true; | |
3836 | spin_unlock(&space_info->lock); | |
663350ac JB |
3837 | ret = may_commit_transaction(root, space_info, |
3838 | orig_bytes, 1); | |
3839 | if (ret) | |
3840 | goto out; | |
3841 | committed = true; | |
3842 | goto again; | |
7e355b83 JB |
3843 | } |
3844 | ||
2bf64758 JB |
3845 | spin_lock(&root->fs_info->free_chunk_lock); |
3846 | avail = root->fs_info->free_chunk_space; | |
3847 | ||
3848 | /* | |
3849 | * If we have dup, raid1 or raid10 then only half of the free | |
3850 | * space is actually useable. | |
3851 | */ | |
3852 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3853 | BTRFS_BLOCK_GROUP_RAID1 | | |
3854 | BTRFS_BLOCK_GROUP_RAID10)) | |
3855 | avail >>= 1; | |
3856 | ||
3857 | /* | |
3858 | * If we aren't flushing don't let us overcommit too much, say | |
3859 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3860 | * 1/2 of the space. | |
3861 | */ | |
3862 | if (flush) | |
3863 | avail >>= 3; | |
3864 | else | |
3865 | avail >>= 1; | |
3866 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3867 | ||
9a82ca65 | 3868 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 | 3869 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3870 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3871 | "space_info", space_info->flags, orig_bytes, 1); |
2bf64758 | 3872 | ret = 0; |
f104d044 JB |
3873 | } else { |
3874 | wait_ordered = true; | |
2bf64758 JB |
3875 | } |
3876 | } | |
3877 | ||
8bb8ab2e JB |
3878 | /* |
3879 | * Couldn't make our reservation, save our place so while we're trying | |
3880 | * to reclaim space we can actually use it instead of somebody else | |
3881 | * stealing it from us. | |
3882 | */ | |
fdb5effd JB |
3883 | if (ret && flush) { |
3884 | flushing = true; | |
3885 | space_info->flush = 1; | |
8bb8ab2e | 3886 | } |
9ed74f2d | 3887 | |
f0486c68 | 3888 | spin_unlock(&space_info->lock); |
9ed74f2d | 3889 | |
fdb5effd | 3890 | if (!ret || !flush) |
8bb8ab2e | 3891 | goto out; |
f0486c68 | 3892 | |
8bb8ab2e JB |
3893 | /* |
3894 | * We do synchronous shrinking since we don't actually unreserve | |
3895 | * metadata until after the IO is completed. | |
3896 | */ | |
663350ac | 3897 | ret = shrink_delalloc(root, num_bytes, wait_ordered); |
fdb5effd | 3898 | if (ret < 0) |
8bb8ab2e | 3899 | goto out; |
f0486c68 | 3900 | |
75c195a2 CM |
3901 | ret = 0; |
3902 | ||
8bb8ab2e JB |
3903 | /* |
3904 | * So if we were overcommitted it's possible that somebody else flushed | |
3905 | * out enough space and we simply didn't have enough space to reclaim, | |
3906 | * so go back around and try again. | |
3907 | */ | |
3908 | if (retries < 2) { | |
f104d044 | 3909 | wait_ordered = true; |
8bb8ab2e JB |
3910 | retries++; |
3911 | goto again; | |
3912 | } | |
f0486c68 | 3913 | |
8bb8ab2e | 3914 | ret = -ENOSPC; |
75c195a2 CM |
3915 | if (committed) |
3916 | goto out; | |
3917 | ||
663350ac | 3918 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); |
38227933 | 3919 | if (!ret) { |
38227933 | 3920 | committed = true; |
8bb8ab2e | 3921 | goto again; |
38227933 | 3922 | } |
8bb8ab2e JB |
3923 | |
3924 | out: | |
fdb5effd | 3925 | if (flushing) { |
8bb8ab2e | 3926 | spin_lock(&space_info->lock); |
fdb5effd JB |
3927 | space_info->flush = 0; |
3928 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3929 | spin_unlock(&space_info->lock); |
f0486c68 | 3930 | } |
f0486c68 YZ |
3931 | return ret; |
3932 | } | |
3933 | ||
79787eaa JM |
3934 | static struct btrfs_block_rsv *get_block_rsv( |
3935 | const struct btrfs_trans_handle *trans, | |
3936 | const struct btrfs_root *root) | |
f0486c68 | 3937 | { |
4c13d758 JB |
3938 | struct btrfs_block_rsv *block_rsv = NULL; |
3939 | ||
3940 | if (root->ref_cows || root == root->fs_info->csum_root) | |
f0486c68 | 3941 | block_rsv = trans->block_rsv; |
4c13d758 JB |
3942 | |
3943 | if (!block_rsv) | |
f0486c68 YZ |
3944 | block_rsv = root->block_rsv; |
3945 | ||
3946 | if (!block_rsv) | |
3947 | block_rsv = &root->fs_info->empty_block_rsv; | |
3948 | ||
3949 | return block_rsv; | |
3950 | } | |
3951 | ||
3952 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3953 | u64 num_bytes) | |
3954 | { | |
3955 | int ret = -ENOSPC; | |
3956 | spin_lock(&block_rsv->lock); | |
3957 | if (block_rsv->reserved >= num_bytes) { | |
3958 | block_rsv->reserved -= num_bytes; | |
3959 | if (block_rsv->reserved < block_rsv->size) | |
3960 | block_rsv->full = 0; | |
3961 | ret = 0; | |
3962 | } | |
3963 | spin_unlock(&block_rsv->lock); | |
3964 | return ret; | |
3965 | } | |
3966 | ||
3967 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3968 | u64 num_bytes, int update_size) | |
3969 | { | |
3970 | spin_lock(&block_rsv->lock); | |
3971 | block_rsv->reserved += num_bytes; | |
3972 | if (update_size) | |
3973 | block_rsv->size += num_bytes; | |
3974 | else if (block_rsv->reserved >= block_rsv->size) | |
3975 | block_rsv->full = 1; | |
3976 | spin_unlock(&block_rsv->lock); | |
3977 | } | |
3978 | ||
8c2a3ca2 JB |
3979 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
3980 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 3981 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
3982 | { |
3983 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3984 | ||
3985 | spin_lock(&block_rsv->lock); | |
3986 | if (num_bytes == (u64)-1) | |
3987 | num_bytes = block_rsv->size; | |
3988 | block_rsv->size -= num_bytes; | |
3989 | if (block_rsv->reserved >= block_rsv->size) { | |
3990 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3991 | block_rsv->reserved = block_rsv->size; | |
3992 | block_rsv->full = 1; | |
3993 | } else { | |
3994 | num_bytes = 0; | |
3995 | } | |
3996 | spin_unlock(&block_rsv->lock); | |
3997 | ||
3998 | if (num_bytes > 0) { | |
3999 | if (dest) { | |
e9e22899 JB |
4000 | spin_lock(&dest->lock); |
4001 | if (!dest->full) { | |
4002 | u64 bytes_to_add; | |
4003 | ||
4004 | bytes_to_add = dest->size - dest->reserved; | |
4005 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4006 | dest->reserved += bytes_to_add; | |
4007 | if (dest->reserved >= dest->size) | |
4008 | dest->full = 1; | |
4009 | num_bytes -= bytes_to_add; | |
4010 | } | |
4011 | spin_unlock(&dest->lock); | |
4012 | } | |
4013 | if (num_bytes) { | |
f0486c68 | 4014 | spin_lock(&space_info->lock); |
fb25e914 | 4015 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4016 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4017 | space_info->flags, num_bytes, 0); |
36e39c40 | 4018 | space_info->reservation_progress++; |
f0486c68 | 4019 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4020 | } |
9ed74f2d | 4021 | } |
f0486c68 | 4022 | } |
4e06bdd6 | 4023 | |
f0486c68 YZ |
4024 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4025 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4026 | { | |
4027 | int ret; | |
9ed74f2d | 4028 | |
f0486c68 YZ |
4029 | ret = block_rsv_use_bytes(src, num_bytes); |
4030 | if (ret) | |
4031 | return ret; | |
9ed74f2d | 4032 | |
f0486c68 | 4033 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4034 | return 0; |
4035 | } | |
4036 | ||
f0486c68 | 4037 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 4038 | { |
f0486c68 YZ |
4039 | memset(rsv, 0, sizeof(*rsv)); |
4040 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
4041 | } |
4042 | ||
4043 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
4044 | { | |
4045 | struct btrfs_block_rsv *block_rsv; | |
4046 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4047 | |
f0486c68 YZ |
4048 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4049 | if (!block_rsv) | |
4050 | return NULL; | |
9ed74f2d | 4051 | |
f0486c68 | 4052 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
4053 | block_rsv->space_info = __find_space_info(fs_info, |
4054 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4055 | return block_rsv; |
4056 | } | |
9ed74f2d | 4057 | |
f0486c68 YZ |
4058 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4059 | struct btrfs_block_rsv *rsv) | |
4060 | { | |
dabdb640 JB |
4061 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4062 | kfree(rsv); | |
9ed74f2d JB |
4063 | } |
4064 | ||
61b520a9 MX |
4065 | static inline int __block_rsv_add(struct btrfs_root *root, |
4066 | struct btrfs_block_rsv *block_rsv, | |
4067 | u64 num_bytes, int flush) | |
9ed74f2d | 4068 | { |
f0486c68 | 4069 | int ret; |
9ed74f2d | 4070 | |
f0486c68 YZ |
4071 | if (num_bytes == 0) |
4072 | return 0; | |
8bb8ab2e | 4073 | |
61b520a9 | 4074 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4075 | if (!ret) { |
4076 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4077 | return 0; | |
4078 | } | |
9ed74f2d | 4079 | |
f0486c68 | 4080 | return ret; |
f0486c68 | 4081 | } |
9ed74f2d | 4082 | |
61b520a9 MX |
4083 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4084 | struct btrfs_block_rsv *block_rsv, | |
4085 | u64 num_bytes) | |
4086 | { | |
4087 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
4088 | } | |
4089 | ||
c06a0e12 JB |
4090 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
4091 | struct btrfs_block_rsv *block_rsv, | |
4092 | u64 num_bytes) | |
f0486c68 | 4093 | { |
61b520a9 | 4094 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
f0486c68 | 4095 | } |
9ed74f2d | 4096 | |
4a92b1b8 | 4097 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4098 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4099 | { |
4100 | u64 num_bytes = 0; | |
f0486c68 | 4101 | int ret = -ENOSPC; |
9ed74f2d | 4102 | |
f0486c68 YZ |
4103 | if (!block_rsv) |
4104 | return 0; | |
9ed74f2d | 4105 | |
f0486c68 | 4106 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4107 | num_bytes = div_factor(block_rsv->size, min_factor); |
4108 | if (block_rsv->reserved >= num_bytes) | |
4109 | ret = 0; | |
4110 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4111 | |
36ba022a JB |
4112 | return ret; |
4113 | } | |
4114 | ||
aa38a711 MX |
4115 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
4116 | struct btrfs_block_rsv *block_rsv, | |
4117 | u64 min_reserved, int flush) | |
36ba022a JB |
4118 | { |
4119 | u64 num_bytes = 0; | |
4120 | int ret = -ENOSPC; | |
4121 | ||
4122 | if (!block_rsv) | |
4123 | return 0; | |
4124 | ||
4125 | spin_lock(&block_rsv->lock); | |
4126 | num_bytes = min_reserved; | |
13553e52 | 4127 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4128 | ret = 0; |
13553e52 | 4129 | else |
f0486c68 | 4130 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4131 | spin_unlock(&block_rsv->lock); |
13553e52 | 4132 | |
f0486c68 YZ |
4133 | if (!ret) |
4134 | return 0; | |
4135 | ||
aa38a711 | 4136 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4137 | if (!ret) { |
4138 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4139 | return 0; |
6a63209f | 4140 | } |
9ed74f2d | 4141 | |
13553e52 | 4142 | return ret; |
f0486c68 YZ |
4143 | } |
4144 | ||
aa38a711 MX |
4145 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4146 | struct btrfs_block_rsv *block_rsv, | |
4147 | u64 min_reserved) | |
4148 | { | |
4149 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
4150 | } | |
4151 | ||
4152 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
4153 | struct btrfs_block_rsv *block_rsv, | |
4154 | u64 min_reserved) | |
4155 | { | |
4156 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
4157 | } | |
4158 | ||
f0486c68 YZ |
4159 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
4160 | struct btrfs_block_rsv *dst_rsv, | |
4161 | u64 num_bytes) | |
4162 | { | |
4163 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4164 | } | |
4165 | ||
4166 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4167 | struct btrfs_block_rsv *block_rsv, | |
4168 | u64 num_bytes) | |
4169 | { | |
4170 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4171 | if (global_rsv->full || global_rsv == block_rsv || | |
4172 | block_rsv->space_info != global_rsv->space_info) | |
4173 | global_rsv = NULL; | |
8c2a3ca2 JB |
4174 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4175 | num_bytes); | |
6a63209f JB |
4176 | } |
4177 | ||
4178 | /* | |
8929ecfa YZ |
4179 | * helper to calculate size of global block reservation. |
4180 | * the desired value is sum of space used by extent tree, | |
4181 | * checksum tree and root tree | |
6a63209f | 4182 | */ |
8929ecfa | 4183 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4184 | { |
8929ecfa YZ |
4185 | struct btrfs_space_info *sinfo; |
4186 | u64 num_bytes; | |
4187 | u64 meta_used; | |
4188 | u64 data_used; | |
6c41761f | 4189 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4190 | |
8929ecfa YZ |
4191 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4192 | spin_lock(&sinfo->lock); | |
4193 | data_used = sinfo->bytes_used; | |
4194 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4195 | |
8929ecfa YZ |
4196 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4197 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4198 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4199 | data_used = 0; | |
8929ecfa YZ |
4200 | meta_used = sinfo->bytes_used; |
4201 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4202 | |
8929ecfa YZ |
4203 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4204 | csum_size * 2; | |
4205 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4206 | |
8929ecfa | 4207 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4208 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4209 | |
8929ecfa YZ |
4210 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4211 | } | |
6a63209f | 4212 | |
8929ecfa YZ |
4213 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4214 | { | |
4215 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4216 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4217 | u64 num_bytes; | |
6a63209f | 4218 | |
8929ecfa | 4219 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4220 | |
8929ecfa | 4221 | spin_lock(&sinfo->lock); |
1f699d38 | 4222 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4223 | |
8929ecfa | 4224 | block_rsv->size = num_bytes; |
4e06bdd6 | 4225 | |
8929ecfa | 4226 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4227 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4228 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4229 | |
4230 | if (sinfo->total_bytes > num_bytes) { | |
4231 | num_bytes = sinfo->total_bytes - num_bytes; | |
4232 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4233 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4234 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4235 | sinfo->flags, num_bytes, 1); |
6a63209f | 4236 | } |
6a63209f | 4237 | |
8929ecfa YZ |
4238 | if (block_rsv->reserved >= block_rsv->size) { |
4239 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4240 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4241 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4242 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4243 | sinfo->reservation_progress++; |
8929ecfa YZ |
4244 | block_rsv->reserved = block_rsv->size; |
4245 | block_rsv->full = 1; | |
4246 | } | |
182608c8 | 4247 | |
8929ecfa | 4248 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4249 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4250 | } |
4251 | ||
f0486c68 | 4252 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4253 | { |
f0486c68 | 4254 | struct btrfs_space_info *space_info; |
6a63209f | 4255 | |
f0486c68 YZ |
4256 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4257 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4258 | |
f0486c68 | 4259 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4260 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4261 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4262 | fs_info->trans_block_rsv.space_info = space_info; |
4263 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4264 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4265 | |
8929ecfa YZ |
4266 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4267 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4268 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4269 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4270 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4271 | |
8929ecfa | 4272 | update_global_block_rsv(fs_info); |
6a63209f JB |
4273 | } |
4274 | ||
8929ecfa | 4275 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4276 | { |
8c2a3ca2 JB |
4277 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4278 | (u64)-1); | |
8929ecfa YZ |
4279 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4280 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4281 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4282 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4283 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4284 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4285 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4286 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4287 | } |
4288 | ||
a22285a6 YZ |
4289 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4290 | struct btrfs_root *root) | |
6a63209f | 4291 | { |
a22285a6 YZ |
4292 | if (!trans->bytes_reserved) |
4293 | return; | |
6a63209f | 4294 | |
e77266e4 | 4295 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4296 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4297 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4298 | trans->bytes_reserved = 0; |
4299 | } | |
6a63209f | 4300 | |
79787eaa | 4301 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4302 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4303 | struct inode *inode) | |
4304 | { | |
4305 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4306 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4307 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4308 | ||
4309 | /* | |
fcb80c2a JB |
4310 | * We need to hold space in order to delete our orphan item once we've |
4311 | * added it, so this takes the reservation so we can release it later | |
4312 | * when we are truly done with the orphan item. | |
d68fc57b | 4313 | */ |
ff5714cc | 4314 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4315 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4316 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4317 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4318 | } |
4319 | ||
d68fc57b | 4320 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4321 | { |
d68fc57b | 4322 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4323 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4324 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4325 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4326 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4327 | } | |
97e728d4 | 4328 | |
a22285a6 YZ |
4329 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4330 | struct btrfs_pending_snapshot *pending) | |
4331 | { | |
4332 | struct btrfs_root *root = pending->root; | |
4333 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4334 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4335 | /* | |
4336 | * two for root back/forward refs, two for directory entries | |
4337 | * and one for root of the snapshot. | |
4338 | */ | |
16cdcec7 | 4339 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4340 | dst_rsv->space_info = src_rsv->space_info; |
4341 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4342 | } |
4343 | ||
7709cde3 JB |
4344 | /** |
4345 | * drop_outstanding_extent - drop an outstanding extent | |
4346 | * @inode: the inode we're dropping the extent for | |
4347 | * | |
4348 | * This is called when we are freeing up an outstanding extent, either called | |
4349 | * after an error or after an extent is written. This will return the number of | |
4350 | * reserved extents that need to be freed. This must be called with | |
4351 | * BTRFS_I(inode)->lock held. | |
4352 | */ | |
9e0baf60 JB |
4353 | static unsigned drop_outstanding_extent(struct inode *inode) |
4354 | { | |
7fd2ae21 | 4355 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4356 | unsigned dropped_extents = 0; |
4357 | ||
9e0baf60 JB |
4358 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4359 | BTRFS_I(inode)->outstanding_extents--; | |
4360 | ||
