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> |
b150a4f1 | 27 | #include <linux/percpu_counter.h> |
4b4e25f2 | 28 | #include "compat.h" |
74493f7a | 29 | #include "hash.h" |
fec577fb CM |
30 | #include "ctree.h" |
31 | #include "disk-io.h" | |
32 | #include "print-tree.h" | |
e089f05c | 33 | #include "transaction.h" |
0b86a832 | 34 | #include "volumes.h" |
53b381b3 | 35 | #include "raid56.h" |
925baedd | 36 | #include "locking.h" |
fa9c0d79 | 37 | #include "free-space-cache.h" |
3fed40cc | 38 | #include "math.h" |
fec577fb | 39 | |
709c0486 AJ |
40 | #undef SCRAMBLE_DELAYED_REFS |
41 | ||
9e622d6b MX |
42 | /* |
43 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
44 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
45 | * if we really need one. | |
46 | * | |
0e4f8f88 CM |
47 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
48 | * if we have very few chunks already allocated. This is | |
49 | * used as part of the clustering code to help make sure | |
50 | * we have a good pool of storage to cluster in, without | |
51 | * filling the FS with empty chunks | |
52 | * | |
9e622d6b MX |
53 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
54 | * | |
0e4f8f88 CM |
55 | */ |
56 | enum { | |
57 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
58 | CHUNK_ALLOC_LIMITED = 1, |
59 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
60 | }; |
61 | ||
fb25e914 JB |
62 | /* |
63 | * Control how reservations are dealt with. | |
64 | * | |
65 | * RESERVE_FREE - freeing a reservation. | |
66 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
67 | * ENOSPC accounting | |
68 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
69 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
70 | */ | |
71 | enum { | |
72 | RESERVE_FREE = 0, | |
73 | RESERVE_ALLOC = 1, | |
74 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
75 | }; | |
76 | ||
c53d613e | 77 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 78 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
79 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
80 | struct btrfs_root *root, | |
81 | u64 bytenr, u64 num_bytes, u64 parent, | |
82 | u64 root_objectid, u64 owner_objectid, | |
83 | u64 owner_offset, int refs_to_drop, | |
84 | struct btrfs_delayed_extent_op *extra_op); | |
85 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
86 | struct extent_buffer *leaf, | |
87 | struct btrfs_extent_item *ei); | |
88 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
89 | struct btrfs_root *root, | |
90 | u64 parent, u64 root_objectid, | |
91 | u64 flags, u64 owner, u64 offset, | |
92 | struct btrfs_key *ins, int ref_mod); | |
93 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
94 | struct btrfs_root *root, | |
95 | u64 parent, u64 root_objectid, | |
96 | u64 flags, struct btrfs_disk_key *key, | |
97 | int level, struct btrfs_key *ins); | |
6a63209f | 98 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
99 | struct btrfs_root *extent_root, u64 flags, |
100 | int force); | |
11833d66 YZ |
101 | static int find_next_key(struct btrfs_path *path, int level, |
102 | struct btrfs_key *key); | |
9ed74f2d JB |
103 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
104 | int dump_block_groups); | |
fb25e914 JB |
105 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
106 | u64 num_bytes, int reserve); | |
5d80366e JB |
107 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
108 | u64 num_bytes); | |
48a3b636 ES |
109 | int btrfs_pin_extent(struct btrfs_root *root, |
110 | u64 bytenr, u64 num_bytes, int reserved); | |
6a63209f | 111 | |
817d52f8 JB |
112 | static noinline int |
113 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
114 | { | |
115 | smp_mb(); | |
116 | return cache->cached == BTRFS_CACHE_FINISHED; | |
117 | } | |
118 | ||
0f9dd46c JB |
119 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
120 | { | |
121 | return (cache->flags & bits) == bits; | |
122 | } | |
123 | ||
62a45b60 | 124 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
125 | { |
126 | atomic_inc(&cache->count); | |
127 | } | |
128 | ||
129 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
130 | { | |
f0486c68 YZ |
131 | if (atomic_dec_and_test(&cache->count)) { |
132 | WARN_ON(cache->pinned > 0); | |
133 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 134 | kfree(cache->free_space_ctl); |
11dfe35a | 135 | kfree(cache); |
f0486c68 | 136 | } |
11dfe35a JB |
137 | } |
138 | ||
0f9dd46c JB |
139 | /* |
140 | * this adds the block group to the fs_info rb tree for the block group | |
141 | * cache | |
142 | */ | |
b2950863 | 143 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
144 | struct btrfs_block_group_cache *block_group) |
145 | { | |
146 | struct rb_node **p; | |
147 | struct rb_node *parent = NULL; | |
148 | struct btrfs_block_group_cache *cache; | |
149 | ||
150 | spin_lock(&info->block_group_cache_lock); | |
151 | p = &info->block_group_cache_tree.rb_node; | |
152 | ||
153 | while (*p) { | |
154 | parent = *p; | |
155 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
156 | cache_node); | |
157 | if (block_group->key.objectid < cache->key.objectid) { | |
158 | p = &(*p)->rb_left; | |
159 | } else if (block_group->key.objectid > cache->key.objectid) { | |
160 | p = &(*p)->rb_right; | |
161 | } else { | |
162 | spin_unlock(&info->block_group_cache_lock); | |
163 | return -EEXIST; | |
164 | } | |
165 | } | |
166 | ||
167 | rb_link_node(&block_group->cache_node, parent, p); | |
168 | rb_insert_color(&block_group->cache_node, | |
169 | &info->block_group_cache_tree); | |
a1897fdd LB |
170 | |
171 | if (info->first_logical_byte > block_group->key.objectid) | |
172 | info->first_logical_byte = block_group->key.objectid; | |
173 | ||
0f9dd46c JB |
174 | spin_unlock(&info->block_group_cache_lock); |
175 | ||
176 | return 0; | |
177 | } | |
178 | ||
179 | /* | |
180 | * This will return the block group at or after bytenr if contains is 0, else | |
181 | * it will return the block group that contains the bytenr | |
182 | */ | |
183 | static struct btrfs_block_group_cache * | |
184 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
185 | int contains) | |
186 | { | |
187 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
188 | struct rb_node *n; | |
189 | u64 end, start; | |
190 | ||
191 | spin_lock(&info->block_group_cache_lock); | |
192 | n = info->block_group_cache_tree.rb_node; | |
193 | ||
194 | while (n) { | |
195 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
196 | cache_node); | |
197 | end = cache->key.objectid + cache->key.offset - 1; | |
198 | start = cache->key.objectid; | |
199 | ||
200 | if (bytenr < start) { | |
201 | if (!contains && (!ret || start < ret->key.objectid)) | |
202 | ret = cache; | |
203 | n = n->rb_left; | |
204 | } else if (bytenr > start) { | |
205 | if (contains && bytenr <= end) { | |
206 | ret = cache; | |
207 | break; | |
208 | } | |
209 | n = n->rb_right; | |
210 | } else { | |
211 | ret = cache; | |
212 | break; | |
213 | } | |
214 | } | |
a1897fdd | 215 | if (ret) { |
11dfe35a | 216 | btrfs_get_block_group(ret); |
a1897fdd LB |
217 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
218 | info->first_logical_byte = ret->key.objectid; | |
219 | } | |
0f9dd46c JB |
220 | spin_unlock(&info->block_group_cache_lock); |
221 | ||
222 | return ret; | |
223 | } | |
224 | ||
11833d66 YZ |
225 | static int add_excluded_extent(struct btrfs_root *root, |
226 | u64 start, u64 num_bytes) | |
817d52f8 | 227 | { |
11833d66 YZ |
228 | u64 end = start + num_bytes - 1; |
229 | set_extent_bits(&root->fs_info->freed_extents[0], | |
230 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
231 | set_extent_bits(&root->fs_info->freed_extents[1], | |
232 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
233 | return 0; | |
234 | } | |
817d52f8 | 235 | |
11833d66 YZ |
236 | static void free_excluded_extents(struct btrfs_root *root, |
237 | struct btrfs_block_group_cache *cache) | |
238 | { | |
239 | u64 start, end; | |
817d52f8 | 240 | |
11833d66 YZ |
241 | start = cache->key.objectid; |
242 | end = start + cache->key.offset - 1; | |
243 | ||
244 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
245 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
246 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
247 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
248 | } |
249 | ||
11833d66 YZ |
250 | static int exclude_super_stripes(struct btrfs_root *root, |
251 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 252 | { |
817d52f8 JB |
253 | u64 bytenr; |
254 | u64 *logical; | |
255 | int stripe_len; | |
256 | int i, nr, ret; | |
257 | ||
06b2331f YZ |
258 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
259 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
260 | cache->bytes_super += stripe_len; | |
261 | ret = add_excluded_extent(root, cache->key.objectid, | |
262 | stripe_len); | |
835d974f JB |
263 | if (ret) |
264 | return ret; | |
06b2331f YZ |
265 | } |
266 | ||
817d52f8 JB |
267 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
268 | bytenr = btrfs_sb_offset(i); | |
269 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
270 | cache->key.objectid, bytenr, | |
271 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
272 | if (ret) |
273 | return ret; | |
11833d66 | 274 | |
817d52f8 | 275 | while (nr--) { |
51bf5f0b JB |
276 | u64 start, len; |
277 | ||
278 | if (logical[nr] > cache->key.objectid + | |
279 | cache->key.offset) | |
280 | continue; | |
281 | ||
282 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
283 | continue; | |
284 | ||
285 | start = logical[nr]; | |
286 | if (start < cache->key.objectid) { | |
287 | start = cache->key.objectid; | |
288 | len = (logical[nr] + stripe_len) - start; | |
289 | } else { | |
290 | len = min_t(u64, stripe_len, | |
291 | cache->key.objectid + | |
292 | cache->key.offset - start); | |
293 | } | |
294 | ||
295 | cache->bytes_super += len; | |
296 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
297 | if (ret) { |
298 | kfree(logical); | |
299 | return ret; | |
300 | } | |
817d52f8 | 301 | } |
11833d66 | 302 | |
817d52f8 JB |
303 | kfree(logical); |
304 | } | |
817d52f8 JB |
305 | return 0; |
306 | } | |
307 | ||
11833d66 YZ |
308 | static struct btrfs_caching_control * |
309 | get_caching_control(struct btrfs_block_group_cache *cache) | |
310 | { | |
311 | struct btrfs_caching_control *ctl; | |
312 | ||
313 | spin_lock(&cache->lock); | |
314 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
315 | spin_unlock(&cache->lock); | |
316 | return NULL; | |
317 | } | |
318 | ||
dde5abee JB |
319 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
320 | if (!cache->caching_ctl) { | |
321 | spin_unlock(&cache->lock); | |
11833d66 YZ |
322 | return NULL; |
323 | } | |
324 | ||
325 | ctl = cache->caching_ctl; | |
326 | atomic_inc(&ctl->count); | |
327 | spin_unlock(&cache->lock); | |
328 | return ctl; | |
329 | } | |
330 | ||
331 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
332 | { | |
333 | if (atomic_dec_and_test(&ctl->count)) | |
334 | kfree(ctl); | |
335 | } | |
336 | ||
0f9dd46c JB |
337 | /* |
338 | * this is only called by cache_block_group, since we could have freed extents | |
339 | * we need to check the pinned_extents for any extents that can't be used yet | |
340 | * since their free space will be released as soon as the transaction commits. | |
341 | */ | |
817d52f8 | 342 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
343 | struct btrfs_fs_info *info, u64 start, u64 end) |
344 | { | |
817d52f8 | 345 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
346 | int ret; |
347 | ||
348 | while (start < end) { | |
11833d66 | 349 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 350 | &extent_start, &extent_end, |
e6138876 JB |
351 | EXTENT_DIRTY | EXTENT_UPTODATE, |
352 | NULL); | |
0f9dd46c JB |
353 | if (ret) |
354 | break; | |
355 | ||
06b2331f | 356 | if (extent_start <= start) { |
0f9dd46c JB |
357 | start = extent_end + 1; |
358 | } else if (extent_start > start && extent_start < end) { | |
359 | size = extent_start - start; | |
817d52f8 | 360 | total_added += size; |
ea6a478e JB |
361 | ret = btrfs_add_free_space(block_group, start, |
362 | size); | |
79787eaa | 363 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
364 | start = extent_end + 1; |
365 | } else { | |
366 | break; | |
367 | } | |
368 | } | |
369 | ||
370 | if (start < end) { | |
371 | size = end - start; | |
817d52f8 | 372 | total_added += size; |
ea6a478e | 373 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 374 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
375 | } |
376 | ||
817d52f8 | 377 | return total_added; |
0f9dd46c JB |
378 | } |
379 | ||
bab39bf9 | 380 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 381 | { |
bab39bf9 JB |
382 | struct btrfs_block_group_cache *block_group; |
383 | struct btrfs_fs_info *fs_info; | |
384 | struct btrfs_caching_control *caching_ctl; | |
385 | struct btrfs_root *extent_root; | |
e37c9e69 | 386 | struct btrfs_path *path; |
5f39d397 | 387 | struct extent_buffer *leaf; |
11833d66 | 388 | struct btrfs_key key; |
817d52f8 | 389 | u64 total_found = 0; |
11833d66 YZ |
390 | u64 last = 0; |
391 | u32 nritems; | |
392 | int ret = 0; | |
f510cfec | 393 | |
bab39bf9 JB |
394 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
395 | block_group = caching_ctl->block_group; | |
396 | fs_info = block_group->fs_info; | |
397 | extent_root = fs_info->extent_root; | |
398 | ||
e37c9e69 CM |
399 | path = btrfs_alloc_path(); |
400 | if (!path) | |
bab39bf9 | 401 | goto out; |
7d7d6068 | 402 | |
817d52f8 | 403 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 404 | |
5cd57b2c | 405 | /* |
817d52f8 JB |
406 | * We don't want to deadlock with somebody trying to allocate a new |
407 | * extent for the extent root while also trying to search the extent | |
408 | * root to add free space. So we skip locking and search the commit | |
409 | * root, since its read-only | |
5cd57b2c CM |
410 | */ |
411 | path->skip_locking = 1; | |
817d52f8 | 412 | path->search_commit_root = 1; |
026fd317 | 413 | path->reada = 1; |
817d52f8 | 414 | |
e4404d6e | 415 | key.objectid = last; |
e37c9e69 | 416 | key.offset = 0; |
11833d66 | 417 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 418 | again: |
11833d66 | 419 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
420 | /* need to make sure the commit_root doesn't disappear */ |
421 | down_read(&fs_info->extent_commit_sem); | |
422 | ||
52ee28d2 | 423 | next: |
11833d66 | 424 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 425 | if (ret < 0) |
ef8bbdfe | 426 | goto err; |
a512bbf8 | 427 | |
11833d66 YZ |
428 | leaf = path->nodes[0]; |
429 | nritems = btrfs_header_nritems(leaf); | |
430 | ||
d397712b | 431 | while (1) { |
7841cb28 | 432 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 433 | last = (u64)-1; |
817d52f8 | 434 | break; |
f25784b3 | 435 | } |
817d52f8 | 436 | |
11833d66 YZ |
437 | if (path->slots[0] < nritems) { |
438 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
439 | } else { | |
440 | ret = find_next_key(path, 0, &key); | |
441 | if (ret) | |
e37c9e69 | 442 | break; |
817d52f8 | 443 | |
0a3896d0 | 444 | if (need_resched()) { |
589d8ade | 445 | caching_ctl->progress = last; |
ff5714cc | 446 | btrfs_release_path(path); |
589d8ade JB |
447 | up_read(&fs_info->extent_commit_sem); |
448 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 449 | cond_resched(); |
589d8ade JB |
450 | goto again; |
451 | } | |
0a3896d0 JB |
452 | |
453 | ret = btrfs_next_leaf(extent_root, path); | |
454 | if (ret < 0) | |
455 | goto err; | |
456 | if (ret) | |
457 | break; | |
589d8ade JB |
458 | leaf = path->nodes[0]; |
459 | nritems = btrfs_header_nritems(leaf); | |
460 | continue; | |
11833d66 | 461 | } |
817d52f8 | 462 | |
52ee28d2 LB |
463 | if (key.objectid < last) { |
464 | key.objectid = last; | |
465 | key.offset = 0; | |
466 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
467 | ||
468 | caching_ctl->progress = last; | |
469 | btrfs_release_path(path); | |
470 | goto next; | |
471 | } | |
472 | ||
11833d66 YZ |
473 | if (key.objectid < block_group->key.objectid) { |
474 | path->slots[0]++; | |
817d52f8 | 475 | continue; |
e37c9e69 | 476 | } |
0f9dd46c | 477 | |
e37c9e69 | 478 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 479 | block_group->key.offset) |
e37c9e69 | 480 | break; |
7d7d6068 | 481 | |
3173a18f JB |
482 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
483 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
484 | total_found += add_new_free_space(block_group, |
485 | fs_info, last, | |
486 | key.objectid); | |
3173a18f JB |
487 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
488 | last = key.objectid + | |
489 | fs_info->tree_root->leafsize; | |
490 | else | |
491 | last = key.objectid + key.offset; | |
817d52f8 | 492 | |
11833d66 YZ |
493 | if (total_found > (1024 * 1024 * 2)) { |
494 | total_found = 0; | |
495 | wake_up(&caching_ctl->wait); | |
496 | } | |
817d52f8 | 497 | } |
e37c9e69 CM |
498 | path->slots[0]++; |
499 | } | |
817d52f8 | 500 | ret = 0; |
e37c9e69 | 501 | |
817d52f8 JB |
502 | total_found += add_new_free_space(block_group, fs_info, last, |
503 | block_group->key.objectid + | |
504 | block_group->key.offset); | |
11833d66 | 505 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
506 | |
507 | spin_lock(&block_group->lock); | |
11833d66 | 508 | block_group->caching_ctl = NULL; |
817d52f8 JB |
509 | block_group->cached = BTRFS_CACHE_FINISHED; |
510 | spin_unlock(&block_group->lock); | |
0f9dd46c | 511 | |
54aa1f4d | 512 | err: |
e37c9e69 | 513 | btrfs_free_path(path); |
276e680d | 514 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 515 | |
11833d66 YZ |
516 | free_excluded_extents(extent_root, block_group); |
517 | ||
518 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 519 | out: |
11833d66 YZ |
520 | wake_up(&caching_ctl->wait); |
521 | ||
522 | put_caching_control(caching_ctl); | |
11dfe35a | 523 | btrfs_put_block_group(block_group); |
817d52f8 JB |
524 | } |
525 | ||
9d66e233 | 526 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 527 | int load_cache_only) |
817d52f8 | 528 | { |
291c7d2f | 529 | DEFINE_WAIT(wait); |
11833d66 YZ |
530 | struct btrfs_fs_info *fs_info = cache->fs_info; |
531 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
532 | int ret = 0; |
533 | ||
291c7d2f | 534 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
535 | if (!caching_ctl) |
536 | return -ENOMEM; | |
291c7d2f JB |
537 | |
538 | INIT_LIST_HEAD(&caching_ctl->list); | |
539 | mutex_init(&caching_ctl->mutex); | |
540 | init_waitqueue_head(&caching_ctl->wait); | |
541 | caching_ctl->block_group = cache; | |
542 | caching_ctl->progress = cache->key.objectid; | |
543 | atomic_set(&caching_ctl->count, 1); | |
544 | caching_ctl->work.func = caching_thread; | |
545 | ||
546 | spin_lock(&cache->lock); | |
547 | /* | |
548 | * This should be a rare occasion, but this could happen I think in the | |
549 | * case where one thread starts to load the space cache info, and then | |
550 | * some other thread starts a transaction commit which tries to do an | |
551 | * allocation while the other thread is still loading the space cache | |
552 | * info. The previous loop should have kept us from choosing this block | |
553 | * group, but if we've moved to the state where we will wait on caching | |
554 | * block groups we need to first check if we're doing a fast load here, | |
555 | * so we can wait for it to finish, otherwise we could end up allocating | |
556 | * from a block group who's cache gets evicted for one reason or | |
557 | * another. | |
558 | */ | |
559 | while (cache->cached == BTRFS_CACHE_FAST) { | |
560 | struct btrfs_caching_control *ctl; | |
561 | ||
562 | ctl = cache->caching_ctl; | |
563 | atomic_inc(&ctl->count); | |
564 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
565 | spin_unlock(&cache->lock); | |
566 | ||
567 | schedule(); | |
568 | ||
569 | finish_wait(&ctl->wait, &wait); | |
570 | put_caching_control(ctl); | |
571 | spin_lock(&cache->lock); | |
572 | } | |
573 | ||
574 | if (cache->cached != BTRFS_CACHE_NO) { | |
575 | spin_unlock(&cache->lock); | |
576 | kfree(caching_ctl); | |
11833d66 | 577 | return 0; |
291c7d2f JB |
578 | } |
579 | WARN_ON(cache->caching_ctl); | |
580 | cache->caching_ctl = caching_ctl; | |
581 | cache->cached = BTRFS_CACHE_FAST; | |
582 | spin_unlock(&cache->lock); | |
11833d66 | 583 | |
d53ba474 | 584 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
585 | ret = load_free_space_cache(fs_info, cache); |
586 | ||
587 | spin_lock(&cache->lock); | |
588 | if (ret == 1) { | |
291c7d2f | 589 | cache->caching_ctl = NULL; |
9d66e233 JB |
590 | cache->cached = BTRFS_CACHE_FINISHED; |
591 | cache->last_byte_to_unpin = (u64)-1; | |
592 | } else { | |
291c7d2f JB |
593 | if (load_cache_only) { |
594 | cache->caching_ctl = NULL; | |
595 | cache->cached = BTRFS_CACHE_NO; | |
596 | } else { | |
597 | cache->cached = BTRFS_CACHE_STARTED; | |
598 | } | |
9d66e233 JB |
599 | } |
600 | spin_unlock(&cache->lock); | |
291c7d2f | 601 | wake_up(&caching_ctl->wait); |
3c14874a | 602 | if (ret == 1) { |
291c7d2f | 603 | put_caching_control(caching_ctl); |
3c14874a | 604 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 605 | return 0; |
3c14874a | 606 | } |
291c7d2f JB |
607 | } else { |
608 | /* | |
609 | * We are not going to do the fast caching, set cached to the | |
610 | * appropriate value and wakeup any waiters. | |
611 | */ | |
612 | spin_lock(&cache->lock); | |
613 | if (load_cache_only) { | |
614 | cache->caching_ctl = NULL; | |
615 | cache->cached = BTRFS_CACHE_NO; | |
616 | } else { | |
617 | cache->cached = BTRFS_CACHE_STARTED; | |
618 | } | |
619 | spin_unlock(&cache->lock); | |
620 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
621 | } |
622 | ||
291c7d2f JB |
623 | if (load_cache_only) { |
624 | put_caching_control(caching_ctl); | |
11833d66 | 625 | return 0; |
817d52f8 | 626 | } |
817d52f8 | 627 | |
11833d66 | 628 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 629 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
630 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
631 | up_write(&fs_info->extent_commit_sem); | |
632 | ||
11dfe35a | 633 | btrfs_get_block_group(cache); |
11833d66 | 634 | |
bab39bf9 | 635 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 636 | |
ef8bbdfe | 637 | return ret; |
e37c9e69 CM |
638 | } |
639 | ||
0f9dd46c JB |
640 | /* |
641 | * return the block group that starts at or after bytenr | |
642 | */ | |
d397712b CM |
643 | static struct btrfs_block_group_cache * |
644 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 645 | { |
0f9dd46c | 646 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 647 | |
0f9dd46c | 648 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 649 | |
0f9dd46c | 650 | return cache; |
0ef3e66b CM |
651 | } |
652 | ||
0f9dd46c | 653 | /* |
9f55684c | 654 | * return the block group that contains the given bytenr |
0f9dd46c | 655 | */ |
d397712b CM |
656 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
657 | struct btrfs_fs_info *info, | |
658 | u64 bytenr) | |
be744175 | 659 | { |
0f9dd46c | 660 | struct btrfs_block_group_cache *cache; |
be744175 | 661 | |
0f9dd46c | 662 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 663 | |
0f9dd46c | 664 | return cache; |
be744175 | 665 | } |
0b86a832 | 666 | |
0f9dd46c JB |
667 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
668 | u64 flags) | |
6324fbf3 | 669 | { |
0f9dd46c | 670 | struct list_head *head = &info->space_info; |
0f9dd46c | 671 | struct btrfs_space_info *found; |
4184ea7f | 672 | |
52ba6929 | 673 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 674 | |
4184ea7f CM |
675 | rcu_read_lock(); |
676 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 677 | if (found->flags & flags) { |
4184ea7f | 678 | rcu_read_unlock(); |
0f9dd46c | 679 | return found; |
4184ea7f | 680 | } |
0f9dd46c | 681 | } |
4184ea7f | 682 | rcu_read_unlock(); |
0f9dd46c | 683 | return NULL; |
6324fbf3 CM |
684 | } |
685 | ||
4184ea7f CM |
686 | /* |
687 | * after adding space to the filesystem, we need to clear the full flags | |
688 | * on all the space infos. | |
689 | */ | |
690 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
691 | { | |
692 | struct list_head *head = &info->space_info; | |
693 | struct btrfs_space_info *found; | |
694 | ||
695 | rcu_read_lock(); | |
696 | list_for_each_entry_rcu(found, head, list) | |
697 | found->full = 0; | |
698 | rcu_read_unlock(); | |
699 | } | |
700 | ||
e02119d5 | 701 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 702 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
703 | { |
704 | int ret; | |
705 | struct btrfs_key key; | |
31840ae1 | 706 | struct btrfs_path *path; |
e02119d5 | 707 | |
31840ae1 | 708 | path = btrfs_alloc_path(); |
d8926bb3 MF |
709 | if (!path) |
710 | return -ENOMEM; | |
711 | ||
e02119d5 CM |
712 | key.objectid = start; |
713 | key.offset = len; | |
3173a18f | 714 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
715 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
716 | 0, 0); | |
3173a18f JB |
717 | if (ret > 0) { |
718 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
719 | if (key.objectid == start && | |
720 | key.type == BTRFS_METADATA_ITEM_KEY) | |
721 | ret = 0; | |
722 | } | |
31840ae1 | 723 | btrfs_free_path(path); |
7bb86316 CM |
724 | return ret; |
725 | } | |
726 | ||
a22285a6 | 727 | /* |
3173a18f | 728 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
729 | * |
730 | * the head node for delayed ref is used to store the sum of all the | |
731 | * reference count modifications queued up in the rbtree. the head | |
732 | * node may also store the extent flags to set. This way you can check | |
733 | * to see what the reference count and extent flags would be if all of | |
734 | * the delayed refs are not processed. | |
735 | */ | |
736 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
737 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 738 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
739 | { |
740 | struct btrfs_delayed_ref_head *head; | |
741 | struct btrfs_delayed_ref_root *delayed_refs; | |
742 | struct btrfs_path *path; | |
743 | struct btrfs_extent_item *ei; | |
744 | struct extent_buffer *leaf; | |
745 | struct btrfs_key key; | |
746 | u32 item_size; | |
747 | u64 num_refs; | |
748 | u64 extent_flags; | |
749 | int ret; | |
750 | ||
3173a18f JB |
751 | /* |
752 | * If we don't have skinny metadata, don't bother doing anything | |
753 | * different | |
754 | */ | |
755 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
756 | offset = root->leafsize; | |
757 | metadata = 0; | |
758 | } | |
759 | ||
a22285a6 YZ |
760 | path = btrfs_alloc_path(); |
761 | if (!path) | |
762 | return -ENOMEM; | |
763 | ||
3173a18f JB |
764 | if (metadata) { |
765 | key.objectid = bytenr; | |
766 | key.type = BTRFS_METADATA_ITEM_KEY; | |
767 | key.offset = offset; | |
768 | } else { | |
769 | key.objectid = bytenr; | |
770 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
771 | key.offset = offset; | |
772 | } | |
773 | ||
a22285a6 YZ |
774 | if (!trans) { |
775 | path->skip_locking = 1; | |
776 | path->search_commit_root = 1; | |
777 | } | |
778 | again: | |
779 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
780 | &key, path, 0, 0); | |
781 | if (ret < 0) | |
782 | goto out_free; | |
783 | ||
3173a18f | 784 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
785 | metadata = 0; |
786 | if (path->slots[0]) { | |
787 | path->slots[0]--; | |
788 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
789 | path->slots[0]); | |
790 | if (key.objectid == bytenr && | |
791 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
792 | key.offset == root->leafsize) | |
793 | ret = 0; | |
794 | } | |
795 | if (ret) { | |
796 | key.objectid = bytenr; | |
797 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
798 | key.offset = root->leafsize; | |
799 | btrfs_release_path(path); | |
800 | goto again; | |
801 | } | |
3173a18f JB |
802 | } |
803 | ||
a22285a6 YZ |
804 | if (ret == 0) { |
805 | leaf = path->nodes[0]; | |
806 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
807 | if (item_size >= sizeof(*ei)) { | |
808 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
809 | struct btrfs_extent_item); | |
810 | num_refs = btrfs_extent_refs(leaf, ei); | |
811 | extent_flags = btrfs_extent_flags(leaf, ei); | |
812 | } else { | |
813 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
814 | struct btrfs_extent_item_v0 *ei0; | |
815 | BUG_ON(item_size != sizeof(*ei0)); | |
816 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
817 | struct btrfs_extent_item_v0); | |
818 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
819 | /* FIXME: this isn't correct for data */ | |
820 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
821 | #else | |
822 | BUG(); | |
823 | #endif | |
824 | } | |
825 | BUG_ON(num_refs == 0); | |
826 | } else { | |
827 | num_refs = 0; | |
828 | extent_flags = 0; | |
829 | ret = 0; | |
830 | } | |
831 | ||
832 | if (!trans) | |
833 | goto out; | |
834 | ||
835 | delayed_refs = &trans->transaction->delayed_refs; | |
836 | spin_lock(&delayed_refs->lock); | |
837 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
838 | if (head) { | |
839 | if (!mutex_trylock(&head->mutex)) { | |
840 | atomic_inc(&head->node.refs); | |
841 | spin_unlock(&delayed_refs->lock); | |
842 | ||
b3b4aa74 | 843 | btrfs_release_path(path); |
a22285a6 | 844 | |
8cc33e5c DS |
845 | /* |
846 | * Mutex was contended, block until it's released and try | |
847 | * again | |
848 | */ | |
a22285a6 YZ |
849 | mutex_lock(&head->mutex); |
850 | mutex_unlock(&head->mutex); | |
851 | btrfs_put_delayed_ref(&head->node); | |
852 | goto again; | |
853 | } | |
854 | if (head->extent_op && head->extent_op->update_flags) | |
855 | extent_flags |= head->extent_op->flags_to_set; | |
856 | else | |
857 | BUG_ON(num_refs == 0); | |
858 | ||
859 | num_refs += head->node.ref_mod; | |
860 | mutex_unlock(&head->mutex); | |
861 | } | |
862 | spin_unlock(&delayed_refs->lock); | |
863 | out: | |
864 | WARN_ON(num_refs == 0); | |
865 | if (refs) | |
866 | *refs = num_refs; | |
867 | if (flags) | |
868 | *flags = extent_flags; | |
869 | out_free: | |
870 | btrfs_free_path(path); | |
871 | return ret; | |
872 | } | |
873 | ||
d8d5f3e1 CM |
874 | /* |
875 | * Back reference rules. Back refs have three main goals: | |
876 | * | |
877 | * 1) differentiate between all holders of references to an extent so that | |
878 | * when a reference is dropped we can make sure it was a valid reference | |
879 | * before freeing the extent. | |
880 | * | |
881 | * 2) Provide enough information to quickly find the holders of an extent | |
882 | * if we notice a given block is corrupted or bad. | |
883 | * | |
884 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
885 | * maintenance. This is actually the same as #2, but with a slightly | |
886 | * different use case. | |
887 | * | |
5d4f98a2 YZ |
888 | * There are two kinds of back refs. The implicit back refs is optimized |
889 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
890 | * back refs of this kind provide information about the block's owner tree | |
891 | * and the pointer's key. These information allow us to find the block by | |
892 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
893 | * referenced by their owner trees. The location of tree block is recorded | |
894 | * in the back refs. Actually the full back refs is generic, and can be | |
895 | * used in all cases the implicit back refs is used. The major shortcoming | |
896 | * of the full back refs is its overhead. Every time a tree block gets | |
897 | * COWed, we have to update back refs entry for all pointers in it. | |
898 | * | |
899 | * For a newly allocated tree block, we use implicit back refs for | |
900 | * pointers in it. This means most tree related operations only involve | |
901 | * implicit back refs. For a tree block created in old transaction, the | |
902 | * only way to drop a reference to it is COW it. So we can detect the | |
903 | * event that tree block loses its owner tree's reference and do the | |
904 | * back refs conversion. | |
905 | * | |
906 | * When a tree block is COW'd through a tree, there are four cases: | |
907 | * | |
908 | * The reference count of the block is one and the tree is the block's | |
909 | * owner tree. Nothing to do in this case. | |
910 | * | |
911 | * The reference count of the block is one and the tree is not the | |
912 | * block's owner tree. In this case, full back refs is used for pointers | |
913 | * in the block. Remove these full back refs, add implicit back refs for | |
914 | * every pointers in the new block. | |
915 | * | |
916 | * The reference count of the block is greater than one and the tree is | |
917 | * the block's owner tree. In this case, implicit back refs is used for | |
918 | * pointers in the block. Add full back refs for every pointers in the | |
919 | * block, increase lower level extents' reference counts. The original | |
920 | * implicit back refs are entailed to the new block. | |
921 | * | |
922 | * The reference count of the block is greater than one and the tree is | |
923 | * not the block's owner tree. Add implicit back refs for every pointer in | |
924 | * the new block, increase lower level extents' reference count. | |
925 | * | |
926 | * Back Reference Key composing: | |
927 | * | |
928 | * The key objectid corresponds to the first byte in the extent, | |
929 | * The key type is used to differentiate between types of back refs. | |
930 | * There are different meanings of the key offset for different types | |
931 | * of back refs. | |
932 | * | |
d8d5f3e1 CM |
933 | * File extents can be referenced by: |
934 | * | |
935 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 936 | * - different files inside a single subvolume |
d8d5f3e1 CM |
937 | * - different offsets inside a file (bookend extents in file.c) |
938 | * | |
5d4f98a2 | 939 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
940 | * |
941 | * - Objectid of the subvolume root | |
d8d5f3e1 | 942 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
943 | * - original offset in the file |
944 | * - how many bookend extents | |
d8d5f3e1 | 945 | * |
5d4f98a2 YZ |
946 | * The key offset for the implicit back refs is hash of the first |
947 | * three fields. | |
d8d5f3e1 | 948 | * |
5d4f98a2 | 949 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 950 | * |
5d4f98a2 | 951 | * - number of pointers in the tree leaf |
d8d5f3e1 | 952 | * |
5d4f98a2 YZ |
953 | * The key offset for the implicit back refs is the first byte of |
954 | * the tree leaf | |
d8d5f3e1 | 955 | * |
5d4f98a2 YZ |
956 | * When a file extent is allocated, The implicit back refs is used. |
957 | * the fields are filled in: | |
d8d5f3e1 | 958 | * |
5d4f98a2 | 959 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 960 | * |
5d4f98a2 YZ |
961 | * When a file extent is removed file truncation, we find the |
962 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 963 | * |
5d4f98a2 | 964 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 965 | * |
5d4f98a2 | 966 | * Btree extents can be referenced by: |
d8d5f3e1 | 967 | * |
5d4f98a2 | 968 | * - Different subvolumes |
d8d5f3e1 | 969 | * |
5d4f98a2 YZ |
970 | * Both the implicit back refs and the full back refs for tree blocks |
971 | * only consist of key. The key offset for the implicit back refs is | |
972 | * objectid of block's owner tree. The key offset for the full back refs | |
973 | * is the first byte of parent block. | |
d8d5f3e1 | 974 | * |
5d4f98a2 YZ |
975 | * When implicit back refs is used, information about the lowest key and |
976 | * level of the tree block are required. These information are stored in | |
977 | * tree block info structure. | |
d8d5f3e1 | 978 | */ |
31840ae1 | 979 | |
5d4f98a2 YZ |
980 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
981 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
982 | struct btrfs_root *root, | |
983 | struct btrfs_path *path, | |
984 | u64 owner, u32 extra_size) | |
7bb86316 | 985 | { |
5d4f98a2 YZ |
986 | struct btrfs_extent_item *item; |
987 | struct btrfs_extent_item_v0 *ei0; | |
988 | struct btrfs_extent_ref_v0 *ref0; | |
989 | struct btrfs_tree_block_info *bi; | |
990 | struct extent_buffer *leaf; | |
7bb86316 | 991 | struct btrfs_key key; |
5d4f98a2 YZ |
992 | struct btrfs_key found_key; |
993 | u32 new_size = sizeof(*item); | |
994 | u64 refs; | |
995 | int ret; | |
996 | ||
997 | leaf = path->nodes[0]; | |
998 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
999 | ||
1000 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1001 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1002 | struct btrfs_extent_item_v0); | |
1003 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1004 | ||
1005 | if (owner == (u64)-1) { | |
1006 | while (1) { | |
1007 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1008 | ret = btrfs_next_leaf(root, path); | |
1009 | if (ret < 0) | |
1010 | return ret; | |
79787eaa | 1011 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1012 | leaf = path->nodes[0]; |
1013 | } | |
1014 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1015 | path->slots[0]); | |
1016 | BUG_ON(key.objectid != found_key.objectid); | |
1017 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1018 | path->slots[0]++; | |
1019 | continue; | |
1020 | } | |
1021 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1022 | struct btrfs_extent_ref_v0); | |
1023 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1024 | break; | |
1025 | } | |
1026 | } | |
b3b4aa74 | 1027 | btrfs_release_path(path); |
5d4f98a2 YZ |
1028 | |
1029 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1030 | new_size += sizeof(*bi); | |
1031 | ||
1032 | new_size -= sizeof(*ei0); | |
1033 | ret = btrfs_search_slot(trans, root, &key, path, | |
1034 | new_size + extra_size, 1); | |
1035 | if (ret < 0) | |
1036 | return ret; | |
79787eaa | 1037 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1038 | |
4b90c680 | 1039 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1040 | |
1041 | leaf = path->nodes[0]; | |
1042 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1043 | btrfs_set_extent_refs(leaf, item, refs); | |
1044 | /* FIXME: get real generation */ | |
1045 | btrfs_set_extent_generation(leaf, item, 0); | |
1046 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1047 | btrfs_set_extent_flags(leaf, item, | |
1048 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1049 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1050 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1051 | /* FIXME: get first key of the block */ | |
1052 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1053 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1054 | } else { | |
1055 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1056 | } | |
1057 | btrfs_mark_buffer_dirty(leaf); | |
1058 | return 0; | |
1059 | } | |
1060 | #endif | |
1061 | ||
1062 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1063 | { | |
1064 | u32 high_crc = ~(u32)0; | |
1065 | u32 low_crc = ~(u32)0; | |
1066 | __le64 lenum; | |
1067 | ||
1068 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1069 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1070 | lenum = cpu_to_le64(owner); |
163e783e | 1071 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1072 | lenum = cpu_to_le64(offset); |
163e783e | 1073 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1074 | |
1075 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1076 | } | |
1077 | ||
1078 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1079 | struct btrfs_extent_data_ref *ref) | |
1080 | { | |
1081 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1082 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1083 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1084 | } | |
1085 | ||
1086 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1087 | struct btrfs_extent_data_ref *ref, | |
1088 | u64 root_objectid, u64 owner, u64 offset) | |
1089 | { | |
1090 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1091 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1092 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1093 | return 0; | |
1094 | return 1; | |
1095 | } | |
1096 | ||
1097 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1098 | struct btrfs_root *root, | |
1099 | struct btrfs_path *path, | |
1100 | u64 bytenr, u64 parent, | |
1101 | u64 root_objectid, | |
1102 | u64 owner, u64 offset) | |
1103 | { | |
1104 | struct btrfs_key key; | |
1105 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1106 | struct extent_buffer *leaf; |
5d4f98a2 | 1107 | u32 nritems; |
74493f7a | 1108 | int ret; |
5d4f98a2 YZ |
1109 | int recow; |
1110 | int err = -ENOENT; | |
74493f7a | 1111 | |
31840ae1 | 1112 | key.objectid = bytenr; |
5d4f98a2 YZ |
1113 | if (parent) { |
1114 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1115 | key.offset = parent; | |
1116 | } else { | |
1117 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1118 | key.offset = hash_extent_data_ref(root_objectid, | |
1119 | owner, offset); | |
1120 | } | |
1121 | again: | |
1122 | recow = 0; | |
1123 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1124 | if (ret < 0) { | |
1125 | err = ret; | |
1126 | goto fail; | |
1127 | } | |
31840ae1 | 1128 | |
5d4f98a2 YZ |
1129 | if (parent) { |
1130 | if (!ret) | |
1131 | return 0; | |
1132 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1133 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1134 | btrfs_release_path(path); |
5d4f98a2 YZ |
1135 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1136 | if (ret < 0) { | |
1137 | err = ret; | |
1138 | goto fail; | |
1139 | } | |
1140 | if (!ret) | |
1141 | return 0; | |
