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> |
74493f7a | 28 | #include "hash.h" |
995946dd | 29 | #include "tree-log.h" |
fec577fb CM |
30 | #include "disk-io.h" |
31 | #include "print-tree.h" | |
0b86a832 | 32 | #include "volumes.h" |
53b381b3 | 33 | #include "raid56.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
1e144fb8 | 36 | #include "free-space-tree.h" |
3fed40cc | 37 | #include "math.h" |
6ab0a202 | 38 | #include "sysfs.h" |
fcebe456 | 39 | #include "qgroup.h" |
fec577fb | 40 | |
709c0486 AJ |
41 | #undef SCRAMBLE_DELAYED_REFS |
42 | ||
9e622d6b MX |
43 | /* |
44 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
45 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
46 | * if we really need one. | |
47 | * | |
0e4f8f88 CM |
48 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
49 | * if we have very few chunks already allocated. This is | |
50 | * used as part of the clustering code to help make sure | |
51 | * we have a good pool of storage to cluster in, without | |
52 | * filling the FS with empty chunks | |
53 | * | |
9e622d6b MX |
54 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
55 | * | |
0e4f8f88 CM |
56 | */ |
57 | enum { | |
58 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
59 | CHUNK_ALLOC_LIMITED = 1, |
60 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
61 | }; |
62 | ||
fb25e914 JB |
63 | /* |
64 | * Control how reservations are dealt with. | |
65 | * | |
66 | * RESERVE_FREE - freeing a reservation. | |
67 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
68 | * ENOSPC accounting | |
69 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
70 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
71 | */ | |
72 | enum { | |
73 | RESERVE_FREE = 0, | |
74 | RESERVE_ALLOC = 1, | |
75 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
76 | }; | |
77 | ||
ce93ec54 JB |
78 | static int update_block_group(struct btrfs_trans_handle *trans, |
79 | struct btrfs_root *root, u64 bytenr, | |
80 | u64 num_bytes, int alloc); | |
5d4f98a2 YZ |
81 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
82 | struct btrfs_root *root, | |
c682f9b3 | 83 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
84 | u64 root_objectid, u64 owner_objectid, |
85 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 86 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
87 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
88 | struct extent_buffer *leaf, | |
89 | struct btrfs_extent_item *ei); | |
90 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
91 | struct btrfs_root *root, | |
92 | u64 parent, u64 root_objectid, | |
93 | u64 flags, u64 owner, u64 offset, | |
94 | struct btrfs_key *ins, int ref_mod); | |
95 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
96 | struct btrfs_root *root, | |
97 | u64 parent, u64 root_objectid, | |
98 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 99 | int level, struct btrfs_key *ins); |
6a63209f | 100 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
101 | struct btrfs_root *extent_root, u64 flags, |
102 | int force); | |
11833d66 YZ |
103 | static int find_next_key(struct btrfs_path *path, int level, |
104 | struct btrfs_key *key); | |
9ed74f2d JB |
105 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
106 | int dump_block_groups); | |
fb25e914 | 107 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 MX |
108 | u64 num_bytes, int reserve, |
109 | int delalloc); | |
5d80366e JB |
110 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
111 | u64 num_bytes); | |
48a3b636 ES |
112 | int btrfs_pin_extent(struct btrfs_root *root, |
113 | u64 bytenr, u64 num_bytes, int reserved); | |
6a63209f | 114 | |
817d52f8 JB |
115 | static noinline int |
116 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
117 | { | |
118 | smp_mb(); | |
36cce922 JB |
119 | return cache->cached == BTRFS_CACHE_FINISHED || |
120 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
121 | } |
122 | ||
0f9dd46c JB |
123 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
124 | { | |
125 | return (cache->flags & bits) == bits; | |
126 | } | |
127 | ||
758f2dfc | 128 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
129 | { |
130 | atomic_inc(&cache->count); | |
131 | } | |
132 | ||
133 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
134 | { | |
f0486c68 YZ |
135 | if (atomic_dec_and_test(&cache->count)) { |
136 | WARN_ON(cache->pinned > 0); | |
137 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 138 | kfree(cache->free_space_ctl); |
11dfe35a | 139 | kfree(cache); |
f0486c68 | 140 | } |
11dfe35a JB |
141 | } |
142 | ||
0f9dd46c JB |
143 | /* |
144 | * this adds the block group to the fs_info rb tree for the block group | |
145 | * cache | |
146 | */ | |
b2950863 | 147 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
148 | struct btrfs_block_group_cache *block_group) |
149 | { | |
150 | struct rb_node **p; | |
151 | struct rb_node *parent = NULL; | |
152 | struct btrfs_block_group_cache *cache; | |
153 | ||
154 | spin_lock(&info->block_group_cache_lock); | |
155 | p = &info->block_group_cache_tree.rb_node; | |
156 | ||
157 | while (*p) { | |
158 | parent = *p; | |
159 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
160 | cache_node); | |
161 | if (block_group->key.objectid < cache->key.objectid) { | |
162 | p = &(*p)->rb_left; | |
163 | } else if (block_group->key.objectid > cache->key.objectid) { | |
164 | p = &(*p)->rb_right; | |
165 | } else { | |
166 | spin_unlock(&info->block_group_cache_lock); | |
167 | return -EEXIST; | |
168 | } | |
169 | } | |
170 | ||
171 | rb_link_node(&block_group->cache_node, parent, p); | |
172 | rb_insert_color(&block_group->cache_node, | |
173 | &info->block_group_cache_tree); | |
a1897fdd LB |
174 | |
175 | if (info->first_logical_byte > block_group->key.objectid) | |
176 | info->first_logical_byte = block_group->key.objectid; | |
177 | ||
0f9dd46c JB |
178 | spin_unlock(&info->block_group_cache_lock); |
179 | ||
180 | return 0; | |
181 | } | |
182 | ||
183 | /* | |
184 | * This will return the block group at or after bytenr if contains is 0, else | |
185 | * it will return the block group that contains the bytenr | |
186 | */ | |
187 | static struct btrfs_block_group_cache * | |
188 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
189 | int contains) | |
190 | { | |
191 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
192 | struct rb_node *n; | |
193 | u64 end, start; | |
194 | ||
195 | spin_lock(&info->block_group_cache_lock); | |
196 | n = info->block_group_cache_tree.rb_node; | |
197 | ||
198 | while (n) { | |
199 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
200 | cache_node); | |
201 | end = cache->key.objectid + cache->key.offset - 1; | |
202 | start = cache->key.objectid; | |
203 | ||
204 | if (bytenr < start) { | |
205 | if (!contains && (!ret || start < ret->key.objectid)) | |
206 | ret = cache; | |
207 | n = n->rb_left; | |
208 | } else if (bytenr > start) { | |
209 | if (contains && bytenr <= end) { | |
210 | ret = cache; | |
211 | break; | |
212 | } | |
213 | n = n->rb_right; | |
214 | } else { | |
215 | ret = cache; | |
216 | break; | |
217 | } | |
218 | } | |
a1897fdd | 219 | if (ret) { |
11dfe35a | 220 | btrfs_get_block_group(ret); |
a1897fdd LB |
221 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
222 | info->first_logical_byte = ret->key.objectid; | |
223 | } | |
0f9dd46c JB |
224 | spin_unlock(&info->block_group_cache_lock); |
225 | ||
226 | return ret; | |
227 | } | |
228 | ||
11833d66 YZ |
229 | static int add_excluded_extent(struct btrfs_root *root, |
230 | u64 start, u64 num_bytes) | |
817d52f8 | 231 | { |
11833d66 YZ |
232 | u64 end = start + num_bytes - 1; |
233 | set_extent_bits(&root->fs_info->freed_extents[0], | |
ceeb0ae7 | 234 | start, end, EXTENT_UPTODATE); |
11833d66 | 235 | set_extent_bits(&root->fs_info->freed_extents[1], |
ceeb0ae7 | 236 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
237 | return 0; |
238 | } | |
817d52f8 | 239 | |
11833d66 YZ |
240 | static void free_excluded_extents(struct btrfs_root *root, |
241 | struct btrfs_block_group_cache *cache) | |
242 | { | |
243 | u64 start, end; | |
817d52f8 | 244 | |
11833d66 YZ |
245 | start = cache->key.objectid; |
246 | end = start + cache->key.offset - 1; | |
247 | ||
248 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
91166212 | 249 | start, end, EXTENT_UPTODATE); |
11833d66 | 250 | clear_extent_bits(&root->fs_info->freed_extents[1], |
91166212 | 251 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
252 | } |
253 | ||
11833d66 YZ |
254 | static int exclude_super_stripes(struct btrfs_root *root, |
255 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 256 | { |
817d52f8 JB |
257 | u64 bytenr; |
258 | u64 *logical; | |
259 | int stripe_len; | |
260 | int i, nr, ret; | |
261 | ||
06b2331f YZ |
262 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
263 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
264 | cache->bytes_super += stripe_len; | |
265 | ret = add_excluded_extent(root, cache->key.objectid, | |
266 | stripe_len); | |
835d974f JB |
267 | if (ret) |
268 | return ret; | |
06b2331f YZ |
269 | } |
270 | ||
817d52f8 JB |
271 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
272 | bytenr = btrfs_sb_offset(i); | |
273 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
274 | cache->key.objectid, bytenr, | |
275 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
276 | if (ret) |
277 | return ret; | |
11833d66 | 278 | |
817d52f8 | 279 | while (nr--) { |
51bf5f0b JB |
280 | u64 start, len; |
281 | ||
282 | if (logical[nr] > cache->key.objectid + | |
283 | cache->key.offset) | |
284 | continue; | |
285 | ||
286 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
287 | continue; | |
288 | ||
289 | start = logical[nr]; | |
290 | if (start < cache->key.objectid) { | |
291 | start = cache->key.objectid; | |
292 | len = (logical[nr] + stripe_len) - start; | |
293 | } else { | |
294 | len = min_t(u64, stripe_len, | |
295 | cache->key.objectid + | |
296 | cache->key.offset - start); | |
297 | } | |
298 | ||
299 | cache->bytes_super += len; | |
300 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
301 | if (ret) { |
302 | kfree(logical); | |
303 | return ret; | |
304 | } | |
817d52f8 | 305 | } |
11833d66 | 306 | |
817d52f8 JB |
307 | kfree(logical); |
308 | } | |
817d52f8 JB |
309 | return 0; |
310 | } | |
311 | ||
11833d66 YZ |
312 | static struct btrfs_caching_control * |
313 | get_caching_control(struct btrfs_block_group_cache *cache) | |
314 | { | |
315 | struct btrfs_caching_control *ctl; | |
316 | ||
317 | spin_lock(&cache->lock); | |
dde5abee JB |
318 | if (!cache->caching_ctl) { |
319 | spin_unlock(&cache->lock); | |
11833d66 YZ |
320 | return NULL; |
321 | } | |
322 | ||
323 | ctl = cache->caching_ctl; | |
324 | atomic_inc(&ctl->count); | |
325 | spin_unlock(&cache->lock); | |
326 | return ctl; | |
327 | } | |
328 | ||
329 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
330 | { | |
331 | if (atomic_dec_and_test(&ctl->count)) | |
332 | kfree(ctl); | |
333 | } | |
334 | ||
d0bd4560 JB |
335 | #ifdef CONFIG_BTRFS_DEBUG |
336 | static void fragment_free_space(struct btrfs_root *root, | |
337 | struct btrfs_block_group_cache *block_group) | |
338 | { | |
339 | u64 start = block_group->key.objectid; | |
340 | u64 len = block_group->key.offset; | |
341 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
342 | root->nodesize : root->sectorsize; | |
343 | u64 step = chunk << 1; | |
344 | ||
345 | while (len > chunk) { | |
346 | btrfs_remove_free_space(block_group, start, chunk); | |
347 | start += step; | |
348 | if (len < step) | |
349 | len = 0; | |
350 | else | |
351 | len -= step; | |
352 | } | |
353 | } | |
354 | #endif | |
355 | ||
0f9dd46c JB |
356 | /* |
357 | * this is only called by cache_block_group, since we could have freed extents | |
358 | * we need to check the pinned_extents for any extents that can't be used yet | |
359 | * since their free space will be released as soon as the transaction commits. | |
360 | */ | |
a5ed9182 OS |
361 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
362 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 363 | { |
817d52f8 | 364 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
365 | int ret; |
366 | ||
367 | while (start < end) { | |
11833d66 | 368 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 369 | &extent_start, &extent_end, |
e6138876 JB |
370 | EXTENT_DIRTY | EXTENT_UPTODATE, |
371 | NULL); | |
0f9dd46c JB |
372 | if (ret) |
373 | break; | |
374 | ||
06b2331f | 375 | if (extent_start <= start) { |
0f9dd46c JB |
376 | start = extent_end + 1; |
377 | } else if (extent_start > start && extent_start < end) { | |
378 | size = extent_start - start; | |
817d52f8 | 379 | total_added += size; |
ea6a478e JB |
380 | ret = btrfs_add_free_space(block_group, start, |
381 | size); | |
79787eaa | 382 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
383 | start = extent_end + 1; |
384 | } else { | |
385 | break; | |
386 | } | |
387 | } | |
388 | ||
389 | if (start < end) { | |
390 | size = end - start; | |
817d52f8 | 391 | total_added += size; |
ea6a478e | 392 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 393 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
394 | } |
395 | ||
817d52f8 | 396 | return total_added; |
0f9dd46c JB |
397 | } |
398 | ||
73fa48b6 | 399 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 400 | { |
bab39bf9 JB |
401 | struct btrfs_block_group_cache *block_group; |
402 | struct btrfs_fs_info *fs_info; | |
bab39bf9 | 403 | struct btrfs_root *extent_root; |
e37c9e69 | 404 | struct btrfs_path *path; |
5f39d397 | 405 | struct extent_buffer *leaf; |
11833d66 | 406 | struct btrfs_key key; |
817d52f8 | 407 | u64 total_found = 0; |
11833d66 YZ |
408 | u64 last = 0; |
409 | u32 nritems; | |
73fa48b6 | 410 | int ret; |
d0bd4560 | 411 | bool wakeup = true; |
f510cfec | 412 | |
bab39bf9 JB |
413 | block_group = caching_ctl->block_group; |
414 | fs_info = block_group->fs_info; | |
415 | extent_root = fs_info->extent_root; | |
416 | ||
e37c9e69 CM |
417 | path = btrfs_alloc_path(); |
418 | if (!path) | |
73fa48b6 | 419 | return -ENOMEM; |
7d7d6068 | 420 | |
817d52f8 | 421 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 422 | |
d0bd4560 JB |
423 | #ifdef CONFIG_BTRFS_DEBUG |
424 | /* | |
425 | * If we're fragmenting we don't want to make anybody think we can | |
426 | * allocate from this block group until we've had a chance to fragment | |
427 | * the free space. | |
428 | */ | |
429 | if (btrfs_should_fragment_free_space(extent_root, block_group)) | |
430 | wakeup = false; | |
431 | #endif | |
5cd57b2c | 432 | /* |
817d52f8 JB |
433 | * We don't want to deadlock with somebody trying to allocate a new |
434 | * extent for the extent root while also trying to search the extent | |
435 | * root to add free space. So we skip locking and search the commit | |
436 | * root, since its read-only | |
5cd57b2c CM |
437 | */ |
438 | path->skip_locking = 1; | |
817d52f8 | 439 | path->search_commit_root = 1; |
e4058b54 | 440 | path->reada = READA_FORWARD; |
817d52f8 | 441 | |
e4404d6e | 442 | key.objectid = last; |
e37c9e69 | 443 | key.offset = 0; |
11833d66 | 444 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 445 | |
52ee28d2 | 446 | next: |
11833d66 | 447 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 448 | if (ret < 0) |
73fa48b6 | 449 | goto out; |
a512bbf8 | 450 | |
11833d66 YZ |
451 | leaf = path->nodes[0]; |
452 | nritems = btrfs_header_nritems(leaf); | |
453 | ||
d397712b | 454 | while (1) { |
7841cb28 | 455 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 456 | last = (u64)-1; |
817d52f8 | 457 | break; |
f25784b3 | 458 | } |
817d52f8 | 459 | |
11833d66 YZ |
460 | if (path->slots[0] < nritems) { |
461 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
462 | } else { | |
463 | ret = find_next_key(path, 0, &key); | |
464 | if (ret) | |
e37c9e69 | 465 | break; |
817d52f8 | 466 | |
c9ea7b24 | 467 | if (need_resched() || |
9e351cc8 | 468 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
469 | if (wakeup) |
470 | caching_ctl->progress = last; | |
ff5714cc | 471 | btrfs_release_path(path); |
9e351cc8 | 472 | up_read(&fs_info->commit_root_sem); |
589d8ade | 473 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 474 | cond_resched(); |
73fa48b6 OS |
475 | mutex_lock(&caching_ctl->mutex); |
476 | down_read(&fs_info->commit_root_sem); | |
477 | goto next; | |
589d8ade | 478 | } |
0a3896d0 JB |
479 | |
480 | ret = btrfs_next_leaf(extent_root, path); | |
481 | if (ret < 0) | |
73fa48b6 | 482 | goto out; |
0a3896d0 JB |
483 | if (ret) |
484 | break; | |
589d8ade JB |
485 | leaf = path->nodes[0]; |
486 | nritems = btrfs_header_nritems(leaf); | |
487 | continue; | |
11833d66 | 488 | } |
817d52f8 | 489 | |
52ee28d2 LB |
490 | if (key.objectid < last) { |
491 | key.objectid = last; | |
492 | key.offset = 0; | |
493 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
494 | ||
d0bd4560 JB |
495 | if (wakeup) |
496 | caching_ctl->progress = last; | |
52ee28d2 LB |
497 | btrfs_release_path(path); |
498 | goto next; | |
499 | } | |
500 | ||
11833d66 YZ |
501 | if (key.objectid < block_group->key.objectid) { |
502 | path->slots[0]++; | |
817d52f8 | 503 | continue; |
e37c9e69 | 504 | } |
0f9dd46c | 505 | |
e37c9e69 | 506 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 507 | block_group->key.offset) |
e37c9e69 | 508 | break; |
7d7d6068 | 509 | |
3173a18f JB |
510 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
511 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
512 | total_found += add_new_free_space(block_group, |
513 | fs_info, last, | |
514 | key.objectid); | |
3173a18f JB |
515 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
516 | last = key.objectid + | |
707e8a07 | 517 | fs_info->tree_root->nodesize; |
3173a18f JB |
518 | else |
519 | last = key.objectid + key.offset; | |
817d52f8 | 520 | |
73fa48b6 | 521 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 522 | total_found = 0; |
d0bd4560 JB |
523 | if (wakeup) |
524 | wake_up(&caching_ctl->wait); | |
11833d66 | 525 | } |
817d52f8 | 526 | } |
e37c9e69 CM |
527 | path->slots[0]++; |
528 | } | |
817d52f8 | 529 | ret = 0; |
e37c9e69 | 530 | |
817d52f8 JB |
531 | total_found += add_new_free_space(block_group, fs_info, last, |
532 | block_group->key.objectid + | |
533 | block_group->key.offset); | |
11833d66 | 534 | caching_ctl->progress = (u64)-1; |
817d52f8 | 535 | |
73fa48b6 OS |
536 | out: |
537 | btrfs_free_path(path); | |
538 | return ret; | |
539 | } | |
540 | ||
541 | static noinline void caching_thread(struct btrfs_work *work) | |
542 | { | |
543 | struct btrfs_block_group_cache *block_group; | |
544 | struct btrfs_fs_info *fs_info; | |
545 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 546 | struct btrfs_root *extent_root; |
73fa48b6 OS |
547 | int ret; |
548 | ||
549 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
550 | block_group = caching_ctl->block_group; | |
551 | fs_info = block_group->fs_info; | |
b4570aa9 | 552 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
553 | |
554 | mutex_lock(&caching_ctl->mutex); | |
555 | down_read(&fs_info->commit_root_sem); | |
556 | ||
1e144fb8 OS |
557 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
558 | ret = load_free_space_tree(caching_ctl); | |
559 | else | |
560 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 561 | |
817d52f8 | 562 | spin_lock(&block_group->lock); |
11833d66 | 563 | block_group->caching_ctl = NULL; |
73fa48b6 | 564 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 565 | spin_unlock(&block_group->lock); |
0f9dd46c | 566 | |
d0bd4560 JB |
567 | #ifdef CONFIG_BTRFS_DEBUG |
568 | if (btrfs_should_fragment_free_space(extent_root, block_group)) { | |
569 | u64 bytes_used; | |
570 | ||
571 | spin_lock(&block_group->space_info->lock); | |
572 | spin_lock(&block_group->lock); | |
573 | bytes_used = block_group->key.offset - | |
574 | btrfs_block_group_used(&block_group->item); | |
575 | block_group->space_info->bytes_used += bytes_used >> 1; | |
576 | spin_unlock(&block_group->lock); | |
577 | spin_unlock(&block_group->space_info->lock); | |
578 | fragment_free_space(extent_root, block_group); | |
579 | } | |
580 | #endif | |
581 | ||
582 | caching_ctl->progress = (u64)-1; | |
11833d66 | 583 | |
9e351cc8 | 584 | up_read(&fs_info->commit_root_sem); |
73fa48b6 | 585 | free_excluded_extents(fs_info->extent_root, block_group); |
11833d66 | 586 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 587 | |
11833d66 YZ |
588 | wake_up(&caching_ctl->wait); |
589 | ||
590 | put_caching_control(caching_ctl); | |
11dfe35a | 591 | btrfs_put_block_group(block_group); |
817d52f8 JB |
592 | } |
593 | ||
9d66e233 | 594 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 595 | int load_cache_only) |
817d52f8 | 596 | { |
291c7d2f | 597 | DEFINE_WAIT(wait); |
11833d66 YZ |
598 | struct btrfs_fs_info *fs_info = cache->fs_info; |
599 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
600 | int ret = 0; |
601 | ||
291c7d2f | 602 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
603 | if (!caching_ctl) |
604 | return -ENOMEM; | |
291c7d2f JB |
605 | |
606 | INIT_LIST_HEAD(&caching_ctl->list); | |
607 | mutex_init(&caching_ctl->mutex); | |
608 | init_waitqueue_head(&caching_ctl->wait); | |
609 | caching_ctl->block_group = cache; | |
610 | caching_ctl->progress = cache->key.objectid; | |
611 | atomic_set(&caching_ctl->count, 1); | |
9e0af237 LB |
612 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
613 | caching_thread, NULL, NULL); | |
291c7d2f JB |
614 | |
615 | spin_lock(&cache->lock); | |
616 | /* | |
617 | * This should be a rare occasion, but this could happen I think in the | |
618 | * case where one thread starts to load the space cache info, and then | |
619 | * some other thread starts a transaction commit which tries to do an | |
620 | * allocation while the other thread is still loading the space cache | |
621 | * info. The previous loop should have kept us from choosing this block | |
622 | * group, but if we've moved to the state where we will wait on caching | |
623 | * block groups we need to first check if we're doing a fast load here, | |
624 | * so we can wait for it to finish, otherwise we could end up allocating | |
625 | * from a block group who's cache gets evicted for one reason or | |
626 | * another. | |
627 | */ | |
628 | while (cache->cached == BTRFS_CACHE_FAST) { | |
629 | struct btrfs_caching_control *ctl; | |
630 | ||
631 | ctl = cache->caching_ctl; | |
632 | atomic_inc(&ctl->count); | |
633 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
634 | spin_unlock(&cache->lock); | |
635 | ||
636 | schedule(); | |
637 | ||
638 | finish_wait(&ctl->wait, &wait); | |
639 | put_caching_control(ctl); | |
640 | spin_lock(&cache->lock); | |
641 | } | |
642 | ||
643 | if (cache->cached != BTRFS_CACHE_NO) { | |
644 | spin_unlock(&cache->lock); | |
645 | kfree(caching_ctl); | |
11833d66 | 646 | return 0; |
291c7d2f JB |
647 | } |
648 | WARN_ON(cache->caching_ctl); | |
649 | cache->caching_ctl = caching_ctl; | |
650 | cache->cached = BTRFS_CACHE_FAST; | |
651 | spin_unlock(&cache->lock); | |
11833d66 | 652 | |
d53ba474 | 653 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 654 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
655 | ret = load_free_space_cache(fs_info, cache); |
656 | ||
657 | spin_lock(&cache->lock); | |
658 | if (ret == 1) { | |
291c7d2f | 659 | cache->caching_ctl = NULL; |
9d66e233 JB |
660 | cache->cached = BTRFS_CACHE_FINISHED; |
661 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 662 | caching_ctl->progress = (u64)-1; |
9d66e233 | 663 | } else { |
291c7d2f JB |
664 | if (load_cache_only) { |
665 | cache->caching_ctl = NULL; | |
666 | cache->cached = BTRFS_CACHE_NO; | |
667 | } else { | |
668 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 669 | cache->has_caching_ctl = 1; |
291c7d2f | 670 | } |
9d66e233 JB |
671 | } |
672 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
673 | #ifdef CONFIG_BTRFS_DEBUG |
674 | if (ret == 1 && | |
675 | btrfs_should_fragment_free_space(fs_info->extent_root, | |
676 | cache)) { | |
677 | u64 bytes_used; | |
678 | ||
679 | spin_lock(&cache->space_info->lock); | |
680 | spin_lock(&cache->lock); | |
681 | bytes_used = cache->key.offset - | |
682 | btrfs_block_group_used(&cache->item); | |
683 | cache->space_info->bytes_used += bytes_used >> 1; | |
684 | spin_unlock(&cache->lock); | |
685 | spin_unlock(&cache->space_info->lock); | |
686 | fragment_free_space(fs_info->extent_root, cache); | |
687 | } | |
688 | #endif | |
cb83b7b8 JB |
689 | mutex_unlock(&caching_ctl->mutex); |
690 | ||
291c7d2f | 691 | wake_up(&caching_ctl->wait); |
3c14874a | 692 | if (ret == 1) { |
291c7d2f | 693 | put_caching_control(caching_ctl); |
3c14874a | 694 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 695 | return 0; |
3c14874a | 696 | } |
291c7d2f JB |
697 | } else { |
698 | /* | |
1e144fb8 OS |
699 | * We're either using the free space tree or no caching at all. |
700 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
701 | */ |
702 | spin_lock(&cache->lock); | |
703 | if (load_cache_only) { | |
704 | cache->caching_ctl = NULL; | |
705 | cache->cached = BTRFS_CACHE_NO; | |
706 | } else { | |
707 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 708 | cache->has_caching_ctl = 1; |
291c7d2f JB |
709 | } |
710 | spin_unlock(&cache->lock); | |
711 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
712 | } |
713 | ||
291c7d2f JB |
714 | if (load_cache_only) { |
715 | put_caching_control(caching_ctl); | |
11833d66 | 716 | return 0; |
817d52f8 | 717 | } |
817d52f8 | 718 | |
9e351cc8 | 719 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 720 | atomic_inc(&caching_ctl->count); |
11833d66 | 721 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 722 | up_write(&fs_info->commit_root_sem); |
11833d66 | 723 | |
11dfe35a | 724 | btrfs_get_block_group(cache); |
11833d66 | 725 | |
e66f0bb1 | 726 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 727 | |
ef8bbdfe | 728 | return ret; |
e37c9e69 CM |
729 | } |
730 | ||
0f9dd46c JB |
731 | /* |
732 | * return the block group that starts at or after bytenr | |
733 | */ | |
d397712b CM |
734 | static struct btrfs_block_group_cache * |
735 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 736 | { |
0f9dd46c | 737 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 738 | |
0f9dd46c | 739 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 740 | |
0f9dd46c | 741 | return cache; |
0ef3e66b CM |
742 | } |
743 | ||
0f9dd46c | 744 | /* |
9f55684c | 745 | * return the block group that contains the given bytenr |
0f9dd46c | 746 | */ |
d397712b CM |
747 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
748 | struct btrfs_fs_info *info, | |
749 | u64 bytenr) | |
be744175 | 750 | { |
0f9dd46c | 751 | struct btrfs_block_group_cache *cache; |
be744175 | 752 | |
0f9dd46c | 753 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 754 | |
0f9dd46c | 755 | return cache; |
be744175 | 756 | } |
0b86a832 | 757 | |
0f9dd46c JB |
758 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
759 | u64 flags) | |
6324fbf3 | 760 | { |
0f9dd46c | 761 | struct list_head *head = &info->space_info; |
0f9dd46c | 762 | struct btrfs_space_info *found; |
4184ea7f | 763 | |
52ba6929 | 764 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 765 | |
4184ea7f CM |
766 | rcu_read_lock(); |
767 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 768 | if (found->flags & flags) { |
4184ea7f | 769 | rcu_read_unlock(); |
0f9dd46c | 770 | return found; |
4184ea7f | 771 | } |
0f9dd46c | 772 | } |
4184ea7f | 773 | rcu_read_unlock(); |
0f9dd46c | 774 | return NULL; |
6324fbf3 CM |
775 | } |
776 | ||
4184ea7f CM |
777 | /* |
778 | * after adding space to the filesystem, we need to clear the full flags | |
779 | * on all the space infos. | |
780 | */ | |
781 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
782 | { | |
783 | struct list_head *head = &info->space_info; | |
784 | struct btrfs_space_info *found; | |
785 | ||
786 | rcu_read_lock(); | |
787 | list_for_each_entry_rcu(found, head, list) | |
788 | found->full = 0; | |
789 | rcu_read_unlock(); | |
790 | } | |
791 | ||
1a4ed8fd FM |
792 | /* simple helper to search for an existing data extent at a given offset */ |
793 | int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len) | |
e02119d5 CM |
794 | { |
795 | int ret; | |
796 | struct btrfs_key key; | |
31840ae1 | 797 | struct btrfs_path *path; |
e02119d5 | 798 | |
31840ae1 | 799 | path = btrfs_alloc_path(); |
d8926bb3 MF |
800 | if (!path) |
801 | return -ENOMEM; | |
802 | ||
e02119d5 CM |
803 | key.objectid = start; |
804 | key.offset = len; | |
3173a18f | 805 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
806 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
807 | 0, 0); | |
31840ae1 | 808 | btrfs_free_path(path); |
7bb86316 CM |
809 | return ret; |
810 | } | |
811 | ||
a22285a6 | 812 | /* |
3173a18f | 813 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
814 | * |
815 | * the head node for delayed ref is used to store the sum of all the | |
816 | * reference count modifications queued up in the rbtree. the head | |
817 | * node may also store the extent flags to set. This way you can check | |
818 | * to see what the reference count and extent flags would be if all of | |
819 | * the delayed refs are not processed. | |
820 | */ | |
821 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
822 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 823 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
824 | { |
825 | struct btrfs_delayed_ref_head *head; | |
826 | struct btrfs_delayed_ref_root *delayed_refs; | |
827 | struct btrfs_path *path; | |
828 | struct btrfs_extent_item *ei; | |
829 | struct extent_buffer *leaf; | |
830 | struct btrfs_key key; | |
831 | u32 item_size; | |
832 | u64 num_refs; | |
833 | u64 extent_flags; | |
834 | int ret; | |
835 | ||
3173a18f JB |
836 | /* |
837 | * If we don't have skinny metadata, don't bother doing anything | |
838 | * different | |
839 | */ | |
840 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
707e8a07 | 841 | offset = root->nodesize; |
3173a18f JB |
842 | metadata = 0; |
843 | } | |
844 | ||
a22285a6 YZ |
845 | path = btrfs_alloc_path(); |
846 | if (!path) | |
847 | return -ENOMEM; | |
848 | ||
a22285a6 YZ |
849 | if (!trans) { |
850 | path->skip_locking = 1; | |
851 | path->search_commit_root = 1; | |
852 | } | |
639eefc8 FDBM |
853 | |
854 | search_again: | |
855 | key.objectid = bytenr; | |
856 | key.offset = offset; | |
857 | if (metadata) | |
858 | key.type = BTRFS_METADATA_ITEM_KEY; | |
859 | else | |
860 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
861 | ||
a22285a6 YZ |
862 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, |
863 | &key, path, 0, 0); | |
864 | if (ret < 0) | |
865 | goto out_free; | |
866 | ||
3173a18f | 867 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
868 | if (path->slots[0]) { |
869 | path->slots[0]--; | |
870 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
871 | path->slots[0]); | |
872 | if (key.objectid == bytenr && | |
873 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
707e8a07 | 874 | key.offset == root->nodesize) |
74be9510 FDBM |
875 | ret = 0; |
876 | } | |
3173a18f JB |
877 | } |
878 | ||
a22285a6 YZ |
879 | if (ret == 0) { |
880 | leaf = path->nodes[0]; | |
881 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
882 | if (item_size >= sizeof(*ei)) { | |
883 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
884 | struct btrfs_extent_item); | |
885 | num_refs = btrfs_extent_refs(leaf, ei); | |
886 | extent_flags = btrfs_extent_flags(leaf, ei); | |
887 | } else { | |
888 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
889 | struct btrfs_extent_item_v0 *ei0; | |
890 | BUG_ON(item_size != sizeof(*ei0)); | |
891 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
892 | struct btrfs_extent_item_v0); | |
893 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
894 | /* FIXME: this isn't correct for data */ | |
895 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
896 | #else | |
897 | BUG(); | |
898 | #endif | |
899 | } | |
900 | BUG_ON(num_refs == 0); | |
901 | } else { | |
902 | num_refs = 0; | |
903 | extent_flags = 0; | |
904 | ret = 0; | |
905 | } | |
906 | ||
907 | if (!trans) | |
908 | goto out; | |
909 | ||
910 | delayed_refs = &trans->transaction->delayed_refs; | |
911 | spin_lock(&delayed_refs->lock); | |
912 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
913 | if (head) { | |
914 | if (!mutex_trylock(&head->mutex)) { | |
915 | atomic_inc(&head->node.refs); | |
916 | spin_unlock(&delayed_refs->lock); | |
917 | ||
b3b4aa74 | 918 | btrfs_release_path(path); |
a22285a6 | 919 | |
8cc33e5c DS |
920 | /* |
921 | * Mutex was contended, block until it's released and try | |
922 | * again | |
923 | */ | |
a22285a6 YZ |
924 | mutex_lock(&head->mutex); |
925 | mutex_unlock(&head->mutex); | |
926 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 927 | goto search_again; |
a22285a6 | 928 | } |
d7df2c79 | 929 | spin_lock(&head->lock); |
a22285a6 YZ |
930 | if (head->extent_op && head->extent_op->update_flags) |
931 | extent_flags |= head->extent_op->flags_to_set; | |
932 | else | |
933 | BUG_ON(num_refs == 0); | |
934 | ||
935 | num_refs += head->node.ref_mod; | |
d7df2c79 | 936 | spin_unlock(&head->lock); |
a22285a6 YZ |
937 | mutex_unlock(&head->mutex); |
938 | } | |
939 | spin_unlock(&delayed_refs->lock); | |
940 | out: | |
941 | WARN_ON(num_refs == 0); | |
942 | if (refs) | |
943 | *refs = num_refs; | |
944 | if (flags) | |
945 | *flags = extent_flags; | |
946 | out_free: | |
947 | btrfs_free_path(path); | |
948 | return ret; | |
949 | } | |
950 | ||
d8d5f3e1 CM |
951 | /* |
952 | * Back reference rules. Back refs have three main goals: | |
953 | * | |
954 | * 1) differentiate between all holders of references to an extent so that | |
955 | * when a reference is dropped we can make sure it was a valid reference | |
956 | * before freeing the extent. | |
957 | * | |
958 | * 2) Provide enough information to quickly find the holders of an extent | |
959 | * if we notice a given block is corrupted or bad. | |
960 | * | |
961 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
962 | * maintenance. This is actually the same as #2, but with a slightly | |
963 | * different use case. | |
964 | * | |
5d4f98a2 YZ |
965 | * There are two kinds of back refs. The implicit back refs is optimized |
966 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
967 | * back refs of this kind provide information about the block's owner tree | |
968 | * and the pointer's key. These information allow us to find the block by | |
969 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
970 | * referenced by their owner trees. The location of tree block is recorded | |
971 | * in the back refs. Actually the full back refs is generic, and can be | |
972 | * used in all cases the implicit back refs is used. The major shortcoming | |
973 | * of the full back refs is its overhead. Every time a tree block gets | |
974 | * COWed, we have to update back refs entry for all pointers in it. | |
975 | * | |
976 | * For a newly allocated tree block, we use implicit back refs for | |
977 | * pointers in it. This means most tree related operations only involve | |
978 | * implicit back refs. For a tree block created in old transaction, the | |
979 | * only way to drop a reference to it is COW it. So we can detect the | |
980 | * event that tree block loses its owner tree's reference and do the | |
981 | * back refs conversion. | |
982 | * | |
01327610 | 983 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
984 | * |
985 | * The reference count of the block is one and the tree is the block's | |
986 | * owner tree. Nothing to do in this case. | |
987 | * | |
988 | * The reference count of the block is one and the tree is not the | |
989 | * block's owner tree. In this case, full back refs is used for pointers | |
990 | * in the block. Remove these full back refs, add implicit back refs for | |
991 | * every pointers in the new block. | |
992 | * | |
993 | * The reference count of the block is greater than one and the tree is | |
994 | * the block's owner tree. In this case, implicit back refs is used for | |
995 | * pointers in the block. Add full back refs for every pointers in the | |
996 | * block, increase lower level extents' reference counts. The original | |
997 | * implicit back refs are entailed to the new block. | |
998 | * | |
999 | * The reference count of the block is greater than one and the tree is | |
1000 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1001 | * the new block, increase lower level extents' reference count. | |
1002 | * | |
1003 | * Back Reference Key composing: | |
1004 | * | |
1005 | * The key objectid corresponds to the first byte in the extent, | |
1006 | * The key type is used to differentiate between types of back refs. | |
1007 | * There are different meanings of the key offset for different types | |
1008 | * of back refs. | |
1009 | * | |
d8d5f3e1 CM |
1010 | * File extents can be referenced by: |
1011 | * | |
1012 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1013 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1014 | * - different offsets inside a file (bookend extents in file.c) |
1015 | * | |
5d4f98a2 | 1016 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1017 | * |
1018 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1019 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1020 | * - original offset in the file |
1021 | * - how many bookend extents | |
d8d5f3e1 | 1022 | * |
5d4f98a2 YZ |
1023 | * The key offset for the implicit back refs is hash of the first |
1024 | * three fields. | |
d8d5f3e1 | 1025 | * |
5d4f98a2 | 1026 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1027 | * |
5d4f98a2 | 1028 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1029 | * |
5d4f98a2 YZ |
1030 | * The key offset for the implicit back refs is the first byte of |
1031 | * the tree leaf | |
d8d5f3e1 | 1032 | * |
5d4f98a2 YZ |
1033 | * When a file extent is allocated, The implicit back refs is used. |
1034 | * the fields are filled in: | |
d8d5f3e1 | 1035 | * |
5d4f98a2 | 1036 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1037 | * |
5d4f98a2 YZ |
1038 | * When a file extent is removed file truncation, we find the |
1039 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1040 | * |
5d4f98a2 | 1041 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1042 | * |
5d4f98a2 | 1043 | * Btree extents can be referenced by: |
d8d5f3e1 | 1044 | * |
5d4f98a2 | 1045 | * - Different subvolumes |
d8d5f3e1 | 1046 | * |
5d4f98a2 YZ |
1047 | * Both the implicit back refs and the full back refs for tree blocks |
1048 | * only consist of key. The key offset for the implicit back refs is | |
1049 | * objectid of block's owner tree. The key offset for the full back refs | |
1050 | * is the first byte of parent block. | |
d8d5f3e1 | 1051 | * |
5d4f98a2 YZ |
1052 | * When implicit back refs is used, information about the lowest key and |
1053 | * level of the tree block are required. These information are stored in | |
1054 | * tree block info structure. | |
d8d5f3e1 | 1055 | */ |
31840ae1 | 1056 | |
5d4f98a2 YZ |
1057 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1058 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1059 | struct btrfs_root *root, | |
1060 | struct btrfs_path *path, | |
1061 | u64 owner, u32 extra_size) | |
7bb86316 | 1062 | { |
5d4f98a2 YZ |
1063 | struct btrfs_extent_item *item; |
1064 | struct btrfs_extent_item_v0 *ei0; | |
1065 | struct btrfs_extent_ref_v0 *ref0; | |
1066 | struct btrfs_tree_block_info *bi; | |
1067 | struct extent_buffer *leaf; | |
7bb86316 | 1068 | struct btrfs_key key; |
5d4f98a2 YZ |
1069 | struct btrfs_key found_key; |
1070 | u32 new_size = sizeof(*item); | |
1071 | u64 refs; | |
1072 | int ret; | |
1073 | ||
1074 | leaf = path->nodes[0]; | |
1075 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1076 | ||
1077 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1078 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1079 | struct btrfs_extent_item_v0); | |
1080 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1081 | ||
1082 | if (owner == (u64)-1) { | |
1083 | while (1) { | |
1084 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1085 | ret = btrfs_next_leaf(root, path); | |
1086 | if (ret < 0) | |
1087 | return ret; | |
79787eaa | 1088 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1089 | leaf = path->nodes[0]; |
1090 | } | |
1091 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1092 | path->slots[0]); | |
1093 | BUG_ON(key.objectid != found_key.objectid); | |
1094 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1095 | path->slots[0]++; | |
1096 | continue; | |
1097 | } | |
1098 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1099 | struct btrfs_extent_ref_v0); | |
1100 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1101 | break; | |
1102 | } | |
1103 | } | |
b3b4aa74 | 1104 | btrfs_release_path(path); |
5d4f98a2 YZ |
1105 | |
1106 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1107 | new_size += sizeof(*bi); | |
1108 | ||
1109 | new_size -= sizeof(*ei0); | |
1110 | ret = btrfs_search_slot(trans, root, &key, path, | |
1111 | new_size + extra_size, 1); | |
1112 | if (ret < 0) | |
1113 | return ret; | |
79787eaa | 1114 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1115 | |
4b90c680 | 1116 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1117 | |
1118 | leaf = path->nodes[0]; | |
1119 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1120 | btrfs_set_extent_refs(leaf, item, refs); | |
1121 | /* FIXME: get real generation */ | |
1122 | btrfs_set_extent_generation(leaf, item, 0); | |
1123 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1124 | btrfs_set_extent_flags(leaf, item, | |
1125 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1126 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1127 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1128 | /* FIXME: get first key of the block */ | |
1129 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1130 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1131 | } else { | |
1132 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1133 | } | |
1134 | btrfs_mark_buffer_dirty(leaf); | |
1135 | return 0; | |
1136 | } | |
1137 | #endif | |
1138 | ||
1139 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1140 | { | |
1141 | u32 high_crc = ~(u32)0; | |
1142 | u32 low_crc = ~(u32)0; | |
1143 | __le64 lenum; | |
1144 | ||
1145 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1146 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1147 | lenum = cpu_to_le64(owner); |
14a958e6 | 1148 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1149 | lenum = cpu_to_le64(offset); |
14a958e6 | 1150 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1151 | |
1152 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1153 | } | |
1154 | ||
1155 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1156 | struct btrfs_extent_data_ref *ref) | |
1157 | { | |
1158 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1159 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1160 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1161 | } | |
1162 | ||
1163 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1164 | struct btrfs_extent_data_ref *ref, | |
1165 | u64 root_objectid, u64 owner, u64 offset) | |
1166 | { | |
1167 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1168 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1169 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1170 | return 0; | |
1171 | return 1; | |
1172 | } | |
1173 | ||
1174 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1175 | struct btrfs_root *root, | |
1176 | struct btrfs_path *path, | |
1177 | u64 bytenr, u64 parent, | |
1178 | u64 root_objectid, | |
1179 | u64 owner, u64 offset) | |
1180 | { | |
1181 | struct btrfs_key key; | |
1182 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1183 | struct extent_buffer *leaf; |
5d4f98a2 | 1184 | u32 nritems; |
74493f7a | 1185 | int ret; |
5d4f98a2 YZ |
1186 | int recow; |
1187 | int err = -ENOENT; | |
74493f7a | 1188 | |
31840ae1 | 1189 | key.objectid = bytenr; |
5d4f98a2 YZ |
1190 | if (parent) { |
1191 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1192 | key.offset = parent; | |
1193 | } else { | |
1194 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1195 | key.offset = hash_extent_data_ref(root_objectid, | |
1196 | owner, offset); | |
1197 | } | |
1198 | again: | |
1199 | recow = 0; | |
1200 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1201 | if (ret < 0) { | |
1202 | err = ret; | |
1203 | goto fail; | |
1204 | } | |
31840ae1 | 1205 | |
5d4f98a2 YZ |
1206 | if (parent) { |
1207 | if (!ret) | |
1208 | return 0; | |
1209 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1210 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1211 | btrfs_release_path(path); |
5d4f98a2 YZ |
1212 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1213 | if (ret < 0) { | |
1214 | err = ret; | |
1215 | goto fail; | |
1216 | } | |
1217 | if (!ret) | |
1218 | return 0; | |
1219 | #endif | |
1220 | goto fail; | |
31840ae1 ZY |
1221 | } |
1222 | ||
1223 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1224 | nritems = btrfs_header_nritems(leaf); |
1225 | while (1) { | |
1226 | if (path->slots[0] >= nritems) { | |
1227 | ret = btrfs_next_leaf(root, path); | |
1228 | if (ret < 0) | |
1229 | err = ret; | |
1230 | if (ret) | |
1231 | goto fail; | |
1232 | ||
1233 | leaf = path->nodes[0]; | |
1234 | nritems = btrfs_header_nritems(leaf); | |
1235 | recow = 1; | |
1236 | } | |
1237 | ||
1238 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1239 | if (key.objectid != bytenr || | |
1240 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1241 | goto fail; | |
1242 | ||
1243 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1244 | struct btrfs_extent_data_ref); | |
1245 | ||
1246 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1247 | owner, offset)) { | |
1248 | if (recow) { | |
b3b4aa74 | 1249 | btrfs_release_path(path); |
5d4f98a2 YZ |
1250 | goto again; |
1251 | } | |
1252 | err = 0; | |
1253 | break; | |
1254 | } | |
1255 | path->slots[0]++; | |
31840ae1 | 1256 | } |
5d4f98a2 YZ |
1257 | fail: |
1258 | return err; | |
31840ae1 ZY |
1259 | } |
1260 | ||
5d4f98a2 YZ |
1261 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1262 | struct btrfs_root *root, | |
1263 | struct btrfs_path *path, | |
1264 | u64 bytenr, u64 parent, | |
1265 | u64 root_objectid, u64 owner, | |
1266 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1267 | { |
1268 | struct btrfs_key key; | |
1269 | struct extent_buffer *leaf; | |
5d4f98a2 | 1270 | u32 size; |
31840ae1 ZY |
1271 | u32 num_refs; |
1272 | int ret; | |
74493f7a | 1273 | |
74493f7a | 1274 | key.objectid = bytenr; |
5d4f98a2 YZ |
1275 | if (parent) { |
1276 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1277 | key.offset = parent; | |
1278 | size = sizeof(struct btrfs_shared_data_ref); | |
1279 | } else { | |
1280 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1281 | key.offset = hash_extent_data_ref(root_objectid, | |
1282 | owner, offset); | |
1283 | size = sizeof(struct btrfs_extent_data_ref); | |
1284 | } | |
74493f7a | 1285 | |
5d4f98a2 YZ |
1286 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1287 | if (ret && ret != -EEXIST) | |
1288 | goto fail; | |
1289 | ||
1290 | leaf = path->nodes[0]; | |
1291 | if (parent) { | |
1292 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1293 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1294 | struct btrfs_shared_data_ref); |
1295 | if (ret == 0) { | |
1296 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1297 | } else { | |
1298 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1299 | num_refs += refs_to_add; | |
1300 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1301 | } |
5d4f98a2 YZ |
1302 | } else { |
1303 | struct btrfs_extent_data_ref *ref; | |
1304 | while (ret == -EEXIST) { | |
1305 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1306 | struct btrfs_extent_data_ref); | |
1307 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1308 | owner, offset)) | |
1309 | break; | |
b3b4aa74 | 1310 | btrfs_release_path(path); |
5d4f98a2 YZ |
1311 | key.offset++; |
1312 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1313 | size); | |
1314 | if (ret && ret != -EEXIST) | |
1315 | goto fail; | |
31840ae1 | 1316 | |
5d4f98a2 YZ |
1317 | leaf = path->nodes[0]; |
1318 | } | |
1319 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1320 | struct btrfs_extent_data_ref); | |
1321 | if (ret == 0) { | |
1322 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1323 | root_objectid); | |
1324 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1325 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1326 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1327 | } else { | |
1328 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1329 | num_refs += refs_to_add; | |
1330 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1331 | } |
31840ae1 | 1332 | } |
5d4f98a2 YZ |
1333 | btrfs_mark_buffer_dirty(leaf); |
1334 | ret = 0; | |
1335 | fail: | |
b3b4aa74 | 1336 | btrfs_release_path(path); |
7bb86316 | 1337 | return ret; |
74493f7a CM |
1338 | } |
1339 | ||
5d4f98a2 YZ |
1340 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1341 | struct btrfs_root *root, | |
1342 | struct btrfs_path *path, | |
fcebe456 | 1343 | int refs_to_drop, int *last_ref) |
31840ae1 | 1344 | { |
5d4f98a2 YZ |
1345 | struct btrfs_key key; |
1346 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1347 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1348 | struct extent_buffer *leaf; |
5d4f98a2 | 1349 | u32 num_refs = 0; |
31840ae1 ZY |
1350 | int ret = 0; |
1351 | ||
1352 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1353 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1354 | ||
1355 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1356 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1357 | struct btrfs_extent_data_ref); | |
1358 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1359 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1360 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1361 | struct btrfs_shared_data_ref); | |
1362 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1363 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1364 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1365 | struct btrfs_extent_ref_v0 *ref0; | |
1366 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1367 | struct btrfs_extent_ref_v0); | |
1368 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1369 | #endif | |
1370 | } else { | |
1371 | BUG(); | |
1372 | } | |
1373 | ||
56bec294 CM |
1374 | BUG_ON(num_refs < refs_to_drop); |
1375 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1376 | |
31840ae1 ZY |
1377 | if (num_refs == 0) { |
1378 | ret = btrfs_del_item(trans, root, path); | |
fcebe456 | 1379 | *last_ref = 1; |
31840ae1 | 1380 | } else { |
5d4f98a2 YZ |
1381 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1382 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1383 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1384 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1385 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1386 | else { | |
1387 | struct btrfs_extent_ref_v0 *ref0; | |
1388 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1389 | struct btrfs_extent_ref_v0); | |
1390 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1391 | } | |
1392 | #endif | |
31840ae1 ZY |
1393 | btrfs_mark_buffer_dirty(leaf); |
1394 | } | |
31840ae1 ZY |
1395 | return ret; |
1396 | } | |
1397 | ||
9ed0dea0 | 1398 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1399 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1400 | { |
5d4f98a2 YZ |
1401 | struct btrfs_key key; |
1402 | struct extent_buffer *leaf; | |
1403 | struct btrfs_extent_data_ref *ref1; | |
1404 | struct btrfs_shared_data_ref *ref2; | |
1405 | u32 num_refs = 0; | |
1406 | ||
1407 | leaf = path->nodes[0]; | |
1408 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1409 | if (iref) { | |
1410 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1411 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1412 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1413 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1414 | } else { | |
1415 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1416 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1417 | } | |