7fd2ae21 | 4361 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4362 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4363 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4364 | drop_inode_space = 1; |
7fd2ae21 | 4365 | |
9e0baf60 JB |
4366 | /* |
4367 | * If we have more or the same amount of outsanding extents than we have | |
4368 | * reserved then we need to leave the reserved extents count alone. | |
4369 | */ | |
4370 | if (BTRFS_I(inode)->outstanding_extents >= | |
4371 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4372 | return drop_inode_space; |
9e0baf60 JB |
4373 | |
4374 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4375 | BTRFS_I(inode)->outstanding_extents; | |
4376 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4377 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4378 | } |
4379 | ||
7709cde3 JB |
4380 | /** |
4381 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4382 | * reserved/free'd for the given bytes. | |
4383 | * @inode: the inode we're manipulating | |
4384 | * @num_bytes: the number of bytes in question | |
4385 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4386 | * | |
4387 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4388 | * correct amount of metadata that must either be reserved or freed. We | |
4389 | * calculate how many checksums we can fit into one leaf and then divide the | |
4390 | * number of bytes that will need to be checksumed by this value to figure out | |
4391 | * how many checksums will be required. If we are adding bytes then the number | |
4392 | * may go up and we will return the number of additional bytes that must be | |
4393 | * reserved. If it is going down we will return the number of bytes that must | |
4394 | * be freed. | |
4395 | * | |
4396 | * This must be called with BTRFS_I(inode)->lock held. | |
4397 | */ | |
4398 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4399 | int reserve) | |
6324fbf3 | 4400 | { |
7709cde3 JB |
4401 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4402 | u64 csum_size; | |
4403 | int num_csums_per_leaf; | |
4404 | int num_csums; | |
4405 | int old_csums; | |
4406 | ||
4407 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4408 | BTRFS_I(inode)->csum_bytes == 0) | |
4409 | return 0; | |
4410 | ||
4411 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4412 | if (reserve) | |
4413 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4414 | else | |
4415 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4416 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4417 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4418 | sizeof(struct btrfs_csum_item) + | |
4419 | sizeof(struct btrfs_disk_key)); | |
4420 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4421 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4422 | num_csums = num_csums / num_csums_per_leaf; | |
4423 | ||
4424 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4425 | old_csums = old_csums / num_csums_per_leaf; | |
4426 | ||
4427 | /* No change, no need to reserve more */ | |
4428 | if (old_csums == num_csums) | |
4429 | return 0; | |
4430 | ||
4431 | if (reserve) | |
4432 | return btrfs_calc_trans_metadata_size(root, | |
4433 | num_csums - old_csums); | |
4434 | ||
4435 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4436 | } |
c146afad | 4437 | |
0ca1f7ce YZ |
4438 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4439 | { | |
4440 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4441 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4442 | u64 to_reserve = 0; |
660d3f6c | 4443 | u64 csum_bytes; |
9e0baf60 | 4444 | unsigned nr_extents = 0; |
660d3f6c | 4445 | int extra_reserve = 0; |
c09544e0 | 4446 | int flush = 1; |
0ca1f7ce | 4447 | int ret; |
6324fbf3 | 4448 | |
660d3f6c | 4449 | /* Need to be holding the i_mutex here if we aren't free space cache */ |
c09544e0 JB |
4450 | if (btrfs_is_free_space_inode(root, inode)) |
4451 | flush = 0; | |
4452 | ||
4453 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4454 | schedule_timeout(1); |
ec44a35c | 4455 | |
f248679e | 4456 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4457 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4458 | |
9e0baf60 JB |
4459 | spin_lock(&BTRFS_I(inode)->lock); |
4460 | BTRFS_I(inode)->outstanding_extents++; | |
4461 | ||
4462 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4463 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4464 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4465 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4466 | |
7fd2ae21 JB |
4467 | /* |
4468 | * Add an item to reserve for updating the inode when we complete the | |
4469 | * delalloc io. | |
4470 | */ | |
72ac3c0d JB |
4471 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4472 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4473 | nr_extents++; |
660d3f6c | 4474 | extra_reserve = 1; |
593060d7 | 4475 | } |
7fd2ae21 JB |
4476 | |
4477 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4478 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4479 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4480 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4481 | |
36ba022a | 4482 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4483 | if (ret) { |
7ed49f18 | 4484 | u64 to_free = 0; |
9e0baf60 | 4485 | unsigned dropped; |
7ed49f18 | 4486 | |
7709cde3 | 4487 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4488 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4489 | /* |
660d3f6c JB |
4490 | * If the inodes csum_bytes is the same as the original |
4491 | * csum_bytes then we know we haven't raced with any free()ers | |
4492 | * so we can just reduce our inodes csum bytes and carry on. | |
4493 | * Otherwise we have to do the normal free thing to account for | |
4494 | * the case that the free side didn't free up its reserve | |
4495 | * because of this outstanding reservation. | |
9e0baf60 | 4496 | */ |
660d3f6c JB |
4497 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4498 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4499 | else | |
4500 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4501 | spin_unlock(&BTRFS_I(inode)->lock); | |
4502 | if (dropped) | |
4503 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4504 | ||
8c2a3ca2 | 4505 | if (to_free) { |
7ed49f18 | 4506 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4507 | trace_btrfs_space_reservation(root->fs_info, |
4508 | "delalloc", | |
4509 | btrfs_ino(inode), | |
4510 | to_free, 0); | |
4511 | } | |
f248679e | 4512 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4513 | return ret; |
9e0baf60 | 4514 | } |
25179201 | 4515 | |
660d3f6c JB |
4516 | spin_lock(&BTRFS_I(inode)->lock); |
4517 | if (extra_reserve) { | |
72ac3c0d JB |
4518 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4519 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4520 | nr_extents--; |
4521 | } | |
4522 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4523 | spin_unlock(&BTRFS_I(inode)->lock); | |
f248679e | 4524 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
660d3f6c | 4525 | |
8c2a3ca2 JB |
4526 | if (to_reserve) |
4527 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4528 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4529 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4530 | ||
0ca1f7ce YZ |
4531 | return 0; |
4532 | } | |
4533 | ||
7709cde3 JB |
4534 | /** |
4535 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4536 | * @inode: the inode to release the reservation for | |
4537 | * @num_bytes: the number of bytes we're releasing | |
4538 | * | |
4539 | * This will release the metadata reservation for an inode. This can be called | |
4540 | * once we complete IO for a given set of bytes to release their metadata | |
4541 | * reservations. | |
4542 | */ | |
0ca1f7ce YZ |
4543 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4544 | { | |
4545 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4546 | u64 to_free = 0; |
4547 | unsigned dropped; | |
0ca1f7ce YZ |
4548 | |
4549 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4550 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4551 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4552 | |
7709cde3 JB |
4553 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4554 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4555 | if (dropped > 0) |
4556 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4557 | |
8c2a3ca2 JB |
4558 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4559 | btrfs_ino(inode), to_free, 0); | |
0ca1f7ce YZ |
4560 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4561 | to_free); | |
4562 | } | |
4563 | ||
7709cde3 JB |
4564 | /** |
4565 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4566 | * @inode: inode we're writing to | |
4567 | * @num_bytes: the number of bytes we want to allocate | |
4568 | * | |
4569 | * This will do the following things | |
4570 | * | |
4571 | * o reserve space in the data space info for num_bytes | |
4572 | * o reserve space in the metadata space info based on number of outstanding | |
4573 | * extents and how much csums will be needed | |
4574 | * o add to the inodes ->delalloc_bytes | |
4575 | * o add it to the fs_info's delalloc inodes list. | |
4576 | * | |
4577 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4578 | */ | |
0ca1f7ce YZ |
4579 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4580 | { | |
4581 | int ret; | |
4582 | ||
4583 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4584 | if (ret) |
0ca1f7ce YZ |
4585 | return ret; |
4586 | ||
4587 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4588 | if (ret) { | |
4589 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4590 | return ret; | |
4591 | } | |
4592 | ||
4593 | return 0; | |
4594 | } | |
4595 | ||
7709cde3 JB |
4596 | /** |
4597 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4598 | * @inode: inode we're releasing space for | |
4599 | * @num_bytes: the number of bytes we want to free up | |
4600 | * | |
4601 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4602 | * called in the case that we don't need the metadata AND data reservations | |
4603 | * anymore. So if there is an error or we insert an inline extent. | |
4604 | * | |
4605 | * This function will release the metadata space that was not used and will | |
4606 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4607 | * list if there are no delalloc bytes left. | |
4608 | */ | |
0ca1f7ce YZ |
4609 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4610 | { | |
4611 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4612 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4613 | } |
4614 | ||
9078a3e1 CM |
4615 | static int update_block_group(struct btrfs_trans_handle *trans, |
4616 | struct btrfs_root *root, | |
f0486c68 | 4617 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4618 | { |
0af3d00b | 4619 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4620 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4621 | u64 total = num_bytes; |
9078a3e1 | 4622 | u64 old_val; |
db94535d | 4623 | u64 byte_in_group; |
0af3d00b | 4624 | int factor; |
3e1ad54f | 4625 | |
5d4f98a2 YZ |
4626 | /* block accounting for super block */ |
4627 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4628 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4629 | if (alloc) |
4630 | old_val += num_bytes; | |
4631 | else | |
4632 | old_val -= num_bytes; | |
6c41761f | 4633 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4634 | spin_unlock(&info->delalloc_lock); |
4635 | ||
d397712b | 4636 | while (total) { |
db94535d | 4637 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4638 | if (!cache) |
79787eaa | 4639 | return -ENOENT; |
b742bb82 YZ |
4640 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4641 | BTRFS_BLOCK_GROUP_RAID1 | | |
4642 | BTRFS_BLOCK_GROUP_RAID10)) | |
4643 | factor = 2; | |
4644 | else | |
4645 | factor = 1; | |
9d66e233 JB |
4646 | /* |
4647 | * If this block group has free space cache written out, we | |
4648 | * need to make sure to load it if we are removing space. This | |
4649 | * is because we need the unpinning stage to actually add the | |
4650 | * space back to the block group, otherwise we will leak space. | |
4651 | */ | |
4652 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4653 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4654 | |
db94535d CM |
4655 | byte_in_group = bytenr - cache->key.objectid; |
4656 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4657 | |
25179201 | 4658 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4659 | spin_lock(&cache->lock); |
0af3d00b | 4660 | |
73bc1876 | 4661 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4662 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4663 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4664 | ||
0f9dd46c | 4665 | cache->dirty = 1; |
9078a3e1 | 4666 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4667 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4668 | if (alloc) { |
db94535d | 4669 | old_val += num_bytes; |
11833d66 YZ |
4670 | btrfs_set_block_group_used(&cache->item, old_val); |
4671 | cache->reserved -= num_bytes; | |
11833d66 | 4672 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4673 | cache->space_info->bytes_used += num_bytes; |
4674 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4675 | spin_unlock(&cache->lock); |
25179201 | 4676 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4677 | } else { |
db94535d | 4678 | old_val -= num_bytes; |
c286ac48 | 4679 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4680 | cache->pinned += num_bytes; |
4681 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4682 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4683 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4684 | spin_unlock(&cache->lock); |
25179201 | 4685 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4686 | |
f0486c68 YZ |
4687 | set_extent_dirty(info->pinned_extents, |
4688 | bytenr, bytenr + num_bytes - 1, | |
4689 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4690 | } |
fa9c0d79 | 4691 | btrfs_put_block_group(cache); |
db94535d CM |
4692 | total -= num_bytes; |
4693 | bytenr += num_bytes; | |
9078a3e1 CM |
4694 | } |
4695 | return 0; | |
4696 | } | |
6324fbf3 | 4697 | |
a061fc8d CM |
4698 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4699 | { | |
0f9dd46c | 4700 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4701 | u64 bytenr; |
0f9dd46c JB |
4702 | |
4703 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4704 | if (!cache) | |
a061fc8d | 4705 | return 0; |
0f9dd46c | 4706 | |
d2fb3437 | 4707 | bytenr = cache->key.objectid; |
fa9c0d79 | 4708 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4709 | |
4710 | return bytenr; | |
a061fc8d CM |
4711 | } |
4712 | ||
f0486c68 YZ |
4713 | static int pin_down_extent(struct btrfs_root *root, |
4714 | struct btrfs_block_group_cache *cache, | |
4715 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4716 | { |
11833d66 YZ |
4717 | spin_lock(&cache->space_info->lock); |
4718 | spin_lock(&cache->lock); | |
4719 | cache->pinned += num_bytes; | |
4720 | cache->space_info->bytes_pinned += num_bytes; | |
4721 | if (reserved) { | |
4722 | cache->reserved -= num_bytes; | |
4723 | cache->space_info->bytes_reserved -= num_bytes; | |
4724 | } | |
4725 | spin_unlock(&cache->lock); | |
4726 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4727 | |
f0486c68 YZ |
4728 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4729 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4730 | return 0; | |
4731 | } | |
68b38550 | 4732 | |
f0486c68 YZ |
4733 | /* |
4734 | * this function must be called within transaction | |
4735 | */ | |
4736 | int btrfs_pin_extent(struct btrfs_root *root, | |
4737 | u64 bytenr, u64 num_bytes, int reserved) | |
4738 | { | |
4739 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4740 | |
f0486c68 | 4741 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 4742 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
4743 | |
4744 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4745 | ||
4746 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4747 | return 0; |
4748 | } | |
4749 | ||
f0486c68 | 4750 | /* |
e688b725 CM |
4751 | * this function must be called within transaction |
4752 | */ | |
4753 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4754 | struct btrfs_root *root, | |
4755 | u64 bytenr, u64 num_bytes) | |
4756 | { | |
4757 | struct btrfs_block_group_cache *cache; | |
4758 | ||
4759 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 4760 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
4761 | |
4762 | /* | |
4763 | * pull in the free space cache (if any) so that our pin | |
4764 | * removes the free space from the cache. We have load_only set | |
4765 | * to one because the slow code to read in the free extents does check | |
4766 | * the pinned extents. | |
4767 | */ | |
4768 | cache_block_group(cache, trans, root, 1); | |
4769 | ||
4770 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4771 | ||
4772 | /* remove us from the free space cache (if we're there at all) */ | |
4773 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4774 | btrfs_put_block_group(cache); | |
4775 | return 0; | |
4776 | } | |
4777 | ||
fb25e914 JB |
4778 | /** |
4779 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4780 | * @cache: The cache we are manipulating | |
4781 | * @num_bytes: The number of bytes in question | |
4782 | * @reserve: One of the reservation enums | |
4783 | * | |
4784 | * This is called by the allocator when it reserves space, or by somebody who is | |
4785 | * freeing space that was never actually used on disk. For example if you | |
4786 | * reserve some space for a new leaf in transaction A and before transaction A | |
4787 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4788 | * clear the reservation. | |
4789 | * | |
4790 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4791 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4792 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4793 | * allocating less disk space for the amount of data we have reserved in the | |
4794 | * case of compression. | |
4795 | * | |
4796 | * If this is a reservation and the block group has become read only we cannot | |
4797 | * make the reservation and return -EAGAIN, otherwise this function always | |
4798 | * succeeds. | |
f0486c68 | 4799 | */ |
fb25e914 JB |
4800 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4801 | u64 num_bytes, int reserve) | |
11833d66 | 4802 | { |
fb25e914 | 4803 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4804 | int ret = 0; |
79787eaa | 4805 | |
fb25e914 JB |
4806 | spin_lock(&space_info->lock); |
4807 | spin_lock(&cache->lock); | |
4808 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4809 | if (cache->ro) { |
4810 | ret = -EAGAIN; | |
4811 | } else { | |
fb25e914 JB |
4812 | cache->reserved += num_bytes; |
4813 | space_info->bytes_reserved += num_bytes; | |
4814 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4815 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
4816 | "space_info", space_info->flags, |
4817 | num_bytes, 0); | |
fb25e914 JB |
4818 | space_info->bytes_may_use -= num_bytes; |
4819 | } | |
f0486c68 | 4820 | } |
fb25e914 JB |
4821 | } else { |
4822 | if (cache->ro) | |
4823 | space_info->bytes_readonly += num_bytes; | |
4824 | cache->reserved -= num_bytes; | |
4825 | space_info->bytes_reserved -= num_bytes; | |
4826 | space_info->reservation_progress++; | |
324ae4df | 4827 | } |
fb25e914 JB |
4828 | spin_unlock(&cache->lock); |
4829 | spin_unlock(&space_info->lock); | |
f0486c68 | 4830 | return ret; |
324ae4df | 4831 | } |
9078a3e1 | 4832 | |
143bede5 | 4833 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 4834 | struct btrfs_root *root) |
e8569813 | 4835 | { |
e8569813 | 4836 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4837 | struct btrfs_caching_control *next; |
4838 | struct btrfs_caching_control *caching_ctl; | |
4839 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4840 | |
11833d66 | 4841 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4842 | |