1142 | #endif | |
1143 | goto fail; | |
31840ae1 ZY |
1144 | } |
1145 | ||
1146 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1147 | nritems = btrfs_header_nritems(leaf); |
1148 | while (1) { | |
1149 | if (path->slots[0] >= nritems) { | |
1150 | ret = btrfs_next_leaf(root, path); | |
1151 | if (ret < 0) | |
1152 | err = ret; | |
1153 | if (ret) | |
1154 | goto fail; | |
1155 | ||
1156 | leaf = path->nodes[0]; | |
1157 | nritems = btrfs_header_nritems(leaf); | |
1158 | recow = 1; | |
1159 | } | |
1160 | ||
1161 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1162 | if (key.objectid != bytenr || | |
1163 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1164 | goto fail; | |
1165 | ||
1166 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1167 | struct btrfs_extent_data_ref); | |
1168 | ||
1169 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1170 | owner, offset)) { | |
1171 | if (recow) { | |
b3b4aa74 | 1172 | btrfs_release_path(path); |
5d4f98a2 YZ |
1173 | goto again; |
1174 | } | |
1175 | err = 0; | |
1176 | break; | |
1177 | } | |
1178 | path->slots[0]++; | |
31840ae1 | 1179 | } |
5d4f98a2 YZ |
1180 | fail: |
1181 | return err; | |
31840ae1 ZY |
1182 | } |
1183 | ||
5d4f98a2 YZ |
1184 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1185 | struct btrfs_root *root, | |
1186 | struct btrfs_path *path, | |
1187 | u64 bytenr, u64 parent, | |
1188 | u64 root_objectid, u64 owner, | |
1189 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1190 | { |
1191 | struct btrfs_key key; | |
1192 | struct extent_buffer *leaf; | |
5d4f98a2 | 1193 | u32 size; |
31840ae1 ZY |
1194 | u32 num_refs; |
1195 | int ret; | |
74493f7a | 1196 | |
74493f7a | 1197 | key.objectid = bytenr; |
5d4f98a2 YZ |
1198 | if (parent) { |
1199 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1200 | key.offset = parent; | |
1201 | size = sizeof(struct btrfs_shared_data_ref); | |
1202 | } else { | |
1203 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1204 | key.offset = hash_extent_data_ref(root_objectid, | |
1205 | owner, offset); | |
1206 | size = sizeof(struct btrfs_extent_data_ref); | |
1207 | } | |
74493f7a | 1208 | |
5d4f98a2 YZ |
1209 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1210 | if (ret && ret != -EEXIST) | |
1211 | goto fail; | |
1212 | ||
1213 | leaf = path->nodes[0]; | |
1214 | if (parent) { | |
1215 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1216 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1217 | struct btrfs_shared_data_ref); |
1218 | if (ret == 0) { | |
1219 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1220 | } else { | |
1221 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1222 | num_refs += refs_to_add; | |
1223 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1224 | } |
5d4f98a2 YZ |
1225 | } else { |
1226 | struct btrfs_extent_data_ref *ref; | |
1227 | while (ret == -EEXIST) { | |
1228 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1229 | struct btrfs_extent_data_ref); | |
1230 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1231 | owner, offset)) | |
1232 | break; | |
b3b4aa74 | 1233 | btrfs_release_path(path); |
5d4f98a2 YZ |
1234 | key.offset++; |
1235 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1236 | size); | |
1237 | if (ret && ret != -EEXIST) | |
1238 | goto fail; | |
31840ae1 | 1239 | |
5d4f98a2 YZ |
1240 | leaf = path->nodes[0]; |
1241 | } | |
1242 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1243 | struct btrfs_extent_data_ref); | |
1244 | if (ret == 0) { | |
1245 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1246 | root_objectid); | |
1247 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1248 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1249 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1250 | } else { | |
1251 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1252 | num_refs += refs_to_add; | |
1253 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1254 | } |
31840ae1 | 1255 | } |
5d4f98a2 YZ |
1256 | btrfs_mark_buffer_dirty(leaf); |
1257 | ret = 0; | |
1258 | fail: | |
b3b4aa74 | 1259 | btrfs_release_path(path); |
7bb86316 | 1260 | return ret; |
74493f7a CM |
1261 | } |
1262 | ||
5d4f98a2 YZ |
1263 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1264 | struct btrfs_root *root, | |
1265 | struct btrfs_path *path, | |
1266 | int refs_to_drop) | |
31840ae1 | 1267 | { |
5d4f98a2 YZ |
1268 | struct btrfs_key key; |
1269 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1270 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1271 | struct extent_buffer *leaf; |
5d4f98a2 | 1272 | u32 num_refs = 0; |
31840ae1 ZY |
1273 | int ret = 0; |
1274 | ||
1275 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1276 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1277 | ||
1278 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1279 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1280 | struct btrfs_extent_data_ref); | |
1281 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1282 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1283 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1284 | struct btrfs_shared_data_ref); | |
1285 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1286 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1287 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1288 | struct btrfs_extent_ref_v0 *ref0; | |
1289 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1290 | struct btrfs_extent_ref_v0); | |
1291 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1292 | #endif | |
1293 | } else { | |
1294 | BUG(); | |
1295 | } | |
1296 | ||
56bec294 CM |
1297 | BUG_ON(num_refs < refs_to_drop); |
1298 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1299 | |
31840ae1 ZY |
1300 | if (num_refs == 0) { |
1301 | ret = btrfs_del_item(trans, root, path); | |
1302 | } else { | |
5d4f98a2 YZ |
1303 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1304 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1305 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1306 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1307 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1308 | else { | |
1309 | struct btrfs_extent_ref_v0 *ref0; | |
1310 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1311 | struct btrfs_extent_ref_v0); | |
1312 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1313 | } | |
1314 | #endif | |
31840ae1 ZY |
1315 | btrfs_mark_buffer_dirty(leaf); |
1316 | } | |
31840ae1 ZY |
1317 | return ret; |
1318 | } | |
1319 | ||
5d4f98a2 YZ |
1320 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1321 | struct btrfs_path *path, | |
1322 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1323 | { |
5d4f98a2 YZ |
1324 | struct btrfs_key key; |
1325 | struct extent_buffer *leaf; | |
1326 | struct btrfs_extent_data_ref *ref1; | |
1327 | struct btrfs_shared_data_ref *ref2; | |
1328 | u32 num_refs = 0; | |
1329 | ||
1330 | leaf = path->nodes[0]; | |
1331 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1332 | if (iref) { | |
1333 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1334 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1335 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1336 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1337 | } else { | |
1338 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1339 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1340 | } | |
1341 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1342 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1343 | struct btrfs_extent_data_ref); | |
1344 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1345 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1346 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1347 | struct btrfs_shared_data_ref); | |
1348 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1349 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1350 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1351 | struct btrfs_extent_ref_v0 *ref0; | |
1352 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1353 | struct btrfs_extent_ref_v0); | |
1354 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1355 | #endif |
5d4f98a2 YZ |
1356 | } else { |
1357 | WARN_ON(1); | |
1358 | } | |
1359 | return num_refs; | |
1360 | } | |
15916de8 | 1361 | |
5d4f98a2 YZ |
1362 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1363 | struct btrfs_root *root, | |
1364 | struct btrfs_path *path, | |
1365 | u64 bytenr, u64 parent, | |
1366 | u64 root_objectid) | |
1f3c79a2 | 1367 | { |
5d4f98a2 | 1368 | struct btrfs_key key; |
1f3c79a2 | 1369 | int ret; |
1f3c79a2 | 1370 | |
5d4f98a2 YZ |
1371 | key.objectid = bytenr; |
1372 | if (parent) { | |
1373 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1374 | key.offset = parent; | |
1375 | } else { | |
1376 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1377 | key.offset = root_objectid; | |
1f3c79a2 LH |
1378 | } |
1379 | ||
5d4f98a2 YZ |
1380 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1381 | if (ret > 0) | |
1382 | ret = -ENOENT; | |
1383 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1384 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1385 | btrfs_release_path(path); |
5d4f98a2 YZ |
1386 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1387 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1388 | if (ret > 0) | |
1389 | ret = -ENOENT; | |
1390 | } | |
1f3c79a2 | 1391 | #endif |
5d4f98a2 | 1392 | return ret; |
1f3c79a2 LH |
1393 | } |
1394 | ||
5d4f98a2 YZ |
1395 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1396 | struct btrfs_root *root, | |
1397 | struct btrfs_path *path, | |
1398 | u64 bytenr, u64 parent, | |
1399 | u64 root_objectid) | |
31840ae1 | 1400 | { |
5d4f98a2 | 1401 | struct btrfs_key key; |
31840ae1 | 1402 | int ret; |
31840ae1 | 1403 | |
5d4f98a2 YZ |
1404 | key.objectid = bytenr; |
1405 | if (parent) { | |
1406 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1407 | key.offset = parent; | |
1408 | } else { | |
1409 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1410 | key.offset = root_objectid; | |
1411 | } | |
1412 | ||
1413 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1414 | btrfs_release_path(path); |
31840ae1 ZY |
1415 | return ret; |
1416 | } | |
1417 | ||
5d4f98a2 | 1418 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1419 | { |
5d4f98a2 YZ |
1420 | int type; |
1421 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1422 | if (parent > 0) | |
1423 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1424 | else | |
1425 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1426 | } else { | |
1427 | if (parent > 0) | |
1428 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1429 | else | |
1430 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1431 | } | |
1432 | return type; | |
31840ae1 | 1433 | } |
56bec294 | 1434 | |
2c47e605 YZ |
1435 | static int find_next_key(struct btrfs_path *path, int level, |
1436 | struct btrfs_key *key) | |
56bec294 | 1437 | |
02217ed2 | 1438 | { |
2c47e605 | 1439 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1440 | if (!path->nodes[level]) |
1441 | break; | |
5d4f98a2 YZ |
1442 | if (path->slots[level] + 1 >= |
1443 | btrfs_header_nritems(path->nodes[level])) | |
1444 | continue; | |
1445 | if (level == 0) | |
1446 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1447 | path->slots[level] + 1); | |
1448 | else | |
1449 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1450 | path->slots[level] + 1); | |
1451 | return 0; | |
1452 | } | |
1453 | return 1; | |
1454 | } | |
037e6390 | 1455 | |
5d4f98a2 YZ |
1456 | /* |
1457 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1458 | * to the address of inline back ref, and 0 is returned. | |
1459 | * | |
1460 | * if back ref isn't found, *ref_ret is set to the address where it | |
1461 | * should be inserted, and -ENOENT is returned. | |
1462 | * | |
1463 | * if insert is true and there are too many inline back refs, the path | |
1464 | * points to the extent item, and -EAGAIN is returned. | |
1465 | * | |
1466 | * NOTE: inline back refs are ordered in the same way that back ref | |
1467 | * items in the tree are ordered. | |
1468 | */ | |
1469 | static noinline_for_stack | |
1470 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1471 | struct btrfs_root *root, | |
1472 | struct btrfs_path *path, | |
1473 | struct btrfs_extent_inline_ref **ref_ret, | |
1474 | u64 bytenr, u64 num_bytes, | |
1475 | u64 parent, u64 root_objectid, | |
1476 | u64 owner, u64 offset, int insert) | |
1477 | { | |
1478 | struct btrfs_key key; | |
1479 | struct extent_buffer *leaf; | |
1480 | struct btrfs_extent_item *ei; | |
1481 | struct btrfs_extent_inline_ref *iref; | |
1482 | u64 flags; | |
1483 | u64 item_size; | |
1484 | unsigned long ptr; | |
1485 | unsigned long end; | |
1486 | int extra_size; | |
1487 | int type; | |
1488 | int want; | |
1489 | int ret; | |
1490 | int err = 0; | |
3173a18f JB |
1491 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1492 | SKINNY_METADATA); | |
26b8003f | 1493 | |
db94535d | 1494 | key.objectid = bytenr; |
31840ae1 | 1495 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1496 | key.offset = num_bytes; |
31840ae1 | 1497 | |
5d4f98a2 YZ |
1498 | want = extent_ref_type(parent, owner); |
1499 | if (insert) { | |
1500 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1501 | path->keep_locks = 1; |
5d4f98a2 YZ |
1502 | } else |
1503 | extra_size = -1; | |
3173a18f JB |
1504 | |
1505 | /* | |
1506 | * Owner is our parent level, so we can just add one to get the level | |
1507 | * for the block we are interested in. | |
1508 | */ | |
1509 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1510 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1511 | key.offset = owner; | |
1512 | } | |
1513 | ||
1514 | again: | |
5d4f98a2 | 1515 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1516 | if (ret < 0) { |
5d4f98a2 YZ |
1517 | err = ret; |
1518 | goto out; | |
1519 | } | |
3173a18f JB |
1520 | |
1521 | /* | |
1522 | * We may be a newly converted file system which still has the old fat | |
1523 | * extent entries for metadata, so try and see if we have one of those. | |
1524 | */ | |
1525 | if (ret > 0 && skinny_metadata) { | |
1526 | skinny_metadata = false; | |
1527 | if (path->slots[0]) { | |
1528 | path->slots[0]--; | |
1529 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1530 | path->slots[0]); | |
1531 | if (key.objectid == bytenr && | |
1532 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1533 | key.offset == num_bytes) | |
1534 | ret = 0; | |
1535 | } | |
1536 | if (ret) { | |
1537 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1538 | key.offset = num_bytes; | |
1539 | btrfs_release_path(path); | |
1540 | goto again; | |
1541 | } | |
1542 | } | |
1543 | ||
79787eaa JM |
1544 | if (ret && !insert) { |
1545 | err = -ENOENT; | |
1546 | goto out; | |
492104c8 JB |
1547 | } else if (ret) { |
1548 | err = -EIO; | |
1549 | WARN_ON(1); | |
1550 | goto out; | |
79787eaa | 1551 | } |
5d4f98a2 YZ |
1552 | |
1553 | leaf = path->nodes[0]; | |
1554 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1555 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1556 | if (item_size < sizeof(*ei)) { | |
1557 | if (!insert) { | |
1558 | err = -ENOENT; | |
1559 | goto out; | |
1560 | } | |
1561 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1562 | extra_size); | |
1563 | if (ret < 0) { | |
1564 | err = ret; | |
1565 | goto out; | |
1566 | } | |
1567 | leaf = path->nodes[0]; | |
1568 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1569 | } | |
1570 | #endif | |
1571 | BUG_ON(item_size < sizeof(*ei)); | |
1572 | ||
5d4f98a2 YZ |
1573 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1574 | flags = btrfs_extent_flags(leaf, ei); | |
1575 | ||
1576 | ptr = (unsigned long)(ei + 1); | |
1577 | end = (unsigned long)ei + item_size; | |
1578 | ||
3173a18f | 1579 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1580 | ptr += sizeof(struct btrfs_tree_block_info); |
1581 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1582 | } |
1583 | ||
1584 | err = -ENOENT; | |
1585 | while (1) { | |
1586 | if (ptr >= end) { | |
1587 | WARN_ON(ptr > end); | |
1588 | break; | |
1589 | } | |
1590 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1591 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1592 | if (want < type) | |
1593 | break; | |
1594 | if (want > type) { | |
1595 | ptr += btrfs_extent_inline_ref_size(type); | |
1596 | continue; | |
1597 | } | |
1598 | ||
1599 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1600 | struct btrfs_extent_data_ref *dref; | |
1601 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1602 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1603 | owner, offset)) { | |
1604 | err = 0; | |
1605 | break; | |
1606 | } | |
1607 | if (hash_extent_data_ref_item(leaf, dref) < | |
1608 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1609 | break; | |
1610 | } else { | |
1611 | u64 ref_offset; | |
1612 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1613 | if (parent > 0) { | |
1614 | if (parent == ref_offset) { | |
1615 | err = 0; | |
1616 | break; | |
1617 | } | |
1618 | if (ref_offset < parent) | |
1619 | break; | |
1620 | } else { | |
1621 | if (root_objectid == ref_offset) { | |
1622 | err = 0; | |
1623 | break; | |
1624 | } | |
1625 | if (ref_offset < root_objectid) | |
1626 | break; | |
1627 | } | |
1628 | } | |
1629 | ptr += btrfs_extent_inline_ref_size(type); | |
1630 | } | |
1631 | if (err == -ENOENT && insert) { | |
1632 | if (item_size + extra_size >= | |
1633 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1634 | err = -EAGAIN; | |
1635 | goto out; | |
1636 | } | |
1637 | /* | |
1638 | * To add new inline back ref, we have to make sure | |
1639 | * there is no corresponding back ref item. | |
1640 | * For simplicity, we just do not add new inline back | |
1641 | * ref if there is any kind of item for this block | |
1642 | */ | |
2c47e605 YZ |
1643 | if (find_next_key(path, 0, &key) == 0 && |
1644 | key.objectid == bytenr && | |
85d4198e | 1645 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1646 | err = -EAGAIN; |
1647 | goto out; | |
1648 | } | |
1649 | } | |
1650 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1651 | out: | |
85d4198e | 1652 | if (insert) { |
5d4f98a2 YZ |
1653 | path->keep_locks = 0; |
1654 | btrfs_unlock_up_safe(path, 1); | |
1655 | } | |
1656 | return err; | |
1657 | } | |
1658 | ||
1659 | /* | |
1660 | * helper to add new inline back ref | |
1661 | */ | |
1662 | static noinline_for_stack | |
fd279fae | 1663 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1664 | struct btrfs_path *path, |
1665 | struct btrfs_extent_inline_ref *iref, | |
1666 | u64 parent, u64 root_objectid, | |
1667 | u64 owner, u64 offset, int refs_to_add, | |
1668 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1669 | { |
1670 | struct extent_buffer *leaf; | |
1671 | struct btrfs_extent_item *ei; | |
1672 | unsigned long ptr; | |
1673 | unsigned long end; | |
1674 | unsigned long item_offset; | |
1675 | u64 refs; | |
1676 | int size; | |
1677 | int type; | |
5d4f98a2 YZ |
1678 | |
1679 | leaf = path->nodes[0]; | |
1680 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1681 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1682 | ||
1683 | type = extent_ref_type(parent, owner); | |
1684 | size = btrfs_extent_inline_ref_size(type); | |
1685 | ||
4b90c680 | 1686 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1687 | |
1688 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1689 | refs = btrfs_extent_refs(leaf, ei); | |
1690 | refs += refs_to_add; | |
1691 | btrfs_set_extent_refs(leaf, ei, refs); | |
1692 | if (extent_op) | |
1693 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1694 | ||
1695 | ptr = (unsigned long)ei + item_offset; | |
1696 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1697 | if (ptr < end - size) | |
1698 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1699 | end - size - ptr); | |
1700 | ||
1701 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1702 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1703 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1704 | struct btrfs_extent_data_ref *dref; | |
1705 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1706 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1707 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1708 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1709 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1710 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1711 | struct btrfs_shared_data_ref *sref; | |
1712 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1713 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1714 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1715 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1716 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1717 | } else { | |
1718 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1719 | } | |
1720 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1721 | } |
1722 | ||
1723 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1724 | struct btrfs_root *root, | |
1725 | struct btrfs_path *path, | |
1726 | struct btrfs_extent_inline_ref **ref_ret, | |
1727 | u64 bytenr, u64 num_bytes, u64 parent, | |
1728 | u64 root_objectid, u64 owner, u64 offset) | |
1729 | { | |
1730 | int ret; | |
1731 | ||
1732 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1733 | bytenr, num_bytes, parent, | |
1734 | root_objectid, owner, offset, 0); | |
1735 | if (ret != -ENOENT) | |
54aa1f4d | 1736 | return ret; |
5d4f98a2 | 1737 | |
b3b4aa74 | 1738 | btrfs_release_path(path); |
5d4f98a2 YZ |
1739 | *ref_ret = NULL; |
1740 | ||
1741 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1742 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1743 | root_objectid); | |
1744 | } else { | |
1745 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1746 | root_objectid, owner, offset); | |
b9473439 | 1747 | } |
5d4f98a2 YZ |
1748 | return ret; |
1749 | } | |
31840ae1 | 1750 | |
5d4f98a2 YZ |
1751 | /* |
1752 | * helper to update/remove inline back ref | |
1753 | */ | |
1754 | static noinline_for_stack | |
afe5fea7 | 1755 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1756 | struct btrfs_path *path, |
1757 | struct btrfs_extent_inline_ref *iref, | |
1758 | int refs_to_mod, | |
1759 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1760 | { |
1761 | struct extent_buffer *leaf; | |
1762 | struct btrfs_extent_item *ei; | |
1763 | struct btrfs_extent_data_ref *dref = NULL; | |
1764 | struct btrfs_shared_data_ref *sref = NULL; | |
1765 | unsigned long ptr; | |
1766 | unsigned long end; | |
1767 | u32 item_size; | |
1768 | int size; | |
1769 | int type; | |
5d4f98a2 YZ |
1770 | u64 refs; |
1771 | ||
1772 | leaf = path->nodes[0]; | |
1773 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1774 | refs = btrfs_extent_refs(leaf, ei); | |
1775 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1776 | refs += refs_to_mod; | |
1777 | btrfs_set_extent_refs(leaf, ei, refs); | |
1778 | if (extent_op) | |
1779 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1780 | ||
1781 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1782 | ||
1783 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1784 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1785 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1786 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1787 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1788 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1789 | } else { | |
1790 | refs = 1; | |
1791 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1792 | } |
31840ae1 | 1793 | |
5d4f98a2 YZ |
1794 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1795 | refs += refs_to_mod; | |
1796 | ||
1797 | if (refs > 0) { | |
1798 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1799 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1800 | else | |
1801 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1802 | } else { | |
1803 | size = btrfs_extent_inline_ref_size(type); | |
1804 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1805 | ptr = (unsigned long)iref; | |
1806 | end = (unsigned long)ei + item_size; | |
1807 | if (ptr + size < end) | |
1808 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1809 | end - ptr - size); | |
1810 | item_size -= size; | |
afe5fea7 | 1811 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1812 | } |
1813 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1814 | } |
1815 | ||
1816 | static noinline_for_stack | |
1817 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1818 | struct btrfs_root *root, | |
1819 | struct btrfs_path *path, | |
1820 | u64 bytenr, u64 num_bytes, u64 parent, | |
1821 | u64 root_objectid, u64 owner, | |
1822 | u64 offset, int refs_to_add, | |
1823 | struct btrfs_delayed_extent_op *extent_op) | |
1824 | { | |
1825 | struct btrfs_extent_inline_ref *iref; | |
1826 | int ret; | |
1827 | ||
1828 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1829 | bytenr, num_bytes, parent, | |
1830 | root_objectid, owner, offset, 1); | |
1831 | if (ret == 0) { | |
1832 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1833 | update_inline_extent_backref(root, path, iref, |
143bede5 | 1834 | refs_to_add, extent_op); |
5d4f98a2 | 1835 | } else if (ret == -ENOENT) { |
fd279fae | 1836 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1837 | root_objectid, owner, offset, |
1838 | refs_to_add, extent_op); | |
1839 | ret = 0; | |
771ed689 | 1840 | } |
5d4f98a2 YZ |
1841 | return ret; |
1842 | } | |
31840ae1 | 1843 | |
5d4f98a2 YZ |
1844 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1845 | struct btrfs_root *root, | |
1846 | struct btrfs_path *path, | |
1847 | u64 bytenr, u64 parent, u64 root_objectid, | |
1848 | u64 owner, u64 offset, int refs_to_add) | |
1849 | { | |
1850 | int ret; | |
1851 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1852 | BUG_ON(refs_to_add != 1); | |
1853 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1854 | parent, root_objectid); | |
1855 | } else { | |
1856 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1857 | parent, root_objectid, | |
1858 | owner, offset, refs_to_add); | |
1859 | } | |
1860 | return ret; | |
1861 | } | |
56bec294 | 1862 | |
5d4f98a2 YZ |
1863 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1864 | struct btrfs_root *root, | |
1865 | struct btrfs_path *path, | |
1866 | struct btrfs_extent_inline_ref *iref, | |
1867 | int refs_to_drop, int is_data) | |
1868 | { | |
143bede5 | 1869 | int ret = 0; |
b9473439 | 1870 | |
5d4f98a2 YZ |
1871 | BUG_ON(!is_data && refs_to_drop != 1); |
1872 | if (iref) { | |
afe5fea7 | 1873 | update_inline_extent_backref(root, path, iref, |
143bede5 | 1874 | -refs_to_drop, NULL); |
5d4f98a2 YZ |
1875 | } else if (is_data) { |
1876 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1877 | } else { | |
1878 | ret = btrfs_del_item(trans, root, path); | |
1879 | } | |
1880 | return ret; | |
1881 | } | |
1882 | ||
5378e607 | 1883 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1884 | u64 start, u64 len) |
1885 | { | |
5378e607 | 1886 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1887 | } |
5d4f98a2 YZ |
1888 | |
1889 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1890 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1891 | { |
5d4f98a2 | 1892 | int ret; |
5378e607 | 1893 | u64 discarded_bytes = 0; |
a1d3c478 | 1894 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1895 | |
e244a0ae | 1896 | |
5d4f98a2 | 1897 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1898 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1899 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1900 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1901 | if (!ret) { |
a1d3c478 | 1902 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1903 | int i; |
1904 | ||
5d4f98a2 | 1905 | |
a1d3c478 | 1906 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1907 | if (!stripe->dev->can_discard) |
1908 | continue; | |
1909 | ||
5378e607 LD |
1910 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1911 | stripe->physical, | |
1912 | stripe->length); | |
1913 | if (!ret) | |
1914 | discarded_bytes += stripe->length; | |
1915 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1916 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1917 | |
1918 | /* | |
1919 | * Just in case we get back EOPNOTSUPP for some reason, | |
1920 | * just ignore the return value so we don't screw up | |
1921 | * people calling discard_extent. | |
1922 | */ | |
1923 | ret = 0; | |
5d4f98a2 | 1924 | } |
a1d3c478 | 1925 | kfree(bbio); |
5d4f98a2 | 1926 | } |
5378e607 LD |
1927 | |
1928 | if (actual_bytes) | |
1929 | *actual_bytes = discarded_bytes; | |
1930 | ||
5d4f98a2 | 1931 | |
53b381b3 DW |
1932 | if (ret == -EOPNOTSUPP) |
1933 | ret = 0; | |
5d4f98a2 | 1934 | return ret; |
5d4f98a2 YZ |
1935 | } |
1936 | ||
79787eaa | 1937 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1938 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1939 | struct btrfs_root *root, | |
1940 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1941 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1942 | { |
1943 | int ret; | |
66d7e7f0 AJ |
1944 | struct btrfs_fs_info *fs_info = root->fs_info; |
1945 | ||
5d4f98a2 YZ |
1946 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1947 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1948 | ||
1949 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1950 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1951 | num_bytes, | |
5d4f98a2 | 1952 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1953 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1954 | } else { |
66d7e7f0 AJ |
1955 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1956 | num_bytes, | |
5d4f98a2 | 1957 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1958 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1959 | } |
1960 | return ret; | |
1961 | } | |
1962 | ||
1963 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1964 | struct btrfs_root *root, | |
1965 | u64 bytenr, u64 num_bytes, | |
1966 | u64 parent, u64 root_objectid, | |
1967 | u64 owner, u64 offset, int refs_to_add, | |
1968 | struct btrfs_delayed_extent_op *extent_op) | |
1969 | { | |
1970 | struct btrfs_path *path; | |
1971 | struct extent_buffer *leaf; | |
1972 | struct btrfs_extent_item *item; | |
1973 | u64 refs; | |
1974 | int ret; | |
1975 | int err = 0; | |
1976 | ||
1977 | path = btrfs_alloc_path(); | |
1978 | if (!path) | |
1979 | return -ENOMEM; | |
1980 | ||
1981 | path->reada = 1; | |
1982 | path->leave_spinning = 1; | |
1983 | /* this will setup the path even if it fails to insert the back ref */ | |
1984 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1985 | path, bytenr, num_bytes, parent, | |
1986 | root_objectid, owner, offset, | |
1987 | refs_to_add, extent_op); | |
1988 | if (ret == 0) | |
1989 | goto out; | |
1990 | ||
1991 | if (ret != -EAGAIN) { | |
1992 | err = ret; | |
1993 | goto out; | |
1994 | } | |
1995 | ||
1996 | leaf = path->nodes[0]; | |
1997 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1998 | refs = btrfs_extent_refs(leaf, item); | |
1999 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2000 | if (extent_op) | |
2001 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2002 | |
5d4f98a2 | 2003 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2004 | btrfs_release_path(path); |
56bec294 CM |
2005 | |
2006 | path->reada = 1; | |
b9473439 CM |
2007 | path->leave_spinning = 1; |
2008 | ||
56bec294 CM |
2009 | /* now insert the actual backref */ |
2010 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2011 | path, bytenr, parent, root_objectid, |
2012 | owner, offset, refs_to_add); | |
79787eaa JM |
2013 | if (ret) |
2014 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2015 | out: |
56bec294 | 2016 | btrfs_free_path(path); |
5d4f98a2 | 2017 | return err; |
56bec294 CM |
2018 | } |
2019 | ||
5d4f98a2 YZ |
2020 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2021 | struct btrfs_root *root, | |
2022 | struct btrfs_delayed_ref_node *node, | |
2023 | struct btrfs_delayed_extent_op *extent_op, | |
2024 | int insert_reserved) | |
56bec294 | 2025 | { |
5d4f98a2 YZ |
2026 | int ret = 0; |
2027 | struct btrfs_delayed_data_ref *ref; | |
2028 | struct btrfs_key ins; | |
2029 | u64 parent = 0; | |
2030 | u64 ref_root = 0; | |
2031 | u64 flags = 0; | |
2032 | ||
2033 | ins.objectid = node->bytenr; | |
2034 | ins.offset = node->num_bytes; | |
2035 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2036 | ||
2037 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2038 | trace_run_delayed_data_ref(node, ref, node->action); |
2039 | ||
5d4f98a2 YZ |
2040 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2041 | parent = ref->parent; | |
2042 | else | |
2043 | ref_root = ref->root; | |
2044 | ||
2045 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2046 | if (extent_op) |
5d4f98a2 | 2047 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2048 | ret = alloc_reserved_file_extent(trans, root, |
2049 | parent, ref_root, flags, | |
2050 | ref->objectid, ref->offset, | |
2051 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
2052 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2053 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2054 | node->num_bytes, parent, | |
2055 | ref_root, ref->objectid, | |
2056 | ref->offset, node->ref_mod, | |
2057 | extent_op); | |
2058 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2059 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2060 | node->num_bytes, parent, | |
2061 | ref_root, ref->objectid, | |
2062 | ref->offset, node->ref_mod, | |
2063 | extent_op); | |
2064 | } else { | |
2065 | BUG(); | |
2066 | } | |
2067 | return ret; | |
2068 | } | |
2069 | ||
2070 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2071 | struct extent_buffer *leaf, | |
2072 | struct btrfs_extent_item *ei) | |
2073 | { | |
2074 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2075 | if (extent_op->update_flags) { | |
2076 | flags |= extent_op->flags_to_set; | |
2077 | btrfs_set_extent_flags(leaf, ei, flags); | |
2078 | } | |
2079 | ||
2080 | if (extent_op->update_key) { | |
2081 | struct btrfs_tree_block_info *bi; | |
2082 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2083 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2084 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2085 | } | |
2086 | } | |
2087 | ||
2088 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2089 | struct btrfs_root *root, | |
2090 | struct btrfs_delayed_ref_node *node, | |
2091 | struct btrfs_delayed_extent_op *extent_op) | |
2092 | { | |
2093 | struct btrfs_key key; | |
2094 | struct btrfs_path *path; | |
2095 | struct btrfs_extent_item *ei; | |
2096 | struct extent_buffer *leaf; | |
2097 | u32 item_size; | |
56bec294 | 2098 | int ret; |
5d4f98a2 | 2099 | int err = 0; |
b1c79e09 | 2100 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2101 | |
79787eaa JM |
2102 | if (trans->aborted) |
2103 | return 0; | |
2104 | ||
3173a18f JB |
2105 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2106 | metadata = 0; | |
2107 | ||
5d4f98a2 YZ |
2108 | path = btrfs_alloc_path(); |
2109 | if (!path) | |
2110 | return -ENOMEM; | |
2111 | ||
2112 | key.objectid = node->bytenr; | |
5d4f98a2 | 2113 | |
3173a18f | 2114 | if (metadata) { |
3173a18f | 2115 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2116 | key.offset = extent_op->level; |
3173a18f JB |
2117 | } else { |
2118 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2119 | key.offset = node->num_bytes; | |
2120 | } | |
2121 | ||
2122 | again: | |
5d4f98a2 YZ |
2123 | path->reada = 1; |
2124 | path->leave_spinning = 1; | |
2125 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2126 | path, 0, 1); | |
2127 | if (ret < 0) { | |
2128 | err = ret; | |
2129 | goto out; | |
2130 | } | |
2131 | if (ret > 0) { | |
3173a18f JB |
2132 | if (metadata) { |
2133 | btrfs_release_path(path); | |
2134 | metadata = 0; | |
2135 | ||
2136 | key.offset = node->num_bytes; | |
2137 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2138 | goto again; | |
2139 | } | |
5d4f98a2 YZ |
2140 | err = -EIO; |
2141 | goto out; | |
2142 | } | |
2143 | ||
2144 | leaf = path->nodes[0]; | |
2145 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2146 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2147 | if (item_size < sizeof(*ei)) { | |
2148 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2149 | path, (u64)-1, 0); | |
2150 | if (ret < 0) { | |
2151 | err = ret; | |
2152 | goto out; | |
2153 | } | |
2154 | leaf = path->nodes[0]; | |
2155 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2156 | } | |
2157 | #endif | |
2158 | BUG_ON(item_size < sizeof(*ei)); | |
2159 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2160 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2161 | |
5d4f98a2 YZ |
2162 | btrfs_mark_buffer_dirty(leaf); |
2163 | out: | |
2164 | btrfs_free_path(path); | |
2165 | return err; | |
56bec294 CM |
2166 | } |
2167 | ||
5d4f98a2 YZ |
2168 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2169 | struct btrfs_root *root, | |
2170 | struct btrfs_delayed_ref_node *node, | |
2171 | struct btrfs_delayed_extent_op *extent_op, | |
2172 | int insert_reserved) | |
56bec294 CM |
2173 | { |
2174 | int ret = 0; | |
5d4f98a2 YZ |
2175 | struct btrfs_delayed_tree_ref *ref; |
2176 | struct btrfs_key ins; | |
2177 | u64 parent = 0; | |
2178 | u64 ref_root = 0; | |
3173a18f JB |
2179 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2180 | SKINNY_METADATA); | |
56bec294 | 2181 | |
5d4f98a2 | 2182 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2183 | trace_run_delayed_tree_ref(node, ref, node->action); |
2184 | ||
5d4f98a2 YZ |
2185 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2186 | parent = ref->parent; | |
2187 | else | |
2188 | ref_root = ref->root; | |
2189 | ||
3173a18f JB |
2190 | ins.objectid = node->bytenr; |
2191 | if (skinny_metadata) { | |
2192 | ins.offset = ref->level; | |
2193 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2194 | } else { | |
2195 | ins.offset = node->num_bytes; | |
2196 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2197 | } | |
2198 | ||
5d4f98a2 YZ |
2199 | BUG_ON(node->ref_mod != 1); |
2200 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2201 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2202 | ret = alloc_reserved_tree_block(trans, root, |
2203 | parent, ref_root, | |
2204 | extent_op->flags_to_set, | |
2205 | &extent_op->key, | |
2206 | ref->level, &ins); | |
5d4f98a2 YZ |
2207 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2208 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2209 | node->num_bytes, parent, ref_root, | |
2210 | ref->level, 0, 1, extent_op); | |
2211 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2212 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2213 | node->num_bytes, parent, ref_root, | |
2214 | ref->level, 0, 1, extent_op); | |
2215 | } else { | |
2216 | BUG(); | |
2217 | } | |
56bec294 CM |
2218 | return ret; |
2219 | } | |
2220 | ||
2221 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2222 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2223 | struct btrfs_root *root, | |
2224 | struct btrfs_delayed_ref_node *node, | |
2225 | struct btrfs_delayed_extent_op *extent_op, | |
2226 | int insert_reserved) | |
56bec294 | 2227 | { |
79787eaa JM |
2228 | int ret = 0; |
2229 | ||
2230 | if (trans->aborted) | |
2231 | return 0; | |
2232 | ||
5d4f98a2 | 2233 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2234 | struct btrfs_delayed_ref_head *head; |
2235 | /* | |
2236 | * we've hit the end of the chain and we were supposed | |
2237 | * to insert this extent into the tree. But, it got | |
2238 | * deleted before we ever needed to insert it, so all | |
2239 | * we have to do is clean up the accounting | |
2240 | */ | |
5d4f98a2 YZ |
2241 | BUG_ON(extent_op); |
2242 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2243 | trace_run_delayed_ref_head(node, head, node->action); |
2244 | ||
56bec294 | 2245 | if (insert_reserved) { |
f0486c68 YZ |
2246 | btrfs_pin_extent(root, node->bytenr, |
2247 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2248 | if (head->is_data) { |
2249 | ret = btrfs_del_csums(trans, root, | |
2250 | node->bytenr, | |
2251 | node->num_bytes); | |
5d4f98a2 | 2252 | } |
56bec294 | 2253 | } |
79787eaa | 2254 | return ret; |
56bec294 CM |
2255 | } |
2256 | ||
5d4f98a2 YZ |
2257 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2258 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2259 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2260 | insert_reserved); | |
2261 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2262 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2263 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2264 | insert_reserved); | |
2265 | else | |
2266 | BUG(); | |
2267 | return ret; | |
56bec294 CM |
2268 | } |
2269 | ||
2270 | static noinline struct btrfs_delayed_ref_node * | |
2271 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2272 | { | |
2273 | struct rb_node *node; | |
2274 | struct btrfs_delayed_ref_node *ref; | |
2275 | int action = BTRFS_ADD_DELAYED_REF; | |
2276 | again: | |
2277 | /* | |
2278 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2279 | * this prevents ref count from going down to zero when | |
2280 | * there still are pending delayed ref. | |
2281 | */ | |
2282 | node = rb_prev(&head->node.rb_node); | |
2283 | while (1) { | |
2284 | if (!node) | |
2285 | break; | |
2286 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2287 | rb_node); | |
2288 | if (ref->bytenr != head->node.bytenr) | |
2289 | break; | |
5d4f98a2 | 2290 | if (ref->action == action) |