1418 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1419 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1420 | struct btrfs_extent_data_ref); | |
1421 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1422 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1423 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1424 | struct btrfs_shared_data_ref); | |
1425 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1426 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1427 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1428 | struct btrfs_extent_ref_v0 *ref0; | |
1429 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1430 | struct btrfs_extent_ref_v0); | |
1431 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1432 | #endif |
5d4f98a2 YZ |
1433 | } else { |
1434 | WARN_ON(1); | |
1435 | } | |
1436 | return num_refs; | |
1437 | } | |
15916de8 | 1438 | |
5d4f98a2 YZ |
1439 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1440 | struct btrfs_root *root, | |
1441 | struct btrfs_path *path, | |
1442 | u64 bytenr, u64 parent, | |
1443 | u64 root_objectid) | |
1f3c79a2 | 1444 | { |
5d4f98a2 | 1445 | struct btrfs_key key; |
1f3c79a2 | 1446 | int ret; |
1f3c79a2 | 1447 | |
5d4f98a2 YZ |
1448 | key.objectid = bytenr; |
1449 | if (parent) { | |
1450 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1451 | key.offset = parent; | |
1452 | } else { | |
1453 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1454 | key.offset = root_objectid; | |
1f3c79a2 LH |
1455 | } |
1456 | ||
5d4f98a2 YZ |
1457 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1458 | if (ret > 0) | |
1459 | ret = -ENOENT; | |
1460 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1461 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1462 | btrfs_release_path(path); |
5d4f98a2 YZ |
1463 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1464 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1465 | if (ret > 0) | |
1466 | ret = -ENOENT; | |
1467 | } | |
1f3c79a2 | 1468 | #endif |
5d4f98a2 | 1469 | return ret; |
1f3c79a2 LH |
1470 | } |
1471 | ||
5d4f98a2 YZ |
1472 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1473 | struct btrfs_root *root, | |
1474 | struct btrfs_path *path, | |
1475 | u64 bytenr, u64 parent, | |
1476 | u64 root_objectid) | |
31840ae1 | 1477 | { |
5d4f98a2 | 1478 | struct btrfs_key key; |
31840ae1 | 1479 | int ret; |
31840ae1 | 1480 | |
5d4f98a2 YZ |
1481 | key.objectid = bytenr; |
1482 | if (parent) { | |
1483 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1484 | key.offset = parent; | |
1485 | } else { | |
1486 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1487 | key.offset = root_objectid; | |
1488 | } | |
1489 | ||
1490 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1491 | btrfs_release_path(path); |
31840ae1 ZY |
1492 | return ret; |
1493 | } | |
1494 | ||
5d4f98a2 | 1495 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1496 | { |
5d4f98a2 YZ |
1497 | int type; |
1498 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1499 | if (parent > 0) | |
1500 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1501 | else | |
1502 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1503 | } else { | |
1504 | if (parent > 0) | |
1505 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1506 | else | |
1507 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1508 | } | |
1509 | return type; | |
31840ae1 | 1510 | } |
56bec294 | 1511 | |
2c47e605 YZ |
1512 | static int find_next_key(struct btrfs_path *path, int level, |
1513 | struct btrfs_key *key) | |
56bec294 | 1514 | |
02217ed2 | 1515 | { |
2c47e605 | 1516 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1517 | if (!path->nodes[level]) |
1518 | break; | |
5d4f98a2 YZ |
1519 | if (path->slots[level] + 1 >= |
1520 | btrfs_header_nritems(path->nodes[level])) | |
1521 | continue; | |
1522 | if (level == 0) | |
1523 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1524 | path->slots[level] + 1); | |
1525 | else | |
1526 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1527 | path->slots[level] + 1); | |
1528 | return 0; | |
1529 | } | |
1530 | return 1; | |
1531 | } | |
037e6390 | 1532 | |
5d4f98a2 YZ |
1533 | /* |
1534 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1535 | * to the address of inline back ref, and 0 is returned. | |
1536 | * | |
1537 | * if back ref isn't found, *ref_ret is set to the address where it | |
1538 | * should be inserted, and -ENOENT is returned. | |
1539 | * | |
1540 | * if insert is true and there are too many inline back refs, the path | |
1541 | * points to the extent item, and -EAGAIN is returned. | |
1542 | * | |
1543 | * NOTE: inline back refs are ordered in the same way that back ref | |
1544 | * items in the tree are ordered. | |
1545 | */ | |
1546 | static noinline_for_stack | |
1547 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1548 | struct btrfs_root *root, | |
1549 | struct btrfs_path *path, | |
1550 | struct btrfs_extent_inline_ref **ref_ret, | |
1551 | u64 bytenr, u64 num_bytes, | |
1552 | u64 parent, u64 root_objectid, | |
1553 | u64 owner, u64 offset, int insert) | |
1554 | { | |
1555 | struct btrfs_key key; | |
1556 | struct extent_buffer *leaf; | |
1557 | struct btrfs_extent_item *ei; | |
1558 | struct btrfs_extent_inline_ref *iref; | |
1559 | u64 flags; | |
1560 | u64 item_size; | |
1561 | unsigned long ptr; | |
1562 | unsigned long end; | |
1563 | int extra_size; | |
1564 | int type; | |
1565 | int want; | |
1566 | int ret; | |
1567 | int err = 0; | |
3173a18f JB |
1568 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1569 | SKINNY_METADATA); | |
26b8003f | 1570 | |
db94535d | 1571 | key.objectid = bytenr; |
31840ae1 | 1572 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1573 | key.offset = num_bytes; |
31840ae1 | 1574 | |
5d4f98a2 YZ |
1575 | want = extent_ref_type(parent, owner); |
1576 | if (insert) { | |
1577 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1578 | path->keep_locks = 1; |
5d4f98a2 YZ |
1579 | } else |
1580 | extra_size = -1; | |
3173a18f JB |
1581 | |
1582 | /* | |
1583 | * Owner is our parent level, so we can just add one to get the level | |
1584 | * for the block we are interested in. | |
1585 | */ | |
1586 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1587 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1588 | key.offset = owner; | |
1589 | } | |
1590 | ||
1591 | again: | |
5d4f98a2 | 1592 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1593 | if (ret < 0) { |
5d4f98a2 YZ |
1594 | err = ret; |
1595 | goto out; | |
1596 | } | |
3173a18f JB |
1597 | |
1598 | /* | |
1599 | * We may be a newly converted file system which still has the old fat | |
1600 | * extent entries for metadata, so try and see if we have one of those. | |
1601 | */ | |
1602 | if (ret > 0 && skinny_metadata) { | |
1603 | skinny_metadata = false; | |
1604 | if (path->slots[0]) { | |
1605 | path->slots[0]--; | |
1606 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1607 | path->slots[0]); | |
1608 | if (key.objectid == bytenr && | |
1609 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1610 | key.offset == num_bytes) | |
1611 | ret = 0; | |
1612 | } | |
1613 | if (ret) { | |
9ce49a0b | 1614 | key.objectid = bytenr; |
3173a18f JB |
1615 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1616 | key.offset = num_bytes; | |
1617 | btrfs_release_path(path); | |
1618 | goto again; | |
1619 | } | |
1620 | } | |
1621 | ||
79787eaa JM |
1622 | if (ret && !insert) { |
1623 | err = -ENOENT; | |
1624 | goto out; | |
fae7f21c | 1625 | } else if (WARN_ON(ret)) { |
492104c8 | 1626 | err = -EIO; |
492104c8 | 1627 | goto out; |
79787eaa | 1628 | } |
5d4f98a2 YZ |
1629 | |
1630 | leaf = path->nodes[0]; | |
1631 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1632 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1633 | if (item_size < sizeof(*ei)) { | |
1634 | if (!insert) { | |
1635 | err = -ENOENT; | |
1636 | goto out; | |
1637 | } | |
1638 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1639 | extra_size); | |
1640 | if (ret < 0) { | |
1641 | err = ret; | |
1642 | goto out; | |
1643 | } | |
1644 | leaf = path->nodes[0]; | |
1645 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1646 | } | |
1647 | #endif | |
1648 | BUG_ON(item_size < sizeof(*ei)); | |
1649 | ||
5d4f98a2 YZ |
1650 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1651 | flags = btrfs_extent_flags(leaf, ei); | |
1652 | ||
1653 | ptr = (unsigned long)(ei + 1); | |
1654 | end = (unsigned long)ei + item_size; | |
1655 | ||
3173a18f | 1656 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1657 | ptr += sizeof(struct btrfs_tree_block_info); |
1658 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1659 | } |
1660 | ||
1661 | err = -ENOENT; | |
1662 | while (1) { | |
1663 | if (ptr >= end) { | |
1664 | WARN_ON(ptr > end); | |
1665 | break; | |
1666 | } | |
1667 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1668 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1669 | if (want < type) | |
1670 | break; | |
1671 | if (want > type) { | |
1672 | ptr += btrfs_extent_inline_ref_size(type); | |
1673 | continue; | |
1674 | } | |
1675 | ||
1676 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1677 | struct btrfs_extent_data_ref *dref; | |
1678 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1679 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1680 | owner, offset)) { | |
1681 | err = 0; | |
1682 | break; | |
1683 | } | |
1684 | if (hash_extent_data_ref_item(leaf, dref) < | |
1685 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1686 | break; | |
1687 | } else { | |
1688 | u64 ref_offset; | |
1689 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1690 | if (parent > 0) { | |
1691 | if (parent == ref_offset) { | |
1692 | err = 0; | |
1693 | break; | |
1694 | } | |
1695 | if (ref_offset < parent) | |
1696 | break; | |
1697 | } else { | |
1698 | if (root_objectid == ref_offset) { | |
1699 | err = 0; | |
1700 | break; | |
1701 | } | |
1702 | if (ref_offset < root_objectid) | |
1703 | break; | |
1704 | } | |
1705 | } | |
1706 | ptr += btrfs_extent_inline_ref_size(type); | |
1707 | } | |
1708 | if (err == -ENOENT && insert) { | |
1709 | if (item_size + extra_size >= | |
1710 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1711 | err = -EAGAIN; | |
1712 | goto out; | |
1713 | } | |
1714 | /* | |
1715 | * To add new inline back ref, we have to make sure | |
1716 | * there is no corresponding back ref item. | |
1717 | * For simplicity, we just do not add new inline back | |
1718 | * ref if there is any kind of item for this block | |
1719 | */ | |
2c47e605 YZ |
1720 | if (find_next_key(path, 0, &key) == 0 && |
1721 | key.objectid == bytenr && | |
85d4198e | 1722 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1723 | err = -EAGAIN; |
1724 | goto out; | |
1725 | } | |
1726 | } | |
1727 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1728 | out: | |
85d4198e | 1729 | if (insert) { |
5d4f98a2 YZ |
1730 | path->keep_locks = 0; |
1731 | btrfs_unlock_up_safe(path, 1); | |
1732 | } | |
1733 | return err; | |
1734 | } | |
1735 | ||
1736 | /* | |
1737 | * helper to add new inline back ref | |
1738 | */ | |
1739 | static noinline_for_stack | |
fd279fae | 1740 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1741 | struct btrfs_path *path, |
1742 | struct btrfs_extent_inline_ref *iref, | |
1743 | u64 parent, u64 root_objectid, | |
1744 | u64 owner, u64 offset, int refs_to_add, | |
1745 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1746 | { |
1747 | struct extent_buffer *leaf; | |
1748 | struct btrfs_extent_item *ei; | |
1749 | unsigned long ptr; | |
1750 | unsigned long end; | |
1751 | unsigned long item_offset; | |
1752 | u64 refs; | |
1753 | int size; | |
1754 | int type; | |
5d4f98a2 YZ |
1755 | |
1756 | leaf = path->nodes[0]; | |
1757 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1758 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1759 | ||
1760 | type = extent_ref_type(parent, owner); | |
1761 | size = btrfs_extent_inline_ref_size(type); | |
1762 | ||
4b90c680 | 1763 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1764 | |
1765 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1766 | refs = btrfs_extent_refs(leaf, ei); | |
1767 | refs += refs_to_add; | |
1768 | btrfs_set_extent_refs(leaf, ei, refs); | |
1769 | if (extent_op) | |
1770 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1771 | ||
1772 | ptr = (unsigned long)ei + item_offset; | |
1773 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1774 | if (ptr < end - size) | |
1775 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1776 | end - size - ptr); | |
1777 | ||
1778 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1779 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1780 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1781 | struct btrfs_extent_data_ref *dref; | |
1782 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1783 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1784 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1785 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1786 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1787 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1788 | struct btrfs_shared_data_ref *sref; | |
1789 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1790 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1791 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1792 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1793 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1794 | } else { | |
1795 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1796 | } | |
1797 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1798 | } |
1799 | ||
1800 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1801 | struct btrfs_root *root, | |
1802 | struct btrfs_path *path, | |
1803 | struct btrfs_extent_inline_ref **ref_ret, | |
1804 | u64 bytenr, u64 num_bytes, u64 parent, | |
1805 | u64 root_objectid, u64 owner, u64 offset) | |
1806 | { | |
1807 | int ret; | |
1808 | ||
1809 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1810 | bytenr, num_bytes, parent, | |
1811 | root_objectid, owner, offset, 0); | |
1812 | if (ret != -ENOENT) | |
54aa1f4d | 1813 | return ret; |
5d4f98a2 | 1814 | |
b3b4aa74 | 1815 | btrfs_release_path(path); |
5d4f98a2 YZ |
1816 | *ref_ret = NULL; |
1817 | ||
1818 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1819 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1820 | root_objectid); | |
1821 | } else { | |
1822 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1823 | root_objectid, owner, offset); | |
b9473439 | 1824 | } |
5d4f98a2 YZ |
1825 | return ret; |
1826 | } | |
31840ae1 | 1827 | |
5d4f98a2 YZ |
1828 | /* |
1829 | * helper to update/remove inline back ref | |
1830 | */ | |
1831 | static noinline_for_stack | |
afe5fea7 | 1832 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1833 | struct btrfs_path *path, |
1834 | struct btrfs_extent_inline_ref *iref, | |
1835 | int refs_to_mod, | |
fcebe456 JB |
1836 | struct btrfs_delayed_extent_op *extent_op, |
1837 | int *last_ref) | |
5d4f98a2 YZ |
1838 | { |
1839 | struct extent_buffer *leaf; | |
1840 | struct btrfs_extent_item *ei; | |
1841 | struct btrfs_extent_data_ref *dref = NULL; | |
1842 | struct btrfs_shared_data_ref *sref = NULL; | |
1843 | unsigned long ptr; | |
1844 | unsigned long end; | |
1845 | u32 item_size; | |
1846 | int size; | |
1847 | int type; | |
5d4f98a2 YZ |
1848 | u64 refs; |
1849 | ||
1850 | leaf = path->nodes[0]; | |
1851 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1852 | refs = btrfs_extent_refs(leaf, ei); | |
1853 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1854 | refs += refs_to_mod; | |
1855 | btrfs_set_extent_refs(leaf, ei, refs); | |
1856 | if (extent_op) | |
1857 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1858 | ||
1859 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1860 | ||
1861 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1862 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1863 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1864 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1865 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1866 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1867 | } else { | |
1868 | refs = 1; | |
1869 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1870 | } |
31840ae1 | 1871 | |
5d4f98a2 YZ |
1872 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1873 | refs += refs_to_mod; | |
1874 | ||
1875 | if (refs > 0) { | |
1876 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1877 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1878 | else | |
1879 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1880 | } else { | |
fcebe456 | 1881 | *last_ref = 1; |
5d4f98a2 YZ |
1882 | size = btrfs_extent_inline_ref_size(type); |
1883 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1884 | ptr = (unsigned long)iref; | |
1885 | end = (unsigned long)ei + item_size; | |
1886 | if (ptr + size < end) | |
1887 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1888 | end - ptr - size); | |
1889 | item_size -= size; | |
afe5fea7 | 1890 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1891 | } |
1892 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1893 | } |
1894 | ||
1895 | static noinline_for_stack | |
1896 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1897 | struct btrfs_root *root, | |
1898 | struct btrfs_path *path, | |
1899 | u64 bytenr, u64 num_bytes, u64 parent, | |
1900 | u64 root_objectid, u64 owner, | |
1901 | u64 offset, int refs_to_add, | |
1902 | struct btrfs_delayed_extent_op *extent_op) | |
1903 | { | |
1904 | struct btrfs_extent_inline_ref *iref; | |
1905 | int ret; | |
1906 | ||
1907 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1908 | bytenr, num_bytes, parent, | |
1909 | root_objectid, owner, offset, 1); | |
1910 | if (ret == 0) { | |
1911 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1912 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1913 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1914 | } else if (ret == -ENOENT) { |
fd279fae | 1915 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1916 | root_objectid, owner, offset, |
1917 | refs_to_add, extent_op); | |
1918 | ret = 0; | |
771ed689 | 1919 | } |
5d4f98a2 YZ |
1920 | return ret; |
1921 | } | |
31840ae1 | 1922 | |
5d4f98a2 YZ |
1923 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1924 | struct btrfs_root *root, | |
1925 | struct btrfs_path *path, | |
1926 | u64 bytenr, u64 parent, u64 root_objectid, | |
1927 | u64 owner, u64 offset, int refs_to_add) | |
1928 | { | |
1929 | int ret; | |
1930 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1931 | BUG_ON(refs_to_add != 1); | |
1932 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1933 | parent, root_objectid); | |
1934 | } else { | |
1935 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1936 | parent, root_objectid, | |
1937 | owner, offset, refs_to_add); | |
1938 | } | |
1939 | return ret; | |
1940 | } | |
56bec294 | 1941 | |
5d4f98a2 YZ |
1942 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1943 | struct btrfs_root *root, | |
1944 | struct btrfs_path *path, | |
1945 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1946 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1947 | { |
143bede5 | 1948 | int ret = 0; |
b9473439 | 1949 | |
5d4f98a2 YZ |
1950 | BUG_ON(!is_data && refs_to_drop != 1); |
1951 | if (iref) { | |
afe5fea7 | 1952 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1953 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1954 | } else if (is_data) { |
fcebe456 JB |
1955 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
1956 | last_ref); | |
5d4f98a2 | 1957 | } else { |
fcebe456 | 1958 | *last_ref = 1; |
5d4f98a2 YZ |
1959 | ret = btrfs_del_item(trans, root, path); |
1960 | } | |
1961 | return ret; | |
1962 | } | |
1963 | ||
86557861 | 1964 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1965 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1966 | u64 *discarded_bytes) | |
5d4f98a2 | 1967 | { |
86557861 JM |
1968 | int j, ret = 0; |
1969 | u64 bytes_left, end; | |
4d89d377 | 1970 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1971 | |
4d89d377 JM |
1972 | if (WARN_ON(start != aligned_start)) { |
1973 | len -= aligned_start - start; | |
1974 | len = round_down(len, 1 << 9); | |
1975 | start = aligned_start; | |
1976 | } | |
d04c6b88 | 1977 | |
4d89d377 | 1978 | *discarded_bytes = 0; |
86557861 JM |
1979 | |
1980 | if (!len) | |
1981 | return 0; | |
1982 | ||
1983 | end = start + len; | |
1984 | bytes_left = len; | |
1985 | ||
1986 | /* Skip any superblocks on this device. */ | |
1987 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1988 | u64 sb_start = btrfs_sb_offset(j); | |
1989 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1990 | u64 size = sb_start - start; | |
1991 | ||
1992 | if (!in_range(sb_start, start, bytes_left) && | |
1993 | !in_range(sb_end, start, bytes_left) && | |
1994 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1995 | continue; | |
1996 | ||
1997 | /* | |
1998 | * Superblock spans beginning of range. Adjust start and | |
1999 | * try again. | |
2000 | */ | |
2001 | if (sb_start <= start) { | |
2002 | start += sb_end - start; | |
2003 | if (start > end) { | |
2004 | bytes_left = 0; | |
2005 | break; | |
2006 | } | |
2007 | bytes_left = end - start; | |
2008 | continue; | |
2009 | } | |
2010 | ||
2011 | if (size) { | |
2012 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2013 | GFP_NOFS, 0); | |
2014 | if (!ret) | |
2015 | *discarded_bytes += size; | |
2016 | else if (ret != -EOPNOTSUPP) | |
2017 | return ret; | |
2018 | } | |
2019 | ||
2020 | start = sb_end; | |
2021 | if (start > end) { | |
2022 | bytes_left = 0; | |
2023 | break; | |
2024 | } | |
2025 | bytes_left = end - start; | |
2026 | } | |
2027 | ||
2028 | if (bytes_left) { | |
2029 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2030 | GFP_NOFS, 0); |
2031 | if (!ret) | |
86557861 | 2032 | *discarded_bytes += bytes_left; |
4d89d377 | 2033 | } |
d04c6b88 | 2034 | return ret; |
5d4f98a2 | 2035 | } |
5d4f98a2 | 2036 | |
1edb647b FM |
2037 | int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
2038 | u64 num_bytes, u64 *actual_bytes) | |
5d4f98a2 | 2039 | { |
5d4f98a2 | 2040 | int ret; |
5378e607 | 2041 | u64 discarded_bytes = 0; |
a1d3c478 | 2042 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2043 | |
e244a0ae | 2044 | |
2999241d FM |
2045 | /* |
2046 | * Avoid races with device replace and make sure our bbio has devices | |
2047 | * associated to its stripes that don't go away while we are discarding. | |
2048 | */ | |
2049 | btrfs_bio_counter_inc_blocked(root->fs_info); | |
5d4f98a2 | 2050 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 2051 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 2052 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 2053 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2054 | if (!ret) { |
a1d3c478 | 2055 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2056 | int i; |
2057 | ||
5d4f98a2 | 2058 | |
a1d3c478 | 2059 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2060 | u64 bytes; |
d5e2003c JB |
2061 | if (!stripe->dev->can_discard) |
2062 | continue; | |
2063 | ||
5378e607 LD |
2064 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2065 | stripe->physical, | |
d04c6b88 JM |
2066 | stripe->length, |
2067 | &bytes); | |
5378e607 | 2068 | if (!ret) |
d04c6b88 | 2069 | discarded_bytes += bytes; |
5378e607 | 2070 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2071 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2072 | |
2073 | /* | |
2074 | * Just in case we get back EOPNOTSUPP for some reason, | |
2075 | * just ignore the return value so we don't screw up | |
2076 | * people calling discard_extent. | |
2077 | */ | |
2078 | ret = 0; | |
5d4f98a2 | 2079 | } |
6e9606d2 | 2080 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2081 | } |
2999241d | 2082 | btrfs_bio_counter_dec(root->fs_info); |
5378e607 LD |
2083 | |
2084 | if (actual_bytes) | |
2085 | *actual_bytes = discarded_bytes; | |
2086 | ||
5d4f98a2 | 2087 | |
53b381b3 DW |
2088 | if (ret == -EOPNOTSUPP) |
2089 | ret = 0; | |
5d4f98a2 | 2090 | return ret; |
5d4f98a2 YZ |
2091 | } |
2092 | ||
79787eaa | 2093 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
2094 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2095 | struct btrfs_root *root, | |
2096 | u64 bytenr, u64 num_bytes, u64 parent, | |
b06c4bf5 | 2097 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2098 | { |
2099 | int ret; | |
66d7e7f0 AJ |
2100 | struct btrfs_fs_info *fs_info = root->fs_info; |
2101 | ||
5d4f98a2 YZ |
2102 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2103 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2104 | ||
2105 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2106 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2107 | num_bytes, | |
5d4f98a2 | 2108 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2109 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2110 | } else { |
66d7e7f0 | 2111 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2112 | num_bytes, parent, root_objectid, |
2113 | owner, offset, 0, | |
b06c4bf5 | 2114 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 YZ |
2115 | } |
2116 | return ret; | |
2117 | } | |
2118 | ||
2119 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2120 | struct btrfs_root *root, | |
c682f9b3 | 2121 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2122 | u64 parent, u64 root_objectid, |
2123 | u64 owner, u64 offset, int refs_to_add, | |
2124 | struct btrfs_delayed_extent_op *extent_op) | |
2125 | { | |
fcebe456 | 2126 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2127 | struct btrfs_path *path; |
2128 | struct extent_buffer *leaf; | |
2129 | struct btrfs_extent_item *item; | |
fcebe456 | 2130 | struct btrfs_key key; |
c682f9b3 QW |
2131 | u64 bytenr = node->bytenr; |
2132 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2133 | u64 refs; |
2134 | int ret; | |
5d4f98a2 YZ |
2135 | |
2136 | path = btrfs_alloc_path(); | |
2137 | if (!path) | |
2138 | return -ENOMEM; | |
2139 | ||
e4058b54 | 2140 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2141 | path->leave_spinning = 1; |
2142 | /* this will setup the path even if it fails to insert the back ref */ | |
fcebe456 JB |
2143 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
2144 | bytenr, num_bytes, parent, | |
5d4f98a2 YZ |
2145 | root_objectid, owner, offset, |
2146 | refs_to_add, extent_op); | |
0ed4792a | 2147 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2148 | goto out; |
fcebe456 JB |
2149 | |
2150 | /* | |
2151 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2152 | * inline extent ref, so just update the reference count and add a | |
2153 | * normal backref. | |
2154 | */ | |
5d4f98a2 | 2155 | leaf = path->nodes[0]; |
fcebe456 | 2156 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2157 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2158 | refs = btrfs_extent_refs(leaf, item); | |
2159 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2160 | if (extent_op) | |
2161 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2162 | |
5d4f98a2 | 2163 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2164 | btrfs_release_path(path); |
56bec294 | 2165 | |
e4058b54 | 2166 | path->reada = READA_FORWARD; |
b9473439 | 2167 | path->leave_spinning = 1; |
56bec294 CM |
2168 | /* now insert the actual backref */ |
2169 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2170 | path, bytenr, parent, root_objectid, |
2171 | owner, offset, refs_to_add); | |
79787eaa JM |
2172 | if (ret) |
2173 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2174 | out: |
56bec294 | 2175 | btrfs_free_path(path); |
30d133fc | 2176 | return ret; |
56bec294 CM |
2177 | } |
2178 | ||
5d4f98a2 YZ |
2179 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2180 | struct btrfs_root *root, | |
2181 | struct btrfs_delayed_ref_node *node, | |
2182 | struct btrfs_delayed_extent_op *extent_op, | |
2183 | int insert_reserved) | |
56bec294 | 2184 | { |
5d4f98a2 YZ |
2185 | int ret = 0; |
2186 | struct btrfs_delayed_data_ref *ref; | |
2187 | struct btrfs_key ins; | |
2188 | u64 parent = 0; | |
2189 | u64 ref_root = 0; | |
2190 | u64 flags = 0; | |
2191 | ||
2192 | ins.objectid = node->bytenr; | |
2193 | ins.offset = node->num_bytes; | |
2194 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2195 | ||
2196 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2197 | trace_run_delayed_data_ref(node, ref, node->action); |
2198 | ||
5d4f98a2 YZ |
2199 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2200 | parent = ref->parent; | |
fcebe456 | 2201 | ref_root = ref->root; |
5d4f98a2 YZ |
2202 | |
2203 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2204 | if (extent_op) |
5d4f98a2 | 2205 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2206 | ret = alloc_reserved_file_extent(trans, root, |
2207 | parent, ref_root, flags, | |
2208 | ref->objectid, ref->offset, | |
2209 | &ins, node->ref_mod); | |
5d4f98a2 | 2210 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 | 2211 | ret = __btrfs_inc_extent_ref(trans, root, node, parent, |
5d4f98a2 YZ |
2212 | ref_root, ref->objectid, |
2213 | ref->offset, node->ref_mod, | |
c682f9b3 | 2214 | extent_op); |
5d4f98a2 | 2215 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 | 2216 | ret = __btrfs_free_extent(trans, root, node, parent, |
5d4f98a2 YZ |
2217 | ref_root, ref->objectid, |
2218 | ref->offset, node->ref_mod, | |
c682f9b3 | 2219 | extent_op); |
5d4f98a2 YZ |
2220 | } else { |
2221 | BUG(); | |
2222 | } | |
2223 | return ret; | |
2224 | } | |
2225 | ||
2226 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2227 | struct extent_buffer *leaf, | |
2228 | struct btrfs_extent_item *ei) | |
2229 | { | |
2230 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2231 | if (extent_op->update_flags) { | |
2232 | flags |= extent_op->flags_to_set; | |
2233 | btrfs_set_extent_flags(leaf, ei, flags); | |
2234 | } | |
2235 | ||
2236 | if (extent_op->update_key) { | |
2237 | struct btrfs_tree_block_info *bi; | |
2238 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2239 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2240 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2241 | } | |
2242 | } | |
2243 | ||
2244 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2245 | struct btrfs_root *root, | |
2246 | struct btrfs_delayed_ref_node *node, | |
2247 | struct btrfs_delayed_extent_op *extent_op) | |
2248 | { | |
2249 | struct btrfs_key key; | |
2250 | struct btrfs_path *path; | |
2251 | struct btrfs_extent_item *ei; | |
2252 | struct extent_buffer *leaf; | |
2253 | u32 item_size; | |
56bec294 | 2254 | int ret; |
5d4f98a2 | 2255 | int err = 0; |
b1c79e09 | 2256 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2257 | |
79787eaa JM |
2258 | if (trans->aborted) |
2259 | return 0; | |
2260 | ||
3173a18f JB |
2261 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2262 | metadata = 0; | |
2263 | ||
5d4f98a2 YZ |
2264 | path = btrfs_alloc_path(); |
2265 | if (!path) | |
2266 | return -ENOMEM; | |
2267 | ||
2268 | key.objectid = node->bytenr; | |
5d4f98a2 | 2269 | |
3173a18f | 2270 | if (metadata) { |
3173a18f | 2271 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2272 | key.offset = extent_op->level; |
3173a18f JB |
2273 | } else { |
2274 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2275 | key.offset = node->num_bytes; | |
2276 | } | |
2277 | ||
2278 | again: | |
e4058b54 | 2279 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2280 | path->leave_spinning = 1; |
2281 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2282 | path, 0, 1); | |
2283 | if (ret < 0) { | |
2284 | err = ret; | |
2285 | goto out; | |
2286 | } | |
2287 | if (ret > 0) { | |
3173a18f | 2288 | if (metadata) { |
55994887 FDBM |
2289 | if (path->slots[0] > 0) { |
2290 | path->slots[0]--; | |
2291 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2292 | path->slots[0]); | |
2293 | if (key.objectid == node->bytenr && | |
2294 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2295 | key.offset == node->num_bytes) | |
2296 | ret = 0; | |
2297 | } | |
2298 | if (ret > 0) { | |
2299 | btrfs_release_path(path); | |
2300 | metadata = 0; | |
3173a18f | 2301 | |
55994887 FDBM |
2302 | key.objectid = node->bytenr; |
2303 | key.offset = node->num_bytes; | |
2304 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2305 | goto again; | |
2306 | } | |
2307 | } else { | |
2308 | err = -EIO; | |
2309 | goto out; | |
3173a18f | 2310 | } |
5d4f98a2 YZ |
2311 | } |
2312 | ||
2313 | leaf = path->nodes[0]; | |
2314 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2315 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2316 | if (item_size < sizeof(*ei)) { | |
2317 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2318 | path, (u64)-1, 0); | |
2319 | if (ret < 0) { | |
2320 | err = ret; | |
2321 | goto out; | |
2322 | } | |
2323 | leaf = path->nodes[0]; | |
2324 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2325 | } | |
2326 | #endif | |
2327 | BUG_ON(item_size < sizeof(*ei)); | |
2328 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2329 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2330 | |
5d4f98a2 YZ |
2331 | btrfs_mark_buffer_dirty(leaf); |
2332 | out: | |
2333 | btrfs_free_path(path); | |
2334 | return err; | |
56bec294 CM |
2335 | } |
2336 | ||
5d4f98a2 YZ |
2337 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2338 | struct btrfs_root *root, | |
2339 | struct btrfs_delayed_ref_node *node, | |
2340 | struct btrfs_delayed_extent_op *extent_op, | |
2341 | int insert_reserved) | |
56bec294 CM |
2342 | { |
2343 | int ret = 0; | |
5d4f98a2 YZ |
2344 | struct btrfs_delayed_tree_ref *ref; |
2345 | struct btrfs_key ins; | |
2346 | u64 parent = 0; | |
2347 | u64 ref_root = 0; | |
3173a18f JB |
2348 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2349 | SKINNY_METADATA); | |
56bec294 | 2350 | |
5d4f98a2 | 2351 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2352 | trace_run_delayed_tree_ref(node, ref, node->action); |
2353 | ||
5d4f98a2 YZ |
2354 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2355 | parent = ref->parent; | |
fcebe456 | 2356 | ref_root = ref->root; |
5d4f98a2 | 2357 | |
3173a18f JB |
2358 | ins.objectid = node->bytenr; |
2359 | if (skinny_metadata) { | |
2360 | ins.offset = ref->level; | |
2361 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2362 | } else { | |
2363 | ins.offset = node->num_bytes; | |
2364 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2365 | } | |
2366 | ||
5d4f98a2 YZ |
2367 | BUG_ON(node->ref_mod != 1); |
2368 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2369 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2370 | ret = alloc_reserved_tree_block(trans, root, |
2371 | parent, ref_root, | |
2372 | extent_op->flags_to_set, | |
2373 | &extent_op->key, | |
b06c4bf5 | 2374 | ref->level, &ins); |
5d4f98a2 | 2375 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 QW |
2376 | ret = __btrfs_inc_extent_ref(trans, root, node, |
2377 | parent, ref_root, | |
2378 | ref->level, 0, 1, | |
fcebe456 | 2379 | extent_op); |
5d4f98a2 | 2380 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 QW |
2381 | ret = __btrfs_free_extent(trans, root, node, |
2382 | parent, ref_root, | |
2383 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2384 | } else { |
2385 | BUG(); | |
2386 | } | |
56bec294 CM |
2387 | return ret; |
2388 | } | |
2389 | ||
2390 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2391 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2392 | struct btrfs_root *root, | |
2393 | struct btrfs_delayed_ref_node *node, | |
2394 | struct btrfs_delayed_extent_op *extent_op, | |
2395 | int insert_reserved) | |
56bec294 | 2396 | { |
79787eaa JM |
2397 | int ret = 0; |
2398 | ||
857cc2fc JB |
2399 | if (trans->aborted) { |
2400 | if (insert_reserved) | |
2401 | btrfs_pin_extent(root, node->bytenr, | |
2402 | node->num_bytes, 1); | |
79787eaa | 2403 | return 0; |
857cc2fc | 2404 | } |
79787eaa | 2405 | |
5d4f98a2 | 2406 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2407 | struct btrfs_delayed_ref_head *head; |
2408 | /* | |
2409 | * we've hit the end of the chain and we were supposed | |
2410 | * to insert this extent into the tree. But, it got | |
2411 | * deleted before we ever needed to insert it, so all | |
2412 | * we have to do is clean up the accounting | |
2413 | */ | |
5d4f98a2 YZ |
2414 | BUG_ON(extent_op); |
2415 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2416 | trace_run_delayed_ref_head(node, head, node->action); |
2417 | ||
56bec294 | 2418 | if (insert_reserved) { |
f0486c68 YZ |
2419 | btrfs_pin_extent(root, node->bytenr, |
2420 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2421 | if (head->is_data) { |
2422 | ret = btrfs_del_csums(trans, root, | |
2423 | node->bytenr, | |
2424 | node->num_bytes); | |
5d4f98a2 | 2425 | } |
56bec294 | 2426 | } |
297d750b QW |
2427 | |
2428 | /* Also free its reserved qgroup space */ | |
2429 | btrfs_qgroup_free_delayed_ref(root->fs_info, | |
2430 | head->qgroup_ref_root, | |
2431 | head->qgroup_reserved); | |
79787eaa | 2432 | return ret; |
56bec294 CM |
2433 | } |
2434 | ||
5d4f98a2 YZ |
2435 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2436 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2437 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2438 | insert_reserved); | |
2439 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2440 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2441 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2442 | insert_reserved); | |
2443 | else | |
2444 | BUG(); | |
2445 | return ret; | |
56bec294 CM |
2446 | } |
2447 | ||
c6fc2454 | 2448 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2449 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2450 | { | |
cffc3374 FM |
2451 | struct btrfs_delayed_ref_node *ref; |
2452 | ||
c6fc2454 QW |
2453 | if (list_empty(&head->ref_list)) |
2454 | return NULL; | |
d7df2c79 | 2455 | |
cffc3374 FM |
2456 | /* |
2457 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2458 | * This is to prevent a ref count from going down to zero, which deletes | |
2459 | * the extent item from the extent tree, when there still are references | |
2460 | * to add, which would fail because they would not find the extent item. | |
2461 | */ | |
2462 | list_for_each_entry(ref, &head->ref_list, list) { | |
2463 | if (ref->action == BTRFS_ADD_DELAYED_REF) | |
2464 | return ref; | |
2465 | } | |
2466 | ||
c6fc2454 QW |
2467 | return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node, |
2468 | list); | |
56bec294 CM |
2469 | } |
2470 | ||
79787eaa JM |
2471 | /* |
2472 | * Returns 0 on success or if called with an already aborted transaction. | |
2473 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2474 | */ | |
d7df2c79 JB |
2475 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2476 | struct btrfs_root *root, | |
2477 | unsigned long nr) | |
56bec294 | 2478 | { |
56bec294 CM |
2479 | struct btrfs_delayed_ref_root *delayed_refs; |
2480 | struct btrfs_delayed_ref_node *ref; | |
2481 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2482 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2483 | struct btrfs_fs_info *fs_info = root->fs_info; |
0a2b2a84 | 2484 | ktime_t start = ktime_get(); |
56bec294 | 2485 | int ret; |
d7df2c79 | 2486 | unsigned long count = 0; |
0a2b2a84 | 2487 | unsigned long actual_count = 0; |
56bec294 | 2488 | int must_insert_reserved = 0; |
56bec294 CM |
2489 | |
2490 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2491 | while (1) { |
2492 | if (!locked_ref) { | |
d7df2c79 | 2493 | if (count >= nr) |
56bec294 | 2494 | break; |
56bec294 | 2495 | |
d7df2c79 JB |
2496 | spin_lock(&delayed_refs->lock); |
2497 | locked_ref = btrfs_select_ref_head(trans); | |
2498 | if (!locked_ref) { | |
2499 | spin_unlock(&delayed_refs->lock); | |
2500 | break; | |
2501 | } | |
c3e69d58 CM |
2502 | |
2503 | /* grab the lock that says we are going to process | |
2504 | * all the refs for this head */ | |
2505 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2506 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2507 | /* |
2508 | * we may have dropped the spin lock to get the head | |
2509 | * mutex lock, and that might have given someone else | |
2510 | * time to free the head. If that's true, it has been | |
2511 | * removed from our list and we can move on. | |
2512 | */ | |
2513 | if (ret == -EAGAIN) { | |
2514 | locked_ref = NULL; | |
2515 | count++; | |
2516 | continue; | |
56bec294 CM |
2517 | } |
2518 | } | |
a28ec197 | 2519 | |
2c3cf7d5 FM |
2520 | /* |
2521 | * We need to try and merge add/drops of the same ref since we | |
2522 | * can run into issues with relocate dropping the implicit ref | |
2523 | * and then it being added back again before the drop can | |
2524 | * finish. If we merged anything we need to re-loop so we can | |
2525 | * get a good ref. | |
2526 | * Or we can get node references of the same type that weren't | |
2527 | * merged when created due to bumps in the tree mod seq, and | |
2528 | * we need to merge them to prevent adding an inline extent | |
2529 | * backref before dropping it (triggering a BUG_ON at | |
2530 | * insert_inline_extent_backref()). | |
2531 | */ | |
d7df2c79 | 2532 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2533 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2534 | locked_ref); | |
ae1e206b | 2535 | |
d1270cd9 AJ |
2536 | /* |
2537 | * locked_ref is the head node, so we have to go one | |
2538 | * node back for any delayed ref updates | |
2539 | */ | |
2540 | ref = select_delayed_ref(locked_ref); | |
2541 | ||
2542 | if (ref && ref->seq && | |
097b8a7c | 2543 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2544 | spin_unlock(&locked_ref->lock); |
093486c4 | 2545 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 JB |
2546 | spin_lock(&delayed_refs->lock); |
2547 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2548 | delayed_refs->num_heads_ready++; |
2549 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2550 | locked_ref = NULL; |
d1270cd9 | 2551 | cond_resched(); |
27a377db | 2552 | count++; |
d1270cd9 AJ |
2553 | continue; |
2554 | } | |
2555 | ||
56bec294 CM |
2556 | /* |
2557 | * record the must insert reserved flag before we | |
2558 | * drop the spin lock. | |
2559 | */ | |
2560 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2561 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2562 | |
5d4f98a2 YZ |
2563 | extent_op = locked_ref->extent_op; |
2564 | locked_ref->extent_op = NULL; | |
2565 | ||
56bec294 | 2566 | if (!ref) { |
d7df2c79 JB |
2567 | |
2568 | ||
56bec294 CM |
2569 | /* All delayed refs have been processed, Go ahead |
2570 | * and send the head node to run_one_delayed_ref, | |
2571 | * so that any accounting fixes can happen | |
2572 | */ | |
2573 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2574 | |
2575 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2576 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2577 | extent_op = NULL; |
2578 | } | |
2579 | ||
2580 | if (extent_op) { | |
d7df2c79 | 2581 | spin_unlock(&locked_ref->lock); |
5d4f98a2 YZ |
2582 | ret = run_delayed_extent_op(trans, root, |
2583 | ref, extent_op); | |
78a6184a | 2584 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2585 | |
79787eaa | 2586 | if (ret) { |
857cc2fc JB |
2587 | /* |
2588 | * Need to reset must_insert_reserved if | |
2589 | * there was an error so the abort stuff | |
2590 | * can cleanup the reserved space | |
2591 | * properly. | |
2592 | */ | |
2593 | if (must_insert_reserved) | |
2594 | locked_ref->must_insert_reserved = 1; | |
d7df2c79 | 2595 | locked_ref->processing = 0; |
c2cf52eb | 2596 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
093486c4 | 2597 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2598 | return ret; |
2599 | } | |
d7df2c79 | 2600 | continue; |
5d4f98a2 | 2601 | } |
02217ed2 | 2602 | |
d7df2c79 | 2603 | /* |
01327610 | 2604 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2605 | * delayed ref lock and then re-check to make sure |
2606 | * nobody got added. | |
2607 | */ | |
2608 | spin_unlock(&locked_ref->lock); | |
2609 | spin_lock(&delayed_refs->lock); | |
2610 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2611 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2612 | locked_ref->extent_op) { |
d7df2c79 JB |
2613 | spin_unlock(&locked_ref->lock); |
2614 | spin_unlock(&delayed_refs->lock); | |
2615 | continue; | |
2616 | } | |
2617 | ref->in_tree = 0; | |
2618 | delayed_refs->num_heads--; | |
c46effa6 LB |
2619 | rb_erase(&locked_ref->href_node, |
2620 | &delayed_refs->href_root); | |
d7df2c79 JB |
2621 | spin_unlock(&delayed_refs->lock); |
2622 | } else { | |
0a2b2a84 | 2623 | actual_count++; |
d7df2c79 | 2624 | ref->in_tree = 0; |
c6fc2454 | 2625 | list_del(&ref->list); |
c46effa6 | 2626 | } |
d7df2c79 JB |
2627 | atomic_dec(&delayed_refs->num_entries); |
2628 | ||
093486c4 | 2629 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2630 | /* |
2631 | * when we play the delayed ref, also correct the | |
2632 | * ref_mod on head | |
2633 | */ | |
2634 | switch (ref->action) { | |
2635 | case BTRFS_ADD_DELAYED_REF: | |
2636 | case BTRFS_ADD_DELAYED_EXTENT: | |
2637 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2638 | break; | |
2639 | case BTRFS_DROP_DELAYED_REF: | |
2640 | locked_ref->node.ref_mod += ref->ref_mod; | |
2641 | break; | |
2642 | default: | |
2643 | WARN_ON(1); | |
2644 | } | |
2645 | } | |
d7df2c79 | 2646 | spin_unlock(&locked_ref->lock); |
925baedd | 2647 | |
5d4f98a2 | 2648 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2649 | must_insert_reserved); |
eb099670 | 2650 | |
78a6184a | 2651 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2652 | if (ret) { |
d7df2c79 | 2653 | locked_ref->processing = 0; |
093486c4 MX |
2654 | btrfs_delayed_ref_unlock(locked_ref); |
2655 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2656 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
79787eaa JM |
2657 | return ret; |
2658 | } | |
2659 | ||
093486c4 MX |
2660 | /* |
2661 | * If this node is a head, that means all the refs in this head | |
2662 | * have been dealt with, and we will pick the next head to deal | |
2663 | * with, so we must unlock the head and drop it from the cluster | |
2664 | * list before we release it. | |
2665 | */ | |
2666 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2667 | if (locked_ref->is_data && |
2668 | locked_ref->total_ref_mod < 0) { | |
2669 | spin_lock(&delayed_refs->lock); | |
2670 | delayed_refs->pending_csums -= ref->num_bytes; | |
2671 | spin_unlock(&delayed_refs->lock); | |
2672 | } | |
093486c4 MX |
2673 | btrfs_delayed_ref_unlock(locked_ref); |
2674 | locked_ref = NULL; | |
2675 | } | |
2676 | btrfs_put_delayed_ref(ref); | |
2677 | count++; | |
c3e69d58 | 2678 | cond_resched(); |
c3e69d58 | 2679 | } |
0a2b2a84 JB |
2680 | |
2681 | /* | |
2682 | * We don't want to include ref heads since we can have empty ref heads | |
2683 | * and those will drastically skew our runtime down since we just do | |
2684 | * accounting, no actual extent tree updates. | |
2685 | */ | |
2686 | if (actual_count > 0) { | |
2687 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2688 | u64 avg; | |
2689 | ||
2690 | /* | |
2691 | * We weigh the current average higher than our current runtime | |
2692 | * to avoid large swings in the average. | |
2693 | */ | |
2694 | spin_lock(&delayed_refs->lock); | |
2695 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2696 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2697 | spin_unlock(&delayed_refs->lock); |
2698 | } | |
d7df2c79 | 2699 | return 0; |
c3e69d58 CM |
2700 | } |
2701 | ||
709c0486 AJ |
2702 | #ifdef SCRAMBLE_DELAYED_REFS |
2703 | /* | |
2704 | * Normally delayed refs get processed in ascending bytenr order. This | |
2705 | * correlates in most cases to the order added. To expose dependencies on this | |
2706 | * order, we start to process the tree in the middle instead of the beginning | |
2707 | */ | |
2708 | static u64 find_middle(struct rb_root *root) | |
2709 | { | |
2710 | struct rb_node *n = root->rb_node; | |
2711 | struct btrfs_delayed_ref_node *entry; | |
2712 | int alt = 1; | |
2713 | u64 middle; | |
2714 | u64 first = 0, last = 0; | |
2715 | ||
2716 | n = rb_first(root); | |
2717 | if (n) { | |
2718 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2719 | first = entry->bytenr; | |
2720 | } | |
2721 | n = rb_last(root); | |
2722 | if (n) { | |
2723 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2724 | last = entry->bytenr; | |
2725 | } | |
2726 | n = root->rb_node; | |
2727 | ||
2728 | while (n) { | |
2729 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2730 | WARN_ON(!entry->in_tree); | |
2731 | ||
2732 | middle = entry->bytenr; | |
2733 | ||
2734 | if (alt) | |
2735 | n = n->rb_left; | |
2736 | else | |
2737 | n = n->rb_right; | |
2738 | ||
2739 | alt = 1 - alt; | |
2740 | } | |
2741 | return middle; | |
2742 | } | |
2743 | #endif | |
2744 | ||
1be41b78 JB |
2745 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2746 | { | |
2747 | u64 num_bytes; | |
2748 | ||
2749 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2750 | sizeof(struct btrfs_extent_inline_ref)); | |
2751 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2752 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2753 | ||
2754 | /* | |
2755 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2756 | * closer to what we're really going to want to use. |
1be41b78 | 2757 | */ |
f8c269d7 | 2758 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); |
1be41b78 JB |
2759 | } |
2760 | ||
1262133b JB |
2761 | /* |
2762 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2763 | * would require to store the csums for that many bytes. | |
2764 | */ | |
28f75a0e | 2765 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes) |
1262133b JB |
2766 | { |
2767 | u64 csum_size; | |
2768 | u64 num_csums_per_leaf; | |
2769 | u64 num_csums; | |
2770 | ||
2771 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
2772 | num_csums_per_leaf = div64_u64(csum_size, | |
2773 | (u64)btrfs_super_csum_size(root->fs_info->super_copy)); | |
2774 | num_csums = div64_u64(csum_bytes, root->sectorsize); | |
2775 | num_csums += num_csums_per_leaf - 1; | |
2776 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2777 | return num_csums; | |
2778 | } | |
2779 | ||
0a2b2a84 | 2780 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
1be41b78 JB |
2781 | struct btrfs_root *root) |
2782 | { | |
2783 | struct btrfs_block_rsv *global_rsv; | |
2784 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2785 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2786 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2787 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2788 | int ret = 0; |
2789 | ||
2790 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2791 | num_heads = heads_to_leaves(root, num_heads); | |
2792 | if (num_heads > 1) | |
707e8a07 | 2793 | num_bytes += (num_heads - 1) * root->nodesize; |
1be41b78 | 2794 | num_bytes <<= 1; |
28f75a0e | 2795 | num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize; |
cb723e49 JB |
2796 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root, |
2797 | num_dirty_bgs); | |
1be41b78 JB |
2798 | global_rsv = &root->fs_info->global_block_rsv; |
2799 | ||
2800 | /* | |
2801 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2802 | * wiggle room since running delayed refs can create more delayed refs. | |
2803 | */ | |
cb723e49 JB |
2804 | if (global_rsv->space_info->full) { |
2805 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2806 | num_bytes <<= 1; |
cb723e49 | 2807 | } |
1be41b78 JB |
2808 | |
2809 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2810 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2811 | ret = 1; |
2812 | spin_unlock(&global_rsv->lock); | |
2813 | return ret; | |
2814 | } | |
2815 | ||
0a2b2a84 JB |
2816 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2817 | struct btrfs_root *root) | |
2818 | { | |
2819 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2820 | u64 num_entries = | |
2821 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2822 | u64 avg_runtime; | |
a79b7d4b | 2823 | u64 val; |
0a2b2a84 JB |
2824 | |
2825 | smp_mb(); | |
2826 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2827 | val = num_entries * avg_runtime; |
0a2b2a84 JB |
2828 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
2829 | return 1; | |