11833d66 YZ |
4843 | list_for_each_entry_safe(caching_ctl, next, |
4844 | &fs_info->caching_block_groups, list) { | |
4845 | cache = caching_ctl->block_group; | |
4846 | if (block_group_cache_done(cache)) { | |
4847 | cache->last_byte_to_unpin = (u64)-1; | |
4848 | list_del_init(&caching_ctl->list); | |
4849 | put_caching_control(caching_ctl); | |
e8569813 | 4850 | } else { |
11833d66 | 4851 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4852 | } |
e8569813 | 4853 | } |
11833d66 YZ |
4854 | |
4855 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4856 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4857 | else | |
4858 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4859 | ||
4860 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4861 | |
4862 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4863 | } |
4864 | ||
11833d66 | 4865 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4866 | { |
11833d66 YZ |
4867 | struct btrfs_fs_info *fs_info = root->fs_info; |
4868 | struct btrfs_block_group_cache *cache = NULL; | |
4869 | u64 len; | |
ccd467d6 | 4870 | |
11833d66 YZ |
4871 | while (start <= end) { |
4872 | if (!cache || | |
4873 | start >= cache->key.objectid + cache->key.offset) { | |
4874 | if (cache) | |
4875 | btrfs_put_block_group(cache); | |
4876 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 4877 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
4878 | } |
4879 | ||
4880 | len = cache->key.objectid + cache->key.offset - start; | |
4881 | len = min(len, end + 1 - start); | |
4882 | ||
4883 | if (start < cache->last_byte_to_unpin) { | |
4884 | len = min(len, cache->last_byte_to_unpin - start); | |
4885 | btrfs_add_free_space(cache, start, len); | |
4886 | } | |
4887 | ||
f0486c68 YZ |
4888 | start += len; |
4889 | ||
11833d66 YZ |
4890 | spin_lock(&cache->space_info->lock); |
4891 | spin_lock(&cache->lock); | |
4892 | cache->pinned -= len; | |
4893 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4894 | if (cache->ro) |
f0486c68 | 4895 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4896 | spin_unlock(&cache->lock); |
4897 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4898 | } |
11833d66 YZ |
4899 | |
4900 | if (cache) | |
4901 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4902 | return 0; |
4903 | } | |
4904 | ||
4905 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4906 | struct btrfs_root *root) |
a28ec197 | 4907 | { |
11833d66 YZ |
4908 | struct btrfs_fs_info *fs_info = root->fs_info; |
4909 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
4910 | u64 start; |
4911 | u64 end; | |
a28ec197 | 4912 | int ret; |
a28ec197 | 4913 | |
79787eaa JM |
4914 | if (trans->aborted) |
4915 | return 0; | |
4916 | ||
11833d66 YZ |
4917 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4918 | unpin = &fs_info->freed_extents[1]; | |
4919 | else | |
4920 | unpin = &fs_info->freed_extents[0]; | |
4921 | ||
d397712b | 4922 | while (1) { |
1a5bc167 CM |
4923 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4924 | EXTENT_DIRTY); | |
4925 | if (ret) | |
a28ec197 | 4926 | break; |
1f3c79a2 | 4927 | |
5378e607 LD |
4928 | if (btrfs_test_opt(root, DISCARD)) |
4929 | ret = btrfs_discard_extent(root, start, | |
4930 | end + 1 - start, NULL); | |
1f3c79a2 | 4931 | |
1a5bc167 | 4932 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4933 | unpin_extent_range(root, start, end); |
b9473439 | 4934 | cond_resched(); |
a28ec197 | 4935 | } |
817d52f8 | 4936 | |
e20d96d6 CM |
4937 | return 0; |
4938 | } | |
4939 | ||
5d4f98a2 YZ |
4940 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4941 | struct btrfs_root *root, | |
4942 | u64 bytenr, u64 num_bytes, u64 parent, | |
4943 | u64 root_objectid, u64 owner_objectid, | |
4944 | u64 owner_offset, int refs_to_drop, | |
4945 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4946 | { |
e2fa7227 | 4947 | struct btrfs_key key; |
5d4f98a2 | 4948 | struct btrfs_path *path; |
1261ec42 CM |
4949 | struct btrfs_fs_info *info = root->fs_info; |
4950 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4951 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4952 | struct btrfs_extent_item *ei; |
4953 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4954 | int ret; |
5d4f98a2 | 4955 | int is_data; |
952fccac CM |
4956 | int extent_slot = 0; |
4957 | int found_extent = 0; | |
4958 | int num_to_del = 1; | |
5d4f98a2 YZ |
4959 | u32 item_size; |
4960 | u64 refs; | |
037e6390 | 4961 | |
5caf2a00 | 4962 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4963 | if (!path) |
4964 | return -ENOMEM; | |
5f26f772 | 4965 | |
3c12ac72 | 4966 | path->reada = 1; |
b9473439 | 4967 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4968 | |
4969 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4970 | BUG_ON(!is_data && refs_to_drop != 1); | |
4971 | ||
4972 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4973 | bytenr, num_bytes, parent, | |
4974 | root_objectid, owner_objectid, | |
4975 | owner_offset); | |
7bb86316 | 4976 | if (ret == 0) { |
952fccac | 4977 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4978 | while (extent_slot >= 0) { |
4979 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4980 | extent_slot); |
5d4f98a2 | 4981 | if (key.objectid != bytenr) |
952fccac | 4982 | break; |
5d4f98a2 YZ |
4983 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4984 | key.offset == num_bytes) { | |
952fccac CM |
4985 | found_extent = 1; |
4986 | break; | |
4987 | } | |
4988 | if (path->slots[0] - extent_slot > 5) | |
4989 | break; | |
5d4f98a2 | 4990 | extent_slot--; |
952fccac | 4991 | } |
5d4f98a2 YZ |
4992 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4993 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4994 | if (found_extent && item_size < sizeof(*ei)) | |
4995 | found_extent = 0; | |
4996 | #endif | |
31840ae1 | 4997 | if (!found_extent) { |
5d4f98a2 | 4998 | BUG_ON(iref); |
56bec294 | 4999 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5000 | NULL, refs_to_drop, |
5001 | is_data); | |
79787eaa JM |
5002 | if (ret) |
5003 | goto abort; | |
b3b4aa74 | 5004 | btrfs_release_path(path); |
b9473439 | 5005 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5006 | |
5007 | key.objectid = bytenr; | |
5008 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5009 | key.offset = num_bytes; | |
5010 | ||
31840ae1 ZY |
5011 | ret = btrfs_search_slot(trans, extent_root, |
5012 | &key, path, -1, 1); | |
f3465ca4 JB |
5013 | if (ret) { |
5014 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5015 | ", was looking for %llu\n", ret, |
5016 | (unsigned long long)bytenr); | |
b783e62d JB |
5017 | if (ret > 0) |
5018 | btrfs_print_leaf(extent_root, | |
5019 | path->nodes[0]); | |
f3465ca4 | 5020 | } |
79787eaa JM |
5021 | if (ret < 0) |
5022 | goto abort; | |
31840ae1 ZY |
5023 | extent_slot = path->slots[0]; |
5024 | } | |
79787eaa | 5025 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5026 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5027 | WARN_ON(1); | |
d397712b | 5028 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5029 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5030 | (unsigned long long)bytenr, |
56bec294 | 5031 | (unsigned long long)parent, |
d397712b | 5032 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5033 | (unsigned long long)owner_objectid, |
5034 | (unsigned long long)owner_offset); | |
79787eaa JM |
5035 | } else { |
5036 | goto abort; | |
7bb86316 | 5037 | } |
5f39d397 CM |
5038 | |
5039 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5040 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5041 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5042 | if (item_size < sizeof(*ei)) { | |
5043 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5044 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5045 | owner_objectid, 0); | |
79787eaa JM |
5046 | if (ret < 0) |
5047 | goto abort; | |
5d4f98a2 | 5048 | |
b3b4aa74 | 5049 | btrfs_release_path(path); |
5d4f98a2 YZ |
5050 | path->leave_spinning = 1; |
5051 | ||
5052 | key.objectid = bytenr; | |
5053 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5054 | key.offset = num_bytes; | |
5055 | ||
5056 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5057 | -1, 1); | |
5058 | if (ret) { | |
5059 | printk(KERN_ERR "umm, got %d back from search" | |
5060 | ", was looking for %llu\n", ret, | |
5061 | (unsigned long long)bytenr); | |
5062 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5063 | } | |
79787eaa JM |
5064 | if (ret < 0) |
5065 | goto abort; | |
5d4f98a2 YZ |
5066 | extent_slot = path->slots[0]; |
5067 | leaf = path->nodes[0]; | |
5068 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5069 | } | |
5070 | #endif | |
5071 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5072 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5073 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5074 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
5075 | struct btrfs_tree_block_info *bi; | |
5076 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5077 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5078 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5079 | } | |
56bec294 | 5080 | |
5d4f98a2 | 5081 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5082 | BUG_ON(refs < refs_to_drop); |
5083 | refs -= refs_to_drop; | |
5f39d397 | 5084 | |
5d4f98a2 YZ |
5085 | if (refs > 0) { |
5086 | if (extent_op) | |
5087 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5088 | /* | |
5089 | * In the case of inline back ref, reference count will | |
5090 | * be updated by remove_extent_backref | |
952fccac | 5091 | */ |
5d4f98a2 YZ |
5092 | if (iref) { |
5093 | BUG_ON(!found_extent); | |
5094 | } else { | |
5095 | btrfs_set_extent_refs(leaf, ei, refs); | |
5096 | btrfs_mark_buffer_dirty(leaf); | |
5097 | } | |
5098 | if (found_extent) { | |
5099 | ret = remove_extent_backref(trans, extent_root, path, | |
5100 | iref, refs_to_drop, | |
5101 | is_data); | |
79787eaa JM |
5102 | if (ret) |
5103 | goto abort; | |
952fccac | 5104 | } |
5d4f98a2 | 5105 | } else { |
5d4f98a2 YZ |
5106 | if (found_extent) { |
5107 | BUG_ON(is_data && refs_to_drop != | |
5108 | extent_data_ref_count(root, path, iref)); | |
5109 | if (iref) { | |
5110 | BUG_ON(path->slots[0] != extent_slot); | |
5111 | } else { | |
5112 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5113 | path->slots[0] = extent_slot; | |
5114 | num_to_del = 2; | |
5115 | } | |
78fae27e | 5116 | } |
b9473439 | 5117 | |
952fccac CM |
5118 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5119 | num_to_del); | |
79787eaa JM |
5120 | if (ret) |
5121 | goto abort; | |
b3b4aa74 | 5122 | btrfs_release_path(path); |
21af804c | 5123 | |
5d4f98a2 | 5124 | if (is_data) { |
459931ec | 5125 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
79787eaa JM |
5126 | if (ret) |
5127 | goto abort; | |
459931ec CM |
5128 | } |
5129 | ||
f0486c68 | 5130 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
79787eaa JM |
5131 | if (ret) |
5132 | goto abort; | |
a28ec197 | 5133 | } |
79787eaa | 5134 | out: |
5caf2a00 | 5135 | btrfs_free_path(path); |
a28ec197 | 5136 | return ret; |
79787eaa JM |
5137 | |
5138 | abort: | |
5139 | btrfs_abort_transaction(trans, extent_root, ret); | |
5140 | goto out; | |
a28ec197 CM |
5141 | } |
5142 | ||
1887be66 | 5143 | /* |
f0486c68 | 5144 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5145 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5146 | * a given extent, and if there are no other delayed refs to be processed, it | |
5147 | * removes it from the tree. | |
5148 | */ | |
5149 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5150 | struct btrfs_root *root, u64 bytenr) | |
5151 | { | |
5152 | struct btrfs_delayed_ref_head *head; | |
5153 | struct btrfs_delayed_ref_root *delayed_refs; | |
5154 | struct btrfs_delayed_ref_node *ref; | |
5155 | struct rb_node *node; | |
f0486c68 | 5156 | int ret = 0; |
1887be66 CM |
5157 | |
5158 | delayed_refs = &trans->transaction->delayed_refs; | |
5159 | spin_lock(&delayed_refs->lock); | |
5160 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5161 | if (!head) | |
5162 | goto out; | |
5163 | ||
5164 | node = rb_prev(&head->node.rb_node); | |
5165 | if (!node) | |
5166 | goto out; | |
5167 | ||
5168 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5169 | ||
5170 | /* there are still entries for this ref, we can't drop it */ | |
5171 | if (ref->bytenr == bytenr) | |
5172 | goto out; | |
5173 | ||
5d4f98a2 YZ |
5174 | if (head->extent_op) { |
5175 | if (!head->must_insert_reserved) | |
5176 | goto out; | |
5177 | kfree(head->extent_op); | |
5178 | head->extent_op = NULL; | |
5179 | } | |
5180 | ||
1887be66 CM |
5181 | /* |
5182 | * waiting for the lock here would deadlock. If someone else has it | |
5183 | * locked they are already in the process of dropping it anyway | |
5184 | */ | |
5185 | if (!mutex_trylock(&head->mutex)) | |
5186 | goto out; | |
5187 | ||
5188 | /* | |
5189 | * at this point we have a head with no other entries. Go | |
5190 | * ahead and process it. | |
5191 | */ | |
5192 | head->node.in_tree = 0; | |
5193 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5194 | |
1887be66 | 5195 | delayed_refs->num_entries--; |
097b8a7c JS |
5196 | if (waitqueue_active(&root->fs_info->tree_mod_seq_wait)) |
5197 | wake_up(&root->fs_info->tree_mod_seq_wait); | |
1887be66 CM |
5198 | |
5199 | /* | |
5200 | * we don't take a ref on the node because we're removing it from the | |
5201 | * tree, so we just steal the ref the tree was holding. | |
5202 | */ | |
c3e69d58 CM |
5203 | delayed_refs->num_heads--; |
5204 | if (list_empty(&head->cluster)) | |
5205 | delayed_refs->num_heads_ready--; | |
5206 | ||
5207 | list_del_init(&head->cluster); | |
1887be66 CM |
5208 | spin_unlock(&delayed_refs->lock); |
5209 | ||
f0486c68 YZ |
5210 | BUG_ON(head->extent_op); |
5211 | if (head->must_insert_reserved) | |
5212 | ret = 1; | |
5213 | ||
5214 | mutex_unlock(&head->mutex); | |
1887be66 | 5215 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5216 | return ret; |
1887be66 CM |
5217 | out: |
5218 | spin_unlock(&delayed_refs->lock); | |
5219 | return 0; | |
5220 | } | |
5221 | ||
f0486c68 YZ |
5222 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5223 | struct btrfs_root *root, | |
5224 | struct extent_buffer *buf, | |
5581a51a | 5225 | u64 parent, int last_ref) |
f0486c68 | 5226 | { |
f0486c68 YZ |
5227 | struct btrfs_block_group_cache *cache = NULL; |
5228 | int ret; | |
5229 | ||
5230 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5231 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5232 | buf->start, buf->len, | |
5233 | parent, root->root_key.objectid, | |
5234 | btrfs_header_level(buf), | |
5581a51a | 5235 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5236 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5237 | } |
5238 | ||
5239 | if (!last_ref) | |
5240 | return; | |
5241 | ||
f0486c68 | 5242 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5243 | |
5244 | if (btrfs_header_generation(buf) == trans->transid) { | |
5245 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5246 | ret = check_ref_cleanup(trans, root, buf->start); | |
5247 | if (!ret) | |
37be25bc | 5248 | goto out; |
f0486c68 YZ |
5249 | } |
5250 | ||
5251 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5252 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5253 | goto out; |
f0486c68 YZ |
5254 | } |
5255 | ||
5256 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5257 | ||
5258 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5259 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5260 | } |
5261 | out: | |
a826d6dc JB |
5262 | /* |
5263 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5264 | * anymore. | |
5265 | */ | |
5266 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5267 | btrfs_put_block_group(cache); |
5268 | } | |
5269 | ||
79787eaa | 5270 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5271 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5272 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5273 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5274 | { |
5275 | int ret; | |
66d7e7f0 | 5276 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5277 | |
56bec294 CM |
5278 | /* |
5279 | * tree log blocks never actually go into the extent allocation | |
5280 | * tree, just update pinning info and exit early. | |
56bec294 | 5281 | */ |
5d4f98a2 YZ |
5282 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5283 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5284 | /* unlocks the pinned mutex */ |
11833d66 | 5285 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5286 | ret = 0; |
5d4f98a2 | 5287 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5288 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5289 | num_bytes, | |
5d4f98a2 | 5290 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5291 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5292 | } else { |
66d7e7f0 AJ |
5293 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5294 | num_bytes, | |
5295 | parent, root_objectid, owner, | |
5296 | offset, BTRFS_DROP_DELAYED_REF, | |
5297 | NULL, for_cow); | |
56bec294 | 5298 | } |
925baedd CM |
5299 | return ret; |
5300 | } | |
5301 | ||
87ee04eb CM |
5302 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5303 | { | |
5304 | u64 mask = ((u64)root->stripesize - 1); | |
5305 | u64 ret = (val + mask) & ~mask; | |
5306 | return ret; | |
5307 | } | |
5308 | ||
817d52f8 JB |
5309 | /* |
5310 | * when we wait for progress in the block group caching, its because | |
5311 | * our allocation attempt failed at least once. So, we must sleep | |
5312 | * and let some progress happen before we try again. | |
5313 | * | |
5314 | * This function will sleep at least once waiting for new free space to | |
5315 | * show up, and then it will check the block group free space numbers | |
5316 | * for our min num_bytes. Another option is to have it go ahead | |
5317 | * and look in the rbtree for a free extent of a given size, but this | |
5318 | * is a good start. | |
5319 | */ | |
5320 | static noinline int | |
5321 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5322 | u64 num_bytes) | |
5323 | { | |
11833d66 | 5324 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5325 | DEFINE_WAIT(wait); |
5326 | ||
11833d66 YZ |
5327 | caching_ctl = get_caching_control(cache); |
5328 | if (!caching_ctl) | |
817d52f8 | 5329 | return 0; |
817d52f8 | 5330 | |
11833d66 | 5331 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5332 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5333 | |
5334 | put_caching_control(caching_ctl); | |
5335 | return 0; | |
5336 | } | |
5337 | ||
5338 | static noinline int | |
5339 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5340 | { | |
5341 | struct btrfs_caching_control *caching_ctl; | |
5342 | DEFINE_WAIT(wait); | |
5343 | ||
5344 | caching_ctl = get_caching_control(cache); | |
5345 | if (!caching_ctl) | |
5346 | return 0; | |
5347 | ||
5348 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5349 | ||
5350 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5351 | return 0; |
5352 | } | |
5353 | ||
7738a53a | 5354 | static int __get_block_group_index(u64 flags) |
b742bb82 YZ |
5355 | { |
5356 | int index; | |
7738a53a ID |
5357 | |
5358 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
b742bb82 | 5359 | index = 0; |
7738a53a | 5360 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
b742bb82 | 5361 | index = 1; |
7738a53a | 5362 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
b742bb82 | 5363 | index = 2; |
7738a53a | 5364 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
b742bb82 YZ |
5365 | index = 3; |
5366 | else | |
5367 | index = 4; | |
7738a53a | 5368 | |
b742bb82 YZ |
5369 | return index; |
5370 | } | |
5371 | ||
7738a53a ID |
5372 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5373 | { | |
5374 | return __get_block_group_index(cache->flags); | |
5375 | } | |
5376 | ||
817d52f8 | 5377 | enum btrfs_loop_type { |
285ff5af JB |
5378 | LOOP_CACHING_NOWAIT = 0, |
5379 | LOOP_CACHING_WAIT = 1, | |