56bec294 CM |
2291 | return ref; |
2292 | node = rb_prev(node); | |
2293 | } | |
2294 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2295 | action = BTRFS_DROP_DELAYED_REF; | |
2296 | goto again; | |
2297 | } | |
2298 | return NULL; | |
2299 | } | |
2300 | ||
79787eaa JM |
2301 | /* |
2302 | * Returns 0 on success or if called with an already aborted transaction. | |
2303 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2304 | */ | |
c3e69d58 CM |
2305 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2306 | struct btrfs_root *root, | |
2307 | struct list_head *cluster) | |
56bec294 | 2308 | { |
56bec294 CM |
2309 | struct btrfs_delayed_ref_root *delayed_refs; |
2310 | struct btrfs_delayed_ref_node *ref; | |
2311 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2312 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2313 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2314 | int ret; |
c3e69d58 | 2315 | int count = 0; |
56bec294 | 2316 | int must_insert_reserved = 0; |
56bec294 CM |
2317 | |
2318 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2319 | while (1) { |
2320 | if (!locked_ref) { | |
c3e69d58 CM |
2321 | /* pick a new head ref from the cluster list */ |
2322 | if (list_empty(cluster)) | |
56bec294 | 2323 | break; |
56bec294 | 2324 | |
c3e69d58 CM |
2325 | locked_ref = list_entry(cluster->next, |
2326 | struct btrfs_delayed_ref_head, cluster); | |
2327 | ||
2328 | /* grab the lock that says we are going to process | |
2329 | * all the refs for this head */ | |
2330 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2331 | ||
2332 | /* | |
2333 | * we may have dropped the spin lock to get the head | |
2334 | * mutex lock, and that might have given someone else | |
2335 | * time to free the head. If that's true, it has been | |
2336 | * removed from our list and we can move on. | |
2337 | */ | |
2338 | if (ret == -EAGAIN) { | |
2339 | locked_ref = NULL; | |
2340 | count++; | |
2341 | continue; | |
56bec294 CM |
2342 | } |
2343 | } | |
a28ec197 | 2344 | |
ae1e206b JB |
2345 | /* |
2346 | * We need to try and merge add/drops of the same ref since we | |
2347 | * can run into issues with relocate dropping the implicit ref | |
2348 | * and then it being added back again before the drop can | |
2349 | * finish. If we merged anything we need to re-loop so we can | |
2350 | * get a good ref. | |
2351 | */ | |
2352 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2353 | locked_ref); | |
2354 | ||
d1270cd9 AJ |
2355 | /* |
2356 | * locked_ref is the head node, so we have to go one | |
2357 | * node back for any delayed ref updates | |
2358 | */ | |
2359 | ref = select_delayed_ref(locked_ref); | |
2360 | ||
2361 | if (ref && ref->seq && | |
097b8a7c | 2362 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2363 | /* |
2364 | * there are still refs with lower seq numbers in the | |
2365 | * process of being added. Don't run this ref yet. | |
2366 | */ | |
2367 | list_del_init(&locked_ref->cluster); | |
093486c4 | 2368 | btrfs_delayed_ref_unlock(locked_ref); |
d1270cd9 AJ |
2369 | locked_ref = NULL; |
2370 | delayed_refs->num_heads_ready++; | |
2371 | spin_unlock(&delayed_refs->lock); | |
2372 | cond_resched(); | |
2373 | spin_lock(&delayed_refs->lock); | |
2374 | continue; | |
2375 | } | |
2376 | ||
56bec294 CM |
2377 | /* |
2378 | * record the must insert reserved flag before we | |
2379 | * drop the spin lock. | |
2380 | */ | |
2381 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2382 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2383 | |
5d4f98a2 YZ |
2384 | extent_op = locked_ref->extent_op; |
2385 | locked_ref->extent_op = NULL; | |
2386 | ||
56bec294 CM |
2387 | if (!ref) { |
2388 | /* All delayed refs have been processed, Go ahead | |
2389 | * and send the head node to run_one_delayed_ref, | |
2390 | * so that any accounting fixes can happen | |
2391 | */ | |
2392 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2393 | |
2394 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2395 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2396 | extent_op = NULL; |
2397 | } | |
2398 | ||
2399 | if (extent_op) { | |
2400 | spin_unlock(&delayed_refs->lock); | |
2401 | ||
2402 | ret = run_delayed_extent_op(trans, root, | |
2403 | ref, extent_op); | |
78a6184a | 2404 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2405 | |
79787eaa | 2406 | if (ret) { |
c2cf52eb | 2407 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
253beebd | 2408 | spin_lock(&delayed_refs->lock); |
093486c4 | 2409 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2410 | return ret; |
2411 | } | |
2412 | ||
203bf287 | 2413 | goto next; |
5d4f98a2 | 2414 | } |
56bec294 | 2415 | } |
02217ed2 | 2416 | |
56bec294 CM |
2417 | ref->in_tree = 0; |
2418 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2419 | delayed_refs->num_entries--; | |
093486c4 | 2420 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2421 | /* |
2422 | * when we play the delayed ref, also correct the | |
2423 | * ref_mod on head | |
2424 | */ | |
2425 | switch (ref->action) { | |
2426 | case BTRFS_ADD_DELAYED_REF: | |
2427 | case BTRFS_ADD_DELAYED_EXTENT: | |
2428 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2429 | break; | |
2430 | case BTRFS_DROP_DELAYED_REF: | |
2431 | locked_ref->node.ref_mod += ref->ref_mod; | |
2432 | break; | |
2433 | default: | |
2434 | WARN_ON(1); | |
2435 | } | |
2436 | } | |
56bec294 | 2437 | spin_unlock(&delayed_refs->lock); |
925baedd | 2438 | |
5d4f98a2 | 2439 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2440 | must_insert_reserved); |
eb099670 | 2441 | |
78a6184a | 2442 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2443 | if (ret) { |
093486c4 MX |
2444 | btrfs_delayed_ref_unlock(locked_ref); |
2445 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2446 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
253beebd | 2447 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2448 | return ret; |
2449 | } | |
2450 | ||
093486c4 MX |
2451 | /* |
2452 | * If this node is a head, that means all the refs in this head | |
2453 | * have been dealt with, and we will pick the next head to deal | |
2454 | * with, so we must unlock the head and drop it from the cluster | |
2455 | * list before we release it. | |
2456 | */ | |
2457 | if (btrfs_delayed_ref_is_head(ref)) { | |
2458 | list_del_init(&locked_ref->cluster); | |
2459 | btrfs_delayed_ref_unlock(locked_ref); | |
2460 | locked_ref = NULL; | |
2461 | } | |
2462 | btrfs_put_delayed_ref(ref); | |
2463 | count++; | |
203bf287 | 2464 | next: |
c3e69d58 CM |
2465 | cond_resched(); |
2466 | spin_lock(&delayed_refs->lock); | |
2467 | } | |
2468 | return count; | |
2469 | } | |
2470 | ||
709c0486 AJ |
2471 | #ifdef SCRAMBLE_DELAYED_REFS |
2472 | /* | |
2473 | * Normally delayed refs get processed in ascending bytenr order. This | |
2474 | * correlates in most cases to the order added. To expose dependencies on this | |
2475 | * order, we start to process the tree in the middle instead of the beginning | |
2476 | */ | |
2477 | static u64 find_middle(struct rb_root *root) | |
2478 | { | |
2479 | struct rb_node *n = root->rb_node; | |
2480 | struct btrfs_delayed_ref_node *entry; | |
2481 | int alt = 1; | |
2482 | u64 middle; | |
2483 | u64 first = 0, last = 0; | |
2484 | ||
2485 | n = rb_first(root); | |
2486 | if (n) { | |
2487 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2488 | first = entry->bytenr; | |
2489 | } | |
2490 | n = rb_last(root); | |
2491 | if (n) { | |
2492 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2493 | last = entry->bytenr; | |
2494 | } | |
2495 | n = root->rb_node; | |
2496 | ||
2497 | while (n) { | |
2498 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2499 | WARN_ON(!entry->in_tree); | |
2500 | ||
2501 | middle = entry->bytenr; | |
2502 | ||
2503 | if (alt) | |
2504 | n = n->rb_left; | |
2505 | else | |
2506 | n = n->rb_right; | |
2507 | ||
2508 | alt = 1 - alt; | |
2509 | } | |
2510 | return middle; | |
2511 | } | |
2512 | #endif | |
2513 | ||
bed92eae AJ |
2514 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2515 | struct btrfs_fs_info *fs_info) | |
2516 | { | |
2517 | struct qgroup_update *qgroup_update; | |
2518 | int ret = 0; | |
2519 | ||
2520 | if (list_empty(&trans->qgroup_ref_list) != | |
2521 | !trans->delayed_ref_elem.seq) { | |
2522 | /* list without seq or seq without list */ | |
c2cf52eb | 2523 | btrfs_err(fs_info, |
fc36ed7e | 2524 | "qgroup accounting update error, list is%s empty, seq is %#x.%x", |
bed92eae | 2525 | list_empty(&trans->qgroup_ref_list) ? "" : " not", |
fc36ed7e JS |
2526 | (u32)(trans->delayed_ref_elem.seq >> 32), |
2527 | (u32)trans->delayed_ref_elem.seq); | |
bed92eae AJ |
2528 | BUG(); |
2529 | } | |
2530 | ||
2531 | if (!trans->delayed_ref_elem.seq) | |
2532 | return 0; | |
2533 | ||
2534 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2535 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2536 | struct qgroup_update, list); | |
2537 | list_del(&qgroup_update->list); | |
2538 | if (!ret) | |
2539 | ret = btrfs_qgroup_account_ref( | |
2540 | trans, fs_info, qgroup_update->node, | |
2541 | qgroup_update->extent_op); | |
2542 | kfree(qgroup_update); | |
2543 | } | |
2544 | ||
2545 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2546 | ||
2547 | return ret; | |
2548 | } | |
2549 | ||
bb721703 CM |
2550 | static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq, |
2551 | int count) | |
2552 | { | |
2553 | int val = atomic_read(&delayed_refs->ref_seq); | |
2554 | ||
2555 | if (val < seq || val >= seq + count) | |
2556 | return 1; | |
2557 | return 0; | |
2558 | } | |
2559 | ||
1be41b78 JB |
2560 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2561 | { | |
2562 | u64 num_bytes; | |
2563 | ||
2564 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2565 | sizeof(struct btrfs_extent_inline_ref)); | |
2566 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2567 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2568 | ||
2569 | /* | |
2570 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
2571 | * closer to what we're really going to want to ouse. | |
2572 | */ | |
2573 | return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); | |
2574 | } | |
2575 | ||
2576 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, | |
2577 | struct btrfs_root *root) | |
2578 | { | |
2579 | struct btrfs_block_rsv *global_rsv; | |
2580 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
2581 | u64 num_bytes; | |
2582 | int ret = 0; | |
2583 | ||
2584 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2585 | num_heads = heads_to_leaves(root, num_heads); | |
2586 | if (num_heads > 1) | |
2587 | num_bytes += (num_heads - 1) * root->leafsize; | |
2588 | num_bytes <<= 1; | |
2589 | global_rsv = &root->fs_info->global_block_rsv; | |
2590 | ||
2591 | /* | |
2592 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2593 | * wiggle room since running delayed refs can create more delayed refs. | |
2594 | */ | |
2595 | if (global_rsv->space_info->full) | |
2596 | num_bytes <<= 1; | |
2597 | ||
2598 | spin_lock(&global_rsv->lock); | |
2599 | if (global_rsv->reserved <= num_bytes) | |
2600 | ret = 1; | |
2601 | spin_unlock(&global_rsv->lock); | |
2602 | return ret; | |
2603 | } | |
2604 | ||
c3e69d58 CM |
2605 | /* |
2606 | * this starts processing the delayed reference count updates and | |
2607 | * extent insertions we have queued up so far. count can be | |
2608 | * 0, which means to process everything in the tree at the start | |
2609 | * of the run (but not newly added entries), or it can be some target | |
2610 | * number you'd like to process. | |
79787eaa JM |
2611 | * |
2612 | * Returns 0 on success or if called with an aborted transaction | |
2613 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2614 | */ |
2615 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2616 | struct btrfs_root *root, unsigned long count) | |
2617 | { | |
2618 | struct rb_node *node; | |
2619 | struct btrfs_delayed_ref_root *delayed_refs; | |
2620 | struct btrfs_delayed_ref_node *ref; | |
2621 | struct list_head cluster; | |
2622 | int ret; | |
a168650c | 2623 | u64 delayed_start; |
c3e69d58 CM |
2624 | int run_all = count == (unsigned long)-1; |
2625 | int run_most = 0; | |
1fa11e26 | 2626 | int loops; |
c3e69d58 | 2627 | |
79787eaa JM |
2628 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2629 | if (trans->aborted) | |
2630 | return 0; | |
2631 | ||
c3e69d58 CM |
2632 | if (root == root->fs_info->extent_root) |
2633 | root = root->fs_info->tree_root; | |
2634 | ||
edf39272 JS |
2635 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2636 | ||
c3e69d58 CM |
2637 | delayed_refs = &trans->transaction->delayed_refs; |
2638 | INIT_LIST_HEAD(&cluster); | |
bb721703 CM |
2639 | if (count == 0) { |
2640 | count = delayed_refs->num_entries * 2; | |
2641 | run_most = 1; | |
2642 | } | |
2643 | ||
2644 | if (!run_all && !run_most) { | |
2645 | int old; | |
2646 | int seq = atomic_read(&delayed_refs->ref_seq); | |
2647 | ||
2648 | progress: | |
2649 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2650 | if (old) { | |
2651 | DEFINE_WAIT(__wait); | |
1be41b78 JB |
2652 | if (delayed_refs->flushing || |
2653 | !btrfs_should_throttle_delayed_refs(trans, root)) | |
bb721703 CM |
2654 | return 0; |
2655 | ||
2656 | prepare_to_wait(&delayed_refs->wait, &__wait, | |
2657 | TASK_UNINTERRUPTIBLE); | |
2658 | ||
2659 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2660 | if (old) { | |
2661 | schedule(); | |
2662 | finish_wait(&delayed_refs->wait, &__wait); | |
2663 | ||
2664 | if (!refs_newer(delayed_refs, seq, 256)) | |
2665 | goto progress; | |
2666 | else | |
2667 | return 0; | |
2668 | } else { | |
2669 | finish_wait(&delayed_refs->wait, &__wait); | |
2670 | goto again; | |
2671 | } | |
2672 | } | |
2673 | ||
2674 | } else { | |
2675 | atomic_inc(&delayed_refs->procs_running_refs); | |
2676 | } | |
2677 | ||
c3e69d58 | 2678 | again: |
1fa11e26 | 2679 | loops = 0; |
c3e69d58 | 2680 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2681 | |
709c0486 AJ |
2682 | #ifdef SCRAMBLE_DELAYED_REFS |
2683 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2684 | #endif | |
2685 | ||
c3e69d58 CM |
2686 | while (1) { |
2687 | if (!(run_all || run_most) && | |
1be41b78 | 2688 | !btrfs_should_throttle_delayed_refs(trans, root)) |
c3e69d58 | 2689 | break; |
eb099670 | 2690 | |
56bec294 | 2691 | /* |
c3e69d58 CM |
2692 | * go find something we can process in the rbtree. We start at |
2693 | * the beginning of the tree, and then build a cluster | |
2694 | * of refs to process starting at the first one we are able to | |
2695 | * lock | |
56bec294 | 2696 | */ |
a168650c | 2697 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2698 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2699 | delayed_refs->run_delayed_start); | |
2700 | if (ret) | |
56bec294 CM |
2701 | break; |
2702 | ||
c3e69d58 | 2703 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa | 2704 | if (ret < 0) { |
093486c4 | 2705 | btrfs_release_ref_cluster(&cluster); |
79787eaa JM |
2706 | spin_unlock(&delayed_refs->lock); |
2707 | btrfs_abort_transaction(trans, root, ret); | |
bb721703 | 2708 | atomic_dec(&delayed_refs->procs_running_refs); |
f971fe29 | 2709 | wake_up(&delayed_refs->wait); |
79787eaa JM |
2710 | return ret; |
2711 | } | |
c3e69d58 | 2712 | |
bb721703 CM |
2713 | atomic_add(ret, &delayed_refs->ref_seq); |
2714 | ||
c3e69d58 CM |
2715 | count -= min_t(unsigned long, ret, count); |
2716 | ||
2717 | if (count == 0) | |
2718 | break; | |
a168650c | 2719 | |
1fa11e26 AJ |
2720 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2721 | if (loops == 0) { | |
2722 | /* | |
2723 | * btrfs_find_ref_cluster looped. let's do one | |
2724 | * more cycle. if we don't run any delayed ref | |
2725 | * during that cycle (because we can't because | |
2726 | * all of them are blocked), bail out. | |
2727 | */ | |
2728 | loops = 1; | |
2729 | } else { | |
2730 | /* | |
2731 | * no runnable refs left, stop trying | |
2732 | */ | |
2733 | BUG_ON(run_all); | |
2734 | break; | |
2735 | } | |
2736 | } | |
2737 | if (ret) { | |
a168650c | 2738 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2739 | loops = 0; |
a168650c | 2740 | } |
eb099670 | 2741 | } |
c3e69d58 | 2742 | |
56bec294 | 2743 | if (run_all) { |
ea658bad JB |
2744 | if (!list_empty(&trans->new_bgs)) { |
2745 | spin_unlock(&delayed_refs->lock); | |
2746 | btrfs_create_pending_block_groups(trans, root); | |
2747 | spin_lock(&delayed_refs->lock); | |
2748 | } | |
2749 | ||
56bec294 | 2750 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2751 | if (!node) |
56bec294 | 2752 | goto out; |
c3e69d58 | 2753 | count = (unsigned long)-1; |
e9d0b13b | 2754 | |
56bec294 CM |
2755 | while (node) { |
2756 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2757 | rb_node); | |
2758 | if (btrfs_delayed_ref_is_head(ref)) { | |
2759 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2760 | |
56bec294 CM |
2761 | head = btrfs_delayed_node_to_head(ref); |
2762 | atomic_inc(&ref->refs); | |
2763 | ||
2764 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2765 | /* |
2766 | * Mutex was contended, block until it's | |
2767 | * released and try again | |
2768 | */ | |
56bec294 CM |
2769 | mutex_lock(&head->mutex); |
2770 | mutex_unlock(&head->mutex); | |
2771 | ||
2772 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2773 | cond_resched(); |
56bec294 CM |
2774 | goto again; |
2775 | } | |
2776 | node = rb_next(node); | |
2777 | } | |
2778 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2779 | schedule_timeout(1); |
2780 | goto again; | |
5f39d397 | 2781 | } |
54aa1f4d | 2782 | out: |
bb721703 CM |
2783 | atomic_dec(&delayed_refs->procs_running_refs); |
2784 | smp_mb(); | |
2785 | if (waitqueue_active(&delayed_refs->wait)) | |
2786 | wake_up(&delayed_refs->wait); | |
2787 | ||
c3e69d58 | 2788 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2789 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2790 | return 0; |
2791 | } | |
2792 | ||
5d4f98a2 YZ |
2793 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2794 | struct btrfs_root *root, | |
2795 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 2796 | int level, int is_data) |
5d4f98a2 YZ |
2797 | { |
2798 | struct btrfs_delayed_extent_op *extent_op; | |
2799 | int ret; | |
2800 | ||
78a6184a | 2801 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2802 | if (!extent_op) |
2803 | return -ENOMEM; | |
2804 | ||
2805 | extent_op->flags_to_set = flags; | |
2806 | extent_op->update_flags = 1; | |
2807 | extent_op->update_key = 0; | |
2808 | extent_op->is_data = is_data ? 1 : 0; | |
b1c79e09 | 2809 | extent_op->level = level; |
5d4f98a2 | 2810 | |
66d7e7f0 AJ |
2811 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2812 | num_bytes, extent_op); | |
5d4f98a2 | 2813 | if (ret) |
78a6184a | 2814 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2815 | return ret; |
2816 | } | |
2817 | ||
2818 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2819 | struct btrfs_root *root, | |
2820 | struct btrfs_path *path, | |
2821 | u64 objectid, u64 offset, u64 bytenr) | |
2822 | { | |
2823 | struct btrfs_delayed_ref_head *head; | |
2824 | struct btrfs_delayed_ref_node *ref; | |
2825 | struct btrfs_delayed_data_ref *data_ref; | |
2826 | struct btrfs_delayed_ref_root *delayed_refs; | |
2827 | struct rb_node *node; | |
2828 | int ret = 0; | |
2829 | ||
2830 | ret = -ENOENT; | |
2831 | delayed_refs = &trans->transaction->delayed_refs; | |
2832 | spin_lock(&delayed_refs->lock); | |
2833 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2834 | if (!head) | |
2835 | goto out; | |
2836 | ||
2837 | if (!mutex_trylock(&head->mutex)) { | |
2838 | atomic_inc(&head->node.refs); | |
2839 | spin_unlock(&delayed_refs->lock); | |
2840 | ||
b3b4aa74 | 2841 | btrfs_release_path(path); |
5d4f98a2 | 2842 | |
8cc33e5c DS |
2843 | /* |
2844 | * Mutex was contended, block until it's released and let | |
2845 | * caller try again | |
2846 | */ | |
5d4f98a2 YZ |
2847 | mutex_lock(&head->mutex); |
2848 | mutex_unlock(&head->mutex); | |
2849 | btrfs_put_delayed_ref(&head->node); | |
2850 | return -EAGAIN; | |
2851 | } | |
2852 | ||
2853 | node = rb_prev(&head->node.rb_node); | |
2854 | if (!node) | |
2855 | goto out_unlock; | |
2856 | ||
2857 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2858 | ||
2859 | if (ref->bytenr != bytenr) | |
2860 | goto out_unlock; | |
2861 | ||
2862 | ret = 1; | |
2863 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2864 | goto out_unlock; | |
2865 | ||
2866 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2867 | ||
2868 | node = rb_prev(node); | |
2869 | if (node) { | |
df57dbe6 LB |
2870 | int seq = ref->seq; |
2871 | ||
5d4f98a2 | 2872 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2873 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2874 | goto out_unlock; |
2875 | } | |
2876 | ||
2877 | if (data_ref->root != root->root_key.objectid || | |
2878 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2879 | goto out_unlock; | |
2880 | ||
2881 | ret = 0; | |
2882 | out_unlock: | |
2883 | mutex_unlock(&head->mutex); | |
2884 | out: | |
2885 | spin_unlock(&delayed_refs->lock); | |
2886 | return ret; | |
2887 | } | |
2888 | ||
2889 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2890 | struct btrfs_root *root, | |
2891 | struct btrfs_path *path, | |
2892 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2893 | { |
2894 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2895 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2896 | struct btrfs_extent_data_ref *ref; |
2897 | struct btrfs_extent_inline_ref *iref; | |
2898 | struct btrfs_extent_item *ei; | |
f321e491 | 2899 | struct btrfs_key key; |
5d4f98a2 | 2900 | u32 item_size; |
be20aa9d | 2901 | int ret; |
925baedd | 2902 | |
be20aa9d | 2903 | key.objectid = bytenr; |
31840ae1 | 2904 | key.offset = (u64)-1; |
f321e491 | 2905 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2906 | |
be20aa9d CM |
2907 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2908 | if (ret < 0) | |
2909 | goto out; | |
79787eaa | 2910 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2911 | |
2912 | ret = -ENOENT; | |
2913 | if (path->slots[0] == 0) | |
31840ae1 | 2914 | goto out; |
be20aa9d | 2915 | |
31840ae1 | 2916 | path->slots[0]--; |
f321e491 | 2917 | leaf = path->nodes[0]; |
5d4f98a2 | 2918 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2919 | |
5d4f98a2 | 2920 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2921 | goto out; |
f321e491 | 2922 | |
5d4f98a2 YZ |
2923 | ret = 1; |
2924 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2925 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2926 | if (item_size < sizeof(*ei)) { | |
2927 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2928 | goto out; | |
2929 | } | |
2930 | #endif | |
2931 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2932 | |
5d4f98a2 YZ |
2933 | if (item_size != sizeof(*ei) + |
2934 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2935 | goto out; | |
be20aa9d | 2936 | |
5d4f98a2 YZ |
2937 | if (btrfs_extent_generation(leaf, ei) <= |
2938 | btrfs_root_last_snapshot(&root->root_item)) | |
2939 | goto out; | |
2940 | ||
2941 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2942 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2943 | BTRFS_EXTENT_DATA_REF_KEY) | |
2944 | goto out; | |
2945 | ||
2946 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2947 | if (btrfs_extent_refs(leaf, ei) != | |
2948 | btrfs_extent_data_ref_count(leaf, ref) || | |
2949 | btrfs_extent_data_ref_root(leaf, ref) != | |
2950 | root->root_key.objectid || | |
2951 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2952 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2953 | goto out; | |
2954 | ||
2955 | ret = 0; | |
2956 | out: | |
2957 | return ret; | |
2958 | } | |
2959 | ||
2960 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2961 | struct btrfs_root *root, | |
2962 | u64 objectid, u64 offset, u64 bytenr) | |
2963 | { | |
2964 | struct btrfs_path *path; | |
2965 | int ret; | |
2966 | int ret2; | |
2967 | ||
2968 | path = btrfs_alloc_path(); | |
2969 | if (!path) | |
2970 | return -ENOENT; | |
2971 | ||
2972 | do { | |
2973 | ret = check_committed_ref(trans, root, path, objectid, | |
2974 | offset, bytenr); | |
2975 | if (ret && ret != -ENOENT) | |
f321e491 | 2976 | goto out; |
80ff3856 | 2977 | |
5d4f98a2 YZ |
2978 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2979 | offset, bytenr); | |
2980 | } while (ret2 == -EAGAIN); | |
2981 | ||
2982 | if (ret2 && ret2 != -ENOENT) { | |
2983 | ret = ret2; | |
2984 | goto out; | |
f321e491 | 2985 | } |
5d4f98a2 YZ |
2986 | |
2987 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2988 | ret = 0; | |
be20aa9d | 2989 | out: |
80ff3856 | 2990 | btrfs_free_path(path); |
f0486c68 YZ |
2991 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2992 | WARN_ON(ret > 0); | |
f321e491 | 2993 | return ret; |
be20aa9d | 2994 | } |
c5739bba | 2995 | |
5d4f98a2 | 2996 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2997 | struct btrfs_root *root, |
5d4f98a2 | 2998 | struct extent_buffer *buf, |
66d7e7f0 | 2999 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
3000 | { |
3001 | u64 bytenr; | |
5d4f98a2 YZ |
3002 | u64 num_bytes; |
3003 | u64 parent; | |
31840ae1 | 3004 | u64 ref_root; |
31840ae1 | 3005 | u32 nritems; |
31840ae1 ZY |
3006 | struct btrfs_key key; |
3007 | struct btrfs_file_extent_item *fi; | |
3008 | int i; | |
3009 | int level; | |
3010 | int ret = 0; | |
31840ae1 | 3011 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 3012 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
3013 | |
3014 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
3015 | nritems = btrfs_header_nritems(buf); |
3016 | level = btrfs_header_level(buf); | |
3017 | ||
5d4f98a2 YZ |
3018 | if (!root->ref_cows && level == 0) |
3019 | return 0; | |
31840ae1 | 3020 | |
5d4f98a2 YZ |
3021 | if (inc) |
3022 | process_func = btrfs_inc_extent_ref; | |
3023 | else | |
3024 | process_func = btrfs_free_extent; | |
31840ae1 | 3025 | |
5d4f98a2 YZ |
3026 | if (full_backref) |
3027 | parent = buf->start; | |
3028 | else | |
3029 | parent = 0; | |
3030 | ||
3031 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3032 | if (level == 0) { |
5d4f98a2 | 3033 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
3034 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
3035 | continue; | |
5d4f98a2 | 3036 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3037 | struct btrfs_file_extent_item); |
3038 | if (btrfs_file_extent_type(buf, fi) == | |
3039 | BTRFS_FILE_EXTENT_INLINE) | |
3040 | continue; | |
3041 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3042 | if (bytenr == 0) | |
3043 | continue; | |
5d4f98a2 YZ |
3044 | |
3045 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3046 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3047 | ret = process_func(trans, root, bytenr, num_bytes, | |
3048 | parent, ref_root, key.objectid, | |
66d7e7f0 | 3049 | key.offset, for_cow); |
31840ae1 ZY |
3050 | if (ret) |
3051 | goto fail; | |
3052 | } else { | |
5d4f98a2 YZ |
3053 | bytenr = btrfs_node_blockptr(buf, i); |
3054 | num_bytes = btrfs_level_size(root, level - 1); | |
3055 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
3056 | parent, ref_root, level - 1, 0, |
3057 | for_cow); | |
31840ae1 ZY |
3058 | if (ret) |
3059 | goto fail; | |
3060 | } | |
3061 | } | |
3062 | return 0; | |
3063 | fail: | |
5d4f98a2 YZ |
3064 | return ret; |
3065 | } | |
3066 | ||
3067 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3068 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3069 | { |
66d7e7f0 | 3070 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
3071 | } |
3072 | ||
3073 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3074 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3075 | { |
66d7e7f0 | 3076 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
3077 | } |
3078 | ||
9078a3e1 CM |
3079 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3080 | struct btrfs_root *root, | |
3081 | struct btrfs_path *path, | |
3082 | struct btrfs_block_group_cache *cache) | |
3083 | { | |
3084 | int ret; | |
9078a3e1 | 3085 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3086 | unsigned long bi; |
3087 | struct extent_buffer *leaf; | |
9078a3e1 | 3088 | |
9078a3e1 | 3089 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
3090 | if (ret < 0) |
3091 | goto fail; | |
79787eaa | 3092 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
3093 | |
3094 | leaf = path->nodes[0]; | |
3095 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3096 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3097 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 3098 | btrfs_release_path(path); |
54aa1f4d | 3099 | fail: |
79787eaa JM |
3100 | if (ret) { |
3101 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 3102 | return ret; |
79787eaa | 3103 | } |
9078a3e1 CM |
3104 | return 0; |
3105 | ||
3106 | } | |
3107 | ||
4a8c9a62 YZ |
3108 | static struct btrfs_block_group_cache * |
3109 | next_block_group(struct btrfs_root *root, | |
3110 | struct btrfs_block_group_cache *cache) | |
3111 | { | |
3112 | struct rb_node *node; | |
3113 | spin_lock(&root->fs_info->block_group_cache_lock); | |
3114 | node = rb_next(&cache->cache_node); | |
3115 | btrfs_put_block_group(cache); | |
3116 | if (node) { | |
3117 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3118 | cache_node); | |
11dfe35a | 3119 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3120 | } else |
3121 | cache = NULL; | |
3122 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3123 | return cache; | |
3124 | } | |
3125 | ||
0af3d00b JB |
3126 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3127 | struct btrfs_trans_handle *trans, | |
3128 | struct btrfs_path *path) | |
3129 | { | |
3130 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3131 | struct inode *inode = NULL; | |
3132 | u64 alloc_hint = 0; | |
2b20982e | 3133 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
3134 | int num_pages = 0; |
3135 | int retries = 0; | |
3136 | int ret = 0; | |
3137 | ||
3138 | /* | |
3139 | * If this block group is smaller than 100 megs don't bother caching the | |
3140 | * block group. | |
3141 | */ | |
3142 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3143 | spin_lock(&block_group->lock); | |
3144 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3145 | spin_unlock(&block_group->lock); | |
3146 | return 0; | |
3147 | } | |
3148 | ||
3149 | again: | |
3150 | inode = lookup_free_space_inode(root, block_group, path); | |
3151 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3152 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3153 | btrfs_release_path(path); |
0af3d00b JB |
3154 | goto out; |
3155 | } | |
3156 | ||
3157 | if (IS_ERR(inode)) { | |
3158 | BUG_ON(retries); | |
3159 | retries++; | |
3160 | ||
3161 | if (block_group->ro) | |
3162 | goto out_free; | |
3163 | ||
3164 | ret = create_free_space_inode(root, trans, block_group, path); | |
3165 | if (ret) | |
3166 | goto out_free; | |
3167 | goto again; | |
3168 | } | |
3169 | ||
5b0e95bf JB |
3170 | /* We've already setup this transaction, go ahead and exit */ |
3171 | if (block_group->cache_generation == trans->transid && | |
3172 | i_size_read(inode)) { | |
3173 | dcs = BTRFS_DC_SETUP; | |
3174 | goto out_put; | |
3175 | } | |
3176 | ||
0af3d00b JB |
3177 | /* |
3178 | * We want to set the generation to 0, that way if anything goes wrong | |
3179 | * from here on out we know not to trust this cache when we load up next | |
3180 | * time. | |
3181 | */ | |
3182 | BTRFS_I(inode)->generation = 0; | |
3183 | ret = btrfs_update_inode(trans, root, inode); | |
3184 | WARN_ON(ret); | |
3185 | ||
3186 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3187 | ret = btrfs_check_trunc_cache_free_space(root, |
3188 | &root->fs_info->global_block_rsv); | |
3189 | if (ret) | |
3190 | goto out_put; | |
3191 | ||
0af3d00b JB |
3192 | ret = btrfs_truncate_free_space_cache(root, trans, path, |
3193 | inode); | |
3194 | if (ret) | |
3195 | goto out_put; | |
3196 | } | |
3197 | ||
3198 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
3199 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
3200 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
3201 | /* | |
3202 | * don't bother trying to write stuff out _if_ | |
3203 | * a) we're not cached, | |
3204 | * b) we're with nospace_cache mount option. | |
3205 | */ | |
2b20982e | 3206 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3207 | spin_unlock(&block_group->lock); |
3208 | goto out_put; | |
3209 | } | |
3210 | spin_unlock(&block_group->lock); | |
3211 | ||
6fc823b1 JB |
3212 | /* |
3213 | * Try to preallocate enough space based on how big the block group is. | |
3214 | * Keep in mind this has to include any pinned space which could end up | |
3215 | * taking up quite a bit since it's not folded into the other space | |
3216 | * cache. | |
3217 | */ | |
3218 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3219 | if (!num_pages) |
3220 | num_pages = 1; | |
3221 | ||
0af3d00b JB |
3222 | num_pages *= 16; |
3223 | num_pages *= PAGE_CACHE_SIZE; | |
3224 | ||
3225 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3226 | if (ret) | |
3227 | goto out_put; | |
3228 | ||
3229 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3230 | num_pages, num_pages, | |
3231 | &alloc_hint); | |
2b20982e JB |
3232 | if (!ret) |
3233 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3234 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3235 | |
0af3d00b JB |
3236 | out_put: |
3237 | iput(inode); | |
3238 | out_free: | |
b3b4aa74 | 3239 | btrfs_release_path(path); |
0af3d00b JB |
3240 | out: |
3241 | spin_lock(&block_group->lock); | |
e65cbb94 | 3242 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3243 | block_group->cache_generation = trans->transid; |
2b20982e | 3244 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3245 | spin_unlock(&block_group->lock); |
3246 | ||
3247 | return ret; | |
3248 | } | |
3249 | ||
96b5179d CM |
3250 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3251 | struct btrfs_root *root) | |
9078a3e1 | 3252 | { |
4a8c9a62 | 3253 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3254 | int err = 0; |
9078a3e1 | 3255 | struct btrfs_path *path; |
96b5179d | 3256 | u64 last = 0; |
9078a3e1 CM |
3257 | |
3258 | path = btrfs_alloc_path(); | |
3259 | if (!path) | |
3260 | return -ENOMEM; | |
3261 | ||
0af3d00b JB |
3262 | again: |
3263 | while (1) { | |
3264 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3265 | while (cache) { | |
3266 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3267 | break; | |
3268 | cache = next_block_group(root, cache); | |
3269 | } | |
3270 | if (!cache) { | |
3271 | if (last == 0) | |
3272 | break; | |
3273 | last = 0; | |
3274 | continue; | |
3275 | } | |
3276 | err = cache_save_setup(cache, trans, path); | |
3277 | last = cache->key.objectid + cache->key.offset; | |
3278 | btrfs_put_block_group(cache); | |
3279 | } | |
3280 | ||
d397712b | 3281 | while (1) { |
4a8c9a62 YZ |
3282 | if (last == 0) { |
3283 | err = btrfs_run_delayed_refs(trans, root, | |
3284 | (unsigned long)-1); | |
79787eaa JM |
3285 | if (err) /* File system offline */ |
3286 | goto out; | |
0f9dd46c | 3287 | } |
54aa1f4d | 3288 | |
4a8c9a62 YZ |
3289 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3290 | while (cache) { | |
0af3d00b JB |
3291 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3292 | btrfs_put_block_group(cache); | |
3293 | goto again; | |
3294 | } | |
3295 | ||
4a8c9a62 YZ |
3296 | if (cache->dirty) |
3297 | break; | |
3298 | cache = next_block_group(root, cache); | |
3299 | } | |
3300 | if (!cache) { | |
3301 | if (last == 0) | |
3302 | break; | |
3303 | last = 0; | |
3304 | continue; | |
3305 | } | |
0f9dd46c | 3306 | |
0cb59c99 JB |
3307 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3308 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3309 | cache->dirty = 0; |
4a8c9a62 | 3310 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3311 | |
4a8c9a62 | 3312 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3313 | if (err) /* File system offline */ |
3314 | goto out; | |
3315 | ||
4a8c9a62 | 3316 | btrfs_put_block_group(cache); |
9078a3e1 | 3317 | } |
4a8c9a62 | 3318 | |
0cb59c99 JB |
3319 | while (1) { |
3320 | /* | |
3321 | * I don't think this is needed since we're just marking our | |
3322 | * preallocated extent as written, but just in case it can't | |
3323 | * hurt. | |
3324 | */ | |
3325 | if (last == 0) { | |
3326 | err = btrfs_run_delayed_refs(trans, root, | |
3327 | (unsigned long)-1); | |
79787eaa JM |
3328 | if (err) /* File system offline */ |
3329 | goto out; | |
0cb59c99 JB |
3330 | } |
3331 | ||
3332 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3333 | while (cache) { | |
3334 | /* | |
3335 | * Really this shouldn't happen, but it could if we | |
3336 | * couldn't write the entire preallocated extent and | |
3337 | * splitting the extent resulted in a new block. | |
3338 | */ | |
3339 | if (cache->dirty) { | |
3340 | btrfs_put_block_group(cache); | |
3341 | goto again; | |
3342 | } | |
3343 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3344 | break; | |
3345 | cache = next_block_group(root, cache); | |
3346 | } | |
3347 | if (!cache) { | |
3348 | if (last == 0) | |
3349 | break; | |
3350 | last = 0; | |
3351 | continue; | |
3352 | } | |
3353 | ||
79787eaa | 3354 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3355 | |
3356 | /* | |
3357 | * If we didn't have an error then the cache state is still | |
3358 | * NEED_WRITE, so we can set it to WRITTEN. | |
3359 | */ | |
79787eaa | 3360 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3361 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3362 | last = cache->key.objectid + cache->key.offset; | |
3363 | btrfs_put_block_group(cache); | |
3364 | } | |
79787eaa | 3365 | out: |
0cb59c99 | 3366 | |
9078a3e1 | 3367 | btrfs_free_path(path); |
79787eaa | 3368 | return err; |
9078a3e1 CM |
3369 | } |
3370 | ||
d2fb3437 YZ |
3371 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3372 | { | |
3373 | struct btrfs_block_group_cache *block_group; | |
3374 | int readonly = 0; | |
3375 | ||
3376 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3377 | if (!block_group || block_group->ro) | |
3378 | readonly = 1; | |
3379 | if (block_group) | |
fa9c0d79 | 3380 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3381 | return readonly; |
3382 | } | |
3383 | ||
593060d7 CM |
3384 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3385 | u64 total_bytes, u64 bytes_used, | |
3386 | struct btrfs_space_info **space_info) | |
3387 | { | |
3388 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3389 | int i; |
3390 | int factor; | |
b150a4f1 | 3391 | int ret; |
b742bb82 YZ |
3392 | |
3393 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3394 | BTRFS_BLOCK_GROUP_RAID10)) | |
3395 | factor = 2; | |
3396 | else | |
3397 | factor = 1; | |
593060d7 CM |
3398 | |
3399 | found = __find_space_info(info, flags); | |
3400 | if (found) { | |
25179201 | 3401 | spin_lock(&found->lock); |
593060d7 | 3402 | found->total_bytes += total_bytes; |
89a55897 | 3403 | found->disk_total += total_bytes * factor; |
593060d7 | 3404 | found->bytes_used += bytes_used; |
b742bb82 | 3405 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3406 | found->full = 0; |
25179201 | 3407 | spin_unlock(&found->lock); |
593060d7 CM |
3408 | *space_info = found; |
3409 | return 0; | |
3410 | } | |
c146afad | 3411 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3412 | if (!found) |
3413 | return -ENOMEM; | |
3414 | ||
b150a4f1 JB |
3415 | ret = percpu_counter_init(&found->total_bytes_pinned, 0); |
3416 | if (ret) { | |
3417 | kfree(found); | |
3418 | return ret; | |
3419 | } | |
3420 | ||
b742bb82 YZ |
3421 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3422 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3423 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3424 | spin_lock_init(&found->lock); |
52ba6929 | 3425 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3426 | found->total_bytes = total_bytes; |
89a55897 | 3427 | found->disk_total = total_bytes * factor; |
593060d7 | 3428 | found->bytes_used = bytes_used; |
b742bb82 | 3429 | found->disk_used = bytes_used * factor; |
593060d7 | 3430 | found->bytes_pinned = 0; |
e8569813 | 3431 | found->bytes_reserved = 0; |
c146afad | 3432 | found->bytes_readonly = 0; |
f0486c68 | 3433 | found->bytes_may_use = 0; |
593060d7 | 3434 | found->full = 0; |
0e4f8f88 | 3435 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3436 | found->chunk_alloc = 0; |
fdb5effd JB |
3437 | found->flush = 0; |
3438 | init_waitqueue_head(&found->wait); | |
593060d7 | 3439 | *space_info = found; |
4184ea7f | 3440 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3441 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3442 | info->data_sinfo = found; | |
593060d7 CM |
3443 | return 0; |
3444 | } | |
3445 | ||
8790d502 CM |
3446 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3447 | { | |
899c81ea ID |
3448 | u64 extra_flags = chunk_to_extended(flags) & |
3449 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3450 | |
de98ced9 | 3451 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3452 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3453 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3454 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3455 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3456 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3457 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3458 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3459 | } |
593060d7 | 3460 | |
fc67c450 ID |
3461 | /* |