a79b7d4b CM |
2830 | if (val >= NSEC_PER_SEC / 2) |
2831 | return 2; | |
0a2b2a84 JB |
2832 | |
2833 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2834 | } | |
2835 | ||
a79b7d4b CM |
2836 | struct async_delayed_refs { |
2837 | struct btrfs_root *root; | |
31b9655f | 2838 | u64 transid; |
a79b7d4b CM |
2839 | int count; |
2840 | int error; | |
2841 | int sync; | |
2842 | struct completion wait; | |
2843 | struct btrfs_work work; | |
2844 | }; | |
2845 | ||
2846 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2847 | { | |
2848 | struct async_delayed_refs *async; | |
2849 | struct btrfs_trans_handle *trans; | |
2850 | int ret; | |
2851 | ||
2852 | async = container_of(work, struct async_delayed_refs, work); | |
2853 | ||
0f873eca CM |
2854 | /* if the commit is already started, we don't need to wait here */ |
2855 | if (btrfs_transaction_blocked(async->root->fs_info)) | |
31b9655f | 2856 | goto done; |
31b9655f | 2857 | |
0f873eca CM |
2858 | trans = btrfs_join_transaction(async->root); |
2859 | if (IS_ERR(trans)) { | |
2860 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2861 | goto done; |
2862 | } | |
2863 | ||
2864 | /* | |
01327610 | 2865 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2866 | * wait on delayed refs |
2867 | */ | |
2868 | trans->sync = true; | |
0f873eca CM |
2869 | |
2870 | /* Don't bother flushing if we got into a different transaction */ | |
2871 | if (trans->transid > async->transid) | |
2872 | goto end; | |
2873 | ||
a79b7d4b CM |
2874 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); |
2875 | if (ret) | |
2876 | async->error = ret; | |
0f873eca | 2877 | end: |
a79b7d4b CM |
2878 | ret = btrfs_end_transaction(trans, async->root); |
2879 | if (ret && !async->error) | |
2880 | async->error = ret; | |
2881 | done: | |
2882 | if (async->sync) | |
2883 | complete(&async->wait); | |
2884 | else | |
2885 | kfree(async); | |
2886 | } | |
2887 | ||
2888 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
31b9655f | 2889 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2890 | { |
2891 | struct async_delayed_refs *async; | |
2892 | int ret; | |
2893 | ||
2894 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2895 | if (!async) | |
2896 | return -ENOMEM; | |
2897 | ||
2898 | async->root = root->fs_info->tree_root; | |
2899 | async->count = count; | |
2900 | async->error = 0; | |
31b9655f | 2901 | async->transid = transid; |
a79b7d4b CM |
2902 | if (wait) |
2903 | async->sync = 1; | |
2904 | else | |
2905 | async->sync = 0; | |
2906 | init_completion(&async->wait); | |
2907 | ||
9e0af237 LB |
2908 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2909 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b CM |
2910 | |
2911 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2912 | ||
2913 | if (wait) { | |
2914 | wait_for_completion(&async->wait); | |
2915 | ret = async->error; | |
2916 | kfree(async); | |
2917 | return ret; | |
2918 | } | |
2919 | return 0; | |
2920 | } | |
2921 | ||
c3e69d58 CM |
2922 | /* |
2923 | * this starts processing the delayed reference count updates and | |
2924 | * extent insertions we have queued up so far. count can be | |
2925 | * 0, which means to process everything in the tree at the start | |
2926 | * of the run (but not newly added entries), or it can be some target | |
2927 | * number you'd like to process. | |
79787eaa JM |
2928 | * |
2929 | * Returns 0 on success or if called with an aborted transaction | |
2930 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2931 | */ |
2932 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2933 | struct btrfs_root *root, unsigned long count) | |
2934 | { | |
2935 | struct rb_node *node; | |
2936 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2937 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2938 | int ret; |
2939 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2940 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2941 | |
79787eaa JM |
2942 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2943 | if (trans->aborted) | |
2944 | return 0; | |
2945 | ||
511711af CM |
2946 | if (root->fs_info->creating_free_space_tree) |
2947 | return 0; | |
2948 | ||
c3e69d58 CM |
2949 | if (root == root->fs_info->extent_root) |
2950 | root = root->fs_info->tree_root; | |
2951 | ||
2952 | delayed_refs = &trans->transaction->delayed_refs; | |
26455d33 | 2953 | if (count == 0) |
d7df2c79 | 2954 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2955 | |
c3e69d58 | 2956 | again: |
709c0486 AJ |
2957 | #ifdef SCRAMBLE_DELAYED_REFS |
2958 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2959 | #endif | |
d9a0540a | 2960 | trans->can_flush_pending_bgs = false; |
d7df2c79 JB |
2961 | ret = __btrfs_run_delayed_refs(trans, root, count); |
2962 | if (ret < 0) { | |
2963 | btrfs_abort_transaction(trans, root, ret); | |
2964 | return ret; | |
eb099670 | 2965 | } |
c3e69d58 | 2966 | |
56bec294 | 2967 | if (run_all) { |
d7df2c79 | 2968 | if (!list_empty(&trans->new_bgs)) |
ea658bad | 2969 | btrfs_create_pending_block_groups(trans, root); |
ea658bad | 2970 | |
d7df2c79 | 2971 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2972 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2973 | if (!node) { |
2974 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2975 | goto out; |
d7df2c79 | 2976 | } |
c3e69d58 | 2977 | count = (unsigned long)-1; |
e9d0b13b | 2978 | |
56bec294 | 2979 | while (node) { |
c46effa6 LB |
2980 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2981 | href_node); | |
2982 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2983 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2984 | |
c46effa6 | 2985 | ref = &head->node; |
56bec294 CM |
2986 | atomic_inc(&ref->refs); |
2987 | ||
2988 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2989 | /* |
2990 | * Mutex was contended, block until it's | |
2991 | * released and try again | |
2992 | */ | |
56bec294 CM |
2993 | mutex_lock(&head->mutex); |
2994 | mutex_unlock(&head->mutex); | |
2995 | ||
2996 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2997 | cond_resched(); |
56bec294 | 2998 | goto again; |
c46effa6 LB |
2999 | } else { |
3000 | WARN_ON(1); | |
56bec294 CM |
3001 | } |
3002 | node = rb_next(node); | |
3003 | } | |
3004 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 3005 | cond_resched(); |
56bec294 | 3006 | goto again; |
5f39d397 | 3007 | } |
54aa1f4d | 3008 | out: |
edf39272 | 3009 | assert_qgroups_uptodate(trans); |
d9a0540a | 3010 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3011 | return 0; |
3012 | } | |
3013 | ||
5d4f98a2 YZ |
3014 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
3015 | struct btrfs_root *root, | |
3016 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 3017 | int level, int is_data) |
5d4f98a2 YZ |
3018 | { |
3019 | struct btrfs_delayed_extent_op *extent_op; | |
3020 | int ret; | |
3021 | ||
78a6184a | 3022 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3023 | if (!extent_op) |
3024 | return -ENOMEM; | |
3025 | ||
3026 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3027 | extent_op->update_flags = true; |
3028 | extent_op->update_key = false; | |
3029 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3030 | extent_op->level = level; |
5d4f98a2 | 3031 | |
66d7e7f0 AJ |
3032 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
3033 | num_bytes, extent_op); | |
5d4f98a2 | 3034 | if (ret) |
78a6184a | 3035 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3036 | return ret; |
3037 | } | |
3038 | ||
3039 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
3040 | struct btrfs_root *root, | |
3041 | struct btrfs_path *path, | |
3042 | u64 objectid, u64 offset, u64 bytenr) | |
3043 | { | |
3044 | struct btrfs_delayed_ref_head *head; | |
3045 | struct btrfs_delayed_ref_node *ref; | |
3046 | struct btrfs_delayed_data_ref *data_ref; | |
3047 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
3048 | int ret = 0; |
3049 | ||
5d4f98a2 YZ |
3050 | delayed_refs = &trans->transaction->delayed_refs; |
3051 | spin_lock(&delayed_refs->lock); | |
3052 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
d7df2c79 JB |
3053 | if (!head) { |
3054 | spin_unlock(&delayed_refs->lock); | |
3055 | return 0; | |
3056 | } | |
5d4f98a2 YZ |
3057 | |
3058 | if (!mutex_trylock(&head->mutex)) { | |
3059 | atomic_inc(&head->node.refs); | |
3060 | spin_unlock(&delayed_refs->lock); | |
3061 | ||
b3b4aa74 | 3062 | btrfs_release_path(path); |
5d4f98a2 | 3063 | |
8cc33e5c DS |
3064 | /* |
3065 | * Mutex was contended, block until it's released and let | |
3066 | * caller try again | |
3067 | */ | |
5d4f98a2 YZ |
3068 | mutex_lock(&head->mutex); |
3069 | mutex_unlock(&head->mutex); | |
3070 | btrfs_put_delayed_ref(&head->node); | |
3071 | return -EAGAIN; | |
3072 | } | |
d7df2c79 | 3073 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3074 | |
d7df2c79 | 3075 | spin_lock(&head->lock); |
c6fc2454 | 3076 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3077 | /* If it's a shared ref we know a cross reference exists */ |
3078 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3079 | ret = 1; | |
3080 | break; | |
3081 | } | |
5d4f98a2 | 3082 | |
d7df2c79 | 3083 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3084 | |
d7df2c79 JB |
3085 | /* |
3086 | * If our ref doesn't match the one we're currently looking at | |
3087 | * then we have a cross reference. | |
3088 | */ | |
3089 | if (data_ref->root != root->root_key.objectid || | |
3090 | data_ref->objectid != objectid || | |
3091 | data_ref->offset != offset) { | |
3092 | ret = 1; | |
3093 | break; | |
3094 | } | |
5d4f98a2 | 3095 | } |
d7df2c79 | 3096 | spin_unlock(&head->lock); |
5d4f98a2 | 3097 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3098 | return ret; |
3099 | } | |
3100 | ||
3101 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
3102 | struct btrfs_root *root, | |
3103 | struct btrfs_path *path, | |
3104 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
3105 | { |
3106 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 3107 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3108 | struct btrfs_extent_data_ref *ref; |
3109 | struct btrfs_extent_inline_ref *iref; | |
3110 | struct btrfs_extent_item *ei; | |
f321e491 | 3111 | struct btrfs_key key; |
5d4f98a2 | 3112 | u32 item_size; |
be20aa9d | 3113 | int ret; |
925baedd | 3114 | |
be20aa9d | 3115 | key.objectid = bytenr; |
31840ae1 | 3116 | key.offset = (u64)-1; |
f321e491 | 3117 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3118 | |
be20aa9d CM |
3119 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3120 | if (ret < 0) | |
3121 | goto out; | |
79787eaa | 3122 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3123 | |
3124 | ret = -ENOENT; | |
3125 | if (path->slots[0] == 0) | |
31840ae1 | 3126 | goto out; |
be20aa9d | 3127 | |
31840ae1 | 3128 | path->slots[0]--; |
f321e491 | 3129 | leaf = path->nodes[0]; |
5d4f98a2 | 3130 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3131 | |
5d4f98a2 | 3132 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3133 | goto out; |
f321e491 | 3134 | |
5d4f98a2 YZ |
3135 | ret = 1; |
3136 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3137 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3138 | if (item_size < sizeof(*ei)) { | |
3139 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3140 | goto out; | |
3141 | } | |
3142 | #endif | |
3143 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3144 | |
5d4f98a2 YZ |
3145 | if (item_size != sizeof(*ei) + |
3146 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3147 | goto out; | |
be20aa9d | 3148 | |
5d4f98a2 YZ |
3149 | if (btrfs_extent_generation(leaf, ei) <= |
3150 | btrfs_root_last_snapshot(&root->root_item)) | |
3151 | goto out; | |
3152 | ||
3153 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3154 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3155 | BTRFS_EXTENT_DATA_REF_KEY) | |
3156 | goto out; | |
3157 | ||
3158 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3159 | if (btrfs_extent_refs(leaf, ei) != | |
3160 | btrfs_extent_data_ref_count(leaf, ref) || | |
3161 | btrfs_extent_data_ref_root(leaf, ref) != | |
3162 | root->root_key.objectid || | |
3163 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3164 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3165 | goto out; | |
3166 | ||
3167 | ret = 0; | |
3168 | out: | |
3169 | return ret; | |
3170 | } | |
3171 | ||
3172 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3173 | struct btrfs_root *root, | |
3174 | u64 objectid, u64 offset, u64 bytenr) | |
3175 | { | |
3176 | struct btrfs_path *path; | |
3177 | int ret; | |
3178 | int ret2; | |
3179 | ||
3180 | path = btrfs_alloc_path(); | |
3181 | if (!path) | |
3182 | return -ENOENT; | |
3183 | ||
3184 | do { | |
3185 | ret = check_committed_ref(trans, root, path, objectid, | |
3186 | offset, bytenr); | |
3187 | if (ret && ret != -ENOENT) | |
f321e491 | 3188 | goto out; |
80ff3856 | 3189 | |
5d4f98a2 YZ |
3190 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3191 | offset, bytenr); | |
3192 | } while (ret2 == -EAGAIN); | |
3193 | ||
3194 | if (ret2 && ret2 != -ENOENT) { | |
3195 | ret = ret2; | |
3196 | goto out; | |
f321e491 | 3197 | } |
5d4f98a2 YZ |
3198 | |
3199 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3200 | ret = 0; | |
be20aa9d | 3201 | out: |
80ff3856 | 3202 | btrfs_free_path(path); |
f0486c68 YZ |
3203 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3204 | WARN_ON(ret > 0); | |
f321e491 | 3205 | return ret; |
be20aa9d | 3206 | } |
c5739bba | 3207 | |
5d4f98a2 | 3208 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3209 | struct btrfs_root *root, |
5d4f98a2 | 3210 | struct extent_buffer *buf, |
e339a6b0 | 3211 | int full_backref, int inc) |
31840ae1 ZY |
3212 | { |
3213 | u64 bytenr; | |
5d4f98a2 YZ |
3214 | u64 num_bytes; |
3215 | u64 parent; | |
31840ae1 | 3216 | u64 ref_root; |
31840ae1 | 3217 | u32 nritems; |
31840ae1 ZY |
3218 | struct btrfs_key key; |
3219 | struct btrfs_file_extent_item *fi; | |
3220 | int i; | |
3221 | int level; | |
3222 | int ret = 0; | |
31840ae1 | 3223 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
b06c4bf5 | 3224 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3225 | |
fccb84c9 DS |
3226 | |
3227 | if (btrfs_test_is_dummy_root(root)) | |
faa2dbf0 | 3228 | return 0; |
fccb84c9 | 3229 | |
31840ae1 | 3230 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3231 | nritems = btrfs_header_nritems(buf); |
3232 | level = btrfs_header_level(buf); | |
3233 | ||
27cdeb70 | 3234 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3235 | return 0; |
31840ae1 | 3236 | |
5d4f98a2 YZ |
3237 | if (inc) |
3238 | process_func = btrfs_inc_extent_ref; | |
3239 | else | |
3240 | process_func = btrfs_free_extent; | |
31840ae1 | 3241 | |
5d4f98a2 YZ |
3242 | if (full_backref) |
3243 | parent = buf->start; | |
3244 | else | |
3245 | parent = 0; | |
3246 | ||
3247 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3248 | if (level == 0) { |
5d4f98a2 | 3249 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3250 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3251 | continue; |
5d4f98a2 | 3252 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3253 | struct btrfs_file_extent_item); |
3254 | if (btrfs_file_extent_type(buf, fi) == | |
3255 | BTRFS_FILE_EXTENT_INLINE) | |
3256 | continue; | |
3257 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3258 | if (bytenr == 0) | |
3259 | continue; | |
5d4f98a2 YZ |
3260 | |
3261 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3262 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3263 | ret = process_func(trans, root, bytenr, num_bytes, | |
3264 | parent, ref_root, key.objectid, | |
b06c4bf5 | 3265 | key.offset); |
31840ae1 ZY |
3266 | if (ret) |
3267 | goto fail; | |
3268 | } else { | |
5d4f98a2 | 3269 | bytenr = btrfs_node_blockptr(buf, i); |
707e8a07 | 3270 | num_bytes = root->nodesize; |
5d4f98a2 | 3271 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3272 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3273 | if (ret) |
3274 | goto fail; | |
3275 | } | |
3276 | } | |
3277 | return 0; | |
3278 | fail: | |
5d4f98a2 YZ |
3279 | return ret; |
3280 | } | |
3281 | ||
3282 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3283 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3284 | { |
e339a6b0 | 3285 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3286 | } |
3287 | ||
3288 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3289 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3290 | { |
e339a6b0 | 3291 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3292 | } |
3293 | ||
9078a3e1 CM |
3294 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3295 | struct btrfs_root *root, | |
3296 | struct btrfs_path *path, | |
3297 | struct btrfs_block_group_cache *cache) | |
3298 | { | |
3299 | int ret; | |
9078a3e1 | 3300 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3301 | unsigned long bi; |
3302 | struct extent_buffer *leaf; | |
9078a3e1 | 3303 | |
9078a3e1 | 3304 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3305 | if (ret) { |
3306 | if (ret > 0) | |
3307 | ret = -ENOENT; | |
54aa1f4d | 3308 | goto fail; |
df95e7f0 | 3309 | } |
5f39d397 CM |
3310 | |
3311 | leaf = path->nodes[0]; | |
3312 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3313 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3314 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3315 | fail: |
24b89d08 | 3316 | btrfs_release_path(path); |
df95e7f0 | 3317 | return ret; |
9078a3e1 CM |
3318 | |
3319 | } | |
3320 | ||
4a8c9a62 YZ |
3321 | static struct btrfs_block_group_cache * |
3322 | next_block_group(struct btrfs_root *root, | |
3323 | struct btrfs_block_group_cache *cache) | |
3324 | { | |
3325 | struct rb_node *node; | |
292cbd51 | 3326 | |
4a8c9a62 | 3327 | spin_lock(&root->fs_info->block_group_cache_lock); |
292cbd51 FM |
3328 | |
3329 | /* If our block group was removed, we need a full search. */ | |
3330 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3331 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3332 | ||
3333 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3334 | btrfs_put_block_group(cache); | |
3335 | cache = btrfs_lookup_first_block_group(root->fs_info, | |
3336 | next_bytenr); | |
3337 | return cache; | |
3338 | } | |
4a8c9a62 YZ |
3339 | node = rb_next(&cache->cache_node); |
3340 | btrfs_put_block_group(cache); | |
3341 | if (node) { | |
3342 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3343 | cache_node); | |
11dfe35a | 3344 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3345 | } else |
3346 | cache = NULL; | |
3347 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3348 | return cache; | |
3349 | } | |
3350 | ||
0af3d00b JB |
3351 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3352 | struct btrfs_trans_handle *trans, | |
3353 | struct btrfs_path *path) | |
3354 | { | |
3355 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3356 | struct inode *inode = NULL; | |
3357 | u64 alloc_hint = 0; | |
2b20982e | 3358 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3359 | u64 num_pages = 0; |
0af3d00b JB |
3360 | int retries = 0; |
3361 | int ret = 0; | |
3362 | ||
3363 | /* | |
3364 | * If this block group is smaller than 100 megs don't bother caching the | |
3365 | * block group. | |
3366 | */ | |
ee22184b | 3367 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3368 | spin_lock(&block_group->lock); |
3369 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3370 | spin_unlock(&block_group->lock); | |
3371 | return 0; | |
3372 | } | |
3373 | ||
0c0ef4bc JB |
3374 | if (trans->aborted) |
3375 | return 0; | |
0af3d00b JB |
3376 | again: |
3377 | inode = lookup_free_space_inode(root, block_group, path); | |
3378 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3379 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3380 | btrfs_release_path(path); |
0af3d00b JB |
3381 | goto out; |
3382 | } | |
3383 | ||
3384 | if (IS_ERR(inode)) { | |
3385 | BUG_ON(retries); | |
3386 | retries++; | |
3387 | ||
3388 | if (block_group->ro) | |
3389 | goto out_free; | |
3390 | ||
3391 | ret = create_free_space_inode(root, trans, block_group, path); | |
3392 | if (ret) | |
3393 | goto out_free; | |
3394 | goto again; | |
3395 | } | |
3396 | ||
5b0e95bf JB |
3397 | /* We've already setup this transaction, go ahead and exit */ |
3398 | if (block_group->cache_generation == trans->transid && | |
3399 | i_size_read(inode)) { | |
3400 | dcs = BTRFS_DC_SETUP; | |
3401 | goto out_put; | |
3402 | } | |
3403 | ||
0af3d00b JB |
3404 | /* |
3405 | * We want to set the generation to 0, that way if anything goes wrong | |
3406 | * from here on out we know not to trust this cache when we load up next | |
3407 | * time. | |
3408 | */ | |
3409 | BTRFS_I(inode)->generation = 0; | |
3410 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3411 | if (ret) { |
3412 | /* | |
3413 | * So theoretically we could recover from this, simply set the | |
3414 | * super cache generation to 0 so we know to invalidate the | |
3415 | * cache, but then we'd have to keep track of the block groups | |
3416 | * that fail this way so we know we _have_ to reset this cache | |
3417 | * before the next commit or risk reading stale cache. So to | |
3418 | * limit our exposure to horrible edge cases lets just abort the | |
3419 | * transaction, this only happens in really bad situations | |
3420 | * anyway. | |
3421 | */ | |
3422 | btrfs_abort_transaction(trans, root, ret); | |
3423 | goto out_put; | |
3424 | } | |
0af3d00b JB |
3425 | WARN_ON(ret); |
3426 | ||
3427 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3428 | ret = btrfs_check_trunc_cache_free_space(root, |
3429 | &root->fs_info->global_block_rsv); | |
3430 | if (ret) | |
3431 | goto out_put; | |
3432 | ||
1bbc621e | 3433 | ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode); |
0af3d00b JB |
3434 | if (ret) |
3435 | goto out_put; | |
3436 | } | |
3437 | ||
3438 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3439 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
e4c88f00 | 3440 | !btrfs_test_opt(root, SPACE_CACHE)) { |
cf7c1ef6 LB |
3441 | /* |
3442 | * don't bother trying to write stuff out _if_ | |
3443 | * a) we're not cached, | |
3444 | * b) we're with nospace_cache mount option. | |
3445 | */ | |
2b20982e | 3446 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3447 | spin_unlock(&block_group->lock); |
3448 | goto out_put; | |
3449 | } | |
3450 | spin_unlock(&block_group->lock); | |
3451 | ||
2968b1f4 JB |
3452 | /* |
3453 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3454 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3455 | */ | |
3456 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3457 | ret = -ENOSPC; | |
3458 | goto out_put; | |
3459 | } | |
3460 | ||
6fc823b1 JB |
3461 | /* |
3462 | * Try to preallocate enough space based on how big the block group is. | |
3463 | * Keep in mind this has to include any pinned space which could end up | |
3464 | * taking up quite a bit since it's not folded into the other space | |
3465 | * cache. | |
3466 | */ | |
ee22184b | 3467 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3468 | if (!num_pages) |
3469 | num_pages = 1; | |
3470 | ||
0af3d00b | 3471 | num_pages *= 16; |
09cbfeaf | 3472 | num_pages *= PAGE_SIZE; |
0af3d00b | 3473 | |
7cf5b976 | 3474 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3475 | if (ret) |
3476 | goto out_put; | |
3477 | ||
3478 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3479 | num_pages, num_pages, | |
3480 | &alloc_hint); | |
2968b1f4 JB |
3481 | /* |
3482 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3483 | * of metadata or split extents when writing the cache out, which means | |
3484 | * we can enospc if we are heavily fragmented in addition to just normal | |
3485 | * out of space conditions. So if we hit this just skip setting up any | |
3486 | * other block groups for this transaction, maybe we'll unpin enough | |
3487 | * space the next time around. | |
3488 | */ | |
2b20982e JB |
3489 | if (!ret) |
3490 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3491 | else if (ret == -ENOSPC) |
3492 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
7cf5b976 | 3493 | btrfs_free_reserved_data_space(inode, 0, num_pages); |
c09544e0 | 3494 | |
0af3d00b JB |
3495 | out_put: |
3496 | iput(inode); | |
3497 | out_free: | |
b3b4aa74 | 3498 | btrfs_release_path(path); |
0af3d00b JB |
3499 | out: |
3500 | spin_lock(&block_group->lock); | |
e65cbb94 | 3501 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3502 | block_group->cache_generation = trans->transid; |
2b20982e | 3503 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3504 | spin_unlock(&block_group->lock); |
3505 | ||
3506 | return ret; | |
3507 | } | |
3508 | ||
dcdf7f6d JB |
3509 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
3510 | struct btrfs_root *root) | |
3511 | { | |
3512 | struct btrfs_block_group_cache *cache, *tmp; | |
3513 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3514 | struct btrfs_path *path; | |
3515 | ||
3516 | if (list_empty(&cur_trans->dirty_bgs) || | |
3517 | !btrfs_test_opt(root, SPACE_CACHE)) | |
3518 | return 0; | |
3519 | ||
3520 | path = btrfs_alloc_path(); | |
3521 | if (!path) | |
3522 | return -ENOMEM; | |
3523 | ||
3524 | /* Could add new block groups, use _safe just in case */ | |
3525 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3526 | dirty_list) { | |
3527 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3528 | cache_save_setup(cache, trans, path); | |
3529 | } | |
3530 | ||
3531 | btrfs_free_path(path); | |
3532 | return 0; | |
3533 | } | |
3534 | ||
1bbc621e CM |
3535 | /* |
3536 | * transaction commit does final block group cache writeback during a | |
3537 | * critical section where nothing is allowed to change the FS. This is | |
3538 | * required in order for the cache to actually match the block group, | |
3539 | * but can introduce a lot of latency into the commit. | |
3540 | * | |
3541 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3542 | * cache IO. There's a chance we'll have to redo some of it if the | |
3543 | * block group changes again during the commit, but it greatly reduces | |
3544 | * the commit latency by getting rid of the easy block groups while | |
3545 | * we're still allowing others to join the commit. | |
3546 | */ | |
3547 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
96b5179d | 3548 | struct btrfs_root *root) |
9078a3e1 | 3549 | { |
4a8c9a62 | 3550 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3551 | struct btrfs_transaction *cur_trans = trans->transaction; |
3552 | int ret = 0; | |
c9dc4c65 | 3553 | int should_put; |
1bbc621e CM |
3554 | struct btrfs_path *path = NULL; |
3555 | LIST_HEAD(dirty); | |
3556 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3557 | int num_started = 0; |
1bbc621e CM |
3558 | int loops = 0; |
3559 | ||
3560 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3561 | if (list_empty(&cur_trans->dirty_bgs)) { |
3562 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3563 | return 0; | |
1bbc621e | 3564 | } |
b58d1a9e | 3565 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3566 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3567 | |
1bbc621e | 3568 | again: |
1bbc621e CM |
3569 | /* |
3570 | * make sure all the block groups on our dirty list actually | |
3571 | * exist | |
3572 | */ | |
3573 | btrfs_create_pending_block_groups(trans, root); | |
3574 | ||
3575 | if (!path) { | |
3576 | path = btrfs_alloc_path(); | |
3577 | if (!path) | |
3578 | return -ENOMEM; | |
3579 | } | |
3580 | ||
b58d1a9e FM |
3581 | /* |
3582 | * cache_write_mutex is here only to save us from balance or automatic | |
3583 | * removal of empty block groups deleting this block group while we are | |
3584 | * writing out the cache | |
3585 | */ | |
3586 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3587 | while (!list_empty(&dirty)) { |
3588 | cache = list_first_entry(&dirty, | |
3589 | struct btrfs_block_group_cache, | |
3590 | dirty_list); | |
1bbc621e CM |
3591 | /* |
3592 | * this can happen if something re-dirties a block | |
3593 | * group that is already under IO. Just wait for it to | |
3594 | * finish and then do it all again | |
3595 | */ | |
3596 | if (!list_empty(&cache->io_list)) { | |
3597 | list_del_init(&cache->io_list); | |
3598 | btrfs_wait_cache_io(root, trans, cache, | |
3599 | &cache->io_ctl, path, | |
3600 | cache->key.objectid); | |
3601 | btrfs_put_block_group(cache); | |
3602 | } | |
3603 | ||
3604 | ||
3605 | /* | |
3606 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3607 | * if it should update the cache_state. Don't delete | |
3608 | * until after we wait. | |
3609 | * | |
3610 | * Since we're not running in the commit critical section | |
3611 | * we need the dirty_bgs_lock to protect from update_block_group | |
3612 | */ | |
3613 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3614 | list_del_init(&cache->dirty_list); | |
3615 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3616 | ||
3617 | should_put = 1; | |
3618 | ||
3619 | cache_save_setup(cache, trans, path); | |
3620 | ||
3621 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3622 | cache->io_ctl.inode = NULL; | |
3623 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3624 | if (ret == 0 && cache->io_ctl.inode) { | |
3625 | num_started++; | |
3626 | should_put = 0; | |
3627 | ||
3628 | /* | |
3629 | * the cache_write_mutex is protecting | |
3630 | * the io_list | |
3631 | */ | |
3632 | list_add_tail(&cache->io_list, io); | |
3633 | } else { | |
3634 | /* | |
3635 | * if we failed to write the cache, the | |
3636 | * generation will be bad and life goes on | |
3637 | */ | |
3638 | ret = 0; | |
3639 | } | |
3640 | } | |
ff1f8250 | 3641 | if (!ret) { |
1bbc621e | 3642 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3643 | /* |
3644 | * Our block group might still be attached to the list | |
3645 | * of new block groups in the transaction handle of some | |
3646 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3647 | * means its block group item isn't yet in the extent | |
3648 | * tree. If this happens ignore the error, as we will | |
3649 | * try again later in the critical section of the | |
3650 | * transaction commit. | |
3651 | */ | |
3652 | if (ret == -ENOENT) { | |
3653 | ret = 0; | |
3654 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3655 | if (list_empty(&cache->dirty_list)) { | |
3656 | list_add_tail(&cache->dirty_list, | |
3657 | &cur_trans->dirty_bgs); | |
3658 | btrfs_get_block_group(cache); | |
3659 | } | |
3660 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3661 | } else if (ret) { | |
3662 | btrfs_abort_transaction(trans, root, ret); | |
3663 | } | |
3664 | } | |
1bbc621e CM |
3665 | |
3666 | /* if its not on the io list, we need to put the block group */ | |
3667 | if (should_put) | |
3668 | btrfs_put_block_group(cache); | |
3669 | ||
3670 | if (ret) | |
3671 | break; | |
b58d1a9e FM |
3672 | |
3673 | /* | |
3674 | * Avoid blocking other tasks for too long. It might even save | |
3675 | * us from writing caches for block groups that are going to be | |
3676 | * removed. | |
3677 | */ | |
3678 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3679 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3680 | } |
b58d1a9e | 3681 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3682 | |
3683 | /* | |
3684 | * go through delayed refs for all the stuff we've just kicked off | |
3685 | * and then loop back (just once) | |
3686 | */ | |
3687 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
3688 | if (!ret && loops == 0) { | |
3689 | loops++; | |
3690 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3691 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3692 | /* |
3693 | * dirty_bgs_lock protects us from concurrent block group | |
3694 | * deletes too (not just cache_write_mutex). | |
3695 | */ | |
3696 | if (!list_empty(&dirty)) { | |
3697 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3698 | goto again; | |
3699 | } | |
1bbc621e | 3700 | spin_unlock(&cur_trans->dirty_bgs_lock); |
1bbc621e CM |
3701 | } |
3702 | ||
3703 | btrfs_free_path(path); | |
3704 | return ret; | |
3705 | } | |
3706 | ||
3707 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3708 | struct btrfs_root *root) | |
3709 | { | |
3710 | struct btrfs_block_group_cache *cache; | |
3711 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3712 | int ret = 0; | |
3713 | int should_put; | |
3714 | struct btrfs_path *path; | |
3715 | struct list_head *io = &cur_trans->io_bgs; | |
3716 | int num_started = 0; | |
9078a3e1 CM |
3717 | |
3718 | path = btrfs_alloc_path(); | |
3719 | if (!path) | |
3720 | return -ENOMEM; | |
3721 | ||
ce93ec54 | 3722 | /* |
e44081ef FM |
3723 | * Even though we are in the critical section of the transaction commit, |
3724 | * we can still have concurrent tasks adding elements to this | |
3725 | * transaction's list of dirty block groups. These tasks correspond to | |
3726 | * endio free space workers started when writeback finishes for a | |
3727 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3728 | * allocate new block groups as a result of COWing nodes of the root | |
3729 | * tree when updating the free space inode. The writeback for the space | |
3730 | * caches is triggered by an earlier call to | |
3731 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3732 | * loop. | |
3733 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3734 | * delayed refs to make sure we have the best chance at doing this all |
3735 | * in one shot. | |
3736 | */ | |
e44081ef | 3737 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3738 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3739 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3740 | struct btrfs_block_group_cache, | |
3741 | dirty_list); | |
c9dc4c65 CM |
3742 | |
3743 | /* | |
3744 | * this can happen if cache_save_setup re-dirties a block | |
3745 | * group that is already under IO. Just wait for it to | |
3746 | * finish and then do it all again | |
3747 | */ | |
3748 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3749 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3750 | list_del_init(&cache->io_list); |
3751 | btrfs_wait_cache_io(root, trans, cache, | |
3752 | &cache->io_ctl, path, | |
3753 | cache->key.objectid); | |
3754 | btrfs_put_block_group(cache); | |
e44081ef | 3755 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3756 | } |
3757 | ||
1bbc621e CM |
3758 | /* |
3759 | * don't remove from the dirty list until after we've waited | |
3760 | * on any pending IO | |
3761 | */ | |
ce93ec54 | 3762 | list_del_init(&cache->dirty_list); |
e44081ef | 3763 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3764 | should_put = 1; |
3765 | ||
1bbc621e | 3766 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3767 | |
ce93ec54 | 3768 | if (!ret) |
c9dc4c65 CM |
3769 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1); |
3770 | ||
3771 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3772 | cache->io_ctl.inode = NULL; | |
3773 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3774 | if (ret == 0 && cache->io_ctl.inode) { | |
3775 | num_started++; | |
3776 | should_put = 0; | |
1bbc621e | 3777 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3778 | } else { |
3779 | /* | |
3780 | * if we failed to write the cache, the | |
3781 | * generation will be bad and life goes on | |
3782 | */ | |
3783 | ret = 0; | |
3784 | } | |
3785 | } | |
ff1f8250 | 3786 | if (!ret) { |
ce93ec54 | 3787 | ret = write_one_cache_group(trans, root, path, cache); |
2bc0bb5f FM |
3788 | /* |
3789 | * One of the free space endio workers might have | |
3790 | * created a new block group while updating a free space | |
3791 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3792 | * and hasn't released its transaction handle yet, in | |
3793 | * which case the new block group is still attached to | |
3794 | * its transaction handle and its creation has not | |
3795 | * finished yet (no block group item in the extent tree | |
3796 | * yet, etc). If this is the case, wait for all free | |
3797 | * space endio workers to finish and retry. This is a | |
3798 | * a very rare case so no need for a more efficient and | |
3799 | * complex approach. | |
3800 | */ | |
3801 | if (ret == -ENOENT) { | |
3802 | wait_event(cur_trans->writer_wait, | |
3803 | atomic_read(&cur_trans->num_writers) == 1); | |
3804 | ret = write_one_cache_group(trans, root, path, | |
3805 | cache); | |
3806 | } | |
ff1f8250 FM |
3807 | if (ret) |
3808 | btrfs_abort_transaction(trans, root, ret); | |
3809 | } | |
c9dc4c65 CM |
3810 | |
3811 | /* if its not on the io list, we need to put the block group */ | |
3812 | if (should_put) | |
3813 | btrfs_put_block_group(cache); | |
e44081ef | 3814 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3815 | } |
e44081ef | 3816 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3817 | |
1bbc621e CM |
3818 | while (!list_empty(io)) { |
3819 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3820 | io_list); |
3821 | list_del_init(&cache->io_list); | |
c9dc4c65 CM |
3822 | btrfs_wait_cache_io(root, trans, cache, |
3823 | &cache->io_ctl, path, cache->key.objectid); | |
0cb59c99 JB |
3824 | btrfs_put_block_group(cache); |
3825 | } | |
3826 | ||
9078a3e1 | 3827 | btrfs_free_path(path); |
ce93ec54 | 3828 | return ret; |
9078a3e1 CM |
3829 | } |
3830 | ||
d2fb3437 YZ |
3831 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3832 | { | |
3833 | struct btrfs_block_group_cache *block_group; | |
3834 | int readonly = 0; | |
3835 | ||
3836 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3837 | if (!block_group || block_group->ro) | |
3838 | readonly = 1; | |
3839 | if (block_group) | |
fa9c0d79 | 3840 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3841 | return readonly; |
3842 | } | |
3843 | ||
f78c436c FM |
3844 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3845 | { | |
3846 | struct btrfs_block_group_cache *bg; | |
3847 | bool ret = true; | |
3848 | ||
3849 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3850 | if (!bg) | |
3851 | return false; | |
3852 | ||
3853 | spin_lock(&bg->lock); | |
3854 | if (bg->ro) | |
3855 | ret = false; | |
3856 | else | |
3857 | atomic_inc(&bg->nocow_writers); | |
3858 | spin_unlock(&bg->lock); | |
3859 | ||
3860 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3861 | if (!ret) | |
3862 | btrfs_put_block_group(bg); | |
3863 | ||
3864 | return ret; | |
3865 | ||
3866 | } | |
3867 | ||
3868 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3869 | { | |
3870 | struct btrfs_block_group_cache *bg; | |
3871 | ||
3872 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3873 | ASSERT(bg); | |
3874 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3875 | wake_up_atomic_t(&bg->nocow_writers); | |
3876 | /* | |
3877 | * Once for our lookup and once for the lookup done by a previous call | |
3878 | * to btrfs_inc_nocow_writers() | |
3879 | */ | |
3880 | btrfs_put_block_group(bg); | |
3881 | btrfs_put_block_group(bg); | |
3882 | } | |
3883 | ||
3884 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3885 | { | |
3886 | schedule(); | |
3887 | return 0; | |
3888 | } | |
3889 | ||
3890 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
3891 | { | |
3892 | wait_on_atomic_t(&bg->nocow_writers, | |
3893 | btrfs_wait_nocow_writers_atomic_t, | |
3894 | TASK_UNINTERRUPTIBLE); | |
3895 | } | |
3896 | ||
6ab0a202 JM |
3897 | static const char *alloc_name(u64 flags) |
3898 | { | |
3899 | switch (flags) { | |
3900 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3901 | return "mixed"; | |
3902 | case BTRFS_BLOCK_GROUP_METADATA: | |
3903 | return "metadata"; | |
3904 | case BTRFS_BLOCK_GROUP_DATA: | |
3905 | return "data"; | |
3906 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3907 | return "system"; | |
3908 | default: | |
3909 | WARN_ON(1); | |
3910 | return "invalid-combination"; | |
3911 | }; | |
3912 | } | |
3913 | ||
593060d7 CM |
3914 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3915 | u64 total_bytes, u64 bytes_used, | |
e40edf2d | 3916 | u64 bytes_readonly, |
593060d7 CM |
3917 | struct btrfs_space_info **space_info) |
3918 | { | |
3919 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3920 | int i; |
3921 | int factor; | |
b150a4f1 | 3922 | int ret; |
b742bb82 YZ |
3923 | |
3924 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3925 | BTRFS_BLOCK_GROUP_RAID10)) | |
3926 | factor = 2; | |
3927 | else | |
3928 | factor = 1; | |
593060d7 CM |
3929 | |
3930 | found = __find_space_info(info, flags); | |
3931 | if (found) { | |
25179201 | 3932 | spin_lock(&found->lock); |
593060d7 | 3933 | found->total_bytes += total_bytes; |
89a55897 | 3934 | found->disk_total += total_bytes * factor; |
593060d7 | 3935 | found->bytes_used += bytes_used; |
b742bb82 | 3936 | found->disk_used += bytes_used * factor; |
e40edf2d | 3937 | found->bytes_readonly += bytes_readonly; |
2e6e5183 FM |
3938 | if (total_bytes > 0) |
3939 | found->full = 0; | |
25179201 | 3940 | spin_unlock(&found->lock); |
593060d7 CM |
3941 | *space_info = found; |
3942 | return 0; | |
3943 | } | |
c146afad | 3944 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3945 | if (!found) |
3946 | return -ENOMEM; | |
3947 | ||
908c7f19 | 3948 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3949 | if (ret) { |
3950 | kfree(found); | |
3951 | return ret; | |
3952 | } | |
3953 | ||
c1895442 | 3954 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3955 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3956 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3957 | spin_lock_init(&found->lock); |
52ba6929 | 3958 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3959 | found->total_bytes = total_bytes; |
89a55897 | 3960 | found->disk_total = total_bytes * factor; |
593060d7 | 3961 | found->bytes_used = bytes_used; |
b742bb82 | 3962 | found->disk_used = bytes_used * factor; |
593060d7 | 3963 | found->bytes_pinned = 0; |
e8569813 | 3964 | found->bytes_reserved = 0; |
e40edf2d | 3965 | found->bytes_readonly = bytes_readonly; |
f0486c68 | 3966 | found->bytes_may_use = 0; |
6af3e3ad | 3967 | found->full = 0; |
4f4db217 | 3968 | found->max_extent_size = 0; |
0e4f8f88 | 3969 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3970 | found->chunk_alloc = 0; |
fdb5effd JB |
3971 | found->flush = 0; |
3972 | init_waitqueue_head(&found->wait); | |
633c0aad | 3973 | INIT_LIST_HEAD(&found->ro_bgs); |
6ab0a202 JM |
3974 | |
3975 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3976 | info->space_info_kobj, "%s", | |
3977 | alloc_name(found->flags)); | |
3978 | if (ret) { | |
3979 | kfree(found); | |
3980 | return ret; | |
3981 | } | |
3982 | ||
593060d7 | 3983 | *space_info = found; |
4184ea7f | 3984 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3985 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3986 | info->data_sinfo = found; | |
6ab0a202 JM |
3987 | |
3988 | return ret; | |
593060d7 CM |
3989 | } |
3990 | ||
8790d502 CM |
3991 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3992 | { | |
899c81ea ID |
3993 | u64 extra_flags = chunk_to_extended(flags) & |
3994 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3995 | |
de98ced9 | 3996 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3997 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3998 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3999 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4000 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4001 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4002 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4003 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4004 | } |
593060d7 | 4005 | |
fc67c450 ID |
4006 | /* |
4007 | * returns target flags in extended format or 0 if restripe for this | |
4008 | * chunk_type is not in progress | |
c6664b42 ID |
4009 | * |
4010 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4011 | */ |
4012 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4013 | { | |
4014 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4015 | u64 target = 0; | |
4016 | ||
fc67c450 ID |
4017 | if (!bctl) |
4018 | return 0; | |
4019 | ||
4020 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4021 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4022 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4023 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4024 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4025 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4026 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4027 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4028 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4029 | } | |
4030 | ||
4031 | return target; | |
4032 | } | |
4033 | ||
a46d11a8 ID |
4034 | /* |
4035 | * @flags: available profiles in extended format (see ctree.h) | |
4036 | * | |
e4d8ec0f ID |
4037 | * Returns reduced profile in chunk format. If profile changing is in |
4038 | * progress (either running or paused) picks the target profile (if it's | |
4039 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4040 | */ |
48a3b636 | 4041 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 4042 | { |
95669976 | 4043 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
fc67c450 | 4044 | u64 target; |
9c170b26 ZL |
4045 | u64 raid_type; |
4046 | u64 allowed = 0; | |
a061fc8d | 4047 | |
fc67c450 ID |
4048 | /* |
4049 | * see if restripe for this chunk_type is in progress, if so | |
4050 | * try to reduce to the target profile | |
4051 | */ | |
e4d8ec0f | 4052 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
4053 | target = get_restripe_target(root->fs_info, flags); |
4054 | if (target) { | |
4055 | /* pick target profile only if it's already available */ | |
4056 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 4057 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 4058 | return extended_to_chunk(target); |
e4d8ec0f ID |
4059 | } |
4060 | } | |
4061 | spin_unlock(&root->fs_info->balance_lock); | |
4062 | ||
53b381b3 | 4063 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4064 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4065 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4066 | allowed |= btrfs_raid_group[raid_type]; | |
4067 | } | |
4068 | allowed &= flags; | |
4069 | ||
4070 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4071 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4072 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4073 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4074 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4075 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4076 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4077 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4078 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4079 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4080 | ||
4081 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4082 | ||
4083 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4084 | } |
4085 | ||
f8213bdc | 4086 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
6a63209f | 4087 | { |
de98ced9 | 4088 | unsigned seq; |
f8213bdc | 4089 | u64 flags; |
de98ced9 MX |
4090 | |
4091 | do { | |
f8213bdc | 4092 | flags = orig_flags; |
de98ced9 MX |
4093 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
4094 | ||
4095 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4096 | flags |= root->fs_info->avail_data_alloc_bits; | |
4097 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4098 | flags |= root->fs_info->avail_system_alloc_bits; | |
4099 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4100 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
4101 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 4102 | |
b742bb82 | 4103 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
4104 | } |
4105 | ||
6d07bcec | 4106 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 4107 | { |
b742bb82 | 4108 | u64 flags; |
53b381b3 | 4109 | u64 ret; |
9ed74f2d | 4110 | |
b742bb82 YZ |
4111 | if (data) |
4112 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4113 | else if (root == root->fs_info->chunk_root) | |
4114 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 4115 | else |
b742bb82 | 4116 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4117 | |
53b381b3 DW |
4118 | ret = get_alloc_profile(root, flags); |
4119 | return ret; | |
6a63209f | 4120 | } |
9ed74f2d | 4121 | |
4ceff079 | 4122 | int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes) |
6a63209f | 4123 | { |
6a63209f | 4124 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 4125 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 4126 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4127 | u64 used; |
94b947b2 | 4128 | int ret = 0; |
c99f1b0c ZL |
4129 | int need_commit = 2; |
4130 | int have_pinned_space; | |
6a63209f | 4131 | |
6a63209f | 4132 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 4133 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 4134 | |
9dced186 | 4135 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4136 | need_commit = 0; |
9dced186 | 4137 | ASSERT(current->journal_info); |
0af3d00b JB |
4138 | } |
4139 | ||
b4d7c3c9 | 4140 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4141 | if (!data_sinfo) |
4142 | goto alloc; | |
9ed74f2d | 4143 | |
6a63209f JB |
4144 | again: |
4145 | /* make sure we have enough space to handle the data first */ | |
4146 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
4147 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
4148 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
4149 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
4150 | |
4151 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4152 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4153 | |
6a63209f JB |
4154 | /* |
4155 | * if we don't have enough free bytes in this space then we need | |
4156 | * to alloc a new chunk. | |
4157 | */ | |
b9fd47cd | 4158 | if (!data_sinfo->full) { |
6a63209f | 4159 | u64 alloc_target; |
9ed74f2d | 4160 | |
0e4f8f88 | 4161 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4162 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4163 | alloc: |
6a63209f | 4164 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
4165 | /* |
4166 | * It is ugly that we don't call nolock join | |
4167 | * transaction for the free space inode case here. | |
4168 | * But it is safe because we only do the data space | |
4169 | * reservation for the free space cache in the | |
4170 | * transaction context, the common join transaction | |
4171 | * just increase the counter of the current transaction | |
4172 | * handler, doesn't try to acquire the trans_lock of | |
4173 | * the fs. | |
4174 | */ | |
7a7eaa40 | 4175 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4176 | if (IS_ERR(trans)) |
4177 | return PTR_ERR(trans); | |
9ed74f2d | 4178 | |
6a63209f | 4179 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
4180 | alloc_target, |
4181 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 4182 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
4183 | if (ret < 0) { |
4184 | if (ret != -ENOSPC) | |
4185 | return ret; | |
c99f1b0c ZL |
4186 | else { |
4187 | have_pinned_space = 1; | |
d52a5b5f | 4188 | goto commit_trans; |
c99f1b0c | 4189 | } |
d52a5b5f | 4190 | } |
9ed74f2d | 4191 | |
b4d7c3c9 LZ |
4192 | if (!data_sinfo) |
4193 | data_sinfo = fs_info->data_sinfo; | |
4194 | ||
6a63209f JB |
4195 | goto again; |
4196 | } | |
f2bb8f5c JB |
4197 | |
4198 | /* | |
b150a4f1 | 4199 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4200 | * allocation, and no removed chunk in current transaction, |
4201 | * don't bother committing the transaction. | |
f2bb8f5c | 4202 | */ |
c99f1b0c ZL |
4203 | have_pinned_space = percpu_counter_compare( |
4204 | &data_sinfo->total_bytes_pinned, | |
4205 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4206 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4207 | |
4e06bdd6 | 4208 | /* commit the current transaction and try again */ |
d52a5b5f | 4209 | commit_trans: |
c99f1b0c | 4210 | if (need_commit && |
a4abeea4 | 4211 | !atomic_read(&root->fs_info->open_ioctl_trans)) { |
c99f1b0c | 4212 | need_commit--; |
b150a4f1 | 4213 | |
e1746e83 ZL |
4214 | if (need_commit > 0) { |
4215 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
578def7c | 4216 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); |
e1746e83 | 4217 | } |
9a4e7276 | 4218 | |
7a7eaa40 | 4219 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4220 | if (IS_ERR(trans)) |
4221 | return PTR_ERR(trans); | |