5380 | LOOP_ALLOC_CHUNK = 2, | |
5381 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5382 | }; |
5383 | ||
fec577fb CM |
5384 | /* |
5385 | * walks the btree of allocated extents and find a hole of a given size. | |
5386 | * The key ins is changed to record the hole: | |
5387 | * ins->objectid == block start | |
62e2749e | 5388 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5389 | * ins->offset == number of blocks |
5390 | * Any available blocks before search_start are skipped. | |
5391 | */ | |
d397712b | 5392 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5393 | struct btrfs_root *orig_root, |
5394 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5395 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5396 | u64 data) |
fec577fb | 5397 | { |
80eb234a | 5398 | int ret = 0; |
d397712b | 5399 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5400 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5401 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5402 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5403 | u64 search_start = 0; |
239b14b3 | 5404 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5405 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5406 | int done_chunk_alloc = 0; |
80eb234a | 5407 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5408 | int loop = 0; |
f0486c68 | 5409 | int index = 0; |
fb25e914 JB |
5410 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5411 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5412 | bool found_uncached_bg = false; |
0a24325e | 5413 | bool failed_cluster_refill = false; |
1cdda9b8 | 5414 | bool failed_alloc = false; |
67377734 | 5415 | bool use_cluster = true; |
60d2adbb | 5416 | bool have_caching_bg = false; |
fec577fb | 5417 | |
db94535d | 5418 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5419 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5420 | ins->objectid = 0; |
5421 | ins->offset = 0; | |
b1a4d965 | 5422 | |
3f7de037 JB |
5423 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5424 | ||
2552d17e | 5425 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5426 | if (!space_info) { |
e0f54067 | 5427 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5428 | return -ENOSPC; |
5429 | } | |
2552d17e | 5430 | |
67377734 JB |
5431 | /* |
5432 | * If the space info is for both data and metadata it means we have a | |
5433 | * small filesystem and we can't use the clustering stuff. | |
5434 | */ | |
5435 | if (btrfs_mixed_space_info(space_info)) | |
5436 | use_cluster = false; | |
5437 | ||
0ef3e66b CM |
5438 | if (orig_root->ref_cows || empty_size) |
5439 | allowed_chunk_alloc = 1; | |
5440 | ||
67377734 | 5441 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5442 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5443 | if (!btrfs_test_opt(root, SSD)) |
5444 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5445 | } |
5446 | ||
67377734 JB |
5447 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5448 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5449 | last_ptr = &root->fs_info->data_alloc_cluster; |
5450 | } | |
0f9dd46c | 5451 | |
239b14b3 | 5452 | if (last_ptr) { |
fa9c0d79 CM |
5453 | spin_lock(&last_ptr->lock); |
5454 | if (last_ptr->block_group) | |
5455 | hint_byte = last_ptr->window_start; | |
5456 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5457 | } |
fa9c0d79 | 5458 | |
a061fc8d | 5459 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5460 | search_start = max(search_start, hint_byte); |
0b86a832 | 5461 | |
817d52f8 | 5462 | if (!last_ptr) |
fa9c0d79 | 5463 | empty_cluster = 0; |
fa9c0d79 | 5464 | |
2552d17e | 5465 | if (search_start == hint_byte) { |
2552d17e JB |
5466 | block_group = btrfs_lookup_block_group(root->fs_info, |
5467 | search_start); | |
274bd4fb | 5468 | used_block_group = block_group; |
817d52f8 JB |
5469 | /* |
5470 | * we don't want to use the block group if it doesn't match our | |
5471 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5472 | * |
5473 | * However if we are re-searching with an ideal block group | |
5474 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5475 | */ |
5476 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5477 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5478 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5479 | if (list_empty(&block_group->list) || |
5480 | block_group->ro) { | |
5481 | /* | |
5482 | * someone is removing this block group, | |
5483 | * we can't jump into the have_block_group | |
5484 | * target because our list pointers are not | |
5485 | * valid | |
5486 | */ | |
5487 | btrfs_put_block_group(block_group); | |
5488 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5489 | } else { |
b742bb82 | 5490 | index = get_block_group_index(block_group); |
44fb5511 | 5491 | goto have_block_group; |
ccf0e725 | 5492 | } |
2552d17e | 5493 | } else if (block_group) { |
fa9c0d79 | 5494 | btrfs_put_block_group(block_group); |
2552d17e | 5495 | } |
42e70e7a | 5496 | } |
2552d17e | 5497 | search: |
60d2adbb | 5498 | have_caching_bg = false; |
80eb234a | 5499 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5500 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5501 | list) { | |
6226cb0a | 5502 | u64 offset; |
817d52f8 | 5503 | int cached; |
8a1413a2 | 5504 | |
274bd4fb | 5505 | used_block_group = block_group; |
11dfe35a | 5506 | btrfs_get_block_group(block_group); |
2552d17e | 5507 | search_start = block_group->key.objectid; |
42e70e7a | 5508 | |
83a50de9 CM |
5509 | /* |
5510 | * this can happen if we end up cycling through all the | |
5511 | * raid types, but we want to make sure we only allocate | |
5512 | * for the proper type. | |
5513 | */ | |
5514 | if (!block_group_bits(block_group, data)) { | |
5515 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5516 | BTRFS_BLOCK_GROUP_RAID1 | | |
5517 | BTRFS_BLOCK_GROUP_RAID10; | |
5518 | ||
5519 | /* | |
5520 | * if they asked for extra copies and this block group | |
5521 | * doesn't provide them, bail. This does allow us to | |
5522 | * fill raid0 from raid1. | |
5523 | */ | |
5524 | if ((data & extra) && !(block_group->flags & extra)) | |
5525 | goto loop; | |
5526 | } | |
5527 | ||
2552d17e | 5528 | have_block_group: |
291c7d2f JB |
5529 | cached = block_group_cache_done(block_group); |
5530 | if (unlikely(!cached)) { | |
291c7d2f | 5531 | found_uncached_bg = true; |
b8399dee | 5532 | ret = cache_block_group(block_group, trans, |
285ff5af | 5533 | orig_root, 0); |
1d4284bd CM |
5534 | BUG_ON(ret < 0); |
5535 | ret = 0; | |
817d52f8 JB |
5536 | } |
5537 | ||
ea6a478e | 5538 | if (unlikely(block_group->ro)) |
2552d17e | 5539 | goto loop; |
0f9dd46c | 5540 | |
0a24325e | 5541 | /* |
062c05c4 AO |
5542 | * Ok we want to try and use the cluster allocator, so |
5543 | * lets look there | |
0a24325e | 5544 | */ |
062c05c4 | 5545 | if (last_ptr) { |
fa9c0d79 CM |
5546 | /* |
5547 | * the refill lock keeps out other | |
5548 | * people trying to start a new cluster | |
5549 | */ | |
5550 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5551 | used_block_group = last_ptr->block_group; |
5552 | if (used_block_group != block_group && | |
5553 | (!used_block_group || | |
5554 | used_block_group->ro || | |
5555 | !block_group_bits(used_block_group, data))) { | |
5556 | used_block_group = block_group; | |
44fb5511 | 5557 | goto refill_cluster; |
274bd4fb AO |
5558 | } |
5559 | ||
5560 | if (used_block_group != block_group) | |
5561 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5562 | |
274bd4fb AO |
5563 | offset = btrfs_alloc_from_cluster(used_block_group, |
5564 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5565 | if (offset) { |
5566 | /* we have a block, we're done */ | |
5567 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5568 | trace_btrfs_reserve_extent_cluster(root, |
5569 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5570 | goto checks; |
5571 | } | |
5572 | ||
274bd4fb AO |
5573 | WARN_ON(last_ptr->block_group != used_block_group); |
5574 | if (used_block_group != block_group) { | |
5575 | btrfs_put_block_group(used_block_group); | |
5576 | used_block_group = block_group; | |
fa9c0d79 | 5577 | } |
44fb5511 | 5578 | refill_cluster: |
274bd4fb | 5579 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5580 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5581 | * set up a new clusters, so lets just skip it | |
5582 | * and let the allocator find whatever block | |
5583 | * it can find. If we reach this point, we | |
5584 | * will have tried the cluster allocator | |
5585 | * plenty of times and not have found | |
5586 | * anything, so we are likely way too | |
5587 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5588 | * anything. |
5589 | * | |
5590 | * However, if the cluster is taken from the | |
5591 | * current block group, release the cluster | |
5592 | * first, so that we stand a better chance of | |
5593 | * succeeding in the unclustered | |
5594 | * allocation. */ | |
5595 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5596 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5597 | spin_unlock(&last_ptr->refill_lock); |
5598 | goto unclustered_alloc; | |
5599 | } | |
5600 | ||
fa9c0d79 CM |
5601 | /* |
5602 | * this cluster didn't work out, free it and | |
5603 | * start over | |
5604 | */ | |
5605 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5606 | ||
a5f6f719 AO |
5607 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5608 | spin_unlock(&last_ptr->refill_lock); | |
5609 | goto unclustered_alloc; | |
5610 | } | |
5611 | ||
fa9c0d79 | 5612 | /* allocate a cluster in this block group */ |
451d7585 | 5613 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5614 | block_group, last_ptr, |
1b22bad7 | 5615 | search_start, num_bytes, |
fa9c0d79 CM |
5616 | empty_cluster + empty_size); |
5617 | if (ret == 0) { | |
5618 | /* | |
5619 | * now pull our allocation out of this | |
5620 | * cluster | |
5621 | */ | |
5622 | offset = btrfs_alloc_from_cluster(block_group, | |
5623 | last_ptr, num_bytes, | |
5624 | search_start); | |
5625 | if (offset) { | |
5626 | /* we found one, proceed */ | |
5627 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5628 | trace_btrfs_reserve_extent_cluster(root, |
5629 | block_group, search_start, | |
5630 | num_bytes); | |
fa9c0d79 CM |
5631 | goto checks; |
5632 | } | |
0a24325e JB |
5633 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5634 | && !failed_cluster_refill) { | |
817d52f8 JB |
5635 | spin_unlock(&last_ptr->refill_lock); |
5636 | ||
0a24325e | 5637 | failed_cluster_refill = true; |
817d52f8 JB |
5638 | wait_block_group_cache_progress(block_group, |
5639 | num_bytes + empty_cluster + empty_size); | |
5640 | goto have_block_group; | |
fa9c0d79 | 5641 | } |
817d52f8 | 5642 | |
fa9c0d79 CM |
5643 | /* |
5644 | * at this point we either didn't find a cluster | |
5645 | * or we weren't able to allocate a block from our | |
5646 | * cluster. Free the cluster we've been trying | |
5647 | * to use, and go to the next block group | |
5648 | */ | |
0a24325e | 5649 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5650 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5651 | goto loop; |
fa9c0d79 CM |
5652 | } |
5653 | ||
062c05c4 | 5654 | unclustered_alloc: |
a5f6f719 AO |
5655 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5656 | if (cached && | |
5657 | block_group->free_space_ctl->free_space < | |
5658 | num_bytes + empty_cluster + empty_size) { | |
5659 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5660 | goto loop; | |
5661 | } | |
5662 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5663 | ||
6226cb0a JB |
5664 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5665 | num_bytes, empty_size); | |
1cdda9b8 JB |
5666 | /* |
5667 | * If we didn't find a chunk, and we haven't failed on this | |
5668 | * block group before, and this block group is in the middle of | |
5669 | * caching and we are ok with waiting, then go ahead and wait | |
5670 | * for progress to be made, and set failed_alloc to true. | |
5671 | * | |
5672 | * If failed_alloc is true then we've already waited on this | |
5673 | * block group once and should move on to the next block group. | |
5674 | */ | |
5675 | if (!offset && !failed_alloc && !cached && | |
5676 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5677 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5678 | num_bytes + empty_size); |
5679 | failed_alloc = true; | |
817d52f8 | 5680 | goto have_block_group; |
1cdda9b8 | 5681 | } else if (!offset) { |
60d2adbb MX |
5682 | if (!cached) |
5683 | have_caching_bg = true; | |
1cdda9b8 | 5684 | goto loop; |
817d52f8 | 5685 | } |
fa9c0d79 | 5686 | checks: |
6226cb0a | 5687 | search_start = stripe_align(root, offset); |
25179201 | 5688 | |
2552d17e JB |
5689 | /* move on to the next group */ |
5690 | if (search_start + num_bytes > | |
274bd4fb AO |
5691 | used_block_group->key.objectid + used_block_group->key.offset) { |
5692 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5693 | goto loop; |
6226cb0a | 5694 | } |
f5a31e16 | 5695 | |
f0486c68 | 5696 | if (offset < search_start) |
274bd4fb | 5697 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5698 | search_start - offset); |
5699 | BUG_ON(offset > search_start); | |
2552d17e | 5700 | |
274bd4fb | 5701 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5702 | alloc_type); |
f0486c68 | 5703 | if (ret == -EAGAIN) { |
274bd4fb | 5704 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5705 | goto loop; |
0f9dd46c | 5706 | } |
0b86a832 | 5707 | |
f0486c68 | 5708 | /* we are all good, lets return */ |
2552d17e JB |
5709 | ins->objectid = search_start; |
5710 | ins->offset = num_bytes; | |
d2fb3437 | 5711 | |
3f7de037 JB |
5712 | trace_btrfs_reserve_extent(orig_root, block_group, |
5713 | search_start, num_bytes); | |
6226cb0a | 5714 | if (offset < search_start) |
274bd4fb | 5715 | btrfs_add_free_space(used_block_group, offset, |
6226cb0a JB |
5716 | search_start - offset); |
5717 | BUG_ON(offset > search_start); | |
274bd4fb AO |
5718 | if (used_block_group != block_group) |
5719 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5720 | btrfs_put_block_group(block_group); |
2552d17e JB |
5721 | break; |
5722 | loop: | |
0a24325e | 5723 | failed_cluster_refill = false; |
1cdda9b8 | 5724 | failed_alloc = false; |
b742bb82 | 5725 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5726 | if (used_block_group != block_group) |
5727 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5728 | btrfs_put_block_group(block_group); |
2552d17e JB |
5729 | } |
5730 | up_read(&space_info->groups_sem); | |
5731 | ||
60d2adbb MX |
5732 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5733 | goto search; | |
5734 | ||
b742bb82 YZ |
5735 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5736 | goto search; | |
5737 | ||
285ff5af | 5738 | /* |
ccf0e725 JB |
5739 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5740 | * caching kthreads as we move along | |
817d52f8 JB |
5741 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5742 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5743 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5744 | * again | |
fa9c0d79 | 5745 | */ |
723bda20 | 5746 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5747 | index = 0; |
723bda20 | 5748 | loop++; |
817d52f8 | 5749 | if (loop == LOOP_ALLOC_CHUNK) { |
723bda20 JB |
5750 | if (allowed_chunk_alloc) { |
5751 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5752 | 2 * 1024 * 1024, data, | |
5753 | CHUNK_ALLOC_LIMITED); | |
79787eaa JM |
5754 | if (ret < 0) { |
5755 | btrfs_abort_transaction(trans, | |
5756 | root, ret); | |
5757 | goto out; | |
5758 | } | |
723bda20 JB |
5759 | allowed_chunk_alloc = 0; |
5760 | if (ret == 1) | |
5761 | done_chunk_alloc = 1; | |
5762 | } else if (!done_chunk_alloc && | |
5763 | space_info->force_alloc == | |
5764 | CHUNK_ALLOC_NO_FORCE) { | |
5765 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5766 | } | |
2552d17e | 5767 | |
723bda20 JB |
5768 | /* |
5769 | * We didn't allocate a chunk, go ahead and drop the | |
5770 | * empty size and loop again. | |
5771 | */ | |
5772 | if (!done_chunk_alloc) | |
5773 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5774 | } |
5775 | ||
723bda20 JB |
5776 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5777 | empty_size = 0; | |
5778 | empty_cluster = 0; | |
fa9c0d79 | 5779 | } |
723bda20 JB |
5780 | |
5781 | goto search; | |
2552d17e JB |
5782 | } else if (!ins->objectid) { |
5783 | ret = -ENOSPC; | |
d82a6f1d | 5784 | } else if (ins->objectid) { |
80eb234a | 5785 | ret = 0; |
be744175 | 5786 | } |
79787eaa | 5787 | out: |
be744175 | 5788 | |
0f70abe2 | 5789 | return ret; |
fec577fb | 5790 | } |
ec44a35c | 5791 | |
9ed74f2d JB |
5792 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5793 | int dump_block_groups) | |
0f9dd46c JB |
5794 | { |
5795 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5796 | int index = 0; |
0f9dd46c | 5797 | |
9ed74f2d | 5798 | spin_lock(&info->lock); |
fb25e914 JB |
5799 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5800 | (unsigned long long)info->flags, | |
d397712b | 5801 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5802 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5803 | info->bytes_readonly), |
d397712b | 5804 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5805 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5806 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5807 | (unsigned long long)info->total_bytes, |
8929ecfa | 5808 | (unsigned long long)info->bytes_used, |
21380931 | 5809 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5810 | (unsigned long long)info->bytes_reserved, |
21380931 | 5811 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5812 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5813 | spin_unlock(&info->lock); |
5814 | ||
5815 | if (!dump_block_groups) | |
5816 | return; | |
0f9dd46c | 5817 | |
80eb234a | 5818 | down_read(&info->groups_sem); |
b742bb82 YZ |
5819 | again: |
5820 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5821 | spin_lock(&cache->lock); |
d397712b CM |
5822 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5823 | "%llu pinned %llu reserved\n", | |
5824 | (unsigned long long)cache->key.objectid, | |
5825 | (unsigned long long)cache->key.offset, | |
5826 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5827 | (unsigned long long)cache->pinned, | |
5828 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5829 | btrfs_dump_free_space(cache, bytes); |
5830 | spin_unlock(&cache->lock); | |
5831 | } | |
b742bb82 YZ |
5832 | if (++index < BTRFS_NR_RAID_TYPES) |
5833 | goto again; | |
80eb234a | 5834 | up_read(&info->groups_sem); |
0f9dd46c | 5835 | } |
e8569813 | 5836 | |
11833d66 YZ |
5837 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5838 | struct btrfs_root *root, | |
5839 | u64 num_bytes, u64 min_alloc_size, | |
5840 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 5841 | struct btrfs_key *ins, u64 data) |
fec577fb | 5842 | { |
9e622d6b | 5843 | bool final_tried = false; |
fec577fb | 5844 | int ret; |
925baedd | 5845 | |
6a63209f | 5846 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5847 | again: |
0ef3e66b CM |
5848 | /* |
5849 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5850 | * is not called recursively on allocations | |
5851 | */ | |
79787eaa | 5852 | if (empty_size || root->ref_cows) { |
6324fbf3 | 5853 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5854 | num_bytes + 2 * 1024 * 1024, data, |
5855 | CHUNK_ALLOC_NO_FORCE); | |
79787eaa JM |
5856 | if (ret < 0 && ret != -ENOSPC) { |
5857 | btrfs_abort_transaction(trans, root, ret); | |
5858 | return ret; | |
5859 | } | |
5860 | } | |
0b86a832 | 5861 | |
db94535d CM |
5862 | WARN_ON(num_bytes < root->sectorsize); |
5863 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 5864 | hint_byte, ins, data); |
3b951516 | 5865 | |
9e622d6b MX |
5866 | if (ret == -ENOSPC) { |
5867 | if (!final_tried) { | |
5868 | num_bytes = num_bytes >> 1; | |
5869 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
5870 | num_bytes = max(num_bytes, min_alloc_size); | |