3462 | * returns target flags in extended format or 0 if restripe for this | |
3463 | * chunk_type is not in progress | |
c6664b42 ID |
3464 | * |
3465 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3466 | */ |
3467 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3468 | { | |
3469 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3470 | u64 target = 0; | |
3471 | ||
fc67c450 ID |
3472 | if (!bctl) |
3473 | return 0; | |
3474 | ||
3475 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3476 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3477 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3478 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3479 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3480 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3481 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3482 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3483 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3484 | } | |
3485 | ||
3486 | return target; | |
3487 | } | |
3488 | ||
a46d11a8 ID |
3489 | /* |
3490 | * @flags: available profiles in extended format (see ctree.h) | |
3491 | * | |
e4d8ec0f ID |
3492 | * Returns reduced profile in chunk format. If profile changing is in |
3493 | * progress (either running or paused) picks the target profile (if it's | |
3494 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3495 | */ |
48a3b636 | 3496 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3497 | { |
cd02dca5 CM |
3498 | /* |
3499 | * we add in the count of missing devices because we want | |
3500 | * to make sure that any RAID levels on a degraded FS | |
3501 | * continue to be honored. | |
3502 | */ | |
3503 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3504 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3505 | u64 target; |
53b381b3 | 3506 | u64 tmp; |
a061fc8d | 3507 | |
fc67c450 ID |
3508 | /* |
3509 | * see if restripe for this chunk_type is in progress, if so | |
3510 | * try to reduce to the target profile | |
3511 | */ | |
e4d8ec0f | 3512 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3513 | target = get_restripe_target(root->fs_info, flags); |
3514 | if (target) { | |
3515 | /* pick target profile only if it's already available */ | |
3516 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3517 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3518 | return extended_to_chunk(target); |
e4d8ec0f ID |
3519 | } |
3520 | } | |
3521 | spin_unlock(&root->fs_info->balance_lock); | |
3522 | ||
53b381b3 | 3523 | /* First, mask out the RAID levels which aren't possible */ |
a061fc8d | 3524 | if (num_devices == 1) |
53b381b3 DW |
3525 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | |
3526 | BTRFS_BLOCK_GROUP_RAID5); | |
3527 | if (num_devices < 3) | |
3528 | flags &= ~BTRFS_BLOCK_GROUP_RAID6; | |
a061fc8d CM |
3529 | if (num_devices < 4) |
3530 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3531 | ||
53b381b3 DW |
3532 | tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | |
3533 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | | |
3534 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); | |
3535 | flags &= ~tmp; | |
ec44a35c | 3536 | |
53b381b3 DW |
3537 | if (tmp & BTRFS_BLOCK_GROUP_RAID6) |
3538 | tmp = BTRFS_BLOCK_GROUP_RAID6; | |
3539 | else if (tmp & BTRFS_BLOCK_GROUP_RAID5) | |
3540 | tmp = BTRFS_BLOCK_GROUP_RAID5; | |
3541 | else if (tmp & BTRFS_BLOCK_GROUP_RAID10) | |
3542 | tmp = BTRFS_BLOCK_GROUP_RAID10; | |
3543 | else if (tmp & BTRFS_BLOCK_GROUP_RAID1) | |
3544 | tmp = BTRFS_BLOCK_GROUP_RAID1; | |
3545 | else if (tmp & BTRFS_BLOCK_GROUP_RAID0) | |
3546 | tmp = BTRFS_BLOCK_GROUP_RAID0; | |
a46d11a8 | 3547 | |
53b381b3 | 3548 | return extended_to_chunk(flags | tmp); |
ec44a35c CM |
3549 | } |
3550 | ||
b742bb82 | 3551 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3552 | { |
de98ced9 MX |
3553 | unsigned seq; |
3554 | ||
3555 | do { | |
3556 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
3557 | ||
3558 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3559 | flags |= root->fs_info->avail_data_alloc_bits; | |
3560 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3561 | flags |= root->fs_info->avail_system_alloc_bits; | |
3562 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3563 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3564 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3565 | |
b742bb82 | 3566 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3567 | } |
3568 | ||
6d07bcec | 3569 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3570 | { |
b742bb82 | 3571 | u64 flags; |
53b381b3 | 3572 | u64 ret; |
9ed74f2d | 3573 | |
b742bb82 YZ |
3574 | if (data) |
3575 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3576 | else if (root == root->fs_info->chunk_root) | |
3577 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3578 | else |
b742bb82 | 3579 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3580 | |
53b381b3 DW |
3581 | ret = get_alloc_profile(root, flags); |
3582 | return ret; | |
6a63209f | 3583 | } |
9ed74f2d | 3584 | |
6a63209f | 3585 | /* |
6a63209f JB |
3586 | * This will check the space that the inode allocates from to make sure we have |
3587 | * enough space for bytes. | |
6a63209f | 3588 | */ |
0ca1f7ce | 3589 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3590 | { |
6a63209f | 3591 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3592 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3593 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3594 | u64 used; |
0af3d00b | 3595 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3596 | |
6a63209f | 3597 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3598 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 3599 | |
82d5902d LZ |
3600 | if (root == root->fs_info->tree_root || |
3601 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3602 | alloc_chunk = 0; |
3603 | committed = 1; | |
3604 | } | |
3605 | ||
b4d7c3c9 | 3606 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3607 | if (!data_sinfo) |
3608 | goto alloc; | |
9ed74f2d | 3609 | |
6a63209f JB |
3610 | again: |
3611 | /* make sure we have enough space to handle the data first */ | |
3612 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3613 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3614 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3615 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3616 | |
3617 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3618 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3619 | |
6a63209f JB |
3620 | /* |
3621 | * if we don't have enough free bytes in this space then we need | |
3622 | * to alloc a new chunk. | |
3623 | */ | |
0af3d00b | 3624 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3625 | u64 alloc_target; |
9ed74f2d | 3626 | |
0e4f8f88 | 3627 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3628 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3629 | alloc: |
6a63209f | 3630 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3631 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3632 | if (IS_ERR(trans)) |
3633 | return PTR_ERR(trans); | |
9ed74f2d | 3634 | |
6a63209f | 3635 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3636 | alloc_target, |
3637 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3638 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3639 | if (ret < 0) { |
3640 | if (ret != -ENOSPC) | |
3641 | return ret; | |
3642 | else | |
3643 | goto commit_trans; | |
3644 | } | |
9ed74f2d | 3645 | |
b4d7c3c9 LZ |
3646 | if (!data_sinfo) |
3647 | data_sinfo = fs_info->data_sinfo; | |
3648 | ||
6a63209f JB |
3649 | goto again; |
3650 | } | |
f2bb8f5c JB |
3651 | |
3652 | /* | |
b150a4f1 JB |
3653 | * If we don't have enough pinned space to deal with this |
3654 | * allocation don't bother committing the transaction. | |
f2bb8f5c | 3655 | */ |
b150a4f1 JB |
3656 | if (percpu_counter_compare(&data_sinfo->total_bytes_pinned, |
3657 | bytes) < 0) | |
f2bb8f5c | 3658 | committed = 1; |
6a63209f | 3659 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3660 | |
4e06bdd6 | 3661 | /* commit the current transaction and try again */ |
d52a5b5f | 3662 | commit_trans: |
a4abeea4 JB |
3663 | if (!committed && |
3664 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3665 | committed = 1; |
b150a4f1 | 3666 | |
7a7eaa40 | 3667 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3668 | if (IS_ERR(trans)) |
3669 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3670 | ret = btrfs_commit_transaction(trans, root); |
3671 | if (ret) | |
3672 | return ret; | |
3673 | goto again; | |
3674 | } | |
9ed74f2d | 3675 | |
6a63209f JB |
3676 | return -ENOSPC; |
3677 | } | |
3678 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3679 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3680 | data_sinfo->flags, bytes, 1); |
6a63209f | 3681 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3682 | |
9ed74f2d | 3683 | return 0; |
9ed74f2d | 3684 | } |
6a63209f | 3685 | |
6a63209f | 3686 | /* |
fb25e914 | 3687 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3688 | */ |
0ca1f7ce | 3689 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3690 | { |
0ca1f7ce | 3691 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3692 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3693 | |
6a63209f | 3694 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3695 | bytes = ALIGN(bytes, root->sectorsize); |
e3ccfa98 | 3696 | |
b4d7c3c9 | 3697 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f | 3698 | spin_lock(&data_sinfo->lock); |
7ee9e440 | 3699 | WARN_ON(data_sinfo->bytes_may_use < bytes); |
6a63209f | 3700 | data_sinfo->bytes_may_use -= bytes; |
8c2a3ca2 | 3701 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3702 | data_sinfo->flags, bytes, 0); |
6a63209f | 3703 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3704 | } |
3705 | ||
97e728d4 | 3706 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3707 | { |
97e728d4 JB |
3708 | struct list_head *head = &info->space_info; |
3709 | struct btrfs_space_info *found; | |
e3ccfa98 | 3710 | |
97e728d4 JB |
3711 | rcu_read_lock(); |
3712 | list_for_each_entry_rcu(found, head, list) { | |
3713 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3714 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3715 | } |
97e728d4 | 3716 | rcu_read_unlock(); |
e3ccfa98 JB |
3717 | } |
3718 | ||
3c76cd84 MX |
3719 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
3720 | { | |
3721 | return (global->size << 1); | |
3722 | } | |
3723 | ||
e5bc2458 | 3724 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3725 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3726 | { |
fb25e914 | 3727 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3728 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3729 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3730 | u64 thresh; |
e3ccfa98 | 3731 | |
0e4f8f88 CM |
3732 | if (force == CHUNK_ALLOC_FORCE) |
3733 | return 1; | |
3734 | ||
fb25e914 JB |
3735 | /* |
3736 | * We need to take into account the global rsv because for all intents | |
3737 | * and purposes it's used space. Don't worry about locking the | |
3738 | * global_rsv, it doesn't change except when the transaction commits. | |
3739 | */ | |
54338b5c | 3740 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 3741 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 3742 | |
0e4f8f88 CM |
3743 | /* |
3744 | * in limited mode, we want to have some free space up to | |
3745 | * about 1% of the FS size. | |
3746 | */ | |
3747 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3748 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3749 | thresh = max_t(u64, 64 * 1024 * 1024, |
3750 | div_factor_fine(thresh, 1)); | |
3751 | ||
3752 | if (num_bytes - num_allocated < thresh) | |
3753 | return 1; | |
3754 | } | |
0e4f8f88 | 3755 | |
698d0082 | 3756 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3757 | return 0; |
424499db | 3758 | return 1; |
32c00aff JB |
3759 | } |
3760 | ||
15d1ff81 LB |
3761 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3762 | { | |
3763 | u64 num_dev; | |
3764 | ||
53b381b3 DW |
3765 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
3766 | BTRFS_BLOCK_GROUP_RAID0 | | |
3767 | BTRFS_BLOCK_GROUP_RAID5 | | |
3768 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
3769 | num_dev = root->fs_info->fs_devices->rw_devices; |
3770 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3771 | num_dev = 2; | |
3772 | else | |
3773 | num_dev = 1; /* DUP or single */ | |
3774 | ||
3775 | /* metadata for updaing devices and chunk tree */ | |
3776 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3777 | } | |
3778 | ||
3779 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3780 | struct btrfs_root *root, u64 type) | |
3781 | { | |
3782 | struct btrfs_space_info *info; | |
3783 | u64 left; | |
3784 | u64 thresh; | |
3785 | ||
3786 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3787 | spin_lock(&info->lock); | |
3788 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3789 | info->bytes_reserved - info->bytes_readonly; | |
3790 | spin_unlock(&info->lock); | |
3791 | ||
3792 | thresh = get_system_chunk_thresh(root, type); | |
3793 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
c2cf52eb SK |
3794 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
3795 | left, thresh, type); | |
15d1ff81 LB |
3796 | dump_space_info(info, 0, 0); |
3797 | } | |
3798 | ||
3799 | if (left < thresh) { | |
3800 | u64 flags; | |
3801 | ||
3802 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3803 | btrfs_alloc_chunk(trans, root, flags); | |
3804 | } | |
3805 | } | |
3806 | ||
6324fbf3 | 3807 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3808 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3809 | { |
6324fbf3 | 3810 | struct btrfs_space_info *space_info; |
97e728d4 | 3811 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3812 | int wait_for_alloc = 0; |
9ed74f2d | 3813 | int ret = 0; |
9ed74f2d | 3814 | |
c6b305a8 JB |
3815 | /* Don't re-enter if we're already allocating a chunk */ |
3816 | if (trans->allocating_chunk) | |
3817 | return -ENOSPC; | |
3818 | ||
6324fbf3 | 3819 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3820 | if (!space_info) { |
3821 | ret = update_space_info(extent_root->fs_info, flags, | |
3822 | 0, 0, &space_info); | |
79787eaa | 3823 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3824 | } |
79787eaa | 3825 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3826 | |
6d74119f | 3827 | again: |
25179201 | 3828 | spin_lock(&space_info->lock); |
9e622d6b | 3829 | if (force < space_info->force_alloc) |
0e4f8f88 | 3830 | force = space_info->force_alloc; |
25179201 JB |
3831 | if (space_info->full) { |
3832 | spin_unlock(&space_info->lock); | |
6d74119f | 3833 | return 0; |
9ed74f2d JB |
3834 | } |
3835 | ||
698d0082 | 3836 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3837 | spin_unlock(&space_info->lock); |
6d74119f JB |
3838 | return 0; |
3839 | } else if (space_info->chunk_alloc) { | |
3840 | wait_for_alloc = 1; | |
3841 | } else { | |
3842 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3843 | } |
0e4f8f88 | 3844 | |
25179201 | 3845 | spin_unlock(&space_info->lock); |
9ed74f2d | 3846 | |
6d74119f JB |
3847 | mutex_lock(&fs_info->chunk_mutex); |
3848 | ||
3849 | /* | |
3850 | * The chunk_mutex is held throughout the entirety of a chunk | |
3851 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3852 | * other guy is done and we need to recheck and see if we should | |
3853 | * allocate. | |
3854 | */ | |
3855 | if (wait_for_alloc) { | |
3856 | mutex_unlock(&fs_info->chunk_mutex); | |
3857 | wait_for_alloc = 0; | |
3858 | goto again; | |
3859 | } | |
3860 | ||
c6b305a8 JB |
3861 | trans->allocating_chunk = true; |
3862 | ||
67377734 JB |
3863 | /* |
3864 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3865 | * allocating mixed chunks instead of individual chunks. | |
3866 | */ | |
3867 | if (btrfs_mixed_space_info(space_info)) | |
3868 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3869 | ||
97e728d4 JB |
3870 | /* |
3871 | * if we're doing a data chunk, go ahead and make sure that | |
3872 | * we keep a reasonable number of metadata chunks allocated in the | |
3873 | * FS as well. | |
3874 | */ | |
9ed74f2d | 3875 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3876 | fs_info->data_chunk_allocations++; |
3877 | if (!(fs_info->data_chunk_allocations % | |
3878 | fs_info->metadata_ratio)) | |
3879 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3880 | } |
3881 | ||
15d1ff81 LB |
3882 | /* |
3883 | * Check if we have enough space in SYSTEM chunk because we may need | |
3884 | * to update devices. | |
3885 | */ | |
3886 | check_system_chunk(trans, extent_root, flags); | |
3887 | ||
2b82032c | 3888 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3889 | trans->allocating_chunk = false; |
92b8e897 | 3890 | |
9ed74f2d | 3891 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
3892 | if (ret < 0 && ret != -ENOSPC) |
3893 | goto out; | |
9ed74f2d | 3894 | if (ret) |
6324fbf3 | 3895 | space_info->full = 1; |
424499db YZ |
3896 | else |
3897 | ret = 1; | |
6d74119f | 3898 | |
0e4f8f88 | 3899 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 3900 | out: |
6d74119f | 3901 | space_info->chunk_alloc = 0; |
9ed74f2d | 3902 | spin_unlock(&space_info->lock); |
a25c75d5 | 3903 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3904 | return ret; |
6324fbf3 | 3905 | } |
9ed74f2d | 3906 | |
a80c8dcf JB |
3907 | static int can_overcommit(struct btrfs_root *root, |
3908 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 3909 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 3910 | { |
96f1bb57 | 3911 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 3912 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3c76cd84 | 3913 | u64 space_size; |
a80c8dcf JB |
3914 | u64 avail; |
3915 | u64 used; | |
70afa399 | 3916 | u64 to_add; |
a80c8dcf JB |
3917 | |
3918 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
3919 | space_info->bytes_pinned + space_info->bytes_readonly; |
3920 | ||
96f1bb57 JB |
3921 | /* |
3922 | * We only want to allow over committing if we have lots of actual space | |
3923 | * free, but if we don't have enough space to handle the global reserve | |
3924 | * space then we could end up having a real enospc problem when trying | |
3925 | * to allocate a chunk or some other such important allocation. | |
3926 | */ | |
3c76cd84 MX |
3927 | spin_lock(&global_rsv->lock); |
3928 | space_size = calc_global_rsv_need_space(global_rsv); | |
3929 | spin_unlock(&global_rsv->lock); | |
3930 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
3931 | return 0; |
3932 | ||
3933 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
3934 | |
3935 | spin_lock(&root->fs_info->free_chunk_lock); | |
3936 | avail = root->fs_info->free_chunk_space; | |
3937 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3938 | ||
3939 | /* | |
3940 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
3941 | * space is actually useable. For raid56, the space info used |
3942 | * doesn't include the parity drive, so we don't have to | |
3943 | * change the math | |
a80c8dcf JB |
3944 | */ |
3945 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3946 | BTRFS_BLOCK_GROUP_RAID1 | | |
3947 | BTRFS_BLOCK_GROUP_RAID10)) | |
3948 | avail >>= 1; | |
3949 | ||
70afa399 JB |
3950 | to_add = space_info->total_bytes; |
3951 | ||
a80c8dcf | 3952 | /* |
561c294d MX |
3953 | * If we aren't flushing all things, let us overcommit up to |
3954 | * 1/2th of the space. If we can flush, don't let us overcommit | |
3955 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 3956 | */ |
08e007d2 | 3957 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
70afa399 | 3958 | to_add >>= 3; |
a80c8dcf | 3959 | else |
70afa399 | 3960 | to_add >>= 1; |
a80c8dcf | 3961 | |
70afa399 JB |
3962 | /* |
3963 | * Limit the overcommit to the amount of free space we could possibly | |
3964 | * allocate for chunks. | |
3965 | */ | |
3966 | to_add = min(avail, to_add); | |
a80c8dcf | 3967 | |
70afa399 | 3968 | if (used + bytes < space_info->total_bytes + to_add) |
a80c8dcf JB |
3969 | return 1; |
3970 | return 0; | |
3971 | } | |
3972 | ||
48a3b636 ES |
3973 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
3974 | unsigned long nr_pages) | |
da633a42 MX |
3975 | { |
3976 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 3977 | |
925a6efb JB |
3978 | if (down_read_trylock(&sb->s_umount)) { |
3979 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
3980 | up_read(&sb->s_umount); | |
3981 | } else { | |
da633a42 MX |
3982 | /* |
3983 | * We needn't worry the filesystem going from r/w to r/o though | |
3984 | * we don't acquire ->s_umount mutex, because the filesystem | |
3985 | * should guarantee the delalloc inodes list be empty after | |
3986 | * the filesystem is readonly(all dirty pages are written to | |
3987 | * the disk). | |
3988 | */ | |
eb73c1b7 | 3989 | btrfs_start_all_delalloc_inodes(root->fs_info, 0); |
98ad69cf | 3990 | if (!current->journal_info) |
199c2a9c | 3991 | btrfs_wait_all_ordered_extents(root->fs_info, 0); |
da633a42 MX |
3992 | } |
3993 | } | |
3994 | ||
9ed74f2d | 3995 | /* |
5da9d01b | 3996 | * shrink metadata reservation for delalloc |
9ed74f2d | 3997 | */ |
f4c738c2 JB |
3998 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3999 | bool wait_ordered) | |
5da9d01b | 4000 | { |
0ca1f7ce | 4001 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4002 | struct btrfs_space_info *space_info; |
663350ac | 4003 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4004 | u64 delalloc_bytes; |
5da9d01b | 4005 | u64 max_reclaim; |
b1953bce | 4006 | long time_left; |
877da174 | 4007 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 4008 | int loops = 0; |
08e007d2 | 4009 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4010 | |
663350ac | 4011 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4012 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4013 | space_info = block_rsv->space_info; |
bf9022e0 CM |
4014 | |
4015 | smp_mb(); | |
963d678b MX |
4016 | delalloc_bytes = percpu_counter_sum_positive( |
4017 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4018 | if (delalloc_bytes == 0) { |
fdb5effd | 4019 | if (trans) |
f4c738c2 | 4020 | return; |
199c2a9c | 4021 | btrfs_wait_all_ordered_extents(root->fs_info, 0); |
f4c738c2 | 4022 | return; |
fdb5effd JB |
4023 | } |
4024 | ||
f4c738c2 JB |
4025 | while (delalloc_bytes && loops < 3) { |
4026 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
4027 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
da633a42 | 4028 | btrfs_writeback_inodes_sb_nr(root, nr_pages); |
dea31f52 JB |
4029 | /* |
4030 | * We need to wait for the async pages to actually start before | |
4031 | * we do anything. | |
4032 | */ | |
4033 | wait_event(root->fs_info->async_submit_wait, | |
4034 | !atomic_read(&root->fs_info->async_delalloc_pages)); | |
4035 | ||
08e007d2 MX |
4036 | if (!trans) |
4037 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4038 | else | |
4039 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4040 | spin_lock(&space_info->lock); |
08e007d2 | 4041 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4042 | spin_unlock(&space_info->lock); |
4043 | break; | |
4044 | } | |
0019f10d | 4045 | spin_unlock(&space_info->lock); |
5da9d01b | 4046 | |
36e39c40 | 4047 | loops++; |
f104d044 | 4048 | if (wait_ordered && !trans) { |
199c2a9c | 4049 | btrfs_wait_all_ordered_extents(root->fs_info, 0); |
f104d044 | 4050 | } else { |
f4c738c2 | 4051 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4052 | if (time_left) |
4053 | break; | |
4054 | } | |
f4c738c2 | 4055 | smp_mb(); |
963d678b MX |
4056 | delalloc_bytes = percpu_counter_sum_positive( |
4057 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4058 | } |
5da9d01b YZ |
4059 | } |
4060 | ||
663350ac JB |
4061 | /** |
4062 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4063 | * @root - the root we're allocating for | |
4064 | * @bytes - the number of bytes we want to reserve | |
4065 | * @force - force the commit | |
8bb8ab2e | 4066 | * |
663350ac JB |
4067 | * This will check to make sure that committing the transaction will actually |
4068 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4069 | * will return -ENOSPC. | |
8bb8ab2e | 4070 | */ |
663350ac JB |
4071 | static int may_commit_transaction(struct btrfs_root *root, |
4072 | struct btrfs_space_info *space_info, | |
4073 | u64 bytes, int force) | |
4074 | { | |
4075 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4076 | struct btrfs_trans_handle *trans; | |
4077 | ||
4078 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4079 | if (trans) | |
4080 | return -EAGAIN; | |
4081 | ||
4082 | if (force) | |
4083 | goto commit; | |
4084 | ||
4085 | /* See if there is enough pinned space to make this reservation */ | |
4086 | spin_lock(&space_info->lock); | |
b150a4f1 JB |
4087 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4088 | bytes) >= 0) { | |
663350ac JB |
4089 | spin_unlock(&space_info->lock); |
4090 | goto commit; | |
4091 | } | |
4092 | spin_unlock(&space_info->lock); | |
4093 | ||
4094 | /* | |
4095 | * See if there is some space in the delayed insertion reservation for | |
4096 | * this reservation. | |
4097 | */ | |
4098 | if (space_info != delayed_rsv->space_info) | |
4099 | return -ENOSPC; | |
4100 | ||
d9b0218f | 4101 | spin_lock(&space_info->lock); |
663350ac | 4102 | spin_lock(&delayed_rsv->lock); |
b150a4f1 JB |
4103 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4104 | bytes - delayed_rsv->size) >= 0) { | |
663350ac | 4105 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 4106 | spin_unlock(&space_info->lock); |
663350ac JB |
4107 | return -ENOSPC; |
4108 | } | |
4109 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 4110 | spin_unlock(&space_info->lock); |
663350ac JB |
4111 | |
4112 | commit: | |
4113 | trans = btrfs_join_transaction(root); | |
4114 | if (IS_ERR(trans)) | |
4115 | return -ENOSPC; | |
4116 | ||
4117 | return btrfs_commit_transaction(trans, root); | |
4118 | } | |
4119 | ||
96c3f433 | 4120 | enum flush_state { |
67b0fd63 JB |
4121 | FLUSH_DELAYED_ITEMS_NR = 1, |
4122 | FLUSH_DELAYED_ITEMS = 2, | |
4123 | FLUSH_DELALLOC = 3, | |
4124 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4125 | ALLOC_CHUNK = 5, |
4126 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4127 | }; |
4128 | ||
4129 | static int flush_space(struct btrfs_root *root, | |
4130 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4131 | u64 orig_bytes, int state) | |
4132 | { | |
4133 | struct btrfs_trans_handle *trans; | |
4134 | int nr; | |
f4c738c2 | 4135 | int ret = 0; |
96c3f433 JB |
4136 | |
4137 | switch (state) { | |
96c3f433 JB |
4138 | case FLUSH_DELAYED_ITEMS_NR: |
4139 | case FLUSH_DELAYED_ITEMS: | |
4140 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
4141 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4142 | ||
4143 | nr = (int)div64_u64(num_bytes, bytes); | |
4144 | if (!nr) | |
4145 | nr = 1; | |
4146 | nr *= 2; | |
4147 | } else { | |
4148 | nr = -1; | |
4149 | } | |
4150 | trans = btrfs_join_transaction(root); | |
4151 | if (IS_ERR(trans)) { | |
4152 | ret = PTR_ERR(trans); | |
4153 | break; | |
4154 | } | |
4155 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4156 | btrfs_end_transaction(trans, root); | |
4157 | break; | |
67b0fd63 JB |
4158 | case FLUSH_DELALLOC: |
4159 | case FLUSH_DELALLOC_WAIT: | |
4160 | shrink_delalloc(root, num_bytes, orig_bytes, | |
4161 | state == FLUSH_DELALLOC_WAIT); | |
4162 | break; | |
ea658bad JB |
4163 | case ALLOC_CHUNK: |
4164 | trans = btrfs_join_transaction(root); | |
4165 | if (IS_ERR(trans)) { | |
4166 | ret = PTR_ERR(trans); | |
4167 | break; | |
4168 | } | |
4169 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4170 | btrfs_get_alloc_profile(root, 0), |
4171 | CHUNK_ALLOC_NO_FORCE); | |
4172 | btrfs_end_transaction(trans, root); | |
4173 | if (ret == -ENOSPC) | |
4174 | ret = 0; | |
4175 | break; | |
96c3f433 JB |
4176 | case COMMIT_TRANS: |
4177 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4178 | break; | |
4179 | default: | |
4180 | ret = -ENOSPC; | |
4181 | break; | |
4182 | } | |
4183 | ||
4184 | return ret; | |
4185 | } | |
4a92b1b8 JB |
4186 | /** |
4187 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4188 | * @root - the root we're allocating for | |
4189 | * @block_rsv - the block_rsv we're allocating for | |
4190 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4191 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4192 | * |
4a92b1b8 JB |
4193 | * This will reserve orgi_bytes number of bytes from the space info associated |
4194 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4195 | * flush out space to make room. It will do this by flushing delalloc if | |
4196 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4197 | * regain reservations will be made and this will fail if there is not enough | |
4198 | * space already. | |
8bb8ab2e | 4199 | */ |
4a92b1b8 | 4200 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4201 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4202 | u64 orig_bytes, |
4203 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4204 | { |
f0486c68 | 4205 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4206 | u64 used; |
8bb8ab2e | 4207 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4208 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4209 | int ret = 0; |
fdb5effd | 4210 | bool flushing = false; |
9ed74f2d | 4211 | |
8bb8ab2e | 4212 | again: |
fdb5effd | 4213 | ret = 0; |
8bb8ab2e | 4214 | spin_lock(&space_info->lock); |
fdb5effd | 4215 | /* |
08e007d2 MX |
4216 | * We only want to wait if somebody other than us is flushing and we |
4217 | * are actually allowed to flush all things. | |
fdb5effd | 4218 | */ |
08e007d2 MX |
4219 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4220 | space_info->flush) { | |
fdb5effd JB |
4221 | spin_unlock(&space_info->lock); |
4222 | /* | |
4223 | * If we have a trans handle we can't wait because the flusher | |
4224 | * may have to commit the transaction, which would mean we would | |
4225 | * deadlock since we are waiting for the flusher to finish, but | |
4226 | * hold the current transaction open. | |
4227 | */ | |
663350ac | 4228 | if (current->journal_info) |
fdb5effd | 4229 | return -EAGAIN; |
b9688bb8 AJ |
4230 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4231 | /* Must have been killed, return */ | |
4232 | if (ret) | |
fdb5effd JB |
4233 | return -EINTR; |
4234 | ||
4235 | spin_lock(&space_info->lock); | |
4236 | } | |
4237 | ||
4238 | ret = -ENOSPC; | |
2bf64758 JB |
4239 | used = space_info->bytes_used + space_info->bytes_reserved + |
4240 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4241 | space_info->bytes_may_use; | |
9ed74f2d | 4242 | |
8bb8ab2e JB |
4243 | /* |
4244 | * The idea here is that we've not already over-reserved the block group | |
4245 | * then we can go ahead and save our reservation first and then start | |
4246 | * flushing if we need to. Otherwise if we've already overcommitted | |
4247 | * lets start flushing stuff first and then come back and try to make | |
4248 | * our reservation. | |
4249 | */ | |
2bf64758 JB |
4250 | if (used <= space_info->total_bytes) { |
4251 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4252 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4253 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4254 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4255 | ret = 0; |
4256 | } else { | |
4257 | /* | |
4258 | * Ok set num_bytes to orig_bytes since we aren't | |
4259 | * overocmmitted, this way we only try and reclaim what | |
4260 | * we need. | |
4261 | */ | |
4262 | num_bytes = orig_bytes; | |
4263 | } | |
4264 | } else { | |
4265 | /* | |
4266 | * Ok we're over committed, set num_bytes to the overcommitted | |
4267 | * amount plus the amount of bytes that we need for this | |
4268 | * reservation. | |
4269 | */ | |
2bf64758 | 4270 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4271 | (orig_bytes * 2); |
8bb8ab2e | 4272 | } |
9ed74f2d | 4273 | |
44734ed1 JB |
4274 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4275 | space_info->bytes_may_use += orig_bytes; | |
4276 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4277 | space_info->flags, orig_bytes, | |
4278 | 1); | |
4279 | ret = 0; | |
2bf64758 JB |
4280 | } |
4281 | ||
8bb8ab2e JB |
4282 | /* |
4283 | * Couldn't make our reservation, save our place so while we're trying | |
4284 | * to reclaim space we can actually use it instead of somebody else | |
4285 | * stealing it from us. | |
08e007d2 MX |
4286 | * |
4287 | * We make the other tasks wait for the flush only when we can flush | |
4288 | * all things. | |
8bb8ab2e | 4289 | */ |
72bcd99d | 4290 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4291 | flushing = true; |
4292 | space_info->flush = 1; | |
8bb8ab2e | 4293 | } |
9ed74f2d | 4294 | |
f0486c68 | 4295 | spin_unlock(&space_info->lock); |
9ed74f2d | 4296 | |
08e007d2 | 4297 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4298 | goto out; |
f0486c68 | 4299 | |
96c3f433 JB |
4300 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4301 | flush_state); | |
4302 | flush_state++; | |
08e007d2 MX |
4303 | |
4304 | /* | |
4305 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4306 | * would happen. So skip delalloc flush. | |
4307 | */ | |
4308 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4309 | (flush_state == FLUSH_DELALLOC || | |
4310 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4311 | flush_state = ALLOC_CHUNK; | |
4312 | ||
96c3f433 | 4313 | if (!ret) |
8bb8ab2e | 4314 | goto again; |
08e007d2 MX |
4315 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4316 | flush_state < COMMIT_TRANS) | |
4317 | goto again; | |
4318 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4319 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4320 | goto again; |
4321 | ||
4322 | out: | |
5d80366e JB |
4323 | if (ret == -ENOSPC && |
4324 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
4325 | struct btrfs_block_rsv *global_rsv = | |
4326 | &root->fs_info->global_block_rsv; | |
4327 | ||
4328 | if (block_rsv != global_rsv && | |
4329 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
4330 | ret = 0; | |
4331 | } | |
fdb5effd | 4332 | if (flushing) { |
8bb8ab2e | 4333 | spin_lock(&space_info->lock); |
fdb5effd JB |
4334 | space_info->flush = 0; |
4335 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4336 | spin_unlock(&space_info->lock); |
f0486c68 | 4337 | } |
f0486c68 YZ |
4338 | return ret; |
4339 | } | |
4340 | ||
79787eaa JM |
4341 | static struct btrfs_block_rsv *get_block_rsv( |
4342 | const struct btrfs_trans_handle *trans, | |
4343 | const struct btrfs_root *root) | |
f0486c68 | 4344 | { |
4c13d758 JB |
4345 | struct btrfs_block_rsv *block_rsv = NULL; |
4346 | ||
0e721106 JB |
4347 | if (root->ref_cows) |
4348 | block_rsv = trans->block_rsv; | |
4349 | ||
4350 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4351 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4352 | |
4353 | if (!block_rsv) | |
f0486c68 YZ |
4354 | block_rsv = root->block_rsv; |
4355 | ||
4356 | if (!block_rsv) | |
4357 | block_rsv = &root->fs_info->empty_block_rsv; | |
4358 | ||
4359 | return block_rsv; | |
4360 | } | |
4361 | ||
4362 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4363 | u64 num_bytes) | |
4364 | { | |
4365 | int ret = -ENOSPC; | |
4366 | spin_lock(&block_rsv->lock); | |
4367 | if (block_rsv->reserved >= num_bytes) { | |
4368 | block_rsv->reserved -= num_bytes; | |
4369 | if (block_rsv->reserved < block_rsv->size) | |
4370 | block_rsv->full = 0; | |
4371 | ret = 0; | |
4372 | } | |
4373 | spin_unlock(&block_rsv->lock); | |
4374 | return ret; | |
4375 | } | |
4376 | ||
4377 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4378 | u64 num_bytes, int update_size) | |
4379 | { | |
4380 | spin_lock(&block_rsv->lock); | |
4381 | block_rsv->reserved += num_bytes; | |
4382 | if (update_size) | |
4383 | block_rsv->size += num_bytes; | |
4384 | else if (block_rsv->reserved >= block_rsv->size) | |
4385 | block_rsv->full = 1; | |
4386 | spin_unlock(&block_rsv->lock); | |
4387 | } | |
4388 | ||
d52be818 JB |
4389 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
4390 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
4391 | int min_factor) | |
4392 | { | |
4393 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
4394 | u64 min_bytes; | |
4395 | ||
4396 | if (global_rsv->space_info != dest->space_info) | |
4397 | return -ENOSPC; | |
4398 | ||
4399 | spin_lock(&global_rsv->lock); | |
4400 | min_bytes = div_factor(global_rsv->size, min_factor); | |
4401 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
4402 | spin_unlock(&global_rsv->lock); | |
4403 | return -ENOSPC; | |
4404 | } | |
4405 | global_rsv->reserved -= num_bytes; | |
4406 | if (global_rsv->reserved < global_rsv->size) | |
4407 | global_rsv->full = 0; | |
4408 | spin_unlock(&global_rsv->lock); | |
4409 | ||
4410 | block_rsv_add_bytes(dest, num_bytes, 1); | |
4411 | return 0; | |
4412 | } | |
4413 | ||
8c2a3ca2 JB |
4414 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4415 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4416 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4417 | { |
4418 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4419 | ||
4420 | spin_lock(&block_rsv->lock); | |
4421 | if (num_bytes == (u64)-1) | |
4422 | num_bytes = block_rsv->size; | |
4423 | block_rsv->size -= num_bytes; | |
4424 | if (block_rsv->reserved >= block_rsv->size) { | |
4425 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4426 | block_rsv->reserved = block_rsv->size; | |
4427 | block_rsv->full = 1; | |
4428 | } else { | |
4429 | num_bytes = 0; | |
4430 | } | |
4431 | spin_unlock(&block_rsv->lock); | |
4432 | ||
4433 | if (num_bytes > 0) { | |
4434 | if (dest) { | |
e9e22899 JB |
4435 | spin_lock(&dest->lock); |
4436 | if (!dest->full) { | |
4437 | u64 bytes_to_add; | |
4438 | ||
4439 | bytes_to_add = dest->size - dest->reserved; | |
4440 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4441 | dest->reserved += bytes_to_add; | |
4442 | if (dest->reserved >= dest->size) | |
4443 | dest->full = 1; | |
4444 | num_bytes -= bytes_to_add; | |
4445 | } | |
4446 | spin_unlock(&dest->lock); | |
4447 | } | |
4448 | if (num_bytes) { | |
f0486c68 | 4449 | spin_lock(&space_info->lock); |
fb25e914 | 4450 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4451 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4452 | space_info->flags, num_bytes, 0); |
36e39c40 | 4453 | space_info->reservation_progress++; |
f0486c68 | 4454 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4455 | } |
9ed74f2d | 4456 | } |
f0486c68 | 4457 | } |
4e06bdd6 | 4458 | |
f0486c68 YZ |
4459 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4460 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4461 | { | |
4462 | int ret; | |
9ed74f2d | 4463 | |
f0486c68 YZ |
4464 | ret = block_rsv_use_bytes(src, num_bytes); |
4465 | if (ret) | |
4466 | return ret; | |
9ed74f2d | 4467 | |
f0486c68 | 4468 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4469 | return 0; |
4470 | } | |
4471 | ||
66d8f3dd | 4472 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4473 | { |
f0486c68 YZ |
4474 | memset(rsv, 0, sizeof(*rsv)); |
4475 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4476 | rsv->type = type; |
f0486c68 YZ |
4477 | } |
4478 | ||
66d8f3dd MX |
4479 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4480 | unsigned short type) | |
f0486c68 YZ |
4481 | { |
4482 | struct btrfs_block_rsv *block_rsv; | |
4483 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4484 | |
f0486c68 YZ |
4485 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4486 | if (!block_rsv) | |
4487 | return NULL; | |
9ed74f2d | 4488 | |
66d8f3dd | 4489 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4490 | block_rsv->space_info = __find_space_info(fs_info, |
4491 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4492 | return block_rsv; |
4493 | } | |
9ed74f2d | 4494 | |
f0486c68 YZ |
4495 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4496 | struct btrfs_block_rsv *rsv) | |
4497 | { | |
2aaa6655 JB |
4498 | if (!rsv) |
4499 | return; | |
dabdb640 JB |
4500 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4501 | kfree(rsv); | |
9ed74f2d JB |
4502 | } |
4503 | ||
08e007d2 MX |
4504 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4505 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4506 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4507 | { |
f0486c68 | 4508 | int ret; |
9ed74f2d | 4509 | |
f0486c68 YZ |
4510 | if (num_bytes == 0) |
4511 | return 0; | |
8bb8ab2e | 4512 | |
61b520a9 | 4513 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4514 | if (!ret) { |
4515 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4516 | return 0; | |
4517 | } | |
9ed74f2d | 4518 | |
f0486c68 | 4519 | return ret; |
f0486c68 | 4520 | } |
9ed74f2d | 4521 | |
4a92b1b8 | 4522 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4523 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4524 | { |
4525 | u64 num_bytes = 0; | |
f0486c68 | 4526 | int ret = -ENOSPC; |
9ed74f2d | 4527 | |
f0486c68 YZ |
4528 | if (!block_rsv) |
4529 | return 0; | |
9ed74f2d | 4530 | |
f0486c68 | 4531 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4532 | num_bytes = div_factor(block_rsv->size, min_factor); |
4533 | if (block_rsv->reserved >= num_bytes) | |
4534 | ret = 0; | |
4535 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4536 | |
36ba022a JB |
4537 | return ret; |
4538 | } | |
4539 | ||
08e007d2 MX |
4540 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4541 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4542 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4543 | { |