c99f1b0c | 4222 | if (have_pinned_space >= 0 || |
3204d33c JB |
4223 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4224 | &trans->transaction->flags) || | |
c99f1b0c | 4225 | need_commit > 0) { |
94b947b2 ZL |
4226 | ret = btrfs_commit_transaction(trans, root); |
4227 | if (ret) | |
4228 | return ret; | |
d7c15171 | 4229 | /* |
c2d6cb16 FM |
4230 | * The cleaner kthread might still be doing iput |
4231 | * operations. Wait for it to finish so that | |
4232 | * more space is released. | |
d7c15171 | 4233 | */ |
c2d6cb16 FM |
4234 | mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex); |
4235 | mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4236 | goto again; |
4237 | } else { | |
4238 | btrfs_end_transaction(trans, root); | |
4239 | } | |
4e06bdd6 | 4240 | } |
9ed74f2d | 4241 | |
cab45e22 JM |
4242 | trace_btrfs_space_reservation(root->fs_info, |
4243 | "space_info:enospc", | |
4244 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4245 | return -ENOSPC; |
4246 | } | |
4247 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 4248 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 4249 | data_sinfo->flags, bytes, 1); |
6a63209f | 4250 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4251 | |
237c0e9f | 4252 | return ret; |
9ed74f2d | 4253 | } |
6a63209f | 4254 | |
4ceff079 QW |
4255 | /* |
4256 | * New check_data_free_space() with ability for precious data reservation | |
4257 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4258 | * add a new function first and then replace it. | |
4259 | */ | |
7cf5b976 | 4260 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 QW |
4261 | { |
4262 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4263 | int ret; | |
4264 | ||
4265 | /* align the range */ | |
4266 | len = round_up(start + len, root->sectorsize) - | |
4267 | round_down(start, root->sectorsize); | |
4268 | start = round_down(start, root->sectorsize); | |
4269 | ||
4270 | ret = btrfs_alloc_data_chunk_ondemand(inode, len); | |
4271 | if (ret < 0) | |
4272 | return ret; | |
4273 | ||
94ed938a QW |
4274 | /* |
4275 | * Use new btrfs_qgroup_reserve_data to reserve precious data space | |
4276 | * | |
4277 | * TODO: Find a good method to avoid reserve data space for NOCOW | |
4278 | * range, but don't impact performance on quota disable case. | |
4279 | */ | |
4ceff079 QW |
4280 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
4281 | return ret; | |
4282 | } | |
4283 | ||
4ceff079 QW |
4284 | /* |
4285 | * Called if we need to clear a data reservation for this inode | |
4286 | * Normally in a error case. | |
4287 | * | |
51773bec QW |
4288 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4289 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4290 | * Like clear_bit_hook(). | |
4ceff079 | 4291 | */ |
51773bec QW |
4292 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4293 | u64 len) | |
4ceff079 QW |
4294 | { |
4295 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4296 | struct btrfs_space_info *data_sinfo; | |
4297 | ||
4298 | /* Make sure the range is aligned to sectorsize */ | |
4299 | len = round_up(start + len, root->sectorsize) - | |
4300 | round_down(start, root->sectorsize); | |
4301 | start = round_down(start, root->sectorsize); | |
4302 | ||
4ceff079 QW |
4303 | data_sinfo = root->fs_info->data_sinfo; |
4304 | spin_lock(&data_sinfo->lock); | |
4305 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4306 | data_sinfo->bytes_may_use = 0; | |
4307 | else | |
4308 | data_sinfo->bytes_may_use -= len; | |
4309 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4310 | data_sinfo->flags, len, 0); | |
4311 | spin_unlock(&data_sinfo->lock); | |
4312 | } | |
4313 | ||
51773bec QW |
4314 | /* |
4315 | * Called if we need to clear a data reservation for this inode | |
4316 | * Normally in a error case. | |
4317 | * | |
01327610 | 4318 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4319 | * space framework. |
4320 | */ | |
4321 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4322 | { | |
4323 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4324 | btrfs_qgroup_free_data(inode, start, len); | |
4325 | } | |
4326 | ||
97e728d4 | 4327 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4328 | { |
97e728d4 JB |
4329 | struct list_head *head = &info->space_info; |
4330 | struct btrfs_space_info *found; | |
e3ccfa98 | 4331 | |
97e728d4 JB |
4332 | rcu_read_lock(); |
4333 | list_for_each_entry_rcu(found, head, list) { | |
4334 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4335 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4336 | } |
97e728d4 | 4337 | rcu_read_unlock(); |
e3ccfa98 JB |
4338 | } |
4339 | ||
3c76cd84 MX |
4340 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4341 | { | |
4342 | return (global->size << 1); | |
4343 | } | |
4344 | ||
e5bc2458 | 4345 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 4346 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4347 | { |
fb25e914 | 4348 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 4349 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4350 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4351 | u64 thresh; |
e3ccfa98 | 4352 | |
0e4f8f88 CM |
4353 | if (force == CHUNK_ALLOC_FORCE) |
4354 | return 1; | |
4355 | ||
fb25e914 JB |
4356 | /* |
4357 | * We need to take into account the global rsv because for all intents | |
4358 | * and purposes it's used space. Don't worry about locking the | |
4359 | * global_rsv, it doesn't change except when the transaction commits. | |
4360 | */ | |
54338b5c | 4361 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4362 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4363 | |
0e4f8f88 CM |
4364 | /* |
4365 | * in limited mode, we want to have some free space up to | |
4366 | * about 1% of the FS size. | |
4367 | */ | |
4368 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 4369 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
ee22184b | 4370 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 CM |
4371 | |
4372 | if (num_bytes - num_allocated < thresh) | |
4373 | return 1; | |
4374 | } | |
0e4f8f88 | 4375 | |
ee22184b | 4376 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) |
14ed0ca6 | 4377 | return 0; |
424499db | 4378 | return 1; |
32c00aff JB |
4379 | } |
4380 | ||
39c2d7fa | 4381 | static u64 get_profile_num_devs(struct btrfs_root *root, u64 type) |
15d1ff81 LB |
4382 | { |
4383 | u64 num_dev; | |
4384 | ||
53b381b3 DW |
4385 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4386 | BTRFS_BLOCK_GROUP_RAID0 | | |
4387 | BTRFS_BLOCK_GROUP_RAID5 | | |
4388 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
4389 | num_dev = root->fs_info->fs_devices->rw_devices; |
4390 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
4391 | num_dev = 2; | |
4392 | else | |
4393 | num_dev = 1; /* DUP or single */ | |
4394 | ||
39c2d7fa | 4395 | return num_dev; |
15d1ff81 LB |
4396 | } |
4397 | ||
39c2d7fa FM |
4398 | /* |
4399 | * If @is_allocation is true, reserve space in the system space info necessary | |
4400 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4401 | * removing a chunk. | |
4402 | */ | |
4403 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
4404 | struct btrfs_root *root, | |
4617ea3a | 4405 | u64 type) |
15d1ff81 LB |
4406 | { |
4407 | struct btrfs_space_info *info; | |
4408 | u64 left; | |
4409 | u64 thresh; | |
4fbcdf66 | 4410 | int ret = 0; |
39c2d7fa | 4411 | u64 num_devs; |
4fbcdf66 FM |
4412 | |
4413 | /* | |
4414 | * Needed because we can end up allocating a system chunk and for an | |
4415 | * atomic and race free space reservation in the chunk block reserve. | |
4416 | */ | |
4417 | ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex)); | |
15d1ff81 LB |
4418 | |
4419 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4420 | spin_lock(&info->lock); | |
4421 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
4fbcdf66 FM |
4422 | info->bytes_reserved - info->bytes_readonly - |
4423 | info->bytes_may_use; | |
15d1ff81 LB |
4424 | spin_unlock(&info->lock); |
4425 | ||
39c2d7fa FM |
4426 | num_devs = get_profile_num_devs(root, type); |
4427 | ||
4428 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
4617ea3a FM |
4429 | thresh = btrfs_calc_trunc_metadata_size(root, num_devs) + |
4430 | btrfs_calc_trans_metadata_size(root, 1); | |
39c2d7fa | 4431 | |
15d1ff81 | 4432 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
c2cf52eb SK |
4433 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
4434 | left, thresh, type); | |
15d1ff81 LB |
4435 | dump_space_info(info, 0, 0); |
4436 | } | |
4437 | ||
4438 | if (left < thresh) { | |
4439 | u64 flags; | |
4440 | ||
4441 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
4fbcdf66 FM |
4442 | /* |
4443 | * Ignore failure to create system chunk. We might end up not | |
4444 | * needing it, as we might not need to COW all nodes/leafs from | |
4445 | * the paths we visit in the chunk tree (they were already COWed | |
4446 | * or created in the current transaction for example). | |
4447 | */ | |
4448 | ret = btrfs_alloc_chunk(trans, root, flags); | |
4449 | } | |
4450 | ||
4451 | if (!ret) { | |
4452 | ret = btrfs_block_rsv_add(root->fs_info->chunk_root, | |
4453 | &root->fs_info->chunk_block_rsv, | |
4454 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
4455 | if (!ret) | |
4456 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4457 | } |
4458 | } | |
4459 | ||
6324fbf3 | 4460 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 4461 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 4462 | { |
6324fbf3 | 4463 | struct btrfs_space_info *space_info; |
97e728d4 | 4464 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 4465 | int wait_for_alloc = 0; |
9ed74f2d | 4466 | int ret = 0; |
9ed74f2d | 4467 | |
c6b305a8 JB |
4468 | /* Don't re-enter if we're already allocating a chunk */ |
4469 | if (trans->allocating_chunk) | |
4470 | return -ENOSPC; | |
4471 | ||
6324fbf3 | 4472 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
4473 | if (!space_info) { |
4474 | ret = update_space_info(extent_root->fs_info, flags, | |
e40edf2d | 4475 | 0, 0, 0, &space_info); |
79787eaa | 4476 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4477 | } |
79787eaa | 4478 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4479 | |
6d74119f | 4480 | again: |
25179201 | 4481 | spin_lock(&space_info->lock); |
9e622d6b | 4482 | if (force < space_info->force_alloc) |
0e4f8f88 | 4483 | force = space_info->force_alloc; |
25179201 | 4484 | if (space_info->full) { |
09fb99a6 FDBM |
4485 | if (should_alloc_chunk(extent_root, space_info, force)) |
4486 | ret = -ENOSPC; | |
4487 | else | |
4488 | ret = 0; | |
25179201 | 4489 | spin_unlock(&space_info->lock); |
09fb99a6 | 4490 | return ret; |
9ed74f2d JB |
4491 | } |
4492 | ||
698d0082 | 4493 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 4494 | spin_unlock(&space_info->lock); |
6d74119f JB |
4495 | return 0; |
4496 | } else if (space_info->chunk_alloc) { | |
4497 | wait_for_alloc = 1; | |
4498 | } else { | |
4499 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4500 | } |
0e4f8f88 | 4501 | |
25179201 | 4502 | spin_unlock(&space_info->lock); |
9ed74f2d | 4503 | |
6d74119f JB |
4504 | mutex_lock(&fs_info->chunk_mutex); |
4505 | ||
4506 | /* | |
4507 | * The chunk_mutex is held throughout the entirety of a chunk | |
4508 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4509 | * other guy is done and we need to recheck and see if we should | |
4510 | * allocate. | |
4511 | */ | |
4512 | if (wait_for_alloc) { | |
4513 | mutex_unlock(&fs_info->chunk_mutex); | |
4514 | wait_for_alloc = 0; | |
4515 | goto again; | |
4516 | } | |
4517 | ||
c6b305a8 JB |
4518 | trans->allocating_chunk = true; |
4519 | ||
67377734 JB |
4520 | /* |
4521 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4522 | * allocating mixed chunks instead of individual chunks. | |
4523 | */ | |
4524 | if (btrfs_mixed_space_info(space_info)) | |
4525 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4526 | ||
97e728d4 JB |
4527 | /* |
4528 | * if we're doing a data chunk, go ahead and make sure that | |
4529 | * we keep a reasonable number of metadata chunks allocated in the | |
4530 | * FS as well. | |
4531 | */ | |
9ed74f2d | 4532 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4533 | fs_info->data_chunk_allocations++; |
4534 | if (!(fs_info->data_chunk_allocations % | |
4535 | fs_info->metadata_ratio)) | |
4536 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4537 | } |
4538 | ||
15d1ff81 LB |
4539 | /* |
4540 | * Check if we have enough space in SYSTEM chunk because we may need | |
4541 | * to update devices. | |
4542 | */ | |
4617ea3a | 4543 | check_system_chunk(trans, extent_root, flags); |
15d1ff81 | 4544 | |
2b82032c | 4545 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 4546 | trans->allocating_chunk = false; |
92b8e897 | 4547 | |
9ed74f2d | 4548 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4549 | if (ret < 0 && ret != -ENOSPC) |
4550 | goto out; | |
9ed74f2d | 4551 | if (ret) |
6324fbf3 | 4552 | space_info->full = 1; |
424499db YZ |
4553 | else |
4554 | ret = 1; | |
6d74119f | 4555 | |
0e4f8f88 | 4556 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4557 | out: |
6d74119f | 4558 | space_info->chunk_alloc = 0; |
9ed74f2d | 4559 | spin_unlock(&space_info->lock); |
a25c75d5 | 4560 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4561 | /* |
4562 | * When we allocate a new chunk we reserve space in the chunk block | |
4563 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4564 | * add new nodes/leafs to it if we end up needing to do it when | |
4565 | * inserting the chunk item and updating device items as part of the | |
4566 | * second phase of chunk allocation, performed by | |
4567 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4568 | * large number of new block groups to create in our transaction | |
4569 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4570 | * in extreme cases - like having a single transaction create many new | |
4571 | * block groups when starting to write out the free space caches of all | |
4572 | * the block groups that were made dirty during the lifetime of the | |
4573 | * transaction. | |
4574 | */ | |
d9a0540a | 4575 | if (trans->can_flush_pending_bgs && |
ee22184b | 4576 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
00d80e34 FM |
4577 | btrfs_create_pending_block_groups(trans, trans->root); |
4578 | btrfs_trans_release_chunk_metadata(trans); | |
4579 | } | |
0f9dd46c | 4580 | return ret; |
6324fbf3 | 4581 | } |
9ed74f2d | 4582 | |
a80c8dcf JB |
4583 | static int can_overcommit(struct btrfs_root *root, |
4584 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4585 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4586 | { |
96f1bb57 | 4587 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 4588 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3c76cd84 | 4589 | u64 space_size; |
a80c8dcf JB |
4590 | u64 avail; |
4591 | u64 used; | |
4592 | ||
4593 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
4594 | space_info->bytes_pinned + space_info->bytes_readonly; |
4595 | ||
96f1bb57 JB |
4596 | /* |
4597 | * We only want to allow over committing if we have lots of actual space | |
4598 | * free, but if we don't have enough space to handle the global reserve | |
4599 | * space then we could end up having a real enospc problem when trying | |
4600 | * to allocate a chunk or some other such important allocation. | |
4601 | */ | |
3c76cd84 MX |
4602 | spin_lock(&global_rsv->lock); |
4603 | space_size = calc_global_rsv_need_space(global_rsv); | |
4604 | spin_unlock(&global_rsv->lock); | |
4605 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4606 | return 0; |
4607 | ||
4608 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
4609 | |
4610 | spin_lock(&root->fs_info->free_chunk_lock); | |
4611 | avail = root->fs_info->free_chunk_space; | |
4612 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4613 | ||
4614 | /* | |
4615 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4616 | * space is actually useable. For raid56, the space info used |
4617 | * doesn't include the parity drive, so we don't have to | |
4618 | * change the math | |
a80c8dcf JB |
4619 | */ |
4620 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4621 | BTRFS_BLOCK_GROUP_RAID1 | | |
4622 | BTRFS_BLOCK_GROUP_RAID10)) | |
4623 | avail >>= 1; | |
4624 | ||
4625 | /* | |
561c294d MX |
4626 | * If we aren't flushing all things, let us overcommit up to |
4627 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4628 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4629 | */ |
08e007d2 | 4630 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4631 | avail >>= 3; |
a80c8dcf | 4632 | else |
14575aef | 4633 | avail >>= 1; |
a80c8dcf | 4634 | |
14575aef | 4635 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4636 | return 1; |
4637 | return 0; | |
4638 | } | |
4639 | ||
48a3b636 | 4640 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
6c255e67 | 4641 | unsigned long nr_pages, int nr_items) |
da633a42 MX |
4642 | { |
4643 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 4644 | |
925a6efb JB |
4645 | if (down_read_trylock(&sb->s_umount)) { |
4646 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4647 | up_read(&sb->s_umount); | |
4648 | } else { | |
da633a42 MX |
4649 | /* |
4650 | * We needn't worry the filesystem going from r/w to r/o though | |
4651 | * we don't acquire ->s_umount mutex, because the filesystem | |
4652 | * should guarantee the delalloc inodes list be empty after | |
4653 | * the filesystem is readonly(all dirty pages are written to | |
4654 | * the disk). | |
4655 | */ | |
6c255e67 | 4656 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
98ad69cf | 4657 | if (!current->journal_info) |
578def7c FM |
4658 | btrfs_wait_ordered_roots(root->fs_info, nr_items, |
4659 | 0, (u64)-1); | |
da633a42 MX |
4660 | } |
4661 | } | |
4662 | ||
18cd8ea6 MX |
4663 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
4664 | { | |
4665 | u64 bytes; | |
4666 | int nr; | |
4667 | ||
4668 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4669 | nr = (int)div64_u64(to_reclaim, bytes); | |
4670 | if (!nr) | |
4671 | nr = 1; | |
4672 | return nr; | |
4673 | } | |
4674 | ||
ee22184b | 4675 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4676 | |
9ed74f2d | 4677 | /* |
5da9d01b | 4678 | * shrink metadata reservation for delalloc |
9ed74f2d | 4679 | */ |
f4c738c2 JB |
4680 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4681 | bool wait_ordered) | |
5da9d01b | 4682 | { |
0ca1f7ce | 4683 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4684 | struct btrfs_space_info *space_info; |
663350ac | 4685 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4686 | u64 delalloc_bytes; |
5da9d01b | 4687 | u64 max_reclaim; |
b1953bce | 4688 | long time_left; |
d3ee29e3 MX |
4689 | unsigned long nr_pages; |
4690 | int loops; | |
b0244199 | 4691 | int items; |
08e007d2 | 4692 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4693 | |
c61a16a7 | 4694 | /* Calc the number of the pages we need flush for space reservation */ |
b0244199 | 4695 | items = calc_reclaim_items_nr(root, to_reclaim); |
8eb0dfdb | 4696 | to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4697 | |
663350ac | 4698 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4699 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4700 | space_info = block_rsv->space_info; |
bf9022e0 | 4701 | |
963d678b MX |
4702 | delalloc_bytes = percpu_counter_sum_positive( |
4703 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4704 | if (delalloc_bytes == 0) { |
fdb5effd | 4705 | if (trans) |
f4c738c2 | 4706 | return; |
38c135af | 4707 | if (wait_ordered) |
578def7c FM |
4708 | btrfs_wait_ordered_roots(root->fs_info, items, |
4709 | 0, (u64)-1); | |
f4c738c2 | 4710 | return; |
fdb5effd JB |
4711 | } |
4712 | ||
d3ee29e3 | 4713 | loops = 0; |
f4c738c2 JB |
4714 | while (delalloc_bytes && loops < 3) { |
4715 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4716 | nr_pages = max_reclaim >> PAGE_SHIFT; |
6c255e67 | 4717 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
dea31f52 JB |
4718 | /* |
4719 | * We need to wait for the async pages to actually start before | |
4720 | * we do anything. | |
4721 | */ | |
9f3a074d MX |
4722 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
4723 | if (!max_reclaim) | |
4724 | goto skip_async; | |
4725 | ||
4726 | if (max_reclaim <= nr_pages) | |
4727 | max_reclaim = 0; | |
4728 | else | |
4729 | max_reclaim -= nr_pages; | |
dea31f52 | 4730 | |
9f3a074d MX |
4731 | wait_event(root->fs_info->async_submit_wait, |
4732 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4733 | (int)max_reclaim); | |
4734 | skip_async: | |
08e007d2 MX |
4735 | if (!trans) |
4736 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4737 | else | |
4738 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4739 | spin_lock(&space_info->lock); |
08e007d2 | 4740 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4741 | spin_unlock(&space_info->lock); |
4742 | break; | |
4743 | } | |
0019f10d | 4744 | spin_unlock(&space_info->lock); |
5da9d01b | 4745 | |
36e39c40 | 4746 | loops++; |
f104d044 | 4747 | if (wait_ordered && !trans) { |
578def7c FM |
4748 | btrfs_wait_ordered_roots(root->fs_info, items, |
4749 | 0, (u64)-1); | |
f104d044 | 4750 | } else { |
f4c738c2 | 4751 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4752 | if (time_left) |
4753 | break; | |
4754 | } | |
963d678b MX |
4755 | delalloc_bytes = percpu_counter_sum_positive( |
4756 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4757 | } |
5da9d01b YZ |
4758 | } |
4759 | ||
663350ac JB |
4760 | /** |
4761 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4762 | * @root - the root we're allocating for | |
4763 | * @bytes - the number of bytes we want to reserve | |
4764 | * @force - force the commit | |
8bb8ab2e | 4765 | * |
663350ac JB |
4766 | * This will check to make sure that committing the transaction will actually |
4767 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4768 | * will return -ENOSPC. | |
8bb8ab2e | 4769 | */ |
663350ac JB |
4770 | static int may_commit_transaction(struct btrfs_root *root, |
4771 | struct btrfs_space_info *space_info, | |
4772 | u64 bytes, int force) | |
4773 | { | |
4774 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4775 | struct btrfs_trans_handle *trans; | |
4776 | ||
4777 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4778 | if (trans) | |
4779 | return -EAGAIN; | |
4780 | ||
4781 | if (force) | |
4782 | goto commit; | |
4783 | ||
4784 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4785 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4786 | bytes) >= 0) |
663350ac | 4787 | goto commit; |
663350ac JB |
4788 | |
4789 | /* | |
4790 | * See if there is some space in the delayed insertion reservation for | |
4791 | * this reservation. | |
4792 | */ | |
4793 | if (space_info != delayed_rsv->space_info) | |
4794 | return -ENOSPC; | |
4795 | ||
4796 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4797 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4798 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4799 | spin_unlock(&delayed_rsv->lock); |
4800 | return -ENOSPC; | |
4801 | } | |
4802 | spin_unlock(&delayed_rsv->lock); | |
4803 | ||
4804 | commit: | |
4805 | trans = btrfs_join_transaction(root); | |
4806 | if (IS_ERR(trans)) | |
4807 | return -ENOSPC; | |
4808 | ||
4809 | return btrfs_commit_transaction(trans, root); | |
4810 | } | |
4811 | ||
96c3f433 | 4812 | enum flush_state { |
67b0fd63 JB |
4813 | FLUSH_DELAYED_ITEMS_NR = 1, |
4814 | FLUSH_DELAYED_ITEMS = 2, | |
4815 | FLUSH_DELALLOC = 3, | |
4816 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4817 | ALLOC_CHUNK = 5, |
4818 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4819 | }; |
4820 | ||
4821 | static int flush_space(struct btrfs_root *root, | |
4822 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4823 | u64 orig_bytes, int state) | |
4824 | { | |
4825 | struct btrfs_trans_handle *trans; | |
4826 | int nr; | |
f4c738c2 | 4827 | int ret = 0; |
96c3f433 JB |
4828 | |
4829 | switch (state) { | |
96c3f433 JB |
4830 | case FLUSH_DELAYED_ITEMS_NR: |
4831 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 MX |
4832 | if (state == FLUSH_DELAYED_ITEMS_NR) |
4833 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4834 | else | |
96c3f433 | 4835 | nr = -1; |
18cd8ea6 | 4836 | |
96c3f433 JB |
4837 | trans = btrfs_join_transaction(root); |
4838 | if (IS_ERR(trans)) { | |
4839 | ret = PTR_ERR(trans); | |
4840 | break; | |
4841 | } | |
4842 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4843 | btrfs_end_transaction(trans, root); | |
4844 | break; | |
67b0fd63 JB |
4845 | case FLUSH_DELALLOC: |
4846 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4847 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4848 | state == FLUSH_DELALLOC_WAIT); |
4849 | break; | |
ea658bad JB |
4850 | case ALLOC_CHUNK: |
4851 | trans = btrfs_join_transaction(root); | |
4852 | if (IS_ERR(trans)) { | |
4853 | ret = PTR_ERR(trans); | |
4854 | break; | |
4855 | } | |
4856 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4857 | btrfs_get_alloc_profile(root, 0), |
4858 | CHUNK_ALLOC_NO_FORCE); | |
4859 | btrfs_end_transaction(trans, root); | |
4860 | if (ret == -ENOSPC) | |
4861 | ret = 0; | |
4862 | break; | |
96c3f433 JB |
4863 | case COMMIT_TRANS: |
4864 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4865 | break; | |
4866 | default: | |
4867 | ret = -ENOSPC; | |
4868 | break; | |
4869 | } | |
4870 | ||
4871 | return ret; | |
4872 | } | |
21c7e756 MX |
4873 | |
4874 | static inline u64 | |
4875 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4876 | struct btrfs_space_info *space_info) | |
4877 | { | |
4878 | u64 used; | |
4879 | u64 expected; | |
4880 | u64 to_reclaim; | |
4881 | ||
ee22184b | 4882 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
21c7e756 MX |
4883 | spin_lock(&space_info->lock); |
4884 | if (can_overcommit(root, space_info, to_reclaim, | |
4885 | BTRFS_RESERVE_FLUSH_ALL)) { | |
4886 | to_reclaim = 0; | |
4887 | goto out; | |
4888 | } | |
4889 | ||
4890 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4891 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4892 | space_info->bytes_may_use; | |
ee22184b | 4893 | if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL)) |
21c7e756 MX |
4894 | expected = div_factor_fine(space_info->total_bytes, 95); |
4895 | else | |
4896 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4897 | ||
4898 | if (used > expected) | |
4899 | to_reclaim = used - expected; | |
4900 | else | |
4901 | to_reclaim = 0; | |
4902 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4903 | space_info->bytes_reserved); | |
4904 | out: | |
4905 | spin_unlock(&space_info->lock); | |
4906 | ||
4907 | return to_reclaim; | |
4908 | } | |
4909 | ||
4910 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4911 | struct btrfs_fs_info *fs_info, u64 used) | |
4912 | { | |
365c5313 JB |
4913 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4914 | ||
4915 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4916 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4917 | return 0; |
4918 | ||
4919 | return (used >= thresh && !btrfs_fs_closing(fs_info) && | |
21c7e756 MX |
4920 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); |
4921 | } | |
4922 | ||
4923 | static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info, | |
25ce459c LB |
4924 | struct btrfs_fs_info *fs_info, |
4925 | int flush_state) | |
21c7e756 MX |
4926 | { |
4927 | u64 used; | |
4928 | ||
4929 | spin_lock(&space_info->lock); | |
25ce459c LB |
4930 | /* |
4931 | * We run out of space and have not got any free space via flush_space, | |
4932 | * so don't bother doing async reclaim. | |
4933 | */ | |
4934 | if (flush_state > COMMIT_TRANS && space_info->full) { | |
4935 | spin_unlock(&space_info->lock); | |
4936 | return 0; | |
4937 | } | |
4938 | ||
21c7e756 MX |
4939 | used = space_info->bytes_used + space_info->bytes_reserved + |
4940 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4941 | space_info->bytes_may_use; | |
4942 | if (need_do_async_reclaim(space_info, fs_info, used)) { | |
4943 | spin_unlock(&space_info->lock); | |
4944 | return 1; | |
4945 | } | |
4946 | spin_unlock(&space_info->lock); | |
4947 | ||
4948 | return 0; | |
4949 | } | |
4950 | ||
4951 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) | |
4952 | { | |
4953 | struct btrfs_fs_info *fs_info; | |
4954 | struct btrfs_space_info *space_info; | |
4955 | u64 to_reclaim; | |
4956 | int flush_state; | |
4957 | ||
4958 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4959 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4960 | ||
4961 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
4962 | space_info); | |
4963 | if (!to_reclaim) | |
4964 | return; | |
4965 | ||
4966 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4967 | do { | |
4968 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
4969 | to_reclaim, flush_state); | |
4970 | flush_state++; | |
25ce459c LB |
4971 | if (!btrfs_need_do_async_reclaim(space_info, fs_info, |
4972 | flush_state)) | |
21c7e756 | 4973 | return; |
365c5313 | 4974 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
4975 | } |
4976 | ||
4977 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
4978 | { | |
4979 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
4980 | } | |
4981 | ||
4a92b1b8 JB |
4982 | /** |
4983 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4984 | * @root - the root we're allocating for | |
4985 | * @block_rsv - the block_rsv we're allocating for | |
4986 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4987 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4988 | * |
01327610 | 4989 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
4990 | * with the block_rsv. If there is not enough space it will make an attempt to |
4991 | * flush out space to make room. It will do this by flushing delalloc if | |
4992 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4993 | * regain reservations will be made and this will fail if there is not enough | |
4994 | * space already. | |
8bb8ab2e | 4995 | */ |
4a92b1b8 | 4996 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4997 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4998 | u64 orig_bytes, |
4999 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5000 | { |
f0486c68 | 5001 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 5002 | u64 used; |
8bb8ab2e | 5003 | u64 num_bytes = orig_bytes; |
67b0fd63 | 5004 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 5005 | int ret = 0; |
fdb5effd | 5006 | bool flushing = false; |
9ed74f2d | 5007 | |
8bb8ab2e | 5008 | again: |
fdb5effd | 5009 | ret = 0; |
8bb8ab2e | 5010 | spin_lock(&space_info->lock); |
fdb5effd | 5011 | /* |
08e007d2 MX |
5012 | * We only want to wait if somebody other than us is flushing and we |
5013 | * are actually allowed to flush all things. | |
fdb5effd | 5014 | */ |
08e007d2 MX |
5015 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
5016 | space_info->flush) { | |
fdb5effd JB |
5017 | spin_unlock(&space_info->lock); |
5018 | /* | |
5019 | * If we have a trans handle we can't wait because the flusher | |
5020 | * may have to commit the transaction, which would mean we would | |
5021 | * deadlock since we are waiting for the flusher to finish, but | |
5022 | * hold the current transaction open. | |
5023 | */ | |
663350ac | 5024 | if (current->journal_info) |
fdb5effd | 5025 | return -EAGAIN; |
b9688bb8 AJ |
5026 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
5027 | /* Must have been killed, return */ | |
5028 | if (ret) | |
fdb5effd JB |
5029 | return -EINTR; |
5030 | ||
5031 | spin_lock(&space_info->lock); | |
5032 | } | |
5033 | ||
5034 | ret = -ENOSPC; | |
2bf64758 JB |
5035 | used = space_info->bytes_used + space_info->bytes_reserved + |
5036 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5037 | space_info->bytes_may_use; | |
9ed74f2d | 5038 | |
8bb8ab2e JB |
5039 | /* |
5040 | * The idea here is that we've not already over-reserved the block group | |
5041 | * then we can go ahead and save our reservation first and then start | |
5042 | * flushing if we need to. Otherwise if we've already overcommitted | |
5043 | * lets start flushing stuff first and then come back and try to make | |
5044 | * our reservation. | |
5045 | */ | |
2bf64758 JB |
5046 | if (used <= space_info->total_bytes) { |
5047 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 5048 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 5049 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 5050 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
5051 | ret = 0; |
5052 | } else { | |
5053 | /* | |
5054 | * Ok set num_bytes to orig_bytes since we aren't | |
5055 | * overocmmitted, this way we only try and reclaim what | |
5056 | * we need. | |
5057 | */ | |
5058 | num_bytes = orig_bytes; | |
5059 | } | |
5060 | } else { | |
5061 | /* | |
5062 | * Ok we're over committed, set num_bytes to the overcommitted | |
5063 | * amount plus the amount of bytes that we need for this | |
5064 | * reservation. | |
5065 | */ | |
2bf64758 | 5066 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 5067 | (orig_bytes * 2); |
8bb8ab2e | 5068 | } |
9ed74f2d | 5069 | |
44734ed1 JB |
5070 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
5071 | space_info->bytes_may_use += orig_bytes; | |
5072 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
5073 | space_info->flags, orig_bytes, | |
5074 | 1); | |
5075 | ret = 0; | |
2bf64758 JB |
5076 | } |
5077 | ||
8bb8ab2e JB |
5078 | /* |
5079 | * Couldn't make our reservation, save our place so while we're trying | |
5080 | * to reclaim space we can actually use it instead of somebody else | |
5081 | * stealing it from us. | |
08e007d2 MX |
5082 | * |
5083 | * We make the other tasks wait for the flush only when we can flush | |
5084 | * all things. | |
8bb8ab2e | 5085 | */ |
72bcd99d | 5086 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
5087 | flushing = true; |
5088 | space_info->flush = 1; | |
21c7e756 MX |
5089 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5090 | used += orig_bytes; | |
f6acfd50 JB |
5091 | /* |
5092 | * We will do the space reservation dance during log replay, | |
5093 | * which means we won't have fs_info->fs_root set, so don't do | |
5094 | * the async reclaim as we will panic. | |
5095 | */ | |
5096 | if (!root->fs_info->log_root_recovering && | |
5097 | need_do_async_reclaim(space_info, root->fs_info, used) && | |
21c7e756 MX |
5098 | !work_busy(&root->fs_info->async_reclaim_work)) |
5099 | queue_work(system_unbound_wq, | |
5100 | &root->fs_info->async_reclaim_work); | |
8bb8ab2e | 5101 | } |
f0486c68 | 5102 | spin_unlock(&space_info->lock); |
9ed74f2d | 5103 | |
08e007d2 | 5104 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 5105 | goto out; |
f0486c68 | 5106 | |
96c3f433 JB |
5107 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
5108 | flush_state); | |
5109 | flush_state++; | |
08e007d2 MX |
5110 | |
5111 | /* | |
5112 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
5113 | * would happen. So skip delalloc flush. | |
5114 | */ | |
5115 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
5116 | (flush_state == FLUSH_DELALLOC || | |
5117 | flush_state == FLUSH_DELALLOC_WAIT)) | |
5118 | flush_state = ALLOC_CHUNK; | |
5119 | ||
96c3f433 | 5120 | if (!ret) |
8bb8ab2e | 5121 | goto again; |
08e007d2 MX |
5122 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
5123 | flush_state < COMMIT_TRANS) | |
5124 | goto again; | |
5125 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
5126 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
5127 | goto again; |
5128 | ||
5129 | out: | |
5d80366e JB |
5130 | if (ret == -ENOSPC && |
5131 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5132 | struct btrfs_block_rsv *global_rsv = | |
5133 | &root->fs_info->global_block_rsv; | |
5134 | ||
5135 | if (block_rsv != global_rsv && | |
5136 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5137 | ret = 0; | |
5138 | } | |
cab45e22 JM |
5139 | if (ret == -ENOSPC) |
5140 | trace_btrfs_space_reservation(root->fs_info, | |
5141 | "space_info:enospc", | |
5142 | space_info->flags, orig_bytes, 1); | |
fdb5effd | 5143 | if (flushing) { |
8bb8ab2e | 5144 | spin_lock(&space_info->lock); |
fdb5effd JB |
5145 | space_info->flush = 0; |
5146 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 5147 | spin_unlock(&space_info->lock); |
f0486c68 | 5148 | } |
f0486c68 YZ |
5149 | return ret; |
5150 | } | |
5151 | ||
79787eaa JM |
5152 | static struct btrfs_block_rsv *get_block_rsv( |
5153 | const struct btrfs_trans_handle *trans, | |
5154 | const struct btrfs_root *root) | |
f0486c68 | 5155 | { |
4c13d758 JB |
5156 | struct btrfs_block_rsv *block_rsv = NULL; |
5157 | ||
e9cf439f AM |
5158 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
5159 | (root == root->fs_info->csum_root && trans->adding_csums) || | |
5160 | (root == root->fs_info->uuid_root)) | |
f7a81ea4 SB |
5161 | block_rsv = trans->block_rsv; |
5162 | ||
4c13d758 | 5163 | if (!block_rsv) |
f0486c68 YZ |
5164 | block_rsv = root->block_rsv; |
5165 | ||
5166 | if (!block_rsv) | |
5167 | block_rsv = &root->fs_info->empty_block_rsv; | |
5168 | ||
5169 | return block_rsv; | |
5170 | } | |
5171 | ||
5172 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5173 | u64 num_bytes) | |
5174 | { | |
5175 | int ret = -ENOSPC; | |
5176 | spin_lock(&block_rsv->lock); | |
5177 | if (block_rsv->reserved >= num_bytes) { | |
5178 | block_rsv->reserved -= num_bytes; | |
5179 | if (block_rsv->reserved < block_rsv->size) | |
5180 | block_rsv->full = 0; | |
5181 | ret = 0; | |
5182 | } | |
5183 | spin_unlock(&block_rsv->lock); | |
5184 | return ret; | |
5185 | } | |
5186 | ||
5187 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5188 | u64 num_bytes, int update_size) | |
5189 | { | |
5190 | spin_lock(&block_rsv->lock); | |
5191 | block_rsv->reserved += num_bytes; | |
5192 | if (update_size) | |
5193 | block_rsv->size += num_bytes; | |
5194 | else if (block_rsv->reserved >= block_rsv->size) | |
5195 | block_rsv->full = 1; | |
5196 | spin_unlock(&block_rsv->lock); | |
5197 | } | |
5198 | ||
d52be818 JB |
5199 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5200 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5201 | int min_factor) | |
5202 | { | |
5203 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5204 | u64 min_bytes; | |
5205 | ||
5206 | if (global_rsv->space_info != dest->space_info) | |
5207 | return -ENOSPC; | |
5208 | ||
5209 | spin_lock(&global_rsv->lock); | |
5210 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5211 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5212 | spin_unlock(&global_rsv->lock); | |
5213 | return -ENOSPC; | |
5214 | } | |
5215 | global_rsv->reserved -= num_bytes; | |
5216 | if (global_rsv->reserved < global_rsv->size) | |
5217 | global_rsv->full = 0; | |
5218 | spin_unlock(&global_rsv->lock); | |
5219 | ||
5220 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5221 | return 0; | |
5222 | } | |
5223 | ||
8c2a3ca2 JB |
5224 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5225 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5226 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5227 | { |
5228 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5229 | ||
5230 | spin_lock(&block_rsv->lock); | |
5231 | if (num_bytes == (u64)-1) | |
5232 | num_bytes = block_rsv->size; | |
5233 | block_rsv->size -= num_bytes; | |
5234 | if (block_rsv->reserved >= block_rsv->size) { | |
5235 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5236 | block_rsv->reserved = block_rsv->size; | |
5237 | block_rsv->full = 1; | |
5238 | } else { | |
5239 | num_bytes = 0; | |
5240 | } | |
5241 | spin_unlock(&block_rsv->lock); | |
5242 | ||
5243 | if (num_bytes > 0) { | |
5244 | if (dest) { | |
e9e22899 JB |
5245 | spin_lock(&dest->lock); |
5246 | if (!dest->full) { | |
5247 | u64 bytes_to_add; | |
5248 | ||
5249 | bytes_to_add = dest->size - dest->reserved; | |
5250 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5251 | dest->reserved += bytes_to_add; | |
5252 | if (dest->reserved >= dest->size) | |
5253 | dest->full = 1; | |
5254 | num_bytes -= bytes_to_add; | |
5255 | } | |
5256 | spin_unlock(&dest->lock); | |
5257 | } | |
5258 | if (num_bytes) { | |
f0486c68 | 5259 | spin_lock(&space_info->lock); |
fb25e914 | 5260 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5261 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5262 | space_info->flags, num_bytes, 0); |
f0486c68 | 5263 | spin_unlock(&space_info->lock); |
4e06bdd6 | 5264 | } |
9ed74f2d | 5265 | } |
f0486c68 | 5266 | } |
4e06bdd6 | 5267 | |
f0486c68 YZ |
5268 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
5269 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
5270 | { | |
5271 | int ret; | |
9ed74f2d | 5272 | |
f0486c68 YZ |
5273 | ret = block_rsv_use_bytes(src, num_bytes); |
5274 | if (ret) | |
5275 | return ret; | |
9ed74f2d | 5276 | |
f0486c68 | 5277 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
5278 | return 0; |
5279 | } | |
5280 | ||
66d8f3dd | 5281 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5282 | { |
f0486c68 YZ |
5283 | memset(rsv, 0, sizeof(*rsv)); |
5284 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5285 | rsv->type = type; |
f0486c68 YZ |
5286 | } |
5287 | ||
66d8f3dd MX |
5288 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
5289 | unsigned short type) | |
f0486c68 YZ |
5290 | { |
5291 | struct btrfs_block_rsv *block_rsv; | |
5292 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 5293 | |
f0486c68 YZ |
5294 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5295 | if (!block_rsv) | |
5296 | return NULL; | |
9ed74f2d | 5297 | |
66d8f3dd | 5298 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5299 | block_rsv->space_info = __find_space_info(fs_info, |
5300 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5301 | return block_rsv; |
5302 | } | |
9ed74f2d | 5303 | |
f0486c68 YZ |
5304 | void btrfs_free_block_rsv(struct btrfs_root *root, |
5305 | struct btrfs_block_rsv *rsv) | |
5306 | { | |
2aaa6655 JB |
5307 | if (!rsv) |
5308 | return; | |
dabdb640 JB |
5309 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
5310 | kfree(rsv); | |
9ed74f2d JB |
5311 | } |
5312 | ||
cdfb080e CM |
5313 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5314 | { | |
5315 | kfree(rsv); | |
5316 | } | |
5317 | ||
08e007d2 MX |
5318 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5319 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5320 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5321 | { |
f0486c68 | 5322 | int ret; |
9ed74f2d | 5323 | |
f0486c68 YZ |
5324 | if (num_bytes == 0) |
5325 | return 0; | |
8bb8ab2e | 5326 | |
61b520a9 | 5327 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5328 | if (!ret) { |
5329 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5330 | return 0; | |
5331 | } | |
9ed74f2d | 5332 | |
f0486c68 | 5333 | return ret; |
f0486c68 | 5334 | } |
9ed74f2d | 5335 | |
4a92b1b8 | 5336 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 5337 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5338 | { |
5339 | u64 num_bytes = 0; | |
f0486c68 | 5340 | int ret = -ENOSPC; |
9ed74f2d | 5341 | |
f0486c68 YZ |
5342 | if (!block_rsv) |
5343 | return 0; | |
9ed74f2d | 5344 | |
f0486c68 | 5345 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5346 | num_bytes = div_factor(block_rsv->size, min_factor); |
5347 | if (block_rsv->reserved >= num_bytes) | |
5348 | ret = 0; | |
5349 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5350 | |
36ba022a JB |
5351 | return ret; |
5352 | } | |
5353 | ||
08e007d2 MX |
5354 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5355 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5356 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5357 | { |
5358 | u64 num_bytes = 0; | |
5359 | int ret = -ENOSPC; | |
5360 | ||
5361 | if (!block_rsv) | |
5362 | return 0; | |
5363 | ||
5364 | spin_lock(&block_rsv->lock); | |
5365 | num_bytes = min_reserved; | |
13553e52 | 5366 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5367 | ret = 0; |
13553e52 | 5368 | else |
f0486c68 | 5369 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5370 | spin_unlock(&block_rsv->lock); |
13553e52 | 5371 | |
f0486c68 YZ |
5372 | if (!ret) |
5373 | return 0; | |
5374 | ||
aa38a711 | 5375 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5376 | if (!ret) { |
5377 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5378 | return 0; |
6a63209f | 5379 | } |
9ed74f2d | 5380 | |
13553e52 | 5381 | return ret; |
f0486c68 YZ |
5382 | } |
5383 | ||
5384 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
5385 | struct btrfs_block_rsv *dst_rsv, | |
5386 | u64 num_bytes) | |
5387 | { | |
5388 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
5389 | } | |
5390 | ||
5391 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
5392 | struct btrfs_block_rsv *block_rsv, | |
5393 | u64 num_bytes) | |
5394 | { | |
5395 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
17504584 | 5396 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5397 | block_rsv->space_info != global_rsv->space_info) |
5398 | global_rsv = NULL; | |
8c2a3ca2 JB |
5399 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
5400 | num_bytes); | |
6a63209f JB |
5401 | } |
5402 | ||
5403 | /* | |
8929ecfa YZ |
5404 | * helper to calculate size of global block reservation. |
5405 | * the desired value is sum of space used by extent tree, | |
5406 | * checksum tree and root tree | |
6a63209f | 5407 | */ |
8929ecfa | 5408 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 5409 | { |
8929ecfa YZ |
5410 | struct btrfs_space_info *sinfo; |
5411 | u64 num_bytes; | |
5412 | u64 meta_used; | |
5413 | u64 data_used; | |
6c41761f | 5414 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 5415 | |
8929ecfa YZ |
5416 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
5417 | spin_lock(&sinfo->lock); | |
5418 | data_used = sinfo->bytes_used; | |
5419 | spin_unlock(&sinfo->lock); | |
33b4d47f | 5420 | |
8929ecfa YZ |
5421 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
5422 | spin_lock(&sinfo->lock); | |
6d48755d JB |
5423 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
5424 | data_used = 0; | |
8929ecfa YZ |
5425 | meta_used = sinfo->bytes_used; |
5426 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 5427 | |
8929ecfa YZ |
5428 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
5429 | csum_size * 2; | |
f8c269d7 | 5430 | num_bytes += div_u64(data_used + meta_used, 50); |
4e06bdd6 | 5431 | |
8929ecfa | 5432 | if (num_bytes * 3 > meta_used) |
f8c269d7 | 5433 | num_bytes = div_u64(meta_used, 3); |
ab6e2410 | 5434 | |
707e8a07 | 5435 | return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10); |
8929ecfa | 5436 | } |
6a63209f | 5437 | |
8929ecfa YZ |
5438 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5439 | { | |
5440 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5441 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5442 | u64 num_bytes; | |
6a63209f | 5443 | |
8929ecfa | 5444 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 5445 | |
8929ecfa | 5446 | spin_lock(&sinfo->lock); |
1f699d38 | 5447 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5448 | |
ee22184b | 5449 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5450 | |
fb4b10e5 JB |
5451 | if (block_rsv->reserved < block_rsv->size) { |
5452 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + | |
5453 | sinfo->bytes_reserved + sinfo->bytes_readonly + | |
5454 | sinfo->bytes_may_use; | |
5455 | if (sinfo->total_bytes > num_bytes) { | |
5456 | num_bytes = sinfo->total_bytes - num_bytes; | |
5457 | num_bytes = min(num_bytes, | |
5458 | block_rsv->size - block_rsv->reserved); | |
5459 | block_rsv->reserved += num_bytes; | |
5460 | sinfo->bytes_may_use += num_bytes; | |
5461 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5462 | sinfo->flags, num_bytes, | |
5463 | 1); | |
5464 | } | |
5465 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5466 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5467 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5468 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5469 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5470 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5471 | } |
182608c8 | 5472 | |
fb4b10e5 JB |
5473 | if (block_rsv->reserved == block_rsv->size) |
5474 | block_rsv->full = 1; | |
5475 | else | |
5476 | block_rsv->full = 0; | |
5477 | ||
8929ecfa | 5478 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5479 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5480 | } |
5481 | ||
f0486c68 | 5482 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5483 | { |
f0486c68 | 5484 | struct btrfs_space_info *space_info; |
6a63209f | 5485 | |
f0486c68 YZ |
5486 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5487 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5488 | |
f0486c68 | 5489 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5490 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5491 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5492 | fs_info->trans_block_rsv.space_info = space_info; |
5493 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5494 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5495 | |
8929ecfa YZ |
5496 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5497 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5498 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5499 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5500 | if (fs_info->quota_root) |
5501 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5502 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5503 | |
8929ecfa | 5504 | update_global_block_rsv(fs_info); |
6a63209f JB |
5505 | } |
5506 | ||
8929ecfa | 5507 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5508 | { |
8c2a3ca2 JB |
5509 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5510 | (u64)-1); | |
8929ecfa YZ |
5511 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5512 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5513 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5514 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5515 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5516 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5517 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5518 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5519 | } |
5520 | ||
a22285a6 YZ |
5521 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
5522 | struct btrfs_root *root) | |
6a63209f | 5523 | { |
0e721106 JB |
5524 | if (!trans->block_rsv) |
5525 | return; | |
5526 | ||
a22285a6 YZ |
5527 | if (!trans->bytes_reserved) |
5528 | return; | |
6a63209f | 5529 | |
e77266e4 | 5530 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 5531 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 5532 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
5533 | trans->bytes_reserved = 0; |
5534 | } | |
6a63209f | 5535 | |
4fbcdf66 FM |
5536 | /* |
5537 | * To be called after all the new block groups attached to the transaction | |
5538 | * handle have been created (btrfs_create_pending_block_groups()). | |
5539 | */ | |
5540 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5541 | { | |
5542 | struct btrfs_fs_info *fs_info = trans->root->fs_info; | |
5543 | ||
5544 | if (!trans->chunk_bytes_reserved) | |
5545 | return; | |
5546 | ||
5547 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5548 | ||
5549 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5550 | trans->chunk_bytes_reserved); | |
5551 | trans->chunk_bytes_reserved = 0; | |
5552 | } | |
5553 | ||
79787eaa | 5554 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
5555 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
5556 | struct inode *inode) | |
5557 | { | |
5558 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5559 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
5560 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
5561 | ||
5562 | /* | |
fcb80c2a JB |
5563 | * We need to hold space in order to delete our orphan item once we've |
5564 | * added it, so this takes the reservation so we can release it later | |
5565 | * when we are truly done with the orphan item. | |
d68fc57b | 5566 | */ |
ff5714cc | 5567 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5568 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5569 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 5570 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