79787eaa | 5871 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
9e622d6b | 5872 | num_bytes, data, CHUNK_ALLOC_FORCE); |
79787eaa JM |
5873 | if (ret < 0 && ret != -ENOSPC) { |
5874 | btrfs_abort_transaction(trans, root, ret); | |
5875 | return ret; | |
5876 | } | |
9e622d6b MX |
5877 | if (num_bytes == min_alloc_size) |
5878 | final_tried = true; | |
5879 | goto again; | |
5880 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
5881 | struct btrfs_space_info *sinfo; | |
5882 | ||
5883 | sinfo = __find_space_info(root->fs_info, data); | |
5884 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
5885 | "wanted %llu\n", (unsigned long long)data, | |
5886 | (unsigned long long)num_bytes); | |
53804280 JM |
5887 | if (sinfo) |
5888 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 5889 | } |
925baedd | 5890 | } |
0f9dd46c | 5891 | |
1abe9b8a | 5892 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5893 | ||
0f9dd46c | 5894 | return ret; |
e6dcd2dc CM |
5895 | } |
5896 | ||
e688b725 CM |
5897 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
5898 | u64 start, u64 len, int pin) | |
65b51a00 | 5899 | { |
0f9dd46c | 5900 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5901 | int ret = 0; |
0f9dd46c | 5902 | |
0f9dd46c JB |
5903 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5904 | if (!cache) { | |
d397712b CM |
5905 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5906 | (unsigned long long)start); | |
0f9dd46c JB |
5907 | return -ENOSPC; |
5908 | } | |
1f3c79a2 | 5909 | |
5378e607 LD |
5910 | if (btrfs_test_opt(root, DISCARD)) |
5911 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5912 | |
e688b725 CM |
5913 | if (pin) |
5914 | pin_down_extent(root, cache, start, len, 1); | |
5915 | else { | |
5916 | btrfs_add_free_space(cache, start, len); | |
5917 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
5918 | } | |
fa9c0d79 | 5919 | btrfs_put_block_group(cache); |
817d52f8 | 5920 | |
1abe9b8a | 5921 | trace_btrfs_reserved_extent_free(root, start, len); |
5922 | ||
e6dcd2dc CM |
5923 | return ret; |
5924 | } | |
5925 | ||
e688b725 CM |
5926 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
5927 | u64 start, u64 len) | |
5928 | { | |
5929 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
5930 | } | |
5931 | ||
5932 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
5933 | u64 start, u64 len) | |
5934 | { | |
5935 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
5936 | } | |
5937 | ||
5d4f98a2 YZ |
5938 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5939 | struct btrfs_root *root, | |
5940 | u64 parent, u64 root_objectid, | |
5941 | u64 flags, u64 owner, u64 offset, | |
5942 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5943 | { |
5944 | int ret; | |
5d4f98a2 | 5945 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5946 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5947 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5948 | struct btrfs_path *path; |
5d4f98a2 YZ |
5949 | struct extent_buffer *leaf; |
5950 | int type; | |
5951 | u32 size; | |
26b8003f | 5952 | |
5d4f98a2 YZ |
5953 | if (parent > 0) |
5954 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5955 | else | |
5956 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5957 | |
5d4f98a2 | 5958 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5959 | |
5960 | path = btrfs_alloc_path(); | |
db5b493a TI |
5961 | if (!path) |
5962 | return -ENOMEM; | |
47e4bb98 | 5963 | |
b9473439 | 5964 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5965 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5966 | ins, size); | |
79787eaa JM |
5967 | if (ret) { |
5968 | btrfs_free_path(path); | |
5969 | return ret; | |
5970 | } | |
0f9dd46c | 5971 | |
5d4f98a2 YZ |
5972 | leaf = path->nodes[0]; |
5973 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5974 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5975 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5976 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5977 | btrfs_set_extent_flags(leaf, extent_item, | |
5978 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5979 | ||
5980 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5981 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5982 | if (parent > 0) { | |
5983 | struct btrfs_shared_data_ref *ref; | |
5984 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5985 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5986 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5987 | } else { | |
5988 | struct btrfs_extent_data_ref *ref; | |
5989 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5990 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5991 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5992 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5993 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5994 | } | |
47e4bb98 CM |
5995 | |
5996 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5997 | btrfs_free_path(path); |
f510cfec | 5998 | |
f0486c68 | 5999 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6000 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
6001 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6002 | "%llu\n", (unsigned long long)ins->objectid, | |
6003 | (unsigned long long)ins->offset); | |
f5947066 CM |
6004 | BUG(); |
6005 | } | |
e6dcd2dc CM |
6006 | return ret; |
6007 | } | |
6008 | ||
5d4f98a2 YZ |
6009 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6010 | struct btrfs_root *root, | |
6011 | u64 parent, u64 root_objectid, | |
6012 | u64 flags, struct btrfs_disk_key *key, | |
6013 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6014 | { |
6015 | int ret; | |
5d4f98a2 YZ |
6016 | struct btrfs_fs_info *fs_info = root->fs_info; |
6017 | struct btrfs_extent_item *extent_item; | |
6018 | struct btrfs_tree_block_info *block_info; | |
6019 | struct btrfs_extent_inline_ref *iref; | |
6020 | struct btrfs_path *path; | |
6021 | struct extent_buffer *leaf; | |
6022 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 6023 | |
5d4f98a2 | 6024 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6025 | if (!path) |
6026 | return -ENOMEM; | |
56bec294 | 6027 | |
5d4f98a2 YZ |
6028 | path->leave_spinning = 1; |
6029 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6030 | ins, size); | |
79787eaa JM |
6031 | if (ret) { |
6032 | btrfs_free_path(path); | |
6033 | return ret; | |
6034 | } | |
5d4f98a2 YZ |
6035 | |
6036 | leaf = path->nodes[0]; | |
6037 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6038 | struct btrfs_extent_item); | |
6039 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6040 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6041 | btrfs_set_extent_flags(leaf, extent_item, | |
6042 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
6043 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6044 | ||
6045 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6046 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6047 | ||
6048 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6049 | if (parent > 0) { | |
6050 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6051 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6052 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6053 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6054 | } else { | |
6055 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6056 | BTRFS_TREE_BLOCK_REF_KEY); | |
6057 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6058 | } | |
6059 | ||
6060 | btrfs_mark_buffer_dirty(leaf); | |
6061 | btrfs_free_path(path); | |
6062 | ||
f0486c68 | 6063 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6064 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6065 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6066 | "%llu\n", (unsigned long long)ins->objectid, | |
6067 | (unsigned long long)ins->offset); | |
6068 | BUG(); | |
6069 | } | |
6070 | return ret; | |
6071 | } | |
6072 | ||
6073 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6074 | struct btrfs_root *root, | |
6075 | u64 root_objectid, u64 owner, | |
6076 | u64 offset, struct btrfs_key *ins) | |
6077 | { | |
6078 | int ret; | |
6079 | ||
6080 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6081 | ||
66d7e7f0 AJ |
6082 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6083 | ins->offset, 0, | |
6084 | root_objectid, owner, offset, | |
6085 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6086 | return ret; |
6087 | } | |
e02119d5 CM |
6088 | |
6089 | /* | |
6090 | * this is used by the tree logging recovery code. It records that | |
6091 | * an extent has been allocated and makes sure to clear the free | |
6092 | * space cache bits as well | |
6093 | */ | |
5d4f98a2 YZ |
6094 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6095 | struct btrfs_root *root, | |
6096 | u64 root_objectid, u64 owner, u64 offset, | |
6097 | struct btrfs_key *ins) | |
e02119d5 CM |
6098 | { |
6099 | int ret; | |
6100 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6101 | struct btrfs_caching_control *caching_ctl; |
6102 | u64 start = ins->objectid; | |
6103 | u64 num_bytes = ins->offset; | |
e02119d5 | 6104 | |
e02119d5 | 6105 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 6106 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 6107 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6108 | |
11833d66 YZ |
6109 | if (!caching_ctl) { |
6110 | BUG_ON(!block_group_cache_done(block_group)); | |
6111 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6112 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6113 | } else { |
6114 | mutex_lock(&caching_ctl->mutex); | |
6115 | ||
6116 | if (start >= caching_ctl->progress) { | |
6117 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6118 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6119 | } else if (start + num_bytes <= caching_ctl->progress) { |
6120 | ret = btrfs_remove_free_space(block_group, | |
6121 | start, num_bytes); | |
79787eaa | 6122 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6123 | } else { |
6124 | num_bytes = caching_ctl->progress - start; | |
6125 | ret = btrfs_remove_free_space(block_group, | |
6126 | start, num_bytes); | |
79787eaa | 6127 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6128 | |
6129 | start = caching_ctl->progress; | |
6130 | num_bytes = ins->objectid + ins->offset - | |
6131 | caching_ctl->progress; | |
6132 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6133 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6134 | } |
6135 | ||
6136 | mutex_unlock(&caching_ctl->mutex); | |
6137 | put_caching_control(caching_ctl); | |
6138 | } | |
6139 | ||
fb25e914 JB |
6140 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6141 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6142 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6143 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6144 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6145 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6146 | return ret; |
6147 | } | |
6148 | ||
65b51a00 CM |
6149 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6150 | struct btrfs_root *root, | |
4008c04a CM |
6151 | u64 bytenr, u32 blocksize, |
6152 | int level) | |
65b51a00 CM |
6153 | { |
6154 | struct extent_buffer *buf; | |
6155 | ||
6156 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6157 | if (!buf) | |
6158 | return ERR_PTR(-ENOMEM); | |
6159 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6160 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6161 | btrfs_tree_lock(buf); |
6162 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6163 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6164 | |
6165 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6166 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6167 | |
d0c803c4 | 6168 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6169 | /* |
6170 | * we allow two log transactions at a time, use different | |
6171 | * EXENT bit to differentiate dirty pages. | |
6172 | */ | |
6173 | if (root->log_transid % 2 == 0) | |
6174 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6175 | buf->start + buf->len - 1, GFP_NOFS); | |
6176 | else | |
6177 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6178 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6179 | } else { |
6180 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6181 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6182 | } |
65b51a00 | 6183 | trans->blocks_used++; |
b4ce94de | 6184 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6185 | return buf; |
6186 | } | |
6187 | ||
f0486c68 YZ |
6188 | static struct btrfs_block_rsv * |
6189 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6190 | struct btrfs_root *root, u32 blocksize) | |
6191 | { | |
6192 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6193 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6194 | int ret; |
6195 | ||
6196 | block_rsv = get_block_rsv(trans, root); | |
6197 | ||
6198 | if (block_rsv->size == 0) { | |
36ba022a | 6199 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
6200 | /* |
6201 | * If we couldn't reserve metadata bytes try and use some from | |
6202 | * the global reserve. | |
6203 | */ | |
6204 | if (ret && block_rsv != global_rsv) { | |
6205 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6206 | if (!ret) | |
6207 | return global_rsv; | |
f0486c68 | 6208 | return ERR_PTR(ret); |
68a82277 | 6209 | } else if (ret) { |
f0486c68 | 6210 | return ERR_PTR(ret); |
68a82277 | 6211 | } |
f0486c68 YZ |
6212 | return block_rsv; |
6213 | } | |
6214 | ||
6215 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6216 | if (!ret) | |
6217 | return block_rsv; | |
68a82277 | 6218 | if (ret) { |
dff51cd1 DS |
6219 | static DEFINE_RATELIMIT_STATE(_rs, |
6220 | DEFAULT_RATELIMIT_INTERVAL, | |
6221 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
6222 | if (__ratelimit(&_rs)) { | |
6223 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
6224 | WARN_ON(1); | |
6225 | } | |
36ba022a | 6226 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 6227 | if (!ret) { |
68a82277 JB |
6228 | return block_rsv; |
6229 | } else if (ret && block_rsv != global_rsv) { | |
6230 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6231 | if (!ret) | |
6232 | return global_rsv; | |
6233 | } | |
6234 | } | |
f0486c68 | 6235 | |
f0486c68 YZ |
6236 | return ERR_PTR(-ENOSPC); |
6237 | } | |
6238 | ||
8c2a3ca2 JB |
6239 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6240 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6241 | { |
6242 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6243 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6244 | } |
6245 | ||
fec577fb | 6246 | /* |
f0486c68 YZ |
6247 | * finds a free extent and does all the dirty work required for allocation |
6248 | * returns the key for the extent through ins, and a tree buffer for | |
6249 | * the first block of the extent through buf. | |
6250 | * | |
fec577fb CM |
6251 | * returns the tree buffer or NULL. |
6252 | */ | |
5f39d397 | 6253 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6254 | struct btrfs_root *root, u32 blocksize, |
6255 | u64 parent, u64 root_objectid, | |
6256 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6257 | u64 hint, u64 empty_size) |
fec577fb | 6258 | { |
e2fa7227 | 6259 | struct btrfs_key ins; |
f0486c68 | 6260 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6261 | struct extent_buffer *buf; |
f0486c68 YZ |
6262 | u64 flags = 0; |
6263 | int ret; | |
6264 | ||
fec577fb | 6265 | |
f0486c68 YZ |
6266 | block_rsv = use_block_rsv(trans, root, blocksize); |
6267 | if (IS_ERR(block_rsv)) | |
6268 | return ERR_CAST(block_rsv); | |
6269 | ||
6270 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6271 | empty_size, hint, &ins, 0); |
fec577fb | 6272 | if (ret) { |
8c2a3ca2 | 6273 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6274 | return ERR_PTR(ret); |
fec577fb | 6275 | } |
55c69072 | 6276 | |
4008c04a CM |
6277 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6278 | blocksize, level); | |
79787eaa | 6279 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6280 | |
6281 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6282 | if (parent == 0) | |
6283 | parent = ins.objectid; | |
6284 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6285 | } else | |
6286 | BUG_ON(parent > 0); | |
6287 | ||
6288 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6289 | struct btrfs_delayed_extent_op *extent_op; | |
6290 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
79787eaa | 6291 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6292 | if (key) |
6293 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6294 | else | |
6295 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6296 | extent_op->flags_to_set = flags; | |
6297 | extent_op->update_key = 1; | |
6298 | extent_op->update_flags = 1; | |
6299 | extent_op->is_data = 0; | |
6300 | ||
66d7e7f0 AJ |
6301 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6302 | ins.objectid, | |
f0486c68 YZ |
6303 | ins.offset, parent, root_objectid, |
6304 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6305 | extent_op, 0); |
79787eaa | 6306 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6307 | } |
fec577fb CM |
6308 | return buf; |
6309 | } | |
a28ec197 | 6310 | |
2c47e605 YZ |
6311 | struct walk_control { |
6312 | u64 refs[BTRFS_MAX_LEVEL]; | |
6313 | u64 flags[BTRFS_MAX_LEVEL]; | |
6314 | struct btrfs_key update_progress; | |
6315 | int stage; | |
6316 | int level; | |
6317 | int shared_level; | |
6318 | int update_ref; | |
6319 | int keep_locks; | |
1c4850e2 YZ |
6320 | int reada_slot; |
6321 | int reada_count; | |
66d7e7f0 | 6322 | int for_reloc; |
2c47e605 YZ |
6323 | }; |
6324 | ||
6325 | #define DROP_REFERENCE 1 | |
6326 | #define UPDATE_BACKREF 2 | |
6327 | ||
1c4850e2 YZ |
6328 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6329 | struct btrfs_root *root, | |
6330 | struct walk_control *wc, | |
6331 | struct btrfs_path *path) | |
6407bf6d | 6332 | { |
1c4850e2 YZ |
6333 | u64 bytenr; |
6334 | u64 generation; | |
6335 | u64 refs; | |
94fcca9f | 6336 | u64 flags; |
5d4f98a2 | 6337 | u32 nritems; |
1c4850e2 YZ |
6338 | u32 blocksize; |
6339 | struct btrfs_key key; | |
6340 | struct extent_buffer *eb; | |
6407bf6d | 6341 | int ret; |
1c4850e2 YZ |
6342 | int slot; |
6343 | int nread = 0; | |
6407bf6d | 6344 | |
1c4850e2 YZ |
6345 | if (path->slots[wc->level] < wc->reada_slot) { |
6346 | wc->reada_count = wc->reada_count * 2 / 3; | |
6347 | wc->reada_count = max(wc->reada_count, 2); | |
6348 | } else { | |
6349 | wc->reada_count = wc->reada_count * 3 / 2; | |
6350 | wc->reada_count = min_t(int, wc->reada_count, | |
6351 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6352 | } | |
7bb86316 | 6353 | |
1c4850e2 YZ |
6354 | eb = path->nodes[wc->level]; |
6355 | nritems = btrfs_header_nritems(eb); | |
6356 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6357 | |
1c4850e2 YZ |
6358 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6359 | if (nread >= wc->reada_count) | |
6360 | break; | |
bd56b302 | 6361 | |
2dd3e67b | 6362 | cond_resched(); |
1c4850e2 YZ |
6363 | bytenr = btrfs_node_blockptr(eb, slot); |
6364 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6365 | |
1c4850e2 YZ |
6366 | if (slot == path->slots[wc->level]) |
6367 | goto reada; | |
5d4f98a2 | 6368 | |
1c4850e2 YZ |
6369 | if (wc->stage == UPDATE_BACKREF && |
6370 | generation <= root->root_key.offset) | |
bd56b302 CM |
6371 | continue; |
6372 | ||
94fcca9f YZ |
6373 | /* We don't lock the tree block, it's OK to be racy here */ |
6374 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6375 | &refs, &flags); | |
79787eaa JM |
6376 | /* We don't care about errors in readahead. */ |
6377 | if (ret < 0) | |
6378 | continue; | |
94fcca9f YZ |
6379 | BUG_ON(refs == 0); |
6380 | ||
1c4850e2 | 6381 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6382 | if (refs == 1) |
6383 | goto reada; | |
bd56b302 | 6384 | |
94fcca9f YZ |
6385 | if (wc->level == 1 && |
6386 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6387 | continue; | |
1c4850e2 YZ |