4544 | u64 num_bytes = 0; | |
4545 | int ret = -ENOSPC; | |
4546 | ||
4547 | if (!block_rsv) | |
4548 | return 0; | |
4549 | ||
4550 | spin_lock(&block_rsv->lock); | |
4551 | num_bytes = min_reserved; | |
13553e52 | 4552 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4553 | ret = 0; |
13553e52 | 4554 | else |
f0486c68 | 4555 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4556 | spin_unlock(&block_rsv->lock); |
13553e52 | 4557 | |
f0486c68 YZ |
4558 | if (!ret) |
4559 | return 0; | |
4560 | ||
aa38a711 | 4561 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4562 | if (!ret) { |
4563 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4564 | return 0; |
6a63209f | 4565 | } |
9ed74f2d | 4566 | |
13553e52 | 4567 | return ret; |
f0486c68 YZ |
4568 | } |
4569 | ||
4570 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4571 | struct btrfs_block_rsv *dst_rsv, | |
4572 | u64 num_bytes) | |
4573 | { | |
4574 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4575 | } | |
4576 | ||
4577 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4578 | struct btrfs_block_rsv *block_rsv, | |
4579 | u64 num_bytes) | |
4580 | { | |
4581 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4582 | if (global_rsv->full || global_rsv == block_rsv || | |
4583 | block_rsv->space_info != global_rsv->space_info) | |
4584 | global_rsv = NULL; | |
8c2a3ca2 JB |
4585 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4586 | num_bytes); | |
6a63209f JB |
4587 | } |
4588 | ||
4589 | /* | |
8929ecfa YZ |
4590 | * helper to calculate size of global block reservation. |
4591 | * the desired value is sum of space used by extent tree, | |
4592 | * checksum tree and root tree | |
6a63209f | 4593 | */ |
8929ecfa | 4594 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4595 | { |
8929ecfa YZ |
4596 | struct btrfs_space_info *sinfo; |
4597 | u64 num_bytes; | |
4598 | u64 meta_used; | |
4599 | u64 data_used; | |
6c41761f | 4600 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4601 | |
8929ecfa YZ |
4602 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4603 | spin_lock(&sinfo->lock); | |
4604 | data_used = sinfo->bytes_used; | |
4605 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4606 | |
8929ecfa YZ |
4607 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4608 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4609 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4610 | data_used = 0; | |
8929ecfa YZ |
4611 | meta_used = sinfo->bytes_used; |
4612 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4613 | |
8929ecfa YZ |
4614 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4615 | csum_size * 2; | |
4616 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4617 | |
8929ecfa | 4618 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4619 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4620 | |
8929ecfa YZ |
4621 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4622 | } | |
6a63209f | 4623 | |
8929ecfa YZ |
4624 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4625 | { | |
4626 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4627 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4628 | u64 num_bytes; | |
6a63209f | 4629 | |
8929ecfa | 4630 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4631 | |
8929ecfa | 4632 | spin_lock(&sinfo->lock); |
1f699d38 | 4633 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4634 | |
fdf30d1c | 4635 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 4636 | |
8929ecfa | 4637 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4638 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4639 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4640 | |
4641 | if (sinfo->total_bytes > num_bytes) { | |
4642 | num_bytes = sinfo->total_bytes - num_bytes; | |
4643 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4644 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4645 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4646 | sinfo->flags, num_bytes, 1); |
6a63209f | 4647 | } |
6a63209f | 4648 | |
8929ecfa YZ |
4649 | if (block_rsv->reserved >= block_rsv->size) { |
4650 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4651 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4652 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4653 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4654 | sinfo->reservation_progress++; |
8929ecfa YZ |
4655 | block_rsv->reserved = block_rsv->size; |
4656 | block_rsv->full = 1; | |
4657 | } | |
182608c8 | 4658 | |
8929ecfa | 4659 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4660 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4661 | } |
4662 | ||
f0486c68 | 4663 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4664 | { |
f0486c68 | 4665 | struct btrfs_space_info *space_info; |
6a63209f | 4666 | |
f0486c68 YZ |
4667 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4668 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4669 | |
f0486c68 | 4670 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4671 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4672 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4673 | fs_info->trans_block_rsv.space_info = space_info; |
4674 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4675 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4676 | |
8929ecfa YZ |
4677 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4678 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4679 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4680 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
4681 | if (fs_info->quota_root) |
4682 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4683 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4684 | |
8929ecfa | 4685 | update_global_block_rsv(fs_info); |
6a63209f JB |
4686 | } |
4687 | ||
8929ecfa | 4688 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4689 | { |
8c2a3ca2 JB |
4690 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4691 | (u64)-1); | |
8929ecfa YZ |
4692 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4693 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4694 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4695 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4696 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4697 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4698 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4699 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4700 | } |
4701 | ||
a22285a6 YZ |
4702 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4703 | struct btrfs_root *root) | |
6a63209f | 4704 | { |
0e721106 JB |
4705 | if (!trans->block_rsv) |
4706 | return; | |
4707 | ||
a22285a6 YZ |
4708 | if (!trans->bytes_reserved) |
4709 | return; | |
6a63209f | 4710 | |
e77266e4 | 4711 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4712 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4713 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4714 | trans->bytes_reserved = 0; |
4715 | } | |
6a63209f | 4716 | |
79787eaa | 4717 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4718 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4719 | struct inode *inode) | |
4720 | { | |
4721 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4722 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4723 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4724 | ||
4725 | /* | |
fcb80c2a JB |
4726 | * We need to hold space in order to delete our orphan item once we've |
4727 | * added it, so this takes the reservation so we can release it later | |
4728 | * when we are truly done with the orphan item. | |
d68fc57b | 4729 | */ |
ff5714cc | 4730 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4731 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4732 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4733 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4734 | } |
4735 | ||
d68fc57b | 4736 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4737 | { |
d68fc57b | 4738 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4739 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4740 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4741 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4742 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4743 | } | |
97e728d4 | 4744 | |
d5c12070 MX |
4745 | /* |
4746 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
4747 | * root: the root of the parent directory | |
4748 | * rsv: block reservation | |
4749 | * items: the number of items that we need do reservation | |
4750 | * qgroup_reserved: used to return the reserved size in qgroup | |
4751 | * | |
4752 | * This function is used to reserve the space for snapshot/subvolume | |
4753 | * creation and deletion. Those operations are different with the | |
4754 | * common file/directory operations, they change two fs/file trees | |
4755 | * and root tree, the number of items that the qgroup reserves is | |
4756 | * different with the free space reservation. So we can not use | |
4757 | * the space reseravtion mechanism in start_transaction(). | |
4758 | */ | |
4759 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
4760 | struct btrfs_block_rsv *rsv, | |
4761 | int items, | |
ee3441b4 JM |
4762 | u64 *qgroup_reserved, |
4763 | bool use_global_rsv) | |
a22285a6 | 4764 | { |
d5c12070 MX |
4765 | u64 num_bytes; |
4766 | int ret; | |
ee3441b4 | 4767 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
4768 | |
4769 | if (root->fs_info->quota_enabled) { | |
4770 | /* One for parent inode, two for dir entries */ | |
4771 | num_bytes = 3 * root->leafsize; | |
4772 | ret = btrfs_qgroup_reserve(root, num_bytes); | |
4773 | if (ret) | |
4774 | return ret; | |
4775 | } else { | |
4776 | num_bytes = 0; | |
4777 | } | |
4778 | ||
4779 | *qgroup_reserved = num_bytes; | |
4780 | ||
4781 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
4782 | rsv->space_info = __find_space_info(root->fs_info, | |
4783 | BTRFS_BLOCK_GROUP_METADATA); | |
4784 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
4785 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
4786 | |
4787 | if (ret == -ENOSPC && use_global_rsv) | |
4788 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
4789 | ||
d5c12070 MX |
4790 | if (ret) { |
4791 | if (*qgroup_reserved) | |
4792 | btrfs_qgroup_free(root, *qgroup_reserved); | |
4793 | } | |
4794 | ||
4795 | return ret; | |
4796 | } | |
4797 | ||
4798 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
4799 | struct btrfs_block_rsv *rsv, | |
4800 | u64 qgroup_reserved) | |
4801 | { | |
4802 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
4803 | if (qgroup_reserved) | |
4804 | btrfs_qgroup_free(root, qgroup_reserved); | |
97e728d4 JB |
4805 | } |
4806 | ||
7709cde3 JB |
4807 | /** |
4808 | * drop_outstanding_extent - drop an outstanding extent | |
4809 | * @inode: the inode we're dropping the extent for | |
4810 | * | |
4811 | * This is called when we are freeing up an outstanding extent, either called | |
4812 | * after an error or after an extent is written. This will return the number of | |
4813 | * reserved extents that need to be freed. This must be called with | |
4814 | * BTRFS_I(inode)->lock held. | |
4815 | */ | |
9e0baf60 JB |
4816 | static unsigned drop_outstanding_extent(struct inode *inode) |
4817 | { | |
7fd2ae21 | 4818 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4819 | unsigned dropped_extents = 0; |
4820 | ||
9e0baf60 JB |
4821 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4822 | BTRFS_I(inode)->outstanding_extents--; | |
4823 | ||
7fd2ae21 | 4824 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4825 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4826 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4827 | drop_inode_space = 1; |
7fd2ae21 | 4828 | |
9e0baf60 JB |
4829 | /* |
4830 | * If we have more or the same amount of outsanding extents than we have | |
4831 | * reserved then we need to leave the reserved extents count alone. | |
4832 | */ | |
4833 | if (BTRFS_I(inode)->outstanding_extents >= | |
4834 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4835 | return drop_inode_space; |
9e0baf60 JB |
4836 | |
4837 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4838 | BTRFS_I(inode)->outstanding_extents; | |
4839 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4840 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4841 | } |
4842 | ||
7709cde3 JB |
4843 | /** |
4844 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4845 | * reserved/free'd for the given bytes. | |
4846 | * @inode: the inode we're manipulating | |
4847 | * @num_bytes: the number of bytes in question | |
4848 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4849 | * | |
4850 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4851 | * correct amount of metadata that must either be reserved or freed. We | |
4852 | * calculate how many checksums we can fit into one leaf and then divide the | |
4853 | * number of bytes that will need to be checksumed by this value to figure out | |
4854 | * how many checksums will be required. If we are adding bytes then the number | |
4855 | * may go up and we will return the number of additional bytes that must be | |
4856 | * reserved. If it is going down we will return the number of bytes that must | |
4857 | * be freed. | |
4858 | * | |
4859 | * This must be called with BTRFS_I(inode)->lock held. | |
4860 | */ | |
4861 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4862 | int reserve) | |
6324fbf3 | 4863 | { |
7709cde3 JB |
4864 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4865 | u64 csum_size; | |
4866 | int num_csums_per_leaf; | |
4867 | int num_csums; | |
4868 | int old_csums; | |
4869 | ||
4870 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4871 | BTRFS_I(inode)->csum_bytes == 0) | |
4872 | return 0; | |
4873 | ||
4874 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4875 | if (reserve) | |
4876 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4877 | else | |
4878 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4879 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4880 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4881 | sizeof(struct btrfs_csum_item) + | |
4882 | sizeof(struct btrfs_disk_key)); | |
4883 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4884 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4885 | num_csums = num_csums / num_csums_per_leaf; | |
4886 | ||
4887 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4888 | old_csums = old_csums / num_csums_per_leaf; | |
4889 | ||
4890 | /* No change, no need to reserve more */ | |
4891 | if (old_csums == num_csums) | |
4892 | return 0; | |
4893 | ||
4894 | if (reserve) | |
4895 | return btrfs_calc_trans_metadata_size(root, | |
4896 | num_csums - old_csums); | |
4897 | ||
4898 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4899 | } |
c146afad | 4900 | |
0ca1f7ce YZ |
4901 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4902 | { | |
4903 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4904 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4905 | u64 to_reserve = 0; |
660d3f6c | 4906 | u64 csum_bytes; |
9e0baf60 | 4907 | unsigned nr_extents = 0; |
660d3f6c | 4908 | int extra_reserve = 0; |
08e007d2 | 4909 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4910 | int ret = 0; |
c64c2bd8 | 4911 | bool delalloc_lock = true; |
88e081bf WS |
4912 | u64 to_free = 0; |
4913 | unsigned dropped; | |
6324fbf3 | 4914 | |
c64c2bd8 JB |
4915 | /* If we are a free space inode we need to not flush since we will be in |
4916 | * the middle of a transaction commit. We also don't need the delalloc | |
4917 | * mutex since we won't race with anybody. We need this mostly to make | |
4918 | * lockdep shut its filthy mouth. | |
4919 | */ | |
4920 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4921 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
4922 | delalloc_lock = false; |
4923 | } | |
c09544e0 | 4924 | |
08e007d2 MX |
4925 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
4926 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4927 | schedule_timeout(1); |
ec44a35c | 4928 | |
c64c2bd8 JB |
4929 | if (delalloc_lock) |
4930 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
4931 | ||
0ca1f7ce | 4932 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4933 | |
9e0baf60 JB |
4934 | spin_lock(&BTRFS_I(inode)->lock); |
4935 | BTRFS_I(inode)->outstanding_extents++; | |
4936 | ||
4937 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4938 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4939 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4940 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4941 | |
7fd2ae21 JB |
4942 | /* |
4943 | * Add an item to reserve for updating the inode when we complete the | |
4944 | * delalloc io. | |
4945 | */ | |
72ac3c0d JB |
4946 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4947 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4948 | nr_extents++; |
660d3f6c | 4949 | extra_reserve = 1; |
593060d7 | 4950 | } |
7fd2ae21 JB |
4951 | |
4952 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4953 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4954 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4955 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4956 | |
88e081bf | 4957 | if (root->fs_info->quota_enabled) { |
c5567237 AJ |
4958 | ret = btrfs_qgroup_reserve(root, num_bytes + |
4959 | nr_extents * root->leafsize); | |
88e081bf WS |
4960 | if (ret) |
4961 | goto out_fail; | |
4962 | } | |
c5567237 | 4963 | |
88e081bf WS |
4964 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
4965 | if (unlikely(ret)) { | |
4966 | if (root->fs_info->quota_enabled) | |
4b5829a8 MX |
4967 | btrfs_qgroup_free(root, num_bytes + |
4968 | nr_extents * root->leafsize); | |
88e081bf | 4969 | goto out_fail; |
9e0baf60 | 4970 | } |
25179201 | 4971 | |
660d3f6c JB |
4972 | spin_lock(&BTRFS_I(inode)->lock); |
4973 | if (extra_reserve) { | |
72ac3c0d JB |
4974 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4975 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4976 | nr_extents--; |
4977 | } | |
4978 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4979 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
4980 | |
4981 | if (delalloc_lock) | |
4982 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 4983 | |
8c2a3ca2 JB |
4984 | if (to_reserve) |
4985 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4986 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4987 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4988 | ||
0ca1f7ce | 4989 | return 0; |
88e081bf WS |
4990 | |
4991 | out_fail: | |
4992 | spin_lock(&BTRFS_I(inode)->lock); | |
4993 | dropped = drop_outstanding_extent(inode); | |
4994 | /* | |
4995 | * If the inodes csum_bytes is the same as the original | |
4996 | * csum_bytes then we know we haven't raced with any free()ers | |
4997 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 4998 | */ |
f4881bc7 | 4999 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 5000 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
5001 | } else { |
5002 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5003 | u64 bytes; | |
5004 | ||
5005 | /* | |
5006 | * This is tricky, but first we need to figure out how much we | |
5007 | * free'd from any free-ers that occured during this | |
5008 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
5009 | * before we dropped our lock, and then call the free for the | |
5010 | * number of bytes that were freed while we were trying our | |
5011 | * reservation. | |
5012 | */ | |
5013 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5014 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5015 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5016 | ||
5017 | ||
5018 | /* | |
5019 | * Now we need to see how much we would have freed had we not | |
5020 | * been making this reservation and our ->csum_bytes were not | |
5021 | * artificially inflated. | |
5022 | */ | |
5023 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5024 | bytes = csum_bytes - orig_csum_bytes; | |
5025 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5026 | ||
5027 | /* | |
5028 | * Now reset ->csum_bytes to what it should be. If bytes is | |
5029 | * more than to_free then we would have free'd more space had we | |
5030 | * not had an artificially high ->csum_bytes, so we need to free | |
5031 | * the remainder. If bytes is the same or less then we don't | |
5032 | * need to do anything, the other free-ers did the correct | |
5033 | * thing. | |
5034 | */ | |
5035 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5036 | if (bytes > to_free) | |
5037 | to_free = bytes - to_free; | |
5038 | else | |
5039 | to_free = 0; | |
5040 | } | |
88e081bf WS |
5041 | spin_unlock(&BTRFS_I(inode)->lock); |
5042 | if (dropped) | |
5043 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
5044 | ||
5045 | if (to_free) { | |
5046 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5047 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5048 | btrfs_ino(inode), to_free, 0); | |
5049 | } | |
5050 | if (delalloc_lock) | |
5051 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5052 | return ret; | |
0ca1f7ce YZ |
5053 | } |
5054 | ||
7709cde3 JB |
5055 | /** |
5056 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5057 | * @inode: the inode to release the reservation for | |
5058 | * @num_bytes: the number of bytes we're releasing | |
5059 | * | |
5060 | * This will release the metadata reservation for an inode. This can be called | |
5061 | * once we complete IO for a given set of bytes to release their metadata | |
5062 | * reservations. | |
5063 | */ | |
0ca1f7ce YZ |
5064 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
5065 | { | |
5066 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
5067 | u64 to_free = 0; |
5068 | unsigned dropped; | |
0ca1f7ce YZ |
5069 | |
5070 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 5071 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 5072 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 5073 | |
0934856d MX |
5074 | if (num_bytes) |
5075 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 5076 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
5077 | if (dropped > 0) |
5078 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 5079 | |
8c2a3ca2 JB |
5080 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5081 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
5082 | if (root->fs_info->quota_enabled) { |
5083 | btrfs_qgroup_free(root, num_bytes + | |
5084 | dropped * root->leafsize); | |
5085 | } | |
5086 | ||
0ca1f7ce YZ |
5087 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5088 | to_free); | |
5089 | } | |
5090 | ||
7709cde3 JB |
5091 | /** |
5092 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
5093 | * @inode: inode we're writing to | |
5094 | * @num_bytes: the number of bytes we want to allocate | |
5095 | * | |
5096 | * This will do the following things | |
5097 | * | |
5098 | * o reserve space in the data space info for num_bytes | |
5099 | * o reserve space in the metadata space info based on number of outstanding | |
5100 | * extents and how much csums will be needed | |
5101 | * o add to the inodes ->delalloc_bytes | |
5102 | * o add it to the fs_info's delalloc inodes list. | |
5103 | * | |
5104 | * This will return 0 for success and -ENOSPC if there is no space left. | |
5105 | */ | |
0ca1f7ce YZ |
5106 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
5107 | { | |
5108 | int ret; | |
5109 | ||
5110 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 5111 | if (ret) |
0ca1f7ce YZ |
5112 | return ret; |
5113 | ||
5114 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
5115 | if (ret) { | |
5116 | btrfs_free_reserved_data_space(inode, num_bytes); | |
5117 | return ret; | |
5118 | } | |
5119 | ||
5120 | return 0; | |
5121 | } | |
5122 | ||
7709cde3 JB |
5123 | /** |
5124 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
5125 | * @inode: inode we're releasing space for | |
5126 | * @num_bytes: the number of bytes we want to free up | |
5127 | * | |
5128 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5129 | * called in the case that we don't need the metadata AND data reservations | |
5130 | * anymore. So if there is an error or we insert an inline extent. | |
5131 | * | |
5132 | * This function will release the metadata space that was not used and will | |
5133 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5134 | * list if there are no delalloc bytes left. | |
5135 | */ | |
0ca1f7ce YZ |
5136 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
5137 | { | |
5138 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
5139 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
5140 | } |
5141 | ||
c53d613e | 5142 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 5143 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 5144 | { |
0af3d00b | 5145 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5146 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5147 | u64 total = num_bytes; |
9078a3e1 | 5148 | u64 old_val; |
db94535d | 5149 | u64 byte_in_group; |
0af3d00b | 5150 | int factor; |
3e1ad54f | 5151 | |
5d4f98a2 | 5152 | /* block accounting for super block */ |
eb73c1b7 | 5153 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5154 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5155 | if (alloc) |
5156 | old_val += num_bytes; | |
5157 | else | |
5158 | old_val -= num_bytes; | |
6c41761f | 5159 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5160 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5161 | |
d397712b | 5162 | while (total) { |
db94535d | 5163 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5164 | if (!cache) |
79787eaa | 5165 | return -ENOENT; |
b742bb82 YZ |
5166 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5167 | BTRFS_BLOCK_GROUP_RAID1 | | |
5168 | BTRFS_BLOCK_GROUP_RAID10)) | |
5169 | factor = 2; | |
5170 | else | |
5171 | factor = 1; | |
9d66e233 JB |
5172 | /* |
5173 | * If this block group has free space cache written out, we | |
5174 | * need to make sure to load it if we are removing space. This | |
5175 | * is because we need the unpinning stage to actually add the | |
5176 | * space back to the block group, otherwise we will leak space. | |
5177 | */ | |
5178 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5179 | cache_block_group(cache, 1); |
0af3d00b | 5180 | |
db94535d CM |
5181 | byte_in_group = bytenr - cache->key.objectid; |
5182 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5183 | |
25179201 | 5184 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5185 | spin_lock(&cache->lock); |
0af3d00b | 5186 | |
73bc1876 | 5187 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5188 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5189 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5190 | ||
0f9dd46c | 5191 | cache->dirty = 1; |
9078a3e1 | 5192 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5193 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5194 | if (alloc) { |
db94535d | 5195 | old_val += num_bytes; |
11833d66 YZ |
5196 | btrfs_set_block_group_used(&cache->item, old_val); |
5197 | cache->reserved -= num_bytes; | |
11833d66 | 5198 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5199 | cache->space_info->bytes_used += num_bytes; |
5200 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5201 | spin_unlock(&cache->lock); |
25179201 | 5202 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5203 | } else { |
db94535d | 5204 | old_val -= num_bytes; |
c286ac48 | 5205 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
5206 | cache->pinned += num_bytes; |
5207 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 5208 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 5209 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 5210 | spin_unlock(&cache->lock); |
25179201 | 5211 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 5212 | |
f0486c68 YZ |
5213 | set_extent_dirty(info->pinned_extents, |
5214 | bytenr, bytenr + num_bytes - 1, | |
5215 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5216 | } |
fa9c0d79 | 5217 | btrfs_put_block_group(cache); |
db94535d CM |
5218 | total -= num_bytes; |
5219 | bytenr += num_bytes; | |
9078a3e1 CM |
5220 | } |
5221 | return 0; | |
5222 | } | |
6324fbf3 | 5223 | |
a061fc8d CM |
5224 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5225 | { | |
0f9dd46c | 5226 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5227 | u64 bytenr; |
0f9dd46c | 5228 | |
a1897fdd LB |
5229 | spin_lock(&root->fs_info->block_group_cache_lock); |
5230 | bytenr = root->fs_info->first_logical_byte; | |
5231 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5232 | ||
5233 | if (bytenr < (u64)-1) | |
5234 | return bytenr; | |
5235 | ||
0f9dd46c JB |
5236 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5237 | if (!cache) | |
a061fc8d | 5238 | return 0; |
0f9dd46c | 5239 | |
d2fb3437 | 5240 | bytenr = cache->key.objectid; |
fa9c0d79 | 5241 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5242 | |
5243 | return bytenr; | |
a061fc8d CM |
5244 | } |
5245 | ||
f0486c68 YZ |
5246 | static int pin_down_extent(struct btrfs_root *root, |
5247 | struct btrfs_block_group_cache *cache, | |
5248 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 5249 | { |
11833d66 YZ |
5250 | spin_lock(&cache->space_info->lock); |
5251 | spin_lock(&cache->lock); | |
5252 | cache->pinned += num_bytes; | |
5253 | cache->space_info->bytes_pinned += num_bytes; | |
5254 | if (reserved) { | |
5255 | cache->reserved -= num_bytes; | |
5256 | cache->space_info->bytes_reserved -= num_bytes; | |
5257 | } | |
5258 | spin_unlock(&cache->lock); | |
5259 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 5260 | |
f0486c68 YZ |
5261 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
5262 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
5263 | return 0; | |
5264 | } | |
68b38550 | 5265 | |
f0486c68 YZ |
5266 | /* |
5267 | * this function must be called within transaction | |
5268 | */ | |
5269 | int btrfs_pin_extent(struct btrfs_root *root, | |
5270 | u64 bytenr, u64 num_bytes, int reserved) | |
5271 | { | |
5272 | struct btrfs_block_group_cache *cache; | |
68b38550 | 5273 | |
f0486c68 | 5274 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 5275 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
5276 | |
5277 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
5278 | ||
5279 | btrfs_put_block_group(cache); | |
11833d66 YZ |
5280 | return 0; |
5281 | } | |
5282 | ||
f0486c68 | 5283 | /* |
e688b725 CM |
5284 | * this function must be called within transaction |
5285 | */ | |
dcfac415 | 5286 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
5287 | u64 bytenr, u64 num_bytes) |
5288 | { | |
5289 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 5290 | int ret; |
e688b725 CM |
5291 | |
5292 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
5293 | if (!cache) |
5294 | return -EINVAL; | |
e688b725 CM |
5295 | |
5296 | /* | |
5297 | * pull in the free space cache (if any) so that our pin | |
5298 | * removes the free space from the cache. We have load_only set | |
5299 | * to one because the slow code to read in the free extents does check | |
5300 | * the pinned extents. | |
5301 | */ | |
f6373bf3 | 5302 | cache_block_group(cache, 1); |
e688b725 CM |
5303 | |
5304 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
5305 | ||
5306 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 5307 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 5308 | btrfs_put_block_group(cache); |
b50c6e25 | 5309 | return ret; |
e688b725 CM |
5310 | } |
5311 | ||
8c2a1a30 JB |
5312 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
5313 | { | |
5314 | int ret; | |
5315 | struct btrfs_block_group_cache *block_group; | |
5316 | struct btrfs_caching_control *caching_ctl; | |
5317 | ||
5318 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
5319 | if (!block_group) | |
5320 | return -EINVAL; | |
5321 | ||
5322 | cache_block_group(block_group, 0); | |
5323 | caching_ctl = get_caching_control(block_group); | |
5324 | ||
5325 | if (!caching_ctl) { | |
5326 | /* Logic error */ | |
5327 | BUG_ON(!block_group_cache_done(block_group)); | |
5328 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5329 | } else { | |
5330 | mutex_lock(&caching_ctl->mutex); | |
5331 | ||
5332 | if (start >= caching_ctl->progress) { | |
5333 | ret = add_excluded_extent(root, start, num_bytes); | |
5334 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5335 | ret = btrfs_remove_free_space(block_group, | |
5336 | start, num_bytes); | |
5337 | } else { | |
5338 | num_bytes = caching_ctl->progress - start; | |
5339 | ret = btrfs_remove_free_space(block_group, | |
5340 | start, num_bytes); | |
5341 | if (ret) | |
5342 | goto out_lock; | |
5343 | ||
5344 | num_bytes = (start + num_bytes) - | |
5345 | caching_ctl->progress; | |
5346 | start = caching_ctl->progress; | |
5347 | ret = add_excluded_extent(root, start, num_bytes); | |
5348 | } | |
5349 | out_lock: | |
5350 | mutex_unlock(&caching_ctl->mutex); | |
5351 | put_caching_control(caching_ctl); | |
5352 | } | |
5353 | btrfs_put_block_group(block_group); | |
5354 | return ret; | |
5355 | } | |
5356 | ||
5357 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
5358 | struct extent_buffer *eb) | |
5359 | { | |
5360 | struct btrfs_file_extent_item *item; | |
5361 | struct btrfs_key key; | |
5362 | int found_type; | |
5363 | int i; | |
5364 | ||
5365 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
5366 | return 0; | |
5367 | ||
5368 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
5369 | btrfs_item_key_to_cpu(eb, &key, i); | |
5370 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
5371 | continue; | |
5372 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
5373 | found_type = btrfs_file_extent_type(eb, item); | |
5374 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
5375 | continue; | |
5376 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
5377 | continue; | |
5378 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
5379 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
5380 | __exclude_logged_extent(log, key.objectid, key.offset); | |
5381 | } | |
5382 | ||
5383 | return 0; | |
5384 | } | |
5385 | ||
fb25e914 JB |
5386 | /** |
5387 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
5388 | * @cache: The cache we are manipulating | |
5389 | * @num_bytes: The number of bytes in question | |
5390 | * @reserve: One of the reservation enums | |
5391 | * | |
5392 | * This is called by the allocator when it reserves space, or by somebody who is | |
5393 | * freeing space that was never actually used on disk. For example if you | |
5394 | * reserve some space for a new leaf in transaction A and before transaction A | |
5395 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
5396 | * clear the reservation. | |
5397 | * | |
5398 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
5399 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
5400 | * delalloc bits in the io_tree. We have to do this since we could end up | |
5401 | * allocating less disk space for the amount of data we have reserved in the | |
5402 | * case of compression. | |
5403 | * | |
5404 | * If this is a reservation and the block group has become read only we cannot | |
5405 | * make the reservation and return -EAGAIN, otherwise this function always | |
5406 | * succeeds. | |
f0486c68 | 5407 | */ |
fb25e914 JB |
5408 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
5409 | u64 num_bytes, int reserve) | |
11833d66 | 5410 | { |
fb25e914 | 5411 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 5412 | int ret = 0; |
79787eaa | 5413 | |
fb25e914 JB |
5414 | spin_lock(&space_info->lock); |
5415 | spin_lock(&cache->lock); | |
5416 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
5417 | if (cache->ro) { |
5418 | ret = -EAGAIN; | |
5419 | } else { | |
fb25e914 JB |
5420 | cache->reserved += num_bytes; |
5421 | space_info->bytes_reserved += num_bytes; | |
5422 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 5423 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
5424 | "space_info", space_info->flags, |
5425 | num_bytes, 0); | |
fb25e914 JB |
5426 | space_info->bytes_may_use -= num_bytes; |
5427 | } | |
f0486c68 | 5428 | } |
fb25e914 JB |
5429 | } else { |
5430 | if (cache->ro) | |
5431 | space_info->bytes_readonly += num_bytes; | |
5432 | cache->reserved -= num_bytes; | |
5433 | space_info->bytes_reserved -= num_bytes; | |
5434 | space_info->reservation_progress++; | |
324ae4df | 5435 | } |
fb25e914 JB |
5436 | spin_unlock(&cache->lock); |
5437 | spin_unlock(&space_info->lock); | |
f0486c68 | 5438 | return ret; |
324ae4df | 5439 | } |
9078a3e1 | 5440 | |
143bede5 | 5441 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 5442 | struct btrfs_root *root) |
e8569813 | 5443 | { |
e8569813 | 5444 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
5445 | struct btrfs_caching_control *next; |
5446 | struct btrfs_caching_control *caching_ctl; | |
5447 | struct btrfs_block_group_cache *cache; | |
b150a4f1 | 5448 | struct btrfs_space_info *space_info; |
e8569813 | 5449 | |
11833d66 | 5450 | down_write(&fs_info->extent_commit_sem); |
25179201 | 5451 | |
11833d66 YZ |
5452 | list_for_each_entry_safe(caching_ctl, next, |
5453 | &fs_info->caching_block_groups, list) { | |
5454 | cache = caching_ctl->block_group; | |
5455 | if (block_group_cache_done(cache)) { | |
5456 | cache->last_byte_to_unpin = (u64)-1; | |
5457 | list_del_init(&caching_ctl->list); | |
5458 | put_caching_control(caching_ctl); | |
e8569813 | 5459 | } else { |
11833d66 | 5460 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5461 | } |
e8569813 | 5462 | } |
11833d66 YZ |
5463 | |
5464 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5465 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5466 | else | |
5467 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5468 | ||
5469 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa | 5470 | |
b150a4f1 JB |
5471 | list_for_each_entry_rcu(space_info, &fs_info->space_info, list) |
5472 | percpu_counter_set(&space_info->total_bytes_pinned, 0); | |
5473 | ||
8929ecfa | 5474 | update_global_block_rsv(fs_info); |
e8569813 ZY |
5475 | } |
5476 | ||
11833d66 | 5477 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5478 | { |
11833d66 YZ |
5479 | struct btrfs_fs_info *fs_info = root->fs_info; |
5480 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5481 | struct btrfs_space_info *space_info; |
5482 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5483 | u64 len; |
7b398f8e | 5484 | bool readonly; |
ccd467d6 | 5485 | |
11833d66 | 5486 | while (start <= end) { |
7b398f8e | 5487 | readonly = false; |
11833d66 YZ |
5488 | if (!cache || |
5489 | start >= cache->key.objectid + cache->key.offset) { | |
5490 | if (cache) | |
5491 | btrfs_put_block_group(cache); | |
5492 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5493 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5494 | } |
5495 | ||
5496 | len = cache->key.objectid + cache->key.offset - start; | |
5497 | len = min(len, end + 1 - start); | |
5498 | ||
5499 | if (start < cache->last_byte_to_unpin) { | |
5500 | len = min(len, cache->last_byte_to_unpin - start); | |
5501 | btrfs_add_free_space(cache, start, len); | |
5502 | } | |
5503 | ||
f0486c68 | 5504 | start += len; |
7b398f8e | 5505 | space_info = cache->space_info; |
f0486c68 | 5506 | |
7b398f8e | 5507 | spin_lock(&space_info->lock); |
11833d66 YZ |
5508 | spin_lock(&cache->lock); |
5509 | cache->pinned -= len; | |
7b398f8e JB |
5510 | space_info->bytes_pinned -= len; |
5511 | if (cache->ro) { | |
5512 | space_info->bytes_readonly += len; | |
5513 | readonly = true; | |
5514 | } | |
11833d66 | 5515 | spin_unlock(&cache->lock); |
7b398f8e JB |
5516 | if (!readonly && global_rsv->space_info == space_info) { |
5517 | spin_lock(&global_rsv->lock); | |
5518 | if (!global_rsv->full) { | |
5519 | len = min(len, global_rsv->size - | |
5520 | global_rsv->reserved); | |
5521 | global_rsv->reserved += len; | |
5522 | space_info->bytes_may_use += len; | |
5523 | if (global_rsv->reserved >= global_rsv->size) | |
5524 | global_rsv->full = 1; | |
5525 | } | |
5526 | spin_unlock(&global_rsv->lock); | |
5527 | } | |
5528 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5529 | } |
11833d66 YZ |
5530 | |
5531 | if (cache) | |
5532 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5533 | return 0; |
5534 | } | |
5535 | ||
5536 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5537 | struct btrfs_root *root) |
a28ec197 | 5538 | { |
11833d66 YZ |
5539 | struct btrfs_fs_info *fs_info = root->fs_info; |
5540 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5541 | u64 start; |
5542 | u64 end; | |
a28ec197 | 5543 | int ret; |
a28ec197 | 5544 | |
79787eaa JM |
5545 | if (trans->aborted) |
5546 | return 0; | |
5547 | ||
11833d66 YZ |
5548 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5549 | unpin = &fs_info->freed_extents[1]; | |
5550 | else | |
5551 | unpin = &fs_info->freed_extents[0]; | |
5552 | ||
d397712b | 5553 | while (1) { |
1a5bc167 | 5554 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5555 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5556 | if (ret) |
a28ec197 | 5557 | break; |
1f3c79a2 | 5558 | |
5378e607 LD |
5559 | if (btrfs_test_opt(root, DISCARD)) |
5560 | ret = btrfs_discard_extent(root, start, | |
5561 | end + 1 - start, NULL); | |
1f3c79a2 | 5562 | |
1a5bc167 | 5563 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5564 | unpin_extent_range(root, start, end); |
b9473439 | 5565 | cond_resched(); |
a28ec197 | 5566 | } |
817d52f8 | 5567 | |
e20d96d6 CM |
5568 | return 0; |
5569 | } | |
5570 | ||