5571 | } |
5572 | ||
d68fc57b | 5573 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 5574 | { |
d68fc57b | 5575 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 5576 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5577 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5578 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
5579 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
5580 | } | |
97e728d4 | 5581 | |
d5c12070 MX |
5582 | /* |
5583 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5584 | * root: the root of the parent directory | |
5585 | * rsv: block reservation | |
5586 | * items: the number of items that we need do reservation | |
5587 | * qgroup_reserved: used to return the reserved size in qgroup | |
5588 | * | |
5589 | * This function is used to reserve the space for snapshot/subvolume | |
5590 | * creation and deletion. Those operations are different with the | |
5591 | * common file/directory operations, they change two fs/file trees | |
5592 | * and root tree, the number of items that the qgroup reserves is | |
5593 | * different with the free space reservation. So we can not use | |
01327610 | 5594 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5595 | */ |
5596 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5597 | struct btrfs_block_rsv *rsv, | |
5598 | int items, | |
ee3441b4 JM |
5599 | u64 *qgroup_reserved, |
5600 | bool use_global_rsv) | |
a22285a6 | 5601 | { |
d5c12070 MX |
5602 | u64 num_bytes; |
5603 | int ret; | |
ee3441b4 | 5604 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
5605 | |
5606 | if (root->fs_info->quota_enabled) { | |
5607 | /* One for parent inode, two for dir entries */ | |
707e8a07 | 5608 | num_bytes = 3 * root->nodesize; |
7174109c | 5609 | ret = btrfs_qgroup_reserve_meta(root, num_bytes); |
d5c12070 MX |
5610 | if (ret) |
5611 | return ret; | |
5612 | } else { | |
5613 | num_bytes = 0; | |
5614 | } | |
5615 | ||
5616 | *qgroup_reserved = num_bytes; | |
5617 | ||
5618 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5619 | rsv->space_info = __find_space_info(root->fs_info, | |
5620 | BTRFS_BLOCK_GROUP_METADATA); | |
5621 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5622 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5623 | |
5624 | if (ret == -ENOSPC && use_global_rsv) | |
5625 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
5626 | ||
7174109c QW |
5627 | if (ret && *qgroup_reserved) |
5628 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5629 | |
5630 | return ret; | |
5631 | } | |
5632 | ||
5633 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5634 | struct btrfs_block_rsv *rsv, | |
5635 | u64 qgroup_reserved) | |
5636 | { | |
5637 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
97e728d4 JB |
5638 | } |
5639 | ||
7709cde3 JB |
5640 | /** |
5641 | * drop_outstanding_extent - drop an outstanding extent | |
5642 | * @inode: the inode we're dropping the extent for | |
01327610 | 5643 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5644 | * |
5645 | * This is called when we are freeing up an outstanding extent, either called | |
5646 | * after an error or after an extent is written. This will return the number of | |
5647 | * reserved extents that need to be freed. This must be called with | |
5648 | * BTRFS_I(inode)->lock held. | |
5649 | */ | |
dcab6a3b | 5650 | static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) |
9e0baf60 | 5651 | { |
7fd2ae21 | 5652 | unsigned drop_inode_space = 0; |
9e0baf60 | 5653 | unsigned dropped_extents = 0; |
dcab6a3b | 5654 | unsigned num_extents = 0; |
9e0baf60 | 5655 | |
dcab6a3b JB |
5656 | num_extents = (unsigned)div64_u64(num_bytes + |
5657 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5658 | BTRFS_MAX_EXTENT_SIZE); | |
5659 | ASSERT(num_extents); | |
5660 | ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); | |
5661 | BTRFS_I(inode)->outstanding_extents -= num_extents; | |
9e0baf60 | 5662 | |
7fd2ae21 | 5663 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
5664 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5665 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 5666 | drop_inode_space = 1; |
7fd2ae21 | 5667 | |
9e0baf60 | 5668 | /* |
01327610 | 5669 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
5670 | * reserved then we need to leave the reserved extents count alone. |
5671 | */ | |
5672 | if (BTRFS_I(inode)->outstanding_extents >= | |
5673 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 5674 | return drop_inode_space; |
9e0baf60 JB |
5675 | |
5676 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5677 | BTRFS_I(inode)->outstanding_extents; | |
5678 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5679 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5680 | } |
5681 | ||
7709cde3 | 5682 | /** |
01327610 NS |
5683 | * calc_csum_metadata_size - return the amount of metadata space that must be |
5684 | * reserved/freed for the given bytes. | |
7709cde3 JB |
5685 | * @inode: the inode we're manipulating |
5686 | * @num_bytes: the number of bytes in question | |
5687 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5688 | * | |
5689 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5690 | * correct amount of metadata that must either be reserved or freed. We | |
5691 | * calculate how many checksums we can fit into one leaf and then divide the | |
5692 | * number of bytes that will need to be checksumed by this value to figure out | |
5693 | * how many checksums will be required. If we are adding bytes then the number | |
5694 | * may go up and we will return the number of additional bytes that must be | |
5695 | * reserved. If it is going down we will return the number of bytes that must | |
5696 | * be freed. | |
5697 | * | |
5698 | * This must be called with BTRFS_I(inode)->lock held. | |
5699 | */ | |
5700 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5701 | int reserve) | |
6324fbf3 | 5702 | { |
7709cde3 | 5703 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1262133b | 5704 | u64 old_csums, num_csums; |
7709cde3 JB |
5705 | |
5706 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5707 | BTRFS_I(inode)->csum_bytes == 0) | |
5708 | return 0; | |
5709 | ||
28f75a0e | 5710 | old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5711 | if (reserve) |
5712 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5713 | else | |
5714 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
28f75a0e | 5715 | num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5716 | |
5717 | /* No change, no need to reserve more */ | |
5718 | if (old_csums == num_csums) | |
5719 | return 0; | |
5720 | ||
5721 | if (reserve) | |
5722 | return btrfs_calc_trans_metadata_size(root, | |
5723 | num_csums - old_csums); | |
5724 | ||
5725 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 5726 | } |
c146afad | 5727 | |
0ca1f7ce YZ |
5728 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
5729 | { | |
5730 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5731 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 5732 | u64 to_reserve = 0; |
660d3f6c | 5733 | u64 csum_bytes; |
9e0baf60 | 5734 | unsigned nr_extents = 0; |
660d3f6c | 5735 | int extra_reserve = 0; |
08e007d2 | 5736 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5737 | int ret = 0; |
c64c2bd8 | 5738 | bool delalloc_lock = true; |
88e081bf WS |
5739 | u64 to_free = 0; |
5740 | unsigned dropped; | |
6324fbf3 | 5741 | |
c64c2bd8 JB |
5742 | /* If we are a free space inode we need to not flush since we will be in |
5743 | * the middle of a transaction commit. We also don't need the delalloc | |
5744 | * mutex since we won't race with anybody. We need this mostly to make | |
5745 | * lockdep shut its filthy mouth. | |
5746 | */ | |
5747 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5748 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
5749 | delalloc_lock = false; |
5750 | } | |
c09544e0 | 5751 | |
08e007d2 MX |
5752 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
5753 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 5754 | schedule_timeout(1); |
ec44a35c | 5755 | |
c64c2bd8 JB |
5756 | if (delalloc_lock) |
5757 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5758 | ||
0ca1f7ce | 5759 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 5760 | |
9e0baf60 | 5761 | spin_lock(&BTRFS_I(inode)->lock); |
6a41dd09 JB |
5762 | nr_extents = (unsigned)div64_u64(num_bytes + |
5763 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5764 | BTRFS_MAX_EXTENT_SIZE); | |
5765 | BTRFS_I(inode)->outstanding_extents += nr_extents; | |
5766 | nr_extents = 0; | |
9e0baf60 JB |
5767 | |
5768 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 5769 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
5770 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
5771 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 5772 | |
7fd2ae21 JB |
5773 | /* |
5774 | * Add an item to reserve for updating the inode when we complete the | |
5775 | * delalloc io. | |
5776 | */ | |
72ac3c0d JB |
5777 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5778 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 5779 | nr_extents++; |
660d3f6c | 5780 | extra_reserve = 1; |
593060d7 | 5781 | } |
7fd2ae21 JB |
5782 | |
5783 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 5784 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 5785 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 5786 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 5787 | |
88e081bf | 5788 | if (root->fs_info->quota_enabled) { |
7174109c QW |
5789 | ret = btrfs_qgroup_reserve_meta(root, |
5790 | nr_extents * root->nodesize); | |
88e081bf WS |
5791 | if (ret) |
5792 | goto out_fail; | |
5793 | } | |
c5567237 | 5794 | |
88e081bf WS |
5795 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
5796 | if (unlikely(ret)) { | |
7174109c | 5797 | btrfs_qgroup_free_meta(root, nr_extents * root->nodesize); |
88e081bf | 5798 | goto out_fail; |
9e0baf60 | 5799 | } |
25179201 | 5800 | |
660d3f6c JB |
5801 | spin_lock(&BTRFS_I(inode)->lock); |
5802 | if (extra_reserve) { | |
72ac3c0d JB |
5803 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5804 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
5805 | nr_extents--; |
5806 | } | |
5807 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5808 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
5809 | |
5810 | if (delalloc_lock) | |
5811 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 5812 | |
8c2a3ca2 | 5813 | if (to_reserve) |
67871254 | 5814 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
8c2a3ca2 | 5815 | btrfs_ino(inode), to_reserve, 1); |
0ca1f7ce YZ |
5816 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
5817 | ||
0ca1f7ce | 5818 | return 0; |
88e081bf WS |
5819 | |
5820 | out_fail: | |
5821 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 5822 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
5823 | /* |
5824 | * If the inodes csum_bytes is the same as the original | |
5825 | * csum_bytes then we know we haven't raced with any free()ers | |
5826 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 5827 | */ |
f4881bc7 | 5828 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 5829 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
5830 | } else { |
5831 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5832 | u64 bytes; | |
5833 | ||
5834 | /* | |
5835 | * This is tricky, but first we need to figure out how much we | |
01327610 | 5836 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
5837 | * reservation, so we reset ->csum_bytes to the csum_bytes |
5838 | * before we dropped our lock, and then call the free for the | |
5839 | * number of bytes that were freed while we were trying our | |
5840 | * reservation. | |
5841 | */ | |
5842 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5843 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5844 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5845 | ||
5846 | ||
5847 | /* | |
5848 | * Now we need to see how much we would have freed had we not | |
5849 | * been making this reservation and our ->csum_bytes were not | |
5850 | * artificially inflated. | |
5851 | */ | |
5852 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5853 | bytes = csum_bytes - orig_csum_bytes; | |
5854 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5855 | ||
5856 | /* | |
5857 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 5858 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
5859 | * not had an artificially high ->csum_bytes, so we need to free |
5860 | * the remainder. If bytes is the same or less then we don't | |
5861 | * need to do anything, the other free-ers did the correct | |
5862 | * thing. | |
5863 | */ | |
5864 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5865 | if (bytes > to_free) | |
5866 | to_free = bytes - to_free; | |
5867 | else | |
5868 | to_free = 0; | |
5869 | } | |
88e081bf | 5870 | spin_unlock(&BTRFS_I(inode)->lock); |
e2d1f923 | 5871 | if (dropped) |
88e081bf WS |
5872 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
5873 | ||
5874 | if (to_free) { | |
5875 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5876 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5877 | btrfs_ino(inode), to_free, 0); | |
5878 | } | |
5879 | if (delalloc_lock) | |
5880 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5881 | return ret; | |
0ca1f7ce YZ |
5882 | } |
5883 | ||
7709cde3 JB |
5884 | /** |
5885 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5886 | * @inode: the inode to release the reservation for | |
5887 | * @num_bytes: the number of bytes we're releasing | |
5888 | * | |
5889 | * This will release the metadata reservation for an inode. This can be called | |
5890 | * once we complete IO for a given set of bytes to release their metadata | |
5891 | * reservations. | |
5892 | */ | |
0ca1f7ce YZ |
5893 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
5894 | { | |
5895 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
5896 | u64 to_free = 0; |
5897 | unsigned dropped; | |
0ca1f7ce YZ |
5898 | |
5899 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 5900 | spin_lock(&BTRFS_I(inode)->lock); |
dcab6a3b | 5901 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 5902 | |
0934856d MX |
5903 | if (num_bytes) |
5904 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 5905 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
5906 | if (dropped > 0) |
5907 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 5908 | |
6a3891c5 JB |
5909 | if (btrfs_test_is_dummy_root(root)) |
5910 | return; | |
5911 | ||
8c2a3ca2 JB |
5912 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5913 | btrfs_ino(inode), to_free, 0); | |
c5567237 | 5914 | |
0ca1f7ce YZ |
5915 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5916 | to_free); | |
5917 | } | |
5918 | ||
1ada3a62 | 5919 | /** |
7cf5b976 | 5920 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
5921 | * delalloc |
5922 | * @inode: inode we're writing to | |
5923 | * @start: start range we are writing to | |
5924 | * @len: how long the range we are writing to | |
5925 | * | |
5926 | * TODO: This function will finally replace old btrfs_delalloc_reserve_space() | |
5927 | * | |
5928 | * This will do the following things | |
5929 | * | |
5930 | * o reserve space in data space info for num bytes | |
5931 | * and reserve precious corresponding qgroup space | |
5932 | * (Done in check_data_free_space) | |
5933 | * | |
5934 | * o reserve space for metadata space, based on the number of outstanding | |
5935 | * extents and how much csums will be needed | |
5936 | * also reserve metadata space in a per root over-reserve method. | |
5937 | * o add to the inodes->delalloc_bytes | |
5938 | * o add it to the fs_info's delalloc inodes list. | |
5939 | * (Above 3 all done in delalloc_reserve_metadata) | |
5940 | * | |
5941 | * Return 0 for success | |
5942 | * Return <0 for error(-ENOSPC or -EQUOT) | |
5943 | */ | |
7cf5b976 | 5944 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
5945 | { |
5946 | int ret; | |
5947 | ||
7cf5b976 | 5948 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
5949 | if (ret < 0) |
5950 | return ret; | |
5951 | ret = btrfs_delalloc_reserve_metadata(inode, len); | |
5952 | if (ret < 0) | |
7cf5b976 | 5953 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
5954 | return ret; |
5955 | } | |
5956 | ||
7709cde3 | 5957 | /** |
7cf5b976 | 5958 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
5959 | * @inode: inode we're releasing space for |
5960 | * @start: start position of the space already reserved | |
5961 | * @len: the len of the space already reserved | |
5962 | * | |
5963 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5964 | * called in the case that we don't need the metadata AND data reservations | |
5965 | * anymore. So if there is an error or we insert an inline extent. | |
5966 | * | |
5967 | * This function will release the metadata space that was not used and will | |
5968 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5969 | * list if there are no delalloc bytes left. | |
5970 | * Also it will handle the qgroup reserved space. | |
5971 | */ | |
7cf5b976 | 5972 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
5973 | { |
5974 | btrfs_delalloc_release_metadata(inode, len); | |
7cf5b976 | 5975 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
5976 | } |
5977 | ||
ce93ec54 JB |
5978 | static int update_block_group(struct btrfs_trans_handle *trans, |
5979 | struct btrfs_root *root, u64 bytenr, | |
5980 | u64 num_bytes, int alloc) | |
9078a3e1 | 5981 | { |
0af3d00b | 5982 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5983 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5984 | u64 total = num_bytes; |
9078a3e1 | 5985 | u64 old_val; |
db94535d | 5986 | u64 byte_in_group; |
0af3d00b | 5987 | int factor; |
3e1ad54f | 5988 | |
5d4f98a2 | 5989 | /* block accounting for super block */ |
eb73c1b7 | 5990 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5991 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5992 | if (alloc) |
5993 | old_val += num_bytes; | |
5994 | else | |
5995 | old_val -= num_bytes; | |
6c41761f | 5996 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5997 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5998 | |
d397712b | 5999 | while (total) { |
db94535d | 6000 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6001 | if (!cache) |
79787eaa | 6002 | return -ENOENT; |
b742bb82 YZ |
6003 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6004 | BTRFS_BLOCK_GROUP_RAID1 | | |
6005 | BTRFS_BLOCK_GROUP_RAID10)) | |
6006 | factor = 2; | |
6007 | else | |
6008 | factor = 1; | |
9d66e233 JB |
6009 | /* |
6010 | * If this block group has free space cache written out, we | |
6011 | * need to make sure to load it if we are removing space. This | |
6012 | * is because we need the unpinning stage to actually add the | |
6013 | * space back to the block group, otherwise we will leak space. | |
6014 | */ | |
6015 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6016 | cache_block_group(cache, 1); |
0af3d00b | 6017 | |
db94535d CM |
6018 | byte_in_group = bytenr - cache->key.objectid; |
6019 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6020 | |
25179201 | 6021 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6022 | spin_lock(&cache->lock); |
0af3d00b | 6023 | |
73bc1876 | 6024 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
6025 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6026 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6027 | ||
9078a3e1 | 6028 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6029 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6030 | if (alloc) { |
db94535d | 6031 | old_val += num_bytes; |
11833d66 YZ |
6032 | btrfs_set_block_group_used(&cache->item, old_val); |
6033 | cache->reserved -= num_bytes; | |
11833d66 | 6034 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6035 | cache->space_info->bytes_used += num_bytes; |
6036 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6037 | spin_unlock(&cache->lock); |
25179201 | 6038 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6039 | } else { |
db94535d | 6040 | old_val -= num_bytes; |
ae0ab003 FM |
6041 | btrfs_set_block_group_used(&cache->item, old_val); |
6042 | cache->pinned += num_bytes; | |
6043 | cache->space_info->bytes_pinned += num_bytes; | |
6044 | cache->space_info->bytes_used -= num_bytes; | |
6045 | cache->space_info->disk_used -= num_bytes * factor; | |
6046 | spin_unlock(&cache->lock); | |
6047 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6048 | |
ae0ab003 FM |
6049 | set_extent_dirty(info->pinned_extents, |
6050 | bytenr, bytenr + num_bytes - 1, | |
6051 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6052 | } |
1bbc621e CM |
6053 | |
6054 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6055 | if (list_empty(&cache->dirty_list)) { | |
6056 | list_add_tail(&cache->dirty_list, | |
6057 | &trans->transaction->dirty_bgs); | |
6058 | trans->transaction->num_dirty_bgs++; | |
6059 | btrfs_get_block_group(cache); | |
6060 | } | |
6061 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6062 | ||
036a9348 FM |
6063 | /* |
6064 | * No longer have used bytes in this block group, queue it for | |
6065 | * deletion. We do this after adding the block group to the | |
6066 | * dirty list to avoid races between cleaner kthread and space | |
6067 | * cache writeout. | |
6068 | */ | |
6069 | if (!alloc && old_val == 0) { | |
6070 | spin_lock(&info->unused_bgs_lock); | |
6071 | if (list_empty(&cache->bg_list)) { | |
6072 | btrfs_get_block_group(cache); | |
6073 | list_add_tail(&cache->bg_list, | |
6074 | &info->unused_bgs); | |
6075 | } | |
6076 | spin_unlock(&info->unused_bgs_lock); | |
6077 | } | |
6078 | ||
fa9c0d79 | 6079 | btrfs_put_block_group(cache); |
db94535d CM |
6080 | total -= num_bytes; |
6081 | bytenr += num_bytes; | |
9078a3e1 CM |
6082 | } |
6083 | return 0; | |
6084 | } | |
6324fbf3 | 6085 | |
a061fc8d CM |
6086 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
6087 | { | |
0f9dd46c | 6088 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6089 | u64 bytenr; |
0f9dd46c | 6090 | |
a1897fdd LB |
6091 | spin_lock(&root->fs_info->block_group_cache_lock); |
6092 | bytenr = root->fs_info->first_logical_byte; | |
6093 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
6094 | ||
6095 | if (bytenr < (u64)-1) | |
6096 | return bytenr; | |
6097 | ||
0f9dd46c JB |
6098 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
6099 | if (!cache) | |
a061fc8d | 6100 | return 0; |
0f9dd46c | 6101 | |
d2fb3437 | 6102 | bytenr = cache->key.objectid; |
fa9c0d79 | 6103 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6104 | |
6105 | return bytenr; | |
a061fc8d CM |
6106 | } |
6107 | ||
f0486c68 YZ |
6108 | static int pin_down_extent(struct btrfs_root *root, |
6109 | struct btrfs_block_group_cache *cache, | |
6110 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6111 | { |
11833d66 YZ |
6112 | spin_lock(&cache->space_info->lock); |
6113 | spin_lock(&cache->lock); | |
6114 | cache->pinned += num_bytes; | |
6115 | cache->space_info->bytes_pinned += num_bytes; | |
6116 | if (reserved) { | |
6117 | cache->reserved -= num_bytes; | |
6118 | cache->space_info->bytes_reserved -= num_bytes; | |
6119 | } | |
6120 | spin_unlock(&cache->lock); | |
6121 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6122 | |
f0486c68 YZ |
6123 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
6124 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
e2d1f923 | 6125 | if (reserved) |
0be5dc67 | 6126 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); |
f0486c68 YZ |
6127 | return 0; |
6128 | } | |
68b38550 | 6129 | |
f0486c68 YZ |
6130 | /* |
6131 | * this function must be called within transaction | |
6132 | */ | |
6133 | int btrfs_pin_extent(struct btrfs_root *root, | |
6134 | u64 bytenr, u64 num_bytes, int reserved) | |
6135 | { | |
6136 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6137 | |
f0486c68 | 6138 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 6139 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
6140 | |
6141 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
6142 | ||
6143 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6144 | return 0; |
6145 | } | |
6146 | ||
f0486c68 | 6147 | /* |
e688b725 CM |
6148 | * this function must be called within transaction |
6149 | */ | |
dcfac415 | 6150 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
6151 | u64 bytenr, u64 num_bytes) |
6152 | { | |
6153 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6154 | int ret; |
e688b725 CM |
6155 | |
6156 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
6157 | if (!cache) |
6158 | return -EINVAL; | |
e688b725 CM |
6159 | |
6160 | /* | |
6161 | * pull in the free space cache (if any) so that our pin | |
6162 | * removes the free space from the cache. We have load_only set | |
6163 | * to one because the slow code to read in the free extents does check | |
6164 | * the pinned extents. | |
6165 | */ | |
f6373bf3 | 6166 | cache_block_group(cache, 1); |
e688b725 CM |
6167 | |
6168 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
6169 | ||
6170 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6171 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6172 | btrfs_put_block_group(cache); |
b50c6e25 | 6173 | return ret; |
e688b725 CM |
6174 | } |
6175 | ||
8c2a1a30 JB |
6176 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
6177 | { | |
6178 | int ret; | |
6179 | struct btrfs_block_group_cache *block_group; | |
6180 | struct btrfs_caching_control *caching_ctl; | |
6181 | ||
6182 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
6183 | if (!block_group) | |
6184 | return -EINVAL; | |
6185 | ||
6186 | cache_block_group(block_group, 0); | |
6187 | caching_ctl = get_caching_control(block_group); | |
6188 | ||
6189 | if (!caching_ctl) { | |
6190 | /* Logic error */ | |
6191 | BUG_ON(!block_group_cache_done(block_group)); | |
6192 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6193 | } else { | |
6194 | mutex_lock(&caching_ctl->mutex); | |
6195 | ||
6196 | if (start >= caching_ctl->progress) { | |
6197 | ret = add_excluded_extent(root, start, num_bytes); | |
6198 | } else if (start + num_bytes <= caching_ctl->progress) { | |
6199 | ret = btrfs_remove_free_space(block_group, | |
6200 | start, num_bytes); | |
6201 | } else { | |
6202 | num_bytes = caching_ctl->progress - start; | |
6203 | ret = btrfs_remove_free_space(block_group, | |
6204 | start, num_bytes); | |
6205 | if (ret) | |
6206 | goto out_lock; | |
6207 | ||
6208 | num_bytes = (start + num_bytes) - | |
6209 | caching_ctl->progress; | |
6210 | start = caching_ctl->progress; | |
6211 | ret = add_excluded_extent(root, start, num_bytes); | |
6212 | } | |
6213 | out_lock: | |
6214 | mutex_unlock(&caching_ctl->mutex); | |
6215 | put_caching_control(caching_ctl); | |
6216 | } | |
6217 | btrfs_put_block_group(block_group); | |
6218 | return ret; | |
6219 | } | |
6220 | ||
6221 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
6222 | struct extent_buffer *eb) | |
6223 | { | |
6224 | struct btrfs_file_extent_item *item; | |
6225 | struct btrfs_key key; | |
6226 | int found_type; | |
6227 | int i; | |
6228 | ||
6229 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
6230 | return 0; | |
6231 | ||
6232 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6233 | btrfs_item_key_to_cpu(eb, &key, i); | |
6234 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6235 | continue; | |
6236 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6237 | found_type = btrfs_file_extent_type(eb, item); | |
6238 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6239 | continue; | |
6240 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6241 | continue; | |
6242 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6243 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
6244 | __exclude_logged_extent(log, key.objectid, key.offset); | |
6245 | } | |
6246 | ||
6247 | return 0; | |
6248 | } | |
6249 | ||
9cfa3e34 FM |
6250 | static void |
6251 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6252 | { | |
6253 | atomic_inc(&bg->reservations); | |
6254 | } | |
6255 | ||
6256 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6257 | const u64 start) | |
6258 | { | |
6259 | struct btrfs_block_group_cache *bg; | |
6260 | ||
6261 | bg = btrfs_lookup_block_group(fs_info, start); | |
6262 | ASSERT(bg); | |
6263 | if (atomic_dec_and_test(&bg->reservations)) | |
6264 | wake_up_atomic_t(&bg->reservations); | |
6265 | btrfs_put_block_group(bg); | |
6266 | } | |
6267 | ||
6268 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6269 | { | |
6270 | schedule(); | |
6271 | return 0; | |
6272 | } | |
6273 | ||
6274 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6275 | { | |
6276 | struct btrfs_space_info *space_info = bg->space_info; | |
6277 | ||
6278 | ASSERT(bg->ro); | |
6279 | ||
6280 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6281 | return; | |
6282 | ||
6283 | /* | |
6284 | * Our block group is read only but before we set it to read only, | |
6285 | * some task might have had allocated an extent from it already, but it | |
6286 | * has not yet created a respective ordered extent (and added it to a | |
6287 | * root's list of ordered extents). | |
6288 | * Therefore wait for any task currently allocating extents, since the | |
6289 | * block group's reservations counter is incremented while a read lock | |
6290 | * on the groups' semaphore is held and decremented after releasing | |
6291 | * the read access on that semaphore and creating the ordered extent. | |
6292 | */ | |
6293 | down_write(&space_info->groups_sem); | |
6294 | up_write(&space_info->groups_sem); | |
6295 | ||
6296 | wait_on_atomic_t(&bg->reservations, | |
6297 | btrfs_wait_bg_reservations_atomic_t, | |
6298 | TASK_UNINTERRUPTIBLE); | |
6299 | } | |
6300 | ||
fb25e914 JB |
6301 | /** |
6302 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
6303 | * @cache: The cache we are manipulating | |
6304 | * @num_bytes: The number of bytes in question | |
6305 | * @reserve: One of the reservation enums | |
e570fd27 | 6306 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 JB |
6307 | * |
6308 | * This is called by the allocator when it reserves space, or by somebody who is | |
6309 | * freeing space that was never actually used on disk. For example if you | |
6310 | * reserve some space for a new leaf in transaction A and before transaction A | |
6311 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
6312 | * clear the reservation. | |
6313 | * | |
6314 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
6315 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
6316 | * delalloc bits in the io_tree. We have to do this since we could end up | |
6317 | * allocating less disk space for the amount of data we have reserved in the | |
6318 | * case of compression. | |
6319 | * | |
6320 | * If this is a reservation and the block group has become read only we cannot | |
6321 | * make the reservation and return -EAGAIN, otherwise this function always | |
6322 | * succeeds. | |
f0486c68 | 6323 | */ |
fb25e914 | 6324 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 | 6325 | u64 num_bytes, int reserve, int delalloc) |
11833d66 | 6326 | { |
fb25e914 | 6327 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6328 | int ret = 0; |
79787eaa | 6329 | |
fb25e914 JB |
6330 | spin_lock(&space_info->lock); |
6331 | spin_lock(&cache->lock); | |
6332 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
6333 | if (cache->ro) { |
6334 | ret = -EAGAIN; | |
6335 | } else { | |
fb25e914 JB |
6336 | cache->reserved += num_bytes; |
6337 | space_info->bytes_reserved += num_bytes; | |
6338 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 6339 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
6340 | "space_info", space_info->flags, |
6341 | num_bytes, 0); | |
fb25e914 JB |
6342 | space_info->bytes_may_use -= num_bytes; |
6343 | } | |
e570fd27 MX |
6344 | |
6345 | if (delalloc) | |
6346 | cache->delalloc_bytes += num_bytes; | |
f0486c68 | 6347 | } |
fb25e914 JB |
6348 | } else { |
6349 | if (cache->ro) | |
6350 | space_info->bytes_readonly += num_bytes; | |
6351 | cache->reserved -= num_bytes; | |
6352 | space_info->bytes_reserved -= num_bytes; | |
e570fd27 MX |
6353 | |
6354 | if (delalloc) | |
6355 | cache->delalloc_bytes -= num_bytes; | |
324ae4df | 6356 | } |
fb25e914 JB |
6357 | spin_unlock(&cache->lock); |
6358 | spin_unlock(&space_info->lock); | |
f0486c68 | 6359 | return ret; |
324ae4df | 6360 | } |
9078a3e1 | 6361 | |
143bede5 | 6362 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 6363 | struct btrfs_root *root) |
e8569813 | 6364 | { |
e8569813 | 6365 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
6366 | struct btrfs_caching_control *next; |
6367 | struct btrfs_caching_control *caching_ctl; | |
6368 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6369 | |
9e351cc8 | 6370 | down_write(&fs_info->commit_root_sem); |
25179201 | 6371 | |
11833d66 YZ |
6372 | list_for_each_entry_safe(caching_ctl, next, |
6373 | &fs_info->caching_block_groups, list) { | |
6374 | cache = caching_ctl->block_group; | |
6375 | if (block_group_cache_done(cache)) { | |
6376 | cache->last_byte_to_unpin = (u64)-1; | |
6377 | list_del_init(&caching_ctl->list); | |
6378 | put_caching_control(caching_ctl); | |
e8569813 | 6379 | } else { |
11833d66 | 6380 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6381 | } |
e8569813 | 6382 | } |
11833d66 YZ |
6383 | |
6384 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6385 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6386 | else | |
6387 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6388 | ||
9e351cc8 | 6389 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6390 | |
6391 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6392 | } |
6393 | ||
c759c4e1 JB |
6394 | /* |
6395 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6396 | * what it should be based on the mount options. | |
6397 | */ | |
6398 | static struct btrfs_free_cluster * | |
6399 | fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info, | |
6400 | u64 *empty_cluster) | |
6401 | { | |
6402 | struct btrfs_free_cluster *ret = NULL; | |
6403 | bool ssd = btrfs_test_opt(root, SSD); | |
6404 | ||
6405 | *empty_cluster = 0; | |
6406 | if (btrfs_mixed_space_info(space_info)) | |
6407 | return ret; | |
6408 | ||
6409 | if (ssd) | |
ee22184b | 6410 | *empty_cluster = SZ_2M; |
c759c4e1 JB |
6411 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
6412 | ret = &root->fs_info->meta_alloc_cluster; | |
6413 | if (!ssd) | |
ee22184b | 6414 | *empty_cluster = SZ_64K; |
c759c4e1 JB |
6415 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { |
6416 | ret = &root->fs_info->data_alloc_cluster; | |
6417 | } | |
6418 | ||
6419 | return ret; | |
6420 | } | |
6421 | ||
678886bd FM |
6422 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end, |
6423 | const bool return_free_space) | |
ccd467d6 | 6424 | { |
11833d66 YZ |
6425 | struct btrfs_fs_info *fs_info = root->fs_info; |
6426 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
6427 | struct btrfs_space_info *space_info; |
6428 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6429 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6430 | u64 len; |
c759c4e1 JB |
6431 | u64 total_unpinned = 0; |
6432 | u64 empty_cluster = 0; | |
7b398f8e | 6433 | bool readonly; |
ccd467d6 | 6434 | |
11833d66 | 6435 | while (start <= end) { |
7b398f8e | 6436 | readonly = false; |
11833d66 YZ |
6437 | if (!cache || |
6438 | start >= cache->key.objectid + cache->key.offset) { | |
6439 | if (cache) | |
6440 | btrfs_put_block_group(cache); | |
c759c4e1 | 6441 | total_unpinned = 0; |
11833d66 | 6442 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6443 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 JB |
6444 | |
6445 | cluster = fetch_cluster_info(root, | |
6446 | cache->space_info, | |
6447 | &empty_cluster); | |
6448 | empty_cluster <<= 1; | |
11833d66 YZ |
6449 | } |
6450 | ||
6451 | len = cache->key.objectid + cache->key.offset - start; | |
6452 | len = min(len, end + 1 - start); | |
6453 | ||
6454 | if (start < cache->last_byte_to_unpin) { | |
6455 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6456 | if (return_free_space) |
6457 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6458 | } |
6459 | ||
f0486c68 | 6460 | start += len; |
c759c4e1 | 6461 | total_unpinned += len; |
7b398f8e | 6462 | space_info = cache->space_info; |
f0486c68 | 6463 | |
c759c4e1 JB |
6464 | /* |
6465 | * If this space cluster has been marked as fragmented and we've | |
6466 | * unpinned enough in this block group to potentially allow a | |
6467 | * cluster to be created inside of it go ahead and clear the | |
6468 | * fragmented check. | |
6469 | */ | |
6470 | if (cluster && cluster->fragmented && | |
6471 | total_unpinned > empty_cluster) { | |
6472 | spin_lock(&cluster->lock); | |
6473 | cluster->fragmented = 0; | |
6474 | spin_unlock(&cluster->lock); | |
6475 | } | |
6476 | ||
7b398f8e | 6477 | spin_lock(&space_info->lock); |
11833d66 YZ |
6478 | spin_lock(&cache->lock); |
6479 | cache->pinned -= len; | |
7b398f8e | 6480 | space_info->bytes_pinned -= len; |
4f4db217 | 6481 | space_info->max_extent_size = 0; |
d288db5d | 6482 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6483 | if (cache->ro) { |
6484 | space_info->bytes_readonly += len; | |
6485 | readonly = true; | |
6486 | } | |
11833d66 | 6487 | spin_unlock(&cache->lock); |
7b398f8e JB |
6488 | if (!readonly && global_rsv->space_info == space_info) { |
6489 | spin_lock(&global_rsv->lock); | |
6490 | if (!global_rsv->full) { | |
6491 | len = min(len, global_rsv->size - | |
6492 | global_rsv->reserved); | |
6493 | global_rsv->reserved += len; | |
6494 | space_info->bytes_may_use += len; | |
6495 | if (global_rsv->reserved >= global_rsv->size) | |
6496 | global_rsv->full = 1; | |
6497 | } | |
6498 | spin_unlock(&global_rsv->lock); | |
6499 | } | |
6500 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6501 | } |
11833d66 YZ |
6502 | |
6503 | if (cache) | |
6504 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6505 | return 0; |
6506 | } | |
6507 | ||
6508 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 6509 | struct btrfs_root *root) |
a28ec197 | 6510 | { |
11833d66 | 6511 | struct btrfs_fs_info *fs_info = root->fs_info; |
e33e17ee JM |
6512 | struct btrfs_block_group_cache *block_group, *tmp; |
6513 | struct list_head *deleted_bgs; | |
11833d66 | 6514 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6515 | u64 start; |
6516 | u64 end; | |
a28ec197 | 6517 | int ret; |
a28ec197 | 6518 | |
11833d66 YZ |
6519 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6520 | unpin = &fs_info->freed_extents[1]; | |
6521 | else | |
6522 | unpin = &fs_info->freed_extents[0]; | |
6523 | ||
e33e17ee | 6524 | while (!trans->aborted) { |
d4b450cd | 6525 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6526 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6527 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6528 | if (ret) { |
6529 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6530 | break; |
d4b450cd | 6531 | } |
1f3c79a2 | 6532 | |
5378e607 LD |
6533 | if (btrfs_test_opt(root, DISCARD)) |
6534 | ret = btrfs_discard_extent(root, start, | |
6535 | end + 1 - start, NULL); | |
1f3c79a2 | 6536 | |
af6f8f60 | 6537 | clear_extent_dirty(unpin, start, end); |
678886bd | 6538 | unpin_extent_range(root, start, end, true); |
d4b450cd | 6539 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6540 | cond_resched(); |
a28ec197 | 6541 | } |
817d52f8 | 6542 | |
e33e17ee JM |
6543 | /* |
6544 | * Transaction is finished. We don't need the lock anymore. We | |
6545 | * do need to clean up the block groups in case of a transaction | |
6546 | * abort. | |
6547 | */ | |
6548 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6549 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6550 | u64 trimmed = 0; | |
6551 | ||
6552 | ret = -EROFS; | |
6553 | if (!trans->aborted) | |
6554 | ret = btrfs_discard_extent(root, | |
6555 | block_group->key.objectid, | |
6556 | block_group->key.offset, | |
6557 | &trimmed); | |
6558 | ||
6559 | list_del_init(&block_group->bg_list); | |
6560 | btrfs_put_block_group_trimming(block_group); | |
6561 | btrfs_put_block_group(block_group); | |
6562 | ||
6563 | if (ret) { | |
6564 | const char *errstr = btrfs_decode_error(ret); | |
6565 | btrfs_warn(fs_info, | |
6566 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6567 | ret, errstr); | |
6568 | } | |
6569 | } | |
6570 | ||
e20d96d6 CM |
6571 | return 0; |
6572 | } | |
6573 | ||
b150a4f1 JB |
6574 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6575 | u64 owner, u64 root_objectid) | |
6576 | { | |
6577 | struct btrfs_space_info *space_info; | |
6578 | u64 flags; | |
6579 | ||
6580 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6581 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6582 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6583 | else | |
6584 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6585 | } else { | |
6586 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6587 | } | |
6588 | ||
6589 | space_info = __find_space_info(fs_info, flags); | |
6590 | BUG_ON(!space_info); /* Logic bug */ | |
6591 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6592 | } | |
6593 | ||
6594 | ||
5d4f98a2 YZ |
6595 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
6596 | struct btrfs_root *root, | |
c682f9b3 | 6597 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6598 | u64 root_objectid, u64 owner_objectid, |
6599 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6600 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6601 | { |
e2fa7227 | 6602 | struct btrfs_key key; |
5d4f98a2 | 6603 | struct btrfs_path *path; |
1261ec42 CM |
6604 | struct btrfs_fs_info *info = root->fs_info; |
6605 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 6606 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6607 | struct btrfs_extent_item *ei; |
6608 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6609 | int ret; |
5d4f98a2 | 6610 | int is_data; |
952fccac CM |
6611 | int extent_slot = 0; |
6612 | int found_extent = 0; | |
6613 | int num_to_del = 1; | |
5d4f98a2 YZ |
6614 | u32 item_size; |
6615 | u64 refs; | |
c682f9b3 QW |
6616 | u64 bytenr = node->bytenr; |
6617 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6618 | int last_ref = 0; |
3173a18f JB |
6619 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6620 | SKINNY_METADATA); | |
037e6390 | 6621 | |
5caf2a00 | 6622 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6623 | if (!path) |
6624 | return -ENOMEM; | |
5f26f772 | 6625 | |
e4058b54 | 6626 | path->reada = READA_FORWARD; |
b9473439 | 6627 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6628 | |
6629 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6630 | BUG_ON(!is_data && refs_to_drop != 1); | |
6631 | ||
3173a18f JB |
6632 | if (is_data) |
6633 | skinny_metadata = 0; | |
6634 | ||
5d4f98a2 YZ |
6635 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
6636 | bytenr, num_bytes, parent, | |
6637 | root_objectid, owner_objectid, | |
6638 | owner_offset); | |
7bb86316 | 6639 | if (ret == 0) { |
952fccac | 6640 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6641 | while (extent_slot >= 0) { |
6642 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6643 | extent_slot); |
5d4f98a2 | 6644 | if (key.objectid != bytenr) |
952fccac | 6645 | break; |
5d4f98a2 YZ |
6646 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6647 | key.offset == num_bytes) { | |
952fccac CM |
6648 | found_extent = 1; |
6649 | break; | |
6650 | } | |
3173a18f JB |
6651 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6652 | key.offset == owner_objectid) { | |
6653 | found_extent = 1; | |
6654 | break; | |
6655 | } | |
952fccac CM |
6656 | if (path->slots[0] - extent_slot > 5) |
6657 | break; | |
5d4f98a2 | 6658 | extent_slot--; |
952fccac | 6659 | } |
5d4f98a2 YZ |
6660 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6661 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6662 | if (found_extent && item_size < sizeof(*ei)) | |
6663 | found_extent = 0; | |
6664 | #endif | |
31840ae1 | 6665 | if (!found_extent) { |
5d4f98a2 | 6666 | BUG_ON(iref); |
56bec294 | 6667 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 | 6668 | NULL, refs_to_drop, |
fcebe456 | 6669 | is_data, &last_ref); |
005d6427 DS |
6670 | if (ret) { |
6671 | btrfs_abort_transaction(trans, extent_root, ret); | |
6672 | goto out; | |
6673 | } | |
b3b4aa74 | 6674 | btrfs_release_path(path); |
b9473439 | 6675 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6676 | |
6677 | key.objectid = bytenr; | |
6678 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6679 | key.offset = num_bytes; | |
6680 | ||
3173a18f JB |
6681 | if (!is_data && skinny_metadata) { |
6682 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6683 | key.offset = owner_objectid; | |
6684 | } | |
6685 | ||
31840ae1 ZY |
6686 | ret = btrfs_search_slot(trans, extent_root, |
6687 | &key, path, -1, 1); | |
3173a18f JB |
6688 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6689 | /* | |
6690 | * Couldn't find our skinny metadata item, | |
6691 | * see if we have ye olde extent item. | |
6692 | */ | |
6693 | path->slots[0]--; | |
6694 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6695 | path->slots[0]); | |
6696 | if (key.objectid == bytenr && | |
6697 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6698 | key.offset == num_bytes) | |
6699 | ret = 0; | |
6700 | } | |
6701 | ||
6702 | if (ret > 0 && skinny_metadata) { | |
6703 | skinny_metadata = false; | |
9ce49a0b | 6704 | key.objectid = bytenr; |
3173a18f JB |
6705 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6706 | key.offset = num_bytes; | |
6707 | btrfs_release_path(path); | |
6708 | ret = btrfs_search_slot(trans, extent_root, | |
6709 | &key, path, -1, 1); | |
6710 | } | |
6711 | ||
f3465ca4 | 6712 | if (ret) { |
c2cf52eb | 6713 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6714 | ret, bytenr); |
b783e62d JB |
6715 | if (ret > 0) |
6716 | btrfs_print_leaf(extent_root, | |
6717 | path->nodes[0]); | |
f3465ca4 | 6718 | } |
005d6427 DS |
6719 | if (ret < 0) { |
6720 | btrfs_abort_transaction(trans, extent_root, ret); | |
6721 | goto out; | |
6722 | } | |
31840ae1 ZY |
6723 | extent_slot = path->slots[0]; |
6724 | } | |
fae7f21c | 6725 | } else if (WARN_ON(ret == -ENOENT)) { |
7bb86316 | 6726 | btrfs_print_leaf(extent_root, path->nodes[0]); |
c2cf52eb SK |
6727 | btrfs_err(info, |
6728 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6729 | bytenr, parent, root_objectid, owner_objectid, |
6730 | owner_offset); | |
c4a050bb JB |
6731 | btrfs_abort_transaction(trans, extent_root, ret); |
6732 | goto out; | |
79787eaa | 6733 | } else { |
005d6427 DS |
6734 | btrfs_abort_transaction(trans, extent_root, ret); |
6735 | goto out; | |
7bb86316 | 6736 | } |
5f39d397 CM |
6737 | |
6738 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6739 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6740 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6741 | if (item_size < sizeof(*ei)) { | |
6742 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
6743 | ret = convert_extent_item_v0(trans, extent_root, path, | |
6744 | owner_objectid, 0); | |
005d6427 DS |
6745 | if (ret < 0) { |
6746 | btrfs_abort_transaction(trans, extent_root, ret); | |
6747 | goto out; | |
6748 | } | |
5d4f98a2 | 6749 | |
b3b4aa74 | 6750 | btrfs_release_path(path); |
5d4f98a2 YZ |
6751 | path->leave_spinning = 1; |
6752 | ||
6753 | key.objectid = bytenr; | |
6754 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6755 | key.offset = num_bytes; | |
6756 | ||
6757 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6758 | -1, 1); | |
6759 | if (ret) { | |
c2cf52eb | 6760 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6761 | ret, bytenr); |
5d4f98a2 YZ |
6762 | btrfs_print_leaf(extent_root, path->nodes[0]); |
6763 | } | |
005d6427 DS |
6764 | if (ret < 0) { |
6765 | btrfs_abort_transaction(trans, extent_root, ret); | |
6766 | goto out; | |
6767 | } | |
6768 | ||
5d4f98a2 YZ |
6769 | extent_slot = path->slots[0]; |
6770 | leaf = path->nodes[0]; | |
6771 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6772 | } | |
6773 | #endif | |
6774 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6775 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6776 | struct btrfs_extent_item); |
3173a18f JB |
6777 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6778 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6779 | struct btrfs_tree_block_info *bi; |
6780 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6781 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6782 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6783 | } | |
56bec294 | 6784 | |
5d4f98a2 | 6785 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
6786 | if (refs < refs_to_drop) { |
6787 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
351fd353 | 6788 | "for bytenr %Lu", refs_to_drop, refs, bytenr); |
32b02538 JB |
6789 | ret = -EINVAL; |
6790 | btrfs_abort_transaction(trans, extent_root, ret); | |
6791 | goto out; | |
6792 | } | |
56bec294 | 6793 | refs -= refs_to_drop; |
5f39d397 | 6794 | |
5d4f98a2 YZ |
6795 | if (refs > 0) { |
6796 | if (extent_op) | |
6797 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6798 | /* | |
6799 | * In the case of inline back ref, reference count will | |
6800 | * be updated by remove_extent_backref | |
952fccac | 6801 | */ |
5d4f98a2 YZ |
6802 | if (iref) { |
6803 | BUG_ON(!found_extent); | |
6804 | } else { | |
6805 | btrfs_set_extent_refs(leaf, ei, refs); | |
6806 | btrfs_mark_buffer_dirty(leaf); | |
6807 | } | |
6808 | if (found_extent) { | |
6809 | ret = remove_extent_backref(trans, extent_root, path, | |
6810 | iref, refs_to_drop, | |
fcebe456 | 6811 | is_data, &last_ref); |
005d6427 DS |
6812 | if (ret) { |
6813 | btrfs_abort_transaction(trans, extent_root, ret); | |
6814 | goto out; | |
6815 | } | |
952fccac | 6816 | } |
b150a4f1 JB |
6817 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
6818 | root_objectid); | |
5d4f98a2 | 6819 | } else { |
5d4f98a2 YZ |
6820 | if (found_extent) { |
6821 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 6822 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
6823 | if (iref) { |
6824 | BUG_ON(path->slots[0] != extent_slot); | |
6825 | } else { | |
6826 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6827 | path->slots[0] = extent_slot; | |
6828 | num_to_del = 2; | |
6829 | } | |
78fae27e | 6830 | } |
b9473439 | 6831 | |
fcebe456 | 6832 | last_ref = 1; |
952fccac CM |
6833 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6834 | num_to_del); | |
005d6427 DS |
6835 | if (ret) { |
6836 | btrfs_abort_transaction(trans, extent_root, ret); | |
6837 | goto out; | |
6838 | } | |
b3b4aa74 | 6839 | btrfs_release_path(path); |
21af804c | 6840 | |
5d4f98a2 | 6841 | if (is_data) { |
459931ec | 6842 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
6843 | if (ret) { |
6844 | btrfs_abort_transaction(trans, extent_root, ret); | |
6845 | goto out; | |
6846 | } | |
459931ec CM |
6847 | } |
6848 | ||
1e144fb8 OS |
6849 | ret = add_to_free_space_tree(trans, root->fs_info, bytenr, |
6850 | num_bytes); | |
6851 | if (ret) { | |
6852 | btrfs_abort_transaction(trans, extent_root, ret); | |
6853 | goto out; | |
6854 | } | |
6855 | ||
ce93ec54 | 6856 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
005d6427 DS |
6857 | if (ret) { |
6858 | btrfs_abort_transaction(trans, extent_root, ret); | |
6859 | goto out; | |
6860 | } | |
a28ec197 | 6861 | } |
fcebe456 JB |
6862 | btrfs_release_path(path); |
6863 | ||
79787eaa | 6864 | out: |
5caf2a00 | 6865 | btrfs_free_path(path); |
a28ec197 CM |
6866 | return ret; |
6867 | } | |
6868 | ||
1887be66 | 6869 | /* |
f0486c68 | 6870 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6871 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6872 | * a given extent, and if there are no other delayed refs to be processed, it | |
6873 | * removes it from the tree. | |
6874 | */ | |
6875 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
6876 | struct btrfs_root *root, u64 bytenr) | |
6877 | { | |
6878 | struct btrfs_delayed_ref_head *head; | |
6879 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6880 | int ret = 0; |
1887be66 CM |
6881 | |
6882 | delayed_refs = &trans->transaction->delayed_refs; | |
6883 | spin_lock(&delayed_refs->lock); | |
6884 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
6885 | if (!head) | |