6388 | if (!wc->update_ref || |
6389 | generation <= root->root_key.offset) | |
6390 | continue; | |
6391 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6392 | ret = btrfs_comp_cpu_keys(&key, | |
6393 | &wc->update_progress); | |
6394 | if (ret < 0) | |
6395 | continue; | |
94fcca9f YZ |
6396 | } else { |
6397 | if (wc->level == 1 && | |
6398 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6399 | continue; | |
6407bf6d | 6400 | } |
1c4850e2 YZ |
6401 | reada: |
6402 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6403 | generation); | |
6404 | if (ret) | |
bd56b302 | 6405 | break; |
1c4850e2 | 6406 | nread++; |
20524f02 | 6407 | } |
1c4850e2 | 6408 | wc->reada_slot = slot; |
20524f02 | 6409 | } |
2c47e605 | 6410 | |
f82d02d9 | 6411 | /* |
2c47e605 YZ |
6412 | * hepler to process tree block while walking down the tree. |
6413 | * | |
2c47e605 YZ |
6414 | * when wc->stage == UPDATE_BACKREF, this function updates |
6415 | * back refs for pointers in the block. | |
6416 | * | |
6417 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6418 | */ |
2c47e605 | 6419 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6420 | struct btrfs_root *root, |
2c47e605 | 6421 | struct btrfs_path *path, |
94fcca9f | 6422 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6423 | { |
2c47e605 YZ |
6424 | int level = wc->level; |
6425 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6426 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6427 | int ret; |
6428 | ||
2c47e605 YZ |
6429 | if (wc->stage == UPDATE_BACKREF && |
6430 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6431 | return 1; | |
f82d02d9 | 6432 | |
2c47e605 YZ |
6433 | /* |
6434 | * when reference count of tree block is 1, it won't increase | |
6435 | * again. once full backref flag is set, we never clear it. | |
6436 | */ | |
94fcca9f YZ |
6437 | if (lookup_info && |
6438 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6439 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6440 | BUG_ON(!path->locks[level]); |
6441 | ret = btrfs_lookup_extent_info(trans, root, | |
6442 | eb->start, eb->len, | |
6443 | &wc->refs[level], | |
6444 | &wc->flags[level]); | |
79787eaa JM |
6445 | BUG_ON(ret == -ENOMEM); |
6446 | if (ret) | |
6447 | return ret; | |
2c47e605 YZ |
6448 | BUG_ON(wc->refs[level] == 0); |
6449 | } | |
5d4f98a2 | 6450 | |
2c47e605 YZ |
6451 | if (wc->stage == DROP_REFERENCE) { |
6452 | if (wc->refs[level] > 1) | |
6453 | return 1; | |
f82d02d9 | 6454 | |
2c47e605 | 6455 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6456 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6457 | path->locks[level] = 0; |
6458 | } | |
6459 | return 0; | |
6460 | } | |
f82d02d9 | 6461 | |
2c47e605 YZ |
6462 | /* wc->stage == UPDATE_BACKREF */ |
6463 | if (!(wc->flags[level] & flag)) { | |
6464 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6465 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6466 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6467 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6468 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6469 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6470 | eb->len, flag, 0); | |
79787eaa | 6471 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6472 | wc->flags[level] |= flag; |
6473 | } | |
6474 | ||
6475 | /* | |
6476 | * the block is shared by multiple trees, so it's not good to | |
6477 | * keep the tree lock | |
6478 | */ | |
6479 | if (path->locks[level] && level > 0) { | |
bd681513 | 6480 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6481 | path->locks[level] = 0; |
6482 | } | |
6483 | return 0; | |
6484 | } | |
6485 | ||
1c4850e2 YZ |
6486 | /* |
6487 | * hepler to process tree block pointer. | |
6488 | * | |
6489 | * when wc->stage == DROP_REFERENCE, this function checks | |
6490 | * reference count of the block pointed to. if the block | |
6491 | * is shared and we need update back refs for the subtree | |
6492 | * rooted at the block, this function changes wc->stage to | |
6493 | * UPDATE_BACKREF. if the block is shared and there is no | |
6494 | * need to update back, this function drops the reference | |
6495 | * to the block. | |
6496 | * | |
6497 | * NOTE: return value 1 means we should stop walking down. | |
6498 | */ | |
6499 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6500 | struct btrfs_root *root, | |
6501 | struct btrfs_path *path, | |
94fcca9f | 6502 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6503 | { |
6504 | u64 bytenr; | |
6505 | u64 generation; | |
6506 | u64 parent; | |
6507 | u32 blocksize; | |
6508 | struct btrfs_key key; | |
6509 | struct extent_buffer *next; | |
6510 | int level = wc->level; | |
6511 | int reada = 0; | |
6512 | int ret = 0; | |
6513 | ||
6514 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6515 | path->slots[level]); | |
6516 | /* | |
6517 | * if the lower level block was created before the snapshot | |
6518 | * was created, we know there is no need to update back refs | |
6519 | * for the subtree | |
6520 | */ | |
6521 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6522 | generation <= root->root_key.offset) { |
6523 | *lookup_info = 1; | |
1c4850e2 | 6524 | return 1; |
94fcca9f | 6525 | } |
1c4850e2 YZ |
6526 | |
6527 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6528 | blocksize = btrfs_level_size(root, level - 1); | |
6529 | ||
6530 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6531 | if (!next) { | |
6532 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6533 | if (!next) |
6534 | return -ENOMEM; | |
1c4850e2 YZ |
6535 | reada = 1; |
6536 | } | |
6537 | btrfs_tree_lock(next); | |
6538 | btrfs_set_lock_blocking(next); | |
6539 | ||
94fcca9f YZ |
6540 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6541 | &wc->refs[level - 1], | |
6542 | &wc->flags[level - 1]); | |
79787eaa JM |
6543 | if (ret < 0) { |
6544 | btrfs_tree_unlock(next); | |
6545 | return ret; | |
6546 | } | |
6547 | ||
94fcca9f YZ |
6548 | BUG_ON(wc->refs[level - 1] == 0); |
6549 | *lookup_info = 0; | |
1c4850e2 | 6550 | |
94fcca9f | 6551 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6552 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6553 | if (level == 1 && |
6554 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6555 | goto skip; | |
6556 | ||
1c4850e2 YZ |
6557 | if (!wc->update_ref || |
6558 | generation <= root->root_key.offset) | |
6559 | goto skip; | |
6560 | ||
6561 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6562 | path->slots[level]); | |
6563 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6564 | if (ret < 0) | |
6565 | goto skip; | |
6566 | ||
6567 | wc->stage = UPDATE_BACKREF; | |
6568 | wc->shared_level = level - 1; | |
6569 | } | |
94fcca9f YZ |
6570 | } else { |
6571 | if (level == 1 && | |
6572 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6573 | goto skip; | |
1c4850e2 YZ |
6574 | } |
6575 | ||
b9fab919 | 6576 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
6577 | btrfs_tree_unlock(next); |
6578 | free_extent_buffer(next); | |
6579 | next = NULL; | |
94fcca9f | 6580 | *lookup_info = 1; |
1c4850e2 YZ |
6581 | } |
6582 | ||
6583 | if (!next) { | |
6584 | if (reada && level == 1) | |
6585 | reada_walk_down(trans, root, wc, path); | |
6586 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6587 | if (!next) |
6588 | return -EIO; | |
1c4850e2 YZ |
6589 | btrfs_tree_lock(next); |
6590 | btrfs_set_lock_blocking(next); | |
6591 | } | |
6592 | ||
6593 | level--; | |
6594 | BUG_ON(level != btrfs_header_level(next)); | |
6595 | path->nodes[level] = next; | |
6596 | path->slots[level] = 0; | |
bd681513 | 6597 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6598 | wc->level = level; |
6599 | if (wc->level == 1) | |
6600 | wc->reada_slot = 0; | |
6601 | return 0; | |
6602 | skip: | |
6603 | wc->refs[level - 1] = 0; | |
6604 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6605 | if (wc->stage == DROP_REFERENCE) { |
6606 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6607 | parent = path->nodes[level]->start; | |
6608 | } else { | |
6609 | BUG_ON(root->root_key.objectid != | |
6610 | btrfs_header_owner(path->nodes[level])); | |
6611 | parent = 0; | |
6612 | } | |
1c4850e2 | 6613 | |
94fcca9f | 6614 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6615 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 6616 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 6617 | } |
1c4850e2 YZ |
6618 | btrfs_tree_unlock(next); |
6619 | free_extent_buffer(next); | |
94fcca9f | 6620 | *lookup_info = 1; |
1c4850e2 YZ |
6621 | return 1; |
6622 | } | |
6623 | ||
2c47e605 YZ |
6624 | /* |
6625 | * hepler to process tree block while walking up the tree. | |
6626 | * | |
6627 | * when wc->stage == DROP_REFERENCE, this function drops | |
6628 | * reference count on the block. | |
6629 | * | |
6630 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6631 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6632 | * to UPDATE_BACKREF previously while processing the block. | |
6633 | * | |
6634 | * NOTE: return value 1 means we should stop walking up. | |
6635 | */ | |
6636 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6637 | struct btrfs_root *root, | |
6638 | struct btrfs_path *path, | |
6639 | struct walk_control *wc) | |
6640 | { | |
f0486c68 | 6641 | int ret; |
2c47e605 YZ |
6642 | int level = wc->level; |
6643 | struct extent_buffer *eb = path->nodes[level]; | |
6644 | u64 parent = 0; | |
6645 | ||
6646 | if (wc->stage == UPDATE_BACKREF) { | |
6647 | BUG_ON(wc->shared_level < level); | |
6648 | if (level < wc->shared_level) | |
6649 | goto out; | |
6650 | ||
2c47e605 YZ |
6651 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6652 | if (ret > 0) | |
6653 | wc->update_ref = 0; | |
6654 | ||
6655 | wc->stage = DROP_REFERENCE; | |
6656 | wc->shared_level = -1; | |
6657 | path->slots[level] = 0; | |
6658 | ||
6659 | /* | |
6660 | * check reference count again if the block isn't locked. | |
6661 | * we should start walking down the tree again if reference | |
6662 | * count is one. | |
6663 | */ | |
6664 | if (!path->locks[level]) { | |
6665 | BUG_ON(level == 0); | |
6666 | btrfs_tree_lock(eb); | |
6667 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6668 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6669 | |
6670 | ret = btrfs_lookup_extent_info(trans, root, | |
6671 | eb->start, eb->len, | |
6672 | &wc->refs[level], | |
6673 | &wc->flags[level]); | |
79787eaa JM |
6674 | if (ret < 0) { |
6675 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
6676 | return ret; | |
6677 | } | |
2c47e605 YZ |
6678 | BUG_ON(wc->refs[level] == 0); |
6679 | if (wc->refs[level] == 1) { | |
bd681513 | 6680 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6681 | return 1; |
6682 | } | |
f82d02d9 | 6683 | } |
2c47e605 | 6684 | } |
f82d02d9 | 6685 | |
2c47e605 YZ |
6686 | /* wc->stage == DROP_REFERENCE */ |
6687 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6688 | |
2c47e605 YZ |
6689 | if (wc->refs[level] == 1) { |
6690 | if (level == 0) { | |
6691 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6692 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6693 | wc->for_reloc); | |
2c47e605 | 6694 | else |
66d7e7f0 AJ |
6695 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6696 | wc->for_reloc); | |
79787eaa | 6697 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6698 | } |
6699 | /* make block locked assertion in clean_tree_block happy */ | |
6700 | if (!path->locks[level] && | |
6701 | btrfs_header_generation(eb) == trans->transid) { | |
6702 | btrfs_tree_lock(eb); | |
6703 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6704 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6705 | } |
6706 | clean_tree_block(trans, root, eb); | |
6707 | } | |
6708 | ||
6709 | if (eb == root->node) { | |
6710 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6711 | parent = eb->start; | |
6712 | else | |
6713 | BUG_ON(root->root_key.objectid != | |
6714 | btrfs_header_owner(eb)); | |
6715 | } else { | |
6716 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6717 | parent = path->nodes[level + 1]->start; | |
6718 | else | |
6719 | BUG_ON(root->root_key.objectid != | |
6720 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6721 | } |
f82d02d9 | 6722 | |
5581a51a | 6723 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6724 | out: |
6725 | wc->refs[level] = 0; | |
6726 | wc->flags[level] = 0; | |
f0486c68 | 6727 | return 0; |
2c47e605 YZ |
6728 | } |
6729 | ||
6730 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6731 | struct btrfs_root *root, | |
6732 | struct btrfs_path *path, | |
6733 | struct walk_control *wc) | |
6734 | { | |
2c47e605 | 6735 | int level = wc->level; |
94fcca9f | 6736 | int lookup_info = 1; |
2c47e605 YZ |
6737 | int ret; |
6738 | ||
6739 | while (level >= 0) { | |
94fcca9f | 6740 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6741 | if (ret > 0) |
6742 | break; | |
6743 | ||
6744 | if (level == 0) | |
6745 | break; | |
6746 | ||
7a7965f8 YZ |
6747 | if (path->slots[level] >= |
6748 | btrfs_header_nritems(path->nodes[level])) | |
6749 | break; | |
6750 | ||
94fcca9f | 6751 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6752 | if (ret > 0) { |
6753 | path->slots[level]++; | |
6754 | continue; | |
90d2c51d MX |
6755 | } else if (ret < 0) |
6756 | return ret; | |
1c4850e2 | 6757 | level = wc->level; |
f82d02d9 | 6758 | } |
f82d02d9 YZ |
6759 | return 0; |
6760 | } | |
6761 | ||
d397712b | 6762 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6763 | struct btrfs_root *root, |
f82d02d9 | 6764 | struct btrfs_path *path, |
2c47e605 | 6765 | struct walk_control *wc, int max_level) |
20524f02 | 6766 | { |
2c47e605 | 6767 | int level = wc->level; |
20524f02 | 6768 | int ret; |
9f3a7427 | 6769 | |
2c47e605 YZ |
6770 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6771 | while (level < max_level && path->nodes[level]) { | |
6772 | wc->level = level; | |
6773 | if (path->slots[level] + 1 < | |
6774 | btrfs_header_nritems(path->nodes[level])) { | |
6775 | path->slots[level]++; | |
20524f02 CM |
6776 | return 0; |
6777 | } else { | |
2c47e605 YZ |
6778 | ret = walk_up_proc(trans, root, path, wc); |
6779 | if (ret > 0) | |
6780 | return 0; | |
bd56b302 | 6781 | |
2c47e605 | 6782 | if (path->locks[level]) { |
bd681513 CM |
6783 | btrfs_tree_unlock_rw(path->nodes[level], |
6784 | path->locks[level]); | |
2c47e605 | 6785 | path->locks[level] = 0; |
f82d02d9 | 6786 | } |
2c47e605 YZ |
6787 | free_extent_buffer(path->nodes[level]); |
6788 | path->nodes[level] = NULL; | |
6789 | level++; | |
20524f02 CM |
6790 | } |
6791 | } | |
6792 | return 1; | |
6793 | } | |
6794 | ||
9aca1d51 | 6795 | /* |
2c47e605 YZ |
6796 | * drop a subvolume tree. |
6797 | * | |
6798 | * this function traverses the tree freeing any blocks that only | |
6799 | * referenced by the tree. | |
6800 | * | |
6801 | * when a shared tree block is found. this function decreases its | |
6802 | * reference count by one. if update_ref is true, this function | |
6803 | * also make sure backrefs for the shared block and all lower level | |
6804 | * blocks are properly updated. | |
9aca1d51 | 6805 | */ |
2c536799 | 6806 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6807 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6808 | int for_reloc) | |
20524f02 | 6809 | { |
5caf2a00 | 6810 | struct btrfs_path *path; |
2c47e605 YZ |
6811 | struct btrfs_trans_handle *trans; |
6812 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6813 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6814 | struct walk_control *wc; |
6815 | struct btrfs_key key; | |
6816 | int err = 0; | |
6817 | int ret; | |
6818 | int level; | |
20524f02 | 6819 | |
5caf2a00 | 6820 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6821 | if (!path) { |
6822 | err = -ENOMEM; | |
6823 | goto out; | |
6824 | } | |
20524f02 | 6825 | |
2c47e605 | 6826 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6827 | if (!wc) { |
6828 | btrfs_free_path(path); | |
cb1b69f4 TI |
6829 | err = -ENOMEM; |
6830 | goto out; | |
38a1a919 | 6831 | } |
2c47e605 | 6832 | |
a22285a6 | 6833 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6834 | if (IS_ERR(trans)) { |
6835 | err = PTR_ERR(trans); | |
6836 | goto out_free; | |
6837 | } | |
98d5dc13 | 6838 | |
3fd0a558 YZ |
6839 | if (block_rsv) |
6840 | trans->block_rsv = block_rsv; | |
2c47e605 | 6841 | |
9f3a7427 | 6842 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6843 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6844 | path->nodes[level] = btrfs_lock_root_node(root); |
6845 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6846 | path->slots[level] = 0; |
bd681513 | 6847 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6848 | memset(&wc->update_progress, 0, |
6849 | sizeof(wc->update_progress)); | |
9f3a7427 | 6850 | } else { |
9f3a7427 | 6851 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6852 | memcpy(&wc->update_progress, &key, |
6853 | sizeof(wc->update_progress)); | |
6854 | ||
6702ed49 | 6855 | level = root_item->drop_level; |
2c47e605 | 6856 | BUG_ON(level == 0); |
6702ed49 | 6857 | path->lowest_level = level; |
2c47e605 YZ |
6858 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6859 | path->lowest_level = 0; | |
6860 | if (ret < 0) { | |
6861 | err = ret; | |
79787eaa | 6862 | goto out_end_trans; |
9f3a7427 | 6863 | } |
1c4850e2 | 6864 | WARN_ON(ret > 0); |
2c47e605 | 6865 | |
7d9eb12c CM |
6866 | /* |
6867 | * unlock our path, this is safe because only this | |
6868 | * function is allowed to delete this snapshot | |
6869 | */ | |
5d4f98a2 | 6870 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6871 | |
6872 | level = btrfs_header_level(root->node); | |
6873 | while (1) { | |
6874 | btrfs_tree_lock(path->nodes[level]); | |
6875 | btrfs_set_lock_blocking(path->nodes[level]); | |
6876 | ||
6877 | ret = btrfs_lookup_extent_info(trans, root, | |
6878 | path->nodes[level]->start, | |
6879 | path->nodes[level]->len, | |
6880 | &wc->refs[level], | |
6881 | &wc->flags[level]); | |
79787eaa JM |
6882 | if (ret < 0) { |
6883 | err = ret; | |
6884 | goto out_end_trans; | |
6885 | } | |
2c47e605 YZ |
6886 | BUG_ON(wc->refs[level] == 0); |
6887 | ||
6888 | if (level == root_item->drop_level) | |
6889 | break; | |
6890 | ||
6891 | btrfs_tree_unlock(path->nodes[level]); | |
6892 | WARN_ON(wc->refs[level] != 1); | |
6893 | level--; | |
6894 | } | |
9f3a7427 | 6895 | } |
2c47e605 YZ |
6896 | |
6897 | wc->level = level; | |
6898 | wc->shared_level = -1; | |
6899 | wc->stage = DROP_REFERENCE; | |
6900 | wc->update_ref = update_ref; | |
6901 | wc->keep_locks = 0; | |
66d7e7f0 | 6902 | wc->for_reloc = for_reloc; |
1c4850e2 | 6903 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6904 | |
d397712b | 6905 | while (1) { |
2c47e605 YZ |
6906 | ret = walk_down_tree(trans, root, path, wc); |
6907 | if (ret < 0) { | |
6908 | err = ret; | |
20524f02 | 6909 | break; |
2c47e605 | 6910 | } |
9aca1d51 | 6911 | |
2c47e605 YZ |
6912 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6913 | if (ret < 0) { | |
6914 | err = ret; | |
20524f02 | 6915 | break; |
2c47e605 YZ |
6916 | } |
6917 | ||
6918 | if (ret > 0) { | |
6919 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6920 | break; |
6921 | } | |
2c47e605 YZ |
6922 | |
6923 | if (wc->stage == DROP_REFERENCE) { | |
6924 | level = wc->level; | |
6925 | btrfs_node_key(path->nodes[level], | |
6926 | &root_item->drop_progress, | |
6927 | path->slots[level]); | |
6928 | root_item->drop_level = level; | |
6929 | } | |
6930 | ||
6931 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6932 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6933 | ret = btrfs_update_root(trans, tree_root, |
6934 | &root->root_key, | |
6935 | root_item); | |
79787eaa JM |
6936 | if (ret) { |
6937 | btrfs_abort_transaction(trans, tree_root, ret); | |
6938 | err = ret; | |
6939 | goto out_end_trans; | |
6940 | } | |
2c47e605 | 6941 | |
3fd0a558 | 6942 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6943 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6944 | if (IS_ERR(trans)) { |
6945 | err = PTR_ERR(trans); | |
6946 | goto out_free; | |
6947 | } | |