b150a4f1 JB |
5571 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
5572 | u64 owner, u64 root_objectid) | |
5573 | { | |
5574 | struct btrfs_space_info *space_info; | |
5575 | u64 flags; | |
5576 | ||
5577 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
5578 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
5579 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
5580 | else | |
5581 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
5582 | } else { | |
5583 | flags = BTRFS_BLOCK_GROUP_DATA; | |
5584 | } | |
5585 | ||
5586 | space_info = __find_space_info(fs_info, flags); | |
5587 | BUG_ON(!space_info); /* Logic bug */ | |
5588 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
5589 | } | |
5590 | ||
5591 | ||
5d4f98a2 YZ |
5592 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5593 | struct btrfs_root *root, | |
5594 | u64 bytenr, u64 num_bytes, u64 parent, | |
5595 | u64 root_objectid, u64 owner_objectid, | |
5596 | u64 owner_offset, int refs_to_drop, | |
5597 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5598 | { |
e2fa7227 | 5599 | struct btrfs_key key; |
5d4f98a2 | 5600 | struct btrfs_path *path; |
1261ec42 CM |
5601 | struct btrfs_fs_info *info = root->fs_info; |
5602 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5603 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5604 | struct btrfs_extent_item *ei; |
5605 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5606 | int ret; |
5d4f98a2 | 5607 | int is_data; |
952fccac CM |
5608 | int extent_slot = 0; |
5609 | int found_extent = 0; | |
5610 | int num_to_del = 1; | |
5d4f98a2 YZ |
5611 | u32 item_size; |
5612 | u64 refs; | |
3173a18f JB |
5613 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
5614 | SKINNY_METADATA); | |
037e6390 | 5615 | |
5caf2a00 | 5616 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5617 | if (!path) |
5618 | return -ENOMEM; | |
5f26f772 | 5619 | |
3c12ac72 | 5620 | path->reada = 1; |
b9473439 | 5621 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5622 | |
5623 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5624 | BUG_ON(!is_data && refs_to_drop != 1); | |
5625 | ||
3173a18f JB |
5626 | if (is_data) |
5627 | skinny_metadata = 0; | |
5628 | ||
5d4f98a2 YZ |
5629 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
5630 | bytenr, num_bytes, parent, | |
5631 | root_objectid, owner_objectid, | |
5632 | owner_offset); | |
7bb86316 | 5633 | if (ret == 0) { |
952fccac | 5634 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5635 | while (extent_slot >= 0) { |
5636 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5637 | extent_slot); |
5d4f98a2 | 5638 | if (key.objectid != bytenr) |
952fccac | 5639 | break; |
5d4f98a2 YZ |
5640 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5641 | key.offset == num_bytes) { | |
952fccac CM |
5642 | found_extent = 1; |
5643 | break; | |
5644 | } | |
3173a18f JB |
5645 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
5646 | key.offset == owner_objectid) { | |
5647 | found_extent = 1; | |
5648 | break; | |
5649 | } | |
952fccac CM |
5650 | if (path->slots[0] - extent_slot > 5) |
5651 | break; | |
5d4f98a2 | 5652 | extent_slot--; |
952fccac | 5653 | } |
5d4f98a2 YZ |
5654 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5655 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5656 | if (found_extent && item_size < sizeof(*ei)) | |
5657 | found_extent = 0; | |
5658 | #endif | |
31840ae1 | 5659 | if (!found_extent) { |
5d4f98a2 | 5660 | BUG_ON(iref); |
56bec294 | 5661 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5662 | NULL, refs_to_drop, |
5663 | is_data); | |
005d6427 DS |
5664 | if (ret) { |
5665 | btrfs_abort_transaction(trans, extent_root, ret); | |
5666 | goto out; | |
5667 | } | |
b3b4aa74 | 5668 | btrfs_release_path(path); |
b9473439 | 5669 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5670 | |
5671 | key.objectid = bytenr; | |
5672 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5673 | key.offset = num_bytes; | |
5674 | ||
3173a18f JB |
5675 | if (!is_data && skinny_metadata) { |
5676 | key.type = BTRFS_METADATA_ITEM_KEY; | |
5677 | key.offset = owner_objectid; | |
5678 | } | |
5679 | ||
31840ae1 ZY |
5680 | ret = btrfs_search_slot(trans, extent_root, |
5681 | &key, path, -1, 1); | |
3173a18f JB |
5682 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
5683 | /* | |
5684 | * Couldn't find our skinny metadata item, | |
5685 | * see if we have ye olde extent item. | |
5686 | */ | |
5687 | path->slots[0]--; | |
5688 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
5689 | path->slots[0]); | |
5690 | if (key.objectid == bytenr && | |
5691 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
5692 | key.offset == num_bytes) | |
5693 | ret = 0; | |
5694 | } | |
5695 | ||
5696 | if (ret > 0 && skinny_metadata) { | |
5697 | skinny_metadata = false; | |
5698 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5699 | key.offset = num_bytes; | |
5700 | btrfs_release_path(path); | |
5701 | ret = btrfs_search_slot(trans, extent_root, | |
5702 | &key, path, -1, 1); | |
5703 | } | |
5704 | ||
f3465ca4 | 5705 | if (ret) { |
c2cf52eb SK |
5706 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
5707 | ret, (unsigned long long)bytenr); | |
b783e62d JB |
5708 | if (ret > 0) |
5709 | btrfs_print_leaf(extent_root, | |
5710 | path->nodes[0]); | |
f3465ca4 | 5711 | } |
005d6427 DS |
5712 | if (ret < 0) { |
5713 | btrfs_abort_transaction(trans, extent_root, ret); | |
5714 | goto out; | |
5715 | } | |
31840ae1 ZY |
5716 | extent_slot = path->slots[0]; |
5717 | } | |
79787eaa | 5718 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5719 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5720 | WARN_ON(1); | |
c2cf52eb SK |
5721 | btrfs_err(info, |
5722 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
5723 | (unsigned long long)bytenr, | |
5724 | (unsigned long long)parent, | |
5725 | (unsigned long long)root_objectid, | |
5726 | (unsigned long long)owner_objectid, | |
5727 | (unsigned long long)owner_offset); | |
79787eaa | 5728 | } else { |
005d6427 DS |
5729 | btrfs_abort_transaction(trans, extent_root, ret); |
5730 | goto out; | |
7bb86316 | 5731 | } |
5f39d397 CM |
5732 | |
5733 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5734 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5735 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5736 | if (item_size < sizeof(*ei)) { | |
5737 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5738 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5739 | owner_objectid, 0); | |
005d6427 DS |
5740 | if (ret < 0) { |
5741 | btrfs_abort_transaction(trans, extent_root, ret); | |
5742 | goto out; | |
5743 | } | |
5d4f98a2 | 5744 | |
b3b4aa74 | 5745 | btrfs_release_path(path); |
5d4f98a2 YZ |
5746 | path->leave_spinning = 1; |
5747 | ||
5748 | key.objectid = bytenr; | |
5749 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5750 | key.offset = num_bytes; | |
5751 | ||
5752 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5753 | -1, 1); | |
5754 | if (ret) { | |
c2cf52eb SK |
5755 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
5756 | ret, (unsigned long long)bytenr); | |
5d4f98a2 YZ |
5757 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5758 | } | |
005d6427 DS |
5759 | if (ret < 0) { |
5760 | btrfs_abort_transaction(trans, extent_root, ret); | |
5761 | goto out; | |
5762 | } | |
5763 | ||
5d4f98a2 YZ |
5764 | extent_slot = path->slots[0]; |
5765 | leaf = path->nodes[0]; | |
5766 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5767 | } | |
5768 | #endif | |
5769 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5770 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5771 | struct btrfs_extent_item); |
3173a18f JB |
5772 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
5773 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
5774 | struct btrfs_tree_block_info *bi; |
5775 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5776 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5777 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5778 | } | |
56bec294 | 5779 | |
5d4f98a2 | 5780 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
5781 | if (refs < refs_to_drop) { |
5782 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
5783 | "for bytenr %Lu\n", refs_to_drop, refs, bytenr); | |
5784 | ret = -EINVAL; | |
5785 | btrfs_abort_transaction(trans, extent_root, ret); | |
5786 | goto out; | |
5787 | } | |
56bec294 | 5788 | refs -= refs_to_drop; |
5f39d397 | 5789 | |
5d4f98a2 YZ |
5790 | if (refs > 0) { |
5791 | if (extent_op) | |
5792 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5793 | /* | |
5794 | * In the case of inline back ref, reference count will | |
5795 | * be updated by remove_extent_backref | |
952fccac | 5796 | */ |
5d4f98a2 YZ |
5797 | if (iref) { |
5798 | BUG_ON(!found_extent); | |
5799 | } else { | |
5800 | btrfs_set_extent_refs(leaf, ei, refs); | |
5801 | btrfs_mark_buffer_dirty(leaf); | |
5802 | } | |
5803 | if (found_extent) { | |
5804 | ret = remove_extent_backref(trans, extent_root, path, | |
5805 | iref, refs_to_drop, | |
5806 | is_data); | |
005d6427 DS |
5807 | if (ret) { |
5808 | btrfs_abort_transaction(trans, extent_root, ret); | |
5809 | goto out; | |
5810 | } | |
952fccac | 5811 | } |
b150a4f1 JB |
5812 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
5813 | root_objectid); | |
5d4f98a2 | 5814 | } else { |
5d4f98a2 YZ |
5815 | if (found_extent) { |
5816 | BUG_ON(is_data && refs_to_drop != | |
5817 | extent_data_ref_count(root, path, iref)); | |
5818 | if (iref) { | |
5819 | BUG_ON(path->slots[0] != extent_slot); | |
5820 | } else { | |
5821 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5822 | path->slots[0] = extent_slot; | |
5823 | num_to_del = 2; | |
5824 | } | |
78fae27e | 5825 | } |
b9473439 | 5826 | |
952fccac CM |
5827 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5828 | num_to_del); | |
005d6427 DS |
5829 | if (ret) { |
5830 | btrfs_abort_transaction(trans, extent_root, ret); | |
5831 | goto out; | |
5832 | } | |
b3b4aa74 | 5833 | btrfs_release_path(path); |
21af804c | 5834 | |
5d4f98a2 | 5835 | if (is_data) { |
459931ec | 5836 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
5837 | if (ret) { |
5838 | btrfs_abort_transaction(trans, extent_root, ret); | |
5839 | goto out; | |
5840 | } | |
459931ec CM |
5841 | } |
5842 | ||
c53d613e | 5843 | ret = update_block_group(root, bytenr, num_bytes, 0); |
005d6427 DS |
5844 | if (ret) { |
5845 | btrfs_abort_transaction(trans, extent_root, ret); | |
5846 | goto out; | |
5847 | } | |
a28ec197 | 5848 | } |
79787eaa | 5849 | out: |
5caf2a00 | 5850 | btrfs_free_path(path); |
a28ec197 CM |
5851 | return ret; |
5852 | } | |
5853 | ||
1887be66 | 5854 | /* |
f0486c68 | 5855 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5856 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5857 | * a given extent, and if there are no other delayed refs to be processed, it | |
5858 | * removes it from the tree. | |
5859 | */ | |
5860 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5861 | struct btrfs_root *root, u64 bytenr) | |
5862 | { | |
5863 | struct btrfs_delayed_ref_head *head; | |
5864 | struct btrfs_delayed_ref_root *delayed_refs; | |
5865 | struct btrfs_delayed_ref_node *ref; | |
5866 | struct rb_node *node; | |
f0486c68 | 5867 | int ret = 0; |
1887be66 CM |
5868 | |
5869 | delayed_refs = &trans->transaction->delayed_refs; | |
5870 | spin_lock(&delayed_refs->lock); | |
5871 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5872 | if (!head) | |
5873 | goto out; | |
5874 | ||
5875 | node = rb_prev(&head->node.rb_node); | |
5876 | if (!node) | |
5877 | goto out; | |
5878 | ||
5879 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5880 | ||
5881 | /* there are still entries for this ref, we can't drop it */ | |
5882 | if (ref->bytenr == bytenr) | |
5883 | goto out; | |
5884 | ||
5d4f98a2 YZ |
5885 | if (head->extent_op) { |
5886 | if (!head->must_insert_reserved) | |
5887 | goto out; | |
78a6184a | 5888 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
5889 | head->extent_op = NULL; |
5890 | } | |
5891 | ||
1887be66 CM |
5892 | /* |
5893 | * waiting for the lock here would deadlock. If someone else has it | |
5894 | * locked they are already in the process of dropping it anyway | |
5895 | */ | |
5896 | if (!mutex_trylock(&head->mutex)) | |
5897 | goto out; | |
5898 | ||
5899 | /* | |
5900 | * at this point we have a head with no other entries. Go | |
5901 | * ahead and process it. | |
5902 | */ | |
5903 | head->node.in_tree = 0; | |
5904 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5905 | |
1887be66 CM |
5906 | delayed_refs->num_entries--; |
5907 | ||
5908 | /* | |
5909 | * we don't take a ref on the node because we're removing it from the | |
5910 | * tree, so we just steal the ref the tree was holding. | |
5911 | */ | |
c3e69d58 CM |
5912 | delayed_refs->num_heads--; |
5913 | if (list_empty(&head->cluster)) | |
5914 | delayed_refs->num_heads_ready--; | |
5915 | ||
5916 | list_del_init(&head->cluster); | |
1887be66 CM |
5917 | spin_unlock(&delayed_refs->lock); |
5918 | ||
f0486c68 YZ |
5919 | BUG_ON(head->extent_op); |
5920 | if (head->must_insert_reserved) | |
5921 | ret = 1; | |
5922 | ||
5923 | mutex_unlock(&head->mutex); | |
1887be66 | 5924 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5925 | return ret; |
1887be66 CM |
5926 | out: |
5927 | spin_unlock(&delayed_refs->lock); | |
5928 | return 0; | |
5929 | } | |
5930 | ||
f0486c68 YZ |
5931 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5932 | struct btrfs_root *root, | |
5933 | struct extent_buffer *buf, | |
5581a51a | 5934 | u64 parent, int last_ref) |
f0486c68 | 5935 | { |
f0486c68 | 5936 | struct btrfs_block_group_cache *cache = NULL; |
b150a4f1 | 5937 | int pin = 1; |
f0486c68 YZ |
5938 | int ret; |
5939 | ||
5940 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5941 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5942 | buf->start, buf->len, | |
5943 | parent, root->root_key.objectid, | |
5944 | btrfs_header_level(buf), | |
5581a51a | 5945 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5946 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5947 | } |
5948 | ||
5949 | if (!last_ref) | |
5950 | return; | |
5951 | ||
f0486c68 | 5952 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5953 | |
5954 | if (btrfs_header_generation(buf) == trans->transid) { | |
5955 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5956 | ret = check_ref_cleanup(trans, root, buf->start); | |
5957 | if (!ret) | |
37be25bc | 5958 | goto out; |
f0486c68 YZ |
5959 | } |
5960 | ||
5961 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5962 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5963 | goto out; |
f0486c68 YZ |
5964 | } |
5965 | ||
5966 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5967 | ||
5968 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5969 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
b150a4f1 | 5970 | pin = 0; |
f0486c68 YZ |
5971 | } |
5972 | out: | |
b150a4f1 JB |
5973 | if (pin) |
5974 | add_pinned_bytes(root->fs_info, buf->len, | |
5975 | btrfs_header_level(buf), | |
5976 | root->root_key.objectid); | |
5977 | ||
a826d6dc JB |
5978 | /* |
5979 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5980 | * anymore. | |
5981 | */ | |
5982 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5983 | btrfs_put_block_group(cache); |
5984 | } | |
5985 | ||
79787eaa | 5986 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5987 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5988 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5989 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5990 | { |
5991 | int ret; | |
66d7e7f0 | 5992 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5993 | |
b150a4f1 JB |
5994 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
5995 | ||
56bec294 CM |
5996 | /* |
5997 | * tree log blocks never actually go into the extent allocation | |
5998 | * tree, just update pinning info and exit early. | |
56bec294 | 5999 | */ |
5d4f98a2 YZ |
6000 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
6001 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 6002 | /* unlocks the pinned mutex */ |
11833d66 | 6003 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 6004 | ret = 0; |
5d4f98a2 | 6005 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
6006 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
6007 | num_bytes, | |
5d4f98a2 | 6008 | parent, root_objectid, (int)owner, |
66d7e7f0 | 6009 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 6010 | } else { |
66d7e7f0 AJ |
6011 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
6012 | num_bytes, | |
6013 | parent, root_objectid, owner, | |
6014 | offset, BTRFS_DROP_DELAYED_REF, | |
6015 | NULL, for_cow); | |
56bec294 | 6016 | } |
925baedd CM |
6017 | return ret; |
6018 | } | |
6019 | ||
53b381b3 DW |
6020 | static u64 stripe_align(struct btrfs_root *root, |
6021 | struct btrfs_block_group_cache *cache, | |
6022 | u64 val, u64 num_bytes) | |
87ee04eb | 6023 | { |
fda2832f | 6024 | u64 ret = ALIGN(val, root->stripesize); |
87ee04eb CM |
6025 | return ret; |
6026 | } | |
6027 | ||
817d52f8 JB |
6028 | /* |
6029 | * when we wait for progress in the block group caching, its because | |
6030 | * our allocation attempt failed at least once. So, we must sleep | |
6031 | * and let some progress happen before we try again. | |
6032 | * | |
6033 | * This function will sleep at least once waiting for new free space to | |
6034 | * show up, and then it will check the block group free space numbers | |
6035 | * for our min num_bytes. Another option is to have it go ahead | |
6036 | * and look in the rbtree for a free extent of a given size, but this | |
6037 | * is a good start. | |
6038 | */ | |
6039 | static noinline int | |
6040 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
6041 | u64 num_bytes) | |
6042 | { | |
11833d66 | 6043 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 6044 | |
11833d66 YZ |
6045 | caching_ctl = get_caching_control(cache); |
6046 | if (!caching_ctl) | |
817d52f8 | 6047 | return 0; |
817d52f8 | 6048 | |
11833d66 | 6049 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 6050 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
6051 | |
6052 | put_caching_control(caching_ctl); | |
6053 | return 0; | |
6054 | } | |
6055 | ||
6056 | static noinline int | |
6057 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6058 | { | |
6059 | struct btrfs_caching_control *caching_ctl; | |
11833d66 YZ |
6060 | |
6061 | caching_ctl = get_caching_control(cache); | |
6062 | if (!caching_ctl) | |
6063 | return 0; | |
6064 | ||
6065 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
6066 | ||
6067 | put_caching_control(caching_ctl); | |
817d52f8 JB |
6068 | return 0; |
6069 | } | |
6070 | ||
31e50229 | 6071 | int __get_raid_index(u64 flags) |
b742bb82 | 6072 | { |
7738a53a | 6073 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 6074 | return BTRFS_RAID_RAID10; |
7738a53a | 6075 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 6076 | return BTRFS_RAID_RAID1; |
7738a53a | 6077 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 6078 | return BTRFS_RAID_DUP; |
7738a53a | 6079 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 6080 | return BTRFS_RAID_RAID0; |
53b381b3 | 6081 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 6082 | return BTRFS_RAID_RAID5; |
53b381b3 | 6083 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 6084 | return BTRFS_RAID_RAID6; |
7738a53a | 6085 | |
e942f883 | 6086 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
6087 | } |
6088 | ||
7738a53a ID |
6089 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
6090 | { | |
31e50229 | 6091 | return __get_raid_index(cache->flags); |
7738a53a ID |
6092 | } |
6093 | ||
817d52f8 | 6094 | enum btrfs_loop_type { |
285ff5af JB |
6095 | LOOP_CACHING_NOWAIT = 0, |
6096 | LOOP_CACHING_WAIT = 1, | |
6097 | LOOP_ALLOC_CHUNK = 2, | |
6098 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
6099 | }; |
6100 | ||
fec577fb CM |
6101 | /* |
6102 | * walks the btree of allocated extents and find a hole of a given size. | |
6103 | * The key ins is changed to record the hole: | |
6104 | * ins->objectid == block start | |
62e2749e | 6105 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
6106 | * ins->offset == number of blocks |
6107 | * Any available blocks before search_start are skipped. | |
6108 | */ | |
d397712b | 6109 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
6110 | struct btrfs_root *orig_root, |
6111 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 6112 | u64 hint_byte, struct btrfs_key *ins, |
b6919a58 | 6113 | u64 flags) |
fec577fb | 6114 | { |
80eb234a | 6115 | int ret = 0; |
d397712b | 6116 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 6117 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 6118 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 6119 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 6120 | u64 search_start = 0; |
239b14b3 | 6121 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 6122 | struct btrfs_space_info *space_info; |
fa9c0d79 | 6123 | int loop = 0; |
b6919a58 DS |
6124 | int index = __get_raid_index(flags); |
6125 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 6126 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
817d52f8 | 6127 | bool found_uncached_bg = false; |
0a24325e | 6128 | bool failed_cluster_refill = false; |
1cdda9b8 | 6129 | bool failed_alloc = false; |
67377734 | 6130 | bool use_cluster = true; |
60d2adbb | 6131 | bool have_caching_bg = false; |
fec577fb | 6132 | |
db94535d | 6133 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 6134 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
6135 | ins->objectid = 0; |
6136 | ins->offset = 0; | |
b1a4d965 | 6137 | |
b6919a58 | 6138 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 6139 | |
b6919a58 | 6140 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 6141 | if (!space_info) { |
b6919a58 | 6142 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
6143 | return -ENOSPC; |
6144 | } | |
2552d17e | 6145 | |
67377734 JB |
6146 | /* |
6147 | * If the space info is for both data and metadata it means we have a | |
6148 | * small filesystem and we can't use the clustering stuff. | |
6149 | */ | |
6150 | if (btrfs_mixed_space_info(space_info)) | |
6151 | use_cluster = false; | |
6152 | ||
b6919a58 | 6153 | if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 6154 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
6155 | if (!btrfs_test_opt(root, SSD)) |
6156 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
6157 | } |
6158 | ||
b6919a58 | 6159 | if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
67377734 | 6160 | btrfs_test_opt(root, SSD)) { |
fa9c0d79 CM |
6161 | last_ptr = &root->fs_info->data_alloc_cluster; |
6162 | } | |
0f9dd46c | 6163 | |
239b14b3 | 6164 | if (last_ptr) { |
fa9c0d79 CM |
6165 | spin_lock(&last_ptr->lock); |
6166 | if (last_ptr->block_group) | |
6167 | hint_byte = last_ptr->window_start; | |
6168 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 6169 | } |
fa9c0d79 | 6170 | |
a061fc8d | 6171 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 6172 | search_start = max(search_start, hint_byte); |
0b86a832 | 6173 | |
817d52f8 | 6174 | if (!last_ptr) |
fa9c0d79 | 6175 | empty_cluster = 0; |
fa9c0d79 | 6176 | |
2552d17e | 6177 | if (search_start == hint_byte) { |
2552d17e JB |
6178 | block_group = btrfs_lookup_block_group(root->fs_info, |
6179 | search_start); | |
274bd4fb | 6180 | used_block_group = block_group; |
817d52f8 JB |
6181 | /* |
6182 | * we don't want to use the block group if it doesn't match our | |
6183 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
6184 | * |
6185 | * However if we are re-searching with an ideal block group | |
6186 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 6187 | */ |
b6919a58 | 6188 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 6189 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 6190 | down_read(&space_info->groups_sem); |
44fb5511 CM |
6191 | if (list_empty(&block_group->list) || |
6192 | block_group->ro) { | |
6193 | /* | |
6194 | * someone is removing this block group, | |
6195 | * we can't jump into the have_block_group | |
6196 | * target because our list pointers are not | |
6197 | * valid | |
6198 | */ | |
6199 | btrfs_put_block_group(block_group); | |
6200 | up_read(&space_info->groups_sem); | |
ccf0e725 | 6201 | } else { |
b742bb82 | 6202 | index = get_block_group_index(block_group); |
44fb5511 | 6203 | goto have_block_group; |
ccf0e725 | 6204 | } |
2552d17e | 6205 | } else if (block_group) { |
fa9c0d79 | 6206 | btrfs_put_block_group(block_group); |
2552d17e | 6207 | } |
42e70e7a | 6208 | } |
2552d17e | 6209 | search: |
60d2adbb | 6210 | have_caching_bg = false; |
80eb234a | 6211 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
6212 | list_for_each_entry(block_group, &space_info->block_groups[index], |
6213 | list) { | |
6226cb0a | 6214 | u64 offset; |
817d52f8 | 6215 | int cached; |
8a1413a2 | 6216 | |
274bd4fb | 6217 | used_block_group = block_group; |
11dfe35a | 6218 | btrfs_get_block_group(block_group); |
2552d17e | 6219 | search_start = block_group->key.objectid; |
42e70e7a | 6220 | |
83a50de9 CM |
6221 | /* |
6222 | * this can happen if we end up cycling through all the | |
6223 | * raid types, but we want to make sure we only allocate | |
6224 | * for the proper type. | |
6225 | */ | |
b6919a58 | 6226 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
6227 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
6228 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
6229 | BTRFS_BLOCK_GROUP_RAID5 | |
6230 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
6231 | BTRFS_BLOCK_GROUP_RAID10; |
6232 | ||
6233 | /* | |
6234 | * if they asked for extra copies and this block group | |
6235 | * doesn't provide them, bail. This does allow us to | |
6236 | * fill raid0 from raid1. | |
6237 | */ | |
b6919a58 | 6238 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
6239 | goto loop; |
6240 | } | |
6241 | ||
2552d17e | 6242 | have_block_group: |
291c7d2f JB |
6243 | cached = block_group_cache_done(block_group); |
6244 | if (unlikely(!cached)) { | |
291c7d2f | 6245 | found_uncached_bg = true; |
f6373bf3 | 6246 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
6247 | BUG_ON(ret < 0); |
6248 | ret = 0; | |
817d52f8 JB |
6249 | } |
6250 | ||
ea6a478e | 6251 | if (unlikely(block_group->ro)) |
2552d17e | 6252 | goto loop; |
0f9dd46c | 6253 | |
0a24325e | 6254 | /* |
062c05c4 AO |
6255 | * Ok we want to try and use the cluster allocator, so |
6256 | * lets look there | |
0a24325e | 6257 | */ |
062c05c4 | 6258 | if (last_ptr) { |
8de972b4 | 6259 | unsigned long aligned_cluster; |
fa9c0d79 CM |
6260 | /* |
6261 | * the refill lock keeps out other | |
6262 | * people trying to start a new cluster | |
6263 | */ | |
6264 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
6265 | used_block_group = last_ptr->block_group; |
6266 | if (used_block_group != block_group && | |
6267 | (!used_block_group || | |
6268 | used_block_group->ro || | |
b6919a58 | 6269 | !block_group_bits(used_block_group, flags))) { |
274bd4fb | 6270 | used_block_group = block_group; |
44fb5511 | 6271 | goto refill_cluster; |
274bd4fb AO |
6272 | } |
6273 | ||
6274 | if (used_block_group != block_group) | |
6275 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 6276 | |
274bd4fb AO |
6277 | offset = btrfs_alloc_from_cluster(used_block_group, |
6278 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
6279 | if (offset) { |
6280 | /* we have a block, we're done */ | |
6281 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6282 | trace_btrfs_reserve_extent_cluster(root, |
6283 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
6284 | goto checks; |
6285 | } | |
6286 | ||
274bd4fb AO |
6287 | WARN_ON(last_ptr->block_group != used_block_group); |
6288 | if (used_block_group != block_group) { | |
6289 | btrfs_put_block_group(used_block_group); | |
6290 | used_block_group = block_group; | |
fa9c0d79 | 6291 | } |
44fb5511 | 6292 | refill_cluster: |
274bd4fb | 6293 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
6294 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
6295 | * set up a new clusters, so lets just skip it | |
6296 | * and let the allocator find whatever block | |
6297 | * it can find. If we reach this point, we | |
6298 | * will have tried the cluster allocator | |
6299 | * plenty of times and not have found | |
6300 | * anything, so we are likely way too | |
6301 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
6302 | * anything. |
6303 | * | |
6304 | * However, if the cluster is taken from the | |
6305 | * current block group, release the cluster | |
6306 | * first, so that we stand a better chance of | |
6307 | * succeeding in the unclustered | |
6308 | * allocation. */ | |
6309 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
6310 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
6311 | spin_unlock(&last_ptr->refill_lock); |
6312 | goto unclustered_alloc; | |
6313 | } | |
6314 | ||
fa9c0d79 CM |
6315 | /* |
6316 | * this cluster didn't work out, free it and | |
6317 | * start over | |
6318 | */ | |
6319 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
6320 | ||
a5f6f719 AO |
6321 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
6322 | spin_unlock(&last_ptr->refill_lock); | |
6323 | goto unclustered_alloc; | |
6324 | } | |
6325 | ||
8de972b4 CM |
6326 | aligned_cluster = max_t(unsigned long, |
6327 | empty_cluster + empty_size, | |
6328 | block_group->full_stripe_len); | |
6329 | ||
fa9c0d79 | 6330 | /* allocate a cluster in this block group */ |
451d7585 | 6331 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 6332 | block_group, last_ptr, |
1b22bad7 | 6333 | search_start, num_bytes, |
8de972b4 | 6334 | aligned_cluster); |
fa9c0d79 CM |
6335 | if (ret == 0) { |
6336 | /* | |
6337 | * now pull our allocation out of this | |
6338 | * cluster | |
6339 | */ | |
6340 | offset = btrfs_alloc_from_cluster(block_group, | |
6341 | last_ptr, num_bytes, | |
6342 | search_start); | |
6343 | if (offset) { | |
6344 | /* we found one, proceed */ | |
6345 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6346 | trace_btrfs_reserve_extent_cluster(root, |
6347 | block_group, search_start, | |
6348 | num_bytes); | |
fa9c0d79 CM |
6349 | goto checks; |
6350 | } | |
0a24325e JB |
6351 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
6352 | && !failed_cluster_refill) { | |
817d52f8 JB |
6353 | spin_unlock(&last_ptr->refill_lock); |
6354 | ||
0a24325e | 6355 | failed_cluster_refill = true; |
817d52f8 JB |
6356 | wait_block_group_cache_progress(block_group, |
6357 | num_bytes + empty_cluster + empty_size); | |
6358 | goto have_block_group; | |
fa9c0d79 | 6359 | } |
817d52f8 | 6360 | |
fa9c0d79 CM |
6361 | /* |
6362 | * at this point we either didn't find a cluster | |
6363 | * or we weren't able to allocate a block from our | |
6364 | * cluster. Free the cluster we've been trying | |
6365 | * to use, and go to the next block group | |
6366 | */ | |
0a24325e | 6367 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 6368 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 6369 | goto loop; |
fa9c0d79 CM |
6370 | } |
6371 | ||
062c05c4 | 6372 | unclustered_alloc: |
a5f6f719 AO |
6373 | spin_lock(&block_group->free_space_ctl->tree_lock); |
6374 | if (cached && | |
6375 | block_group->free_space_ctl->free_space < | |
6376 | num_bytes + empty_cluster + empty_size) { | |
6377 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6378 | goto loop; | |
6379 | } | |
6380 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6381 | ||
6226cb0a JB |
6382 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
6383 | num_bytes, empty_size); | |
1cdda9b8 JB |
6384 | /* |
6385 | * If we didn't find a chunk, and we haven't failed on this | |
6386 | * block group before, and this block group is in the middle of | |
6387 | * caching and we are ok with waiting, then go ahead and wait | |
6388 | * for progress to be made, and set failed_alloc to true. | |
6389 | * | |
6390 | * If failed_alloc is true then we've already waited on this | |
6391 | * block group once and should move on to the next block group. | |
6392 | */ | |
6393 | if (!offset && !failed_alloc && !cached && | |
6394 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 6395 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
6396 | num_bytes + empty_size); |
6397 | failed_alloc = true; | |
817d52f8 | 6398 | goto have_block_group; |
1cdda9b8 | 6399 | } else if (!offset) { |
60d2adbb MX |
6400 | if (!cached) |
6401 | have_caching_bg = true; | |
1cdda9b8 | 6402 | goto loop; |
817d52f8 | 6403 | } |
fa9c0d79 | 6404 | checks: |
53b381b3 DW |
6405 | search_start = stripe_align(root, used_block_group, |
6406 | offset, num_bytes); | |
25179201 | 6407 | |
2552d17e JB |
6408 | /* move on to the next group */ |
6409 | if (search_start + num_bytes > | |
274bd4fb AO |
6410 | used_block_group->key.objectid + used_block_group->key.offset) { |
6411 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 6412 | goto loop; |
6226cb0a | 6413 | } |
f5a31e16 | 6414 | |
f0486c68 | 6415 | if (offset < search_start) |
274bd4fb | 6416 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
6417 | search_start - offset); |
6418 | BUG_ON(offset > search_start); | |
2552d17e | 6419 | |
274bd4fb | 6420 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 6421 | alloc_type); |
f0486c68 | 6422 | if (ret == -EAGAIN) { |
274bd4fb | 6423 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 6424 | goto loop; |
0f9dd46c | 6425 | } |
0b86a832 | 6426 | |
f0486c68 | 6427 | /* we are all good, lets return */ |
2552d17e JB |
6428 | ins->objectid = search_start; |
6429 | ins->offset = num_bytes; | |
d2fb3437 | 6430 | |
3f7de037 JB |
6431 | trace_btrfs_reserve_extent(orig_root, block_group, |
6432 | search_start, num_bytes); | |
274bd4fb AO |
6433 | if (used_block_group != block_group) |
6434 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 6435 | btrfs_put_block_group(block_group); |
2552d17e JB |
6436 | break; |
6437 | loop: | |
0a24325e | 6438 | failed_cluster_refill = false; |
1cdda9b8 | 6439 | failed_alloc = false; |
b742bb82 | 6440 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
6441 | if (used_block_group != block_group) |
6442 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 6443 | btrfs_put_block_group(block_group); |
2552d17e JB |
6444 | } |
6445 | up_read(&space_info->groups_sem); | |
6446 | ||
60d2adbb MX |
6447 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
6448 | goto search; | |
6449 | ||
b742bb82 YZ |
6450 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
6451 | goto search; | |
6452 | ||
285ff5af | 6453 | /* |
ccf0e725 JB |
6454 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
6455 | * caching kthreads as we move along | |
817d52f8 JB |
6456 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
6457 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
6458 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
6459 | * again | |
fa9c0d79 | 6460 | */ |
723bda20 | 6461 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 6462 | index = 0; |
723bda20 | 6463 | loop++; |
817d52f8 | 6464 | if (loop == LOOP_ALLOC_CHUNK) { |
b6919a58 | 6465 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad JB |
6466 | CHUNK_ALLOC_FORCE); |
6467 | /* | |
6468 | * Do not bail out on ENOSPC since we | |
6469 | * can do more things. | |
6470 | */ | |
6471 | if (ret < 0 && ret != -ENOSPC) { | |
6472 | btrfs_abort_transaction(trans, | |
6473 | root, ret); | |
6474 | goto out; | |
723bda20 | 6475 | } |
2552d17e JB |
6476 | } |
6477 | ||
723bda20 JB |
6478 | if (loop == LOOP_NO_EMPTY_SIZE) { |
6479 | empty_size = 0; | |
6480 | empty_cluster = 0; | |
fa9c0d79 | 6481 | } |
723bda20 JB |
6482 | |
6483 | goto search; | |
2552d17e JB |
6484 | } else if (!ins->objectid) { |
6485 | ret = -ENOSPC; | |
d82a6f1d | 6486 | } else if (ins->objectid) { |
80eb234a | 6487 | ret = 0; |
be744175 | 6488 | } |
79787eaa | 6489 | out: |
be744175 | 6490 | |
0f70abe2 | 6491 | return ret; |
fec577fb | 6492 | } |
ec44a35c | 6493 | |
9ed74f2d JB |
6494 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
6495 | int dump_block_groups) | |
0f9dd46c JB |
6496 | { |
6497 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 6498 | int index = 0; |
0f9dd46c | 6499 | |
9ed74f2d | 6500 | spin_lock(&info->lock); |
fb25e914 JB |
6501 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
6502 | (unsigned long long)info->flags, | |
d397712b | 6503 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 6504 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 6505 | info->bytes_readonly), |
d397712b | 6506 | (info->full) ? "" : "not "); |
8929ecfa YZ |
6507 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
6508 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 6509 | (unsigned long long)info->total_bytes, |
8929ecfa | 6510 | (unsigned long long)info->bytes_used, |
21380931 | 6511 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 6512 | (unsigned long long)info->bytes_reserved, |
21380931 | 6513 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 6514 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
6515 | spin_unlock(&info->lock); |
6516 | ||
6517 | if (!dump_block_groups) | |
6518 | return; | |
0f9dd46c | 6519 | |
80eb234a | 6520 | down_read(&info->groups_sem); |
b742bb82 YZ |
6521 | again: |
6522 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 6523 | spin_lock(&cache->lock); |
799ffc3c | 6524 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
6525 | (unsigned long long)cache->key.objectid, |
6526 | (unsigned long long)cache->key.offset, | |
6527 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
6528 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
6529 | (unsigned long long)cache->reserved, |
6530 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
6531 | btrfs_dump_free_space(cache, bytes); |
6532 | spin_unlock(&cache->lock); | |
6533 | } | |
b742bb82 YZ |
6534 | if (++index < BTRFS_NR_RAID_TYPES) |
6535 | goto again; | |
80eb234a | 6536 | up_read(&info->groups_sem); |
0f9dd46c | 6537 | } |
e8569813 | 6538 | |
11833d66 YZ |
6539 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
6540 | struct btrfs_root *root, | |
6541 | u64 num_bytes, u64 min_alloc_size, | |
6542 | u64 empty_size, u64 hint_byte, | |
b6919a58 | 6543 | struct btrfs_key *ins, int is_data) |
fec577fb | 6544 | { |
9e622d6b | 6545 | bool final_tried = false; |
b6919a58 | 6546 | u64 flags; |
fec577fb | 6547 | int ret; |
925baedd | 6548 | |
b6919a58 | 6549 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 6550 | again: |
db94535d CM |
6551 | WARN_ON(num_bytes < root->sectorsize); |
6552 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
b6919a58 | 6553 | hint_byte, ins, flags); |
3b951516 | 6554 | |
9e622d6b MX |
6555 | if (ret == -ENOSPC) { |
6556 | if (!final_tried) { | |
6557 | num_bytes = num_bytes >> 1; | |
24542bf7 | 6558 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 6559 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
6560 | if (num_bytes == min_alloc_size) |
6561 | final_tried = true; | |
6562 | goto again; | |
6563 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6564 | struct btrfs_space_info *sinfo; | |
6565 | ||
b6919a58 | 6566 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 6567 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
b6919a58 | 6568 | (unsigned long long)flags, |
c2cf52eb | 6569 | (unsigned long long)num_bytes); |
53804280 JM |
6570 | if (sinfo) |
6571 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6572 | } |
925baedd | 6573 | } |
0f9dd46c | 6574 | |
1abe9b8a | 6575 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
6576 | ||
0f9dd46c | 6577 | return ret; |
e6dcd2dc CM |
6578 | } |
6579 | ||
e688b725 CM |
6580 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6581 | u64 start, u64 len, int pin) | |
65b51a00 | 6582 | { |
0f9dd46c | 6583 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6584 | int ret = 0; |
0f9dd46c | 6585 | |
0f9dd46c JB |
6586 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6587 | if (!cache) { | |
c2cf52eb SK |
6588 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
6589 | (unsigned long long)start); | |
0f9dd46c JB |
6590 | return -ENOSPC; |
6591 | } | |
1f3c79a2 | 6592 | |
5378e607 LD |
6593 | if (btrfs_test_opt(root, DISCARD)) |
6594 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6595 | |
e688b725 CM |
6596 | if (pin) |
6597 | pin_down_extent(root, cache, start, len, 1); | |
6598 | else { | |
6599 | btrfs_add_free_space(cache, start, len); | |
6600 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6601 | } | |
fa9c0d79 | 6602 | btrfs_put_block_group(cache); |
817d52f8 | 6603 | |
1abe9b8a | 6604 | trace_btrfs_reserved_extent_free(root, start, len); |
6605 | ||
e6dcd2dc CM |
6606 | return ret; |
6607 | } | |
6608 | ||
e688b725 CM |
6609 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6610 | u64 start, u64 len) | |
6611 | { | |
6612 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6613 | } | |
6614 | ||
6615 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6616 | u64 start, u64 len) | |
6617 | { | |
6618 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6619 | } | |
6620 | ||
5d4f98a2 YZ |
6621 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6622 | struct btrfs_root *root, | |
6623 | u64 parent, u64 root_objectid, | |
6624 | u64 flags, u64 owner, u64 offset, | |
6625 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6626 | { |
6627 | int ret; | |
5d4f98a2 | 6628 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6629 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6630 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6631 | struct btrfs_path *path; |