cf93da7b | 6886 | goto out_delayed_unlock; |
1887be66 | 6887 | |
d7df2c79 | 6888 | spin_lock(&head->lock); |
c6fc2454 | 6889 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
6890 | goto out; |
6891 | ||
5d4f98a2 YZ |
6892 | if (head->extent_op) { |
6893 | if (!head->must_insert_reserved) | |
6894 | goto out; | |
78a6184a | 6895 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
6896 | head->extent_op = NULL; |
6897 | } | |
6898 | ||
1887be66 CM |
6899 | /* |
6900 | * waiting for the lock here would deadlock. If someone else has it | |
6901 | * locked they are already in the process of dropping it anyway | |
6902 | */ | |
6903 | if (!mutex_trylock(&head->mutex)) | |
6904 | goto out; | |
6905 | ||
6906 | /* | |
6907 | * at this point we have a head with no other entries. Go | |
6908 | * ahead and process it. | |
6909 | */ | |
6910 | head->node.in_tree = 0; | |
c46effa6 | 6911 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 6912 | |
d7df2c79 | 6913 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
6914 | |
6915 | /* | |
6916 | * we don't take a ref on the node because we're removing it from the | |
6917 | * tree, so we just steal the ref the tree was holding. | |
6918 | */ | |
c3e69d58 | 6919 | delayed_refs->num_heads--; |
d7df2c79 | 6920 | if (head->processing == 0) |
c3e69d58 | 6921 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
6922 | head->processing = 0; |
6923 | spin_unlock(&head->lock); | |
1887be66 CM |
6924 | spin_unlock(&delayed_refs->lock); |
6925 | ||
f0486c68 YZ |
6926 | BUG_ON(head->extent_op); |
6927 | if (head->must_insert_reserved) | |
6928 | ret = 1; | |
6929 | ||
6930 | mutex_unlock(&head->mutex); | |
1887be66 | 6931 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 6932 | return ret; |
1887be66 | 6933 | out: |
d7df2c79 | 6934 | spin_unlock(&head->lock); |
cf93da7b CM |
6935 | |
6936 | out_delayed_unlock: | |
1887be66 CM |
6937 | spin_unlock(&delayed_refs->lock); |
6938 | return 0; | |
6939 | } | |
6940 | ||
f0486c68 YZ |
6941 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
6942 | struct btrfs_root *root, | |
6943 | struct extent_buffer *buf, | |
5581a51a | 6944 | u64 parent, int last_ref) |
f0486c68 | 6945 | { |
b150a4f1 | 6946 | int pin = 1; |
f0486c68 YZ |
6947 | int ret; |
6948 | ||
6949 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
6950 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6951 | buf->start, buf->len, | |
6952 | parent, root->root_key.objectid, | |
6953 | btrfs_header_level(buf), | |
b06c4bf5 | 6954 | BTRFS_DROP_DELAYED_REF, NULL); |
79787eaa | 6955 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
6956 | } |
6957 | ||
6958 | if (!last_ref) | |
6959 | return; | |
6960 | ||
f0486c68 | 6961 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
6962 | struct btrfs_block_group_cache *cache; |
6963 | ||
f0486c68 YZ |
6964 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
6965 | ret = check_ref_cleanup(trans, root, buf->start); | |
6966 | if (!ret) | |
37be25bc | 6967 | goto out; |
f0486c68 YZ |
6968 | } |
6969 | ||
6219872d FM |
6970 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
6971 | ||
f0486c68 YZ |
6972 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6973 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
6219872d | 6974 | btrfs_put_block_group(cache); |
37be25bc | 6975 | goto out; |
f0486c68 YZ |
6976 | } |
6977 | ||
6978 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
6979 | ||
6980 | btrfs_add_free_space(cache, buf->start, buf->len); | |
e570fd27 | 6981 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0); |
6219872d | 6982 | btrfs_put_block_group(cache); |
0be5dc67 | 6983 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 6984 | pin = 0; |
f0486c68 YZ |
6985 | } |
6986 | out: | |
b150a4f1 JB |
6987 | if (pin) |
6988 | add_pinned_bytes(root->fs_info, buf->len, | |
6989 | btrfs_header_level(buf), | |
6990 | root->root_key.objectid); | |
6991 | ||
a826d6dc JB |
6992 | /* |
6993 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
6994 | * anymore. | |
6995 | */ | |
6996 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
6997 | } |
6998 | ||
79787eaa | 6999 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
7000 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
7001 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
b06c4bf5 | 7002 | u64 owner, u64 offset) |
925baedd CM |
7003 | { |
7004 | int ret; | |
66d7e7f0 | 7005 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 7006 | |
fccb84c9 | 7007 | if (btrfs_test_is_dummy_root(root)) |
faa2dbf0 | 7008 | return 0; |
fccb84c9 | 7009 | |
b150a4f1 JB |
7010 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
7011 | ||
56bec294 CM |
7012 | /* |
7013 | * tree log blocks never actually go into the extent allocation | |
7014 | * tree, just update pinning info and exit early. | |
56bec294 | 7015 | */ |
5d4f98a2 YZ |
7016 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7017 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7018 | /* unlocks the pinned mutex */ |
11833d66 | 7019 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 7020 | ret = 0; |
5d4f98a2 | 7021 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
7022 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7023 | num_bytes, | |
5d4f98a2 | 7024 | parent, root_objectid, (int)owner, |
b06c4bf5 | 7025 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 7026 | } else { |
66d7e7f0 AJ |
7027 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7028 | num_bytes, | |
7029 | parent, root_objectid, owner, | |
5846a3c2 QW |
7030 | offset, 0, |
7031 | BTRFS_DROP_DELAYED_REF, NULL); | |
56bec294 | 7032 | } |
925baedd CM |
7033 | return ret; |
7034 | } | |
7035 | ||
817d52f8 JB |
7036 | /* |
7037 | * when we wait for progress in the block group caching, its because | |
7038 | * our allocation attempt failed at least once. So, we must sleep | |
7039 | * and let some progress happen before we try again. | |
7040 | * | |
7041 | * This function will sleep at least once waiting for new free space to | |
7042 | * show up, and then it will check the block group free space numbers | |
7043 | * for our min num_bytes. Another option is to have it go ahead | |
7044 | * and look in the rbtree for a free extent of a given size, but this | |
7045 | * is a good start. | |
36cce922 JB |
7046 | * |
7047 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7048 | * any of the information in this block group. | |
817d52f8 | 7049 | */ |
36cce922 | 7050 | static noinline void |
817d52f8 JB |
7051 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7052 | u64 num_bytes) | |
7053 | { | |
11833d66 | 7054 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7055 | |
11833d66 YZ |
7056 | caching_ctl = get_caching_control(cache); |
7057 | if (!caching_ctl) | |
36cce922 | 7058 | return; |
817d52f8 | 7059 | |
11833d66 | 7060 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7061 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7062 | |
7063 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7064 | } |
7065 | ||
7066 | static noinline int | |
7067 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7068 | { | |
7069 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7070 | int ret = 0; |
11833d66 YZ |
7071 | |
7072 | caching_ctl = get_caching_control(cache); | |
7073 | if (!caching_ctl) | |
36cce922 | 7074 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7075 | |
7076 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7077 | if (cache->cached == BTRFS_CACHE_ERROR) |
7078 | ret = -EIO; | |
11833d66 | 7079 | put_caching_control(caching_ctl); |
36cce922 | 7080 | return ret; |
817d52f8 JB |
7081 | } |
7082 | ||
31e50229 | 7083 | int __get_raid_index(u64 flags) |
b742bb82 | 7084 | { |
7738a53a | 7085 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7086 | return BTRFS_RAID_RAID10; |
7738a53a | 7087 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7088 | return BTRFS_RAID_RAID1; |
7738a53a | 7089 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7090 | return BTRFS_RAID_DUP; |
7738a53a | 7091 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7092 | return BTRFS_RAID_RAID0; |
53b381b3 | 7093 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7094 | return BTRFS_RAID_RAID5; |
53b381b3 | 7095 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7096 | return BTRFS_RAID_RAID6; |
7738a53a | 7097 | |
e942f883 | 7098 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7099 | } |
7100 | ||
6ab0a202 | 7101 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7102 | { |
31e50229 | 7103 | return __get_raid_index(cache->flags); |
7738a53a ID |
7104 | } |
7105 | ||
6ab0a202 JM |
7106 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7107 | [BTRFS_RAID_RAID10] = "raid10", | |
7108 | [BTRFS_RAID_RAID1] = "raid1", | |
7109 | [BTRFS_RAID_DUP] = "dup", | |
7110 | [BTRFS_RAID_RAID0] = "raid0", | |
7111 | [BTRFS_RAID_SINGLE] = "single", | |
7112 | [BTRFS_RAID_RAID5] = "raid5", | |
7113 | [BTRFS_RAID_RAID6] = "raid6", | |
7114 | }; | |
7115 | ||
1b8e5df6 | 7116 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7117 | { |
7118 | if (type >= BTRFS_NR_RAID_TYPES) | |
7119 | return NULL; | |
7120 | ||
7121 | return btrfs_raid_type_names[type]; | |
7122 | } | |
7123 | ||
817d52f8 | 7124 | enum btrfs_loop_type { |
285ff5af JB |
7125 | LOOP_CACHING_NOWAIT = 0, |
7126 | LOOP_CACHING_WAIT = 1, | |
7127 | LOOP_ALLOC_CHUNK = 2, | |
7128 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7129 | }; |
7130 | ||
e570fd27 MX |
7131 | static inline void |
7132 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7133 | int delalloc) | |
7134 | { | |
7135 | if (delalloc) | |
7136 | down_read(&cache->data_rwsem); | |
7137 | } | |
7138 | ||
7139 | static inline void | |
7140 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7141 | int delalloc) | |
7142 | { | |
7143 | btrfs_get_block_group(cache); | |
7144 | if (delalloc) | |
7145 | down_read(&cache->data_rwsem); | |
7146 | } | |
7147 | ||
7148 | static struct btrfs_block_group_cache * | |
7149 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7150 | struct btrfs_free_cluster *cluster, | |
7151 | int delalloc) | |
7152 | { | |
89771cc9 | 7153 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7154 | |
e570fd27 | 7155 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7156 | while (1) { |
7157 | used_bg = cluster->block_group; | |
7158 | if (!used_bg) | |
7159 | return NULL; | |
7160 | ||
7161 | if (used_bg == block_group) | |
e570fd27 MX |
7162 | return used_bg; |
7163 | ||
6719afdc | 7164 | btrfs_get_block_group(used_bg); |
e570fd27 | 7165 | |
6719afdc GU |
7166 | if (!delalloc) |
7167 | return used_bg; | |
e570fd27 | 7168 | |
6719afdc GU |
7169 | if (down_read_trylock(&used_bg->data_rwsem)) |
7170 | return used_bg; | |
e570fd27 | 7171 | |
6719afdc | 7172 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7173 | |
6719afdc | 7174 | down_read(&used_bg->data_rwsem); |
e570fd27 | 7175 | |
6719afdc GU |
7176 | spin_lock(&cluster->refill_lock); |
7177 | if (used_bg == cluster->block_group) | |
7178 | return used_bg; | |
e570fd27 | 7179 | |
6719afdc GU |
7180 | up_read(&used_bg->data_rwsem); |
7181 | btrfs_put_block_group(used_bg); | |
7182 | } | |
e570fd27 MX |
7183 | } |
7184 | ||
7185 | static inline void | |
7186 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7187 | int delalloc) | |
7188 | { | |
7189 | if (delalloc) | |
7190 | up_read(&cache->data_rwsem); | |
7191 | btrfs_put_block_group(cache); | |
7192 | } | |
7193 | ||
fec577fb CM |
7194 | /* |
7195 | * walks the btree of allocated extents and find a hole of a given size. | |
7196 | * The key ins is changed to record the hole: | |
a4820398 | 7197 | * ins->objectid == start position |
62e2749e | 7198 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7199 | * ins->offset == the size of the hole. |
fec577fb | 7200 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7201 | * |
7202 | * If there is no suitable free space, we will record the max size of | |
7203 | * the free space extent currently. | |
fec577fb | 7204 | */ |
00361589 | 7205 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
98ed5174 | 7206 | u64 num_bytes, u64 empty_size, |
98ed5174 | 7207 | u64 hint_byte, struct btrfs_key *ins, |
e570fd27 | 7208 | u64 flags, int delalloc) |
fec577fb | 7209 | { |
80eb234a | 7210 | int ret = 0; |
d397712b | 7211 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 7212 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7213 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7214 | u64 search_start = 0; |
a4820398 | 7215 | u64 max_extent_size = 0; |
c759c4e1 | 7216 | u64 empty_cluster = 0; |
80eb234a | 7217 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7218 | int loop = 0; |
b6919a58 DS |
7219 | int index = __get_raid_index(flags); |
7220 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 7221 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
0a24325e | 7222 | bool failed_cluster_refill = false; |
1cdda9b8 | 7223 | bool failed_alloc = false; |
67377734 | 7224 | bool use_cluster = true; |
60d2adbb | 7225 | bool have_caching_bg = false; |
13a0db5a | 7226 | bool orig_have_caching_bg = false; |
a5e681d9 | 7227 | bool full_search = false; |
fec577fb | 7228 | |
db94535d | 7229 | WARN_ON(num_bytes < root->sectorsize); |
962a298f | 7230 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7231 | ins->objectid = 0; |
7232 | ins->offset = 0; | |
b1a4d965 | 7233 | |
b6919a58 | 7234 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 7235 | |
b6919a58 | 7236 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 7237 | if (!space_info) { |
b6919a58 | 7238 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7239 | return -ENOSPC; |
7240 | } | |
2552d17e | 7241 | |
67377734 | 7242 | /* |
4f4db217 JB |
7243 | * If our free space is heavily fragmented we may not be able to make |
7244 | * big contiguous allocations, so instead of doing the expensive search | |
7245 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7246 | * can go ahead and search for a more manageable chunk. | |
7247 | * | |
7248 | * If our max_extent_size is large enough for our allocation simply | |
7249 | * disable clustering since we will likely not be able to find enough | |
7250 | * space to create a cluster and induce latency trying. | |
67377734 | 7251 | */ |
4f4db217 JB |
7252 | if (unlikely(space_info->max_extent_size)) { |
7253 | spin_lock(&space_info->lock); | |
7254 | if (space_info->max_extent_size && | |
7255 | num_bytes > space_info->max_extent_size) { | |
7256 | ins->offset = space_info->max_extent_size; | |
7257 | spin_unlock(&space_info->lock); | |
7258 | return -ENOSPC; | |
7259 | } else if (space_info->max_extent_size) { | |
7260 | use_cluster = false; | |
7261 | } | |
7262 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7263 | } |
0f9dd46c | 7264 | |
c759c4e1 | 7265 | last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster); |
239b14b3 | 7266 | if (last_ptr) { |
fa9c0d79 CM |
7267 | spin_lock(&last_ptr->lock); |
7268 | if (last_ptr->block_group) | |
7269 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7270 | if (last_ptr->fragmented) { |
7271 | /* | |
7272 | * We still set window_start so we can keep track of the | |
7273 | * last place we found an allocation to try and save | |
7274 | * some time. | |
7275 | */ | |
7276 | hint_byte = last_ptr->window_start; | |
7277 | use_cluster = false; | |
7278 | } | |
fa9c0d79 | 7279 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7280 | } |
fa9c0d79 | 7281 | |
a061fc8d | 7282 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 7283 | search_start = max(search_start, hint_byte); |
2552d17e | 7284 | if (search_start == hint_byte) { |
2552d17e JB |
7285 | block_group = btrfs_lookup_block_group(root->fs_info, |
7286 | search_start); | |
817d52f8 JB |
7287 | /* |
7288 | * we don't want to use the block group if it doesn't match our | |
7289 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7290 | * |
7291 | * However if we are re-searching with an ideal block group | |
7292 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7293 | */ |
b6919a58 | 7294 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7295 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7296 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7297 | if (list_empty(&block_group->list) || |
7298 | block_group->ro) { | |
7299 | /* | |
7300 | * someone is removing this block group, | |
7301 | * we can't jump into the have_block_group | |
7302 | * target because our list pointers are not | |
7303 | * valid | |
7304 | */ | |
7305 | btrfs_put_block_group(block_group); | |
7306 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7307 | } else { |
b742bb82 | 7308 | index = get_block_group_index(block_group); |
e570fd27 | 7309 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7310 | goto have_block_group; |
ccf0e725 | 7311 | } |
2552d17e | 7312 | } else if (block_group) { |
fa9c0d79 | 7313 | btrfs_put_block_group(block_group); |
2552d17e | 7314 | } |
42e70e7a | 7315 | } |
2552d17e | 7316 | search: |
60d2adbb | 7317 | have_caching_bg = false; |
a5e681d9 JB |
7318 | if (index == 0 || index == __get_raid_index(flags)) |
7319 | full_search = true; | |
80eb234a | 7320 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7321 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7322 | list) { | |
6226cb0a | 7323 | u64 offset; |
817d52f8 | 7324 | int cached; |
8a1413a2 | 7325 | |
e570fd27 | 7326 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7327 | search_start = block_group->key.objectid; |
42e70e7a | 7328 | |
83a50de9 CM |
7329 | /* |
7330 | * this can happen if we end up cycling through all the | |
7331 | * raid types, but we want to make sure we only allocate | |
7332 | * for the proper type. | |
7333 | */ | |
b6919a58 | 7334 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7335 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7336 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7337 | BTRFS_BLOCK_GROUP_RAID5 | |
7338 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7339 | BTRFS_BLOCK_GROUP_RAID10; |
7340 | ||
7341 | /* | |
7342 | * if they asked for extra copies and this block group | |
7343 | * doesn't provide them, bail. This does allow us to | |
7344 | * fill raid0 from raid1. | |
7345 | */ | |
b6919a58 | 7346 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7347 | goto loop; |
7348 | } | |
7349 | ||
2552d17e | 7350 | have_block_group: |
291c7d2f JB |
7351 | cached = block_group_cache_done(block_group); |
7352 | if (unlikely(!cached)) { | |
a5e681d9 | 7353 | have_caching_bg = true; |
f6373bf3 | 7354 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7355 | BUG_ON(ret < 0); |
7356 | ret = 0; | |
817d52f8 JB |
7357 | } |
7358 | ||
36cce922 JB |
7359 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7360 | goto loop; | |
ea6a478e | 7361 | if (unlikely(block_group->ro)) |
2552d17e | 7362 | goto loop; |
0f9dd46c | 7363 | |
0a24325e | 7364 | /* |
062c05c4 AO |
7365 | * Ok we want to try and use the cluster allocator, so |
7366 | * lets look there | |
0a24325e | 7367 | */ |
c759c4e1 | 7368 | if (last_ptr && use_cluster) { |
215a63d1 | 7369 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7370 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7371 | /* |
7372 | * the refill lock keeps out other | |
7373 | * people trying to start a new cluster | |
7374 | */ | |
e570fd27 MX |
7375 | used_block_group = btrfs_lock_cluster(block_group, |
7376 | last_ptr, | |
7377 | delalloc); | |
7378 | if (!used_block_group) | |
44fb5511 | 7379 | goto refill_cluster; |
274bd4fb | 7380 | |
e570fd27 MX |
7381 | if (used_block_group != block_group && |
7382 | (used_block_group->ro || | |
7383 | !block_group_bits(used_block_group, flags))) | |
7384 | goto release_cluster; | |
44fb5511 | 7385 | |
274bd4fb | 7386 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7387 | last_ptr, |
7388 | num_bytes, | |
7389 | used_block_group->key.objectid, | |
7390 | &max_extent_size); | |
fa9c0d79 CM |
7391 | if (offset) { |
7392 | /* we have a block, we're done */ | |
7393 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 | 7394 | trace_btrfs_reserve_extent_cluster(root, |
89d4346a MX |
7395 | used_block_group, |
7396 | search_start, num_bytes); | |
215a63d1 | 7397 | if (used_block_group != block_group) { |
e570fd27 MX |
7398 | btrfs_release_block_group(block_group, |
7399 | delalloc); | |
215a63d1 MX |
7400 | block_group = used_block_group; |
7401 | } | |
fa9c0d79 CM |
7402 | goto checks; |
7403 | } | |
7404 | ||
274bd4fb | 7405 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7406 | release_cluster: |
062c05c4 AO |
7407 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7408 | * set up a new clusters, so lets just skip it | |
7409 | * and let the allocator find whatever block | |
7410 | * it can find. If we reach this point, we | |
7411 | * will have tried the cluster allocator | |
7412 | * plenty of times and not have found | |
7413 | * anything, so we are likely way too | |
7414 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7415 | * anything. |
7416 | * | |
7417 | * However, if the cluster is taken from the | |
7418 | * current block group, release the cluster | |
7419 | * first, so that we stand a better chance of | |
7420 | * succeeding in the unclustered | |
7421 | * allocation. */ | |
7422 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7423 | used_block_group != block_group) { |
062c05c4 | 7424 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7425 | btrfs_release_block_group(used_block_group, |
7426 | delalloc); | |
062c05c4 AO |
7427 | goto unclustered_alloc; |
7428 | } | |
7429 | ||
fa9c0d79 CM |
7430 | /* |
7431 | * this cluster didn't work out, free it and | |
7432 | * start over | |
7433 | */ | |
7434 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7435 | ||
e570fd27 MX |
7436 | if (used_block_group != block_group) |
7437 | btrfs_release_block_group(used_block_group, | |
7438 | delalloc); | |
7439 | refill_cluster: | |
a5f6f719 AO |
7440 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7441 | spin_unlock(&last_ptr->refill_lock); | |
7442 | goto unclustered_alloc; | |
7443 | } | |
7444 | ||
8de972b4 CM |
7445 | aligned_cluster = max_t(unsigned long, |
7446 | empty_cluster + empty_size, | |
7447 | block_group->full_stripe_len); | |
7448 | ||
fa9c0d79 | 7449 | /* allocate a cluster in this block group */ |
00361589 JB |
7450 | ret = btrfs_find_space_cluster(root, block_group, |
7451 | last_ptr, search_start, | |
7452 | num_bytes, | |
7453 | aligned_cluster); | |
fa9c0d79 CM |
7454 | if (ret == 0) { |
7455 | /* | |
7456 | * now pull our allocation out of this | |
7457 | * cluster | |
7458 | */ | |
7459 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7460 | last_ptr, |
7461 | num_bytes, | |
7462 | search_start, | |
7463 | &max_extent_size); | |
fa9c0d79 CM |
7464 | if (offset) { |
7465 | /* we found one, proceed */ | |
7466 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
7467 | trace_btrfs_reserve_extent_cluster(root, |
7468 | block_group, search_start, | |
7469 | num_bytes); | |
fa9c0d79 CM |
7470 | goto checks; |
7471 | } | |
0a24325e JB |
7472 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7473 | && !failed_cluster_refill) { | |
817d52f8 JB |
7474 | spin_unlock(&last_ptr->refill_lock); |
7475 | ||
0a24325e | 7476 | failed_cluster_refill = true; |
817d52f8 JB |
7477 | wait_block_group_cache_progress(block_group, |
7478 | num_bytes + empty_cluster + empty_size); | |
7479 | goto have_block_group; | |
fa9c0d79 | 7480 | } |
817d52f8 | 7481 | |
fa9c0d79 CM |
7482 | /* |
7483 | * at this point we either didn't find a cluster | |
7484 | * or we weren't able to allocate a block from our | |
7485 | * cluster. Free the cluster we've been trying | |
7486 | * to use, and go to the next block group | |
7487 | */ | |
0a24325e | 7488 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7489 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7490 | goto loop; |
fa9c0d79 CM |
7491 | } |
7492 | ||
062c05c4 | 7493 | unclustered_alloc: |
c759c4e1 JB |
7494 | /* |
7495 | * We are doing an unclustered alloc, set the fragmented flag so | |
7496 | * we don't bother trying to setup a cluster again until we get | |
7497 | * more space. | |
7498 | */ | |
7499 | if (unlikely(last_ptr)) { | |
7500 | spin_lock(&last_ptr->lock); | |
7501 | last_ptr->fragmented = 1; | |
7502 | spin_unlock(&last_ptr->lock); | |
7503 | } | |
a5f6f719 AO |
7504 | spin_lock(&block_group->free_space_ctl->tree_lock); |
7505 | if (cached && | |
7506 | block_group->free_space_ctl->free_space < | |
7507 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
7508 | if (block_group->free_space_ctl->free_space > |
7509 | max_extent_size) | |
7510 | max_extent_size = | |
7511 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
7512 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
7513 | goto loop; | |
7514 | } | |
7515 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7516 | ||
6226cb0a | 7517 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7518 | num_bytes, empty_size, |
7519 | &max_extent_size); | |
1cdda9b8 JB |
7520 | /* |
7521 | * If we didn't find a chunk, and we haven't failed on this | |
7522 | * block group before, and this block group is in the middle of | |
7523 | * caching and we are ok with waiting, then go ahead and wait | |
7524 | * for progress to be made, and set failed_alloc to true. | |
7525 | * | |
7526 | * If failed_alloc is true then we've already waited on this | |
7527 | * block group once and should move on to the next block group. | |
7528 | */ | |
7529 | if (!offset && !failed_alloc && !cached && | |
7530 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7531 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7532 | num_bytes + empty_size); |
7533 | failed_alloc = true; | |
817d52f8 | 7534 | goto have_block_group; |
1cdda9b8 JB |
7535 | } else if (!offset) { |
7536 | goto loop; | |
817d52f8 | 7537 | } |
fa9c0d79 | 7538 | checks: |
4e54b17a | 7539 | search_start = ALIGN(offset, root->stripesize); |
25179201 | 7540 | |
2552d17e JB |
7541 | /* move on to the next group */ |
7542 | if (search_start + num_bytes > | |
215a63d1 MX |
7543 | block_group->key.objectid + block_group->key.offset) { |
7544 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7545 | goto loop; |
6226cb0a | 7546 | } |
f5a31e16 | 7547 | |
f0486c68 | 7548 | if (offset < search_start) |
215a63d1 | 7549 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7550 | search_start - offset); |
7551 | BUG_ON(offset > search_start); | |
2552d17e | 7552 | |
215a63d1 | 7553 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
e570fd27 | 7554 | alloc_type, delalloc); |
f0486c68 | 7555 | if (ret == -EAGAIN) { |
215a63d1 | 7556 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7557 | goto loop; |
0f9dd46c | 7558 | } |
9cfa3e34 | 7559 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7560 | |
f0486c68 | 7561 | /* we are all good, lets return */ |
2552d17e JB |
7562 | ins->objectid = search_start; |
7563 | ins->offset = num_bytes; | |
d2fb3437 | 7564 | |
3f7de037 JB |
7565 | trace_btrfs_reserve_extent(orig_root, block_group, |
7566 | search_start, num_bytes); | |
e570fd27 | 7567 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7568 | break; |
7569 | loop: | |
0a24325e | 7570 | failed_cluster_refill = false; |
1cdda9b8 | 7571 | failed_alloc = false; |
b742bb82 | 7572 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7573 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7574 | } |
7575 | up_read(&space_info->groups_sem); | |
7576 | ||
13a0db5a | 7577 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7578 | && !orig_have_caching_bg) | |
7579 | orig_have_caching_bg = true; | |
7580 | ||
60d2adbb MX |
7581 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7582 | goto search; | |
7583 | ||
b742bb82 YZ |
7584 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7585 | goto search; | |
7586 | ||
285ff5af | 7587 | /* |
ccf0e725 JB |
7588 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7589 | * caching kthreads as we move along | |
817d52f8 JB |
7590 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7591 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7592 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7593 | * again | |
fa9c0d79 | 7594 | */ |
723bda20 | 7595 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7596 | index = 0; |
a5e681d9 JB |
7597 | if (loop == LOOP_CACHING_NOWAIT) { |
7598 | /* | |
7599 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7600 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7601 | * full search through. |
7602 | */ | |
13a0db5a | 7603 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7604 | loop = LOOP_CACHING_WAIT; |
7605 | else | |
7606 | loop = LOOP_ALLOC_CHUNK; | |
7607 | } else { | |
7608 | loop++; | |
7609 | } | |
7610 | ||
817d52f8 | 7611 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7612 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7613 | int exist = 0; |
7614 | ||
7615 | trans = current->journal_info; | |
7616 | if (trans) | |
7617 | exist = 1; | |
7618 | else | |
7619 | trans = btrfs_join_transaction(root); | |
00361589 | 7620 | |
00361589 JB |
7621 | if (IS_ERR(trans)) { |
7622 | ret = PTR_ERR(trans); | |
7623 | goto out; | |
7624 | } | |
7625 | ||
b6919a58 | 7626 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad | 7627 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7628 | |
7629 | /* | |
7630 | * If we can't allocate a new chunk we've already looped | |
7631 | * through at least once, move on to the NO_EMPTY_SIZE | |
7632 | * case. | |
7633 | */ | |
7634 | if (ret == -ENOSPC) | |
7635 | loop = LOOP_NO_EMPTY_SIZE; | |
7636 | ||
ea658bad JB |
7637 | /* |
7638 | * Do not bail out on ENOSPC since we | |
7639 | * can do more things. | |
7640 | */ | |
00361589 | 7641 | if (ret < 0 && ret != -ENOSPC) |
ea658bad JB |
7642 | btrfs_abort_transaction(trans, |
7643 | root, ret); | |
00361589 JB |
7644 | else |
7645 | ret = 0; | |
f017f15f WS |
7646 | if (!exist) |
7647 | btrfs_end_transaction(trans, root); | |
00361589 | 7648 | if (ret) |
ea658bad | 7649 | goto out; |
2552d17e JB |
7650 | } |
7651 | ||
723bda20 | 7652 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7653 | /* |
7654 | * Don't loop again if we already have no empty_size and | |
7655 | * no empty_cluster. | |
7656 | */ | |
7657 | if (empty_size == 0 && | |
7658 | empty_cluster == 0) { | |
7659 | ret = -ENOSPC; | |
7660 | goto out; | |
7661 | } | |
723bda20 JB |
7662 | empty_size = 0; |
7663 | empty_cluster = 0; | |
fa9c0d79 | 7664 | } |
723bda20 JB |
7665 | |
7666 | goto search; | |
2552d17e JB |
7667 | } else if (!ins->objectid) { |
7668 | ret = -ENOSPC; | |
d82a6f1d | 7669 | } else if (ins->objectid) { |
c759c4e1 JB |
7670 | if (!use_cluster && last_ptr) { |
7671 | spin_lock(&last_ptr->lock); | |
7672 | last_ptr->window_start = ins->objectid; | |
7673 | spin_unlock(&last_ptr->lock); | |
7674 | } | |
80eb234a | 7675 | ret = 0; |
be744175 | 7676 | } |
79787eaa | 7677 | out: |
4f4db217 JB |
7678 | if (ret == -ENOSPC) { |
7679 | spin_lock(&space_info->lock); | |
7680 | space_info->max_extent_size = max_extent_size; | |
7681 | spin_unlock(&space_info->lock); | |
a4820398 | 7682 | ins->offset = max_extent_size; |
4f4db217 | 7683 | } |
0f70abe2 | 7684 | return ret; |
fec577fb | 7685 | } |
ec44a35c | 7686 | |
9ed74f2d JB |
7687 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
7688 | int dump_block_groups) | |
0f9dd46c JB |
7689 | { |
7690 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7691 | int index = 0; |
0f9dd46c | 7692 | |
9ed74f2d | 7693 | spin_lock(&info->lock); |
efe120a0 | 7694 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", |
c1c9ff7c GU |
7695 | info->flags, |
7696 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
7697 | info->bytes_reserved - info->bytes_readonly, | |
d397712b | 7698 | (info->full) ? "" : "not "); |
efe120a0 | 7699 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " |
8929ecfa | 7700 | "reserved=%llu, may_use=%llu, readonly=%llu\n", |
c1c9ff7c GU |
7701 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
7702 | info->bytes_reserved, info->bytes_may_use, | |
7703 | info->bytes_readonly); | |
9ed74f2d JB |
7704 | spin_unlock(&info->lock); |
7705 | ||
7706 | if (!dump_block_groups) | |
7707 | return; | |
0f9dd46c | 7708 | |
80eb234a | 7709 | down_read(&info->groups_sem); |
b742bb82 YZ |
7710 | again: |
7711 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7712 | spin_lock(&cache->lock); |
efe120a0 FH |
7713 | printk(KERN_INFO "BTRFS: " |
7714 | "block group %llu has %llu bytes, " | |
7715 | "%llu used %llu pinned %llu reserved %s\n", | |
c1c9ff7c GU |
7716 | cache->key.objectid, cache->key.offset, |
7717 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7718 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7719 | btrfs_dump_free_space(cache, bytes); |
7720 | spin_unlock(&cache->lock); | |
7721 | } | |
b742bb82 YZ |
7722 | if (++index < BTRFS_NR_RAID_TYPES) |
7723 | goto again; | |
80eb234a | 7724 | up_read(&info->groups_sem); |
0f9dd46c | 7725 | } |
e8569813 | 7726 | |
00361589 | 7727 | int btrfs_reserve_extent(struct btrfs_root *root, |
11833d66 YZ |
7728 | u64 num_bytes, u64 min_alloc_size, |
7729 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7730 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7731 | { |
36af4e07 | 7732 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7733 | u64 flags; |
fec577fb | 7734 | int ret; |
925baedd | 7735 | |
b6919a58 | 7736 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 7737 | again: |
db94535d | 7738 | WARN_ON(num_bytes < root->sectorsize); |
00361589 | 7739 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
e570fd27 | 7740 | flags, delalloc); |
9cfa3e34 FM |
7741 | if (!ret && !is_data) { |
7742 | btrfs_dec_block_group_reservations(root->fs_info, | |
7743 | ins->objectid); | |
7744 | } else if (ret == -ENOSPC) { | |
a4820398 MX |
7745 | if (!final_tried && ins->offset) { |
7746 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 7747 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 7748 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
7749 | if (num_bytes == min_alloc_size) |
7750 | final_tried = true; | |
7751 | goto again; | |
7752 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
7753 | struct btrfs_space_info *sinfo; | |
7754 | ||
b6919a58 | 7755 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 7756 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
c1c9ff7c | 7757 | flags, num_bytes); |
53804280 JM |
7758 | if (sinfo) |
7759 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 7760 | } |
925baedd | 7761 | } |
0f9dd46c JB |
7762 | |
7763 | return ret; | |
e6dcd2dc CM |
7764 | } |
7765 | ||
e688b725 | 7766 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 MX |
7767 | u64 start, u64 len, |
7768 | int pin, int delalloc) | |
65b51a00 | 7769 | { |
0f9dd46c | 7770 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7771 | int ret = 0; |
0f9dd46c | 7772 | |
0f9dd46c JB |
7773 | cache = btrfs_lookup_block_group(root->fs_info, start); |
7774 | if (!cache) { | |
c2cf52eb | 7775 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 7776 | start); |
0f9dd46c JB |
7777 | return -ENOSPC; |
7778 | } | |
1f3c79a2 | 7779 | |
e688b725 CM |
7780 | if (pin) |
7781 | pin_down_extent(root, cache, start, len, 1); | |
7782 | else { | |
dcc82f47 FM |
7783 | if (btrfs_test_opt(root, DISCARD)) |
7784 | ret = btrfs_discard_extent(root, start, len, NULL); | |
e688b725 | 7785 | btrfs_add_free_space(cache, start, len); |
e570fd27 | 7786 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc); |
e688b725 | 7787 | } |
31193213 | 7788 | |
fa9c0d79 | 7789 | btrfs_put_block_group(cache); |
817d52f8 | 7790 | |
1abe9b8a | 7791 | trace_btrfs_reserved_extent_free(root, start, len); |
7792 | ||
e6dcd2dc CM |
7793 | return ret; |
7794 | } | |
7795 | ||
e688b725 | 7796 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 | 7797 | u64 start, u64 len, int delalloc) |
e688b725 | 7798 | { |
e570fd27 | 7799 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); |
e688b725 CM |
7800 | } |
7801 | ||
7802 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
7803 | u64 start, u64 len) | |
7804 | { | |
e570fd27 | 7805 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); |
e688b725 CM |
7806 | } |
7807 | ||
5d4f98a2 YZ |
7808 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
7809 | struct btrfs_root *root, | |
7810 | u64 parent, u64 root_objectid, | |
7811 | u64 flags, u64 owner, u64 offset, | |
7812 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
7813 | { |
7814 | int ret; | |
5d4f98a2 | 7815 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 7816 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 7817 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 7818 | struct btrfs_path *path; |
5d4f98a2 YZ |
7819 | struct extent_buffer *leaf; |
7820 | int type; | |
7821 | u32 size; | |
26b8003f | 7822 | |
5d4f98a2 YZ |
7823 | if (parent > 0) |
7824 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7825 | else | |
7826 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7827 | |
5d4f98a2 | 7828 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7829 | |
7830 | path = btrfs_alloc_path(); | |
db5b493a TI |
7831 | if (!path) |
7832 | return -ENOMEM; | |
47e4bb98 | 7833 | |
b9473439 | 7834 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7835 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7836 | ins, size); | |
79787eaa JM |
7837 | if (ret) { |
7838 | btrfs_free_path(path); | |
7839 | return ret; | |
7840 | } | |
0f9dd46c | 7841 | |
5d4f98a2 YZ |
7842 | leaf = path->nodes[0]; |
7843 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7844 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7845 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7846 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7847 | btrfs_set_extent_flags(leaf, extent_item, | |
7848 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7849 | ||
7850 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7851 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7852 | if (parent > 0) { | |
7853 | struct btrfs_shared_data_ref *ref; | |
7854 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7855 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7856 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7857 | } else { | |
7858 | struct btrfs_extent_data_ref *ref; | |
7859 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7860 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7861 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7862 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7863 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7864 | } | |
47e4bb98 CM |
7865 | |
7866 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7867 | btrfs_free_path(path); |
f510cfec | 7868 | |
1e144fb8 OS |
7869 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
7870 | ins->offset); | |
7871 | if (ret) | |
7872 | return ret; | |
7873 | ||
ce93ec54 | 7874 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 7875 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7876 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7877 | ins->objectid, ins->offset); |
f5947066 CM |
7878 | BUG(); |
7879 | } | |
0be5dc67 | 7880 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
7881 | return ret; |
7882 | } | |
7883 | ||
5d4f98a2 YZ |
7884 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
7885 | struct btrfs_root *root, | |
7886 | u64 parent, u64 root_objectid, | |
7887 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 7888 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
7889 | { |
7890 | int ret; | |
5d4f98a2 YZ |
7891 | struct btrfs_fs_info *fs_info = root->fs_info; |
7892 | struct btrfs_extent_item *extent_item; | |
7893 | struct btrfs_tree_block_info *block_info; | |
7894 | struct btrfs_extent_inline_ref *iref; | |
7895 | struct btrfs_path *path; | |
7896 | struct extent_buffer *leaf; | |
3173a18f | 7897 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 7898 | u64 num_bytes = ins->offset; |
3173a18f JB |
7899 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7900 | SKINNY_METADATA); | |
7901 | ||
7902 | if (!skinny_metadata) | |
7903 | size += sizeof(*block_info); | |
1c2308f8 | 7904 | |
5d4f98a2 | 7905 | path = btrfs_alloc_path(); |
857cc2fc JB |
7906 | if (!path) { |
7907 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
707e8a07 | 7908 | root->nodesize); |
d8926bb3 | 7909 | return -ENOMEM; |
857cc2fc | 7910 | } |
56bec294 | 7911 | |
5d4f98a2 YZ |
7912 | path->leave_spinning = 1; |
7913 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
7914 | ins, size); | |
79787eaa | 7915 | if (ret) { |
dd825259 | 7916 | btrfs_free_path(path); |
857cc2fc | 7917 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
707e8a07 | 7918 | root->nodesize); |
79787eaa JM |
7919 | return ret; |
7920 | } | |
5d4f98a2 YZ |
7921 | |
7922 | leaf = path->nodes[0]; | |
7923 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7924 | struct btrfs_extent_item); | |
7925 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7926 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7927 | btrfs_set_extent_flags(leaf, extent_item, | |
7928 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 7929 | |
3173a18f JB |
7930 | if (skinny_metadata) { |
7931 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
707e8a07 | 7932 | num_bytes = root->nodesize; |
3173a18f JB |
7933 | } else { |
7934 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
7935 | btrfs_set_tree_block_key(leaf, block_info, key); | |
7936 | btrfs_set_tree_block_level(leaf, block_info, level); | |
7937 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
7938 | } | |
5d4f98a2 | 7939 | |
5d4f98a2 YZ |
7940 | if (parent > 0) { |
7941 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
7942 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7943 | BTRFS_SHARED_BLOCK_REF_KEY); | |
7944 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7945 | } else { | |
7946 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7947 | BTRFS_TREE_BLOCK_REF_KEY); | |
7948 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
7949 | } | |
7950 | ||
7951 | btrfs_mark_buffer_dirty(leaf); | |
7952 | btrfs_free_path(path); | |
7953 | ||
1e144fb8 OS |
7954 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
7955 | num_bytes); | |
7956 | if (ret) | |
7957 | return ret; | |
7958 | ||
ce93ec54 JB |
7959 | ret = update_block_group(trans, root, ins->objectid, root->nodesize, |
7960 | 1); | |
79787eaa | 7961 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7962 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7963 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
7964 | BUG(); |
7965 | } | |
0be5dc67 | 7966 | |
707e8a07 | 7967 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize); |
5d4f98a2 YZ |
7968 | return ret; |
7969 | } | |
7970 | ||
7971 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
7972 | struct btrfs_root *root, | |
7973 | u64 root_objectid, u64 owner, | |
5846a3c2 QW |
7974 | u64 offset, u64 ram_bytes, |
7975 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
7976 | { |
7977 | int ret; | |
7978 | ||
7979 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
7980 | ||
66d7e7f0 AJ |
7981 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
7982 | ins->offset, 0, | |
7983 | root_objectid, owner, offset, | |
5846a3c2 QW |
7984 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT, |
7985 | NULL); | |
e6dcd2dc CM |
7986 | return ret; |
7987 | } | |
e02119d5 CM |
7988 | |
7989 | /* | |
7990 | * this is used by the tree logging recovery code. It records that | |
7991 | * an extent has been allocated and makes sure to clear the free | |
7992 | * space cache bits as well | |
7993 | */ | |
5d4f98a2 YZ |
7994 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
7995 | struct btrfs_root *root, | |
7996 | u64 root_objectid, u64 owner, u64 offset, | |
7997 | struct btrfs_key *ins) | |
e02119d5 CM |
7998 | { |
7999 | int ret; | |
8000 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 8001 | |
8c2a1a30 JB |
8002 | /* |
8003 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8004 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 JB |
8005 | */ |
8006 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
8007 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 8008 | if (ret) |
8c2a1a30 | 8009 | return ret; |
11833d66 YZ |
8010 | } |
8011 | ||
8c2a1a30 JB |
8012 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
8013 | if (!block_group) | |
8014 | return -EINVAL; | |
8015 | ||
fb25e914 | 8016 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
e570fd27 | 8017 | RESERVE_ALLOC_NO_ACCOUNT, 0); |
79787eaa | 8018 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
8019 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
8020 | 0, owner, offset, ins, 1); | |
b50c6e25 | 8021 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8022 | return ret; |
8023 | } | |
8024 | ||
48a3b636 ES |
8025 | static struct extent_buffer * |
8026 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8027 | u64 bytenr, int level) |
65b51a00 CM |
8028 | { |
8029 | struct extent_buffer *buf; | |
8030 | ||
a83fffb7 | 8031 | buf = btrfs_find_create_tree_block(root, bytenr); |
c871b0f2 LB |
8032 | if (IS_ERR(buf)) |
8033 | return buf; | |
8034 | ||
65b51a00 | 8035 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8036 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8037 | btrfs_tree_lock(buf); |
01d58472 | 8038 | clean_tree_block(trans, root->fs_info, buf); |
3083ee2e | 8039 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8040 | |
8041 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8042 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8043 | |
d0c803c4 | 8044 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8045 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8046 | /* |
8047 | * we allow two log transactions at a time, use different | |
8048 | * EXENT bit to differentiate dirty pages. | |
8049 | */ | |
656f30db | 8050 | if (buf->log_index == 0) |
8cef4e16 YZ |
8051 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8052 | buf->start + buf->len - 1, GFP_NOFS); | |
8053 | else | |
8054 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8055 | buf->start + buf->len - 1); |
d0c803c4 | 8056 | } else { |
656f30db | 8057 | buf->log_index = -1; |
d0c803c4 | 8058 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8059 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8060 | } |
64c12921 | 8061 | trans->dirty = true; |
b4ce94de | 8062 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8063 | return buf; |
8064 | } | |
8065 | ||
f0486c68 YZ |
8066 | static struct btrfs_block_rsv * |
8067 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8068 | struct btrfs_root *root, u32 blocksize) | |
8069 | { | |
8070 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 8071 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 8072 | int ret; |
d88033db | 8073 | bool global_updated = false; |
f0486c68 YZ |
8074 | |
8075 | block_rsv = get_block_rsv(trans, root); | |
8076 | ||
b586b323 MX |
8077 | if (unlikely(block_rsv->size == 0)) |
8078 | goto try_reserve; | |
d88033db | 8079 | again: |
f0486c68 YZ |
8080 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8081 | if (!ret) | |
8082 | return block_rsv; | |
8083 | ||
b586b323 MX |
8084 | if (block_rsv->failfast) |
8085 | return ERR_PTR(ret); | |
8086 | ||
d88033db MX |
8087 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8088 | global_updated = true; | |
8089 | update_global_block_rsv(root->fs_info); | |
8090 | goto again; | |
8091 | } | |
8092 | ||
b586b323 MX |
8093 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
8094 | static DEFINE_RATELIMIT_STATE(_rs, | |
8095 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8096 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8097 | if (__ratelimit(&_rs)) | |
8098 | WARN(1, KERN_DEBUG | |
efe120a0 | 8099 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8100 | } |
8101 | try_reserve: | |
8102 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8103 | BTRFS_RESERVE_NO_FLUSH); | |
8104 | if (!ret) | |
8105 | return block_rsv; | |
8106 | /* | |
8107 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8108 | * the global reserve if its space type is the same as the global |
8109 | * reservation. | |
b586b323 | 8110 | */ |
5881cfc9 MX |
8111 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8112 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8113 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8114 | if (!ret) | |
8115 | return global_rsv; | |
8116 | } | |
8117 | return ERR_PTR(ret); | |
f0486c68 YZ |
8118 | } |
8119 | ||
8c2a3ca2 JB |
8120 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8121 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8122 | { |
8123 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8124 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8125 | } |
8126 | ||
fec577fb | 8127 | /* |
f0486c68 | 8128 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8129 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8130 | */ |
4d75f8a9 DS |
8131 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
8132 | struct btrfs_root *root, | |
5d4f98a2 YZ |
8133 | u64 parent, u64 root_objectid, |
8134 | struct btrfs_disk_key *key, int level, | |
5581a51a | 8135 | u64 hint, u64 empty_size) |
fec577fb | 8136 | { |
e2fa7227 | 8137 | struct btrfs_key ins; |
f0486c68 | 8138 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8139 | struct extent_buffer *buf; |
67b7859e | 8140 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8141 | u64 flags = 0; |
8142 | int ret; | |
4d75f8a9 | 8143 | u32 blocksize = root->nodesize; |
3173a18f JB |
8144 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
8145 | SKINNY_METADATA); | |
fec577fb | 8146 | |
fccb84c9 | 8147 | if (btrfs_test_is_dummy_root(root)) { |
faa2dbf0 | 8148 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8149 | level); |
faa2dbf0 JB |
8150 | if (!IS_ERR(buf)) |
8151 | root->alloc_bytenr += blocksize; | |
8152 | return buf; | |
8153 | } | |
fccb84c9 | 8154 | |
f0486c68 YZ |
8155 | block_rsv = use_block_rsv(trans, root, blocksize); |
8156 | if (IS_ERR(block_rsv)) | |
8157 | return ERR_CAST(block_rsv); | |
8158 | ||
00361589 | 8159 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
e570fd27 | 8160 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8161 | if (ret) |
8162 | goto out_unuse; | |
55c69072 | 8163 | |
fe864576 | 8164 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8165 | if (IS_ERR(buf)) { |
8166 | ret = PTR_ERR(buf); | |
8167 | goto out_free_reserved; | |
8168 | } | |
f0486c68 YZ |
8169 | |
8170 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8171 | if (parent == 0) | |
8172 | parent = ins.objectid; | |
8173 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8174 | } else | |
8175 | BUG_ON(parent > 0); | |
8176 | ||
8177 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8178 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8179 | if (!extent_op) { |
8180 | ret = -ENOMEM; | |
8181 | goto out_free_buf; | |
8182 | } | |
f0486c68 YZ |
8183 | if (key) |
8184 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8185 | else | |
8186 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8187 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8188 | extent_op->update_key = skinny_metadata ? false : true; |
8189 | extent_op->update_flags = true; | |
8190 | extent_op->is_data = false; | |
b1c79e09 | 8191 | extent_op->level = level; |
f0486c68 | 8192 | |
66d7e7f0 | 8193 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
67b7859e OS |
8194 | ins.objectid, ins.offset, |
8195 | parent, root_objectid, level, | |
8196 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 8197 | extent_op); |
67b7859e OS |
8198 | if (ret) |
8199 | goto out_free_delayed; | |
f0486c68 | 8200 | } |
fec577fb | 8201 | return buf; |
67b7859e OS |
8202 | |
8203 | out_free_delayed: | |
8204 | btrfs_free_delayed_extent_op(extent_op); | |
8205 | out_free_buf: | |
8206 | free_extent_buffer(buf); | |
8207 | out_free_reserved: | |
8208 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0); | |
8209 | out_unuse: | |
8210 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); | |
8211 | return ERR_PTR(ret); | |
fec577fb | 8212 | } |
a28ec197 | 8213 | |
2c47e605 YZ |
8214 | struct walk_control { |
8215 | u64 refs[BTRFS_MAX_LEVEL]; | |
8216 | u64 flags[BTRFS_MAX_LEVEL]; | |
8217 | struct btrfs_key update_progress; | |
8218 | int stage; | |
8219 | int level; | |
8220 | int shared_level; | |
8221 | int update_ref; | |
8222 | int keep_locks; | |