3fd0a558 YZ |
6948 | if (block_rsv) |
6949 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6950 | } |
20524f02 | 6951 | } |
b3b4aa74 | 6952 | btrfs_release_path(path); |
79787eaa JM |
6953 | if (err) |
6954 | goto out_end_trans; | |
2c47e605 YZ |
6955 | |
6956 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
6957 | if (ret) { |
6958 | btrfs_abort_transaction(trans, tree_root, ret); | |
6959 | goto out_end_trans; | |
6960 | } | |
2c47e605 | 6961 | |
76dda93c YZ |
6962 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6963 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6964 | NULL, NULL); | |
79787eaa JM |
6965 | if (ret < 0) { |
6966 | btrfs_abort_transaction(trans, tree_root, ret); | |
6967 | err = ret; | |
6968 | goto out_end_trans; | |
6969 | } else if (ret > 0) { | |
84cd948c JB |
6970 | /* if we fail to delete the orphan item this time |
6971 | * around, it'll get picked up the next time. | |
6972 | * | |
6973 | * The most common failure here is just -ENOENT. | |
6974 | */ | |
6975 | btrfs_del_orphan_item(trans, tree_root, | |
6976 | root->root_key.objectid); | |
76dda93c YZ |
6977 | } |
6978 | } | |
6979 | ||
6980 | if (root->in_radix) { | |
6981 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6982 | } else { | |
6983 | free_extent_buffer(root->node); | |
6984 | free_extent_buffer(root->commit_root); | |
6985 | kfree(root); | |
6986 | } | |
79787eaa | 6987 | out_end_trans: |
3fd0a558 | 6988 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 6989 | out_free: |
2c47e605 | 6990 | kfree(wc); |
5caf2a00 | 6991 | btrfs_free_path(path); |
cb1b69f4 TI |
6992 | out: |
6993 | if (err) | |
6994 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 6995 | return err; |
20524f02 | 6996 | } |
9078a3e1 | 6997 | |
2c47e605 YZ |
6998 | /* |
6999 | * drop subtree rooted at tree block 'node'. | |
7000 | * | |
7001 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7002 | * only used by relocation code |
2c47e605 | 7003 | */ |
f82d02d9 YZ |
7004 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7005 | struct btrfs_root *root, | |
7006 | struct extent_buffer *node, | |
7007 | struct extent_buffer *parent) | |
7008 | { | |
7009 | struct btrfs_path *path; | |
2c47e605 | 7010 | struct walk_control *wc; |
f82d02d9 YZ |
7011 | int level; |
7012 | int parent_level; | |
7013 | int ret = 0; | |
7014 | int wret; | |
7015 | ||
2c47e605 YZ |
7016 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7017 | ||
f82d02d9 | 7018 | path = btrfs_alloc_path(); |
db5b493a TI |
7019 | if (!path) |
7020 | return -ENOMEM; | |
f82d02d9 | 7021 | |
2c47e605 | 7022 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7023 | if (!wc) { |
7024 | btrfs_free_path(path); | |
7025 | return -ENOMEM; | |
7026 | } | |
2c47e605 | 7027 | |
b9447ef8 | 7028 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7029 | parent_level = btrfs_header_level(parent); |
7030 | extent_buffer_get(parent); | |
7031 | path->nodes[parent_level] = parent; | |
7032 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7033 | ||
b9447ef8 | 7034 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7035 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7036 | path->nodes[level] = node; |
7037 | path->slots[level] = 0; | |
bd681513 | 7038 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7039 | |
7040 | wc->refs[parent_level] = 1; | |
7041 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7042 | wc->level = level; | |
7043 | wc->shared_level = -1; | |
7044 | wc->stage = DROP_REFERENCE; | |
7045 | wc->update_ref = 0; | |
7046 | wc->keep_locks = 1; | |
66d7e7f0 | 7047 | wc->for_reloc = 1; |
1c4850e2 | 7048 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7049 | |
7050 | while (1) { | |
2c47e605 YZ |
7051 | wret = walk_down_tree(trans, root, path, wc); |
7052 | if (wret < 0) { | |
f82d02d9 | 7053 | ret = wret; |
f82d02d9 | 7054 | break; |
2c47e605 | 7055 | } |
f82d02d9 | 7056 | |
2c47e605 | 7057 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7058 | if (wret < 0) |
7059 | ret = wret; | |
7060 | if (wret != 0) | |
7061 | break; | |
7062 | } | |
7063 | ||
2c47e605 | 7064 | kfree(wc); |
f82d02d9 YZ |
7065 | btrfs_free_path(path); |
7066 | return ret; | |
7067 | } | |
7068 | ||
ec44a35c CM |
7069 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7070 | { | |
7071 | u64 num_devices; | |
fc67c450 | 7072 | u64 stripped; |
e4d8ec0f | 7073 | |
fc67c450 ID |
7074 | /* |
7075 | * if restripe for this chunk_type is on pick target profile and | |
7076 | * return, otherwise do the usual balance | |
7077 | */ | |
7078 | stripped = get_restripe_target(root->fs_info, flags); | |
7079 | if (stripped) | |
7080 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7081 | |
cd02dca5 CM |
7082 | /* |
7083 | * we add in the count of missing devices because we want | |
7084 | * to make sure that any RAID levels on a degraded FS | |
7085 | * continue to be honored. | |
7086 | */ | |
7087 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7088 | root->fs_info->fs_devices->missing_devices; | |
7089 | ||
fc67c450 ID |
7090 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
7091 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7092 | ||
ec44a35c CM |
7093 | if (num_devices == 1) { |
7094 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7095 | stripped = flags & ~stripped; | |
7096 | ||
7097 | /* turn raid0 into single device chunks */ | |
7098 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7099 | return stripped; | |
7100 | ||
7101 | /* turn mirroring into duplication */ | |
7102 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7103 | BTRFS_BLOCK_GROUP_RAID10)) | |
7104 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7105 | } else { |
7106 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7107 | if (flags & stripped) |
7108 | return flags; | |
7109 | ||
7110 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7111 | stripped = flags & ~stripped; | |
7112 | ||
7113 | /* switch duplicated blocks with raid1 */ | |
7114 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7115 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7116 | ||
e3176ca2 | 7117 | /* this is drive concat, leave it alone */ |
ec44a35c | 7118 | } |
e3176ca2 | 7119 | |
ec44a35c CM |
7120 | return flags; |
7121 | } | |
7122 | ||
199c36ea | 7123 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7124 | { |
f0486c68 YZ |
7125 | struct btrfs_space_info *sinfo = cache->space_info; |
7126 | u64 num_bytes; | |
199c36ea | 7127 | u64 min_allocable_bytes; |
f0486c68 | 7128 | int ret = -ENOSPC; |
0ef3e66b | 7129 | |
c286ac48 | 7130 | |
199c36ea MX |
7131 | /* |
7132 | * We need some metadata space and system metadata space for | |
7133 | * allocating chunks in some corner cases until we force to set | |
7134 | * it to be readonly. | |
7135 | */ | |
7136 | if ((sinfo->flags & | |
7137 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7138 | !force) | |
7139 | min_allocable_bytes = 1 * 1024 * 1024; | |
7140 | else | |
7141 | min_allocable_bytes = 0; | |
7142 | ||
f0486c68 YZ |
7143 | spin_lock(&sinfo->lock); |
7144 | spin_lock(&cache->lock); | |
61cfea9b W |
7145 | |
7146 | if (cache->ro) { | |
7147 | ret = 0; | |
7148 | goto out; | |
7149 | } | |
7150 | ||
f0486c68 YZ |
7151 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7152 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7153 | ||
7154 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7155 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7156 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7157 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7158 | cache->ro = 1; |
7159 | ret = 0; | |
7160 | } | |
61cfea9b | 7161 | out: |
f0486c68 YZ |
7162 | spin_unlock(&cache->lock); |
7163 | spin_unlock(&sinfo->lock); | |
7164 | return ret; | |
7165 | } | |
7d9eb12c | 7166 | |
f0486c68 YZ |
7167 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7168 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7169 | |
f0486c68 YZ |
7170 | { |
7171 | struct btrfs_trans_handle *trans; | |
7172 | u64 alloc_flags; | |
7173 | int ret; | |
7d9eb12c | 7174 | |
f0486c68 | 7175 | BUG_ON(cache->ro); |
0ef3e66b | 7176 | |
ff5714cc | 7177 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7178 | if (IS_ERR(trans)) |
7179 | return PTR_ERR(trans); | |
5d4f98a2 | 7180 | |
f0486c68 | 7181 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa JM |
7182 | if (alloc_flags != cache->flags) { |
7183 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, | |
7184 | CHUNK_ALLOC_FORCE); | |
7185 | if (ret < 0) | |
7186 | goto out; | |
7187 | } | |
5d4f98a2 | 7188 | |
199c36ea | 7189 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7190 | if (!ret) |
7191 | goto out; | |
7192 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
7193 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7194 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
7195 | if (ret < 0) |
7196 | goto out; | |
199c36ea | 7197 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7198 | out: |
7199 | btrfs_end_transaction(trans, root); | |
7200 | return ret; | |
7201 | } | |
5d4f98a2 | 7202 | |
c87f08ca CM |
7203 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7204 | struct btrfs_root *root, u64 type) | |
7205 | { | |
7206 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
7207 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7208 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
7209 | } |
7210 | ||
6d07bcec MX |
7211 | /* |
7212 | * helper to account the unused space of all the readonly block group in the | |
7213 | * list. takes mirrors into account. | |
7214 | */ | |
7215 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7216 | { | |
7217 | struct btrfs_block_group_cache *block_group; | |
7218 | u64 free_bytes = 0; | |
7219 | int factor; | |
7220 | ||
7221 | list_for_each_entry(block_group, groups_list, list) { | |
7222 | spin_lock(&block_group->lock); | |
7223 | ||
7224 | if (!block_group->ro) { | |
7225 | spin_unlock(&block_group->lock); | |
7226 | continue; | |
7227 | } | |
7228 | ||
7229 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7230 | BTRFS_BLOCK_GROUP_RAID10 | | |
7231 | BTRFS_BLOCK_GROUP_DUP)) | |
7232 | factor = 2; | |
7233 | else | |
7234 | factor = 1; | |
7235 | ||
7236 | free_bytes += (block_group->key.offset - | |
7237 | btrfs_block_group_used(&block_group->item)) * | |
7238 | factor; | |
7239 | ||
7240 | spin_unlock(&block_group->lock); | |
7241 | } | |
7242 | ||
7243 | return free_bytes; | |
7244 | } | |
7245 | ||
7246 | /* | |
7247 | * helper to account the unused space of all the readonly block group in the | |
7248 | * space_info. takes mirrors into account. | |
7249 | */ | |
7250 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7251 | { | |
7252 | int i; | |
7253 | u64 free_bytes = 0; | |
7254 | ||
7255 | spin_lock(&sinfo->lock); | |
7256 | ||
7257 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7258 | if (!list_empty(&sinfo->block_groups[i])) | |
7259 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7260 | &sinfo->block_groups[i]); | |
7261 | ||
7262 | spin_unlock(&sinfo->lock); | |
7263 | ||
7264 | return free_bytes; | |
7265 | } | |
7266 | ||
143bede5 | 7267 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7268 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7269 | { |
f0486c68 YZ |
7270 | struct btrfs_space_info *sinfo = cache->space_info; |
7271 | u64 num_bytes; | |
7272 | ||
7273 | BUG_ON(!cache->ro); | |
7274 | ||
7275 | spin_lock(&sinfo->lock); | |
7276 | spin_lock(&cache->lock); | |
7277 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7278 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7279 | sinfo->bytes_readonly -= num_bytes; | |
7280 | cache->ro = 0; | |
7281 | spin_unlock(&cache->lock); | |
7282 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7283 | } |
7284 | ||
ba1bf481 JB |
7285 | /* |
7286 | * checks to see if its even possible to relocate this block group. | |
7287 | * | |
7288 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7289 | * ok to go ahead and try. | |
7290 | */ | |
7291 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7292 | { |
ba1bf481 JB |
7293 | struct btrfs_block_group_cache *block_group; |
7294 | struct btrfs_space_info *space_info; | |
7295 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7296 | struct btrfs_device *device; | |
cdcb725c | 7297 | u64 min_free; |
6719db6a JB |
7298 | u64 dev_min = 1; |
7299 | u64 dev_nr = 0; | |
4a5e98f5 | 7300 | u64 target; |
cdcb725c | 7301 | int index; |
ba1bf481 JB |
7302 | int full = 0; |
7303 | int ret = 0; | |
1a40e23b | 7304 | |
ba1bf481 | 7305 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7306 | |
ba1bf481 JB |
7307 | /* odd, couldn't find the block group, leave it alone */ |
7308 | if (!block_group) | |
7309 | return -1; | |
1a40e23b | 7310 | |
cdcb725c | 7311 | min_free = btrfs_block_group_used(&block_group->item); |
7312 | ||
ba1bf481 | 7313 | /* no bytes used, we're good */ |
cdcb725c | 7314 | if (!min_free) |
1a40e23b ZY |
7315 | goto out; |
7316 | ||
ba1bf481 JB |
7317 | space_info = block_group->space_info; |
7318 | spin_lock(&space_info->lock); | |
17d217fe | 7319 | |
ba1bf481 | 7320 | full = space_info->full; |
17d217fe | 7321 | |
ba1bf481 JB |
7322 | /* |
7323 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7324 | * relocate it unless we're able to allocate a new chunk below. |
7325 | * | |
7326 | * Otherwise, we need to make sure we have room in the space to handle | |
7327 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7328 | */ |
7ce618db | 7329 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7330 | (space_info->bytes_used + space_info->bytes_reserved + |
7331 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7332 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7333 | spin_unlock(&space_info->lock); |
7334 | goto out; | |
17d217fe | 7335 | } |
ba1bf481 | 7336 | spin_unlock(&space_info->lock); |
ea8c2819 | 7337 | |
ba1bf481 JB |
7338 | /* |
7339 | * ok we don't have enough space, but maybe we have free space on our | |
7340 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7341 | * alloc devices and guess if we have enough space. if this block |
7342 | * group is going to be restriped, run checks against the target | |
7343 | * profile instead of the current one. | |
ba1bf481 JB |
7344 | */ |
7345 | ret = -1; | |
ea8c2819 | 7346 | |
cdcb725c | 7347 | /* |
7348 | * index: | |
7349 | * 0: raid10 | |
7350 | * 1: raid1 | |
7351 | * 2: dup | |
7352 | * 3: raid0 | |
7353 | * 4: single | |
7354 | */ | |
4a5e98f5 ID |
7355 | target = get_restripe_target(root->fs_info, block_group->flags); |
7356 | if (target) { | |
7357 | index = __get_block_group_index(extended_to_chunk(target)); | |
7358 | } else { | |
7359 | /* | |
7360 | * this is just a balance, so if we were marked as full | |
7361 | * we know there is no space for a new chunk | |
7362 | */ | |
7363 | if (full) | |
7364 | goto out; | |
7365 | ||
7366 | index = get_block_group_index(block_group); | |
7367 | } | |
7368 | ||
cdcb725c | 7369 | if (index == 0) { |
7370 | dev_min = 4; | |
6719db6a JB |
7371 | /* Divide by 2 */ |
7372 | min_free >>= 1; | |
cdcb725c | 7373 | } else if (index == 1) { |
7374 | dev_min = 2; | |
7375 | } else if (index == 2) { | |
6719db6a JB |
7376 | /* Multiply by 2 */ |
7377 | min_free <<= 1; | |
cdcb725c | 7378 | } else if (index == 3) { |
7379 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7380 | do_div(min_free, dev_min); |
cdcb725c | 7381 | } |
7382 | ||
ba1bf481 JB |
7383 | mutex_lock(&root->fs_info->chunk_mutex); |
7384 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7385 | u64 dev_offset; |
56bec294 | 7386 | |
ba1bf481 JB |
7387 | /* |
7388 | * check to make sure we can actually find a chunk with enough | |
7389 | * space to fit our block group in. | |
7390 | */ | |
7391 | if (device->total_bytes > device->bytes_used + min_free) { | |
125ccb0a | 7392 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7393 | &dev_offset, NULL); |
ba1bf481 | 7394 | if (!ret) |
cdcb725c | 7395 | dev_nr++; |
7396 | ||
7397 | if (dev_nr >= dev_min) | |
73e48b27 | 7398 | break; |
cdcb725c | 7399 | |
ba1bf481 | 7400 | ret = -1; |
725c8463 | 7401 | } |
edbd8d4e | 7402 | } |
ba1bf481 | 7403 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7404 | out: |
ba1bf481 | 7405 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7406 | return ret; |
7407 | } | |
7408 | ||
b2950863 CH |
7409 | static int find_first_block_group(struct btrfs_root *root, |
7410 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7411 | { |
925baedd | 7412 | int ret = 0; |
0b86a832 CM |
7413 | struct btrfs_key found_key; |
7414 | struct extent_buffer *leaf; | |
7415 | int slot; | |
edbd8d4e | 7416 | |
0b86a832 CM |
7417 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7418 | if (ret < 0) | |
925baedd CM |
7419 | goto out; |
7420 | ||
d397712b | 7421 | while (1) { |
0b86a832 | 7422 | slot = path->slots[0]; |
edbd8d4e | 7423 | leaf = path->nodes[0]; |
0b86a832 CM |
7424 | if (slot >= btrfs_header_nritems(leaf)) { |
7425 | ret = btrfs_next_leaf(root, path); | |
7426 | if (ret == 0) | |
7427 | continue; | |
7428 | if (ret < 0) | |
925baedd | 7429 | goto out; |
0b86a832 | 7430 | break; |
edbd8d4e | 7431 | } |
0b86a832 | 7432 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7433 | |
0b86a832 | 7434 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7435 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7436 | ret = 0; | |
7437 | goto out; | |
7438 | } | |
0b86a832 | 7439 | path->slots[0]++; |
edbd8d4e | 7440 | } |
925baedd | 7441 | out: |
0b86a832 | 7442 | return ret; |
edbd8d4e CM |
7443 | } |
7444 | ||
0af3d00b JB |
7445 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7446 | { | |
7447 | struct btrfs_block_group_cache *block_group; | |
7448 | u64 last = 0; | |
7449 | ||
7450 | while (1) { | |
7451 | struct inode *inode; | |
7452 | ||
7453 | block_group = btrfs_lookup_first_block_group(info, last); | |
7454 | while (block_group) { | |
7455 | spin_lock(&block_group->lock); | |
7456 | if (block_group->iref) | |
7457 | break; | |
7458 | spin_unlock(&block_group->lock); | |
7459 | block_group = next_block_group(info->tree_root, | |
7460 | block_group); | |
7461 | } | |
7462 | if (!block_group) { | |
7463 | if (last == 0) | |
7464 | break; | |
7465 | last = 0; | |
7466 | continue; | |
7467 | } | |
7468 | ||
7469 | inode = block_group->inode; | |
7470 | block_group->iref = 0; | |
7471 | block_group->inode = NULL; | |
7472 | spin_unlock(&block_group->lock); | |
7473 | iput(inode); | |
7474 | last = block_group->key.objectid + block_group->key.offset; | |
7475 | btrfs_put_block_group(block_group); | |
7476 | } | |
7477 | } | |
7478 | ||
1a40e23b ZY |
7479 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7480 | { | |
7481 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7482 | struct btrfs_space_info *space_info; |
11833d66 | 7483 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7484 | struct rb_node *n; |
7485 | ||
11833d66 YZ |
7486 | down_write(&info->extent_commit_sem); |
7487 | while (!list_empty(&info->caching_block_groups)) { | |
7488 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7489 | struct btrfs_caching_control, list); | |
7490 | list_del(&caching_ctl->list); | |
7491 | put_caching_control(caching_ctl); | |
7492 | } | |
7493 | up_write(&info->extent_commit_sem); | |
7494 | ||
1a40e23b ZY |
7495 | spin_lock(&info->block_group_cache_lock); |
7496 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7497 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7498 | cache_node); | |
1a40e23b ZY |
7499 | rb_erase(&block_group->cache_node, |
7500 | &info->block_group_cache_tree); | |
d899e052 YZ |
7501 | spin_unlock(&info->block_group_cache_lock); |
7502 | ||
80eb234a | 7503 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7504 | list_del(&block_group->list); |
80eb234a | 7505 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7506 | |
817d52f8 | 7507 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7508 | wait_block_group_cache_done(block_group); |
817d52f8 | 7509 | |
3c14874a JB |
7510 | /* |
7511 | * We haven't cached this block group, which means we could | |