5d4f98a2 YZ |
6632 | struct extent_buffer *leaf; |
6633 | int type; | |
6634 | u32 size; | |
26b8003f | 6635 | |
5d4f98a2 YZ |
6636 | if (parent > 0) |
6637 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6638 | else | |
6639 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6640 | |
5d4f98a2 | 6641 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6642 | |
6643 | path = btrfs_alloc_path(); | |
db5b493a TI |
6644 | if (!path) |
6645 | return -ENOMEM; | |
47e4bb98 | 6646 | |
b9473439 | 6647 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6648 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6649 | ins, size); | |
79787eaa JM |
6650 | if (ret) { |
6651 | btrfs_free_path(path); | |
6652 | return ret; | |
6653 | } | |
0f9dd46c | 6654 | |
5d4f98a2 YZ |
6655 | leaf = path->nodes[0]; |
6656 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6657 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6658 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6659 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6660 | btrfs_set_extent_flags(leaf, extent_item, | |
6661 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6662 | ||
6663 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6664 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6665 | if (parent > 0) { | |
6666 | struct btrfs_shared_data_ref *ref; | |
6667 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6668 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6669 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6670 | } else { | |
6671 | struct btrfs_extent_data_ref *ref; | |
6672 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6673 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6674 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6675 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6676 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6677 | } | |
47e4bb98 CM |
6678 | |
6679 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6680 | btrfs_free_path(path); |
f510cfec | 6681 | |
c53d613e | 6682 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 6683 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb SK |
6684 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
6685 | (unsigned long long)ins->objectid, | |
6686 | (unsigned long long)ins->offset); | |
f5947066 CM |
6687 | BUG(); |
6688 | } | |
e6dcd2dc CM |
6689 | return ret; |
6690 | } | |
6691 | ||
5d4f98a2 YZ |
6692 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6693 | struct btrfs_root *root, | |
6694 | u64 parent, u64 root_objectid, | |
6695 | u64 flags, struct btrfs_disk_key *key, | |
6696 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6697 | { |
6698 | int ret; | |
5d4f98a2 YZ |
6699 | struct btrfs_fs_info *fs_info = root->fs_info; |
6700 | struct btrfs_extent_item *extent_item; | |
6701 | struct btrfs_tree_block_info *block_info; | |
6702 | struct btrfs_extent_inline_ref *iref; | |
6703 | struct btrfs_path *path; | |
6704 | struct extent_buffer *leaf; | |
3173a18f JB |
6705 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
6706 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
6707 | SKINNY_METADATA); | |
6708 | ||
6709 | if (!skinny_metadata) | |
6710 | size += sizeof(*block_info); | |
1c2308f8 | 6711 | |
5d4f98a2 | 6712 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6713 | if (!path) |
6714 | return -ENOMEM; | |
56bec294 | 6715 | |
5d4f98a2 YZ |
6716 | path->leave_spinning = 1; |
6717 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6718 | ins, size); | |
79787eaa JM |
6719 | if (ret) { |
6720 | btrfs_free_path(path); | |
6721 | return ret; | |
6722 | } | |
5d4f98a2 YZ |
6723 | |
6724 | leaf = path->nodes[0]; | |
6725 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6726 | struct btrfs_extent_item); | |
6727 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6728 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6729 | btrfs_set_extent_flags(leaf, extent_item, | |
6730 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 6731 | |
3173a18f JB |
6732 | if (skinny_metadata) { |
6733 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6734 | } else { | |
6735 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6736 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6737 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6738 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6739 | } | |
5d4f98a2 | 6740 | |
5d4f98a2 YZ |
6741 | if (parent > 0) { |
6742 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6743 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6744 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6745 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6746 | } else { | |
6747 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6748 | BTRFS_TREE_BLOCK_REF_KEY); | |
6749 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6750 | } | |
6751 | ||
6752 | btrfs_mark_buffer_dirty(leaf); | |
6753 | btrfs_free_path(path); | |
6754 | ||
3173a18f | 6755 | ret = update_block_group(root, ins->objectid, root->leafsize, 1); |
79787eaa | 6756 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb SK |
6757 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
6758 | (unsigned long long)ins->objectid, | |
6759 | (unsigned long long)ins->offset); | |
5d4f98a2 YZ |
6760 | BUG(); |
6761 | } | |
6762 | return ret; | |
6763 | } | |
6764 | ||
6765 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6766 | struct btrfs_root *root, | |
6767 | u64 root_objectid, u64 owner, | |
6768 | u64 offset, struct btrfs_key *ins) | |
6769 | { | |
6770 | int ret; | |
6771 | ||
6772 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6773 | ||
66d7e7f0 AJ |
6774 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6775 | ins->offset, 0, | |
6776 | root_objectid, owner, offset, | |
6777 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6778 | return ret; |
6779 | } | |
e02119d5 CM |
6780 | |
6781 | /* | |
6782 | * this is used by the tree logging recovery code. It records that | |
6783 | * an extent has been allocated and makes sure to clear the free | |
6784 | * space cache bits as well | |
6785 | */ | |
5d4f98a2 YZ |
6786 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6787 | struct btrfs_root *root, | |
6788 | u64 root_objectid, u64 owner, u64 offset, | |
6789 | struct btrfs_key *ins) | |
e02119d5 CM |
6790 | { |
6791 | int ret; | |
6792 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 6793 | |
8c2a1a30 JB |
6794 | /* |
6795 | * Mixed block groups will exclude before processing the log so we only | |
6796 | * need to do the exlude dance if this fs isn't mixed. | |
6797 | */ | |
6798 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
6799 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 6800 | if (ret) |
8c2a1a30 | 6801 | return ret; |
11833d66 YZ |
6802 | } |
6803 | ||
8c2a1a30 JB |
6804 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
6805 | if (!block_group) | |
6806 | return -EINVAL; | |
6807 | ||
fb25e914 JB |
6808 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6809 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6810 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
6811 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6812 | 0, owner, offset, ins, 1); | |
b50c6e25 | 6813 | btrfs_put_block_group(block_group); |
e02119d5 CM |
6814 | return ret; |
6815 | } | |
6816 | ||
48a3b636 ES |
6817 | static struct extent_buffer * |
6818 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
6819 | u64 bytenr, u32 blocksize, int level) | |
65b51a00 CM |
6820 | { |
6821 | struct extent_buffer *buf; | |
6822 | ||
6823 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6824 | if (!buf) | |
6825 | return ERR_PTR(-ENOMEM); | |
6826 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6827 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6828 | btrfs_tree_lock(buf); |
6829 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6830 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6831 | |
6832 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6833 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6834 | |
d0c803c4 | 6835 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6836 | /* |
6837 | * we allow two log transactions at a time, use different | |
6838 | * EXENT bit to differentiate dirty pages. | |
6839 | */ | |
6840 | if (root->log_transid % 2 == 0) | |
6841 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6842 | buf->start + buf->len - 1, GFP_NOFS); | |
6843 | else | |
6844 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6845 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6846 | } else { |
6847 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6848 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6849 | } |
65b51a00 | 6850 | trans->blocks_used++; |
b4ce94de | 6851 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6852 | return buf; |
6853 | } | |
6854 | ||
f0486c68 YZ |
6855 | static struct btrfs_block_rsv * |
6856 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6857 | struct btrfs_root *root, u32 blocksize) | |
6858 | { | |
6859 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6860 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 6861 | int ret; |
d88033db | 6862 | bool global_updated = false; |
f0486c68 YZ |
6863 | |
6864 | block_rsv = get_block_rsv(trans, root); | |
6865 | ||
b586b323 MX |
6866 | if (unlikely(block_rsv->size == 0)) |
6867 | goto try_reserve; | |
d88033db | 6868 | again: |
f0486c68 YZ |
6869 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
6870 | if (!ret) | |
6871 | return block_rsv; | |
6872 | ||
b586b323 MX |
6873 | if (block_rsv->failfast) |
6874 | return ERR_PTR(ret); | |
6875 | ||
d88033db MX |
6876 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
6877 | global_updated = true; | |
6878 | update_global_block_rsv(root->fs_info); | |
6879 | goto again; | |
6880 | } | |
6881 | ||
b586b323 MX |
6882 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
6883 | static DEFINE_RATELIMIT_STATE(_rs, | |
6884 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
6885 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
6886 | if (__ratelimit(&_rs)) | |
6887 | WARN(1, KERN_DEBUG | |
6888 | "btrfs: block rsv returned %d\n", ret); | |
6889 | } | |
6890 | try_reserve: | |
6891 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
6892 | BTRFS_RESERVE_NO_FLUSH); | |
6893 | if (!ret) | |
6894 | return block_rsv; | |
6895 | /* | |
6896 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
6897 | * the global reserve if its space type is the same as the global |
6898 | * reservation. | |
b586b323 | 6899 | */ |
5881cfc9 MX |
6900 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
6901 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
6902 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
6903 | if (!ret) | |
6904 | return global_rsv; | |
6905 | } | |
6906 | return ERR_PTR(ret); | |
f0486c68 YZ |
6907 | } |
6908 | ||
8c2a3ca2 JB |
6909 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6910 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6911 | { |
6912 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6913 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6914 | } |
6915 | ||
fec577fb | 6916 | /* |
f0486c68 YZ |
6917 | * finds a free extent and does all the dirty work required for allocation |
6918 | * returns the key for the extent through ins, and a tree buffer for | |
6919 | * the first block of the extent through buf. | |
6920 | * | |
fec577fb CM |
6921 | * returns the tree buffer or NULL. |
6922 | */ | |
5f39d397 | 6923 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6924 | struct btrfs_root *root, u32 blocksize, |
6925 | u64 parent, u64 root_objectid, | |
6926 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6927 | u64 hint, u64 empty_size) |
fec577fb | 6928 | { |
e2fa7227 | 6929 | struct btrfs_key ins; |
f0486c68 | 6930 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6931 | struct extent_buffer *buf; |
f0486c68 YZ |
6932 | u64 flags = 0; |
6933 | int ret; | |
3173a18f JB |
6934 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6935 | SKINNY_METADATA); | |
fec577fb | 6936 | |
f0486c68 YZ |
6937 | block_rsv = use_block_rsv(trans, root, blocksize); |
6938 | if (IS_ERR(block_rsv)) | |
6939 | return ERR_CAST(block_rsv); | |
6940 | ||
6941 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6942 | empty_size, hint, &ins, 0); |
fec577fb | 6943 | if (ret) { |
8c2a3ca2 | 6944 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6945 | return ERR_PTR(ret); |
fec577fb | 6946 | } |
55c69072 | 6947 | |
4008c04a CM |
6948 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6949 | blocksize, level); | |
79787eaa | 6950 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6951 | |
6952 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6953 | if (parent == 0) | |
6954 | parent = ins.objectid; | |
6955 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6956 | } else | |
6957 | BUG_ON(parent > 0); | |
6958 | ||
6959 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6960 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 6961 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 6962 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6963 | if (key) |
6964 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6965 | else | |
6966 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6967 | extent_op->flags_to_set = flags; | |
3173a18f JB |
6968 | if (skinny_metadata) |
6969 | extent_op->update_key = 0; | |
6970 | else | |
6971 | extent_op->update_key = 1; | |
f0486c68 YZ |
6972 | extent_op->update_flags = 1; |
6973 | extent_op->is_data = 0; | |
b1c79e09 | 6974 | extent_op->level = level; |
f0486c68 | 6975 | |
66d7e7f0 AJ |
6976 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6977 | ins.objectid, | |
f0486c68 YZ |
6978 | ins.offset, parent, root_objectid, |
6979 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6980 | extent_op, 0); |
79787eaa | 6981 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6982 | } |
fec577fb CM |
6983 | return buf; |
6984 | } | |
a28ec197 | 6985 | |
2c47e605 YZ |
6986 | struct walk_control { |
6987 | u64 refs[BTRFS_MAX_LEVEL]; | |
6988 | u64 flags[BTRFS_MAX_LEVEL]; | |
6989 | struct btrfs_key update_progress; | |
6990 | int stage; | |
6991 | int level; | |
6992 | int shared_level; | |
6993 | int update_ref; | |
6994 | int keep_locks; | |
1c4850e2 YZ |
6995 | int reada_slot; |
6996 | int reada_count; | |
66d7e7f0 | 6997 | int for_reloc; |
2c47e605 YZ |
6998 | }; |
6999 | ||
7000 | #define DROP_REFERENCE 1 | |
7001 | #define UPDATE_BACKREF 2 | |
7002 | ||
1c4850e2 YZ |
7003 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
7004 | struct btrfs_root *root, | |
7005 | struct walk_control *wc, | |
7006 | struct btrfs_path *path) | |
6407bf6d | 7007 | { |
1c4850e2 YZ |
7008 | u64 bytenr; |
7009 | u64 generation; | |
7010 | u64 refs; | |
94fcca9f | 7011 | u64 flags; |
5d4f98a2 | 7012 | u32 nritems; |
1c4850e2 YZ |
7013 | u32 blocksize; |
7014 | struct btrfs_key key; | |
7015 | struct extent_buffer *eb; | |
6407bf6d | 7016 | int ret; |
1c4850e2 YZ |
7017 | int slot; |
7018 | int nread = 0; | |
6407bf6d | 7019 | |
1c4850e2 YZ |
7020 | if (path->slots[wc->level] < wc->reada_slot) { |
7021 | wc->reada_count = wc->reada_count * 2 / 3; | |
7022 | wc->reada_count = max(wc->reada_count, 2); | |
7023 | } else { | |
7024 | wc->reada_count = wc->reada_count * 3 / 2; | |
7025 | wc->reada_count = min_t(int, wc->reada_count, | |
7026 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
7027 | } | |
7bb86316 | 7028 | |
1c4850e2 YZ |
7029 | eb = path->nodes[wc->level]; |
7030 | nritems = btrfs_header_nritems(eb); | |
7031 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 7032 | |
1c4850e2 YZ |
7033 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
7034 | if (nread >= wc->reada_count) | |
7035 | break; | |
bd56b302 | 7036 | |
2dd3e67b | 7037 | cond_resched(); |
1c4850e2 YZ |
7038 | bytenr = btrfs_node_blockptr(eb, slot); |
7039 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 7040 | |
1c4850e2 YZ |
7041 | if (slot == path->slots[wc->level]) |
7042 | goto reada; | |
5d4f98a2 | 7043 | |
1c4850e2 YZ |
7044 | if (wc->stage == UPDATE_BACKREF && |
7045 | generation <= root->root_key.offset) | |
bd56b302 CM |
7046 | continue; |
7047 | ||
94fcca9f | 7048 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
7049 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
7050 | wc->level - 1, 1, &refs, | |
7051 | &flags); | |
79787eaa JM |
7052 | /* We don't care about errors in readahead. */ |
7053 | if (ret < 0) | |
7054 | continue; | |
94fcca9f YZ |
7055 | BUG_ON(refs == 0); |
7056 | ||
1c4850e2 | 7057 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
7058 | if (refs == 1) |
7059 | goto reada; | |
bd56b302 | 7060 | |
94fcca9f YZ |
7061 | if (wc->level == 1 && |
7062 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7063 | continue; | |
1c4850e2 YZ |
7064 | if (!wc->update_ref || |
7065 | generation <= root->root_key.offset) | |
7066 | continue; | |
7067 | btrfs_node_key_to_cpu(eb, &key, slot); | |
7068 | ret = btrfs_comp_cpu_keys(&key, | |
7069 | &wc->update_progress); | |
7070 | if (ret < 0) | |
7071 | continue; | |
94fcca9f YZ |
7072 | } else { |
7073 | if (wc->level == 1 && | |
7074 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7075 | continue; | |
6407bf6d | 7076 | } |
1c4850e2 YZ |
7077 | reada: |
7078 | ret = readahead_tree_block(root, bytenr, blocksize, | |
7079 | generation); | |
7080 | if (ret) | |
bd56b302 | 7081 | break; |
1c4850e2 | 7082 | nread++; |
20524f02 | 7083 | } |
1c4850e2 | 7084 | wc->reada_slot = slot; |
20524f02 | 7085 | } |
2c47e605 | 7086 | |
f82d02d9 | 7087 | /* |
2c016dc2 | 7088 | * helper to process tree block while walking down the tree. |
2c47e605 | 7089 | * |
2c47e605 YZ |
7090 | * when wc->stage == UPDATE_BACKREF, this function updates |
7091 | * back refs for pointers in the block. | |
7092 | * | |
7093 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 7094 | */ |
2c47e605 | 7095 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 7096 | struct btrfs_root *root, |
2c47e605 | 7097 | struct btrfs_path *path, |
94fcca9f | 7098 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 7099 | { |
2c47e605 YZ |
7100 | int level = wc->level; |
7101 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 7102 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
7103 | int ret; |
7104 | ||
2c47e605 YZ |
7105 | if (wc->stage == UPDATE_BACKREF && |
7106 | btrfs_header_owner(eb) != root->root_key.objectid) | |
7107 | return 1; | |
f82d02d9 | 7108 | |
2c47e605 YZ |
7109 | /* |
7110 | * when reference count of tree block is 1, it won't increase | |
7111 | * again. once full backref flag is set, we never clear it. | |
7112 | */ | |
94fcca9f YZ |
7113 | if (lookup_info && |
7114 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
7115 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
7116 | BUG_ON(!path->locks[level]); |
7117 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7118 | eb->start, level, 1, |
2c47e605 YZ |
7119 | &wc->refs[level], |
7120 | &wc->flags[level]); | |
79787eaa JM |
7121 | BUG_ON(ret == -ENOMEM); |
7122 | if (ret) | |
7123 | return ret; | |
2c47e605 YZ |
7124 | BUG_ON(wc->refs[level] == 0); |
7125 | } | |
5d4f98a2 | 7126 | |
2c47e605 YZ |
7127 | if (wc->stage == DROP_REFERENCE) { |
7128 | if (wc->refs[level] > 1) | |
7129 | return 1; | |
f82d02d9 | 7130 | |
2c47e605 | 7131 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 7132 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7133 | path->locks[level] = 0; |
7134 | } | |
7135 | return 0; | |
7136 | } | |
f82d02d9 | 7137 | |
2c47e605 YZ |
7138 | /* wc->stage == UPDATE_BACKREF */ |
7139 | if (!(wc->flags[level] & flag)) { | |
7140 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 7141 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 7142 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 7143 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 7144 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 7145 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
7146 | eb->len, flag, |
7147 | btrfs_header_level(eb), 0); | |
79787eaa | 7148 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7149 | wc->flags[level] |= flag; |
7150 | } | |
7151 | ||
7152 | /* | |
7153 | * the block is shared by multiple trees, so it's not good to | |
7154 | * keep the tree lock | |
7155 | */ | |
7156 | if (path->locks[level] && level > 0) { | |
bd681513 | 7157 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7158 | path->locks[level] = 0; |
7159 | } | |
7160 | return 0; | |
7161 | } | |
7162 | ||
1c4850e2 | 7163 | /* |
2c016dc2 | 7164 | * helper to process tree block pointer. |
1c4850e2 YZ |
7165 | * |
7166 | * when wc->stage == DROP_REFERENCE, this function checks | |
7167 | * reference count of the block pointed to. if the block | |
7168 | * is shared and we need update back refs for the subtree | |
7169 | * rooted at the block, this function changes wc->stage to | |
7170 | * UPDATE_BACKREF. if the block is shared and there is no | |
7171 | * need to update back, this function drops the reference | |
7172 | * to the block. | |
7173 | * | |
7174 | * NOTE: return value 1 means we should stop walking down. | |
7175 | */ | |
7176 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
7177 | struct btrfs_root *root, | |
7178 | struct btrfs_path *path, | |
94fcca9f | 7179 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
7180 | { |
7181 | u64 bytenr; | |
7182 | u64 generation; | |
7183 | u64 parent; | |
7184 | u32 blocksize; | |
7185 | struct btrfs_key key; | |
7186 | struct extent_buffer *next; | |
7187 | int level = wc->level; | |
7188 | int reada = 0; | |
7189 | int ret = 0; | |
7190 | ||
7191 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
7192 | path->slots[level]); | |
7193 | /* | |
7194 | * if the lower level block was created before the snapshot | |
7195 | * was created, we know there is no need to update back refs | |
7196 | * for the subtree | |
7197 | */ | |
7198 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
7199 | generation <= root->root_key.offset) { |
7200 | *lookup_info = 1; | |
1c4850e2 | 7201 | return 1; |
94fcca9f | 7202 | } |
1c4850e2 YZ |
7203 | |
7204 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
7205 | blocksize = btrfs_level_size(root, level - 1); | |
7206 | ||
7207 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
7208 | if (!next) { | |
7209 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
7210 | if (!next) |
7211 | return -ENOMEM; | |
b2aaaa3b JB |
7212 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
7213 | level - 1); | |
1c4850e2 YZ |
7214 | reada = 1; |
7215 | } | |
7216 | btrfs_tree_lock(next); | |
7217 | btrfs_set_lock_blocking(next); | |
7218 | ||
3173a18f | 7219 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
7220 | &wc->refs[level - 1], |
7221 | &wc->flags[level - 1]); | |
79787eaa JM |
7222 | if (ret < 0) { |
7223 | btrfs_tree_unlock(next); | |
7224 | return ret; | |
7225 | } | |
7226 | ||
c2cf52eb SK |
7227 | if (unlikely(wc->refs[level - 1] == 0)) { |
7228 | btrfs_err(root->fs_info, "Missing references."); | |
7229 | BUG(); | |
7230 | } | |
94fcca9f | 7231 | *lookup_info = 0; |
1c4850e2 | 7232 | |
94fcca9f | 7233 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 7234 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
7235 | if (level == 1 && |
7236 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7237 | goto skip; | |
7238 | ||
1c4850e2 YZ |
7239 | if (!wc->update_ref || |
7240 | generation <= root->root_key.offset) | |
7241 | goto skip; | |
7242 | ||
7243 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
7244 | path->slots[level]); | |
7245 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
7246 | if (ret < 0) | |
7247 | goto skip; | |
7248 | ||
7249 | wc->stage = UPDATE_BACKREF; | |
7250 | wc->shared_level = level - 1; | |
7251 | } | |
94fcca9f YZ |
7252 | } else { |
7253 | if (level == 1 && | |
7254 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7255 | goto skip; | |
1c4850e2 YZ |
7256 | } |
7257 | ||
b9fab919 | 7258 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
7259 | btrfs_tree_unlock(next); |
7260 | free_extent_buffer(next); | |
7261 | next = NULL; | |
94fcca9f | 7262 | *lookup_info = 1; |
1c4850e2 YZ |
7263 | } |
7264 | ||
7265 | if (!next) { | |
7266 | if (reada && level == 1) | |
7267 | reada_walk_down(trans, root, wc, path); | |
7268 | next = read_tree_block(root, bytenr, blocksize, generation); | |
416bc658 JB |
7269 | if (!next || !extent_buffer_uptodate(next)) { |
7270 | free_extent_buffer(next); | |
97d9a8a4 | 7271 | return -EIO; |
416bc658 | 7272 | } |
1c4850e2 YZ |
7273 | btrfs_tree_lock(next); |
7274 | btrfs_set_lock_blocking(next); | |
7275 | } | |
7276 | ||
7277 | level--; | |
7278 | BUG_ON(level != btrfs_header_level(next)); | |
7279 | path->nodes[level] = next; | |
7280 | path->slots[level] = 0; | |
bd681513 | 7281 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
7282 | wc->level = level; |
7283 | if (wc->level == 1) | |
7284 | wc->reada_slot = 0; | |
7285 | return 0; | |
7286 | skip: | |
7287 | wc->refs[level - 1] = 0; | |
7288 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
7289 | if (wc->stage == DROP_REFERENCE) { |
7290 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
7291 | parent = path->nodes[level]->start; | |
7292 | } else { | |
7293 | BUG_ON(root->root_key.objectid != | |
7294 | btrfs_header_owner(path->nodes[level])); | |
7295 | parent = 0; | |
7296 | } | |
1c4850e2 | 7297 | |
94fcca9f | 7298 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 7299 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 7300 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 7301 | } |
1c4850e2 YZ |
7302 | btrfs_tree_unlock(next); |
7303 | free_extent_buffer(next); | |
94fcca9f | 7304 | *lookup_info = 1; |
1c4850e2 YZ |
7305 | return 1; |
7306 | } | |
7307 | ||
2c47e605 | 7308 | /* |
2c016dc2 | 7309 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
7310 | * |
7311 | * when wc->stage == DROP_REFERENCE, this function drops | |
7312 | * reference count on the block. | |
7313 | * | |
7314 | * when wc->stage == UPDATE_BACKREF, this function changes | |
7315 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
7316 | * to UPDATE_BACKREF previously while processing the block. | |
7317 | * | |
7318 | * NOTE: return value 1 means we should stop walking up. | |
7319 | */ | |
7320 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
7321 | struct btrfs_root *root, | |
7322 | struct btrfs_path *path, | |
7323 | struct walk_control *wc) | |
7324 | { | |
f0486c68 | 7325 | int ret; |
2c47e605 YZ |
7326 | int level = wc->level; |
7327 | struct extent_buffer *eb = path->nodes[level]; | |
7328 | u64 parent = 0; | |
7329 | ||
7330 | if (wc->stage == UPDATE_BACKREF) { | |
7331 | BUG_ON(wc->shared_level < level); | |
7332 | if (level < wc->shared_level) | |
7333 | goto out; | |
7334 | ||
2c47e605 YZ |
7335 | ret = find_next_key(path, level + 1, &wc->update_progress); |
7336 | if (ret > 0) | |
7337 | wc->update_ref = 0; | |
7338 | ||
7339 | wc->stage = DROP_REFERENCE; | |
7340 | wc->shared_level = -1; | |
7341 | path->slots[level] = 0; | |
7342 | ||
7343 | /* | |
7344 | * check reference count again if the block isn't locked. | |
7345 | * we should start walking down the tree again if reference | |
7346 | * count is one. | |
7347 | */ | |
7348 | if (!path->locks[level]) { | |
7349 | BUG_ON(level == 0); | |
7350 | btrfs_tree_lock(eb); | |
7351 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7352 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7353 | |
7354 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7355 | eb->start, level, 1, |
2c47e605 YZ |
7356 | &wc->refs[level], |
7357 | &wc->flags[level]); | |
79787eaa JM |
7358 | if (ret < 0) { |
7359 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 7360 | path->locks[level] = 0; |
79787eaa JM |
7361 | return ret; |
7362 | } | |
2c47e605 YZ |
7363 | BUG_ON(wc->refs[level] == 0); |
7364 | if (wc->refs[level] == 1) { | |
bd681513 | 7365 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 7366 | path->locks[level] = 0; |
2c47e605 YZ |
7367 | return 1; |
7368 | } | |
f82d02d9 | 7369 | } |
2c47e605 | 7370 | } |
f82d02d9 | 7371 | |
2c47e605 YZ |
7372 | /* wc->stage == DROP_REFERENCE */ |
7373 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 7374 | |
2c47e605 YZ |
7375 | if (wc->refs[level] == 1) { |
7376 | if (level == 0) { | |
7377 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
7378 | ret = btrfs_dec_ref(trans, root, eb, 1, |
7379 | wc->for_reloc); | |
2c47e605 | 7380 | else |
66d7e7f0 AJ |
7381 | ret = btrfs_dec_ref(trans, root, eb, 0, |
7382 | wc->for_reloc); | |
79787eaa | 7383 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7384 | } |
7385 | /* make block locked assertion in clean_tree_block happy */ | |
7386 | if (!path->locks[level] && | |
7387 | btrfs_header_generation(eb) == trans->transid) { | |
7388 | btrfs_tree_lock(eb); | |
7389 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7390 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7391 | } |
7392 | clean_tree_block(trans, root, eb); | |
7393 | } | |
7394 | ||
7395 | if (eb == root->node) { | |
7396 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7397 | parent = eb->start; | |
7398 | else | |
7399 | BUG_ON(root->root_key.objectid != | |
7400 | btrfs_header_owner(eb)); | |
7401 | } else { | |
7402 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7403 | parent = path->nodes[level + 1]->start; | |
7404 | else | |
7405 | BUG_ON(root->root_key.objectid != | |
7406 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 7407 | } |
f82d02d9 | 7408 | |
5581a51a | 7409 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
7410 | out: |
7411 | wc->refs[level] = 0; | |
7412 | wc->flags[level] = 0; | |
f0486c68 | 7413 | return 0; |
2c47e605 YZ |
7414 | } |
7415 | ||
7416 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
7417 | struct btrfs_root *root, | |
7418 | struct btrfs_path *path, | |
7419 | struct walk_control *wc) | |
7420 | { | |
2c47e605 | 7421 | int level = wc->level; |
94fcca9f | 7422 | int lookup_info = 1; |
2c47e605 YZ |
7423 | int ret; |
7424 | ||
7425 | while (level >= 0) { | |
94fcca9f | 7426 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
7427 | if (ret > 0) |
7428 | break; | |
7429 | ||
7430 | if (level == 0) | |
7431 | break; | |
7432 | ||
7a7965f8 YZ |
7433 | if (path->slots[level] >= |
7434 | btrfs_header_nritems(path->nodes[level])) | |
7435 | break; | |
7436 | ||
94fcca9f | 7437 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
7438 | if (ret > 0) { |
7439 | path->slots[level]++; | |
7440 | continue; | |
90d2c51d MX |
7441 | } else if (ret < 0) |
7442 | return ret; | |
1c4850e2 | 7443 | level = wc->level; |
f82d02d9 | 7444 | } |
f82d02d9 YZ |
7445 | return 0; |
7446 | } | |
7447 | ||
d397712b | 7448 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 7449 | struct btrfs_root *root, |
f82d02d9 | 7450 | struct btrfs_path *path, |
2c47e605 | 7451 | struct walk_control *wc, int max_level) |
20524f02 | 7452 | { |
2c47e605 | 7453 | int level = wc->level; |
20524f02 | 7454 | int ret; |
9f3a7427 | 7455 | |
2c47e605 YZ |
7456 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
7457 | while (level < max_level && path->nodes[level]) { | |
7458 | wc->level = level; | |
7459 | if (path->slots[level] + 1 < | |
7460 | btrfs_header_nritems(path->nodes[level])) { | |
7461 | path->slots[level]++; | |
20524f02 CM |
7462 | return 0; |
7463 | } else { | |
2c47e605 YZ |
7464 | ret = walk_up_proc(trans, root, path, wc); |
7465 | if (ret > 0) | |
7466 | return 0; | |
bd56b302 | 7467 | |
2c47e605 | 7468 | if (path->locks[level]) { |
bd681513 CM |
7469 | btrfs_tree_unlock_rw(path->nodes[level], |
7470 | path->locks[level]); | |
2c47e605 | 7471 | path->locks[level] = 0; |
f82d02d9 | 7472 | } |
2c47e605 YZ |
7473 | free_extent_buffer(path->nodes[level]); |
7474 | path->nodes[level] = NULL; | |
7475 | level++; | |
20524f02 CM |
7476 | } |
7477 | } | |
7478 | return 1; | |
7479 | } | |
7480 | ||
9aca1d51 | 7481 | /* |
2c47e605 YZ |
7482 | * drop a subvolume tree. |
7483 | * | |
7484 | * this function traverses the tree freeing any blocks that only | |
7485 | * referenced by the tree. | |
7486 | * | |
7487 | * when a shared tree block is found. this function decreases its | |
7488 | * reference count by one. if update_ref is true, this function | |
7489 | * also make sure backrefs for the shared block and all lower level | |
7490 | * blocks are properly updated. | |
9d1a2a3a DS |
7491 | * |
7492 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 7493 | */ |
2c536799 | 7494 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
7495 | struct btrfs_block_rsv *block_rsv, int update_ref, |
7496 | int for_reloc) | |
20524f02 | 7497 | { |
5caf2a00 | 7498 | struct btrfs_path *path; |
2c47e605 YZ |
7499 | struct btrfs_trans_handle *trans; |
7500 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 7501 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
7502 | struct walk_control *wc; |
7503 | struct btrfs_key key; | |
7504 | int err = 0; | |
7505 | int ret; | |
7506 | int level; | |
d29a9f62 | 7507 | bool root_dropped = false; |
20524f02 | 7508 | |
5caf2a00 | 7509 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
7510 | if (!path) { |
7511 | err = -ENOMEM; | |
7512 | goto out; | |
7513 | } | |
20524f02 | 7514 | |
2c47e605 | 7515 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
7516 | if (!wc) { |
7517 | btrfs_free_path(path); | |
cb1b69f4 TI |
7518 | err = -ENOMEM; |
7519 | goto out; | |
38a1a919 | 7520 | } |
2c47e605 | 7521 | |
a22285a6 | 7522 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7523 | if (IS_ERR(trans)) { |
7524 | err = PTR_ERR(trans); | |
7525 | goto out_free; | |
7526 | } | |
98d5dc13 | 7527 | |
3fd0a558 YZ |
7528 | if (block_rsv) |
7529 | trans->block_rsv = block_rsv; | |
2c47e605 | 7530 | |
9f3a7427 | 7531 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 7532 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
7533 | path->nodes[level] = btrfs_lock_root_node(root); |
7534 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 7535 | path->slots[level] = 0; |
bd681513 | 7536 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7537 | memset(&wc->update_progress, 0, |
7538 | sizeof(wc->update_progress)); | |
9f3a7427 | 7539 | } else { |
9f3a7427 | 7540 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
7541 | memcpy(&wc->update_progress, &key, |
7542 | sizeof(wc->update_progress)); | |
7543 | ||
6702ed49 | 7544 | level = root_item->drop_level; |
2c47e605 | 7545 | BUG_ON(level == 0); |
6702ed49 | 7546 | path->lowest_level = level; |
2c47e605 YZ |
7547 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7548 | path->lowest_level = 0; | |
7549 | if (ret < 0) { | |
7550 | err = ret; | |
79787eaa | 7551 | goto out_end_trans; |
9f3a7427 | 7552 | } |
1c4850e2 | 7553 | WARN_ON(ret > 0); |
2c47e605 | 7554 | |
7d9eb12c CM |
7555 | /* |
7556 | * unlock our path, this is safe because only this | |
7557 | * function is allowed to delete this snapshot | |
7558 | */ | |
5d4f98a2 | 7559 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7560 | |
7561 | level = btrfs_header_level(root->node); | |
7562 | while (1) { | |
7563 | btrfs_tree_lock(path->nodes[level]); | |
7564 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 7565 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7566 | |
7567 | ret = btrfs_lookup_extent_info(trans, root, | |
7568 | path->nodes[level]->start, | |
3173a18f | 7569 | level, 1, &wc->refs[level], |
2c47e605 | 7570 | &wc->flags[level]); |
79787eaa JM |
7571 | if (ret < 0) { |
7572 | err = ret; | |
7573 | goto out_end_trans; | |
7574 | } | |
2c47e605 YZ |
7575 | BUG_ON(wc->refs[level] == 0); |
7576 | ||
7577 | if (level == root_item->drop_level) | |
7578 | break; | |
7579 | ||
7580 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 7581 | path->locks[level] = 0; |
2c47e605 YZ |
7582 | WARN_ON(wc->refs[level] != 1); |
7583 | level--; | |
7584 | } | |
9f3a7427 | 7585 | } |
2c47e605 YZ |
7586 | |
7587 | wc->level = level; | |
7588 | wc->shared_level = -1; | |
7589 | wc->stage = DROP_REFERENCE; | |
7590 | wc->update_ref = update_ref; | |
7591 | wc->keep_locks = 0; | |
66d7e7f0 | 7592 | wc->for_reloc = for_reloc; |
1c4850e2 | 7593 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7594 | |
d397712b | 7595 | while (1) { |
9d1a2a3a | 7596 | |
2c47e605 YZ |
7597 | ret = walk_down_tree(trans, root, path, wc); |
7598 | if (ret < 0) { | |
7599 | err = ret; | |
20524f02 | 7600 | break; |
2c47e605 | 7601 | } |
9aca1d51 | 7602 | |
2c47e605 YZ |
7603 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7604 | if (ret < 0) { | |
7605 | err = ret; | |
20524f02 | 7606 | break; |
2c47e605 YZ |
7607 | } |
7608 | ||
7609 | if (ret > 0) { | |
7610 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7611 | break; |
7612 | } | |
2c47e605 YZ |
7613 | |
7614 | if (wc->stage == DROP_REFERENCE) { | |
7615 | level = wc->level; | |
7616 | btrfs_node_key(path->nodes[level], | |
7617 | &root_item->drop_progress, | |
7618 | path->slots[level]); | |
7619 | root_item->drop_level = level; | |
7620 | } | |
7621 | ||
7622 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
7623 | if (btrfs_should_end_transaction(trans, tree_root) || |
7624 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
7625 | ret = btrfs_update_root(trans, tree_root, |
7626 | &root->root_key, | |
7627 | root_item); | |
79787eaa JM |
7628 | if (ret) { |
7629 | btrfs_abort_transaction(trans, tree_root, ret); | |
7630 | err = ret; | |
7631 | goto out_end_trans; | |
7632 | } | |
2c47e605 | 7633 | |
3fd0a558 | 7634 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 JB |
7635 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
7636 | pr_debug("btrfs: drop snapshot early exit\n"); | |
7637 | err = -EAGAIN; | |
7638 | goto out_free; | |
7639 | } | |
7640 | ||
a22285a6 | 7641 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7642 | if (IS_ERR(trans)) { |
7643 | err = PTR_ERR(trans); | |
7644 | goto out_free; | |
7645 | } | |
3fd0a558 YZ |
7646 | if (block_rsv) |
7647 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7648 | } |
20524f02 | 7649 | } |
b3b4aa74 | 7650 | btrfs_release_path(path); |
79787eaa JM |
7651 | if (err) |
7652 | goto out_end_trans; | |
2c47e605 YZ |
7653 | |
7654 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7655 | if (ret) { |
7656 | btrfs_abort_transaction(trans, tree_root, ret); | |
7657 | goto out_end_trans; | |
7658 | } | |
2c47e605 | 7659 | |
76dda93c | 7660 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
7661 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
7662 | NULL, NULL); | |
79787eaa JM |
7663 | if (ret < 0) { |
7664 | btrfs_abort_transaction(trans, tree_root, ret); | |
7665 | err = ret; | |
7666 | goto out_end_trans; | |
7667 | } else if (ret > 0) { | |
84cd948c JB |
7668 | /* if we fail to delete the orphan item this time |
7669 | * around, it'll get picked up the next time. | |
7670 | * | |
7671 | * The most common failure here is just -ENOENT. | |
7672 | */ | |
7673 | btrfs_del_orphan_item(trans, tree_root, | |
7674 | root->root_key.objectid); | |
76dda93c YZ |
7675 | } |
7676 | } | |
7677 | ||
7678 | if (root->in_radix) { | |
cb517eab | 7679 | btrfs_drop_and_free_fs_root(tree_root->fs_info, root); |
76dda93c YZ |
7680 | } else { |
7681 | free_extent_buffer(root->node); | |
7682 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 7683 | btrfs_put_fs_root(root); |
76dda93c | 7684 | } |
d29a9f62 | 7685 | root_dropped = true; |
79787eaa | 7686 | out_end_trans: |
3fd0a558 | 7687 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7688 | out_free: |
2c47e605 | 7689 | kfree(wc); |
5caf2a00 | 7690 | btrfs_free_path(path); |
cb1b69f4 | 7691 | out: |
d29a9f62 JB |
7692 | /* |
7693 | * So if we need to stop dropping the snapshot for whatever reason we | |
7694 | * need to make sure to add it back to the dead root list so that we | |
7695 | * keep trying to do the work later. This also cleans up roots if we | |
7696 | * don't have it in the radix (like when we recover after a power fail | |
7697 | * or unmount) so we don't leak memory. | |
7698 | */ | |
7699 | if (root_dropped == false) | |
7700 | btrfs_add_dead_root(root); | |
cb1b69f4 TI |
7701 | if (err) |
7702 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7703 | return err; |
20524f02 | 7704 | } |
9078a3e1 | 7705 | |
2c47e605 YZ |
7706 | /* |
7707 | * drop subtree rooted at tree block 'node'. | |
7708 | * | |
7709 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7710 | * only used by relocation code |
2c47e605 | 7711 | */ |
f82d02d9 YZ |
7712 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7713 | struct btrfs_root *root, | |
7714 | struct extent_buffer *node, | |
7715 | struct extent_buffer *parent) | |
7716 | { | |
7717 | struct btrfs_path *path; | |
2c47e605 | 7718 | struct walk_control *wc; |
f82d02d9 YZ |
7719 | int level; |
7720 | int parent_level; | |
7721 | int ret = 0; | |
7722 | int wret; | |
7723 | ||
2c47e605 YZ |
7724 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7725 | ||
f82d02d9 | 7726 | path = btrfs_alloc_path(); |
db5b493a TI |
7727 | if (!path) |
7728 | return -ENOMEM; | |
f82d02d9 | 7729 | |
2c47e605 | 7730 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7731 | if (!wc) { |
7732 | btrfs_free_path(path); | |
7733 | return -ENOMEM; | |
7734 | } | |
2c47e605 | 7735 | |
b9447ef8 | 7736 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7737 | parent_level = btrfs_header_level(parent); |
7738 | extent_buffer_get(parent); | |
7739 | path->nodes[parent_level] = parent; | |
7740 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7741 | ||
b9447ef8 | 7742 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7743 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7744 | path->nodes[level] = node; |