1c4850e2 YZ |
8223 | int reada_slot; |
8224 | int reada_count; | |
66d7e7f0 | 8225 | int for_reloc; |
2c47e605 YZ |
8226 | }; |
8227 | ||
8228 | #define DROP_REFERENCE 1 | |
8229 | #define UPDATE_BACKREF 2 | |
8230 | ||
1c4850e2 YZ |
8231 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8232 | struct btrfs_root *root, | |
8233 | struct walk_control *wc, | |
8234 | struct btrfs_path *path) | |
6407bf6d | 8235 | { |
1c4850e2 YZ |
8236 | u64 bytenr; |
8237 | u64 generation; | |
8238 | u64 refs; | |
94fcca9f | 8239 | u64 flags; |
5d4f98a2 | 8240 | u32 nritems; |
1c4850e2 YZ |
8241 | u32 blocksize; |
8242 | struct btrfs_key key; | |
8243 | struct extent_buffer *eb; | |
6407bf6d | 8244 | int ret; |
1c4850e2 YZ |
8245 | int slot; |
8246 | int nread = 0; | |
6407bf6d | 8247 | |
1c4850e2 YZ |
8248 | if (path->slots[wc->level] < wc->reada_slot) { |
8249 | wc->reada_count = wc->reada_count * 2 / 3; | |
8250 | wc->reada_count = max(wc->reada_count, 2); | |
8251 | } else { | |
8252 | wc->reada_count = wc->reada_count * 3 / 2; | |
8253 | wc->reada_count = min_t(int, wc->reada_count, | |
8254 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
8255 | } | |
7bb86316 | 8256 | |
1c4850e2 YZ |
8257 | eb = path->nodes[wc->level]; |
8258 | nritems = btrfs_header_nritems(eb); | |
707e8a07 | 8259 | blocksize = root->nodesize; |
bd56b302 | 8260 | |
1c4850e2 YZ |
8261 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8262 | if (nread >= wc->reada_count) | |
8263 | break; | |
bd56b302 | 8264 | |
2dd3e67b | 8265 | cond_resched(); |
1c4850e2 YZ |
8266 | bytenr = btrfs_node_blockptr(eb, slot); |
8267 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8268 | |
1c4850e2 YZ |
8269 | if (slot == path->slots[wc->level]) |
8270 | goto reada; | |
5d4f98a2 | 8271 | |
1c4850e2 YZ |
8272 | if (wc->stage == UPDATE_BACKREF && |
8273 | generation <= root->root_key.offset) | |
bd56b302 CM |
8274 | continue; |
8275 | ||
94fcca9f | 8276 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
8277 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
8278 | wc->level - 1, 1, &refs, | |
8279 | &flags); | |
79787eaa JM |
8280 | /* We don't care about errors in readahead. */ |
8281 | if (ret < 0) | |
8282 | continue; | |
94fcca9f YZ |
8283 | BUG_ON(refs == 0); |
8284 | ||
1c4850e2 | 8285 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8286 | if (refs == 1) |
8287 | goto reada; | |
bd56b302 | 8288 | |
94fcca9f YZ |
8289 | if (wc->level == 1 && |
8290 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8291 | continue; | |
1c4850e2 YZ |
8292 | if (!wc->update_ref || |
8293 | generation <= root->root_key.offset) | |
8294 | continue; | |
8295 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8296 | ret = btrfs_comp_cpu_keys(&key, | |
8297 | &wc->update_progress); | |
8298 | if (ret < 0) | |
8299 | continue; | |
94fcca9f YZ |
8300 | } else { |
8301 | if (wc->level == 1 && | |
8302 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8303 | continue; | |
6407bf6d | 8304 | } |
1c4850e2 | 8305 | reada: |
d3e46fea | 8306 | readahead_tree_block(root, bytenr); |
1c4850e2 | 8307 | nread++; |
20524f02 | 8308 | } |
1c4850e2 | 8309 | wc->reada_slot = slot; |
20524f02 | 8310 | } |
2c47e605 | 8311 | |
0ed4792a | 8312 | /* |
82bd101b MF |
8313 | * These may not be seen by the usual inc/dec ref code so we have to |
8314 | * add them here. | |
0ed4792a | 8315 | */ |
82bd101b MF |
8316 | static int record_one_subtree_extent(struct btrfs_trans_handle *trans, |
8317 | struct btrfs_root *root, u64 bytenr, | |
8318 | u64 num_bytes) | |
8319 | { | |
8320 | struct btrfs_qgroup_extent_record *qrecord; | |
8321 | struct btrfs_delayed_ref_root *delayed_refs; | |
8322 | ||
8323 | qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS); | |
8324 | if (!qrecord) | |
8325 | return -ENOMEM; | |
8326 | ||
8327 | qrecord->bytenr = bytenr; | |
8328 | qrecord->num_bytes = num_bytes; | |
8329 | qrecord->old_roots = NULL; | |
8330 | ||
8331 | delayed_refs = &trans->transaction->delayed_refs; | |
8332 | spin_lock(&delayed_refs->lock); | |
8333 | if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord)) | |
8334 | kfree(qrecord); | |
8335 | spin_unlock(&delayed_refs->lock); | |
8336 | ||
8337 | return 0; | |
8338 | } | |
8339 | ||
1152651a MF |
8340 | static int account_leaf_items(struct btrfs_trans_handle *trans, |
8341 | struct btrfs_root *root, | |
8342 | struct extent_buffer *eb) | |
8343 | { | |
8344 | int nr = btrfs_header_nritems(eb); | |
82bd101b | 8345 | int i, extent_type, ret; |
1152651a MF |
8346 | struct btrfs_key key; |
8347 | struct btrfs_file_extent_item *fi; | |
8348 | u64 bytenr, num_bytes; | |
8349 | ||
82bd101b MF |
8350 | /* We can be called directly from walk_up_proc() */ |
8351 | if (!root->fs_info->quota_enabled) | |
8352 | return 0; | |
8353 | ||
1152651a MF |
8354 | for (i = 0; i < nr; i++) { |
8355 | btrfs_item_key_to_cpu(eb, &key, i); | |
8356 | ||
8357 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
8358 | continue; | |
8359 | ||
8360 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
8361 | /* filter out non qgroup-accountable extents */ | |
8362 | extent_type = btrfs_file_extent_type(eb, fi); | |
8363 | ||
8364 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
8365 | continue; | |
8366 | ||
8367 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
8368 | if (!bytenr) | |
8369 | continue; | |
8370 | ||
8371 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
82bd101b MF |
8372 | |
8373 | ret = record_one_subtree_extent(trans, root, bytenr, num_bytes); | |
8374 | if (ret) | |
8375 | return ret; | |
1152651a MF |
8376 | } |
8377 | return 0; | |
8378 | } | |
8379 | ||
8380 | /* | |
8381 | * Walk up the tree from the bottom, freeing leaves and any interior | |
8382 | * nodes which have had all slots visited. If a node (leaf or | |
8383 | * interior) is freed, the node above it will have it's slot | |
8384 | * incremented. The root node will never be freed. | |
8385 | * | |
8386 | * At the end of this function, we should have a path which has all | |
8387 | * slots incremented to the next position for a search. If we need to | |
8388 | * read a new node it will be NULL and the node above it will have the | |
8389 | * correct slot selected for a later read. | |
8390 | * | |
8391 | * If we increment the root nodes slot counter past the number of | |
8392 | * elements, 1 is returned to signal completion of the search. | |
8393 | */ | |
8394 | static int adjust_slots_upwards(struct btrfs_root *root, | |
8395 | struct btrfs_path *path, int root_level) | |
8396 | { | |
8397 | int level = 0; | |
8398 | int nr, slot; | |
8399 | struct extent_buffer *eb; | |
8400 | ||
8401 | if (root_level == 0) | |
8402 | return 1; | |
8403 | ||
8404 | while (level <= root_level) { | |
8405 | eb = path->nodes[level]; | |
8406 | nr = btrfs_header_nritems(eb); | |
8407 | path->slots[level]++; | |
8408 | slot = path->slots[level]; | |
8409 | if (slot >= nr || level == 0) { | |
8410 | /* | |
8411 | * Don't free the root - we will detect this | |
8412 | * condition after our loop and return a | |
8413 | * positive value for caller to stop walking the tree. | |
8414 | */ | |
8415 | if (level != root_level) { | |
8416 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8417 | path->locks[level] = 0; | |
8418 | ||
8419 | free_extent_buffer(eb); | |
8420 | path->nodes[level] = NULL; | |
8421 | path->slots[level] = 0; | |
8422 | } | |
8423 | } else { | |
8424 | /* | |
8425 | * We have a valid slot to walk back down | |
8426 | * from. Stop here so caller can process these | |
8427 | * new nodes. | |
8428 | */ | |
8429 | break; | |
8430 | } | |
8431 | ||
8432 | level++; | |
8433 | } | |
8434 | ||
8435 | eb = path->nodes[root_level]; | |
8436 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
8437 | return 1; | |
8438 | ||
8439 | return 0; | |
8440 | } | |
8441 | ||
8442 | /* | |
8443 | * root_eb is the subtree root and is locked before this function is called. | |
8444 | */ | |
8445 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
8446 | struct btrfs_root *root, | |
8447 | struct extent_buffer *root_eb, | |
8448 | u64 root_gen, | |
8449 | int root_level) | |
8450 | { | |
8451 | int ret = 0; | |
8452 | int level; | |
8453 | struct extent_buffer *eb = root_eb; | |
8454 | struct btrfs_path *path = NULL; | |
8455 | ||
8456 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
8457 | BUG_ON(root_eb == NULL); | |
8458 | ||
8459 | if (!root->fs_info->quota_enabled) | |
8460 | return 0; | |
8461 | ||
8462 | if (!extent_buffer_uptodate(root_eb)) { | |
8463 | ret = btrfs_read_buffer(root_eb, root_gen); | |
8464 | if (ret) | |
8465 | goto out; | |
8466 | } | |
8467 | ||
8468 | if (root_level == 0) { | |
8469 | ret = account_leaf_items(trans, root, root_eb); | |
8470 | goto out; | |
8471 | } | |
8472 | ||
8473 | path = btrfs_alloc_path(); | |
8474 | if (!path) | |
8475 | return -ENOMEM; | |
8476 | ||
8477 | /* | |
8478 | * Walk down the tree. Missing extent blocks are filled in as | |
8479 | * we go. Metadata is accounted every time we read a new | |
8480 | * extent block. | |
8481 | * | |
8482 | * When we reach a leaf, we account for file extent items in it, | |
8483 | * walk back up the tree (adjusting slot pointers as we go) | |
8484 | * and restart the search process. | |
8485 | */ | |
8486 | extent_buffer_get(root_eb); /* For path */ | |
8487 | path->nodes[root_level] = root_eb; | |
8488 | path->slots[root_level] = 0; | |
8489 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
8490 | walk_down: | |
8491 | level = root_level; | |
8492 | while (level >= 0) { | |
8493 | if (path->nodes[level] == NULL) { | |
1152651a MF |
8494 | int parent_slot; |
8495 | u64 child_gen; | |
8496 | u64 child_bytenr; | |
8497 | ||
8498 | /* We need to get child blockptr/gen from | |
8499 | * parent before we can read it. */ | |
8500 | eb = path->nodes[level + 1]; | |
8501 | parent_slot = path->slots[level + 1]; | |
8502 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
8503 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
8504 | ||
ce86cd59 | 8505 | eb = read_tree_block(root, child_bytenr, child_gen); |
64c043de LB |
8506 | if (IS_ERR(eb)) { |
8507 | ret = PTR_ERR(eb); | |
8508 | goto out; | |
8509 | } else if (!extent_buffer_uptodate(eb)) { | |
8635eda9 | 8510 | free_extent_buffer(eb); |
64c043de | 8511 | ret = -EIO; |
1152651a MF |
8512 | goto out; |
8513 | } | |
8514 | ||
8515 | path->nodes[level] = eb; | |
8516 | path->slots[level] = 0; | |
8517 | ||
8518 | btrfs_tree_read_lock(eb); | |
8519 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
8520 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
82bd101b MF |
8521 | |
8522 | ret = record_one_subtree_extent(trans, root, child_bytenr, | |
8523 | root->nodesize); | |
8524 | if (ret) | |
8525 | goto out; | |
1152651a MF |
8526 | } |
8527 | ||
8528 | if (level == 0) { | |
8529 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
8530 | if (ret) | |
8531 | goto out; | |
8532 | ||
8533 | /* Nonzero return here means we completed our search */ | |
8534 | ret = adjust_slots_upwards(root, path, root_level); | |
8535 | if (ret) | |
8536 | break; | |
8537 | ||
8538 | /* Restart search with new slots */ | |
8539 | goto walk_down; | |
8540 | } | |
8541 | ||
8542 | level--; | |
8543 | } | |
8544 | ||
8545 | ret = 0; | |
8546 | out: | |
8547 | btrfs_free_path(path); | |
8548 | ||
8549 | return ret; | |
8550 | } | |
8551 | ||
f82d02d9 | 8552 | /* |
2c016dc2 | 8553 | * helper to process tree block while walking down the tree. |
2c47e605 | 8554 | * |
2c47e605 YZ |
8555 | * when wc->stage == UPDATE_BACKREF, this function updates |
8556 | * back refs for pointers in the block. | |
8557 | * | |
8558 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8559 | */ |
2c47e605 | 8560 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8561 | struct btrfs_root *root, |
2c47e605 | 8562 | struct btrfs_path *path, |
94fcca9f | 8563 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8564 | { |
2c47e605 YZ |
8565 | int level = wc->level; |
8566 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8567 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8568 | int ret; |
8569 | ||
2c47e605 YZ |
8570 | if (wc->stage == UPDATE_BACKREF && |
8571 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8572 | return 1; | |
f82d02d9 | 8573 | |
2c47e605 YZ |
8574 | /* |
8575 | * when reference count of tree block is 1, it won't increase | |
8576 | * again. once full backref flag is set, we never clear it. | |
8577 | */ | |
94fcca9f YZ |
8578 | if (lookup_info && |
8579 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8580 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
8581 | BUG_ON(!path->locks[level]); |
8582 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8583 | eb->start, level, 1, |
2c47e605 YZ |
8584 | &wc->refs[level], |
8585 | &wc->flags[level]); | |
79787eaa JM |
8586 | BUG_ON(ret == -ENOMEM); |
8587 | if (ret) | |
8588 | return ret; | |
2c47e605 YZ |
8589 | BUG_ON(wc->refs[level] == 0); |
8590 | } | |
5d4f98a2 | 8591 | |
2c47e605 YZ |
8592 | if (wc->stage == DROP_REFERENCE) { |
8593 | if (wc->refs[level] > 1) | |
8594 | return 1; | |
f82d02d9 | 8595 | |
2c47e605 | 8596 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8597 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8598 | path->locks[level] = 0; |
8599 | } | |
8600 | return 0; | |
8601 | } | |
f82d02d9 | 8602 | |
2c47e605 YZ |
8603 | /* wc->stage == UPDATE_BACKREF */ |
8604 | if (!(wc->flags[level] & flag)) { | |
8605 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8606 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8607 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8608 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8609 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 8610 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
8611 | eb->len, flag, |
8612 | btrfs_header_level(eb), 0); | |
79787eaa | 8613 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8614 | wc->flags[level] |= flag; |
8615 | } | |
8616 | ||
8617 | /* | |
8618 | * the block is shared by multiple trees, so it's not good to | |
8619 | * keep the tree lock | |
8620 | */ | |
8621 | if (path->locks[level] && level > 0) { | |
bd681513 | 8622 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8623 | path->locks[level] = 0; |
8624 | } | |
8625 | return 0; | |
8626 | } | |
8627 | ||
1c4850e2 | 8628 | /* |
2c016dc2 | 8629 | * helper to process tree block pointer. |
1c4850e2 YZ |
8630 | * |
8631 | * when wc->stage == DROP_REFERENCE, this function checks | |
8632 | * reference count of the block pointed to. if the block | |
8633 | * is shared and we need update back refs for the subtree | |
8634 | * rooted at the block, this function changes wc->stage to | |
8635 | * UPDATE_BACKREF. if the block is shared and there is no | |
8636 | * need to update back, this function drops the reference | |
8637 | * to the block. | |
8638 | * | |
8639 | * NOTE: return value 1 means we should stop walking down. | |
8640 | */ | |
8641 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8642 | struct btrfs_root *root, | |
8643 | struct btrfs_path *path, | |
94fcca9f | 8644 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
8645 | { |
8646 | u64 bytenr; | |
8647 | u64 generation; | |
8648 | u64 parent; | |
8649 | u32 blocksize; | |
8650 | struct btrfs_key key; | |
8651 | struct extent_buffer *next; | |
8652 | int level = wc->level; | |
8653 | int reada = 0; | |
8654 | int ret = 0; | |
1152651a | 8655 | bool need_account = false; |
1c4850e2 YZ |
8656 | |
8657 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8658 | path->slots[level]); | |
8659 | /* | |
8660 | * if the lower level block was created before the snapshot | |
8661 | * was created, we know there is no need to update back refs | |
8662 | * for the subtree | |
8663 | */ | |
8664 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8665 | generation <= root->root_key.offset) { |
8666 | *lookup_info = 1; | |
1c4850e2 | 8667 | return 1; |
94fcca9f | 8668 | } |
1c4850e2 YZ |
8669 | |
8670 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
707e8a07 | 8671 | blocksize = root->nodesize; |
1c4850e2 | 8672 | |
01d58472 | 8673 | next = btrfs_find_tree_block(root->fs_info, bytenr); |
1c4850e2 | 8674 | if (!next) { |
a83fffb7 | 8675 | next = btrfs_find_create_tree_block(root, bytenr); |
c871b0f2 LB |
8676 | if (IS_ERR(next)) |
8677 | return PTR_ERR(next); | |
8678 | ||
b2aaaa3b JB |
8679 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8680 | level - 1); | |
1c4850e2 YZ |
8681 | reada = 1; |
8682 | } | |
8683 | btrfs_tree_lock(next); | |
8684 | btrfs_set_lock_blocking(next); | |
8685 | ||
3173a18f | 8686 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
8687 | &wc->refs[level - 1], |
8688 | &wc->flags[level - 1]); | |
79787eaa JM |
8689 | if (ret < 0) { |
8690 | btrfs_tree_unlock(next); | |
8691 | return ret; | |
8692 | } | |
8693 | ||
c2cf52eb SK |
8694 | if (unlikely(wc->refs[level - 1] == 0)) { |
8695 | btrfs_err(root->fs_info, "Missing references."); | |
8696 | BUG(); | |
8697 | } | |
94fcca9f | 8698 | *lookup_info = 0; |
1c4850e2 | 8699 | |
94fcca9f | 8700 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8701 | if (wc->refs[level - 1] > 1) { |
1152651a | 8702 | need_account = true; |
94fcca9f YZ |
8703 | if (level == 1 && |
8704 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8705 | goto skip; | |
8706 | ||
1c4850e2 YZ |
8707 | if (!wc->update_ref || |
8708 | generation <= root->root_key.offset) | |
8709 | goto skip; | |
8710 | ||
8711 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8712 | path->slots[level]); | |
8713 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8714 | if (ret < 0) | |
8715 | goto skip; | |
8716 | ||
8717 | wc->stage = UPDATE_BACKREF; | |
8718 | wc->shared_level = level - 1; | |
8719 | } | |
94fcca9f YZ |
8720 | } else { |
8721 | if (level == 1 && | |
8722 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8723 | goto skip; | |
1c4850e2 YZ |
8724 | } |
8725 | ||
b9fab919 | 8726 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8727 | btrfs_tree_unlock(next); |
8728 | free_extent_buffer(next); | |
8729 | next = NULL; | |
94fcca9f | 8730 | *lookup_info = 1; |
1c4850e2 YZ |
8731 | } |
8732 | ||
8733 | if (!next) { | |
8734 | if (reada && level == 1) | |
8735 | reada_walk_down(trans, root, wc, path); | |
ce86cd59 | 8736 | next = read_tree_block(root, bytenr, generation); |
64c043de LB |
8737 | if (IS_ERR(next)) { |
8738 | return PTR_ERR(next); | |
8739 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8740 | free_extent_buffer(next); |
97d9a8a4 | 8741 | return -EIO; |
416bc658 | 8742 | } |
1c4850e2 YZ |
8743 | btrfs_tree_lock(next); |
8744 | btrfs_set_lock_blocking(next); | |
8745 | } | |
8746 | ||
8747 | level--; | |
8748 | BUG_ON(level != btrfs_header_level(next)); | |
8749 | path->nodes[level] = next; | |
8750 | path->slots[level] = 0; | |
bd681513 | 8751 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8752 | wc->level = level; |
8753 | if (wc->level == 1) | |
8754 | wc->reada_slot = 0; | |
8755 | return 0; | |
8756 | skip: | |
8757 | wc->refs[level - 1] = 0; | |
8758 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8759 | if (wc->stage == DROP_REFERENCE) { |
8760 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8761 | parent = path->nodes[level]->start; | |
8762 | } else { | |
8763 | BUG_ON(root->root_key.objectid != | |
8764 | btrfs_header_owner(path->nodes[level])); | |
8765 | parent = 0; | |
8766 | } | |
1c4850e2 | 8767 | |
1152651a MF |
8768 | if (need_account) { |
8769 | ret = account_shared_subtree(trans, root, next, | |
8770 | generation, level - 1); | |
8771 | if (ret) { | |
94647322 DS |
8772 | btrfs_err_rl(root->fs_info, |
8773 | "Error " | |
1152651a | 8774 | "%d accounting shared subtree. Quota " |
94647322 DS |
8775 | "is out of sync, rescan required.", |
8776 | ret); | |
1152651a MF |
8777 | } |
8778 | } | |
94fcca9f | 8779 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
b06c4bf5 | 8780 | root->root_key.objectid, level - 1, 0); |
79787eaa | 8781 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 8782 | } |
1c4850e2 YZ |
8783 | btrfs_tree_unlock(next); |
8784 | free_extent_buffer(next); | |
94fcca9f | 8785 | *lookup_info = 1; |
1c4850e2 YZ |
8786 | return 1; |
8787 | } | |
8788 | ||
2c47e605 | 8789 | /* |
2c016dc2 | 8790 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8791 | * |
8792 | * when wc->stage == DROP_REFERENCE, this function drops | |
8793 | * reference count on the block. | |
8794 | * | |
8795 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8796 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8797 | * to UPDATE_BACKREF previously while processing the block. | |
8798 | * | |
8799 | * NOTE: return value 1 means we should stop walking up. | |
8800 | */ | |
8801 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8802 | struct btrfs_root *root, | |
8803 | struct btrfs_path *path, | |
8804 | struct walk_control *wc) | |
8805 | { | |
f0486c68 | 8806 | int ret; |
2c47e605 YZ |
8807 | int level = wc->level; |
8808 | struct extent_buffer *eb = path->nodes[level]; | |
8809 | u64 parent = 0; | |
8810 | ||
8811 | if (wc->stage == UPDATE_BACKREF) { | |
8812 | BUG_ON(wc->shared_level < level); | |
8813 | if (level < wc->shared_level) | |
8814 | goto out; | |
8815 | ||
2c47e605 YZ |
8816 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8817 | if (ret > 0) | |
8818 | wc->update_ref = 0; | |
8819 | ||
8820 | wc->stage = DROP_REFERENCE; | |
8821 | wc->shared_level = -1; | |
8822 | path->slots[level] = 0; | |
8823 | ||
8824 | /* | |
8825 | * check reference count again if the block isn't locked. | |
8826 | * we should start walking down the tree again if reference | |
8827 | * count is one. | |
8828 | */ | |
8829 | if (!path->locks[level]) { | |
8830 | BUG_ON(level == 0); | |
8831 | btrfs_tree_lock(eb); | |
8832 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8833 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8834 | |
8835 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8836 | eb->start, level, 1, |
2c47e605 YZ |
8837 | &wc->refs[level], |
8838 | &wc->flags[level]); | |
79787eaa JM |
8839 | if (ret < 0) { |
8840 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8841 | path->locks[level] = 0; |
79787eaa JM |
8842 | return ret; |
8843 | } | |
2c47e605 YZ |
8844 | BUG_ON(wc->refs[level] == 0); |
8845 | if (wc->refs[level] == 1) { | |
bd681513 | 8846 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8847 | path->locks[level] = 0; |
2c47e605 YZ |
8848 | return 1; |
8849 | } | |
f82d02d9 | 8850 | } |
2c47e605 | 8851 | } |
f82d02d9 | 8852 | |
2c47e605 YZ |
8853 | /* wc->stage == DROP_REFERENCE */ |
8854 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8855 | |
2c47e605 YZ |
8856 | if (wc->refs[level] == 1) { |
8857 | if (level == 0) { | |
8858 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8859 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8860 | else |
e339a6b0 | 8861 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8862 | BUG_ON(ret); /* -ENOMEM */ |
1152651a MF |
8863 | ret = account_leaf_items(trans, root, eb); |
8864 | if (ret) { | |
94647322 DS |
8865 | btrfs_err_rl(root->fs_info, |
8866 | "error " | |
1152651a | 8867 | "%d accounting leaf items. Quota " |
94647322 DS |
8868 | "is out of sync, rescan required.", |
8869 | ret); | |
1152651a | 8870 | } |
2c47e605 YZ |
8871 | } |
8872 | /* make block locked assertion in clean_tree_block happy */ | |
8873 | if (!path->locks[level] && | |
8874 | btrfs_header_generation(eb) == trans->transid) { | |
8875 | btrfs_tree_lock(eb); | |
8876 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8877 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8878 | } |
01d58472 | 8879 | clean_tree_block(trans, root->fs_info, eb); |
2c47e605 YZ |
8880 | } |
8881 | ||
8882 | if (eb == root->node) { | |
8883 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8884 | parent = eb->start; | |
8885 | else | |
8886 | BUG_ON(root->root_key.objectid != | |
8887 | btrfs_header_owner(eb)); | |
8888 | } else { | |
8889 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8890 | parent = path->nodes[level + 1]->start; | |
8891 | else | |
8892 | BUG_ON(root->root_key.objectid != | |
8893 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8894 | } |
f82d02d9 | 8895 | |
5581a51a | 8896 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8897 | out: |
8898 | wc->refs[level] = 0; | |
8899 | wc->flags[level] = 0; | |
f0486c68 | 8900 | return 0; |
2c47e605 YZ |
8901 | } |
8902 | ||
8903 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8904 | struct btrfs_root *root, | |
8905 | struct btrfs_path *path, | |
8906 | struct walk_control *wc) | |
8907 | { | |
2c47e605 | 8908 | int level = wc->level; |
94fcca9f | 8909 | int lookup_info = 1; |
2c47e605 YZ |
8910 | int ret; |
8911 | ||
8912 | while (level >= 0) { | |
94fcca9f | 8913 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8914 | if (ret > 0) |
8915 | break; | |
8916 | ||
8917 | if (level == 0) | |
8918 | break; | |
8919 | ||
7a7965f8 YZ |
8920 | if (path->slots[level] >= |
8921 | btrfs_header_nritems(path->nodes[level])) | |
8922 | break; | |
8923 | ||
94fcca9f | 8924 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8925 | if (ret > 0) { |
8926 | path->slots[level]++; | |
8927 | continue; | |
90d2c51d MX |
8928 | } else if (ret < 0) |
8929 | return ret; | |
1c4850e2 | 8930 | level = wc->level; |
f82d02d9 | 8931 | } |
f82d02d9 YZ |
8932 | return 0; |
8933 | } | |
8934 | ||
d397712b | 8935 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8936 | struct btrfs_root *root, |
f82d02d9 | 8937 | struct btrfs_path *path, |
2c47e605 | 8938 | struct walk_control *wc, int max_level) |
20524f02 | 8939 | { |
2c47e605 | 8940 | int level = wc->level; |
20524f02 | 8941 | int ret; |
9f3a7427 | 8942 | |
2c47e605 YZ |
8943 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8944 | while (level < max_level && path->nodes[level]) { | |
8945 | wc->level = level; | |
8946 | if (path->slots[level] + 1 < | |
8947 | btrfs_header_nritems(path->nodes[level])) { | |
8948 | path->slots[level]++; | |
20524f02 CM |
8949 | return 0; |
8950 | } else { | |
2c47e605 YZ |
8951 | ret = walk_up_proc(trans, root, path, wc); |
8952 | if (ret > 0) | |
8953 | return 0; | |
bd56b302 | 8954 | |
2c47e605 | 8955 | if (path->locks[level]) { |
bd681513 CM |
8956 | btrfs_tree_unlock_rw(path->nodes[level], |
8957 | path->locks[level]); | |
2c47e605 | 8958 | path->locks[level] = 0; |
f82d02d9 | 8959 | } |
2c47e605 YZ |
8960 | free_extent_buffer(path->nodes[level]); |
8961 | path->nodes[level] = NULL; | |
8962 | level++; | |
20524f02 CM |
8963 | } |
8964 | } | |
8965 | return 1; | |
8966 | } | |
8967 | ||
9aca1d51 | 8968 | /* |
2c47e605 YZ |
8969 | * drop a subvolume tree. |
8970 | * | |
8971 | * this function traverses the tree freeing any blocks that only | |
8972 | * referenced by the tree. | |
8973 | * | |
8974 | * when a shared tree block is found. this function decreases its | |
8975 | * reference count by one. if update_ref is true, this function | |
8976 | * also make sure backrefs for the shared block and all lower level | |
8977 | * blocks are properly updated. | |
9d1a2a3a DS |
8978 | * |
8979 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8980 | */ |
2c536799 | 8981 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8982 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8983 | int for_reloc) | |
20524f02 | 8984 | { |
5caf2a00 | 8985 | struct btrfs_path *path; |
2c47e605 YZ |
8986 | struct btrfs_trans_handle *trans; |
8987 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 8988 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8989 | struct walk_control *wc; |
8990 | struct btrfs_key key; | |
8991 | int err = 0; | |
8992 | int ret; | |
8993 | int level; | |
d29a9f62 | 8994 | bool root_dropped = false; |
20524f02 | 8995 | |
1152651a MF |
8996 | btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid); |
8997 | ||
5caf2a00 | 8998 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8999 | if (!path) { |
9000 | err = -ENOMEM; | |
9001 | goto out; | |
9002 | } | |
20524f02 | 9003 | |
2c47e605 | 9004 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9005 | if (!wc) { |
9006 | btrfs_free_path(path); | |
cb1b69f4 TI |
9007 | err = -ENOMEM; |
9008 | goto out; | |
38a1a919 | 9009 | } |
2c47e605 | 9010 | |
a22285a6 | 9011 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9012 | if (IS_ERR(trans)) { |
9013 | err = PTR_ERR(trans); | |
9014 | goto out_free; | |
9015 | } | |
98d5dc13 | 9016 | |
3fd0a558 YZ |
9017 | if (block_rsv) |
9018 | trans->block_rsv = block_rsv; | |
2c47e605 | 9019 | |
9f3a7427 | 9020 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9021 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9022 | path->nodes[level] = btrfs_lock_root_node(root); |
9023 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9024 | path->slots[level] = 0; |
bd681513 | 9025 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9026 | memset(&wc->update_progress, 0, |
9027 | sizeof(wc->update_progress)); | |
9f3a7427 | 9028 | } else { |
9f3a7427 | 9029 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9030 | memcpy(&wc->update_progress, &key, |
9031 | sizeof(wc->update_progress)); | |
9032 | ||
6702ed49 | 9033 | level = root_item->drop_level; |
2c47e605 | 9034 | BUG_ON(level == 0); |
6702ed49 | 9035 | path->lowest_level = level; |
2c47e605 YZ |
9036 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9037 | path->lowest_level = 0; | |
9038 | if (ret < 0) { | |
9039 | err = ret; | |
79787eaa | 9040 | goto out_end_trans; |
9f3a7427 | 9041 | } |
1c4850e2 | 9042 | WARN_ON(ret > 0); |
2c47e605 | 9043 | |
7d9eb12c CM |
9044 | /* |
9045 | * unlock our path, this is safe because only this | |
9046 | * function is allowed to delete this snapshot | |
9047 | */ | |
5d4f98a2 | 9048 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9049 | |
9050 | level = btrfs_header_level(root->node); | |
9051 | while (1) { | |
9052 | btrfs_tree_lock(path->nodes[level]); | |
9053 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9054 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9055 | |
9056 | ret = btrfs_lookup_extent_info(trans, root, | |
9057 | path->nodes[level]->start, | |
3173a18f | 9058 | level, 1, &wc->refs[level], |
2c47e605 | 9059 | &wc->flags[level]); |
79787eaa JM |
9060 | if (ret < 0) { |
9061 | err = ret; | |
9062 | goto out_end_trans; | |
9063 | } | |
2c47e605 YZ |
9064 | BUG_ON(wc->refs[level] == 0); |
9065 | ||
9066 | if (level == root_item->drop_level) | |
9067 | break; | |
9068 | ||
9069 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9070 | path->locks[level] = 0; |
2c47e605 YZ |
9071 | WARN_ON(wc->refs[level] != 1); |
9072 | level--; | |
9073 | } | |
9f3a7427 | 9074 | } |
2c47e605 YZ |
9075 | |
9076 | wc->level = level; | |
9077 | wc->shared_level = -1; | |
9078 | wc->stage = DROP_REFERENCE; | |
9079 | wc->update_ref = update_ref; | |
9080 | wc->keep_locks = 0; | |
66d7e7f0 | 9081 | wc->for_reloc = for_reloc; |
1c4850e2 | 9082 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 9083 | |
d397712b | 9084 | while (1) { |
9d1a2a3a | 9085 | |
2c47e605 YZ |
9086 | ret = walk_down_tree(trans, root, path, wc); |
9087 | if (ret < 0) { | |
9088 | err = ret; | |
20524f02 | 9089 | break; |
2c47e605 | 9090 | } |
9aca1d51 | 9091 | |
2c47e605 YZ |
9092 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9093 | if (ret < 0) { | |
9094 | err = ret; | |
20524f02 | 9095 | break; |
2c47e605 YZ |
9096 | } |
9097 | ||
9098 | if (ret > 0) { | |
9099 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9100 | break; |
9101 | } | |
2c47e605 YZ |
9102 | |
9103 | if (wc->stage == DROP_REFERENCE) { | |
9104 | level = wc->level; | |
9105 | btrfs_node_key(path->nodes[level], | |
9106 | &root_item->drop_progress, | |
9107 | path->slots[level]); | |
9108 | root_item->drop_level = level; | |
9109 | } | |
9110 | ||
9111 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
9112 | if (btrfs_should_end_transaction(trans, tree_root) || |
9113 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
9114 | ret = btrfs_update_root(trans, tree_root, |
9115 | &root->root_key, | |
9116 | root_item); | |
79787eaa JM |
9117 | if (ret) { |
9118 | btrfs_abort_transaction(trans, tree_root, ret); | |
9119 | err = ret; | |
9120 | goto out_end_trans; | |
9121 | } | |
2c47e605 | 9122 | |
3fd0a558 | 9123 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 | 9124 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
efe120a0 | 9125 | pr_debug("BTRFS: drop snapshot early exit\n"); |
3c8f2422 JB |
9126 | err = -EAGAIN; |
9127 | goto out_free; | |
9128 | } | |
9129 | ||
a22285a6 | 9130 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9131 | if (IS_ERR(trans)) { |
9132 | err = PTR_ERR(trans); | |
9133 | goto out_free; | |
9134 | } | |
3fd0a558 YZ |
9135 | if (block_rsv) |
9136 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9137 | } |
20524f02 | 9138 | } |
b3b4aa74 | 9139 | btrfs_release_path(path); |
79787eaa JM |
9140 | if (err) |
9141 | goto out_end_trans; | |
2c47e605 YZ |
9142 | |
9143 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
9144 | if (ret) { |
9145 | btrfs_abort_transaction(trans, tree_root, ret); | |
9146 | goto out_end_trans; | |
9147 | } | |
2c47e605 | 9148 | |
76dda93c | 9149 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9150 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9151 | NULL, NULL); | |
79787eaa JM |
9152 | if (ret < 0) { |
9153 | btrfs_abort_transaction(trans, tree_root, ret); | |
9154 | err = ret; | |
9155 | goto out_end_trans; | |
9156 | } else if (ret > 0) { | |
84cd948c JB |
9157 | /* if we fail to delete the orphan item this time |
9158 | * around, it'll get picked up the next time. | |
9159 | * | |
9160 | * The most common failure here is just -ENOENT. | |
9161 | */ | |
9162 | btrfs_del_orphan_item(trans, tree_root, | |
9163 | root->root_key.objectid); | |
76dda93c YZ |
9164 | } |
9165 | } | |
9166 | ||
27cdeb70 | 9167 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9168 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9169 | } else { |
9170 | free_extent_buffer(root->node); | |
9171 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9172 | btrfs_put_fs_root(root); |
76dda93c | 9173 | } |
d29a9f62 | 9174 | root_dropped = true; |
79787eaa | 9175 | out_end_trans: |
3fd0a558 | 9176 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 9177 | out_free: |
2c47e605 | 9178 | kfree(wc); |
5caf2a00 | 9179 | btrfs_free_path(path); |
cb1b69f4 | 9180 | out: |
d29a9f62 JB |
9181 | /* |
9182 | * So if we need to stop dropping the snapshot for whatever reason we | |
9183 | * need to make sure to add it back to the dead root list so that we | |
9184 | * keep trying to do the work later. This also cleans up roots if we | |
9185 | * don't have it in the radix (like when we recover after a power fail | |
9186 | * or unmount) so we don't leak memory. | |
9187 | */ | |
b37b39cd | 9188 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9189 | btrfs_add_dead_root(root); |
90515e7f | 9190 | if (err && err != -EAGAIN) |
34d97007 | 9191 | btrfs_handle_fs_error(root->fs_info, err, NULL); |
2c536799 | 9192 | return err; |
20524f02 | 9193 | } |
9078a3e1 | 9194 | |
2c47e605 YZ |
9195 | /* |
9196 | * drop subtree rooted at tree block 'node'. | |
9197 | * | |
9198 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9199 | * only used by relocation code |
2c47e605 | 9200 | */ |
f82d02d9 YZ |
9201 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9202 | struct btrfs_root *root, | |
9203 | struct extent_buffer *node, | |
9204 | struct extent_buffer *parent) | |
9205 | { | |
9206 | struct btrfs_path *path; | |
2c47e605 | 9207 | struct walk_control *wc; |
f82d02d9 YZ |
9208 | int level; |
9209 | int parent_level; | |
9210 | int ret = 0; | |
9211 | int wret; | |
9212 | ||
2c47e605 YZ |
9213 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9214 | ||
f82d02d9 | 9215 | path = btrfs_alloc_path(); |
db5b493a TI |
9216 | if (!path) |
9217 | return -ENOMEM; | |
f82d02d9 | 9218 | |
2c47e605 | 9219 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9220 | if (!wc) { |
9221 | btrfs_free_path(path); | |
9222 | return -ENOMEM; | |
9223 | } | |
2c47e605 | 9224 | |
b9447ef8 | 9225 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9226 | parent_level = btrfs_header_level(parent); |
9227 | extent_buffer_get(parent); | |
9228 | path->nodes[parent_level] = parent; | |
9229 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9230 | ||
b9447ef8 | 9231 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9232 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9233 | path->nodes[level] = node; |
9234 | path->slots[level] = 0; | |
bd681513 | 9235 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9236 | |
9237 | wc->refs[parent_level] = 1; | |
9238 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9239 | wc->level = level; | |
9240 | wc->shared_level = -1; | |
9241 | wc->stage = DROP_REFERENCE; | |
9242 | wc->update_ref = 0; | |
9243 | wc->keep_locks = 1; | |
66d7e7f0 | 9244 | wc->for_reloc = 1; |
1c4850e2 | 9245 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
9246 | |
9247 | while (1) { | |
2c47e605 YZ |
9248 | wret = walk_down_tree(trans, root, path, wc); |
9249 | if (wret < 0) { | |
f82d02d9 | 9250 | ret = wret; |
f82d02d9 | 9251 | break; |
2c47e605 | 9252 | } |
f82d02d9 | 9253 | |
2c47e605 | 9254 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9255 | if (wret < 0) |
9256 | ret = wret; | |
9257 | if (wret != 0) | |
9258 | break; | |
9259 | } | |
9260 | ||
2c47e605 | 9261 | kfree(wc); |
f82d02d9 YZ |
9262 | btrfs_free_path(path); |
9263 | return ret; | |
9264 | } | |
9265 | ||
ec44a35c CM |
9266 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
9267 | { | |
9268 | u64 num_devices; | |
fc67c450 | 9269 | u64 stripped; |
e4d8ec0f | 9270 | |
fc67c450 ID |
9271 | /* |
9272 | * if restripe for this chunk_type is on pick target profile and | |
9273 | * return, otherwise do the usual balance | |
9274 | */ | |
9275 | stripped = get_restripe_target(root->fs_info, flags); | |
9276 | if (stripped) | |
9277 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9278 | |
95669976 | 9279 | num_devices = root->fs_info->fs_devices->rw_devices; |
cd02dca5 | 9280 | |
fc67c450 | 9281 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9282 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9283 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9284 | ||
ec44a35c CM |
9285 | if (num_devices == 1) { |
9286 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9287 | stripped = flags & ~stripped; | |
9288 | ||
9289 | /* turn raid0 into single device chunks */ | |
9290 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9291 | return stripped; | |
9292 | ||
9293 | /* turn mirroring into duplication */ | |
9294 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9295 | BTRFS_BLOCK_GROUP_RAID10)) | |
9296 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9297 | } else { |
9298 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9299 | if (flags & stripped) |
9300 | return flags; | |
9301 | ||
9302 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9303 | stripped = flags & ~stripped; | |
9304 | ||
9305 | /* switch duplicated blocks with raid1 */ | |
9306 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9307 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9308 | ||
e3176ca2 | 9309 | /* this is drive concat, leave it alone */ |
ec44a35c | 9310 | } |
e3176ca2 | 9311 | |
ec44a35c CM |
9312 | return flags; |
9313 | } | |
9314 | ||
868f401a | 9315 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9316 | { |
f0486c68 YZ |
9317 | struct btrfs_space_info *sinfo = cache->space_info; |
9318 | u64 num_bytes; | |
199c36ea | 9319 | u64 min_allocable_bytes; |
f0486c68 | 9320 | int ret = -ENOSPC; |
0ef3e66b | 9321 | |
199c36ea MX |
9322 | /* |
9323 | * We need some metadata space and system metadata space for | |
9324 | * allocating chunks in some corner cases until we force to set | |
9325 | * it to be readonly. | |
9326 | */ | |
9327 | if ((sinfo->flags & | |
9328 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9329 | !force) | |
ee22184b | 9330 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9331 | else |
9332 | min_allocable_bytes = 0; | |
9333 | ||
f0486c68 YZ |
9334 | spin_lock(&sinfo->lock); |
9335 | spin_lock(&cache->lock); | |
61cfea9b W |
9336 | |
9337 | if (cache->ro) { | |
868f401a | 9338 | cache->ro++; |
61cfea9b W |
9339 | ret = 0; |
9340 | goto out; | |
9341 | } | |
9342 | ||
f0486c68 YZ |
9343 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9344 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9345 | ||
9346 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
9347 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
9348 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 9349 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9350 | cache->ro++; |
633c0aad | 9351 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9352 | ret = 0; |
9353 | } | |
61cfea9b | 9354 | out: |
f0486c68 YZ |
9355 | spin_unlock(&cache->lock); |
9356 | spin_unlock(&sinfo->lock); | |
9357 | return ret; | |
9358 | } | |
7d9eb12c | 9359 | |
868f401a | 9360 | int btrfs_inc_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9361 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9362 | |
f0486c68 YZ |
9363 | { |
9364 | struct btrfs_trans_handle *trans; | |
9365 | u64 alloc_flags; | |
9366 | int ret; | |
7d9eb12c | 9367 | |
1bbc621e | 9368 | again: |
ff5714cc | 9369 | trans = btrfs_join_transaction(root); |
79787eaa JM |
9370 | if (IS_ERR(trans)) |
9371 | return PTR_ERR(trans); | |
5d4f98a2 | 9372 | |
1bbc621e CM |
9373 | /* |
9374 | * we're not allowed to set block groups readonly after the dirty | |
9375 | * block groups cache has started writing. If it already started, | |
9376 | * back off and let this transaction commit | |
9377 | */ | |
9378 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c | 9379 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9380 | u64 transid = trans->transid; |
9381 | ||
9382 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9383 | btrfs_end_transaction(trans, root); | |
9384 | ||
9385 | ret = btrfs_wait_for_commit(root, transid); | |
9386 | if (ret) | |
9387 | return ret; | |
9388 | goto again; | |
9389 | } | |
9390 | ||
153c35b6 CM |
9391 | /* |
9392 | * if we are changing raid levels, try to allocate a corresponding | |
9393 | * block group with the new raid level. | |
9394 | */ | |
9395 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9396 | if (alloc_flags != cache->flags) { | |
9397 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9398 | CHUNK_ALLOC_FORCE); | |
9399 | /* | |
9400 | * ENOSPC is allowed here, we may have enough space | |
9401 | * already allocated at the new raid level to | |
9402 | * carry on | |
9403 | */ | |
9404 | if (ret == -ENOSPC) | |
9405 | ret = 0; | |
9406 | if (ret < 0) | |
9407 | goto out; | |
9408 | } | |
1bbc621e | 9409 | |
868f401a | 9410 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9411 | if (!ret) |
9412 | goto out; | |
9413 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 9414 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9415 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9416 | if (ret < 0) |
9417 | goto out; | |
868f401a | 9418 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9419 | out: |
2f081088 SL |
9420 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
9421 | alloc_flags = update_block_group_flags(root, cache->flags); | |
a9629596 | 9422 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 9423 | check_system_chunk(trans, root, alloc_flags); |
a9629596 | 9424 | unlock_chunks(root->fs_info->chunk_root); |
2f081088 | 9425 | } |
1bbc621e | 9426 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
2f081088 | 9427 | |
f0486c68 YZ |
9428 | btrfs_end_transaction(trans, root); |
9429 | return ret; | |
9430 | } | |
5d4f98a2 | 9431 | |
c87f08ca CM |
9432 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
9433 | struct btrfs_root *root, u64 type) | |
9434 | { | |
9435 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 9436 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9437 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9438 | } |
9439 | ||
6d07bcec MX |
9440 | /* |
9441 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9442 | * space_info. takes mirrors into account. |
6d07bcec | 9443 | */ |
633c0aad | 9444 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9445 | { |
9446 | struct btrfs_block_group_cache *block_group; | |
9447 | u64 free_bytes = 0; | |
9448 | int factor; | |
9449 | ||
01327610 | 9450 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9451 | if (list_empty(&sinfo->ro_bgs)) |
9452 | return 0; | |
9453 | ||
9454 | spin_lock(&sinfo->lock); | |
9455 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9456 | spin_lock(&block_group->lock); |
9457 | ||
9458 | if (!block_group->ro) { | |
9459 | spin_unlock(&block_group->lock); | |
9460 | continue; | |
9461 | } | |
9462 | ||
9463 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9464 | BTRFS_BLOCK_GROUP_RAID10 | | |
9465 | BTRFS_BLOCK_GROUP_DUP)) | |
9466 | factor = 2; | |
9467 | else | |
9468 | factor = 1; | |
9469 | ||
9470 | free_bytes += (block_group->key.offset - | |
9471 | btrfs_block_group_used(&block_group->item)) * | |
9472 | factor; | |
9473 | ||
9474 | spin_unlock(&block_group->lock); | |
9475 | } | |
6d07bcec MX |
9476 | spin_unlock(&sinfo->lock); |
9477 | ||
9478 | return free_bytes; | |
9479 | } | |
9480 | ||
868f401a | 9481 | void btrfs_dec_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9482 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9483 | { |
f0486c68 YZ |
9484 | struct btrfs_space_info *sinfo = cache->space_info; |
9485 | u64 num_bytes; | |
9486 | ||
9487 | BUG_ON(!cache->ro); | |
9488 | ||
9489 | spin_lock(&sinfo->lock); | |
9490 | spin_lock(&cache->lock); | |
868f401a Z |
9491 | if (!--cache->ro) { |
9492 | num_bytes = cache->key.offset - cache->reserved - | |
9493 | cache->pinned - cache->bytes_super - | |
9494 | btrfs_block_group_used(&cache->item); | |
9495 | sinfo->bytes_readonly -= num_bytes; | |
9496 | list_del_init(&cache->ro_list); | |
9497 | } | |
f0486c68 YZ |
9498 | spin_unlock(&cache->lock); |
9499 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9500 | } |
9501 | ||
ba1bf481 JB |
9502 | /* |
9503 | * checks to see if its even possible to relocate this block group. | |
9504 | * | |
9505 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9506 | * ok to go ahead and try. | |
9507 | */ | |
9508 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 9509 | { |
ba1bf481 JB |
9510 | struct btrfs_block_group_cache *block_group; |
9511 | struct btrfs_space_info *space_info; | |
9512 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
9513 | struct btrfs_device *device; | |
6df9a95e | 9514 | struct btrfs_trans_handle *trans; |
cdcb725c | 9515 | u64 min_free; |
6719db6a JB |
9516 | u64 dev_min = 1; |
9517 | u64 dev_nr = 0; | |
4a5e98f5 | 9518 | u64 target; |
0305bc27 | 9519 | int debug; |
cdcb725c | 9520 | int index; |
ba1bf481 JB |
9521 | int full = 0; |
9522 | int ret = 0; | |
1a40e23b | 9523 | |
0305bc27 QW |
9524 | debug = btrfs_test_opt(root, ENOSPC_DEBUG); |
9525 | ||
ba1bf481 | 9526 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 9527 | |
ba1bf481 | 9528 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9529 | if (!block_group) { |
9530 | if (debug) | |
9531 | btrfs_warn(root->fs_info, | |
9532 | "can't find block group for bytenr %llu", | |
9533 | bytenr); | |
ba1bf481 | 9534 | return -1; |
0305bc27 | 9535 | } |
1a40e23b | 9536 | |
cdcb725c | 9537 | min_free = btrfs_block_group_used(&block_group->item); |
9538 | ||
ba1bf481 | 9539 | /* no bytes used, we're good */ |
cdcb725c | 9540 | if (!min_free) |
1a40e23b ZY |
9541 | goto out; |
9542 | ||
ba1bf481 JB |
9543 | space_info = block_group->space_info; |
9544 | spin_lock(&space_info->lock); | |
17d217fe | 9545 | |
ba1bf481 | 9546 | full = space_info->full; |
17d217fe | 9547 | |
ba1bf481 JB |
9548 | /* |
9549 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9550 | * relocate it unless we're able to allocate a new chunk below. |
9551 | * | |
9552 | * Otherwise, we need to make sure we have room in the space to handle | |
9553 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9554 | */ |
7ce618db | 9555 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 9556 | (space_info->bytes_used + space_info->bytes_reserved + |
9557 | space_info->bytes_pinned + space_info->bytes_readonly + | |
9558 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
9559 | spin_unlock(&space_info->lock); |
9560 | goto out; | |
17d217fe | 9561 | } |
ba1bf481 | 9562 | spin_unlock(&space_info->lock); |
ea8c2819 | 9563 | |
ba1bf481 JB |
9564 | /* |
9565 | * ok we don't have enough space, but maybe we have free space on our | |
9566 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9567 | * alloc devices and guess if we have enough space. if this block |
9568 | * group is going to be restriped, run checks against the target | |
9569 | * profile instead of the current one. | |
ba1bf481 JB |
9570 | */ |
9571 | ret = -1; | |
ea8c2819 | 9572 | |
cdcb725c | 9573 | /* |
9574 | * index: | |
9575 | * 0: raid10 | |
9576 | * 1: raid1 | |
9577 | * 2: dup | |
9578 | * 3: raid0 | |
9579 | * 4: single | |
9580 | */ | |
4a5e98f5 ID |
9581 | target = get_restripe_target(root->fs_info, block_group->flags); |
9582 | if (target) { | |
31e50229 | 9583 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9584 | } else { |
9585 | /* | |
9586 | * this is just a balance, so if we were marked as full | |
9587 | * we know there is no space for a new chunk | |
9588 | */ | |
0305bc27 QW |
9589 | if (full) { |
9590 | if (debug) | |
9591 | btrfs_warn(root->fs_info, | |
9592 | "no space to alloc new chunk for block group %llu", | |
9593 | block_group->key.objectid); | |
4a5e98f5 | 9594 | goto out; |
0305bc27 | 9595 | } |
4a5e98f5 ID |
9596 | |
9597 | index = get_block_group_index(block_group); | |
9598 | } | |
9599 | ||
e6ec716f | 9600 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9601 | dev_min = 4; |
6719db6a JB |
9602 | /* Divide by 2 */ |
9603 | min_free >>= 1; | |
e6ec716f | 9604 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9605 | dev_min = 2; |
e6ec716f | 9606 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9607 | /* Multiply by 2 */ |
9608 | min_free <<= 1; | |
e6ec716f | 9609 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9610 | dev_min = fs_devices->rw_devices; |
47c5713f | 9611 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9612 | } |
9613 | ||
6df9a95e JB |
9614 | /* We need to do this so that we can look at pending chunks */ |
9615 | trans = btrfs_join_transaction(root); | |
9616 | if (IS_ERR(trans)) { | |
9617 | ret = PTR_ERR(trans); | |
9618 | goto out; | |
9619 | } | |
9620 | ||
ba1bf481 JB |
9621 | mutex_lock(&root->fs_info->chunk_mutex); |
9622 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 9623 | u64 dev_offset; |
56bec294 | 9624 | |
ba1bf481 JB |
9625 | /* |
9626 | * check to make sure we can actually find a chunk with enough | |
9627 | * space to fit our block group in. | |
9628 | */ | |
63a212ab SB |
9629 | if (device->total_bytes > device->bytes_used + min_free && |
9630 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9631 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9632 | &dev_offset, NULL); |
ba1bf481 | 9633 | if (!ret) |
cdcb725c | 9634 | dev_nr++; |
9635 | ||
9636 | if (dev_nr >= dev_min) | |
73e48b27 | 9637 | break; |
cdcb725c | 9638 | |
ba1bf481 | 9639 | ret = -1; |
725c8463 | 9640 | } |
edbd8d4e | 9641 | } |
0305bc27 QW |
9642 | if (debug && ret == -1) |
9643 | btrfs_warn(root->fs_info, | |
9644 | "no space to allocate a new chunk for block group %llu", | |
9645 | block_group->key.objectid); | |
ba1bf481 | 9646 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 9647 | btrfs_end_transaction(trans, root); |
edbd8d4e | 9648 | out: |
ba1bf481 | 9649 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9650 | return ret; |
9651 | } | |
9652 | ||
b2950863 CH |
9653 | static int find_first_block_group(struct btrfs_root *root, |
9654 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 9655 | { |
925baedd | 9656 | int ret = 0; |
0b86a832 CM |
9657 | struct btrfs_key found_key; |
9658 | struct extent_buffer *leaf; | |
9659 | int slot; | |
edbd8d4e | 9660 | |
0b86a832 CM |
9661 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9662 | if (ret < 0) | |
925baedd CM |
9663 | goto out; |
9664 | ||
d397712b | 9665 | while (1) { |
0b86a832 | 9666 | slot = path->slots[0]; |
edbd8d4e | 9667 | leaf = path->nodes[0]; |