7512 | * possibly have excluded extents on this block group. | |
7513 | */ | |
7514 | if (block_group->cached == BTRFS_CACHE_NO) | |
7515 | free_excluded_extents(info->extent_root, block_group); | |
7516 | ||
817d52f8 | 7517 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7518 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7519 | |
7520 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7521 | } |
7522 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7523 | |
7524 | /* now that all the block groups are freed, go through and | |
7525 | * free all the space_info structs. This is only called during | |
7526 | * the final stages of unmount, and so we know nobody is | |
7527 | * using them. We call synchronize_rcu() once before we start, | |
7528 | * just to be on the safe side. | |
7529 | */ | |
7530 | synchronize_rcu(); | |
7531 | ||
8929ecfa YZ |
7532 | release_global_block_rsv(info); |
7533 | ||
4184ea7f CM |
7534 | while(!list_empty(&info->space_info)) { |
7535 | space_info = list_entry(info->space_info.next, | |
7536 | struct btrfs_space_info, | |
7537 | list); | |
f0486c68 | 7538 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7539 | space_info->bytes_reserved > 0 || |
7540 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7541 | WARN_ON(1); |
7542 | dump_space_info(space_info, 0, 0); | |
7543 | } | |
4184ea7f CM |
7544 | list_del(&space_info->list); |
7545 | kfree(space_info); | |
7546 | } | |
1a40e23b ZY |
7547 | return 0; |
7548 | } | |
7549 | ||
b742bb82 YZ |
7550 | static void __link_block_group(struct btrfs_space_info *space_info, |
7551 | struct btrfs_block_group_cache *cache) | |
7552 | { | |
7553 | int index = get_block_group_index(cache); | |
7554 | ||
7555 | down_write(&space_info->groups_sem); | |
7556 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7557 | up_write(&space_info->groups_sem); | |
7558 | } | |
7559 | ||
9078a3e1 CM |
7560 | int btrfs_read_block_groups(struct btrfs_root *root) |
7561 | { | |
7562 | struct btrfs_path *path; | |
7563 | int ret; | |
9078a3e1 | 7564 | struct btrfs_block_group_cache *cache; |
be744175 | 7565 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7566 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7567 | struct btrfs_key key; |
7568 | struct btrfs_key found_key; | |
5f39d397 | 7569 | struct extent_buffer *leaf; |
0af3d00b JB |
7570 | int need_clear = 0; |
7571 | u64 cache_gen; | |
96b5179d | 7572 | |
be744175 | 7573 | root = info->extent_root; |
9078a3e1 | 7574 | key.objectid = 0; |
0b86a832 | 7575 | key.offset = 0; |
9078a3e1 | 7576 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7577 | path = btrfs_alloc_path(); |
7578 | if (!path) | |
7579 | return -ENOMEM; | |
026fd317 | 7580 | path->reada = 1; |
9078a3e1 | 7581 | |
6c41761f | 7582 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7583 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7584 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7585 | need_clear = 1; |
88c2ba3b JB |
7586 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7587 | need_clear = 1; | |
0af3d00b | 7588 | |
d397712b | 7589 | while (1) { |
0b86a832 | 7590 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7591 | if (ret > 0) |
7592 | break; | |
0b86a832 CM |
7593 | if (ret != 0) |
7594 | goto error; | |
5f39d397 CM |
7595 | leaf = path->nodes[0]; |
7596 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7597 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7598 | if (!cache) { |
0b86a832 | 7599 | ret = -ENOMEM; |
f0486c68 | 7600 | goto error; |
9078a3e1 | 7601 | } |
34d52cb6 LZ |
7602 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7603 | GFP_NOFS); | |
7604 | if (!cache->free_space_ctl) { | |
7605 | kfree(cache); | |
7606 | ret = -ENOMEM; | |
7607 | goto error; | |
7608 | } | |
3e1ad54f | 7609 | |
d2fb3437 | 7610 | atomic_set(&cache->count, 1); |
c286ac48 | 7611 | spin_lock_init(&cache->lock); |
817d52f8 | 7612 | cache->fs_info = info; |
0f9dd46c | 7613 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7614 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7615 | |
0af3d00b JB |
7616 | if (need_clear) |
7617 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
7618 | ||
5f39d397 CM |
7619 | read_extent_buffer(leaf, &cache->item, |
7620 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7621 | sizeof(cache->item)); | |
9078a3e1 | 7622 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7623 | |
9078a3e1 | 7624 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7625 | btrfs_release_path(path); |
0b86a832 | 7626 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7627 | cache->sectorsize = root->sectorsize; |
7628 | ||
34d52cb6 LZ |
7629 | btrfs_init_free_space_ctl(cache); |
7630 | ||
3c14874a JB |
7631 | /* |
7632 | * We need to exclude the super stripes now so that the space | |
7633 | * info has super bytes accounted for, otherwise we'll think | |
7634 | * we have more space than we actually do. | |
7635 | */ | |
7636 | exclude_super_stripes(root, cache); | |
7637 | ||
817d52f8 JB |
7638 | /* |
7639 | * check for two cases, either we are full, and therefore | |
7640 | * don't need to bother with the caching work since we won't | |
7641 | * find any space, or we are empty, and we can just add all | |
7642 | * the space in and be done with it. This saves us _alot_ of | |
7643 | * time, particularly in the full case. | |
7644 | */ | |
7645 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7646 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7647 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7648 | free_excluded_extents(root, cache); |
817d52f8 | 7649 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7650 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7651 | cache->cached = BTRFS_CACHE_FINISHED; |
7652 | add_new_free_space(cache, root->fs_info, | |
7653 | found_key.objectid, | |
7654 | found_key.objectid + | |
7655 | found_key.offset); | |
11833d66 | 7656 | free_excluded_extents(root, cache); |
817d52f8 | 7657 | } |
96b5179d | 7658 | |
6324fbf3 CM |
7659 | ret = update_space_info(info, cache->flags, found_key.offset, |
7660 | btrfs_block_group_used(&cache->item), | |
7661 | &space_info); | |
79787eaa | 7662 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 7663 | cache->space_info = space_info; |
1b2da372 | 7664 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7665 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7666 | spin_unlock(&cache->space_info->lock); |
7667 | ||
b742bb82 | 7668 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7669 | |
7670 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 7671 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
7672 | |
7673 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7674 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7675 | set_block_group_ro(cache, 1); |
9078a3e1 | 7676 | } |
b742bb82 YZ |
7677 | |
7678 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7679 | if (!(get_alloc_profile(root, space_info->flags) & | |
7680 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7681 | BTRFS_BLOCK_GROUP_RAID1 | | |
7682 | BTRFS_BLOCK_GROUP_DUP))) | |
7683 | continue; | |
7684 | /* | |
7685 | * avoid allocating from un-mirrored block group if there are | |
7686 | * mirrored block groups. | |
7687 | */ | |
7688 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7689 | set_block_group_ro(cache, 1); |
b742bb82 | 7690 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7691 | set_block_group_ro(cache, 1); |
9078a3e1 | 7692 | } |
f0486c68 YZ |
7693 | |
7694 | init_global_block_rsv(info); | |
0b86a832 CM |
7695 | ret = 0; |
7696 | error: | |
9078a3e1 | 7697 | btrfs_free_path(path); |
0b86a832 | 7698 | return ret; |
9078a3e1 | 7699 | } |
6324fbf3 CM |
7700 | |
7701 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7702 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7703 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7704 | u64 size) |
7705 | { | |
7706 | int ret; | |
6324fbf3 CM |
7707 | struct btrfs_root *extent_root; |
7708 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7709 | |
7710 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7711 | |
12fcfd22 | 7712 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7713 | |
8f18cf13 | 7714 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7715 | if (!cache) |
7716 | return -ENOMEM; | |
34d52cb6 LZ |
7717 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7718 | GFP_NOFS); | |
7719 | if (!cache->free_space_ctl) { | |
7720 | kfree(cache); | |
7721 | return -ENOMEM; | |
7722 | } | |
0f9dd46c | 7723 | |
e17cade2 | 7724 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7725 | cache->key.offset = size; |
d2fb3437 | 7726 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7727 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7728 | cache->fs_info = root->fs_info; |
96303081 | 7729 | |
d2fb3437 | 7730 | atomic_set(&cache->count, 1); |
c286ac48 | 7731 | spin_lock_init(&cache->lock); |
0f9dd46c | 7732 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7733 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7734 | |
34d52cb6 LZ |
7735 | btrfs_init_free_space_ctl(cache); |
7736 | ||
6324fbf3 | 7737 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7738 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7739 | cache->flags = type; | |
7740 | btrfs_set_block_group_flags(&cache->item, type); | |
7741 | ||
11833d66 | 7742 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7743 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7744 | exclude_super_stripes(root, cache); |
96303081 | 7745 | |
817d52f8 JB |
7746 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7747 | chunk_offset + size); | |
7748 | ||
11833d66 YZ |
7749 | free_excluded_extents(root, cache); |
7750 | ||
6324fbf3 CM |
7751 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7752 | &cache->space_info); | |
79787eaa | 7753 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 7754 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7755 | |
7756 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7757 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7758 | spin_unlock(&cache->space_info->lock); |
7759 | ||
b742bb82 | 7760 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7761 | |
0f9dd46c | 7762 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 7763 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 7764 | |
6324fbf3 CM |
7765 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7766 | sizeof(cache->item)); | |
79787eaa JM |
7767 | if (ret) { |
7768 | btrfs_abort_transaction(trans, extent_root, ret); | |
7769 | return ret; | |
7770 | } | |
6324fbf3 | 7771 | |
d18a2c44 | 7772 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7773 | |
6324fbf3 CM |
7774 | return 0; |
7775 | } | |
1a40e23b | 7776 | |
10ea00f5 ID |
7777 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7778 | { | |
899c81ea ID |
7779 | u64 extra_flags = chunk_to_extended(flags) & |
7780 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 ID |
7781 | |
7782 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7783 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7784 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7785 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7786 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7787 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7788 | } | |
7789 | ||
1a40e23b ZY |
7790 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7791 | struct btrfs_root *root, u64 group_start) | |
7792 | { | |
7793 | struct btrfs_path *path; | |
7794 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7795 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7796 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7797 | struct btrfs_key key; |
0af3d00b | 7798 | struct inode *inode; |
1a40e23b | 7799 | int ret; |
10ea00f5 | 7800 | int index; |
89a55897 | 7801 | int factor; |
1a40e23b | 7802 | |
1a40e23b ZY |
7803 | root = root->fs_info->extent_root; |
7804 | ||
7805 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7806 | BUG_ON(!block_group); | |
c146afad | 7807 | BUG_ON(!block_group->ro); |
1a40e23b | 7808 | |
9f7c43c9 | 7809 | /* |
7810 | * Free the reserved super bytes from this block group before | |
7811 | * remove it. | |
7812 | */ | |
7813 | free_excluded_extents(root, block_group); | |
7814 | ||
1a40e23b | 7815 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7816 | index = get_block_group_index(block_group); |
89a55897 JB |
7817 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7818 | BTRFS_BLOCK_GROUP_RAID1 | | |
7819 | BTRFS_BLOCK_GROUP_RAID10)) | |
7820 | factor = 2; | |
7821 | else | |
7822 | factor = 1; | |
1a40e23b | 7823 | |
44fb5511 CM |
7824 | /* make sure this block group isn't part of an allocation cluster */ |
7825 | cluster = &root->fs_info->data_alloc_cluster; | |
7826 | spin_lock(&cluster->refill_lock); | |
7827 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7828 | spin_unlock(&cluster->refill_lock); | |
7829 | ||
7830 | /* | |
7831 | * make sure this block group isn't part of a metadata | |
7832 | * allocation cluster | |
7833 | */ | |
7834 | cluster = &root->fs_info->meta_alloc_cluster; | |
7835 | spin_lock(&cluster->refill_lock); | |
7836 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7837 | spin_unlock(&cluster->refill_lock); | |
7838 | ||
1a40e23b | 7839 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7840 | if (!path) { |
7841 | ret = -ENOMEM; | |
7842 | goto out; | |
7843 | } | |
1a40e23b | 7844 | |
10b2f34d | 7845 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7846 | if (!IS_ERR(inode)) { |
b532402e | 7847 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
7848 | if (ret) { |
7849 | btrfs_add_delayed_iput(inode); | |
7850 | goto out; | |
7851 | } | |
0af3d00b JB |
7852 | clear_nlink(inode); |
7853 | /* One for the block groups ref */ | |
7854 | spin_lock(&block_group->lock); | |
7855 | if (block_group->iref) { | |
7856 | block_group->iref = 0; | |
7857 | block_group->inode = NULL; | |
7858 | spin_unlock(&block_group->lock); | |
7859 | iput(inode); | |
7860 | } else { | |
7861 | spin_unlock(&block_group->lock); | |
7862 | } | |
7863 | /* One for our lookup ref */ | |
455757c3 | 7864 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7865 | } |
7866 | ||
7867 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7868 | key.offset = block_group->key.objectid; | |
7869 | key.type = 0; | |
7870 | ||
7871 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7872 | if (ret < 0) | |
7873 | goto out; | |
7874 | if (ret > 0) | |
b3b4aa74 | 7875 | btrfs_release_path(path); |
0af3d00b JB |
7876 | if (ret == 0) { |
7877 | ret = btrfs_del_item(trans, tree_root, path); | |
7878 | if (ret) | |
7879 | goto out; | |
b3b4aa74 | 7880 | btrfs_release_path(path); |
0af3d00b JB |
7881 | } |
7882 | ||
3dfdb934 | 7883 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7884 | rb_erase(&block_group->cache_node, |
7885 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7886 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7887 | |
80eb234a | 7888 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7889 | /* |
7890 | * we must use list_del_init so people can check to see if they | |
7891 | * are still on the list after taking the semaphore | |
7892 | */ | |
7893 | list_del_init(&block_group->list); | |
10ea00f5 ID |
7894 | if (list_empty(&block_group->space_info->block_groups[index])) |
7895 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 7896 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7897 | |
817d52f8 | 7898 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7899 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7900 | |
7901 | btrfs_remove_free_space_cache(block_group); | |
7902 | ||
c146afad YZ |
7903 | spin_lock(&block_group->space_info->lock); |
7904 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7905 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7906 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7907 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7908 | |
0af3d00b JB |
7909 | memcpy(&key, &block_group->key, sizeof(key)); |
7910 | ||
283bb197 | 7911 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7912 | |
fa9c0d79 CM |
7913 | btrfs_put_block_group(block_group); |
7914 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7915 | |
7916 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7917 | if (ret > 0) | |
7918 | ret = -EIO; | |
7919 | if (ret < 0) | |
7920 | goto out; | |
7921 | ||
7922 | ret = btrfs_del_item(trans, root, path); | |
7923 | out: | |
7924 | btrfs_free_path(path); | |
7925 | return ret; | |
7926 | } | |
acce952b | 7927 | |
c59021f8 | 7928 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7929 | { | |
7930 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7931 | struct btrfs_super_block *disk_super; |
7932 | u64 features; | |
7933 | u64 flags; | |
7934 | int mixed = 0; | |
c59021f8 | 7935 | int ret; |
7936 | ||
6c41761f | 7937 | disk_super = fs_info->super_copy; |
1aba86d6 | 7938 | if (!btrfs_super_root(disk_super)) |
7939 | return 1; | |
c59021f8 | 7940 | |
1aba86d6 | 7941 | features = btrfs_super_incompat_flags(disk_super); |
7942 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7943 | mixed = 1; | |
c59021f8 | 7944 | |
1aba86d6 | 7945 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7946 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7947 | if (ret) |
1aba86d6 | 7948 | goto out; |
c59021f8 | 7949 | |
1aba86d6 | 7950 | if (mixed) { |
7951 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7952 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7953 | } else { | |
7954 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7955 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7956 | if (ret) | |
7957 | goto out; | |
7958 | ||
7959 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7960 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7961 | } | |
7962 | out: | |
c59021f8 | 7963 | return ret; |
7964 | } | |
7965 | ||
acce952b | 7966 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7967 | { | |
7968 | return unpin_extent_range(root, start, end); | |
7969 | } | |
7970 | ||
7971 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7972 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7973 | { |
5378e607 | 7974 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7975 | } |
f7039b1d LD |
7976 | |
7977 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7978 | { | |
7979 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7980 | struct btrfs_block_group_cache *cache = NULL; | |
7981 | u64 group_trimmed; | |
7982 | u64 start; | |
7983 | u64 end; | |
7984 | u64 trimmed = 0; | |
2cac13e4 | 7985 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
7986 | int ret = 0; |
7987 | ||
2cac13e4 LB |
7988 | /* |
7989 | * try to trim all FS space, our block group may start from non-zero. | |
7990 | */ | |
7991 | if (range->len == total_bytes) | |
7992 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
7993 | else | |
7994 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
7995 | |
7996 | while (cache) { | |
7997 | if (cache->key.objectid >= (range->start + range->len)) { | |
7998 | btrfs_put_block_group(cache); | |
7999 | break; | |
8000 | } | |
8001 | ||
8002 | start = max(range->start, cache->key.objectid); | |
8003 | end = min(range->start + range->len, | |
8004 | cache->key.objectid + cache->key.offset); | |
8005 | ||
8006 | if (end - start >= range->minlen) { | |
8007 | if (!block_group_cache_done(cache)) { | |
8008 | ret = cache_block_group(cache, NULL, root, 0); | |
8009 | if (!ret) | |
8010 | wait_block_group_cache_done(cache); | |
8011 | } | |
8012 | ret = btrfs_trim_block_group(cache, | |
8013 | &group_trimmed, | |
8014 | start, | |
8015 | end, | |
8016 | range->minlen); | |
8017 | ||
8018 | trimmed += group_trimmed; | |
8019 | if (ret) { | |
8020 | btrfs_put_block_group(cache); | |
8021 | break; | |
8022 | } | |
8023 | } | |
8024 | ||
8025 | cache = next_block_group(fs_info->tree_root, cache); | |
8026 | } | |
8027 | ||
8028 | range->len = trimmed; | |
8029 | return ret; | |
8030 | } |