7745 | path->slots[level] = 0; | |
bd681513 | 7746 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7747 | |
7748 | wc->refs[parent_level] = 1; | |
7749 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7750 | wc->level = level; | |
7751 | wc->shared_level = -1; | |
7752 | wc->stage = DROP_REFERENCE; | |
7753 | wc->update_ref = 0; | |
7754 | wc->keep_locks = 1; | |
66d7e7f0 | 7755 | wc->for_reloc = 1; |
1c4850e2 | 7756 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7757 | |
7758 | while (1) { | |
2c47e605 YZ |
7759 | wret = walk_down_tree(trans, root, path, wc); |
7760 | if (wret < 0) { | |
f82d02d9 | 7761 | ret = wret; |
f82d02d9 | 7762 | break; |
2c47e605 | 7763 | } |
f82d02d9 | 7764 | |
2c47e605 | 7765 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7766 | if (wret < 0) |
7767 | ret = wret; | |
7768 | if (wret != 0) | |
7769 | break; | |
7770 | } | |
7771 | ||
2c47e605 | 7772 | kfree(wc); |
f82d02d9 YZ |
7773 | btrfs_free_path(path); |
7774 | return ret; | |
7775 | } | |
7776 | ||
ec44a35c CM |
7777 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7778 | { | |
7779 | u64 num_devices; | |
fc67c450 | 7780 | u64 stripped; |
e4d8ec0f | 7781 | |
fc67c450 ID |
7782 | /* |
7783 | * if restripe for this chunk_type is on pick target profile and | |
7784 | * return, otherwise do the usual balance | |
7785 | */ | |
7786 | stripped = get_restripe_target(root->fs_info, flags); | |
7787 | if (stripped) | |
7788 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7789 | |
cd02dca5 CM |
7790 | /* |
7791 | * we add in the count of missing devices because we want | |
7792 | * to make sure that any RAID levels on a degraded FS | |
7793 | * continue to be honored. | |
7794 | */ | |
7795 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7796 | root->fs_info->fs_devices->missing_devices; | |
7797 | ||
fc67c450 | 7798 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 7799 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
7800 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
7801 | ||
ec44a35c CM |
7802 | if (num_devices == 1) { |
7803 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7804 | stripped = flags & ~stripped; | |
7805 | ||
7806 | /* turn raid0 into single device chunks */ | |
7807 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7808 | return stripped; | |
7809 | ||
7810 | /* turn mirroring into duplication */ | |
7811 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7812 | BTRFS_BLOCK_GROUP_RAID10)) | |
7813 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7814 | } else { |
7815 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7816 | if (flags & stripped) |
7817 | return flags; | |
7818 | ||
7819 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7820 | stripped = flags & ~stripped; | |
7821 | ||
7822 | /* switch duplicated blocks with raid1 */ | |
7823 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7824 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7825 | ||
e3176ca2 | 7826 | /* this is drive concat, leave it alone */ |
ec44a35c | 7827 | } |
e3176ca2 | 7828 | |
ec44a35c CM |
7829 | return flags; |
7830 | } | |
7831 | ||
199c36ea | 7832 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7833 | { |
f0486c68 YZ |
7834 | struct btrfs_space_info *sinfo = cache->space_info; |
7835 | u64 num_bytes; | |
199c36ea | 7836 | u64 min_allocable_bytes; |
f0486c68 | 7837 | int ret = -ENOSPC; |
0ef3e66b | 7838 | |
c286ac48 | 7839 | |
199c36ea MX |
7840 | /* |
7841 | * We need some metadata space and system metadata space for | |
7842 | * allocating chunks in some corner cases until we force to set | |
7843 | * it to be readonly. | |
7844 | */ | |
7845 | if ((sinfo->flags & | |
7846 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7847 | !force) | |
7848 | min_allocable_bytes = 1 * 1024 * 1024; | |
7849 | else | |
7850 | min_allocable_bytes = 0; | |
7851 | ||
f0486c68 YZ |
7852 | spin_lock(&sinfo->lock); |
7853 | spin_lock(&cache->lock); | |
61cfea9b W |
7854 | |
7855 | if (cache->ro) { | |
7856 | ret = 0; | |
7857 | goto out; | |
7858 | } | |
7859 | ||
f0486c68 YZ |
7860 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7861 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7862 | ||
7863 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7864 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7865 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7866 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7867 | cache->ro = 1; |
7868 | ret = 0; | |
7869 | } | |
61cfea9b | 7870 | out: |
f0486c68 YZ |
7871 | spin_unlock(&cache->lock); |
7872 | spin_unlock(&sinfo->lock); | |
7873 | return ret; | |
7874 | } | |
7d9eb12c | 7875 | |
f0486c68 YZ |
7876 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7877 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7878 | |
f0486c68 YZ |
7879 | { |
7880 | struct btrfs_trans_handle *trans; | |
7881 | u64 alloc_flags; | |
7882 | int ret; | |
7d9eb12c | 7883 | |
f0486c68 | 7884 | BUG_ON(cache->ro); |
0ef3e66b | 7885 | |
ff5714cc | 7886 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7887 | if (IS_ERR(trans)) |
7888 | return PTR_ERR(trans); | |
5d4f98a2 | 7889 | |
f0486c68 | 7890 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 7891 | if (alloc_flags != cache->flags) { |
698d0082 | 7892 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
7893 | CHUNK_ALLOC_FORCE); |
7894 | if (ret < 0) | |
7895 | goto out; | |
7896 | } | |
5d4f98a2 | 7897 | |
199c36ea | 7898 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7899 | if (!ret) |
7900 | goto out; | |
7901 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 7902 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7903 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7904 | if (ret < 0) |
7905 | goto out; | |
199c36ea | 7906 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7907 | out: |
7908 | btrfs_end_transaction(trans, root); | |
7909 | return ret; | |
7910 | } | |
5d4f98a2 | 7911 | |
c87f08ca CM |
7912 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7913 | struct btrfs_root *root, u64 type) | |
7914 | { | |
7915 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 7916 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7917 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
7918 | } |
7919 | ||
6d07bcec MX |
7920 | /* |
7921 | * helper to account the unused space of all the readonly block group in the | |
7922 | * list. takes mirrors into account. | |
7923 | */ | |
7924 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7925 | { | |
7926 | struct btrfs_block_group_cache *block_group; | |
7927 | u64 free_bytes = 0; | |
7928 | int factor; | |
7929 | ||
7930 | list_for_each_entry(block_group, groups_list, list) { | |
7931 | spin_lock(&block_group->lock); | |
7932 | ||
7933 | if (!block_group->ro) { | |
7934 | spin_unlock(&block_group->lock); | |
7935 | continue; | |
7936 | } | |
7937 | ||
7938 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7939 | BTRFS_BLOCK_GROUP_RAID10 | | |
7940 | BTRFS_BLOCK_GROUP_DUP)) | |
7941 | factor = 2; | |
7942 | else | |
7943 | factor = 1; | |
7944 | ||
7945 | free_bytes += (block_group->key.offset - | |
7946 | btrfs_block_group_used(&block_group->item)) * | |
7947 | factor; | |
7948 | ||
7949 | spin_unlock(&block_group->lock); | |
7950 | } | |
7951 | ||
7952 | return free_bytes; | |
7953 | } | |
7954 | ||
7955 | /* | |
7956 | * helper to account the unused space of all the readonly block group in the | |
7957 | * space_info. takes mirrors into account. | |
7958 | */ | |
7959 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7960 | { | |
7961 | int i; | |
7962 | u64 free_bytes = 0; | |
7963 | ||
7964 | spin_lock(&sinfo->lock); | |
7965 | ||
7966 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7967 | if (!list_empty(&sinfo->block_groups[i])) | |
7968 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7969 | &sinfo->block_groups[i]); | |
7970 | ||
7971 | spin_unlock(&sinfo->lock); | |
7972 | ||
7973 | return free_bytes; | |
7974 | } | |
7975 | ||
143bede5 | 7976 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7977 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7978 | { |
f0486c68 YZ |
7979 | struct btrfs_space_info *sinfo = cache->space_info; |
7980 | u64 num_bytes; | |
7981 | ||
7982 | BUG_ON(!cache->ro); | |
7983 | ||
7984 | spin_lock(&sinfo->lock); | |
7985 | spin_lock(&cache->lock); | |
7986 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7987 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7988 | sinfo->bytes_readonly -= num_bytes; | |
7989 | cache->ro = 0; | |
7990 | spin_unlock(&cache->lock); | |
7991 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7992 | } |
7993 | ||
ba1bf481 JB |
7994 | /* |
7995 | * checks to see if its even possible to relocate this block group. | |
7996 | * | |
7997 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7998 | * ok to go ahead and try. | |
7999 | */ | |
8000 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 8001 | { |
ba1bf481 JB |
8002 | struct btrfs_block_group_cache *block_group; |
8003 | struct btrfs_space_info *space_info; | |
8004 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
8005 | struct btrfs_device *device; | |
6df9a95e | 8006 | struct btrfs_trans_handle *trans; |
cdcb725c | 8007 | u64 min_free; |
6719db6a JB |
8008 | u64 dev_min = 1; |
8009 | u64 dev_nr = 0; | |
4a5e98f5 | 8010 | u64 target; |
cdcb725c | 8011 | int index; |
ba1bf481 JB |
8012 | int full = 0; |
8013 | int ret = 0; | |
1a40e23b | 8014 | |
ba1bf481 | 8015 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 8016 | |
ba1bf481 JB |
8017 | /* odd, couldn't find the block group, leave it alone */ |
8018 | if (!block_group) | |
8019 | return -1; | |
1a40e23b | 8020 | |
cdcb725c | 8021 | min_free = btrfs_block_group_used(&block_group->item); |
8022 | ||
ba1bf481 | 8023 | /* no bytes used, we're good */ |
cdcb725c | 8024 | if (!min_free) |
1a40e23b ZY |
8025 | goto out; |
8026 | ||
ba1bf481 JB |
8027 | space_info = block_group->space_info; |
8028 | spin_lock(&space_info->lock); | |
17d217fe | 8029 | |
ba1bf481 | 8030 | full = space_info->full; |
17d217fe | 8031 | |
ba1bf481 JB |
8032 | /* |
8033 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
8034 | * relocate it unless we're able to allocate a new chunk below. |
8035 | * | |
8036 | * Otherwise, we need to make sure we have room in the space to handle | |
8037 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 8038 | */ |
7ce618db | 8039 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 8040 | (space_info->bytes_used + space_info->bytes_reserved + |
8041 | space_info->bytes_pinned + space_info->bytes_readonly + | |
8042 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
8043 | spin_unlock(&space_info->lock); |
8044 | goto out; | |
17d217fe | 8045 | } |
ba1bf481 | 8046 | spin_unlock(&space_info->lock); |
ea8c2819 | 8047 | |
ba1bf481 JB |
8048 | /* |
8049 | * ok we don't have enough space, but maybe we have free space on our | |
8050 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
8051 | * alloc devices and guess if we have enough space. if this block |
8052 | * group is going to be restriped, run checks against the target | |
8053 | * profile instead of the current one. | |
ba1bf481 JB |
8054 | */ |
8055 | ret = -1; | |
ea8c2819 | 8056 | |
cdcb725c | 8057 | /* |
8058 | * index: | |
8059 | * 0: raid10 | |
8060 | * 1: raid1 | |
8061 | * 2: dup | |
8062 | * 3: raid0 | |
8063 | * 4: single | |
8064 | */ | |
4a5e98f5 ID |
8065 | target = get_restripe_target(root->fs_info, block_group->flags); |
8066 | if (target) { | |
31e50229 | 8067 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
8068 | } else { |
8069 | /* | |
8070 | * this is just a balance, so if we were marked as full | |
8071 | * we know there is no space for a new chunk | |
8072 | */ | |
8073 | if (full) | |
8074 | goto out; | |
8075 | ||
8076 | index = get_block_group_index(block_group); | |
8077 | } | |
8078 | ||
e6ec716f | 8079 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 8080 | dev_min = 4; |
6719db6a JB |
8081 | /* Divide by 2 */ |
8082 | min_free >>= 1; | |
e6ec716f | 8083 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 8084 | dev_min = 2; |
e6ec716f | 8085 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
8086 | /* Multiply by 2 */ |
8087 | min_free <<= 1; | |
e6ec716f | 8088 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 8089 | dev_min = fs_devices->rw_devices; |
6719db6a | 8090 | do_div(min_free, dev_min); |
cdcb725c | 8091 | } |
8092 | ||
6df9a95e JB |
8093 | /* We need to do this so that we can look at pending chunks */ |
8094 | trans = btrfs_join_transaction(root); | |
8095 | if (IS_ERR(trans)) { | |
8096 | ret = PTR_ERR(trans); | |
8097 | goto out; | |
8098 | } | |
8099 | ||
ba1bf481 JB |
8100 | mutex_lock(&root->fs_info->chunk_mutex); |
8101 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 8102 | u64 dev_offset; |
56bec294 | 8103 | |
ba1bf481 JB |
8104 | /* |
8105 | * check to make sure we can actually find a chunk with enough | |
8106 | * space to fit our block group in. | |
8107 | */ | |
63a212ab SB |
8108 | if (device->total_bytes > device->bytes_used + min_free && |
8109 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 8110 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 8111 | &dev_offset, NULL); |
ba1bf481 | 8112 | if (!ret) |
cdcb725c | 8113 | dev_nr++; |
8114 | ||
8115 | if (dev_nr >= dev_min) | |
73e48b27 | 8116 | break; |
cdcb725c | 8117 | |
ba1bf481 | 8118 | ret = -1; |
725c8463 | 8119 | } |
edbd8d4e | 8120 | } |
ba1bf481 | 8121 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 8122 | btrfs_end_transaction(trans, root); |
edbd8d4e | 8123 | out: |
ba1bf481 | 8124 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
8125 | return ret; |
8126 | } | |
8127 | ||
b2950863 CH |
8128 | static int find_first_block_group(struct btrfs_root *root, |
8129 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 8130 | { |
925baedd | 8131 | int ret = 0; |
0b86a832 CM |
8132 | struct btrfs_key found_key; |
8133 | struct extent_buffer *leaf; | |
8134 | int slot; | |
edbd8d4e | 8135 | |
0b86a832 CM |
8136 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
8137 | if (ret < 0) | |
925baedd CM |
8138 | goto out; |
8139 | ||
d397712b | 8140 | while (1) { |
0b86a832 | 8141 | slot = path->slots[0]; |
edbd8d4e | 8142 | leaf = path->nodes[0]; |
0b86a832 CM |
8143 | if (slot >= btrfs_header_nritems(leaf)) { |
8144 | ret = btrfs_next_leaf(root, path); | |
8145 | if (ret == 0) | |
8146 | continue; | |
8147 | if (ret < 0) | |
925baedd | 8148 | goto out; |
0b86a832 | 8149 | break; |
edbd8d4e | 8150 | } |
0b86a832 | 8151 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 8152 | |
0b86a832 | 8153 | if (found_key.objectid >= key->objectid && |
925baedd CM |
8154 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
8155 | ret = 0; | |
8156 | goto out; | |
8157 | } | |
0b86a832 | 8158 | path->slots[0]++; |
edbd8d4e | 8159 | } |
925baedd | 8160 | out: |
0b86a832 | 8161 | return ret; |
edbd8d4e CM |
8162 | } |
8163 | ||
0af3d00b JB |
8164 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
8165 | { | |
8166 | struct btrfs_block_group_cache *block_group; | |
8167 | u64 last = 0; | |
8168 | ||
8169 | while (1) { | |
8170 | struct inode *inode; | |
8171 | ||
8172 | block_group = btrfs_lookup_first_block_group(info, last); | |
8173 | while (block_group) { | |
8174 | spin_lock(&block_group->lock); | |
8175 | if (block_group->iref) | |
8176 | break; | |
8177 | spin_unlock(&block_group->lock); | |
8178 | block_group = next_block_group(info->tree_root, | |
8179 | block_group); | |
8180 | } | |
8181 | if (!block_group) { | |
8182 | if (last == 0) | |
8183 | break; | |
8184 | last = 0; | |
8185 | continue; | |
8186 | } | |
8187 | ||
8188 | inode = block_group->inode; | |
8189 | block_group->iref = 0; | |
8190 | block_group->inode = NULL; | |
8191 | spin_unlock(&block_group->lock); | |
8192 | iput(inode); | |
8193 | last = block_group->key.objectid + block_group->key.offset; | |
8194 | btrfs_put_block_group(block_group); | |
8195 | } | |
8196 | } | |
8197 | ||
1a40e23b ZY |
8198 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8199 | { | |
8200 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8201 | struct btrfs_space_info *space_info; |
11833d66 | 8202 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8203 | struct rb_node *n; |
8204 | ||
11833d66 YZ |
8205 | down_write(&info->extent_commit_sem); |
8206 | while (!list_empty(&info->caching_block_groups)) { | |
8207 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8208 | struct btrfs_caching_control, list); | |
8209 | list_del(&caching_ctl->list); | |
8210 | put_caching_control(caching_ctl); | |
8211 | } | |
8212 | up_write(&info->extent_commit_sem); | |
8213 | ||
1a40e23b ZY |
8214 | spin_lock(&info->block_group_cache_lock); |
8215 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8216 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8217 | cache_node); | |
1a40e23b ZY |
8218 | rb_erase(&block_group->cache_node, |
8219 | &info->block_group_cache_tree); | |
d899e052 YZ |
8220 | spin_unlock(&info->block_group_cache_lock); |
8221 | ||
80eb234a | 8222 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8223 | list_del(&block_group->list); |
80eb234a | 8224 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8225 | |
817d52f8 | 8226 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8227 | wait_block_group_cache_done(block_group); |
817d52f8 | 8228 | |
3c14874a JB |
8229 | /* |
8230 | * We haven't cached this block group, which means we could | |
8231 | * possibly have excluded extents on this block group. | |
8232 | */ | |
8233 | if (block_group->cached == BTRFS_CACHE_NO) | |
8234 | free_excluded_extents(info->extent_root, block_group); | |
8235 | ||
817d52f8 | 8236 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 8237 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8238 | |
8239 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8240 | } |
8241 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8242 | |
8243 | /* now that all the block groups are freed, go through and | |
8244 | * free all the space_info structs. This is only called during | |
8245 | * the final stages of unmount, and so we know nobody is | |
8246 | * using them. We call synchronize_rcu() once before we start, | |
8247 | * just to be on the safe side. | |
8248 | */ | |
8249 | synchronize_rcu(); | |
8250 | ||
8929ecfa YZ |
8251 | release_global_block_rsv(info); |
8252 | ||
4184ea7f CM |
8253 | while(!list_empty(&info->space_info)) { |
8254 | space_info = list_entry(info->space_info.next, | |
8255 | struct btrfs_space_info, | |
8256 | list); | |
b069e0c3 DS |
8257 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
8258 | if (space_info->bytes_pinned > 0 || | |
8259 | space_info->bytes_reserved > 0 || | |
8260 | space_info->bytes_may_use > 0) { | |
8261 | WARN_ON(1); | |
8262 | dump_space_info(space_info, 0, 0); | |
8263 | } | |
f0486c68 | 8264 | } |
b150a4f1 | 8265 | percpu_counter_destroy(&space_info->total_bytes_pinned); |
4184ea7f CM |
8266 | list_del(&space_info->list); |
8267 | kfree(space_info); | |
8268 | } | |
1a40e23b ZY |
8269 | return 0; |
8270 | } | |
8271 | ||
b742bb82 YZ |
8272 | static void __link_block_group(struct btrfs_space_info *space_info, |
8273 | struct btrfs_block_group_cache *cache) | |
8274 | { | |
8275 | int index = get_block_group_index(cache); | |
8276 | ||
8277 | down_write(&space_info->groups_sem); | |
8278 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8279 | up_write(&space_info->groups_sem); | |
8280 | } | |
8281 | ||
9078a3e1 CM |
8282 | int btrfs_read_block_groups(struct btrfs_root *root) |
8283 | { | |
8284 | struct btrfs_path *path; | |
8285 | int ret; | |
9078a3e1 | 8286 | struct btrfs_block_group_cache *cache; |
be744175 | 8287 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 8288 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8289 | struct btrfs_key key; |
8290 | struct btrfs_key found_key; | |
5f39d397 | 8291 | struct extent_buffer *leaf; |
0af3d00b JB |
8292 | int need_clear = 0; |
8293 | u64 cache_gen; | |
96b5179d | 8294 | |
be744175 | 8295 | root = info->extent_root; |
9078a3e1 | 8296 | key.objectid = 0; |
0b86a832 | 8297 | key.offset = 0; |
9078a3e1 | 8298 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
8299 | path = btrfs_alloc_path(); |
8300 | if (!path) | |
8301 | return -ENOMEM; | |
026fd317 | 8302 | path->reada = 1; |
9078a3e1 | 8303 | |
6c41761f | 8304 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 8305 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 8306 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 8307 | need_clear = 1; |
88c2ba3b JB |
8308 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
8309 | need_clear = 1; | |
0af3d00b | 8310 | |
d397712b | 8311 | while (1) { |
0b86a832 | 8312 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
8313 | if (ret > 0) |
8314 | break; | |
0b86a832 CM |
8315 | if (ret != 0) |
8316 | goto error; | |
5f39d397 CM |
8317 | leaf = path->nodes[0]; |
8318 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 8319 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 8320 | if (!cache) { |
0b86a832 | 8321 | ret = -ENOMEM; |
f0486c68 | 8322 | goto error; |
9078a3e1 | 8323 | } |
34d52cb6 LZ |
8324 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8325 | GFP_NOFS); | |
8326 | if (!cache->free_space_ctl) { | |
8327 | kfree(cache); | |
8328 | ret = -ENOMEM; | |
8329 | goto error; | |
8330 | } | |
3e1ad54f | 8331 | |
d2fb3437 | 8332 | atomic_set(&cache->count, 1); |
c286ac48 | 8333 | spin_lock_init(&cache->lock); |
817d52f8 | 8334 | cache->fs_info = info; |
0f9dd46c | 8335 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8336 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 8337 | |
cf7c1ef6 LB |
8338 | if (need_clear) { |
8339 | /* | |
8340 | * When we mount with old space cache, we need to | |
8341 | * set BTRFS_DC_CLEAR and set dirty flag. | |
8342 | * | |
8343 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
8344 | * truncate the old free space cache inode and | |
8345 | * setup a new one. | |
8346 | * b) Setting 'dirty flag' makes sure that we flush | |
8347 | * the new space cache info onto disk. | |
8348 | */ | |
0af3d00b | 8349 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
8350 | if (btrfs_test_opt(root, SPACE_CACHE)) |
8351 | cache->dirty = 1; | |
8352 | } | |
0af3d00b | 8353 | |
5f39d397 CM |
8354 | read_extent_buffer(leaf, &cache->item, |
8355 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8356 | sizeof(cache->item)); | |
9078a3e1 | 8357 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 8358 | |
9078a3e1 | 8359 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 8360 | btrfs_release_path(path); |
0b86a832 | 8361 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 | 8362 | cache->sectorsize = root->sectorsize; |
53b381b3 DW |
8363 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8364 | &root->fs_info->mapping_tree, | |
8365 | found_key.objectid); | |
34d52cb6 LZ |
8366 | btrfs_init_free_space_ctl(cache); |
8367 | ||
3c14874a JB |
8368 | /* |
8369 | * We need to exclude the super stripes now so that the space | |
8370 | * info has super bytes accounted for, otherwise we'll think | |
8371 | * we have more space than we actually do. | |
8372 | */ | |
835d974f JB |
8373 | ret = exclude_super_stripes(root, cache); |
8374 | if (ret) { | |
8375 | /* | |
8376 | * We may have excluded something, so call this just in | |
8377 | * case. | |
8378 | */ | |
8379 | free_excluded_extents(root, cache); | |
8380 | kfree(cache->free_space_ctl); | |
8381 | kfree(cache); | |
8382 | goto error; | |
8383 | } | |
3c14874a | 8384 | |
817d52f8 JB |
8385 | /* |
8386 | * check for two cases, either we are full, and therefore | |
8387 | * don't need to bother with the caching work since we won't | |
8388 | * find any space, or we are empty, and we can just add all | |
8389 | * the space in and be done with it. This saves us _alot_ of | |
8390 | * time, particularly in the full case. | |
8391 | */ | |
8392 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 8393 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8394 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 8395 | free_excluded_extents(root, cache); |
817d52f8 | 8396 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 8397 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
8398 | cache->cached = BTRFS_CACHE_FINISHED; |
8399 | add_new_free_space(cache, root->fs_info, | |
8400 | found_key.objectid, | |
8401 | found_key.objectid + | |
8402 | found_key.offset); | |
11833d66 | 8403 | free_excluded_extents(root, cache); |
817d52f8 | 8404 | } |
96b5179d | 8405 | |
8c579fe7 JB |
8406 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8407 | if (ret) { | |
8408 | btrfs_remove_free_space_cache(cache); | |
8409 | btrfs_put_block_group(cache); | |
8410 | goto error; | |
8411 | } | |
8412 | ||
6324fbf3 CM |
8413 | ret = update_space_info(info, cache->flags, found_key.offset, |
8414 | btrfs_block_group_used(&cache->item), | |
8415 | &space_info); | |
8c579fe7 JB |
8416 | if (ret) { |
8417 | btrfs_remove_free_space_cache(cache); | |
8418 | spin_lock(&info->block_group_cache_lock); | |
8419 | rb_erase(&cache->cache_node, | |
8420 | &info->block_group_cache_tree); | |
8421 | spin_unlock(&info->block_group_cache_lock); | |
8422 | btrfs_put_block_group(cache); | |
8423 | goto error; | |
8424 | } | |
8425 | ||
6324fbf3 | 8426 | cache->space_info = space_info; |
1b2da372 | 8427 | spin_lock(&cache->space_info->lock); |
f0486c68 | 8428 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8429 | spin_unlock(&cache->space_info->lock); |
8430 | ||
b742bb82 | 8431 | __link_block_group(space_info, cache); |
0f9dd46c | 8432 | |
75ccf47d | 8433 | set_avail_alloc_bits(root->fs_info, cache->flags); |
2b82032c | 8434 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 8435 | set_block_group_ro(cache, 1); |
9078a3e1 | 8436 | } |
b742bb82 YZ |
8437 | |
8438 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
8439 | if (!(get_alloc_profile(root, space_info->flags) & | |
8440 | (BTRFS_BLOCK_GROUP_RAID10 | | |
8441 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
8442 | BTRFS_BLOCK_GROUP_RAID5 | |
8443 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
8444 | BTRFS_BLOCK_GROUP_DUP))) |
8445 | continue; | |
8446 | /* | |
8447 | * avoid allocating from un-mirrored block group if there are | |
8448 | * mirrored block groups. | |
8449 | */ | |
1095cc0d | 8450 | list_for_each_entry(cache, |
8451 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
8452 | list) | |
199c36ea | 8453 | set_block_group_ro(cache, 1); |
1095cc0d | 8454 | list_for_each_entry(cache, |
8455 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
8456 | list) | |
199c36ea | 8457 | set_block_group_ro(cache, 1); |
9078a3e1 | 8458 | } |
f0486c68 YZ |
8459 | |
8460 | init_global_block_rsv(info); | |
0b86a832 CM |
8461 | ret = 0; |
8462 | error: | |
9078a3e1 | 8463 | btrfs_free_path(path); |
0b86a832 | 8464 | return ret; |
9078a3e1 | 8465 | } |
6324fbf3 | 8466 | |
ea658bad JB |
8467 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
8468 | struct btrfs_root *root) | |
8469 | { | |
8470 | struct btrfs_block_group_cache *block_group, *tmp; | |
8471 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
8472 | struct btrfs_block_group_item item; | |
8473 | struct btrfs_key key; | |
8474 | int ret = 0; | |
8475 | ||
8476 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
8477 | new_bg_list) { | |
8478 | list_del_init(&block_group->new_bg_list); | |
8479 | ||
8480 | if (ret) | |
8481 | continue; | |
8482 | ||
8483 | spin_lock(&block_group->lock); | |
8484 | memcpy(&item, &block_group->item, sizeof(item)); | |
8485 | memcpy(&key, &block_group->key, sizeof(key)); | |
8486 | spin_unlock(&block_group->lock); | |
8487 | ||
8488 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
8489 | sizeof(item)); | |
8490 | if (ret) | |
8491 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
8492 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
8493 | key.objectid, key.offset); | |
8494 | if (ret) | |
8495 | btrfs_abort_transaction(trans, extent_root, ret); | |
ea658bad JB |
8496 | } |
8497 | } | |
8498 | ||
6324fbf3 CM |
8499 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
8500 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 8501 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
8502 | u64 size) |
8503 | { | |
8504 | int ret; | |
6324fbf3 CM |
8505 | struct btrfs_root *extent_root; |
8506 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
8507 | |
8508 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 8509 | |
12fcfd22 | 8510 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 8511 | |
8f18cf13 | 8512 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
8513 | if (!cache) |
8514 | return -ENOMEM; | |
34d52cb6 LZ |
8515 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8516 | GFP_NOFS); | |
8517 | if (!cache->free_space_ctl) { | |
8518 | kfree(cache); | |
8519 | return -ENOMEM; | |
8520 | } | |
0f9dd46c | 8521 | |
e17cade2 | 8522 | cache->key.objectid = chunk_offset; |
6324fbf3 | 8523 | cache->key.offset = size; |
d2fb3437 | 8524 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 8525 | cache->sectorsize = root->sectorsize; |
0af3d00b | 8526 | cache->fs_info = root->fs_info; |
53b381b3 DW |
8527 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8528 | &root->fs_info->mapping_tree, | |
8529 | chunk_offset); | |
96303081 | 8530 | |
d2fb3437 | 8531 | atomic_set(&cache->count, 1); |
c286ac48 | 8532 | spin_lock_init(&cache->lock); |
0f9dd46c | 8533 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8534 | INIT_LIST_HEAD(&cache->cluster_list); |
ea658bad | 8535 | INIT_LIST_HEAD(&cache->new_bg_list); |
0ef3e66b | 8536 | |
34d52cb6 LZ |
8537 | btrfs_init_free_space_ctl(cache); |
8538 | ||
6324fbf3 | 8539 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
8540 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
8541 | cache->flags = type; | |
8542 | btrfs_set_block_group_flags(&cache->item, type); | |
8543 | ||
11833d66 | 8544 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8545 | cache->cached = BTRFS_CACHE_FINISHED; |
835d974f JB |
8546 | ret = exclude_super_stripes(root, cache); |
8547 | if (ret) { | |
8548 | /* | |
8549 | * We may have excluded something, so call this just in | |
8550 | * case. | |
8551 | */ | |
8552 | free_excluded_extents(root, cache); | |
8553 | kfree(cache->free_space_ctl); | |
8554 | kfree(cache); | |
8555 | return ret; | |
8556 | } | |
96303081 | 8557 | |
817d52f8 JB |
8558 | add_new_free_space(cache, root->fs_info, chunk_offset, |
8559 | chunk_offset + size); | |
8560 | ||
11833d66 YZ |
8561 | free_excluded_extents(root, cache); |
8562 | ||
8c579fe7 JB |
8563 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8564 | if (ret) { | |
8565 | btrfs_remove_free_space_cache(cache); | |
8566 | btrfs_put_block_group(cache); | |
8567 | return ret; | |
8568 | } | |
8569 | ||
6324fbf3 CM |
8570 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
8571 | &cache->space_info); | |
8c579fe7 JB |
8572 | if (ret) { |
8573 | btrfs_remove_free_space_cache(cache); | |
8574 | spin_lock(&root->fs_info->block_group_cache_lock); | |
8575 | rb_erase(&cache->cache_node, | |
8576 | &root->fs_info->block_group_cache_tree); | |
8577 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
8578 | btrfs_put_block_group(cache); | |
8579 | return ret; | |
8580 | } | |
c7c144db | 8581 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
8582 | |
8583 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 8584 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8585 | spin_unlock(&cache->space_info->lock); |
8586 | ||
b742bb82 | 8587 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 8588 | |
ea658bad | 8589 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 8590 | |
d18a2c44 | 8591 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 8592 | |
6324fbf3 CM |
8593 | return 0; |
8594 | } | |
1a40e23b | 8595 | |
10ea00f5 ID |
8596 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
8597 | { | |
899c81ea ID |
8598 | u64 extra_flags = chunk_to_extended(flags) & |
8599 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 8600 | |
de98ced9 | 8601 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8602 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
8603 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
8604 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
8605 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
8606 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
8607 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 8608 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8609 | } |
8610 | ||
1a40e23b ZY |
8611 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
8612 | struct btrfs_root *root, u64 group_start) | |
8613 | { | |
8614 | struct btrfs_path *path; | |
8615 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8616 | struct btrfs_free_cluster *cluster; |
0af3d00b | 8617 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 8618 | struct btrfs_key key; |
0af3d00b | 8619 | struct inode *inode; |
1a40e23b | 8620 | int ret; |
10ea00f5 | 8621 | int index; |
89a55897 | 8622 | int factor; |
1a40e23b | 8623 | |
1a40e23b ZY |
8624 | root = root->fs_info->extent_root; |
8625 | ||
8626 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8627 | BUG_ON(!block_group); | |
c146afad | 8628 | BUG_ON(!block_group->ro); |
1a40e23b | 8629 | |
9f7c43c9 | 8630 | /* |
8631 | * Free the reserved super bytes from this block group before | |
8632 | * remove it. | |
8633 | */ | |
8634 | free_excluded_extents(root, block_group); | |
8635 | ||
1a40e23b | 8636 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 8637 | index = get_block_group_index(block_group); |
89a55897 JB |
8638 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8639 | BTRFS_BLOCK_GROUP_RAID1 | | |
8640 | BTRFS_BLOCK_GROUP_RAID10)) | |
8641 | factor = 2; | |
8642 | else | |
8643 | factor = 1; | |
1a40e23b | 8644 | |
44fb5511 CM |
8645 | /* make sure this block group isn't part of an allocation cluster */ |
8646 | cluster = &root->fs_info->data_alloc_cluster; | |
8647 | spin_lock(&cluster->refill_lock); | |
8648 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8649 | spin_unlock(&cluster->refill_lock); | |
8650 | ||
8651 | /* | |
8652 | * make sure this block group isn't part of a metadata | |
8653 | * allocation cluster | |
8654 | */ | |
8655 | cluster = &root->fs_info->meta_alloc_cluster; | |
8656 | spin_lock(&cluster->refill_lock); | |
8657 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8658 | spin_unlock(&cluster->refill_lock); | |
8659 | ||
1a40e23b | 8660 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8661 | if (!path) { |
8662 | ret = -ENOMEM; | |
8663 | goto out; | |
8664 | } | |
1a40e23b | 8665 | |
10b2f34d | 8666 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 8667 | if (!IS_ERR(inode)) { |
b532402e | 8668 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
8669 | if (ret) { |
8670 | btrfs_add_delayed_iput(inode); | |
8671 | goto out; | |
8672 | } | |
0af3d00b JB |
8673 | clear_nlink(inode); |
8674 | /* One for the block groups ref */ | |
8675 | spin_lock(&block_group->lock); | |
8676 | if (block_group->iref) { | |
8677 | block_group->iref = 0; | |
8678 | block_group->inode = NULL; | |
8679 | spin_unlock(&block_group->lock); | |
8680 | iput(inode); | |
8681 | } else { | |
8682 | spin_unlock(&block_group->lock); | |
8683 | } | |
8684 | /* One for our lookup ref */ | |
455757c3 | 8685 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
8686 | } |
8687 | ||
8688 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8689 | key.offset = block_group->key.objectid; | |
8690 | key.type = 0; | |
8691 | ||
8692 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8693 | if (ret < 0) | |
8694 | goto out; | |
8695 | if (ret > 0) | |
b3b4aa74 | 8696 | btrfs_release_path(path); |
0af3d00b JB |
8697 | if (ret == 0) { |
8698 | ret = btrfs_del_item(trans, tree_root, path); | |
8699 | if (ret) | |
8700 | goto out; | |
b3b4aa74 | 8701 | btrfs_release_path(path); |
0af3d00b JB |
8702 | } |
8703 | ||
3dfdb934 | 8704 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8705 | rb_erase(&block_group->cache_node, |
8706 | &root->fs_info->block_group_cache_tree); | |
a1897fdd LB |
8707 | |
8708 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
8709 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 8710 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8711 | |
80eb234a | 8712 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8713 | /* |
8714 | * we must use list_del_init so people can check to see if they | |
8715 | * are still on the list after taking the semaphore | |
8716 | */ | |
8717 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8718 | if (list_empty(&block_group->space_info->block_groups[index])) |
8719 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8720 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8721 | |
817d52f8 | 8722 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8723 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8724 | |
8725 | btrfs_remove_free_space_cache(block_group); | |
8726 | ||
c146afad YZ |
8727 | spin_lock(&block_group->space_info->lock); |
8728 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8729 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8730 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8731 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8732 | |
0af3d00b JB |
8733 | memcpy(&key, &block_group->key, sizeof(key)); |
8734 | ||
283bb197 | 8735 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8736 | |
fa9c0d79 CM |
8737 | btrfs_put_block_group(block_group); |
8738 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8739 | |
8740 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8741 | if (ret > 0) | |
8742 | ret = -EIO; | |
8743 | if (ret < 0) | |
8744 | goto out; | |
8745 | ||
8746 | ret = btrfs_del_item(trans, root, path); | |
8747 | out: | |
8748 | btrfs_free_path(path); | |
8749 | return ret; | |
8750 | } | |
acce952b | 8751 | |
c59021f8 | 8752 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8753 | { | |
8754 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8755 | struct btrfs_super_block *disk_super; |
8756 | u64 features; | |
8757 | u64 flags; | |
8758 | int mixed = 0; | |
c59021f8 | 8759 | int ret; |
8760 | ||
6c41761f | 8761 | disk_super = fs_info->super_copy; |
1aba86d6 | 8762 | if (!btrfs_super_root(disk_super)) |
8763 | return 1; | |
c59021f8 | 8764 | |
1aba86d6 | 8765 | features = btrfs_super_incompat_flags(disk_super); |
8766 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8767 | mixed = 1; | |
c59021f8 | 8768 | |
1aba86d6 | 8769 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8770 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8771 | if (ret) |
1aba86d6 | 8772 | goto out; |
c59021f8 | 8773 | |
1aba86d6 | 8774 | if (mixed) { |
8775 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8776 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8777 | } else { | |
8778 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8779 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8780 | if (ret) | |
8781 | goto out; | |
8782 | ||
8783 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8784 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8785 | } | |
8786 | out: | |
c59021f8 | 8787 | return ret; |
8788 | } | |
8789 | ||
acce952b | 8790 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8791 | { | |
8792 | return unpin_extent_range(root, start, end); | |
8793 | } | |
8794 | ||
8795 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8796 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8797 | { |
5378e607 | 8798 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8799 | } |
f7039b1d LD |
8800 | |
8801 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8802 | { | |
8803 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8804 | struct btrfs_block_group_cache *cache = NULL; | |
8805 | u64 group_trimmed; | |
8806 | u64 start; | |
8807 | u64 end; | |
8808 | u64 trimmed = 0; | |
2cac13e4 | 8809 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8810 | int ret = 0; |
8811 | ||
2cac13e4 LB |
8812 | /* |
8813 | * try to trim all FS space, our block group may start from non-zero. | |
8814 | */ | |
8815 | if (range->len == total_bytes) | |
8816 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8817 | else | |
8818 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8819 | |
8820 | while (cache) { | |
8821 | if (cache->key.objectid >= (range->start + range->len)) { | |
8822 | btrfs_put_block_group(cache); | |
8823 | break; | |
8824 | } | |
8825 | ||
8826 | start = max(range->start, cache->key.objectid); | |
8827 | end = min(range->start + range->len, | |
8828 | cache->key.objectid + cache->key.offset); | |
8829 | ||
8830 | if (end - start >= range->minlen) { | |
8831 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 8832 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
8833 | if (ret) { |
8834 | btrfs_put_block_group(cache); | |
8835 | break; | |
8836 | } | |
8837 | ret = wait_block_group_cache_done(cache); | |
8838 | if (ret) { | |
8839 | btrfs_put_block_group(cache); | |
8840 | break; | |
8841 | } | |
f7039b1d LD |
8842 | } |
8843 | ret = btrfs_trim_block_group(cache, | |
8844 | &group_trimmed, | |
8845 | start, | |
8846 | end, | |
8847 | range->minlen); | |
8848 | ||
8849 | trimmed += group_trimmed; | |
8850 | if (ret) { | |
8851 | btrfs_put_block_group(cache); | |
8852 | break; | |
8853 | } | |
8854 | } | |
8855 | ||
8856 | cache = next_block_group(fs_info->tree_root, cache); | |
8857 | } | |
8858 | ||
8859 | range->len = trimmed; | |
8860 | return ret; | |
8861 | } |