0b86a832 CM |
9668 | if (slot >= btrfs_header_nritems(leaf)) { |
9669 | ret = btrfs_next_leaf(root, path); | |
9670 | if (ret == 0) | |
9671 | continue; | |
9672 | if (ret < 0) | |
925baedd | 9673 | goto out; |
0b86a832 | 9674 | break; |
edbd8d4e | 9675 | } |
0b86a832 | 9676 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9677 | |
0b86a832 | 9678 | if (found_key.objectid >= key->objectid && |
925baedd CM |
9679 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
9680 | ret = 0; | |
9681 | goto out; | |
9682 | } | |
0b86a832 | 9683 | path->slots[0]++; |
edbd8d4e | 9684 | } |
925baedd | 9685 | out: |
0b86a832 | 9686 | return ret; |
edbd8d4e CM |
9687 | } |
9688 | ||
0af3d00b JB |
9689 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9690 | { | |
9691 | struct btrfs_block_group_cache *block_group; | |
9692 | u64 last = 0; | |
9693 | ||
9694 | while (1) { | |
9695 | struct inode *inode; | |
9696 | ||
9697 | block_group = btrfs_lookup_first_block_group(info, last); | |
9698 | while (block_group) { | |
9699 | spin_lock(&block_group->lock); | |
9700 | if (block_group->iref) | |
9701 | break; | |
9702 | spin_unlock(&block_group->lock); | |
9703 | block_group = next_block_group(info->tree_root, | |
9704 | block_group); | |
9705 | } | |
9706 | if (!block_group) { | |
9707 | if (last == 0) | |
9708 | break; | |
9709 | last = 0; | |
9710 | continue; | |
9711 | } | |
9712 | ||
9713 | inode = block_group->inode; | |
9714 | block_group->iref = 0; | |
9715 | block_group->inode = NULL; | |
9716 | spin_unlock(&block_group->lock); | |
9717 | iput(inode); | |
9718 | last = block_group->key.objectid + block_group->key.offset; | |
9719 | btrfs_put_block_group(block_group); | |
9720 | } | |
9721 | } | |
9722 | ||
1a40e23b ZY |
9723 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9724 | { | |
9725 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9726 | struct btrfs_space_info *space_info; |
11833d66 | 9727 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9728 | struct rb_node *n; |
9729 | ||
9e351cc8 | 9730 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9731 | while (!list_empty(&info->caching_block_groups)) { |
9732 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9733 | struct btrfs_caching_control, list); | |
9734 | list_del(&caching_ctl->list); | |
9735 | put_caching_control(caching_ctl); | |
9736 | } | |
9e351cc8 | 9737 | up_write(&info->commit_root_sem); |
11833d66 | 9738 | |
47ab2a6c JB |
9739 | spin_lock(&info->unused_bgs_lock); |
9740 | while (!list_empty(&info->unused_bgs)) { | |
9741 | block_group = list_first_entry(&info->unused_bgs, | |
9742 | struct btrfs_block_group_cache, | |
9743 | bg_list); | |
9744 | list_del_init(&block_group->bg_list); | |
9745 | btrfs_put_block_group(block_group); | |
9746 | } | |
9747 | spin_unlock(&info->unused_bgs_lock); | |
9748 | ||
1a40e23b ZY |
9749 | spin_lock(&info->block_group_cache_lock); |
9750 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9751 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9752 | cache_node); | |
1a40e23b ZY |
9753 | rb_erase(&block_group->cache_node, |
9754 | &info->block_group_cache_tree); | |
01eacb27 | 9755 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9756 | spin_unlock(&info->block_group_cache_lock); |
9757 | ||
80eb234a | 9758 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9759 | list_del(&block_group->list); |
80eb234a | 9760 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9761 | |
817d52f8 | 9762 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9763 | wait_block_group_cache_done(block_group); |
817d52f8 | 9764 | |
3c14874a JB |
9765 | /* |
9766 | * We haven't cached this block group, which means we could | |
9767 | * possibly have excluded extents on this block group. | |
9768 | */ | |
36cce922 JB |
9769 | if (block_group->cached == BTRFS_CACHE_NO || |
9770 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
9771 | free_excluded_extents(info->extent_root, block_group); |
9772 | ||
817d52f8 | 9773 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 9774 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9775 | |
9776 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9777 | } |
9778 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9779 | |
9780 | /* now that all the block groups are freed, go through and | |
9781 | * free all the space_info structs. This is only called during | |
9782 | * the final stages of unmount, and so we know nobody is | |
9783 | * using them. We call synchronize_rcu() once before we start, | |
9784 | * just to be on the safe side. | |
9785 | */ | |
9786 | synchronize_rcu(); | |
9787 | ||
8929ecfa YZ |
9788 | release_global_block_rsv(info); |
9789 | ||
67871254 | 9790 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9791 | int i; |
9792 | ||
4184ea7f CM |
9793 | space_info = list_entry(info->space_info.next, |
9794 | struct btrfs_space_info, | |
9795 | list); | |
b069e0c3 | 9796 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
fae7f21c | 9797 | if (WARN_ON(space_info->bytes_pinned > 0 || |
b069e0c3 | 9798 | space_info->bytes_reserved > 0 || |
fae7f21c | 9799 | space_info->bytes_may_use > 0)) { |
b069e0c3 DS |
9800 | dump_space_info(space_info, 0, 0); |
9801 | } | |
f0486c68 | 9802 | } |
4184ea7f | 9803 | list_del(&space_info->list); |
6ab0a202 JM |
9804 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9805 | struct kobject *kobj; | |
c1895442 JM |
9806 | kobj = space_info->block_group_kobjs[i]; |
9807 | space_info->block_group_kobjs[i] = NULL; | |
9808 | if (kobj) { | |
6ab0a202 JM |
9809 | kobject_del(kobj); |
9810 | kobject_put(kobj); | |
9811 | } | |
9812 | } | |
9813 | kobject_del(&space_info->kobj); | |
9814 | kobject_put(&space_info->kobj); | |
4184ea7f | 9815 | } |
1a40e23b ZY |
9816 | return 0; |
9817 | } | |
9818 | ||
b742bb82 YZ |
9819 | static void __link_block_group(struct btrfs_space_info *space_info, |
9820 | struct btrfs_block_group_cache *cache) | |
9821 | { | |
9822 | int index = get_block_group_index(cache); | |
ed55b6ac | 9823 | bool first = false; |
b742bb82 YZ |
9824 | |
9825 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9826 | if (list_empty(&space_info->block_groups[index])) |
9827 | first = true; | |
9828 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9829 | up_write(&space_info->groups_sem); | |
9830 | ||
9831 | if (first) { | |
c1895442 | 9832 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9833 | int ret; |
9834 | ||
c1895442 JM |
9835 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9836 | if (!rkobj) | |
9837 | goto out_err; | |
9838 | rkobj->raid_type = index; | |
9839 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9840 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9841 | "%s", get_raid_name(index)); | |
6ab0a202 | 9842 | if (ret) { |
c1895442 JM |
9843 | kobject_put(&rkobj->kobj); |
9844 | goto out_err; | |
6ab0a202 | 9845 | } |
c1895442 | 9846 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9847 | } |
c1895442 JM |
9848 | |
9849 | return; | |
9850 | out_err: | |
9851 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); | |
b742bb82 YZ |
9852 | } |
9853 | ||
920e4a58 MX |
9854 | static struct btrfs_block_group_cache * |
9855 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
9856 | { | |
9857 | struct btrfs_block_group_cache *cache; | |
9858 | ||
9859 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9860 | if (!cache) | |
9861 | return NULL; | |
9862 | ||
9863 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9864 | GFP_NOFS); | |
9865 | if (!cache->free_space_ctl) { | |
9866 | kfree(cache); | |
9867 | return NULL; | |
9868 | } | |
9869 | ||
9870 | cache->key.objectid = start; | |
9871 | cache->key.offset = size; | |
9872 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9873 | ||
9874 | cache->sectorsize = root->sectorsize; | |
9875 | cache->fs_info = root->fs_info; | |
9876 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
9877 | &root->fs_info->mapping_tree, | |
9878 | start); | |
1e144fb8 OS |
9879 | set_free_space_tree_thresholds(cache); |
9880 | ||
920e4a58 MX |
9881 | atomic_set(&cache->count, 1); |
9882 | spin_lock_init(&cache->lock); | |
e570fd27 | 9883 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9884 | INIT_LIST_HEAD(&cache->list); |
9885 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9886 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9887 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9888 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9889 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9890 | btrfs_init_free_space_ctl(cache); |
04216820 | 9891 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9892 | mutex_init(&cache->free_space_lock); |
920e4a58 MX |
9893 | |
9894 | return cache; | |
9895 | } | |
9896 | ||
9078a3e1 CM |
9897 | int btrfs_read_block_groups(struct btrfs_root *root) |
9898 | { | |
9899 | struct btrfs_path *path; | |
9900 | int ret; | |
9078a3e1 | 9901 | struct btrfs_block_group_cache *cache; |
be744175 | 9902 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 9903 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9904 | struct btrfs_key key; |
9905 | struct btrfs_key found_key; | |
5f39d397 | 9906 | struct extent_buffer *leaf; |
0af3d00b JB |
9907 | int need_clear = 0; |
9908 | u64 cache_gen; | |
96b5179d | 9909 | |
be744175 | 9910 | root = info->extent_root; |
9078a3e1 | 9911 | key.objectid = 0; |
0b86a832 | 9912 | key.offset = 0; |
962a298f | 9913 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9914 | path = btrfs_alloc_path(); |
9915 | if (!path) | |
9916 | return -ENOMEM; | |
e4058b54 | 9917 | path->reada = READA_FORWARD; |
9078a3e1 | 9918 | |
6c41761f | 9919 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 9920 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 9921 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 9922 | need_clear = 1; |
88c2ba3b JB |
9923 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
9924 | need_clear = 1; | |
0af3d00b | 9925 | |
d397712b | 9926 | while (1) { |
0b86a832 | 9927 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
9928 | if (ret > 0) |
9929 | break; | |
0b86a832 CM |
9930 | if (ret != 0) |
9931 | goto error; | |
920e4a58 | 9932 | |
5f39d397 CM |
9933 | leaf = path->nodes[0]; |
9934 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 MX |
9935 | |
9936 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
9937 | found_key.offset); | |
9078a3e1 | 9938 | if (!cache) { |
0b86a832 | 9939 | ret = -ENOMEM; |
f0486c68 | 9940 | goto error; |
9078a3e1 | 9941 | } |
96303081 | 9942 | |
cf7c1ef6 LB |
9943 | if (need_clear) { |
9944 | /* | |
9945 | * When we mount with old space cache, we need to | |
9946 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9947 | * | |
9948 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9949 | * truncate the old free space cache inode and | |
9950 | * setup a new one. | |
9951 | * b) Setting 'dirty flag' makes sure that we flush | |
9952 | * the new space cache info onto disk. | |
9953 | */ | |
cf7c1ef6 | 9954 | if (btrfs_test_opt(root, SPACE_CACHE)) |
ce93ec54 | 9955 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 9956 | } |
0af3d00b | 9957 | |
5f39d397 CM |
9958 | read_extent_buffer(leaf, &cache->item, |
9959 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9960 | sizeof(cache->item)); | |
920e4a58 | 9961 | cache->flags = btrfs_block_group_flags(&cache->item); |
0b86a832 | 9962 | |
9078a3e1 | 9963 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 9964 | btrfs_release_path(path); |
34d52cb6 | 9965 | |
3c14874a JB |
9966 | /* |
9967 | * We need to exclude the super stripes now so that the space | |
9968 | * info has super bytes accounted for, otherwise we'll think | |
9969 | * we have more space than we actually do. | |
9970 | */ | |
835d974f JB |
9971 | ret = exclude_super_stripes(root, cache); |
9972 | if (ret) { | |
9973 | /* | |
9974 | * We may have excluded something, so call this just in | |
9975 | * case. | |
9976 | */ | |
9977 | free_excluded_extents(root, cache); | |
920e4a58 | 9978 | btrfs_put_block_group(cache); |
835d974f JB |
9979 | goto error; |
9980 | } | |
3c14874a | 9981 | |
817d52f8 JB |
9982 | /* |
9983 | * check for two cases, either we are full, and therefore | |
9984 | * don't need to bother with the caching work since we won't | |
9985 | * find any space, or we are empty, and we can just add all | |
9986 | * the space in and be done with it. This saves us _alot_ of | |
9987 | * time, particularly in the full case. | |
9988 | */ | |
9989 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 9990 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9991 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 9992 | free_excluded_extents(root, cache); |
817d52f8 | 9993 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 9994 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
9995 | cache->cached = BTRFS_CACHE_FINISHED; |
9996 | add_new_free_space(cache, root->fs_info, | |
9997 | found_key.objectid, | |
9998 | found_key.objectid + | |
9999 | found_key.offset); | |
11833d66 | 10000 | free_excluded_extents(root, cache); |
817d52f8 | 10001 | } |
96b5179d | 10002 | |
8c579fe7 JB |
10003 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10004 | if (ret) { | |
10005 | btrfs_remove_free_space_cache(cache); | |
10006 | btrfs_put_block_group(cache); | |
10007 | goto error; | |
10008 | } | |
10009 | ||
6324fbf3 CM |
10010 | ret = update_space_info(info, cache->flags, found_key.offset, |
10011 | btrfs_block_group_used(&cache->item), | |
e40edf2d | 10012 | cache->bytes_super, &space_info); |
8c579fe7 JB |
10013 | if (ret) { |
10014 | btrfs_remove_free_space_cache(cache); | |
10015 | spin_lock(&info->block_group_cache_lock); | |
10016 | rb_erase(&cache->cache_node, | |
10017 | &info->block_group_cache_tree); | |
01eacb27 | 10018 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10019 | spin_unlock(&info->block_group_cache_lock); |
10020 | btrfs_put_block_group(cache); | |
10021 | goto error; | |
10022 | } | |
10023 | ||
6324fbf3 | 10024 | cache->space_info = space_info; |
1b2da372 | 10025 | |
b742bb82 | 10026 | __link_block_group(space_info, cache); |
0f9dd46c | 10027 | |
75ccf47d | 10028 | set_avail_alloc_bits(root->fs_info, cache->flags); |
47ab2a6c | 10029 | if (btrfs_chunk_readonly(root, cache->key.objectid)) { |
868f401a | 10030 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10031 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10032 | spin_lock(&info->unused_bgs_lock); | |
10033 | /* Should always be true but just in case. */ | |
10034 | if (list_empty(&cache->bg_list)) { | |
10035 | btrfs_get_block_group(cache); | |
10036 | list_add_tail(&cache->bg_list, | |
10037 | &info->unused_bgs); | |
10038 | } | |
10039 | spin_unlock(&info->unused_bgs_lock); | |
10040 | } | |
9078a3e1 | 10041 | } |
b742bb82 YZ |
10042 | |
10043 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
10044 | if (!(get_alloc_profile(root, space_info->flags) & | |
10045 | (BTRFS_BLOCK_GROUP_RAID10 | | |
10046 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10047 | BTRFS_BLOCK_GROUP_RAID5 | |
10048 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10049 | BTRFS_BLOCK_GROUP_DUP))) |
10050 | continue; | |
10051 | /* | |
10052 | * avoid allocating from un-mirrored block group if there are | |
10053 | * mirrored block groups. | |
10054 | */ | |
1095cc0d | 10055 | list_for_each_entry(cache, |
10056 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10057 | list) | |
868f401a | 10058 | inc_block_group_ro(cache, 1); |
1095cc0d | 10059 | list_for_each_entry(cache, |
10060 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10061 | list) | |
868f401a | 10062 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10063 | } |
f0486c68 YZ |
10064 | |
10065 | init_global_block_rsv(info); | |
0b86a832 CM |
10066 | ret = 0; |
10067 | error: | |
9078a3e1 | 10068 | btrfs_free_path(path); |
0b86a832 | 10069 | return ret; |
9078a3e1 | 10070 | } |
6324fbf3 | 10071 | |
ea658bad JB |
10072 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
10073 | struct btrfs_root *root) | |
10074 | { | |
10075 | struct btrfs_block_group_cache *block_group, *tmp; | |
10076 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
10077 | struct btrfs_block_group_item item; | |
10078 | struct btrfs_key key; | |
10079 | int ret = 0; | |
d9a0540a | 10080 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10081 | |
d9a0540a | 10082 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10083 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10084 | if (ret) |
c92f6be3 | 10085 | goto next; |
ea658bad JB |
10086 | |
10087 | spin_lock(&block_group->lock); | |
10088 | memcpy(&item, &block_group->item, sizeof(item)); | |
10089 | memcpy(&key, &block_group->key, sizeof(key)); | |
10090 | spin_unlock(&block_group->lock); | |
10091 | ||
10092 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10093 | sizeof(item)); | |
10094 | if (ret) | |
10095 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
10096 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
10097 | key.objectid, key.offset); | |
10098 | if (ret) | |
10099 | btrfs_abort_transaction(trans, extent_root, ret); | |
1e144fb8 OS |
10100 | add_block_group_free_space(trans, root->fs_info, block_group); |
10101 | /* already aborted the transaction if it failed. */ | |
c92f6be3 FM |
10102 | next: |
10103 | list_del_init(&block_group->bg_list); | |
ea658bad | 10104 | } |
d9a0540a | 10105 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10106 | } |
10107 | ||
6324fbf3 CM |
10108 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
10109 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 10110 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
10111 | u64 size) |
10112 | { | |
10113 | int ret; | |
6324fbf3 CM |
10114 | struct btrfs_root *extent_root; |
10115 | struct btrfs_block_group_cache *cache; | |
6324fbf3 | 10116 | extent_root = root->fs_info->extent_root; |
6324fbf3 | 10117 | |
995946dd | 10118 | btrfs_set_log_full_commit(root->fs_info, trans); |
e02119d5 | 10119 | |
920e4a58 | 10120 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
0f9dd46c JB |
10121 | if (!cache) |
10122 | return -ENOMEM; | |
34d52cb6 | 10123 | |
6324fbf3 | 10124 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 10125 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
10126 | btrfs_set_block_group_flags(&cache->item, type); |
10127 | ||
920e4a58 | 10128 | cache->flags = type; |
11833d66 | 10129 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10130 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10131 | cache->needs_free_space = 1; |
835d974f JB |
10132 | ret = exclude_super_stripes(root, cache); |
10133 | if (ret) { | |
10134 | /* | |
10135 | * We may have excluded something, so call this just in | |
10136 | * case. | |
10137 | */ | |
10138 | free_excluded_extents(root, cache); | |
920e4a58 | 10139 | btrfs_put_block_group(cache); |
835d974f JB |
10140 | return ret; |
10141 | } | |
96303081 | 10142 | |
817d52f8 JB |
10143 | add_new_free_space(cache, root->fs_info, chunk_offset, |
10144 | chunk_offset + size); | |
10145 | ||
11833d66 YZ |
10146 | free_excluded_extents(root, cache); |
10147 | ||
d0bd4560 JB |
10148 | #ifdef CONFIG_BTRFS_DEBUG |
10149 | if (btrfs_should_fragment_free_space(root, cache)) { | |
10150 | u64 new_bytes_used = size - bytes_used; | |
10151 | ||
10152 | bytes_used += new_bytes_used >> 1; | |
10153 | fragment_free_space(root, cache); | |
10154 | } | |
10155 | #endif | |
2e6e5183 FM |
10156 | /* |
10157 | * Call to ensure the corresponding space_info object is created and | |
10158 | * assigned to our block group, but don't update its counters just yet. | |
10159 | * We want our bg to be added to the rbtree with its ->space_info set. | |
10160 | */ | |
e40edf2d | 10161 | ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0, |
2e6e5183 FM |
10162 | &cache->space_info); |
10163 | if (ret) { | |
10164 | btrfs_remove_free_space_cache(cache); | |
10165 | btrfs_put_block_group(cache); | |
10166 | return ret; | |
10167 | } | |
10168 | ||
8c579fe7 JB |
10169 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10170 | if (ret) { | |
10171 | btrfs_remove_free_space_cache(cache); | |
10172 | btrfs_put_block_group(cache); | |
10173 | return ret; | |
10174 | } | |
10175 | ||
2e6e5183 FM |
10176 | /* |
10177 | * Now that our block group has its ->space_info set and is inserted in | |
10178 | * the rbtree, update the space info's counters. | |
10179 | */ | |
6324fbf3 | 10180 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10181 | cache->bytes_super, &cache->space_info); |
8c579fe7 JB |
10182 | if (ret) { |
10183 | btrfs_remove_free_space_cache(cache); | |
10184 | spin_lock(&root->fs_info->block_group_cache_lock); | |
10185 | rb_erase(&cache->cache_node, | |
10186 | &root->fs_info->block_group_cache_tree); | |
01eacb27 | 10187 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10188 | spin_unlock(&root->fs_info->block_group_cache_lock); |
10189 | btrfs_put_block_group(cache); | |
10190 | return ret; | |
10191 | } | |
c7c144db | 10192 | update_global_block_rsv(root->fs_info); |
1b2da372 | 10193 | |
b742bb82 | 10194 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10195 | |
47ab2a6c | 10196 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10197 | |
d18a2c44 | 10198 | set_avail_alloc_bits(extent_root->fs_info, type); |
6324fbf3 CM |
10199 | return 0; |
10200 | } | |
1a40e23b | 10201 | |
10ea00f5 ID |
10202 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10203 | { | |
899c81ea ID |
10204 | u64 extra_flags = chunk_to_extended(flags) & |
10205 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10206 | |
de98ced9 | 10207 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10208 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10209 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10210 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10211 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10212 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10213 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10214 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10215 | } |
10216 | ||
1a40e23b | 10217 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
04216820 FM |
10218 | struct btrfs_root *root, u64 group_start, |
10219 | struct extent_map *em) | |
1a40e23b ZY |
10220 | { |
10221 | struct btrfs_path *path; | |
10222 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10223 | struct btrfs_free_cluster *cluster; |
0af3d00b | 10224 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 10225 | struct btrfs_key key; |
0af3d00b | 10226 | struct inode *inode; |
c1895442 | 10227 | struct kobject *kobj = NULL; |
1a40e23b | 10228 | int ret; |
10ea00f5 | 10229 | int index; |
89a55897 | 10230 | int factor; |
4f69cb98 | 10231 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10232 | bool remove_em; |
1a40e23b | 10233 | |
1a40e23b ZY |
10234 | root = root->fs_info->extent_root; |
10235 | ||
10236 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
10237 | BUG_ON(!block_group); | |
c146afad | 10238 | BUG_ON(!block_group->ro); |
1a40e23b | 10239 | |
9f7c43c9 | 10240 | /* |
10241 | * Free the reserved super bytes from this block group before | |
10242 | * remove it. | |
10243 | */ | |
10244 | free_excluded_extents(root, block_group); | |
10245 | ||
1a40e23b | 10246 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10247 | index = get_block_group_index(block_group); |
89a55897 JB |
10248 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10249 | BTRFS_BLOCK_GROUP_RAID1 | | |
10250 | BTRFS_BLOCK_GROUP_RAID10)) | |
10251 | factor = 2; | |
10252 | else | |
10253 | factor = 1; | |
1a40e23b | 10254 | |
44fb5511 CM |
10255 | /* make sure this block group isn't part of an allocation cluster */ |
10256 | cluster = &root->fs_info->data_alloc_cluster; | |
10257 | spin_lock(&cluster->refill_lock); | |
10258 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10259 | spin_unlock(&cluster->refill_lock); | |
10260 | ||
10261 | /* | |
10262 | * make sure this block group isn't part of a metadata | |
10263 | * allocation cluster | |
10264 | */ | |
10265 | cluster = &root->fs_info->meta_alloc_cluster; | |
10266 | spin_lock(&cluster->refill_lock); | |
10267 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10268 | spin_unlock(&cluster->refill_lock); | |
10269 | ||
1a40e23b | 10270 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10271 | if (!path) { |
10272 | ret = -ENOMEM; | |
10273 | goto out; | |
10274 | } | |
1a40e23b | 10275 | |
1bbc621e CM |
10276 | /* |
10277 | * get the inode first so any iput calls done for the io_list | |
10278 | * aren't the final iput (no unlinks allowed now) | |
10279 | */ | |
10b2f34d | 10280 | inode = lookup_free_space_inode(tree_root, block_group, path); |
1bbc621e CM |
10281 | |
10282 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10283 | /* | |
10284 | * make sure our free spache cache IO is done before remove the | |
10285 | * free space inode | |
10286 | */ | |
10287 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10288 | if (!list_empty(&block_group->io_list)) { | |
10289 | list_del_init(&block_group->io_list); | |
10290 | ||
10291 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10292 | ||
10293 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10294 | btrfs_wait_cache_io(root, trans, block_group, | |
10295 | &block_group->io_ctl, path, | |
10296 | block_group->key.objectid); | |
10297 | btrfs_put_block_group(block_group); | |
10298 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10299 | } | |
10300 | ||
10301 | if (!list_empty(&block_group->dirty_list)) { | |
10302 | list_del_init(&block_group->dirty_list); | |
10303 | btrfs_put_block_group(block_group); | |
10304 | } | |
10305 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10306 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10307 | ||
0af3d00b | 10308 | if (!IS_ERR(inode)) { |
b532402e | 10309 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
10310 | if (ret) { |
10311 | btrfs_add_delayed_iput(inode); | |
10312 | goto out; | |
10313 | } | |
0af3d00b JB |
10314 | clear_nlink(inode); |
10315 | /* One for the block groups ref */ | |
10316 | spin_lock(&block_group->lock); | |
10317 | if (block_group->iref) { | |
10318 | block_group->iref = 0; | |
10319 | block_group->inode = NULL; | |
10320 | spin_unlock(&block_group->lock); | |
10321 | iput(inode); | |
10322 | } else { | |
10323 | spin_unlock(&block_group->lock); | |
10324 | } | |
10325 | /* One for our lookup ref */ | |
455757c3 | 10326 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10327 | } |
10328 | ||
10329 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10330 | key.offset = block_group->key.objectid; | |
10331 | key.type = 0; | |
10332 | ||
10333 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10334 | if (ret < 0) | |
10335 | goto out; | |
10336 | if (ret > 0) | |
b3b4aa74 | 10337 | btrfs_release_path(path); |
0af3d00b JB |
10338 | if (ret == 0) { |
10339 | ret = btrfs_del_item(trans, tree_root, path); | |
10340 | if (ret) | |
10341 | goto out; | |
b3b4aa74 | 10342 | btrfs_release_path(path); |
0af3d00b JB |
10343 | } |
10344 | ||
3dfdb934 | 10345 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
10346 | rb_erase(&block_group->cache_node, |
10347 | &root->fs_info->block_group_cache_tree); | |
292cbd51 | 10348 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd LB |
10349 | |
10350 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
10351 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 10352 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 10353 | |
80eb234a | 10354 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10355 | /* |
10356 | * we must use list_del_init so people can check to see if they | |
10357 | * are still on the list after taking the semaphore | |
10358 | */ | |
10359 | list_del_init(&block_group->list); | |
6ab0a202 | 10360 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10361 | kobj = block_group->space_info->block_group_kobjs[index]; |
10362 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10ea00f5 | 10363 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
6ab0a202 | 10364 | } |
80eb234a | 10365 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10366 | if (kobj) { |
10367 | kobject_del(kobj); | |
10368 | kobject_put(kobj); | |
10369 | } | |
1a40e23b | 10370 | |
4f69cb98 FM |
10371 | if (block_group->has_caching_ctl) |
10372 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10373 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10374 | wait_block_group_cache_done(block_group); |
4f69cb98 FM |
10375 | if (block_group->has_caching_ctl) { |
10376 | down_write(&root->fs_info->commit_root_sem); | |
10377 | if (!caching_ctl) { | |
10378 | struct btrfs_caching_control *ctl; | |
10379 | ||
10380 | list_for_each_entry(ctl, | |
10381 | &root->fs_info->caching_block_groups, list) | |
10382 | if (ctl->block_group == block_group) { | |
10383 | caching_ctl = ctl; | |
10384 | atomic_inc(&caching_ctl->count); | |
10385 | break; | |
10386 | } | |
10387 | } | |
10388 | if (caching_ctl) | |
10389 | list_del_init(&caching_ctl->list); | |
10390 | up_write(&root->fs_info->commit_root_sem); | |
10391 | if (caching_ctl) { | |
10392 | /* Once for the caching bgs list and once for us. */ | |
10393 | put_caching_control(caching_ctl); | |
10394 | put_caching_control(caching_ctl); | |
10395 | } | |
10396 | } | |
817d52f8 | 10397 | |
ce93ec54 JB |
10398 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10399 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10400 | WARN_ON(1); |
10401 | } | |
10402 | if (!list_empty(&block_group->io_list)) { | |
10403 | WARN_ON(1); | |
ce93ec54 JB |
10404 | } |
10405 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10406 | btrfs_remove_free_space_cache(block_group); |
10407 | ||
c146afad | 10408 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10409 | list_del_init(&block_group->ro_list); |
18d018ad ZL |
10410 | |
10411 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
10412 | WARN_ON(block_group->space_info->total_bytes | |
10413 | < block_group->key.offset); | |
10414 | WARN_ON(block_group->space_info->bytes_readonly | |
10415 | < block_group->key.offset); | |
10416 | WARN_ON(block_group->space_info->disk_total | |
10417 | < block_group->key.offset * factor); | |
10418 | } | |
c146afad YZ |
10419 | block_group->space_info->total_bytes -= block_group->key.offset; |
10420 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10421 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10422 | |
c146afad | 10423 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10424 | |
0af3d00b JB |
10425 | memcpy(&key, &block_group->key, sizeof(key)); |
10426 | ||
04216820 | 10427 | lock_chunks(root); |
495e64f4 FM |
10428 | if (!list_empty(&em->list)) { |
10429 | /* We're in the transaction->pending_chunks list. */ | |
10430 | free_extent_map(em); | |
10431 | } | |
04216820 FM |
10432 | spin_lock(&block_group->lock); |
10433 | block_group->removed = 1; | |
10434 | /* | |
10435 | * At this point trimming can't start on this block group, because we | |
10436 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10437 | * so no one can't find it anymore and even if someone already got this | |
10438 | * block group before we removed it from the rbtree, they have already | |
10439 | * incremented block_group->trimming - if they didn't, they won't find | |
10440 | * any free space entries because we already removed them all when we | |
10441 | * called btrfs_remove_free_space_cache(). | |
10442 | * | |
10443 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10444 | * to prevent the same logical address range and physical device space | |
10445 | * ranges from being reused for a new block group. This is because our | |
10446 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10447 | * completely transactionless, so while it is trimming a range the | |
10448 | * currently running transaction might finish and a new one start, | |
10449 | * allowing for new block groups to be created that can reuse the same | |
10450 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10451 | * |
10452 | * There may also be an implicit trim operation if the file system | |
10453 | * is mounted with -odiscard. The same protections must remain | |
10454 | * in place until the extents have been discarded completely when | |
10455 | * the transaction commit has completed. | |
04216820 FM |
10456 | */ |
10457 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10458 | /* | |
10459 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10460 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10461 | * before checking block_group->removed). | |
10462 | */ | |
10463 | if (!remove_em) { | |
10464 | /* | |
10465 | * Our em might be in trans->transaction->pending_chunks which | |
10466 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10467 | * and so is the fs_info->pinned_chunks list. | |
10468 | * | |
10469 | * So at this point we must be holding the chunk_mutex to avoid | |
10470 | * any races with chunk allocation (more specifically at | |
10471 | * volumes.c:contains_pending_extent()), to ensure it always | |
10472 | * sees the em, either in the pending_chunks list or in the | |
10473 | * pinned_chunks list. | |
10474 | */ | |
10475 | list_move_tail(&em->list, &root->fs_info->pinned_chunks); | |
10476 | } | |
10477 | spin_unlock(&block_group->lock); | |
04216820 FM |
10478 | |
10479 | if (remove_em) { | |
10480 | struct extent_map_tree *em_tree; | |
10481 | ||
10482 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
10483 | write_lock(&em_tree->lock); | |
8dbcd10f FM |
10484 | /* |
10485 | * The em might be in the pending_chunks list, so make sure the | |
10486 | * chunk mutex is locked, since remove_extent_mapping() will | |
10487 | * delete us from that list. | |
10488 | */ | |
04216820 FM |
10489 | remove_extent_mapping(em_tree, em); |
10490 | write_unlock(&em_tree->lock); | |
10491 | /* once for the tree */ | |
10492 | free_extent_map(em); | |
10493 | } | |
10494 | ||
8dbcd10f FM |
10495 | unlock_chunks(root); |
10496 | ||
1e144fb8 OS |
10497 | ret = remove_block_group_free_space(trans, root->fs_info, block_group); |
10498 | if (ret) | |
10499 | goto out; | |
10500 | ||
fa9c0d79 CM |
10501 | btrfs_put_block_group(block_group); |
10502 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10503 | |
10504 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10505 | if (ret > 0) | |
10506 | ret = -EIO; | |
10507 | if (ret < 0) | |
10508 | goto out; | |
10509 | ||
10510 | ret = btrfs_del_item(trans, root, path); | |
10511 | out: | |
10512 | btrfs_free_path(path); | |
10513 | return ret; | |
10514 | } | |
acce952b | 10515 | |
8eab77ff | 10516 | struct btrfs_trans_handle * |
7fd01182 FM |
10517 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10518 | const u64 chunk_offset) | |
8eab77ff | 10519 | { |
7fd01182 FM |
10520 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10521 | struct extent_map *em; | |
10522 | struct map_lookup *map; | |
10523 | unsigned int num_items; | |
10524 | ||
10525 | read_lock(&em_tree->lock); | |
10526 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10527 | read_unlock(&em_tree->lock); | |
10528 | ASSERT(em && em->start == chunk_offset); | |
10529 | ||
8eab77ff | 10530 | /* |
7fd01182 FM |
10531 | * We need to reserve 3 + N units from the metadata space info in order |
10532 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10533 | * btrfs_remove_block_group()), which are used for: | |
10534 | * | |
8eab77ff FM |
10535 | * 1 unit for adding the free space inode's orphan (located in the tree |
10536 | * of tree roots). | |
7fd01182 FM |
10537 | * 1 unit for deleting the block group item (located in the extent |
10538 | * tree). | |
10539 | * 1 unit for deleting the free space item (located in tree of tree | |
10540 | * roots). | |
10541 | * N units for deleting N device extent items corresponding to each | |
10542 | * stripe (located in the device tree). | |
10543 | * | |
10544 | * In order to remove a block group we also need to reserve units in the | |
10545 | * system space info in order to update the chunk tree (update one or | |
10546 | * more device items and remove one chunk item), but this is done at | |
10547 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10548 | */ |
95617d69 | 10549 | map = em->map_lookup; |
7fd01182 FM |
10550 | num_items = 3 + map->num_stripes; |
10551 | free_extent_map(em); | |
10552 | ||
8eab77ff | 10553 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10554 | num_items, 1); |
8eab77ff FM |
10555 | } |
10556 | ||
47ab2a6c JB |
10557 | /* |
10558 | * Process the unused_bgs list and remove any that don't have any allocated | |
10559 | * space inside of them. | |
10560 | */ | |
10561 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10562 | { | |
10563 | struct btrfs_block_group_cache *block_group; | |
10564 | struct btrfs_space_info *space_info; | |
10565 | struct btrfs_root *root = fs_info->extent_root; | |
10566 | struct btrfs_trans_handle *trans; | |
10567 | int ret = 0; | |
10568 | ||
10569 | if (!fs_info->open) | |
10570 | return; | |
10571 | ||
10572 | spin_lock(&fs_info->unused_bgs_lock); | |
10573 | while (!list_empty(&fs_info->unused_bgs)) { | |
10574 | u64 start, end; | |
e33e17ee | 10575 | int trimming; |
47ab2a6c JB |
10576 | |
10577 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10578 | struct btrfs_block_group_cache, | |
10579 | bg_list); | |
47ab2a6c | 10580 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10581 | |
10582 | space_info = block_group->space_info; | |
10583 | ||
47ab2a6c JB |
10584 | if (ret || btrfs_mixed_space_info(space_info)) { |
10585 | btrfs_put_block_group(block_group); | |
10586 | continue; | |
10587 | } | |
10588 | spin_unlock(&fs_info->unused_bgs_lock); | |
10589 | ||
d5f2e33b | 10590 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10591 | |
47ab2a6c JB |
10592 | /* Don't want to race with allocators so take the groups_sem */ |
10593 | down_write(&space_info->groups_sem); | |
10594 | spin_lock(&block_group->lock); | |
10595 | if (block_group->reserved || | |
10596 | btrfs_block_group_used(&block_group->item) || | |
aefbe9a6 ZL |
10597 | block_group->ro || |
10598 | list_is_singular(&block_group->list)) { | |
47ab2a6c JB |
10599 | /* |
10600 | * We want to bail if we made new allocations or have | |
10601 | * outstanding allocations in this block group. We do | |
10602 | * the ro check in case balance is currently acting on | |
10603 | * this block group. | |
10604 | */ | |
10605 | spin_unlock(&block_group->lock); | |
10606 | up_write(&space_info->groups_sem); | |
10607 | goto next; | |
10608 | } | |
10609 | spin_unlock(&block_group->lock); | |
10610 | ||
10611 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10612 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10613 | up_write(&space_info->groups_sem); |
10614 | if (ret < 0) { | |
10615 | ret = 0; | |
10616 | goto next; | |
10617 | } | |
10618 | ||
10619 | /* | |
10620 | * Want to do this before we do anything else so we can recover | |
10621 | * properly if we fail to join the transaction. | |
10622 | */ | |
7fd01182 FM |
10623 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10624 | block_group->key.objectid); | |
47ab2a6c | 10625 | if (IS_ERR(trans)) { |
868f401a | 10626 | btrfs_dec_block_group_ro(root, block_group); |
47ab2a6c JB |
10627 | ret = PTR_ERR(trans); |
10628 | goto next; | |
10629 | } | |
10630 | ||
10631 | /* | |
10632 | * We could have pending pinned extents for this block group, | |
10633 | * just delete them, we don't care about them anymore. | |
10634 | */ | |
10635 | start = block_group->key.objectid; | |
10636 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10637 | /* |
10638 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10639 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10640 | * another task might be running finish_extent_commit() for the | |
10641 | * previous transaction N - 1, and have seen a range belonging | |
10642 | * to the block group in freed_extents[] before we were able to | |
10643 | * clear the whole block group range from freed_extents[]. This | |
10644 | * means that task can lookup for the block group after we | |
10645 | * unpinned it from freed_extents[] and removed it, leading to | |
10646 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10647 | */ | |
10648 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10649 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10650 | EXTENT_DIRTY); |
758eb51e | 10651 | if (ret) { |
d4b450cd | 10652 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10653 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10654 | goto end_trans; |
10655 | } | |
10656 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10657 | EXTENT_DIRTY); |
758eb51e | 10658 | if (ret) { |
d4b450cd | 10659 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10660 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10661 | goto end_trans; |
10662 | } | |
d4b450cd | 10663 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10664 | |
10665 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10666 | spin_lock(&space_info->lock); |
10667 | spin_lock(&block_group->lock); | |
10668 | ||
10669 | space_info->bytes_pinned -= block_group->pinned; | |
10670 | space_info->bytes_readonly += block_group->pinned; | |
10671 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10672 | -block_group->pinned); | |
47ab2a6c JB |
10673 | block_group->pinned = 0; |
10674 | ||
c30666d4 ZL |
10675 | spin_unlock(&block_group->lock); |
10676 | spin_unlock(&space_info->lock); | |
10677 | ||
e33e17ee JM |
10678 | /* DISCARD can flip during remount */ |
10679 | trimming = btrfs_test_opt(root, DISCARD); | |
10680 | ||
10681 | /* Implicit trim during transaction commit. */ | |
10682 | if (trimming) | |
10683 | btrfs_get_block_group_trimming(block_group); | |
10684 | ||
47ab2a6c JB |
10685 | /* |
10686 | * Btrfs_remove_chunk will abort the transaction if things go | |
10687 | * horribly wrong. | |
10688 | */ | |
10689 | ret = btrfs_remove_chunk(trans, root, | |
10690 | block_group->key.objectid); | |
e33e17ee JM |
10691 | |
10692 | if (ret) { | |
10693 | if (trimming) | |
10694 | btrfs_put_block_group_trimming(block_group); | |
10695 | goto end_trans; | |
10696 | } | |
10697 | ||
10698 | /* | |
10699 | * If we're not mounted with -odiscard, we can just forget | |
10700 | * about this block group. Otherwise we'll need to wait | |
10701 | * until transaction commit to do the actual discard. | |
10702 | */ | |
10703 | if (trimming) { | |
348a0013 FM |
10704 | spin_lock(&fs_info->unused_bgs_lock); |
10705 | /* | |
10706 | * A concurrent scrub might have added us to the list | |
10707 | * fs_info->unused_bgs, so use a list_move operation | |
10708 | * to add the block group to the deleted_bgs list. | |
10709 | */ | |
e33e17ee JM |
10710 | list_move(&block_group->bg_list, |
10711 | &trans->transaction->deleted_bgs); | |
348a0013 | 10712 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10713 | btrfs_get_block_group(block_group); |
10714 | } | |
758eb51e | 10715 | end_trans: |
47ab2a6c JB |
10716 | btrfs_end_transaction(trans, root); |
10717 | next: | |
d5f2e33b | 10718 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10719 | btrfs_put_block_group(block_group); |
10720 | spin_lock(&fs_info->unused_bgs_lock); | |
10721 | } | |
10722 | spin_unlock(&fs_info->unused_bgs_lock); | |
10723 | } | |
10724 | ||
c59021f8 | 10725 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10726 | { | |
10727 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10728 | struct btrfs_super_block *disk_super; |
10729 | u64 features; | |
10730 | u64 flags; | |
10731 | int mixed = 0; | |
c59021f8 | 10732 | int ret; |
10733 | ||
6c41761f | 10734 | disk_super = fs_info->super_copy; |
1aba86d6 | 10735 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10736 | return -EINVAL; |
c59021f8 | 10737 | |
1aba86d6 | 10738 | features = btrfs_super_incompat_flags(disk_super); |
10739 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10740 | mixed = 1; | |
c59021f8 | 10741 | |
1aba86d6 | 10742 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
e40edf2d | 10743 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
c59021f8 | 10744 | if (ret) |
1aba86d6 | 10745 | goto out; |
c59021f8 | 10746 | |
1aba86d6 | 10747 | if (mixed) { |
10748 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10749 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10750 | } else { |
10751 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
e40edf2d | 10752 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10753 | if (ret) |
10754 | goto out; | |
10755 | ||
10756 | flags = BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10757 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10758 | } |
10759 | out: | |
c59021f8 | 10760 | return ret; |
10761 | } | |
10762 | ||
acce952b | 10763 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
10764 | { | |
678886bd | 10765 | return unpin_extent_range(root, start, end, false); |
acce952b | 10766 | } |
10767 | ||
499f377f JM |
10768 | /* |
10769 | * It used to be that old block groups would be left around forever. | |
10770 | * Iterating over them would be enough to trim unused space. Since we | |
10771 | * now automatically remove them, we also need to iterate over unallocated | |
10772 | * space. | |
10773 | * | |
10774 | * We don't want a transaction for this since the discard may take a | |
10775 | * substantial amount of time. We don't require that a transaction be | |
10776 | * running, but we do need to take a running transaction into account | |
10777 | * to ensure that we're not discarding chunks that were released in | |
10778 | * the current transaction. | |
10779 | * | |
10780 | * Holding the chunks lock will prevent other threads from allocating | |
10781 | * or releasing chunks, but it won't prevent a running transaction | |
10782 | * from committing and releasing the memory that the pending chunks | |
10783 | * list head uses. For that, we need to take a reference to the | |
10784 | * transaction. | |
10785 | */ | |
10786 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10787 | u64 minlen, u64 *trimmed) | |
10788 | { | |
10789 | u64 start = 0, len = 0; | |
10790 | int ret; | |
10791 | ||
10792 | *trimmed = 0; | |
10793 | ||
10794 | /* Not writeable = nothing to do. */ | |
10795 | if (!device->writeable) | |
10796 | return 0; | |
10797 | ||
10798 | /* No free space = nothing to do. */ | |
10799 | if (device->total_bytes <= device->bytes_used) | |
10800 | return 0; | |
10801 | ||
10802 | ret = 0; | |
10803 | ||
10804 | while (1) { | |
10805 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
10806 | struct btrfs_transaction *trans; | |
10807 | u64 bytes; | |
10808 | ||
10809 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10810 | if (ret) | |
10811 | return ret; | |
10812 | ||
10813 | down_read(&fs_info->commit_root_sem); | |
10814 | ||
10815 | spin_lock(&fs_info->trans_lock); | |
10816 | trans = fs_info->running_transaction; | |
10817 | if (trans) | |
10818 | atomic_inc(&trans->use_count); | |
10819 | spin_unlock(&fs_info->trans_lock); | |
10820 | ||
10821 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10822 | &start, &len); | |
10823 | if (trans) | |
10824 | btrfs_put_transaction(trans); | |
10825 | ||
10826 | if (ret) { | |
10827 | up_read(&fs_info->commit_root_sem); | |
10828 | mutex_unlock(&fs_info->chunk_mutex); | |
10829 | if (ret == -ENOSPC) | |
10830 | ret = 0; | |
10831 | break; | |
10832 | } | |
10833 | ||
10834 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10835 | up_read(&fs_info->commit_root_sem); | |
10836 | mutex_unlock(&fs_info->chunk_mutex); | |
10837 | ||
10838 | if (ret) | |
10839 | break; | |
10840 | ||
10841 | start += len; | |
10842 | *trimmed += bytes; | |
10843 | ||
10844 | if (fatal_signal_pending(current)) { | |
10845 | ret = -ERESTARTSYS; | |
10846 | break; | |
10847 | } | |
10848 | ||
10849 | cond_resched(); | |
10850 | } | |
10851 | ||
10852 | return ret; | |
10853 | } | |
10854 | ||
f7039b1d LD |
10855 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) |
10856 | { | |
10857 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10858 | struct btrfs_block_group_cache *cache = NULL; | |
499f377f JM |
10859 | struct btrfs_device *device; |
10860 | struct list_head *devices; | |
f7039b1d LD |
10861 | u64 group_trimmed; |
10862 | u64 start; | |
10863 | u64 end; | |
10864 | u64 trimmed = 0; | |
2cac13e4 | 10865 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10866 | int ret = 0; |
10867 | ||
2cac13e4 LB |
10868 | /* |
10869 | * try to trim all FS space, our block group may start from non-zero. | |
10870 | */ | |
10871 | if (range->len == total_bytes) | |
10872 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10873 | else | |
10874 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10875 | |
10876 | while (cache) { | |
10877 | if (cache->key.objectid >= (range->start + range->len)) { | |
10878 | btrfs_put_block_group(cache); | |
10879 | break; | |
10880 | } | |
10881 | ||
10882 | start = max(range->start, cache->key.objectid); | |
10883 | end = min(range->start + range->len, | |
10884 | cache->key.objectid + cache->key.offset); | |
10885 | ||
10886 | if (end - start >= range->minlen) { | |
10887 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10888 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10889 | if (ret) { |
10890 | btrfs_put_block_group(cache); | |
10891 | break; | |
10892 | } | |
10893 | ret = wait_block_group_cache_done(cache); | |
10894 | if (ret) { | |
10895 | btrfs_put_block_group(cache); | |
10896 | break; | |
10897 | } | |
f7039b1d LD |
10898 | } |
10899 | ret = btrfs_trim_block_group(cache, | |
10900 | &group_trimmed, | |
10901 | start, | |
10902 | end, | |
10903 | range->minlen); | |
10904 | ||
10905 | trimmed += group_trimmed; | |
10906 | if (ret) { | |
10907 | btrfs_put_block_group(cache); | |
10908 | break; | |
10909 | } | |
10910 | } | |
10911 | ||
10912 | cache = next_block_group(fs_info->tree_root, cache); | |
10913 | } | |
10914 | ||
499f377f JM |
10915 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
10916 | devices = &root->fs_info->fs_devices->alloc_list; | |
10917 | list_for_each_entry(device, devices, dev_alloc_list) { | |
10918 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10919 | &group_trimmed); | |
10920 | if (ret) | |
10921 | break; | |
10922 | ||
10923 | trimmed += group_trimmed; | |
10924 | } | |
10925 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
10926 | ||
f7039b1d LD |
10927 | range->len = trimmed; |
10928 | return ret; | |
10929 | } | |
8257b2dc MX |
10930 | |
10931 | /* | |
9ea24bbe FM |
10932 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
10933 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
10934 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10935 | * but flush the data into disk after the snapshot creation, or to prevent | |
10936 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
10937 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 10938 | */ |
9ea24bbe | 10939 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
10940 | { |
10941 | percpu_counter_dec(&root->subv_writers->counter); | |
10942 | /* | |
a83342aa | 10943 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
10944 | */ |
10945 | smp_mb(); | |
10946 | if (waitqueue_active(&root->subv_writers->wait)) | |
10947 | wake_up(&root->subv_writers->wait); | |
10948 | } | |
10949 | ||
9ea24bbe | 10950 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 10951 | { |
ee39b432 | 10952 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
10953 | return 0; |
10954 | ||
10955 | percpu_counter_inc(&root->subv_writers->counter); | |
10956 | /* | |
10957 | * Make sure counter is updated before we check for snapshot creation. | |
10958 | */ | |
10959 | smp_mb(); | |
ee39b432 | 10960 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 10961 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
10962 | return 0; |
10963 | } | |
10964 | return 1; | |
10965 | } | |
0bc19f90 ZL |
10966 | |
10967 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
10968 | { | |
10969 | schedule(); | |
10970 | return 0; | |
10971 | } | |
10972 | ||
10973 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
10974 | { | |
10975 | while (true) { | |
10976 | int ret; | |
10977 | ||
10978 | ret = btrfs_start_write_no_snapshoting(root); | |
10979 | if (ret) | |
10980 | break; | |
10981 | wait_on_atomic_t(&root->will_be_snapshoted, | |
10982 | wait_snapshoting_atomic_t, | |
10983 | TASK_UNINTERRUPTIBLE); | |
10984 | } | |
10985 | } |